When throttle group limits are updated through cgroups, a thread is
woken up to process these updates. While reviewing that code, oleg noted
couple of race conditions existed in the code and he also suggested that
code can be simplified.
This patch fixes the races simplifies the code based on Oleg's suggestions:
- Use xchg().
- Introduced a common function throtl_update_blkio_group_common()
which is shared now by all iops/bps update functions.
Reviewed-by: Oleg Nesterov <oleg@redhat.com>
Reviewed-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
Fixed a merge issue, throtl_schedule_delayed_work() takes throtl_data
as the argument now, not the queue.
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
With the help of cgroup interface one can go and upate the bps/iops
limits of existing group. Once the limits are udpated, a thread is
woken up to see if some blocked group needs recalculation based on new
limits and needs to be requeued.
There was also a piece of code where I was checking for group limit
update when a fresh bio comes in. This patch gets rid of that piece of
code and keeps processing the limit change at one place
throtl_process_limit_change(). It just keeps the code simple and easy
to understand.
Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
The update_vdisktime logic is broken since commit
b54ce60eb7, st->min_vdisktime never makes
a progress. Fix it.
Thanks Vivek for pointing it out.
Signed-off-by: Gui Jianfeng <guijianfen@cn.fujitsu.com>
Acked-by: Vivek Goyal <vgoyal@redhat.com>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
If there are a sync and an async queue and the sync queue's think time
is small, we can ignore the sync queue's dispatch quantum. Because the
sync queue will always preempt the async queue, we don't need to care
about async's latency. This can fix a performance regression of
aiostress test, which is introduced by commit f8ae6e3eb8. The issue
should exist even without the commit, but the commit amplifies the
impact.
The initial post does the same optimization for RT queue too, but since
I have no real workload for it, Vivek suggests to drop it.
Signed-off-by: Shaohua Li <shaohua.li@intel.com>
Reviewed-by: Gui Jianfeng <guijianfeng@cn.fujitsu.com>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
This merge creates two set of conflicts. One is simple context
conflicts caused by removal of throtl_scheduled_delayed_work() in
for-linus and removal of throtl_shutdown_timer_wq() in
for-2.6.39/core.
The other is caused by commit 255bb490c8 (block: blk-flush shouldn't
call directly into q->request_fn() __blk_run_queue()) in for-linus
crashing with FLUSH reimplementation in for-2.6.39/core. The conflict
isn't trivial but the resolution is straight-forward.
* __blk_run_queue() calls in flush_end_io() and flush_data_end_io()
should be called with @force_kblockd set to %true.
* elv_insert() in blk_kick_flush() should use
%ELEVATOR_INSERT_REQUEUE.
Both changes are to avoid invoking ->request_fn() directly from
request completion path and closely match the changes in the commit
255bb490c8.
Signed-off-by: Tejun Heo <tj@kernel.org>
If we enable trace events to trace block actions, We use
blk_fill_rwbs_rq to analyze the corresponding actions
in request's cmd_flags, but we only choose the minor 2 bits
from it, so most of other flags(e.g, REQ_SYNC) are missing.
For example, with a sync write we get:
write_test-2409 [001] 160.013869: block_rq_insert: 3,64 W 0 () 258135 + =
8 [write_test]
Since now we have integrated the flags of both bio and request,
it is safe to pass rq->cmd_flags directly to blk_fill_rwbs and
blk_fill_rwbs_rq isn't needed any more.
With this patch, after a sync write we get:
write_test-2417 [000] 226.603878: block_rq_insert: 3,64 WS 0 () 258135 +=
8 [write_test]
Signed-off-by: Tao Ma <boyu.mt@taobao.com>
Acked-by: Jeff Moyer <jmoyer@redhat.com>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
Move blk_throtl_exit() in blk_cleanup_queue() as blk_throtl_exit() is
written in such a way that it needs queue lock. In blk_release_queue()
there is no gurantee that ->queue_lock is still around.
