Currently when the last queue of a group has no request, we don't expire
the queue to hope request from the group comes soon, so the group doesn't
miss its share. But if the think time is big, the assumption isn't correct
and we just waste bandwidth. In such case, we don't do idle.
[global]
runtime=30
direct=1
[test1]
cgroup=test1
cgroup_weight=1000
rw=randread
ioengine=libaio
size=500m
runtime=30
directory=/mnt
filename=file1
thinktime=9000
[test2]
cgroup=test2
cgroup_weight=1000
rw=randread
ioengine=libaio
size=500m
runtime=30
directory=/mnt
filename=file2
patched base
test1 64k 39k
test2 548k 540k
total 604k 578k
group1 gets much better throughput because it waits less time.
To check if the patch changes behavior of queue without think time. I also
tried to give test1 2ms think time or no think time. The test result is stable.
The thoughput doesn't change with/without the patch.
Signed-off-by: Shaohua Li <shaohua.li@intel.com>
Acked-by: Vivek Goyal <vgoyal@redhat.com>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
Currently when the last queue of a service tree has no request, we don't
expire the queue to hope request from the service tree comes soon, so the
service tree doesn't miss its share. But if the think time is big, the
assumption isn't correct and we just waste bandwidth. In such case, we
don't do idle.
[global]
runtime=10
direct=1
[test1]
rw=randread
ioengine=libaio
size=500m
directory=/mnt
filename=file1
thinktime=9000
[test2]
rw=read
ioengine=libaio
size=1G
directory=/mnt
filename=file2
patched base
test1 41k/s 33k/s
test2 15868k/s 15789k/s
total 15902k/s 15817k/s
A slightly better
To check if the patch changes behavior of queue without think time. I also
tried to give test1 2ms think time or no think time. The test has variation
even without the patch, but the average throughput doesn't change with/without
the patch.
Signed-off-by: Shaohua Li <shaohua.li@intel.com>
Acked-by: Vivek Goyal <vgoyal@redhat.com>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
Move the variables to do think time check to a sepatate struct. This is
to prepare adding think time check for service tree and group. No
functional change.
Signed-off-by: Shaohua Li <shaohua.li@intel.com>
Acked-by: Vivek Goyal <vgoyal@redhat.com>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
fs_excl is a poor man's priority inheritance for filesystems to hint to
the block layer that an operation is important. It was never clearly
specified, not widely adopted, and will not prevent starvation in many
cases (like across cgroups).
fs_excl was introduced with the time sliced CFQ IO scheduler, to
indicate when a process held FS exclusive resources and thus needed
a boost.
It doesn't cover all file systems, and it was never fully complete.
Lets kill it.
Signed-off-by: Justin TerAvest <teravest@google.com>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
There is no consistency among filesystems from what bios (or requests)
are marked as being metadata. It's interesting to expose this in traces,
but we shouldn't schedule the requests differently based on whether or
not they're marked as being metadata.
Signed-off-by: Justin TerAvest <teravest@google.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>
Due to the recently identified overflow in read_capacity_16() it was
possible for max_discard_sectors to be zero but still have discards
enabled on the associated device's queue.
Eliminate the possibility for blkdev_issue_discard to infinitely loop.
Interestingly this issue wasn't identified until a device, whose
discard_granularity was 0 due to read_capacity_16 overflow, was consumed
by blk_stack_limits() to construct limits for a higher-level DM
multipath device. The multipath device's resulting limits never had the
discard limits stacked because blk_stack_limits() will only do so if
the bottom device's discard_granularity != 0. This resulted in the
multipath device's limits.max_discard_sectors being 0.
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
On Linux x86_64 host with 32bit userspace, running
qemu or even just "qemu-img create -f qcow2 some.img 1G"
causes a kernel warning:
ioctl32(qemu-img:5296): Unknown cmd fd(3) cmd(00005326){t:'S';sz:0} arg(7fffffff) on some.img
ioctl32(qemu-img:5296): Unknown cmd fd(3) cmd(801c0204){t:02;sz:28} arg(fff77350) on some.img
ioctl 00005326 is CDROM_DRIVE_STATUS,
ioctl 801c0204 is FDGETPRM.
The warning appears because the Linux compat-ioctl handler for these
ioctls only applies to block devices, while qemu also uses the ioctls on
plain files.
Signed-off-by: Johannes Stezenbach <js@sig21.net>
Acked-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
Currently, only open(2) is defined as the 'clearing' point. It has
two roles - first, it's an acknowledgement from userland indicating
that the event has been received and kernel can clear pending states
and proceed to generate more events. Secondly, it's passed on to
device drivers as a hint indicating that a synchronization point has
been reached and it might want to take a deeper look at the device.
