In futex_wake() there is clearly no point in taking the hb->lock
if we know beforehand that there are no tasks to be woken. While
the hash bucket's plist head is a cheap way of knowing this, we
cannot rely 100% on it as there is a racy window between the
futex_wait call and when the task is actually added to the
plist. To this end, we couple it with the spinlock check as
tasks trying to enter the critical region are most likely
potential waiters that will be added to the plist, thus
preventing tasks sleeping forever if wakers don't acknowledge
all possible waiters.
Furthermore, the futex ordering guarantees are preserved,
ensuring that waiters either observe the changed user space
value before blocking or is woken by a concurrent waker. For
wakers, this is done by relying on the barriers in
get_futex_key_refs() -- for archs that do not have implicit mb
in atomic_inc(), we explicitly add them through a new
futex_get_mm function. For waiters we rely on the fact that
spin_lock calls already update the head counter, so spinners
are visible even if the lock hasn't been acquired yet.
For more details please refer to the updated comments in the
code and related discussion:
https://lkml.org/lkml/2013/11/26/556
Special thanks to tglx for careful review and feedback.
Suggested-by: Linus Torvalds <torvalds@linux-foundation.org>
Reviewed-by: Darren Hart <dvhart@linux.intel.com>
Reviewed-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Davidlohr Bueso <davidlohr@hp.com>
Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Jeff Mahoney <jeffm@suse.com>
Cc: Scott Norton <scott.norton@hp.com>
Cc: Tom Vaden <tom.vaden@hp.com>
Cc: Aswin Chandramouleeswaran <aswin@hp.com>
Cc: Waiman Long <Waiman.Long@hp.com>
Cc: Jason Low <jason.low2@hp.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Link: http://lkml.kernel.org/r/1389569486-25487-5-git-send-email-davidlohr@hp.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Currently, the futex global hash table suffers from its fixed,
smallish (for today's standards) size of 256 entries, as well as
its lack of NUMA awareness. Large systems, using many futexes,
can be prone to high amounts of collisions; where these futexes
hash to the same bucket and lead to extra contention on the same
hb->lock. Furthermore, cacheline bouncing is a reality when we
have multiple hb->locks residing on the same cacheline and
different futexes hash to adjacent buckets.
This patch keeps the current static size of 16 entries for small
systems, or otherwise, 256 * ncpus (or larger as we need to
round the number to a power of 2). Note that this number of CPUs
accounts for all CPUs that can ever be available in the system,
taking into consideration things like hotpluging. While we do
impose extra overhead at bootup by making the hash table larger,
this is a one time thing, and does not shadow the benefits of
this patch.
Furthermore, as suggested by tglx, by cache aligning the hash
buckets we can avoid access across cacheline boundaries and also
avoid massive cache line bouncing if multiple cpus are hammering
away at different hash buckets which happen to reside in the
same cache line.
Also, similar to other core kernel components (pid, dcache,
tcp), by using alloc_large_system_hash() we benefit from its
NUMA awareness and thus the table is distributed among the nodes
instead of in a single one.
For a custom microbenchmark that pounds on the uaddr hashing --
making the wait path fail at futex_wait_setup() returning
-EWOULDBLOCK for large amounts of futexes, we can see the
following benefits on a 80-core, 8-socket 1Tb server:
+---------+--------------------+------------------------+-----------------------+-------------------------------+
| threads | baseline (ops/sec) | aligned-only (ops/sec) | large table (ops/sec) | large table+aligned (ops/sec) |
+---------+--------------------+------------------------+-----------------------+-------------------------------+
| 512 | 32426 | 50531 (+55.8%) | 255274 (+687.2%) | 292553 (+802.2%) |
| 256 | 65360 | 99588 (+52.3%) | 443563 (+578.6%) | 508088 (+677.3%) |
| 128 | 125635 | 200075 (+59.2%) | 742613 (+491.1%) | 835452 (+564.9%) |
| 80 | 193559 | 323425 (+67.1%) | 1028147 (+431.1%) | 1130304 (+483.9%) |
| 64 | 247667 | 443740 (+79.1%) | 997300 (+302.6%) | 1145494 (+362.5%) |
| 32 | 628412 | 721401 (+14.7%) | 965996 (+53.7%) | 1122115 (+78.5%) |
+---------+--------------------+------------------------+-----------------------+-------------------------------+
Reviewed-by: Darren Hart <dvhart@linux.intel.com>
Reviewed-by: Peter Zijlstra <peterz@infradead.org>
Reviewed-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Reviewed-by: Waiman Long <Waiman.Long@hp.com>
Reviewed-and-tested-by: Jason Low <jason.low2@hp.com>
Reviewed-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Davidlohr Bueso <davidlohr@hp.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Jeff Mahoney <jeffm@suse.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Scott Norton <scott.norton@hp.com>
Cc: Tom Vaden <tom.vaden@hp.com>
Cc: Aswin Chandramouleeswaran <aswin@hp.com>
Link: http://lkml.kernel.org/r/1389569486-25487-3-git-send-email-davidlohr@hp.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
This patch fixes a problem with the initialization of the
struct perf_event active_entry field. It is defined inside
an anonymous union and was initialized in perf_event_alloc()
using INIT_LIST_HEAD(). However at that time, we do not know
whether the event is going to use active_entry or hlist_entry (SW).
