Re: Patches to use local APIC timer cause SCHED_ULE panic at boot

On Tue, 22 Feb 2005 14:00:52 -0500, John Baldwin <jhb_at_freebsd.org> wrote:
> On Monday 21 February 2005 03:35 pm, Arjan Van Leeuwen wrote:
> > Hey,
> >
> > The recently committed patches to use the local APIC timer to drive
> > the various kernel clocks on SMP machines (by jhb) panic my SMP system
> > at boot when
> > 1) SCHED_ULE is used instead of SCHED_4BSD
> > 2) INVARIANTS_* and WITNESS_* are disabled (I don't know which one is
> > the culprit, but if I enable them all the system doesn't panic).
> >
> > I can't provide a dump, since (I assume) the hard drive isn't
> > initialized yet, but I have a debug kernel if that's useful.
> >
> > Here is the panic:
> > Timecounters tick every 1.000 msec
> > panic: Thread not on runq.
> > cpuid = 0
> > KDB: enter: panic
> > [thread pid 0 tid 0]
> > Stopped at    kdb_enter+0x30: leave
> > db> trace
> > Tracing pid 0 tid 0 td 0xc0769020
> > kdb_enter(c071398d,0,c07147a2,c0c207d4,c1a12f60) at kdb_enter+0x30
> > panic(c07147a2,369e99,c0769020,1,c0746120) at panic+0x14e
> > sched_switch(c0769020,0,2,b43e27c0,87c7fc) at sched_switch+0x85
> > mi_switch(2,0,ffffffff,7fff0000,ffffffff) at mi_switch+0x1d9
> > critical_exit(c0c20888,c0c20890,c06e8dfe) at critical_exit+0xbf
> > lapic_handle_timer(0) at lapic_handle_timer+0xf5
> > Xtimerint(c0c20938,c0714fc7,a,61c2091c,c05c4000) at Xtimerint+0x30
> > vsscanf(c0886103,c0714fb3,c0c20a70,c0c20b50,c0564312) at vsscanf+0x1fc
> > sscanf(c0886103,c0714fb3,c0c20b24,c0c20a80,c0c20b04) at sscanf+0x1f
> > res_find(c0c20bf8,c0c20bd0,c0719124,0,0) at res_find+0x262
> > resource_find(c0c20bf8,c0c20bd0,c0719124,0,0) at resource_find+0x67
> > resource_find_dev(c0c20bf8,c0719124,c0c20bfc,0,0) at resource_find_dev+0x6a
> > if_findindex(c1c1ec00,0,0,0,c0703b9c) at if_findindex+0x126
> > if_attach(c1c1ec00,c07044a6,0,0,c074cea0) at if_attach+0x1ac
> > lo_clone_create(c074cea0,0,d,0,c074ce88) at lo_create+0x79
> > ifc_simple_create(c074cea0,c0c20cf8,10,c07044a6,0) at
> > ifc_simple_create+0x66 ifc_simple_attach(c074cea0,c0718afd,0,0,c0777a20) at
> > ifc_simple_attach+0x55 if_clone_attach(c074cea0,c0718d90,0,0,c0c20d80) at
> > if_clone_attach+0x1cf loop_modevent(c198b8c0,0,0,c28000,c0c20d80) at
> > loop_modevent+0x6a
> > module_register_init(c074cef4,c1e000,c1ec00,c1e000,0) at
> > module_register_init+0x81
> > mi_startup() at mi_startup+0xb5
> > begin() at begin+0x2c
> >
(...)
> 
> This doesn't look to be specific to the local APIC timer.  The softclock
> thread should be on a run queue since it was being preempted to via
> critical_exit().  KTR_SCHED traces would probably be helpful.

I'm reading up on KTR_SCHED, but before I start trying to do anything
with it, will it be able to do anything useful when the system isn't
in multi-user yet? I understand from
http://www.watson.org/~robert/freebsd/netperf/ktr/ that I'll have to
set some sysctls that don't seem to be loader tunables, and I won't be
able to run ktrdump.

Thanks,

Arjan

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