Traversing an entire package is not only expensive, it also leads to tasks
bouncing all over a partially idle and possible quite large package. Fix
that up by assigning a 'buddy' CPU to try to motivate. Each buddy may try
to motivate that one other CPU, if it's busy, tough, it may then try its
SMT sibling, but that's all this optimization is allowed to cost.
Sibling cache buddies are cross-wired to prevent bouncing.
4 socket 40 core + SMT Westmere box, single 30 sec tbench runs, higher is better:
clients 1 2 4 8 16 32 64 128
..........................................................................
pre 30 41 118 645 3769 6214 12233 14312
post 299 603 1211 2418 4697 6847 11606 14557
A nice increase in performance.
Signed-off-by: Mike Galbraith <efault@gmx.de>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/1339471112.7352.32.camel@marge.simpson.net
Signed-off-by: Ingo Molnar <mingo@kernel.org>
At present, the functions that deal with cpusets during CPU/Mem hotplug
are quite messy, since a lot of the functionality is mixed up without clear
separation. And this takes a toll on optimization as well. For example,
the function cpuset_update_active_cpus() is called on both CPU offline and CPU
online events; and it invokes scan_for_empty_cpusets(), which makes sense
only for CPU offline events. And hence, the current code ends up unnecessarily
traversing the cpuset tree during CPU online also.
As a first step towards cleaning up those functions, encapsulate the cpuset
tree traversal in a helper function, so as to facilitate upcoming changes.
Signed-off-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/20120524141635.3692.893.stgit@srivatsabhat.in.ibm.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
In the event of CPU hotplug, the kernel modifies the cpusets' cpus_allowed
masks as and when necessary to ensure that the tasks belonging to the cpusets
have some place (online CPUs) to run on. And regular CPU hotplug is
destructive in the sense that the kernel doesn't remember the original cpuset
configurations set by the user, across hotplug operations.
However, suspend/resume (which uses CPU hotplug) is a special case in which
the kernel has the responsibility to restore the system (during resume), to
exactly the same state it was in before suspend.
In order to achieve that, do the following:
1. Don't modify cpusets during suspend/resume. At all.
In particular, don't move the tasks from one cpuset to another, and
don't modify any cpuset's cpus_allowed mask. So, simply ignore cpusets
during the CPU hotplug operations that are carried out in the
suspend/resume path.
2. However, cpusets and sched domains are related. We just want to avoid
altering cpusets alone. So, to keep the sched domains updated, build
a single sched domain (containing all active cpus) during each of the
CPU hotplug operations carried out in s/r path, effectively ignoring
the cpusets' cpus_allowed masks.
(Since userspace is frozen while doing all this, it will go unnoticed.)
3. During the last CPU online operation during resume, build the sched
domains by looking up the (unaltered) cpusets' cpus_allowed masks.
That will bring back the system to the same original state as it was in
before suspend.
Ultimately, this will not only solve the cpuset problem related to suspend
resume (ie., restores the cpusets to exactly what it was before suspend, by
not touching it at all) but also speeds up suspend/resume because we avoid
running cpuset update code for every CPU being offlined/onlined.
Signed-off-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/20120524141611.3692.20155.stgit@srivatsabhat.in.ibm.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Merge emailed kgdb dmesg fixups patches from Anton Vorontsov:
"The dmesg command appears to be broken after the printk rework. The
old logic in the kdb code makes no sense in terms of current
printk/logging storage format, and KDB simply hangs forever upon
entering 'dmesg' command.
The first patch revives the command by switching to kmsg_dumper
iterator. As a side-effect, the code is now much more simpler.
A few changes were needed in the printk.c: we needed unlocked variant
of the kmsg_dumper iterator, but these can surely wait for 3.6.
It's probably too late even for the first patch to go to 3.5, but I'll
try to convince otherwise. :-) Here we go:
- The current code is broken for sure, and has no hope to work at
all. It is a regression
- The new code works for me, and probably works for everyone else;
- If it compiles (and I urge everyone to compile-test it on your
setup), it hardly can make things worse."
* Merge emailed patches from Anton Vorontsov: (4 commits)
kdb: Switch to nolock variants of kmsg_dump functions
printk: Implement some unlocked kmsg_dump functions
printk: Remove kdb_syslog_data
kdb: Revive dmesg command
The locked variants are prone to deadlocks (suppose we got to the
debugger w/ the logbuf lock held), so let's switch to nolock variants.
Signed-off-by: Anton Vorontsov <anton.vorontsov@linaro.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
If used from KDB, the locked variants are prone to deadlocks (suppose we
got to the debugger w/ the logbuf lock held).
So, we have to implement a few routines that grab no logbuf lock.
Yet we don't need these functions in modules, so we don't export them.
Signed-off-by: Anton Vorontsov <anton.vorontsov@linaro.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The kgdb dmesg command is broken after the printk rework. The old logic
in kdb code makes no sense in terms of current printk/logging storage
format, and KDB simply hangs forever.
