Re: panic upon kldunload snd_ich (lor # 159)

On Tue, Sep 13, 2005 at 07:35:50PM +0000, Ben Kaduk wrote:

[...]

 > Pyun,
 > 
 > I was able to try your patch -- it did not apply cleanly, but was simple to 
 > manually enter.
 > After building a new kernel, the panic is gone, but the LOR remains -- it 

Thanks for testing. Fix committed to HEAD(sndstat.c rev. 1.21)

 > does seem that they were unrelated.
 > For the records, the backtrace of the LOR (#159) is:
 > lock order reversal
 > 1st 0xc1716bc0 pcm0 (sound cdev) _at_ 
 > /usr/src/sys/modules/sound/sound/../../../de
 > v/sound/pcm/sound.c:761
 > 2nd 0xc196b540 sndstat (sndstat) _at_ 
 > /usr/src/sys/modules/sound/sound/../../../de
 > v/sound/pcm/sndstat.c:256
 > KDB: stack backtrace:
 > kdb_backtrace(c06ef80d,c196b540,c1965a62,c1965a62,c1965a6a) at 
 > kdb_backtrace+0x2
 > f
 > witness_checkorder(c196b540,9,c1965a6a,100,c0725260) at 
 > witness_checkorder+0x68f
 > _sx_xlock(c196b540,c1965a6a,100,c197a380,c18fca00) at _sx_xlock+0x7f
 > sndstat_unregister(c171eb00,c197a380,c1965b77,2f9,c171eb00) at 
 > sndstat_unregiste
 > r+0x27
 > pcm_unregister(c171eb00,c761bbfc,c0557f32,c1943820,c171eb00) at 
 > pcm_unregister+0
 > x112
 > ich_pci_detach(c171eb00,c18e0050,c0721ba8,961,c1909ab0) at 
 > ich_pci_detach+0x13
 > device_detach(c171eb00,c1942428,c171eb00,c1683400,c194380c) at 
 > device_detach+0x8
 > f
 > devclass_delete_driver(c1683400,c1943820,1,c1716280,c1716280) at 
 > devclass_delete
 > _driver+0x8e
 > driver_module_handler(c1716280,1,c194380c) at driver_module_handler+0xe7
 > module_unload(c1716280,0,1fb,0,0) at module_unload+0x61
 > linker_file_unload(c193ca00,0,c06e9d40,327,0) at linker_file_unload+0x89
 > kern_kldunload(c16ebaf0,5,0,c761bd30,c06afce5) at kern_kldunload+0x96
 > kldunloadf(c16ebaf0,c761bd04,8,422,2) at kldunloadf+0x2c
 > syscall(3b,3b,3b,5,bfbfef12) at syscall+0x295
 > Xint0x80_syscall() at Xint0x80_syscall+0x1f
 > --- syscall (444, FreeBSD ELF32, kldunloadf), eip = 0x280b700b, esp = 
 > 0xbfbfe9dc
 > , ebp = 0xbfbfee48 ---
 > pcm0: detached
 > 
 > 

AFAIK, the LOR is well known to sound developers. Though it's not real
LOR(grabbing sx lock while holding mutex) it's difficult to fix.

-- 
Regards,
Pyun YongHyeon