Re: Acquiring lockmgr lock "ufs" with non-sleepable locks

Hi,

On Jun 12, 2005, at 4:18 AM, Jun Kuriyama wrote:

>
> As of today's current:
>
> Acquiring lockmgr lock "ufs" with the following non-sleepable locks  
> held:
> exclusive sleep mutex vnode pollinfo r = 0 (0xc4814000) locked _at_ / 
> usr/src/sys/kern/kern_event.c:881
> KDB: stack backtrace:
> kdb_backtrace(1,3001,c3fa229c,c3a36780,ecf71a5c) at kdb_backtrace+0x29
> witness_warn(5,c070c378,c06a435b,c06ad03d) at witness_warn+0x18e
> lockmgr(c3fa2278,3001,c3fa229c,c3a36780,ecf71a88) at lockmgr+0x8d
> vop_stdlock(ecf71acc,c06f58c0,ecf71acc,ecf71a98,c0612184) at  
> vop_stdlock+0x1e
> VOP_LOCK_APV(c06f5e00,ecf71acc,ecf71aac,c067caff,ecf71acc) at  
> VOP_LOCK_APV+0x87
> ffs_lock(ecf71acc,1001,c3fa2220,ecf71ae8,c0566634) at ffs_lock+0x10
> VOP_LOCK_APV(c06f58c0,ecf71acc) at VOP_LOCK_APV+0x87
> vn_lock(c3fa2220,1001,c3a36780) at vn_lock+0xa4
> filt_vfsread(c3ac9264,0,0,ffffffff,0) at filt_vfsread+0x3a
> kqueue_register(c3dd6480,ecf71bf4,c3a36780,1,0) at kqueue_register 
> +0x668
> kern_kevent(c3a36780,4,1,0,ecf71cc8) at kern_kevent+0xc9
> kevent(c3a36780,ecf71d04,6,1,292) at kevent+0x55
> syscall(3b,3b,3b,1,804e068) at syscall+0x22f
> Xint0x80_syscall() at Xint0x80_syscall+0x1f
> --- syscall (363, FreeBSD ELF32, kevent), eip = 0x280b9503, esp =  
> 0xbfbfe8ec, ebp = 0xbfbfe938 ---

This was introduced by my commit to take out the various kqueue  
operations outside of ufs. It is because we are trying to lock the  
vnode while holding the knlist mutex. I'll work on a fix on  
wednesday, as I'll unfortunately be a bit busy until then. I hope  
this won't cause you any inconvenience.

Bye.
--
Suleiman Souhlal     | ssouhlal_at_vt.edu
The FreeBSD Project  | ssouhlal_at_FreeBSD.org