Re: 8.0-CURRENT kernel panic on system boot

Kostik Belousov wrote:
> On Thu, Aug 21, 2008 at 10:15:46PM +0800, kevin wrote:
>   
>> Kostik Belousov wrote:
>>     
>>> On Thu, Aug 21, 2008 at 09:33:55PM +0800, kevin wrote:
>>>  
>>>       
>>>> Hi,
>>>>   I update to the newest source tree and rebuild kernel(with zfs 
>>>> patch).kernel panic on system startup.Should I rebuild world?
>>>> ......
>>>> Starting gdm.
>>>> Sleeping thread (tid 100068,pid 1145) owns a non-sleepable lock
>>>> sched_switch() at sched_switch+0x184
>>>> mi_switch() at mi_switch+0x21b
>>>> sleepq_switch() at sleepq_switch+0xfa
>>>> sleepq_wait() at sleepq_wait+0x46
>>>> _sx_slock_hard() at _sx_slock_hard+0x169
>>>> _sx_slock() at _sx_slock+0x6b
>>>> acct_process() at acct_process+0x24b
>>>> exit1() at exit1+0x3e8
>>>> sys_exit() at sys_exit+0xe
>>>> syscall() at syscall+0x1dd
>>>> Xfast_syscall() at Xfast_syscall+0xab
>>>> --- syscall (1, FreeBSD ELF64, sys_exit), rip = 0x80094dd0c, rsp = 
>>>> 0x7fffffffcdf8, rbp = 0x62d548 ---
>>>> panic: sleeping thread
>>>> cpuid = 0
>>>> KDB: enter: panic
>>>> [thread pid 1143 tid 100127 ]
>>>> Stopped at       kdb_enter+0x3d: movq    $0,0x6cebb4(%rip)
>>>> db>bt
>>>> Tracing pid 1143 tid 100127 td 0xfffff000f073700
>>>> kdb_enter() at kdb_enter()+0x3d
>>>> panic() at panic()+0x16c
>>>> propagate_priority() propagate_priority+0x221
>>>> turnstile_wait() at turnstile_wait+0x212
>>>> _mtx_lock_sleep() at _mtx_lock_sleep+0xe0
>>>> _mtx_lock_flags() at _mtx_lock_flags+0xc0
>>>> kern_wait() at kern_wait+0x11b
>>>> wait4() at wait4+0x33
>>>> syscall() at syscall()+0x1dd
>>>> Xfast_syscall() at Xfast_syscall()+0xab
>>>> --- syscall (7, FreeBSD ELF64, wait4), rip = 0x8009115c, rsp = 
>>>> 0x7fffffffcf18, rbp = 0x26 ---
>>>> db>show locks
>>>> exclusive sx proctree (proctree) r = 0 (0xffffffff8099fc80) locked 
>>>> _at_/usr/src/sys/kern/kern_exit.c:702
>>>> db>show allpcpu
>>>> Current CPU:0
>>>> cpuid   = 0
>>>> curthread =0xffffff000f073700:pid 1143 "sh"
>>>> curpcb    =0xfffffffea6084d40
>>>> fpcurthread  = none
>>>> idlethread  = 0xffffff0002395380:pid 11 "idle: cpu0"
>>>> spin locks held:
>>>>
>>>> cpuid   = 1
>>>> curthread =0xffffff0002395700:pid 1143 "idle: cpu1"
>>>> curpcb    =0xfffffffe4001dd40
>>>> fpcurthread  = none
>>>> idlethread  = 0xffffff0002395700:pid 11 "idle: cpu1"
>>>> spin locks held:
>>>>
>>>>    
>>>>         
>>> No, this is a consequence of taking sx after proc lock in kern_acct.c,
>>> acct_process(). You may turn off accounting, or try this patch:
>>>
>>> diff --git a/sys/kern/kern_acct.c b/sys/kern/kern_acct.c
>>> index e505f62..d45e3b7 100644
>>> --- a/sys/kern/kern_acct.c
>>> +++ b/sys/kern/kern_acct.c
>>> _at__at_ -366,16 +366,25 _at__at_ acct_process(struct thread *td)
>>> 	 * Get process accounting information.
>>> 	 */
>>>
>>> +	sx_slock(&proctree_lock);
>>> 	PROC_LOCK(p);
>>> -	/* (1) The name of the command that ran */
>>> +
>>> +	/* (1) The terminal from which the process was started */
>>> +	if ((p->p_flag & P_CONTROLT) && p->p_pgrp->pg_session->s_ttyp)
>>> +		acct.ac_tty = tty_udev(p->p_pgrp->pg_session->s_ttyp);
>>> +	else
>>> +		acct.ac_tty = NODEV;
>>> +	sx_sunlock(&proctree_lock);
>>> +
>>> +	/* (2) The name of the command that ran */
>>> 	bcopy(p->p_comm, acct.ac_comm, sizeof acct.ac_comm);
>>>
>>> -	/* (2) The amount of user and system time that was used */
>>> +	/* (3) The amount of user and system time that was used */
>>> 	rufetchcalc(p, &ru, &ut, &st);
>>> 	acct.ac_utime = encode_timeval(ut);
>>> 	acct.ac_stime = encode_timeval(st);
>>>
>>> -	/* (3) The elapsed time the command ran (and its starting time) */
>>> +	/* (4) The elapsed time the command ran (and its starting time) */
>>> 	tmp = boottime;
>>> 	timevaladd(&tmp, &p->p_stats->p_start);
>>> 	acct.ac_btime = tmp.tv_sec;
>>> _at__at_ -383,7 +392,7 _at__at_ acct_process(struct thread *td)
>>> 	timevalsub(&tmp, &p->p_stats->p_start);
>>> 	acct.ac_etime = encode_timeval(tmp);
>>>
>>> -	/* (4) The average amount of memory used */
>>> +	/* (5) The average amount of memory used */
>>> 	tmp = ut;
>>> 	timevaladd(&tmp, &st);
>>> 	/* Convert tmp (i.e. u + s) into hz units to match ru_i*. */
>>> _at__at_ -394,21 +403,13 _at__at_ acct_process(struct thread *td)
>>> 	else
>>> 		acct.ac_mem = 0;
>>>
>>> -	/* (5) The number of disk I/O operations done */
>>> +	/* (6) The number of disk I/O operations done */
>>> 	acct.ac_io = encode_long(ru.ru_inblock + ru.ru_oublock);
>>>
>>> -	/* (6) The UID and GID of the process */
>>> +	/* (7) The UID and GID of the process */
>>> 	acct.ac_uid = p->p_ucred->cr_ruid;
>>> 	acct.ac_gid = p->p_ucred->cr_rgid;
>>>
>>> -	/* (7) The terminal from which the process was started */
>>> -	sx_slock(&proctree_lock);
>>> -	if ((p->p_flag & P_CONTROLT) && p->p_pgrp->pg_session->s_ttyp)
>>> -		acct.ac_tty = tty_udev(p->p_pgrp->pg_session->s_ttyp);
>>> -	else
>>> -		acct.ac_tty = NODEV;
>>> -	sx_sunlock(&proctree_lock);
>>> -
>>> 	/* (8) The boolean flags that tell how the process terminated, etc. 
>>> 	*/
>>> 	acct.ac_flagx = p->p_acflag;
>>> 	PROC_UNLOCK(p);
>>>  
>>>       
>> Thanks. After I turned off accounting in /etc/rc.conf. It works.
>>     
>
> And, could you, please, give us a service and try to turn accounting
> on the patched kernel ?
>   
yes  I rebuilt the kernel and it seems well now.