Initially blk_throtl_exit() was in blk_cleanup_queue() but Ingo reported
one problem.
https://lkml.org/lkml/2010/10/23/86
And a quick fix moved blk_throtl_exit() to blk_release_queue().
commit 7ad58c0286
Author: Jens Axboe <jaxboe@fusionio.com>
Date: Sat Oct 23 20:40:26 2010 +0200
block: fix use-after-free bug in blk throttle code
This patch reverts above change and does not try to shutdown the
throtl work in blk_sync_queue(). By avoiding call to
throtl_shutdown_timer_wq() from blk_sync_queue(), we should also avoid
the problem reported by Ingo.
blk_sync_queue() seems to be used only by md driver and it seems to be
using it to make sure q->unplug_fn is not called as md registers its
own unplug functions and it is about to free up the data structures
used by unplug_fn(). Block throttle does not call back into unplug_fn()
or into md. So there is no need to cancel blk throttle work.
In fact I think cancelling block throttle work is bad because it might
happen that some bios are throttled and scheduled to be dispatched later
with the help of pending work and if work is cancelled, these bios might
never be dispatched.
Block layer also uses blk_sync_queue() during blk_cleanup_queue() and
blk_release_queue() time. That should be safe as we are also calling
blk_throtl_exit() which should make sure all the throttling related
data structures are cleaned up.
Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
Now we initialize ->queue_lock at queue allocation time so driver does
not have to worry about initializing it before calling
blk_cleanup_queue().
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
There does not seem to be a clear convention whether q->queue_lock is
initialized or not when blk_cleanup_queue() is called. In the past it
was not necessary but now blk_throtl_exit() takes up queue lock by
default and needs queue lock to be available.
In fact elevator_exit() code also has similar requirement just that it
is less stringent in the sense that elevator_exit() is called only if
elevator is initialized.
Two problems have been noticed because of ambiguity about spin lock
status.
- If a driver calls blk_alloc_queue() and then soon calls
blk_cleanup_queue() almost immediately, (because some other
driver structure allocation failed or some other error happened)
then blk_throtl_exit() will run into issues as queue lock is not
initialized. Loop driver ran into this issue recently and I
noticed error paths in md driver too. Similar error paths should
exist in other drivers too.
- If some driver provided external spin lock and zapped the lock
before blk_cleanup_queue(), then it can lead to issues.
So this patch initializes the default queue lock at queue allocation time.
block throttling code is one of the users of queue lock and it is
initialized at the queue allocation time, so it makes sense to
initialize ->queue_lock also to internal lock. A driver can overide that
lock later. This will take care of the issue where a driver does not have
to worry about initializing the queue lock to default before calling
blk_cleanup_queue()
Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
Rename the numerals in the diskstats_show() into the macros.
Cc: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Liu Yuan <tailai.ly@taobao.com>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
blk-flush decomposes a flush into sequence of multiple requests. On
completion of a request, the next one is queued; however, block layer
must not implicitly call into q->request_fn() directly from completion
path. This makes the queue behave unexpectedly when seen from the
drivers and violates the assumption that q->request_fn() is called
with process context + queue_lock.
This patch makes blk-flush the following two changes to make sure
q->request_fn() is not called directly from request completion path.
- blk_flush_complete_seq_end_io() now asks __blk_run_queue() to always
use kblockd instead of calling directly into q->request_fn().
- queue_next_fseq() uses ELEVATOR_INSERT_REQUEUE instead of
ELEVATOR_INSERT_FRONT so that elv_insert() doesn't try to unplug the
request queue directly.
Reported by Jan in the following threads.
http://thread.gmane.org/gmane.linux.ide/48778http://thread.gmane.org/gmane.linux.ide/48786
stable: applicable to v2.6.37.
Signed-off-by: Tejun Heo <tj@kernel.org>
Reported-by: Jan Beulich <JBeulich@novell.com>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: stable@kernel.org
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
__blk_run_queue() automatically either calls q->request_fn() directly
or schedules kblockd depending on whether the function is recursed.
blk-flush implementation needs to be able to explicitly choose
kblockd. Add @force_kblockd.
All the current users are converted to specify %false for the
parameter and this patch doesn't introduce any behavior change.
stable: This is prerequisite for fixing ide oops caused by the new
blk-flush implementation.
Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: Jan Beulich <JBeulich@novell.com>
Cc: James Bottomley <James.Bottomley@HansenPartnership.com>
Cc: stable@kernel.org
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
Effectively, make group_isolation=1 the default and remove the tunable.