The latter currently is only used by sr which uses two different
mechanisms - GET_EVENT_MEDIA_STATUS_NOTIFICATION and TEST_UNIT_READY
to discover events, where the former is lighter weight and safe to be
used repeatedly but may not provide full coverage. Among other
things, GET_EVENT can't detect media removal while TUR can.
This patch makes close(2) - blkdev_put() - indicate clearing hint for
MEDIA_CHANGE to drivers. disk_check_events() is renamed to
disk_flush_events() and updated to take @mask for events to flush
which is or'd to ev->clearing and will be passed to the driver on the
next ->check_events() invocation.
This change makes sr generate MEDIA_CHANGE when media is ejected from
userland - e.g. with eject(1).
Note: Given the current usage, it seems @clearing hint is needlessly
complex. disk_clear_events() can simply clear all events and the hint
can be boolean @flush.
Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: Kay Sievers <kay.sievers@vrfy.org>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
ioc->ioc_data is rcu protectd, so uses correct API to access it.
This doesn't change any behavior, but just make code consistent.
Signed-off-by: Shaohua Li <shaohua.li@intel.com>
Cc: stable@kernel.org # after ab4bd22d
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
I got a rcu warnning at boot. the ioc->ioc_data is rcu_deferenced, but
doesn't hold rcu_read_lock.
Signed-off-by: Shaohua Li <shaohua.li@intel.com>
Cc: stable@kernel.org # after ab4bd22d
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
The total of two unsigned values should also be unsigned.
Update throtl_log output to unsigned.
Update total_nr_queued test to non-zero to be the
same as the other total_nr_queued tests.
Signed-off-by: Joe Perches <joe@perches.com>
Acked-by: Vivek Goyal <vgoyal@redhat.com>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
Use the compiler to verify format strings and arguments.
Fix fallout.
Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
Use the compiler to verify format strings and arguments.
Fix fallout.
Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
disk_block_events() should guarantee that the event work is not in
flight on return and once blocked it shouldn't issue further
cancellations.
Because there was no synchronization between the first blocker doing
cancel_delayed_work_sync() and the following blockers, the following
blockers could finish before cancellation was complete, which broke
both guarantees - event work could be in flight and cancellation could
happen after return.
This bug triggered WARN_ON_ONCE() in disk_clear_events() reported in
bug#34662.
https://bugzilla.kernel.org/show_bug.cgi?id=34662
Fix it by adding an outer mutex which protects both block count
manipulation and work cancellation.
-v2: Use outer mutex instead of bit waitqueue per Linus.
Signed-off-by: Tejun Heo <tj@kernel.org>
Tested-by: Sitsofe Wheeler <sitsofe@yahoo.com>
Reported-by: Sitsofe Wheeler <sitsofe@yahoo.com>
Reported-by: Borislav Petkov <bp@alien8.de>
Reported-by: Meelis Roos <mroos@linux.ee>
Reported-by: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Kay Sievers <kay.sievers@vrfy.org>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
After the previous update to disk_check_events(), nobody is using
non-syncing __disk_block_events(). Remove @sync and, as this makes
__disk_block_events() virtually identical to disk_block_events(),
remove the underscore prefixed version.
Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
This patch is part of fix for triggering of WARN_ON_ONCE() in
disk_clear_events() reported in bug#34662.
https://bugzilla.kernel.org/show_bug.cgi?id=34662
disk_clear_events() blocks events, schedules and flushes the event
work. It expects the work to have started execution on schedule and
finished on return from flush. WARN_ON_ONCE() triggers if the event
work hasn't executed as expected. This problem happens because
__disk_block_events() fails to guarantee that the event work item is
not in flight on return from the function in race-free manner. The
problem is two-fold and this patch addresses one of them.
When __disk_block_events() is called with @sync == %false, it bumps
event block count, calls cancel_delayed_work() and return. This makes
it impossible to guarantee that event polling is not in flight on
return from syncing __disk_block_events() - if the first blocker was
non-syncing, polling could still be in progress and later syncing ones
would assume that the first blocker already canceled it.
Making __disk_block_events() cancel_sync regardless of block count
isn't feasible either as it may race with forced event checking in
disk_clear_events().
As disk_check_events() is the only user of non-syncing
__disk_block_events(), updating it to directly cancel and schedule
event work is the easiest way to solve the issue.
Note that there's another bug in __disk_block_events() and this patch
doesn't fix the issue completely. Later patch will fix the other bug.
