OpenCloudOS-Kernel/drivers/scsi/lpfc/lpfc_debugfs.c

6469 lines
184 KiB
C
Raw Normal View History

/*******************************************************************
* This file is part of the Emulex Linux Device Driver for *
* Fibre Channel Host Bus Adapters. *
* Copyright (C) 2017-2019 Broadcom. All Rights Reserved. The term *
* Broadcom refers to Broadcom Inc. and/or its subsidiaries. *
* Copyright (C) 2007-2015 Emulex. All rights reserved. *
* EMULEX and SLI are trademarks of Emulex. *
* www.broadcom.com *
* *
* This program is free software; you can redistribute it and/or *
* modify it under the terms of version 2 of the GNU General *
* Public License as published by the Free Software Foundation. *
* This program is distributed in the hope that it will be useful. *
* ALL EXPRESS OR IMPLIED CONDITIONS, REPRESENTATIONS AND *
* WARRANTIES, INCLUDING ANY IMPLIED WARRANTY OF MERCHANTABILITY, *
* FITNESS FOR A PARTICULAR PURPOSE, OR NON-INFRINGEMENT, ARE *
* DISCLAIMED, EXCEPT TO THE EXTENT THAT SUCH DISCLAIMERS ARE HELD *
* TO BE LEGALLY INVALID. See the GNU General Public License for *
* more details, a copy of which can be found in the file COPYING *
* included with this package. *
*******************************************************************/
#include <linux/blkdev.h>
#include <linux/delay.h>
#include <linux/module.h>
#include <linux/dma-mapping.h>
#include <linux/idr.h>
#include <linux/interrupt.h>
#include <linux/kthread.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>
#include <linux/pci.h>
#include <linux/spinlock.h>
#include <linux/ctype.h>
#include <scsi/scsi.h>
#include <scsi/scsi_device.h>
#include <scsi/scsi_host.h>
#include <scsi/scsi_transport_fc.h>
#include <scsi/fc/fc_fs.h>
#include <linux/nvme-fc-driver.h>
#include "lpfc_hw4.h"
#include "lpfc_hw.h"
#include "lpfc_sli.h"
#include "lpfc_sli4.h"
#include "lpfc_nl.h"
#include "lpfc_disc.h"
#include "lpfc.h"
#include "lpfc_scsi.h"
#include "lpfc_nvme.h"
#include "lpfc_nvmet.h"
#include "lpfc_logmsg.h"
#include "lpfc_crtn.h"
#include "lpfc_vport.h"
#include "lpfc_version.h"
#include "lpfc_compat.h"
#include "lpfc_debugfs.h"
#include "lpfc_bsg.h"
#ifdef CONFIG_SCSI_LPFC_DEBUG_FS
/*
* debugfs interface
*
* To access this interface the user should:
* # mount -t debugfs none /sys/kernel/debug
*
* The lpfc debugfs directory hierarchy is:
* /sys/kernel/debug/lpfc/fnX/vportY
* where X is the lpfc hba function unique_id
* where Y is the vport VPI on that hba
*
* Debugging services available per vport:
* discovery_trace
* This is an ACSII readable file that contains a trace of the last
* lpfc_debugfs_max_disc_trc events that happened on a specific vport.
* See lpfc_debugfs.h for different categories of discovery events.
* To enable the discovery trace, the following module parameters must be set:
* lpfc_debugfs_enable=1 Turns on lpfc debugfs filesystem support
* lpfc_debugfs_max_disc_trc=X Where X is the event trace depth for
* EACH vport. X MUST also be a power of 2.
* lpfc_debugfs_mask_disc_trc=Y Where Y is an event mask as defined in
* lpfc_debugfs.h .
*
* slow_ring_trace
* This is an ACSII readable file that contains a trace of the last
* lpfc_debugfs_max_slow_ring_trc events that happened on a specific HBA.
* To enable the slow ring trace, the following module parameters must be set:
* lpfc_debugfs_enable=1 Turns on lpfc debugfs filesystem support
* lpfc_debugfs_max_slow_ring_trc=X Where X is the event trace depth for
* the HBA. X MUST also be a power of 2.
*/
static int lpfc_debugfs_enable = 1;
module_param(lpfc_debugfs_enable, int, S_IRUGO);
MODULE_PARM_DESC(lpfc_debugfs_enable, "Enable debugfs services");
/* This MUST be a power of 2 */
static int lpfc_debugfs_max_disc_trc;
module_param(lpfc_debugfs_max_disc_trc, int, S_IRUGO);
MODULE_PARM_DESC(lpfc_debugfs_max_disc_trc,
"Set debugfs discovery trace depth");
/* This MUST be a power of 2 */
static int lpfc_debugfs_max_slow_ring_trc;
module_param(lpfc_debugfs_max_slow_ring_trc, int, S_IRUGO);
MODULE_PARM_DESC(lpfc_debugfs_max_slow_ring_trc,
"Set debugfs slow ring trace depth");
/* This MUST be a power of 2 */
static int lpfc_debugfs_max_nvmeio_trc;
module_param(lpfc_debugfs_max_nvmeio_trc, int, 0444);
MODULE_PARM_DESC(lpfc_debugfs_max_nvmeio_trc,
"Set debugfs NVME IO trace depth");
[SCSI] lpfc 8.3.1: misc fixes/changes 8.3.1 Fixes/Changes : - Fix incorrect byte-swapping on word 4 of IOCB (data length) which caused LUNs to not be discovered on big-endian (e.g. PPC) - Remove a bad cast of MBslimaddr which loses the __iomem (sparse) - Make lpfc_debugfs_mask_disc_trc static (sparse) - Correct misspelled word BlockGuard in lpfc_logmsg.h comment - Replaced repeated code segment for canceling IOCBs from a list with a function call, lpfc_sli_cancel_iocbs(). - Increased HBQ buffers to support 40KB SSC sequences. - Added sysfs interface to update speed and topology parameter without link bounce. - Fixed bug with sysfs fc_host WWNs not being updated after changing the WWNs. - Check if the active mailbox is NULL in the beginning of the mailbox timeout handler - fixes panic in the mailbox timeout handler while running IO stress test - Fixed system panic in lpfc_pci_remove_one() due to ndlp indirect reference to phba through vport - Removed de-reference of scsi device after call to scsi_done() to fix panic in scsi completion path while accessing scsi device after scsi_done is called. - Fixed "Nodelist not empty" message when unloading the driver after target reboot test - Added LP2105 HBA model description - Added code to print all 16 words of unrecognized ASYNC events - Fixed memory leak in vport create + delete loop - Added support for handling dual error bit from HBA - Fixed a driver NULL pointer dereference in lpfc_sli_process_sol_iocb - Fixed a discovery bug with FC switch reboot in lpfc_setup_disc_node - Take NULL termintator into account when calculating available buffer space Signed-off-by: James Smart <james.smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2009-04-07 06:48:10 +08:00
static int lpfc_debugfs_mask_disc_trc;
module_param(lpfc_debugfs_mask_disc_trc, int, S_IRUGO);
MODULE_PARM_DESC(lpfc_debugfs_mask_disc_trc,
"Set debugfs discovery trace mask");
#include <linux/debugfs.h>
static atomic_t lpfc_debugfs_seq_trc_cnt = ATOMIC_INIT(0);
static unsigned long lpfc_debugfs_start_time = 0L;
/* iDiag */
static struct lpfc_idiag idiag;
/**
* lpfc_debugfs_disc_trc_data - Dump discovery logging to a buffer
* @vport: The vport to gather the log info from.
* @buf: The buffer to dump log into.
* @size: The maximum amount of data to process.
*
* Description:
* This routine gathers the lpfc discovery debugfs data from the @vport and
* dumps it to @buf up to @size number of bytes. It will start at the next entry
* in the log and process the log until the end of the buffer. Then it will
* gather from the beginning of the log and process until the current entry.
*
* Notes:
* Discovery logging will be disabled while while this routine dumps the log.
*
* Return Value:
* This routine returns the amount of bytes that were dumped into @buf and will
* not exceed @size.
**/
static int
lpfc_debugfs_disc_trc_data(struct lpfc_vport *vport, char *buf, int size)
{
int i, index, len, enable;
uint32_t ms;
struct lpfc_debugfs_trc *dtp;
char *buffer;
buffer = kmalloc(LPFC_DEBUG_TRC_ENTRY_SIZE, GFP_KERNEL);
if (!buffer)
return 0;
enable = lpfc_debugfs_enable;
lpfc_debugfs_enable = 0;
len = 0;
index = (atomic_read(&vport->disc_trc_cnt) + 1) &
(lpfc_debugfs_max_disc_trc - 1);
for (i = index; i < lpfc_debugfs_max_disc_trc; i++) {
dtp = vport->disc_trc + i;
if (!dtp->fmt)
continue;
ms = jiffies_to_msecs(dtp->jif - lpfc_debugfs_start_time);
snprintf(buffer,
LPFC_DEBUG_TRC_ENTRY_SIZE, "%010d:%010d ms:%s\n",
dtp->seq_cnt, ms, dtp->fmt);
len += snprintf(buf+len, size-len, buffer,
dtp->data1, dtp->data2, dtp->data3);
}
for (i = 0; i < index; i++) {
dtp = vport->disc_trc + i;
if (!dtp->fmt)
continue;
ms = jiffies_to_msecs(dtp->jif - lpfc_debugfs_start_time);
snprintf(buffer,
LPFC_DEBUG_TRC_ENTRY_SIZE, "%010d:%010d ms:%s\n",
dtp->seq_cnt, ms, dtp->fmt);
len += snprintf(buf+len, size-len, buffer,
dtp->data1, dtp->data2, dtp->data3);
}
lpfc_debugfs_enable = enable;
kfree(buffer);
return len;
}
/**
* lpfc_debugfs_slow_ring_trc_data - Dump slow ring logging to a buffer
* @phba: The HBA to gather the log info from.
* @buf: The buffer to dump log into.
* @size: The maximum amount of data to process.
*
* Description:
* This routine gathers the lpfc slow ring debugfs data from the @phba and
* dumps it to @buf up to @size number of bytes. It will start at the next entry
* in the log and process the log until the end of the buffer. Then it will
* gather from the beginning of the log and process until the current entry.
*
* Notes:
* Slow ring logging will be disabled while while this routine dumps the log.
*
* Return Value:
* This routine returns the amount of bytes that were dumped into @buf and will
* not exceed @size.
**/
static int
lpfc_debugfs_slow_ring_trc_data(struct lpfc_hba *phba, char *buf, int size)
{
int i, index, len, enable;
uint32_t ms;
struct lpfc_debugfs_trc *dtp;
char *buffer;
buffer = kmalloc(LPFC_DEBUG_TRC_ENTRY_SIZE, GFP_KERNEL);
if (!buffer)
return 0;
enable = lpfc_debugfs_enable;
lpfc_debugfs_enable = 0;
len = 0;
index = (atomic_read(&phba->slow_ring_trc_cnt) + 1) &
(lpfc_debugfs_max_slow_ring_trc - 1);
for (i = index; i < lpfc_debugfs_max_slow_ring_trc; i++) {
dtp = phba->slow_ring_trc + i;
if (!dtp->fmt)
continue;
ms = jiffies_to_msecs(dtp->jif - lpfc_debugfs_start_time);
snprintf(buffer,
LPFC_DEBUG_TRC_ENTRY_SIZE, "%010d:%010d ms:%s\n",
dtp->seq_cnt, ms, dtp->fmt);
len += snprintf(buf+len, size-len, buffer,
dtp->data1, dtp->data2, dtp->data3);
}
for (i = 0; i < index; i++) {
dtp = phba->slow_ring_trc + i;
if (!dtp->fmt)
continue;
ms = jiffies_to_msecs(dtp->jif - lpfc_debugfs_start_time);
snprintf(buffer,
LPFC_DEBUG_TRC_ENTRY_SIZE, "%010d:%010d ms:%s\n",
dtp->seq_cnt, ms, dtp->fmt);
len += snprintf(buf+len, size-len, buffer,
dtp->data1, dtp->data2, dtp->data3);
}
lpfc_debugfs_enable = enable;
kfree(buffer);
return len;
}
static int lpfc_debugfs_last_hbq = -1;
/**
* lpfc_debugfs_hbqinfo_data - Dump host buffer queue info to a buffer
* @phba: The HBA to gather host buffer info from.
* @buf: The buffer to dump log into.
* @size: The maximum amount of data to process.
*
* Description:
* This routine dumps the host buffer queue info from the @phba to @buf up to
* @size number of bytes. A header that describes the current hbq state will be
* dumped to @buf first and then info on each hbq entry will be dumped to @buf
* until @size bytes have been dumped or all the hbq info has been dumped.
*
* Notes:
* This routine will rotate through each configured HBQ each time called.
*
* Return Value:
* This routine returns the amount of bytes that were dumped into @buf and will
* not exceed @size.
**/
static int
lpfc_debugfs_hbqinfo_data(struct lpfc_hba *phba, char *buf, int size)
{
int len = 0;
int i, j, found, posted, low;
uint32_t phys, raw_index, getidx;
struct lpfc_hbq_init *hip;
struct hbq_s *hbqs;
struct lpfc_hbq_entry *hbqe;
struct lpfc_dmabuf *d_buf;
struct hbq_dmabuf *hbq_buf;
if (phba->sli_rev != 3)
return 0;
spin_lock_irq(&phba->hbalock);
/* toggle between multiple hbqs, if any */
i = lpfc_sli_hbq_count();
if (i > 1) {
lpfc_debugfs_last_hbq++;
if (lpfc_debugfs_last_hbq >= i)
lpfc_debugfs_last_hbq = 0;
}
else
lpfc_debugfs_last_hbq = 0;
i = lpfc_debugfs_last_hbq;
len += snprintf(buf+len, size-len, "HBQ %d Info\n", i);
hbqs = &phba->hbqs[i];
posted = 0;
list_for_each_entry(d_buf, &hbqs->hbq_buffer_list, list)
posted++;
hip = lpfc_hbq_defs[i];
len += snprintf(buf+len, size-len,
"idx:%d prof:%d rn:%d bufcnt:%d icnt:%d acnt:%d posted %d\n",
hip->hbq_index, hip->profile, hip->rn,
hip->buffer_count, hip->init_count, hip->add_count, posted);
raw_index = phba->hbq_get[i];
getidx = le32_to_cpu(raw_index);
len += snprintf(buf+len, size-len,
"entries:%d bufcnt:%d Put:%d nPut:%d localGet:%d hbaGet:%d\n",
hbqs->entry_count, hbqs->buffer_count, hbqs->hbqPutIdx,
hbqs->next_hbqPutIdx, hbqs->local_hbqGetIdx, getidx);
hbqe = (struct lpfc_hbq_entry *) phba->hbqs[i].hbq_virt;
for (j=0; j<hbqs->entry_count; j++) {
len += snprintf(buf+len, size-len,
"%03d: %08x %04x %05x ", j,
le32_to_cpu(hbqe->bde.addrLow),
le32_to_cpu(hbqe->bde.tus.w),
le32_to_cpu(hbqe->buffer_tag));
i = 0;
found = 0;
/* First calculate if slot has an associated posted buffer */
low = hbqs->hbqPutIdx - posted;
if (low >= 0) {
if ((j >= hbqs->hbqPutIdx) || (j < low)) {
len += snprintf(buf+len, size-len, "Unused\n");
goto skipit;
}
}
else {
if ((j >= hbqs->hbqPutIdx) &&
(j < (hbqs->entry_count+low))) {
len += snprintf(buf+len, size-len, "Unused\n");
goto skipit;
}
}
/* Get the Buffer info for the posted buffer */
list_for_each_entry(d_buf, &hbqs->hbq_buffer_list, list) {
hbq_buf = container_of(d_buf, struct hbq_dmabuf, dbuf);
phys = ((uint64_t)hbq_buf->dbuf.phys & 0xffffffff);
if (phys == le32_to_cpu(hbqe->bde.addrLow)) {
len += snprintf(buf+len, size-len,
"Buf%d: %p %06x\n", i,
hbq_buf->dbuf.virt, hbq_buf->tag);
found = 1;
break;
}
i++;
}
if (!found) {
len += snprintf(buf+len, size-len, "No DMAinfo?\n");
}
skipit:
hbqe++;
if (len > LPFC_HBQINFO_SIZE - 54)
break;
}
spin_unlock_irq(&phba->hbalock);
return len;
}
static int lpfc_debugfs_last_xripool;
/**
* lpfc_debugfs_common_xri_data - Dump Hardware Queue info to a buffer
* @phba: The HBA to gather host buffer info from.
* @buf: The buffer to dump log into.
* @size: The maximum amount of data to process.
*
* Description:
* This routine dumps the Hardware Queue info from the @phba to @buf up to
* @size number of bytes. A header that describes the current hdwq state will be
* dumped to @buf first and then info on each hdwq entry will be dumped to @buf
* until @size bytes have been dumped or all the hdwq info has been dumped.
*
* Notes:
* This routine will rotate through each configured Hardware Queue each
* time called.
*
* Return Value:
* This routine returns the amount of bytes that were dumped into @buf and will
* not exceed @size.
**/
static int
lpfc_debugfs_commonxripools_data(struct lpfc_hba *phba, char *buf, int size)
{
struct lpfc_sli4_hdw_queue *qp;
int len = 0;
int i, out;
unsigned long iflag;
for (i = 0; i < phba->cfg_hdw_queue; i++) {
if (len > (LPFC_DUMP_MULTIXRIPOOL_SIZE - 80))
break;
qp = &phba->sli4_hba.hdwq[lpfc_debugfs_last_xripool];
len += snprintf(buf + len, size - len, "HdwQ %d Info ", i);
spin_lock_irqsave(&qp->abts_scsi_buf_list_lock, iflag);
spin_lock(&qp->abts_nvme_buf_list_lock);
spin_lock(&qp->io_buf_list_get_lock);
spin_lock(&qp->io_buf_list_put_lock);
out = qp->total_io_bufs - (qp->get_io_bufs + qp->put_io_bufs +
qp->abts_scsi_io_bufs + qp->abts_nvme_io_bufs);
len += snprintf(buf + len, size - len,
"tot:%d get:%d put:%d mt:%d "
"ABTS scsi:%d nvme:%d Out:%d\n",
qp->total_io_bufs, qp->get_io_bufs, qp->put_io_bufs,
qp->empty_io_bufs, qp->abts_scsi_io_bufs,
qp->abts_nvme_io_bufs, out);
spin_unlock(&qp->io_buf_list_put_lock);
spin_unlock(&qp->io_buf_list_get_lock);
spin_unlock(&qp->abts_nvme_buf_list_lock);
spin_unlock_irqrestore(&qp->abts_scsi_buf_list_lock, iflag);
lpfc_debugfs_last_xripool++;
if (lpfc_debugfs_last_xripool >= phba->cfg_hdw_queue)
lpfc_debugfs_last_xripool = 0;
}
return len;
}
scsi: lpfc: Adapt partitioned XRI lists to efficient sharing The XRI get/put lists were partitioned per hardware queue. However, the adapter rarely had sufficient resources to give a large number of resources per queue. As such, it became common for a cpu to encounter a lack of XRI resource and request the upper io stack to retry after returning a BUSY condition. This occurred even though other cpus were idle and not using their resources. Create as efficient a scheme as possible to move resources to the cpus that need them. Each cpu maintains a small private pool which it allocates from for io. There is a watermark that the cpu attempts to keep in the private pool. The private pool, when empty, pulls from a global pool from the cpu. When the cpu's global pool is empty it will pull from other cpu's global pool. As there many cpu global pools (1 per cpu or hardware queue count) and as each cpu selects what cpu to pull from at different rates and at different times, it creates a radomizing effect that minimizes the number of cpu's that will contend with each other when the steal XRI's from another cpu's global pool. On io completion, a cpu will push the XRI back on to its private pool. A watermark level is maintained for the private pool such that when it is exceeded it will move XRI's to the CPU global pool so that other cpu's may allocate them. On NVME, as heartbeat commands are critical to get placed on the wire, a single expedite pool is maintained. When a heartbeat is to be sent, it will allocate an XRI from the expedite pool rather than the normal cpu private/global pools. On any io completion, if a reduction in the expedite pools is seen, it will be replenished before the XRI is placed on the cpu private pool. Statistics are added to aid understanding the XRI levels on each cpu and their behaviors. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <jsmart2021@gmail.com> Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2019-01-29 03:14:28 +08:00
/**
* lpfc_debugfs_multixripools_data - Display multi-XRI pools information
* @phba: The HBA to gather host buffer info from.
* @buf: The buffer to dump log into.
* @size: The maximum amount of data to process.
*
* Description:
* This routine displays current multi-XRI pools information including XRI
* count in public, private and txcmplq. It also displays current high and
* low watermark.
*
* Return Value:
* This routine returns the amount of bytes that were dumped into @buf and will
* not exceed @size.
**/
static int
lpfc_debugfs_multixripools_data(struct lpfc_hba *phba, char *buf, int size)
{
u32 i;
u32 hwq_count;
struct lpfc_sli4_hdw_queue *qp;
struct lpfc_multixri_pool *multixri_pool;
struct lpfc_pvt_pool *pvt_pool;
struct lpfc_pbl_pool *pbl_pool;
u32 txcmplq_cnt;
char tmp[LPFC_DEBUG_OUT_LINE_SZ] = {0};
if (phba->sli_rev != LPFC_SLI_REV4)
return 0;
if (!phba->sli4_hba.hdwq)
return 0;
if (!phba->cfg_xri_rebalancing) {
i = lpfc_debugfs_commonxripools_data(phba, buf, size);
return i;
}
scsi: lpfc: Adapt partitioned XRI lists to efficient sharing The XRI get/put lists were partitioned per hardware queue. However, the adapter rarely had sufficient resources to give a large number of resources per queue. As such, it became common for a cpu to encounter a lack of XRI resource and request the upper io stack to retry after returning a BUSY condition. This occurred even though other cpus were idle and not using their resources. Create as efficient a scheme as possible to move resources to the cpus that need them. Each cpu maintains a small private pool which it allocates from for io. There is a watermark that the cpu attempts to keep in the private pool. The private pool, when empty, pulls from a global pool from the cpu. When the cpu's global pool is empty it will pull from other cpu's global pool. As there many cpu global pools (1 per cpu or hardware queue count) and as each cpu selects what cpu to pull from at different rates and at different times, it creates a radomizing effect that minimizes the number of cpu's that will contend with each other when the steal XRI's from another cpu's global pool. On io completion, a cpu will push the XRI back on to its private pool. A watermark level is maintained for the private pool such that when it is exceeded it will move XRI's to the CPU global pool so that other cpu's may allocate them. On NVME, as heartbeat commands are critical to get placed on the wire, a single expedite pool is maintained. When a heartbeat is to be sent, it will allocate an XRI from the expedite pool rather than the normal cpu private/global pools. On any io completion, if a reduction in the expedite pools is seen, it will be replenished before the XRI is placed on the cpu private pool. Statistics are added to aid understanding the XRI levels on each cpu and their behaviors. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <jsmart2021@gmail.com> Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2019-01-29 03:14:28 +08:00
/*
* Pbl: Current number of free XRIs in public pool
* Pvt: Current number of free XRIs in private pool
* Busy: Current number of outstanding XRIs
* HWM: Current high watermark
* pvt_empty: Incremented by 1 when IO submission fails (no xri)
* pbl_empty: Incremented by 1 when all pbl_pool are empty during
* IO submission
*/
scnprintf(tmp, sizeof(tmp),
"HWQ: Pbl Pvt Busy HWM | pvt_empty pbl_empty ");
if (strlcat(buf, tmp, size) >= size)
return strnlen(buf, size);
#ifdef LPFC_MXP_STAT
/*
* MAXH: Max high watermark seen so far
* above_lmt: Incremented by 1 if xri_owned > xri_limit during
* IO submission
* below_lmt: Incremented by 1 if xri_owned <= xri_limit during
* IO submission
* locPbl_hit: Incremented by 1 if successfully get a batch of XRI from
* local pbl_pool
* othPbl_hit: Incremented by 1 if successfully get a batch of XRI from
* other pbl_pool
*/
scnprintf(tmp, sizeof(tmp),
"MAXH above_lmt below_lmt locPbl_hit othPbl_hit");
if (strlcat(buf, tmp, size) >= size)
return strnlen(buf, size);
/*
* sPbl: snapshot of Pbl 15 sec after stat gets cleared
* sPvt: snapshot of Pvt 15 sec after stat gets cleared
* sBusy: snapshot of Busy 15 sec after stat gets cleared
*/
scnprintf(tmp, sizeof(tmp),
" | sPbl sPvt sBusy");
if (strlcat(buf, tmp, size) >= size)
return strnlen(buf, size);
#endif
scnprintf(tmp, sizeof(tmp), "\n");
if (strlcat(buf, tmp, size) >= size)
return strnlen(buf, size);
hwq_count = phba->cfg_hdw_queue;
for (i = 0; i < hwq_count; i++) {
qp = &phba->sli4_hba.hdwq[i];
multixri_pool = qp->p_multixri_pool;
if (!multixri_pool)
continue;
pbl_pool = &multixri_pool->pbl_pool;
pvt_pool = &multixri_pool->pvt_pool;
txcmplq_cnt = qp->fcp_wq->pring->txcmplq_cnt;
if (qp->nvme_wq)
txcmplq_cnt += qp->nvme_wq->pring->txcmplq_cnt;
scnprintf(tmp, sizeof(tmp),
"%03d: %4d %4d %4d %4d | %10d %10d ",
i, pbl_pool->count, pvt_pool->count,
txcmplq_cnt, pvt_pool->high_watermark,
qp->empty_io_bufs, multixri_pool->pbl_empty_count);
if (strlcat(buf, tmp, size) >= size)
break;
#ifdef LPFC_MXP_STAT
scnprintf(tmp, sizeof(tmp),
"%4d %10d %10d %10d %10d",
multixri_pool->stat_max_hwm,
multixri_pool->above_limit_count,
multixri_pool->below_limit_count,
multixri_pool->local_pbl_hit_count,
multixri_pool->other_pbl_hit_count);
if (strlcat(buf, tmp, size) >= size)
break;
scnprintf(tmp, sizeof(tmp),
" | %4d %4d %5d",
multixri_pool->stat_pbl_count,
multixri_pool->stat_pvt_count,
multixri_pool->stat_busy_count);
if (strlcat(buf, tmp, size) >= size)
break;
#endif
scnprintf(tmp, sizeof(tmp), "\n");
if (strlcat(buf, tmp, size) >= size)
break;
}
return strnlen(buf, size);
}
#ifdef LPFC_HDWQ_LOCK_STAT
static int lpfc_debugfs_last_lock;
/**
* lpfc_debugfs_lockstat_data - Dump Hardware Queue info to a buffer
* @phba: The HBA to gather host buffer info from.
* @buf: The buffer to dump log into.
* @size: The maximum amount of data to process.
*
* Description:
* This routine dumps the Hardware Queue info from the @phba to @buf up to
* @size number of bytes. A header that describes the current hdwq state will be
* dumped to @buf first and then info on each hdwq entry will be dumped to @buf
* until @size bytes have been dumped or all the hdwq info has been dumped.
*
* Notes:
* This routine will rotate through each configured Hardware Queue each
* time called.
*
* Return Value:
* This routine returns the amount of bytes that were dumped into @buf and will
* not exceed @size.
**/
static int
lpfc_debugfs_lockstat_data(struct lpfc_hba *phba, char *buf, int size)
{
struct lpfc_sli4_hdw_queue *qp;
int len = 0;
int i;
if (phba->sli_rev != LPFC_SLI_REV4)
return 0;
if (!phba->sli4_hba.hdwq)
return 0;
for (i = 0; i < phba->cfg_hdw_queue; i++) {
if (len > (LPFC_HDWQINFO_SIZE - 100))
break;
qp = &phba->sli4_hba.hdwq[lpfc_debugfs_last_lock];
len += snprintf(buf + len, size - len, "HdwQ %03d Lock ", i);
if (phba->cfg_xri_rebalancing) {
len += snprintf(buf + len, size - len,
"get_pvt:%d mv_pvt:%d "
"mv2pub:%d mv2pvt:%d "
"put_pvt:%d put_pub:%d wq:%d\n",
qp->lock_conflict.alloc_pvt_pool,
qp->lock_conflict.mv_from_pvt_pool,
qp->lock_conflict.mv_to_pub_pool,
qp->lock_conflict.mv_to_pvt_pool,
qp->lock_conflict.free_pvt_pool,
qp->lock_conflict.free_pub_pool,
qp->lock_conflict.wq_access);
} else {
len += snprintf(buf + len, size - len,
"get:%d put:%d free:%d wq:%d\n",
qp->lock_conflict.alloc_xri_get,
qp->lock_conflict.alloc_xri_put,
qp->lock_conflict.free_xri,
qp->lock_conflict.wq_access);
}
lpfc_debugfs_last_lock++;
if (lpfc_debugfs_last_lock >= phba->cfg_hdw_queue)
lpfc_debugfs_last_lock = 0;
}
return len;
}
#endif
static int lpfc_debugfs_last_hba_slim_off;
/**
* lpfc_debugfs_dumpHBASlim_data - Dump HBA SLIM info to a buffer
* @phba: The HBA to gather SLIM info from.
* @buf: The buffer to dump log into.
* @size: The maximum amount of data to process.
*
* Description:
* This routine dumps the current contents of HBA SLIM for the HBA associated
* with @phba to @buf up to @size bytes of data. This is the raw HBA SLIM data.
*
* Notes:
* This routine will only dump up to 1024 bytes of data each time called and
* should be called multiple times to dump the entire HBA SLIM.
*
* Return Value:
* This routine returns the amount of bytes that were dumped into @buf and will
* not exceed @size.
**/
static int
lpfc_debugfs_dumpHBASlim_data(struct lpfc_hba *phba, char *buf, int size)
{
int len = 0;
int i, off;
uint32_t *ptr;
char *buffer;
buffer = kmalloc(1024, GFP_KERNEL);
if (!buffer)
return 0;
off = 0;
spin_lock_irq(&phba->hbalock);
len += snprintf(buf+len, size-len, "HBA SLIM\n");
lpfc_memcpy_from_slim(buffer,
[SCSI] lpfc 8.3.1: misc fixes/changes 8.3.1 Fixes/Changes : - Fix incorrect byte-swapping on word 4 of IOCB (data length) which caused LUNs to not be discovered on big-endian (e.g. PPC) - Remove a bad cast of MBslimaddr which loses the __iomem (sparse) - Make lpfc_debugfs_mask_disc_trc static (sparse) - Correct misspelled word BlockGuard in lpfc_logmsg.h comment - Replaced repeated code segment for canceling IOCBs from a list with a function call, lpfc_sli_cancel_iocbs(). - Increased HBQ buffers to support 40KB SSC sequences. - Added sysfs interface to update speed and topology parameter without link bounce. - Fixed bug with sysfs fc_host WWNs not being updated after changing the WWNs. - Check if the active mailbox is NULL in the beginning of the mailbox timeout handler - fixes panic in the mailbox timeout handler while running IO stress test - Fixed system panic in lpfc_pci_remove_one() due to ndlp indirect reference to phba through vport - Removed de-reference of scsi device after call to scsi_done() to fix panic in scsi completion path while accessing scsi device after scsi_done is called. - Fixed "Nodelist not empty" message when unloading the driver after target reboot test - Added LP2105 HBA model description - Added code to print all 16 words of unrecognized ASYNC events - Fixed memory leak in vport create + delete loop - Added support for handling dual error bit from HBA - Fixed a driver NULL pointer dereference in lpfc_sli_process_sol_iocb - Fixed a discovery bug with FC switch reboot in lpfc_setup_disc_node - Take NULL termintator into account when calculating available buffer space Signed-off-by: James Smart <james.smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2009-04-07 06:48:10 +08:00
phba->MBslimaddr + lpfc_debugfs_last_hba_slim_off, 1024);
ptr = (uint32_t *)&buffer[0];
off = lpfc_debugfs_last_hba_slim_off;
/* Set it up for the next time */
lpfc_debugfs_last_hba_slim_off += 1024;
if (lpfc_debugfs_last_hba_slim_off >= 4096)
lpfc_debugfs_last_hba_slim_off = 0;
i = 1024;
while (i > 0) {
len += snprintf(buf+len, size-len,
"%08x: %08x %08x %08x %08x %08x %08x %08x %08x\n",
off, *ptr, *(ptr+1), *(ptr+2), *(ptr+3), *(ptr+4),
*(ptr+5), *(ptr+6), *(ptr+7));
ptr += 8;
i -= (8 * sizeof(uint32_t));
off += (8 * sizeof(uint32_t));
}
spin_unlock_irq(&phba->hbalock);
kfree(buffer);
return len;
}
/**
* lpfc_debugfs_dumpHostSlim_data - Dump host SLIM info to a buffer
* @phba: The HBA to gather Host SLIM info from.
* @buf: The buffer to dump log into.
* @size: The maximum amount of data to process.
*
* Description:
* This routine dumps the current contents of host SLIM for the host associated
* with @phba to @buf up to @size bytes of data. The dump will contain the
* Mailbox, PCB, Rings, and Registers that are located in host memory.
*
* Return Value:
* This routine returns the amount of bytes that were dumped into @buf and will
* not exceed @size.
**/
static int
lpfc_debugfs_dumpHostSlim_data(struct lpfc_hba *phba, char *buf, int size)
{
int len = 0;
int i, off;
uint32_t word0, word1, word2, word3;
uint32_t *ptr;
struct lpfc_pgp *pgpp;
struct lpfc_sli *psli = &phba->sli;
struct lpfc_sli_ring *pring;
off = 0;
spin_lock_irq(&phba->hbalock);
len += snprintf(buf+len, size-len, "SLIM Mailbox\n");
ptr = (uint32_t *)phba->slim2p.virt;
i = sizeof(MAILBOX_t);
while (i > 0) {
len += snprintf(buf+len, size-len,
"%08x: %08x %08x %08x %08x %08x %08x %08x %08x\n",
off, *ptr, *(ptr+1), *(ptr+2), *(ptr+3), *(ptr+4),
*(ptr+5), *(ptr+6), *(ptr+7));
ptr += 8;
i -= (8 * sizeof(uint32_t));
off += (8 * sizeof(uint32_t));
}
len += snprintf(buf+len, size-len, "SLIM PCB\n");
ptr = (uint32_t *)phba->pcb;
i = sizeof(PCB_t);
while (i > 0) {
len += snprintf(buf+len, size-len,
"%08x: %08x %08x %08x %08x %08x %08x %08x %08x\n",
off, *ptr, *(ptr+1), *(ptr+2), *(ptr+3), *(ptr+4),
*(ptr+5), *(ptr+6), *(ptr+7));
ptr += 8;
i -= (8 * sizeof(uint32_t));
off += (8 * sizeof(uint32_t));
}
if (phba->sli_rev <= LPFC_SLI_REV3) {
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-13 05:52:30 +08:00
for (i = 0; i < 4; i++) {
pgpp = &phba->port_gp[i];
pring = &psli->sli3_ring[i];
len += snprintf(buf+len, size-len,
"Ring %d: CMD GetInx:%d "
"(Max:%d Next:%d "
"Local:%d flg:x%x) "
"RSP PutInx:%d Max:%d\n",
i, pgpp->cmdGetInx,
pring->sli.sli3.numCiocb,
pring->sli.sli3.next_cmdidx,
pring->sli.sli3.local_getidx,
pring->flag, pgpp->rspPutInx,
pring->sli.sli3.numRiocb);
}
word0 = readl(phba->HAregaddr);
word1 = readl(phba->CAregaddr);
word2 = readl(phba->HSregaddr);
word3 = readl(phba->HCregaddr);
len += snprintf(buf+len, size-len, "HA:%08x CA:%08x HS:%08x "
"HC:%08x\n", word0, word1, word2, word3);
}
spin_unlock_irq(&phba->hbalock);
return len;
}
/**
* lpfc_debugfs_nodelist_data - Dump target node list to a buffer
* @vport: The vport to gather target node info from.
* @buf: The buffer to dump log into.
* @size: The maximum amount of data to process.
*
* Description:
* This routine dumps the current target node list associated with @vport to
* @buf up to @size bytes of data. Each node entry in the dump will contain a
* node state, DID, WWPN, WWNN, RPI, flags, type, and other useful fields.
*
* Return Value:
* This routine returns the amount of bytes that were dumped into @buf and will
* not exceed @size.
**/
static int
lpfc_debugfs_nodelist_data(struct lpfc_vport *vport, char *buf, int size)
{
int len = 0;
int i, iocnt, outio, cnt;
struct Scsi_Host *shost = lpfc_shost_from_vport(vport);
struct lpfc_hba *phba = vport->phba;
struct lpfc_nodelist *ndlp;
unsigned char *statep;
struct nvme_fc_local_port *localport;
struct nvme_fc_remote_port *nrport = NULL;
struct lpfc_nvme_rport *rport;
cnt = (LPFC_NODELIST_SIZE / LPFC_NODELIST_ENTRY_SIZE);
outio = 0;
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-13 05:52:30 +08:00
len += snprintf(buf+len, size-len, "\nFCP Nodelist Entries ...\n");
spin_lock_irq(shost->host_lock);
list_for_each_entry(ndlp, &vport->fc_nodes, nlp_listp) {
iocnt = 0;
if (!cnt) {
len += snprintf(buf+len, size-len,
"Missing Nodelist Entries\n");
break;
}
cnt--;
switch (ndlp->nlp_state) {
case NLP_STE_UNUSED_NODE:
statep = "UNUSED";
break;
case NLP_STE_PLOGI_ISSUE:
statep = "PLOGI ";
break;
case NLP_STE_ADISC_ISSUE:
statep = "ADISC ";
break;
case NLP_STE_REG_LOGIN_ISSUE:
statep = "REGLOG";
break;
case NLP_STE_PRLI_ISSUE:
statep = "PRLI ";
break;
case NLP_STE_LOGO_ISSUE:
statep = "LOGO ";
break;
case NLP_STE_UNMAPPED_NODE:
statep = "UNMAP ";
iocnt = 1;
break;
case NLP_STE_MAPPED_NODE:
statep = "MAPPED";
iocnt = 1;
break;
case NLP_STE_NPR_NODE:
statep = "NPR ";
break;
default:
statep = "UNKNOWN";
}
len += snprintf(buf+len, size-len, "%s DID:x%06x ",
statep, ndlp->nlp_DID);
len += snprintf(buf+len, size-len,
"WWPN x%llx ",
wwn_to_u64(ndlp->nlp_portname.u.wwn));
len += snprintf(buf+len, size-len,
"WWNN x%llx ",
wwn_to_u64(ndlp->nlp_nodename.u.wwn));
if (ndlp->nlp_flag & NLP_RPI_REGISTERED)
len += snprintf(buf+len, size-len, "RPI:%03d ",
ndlp->nlp_rpi);
else
len += snprintf(buf+len, size-len, "RPI:none ");
len += snprintf(buf+len, size-len, "flag:x%08x ",
ndlp->nlp_flag);
if (!ndlp->nlp_type)
len += snprintf(buf+len, size-len, "UNKNOWN_TYPE ");
if (ndlp->nlp_type & NLP_FC_NODE)
len += snprintf(buf+len, size-len, "FC_NODE ");
if (ndlp->nlp_type & NLP_FABRIC) {
len += snprintf(buf+len, size-len, "FABRIC ");
iocnt = 0;
}
if (ndlp->nlp_type & NLP_FCP_TARGET)
len += snprintf(buf+len, size-len, "FCP_TGT sid:%d ",
ndlp->nlp_sid);
if (ndlp->nlp_type & NLP_FCP_INITIATOR)
len += snprintf(buf+len, size-len, "FCP_INITIATOR ");
if (ndlp->nlp_type & NLP_NVME_TARGET)
len += snprintf(buf + len,
size - len, "NVME_TGT sid:%d ",
NLP_NO_SID);
if (ndlp->nlp_type & NLP_NVME_INITIATOR)
len += snprintf(buf + len,
size - len, "NVME_INITIATOR ");
[SCSI] lpfc 8.2.6 : Multiple discovery fixes Multiple Discovery Fixes: - Fix race on discovery due to link events coinciding with vport_delete. - Use NLP_FABRIC state to filter out switch-based pseudo initiators that reuse the same WWNs. - Correct erroneous setting of DID=0 in lpfc_matchdid() - Correct extra reference count that was in the lookup path for the remoteid from an unsolicited ELS. - Correct double-free bug in els abort path. - Correct FDMI server discovery logic for switch that return a WWN of 0. - Fix bugs in ndlp mgmt when a node changes address - Correct bug that did not delete RSCNs for vports upon link transitions - Fix "0216 Link event during NS query" error which pops up when vports are swapped to different switch ports. - Add sanity checks on ndlp structures - Fix devloss log message to dump WWN correctly - Hold off mgmt commands that were interferring with discovery mailbox cmds - Remove unnecessary FC_ESTABLISH_LINK logic. - Correct some race conditions in the worker thread, resulting in devloss: - Clear the work_port_events field before handling the work port events - Clear the deferred ring event before handling a deferred ring event - Hold the hba lock when waking up the work thread - Send an acc for the rscn even when we aren't going to handle it - Fix locking behavior that was not properly protecting the ACTIVE flag, thus allowing mailbox command order to shift. Signed-off-by: James Smart <james.smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2008-04-07 22:15:56 +08:00
len += snprintf(buf+len, size-len, "usgmap:%x ",
ndlp->nlp_usg_map);
len += snprintf(buf+len, size-len, "refcnt:%x",
kref_read(&ndlp->kref));
if (iocnt) {
i = atomic_read(&ndlp->cmd_pending);
len += snprintf(buf + len, size - len,
" OutIO:x%x Qdepth x%x",
i, ndlp->cmd_qdepth);
outio += i;
}
len += snprintf(buf + len, size - len, "defer:%x ",
ndlp->nlp_defer_did);
len += snprintf(buf+len, size-len, "\n");
}
spin_unlock_irq(shost->host_lock);
len += snprintf(buf + len, size - len,
"\nOutstanding IO x%x\n", outio);
if (phba->nvmet_support && phba->targetport && (vport == phba->pport)) {
len += snprintf(buf + len, size - len,
"\nNVME Targetport Entry ...\n");
/* Port state is only one of two values for now. */
if (phba->targetport->port_id)
statep = "REGISTERED";
else
statep = "INIT";
len += snprintf(buf + len, size - len,
"TGT WWNN x%llx WWPN x%llx State %s\n",
wwn_to_u64(vport->fc_nodename.u.wwn),
wwn_to_u64(vport->fc_portname.u.wwn),
statep);
len += snprintf(buf + len, size - len,
" Targetport DID x%06x\n",
phba->targetport->port_id);
goto out_exit;
}
len += snprintf(buf + len, size - len,
"\nNVME Lport/Rport Entries ...\n");
localport = vport->localport;
if (!localport)
goto out_exit;
spin_lock_irq(shost->host_lock);
/* Port state is only one of two values for now. */
if (localport->port_id)
statep = "ONLINE";
else
statep = "UNKNOWN ";
len += snprintf(buf + len, size - len,
"Lport DID x%06x PortState %s\n",
localport->port_id, statep);
len += snprintf(buf + len, size - len, "\tRport List:\n");
list_for_each_entry(ndlp, &vport->fc_nodes, nlp_listp) {
/* local short-hand pointer. */
spin_lock(&phba->hbalock);
rport = lpfc_ndlp_get_nrport(ndlp);
if (rport)
nrport = rport->remoteport;
else
nrport = NULL;
spin_unlock(&phba->hbalock);
if (!nrport)
continue;
/* Port state is only one of two values for now. */
switch (nrport->port_state) {
case FC_OBJSTATE_ONLINE:
statep = "ONLINE";
break;
case FC_OBJSTATE_UNKNOWN:
statep = "UNKNOWN ";
break;
default:
statep = "UNSUPPORTED";
break;
}
/* Tab in to show lport ownership. */
len += snprintf(buf + len, size - len,
"\t%s Port ID:x%06x ",
statep, nrport->port_id);
len += snprintf(buf + len, size - len, "WWPN x%llx ",
nrport->port_name);
len += snprintf(buf + len, size - len, "WWNN x%llx ",
nrport->node_name);
/* An NVME rport can have multiple roles. */
if (nrport->port_role & FC_PORT_ROLE_NVME_INITIATOR)
len += snprintf(buf + len, size - len,
"INITIATOR ");
if (nrport->port_role & FC_PORT_ROLE_NVME_TARGET)
len += snprintf(buf + len, size - len,
"TARGET ");
if (nrport->port_role & FC_PORT_ROLE_NVME_DISCOVERY)
len += snprintf(buf + len, size - len,
"DISCSRVC ");
if (nrport->port_role & ~(FC_PORT_ROLE_NVME_INITIATOR |
FC_PORT_ROLE_NVME_TARGET |
FC_PORT_ROLE_NVME_DISCOVERY))
len += snprintf(buf + len, size - len,
"UNKNOWN ROLE x%x",
nrport->port_role);
/* Terminate the string. */
len += snprintf(buf + len, size - len, "\n");
}
spin_unlock_irq(shost->host_lock);
out_exit:
return len;
}
/**
* lpfc_debugfs_nvmestat_data - Dump target node list to a buffer
* @vport: The vport to gather target node info from.
