Closed
Bug 660850
Opened 14 years ago
Closed 14 years ago
TI: Crash [@ js::mjit::JITScript::nativeToPC]
Categories
(Core :: JavaScript Engine, defect)
Core
JavaScript Engine
Tracking
()
RESOLVED
FIXED
People
(Reporter: gkw, Unassigned)
References
Details
(Keywords: crash, testcase)
Crash Data
Attachments
(1 file)
1.96 KB,
text/plain
|
Details |
The attached testcase in the next comment crashes js debug shell on JM changeset 0cc71b0c30f4 with -m, -a and -n at js::mjit::JITScript::nativeToPC
![]() |
Reporter | |
Comment 2•14 years ago
|
||
debug stack on Linux
Comment 3•14 years ago
|
||
The root problem here is that CompileFunction did not respect several invariants used in other call paths to make sure we are able to recompile a caller while in the middle of compiling its callee. The VMFrame's regs were incorrect, the callee frame was being pushed onto f.regs inappropriately, the writes to ncode happened differently. Other places in the recompiler code were hacking around this already, but these were incomplete. Rather than pile more complexity onto CompileFunction, the fix below guts it so that it just pops the partially initialized frame and invokes UncachedNew/UncachedCall (a fix I've wanted to do for some time).
http://hg.mozilla.org/projects/jaegermonkey/rev/750eb028fb06
Status: NEW → RESOLVED
Closed: 14 years ago
Resolution: --- → FIXED
Updated•14 years ago
|
Crash Signature: [@ js::mjit::JITScript::nativeToPC]
You need to log in
before you can comment on or make changes to this bug.
Description
•