1435693 - Sometimes failed to create Pbuffer of back buffer for WebRender

Status: RESOLVED DUPLICATE of bug 1441056

(Core :: Graphics: WebRender, defect, P1)

Description

[Sotaro Ikeda [:sotaro]]

When I updated nightly, sometime it failed to create Pbuffer of back buffer for WebRender and fallbacked to normal compositor.

When it happened, about:support had the following failure log.

> GP+[GFX1]: Failed to create Pbuffer of back buffer error: 0x300c Size : Size(292,0)

Comment 1

[Sotaro Ikeda [:sotaro]]

When sanityTest of SanityTest.js run, the problem seems to happen.

Comment 2

[Sotaro Ikeda [:sotaro]]

(In reply to Sotaro Ikeda [:sotaro] from comment #1)
> When sanityTest of SanityTest.js run, the problem seems to happen.

verifyLayersRendering() always seemed to be failed. Then triggerDeviceReset() was called. It end up to GPUProcessManager::SimulateDeviceReset() calls. It tries to re-create process sessions.
  https://dxr.mozilla.org/mozilla-central/source/toolkit/components/gfx/SanityTest.js#142

And tries to recreate WebRenderBridgeParent. But it failed because Bug 1364504 is not addressed yet. Then nsBaseWidget::CreateCompositorSession() fallbacked to normal compositor session. but the normal compositor session did not work some how.

Comment 3

[Sotaro Ikeda [:sotaro]]

(In reply to Sotaro Ikeda [:sotaro] from comment #2)
> And tries to recreate WebRenderBridgeParent. But it failed because Bug
> 1364504 is not addressed yet. Then nsBaseWidget::CreateCompositorSession()
> fallbacked to normal compositor session. but the normal compositor session
> did not work some how.

When WebRender was fallback to normal compositor, the compositor seems to work as normal, but window was not updated. I tried a lot of RenderCompositorANGLE code modification patterns to find out a cause of it. It seems that DXGI_SWAP_EFFECT_FLIP_SEQUENTIAL seems to cause the problem. When I changed DXGI_SWAP_EFFECT_FLIP_SEQUENTIAL to DXGI_SWAP_EFFECT_SEQUENTIAL, the window update problem did not happen.

MDN has the following. It seems to be related.

 > Use one flip model swap chain per HWND. Do not target multiple flip model swap chains to the same HWND.
 https://msdn.microsoft.com/ja-jp/library/windows/desktop/hh706346

Comment 4

[Sotaro Ikeda [:sotaro]]

Created Bug 1435995 for Comment 3.

Comment 5

[Sotaro Ikeda [:sotaro]]

Attachment: temporal patch - Run SanityTest for every launch

Comment 6

[Sotaro Ikeda [:sotaro]]

During SanityTest, verifyLayersRendering() always failed. I noticed it, since with SanityTest, nsDOMWindowUtils::TriggerDeviceReset() was always called.

Comment 7

[Sotaro Ikeda [:sotaro]]

(In reply to Sotaro Ikeda [:sotaro] from comment #6)
> During SanityTest, verifyLayersRendering() always failed. I noticed it,
> since with SanityTest, nsDOMWindowUtils::TriggerDeviceReset() was always
> called.

Bug 1436020 is created for it.

Comment 9

[Sotaro Ikeda [:sotaro]]

The problem seems to be addressed by Bug 1441056.

Comment 10

[Vishinko]

(In reply to Sotaro Ikeda [:sotaro] from comment #9)
> The problem seems to be addressed by Bug 1441056.

I confirm that I no longer get this error after that, however WebRender still fails to initialize as I'm now hitting this bug: Bug 1441827.

> GP+[GFX1-]: Failed to load EGL library: FEATURE_FAILURE_NO_DISPLAY

Comment 11

[Sotaro Ikeda [:sotaro]]

Bug 1441827 is a different bug, then I close this bug.