This patch revives the command by switching to kmsg_dumper iterator.
The code is now much more simpler and shorter.
Signed-off-by: Anton Vorontsov <anton.vorontsov@linaro.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Commit a7a20d1039 ("sd: limit the scope of the async probe domain")
make the SCSI device probing run device discovery in it's own async
domain.
However, as a result, the partition detection was no longer synchronized
by async_synchronize_full() (which, despite the name, only synchronizes
the global async space, not all of them). Which in turn meant that
"wait_for_device_probe()" would not wait for the SCSI partitions to be
parsed.
And "wait_for_device_probe()" was what the boot time init code relied on
for mounting the root filesystem.
Now, most people never noticed this, because not only is it
timing-dependent, but modern distributions all use initrd. So the root
filesystem isn't actually on a disk at all. And then before they
actually mount the final disk filesystem, they will have loaded the
scsi-wait-scan module, which not only does the expected
wait_for_device_probe(), but also does scsi_complete_async_scans().
[ Side note: scsi_complete_async_scans() had also been partially broken,
but that was fixed in commit 43a8d39d01 ("fix async probe
regression"), so that same commit a7a20d1039 had actually broken
setups even if you used scsi-wait-scan explicitly ]
Solve this problem by just moving the scsi_complete_async_scans() call
into wait_for_device_probe(). Everybody who wants to wait for device
probing to finish really wants the SCSI probing to complete, so there's
no reason not to do this.
So now "wait_for_device_probe()" really does what the name implies, and
properly waits for device probing to finish. This also removes the now
unnecessary extra calls to scsi_complete_async_scans().
Reported-and-tested-by: Artem S. Tashkinov <t.artem@mailcity.com>
Cc: Dan Williams <dan.j.williams@gmail.com>
Cc: Alan Stern <stern@rowland.harvard.edu>
Cc: James Bottomley <jbottomley@parallels.com>
Cc: Borislav Petkov <bp@amd64.org>
Cc: linux-scsi <linux-scsi@vger.kernel.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
One more time/ntp fix pulled from Ingo Molnar.
* 'timers-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
ntp: Fix STA_INS/DEL clearing bug
Pull RCU, perf, and scheduler fixes from Ingo Molnar.
The RCU fix is a revert for an optimization that could cause deadlocks.
One of the scheduler commits (164c33c6ad "sched: Fix fork() error path
to not crash") is correct but not complete (some architectures like Tile
are not covered yet) - the resulting additional fixes are still WIP and
Ingo did not want to delay these pending fixes. See this thread on
lkml:
[PATCH] fork: fix error handling in dup_task()
The perf fixes are just trivial oneliners.
* 'core-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
Revert "rcu: Move PREEMPT_RCU preemption to switch_to() invocation"
* 'perf-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
perf kvm: Fix segfault with report and mixed guestmount use
perf kvm: Fix regression with guest machine creation
perf script: Fix format regression due to libtraceevent merge
ring-buffer: Fix accounting of entries when removing pages
ring-buffer: Fix crash due to uninitialized new_pages list head
* 'sched-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
MAINTAINERS/sched: Update scheduler file pattern
sched/nohz: Rewrite and fix load-avg computation -- again
sched: Fix fork() error path to not crash
Pull the leap second fixes from Thomas Gleixner:
"It's a rather large series, but well discussed, refined and reviewed.
It got a massive testing by John, Prarit and tip.
In theory we could split it into two parts. The first two patches
f55a6faa38: hrtimer: Provide clock_was_set_delayed()
4873fa070a: timekeeping: Fix leapsecond triggered load spike issue
are merely preventing the stuff loops forever issues, which people
have observed.
But there is no point in delaying the other 4 commits which achieve
full correctness into 3.6 as they are tagged for stable anyway. And I
rather prefer to have the full fixes merged in bulk than a "prevent
the observable wreckage and deal with the hidden fallout later"
approach."
* 'timers-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
hrtimer: Update hrtimer base offsets each hrtimer_interrupt
timekeeping: Provide hrtimer update function
hrtimers: Move lock held region in hrtimer_interrupt()
timekeeping: Maintain ktime_t based offsets for hrtimers
timekeeping: Fix leapsecond triggered load spike issue
hrtimer: Provide clock_was_set_delayed()
"no other files mapped" requirement from my previous patch (c/r: prctl:
update prctl_set_mm_exe_file() after mm->num_exe_file_vmas removal) is too
paranoid, it forbids operation even if there mapped one shared-anon vma.
Let's check that current mm->exe_file already unmapped, in this case
exe_file symlink already outdated and its changing is reasonable.
Plus, this patch fixes exit code in case operation success.
