OpenCloudOS-Kernel/drivers/mtd/ubi/build.c

1487 lines
43 KiB
C
Raw Normal View History

UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
/*
* Copyright (c) International Business Machines Corp., 2006
* Copyright (c) Nokia Corporation, 2007
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation; either version 2 of the License, or
* (at your option) any later version.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See
* the GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program; if not, write to the Free Software
* Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
*
* Author: Artem Bityutskiy (Битюцкий Артём),
* Frank Haverkamp
*/
/*
* This file includes UBI initialization and building of UBI devices.
*
* When UBI is initialized, it attaches all the MTD devices specified as the
* module load parameters or the kernel boot parameters. If MTD devices were
* specified, UBI does not attach any MTD device, but it is possible to do
* later using the "UBI control device".
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
*/
#include <linux/err.h>
#include <linux/module.h>
#include <linux/moduleparam.h>
#include <linux/stringify.h>
#include <linux/namei.h>
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
#include <linux/stat.h>
#include <linux/miscdevice.h>
UBI: use the whole MTD device size to get bad_peb_limit On NAND flash devices, UBI reserves some physical erase blocks (PEB) for bad block handling. Today, the number of reserved PEB can only be set as a percentage of the total number of PEB in each MTD partition. For example, for a NAND flash with 128KiB PEB, 2 MTD partition of 20MiB (mtd0) and 100MiB (mtd1) and 2% reserved PEB: - the UBI device on mtd0 will have 2 PEB reserved - the UBI device on mtd1 will have 16 PEB reserved The problem with this behaviour is that NAND flash manufacturers give a minimum number of valid block (NVB) during the endurance life of the device, e.g.: Parameter Symbol Min Max Unit Notes -------------------------------------------------------------- Valid block number NVB 1004 1024 Blocks 1 From this number we can deduce the maximum number of bad PEB that a device will contain during its endurance life: a 128MiB NAND flash (1024 PEB) will not have less than 20 bad blocks during the flash endurance life. But the manufacturer doesn't tell where those bad block will appear. He doesn't say either if they will be equally disposed on the whole device (and I'm pretty sure they won't). So, according to the datasheets, we should reserve the maximum number of bad PEB for each UBI device (worst case scenario: 20 bad blocks appears on the smallest MTD partition). So this patch make UBI use the whole MTD device size to calculate the maximum bad expected eraseblocks. The Kconfig option is in per1024 blocks, thus it can have a default value of 20 which is *very* common for NAND devices. Signed-off-by: Richard Genoud <richard.genoud@gmail.com> Signed-off-by: Artem Bityutskiy <artem.bityutskiy@linux.intel.com>
2012-07-11 00:23:41 +08:00
#include <linux/mtd/partitions.h>
#include <linux/log2.h>
#include <linux/kthread.h>
#include <linux/kernel.h>
include cleanup: Update gfp.h and slab.h includes to prepare for breaking implicit slab.h inclusion from percpu.h percpu.h is included by sched.h and module.h and thus ends up being included when building most .c files. percpu.h includes slab.h which in turn includes gfp.h making everything defined by the two files universally available and complicating inclusion dependencies. percpu.h -> slab.h dependency is about to be removed. Prepare for this change by updating users of gfp and slab facilities include those headers directly instead of assuming availability. As this conversion needs to touch large number of source files, the following script is used as the basis of conversion. http://userweb.kernel.org/~tj/misc/slabh-sweep.py The script does the followings. * Scan files for gfp and slab usages and update includes such that only the necessary includes are there. ie. if only gfp is used, gfp.h, if slab is used, slab.h. * When the script inserts a new include, it looks at the include blocks and try to put the new include such that its order conforms to its surrounding. It's put in the include block which contains core kernel includes, in the same order that the rest are ordered - alphabetical, Christmas tree, rev-Xmas-tree or at the end if there doesn't seem to be any matching order. * If the script can't find a place to put a new include (mostly because the file doesn't have fitting include block), it prints out an error message indicating which .h file needs to be added to the file. The conversion was done in the following steps. 1. The initial automatic conversion of all .c files updated slightly over 4000 files, deleting around 700 includes and adding ~480 gfp.h and ~3000 slab.h inclusions. The script emitted errors for ~400 files. 2. Each error was manually checked. Some didn't need the inclusion, some needed manual addition while adding it to implementation .h or embedding .c file was more appropriate for others. This step added inclusions to around 150 files. 3. The script was run again and the output was compared to the edits from #2 to make sure no file was left behind. 4. Several build tests were done and a couple of problems were fixed. e.g. lib/decompress_*.c used malloc/free() wrappers around slab APIs requiring slab.h to be added manually. 5. The script was run on all .h files but without automatically editing them as sprinkling gfp.h and slab.h inclusions around .h files could easily lead to inclusion dependency hell. Most gfp.h inclusion directives were ignored as stuff from gfp.h was usually wildly available and often used in preprocessor macros. Each slab.h inclusion directive was examined and added manually as necessary. 6. percpu.h was updated not to include slab.h. 7. Build test were done on the following configurations and failures were fixed. CONFIG_GCOV_KERNEL was turned off for all tests (as my distributed build env didn't work with gcov compiles) and a few more options had to be turned off depending on archs to make things build (like ipr on powerpc/64 which failed due to missing writeq). * x86 and x86_64 UP and SMP allmodconfig and a custom test config. * powerpc and powerpc64 SMP allmodconfig * sparc and sparc64 SMP allmodconfig * ia64 SMP allmodconfig * s390 SMP allmodconfig * alpha SMP allmodconfig * um on x86_64 SMP allmodconfig 8. percpu.h modifications were reverted so that it could be applied as a separate patch and serve as bisection point. Given the fact that I had only a couple of failures from tests on step 6, I'm fairly confident about the coverage of this conversion patch. If there is a breakage, it's likely to be something in one of the arch headers which should be easily discoverable easily on most builds of the specific arch. Signed-off-by: Tejun Heo <tj@kernel.org> Guess-its-ok-by: Christoph Lameter <cl@linux-foundation.org> Cc: Ingo Molnar <mingo@redhat.com> Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
2010-03-24 16:04:11 +08:00
#include <linux/slab.h>
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
#include "ubi.h"
/* Maximum length of the 'mtd=' parameter */
#define MTD_PARAM_LEN_MAX 64
/* Maximum number of comma-separated items in the 'mtd=' parameter */
#define MTD_PARAM_MAX_COUNT 4
/* Maximum value for the number of bad PEBs per 1024 PEBs */
#define MAX_MTD_UBI_BEB_LIMIT 768
#ifdef CONFIG_MTD_UBI_MODULE
#define ubi_is_module() 1
#else
#define ubi_is_module() 0
#endif
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
/**
* struct mtd_dev_param - MTD device parameter description data structure.
