[PATCH -tip v2 0/5] rwsem: Fine tuning

From: Davidlohr Bueso
Date: Fri Jan 30 2015 - 04:17:16 EST


Hello,

First two patches are self descriptive obvious add-ons.

The rest are performance enhancements for write-mostly workloads.
While this is not our priority (use mutexes instead!!), there are
cases where heavy use of writers can severely hurt rwsem performance.
For instance, we got reports[1] of writer only issues when converting
the i_mmap_rwsem from mutex, on a workload that pathologically pounds
on this lock for vma operations:

- 81.20% execl [kernel.kallsyms] [k] osq_lock
- 100.00% mutex_optimistic_spin
__mutex_lock_slowpath
- mutex_lock
+ 47.71% unlink_file_vma
+ 34.91% vma_adjust
+ 17.38% vma_link

This is enough to make small differences painfully evident. These changes
(particularly patch 5/5) recover most (~75%) of the performance regression.
Patches 3- 5 deal with optimistic spinning fine tuning. Details are in the
individual patches. Passes multiple x86-64 tests.

Thanks!

Changes since v1 (https://lkml.org/lkml/2015/1/26/34):

- Redid patch 2.

- Improved "Check for active lock before bailing on spinning" patch
to consider checking if the lock has a new owner.

- Dropped patch 5, will send a new patch later disabling preemption in the
slowpath to get rid of the barriers (assuming testing goes well) without
missing wakeups. It's not very related to the patches in this series anyway.

Davidlohr Bueso (5):
locking/rwsem: Use task->state helpers
locking/rwsem: Document barrier need when waking tasks
locking/rwsem: Set lock ownership ASAP
locking/rwsem: Avoid deceiving lock spinners
locking/rwsem: Check for active lock before bailing on spinning

kernel/locking/mutex.c | 2 +-
kernel/locking/rwsem-spinlock.c | 9 +++++-
kernel/locking/rwsem-xadd.c | 66 ++++++++++++++++++++++++++++-------------
kernel/locking/rwsem.c | 22 +-------------
kernel/locking/rwsem.h | 20 +++++++++++++
5 files changed, 76 insertions(+), 43 deletions(-)
create mode 100644 kernel/locking/rwsem.h

--
2.1.4

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/