For request_fn based devices, the block layer exports a 'nr_requests'
file through sysfs to allow adjusting of queue depth on the fly.
Currently this returns -EINVAL for blk-mq, since it's not wired up.
Wire this up for blk-mq, so that it now also always dynamic
adjustments of the allowed queue depth for any given block device
managed by blk-mq.
Signed-off-by: NJens Axboe <axboe@fb.com>

Each hardware queue has a bitmap of software queues with pending
requests. When new IO is queued on a software queue, the bit is
set, and when IO is pruned on a hardware queue run, the bit is
cleared. This causes a lot of traffic. Switch this from the regular
BITS_PER_LONG bitmap to a sparser layout, similarly to what was
done for blk-mq tagging.

20% performance increase was observed for single threaded IO, and
about 15% performanc increase on multiple threads driving the
same device.
Signed-off-by: NJens Axboe <axboe@fb.com>

This adds support for active queue tracking, meaning that the
blk-mq tagging maintains a count of active users of a tag set.
This allows us to maintain a notion of fairness between users,
so that we can distribute the tag depth evenly without starving
some users while allowing others to try unfair deep queues.

If sharing of a tag set is detected, each hardware queue will
track the depth of its own queue. And if this exceeds the total
depth divided by the number of active queues, the user is actively
throttled down.

The active queue count is done lazily to avoid bouncing that data
between submitter and completer. Each hardware queue gets marked
active when it allocates its first tag, and gets marked inactive
when 1) the last tag is cleared, and 2) the queue timeout grace
period has passed.
Signed-off-by: NJens Axboe <axboe@fb.com>

blk-mq currently uses percpu_ida for tag allocation. But that only
works well if the ratio between tag space and number of CPUs is
sufficiently high. For most devices and systems, that is not the
case. The end result if that we either only utilize the tag space
partially, or we end up attempting to fully exhaust it and run
into lots of lock contention with stealing between CPUs. This is
not optimal.

This new tagging scheme is a hybrid bitmap allocator. It uses
two tricks to both be SMP friendly and allow full exhaustion
of the space:

1) We cache the last allocated (or freed) tag on a per blk-mq
   software context basis. This allows us to limit the space
   we have to search. The key element here is not caching it
   in the shared tag structure, otherwise we end up dirtying
   more shared cache lines on each allocate/free operation.

2) The tag space is split into cache line sized groups, and
   each context will start off randomly in that space. Even up
   to full utilization of the space, this divides the tag users
   efficiently into cache line groups, avoiding dirtying the same
   one both between allocators and between allocator and freeer.

This scheme shows drastically better behaviour, both on small
tag spaces but on large ones as well. It has been tested extensively
to show better performance for all the cases blk-mq cares about.
Signed-off-by: NJens Axboe <axboe@fb.com>

Right now we just pick the first CPU in the mask, but that can
easily overload that one. Add some basic batching and round-robin
all the entries in the mask instead.
Signed-off-by: NJens Axboe <axboe@fb.com>

The blk-mq code is using it's own version of the I/O completion affinity
tunables, which causes a few issues:

 - the rq_affinity sysfs file doesn't work for blk-mq devices, even if it
   still is present, thus breaking existing tuning setups.
 - the rq_affinity = 1 mode, which is the defauly for legacy request based
   drivers isn't implemented at all.
 - blk-mq drivers don't implement any completion affinity with the default
   flag settings.

This patches removes the blk-mq ipi_redirect flag and sysfs file, as well
as the internal BLK_MQ_F_SHOULD_IPI flag and replaces it with code that
respects the queue-wide rq_affinity flags and also implements the
rq_affinity = 1 mode.

This means I/O completion affinity can now only be tuned block-queue wide
instead of per context, which seems more sensible to me anyway.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NJens Axboe <axboe@fb.com>

This allows to requeue a request that has been accepted by ->queue_rq
earlier.  This is needed by the SCSI layer in various error conditions.

