Re: 2.6.12-rc6-mm1

[Andrew Morton <akpm@osdl.org>]

Andrew Morton <akpm@osdl.org> wrote:
>
> Dominik Karall <dominik.karall@gmx.net> wrote:
> >
> > On Friday 29 July 2005 23:27, Andrew Morton wrote:
> > > Dominik Karall <dominik.karall@gmx.net> wrote:
> > > > On Friday 29 July 2005 20:22, Andrew Morton wrote:
> > > > > Dominik Karall <dominik.karall@gmx.net> wrote:
> > > > > > On Friday 29 July 2005 06:54, Andrew Morton wrote:
> > > > > > > Dominik Karall <dominik.karall@gmx.net> wrote:
> > > > > > > > On Tuesday 07 June 2005 13:29, Andrew Morton wrote:
> > > > > > > > > ftp://ftp.kernel.org/pub/linux/kernel/people/akpm/patches/2.6/2
> > > > > > > > >.6.1 2-rc 6/2. 6.12-rc6-mm1/
> > > > > > > >
> > > > > > > > After looking in my dmesg output today, I saw following error
> > > > > > > > with 2.6.12-rc6-mm1, maybe it's usefull to you. I don't know when
> > > > > > > > it exactly happens, cause I never used mono last time, I just did
> > > > > > > > an emerge mono on my gentoo system, maybe this forced the
> > > > > > > > failure.
> > > > > > > >
> > > > > > > > note: mono[26736] exited with preempt_count 1
> > > > > > > > scheduling while atomic: mono/0x10000001/26736
> > > > > > > >
> > > > > > > > Call Trace:<ffffffff803e13ea>{schedule+122}
> > > > > > > > <ffffffff8013197b>{vprintk+635}
> > > > > > > > <ffffffff803e2738>{cond_resched+56}
> > > > > > > > <ffffffff80164de3>{unmap_vmas+1587}
> > > > > > > > <ffffffff8016a560>{exit_mmap+128} <ffffffff8012e7bf>{mmput+31}
> > > > > > > > <ffffffff80133466>{do_exit+438}
> > > > > > > > <ffffffff8013bf25>{__dequeue_signal+501}
> > > > > > > >        <ffffffff801340c8>{do_group_exit+280}
> > > > > > > > <ffffffff8013e147>{get_signal_to_deliver+1575}
> > > > > > > >        <ffffffff8010de92>{do_signal+162}
> > > > > > > > <ffffffff8012d1e0>{default_wake_function+0}
> > > > > > > >        <ffffffff8010e8e1>{sys_rt_sigreturn+577}
> > > > > > > > <ffffffff8010eb3f>{sysret_signal+28}
> > > > > > > >        <ffffffff8010ee27>{ptregscall_common+103}
> > > > > > >
> > > > > > > A couple of people reported this, but all seems to have gone quiet.
> > > > > > >  Is it fixed in later -mm's?   Is 2.6.13-rc4 running OK?
> > > > > > >
> > > > > > > Thanks.
> > > > > >
> > > > > > hi andrew!
> > > > > >
> > > > > > I'm sorry, but it's not fixed in current 2.6.13-rc3-mm3. I did an
> > > > > > emerge mono right now to test it, and I got this one:
> > > > > > Jul 29 15:26:37 [kernel] note: mono[11138] exited with preempt_count
> > > > > > 1 Jul 29 15:26:50 [kernel] file[14627]: segfault at 00002aaaab453000
> > > > > > rip 00002aaaaaf652cf rsp 00007fffffe43b50 error 4
> > > > > > Jul 29 15:26:50 [kernel] file[14633]: segfault at 00002aaaab453000
> > > > > > rip 00002aaaaaf652cf rsp 00007fffffcc87a0 error 4
> > > > > > Jul 29 15:26:51 [kernel] file[14669]: segfault at 00002aaaab453000
> > > > > > rip 00002aaaaaf652cf rsp 00007fffff905f80 error 4
> > > > > >
> > > > > > DEBUG_KERNEL/ PREEMPT/ SPINLOCK are enabled, but I didn't get more
> > > > > > info about the bug. Did I forget any debug option?
> > > > >
> > > > > Gee, I don't know how to find this one.  Do you know if the problem is
> > > > > specific to -mm?
> > > >
> > > > Tested with 2.6.13-rc4 and it seems to work. Didn't get any error.
> > >
> > > Great, thanks for that.
> > >
> > > > So it seems to be -mm related. Do you suspect any patch which could cause
> > > > the error?
> > >
> > > I wouldn't know, sorry.  Possible the scheduler patches, possibly an
> > > x86_64-specific patch.  Is the problem repeatable?  If so, a binary search
> > > would only take ten build-n-boots ;)
> > 
> > Yes, it is repeatable. I tested on lastest -mm about 4 times. Ok, I will try 
> > to find the right patch tomorrow, 10 build-n-boots would end up in morning ;)
> > 
> > btw, as the error occured in 2.6.12-rc6-mm1 too, it must be an old patch which 
> > wasn't merged to linus tree till now...hope there aren't a lot of them :)
> > 
> 
> Any progress on this?  It kinda measn that the whole of the -mm lineup is
> stuck until we can identify the offending patch.  We have a couple of weeks
> in which to do this but if you can identify the bad patch it'd help
> enormously, thanks.
> 

OK, Bartosz Taudul tells me that he's occasionally seeing this on stock
2.6.12 (thanks!).  So there's not a lot of point in doing the -mm bisection
search.

I think Ingo was planning on coming up with some infrastructure which would
allow us to debug this further.