Bug 278738 – [ Sonoma wk2 x86_64]: fast/canvas/image-buffer-backend-variants.html is a flaky crash.

• Status: NEW
• Component: JavaScriptCore

[Anfernee Viduya]

fast/canvas/image-buffer-backend-variants.html 
has been a flaky crash. 

HISTORY: 
https://results.webkit.org/?platform=mac&suite=layout-tests&test=fast%2Fcanvas%2Fimage-buffer-backend-variants.html

SYMBOLICATED LOG: 
0   com.apple.JavaScriptCore      	       0x59be75778    WTFCrashWithInfoImpl(int, char const*, char const*, int, unsigned long long)
1   com.apple.JavaScriptCore      	       0x59c3ccd26    JSC::LinkBuffer::linkCode(JSC::MacroAssembler&, JSC::JITCompilationEffort)
2   com.apple.JavaScriptCore      	       0x59cba2fa4    JSC::JIT::compileAndLinkWithoutFinalizing(JSC::JITCompilationEffort)
3   com.apple.JavaScriptCore      	       0x59cb8633c    JSC::BaselineJITPlan::compileInThreadImpl(JSC::JITCompilationEffort)
4   com.apple.JavaScriptCore      	       0x59cc7aefe    JSC::JITPlan::compileInThread(JSC::JITWorklistThread*)

LINK: 
https://build.webkit.org/results/Apple-Sonoma-Release-WK2-Tests/282788@main%20(4335)/fast/canvas/image-buffer-backend-variants-crash-log.txt

REPRODUCIBILITY: 
Currently I am setting up a machine that can reproduce this crash. Will Update bug when updated. 

DESCRIPTION: 
This test has been crashing on this queue for some time now. No exact regression point was determined.

[Radar WebKit Bug Importer]

<rdar://problem/134796136>

[EWS]

Test gardening commit 282808@main (a1ca824c11c2): <https://commits.webkit.org/282808@main>

Reviewed commits have been landed. Closing PR #32775 and removing active labels.

[Anfernee Viduya]

REPRODUCIBILITY: 
I was able to reproduce this issue on ToT using command. 
run-webkit-test fast/canvas/image-buffer-backend-variants.html --iterations=10

[Fujii Hironori]

The release assertion fails in performJITMemcpy:
> RELEASE_ASSERT(runLength <= maxZeroByteRunLength, buffer);

[Kimmo Kinnunen]

Marcus, could you take a look. Is the added release assert  exposing an older issue or is it a problem from the commit in bug  276913?

[Marcus Plutowski]

I'll take a look. This assert firing means that the AssemblerBuffer (where we emit code before it gets linked) contains an unexpectedly long string of 0s -- 16 or more, which should not be valid x86 ASM. A few questions:
  1. Does this reproduce locally, or just on the test runner?
  2. Does this reproduce when run alone, or only when run as part of a large batch of tests running in parallel?
  3. Is it possible for me to see the .ips files which resulted from the crashes in the link https://results.webkit.org/?platform=mac&suite=layout-tests&test=fast%2Fcanvas%2Fimage-buffer-backend-variants.html ?

[Anfernee Viduya]

(In reply to Marcus Plutowski from comment #6)
>   1. Does this reproduce locally, or just on the test runner?
It reproduces locally on specified machine configurations

>   2. Does this reproduce when run alone, or only when run as part of a large
> batch of tests running in parallel?
It does reproduce on its own, not needing a test batch to crash. 

>   3. Is it possible for me to see the .ips files which resulted from the
> crashes in the link
> https://results.webkit.org/?platform=mac&suite=layout-
> tests&test=fast%2Fcanvas%2Fimage-buffer-backend-variants.html ?
I'll try to get that file.