The setting group_isolation=0 was because by default we idle on
sync-noidle tree and on fast devices, this can be very harmful for
throughput.
However, this problem can also be addressed by tuning slice_idle and
possibly group_idle on faster storage devices.
This change simplifies the CFQ code by removing the feature entirely.
Signed-off-by: Justin TerAvest <teravest@google.com>
Acked-by: Vivek Goyal <vgoyal@redhat.com>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
o Dominik Klein reported a system hang issue while doing some blkio
throttling testing.
https://lkml.org/lkml/2011/2/24/173
o Some tracing revealed that CFQ was not dispatching any more jobs as
queue unplug was not happening. And queue unplug was not happening
because unplug work was not being called as there was one throttling
work on same cpu which as not finished yet. And throttling work had not
finished as it was tyring to dispatch a bio to CFQ but all the request
descriptors were consume to it was put to sleep.
o So basically it is a cyclic dependecny between CFQ unplug work and
throtl dispatch work. Tejun suggested that use separate workqueue for
such cases.
o This patch uses a separate workqueue for throttle related work and
does not rely on kblockd workqueue anymore.
Cc: stable@kernel.org
Reported-by: Dominik Klein <dk@in-telegence.net>
Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
Acked-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
Fix new kernel-doc warning in fs/block_dev.c:
Warning(fs/block_dev.c:937): No description found for parameter 'kill_dirty'
Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Several ACPI drivers fail to build if CONFIG_NET is unset, because
they refer to things depending on CONFIG_THERMAL that in turn depends
on CONFIG_NET. However, CONFIG_THERMAL doesn't really need to depend
on CONFIG_NET, because the only part of it requiring CONFIG_NET is
the netlink interface in thermal_sys.c.
Put the netlink interface in thermal_sys.c under #ifdef CONFIG_NET
and remove the dependency of CONFIG_THERMAL on CONFIG_NET from
drivers/thermal/Kconfig.
Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Acked-by: Randy Dunlap <randy.dunlap@oracle.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Len Brown <lenb@kernel.org>
Cc: Stephen Rothwell <sfr@canb.auug.org.au>
Cc: Luming Yu <luming.yu@intel.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
* 'drm-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/airlied/drm-2.6:
drm: fix unsigned vs signed comparison issue in modeset ctl ioctl.
drm/nv50-nvc0: make sure vma is definitely unmapped when destroying bo
* 'perf-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
perf timechart: Fix max number of cpus
perf timechart: Fix black idle boxes in the title
perf hists: Print number of samples, not the period sum
* 'x86-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
x86: Use u32 instead of long to set reset vector back to 0
* 'timers-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
clockevents: Prevent oneshot mode when broadcast device is periodic
* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mszeredi/fuse:
fuse: fix truncate after open
fuse: fix hang of single threaded fuseblk filesystem
* 'upstream-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jlbec/ocfs2:
ocfs2: Check heartbeat mode for kernel stacks only
Ocfs2/refcounttree: Fix a bug for refcounttree to writeback clusters in a right number.
ocfs2: Fix estimate of necessary credits for mkdir
* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tiwai/sound-2.6:
eukrea-tlv320: fix platform_name
ASoC: correct pxa AC97 DAI names
ALSA: hda - Add support for new IDT 92HD98 and 92HD99 codecs
ALSA: HDA: Add ideapad quirk for two Dell machines
ALSA: HDA: Add a new Conexant codec 506e (20590)
ALSA: usb-audio: fix oops due to cleanup race when disconnecting
ASoC: Hook wm_hubs micbiases up to CLK_SYS
ASoC: Correct definition of WM8903_VMID_RES_5K
ASoC: Fix WM8958 default microphone detection argument ordering
ALSA: HDA: Fix mic initialization in VIA auto parser
ALSA: fix one memory leak in sound jack
Commit e2cda32264 ("thp: add pmd mangling generic functions") replaced
some macros in <asm-generic/pgtable.h> with inline functions.
If the functions are to be defined (not all architectures need them)
then struct vm_area_struct must be defined first. So include
<linux/mm_types.h>.
