1956094 - Crash in [@ mozilla::gfx::GPUProcessManager::OnProcessUnexpectedShutdown]

Status: NEW

(Core :: Graphics, defect)

Description

[osahashi]

Crash report: https://crash-stats.mozilla.org/report/index/7209acd2-056b-458f-9daa-1aab60250324

I apologize for questions or mistakes, I have never done a report like this before.

Since I have an AMD GPU I've been struggling with problems in Firefox, I've tried hundreds of things, nothing has solved the problem. Finally I was able to visualize something through logs, so I hope to get some help now.

MOZ_CRASH Reason:

When playing YouTube or Twitch, the AMD driver crashed with a timeout, the last two drivers no longer give an error message (timeout), they just crash and then report a crash. Of course, I have reported every crash with data to AMD, but so far AMD has neither responded nor fixed the crashes in driver updates.

The crashes only occur when media is playing and something is being done in another tab or window on another screen. I don't know if the crashes are reproducible, I hope you can read any information or draw any conclusions from the crash report data.

This crash problem has persisted for half a year, no Firefox version, no Windows update, no AMD update, no Bios update, no tweaks, no software, nothing has solved the problem

I also have the following data:
Failure Log

(#0) 	GP+[GFX1-]: WaitForFrameGPUQuery() failed
(#1) 	GP+[GFX1-]: video Present failed: 0x887a0005
(#2) 	GP+[GFX1-]: Could not create D3D11_QUERY_EVENT: 0x887a0005
(#3) Error	Detect DeviceReset DeviceResetReason::RESET DeviceResetDetectPlace::WR_POST_UPDATE in GPU process
(#4) 	GP+[GFX1-]: [D3D11] 2 CreateTexture2D failure Size: Size(1920,1080)texture11: false Code: 0x887a0005

MOZ_CRASH(GPU process crashed and pref is set to crash the browser.)

Top 10 frames:

0  xul.dll  mozilla::gfx::GPUProcessManager::OnProcessUnexpectedShutdown(mozilla::gfx::GP...  gfx/ipc/GPUProcessManager.cpp:909
1  xul.dll  mozilla::gfx::GPUProcessHost::OnChannelClosed()  gfx/ipc/GPUProcessHost.cpp:204
2  xul.dll  mozilla::gfx::GPUChild::ActorDestroy(mozilla::ipc::IProtocol::ActorDestroyRea...  gfx/ipc/GPUChild.cpp:347
3  xul.dll  mozilla::ipc::IProtocol::ActorDisconnected::<lambda_0>::operator()(mozilla::i...  ipc/glue/ProtocolUtils.cpp:630
3  xul.dll  mozilla::ipc::IProtocol::ActorDisconnected(mozilla::ipc::IProtocol::ActorDest...  ipc/glue/ProtocolUtils.cpp:657
4  xul.dll  mozilla::ipc::IToplevelProtocol::OnChannelError()  ipc/glue/ProtocolUtils.h:544
4  xul.dll  mozilla::ipc::MessageChannel::NotifyMaybeChannelError(mozilla::ReleasableMoni...  ipc/glue/MessageChannel.cpp:2035
5  xul.dll  mozilla::ipc::MessageChannel::OnNotifyMaybeChannelError()  ipc/glue/MessageChannel.cpp:2061
6  xul.dll  mozilla::detail::RunnableMethodArguments<>::apply<mozilla::ipc::MessageChanne...  xpcom/threads/nsThreadUtils.h:1085
6  xul.dll  std::invoke(mozilla::detail::RunnableMethodArguments<>::apply<mozilla::ipc::M...  /builds/worker/fetches/vs/VC/Tools/MSVC/14.39.33519/include/type_traits:1729

Comment 1

[BugBot [:suhaib / :marco/ :calixte]]

The Bugbug bot thinks this bug should belong to the 'Core::Graphics' component, and is moving the bug to that component. Please correct in case you think the bot is wrong.

Comment 2

[osahashi]

Failure Log
(#0) 	GP+[GFX1-]: WaitForFrameGPUQuery() failed
(#1) Error	Handling webrender error 5
(#2) 	GP+[GFX1-]: ID3D11VideoProcessorInputView creation failed: 0x80070057
(#3) 	GP+[GFX1-]: Could not create D3D11_QUERY_EVENT: 0x887a0005
(#4) Error	Detect DeviceReset DeviceResetReason::RESET DeviceResetDetectPlace::WR_POST_UPDATE in GPU process
(#5) 	GP+[GFX1-]: [D3D11] 2 CreateTexture2D failure Size: Size(1280,720)texture11: false Code: 0x887a0005

Comment 3

[osahashi]

Another crash report: https://crash-stats.mozilla.org/report/index/288b22b2-793d-41d7-bee5-6cb150250325

Comment 4

[BugBot [:suhaib / :marco/ :calixte]]

The bug has a crash signature, thus the bug will be considered confirmed.