Signed-off-by: Konstantin Khlebnikov <khlebnikov@openvz.org>
Reported-by: Cyrill Gorcunov <gorcunov@openvz.org>
Tested-by: Cyrill Gorcunov <gorcunov@openvz.org>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Matt Helsley <matthltc@us.ibm.com>
Cc: Kees Cook <keescook@chromium.org>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: Pavel Emelyanov <xemul@parallels.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The update of the hrtimer base offsets on all cpus cannot be made
atomically from the timekeeper.lock held and interrupt disabled region
as smp function calls are not allowed there.
clock_was_set(), which enforces the update on all cpus, is called
either from preemptible process context in case of do_settimeofday()
or from the softirq context when the offset modification happened in
the timer interrupt itself due to a leap second.
In both cases there is a race window for an hrtimer interrupt between
dropping timekeeper lock, enabling interrupts and clock_was_set()
issuing the updates. Any interrupt which arrives in that window will
see the new time but operate on stale offsets.
So we need to make sure that an hrtimer interrupt always sees a
consistent state of time and offsets.
ktime_get_update_offsets() allows us to get the current monotonic time
and update the per cpu hrtimer base offsets from hrtimer_interrupt()
to capture a consistent state of monotonic time and the offsets. The
function replaces the existing ktime_get() calls in hrtimer_interrupt().
The overhead of the new function vs. ktime_get() is minimal as it just
adds two store operations.
This ensures that any changes to realtime or boottime offsets are
noticed and stored into the per-cpu hrtimer base structures, prior to
any hrtimer expiration and guarantees that timers are not expired early.
Signed-off-by: John Stultz <johnstul@us.ibm.com>
Reviewed-by: Ingo Molnar <mingo@kernel.org>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Acked-by: Prarit Bhargava <prarit@redhat.com>
Cc: stable@vger.kernel.org
Link: http://lkml.kernel.org/r/1341960205-56738-8-git-send-email-johnstul@us.ibm.com
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
To finally fix the infamous leap second issue and other race windows
caused by functions which change the offsets between the various time
bases (CLOCK_MONOTONIC, CLOCK_REALTIME and CLOCK_BOOTTIME) we need a
function which atomically gets the current monotonic time and updates
the offsets of CLOCK_REALTIME and CLOCK_BOOTTIME with minimalistic
overhead. The previous patch which provides ktime_t offsets allows us
to make this function almost as cheap as ktime_get() which is going to
be replaced in hrtimer_interrupt().
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Ingo Molnar <mingo@kernel.org>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Acked-by: Prarit Bhargava <prarit@redhat.com>
Cc: stable@vger.kernel.org
Signed-off-by: John Stultz <johnstul@us.ibm.com>
Link: http://lkml.kernel.org/r/1341960205-56738-7-git-send-email-johnstul@us.ibm.com
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
The timekeeping code misses an update of the hrtimer subsystem after a
leap second happened. Due to that timers based on CLOCK_REALTIME are
either expiring a second early or late depending on whether a leap
second has been inserted or deleted until an operation is initiated
which causes that update. Unless the update happens by some other
means this discrepancy between the timekeeping and the hrtimer data
stays forever and timers are expired either early or late.
The reported immediate workaround - $ data -s "`date`" - is causing a
call to clock_was_set() which updates the hrtimer data structures.
See: http://www.sheeri.com/content/mysql-and-leap-second-high-cpu-and-fix
Add the missing clock_was_set() call to update_wall_time() in case of
a leap second event. The actual update is deferred to softirq context
as the necessary smp function call cannot be invoked from hard
interrupt context.
Signed-off-by: John Stultz <johnstul@us.ibm.com>
Reported-by: Jan Engelhardt <jengelh@inai.de>
Reviewed-by: Ingo Molnar <mingo@kernel.org>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Acked-by: Prarit Bhargava <prarit@redhat.com>
Cc: stable@vger.kernel.org
Link: http://lkml.kernel.org/r/1341960205-56738-3-git-send-email-johnstul@us.ibm.com
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
clock_was_set() cannot be called from hard interrupt context because
it calls on_each_cpu().
For fixing the widely reported leap seconds issue it is necessary to
call it from hard interrupt context, i.e. the timer tick code, which
does the timekeeping updates.
Provide a new function which denotes it in the hrtimer cpu base
structure of the cpu on which it is called and raise the hrtimer
softirq. We then execute the clock_was_set() notificiation from
softirq context in run_hrtimer_softirq(). The hrtimer softirq is
rarely used, so polling the flag there is not a performance issue.
[ tglx: Made it depend on CONFIG_HIGH_RES_TIMERS. We really should get
rid of all this ifdeffery ASAP ]
Signed-off-by: John Stultz <johnstul@us.ibm.com>
Reported-by: Jan Engelhardt <jengelh@inai.de>
Reviewed-by: Ingo Molnar <mingo@kernel.org>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Acked-by: Prarit Bhargava <prarit@redhat.com>
Cc: stable@vger.kernel.org
Link: http://lkml.kernel.org/r/1341960205-56738-2-git-send-email-johnstul@us.ibm.com
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Prashanth Nageshappa
Mike Galbraith
Srivatsa S. Bhat
Linus Torvalds
Anton Vorontsov
Thomas Gleixner
John Stultz
Konstantin Khlebnikov