Or at last, we don't want to make that determination there.
The problem is that hlist and list_head are not initialized
the same way. One is okay with NULL (from kzmalloc), the other
needs to pointers to point to self.
This patch resolves this problem by dropping the union.
This will avoid problems later on, if someone starts using
active_entry or hlist_entry without verifying that they
actually overlap. This also solves the initialization
problem.
Signed-off-by: Stephane Eranian <eranian@google.com>
Cc: ak@linux.intel.com
Cc: acme@redhat.com
Cc: jolsa@redhat.com
Cc: zheng.z.yan@intel.com
Cc: bp@alien8.de
Cc: vincent.weaver@maine.edu
Cc: maria.n.dimakopoulou@gmail.com
Signed-off-by: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1389176153-3128-2-git-send-email-eranian@google.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Thomas Hellstrom bisected a regression where erratic 3D performance is
experienced on virtual machines as measured by glxgears. It identified
commit 58d081b5 ("sched/numa: Avoid overloading CPUs on a preferred NUMA
node") as the problem which had modified the behaviour of effective_load.
Effective load calculates the difference to the system-wide load if a
scheduling entity was moved to another CPU. The task group is not heavier
as a result of the move but overall system load can increase/decrease as a
result of the change. Commit 58d081b5 ("sched/numa: Avoid overloading CPUs
on a preferred NUMA node") changed effective_load to make it suitable for
calculating if a particular NUMA node was compute overloaded. To reduce
the cost of the function, it assumed that a current sched entity weight
of 0 was uninteresting but that is not the case.
wake_affine() uses a weight of 0 for sync wakeups on the grounds that it
is assuming the waking task will sleep and not contribute to load in the
near future. In this case, we still want to calculate the effective load
of the sched entity hierarchy. As effective_load is no longer used by
task_numa_compare since commit fb13c7ee (sched/numa: Use a system-wide
search to find swap/migration candidates), this patch simply restores the
historical behaviour.
Reported-and-tested-by: Thomas Hellstrom <thellstrom@vmware.com>
Signed-off-by: Rik van Riel <riel@redhat.com>
[ Wrote changelog]
Signed-off-by: Mel Gorman <mgorman@suse.de>
Signed-off-by: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20140106113912.GC6178@suse.de
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Pull ACPI and power management fixes and new device IDs from Rafael Wysocki:
- Fix for a cpufreq regression causing stale sysfs files to be left
behind during system resume if cpufreq_add_dev() fails for one or
more CPUs from Viresh Kumar.
- Fix for a bug in cpufreq causing CONFIG_CPU_FREQ_DEFAULT_* to be
ignored when the intel_pstate driver is used from Jason Baron.
- System suspend fix for a memory leak in pm_vt_switch_unregister()
that forgot to release objects after removing them from
pm_vt_switch_list. From Masami Ichikawa.
- Intel Valley View device ID and energy unit encoding update for the
(recently added) Intel RAPL (Running Average Power Limit) driver from
Jacob Pan.
