1316327 - Crash in mozilla::WebGLFramebuffer::BlitFramebuffer (Webgl2.0 regression)

Status: RESOLVED FIXED

(Core :: Graphics: CanvasWebGL, defect, P3)

Description

[marcot]

User Agent: Mozilla/5.0 (Windows NT 10.0; WOW64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/53.0.2785.143 Safari/537.36

Steps to reproduce:

- enable webgl 2.0 in Firefox 52
- open the attached repro


Actual results:

tab crashed


Expected results:

renders a scene with some lights. See firefox 49 for expected visuals.

Comment 1

[marcot]

More findings:

1) if webgl.enable-webgl2;false on Firefox52-Windows, lights don't render and I get errors like:
Error: WebGL: drawElements: Feedback loop detected between tex target 0x0de1, tex unit 0, levels 0-0; and framebuffer attachment 0x821a, level 0.

2) If webgl.enable-webgl2;true and webgl.disable-angle;true on Firefox49-Windows: Everything renders fine.

I should also mention that this repro will try to use WEBGL_draw_buffers, so issue #2 probably indicates a bug with WEBGL_draw_buffers and Angle (can repro on Chrome-Windows but not on Edge)

Comment 2

[Loic]

Could you attach the testcase, please.

Comment 3

[marcot]

It looks like the test case failed to upload because exceeded the size limit. You can download it here: https://oc.unity3d.com/index.php/s/hZnlYroDEspI3zT

Comment 4

[Liviu Cirdei]

Thanks for the testcase. 

I can confirm the issue on Windows 10 (Intel(R) HD Graphics) on latest Nightly 52 (Build ID 20161109030210).

I did a regression range and this is what I found:

Pushlog: https://hg.mozilla.org/integration/mozilla-inbound/pushloghtml?fromchange=25f2e739a961677e6eb1f296ddaf59c46207bcbc&tochange=b32a687a0a3ebd6db5806e69621563a0300fe6ce

Regressed by bug 1300946.

The issue is reproducible also on Aurora 51 (Build ID 20161108004019), probably because of the uplift on that bug. Not sure if Beta is affected.

Comment 5

[Liviu Cirdei]

The issue is also reproducible on Mac 10.12.1 and Ubuntu 16.04 (Nightly and Aurora). Beta is not affected.

Comment 6

[Loic]

https://crash-stats.mozilla.com/report/index/9a243261-d658-4137-a9b7-3388b2161110

Comment 7

[Marcia Knous [:marcia]]

Tracking 52+ for this reproducible crash.

Comment 8

[Gerry Chang [:gchang]]

Hi Peter,
Can you help find an owner for this?

Comment 9

[Gerry Chang [:gchang]]

Track 51+ as regression related to WebGL 2.

Comment 10

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

Assertion failure: IsDefined(), at /Users/msreckovic/Repos/mozilla-central/dom/canvas/WebGLFramebuffer.cpp:75
#01: mozilla::WebGLFBAttachPoint::Format() const[/Users/msreckovic/Repos/mozilla-central/obj-x86_64-apple-darwin15.6.0/dist/NightlyDebug.app/Contents/MacOS/XUL +0x36728ec]
#02: mozilla::WebGLFramebuffer::BlitFramebuffer(mozilla::WebGLContext*, mozilla::WebGLFramebuffer const*, int, int, int, int, mozilla::WebGLFramebuffer const*, int, int, int, int, unsigned int, unsigned int)[/Users/msreckovic/Repos/mozilla-central/obj-x86_64-apple-darwin15.6.0/dist/NightlyDebug.app/Contents/MacOS/XUL +0x367a091]
#03: mozilla::WebGL2Context::BlitFramebuffer(int, int, int, int, int, int, int, int, unsigned int, unsigned int)[/Users/msreckovic/Repos/mozilla-central/obj-x86_64-apple-darwin15.6.0/dist/NightlyDebug.app/Contents/MacOS/XUL +0x360ac0d]
#04: mozilla::dom::WebGL2RenderingContextBinding::blitFramebuffer(JSContext*, JS::Handle<JSObject*>, mozilla::WebGL2Context*, JSJitMethodCallArgs const&)[/Users/msreckovic/Repos/mozilla-central/obj-x86_64-apple-darwin15.6.0/dist/NightlyDebug.app/Contents/MacOS/XUL +0x2f4b215]
...

Comment 11

[Peter Chang[:pchang]]

Michael, please take a look.

Comment 12

[Michael Leu[:Lenzak](UTC+8)]

I can reproduce it on Mac OSX 10.11, too.

I'll look into this one.

Comment 13

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

https://dxr.mozilla.org/mozilla-central/source/dom/canvas/WebGLFramebuffer.cpp#1625 :

for (const auto& drawBuffer : dstFB->mColorDrawBuffers) {
    fnCheckColorFormat(drawBuffer->Format()->format);   <---- drawBuffer->IsDefined() fails here
}

I'm guessing the WebGLFramebuffer constructor should not do this:

mColorDrawBuffers.push_back(&mColorAttachments[0]);

when the color attachment isn't defined.

Comment 14

[Peter Chang[:pchang]]

Michael, any update?

Comment 15

[Michael Leu[:Lenzak](UTC+8)]

I am still investigating this crash.

Trying to figure out why the draw buffer is undefined here.

We can skip undefined buffers by adding if-checks and the testcase would not crash, but I don't think it is a right solution.