* @name: MTD character device node path, MTD device name, or MTD device number
* string
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
* @vid_hdr_offs: VID header offset
* @max_beb_per1024: maximum expected number of bad PEBs per 1024 PEBs
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
*/
struct mtd_dev_param {
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
char name[MTD_PARAM_LEN_MAX];
int ubi_num;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
int vid_hdr_offs;
int max_beb_per1024;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
};
/* Numbers of elements set in the @mtd_dev_param array */
static int __initdata mtd_devs;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
/* MTD devices specification parameters */
static struct mtd_dev_param __initdata mtd_dev_param[UBI_MAX_DEVICES];
#ifdef CONFIG_MTD_UBI_FASTMAP
/* UBI module parameter to enable fastmap automatically on non-fastmap images */
static bool fm_autoconvert;
#endif
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
/* Root UBI "class" object (corresponds to '/<sysfs>/class/ubi/') */
struct class *ubi_class;
/* Slab cache for wear-leveling entries */
struct kmem_cache *ubi_wl_entry_slab;
/* UBI control character device */
static struct miscdevice ubi_ctrl_cdev = {
.minor = MISC_DYNAMIC_MINOR,
.name = "ubi_ctrl",
.fops = &ubi_ctrl_cdev_operations,
};
/* All UBI devices in system */
static struct ubi_device *ubi_devices[UBI_MAX_DEVICES];
/* Serializes UBI devices creations and removals */
DEFINE_MUTEX(ubi_devices_mutex);
/* Protects @ubi_devices and @ubi->ref_count */
static DEFINE_SPINLOCK(ubi_devices_lock);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
/* "Show" method for files in '/<sysfs>/class/ubi/' */
static ssize_t ubi_version_show(struct class *class,
struct class_attribute *attr, char *buf)
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
{
return sprintf(buf, "%d\n", UBI_VERSION);
}
/* UBI version attribute ('/<sysfs>/class/ubi/version') */
static struct class_attribute ubi_version =
__ATTR(version, S_IRUGO, ubi_version_show, NULL);
static ssize_t dev_attribute_show(struct device *dev,
struct device_attribute *attr, char *buf);
/* UBI device attributes (correspond to files in '/<sysfs>/class/ubi/ubiX') */
static struct device_attribute dev_eraseblock_size =
__ATTR(eraseblock_size, S_IRUGO, dev_attribute_show, NULL);
static struct device_attribute dev_avail_eraseblocks =
__ATTR(avail_eraseblocks, S_IRUGO, dev_attribute_show, NULL);
static struct device_attribute dev_total_eraseblocks =
__ATTR(total_eraseblocks, S_IRUGO, dev_attribute_show, NULL);
static struct device_attribute dev_volumes_count =
__ATTR(volumes_count, S_IRUGO, dev_attribute_show, NULL);
static struct device_attribute dev_max_ec =
__ATTR(max_ec, S_IRUGO, dev_attribute_show, NULL);
static struct device_attribute dev_reserved_for_bad =
__ATTR(reserved_for_bad, S_IRUGO, dev_attribute_show, NULL);
static struct device_attribute dev_bad_peb_count =
__ATTR(bad_peb_count, S_IRUGO, dev_attribute_show, NULL);
static struct device_attribute dev_max_vol_count =
__ATTR(max_vol_count, S_IRUGO, dev_attribute_show, NULL);
static struct device_attribute dev_min_io_size =
__ATTR(min_io_size, S_IRUGO, dev_attribute_show, NULL);
static struct device_attribute dev_bgt_enabled =
__ATTR(bgt_enabled, S_IRUGO, dev_attribute_show, NULL);
static struct device_attribute dev_mtd_num =
__ATTR(mtd_num, S_IRUGO, dev_attribute_show, NULL);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
/**
* ubi_volume_notify - send a volume change notification.
* @ubi: UBI device description object
* @vol: volume description object of the changed volume
* @ntype: notification type to send (%UBI_VOLUME_ADDED, etc)
*
* This is a helper function which notifies all subscribers about a volume
* change event (creation, removal, re-sizing, re-naming, updating). Returns
* zero in case of success and a negative error code in case of failure.
*/
int ubi_volume_notify(struct ubi_device *ubi, struct ubi_volume *vol, int ntype)
{
struct ubi_notification nt;
ubi_do_get_device_info(ubi, &nt.di);
ubi_do_get_volume_info(ubi, vol, &nt.vi);
#ifdef CONFIG_MTD_UBI_FASTMAP
switch (ntype) {
case UBI_VOLUME_ADDED:
case UBI_VOLUME_REMOVED:
case UBI_VOLUME_RESIZED:
case UBI_VOLUME_RENAMED:
if (ubi_update_fastmap(ubi)) {
ubi_err("Unable to update fastmap!");
ubi_ro_mode(ubi);
}
}
#endif
return blocking_notifier_call_chain(&ubi_notifiers, ntype, &nt);
}
/**
* ubi_notify_all - send a notification to all volumes.
* @ubi: UBI device description object
* @ntype: notification type to send (%UBI_VOLUME_ADDED, etc)
* @nb: the notifier to call
*
* This function walks all volumes of UBI device @ubi and sends the @ntype
* notification for each volume. If @nb is %NULL, then all registered notifiers
* are called, otherwise only the @nb notifier is called. Returns the number of
* sent notifications.
*/
int ubi_notify_all(struct ubi_device *ubi, int ntype, struct notifier_block *nb)
{
struct ubi_notification nt;
int i, count = 0;
ubi_do_get_device_info(ubi, &nt.di);
mutex_lock(&ubi->device_mutex);
for (i = 0; i < ubi->vtbl_slots; i++) {
/*
* Since the @ubi->device is locked, and we are not going to
* change @ubi->volumes, we do not have to lock
* @ubi->volumes_lock.
*/
if (!ubi->volumes[i])
continue;
ubi_do_get_volume_info(ubi, ubi->volumes[i], &nt.vi);
if (nb)
nb->notifier_call(nb, ntype, &nt);
else
blocking_notifier_call_chain(&ubi_notifiers, ntype,
&nt);
count += 1;
}
mutex_unlock(&ubi->device_mutex);
return count;
}
/**
* ubi_enumerate_volumes - send "add" notification for all existing volumes.
* @nb: the notifier to call
*
* This function walks all UBI devices and volumes and sends the
* %UBI_VOLUME_ADDED notification for each volume. If @nb is %NULL, then all
* registered notifiers are called, otherwise only the @nb notifier is called.
* Returns the number of sent notifications.
*/
int ubi_enumerate_volumes(struct notifier_block *nb)
{
int i, count = 0;
/*
* Since the @ubi_devices_mutex is locked, and we are not going to
* change @ubi_devices, we do not have to lock @ubi_devices_lock.
*/
for (i = 0; i < UBI_MAX_DEVICES; i++) {
struct ubi_device *ubi = ubi_devices[i];
if (!ubi)
continue;
count += ubi_notify_all(ubi, UBI_VOLUME_ADDED, nb);
}
return count;
}
/**
* ubi_get_device - get UBI device.
* @ubi_num: UBI device number
*
* This function returns UBI device description object for UBI device number
* @ubi_num, or %NULL if the device does not exist. This function increases the
* device reference count to prevent removal of the device. In other words, the
* device cannot be removed if its reference count is not zero.
*/
struct ubi_device *ubi_get_device(int ubi_num)
{
struct ubi_device *ubi;
spin_lock(&ubi_devices_lock);
ubi = ubi_devices[ubi_num];
if (ubi) {
ubi_assert(ubi->ref_count >= 0);
ubi->ref_count += 1;
get_device(&ubi->dev);
}
spin_unlock(&ubi_devices_lock);
return ubi;
}
/**
* ubi_put_device - drop an UBI device reference.