The existing internal blk_mq_requeue_request is renamed to
__blk_mq_requeue_request as it is a lower level building block for this
funtionality.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NJens Axboe <axboe@fb.com>

Add a helper to unconditionally kick contexts of a queue.  This will
be needed by the SCSI layer to provide fair queueing between multiple
devices on a single host.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NJens Axboe <axboe@fb.com>

Add a blk-mq equivalent to blk_delay_queue so that the scsi layer can ask
to be kicked again after a delay.
Signed-off-by: NChristoph Hellwig <hch@lst.de>

Modified by me to kill the unnecessary preempt disable/enable
in the delayed workqueue handler.
Signed-off-by: NJens Axboe <axboe@fb.com>

Signed-off-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NJens Axboe <axboe@fb.com>

Split out the bottom half of blk_mq_end_io so that drivers can perform
work when they know a request has been completed, but before it has been
freed.  This also obsoletes blk_mq_end_io_partial as drivers can now
pass any value to blk_update_request directly.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NJens Axboe <axboe@fb.com>

Add a new blk_mq_tag_set structure that gets set up before we initialize
the queue.  A single blk_mq_tag_set structure can be shared by multiple
queues.
Signed-off-by: NChristoph Hellwig <hch@lst.de>

Modular export of blk_mq_{alloc,free}_tagset added by me.
Signed-off-by: NJens Axboe <axboe@fb.com>

The current blk_mq_init_commands/blk_mq_free_commands interface has a
two problems:

 1) Because only the constructor is passed to blk_mq_init_commands there
    is no easy way to clean up when a comman initialization failed.  The
    current code simply leaks the allocations done in the constructor.

 2) There is no good place to call blk_mq_free_commands: before
    blk_cleanup_queue there is no guarantee that all outstanding
    commands have completed, so we can't free them yet.  After
    blk_cleanup_queue the queue has usually been freed.  This can be
    worked around by grabbing an unconditional reference before calling
    blk_cleanup_queue and dropping it after blk_mq_free_commands is
    done, although that's not exatly pretty and driver writers are
    guaranteed to get it wrong sooner or later.

Both issues are easily fixed by making the request constructor and
destructor normal blk_mq_ops methods.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NJens Axboe <axboe@fb.com>

Instead of providing soft mappings with no guarantees on hardware
queues always being run on the right CPU, switch to a hard mapping
guarantee that ensure that we always run the hardware queue on
(one of, if more) the mapped CPU.
Signed-off-by: NJens Axboe <axboe@fb.com>

It's almost identical to blk_mq_insert_request, so fold the two into one
slightly more generic function by making the flush special case a bit
smarted.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NJens Axboe <axboe@fb.com>

BLK_MQ_F_* flags are for hctx->flags, and are non-atomic and
set at registration time. BLK_MQ_S_* flags are dynamic and
atomic, and are accessed through hctx->state.

Some of the BLK_MQ_S_STOPPED uses were wrong. Additionally,
the header file should not use a bit shift for the _S_ flags,
as they are done through the set/test_bit functions.
Signed-off-by: NJens Axboe <axboe@fb.com>

If drivers do dynamic allocation in the hardware command init
path, then we need to be able to handle and return failures.

And if they do allocations or mappings in the init command path,
then we need a cleanup function to free up that space at exit
time. So add blk_mq_free_commands() as the cleanup function.

This is required for the mtip32xx driver conversion to blk-mq.
Signed-off-by: NJens Axboe <axboe@fb.com>

Add a new blk_mq_end_io_partial function to partially complete requests
as needed by the SCSI layer.  We do this by reusing blk_update_request
to advance the bio instead of having a simplified version of it in
the blk-mq code.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NJens Axboe <axboe@fb.com>

It's almost identical to blk_mq_insert_request, so fold the two into one
slightly more generic function by making the flush special case a bit
smarted.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NJens Axboe <axboe@fb.com>

Witch to using a preallocated flush_rq for blk-mq similar to what's done
with the old request path.  This allows us to set up the request properly
with a tag from the actually allowed range and ->rq_disk as needed by
some drivers.  To make life easier we also switch to dynamic allocation
of ->flush_rq for the old path.

This effectively reverts most of

    "blk-mq: fix for flush deadlock"

and

    "blk-mq: Don't reserve a tag for flush request"
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NJens Axboe <axboe@fb.com>

Rework I/O completions to work more like the old code path.  blk_mq_end_io
now stays out of the business of deferring completions to others CPUs
and calling blk_mark_rq_complete.  The latter is very important to allow
completing requests that have timed out and thus are already marked completed,
the former allows using the IPI callout even for driver specific completions
instead of having to reimplement them.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NJens Axboe <axboe@fb.com>

This is neede for proper SG_IO operation as well as various uses of
blk_execute_rq from the SCSI midlayer.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NJens Axboe <axboe@fb.com>

Reserving a tag (request) for flush to avoid dead lock is a overkill. A
tag is valuable resource. We can track the number of flush requests and
disallow having too many pending flush requests allocated. With this
patch, blk_mq_alloc_request_pinned() could do a busy nop (but not a dead
loop) if too many pending requests are allocated and new flush request
is allocated. But this should not be a problem, too many pending flush
requests are very rare case.

