Unkillable KSE threaded proc

[Julian Elischer]

"bugger"


Andrew Gallatin wrote:
> Julian Elischer writes:
>  > Andrew Gallatin wrote:
>  > > Julian Elischer writes:
>  > >  > >
>  > >  > >Maybe this would be easier to debug if I disabled preemption?
>  > >  > >
>  > >  > 
>  > >  > 
>  > >  > I think that this would possibly GO AWAY of you disab;ed preemption. 
>  > >  > which would make it very hard to debug :-)
>  > >  > 
>  > > 
>  > > Yes and no.  You initially asked me to try in -current because of
>  > > some changes you'd made to the exit code.  RELENG_5 (with the old
>  > > exit code and no preemption) shows a different problem (proc is
>  > > just not killable).    If the proc was killable without preemption,
>  > > that would at least show your new code is better..
>  > 
>  > try the attached diff:
>  > 
> 
> This is worse..
> 
> Its worse in that the application never starts running fully, and that
> it seems to ignore signals entirely.  I can't attach a debugger to it
> to see how far it got before hanging due to the signal problem.  When
> it hangs, (both before and after a signal is sent) the CPU utilization
> is 0%..  Before its sent a signal, it looks like this:
> 
>   573 c1f3b8c0 e88ae000 1387   517   573 000c082 (threaded) mx_pingpong
>    thread 0xc1f3e320 ksegrp 0xc19ead20 [RUNQ]
>    thread 0xc1f3e4b0 ksegrp 0xc19ead20 [RUNQ]
>    thread 0xc1f3e640 ksegrp 0xc19eaaf0 [SLPQ ksesigwait 0xc1f3b9c0][SLP]
> 
> 
> db> call db_trace_thread(0xc1f3e320, -1)
> sched_switch(c1f3e320,0,1,1862ccb2,994777d8) at sched_switch+0x137
> mi_switch(1,0,c05fdf59,804c000,c2b8c2ec) at mi_switch+0x1ce
> turnstile_wait(c1a518c0,c06c53e0,c1a4d7d0,0,1) at turnstile_wait+0x339
> _mtx_lock_sleep(c06c53e0,c1f3e320,0,0,0) at _mtx_lock_sleep+0x122
> vm_fault(c187a5dc,804c000,1,0,0) at vm_fault+0x214
> trap_pfault(e88b8d48,1,804c800,3,804c800) at trap_pfault+0x136
> trap(2f,2f,2f,805d13c,805d13c) at trap+0x201
> calltrap() at calltrap+0x5
> --- trap 0xc, eip = 0x804c800, esp = 0xbfbfe66c, ebp = 0xbfbfe678 ---
> 0
> 
> db> call db_trace_thread(0xc1f3e4b0, -1)
> sched_switch(c1f3e4b0,0,1,f0007932,9935c3e9) at sched_switch+0x137
> mi_switch(1,0,c19ead60,e88bbc5c,c1f3e4b0) at mi_switch+0x1ce
> sleepq_switch(c19ead60,c1f3e4b0,0,e88bbc94,c04e5da6) at sleepq_switch+0x171
> sleepq_timedwait_sig(c19ead60,0,c1f3b92c,c0677640,100) at sleepq_timedwait_sig+0x13
> msleep(c19ead60,c1f3b92c,168,c0677640,1771) at msleep+0x37b
> kse_release(c1f3e4b0,e88bbd14,4,c04c47ab,0) at kse_release+0x29b
> syscall(2f,2f,2f,8054200,0) at syscall+0x2fc
> Xint0x80_syscall() at Xint0x80_syscall+0x1f
> --- syscall (383, FreeBSD ELF32, kse_release), eip = 0x280a3d4f, esp = 0x8194f80, ebp = 0x8194fbc ---
