Introduce API to temporarily interrupt long running JavaScript code.

Introduce API to temporarily interrupt long running JavaScript code.

It is different from termination API as interrupted JavaScript will continue to execute normally when registered InterruptCallback returns.

  /**
   * Request V8 to interrupt long running JavaScript code and invoke
   * the given |callback| passing the given |data| to it. After |callback|
   * returns control will be returned to the JavaScript code.
   * At any given moment V8 can remember only a single callback for the very
   * last interrupt request.
   * Can be called from another thread without acquiring a |Locker|.
   * Registered |callback| must not reenter interrupted Isolate.
   */
  void RequestInterrupt(InterruptCallback callback, void* data);

  /**
   * Clear interrupt request created by |RequestInterrupt|.
   * Can be called from another thread without acquiring a |Locker|.
   */
  void ClearInterrupt();

Fix Hydrogen SCE pass to avoid eliminating stack guards too aggressively. Only normal JavaScript functions are guaranteed to have stack guard in the prologue. If function is a builtin or has a custom call IC it will lack one.

BUG=
R=danno@chromium.org, dcarney@chromium.org

Committed: https://code.google.com/p/v8/source/detail?r=18363

[Vyacheslav Egorov (Google), 2013-12-04 21:30:24]

Hi!

Dan, please take a look at the API part of the change.

Danno, could you please review or find a reviewer for a Hydrogen part of this
change?

Alternative to fixing SCE would be to insert stack guards into each CallIC stub
and each builtin which I think is not robust and will break when someone decides
to add a new stubs/builtins and forgets to add stack guard (as stubs usually
have none). 

Measurements indicate that SCE fix is performance neutral (which is rather to be
expected).

[dcarney, 2013-12-05 08:06:27]

lgtm

https://codereview.chromium.org/102063004/diff/1/test/cctest/test-api.cc
File test/cctest/test-api.cc (right):

https://codereview.chromium.org/102063004/diff/1/test/cctest/test-api.cc#newc...
test/cctest/test-api.cc:21057: #endif  // V8_OS_POSIX
you've got this inside the V8_OS_POSIX ifdef, which I presume is wrong?  That
#endif is already in the wrong spot itself...

[danno, 2013-12-05 10:14:41]

lgtm from me too

[Vyacheslav Egorov (Chromium), 2013-12-05 12:08:20]

Thanks for review. 

Landing.

https://codereview.chromium.org/102063004/diff/1/test/cctest/test-api.cc
File test/cctest/test-api.cc (right):

https://codereview.chromium.org/102063004/diff/1/test/cctest/test-api.cc#newc...
test/cctest/test-api.cc:21057: #endif  // V8_OS_POSIX
On 2013/12/05 08:06:28, dcarney wrote:
> you've got this inside the V8_OS_POSIX ifdef, which I presume is wrong?  That
> #endif is already in the wrong spot itself...

Thanks! Copy&paste mistake. Fixed.

[Vyacheslav Egorov (Chromium), 2013-12-05 12:45:22]

The tree is throttled right now. I will just wait until branch point has past
and tree is open again.

[dcarney, 2013-12-18 09:49:58]

On 2013/12/05 12:45:22, Vyacheslav Egorov wrote:
> The tree is throttled right now. I will just wait until branch point has past
> and tree is open again.

tree is open again (off an on throttled, though)
talk to to hpayer about when to land your changes - they are low risk changes,
so no real reason not to land them immediately

[Vyacheslav Egorov (Chromium), 2013-12-18 18:38:47]

Committed patchset #1 manually as r18363 (presubmit successful).