* @ubi: UBI device description object
*/
void ubi_put_device(struct ubi_device *ubi)
{
spin_lock(&ubi_devices_lock);
ubi->ref_count -= 1;
put_device(&ubi->dev);
spin_unlock(&ubi_devices_lock);
}
/**
* ubi_get_by_major - get UBI device by character device major number.
* @major: major number
*
* This function is similar to 'ubi_get_device()', but it searches the device
* by its major number.
*/
struct ubi_device *ubi_get_by_major(int major)
{
int i;
struct ubi_device *ubi;
spin_lock(&ubi_devices_lock);
for (i = 0; i < UBI_MAX_DEVICES; i++) {
ubi = ubi_devices[i];
if (ubi && MAJOR(ubi->cdev.dev) == major) {
ubi_assert(ubi->ref_count >= 0);
ubi->ref_count += 1;
get_device(&ubi->dev);
spin_unlock(&ubi_devices_lock);
return ubi;
}
}
spin_unlock(&ubi_devices_lock);
return NULL;
}
/**
* ubi_major2num - get UBI device number by character device major number.
* @major: major number
*
* This function searches UBI device number object by its major number. If UBI
* device was not found, this function returns -ENODEV, otherwise the UBI device
* number is returned.
*/
int ubi_major2num(int major)
{
int i, ubi_num = -ENODEV;
spin_lock(&ubi_devices_lock);
for (i = 0; i < UBI_MAX_DEVICES; i++) {
struct ubi_device *ubi = ubi_devices[i];
if (ubi && MAJOR(ubi->cdev.dev) == major) {
ubi_num = ubi->ubi_num;
break;
}
}
spin_unlock(&ubi_devices_lock);
return ubi_num;
}
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
/* "Show" method for files in '/<sysfs>/class/ubi/ubiX/' */
static ssize_t dev_attribute_show(struct device *dev,
struct device_attribute *attr, char *buf)
{
ssize_t ret;
struct ubi_device *ubi;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
/*
* The below code looks weird, but it actually makes sense. We get the
* UBI device reference from the contained 'struct ubi_device'. But it
* is unclear if the device was removed or not yet. Indeed, if the
* device was removed before we increased its reference count,
* 'ubi_get_device()' will return -ENODEV and we fail.
*
* Remember, 'struct ubi_device' is freed in the release function, so
* we still can use 'ubi->ubi_num'.
*/
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
ubi = container_of(dev, struct ubi_device, dev);
ubi = ubi_get_device(ubi->ubi_num);
if (!ubi)
return -ENODEV;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
if (attr == &dev_eraseblock_size)
ret = sprintf(buf, "%d\n", ubi->leb_size);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
else if (attr == &dev_avail_eraseblocks)
ret = sprintf(buf, "%d\n", ubi->avail_pebs);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
else if (attr == &dev_total_eraseblocks)
ret = sprintf(buf, "%d\n", ubi->good_peb_count);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
else if (attr == &dev_volumes_count)
ret = sprintf(buf, "%d\n", ubi->vol_count - UBI_INT_VOL_COUNT);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
else if (attr == &dev_max_ec)
ret = sprintf(buf, "%d\n", ubi->max_ec);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
else if (attr == &dev_reserved_for_bad)
ret = sprintf(buf, "%d\n", ubi->beb_rsvd_pebs);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
else if (attr == &dev_bad_peb_count)
ret = sprintf(buf, "%d\n", ubi->bad_peb_count);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
else if (attr == &dev_max_vol_count)
ret = sprintf(buf, "%d\n", ubi->vtbl_slots);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
else if (attr == &dev_min_io_size)
ret = sprintf(buf, "%d\n", ubi->min_io_size);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
else if (attr == &dev_bgt_enabled)
ret = sprintf(buf, "%d\n", ubi->thread_enabled);
else if (attr == &dev_mtd_num)
ret = sprintf(buf, "%d\n", ubi->mtd->index);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
else
ret = -EINVAL;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
ubi_put_device(ubi);
return ret;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
}
static void dev_release(struct device *dev)
{
struct ubi_device *ubi = container_of(dev, struct ubi_device, dev);
kfree(ubi);
}
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
/**
* ubi_sysfs_init - initialize sysfs for an UBI device.
* @ubi: UBI device description object
* @ref: set to %1 on exit in case of failure if a reference to @ubi->dev was
* taken
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
*
* This function returns zero in case of success and a negative error code in
* case of failure.
*/
static int ubi_sysfs_init(struct ubi_device *ubi, int *ref)
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
{
int err;
ubi->dev.release = dev_release;
ubi->dev.devt = ubi->cdev.dev;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
ubi->dev.class = ubi_class;
dev_set_name(&ubi->dev, UBI_NAME_STR"%d", ubi->ubi_num);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
err = device_register(&ubi->dev);
if (err)
return err;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
*ref = 1;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
err = device_create_file(&ubi->dev, &dev_eraseblock_size);
if (err)
return err;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
err = device_create_file(&ubi->dev, &dev_avail_eraseblocks);
if (err)
return err;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
err = device_create_file(&ubi->dev, &dev_total_eraseblocks);
if (err)
return err;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
err = device_create_file(&ubi->dev, &dev_volumes_count);
if (err)
return err;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
err = device_create_file(&ubi->dev, &dev_max_ec);
if (err)
return err;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
err = device_create_file(&ubi->dev, &dev_reserved_for_bad);
if (err)
return err;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
err = device_create_file(&ubi->dev, &dev_bad_peb_count);
if (err)
return err;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
err = device_create_file(&ubi->dev, &dev_max_vol_count);
if (err)
return err;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
err = device_create_file(&ubi->dev, &dev_min_io_size);
if (err)
return err;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
err = device_create_file(&ubi->dev, &dev_bgt_enabled);
if (err)
return err;
err = device_create_file(&ubi->dev, &dev_mtd_num);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
return err;
}
/**
* ubi_sysfs_close - close sysfs for an UBI device.
* @ubi: UBI device description object
*/
static void ubi_sysfs_close(struct ubi_device *ubi)
{
device_remove_file(&ubi->dev, &dev_mtd_num);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
device_remove_file(&ubi->dev, &dev_bgt_enabled);
device_remove_file(&ubi->dev, &dev_min_io_size);
device_remove_file(&ubi->dev, &dev_max_vol_count);
device_remove_file(&ubi->dev, &dev_bad_peb_count);
device_remove_file(&ubi->dev, &dev_reserved_for_bad);
device_remove_file(&ubi->dev, &dev_max_ec);
device_remove_file(&ubi->dev, &dev_volumes_count);
device_remove_file(&ubi->dev, &dev_total_eraseblocks);
device_remove_file(&ubi->dev, &dev_avail_eraseblocks);
device_remove_file(&ubi->dev, &dev_eraseblock_size);
device_unregister(&ubi->dev);
}
/**
* kill_volumes - destroy all user volumes.
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
* @ubi: UBI device description object
*/
static void kill_volumes(struct ubi_device *ubi)
{
int i;
for (i = 0; i < ubi->vtbl_slots; i++)
if (ubi->volumes[i])
ubi_free_volume(ubi, ubi->volumes[i]);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
}
/**
* uif_init - initialize user interfaces for an UBI device.
