REGRESSION(r137994): It made JSC tests hang and layout tests fail on ARM Thumb2 / Linux

[Csaba Osztrogonác]

Tested on QtWebKit, but it might be valid on other platforms too.

- r137993 (with r138032 buildfix applied): http://build.webkit.sed.hu/builders/ARMv7%20Linux%20Qt5%20Release%20%28Test%29/builds/7810
- r137994 (with r138032 buildfix applied): http://build.webkit.sed.hu/builders/ARMv7%20Linux%20Qt5%20Release%20%28Test%29/builds/7703

[Csaba Osztrogonác]

- ToT (r141579) - http://cute1.inf.u-szeged.hu:8010/builders/ARMv7%20Linux%20Qt5%20Release%20%28Test%29/builds/7809 - JSC tests still hang and inspector tests still crash
- ToT (r141590) with reverted r137994 and r138474 - http://build.webkit.sed.hu/builders/ARMv7%20Linux%20Qt5%20Release%20%28Test%29/builds/7811 - JSC tests pass, but inspector tests still crash (maybe related to other bug too)

[Csaba Osztrogonác]

The bug is still valid on ToT . :-/ Are you going to fix it or should 
we revert the buggy patch until somebody find the proper fix?

r142785 - http://build.webkit.sed.hu/builders/ARMv7%20Linux%20Qt5%20Release%20%28Test%29/builds/7876
r142785 (with reverted r137994 and r138474) - http://trac.webkit.org/changeset/142785

[Csaba Osztrogonác]

(In reply to comment #2)
> The bug is still valid on ToT . :-/ Are you going to fix it or should 
> we revert the buggy patch until somebody find the proper fix?
> 
> r142785 - http://build.webkit.sed.hu/builders/ARMv7%20Linux%20Qt5%20Release%20%28Test%29/builds/7876
> r142785 (with reverted r137994 and r138474) - 
ooops, wrong URL, here is the correct one: http://build.webkit.sed.hu/builders/ARMv7%20Linux%20Qt5%20Release%20%28Test%29/builds/7877

[Uli Schlachter]

Assuming you mean me with "you": Nope, no idea what I could do about this. I don't see how the patch could cause hangs unless mprotect() is painfully slow on ARM and gets called too often. Also, I don't have any ARM devices to experiment with.

So the best that I can suggest: Restrict the #if OS(LINUX) in OSAllocatorPosix.cpp to also exclude ARM so that on these devices overcommit is still used.

[Simon Hausmann]

Created attachment 190062 [details]
Patch

Patch re-enables overcommit on ARM

[Csaba Osztrogonác]

Just a note: It caused problem with only with ARM Thumb2 assembler, but didn't with ARM traditional assembler on the same hardware. So I think the problem
is something else here. Maybe a hidden bug was revealed by changing allocator.

[Csaba Osztrogonác]

Comment on attachment 190062 [details]
Patch

rs=me as workaround until proper fix

[Csaba Osztrogonác]

I checked this patch on the bot and it fixed most of the failures on Thumb2:
http://build.webkit.sed.hu/builders/ARMv7%20Linux%20Qt5%20Release%20%28Test%29/builds/7938

The remaining ~150 inspector crash is unrelated to this bug, needs more investigation.

[WebKit Review Bot]

Comment on attachment 190062 [details]
Patch

Clearing flags on attachment: 190062

Committed r144170: <http://trac.webkit.org/changeset/144170>

[WebKit Review Bot]

All reviewed patches have been landed.  Closing bug.

[Csaba Osztrogonác]

Reopen, because only the workaround landed, we need to find the root 
of the problem. (rgabor said he will check it)

[Gabor Rapcsanyi]

I opened a new bug for this: https://bugs.webkit.org/show_bug.cgi?id=111441

[Csaba Osztrogonác]

*** This bug has been marked as a duplicate of bug 111441 ***