1980145 - Corrupted content sometimes on Zoom Web calls or Google docs (Linux AMD)

Status: RESOLVED INCOMPLETE

(Core :: Graphics, defect)

Description

[:gerard-majax]

Attachment: image.png

Attached is an example

I have parts of Google docs that are like this. Reloading the page make it go away (sometimes). Likely related are phantom data during Zoom Web calls, where e.g. there's a greenish+pinkish (typical of stride / pixels misreading) shadow of one of the callers that gets printed over other's thumbnails.

This started when I upgraded my ThinkPad P14s Gen2 AMD (Ryzen 5850U) to ThinkPad P14s Gen6 AMD (Ryzen 9 HX 370). The OS is exactly the same (Ubuntu 25.04 uptodate).

Comment 1

[:gerard-majax]

Attachment: about:support

Comment 2

[:gerard-majax]

I've ran a snap stable in parallel since yesterday: no bug after hours on google docs, zoom web is fine so far. I'll have to verify with non snap stable, so we can know if it's firefox or mesa

Comment 3

[:gerard-majax]

Still happening with newer kernel 6.16. How can we debug? I could not reproduce easily outside of long-running sessions of nightly

Comment 4

[Nicolas Silva [:nical]]

I had a look at a screenshot of the glitch with Ashley. Here are a few notes:

it looks like a YUV -> RGB conversion shader is being fed the wrong planes and sometimes the planes are mirrored.
The zoom call has multiple video feeds. one of the feeds is a horizontally mirrored version of another. All of them look like they have they don't have luminance information. One of the feeds has superposition of two videos (probably sampling planes from two different videos).

The rest of the page does not look like it is affected in this particular case so it may be an issue with the interaction between canvas and WR (or between the video and canvas, but that would not explain the gdocs issues).

Good next steps would include checking what packaging configurations the issue appears with (making sure that important prefs are the same in these configurations). :gerard-majax already noted that the bug happens with our nightly builds, but probably not in stable snap builds.
Quoting him:

mesa on snap is 24.xx while on the system it's 25.xx

It would also be useful to test with/without prefs that might affect this:

• widget.dmabuf-webgl.enabled
• webgl.threadsafe-gl.force-disabled

The attached about:supports shows DMABUF_SURFACE_EXPORT being disabled because of broken drivers. I wonder if we are using dmabuf in some other way here and it's also buggy at the driver level.

:gerard-majax also noted that the issue only happens after a while.

Comment 5

[:gerard-majax]

After a while running with widget.dmabuf-webgl.enabled=false, I still cannot see corruption. Could there be a regression on Mesa side here, between 24.xx (as in Ubuntu 24.04 LTS) and 25.xx (as in current Ubuntu 25.04 uptodate)

Comment 6

[:gerard-majax]

(In reply to :gerard-majax from comment #5)

After a while running with widget.dmabuf-webgl.enabled=false, I still cannot see corruption. Could there be a regression on Mesa side here, between 24.xx (as in Ubuntu 24.04 LTS) and 25.xx (as in current Ubuntu 25.04 uptodate)

I can confirm Zoom Web call is fine with this pref set to false.

Comment 7

[BugBot [:suhaib / :marco/ :calixte]]

The severity field is not set for this bug.
:bhood, could you have a look please?

For more information, please visit BugBot documentation.

Comment 8

[:gerard-majax]

(In reply to :gerard-majax from comment #6)

(In reply to :gerard-majax from comment #5)

After a while running with widget.dmabuf-webgl.enabled=false, I still cannot see corruption. Could there be a regression on Mesa side here, between 24.xx (as in Ubuntu 24.04 LTS) and 25.xx (as in current Ubuntu 25.04 uptodate)

I can confirm Zoom Web call is fine with this pref set to false.

Just upgraded to Ubuntu 25.10 and this is still happening. What can we do further?

Comment 9

[:gerard-majax]

Attachment: dmabuf.log.zip

Reproduced on a zoom call, around 1500~1501 Paris time. Logs are timestamped on UTC so it's around 1400-1401.

