Firefox segfaults when opening many tabs at once if memory-hotplug is enabled in domU about qubes-issues HOT 17 OPEN

No, disabling hotplug also solves the issue with low init memory settings, the system seems fully stable with the feature turned off.

I just didn't know if memory hotplug and memory balancing were doing something similar, turning off memory balancing and/or increasing the init memory also seems to improve stability, when running your test.

from qubes-issues.

Ouch.

Can you provide a way to reproduce this with a completely fresh AppVM?

from qubes-issues.

I have the same issue, it stated happening recently maybe a week ago, I'm having multiple FF crashes daily.

Isolated Web Co[1205]: segfault at 1a39be38a0d8 ip 00001a39be38a0d8 sp 00007ffc3eea83b8 error 15 likely on CPU 2 (core 2, socket 0)
Code: fe ff 00 0d 3c be 39 1a 00 00 90 d0 59 1a 92 7b 00 00 c0 d1 59 1a 92 7b 00 00 e8 91 38 be 39 1a fe ff e8 91 38 be 39 1a fe ff <c0> 6f 3c be 39 1a 00 00 90 d0 59 1a 92 7b 00 00 c0 d1 59 1a 92 7b

It happens with a single tab open, when streaming video or using JS heavy sites, and it seems to happen randomly.

from qubes-issues.

@renehoj Are you using the Fedora Firefox package? I suspect this is a Fedora packaging bug.

from qubes-issues.

No, I'm using Debian 12 minimal with Firefox-ESR. I tried giving my browser qubes 6GB memory, it didn't stop the crashes.

from qubes-issues.

Ouch.

Can you provide a way to reproduce this with a completely fresh AppVM?

I updated the steps to reproduce section.

from qubes-issues.

I have the same issue, it stated happening recently maybe a week ago, I'm having multiple FF crashes daily.

Isolated Web Co[1205]: segfault at 1a39be38a0d8 ip 00001a39be38a0d8 sp 00007ffc3eea83b8 error 15 likely on CPU 2 (core 2, socket 0)
Code: fe ff 00 0d 3c be 39 1a 00 00 90 d0 59 1a 92 7b 00 00 c0 d1 59 1a 92 7b 00 00 e8 91 38 be 39 1a fe ff e8 91 38 be 39 1a fe ff <c0> 6f 3c be 39 1a 00 00 90 d0 59 1a 92 7b 00 00 c0 d1 59 1a 92 7b

It happens with a single tab open, when streaming video or using JS heavy sites, and it seems to happen randomly.

I too noticed it at first with random crashes, then I recreated Firefox profiles, tried older versions, Flatpak's, suspected hardware failure, but what I finally ended up with was memory hotplug.

Disabling the memory hotplug fixes the problem like a charm, and I can reproduce it on another laptop with R4.2 installed, which is why I'm reporting the problem.

from qubes-issues.

Could this be due to a memory allocation failure?

from qubes-issues.

@noskb Is disabling hotplug the same as memory balancing?

Your test pass on my system with 8 GB initial memory, and balancing enabled, but fails with low values like 800 MB.

from qubes-issues.

@renehoj Does that mean that even with memory hotplug feature disabled, it still fails if the init memory value is low?

from qubes-issues.

I'm still having crashes, even with qvm-features memory-hotplug ''

During the weekend, I've had Firefox fully crash 3 times, not just a single tab.

It doesn't leave any information in the logs except for

Feb 25 09:49:08 browser-streaming qubes.StartApp+firefox-esr-dom0[882]: ExceptionHandler::SendContinueSignalToChild sent continue signal to child
Feb 25 09:49:08 browser-streaming qubes.StartApp+firefox-esr-dom0[882]: ExceptionHandler::GenerateDump cloned child 2312
Feb 25 09:49:08 browser-streaming qubes.StartApp+firefox-esr-dom0[882]: ExceptionHandler::WaitForContinueSignal waiting for continue signal...

from qubes-issues.

I'm still having crashes, even with qvm-features memory-hotplug ''

During the weekend, I've had Firefox fully crash 3 times, not just a single tab.

It doesn't leave any information in the logs except for

Feb 25 09:49:08 browser-streaming qubes.StartApp+firefox-esr-dom0[882]: ExceptionHandler::SendContinueSignalToChild sent continue signal to child
Feb 25 09:49:08 browser-streaming qubes.StartApp+firefox-esr-dom0[882]: ExceptionHandler::GenerateDump cloned child 2312
Feb 25 09:49:08 browser-streaming qubes.StartApp+firefox-esr-dom0[882]: ExceptionHandler::WaitForContinueSignal waiting for continue signal...

Even with memory balancing disabled and allocating memory statically to AppVM, does firefox still crash during normal use? If so, the most likely cause is a firefox problem or hardware failure.

from qubes-issues.

I only had memory-hotplug disabled, now I'm trying with memory balancing disabled as well.

My guess is that it started after an update this month, I didn't use to have any issues with Firefox, and suddenly it becomes noticeably unstable. It is a problem specifically with Firefox, no other application is crashing, but it could have started after updates to the Linux kernel or Xen.

from qubes-issues.

Disabling both memory-hotplug and memory balancing didn't stop the crashes.

I ended up downgrading the kernel in my browser qubes to 6.6.2-1, and now the crashes seem to have stopped.

from qubes-issues.

@renehoj Okay, so a kernel regression.

Can you (in a test standalone VM) try doing a kernel bisection to see which upstream commit broke things?

from qubes-issues.

@DemiMarie I spoke too soon, I just had libxul.so crash again.

Changing the kernel, just like disabling memory_hotplug, will allow the browser to pass noskb's test, but it doesn't stop the crashes.

from qubes-issues.

@renehoj Ouch.

The usual advice for this kind of problem is “record an rr trace” but that:

1. Is insecure (it breaks the Firefox sandbox), so I don’t recommend it outside of a disposable VM, if at all.
2. Does not work on Xen (in any VM) because vPMU is unsupported.

from qubes-issues.