OpenCloudOS-Kernel/drivers/md/dm-kcopyd.c

986 lines
22 KiB
C
Raw Normal View History

/*
* Copyright (C) 2002 Sistina Software (UK) Limited.
* Copyright (C) 2006 Red Hat GmbH
*
* This file is released under the GPL.
*
* Kcopyd provides a simple interface for copying an area of one
* block-device to one or more other block-devices, with an asynchronous
* completion notification.
*/
#include <linux/types.h>
#include <linux/atomic.h>
#include <linux/blkdev.h>
#include <linux/fs.h>
#include <linux/init.h>
#include <linux/list.h>
#include <linux/mempool.h>
#include <linux/module.h>
#include <linux/pagemap.h>
#include <linux/slab.h>
#include <linux/vmalloc.h>
#include <linux/workqueue.h>
#include <linux/mutex.h>
#include <linux/delay.h>
#include <linux/device-mapper.h>
#include <linux/dm-kcopyd.h>
#include "dm-core.h"
#define SPLIT_COUNT 8
#define MIN_JOBS 8
dm kcopyd: Increase default sub-job size to 512KB Currently, kcopyd has a sub-job size of 64KB and a maximum number of 8 sub-jobs. As a result, for any kcopyd job, we have a maximum of 512KB of I/O in flight. This upper limit to the amount of in-flight I/O under-utilizes fast devices and results in decreased throughput, e.g., when writing to a snapshotted thin LV with I/O size less than the pool's block size (so COW is performed using kcopyd). Increase kcopyd's default sub-job size to 512KB, so we have a maximum of 4MB of I/O in flight for each kcopyd job. This results in an up to 96% improvement of bandwidth when writing to a snapshotted thin LV, with I/O sizes less than the pool's block size. Also, add dm_mod.kcopyd_subjob_size_kb module parameter to allow users to fine tune the sub-job size of kcopyd. The default value of this parameter is 512KB and the maximum allowed value is 1024KB. We evaluate the performance impact of the change by running the snap_breaking_throughput benchmark, from the device mapper test suite [1]. The benchmark: 1. Creates a 1G thin LV 2. Provisions the thin LV 3. Takes a snapshot of the thin LV 4. Writes to the thin LV with: dd if=/dev/zero of=/dev/vg/thin_lv oflag=direct bs=<I/O size> Running this benchmark with various thin pool block sizes and dd I/O sizes (all combinations triggering the use of kcopyd) we get the following results: +-----------------+-------------+------------------+-----------------+ | Pool block size | dd I/O size | BW before (MB/s) | BW after (MB/s) | +-----------------+-------------+------------------+-----------------+ | 1 MB | 256 KB | 242 | 280 | | 1 MB | 512 KB | 238 | 295 | | | | | | | 2 MB | 256 KB | 238 | 354 | | 2 MB | 512 KB | 241 | 380 | | 2 MB | 1 MB | 245 | 394 | | | | | | | 4 MB | 256 KB | 248 | 412 | | 4 MB | 512 KB | 234 | 432 | | 4 MB | 1 MB | 251 | 474 | | 4 MB | 2 MB | 257 | 504 | | | | | | | 8 MB | 256 KB | 239 | 420 | | 8 MB | 512 KB | 256 | 431 | | 8 MB | 1 MB | 264 | 467 | | 8 MB | 2 MB | 264 | 502 | | 8 MB | 4 MB | 281 | 537 | +-----------------+-------------+------------------+-----------------+ [1] https://github.com/jthornber/device-mapper-test-suite Signed-off-by: Nikos Tsironis <ntsironis@arrikto.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2019-07-17 19:24:10 +08:00
#define DEFAULT_SUB_JOB_SIZE_KB 512
#define MAX_SUB_JOB_SIZE_KB 1024
static unsigned kcopyd_subjob_size_kb = DEFAULT_SUB_JOB_SIZE_KB;
module_param(kcopyd_subjob_size_kb, uint, S_IRUGO | S_IWUSR);
MODULE_PARM_DESC(kcopyd_subjob_size_kb, "Sub-job size for dm-kcopyd clients");
static unsigned dm_get_kcopyd_subjob_size(void)
{
unsigned sub_job_size_kb;
sub_job_size_kb = __dm_get_module_param(&kcopyd_subjob_size_kb,
DEFAULT_SUB_JOB_SIZE_KB,
MAX_SUB_JOB_SIZE_KB);
return sub_job_size_kb << 1;
}
/*-----------------------------------------------------------------
* Each kcopyd client has its own little pool of preallocated
* pages for kcopyd io.
*---------------------------------------------------------------*/
struct dm_kcopyd_client {
struct page_list *pages;
unsigned nr_reserved_pages;
unsigned nr_free_pages;
dm kcopyd: Increase default sub-job size to 512KB Currently, kcopyd has a sub-job size of 64KB and a maximum number of 8 sub-jobs. As a result, for any kcopyd job, we have a maximum of 512KB of I/O in flight. This upper limit to the amount of in-flight I/O under-utilizes fast devices and results in decreased throughput, e.g., when writing to a snapshotted thin LV with I/O size less than the pool's block size (so COW is performed using kcopyd). Increase kcopyd's default sub-job size to 512KB, so we have a maximum of 4MB of I/O in flight for each kcopyd job. This results in an up to 96% improvement of bandwidth when writing to a snapshotted thin LV, with I/O sizes less than the pool's block size. Also, add dm_mod.kcopyd_subjob_size_kb module parameter to allow users to fine tune the sub-job size of kcopyd. The default value of this parameter is 512KB and the maximum allowed value is 1024KB. We evaluate the performance impact of the change by running the snap_breaking_throughput benchmark, from the device mapper test suite [1]. The benchmark: 1. Creates a 1G thin LV 2. Provisions the thin LV 3. Takes a snapshot of the thin LV 4. Writes to the thin LV with: dd if=/dev/zero of=/dev/vg/thin_lv oflag=direct bs=<I/O size> Running this benchmark with various thin pool block sizes and dd I/O sizes (all combinations triggering the use of kcopyd) we get the following results: +-----------------+-------------+------------------+-----------------+ | Pool block size | dd I/O size | BW before (MB/s) | BW after (MB/s) | +-----------------+-------------+------------------+-----------------+ | 1 MB | 256 KB | 242 | 280 | | 1 MB | 512 KB | 238 | 295 | | | | | | | 2 MB | 256 KB | 238 | 354 | | 2 MB | 512 KB | 241 | 380 | | 2 MB | 1 MB | 245 | 394 | | | | | | | 4 MB | 256 KB | 248 | 412 | | 4 MB | 512 KB | 234 | 432 | | 4 MB | 1 MB | 251 | 474 | | 4 MB | 2 MB | 257 | 504 | | | | | | | 8 MB | 256 KB | 239 | 420 | | 8 MB | 512 KB | 256 | 431 | | 8 MB | 1 MB | 264 | 467 | | 8 MB | 2 MB | 264 | 502 | | 8 MB | 4 MB | 281 | 537 | +-----------------+-------------+------------------+-----------------+ [1] https://github.com/jthornber/device-mapper-test-suite Signed-off-by: Nikos Tsironis <ntsironis@arrikto.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2019-07-17 19:24:10 +08:00
unsigned sub_job_size;
struct dm_io_client *io_client;
wait_queue_head_t destroyq;
mempool_t job_pool;
struct workqueue_struct *kcopyd_wq;
struct work_struct kcopyd_work;
struct dm_kcopyd_throttle *throttle;
atomic_t nr_jobs;
/*
dm kcopyd: Fix bug causing workqueue stalls When using kcopyd to run callbacks through dm_kcopyd_do_callback() or submitting copy jobs with a source size of 0, the jobs are pushed directly to the complete_jobs list, which could be under processing by the kcopyd thread. As a result, the kcopyd thread can continue running completed jobs indefinitely, without releasing the CPU, as long as someone keeps submitting new completed jobs through the aforementioned paths. Processing of work items, queued for execution on the same CPU as the currently running kcopyd thread, is thus stalled for excessive amounts of time, hurting performance. Running the following test, from the device mapper test suite [1], dmtest run --suite snapshot -n parallel_io_to_many_snaps_N , with 8 active snapshots, we get, in dmesg, messages like the following: [68899.948523] BUG: workqueue lockup - pool cpus=0 node=0 flags=0x0 nice=0 stuck for 95s! [68899.949282] Showing busy workqueues and worker pools: [68899.949288] workqueue events: flags=0x0 [68899.949295] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=2/256 [68899.949306] pending: vmstat_shepherd, cache_reap [68899.949331] workqueue mm_percpu_wq: flags=0x8 [68899.949337] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=1/256 [68899.949345] pending: vmstat_update [68899.949387] workqueue dm_bufio_cache: flags=0x8 [68899.949392] pwq 4: cpus=2 node=0 flags=0x0 nice=0 active=1/256 [68899.949400] pending: work_fn [dm_bufio] [68899.949423] workqueue kcopyd: flags=0x8 [68899.949429] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=1/256 [68899.949437] pending: do_work [dm_mod] [68899.949452] workqueue kcopyd: flags=0x8 [68899.949458] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=2/256 [68899.949466] in-flight: 13:do_work [dm_mod] [68899.949474] pending: do_work [dm_mod] [68899.949487] workqueue kcopyd: flags=0x8 [68899.949493] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=1/256 [68899.949501] pending: do_work [dm_mod] [68899.949515] workqueue kcopyd: flags=0x8 [68899.949521] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=1/256 [68899.949529] pending: do_work [dm_mod] [68899.949541] workqueue kcopyd: flags=0x8 [68899.949547] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=1/256 [68899.949555] pending: do_work [dm_mod] [68899.949568] pool 0: cpus=0 node=0 flags=0x0 nice=0 hung=95s workers=4 idle: 27130 27223 1084 Fix this by splitting the complete_jobs list into two parts: A user facing part, named callback_jobs, and one used internally by kcopyd, retaining the name complete_jobs. dm_kcopyd_do_callback() and dispatch_job() now push their jobs to the callback_jobs list, which is spliced to the complete_jobs list once, every time the kcopyd thread wakes up. This prevents kcopyd from hogging the CPU indefinitely and causing workqueue stalls. Re-running the aforementioned test: * Workqueue stalls are eliminated * The maximum writing time among all targets is reduced from 09m37.10s to 06m04.85s and the total run time of the test is reduced from 10m43.591s to 7m19.199s [1] https://github.com/jthornber/device-mapper-test-suite Signed-off-by: Nikos Tsironis <ntsironis@arrikto.com> Signed-off-by: Ilias Tsitsimpis <iliastsi@arrikto.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2018-11-01 05:53:09 +08:00
* We maintain four lists of jobs:
*
* i) jobs waiting for pages
* ii) jobs that have pages, and are waiting for the io to be issued.