Comment 10

[:gerard-majax]

Martin, anything hinting from the logs?

Comment 11

[Martin Stránský [:stransky] (ni? me)]

The logs look clear.

2025-11-03 14:01:03.072583 UTC - [Parent 1829475: Renderer]: D/Dmabuf [7256c5df7660]: DMABufSurfaceRGBA::CreateGBM() UID 1830987 size 1753 x 1089 format 0x34325241 modifiers 1
2025-11-03 14:01:03.072584 UTC - [Parent 1829475: Renderer]: D/Dmabuf [7256c5df7660]:     Creating with modifiers
2025-11-03 14:01:03.072823 UTC - [Parent 1829475: Renderer]: D/Dmabuf [7256c5df7660]:     Planes count 2
2025-11-03 14:01:03.072845 UTC - [Parent 1829475: Renderer]: D/Dmabuf [7256c5df7660]:     Success

Looks like we're creating two-plane dmabuf RGBA surface here (surface width 1753 doesn't look very rendering friendly - do we need aligned buffers here?).

Can you try to use nightly and set widget.dmabuf-export.force-enabled to true? (that pref is nightly only).

Comment 12

[:gerard-majax]

Right, reverted widget.dmabuf-webgl.enabled to true and forced widget.dmabuf-export.force-enabled to true. Let's see.

Comment 13

[:gerard-majax]

(In reply to :gerard-majax from comment #12)

Right, reverted widget.dmabuf-webgl.enabled to true and forced widget.dmabuf-export.force-enabled to true. Let's see.

so I have not reproduced corruption for now, but google maps for example flashes tiles when moving the map and shows black content when stopping

Comment 14

[:gerard-majax]

(In reply to :gerard-majax from comment #13)

(In reply to :gerard-majax from comment #12)

Right, reverted widget.dmabuf-webgl.enabled to true and forced widget.dmabuf-export.force-enabled to true. Let's see.

so I have not reproduced corruption for now, but google maps for example flashes tiles when moving the map and shows black content when stopping

So corruption still happens, but it seems to be taking more time and be maybe bit less bad.

Comment 15

[Martin Stránský [:stransky] (ni? me)]

(In reply to :gerard-majax from comment #14)

(In reply to :gerard-majax from comment #13)

(In reply to :gerard-majax from comment #12)

Right, reverted widget.dmabuf-webgl.enabled to true and forced widget.dmabuf-export.force-enabled to true. Let's see.

so I have not reproduced corruption for now, but google maps for example flashes tiles when moving the map and shows black content when stopping

So corruption still happens, but it seems to be taking more time and be maybe bit less bad.

It's really hard to say and looks like gfx driver issue to me. dmabuf-export is known to be broken on AMD hardware. I recommend to file at MESA tracker, I don't think it's actionable on our side.

Comment 16

[James Graham [:jgraham]]

I've seen a few reports of this internally, so I'm triaging it as a webcompat issue affecting google docs (although I suspect in practice it could affect any site which is using canvas). I understand that it's not exactly a platform bug, but from a webcompat point of view that's the closest approximation. I'll also mark it as blocked since it sounds like we don't think it's possible to fix on our side.

Comment 17

[Glenn Watson [:gw]]

Stransky, should we block those dmabuf prefs on AMD hardware (or some subset of AMD hardware)?

Comment 18

[:gerard-majax]

I can confirm widget.dmabuf-export.force-enabled=true has helped a lot in the past weeks, so much I forgot about this bug ...

Comment 19

[Martin Stránský [:stransky] (ni? me)]

First of all please file at https://gitlab.freedesktop.org/mesa/mesa/-/issues

If widget.dmabuf-export.force-enabled=true works and widget.dmabuf-export.force-enabled=false is broken then we're having problem with GBM dmabuf backend. Please file that at MESA.

