rwsem-spinlock: Implement writer lock-stealing for better scalability

We (Linux Kernel Performance project) found a regression
introduced by commit:

  5a505085f043 mm/rmap: Convert the struct anon_vma::mutex to an rwsem

which converted all anon_vma::mutex locks rwsem write locks.

The semantics are the same, but the behavioral difference is
quite huge in some cases. After investigating it we found the
root cause: mutexes support lock stealing while rwsems don't.

Here is the link for the detailed regression report:

  https://lkml.org/lkml/2013/1/29/84

Ingo suggested adding write lock stealing to rwsems:

    "I think we should allow lock-steal between rwsem writers - that
     will not hurt fairness as most rwsem fairness concerns relate to
     reader vs. writer fairness"

And here is the rwsem-spinlock version.

With this patch, we got a double performance increase in one
test box with following aim7 workfile:

    FILESIZE: 1M
    POOLSIZE: 10M
    10 fork_test

 /usr/bin/time output w/o patch                       /usr/bin/time_output with patch
 -- Percent of CPU this job got: 369%                 Percent of CPU this job got: 537%
 Voluntary context switches: 640595016                Voluntary context switches: 157915561

We got a 45% increase in CPU usage and saved about 3/4 voluntary context switches.

Reported-by: LKP project <[email protected]>
Suggested-by: Ingo Molnar <[email protected]>
Signed-off-by: Yuanhan Liu <[email protected]>
Cc: Alex Shi <[email protected]>
Cc: David Howells <[email protected]>
Cc: Michel Lespinasse <[email protected]>
Cc: Linus Torvalds <[email protected]>
Cc: Andrew Morton <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Anton Blanchard <[email protected]>
Cc: Arjan van de Ven <[email protected]>
Cc: [email protected]
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Ingo Molnar <[email protected]>

0 files changed, 0 insertions, 0 deletions