I verified this can fix the deadlock caused by too many pending flush
requests.
Signed-off-by: NShaohua Li <shli@fusionio.com>
Signed-off-by: NJens Axboe <axboe@kernel.dk>

I observed that there are for_each macros that do an extra memory access
beyond the defined area.
Normally this does not cause problems.
But, this can cause exceptions. For example: if the area is allocated at
the end of a page and the next page is not accessible.

For correctness, I suggest changing the arguments of the 'for loop' like
others 'for_each' do in the kernel.
Signed-off-by: NJose Alonso <joalonsof@gmail.com>
Signed-off-by: NJens Axboe <axboe@kernel.dk>

blk_mq_free_queue() is called from release handler of
queue kobject, so it needn't be called from drivers.

Cc: Jens Axboe <axboe@kernel.dk>
Signed-off-by: NMing Lei <tom.leiming@gmail.com>
Signed-off-by: NJens Axboe <axboe@kernel.dk>

The flush state machine takes in a struct request, which then is
submitted multiple times to the underling driver.  The old block code
requeses the same request for each of those, so it does not have an
issue with tapping into the request pool.  The new one on the other hand
allocates a new request for each of the actualy steps of the flush
sequence. If have already allocated all of the tags for IO, we will
fail allocating the flush request.

Set aside a reserved request just for flushes.
Signed-off-by: NJens Axboe <axboe@kernel.dk>

Add a helper to iterate over all hw queues and stop them.  This is useful
for driver that implement PM suspend functionality.
Signed-off-by: NChristoph Hellwig <hch@lst.de>

Modified to just call blk_mq_stop_hw_queue() by Jens.
Signed-off-by: NJens Axboe <axboe@kernel.dk>

Linux currently has two models for block devices:

- The classic request_fn based approach, where drivers use struct
  request units for IO. The block layer provides various helper
  functionalities to let drivers share code, things like tag
  management, timeout handling, queueing, etc.

- The "stacked" approach, where a driver squeezes in between the
  block layer and IO submitter. Since this bypasses the IO stack,
  driver generally have to manage everything themselves.

With drivers being written for new high IOPS devices, the classic
request_fn based driver doesn't work well enough. The design dates
back to when both SMP and high IOPS was rare. It has problems with
scaling to bigger machines, and runs into scaling issues even on
smaller machines when you have IOPS in the hundreds of thousands
per device.

The stacked approach is then most often selected as the model
for the driver. But this means that everybody has to re-invent
everything, and along with that we get all the problems again
that the shared approach solved.

This commit introduces blk-mq, block multi queue support. The
design is centered around per-cpu queues for queueing IO, which
then funnel down into x number of hardware submission queues.
We might have a 1:1 mapping between the two, or it might be
an N:M mapping. That all depends on what the hardware supports.

blk-mq provides various helper functions, which include:

- Scalable support for request tagging. Most devices need to
  be able to uniquely identify a request both in the driver and
  to the hardware. The tagging uses per-cpu caches for freed
  tags, to enable cache hot reuse.

- Timeout handling without tracking request on a per-device
  basis. Basically the driver should be able to get a notification,
  if a request happens to fail.

- Optional support for non 1:1 mappings between issue and
  submission queues. blk-mq can redirect IO completions to the
  desired location.

- Support for per-request payloads. Drivers almost always need
  to associate a request structure with some driver private
  command structure. Drivers can tell blk-mq this at init time,
  and then any request handed to the driver will have the
  required size of memory associated with it.

- Support for merging of IO, and plugging. The stacked model
  gets neither of these. Even for high IOPS devices, merging
  sequential IO reduces per-command overhead and thus
  increases bandwidth.

For now, this is provided as a potential 3rd queueing model, with
the hope being that, as it matures, it can replace both the classic
and stacked model. That would get us back to having just 1 real
model for block devices, leaving the stacked approach to dm/md
devices (as it was originally intended).

Contributions in this patch from the following people:

Shaohua Li <shli@fusionio.com>
Alexander Gordeev <agordeev@redhat.com>
Christoph Hellwig <hch@infradead.org>
Mike Christie <michaelc@cs.wisc.edu>
Matias Bjorling <m@bjorling.me>
Jeff Moyer <jmoyer@redhat.com>
Acked-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NJens Axboe <axboe@kernel.dk>