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>
kcopyd has no upper limit to the number of jobs one can allocate and
issue. Under certain workloads this can lead to excessive memory usage
and workqueue stalls. For example, when creating multiple dm-snapshot
targets with a 4K chunk size and then writing to the origin through the
page cache. Syncing the page cache causes a large number of BIOs to be
issued to the dm-snapshot origin target, which itself issues an even
larger (because of the BIO splitting taking place) number of kcopyd
jobs.
Running the following test, from the device mapper test suite [1],
dmtest run --suite snapshot -n many_snapshots_of_same_volume_N
, with 8 active snapshots, results in the kcopyd job slab cache growing
to 10G. Depending on the available system RAM this can lead to the OOM
killer killing user processes:
[463.492878] kthreadd invoked oom-killer: gfp_mask=0x6040c0(GFP_KERNEL|__GFP_COMP),
nodemask=(null), order=1, oom_score_adj=0
[463.492894] kthreadd cpuset=/ mems_allowed=0
[463.492948] CPU: 7 PID: 2 Comm: kthreadd Not tainted 4.19.0-rc7 #3
[463.492950] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.10.2-1 04/01/2014
[463.492952] Call Trace:
[463.492964] dump_stack+0x7d/0xbb
[463.492973] dump_header+0x6b/0x2fc
[463.492987] ? lockdep_hardirqs_on+0xee/0x190
[463.493012] oom_kill_process+0x302/0x370
[463.493021] out_of_memory+0x113/0x560
[463.493030] __alloc_pages_slowpath+0xf40/0x1020
[463.493055] __alloc_pages_nodemask+0x348/0x3c0
[463.493067] cache_grow_begin+0x81/0x8b0
[463.493072] ? cache_grow_begin+0x874/0x8b0
[463.493078] fallback_alloc+0x1e4/0x280
[463.493092] kmem_cache_alloc_node+0xd6/0x370
[463.493098] ? copy_process.part.31+0x1c5/0x20d0
[463.493105] copy_process.part.31+0x1c5/0x20d0
[463.493115] ? __lock_acquire+0x3cc/0x1550
[463.493121] ? __switch_to_asm+0x34/0x70
[463.493129] ? kthread_create_worker_on_cpu+0x70/0x70
[463.493135] ? finish_task_switch+0x90/0x280
[463.493165] _do_fork+0xe0/0x6d0
[463.493191] ? kthreadd+0x19f/0x220
[463.493233] kernel_thread+0x25/0x30
[463.493235] kthreadd+0x1bf/0x220
[463.493242] ? kthread_create_on_cpu+0x90/0x90
[463.493248] ret_from_fork+0x3a/0x50
[463.493279] Mem-Info:
[463.493285] active_anon:20631 inactive_anon:4831 isolated_anon:0
[463.493285] active_file:80216 inactive_file:80107 isolated_file:435
[463.493285] unevictable:0 dirty:51266 writeback:109372 unstable:0
[463.493285] slab_reclaimable:31191 slab_unreclaimable:3483521
[463.493285] mapped:526 shmem:4903 pagetables:1759 bounce:0
[463.493285] free:33623 free_pcp:2392 free_cma:0
...
[463.493489] Unreclaimable slab info:
[463.493513] Name Used Total
[463.493522] bio-6 1028KB 1028KB
[463.493525] bio-5 1028KB 1028KB
[463.493528] dm_snap_pending_exception 236783KB 243789KB
[463.493531] dm_exception 41KB 42KB
[463.493534] bio-4 1216KB 1216KB
[463.493537] bio-3 439396KB 439396KB
[463.493539] kcopyd_job 6973427KB 6973427KB
...
[463.494340] Out of memory: Kill process 1298 (ruby2.3) score 1 or sacrifice child
[463.494673] Killed process 1298 (ruby2.3) total-vm:435740kB, anon-rss:20180kB, file-rss:4kB, shmem-rss:0kB
[463.506437] oom_reaper: reaped process 1298 (ruby2.3), now anon-rss:0kB, file-rss:0kB, shmem-rss:0kB
Moreover, issuing a large number of kcopyd jobs results in kcopyd
hogging the CPU, while processing them. As a result, processing of work
items, queued for execution on the same CPU as the currently running
kcopyd thread, is stalled for long periods of time, hurting performance.
