openwrt/staging/blogic.git
7 years agonvme-fabrics: Allow ctrl loss timeout configuration
Sagi Grimberg [Sat, 18 Mar 2017 18:52:36 +0000 (20:52 +0200)]
nvme-fabrics: Allow ctrl loss timeout configuration

When a host sense that its controller session is damaged,
it tries to re-establish it periodically (reconnect every
reconnect_delay). It may very well be that the controller
is gone and never coming back, in this case the host will
try to reconnect forever.

Add a ctrl_loss_tmo to bound the number of reconnect attempts
to a specific controller (default to a reasonable 10 minutes).
The timeout configuration is actually translated into number of
reconnect attempts and not a schedule on its own but rather
divided with reconnect_delay. This is useful to prevent
racing flows of remove and reconnect, and it doesn't really
matter if we remove slightly sooner than what the user requested.

Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agonvme-rdma: get rid of local reconnect_delay
Sagi Grimberg [Sat, 18 Mar 2017 21:47:22 +0000 (23:47 +0200)]
nvme-rdma: get rid of local reconnect_delay

we already have it in opts.

Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agonvme-loop: retrieve iod from the cqe command_id
Sagi Grimberg [Mon, 27 Feb 2017 16:28:25 +0000 (18:28 +0200)]
nvme-loop: retrieve iod from the cqe command_id

useful to validate that the we didn't mess up
the command_id.

Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agonvme-loop: remove unneeded includes
Sagi Grimberg [Sun, 19 Mar 2017 04:32:09 +0000 (06:32 +0200)]
nvme-loop: remove unneeded includes

Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agonvme-fc: fix module_init (theoretical) error path
Sagi Grimberg [Sun, 19 Mar 2017 12:16:05 +0000 (14:16 +0200)]
nvme-fc: fix module_init (theoretical) error path

If nvmf_register_transport happened to fail
(it can't, but theoretically) we leak memory.

Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agonvme-loop: fix module_init (theoretical) error path
Sagi Grimberg [Sun, 19 Mar 2017 04:26:28 +0000 (06:26 +0200)]
nvme-loop: fix module_init (theoretical) error path

if nvmf_register_transport happend to fail, we
need to nvmet_unregister_transport as well.

Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agonvme-rdma: fix module_init (theoretical) error path
Sagi Grimberg [Sun, 19 Mar 2017 04:21:42 +0000 (06:21 +0200)]
nvme-rdma: fix module_init (theoretical) error path

If nvmf_register_transport happened to fail
(it can't, but theoretically) we leak memory.

Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agonvmet: use symbolic constants for log identifiers
Max Gurtovoy [Sun, 5 Mar 2017 22:30:38 +0000 (00:30 +0200)]
nvmet: use symbolic constants for log identifiers

Signed-off-by: Max Gurtovoy <maxg@mellanox.com>
Reviewed-by: Parav Pandit <parav@mellanox.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agonvmet: Introduced helper routine for controller status check.
Parav Pandit [Tue, 28 Feb 2017 05:21:33 +0000 (23:21 -0600)]
nvmet: Introduced helper routine for controller status check.

This patch introduces helper function for checking controller
status during admin and io command processing which returns u16
status. As to bring consistency on returning status, other
friend functions also now return u16 status instead of int
to match the spec.

As part of the theseerror log prints in also prints qid on
which command error occured.

Signed-off-by: Parav Pandit <parav@mellanox.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agonvmet: Fixed avoided printing nvmet: twice in error logs.
Parav Pandit [Tue, 28 Feb 2017 05:21:02 +0000 (23:21 -0600)]
nvmet: Fixed avoided printing nvmet: twice in error logs.

This patch avoids printing "nvmet:" twice in error logs as its already
coming through pr_fmt macro.

Reviewed-by: Max Gurtovoy <maxg@mellanox.com>
Signed-off-by: Parav Pandit <parav@mellanox.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agoiscsi-target: use generic inet_pton_with_scope
Sagi Grimberg [Mon, 6 Feb 2017 10:51:03 +0000 (12:51 +0200)]
iscsi-target: use generic inet_pton_with_scope

Instead of parsing address strings, use a generic
helper.

Acked-by: Nicholas Bellinger <nab@linux-iscsi.org>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agonvme-rdma: use inet_pton_with_scope helper
Sagi Grimberg [Sun, 5 Feb 2017 19:49:32 +0000 (21:49 +0200)]
nvme-rdma: use inet_pton_with_scope helper

Both the destination and the host addresses are now
parsed using inet_pton_with_scope helper. We also
get ipv6 (with address scopes support).

Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agonvmet-rdma: use generic inet_pton_with_scope
Sagi Grimberg [Sun, 5 Feb 2017 19:49:04 +0000 (21:49 +0200)]
nvmet-rdma: use generic inet_pton_with_scope

Instead of parsing address strings, use a generic
helper. This also adds ipv6 (with address scopes)
support.

Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agonet/utils: generic inet_pton_with_scope helper
Sagi Grimberg [Sun, 5 Feb 2017 19:47:22 +0000 (21:47 +0200)]
net/utils: generic inet_pton_with_scope helper

Several locations in the stack need to handle ipv4/ipv6
(with scope) and port strings conversion to sockaddr.
Add a helper that takes either AF_INET, AF_INET6 or
AF_UNSPEC (for wildcard) to centralize this handling.

Suggested-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Acked-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agonvme-loop: remove some code duplication
Sagi Grimberg [Mon, 13 Mar 2017 13:43:44 +0000 (15:43 +0200)]
nvme-loop: remove some code duplication

Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agonvme-rdma: Give some more grace for rdma connection establishment
Sagi Grimberg [Tue, 21 Mar 2017 14:32:38 +0000 (16:32 +0200)]
nvme-rdma: Give some more grace for rdma connection establishment

The target might be occupied with multiple hosts so lets
give it some more grace before failing the connection
establishment.

Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agonvmet-rdma: occasionally flush ongoing controller teardown
Sagi Grimberg [Tue, 21 Mar 2017 14:29:49 +0000 (16:29 +0200)]
nvmet-rdma: occasionally flush ongoing controller teardown

If we are attacked with establishments/teradowns we need to
make sure we do not consume too much system memory. Thus
let ongoing controller teardowns complete before accepting
new controller establishments.

Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agonvme-rdma: handle cpu unplug when re-establishing the controller
Sagi Grimberg [Thu, 9 Mar 2017 11:26:07 +0000 (13:26 +0200)]
nvme-rdma: handle cpu unplug when re-establishing the controller

If a cpu unplug event has occured, we need to take the minimum
of the provided nr_io_queues and the number of online cpus,
otherwise we won't be able to connect them as blk-mq mapping
won't dispatch to those queues.

Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agonvmet-rdma: Fix a possible uninitialized variable dereference
Sagi Grimberg [Thu, 9 Mar 2017 11:45:52 +0000 (13:45 +0200)]
nvmet-rdma: Fix a possible uninitialized variable dereference

When handling a new recv command, we grab a new rsp resource and
check for the queue state being live. In case the queue is not in
live state, we simply restore the rsp back to the free list. However
in this flow we didn't set rsp->queue yet, so we cannot dereference it.

Instead, make sure to initialize rsp->queue (and other rsp members)
as soon as possible so we won't reference uninitialized variables.

Reported-by: Yi Zhang <yizhan@redhat.com>
Reported-by: Raju Rangoju <rajur@chelsio.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Tested-by: Raju Rangoju <rajur@chelsio.com>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agonvmet: confirm sq percpu has scheduled and switched to atomic
Sagi Grimberg [Mon, 6 Mar 2017 16:46:20 +0000 (18:46 +0200)]
nvmet: confirm sq percpu has scheduled and switched to atomic

percpu_ref_kill is not enough to prevent subsequent
percpu_ref_tryget_live from failing. Hence call
perfcpu_ref_kill_confirm to make it safe.

Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agonvme-loop: handle cpu unplug when re-establishing the controller
Sagi Grimberg [Mon, 13 Mar 2017 11:27:51 +0000 (13:27 +0200)]
nvme-loop: handle cpu unplug when re-establishing the controller

If a cpu unplug event has occured, we need to take the minimum
of the provided nr_io_queues and the number of online cpus,
otherwise we won't be able to connect them as blk-mq mapping
won't dispatch to those queues.

Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
7 years agonvme-loop: fix a possible use-after-free when destroying the admin queue
Sagi Grimberg [Mon, 27 Feb 2017 16:44:45 +0000 (18:44 +0200)]
nvme-loop: fix a possible use-after-free when destroying the admin queue

we need to destroy the nvmet sq and let it finish gracefully
before continue to cleanup the queue.

Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
7 years agoblk-mq: constify struct blk_mq_ops
Eric Biggers [Thu, 30 Mar 2017 20:39:16 +0000 (13:39 -0700)]
blk-mq: constify struct blk_mq_ops

Constify all instances of blk_mq_ops, as they are never modified.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agonull_blk: add blocking mode
Jens Axboe [Thu, 30 Mar 2017 19:44:26 +0000 (13:44 -0600)]
null_blk: add blocking mode

This adds a new module parameter to null_blk, blocking. If set, null_blk
will set the BLK_MQ_F_BLOCKING flag, indicating that it sometimes/always
needs to block in its ->queue_rq() function.  The intent is to help find
regressions in blocking drivers, since not many of them exist.