* @buf: The buffer to dump log into.
* @size: The maximum amount of data to process.
*
* Description:
* This routine dumps the NVME statistics associated with @vport
*
* Return Value:
* This routine returns the amount of bytes that were dumped into @buf and will
* not exceed @size.
**/
static int
lpfc_debugfs_nvmestat_data(struct lpfc_vport *vport, char *buf, int size)
{
struct lpfc_hba *phba = vport->phba;
struct lpfc_nvmet_tgtport *tgtp;
Update ABORT processing for NVMET. The driver with nvme had this routine stubbed. Right now XRI_ABORTED_CQE is not handled and the FC NVMET Transport has a new API for the driver. Missing code path, new NVME abort API Update ABORT processing for NVMET There are 3 new FC NVMET Transport API/ template routines for NVMET: lpfc_nvmet_xmt_fcp_release This NVMET template callback routine called to release context associated with an IO This routine is ALWAYS called last, even if the IO was aborted or completed in error. lpfc_nvmet_xmt_fcp_abort This NVMET template callback routine called to abort an exchange that has an IO in progress nvmet_fc_rcv_fcp_req When the lpfc driver receives an ABTS, this NVME FC transport layer callback routine is called. For this case there are 2 paths thru the driver: the driver either has an outstanding exchange / context for the XRI to be aborted or not. If not, a BA_RJT is issued otherwise a BA_ACC NVMET Driver abort paths: There are 2 paths for aborting an IO. The first one is we receive an IO and decide not to process it because of lack of resources. An unsolicated ABTS is immediately sent back to the initiator as a response. lpfc_nvmet_unsol_fcp_buffer lpfc_nvmet_unsol_issue_abort (XMIT_SEQUENCE_WQE) The second one is we sent the IO up to the NVMET transport layer to process, and for some reason the NVME Transport layer decided to abort the IO before it completes all its phases. For this case there are 2 paths thru the driver: the driver either has an outstanding TSEND/TRECEIVE/TRSP WQE or no outstanding WQEs are present for the exchange / context. lpfc_nvmet_xmt_fcp_abort if (LPFC_NVMET_IO_INP) lpfc_nvmet_sol_fcp_issue_abort (ABORT_WQE) lpfc_nvmet_sol_fcp_abort_cmp else lpfc_nvmet_unsol_fcp_issue_abort lpfc_nvmet_unsol_issue_abort (XMIT_SEQUENCE_WQE) lpfc_nvmet_unsol_fcp_abort_cmp Context flags: LPFC_NVMET_IOP - his flag signifies an IO is in progress on the exchange. LPFC_NVMET_XBUSY - this flag indicates the IO completed but the firmware is still busy with the corresponding exchange. The exchange should not be reused until after a XRI_ABORTED_CQE is received for that exchange. LPFC_NVMET_ABORT_OP - this flag signifies an ABORT_WQE was issued on the exchange. LPFC_NVMET_CTX_RLS - this flag signifies a context free was requested, but we are deferring it due to an XBUSY or ABORT in progress. A ctxlock is added to the context structure that is used whenever these flags are set/read within the context of an IO. The LPFC_NVMET_CTX_RLS flag is only set in the defer_relase routine when the transport has resolved all IO associated with the buffer. The flag is cleared when the CTX is associated with a new IO. An exchange can has both an LPFC_NVMET_XBUSY and a LPFC_NVMET_ABORT_OP condition active simultaneously. Both conditions must complete before the exchange is freed. When the abort callback (lpfc_nvmet_xmt_fcp_abort) is envoked: If there is an outstanding IO, the driver will issue an ABORT_WQE. This should result in 3 completions for the exchange: 1) IO cmpl with XB bit set 2) Abort WQE cmpl 3) XRI_ABORTED_CQE cmpl For this scenerio, after completion #1, the NVMET Transport IO rsp callback is called. After completion #2, no action is taken with respect to the exchange / context. After completion #3, the exchange context is free for re-use on another IO. If there is no outstanding activity on the exchange, the driver will send a ABTS to the Initiator. Upon completion of this WQE, the exchange / context is freed for re-use on another IO. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
2017-04-22 07:05:04 +08:00
struct lpfc_nvmet_rcv_ctx *ctxp, *next_ctxp;
struct nvme_fc_local_port *localport;
struct lpfc_fc4_ctrl_stat *cstat;
struct lpfc_nvme_lport *lport;
uint64_t data1, data2, data3;
uint64_t tot, totin, totout;
int cnt, i;
int len = 0;
if (phba->nvmet_support) {
if (!phba->targetport)
return len;
tgtp = (struct lpfc_nvmet_tgtport *)phba->targetport->private;
Update ABORT processing for NVMET. The driver with nvme had this routine stubbed. Right now XRI_ABORTED_CQE is not handled and the FC NVMET Transport has a new API for the driver. Missing code path, new NVME abort API Update ABORT processing for NVMET There are 3 new FC NVMET Transport API/ template routines for NVMET: lpfc_nvmet_xmt_fcp_release This NVMET template callback routine called to release context associated with an IO This routine is ALWAYS called last, even if the IO was aborted or completed in error. lpfc_nvmet_xmt_fcp_abort This NVMET template callback routine called to abort an exchange that has an IO in progress nvmet_fc_rcv_fcp_req When the lpfc driver receives an ABTS, this NVME FC transport layer callback routine is called. For this case there are 2 paths thru the driver: the driver either has an outstanding exchange / context for the XRI to be aborted or not. If not, a BA_RJT is issued otherwise a BA_ACC NVMET Driver abort paths: There are 2 paths for aborting an IO. The first one is we receive an IO and decide not to process it because of lack of resources. An unsolicated ABTS is immediately sent back to the initiator as a response. lpfc_nvmet_unsol_fcp_buffer lpfc_nvmet_unsol_issue_abort (XMIT_SEQUENCE_WQE) The second one is we sent the IO up to the NVMET transport layer to process, and for some reason the NVME Transport layer decided to abort the IO before it completes all its phases. For this case there are 2 paths thru the driver: the driver either has an outstanding TSEND/TRECEIVE/TRSP WQE or no outstanding WQEs are present for the exchange / context. lpfc_nvmet_xmt_fcp_abort if (LPFC_NVMET_IO_INP) lpfc_nvmet_sol_fcp_issue_abort (ABORT_WQE) lpfc_nvmet_sol_fcp_abort_cmp else lpfc_nvmet_unsol_fcp_issue_abort lpfc_nvmet_unsol_issue_abort (XMIT_SEQUENCE_WQE) lpfc_nvmet_unsol_fcp_abort_cmp Context flags: LPFC_NVMET_IOP - his flag signifies an IO is in progress on the exchange. LPFC_NVMET_XBUSY - this flag indicates the IO completed but the firmware is still busy with the corresponding exchange. The exchange should not be reused until after a XRI_ABORTED_CQE is received for that exchange. LPFC_NVMET_ABORT_OP - this flag signifies an ABORT_WQE was issued on the exchange. LPFC_NVMET_CTX_RLS - this flag signifies a context free was requested, but we are deferring it due to an XBUSY or ABORT in progress. A ctxlock is added to the context structure that is used whenever these flags are set/read within the context of an IO. The LPFC_NVMET_CTX_RLS flag is only set in the defer_relase routine when the transport has resolved all IO associated with the buffer. The flag is cleared when the CTX is associated with a new IO. An exchange can has both an LPFC_NVMET_XBUSY and a LPFC_NVMET_ABORT_OP condition active simultaneously. Both conditions must complete before the exchange is freed. When the abort callback (lpfc_nvmet_xmt_fcp_abort) is envoked: If there is an outstanding IO, the driver will issue an ABORT_WQE. This should result in 3 completions for the exchange: 1) IO cmpl with XB bit set 2) Abort WQE cmpl 3) XRI_ABORTED_CQE cmpl For this scenerio, after completion #1, the NVMET Transport IO rsp callback is called. After completion #2, no action is taken with respect to the exchange / context. After completion #3, the exchange context is free for re-use on another IO. If there is no outstanding activity on the exchange, the driver will send a ABTS to the Initiator. Upon completion of this WQE, the exchange / context is freed for re-use on another IO. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
2017-04-22 07:05:04 +08:00
len += snprintf(buf + len, size - len,
"\nNVME Targetport Statistics\n");
Update ABORT processing for NVMET. The driver with nvme had this routine stubbed. Right now XRI_ABORTED_CQE is not handled and the FC NVMET Transport has a new API for the driver. Missing code path, new NVME abort API Update ABORT processing for NVMET There are 3 new FC NVMET Transport API/ template routines for NVMET: lpfc_nvmet_xmt_fcp_release This NVMET template callback routine called to release context associated with an IO This routine is ALWAYS called last, even if the IO was aborted or completed in error. lpfc_nvmet_xmt_fcp_abort This NVMET template callback routine called to abort an exchange that has an IO in progress nvmet_fc_rcv_fcp_req When the lpfc driver receives an ABTS, this NVME FC transport layer callback routine is called. For this case there are 2 paths thru the driver: the driver either has an outstanding exchange / context for the XRI to be aborted or not. If not, a BA_RJT is issued otherwise a BA_ACC NVMET Driver abort paths: There are 2 paths for aborting an IO. The first one is we receive an IO and decide not to process it because of lack of resources. An unsolicated ABTS is immediately sent back to the initiator as a response. lpfc_nvmet_unsol_fcp_buffer lpfc_nvmet_unsol_issue_abort (XMIT_SEQUENCE_WQE) The second one is we sent the IO up to the NVMET transport layer to process, and for some reason the NVME Transport layer decided to abort the IO before it completes all its phases. For this case there are 2 paths thru the driver: the driver either has an outstanding TSEND/TRECEIVE/TRSP WQE or no outstanding WQEs are present for the exchange / context. lpfc_nvmet_xmt_fcp_abort if (LPFC_NVMET_IO_INP) lpfc_nvmet_sol_fcp_issue_abort (ABORT_WQE) lpfc_nvmet_sol_fcp_abort_cmp else lpfc_nvmet_unsol_fcp_issue_abort lpfc_nvmet_unsol_issue_abort (XMIT_SEQUENCE_WQE) lpfc_nvmet_unsol_fcp_abort_cmp Context flags: LPFC_NVMET_IOP - his flag signifies an IO is in progress on the exchange. LPFC_NVMET_XBUSY - this flag indicates the IO completed but the firmware is still busy with the corresponding exchange. The exchange should not be reused until after a XRI_ABORTED_CQE is received for that exchange. LPFC_NVMET_ABORT_OP - this flag signifies an ABORT_WQE was issued on the exchange. LPFC_NVMET_CTX_RLS - this flag signifies a context free was requested, but we are deferring it due to an XBUSY or ABORT in progress. A ctxlock is added to the context structure that is used whenever these flags are set/read within the context of an IO. The LPFC_NVMET_CTX_RLS flag is only set in the defer_relase routine when the transport has resolved all IO associated with the buffer. The flag is cleared when the CTX is associated with a new IO. An exchange can has both an LPFC_NVMET_XBUSY and a LPFC_NVMET_ABORT_OP condition active simultaneously. Both conditions must complete before the exchange is freed. When the abort callback (lpfc_nvmet_xmt_fcp_abort) is envoked: If there is an outstanding IO, the driver will issue an ABORT_WQE. This should result in 3 completions for the exchange: 1) IO cmpl with XB bit set 2) Abort WQE cmpl 3) XRI_ABORTED_CQE cmpl For this scenerio, after completion #1, the NVMET Transport IO rsp callback is called. After completion #2, no action is taken with respect to the exchange / context. After completion #3, the exchange context is free for re-use on another IO. If there is no outstanding activity on the exchange, the driver will send a ABTS to the Initiator. Upon completion of this WQE, the exchange / context is freed for re-use on another IO. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
2017-04-22 07:05:04 +08:00
len += snprintf(buf + len, size - len,
"LS: Rcv %08x Drop %08x Abort %08x\n",
atomic_read(&tgtp->rcv_ls_req_in),
atomic_read(&tgtp->rcv_ls_req_drop),
atomic_read(&tgtp->xmt_ls_abort));
if (atomic_read(&tgtp->rcv_ls_req_in) !=
atomic_read(&tgtp->rcv_ls_req_out)) {
Update ABORT processing for NVMET. The driver with nvme had this routine stubbed. Right now XRI_ABORTED_CQE is not handled and the FC NVMET Transport has a new API for the driver. Missing code path, new NVME abort API Update ABORT processing for NVMET There are 3 new FC NVMET Transport API/ template routines for NVMET: lpfc_nvmet_xmt_fcp_release This NVMET template callback routine called to release context associated with an IO This routine is ALWAYS called last, even if the IO was aborted or completed in error. lpfc_nvmet_xmt_fcp_abort This NVMET template callback routine called to abort an exchange that has an IO in progress nvmet_fc_rcv_fcp_req When the lpfc driver receives an ABTS, this NVME FC transport layer callback routine is called. For this case there are 2 paths thru the driver: the driver either has an outstanding exchange / context for the XRI to be aborted or not. If not, a BA_RJT is issued otherwise a BA_ACC NVMET Driver abort paths: There are 2 paths for aborting an IO. The first one is we receive an IO and decide not to process it because of lack of resources. An unsolicated ABTS is immediately sent back to the initiator as a response. lpfc_nvmet_unsol_fcp_buffer lpfc_nvmet_unsol_issue_abort (XMIT_SEQUENCE_WQE) The second one is we sent the IO up to the NVMET transport layer to process, and for some reason the NVME Transport layer decided to abort the IO before it completes all its phases. For this case there are 2 paths thru the driver: the driver either has an outstanding TSEND/TRECEIVE/TRSP WQE or no outstanding WQEs are present for the exchange / context. lpfc_nvmet_xmt_fcp_abort if (LPFC_NVMET_IO_INP) lpfc_nvmet_sol_fcp_issue_abort (ABORT_WQE) lpfc_nvmet_sol_fcp_abort_cmp else lpfc_nvmet_unsol_fcp_issue_abort lpfc_nvmet_unsol_issue_abort (XMIT_SEQUENCE_WQE) lpfc_nvmet_unsol_fcp_abort_cmp Context flags: LPFC_NVMET_IOP - his flag signifies an IO is in progress on the exchange. LPFC_NVMET_XBUSY - this flag indicates the IO completed but the firmware is still busy with the corresponding exchange. The exchange should not be reused until after a XRI_ABORTED_CQE is received for that exchange. LPFC_NVMET_ABORT_OP - this flag signifies an ABORT_WQE was issued on the exchange. LPFC_NVMET_CTX_RLS - this flag signifies a context free was requested, but we are deferring it due to an XBUSY or ABORT in progress. A ctxlock is added to the context structure that is used whenever these flags are set/read within the context of an IO. The LPFC_NVMET_CTX_RLS flag is only set in the defer_relase routine when the transport has resolved all IO associated with the buffer. The flag is cleared when the CTX is associated with a new IO. An exchange can has both an LPFC_NVMET_XBUSY and a LPFC_NVMET_ABORT_OP condition active simultaneously. Both conditions must complete before the exchange is freed. When the abort callback (lpfc_nvmet_xmt_fcp_abort) is envoked: If there is an outstanding IO, the driver will issue an ABORT_WQE. This should result in 3 completions for the exchange: 1) IO cmpl with XB bit set 2) Abort WQE cmpl 3) XRI_ABORTED_CQE cmpl For this scenerio, after completion #1, the NVMET Transport IO rsp callback is called. After completion #2, no action is taken with respect to the exchange / context. After completion #3, the exchange context is free for re-use on another IO. If there is no outstanding activity on the exchange, the driver will send a ABTS to the Initiator. Upon completion of this WQE, the exchange / context is freed for re-use on another IO. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
2017-04-22 07:05:04 +08:00
len += snprintf(buf + len, size - len,
"Rcv LS: in %08x != out %08x\n",
atomic_read(&tgtp->rcv_ls_req_in),
atomic_read(&tgtp->rcv_ls_req_out));
}
Update ABORT processing for NVMET. The driver with nvme had this routine stubbed. Right now XRI_ABORTED_CQE is not handled and the FC NVMET Transport has a new API for the driver. Missing code path, new NVME abort API Update ABORT processing for NVMET There are 3 new FC NVMET Transport API/ template routines for NVMET: lpfc_nvmet_xmt_fcp_release This NVMET template callback routine called to release context associated with an IO This routine is ALWAYS called last, even if the IO was aborted or completed in error. lpfc_nvmet_xmt_fcp_abort This NVMET template callback routine called to abort an exchange that has an IO in progress nvmet_fc_rcv_fcp_req When the lpfc driver receives an ABTS, this NVME FC transport layer callback routine is called. For this case there are 2 paths thru the driver: the driver either has an outstanding exchange / context for the XRI to be aborted or not. If not, a BA_RJT is issued otherwise a BA_ACC NVMET Driver abort paths: There are 2 paths for aborting an IO. The first one is we receive an IO and decide not to process it because of lack of resources. An unsolicated ABTS is immediately sent back to the initiator as a response. lpfc_nvmet_unsol_fcp_buffer lpfc_nvmet_unsol_issue_abort (XMIT_SEQUENCE_WQE) The second one is we sent the IO up to the NVMET transport layer to process, and for some reason the NVME Transport layer decided to abort the IO before it completes all its phases. For this case there are 2 paths thru the driver: the driver either has an outstanding TSEND/TRECEIVE/TRSP WQE or no outstanding WQEs are present for the exchange / context. lpfc_nvmet_xmt_fcp_abort if (LPFC_NVMET_IO_INP) lpfc_nvmet_sol_fcp_issue_abort (ABORT_WQE) lpfc_nvmet_sol_fcp_abort_cmp else lpfc_nvmet_unsol_fcp_issue_abort lpfc_nvmet_unsol_issue_abort (XMIT_SEQUENCE_WQE) lpfc_nvmet_unsol_fcp_abort_cmp Context flags: LPFC_NVMET_IOP - his flag signifies an IO is in progress on the exchange. LPFC_NVMET_XBUSY - this flag indicates the IO completed but the firmware is still busy with the corresponding exchange. The exchange should not be reused until after a XRI_ABORTED_CQE is received for that exchange. LPFC_NVMET_ABORT_OP - this flag signifies an ABORT_WQE was issued on the exchange. LPFC_NVMET_CTX_RLS - this flag signifies a context free was requested, but we are deferring it due to an XBUSY or ABORT in progress. A ctxlock is added to the context structure that is used whenever these flags are set/read within the context of an IO. The LPFC_NVMET_CTX_RLS flag is only set in the defer_relase routine when the transport has resolved all IO associated with the buffer. The flag is cleared when the CTX is associated with a new IO. An exchange can has both an LPFC_NVMET_XBUSY and a LPFC_NVMET_ABORT_OP condition active simultaneously. Both conditions must complete before the exchange is freed. When the abort callback (lpfc_nvmet_xmt_fcp_abort) is envoked: If there is an outstanding IO, the driver will issue an ABORT_WQE. This should result in 3 completions for the exchange: 1) IO cmpl with XB bit set 2) Abort WQE cmpl 3) XRI_ABORTED_CQE cmpl For this scenerio, after completion #1, the NVMET Transport IO rsp callback is called. After completion #2, no action is taken with respect to the exchange / context. After completion #3, the exchange context is free for re-use on another IO. If there is no outstanding activity on the exchange, the driver will send a ABTS to the Initiator. Upon completion of this WQE, the exchange / context is freed for re-use on another IO. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
2017-04-22 07:05:04 +08:00
len += snprintf(buf + len, size - len,
"LS: Xmt %08x Drop %08x Cmpl %08x\n",
atomic_read(&tgtp->xmt_ls_rsp),
atomic_read(&tgtp->xmt_ls_drop),
atomic_read(&tgtp->xmt_ls_rsp_cmpl));
len += snprintf(buf + len, size - len,
"LS: RSP Abort %08x xb %08x Err %08x\n",
atomic_read(&tgtp->xmt_ls_rsp_aborted),
atomic_read(&tgtp->xmt_ls_rsp_xb_set),
atomic_read(&tgtp->xmt_ls_rsp_error));
Update ABORT processing for NVMET. The driver with nvme had this routine stubbed. Right now XRI_ABORTED_CQE is not handled and the FC NVMET Transport has a new API for the driver. Missing code path, new NVME abort API Update ABORT processing for NVMET There are 3 new FC NVMET Transport API/ template routines for NVMET: lpfc_nvmet_xmt_fcp_release This NVMET template callback routine called to release context associated with an IO This routine is ALWAYS called last, even if the IO was aborted or completed in error. lpfc_nvmet_xmt_fcp_abort This NVMET template callback routine called to abort an exchange that has an IO in progress nvmet_fc_rcv_fcp_req When the lpfc driver receives an ABTS, this NVME FC transport layer callback routine is called. For this case there are 2 paths thru the driver: the driver either has an outstanding exchange / context for the XRI to be aborted or not. If not, a BA_RJT is issued otherwise a BA_ACC NVMET Driver abort paths: There are 2 paths for aborting an IO. The first one is we receive an IO and decide not to process it because of lack of resources. An unsolicated ABTS is immediately sent back to the initiator as a response. lpfc_nvmet_unsol_fcp_buffer lpfc_nvmet_unsol_issue_abort (XMIT_SEQUENCE_WQE) The second one is we sent the IO up to the NVMET transport layer to process, and for some reason the NVME Transport layer decided to abort the IO before it completes all its phases. For this case there are 2 paths thru the driver: the driver either has an outstanding TSEND/TRECEIVE/TRSP WQE or no outstanding WQEs are present for the exchange / context. lpfc_nvmet_xmt_fcp_abort if (LPFC_NVMET_IO_INP) lpfc_nvmet_sol_fcp_issue_abort (ABORT_WQE) lpfc_nvmet_sol_fcp_abort_cmp else lpfc_nvmet_unsol_fcp_issue_abort lpfc_nvmet_unsol_issue_abort (XMIT_SEQUENCE_WQE) lpfc_nvmet_unsol_fcp_abort_cmp Context flags: LPFC_NVMET_IOP - his flag signifies an IO is in progress on the exchange. LPFC_NVMET_XBUSY - this flag indicates the IO completed but the firmware is still busy with the corresponding exchange. The exchange should not be reused until after a XRI_ABORTED_CQE is received for that exchange. LPFC_NVMET_ABORT_OP - this flag signifies an ABORT_WQE was issued on the exchange. LPFC_NVMET_CTX_RLS - this flag signifies a context free was requested, but we are deferring it due to an XBUSY or ABORT in progress. A ctxlock is added to the context structure that is used whenever these flags are set/read within the context of an IO. The LPFC_NVMET_CTX_RLS flag is only set in the defer_relase routine when the transport has resolved all IO associated with the buffer. The flag is cleared when the CTX is associated with a new IO. An exchange can has both an LPFC_NVMET_XBUSY and a LPFC_NVMET_ABORT_OP condition active simultaneously. Both conditions must complete before the exchange is freed. When the abort callback (lpfc_nvmet_xmt_fcp_abort) is envoked: If there is an outstanding IO, the driver will issue an ABORT_WQE. This should result in 3 completions for the exchange: 1) IO cmpl with XB bit set 2) Abort WQE cmpl 3) XRI_ABORTED_CQE cmpl For this scenerio, after completion #1, the NVMET Transport IO rsp callback is called. After completion #2, no action is taken with respect to the exchange / context. After completion #3, the exchange context is free for re-use on another IO. If there is no outstanding activity on the exchange, the driver will send a ABTS to the Initiator. Upon completion of this WQE, the exchange / context is freed for re-use on another IO. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
2017-04-22 07:05:04 +08:00
len += snprintf(buf + len, size - len,
"FCP: Rcv %08x Defer %08x Release %08x "
"Drop %08x\n",
atomic_read(&tgtp->rcv_fcp_cmd_in),
atomic_read(&tgtp->rcv_fcp_cmd_defer),
atomic_read(&tgtp->xmt_fcp_release),
atomic_read(&tgtp->rcv_fcp_cmd_drop));
if (atomic_read(&tgtp->rcv_fcp_cmd_in) !=
atomic_read(&tgtp->rcv_fcp_cmd_out)) {
Update ABORT processing for NVMET. The driver with nvme had this routine stubbed. Right now XRI_ABORTED_CQE is not handled and the FC NVMET Transport has a new API for the driver. Missing code path, new NVME abort API Update ABORT processing for NVMET There are 3 new FC NVMET Transport API/ template routines for NVMET: lpfc_nvmet_xmt_fcp_release This NVMET template callback routine called to release context associated with an IO This routine is ALWAYS called last, even if the IO was aborted or completed in error. lpfc_nvmet_xmt_fcp_abort This NVMET template callback routine called to abort an exchange that has an IO in progress nvmet_fc_rcv_fcp_req When the lpfc driver receives an ABTS, this NVME FC transport layer callback routine is called. For this case there are 2 paths thru the driver: the driver either has an outstanding exchange / context for the XRI to be aborted or not. If not, a BA_RJT is issued otherwise a BA_ACC NVMET Driver abort paths: There are 2 paths for aborting an IO. The first one is we receive an IO and decide not to process it because of lack of resources. An unsolicated ABTS is immediately sent back to the initiator as a response. lpfc_nvmet_unsol_fcp_buffer lpfc_nvmet_unsol_issue_abort (XMIT_SEQUENCE_WQE) The second one is we sent the IO up to the NVMET transport layer to process, and for some reason the NVME Transport layer decided to abort the IO before it completes all its phases. For this case there are 2 paths thru the driver: the driver either has an outstanding TSEND/TRECEIVE/TRSP WQE or no outstanding WQEs are present for the exchange / context. lpfc_nvmet_xmt_fcp_abort if (LPFC_NVMET_IO_INP) lpfc_nvmet_sol_fcp_issue_abort (ABORT_WQE) lpfc_nvmet_sol_fcp_abort_cmp else lpfc_nvmet_unsol_fcp_issue_abort lpfc_nvmet_unsol_issue_abort (XMIT_SEQUENCE_WQE) lpfc_nvmet_unsol_fcp_abort_cmp Context flags: LPFC_NVMET_IOP - his flag signifies an IO is in progress on the exchange. LPFC_NVMET_XBUSY - this flag indicates the IO completed but the firmware is still busy with the corresponding exchange. The exchange should not be reused until after a XRI_ABORTED_CQE is received for that exchange. LPFC_NVMET_ABORT_OP - this flag signifies an ABORT_WQE was issued on the exchange. LPFC_NVMET_CTX_RLS - this flag signifies a context free was requested, but we are deferring it due to an XBUSY or ABORT in progress. A ctxlock is added to the context structure that is used whenever these flags are set/read within the context of an IO. The LPFC_NVMET_CTX_RLS flag is only set in the defer_relase routine when the transport has resolved all IO associated with the buffer. The flag is cleared when the CTX is associated with a new IO. An exchange can has both an LPFC_NVMET_XBUSY and a LPFC_NVMET_ABORT_OP condition active simultaneously. Both conditions must complete before the exchange is freed. When the abort callback (lpfc_nvmet_xmt_fcp_abort) is envoked: If there is an outstanding IO, the driver will issue an ABORT_WQE. This should result in 3 completions for the exchange: 1) IO cmpl with XB bit set 2) Abort WQE cmpl 3) XRI_ABORTED_CQE cmpl For this scenerio, after completion #1, the NVMET Transport IO rsp callback is called. After completion #2, no action is taken with respect to the exchange / context. After completion #3, the exchange context is free for re-use on another IO. If there is no outstanding activity on the exchange, the driver will send a ABTS to the Initiator. Upon completion of this WQE, the exchange / context is freed for re-use on another IO. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
2017-04-22 07:05:04 +08:00
len += snprintf(buf + len, size - len,
"Rcv FCP: in %08x != out %08x\n",
atomic_read(&tgtp->rcv_fcp_cmd_in),
atomic_read(&tgtp->rcv_fcp_cmd_out));
}
Update ABORT processing for NVMET. The driver with nvme had this routine stubbed. Right now XRI_ABORTED_CQE is not handled and the FC NVMET Transport has a new API for the driver. Missing code path, new NVME abort API Update ABORT processing for NVMET There are 3 new FC NVMET Transport API/ template routines for NVMET: lpfc_nvmet_xmt_fcp_release This NVMET template callback routine called to release context associated with an IO This routine is ALWAYS called last, even if the IO was aborted or completed in error. lpfc_nvmet_xmt_fcp_abort This NVMET template callback routine called to abort an exchange that has an IO in progress nvmet_fc_rcv_fcp_req When the lpfc driver receives an ABTS, this NVME FC transport layer callback routine is called. For this case there are 2 paths thru the driver: the driver either has an outstanding exchange / context for the XRI to be aborted or not. If not, a BA_RJT is issued otherwise a BA_ACC NVMET Driver abort paths: There are 2 paths for aborting an IO. The first one is we receive an IO and decide not to process it because of lack of resources. An unsolicated ABTS is immediately sent back to the initiator as a response. lpfc_nvmet_unsol_fcp_buffer lpfc_nvmet_unsol_issue_abort (XMIT_SEQUENCE_WQE) The second one is we sent the IO up to the NVMET transport layer to process, and for some reason the NVME Transport layer decided to abort the IO before it completes all its phases. For this case there are 2 paths thru the driver: the driver either has an outstanding TSEND/TRECEIVE/TRSP WQE or no outstanding WQEs are present for the exchange / context. lpfc_nvmet_xmt_fcp_abort if (LPFC_NVMET_IO_INP) lpfc_nvmet_sol_fcp_issue_abort (ABORT_WQE) lpfc_nvmet_sol_fcp_abort_cmp else lpfc_nvmet_unsol_fcp_issue_abort lpfc_nvmet_unsol_issue_abort (XMIT_SEQUENCE_WQE) lpfc_nvmet_unsol_fcp_abort_cmp Context flags: LPFC_NVMET_IOP - his flag signifies an IO is in progress on the exchange. LPFC_NVMET_XBUSY - this flag indicates the IO completed but the firmware is still busy with the corresponding exchange. The exchange should not be reused until after a XRI_ABORTED_CQE is received for that exchange. LPFC_NVMET_ABORT_OP - this flag signifies an ABORT_WQE was issued on the exchange. LPFC_NVMET_CTX_RLS - this flag signifies a context free was requested, but we are deferring it due to an XBUSY or ABORT in progress. A ctxlock is added to the context structure that is used whenever these flags are set/read within the context of an IO. The LPFC_NVMET_CTX_RLS flag is only set in the defer_relase routine when the transport has resolved all IO associated with the buffer. The flag is cleared when the CTX is associated with a new IO. An exchange can has both an LPFC_NVMET_XBUSY and a LPFC_NVMET_ABORT_OP condition active simultaneously. Both conditions must complete before the exchange is freed. When the abort callback (lpfc_nvmet_xmt_fcp_abort) is envoked: If there is an outstanding IO, the driver will issue an ABORT_WQE. This should result in 3 completions for the exchange: 1) IO cmpl with XB bit set 2) Abort WQE cmpl 3) XRI_ABORTED_CQE cmpl For this scenerio, after completion #1, the NVMET Transport IO rsp callback is called. After completion #2, no action is taken with respect to the exchange / context. After completion #3, the exchange context is free for re-use on another IO. If there is no outstanding activity on the exchange, the driver will send a ABTS to the Initiator. Upon completion of this WQE, the exchange / context is freed for re-use on another IO. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
2017-04-22 07:05:04 +08:00
len += snprintf(buf + len, size - len,
"FCP Rsp: read %08x readrsp %08x "
"write %08x rsp %08x\n",
atomic_read(&tgtp->xmt_fcp_read),
atomic_read(&tgtp->xmt_fcp_read_rsp),
atomic_read(&tgtp->xmt_fcp_write),
atomic_read(&tgtp->xmt_fcp_rsp));
Update ABORT processing for NVMET. The driver with nvme had this routine stubbed. Right now XRI_ABORTED_CQE is not handled and the FC NVMET Transport has a new API for the driver. Missing code path, new NVME abort API Update ABORT processing for NVMET There are 3 new FC NVMET Transport API/ template routines for NVMET: lpfc_nvmet_xmt_fcp_release This NVMET template callback routine called to release context associated with an IO This routine is ALWAYS called last, even if the IO was aborted or completed in error. lpfc_nvmet_xmt_fcp_abort This NVMET template callback routine called to abort an exchange that has an IO in progress nvmet_fc_rcv_fcp_req When the lpfc driver receives an ABTS, this NVME FC transport layer callback routine is called. For this case there are 2 paths thru the driver: the driver either has an outstanding exchange / context for the XRI to be aborted or not. If not, a BA_RJT is issued otherwise a BA_ACC NVMET Driver abort paths: There are 2 paths for aborting an IO. The first one is we receive an IO and decide not to process it because of lack of resources. An unsolicated ABTS is immediately sent back to the initiator as a response. lpfc_nvmet_unsol_fcp_buffer lpfc_nvmet_unsol_issue_abort (XMIT_SEQUENCE_WQE) The second one is we sent the IO up to the NVMET transport layer to process, and for some reason the NVME Transport layer decided to abort the IO before it completes all its phases. For this case there are 2 paths thru the driver: the driver either has an outstanding TSEND/TRECEIVE/TRSP WQE or no outstanding WQEs are present for the exchange / context. lpfc_nvmet_xmt_fcp_abort if (LPFC_NVMET_IO_INP) lpfc_nvmet_sol_fcp_issue_abort (ABORT_WQE) lpfc_nvmet_sol_fcp_abort_cmp else lpfc_nvmet_unsol_fcp_issue_abort lpfc_nvmet_unsol_issue_abort (XMIT_SEQUENCE_WQE) lpfc_nvmet_unsol_fcp_abort_cmp Context flags: LPFC_NVMET_IOP - his flag signifies an IO is in progress on the exchange. LPFC_NVMET_XBUSY - this flag indicates the IO completed but the firmware is still busy with the corresponding exchange. The exchange should not be reused until after a XRI_ABORTED_CQE is received for that exchange. LPFC_NVMET_ABORT_OP - this flag signifies an ABORT_WQE was issued on the exchange. LPFC_NVMET_CTX_RLS - this flag signifies a context free was requested, but we are deferring it due to an XBUSY or ABORT in progress. A ctxlock is added to the context structure that is used whenever these flags are set/read within the context of an IO. The LPFC_NVMET_CTX_RLS flag is only set in the defer_relase routine when the transport has resolved all IO associated with the buffer. The flag is cleared when the CTX is associated with a new IO. An exchange can has both an LPFC_NVMET_XBUSY and a LPFC_NVMET_ABORT_OP condition active simultaneously. Both conditions must complete before the exchange is freed. When the abort callback (lpfc_nvmet_xmt_fcp_abort) is envoked: If there is an outstanding IO, the driver will issue an ABORT_WQE. This should result in 3 completions for the exchange: 1) IO cmpl with XB bit set 2) Abort WQE cmpl 3) XRI_ABORTED_CQE cmpl For this scenerio, after completion #1, the NVMET Transport IO rsp callback is called. After completion #2, no action is taken with respect to the exchange / context. After completion #3, the exchange context is free for re-use on another IO. If there is no outstanding activity on the exchange, the driver will send a ABTS to the Initiator. Upon completion of this WQE, the exchange / context is freed for re-use on another IO. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
2017-04-22 07:05:04 +08:00
len += snprintf(buf + len, size - len,
"FCP Rsp Cmpl: %08x err %08x drop %08x\n",
atomic_read(&tgtp->xmt_fcp_rsp_cmpl),
atomic_read(&tgtp->xmt_fcp_rsp_error),
atomic_read(&tgtp->xmt_fcp_rsp_drop));
len += snprintf(buf + len, size - len,
"FCP Rsp Abort: %08x xb %08x xricqe %08x\n",
atomic_read(&tgtp->xmt_fcp_rsp_aborted),
atomic_read(&tgtp->xmt_fcp_rsp_xb_set),
atomic_read(&tgtp->xmt_fcp_xri_abort_cqe));
Update ABORT processing for NVMET. The driver with nvme had this routine stubbed. Right now XRI_ABORTED_CQE is not handled and the FC NVMET Transport has a new API for the driver. Missing code path, new NVME abort API Update ABORT processing for NVMET There are 3 new FC NVMET Transport API/ template routines for NVMET: lpfc_nvmet_xmt_fcp_release This NVMET template callback routine called to release context associated with an IO This routine is ALWAYS called last, even if the IO was aborted or completed in error. lpfc_nvmet_xmt_fcp_abort This NVMET template callback routine called to abort an exchange that has an IO in progress nvmet_fc_rcv_fcp_req When the lpfc driver receives an ABTS, this NVME FC transport layer callback routine is called. For this case there are 2 paths thru the driver: the driver either has an outstanding exchange / context for the XRI to be aborted or not. If not, a BA_RJT is issued otherwise a BA_ACC NVMET Driver abort paths: There are 2 paths for aborting an IO. The first one is we receive an IO and decide not to process it because of lack of resources. An unsolicated ABTS is immediately sent back to the initiator as a response. lpfc_nvmet_unsol_fcp_buffer lpfc_nvmet_unsol_issue_abort (XMIT_SEQUENCE_WQE) The second one is we sent the IO up to the NVMET transport layer to process, and for some reason the NVME Transport layer decided to abort the IO before it completes all its phases. For this case there are 2 paths thru the driver: the driver either has an outstanding TSEND/TRECEIVE/TRSP WQE or no outstanding WQEs are present for the exchange / context. lpfc_nvmet_xmt_fcp_abort if (LPFC_NVMET_IO_INP) lpfc_nvmet_sol_fcp_issue_abort (ABORT_WQE) lpfc_nvmet_sol_fcp_abort_cmp else lpfc_nvmet_unsol_fcp_issue_abort lpfc_nvmet_unsol_issue_abort (XMIT_SEQUENCE_WQE) lpfc_nvmet_unsol_fcp_abort_cmp Context flags: LPFC_NVMET_IOP - his flag signifies an IO is in progress on the exchange. LPFC_NVMET_XBUSY - this flag indicates the IO completed but the firmware is still busy with the corresponding exchange. The exchange should not be reused until after a XRI_ABORTED_CQE is received for that exchange. LPFC_NVMET_ABORT_OP - this flag signifies an ABORT_WQE was issued on the exchange. LPFC_NVMET_CTX_RLS - this flag signifies a context free was requested, but we are deferring it due to an XBUSY or ABORT in progress. A ctxlock is added to the context structure that is used whenever these flags are set/read within the context of an IO. The LPFC_NVMET_CTX_RLS flag is only set in the defer_relase routine when the transport has resolved all IO associated with the buffer. The flag is cleared when the CTX is associated with a new IO. An exchange can has both an LPFC_NVMET_XBUSY and a LPFC_NVMET_ABORT_OP condition active simultaneously. Both conditions must complete before the exchange is freed. When the abort callback (lpfc_nvmet_xmt_fcp_abort) is envoked: If there is an outstanding IO, the driver will issue an ABORT_WQE. This should result in 3 completions for the exchange: 1) IO cmpl with XB bit set 2) Abort WQE cmpl 3) XRI_ABORTED_CQE cmpl For this scenerio, after completion #1, the NVMET Transport IO rsp callback is called. After completion #2, no action is taken with respect to the exchange / context. After completion #3, the exchange context is free for re-use on another IO. If there is no outstanding activity on the exchange, the driver will send a ABTS to the Initiator. Upon completion of this WQE, the exchange / context is freed for re-use on another IO. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
