kernel panic ubuntu 10.10

lexvictory · 12-27-2010, 08:01 PM

I'm not sure if this is really the right place for a kernel panic question, but since it only occurred after updating form 10.04->10.10 i figured this might be the place (I don't know if its a bug or a hardware problem)

Since upgrading, running ffmpeg seems to cause a kernel panic, usually with a message similar to "BUG: unable to handle kernel NULL pointer dereference at <address>" or "unable to handle kernel paging request" (the 2nd is from memory, I have a photo of the first)
This is really troubling as I can't transcode things i record off tv (using mythtv), also I didn't think a userland program should crash the kernel...

this is the ffmpeg line that has been causing either a segfault in ffmpeg, or the kernel to crash (whether i compile it and libx264 myself or use the repo's version)

Code:

ffmpeg -y -i /home/lex/mythout/mytranscode.m2v -i /home/lex/mythout/mytranscode.ac3 -acodec copy -vcodec libx264  -vpre normal  -crf 20 -b 4000k -deinterlace  -s hd720  -threads 0 "/path/to/out/file.mkv"

(also seems to happen on SD recordings - which have no resize options present)

What should I do? or What extra info should I supply?
The computer is used as a server; no desktop environment (the kdm upstart job file has been renamed)

jschiwal · 12-28-2010, 12:51 PM

You might try running an strace on ffmpeg. The last system call used may be the culprit.

Code:

strace /usr/bin/ffmpeg -y -i /home/lex/mythout/mytranscode.m2v -i /home/lex/mythout/mytranscode.ac3 \ 
  -acodec copy -vcodec libx264  -vpre normal  -crf 20 -b 4000k -deinterlace  -s hd720  -threads 0 "/path/to/out/file.mkv" 2>ffmpeg.strace

lexvictory · 12-28-2010, 07:43 PM

here are the last few lines: (standard def recording, crashing just ffmpeg rather than whole kernel like HD ones do)

Code:

clone(child_stack=0xb29d8484, flags=CLONE_VM|CLONE_FS|CLONE_FILES|CLONE_SIGHAND|CLONE_THREAD|CLONE_SYSVSEM|CLONE_SETTLS|CLONE_PARENT_SETTID|CLONE_CHILD_CLEARTID, parent_tidptr=0xb29d8bd8, {entry_number:6, base_addr:0xb29d8b70, limit:1048575, seg_32bit:1, contents:0, read_exec_only:0, limit_in_pages:1, seg_not_present:0, useable:1}, child_tidptr=0xb29d8bd8) = 19825
gettimeofday({1293586801, 84364}, NULL) = 0
select(1, [0], NULL, NULL, {0, 0})      = 0 (Timeout)
write(5, "\365\201\270\276\313\213\244\234\303\247N\301\354'\235j_\312\2\310@\33\216\266X\240F\337\204m\201\323"..., 32768) = 32768
gettimeofday({1293586801, 84898}, NULL) = 0
select(1, [0], NULL, NULL, {0, 0})      = 0 (Timeout)
gettimeofday({1293586801, 85115}, NULL) = 0
select(1, [0], NULL, NULL, {0, 0})      = 0 (Timeout)
read(3, "r\371\343\221[\1L\3\200(\30\374T5\16\7\266\315\26\371=\372\250\357\3561\230\1\201d<\32"..., 32768) = 32768
futex(0x9dfc368, FUTEX_CMP_REQUEUE_PRIVATE, 1, 2147483647, 0x9dfc34c, 15136) = 1
clone(child_stack=0xb1eb9484, flags=CLONE_VM|CLONE_FS|CLONE_FILES|CLONE_SIGHAND|CLONE_THREAD|CLONE_SYSVSEM|CLONE_SETTLS|CLONE_PARENT_SETTID|CLONE_CHILD_CLEARTID, parent_tidptr=0xb1eb9bd8, {entry_number:6, base_addr:0xb1eb9b70, limit:1048575, seg_32bit:1, contents:0, read_exec_only:0, limit_in_pages:1, seg_not_present:0, useable:1}, child_tidptr=0xb1eb9bd8) = 19826
futex(0xb139abd8, FUTEX_WAIT, 19824, NULL <unfinished ...>
+++ killed by SIGSEGV +++

jschiwal · 12-29-2010, 08:23 AM

The , NULL <unfinished ...> of the futex command caught my attention, but according to the manpage, for FUTEX_WAIT, that argument, and the next are ignored.

I'd try repeating the experiment and see if the code bails with the same command. If so, maybe you found a bug. I don't know it it would be in ffmpeg or the kernel. The futex command runs in user space if there is no contention, but some code runs in kernel space to handle contentions to the same resource.

lexvictory · 12-29-2010, 08:31 PM

after some further experimenting, i found that it doesnt crash on my x64 desktop, so i migrated to x64 on my server, but ffmpeg still crashes (well the error is actually in libx264), but so far no kernel panics, just "illegal instruction" or "segmentation fault"...
using gdb on x264 seems to reveal a different crash line every time...

Could this indicate a hardware issue?
Or an asm instruction that is somehow being used and isnt supported? (if so how do i get gdb or some other tool to tell me what instruction?)

jschiwal · 12-31-2010, 08:22 PM

You might want to start by testing your ram. Bad ram will cause intermittent errors.

lexvictory · 01-04-2011, 12:12 AM

well it appears to have been a RAM problem.. Tho the RAM stick isn't that old and its only since the 10.10 upgrade its been a problem..
Memtest didnt report any problems, but swapping out the stick with the newest from my desktop is letting x264 run properly so far...
Maybe its something to do with it being a 2gb stick instead of a 1gb :S