[Oilpan] Set stack limit using estimated sizes (Re-land)

Set the stack limit size based on estimated sizes.

To check if the stack memory has enough room for recursive
calls, we tried to allocate a large object and assumed the size.
This CL estimate the real thread stack limit using APIs.
On some platforms where such APIs do not work correctly,
we set the limit using the old algorithm.

---
This CL is re-land of https://codereview.chromium.org/910663002/
with fixing a test.
We needed a longer call chain of trace() to expect
to use an alternative marking method, because we got wider
stack memory area to go eagerly.

Committed: https://src.chromium.org/viewvc/blink?view=rev&revision=190388

[peria, 2015-02-17 04:26:59]

peria@chromium.org changed reviewers:
+ kouhei@chromium.org, oilpan-reviews@chromium.org

[peria, 2015-02-17 04:27:00]

PTL

[kouhei (in TOK), 2015-02-17 04:28:37]

One quick question: have you checked the test's execution time?

[peria, 2015-02-17 04:38:46]

On 2015/02/17 04:28:37, kouhei wrote:
> One quick question: have you checked the test's execution time?

I didn't see the real number, but I felt that HeapTest finished in 500ms or so
on my workstation Z620.

[kouhei (in TOK), 2015-02-17 06:00:05]

On 2015/02/17 04:38:46, peria wrote:
> On 2015/02/17 04:28:37, kouhei wrote:
> > One quick question: have you checked the test's execution time?
> 
> I didn't see the real number, but I felt that HeapTest finished in 500ms or so
> on my workstation Z620.

I feel 500ms is too long.

https://codereview.chromium.org/927213002/diff/1/Source/platform/heap/HeapTes...
File Source/platform/heap/HeapTest.cpp (right):

https://codereview.chromium.org/927213002/diff/1/Source/platform/heap/HeapTes...
Source/platform/heap/HeapTest.cpp:5460: int calls = ++sTraceCalls;
Would you try allocating a large array in stack here, instead of increasing
depth?

[Hitoshi Yoshida, 2015-02-17 10:09:49]

peria@google.com changed reviewers:
+ peria@google.com

[Hitoshi Yoshida, 2015-02-17 10:09:50]

https://codereview.chromium.org/927213002/diff/1/Source/platform/heap/HeapTes...
File Source/platform/heap/HeapTest.cpp (right):

https://codereview.chromium.org/927213002/diff/1/Source/platform/heap/HeapTes...
Source/platform/heap/HeapTest.cpp:5460: int calls = ++sTraceCalls;
On 2015/02/17 06:00:05, kouhei wrote:
> Would you try allocating a large array in stack here, instead of increasing
> depth?

Parameters (the number of loops and the size of the object) need
to be optimized to pass this test, and they may depend on the platform.
Otherwise, this approach causes a crush or fails the test.


Current settings (chaining 2*10^6 DeepEagerly objects) takes 87ms 
in HeapTest.TraceDeepEagerly, while few other tests take longer.

[kouhei (in TOK), 2015-02-17 13:11:44]

On 2015/02/17 10:09:50, Hitoshi Yoshida wrote:
>
https://codereview.chromium.org/927213002/diff/1/Source/platform/heap/HeapTes...
> File Source/platform/heap/HeapTest.cpp (right):
> 
>
https://codereview.chromium.org/927213002/diff/1/Source/platform/heap/HeapTes...
> Source/platform/heap/HeapTest.cpp:5460: int calls = ++sTraceCalls;
> On 2015/02/17 06:00:05, kouhei wrote:
> > Would you try allocating a large array in stack here, instead of increasing
> > depth?
> 
> Parameters (the number of loops and the size of the object) need
> to be optimized to pass this test, and they may depend on the platform.
> Otherwise, this approach causes a crush or fails the test.
> 
> 
> Current settings (chaining 2*10^6 DeepEagerly objects) takes 87ms 
> in HeapTest.TraceDeepEagerly, while few other tests take longer.

lgtm given that this test isn't the worst test.

[haraken, 2015-02-17 14:48:48]

haraken@chromium.org changed reviewers:
+ haraken@chromium.org

[haraken, 2015-02-17 14:48:49]

LGTM

https://codereview.chromium.org/927213002/diff/20001/Source/platform/heap/Thr...
File Source/platform/heap/ThreadState.cpp (right):

https://codereview.chromium.org/927213002/diff/20001/Source/platform/heap/Thr...
Source/platform/heap/ThreadState.cpp:258: if (underestimatedStackSize > 0)

underestimatedStackSize >= sizeof(void*)

[peria, 2015-02-18 02:45:29]

New patchsets have been uploaded after l-g-t-m from
kouhei@chromium.org,haraken@chromium.org

[peria, 2015-02-18 02:45:38]

The CQ bit was checked by peria@chromium.org

[peria, 2015-02-18 02:45:59]

https://codereview.chromium.org/927213002/diff/20001/Source/platform/heap/Thr...
File Source/platform/heap/ThreadState.cpp (right):

https://codereview.chromium.org/927213002/diff/20001/Source/platform/heap/Thr...
Source/platform/heap/ThreadState.cpp:258: if (underestimatedStackSize > 0)
On 2015/02/17 14:48:49, haraken wrote:
> 
> underestimatedStackSize >= sizeof(void*)

Done.

[commit-bot: I haz the power, 2015-02-18 02:46:56]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/927213002/40001

[commit-bot: I haz the power, 2015-02-18 04:41:52]

Committed patchset #3 (id:40001) as
https://src.chromium.org/viewvc/blink?view=rev&revision=190388

[yhirano, 2015-02-18 08:06:38]

A revert of this CL (patchset #3 id:40001) has been created in
https://codereview.chromium.org/934193002/ by yhirano@chromium.org.

The reason for reverting is: Broke a heap test on Android Nexus4

https://build.chromium.org/p/chromium.webkit/builders/WebKit%20Android%20(Nex...

https://build.chromium.org/p/chromium.webkit/builders/WebKit%20Android%20%28N....