Comment 5

[osahashi]

AMD GPU driver crash watching Twitch, no crash report, just:

Failure Log
(#0) 	GP+[GFX1-]: WaitForFrameGPUQuery() failed
(#85) 	GP+[GFX1-]: RenderDXGITextureHost init stream failed
(#86) 	GP+[GFX1-]: RenderDXGITextureHost init stream failed
(#87) 	GP+[GFX1-]: RenderDXGITextureHost init stream failed
(#88) 	GP+[GFX1-]: RenderDXGITextureHost init stream failed
(#89) 	GP+[GFX1-]: RenderDXGITextureHost init stream failed
(#90) 	GP+[GFX1-]: RenderDXGITextureHost init stream failed
(#91) 	GP+[GFX1-]: RenderDXGITextureHost init stream failed
(#92) 	GP+[GFX1-]: RenderDXGITextureHost init stream failed
(#93) 	GP+[GFX1-]: RenderDXGITextureHost init stream failed
(#94) 	GP+[GFX1-]: RenderDXGITextureHost init stream failed
(#95) 	GP+[GFX1-]: RenderDXGITextureHost init stream failed
(#96) 	GP+[GFX1-]: RenderDXGITextureHost init stream failed
(#97) 	GP+[GFX1-]: RenderDXGITextureHost init stream failed
(#98) 	GP+[GFX1-]: RenderDXGITextureHost init stream failed
(#99) 	GP+[GFX1-]: RenderDXGITextureHost init stream failed

Comment 6

[Lee Salzman [:lsalzman]]

Sotaro, do you have ideas what might be going on here?

Comment 7

[osahashi]

I made a completely new profile with the 136.0.4 (64-bit), no settings changed, everything default, the GPU driver crashed on a YouTube video, only the YouTube video was open in a single tab. No crash report was created. Only the information:

GP+[GFX1-]: WaitForFrameGPUQuery() failed
GP+[GFX1-]: RenderDXGITextureHost init stream failed

Maybe also helpful, in the normal profile a new error message appeared when crashing on Twitch:

(#0) Error	Detect DeviceReset DeviceResetReason::RESET DeviceResetDetectPlace::WR_POST_UPDATE in GPU process
(#1) 	GP+[GFX1-]: Failed to make render context current during destroying.
(#2) 	GP+[GFX1-]: Error in eglMakeCurrent: 0x300e
(#3) 	GP+[GFX1-]: Error in eglDestroySurface: 0x300e

I have also sent another bug report to AMD with reference to this bug report and contact details to me, maybe someone will get back to me this time.

I don't know what to do anymore, stability tests say the GPU runs without problems and except for Firefox in media playback there are no problems. Windows finds no problems in tests. I have tested the memory without any problems. Tried every setting, set everything to default, redid drivers and carefully removed them beforehand.

Comment 8

[Kelsey Gilbert [:jgilbert]]

Bob, can you ask sotaro?

Comment 9

[Ashley Hale [:ahale]]

Can you go to about:support and click the Copy text to clipboard button near the top of the page, then go to this bug and click Attach New File and paste into that so we can see your about:support text on the bug? It would be helpful to see several details of your hardware and video codec support in understanding the bug.

I would like to better understand why the Direct3D device reset is happening.

As a workaround you can try going to about:config and turning off any of these options you wish to experiment with:
media.hardware-video-decoding.enabled
gfx.webrender.dcomp-video-hw-overlay-win
gfx.webrender.dcomp-video-sw-overlay-win
gfx.webrender.dcomp-video-vp-scaling-win
gfx.webrender.dcomp-win.enabled

Turning off any of those options may have a significant performance impact, but it would be good to know which ones mitigate the crash, so if you don't mind experimenting it would be valuable data and narrows the possible causes - and also lets you use the browser while we investigate this bug.

In the worst case you can try going to about:config and toggling this option but it should not be necessary to do so:
gfx.webrender.software

Comment 10

[osahashi]

Attachment: Application Basics

Comment 11

[osahashi]

The last few days I did a CMOS, rearranged the ram bars and moved the graphics card from slot 1 to slot 2, since then there has been no crash. It is possible that PCIe slot 1 on my mainboard is defect/broken due to the change of graphics card. However, I am cautious as to whether this will really solve the problem in the long term, as the crashes have always been sporadic. But I hope so.

If it comes to crashes again I will try the about:config settings and test if there is a change.

Comment 12

[osahashi]

Attachment: Crash before new settings test - after moving GPU to new slot.txt

Maybe some useful data

Comment 13

[osahashi]

Attachment: Crash AFTER new settings test - off.txt

Microsoft updated their mediacodec pugins yesterday, maybe it will help somehow, wait and see. No feedback from AMD, no new GPU driver released yet.

I apologize for the workload of the ticket.

Comment 14

[Brad Werth [:bradwerth]]

Thank you for filing. If you can reproduce this consistently, there's something I'd like you to try. This crash only happens when the layers_gpu_process_crash_also_crashes_browser pref is set. Would you please navigate to about:config and change that pref to false? Then try to reproduce the crash. What happens in that case? Does everything "just work", or is there some other manifestation of the problem?

Comment 15

[osahashi]

(In reply to Brad Werth [:bradwerth] from comment #14)

Thank you for filing. If you can reproduce this consistently, there's something I'd like you to try. This crash only happens when the layers_gpu_process_crash_also_crashes_browser pref is set. Would you please navigate to about:config and change that pref to false? Then try to reproduce the crash. What happens in that case? Does everything "just work", or is there some other manifestation of the problem?

Firefox does not create a report when the crash occurs, so I used “layers.gpu-process.crash-also-crashes-browser” for the crash reports, even with this setting activated a report is not always created. After I created this ticket here, I deactivated the setting again. the setting “layers.gpu-process.crash-also-crashes-browser” has no influence on the crashes. In the attachments you can see that I have deactivated this setting and the crashes still happen.

The settings:
gfx.webrender.dcomp-video-hw-overlay-win
gfx.webrender.dcomp-video-sw-overlay-win
gfx.webrender.dcomp-video-vp-scaling-win
gfx.webrender.dcomp-win.enabled
I activated it again a few days ago. There has not been a single crash since then. So apparently they have no influence on the problem, whether on or off. It crashed with on and with off.

I have also tested a new profile, 28 days ago, I had a crash with all default settings, so I don't think it was due to a setting. If the crashes start again I can test again with a new profile and upload all the information.