linux/kernel/locking
Waiman Long 11752adb68 locking/pvqspinlock: Implement hybrid PV queued/unfair locks
Currently, all the lock waiters entering the slowpath will do one
lock stealing attempt to acquire the lock. That helps performance,
especially in VMs with over-committed vCPUs. However, the current
pvqspinlocks still don't perform as good as unfair locks in many cases.
On the other hands, unfair locks do have the problem of lock starvation
that pvqspinlocks don't have.

This patch combines the best attributes of an unfair lock and a
pvqspinlock into a hybrid lock with 2 modes - queued mode & unfair
mode. A lock waiter goes into the unfair mode when there are waiters
in the wait queue but the pending bit isn't set. Otherwise, it will
go into the queued mode waiting in the queue for its turn.

On a 2-socket 36-core E5-2699 v3 system (HT off), a kernel build
(make -j<n>) was done in a VM with unpinned vCPUs 3 times with the
best time selected and <n> is the number of vCPUs available. The build
times of the original pvqspinlock, hybrid pvqspinlock and unfair lock
with various number of vCPUs are as follows:

  vCPUs    pvqlock     hybrid pvqlock    unfair lock
  -----    -------     --------------    -----------
    30      342.1s         329.1s          329.1s
    36      314.1s         305.3s          307.3s
    45      345.0s         302.1s          306.6s
    54      365.4s         308.6s          307.8s
    72      358.9s         293.6s          303.9s
   108      343.0s         285.9s          304.2s

The hybrid pvqspinlock performs better or comparable to the unfair
lock.

By turning on QUEUED_LOCK_STAT, the table below showed the number
of lock acquisitions in unfair mode and queue mode after a kernel
build with various number of vCPUs.

  vCPUs    queued mode  unfair mode
  -----    -----------  -----------
    30      9,130,518      294,954
    36     10,856,614      386,809
    45      8,467,264   11,475,373
    54      6,409,987   19,670,855
    72      4,782,063   25,712,180

It can be seen that as the VM became more and more over-committed,
the ratio of locks acquired in unfair mode increases. This is all
done automatically to get the best overall performance as possible.

Using a kernel locking microbenchmark with number of locking
threads equals to the number of vCPUs available on the same machine,
the minimum, average and maximum (min/avg/max) numbers of locking
operations done per thread in a 5-second testing interval are shown
below:

  vCPUs         hybrid pvqlock             unfair lock
  -----         --------------             -----------
    36     822,135/881,063/950,363    75,570/313,496/  690,465
    54     542,435/581,664/625,937    35,460/204,280/  457,172
    72     397,500/428,177/499,299    17,933/150,679/  708,001
   108     257,898/288,150/340,871     3,085/181,176/1,257,109

It can be seen that the hybrid pvqspinlocks are more fair and
performant than the unfair locks in this test.

The table below shows the kernel build times on a smaller 2-socket
16-core 32-thread E5-2620 v4 system.

  vCPUs    pvqlock     hybrid pvqlock    unfair lock
  -----    -------     --------------    -----------
    16      436.8s         433.4s          435.6s
    36      366.2s         364.8s          364.5s
    48      423.6s         376.3s          370.2s
    64      433.1s         376.6s          376.8s

Again, the performance of the hybrid pvqspinlock was comparable to
that of the unfair lock.

