blk_mq_wait_for_tags() is only able to wait for "normal" tags,
not reserved tags. Pass in which one we should attempt to get
a tag for, so that waiting for reserved tags will work.
Reserved tags are used for internal commands, which are usually
serialized. Hence no waiting generally takes place, but we should
ensure that it actually works if users need that functionality.
Signed-off-by: Jens 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: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@fb.com>
If a requeue event races with a timeout, we can get into the
situation where we attempt to complete a request from the
timeout handler when it's not start anymore. This causes a crash.
So have the timeout handler check that REQ_ATOM_STARTED is still
set on the request - if not, we ignore the event. If this happens,
the request has now been marked as complete. As a consequence, we
need to ensure to clear REQ_ATOM_COMPLETE in blk_mq_start_request(),
as to maintain proper request state.
Signed-off-by: Jens Axboe <axboe@fb.com>
This reverts commit 6a3c8a3ac0.
We need selective clearing of the request to make the init-at-free
time completely safe. Otherwise we end up stomping on
rq->atomic_flags, which we don't want to do.
The patch basically reverts the patch of(blk-mq:
initialize request on allocation) in Jens's tree(already
in -next), and only initialize req->q in allocation
for two reasons:
- presumed cache hotness on completion
- blk_rq_tagged(rq) depends on reset of req->mq_ctx
Signed-off-by: Ming Lei <tom.leiming@gmail.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
type of set->tags is struct blk_mq_tags **.
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Ming Lei <tom.leiming@gmail.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
Avoid memory leak in the failure path.
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Ming Lei <tom.leiming@gmail.com>
Signed-off-by: Jens 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: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens 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: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens 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: Christoph Hellwig <hch@lst.de>
Modified by me to kill the unnecessary preempt disable/enable
in the delayed workqueue handler.
Signed-off-by: Jens Axboe <axboe@fb.com>
Add two unlinkely branches to make sure the resid is initialized correctly
for bidi request pairs, and the second request gets properly freed.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens 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: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@fb.com>
blk_mq_work_fn() is always invoked off the bounded workqueues,
so it can happily preempt among the queues in that set without
causing any issues for blk-mq.
Signed-off-by: Jens Axboe <axboe@fb.com>
UP or CONFIG_PREEMPT_NONE will return 0, and what we really
want to check is whether or not we are on the right CPU.
So don't make PREEMPT part of this, just test the CPU in
the mask directly.
Signed-off-by: Jens 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: Christoph Hellwig <hch@lst.de>
Modular export of blk_mq_{alloc,free}_tagset added by me.
Signed-off-by: Jens Axboe <axboe@fb.com>
If we want to share tag and request allocation between queues we cannot
initialize the request at init/free time, but need to initialize it
at allocation time as it might get used for different queues over its
lifetime.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens 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: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@fb.com>
Drivers shouldn't have to care about the block layer setting aside a
request to implement the flush state machine. We already override the
mq context and tag to make it more transparent, but so far haven't deal
with the driver private data in the request. Make sure to override this
as well, and while we're at it add a proper helper sitting in blk-mq.c
that implements the full impersonation.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@fb.com>
Drivers can reach their private data easily using the blk_mq_rq_to_pdu
helper and don't need req->special. By not initializing it code can
be simplified nicely, and we also shave off a few more instructions from
the I/O path.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens 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: Jens Axboe <axboe@fb.com>
When a CPU is unplugged, we move the blk_mq_ctx request entries
to the current queue. The current code forgets to remap the
blk_mq_hw_ctx before marking the software context pending,
which breaks if old-cpu and new-cpu don't map to the same
hardware queue.
Additionally, if we mark entries as pending in the new
hardware queue, then make sure we schedule it for running.
Otherwise request could be sitting there until someone else
queues IO for that hardware queue.
Signed-off-by: Jens Axboe <axboe@fb.com>
Add REQ_SYNC early, so rq_dispatched[] in blk_mq_rq_ctx_init
is set correctly.
Signed-off-by: Shaohua Li<shli@fusionio.com>
Signed-off-by: Jens 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: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens 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: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@fb.com>
There's only one caller, which is a straight wrapper and fits the naming
scheme of the related functions a lot better.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@fb.com>
Now that we are out of initial debug/bringup mode, remove
the verbose dump of the mapping table.
Provide the mapping table in sysfs, under the hardware queue
directory, in the cpu_list file.
Signed-off-by: Jens 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: Jens 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: Jens Axboe <axboe@fb.com>
trace_block_rq_complete does not take into account that request can
be partially completed, so we can get the following incorrect output
of blkparser:
C R 232 + 240 [0]
C R 240 + 232 [0]
C R 248 + 224 [0]
C R 256 + 216 [0]
but should be:
C R 232 + 8 [0]
C R 240 + 8 [0]
C R 248 + 8 [0]
C R 256 + 8 [0]
Also, the whole output summary statistics of completed requests and
final throughput will be incorrect.