dm kcopyd: Fix bug causing workqueue stalls When using kcopyd to run callbacks through dm_kcopyd_do_callback() or submitting copy jobs with a source size of 0, the jobs are pushed directly to the complete_jobs list, which could be under processing by the kcopyd thread. As a result, the kcopyd thread can continue running completed jobs indefinitely, without releasing the CPU, as long as someone keeps submitting new completed jobs through the aforementioned paths. Processing of work items, queued for execution on the same CPU as the currently running kcopyd thread, is thus stalled for excessive amounts of time, hurting performance. Running the following test, from the device mapper test suite [1], dmtest run --suite snapshot -n parallel_io_to_many_snaps_N , with 8 active snapshots, we get, in dmesg, messages like the following: [68899.948523] BUG: workqueue lockup - pool cpus=0 node=0 flags=0x0 nice=0 stuck for 95s! [68899.949282] Showing busy workqueues and worker pools: [68899.949288] workqueue events: flags=0x0 [68899.949295] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=2/256 [68899.949306] pending: vmstat_shepherd, cache_reap [68899.949331] workqueue mm_percpu_wq: flags=0x8 [68899.949337] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=1/256 [68899.949345] pending: vmstat_update [68899.949387] workqueue dm_bufio_cache: flags=0x8 [68899.949392] pwq 4: cpus=2 node=0 flags=0x0 nice=0 active=1/256 [68899.949400] pending: work_fn [dm_bufio] [68899.949423] workqueue kcopyd: flags=0x8 [68899.949429] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=1/256 [68899.949437] pending: do_work [dm_mod] [68899.949452] workqueue kcopyd: flags=0x8 [68899.949458] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=2/256 [68899.949466] in-flight: 13:do_work [dm_mod] [68899.949474] pending: do_work [dm_mod] [68899.949487] workqueue kcopyd: flags=0x8 [68899.949493] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=1/256 [68899.949501] pending: do_work [dm_mod] [68899.949515] workqueue kcopyd: flags=0x8 [68899.949521] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=1/256 [68899.949529] pending: do_work [dm_mod] [68899.949541] workqueue kcopyd: flags=0x8 [68899.949547] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=1/256 [68899.949555] pending: do_work [dm_mod] [68899.949568] pool 0: cpus=0 node=0 flags=0x0 nice=0 hung=95s workers=4 idle: 27130 27223 1084 Fix this by splitting the complete_jobs list into two parts: A user facing part, named callback_jobs, and one used internally by kcopyd, retaining the name complete_jobs. dm_kcopyd_do_callback() and dispatch_job() now push their jobs to the callback_jobs list, which is spliced to the complete_jobs list once, every time the kcopyd thread wakes up. This prevents kcopyd from hogging the CPU indefinitely and causing workqueue stalls. Re-running the aforementioned test: * Workqueue stalls are eliminated * The maximum writing time among all targets is reduced from 09m37.10s to 06m04.85s and the total run time of the test is reduced from 10m43.591s to 7m19.199s [1] https://github.com/jthornber/device-mapper-test-suite Signed-off-by: Nikos Tsironis <ntsironis@arrikto.com> Signed-off-by: Ilias Tsitsimpis <iliastsi@arrikto.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2018-11-01 05:53:09 +08:00
* iii) jobs that don't need to do any IO and just run a callback
* iv) jobs that have completed.
*
dm kcopyd: Fix bug causing workqueue stalls When using kcopyd to run callbacks through dm_kcopyd_do_callback() or submitting copy jobs with a source size of 0, the jobs are pushed directly to the complete_jobs list, which could be under processing by the kcopyd thread. As a result, the kcopyd thread can continue running completed jobs indefinitely, without releasing the CPU, as long as someone keeps submitting new completed jobs through the aforementioned paths. Processing of work items, queued for execution on the same CPU as the currently running kcopyd thread, is thus stalled for excessive amounts of time, hurting performance. Running the following test, from the device mapper test suite [1], dmtest run --suite snapshot -n parallel_io_to_many_snaps_N , with 8 active snapshots, we get, in dmesg, messages like the following: [68899.948523] BUG: workqueue lockup - pool cpus=0 node=0 flags=0x0 nice=0 stuck for 95s! [68899.949282] Showing busy workqueues and worker pools: [68899.949288] workqueue events: flags=0x0 [68899.949295] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=2/256 [68899.949306] pending: vmstat_shepherd, cache_reap [68899.949331] workqueue mm_percpu_wq: flags=0x8 [68899.949337] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=1/256 [68899.949345] pending: vmstat_update [68899.949387] workqueue dm_bufio_cache: flags=0x8 [68899.949392] pwq 4: cpus=2 node=0 flags=0x0 nice=0 active=1/256 [68899.949400] pending: work_fn [dm_bufio] [68899.949423] workqueue kcopyd: flags=0x8 [68899.949429] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=1/256 [68899.949437] pending: do_work [dm_mod] [68899.949452] workqueue kcopyd: flags=0x8 [68899.949458] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=2/256 [68899.949466] in-flight: 13:do_work [dm_mod] [68899.949474] pending: do_work [dm_mod] [68899.949487] workqueue kcopyd: flags=0x8 [68899.949493] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=1/256 [68899.949501] pending: do_work [dm_mod] [68899.949515] workqueue kcopyd: flags=0x8 [68899.949521] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=1/256 [68899.949529] pending: do_work [dm_mod] [68899.949541] workqueue kcopyd: flags=0x8 [68899.949547] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=1/256 [68899.949555] pending: do_work [dm_mod] [68899.949568] pool 0: cpus=0 node=0 flags=0x0 nice=0 hung=95s workers=4 idle: 27130 27223 1084 Fix this by splitting the complete_jobs list into two parts: A user facing part, named callback_jobs, and one used internally by kcopyd, retaining the name complete_jobs. dm_kcopyd_do_callback() and dispatch_job() now push their jobs to the callback_jobs list, which is spliced to the complete_jobs list once, every time the kcopyd thread wakes up. This prevents kcopyd from hogging the CPU indefinitely and causing workqueue stalls. Re-running the aforementioned test: * Workqueue stalls are eliminated * The maximum writing time among all targets is reduced from 09m37.10s to 06m04.85s and the total run time of the test is reduced from 10m43.591s to 7m19.199s [1] https://github.com/jthornber/device-mapper-test-suite Signed-off-by: Nikos Tsironis <ntsironis@arrikto.com> Signed-off-by: Ilias Tsitsimpis <iliastsi@arrikto.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2018-11-01 05:53:09 +08:00
* All four of these are protected by job_lock.
*/
spinlock_t job_lock;
dm kcopyd: Fix bug causing workqueue stalls When using kcopyd to run callbacks through dm_kcopyd_do_callback() or submitting copy jobs with a source size of 0, the jobs are pushed directly to the complete_jobs list, which could be under processing by the kcopyd thread. As a result, the kcopyd thread can continue running completed jobs indefinitely, without releasing the CPU, as long as someone keeps submitting new completed jobs through the aforementioned paths. Processing of work items, queued for execution on the same CPU as the currently running kcopyd thread, is thus stalled for excessive amounts of time, hurting performance. Running the following test, from the device mapper test suite [1], dmtest run --suite snapshot -n parallel_io_to_many_snaps_N , with 8 active snapshots, we get, in dmesg, messages like the following: [68899.948523] BUG: workqueue lockup - pool cpus=0 node=0 flags=0x0 nice=0 stuck for 95s! [68899.949282] Showing busy workqueues and worker pools: [68899.949288] workqueue events: flags=0x0 [68899.949295] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=2/256 [68899.949306] pending: vmstat_shepherd, cache_reap [68899.949331] workqueue mm_percpu_wq: flags=0x8 [68899.949337] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=1/256 [68899.949345] pending: vmstat_update [68899.949387] workqueue dm_bufio_cache: flags=0x8 [68899.949392] pwq 4: cpus=2 node=0 flags=0x0 nice=0 active=1/256 [68899.949400] pending: work_fn [dm_bufio] [68899.949423] workqueue kcopyd: flags=0x8 [68899.949429] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=1/256 [68899.949437] pending: do_work [dm_mod] [68899.949452] workqueue kcopyd: flags=0x8 [68899.949458] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=2/256 [68899.949466] in-flight: 13:do_work [dm_mod] [68899.949474] pending: do_work [dm_mod] [68899.949487] workqueue kcopyd: flags=0x8 [68899.949493] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=1/256 [68899.949501] pending: do_work [dm_mod] [68899.949515] workqueue kcopyd: flags=0x8 [68899.949521] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=1/256 [68899.949529] pending: do_work [dm_mod] [68899.949541] workqueue kcopyd: flags=0x8 [68899.949547] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=1/256 [68899.949555] pending: do_work [dm_mod] [68899.949568] pool 0: cpus=0 node=0 flags=0x0 nice=0 hung=95s workers=4 idle: 27130 27223 1084 Fix this by splitting the complete_jobs list into two parts: A user facing part, named callback_jobs, and one used internally by kcopyd, retaining the name complete_jobs. dm_kcopyd_do_callback() and dispatch_job() now push their jobs to the callback_jobs list, which is spliced to the complete_jobs list once, every time the kcopyd thread wakes up. This prevents kcopyd from hogging the CPU indefinitely and causing workqueue stalls. Re-running the aforementioned test: * Workqueue stalls are eliminated * The maximum writing time among all targets is reduced from 09m37.10s to 06m04.85s and the total run time of the test is reduced from 10m43.591s to 7m19.199s [1] https://github.com/jthornber/device-mapper-test-suite Signed-off-by: Nikos Tsironis <ntsironis@arrikto.com> Signed-off-by: Ilias Tsitsimpis <iliastsi@arrikto.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2018-11-01 05:53:09 +08:00
struct list_head callback_jobs;
struct list_head complete_jobs;
struct list_head io_jobs;
struct list_head pages_jobs;
};
static struct page_list zero_page_list;
static DEFINE_SPINLOCK(throttle_spinlock);
/*
* IO/IDLE accounting slowly decays after (1 << ACCOUNT_INTERVAL_SHIFT) period.
* When total_period >= (1 << ACCOUNT_INTERVAL_SHIFT) the counters are divided
* by 2.
*/
#define ACCOUNT_INTERVAL_SHIFT SHIFT_HZ
/*
* Sleep this number of milliseconds.
*
* The value was decided experimentally.
* Smaller values seem to cause an increased copy rate above the limit.
* The reason for this is unknown but possibly due to jiffies rounding errors
* or read/write cache inside the disk.
*/
#define SLEEP_MSEC 100
/*
* Maximum number of sleep events. There is a theoretical livelock if more
* kcopyd clients do work simultaneously which this limit avoids.
