TCMalloc: reduce waiting time in SpinLockDelay on Linux

TCMalloc: reduce waiting time in SpinLockDelay on Linux

Because of an optimization in Unlock(), we can't expect explicit
wake-ups and shouldn't wait for too long in SpinLockDelay.

BUG=168139
NOTRY=true


Committed: https://src.chromium.org/viewvc/chrome?view=rev&revision=175543

[davidjames, 2013-01-04 23:42:00]

lgtm

[commit-bot: I haz the power, 2013-01-07 23:46:54]

No LGTM from a valid reviewer yet. Only full committers are accepted.
Even if an LGTM may have been provided, it was from a non-committer or
a lowly provisional committer, _not_ a full super star committer.
See http://www.chromium.org/getting-involved/become-a-committer
Note that this has nothing to do with OWNERS files.

[Markus (顧孟勤), 2013-01-08 00:38:30]

lgtm

"go/spinlock_latency" has a discussion of (some of) the issues involved.

I do agree that making the time out shorter is likely always better for Chrome,
so I am in favor of this CL.

It is my understanding that the large value for the timeout was chosen, because
it avoids costly operations when dealing with thousands of threads fighting over
a common resource. This is unlikely to be a common problem for Chrome, and we
care more about deterministic upper bounds on latency.

In fact, I am not convinced that tweaking the time out value is sufficient. I'd
much rather see a CL that makes the lock more deterministic or maybe even fair.

I wouldn't be opposed to a CL that uses an atomic operating when releasing the
lock.

And neither would I be opposed to a CL that replaces the spinlock with a
(queued?) mutex.

It should be kept in mind that this is generic code in tcmalloc that was not
tuned especially for Chrome. So, some of the trade-offs are bound to be
suboptimal.

Those changes are more invasive though, and probably warrant extensive
performance testing. So, as an immediate measure, we should reduce the timeout
and then investigate what other changes are appropriate.

[commit-bot: I haz the power, 2013-01-08 00:41:54]

CQ is trying da patch. Follow status at
https://chromium-status.appspot.com/cq/jln@chromium.org/11784009/1

[commit-bot: I haz the power, 2013-01-08 17:26:25]

CQ is trying da patch. Follow status at
https://chromium-status.appspot.com/cq/jln@chromium.org/11784009/1

[commit-bot: I haz the power, 2013-01-08 17:27:20]

Change committed as 175543