1920664 - Firefox for Windows ARM64 randomly hangs and causing crash in QCDX Display Driver

Status: UNCONFIRMED

(Core :: Graphics: WebRender, defect, P1)

Description

[zeroaurora233]

This is probably a dup of #1798465 but I an reopening an issue since the old one is closed.

Firefox randomly hangs the whole system while browsing, for around 10 seconds. Then pages recover without crashing. But looking into Windows event viewer, I found that the QCDX display driver stopped responding and recovered successfully.

Windows Error Reporting also collected the information. I will try to provide if needed (I'm figuring out how).

System spec: Xiaomi Boox S 12.4, Snapdragon 8cx Gen 2

Comment 1

[zeroaurora233]

typo: Xiaomi Book S 12.4 sorry

Comment 2

[Mayank Bansal]

In general, try to update your graphics drivers to the latest version.

can you type "about:support" in your firefox browsers address bar, and copy-paste its contents here in the following two scenarios :

1. Before you get the random freeze
2. After you get the random freeze

Comment 3

[zeroaurora233]

Driver is already the latest version. i.e. I couldn't find any updates in Windows Update.
I reproduced a freeze under some high load and can confirm driver crash in system event viewer. Posting support data as files.

Comment 4

[zeroaurora233]

Attachment: before-freeze.json

Comment 5

[zeroaurora233]

Attachment: after-freeze.json

Comment 6

[zeroaurora233]

found a interesting diff:

    "failures": [
      "[GFX1-]: Detect DeviceReset DeviceResetReason::HUNG DeviceResetDetectPlace::WR_POST_UPDATE in GPU process",
      "GP+[GFX1-]: Failed to make render context current during destroying.",
      "GP+[GFX1-]: Error in eglMakeCurrent: 0x300e",
      "GP+[GFX1-]: Error in eglDestroySurface: 0x300e"
    ],
    "indices": [
      0,
      1,
      2,
      3
    ],

Comment 7

[zeroaurora233]

Attachment: wer_dump.zip

I collected a copy of windows error report dump from Reliability Monitor (inside control panel), also uploading wishing it helps

Comment 8

[Mayank Bansal]

Reading Windows crash dumps - > gsvelto and yjuglaret

Comment 9

[Gabriele Svelto [:gsvelto]]

The dump file within attachment 9426761 [details] is a kernel crash dump file from what I can tell. We don't have tooling to analyze these, and a quick check with windbg yielded a short and not very interesting stack.

Comment 10

[Kelsey Gilbert [:jgilbert]]

ANGLE and WR are probably both under-tested on Windows+Qualcomm. (Maybe @gw knows more, ni'd)
I'm going to mark this S2, since I worry that people will just not use Firefox, and not try to tell us about it, running into this type of error.
Maybe it should be S3 since this isn't a super common config, but in that case it would be better so punt it to swgl, probably.
We should try very hard to Just Work.

Comment 11

[Kelsey Gilbert [:jgilbert]]

Reporter: Do you still have this problem in Firefox's Safe Mode?

Comment 12

[Glenn Watson [:gw]]

Could we get the output of about:support, we can check what driver version is in use - perhaps there is an update available?

Comment 13

[zeroaurora233]

(In reply to Kelsey Gilbert [:jgilbert] from comment #11)

Reporter: Do you still have this problem in Firefox's Safe Mode?

I stressed Firefox in Safe Mode and while there are stucks sometimes, there are no crashes. Software rendering here seems to work.

(In reply to Glenn Watson [:gw] from comment #12)

Could we get the output of about:support, we can check what driver version is in use - perhaps there is an update available?

See file before-freeze.json and after-freeze.json. These are export files from about:support.

Comment 14

[Kelsey Gilbert [:jgilbert]]

We are thinking that it would be safest to blocklist here until we're more confident in it.
@ahale volunteers to help with adding things to blocklist.

Comment 15

[Bob Hood [:bhood]]

(re)Assigning to Ashley, and setting to P1 for follow-up.

Comment 16

[Bob Hood [:bhood]]

Ashley, based on comment 14, how quickly could we update the blocklist for this?

Comment 17

[Bob Hood [:bhood]]

Setting this to stalled until we have the personnel/cycles to focus on it again.

Comment 18

[Peter Meinl]

FF ARM on Windows is still not stable playing YouTube videos.
FF v146
Windows 25H5 26200.7309
all available driver updates installed

I don't know if this is a general problem for ARM versions on Windows or specific to my Windows Dev Kit 2023 ARM Mini-PC.

No dumps etc. needed to repro. Simply watch some vids for a couple of minutes.

Comment 19

[Gabriele Svelto [:gsvelto]]

FYI I have the same issue on a Lenovo Yoga C630 (based on the Snapdragon 850). I get the exact same behavior as in comment 0. It feels like a problem in Qualcomm's drivers, but given they are not releasing updates for this particular device I doubt it will ever be fixed.

Comment 20

[Peter Meinl]

The driver may be the root cause. But other Browsers like Edge, can handle this problem.

Comment 21

[Peter Meinl]

@Gabriele
Does your affected Yoga also have the problem of YouTube audio and video increasingly getting out of sync while playing a video?

Comment 22

[Gabriele Svelto [:gsvelto]]

No, that doesn't happen to me. However when the graphics driver encounters an issue everything freezes - including the cursor.