*/
#define MAX_SLEEPS 10
static void io_job_start(struct dm_kcopyd_throttle *t)
{
unsigned throttle, now, difference;
int slept = 0, skew;
if (unlikely(!t))
return;
try_again:
spin_lock_irq(&throttle_spinlock);
locking/atomics: COCCINELLE/treewide: Convert trivial ACCESS_ONCE() patterns to READ_ONCE()/WRITE_ONCE() Please do not apply this to mainline directly, instead please re-run the coccinelle script shown below and apply its output. For several reasons, it is desirable to use {READ,WRITE}_ONCE() in preference to ACCESS_ONCE(), and new code is expected to use one of the former. So far, there's been no reason to change most existing uses of ACCESS_ONCE(), as these aren't harmful, and changing them results in churn. However, for some features, the read/write distinction is critical to correct operation. To distinguish these cases, separate read/write accessors must be used. This patch migrates (most) remaining ACCESS_ONCE() instances to {READ,WRITE}_ONCE(), using the following coccinelle script: ---- // Convert trivial ACCESS_ONCE() uses to equivalent READ_ONCE() and // WRITE_ONCE() // $ make coccicheck COCCI=/home/mark/once.cocci SPFLAGS="--include-headers" MODE=patch virtual patch @ depends on patch @ expression E1, E2; @@ - ACCESS_ONCE(E1) = E2 + WRITE_ONCE(E1, E2) @ depends on patch @ expression E; @@ - ACCESS_ONCE(E) + READ_ONCE(E) ---- Signed-off-by: Mark Rutland <mark.rutland@arm.com> Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: davem@davemloft.net Cc: linux-arch@vger.kernel.org Cc: mpe@ellerman.id.au Cc: shuah@kernel.org Cc: snitzer@redhat.com Cc: thor.thayer@linux.intel.com Cc: tj@kernel.org Cc: viro@zeniv.linux.org.uk Cc: will.deacon@arm.com Link: http://lkml.kernel.org/r/1508792849-3115-19-git-send-email-paulmck@linux.vnet.ibm.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2017-10-24 05:07:29 +08:00
throttle = READ_ONCE(t->throttle);
if (likely(throttle >= 100))
goto skip_limit;
now = jiffies;
difference = now - t->last_jiffies;
t->last_jiffies = now;
if (t->num_io_jobs)
t->io_period += difference;
t->total_period += difference;
/*
* Maintain sane values if we got a temporary overflow.
*/
if (unlikely(t->io_period > t->total_period))
t->io_period = t->total_period;
if (unlikely(t->total_period >= (1 << ACCOUNT_INTERVAL_SHIFT))) {
int shift = fls(t->total_period >> ACCOUNT_INTERVAL_SHIFT);
t->total_period >>= shift;
t->io_period >>= shift;
}
skew = t->io_period - throttle * t->total_period / 100;
if (unlikely(skew > 0) && slept < MAX_SLEEPS) {
slept++;
spin_unlock_irq(&throttle_spinlock);
msleep(SLEEP_MSEC);
goto try_again;
}
skip_limit:
t->num_io_jobs++;
spin_unlock_irq(&throttle_spinlock);
}
static void io_job_finish(struct dm_kcopyd_throttle *t)
{
unsigned long flags;
if (unlikely(!t))
return;
spin_lock_irqsave(&throttle_spinlock, flags);
t->num_io_jobs--;
locking/atomics: COCCINELLE/treewide: Convert trivial ACCESS_ONCE() patterns to READ_ONCE()/WRITE_ONCE() Please do not apply this to mainline directly, instead please re-run the coccinelle script shown below and apply its output. For several reasons, it is desirable to use {READ,WRITE}_ONCE() in preference to ACCESS_ONCE(), and new code is expected to use one of the former. So far, there's been no reason to change most existing uses of ACCESS_ONCE(), as these aren't harmful, and changing them results in churn. However, for some features, the read/write distinction is critical to correct operation. To distinguish these cases, separate read/write accessors must be used. This patch migrates (most) remaining ACCESS_ONCE() instances to {READ,WRITE}_ONCE(), using the following coccinelle script: ---- // Convert trivial ACCESS_ONCE() uses to equivalent READ_ONCE() and // WRITE_ONCE() // $ make coccicheck COCCI=/home/mark/once.cocci SPFLAGS="--include-headers" MODE=patch virtual patch @ depends on patch @ expression E1, E2; @@ - ACCESS_ONCE(E1) = E2 + WRITE_ONCE(E1, E2) @ depends on patch @ expression E; @@ - ACCESS_ONCE(E) + READ_ONCE(E) ---- Signed-off-by: Mark Rutland <mark.rutland@arm.com> Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: davem@davemloft.net Cc: linux-arch@vger.kernel.org Cc: mpe@ellerman.id.au Cc: shuah@kernel.org Cc: snitzer@redhat.com Cc: thor.thayer@linux.intel.com Cc: tj@kernel.org Cc: viro@zeniv.linux.org.uk Cc: will.deacon@arm.com Link: http://lkml.kernel.org/r/1508792849-3115-19-git-send-email-paulmck@linux.vnet.ibm.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2017-10-24 05:07:29 +08:00
if (likely(READ_ONCE(t->throttle) >= 100))
goto skip_limit;
if (!t->num_io_jobs) {
unsigned now, difference;
now = jiffies;
difference = now - t->last_jiffies;
t->last_jiffies = now;
t->io_period += difference;
t->total_period += difference;
/*
* Maintain sane values if we got a temporary overflow.
*/
if (unlikely(t->io_period > t->total_period))
t->io_period = t->total_period;
}
skip_limit:
spin_unlock_irqrestore(&throttle_spinlock, flags);
}
static void wake(struct dm_kcopyd_client *kc)
{
queue_work(kc->kcopyd_wq, &kc->kcopyd_work);
}
/*
* Obtain one page for the use of kcopyd.
*/
static struct page_list *alloc_pl(gfp_t gfp)
{
struct page_list *pl;
pl = kmalloc(sizeof(*pl), gfp);
if (!pl)
return NULL;
pl->page = alloc_page(gfp);
if (!pl->page) {
kfree(pl);
return NULL;
}
return pl;
}
static void free_pl(struct page_list *pl)
{
__free_page(pl->page);
kfree(pl);
}
/*
* Add the provided pages to a client's free page list, releasing
* back to the system any beyond the reserved_pages limit.
*/
static void kcopyd_put_pages(struct dm_kcopyd_client *kc, struct page_list *pl)
{
struct page_list *next;
do {
next = pl->next;
if (kc->nr_free_pages >= kc->nr_reserved_pages)
free_pl(pl);
else {
pl->next = kc->pages;
kc->pages = pl;
kc->nr_free_pages++;
}
pl = next;
} while (pl);
}
static int kcopyd_get_pages(struct dm_kcopyd_client *kc,
unsigned int nr, struct page_list **pages)
{
struct page_list *pl;
*pages = NULL;
do {
mm, page_alloc: distinguish between being unable to sleep, unwilling to sleep and avoiding waking kswapd __GFP_WAIT has been used to identify atomic context in callers that hold spinlocks or are in interrupts. They are expected to be high priority and have access one of two watermarks lower than "min" which can be referred to as the "atomic reserve". __GFP_HIGH users get access to the first lower watermark and can be called the "high priority reserve". Over time, callers had a requirement to not block when fallback options were available. Some have abused __GFP_WAIT leading to a situation where an optimisitic allocation with a fallback option can access atomic reserves. This patch uses __GFP_ATOMIC to identify callers that are truely atomic, cannot sleep and have no alternative. High priority users continue to use __GFP_HIGH. __GFP_DIRECT_RECLAIM identifies callers that can sleep and are willing to enter direct reclaim. __GFP_KSWAPD_RECLAIM to identify callers that want to wake kswapd for background reclaim. __GFP_WAIT is redefined as a caller that is willing to enter direct reclaim and wake kswapd for background reclaim. This patch then converts a number of sites o __GFP_ATOMIC is used by callers that are high priority and have memory pools for those requests. GFP_ATOMIC uses this flag. o Callers that have a limited mempool to guarantee forward progress clear __GFP_DIRECT_RECLAIM but keep __GFP_KSWAPD_RECLAIM. bio allocations fall into this category where kswapd will still be woken but atomic reserves are not used as there is a one-entry mempool to guarantee progress. o Callers that are checking if they are non-blocking should use the helper gfpflags_allow_blocking() where possible. This is because checking for __GFP_WAIT as was done historically now can trigger false positives. Some exceptions like dm-crypt.c exist where the code intent is clearer if __GFP_DIRECT_RECLAIM is used instead of the helper due to flag manipulations. o Callers that built their own GFP flags instead of starting with GFP_KERNEL and friends now also need to specify __GFP_KSWAPD_RECLAIM. The first key hazard to watch out for is callers that removed __GFP_WAIT and was depending on access to atomic reserves for inconspicuous reasons. In some cases it may be appropriate for them to use __GFP_HIGH. The second key hazard is callers that assembled their own combination of GFP flags instead of starting with something like GFP_KERNEL. They may now wish to specify __GFP_KSWAPD_RECLAIM. It's almost certainly harmless if it's missed in most cases as other activity will wake kswapd. Signed-off-by: Mel Gorman <mgorman@techsingularity.net> Acked-by: Vlastimil Babka <vbabka@suse.cz> Acked-by: Michal Hocko <mhocko@suse.com> Acked-by: Johannes Weiner <hannes@cmpxchg.org> Cc: Christoph Lameter <cl@linux.com> Cc: David Rientjes <rientjes@google.com> Cc: Vitaly Wool <vitalywool@gmail.com> Cc: Rik van Riel <riel@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2015-11-07 08:28:21 +08:00
pl = alloc_pl(__GFP_NOWARN | __GFP_NORETRY | __GFP_KSWAPD_RECLAIM);
if (unlikely(!pl)) {
/* Use reserved pages */
pl = kc->pages;
if (unlikely(!pl))
goto out_of_memory;
kc->pages = pl->next;
kc->nr_free_pages--;
}
pl->next = *pages;
*pages = pl;
} while (--nr);
return 0;
out_of_memory:
if (*pages)
kcopyd_put_pages(kc, *pages);
return -ENOMEM;
}
/*
* These three functions resize the page pool.
*/
static void drop_pages(struct page_list *pl)
{
struct page_list *next;
while (pl) {
next = pl->next;
free_pl(pl);
pl = next;
}
}
/*
* Allocate and reserve nr_pages for the use of a specific client.