If null_blk is loaded with submit_queues > 1 and blocking=1, this
shows the regression recently fixed by bf4907c05e61.

Reviewed-by: Bart Van Assche <bart.vanassche@sandisk.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agoblk-mq: fix schedule-under-preempt for blocking drivers
Jens Axboe [Thu, 30 Mar 2017 18:30:39 +0000 (12:30 -0600)]
blk-mq: fix schedule-under-preempt for blocking drivers

Commit a4d907b6a33b unified the single and multi queue request handlers,
but in the process, it also screwed up the locking balance and calls
blk_mq_try_issue_directly() with the ctx preempt lock held. This is a
problem for drivers that have set BLK_MQ_F_BLOCKING, since now they
can't reliably sleep.

While in there, protect against similar issues in the future, by adding
a might_sleep() trigger in the BLOCKING path for direct issue or queue
run.

Reported-by: Josef Bacik <josef@toxicpanda.com>
Tested-by: Josef Bacik <josef@toxicpanda.com>
Fixes: a4d907b6a33b ("blk-mq: streamline blk_mq_make_request")
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agoblock/sed-opal: fix spelling mistake: "Lifcycle" -> "Lifecycle"
Colin Ian King [Thu, 30 Mar 2017 09:58:08 +0000 (10:58 +0100)]
block/sed-opal: fix spelling mistake: "Lifcycle" -> "Lifecycle"

trivial fix to spelling mistake in pr_err error message

Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agoblock: do not put mq context in blk_mq_alloc_request_hctx
Minchan Kim [Thu, 30 Mar 2017 05:20:45 +0000 (14:20 +0900)]
block: do not put mq context in blk_mq_alloc_request_hctx

In blk_mq_alloc_request_hctx, blk_mq_sched_get_request doesn't
get sw context so we don't need to put the context with
blk_mq_put_ctx. Unless, we will see preempt counter underflow.

Cc: Omar Sandoval <osandov@fb.com>
Signed-off-by: Minchan Kim <minchan@kernel.org>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agoblk-mq: include errors in did_work calculation
Jens Axboe [Fri, 24 Mar 2017 18:04:19 +0000 (12:04 -0600)]
blk-mq: include errors in did_work calculation

Currently we return true in blk_mq_dispatch_rq_list() if we queued IO
successfully, but we really want to return whether or not the we made
progress. Progress includes if we got an error return.  If we don't,
this can lead to a hang in blk_mq_sched_dispatch_requests() when a
driver is draining IO by returning BLK_MQ_QUEUE_ERROR instead of
manually ending the IO in error and return BLK_MQ_QUEUE_OK.

Tested-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: Bart Van Assche <bart.vanassche@sandisk.com>
Reviewed-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agoblock-mq: don't re-queue if we get a queue error
Josef Bacik [Tue, 28 Mar 2017 20:37:52 +0000 (16:37 -0400)]
block-mq: don't re-queue if we get a queue error

When try to issue a request directly and we fail we will requeue the
request, but call blk_mq_end_request() as well.  This leads to the
completed request being on a queuelist and getting ended twice, which
causes list corruption in schedulers and other shenanigans.

Signed-off-by: Josef Bacik <jbacik@fb.com>
Reviewed-by: Ming Lei <tom.leiming@gmail.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agoblkcg: allocate struct blkcg_gq outside request queue spinlock
Tahsin Erdogan [Wed, 29 Mar 2017 17:27:19 +0000 (11:27 -0600)]
blkcg: allocate struct blkcg_gq outside request queue spinlock

blkg_conf_prep() currently calls blkg_lookup_create() while holding
request queue spinlock. This means allocating memory for struct
blkcg_gq has to be made non-blocking. This causes occasional -ENOMEM
failures in call paths like below:

  pcpu_alloc+0x68f/0x710
  __alloc_percpu_gfp+0xd/0x10
  __percpu_counter_init+0x55/0xc0
  cfq_pd_alloc+0x3b2/0x4e0
  blkg_alloc+0x187/0x230
  blkg_create+0x489/0x670
  blkg_lookup_create+0x9a/0x230
  blkg_conf_prep+0x1fb/0x240
  __cfqg_set_weight_device.isra.105+0x5c/0x180
  cfq_set_weight_on_dfl+0x69/0xc0
  cgroup_file_write+0x39/0x1c0
  kernfs_fop_write+0x13f/0x1d0
  __vfs_write+0x23/0x120
  vfs_write+0xc2/0x1f0
  SyS_write+0x44/0xb0
  entry_SYSCALL_64_fastpath+0x18/0xad

In the code path above, percpu allocator cannot call vmalloc() due to
queue spinlock.

A failure in this call path gives grief to tools which are trying to
configure io weights. We see occasional failures happen shortly after
reboots even when system is not under any memory pressure. Machines
with a lot of cpus are more vulnerable to this condition.

Do struct blkcg_gq allocations outside the queue spinlock to allow
blocking during memory allocations.

Suggested-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Tahsin Erdogan <tahsin@google.com>
Acked-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agoRevert "blkcg: allocate struct blkcg_gq outside request queue spinlock"
Jens Axboe [Wed, 29 Mar 2017 17:25:48 +0000 (11:25 -0600)]
Revert "blkcg: allocate struct blkcg_gq outside request queue spinlock"

I inadvertently applied the v5 version of this patch, whereas
the agreed upon version was v5. Revert this one so we can apply
the right one.

This reverts commit 7fc6b87a9ff537e7df32b1278118ce9c5bcd6788.

7 years agoblk-mq: fix a typo and a spelling mistake
Jens Axboe [Wed, 29 Mar 2017 17:10:34 +0000 (11:10 -0600)]
blk-mq: fix a typo and a spelling mistake

Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agoblk-mq-pci: Fix two spelling mistakes
Sagi Grimberg [Wed, 29 Mar 2017 17:04:36 +0000 (20:04 +0300)]
blk-mq-pci: Fix two spelling mistakes

Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agoblock: fix leak of q->rq_wb
Omar Sandoval [Tue, 28 Mar 2017 23:12:17 +0000 (16:12 -0700)]
block: fix leak of q->rq_wb

CONFIG_DEBUG_TEST_DRIVER_REMOVE found a possible leak of q->rq_wb when a
request queue is reregistered. This has been a problem since wbt was
introduced, but the WARN_ON(!list_empty(&stats->callbacks)) in the
blk-stat rework exposed it. Fix it by cleaning up wbt when we unregister
the queue.

Fixes: 87760e5eef35 ("block: hook up writeback throttling")
Signed-off-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agoblk-mq: fix leak of q->stats
Omar Sandoval [Tue, 28 Mar 2017 23:12:16 +0000 (16:12 -0700)]
blk-mq: fix leak of q->stats

blk_alloc_queue_node() already allocates q->stats, so
blk_mq_init_allocated_queue() is overwriting it with a new allocation.

Fixes: a83b576c9c25 ("block: fix stacked driver stats init and free")
Reviewed-by: Ming Lei <tom.leiming@gmail.com>
Signed-off-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agoblock: warn if sharing request queue across gendisks
Omar Sandoval [Tue, 28 Mar 2017 23:12:15 +0000 (16:12 -0700)]
block: warn if sharing request queue across gendisks

Now that the remaining drivers have been converted to one request queue
per gendisk, let's warn if a request queue gets registered more than
once. This will catch future drivers which might do it inadvertently or
any old drivers that I may have missed.

Signed-off-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agoblock: block new I/O just after queue is set as dying
Ming Lei [Mon, 27 Mar 2017 12:06:58 +0000 (20:06 +0800)]
block: block new I/O just after queue is set as dying

Before commit 780db2071a(blk-mq: decouble blk-mq freezing
from generic bypassing), the dying flag is checked before
entering queue, and Tejun converts the checking into .mq_freeze_depth,
and assumes the counter is increased just after dying flag
is set. Unfortunately we doesn't do that in blk_set_queue_dying().

This patch calls blk_freeze_queue_start() in blk_set_queue_dying(),
so that we can block new I/O coming once the queue is set as dying.

Given blk_set_queue_dying() is always called in remove path
of block device, and queue will be cleaned up later, we don't
need to worry about undoing the counter.

Cc: Tejun Heo <tj@kernel.org>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Ming Lei <tom.leiming@gmail.com>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: Bart Van Assche <bart.vanassche@sandisk.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agoblock: rename blk_mq_freeze_queue_start()
Ming Lei [Mon, 27 Mar 2017 12:06:57 +0000 (20:06 +0800)]
block: rename blk_mq_freeze_queue_start()

As the .q_usage_counter is used by both legacy and
mq path, we need to block new I/O if queue becomes
dead in blk_queue_enter().

