Commit ae1b153962, block: reimplement
FLUSH/FUA to support merge, introduced a performance regression when
running any sort of fsyncing workload using dm-multipath and certain
storage (in our case, an HP EVA). The test I ran was fs_mark, and it
dropped from ~800 files/sec on ext4 to ~100 files/sec. It turns out
that dm-multipath always advertised flush+fua support, and passed
commands on down the stack, where those flags used to get stripped off.
The above commit changed that behavior:
static inline struct request *__elv_next_request(struct request_queue *q)
{
struct request *rq;
while (1) {
- while (!list_empty(&q->queue_head)) {
+ if (!list_empty(&q->queue_head)) {
rq = list_entry_rq(q->queue_head.next);
- if (!(rq->cmd_flags & (REQ_FLUSH | REQ_FUA)) ||
- (rq->cmd_flags & REQ_FLUSH_SEQ))
- return rq;
- rq = blk_do_flush(q, rq);
- if (rq)
- return rq;
+ return rq;
}
Note that previously, a command would come in here, have
REQ_FLUSH|REQ_FUA set, and then get handed off to blk_do_flush:
struct request *blk_do_flush(struct request_queue *q, struct request *rq)
{
unsigned int fflags = q->flush_flags; /* may change, cache it */
bool has_flush = fflags & REQ_FLUSH, has_fua = fflags & REQ_FUA;
bool do_preflush = has_flush && (rq->cmd_flags & REQ_FLUSH);
bool do_postflush = has_flush && !has_fua && (rq->cmd_flags &
REQ_FUA);
unsigned skip = 0;
...
if (blk_rq_sectors(rq) && !do_preflush && !do_postflush) {
rq->cmd_flags &= ~REQ_FLUSH;
if (!has_fua)
rq->cmd_flags &= ~REQ_FUA;
return rq;
}
So, the flush machinery was bypassed in such cases (q->flush_flags == 0
&& rq->cmd_flags & (REQ_FLUSH|REQ_FUA)).
Now, however, we don't get into the flush machinery at all. Instead,
__elv_next_request just hands a request with flush and fua bits set to
the scsi_request_fn, even if the underlying request_queue does not
support flush or fua.
The agreed upon approach is to fix the flush machinery to allow
stacking. While this isn't used in practice (since there is only one
request-based dm target, and that target will now reflect the flush
flags of the underlying device), it does future-proof the solution, and
make it function as designed.
In order to make this work, I had to add a field to the struct request,
inside the flush structure (to store the original req->end_io). Shaohua
had suggested overloading the union with rb_node and completion_data,
but the completion data is used by device mapper and can also be used by
other drivers. So, I didn't see a way around the additional field.
I tested this patch on an HP EVA with both ext4 and xfs, and it recovers
the lost performance. Comments and other testers, as always, are
appreciated.
Cheers,
Jeff
Signed-off-by: Jeff Moyer <jmoyer@redhat.com>
Acked-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
This moves the FC classes bsg code to the block layer and
makes it a lib so that other classes like iscsi and SAS can use it.
It is helpful because working with the request queue, bios,
creating scatterlists, etc are a pain that the LLD does not
have to worry about with normal IOs and should not have to
worry about for bsg requests.
Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
Some systems benefit from completions always being steered to the strict
requester cpu rather than the looser "per-socket" steering that
blk_cpu_to_group() attempts by default. This is because the first
CPU in the group mask ends up being completely overloaded with work,
while the others (including the original submitter) has power left
to spare.
Allow the strict mode to be set by writing '2' to the sysfs control
file. This is identical to the scheme used for the nomerges file,
where '2' is a more aggressive setting than just being turned on.
echo 2 > /sys/block/<bdev>/queue/rq_affinity
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Roland Dreier <roland@purestorage.com>
Tested-by: Dave Jiang <dave.jiang@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
Reorder request_queue to remove 16 bytes of alignment padding in 64 bit
builds.
On my config this shrinks the size of this structure from 1608 to 1592
bytes and therefore needs one fewer cachelines.
Also trivially move the open bracket { to be on the same line as the
structure name to make it easier to grep.
Signed-off-by: Richard Kennedy <richard@rsk.demon.co.uk>
Acked-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
I'm often confused why not disable preempt when changing blk_plug list. It
would be better to add comments here in case others have the similar concerns.
Signed-off-by: Shaohua Li <shaohua.li@intel.com>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
When I test fio script with big I/O depth, I found the total throughput drops
compared to some relative small I/O depth. The reason is the thread accumulates
big requests in its plug list and causes some delays (surely this depends
on CPU speed).
I thought we'd better have a threshold for requests. When a threshold reaches,
this means there is no request merge and queue lock contention isn't severe
when pushing per-task requests to queue, so the main advantages of blk plug
don't exist. We can force a plug list flush in this case.
With this, my test throughput actually increases and almost equals to small
I/O depth. Another side effect is irq off time decreases in blk_flush_plug_list()
for big I/O depth.
The BLK_MAX_REQUEST_COUNT is choosen arbitarily, but 16 is efficiently to
reduce lock contention to me. But I'm open here, 32 is ok in my test too.
Signed-off-by: Shaohua Li <shaohua.li@intel.com>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
There is not a function rq_init but blk_rq_init in block/blk-core.c.
Signed-off-by: Wanlong Gao <wanlong.gao@gmail.com>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
Since those defined functions require additional semicolon
from the caller, they could cause potential syntax errors
when used in if-else statements.