2017-04-22 07:05:04 +08:00
len += snprintf(buf + len, size - len,
"ABORT: Xmt %08x Cmpl %08x\n",
atomic_read(&tgtp->xmt_fcp_abort),
atomic_read(&tgtp->xmt_fcp_abort_cmpl));
len += snprintf(buf + len, size - len,
"ABORT: Sol %08x Usol %08x Err %08x Cmpl %08x",
atomic_read(&tgtp->xmt_abort_sol),
atomic_read(&tgtp->xmt_abort_unsol),
atomic_read(&tgtp->xmt_abort_rsp),
atomic_read(&tgtp->xmt_abort_rsp_error));
Update ABORT processing for NVMET. The driver with nvme had this routine stubbed. Right now XRI_ABORTED_CQE is not handled and the FC NVMET Transport has a new API for the driver. Missing code path, new NVME abort API Update ABORT processing for NVMET There are 3 new FC NVMET Transport API/ template routines for NVMET: lpfc_nvmet_xmt_fcp_release This NVMET template callback routine called to release context associated with an IO This routine is ALWAYS called last, even if the IO was aborted or completed in error. lpfc_nvmet_xmt_fcp_abort This NVMET template callback routine called to abort an exchange that has an IO in progress nvmet_fc_rcv_fcp_req When the lpfc driver receives an ABTS, this NVME FC transport layer callback routine is called. For this case there are 2 paths thru the driver: the driver either has an outstanding exchange / context for the XRI to be aborted or not. If not, a BA_RJT is issued otherwise a BA_ACC NVMET Driver abort paths: There are 2 paths for aborting an IO. The first one is we receive an IO and decide not to process it because of lack of resources. An unsolicated ABTS is immediately sent back to the initiator as a response. lpfc_nvmet_unsol_fcp_buffer lpfc_nvmet_unsol_issue_abort (XMIT_SEQUENCE_WQE) The second one is we sent the IO up to the NVMET transport layer to process, and for some reason the NVME Transport layer decided to abort the IO before it completes all its phases. For this case there are 2 paths thru the driver: the driver either has an outstanding TSEND/TRECEIVE/TRSP WQE or no outstanding WQEs are present for the exchange / context. lpfc_nvmet_xmt_fcp_abort if (LPFC_NVMET_IO_INP) lpfc_nvmet_sol_fcp_issue_abort (ABORT_WQE) lpfc_nvmet_sol_fcp_abort_cmp else lpfc_nvmet_unsol_fcp_issue_abort lpfc_nvmet_unsol_issue_abort (XMIT_SEQUENCE_WQE) lpfc_nvmet_unsol_fcp_abort_cmp Context flags: LPFC_NVMET_IOP - his flag signifies an IO is in progress on the exchange. LPFC_NVMET_XBUSY - this flag indicates the IO completed but the firmware is still busy with the corresponding exchange. The exchange should not be reused until after a XRI_ABORTED_CQE is received for that exchange. LPFC_NVMET_ABORT_OP - this flag signifies an ABORT_WQE was issued on the exchange. LPFC_NVMET_CTX_RLS - this flag signifies a context free was requested, but we are deferring it due to an XBUSY or ABORT in progress. A ctxlock is added to the context structure that is used whenever these flags are set/read within the context of an IO. The LPFC_NVMET_CTX_RLS flag is only set in the defer_relase routine when the transport has resolved all IO associated with the buffer. The flag is cleared when the CTX is associated with a new IO. An exchange can has both an LPFC_NVMET_XBUSY and a LPFC_NVMET_ABORT_OP condition active simultaneously. Both conditions must complete before the exchange is freed. When the abort callback (lpfc_nvmet_xmt_fcp_abort) is envoked: If there is an outstanding IO, the driver will issue an ABORT_WQE. This should result in 3 completions for the exchange: 1) IO cmpl with XB bit set 2) Abort WQE cmpl 3) XRI_ABORTED_CQE cmpl For this scenerio, after completion #1, the NVMET Transport IO rsp callback is called. After completion #2, no action is taken with respect to the exchange / context. After completion #3, the exchange context is free for re-use on another IO. If there is no outstanding activity on the exchange, the driver will send a ABTS to the Initiator. Upon completion of this WQE, the exchange / context is freed for re-use on another IO. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
2017-04-22 07:05:04 +08:00
len += snprintf(buf + len, size - len, "\n");
cnt = 0;
spin_lock(&phba->sli4_hba.abts_nvmet_buf_list_lock);
Update ABORT processing for NVMET. The driver with nvme had this routine stubbed. Right now XRI_ABORTED_CQE is not handled and the FC NVMET Transport has a new API for the driver. Missing code path, new NVME abort API Update ABORT processing for NVMET There are 3 new FC NVMET Transport API/ template routines for NVMET: lpfc_nvmet_xmt_fcp_release This NVMET template callback routine called to release context associated with an IO This routine is ALWAYS called last, even if the IO was aborted or completed in error. lpfc_nvmet_xmt_fcp_abort This NVMET template callback routine called to abort an exchange that has an IO in progress nvmet_fc_rcv_fcp_req When the lpfc driver receives an ABTS, this NVME FC transport layer callback routine is called. For this case there are 2 paths thru the driver: the driver either has an outstanding exchange / context for the XRI to be aborted or not. If not, a BA_RJT is issued otherwise a BA_ACC NVMET Driver abort paths: There are 2 paths for aborting an IO. The first one is we receive an IO and decide not to process it because of lack of resources. An unsolicated ABTS is immediately sent back to the initiator as a response. lpfc_nvmet_unsol_fcp_buffer lpfc_nvmet_unsol_issue_abort (XMIT_SEQUENCE_WQE) The second one is we sent the IO up to the NVMET transport layer to process, and for some reason the NVME Transport layer decided to abort the IO before it completes all its phases. For this case there are 2 paths thru the driver: the driver either has an outstanding TSEND/TRECEIVE/TRSP WQE or no outstanding WQEs are present for the exchange / context. lpfc_nvmet_xmt_fcp_abort if (LPFC_NVMET_IO_INP) lpfc_nvmet_sol_fcp_issue_abort (ABORT_WQE) lpfc_nvmet_sol_fcp_abort_cmp else lpfc_nvmet_unsol_fcp_issue_abort lpfc_nvmet_unsol_issue_abort (XMIT_SEQUENCE_WQE) lpfc_nvmet_unsol_fcp_abort_cmp Context flags: LPFC_NVMET_IOP - his flag signifies an IO is in progress on the exchange. LPFC_NVMET_XBUSY - this flag indicates the IO completed but the firmware is still busy with the corresponding exchange. The exchange should not be reused until after a XRI_ABORTED_CQE is received for that exchange. LPFC_NVMET_ABORT_OP - this flag signifies an ABORT_WQE was issued on the exchange. LPFC_NVMET_CTX_RLS - this flag signifies a context free was requested, but we are deferring it due to an XBUSY or ABORT in progress. A ctxlock is added to the context structure that is used whenever these flags are set/read within the context of an IO. The LPFC_NVMET_CTX_RLS flag is only set in the defer_relase routine when the transport has resolved all IO associated with the buffer. The flag is cleared when the CTX is associated with a new IO. An exchange can has both an LPFC_NVMET_XBUSY and a LPFC_NVMET_ABORT_OP condition active simultaneously. Both conditions must complete before the exchange is freed. When the abort callback (lpfc_nvmet_xmt_fcp_abort) is envoked: If there is an outstanding IO, the driver will issue an ABORT_WQE. This should result in 3 completions for the exchange: 1) IO cmpl with XB bit set 2) Abort WQE cmpl 3) XRI_ABORTED_CQE cmpl For this scenerio, after completion #1, the NVMET Transport IO rsp callback is called. After completion #2, no action is taken with respect to the exchange / context. After completion #3, the exchange context is free for re-use on another IO. If there is no outstanding activity on the exchange, the driver will send a ABTS to the Initiator. Upon completion of this WQE, the exchange / context is freed for re-use on another IO. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
2017-04-22 07:05:04 +08:00
list_for_each_entry_safe(ctxp, next_ctxp,
&phba->sli4_hba.lpfc_abts_nvmet_ctx_list,
list) {
cnt++;
}
spin_unlock(&phba->sli4_hba.abts_nvmet_buf_list_lock);
Update ABORT processing for NVMET. The driver with nvme had this routine stubbed. Right now XRI_ABORTED_CQE is not handled and the FC NVMET Transport has a new API for the driver. Missing code path, new NVME abort API Update ABORT processing for NVMET There are 3 new FC NVMET Transport API/ template routines for NVMET: lpfc_nvmet_xmt_fcp_release This NVMET template callback routine called to release context associated with an IO This routine is ALWAYS called last, even if the IO was aborted or completed in error. lpfc_nvmet_xmt_fcp_abort This NVMET template callback routine called to abort an exchange that has an IO in progress nvmet_fc_rcv_fcp_req When the lpfc driver receives an ABTS, this NVME FC transport layer callback routine is called. For this case there are 2 paths thru the driver: the driver either has an outstanding exchange / context for the XRI to be aborted or not. If not, a BA_RJT is issued otherwise a BA_ACC NVMET Driver abort paths: There are 2 paths for aborting an IO. The first one is we receive an IO and decide not to process it because of lack of resources. An unsolicated ABTS is immediately sent back to the initiator as a response. lpfc_nvmet_unsol_fcp_buffer lpfc_nvmet_unsol_issue_abort (XMIT_SEQUENCE_WQE) The second one is we sent the IO up to the NVMET transport layer to process, and for some reason the NVME Transport layer decided to abort the IO before it completes all its phases. For this case there are 2 paths thru the driver: the driver either has an outstanding TSEND/TRECEIVE/TRSP WQE or no outstanding WQEs are present for the exchange / context. lpfc_nvmet_xmt_fcp_abort if (LPFC_NVMET_IO_INP) lpfc_nvmet_sol_fcp_issue_abort (ABORT_WQE) lpfc_nvmet_sol_fcp_abort_cmp else lpfc_nvmet_unsol_fcp_issue_abort lpfc_nvmet_unsol_issue_abort (XMIT_SEQUENCE_WQE) lpfc_nvmet_unsol_fcp_abort_cmp Context flags: LPFC_NVMET_IOP - his flag signifies an IO is in progress on the exchange. LPFC_NVMET_XBUSY - this flag indicates the IO completed but the firmware is still busy with the corresponding exchange. The exchange should not be reused until after a XRI_ABORTED_CQE is received for that exchange. LPFC_NVMET_ABORT_OP - this flag signifies an ABORT_WQE was issued on the exchange. LPFC_NVMET_CTX_RLS - this flag signifies a context free was requested, but we are deferring it due to an XBUSY or ABORT in progress. A ctxlock is added to the context structure that is used whenever these flags are set/read within the context of an IO. The LPFC_NVMET_CTX_RLS flag is only set in the defer_relase routine when the transport has resolved all IO associated with the buffer. The flag is cleared when the CTX is associated with a new IO. An exchange can has both an LPFC_NVMET_XBUSY and a LPFC_NVMET_ABORT_OP condition active simultaneously. Both conditions must complete before the exchange is freed. When the abort callback (lpfc_nvmet_xmt_fcp_abort) is envoked: If there is an outstanding IO, the driver will issue an ABORT_WQE. This should result in 3 completions for the exchange: 1) IO cmpl with XB bit set 2) Abort WQE cmpl 3) XRI_ABORTED_CQE cmpl For this scenerio, after completion #1, the NVMET Transport IO rsp callback is called. After completion #2, no action is taken with respect to the exchange / context. After completion #3, the exchange context is free for re-use on another IO. If there is no outstanding activity on the exchange, the driver will send a ABTS to the Initiator. Upon completion of this WQE, the exchange / context is freed for re-use on another IO. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
2017-04-22 07:05:04 +08:00
if (cnt) {
len += snprintf(buf + len, size - len,
"ABORT: %d ctx entries\n", cnt);
spin_lock(&phba->sli4_hba.abts_nvmet_buf_list_lock);
Update ABORT processing for NVMET. The driver with nvme had this routine stubbed. Right now XRI_ABORTED_CQE is not handled and the FC NVMET Transport has a new API for the driver. Missing code path, new NVME abort API Update ABORT processing for NVMET There are 3 new FC NVMET Transport API/ template routines for NVMET: lpfc_nvmet_xmt_fcp_release This NVMET template callback routine called to release context associated with an IO This routine is ALWAYS called last, even if the IO was aborted or completed in error. lpfc_nvmet_xmt_fcp_abort This NVMET template callback routine called to abort an exchange that has an IO in progress nvmet_fc_rcv_fcp_req When the lpfc driver receives an ABTS, this NVME FC transport layer callback routine is called. For this case there are 2 paths thru the driver: the driver either has an outstanding exchange / context for the XRI to be aborted or not. If not, a BA_RJT is issued otherwise a BA_ACC NVMET Driver abort paths: There are 2 paths for aborting an IO. The first one is we receive an IO and decide not to process it because of lack of resources. An unsolicated ABTS is immediately sent back to the initiator as a response. lpfc_nvmet_unsol_fcp_buffer lpfc_nvmet_unsol_issue_abort (XMIT_SEQUENCE_WQE) The second one is we sent the IO up to the NVMET transport layer to process, and for some reason the NVME Transport layer decided to abort the IO before it completes all its phases. For this case there are 2 paths thru the driver: the driver either has an outstanding TSEND/TRECEIVE/TRSP WQE or no outstanding WQEs are present for the exchange / context. lpfc_nvmet_xmt_fcp_abort if (LPFC_NVMET_IO_INP) lpfc_nvmet_sol_fcp_issue_abort (ABORT_WQE) lpfc_nvmet_sol_fcp_abort_cmp else lpfc_nvmet_unsol_fcp_issue_abort lpfc_nvmet_unsol_issue_abort (XMIT_SEQUENCE_WQE) lpfc_nvmet_unsol_fcp_abort_cmp Context flags: LPFC_NVMET_IOP - his flag signifies an IO is in progress on the exchange. LPFC_NVMET_XBUSY - this flag indicates the IO completed but the firmware is still busy with the corresponding exchange. The exchange should not be reused until after a XRI_ABORTED_CQE is received for that exchange. LPFC_NVMET_ABORT_OP - this flag signifies an ABORT_WQE was issued on the exchange. LPFC_NVMET_CTX_RLS - this flag signifies a context free was requested, but we are deferring it due to an XBUSY or ABORT in progress. A ctxlock is added to the context structure that is used whenever these flags are set/read within the context of an IO. The LPFC_NVMET_CTX_RLS flag is only set in the defer_relase routine when the transport has resolved all IO associated with the buffer. The flag is cleared when the CTX is associated with a new IO. An exchange can has both an LPFC_NVMET_XBUSY and a LPFC_NVMET_ABORT_OP condition active simultaneously. Both conditions must complete before the exchange is freed. When the abort callback (lpfc_nvmet_xmt_fcp_abort) is envoked: If there is an outstanding IO, the driver will issue an ABORT_WQE. This should result in 3 completions for the exchange: 1) IO cmpl with XB bit set 2) Abort WQE cmpl 3) XRI_ABORTED_CQE cmpl For this scenerio, after completion #1, the NVMET Transport IO rsp callback is called. After completion #2, no action is taken with respect to the exchange / context. After completion #3, the exchange context is free for re-use on another IO. If there is no outstanding activity on the exchange, the driver will send a ABTS to the Initiator. Upon completion of this WQE, the exchange / context is freed for re-use on another IO. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
2017-04-22 07:05:04 +08:00
list_for_each_entry_safe(ctxp, next_ctxp,
&phba->sli4_hba.lpfc_abts_nvmet_ctx_list,
list) {
if (len >= (size - LPFC_DEBUG_OUT_LINE_SZ))
break;
len += snprintf(buf + len, size - len,
"Entry: oxid %x state %x "
"flag %x\n",
ctxp->oxid, ctxp->state,
ctxp->flag);
}
spin_unlock(&phba->sli4_hba.abts_nvmet_buf_list_lock);
Update ABORT processing for NVMET. The driver with nvme had this routine stubbed. Right now XRI_ABORTED_CQE is not handled and the FC NVMET Transport has a new API for the driver. Missing code path, new NVME abort API Update ABORT processing for NVMET There are 3 new FC NVMET Transport API/ template routines for NVMET: lpfc_nvmet_xmt_fcp_release This NVMET template callback routine called to release context associated with an IO This routine is ALWAYS called last, even if the IO was aborted or completed in error. lpfc_nvmet_xmt_fcp_abort This NVMET template callback routine called to abort an exchange that has an IO in progress nvmet_fc_rcv_fcp_req When the lpfc driver receives an ABTS, this NVME FC transport layer callback routine is called. For this case there are 2 paths thru the driver: the driver either has an outstanding exchange / context for the XRI to be aborted or not. If not, a BA_RJT is issued otherwise a BA_ACC NVMET Driver abort paths: There are 2 paths for aborting an IO. The first one is we receive an IO and decide not to process it because of lack of resources. An unsolicated ABTS is immediately sent back to the initiator as a response. lpfc_nvmet_unsol_fcp_buffer lpfc_nvmet_unsol_issue_abort (XMIT_SEQUENCE_WQE) The second one is we sent the IO up to the NVMET transport layer to process, and for some reason the NVME Transport layer decided to abort the IO before it completes all its phases. For this case there are 2 paths thru the driver: the driver either has an outstanding TSEND/TRECEIVE/TRSP WQE or no outstanding WQEs are present for the exchange / context. lpfc_nvmet_xmt_fcp_abort if (LPFC_NVMET_IO_INP) lpfc_nvmet_sol_fcp_issue_abort (ABORT_WQE) lpfc_nvmet_sol_fcp_abort_cmp else lpfc_nvmet_unsol_fcp_issue_abort lpfc_nvmet_unsol_issue_abort (XMIT_SEQUENCE_WQE) lpfc_nvmet_unsol_fcp_abort_cmp Context flags: LPFC_NVMET_IOP - his flag signifies an IO is in progress on the exchange. LPFC_NVMET_XBUSY - this flag indicates the IO completed but the firmware is still busy with the corresponding exchange. The exchange should not be reused until after a XRI_ABORTED_CQE is received for that exchange. LPFC_NVMET_ABORT_OP - this flag signifies an ABORT_WQE was issued on the exchange. LPFC_NVMET_CTX_RLS - this flag signifies a context free was requested, but we are deferring it due to an XBUSY or ABORT in progress. A ctxlock is added to the context structure that is used whenever these flags are set/read within the context of an IO. The LPFC_NVMET_CTX_RLS flag is only set in the defer_relase routine when the transport has resolved all IO associated with the buffer. The flag is cleared when the CTX is associated with a new IO. An exchange can has both an LPFC_NVMET_XBUSY and a LPFC_NVMET_ABORT_OP condition active simultaneously. Both conditions must complete before the exchange is freed. When the abort callback (lpfc_nvmet_xmt_fcp_abort) is envoked: If there is an outstanding IO, the driver will issue an ABORT_WQE. This should result in 3 completions for the exchange: 1) IO cmpl with XB bit set 2) Abort WQE cmpl 3) XRI_ABORTED_CQE cmpl For this scenerio, after completion #1, the NVMET Transport IO rsp callback is called. After completion #2, no action is taken with respect to the exchange / context. After completion #3, the exchange context is free for re-use on another IO. If there is no outstanding activity on the exchange, the driver will send a ABTS to the Initiator. Upon completion of this WQE, the exchange / context is freed for re-use on another IO. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
2017-04-22 07:05:04 +08:00
}
/* Calculate outstanding IOs */
tot = atomic_read(&tgtp->rcv_fcp_cmd_drop);
tot += atomic_read(&tgtp->xmt_fcp_release);
tot = atomic_read(&tgtp->rcv_fcp_cmd_in) - tot;
len += snprintf(buf + len, size - len,
"IO_CTX: %08x WAIT: cur %08x tot %08x\n"
"CTX Outstanding %08llx\n",
phba->sli4_hba.nvmet_xri_cnt,
phba->sli4_hba.nvmet_io_wait_cnt,
phba->sli4_hba.nvmet_io_wait_total,
tot);
} else {
if (!(vport->cfg_enable_fc4_type & LPFC_ENABLE_NVME))
return len;
localport = vport->localport;
if (!localport)
return len;
lport = (struct lpfc_nvme_lport *)localport->private;
if (!lport)
return len;
len += snprintf(buf + len, size - len,
"\nNVME HDWQ Statistics\n");
len += snprintf(buf + len, size - len,
"LS: Xmt %016x Cmpl %016x\n",
atomic_read(&lport->fc4NvmeLsRequests),
atomic_read(&lport->fc4NvmeLsCmpls));
totin = 0;
totout = 0;
2019-01-29 03:14:21 +08:00
for (i = 0; i < phba->cfg_hdw_queue; i++) {
cstat = &phba->sli4_hba.hdwq[i].nvme_cstat;
tot = cstat->io_cmpls;
totin += tot;
data1 = cstat->input_requests;
data2 = cstat->output_requests;
data3 = cstat->control_requests;
totout += (data1 + data2 + data3);
/* Limit to 32, debugfs display buffer limitation */
if (i >= 32)
continue;
len += snprintf(buf + len, PAGE_SIZE - len,
"HDWQ (%d): Rd %016llx Wr %016llx "
"IO %016llx ",
i, data1, data2, data3);
len += snprintf(buf + len, PAGE_SIZE - len,
"Cmpl %016llx OutIO %016llx\n",
tot, ((data1 + data2 + data3) - tot));
}
len += snprintf(buf + len, PAGE_SIZE - len,
"Total FCP Cmpl %016llx Issue %016llx "
"OutIO %016llx\n",
totin, totout, totout - totin);
len += snprintf(buf + len, size - len,
"LS Xmt Err: Abrt %08x Err %08x "
"Cmpl Err: xb %08x Err %08x\n",
atomic_read(&lport->xmt_ls_abort),
atomic_read(&lport->xmt_ls_err),
atomic_read(&lport->cmpl_ls_xb),
atomic_read(&lport->cmpl_ls_err));
len += snprintf(buf + len, size - len,
"FCP Xmt Err: noxri %06x nondlp %06x "
"qdepth %06x wqerr %06x err %06x Abrt %06x\n",
atomic_read(&lport->xmt_fcp_noxri),
atomic_read(&lport->xmt_fcp_bad_ndlp),
atomic_read(&lport->xmt_fcp_qdepth),
atomic_read(&lport->xmt_fcp_wqerr),
atomic_read(&lport->xmt_fcp_err),
atomic_read(&lport->xmt_fcp_abort));
len += snprintf(buf + len, size - len,
"FCP Cmpl Err: xb %08x Err %08x\n",
atomic_read(&lport->cmpl_fcp_xb),
atomic_read(&lport->cmpl_fcp_err));
}
return len;
}
/**
* lpfc_debugfs_scsistat_data - Dump target node list to a buffer
* @vport: The vport to gather target node info from.
* @buf: The buffer to dump log into.
* @size: The maximum amount of data to process.
*
* Description:
* This routine dumps the SCSI statistics associated with @vport
*
* Return Value:
* This routine returns the amount of bytes that were dumped into @buf and will
* not exceed @size.
**/
static int
lpfc_debugfs_scsistat_data(struct lpfc_vport *vport, char *buf, int size)
{
int len;
struct lpfc_hba *phba = vport->phba;
struct lpfc_fc4_ctrl_stat *cstat;
u64 data1, data2, data3;
u64 tot, totin, totout;
int i;
char tmp[LPFC_MAX_SCSI_INFO_TMP_LEN] = {0};
if (!(vport->cfg_enable_fc4_type & LPFC_ENABLE_FCP) ||
(phba->sli_rev != LPFC_SLI_REV4))
return 0;
scnprintf(buf, size, "SCSI HDWQ Statistics\n");
totin = 0;
totout = 0;
for (i = 0; i < phba->cfg_hdw_queue; i++) {
cstat = &phba->sli4_hba.hdwq[i].scsi_cstat;
tot = cstat->io_cmpls;
totin += tot;
data1 = cstat->input_requests;
data2 = cstat->output_requests;
data3 = cstat->control_requests;
totout += (data1 + data2 + data3);
scnprintf(tmp, sizeof(tmp), "HDWQ (%d): Rd %016llx Wr %016llx "
"IO %016llx ", i, data1, data2, data3);
if (strlcat(buf, tmp, size) >= size)
goto buffer_done;
scnprintf(tmp, sizeof(tmp), "Cmpl %016llx OutIO %016llx\n",
tot, ((data1 + data2 + data3) - tot));
if (strlcat(buf, tmp, size) >= size)
goto buffer_done;
}
scnprintf(tmp, sizeof(tmp), "Total FCP Cmpl %016llx Issue %016llx "
"OutIO %016llx\n", totin, totout, totout - totin);
strlcat(buf, tmp, size);
buffer_done:
len = strnlen(buf, size);
return len;
}
/**
* lpfc_debugfs_nvmektime_data - Dump target node list to a buffer
* @vport: The vport to gather target node info from.
* @buf: The buffer to dump log into.
* @size: The maximum amount of data to process.
*
* Description:
* This routine dumps the NVME statistics associated with @vport
*
* Return Value:
* This routine returns the amount of bytes that were dumped into @buf and will
* not exceed @size.
**/
static int
lpfc_debugfs_nvmektime_data(struct lpfc_vport *vport, char *buf, int size)
{
struct lpfc_hba *phba = vport->phba;
int len = 0;
if (phba->nvmet_support == 0) {
/* NVME Initiator */
len += snprintf(buf + len, PAGE_SIZE - len,
"ktime %s: Total Samples: %lld\n",
(phba->ktime_on ? "Enabled" : "Disabled"),
phba->ktime_data_samples);
if (phba->ktime_data_samples == 0)
return len;
len += snprintf(
buf + len, PAGE_SIZE - len,
"Segment 1: Last NVME Cmd cmpl "
"done -to- Start of next NVME cnd (in driver)\n");
len += snprintf(
buf + len, PAGE_SIZE - len,
"avg:%08lld min:%08lld max %08lld\n",
div_u64(phba->ktime_seg1_total,
phba->ktime_data_samples),
phba->ktime_seg1_min,
phba->ktime_seg1_max);
len += snprintf(
buf + len, PAGE_SIZE - len,
"Segment 2: Driver start of NVME cmd "
"-to- Firmware WQ doorbell\n");
len += snprintf(
buf + len, PAGE_SIZE - len,
"avg:%08lld min:%08lld max %08lld\n",
div_u64(phba->ktime_seg2_total,
phba->ktime_data_samples),
phba->ktime_seg2_min,
phba->ktime_seg2_max);
len += snprintf(
buf + len, PAGE_SIZE - len,
"Segment 3: Firmware WQ doorbell -to- "
"MSI-X ISR cmpl\n");
len += snprintf(
buf + len, PAGE_SIZE - len,
"avg:%08lld min:%08lld max %08lld\n",
div_u64(phba->ktime_seg3_total,
phba->ktime_data_samples),
phba->ktime_seg3_min,
phba->ktime_seg3_max);
len += snprintf(
buf + len, PAGE_SIZE - len,
"Segment 4: MSI-X ISR cmpl -to- "
"NVME cmpl done\n");
len += snprintf(
buf + len, PAGE_SIZE - len,
"avg:%08lld min:%08lld max %08lld\n",
div_u64(phba->ktime_seg4_total,
phba->ktime_data_samples),
phba->ktime_seg4_min,
phba->ktime_seg4_max);
len += snprintf(
buf + len, PAGE_SIZE - len,
"Total IO avg time: %08lld\n",
div_u64(phba->ktime_seg1_total +
phba->ktime_seg2_total +
phba->ktime_seg3_total +
phba->ktime_seg4_total,
phba->ktime_data_samples));
return len;
}
/* NVME Target */
len += snprintf(buf + len, PAGE_SIZE-len,
"ktime %s: Total Samples: %lld %lld\n",
(phba->ktime_on ? "Enabled" : "Disabled"),
phba->ktime_data_samples,
phba->ktime_status_samples);
if (phba->ktime_data_samples == 0)
return len;
len += snprintf(buf + len, PAGE_SIZE-len,
"Segment 1: MSI-X ISR Rcv cmd -to- "
"cmd pass to NVME Layer\n");
len += snprintf(buf + len, PAGE_SIZE-len,
"avg:%08lld min:%08lld max %08lld\n",
div_u64(phba->ktime_seg1_total,
phba->ktime_data_samples),
phba->ktime_seg1_min,
phba->ktime_seg1_max);
len += snprintf(buf + len, PAGE_SIZE-len,
"Segment 2: cmd pass to NVME Layer- "
"-to- Driver rcv cmd OP (action)\n");
len += snprintf(buf + len, PAGE_SIZE-len,
"avg:%08lld min:%08lld max %08lld\n",
div_u64(phba->ktime_seg2_total,
phba->ktime_data_samples),
phba->ktime_seg2_min,
phba->ktime_seg2_max);
len += snprintf(buf + len, PAGE_SIZE-len,
"Segment 3: Driver rcv cmd OP -to- "
"Firmware WQ doorbell: cmd\n");
len += snprintf(buf + len, PAGE_SIZE-len,
"avg:%08lld min:%08lld max %08lld\n",
div_u64(phba->ktime_seg3_total,
phba->ktime_data_samples),
phba->ktime_seg3_min,
phba->ktime_seg3_max);
len += snprintf(buf + len, PAGE_SIZE-len,
"Segment 4: Firmware WQ doorbell: cmd "
"-to- MSI-X ISR for cmd cmpl\n");
len += snprintf(buf + len, PAGE_SIZE-len,
"avg:%08lld min:%08lld max %08lld\n",
div_u64(phba->ktime_seg4_total,
phba->ktime_data_samples),
phba->ktime_seg4_min,
phba->ktime_seg4_max);
len += snprintf(buf + len, PAGE_SIZE-len,
"Segment 5: MSI-X ISR for cmd cmpl "
"-to- NVME layer passed cmd done\n");
len += snprintf(buf + len, PAGE_SIZE-len,
"avg:%08lld min:%08lld max %08lld\n",
div_u64(phba->ktime_seg5_total,
phba->ktime_data_samples),
phba->ktime_seg5_min,
phba->ktime_seg5_max);
if (phba->ktime_status_samples == 0) {
len += snprintf(buf + len, PAGE_SIZE-len,
"Total: cmd received by MSI-X ISR "
"-to- cmd completed on wire\n");
len += snprintf(buf + len, PAGE_SIZE-len,
"avg:%08lld min:%08lld "
"max %08lld\n",
div_u64(phba->ktime_seg10_total,
phba->ktime_data_samples),
phba->ktime_seg10_min,
phba->ktime_seg10_max);
return len;
}
len += snprintf(buf + len, PAGE_SIZE-len,
"Segment 6: NVME layer passed cmd done "
"-to- Driver rcv rsp status OP\n");
len += snprintf(buf + len, PAGE_SIZE-len,
"avg:%08lld min:%08lld max %08lld\n",
div_u64(phba->ktime_seg6_total,
phba->ktime_status_samples),
phba->ktime_seg6_min,
phba->ktime_seg6_max);
len += snprintf(buf + len, PAGE_SIZE-len,
"Segment 7: Driver rcv rsp status OP "
"-to- Firmware WQ doorbell: status\n");
len += snprintf(buf + len, PAGE_SIZE-len,
"avg:%08lld min:%08lld max %08lld\n",
div_u64(phba->ktime_seg7_total,
phba->ktime_status_samples),
phba->ktime_seg7_min,
phba->ktime_seg7_max);
len += snprintf(buf + len, PAGE_SIZE-len,
"Segment 8: Firmware WQ doorbell: status"
" -to- MSI-X ISR for status cmpl\n");
len += snprintf(buf + len, PAGE_SIZE-len,
"avg:%08lld min:%08lld max %08lld\n",
div_u64(phba->ktime_seg8_total,
phba->ktime_status_samples),
phba->ktime_seg8_min,
phba->ktime_seg8_max);
len += snprintf(buf + len, PAGE_SIZE-len,
"Segment 9: MSI-X ISR for status cmpl "
"-to- NVME layer passed status done\n");
len += snprintf(buf + len, PAGE_SIZE-len,
"avg:%08lld min:%08lld max %08lld\n",
div_u64(phba->ktime_seg9_total,
phba->ktime_status_samples),
phba->ktime_seg9_min,
phba->ktime_seg9_max);
len += snprintf(buf + len, PAGE_SIZE-len,
"Total: cmd received by MSI-X ISR -to- "
"cmd completed on wire\n");
len += snprintf(buf + len, PAGE_SIZE-len,
"avg:%08lld min:%08lld max %08lld\n",
div_u64(phba->ktime_seg10_total,
phba->ktime_status_samples),
phba->ktime_seg10_min,
phba->ktime_seg10_max);
return len;
}
/**
* lpfc_debugfs_nvmeio_trc_data - Dump NVME IO trace list to a buffer
* @phba: The phba to gather target node info from.
* @buf: The buffer to dump log into.
* @size: The maximum amount of data to process.
*
* Description:
* This routine dumps the NVME IO trace associated with @phba
*
* Return Value:
* This routine returns the amount of bytes that were dumped into @buf and will
* not exceed @size.
**/
static int
lpfc_debugfs_nvmeio_trc_data(struct lpfc_hba *phba, char *buf, int size)
{
struct lpfc_debugfs_nvmeio_trc *dtp;
int i, state, index, skip;
int len = 0;
state = phba->nvmeio_trc_on;
index = (atomic_read(&phba->nvmeio_trc_cnt) + 1) &
(phba->nvmeio_trc_size - 1);
skip = phba->nvmeio_trc_output_idx;
len += snprintf(buf + len, size - len,
"%s IO Trace %s: next_idx %d skip %d size %d\n",
(phba->nvmet_support ? "NVME" : "NVMET"),
(state ? "Enabled" : "Disabled"),
index, skip, phba->nvmeio_trc_size);
if (!phba->nvmeio_trc || state)
return len;
/* trace MUST bhe off to continue */
for (i = index; i < phba->nvmeio_trc_size; i++) {
if (skip) {
skip--;
continue;
}
dtp = phba->nvmeio_trc + i;
phba->nvmeio_trc_output_idx++;
if (!dtp->fmt)
continue;
len += snprintf(buf + len, size - len, dtp->fmt,
dtp->data1, dtp->data2, dtp->data3);
if (phba->nvmeio_trc_output_idx >= phba->nvmeio_trc_size) {
phba->nvmeio_trc_output_idx = 0;
len += snprintf(buf + len, size - len,
"Trace Complete\n");
goto out;
}
if (len >= (size - LPFC_DEBUG_OUT_LINE_SZ)) {
len += snprintf(buf + len, size - len,
"Trace Continue (%d of %d)\n",
phba->nvmeio_trc_output_idx,
phba->nvmeio_trc_size);
goto out;
}
}
for (i = 0; i < index; i++) {
if (skip) {
skip--;
continue;
}
dtp = phba->nvmeio_trc + i;
phba->nvmeio_trc_output_idx++;
if (!dtp->fmt)
continue;
len += snprintf(buf + len, size - len, dtp->fmt,
dtp->data1, dtp->data2, dtp->data3);
if (phba->nvmeio_trc_output_idx >= phba->nvmeio_trc_size) {
phba->nvmeio_trc_output_idx = 0;
len += snprintf(buf + len, size - len,
"Trace Complete\n");
goto out;
}
if (len >= (size - LPFC_DEBUG_OUT_LINE_SZ)) {
len += snprintf(buf + len, size - len,
"Trace Continue (%d of %d)\n",
phba->nvmeio_trc_output_idx,
phba->nvmeio_trc_size);
goto out;
}
}
len += snprintf(buf + len, size - len,
"Trace Done\n");
out:
return len;
}
/**
* lpfc_debugfs_cpucheck_data - Dump target node list to a buffer
* @vport: The vport to gather target node info from.
* @buf: The buffer to dump log into.
* @size: The maximum amount of data to process.
*
* Description:
* This routine dumps the NVME statistics associated with @vport
*
* Return Value:
* This routine returns the amount of bytes that were dumped into @buf and will
* not exceed @size.
**/
static int
lpfc_debugfs_cpucheck_data(struct lpfc_vport *vport, char *buf, int size)
{
struct lpfc_hba *phba = vport->phba;
struct lpfc_sli4_hdw_queue *qp;
int i, j, max_cnt;
int len = 0;
uint32_t tot_xmt;
uint32_t tot_rcv;
uint32_t tot_cmpl;
len += snprintf(buf + len, PAGE_SIZE - len,
"CPUcheck %s ",
(phba->cpucheck_on & LPFC_CHECK_NVME_IO ?
"Enabled" : "Disabled"));
if (phba->nvmet_support) {
len += snprintf(buf + len, PAGE_SIZE - len,
"%s\n",
(phba->cpucheck_on & LPFC_CHECK_NVMET_RCV ?
"Rcv Enabled\n" : "Rcv Disabled\n"));
} else {
len += snprintf(buf + len, PAGE_SIZE - len, "\n");
}
max_cnt = size - LPFC_DEBUG_OUT_LINE_SZ;
for (i = 0; i < phba->cfg_hdw_queue; i++) {
qp = &phba->sli4_hba.hdwq[i];
tot_rcv = 0;
tot_xmt = 0;
tot_cmpl = 0;
for (j = 0; j < LPFC_CHECK_CPU_CNT; j++) {
tot_xmt += qp->cpucheck_xmt_io[j];
tot_cmpl += qp->cpucheck_cmpl_io[j];
if (phba->nvmet_support)
tot_rcv += qp->cpucheck_rcv_io[j];
}
/* Only display Hardware Qs with something */
if (!tot_xmt && !tot_cmpl && !tot_rcv)
continue;
len += snprintf(buf + len, PAGE_SIZE - len,
"HDWQ %03d: ", i);
for (j = 0; j < LPFC_CHECK_CPU_CNT; j++) {
/* Only display non-zero counters */
if (!qp->cpucheck_xmt_io[j] &&
!qp->cpucheck_cmpl_io[j] &&
!qp->cpucheck_rcv_io[j])
continue;
if (phba->nvmet_support) {
len += snprintf(buf + len, PAGE_SIZE - len,
"CPU %03d: %x/%x/%x ", j,
qp->cpucheck_rcv_io[j],
qp->cpucheck_xmt_io[j],
qp->cpucheck_cmpl_io[j]);
} else {
len += snprintf(buf + len, PAGE_SIZE - len,
"CPU %03d: %x/%x ", j,
qp->cpucheck_xmt_io[j],
qp->cpucheck_cmpl_io[j]);
}
}
len += snprintf(buf + len, PAGE_SIZE - len,
"Total: %x\n", tot_xmt);
if (len >= max_cnt) {
len += snprintf(buf + len, PAGE_SIZE - len,
"Truncated ...\n");
return len;
}
}
return len;
}
#endif
/**
* lpfc_debugfs_disc_trc - Store discovery trace log
* @vport: The vport to associate this trace string with for retrieval.
* @mask: Log entry classification.
* @fmt: Format string to be displayed when dumping the log.
* @data1: 1st data parameter to be applied to @fmt.
* @data2: 2nd data parameter to be applied to @fmt.
* @data3: 3rd data parameter to be applied to @fmt.
*
* Description:
* This routine is used by the driver code to add a debugfs log entry to the
* discovery trace buffer associated with @vport. Only entries with a @mask that
* match the current debugfs discovery mask will be saved. Entries that do not
* match will be thrown away. @fmt, @data1, @data2, and @data3 are used like
* printf when displaying the log.
**/
inline void
lpfc_debugfs_disc_trc(struct lpfc_vport *vport, int mask, char *fmt,
uint32_t data1, uint32_t data2, uint32_t data3)
{
#ifdef CONFIG_SCSI_LPFC_DEBUG_FS
struct lpfc_debugfs_trc *dtp;
int index;
if (!(lpfc_debugfs_mask_disc_trc & mask))
return;
if (!lpfc_debugfs_enable || !lpfc_debugfs_max_disc_trc ||
!vport || !vport->disc_trc)
return;
index = atomic_inc_return(&vport->disc_trc_cnt) &
(lpfc_debugfs_max_disc_trc - 1);
dtp = vport->disc_trc + index;
dtp->fmt = fmt;
dtp->data1 = data1;
dtp->data2 = data2;
dtp->data3 = data3;
dtp->seq_cnt = atomic_inc_return(&lpfc_debugfs_seq_trc_cnt);
dtp->jif = jiffies;
#endif
return;
}
/**
* lpfc_debugfs_slow_ring_trc - Store slow ring trace log
* @phba: The phba to associate this trace string with for retrieval.
* @fmt: Format string to be displayed when dumping the log.
* @data1: 1st data parameter to be applied to @fmt.
* @data2: 2nd data parameter to be applied to @fmt.
* @data3: 3rd data parameter to be applied to @fmt.
*
* Description:
* This routine is used by the driver code to add a debugfs log entry to the
* discovery trace buffer associated with @vport. @fmt, @data1, @data2, and
* @data3 are used like printf when displaying the log.
**/
inline void
lpfc_debugfs_slow_ring_trc(struct lpfc_hba *phba, char *fmt,
uint32_t data1, uint32_t data2, uint32_t data3)
{
#ifdef CONFIG_SCSI_LPFC_DEBUG_FS
struct lpfc_debugfs_trc *dtp;
int index;
if (!lpfc_debugfs_enable || !lpfc_debugfs_max_slow_ring_trc ||
!phba || !phba->slow_ring_trc)
return;
index = atomic_inc_return(&phba->slow_ring_trc_cnt) &
(lpfc_debugfs_max_slow_ring_trc - 1);
dtp = phba->slow_ring_trc + index;
dtp->fmt = fmt;
dtp->data1 = data1;
dtp->data2 = data2;
dtp->data3 = data3;
dtp->seq_cnt = atomic_inc_return(&lpfc_debugfs_seq_trc_cnt);
dtp->jif = jiffies;
#endif
return;
}
/**
* lpfc_debugfs_nvme_trc - Store NVME/NVMET trace log
* @phba: The phba to associate this trace string with for retrieval.
* @fmt: Format string to be displayed when dumping the log.
