On 2025-08-15 at 21:20, Van Snyder wrote: > On Fri, 2025-08-15 at 19:50 -0400, The Wanderer wrote: > >> On 2025-08-15 at 19:39, Van Snyder wrote: >>> When the system crashes, it is well and truly crashed. The mouse >>> cursor doesn't move. The keyboard doesn't do anything. Tapping >>> the power button doesn't do anything. The graphs in GKrellM >>> aren't moving,... And I can't log in from another computer using >>> ssh. So it's hard to return to the console and ask what top is >>> telling me. >> >> The ssh bit is a good detail to know. Does *ping* still get >> responses when the computer is in the frozen state? > > No response to ping. So it's a lockup all the way down the networking stack, probably into the kernel. That basically confirms that it's either hardware or drivers, or *maybe* firmware but that seems a bit less likely. Firefox cannot plausibly be the actual root cause of the issue, it's just the triggering element. I think the "disable GPU acceleration" thing suggested elsewhere would be worth trying. If that doesn't help, I'm not actually sure what would be best to try next. >> This is looking like one of three things: a hardware issue (which >> as you note, seems less likely given that it's happening on two >> different computers), a drivers issue (most likely GPU drivers, >> given what Firefox would be likely to interact with), or a >> *firmware* issue (potentially even CPU microcode). > > That was my first guess so I replaced my NVidia K2200 with an AMD > Radeon RX580. That doesn't entirely rule out the possibility that it's a GPU issue, but does make it a lot less likely. It would be interesting to see what happens if you try starting a bare-bones graphical environment - with a minimalist window manager (not whichever one you normally use), or even with the system told to launch Firefox *instead of* the window manager (I believe that's possible with X, but no idea about Wayland) - and see if the issue reproduces with *nothing* else but Firefox running. All of this is, however, "try to get more information" stabbing around in the dark; I don't hold out a lot of hope of any of it actually making a difference. I know Firefox has some internal debugging/etc. pages that can be accessed via the about: namespace (or it did in previous versions, and I'd be surprised if they had all been removed). I don't know of any offhand that would be likely to be helpful for this, but if they exist in the Firefox you're running, they should be listed under 'about:about'; it might be worth looking at that list of available pages and seeing whether any of them look potentially relevant enough to bother checking whether they have any records of events that occurred in a previous session, such as the sessions where the freezes occurred. >> The canonical way to try to get more information would probably to >> be to connect a serial console, have a relevant log (or other >> data-source stream) feeding out over it as info comes in to that >> source, and see what happens - or, at minimum, what *has just* >> happened - in that data stream when the freeze event occurs. There >> are *multiple* parts of that which are easier said than done, >> however, and I would barely know where to get started trying to >> find out how to actually do it. > > Not having any serial ports, I could put a log onto a USB disk. Is > there a way to ask for flushing after every write so the interesting > stuff doesn't die a pointless death in the OS's disk cache? Serial console can, I believe, nowadays be done over USB - though I have no information as to exactly how. If I needed to find out, I'd go looking for documentation from the Linux kernel. (netconsole is also a thing, but since we know networking is affected by the lockup, I don't know whether that would be any more likely to get useful information out.) Adding filesystem write into the process would add enough moving pieces to the process that the odds of getting the interesting information out before the freeze stops things would be lower than I'd like. I believe there *is* a way to specify always-sync-immediately, possibly as a filesystem-create-time option and possibly as a mount option, but I don't remember offhand what it is. -- The Wanderer The reasonable man adapts himself to the world; the unreasonable one persists in trying to adapt the world to himself. Therefore all progress depends on the unreasonable man. -- George Bernard Shaw
Attachment:
signature.asc
Description: OpenPGP digital signature