- Intel Bay Trail SoC GPIO and ACPI device IDs for the Low Power
Subsystem (LPSS) ACPI driver from Paul Drews.
* tag 'pm+acpi-3.13-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm:
powercap / RAPL: add support for ValleyView Soc
PM / sleep: Fix memory leak in pm_vt_switch_unregister().
cpufreq: Use CONFIG_CPU_FREQ_DEFAULT_* to set initial policy for setpolicy drivers
cpufreq: remove sysfs files for CPUs which failed to come back after resume
ACPI: Add BayTrail SoC GPIO and LPSS ACPI IDs
* pm-cpufreq:
cpufreq: Use CONFIG_CPU_FREQ_DEFAULT_* to set initial policy for setpolicy drivers
cpufreq: remove sysfs files for CPUs which failed to come back after resume
* pm-sleep:
PM / sleep: Fix memory leak in pm_vt_switch_unregister().
Pull cgroup fixes from Tejun Heo:
"Two fixes. One fixes a bug in the error path of cgroup_create(). The
other changes cgrp->id lifetime rule so that the id doesn't get
recycled before all controller states are destroyed. This premature
id recycling made memcg malfunction"
* 'for-3.13-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/tj/cgroup:
cgroup: don't recycle cgroup id until all csses' have been destroyed
cgroup: fix cgroup_create() error handling path
Pull libata fixes from Tejun Heo:
"There's one interseting commit - "libata, freezer: avoid block device
removal while system is frozen". It's an ugly hack working around a
deadlock condition between driver core resume and block layer device
removal paths through freezer which was made more reproducible by
writeback being converted to workqueue some releases ago. The bug has
nothing to do with libata but it's just an workaround which is easy to
backport. After discussion, Rafael and I seem to agree that we don't
really need kernel freezables - both kthread and workqueue. There are
few specific workqueues which constitute PM operations and require
freezing, which will be converted to use workqueue_set_max_active()
instead. All other kernel freezer uses are planned to be removed,
followed by the removal of kthread and workqueue freezer support,
hopefully.
Others are device-specific fixes. The most notable is the addition of
NO_NCQ_TRIM which is used to disable queued TRIM commands to Micro
M500 SSDs which otherwise suffers data corruption"
* 'for-3.13-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/tj/libata:
libata, freezer: avoid block device removal while system is frozen
libata: implement ATA_HORKAGE_NO_NCQ_TRIM and apply it to Micro M500 SSDs
libata: disable a disk via libata.force params
ahci: bail out on ICH6 before using AHCI BAR
ahci: imx: Explicitly clear IMX6Q_GPR13_SATA_MPLL_CLK_EN
libata: add ATA_HORKAGE_BROKEN_FPDMA_AA quirk for Seagate Momentus SpinPoint M8
Prior to 92bb1fcf57 (Only
do nanosecond rounding on GENERIC_TIME_VSYSCALL_OLD
systems), the comment here was accuate, but now we can
mostly avoid the extra rounding which causes the unlikey
to be actually likely here.
So remove the out of date comment.
Signed-off-by: John Stultz <john.stultz@linaro.org>
Since 48cdc135d4 (Implement a shadow timekeeper), we have to
call timekeeping_update() after any adjustment to the timekeeping
structure in order to make sure that any adjustments to the structure
persist.
In the timekeeping suspend path, we udpate the timekeeper
structure, so we should be sure to update the shadow-timekeeper
before releasing the timekeeping locks. Currently this isn't done.
In most cases, the next time related code to run would be
timekeeping_resume, which does update the shadow-timekeeper, but
in an abundence of caution, this patch adds the call to
timekeeping_update() in the suspend path.
Cc: Sasha Levin <sasha.levin@oracle.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Prarit Bhargava <prarit@redhat.com>
Cc: Richard Cochran <richardcochran@gmail.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: stable <stable@vger.kernel.org> #3.10+
Signed-off-by: John Stultz <john.stultz@linaro.org>
Since the xtime lock was split into the timekeeping lock and
the jiffies lock, we no longer need to call update_wall_time()
while holding the jiffies lock.
Thus, this patch splits update_wall_time() out from do_timer().
This allows us to get away from calling clock_was_set_delayed()
in update_wall_time() and instead use the standard clock_was_set()
call that previously would deadlock, as it causes the jiffies lock
to be acquired.