So rename it and we can use this function in both
paths.

Reviewed-by: Bart Van Assche <bart.vanassche@sandisk.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Ming Lei <tom.leiming@gmail.com>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agoblock: add a read barrier in blk_queue_enter()
Ming Lei [Mon, 27 Mar 2017 12:06:56 +0000 (20:06 +0800)]
block: add a read barrier in blk_queue_enter()

Without the barrier, reading DEAD flag of .q_usage_counter
and reading .mq_freeze_depth may be reordered, then the
following wait_event_interruptible() may never return.

Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Ming Lei <tom.leiming@gmail.com>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: Bart Van Assche <bart.vanassche@sandisk.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agoblk-mq: comment on races related with timeout handler
Ming Lei [Mon, 27 Mar 2017 12:06:55 +0000 (20:06 +0800)]
blk-mq: comment on races related with timeout handler

This patch adds comment on two races related with
timeout handler:

- requeue from queue busy vs. timeout
- rq free & reallocation vs. timeout

Both the races themselves and current solution aren't
explicit enough, so add comments on them.

Cc: Bart Van Assche <bart.vanassche@sandisk.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Ming Lei <tom.leiming@gmail.com>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agoblk-mq: don't complete un-started request in timeout handler
Ming Lei [Wed, 22 Mar 2017 02:14:43 +0000 (10:14 +0800)]
blk-mq: don't complete un-started request in timeout handler

When iterating busy requests in timeout handler,
if the STARTED flag of one request isn't set, that means
the request is being processed in block layer or driver, and
isn't submitted to hardware yet.

In current implementation of blk_mq_check_expired(),
if the request queue becomes dying, un-started requests are
handled as being completed/freed immediately. This way is
wrong, and can cause rq corruption or double allocation[1][2],
when doing I/O and removing&resetting NVMe device at the sametime.

This patch fixes several issues reported by Yi Zhang.

[1]. oops log 1
[  581.789754] ------------[ cut here ]------------
[  581.789758] kernel BUG at block/blk-mq.c:374!
[  581.789760] invalid opcode: 0000 [#1] SMP
[  581.789761] Modules linked in: vfat fat ipmi_ssif intel_rapl sb_edac
edac_core x86_pkg_temp_thermal intel_powerclamp coretemp kvm_intel kvm nvme
irqbypass crct10dif_pclmul nvme_core crc32_pclmul ghash_clmulni_intel
intel_cstate ipmi_si mei_me ipmi_devintf intel_uncore sg ipmi_msghandler
intel_rapl_perf iTCO_wdt mei iTCO_vendor_support mxm_wmi lpc_ich dcdbas shpchp
pcspkr acpi_power_meter wmi nfsd auth_rpcgss nfs_acl lockd dm_multipath grace
sunrpc ip_tables xfs libcrc32c sd_mod mgag200 i2c_algo_bit drm_kms_helper
syscopyarea sysfillrect sysimgblt fb_sys_fops ttm drm ahci libahci
crc32c_intel tg3 libata megaraid_sas i2c_core ptp fjes pps_core dm_mirror
dm_region_hash dm_log dm_mod
[  581.789796] CPU: 1 PID: 1617 Comm: kworker/1:1H Not tainted 4.10.0.bz1420297+ #4
[  581.789797] Hardware name: Dell Inc. PowerEdge R730xd/072T6D, BIOS 2.2.5 09/06/2016
[  581.789804] Workqueue: kblockd blk_mq_timeout_work
[  581.789806] task: ffff8804721c8000 task.stack: ffffc90006ee4000
[  581.789809] RIP: 0010:blk_mq_end_request+0x58/0x70
[  581.789810] RSP: 0018:ffffc90006ee7d50 EFLAGS: 00010202
[  581.789811] RAX: 0000000000000001 RBX: ffff8802e4195340 RCX: ffff88028e2f4b88
[  581.789812] RDX: 0000000000001000 RSI: 0000000000001000 RDI: 0000000000000000
[  581.789813] RBP: ffffc90006ee7d60 R08: 0000000000000003 R09: ffff88028e2f4b00
[  581.789814] R10: 0000000000001000 R11: 0000000000000001 R12: 00000000fffffffb
[  581.789815] R13: ffff88042abe5780 R14: 000000000000002d R15: ffff88046fbdff80
[  581.789817] FS:  0000000000000000(0000) GS:ffff88047fc00000(0000) knlGS:0000000000000000
[  581.789818] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[  581.789819] CR2: 00007f64f403a008 CR3: 000000014d078000 CR4: 00000000001406e0
[  581.789820] Call Trace:
[  581.789825]  blk_mq_check_expired+0x76/0x80
[  581.789828]  bt_iter+0x45/0x50
[  581.789830]  blk_mq_queue_tag_busy_iter+0xdd/0x1f0
[  581.789832]  ? blk_mq_rq_timed_out+0x70/0x70
[  581.789833]  ? blk_mq_rq_timed_out+0x70/0x70
[  581.789840]  ? __switch_to+0x140/0x450
[  581.789841]  blk_mq_timeout_work+0x88/0x170
[  581.789845]  process_one_work+0x165/0x410
[  581.789847]  worker_thread+0x137/0x4c0
[  581.789851]  kthread+0x101/0x140
[  581.789853]  ? rescuer_thread+0x3b0/0x3b0
[  581.789855]  ? kthread_park+0x90/0x90
[  581.789860]  ret_from_fork+0x2c/0x40
[  581.789861] Code: 48 85 c0 74 0d 44 89 e6 48 89 df ff d0 5b 41 5c 5d c3 48
8b bb 70 01 00 00 48 85 ff 75 0f 48 89 df e8 7d f0 ff ff 5b 41 5c 5d c3 <0f>
0b e8 71 f0 ff ff 90 eb e9 0f 1f 40 00 66 2e 0f 1f 84 00 00
[  581.789882] RIP: blk_mq_end_request+0x58/0x70 RSP: ffffc90006ee7d50
[  581.789889] ---[ end trace bcaf03d9a14a0a70 ]---

[2]. oops log2
[ 6984.857362] BUG: unable to handle kernel NULL pointer dereference at 0000000000000010
[ 6984.857372] IP: nvme_queue_rq+0x6e6/0x8cd [nvme]
[ 6984.857373] PGD 0
[ 6984.857374]
[ 6984.857376] Oops: 0000 [#1] SMP
[ 6984.857379] Modules linked in: ipmi_ssif vfat fat intel_rapl sb_edac
edac_core x86_pkg_temp_thermal intel_powerclamp coretemp kvm_intel kvm
irqbypass crct10dif_pclmul crc32_pclmul ghash_clmulni_intel ipmi_si iTCO_wdt
iTCO_vendor_support mxm_wmi ipmi_devintf intel_cstate sg dcdbas intel_uncore
mei_me intel_rapl_perf mei pcspkr lpc_ich ipmi_msghandler shpchp
acpi_power_meter wmi nfsd auth_rpcgss dm_multipath nfs_acl lockd grace sunrpc
ip_tables xfs libcrc32c sd_mod mgag200 i2c_algo_bit drm_kms_helper syscopyarea
sysfillrect crc32c_intel sysimgblt fb_sys_fops ttm nvme drm nvme_core ahci
libahci i2c_core tg3 libata ptp megaraid_sas pps_core fjes dm_mirror
dm_region_hash dm_log dm_mod
[ 6984.857416] CPU: 7 PID: 1635 Comm: kworker/7:1H Not tainted
4.10.0-2.el7.bz1420297.x86_64 #1
[ 6984.857417] Hardware name: Dell Inc. PowerEdge R730xd/072T6D, BIOS 2.2.5 09/06/2016
[ 6984.857427] Workqueue: kblockd blk_mq_run_work_fn
[ 6984.857429] task: ffff880476e3da00 task.stack: ffffc90002e90000
[ 6984.857432] RIP: 0010:nvme_queue_rq+0x6e6/0x8cd [nvme]
[ 6984.857433] RSP: 0018:ffffc90002e93c50 EFLAGS: 00010246
[ 6984.857434] RAX: 0000000000000000 RBX: ffff880275646600 RCX: 0000000000001000
[ 6984.857435] RDX: 0000000000000fff RSI: 00000002fba2a000 RDI: ffff8804734e6950
[ 6984.857436] RBP: ffffc90002e93d30 R08: 0000000000002000 R09: 0000000000001000
[ 6984.857437] R10: 0000000000001000 R11: 0000000000000000 R12: ffff8804741d8000
[ 6984.857438] R13: 0000000000000040 R14: ffff880475649f80 R15: ffff8804734e6780
[ 6984.857439] FS:  0000000000000000(0000) GS:ffff88047fcc0000(0000) knlGS:0000000000000000
[ 6984.857440] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[ 6984.857442] CR2: 0000000000000010 CR3: 0000000001c09000 CR4: 00000000001406e0
[ 6984.857443] Call Trace:
[ 6984.857451]  ? mempool_free+0x2b/0x80
[ 6984.857455]  ? bio_free+0x4e/0x60
[ 6984.857459]  blk_mq_dispatch_rq_list+0xf5/0x230
[ 6984.857462]  blk_mq_process_rq_list+0x133/0x170
[ 6984.857465]  __blk_mq_run_hw_queue+0x8c/0xa0
[ 6984.857467]  blk_mq_run_work_fn+0x12/0x20
[ 6984.857473]  process_one_work+0x165/0x410
[ 6984.857475]  worker_thread+0x137/0x4c0
[ 6984.857478]  kthread+0x101/0x140
[ 6984.857480]  ? rescuer_thread+0x3b0/0x3b0
[ 6984.857481]  ? kthread_park+0x90/0x90
[ 6984.857489]  ret_from_fork+0x2c/0x40
[ 6984.857490] Code: 8b bd 70 ff ff ff 89 95 50 ff ff ff 89 8d 58 ff ff ff 44
89 95 60 ff ff ff e8 b7 dd 12 e1 8b 95 50 ff ff ff 48 89 85 68 ff ff ff <4c>
8b 48 10 44 8b 58 18 8b 8d 58 ff ff ff 44 8b 95 60 ff ff ff
[ 6984.857511] RIP: nvme_queue_rq+0x6e6/0x8cd [nvme] RSP: ffffc90002e93c50
[ 6984.857512] CR2: 0000000000000010
[ 6984.895359] ---[ end trace 2d7ceb528432bf83 ]---

