Bug 216406

Summary:	REGRESSION(r266770): construct-return-early-from-infinite-loop-for-fuzzer.js loops on x86_64 cloop
Product:	WebKit	Reporter:	Michael Catanzaro <mcatanzaro>
Component:	JavaScriptCore	Assignee:	Nobody <webkit-unassigned>
Status:	RESOLVED FIXED
Severity:	Normal	CC:	guijemont, mcatanzaro, webkit-bug-importer, ysuzuki
Priority:	P2	Keywords:	InRadar
Version:	WebKit Nightly Build
Hardware:	PC
OS:	Linux
See Also:	https://bugs.webkit.org/show_bug.cgi?id=216289 https://bugs.webkit.org/show_bug.cgi?id=258179

Description Michael Catanzaro 2020-09-11 09:07:40 PDT

The new construct-return-early-from-infinite-loop-for-fuzzer.js test added in r266770 "[JSC] returnEarlyFromInfiniteLoopsForFuzzing should return object" loops forever on llint, but only on x86_64. Other architectures seems to be fine. (So this doesn't affect any architectures that commonly use llint, but x86_64 is great to have working as a baseline to see when problems occur only on a particular architecture.)

Comment 1 Guillaume Emont 2020-09-14 07:28:23 PDT

(In reply to Michael Catanzaro from comment #0)
> The new construct-return-early-from-infinite-loop-for-fuzzer.js test added
> in r266770 "[JSC] returnEarlyFromInfiniteLoopsForFuzzing should return
> object" loops forever on llint, but only on x86_64. Other architectures
> seems to be fine. (So this doesn't affect any architectures that commonly
> use llint, but x86_64 is great to have working as a baseline to see when
> problems occur only on a particular architecture.)

Did you mean when using CLoop (JIT disabled at compile time)? Or with LLInt compiled with the x86_64 backend (default) and disabling higher tiers at runtime? FWIW x86_64 does use LLInt as its first tier by default.

Comment 2 Michael Catanzaro 2020-09-14 08:03:04 PDT

Yes, this is with cloop enabled and JIT disabled. (The point of testing cloop on x86_64 is to distinguish between architecture-specific bugs and general cloop bugs.)

Comment 3 Michael Catanzaro 2020-09-15 16:47:19 PDT

Is there anything I should try running that could maybe help with debugging this? It hangs our internal Jenkins builder, and I'm starting to get nervous about missing CI coverage since it's now been a week since the regression landed.

Comment 4 Yusuke Suzuki 2020-09-15 16:48:44 PDT

I think maybe returnEarlyFromInfiniteLoopsForFuzzing's loop-hint thing is not supported in CLoop...? For now, I'll put `isUseJIT()` check

Comment 5 Yusuke Suzuki 2020-09-15 16:51:48 PDT

Committed r267115: <https://trac.webkit.org/changeset/267115>

Comment 6 Radar WebKit Bug Importer 2020-09-15 16:52:19 PDT

<rdar://problem/68950461>

Comment 7 Michael Catanzaro 2020-09-15 17:52:44 PDT

OK, thanks Yusuke! $vm.useJIT() will certainly guarantee the test doesn't run.