panic: spin lock held too long

Hi

I've got this panic on 2 out of 3 make world (-j8) attempts on my
SMP system using SCHED_BSD.

I'm compiling a kernel with options BREAK_TO_DEBUGGER to see if I
can reproduce the panic once again and break into the debugger to
force a coredump.  It seems to not stop CPU0 because 'spin lock
sleepq chain held by xxxx for > 5 seconds' message keeps on appearing
after the panic.

Any ideas or requests for specific information are welcome.

I have private mods to the following files:

M sys/dev/usb/uhci_pci.c
M sys/kern/kern_synch.c
M sys/netinet/ip_fastfwd.c
M sys/netinet/ip_fw.h
M sys/netinet/ip_fw2.c
M sys/netinet/ip_input.c
M sys/netinet/ip_output.c

The modification to sys/kern/kern_synch.c was suggested by jhb to
stop another panic (spin lock held too long) relating to a deadlock
in the swapper.  USB doesn't work and the rest fixes ipfw tee.  I
can back these out if people think they might be interfering.

Fatal trap 1spin lock sched lock held by 0xc0fc4690 for > 5 seconds
panic: spin lock held too long
cpuid = 1; 
Stack backtrace:
backtrace(100,c23032a0,3938700,c065f520,4) at backtrace+0x12
panic(c06113fe,c06113d5,c061141c,c0fc4690,c23032a0) at panic+0x11e
_mtx_lock_spin(c065f520,2,0,0) at _mtx_lock_spin+0x80
hardclock_process(d1b44c60,bd69f,d1b44cb0,c05c7a20,0) at hardclock_process+0x60
forwarded_hardclock(0) at forwarded_hardclock+0x17
Xhardclock(c065f520,0,0,0) at Xhardclock+0x30
sched_userret(c23032a0) at sched_userret+0x6c
userret(c23032a0,d1b44d48,0,1,0) at userret+0x11
syscall(805002f,2f,805002f,8057000,0) at syscall+0x359
Xint0x80_syscall() at Xint0x80_syscall+0x1f
--- syscall (0), eip = 0x28107473, esp = 0x805bf80, ebp = 0x805bfbc ---
spin lock sched lock held by 0xc0fc4690 for > 5 seconds
panic: spin lock held too long
cpuid = 1; 
spin lock sched lock held by 0xc0fc4690 for > 5 seconds
panic: spin lock held too long
cpuid = 1; 
....

Ian

--
Ian Freislich