*/
static int client_reserve_pages(struct dm_kcopyd_client *kc, unsigned nr_pages)
{
unsigned i;
struct page_list *pl = NULL, *next;
for (i = 0; i < nr_pages; i++) {
next = alloc_pl(GFP_KERNEL);
if (!next) {
if (pl)
drop_pages(pl);
return -ENOMEM;
}
next->next = pl;
pl = next;
}
kc->nr_reserved_pages += nr_pages;
kcopyd_put_pages(kc, pl);
return 0;
}
static void client_free_pages(struct dm_kcopyd_client *kc)
{
BUG_ON(kc->nr_free_pages != kc->nr_reserved_pages);
drop_pages(kc->pages);
kc->pages = NULL;
kc->nr_free_pages = kc->nr_reserved_pages = 0;
}
/*-----------------------------------------------------------------
* kcopyd_jobs need to be allocated by the *clients* of kcopyd,
* for this reason we use a mempool to prevent the client from
* ever having to do io (which could cause a deadlock).
*---------------------------------------------------------------*/
struct kcopyd_job {
struct dm_kcopyd_client *kc;
struct list_head list;
unsigned long flags;
/*
* Error state of the job.
*/
int read_err;
unsigned long write_err;
/*
* Either READ or WRITE
*/
int rw;
struct dm_io_region source;
/*
* The destinations for the transfer.
*/
unsigned int num_dests;
struct dm_io_region dests[DM_KCOPYD_MAX_REGIONS];
struct page_list *pages;
/*
* Set this to ensure you are notified when the job has
* completed. 'context' is for callback to use.
*/
dm_kcopyd_notify_fn fn;
void *context;
/*
* These fields are only used if the job has been split
* into more manageable parts.
*/
struct mutex lock;
atomic_t sub_jobs;
sector_t progress;
sector_t write_offset;
struct kcopyd_job *master_job;
};
static struct kmem_cache *_job_cache;
int __init dm_kcopyd_init(void)
{
_job_cache = kmem_cache_create("kcopyd_job",
sizeof(struct kcopyd_job) * (SPLIT_COUNT + 1),
__alignof__(struct kcopyd_job), 0, NULL);
if (!_job_cache)
return -ENOMEM;
zero_page_list.next = &zero_page_list;
zero_page_list.page = ZERO_PAGE(0);
return 0;
}
void dm_kcopyd_exit(void)
{
kmem_cache_destroy(_job_cache);
_job_cache = NULL;
}
/*
* Functions to push and pop a job onto the head of a given job
* list.
*/
static struct kcopyd_job *pop_io_job(struct list_head *jobs,
struct dm_kcopyd_client *kc)
{
struct kcopyd_job *job;
/*
* For I/O jobs, pop any read, any write without sequential write
* constraint and sequential writes that are at the right position.
*/
list_for_each_entry(job, jobs, list) {
if (job->rw == READ || !test_bit(DM_KCOPYD_WRITE_SEQ, &job->flags)) {
list_del(&job->list);
return job;
}
if (job->write_offset == job->master_job->write_offset) {
job->master_job->write_offset += job->source.count;
list_del(&job->list);
return job;
}
}
return NULL;
}
static struct kcopyd_job *pop(struct list_head *jobs,
struct dm_kcopyd_client *kc)
{
struct kcopyd_job *job = NULL;
unsigned long flags;
spin_lock_irqsave(&kc->job_lock, flags);
if (!list_empty(jobs)) {
if (jobs == &kc->io_jobs)
job = pop_io_job(jobs, kc);
else {
job = list_entry(jobs->next, struct kcopyd_job, list);
list_del(&job->list);
}
}
spin_unlock_irqrestore(&kc->job_lock, flags);
return job;
}
static void push(struct list_head *jobs, struct kcopyd_job *job)
{
unsigned long flags;
struct dm_kcopyd_client *kc = job->kc;
spin_lock_irqsave(&kc->job_lock, flags);
list_add_tail(&job->list, jobs);
spin_unlock_irqrestore(&kc->job_lock, flags);
}
dm kcopyd: avoid queue shuffle Write throughput to LVM snapshot origin volume is an order of magnitude slower than those to LV without snapshots or snapshot target volumes, especially in the case of sequential writes with O_SYNC on. The following patch originally written by Kevin Jamieson and Jan Blunck and slightly modified for the current RCs by myself tries to improve the performance by modifying the behaviour of kcopyd, so that it pushes back an I/O job to the head of the job queue instead of the tail as process_jobs() currently does when it has to wait for free pages. This way, write requests aren't shuffled to cause extra seeks. I tested the patch against 2.6.27-rc5 and got the following results. The test is a dd command writing to snapshot origin followed by fsync to the file just created/updated. A couple of filesystem benchmarks gave me similar results in case of sequential writes, while random writes didn't suffer much. dd if=/dev/zero of=<somewhere on snapshot origin> bs=4096 count=... [conv=notrunc when updating] 1) linux 2.6.27-rc5 without the patch, write to snapshot origin, average throughput (MB/s) 10M 100M 1000M create,dd 511.46 610.72 11.81 create,dd+fsync 7.10 6.77 8.13 update,dd 431.63 917.41 12.75 update,dd+fsync 7.79 7.43 8.12 compared with write throughput to LV without any snapshots, all dd+fsync and 1000 MiB writes perform very poorly. 10M 100M 1000M create,dd 555.03 608.98 123.29 create,dd+fsync 114.27 72.78 76.65 update,dd 152.34 1267.27 124.04 update,dd+fsync 130.56 77.81 77.84 2) linux 2.6.27-rc5 with the patch, write to snapshot origin, average throughput (MB/s) 10M 100M 1000M create,dd 537.06 589.44 46.21 create,dd+fsync 31.63 29.19 29.23 update,dd 487.59 897.65 37.76 update,dd+fsync 34.12 30.07 26.85 Although still not on par with plain LV performance - cannot be avoided because it's copy on write anyway - this simple patch successfully improves throughtput of dd+fsync while not affecting the rest. Signed-off-by: Jan Blunck <jblunck@suse.de> Signed-off-by: Kazuo Ito <ito.kazuo@oss.ntt.co.jp> Signed-off-by: Alasdair G Kergon <agk@redhat.com> Cc: stable@kernel.org
2008-10-22 00:44:50 +08:00
static void push_head(struct list_head *jobs, struct kcopyd_job *job)
{
unsigned long flags;
struct dm_kcopyd_client *kc = job->kc;
spin_lock_irqsave(&kc->job_lock, flags);
list_add(&job->list, jobs);
spin_unlock_irqrestore(&kc->job_lock, flags);
}
/*
* These three functions process 1 item from the corresponding
* job list.
*
* They return:
* < 0: error
* 0: success
* > 0: can't process yet.
*/
static int run_complete_job(struct kcopyd_job *job)
{
void *context = job->context;
int read_err = job->read_err;
unsigned long write_err = job->write_err;
dm_kcopyd_notify_fn fn = job->fn;
struct dm_kcopyd_client *kc = job->kc;
if (job->pages && job->pages != &zero_page_list)
kcopyd_put_pages(kc, job->pages);
/*
* If this is the master job, the sub jobs have already
* completed so we can free everything.
*/
if (job->master_job == job) {
mutex_destroy(&job->lock);
mempool_free(job, &kc->job_pool);
}
fn(read_err, write_err, context);
if (atomic_dec_and_test(&kc->nr_jobs))
wake_up(&kc->destroyq);
cond_resched();
return 0;
}
static void complete_io(unsigned long error, void *context)
{
struct kcopyd_job *job = (struct kcopyd_job *) context;
struct dm_kcopyd_client *kc = job->kc;
io_job_finish(kc->throttle);
if (error) {
if (op_is_write(job->rw))
job->write_err |= error;
else
job->read_err = 1;
if (!test_bit(DM_KCOPYD_IGNORE_ERROR, &job->flags)) {
push(&kc->complete_jobs, job);
wake(kc);
return;
}
}
if (op_is_write(job->rw))
push(&kc->complete_jobs, job);
else {
job->rw = WRITE;
push(&kc->io_jobs, job);
}
wake(kc);
}
/*
* Request io on as many buffer heads as we can currently get for
* a particular job.
*/
static int run_io_job(struct kcopyd_job *job)
{
int r;
struct dm_io_request io_req = {
.bi_op = job->rw,
.bi_op_flags = 0,
.mem.type = DM_IO_PAGE_LIST,
.mem.ptr.pl = job->pages,
.mem.offset = 0,
.notify.fn = complete_io,
.notify.context = job,
.client = job->kc->io_client,
};
/*
* If we need to write sequentially and some reads or writes failed,
* no point in continuing.
*/
if (test_bit(DM_KCOPYD_WRITE_SEQ, &job->flags) &&
job->master_job->write_err) {
job->write_err = job->master_job->write_err;
return -EIO;
}
io_job_start(job->kc->throttle);
if (job->rw == READ)
r = dm_io(&io_req, 1, &job->source, NULL);
else
r = dm_io(&io_req, job->num_dests, job->dests, NULL);
return r;
}
static int run_pages_job(struct kcopyd_job *job)
{
int r;
unsigned nr_pages = dm_div_up(job->dests[0].count, PAGE_SIZE >> 9);
r = kcopyd_get_pages(job->kc, nr_pages, &job->pages);
if (!r) {
/* this job is ready for io */
push(&job->kc->io_jobs, job);
return 0;
}
if (r == -ENOMEM)
/* can't complete now */
return 1;
return r;
}
/*
* Run through a list for as long as possible. Returns the count
* of successful jobs.
*/
static int process_jobs(struct list_head *jobs, struct dm_kcopyd_client *kc,
int (*fn) (struct kcopyd_job *))
{
struct kcopyd_job *job;
int r, count = 0;
while ((job = pop(jobs, kc))) {
r = fn(job);
if (r < 0) {
/* error this rogue job */
if (op_is_write(job->rw))
job->write_err = (unsigned long) -1L;
else
job->read_err = 1;
push(&kc->complete_jobs, job);
wake(kc);
break;
}
if (r > 0) {
/*
* We couldn't service this job ATM, so
* push this job back onto the list.