Signed-off-by: Tejun Heo <tj@kernel.org>
Tested-by: Sitsofe Wheeler <sitsofe@yahoo.com>
Reported-by: Sitsofe Wheeler <sitsofe@yahoo.com>
Reported-by: Borislav Petkov <bp@alien8.de>
Reported-by: Meelis Roos <mroos@linux.ee>
Reported-by: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Kay Sievers <kay.sievers@vrfy.org>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
If we rename the return of alloc_io_context() and get_io_context() from
"ret" to "ioc" the code get's (a bit) more readable and (a lot) more
grepable.
Signed-off-by: Paul Bolle <pebolle@tiscali.nl>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
Since we are modifying this RCU pointer, we need to hold
the lock protecting it around it.
This fixes a potential reuse and double free of a cfq
io_context structure. The bug has been in CFQ for a long
time, it hit very few people but those it did hit seemed
to see it a lot.
Tracked in RH bugzilla here:
https://bugzilla.redhat.com/show_bug.cgi?id=577968
Credit goes to Paul Bolle for figuring out that the issue
was around the one-hit ioc->ioc_data cache. Thanks to his
hard work the issue is now fixed.
Cc: stable@kernel.org
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
Since we are modifying this RCU pointer, we need to hold
the lock protecting it around it.
This fixes a potential reuse and double free of a cfq
io_context structure. The bug has been in CFQ for a long
time, it hit very few people but those it did hit seemed
to see it a lot.
Tracked in RH bugzilla here:
https://bugzilla.redhat.com/show_bug.cgi?id=577968
Credit goes to Paul Bolle for figuring out that the issue
was around the one-hit ioc->ioc_data cache. Thanks to his
hard work the issue is now fixed.
Cc: stable@kernel.org
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
Hi, Jens,
If you recall, I posted an RFC patch for this back in July of last year:
http://lkml.org/lkml/2010/7/13/279
The basic problem is that a process can issue a never-ending stream of
async direct I/Os to the same sector on a device, thus starving out
other I/O in the system (due to the way the alias handling works in both
cfq and deadline). The solution I proposed back then was to start
dispatching from the fifo after a certain number of aliases had been
dispatched. Vivek asked why we had to treat aliases differently at all,
and I never had a good answer. So, I put together a simple patch which
allows aliases to be added to the rb tree (it adds them to the right,
though that doesn't matter as the order isn't guaranteed anyway). I
think this is the preferred solution, as it doesn't break up time slices
in CFQ or batches in deadline. I've tested it, and it does solve the
starvation issue. Let me know what you think.
Cheers,
Jeff
Signed-off-by: Jeff Moyer <jmoyer@redhat.com>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
list_entry() and hlist_entry() are both simply aliases for
container_of(), but since io_context.cic_list.first is an hlist_node one
should at least use the correct alias.
Signed-off-by: Paul Bolle <pebolle@tiscali.nl>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
queue_fail can only be reached if cic is NULL, so its check for cic must
be bogus.
Signed-off-by: Paul Bolle <pebolle@tiscali.nl>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
Add cgroup subsystem callbacks for per-thread attachment in atomic contexts
Add can_attach_task(), pre_attach(), and attach_task() as new callbacks
for cgroups's subsystem interface. Unlike can_attach and attach, these
are for per-thread operations, to be called potentially many times when
attaching an entire threadgroup.
Also, the old "bool threadgroup" interface is removed, as replaced by
this. All subsystems are modified for the new interface - of note is
cpuset, which requires from/to nodemasks for attach to be globally scoped
(though per-cpuset would work too) to persist from its pre_attach to
attach_task and attach.
This is a pre-patch for cgroup-procs-writable.patch.
Signed-off-by: Ben Blum <bblum@andrew.cmu.edu>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Matt Helsley <matthltc@us.ibm.com>
Reviewed-by: Paul Menage <menage@google.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9fd097b149 (block: unexport DISK_EVENT_MEDIA_CHANGE for legacy/fringe
drivers) removed DISK_EVENT_MEDIA_CHANGE from legacy/fringe block
drivers which have inadequate ->check_events(). Combined with earlier
change 7c88a168da (block: don't propagate unlisted DISK_EVENTs to
userland), this enables using ->check_events() for internal processing
while avoiding enabling in-kernel block event polling which can lead
to infinite event loop.
Unfortunately, this made many drivers including floppy without any bit
set in disk->events and ->async_events in which case disk_add_events()
simply skipped allocation of disk->ev, which disables whole event
handling. As ->check_events() is still used during open processing
for revalidation, this can lead to open failure.