* @ubi: UBI device description object
* @ref: set to %1 on exit in case of failure if a reference to @ubi->dev was
* taken, otherwise set to %0
*
* This function initializes various user interfaces for an UBI device. If the
* initialization fails at an early stage, this function frees all the
* resources it allocated, returns an error, and @ref is set to %0. However,
* if the initialization fails after the UBI device was registered in the
* driver core subsystem, this function takes a reference to @ubi->dev, because
* otherwise the release function ('dev_release()') would free whole @ubi
* object. The @ref argument is set to %1 in this case. The caller has to put
* this reference.
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
*
* This function returns zero in case of success and a negative error code in
* case of failure.
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
*/
static int uif_init(struct ubi_device *ubi, int *ref)
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
{
int i, err;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
dev_t dev;
*ref = 0;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
sprintf(ubi->ubi_name, UBI_NAME_STR "%d", ubi->ubi_num);
/*
* Major numbers for the UBI character devices are allocated
* dynamically. Major numbers of volume character devices are
* equivalent to ones of the corresponding UBI character device. Minor
* numbers of UBI character devices are 0, while minor numbers of
* volume character devices start from 1. Thus, we allocate one major
* number and ubi->vtbl_slots + 1 minor numbers.
*/
err = alloc_chrdev_region(&dev, 0, ubi->vtbl_slots + 1, ubi->ubi_name);
if (err) {
ubi_err("cannot register UBI character devices");
return err;
}
ubi_assert(MINOR(dev) == 0);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
cdev_init(&ubi->cdev, &ubi_cdev_operations);
dbg_gen("%s major is %u", ubi->ubi_name, MAJOR(dev));
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
ubi->cdev.owner = THIS_MODULE;
err = cdev_add(&ubi->cdev, dev, 1);
if (err) {
ubi_err("cannot add character device");
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
goto out_unreg;
}
err = ubi_sysfs_init(ubi, ref);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
if (err)
goto out_sysfs;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
for (i = 0; i < ubi->vtbl_slots; i++)
if (ubi->volumes[i]) {
err = ubi_add_volume(ubi, ubi->volumes[i]);
if (err) {
ubi_err("cannot add volume %d", i);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
goto out_volumes;
}
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
}
return 0;
out_volumes:
kill_volumes(ubi);
out_sysfs:
if (*ref)
get_device(&ubi->dev);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
ubi_sysfs_close(ubi);
cdev_del(&ubi->cdev);
out_unreg:
unregister_chrdev_region(ubi->cdev.dev, ubi->vtbl_slots + 1);
ubi_err("cannot initialize UBI %s, error %d", ubi->ubi_name, err);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
return err;
}
/**
* uif_close - close user interfaces for an UBI device.
* @ubi: UBI device description object
*
* Note, since this function un-registers UBI volume device objects (@vol->dev),
* the memory allocated voe the volumes is freed as well (in the release
* function).
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
*/
static void uif_close(struct ubi_device *ubi)
{
kill_volumes(ubi);
ubi_sysfs_close(ubi);
cdev_del(&ubi->cdev);
unregister_chrdev_region(ubi->cdev.dev, ubi->vtbl_slots + 1);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
}
/**
* ubi_free_internal_volumes - free internal volumes.
* @ubi: UBI device description object
*/
void ubi_free_internal_volumes(struct ubi_device *ubi)
{
int i;
for (i = ubi->vtbl_slots;
i < ubi->vtbl_slots + UBI_INT_VOL_COUNT; i++) {
kfree(ubi->volumes[i]->eba_tbl);
kfree(ubi->volumes[i]);
}
}
static int get_bad_peb_limit(const struct ubi_device *ubi, int max_beb_per1024)
{
int limit, device_pebs;
uint64_t device_size;
if (!max_beb_per1024)
return 0;
/*
* Here we are using size of the entire flash chip and
* not just the MTD partition size because the maximum
* number of bad eraseblocks is a percentage of the
* whole device and bad eraseblocks are not fairly
* distributed over the flash chip. So the worst case
* is that all the bad eraseblocks of the chip are in
* the MTD partition we are attaching (ubi->mtd).
*/
device_size = mtd_get_device_size(ubi->mtd);
device_pebs = mtd_div_by_eb(device_size, ubi->mtd);
limit = mult_frac(device_pebs, max_beb_per1024, 1024);
/* Round it up */
if (mult_frac(limit, 1024, max_beb_per1024) < device_pebs)
limit += 1;
return limit;
}
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
/**
* io_init - initialize I/O sub-system for a given UBI device.
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
* @ubi: UBI device description object
* @max_beb_per1024: maximum expected number of bad PEB per 1024 PEBs
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
*
* If @ubi->vid_hdr_offset or @ubi->leb_start is zero, default offsets are
* assumed:
* o EC header is always at offset zero - this cannot be changed;
* o VID header starts just after the EC header at the closest address
* aligned to @io->hdrs_min_io_size;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
* o data starts just after the VID header at the closest address aligned to
* @io->min_io_size
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
*
* This function returns zero in case of success and a negative error code in
* case of failure.
*/
static int io_init(struct ubi_device *ubi, int max_beb_per1024)
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
{
dbg_gen("sizeof(struct ubi_ainf_peb) %zu", sizeof(struct ubi_ainf_peb));
dbg_gen("sizeof(struct ubi_wl_entry) %zu", sizeof(struct ubi_wl_entry));
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
if (ubi->mtd->numeraseregions != 0) {
/*
* Some flashes have several erase regions. Different regions
* may have different eraseblock size and other
* characteristics. It looks like mostly multi-region flashes
* have one "main" region and one or more small regions to
* store boot loader code or boot parameters or whatever. I
* guess we should just pick the largest region. But this is
* not implemented.
*/
ubi_err("multiple regions, not implemented");
return -EINVAL;
}
if (ubi->vid_hdr_offset < 0)
return -EINVAL;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
/*
* Note, in this implementation we support MTD devices with 0x7FFFFFFF
* physical eraseblocks maximum.
*/
ubi->peb_size = ubi->mtd->erasesize;
ubi->peb_count = mtd_div_by_eb(ubi->mtd->size, ubi->mtd);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
ubi->flash_size = ubi->mtd->size;
if (mtd_can_have_bb(ubi->mtd)) {
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
ubi->bad_allowed = 1;
ubi->bad_peb_limit = get_bad_peb_limit(ubi, max_beb_per1024);
}
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
if (ubi->mtd->type == MTD_NORFLASH) {
ubi_assert(ubi->mtd->writesize == 1);
ubi->nor_flash = 1;
}
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
ubi->min_io_size = ubi->mtd->writesize;
ubi->hdrs_min_io_size = ubi->mtd->writesize >> ubi->mtd->subpage_sft;
/*
* Make sure minimal I/O unit is power of 2. Note, there is no
* fundamental reason for this assumption. It is just an optimization
* which allows us to avoid costly division operations.
*/
if (!is_power_of_2(ubi->min_io_size)) {
ubi_err("min. I/O unit (%d) is not power of 2",
ubi->min_io_size);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
return -EINVAL;
}
ubi_assert(ubi->hdrs_min_io_size > 0);
ubi_assert(ubi->hdrs_min_io_size <= ubi->min_io_size);
ubi_assert(ubi->min_io_size % ubi->hdrs_min_io_size == 0);
ubi->max_write_size = ubi->mtd->writebufsize;
/*
* Maximum write size has to be greater or equivalent to min. I/O
* size, and be multiple of min. I/O size.