* @data1: 1st data parameter to be applied to @fmt.
* @data2: 2nd data parameter to be applied to @fmt.
* @data3: 3rd data parameter to be applied to @fmt.
*
* Description:
* This routine is used by the driver code to add a debugfs log entry to the
* nvme trace buffer associated with @phba. @fmt, @data1, @data2, and
* @data3 are used like printf when displaying the log.
**/
inline void
lpfc_debugfs_nvme_trc(struct lpfc_hba *phba, char *fmt,
uint16_t data1, uint16_t data2, uint32_t data3)
{
#ifdef CONFIG_SCSI_LPFC_DEBUG_FS
struct lpfc_debugfs_nvmeio_trc *dtp;
int index;
if (!phba->nvmeio_trc_on || !phba->nvmeio_trc)
return;
index = atomic_inc_return(&phba->nvmeio_trc_cnt) &
(phba->nvmeio_trc_size - 1);
dtp = phba->nvmeio_trc + index;
dtp->fmt = fmt;
dtp->data1 = data1;
dtp->data2 = data2;
dtp->data3 = data3;
#endif
}
#ifdef CONFIG_SCSI_LPFC_DEBUG_FS
/**
* lpfc_debugfs_disc_trc_open - Open the discovery trace log
* @inode: The inode pointer that contains a vport pointer.
* @file: The file pointer to attach the log output.
*
* Description:
* This routine is the entry point for the debugfs open file operation. It gets
* the vport from the i_private field in @inode, allocates the necessary buffer
* for the log, fills the buffer from the in-memory log for this vport, and then
* returns a pointer to that log in the private_data field in @file.
*
* Returns:
* This function returns zero if successful. On error it will return a negative
* error value.
**/
static int
lpfc_debugfs_disc_trc_open(struct inode *inode, struct file *file)
{
struct lpfc_vport *vport = inode->i_private;
struct lpfc_debug *debug;
int size;
int rc = -ENOMEM;
if (!lpfc_debugfs_max_disc_trc) {
rc = -ENOSPC;
goto out;
}
debug = kmalloc(sizeof(*debug), GFP_KERNEL);
if (!debug)
goto out;
/* Round to page boundary */
size = (lpfc_debugfs_max_disc_trc * LPFC_DEBUG_TRC_ENTRY_SIZE);
size = PAGE_ALIGN(size);
debug->buffer = kmalloc(size, GFP_KERNEL);
if (!debug->buffer) {
kfree(debug);
goto out;
}
debug->len = lpfc_debugfs_disc_trc_data(vport, debug->buffer, size);
file->private_data = debug;
rc = 0;
out:
return rc;
}
/**
* lpfc_debugfs_slow_ring_trc_open - Open the Slow Ring trace log
* @inode: The inode pointer that contains a vport pointer.
* @file: The file pointer to attach the log output.
*
* Description:
* This routine is the entry point for the debugfs open file operation. It gets
* the vport from the i_private field in @inode, allocates the necessary buffer
* for the log, fills the buffer from the in-memory log for this vport, and then
* returns a pointer to that log in the private_data field in @file.
*
* Returns:
* This function returns zero if successful. On error it will return a negative
* error value.
**/
static int
lpfc_debugfs_slow_ring_trc_open(struct inode *inode, struct file *file)
{
struct lpfc_hba *phba = inode->i_private;
struct lpfc_debug *debug;
int size;
int rc = -ENOMEM;
if (!lpfc_debugfs_max_slow_ring_trc) {
rc = -ENOSPC;
goto out;
}
debug = kmalloc(sizeof(*debug), GFP_KERNEL);
if (!debug)
goto out;
/* Round to page boundary */
size = (lpfc_debugfs_max_slow_ring_trc * LPFC_DEBUG_TRC_ENTRY_SIZE);
size = PAGE_ALIGN(size);
debug->buffer = kmalloc(size, GFP_KERNEL);
if (!debug->buffer) {
kfree(debug);
goto out;
}
debug->len = lpfc_debugfs_slow_ring_trc_data(phba, debug->buffer, size);
file->private_data = debug;
rc = 0;
out:
return rc;
}
/**
* lpfc_debugfs_hbqinfo_open - Open the hbqinfo debugfs buffer
* @inode: The inode pointer that contains a vport pointer.
* @file: The file pointer to attach the log output.
*
* Description:
* This routine is the entry point for the debugfs open file operation. It gets
* the vport from the i_private field in @inode, allocates the necessary buffer
* for the log, fills the buffer from the in-memory log for this vport, and then
* returns a pointer to that log in the private_data field in @file.
*
* Returns:
* This function returns zero if successful. On error it will return a negative
* error value.
**/
static int
lpfc_debugfs_hbqinfo_open(struct inode *inode, struct file *file)
{
struct lpfc_hba *phba = inode->i_private;
struct lpfc_debug *debug;
int rc = -ENOMEM;
debug = kmalloc(sizeof(*debug), GFP_KERNEL);
if (!debug)
goto out;
/* Round to page boundary */
debug->buffer = kmalloc(LPFC_HBQINFO_SIZE, GFP_KERNEL);
if (!debug->buffer) {
kfree(debug);
goto out;
}
debug->len = lpfc_debugfs_hbqinfo_data(phba, debug->buffer,
LPFC_HBQINFO_SIZE);
file->private_data = debug;
rc = 0;
out:
return rc;
}
scsi: lpfc: Adapt partitioned XRI lists to efficient sharing The XRI get/put lists were partitioned per hardware queue. However, the adapter rarely had sufficient resources to give a large number of resources per queue. As such, it became common for a cpu to encounter a lack of XRI resource and request the upper io stack to retry after returning a BUSY condition. This occurred even though other cpus were idle and not using their resources. Create as efficient a scheme as possible to move resources to the cpus that need them. Each cpu maintains a small private pool which it allocates from for io. There is a watermark that the cpu attempts to keep in the private pool. The private pool, when empty, pulls from a global pool from the cpu. When the cpu's global pool is empty it will pull from other cpu's global pool. As there many cpu global pools (1 per cpu or hardware queue count) and as each cpu selects what cpu to pull from at different rates and at different times, it creates a radomizing effect that minimizes the number of cpu's that will contend with each other when the steal XRI's from another cpu's global pool. On io completion, a cpu will push the XRI back on to its private pool. A watermark level is maintained for the private pool such that when it is exceeded it will move XRI's to the CPU global pool so that other cpu's may allocate them. On NVME, as heartbeat commands are critical to get placed on the wire, a single expedite pool is maintained. When a heartbeat is to be sent, it will allocate an XRI from the expedite pool rather than the normal cpu private/global pools. On any io completion, if a reduction in the expedite pools is seen, it will be replenished before the XRI is placed on the cpu private pool. Statistics are added to aid understanding the XRI levels on each cpu and their behaviors. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <jsmart2021@gmail.com> Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2019-01-29 03:14:28 +08:00
/**
* lpfc_debugfs_multixripools_open - Open the multixripool debugfs buffer
* @inode: The inode pointer that contains a hba pointer.
* @file: The file pointer to attach the log output.
*
* Description:
* This routine is the entry point for the debugfs open file operation. It gets
* the hba from the i_private field in @inode, allocates the necessary buffer
* for the log, fills the buffer from the in-memory log for this hba, and then
* returns a pointer to that log in the private_data field in @file.
*
* Returns:
* This function returns zero if successful. On error it will return a negative
* error value.
**/
static int
lpfc_debugfs_multixripools_open(struct inode *inode, struct file *file)
{
struct lpfc_hba *phba = inode->i_private;
struct lpfc_debug *debug;
int rc = -ENOMEM;
debug = kmalloc(sizeof(*debug), GFP_KERNEL);
if (!debug)
goto out;
/* Round to page boundary */
debug->buffer = kzalloc(LPFC_DUMP_MULTIXRIPOOL_SIZE, GFP_KERNEL);
if (!debug->buffer) {
kfree(debug);
goto out;
}
debug->len = lpfc_debugfs_multixripools_data(
phba, debug->buffer, LPFC_DUMP_MULTIXRIPOOL_SIZE);
scsi: lpfc: Adapt partitioned XRI lists to efficient sharing The XRI get/put lists were partitioned per hardware queue. However, the adapter rarely had sufficient resources to give a large number of resources per queue. As such, it became common for a cpu to encounter a lack of XRI resource and request the upper io stack to retry after returning a BUSY condition. This occurred even though other cpus were idle and not using their resources. Create as efficient a scheme as possible to move resources to the cpus that need them. Each cpu maintains a small private pool which it allocates from for io. There is a watermark that the cpu attempts to keep in the private pool. The private pool, when empty, pulls from a global pool from the cpu. When the cpu's global pool is empty it will pull from other cpu's global pool. As there many cpu global pools (1 per cpu or hardware queue count) and as each cpu selects what cpu to pull from at different rates and at different times, it creates a radomizing effect that minimizes the number of cpu's that will contend with each other when the steal XRI's from another cpu's global pool. On io completion, a cpu will push the XRI back on to its private pool. A watermark level is maintained for the private pool such that when it is exceeded it will move XRI's to the CPU global pool so that other cpu's may allocate them. On NVME, as heartbeat commands are critical to get placed on the wire, a single expedite pool is maintained. When a heartbeat is to be sent, it will allocate an XRI from the expedite pool rather than the normal cpu private/global pools. On any io completion, if a reduction in the expedite pools is seen, it will be replenished before the XRI is placed on the cpu private pool. Statistics are added to aid understanding the XRI levels on each cpu and their behaviors. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <jsmart2021@gmail.com> Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2019-01-29 03:14:28 +08:00
debug->i_private = inode->i_private;
file->private_data = debug;
rc = 0;
out:
return rc;
}
#ifdef LPFC_HDWQ_LOCK_STAT
/**
* lpfc_debugfs_lockstat_open - Open the lockstat debugfs buffer
* @inode: The inode pointer that contains a vport pointer.
* @file: The file pointer to attach the log output.
*
* Description:
* This routine is the entry point for the debugfs open file operation. It gets
* the vport from the i_private field in @inode, allocates the necessary buffer
* for the log, fills the buffer from the in-memory log for this vport, and then
* returns a pointer to that log in the private_data field in @file.
*
* Returns:
* This function returns zero if successful. On error it will return a negative
* error value.
**/
static int
lpfc_debugfs_lockstat_open(struct inode *inode, struct file *file)
{
struct lpfc_hba *phba = inode->i_private;
struct lpfc_debug *debug;
int rc = -ENOMEM;
debug = kmalloc(sizeof(*debug), GFP_KERNEL);
if (!debug)
goto out;
/* Round to page boundary */
debug->buffer = kmalloc(LPFC_HDWQINFO_SIZE, GFP_KERNEL);
if (!debug->buffer) {
kfree(debug);
goto out;
}
debug->len = lpfc_debugfs_lockstat_data(phba, debug->buffer,
LPFC_HBQINFO_SIZE);
file->private_data = debug;
rc = 0;
out:
return rc;
}
static ssize_t
lpfc_debugfs_lockstat_write(struct file *file, const char __user *buf,
size_t nbytes, loff_t *ppos)
{
struct lpfc_debug *debug = file->private_data;
struct lpfc_hba *phba = (struct lpfc_hba *)debug->i_private;
struct lpfc_sli4_hdw_queue *qp;
char mybuf[64];
char *pbuf;
int i;
/* Protect copy from user */
if (!access_ok(buf, nbytes))
return -EFAULT;
memset(mybuf, 0, sizeof(mybuf));
if (copy_from_user(mybuf, buf, nbytes))
return -EFAULT;
pbuf = &mybuf[0];
if ((strncmp(pbuf, "reset", strlen("reset")) == 0) ||
(strncmp(pbuf, "zero", strlen("zero")) == 0)) {
for (i = 0; i < phba->cfg_hdw_queue; i++) {
qp = &phba->sli4_hba.hdwq[i];
qp->lock_conflict.alloc_xri_get = 0;
qp->lock_conflict.alloc_xri_put = 0;
qp->lock_conflict.free_xri = 0;
qp->lock_conflict.wq_access = 0;
qp->lock_conflict.alloc_pvt_pool = 0;
qp->lock_conflict.mv_from_pvt_pool = 0;
qp->lock_conflict.mv_to_pub_pool = 0;
qp->lock_conflict.mv_to_pvt_pool = 0;
qp->lock_conflict.free_pvt_pool = 0;
qp->lock_conflict.free_pub_pool = 0;
qp->lock_conflict.wq_access = 0;
}
}
return nbytes;
}
#endif
/**
* lpfc_debugfs_dumpHBASlim_open - Open the Dump HBA SLIM debugfs buffer
* @inode: The inode pointer that contains a vport pointer.
* @file: The file pointer to attach the log output.
*
* Description:
* This routine is the entry point for the debugfs open file operation. It gets
* the vport from the i_private field in @inode, allocates the necessary buffer
* for the log, fills the buffer from the in-memory log for this vport, and then
* returns a pointer to that log in the private_data field in @file.
*
* Returns:
* This function returns zero if successful. On error it will return a negative
* error value.
**/
static int
lpfc_debugfs_dumpHBASlim_open(struct inode *inode, struct file *file)
{
struct lpfc_hba *phba = inode->i_private;
struct lpfc_debug *debug;
int rc = -ENOMEM;
debug = kmalloc(sizeof(*debug), GFP_KERNEL);
if (!debug)
goto out;
/* Round to page boundary */
debug->buffer = kmalloc(LPFC_DUMPHBASLIM_SIZE, GFP_KERNEL);
if (!debug->buffer) {
kfree(debug);
goto out;
}
debug->len = lpfc_debugfs_dumpHBASlim_data(phba, debug->buffer,
LPFC_DUMPHBASLIM_SIZE);
file->private_data = debug;
rc = 0;
out:
return rc;
}
/**
* lpfc_debugfs_dumpHostSlim_open - Open the Dump Host SLIM debugfs buffer
* @inode: The inode pointer that contains a vport pointer.
* @file: The file pointer to attach the log output.
*
* Description:
* This routine is the entry point for the debugfs open file operation. It gets
* the vport from the i_private field in @inode, allocates the necessary buffer
* for the log, fills the buffer from the in-memory log for this vport, and then
* returns a pointer to that log in the private_data field in @file.
*
* Returns:
* This function returns zero if successful. On error it will return a negative
* error value.
**/
static int
lpfc_debugfs_dumpHostSlim_open(struct inode *inode, struct file *file)
{
struct lpfc_hba *phba = inode->i_private;
struct lpfc_debug *debug;
int rc = -ENOMEM;
debug = kmalloc(sizeof(*debug), GFP_KERNEL);
if (!debug)
goto out;
/* Round to page boundary */
debug->buffer = kmalloc(LPFC_DUMPHOSTSLIM_SIZE, GFP_KERNEL);
if (!debug->buffer) {
kfree(debug);
goto out;
}
debug->len = lpfc_debugfs_dumpHostSlim_data(phba, debug->buffer,
LPFC_DUMPHOSTSLIM_SIZE);
file->private_data = debug;
rc = 0;
out:
return rc;
}
static int
lpfc_debugfs_dumpData_open(struct inode *inode, struct file *file)
{
struct lpfc_debug *debug;
int rc = -ENOMEM;
if (!_dump_buf_data)
return -EBUSY;
debug = kmalloc(sizeof(*debug), GFP_KERNEL);
if (!debug)
goto out;
/* Round to page boundary */
pr_err("9059 BLKGRD: %s: _dump_buf_data=0x%p\n",
__func__, _dump_buf_data);
debug->buffer = _dump_buf_data;
if (!debug->buffer) {
kfree(debug);
goto out;
}
debug->len = (1 << _dump_buf_data_order) << PAGE_SHIFT;
file->private_data = debug;
rc = 0;
out:
return rc;
}
static int
lpfc_debugfs_dumpDif_open(struct inode *inode, struct file *file)
{
struct lpfc_debug *debug;
int rc = -ENOMEM;
if (!_dump_buf_dif)
return -EBUSY;
debug = kmalloc(sizeof(*debug), GFP_KERNEL);
if (!debug)
goto out;
/* Round to page boundary */
pr_err("9060 BLKGRD: %s: _dump_buf_dif=0x%p file=%pD\n",
__func__, _dump_buf_dif, file);
debug->buffer = _dump_buf_dif;
if (!debug->buffer) {
kfree(debug);
goto out;
}
debug->len = (1 << _dump_buf_dif_order) << PAGE_SHIFT;
file->private_data = debug;
rc = 0;
out:
return rc;
}
static ssize_t
lpfc_debugfs_dumpDataDif_write(struct file *file, const char __user *buf,
size_t nbytes, loff_t *ppos)
{
/*
* The Data/DIF buffers only save one failing IO
* The write op is used as a reset mechanism after an IO has
* already been saved to the next one can be saved
*/
spin_lock(&_dump_buf_lock);
memset((void *)_dump_buf_data, 0,
((1 << PAGE_SHIFT) << _dump_buf_data_order));
memset((void *)_dump_buf_dif, 0,
((1 << PAGE_SHIFT) << _dump_buf_dif_order));
_dump_buf_done = 0;
spin_unlock(&_dump_buf_lock);
return nbytes;
}
static ssize_t
lpfc_debugfs_dif_err_read(struct file *file, char __user *buf,
size_t nbytes, loff_t *ppos)
{
struct dentry *dent = file->f_path.dentry;
struct lpfc_hba *phba = file->private_data;
char cbuf[32];
uint64_t tmp = 0;
int cnt = 0;
if (dent == phba->debug_writeGuard)
cnt = snprintf(cbuf, 32, "%u\n", phba->lpfc_injerr_wgrd_cnt);
else if (dent == phba->debug_writeApp)
cnt = snprintf(cbuf, 32, "%u\n", phba->lpfc_injerr_wapp_cnt);
else if (dent == phba->debug_writeRef)
cnt = snprintf(cbuf, 32, "%u\n", phba->lpfc_injerr_wref_cnt);
else if (dent == phba->debug_readGuard)
cnt = snprintf(cbuf, 32, "%u\n", phba->lpfc_injerr_rgrd_cnt);
else if (dent == phba->debug_readApp)
cnt = snprintf(cbuf, 32, "%u\n", phba->lpfc_injerr_rapp_cnt);
else if (dent == phba->debug_readRef)
cnt = snprintf(cbuf, 32, "%u\n", phba->lpfc_injerr_rref_cnt);
else if (dent == phba->debug_InjErrNPortID)
cnt = snprintf(cbuf, 32, "0x%06x\n", phba->lpfc_injerr_nportid);
else if (dent == phba->debug_InjErrWWPN) {
memcpy(&tmp, &phba->lpfc_injerr_wwpn, sizeof(struct lpfc_name));
tmp = cpu_to_be64(tmp);
cnt = snprintf(cbuf, 32, "0x%016llx\n", tmp);
} else if (dent == phba->debug_InjErrLBA) {
if (phba->lpfc_injerr_lba == (sector_t)(-1))
cnt = snprintf(cbuf, 32, "off\n");
else
cnt = snprintf(cbuf, 32, "0x%llx\n",
(uint64_t) phba->lpfc_injerr_lba);
} else
lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
"0547 Unknown debugfs error injection entry\n");
return simple_read_from_buffer(buf, nbytes, ppos, &cbuf, cnt);
}
static ssize_t
lpfc_debugfs_dif_err_write(struct file *file, const char __user *buf,
size_t nbytes, loff_t *ppos)
{
struct dentry *dent = file->f_path.dentry;
struct lpfc_hba *phba = file->private_data;
char dstbuf[33];
uint64_t tmp = 0;
int size;
memset(dstbuf, 0, 33);
size = (nbytes < 32) ? nbytes : 32;
if (copy_from_user(dstbuf, buf, size))
return 0;
if (dent == phba->debug_InjErrLBA) {
if ((buf[0] == 'o') && (buf[1] == 'f') && (buf[2] == 'f'))
tmp = (uint64_t)(-1);
}
if ((tmp == 0) && (kstrtoull(dstbuf, 0, &tmp)))
return 0;
if (dent == phba->debug_writeGuard)
phba->lpfc_injerr_wgrd_cnt = (uint32_t)tmp;
else if (dent == phba->debug_writeApp)
phba->lpfc_injerr_wapp_cnt = (uint32_t)tmp;
else if (dent == phba->debug_writeRef)
phba->lpfc_injerr_wref_cnt = (uint32_t)tmp;
else if (dent == phba->debug_readGuard)
phba->lpfc_injerr_rgrd_cnt = (uint32_t)tmp;
else if (dent == phba->debug_readApp)
phba->lpfc_injerr_rapp_cnt = (uint32_t)tmp;
else if (dent == phba->debug_readRef)
phba->lpfc_injerr_rref_cnt = (uint32_t)tmp;
else if (dent == phba->debug_InjErrLBA)
phba->lpfc_injerr_lba = (sector_t)tmp;
else if (dent == phba->debug_InjErrNPortID)
phba->lpfc_injerr_nportid = (uint32_t)(tmp & Mask_DID);
else if (dent == phba->debug_InjErrWWPN) {
tmp = cpu_to_be64(tmp);
memcpy(&phba->lpfc_injerr_wwpn, &tmp, sizeof(struct lpfc_name));
} else
lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
"0548 Unknown debugfs error injection entry\n");
return nbytes;
}
static int
lpfc_debugfs_dif_err_release(struct inode *inode, struct file *file)
{
return 0;
}
/**
* lpfc_debugfs_nodelist_open - Open the nodelist debugfs file
* @inode: The inode pointer that contains a vport pointer.
* @file: The file pointer to attach the log output.
*
* Description:
* This routine is the entry point for the debugfs open file operation. It gets
* the vport from the i_private field in @inode, allocates the necessary buffer
* for the log, fills the buffer from the in-memory log for this vport, and then
* returns a pointer to that log in the private_data field in @file.
*
* Returns:
* This function returns zero if successful. On error it will return a negative
* error value.
**/
static int
lpfc_debugfs_nodelist_open(struct inode *inode, struct file *file)
{
struct lpfc_vport *vport = inode->i_private;
struct lpfc_debug *debug;
int rc = -ENOMEM;
debug = kmalloc(sizeof(*debug), GFP_KERNEL);
if (!debug)
goto out;
/* Round to page boundary */
debug->buffer = kmalloc(LPFC_NODELIST_SIZE, GFP_KERNEL);
if (!debug->buffer) {
kfree(debug);
goto out;
}
debug->len = lpfc_debugfs_nodelist_data(vport, debug->buffer,
LPFC_NODELIST_SIZE);
file->private_data = debug;
rc = 0;
out:
return rc;
}
/**
* lpfc_debugfs_lseek - Seek through a debugfs file
* @file: The file pointer to seek through.
* @off: The offset to seek to or the amount to seek by.
* @whence: Indicates how to seek.
*
* Description:
* This routine is the entry point for the debugfs lseek file operation. The
* @whence parameter indicates whether @off is the offset to directly seek to,
* or if it is a value to seek forward or reverse by. This function figures out
* what the new offset of the debugfs file will be and assigns that value to the
* f_pos field of @file.
*
* Returns:
* This function returns the new offset if successful and returns a negative
* error if unable to process the seek.
**/
static loff_t
lpfc_debugfs_lseek(struct file *file, loff_t off, int whence)
{
struct lpfc_debug *debug = file->private_data;
return fixed_size_llseek(file, off, whence, debug->len);
}
/**
* lpfc_debugfs_read - Read a debugfs file
* @file: The file pointer to read from.
* @buf: The buffer to copy the data to.
* @nbytes: The number of bytes to read.
* @ppos: The position in the file to start reading from.
*
* Description:
* This routine reads data from from the buffer indicated in the private_data
* field of @file. It will start reading at @ppos and copy up to @nbytes of
* data to @buf.
*
* Returns:
* This function returns the amount of data that was read (this could be less
* than @nbytes if the end of the file was reached) or a negative error value.
**/
static ssize_t
lpfc_debugfs_read(struct file *file, char __user *buf,
size_t nbytes, loff_t *ppos)
{
struct lpfc_debug *debug = file->private_data;
return simple_read_from_buffer(buf, nbytes, ppos, debug->buffer,
debug->len);
}
/**
* lpfc_debugfs_release - Release the buffer used to store debugfs file data
* @inode: The inode pointer that contains a vport pointer. (unused)
* @file: The file pointer that contains the buffer to release.
*
* Description:
* This routine frees the buffer that was allocated when the debugfs file was
* opened.
*
* Returns:
* This function returns zero.
**/
static int
lpfc_debugfs_release(struct inode *inode, struct file *file)
{
struct lpfc_debug *debug = file->private_data;
kfree(debug->buffer);
kfree(debug);
return 0;
}
static int
lpfc_debugfs_dumpDataDif_release(struct inode *inode, struct file *file)
{
struct lpfc_debug *debug = file->private_data;
debug->buffer = NULL;
kfree(debug);
return 0;
}
scsi: lpfc: Adapt partitioned XRI lists to efficient sharing The XRI get/put lists were partitioned per hardware queue. However, the adapter rarely had sufficient resources to give a large number of resources per queue. As such, it became common for a cpu to encounter a lack of XRI resource and request the upper io stack to retry after returning a BUSY condition. This occurred even though other cpus were idle and not using their resources. Create as efficient a scheme as possible to move resources to the cpus that need them. Each cpu maintains a small private pool which it allocates from for io. There is a watermark that the cpu attempts to keep in the private pool. The private pool, when empty, pulls from a global pool from the cpu. When the cpu's global pool is empty it will pull from other cpu's global pool. As there many cpu global pools (1 per cpu or hardware queue count) and as each cpu selects what cpu to pull from at different rates and at different times, it creates a radomizing effect that minimizes the number of cpu's that will contend with each other when the steal XRI's from another cpu's global pool. On io completion, a cpu will push the XRI back on to its private pool. A watermark level is maintained for the private pool such that when it is exceeded it will move XRI's to the CPU global pool so that other cpu's may allocate them. On NVME, as heartbeat commands are critical to get placed on the wire, a single expedite pool is maintained. When a heartbeat is to be sent, it will allocate an XRI from the expedite pool rather than the normal cpu private/global pools. On any io completion, if a reduction in the expedite pools is seen, it will be replenished before the XRI is placed on the cpu private pool. Statistics are added to aid understanding the XRI levels on each cpu and their behaviors. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <jsmart2021@gmail.com> Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2019-01-29 03:14:28 +08:00
/**
* lpfc_debugfs_multixripools_write - Clear multi-XRI pools statistics
* @file: The file pointer to read from.
* @buf: The buffer to copy the user data from.
* @nbytes: The number of bytes to get.
* @ppos: The position in the file to start reading from.
*
* Description:
* This routine clears multi-XRI pools statistics when buf contains "clear".
*
* Return Value:
* It returns the @nbytges passing in from debugfs user space when successful.
* In case of error conditions, it returns proper error code back to the user
* space.
**/
static ssize_t
lpfc_debugfs_multixripools_write(struct file *file, const char __user *buf,
size_t nbytes, loff_t *ppos)
{
struct lpfc_debug *debug = file->private_data;
struct lpfc_hba *phba = (struct lpfc_hba *)debug->i_private;
char mybuf[64];
char *pbuf;
u32 i;
u32 hwq_count;
struct lpfc_sli4_hdw_queue *qp;
struct lpfc_multixri_pool *multixri_pool;
if (nbytes > 64)
nbytes = 64;
/* Protect copy from user */
if (!access_ok(buf, nbytes))
return -EFAULT;
memset(mybuf, 0, sizeof(mybuf));
if (copy_from_user(mybuf, buf, nbytes))
return -EFAULT;
pbuf = &mybuf[0];
if ((strncmp(pbuf, "clear", strlen("clear"))) == 0) {
hwq_count = phba->cfg_hdw_queue;
for (i = 0; i < hwq_count; i++) {
qp = &phba->sli4_hba.hdwq[i];
multixri_pool = qp->p_multixri_pool;
if (!multixri_pool)
continue;
qp->empty_io_bufs = 0;
multixri_pool->pbl_empty_count = 0;
#ifdef LPFC_MXP_STAT
multixri_pool->above_limit_count = 0;
multixri_pool->below_limit_count = 0;
multixri_pool->stat_max_hwm = 0;
multixri_pool->local_pbl_hit_count = 0;
multixri_pool->other_pbl_hit_count = 0;
multixri_pool->stat_pbl_count = 0;
multixri_pool->stat_pvt_count = 0;
multixri_pool->stat_busy_count = 0;
multixri_pool->stat_snapshot_taken = 0;
#endif
}
return strlen(pbuf);
}
return -EINVAL;
}
static int
lpfc_debugfs_nvmestat_open(struct inode *inode, struct file *file)
{
struct lpfc_vport *vport = inode->i_private;
struct lpfc_debug *debug;
int rc = -ENOMEM;
debug = kmalloc(sizeof(*debug), GFP_KERNEL);
if (!debug)
goto out;
/* Round to page boundary */
debug->buffer = kmalloc(LPFC_NVMESTAT_SIZE, GFP_KERNEL);
if (!debug->buffer) {
kfree(debug);
goto out;
}
debug->len = lpfc_debugfs_nvmestat_data(vport, debug->buffer,
LPFC_NVMESTAT_SIZE);
debug->i_private = inode->i_private;
file->private_data = debug;
rc = 0;
out:
return rc;
}
static ssize_t
lpfc_debugfs_nvmestat_write(struct file *file, const char __user *buf,
size_t nbytes, loff_t *ppos)
{
struct lpfc_debug *debug = file->private_data;
struct lpfc_vport *vport = (struct lpfc_vport *)debug->i_private;
struct lpfc_hba *phba = vport->phba;
struct lpfc_nvmet_tgtport *tgtp;
char mybuf[64];
char *pbuf;
if (!phba->targetport)
return -ENXIO;
if (nbytes > 64)
nbytes = 64;
memset(mybuf, 0, sizeof(mybuf));
if (copy_from_user(mybuf, buf, nbytes))
return -EFAULT;
pbuf = &mybuf[0];
tgtp = (struct lpfc_nvmet_tgtport *)phba->targetport->private;
if ((strncmp(pbuf, "reset", strlen("reset")) == 0) ||
(strncmp(pbuf, "zero", strlen("zero")) == 0)) {
atomic_set(&tgtp->rcv_ls_req_in, 0);
atomic_set(&tgtp->rcv_ls_req_out, 0);
atomic_set(&tgtp->rcv_ls_req_drop, 0);
atomic_set(&tgtp->xmt_ls_abort, 0);
atomic_set(&tgtp->xmt_ls_abort_cmpl, 0);
atomic_set(&tgtp->xmt_ls_rsp, 0);
atomic_set(&tgtp->xmt_ls_drop, 0);
atomic_set(&tgtp->xmt_ls_rsp_error, 0);
atomic_set(&tgtp->xmt_ls_rsp_cmpl, 0);
atomic_set(&tgtp->rcv_fcp_cmd_in, 0);
atomic_set(&tgtp->rcv_fcp_cmd_out, 0);
atomic_set(&tgtp->rcv_fcp_cmd_drop, 0);
atomic_set(&tgtp->xmt_fcp_drop, 0);
atomic_set(&tgtp->xmt_fcp_read_rsp, 0);
atomic_set(&tgtp->xmt_fcp_read, 0);
atomic_set(&tgtp->xmt_fcp_write, 0);
atomic_set(&tgtp->xmt_fcp_rsp, 0);
atomic_set(&tgtp->xmt_fcp_release, 0);
atomic_set(&tgtp->xmt_fcp_rsp_cmpl, 0);
atomic_set(&tgtp->xmt_fcp_rsp_error, 0);
atomic_set(&tgtp->xmt_fcp_rsp_drop, 0);
atomic_set(&tgtp->xmt_fcp_abort, 0);
atomic_set(&tgtp->xmt_fcp_abort_cmpl, 0);
atomic_set(&tgtp->xmt_abort_sol, 0);
atomic_set(&tgtp->xmt_abort_unsol, 0);
atomic_set(&tgtp->xmt_abort_rsp, 0);
atomic_set(&tgtp->xmt_abort_rsp_error, 0);
}
return nbytes;
}
static int
lpfc_debugfs_scsistat_open(struct inode *inode, struct file *file)
{
struct lpfc_vport *vport = inode->i_private;
struct lpfc_debug *debug;
int rc = -ENOMEM;
debug = kmalloc(sizeof(*debug), GFP_KERNEL);
if (!debug)
goto out;
/* Round to page boundary */
debug->buffer = kzalloc(LPFC_SCSISTAT_SIZE, GFP_KERNEL);
if (!debug->buffer) {
kfree(debug);
goto out;
}
debug->len = lpfc_debugfs_scsistat_data(vport, debug->buffer,
LPFC_SCSISTAT_SIZE);
debug->i_private = inode->i_private;
file->private_data = debug;
rc = 0;
out:
return rc;
}
static ssize_t
lpfc_debugfs_scsistat_write(struct file *file, const char __user *buf,
size_t nbytes, loff_t *ppos)
{
struct lpfc_debug *debug = file->private_data;
struct lpfc_vport *vport = (struct lpfc_vport *)debug->i_private;
struct lpfc_hba *phba = vport->phba;
char mybuf[6] = {0};
int i;
/* Protect copy from user */
if (!access_ok(buf, nbytes))
return -EFAULT;
if (copy_from_user(mybuf, buf, (nbytes >= sizeof(mybuf)) ?
(sizeof(mybuf) - 1) : nbytes))
return -EFAULT;
if ((strncmp(&mybuf[0], "reset", strlen("reset")) == 0) ||
(strncmp(&mybuf[0], "zero", strlen("zero")) == 0)) {
for (i = 0; i < phba->cfg_hdw_queue; i++) {
memset(&phba->sli4_hba.hdwq[i].scsi_cstat, 0,
sizeof(phba->sli4_hba.hdwq[i].scsi_cstat));
}
}
return nbytes;
}
static int
lpfc_debugfs_nvmektime_open(struct inode *inode, struct file *file)
{
struct lpfc_vport *vport = inode->i_private;
struct lpfc_debug *debug;
int rc = -ENOMEM;
debug = kmalloc(sizeof(*debug), GFP_KERNEL);
if (!debug)
goto out;
/* Round to page boundary */
debug->buffer = kmalloc(LPFC_NVMEKTIME_SIZE, GFP_KERNEL);
if (!debug->buffer) {
kfree(debug);
goto out;
}
debug->len = lpfc_debugfs_nvmektime_data(vport, debug->buffer,
LPFC_NVMEKTIME_SIZE);
debug->i_private = inode->i_private;
file->private_data = debug;
rc = 0;
out:
return rc;
}
static ssize_t
lpfc_debugfs_nvmektime_write(struct file *file, const char __user *buf,
size_t nbytes, loff_t *ppos)
{
struct lpfc_debug *debug = file->private_data;
struct lpfc_vport *vport = (struct lpfc_vport *)debug->i_private;
struct lpfc_hba *phba = vport->phba;
char mybuf[64];
char *pbuf;
if (nbytes > 64)
nbytes = 64;
memset(mybuf, 0, sizeof(mybuf));
if (copy_from_user(mybuf, buf, nbytes))
return -EFAULT;
pbuf = &mybuf[0];
if ((strncmp(pbuf, "on", sizeof("on") - 1) == 0)) {
phba->ktime_data_samples = 0;
phba->ktime_status_samples = 0;
phba->ktime_seg1_total = 0;
phba->ktime_seg1_max = 0;
phba->ktime_seg1_min = 0xffffffff;
phba->ktime_seg2_total = 0;
phba->ktime_seg2_max = 0;
phba->ktime_seg2_min = 0xffffffff;
phba->ktime_seg3_total = 0;
phba->ktime_seg3_max = 0;
phba->ktime_seg3_min = 0xffffffff;
phba->ktime_seg4_total = 0;
phba->ktime_seg4_max = 0;
phba->ktime_seg4_min = 0xffffffff;
phba->ktime_seg5_total = 0;
phba->ktime_seg5_max = 0;
phba->ktime_seg5_min = 0xffffffff;
phba->ktime_seg6_total = 0;
phba->ktime_seg6_max = 0;
phba->ktime_seg6_min = 0xffffffff;
phba->ktime_seg7_total = 0;
phba->ktime_seg7_max = 0;
phba->ktime_seg7_min = 0xffffffff;
phba->ktime_seg8_total = 0;
phba->ktime_seg8_max = 0;
phba->ktime_seg8_min = 0xffffffff;
phba->ktime_seg9_total = 0;
phba->ktime_seg9_max = 0;
phba->ktime_seg9_min = 0xffffffff;
phba->ktime_seg10_total = 0;
phba->ktime_seg10_max = 0;
phba->ktime_seg10_min = 0xffffffff;
phba->ktime_on = 1;
return strlen(pbuf);
} else if ((strncmp(pbuf, "off",
sizeof("off") - 1) == 0)) {
phba->ktime_on = 0;
return strlen(pbuf);
} else if ((strncmp(pbuf, "zero",
sizeof("zero") - 1) == 0)) {
phba->ktime_data_samples = 0;
phba->ktime_status_samples = 0;
phba->ktime_seg1_total = 0;
phba->ktime_seg1_max = 0;
phba->ktime_seg1_min = 0xffffffff;
phba->ktime_seg2_total = 0;
phba->ktime_seg2_max = 0;
phba->ktime_seg2_min = 0xffffffff;
phba->ktime_seg3_total = 0;
phba->ktime_seg3_max = 0;
phba->ktime_seg3_min = 0xffffffff;
phba->ktime_seg4_total = 0;
phba->ktime_seg4_max = 0;
phba->ktime_seg4_min = 0xffffffff;
phba->ktime_seg5_total = 0;
phba->ktime_seg5_max = 0;
phba->ktime_seg5_min = 0xffffffff;
phba->ktime_seg6_total = 0;
phba->ktime_seg6_max = 0;
phba->ktime_seg6_min = 0xffffffff;
phba->ktime_seg7_total = 0;
phba->ktime_seg7_max = 0;
phba->ktime_seg7_min = 0xffffffff;
phba->ktime_seg8_total = 0;
phba->ktime_seg8_max = 0;
phba->ktime_seg8_min = 0xffffffff;
phba->ktime_seg9_total = 0;
phba->ktime_seg9_max = 0;
phba->ktime_seg9_min = 0xffffffff;
phba->ktime_seg10_total = 0;
phba->ktime_seg10_max = 0;
phba->ktime_seg10_min = 0xffffffff;
return strlen(pbuf);
}
return -EINVAL;
}
static int
lpfc_debugfs_nvmeio_trc_open(struct inode *inode, struct file *file)
{
struct lpfc_hba *phba = inode->i_private;
struct lpfc_debug *debug;
int rc = -ENOMEM;
debug = kmalloc(sizeof(*debug), GFP_KERNEL);
if (!debug)
goto out;
/* Round to page boundary */
debug->buffer = kmalloc(LPFC_NVMEIO_TRC_SIZE, GFP_KERNEL);
if (!debug->buffer) {
kfree(debug);
goto out;
}
debug->len = lpfc_debugfs_nvmeio_trc_data(phba, debug->buffer,
LPFC_NVMEIO_TRC_SIZE);
debug->i_private = inode->i_private;
file->private_data = debug;
rc = 0;
out:
return rc;
}
static ssize_t
lpfc_debugfs_nvmeio_trc_write(struct file *file, const char __user *buf,
size_t nbytes, loff_t *ppos)
{
struct lpfc_debug *debug = file->private_data;
struct lpfc_hba *phba = (struct lpfc_hba *)debug->i_private;
int i;
unsigned long sz;
char mybuf[64];
char *pbuf;
if (nbytes > 64)
nbytes = 64;
memset(mybuf, 0, sizeof(mybuf));
if (copy_from_user(mybuf, buf, nbytes))
return -EFAULT;
pbuf = &mybuf[0];
if ((strncmp(pbuf, "off", sizeof("off") - 1) == 0)) {
lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
"0570 nvmeio_trc_off\n");
phba->nvmeio_trc_output_idx = 0;
phba->nvmeio_trc_on = 0;
return strlen(pbuf);
} else if ((strncmp(pbuf, "on", sizeof("on") - 1) == 0)) {
lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
"0571 nvmeio_trc_on\n");
phba->nvmeio_trc_output_idx = 0;
phba->nvmeio_trc_on = 1;
return strlen(pbuf);
}
/* We must be off to allocate the trace buffer */
if (phba->nvmeio_trc_on != 0)
return -EINVAL;
/* If not on or off, the parameter is the trace buffer size */
i = kstrtoul(pbuf, 0, &sz);
if (i)
return -EINVAL;
phba->nvmeio_trc_size = (uint32_t)sz;
/* It must be a power of 2 - round down */
i = 0;
while (sz > 1) {
sz = sz >> 1;
i++;
}
sz = (1 << i);
if (phba->nvmeio_trc_size != sz)
lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
"0572 nvmeio_trc_size changed to %ld\n",
sz);
phba->nvmeio_trc_size = (uint32_t)sz;
/* If one previously exists, free it */
kfree(phba->nvmeio_trc);
/* Allocate new trace buffer and initialize */
phba->nvmeio_trc = kzalloc((sizeof(struct lpfc_debugfs_nvmeio_trc) *
sz), GFP_KERNEL);
if (!phba->nvmeio_trc) {
lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
"0573 Cannot create debugfs "
"nvmeio_trc buffer\n");
return -ENOMEM;
}
atomic_set(&phba->nvmeio_trc_cnt, 0);
phba->nvmeio_trc_on = 0;
phba->nvmeio_trc_output_idx = 0;
return strlen(pbuf);
}
static int
lpfc_debugfs_cpucheck_open(struct inode *inode, struct file *file)
{
struct lpfc_vport *vport = inode->i_private;
struct lpfc_debug *debug;
int rc = -ENOMEM;
debug = kmalloc(sizeof(*debug), GFP_KERNEL);
if (!debug)
goto out;
/* Round to page boundary */
debug->buffer = kmalloc(LPFC_CPUCHECK_SIZE, GFP_KERNEL);
if (!debug->buffer) {
kfree(debug);
goto out;
}
debug->len = lpfc_debugfs_cpucheck_data(vport, debug->buffer,
LPFC_CPUCHECK_SIZE);
debug->i_private = inode->i_private;
file->private_data = debug;
rc = 0;
out:
return rc;
}
static ssize_t
lpfc_debugfs_cpucheck_write(struct file *file, const char __user *buf,
size_t nbytes, loff_t *ppos)
{
struct lpfc_debug *debug = file->private_data;
struct lpfc_vport *vport = (struct lpfc_vport *)debug->i_private;
struct lpfc_hba *phba = vport->phba;
struct lpfc_sli4_hdw_queue *qp;
char mybuf[64];
char *pbuf;
int i, j;
if (nbytes > 64)
nbytes = 64;
memset(mybuf, 0, sizeof(mybuf));
if (copy_from_user(mybuf, buf, nbytes))
return -EFAULT;
pbuf = &mybuf[0];
if ((strncmp(pbuf, "on", sizeof("on") - 1) == 0)) {
if (phba->nvmet_support)
phba->cpucheck_on |= LPFC_CHECK_NVMET_IO;
else
phba->cpucheck_on |= (LPFC_CHECK_NVME_IO |
LPFC_CHECK_SCSI_IO);
return strlen(pbuf);
} else if ((strncmp(pbuf, "nvme_on", sizeof("nvme_on") - 1) == 0)) {
if (phba->nvmet_support)
phba->cpucheck_on |= LPFC_CHECK_NVMET_IO;
else
phba->cpucheck_on |= LPFC_CHECK_NVME_IO;
return strlen(pbuf);
} else if ((strncmp(pbuf, "scsi_on", sizeof("scsi_on") - 1) == 0)) {
phba->cpucheck_on |= LPFC_CHECK_SCSI_IO;
return strlen(pbuf);
} else if ((strncmp(pbuf, "rcv",
sizeof("rcv") - 1) == 0)) {
if (phba->nvmet_support)
phba->cpucheck_on |= LPFC_CHECK_NVMET_RCV;
else
return -EINVAL;
return strlen(pbuf);
} else if ((strncmp(pbuf, "off",
sizeof("off") - 1) == 0)) {
phba->cpucheck_on = LPFC_CHECK_OFF;
return strlen(pbuf);
} else if ((strncmp(pbuf, "zero",
sizeof("zero") - 1) == 0)) {
for (i = 0; i < phba->cfg_hdw_queue; i++) {
qp = &phba->sli4_hba.hdwq[i];
for (j = 0; j < LPFC_CHECK_CPU_CNT; j++) {
qp->cpucheck_rcv_io[j] = 0;
qp->cpucheck_xmt_io[j] = 0;
qp->cpucheck_cmpl_io[j] = 0;
}
}
return strlen(pbuf);
}
return -EINVAL;
}
/*
* ---------------------------------
* iDiag debugfs file access methods
* ---------------------------------
*
* All access methods are through the proper SLI4 PCI function's debugfs
* iDiag directory:
*
* /sys/kernel/debug/lpfc/fn<#>/iDiag
*/
/**
* lpfc_idiag_cmd_get - Get and parse idiag debugfs comands from user space
* @buf: The pointer to the user space buffer.