Fixes a build failure seen in Debian:
CC [M] drivers/media/dvb/mantis/mantis_pci.o
In file included from arch/arm/include/asm/pgtable.h:460,
from drivers/media/dvb/mantis/mantis_pci.c:25:
include/asm-generic/pgtable.h: In function 'ptep_test_and_clear_young':
include/asm-generic/pgtable.h:29: error: dereferencing pointer to incomplete type
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
A customer of ours, complained that when setting the reset
vector back to 0, it trashed other data and hung their box.
They noticed when only 4 bytes were set to 0 instead of 8,
everything worked correctly.
Mathew pointed out:
|
| We're supposed to be resetting trampoline_phys_low and
| trampoline_phys_high here, which are two 16-bit values.
| Writing 64 bits is definitely going to overwrite space
| that we're not supposed to be touching.
|
So limit the area modified to u32.
Signed-off-by: Don Zickus <dzickus@redhat.com>
Acked-by: Matthew Garrett <mjg@redhat.com>
Cc: <stable@kernel.org>
LKML-Reference: <1297139100-424-1-git-send-email-dzickus@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Currently numcpus is determined in pid_put_sample which is only
called on sched_switch/sched_wakeup sample processing.
On a machine with a lot cpus I often saw the last cpu missing.
Check for (max) numcpus on every event happening and in the
beginning. -> fixes the issue for me.
Signed-off-by: Thomas Renninger <trenn@suse.de>
Cc: Arjan van de Ven <arjan@linux.intel.com>
Cc: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Cc: lenb@kernel.org
LKML-Reference: <1298842606-55712-6-git-send-email-trenn@suse.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
This fix is needed for eye of gnome and firefox svg viewers.
Only Inkscape can handle the broken case.
Compare with the other svg_legenda_box declarations, looks
like a typo slipped in at this place.
Signed-off-by: Thomas Renninger <trenn@suse.de>
Cc: Arjan van de Ven <arjan@linux.intel.com>
Cc: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Cc: lenb@kernel.org
LKML-Reference: <1298842606-55712-5-git-send-email-trenn@suse.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Somehow fixes a misrendering + hang at GDM startup on my NVA8...
My first guess would have been stale TLB entries laying around that a new
bo then accidentally inherits. That doesn't make a great deal of sense
however, as when we mapped the pages for the new bo the TLBs would've
gotten flushed anyway.
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
The device table is required to load modules based on modaliases.
Signed-off-by: Axel Lin <axel.lin@gmail.com>
Acked-by: Wolfram Sang <w.sang@pengutronix.de>
Signed-off-by: Guenter Roeck <guenter.roeck@ericsson.com>
The device table is required to load modules based on modaliases.
Signed-off-by: Axel Lin <axel.lin@gmail.com>
Signed-off-by: Guenter Roeck <guenter.roeck@ericsson.com>
When the per cpu timer is marked CLOCK_EVT_FEAT_C3STOP, then we only
can switch into oneshot mode, when the backup broadcast device
supports oneshot mode as well. Otherwise we would try to switch the
broadcast device into an unsupported mode unconditionally. This went
unnoticed so far as the current available broadcast devices support
oneshot mode. Seth unearthed this problem while debugging and working
around an hpet related BIOS wreckage.
Add the necessary check to tick_is_oneshot_available().
Reported-and-tested-by: Seth Forshee <seth.forshee@canonical.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
LKML-Reference: <alpine.LFD.2.00.1102252231200.2701@localhost6.localdomain6>
Cc: stable@kernel.org # .21 ->
* 'pm-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/suspend-2.6:
PM: Make ACPI wakeup from S5 work again when CONFIG_PM_SLEEP is unset
Fixes sysfs config attribute to allow access to entire 16MB maintenance
space of RapidIO devices.
Signed-off-by: Alexandre Bounine <alexandre.bounine@idt.com>
Cc: Kumar Gala <galak@kernel.crashing.org>
Cc: Matt Porter <mporter@kernel.crashing.org>
Cc: Li Yang <leoli@freescale.com>
Cc: Thomas Moll <thomas.moll@sysgo.com>
Cc: Micha Nelissen <micha@neli.hopto.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Initialize ts_real.flags to fix compiler warning about possible
uninitialized use of this field.
