Closed Bug 1203199 Opened 4 years ago Closed 4 years ago
crash in Deallocate
CB on Win7 with Intel drivers 9 .17 .10 .x
This bug was filed from the Socorro interface and is report bp-fdb603d4-12b0-40e8-b22d-13fa02150908. ============================================================= 0 d3d9.dll DeallocateCB Ø 1 igdumd32.dll igdumd32.dll@0x63d9 Ø 2 igdumd32.dll igdumd32.dll@0x11e18 Ø 3 igdumd32.dll igdumd32.dll@0x1108e 4 d3d9.dll CBatchFilterI::LHBatchDestroyResource(void*, void*) 5 d3d9.dll DestroyResource 6 d3d9.dll DdDestroySurfaceLH 7 d3d9.dll CDriverSurface::~CDriverSurface() 8 d3d9.dll CDriverSurface::`scalar deleting destructor'(unsigned int) 9 d3d9.dll CBaseObject::~CBaseObject() 10 dxva2.dll RELEASE<IDirect3DVideoDevice9>(IDirect3DVideoDevice9*&) 11 dxva2.dll CVideoDecoderDevice::DecoderCleanUp() 12 dxva2.dll CVideoDecoderDevice::CVideoDecoderDevice(CVideoAccelerationService*, _GUID const&, _DXVA2_VideoDesc const*, _DXVA2_ConfigPictureDecode const*, IDirect3DSurface9**, unsigned int, long*) 13 dxva2.dll CVideoAccelerationService::CreateVideoDecoder(_GUID const&, _DXVA2_VideoDesc const*, _DXVA2_ConfigPictureDecode const*, IDirect3DSurface9**, unsigned int, IDirectXVideoDecoder**) 14 msmpeg2vdec.dll CMFTH264DecoderTransform::CheckDecoderConfiguration(IDirectXVideoDecoderService*, _GUID, _DXVA2_VideoDesc*, _DXVA2_ConfigPictureDecode*) 15 msmpeg2vdec.dll CMFTH264DecoderTransform::ConfigureDXVA2Accelerator() 16 msmpeg2vdec.dll CMFTH264DecoderTransform::ProcessMessage(_MFT_MESSAGE_TYPE, unsigned long) 17 xul.dll mozilla::MFTDecoder::SendMFTMessage(_MFT_MESSAGE_TYPE, unsigned long) dom/media/platforms/wmf/MFTDecoder.cpp 18 xul.dll mozilla::WMFVideoMFTManager::InitInternal(bool) dom/media/platforms/wmf/WMFVideoMFTManager.cpp 19 @0x1b82f813 20 xul.dll mozilla::WMFMediaDataDecoder::Init() dom/media/platforms/wmf/WMFMediaDataDecoder.cpp 21 xul.dll mozilla::MediaFormatReader::EnsureDecodersSetup() dom/media/MediaFormatReader.cpp 22 xul.dll mozilla::MediaFormatReader::OnDemuxerInitDone(nsresult) dom/media/MediaFormatReader.cpp
This is almost entirely on Win7 and Win7SP1 so at first glance I'm going to assume it's different from bug 1092166 which was filed on Win8. Matt, this is both gfx and media so I figure you're a good starting point. This has been around for several releases but is now becoming prominent on the crash charts (maybe because we've fixed other bugs). About 2/3 of URLs contain facebook. Rank Adapter driver version Count % 1 126.96.36.19943 12300 51.65 % 2 188.8.131.5228 4194 17.61 % 3 184.108.40.20617 2609 10.95 % 4 220.127.116.1157 2165 9.09 % 5 18.104.22.16849 1636 6.87 % 6 22.214.171.12492 571 2.40 %
I should clarify: This is seen on 39 and 40 release, 41 beta, and in low volume on 42 aurora. Not seen on 43 nightly but that (and the aurora volume too) may be due to a difference in user populations.
I believe this is the same as bug 1092166, the implementation of CMFTH264DecoderTransform has changed a fair bit between windows 7 and 8, which explains the different stacks. Do the driver versions match up between the two bugs? This callstack happens before we get anything specific to the video fed into the decoder, so it seems like a driver bug. Blacklisting seems like the right way to go, do we know what % of our intel users that would be?
> Do the driver versions match up between the two bugs? Yes, both that bug and its duplicate were 126.96.36.19943. > Blacklisting seems like the right way to go, do we know what % of our intel > users that would be? Looking at crashes (which is an imperfect approximation) it's 3%.
Starting to increase in term of # of crash, tracking.
If we're going to blocklist video on these systems, let's just do it.
Component: Graphics → Audio/Video
Ok. It's not clear exactly which driver versions to blacklist here, comment 1 lists a few but comment 4 only mentions the one. Should we blacklist all driver versions before 188.8.131.5249? Jeff, could you please check with supersearch to make sure that's what we want? :)
Flags: needinfo?(matt.woodrow) → needinfo?(jmuizelaar)
Assignee: nobody → matt.woodrow
Attachment #8677159 - Flags: review?(jmuizelaar)
This is a wontfix for 42. However, we would be happy to take a patch in 43.
Comment on attachment 8677159 [details] [diff] [review] blacklist-intel Review of attachment 8677159 [details] [diff] [review]: ----------------------------------------------------------------- Sorry about the delay.
Attachment #8677159 - Flags: review?(jmuizelaar) → review+
https://hg.mozilla.org/integration/mozilla-inbound/rev/92730da4e7ebdc02f0d29423e19fa950e24c897a Backed out changeset 7b99659430b1 (bug 1203199) https://hg.mozilla.org/integration/mozilla-inbound/rev/1a41401462ca2647f2690853688709b334bb63d8 Backed out changeset ad08ab3d158e (bug 1203199) for widespread reftest bustage
Matt, do you want to uplift this to 43? Looks like 43 is affected, and the crash signature is still showing up in 43 beta.
Comment on attachment 8677159 [details] [diff] [review] blacklist-intel Approval Request Comment [Feature/regressing bug #]: DXVA video [User impact if declined]: Crashes with some drivers [Describe test coverage new/current, TreeHerder]: None, just crash stats. [Risks and why]: Simple blacklist change, very low risk. [String/UUID change made/needed]: None.
Comment on attachment 8677159 [details] [diff] [review] blacklist-intel Blacklist particular drivers for crash fix. Approved for uplift to aurora and beta.
Drivers beyond .2849 are crashing as well; here's one with 184.108.40.20657 - https://crash-stats.mozilla.com/report/index/06ccd70d-1c9e-4b58-bd4d-0f4ee2160926
(In reply to Milan Sreckovic [:milan] from comment #25) > Drivers beyond .2849 are crashing as well; here's one with 220.127.116.1157 - > https://crash-stats.mozilla.com/report/index/06ccd70d-1c9e-4b58-bd4d- > 0f4ee2160926 Do you have any idea of what the highest version number we have a crash with is?
That's the highest I'm seeing for this crash - 18.104.22.16857 (the highest driver version in that branch outside of this crash seems to be 22.214.171.12459, but nothing above 2857 shows up for this crash.)
3 years ago
You need to log in before you can comment on or make changes to this bug.