*/
if (ubi->max_write_size < ubi->min_io_size ||
ubi->max_write_size % ubi->min_io_size ||
!is_power_of_2(ubi->max_write_size)) {
ubi_err("bad write buffer size %d for %d min. I/O unit",
ubi->max_write_size, ubi->min_io_size);
return -EINVAL;
}
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
/* Calculate default aligned sizes of EC and VID headers */
ubi->ec_hdr_alsize = ALIGN(UBI_EC_HDR_SIZE, ubi->hdrs_min_io_size);
ubi->vid_hdr_alsize = ALIGN(UBI_VID_HDR_SIZE, ubi->hdrs_min_io_size);
dbg_gen("min_io_size %d", ubi->min_io_size);
dbg_gen("max_write_size %d", ubi->max_write_size);
dbg_gen("hdrs_min_io_size %d", ubi->hdrs_min_io_size);
dbg_gen("ec_hdr_alsize %d", ubi->ec_hdr_alsize);
dbg_gen("vid_hdr_alsize %d", ubi->vid_hdr_alsize);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
if (ubi->vid_hdr_offset == 0)
/* Default offset */
ubi->vid_hdr_offset = ubi->vid_hdr_aloffset =
ubi->ec_hdr_alsize;
else {
ubi->vid_hdr_aloffset = ubi->vid_hdr_offset &
~(ubi->hdrs_min_io_size - 1);
ubi->vid_hdr_shift = ubi->vid_hdr_offset -
ubi->vid_hdr_aloffset;
}
/* Similar for the data offset */
ubi->leb_start = ubi->vid_hdr_offset + UBI_VID_HDR_SIZE;
ubi->leb_start = ALIGN(ubi->leb_start, ubi->min_io_size);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
dbg_gen("vid_hdr_offset %d", ubi->vid_hdr_offset);
dbg_gen("vid_hdr_aloffset %d", ubi->vid_hdr_aloffset);
dbg_gen("vid_hdr_shift %d", ubi->vid_hdr_shift);
dbg_gen("leb_start %d", ubi->leb_start);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
/* The shift must be aligned to 32-bit boundary */
if (ubi->vid_hdr_shift % 4) {
ubi_err("unaligned VID header shift %d",
ubi->vid_hdr_shift);
return -EINVAL;
}
/* Check sanity */
if (ubi->vid_hdr_offset < UBI_EC_HDR_SIZE ||
ubi->leb_start < ubi->vid_hdr_offset + UBI_VID_HDR_SIZE ||
ubi->leb_start > ubi->peb_size - UBI_VID_HDR_SIZE ||
ubi->leb_start & (ubi->min_io_size - 1)) {
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
ubi_err("bad VID header (%d) or data offsets (%d)",
ubi->vid_hdr_offset, ubi->leb_start);
return -EINVAL;
}
/*
* Set maximum amount of physical erroneous eraseblocks to be 10%.
* Erroneous PEB are those which have read errors.
*/
ubi->max_erroneous = ubi->peb_count / 10;
if (ubi->max_erroneous < 16)
ubi->max_erroneous = 16;
dbg_gen("max_erroneous %d", ubi->max_erroneous);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
/*
* It may happen that EC and VID headers are situated in one minimal
* I/O unit. In this case we can only accept this UBI image in
* read-only mode.
*/
if (ubi->vid_hdr_offset + UBI_VID_HDR_SIZE <= ubi->hdrs_min_io_size) {
ubi_warn("EC and VID headers are in the same minimal I/O unit, switch to read-only mode");
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
ubi->ro_mode = 1;
}
ubi->leb_size = ubi->peb_size - ubi->leb_start;
if (!(ubi->mtd->flags & MTD_WRITEABLE)) {
ubi_msg("MTD device %d is write-protected, attach in read-only mode",
ubi->mtd->index);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
ubi->ro_mode = 1;
}
/*
* Note, ideally, we have to initialize @ubi->bad_peb_count here. But
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
* unfortunately, MTD does not provide this information. We should loop
* over all physical eraseblocks and invoke mtd->block_is_bad() for
* each physical eraseblock. So, we leave @ubi->bad_peb_count
* uninitialized so far.
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
*/
return 0;
}
/**
* autoresize - re-size the volume which has the "auto-resize" flag set.
* @ubi: UBI device description object
* @vol_id: ID of the volume to re-size
*
* This function re-sizes the volume marked by the %UBI_VTBL_AUTORESIZE_FLG in
* the volume table to the largest possible size. See comments in ubi-header.h
* for more description of the flag. Returns zero in case of success and a
* negative error code in case of failure.
*/
static int autoresize(struct ubi_device *ubi, int vol_id)
{
struct ubi_volume_desc desc;
struct ubi_volume *vol = ubi->volumes[vol_id];
int err, old_reserved_pebs = vol->reserved_pebs;
if (ubi->ro_mode) {
ubi_warn("skip auto-resize because of R/O mode");
return 0;
}
/*
* Clear the auto-resize flag in the volume in-memory copy of the
* volume table, and 'ubi_resize_volume()' will propagate this change
* to the flash.
*/
ubi->vtbl[vol_id].flags &= ~UBI_VTBL_AUTORESIZE_FLG;
if (ubi->avail_pebs == 0) {
struct ubi_vtbl_record vtbl_rec;
/*
* No available PEBs to re-size the volume, clear the flag on
* flash and exit.
*/
vtbl_rec = ubi->vtbl[vol_id];
err = ubi_change_vtbl_record(ubi, vol_id, &vtbl_rec);
if (err)
ubi_err("cannot clean auto-resize flag for volume %d",
vol_id);
} else {
desc.vol = vol;
err = ubi_resize_volume(&desc,
old_reserved_pebs + ubi->avail_pebs);
if (err)
ubi_err("cannot auto-resize volume %d", vol_id);
}
if (err)
return err;
ubi_msg("volume %d (\"%s\") re-sized from %d to %d LEBs", vol_id,
vol->name, old_reserved_pebs, vol->reserved_pebs);
return 0;
}
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
/**
* ubi_attach_mtd_dev - attach an MTD device.
* @mtd: MTD device description object
* @ubi_num: number to assign to the new UBI device
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
* @vid_hdr_offset: VID header offset
* @max_beb_per1024: maximum expected number of bad PEB per 1024 PEBs
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
*
* This function attaches MTD device @mtd_dev to UBI and assign @ubi_num number
* to the newly created UBI device, unless @ubi_num is %UBI_DEV_NUM_AUTO, in
* which case this function finds a vacant device number and assigns it
* automatically. Returns the new UBI device number in case of success and a
* negative error code in case of failure.
*
* Note, the invocations of this function has to be serialized by the
* @ubi_devices_mutex.
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
*/
int ubi_attach_mtd_dev(struct mtd_info *mtd, int ubi_num,
int vid_hdr_offset, int max_beb_per1024)
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
{
struct ubi_device *ubi;
int i, err, ref = 0;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
if (max_beb_per1024 < 0 || max_beb_per1024 > MAX_MTD_UBI_BEB_LIMIT)
return -EINVAL;
if (!max_beb_per1024)
max_beb_per1024 = CONFIG_MTD_UBI_BEB_LIMIT;
/*
* Check if we already have the same MTD device attached.