*/
dm kcopyd: avoid queue shuffle Write throughput to LVM snapshot origin volume is an order of magnitude slower than those to LV without snapshots or snapshot target volumes, especially in the case of sequential writes with O_SYNC on. The following patch originally written by Kevin Jamieson and Jan Blunck and slightly modified for the current RCs by myself tries to improve the performance by modifying the behaviour of kcopyd, so that it pushes back an I/O job to the head of the job queue instead of the tail as process_jobs() currently does when it has to wait for free pages. This way, write requests aren't shuffled to cause extra seeks. I tested the patch against 2.6.27-rc5 and got the following results. The test is a dd command writing to snapshot origin followed by fsync to the file just created/updated. A couple of filesystem benchmarks gave me similar results in case of sequential writes, while random writes didn't suffer much. dd if=/dev/zero of=<somewhere on snapshot origin> bs=4096 count=... [conv=notrunc when updating] 1) linux 2.6.27-rc5 without the patch, write to snapshot origin, average throughput (MB/s) 10M 100M 1000M create,dd 511.46 610.72 11.81 create,dd+fsync 7.10 6.77 8.13 update,dd 431.63 917.41 12.75 update,dd+fsync 7.79 7.43 8.12 compared with write throughput to LV without any snapshots, all dd+fsync and 1000 MiB writes perform very poorly. 10M 100M 1000M create,dd 555.03 608.98 123.29 create,dd+fsync 114.27 72.78 76.65 update,dd 152.34 1267.27 124.04 update,dd+fsync 130.56 77.81 77.84 2) linux 2.6.27-rc5 with the patch, write to snapshot origin, average throughput (MB/s) 10M 100M 1000M create,dd 537.06 589.44 46.21 create,dd+fsync 31.63 29.19 29.23 update,dd 487.59 897.65 37.76 update,dd+fsync 34.12 30.07 26.85 Although still not on par with plain LV performance - cannot be avoided because it's copy on write anyway - this simple patch successfully improves throughtput of dd+fsync while not affecting the rest. Signed-off-by: Jan Blunck <jblunck@suse.de> Signed-off-by: Kazuo Ito <ito.kazuo@oss.ntt.co.jp> Signed-off-by: Alasdair G Kergon <agk@redhat.com> Cc: stable@kernel.org
2008-10-22 00:44:50 +08:00
push_head(jobs, job);
break;
}
count++;
}
return count;
}
/*
* kcopyd does this every time it's woken up.
*/
static void do_work(struct work_struct *work)
{
struct dm_kcopyd_client *kc = container_of(work,
struct dm_kcopyd_client, kcopyd_work);
struct blk_plug plug;
dm kcopyd: Fix bug causing workqueue stalls When using kcopyd to run callbacks through dm_kcopyd_do_callback() or submitting copy jobs with a source size of 0, the jobs are pushed directly to the complete_jobs list, which could be under processing by the kcopyd thread. As a result, the kcopyd thread can continue running completed jobs indefinitely, without releasing the CPU, as long as someone keeps submitting new completed jobs through the aforementioned paths. Processing of work items, queued for execution on the same CPU as the currently running kcopyd thread, is thus stalled for excessive amounts of time, hurting performance. Running the following test, from the device mapper test suite [1], dmtest run --suite snapshot -n parallel_io_to_many_snaps_N , with 8 active snapshots, we get, in dmesg, messages like the following: [68899.948523] BUG: workqueue lockup - pool cpus=0 node=0 flags=0x0 nice=0 stuck for 95s! [68899.949282] Showing busy workqueues and worker pools: [68899.949288] workqueue events: flags=0x0 [68899.949295] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=2/256 [68899.949306] pending: vmstat_shepherd, cache_reap [68899.949331] workqueue mm_percpu_wq: flags=0x8 [68899.949337] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=1/256 [68899.949345] pending: vmstat_update [68899.949387] workqueue dm_bufio_cache: flags=0x8 [68899.949392] pwq 4: cpus=2 node=0 flags=0x0 nice=0 active=1/256 [68899.949400] pending: work_fn [dm_bufio] [68899.949423] workqueue kcopyd: flags=0x8 [68899.949429] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=1/256 [68899.949437] pending: do_work [dm_mod] [68899.949452] workqueue kcopyd: flags=0x8 [68899.949458] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=2/256 [68899.949466] in-flight: 13:do_work [dm_mod] [68899.949474] pending: do_work [dm_mod] [68899.949487] workqueue kcopyd: flags=0x8 [68899.949493] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=1/256 [68899.949501] pending: do_work [dm_mod] [68899.949515] workqueue kcopyd: flags=0x8 [68899.949521] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=1/256 [68899.949529] pending: do_work [dm_mod] [68899.949541] workqueue kcopyd: flags=0x8 [68899.949547] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=1/256 [68899.949555] pending: do_work [dm_mod] [68899.949568] pool 0: cpus=0 node=0 flags=0x0 nice=0 hung=95s workers=4 idle: 27130 27223 1084 Fix this by splitting the complete_jobs list into two parts: A user facing part, named callback_jobs, and one used internally by kcopyd, retaining the name complete_jobs. dm_kcopyd_do_callback() and dispatch_job() now push their jobs to the callback_jobs list, which is spliced to the complete_jobs list once, every time the kcopyd thread wakes up. This prevents kcopyd from hogging the CPU indefinitely and causing workqueue stalls. Re-running the aforementioned test: * Workqueue stalls are eliminated * The maximum writing time among all targets is reduced from 09m37.10s to 06m04.85s and the total run time of the test is reduced from 10m43.591s to 7m19.199s [1] https://github.com/jthornber/device-mapper-test-suite Signed-off-by: Nikos Tsironis <ntsironis@arrikto.com> Signed-off-by: Ilias Tsitsimpis <iliastsi@arrikto.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2018-11-01 05:53:09 +08:00
unsigned long flags;
/*
* The order that these are called is *very* important.
* complete jobs can free some pages for pages jobs.
* Pages jobs when successful will jump onto the io jobs
* list. io jobs call wake when they complete and it all
* starts again.
*/
dm kcopyd: Fix bug causing workqueue stalls When using kcopyd to run callbacks through dm_kcopyd_do_callback() or submitting copy jobs with a source size of 0, the jobs are pushed directly to the complete_jobs list, which could be under processing by the kcopyd thread. As a result, the kcopyd thread can continue running completed jobs indefinitely, without releasing the CPU, as long as someone keeps submitting new completed jobs through the aforementioned paths. Processing of work items, queued for execution on the same CPU as the currently running kcopyd thread, is thus stalled for excessive amounts of time, hurting performance. Running the following test, from the device mapper test suite [1], dmtest run --suite snapshot -n parallel_io_to_many_snaps_N , with 8 active snapshots, we get, in dmesg, messages like the following: [68899.948523] BUG: workqueue lockup - pool cpus=0 node=0 flags=0x0 nice=0 stuck for 95s! [68899.949282] Showing busy workqueues and worker pools: [68899.949288] workqueue events: flags=0x0 [68899.949295] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=2/256 [68899.949306] pending: vmstat_shepherd, cache_reap [68899.949331] workqueue mm_percpu_wq: flags=0x8 [68899.949337] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=1/256 [68899.949345] pending: vmstat_update [68899.949387] workqueue dm_bufio_cache: flags=0x8 [68899.949392] pwq 4: cpus=2 node=0 flags=0x0 nice=0 active=1/256 [68899.949400] pending: work_fn [dm_bufio] [68899.949423] workqueue kcopyd: flags=0x8 [68899.949429] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=1/256 [68899.949437] pending: do_work [dm_mod] [68899.949452] workqueue kcopyd: flags=0x8 [68899.949458] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=2/256 [68899.949466] in-flight: 13:do_work [dm_mod] [68899.949474] pending: do_work [dm_mod] [68899.949487] workqueue kcopyd: flags=0x8 [68899.949493] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=1/256 [68899.949501] pending: do_work [dm_mod] [68899.949515] workqueue kcopyd: flags=0x8 [68899.949521] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=1/256 [68899.949529] pending: do_work [dm_mod] [68899.949541] workqueue kcopyd: flags=0x8 [68899.949547] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=1/256 [68899.949555] pending: do_work [dm_mod] [68899.949568] pool 0: cpus=0 node=0 flags=0x0 nice=0 hung=95s workers=4 idle: 27130 27223 1084 Fix this by splitting the complete_jobs list into two parts: A user facing part, named callback_jobs, and one used internally by kcopyd, retaining the name complete_jobs. dm_kcopyd_do_callback() and dispatch_job() now push their jobs to the callback_jobs list, which is spliced to the complete_jobs list once, every time the kcopyd thread wakes up. This prevents kcopyd from hogging the CPU indefinitely and causing workqueue stalls. Re-running the aforementioned test: * Workqueue stalls are eliminated * The maximum writing time among all targets is reduced from 09m37.10s to 06m04.85s and the total run time of the test is reduced from 10m43.591s to 7m19.199s [1] https://github.com/jthornber/device-mapper-test-suite Signed-off-by: Nikos Tsironis <ntsironis@arrikto.com> Signed-off-by: Ilias Tsitsimpis <iliastsi@arrikto.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2018-11-01 05:53:09 +08:00
spin_lock_irqsave(&kc->job_lock, flags);
list_splice_tail_init(&kc->callback_jobs, &kc->complete_jobs);
spin_unlock_irqrestore(&kc->job_lock, flags);
blk_start_plug(&plug);
process_jobs(&kc->complete_jobs, kc, run_complete_job);
process_jobs(&kc->pages_jobs, kc, run_pages_job);
process_jobs(&kc->io_jobs, kc, run_io_job);
blk_finish_plug(&plug);
}
/*
* If we are copying a small region we just dispatch a single job
* to do the copy, otherwise the io has to be split up into many
* jobs.
*/
static void dispatch_job(struct kcopyd_job *job)
{
struct dm_kcopyd_client *kc = job->kc;
atomic_inc(&kc->nr_jobs);
if (unlikely(!job->source.count))
dm kcopyd: Fix bug causing workqueue stalls When using kcopyd to run callbacks through dm_kcopyd_do_callback() or submitting copy jobs with a source size of 0, the jobs are pushed directly to the complete_jobs list, which could be under processing by the kcopyd thread. As a result, the kcopyd thread can continue running completed jobs indefinitely, without releasing the CPU, as long as someone keeps submitting new completed jobs through the aforementioned paths. Processing of work items, queued for execution on the same CPU as the currently running kcopyd thread, is thus stalled for excessive amounts of time, hurting performance. Running the following test, from the device mapper test suite [1], dmtest run --suite snapshot -n parallel_io_to_many_snaps_N , with 8 active snapshots, we get, in dmesg, messages like the following: [68899.948523] BUG: workqueue lockup - pool cpus=0 node=0 flags=0x0 nice=0 stuck for 95s! [68899.949282] Showing busy workqueues and worker pools: [68899.949288] workqueue events: flags=0x0 [68899.949295] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=2/256 [68899.949306] pending: vmstat_shepherd, cache_reap [68899.949331] workqueue mm_percpu_wq: flags=0x8 [68899.949337] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=1/256 [68899.949345] pending: vmstat_update [68899.949387] workqueue dm_bufio_cache: flags=0x8 [68899.949392] pwq 4: cpus=2 node=0 flags=0x0 nice=0 active=1/256 [68899.949400] pending: work_fn [dm_bufio] [68899.949423] workqueue kcopyd: flags=0x8 [68899.949429] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=1/256 [68899.949437] pending: do_work [dm_mod] [68899.949452] workqueue kcopyd: flags=0x8 [68899.949458] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=2/256 [68899.949466] in-flight: 13:do_work [dm_mod] [68899.949474] pending: do_work [dm_mod] [68899.949487] workqueue kcopyd: flags=0x8 [68899.949493] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=1/256 [68899.949501] pending: do_work [dm_mod] [68899.949515] workqueue kcopyd: flags=0x8 [68899.949521] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=1/256 [68899.949529] pending: do_work [dm_mod] [68899.949541] workqueue kcopyd: flags=0x8 [68899.949547] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=1/256 [68899.949555] pending: do_work [dm_mod] [68899.949568] pool 0: cpus=0 node=0 flags=0x0 nice=0 hung=95s workers=4 idle: 27130 27223 1084 Fix this by splitting the complete_jobs list into two parts: A user facing part, named callback_jobs, and one used internally by kcopyd, retaining the name complete_jobs. dm_kcopyd_do_callback() and dispatch_job() now push their jobs to the callback_jobs list, which is spliced to the complete_jobs list once, every time the kcopyd thread wakes up. This prevents kcopyd from hogging the CPU indefinitely and causing workqueue stalls. Re-running the aforementioned test: * Workqueue stalls are eliminated * The maximum writing time among all targets is reduced from 09m37.10s to 06m04.85s and the total run time of the test is reduced from 10m43.591s to 7m19.199s [1] https://github.com/jthornber/device-mapper-test-suite Signed-off-by: Nikos Tsironis <ntsironis@arrikto.com> Signed-off-by: Ilias Tsitsimpis <iliastsi@arrikto.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2018-11-01 05:53:09 +08:00
push(&kc->callback_jobs, job);
else if (job->pages == &zero_page_list)
push(&kc->io_jobs, job);
else
push(&kc->pages_jobs, job);
wake(kc);
}
static void segment_complete(int read_err, unsigned long write_err,
void *context)
{
/* FIXME: tidy this function */
sector_t progress = 0;
sector_t count = 0;
struct kcopyd_job *sub_job = (struct kcopyd_job *) context;
struct kcopyd_job *job = sub_job->master_job;
struct dm_kcopyd_client *kc = job->kc;
mutex_lock(&job->lock);
/* update the error */
if (read_err)
job->read_err = 1;
if (write_err)
job->write_err |= write_err;
/*
* Only dispatch more work if there hasn't been an error.