This patch takes into account real completion size of the request and
fixes wrong completion accounting.
Signed-off-by: Roman Pen <r.peniaev@gmail.com>
CC: Steven Rostedt <rostedt@goodmis.org>
CC: Frederic Weisbecker <fweisbec@gmail.com>
CC: Ingo Molnar <mingo@redhat.com>
CC: linux-kernel@vger.kernel.org
Cc: stable@kernel.org
Signed-off-by: Jens Axboe <axboe@fb.com>
The name __smp_call_function_single() doesn't tell much about the
properties of this function, especially when compared to
smp_call_function_single().
The comments above the implementation are also misleading. The main
point of this function is actually not to be able to embed the csd
in an object. This is actually a requirement that result from the
purpose of this function which is to raise an IPI asynchronously.
As such it can be called with interrupts disabled. And this feature
comes at the cost of the caller who then needs to serialize the
IPIs on this csd.
Lets rename the function and enhance the comments so that they reflect
these properties.
Suggested-by: Christoph Hellwig <hch@infradead.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jan Kara <jack@suse.cz>
Cc: Jens Axboe <axboe@fb.com>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
The main point of calling __smp_call_function_single() is to send
an IPI in a pure asynchronous way. By embedding a csd in an object,
a caller can send the IPI without waiting for a previous one to complete
as is required by smp_call_function_single() for example. As such,
sending this kind of IPI can be safe even when irqs are disabled.
This flexibility comes at the expense of the caller who then needs to
synchronize the csd lifecycle by himself and make sure that IPIs on a
single csd are serialized.
This is how __smp_call_function_single() works when wait = 0 and this
usecase is relevant.
Now there don't seem to be any usecase with wait = 1 that can't be
covered by smp_call_function_single() instead, which is safer. Lets look
at the two possible scenario:
1) The user calls __smp_call_function_single(wait = 1) on a csd embedded
in an object. It looks like a nice and convenient pattern at the first
sight because we can then retrieve the object from the IPI handler easily.
But actually it is a waste of memory space in the object since the csd
can be allocated from the stack by smp_call_function_single(wait = 1)
and the object can be passed an the IPI argument.
Besides that, embedding the csd in an object is more error prone
because the caller must take care of the serialization of the IPIs
for this csd.
2) The user calls __smp_call_function_single(wait = 1) on a csd that
is allocated on the stack. It's ok but smp_call_function_single()
can do it as well and it already takes care of the allocation on the
stack. Again it's more simple and less error prone.
Therefore, using the underscore prepend API version with wait = 1
is a bad pattern and a sign that the caller can do safer and more
simple.
There was a single user of that which has just been converted.
So lets remove this option to discourage further users.
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jan Kara <jack@suse.cz>
Cc: Jens Axboe <axboe@fb.com>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
Make sure we have a proper pairing between starting and requeueing
requests. Move the dma drain and REQ_END setup into blk_mq_start_request,
and make sure blk_mq_requeue_request properly undoes them, giving us
a pair of function to prepare and unprepare a request without leaving
side effects.
Together this ensures we always clean up properly after
BLK_MQ_RQ_QUEUE_BUSY returns from ->queue_rq.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@fb.com>
rq->errors never has been part of the communication protocol between drivers
and the block stack and most drivers will not have initialized it.
Return -EIO to upper layers when the driver returns BLK_MQ_RQ_QUEUE_ERROR
unconditionally. If a driver want to return a different error it can easily
do so by returning success after calling blk_mq_end_io itself.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens 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: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens 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: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@fb.com>
This patch adds the missing bio_integrity_enabled() +
bio_integrity_prep() setup into blk_mq_make_request()
in order to use DIF protection with scsi-mq.
Cc: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Nicholas Bellinger <nab@linux-iscsi.org>
Signed-off-by: Jens Axboe <axboe@fb.com>
Make blk-mq handle the dma_drain_size field the same way as the old request
path.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens 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: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens 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: Shaohua Li <shli@fusionio.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
'struct page' has two list_head fields: 'lru' and 'list'. Conveniently,
they are unioned together. This means that code can use them
interchangably, which gets horribly confusing.
The blk-mq made the logical decision to try to use page->list. But, that
field was actually introduced just for the slub code. ->lru is the right
field to use outside of slab/slub.
Signed-off-by: Dave Hansen <dave.hansen@linux.intel.com>
Acked-by: David Rientjes <rientjes@google.com>
Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
__smp_call_function_single already avoids multiple IPIs by internally
queing up the items, and now also is available for non-SMP builds as
a trivially correct stub, so there is no need to wrap it. If the
additional lock roundtrip cause problems my patch to convert the
generic IPI code to llists is waiting to get merged will fix it.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
blk_rq_init() is called in req's complete handler to initialize
the request, so the members of start_time and start_time_ns might
become inaccurate when it is allocated in future.
The patch initializes the two members in blk_mq_rq_ctx_init() to
fix the problem.