*
* Note, this function assumes that UBI devices creations and deletions
* are serialized, so it does not take the &ubi_devices_lock.
*/
for (i = 0; i < UBI_MAX_DEVICES; i++) {
ubi = ubi_devices[i];
if (ubi && mtd->index == ubi->mtd->index) {
ubi_err("mtd%d is already attached to ubi%d",
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
mtd->index, i);
return -EEXIST;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
}
}
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
/*
* Make sure this MTD device is not emulated on top of an UBI volume
* already. Well, generally this recursion works fine, but there are
* different problems like the UBI module takes a reference to itself
* by attaching (and thus, opening) the emulated MTD device. This
* results in inability to unload the module. And in general it makes
* no sense to attach emulated MTD devices, so we prohibit this.
*/
if (mtd->type == MTD_UBIVOLUME) {
ubi_err("refuse attaching mtd%d - it is already emulated on top of UBI",
mtd->index);
return -EINVAL;
}
if (ubi_num == UBI_DEV_NUM_AUTO) {
/* Search for an empty slot in the @ubi_devices array */
for (ubi_num = 0; ubi_num < UBI_MAX_DEVICES; ubi_num++)
if (!ubi_devices[ubi_num])
break;
if (ubi_num == UBI_MAX_DEVICES) {
ubi_err("only %d UBI devices may be created",
UBI_MAX_DEVICES);
return -ENFILE;
}
} else {
if (ubi_num >= UBI_MAX_DEVICES)
return -EINVAL;
/* Make sure ubi_num is not busy */
if (ubi_devices[ubi_num]) {
ubi_err("ubi%d already exists", ubi_num);
return -EEXIST;
}
}
ubi = kzalloc(sizeof(struct ubi_device), GFP_KERNEL);
if (!ubi)
return -ENOMEM;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
ubi->mtd = mtd;
ubi->ubi_num = ubi_num;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
ubi->vid_hdr_offset = vid_hdr_offset;
ubi->autoresize_vol_id = -1;
#ifdef CONFIG_MTD_UBI_FASTMAP
ubi->fm_pool.used = ubi->fm_pool.size = 0;
ubi->fm_wl_pool.used = ubi->fm_wl_pool.size = 0;
/*
* fm_pool.max_size is 5% of the total number of PEBs but it's also
* between UBI_FM_MAX_POOL_SIZE and UBI_FM_MIN_POOL_SIZE.
*/
ubi->fm_pool.max_size = min(((int)mtd_div_by_eb(ubi->mtd->size,
ubi->mtd) / 100) * 5, UBI_FM_MAX_POOL_SIZE);
if (ubi->fm_pool.max_size < UBI_FM_MIN_POOL_SIZE)
ubi->fm_pool.max_size = UBI_FM_MIN_POOL_SIZE;
ubi->fm_wl_pool.max_size = UBI_FM_WL_POOL_SIZE;
ubi->fm_disabled = !fm_autoconvert;
if (!ubi->fm_disabled && (int)mtd_div_by_eb(ubi->mtd->size, ubi->mtd)
<= UBI_FM_MAX_START) {
ubi_err("More than %i PEBs are needed for fastmap, sorry.",
UBI_FM_MAX_START);
ubi->fm_disabled = 1;
}
ubi_msg("default fastmap pool size: %d", ubi->fm_pool.max_size);
ubi_msg("default fastmap WL pool size: %d", ubi->fm_wl_pool.max_size);
#else
ubi->fm_disabled = 1;
#endif
mutex_init(&ubi->buf_mutex);
mutex_init(&ubi->ckvol_mutex);
mutex_init(&ubi->device_mutex);
spin_lock_init(&ubi->volumes_lock);
mutex_init(&ubi->fm_mutex);
init_rwsem(&ubi->fm_sem);
ubi_msg("attaching mtd%d to ubi%d", mtd->index, ubi_num);
err = io_init(ubi, max_beb_per1024);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
if (err)
goto out_free;
err = -ENOMEM;
ubi->peb_buf = vmalloc(ubi->peb_size);
if (!ubi->peb_buf)
goto out_free;
#ifdef CONFIG_MTD_UBI_FASTMAP
ubi->fm_size = ubi_calc_fm_size(ubi);
ubi->fm_buf = vzalloc(ubi->fm_size);
if (!ubi->fm_buf)
goto out_free;
#endif
err = ubi_attach(ubi, 0);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
if (err) {
ubi_err("failed to attach mtd%d, error %d", mtd->index, err);
goto out_free;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
}
if (ubi->autoresize_vol_id != -1) {
err = autoresize(ubi, ubi->autoresize_vol_id);
if (err)
goto out_detach;
}
err = uif_init(ubi, &ref);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
if (err)
goto out_detach;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
err = ubi_debugfs_init_dev(ubi);
if (err)
goto out_uif;
ubi->bgt_thread = kthread_create(ubi_thread, ubi, "%s", ubi->bgt_name);
if (IS_ERR(ubi->bgt_thread)) {
err = PTR_ERR(ubi->bgt_thread);
ubi_err("cannot spawn \"%s\", error %d", ubi->bgt_name,
err);
goto out_debugfs;
}
ubi_msg("attached mtd%d (name \"%s\", size %llu MiB) to ubi%d",
mtd->index, mtd->name, ubi->flash_size >> 20, ubi_num);
ubi_msg("PEB size: %d bytes (%d KiB), LEB size: %d bytes",
ubi->peb_size, ubi->peb_size >> 10, ubi->leb_size);
ubi_msg("min./max. I/O unit sizes: %d/%d, sub-page size %d",
ubi->min_io_size, ubi->max_write_size, ubi->hdrs_min_io_size);
ubi_msg("VID header offset: %d (aligned %d), data offset: %d",
ubi->vid_hdr_offset, ubi->vid_hdr_aloffset, ubi->leb_start);
ubi_msg("good PEBs: %d, bad PEBs: %d, corrupted PEBs: %d",
ubi->good_peb_count, ubi->bad_peb_count, ubi->corr_peb_count);
ubi_msg("user volume: %d, internal volumes: %d, max. volumes count: %d",
ubi->vol_count - UBI_INT_VOL_COUNT, UBI_INT_VOL_COUNT,
ubi->vtbl_slots);
ubi_msg("max/mean erase counter: %d/%d, WL threshold: %d, image sequence number: %u",
ubi->max_ec, ubi->mean_ec, CONFIG_MTD_UBI_WL_THRESHOLD,
ubi->image_seq);
ubi_msg("available PEBs: %d, total reserved PEBs: %d, PEBs reserved for bad PEB handling: %d",
ubi->avail_pebs, ubi->rsvd_pebs, ubi->beb_rsvd_pebs);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
/*
* The below lock makes sure we do not race with 'ubi_thread()' which
* checks @ubi->thread_enabled. Otherwise we may fail to wake it up.
