Use timerfd to handler timeouts on Linux.

Use timerfd to handle timeouts on Linux.

BUG=
R=sgjesse@google.com

Committed: https://code.google.com/p/dart/source/detail?r=30884

[Anders Johnsen, 2013-12-04 20:27:57]

I think we should land this and see how it behaves, before we port it to other
platforms.

This lowers the number of gettimeofday calls made by the event-handler. In a
simple HTTP hello world example, it's reduced by half.

[Søren Gjesse, 2013-12-05 07:39:34]

LGTM - This looks much cleaner

https://codereview.chromium.org/104803005/diff/10001/runtime/bin/eventhandler...
File runtime/bin/eventhandler_linux.cc (right):

https://codereview.chromium.org/104803005/diff/10001/runtime/bin/eventhandler...
runtime/bin/eventhandler_linux.cc:208: if (msg[i].id == kTimerId) {
With this approach do we even need to send this through the pipe?

https://codereview.chromium.org/104803005/diff/10001/runtime/bin/eventhandler...
runtime/bin/eventhandler_linux.cc:217: timerfd_settime(timer_fd_,
TFD_TIMER_ABSTIME, &it, NULL);
If the abs time set here has just passed will the timer still fire?

https://codereview.chromium.org/104803005/diff/10001/runtime/bin/eventhandler...
runtime/bin/eventhandler_linux.cc:367: VOID_TEMP_FAILURE_RETRY(read(timer_fd_,
&val, 8));
Should we have a constant for '8' here just to indicate what this is? Maybe
there is one in the .h file.

https://codereview.chromium.org/104803005/diff/10001/runtime/bin/eventhandler...
runtime/bin/eventhandler_linux.cc:403: -1));
Is there a constant for infinite?

[Anders Johnsen, 2013-12-05 09:18:27]

https://codereview.chromium.org/104803005/diff/10001/runtime/bin/eventhandler...
File runtime/bin/eventhandler_linux.cc (right):

https://codereview.chromium.org/104803005/diff/10001/runtime/bin/eventhandler...
runtime/bin/eventhandler_linux.cc:208: if (msg[i].id == kTimerId) {
On 2013/12/05 07:39:34, Søren Gjesse wrote:
> With this approach do we even need to send this through the pipe?

The alternative is to have a FD per isolate, but then we need a isolate-shutdown
hook. Let's keep it as this for now, but it may be worth looking into the
alternative in the future.

https://codereview.chromium.org/104803005/diff/10001/runtime/bin/eventhandler...
runtime/bin/eventhandler_linux.cc:217: timerfd_settime(timer_fd_,
TFD_TIMER_ABSTIME, &it, NULL);
On 2013/12/05 07:39:34, Søren Gjesse wrote:
> If the abs time set here has just passed will the timer still fire?

Yes, from spec:

  "If the specified absolute time has
   already passed, then the timer expires immediately, and the 
   overrun count (see timer_getoverrun(2)) will be set correctly"

https://codereview.chromium.org/104803005/diff/10001/runtime/bin/eventhandler...
runtime/bin/eventhandler_linux.cc:367: VOID_TEMP_FAILURE_RETRY(read(timer_fd_,
&val, 8));
On 2013/12/05 07:39:34, Søren Gjesse wrote:
> Should we have a constant for '8' here just to indicate what this is? Maybe
> there is one in the .h file.

I'll use sizeof val.

https://codereview.chromium.org/104803005/diff/10001/runtime/bin/eventhandler...
runtime/bin/eventhandler_linux.cc:403: -1));
On 2013/12/05 07:39:34, Søren Gjesse wrote:
> Is there a constant for infinite?

-1 is special-case for epoll_wait, to wait forever. I think we should keep it
like this, as it's strictly related to the epoll call.

[Anders Johnsen, 2013-12-05 09:22:13]

Committed patchset #3 manually as r30884 (presubmit successful).