Signed-off-by: Waiman Long <longman@redhat.com>
Reviewed-by: Juergen Gross <jgross@suse.com>
Reviewed-by: Eduardo Valentin <eduval@amazon.com>
Acked-by: Peter Zijlstra <peterz@infradead.org>
Cc: Boris Ostrovsky <boris.ostrovsky@oracle.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Radim Krčmář <rkrcmar@redhat.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/1510089486-3466-1-git-send-email-longman@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
2017-11-08 10:10:04 +01:00
..
lockdep_internals.h License cleanup: add SPDX GPL-2.0 license identifier to files with no license 2017-11-02 11:10:55 +01:00
lockdep_proc.c License cleanup: add SPDX GPL-2.0 license identifier to files with no license 2017-11-02 11:10:55 +01:00
lockdep_states.h locking/lockdep: Rework FS_RECLAIM annotation 2017-08-10 12:29:03 +02:00
lockdep.c locking/lockdep: Introduce CONFIG_BOOTPARAM_LOCKDEP_CROSSRELEASE_FULLSTACK=y 2017-10-25 12:19:02 +02:00
locktorture.c sched/headers: Prepare for the removal of <linux/rtmutex.h> from <linux/sched.h> 2017-03-02 08:42:32 +01:00
Makefile License cleanup: add SPDX GPL-2.0 license identifier to files with no license 2017-11-02 11:10:55 +01:00
mcs_spinlock.h License cleanup: add SPDX GPL-2.0 license identifier to files with no license 2017-11-02 11:10:55 +01:00
mutex-debug.c locking/mutex: Rework mutex::owner 2016-10-25 11:31:50 +02:00
mutex-debug.h License cleanup: add SPDX GPL-2.0 license identifier to files with no license 2017-11-02 11:10:55 +01:00
mutex.c mutex, futex: adjust kernel-doc markups to generate ReST 2017-05-16 08:43:25 -03:00
mutex.h License cleanup: add SPDX GPL-2.0 license identifier to files with no license 2017-11-02 11:10:55 +01:00
osq_lock.c License cleanup: add SPDX GPL-2.0 license identifier to files with no license 2017-11-02 11:10:55 +01:00
percpu-rwsem.c locking/percpu-rwsem: Replace waitqueue with rcuwait 2017-01-14 11:14:35 +01:00
qrwlock.c locking/qrwlock: Prevent slowpath writers getting held up by fastpath 2017-10-25 10:57:25 +02:00
qspinlock_paravirt.h locking/pvqspinlock: Implement hybrid PV queued/unfair locks 2017-11-08 10:10:04 +01:00
qspinlock_stat.h sched/headers: Prepare for new header dependencies before moving code to <linux/sched/clock.h> 2017-03-02 08:42:27 +01:00
qspinlock.c locking: Remove spin_unlock_wait() generic definitions 2017-08-17 08:08:58 -07:00
rtmutex_common.h License cleanup: add SPDX GPL-2.0 license identifier to files with no license 2017-11-02 11:10:55 +01:00
rtmutex-debug.c License cleanup: add SPDX GPL-2.0 license identifier to files with no license 2017-11-02 11:10:55 +01:00
rtmutex-debug.h License cleanup: add SPDX GPL-2.0 license identifier to files with no license 2017-11-02 11:10:55 +01:00
rtmutex.c locking/rtmutex: replace top-waiter and pi_waiters leftmost caching 2017-09-08 18:26:49 -07:00
rtmutex.h License cleanup: add SPDX GPL-2.0 license identifier to files with no license 2017-11-02 11:10:55 +01:00
rwsem-spinlock.c License cleanup: add SPDX GPL-2.0 license identifier to files with no license 2017-11-02 11:10:55 +01:00
rwsem-xadd.c License cleanup: add SPDX GPL-2.0 license identifier to files with no license 2017-11-02 11:10:55 +01:00
rwsem.c Merge branch 'linus' into locking/core, to resolve conflicts 2017-11-07 10:32:44 +01:00
rwsem.h License cleanup: add SPDX GPL-2.0 license identifier to files with no license 2017-11-02 11:10:55 +01:00
semaphore.c sched/headers: Prepare for new header dependencies before moving code to <linux/sched/debug.h> 2017-03-02 08:42:34 +01:00
spinlock_debug.c locking/spinlock/debug: Remove spinlock lockup detection code 2017-02-10 09:09:49 +01:00
spinlock.c locking/rwlocks: Fix comments 2017-11-07 12:22:21 +01:00
test-ww_mutex.c mm: treewide: remove GFP_TEMPORARY allocation flag 2017-09-13 18:53:16 -07:00