*/
if ((!job->read_err && !job->write_err) ||
test_bit(DM_KCOPYD_IGNORE_ERROR, &job->flags)) {
/* get the next chunk of work */
progress = job->progress;
count = job->source.count - progress;
if (count) {
dm kcopyd: Increase default sub-job size to 512KB Currently, kcopyd has a sub-job size of 64KB and a maximum number of 8 sub-jobs. As a result, for any kcopyd job, we have a maximum of 512KB of I/O in flight. This upper limit to the amount of in-flight I/O under-utilizes fast devices and results in decreased throughput, e.g., when writing to a snapshotted thin LV with I/O size less than the pool's block size (so COW is performed using kcopyd). Increase kcopyd's default sub-job size to 512KB, so we have a maximum of 4MB of I/O in flight for each kcopyd job. This results in an up to 96% improvement of bandwidth when writing to a snapshotted thin LV, with I/O sizes less than the pool's block size. Also, add dm_mod.kcopyd_subjob_size_kb module parameter to allow users to fine tune the sub-job size of kcopyd. The default value of this parameter is 512KB and the maximum allowed value is 1024KB. We evaluate the performance impact of the change by running the snap_breaking_throughput benchmark, from the device mapper test suite [1]. The benchmark: 1. Creates a 1G thin LV 2. Provisions the thin LV 3. Takes a snapshot of the thin LV 4. Writes to the thin LV with: dd if=/dev/zero of=/dev/vg/thin_lv oflag=direct bs=<I/O size> Running this benchmark with various thin pool block sizes and dd I/O sizes (all combinations triggering the use of kcopyd) we get the following results: +-----------------+-------------+------------------+-----------------+ | Pool block size | dd I/O size | BW before (MB/s) | BW after (MB/s) | +-----------------+-------------+------------------+-----------------+ | 1 MB | 256 KB | 242 | 280 | | 1 MB | 512 KB | 238 | 295 | | | | | | | 2 MB | 256 KB | 238 | 354 | | 2 MB | 512 KB | 241 | 380 | | 2 MB | 1 MB | 245 | 394 | | | | | | | 4 MB | 256 KB | 248 | 412 | | 4 MB | 512 KB | 234 | 432 | | 4 MB | 1 MB | 251 | 474 | | 4 MB | 2 MB | 257 | 504 | | | | | | | 8 MB | 256 KB | 239 | 420 | | 8 MB | 512 KB | 256 | 431 | | 8 MB | 1 MB | 264 | 467 | | 8 MB | 2 MB | 264 | 502 | | 8 MB | 4 MB | 281 | 537 | +-----------------+-------------+------------------+-----------------+ [1] https://github.com/jthornber/device-mapper-test-suite Signed-off-by: Nikos Tsironis <ntsironis@arrikto.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2019-07-17 19:24:10 +08:00
if (count > kc->sub_job_size)
count = kc->sub_job_size;
job->progress += count;
}
}
mutex_unlock(&job->lock);
if (count) {
int i;
*sub_job = *job;
sub_job->write_offset = progress;
sub_job->source.sector += progress;
sub_job->source.count = count;
for (i = 0; i < job->num_dests; i++) {
sub_job->dests[i].sector += progress;
sub_job->dests[i].count = count;
}
sub_job->fn = segment_complete;
sub_job->context = sub_job;
dispatch_job(sub_job);
} else if (atomic_dec_and_test(&job->sub_jobs)) {
/*
* Queue the completion callback to the kcopyd thread.
*
* Some callers assume that all the completions are called
* from a single thread and don't race with each other.
*
* We must not call the callback directly here because this
* code may not be executing in the thread.
*/
push(&kc->complete_jobs, job);
wake(kc);
}
}
/*
* Create some sub jobs to share the work between them.
*/
static void split_job(struct kcopyd_job *master_job)
{
int i;
atomic_inc(&master_job->kc->nr_jobs);
atomic_set(&master_job->sub_jobs, SPLIT_COUNT);
for (i = 0; i < SPLIT_COUNT; i++) {
master_job[i + 1].master_job = master_job;
segment_complete(0, 0u, &master_job[i + 1]);
}
}
void dm_kcopyd_copy(struct dm_kcopyd_client *kc, struct dm_io_region *from,
unsigned int num_dests, struct dm_io_region *dests,
unsigned int flags, dm_kcopyd_notify_fn fn, void *context)
{
struct kcopyd_job *job;
int i;
/*
* Allocate an array of jobs consisting of one master job
* followed by SPLIT_COUNT sub jobs.
*/
job = mempool_alloc(&kc->job_pool, GFP_NOIO);
mutex_init(&job->lock);
/*
* set up for the read.
*/
job->kc = kc;
job->flags = flags;
job->read_err = 0;
job->write_err = 0;
job->num_dests = num_dests;
memcpy(&job->dests, dests, sizeof(*dests) * num_dests);
/*
* If one of the destination is a host-managed zoned block device,
* we need to write sequentially. If one of the destination is a
* host-aware device, then leave it to the caller to choose what to do.
*/
if (!test_bit(DM_KCOPYD_WRITE_SEQ, &job->flags)) {
for (i = 0; i < job->num_dests; i++) {
if (bdev_zoned_model(dests[i].bdev) == BLK_ZONED_HM) {
set_bit(DM_KCOPYD_WRITE_SEQ, &job->flags);
break;
}
}
}
/*
* If we need to write sequentially, errors cannot be ignored.
*/
if (test_bit(DM_KCOPYD_WRITE_SEQ, &job->flags) &&
test_bit(DM_KCOPYD_IGNORE_ERROR, &job->flags))
clear_bit(DM_KCOPYD_IGNORE_ERROR, &job->flags);
if (from) {
job->source = *from;
job->pages = NULL;
job->rw = READ;
} else {
memset(&job->source, 0, sizeof job->source);
job->source.count = job->dests[0].count;
job->pages = &zero_page_list;
/*
* Use WRITE ZEROES to optimize zeroing if all dests support it.
