Closed Bug 1898164 Opened 6 months ago Closed 4 months ago

Display problems with Beta 127 on a Windows ARM PC (Adreno 685 GPU)

Categories

(Core :: Graphics, defect)

Firefox 127
defect

Tracking

()

RESOLVED DUPLICATE of bug 1897444

People

(Reporter: m.ernst, Unassigned)

References

(Blocks 1 open bug)

Details

User Agent: Mozilla/5.0 (Windows NT 10.0; Win64; x64; rv:127.0) Gecko/20100101 Firefox/127.0

Steps to reproduce:

Open Firefox Beta 127

Actual results:

I have display problems with Beta 127 (b4) on a Windows ARM PC (Adreno 685 GPU). White FF itself and websites: white areas, text not readable.
Version 126 and before runs fine.
These errors do !not! occur when hardware acceleration is switched off!

The Bugbug bot thinks this bug should belong to the 'Core::Graphics' component, and is moving the bug to that component. Please correct in case you think the bot is wrong.

Component: Untriaged → Graphics
Product: Firefox → Core

:gw, can you comment to the bug?

Flags: needinfo?(gwatson)

Could you supply the content of your about:support. Jamie, do we have any known issues with that GPU on android devices?

Flags: needinfo?(m.ernst)
Flags: needinfo?(jnicol)
Flags: needinfo?(gwatson)
See Also: → 1895163

Nope can't see anything for that GPU or similar. One extension we use on Adreno which is a bit niche is QCOM_tiled_rendering. But the about:support in bug 1895163 (for an Adreno 680) doesn't indicate that is advertised as supported, so I doubt it's that.

@m.ernst, one thing you could try is setting the pref gfx.webrender.panic-on-gl-error to true, and see if firefox crashes. (and if so, submit the report and send us the link to it here.

This might make firefox permanently crash. So it would be best to create a user.js file in your profile directory (which you can find by going to about:support. Create a file in that directory called user.js and add the line user_pref("gfx.webrender.panic-on-gl-error", true); to it. Then open firefox, browse around like you would when the rendering is broken, and if it crashes submit the crash report. Then delete the user.js file (or remove the line from it), reopen firefox, go to about:crashes and see if there is a crash report there. If so, give us the link.

Sorry that's a bit complicated, let me know if you need any help with that!

Flags: needinfo?(jnicol)

I updated to the latest Adreno 680 GPU driver -> no change.
I tried the setting "gfx.webrender.panic-on-gl-error" -> no crash / no change.
And FF 126 runs well!

Flags: needinfo?(m.ernst)

Thank you! It's really useful to know that version 126 works. We have a tool called mozregression that can find out which specific change to the code caused the issue. Unfortunately I'm not actually sure whether it will work on an arm windows machine, but it would be really useful if it does, so if you're willing to try it that would be great!

You'll need to install python, if you don't have it already. On intel windows at least, opening the command prompt from the start menu, then typing python will prompt to install python. Once that's installed, type pip install mozregression. And once that's installed, type mozregression --good 126 --bad 127.

If that works, it will download a run a series of versions of firefox. For each one you type whether it was "good" or "bad", and eventually it will give you a link to the changes that caused the bug. Let me know if you need any help with that! Thanks!

Flags: needinfo?(m.ernst)

(In reply to Jamie Nicol [:jnicol] from comment #6)
I wil try it...hopefully your mozregression will work on an Arm-System...

Flags: needinfo?(m.ernst)

(In reply to Jamie Nicol [:jnicol] from comment #6)

You'll need to install python, if you don't have it already. On intel windows at least, opening the command prompt from the start menu, then typing python will prompt to install python. Once that's installed, type pip install mozregression. And once that's installed, type mozregression --good 126 --bad 127.

I have installed mozregression - how / where do I start the command (mozregression --good 126 --bad 127)? And where do I see the result?

How did you install mozregression? If from the command prompt using the pip install mozregression command I wrote above, then you should be able to paste mozregression --good 126 --bad 127 in to the same command prompt and press enter. (I think you should be able to open the command prompt by typing "command" or "cmd" in to the start menu and clicking it, sorry I don't have a windows machine right now to verify, nor an arm one at all)

If you installed it by downloading from github as linked to on the mozregression website, then it should make a start menu entry. So type mozregression in to the start menu and click it. That will open the mozregression GUI. I think this has less chance of working on ARM, though. But if it does work, when the mozregression window opens you click the scissor icon to "start a new bisection". I can't remember the order the options appear in but when you get prompted for a good and bad version you need to change the dropdown from "date" to "release" and enter 126 and 127 respectively

I first installed Python from the MS Store. No idea if that was necessary...
Then I ran "pip install mozregression" from the Windows command prompt (CMD) - mozregression was installed.
Correct so far?
Where do I now execute the command "mozregression --good 126 --bad 127"? It does not work in the Windows command prompt.

That all sounds correct so far! What does it say when you try to run it on the command prompt?

(In reply to Jamie Nicol [:jnicol] from comment #11)

That all sounds correct so far! What does it say when you try to run it on the command prompt?

windows (in cmd) does not recognise mozregression...

Ah I don’t think windows configures the PATH environment variable correctly. If you see the top answer in this link that might help fix it?

Unfortunately I cannot start mozregression. Windows / CMD does not recognise the command. Any ideas?

UPDATE: I didn't see Jamies last comment. Will try this...

I'm sorry, but that's too much for me... ;-)

  1. I have to go through all 127-dailies! That takes ages. Also, the error is already in the 1st 127-daily!
  2. mozregression tries to write results to my user directory - but has no authorisation (access denied). In other words, I have no detailed results.

Fact is: 126 works, from 127 (first daily) no more. 128 of course not either.

(In reply to m.ernst from comment #16)

I'm sorry, but that's too much for me... ;-)

  1. I have to go through all 127-dailies! That takes ages. Also, the error is already in the 1st 127-daily!
  2. mozregression tries to write results to my user directory - but has no authorisation (access denied). In other words, I have no detailed results.

Fact is: 126 works, from 127 (first daily) no more. 128 of course not either.

daily --> nighly!

It uses a binary search algorithm so thankfully it’s much much quicker than testing every nightly version! If the first 127 nightly has the bug then it was likely introduced in a 126 nightly and it should be able to find which one.

What is the specific error it gives you about authorisation? Can you share a screenshot? Thanks!

Blocks: gfx-triage
Severity: -- → S3

You should only have to do a couple before it says "you're done!".
127 is the maximum number, not the number that you'll have to go through! :)

If the command-line install of mozregression is not working well for you, that link also leads to a GUI tool that might work better for you. There is a video on that page that shows how to install and use the GUI tool. Please do give it a try as getting the regression will rapidly identify the code change that caused this issue.

Flags: needinfo?(m.ernst)

Let's move this out of triage if we don't here from the reporter next week.

See Also: → 1903048

Ok, close it! Unfortunately, I don't have the expertise or the time to test this error extensively...
Can't you do it yourself? Don't you have such devices available?
The fact is, since 127 it no longer runs on ARM PCs (Adreno 685 GPU)!
I guess I have to use Chrome now - at least on my MS-Surface.

Flags: needinfo?(m.ernst)

I think that this is the same as bug 1897444

Hi guys, Beta 129 is running again with Windows ARM PC (Adreno 685 GPU). You have changed something!
So far so good! It's a bit slow on page load, but no errors!

Status: UNCONFIRMED → RESOLVED
Closed: 4 months ago
Duplicate of bug: 1897444
Resolution: --- → DUPLICATE
You need to log in before you can comment on or make changes to this bug.