* @nbytes: The number of bytes in the user space buffer.
* @idiag_cmd: pointer to the idiag command struct.
*
* This routine reads data from debugfs user space buffer and parses the
* buffer for getting the idiag command and arguments. The while space in
* between the set of data is used as the parsing separator.
*
* This routine returns 0 when successful, it returns proper error code
* back to the user space in error conditions.
*/
static int lpfc_idiag_cmd_get(const char __user *buf, size_t nbytes,
struct lpfc_idiag_cmd *idiag_cmd)
{
char mybuf[64];
char *pbuf, *step_str;
int i;
size_t bsize;
memset(mybuf, 0, sizeof(mybuf));
memset(idiag_cmd, 0, sizeof(*idiag_cmd));
bsize = min(nbytes, (sizeof(mybuf)-1));
if (copy_from_user(mybuf, buf, bsize))
return -EFAULT;
pbuf = &mybuf[0];
step_str = strsep(&pbuf, "\t ");
/* The opcode must present */
if (!step_str)
return -EINVAL;
idiag_cmd->opcode = simple_strtol(step_str, NULL, 0);
if (idiag_cmd->opcode == 0)
return -EINVAL;
for (i = 0; i < LPFC_IDIAG_CMD_DATA_SIZE; i++) {
step_str = strsep(&pbuf, "\t ");
if (!step_str)
return i;
idiag_cmd->data[i] = simple_strtol(step_str, NULL, 0);
}
return i;
}
/**
* lpfc_idiag_open - idiag open debugfs
* @inode: The inode pointer that contains a pointer to phba.
* @file: The file pointer to attach the file operation.
*
* Description:
* This routine is the entry point for the debugfs open file operation. It
* gets the reference to phba from the i_private field in @inode, it then
* allocates buffer for the file operation, performs the necessary PCI config
* space read into the allocated buffer according to the idiag user command
* setup, and then returns a pointer to buffer in the private_data field in
* @file.
*
* Returns:
* This function returns zero if successful. On error it will return an
* negative error value.
**/
static int
lpfc_idiag_open(struct inode *inode, struct file *file)
{
struct lpfc_debug *debug;
debug = kmalloc(sizeof(*debug), GFP_KERNEL);
if (!debug)
return -ENOMEM;
debug->i_private = inode->i_private;
debug->buffer = NULL;
file->private_data = debug;
return 0;
}
/**
* lpfc_idiag_release - Release idiag access file operation
* @inode: The inode pointer that contains a vport pointer. (unused)
* @file: The file pointer that contains the buffer to release.
*
* Description:
* This routine is the generic release routine for the idiag access file
* operation, it frees the buffer that was allocated when the debugfs file
* was opened.
*
* Returns:
* This function returns zero.
**/
static int
lpfc_idiag_release(struct inode *inode, struct file *file)
{
struct lpfc_debug *debug = file->private_data;
/* Free the buffers to the file operation */
kfree(debug->buffer);
kfree(debug);
return 0;
}
/**
* lpfc_idiag_cmd_release - Release idiag cmd access file operation
* @inode: The inode pointer that contains a vport pointer. (unused)
* @file: The file pointer that contains the buffer to release.
*
* Description:
* This routine frees the buffer that was allocated when the debugfs file
* was opened. It also reset the fields in the idiag command struct in the
* case of command for write operation.
*
* Returns:
* This function returns zero.
**/
static int
lpfc_idiag_cmd_release(struct inode *inode, struct file *file)
{
struct lpfc_debug *debug = file->private_data;
if (debug->op == LPFC_IDIAG_OP_WR) {
switch (idiag.cmd.opcode) {
case LPFC_IDIAG_CMD_PCICFG_WR:
case LPFC_IDIAG_CMD_PCICFG_ST:
case LPFC_IDIAG_CMD_PCICFG_CL:
case LPFC_IDIAG_CMD_QUEACC_WR:
case LPFC_IDIAG_CMD_QUEACC_ST:
case LPFC_IDIAG_CMD_QUEACC_CL:
memset(&idiag, 0, sizeof(idiag));
break;
default:
break;
}
}
/* Free the buffers to the file operation */
kfree(debug->buffer);
kfree(debug);
return 0;
}
/**
* lpfc_idiag_pcicfg_read - idiag debugfs read pcicfg
* @file: The file pointer to read from.
* @buf: The buffer to copy the data to.
* @nbytes: The number of bytes to read.
* @ppos: The position in the file to start reading from.
*
* Description:
* This routine reads data from the @phba pci config space according to the
* idiag command, and copies to user @buf. Depending on the PCI config space
* read command setup, it does either a single register read of a byte
* (8 bits), a word (16 bits), or a dword (32 bits) or browsing through all
* registers from the 4K extended PCI config space.
*
* Returns:
* This function returns the amount of data that was read (this could be less
* than @nbytes if the end of the file was reached) or a negative error value.
**/
static ssize_t
lpfc_idiag_pcicfg_read(struct file *file, char __user *buf, size_t nbytes,
loff_t *ppos)
{
struct lpfc_debug *debug = file->private_data;
struct lpfc_hba *phba = (struct lpfc_hba *)debug->i_private;
int offset_label, offset, len = 0, index = LPFC_PCI_CFG_RD_SIZE;
int where, count;
char *pbuffer;
struct pci_dev *pdev;
uint32_t u32val;
uint16_t u16val;
uint8_t u8val;
pdev = phba->pcidev;
if (!pdev)
return 0;
/* This is a user read operation */
debug->op = LPFC_IDIAG_OP_RD;
if (!debug->buffer)
debug->buffer = kmalloc(LPFC_PCI_CFG_SIZE, GFP_KERNEL);
if (!debug->buffer)
return 0;
pbuffer = debug->buffer;
if (*ppos)
return 0;
if (idiag.cmd.opcode == LPFC_IDIAG_CMD_PCICFG_RD) {
where = idiag.cmd.data[IDIAG_PCICFG_WHERE_INDX];
count = idiag.cmd.data[IDIAG_PCICFG_COUNT_INDX];
} else
return 0;
/* Read single PCI config space register */
switch (count) {
case SIZE_U8: /* byte (8 bits) */
pci_read_config_byte(pdev, where, &u8val);
len += snprintf(pbuffer+len, LPFC_PCI_CFG_SIZE-len,
"%03x: %02x\n", where, u8val);
break;
case SIZE_U16: /* word (16 bits) */
pci_read_config_word(pdev, where, &u16val);
len += snprintf(pbuffer+len, LPFC_PCI_CFG_SIZE-len,
"%03x: %04x\n", where, u16val);
break;
case SIZE_U32: /* double word (32 bits) */
pci_read_config_dword(pdev, where, &u32val);
len += snprintf(pbuffer+len, LPFC_PCI_CFG_SIZE-len,
"%03x: %08x\n", where, u32val);
break;
case LPFC_PCI_CFG_BROWSE: /* browse all */
goto pcicfg_browse;
break;
default:
/* illegal count */
len = 0;
break;
}
return simple_read_from_buffer(buf, nbytes, ppos, pbuffer, len);
pcicfg_browse:
/* Browse all PCI config space registers */
offset_label = idiag.offset.last_rd;
offset = offset_label;
/* Read PCI config space */
len += snprintf(pbuffer+len, LPFC_PCI_CFG_SIZE-len,
"%03x: ", offset_label);
while (index > 0) {
pci_read_config_dword(pdev, offset, &u32val);
len += snprintf(pbuffer+len, LPFC_PCI_CFG_SIZE-len,
"%08x ", u32val);
offset += sizeof(uint32_t);
if (offset >= LPFC_PCI_CFG_SIZE) {
len += snprintf(pbuffer+len,
LPFC_PCI_CFG_SIZE-len, "\n");
break;
}
index -= sizeof(uint32_t);
if (!index)
len += snprintf(pbuffer+len, LPFC_PCI_CFG_SIZE-len,
"\n");
else if (!(index % (8 * sizeof(uint32_t)))) {
offset_label += (8 * sizeof(uint32_t));
len += snprintf(pbuffer+len, LPFC_PCI_CFG_SIZE-len,
"\n%03x: ", offset_label);
}
}
/* Set up the offset for next portion of pci cfg read */
if (index == 0) {
idiag.offset.last_rd += LPFC_PCI_CFG_RD_SIZE;
if (idiag.offset.last_rd >= LPFC_PCI_CFG_SIZE)
idiag.offset.last_rd = 0;
} else
idiag.offset.last_rd = 0;
return simple_read_from_buffer(buf, nbytes, ppos, pbuffer, len);
}
/**
* lpfc_idiag_pcicfg_write - Syntax check and set up idiag pcicfg commands
* @file: The file pointer to read from.
* @buf: The buffer to copy the user data from.
* @nbytes: The number of bytes to get.
* @ppos: The position in the file to start reading from.
*
* This routine get the debugfs idiag command struct from user space and
* then perform the syntax check for PCI config space read or write command
* accordingly. In the case of PCI config space read command, it sets up
* the command in the idiag command struct for the debugfs read operation.
* In the case of PCI config space write operation, it executes the write
* operation into the PCI config space accordingly.
*
* It returns the @nbytges passing in from debugfs user space when successful.
* In case of error conditions, it returns proper error code back to the user
* space.
*/
static ssize_t
lpfc_idiag_pcicfg_write(struct file *file, const char __user *buf,
size_t nbytes, loff_t *ppos)
{
struct lpfc_debug *debug = file->private_data;
struct lpfc_hba *phba = (struct lpfc_hba *)debug->i_private;
uint32_t where, value, count;
uint32_t u32val;
uint16_t u16val;
uint8_t u8val;
struct pci_dev *pdev;
int rc;
pdev = phba->pcidev;
if (!pdev)
return -EFAULT;
/* This is a user write operation */
debug->op = LPFC_IDIAG_OP_WR;
rc = lpfc_idiag_cmd_get(buf, nbytes, &idiag.cmd);
if (rc < 0)
return rc;
if (idiag.cmd.opcode == LPFC_IDIAG_CMD_PCICFG_RD) {
/* Sanity check on PCI config read command line arguments */
if (rc != LPFC_PCI_CFG_RD_CMD_ARG)
goto error_out;
/* Read command from PCI config space, set up command fields */
where = idiag.cmd.data[IDIAG_PCICFG_WHERE_INDX];
count = idiag.cmd.data[IDIAG_PCICFG_COUNT_INDX];
if (count == LPFC_PCI_CFG_BROWSE) {
if (where % sizeof(uint32_t))
goto error_out;
/* Starting offset to browse */
idiag.offset.last_rd = where;
} else if ((count != sizeof(uint8_t)) &&
(count != sizeof(uint16_t)) &&
(count != sizeof(uint32_t)))
goto error_out;
if (count == sizeof(uint8_t)) {
if (where > LPFC_PCI_CFG_SIZE - sizeof(uint8_t))
goto error_out;
if (where % sizeof(uint8_t))
goto error_out;
}
if (count == sizeof(uint16_t)) {
if (where > LPFC_PCI_CFG_SIZE - sizeof(uint16_t))
goto error_out;
if (where % sizeof(uint16_t))
goto error_out;
}
if (count == sizeof(uint32_t)) {
if (where > LPFC_PCI_CFG_SIZE - sizeof(uint32_t))
goto error_out;
if (where % sizeof(uint32_t))
goto error_out;
}
} else if (idiag.cmd.opcode == LPFC_IDIAG_CMD_PCICFG_WR ||
idiag.cmd.opcode == LPFC_IDIAG_CMD_PCICFG_ST ||
idiag.cmd.opcode == LPFC_IDIAG_CMD_PCICFG_CL) {
/* Sanity check on PCI config write command line arguments */
if (rc != LPFC_PCI_CFG_WR_CMD_ARG)
goto error_out;
/* Write command to PCI config space, read-modify-write */
where = idiag.cmd.data[IDIAG_PCICFG_WHERE_INDX];
count = idiag.cmd.data[IDIAG_PCICFG_COUNT_INDX];
value = idiag.cmd.data[IDIAG_PCICFG_VALUE_INDX];
/* Sanity checks */
if ((count != sizeof(uint8_t)) &&
(count != sizeof(uint16_t)) &&
(count != sizeof(uint32_t)))
goto error_out;
if (count == sizeof(uint8_t)) {
if (where > LPFC_PCI_CFG_SIZE - sizeof(uint8_t))
goto error_out;
if (where % sizeof(uint8_t))
goto error_out;
if (idiag.cmd.opcode == LPFC_IDIAG_CMD_PCICFG_WR)
pci_write_config_byte(pdev, where,
(uint8_t)value);
if (idiag.cmd.opcode == LPFC_IDIAG_CMD_PCICFG_ST) {
rc = pci_read_config_byte(pdev, where, &u8val);
if (!rc) {
u8val |= (uint8_t)value;
pci_write_config_byte(pdev, where,
u8val);
}
}
if (idiag.cmd.opcode == LPFC_IDIAG_CMD_PCICFG_CL) {
rc = pci_read_config_byte(pdev, where, &u8val);
if (!rc) {
u8val &= (uint8_t)(~value);
pci_write_config_byte(pdev, where,
u8val);
}
}
}
if (count == sizeof(uint16_t)) {
if (where > LPFC_PCI_CFG_SIZE - sizeof(uint16_t))
goto error_out;
if (where % sizeof(uint16_t))
goto error_out;
if (idiag.cmd.opcode == LPFC_IDIAG_CMD_PCICFG_WR)
pci_write_config_word(pdev, where,
(uint16_t)value);
if (idiag.cmd.opcode == LPFC_IDIAG_CMD_PCICFG_ST) {
rc = pci_read_config_word(pdev, where, &u16val);
if (!rc) {
u16val |= (uint16_t)value;
pci_write_config_word(pdev, where,
u16val);
}
}
if (idiag.cmd.opcode == LPFC_IDIAG_CMD_PCICFG_CL) {
rc = pci_read_config_word(pdev, where, &u16val);
if (!rc) {
u16val &= (uint16_t)(~value);
pci_write_config_word(pdev, where,
u16val);
}
}
}
if (count == sizeof(uint32_t)) {
if (where > LPFC_PCI_CFG_SIZE - sizeof(uint32_t))
goto error_out;
if (where % sizeof(uint32_t))
goto error_out;
if (idiag.cmd.opcode == LPFC_IDIAG_CMD_PCICFG_WR)
pci_write_config_dword(pdev, where, value);
if (idiag.cmd.opcode == LPFC_IDIAG_CMD_PCICFG_ST) {
rc = pci_read_config_dword(pdev, where,
&u32val);
if (!rc) {
u32val |= value;
pci_write_config_dword(pdev, where,
u32val);
}
}
if (idiag.cmd.opcode == LPFC_IDIAG_CMD_PCICFG_CL) {
rc = pci_read_config_dword(pdev, where,
&u32val);
if (!rc) {
u32val &= ~value;
pci_write_config_dword(pdev, where,
u32val);
}
}
}
} else
/* All other opecodes are illegal for now */
goto error_out;
return nbytes;
error_out:
memset(&idiag, 0, sizeof(idiag));
return -EINVAL;
}
/**
* lpfc_idiag_baracc_read - idiag debugfs pci bar access read
* @file: The file pointer to read from.
* @buf: The buffer to copy the data to.
* @nbytes: The number of bytes to read.
* @ppos: The position in the file to start reading from.
*
* Description:
* This routine reads data from the @phba pci bar memory mapped space
* according to the idiag command, and copies to user @buf.
*
* Returns:
* This function returns the amount of data that was read (this could be less
* than @nbytes if the end of the file was reached) or a negative error value.
**/
static ssize_t
lpfc_idiag_baracc_read(struct file *file, char __user *buf, size_t nbytes,
loff_t *ppos)
{
struct lpfc_debug *debug = file->private_data;
struct lpfc_hba *phba = (struct lpfc_hba *)debug->i_private;
int offset_label, offset, offset_run, len = 0, index;
int bar_num, acc_range, bar_size;
char *pbuffer;
void __iomem *mem_mapped_bar;
uint32_t if_type;
struct pci_dev *pdev;
uint32_t u32val;
pdev = phba->pcidev;
if (!pdev)
return 0;
/* This is a user read operation */
debug->op = LPFC_IDIAG_OP_RD;
if (!debug->buffer)
debug->buffer = kmalloc(LPFC_PCI_BAR_RD_BUF_SIZE, GFP_KERNEL);
if (!debug->buffer)
return 0;
pbuffer = debug->buffer;
if (*ppos)
return 0;
if (idiag.cmd.opcode == LPFC_IDIAG_CMD_BARACC_RD) {
bar_num = idiag.cmd.data[IDIAG_BARACC_BAR_NUM_INDX];
offset = idiag.cmd.data[IDIAG_BARACC_OFF_SET_INDX];
acc_range = idiag.cmd.data[IDIAG_BARACC_ACC_MOD_INDX];
bar_size = idiag.cmd.data[IDIAG_BARACC_BAR_SZE_INDX];
} else
return 0;
if (acc_range == 0)
return 0;
if_type = bf_get(lpfc_sli_intf_if_type, &phba->sli4_hba.sli_intf);
if (if_type == LPFC_SLI_INTF_IF_TYPE_0) {
if (bar_num == IDIAG_BARACC_BAR_0)
mem_mapped_bar = phba->sli4_hba.conf_regs_memmap_p;
else if (bar_num == IDIAG_BARACC_BAR_1)
mem_mapped_bar = phba->sli4_hba.ctrl_regs_memmap_p;
else if (bar_num == IDIAG_BARACC_BAR_2)
mem_mapped_bar = phba->sli4_hba.drbl_regs_memmap_p;
else
return 0;
} else if (if_type == LPFC_SLI_INTF_IF_TYPE_2) {
if (bar_num == IDIAG_BARACC_BAR_0)
mem_mapped_bar = phba->sli4_hba.conf_regs_memmap_p;
else
return 0;
} else
return 0;
/* Read single PCI bar space register */
if (acc_range == SINGLE_WORD) {
offset_run = offset;
u32val = readl(mem_mapped_bar + offset_run);
len += snprintf(pbuffer+len, LPFC_PCI_BAR_RD_BUF_SIZE-len,
"%05x: %08x\n", offset_run, u32val);
} else
goto baracc_browse;
return simple_read_from_buffer(buf, nbytes, ppos, pbuffer, len);
baracc_browse:
/* Browse all PCI bar space registers */
offset_label = idiag.offset.last_rd;
offset_run = offset_label;
/* Read PCI bar memory mapped space */
len += snprintf(pbuffer+len, LPFC_PCI_BAR_RD_BUF_SIZE-len,
"%05x: ", offset_label);
index = LPFC_PCI_BAR_RD_SIZE;
while (index > 0) {
u32val = readl(mem_mapped_bar + offset_run);
len += snprintf(pbuffer+len, LPFC_PCI_BAR_RD_BUF_SIZE-len,
"%08x ", u32val);
offset_run += sizeof(uint32_t);
if (acc_range == LPFC_PCI_BAR_BROWSE) {
if (offset_run >= bar_size) {
len += snprintf(pbuffer+len,
LPFC_PCI_BAR_RD_BUF_SIZE-len, "\n");
break;
}
} else {
if (offset_run >= offset +
(acc_range * sizeof(uint32_t))) {
len += snprintf(pbuffer+len,
LPFC_PCI_BAR_RD_BUF_SIZE-len, "\n");
break;
}
}
index -= sizeof(uint32_t);
if (!index)
len += snprintf(pbuffer+len,
LPFC_PCI_BAR_RD_BUF_SIZE-len, "\n");
else if (!(index % (8 * sizeof(uint32_t)))) {
offset_label += (8 * sizeof(uint32_t));
len += snprintf(pbuffer+len,
LPFC_PCI_BAR_RD_BUF_SIZE-len,
"\n%05x: ", offset_label);
}
}
/* Set up the offset for next portion of pci bar read */
if (index == 0) {
idiag.offset.last_rd += LPFC_PCI_BAR_RD_SIZE;
if (acc_range == LPFC_PCI_BAR_BROWSE) {
if (idiag.offset.last_rd >= bar_size)
idiag.offset.last_rd = 0;
} else {
if (offset_run >= offset +
(acc_range * sizeof(uint32_t)))
idiag.offset.last_rd = offset;
}
} else {
if (acc_range == LPFC_PCI_BAR_BROWSE)
idiag.offset.last_rd = 0;
else
idiag.offset.last_rd = offset;
}
return simple_read_from_buffer(buf, nbytes, ppos, pbuffer, len);
}
/**
* lpfc_idiag_baracc_write - Syntax check and set up idiag bar access commands
* @file: The file pointer to read from.
* @buf: The buffer to copy the user data from.
* @nbytes: The number of bytes to get.
* @ppos: The position in the file to start reading from.
*
* This routine get the debugfs idiag command struct from user space and
* then perform the syntax check for PCI bar memory mapped space read or
* write command accordingly. In the case of PCI bar memory mapped space
* read command, it sets up the command in the idiag command struct for
* the debugfs read operation. In the case of PCI bar memorpy mapped space
* write operation, it executes the write operation into the PCI bar memory
* mapped space accordingly.
*
* It returns the @nbytges passing in from debugfs user space when successful.
* In case of error conditions, it returns proper error code back to the user
* space.
*/
static ssize_t
lpfc_idiag_baracc_write(struct file *file, const char __user *buf,
size_t nbytes, loff_t *ppos)
{
struct lpfc_debug *debug = file->private_data;
struct lpfc_hba *phba = (struct lpfc_hba *)debug->i_private;
uint32_t bar_num, bar_size, offset, value, acc_range;
struct pci_dev *pdev;
void __iomem *mem_mapped_bar;
uint32_t if_type;
uint32_t u32val;
int rc;
pdev = phba->pcidev;
if (!pdev)
return -EFAULT;
/* This is a user write operation */
debug->op = LPFC_IDIAG_OP_WR;
rc = lpfc_idiag_cmd_get(buf, nbytes, &idiag.cmd);
if (rc < 0)
return rc;
if_type = bf_get(lpfc_sli_intf_if_type, &phba->sli4_hba.sli_intf);
bar_num = idiag.cmd.data[IDIAG_BARACC_BAR_NUM_INDX];
if (if_type == LPFC_SLI_INTF_IF_TYPE_0) {
if ((bar_num != IDIAG_BARACC_BAR_0) &&
(bar_num != IDIAG_BARACC_BAR_1) &&
(bar_num != IDIAG_BARACC_BAR_2))
goto error_out;
} else if (if_type == LPFC_SLI_INTF_IF_TYPE_2) {
if (bar_num != IDIAG_BARACC_BAR_0)
goto error_out;
} else
goto error_out;
if (if_type == LPFC_SLI_INTF_IF_TYPE_0) {
if (bar_num == IDIAG_BARACC_BAR_0) {
idiag.cmd.data[IDIAG_BARACC_BAR_SZE_INDX] =
LPFC_PCI_IF0_BAR0_SIZE;
mem_mapped_bar = phba->sli4_hba.conf_regs_memmap_p;
} else if (bar_num == IDIAG_BARACC_BAR_1) {
idiag.cmd.data[IDIAG_BARACC_BAR_SZE_INDX] =
LPFC_PCI_IF0_BAR1_SIZE;
mem_mapped_bar = phba->sli4_hba.ctrl_regs_memmap_p;
} else if (bar_num == IDIAG_BARACC_BAR_2) {
idiag.cmd.data[IDIAG_BARACC_BAR_SZE_INDX] =
LPFC_PCI_IF0_BAR2_SIZE;
mem_mapped_bar = phba->sli4_hba.drbl_regs_memmap_p;
} else
goto error_out;
} else if (if_type == LPFC_SLI_INTF_IF_TYPE_2) {
if (bar_num == IDIAG_BARACC_BAR_0) {
idiag.cmd.data[IDIAG_BARACC_BAR_SZE_INDX] =
LPFC_PCI_IF2_BAR0_SIZE;
mem_mapped_bar = phba->sli4_hba.conf_regs_memmap_p;
} else
goto error_out;
} else
goto error_out;
offset = idiag.cmd.data[IDIAG_BARACC_OFF_SET_INDX];
if (offset % sizeof(uint32_t))
goto error_out;
bar_size = idiag.cmd.data[IDIAG_BARACC_BAR_SZE_INDX];
if (idiag.cmd.opcode == LPFC_IDIAG_CMD_BARACC_RD) {
/* Sanity check on PCI config read command line arguments */
if (rc != LPFC_PCI_BAR_RD_CMD_ARG)
goto error_out;
acc_range = idiag.cmd.data[IDIAG_BARACC_ACC_MOD_INDX];
if (acc_range == LPFC_PCI_BAR_BROWSE) {
if (offset > bar_size - sizeof(uint32_t))
goto error_out;
/* Starting offset to browse */
idiag.offset.last_rd = offset;
} else if (acc_range > SINGLE_WORD) {
if (offset + acc_range * sizeof(uint32_t) > bar_size)
goto error_out;
/* Starting offset to browse */
idiag.offset.last_rd = offset;
} else if (acc_range != SINGLE_WORD)
goto error_out;
} else if (idiag.cmd.opcode == LPFC_IDIAG_CMD_BARACC_WR ||
idiag.cmd.opcode == LPFC_IDIAG_CMD_BARACC_ST ||
idiag.cmd.opcode == LPFC_IDIAG_CMD_BARACC_CL) {
/* Sanity check on PCI bar write command line arguments */
if (rc != LPFC_PCI_BAR_WR_CMD_ARG)
goto error_out;
/* Write command to PCI bar space, read-modify-write */
acc_range = SINGLE_WORD;
value = idiag.cmd.data[IDIAG_BARACC_REG_VAL_INDX];
if (idiag.cmd.opcode == LPFC_IDIAG_CMD_BARACC_WR) {
writel(value, mem_mapped_bar + offset);
readl(mem_mapped_bar + offset);
}
if (idiag.cmd.opcode == LPFC_IDIAG_CMD_BARACC_ST) {
u32val = readl(mem_mapped_bar + offset);
u32val |= value;
writel(u32val, mem_mapped_bar + offset);
readl(mem_mapped_bar + offset);
}
if (idiag.cmd.opcode == LPFC_IDIAG_CMD_BARACC_CL) {
u32val = readl(mem_mapped_bar + offset);
u32val &= ~value;
writel(u32val, mem_mapped_bar + offset);
readl(mem_mapped_bar + offset);
}
} else
/* All other opecodes are illegal for now */
goto error_out;
return nbytes;
error_out:
memset(&idiag, 0, sizeof(idiag));
return -EINVAL;
}
static int
__lpfc_idiag_print_wq(struct lpfc_queue *qp, char *wqtype,
char *pbuffer, int len)
{
if (!qp)
return len;
len += snprintf(pbuffer + len, LPFC_QUE_INFO_GET_BUF_SIZE - len,
"\t\t%s WQ info: ", wqtype);
len += snprintf(pbuffer + len, LPFC_QUE_INFO_GET_BUF_SIZE - len,
"AssocCQID[%04d]: WQ-STAT[oflow:x%x posted:x%llx]\n",
qp->assoc_qid, qp->q_cnt_1,
(unsigned long long)qp->q_cnt_4);
len += snprintf(pbuffer + len, LPFC_QUE_INFO_GET_BUF_SIZE - len,
"\t\tWQID[%02d], QE-CNT[%04d], QE-SZ[%04d], "
scsi: lpfc: Rework EQ/CQ processing to address interrupt coalescing When driving high iop counts, auto_imax coalescing kicks in and drives the performance to extremely small iops levels. There are two issues: 1) auto_imax is enabled by default. The auto algorithm, when iops gets high, divides the iops by the hdwq count and uses that value to calculate EQ_Delay. The EQ_Delay is set uniformly on all EQs whether they have load or not. The EQ_delay is only manipulated every 5s (a long time). Thus there were large 5s swings of no interrupt delay followed by large/maximum delay, before repeating. 2) When processing a CQ, the driver got mixed up on the rate of when to ring the doorbell to keep the chip appraised of the eqe or cqe consumption as well as how how long to sit in the thread and process queue entries. Currently, the driver capped its work at 64 entries (very small) and exited/rearmed the CQ. Thus, on heavy loads, additional overheads were taken to exit and re-enter the interrupt handler. Worse, if in the large/maximum coalescing windows,k it could be a while before getting back to servicing. The issues are corrected by the following: - A change in defaults. Auto_imax is turned OFF and fcp_imax is set to 0. Thus all interrupts are immediate. - Cleanup of field names and their meanings. Existing names were non-intuitive or used for duplicate things. - Added max_proc_limit field, to control the length of time the handlers would service completions. - Reworked EQ handling: Added common routine that walks eq, applying notify interval and max processing limits. Use queue_claimed to claim ownership of the queue while processing. Always rearm the queue whenever the common routine is called. Rework queue element processing, namely to eliminate hba_index vs host_index. Only one index is necessary. The queue entry can be marked invalid and the host_index updated immediately after eqe processing. After rework, xx_release routines are now DB write functions. Renamed the routines as such. Moved lpfc_sli4_eq_flush(), which does similar action, to same area. Replaced the 2 individual loops that walk an eq with a call to the common routine. Slightly revised lpfc_sli4_hba_handle_eqe() calling syntax. Added per-cpu counters to detect interrupt rates and scale interrupt coalescing values. - Reworked CQ handling: Added common routine that walks cq, applying notify interval and max processing limits. Use queue_claimed to claim ownership of the queue while processing. Always rearm the queue whenever the common routine is called. Rework queue element processing, namely to eliminate hba_index vs host_index. Only one index is necessary. The queue entry can be marked invalid and the host_index updated immediately after cqe processing. After rework, xx_release routines are now DB write functions. Renamed the routines as such. Replaced the 3 individual loops that walk a cq with a call to the common routine. Redefined lpfc_sli4_sp_handle_mcqe() to commong handler definition with queue reference. Add increment for mbox completion to handler. - Added a new module/sysfs attribute: lpfc_cq_max_proc_limit To allow dynamic changing of the CQ max_proc_limit value being used. Although this leaves an EQ as an immediate interrupt, that interrupt will only occur if a CQ bound to it is in an armed state and has cqe's to process. By staying in the cq processing routine longer, high loads will avoid generating more interrupts as they will only rearm as the processing thread exits. The immediately interrupt is also beneficial to idle or lower-processing CQ's as they get serviced immediately without being penalized by sharing an EQ with a more loaded CQ. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <jsmart2021@gmail.com> Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2019-01-29 03:14:33 +08:00
"HST-IDX[%04d], PRT-IDX[%04d], NTFI[%03d]",
qp->queue_id, qp->entry_count,
qp->entry_size, qp->host_index,
scsi: lpfc: Rework EQ/CQ processing to address interrupt coalescing When driving high iop counts, auto_imax coalescing kicks in and drives the performance to extremely small iops levels. There are two issues: 1) auto_imax is enabled by default. The auto algorithm, when iops gets high, divides the iops by the hdwq count and uses that value to calculate EQ_Delay. The EQ_Delay is set uniformly on all EQs whether they have load or not. The EQ_delay is only manipulated every 5s (a long time). Thus there were large 5s swings of no interrupt delay followed by large/maximum delay, before repeating. 2) When processing a CQ, the driver got mixed up on the rate of when to ring the doorbell to keep the chip appraised of the eqe or cqe consumption as well as how how long to sit in the thread and process queue entries. Currently, the driver capped its work at 64 entries (very small) and exited/rearmed the CQ. Thus, on heavy loads, additional overheads were taken to exit and re-enter the interrupt handler. Worse, if in the large/maximum coalescing windows,k it could be a while before getting back to servicing. The issues are corrected by the following: - A change in defaults. Auto_imax is turned OFF and fcp_imax is set to 0. Thus all interrupts are immediate. - Cleanup of field names and their meanings. Existing names were non-intuitive or used for duplicate things. - Added max_proc_limit field, to control the length of time the handlers would service completions. - Reworked EQ handling: Added common routine that walks eq, applying notify interval and max processing limits. Use queue_claimed to claim ownership of the queue while processing. Always rearm the queue whenever the common routine is called. Rework queue element processing, namely to eliminate hba_index vs host_index. Only one index is necessary. The queue entry can be marked invalid and the host_index updated immediately after eqe processing. After rework, xx_release routines are now DB write functions. Renamed the routines as such. Moved lpfc_sli4_eq_flush(), which does similar action, to same area. Replaced the 2 individual loops that walk an eq with a call to the common routine. Slightly revised lpfc_sli4_hba_handle_eqe() calling syntax. Added per-cpu counters to detect interrupt rates and scale interrupt coalescing values. - Reworked CQ handling: Added common routine that walks cq, applying notify interval and max processing limits. Use queue_claimed to claim ownership of the queue while processing. Always rearm the queue whenever the common routine is called. Rework queue element processing, namely to eliminate hba_index vs host_index. Only one index is necessary. The queue entry can be marked invalid and the host_index updated immediately after cqe processing. After rework, xx_release routines are now DB write functions. Renamed the routines as such. Replaced the 3 individual loops that walk a cq with a call to the common routine. Redefined lpfc_sli4_sp_handle_mcqe() to commong handler definition with queue reference. Add increment for mbox completion to handler. - Added a new module/sysfs attribute: lpfc_cq_max_proc_limit To allow dynamic changing of the CQ max_proc_limit value being used. Although this leaves an EQ as an immediate interrupt, that interrupt will only occur if a CQ bound to it is in an armed state and has cqe's to process. By staying in the cq processing routine longer, high loads will avoid generating more interrupts as they will only rearm as the processing thread exits. The immediately interrupt is also beneficial to idle or lower-processing CQ's as they get serviced immediately without being penalized by sharing an EQ with a more loaded CQ. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <jsmart2021@gmail.com> Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2019-01-29 03:14:33 +08:00
qp->hba_index, qp->notify_interval);
len += snprintf(pbuffer + len,
LPFC_QUE_INFO_GET_BUF_SIZE - len, "\n");
return len;
}
static int
lpfc_idiag_wqs_for_cq(struct lpfc_hba *phba, char *wqtype, char *pbuffer,
int *len, int max_cnt, int cq_id)
{
struct lpfc_queue *qp;
int qidx;
2019-01-29 03:14:21 +08:00
for (qidx = 0; qidx < phba->cfg_hdw_queue; qidx++) {
qp = phba->sli4_hba.hdwq[qidx].fcp_wq;
if (qp->assoc_qid != cq_id)
continue;
*len = __lpfc_idiag_print_wq(qp, wqtype, pbuffer, *len);
if (*len >= max_cnt)
return 1;
}
2019-01-29 03:14:21 +08:00
if (phba->cfg_enable_fc4_type & LPFC_ENABLE_NVME) {
for (qidx = 0; qidx < phba->cfg_hdw_queue; qidx++) {
qp = phba->sli4_hba.hdwq[qidx].nvme_wq;
if (qp->assoc_qid != cq_id)
continue;
*len = __lpfc_idiag_print_wq(qp, wqtype, pbuffer, *len);
if (*len >= max_cnt)
return 1;
}
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-13 05:52:30 +08:00
}
return 0;
}
static int
__lpfc_idiag_print_cq(struct lpfc_queue *qp, char *cqtype,
char *pbuffer, int len)
{
if (!qp)
return len;
len += snprintf(pbuffer + len, LPFC_QUE_INFO_GET_BUF_SIZE - len,
"\t%s CQ info: ", cqtype);
len += snprintf(pbuffer + len, LPFC_QUE_INFO_GET_BUF_SIZE - len,
"AssocEQID[%02d]: CQ STAT[max:x%x relw:x%x "
"xabt:x%x wq:x%llx]\n",
qp->assoc_qid, qp->q_cnt_1, qp->q_cnt_2,
qp->q_cnt_3, (unsigned long long)qp->q_cnt_4);
len += snprintf(pbuffer + len, LPFC_QUE_INFO_GET_BUF_SIZE - len,
"\tCQID[%02d], QE-CNT[%04d], QE-SZ[%04d], "
scsi: lpfc: Rework EQ/CQ processing to address interrupt coalescing When driving high iop counts, auto_imax coalescing kicks in and drives the performance to extremely small iops levels. There are two issues: 1) auto_imax is enabled by default. The auto algorithm, when iops gets high, divides the iops by the hdwq count and uses that value to calculate EQ_Delay. The EQ_Delay is set uniformly on all EQs whether they have load or not. The EQ_delay is only manipulated every 5s (a long time). Thus there were large 5s swings of no interrupt delay followed by large/maximum delay, before repeating. 2) When processing a CQ, the driver got mixed up on the rate of when to ring the doorbell to keep the chip appraised of the eqe or cqe consumption as well as how how long to sit in the thread and process queue entries. Currently, the driver capped its work at 64 entries (very small) and exited/rearmed the CQ. Thus, on heavy loads, additional overheads were taken to exit and re-enter the interrupt handler. Worse, if in the large/maximum coalescing windows,k it could be a while before getting back to servicing. The issues are corrected by the following: - A change in defaults. Auto_imax is turned OFF and fcp_imax is set to 0. Thus all interrupts are immediate. - Cleanup of field names and their meanings. Existing names were non-intuitive or used for duplicate things. - Added max_proc_limit field, to control the length of time the handlers would service completions. - Reworked EQ handling: Added common routine that walks eq, applying notify interval and max processing limits. Use queue_claimed to claim ownership of the queue while processing. Always rearm the queue whenever the common routine is called. Rework queue element processing, namely to eliminate hba_index vs host_index. Only one index is necessary. The queue entry can be marked invalid and the host_index updated immediately after eqe processing. After rework, xx_release routines are now DB write functions. Renamed the routines as such. Moved lpfc_sli4_eq_flush(), which does similar action, to same area. Replaced the 2 individual loops that walk an eq with a call to the common routine. Slightly revised lpfc_sli4_hba_handle_eqe() calling syntax. Added per-cpu counters to detect interrupt rates and scale interrupt coalescing values. - Reworked CQ handling: Added common routine that walks cq, applying notify interval and max processing limits. Use queue_claimed to claim ownership of the queue while processing. Always rearm the queue whenever the common routine is called. Rework queue element processing, namely to eliminate hba_index vs host_index. Only one index is necessary. The queue entry can be marked invalid and the host_index updated immediately after cqe processing. After rework, xx_release routines are now DB write functions. Renamed the routines as such. Replaced the 3 individual loops that walk a cq with a call to the common routine. Redefined lpfc_sli4_sp_handle_mcqe() to commong handler definition with queue reference. Add increment for mbox completion to handler. - Added a new module/sysfs attribute: lpfc_cq_max_proc_limit To allow dynamic changing of the CQ max_proc_limit value being used. Although this leaves an EQ as an immediate interrupt, that interrupt will only occur if a CQ bound to it is in an armed state and has cqe's to process. By staying in the cq processing routine longer, high loads will avoid generating more interrupts as they will only rearm as the processing thread exits. The immediately interrupt is also beneficial to idle or lower-processing CQ's as they get serviced immediately without being penalized by sharing an EQ with a more loaded CQ. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <jsmart2021@gmail.com> Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2019-01-29 03:14:33 +08:00
"HST-IDX[%04d], NTFI[%03d], PLMT[%03d]",
qp->queue_id, qp->entry_count,
qp->entry_size, qp->host_index,
scsi: lpfc: Rework EQ/CQ processing to address interrupt coalescing When driving high iop counts, auto_imax coalescing kicks in and drives the performance to extremely small iops levels. There are two issues: 1) auto_imax is enabled by default. The auto algorithm, when iops gets high, divides the iops by the hdwq count and uses that value to calculate EQ_Delay. The EQ_Delay is set uniformly on all EQs whether they have load or not. The EQ_delay is only manipulated every 5s (a long time). Thus there were large 5s swings of no interrupt delay followed by large/maximum delay, before repeating. 2) When processing a CQ, the driver got mixed up on the rate of when to ring the doorbell to keep the chip appraised of the eqe or cqe consumption as well as how how long to sit in the thread and process queue entries. Currently, the driver capped its work at 64 entries (very small) and exited/rearmed the CQ. Thus, on heavy loads, additional overheads were taken to exit and re-enter the interrupt handler. Worse, if in the large/maximum coalescing windows,k it could be a while before getting back to servicing. The issues are corrected by the following: - A change in defaults. Auto_imax is turned OFF and fcp_imax is set to 0. Thus all interrupts are immediate. - Cleanup of field names and their meanings. Existing names were non-intuitive or used for duplicate things. - Added max_proc_limit field, to control the length of time the handlers would service completions. - Reworked EQ handling: Added common routine that walks eq, applying notify interval and max processing limits. Use queue_claimed to claim ownership of the queue while processing. Always rearm the queue whenever the common routine is called. Rework queue element processing, namely to eliminate hba_index vs host_index. Only one index is necessary. The queue entry can be marked invalid and the host_index updated immediately after eqe processing. After rework, xx_release routines are now DB write functions. Renamed the routines as such. Moved lpfc_sli4_eq_flush(), which does similar action, to same area. Replaced the 2 individual loops that walk an eq with a call to the common routine. Slightly revised lpfc_sli4_hba_handle_eqe() calling syntax. Added per-cpu counters to detect interrupt rates and scale interrupt coalescing values. - Reworked CQ handling: Added common routine that walks cq, applying notify interval and max processing limits. Use queue_claimed to claim ownership of the queue while processing. Always rearm the queue whenever the common routine is called. Rework queue element processing, namely to eliminate hba_index vs host_index. Only one index is necessary. The queue entry can be marked invalid and the host_index updated immediately after cqe processing. After rework, xx_release routines are now DB write functions. Renamed the routines as such. Replaced the 3 individual loops that walk a cq with a call to the common routine. Redefined lpfc_sli4_sp_handle_mcqe() to commong handler definition with queue reference. Add increment for mbox completion to handler. - Added a new module/sysfs attribute: lpfc_cq_max_proc_limit To allow dynamic changing of the CQ max_proc_limit value being used. Although this leaves an EQ as an immediate interrupt, that interrupt will only occur if a CQ bound to it is in an armed state and has cqe's to process. By staying in the cq processing routine longer, high loads will avoid generating more interrupts as they will only rearm as the processing thread exits. The immediately interrupt is also beneficial to idle or lower-processing CQ's as they get serviced immediately without being penalized by sharing an EQ with a more loaded CQ. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <jsmart2021@gmail.com> Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2019-01-29 03:14:33 +08:00
qp->notify_interval, qp->max_proc_limit);
len += snprintf(pbuffer + len, LPFC_QUE_INFO_GET_BUF_SIZE - len, "\n");
return len;
}
static int
__lpfc_idiag_print_rqpair(struct lpfc_queue *qp, struct lpfc_queue *datqp,
char *rqtype, char *pbuffer, int len)
{
if (!qp || !datqp)
return len;
len += snprintf(pbuffer + len, LPFC_QUE_INFO_GET_BUF_SIZE - len,
"\t\t%s RQ info: ", rqtype);
len += snprintf(pbuffer + len, LPFC_QUE_INFO_GET_BUF_SIZE - len,
"AssocCQID[%02d]: RQ-STAT[nopost:x%x nobuf:x%x "
"posted:x%x rcv:x%llx]\n",
qp->assoc_qid, qp->q_cnt_1, qp->q_cnt_2,
qp->q_cnt_3, (unsigned long long)qp->q_cnt_4);
len += snprintf(pbuffer + len, LPFC_QUE_INFO_GET_BUF_SIZE - len,
"\t\tHQID[%02d], QE-CNT[%04d], QE-SZ[%04d], "
scsi: lpfc: Rework EQ/CQ processing to address interrupt coalescing When driving high iop counts, auto_imax coalescing kicks in and drives the performance to extremely small iops levels. There are two issues: 1) auto_imax is enabled by default. The auto algorithm, when iops gets high, divides the iops by the hdwq count and uses that value to calculate EQ_Delay. The EQ_Delay is set uniformly on all EQs whether they have load or not. The EQ_delay is only manipulated every 5s (a long time). Thus there were large 5s swings of no interrupt delay followed by large/maximum delay, before repeating. 2) When processing a CQ, the driver got mixed up on the rate of when to ring the doorbell to keep the chip appraised of the eqe or cqe consumption as well as how how long to sit in the thread and process queue entries. Currently, the driver capped its work at 64 entries (very small) and exited/rearmed the CQ. Thus, on heavy loads, additional overheads were taken to exit and re-enter the interrupt handler. Worse, if in the large/maximum coalescing windows,k it could be a while before getting back to servicing. The issues are corrected by the following: - A change in defaults. Auto_imax is turned OFF and fcp_imax is set to 0. Thus all interrupts are immediate. - Cleanup of field names and their meanings. Existing names were non-intuitive or used for duplicate things. - Added max_proc_limit field, to control the length of time the handlers would service completions. - Reworked EQ handling: Added common routine that walks eq, applying notify interval and max processing limits. Use queue_claimed to claim ownership of the queue while processing. Always rearm the queue whenever the common routine is called. Rework queue element processing, namely to eliminate hba_index vs host_index. Only one index is necessary. The queue entry can be marked invalid and the host_index updated immediately after eqe processing. After rework, xx_release routines are now DB write functions. Renamed the routines as such. Moved lpfc_sli4_eq_flush(), which does similar action, to same area. Replaced the 2 individual loops that walk an eq with a call to the common routine. Slightly revised lpfc_sli4_hba_handle_eqe() calling syntax. Added per-cpu counters to detect interrupt rates and scale interrupt coalescing values. - Reworked CQ handling: Added common routine that walks cq, applying notify interval and max processing limits. Use queue_claimed to claim ownership of the queue while processing. Always rearm the queue whenever the common routine is called. Rework queue element processing, namely to eliminate hba_index vs host_index. Only one index is necessary. The queue entry can be marked invalid and the host_index updated immediately after cqe processing. After rework, xx_release routines are now DB write functions. Renamed the routines as such. Replaced the 3 individual loops that walk a cq with a call to the common routine. Redefined lpfc_sli4_sp_handle_mcqe() to commong handler definition with queue reference. Add increment for mbox completion to handler. - Added a new module/sysfs attribute: lpfc_cq_max_proc_limit To allow dynamic changing of the CQ max_proc_limit value being used. Although this leaves an EQ as an immediate interrupt, that interrupt will only occur if a CQ bound to it is in an armed state and has cqe's to process. By staying in the cq processing routine longer, high loads will avoid generating more interrupts as they will only rearm as the processing thread exits. The immediately interrupt is also beneficial to idle or lower-processing CQ's as they get serviced immediately without being penalized by sharing an EQ with a more loaded CQ. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <jsmart2021@gmail.com> Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2019-01-29 03:14:33 +08:00
"HST-IDX[%04d], PRT-IDX[%04d], NTFI[%03d]\n",
qp->queue_id, qp->entry_count, qp->entry_size,
scsi: lpfc: Rework EQ/CQ processing to address interrupt coalescing When driving high iop counts, auto_imax coalescing kicks in and drives the performance to extremely small iops levels. There are two issues: 1) auto_imax is enabled by default. The auto algorithm, when iops gets high, divides the iops by the hdwq count and uses that value to calculate EQ_Delay. The EQ_Delay is set uniformly on all EQs whether they have load or not. The EQ_delay is only manipulated every 5s (a long time). Thus there were large 5s swings of no interrupt delay followed by large/maximum delay, before repeating. 2) When processing a CQ, the driver got mixed up on the rate of when to ring the doorbell to keep the chip appraised of the eqe or cqe consumption as well as how how long to sit in the thread and process queue entries. Currently, the driver capped its work at 64 entries (very small) and exited/rearmed the CQ. Thus, on heavy loads, additional overheads were taken to exit and re-enter the interrupt handler. Worse, if in the large/maximum coalescing windows,k it could be a while before getting back to servicing. The issues are corrected by the following: - A change in defaults. Auto_imax is turned OFF and fcp_imax is set to 0. Thus all interrupts are immediate. - Cleanup of field names and their meanings. Existing names were non-intuitive or used for duplicate things. - Added max_proc_limit field, to control the length of time the handlers would service completions. - Reworked EQ handling: Added common routine that walks eq, applying notify interval and max processing limits. Use queue_claimed to claim ownership of the queue while processing. Always rearm the queue whenever the common routine is called. Rework queue element processing, namely to eliminate hba_index vs host_index. Only one index is necessary. The queue entry can be marked invalid and the host_index updated immediately after eqe processing. After rework, xx_release routines are now DB write functions. Renamed the routines as such. Moved lpfc_sli4_eq_flush(), which does similar action, to same area. Replaced the 2 individual loops that walk an eq with a call to the common routine. Slightly revised lpfc_sli4_hba_handle_eqe() calling syntax. Added per-cpu counters to detect interrupt rates and scale interrupt coalescing values. - Reworked CQ handling: Added common routine that walks cq, applying notify interval and max processing limits. Use queue_claimed to claim ownership of the queue while processing. Always rearm the queue whenever the common routine is called. Rework queue element processing, namely to eliminate hba_index vs host_index. Only one index is necessary. The queue entry can be marked invalid and the host_index updated immediately after cqe processing. After rework, xx_release routines are now DB write functions. Renamed the routines as such. Replaced the 3 individual loops that walk a cq with a call to the common routine. Redefined lpfc_sli4_sp_handle_mcqe() to commong handler definition with queue reference. Add increment for mbox completion to handler. - Added a new module/sysfs attribute: lpfc_cq_max_proc_limit To allow dynamic changing of the CQ max_proc_limit value being used. Although this leaves an EQ as an immediate interrupt, that interrupt will only occur if a CQ bound to it is in an armed state and has cqe's to process. By staying in the cq processing routine longer, high loads will avoid generating more interrupts as they will only rearm as the processing thread exits. The immediately interrupt is also beneficial to idle or lower-processing CQ's as they get serviced immediately without being penalized by sharing an EQ with a more loaded CQ. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <jsmart2021@gmail.com> Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2019-01-29 03:14:33 +08:00
qp->host_index, qp->hba_index, qp->notify_interval);
len += snprintf(pbuffer + len, LPFC_QUE_INFO_GET_BUF_SIZE - len,
"\t\tDQID[%02d], QE-CNT[%04d], QE-SZ[%04d], "
scsi: lpfc: Rework EQ/CQ processing to address interrupt coalescing When driving high iop counts, auto_imax coalescing kicks in and drives the performance to extremely small iops levels. There are two issues: 1) auto_imax is enabled by default. The auto algorithm, when iops gets high, divides the iops by the hdwq count and uses that value to calculate EQ_Delay. The EQ_Delay is set uniformly on all EQs whether they have load or not. The EQ_delay is only manipulated every 5s (a long time). Thus there were large 5s swings of no interrupt delay followed by large/maximum delay, before repeating. 2) When processing a CQ, the driver got mixed up on the rate of when to ring the doorbell to keep the chip appraised of the eqe or cqe consumption as well as how how long to sit in the thread and process queue entries. Currently, the driver capped its work at 64 entries (very small) and exited/rearmed the CQ. Thus, on heavy loads, additional overheads were taken to exit and re-enter the interrupt handler. Worse, if in the large/maximum coalescing windows,k it could be a while before getting back to servicing. The issues are corrected by the following: - A change in defaults. Auto_imax is turned OFF and fcp_imax is set to 0. Thus all interrupts are immediate. - Cleanup of field names and their meanings. Existing names were non-intuitive or used for duplicate things. - Added max_proc_limit field, to control the length of time the handlers would service completions. - Reworked EQ handling: Added common routine that walks eq, applying notify interval and max processing limits. Use queue_claimed to claim ownership of the queue while processing. Always rearm the queue whenever the common routine is called. Rework queue element processing, namely to eliminate hba_index vs host_index. Only one index is necessary. The queue entry can be marked invalid and the host_index updated immediately after eqe processing. After rework, xx_release routines are now DB write functions. Renamed the routines as such. Moved lpfc_sli4_eq_flush(), which does similar action, to same area. Replaced the 2 individual loops that walk an eq with a call to the common routine. Slightly revised lpfc_sli4_hba_handle_eqe() calling syntax. Added per-cpu counters to detect interrupt rates and scale interrupt coalescing values. - Reworked CQ handling: Added common routine that walks cq, applying notify interval and max processing limits. Use queue_claimed to claim ownership of the queue while processing. Always rearm the queue whenever the common routine is called. Rework queue element processing, namely to eliminate hba_index vs host_index. Only one index is necessary. The queue entry can be marked invalid and the host_index updated immediately after cqe processing. After rework, xx_release routines are now DB write functions. Renamed the routines as such. Replaced the 3 individual loops that walk a cq with a call to the common routine. Redefined lpfc_sli4_sp_handle_mcqe() to commong handler definition with queue reference. Add increment for mbox completion to handler. - Added a new module/sysfs attribute: lpfc_cq_max_proc_limit To allow dynamic changing of the CQ max_proc_limit value being used. Although this leaves an EQ as an immediate interrupt, that interrupt will only occur if a CQ bound to it is in an armed state and has cqe's to process. By staying in the cq processing routine longer, high loads will avoid generating more interrupts as they will only rearm as the processing thread exits. The immediately interrupt is also beneficial to idle or lower-processing CQ's as they get serviced immediately without being penalized by sharing an EQ with a more loaded CQ. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <jsmart2021@gmail.com> Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2019-01-29 03:14:33 +08:00
"HST-IDX[%04d], PRT-IDX[%04d], NTFI[%03d]\n",
datqp->queue_id, datqp->entry_count,
datqp->entry_size, datqp->host_index,
scsi: lpfc: Rework EQ/CQ processing to address interrupt coalescing When driving high iop counts, auto_imax coalescing kicks in and drives the performance to extremely small iops levels. There are two issues: 1) auto_imax is enabled by default. The auto algorithm, when iops gets high, divides the iops by the hdwq count and uses that value to calculate EQ_Delay. The EQ_Delay is set uniformly on all EQs whether they have load or not. The EQ_delay is only manipulated every 5s (a long time). Thus there were large 5s swings of no interrupt delay followed by large/maximum delay, before repeating. 2) When processing a CQ, the driver got mixed up on the rate of when to ring the doorbell to keep the chip appraised of the eqe or cqe consumption as well as how how long to sit in the thread and process queue entries. Currently, the driver capped its work at 64 entries (very small) and exited/rearmed the CQ. Thus, on heavy loads, additional overheads were taken to exit and re-enter the interrupt handler. Worse, if in the large/maximum coalescing windows,k it could be a while before getting back to servicing. The issues are corrected by the following: - A change in defaults. Auto_imax is turned OFF and fcp_imax is set to 0. Thus all interrupts are immediate. - Cleanup of field names and their meanings. Existing names were non-intuitive or used for duplicate things. - Added max_proc_limit field, to control the length of time the handlers would service completions. - Reworked EQ handling: Added common routine that walks eq, applying notify interval and max processing limits. Use queue_claimed to claim ownership of the queue while processing. Always rearm the queue whenever the common routine is called. Rework queue element processing, namely to eliminate hba_index vs host_index. Only one index is necessary. The queue entry can be marked invalid and the host_index updated immediately after eqe processing. After rework, xx_release routines are now DB write functions. Renamed the routines as such. Moved lpfc_sli4_eq_flush(), which does similar action, to same area. Replaced the 2 individual loops that walk an eq with a call to the common routine. Slightly revised lpfc_sli4_hba_handle_eqe() calling syntax. Added per-cpu counters to detect interrupt rates and scale interrupt coalescing values. - Reworked CQ handling: Added common routine that walks cq, applying notify interval and max processing limits. Use queue_claimed to claim ownership of the queue while processing. Always rearm the queue whenever the common routine is called. Rework queue element processing, namely to eliminate hba_index vs host_index. Only one index is necessary. The queue entry can be marked invalid and the host_index updated immediately after cqe processing. After rework, xx_release routines are now DB write functions. Renamed the routines as such. Replaced the 3 individual loops that walk a cq with a call to the common routine. Redefined lpfc_sli4_sp_handle_mcqe() to commong handler definition with queue reference. Add increment for mbox completion to handler. - Added a new module/sysfs attribute: lpfc_cq_max_proc_limit To allow dynamic changing of the CQ max_proc_limit value being used. Although this leaves an EQ as an immediate interrupt, that interrupt will only occur if a CQ bound to it is in an armed state and has cqe's to process. By staying in the cq processing routine longer, high loads will avoid generating more interrupts as they will only rearm as the processing thread exits. The immediately interrupt is also beneficial to idle or lower-processing CQ's as they get serviced immediately without being penalized by sharing an EQ with a more loaded CQ. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <jsmart2021@gmail.com> Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2019-01-29 03:14:33 +08:00
datqp->hba_index, datqp->notify_interval);
return len;
}
static int
lpfc_idiag_cqs_for_eq(struct lpfc_hba *phba, char *pbuffer,
int *len, int max_cnt, int eqidx, int eq_id)
{
struct lpfc_queue *qp;
int rc;
qp = phba->sli4_hba.hdwq[eqidx].fcp_cq;
*len = __lpfc_idiag_print_cq(qp, "FCP", pbuffer, *len);
/* Reset max counter */
qp->CQ_max_cqe = 0;
if (*len >= max_cnt)
return 1;
rc = lpfc_idiag_wqs_for_cq(phba, "FCP", pbuffer, len,
max_cnt, qp->queue_id);
if (rc)
return 1;
2019-01-29 03:14:21 +08:00
if (phba->cfg_enable_fc4_type & LPFC_ENABLE_NVME) {
qp = phba->sli4_hba.hdwq[eqidx].nvme_cq;
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-13 05:52:30 +08:00
*len = __lpfc_idiag_print_cq(qp, "NVME", pbuffer, *len);
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-13 05:52:30 +08:00
/* Reset max counter */
qp->CQ_max_cqe = 0;
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-13 05:52:30 +08:00
if (*len >= max_cnt)
return 1;
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-13 05:52:30 +08:00
rc = lpfc_idiag_wqs_for_cq(phba, "NVME", pbuffer, len,
max_cnt, qp->queue_id);
if (rc)
return 1;
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-13 05:52:30 +08:00
}
if ((eqidx < phba->cfg_nvmet_mrq) && phba->nvmet_support) {
/* NVMET CQset */
qp = phba->sli4_hba.nvmet_cqset[eqidx];
*len = __lpfc_idiag_print_cq(qp, "NVMET CQset", pbuffer, *len);
/* Reset max counter */
qp->CQ_max_cqe = 0;
if (*len >= max_cnt)
return 1;
/* RQ header */
qp = phba->sli4_hba.nvmet_mrq_hdr[eqidx];
*len = __lpfc_idiag_print_rqpair(qp,
phba->sli4_hba.nvmet_mrq_data[eqidx],
"NVMET MRQ", pbuffer, *len);
if (*len >= max_cnt)
return 1;
}
return 0;
}
static int
__lpfc_idiag_print_eq(struct lpfc_queue *qp, char *eqtype,
char *pbuffer, int len)
{
if (!qp)
return len;
len += snprintf(pbuffer + len, LPFC_QUE_INFO_GET_BUF_SIZE - len,
"\n%s EQ info: EQ-STAT[max:x%x noE:x%x "
"cqe_proc:x%x eqe_proc:x%llx eqd %d]\n",
eqtype, qp->q_cnt_1, qp->q_cnt_2, qp->q_cnt_3,
(unsigned long long)qp->q_cnt_4, qp->q_mode);
len += snprintf(pbuffer + len, LPFC_QUE_INFO_GET_BUF_SIZE - len,
"EQID[%02d], QE-CNT[%04d], QE-SZ[%04d], "
scsi: lpfc: Rework EQ/CQ processing to address interrupt coalescing When driving high iop counts, auto_imax coalescing kicks in and drives the performance to extremely small iops levels. There are two issues: 1) auto_imax is enabled by default. The auto algorithm, when iops gets high, divides the iops by the hdwq count and uses that value to calculate EQ_Delay. The EQ_Delay is set uniformly on all EQs whether they have load or not. The EQ_delay is only manipulated every 5s (a long time). Thus there were large 5s swings of no interrupt delay followed by large/maximum delay, before repeating. 2) When processing a CQ, the driver got mixed up on the rate of when to ring the doorbell to keep the chip appraised of the eqe or cqe consumption as well as how how long to sit in the thread and process queue entries. Currently, the driver capped its work at 64 entries (very small) and exited/rearmed the CQ. Thus, on heavy loads, additional overheads were taken to exit and re-enter the interrupt handler. Worse, if in the large/maximum coalescing windows,k it could be a while before getting back to servicing. The issues are corrected by the following: - A change in defaults. Auto_imax is turned OFF and fcp_imax is set to 0. Thus all interrupts are immediate. - Cleanup of field names and their meanings. Existing names were non-intuitive or used for duplicate things. - Added max_proc_limit field, to control the length of time the handlers would service completions. - Reworked EQ handling: Added common routine that walks eq, applying notify interval and max processing limits. Use queue_claimed to claim ownership of the queue while processing. Always rearm the queue whenever the common routine is called. Rework queue element processing, namely to eliminate hba_index vs host_index. Only one index is necessary. The queue entry can be marked invalid and the host_index updated immediately after eqe processing. After rework, xx_release routines are now DB write functions. Renamed the routines as such. Moved lpfc_sli4_eq_flush(), which does similar action, to same area. Replaced the 2 individual loops that walk an eq with a call to the common routine. Slightly revised lpfc_sli4_hba_handle_eqe() calling syntax. Added per-cpu counters to detect interrupt rates and scale interrupt coalescing values. - Reworked CQ handling: Added common routine that walks cq, applying notify interval and max processing limits. Use queue_claimed to claim ownership of the queue while processing. Always rearm the queue whenever the common routine is called. Rework queue element processing, namely to eliminate hba_index vs host_index. Only one index is necessary. The queue entry can be marked invalid and the host_index updated immediately after cqe processing. After rework, xx_release routines are now DB write functions. Renamed the routines as such. Replaced the 3 individual loops that walk a cq with a call to the common routine. Redefined lpfc_sli4_sp_handle_mcqe() to commong handler definition with queue reference. Add increment for mbox completion to handler. - Added a new module/sysfs attribute: lpfc_cq_max_proc_limit To allow dynamic changing of the CQ max_proc_limit value being used. Although this leaves an EQ as an immediate interrupt, that interrupt will only occur if a CQ bound to it is in an armed state and has cqe's to process. By staying in the cq processing routine longer, high loads will avoid generating more interrupts as they will only rearm as the processing thread exits. The immediately interrupt is also beneficial to idle or lower-processing CQ's as they get serviced immediately without being penalized by sharing an EQ with a more loaded CQ. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <jsmart2021@gmail.com> Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2019-01-29 03:14:33 +08:00
"HST-IDX[%04d], NTFI[%03d], PLMT[%03d], AFFIN[%03d]",
qp->queue_id, qp->entry_count, qp->entry_size,
scsi: lpfc: Rework EQ/CQ processing to address interrupt coalescing When driving high iop counts, auto_imax coalescing kicks in and drives the performance to extremely small iops levels. There are two issues: 1) auto_imax is enabled by default. The auto algorithm, when iops gets high, divides the iops by the hdwq count and uses that value to calculate EQ_Delay. The EQ_Delay is set uniformly on all EQs whether they have load or not. The EQ_delay is only manipulated every 5s (a long time). Thus there were large 5s swings of no interrupt delay followed by large/maximum delay, before repeating. 2) When processing a CQ, the driver got mixed up on the rate of when to ring the doorbell to keep the chip appraised of the eqe or cqe consumption as well as how how long to sit in the thread and process queue entries. Currently, the driver capped its work at 64 entries (very small) and exited/rearmed the CQ. Thus, on heavy loads, additional overheads were taken to exit and re-enter the interrupt handler. Worse, if in the large/maximum coalescing windows,k it could be a while before getting back to servicing. The issues are corrected by the following: - A change in defaults. Auto_imax is turned OFF and fcp_imax is set to 0. Thus all interrupts are immediate. - Cleanup of field names and their meanings. Existing names were non-intuitive or used for duplicate things. - Added max_proc_limit field, to control the length of time the handlers would service completions. - Reworked EQ handling: Added common routine that walks eq, applying notify interval and max processing limits. Use queue_claimed to claim ownership of the queue while processing. Always rearm the queue whenever the common routine is called. Rework queue element processing, namely to eliminate hba_index vs host_index. Only one index is necessary. The queue entry can be marked invalid and the host_index updated immediately after eqe processing. After rework, xx_release routines are now DB write functions. Renamed the routines as such. Moved lpfc_sli4_eq_flush(), which does similar action, to same area. Replaced the 2 individual loops that walk an eq with a call to the common routine. Slightly revised lpfc_sli4_hba_handle_eqe() calling syntax. Added per-cpu counters to detect interrupt rates and scale interrupt coalescing values. - Reworked CQ handling: Added common routine that walks cq, applying notify interval and max processing limits. Use queue_claimed to claim ownership of the queue while processing. Always rearm the queue whenever the common routine is called. Rework queue element processing, namely to eliminate hba_index vs host_index. Only one index is necessary. The queue entry can be marked invalid and the host_index updated immediately after cqe processing. After rework, xx_release routines are now DB write functions. Renamed the routines as such. Replaced the 3 individual loops that walk a cq with a call to the common routine. Redefined lpfc_sli4_sp_handle_mcqe() to commong handler definition with queue reference. Add increment for mbox completion to handler. - Added a new module/sysfs attribute: lpfc_cq_max_proc_limit To allow dynamic changing of the CQ max_proc_limit value being used. Although this leaves an EQ as an immediate interrupt, that interrupt will only occur if a CQ bound to it is in an armed state and has cqe's to process. By staying in the cq processing routine longer, high loads will avoid generating more interrupts as they will only rearm as the processing thread exits. The immediately interrupt is also beneficial to idle or lower-processing CQ's as they get serviced immediately without being penalized by sharing an EQ with a more loaded CQ. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <jsmart2021@gmail.com> Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2019-01-29 03:14:33 +08:00
qp->host_index, qp->notify_interval,
qp->max_proc_limit, qp->chann);
len += snprintf(pbuffer + len, LPFC_QUE_INFO_GET_BUF_SIZE - len, "\n");
return len;
}
/**
* lpfc_idiag_queinfo_read - idiag debugfs read queue information
* @file: The file pointer to read from.
* @buf: The buffer to copy the data to.
* @nbytes: The number of bytes to read.
* @ppos: The position in the file to start reading from.
*
* Description:
* This routine reads data from the @phba SLI4 PCI function queue information,
* and copies to user @buf.
* This routine only returns 1 EQs worth of information. It remembers the last
* EQ read and jumps to the next EQ. Thus subsequent calls to queInfo will
* retrieve all EQs allocated for the phba.
*
* Returns:
* This function returns the amount of data that was read (this could be less
* than @nbytes if the end of the file was reached) or a negative error value.
**/
static ssize_t
lpfc_idiag_queinfo_read(struct file *file, char __user *buf, size_t nbytes,
loff_t *ppos)
{
struct lpfc_debug *debug = file->private_data;
struct lpfc_hba *phba = (struct lpfc_hba *)debug->i_private;
char *pbuffer;
int max_cnt, rc, x, len = 0;
struct lpfc_queue *qp = NULL;
if (!debug->buffer)
debug->buffer = kmalloc(LPFC_QUE_INFO_GET_BUF_SIZE, GFP_KERNEL);
if (!debug->buffer)
return 0;
pbuffer = debug->buffer;
max_cnt = LPFC_QUE_INFO_GET_BUF_SIZE - 256;
if (*ppos)
return 0;
spin_lock_irq(&phba->hbalock);
/* Fast-path event queue */
2019-01-29 03:14:21 +08:00
if (phba->sli4_hba.hdwq && phba->cfg_hdw_queue) {
x = phba->lpfc_idiag_last_eq;
phba->lpfc_idiag_last_eq++;
2019-01-29 03:14:21 +08:00
if (phba->lpfc_idiag_last_eq >= phba->cfg_hdw_queue)
phba->lpfc_idiag_last_eq = 0;
len += snprintf(pbuffer + len, LPFC_QUE_INFO_GET_BUF_SIZE - len,
"HDWQ %d out of %d HBA HDWQs\n",
2019-01-29 03:14:21 +08:00
x, phba->cfg_hdw_queue);
/* Fast-path EQ */
2019-01-29 03:14:21 +08:00
qp = phba->sli4_hba.hdwq[x].hba_eq;
if (!qp)
goto out;
len = __lpfc_idiag_print_eq(qp, "HBA", pbuffer, len);
/* Reset max counter */
qp->EQ_max_eqe = 0;
if (len >= max_cnt)
goto too_big;
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-13 05:52:30 +08:00
/* will dump both fcp and nvme cqs/wqs for the eq */
rc = lpfc_idiag_cqs_for_eq(phba, pbuffer, &len,
max_cnt, x, qp->queue_id);
if (rc)
goto too_big;
/* Only EQ 0 has slow path CQs configured */
if (x)
goto out;
/* Slow-path mailbox CQ */
qp = phba->sli4_hba.mbx_cq;
len = __lpfc_idiag_print_cq(qp, "MBX", pbuffer, len);
if (len >= max_cnt)
goto too_big;
/* Slow-path MBOX MQ */
qp = phba->sli4_hba.mbx_wq;
len = __lpfc_idiag_print_wq(qp, "MBX", pbuffer, len);
if (len >= max_cnt)
goto too_big;
/* Slow-path ELS response CQ */
qp = phba->sli4_hba.els_cq;
len = __lpfc_idiag_print_cq(qp, "ELS", pbuffer, len);
/* Reset max counter */
if (qp)
qp->CQ_max_cqe = 0;
if (len >= max_cnt)
goto too_big;
/* Slow-path ELS WQ */
qp = phba->sli4_hba.els_wq;
len = __lpfc_idiag_print_wq(qp, "ELS", pbuffer, len);
if (len >= max_cnt)
goto too_big;
qp = phba->sli4_hba.hdr_rq;
len = __lpfc_idiag_print_rqpair(qp, phba->sli4_hba.dat_rq,
"ELS RQpair", pbuffer, len);
if (len >= max_cnt)
goto too_big;
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-13 05:52:30 +08:00
/* Slow-path NVME LS response CQ */
qp = phba->sli4_hba.nvmels_cq;
len = __lpfc_idiag_print_cq(qp, "NVME LS",
pbuffer, len);
/* Reset max counter */
if (qp)
qp->CQ_max_cqe = 0;
if (len >= max_cnt)
goto too_big;
/* Slow-path NVME LS WQ */
qp = phba->sli4_hba.nvmels_wq;
len = __lpfc_idiag_print_wq(qp, "NVME LS",
pbuffer, len);
if (len >= max_cnt)
goto too_big;
goto out;
}
spin_unlock_irq(&phba->hbalock);
return simple_read_from_buffer(buf, nbytes, ppos, pbuffer, len);
too_big:
len += snprintf(pbuffer + len,
LPFC_QUE_INFO_GET_BUF_SIZE - len, "Truncated ...\n");
out:
spin_unlock_irq(&phba->hbalock);
return simple_read_from_buffer(buf, nbytes, ppos, pbuffer, len);
}
/**
* lpfc_idiag_que_param_check - queue access command parameter sanity check
* @q: The pointer to queue structure.
* @index: The index into a queue entry.
* @count: The number of queue entries to access.
*
* Description:
* The routine performs sanity check on device queue access method commands.
*
* Returns:
* This function returns -EINVAL when fails the sanity check, otherwise, it
* returns 0.
**/
static int
lpfc_idiag_que_param_check(struct lpfc_queue *q, int index, int count)
{
/* Only support single entry read or browsing */
if ((count != 1) && (count != LPFC_QUE_ACC_BROWSE))
return -EINVAL;
if (index > q->entry_count - 1)
return -EINVAL;
return 0;
}
/**
* lpfc_idiag_queacc_read_qe - read a single entry from the given queue index
* @pbuffer: The pointer to buffer to copy the read data into.
* @pque: The pointer to the queue to be read.
* @index: The index into the queue entry.
*
* Description:
* This routine reads out a single entry from the given queue's index location
* and copies it into the buffer provided.
*
* Returns:
* This function returns 0 when it fails, otherwise, it returns the length of
* the data read into the buffer provided.
**/
static int
lpfc_idiag_queacc_read_qe(char *pbuffer, int len, struct lpfc_queue *pque,
uint32_t index)
{
int offset, esize;
uint32_t *pentry;
if (!pbuffer || !pque)
return 0;
esize = pque->entry_size;
len += snprintf(pbuffer+len, LPFC_QUE_ACC_BUF_SIZE-len,
"QE-INDEX[%04d]:\n", index);
offset = 0;
pentry = lpfc_sli4_qe(pque, index);
while (esize > 0) {
len += snprintf(pbuffer+len, LPFC_QUE_ACC_BUF_SIZE-len,
"%08x ", *pentry);
pentry++;
offset += sizeof(uint32_t);
esize -= sizeof(uint32_t);
if (esize > 0 && !(offset % (4 * sizeof(uint32_t))))
len += snprintf(pbuffer+len,
LPFC_QUE_ACC_BUF_SIZE-len, "\n");
}
len += snprintf(pbuffer+len, LPFC_QUE_ACC_BUF_SIZE-len, "\n");
return len;
}
/**
* lpfc_idiag_queacc_read - idiag debugfs read port queue
* @file: The file pointer to read from.
* @buf: The buffer to copy the data to.
* @nbytes: The number of bytes to read.
* @ppos: The position in the file to start reading from.
*
* Description:
* This routine reads data from the @phba device queue memory according to the
* idiag command, and copies to user @buf. Depending on the queue dump read
* command setup, it does either a single queue entry read or browing through
* all entries of the queue.
*
* Returns:
* This function returns the amount of data that was read (this could be less
* than @nbytes if the end of the file was reached) or a negative error value.
**/
static ssize_t
lpfc_idiag_queacc_read(struct file *file, char __user *buf, size_t nbytes,
loff_t *ppos)
{
struct lpfc_debug *debug = file->private_data;
uint32_t last_index, index, count;
struct lpfc_queue *pque = NULL;
char *pbuffer;
int len = 0;
/* This is a user read operation */
debug->op = LPFC_IDIAG_OP_RD;
if (!debug->buffer)
debug->buffer = kmalloc(LPFC_QUE_ACC_BUF_SIZE, GFP_KERNEL);
if (!debug->buffer)
return 0;
pbuffer = debug->buffer;
if (*ppos)
return 0;
if (idiag.cmd.opcode == LPFC_IDIAG_CMD_QUEACC_RD) {
index = idiag.cmd.data[IDIAG_QUEACC_INDEX_INDX];
count = idiag.cmd.data[IDIAG_QUEACC_COUNT_INDX];
pque = (struct lpfc_queue *)idiag.ptr_private;
} else
return 0;
/* Browse the queue starting from index */
if (count == LPFC_QUE_ACC_BROWSE)
goto que_browse;
/* Read a single entry from the queue */
len = lpfc_idiag_queacc_read_qe(pbuffer, len, pque, index);
return simple_read_from_buffer(buf, nbytes, ppos, pbuffer, len);
que_browse:
/* Browse all entries from the queue */
last_index = idiag.offset.last_rd;
index = last_index;
while (len < LPFC_QUE_ACC_SIZE - pque->entry_size) {
len = lpfc_idiag_queacc_read_qe(pbuffer, len, pque, index);
index++;
if (index > pque->entry_count - 1)
break;
}
/* Set up the offset for next portion of pci cfg read */
if (index > pque->entry_count - 1)
index = 0;
idiag.offset.last_rd = index;
return simple_read_from_buffer(buf, nbytes, ppos, pbuffer, len);
}
/**
* lpfc_idiag_queacc_write - Syntax check and set up idiag queacc commands
* @file: The file pointer to read from.
* @buf: The buffer to copy the user data from.
* @nbytes: The number of bytes to get.
* @ppos: The position in the file to start reading from.
*
* This routine get the debugfs idiag command struct from user space and then
* perform the syntax check for port queue read (dump) or write (set) command
* accordingly. In the case of port queue read command, it sets up the command
* in the idiag command struct for the following debugfs read operation. In
* the case of port queue write operation, it executes the write operation
* into the port queue entry accordingly.
*
* It returns the @nbytges passing in from debugfs user space when successful.
* In case of error conditions, it returns proper error code back to the user
* space.
**/
static ssize_t
lpfc_idiag_queacc_write(struct file *file, const char __user *buf,
size_t nbytes, loff_t *ppos)
{
struct lpfc_debug *debug = file->private_data;
struct lpfc_hba *phba = (struct lpfc_hba *)debug->i_private;
uint32_t qidx, quetp, queid, index, count, offset, value;
uint32_t *pentry;
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-13 05:52:30 +08:00
struct lpfc_queue *pque, *qp;
int rc;
/* This is a user write operation */
debug->op = LPFC_IDIAG_OP_WR;
rc = lpfc_idiag_cmd_get(buf, nbytes, &idiag.cmd);
if (rc < 0)
return rc;
/* Get and sanity check on command feilds */
quetp = idiag.cmd.data[IDIAG_QUEACC_QUETP_INDX];
queid = idiag.cmd.data[IDIAG_QUEACC_QUEID_INDX];
index = idiag.cmd.data[IDIAG_QUEACC_INDEX_INDX];
count = idiag.cmd.data[IDIAG_QUEACC_COUNT_INDX];
offset = idiag.cmd.data[IDIAG_QUEACC_OFFST_INDX];
value = idiag.cmd.data[IDIAG_QUEACC_VALUE_INDX];
/* Sanity check on command line arguments */
if (idiag.cmd.opcode == LPFC_IDIAG_CMD_QUEACC_WR ||
idiag.cmd.opcode == LPFC_IDIAG_CMD_QUEACC_ST ||
idiag.cmd.opcode == LPFC_IDIAG_CMD_QUEACC_CL) {
if (rc != LPFC_QUE_ACC_WR_CMD_ARG)
goto error_out;
if (count != 1)
goto error_out;
} else if (idiag.cmd.opcode == LPFC_IDIAG_CMD_QUEACC_RD) {
if (rc != LPFC_QUE_ACC_RD_CMD_ARG)
goto error_out;
} else
goto error_out;
switch (quetp) {
case LPFC_IDIAG_EQ:
/* HBA event queue */
2019-01-29 03:14:21 +08:00
if (phba->sli4_hba.hdwq) {
for (qidx = 0; qidx < phba->cfg_hdw_queue; qidx++) {
qp = phba->sli4_hba.hdwq[qidx].hba_eq;
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-13 05:52:30 +08:00
if (qp && qp->queue_id == queid) {
/* Sanity check */
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-13 05:52:30 +08:00
rc = lpfc_idiag_que_param_check(qp,
index, count);
if (rc)
goto error_out;
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-13 05:52:30 +08:00
idiag.ptr_private = qp;
goto pass_check;
}
}
}
goto error_out;
break;
case LPFC_IDIAG_CQ:
/* MBX complete queue */
if (phba->sli4_hba.mbx_cq &&
phba->sli4_hba.mbx_cq->queue_id == queid) {
/* Sanity check */
rc = lpfc_idiag_que_param_check(
phba->sli4_hba.mbx_cq, index, count);
if (rc)
goto error_out;
idiag.ptr_private = phba->sli4_hba.mbx_cq;
goto pass_check;
}
/* ELS complete queue */
if (phba->sli4_hba.els_cq &&
phba->sli4_hba.els_cq->queue_id == queid) {
/* Sanity check */
rc = lpfc_idiag_que_param_check(
phba->sli4_hba.els_cq, index, count);
if (rc)
goto error_out;
idiag.ptr_private = phba->sli4_hba.els_cq;
goto pass_check;
}
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-13 05:52:30 +08:00
/* NVME LS complete queue */
if (phba->sli4_hba.nvmels_cq &&
phba->sli4_hba.nvmels_cq->queue_id == queid) {
/* Sanity check */
rc = lpfc_idiag_que_param_check(
phba->sli4_hba.nvmels_cq, index, count);
if (rc)
goto error_out;
idiag.ptr_private = phba->sli4_hba.nvmels_cq;
goto pass_check;
}
/* FCP complete queue */
2019-01-29 03:14:21 +08:00
if (phba->sli4_hba.hdwq) {
for (qidx = 0; qidx < phba->cfg_hdw_queue;
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-13 05:52:30 +08:00
qidx++) {
2019-01-29 03:14:21 +08:00
qp = phba->sli4_hba.hdwq[qidx].fcp_cq;
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-13 05:52:30 +08:00
if (qp && qp->queue_id == queid) {
/* Sanity check */
rc = lpfc_idiag_que_param_check(
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-13 05:52:30 +08:00
qp, index, count);
if (rc)
goto error_out;
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-13 05:52:30 +08:00
idiag.ptr_private = qp;
goto pass_check;
}
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-13 05:52:30 +08:00
}
}
/* NVME complete queue */
2019-01-29 03:14:21 +08:00
if (phba->sli4_hba.hdwq) {
qidx = 0;
do {
2019-01-29 03:14:21 +08:00
qp = phba->sli4_hba.hdwq[qidx].nvme_cq;
if (qp && qp->queue_id == queid) {
/* Sanity check */
rc = lpfc_idiag_que_param_check(
2019-01-29 03:14:21 +08:00
qp, index, count);
if (rc)
goto error_out;
2019-01-29 03:14:21 +08:00
idiag.ptr_private = qp;
goto pass_check;
}
2019-01-29 03:14:21 +08:00
} while (++qidx < phba->cfg_hdw_queue);
}
goto error_out;
break;
case LPFC_IDIAG_MQ:
/* MBX work queue */
if (phba->sli4_hba.mbx_wq &&
phba->sli4_hba.mbx_wq->queue_id == queid) {
/* Sanity check */
rc = lpfc_idiag_que_param_check(
phba->sli4_hba.mbx_wq, index, count);
if (rc)
goto error_out;
idiag.ptr_private = phba->sli4_hba.mbx_wq;
goto pass_check;
}
goto error_out;
break;
case LPFC_IDIAG_WQ:
/* ELS work queue */
if (phba->sli4_hba.els_wq &&
phba->sli4_hba.els_wq->queue_id == queid) {
/* Sanity check */
rc = lpfc_idiag_que_param_check(
phba->sli4_hba.els_wq, index, count);
if (rc)
goto error_out;
idiag.ptr_private = phba->sli4_hba.els_wq;
goto pass_check;
}
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-13 05:52:30 +08:00
/* NVME LS work queue */
if (phba->sli4_hba.nvmels_wq &&
phba->sli4_hba.nvmels_wq->queue_id == queid) {
/* Sanity check */
rc = lpfc_idiag_que_param_check(
phba->sli4_hba.nvmels_wq, index, count);
if (rc)
goto error_out;
idiag.ptr_private = phba->sli4_hba.nvmels_wq;
goto pass_check;
}
2019-01-29 03:14:21 +08:00
if (phba->sli4_hba.hdwq) {
/* FCP/SCSI work queue */
for (qidx = 0; qidx < phba->cfg_hdw_queue; qidx++) {
qp = phba->sli4_hba.hdwq[qidx].fcp_wq;
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-13 05:52:30 +08:00
if (qp && qp->queue_id == queid) {
/* Sanity check */
rc = lpfc_idiag_que_param_check(
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-13 05:52:30 +08:00
qp, index, count);
if (rc)
goto error_out;
idiag.ptr_private = qp;
goto pass_check;
}
}
2019-01-29 03:14:21 +08:00
/* NVME work queue */
for (qidx = 0; qidx < phba->cfg_hdw_queue; qidx++) {
qp = phba->sli4_hba.hdwq[qidx].nvme_wq;
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-13 05:52:30 +08:00
if (qp && qp->queue_id == queid) {
/* Sanity check */
rc = lpfc_idiag_que_param_check(
qp, index, count);
if (rc)
goto error_out;
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-13 05:52:30 +08:00
idiag.ptr_private = qp;
goto pass_check;
}
}
}
goto error_out;
break;
case LPFC_IDIAG_RQ:
/* HDR queue */
if (phba->sli4_hba.hdr_rq &&
phba->sli4_hba.hdr_rq->queue_id == queid) {
/* Sanity check */
rc = lpfc_idiag_que_param_check(
phba->sli4_hba.hdr_rq, index, count);
if (rc)
goto error_out;
idiag.ptr_private = phba->sli4_hba.hdr_rq;
goto pass_check;
}
/* DAT queue */
if (phba->sli4_hba.dat_rq &&
phba->sli4_hba.dat_rq->queue_id == queid) {
/* Sanity check */
rc = lpfc_idiag_que_param_check(
phba->sli4_hba.dat_rq, index, count);
if (rc)
goto error_out;
idiag.ptr_private = phba->sli4_hba.dat_rq;
goto pass_check;
}
goto error_out;
break;
default:
goto error_out;
break;
}
pass_check:
if (idiag.cmd.opcode == LPFC_IDIAG_CMD_QUEACC_RD) {
if (count == LPFC_QUE_ACC_BROWSE)
idiag.offset.last_rd = index;
}
if (idiag.cmd.opcode == LPFC_IDIAG_CMD_QUEACC_WR ||
idiag.cmd.opcode == LPFC_IDIAG_CMD_QUEACC_ST ||
idiag.cmd.opcode == LPFC_IDIAG_CMD_QUEACC_CL) {
/* Additional sanity checks on write operation */
pque = (struct lpfc_queue *)idiag.ptr_private;
if (offset > pque->entry_size/sizeof(uint32_t) - 1)
goto error_out;
pentry = lpfc_sli4_qe(pque, index);
pentry += offset;
if (idiag.cmd.opcode == LPFC_IDIAG_CMD_QUEACC_WR)
*pentry = value;
if (idiag.cmd.opcode == LPFC_IDIAG_CMD_QUEACC_ST)
*pentry |= value;
if (idiag.cmd.opcode == LPFC_IDIAG_CMD_QUEACC_CL)
*pentry &= ~value;
}
return nbytes;
error_out:
/* Clean out command structure on command error out */
memset(&idiag, 0, sizeof(idiag));
return -EINVAL;
}
/**
* lpfc_idiag_drbacc_read_reg - idiag debugfs read a doorbell register
* @phba: The pointer to hba structure.
* @pbuffer: The pointer to the buffer to copy the data to.
* @len: The lenght of bytes to copied.
* @drbregid: The id to doorbell registers.
*
* Description:
* This routine reads a doorbell register and copies its content to the
* user buffer pointed to by @pbuffer.
*
* Returns:
* This function returns the amount of data that was copied into @pbuffer.
**/
static int
lpfc_idiag_drbacc_read_reg(struct lpfc_hba *phba, char *pbuffer,
int len, uint32_t drbregid)
{
if (!pbuffer)
return 0;
switch (drbregid) {
case LPFC_DRB_EQ:
len += snprintf(pbuffer + len, LPFC_DRB_ACC_BUF_SIZE-len,
"EQ-DRB-REG: 0x%08x\n",
readl(phba->sli4_hba.EQDBregaddr));
break;
case LPFC_DRB_CQ:
len += snprintf(pbuffer + len, LPFC_DRB_ACC_BUF_SIZE - len,
"CQ-DRB-REG: 0x%08x\n",
readl(phba->sli4_hba.CQDBregaddr));
break;
case LPFC_DRB_MQ:
len += snprintf(pbuffer+len, LPFC_DRB_ACC_BUF_SIZE-len,
"MQ-DRB-REG: 0x%08x\n",
readl(phba->sli4_hba.MQDBregaddr));
break;
case LPFC_DRB_WQ:
len += snprintf(pbuffer+len, LPFC_DRB_ACC_BUF_SIZE-len,
"WQ-DRB-REG: 0x%08x\n",
readl(phba->sli4_hba.WQDBregaddr));
break;
case LPFC_DRB_RQ:
len += snprintf(pbuffer+len, LPFC_DRB_ACC_BUF_SIZE-len,
"RQ-DRB-REG: 0x%08x\n",
readl(phba->sli4_hba.RQDBregaddr));
break;
default:
break;
}
return len;
}
/**
* lpfc_idiag_drbacc_read - idiag debugfs read port doorbell
* @file: The file pointer to read from.
* @buf: The buffer to copy the data to.
* @nbytes: The number of bytes to read.
* @ppos: The position in the file to start reading from.
*
* Description:
* This routine reads data from the @phba device doorbell register according
* to the idiag command, and copies to user @buf. Depending on the doorbell
* register read command setup, it does either a single doorbell register
* read or dump all doorbell registers.
*
* Returns:
* This function returns the amount of data that was read (this could be less
* than @nbytes if the end of the file was reached) or a negative error value.
**/
static ssize_t
lpfc_idiag_drbacc_read(struct file *file, char __user *buf, size_t nbytes,
loff_t *ppos)
{
struct lpfc_debug *debug = file->private_data;
struct lpfc_hba *phba = (struct lpfc_hba *)debug->i_private;
uint32_t drb_reg_id, i;
char *pbuffer;
int len = 0;
/* This is a user read operation */
debug->op = LPFC_IDIAG_OP_RD;
if (!debug->buffer)
debug->buffer = kmalloc(LPFC_DRB_ACC_BUF_SIZE, GFP_KERNEL);
if (!debug->buffer)
return 0;
pbuffer = debug->buffer;
if (*ppos)
return 0;
if (idiag.cmd.opcode == LPFC_IDIAG_CMD_DRBACC_RD)
drb_reg_id = idiag.cmd.data[IDIAG_DRBACC_REGID_INDX];
else
return 0;
if (drb_reg_id == LPFC_DRB_ACC_ALL)
for (i = 1; i <= LPFC_DRB_MAX; i++)
len = lpfc_idiag_drbacc_read_reg(phba,
pbuffer, len, i);
else
len = lpfc_idiag_drbacc_read_reg(phba,
pbuffer, len, drb_reg_id);
return simple_read_from_buffer(buf, nbytes, ppos, pbuffer, len);
}
/**
* lpfc_idiag_drbacc_write - Syntax check and set up idiag drbacc commands
* @file: The file pointer to read from.
* @buf: The buffer to copy the user data from.
* @nbytes: The number of bytes to get.
* @ppos: The position in the file to start reading from.
*
* This routine get the debugfs idiag command struct from user space and then
* perform the syntax check for port doorbell register read (dump) or write
* (set) command accordingly. In the case of port queue read command, it sets
* up the command in the idiag command struct for the following debugfs read
* operation. In the case of port doorbell register write operation, it
* executes the write operation into the port doorbell register accordingly.
*
* It returns the @nbytges passing in from debugfs user space when successful.
* In case of error conditions, it returns proper error code back to the user
* space.
**/
static ssize_t
lpfc_idiag_drbacc_write(struct file *file, const char __user *buf,
size_t nbytes, loff_t *ppos)
{
struct lpfc_debug *debug = file->private_data;
struct lpfc_hba *phba = (struct lpfc_hba *)debug->i_private;
uint32_t drb_reg_id, value, reg_val = 0;
void __iomem *drb_reg;
int rc;
/* This is a user write operation */
debug->op = LPFC_IDIAG_OP_WR;
rc = lpfc_idiag_cmd_get(buf, nbytes, &idiag.cmd);
if (rc < 0)
return rc;
/* Sanity check on command line arguments */
drb_reg_id = idiag.cmd.data[IDIAG_DRBACC_REGID_INDX];
value = idiag.cmd.data[IDIAG_DRBACC_VALUE_INDX];
if (idiag.cmd.opcode == LPFC_IDIAG_CMD_DRBACC_WR ||
idiag.cmd.opcode == LPFC_IDIAG_CMD_DRBACC_ST ||
idiag.cmd.opcode == LPFC_IDIAG_CMD_DRBACC_CL) {
if (rc != LPFC_DRB_ACC_WR_CMD_ARG)
goto error_out;
if (drb_reg_id > LPFC_DRB_MAX)
goto error_out;
} else if (idiag.cmd.opcode == LPFC_IDIAG_CMD_DRBACC_RD) {
if (rc != LPFC_DRB_ACC_RD_CMD_ARG)
goto error_out;
if ((drb_reg_id > LPFC_DRB_MAX) &&
(drb_reg_id != LPFC_DRB_ACC_ALL))
goto error_out;
} else
goto error_out;
/* Perform the write access operation */
if (idiag.cmd.opcode == LPFC_IDIAG_CMD_DRBACC_WR ||
idiag.cmd.opcode == LPFC_IDIAG_CMD_DRBACC_ST ||
idiag.cmd.opcode == LPFC_IDIAG_CMD_DRBACC_CL) {
switch (drb_reg_id) {
case LPFC_DRB_EQ:
drb_reg = phba->sli4_hba.EQDBregaddr;
break;
case LPFC_DRB_CQ:
drb_reg = phba->sli4_hba.CQDBregaddr;
break;
case LPFC_DRB_MQ:
drb_reg = phba->sli4_hba.MQDBregaddr;
break;
case LPFC_DRB_WQ:
drb_reg = phba->sli4_hba.WQDBregaddr;
break;
case LPFC_DRB_RQ:
drb_reg = phba->sli4_hba.RQDBregaddr;
break;
default:
goto error_out;
}
if (idiag.cmd.opcode == LPFC_IDIAG_CMD_DRBACC_WR)
reg_val = value;
if (idiag.cmd.opcode == LPFC_IDIAG_CMD_DRBACC_ST) {
reg_val = readl(drb_reg);
reg_val |= value;
}
if (idiag.cmd.opcode == LPFC_IDIAG_CMD_DRBACC_CL) {
reg_val = readl(drb_reg);
reg_val &= ~value;
}
writel(reg_val, drb_reg);
readl(drb_reg); /* flush */
}
return nbytes;
error_out:
/* Clean out command structure on command error out */
memset(&idiag, 0, sizeof(idiag));
return -EINVAL;
}
/**
* lpfc_idiag_ctlacc_read_reg - idiag debugfs read a control registers
* @phba: The pointer to hba structure.
* @pbuffer: The pointer to the buffer to copy the data to.
* @len: The lenght of bytes to copied.
* @drbregid: The id to doorbell registers.
*
* Description:
* This routine reads a control register and copies its content to the
* user buffer pointed to by @pbuffer.
*
* Returns:
* This function returns the amount of data that was copied into @pbuffer.
**/
static int
lpfc_idiag_ctlacc_read_reg(struct lpfc_hba *phba, char *pbuffer,
int len, uint32_t ctlregid)
{
if (!pbuffer)
return 0;
switch (ctlregid) {
case LPFC_CTL_PORT_SEM:
len += snprintf(pbuffer+len, LPFC_CTL_ACC_BUF_SIZE-len,
"Port SemReg: 0x%08x\n",
readl(phba->sli4_hba.conf_regs_memmap_p +
LPFC_CTL_PORT_SEM_OFFSET));
break;
case LPFC_CTL_PORT_STA:
len += snprintf(pbuffer+len, LPFC_CTL_ACC_BUF_SIZE-len,
"Port StaReg: 0x%08x\n",
readl(phba->sli4_hba.conf_regs_memmap_p +
LPFC_CTL_PORT_STA_OFFSET));
break;
case LPFC_CTL_PORT_CTL:
len += snprintf(pbuffer+len, LPFC_CTL_ACC_BUF_SIZE-len,
"Port CtlReg: 0x%08x\n",
readl(phba->sli4_hba.conf_regs_memmap_p +
LPFC_CTL_PORT_CTL_OFFSET));
break;
case LPFC_CTL_PORT_ER1:
len += snprintf(pbuffer+len, LPFC_CTL_ACC_BUF_SIZE-len,
"Port Er1Reg: 0x%08x\n",
readl(phba->sli4_hba.conf_regs_memmap_p +
LPFC_CTL_PORT_ER1_OFFSET));
break;
case LPFC_CTL_PORT_ER2:
len += snprintf(pbuffer+len, LPFC_CTL_ACC_BUF_SIZE-len,
"Port Er2Reg: 0x%08x\n",
readl(phba->sli4_hba.conf_regs_memmap_p +
LPFC_CTL_PORT_ER2_OFFSET));
break;
case LPFC_CTL_PDEV_CTL:
len += snprintf(pbuffer+len, LPFC_CTL_ACC_BUF_SIZE-len,
"PDev CtlReg: 0x%08x\n",
readl(phba->sli4_hba.conf_regs_memmap_p +
LPFC_CTL_PDEV_CTL_OFFSET));
break;
default:
break;
}
return len;
}
/**
* lpfc_idiag_ctlacc_read - idiag debugfs read port and device control register
* @file: The file pointer to read from.
* @buf: The buffer to copy the data to.
* @nbytes: The number of bytes to read.
* @ppos: The position in the file to start reading from.
*
* Description:
* This routine reads data from the @phba port and device registers according
* to the idiag command, and copies to user @buf.
*
* Returns:
* This function returns the amount of data that was read (this could be less
* than @nbytes if the end of the file was reached) or a negative error value.
**/
static ssize_t
lpfc_idiag_ctlacc_read(struct file *file, char __user *buf, size_t nbytes,
loff_t *ppos)
{
struct lpfc_debug *debug = file->private_data;
struct lpfc_hba *phba = (struct lpfc_hba *)debug->i_private;
uint32_t ctl_reg_id, i;
char *pbuffer;
int len = 0;
/* This is a user read operation */
debug->op = LPFC_IDIAG_OP_RD;
if (!debug->buffer)
debug->buffer = kmalloc(LPFC_CTL_ACC_BUF_SIZE, GFP_KERNEL);
if (!debug->buffer)
return 0;
pbuffer = debug->buffer;
if (*ppos)
return 0;
if (idiag.cmd.opcode == LPFC_IDIAG_CMD_CTLACC_RD)
ctl_reg_id = idiag.cmd.data[IDIAG_CTLACC_REGID_INDX];
else
return 0;
if (ctl_reg_id == LPFC_CTL_ACC_ALL)
for (i = 1; i <= LPFC_CTL_MAX; i++)
len = lpfc_idiag_ctlacc_read_reg(phba,
pbuffer, len, i);
else
len = lpfc_idiag_ctlacc_read_reg(phba,
pbuffer, len, ctl_reg_id);
return simple_read_from_buffer(buf, nbytes, ppos, pbuffer, len);
}
/**
* lpfc_idiag_ctlacc_write - Syntax check and set up idiag ctlacc commands
* @file: The file pointer to read from.
* @buf: The buffer to copy the user data from.
* @nbytes: The number of bytes to get.
* @ppos: The position in the file to start reading from.
*
* This routine get the debugfs idiag command struct from user space and then
* perform the syntax check for port and device control register read (dump)
* or write (set) command accordingly.
*
* It returns the @nbytges passing in from debugfs user space when successful.
* In case of error conditions, it returns proper error code back to the user
* space.
**/
static ssize_t
lpfc_idiag_ctlacc_write(struct file *file, const char __user *buf,
size_t nbytes, loff_t *ppos)
{
struct lpfc_debug *debug = file->private_data;
struct lpfc_hba *phba = (struct lpfc_hba *)debug->i_private;
uint32_t ctl_reg_id, value, reg_val = 0;
void __iomem *ctl_reg;
int rc;
/* This is a user write operation */
debug->op = LPFC_IDIAG_OP_WR;
rc = lpfc_idiag_cmd_get(buf, nbytes, &idiag.cmd);
if (rc < 0)
return rc;
/* Sanity check on command line arguments */
ctl_reg_id = idiag.cmd.data[IDIAG_CTLACC_REGID_INDX];
value = idiag.cmd.data[IDIAG_CTLACC_VALUE_INDX];
if (idiag.cmd.opcode == LPFC_IDIAG_CMD_CTLACC_WR ||
idiag.cmd.opcode == LPFC_IDIAG_CMD_CTLACC_ST ||
idiag.cmd.opcode == LPFC_IDIAG_CMD_CTLACC_CL) {
if (rc != LPFC_CTL_ACC_WR_CMD_ARG)
goto error_out;
if (ctl_reg_id > LPFC_CTL_MAX)
goto error_out;
} else if (idiag.cmd.opcode == LPFC_IDIAG_CMD_CTLACC_RD) {
if (rc != LPFC_CTL_ACC_RD_CMD_ARG)
goto error_out;
if ((ctl_reg_id > LPFC_CTL_MAX) &&
(ctl_reg_id != LPFC_CTL_ACC_ALL))
goto error_out;
} else
goto error_out;
/* Perform the write access operation */
if (idiag.cmd.opcode == LPFC_IDIAG_CMD_CTLACC_WR ||
idiag.cmd.opcode == LPFC_IDIAG_CMD_CTLACC_ST ||
idiag.cmd.opcode == LPFC_IDIAG_CMD_CTLACC_CL) {
switch (ctl_reg_id) {
case LPFC_CTL_PORT_SEM:
ctl_reg = phba->sli4_hba.conf_regs_memmap_p +
LPFC_CTL_PORT_SEM_OFFSET;
break;
case LPFC_CTL_PORT_STA:
ctl_reg = phba->sli4_hba.conf_regs_memmap_p +
LPFC_CTL_PORT_STA_OFFSET;
break;
case LPFC_CTL_PORT_CTL:
ctl_reg = phba->sli4_hba.conf_regs_memmap_p +
LPFC_CTL_PORT_CTL_OFFSET;
break;
case LPFC_CTL_PORT_ER1:
ctl_reg = phba->sli4_hba.conf_regs_memmap_p +
LPFC_CTL_PORT_ER1_OFFSET;
break;
case LPFC_CTL_PORT_ER2:
ctl_reg = phba->sli4_hba.conf_regs_memmap_p +
LPFC_CTL_PORT_ER2_OFFSET;
break;
case LPFC_CTL_PDEV_CTL:
ctl_reg = phba->sli4_hba.conf_regs_memmap_p +
LPFC_CTL_PDEV_CTL_OFFSET;
break;
default:
goto error_out;
}
if (idiag.cmd.opcode == LPFC_IDIAG_CMD_CTLACC_WR)
reg_val = value;
if (idiag.cmd.opcode == LPFC_IDIAG_CMD_CTLACC_ST) {
reg_val = readl(ctl_reg);
reg_val |= value;
}
if (idiag.cmd.opcode == LPFC_IDIAG_CMD_CTLACC_CL) {
reg_val = readl(ctl_reg);
reg_val &= ~value;
}
writel(reg_val, ctl_reg);
readl(ctl_reg); /* flush */
}
return nbytes;
error_out:
/* Clean out command structure on command error out */
memset(&idiag, 0, sizeof(idiag));
return -EINVAL;
}
/**
* lpfc_idiag_mbxacc_get_setup - idiag debugfs get mailbox access setup
* @phba: Pointer to HBA context object.
* @pbuffer: Pointer to data buffer.
*
* Description:
* This routine gets the driver mailbox access debugfs setup information.
*
* Returns:
* This function returns the amount of data that was read (this could be less
* than @nbytes if the end of the file was reached) or a negative error value.
**/
static int
lpfc_idiag_mbxacc_get_setup(struct lpfc_hba *phba, char *pbuffer)
{
uint32_t mbx_dump_map, mbx_dump_cnt, mbx_word_cnt, mbx_mbox_cmd;
int len = 0;
mbx_mbox_cmd = idiag.cmd.data[IDIAG_MBXACC_MBCMD_INDX];
mbx_dump_map = idiag.cmd.data[IDIAG_MBXACC_DPMAP_INDX];
mbx_dump_cnt = idiag.cmd.data[IDIAG_MBXACC_DPCNT_INDX];
mbx_word_cnt = idiag.cmd.data[IDIAG_MBXACC_WDCNT_INDX];
len += snprintf(pbuffer+len, LPFC_MBX_ACC_BUF_SIZE-len,
"mbx_dump_map: 0x%08x\n", mbx_dump_map);
len += snprintf(pbuffer+len, LPFC_MBX_ACC_BUF_SIZE-len,
"mbx_dump_cnt: %04d\n", mbx_dump_cnt);
len += snprintf(pbuffer+len, LPFC_MBX_ACC_BUF_SIZE-len,
"mbx_word_cnt: %04d\n", mbx_word_cnt);
len += snprintf(pbuffer+len, LPFC_MBX_ACC_BUF_SIZE-len,
"mbx_mbox_cmd: 0x%02x\n", mbx_mbox_cmd);
return len;
}
/**
* lpfc_idiag_mbxacc_read - idiag debugfs read on mailbox access
* @file: The file pointer to read from.
* @buf: The buffer to copy the data to.
* @nbytes: The number of bytes to read.
* @ppos: The position in the file to start reading from.
*
* Description:
* This routine reads data from the @phba driver mailbox access debugfs setup
* information.
*
* Returns:
* This function returns the amount of data that was read (this could be less
* than @nbytes if the end of the file was reached) or a negative error value.
**/
static ssize_t
lpfc_idiag_mbxacc_read(struct file *file, char __user *buf, size_t nbytes,
loff_t *ppos)
{
struct lpfc_debug *debug = file->private_data;
struct lpfc_hba *phba = (struct lpfc_hba *)debug->i_private;
char *pbuffer;
int len = 0;
/* This is a user read operation */
debug->op = LPFC_IDIAG_OP_RD;
if (!debug->buffer)
debug->buffer = kmalloc(LPFC_MBX_ACC_BUF_SIZE, GFP_KERNEL);
if (!debug->buffer)
return 0;
pbuffer = debug->buffer;
if (*ppos)
return 0;
if ((idiag.cmd.opcode != LPFC_IDIAG_CMD_MBXACC_DP) &&
(idiag.cmd.opcode != LPFC_IDIAG_BSG_MBXACC_DP))
return 0;
len = lpfc_idiag_mbxacc_get_setup(phba, pbuffer);
return simple_read_from_buffer(buf, nbytes, ppos, pbuffer, len);
}
/**
* lpfc_idiag_mbxacc_write - Syntax check and set up idiag mbxacc commands
* @file: The file pointer to read from.
* @buf: The buffer to copy the user data from.
* @nbytes: The number of bytes to get.
* @ppos: The position in the file to start reading from.
*
* This routine get the debugfs idiag command struct from user space and then
* perform the syntax check for driver mailbox command (dump) and sets up the
* necessary states in the idiag command struct accordingly.
*
* It returns the @nbytges passing in from debugfs user space when successful.
* In case of error conditions, it returns proper error code back to the user
* space.
**/
static ssize_t
lpfc_idiag_mbxacc_write(struct file *file, const char __user *buf,
size_t nbytes, loff_t *ppos)
{
struct lpfc_debug *debug = file->private_data;
uint32_t mbx_dump_map, mbx_dump_cnt, mbx_word_cnt, mbx_mbox_cmd;
int rc;
/* This is a user write operation */
debug->op = LPFC_IDIAG_OP_WR;
rc = lpfc_idiag_cmd_get(buf, nbytes, &idiag.cmd);
if (rc < 0)
return rc;
/* Sanity check on command line arguments */
mbx_mbox_cmd = idiag.cmd.data[IDIAG_MBXACC_MBCMD_INDX];
mbx_dump_map = idiag.cmd.data[IDIAG_MBXACC_DPMAP_INDX];
mbx_dump_cnt = idiag.cmd.data[IDIAG_MBXACC_DPCNT_INDX];
mbx_word_cnt = idiag.cmd.data[IDIAG_MBXACC_WDCNT_INDX];
if (idiag.cmd.opcode == LPFC_IDIAG_CMD_MBXACC_DP) {
if (!(mbx_dump_map & LPFC_MBX_DMP_MBX_ALL))
goto error_out;
if ((mbx_dump_map & ~LPFC_MBX_DMP_MBX_ALL) &&
(mbx_dump_map != LPFC_MBX_DMP_ALL))
goto error_out;
if (mbx_word_cnt > sizeof(MAILBOX_t))
goto error_out;
} else if (idiag.cmd.opcode == LPFC_IDIAG_BSG_MBXACC_DP) {
if (!(mbx_dump_map & LPFC_BSG_DMP_MBX_ALL))
goto error_out;
if ((mbx_dump_map & ~LPFC_BSG_DMP_MBX_ALL) &&
(mbx_dump_map != LPFC_MBX_DMP_ALL))
goto error_out;
if (mbx_word_cnt > (BSG_MBOX_SIZE)/4)
goto error_out;
if (mbx_mbox_cmd != 0x9b)
goto error_out;
} else
goto error_out;
if (mbx_word_cnt == 0)
goto error_out;
if (rc != LPFC_MBX_DMP_ARG)
goto error_out;
if (mbx_mbox_cmd & ~0xff)
goto error_out;
/* condition for stop mailbox dump */
if (mbx_dump_cnt == 0)
goto reset_out;
return nbytes;
reset_out:
/* Clean out command structure on command error out */
memset(&idiag, 0, sizeof(idiag));
return nbytes;
error_out:
/* Clean out command structure on command error out */
memset(&idiag, 0, sizeof(idiag));
return -EINVAL;
}
/**
* lpfc_idiag_extacc_avail_get - get the available extents information
* @phba: pointer to lpfc hba data structure.
* @pbuffer: pointer to internal buffer.
* @len: length into the internal buffer data has been copied.
*
* Description:
* This routine is to get the available extent information.
*
* Returns:
* overall lenth of the data read into the internal buffer.
**/
static int
lpfc_idiag_extacc_avail_get(struct lpfc_hba *phba, char *pbuffer, int len)
{
uint16_t ext_cnt, ext_size;
len += snprintf(pbuffer+len, LPFC_EXT_ACC_BUF_SIZE-len,
"\nAvailable Extents Information:\n");
len += snprintf(pbuffer+len, LPFC_EXT_ACC_BUF_SIZE-len,
"\tPort Available VPI extents: ");
lpfc_sli4_get_avail_extnt_rsrc(phba, LPFC_RSC_TYPE_FCOE_VPI,
&ext_cnt, &ext_size);
len += snprintf(pbuffer+len, LPFC_EXT_ACC_BUF_SIZE-len,
"Count %3d, Size %3d\n", ext_cnt, ext_size);
len += snprintf(pbuffer+len, LPFC_EXT_ACC_BUF_SIZE-len,
"\tPort Available VFI extents: ");
lpfc_sli4_get_avail_extnt_rsrc(phba, LPFC_RSC_TYPE_FCOE_VFI,
&ext_cnt, &ext_size);
len += snprintf(pbuffer+len, LPFC_EXT_ACC_BUF_SIZE-len,
"Count %3d, Size %3d\n", ext_cnt, ext_size);
len += snprintf(pbuffer+len, LPFC_EXT_ACC_BUF_SIZE-len,
"\tPort Available RPI extents: ");
lpfc_sli4_get_avail_extnt_rsrc(phba, LPFC_RSC_TYPE_FCOE_RPI,
&ext_cnt, &ext_size);
len += snprintf(pbuffer+len, LPFC_EXT_ACC_BUF_SIZE-len,
"Count %3d, Size %3d\n", ext_cnt, ext_size);
len += snprintf(pbuffer+len, LPFC_EXT_ACC_BUF_SIZE-len,
"\tPort Available XRI extents: ");
lpfc_sli4_get_avail_extnt_rsrc(phba, LPFC_RSC_TYPE_FCOE_XRI,
&ext_cnt, &ext_size);
len += snprintf(pbuffer+len, LPFC_EXT_ACC_BUF_SIZE-len,
"Count %3d, Size %3d\n", ext_cnt, ext_size);
return len;
}
/**
* lpfc_idiag_extacc_alloc_get - get the allocated extents information
* @phba: pointer to lpfc hba data structure.
* @pbuffer: pointer to internal buffer.
* @len: length into the internal buffer data has been copied.
*
* Description:
* This routine is to get the allocated extent information.
*
* Returns:
* overall lenth of the data read into the internal buffer.
**/
static int
lpfc_idiag_extacc_alloc_get(struct lpfc_hba *phba, char *pbuffer, int len)
{
uint16_t ext_cnt, ext_size;
int rc;
len += snprintf(pbuffer+len, LPFC_EXT_ACC_BUF_SIZE-len,
"\nAllocated Extents Information:\n");
len += snprintf(pbuffer+len, LPFC_EXT_ACC_BUF_SIZE-len,
"\tHost Allocated VPI extents: ");
rc = lpfc_sli4_get_allocated_extnts(phba, LPFC_RSC_TYPE_FCOE_VPI,
&ext_cnt, &ext_size);
if (!rc)
len += snprintf(pbuffer+len, LPFC_EXT_ACC_BUF_SIZE-len,
"Port %d Extent %3d, Size %3d\n",
phba->brd_no, ext_cnt, ext_size);
else
len += snprintf(pbuffer+len, LPFC_EXT_ACC_BUF_SIZE-len,
"N/A\n");
len += snprintf(pbuffer+len, LPFC_EXT_ACC_BUF_SIZE-len,
"\tHost Allocated VFI extents: ");
rc = lpfc_sli4_get_allocated_extnts(phba, LPFC_RSC_TYPE_FCOE_VFI,
&ext_cnt, &ext_size);
if (!rc)
len += snprintf(pbuffer+len, LPFC_EXT_ACC_BUF_SIZE-len,
"Port %d Extent %3d, Size %3d\n",
phba->brd_no, ext_cnt, ext_size);
else
len += snprintf(pbuffer+len, LPFC_EXT_ACC_BUF_SIZE-len,
"N/A\n");
len += snprintf(pbuffer+len, LPFC_EXT_ACC_BUF_SIZE-len,
"\tHost Allocated RPI extents: ");
rc = lpfc_sli4_get_allocated_extnts(phba, LPFC_RSC_TYPE_FCOE_RPI,
&ext_cnt, &ext_size);
if (!rc)
len += snprintf(pbuffer+len, LPFC_EXT_ACC_BUF_SIZE-len,
"Port %d Extent %3d, Size %3d\n",
phba->brd_no, ext_cnt, ext_size);
else
len += snprintf(pbuffer+len, LPFC_EXT_ACC_BUF_SIZE-len,
"N/A\n");
len += snprintf(pbuffer+len, LPFC_EXT_ACC_BUF_SIZE-len,
"\tHost Allocated XRI extents: ");
rc = lpfc_sli4_get_allocated_extnts(phba, LPFC_RSC_TYPE_FCOE_XRI,
&ext_cnt, &ext_size);
if (!rc)
len += snprintf(pbuffer+len, LPFC_EXT_ACC_BUF_SIZE-len,
"Port %d Extent %3d, Size %3d\n",
phba->brd_no, ext_cnt, ext_size);
else
len += snprintf(pbuffer+len, LPFC_EXT_ACC_BUF_SIZE-len,
"N/A\n");
return len;
}
/**
* lpfc_idiag_extacc_drivr_get - get driver extent information
* @phba: pointer to lpfc hba data structure.
* @pbuffer: pointer to internal buffer.
* @len: length into the internal buffer data has been copied.
*
* Description:
* This routine is to get the driver extent information.
*
* Returns:
* overall lenth of the data read into the internal buffer.
**/
static int
lpfc_idiag_extacc_drivr_get(struct lpfc_hba *phba, char *pbuffer, int len)
{
struct lpfc_rsrc_blks *rsrc_blks;
int index;
len += snprintf(pbuffer+len, LPFC_EXT_ACC_BUF_SIZE-len,
"\nDriver Extents Information:\n");
len += snprintf(pbuffer+len, LPFC_EXT_ACC_BUF_SIZE-len,
"\tVPI extents:\n");
index = 0;
list_for_each_entry(rsrc_blks, &phba->lpfc_vpi_blk_list, list) {
len += snprintf(pbuffer+len, LPFC_EXT_ACC_BUF_SIZE-len,
"\t\tBlock %3d: Start %4d, Count %4d\n",
index, rsrc_blks->rsrc_start,
rsrc_blks->rsrc_size);
index++;
}
len += snprintf(pbuffer+len, LPFC_EXT_ACC_BUF_SIZE-len,
"\tVFI extents:\n");
index = 0;
list_for_each_entry(rsrc_blks, &phba->sli4_hba.lpfc_vfi_blk_list,
list) {
len += snprintf(pbuffer+len, LPFC_EXT_ACC_BUF_SIZE-len,
"\t\tBlock %3d: Start %4d, Count %4d\n",
index, rsrc_blks->rsrc_start,
rsrc_blks->rsrc_size);
index++;
}
len += snprintf(pbuffer+len, LPFC_EXT_ACC_BUF_SIZE-len,
"\tRPI extents:\n");
index = 0;
list_for_each_entry(rsrc_blks, &phba->sli4_hba.lpfc_rpi_blk_list,
list) {
len += snprintf(pbuffer+len, LPFC_EXT_ACC_BUF_SIZE-len,
"\t\tBlock %3d: Start %4d, Count %4d\n",
index, rsrc_blks->rsrc_start,
rsrc_blks->rsrc_size);
index++;
}
len += snprintf(pbuffer+len, LPFC_EXT_ACC_BUF_SIZE-len,
"\tXRI extents:\n");
index = 0;
list_for_each_entry(rsrc_blks, &phba->sli4_hba.lpfc_xri_blk_list,
list) {
len += snprintf(pbuffer+len, LPFC_EXT_ACC_BUF_SIZE-len,
"\t\tBlock %3d: Start %4d, Count %4d\n",
index, rsrc_blks->rsrc_start,
rsrc_blks->rsrc_size);
index++;
}
return len;
}
/**
* lpfc_idiag_extacc_write - Syntax check and set up idiag extacc commands
* @file: The file pointer to read from.
* @buf: The buffer to copy the user data from.
* @nbytes: The number of bytes to get.
* @ppos: The position in the file to start reading from.
*
* This routine get the debugfs idiag command struct from user space and then
* perform the syntax check for extent information access commands and sets
* up the necessary states in the idiag command struct accordingly.
*
* It returns the @nbytges passing in from debugfs user space when successful.
* In case of error conditions, it returns proper error code back to the user
* space.
**/
static ssize_t
lpfc_idiag_extacc_write(struct file *file, const char __user *buf,
size_t nbytes, loff_t *ppos)
{
struct lpfc_debug *debug = file->private_data;
uint32_t ext_map;
int rc;
/* This is a user write operation */
debug->op = LPFC_IDIAG_OP_WR;
rc = lpfc_idiag_cmd_get(buf, nbytes, &idiag.cmd);
if (rc < 0)
return rc;
ext_map = idiag.cmd.data[IDIAG_EXTACC_EXMAP_INDX];
if (idiag.cmd.opcode != LPFC_IDIAG_CMD_EXTACC_RD)
goto error_out;
if (rc != LPFC_EXT_ACC_CMD_ARG)
goto error_out;
if (!(ext_map & LPFC_EXT_ACC_ALL))
goto error_out;
return nbytes;
error_out:
/* Clean out command structure on command error out */
memset(&idiag, 0, sizeof(idiag));
return -EINVAL;
}
/**
* lpfc_idiag_extacc_read - idiag debugfs read access to extent information
* @file: The file pointer to read from.
* @buf: The buffer to copy the data to.
* @nbytes: The number of bytes to read.
* @ppos: The position in the file to start reading from.
*
* Description:
* This routine reads data from the proper extent information according to
* the idiag command, and copies to user @buf.
*
* Returns:
* This function returns the amount of data that was read (this could be less
* than @nbytes if the end of the file was reached) or a negative error value.
