1753043 - Crash when running the profiler under rr

Status: NEW

(Core :: Gecko Profiler, defect, P5)

Description

[Julien Wajsberg [:julienw]]

STR:

1. Run a firefox build with rr
2. Start the profiler

=> crash

Here is a pernosco link for this crash for a debug build: https://pernos.co/debug/gefrJF45YPulQnkvWBJabQ/index.html

Here is a pernosco link for this crash for a no-debug build: https://pernos.co/debug/4hZOME8ATDRlstWWcAqBhA/index.html

Comment 1

[Gerald Squelart (he/him) (not at Mozilla since 2022-09-15)]

In both Pernosco traces, DoLULBacktrace wants to copy > 261,700 bytes, it's restricted to N_STACK_BYTES=160KB, but then the memcpy triggers a SIGSEGV. Pernosco links near the N_STACK_BYTES check and just before the SIGSEGV: debug, no-debug.

We had had this problem in bug 1726125, where I was able to avoid these crashes by giving up if it wanted to copy >1MB. But it seems it's crashing now with a smaller copy.

Markus, you know more about LUL (and rr?), would you mind having a look?
Should we also give up when reaching N_STACK_BYTES? Is there a way to predict the SIGSEGV?

Comment 2

[Gerald Squelart (he/him) (not at Mozilla since 2022-09-15)]

Kyle, any idea about this please?

Comment 3

[Kyle Huey [:khuey]]

The profiler is not compatible with rr's syscall buffering. Use rr record -n.