> 0
> 
> db> call db_trace_thread(0xc1f3e640, -1)
> sched_switch(c1f3e640,0,1,bc7c14b2,97d6ec54) at sched_switch+0x137
> mi_switch(1,0,0,0,0) at mi_switch+0x1ce
> sleepq_switch(c1f3b9c0,c1f3e640,0,e88bec94,c04e5da6) at sleepq_switch+0x171
> sleepq_timedwait_sig(c1f3b9c0,0,0,0,0) at sleepq_timedwait_sig+0x13
> msleep(c1f3b9c0,c1f3b92c,168,c0677635,bb9) at msleep+0x37b
> kse_release(c1f3e640,e88bed14,4,c04c47ab,0) at kse_release+0x1a1
> syscall(2f,2f,2f,1,81) at syscall+0x2fc
> Xint0x80_syscall() at Xint0x80_syscall+0x1f
> --- syscall (383, FreeBSD ELF32, kse_release), eip = 0x280a3d4f, esp = 0xbfafef30, ebp = 0xbfafef8c ---
> 0
> 
> 
> A different run, but after sending it a ^C from the command line:
> 
>  547 c1f3b1c0 e88aa000    0     1   547 000c482 (threaded)  mx_pingpong
>    thread 0xc1f3e960 ksegrp 0xc19eaee0 [RUNQ]
>    thread 0xc1f3eaf0 ksegrp 0xc19eaee0 [RUNQ]
>    thread 0xc1f3ec80 ksegrp 0xc19eab60 [SUSP]
> 
> db> call db_trace_thread(0xc1f3e960, -1)
> sched_switch(c1f3e960,0,2,e7ff39b6,d6d80c8c) at sched_switch+0x137
> mi_switch(2,0,0,0,0) at mi_switch+0x1ce
> ast(e88c4d48) at ast+0x4eb
> doreti_ast() at doreti_ast+0x17
> 0
> db> call db_trace_thread(0xc1f3eaf0, -1)
> sched_switch(c1f3eaf0,0,1,6e2ca4e6,d6924d2f) at sched_switch+0x137
> mi_switch(1,0,c19eaf20,e88c7c5c,c1f3eaf0) at mi_switch+0x1ce
> sleepq_switch(c19eaf20,c1f3eaf0,0,e88c7c94,c04e5da6) at sleepq_switch+0x171
> sleepq_timedwait_sig(c19eaf20,0,c1f3b22c,c0677640,100) at sleepq_timedwait_sig+0x13
> msleep(c19eaf20,c1f3b22c,168,c0677640,1771) at msleep+0x37b
> kse_release(c1f3eaf0,e88c7d14,4,c04c47ab,0) at kse_release+0x29b
> syscall(2f,2f,2f,8054200,0) at syscall+0x2fc
> Xint0x80_syscall() at Xint0x80_syscall+0x1f
> --- syscall (383, FreeBSD ELF32, kse_release), eip = 0x280a3d4f, esp = 0x8194f80, ebp = 0x8194fbc ---
> 0
> db> call db_trace_thread(0xc1f3ec80, -1)
> sched_switch(c1f3ec80,0,1,26e24232,4249ca0b) at sched_switch+0x137
> mi_switch(1,0,0,0,0) at mi_switch+0x1ce
> thread_single(1,c1f3ec80,c1f3b1c0,e88cac5c,c0500581) at thread_single+0x1d7
> exit1(c1f3ec80,2,e88cacb8,c04f1736,0) at exit1+0x115
> expand_name(c1f3ec80,2,c1f3ec80,e88cad48,0) at expand_name
> kse_thr_interrupt(c1f3ec80,e88cad14,c,c1f3ec80,e88cad3c) at kse_thr_interrupt+0x329
> syscall(2f,2f,2f,8054100,805a800) at syscall+0x2fc
> Xint0x80_syscall() at Xint0x80_syscall+0x1f
> --- syscall (382, FreeBSD ELF32, kse_thr_interrupt), eip = 0x280a3d6f, esp = 0xbfafee60, ebp = 0xbfafeefc ---
> 0
> 
> 
> If you want line number translations, please let me know.  I saved the
> kernel that this came from and also took a dump.
> 
> Drew