This patch always allocates disk->ev if ->check_events is implemented.
In the long term, it would make sense to simply include the event
structure inline into genhd as it's now used by virtually all block
devices.
Signed-off-by: Tejun Heo <tj@kernel.org>
Reported-by: Ondrej Zary <linux@rainbow-software.org>
Reported-by: Alex Villacis Lasso <avillaci@ceibo.fiec.espol.edu.ec>
Cc: stable@kernel.org
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
When struct cfq_data allocation fails, cic_index need to be freed.
Signed-off-by: Namhyung Kim <namhyung@gmail.com>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
The 'group_changed' variable is initialized to 0 and never changed, so
checking the variable is meaningless.
It is a leftover from 0bbfeb8320 ("cfq-iosched: Always provide group
iosolation."). Let's get rid of it.
Signed-off-by: Namhyung Kim <namhyung@gmail.com>
Cc: Justin TerAvest <teravest@google.com>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
Reduce the number of bit operations in cfq_choose_req() on average
(and worst) cases.
Signed-off-by: Namhyung Kim <namhyung@gmail.com>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
Simplify the calculation in cfq_prio_to_maxrq(), plus replace CFQ_PRIO_LISTS to
IOPRIO_BE_NR since they are the same and IOPRIO_BE_NR looks more reasonable in
this context IMHO.
Signed-off-by: Namhyung Kim <namhyung@gmail.com>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
If we don't explicitly initialize it to zero, CFQ might think that
cgroup of ioc has changed and it generates lots of unnecessary calls
to call_for_each_cic(changed_cgroup). Fix it.
cfq_get_io_context()
cfq_ioc_set_cgroup()
call_for_each_cic(ioc, changed_cgroup)
Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
Commit 73c1010119 ("block: initial patch for on-stack per-task plugging")
removed calls to elv_bio_merged() when @bio merged with @req. Re-add them.
This in turn will update merged stats in associated group. That
should be safe as long as request has got reference to the blkio_group.
Signed-off-by: Namhyung Kim <namhyung@gmail.com>
Cc: Divyesh Shah <dpshah@google.com>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
Make BLKIO_STAT_MERGED per cpu hence gettring rid of need of taking
blkg->stats_lock.
Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
We allocated per cpu stats struct for root group but did not free it.
Fix it.
Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
We don't need them anymore, so kill:
- REQ_ON_PLUG checks in various places
- !rq_mergeable() check in plug merging
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
Currently we take a queue lock on each bio to check if there are any
throttling rules associated with the group and also update the stats.
Now access the group under rcu and update the stats without taking
the queue lock. Queue lock is taken only if there are throttling rules
associated with the group.
So the common case of root group when there are no rules, save
unnecessary pounding of request queue lock.
Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
Now dispatch stats update is lock free. But reset of these stats still
takes blkg->stats_lock and is dependent on that. As stats are per cpu,
we should be able to just reset the stats on each cpu without any locks.
(Atleast for 64bit arch).
On 32bit arch there is a small race where 64bit updates are not atomic.
The result of this race can be that in the presence of other writers,
one might not get 0 value after reset of a stat and might see something
intermediate
One can write more complicated code to cover this race like sending IPI
to other cpus to reset stats and for offline cpus, reset these directly.
Right not I am not taking that path because reset_update is more of a
debug feature and it can happen only on 32bit arch and possibility of
it happening is small. Will fix it if it becomes a real problem. For
the time being going for code simplicity.
Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
Some of the stats are 64bit and updation will be non atomic on 32bit
architecture. Use sequence counters on 32bit arch to make reading
of stats safe.
Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
Currently we take blkg_stat lock for even updating the stats. So even if
a group has no throttling rules (common case for root group), we end
up taking blkg_lock, for updating the stats.
Make dispatch stats per cpu so that these can be updated without taking
blkg lock.
If cpu goes offline, these stats simply disappear. No protection has
been provided for that yet. Do we really need anything for that?
Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
Soon we will allow accessing a throtl_grp under rcu_read_lock(). Hence
start freeing up throtl_grp after one rcu grace period.
Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
Use same helper function for root group as we use with dynamically
allocated groups to add it to various lists.
Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
A helper function for the code which is used at 2-3 places. Makes reading
code little easier.
Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
Currently, we allocate root throtl_grp statically. But as we will be
introducing per cpu stat pointers and that will be allocated
dynamically even for root group, we might as well make whole root
throtl_grp allocation dynamic and treat it in same manner as other
groups.
Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>