Bug#584314: base: System freezes at random time after Resume from Suspend (Regression)
notfound 584314 linux-2.6/2.6.32-32
notfixed 584314 linux-2.6/2.6.38-3
found 584314 linux-2.6/2.6.32-30
fixed 584314 linux-2.6/2.6.38-5
quit
Andreas Berger wrote:
> in linux-image-2.6.32-5-686, version 2.6.32-30, the bug was still there,
> in linux-image-2.6.38-2-686, version 2.6.38-5, the bug was no longer there,
>
> in between the two, i don't know, but if it helps, i can narrow it down as
> soon as i get home to a spare hard drive.
Sure, it would help to narrow the search for the fix (but see below to
save some time).
> On Thursday, July 28, 2011 04:19:39 Jonathan Nieder wrote:
>> - could you send a photo of the screen during the oops, so we can read
>> the backtrace?
>
> i typed it off the screen and included it in my previous mail here:
> http://bugs.debian.org/cgi-bin/bugreport.cgi?msg=22;bug=584314
>
> is that not what you mean?
Unfortunately what you typed doesn't include the call trace (or maybe
there was none). It does include the code, which when passed through
scripts/decodecode looks like this:
| kernel:[ 496.263433] Code: 04 01 00 00 00 66 83 7c 24 28 00 79 37 89 f5 31 db eb 2b ba 03 00 00 00 89 e8 e8 ee 73 fa ff b9 00 04 00 00 89 04 24 89 c7 31 c0 <f3> ab 8b 04 24 ba 03 00 00 00 43 83 c5 20 e8 20 72 fa ff 3b 5c
[...]
| 11: eb 2b jmp 0x3e
| 13: ba 03 00 00 00 mov $0x3,%edx
| 18: 89 e8 mov %ebp,%eax
| 1a: e8 ee 73 fa ff callq 0xfffffffffffa740d
| 1f: b9 00 04 00 00 mov $0x400,%ecx
| 24: 89 04 24 mov %eax,(%rsp)
| 27: 89 c7 mov %eax,%edi
| 29: 31 c0 xor %eax,%eax
| 2b:* f3 ab rep stos %eax,%es <-- trapping instruction:(%rdi)
| 2d: 8b 04 24 mov (%rsp),%eax
Building mm/page_alloc.s and comparing, we see that this is in
"clear_highpage"; the function call starting on line 13 is to
kmap_atomic and the trapping rep stos is memset(page, 0, PAGE_SIZE).
Unwinding a little: clear_highpage is called by prep_zero_page,
which is called by prep_new_page, which is called by buffered_rmqueue,
which is called by get_page_from_freelist for each potentially
free page.
I suspect memory corruption. Maybe v2.6.37-rc5~3^2 (PM / Hibernate:
Fix memory corruption related to swap, 2010-12-03) fixes it. Could
you test 2.6.37-rc5 and 2.6.37-rc4?
Thanks,
Jonathan
Reply to: