2015-06-01 03:02:11 +08:00
|
|
|
/*
|
|
|
|
* Copyright(c) 2013-2015 Intel Corporation. All rights reserved.
|
|
|
|
*
|
|
|
|
* 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, but
|
|
|
|
* WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
|
|
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
|
|
|
* General Public License for more details.
|
|
|
|
*/
|
nd_btt: atomic sector updates
BTT stands for Block Translation Table, and is a way to provide power
fail sector atomicity semantics for block devices that have the ability
to perform byte granularity IO. It relies on the capability of libnvdimm
namespace devices to do byte aligned IO.
The BTT works as a stacked blocked device, and reserves a chunk of space
from the backing device for its accounting metadata. It is a bio-based
driver because all IO is done synchronously, and there is no queuing or
asynchronous completions at either the device or the driver level.
The BTT uses 'lanes' to index into various 'on-disk' data structures,
and lanes also act as a synchronization mechanism in case there are more
CPUs than available lanes. We did a comparison between two lane lock
strategies - first where we kept an atomic counter around that tracked
which was the last lane that was used, and 'our' lane was determined by
atomically incrementing that. That way, for the nr_cpus > nr_lanes case,
theoretically, no CPU would be blocked waiting for a lane. The other
strategy was to use the cpu number we're scheduled on to and hash it to
a lane number. Theoretically, this could block an IO that could've
otherwise run using a different, free lane. But some fio workloads
showed that the direct cpu -> lane hash performed faster than tracking
'last lane' - my reasoning is the cache thrash caused by moving the
atomic variable made that approach slower than simply waiting out the
in-progress IO. This supports the conclusion that the driver can be a
very simple bio-based one that does synchronous IOs instead of queuing.
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Boaz Harrosh <boaz@plexistor.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Jens Axboe <axboe@fb.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Neil Brown <neilb@suse.de>
Cc: Jeff Moyer <jmoyer@redhat.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Greg KH <gregkh@linuxfoundation.org>
[jmoyer: fix nmi watchdog timeout in btt_map_init]
[jmoyer: move btt initialization to module load path]
[jmoyer: fix memory leak in the btt initialization path]
[jmoyer: Don't overwrite corrupted arenas]
Signed-off-by: Vishal Verma <vishal.l.verma@linux.intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2015-06-25 16:20:32 +08:00
|
|
|
#include <linux/cpumask.h>
|
2015-06-01 03:02:11 +08:00
|
|
|
#include <linux/module.h>
|
|
|
|
#include <linux/device.h>
|
|
|
|
#include <linux/nd.h>
|
|
|
|
#include "nd.h"
|
|
|
|
|
|
|
|
static int nd_region_probe(struct device *dev)
|
|
|
|
{
|
2015-06-25 16:21:02 +08:00
|
|
|
int err, rc;
|
nd_btt: atomic sector updates
BTT stands for Block Translation Table, and is a way to provide power
fail sector atomicity semantics for block devices that have the ability
to perform byte granularity IO. It relies on the capability of libnvdimm
namespace devices to do byte aligned IO.
The BTT works as a stacked blocked device, and reserves a chunk of space
from the backing device for its accounting metadata. It is a bio-based
driver because all IO is done synchronously, and there is no queuing or
asynchronous completions at either the device or the driver level.
The BTT uses 'lanes' to index into various 'on-disk' data structures,
and lanes also act as a synchronization mechanism in case there are more
CPUs than available lanes. We did a comparison between two lane lock
strategies - first where we kept an atomic counter around that tracked
which was the last lane that was used, and 'our' lane was determined by
atomically incrementing that. That way, for the nr_cpus > nr_lanes case,
theoretically, no CPU would be blocked waiting for a lane. The other
strategy was to use the cpu number we're scheduled on to and hash it to
a lane number. Theoretically, this could block an IO that could've
otherwise run using a different, free lane. But some fio workloads
showed that the direct cpu -> lane hash performed faster than tracking
'last lane' - my reasoning is the cache thrash caused by moving the
atomic variable made that approach slower than simply waiting out the
in-progress IO. This supports the conclusion that the driver can be a
very simple bio-based one that does synchronous IOs instead of queuing.
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Boaz Harrosh <boaz@plexistor.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Jens Axboe <axboe@fb.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Neil Brown <neilb@suse.de>
Cc: Jeff Moyer <jmoyer@redhat.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Greg KH <gregkh@linuxfoundation.org>
[jmoyer: fix nmi watchdog timeout in btt_map_init]
[jmoyer: move btt initialization to module load path]
[jmoyer: fix memory leak in the btt initialization path]
[jmoyer: Don't overwrite corrupted arenas]
Signed-off-by: Vishal Verma <vishal.l.verma@linux.intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2015-06-25 16:20:32 +08:00
|
|
|
static unsigned long once;
|
2015-06-01 03:02:11 +08:00
|
|
|
struct nd_region_namespaces *num_ns;
|
|
|
|
struct nd_region *nd_region = to_nd_region(dev);
|
|
|
|
|
nd_btt: atomic sector updates
BTT stands for Block Translation Table, and is a way to provide power
fail sector atomicity semantics for block devices that have the ability
to perform byte granularity IO. It relies on the capability of libnvdimm
namespace devices to do byte aligned IO.
The BTT works as a stacked blocked device, and reserves a chunk of space
from the backing device for its accounting metadata. It is a bio-based
driver because all IO is done synchronously, and there is no queuing or
asynchronous completions at either the device or the driver level.
The BTT uses 'lanes' to index into various 'on-disk' data structures,
and lanes also act as a synchronization mechanism in case there are more
CPUs than available lanes. We did a comparison between two lane lock
strategies - first where we kept an atomic counter around that tracked
which was the last lane that was used, and 'our' lane was determined by
atomically incrementing that. That way, for the nr_cpus > nr_lanes case,
theoretically, no CPU would be blocked waiting for a lane. The other
strategy was to use the cpu number we're scheduled on to and hash it to
a lane number. Theoretically, this could block an IO that could've
otherwise run using a different, free lane. But some fio workloads
showed that the direct cpu -> lane hash performed faster than tracking
'last lane' - my reasoning is the cache thrash caused by moving the
atomic variable made that approach slower than simply waiting out the
in-progress IO. This supports the conclusion that the driver can be a
very simple bio-based one that does synchronous IOs instead of queuing.
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Boaz Harrosh <boaz@plexistor.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Jens Axboe <axboe@fb.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Neil Brown <neilb@suse.de>
Cc: Jeff Moyer <jmoyer@redhat.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Greg KH <gregkh@linuxfoundation.org>
[jmoyer: fix nmi watchdog timeout in btt_map_init]
[jmoyer: move btt initialization to module load path]
[jmoyer: fix memory leak in the btt initialization path]
[jmoyer: Don't overwrite corrupted arenas]
Signed-off-by: Vishal Verma <vishal.l.verma@linux.intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2015-06-25 16:20:32 +08:00
|
|
|
if (nd_region->num_lanes > num_online_cpus()
|
|
|
|
&& nd_region->num_lanes < num_possible_cpus()
|
|
|
|
&& !test_and_set_bit(0, &once)) {
|
|
|
|
dev_info(dev, "online cpus (%d) < concurrent i/o lanes (%d) < possible cpus (%d)\n",
|
|
|
|
num_online_cpus(), nd_region->num_lanes,
|
|
|
|
num_possible_cpus());
|
|
|
|
dev_info(dev, "setting nr_cpus=%d may yield better libnvdimm device performance\n",
|
|
|
|
nd_region->num_lanes);
|
|
|
|
}
|
|
|
|
|
2015-06-25 16:21:02 +08:00
|
|
|
rc = nd_blk_region_init(nd_region);
|
|
|
|
if (rc)
|
|
|
|
return rc;
|
|
|
|
|
|
|
|
rc = nd_region_register_namespaces(nd_region, &err);
|
2015-06-01 03:02:11 +08:00
|
|
|
num_ns = devm_kzalloc(dev, sizeof(*num_ns), GFP_KERNEL);
|
|
|
|
if (!num_ns)
|
|
|
|
return -ENOMEM;
|
|
|
|
|
|
|
|
if (rc < 0)
|
|
|
|
return rc;
|
|
|
|
|
|
|
|
num_ns->active = rc;
|
|
|
|
num_ns->count = rc + err;
|
|
|
|
dev_set_drvdata(dev, num_ns);
|
|
|
|
|
2015-06-25 16:20:04 +08:00
|
|
|
if (rc && err && rc == err)
|
|
|
|
return -ENODEV;
|
|
|
|
|
|
|
|
nd_region->btt_seed = nd_btt_create(nd_region);
|
2015-07-31 05:57:47 +08:00
|
|
|
nd_region->pfn_seed = nd_pfn_create(nd_region);
|
2015-06-01 03:02:11 +08:00
|
|
|
if (err == 0)
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Given multiple namespaces per region, we do not want to
|
|
|
|
* disable all the successfully registered peer namespaces upon
|
|
|
|
* a single registration failure. If userspace is missing a
|
|
|
|
* namespace that it expects it can disable/re-enable the region
|
|
|
|
* to retry discovery after correcting the failure.