Cc: Sasha Levin <sasha.levin@oracle.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Prarit Bhargava <prarit@redhat.com>
Cc: Richard Cochran <richardcochran@gmail.com>
Cc: Ingo Molnar <mingo@kernel.org>
Signed-off-by: John Stultz <john.stultz@linaro.org>
As part of normal operaions, the hrtimer subsystem frequently calls
into the timekeeping code, creating a locking order of
hrtimer locks -> timekeeping locks
clock_was_set_delayed() was suppoed to allow us to avoid deadlocks
between the timekeeping the hrtimer subsystem, so that we could
notify the hrtimer subsytem the time had changed while holding
the timekeeping locks. This was done by scheduling delayed work
that would run later once we were out of the timekeeing code.
But unfortunately the lock chains are complex enoguh that in
scheduling delayed work, we end up eventually trying to grab
an hrtimer lock.
Sasha Levin noticed this in testing when the new seqlock lockdep
enablement triggered the following (somewhat abrieviated) message:
[ 251.100221] ======================================================
[ 251.100221] [ INFO: possible circular locking dependency detected ]
[ 251.100221] 3.13.0-rc2-next-20131206-sasha-00005-g8be2375-dirty #4053 Not tainted
[ 251.101967] -------------------------------------------------------
[ 251.101967] kworker/10:1/4506 is trying to acquire lock:
[ 251.101967] (timekeeper_seq){----..}, at: [<ffffffff81160e96>] retrigger_next_event+0x56/0x70
[ 251.101967]
[ 251.101967] but task is already holding lock:
[ 251.101967] (hrtimer_bases.lock#11){-.-...}, at: [<ffffffff81160e7c>] retrigger_next_event+0x3c/0x70
[ 251.101967]
[ 251.101967] which lock already depends on the new lock.
[ 251.101967]
[ 251.101967]
[ 251.101967] the existing dependency chain (in reverse order) is:
[ 251.101967]
-> #5 (hrtimer_bases.lock#11){-.-...}:
[snipped]
-> #4 (&rt_b->rt_runtime_lock){-.-...}:
[snipped]
-> #3 (&rq->lock){-.-.-.}:
[snipped]
-> #2 (&p->pi_lock){-.-.-.}:
[snipped]
-> #1 (&(&pool->lock)->rlock){-.-...}:
[ 251.101967] [<ffffffff81194803>] validate_chain+0x6c3/0x7b0
[ 251.101967] [<ffffffff81194d9d>] __lock_acquire+0x4ad/0x580
[ 251.101967] [<ffffffff81194ff2>] lock_acquire+0x182/0x1d0
[ 251.101967] [<ffffffff84398500>] _raw_spin_lock+0x40/0x80
[ 251.101967] [<ffffffff81153e69>] __queue_work+0x1a9/0x3f0
[ 251.101967] [<ffffffff81154168>] queue_work_on+0x98/0x120
[ 251.101967] [<ffffffff81161351>] clock_was_set_delayed+0x21/0x30
[ 251.101967] [<ffffffff811c4bd1>] do_adjtimex+0x111/0x160
[ 251.101967] [<ffffffff811e2711>] compat_sys_adjtimex+0x41/0x70
[ 251.101967] [<ffffffff843a4b49>] ia32_sysret+0x0/0x5
[ 251.101967]
-> #0 (timekeeper_seq){----..}:
[snipped]
[ 251.101967] other info that might help us debug this:
[ 251.101967]
[ 251.101967] Chain exists of:
timekeeper_seq --> &rt_b->rt_runtime_lock --> hrtimer_bases.lock#11
[ 251.101967] Possible unsafe locking scenario:
[ 251.101967]
[ 251.101967] CPU0 CPU1
[ 251.101967] ---- ----
[ 251.101967] lock(hrtimer_bases.lock#11);
[ 251.101967] lock(&rt_b->rt_runtime_lock);
[ 251.101967] lock(hrtimer_bases.lock#11);
[ 251.101967] lock(timekeeper_seq);
[ 251.101967]
[ 251.101967] *** DEADLOCK ***
[ 251.101967]
[ 251.101967] 3 locks held by kworker/10:1/4506:
[ 251.101967] #0: (events){.+.+.+}, at: [<ffffffff81154960>] process_one_work+0x200/0x530
[ 251.101967] #1: (hrtimer_work){+.+...}, at: [<ffffffff81154960>] process_one_work+0x200/0x530
[ 251.101967] #2: (hrtimer_bases.lock#11){-.-...}, at: [<ffffffff81160e7c>] retrigger_next_event+0x3c/0x70
[ 251.101967]
[ 251.101967] stack backtrace:
[ 251.101967] CPU: 10 PID: 4506 Comm: kworker/10:1 Not tainted 3.13.0-rc2-next-20131206-sasha-00005-g8be2375-dirty #4053
[ 251.101967] Workqueue: events clock_was_set_work
So the best solution is to avoid calling clock_was_set_delayed() while
holding the timekeeping lock, and instead using a flag variable to
decide if we should call clock_was_set() once we've released the locks.