**/
static ssize_t
lpfc_idiag_extacc_read(struct file *file, char __user *buf, size_t nbytes,
loff_t *ppos)
{
struct lpfc_debug *debug = file->private_data;
struct lpfc_hba *phba = (struct lpfc_hba *)debug->i_private;
char *pbuffer;
uint32_t ext_map;
int len = 0;
/* This is a user read operation */
debug->op = LPFC_IDIAG_OP_RD;
if (!debug->buffer)
debug->buffer = kmalloc(LPFC_EXT_ACC_BUF_SIZE, GFP_KERNEL);
if (!debug->buffer)
return 0;
pbuffer = debug->buffer;
if (*ppos)
return 0;
if (idiag.cmd.opcode != LPFC_IDIAG_CMD_EXTACC_RD)
return 0;
ext_map = idiag.cmd.data[IDIAG_EXTACC_EXMAP_INDX];
if (ext_map & LPFC_EXT_ACC_AVAIL)
len = lpfc_idiag_extacc_avail_get(phba, pbuffer, len);
if (ext_map & LPFC_EXT_ACC_ALLOC)
len = lpfc_idiag_extacc_alloc_get(phba, pbuffer, len);
if (ext_map & LPFC_EXT_ACC_DRIVR)
len = lpfc_idiag_extacc_drivr_get(phba, pbuffer, len);
return simple_read_from_buffer(buf, nbytes, ppos, pbuffer, len);
}
#undef lpfc_debugfs_op_disc_trc
static const struct file_operations lpfc_debugfs_op_disc_trc = {
.owner = THIS_MODULE,
.open = lpfc_debugfs_disc_trc_open,
.llseek = lpfc_debugfs_lseek,
.read = lpfc_debugfs_read,
.release = lpfc_debugfs_release,
};
#undef lpfc_debugfs_op_nodelist
static const struct file_operations lpfc_debugfs_op_nodelist = {
.owner = THIS_MODULE,
.open = lpfc_debugfs_nodelist_open,
.llseek = lpfc_debugfs_lseek,
.read = lpfc_debugfs_read,
.release = lpfc_debugfs_release,
};
scsi: lpfc: Adapt partitioned XRI lists to efficient sharing The XRI get/put lists were partitioned per hardware queue. However, the adapter rarely had sufficient resources to give a large number of resources per queue. As such, it became common for a cpu to encounter a lack of XRI resource and request the upper io stack to retry after returning a BUSY condition. This occurred even though other cpus were idle and not using their resources. Create as efficient a scheme as possible to move resources to the cpus that need them. Each cpu maintains a small private pool which it allocates from for io. There is a watermark that the cpu attempts to keep in the private pool. The private pool, when empty, pulls from a global pool from the cpu. When the cpu's global pool is empty it will pull from other cpu's global pool. As there many cpu global pools (1 per cpu or hardware queue count) and as each cpu selects what cpu to pull from at different rates and at different times, it creates a radomizing effect that minimizes the number of cpu's that will contend with each other when the steal XRI's from another cpu's global pool. On io completion, a cpu will push the XRI back on to its private pool. A watermark level is maintained for the private pool such that when it is exceeded it will move XRI's to the CPU global pool so that other cpu's may allocate them. On NVME, as heartbeat commands are critical to get placed on the wire, a single expedite pool is maintained. When a heartbeat is to be sent, it will allocate an XRI from the expedite pool rather than the normal cpu private/global pools. On any io completion, if a reduction in the expedite pools is seen, it will be replenished before the XRI is placed on the cpu private pool. Statistics are added to aid understanding the XRI levels on each cpu and their behaviors. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <jsmart2021@gmail.com> Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2019-01-29 03:14:28 +08:00
#undef lpfc_debugfs_op_multixripools
static const struct file_operations lpfc_debugfs_op_multixripools = {
.owner = THIS_MODULE,
.open = lpfc_debugfs_multixripools_open,
.llseek = lpfc_debugfs_lseek,
.read = lpfc_debugfs_read,
.write = lpfc_debugfs_multixripools_write,
.release = lpfc_debugfs_release,
};
#undef lpfc_debugfs_op_hbqinfo
static const struct file_operations lpfc_debugfs_op_hbqinfo = {
.owner = THIS_MODULE,
.open = lpfc_debugfs_hbqinfo_open,
.llseek = lpfc_debugfs_lseek,
.read = lpfc_debugfs_read,
.release = lpfc_debugfs_release,
};
#ifdef LPFC_HDWQ_LOCK_STAT
#undef lpfc_debugfs_op_lockstat
static const struct file_operations lpfc_debugfs_op_lockstat = {
.owner = THIS_MODULE,
.open = lpfc_debugfs_lockstat_open,
.llseek = lpfc_debugfs_lseek,
.read = lpfc_debugfs_read,
.write = lpfc_debugfs_lockstat_write,
.release = lpfc_debugfs_release,
};
#endif
#undef lpfc_debugfs_op_dumpHBASlim
static const struct file_operations lpfc_debugfs_op_dumpHBASlim = {
.owner = THIS_MODULE,
.open = lpfc_debugfs_dumpHBASlim_open,
.llseek = lpfc_debugfs_lseek,
.read = lpfc_debugfs_read,
.release = lpfc_debugfs_release,
};
#undef lpfc_debugfs_op_dumpHostSlim
static const struct file_operations lpfc_debugfs_op_dumpHostSlim = {
.owner = THIS_MODULE,
.open = lpfc_debugfs_dumpHostSlim_open,
.llseek = lpfc_debugfs_lseek,
.read = lpfc_debugfs_read,
.release = lpfc_debugfs_release,
};
#undef lpfc_debugfs_op_nvmestat
static const struct file_operations lpfc_debugfs_op_nvmestat = {
.owner = THIS_MODULE,
.open = lpfc_debugfs_nvmestat_open,
.llseek = lpfc_debugfs_lseek,
.read = lpfc_debugfs_read,
.write = lpfc_debugfs_nvmestat_write,
.release = lpfc_debugfs_release,
};
#undef lpfc_debugfs_op_scsistat
static const struct file_operations lpfc_debugfs_op_scsistat = {
.owner = THIS_MODULE,
.open = lpfc_debugfs_scsistat_open,
.llseek = lpfc_debugfs_lseek,
.read = lpfc_debugfs_read,
.write = lpfc_debugfs_scsistat_write,
.release = lpfc_debugfs_release,
};
#undef lpfc_debugfs_op_nvmektime
static const struct file_operations lpfc_debugfs_op_nvmektime = {
.owner = THIS_MODULE,
.open = lpfc_debugfs_nvmektime_open,
.llseek = lpfc_debugfs_lseek,
.read = lpfc_debugfs_read,
.write = lpfc_debugfs_nvmektime_write,
.release = lpfc_debugfs_release,
};
#undef lpfc_debugfs_op_nvmeio_trc
static const struct file_operations lpfc_debugfs_op_nvmeio_trc = {
.owner = THIS_MODULE,
.open = lpfc_debugfs_nvmeio_trc_open,
.llseek = lpfc_debugfs_lseek,
.read = lpfc_debugfs_read,
.write = lpfc_debugfs_nvmeio_trc_write,
.release = lpfc_debugfs_release,
};
#undef lpfc_debugfs_op_cpucheck
static const struct file_operations lpfc_debugfs_op_cpucheck = {
.owner = THIS_MODULE,
.open = lpfc_debugfs_cpucheck_open,
.llseek = lpfc_debugfs_lseek,
.read = lpfc_debugfs_read,
.write = lpfc_debugfs_cpucheck_write,
.release = lpfc_debugfs_release,
};
#undef lpfc_debugfs_op_dumpData
static const struct file_operations lpfc_debugfs_op_dumpData = {
.owner = THIS_MODULE,
.open = lpfc_debugfs_dumpData_open,
.llseek = lpfc_debugfs_lseek,
.read = lpfc_debugfs_read,
.write = lpfc_debugfs_dumpDataDif_write,
.release = lpfc_debugfs_dumpDataDif_release,
};
#undef lpfc_debugfs_op_dumpDif
static const struct file_operations lpfc_debugfs_op_dumpDif = {
.owner = THIS_MODULE,
.open = lpfc_debugfs_dumpDif_open,
.llseek = lpfc_debugfs_lseek,
.read = lpfc_debugfs_read,
.write = lpfc_debugfs_dumpDataDif_write,
.release = lpfc_debugfs_dumpDataDif_release,
};
#undef lpfc_debugfs_op_dif_err
static const struct file_operations lpfc_debugfs_op_dif_err = {
.owner = THIS_MODULE,
.open = simple_open,
.llseek = lpfc_debugfs_lseek,
.read = lpfc_debugfs_dif_err_read,
.write = lpfc_debugfs_dif_err_write,
.release = lpfc_debugfs_dif_err_release,
};
#undef lpfc_debugfs_op_slow_ring_trc
static const struct file_operations lpfc_debugfs_op_slow_ring_trc = {
.owner = THIS_MODULE,
.open = lpfc_debugfs_slow_ring_trc_open,
.llseek = lpfc_debugfs_lseek,
.read = lpfc_debugfs_read,
.release = lpfc_debugfs_release,
};
static struct dentry *lpfc_debugfs_root = NULL;
static atomic_t lpfc_debugfs_hba_count;
/*
* File operations for the iDiag debugfs
*/
#undef lpfc_idiag_op_pciCfg
static const struct file_operations lpfc_idiag_op_pciCfg = {
.owner = THIS_MODULE,
.open = lpfc_idiag_open,
.llseek = lpfc_debugfs_lseek,
.read = lpfc_idiag_pcicfg_read,
.write = lpfc_idiag_pcicfg_write,
.release = lpfc_idiag_cmd_release,
};
#undef lpfc_idiag_op_barAcc
static const struct file_operations lpfc_idiag_op_barAcc = {
.owner = THIS_MODULE,
.open = lpfc_idiag_open,
.llseek = lpfc_debugfs_lseek,
.read = lpfc_idiag_baracc_read,
.write = lpfc_idiag_baracc_write,
.release = lpfc_idiag_cmd_release,
};
#undef lpfc_idiag_op_queInfo
static const struct file_operations lpfc_idiag_op_queInfo = {
.owner = THIS_MODULE,
.open = lpfc_idiag_open,
.read = lpfc_idiag_queinfo_read,
.release = lpfc_idiag_release,
};
#undef lpfc_idiag_op_queAcc
static const struct file_operations lpfc_idiag_op_queAcc = {
.owner = THIS_MODULE,
.open = lpfc_idiag_open,
.llseek = lpfc_debugfs_lseek,
.read = lpfc_idiag_queacc_read,
.write = lpfc_idiag_queacc_write,
.release = lpfc_idiag_cmd_release,
};
#undef lpfc_idiag_op_drbAcc
static const struct file_operations lpfc_idiag_op_drbAcc = {
.owner = THIS_MODULE,
.open = lpfc_idiag_open,
.llseek = lpfc_debugfs_lseek,
.read = lpfc_idiag_drbacc_read,
.write = lpfc_idiag_drbacc_write,
.release = lpfc_idiag_cmd_release,
};
#undef lpfc_idiag_op_ctlAcc
static const struct file_operations lpfc_idiag_op_ctlAcc = {
.owner = THIS_MODULE,
.open = lpfc_idiag_open,
.llseek = lpfc_debugfs_lseek,
.read = lpfc_idiag_ctlacc_read,
.write = lpfc_idiag_ctlacc_write,
.release = lpfc_idiag_cmd_release,
};
#undef lpfc_idiag_op_mbxAcc
static const struct file_operations lpfc_idiag_op_mbxAcc = {
.owner = THIS_MODULE,
.open = lpfc_idiag_open,
.llseek = lpfc_debugfs_lseek,
.read = lpfc_idiag_mbxacc_read,
.write = lpfc_idiag_mbxacc_write,
.release = lpfc_idiag_cmd_release,
};
#undef lpfc_idiag_op_extAcc
static const struct file_operations lpfc_idiag_op_extAcc = {
.owner = THIS_MODULE,
.open = lpfc_idiag_open,
.llseek = lpfc_debugfs_lseek,
.read = lpfc_idiag_extacc_read,
.write = lpfc_idiag_extacc_write,
.release = lpfc_idiag_cmd_release,
};
#endif
/* lpfc_idiag_mbxacc_dump_bsg_mbox - idiag debugfs dump bsg mailbox command
* @phba: Pointer to HBA context object.
* @dmabuf: Pointer to a DMA buffer descriptor.
*
* Description:
* This routine dump a bsg pass-through non-embedded mailbox command with
* external buffer.
**/
void
lpfc_idiag_mbxacc_dump_bsg_mbox(struct lpfc_hba *phba, enum nemb_type nemb_tp,
enum mbox_type mbox_tp, enum dma_type dma_tp,
enum sta_type sta_tp,
struct lpfc_dmabuf *dmabuf, uint32_t ext_buf)
{
#ifdef CONFIG_SCSI_LPFC_DEBUG_FS
uint32_t *mbx_mbox_cmd, *mbx_dump_map, *mbx_dump_cnt, *mbx_word_cnt;
char line_buf[LPFC_MBX_ACC_LBUF_SZ];
int len = 0;
uint32_t do_dump = 0;
uint32_t *pword;
uint32_t i;
if (idiag.cmd.opcode != LPFC_IDIAG_BSG_MBXACC_DP)
return;
mbx_mbox_cmd = &idiag.cmd.data[IDIAG_MBXACC_MBCMD_INDX];
mbx_dump_map = &idiag.cmd.data[IDIAG_MBXACC_DPMAP_INDX];
mbx_dump_cnt = &idiag.cmd.data[IDIAG_MBXACC_DPCNT_INDX];
mbx_word_cnt = &idiag.cmd.data[IDIAG_MBXACC_WDCNT_INDX];
if (!(*mbx_dump_map & LPFC_MBX_DMP_ALL) ||
(*mbx_dump_cnt == 0) ||
(*mbx_word_cnt == 0))
return;
if (*mbx_mbox_cmd != 0x9B)
return;
if ((mbox_tp == mbox_rd) && (dma_tp == dma_mbox)) {
if (*mbx_dump_map & LPFC_BSG_DMP_MBX_RD_MBX) {
do_dump |= LPFC_BSG_DMP_MBX_RD_MBX;
pr_err("\nRead mbox command (x%x), "
"nemb:0x%x, extbuf_cnt:%d:\n",
sta_tp, nemb_tp, ext_buf);
}
}
if ((mbox_tp == mbox_rd) && (dma_tp == dma_ebuf)) {
if (*mbx_dump_map & LPFC_BSG_DMP_MBX_RD_BUF) {
do_dump |= LPFC_BSG_DMP_MBX_RD_BUF;
pr_err("\nRead mbox buffer (x%x), "
"nemb:0x%x, extbuf_seq:%d:\n",
sta_tp, nemb_tp, ext_buf);
}
}
if ((mbox_tp == mbox_wr) && (dma_tp == dma_mbox)) {
if (*mbx_dump_map & LPFC_BSG_DMP_MBX_WR_MBX) {
do_dump |= LPFC_BSG_DMP_MBX_WR_MBX;
pr_err("\nWrite mbox command (x%x), "
"nemb:0x%x, extbuf_cnt:%d:\n",
sta_tp, nemb_tp, ext_buf);
}
}
if ((mbox_tp == mbox_wr) && (dma_tp == dma_ebuf)) {
if (*mbx_dump_map & LPFC_BSG_DMP_MBX_WR_BUF) {
do_dump |= LPFC_BSG_DMP_MBX_WR_BUF;
pr_err("\nWrite mbox buffer (x%x), "
"nemb:0x%x, extbuf_seq:%d:\n",
sta_tp, nemb_tp, ext_buf);
}
}
/* dump buffer content */
if (do_dump) {
pword = (uint32_t *)dmabuf->virt;
for (i = 0; i < *mbx_word_cnt; i++) {
if (!(i % 8)) {
if (i != 0)
pr_err("%s\n", line_buf);
len = 0;
len += snprintf(line_buf+len,
LPFC_MBX_ACC_LBUF_SZ-len,
"%03d: ", i);
}
len += snprintf(line_buf+len, LPFC_MBX_ACC_LBUF_SZ-len,
"%08x ", (uint32_t)*pword);
pword++;
}
if ((i - 1) % 8)
pr_err("%s\n", line_buf);
(*mbx_dump_cnt)--;
}
/* Clean out command structure on reaching dump count */
if (*mbx_dump_cnt == 0)
memset(&idiag, 0, sizeof(idiag));
return;
#endif
}
/* lpfc_idiag_mbxacc_dump_issue_mbox - idiag debugfs dump issue mailbox command
* @phba: Pointer to HBA context object.
* @dmabuf: Pointer to a DMA buffer descriptor.
*
* Description:
* This routine dump a pass-through non-embedded mailbox command from issue
* mailbox command.
**/
void
lpfc_idiag_mbxacc_dump_issue_mbox(struct lpfc_hba *phba, MAILBOX_t *pmbox)
{
#ifdef CONFIG_SCSI_LPFC_DEBUG_FS
uint32_t *mbx_dump_map, *mbx_dump_cnt, *mbx_word_cnt, *mbx_mbox_cmd;
char line_buf[LPFC_MBX_ACC_LBUF_SZ];
int len = 0;
uint32_t *pword;
uint8_t *pbyte;
uint32_t i, j;
if (idiag.cmd.opcode != LPFC_IDIAG_CMD_MBXACC_DP)
return;
mbx_mbox_cmd = &idiag.cmd.data[IDIAG_MBXACC_MBCMD_INDX];
mbx_dump_map = &idiag.cmd.data[IDIAG_MBXACC_DPMAP_INDX];
mbx_dump_cnt = &idiag.cmd.data[IDIAG_MBXACC_DPCNT_INDX];
mbx_word_cnt = &idiag.cmd.data[IDIAG_MBXACC_WDCNT_INDX];
if (!(*mbx_dump_map & LPFC_MBX_DMP_MBX_ALL) ||
(*mbx_dump_cnt == 0) ||
(*mbx_word_cnt == 0))
return;
if ((*mbx_mbox_cmd != LPFC_MBX_ALL_CMD) &&
(*mbx_mbox_cmd != pmbox->mbxCommand))
return;
/* dump buffer content */
if (*mbx_dump_map & LPFC_MBX_DMP_MBX_WORD) {
pr_err("Mailbox command:0x%x dump by word:\n",
pmbox->mbxCommand);
pword = (uint32_t *)pmbox;
for (i = 0; i < *mbx_word_cnt; i++) {
if (!(i % 8)) {
if (i != 0)
pr_err("%s\n", line_buf);
len = 0;
memset(line_buf, 0, LPFC_MBX_ACC_LBUF_SZ);
len += snprintf(line_buf+len,
LPFC_MBX_ACC_LBUF_SZ-len,
"%03d: ", i);
}
len += snprintf(line_buf+len, LPFC_MBX_ACC_LBUF_SZ-len,
"%08x ",
((uint32_t)*pword) & 0xffffffff);
pword++;
}
if ((i - 1) % 8)
pr_err("%s\n", line_buf);
pr_err("\n");
}
if (*mbx_dump_map & LPFC_MBX_DMP_MBX_BYTE) {
pr_err("Mailbox command:0x%x dump by byte:\n",
pmbox->mbxCommand);
pbyte = (uint8_t *)pmbox;
for (i = 0; i < *mbx_word_cnt; i++) {
if (!(i % 8)) {
if (i != 0)
pr_err("%s\n", line_buf);
len = 0;
memset(line_buf, 0, LPFC_MBX_ACC_LBUF_SZ);
len += snprintf(line_buf+len,
LPFC_MBX_ACC_LBUF_SZ-len,
"%03d: ", i);
}
for (j = 0; j < 4; j++) {
len += snprintf(line_buf+len,
LPFC_MBX_ACC_LBUF_SZ-len,
"%02x",
((uint8_t)*pbyte) & 0xff);
pbyte++;
}
len += snprintf(line_buf+len,
LPFC_MBX_ACC_LBUF_SZ-len, " ");
}
if ((i - 1) % 8)
pr_err("%s\n", line_buf);
pr_err("\n");
}
(*mbx_dump_cnt)--;
/* Clean out command structure on reaching dump count */
if (*mbx_dump_cnt == 0)
memset(&idiag, 0, sizeof(idiag));
return;
#endif
}
/**
* lpfc_debugfs_initialize - Initialize debugfs for a vport
* @vport: The vport pointer to initialize.
*
* Description:
* When Debugfs is configured this routine sets up the lpfc debugfs file system.
* If not already created, this routine will create the lpfc directory, and
* lpfcX directory (for this HBA), and vportX directory for this vport. It will
* also create each file used to access lpfc specific debugfs information.
**/
inline void
lpfc_debugfs_initialize(struct lpfc_vport *vport)
{
#ifdef CONFIG_SCSI_LPFC_DEBUG_FS
struct lpfc_hba *phba = vport->phba;
char name[64];
uint32_t num, i;
bool pport_setup = false;
if (!lpfc_debugfs_enable)
return;
/* Setup lpfc root directory */
if (!lpfc_debugfs_root) {
lpfc_debugfs_root = debugfs_create_dir("lpfc", NULL);
atomic_set(&lpfc_debugfs_hba_count, 0);
}
if (!lpfc_debugfs_start_time)
lpfc_debugfs_start_time = jiffies;
/* Setup funcX directory for specific HBA PCI function */
snprintf(name, sizeof(name), "fn%d", phba->brd_no);
if (!phba->hba_debugfs_root) {
pport_setup = true;
phba->hba_debugfs_root =
debugfs_create_dir(name, lpfc_debugfs_root);
atomic_inc(&lpfc_debugfs_hba_count);
atomic_set(&phba->debugfs_vport_count, 0);
scsi: lpfc: Adapt partitioned XRI lists to efficient sharing The XRI get/put lists were partitioned per hardware queue. However, the adapter rarely had sufficient resources to give a large number of resources per queue. As such, it became common for a cpu to encounter a lack of XRI resource and request the upper io stack to retry after returning a BUSY condition. This occurred even though other cpus were idle and not using their resources. Create as efficient a scheme as possible to move resources to the cpus that need them. Each cpu maintains a small private pool which it allocates from for io. There is a watermark that the cpu attempts to keep in the private pool. The private pool, when empty, pulls from a global pool from the cpu. When the cpu's global pool is empty it will pull from other cpu's global pool. As there many cpu global pools (1 per cpu or hardware queue count) and as each cpu selects what cpu to pull from at different rates and at different times, it creates a radomizing effect that minimizes the number of cpu's that will contend with each other when the steal XRI's from another cpu's global pool. On io completion, a cpu will push the XRI back on to its private pool. A watermark level is maintained for the private pool such that when it is exceeded it will move XRI's to the CPU global pool so that other cpu's may allocate them. On NVME, as heartbeat commands are critical to get placed on the wire, a single expedite pool is maintained. When a heartbeat is to be sent, it will allocate an XRI from the expedite pool rather than the normal cpu private/global pools. On any io completion, if a reduction in the expedite pools is seen, it will be replenished before the XRI is placed on the cpu private pool. Statistics are added to aid understanding the XRI levels on each cpu and their behaviors. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <jsmart2021@gmail.com> Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2019-01-29 03:14:28 +08:00
/* Multi-XRI pools */
snprintf(name, sizeof(name), "multixripools");
phba->debug_multixri_pools =
debugfs_create_file(name, S_IFREG | 0644,
phba->hba_debugfs_root,
phba,
&lpfc_debugfs_op_multixripools);
if (!phba->debug_multixri_pools) {
lpfc_printf_vlog(vport, KERN_ERR, LOG_INIT,
"0527 Cannot create debugfs multixripools\n");
goto debug_failed;
}
/* Setup hbqinfo */
snprintf(name, sizeof(name), "hbqinfo");
phba->debug_hbqinfo =
debugfs_create_file(name, S_IFREG | 0644,
phba->hba_debugfs_root,
phba, &lpfc_debugfs_op_hbqinfo);
#ifdef LPFC_HDWQ_LOCK_STAT
/* Setup lockstat */
snprintf(name, sizeof(name), "lockstat");
phba->debug_lockstat =
debugfs_create_file(name, S_IFREG | 0644,
phba->hba_debugfs_root,
phba, &lpfc_debugfs_op_lockstat);
if (!phba->debug_lockstat) {
lpfc_printf_vlog(vport, KERN_ERR, LOG_INIT,
"0913 Cant create debugfs lockstat\n");
goto debug_failed;
}
#endif
/* Setup dumpHBASlim */
if (phba->sli_rev < LPFC_SLI_REV4) {
snprintf(name, sizeof(name), "dumpHBASlim");
phba->debug_dumpHBASlim =
debugfs_create_file(name,
S_IFREG|S_IRUGO|S_IWUSR,
phba->hba_debugfs_root,
phba, &lpfc_debugfs_op_dumpHBASlim);
} else
phba->debug_dumpHBASlim = NULL;
/* Setup dumpHostSlim */
if (phba->sli_rev < LPFC_SLI_REV4) {
snprintf(name, sizeof(name), "dumpHostSlim");
phba->debug_dumpHostSlim =
debugfs_create_file(name,
S_IFREG|S_IRUGO|S_IWUSR,
phba->hba_debugfs_root,
phba, &lpfc_debugfs_op_dumpHostSlim);
} else
phba->debug_dumpHostSlim = NULL;
/* Setup dumpData */
snprintf(name, sizeof(name), "dumpData");
phba->debug_dumpData =
debugfs_create_file(name, S_IFREG|S_IRUGO|S_IWUSR,
phba->hba_debugfs_root,
phba, &lpfc_debugfs_op_dumpData);
/* Setup dumpDif */
snprintf(name, sizeof(name), "dumpDif");
phba->debug_dumpDif =
debugfs_create_file(name, S_IFREG|S_IRUGO|S_IWUSR,
phba->hba_debugfs_root,
phba, &lpfc_debugfs_op_dumpDif);
/* Setup DIF Error Injections */
snprintf(name, sizeof(name), "InjErrLBA");
phba->debug_InjErrLBA =
debugfs_create_file(name, S_IFREG|S_IRUGO|S_IWUSR,
phba->hba_debugfs_root,
phba, &lpfc_debugfs_op_dif_err);
phba->lpfc_injerr_lba = LPFC_INJERR_LBA_OFF;
snprintf(name, sizeof(name), "InjErrNPortID");
phba->debug_InjErrNPortID =
debugfs_create_file(name, S_IFREG|S_IRUGO|S_IWUSR,
phba->hba_debugfs_root,
phba, &lpfc_debugfs_op_dif_err);
snprintf(name, sizeof(name), "InjErrWWPN");
phba->debug_InjErrWWPN =
debugfs_create_file(name, S_IFREG|S_IRUGO|S_IWUSR,
phba->hba_debugfs_root,
phba, &lpfc_debugfs_op_dif_err);
snprintf(name, sizeof(name), "writeGuardInjErr");
phba->debug_writeGuard =
debugfs_create_file(name, S_IFREG|S_IRUGO|S_IWUSR,
phba->hba_debugfs_root,
phba, &lpfc_debugfs_op_dif_err);
snprintf(name, sizeof(name), "writeAppInjErr");
phba->debug_writeApp =
debugfs_create_file(name, S_IFREG|S_IRUGO|S_IWUSR,
phba->hba_debugfs_root,
phba, &lpfc_debugfs_op_dif_err);
snprintf(name, sizeof(name), "writeRefInjErr");
phba->debug_writeRef =
debugfs_create_file(name, S_IFREG|S_IRUGO|S_IWUSR,
phba->hba_debugfs_root,
phba, &lpfc_debugfs_op_dif_err);
snprintf(name, sizeof(name), "readGuardInjErr");
phba->debug_readGuard =
debugfs_create_file(name, S_IFREG|S_IRUGO|S_IWUSR,
phba->hba_debugfs_root,
phba, &lpfc_debugfs_op_dif_err);
snprintf(name, sizeof(name), "readAppInjErr");
phba->debug_readApp =
debugfs_create_file(name, S_IFREG|S_IRUGO|S_IWUSR,
phba->hba_debugfs_root,
phba, &lpfc_debugfs_op_dif_err);
snprintf(name, sizeof(name), "readRefInjErr");
phba->debug_readRef =
debugfs_create_file(name, S_IFREG|S_IRUGO|S_IWUSR,
phba->hba_debugfs_root,
phba, &lpfc_debugfs_op_dif_err);
/* Setup slow ring trace */
if (lpfc_debugfs_max_slow_ring_trc) {
num = lpfc_debugfs_max_slow_ring_trc - 1;
if (num & lpfc_debugfs_max_slow_ring_trc) {
/* Change to be a power of 2 */
num = lpfc_debugfs_max_slow_ring_trc;
i = 0;
while (num > 1) {
num = num >> 1;
i++;
}
lpfc_debugfs_max_slow_ring_trc = (1 << i);
pr_err("lpfc_debugfs_max_disc_trc changed to "
"%d\n", lpfc_debugfs_max_disc_trc);
}
}
snprintf(name, sizeof(name), "slow_ring_trace");
phba->debug_slow_ring_trc =
debugfs_create_file(name, S_IFREG|S_IRUGO|S_IWUSR,
phba->hba_debugfs_root,
phba, &lpfc_debugfs_op_slow_ring_trc);
if (!phba->slow_ring_trc) {
phba->slow_ring_trc = kmalloc(
(sizeof(struct lpfc_debugfs_trc) *
lpfc_debugfs_max_slow_ring_trc),
GFP_KERNEL);
if (!phba->slow_ring_trc) {
lpfc_printf_vlog(vport, KERN_ERR, LOG_INIT,
"0416 Cannot create debugfs "
"slow_ring buffer\n");
goto debug_failed;
}
atomic_set(&phba->slow_ring_trc_cnt, 0);
memset(phba->slow_ring_trc, 0,
(sizeof(struct lpfc_debugfs_trc) *
lpfc_debugfs_max_slow_ring_trc));
}
snprintf(name, sizeof(name), "nvmeio_trc");
phba->debug_nvmeio_trc =
debugfs_create_file(name, 0644,
phba->hba_debugfs_root,
phba, &lpfc_debugfs_op_nvmeio_trc);
atomic_set(&phba->nvmeio_trc_cnt, 0);
if (lpfc_debugfs_max_nvmeio_trc) {
num = lpfc_debugfs_max_nvmeio_trc - 1;
if (num & lpfc_debugfs_max_disc_trc) {
/* Change to be a power of 2 */
num = lpfc_debugfs_max_nvmeio_trc;
i = 0;
while (num > 1) {
num = num >> 1;
i++;
}
lpfc_debugfs_max_nvmeio_trc = (1 << i);
lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
"0575 lpfc_debugfs_max_nvmeio_trc "
"changed to %d\n",
lpfc_debugfs_max_nvmeio_trc);
}
phba->nvmeio_trc_size = lpfc_debugfs_max_nvmeio_trc;
/* Allocate trace buffer and initialize */
phba->nvmeio_trc = kzalloc(
(sizeof(struct lpfc_debugfs_nvmeio_trc) *
phba->nvmeio_trc_size), GFP_KERNEL);
if (!phba->nvmeio_trc) {
lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
"0576 Cannot create debugfs "
"nvmeio_trc buffer\n");
goto nvmeio_off;
}
phba->nvmeio_trc_on = 1;
phba->nvmeio_trc_output_idx = 0;
phba->nvmeio_trc = NULL;
} else {
nvmeio_off:
phba->nvmeio_trc_size = 0;
phba->nvmeio_trc_on = 0;
phba->nvmeio_trc_output_idx = 0;
phba->nvmeio_trc = NULL;
}
}
snprintf(name, sizeof(name), "vport%d", vport->vpi);
if (!vport->vport_debugfs_root) {
vport->vport_debugfs_root =
debugfs_create_dir(name, phba->hba_debugfs_root);
atomic_inc(&phba->debugfs_vport_count);
}
if (lpfc_debugfs_max_disc_trc) {
num = lpfc_debugfs_max_disc_trc - 1;
if (num & lpfc_debugfs_max_disc_trc) {
/* Change to be a power of 2 */
num = lpfc_debugfs_max_disc_trc;
i = 0;
while (num > 1) {
num = num >> 1;
i++;
}
lpfc_debugfs_max_disc_trc = (1 << i);
pr_err("lpfc_debugfs_max_disc_trc changed to %d\n",
lpfc_debugfs_max_disc_trc);
}
}
vport->disc_trc = kzalloc(
(sizeof(struct lpfc_debugfs_trc) * lpfc_debugfs_max_disc_trc),
GFP_KERNEL);
if (!vport->disc_trc) {
lpfc_printf_vlog(vport, KERN_ERR, LOG_INIT,
"0418 Cannot create debugfs disc trace "
"buffer\n");
goto debug_failed;
}
atomic_set(&vport->disc_trc_cnt, 0);
snprintf(name, sizeof(name), "discovery_trace");
vport->debug_disc_trc =
debugfs_create_file(name, S_IFREG|S_IRUGO|S_IWUSR,
vport->vport_debugfs_root,
vport, &lpfc_debugfs_op_disc_trc);
snprintf(name, sizeof(name), "nodelist");
vport->debug_nodelist =
debugfs_create_file(name, S_IFREG|S_IRUGO|S_IWUSR,
vport->vport_debugfs_root,
vport, &lpfc_debugfs_op_nodelist);
snprintf(name, sizeof(name), "nvmestat");
vport->debug_nvmestat =
debugfs_create_file(name, 0644,
vport->vport_debugfs_root,
vport, &lpfc_debugfs_op_nvmestat);
snprintf(name, sizeof(name), "scsistat");
vport->debug_scsistat =
debugfs_create_file(name, 0644,
vport->vport_debugfs_root,
vport, &lpfc_debugfs_op_scsistat);
if (!vport->debug_scsistat) {
lpfc_printf_vlog(vport, KERN_ERR, LOG_INIT,
"0914 Cannot create debugfs scsistat\n");
goto debug_failed;
}
snprintf(name, sizeof(name), "nvmektime");
vport->debug_nvmektime =
debugfs_create_file(name, 0644,
vport->vport_debugfs_root,
vport, &lpfc_debugfs_op_nvmektime);
snprintf(name, sizeof(name), "cpucheck");
vport->debug_cpucheck =
debugfs_create_file(name, 0644,
vport->vport_debugfs_root,
vport, &lpfc_debugfs_op_cpucheck);
/*
* The following section is for additional directories/files for the
* physical port.
*/
if (!pport_setup)
goto debug_failed;
/*
* iDiag debugfs root entry points for SLI4 device only
*/
if (phba->sli_rev < LPFC_SLI_REV4)
goto debug_failed;
snprintf(name, sizeof(name), "iDiag");
if (!phba->idiag_root) {
phba->idiag_root =
debugfs_create_dir(name, phba->hba_debugfs_root);
/* Initialize iDiag data structure */
memset(&idiag, 0, sizeof(idiag));
}
/* iDiag read PCI config space */
snprintf(name, sizeof(name), "pciCfg");
if (!phba->idiag_pci_cfg) {
phba->idiag_pci_cfg =
debugfs_create_file(name, S_IFREG|S_IRUGO|S_IWUSR,
phba->idiag_root, phba, &lpfc_idiag_op_pciCfg);
idiag.offset.last_rd = 0;
}
/* iDiag PCI BAR access */
snprintf(name, sizeof(name), "barAcc");
if (!phba->idiag_bar_acc) {
phba->idiag_bar_acc =
debugfs_create_file(name, S_IFREG|S_IRUGO|S_IWUSR,
phba->idiag_root, phba, &lpfc_idiag_op_barAcc);
idiag.offset.last_rd = 0;
}
/* iDiag get PCI function queue information */
snprintf(name, sizeof(name), "queInfo");
if (!phba->idiag_que_info) {
phba->idiag_que_info =
debugfs_create_file(name, S_IFREG|S_IRUGO,
phba->idiag_root, phba, &lpfc_idiag_op_queInfo);
}
/* iDiag access PCI function queue */
snprintf(name, sizeof(name), "queAcc");
if (!phba->idiag_que_acc) {
phba->idiag_que_acc =
debugfs_create_file(name, S_IFREG|S_IRUGO|S_IWUSR,
phba->idiag_root, phba, &lpfc_idiag_op_queAcc);
}
/* iDiag access PCI function doorbell registers */
snprintf(name, sizeof(name), "drbAcc");
if (!phba->idiag_drb_acc) {
phba->idiag_drb_acc =
debugfs_create_file(name, S_IFREG|S_IRUGO|S_IWUSR,
phba->idiag_root, phba, &lpfc_idiag_op_drbAcc);
}
/* iDiag access PCI function control registers */
snprintf(name, sizeof(name), "ctlAcc");
if (!phba->idiag_ctl_acc) {
phba->idiag_ctl_acc =
debugfs_create_file(name, S_IFREG|S_IRUGO|S_IWUSR,
phba->idiag_root, phba, &lpfc_idiag_op_ctlAcc);
}
/* iDiag access mbox commands */
snprintf(name, sizeof(name), "mbxAcc");
if (!phba->idiag_mbx_acc) {
phba->idiag_mbx_acc =
debugfs_create_file(name, S_IFREG|S_IRUGO|S_IWUSR,
phba->idiag_root, phba, &lpfc_idiag_op_mbxAcc);
}
/* iDiag extents access commands */
if (phba->sli4_hba.extents_in_use) {
snprintf(name, sizeof(name), "extAcc");
if (!phba->idiag_ext_acc) {
phba->idiag_ext_acc =
debugfs_create_file(name,
S_IFREG|S_IRUGO|S_IWUSR,
phba->idiag_root, phba,
&lpfc_idiag_op_extAcc);
}
}
debug_failed:
return;
#endif
}
/**
* lpfc_debugfs_terminate - Tear down debugfs infrastructure for this vport
* @vport: The vport pointer to remove from debugfs.
*
* Description:
* When Debugfs is configured this routine removes debugfs file system elements
* that are specific to this vport. It also checks to see if there are any
* users left for the debugfs directories associated with the HBA and driver. If
* this is the last user of the HBA directory or driver directory then it will
* remove those from the debugfs infrastructure as well.
**/
inline void
lpfc_debugfs_terminate(struct lpfc_vport *vport)
{
#ifdef CONFIG_SCSI_LPFC_DEBUG_FS
struct lpfc_hba *phba = vport->phba;
kfree(vport->disc_trc);
vport->disc_trc = NULL;
debugfs_remove(vport->debug_disc_trc); /* discovery_trace */
vport->debug_disc_trc = NULL;
debugfs_remove(vport->debug_nodelist); /* nodelist */
vport->debug_nodelist = NULL;
debugfs_remove(vport->debug_nvmestat); /* nvmestat */
vport->debug_nvmestat = NULL;
debugfs_remove(vport->debug_scsistat); /* scsistat */
vport->debug_scsistat = NULL;
debugfs_remove(vport->debug_nvmektime); /* nvmektime */
vport->debug_nvmektime = NULL;
debugfs_remove(vport->debug_cpucheck); /* cpucheck */
vport->debug_cpucheck = NULL;
if (vport->vport_debugfs_root) {
debugfs_remove(vport->vport_debugfs_root); /* vportX */
vport->vport_debugfs_root = NULL;
atomic_dec(&phba->debugfs_vport_count);
}
if (atomic_read(&phba->debugfs_vport_count) == 0) {
scsi: lpfc: Adapt partitioned XRI lists to efficient sharing The XRI get/put lists were partitioned per hardware queue. However, the adapter rarely had sufficient resources to give a large number of resources per queue. As such, it became common for a cpu to encounter a lack of XRI resource and request the upper io stack to retry after returning a BUSY condition. This occurred even though other cpus were idle and not using their resources. Create as efficient a scheme as possible to move resources to the cpus that need them. Each cpu maintains a small private pool which it allocates from for io. There is a watermark that the cpu attempts to keep in the private pool. The private pool, when empty, pulls from a global pool from the cpu. When the cpu's global pool is empty it will pull from other cpu's global pool. As there many cpu global pools (1 per cpu or hardware queue count) and as each cpu selects what cpu to pull from at different rates and at different times, it creates a radomizing effect that minimizes the number of cpu's that will contend with each other when the steal XRI's from another cpu's global pool. On io completion, a cpu will push the XRI back on to its private pool. A watermark level is maintained for the private pool such that when it is exceeded it will move XRI's to the CPU global pool so that other cpu's may allocate them. On NVME, as heartbeat commands are critical to get placed on the wire, a single expedite pool is maintained. When a heartbeat is to be sent, it will allocate an XRI from the expedite pool rather than the normal cpu private/global pools. On any io completion, if a reduction in the expedite pools is seen, it will be replenished before the XRI is placed on the cpu private pool. Statistics are added to aid understanding the XRI levels on each cpu and their behaviors. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <jsmart2021@gmail.com> Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2019-01-29 03:14:28 +08:00
debugfs_remove(phba->debug_multixri_pools); /* multixripools*/
phba->debug_multixri_pools = NULL;
debugfs_remove(phba->debug_hbqinfo); /* hbqinfo */
phba->debug_hbqinfo = NULL;
#ifdef LPFC_HDWQ_LOCK_STAT
debugfs_remove(phba->debug_lockstat); /* lockstat */
phba->debug_lockstat = NULL;
#endif
debugfs_remove(phba->debug_dumpHBASlim); /* HBASlim */
phba->debug_dumpHBASlim = NULL;
debugfs_remove(phba->debug_dumpHostSlim); /* HostSlim */
phba->debug_dumpHostSlim = NULL;
debugfs_remove(phba->debug_dumpData); /* dumpData */
phba->debug_dumpData = NULL;
debugfs_remove(phba->debug_dumpDif); /* dumpDif */
phba->debug_dumpDif = NULL;
debugfs_remove(phba->debug_InjErrLBA); /* InjErrLBA */
phba->debug_InjErrLBA = NULL;
debugfs_remove(phba->debug_InjErrNPortID);
phba->debug_InjErrNPortID = NULL;
debugfs_remove(phba->debug_InjErrWWPN); /* InjErrWWPN */
phba->debug_InjErrWWPN = NULL;
debugfs_remove(phba->debug_writeGuard); /* writeGuard */
phba->debug_writeGuard = NULL;
debugfs_remove(phba->debug_writeApp); /* writeApp */
phba->debug_writeApp = NULL;
debugfs_remove(phba->debug_writeRef); /* writeRef */
phba->debug_writeRef = NULL;
debugfs_remove(phba->debug_readGuard); /* readGuard */
phba->debug_readGuard = NULL;
debugfs_remove(phba->debug_readApp); /* readApp */
phba->debug_readApp = NULL;
debugfs_remove(phba->debug_readRef); /* readRef */
phba->debug_readRef = NULL;
kfree(phba->slow_ring_trc);
phba->slow_ring_trc = NULL;
/* slow_ring_trace */
debugfs_remove(phba->debug_slow_ring_trc);
phba->debug_slow_ring_trc = NULL;
debugfs_remove(phba->debug_nvmeio_trc);
phba->debug_nvmeio_trc = NULL;
kfree(phba->nvmeio_trc);
phba->nvmeio_trc = NULL;
/*
* iDiag release
*/
if (phba->sli_rev == LPFC_SLI_REV4) {
/* iDiag extAcc */
debugfs_remove(phba->idiag_ext_acc);
phba->idiag_ext_acc = NULL;
/* iDiag mbxAcc */
debugfs_remove(phba->idiag_mbx_acc);
phba->idiag_mbx_acc = NULL;
/* iDiag ctlAcc */
debugfs_remove(phba->idiag_ctl_acc);
phba->idiag_ctl_acc = NULL;
/* iDiag drbAcc */
debugfs_remove(phba->idiag_drb_acc);
phba->idiag_drb_acc = NULL;
/* iDiag queAcc */
debugfs_remove(phba->idiag_que_acc);
phba->idiag_que_acc = NULL;
/* iDiag queInfo */
debugfs_remove(phba->idiag_que_info);
phba->idiag_que_info = NULL;
/* iDiag barAcc */
debugfs_remove(phba->idiag_bar_acc);
phba->idiag_bar_acc = NULL;
/* iDiag pciCfg */
debugfs_remove(phba->idiag_pci_cfg);
phba->idiag_pci_cfg = NULL;
/* Finally remove the iDiag debugfs root */
debugfs_remove(phba->idiag_root);
phba->idiag_root = NULL;
}
if (phba->hba_debugfs_root) {
debugfs_remove(phba->hba_debugfs_root); /* fnX */
phba->hba_debugfs_root = NULL;
atomic_dec(&lpfc_debugfs_hba_count);
}
if (atomic_read(&lpfc_debugfs_hba_count) == 0) {
debugfs_remove(lpfc_debugfs_root); /* lpfc */
lpfc_debugfs_root = NULL;
}
}
#endif
return;
}
/*
* Driver debug utility routines outside of debugfs. The debug utility
* routines implemented here is intended to be used in the instrumented
* debug driver for debugging host or port issues.
*/
/**
* lpfc_debug_dump_all_queues - dump all the queues with a hba
* @phba: Pointer to HBA context object.
*
* This function dumps entries of all the queues asociated with the @phba.
**/
void
lpfc_debug_dump_all_queues(struct lpfc_hba *phba)
{
int idx;
/*
* Dump Work Queues (WQs)
*/
lpfc_debug_dump_wq(phba, DUMP_MBX, 0);
lpfc_debug_dump_wq(phba, DUMP_ELS, 0);
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-13 05:52:30 +08:00
lpfc_debug_dump_wq(phba, DUMP_NVMELS, 0);
2019-01-29 03:14:21 +08:00
for (idx = 0; idx < phba->cfg_hdw_queue; idx++)
lpfc_debug_dump_wq(phba, DUMP_FCP, idx);
2019-01-29 03:14:21 +08:00
if (phba->cfg_enable_fc4_type & LPFC_ENABLE_NVME) {
for (idx = 0; idx < phba->cfg_hdw_queue; idx++)
lpfc_debug_dump_wq(phba, DUMP_NVME, idx);
}
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-13 05:52:30 +08:00
lpfc_debug_dump_hdr_rq(phba);
lpfc_debug_dump_dat_rq(phba);
/*
* Dump Complete Queues (CQs)
*/
lpfc_debug_dump_cq(phba, DUMP_MBX, 0);
lpfc_debug_dump_cq(phba, DUMP_ELS, 0);
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-13 05:52:30 +08:00
lpfc_debug_dump_cq(phba, DUMP_NVMELS, 0);
2019-01-29 03:14:21 +08:00
for (idx = 0; idx < phba->cfg_hdw_queue; idx++)
lpfc_debug_dump_cq(phba, DUMP_FCP, idx);
2019-01-29 03:14:21 +08:00
if (phba->cfg_enable_fc4_type & LPFC_ENABLE_NVME) {
for (idx = 0; idx < phba->cfg_hdw_queue; idx++)
lpfc_debug_dump_cq(phba, DUMP_NVME, idx);
}
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-13 05:52:30 +08:00
/*
* Dump Event Queues (EQs)
*/
2019-01-29 03:14:21 +08:00
for (idx = 0; idx < phba->cfg_hdw_queue; idx++)
lpfc_debug_dump_hba_eq(phba, idx);
}