Running the aforementioned test we get, in dmesg, messages like the
following:
[67501.194592] BUG: workqueue lockup - pool cpus=4 node=0 flags=0x0 nice=0 stuck for 27s!
[67501.195586] Showing busy workqueues and worker pools:
[67501.195591] workqueue events: flags=0x0
[67501.195597] pwq 8: cpus=4 node=0 flags=0x0 nice=0 active=1/256
[67501.195611] pending: cache_reap
[67501.195641] workqueue mm_percpu_wq: flags=0x8
[67501.195645] pwq 8: cpus=4 node=0 flags=0x0 nice=0 active=1/256
[67501.195656] pending: vmstat_update
[67501.195682] workqueue kblockd: flags=0x18
[67501.195687] pwq 5: cpus=2 node=0 flags=0x0 nice=-20 active=1/256
[67501.195698] pending: blk_timeout_work
[67501.195753] workqueue kcopyd: flags=0x8
[67501.195757] pwq 8: cpus=4 node=0 flags=0x0 nice=0 active=1/256
[67501.195768] pending: do_work [dm_mod]
[67501.195802] workqueue kcopyd: flags=0x8
[67501.195806] pwq 8: cpus=4 node=0 flags=0x0 nice=0 active=1/256
[67501.195817] pending: do_work [dm_mod]
[67501.195834] workqueue kcopyd: flags=0x8
[67501.195838] pwq 8: cpus=4 node=0 flags=0x0 nice=0 active=1/256
[67501.195848] pending: do_work [dm_mod]
[67501.195881] workqueue kcopyd: flags=0x8
[67501.195885] pwq 8: cpus=4 node=0 flags=0x0 nice=0 active=1/256
[67501.195896] pending: do_work [dm_mod]
[67501.195920] workqueue kcopyd: flags=0x8
[67501.195924] pwq 8: cpus=4 node=0 flags=0x0 nice=0 active=2/256
[67501.195935] in-flight: 67:do_work [dm_mod]
[67501.195945] pending: do_work [dm_mod]
[67501.195961] pool 8: cpus=4 node=0 flags=0x0 nice=0 hung=27s workers=3 idle: 129 23765
The root cause for these issues is the way dm-snapshot uses kcopyd. In
particular, the lack of an explicit or implicit limit to the maximum
number of in-flight COW jobs. The merging path is not affected because
it implicitly limits the in-flight kcopyd jobs to one.
Fix these issues by using a semaphore to limit the maximum number of
in-flight kcopyd jobs. We grab the semaphore before allocating a new
kcopyd job in start_copy() and start_full_bio() and release it after the
job finishes in copy_callback().
The initial semaphore value is configurable through a module parameter,
to allow fine tuning the maximum number of in-flight COW jobs. Setting
this parameter to zero initializes the semaphore to INT_MAX.
A default value of 2048 maximum in-flight kcopyd jobs was chosen. This
value was decided experimentally as a trade-off between memory
consumption, stalling the kernel's workqueues and maintaining a high
enough throughput.
Re-running the aforementioned test:
* Workqueue stalls are eliminated
* kcopyd's job slab cache uses a maximum of 130MB
* The time taken by the test to write to the snapshot-origin target is
reduced from 05m20.48s to 03m26.38s
[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>
* Hashtable has been replaced by rbtree to manage buffers.
Update the comment.
* Fix typo in the comment for dm_bufio_issue_flush
Signed-off-by: Shenghui Wang <shhuiw@foxmail.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
The error msg should be "flush thread" instead of "endio thread"
for writecache_flush_thread.
Signed-off-by: Shenghui Wang <shhuiw@foxmail.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
The workqueues are shared by many multipath devices, only flush whole
workqueue when necessary. Otherwise, we just flush works as needed.
Signed-off-by: wuzhouhui <wuzhouhui14@mails.ucas.ac.cn>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Indirect calls are inefficient because of retpolines that are used for
spectre workaround. This patch replaces an indirect call with a condition
(that can be predicted by the branch predictor).
Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
The queue mapping of type poll only exists when set->map[HCTX_TYPE_POLL].nr_queues
is bigger than zero, so enhance the constraint by checking .nr_queues of type poll
before enabling IO poll.
Otherwise IO race & timeout can be observed when running block/007.
Cc: Jeff Moyer <jmoyer@redhat.com>
Cc: Christoph Hellwig <hch@lst.de>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
There's a single user of this function, dm, and dm just wants
to check if IO is inflight, not that it's just allocated.