Cc: stable@vger.kernel.org
Reported-by: Yi Zhang <yizhan@redhat.com>
Tested-by: Yi Zhang <yizhan@redhat.com>
Reviewed-by: Bart Van Assche <bart.vanassche@sandisk.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Ming Lei <tom.leiming@gmail.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agoblkcg: allocate struct blkcg_gq outside request queue spinlock
Tahsin Erdogan [Thu, 9 Mar 2017 08:05:31 +0000 (00:05 -0800)]
blkcg: allocate struct blkcg_gq outside request queue spinlock

blkg_conf_prep() currently calls blkg_lookup_create() while holding
request queue spinlock. This means allocating memory for struct
blkcg_gq has to be made non-blocking. This causes occasional -ENOMEM
failures in call paths like below:

  pcpu_alloc+0x68f/0x710
  __alloc_percpu_gfp+0xd/0x10
  __percpu_counter_init+0x55/0xc0
  cfq_pd_alloc+0x3b2/0x4e0
  blkg_alloc+0x187/0x230
  blkg_create+0x489/0x670
  blkg_lookup_create+0x9a/0x230
  blkg_conf_prep+0x1fb/0x240
  __cfqg_set_weight_device.isra.105+0x5c/0x180
  cfq_set_weight_on_dfl+0x69/0xc0
  cgroup_file_write+0x39/0x1c0
  kernfs_fop_write+0x13f/0x1d0
  __vfs_write+0x23/0x120
  vfs_write+0xc2/0x1f0
  SyS_write+0x44/0xb0
  entry_SYSCALL_64_fastpath+0x18/0xad

In the code path above, percpu allocator cannot call vmalloc() due to
queue spinlock.

A failure in this call path gives grief to tools which are trying to
configure io weights. We see occasional failures happen shortly after
reboots even when system is not under any memory pressure. Machines
with a lot of cpus are more vulnerable to this condition.

Update blkg_create() function to temporarily drop the rcu and queue
locks when it is allowed by gfp mask.

Suggested-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Tahsin Erdogan <tahsin@google.com>
Acked-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agojsflash: stop sharing request queue across multiple gendisks
Omar Sandoval [Tue, 28 Mar 2017 06:28:47 +0000 (23:28 -0700)]
jsflash: stop sharing request queue across multiple gendisks

Compile-tested only (by hacking it to compile on x86).

Cc: David S. Miller <davem@davemloft.net>
Signed-off-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agoswim: stop sharing request queue across multiple gendisks
Omar Sandoval [Tue, 28 Mar 2017 06:28:46 +0000 (23:28 -0700)]
swim: stop sharing request queue across multiple gendisks

Compile-tested only (by hacking it to compile on x86).

Signed-off-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agoparport/pf: stop sharing request queue across multiple gendisks
Omar Sandoval [Tue, 28 Mar 2017 06:28:45 +0000 (23:28 -0700)]
parport/pf: stop sharing request queue across multiple gendisks

Compile-tested only.

Cc: Tim Waugh <tim@cyberelk.net>
Signed-off-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agoparport/pcd: stop sharing request queue across multiple gendisks
Omar Sandoval [Tue, 28 Mar 2017 06:28:44 +0000 (23:28 -0700)]
parport/pcd: stop sharing request queue across multiple gendisks

Compile-tested only.

Cc: Tim Waugh <tim@cyberelk.net>
Signed-off-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agoparport/pd: stop sharing request queue across multiple gendisks
Omar Sandoval [Tue, 28 Mar 2017 06:28:43 +0000 (23:28 -0700)]
parport/pd: stop sharing request queue across multiple gendisks

Compile-tested only.

Cc: Tim Waugh <tim@cyberelk.net>
Signed-off-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agohd: stop sharing request queue across multiple gendisks
Omar Sandoval [Tue, 28 Mar 2017 06:28:42 +0000 (23:28 -0700)]
hd: stop sharing request queue across multiple gendisks

Compile-tested only.

Signed-off-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agoblk-throttle: add latency target support
Shaohua Li [Mon, 27 Mar 2017 22:19:43 +0000 (15:19 -0700)]
blk-throttle: add latency target support

One hard problem adding .low limit is to detect idle cgroup. If one
cgroup doesn't dispatch enough IO against its low limit, we must have a
mechanism to determine if other cgroups dispatch more IO. We added the
think time detection mechanism before, but it doesn't work for all
workloads. Here we add a latency based approach.

We already have mechanism to calculate latency threshold for each IO
size. For every IO dispatched from a cgorup, we compare its latency
against its threshold and record the info. If most IO latency is below
threshold (in the code I use 75%), the cgroup could be treated idle and
other cgroups can dispatch more IO.

Currently this latency target check is only for SSD as we can't
calcualte the latency target for hard disk. And this is only for cgroup
leaf node so far.

Signed-off-by: Shaohua Li <shli@fb.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agoblk-throttle: add a mechanism to estimate IO latency
Shaohua Li [Mon, 27 Mar 2017 22:19:42 +0000 (15:19 -0700)]
blk-throttle: add a mechanism to estimate IO latency

User configures latency target, but the latency threshold for each
request size isn't fixed. For a SSD, the IO latency highly depends on
request size. To calculate latency threshold, we sample some data, eg,
average latency for request size 4k, 8k, 16k, 32k .. 1M. The latency
threshold of each request size will be the sample latency (I'll call it
base latency) plus latency target. For example, the base latency for
request size 4k is 80us and user configures latency target 60us. The 4k
latency threshold will be 80 + 60 = 140us.

To sample data, we calculate the order base 2 of rounded up IO sectors.
If the IO size is bigger than 1M, it will be accounted as 1M. Since the
calculation does round up, the base latency will be slightly smaller
than actual value. Also if there isn't any IO dispatched for a specific
IO size, we will use the base latency of smaller IO size for this IO
size.

But we shouldn't sample data at any time. The base latency is supposed
to be latency where disk isn't congested, because we use latency
threshold to schedule IOs between cgroups. If disk is congested, the
latency is higher, using it for scheduling is meaningless. Hence we only
do the sampling when block throttling is in the LOW limit, with
assumption disk isn't congested in such state. If the assumption isn't
true, eg, low limit is too high, calculated latency threshold will be
higher.

Hard disk is completely different. Latency depends on spindle seek
instead of request size. Currently this feature is SSD only, we probably
can use a fixed threshold like 4ms for hard disk though.

Signed-off-by: Shaohua Li <shli@fb.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agoblock: track request size in blk_issue_stat
Shaohua Li [Mon, 27 Mar 2017 22:19:41 +0000 (15:19 -0700)]
block: track request size in blk_issue_stat

Currently there is no way to know the request size when the request is
finished. Next patch will need this info. We could add extra field to
record the size, but blk_issue_stat has enough space to record it, so
this patch just overloads blk_issue_stat. With this, we will have 49bits
to track time, which still is very long time.

Signed-off-by: Shaohua Li <shli@fb.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agoblk-throttle: add interface for per-cgroup target latency
Shaohua Li [Mon, 27 Mar 2017 17:51:44 +0000 (10:51 -0700)]
blk-throttle: add interface for per-cgroup target latency

Here we introduce per-cgroup latency target. The target determines how a
cgroup can afford latency increasement. We will use the target latency
to calculate a threshold and use it to schedule IO for cgroups. If a
cgroup's bandwidth is below its low limit but its average latency is
below the threshold, other cgroups can safely dispatch more IO even
their bandwidth is higher than their low limits. On the other hand, if
the first cgroup's latency is higher than the threshold, other cgroups
are throttled to their low limits. So the target latency determines how
we efficiently utilize free disk resource without sacifice of worload's
IO latency.