Signed-off-by: Namhyung Kim <namhyung@gmail.com>
Acked-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
Since for-2.6.40/core was forked off the 2.6.39 devel tree, we've
had churn in the core area that makes it difficult to handle
patches for eg cfq or blk-throttle. Instead of requiring that they
be based in older versions with bugs that have been fixed later
in the rc cycle, merge in 2.6.39 final.
Also fixes up conflicts in the below files.
Conflicts:
drivers/block/paride/pcd.c
drivers/cdrom/viocd.c
drivers/ide/ide-cd.c
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
In some cases we would end up stacking discard_zeroes_data incorrectly.
Fix this by enabling the feature by default for stacking drivers and
clearing it for low-level drivers. Incorporating a device that does not
support dzd will then cause the feature to be disabled in the stacking
driver.
Also ensure that the maximum discard value does not overflow when
exported in sysfs and return 0 in the alignment and dzd fields for
devices that don't support discard.
Reported-by: Lukas Czerner <lczerner@redhat.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Acked-by: Mike Snitzer <snitzer@redhat.com>
Cc: stable@kernel.org
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
In some drives, flush requests are non-queueable. When flush request is
running, normal read/write requests can't run. If block layer dispatches
such request, driver can't handle it and requeue it. Tejun suggested we
can hold the queue when flush is running. This can avoid unnecessary
requeue. Also this can improve performance. For example, we have
request flush1, write1, flush 2. flush1 is dispatched, then queue is
hold, write1 isn't inserted to queue. After flush1 is finished, flush2
will be dispatched. Since disk cache is already clean, flush2 will be
finished very soon, so looks like flush2 is folded to flush1.
In my test, the queue holding completely solves a regression introduced by
commit 53d63e6b0d:
block: make the flush insertion use the tail of the dispatch list
It's not a preempt type request, in fact we have to insert it
behind requests that do specify INSERT_FRONT.
which causes about 20% regression running a sysbench fileio
workload.
Stable: 2.6.39 only
Cc: stable@kernel.org
Signed-off-by: Shaohua Li <shaohua.li@intel.com>
Acked-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
flush request isn't queueable in some drives. Add a flag to let driver
notify block layer about this. We can optimize flush performance with the
knowledge.
Stable: 2.6.39 only
Cc: stable@kernel.org
Signed-off-by: Shaohua Li <shaohua.li@intel.com>
Acked-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
We are currently using this flag to check whether it's safe
to call into ->request_fn(). If it is set, we punt to kblockd.
But we get a lot of false positives and excessive punts to
kblockd, which hurts performance.
The only real abuser of this infrastructure is SCSI. So export
the async queue run and convert SCSI over to use that. There's
room for improvement in that SCSI need not always use the async
call, but this fixes our performance issue and they can fix that
up in due time.
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
Instead of overloading __blk_run_queue to force an offload to kblockd
add a new blk_run_queue_async helper to do it explicitly. I've kept
the blk_queue_stopped check for now, but I suspect it's not needed
as the check we do when the workqueue items runs should be enough.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
MD can't use this since it really requires us to be able to
keep more than a single piece of state for the unplug. Commit
048c9374 added the required support for MD, so get rid of this
now unused code.
This reverts commit f75664570d.
Conflicts:
block/blk-core.c
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
md/raid requires an unplug callback, but as it does not uses
requests the current code cannot provide one.
So allow arbitrary callbacks to be attached to the blk_plug.
Signed-off-by: NeilBrown <neilb@suse.de>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
Linus correctly observes that the most important dispatch cases
are now done from kblockd, this isn't ideal for latency reasons.
The original reason for switching dispatches out-of-line was to
avoid too deep a stack, so by _only_ letting the "accidental"
flush directly in schedule() be guarded by offload to kblockd,
we should be able to get the best of both worlds.
So add a blk_schedule_flush_plug() that offloads to kblockd,
and only use that from the schedule() path.
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
For the explicit unplugging, we'd prefer to kick things off
immediately and not pay the penalty of the latency to switch
to kblockd. So let blk_finish_plug() do the run inline, while
the implicit-on-schedule-out unplug will punt to kblockd.
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
It's a bit of a mess currently. task->plug is being cleared
and reset in __blk_finish_plug(), and blk_finish_plug() is
testing for a NULL plug which cannot happen even from schedule()
anymore since it uses blk_needs_flush_plug() to determine
whether to call into this function at all.
So get rid of some of the cruft.
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
MD would like to know when a queue is unplugged, so it can flush
it's bitmap writes. Add such a callback.
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
The current block integrity (DIF/DIX) support in DM is verifying that
all devices' integrity profiles match during DM device resume (which
is past the point of no return). To some degree that is unavoidable
(stacked DM devices force this late checking). But for most DM
devices (which aren't stacking on other DM devices) the ideal time to
verify all integrity profiles match is during table load.
Introduce the notion of an "initialized" integrity profile: a profile
that was blk_integrity_register()'d with a non-NULL 'blk_integrity'
template. Add blk_integrity_is_initialized() to allow checking if a
profile was initialized.
Update DM integrity support to:
- check all devices with _initialized_ integrity profiles match
during table load; uninitialized profiles (e.g. for underlying DM
device(s) of a stacked DM device) are ignored.
- disallow a table load that would result in an integrity profile that
conflicts with a DM device's existing (in-use) integrity profile
- avoid clearing an existing integrity profile
- validate all integrity profiles match during resume; but if they
don't all we can do is report the mismatch (during resume we're past
the point of no return)
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Cc: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
Code has been converted over to the new explicit on-stack plugging,
and delay users have been converted to use the new API for that.
So lets kill off the old plugging along with aops->sync_page().
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