This fixes a hang with srp/002 in blktests with dm, where it tries
to suspend but waits for inflight IO to finish first. As it checks
for just allocated requests, this fails.
Tested-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
From 7e849dd9cf ("nvme-pci: don't share queue maps"), the mapping
table won't be initialized actually if map->nr_queues is zero, so
we can't use blk_mq_map_queue_type() to retrieve hctx any more.
This way still may cause broken mapping, fix it by skipping zero-queues
maps in blk_mq_map_swqueue().
Cc: Jeff Moyer <jmoyer@redhat.com>
Cc: Mike Snitzer <snitzer@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The blk-iolatency controller measures the time from rq_qos_throttle() to
rq_qos_done_bio() and attributes this time to the first bio that needs
to create the request. This means if a bio is plug-mergeable or
bio-mergeable, it gets to bypass the blk-iolatency controller.
The recent series [1], to tag all bios w/ blkgs undermined how iolatency
was determining which bios it was charging and should process in
rq_qos_done_bio(). Because all bios are being tagged, this caused the
atomic_t for the struct rq_wait inflight count to underflow and result
in a stall.
This patch adds a new flag BIO_TRACKED to let controllers know that a
bio is going through the rq_qos path. blk-iolatency now checks if this
flag is set to see if it should process the bio in rq_qos_done_bio().
Overloading BLK_QUEUE_ENTERED works, but makes the flag rules confusing.
BIO_THROTTLED was another candidate, but the flag is set for all bios
that have gone through blk-throttle code. Overloading a flag comes with
the burden of making sure that when either implementation changes, a
change in setting rules for one doesn't cause a bug in the other. So
here, we unfortunately opt for adding a new flag.
[1] https://lore.kernel.org/lkml/20181205171039.73066-1-dennis@kernel.org/
Fixes: 5cdf2e3fea ("blkcg: associate blkg when associating a device")
Signed-off-by: Dennis Zhou <dennis@kernel.org>
Cc: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
When a request is added to rq list of sw queue(ctx), the rq may be from
a different type of hctx, especially after multi queue mapping is
introduced.
So when dispach request from sw queue via blk_mq_flush_busy_ctxs() or
blk_mq_dequeue_from_ctx(), one request belonging to other queue type of
hctx can be dispatched to current hctx in case that read queue or poll
queue is enabled.
This patch fixes this issue by introducing per-queue-type list.
Cc: Christoph Hellwig <hch@lst.de>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Changed by me to not use separately cacheline aligned lists, just
place them all in the same cacheline where we had just the one list
and lock before.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
For a zoned block device using mq-deadline, if a write request for a
zone is received while another write was already dispatched for the same
zone, dd_dispatch_request() will return NULL and the newly inserted
write request is kept in the scheduler queue waiting for the ongoing
zone write to complete. With this behavior, when no other request has
been dispatched, rq_list in blk_mq_sched_dispatch_requests() is empty
and blk_mq_sched_mark_restart_hctx() not called. This in turn leads to
__blk_mq_free_request() call of blk_mq_sched_restart() to not run the
queue when the already dispatched write request completes. The newly
dispatched request stays stuck in the scheduler queue until eventually
another request is submitted.
This problem does not affect SCSI disk as the SCSI stack handles queue
restart on request completion. However, this problem is can be triggered
the nullblk driver with zoned mode enabled.
Fix this by always requesting a queue restart in dd_dispatch_request()
if no request was dispatched while WRITE requests are queued.
Fixes: 5700f69178 ("mq-deadline: Introduce zone locking support")
Cc: <stable@vger.kernel.org>
Signed-off-by: Damien Le Moal <damien.lemoal@wdc.com>
Add missing export of blk_mq_sched_restart()
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Now that the block layer checks if a queue map has any queues inside
it there is no more reason to duplicate the maps for the non-default
types.
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
We should check if a given queue map actually has queues enabled before
dispatching to it. This allows drivers to not initialize optional but
not used map types, which subsequently will allow fixing problems with
queue map rebuilds for that case.
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Now we only export hctx->type via sysfs, and there isn't such info
in hctx entry under debugfs. We often use debugfs only to diagnose
queue mapping issue, so add the support in debugfs.
Queue mapping becomes a bit more complicated after multiple queue
mapping is supported, we may write blktest to verify if queue mapping
is valid based on blk-mq-debugfs.