Signed-off-by: Alexander Gordeev <lasaine@lvk.cs.msu.su>
Cc: john stultz <johnstul@us.ibm.com>
Cc: Rodolfo Giometti <giometti@linux.it>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
It seems odd that truncate_inode_pages_range(), called not only when
truncating but also when evicting inodes, has mem_cgroup_uncharge_start
and _end() batching in its second loop to clear up a few leftovers, but
not in its first loop that does almost all the work: add them there too.
Signed-off-by: Hugh Dickins <hughd@google.com>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: Balbir Singh <balbir@linux.vnet.ibm.com>
Acked-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The THP code didn't pass the correct interleaving shift to the memory
policy code. Fix this here by adjusting for the order.
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Reviewed-by: Christoph Lameter <cl@linux.com>
Acked-by: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
A race can occur when io_submit() races with io_destroy():
CPU1 CPU2
io_submit()
do_io_submit()
...
ctx = lookup_ioctx(ctx_id);
io_destroy()
Now do_io_submit() holds the last reference to ctx.
...
queue new AIO
put_ioctx(ctx) - frees ctx with active AIOs
We solve this issue by checking whether ctx is being destroyed in AIO
submission path after adding new AIO to ctx. Then we are guaranteed that
either io_destroy() waits for new AIO or we see that ctx is being
destroyed and bail out.
Cc: Nick Piggin <npiggin@kernel.dk>
Reviewed-by: Jeff Moyer <jmoyer@redhat.com>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
aio-dio-invalidate-failure GPFs in aio_put_req from io_submit.
lookup_ioctx doesn't implement the rcu lookup pattern properly.
rcu_read_lock does not prevent refcount going to zero, so we might take
a refcount on a zero count ioctx.
Fix the bug by atomically testing for zero refcount before incrementing.
[jack@suse.cz: added comment into the code]
Reviewed-by: Jeff Moyer <jmoyer@redhat.com>
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
In linux rtc_time struct, tm_mon range is 0~11, tm_wday range is 0~6,
while in RTC HW REG, month range is 1~12, day of the week range is 1~7,
this patch adjusts difference of them.
The efect of this bug was that most of month will be operated on as the
next month by the hardware (When in Jan it maybe even worse). For
example, if in May, software wrote 4 to the hardware, which handled it as
April. Then the logic would be different between software and hardware,
which would cause weird things to happen.
Signed-off-by: Lei Xu <B33228@freescale.com>
Cc: Alessandro Zummo <a.zummo@towertech.it>
Cc: john stultz <johnstul@us.ibm.com>
Cc: Jack Lan <jack.lan@freescale.com>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The kernel automatically evaluates partition tables of storage devices.
The code for evaluating LDM partitions (in fs/partitions/ldm.c) contains
a bug that causes a kernel oops on certain corrupted LDM partitions. A
kernel subsystem seems to crash, because, after the oops, the kernel no
longer recognizes newly connected storage devices.
The patch changes ldm_parse_vmdb() to Validate the value of vblk_size.
Signed-off-by: Timo Warns <warns@pre-sense.de>
Cc: Eugene Teo <eugeneteo@kernel.sg>
Acked-by: Richard Russon <ldm@flatcap.org>
Cc: Harvey Harrison <harvey.harrison@gmail.com>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
should_continue_reclaim() for reclaim/compaction allows scanning to
continue even if pages are not being reclaimed until the full list is
scanned. In terms of allocation success, this makes sense but potentially
it introduces unwanted latency for high-order allocations such as
transparent hugepages and network jumbo frames that would prefer to fail
the allocation attempt and fallback to order-0 pages. Worse, there is a
potential that the full LRU scan will clear all the young bits, distort
page aging information and potentially push pages into swap that would
have otherwise remained resident.
This patch will stop reclaim/compaction if no pages were reclaimed in the
last SWAP_CLUSTER_MAX pages that were considered. For allocations such as
hugetlbfs that use __GFP_REPEAT and have fewer fallback options, the full
LRU list may still be scanned.
