Re: commit 1c29a32ce65f4cd0f1c causes Bad rss-counter state and firefox-esr crash in linux-next-20240613

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



* Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx> [240613 20:03]:
> On Fri, 14 Jun 2024 01:40:54 +0200 Bert Karwatzki <spasswolf@xxxxxx> wrote:
> 
> > Since linux-next-20240613 firefox-esr crashes after several minutes of browsing
> > giving the following error messages in dmesg:
> > [ T2343] BUG: Bad rss-counter state mm:00000000babe0c39 type:MM_ANONPAGES val:86
> > [ T4063] show_signal_msg: 16 callbacks suppressed
> > [ T4063] Isolated Web Co[4063]: segfault at 396d1686c000 ip 0000396d1686c000 sp
> > 00007ffd767b30a8 error 14 likely on CPU 7 (core 3, socket 0)
> > [ T4063] Code: Unable to access opcode bytes at 0x396d1686bfd6.
> > [ T4211] BUG: Bad rss-counter state mm:00000000cd9fc541 type:MM_ANONPAGES
> > val:817
> > [ T3798] BUG: Bad rss-counter state mm:00000000432d87c2 type:MM_ANONPAGES
> > val:181
> > [ T5548] BUG: Bad rss-counter state mm:00000000034aa27a type:MM_ANONPAGES
> > val:242
> > [ T3823] BUG: Bad rss-counter state mm:0000000099734197 type:MM_ANONPAGES
> > val:137
> > [    T1] BUG: Bad rss-counter state mm:000000005e5e2f2f type:MM_ANONPAGES val:28
> 
> Let's hope Linus doesn't read this.  Why are we nuking the entire
> planet just because some counter went wonky?

I think I know what's going on, and it's more than just the counters
being off here.  The counters are the symptom of what is happening.

> 
> > (these are the error messages of several crashes and the error seems to affect
> > other processes, too (T1))
> > 
> > The crash can be provoked to appear in ~1min by opening large numbers of tabs in
> > firefox-esr (by holding pressing ctrl+t for some time). With this I bisected the
> > error to commit "1c29a32ce65f mm/mmap: use split munmap calls for MAP_FIXED" and
> > reverting this commit in linux-next-20240613 fixes the issue for me.
> 
> Thanks, that must have taken a lot of time.

Yes, thank you for all that work and apologies in creating this
frustrating situation.

Andrew, please drop the set from your branch.  I need to write some more
tests, but I suspect I will need to do some work around the vma_merge()
function, which is never a fun endeavor.

Regards,
Liam




[Index of Archives]     [Linux Kernel]     [Linux USB Development]     [Yosemite News]     [Linux SCSI]

  Powered by Linux