Given not necessary to export hctx->type twice, so remove the export
from sysfs.
Cc: Jeff Moyer <jmoyer@redhat.com>
Cc: Mike Snitzer <snitzer@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Type of each element in queue mapping table is 'unsigned int,
intead of 'struct blk_mq_queue_map)', so fix it.
Cc: Jeff Moyer <jmoyer@redhat.com>
Cc: Mike Snitzer <snitzer@redhat.com>
Cc: Christoph Hellwig <hch@lst.de>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
This information is helpful to either investigate issues, or understand
wbt's internal behaviour.
Cc: Bart Van Assche <bart.vanassche@wdc.com>
Cc: Omar Sandoval <osandov@fb.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
blk-mq-debugfs has been proved as very helpful for debug some
tough issues, such as IO hang.
We have seen blk-wbt related IO hang several times, even inside
Red Hat BZ, there is such report not sovled yet, so this patch
adds support debugfs on rq_qos.
Cc: Bart Van Assche <bart.vanassche@wdc.com>
Cc: Omar Sandoval <osandov@fb.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Add the description of the zoned, nr_zones and chunk_sectors sysfs queue
attributes to Documentation/block/queue-sysfs.txt. The description of
the zoned and chunk_sector attributes are mostly copied from
ABI/testing/sysfs-block (added a typo fix). While at it, also fix a
typo in the description of the io_poll_delay attribute.
nr_zones description is also added to ABI/testing/sysfs-block and
contact email address updated for the zoned attribute.
Signed-off-by: Damien Le Moal <damien.lemoal@wdc.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
blk_mq_init_queue() will not return NULL pointer to its caller,
so it's better to replace IS_ERR_OR_NULL using IS_ERR in loop_add().
If in the future things change to check NULL pointer inside loop_add(),
we should return -ENOMEM as return code instead of PTR_ERR(NULL).
Signed-off-by: Chengguang Xu <cgxu519@gmx.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Add __exit annotation to cleanup helper which
is only called once in the module.
Signed-off-by: Chengguang Xu <cgxu519@gmx.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
This prevents a HIPRI bio from being submitted through a stacking
driver that does not support polling and thus won't poll for I/O
completion.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Replace blk_mq_request_issue_directly with blk_mq_try_issue_directly
in blk_insert_cloned_request and kill it as nobody uses it any more.
Signed-off-by: Jianchao Wang <jianchao.w.wang@oracle.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
It is not necessary to issue request directly with bypass 'true'
in blk_mq_sched_insert_requests and handle the non-issued requests
itself. Just set bypass to 'false' and let blk_mq_try_issue_directly
handle them totally. Remove the blk_rq_can_direct_dispatch check,
because blk_mq_try_issue_directly can handle it well.If request is
direct-issued unsuccessfully, insert the reset.
Signed-off-by: Jianchao Wang <jianchao.w.wang@oracle.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Merge blk_mq_try_issue_directly and __blk_mq_try_issue_directly
into one interface to unify the interfaces to issue requests
directly. The merged interface takes over the requests totally,
it could insert, end or do nothing based on the return value of
.queue_rq and 'bypass' parameter. Then caller needn't any other
handling any more and then code could be cleaned up.
And also the commit c616cbee ( blk-mq: punt failed direct issue
to dispatch list ) always inserts requests to hctx dispatch list
whenever get a BLK_STS_RESOURCE or BLK_STS_DEV_RESOURCE, this is
overkill and will harm the merging. We just need to do that for
the requests that has been through .queue_rq. This patch also
could fix this.
Signed-off-by: Jianchao Wang <jianchao.w.wang@oracle.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Pull NVMe updates from Christoph:
"Here is the second large chunk of nvme updates for 4.21:
- host and target support for NVMe over TCP (Sagi Grimberg,
Roy Shterman, Solganik Alexander)
- error log page support in target (Chaitanya Kulkarni)
plus small fixes and improvements from Jens Axboe and Chengguang Xu."