Order-0 allocation should not be affected because RECLAIM_MODE_COMPACTION
is not set so the following avoids the gfp_mask being examined:
if (!(sc->reclaim_mode & RECLAIM_MODE_COMPACTION))
return false;
A tool was developed based on ftrace that tracked the latency of
high-order allocations while transparent hugepage support was enabled and
three benchmarks were run. The "fix-infinite" figures are 2.6.38-rc4 with
Johannes's patch "vmscan: fix zone shrinking exit when scan work is done"
applied.
STREAM Highorder Allocation Latency Statistics
fix-infinite break-early
1 :: Count 10298 10229
1 :: Min 0.4560 0.4640
1 :: Mean 1.0589 1.0183
1 :: Max 14.5990 11.7510
1 :: Stddev 0.5208 0.4719
2 :: Count 2 1
2 :: Min 1.8610 3.7240
2 :: Mean 3.4325 3.7240
2 :: Max 5.0040 3.7240
2 :: Stddev 1.5715 0.0000
9 :: Count 111696 111694
9 :: Min 0.5230 0.4110
9 :: Mean 10.5831 10.5718
9 :: Max 38.4480 43.2900
9 :: Stddev 1.1147 1.1325
Mean time for order-1 allocations is reduced. order-2 looks increased but
with so few allocations, it's not particularly significant. THP mean
allocation latency is also reduced. That said, allocation time varies so
significantly that the reductions are within noise.
Max allocation time is reduced by a significant amount for low-order
allocations but reduced for THP allocations which presumably are now
breaking before reclaim has done enough work.
SysBench Highorder Allocation Latency Statistics
fix-infinite break-early
1 :: Count 15745 15677
1 :: Min 0.4250 0.4550
1 :: Mean 1.1023 1.0810
1 :: Max 14.4590 10.8220
1 :: Stddev 0.5117 0.5100
2 :: Count 1 1
2 :: Min 3.0040 2.1530
2 :: Mean 3.0040 2.1530
2 :: Max 3.0040 2.1530
2 :: Stddev 0.0000 0.0000
9 :: Count 2017 1931
9 :: Min 0.4980 0.7480
9 :: Mean 10.4717 10.3840
9 :: Max 24.9460 26.2500
9 :: Stddev 1.1726 1.1966
Again, mean time for order-1 allocations is reduced while order-2
allocations are too few to draw conclusions from. The mean time for THP
allocations is also slightly reduced albeit the reductions are within
varianes.
Once again, our maximum allocation time is significantly reduced for
low-order allocations and slightly increased for THP allocations.
Anon stream mmap reference Highorder Allocation Latency Statistics
1 :: Count 1376 1790
1 :: Min 0.4940 0.5010
1 :: Mean 1.0289 0.9732
1 :: Max 6.2670 4.2540
1 :: Stddev 0.4142 0.2785
2 :: Count 1 -
2 :: Min 1.9060 -
2 :: Mean 1.9060 -
2 :: Max 1.9060 -
2 :: Stddev 0.0000 -
9 :: Count 11266 11257
9 :: Min 0.4990 0.4940
9 :: Mean 27250.4669 24256.1919
9 :: Max 11439211.0000 6008885.0000
9 :: Stddev 226427.4624 186298.1430
This benchmark creates one thread per CPU which references an amount of
anonymous memory 1.5 times the size of physical RAM. This pounds swap
quite heavily and is intended to exercise THP a bit.
Mean allocation time for order-1 is reduced as before. It's also reduced
for THP allocations but the variations here are pretty massive due to
swap. As before, maximum allocation times are significantly reduced.
Overall, the patch reduces the mean and maximum allocation latencies for
the smaller high-order allocations. This was with Slab configured so it
would be expected to be more significant with Slub which uses these size
allocations more aggressively.
The mean allocation times for THP allocations are also slightly reduced.
The maximum latency was slightly increased as predicted by the comments
due to reclaim/compaction breaking early. However, workloads care more
about the latency of lower-order allocations than THP so it's an
acceptable trade-off.
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Acked-by: Andrea Arcangeli <aarcange@redhat.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Acked-by: Andrea Arcangeli <aarcange@redhat.com>
Acked-by: Rik van Riel <riel@redhat.com>
Cc: Michal Hocko <mhocko@suse.cz>
Cc: Kent Overstreet <kent.overstreet@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
