1229533 - When swap chain ResizeBuffers fails (as the first failure)

Status: RESOLVED FIXED

(Core :: Graphics: Layers, defect)

Description

[Milan Sreckovic [:milan] (needinfo for best results)]

In bug 1204922 we see mSwapChain->ResizeBuffers fails sometimes (CompositorD3D11), and we try to deal with the consequences.

But, why does ResizeBuffers fail in the first place?

For example, https://crash-stats.mozilla.com/report/index/909ecaa4-6f7c-4478-bbab-a05da2151201 fails with error 0x887a0001 (DXGI_ERROR_INVALID_CALL), on a fairly reasonably sized request (340x74).

Comment 1

[Milan Sreckovic [:milan] (needinfo for best results)]

Is it possible we somehow still didn't release the texture, despite our efforts just before calling ResizeBuffers in CompositorD3D11::VerifyBufferSize?  Any additional information we can get to help us decipher this?

Comment 2

[Milan Sreckovic [:milan] (needinfo for best results)]

Indirect references included...

Comment 3

[Bas Schouten (:bas.schouten)]

(In reply to Milan Sreckovic [:milan] from comment #1)
> Is it possible we somehow still didn't release the texture, despite our
> efforts just before calling ResizeBuffers in
> CompositorD3D11::VerifyBufferSize?  Any additional information we can get to
> help us decipher this?

We could log the refcount on -our- last release and see if it matches our expectations.

Comment 4

[Bas Schouten (:bas.schouten)]

Attachment: Ensure our rendertargetview for the backbuffer and the backbuffer itself have no lingering references, and log if they do.

Comment 5

[Milan Sreckovic [:milan] (needinfo for best results)]

Comment on attachment 8695336 [details] [diff] [review]
Ensure our rendertargetview for the backbuffer and the backbuffer itself have no lingering references, and log if they do.

Review of attachment 8695336 [details] [diff] [review]:
-----------------------------------------------------------------

R+, but either clarify or protect against the null pointer dereference in a couple of places.

::: gfx/layers/d3d11/CompositorD3D11.cpp
@@ +1234,5 @@
> +  ID3D11RenderTargetView* view = nullptr;
> +  mContext->OMSetRenderTargets(1, &view, nullptr);
> +
> +  RefPtr<ID3D11RenderTargetView> rtView = mDefaultRT->mRTView;
> +

It seems mDefaultRT can be null, given the test right after this, so we probably shouldn't just mDefaultRT->mRTView?

Can mRTView/rtView be null?  Because we use it below as if we guarantee it is always valid.

Comment 6

[Pulsebot]

https://hg.mozilla.org/integration/mozilla-inbound/rev/6899517c16a2

Comment 7

[Carsten Book [:Tomcat]]

https://hg.mozilla.org/mozilla-central/rev/6899517c16a2