panic: PHOLD of exiting process

[John Baldwin]

On Sunday 26 February 2006 04:35, Peter Holm wrote:
> On Fri, Feb 24, 2006 at 03:53:26PM -0500, John Baldwin wrote:
> > On Friday 24 February 2006 14:50, Peter Holm wrote:
> > > GENERIC HEAD from Feb 23 18:08 UTC
> > > 
> > > panic: PHOLD of exiting process
> > > cpuid = 0
> > > KDB: enter: panic
> > > [thread pid 75139 tid 100298 ]
> > > Stopped at      kdb_enter+0x2b: nop
> > > db> where
> > > Tracing pid 75139 tid 100298 td 0xc24244e0
> > > kdb_enter(c0888514) at kdb_enter+0x2b
> > > panic(c084a3c9,d57,c09af2d8,e,c08904ae) at panic+0x14b
> > > softdep_disk_io_initiation(c72ea010) at
> > > softdep_disk_io_initiation+0x7a
> > > ffs_geom_strategy(c242beec,c72ea010) at ffs_geom_strategy+0x32
> > > bufwrite(c72ea010,0,0,cd5419b4,c0784aa1) at bufwrite+0x15a
> > > ffs_bufwrite(c72ea010) at ffs_bufwrite+0x282
> > > ffs_update(c3121b2c,0,c3121b2c,c23d8c00,1) at ffs_update+0x3a5
> > > ufs_inactive(cd541a04) at ufs_inactive+0x178
> > > VOP_INACTIVE_APV(c092b5e0,cd541a04) at VOP_INACTIVE_APV+0x7e
> > > vinactive(c3121b2c,c24244e0) at vinactive+0x72
> > > vput(c3121b2c,c23d8c00,c093a720,c3121b2c,1) at vput+0x188
> > > vn_close(c3121b2c,1,c2d74600,c24244e0,cd541acc) at vn_close+0x96
> > > vn_closefile(c26b1e10,c24244e0) at vn_closefile+0xca
> > > fdrop_locked(c26b1e10,c24244e0,c219a3a0,0,c0884d50) at
> > > fdrop_locked+0x88
> > > fdrop(c26b1e10,c24244e0,6b5,c095b034,0) at fdrop+0x24
> > > closef(c26b1e10,c24244e0) at closef+0x367
> > > fdfree(c24244e0) at fdfree+0x4a3
> > > exit1(c24244e0,0,cd541d30,c081b34e,c24244e0) at exit1+0x450
> > > exit1(c24244e0,cd541d04,c3a77890,c,c24244e0) at exit1
> > > syscall(3b,3b,3b,2804eaec,bfbfebe8) at syscall+0x27a
> > > 
> > > http://people.freebsd.org/~pho/stress/log/cons188.html
> > 
> > I'm talking with Jeff and Kris about it.  This is my current thinking:
> > 
> > --- //depot/vendor/freebsd/src/sys/sys/proc.h	2006/02/22 19:00:49
> > +++ //depot/projects/smpng/sys/sys/proc.h	2006/02/24 20:11:28
> > @@ -792,7 +792,8 @@
> >  } while (0)
> >  #define	_PHOLD(p) do {							\
> >  	PROC_LOCK_ASSERT((p), MA_OWNED);				\
> > -	KASSERT(!((p)->p_flag & P_WEXIT), ("PHOLD of exiting process"));\
> > +	KASSERT(!((p)->p_flag & P_WEXIT) || (p) == curproc,		\
> > +	    ("PHOLD of exiting process"));				\
> >  	(p)->p_lock++;							\
> >  	if (((p)->p_sflag & PS_INMEM) == 0)				\
> >  		faultin((p));						\
> > 
> > -- 
> > John Baldwin <jhb at FreeBSD.org>  <><  http://www.FreeBSD.org/~jhb/
> > "Power Users Use the Power to Serve"  =  http://www.FreeBSD.org
> 
> Without your patch I get two of these panics within a short time frame
> of testing. With your patch I have not seen this problem for a 24 hour
> period of testing.

I'll commit it along with a note of why the assertion is still safe.

-- 
John Baldwin <jhb at FreeBSD.org>  <><  http://www.FreeBSD.org/~jhb/
"Power Users Use the Power to Serve"  =  http://www.FreeBSD.org