*/
job->rw = REQ_OP_WRITE_ZEROES;
for (i = 0; i < job->num_dests; i++)
if (!bdev_write_zeroes_sectors(job->dests[i].bdev)) {
job->rw = WRITE;
break;
}
}
job->fn = fn;
job->context = context;
job->master_job = job;
job->write_offset = 0;
dm kcopyd: Increase default sub-job size to 512KB Currently, kcopyd has a sub-job size of 64KB and a maximum number of 8 sub-jobs. As a result, for any kcopyd job, we have a maximum of 512KB of I/O in flight. This upper limit to the amount of in-flight I/O under-utilizes fast devices and results in decreased throughput, e.g., when writing to a snapshotted thin LV with I/O size less than the pool's block size (so COW is performed using kcopyd). Increase kcopyd's default sub-job size to 512KB, so we have a maximum of 4MB of I/O in flight for each kcopyd job. This results in an up to 96% improvement of bandwidth when writing to a snapshotted thin LV, with I/O sizes less than the pool's block size. Also, add dm_mod.kcopyd_subjob_size_kb module parameter to allow users to fine tune the sub-job size of kcopyd. The default value of this parameter is 512KB and the maximum allowed value is 1024KB. We evaluate the performance impact of the change by running the snap_breaking_throughput benchmark, from the device mapper test suite [1]. The benchmark: 1. Creates a 1G thin LV 2. Provisions the thin LV 3. Takes a snapshot of the thin LV 4. Writes to the thin LV with: dd if=/dev/zero of=/dev/vg/thin_lv oflag=direct bs=<I/O size> Running this benchmark with various thin pool block sizes and dd I/O sizes (all combinations triggering the use of kcopyd) we get the following results: +-----------------+-------------+------------------+-----------------+ | Pool block size | dd I/O size | BW before (MB/s) | BW after (MB/s) | +-----------------+-------------+------------------+-----------------+ | 1 MB | 256 KB | 242 | 280 | | 1 MB | 512 KB | 238 | 295 | | | | | | | 2 MB | 256 KB | 238 | 354 | | 2 MB | 512 KB | 241 | 380 | | 2 MB | 1 MB | 245 | 394 | | | | | | | 4 MB | 256 KB | 248 | 412 | | 4 MB | 512 KB | 234 | 432 | | 4 MB | 1 MB | 251 | 474 | | 4 MB | 2 MB | 257 | 504 | | | | | | | 8 MB | 256 KB | 239 | 420 | | 8 MB | 512 KB | 256 | 431 | | 8 MB | 1 MB | 264 | 467 | | 8 MB | 2 MB | 264 | 502 | | 8 MB | 4 MB | 281 | 537 | +-----------------+-------------+------------------+-----------------+ [1] https://github.com/jthornber/device-mapper-test-suite Signed-off-by: Nikos Tsironis <ntsironis@arrikto.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2019-07-17 19:24:10 +08:00
if (job->source.count <= kc->sub_job_size)
dispatch_job(job);
else {
job->progress = 0;
split_job(job);
}
}
EXPORT_SYMBOL(dm_kcopyd_copy);
void dm_kcopyd_zero(struct dm_kcopyd_client *kc,
unsigned num_dests, struct dm_io_region *dests,
unsigned flags, dm_kcopyd_notify_fn fn, void *context)
{
dm_kcopyd_copy(kc, NULL, num_dests, dests, flags, fn, context);
}
EXPORT_SYMBOL(dm_kcopyd_zero);
void *dm_kcopyd_prepare_callback(struct dm_kcopyd_client *kc,
dm_kcopyd_notify_fn fn, void *context)
{
struct kcopyd_job *job;
job = mempool_alloc(&kc->job_pool, GFP_NOIO);
memset(job, 0, sizeof(struct kcopyd_job));
job->kc = kc;
job->fn = fn;
job->context = context;
job->master_job = job;
atomic_inc(&kc->nr_jobs);
return job;
}
EXPORT_SYMBOL(dm_kcopyd_prepare_callback);
void dm_kcopyd_do_callback(void *j, int read_err, unsigned long write_err)
{
struct kcopyd_job *job = j;
struct dm_kcopyd_client *kc = job->kc;
job->read_err = read_err;
job->write_err = write_err;
dm kcopyd: Fix bug causing workqueue stalls When using kcopyd to run callbacks through dm_kcopyd_do_callback() or submitting copy jobs with a source size of 0, the jobs are pushed directly to the complete_jobs list, which could be under processing by the kcopyd thread. As a result, the kcopyd thread can continue running completed jobs indefinitely, without releasing the CPU, as long as someone keeps submitting new completed jobs through the aforementioned paths. Processing of work items, queued for execution on the same CPU as the currently running kcopyd thread, is thus stalled for excessive amounts of time, hurting performance. Running the following test, from the device mapper test suite [1], dmtest run --suite snapshot -n parallel_io_to_many_snaps_N , with 8 active snapshots, we get, in dmesg, messages like the following: [68899.948523] BUG: workqueue lockup - pool cpus=0 node=0 flags=0x0 nice=0 stuck for 95s! [68899.949282] Showing busy workqueues and worker pools: [68899.949288] workqueue events: flags=0x0 [68899.949295] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=2/256 [68899.949306] pending: vmstat_shepherd, cache_reap [68899.949331] workqueue mm_percpu_wq: flags=0x8 [68899.949337] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=1/256 [68899.949345] pending: vmstat_update [68899.949387] workqueue dm_bufio_cache: flags=0x8 [68899.949392] pwq 4: cpus=2 node=0 flags=0x0 nice=0 active=1/256 [68899.949400] pending: work_fn [dm_bufio] [68899.949423] workqueue kcopyd: flags=0x8 [68899.949429] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=1/256 [68899.949437] pending: do_work [dm_mod] [68899.949452] workqueue kcopyd: flags=0x8 [68899.949458] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=2/256 [68899.949466] in-flight: 13:do_work [dm_mod] [68899.949474] pending: do_work [dm_mod] [68899.949487] workqueue kcopyd: flags=0x8 [68899.949493] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=1/256 [68899.949501] pending: do_work [dm_mod] [68899.949515] workqueue kcopyd: flags=0x8 [68899.949521] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=1/256 [68899.949529] pending: do_work [dm_mod] [68899.949541] workqueue kcopyd: flags=0x8 [68899.949547] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=1/256 [68899.949555] pending: do_work [dm_mod] [68899.949568] pool 0: cpus=0 node=0 flags=0x0 nice=0 hung=95s workers=4 idle: 27130 27223 1084 Fix this by splitting the complete_jobs list into two parts: A user facing part, named callback_jobs, and one used internally by kcopyd, retaining the name complete_jobs. dm_kcopyd_do_callback() and dispatch_job() now push their jobs to the callback_jobs list, which is spliced to the complete_jobs list once, every time the kcopyd thread wakes up. This prevents kcopyd from hogging the CPU indefinitely and causing workqueue stalls. Re-running the aforementioned test: * Workqueue stalls are eliminated * The maximum writing time among all targets is reduced from 09m37.10s to 06m04.85s and the total run time of the test is reduced from 10m43.591s to 7m19.199s [1] https://github.com/jthornber/device-mapper-test-suite Signed-off-by: Nikos Tsironis <ntsironis@arrikto.com> Signed-off-by: Ilias Tsitsimpis <iliastsi@arrikto.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2018-11-01 05:53:09 +08:00
push(&kc->callback_jobs, job);
wake(kc);
}
EXPORT_SYMBOL(dm_kcopyd_do_callback);
/*
* Cancels a kcopyd job, eg. someone might be deactivating a
* mirror.
*/
#if 0
int kcopyd_cancel(struct kcopyd_job *job, int block)
{
/* FIXME: finish */
return -1;
}
#endif /* 0 */
/*-----------------------------------------------------------------
* Client setup
*---------------------------------------------------------------*/
struct dm_kcopyd_client *dm_kcopyd_client_create(struct dm_kcopyd_throttle *throttle)
{
int r;
dm kcopyd: Increase default sub-job size to 512KB Currently, kcopyd has a sub-job size of 64KB and a maximum number of 8 sub-jobs. As a result, for any kcopyd job, we have a maximum of 512KB of I/O in flight. This upper limit to the amount of in-flight I/O under-utilizes fast devices and results in decreased throughput, e.g., when writing to a snapshotted thin LV with I/O size less than the pool's block size (so COW is performed using kcopyd). Increase kcopyd's default sub-job size to 512KB, so we have a maximum of 4MB of I/O in flight for each kcopyd job. This results in an up to 96% improvement of bandwidth when writing to a snapshotted thin LV, with I/O sizes less than the pool's block size. Also, add dm_mod.kcopyd_subjob_size_kb module parameter to allow users to fine tune the sub-job size of kcopyd. The default value of this parameter is 512KB and the maximum allowed value is 1024KB. We evaluate the performance impact of the change by running the snap_breaking_throughput benchmark, from the device mapper test suite [1]. The benchmark: 1. Creates a 1G thin LV 2. Provisions the thin LV 3. Takes a snapshot of the thin LV 4. Writes to the thin LV with: dd if=/dev/zero of=/dev/vg/thin_lv oflag=direct bs=<I/O size> Running this benchmark with various thin pool block sizes and dd I/O sizes (all combinations triggering the use of kcopyd) we get the following results: +-----------------+-------------+------------------+-----------------+ | Pool block size | dd I/O size | BW before (MB/s) | BW after (MB/s) | +-----------------+-------------+------------------+-----------------+ | 1 MB | 256 KB | 242 | 280 | | 1 MB | 512 KB | 238 | 295 | | | | | | | 2 MB | 256 KB | 238 | 354 | | 2 MB | 512 KB | 241 | 380 | | 2 MB | 1 MB | 245 | 394 | | | | | | | 4 MB | 256 KB | 248 | 412 | | 4 MB | 512 KB | 234 | 432 | | 4 MB | 1 MB | 251 | 474 | | 4 MB | 2 MB | 257 | 504 | | | | | | | 8 MB | 256 KB | 239 | 420 | | 8 MB | 512 KB | 256 | 431 | | 8 MB | 1 MB | 264 | 467 | | 8 MB | 2 MB | 264 | 502 | | 8 MB | 4 MB | 281 | 537 | +-----------------+-------------+------------------+-----------------+ [1] https://github.com/jthornber/device-mapper-test-suite Signed-off-by: Nikos Tsironis <ntsironis@arrikto.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2019-07-17 19:24:10 +08:00
unsigned reserve_pages;
struct dm_kcopyd_client *kc;
kc = kzalloc(sizeof(*kc), GFP_KERNEL);
if (!kc)
return ERR_PTR(-ENOMEM);
spin_lock_init(&kc->job_lock);