This works for the case here, where the do_adjtimex() was the deadlock
trigger point. Unfortuantely, in update_wall_time() we still hold
the jiffies lock, which would deadlock with the ipi triggered by
clock_was_set(), preventing us from calling it even after we drop the
timekeeping lock. So instead call clock_was_set_delayed() at that point.
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Prarit Bhargava <prarit@redhat.com>
Cc: Richard Cochran <richardcochran@gmail.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Sasha Levin <sasha.levin@oracle.com>
Cc: stable <stable@vger.kernel.org> #3.10+
Reported-by: Sasha Levin <sasha.levin@oracle.com>
Tested-by: Sasha Levin <sasha.levin@oracle.com>
Signed-off-by: John Stultz <john.stultz@linaro.org>
In 780427f0e1 (Indicate that clock was set in the pvclock
gtod notifier), logic was added to pass a CLOCK_WAS_SET
notification to the pvclock notifier chain.
While that patch added a action flag returned from
accumulate_nsecs_to_secs(), it only uses the returned value
in one location, and not in the logarithmic accumulation.
This means if a leap second triggered during the logarithmic
accumulation (which is most likely where it would happen),
the notification that the clock was set would not make it to
the pv notifiers.
This patch extends the logarithmic_accumulation pass down
that action flag so proper notification will occur.
This patch also changes the varialbe action -> clock_set
per Ingo's suggestion.
Cc: Sasha Levin <sasha.levin@oracle.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: David Vrabel <david.vrabel@citrix.com>
Cc: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Cc: Prarit Bhargava <prarit@redhat.com>
Cc: Richard Cochran <richardcochran@gmail.com>
Cc: <xen-devel@lists.xen.org>
Cc: stable <stable@vger.kernel.org> #3.11+
Signed-off-by: John Stultz <john.stultz@linaro.org>
Since 48cdc135d4 (Implement a shadow timekeeper), we have to
call timekeeping_update() after any adjustment to the timekeeping
structure in order to make sure that any adjustments to the structure
persist.
Unfortunately, the updates to the tai offset via adjtimex do not
trigger this update, causing adjustments to the tai offset to be
made and then over-written by the previous value at the next
update_wall_time() call.
This patch resovles the issue by calling timekeeping_update()
right after setting the tai offset.
Cc: Sasha Levin <sasha.levin@oracle.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Prarit Bhargava <prarit@redhat.com>
Cc: Richard Cochran <richardcochran@gmail.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: stable <stable@vger.kernel.org> #3.10+
Signed-off-by: John Stultz <john.stultz@linaro.org>
In struct page we have enough space to fit long-size page->ptl there,
but we use dynamically-allocated page->ptl if size(spinlock_t) is larger
than sizeof(int).
It hurts 64-bit architectures with CONFIG_GENERIC_LOCKBREAK, where
sizeof(spinlock_t) == 8, but it easily fits into struct page.
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Acked-by: Hugh Dickins <hughd@google.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Davidlohr Bueso
Thomas Gleixner
Jason Low
Ingo Molnar
Yann Droneaud
Stephane Eranian
John Stultz
Rik van Riel
Linus Torvalds
Rafael J. Wysocki
Yijing Wang
Masami Ichikawa
Kirill A. Shutemov