[lkml]   [2013]   [Nov]   [25]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
Subject[RFC patch 0/5] futex: Allow lockless empty check of hashbucket plist in futex_wake()
The patch set from Davidlohr [1] tried to attempt the same via an
atomic counter of waiters in a hash bucket. The atomic counter access
provided enough serialization for x86 so that a failure is not
observable in testing, but does not provide any guarantees.

The same can be achieved with a smp_mb() pair including proper
guarantees for all architectures.

The following series provides an incremental approach to this and adds
documentation of the ordering guarantees of futexes.

Note, this is RFC and needs a lot of review, testing and proper
performance numbers for the following scenarios:

1) Test case where a single waiter is about to queue itself, i.e. the
test case Davidlohr used to gather his numbers.

2) Test case where the hash bucket is always not empty. This allows us
to determine the smp_mb() overhead for cases which are not
optimized by the singler waiter per bucket.

These tests need to be done on x86 AND on other architectures where
the smp_mb() might be more expensive than on x86.




 \ /
  Last update: 2013-11-25 22:21    [W:0.100 / U:0.348 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site