RE: hmm.... spinlocks..
On 31-Mar-01 Matthew Jacob wrote: > > On Fri, 30 Mar 2001, John Baldwin wrote: > >> >> On 30-Mar-01 Matthew Jacob wrote: >> > >> > >> > pic_initialize(): >> > lint0: 0x00010700 lint1: 0x00010400 TPR: 0x0010 SVR: 0x01ff >> > kernel trap 12 with interrupts disabled >> > panic: spin lock sched lock held by 0xc7ba8a60 for > 5 seconds >> > cpuid = 0; lapic.id = >> > Debugger("panic") >> > >> > CPU0 stopping CPUs: 0x0002... stopped. >> > Stopped at Debugger+0x45: pushl %ebx >> > db> t >> > Debugger(c02e3759) at Debugger+0x45 >> > panic(c02e2d60,c02fb929,c7ba8a60,c7ba7b80,fffea000) at panic+0xd0 >> > _mtx_lock_spin(c0357400,0,80246,c02e3a48,2fb) at _mtx_lock_spin+0x6e >> > wakeup(c3709378,c3709378,c0eec000,c0f71400,c3709454) at wakeup+0x67 >> > bufdone(c3709378,c7fbff40,c0131efb,c3709378,c0f71400) at bufdone+0x385 >> > bufdonebio(c3709378) at bufdonebio+0xe >> > dadone(c0f5d100,c0f71400) at dadone+0x1f7 >> > camisr(c032a834) at camisr+0x231 >> > ithread_loop(c0b2f080,c7fbffa8) at ithread_loop+0x247 >> > fork_exit(c019898c,c0b2f080,c7fbffa8) at fork_exit+0x83 >> > fork_trampoline() at fork_trampoline+0x8 >> >> Did you get a crashdump (probably not.) If you look at sched_lock, one of >> the >> words will be a pointer to the process owning the lock in question. >> Unfortunately it's not the first word anymore (something I may change in the >> future). On the alpha it would be 'sched_lock+48'. The pointer there >> points >> to the process owning the lock (and you can look up the process via ps). If >> you have a crash dump then I have some gdb macros that make it easy to get a >> backtrace of that process. If not, then, well, it gets harder. :-P Hmm, it >> might be nice to be able to ask ddb to give a backtrace of any arbitrary >> process. Maybe I'll add a new command for that.. >> >> The trick is that we want to know who grabbed sched_lock where and then >> started >> spinning with it. Using KTR with KTR_LOCK turned on and using the 'show >> ktr' >> command in ddb could also be used to see which process was the last to grab >> sched_lock and where it was grabbed. > > If it happens again, I''ll try. This was, btw, i386. Oh, duh. Umm, on i386 it's 'sched_lock + 28'. -- John Baldwin <[EMAIL PROTECTED]> -- http://www.FreeBSD.org/~jhb/ PGP Key: http://www.baldwin.cx/~john/pgpkey.asc "Power Users Use the Power to Serve!" - http://www.FreeBSD.org/ To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
RE: hmm.... spinlocks..