// FEATURE_DMABUF_SURFACE_EXPORT
// Disabled on all Mesa drivers due to various issue, among them:
// https://gitlab.freedesktop.org/mesa/mesa/-/issues/6666
// https://gitlab.freedesktop.org/mesa/mesa/-/issues/6796
// https://gitlab.freedesktop.org/mesa/mesa/-/issues/6688
// https://gitlab.freedesktop.org/mesa/mesa/-/issues/6988

We disabled FEATURE_DMABUF_SURFACE_EXPORT on MESA dues to various issues - maybe it's fixed already?

Comment 20

[Martin Stránský [:stransky] (ni? me)]

dri2_export_dma_buf_image_query_mesa / eglExportDMABUFImage* has been historically broken on AMD hardware so we use GBM there. Not sure what's the best way how to create dmabuf based textures/framebuffers to render to.

Comment 21

[Brad Werth [:bradwerth]]

Badly need a testcase so we can provide a coherent bug note to Mesa.

Comment 22

[Camelia Badau [:cbadau], Desktop Test Engineering]

Attachment: aboutSupport.txt

Hello,
I tried to reproduce the issue on Firefox Nightly 149.0a1 (2026-02-06) on Ubuntu 22.04.
Hardware / drivers
• Laptop default GPU: Mesa Intel(R) UHD Graphics 620
• Discrete GPU used for testing: Mesa AMD Radeon R5 M430
- Nightly was launched on the AMD GPU using DRI_PRIME=1

I was not able to reproduce the original issue described in the bug. I tested extensively on: Google Docs, Google Slides, Web Zoom call, Google Maps, Canva.
However, I encountered frequent flickering and rendering artifacts during testing:

• Google Docs: no issues related to this bug
• Google Slides: noticeable and frequent flickering
• Canva: slight flickering
• Google Maps: works correctly, no black content
• Web Zoom call: works fine

Additionally, the entire browser shows flickering/artifacts when switching between tabs or scrolling on various pages.
I tested with the following prefs:
• widget.dmabuf-webgl.enabled = true
• widget.dmabuf-webgl.enabled = false
• widget.dmabuf-export.force-enabled = true + widget.dmabuf-webgl.enabled = true
But all configurations showed the same general flickering behavior.

You can see here the video with flickers/artifacts that I have encountered: https://drive.google.com/file/d/1HhIl__0hYSMPvHzPHHLjam7m30SlspA8/view?t=18. Also, I attached the about:support information.
Please let me know if there is anything else I can test or provide.

Comment 23

[:gerard-majax]

(In reply to Camelia Badau [:cbadau], Desktop Test Engineering from comment #22)

[...]

I tried to reproduce the issue on Firefox Nightly 149.0a1 (2026-02-06) on Ubuntu 22.04.

I cannot speak for the OS, but I can state that this started when I upgraded my laptop, so I suspect you may at least need the same version of Ubuntu as I was using (25.04 uptodate as of the moment then) with the same hardware (Ryzen 9 HX 370).

I have been running with the pref mentionneed earlier for some months and have not suffered anything. Maybe MESA fixed it? i'll need to revisit the preef.

Comment 24

[Ashley Hale [:ahale]]

Paul, would someone at AMD like to take a look at this? We're not 100% sure if this bug is in Mesa or DRM as either spot could lead to the result.

Comment 25

[Ashley Hale [:ahale]]

(In reply to Camelia Badau [:cbadau], Desktop Test Engineering from comment #22)

Created attachment 9543161 [details]
aboutSupport.txt

Hello,
I tried to reproduce the issue on Firefox Nightly 149.0a1 (2026-02-06) on Ubuntu 22.04.
Hardware / drivers
• Laptop default GPU: Mesa Intel(R) UHD Graphics 620
• Discrete GPU used for testing: Mesa AMD Radeon R5 M430
- Nightly was launched on the AMD GPU using DRI_PRIME=1

I will note that on laptops that have Intel integrated graphics typically the physical video ports (HDMI / Displayport) or USB-C ports are driven by the Intel integrated GPU and not the AMD dedicated GPU, so testing on this type of laptop can hit bugs with dmabuf or GBM (the two alternative approaches being discussed in this bug) in drivers for either of these GPUs, good to test this configuration, may not be typical but I'm glad you have that coverage.

