964355 - crash in RtlInterlockedFlushSList | winmm_buffer_thread explosive in Aurora 28 since 2014-01-24

Status: RESOLVED INCOMPLETE

(Core :: Audio/Video: cubeb, defect)

Description

[u279076]

This bug was filed from the Socorro interface and is 
report bp-cadc57c3-c685-4509-b193-beb7d2140127.
=============================================================
0 	ntdll.dll 	RtlInterlockedFlushSList 	
1 	gkmedias.dll 	winmm_buffer_thread 	media/libcubeb/src/cubeb_winmm.c
2 	msvcr100.dll 	_callthreadstartex 	f:\dd\vctools\crt_bld\self_x86\crt\src\threadex.c
3 	msvcr100.dll 	_threadstartex 	f:\dd\vctools\crt_bld\self_x86\crt\src\threadex.c
4 	kernel32.dll 	BaseThreadStart

More Reports:
https://crash-stats.mozilla.com/report/list?product=Firefox&signature=RtlInterlockedFlushSList+%7C+winmm_buffer_thread

Current Rank:
 * Firefox 29: N/A (0 crashes reported)
 * Firefox 28: #19 @ 0.59%
 * Firefox 27: Low volume (23 crashes reported)
 * Firefox 26: Low volume (93 crashes reported)

Platforms:
 * Windows XP: 100%

Correlations:
 * null

URLs:
 * mostly facebook.com

Pushlog:
http://hg.mozilla.org/releases/mozilla-aurora/pushloghtml?fromchange=e8f6bdf8db3d&tochange=a73b697b50b3
=============================================================

This crash is not yet in topcrash territory but it has shown up at the top of the explosiveness reports for Aurora 28. It first showed up on 2014-01-24 with ~177 crashes per 1MM ADU and has remained around there since.

Comment 1

[(Away)]

It is suspicious that all of the reports are on Windows XP. (Though I do see both SP2 and SP3, so it's not specific down to a single build.)

The exception code is EXCEPTION_IN_PAGE_ERROR which is not a typical one. We failed to page in the code for ntdll!RtlInterlockedFlushSList. !analyze on various reports says: IO_ERROR: (NTSTATUS) 0xc000026e - An operation was attempted to a volume after it was dismounted.

I don't think we can really expect to survive a dismount of the system volume. Maybe newer Windows made things more resilient.

The "spike" on Aurora all seems to be coming from the same user. (It's across three installations but the locale and CPU model are the same) While I'm curious why that person's volume gets dismounted so often, this may not be worth spending time on unless it becomes more widespread.

Comment 2

[Lukas Blakk [:lsblakk] use ?needinfo]

again, not tracking if this is mostly from one user causing a false "spike" on a small-population channel, renom if this arises in beta 28.