dm kcopyd: Fix bug causing workqueue stalls When using kcopyd to run callbacks through dm_kcopyd_do_callback() or submitting copy jobs with a source size of 0, the jobs are pushed directly to the complete_jobs list, which could be under processing by the kcopyd thread. As a result, the kcopyd thread can continue running completed jobs indefinitely, without releasing the CPU, as long as someone keeps submitting new completed jobs through the aforementioned paths. Processing of work items, queued for execution on the same CPU as the currently running kcopyd thread, is thus stalled for excessive amounts of time, hurting performance. Running the following test, from the device mapper test suite [1], dmtest run --suite snapshot -n parallel_io_to_many_snaps_N , with 8 active snapshots, we get, in dmesg, messages like the following: [68899.948523] BUG: workqueue lockup - pool cpus=0 node=0 flags=0x0 nice=0 stuck for 95s! [68899.949282] Showing busy workqueues and worker pools: [68899.949288] workqueue events: flags=0x0 [68899.949295] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=2/256 [68899.949306] pending: vmstat_shepherd, cache_reap [68899.949331] workqueue mm_percpu_wq: flags=0x8 [68899.949337] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=1/256 [68899.949345] pending: vmstat_update [68899.949387] workqueue dm_bufio_cache: flags=0x8 [68899.949392] pwq 4: cpus=2 node=0 flags=0x0 nice=0 active=1/256 [68899.949400] pending: work_fn [dm_bufio] [68899.949423] workqueue kcopyd: flags=0x8 [68899.949429] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=1/256 [68899.949437] pending: do_work [dm_mod] [68899.949452] workqueue kcopyd: flags=0x8 [68899.949458] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=2/256 [68899.949466] in-flight: 13:do_work [dm_mod] [68899.949474] pending: do_work [dm_mod] [68899.949487] workqueue kcopyd: flags=0x8 [68899.949493] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=1/256 [68899.949501] pending: do_work [dm_mod] [68899.949515] workqueue kcopyd: flags=0x8 [68899.949521] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=1/256 [68899.949529] pending: do_work [dm_mod] [68899.949541] workqueue kcopyd: flags=0x8 [68899.949547] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=1/256 [68899.949555] pending: do_work [dm_mod] [68899.949568] pool 0: cpus=0 node=0 flags=0x0 nice=0 hung=95s workers=4 idle: 27130 27223 1084 Fix this by splitting the complete_jobs list into two parts: A user facing part, named callback_jobs, and one used internally by kcopyd, retaining the name complete_jobs. dm_kcopyd_do_callback() and dispatch_job() now push their jobs to the callback_jobs list, which is spliced to the complete_jobs list once, every time the kcopyd thread wakes up. This prevents kcopyd from hogging the CPU indefinitely and causing workqueue stalls. Re-running the aforementioned test: * Workqueue stalls are eliminated * The maximum writing time among all targets is reduced from 09m37.10s to 06m04.85s and the total run time of the test is reduced from 10m43.591s to 7m19.199s [1] https://github.com/jthornber/device-mapper-test-suite Signed-off-by: Nikos Tsironis <ntsironis@arrikto.com> Signed-off-by: Ilias Tsitsimpis <iliastsi@arrikto.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2018-11-01 05:53:09 +08:00
INIT_LIST_HEAD(&kc->callback_jobs);
INIT_LIST_HEAD(&kc->complete_jobs);
INIT_LIST_HEAD(&kc->io_jobs);
INIT_LIST_HEAD(&kc->pages_jobs);
kc->throttle = throttle;
r = mempool_init_slab_pool(&kc->job_pool, MIN_JOBS, _job_cache);
if (r)
goto bad_slab;
INIT_WORK(&kc->kcopyd_work, do_work);
kc->kcopyd_wq = alloc_workqueue("kcopyd", WQ_MEM_RECLAIM, 0);
if (!kc->kcopyd_wq) {
r = -ENOMEM;
goto bad_workqueue;
}
dm kcopyd: Increase default sub-job size to 512KB Currently, kcopyd has a sub-job size of 64KB and a maximum number of 8 sub-jobs. As a result, for any kcopyd job, we have a maximum of 512KB of I/O in flight. This upper limit to the amount of in-flight I/O under-utilizes fast devices and results in decreased throughput, e.g., when writing to a snapshotted thin LV with I/O size less than the pool's block size (so COW is performed using kcopyd). Increase kcopyd's default sub-job size to 512KB, so we have a maximum of 4MB of I/O in flight for each kcopyd job. This results in an up to 96% improvement of bandwidth when writing to a snapshotted thin LV, with I/O sizes less than the pool's block size. Also, add dm_mod.kcopyd_subjob_size_kb module parameter to allow users to fine tune the sub-job size of kcopyd. The default value of this parameter is 512KB and the maximum allowed value is 1024KB. We evaluate the performance impact of the change by running the snap_breaking_throughput benchmark, from the device mapper test suite [1]. The benchmark: 1. Creates a 1G thin LV 2. Provisions the thin LV 3. Takes a snapshot of the thin LV 4. Writes to the thin LV with: dd if=/dev/zero of=/dev/vg/thin_lv oflag=direct bs=<I/O size> Running this benchmark with various thin pool block sizes and dd I/O sizes (all combinations triggering the use of kcopyd) we get the following results: +-----------------+-------------+------------------+-----------------+ | Pool block size | dd I/O size | BW before (MB/s) | BW after (MB/s) | +-----------------+-------------+------------------+-----------------+ | 1 MB | 256 KB | 242 | 280 | | 1 MB | 512 KB | 238 | 295 | | | | | | | 2 MB | 256 KB | 238 | 354 | | 2 MB | 512 KB | 241 | 380 | | 2 MB | 1 MB | 245 | 394 | | | | | | | 4 MB | 256 KB | 248 | 412 | | 4 MB | 512 KB | 234 | 432 | | 4 MB | 1 MB | 251 | 474 | | 4 MB | 2 MB | 257 | 504 | | | | | | | 8 MB | 256 KB | 239 | 420 | | 8 MB | 512 KB | 256 | 431 | | 8 MB | 1 MB | 264 | 467 | | 8 MB | 2 MB | 264 | 502 | | 8 MB | 4 MB | 281 | 537 | +-----------------+-------------+------------------+-----------------+ [1] https://github.com/jthornber/device-mapper-test-suite Signed-off-by: Nikos Tsironis <ntsironis@arrikto.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2019-07-17 19:24:10 +08:00
kc->sub_job_size = dm_get_kcopyd_subjob_size();
reserve_pages = DIV_ROUND_UP(kc->sub_job_size << SECTOR_SHIFT, PAGE_SIZE);
kc->pages = NULL;
kc->nr_reserved_pages = kc->nr_free_pages = 0;
dm kcopyd: Increase default sub-job size to 512KB Currently, kcopyd has a sub-job size of 64KB and a maximum number of 8 sub-jobs. As a result, for any kcopyd job, we have a maximum of 512KB of I/O in flight. This upper limit to the amount of in-flight I/O under-utilizes fast devices and results in decreased throughput, e.g., when writing to a snapshotted thin LV with I/O size less than the pool's block size (so COW is performed using kcopyd). Increase kcopyd's default sub-job size to 512KB, so we have a maximum of 4MB of I/O in flight for each kcopyd job. This results in an up to 96% improvement of bandwidth when writing to a snapshotted thin LV, with I/O sizes less than the pool's block size. Also, add dm_mod.kcopyd_subjob_size_kb module parameter to allow users to fine tune the sub-job size of kcopyd. The default value of this parameter is 512KB and the maximum allowed value is 1024KB. We evaluate the performance impact of the change by running the snap_breaking_throughput benchmark, from the device mapper test suite [1]. The benchmark: 1. Creates a 1G thin LV 2. Provisions the thin LV 3. Takes a snapshot of the thin LV 4. Writes to the thin LV with: dd if=/dev/zero of=/dev/vg/thin_lv oflag=direct bs=<I/O size> Running this benchmark with various thin pool block sizes and dd I/O sizes (all combinations triggering the use of kcopyd) we get the following results: +-----------------+-------------+------------------+-----------------+ | Pool block size | dd I/O size | BW before (MB/s) | BW after (MB/s) | +-----------------+-------------+------------------+-----------------+ | 1 MB | 256 KB | 242 | 280 | | 1 MB | 512 KB | 238 | 295 | | | | | | | 2 MB | 256 KB | 238 | 354 | | 2 MB | 512 KB | 241 | 380 | | 2 MB | 1 MB | 245 | 394 | | | | | | | 4 MB | 256 KB | 248 | 412 | | 4 MB | 512 KB | 234 | 432 | | 4 MB | 1 MB | 251 | 474 | | 4 MB | 2 MB | 257 | 504 | | | | | | | 8 MB | 256 KB | 239 | 420 | | 8 MB | 512 KB | 256 | 431 | | 8 MB | 1 MB | 264 | 467 | | 8 MB | 2 MB | 264 | 502 | | 8 MB | 4 MB | 281 | 537 | +-----------------+-------------+------------------+-----------------+ [1] https://github.com/jthornber/device-mapper-test-suite Signed-off-by: Nikos Tsironis <ntsironis@arrikto.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2019-07-17 19:24:10 +08:00
r = client_reserve_pages(kc, reserve_pages);
if (r)
goto bad_client_pages;
kc->io_client = dm_io_client_create();
if (IS_ERR(kc->io_client)) {
r = PTR_ERR(kc->io_client);
goto bad_io_client;
}
init_waitqueue_head(&kc->destroyq);
atomic_set(&kc->nr_jobs, 0);
return kc;
bad_io_client:
client_free_pages(kc);
bad_client_pages:
destroy_workqueue(kc->kcopyd_wq);
bad_workqueue:
mempool_exit(&kc->job_pool);
bad_slab:
kfree(kc);
return ERR_PTR(r);
}
EXPORT_SYMBOL(dm_kcopyd_client_create);
void dm_kcopyd_client_destroy(struct dm_kcopyd_client *kc)
{
/* Wait for completion of all jobs submitted by this client. */
wait_event(kc->destroyq, !atomic_read(&kc->nr_jobs));
dm kcopyd: Fix bug causing workqueue stalls When using kcopyd to run callbacks through dm_kcopyd_do_callback() or submitting copy jobs with a source size of 0, the jobs are pushed directly to the complete_jobs list, which could be under processing by the kcopyd thread. As a result, the kcopyd thread can continue running completed jobs indefinitely, without releasing the CPU, as long as someone keeps submitting new completed jobs through the aforementioned paths. Processing of work items, queued for execution on the same CPU as the currently running kcopyd thread, is thus stalled for excessive amounts of time, hurting performance. Running the following test, from the device mapper test suite [1], dmtest run --suite snapshot -n parallel_io_to_many_snaps_N , with 8 active snapshots, we get, in dmesg, messages like the following: [68899.948523] BUG: workqueue lockup - pool cpus=0 node=0 flags=0x0 nice=0 stuck for 95s! [68899.949282] Showing busy workqueues and worker pools: [68899.949288] workqueue events: flags=0x0 [68899.949295] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=2/256 [68899.949306] pending: vmstat_shepherd, cache_reap [68899.949331] workqueue mm_percpu_wq: flags=0x8 [68899.949337] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=1/256 [68899.949345] pending: vmstat_update [68899.949387] workqueue dm_bufio_cache: flags=0x8 [68899.949392] pwq 4: cpus=2 node=0 flags=0x0 nice=0 active=1/256 [68899.949400] pending: work_fn [dm_bufio] [68899.949423] workqueue kcopyd: flags=0x8 [68899.949429] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=1/256 [68899.949437] pending: do_work [dm_mod] [68899.949452] workqueue kcopyd: flags=0x8 [68899.949458] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=2/256 [68899.949466] in-flight: 13:do_work [dm_mod] [68899.949474] pending: do_work [dm_mod] [68899.949487] workqueue kcopyd: flags=0x8 [68899.949493] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=1/256 [68899.949501] pending: do_work [dm_mod] [68899.949515] workqueue kcopyd: flags=0x8 [68899.949521] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=1/256 [68899.949529] pending: do_work [dm_mod] [68899.949541] workqueue kcopyd: flags=0x8 [68899.949547] pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=1/256 [68899.949555] pending: do_work [dm_mod] [68899.949568] pool 0: cpus=0 node=0 flags=0x0 nice=0 hung=95s workers=4 idle: 27130 27223 1084 Fix this by splitting the complete_jobs list into two parts: A user facing part, named callback_jobs, and one used internally by kcopyd, retaining the name complete_jobs. dm_kcopyd_do_callback() and dispatch_job() now push their jobs to the callback_jobs list, which is spliced to the complete_jobs list once, every time the kcopyd thread wakes up. This prevents kcopyd from hogging the CPU indefinitely and causing workqueue stalls. Re-running the aforementioned test: * Workqueue stalls are eliminated * The maximum writing time among all targets is reduced from 09m37.10s to 06m04.85s and the total run time of the test is reduced from 10m43.591s to 7m19.199s [1] https://github.com/jthornber/device-mapper-test-suite Signed-off-by: Nikos Tsironis <ntsironis@arrikto.com> Signed-off-by: Ilias Tsitsimpis <iliastsi@arrikto.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2018-11-01 05:53:09 +08:00
BUG_ON(!list_empty(&kc->callback_jobs));
BUG_ON(!list_empty(&kc->complete_jobs));
BUG_ON(!list_empty(&kc->io_jobs));
BUG_ON(!list_empty(&kc->pages_jobs));
destroy_workqueue(kc->kcopyd_wq);
dm_io_client_destroy(kc->io_client);
client_free_pages(kc);
mempool_exit(&kc->job_pool);
kfree(kc);
}
EXPORT_SYMBOL(dm_kcopyd_client_destroy);