For example, assume 4k IO average latency is 50us when disk isn't
congested. A cgroup sets the target latency to 30us. Then the cgroup can
accept 50+30=80us IO latency. If the cgroupt's average IO latency is
90us and its bandwidth is below low limit, other cgroups are throttled
to their low limit. If the cgroup's average IO latency is 60us, other
cgroups are allowed to dispatch more IO. When other cgroups dispatch
more IO, the first cgroup's IO latency will increase. If it increases to
81us, we then throttle other cgroups.

User will configure the interface in this way:
echo "8:16 rbps=2097152 wbps=max latency=100 idle=200" > io.low

latency is in microsecond unit

By default, latency target is 0, which means to guarantee IO latency.

Signed-off-by: Shaohua Li <shli@fb.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agoblk-throttle: ignore idle cgroup limit
Shaohua Li [Mon, 27 Mar 2017 17:51:43 +0000 (10:51 -0700)]
blk-throttle: ignore idle cgroup limit

Last patch introduces a way to detect idle cgroup. We use it to make
upgrade/downgrade decision. And the new algorithm can detect completely
idle cgroup too, so we can delete the corresponding code.

Signed-off-by: Shaohua Li <shli@fb.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agoblk-throttle: add interface to configure idle time threshold
Shaohua Li [Mon, 27 Mar 2017 17:51:42 +0000 (10:51 -0700)]
blk-throttle: add interface to configure idle time threshold

Add interface to configure the threshold. The io.low interface will
like:
echo "8:16 rbps=2097152 wbps=max idle=2000" > io.low

idle is in microsecond unit.

Signed-off-by: Shaohua Li <shli@fb.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agoblk-throttle: add a simple idle detection
Shaohua Li [Mon, 27 Mar 2017 17:51:41 +0000 (10:51 -0700)]
blk-throttle: add a simple idle detection

A cgroup gets assigned a low limit, but the cgroup could never dispatch
enough IO to cross the low limit. In such case, the queue state machine
will remain in LIMIT_LOW state and all other cgroups will be throttled
according to low limit. This is unfair for other cgroups. We should
treat the cgroup idle and upgrade the state machine to lower state.

We also have a downgrade logic. If the state machine upgrades because of
cgroup idle (real idle), the state machine will downgrade soon as the
cgroup is below its low limit. This isn't what we want. A more
complicated case is cgroup isn't idle when queue is in LIMIT_LOW. But
when queue gets upgraded to lower state, other cgroups could dispatch
more IO and this cgroup can't dispatch enough IO, so the cgroup is below
its low limit and looks like idle (fake idle). In this case, the queue
should downgrade soon. The key to determine if we should do downgrade is
to detect if cgroup is truely idle.

Unfortunately it's very hard to determine if a cgroup is real idle. This
patch uses the 'think time check' idea from CFQ for the purpose. Please
note, the idea doesn't work for all workloads. For example, a workload
with io depth 8 has disk utilization 100%, hence think time is 0, eg,
not idle. But the workload can run higher bandwidth with io depth 16.
Compared to io depth 16, the io depth 8 workload is idle. We use the
idea to roughly determine if a cgroup is idle.

We treat a cgroup idle if its think time is above a threshold (by
default 1ms for SSD and 100ms for HD). The idea is think time above the
threshold will start to harm performance. HD is much slower so a longer
think time is ok.

The patch (and the latter patches) uses 'unsigned long' to track time.
We convert 'ns' to 'us' with 'ns >> 10'. This is fast but loses
precision, should not a big deal.

Signed-off-by: Shaohua Li <shli@fb.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agoblk-throttle: make bandwidth change smooth
Shaohua Li [Mon, 27 Mar 2017 17:51:40 +0000 (10:51 -0700)]
blk-throttle: make bandwidth change smooth

When cgroups all reach low limit, cgroups can dispatch more IO. This
could make some cgroups dispatch more IO but others not, and even some
cgroups could dispatch less IO than their low limit. For example, cg1
low limit 10MB/s, cg2 limit 80MB/s, assume disk maximum bandwidth is
120M/s for the workload. Their bps could something like this:

cg1/cg2 bps: T1: 10/80 -> T2: 60/60 -> T3: 10/80

At T1, all cgroups reach low limit, so they can dispatch more IO later.
Then cg1 dispatch more IO and cg2 has no room to dispatch enough IO. At
T2, cg2 only dispatches 60M/s. Since We detect cg2 dispatches less IO
than its low limit 80M/s, we downgrade the queue from LIMIT_MAX to
LIMIT_LOW, then all cgroups are throttled to their low limit (T3). cg2
will have bandwidth below its low limit at most time.

The big problem here is we don't know the maximum bandwidth of the
workload, so we can't make smart decision to avoid the situation. This
patch makes cgroup bandwidth change smooth. After disk upgrades from
LIMIT_LOW to LIMIT_MAX, we don't allow cgroups use all bandwidth upto
their max limit immediately. Their bandwidth limit will be increased
gradually to avoid above situation. So above example will became
something like:

cg1/cg2 bps: 10/80 -> 15/105 -> 20/100 -> 25/95 -> 30/90 -> 35/85 -> 40/80
-> 45/75 -> 22/98

In this way cgroups bandwidth will be above their limit in majority
time, this still doesn't fully utilize disk bandwidth, but that's
something we pay for sharing.

Scale up is linear. The limit scales up 1/2 .low limit every
throtl_slice after upgrade. The scale up will stop if the adjusted limit
hits .max limit. Scale down is exponential. We cut the scale value half
if a cgroup doesn't hit its .low limit. If the scale becomes 0, we then
fully downgrade the queue to LIMIT_LOW state.

Note this doesn't completely avoid cgroup running under its low limit.
The best way to guarantee cgroup doesn't run under its limit is to set
max limit. For example, if we set cg1 max limit to 40, cg2 will never
run under its low limit.

Signed-off-by: Shaohua Li <shli@fb.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agoblk-throttle: detect completed idle cgroup
Shaohua Li [Mon, 27 Mar 2017 17:51:39 +0000 (10:51 -0700)]
blk-throttle: detect completed idle cgroup

cgroup could be assigned a limit, but doesn't dispatch enough IO, eg the
cgroup is idle. When this happens, the cgroup doesn't hit its limit, so
we can't move the state machine to higher level and all cgroups will be
throttled to their lower limit, so we waste bandwidth. Detecting idle
cgroup is hard. This patch handles a simple case, a cgroup doesn't
dispatch any IO. We ignore such cgroup's limit, so other cgroups can use
the bandwidth.

Please note this will be replaced with a more sophisticated algorithm
later, but this demonstrates the idea how we handle idle cgroups, so I
leave it here.

Signed-off-by: Shaohua Li <shli@fb.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agoblk-throttle: choose a small throtl_slice for SSD
Shaohua Li [Mon, 27 Mar 2017 17:51:38 +0000 (10:51 -0700)]
blk-throttle: choose a small throtl_slice for SSD

The throtl_slice is 100ms by default. This is a long time for SSD, a lot
of IO can run. To make cgroups have smoother throughput, we choose a
small value (20ms) for SSD.

Signed-off-by: Shaohua Li <shli@fb.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agoblk-throttle: make throtl_slice tunable
Shaohua Li [Mon, 27 Mar 2017 17:51:37 +0000 (10:51 -0700)]
blk-throttle: make throtl_slice tunable

throtl_slice is important for blk-throttling. It's called slice
internally but it really is a time window blk-throttling samples data.
blk-throttling will make decision based on the samplings. An example is
bandwidth measurement. A cgroup's bandwidth is measured in the time
interval of throtl_slice.

A small throtl_slice meanse cgroups have smoother throughput but burn
more CPUs. It has 100ms default value, which is not appropriate for all
disks. A fast SSD can dispatch a lot of IOs in 100ms. This patch makes
it tunable.

Since throtl_slice isn't a time slice, the sysfs name
'throttle_sample_time' reflects its character better.

Signed-off-by: Shaohua Li <shli@fb.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agoblk-throttle: make sure expire time isn't too big
Shaohua Li [Mon, 27 Mar 2017 17:51:36 +0000 (10:51 -0700)]
blk-throttle: make sure expire time isn't too big

cgroup could be throttled to a limit but when all cgroups cross high
limit, queue enters a higher state and so the group should be throttled
to a higher limit. It's possible the cgroup is sleeping because of
throttle and other cgroups don't dispatch IO any more. In this case,
nobody can trigger current downgrade/upgrade logic. To fix this issue,
we could either set up a timer to wakeup the cgroup if other cgroups are
idle or make sure this cgroup doesn't sleep too long. Setting up a timer
means we must change the timer very frequently. This patch chooses the
latter. Making cgroup sleep time not too big wouldn't change cgroup
bps/iops, but could make it wakeup more frequently, which isn't a big
issue because throtl_slice * 8 is already quite big.