I was not able to reproduce the original issue described in the bug. I tested extensively on: Google Docs, Google Slides, Web Zoom call, Google Maps, Canva.
However, I encountered frequent flickering and rendering artifacts during testing:

• Google Docs: no issues related to this bug
• Google Slides: noticeable and frequent flickering
• Canva: slight flickering
• Google Maps: works correctly, no black content
• Web Zoom call: works fine

Additionally, the entire browser shows flickering/artifacts when switching between tabs or scrolling on various pages.
I tested with the following prefs:
• widget.dmabuf-webgl.enabled = true
• widget.dmabuf-webgl.enabled = false
• widget.dmabuf-export.force-enabled = true + widget.dmabuf-webgl.enabled = true
But all configurations showed the same general flickering behavior.

You can see here the video with flickers/artifacts that I have encountered: https://drive.google.com/file/d/1HhIl__0hYSMPvHzPHHLjam7m30SlspA8/view?t=18. Also, I attached the about:support information.
Please let me know if there is anything else I can test or provide.

Your video illustrates exactly the kind of bugs we'd expect when this functionality malfunctions, does that occur with default prefs on that laptop? Does it occur with DRI_PRIME=1 only or also without that option? I ask because if the default prefs behave that way on Intel UHD Graphics 620 that is another similar bug we would need to look into as well.

Comment 26

[Camelia Badau [:cbadau], Desktop Test Engineering]

@:gerard-majax - Unfortunately, our QA team does not have a laptop with hardware: Ryzen 9 HX 370, so we are unable to test on that configuration.

@Ashley Hale - I tested on latest Nightly 149.0a1 with default preferences on the same laptop, without DRI_PRIME=1 enabled.

Graphics: Intel — Mesa Intel(R) UHD Graphics 620 (KBL GT2)
- I did not observe any flickering or rendering artifacts; everything looks good. Testing covered the same use cases as before: Google Docs, Google Slides, Web Zoom calls, Google Maps, and Canva.

Based on these results, the issue appears to reproduce only with DRI_PRIME=1 option.

Comment 27

[Martin Stránský [:stransky] (ni? me)]

I have this hardware available so will test this week on google meet. Fedora 44/ Gnome 50. Ryzen PRO 9 HX.

Comment 28

[Martin Stránský [:stransky] (ni? me)]

Tested yesterday on Google Meet for long time and I haven't see any corruption. Fedora 44 / Firefox 147.0.3 / Ryzen PRO 9 HX.

Comment 29

[:gerard-majax]

I have reverted the prefs to their default value and will monitor

Comment 30

[Ashley Hale [:ahale]]

Possible theory - is this a mismatch between the Mesa inside a snap package and the host DRM? (i.e. outdated Snap base package)

We've seen that kind of bug before on AMD in particular.

Comment 31

[Bob Hood [:bhood]]

Ashley will attempt to file upstream with Mesa.

Comment 32

[:gerard-majax]

(In reply to Ashley Hale [:ahale] from comment #30)

Possible theory - is this a mismatch between the Mesa inside a snap package and the host DRM? (i.e. outdated Snap base package)

We've seen that kind of bug before on AMD in particular.

This may explain some but I was hitting the issue on nightly non snap

Comment 33

[:gerard-majax]

I dont seem to be able to reproduce the issue anymore. Either a kernel fix or a MESA fix was included during 25.10 lifetime, I'm currently at 6.17.0-8-generic kernel-wise and 25.2.8-0ubuntu0.25.10.1 for mesa

Comment 34

[Brad Werth [:bradwerth]]

(In reply to :gerard-majax from comment #33)

I dont seem to be able to reproduce the issue anymore.
Resovling INCOMPLETE for now. Please re-open if it re-occurs.