On Fri, 30 Mar 2001, John Baldwin wrote: > > On 30-Mar-01 Matthew Jacob wrote: > > > > > > pic_initialize(): > > lint0: 0x00010700 lint1: 0x00010400 TPR: 0x0010 SVR: 0x01ff > > kernel trap 12 with interrupts disabled > > panic: spin lock sched lock held by 0xc7ba8a60 for > 5 seconds > > cpuid = 0; lapic.id = > > Debugger("panic") > > > > CPU0 stopping CPUs: 0x0002... stopped. > > Stopped at Debugger+0x45: pushl %ebx > > db> t > > Debugger(c02e3759) at Debugger+0x45 > > panic(c02e2d60,c02fb929,c7ba8a60,c7ba7b80,fffea000) at panic+0xd0 > > _mtx_lock_spin(c0357400,0,80246,c02e3a48,2fb) at _mtx_lock_spin+0x6e > > wakeup(c3709378,c3709378,c0eec000,c0f71400,c3709454) at wakeup+0x67 > > bufdone(c3709378,c7fbff40,c0131efb,c3709378,c0f71400) at bufdone+0x385 > > bufdonebio(c3709378) at bufdonebio+0xe > > dadone(c0f5d100,c0f71400) at dadone+0x1f7 > > camisr(c032a834) at camisr+0x231 > > ithread_loop(c0b2f080,c7fbffa8) at ithread_loop+0x247 > > fork_exit(c019898c,c0b2f080,c7fbffa8) at fork_exit+0x83 > > fork_trampoline() at fork_trampoline+0x8 > > Did you get a crashdump (probably not.) If you look at sched_lock, one of the > words will be a pointer to the process owning the lock in question. > Unfortunately it's not the first word anymore (something I may change in the > future). On the alpha it would be 'sched_lock+48'. The pointer there points > to the process owning the lock (and you can look up the process via ps). If > you have a crash dump then I have some gdb macros that make it easy to get a > backtrace of that process. If not, then, well, it gets harder. :-P Hmm, it > might be nice to be able to ask ddb to give a backtrace of any arbitrary > process. Maybe I'll add a new command for that.. > > The trick is that we want to know who grabbed sched_lock where and then started > spinning with it. Using KTR with KTR_LOCK turned on and using the 'show ktr' > command in ddb could also be used to see which process was the last to grab > sched_lock and where it was grabbed. If it happens again, I''ll try. This was, btw, i386. To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
RE: hmm.... spinlocks..
On 30-Mar-01 Matthew Jacob wrote: > > > pic_initialize(): > lint0: 0x00010700 lint1: 0x00010400 TPR: 0x0010 SVR: 0x01ff > kernel trap 12 with interrupts disabled > panic: spin lock sched lock held by 0xc7ba8a60 for > 5 seconds > cpuid = 0; lapic.id = > Debugger("panic") > > CPU0 stopping CPUs: 0x0002... stopped. > Stopped at Debugger+0x45: pushl %ebx > db> t > Debugger(c02e3759) at Debugger+0x45 > panic(c02e2d60,c02fb929,c7ba8a60,c7ba7b80,fffea000) at panic+0xd0 > _mtx_lock_spin(c0357400,0,80246,c02e3a48,2fb) at _mtx_lock_spin+0x6e > wakeup(c3709378,c3709378,c0eec000,c0f71400,c3709454) at wakeup+0x67 > bufdone(c3709378,c7fbff40,c0131efb,c3709378,c0f71400) at bufdone+0x385 > bufdonebio(c3709378) at bufdonebio+0xe > dadone(c0f5d100,c0f71400) at dadone+0x1f7 > camisr(c032a834) at camisr+0x231 > ithread_loop(c0b2f080,c7fbffa8) at ithread_loop+0x247 > fork_exit(c019898c,c0b2f080,c7fbffa8) at fork_exit+0x83 > fork_trampoline() at fork_trampoline+0x8 Did you get a crashdump (probably not.) If you look at sched_lock, one of the words will be a pointer to the process owning the lock in question. Unfortunately it's not the first word anymore (something I may change in the future). On the alpha it would be 'sched_lock+48'. The pointer there points to the process owning the lock (and you can look up the process via ps). If you have a crash dump then I have some gdb macros that make it easy to get a backtrace of that process. If not, then, well, it gets harder. :-P Hmm, it might be nice to be able to ask ddb to give a backtrace of any arbitrary process. Maybe I'll add a new command for that.. The trick is that we want to know who grabbed sched_lock where and then started spinning with it. Using KTR with KTR_LOCK turned on and using the 'show ktr' command in ddb could also be used to see which process was the last to grab sched_lock and where it was grabbed. -- John Baldwin <[EMAIL PROTECTED]> -- http://www.FreeBSD.org/~jhb/ PGP Key: http://www.baldwin.cx/~john/pgpkey.asc "Power Users Use the Power to Serve!" - http://www.FreeBSD.org/ To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message