Signed-off-by: Shaohua Li <shli@fb.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agoblk-throttle: add downgrade logic
Shaohua Li [Mon, 27 Mar 2017 17:51:35 +0000 (10:51 -0700)]
blk-throttle: add downgrade logic

When queue state machine is in LIMIT_MAX state, but a cgroup is below
its low limit for some time, the queue should be downgraded to lower
state as one cgroup's low limit isn't met.

Signed-off-by: Shaohua Li <shli@fb.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agoblk-throttle: add upgrade logic for LIMIT_LOW state
Shaohua Li [Mon, 27 Mar 2017 17:51:34 +0000 (10:51 -0700)]
blk-throttle: add upgrade logic for LIMIT_LOW state

When queue is in LIMIT_LOW state and all cgroups with low limit cross
the bps/iops limitation, we will upgrade queue's state to
LIMIT_MAX. To determine if a cgroup exceeds its limitation, we check if
the cgroup has pending request. Since cgroup is throttled according to
the limit, pending request means the cgroup reaches the limit.

If a cgroup has limit set for both read and write, we consider the
combination of them for upgrade. The reason is read IO and write IO can
interfere with each other. If we do the upgrade based in one direction
IO, the other direction IO could be severly harmed.

For a cgroup hierarchy, there are two cases. Children has lower low
limit than parent. Parent's low limit is meaningless. If children's
bps/iops cross low limit, we can upgrade queue state. The other case is
children has higher low limit than parent. Children's low limit is
meaningless. As long as parent's bps/iops (which is a sum of childrens
bps/iops) cross low limit, we can upgrade queue state.

Signed-off-by: Shaohua Li <shli@fb.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agoblk-throttle: configure bps/iops limit for cgroup in low limit
Shaohua Li [Mon, 27 Mar 2017 17:51:33 +0000 (10:51 -0700)]
blk-throttle: configure bps/iops limit for cgroup in low limit

each queue will have a state machine. Initially queue is in LIMIT_LOW
state, which means all cgroups will be throttled according to their low
limit. After all cgroups with low limit cross the limit, the queue state
gets upgraded to LIMIT_MAX state.
For max limit, cgroup will use the limit configured by user.
For low limit, cgroup will use the minimal value between low limit and
max limit configured by user. If the minimal value is 0, which means the
cgroup doesn't configure low limit, we will use max limit to throttle
the cgroup and the cgroup is ready to upgrade to LIMIT_MAX

Signed-off-by: Shaohua Li <shli@fb.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agoblk-throttle: add .low interface
Shaohua Li [Mon, 27 Mar 2017 17:51:32 +0000 (10:51 -0700)]
blk-throttle: add .low interface

Add low limit for cgroup and corresponding cgroup interface. To be
consistent with memcg, we allow users configure .low limit higher than
.max limit. But the internal logic always assumes .low limit is lower
than .max limit. So we add extra bps/iops_conf fields in throtl_grp for
userspace configuration. Old bps/iops fields in throtl_grp will be the
actual limit we use for throttling.

Signed-off-by: Shaohua Li <shli@fb.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agoblk-throttle: add configure option for new .low interface
Shaohua Li [Mon, 27 Mar 2017 17:51:31 +0000 (10:51 -0700)]
blk-throttle: add configure option for new .low interface

As discussed in LSF, add configure option for the interface and mark it
as experimental, so people can try/test.

Signed-off-by: Shaohua Li <shli@fb.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agoblk-throttle: prepare support multiple limits
Shaohua Li [Mon, 27 Mar 2017 17:51:30 +0000 (10:51 -0700)]
blk-throttle: prepare support multiple limits

We are going to support low/max limit, each cgroup will have 2 limits
after that. This patch prepares for the multiple limits change.

Signed-off-by: Shaohua Li <shli@fb.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agoblk-throttle: use U64_MAX/UINT_MAX to replace -1
Shaohua Li [Mon, 27 Mar 2017 17:51:29 +0000 (10:51 -0700)]
blk-throttle: use U64_MAX/UINT_MAX to replace -1

clean up the code to avoid using -1

Signed-off-by: Shaohua Li <shli@fb.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agoblock: constify struct blk_integrity_profile
Eric Biggers [Sat, 25 Mar 2017 01:03:48 +0000 (18:03 -0700)]
block: constify struct blk_integrity_profile

blk_integrity_profile's are never modified, so mark them 'const' so that
they are placed in .rodata and benefit from memory protection.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agoblock: remove outdated part of blkdev_issue_flush() comment
Eric Biggers [Mon, 23 Jan 2017 19:43:21 +0000 (11:43 -0800)]
block: remove outdated part of blkdev_issue_flush() comment

blkdev_issue_flush() is now always synchronous, and it no longer has a
flags argument.  So remove the part of the comment about the WAIT flag.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agoblock: correct documentation for blkdev_issue_discard() flags
Eric Biggers [Mon, 23 Jan 2017 19:41:39 +0000 (11:41 -0800)]
block: correct documentation for blkdev_issue_discard() flags

BLKDEV_IFL_* flags no longer exist; blkdev_issue_discard() now actually
takes BLKDEV_DISCARD_* flags.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agomg_disk: use setup_timer
Geliang Tang [Fri, 24 Mar 2017 14:15:13 +0000 (22:15 +0800)]
mg_disk: use setup_timer

Use setup_timer() instead of init_timer() to simplify the code.

Signed-off-by: Geliang Tang <geliangtang@gmail.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agoblock: floppy: use setup_timer
Geliang Tang [Fri, 24 Mar 2017 14:15:10 +0000 (22:15 +0800)]
block: floppy: use setup_timer

Use setup_timer() instead of init_timer() to simplify the code.

Signed-off-by: Geliang Tang <geliangtang@gmail.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agoblock: make nr_iovecs unsigned in bio_alloc_bioset()
Dan Carpenter [Thu, 23 Mar 2017 10:24:55 +0000 (13:24 +0300)]
block: make nr_iovecs unsigned in bio_alloc_bioset()

There isn't a bug here, but Smatch is not smart enough to know that
"nr_iovecs" can't be negative so it complains about underflows.
Really, it's slightly cleaner to make this parameter unsigned.

Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agoblk-mq: streamline blk_mq_make_request
Christoph Hellwig [Wed, 22 Mar 2017 19:01:53 +0000 (15:01 -0400)]
blk-mq: streamline blk_mq_make_request

Turn the different ways of merging or issuing I/O into a series of if/else
statements instead of the current maze of gotos.  Note that this means we
pin the CPU a little longer for some cases as the CTX put is moved to
common code at the end of the function.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agoblk-mq: split the plug and sync cases in blk_mq_make_request
Christoph Hellwig [Wed, 22 Mar 2017 19:01:52 +0000 (15:01 -0400)]
blk-mq: split the plug and sync cases in blk_mq_make_request

Now that we have a nice direct issue heper this helps simplifying
the code a bit, and also gets rid of the old_rq variable.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agoblk-mq: improve blk_mq_try_issue_directly
Christoph Hellwig [Wed, 22 Mar 2017 19:01:51 +0000 (15:01 -0400)]
blk-mq: improve blk_mq_try_issue_directly

Rename blk_mq_try_issue_directly to __blk_mq_try_issue_directly and add a
new wrapper that takes care of RCU / SRCU locking to avoid having
boileplate code in the caller which would get duplicated with new callers.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Bart Van Assche <bart.vanassche@sandisk.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agoblk-mq: merge mq and sq make_request instances
Christoph Hellwig [Wed, 22 Mar 2017 19:01:50 +0000 (15:01 -0400)]
blk-mq: merge mq and sq make_request instances

They are mostly the same code anyway - this just one small conditional
for the plug case that is different for both variants.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agoblk-mq: remove BLK_MQ_F_DEFER_ISSUE
Christoph Hellwig [Wed, 22 Mar 2017 19:01:49 +0000 (15:01 -0400)]
blk-mq: remove BLK_MQ_F_DEFER_ISSUE

This flag was never used since it was introduced.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Bart Van Assche <bart.vanassche@sandisk.com>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agoblock: Fix oops scsi_disk_get()
Jan Kara [Thu, 23 Mar 2017 00:37:02 +0000 (01:37 +0100)]
block: Fix oops scsi_disk_get()

When device open races with device shutdown, we can get the following
oops in scsi_disk_get():

