moving from misc@ to bugs@
On Sun, Jun 24, 2018 at 12:37:45PM +0200, Walter Alejandro Iglesias wrote:
> Hello,
>
> I had a kernel panic while reproducing a video with mpv.
>
> It's my first kernel panic with OpenBSD, so I didn't know how to use
> ddb(4). Since I'm running my http and smtp server in this machine I
> cannot entertain myself too much reproducing the panic to get more info.
> That's why I don't include the per cpu trace and other additonal info as
> explained in ddb.html, sorry! But, if you need it let me knonw and I'll
> try my best.
>
>
> Message automatically dumped:
> ===============================================================
> panic: mtx 0xffffffff81c86470: locking against myself
> Stopped at db_enter+0x12: popq %r11
> TID PID UID PRFLAGS PFLAGS CPU COMMAND
>
>
> 104021 96401 1000 0x3 0x4000000 2 mpv
>
>
> *402610 50624 1000 0x32 0 0K Xorg
>
>
> db_enter() at db_enter+0x12
> panic() at panic+0x138
> __mtx_enter_try(53b9235709d40154) at __mtx_enter_try+0xb5
> _mtx_enter(ffffffff81cf3e60,ffffffff81a5d6a2,0) at _mtx_enter+0x5a
> printf(c9ef1007dec621e0) at printf+0x70
> witness_checkorder(2e4447d1b3cbb9af,ffffffff81c2ac7c,32a,0,ffffffff81da6d00)
> at
> witness_checkorder+0x943
> ___mp_lock(ffff8000330cd760,d,7) at ___mp_lock+0x70
> selwakeup(e80faaebded7c1a2) at selwakeup+0x9c
> ptsstart(8ce5939828d5e23) at ptsstart+0x79
> tputchar(174549bf676e909c,ffff800000afa400) at tputchar+0x85
> kputchar(75d50501b895e9e4,0,ffffffff81a5d6a2) at kputchar+0x91
> kprintf() at kprintf+0xe8
> printf(c9ef1007dec621e0) at printf+0x85
> witness_checkorder(2e4447d1b3cba2fe,ffffffff81af9df1,298,ffffffff81c8a678,ffffff
> ff81c8a688) at witness_checkorder+0x943
> end trace frame: 0xffff80003302e978, count: 0
>
I think I might already see such problem while doing some debug-printf
in kernel. so I should not be related to mpv.
It is a witness issue when a lock ordering problem is found while under
printf(9).
kern/subr_prf.c
503 int
504 printf(const char *fmt, ...)
505 {
506 va_list ap;
507 int retval;
508
509
510 va_start(ap, fmt);
511 mtx_enter(&kprintf_mutex);
512 retval = kprintf(fmt, TOCONS | TOLOG, NULL, NULL, ap);
513 mtx_leave(&kprintf_mutex);
514 va_end(ap);
515 if (!panicstr)
516 logwakeup();
517
518
519 return(retval);
520 }
printf(9) enter in `kprintf_mutex' to use kprintf().
but as witness_checkorder() might want to use printf(9), it could result
in deadlock situation (calling printf(9) from printf(9)).
I dunno what would be the right solution. Maybe witness should use
kprintf() directly (without taking kprintf_mutex) ? but it could result
in mangled output.
--
Sebastien Marie