* 'nvme-4.21' of git://git.infradead.org/nvme: (33 commits)
nvme-rdma: support separate queue maps for read and write
nvme-tcp: support separate queue maps for read and write
nvme-fabrics: allow user to set nr_write_queues for separate queue maps
nvme-fabrics: add missing nvmf_ctrl_options documentation
blk-mq-rdma: pass in queue map to blk_mq_rdma_map_queues
nvmet: update smart log with num err log entries
nvmet: add error log page cmd handler
nvmet: add error log support for file backend
nvmet: add error log support for bdev backend
nvmet: add error log support for admin-cmd
nvmet: add error log support for rdma backend
nvmet: add error log support for fabrics-cmd
nvmet: add error log support in the core
nvmet: add interface to update error-log page
nvmet: add error-log definitions
nvme: add error log page slot definition
nvme: remove nvme_common command cdw10 array
nvmet: remove unused variable
nvme: provide fallback for discard alloc failure
nvme: add __exit annotation
...
Sometimes flush journal may be very frequent, so it's useful to dump
number of keys every time write journal.
Signed-off-by: Guoju Fang <fangguoju@gmail.com>
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Because CUTOFF_WRITEBACK is defined as 40, so before the changes of
dynamic cutoff writeback values, writeback_percent is limited to [0,
CUTOFF_WRITEBACK]. Any value larger than CUTOFF_WRITEBACK will be fixed
up to 40.
Now cutof writeback limit is a dynamic value bch_cutoff_writeback, so
the range of writeback_percent can be a more flexible range as [0,
bch_cutoff_writeback]. The flexibility is, it can be expended to a
larger or smaller range than [0, 40], depends on how value
bch_cutoff_writeback is specified.
The default value is still strongly recommended to most of users for
most of workloads. But for people who want to do research on bcache
writeback perforamnce tuning, they may have chance to specify more
flexible writeback_percent in range [0, 70].
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Currently the cutoff writeback and cutoff writeback sync thresholds are
defined by CUTOFF_WRITEBACK (40) and CUTOFF_WRITEBACK_SYNC (70) as
static values. Most of time these they work fine, but when people want
to do research on bcache writeback mode performance tuning, there is no
chance to modify the soft and hard cutoff writeback values.
This patch introduces two module parameters bch_cutoff_writeback_sync
and bch_cutoff_writeback which permit people to tune the values when
loading bcache.ko. If they are not specified by module loading, current
values CUTOFF_WRITEBACK_SYNC and CUTOFF_WRITEBACK will be used as
default and nothing changes.
When people want to tune this two values,
- cutoff_writeback can be set in range [1, 70]
- cutoff_writeback_sync can be set in range [1, 90]
- cutoff_writeback always <= cutoff_writeback_sync
The default values are strongly recommended to most of users for most of
workloads. Anyway, if people wants to take their own risk to do research
on new writeback cutoff tuning for their own workload, now they can make
it.
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
This patch moves MODULE_AUTHOR and MODULE_LICENSE to end of super.c, and
add MODULE_DESCRIPTION("Bcache: a Linux block layer cache").
This is preparation for adding module parameters.
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The option gc_after_writeback is disabled by default, because garbage
collection will discard SSD data which drops cached data.
Echo 1 into /sys/fs/bcache/<UUID>/internal/gc_after_writeback will
enable this option, which wakes up gc thread when writeback accomplished
and all cached data is clean.
This option is helpful for people who cares writing performance more. In
heavy writing workload, all cached data can be clean only happens when
writeback thread cleans all cached data in I/O idle time. In such
situation a following gc running may help to shrink bcache B+ tree and
discard more clean data, which may be helpful for future writing
requests.
If you are not sure whether this is helpful for your own workload,
please leave it as disabled by default.
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Garbage collection thread starts to work when c->sectors_to_gc is
negative value, otherwise nothing will happen even the gc thread is
woken up by wake_up_gc().
force_wake_up_gc() sets c->sectors_to_gc to -1 before calling
wake_up_gc(), then gc thread may have chance to run if no one else sets
c->sectors_to_gc to a positive value before gc_should_run().
This routine can be called where the gc thread is woken up and required
to run in force.
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
"echo 1 > writeback_running" marks writeback_running even if no
writeback kthread created as "d_strtoul(writeback_running)" will simply
set dc-> writeback_running without checking the existence of
dc->writeback_thread.
Add check for setting writeback_running via sysfs: if no writeback
kthread available, reject setting to 1.
v2 -> v3:
* Make message on wrong assignment more clear.
* Print name of bcache device instead of name of backing device.
Signed-off-by: Shenghui Wang <shhuiw@foxmail.com>
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
A fresh backing device is not attached to any cache_set, and
has no writeback kthread created until first attached to some
cache_set.
But bch_cached_dev_writeback_init run
"
dc->writeback_running = true;
WARN_ON(test_and_clear_bit(BCACHE_DEV_WB_RUNNING,
&dc->disk.flags));
"
for any newly formatted backing devices.
For a fresh standalone backing device, we can get something like
following even if no writeback kthread created:
------------------------
/sys/block/bcache0/bcache# cat writeback_running
1
/sys/block/bcache0/bcache# cat writeback_rate_debug
rate: 512.0k/sec
dirty: 0.0k
target: 0.0k
proportional: 0.0k
integral: 0.0k
change: 0.0k/sec
next io: -15427384ms
The none ZERO fields are misleading as no alive writeback kthread yet.
Set dc->writeback_running false as no writeback thread created in
bch_cached_dev_writeback_init().
We have writeback thread created and woken up in bch_cached_dev_writeback
_start(). Set dc->writeback_running true before bch_writeback_queue()
called, as a writeback thread will check if dc->writeback_running is true
before writing back dirty data, and hung if false detected.
After the change, we can get the following output for a fresh standalone
backing device:
-----------------------
/sys/block/bcache0/bcache$ cat writeback_running
0
/sys/block/bcache0/bcache# cat writeback_rate_debug
rate: 0.0k/sec
dirty: 0.0k
target: 0.0k
proportional: 0.0k
integral: 0.0k
change: 0.0k/sec
next io: 0ms
v1 -> v2:
Set dc->writeback_running before bch_writeback_queue() called,
Signed-off-by: Shenghui Wang <shhuiw@foxmail.com>
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
We have struct cached_dev allocated by kzalloc in register_bcache(),
which initializes all the fields of cached_dev with 0s. And commit
ce4c3e19e5 ("bcache: Replace bch_read_string_list() by
__sysfs_match_string()") has remove the string "default".
Update the comment.
Signed-off-by: Shenghui Wang <shhuiw@foxmail.com>
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
commit 220bb38c21 ("bcache: Break up struct search") introduced
changes to struct search and s->iop. bypass/bio are fields of struct
data_insert_op now. Update the comment.
Signed-off-by: Shenghui Wang <shhuiw@foxmail.com>
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
debugfs_remove and debugfs_remove_recursive will check if the dentry
pointer is NULL or ERR, and will do nothing in that case.
Remove the check in cache_set_free and bch_debug_init.
Signed-off-by: Shenghui Wang <shhuiw@foxmail.com>
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
We have the following define for btree iterator:
struct btree_iter {
size_t size, used;
#ifdef CONFIG_BCACHE_DEBUG
struct btree_keys *b;
#endif
struct btree_iter_set {
struct bkey *k, *end;
} data[MAX_BSETS];
};
We can see that the length of data[] field is static MAX_BSETS, which is
defined as 4 currently.
But a btree node on disk could have too many bsets for an iterator to fit
on the stack - maybe far more that MAX_BSETS. Have to dynamically allocate
space to host more btree_iter_sets.
bch_cache_set_alloc() will make sure the pool cache_set->fill_iter can
allocate an iterator equipped with enough room that can host
(sb.bucket_size / sb.block_size)
btree_iter_sets, which is more than static MAX_BSETS.
bch_btree_node_read_done() will use that pool to allocate one iterator, to
host many bsets in one btree node.
Add more comment around cache_set->fill_iter to make code less confusing.
Signed-off-by: Shenghui Wang <shhuiw@foxmail.com>
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
llow NVMF_OPT_NR_WRITE_QUEUES to describe additional write queues. In
addition, implement .map_queues that will apply 2 queue maps for read
and write queue sets.
Note that with the separate queue map, HCTX_TYPE_READ will always use
nr_io_queues and HCTX_TYPE_DEFAULT will use nr_write_queues.
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Allow NVMF_OPT_NR_WRITE_QUEUES to describe additional write queues. In
addition, implement .map_queues that will apply 2 queue maps for read
and write queue sets.
Note that with the separate queue map, HCTX_TYPE_READ will always use
nr_io_queues and HCTX_TYPE_DEFAULT will use nr_write_queues.
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Christoph Hellwig <hch@lst.de>
This argument will specify how many I/O queues will be connected in
create_ctrl in addition to nr_io_queues. With this configuration, I/O
that carries payload from the host to the target, will use the default
hctx queue map, and I/O that involves target to host transfers will use
the read hctx queue map.
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Christoph Hellwig <hch@lst.de>