[11863.044351] general protection fault: 0000 [#1] SMP
[11863.045561] Modules linked in: scsi_debug xfs libcrc32c netconsole btrfs raid6_pq zlib_deflate lzo_compress xor [last unloaded: loop]
[11863.047853] CPU: 3 PID: 13042 Comm: hald-probe-stor Tainted: G W      4.10.0-rc2-xen+ #35
[11863.048030] Hardware name: Bochs Bochs, BIOS Bochs 01/01/2011
[11863.048030] task: ffff88007f438200 task.stack: ffffc90000fd0000
[11863.048030] RIP: 0010:scsi_disk_get+0x43/0x70
[11863.048030] RSP: 0018:ffffc90000fd3a08 EFLAGS: 00010202
[11863.048030] RAX: 6b6b6b6b6b6b6b6b RBX: ffff88007f56d000 RCX: 0000000000000000
[11863.048030] RDX: 0000000000000001 RSI: 0000000000000004 RDI: ffffffff81a8d880
[11863.048030] RBP: ffffc90000fd3a18 R08: 0000000000000000 R09: 0000000000000001
[11863.059217] R10: 0000000000000000 R11: 0000000000000000 R12: 00000000fffffffa
[11863.059217] R13: ffff880078872800 R14: ffff880070915540 R15: 000000000000001d
[11863.059217] FS:  00007f2611f71800(0000) GS:ffff88007f0c0000(0000) knlGS:0000000000000000
[11863.059217] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[11863.059217] CR2: 000000000060e048 CR3: 00000000778d4000 CR4: 00000000000006e0
[11863.059217] Call Trace:
[11863.059217]  ? disk_get_part+0x22/0x1f0
[11863.059217]  sd_open+0x39/0x130
[11863.059217]  __blkdev_get+0x69/0x430
[11863.059217]  ? bd_acquire+0x7f/0xc0
[11863.059217]  ? bd_acquire+0x96/0xc0
[11863.059217]  ? blkdev_get+0x350/0x350
[11863.059217]  blkdev_get+0x126/0x350
[11863.059217]  ? _raw_spin_unlock+0x2b/0x40
[11863.059217]  ? bd_acquire+0x7f/0xc0
[11863.059217]  ? blkdev_get+0x350/0x350
[11863.059217]  blkdev_open+0x65/0x80
...

As you can see RAX value is already poisoned showing that gendisk we got
is already freed. The problem is that get_gendisk() looks up device
number in ext_devt_idr and then does get_disk() which does kobject_get()
on the disks kobject. However the disk gets removed from ext_devt_idr
only in disk_release() (through blk_free_devt()) at which moment it has
already 0 refcount and is already on its way to be freed. Indeed we've
got a warning from kobject_get() about 0 refcount shortly before the
oops.

We fix the problem by using kobject_get_unless_zero() in get_disk() so
that get_disk() cannot get reference on a disk that is already being
freed.

Tested-by: Lekshmi Pillai <lekshmicpillai@in.ibm.com>
Reviewed-by: Bart Van Assche <bart.vanassche@sandisk.com>
Acked-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agokobject: Export kobject_get_unless_zero()
Jan Kara [Thu, 23 Mar 2017 00:37:01 +0000 (01:37 +0100)]
kobject: Export kobject_get_unless_zero()

Make the function available for outside use and fortify it against NULL
kobject.

CC: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Reviewed-by: Bart Van Assche <bart.vanassche@sandisk.com>
Acked-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agoblock: Fix oops in locked_inode_to_wb_and_lock_list()
Jan Kara [Thu, 23 Mar 2017 00:37:00 +0000 (01:37 +0100)]
block: Fix oops in locked_inode_to_wb_and_lock_list()

When block device is closed, we call inode_detach_wb() in __blkdev_put()
which sets inode->i_wb to NULL. That is contrary to expectations that
inode->i_wb stays valid once set during the whole inode's lifetime and
leads to oops in wb_get() in locked_inode_to_wb_and_lock_list() because
inode_to_wb() returned NULL.

The reason why we called inode_detach_wb() is not valid anymore though.
BDI is guaranteed to stay along until we call bdi_put() from
bdev_evict_inode() so we can postpone calling inode_detach_wb() to that
moment.

Also add a warning to catch if someone uses inode_detach_wb() in a
dangerous way.

Reported-by: Thiago Jung Bauermann <bauerman@linux.vnet.ibm.com>
Acked-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agobdi: Rename cgwb_bdi_destroy() to cgwb_bdi_unregister()
Jan Kara [Thu, 23 Mar 2017 00:36:59 +0000 (01:36 +0100)]
bdi: Rename cgwb_bdi_destroy() to cgwb_bdi_unregister()

Rename cgwb_bdi_destroy() to cgwb_bdi_unregister() as it gets called
from bdi_unregister() which is not necessarily called from bdi_destroy()
and thus the name is somewhat misleading.

Acked-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agobdi: Do not wait for cgwbs release in bdi_unregister()
Jan Kara [Thu, 23 Mar 2017 00:36:58 +0000 (01:36 +0100)]
bdi: Do not wait for cgwbs release in bdi_unregister()

Currently we wait for all cgwbs to get released in cgwb_bdi_destroy()
(called from bdi_unregister()). That is however unnecessary now when
cgwb->bdi is a proper refcounted reference (thus bdi cannot get
released before all cgwbs are released) and when cgwb_bdi_destroy()
shuts down writeback directly.

Acked-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agobdi: Shutdown writeback on all cgwbs in cgwb_bdi_destroy()
Jan Kara [Thu, 23 Mar 2017 00:36:57 +0000 (01:36 +0100)]
bdi: Shutdown writeback on all cgwbs in cgwb_bdi_destroy()

Currently we waited for all cgwbs to get freed in cgwb_bdi_destroy()
which also means that writeback has been shutdown on them. Since this
wait is going away, directly shutdown writeback on cgwbs from
cgwb_bdi_destroy() to avoid live writeback structures after
bdi_unregister() has finished. To make that safe with concurrent
shutdown from cgwb_release_workfn(), we also have to make sure
wb_shutdown() returns only after the bdi_writeback structure is really
shutdown.

Acked-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agobdi: Unify bdi->wb_list handling for root wb_writeback
Jan Kara [Thu, 23 Mar 2017 00:36:56 +0000 (01:36 +0100)]
bdi: Unify bdi->wb_list handling for root wb_writeback

Currently root wb_writeback structure is added to bdi->wb_list in
bdi_init() and never removed. That is different from all other
wb_writeback structures which get added to the list when created and
removed from it before wb_shutdown().

So move list addition of root bdi_writeback to bdi_register() and list
removal of all wb_writeback structures to wb_shutdown(). That way a
wb_writeback structure is on bdi->wb_list if and only if it can handle
writeback and it will make it easier for us to handle shutdown of all
wb_writeback structures in bdi_unregister().

Acked-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agobdi: Make wb->bdi a proper reference
Jan Kara [Thu, 23 Mar 2017 00:36:55 +0000 (01:36 +0100)]
bdi: Make wb->bdi a proper reference

Make wb->bdi a proper refcounted reference to bdi for all bdi_writeback
structures except for the one embedded inside struct backing_dev_info.
That will allow us to simplify bdi unregistration.

Acked-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agobdi: Mark congested->bdi as internal
Jan Kara [Thu, 23 Mar 2017 00:36:54 +0000 (01:36 +0100)]
bdi: Mark congested->bdi as internal

congested->bdi pointer is used only to be able to remove congested
structure from bdi->cgwb_congested_tree on structure release. Moreover
the pointer can become NULL when we unregister the bdi. Rename the field
to __bdi and add a comment to make it more explicit this is internal
stuff of memcg writeback code and people should not use the field as
such use will be likely race prone.

We do not bother with converting congested->bdi to a proper refcounted
reference. It will be slightly ugly to special-case bdi->wb.congested to
avoid effectively a cyclic reference of bdi to itself and the reference
gets cleared from bdi_unregister() making it impossible to reference
a freed bdi.

Acked-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agoblock: Fix bdi assignment to bdev inode when racing with disk delete
Jan Kara [Thu, 23 Mar 2017 00:36:53 +0000 (01:36 +0100)]
block: Fix bdi assignment to bdev inode when racing with disk delete

When disk->fops->open() in __blkdev_get() returns -ERESTARTSYS, we
restart the process of opening the block device. However we forget to
switch bdev->bd_bdi back to noop_backing_dev_info and as a result bdev
inode will be pointing to a stale bdi. Fix the problem by setting
bdev->bd_bdi later when __blkdev_get() is already guaranteed to succeed.

Acked-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agoblock: fix stacked driver stats init and free
Jens Axboe [Tue, 21 Mar 2017 23:20:01 +0000 (17:20 -0600)]
block: fix stacked driver stats init and free

If a driver allocates a queue for stacked usage, then it does
not currently get stats allocated. This causes the later init
of, eg, writeback throttling to blow up. Move the init to the
queue allocation instead.

Additionally, allow a NULL callback unregistration. This avoids
having the caller check for that, fixing another oops on
removal of a block device that doesn't have poll stats allocated.

Fixes: 34dbad5d26e2 ("blk-stat: convert to callback-based statistics reporting")
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agoblk-stat: convert to callback-based statistics reporting
Omar Sandoval [Tue, 21 Mar 2017 15:56:08 +0000 (08:56 -0700)]
blk-stat: convert to callback-based statistics reporting

Currently, statistics are gathered in ~0.13s windows, and users grab the
statistics whenever they need them. This is not ideal for both in-tree
users:

1. Writeback throttling wants its own dynamically sized window of
   statistics. Since the blk-stats statistics are reset after every
   window and the wbt windows don't line up with the blk-stats windows,
   wbt doesn't see every I/O.
2. Polling currently grabs the statistics on every I/O. Again, depending
   on how the window lines up, we may miss some I/Os. It's also
   unnecessary overhead to get the statistics on every I/O; the hybrid
   polling heuristic would be just as happy with the statistics from the
   previous full window.

