Closed Bug 1307543 Opened 8 years ago Closed 8 years ago

Regression in D3D11DeviceContext crashes with 50.0b

Categories

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

50 Branch
defect

Tracking

()

RESOLVED FIXED
Tracking Status
firefox50 + fixed
firefox51 - fixed
firefox52 --- fixed

People

(Reporter: marco, Unassigned)

References

Details

(Keywords: crash, Whiteboard: [gfx-noted])

We have a regression in 50.0b3 with a few different signatures:
https://crash-stats.mozilla.com/search/?signature=~UpdateSubresource&proto_signature=~UpdateImage&product=Firefox&version=50.0b&process_type=browser&_sort=-date&_facets=signature&_facets=user_comments&_facets=adapter_vendor_id&_facets=platform_pretty_version&_columns=date&_columns=signature#facet-signature

Some might be OOMs (bug 1282562 and bug 1282564).

There's a large chunk of the crashes which have 'mozilla::layers::ImageClientSingle::UpdateImage(mozilla::layers::ImageContainer*, unsigned int)' in the stack trace:
https://crash-stats.mozilla.com/search/?signature=~UpdateSubresource&proto_signature=~UpdateImage&product=Firefox&version=50.0b&process_type=browser&_sort=-date&_facets=signature&_facets=user_comments&_facets=adapter_vendor_id&_facets=platform_pretty_version&_columns=date&_columns=signature#facet-signature

They affect 49 as well, but looks like the volume is higher on 50.

Correlations for signatures containing 'UpdateSubresource':
(58.83% in signature vs 10.43% overall) reason="EXCEPTION_ACCESS_VIOLATION_WRITE"
(96.85% in signature vs 48.62% overall) "D3D11 Layers+" in app_notes="True"
(96.85% in signature vs 49.95% overall) "D3D11 Layers?" in app_notes="True"
(71.03% in signature vs 26.28% overall) "DXVA2D3D9?" in app_notes="True"
(71.82% in signature vs 35.73% overall) "D2D1.1+" in app_notes="True"
(71.82% in signature vs 35.74% overall) "DWrite+" in app_notes="True"
(71.82% in signature vs 35.74% overall) "DWrite?" in app_notes="True"
(28.73% in signature vs 57.99% overall) "D2D1.1-" in app_notes="True"
(29.96% in signature vs 01.32% overall) address="0x1"
(63.70% in signature vs 37.44% overall) os_arch="amd64"
(47.71% in signature vs 21.48% overall) "DXVA2D3D9+" in app_notes="True"
(40.63% in signature vs 15.85% overall) adapter_vendor_id="0x10de"
(04.48% in signature vs 27.71% overall) dom_ipc_enabled="1"
(40.63% in signature vs 20.03% overall) "EGL+" in app_notes="True"
(40.63% in signature vs 20.33% overall) "WebGL+" in app_notes="True"
(40.63% in signature vs 20.87% overall) "GL Context+" in app_notes="True"
(40.63% in signature vs 20.87% overall) "GL Context?" in app_notes="True"
(40.63% in signature vs 20.93% overall) "EGL?" in app_notes="True"
(40.63% in signature vs 21.33% overall) "WebGL?" in app_notes="True"
(24.45% in signature vs 05.27% overall) "DXVA2D3D9-" in app_notes="True"
(27.89% in signature vs 09.45% overall) address="0x0"
(05.31% in signature vs 23.47% overall) url=""
(21.84% in signature vs 04.15% overall) GFX_ERROR "DXVA2D3D9 video decoding is disabled due to a previous crash."="True"
(36.15% in signature vs 53.11% overall) os_arch="x86"
(29.41% in signature vs 14.35% overall) platform_pretty_version="Windows 10"

(21.84% in signature vs 04.09% overall) GFX_ERROR "DXVA2D3D9 video decoding is disabled due to a previous crash."="True" ∧ "DXVA2D3D9-" in app_notes="True"
(29.32% in signature vs 12.37% overall) platform_pretty_version="Windows 10" ∧ "D2D1.1+" in app_notes="True"

There might be different crashes hidden in these signatures. More investigation
is needed to see what buckets became more frequent between 49 and 50.
Summary: Regression in D3D11DeviceContext crashes with 50.0b3 → Regression in D3D11DeviceContext crashes with 50.0b
The patch from bug 1291084 comment 60 might fix this bug.
beta 5 seems to have largely addressed the intel gpu-related bucket of this crash.
Depends on: 1291084
Severity: normal → major
Keywords: crash
Priority: -- → P3
Whiteboard: [gfx-noted]
See Also: → 1310600
[Tracking Requested - why for this release]: The crash rates on 50 seem acceptable at the moment. Let's try to do something in 51 for this crash signature.
This was probably fixed by bug 1308418 too, as bug 1310600.

The signatures are still there, but they are not looking worse in 50 compared to 49.
Status: NEW → RESOLVED
Closed: 8 years ago
Resolution: --- → FIXED
Looks like this was fixed or mostly mitigated by the bugs marco mentions in comment 4.
You need to log in before you can comment on or make changes to this bug.