*/
spin_lock(&ubi->wl_lock);
ubi->thread_enabled = 1;
wake_up_process(ubi->bgt_thread);
spin_unlock(&ubi->wl_lock);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
ubi_devices[ubi_num] = ubi;
ubi_notify_all(ubi, UBI_VOLUME_ADDED, NULL);
return ubi_num;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
out_debugfs:
ubi_debugfs_exit_dev(ubi);
out_uif:
get_device(&ubi->dev);
ubi_assert(ref);
uif_close(ubi);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
out_detach:
ubi_wl_close(ubi);
ubi_free_internal_volumes(ubi);
vfree(ubi->vtbl);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
out_free:
vfree(ubi->peb_buf);
vfree(ubi->fm_buf);
if (ref)
put_device(&ubi->dev);
else
kfree(ubi);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
return err;
}
/**
* ubi_detach_mtd_dev - detach an MTD device.
* @ubi_num: UBI device number to detach from
* @anyway: detach MTD even if device reference count is not zero
*
* This function destroys an UBI device number @ubi_num and detaches the
* underlying MTD device. Returns zero in case of success and %-EBUSY if the
* UBI device is busy and cannot be destroyed, and %-EINVAL if it does not
* exist.
*
* Note, the invocations of this function has to be serialized by the
* @ubi_devices_mutex.
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
*/
int ubi_detach_mtd_dev(int ubi_num, int anyway)
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
{
struct ubi_device *ubi;
if (ubi_num < 0 || ubi_num >= UBI_MAX_DEVICES)
return -EINVAL;
ubi = ubi_get_device(ubi_num);
if (!ubi)
return -EINVAL;
spin_lock(&ubi_devices_lock);
put_device(&ubi->dev);
ubi->ref_count -= 1;
if (ubi->ref_count) {
if (!anyway) {
spin_unlock(&ubi_devices_lock);
return -EBUSY;
}
/* This may only happen if there is a bug */
ubi_err("%s reference count %d, destroy anyway",
ubi->ubi_name, ubi->ref_count);
}
ubi_devices[ubi_num] = NULL;
spin_unlock(&ubi_devices_lock);
ubi_assert(ubi_num == ubi->ubi_num);
ubi_notify_all(ubi, UBI_VOLUME_REMOVED, NULL);
ubi_msg("detaching mtd%d from ubi%d", ubi->mtd->index, ubi_num);
#ifdef CONFIG_MTD_UBI_FASTMAP
/* If we don't write a new fastmap at detach time we lose all
* EC updates that have been made since the last written fastmap. */
ubi_update_fastmap(ubi);
#endif
/*
* Before freeing anything, we have to stop the background thread to
* prevent it from doing anything on this device while we are freeing.
*/
if (ubi->bgt_thread)
kthread_stop(ubi->bgt_thread);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
/*
* Get a reference to the device in order to prevent 'dev_release()'
* from freeing the @ubi object.
*/
get_device(&ubi->dev);
ubi_debugfs_exit_dev(ubi);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
uif_close(ubi);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
ubi_wl_close(ubi);
ubi_free_internal_volumes(ubi);
vfree(ubi->vtbl);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
put_mtd_device(ubi->mtd);
vfree(ubi->peb_buf);
vfree(ubi->fm_buf);
ubi_msg("mtd%d is detached from ubi%d", ubi->mtd->index, ubi->ubi_num);
put_device(&ubi->dev);
return 0;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
}
/**
* open_mtd_by_chdev - open an MTD device by its character device node path.
* @mtd_dev: MTD character device node path
*
* This helper function opens an MTD device by its character node device path.
* Returns MTD device description object in case of success and a negative
* error code in case of failure.
*/
static struct mtd_info * __init open_mtd_by_chdev(const char *mtd_dev)
{
int err, major, minor, mode;
struct path path;
/* Probably this is an MTD character device node path */
err = kern_path(mtd_dev, LOOKUP_FOLLOW, &path);
if (err)
return ERR_PTR(err);
/* MTD device number is defined by the major / minor numbers */
major = imajor(path.dentry->d_inode);
minor = iminor(path.dentry->d_inode);
mode = path.dentry->d_inode->i_mode;
path_put(&path);
if (major != MTD_CHAR_MAJOR || !S_ISCHR(mode))
return ERR_PTR(-EINVAL);
if (minor & 1)
/*
* Just do not think the "/dev/mtdrX" devices support is need,
* so do not support them to avoid doing extra work.
*/
return ERR_PTR(-EINVAL);
return get_mtd_device(NULL, minor / 2);
}
/**
* open_mtd_device - open MTD device by name, character device path, or number.
* @mtd_dev: name, character device node path, or MTD device device number
*
* This function tries to open and MTD device described by @mtd_dev string,
* which is first treated as ASCII MTD device number, and if it is not true, it
* is treated as MTD device name, and if that is also not true, it is treated
* as MTD character device node path. Returns MTD device description object in
* case of success and a negative error code in case of failure.
*/
static struct mtd_info * __init open_mtd_device(const char *mtd_dev)
{
struct mtd_info *mtd;
int mtd_num;
char *endp;
mtd_num = simple_strtoul(mtd_dev, &endp, 0);
if (*endp != '\0' || mtd_dev == endp) {
/*
* This does not look like an ASCII integer, probably this is
* MTD device name.
*/
mtd = get_mtd_device_nm(mtd_dev);
if (IS_ERR(mtd) && PTR_ERR(mtd) == -ENODEV)
/* Probably this is an MTD character device node path */
mtd = open_mtd_by_chdev(mtd_dev);
} else
mtd = get_mtd_device(NULL, mtd_num);
return mtd;
}
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
static int __init ubi_init(void)
{
int err, i, k;
/* Ensure that EC and VID headers have correct size */
BUILD_BUG_ON(sizeof(struct ubi_ec_hdr) != 64);
BUILD_BUG_ON(sizeof(struct ubi_vid_hdr) != 64);
if (mtd_devs > UBI_MAX_DEVICES) {
ubi_err("too many MTD devices, maximum is %d", UBI_MAX_DEVICES);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
return -EINVAL;
}
/* Create base sysfs directory and sysfs files */
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
ubi_class = class_create(THIS_MODULE, UBI_NAME_STR);
if (IS_ERR(ubi_class)) {
err = PTR_ERR(ubi_class);
ubi_err("cannot create UBI class");
goto out;
}
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
err = class_create_file(ubi_class, &ubi_version);
if (err) {
ubi_err("cannot create sysfs file");
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
goto out_class;
}
err = misc_register(&ubi_ctrl_cdev);
if (err) {
ubi_err("cannot register device");
goto out_version;
}
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
ubi_wl_entry_slab = kmem_cache_create("ubi_wl_entry_slab",
sizeof(struct ubi_wl_entry),
0, 0, NULL);
if (!ubi_wl_entry_slab)
goto out_dev_unreg;
err = ubi_debugfs_init();
if (err)
goto out_slab;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
/* Attach MTD devices */
for (i = 0; i < mtd_devs; i++) {
struct mtd_dev_param *p = &mtd_dev_param[i];
struct mtd_info *mtd;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
cond_resched();
mtd = open_mtd_device(p->name);
if (IS_ERR(mtd)) {
err = PTR_ERR(mtd);
ubi_err("cannot open mtd %s, error %d", p->name, err);
/* See comment below re-ubi_is_module(). */
if (ubi_is_module())
goto out_detach;
continue;
}
mutex_lock(&ubi_devices_mutex);
err = ubi_attach_mtd_dev(mtd, p->ubi_num,
p->vid_hdr_offs, p->max_beb_per1024);
mutex_unlock(&ubi_devices_mutex);
if (err < 0) {
ubi_err("cannot attach mtd%d", mtd->index);
put_mtd_device(mtd);
/*
* Originally UBI stopped initializing on any error.