Signed-off-by: Ming Lei <tom.leiming@gmail.com>
Signed-off-by: Jens 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: Ming Lei <tom.leiming@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
This patch moves synchronization on mq->delay_work
from blk_mq_free_queue() to blk_sync_queue(), so that
blk_sync_queue can work on mq.
Cc: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Ming Lei <tom.leiming@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
blk_mq_drain_queue() is introduced so that we can drain
mq queue inside blk_cleanup_queue().
Also don't accept new requests any more if queue is marked
as dying.
Cc: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Ming Lei <tom.leiming@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1
iQEcBAABAgAGBQJSwLfoAAoJEHm+PkMAQRiGi6QH/1U1B7lmHChDTw3jj1lfm9gA
189Si4QJlnxFWCKHvKEL+pcaVuACU+aMGI8+KyMYK4/JfuWVjjj5fr/SvyHH2/8m
LdSK8aHMhJ46uBS4WJ/l6v46qQa5e2vn8RKSBAyKm/h4vpt+hd6zJdoFrFai4th7
k/TAwOAEHI5uzexUChwLlUBRTvbq4U8QUvDu+DeifC8cT63CGaaJ4qVzjOZrx1an
eP6UXZrKDASZs7RU950i7xnFVDQu4PsjlZi25udsbeiKcZJgPqGgXz5ULf8ZH8RQ
YCi1JOnTJRGGjyIOyLj7pyB01h7XiSM2+eMQ0S7g54F2s7gCJ58c2UwQX45vRWU=
=/4/R
-----END PGP SIGNATURE-----
Merge tag 'v3.13-rc6' into for-3.14/core
Needed to bring blk-mq uptodate, since changes have been going in
since for-3.14/core was established.
Fixup merge issues related to the immutable biovec changes.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Conflicts:
block/blk-flush.c
fs/btrfs/check-integrity.c
fs/btrfs/extent_io.c
fs/btrfs/scrub.c
fs/logfs/dev_bdev.c
If accounting is on, we will do the IO completion accounting after
we have freed the request. Fix that by moving it sooner instead.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
If __GFP_WAIT isn't set and we fail allocating, when we go
to drop the reference on the ctx, we will attempt to dereference
the NULL rq. Fix that.
Signed-off-by: Jeff Moyer <jmoyer@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
If disk stats are enabled on the queue, a request needs to
be marked with REQ_IO_STAT for accounting to be active on
that request. This fixes an issue with virtio-blk not
showing up in /proc/diskstats after the conversion to
blk-mq.
Add QUEUE_FLAG_MQ_DEFAULT, setting stats and same cpu-group
completion on by default.
Reported-by: Dave Chinner <david@fromorbit.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Pull second round of block driver updates from Jens Axboe:
"As mentioned in the original pull request, the bcache bits were pulled
because of their dependency on the immutable bio vecs. Kent re-did
this part and resubmitted it, so here's the 2nd round of (mostly)
driver updates for 3.13. It contains:
- The bcache work from Kent.
- Conversion of virtio-blk to blk-mq. This removes the bio and request
path, and substitutes with the blk-mq path instead. The end result
almost 200 deleted lines. Patch is acked by Asias and Christoph, who
both did a bunch of testing.
- A removal of bootmem.h include from Grygorii Strashko, part of a
larger series of his killing the dependency on that header file.
- Removal of __cpuinit from blk-mq from Paul Gortmaker"
* 'for-linus' of git://git.kernel.dk/linux-block: (56 commits)
virtio_blk: blk-mq support
blk-mq: remove newly added instances of __cpuinit
bcache: defensively handle format strings
bcache: Bypass torture test
bcache: Delete some slower inline asm
bcache: Use ida for bcache block dev minor
bcache: Fix sysfs splat on shutdown with flash only devs
bcache: Better full stripe scanning
bcache: Have btree_split() insert into parent directly
bcache: Move spinlock into struct time_stats
bcache: Kill sequential_merge option
bcache: Kill bch_next_recurse_key()
bcache: Avoid deadlocking in garbage collection
bcache: Incremental gc
bcache: Add make_btree_freeing_key()
bcache: Add btree_node_write_sync()
bcache: PRECEDING_KEY()
bcache: bch_(btree|extent)_ptr_invalid()
bcache: Don't bother with bucket refcount for btree node allocations
bcache: Debug code improvements
...
We've switched over every architecture that supports SMP to it, so
remove the new useless config variable.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Cc: Jan Kara <jack@suse.cz>
Cc: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The new blk-mq code added new instances of __cpuinit usage.
We removed this a couple versions ago; we now want to remove
the compat no-op stubs. Introducing new users is not what
we want to see at this point in time, as it will break once
the stubs are gone.
Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
We switched to plug mq_list for mq, but some code are still using old list.
Signed-off-by: Shaohua Li <shli@fusionio.com>
Signed-off-by: Jens 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: Jens 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: Christoph Hellwig <hch@lst.de>
Modified to just call blk_mq_stop_hw_queue() by Jens.
Signed-off-by: Jens 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: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>