Comment 16

[Michael Leu[:Lenzak](UTC+8)]

Attachment: Check colorbuffer is defined or not when constructing WebGLFramebuffer

Comment 17

[Michael Leu[:Lenzak](UTC+8)]

I thought it must encounter some failure which prevent ColorBuffer from being initialized and defined correctly, but I didn't see any failure about framebuffer manipulation before the crash happens. Maybe the ColorBuffer was not going to be defined throughout the WebGL program execution.

So I think Milan's idea is right, we just simply skip adding those undefined buffers when we create WebGLFramebuffer.

Comment 18

[Kelsey Gilbert [:jgilbert]]

Comment on attachment 8812681 [details] [diff] [review]
Check colorbuffer is defined or not when constructing WebGLFramebuffer

Review of attachment 8812681 [details] [diff] [review]:
-----------------------------------------------------------------

::: dom/canvas/WebGLFramebuffer.cpp
@@ +623,5 @@
> +    if (mColorAttachments[0].IsDefined()) {
> +        mColorDrawBuffers.push_back(&mColorAttachments[0]);
> +        mColorReadBuffer = &mColorAttachments[0];
> +    } else {
> +        mColorReadBuffer = nullptr;

mColorReadBuffer makes no assumptions about whether it is defined. It is simply how we track which attachment the ReadBuffer is pointed at.

See WebGLFramebuffer::ResolvedData for where we gather a list of non-empty attachments.

Comment 19

[Michael Leu[:Lenzak](UTC+8)]

Attachment: Iterate through defined ColorBuffers from mResolvedCompleteData instead of mColorDrawBuffers.

Comment 20

[Michael Leu[:Lenzak](UTC+8)]

Comment on attachment 8813971 [details] [diff] [review]
Iterate through defined ColorBuffers from mResolvedCompleteData instead of mColorDrawBuffers.

Review of attachment 8813971 [details] [diff] [review]:
-----------------------------------------------------------------

Access ColorDrawBuffers by mResolvedCompleteData to prevent accessing undefined buffer.

Comment 21

[Kelsey Gilbert [:jgilbert]]

Comment on attachment 8813971 [details] [diff] [review]
Iterate through defined ColorBuffers from mResolvedCompleteData instead of mColorDrawBuffers.

Review of attachment 8813971 [details] [diff] [review]:
-----------------------------------------------------------------

::: dom/canvas/WebGLFramebuffer.cpp
@@ +1632,4 @@
>  
>          dstDepthFormat = fnGetFormat(dstFB->mResolvedCompleteData->depthBuffer);
>          dstStencilFormat = fnGetFormat(dstFB->mResolvedCompleteData->stencilBuffer);
> +        if (dstFB->mResolvedCompleteData) {

This check is either not needed, or is needed in the two lines above.
Since it's not used in the above two lines, I suspect it's not needed below.

Comment 22

[Michael Leu[:Lenzak](UTC+8)]

Attachment: Iterate through defined ColorBuffers from mResolvedCompleteData instead of mColorDrawBuffers.

Remove unnecessary null check.

Comment 23

[Michael Leu[:Lenzak](UTC+8)]

Attachment: Iterate through defined ColorBuffers from mResolvedCompleteData instead of mColorDrawBuffers.

Comment 24

[Michael Leu[:Lenzak](UTC+8)]

Comment on attachment 8815601 [details] [diff] [review]
Iterate through defined ColorBuffers from mResolvedCompleteData instead of mColorDrawBuffers.

Review of attachment 8815601 [details] [diff] [review]:
-----------------------------------------------------------------

I think the null check can be just removed.

Comment 25

[Michael Leu[:Lenzak](UTC+8)]

Here is try.
https://treeherder.mozilla.org/#/jobs?repo=try&revision=115469d212892173bb0b8da43d30eade46df28d0

Comment 26

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/mozilla-inbound/rev/57509edabda2
Iterate through defined ColorBuffers from mResolvedCompleteData instead of mColorBuffers. r=jgilbert

Comment 27

[Wes Kocher (:KWierso) (Not reading bugmail; email directly if needed)]

https://hg.mozilla.org/mozilla-central/rev/57509edabda2

Comment 28

[Gerry Chang [:gchang]]

Hi :jgilbert,
Is this worth uplifting to Beta51?

Comment 29

[Kelsey Gilbert [:jgilbert]]

Comment on attachment 8815601 [details] [diff] [review]
Iterate through defined ColorBuffers from mResolvedCompleteData instead of mColorDrawBuffers.

Approval Request Comment
[Feature/Bug causing the regression]: webgl2 crash
[User impact if declined]:
[Is this code covered by automated tests?]:
[Has the fix been verified in Nightly?]:
[Needs manual test from QE? If yes, steps to reproduce]: 
[List of other uplifts needed for the feature/fix]:
[Is the change risky?]:
[Why is the change risky/not risky?]:
[String changes made/needed]:

Comment 30

[Gerry Chang [:gchang]]

Comment on attachment 8815601 [details] [diff] [review]
Iterate through defined ColorBuffers from mResolvedCompleteData instead of mColorDrawBuffers.

WebGL2 related patch. Beta51+ & Aurora52+. Should be in 51 beta 8.

Comment 31

[Carsten Book [:Tomcat]]

https://hg.mozilla.org/releases/mozilla-aurora/rev/d0afc1865fcc

Comment 32

[Carsten Book [:Tomcat]]

https://hg.mozilla.org/releases/mozilla-beta/rev/6d12cc5b5087