1361519 - Crash in libovrrt32_1.dll@0x2eba9

Status: RESOLVED FIXED

(Core :: Graphics, defect, P3)

Description

[Marcia Knous [:marcia]]

This bug was filed from the Socorro interface and is 
report bp-a12eabc5-55db-4cb3-98cc-a91a50170502.
=============================================================

Seen while looking at nightly crash stats - looks as if it is related to WebVR: http://bit.ly/2qpQZFL

29 crashes/10 installs. Crashes on other branches, but not as many as nightly.

Comment 1

[Vincent Liu[:vliu]]

All crash reason were pointed to EXCEPTION_ACCESS_VIOLATION_READ. It is bad thing that the last back trace frame was stopped in driver(libovrrt32_1.dll) so I couldn't see anything. The last back trace frame in Gecko were all in [1].

[1]: http://searchfox.org/mozilla-central/rev/abe68d5dad139e376d5521ca1d4b7892e1e7f1ba/gfx/vr/gfxVROculus.cpp#824

In minidump, access violation reading location pointed to 0x00000001. The address of sourceD3D11 also pointed to 0x00000001 but I am not sure there was any connection.


Kip, Daosheng, do your have any idea for this issue? Really thanks.

Comment 2

[Daosheng Mu[:daoshengmu]]

The 0x00000001 address of sourceD3D11 looks very strange. It seems that the IPC actor is destroyed at somewhere. Probably, we should check (mIPCOpen) at https://dxr.mozilla.org/mozilla-central/rev/a748acbebbde373a88868dc02910fb2bc5e6a023/gfx/vr/ipc/VRLayerParent.cpp#51.

Moreover, I think it could be a little bit risky if we don't check aSource->GetShaderResourceView() is nullptr at http://searchfox.org/mozilla-central/rev/abe68d5dad139e376d5521ca1d4b7892e1e7f1ba/gfx/vr/gfxVROculus.cpp#765.

This is just my two cent. Kip, might have more thought about this.

Comment 3

[:kip (Kearwood Gilbert)]

It wouldn't hurt to have some additional validation for cases when DX11 resources fail to be allocated.  Hopefully this will reduce the occurrence of this crash.  I'd suggest we first implement the validation before looking for more complex causes.

Another possibility is that these are cases where the Oculus runtime is being updated while Firefox has an active VR presentation.  We no longer keep the Oculus runtime dll's loaded when presentations are not active; however, this is still possible during a normally active VR presentation.  Oculus has shared some details on these cases in the past; perhaps with more information we could find a correlation.

Comment 4

[:kip (Kearwood Gilbert)]

I am finalizing work on Bug 1362213, and will take a look at this next if it hasn't been picked up by anyone else.

Comment 5

[:kip (Kearwood Gilbert)]

Bug 1287944 added a lot of error handling for the cases where DX11 resources fail to be allocated.
Bug 1321275 fixed issues with IPC shutdown during VR presentation.

The combination of these has likely eliminated this crash.