|
|
|
|
* <regionX>/namespaces returns the current
|
|
|
|
* "<async-registered>/<total>" namespace count.
|
|
|
|
*/
|
|
|
|
dev_err(dev, "failed to register %d namespace%s, continuing...\n",
|
|
|
|
err, err == 1 ? "" : "s");
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int child_unregister(struct device *dev, void *data)
|
|
|
|
{
|
|
|
|
nd_device_unregister(dev, ND_SYNC);
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int nd_region_remove(struct device *dev)
|
|
|
|
{
|
2015-06-18 05:14:46 +08:00
|
|
|
struct nd_region *nd_region = to_nd_region(dev);
|
|
|
|
|
2015-06-01 03:02:11 +08:00
|
|
|
/* flush attribute readers and disable */
|
|
|
|
nvdimm_bus_lock(dev);
|
2015-06-18 05:14:46 +08:00
|
|
|
nd_region->ns_seed = NULL;
|
2015-06-25 16:20:04 +08:00
|
|
|
nd_region->btt_seed = NULL;
|
2015-07-31 05:57:47 +08:00
|
|
|
nd_region->pfn_seed = NULL;
|
2015-06-01 03:02:11 +08:00
|
|
|
dev_set_drvdata(dev, NULL);
|
|
|
|
nvdimm_bus_unlock(dev);
|
|
|
|
|
|
|
|
device_for_each_child(dev, NULL, child_unregister);
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2016-02-19 02:29:49 +08:00
|
|
|
static int child_notify(struct device *dev, void *data)
|
|
|
|
{
|
|
|
|
nd_device_notify(dev, *(enum nvdimm_event *) data);
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void nd_region_notify(struct device *dev, enum nvdimm_event event)
|
|
|
|
{
|
|
|
|
device_for_each_child(dev, &event, child_notify);
|
|
|
|
}
|
|
|
|
|
2015-06-01 03:02:11 +08:00
|
|
|
static struct nd_device_driver nd_region_driver = {
|
|
|
|
.probe = nd_region_probe,
|
|
|
|
.remove = nd_region_remove,
|
2016-02-19 02:29:49 +08:00
|
|
|
.notify = nd_region_notify,
|
2015-06-01 03:02:11 +08:00
|
|
|
.drv = {
|
|
|
|
.name = "nd_region",
|
|
|
|
},
|
|
|
|
.type = ND_DRIVER_REGION_BLK | ND_DRIVER_REGION_PMEM,
|
|
|
|
};
|
|
|
|
|
|
|
|
int __init nd_region_init(void)
|
|
|
|
{
|
|
|
|
return nd_driver_register(&nd_region_driver);
|
|
|
|
}
|
|
|
|
|
|
|
|
void nd_region_exit(void)
|
|
|
|
{
|
|
|
|
driver_unregister(&nd_region_driver.drv);
|
|
|
|
}
|
|
|
|
|
|
|
|
MODULE_ALIAS_ND_DEVICE(ND_DEVICE_REGION_PMEM);
|
|
|
|
MODULE_ALIAS_ND_DEVICE(ND_DEVICE_REGION_BLK);
|