This reworks the blk-stats infrastructure to be callback-based: users
register a callback that they want called at a given time with all of
the statistics from the window during which the callback was active.
Users can dynamically bucketize the statistics. wbt and polling both
currently use read vs. write, but polling can be extended to further
subdivide based on request size.

The callbacks are kept on an RCU list, and each callback has percpu
stats buffers. There will only be a few users, so the overhead on the
I/O completion side is low. The stats flushing is also simplified
considerably: since the timer function is responsible for clearing the
statistics, we don't have to worry about stale statistics.

wbt is a trivial conversion. After the conversion, the windowing problem
mentioned above is fixed.

For polling, we register an extra callback that caches the previous
window's statistics in the struct request_queue for the hybrid polling
heuristic to use.

Since we no longer have a single stats buffer for the request queue,
this also removes the sysfs and debugfs stats entries. To replace those,
we add a debugfs entry for the poll statistics.

Signed-off-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agoblk-stat: move BLK_RQ_STAT_BATCH definition to blk-stat.c
Omar Sandoval [Tue, 21 Mar 2017 15:56:07 +0000 (08:56 -0700)]
blk-stat: move BLK_RQ_STAT_BATCH definition to blk-stat.c

This is an implementation detail that no-one outside of blk-stat.c uses.

Signed-off-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agoblk-stat: use READ and WRITE instead of BLK_STAT_{READ,WRITE}
Omar Sandoval [Tue, 21 Mar 2017 15:56:06 +0000 (08:56 -0700)]
blk-stat: use READ and WRITE instead of BLK_STAT_{READ,WRITE}

The stats buckets will become generic soon, so make the existing users
use the common READ and WRITE definitions instead of one internal to
blk-stat.

Signed-off-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agoblock: remove extra calls to wbt_exit()
Omar Sandoval [Tue, 21 Mar 2017 15:56:05 +0000 (08:56 -0700)]
block: remove extra calls to wbt_exit()

We always call wbt_exit() from blk_release_queue(), so these are
unnecessary.

Signed-off-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agoblk-stat: fix blk_stat_sum() if all samples are batched
Omar Sandoval [Thu, 16 Mar 2017 15:46:14 +0000 (09:46 -0600)]
blk-stat: fix blk_stat_sum() if all samples are batched

We need to flush the batch _before_ we check the number of samples,
otherwise we'll miss all of the batched samples.

Fixes: cf43e6b ("block: add scalable completion tracking of requests")
Signed-off-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
7 years agoLinux 4.11-rc3
Linus Torvalds [Mon, 20 Mar 2017 02:09:39 +0000 (19:09 -0700)]
Linux 4.11-rc3

7 years agomm/swap: don't BUG_ON() due to uninitialized swap slot cache
Linus Torvalds [Mon, 20 Mar 2017 02:00:47 +0000 (19:00 -0700)]
mm/swap: don't BUG_ON() due to uninitialized swap slot cache

This BUG_ON() triggered for me once at shutdown, and I don't see a
reason for the check.  The code correctly checks whether the swap slot
cache is usable or not, so an uninitialized swap slot cache is not
actually problematic afaik.

I've temporarily just switched the BUG_ON() to a WARN_ON_ONCE(), since
I'm not sure why that seemingly pointless check was there.  I suspect
the real fix is to just remove it entirely, but for now we'll warn about
it but not bring the machine down.

Cc: "Huang, Ying" <ying.huang@intel.com>
Cc: Tim Chen <tim.c.chen@linux.intel.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agoMerge tag 'powerpc-4.11-5' of git://git.kernel.org/pub/scm/linux/kernel/git/powerpc...
Linus Torvalds [Mon, 20 Mar 2017 01:49:28 +0000 (18:49 -0700)]
Merge tag 'powerpc-4.11-5' of git://git./linux/kernel/git/powerpc/linux

Pull more powerpc fixes from Michael Ellerman:
 "A couple of minor powerpc fixes for 4.11:

   - wire up statx() syscall

   - don't print a warning on memory hotplug when HPT resizing isn't
     available

  Thanks to: David Gibson, Chandan Rajendra"

* tag 'powerpc-4.11-5' of git://git.kernel.org/pub/scm/linux/kernel/git/powerpc/linux:
  powerpc/pseries: Don't give a warning when HPT resizing isn't available
  powerpc: Wire up statx() syscall

7 years agoMerge branch 'parisc-4.11-2' of git://git.kernel.org/pub/scm/linux/kernel/git/deller...
Linus Torvalds [Mon, 20 Mar 2017 01:11:13 +0000 (18:11 -0700)]
Merge branch 'parisc-4.11-2' of git://git./linux/kernel/git/deller/parisc-linux

Pull parisc fixes from Helge Deller:

 - Mikulas Patocka added support for R_PARISC_SECREL32 relocations in
   modules with CONFIG_MODVERSIONS.

 - Dave Anglin optimized the cache flushing for vmap ranges.

 - Arvind Yadav provided a fix for a potential NULL pointer dereference
   in the parisc perf code (and some code cleanups).

 - I wired up the new statx system call, fixed some compiler warnings
   with the access_ok() macro and fixed shutdown code to really halt a
   system at shutdown instead of crashing & rebooting.

* 'parisc-4.11-2' of git://git.kernel.org/pub/scm/linux/kernel/git/deller/parisc-linux:
  parisc: Fix system shutdown halt
  parisc: perf: Fix potential NULL pointer dereference
  parisc: Avoid compiler warnings with access_ok()
  parisc: Wire up statx system call
  parisc: Optimize flush_kernel_vmap_range and invalidate_kernel_vmap_range
  parisc: support R_PARISC_SECREL32 relocation in modules

7 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/nab/target-pending
Linus Torvalds [Mon, 20 Mar 2017 01:06:31 +0000 (18:06 -0700)]
Merge git://git./linux/kernel/git/nab/target-pending

Pull SCSI target fixes from Nicholas Bellinger:
 "The bulk of the changes are in qla2xxx target driver code to address
  various issues found during Cavium/QLogic's internal testing (stable
  CC's included), along with a few other stability and smaller
  miscellaneous improvements.

  There are also a couple of different patch sets from Mike Christie,
  which have been a result of his work to use target-core ALUA logic
  together with tcm-user backend driver.

  Finally, a patch to address some long standing issues with
  pass-through SCSI export of TYPE_TAPE + TYPE_MEDIUM_CHANGER devices,
  which will make folks using physical (or virtual) magnetic tape happy"

* git://git.kernel.org/pub/scm/linux/kernel/git/nab/target-pending: (28 commits)
  qla2xxx: Update driver version to 9.00.00.00-k
  qla2xxx: Fix delayed response to command for loop mode/direct connect.
  qla2xxx: Change scsi host lookup method.
  qla2xxx: Add DebugFS node to display Port Database
  qla2xxx: Use IOCB interface to submit non-critical MBX.
  qla2xxx: Add async new target notification
  qla2xxx: Export DIF stats via debugfs
  qla2xxx: Improve T10-DIF/PI handling in driver.
  qla2xxx: Allow relogin to proceed if remote login did not finish
  qla2xxx: Fix sess_lock & hardware_lock lock order problem.
  qla2xxx: Fix inadequate lock protection for ABTS.
  qla2xxx: Fix request queue corruption.
  qla2xxx: Fix memory leak for abts processing
  qla2xxx: Allow vref count to timeout on vport delete.
  tcmu: Convert cmd_time_out into backend device attribute
  tcmu: make cmd timeout configurable
  tcmu: add helper to check if dev was configured
  target: fix race during implicit transition work flushes
  target: allow userspace to set state to transitioning
  target: fix ALUA transition timeout handling
  ...

7 years agoMerge branch 'libnvdimm-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/nvdim...
Linus Torvalds [Sun, 19 Mar 2017 22:45:02 +0000 (15:45 -0700)]
Merge branch 'libnvdimm-fixes' of git://git./linux/kernel/git/nvdimm/nvdimm

Pull device-dax fixes from Dan Williams:
 "The device-dax driver was not being careful to handle falling back to
  smaller fault-granularity sizes.

  The driver already fails fault attempts that are smaller than the
  device's alignment, but it also needs to handle the cases where a
  larger page mapping could be established. For simplicity of the
  immediate fix the implementation just signals VM_FAULT_FALLBACK until
  fault-size == device-alignment.

  One fix is for -stable to address pmd-to-pte fallback from the
  original implementation, another fix is for the new (introduced in
  4.11-rc1) pud-to-pmd regression, and a typo fix comes along for the
  ride.

  These have received a build success notification from the kbuild
  robot"

* 'libnvdimm-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/nvdimm/nvdimm:
  device-dax: fix debug output typo
  device-dax: fix pud fault fallback handling
  device-dax: fix pmd/pte fault fallback handling