* However, later on it was found out that this
* behavior is not very good when UBI is compiled into
* the kernel and the MTD devices to attach are passed
* through the command line. Indeed, UBI failure
* stopped whole boot sequence.
*
* To fix this, we changed the behavior for the
* non-module case, but preserved the old behavior for
* the module case, just for compatibility. This is a
* little inconsistent, though.
*/
if (ubi_is_module())
goto out_detach;
}
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
}
return 0;
out_detach:
for (k = 0; k < i; k++)
if (ubi_devices[k]) {
mutex_lock(&ubi_devices_mutex);
ubi_detach_mtd_dev(ubi_devices[k]->ubi_num, 1);
mutex_unlock(&ubi_devices_mutex);
}
ubi_debugfs_exit();
out_slab:
kmem_cache_destroy(ubi_wl_entry_slab);
out_dev_unreg:
misc_deregister(&ubi_ctrl_cdev);
out_version:
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
class_remove_file(ubi_class, &ubi_version);
out_class:
class_destroy(ubi_class);
out:
ubi_err("cannot initialize UBI, error %d", err);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
return err;
}
late_initcall(ubi_init);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
static void __exit ubi_exit(void)
{
int i;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
for (i = 0; i < UBI_MAX_DEVICES; i++)
if (ubi_devices[i]) {
mutex_lock(&ubi_devices_mutex);
ubi_detach_mtd_dev(ubi_devices[i]->ubi_num, 1);
mutex_unlock(&ubi_devices_mutex);
}
ubi_debugfs_exit();
kmem_cache_destroy(ubi_wl_entry_slab);
misc_deregister(&ubi_ctrl_cdev);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
class_remove_file(ubi_class, &ubi_version);
class_destroy(ubi_class);
}
module_exit(ubi_exit);
/**
* bytes_str_to_int - convert a number of bytes string into an integer.
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
* @str: the string to convert
*
* This function returns positive resulting integer in case of success and a
* negative error code in case of failure.
*/
static int __init bytes_str_to_int(const char *str)
{
char *endp;
unsigned long result;
result = simple_strtoul(str, &endp, 0);
if (str == endp || result >= INT_MAX) {
ubi_err("incorrect bytes count: \"%s\"\n", str);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
return -EINVAL;
}
switch (*endp) {
case 'G':
result *= 1024;
case 'M':
result *= 1024;
case 'K':
result *= 1024;
if (endp[1] == 'i' && endp[2] == 'B')
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
endp += 2;
case '\0':
break;
default:
ubi_err("incorrect bytes count: \"%s\"\n", str);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
return -EINVAL;
}
return result;
}
/**
* ubi_mtd_param_parse - parse the 'mtd=' UBI parameter.
* @val: the parameter value to parse
* @kp: not used
*
* This function returns zero in case of success and a negative error code in
* case of error.
*/
static int __init ubi_mtd_param_parse(const char *val, struct kernel_param *kp)
{
int i, len;
struct mtd_dev_param *p;
char buf[MTD_PARAM_LEN_MAX];
char *pbuf = &buf[0];
char *tokens[MTD_PARAM_MAX_COUNT], *token;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
if (!val)
return -EINVAL;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
if (mtd_devs == UBI_MAX_DEVICES) {
ubi_err("too many parameters, max. is %d\n",
UBI_MAX_DEVICES);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
return -EINVAL;
}
len = strnlen(val, MTD_PARAM_LEN_MAX);
if (len == MTD_PARAM_LEN_MAX) {
ubi_err("parameter \"%s\" is too long, max. is %d\n",
val, MTD_PARAM_LEN_MAX);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
return -EINVAL;
}
if (len == 0) {
pr_warn("UBI warning: empty 'mtd=' parameter - ignored\n");
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
return 0;
}
strcpy(buf, val);
/* Get rid of the final newline */
if (buf[len - 1] == '\n')
buf[len - 1] = '\0';
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
for (i = 0; i < MTD_PARAM_MAX_COUNT; i++)
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
tokens[i] = strsep(&pbuf, ",");
if (pbuf) {
ubi_err("too many arguments at \"%s\"\n", val);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
return -EINVAL;
}
p = &mtd_dev_param[mtd_devs];
strcpy(&p->name[0], tokens[0]);
token = tokens[1];
if (token) {
p->vid_hdr_offs = bytes_str_to_int(token);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
if (p->vid_hdr_offs < 0)
return p->vid_hdr_offs;
}
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
token = tokens[2];
if (token) {
int err = kstrtoint(token, 10, &p->max_beb_per1024);
if (err) {
ubi_err("bad value for max_beb_per1024 parameter: %s",
token);
return -EINVAL;
}
}
token = tokens[3];
if (token) {
int err = kstrtoint(token, 10, &p->ubi_num);
if (err) {
ubi_err("bad value for ubi_num parameter: %s", token);
return -EINVAL;
}
} else
p->ubi_num = UBI_DEV_NUM_AUTO;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
mtd_devs += 1;
return 0;
}
module_param_call(mtd, ubi_mtd_param_parse, NULL, NULL, 000);
MODULE_PARM_DESC(mtd, "MTD devices to attach. Parameter format: mtd=<name|num|path>[,<vid_hdr_offs>[,max_beb_per1024[,ubi_num]]].\n"
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
"Multiple \"mtd\" parameters may be specified.\n"
"MTD devices may be specified by their number, name, or path to the MTD character device node.\n"
"Optional \"vid_hdr_offs\" parameter specifies UBI VID header position to be used by UBI. (default value if 0)\n"
"Optional \"max_beb_per1024\" parameter specifies the maximum expected bad eraseblock per 1024 eraseblocks. (default value ("
__stringify(CONFIG_MTD_UBI_BEB_LIMIT) ") if 0)\n"
"Optional \"ubi_num\" parameter specifies UBI device number which have to be assigned to the newly created UBI device (assigned automatically by default)\n"
"\n"
"Example 1: mtd=/dev/mtd0 - attach MTD device /dev/mtd0.\n"
"Example 2: mtd=content,1984 mtd=4 - attach MTD device with name \"content\" using VID header offset 1984, and MTD device number 4 with default VID header offset.\n"
"Example 3: mtd=/dev/mtd1,0,25 - attach MTD device /dev/mtd1 using default VID header offset and reserve 25*nand_size_in_blocks/1024 erase blocks for bad block handling.\n"
"Example 4: mtd=/dev/mtd1,0,0,5 - attach MTD device /dev/mtd1 to UBI 5 and using default values for the other fields.\n"
"\t(e.g. if the NAND *chipset* has 4096 PEB, 100 will be reserved for this UBI device).");
#ifdef CONFIG_MTD_UBI_FASTMAP
module_param(fm_autoconvert, bool, 0644);
MODULE_PARM_DESC(fm_autoconvert, "Set this parameter to enable fastmap automatically on images without a fastmap.");
#endif
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 16:22:22 +08:00
MODULE_VERSION(__stringify(UBI_VERSION));
MODULE_DESCRIPTION("UBI - Unsorted Block Images");
MODULE_AUTHOR("Artem Bityutskiy");
MODULE_LICENSE("GPL");