Messages in this thread | | | Date | Fri, 12 Dec 2008 12:31:26 +0100 | From | Ingo Molnar <> | Subject | Re: [crash] Re: [PATCH 1/3] tracing/function-graph-tracer: introduce __notrace_funcgraph to filter special functions |
| |
* Steven Rostedt <rostedt@goodmis.org> wrote:
> > On Tue, 9 Dec 2008, Fr?d?ric Weisbecker wrote: > > 2008/12/9 Fr?d?ric Weisbecker <fweisbec@gmail.com>: > > > > > > I built this config on an smp x86-64 and booted several times (and > > tested the function tracer manually) with and without the following > > options: > > > > initcall_debug apic=verbose sysrq_always_enabled ignore_loglevel > > selinux=0 nmi_watchdog=2 idle=poll > > > > And I can't reproduce this bug.... That might be a very nasty race condition... > > Neither can I :-( > > Ingo, is this on the box that stresses NMIs?
yes. You could try to inject even more NMIs than i normally do, artificially, by picking up tip/master and running KernelTop with a 100,000 cycles IRQ interval:
http://redhat.com/~mingo/perfcounters/kerneltop.c
run it like this:
./kerneltop -c 100000
and run an infinite loop per process:
while :; do :; done &
to get the NMIs going intensely:
------------------------------------------------------------------------------ KernelTop: 115437 irqs/sec [NMI, 100000 cycles], (all, 16 CPUs) ------------------------------------------------------------------------------
events RIP kernel function ______ ______ ________________ _______________
81224.00 - ffffffffff6001a4 : vread_hpet 3119.00 - ffffffff8027cf99 : perf_read 2484.00 - ffffffff802227d5 : read_hpet 1157.00 - ffffffff8056b69a : mutex_lock 1041.00 - ffffffffff600000 : vgettimeofday 986.00 - ffffffff80333db5 : avc_has_perm_noaudit 896.00 - ffffffff8026be2d : audit_syscall_exit 653.00 - ffffffff8056c8f5 : _spin_lock 577.00 - ffffffff802ecd14 : dnotify_parent 528.00 - ffffffff8033679c : file_has_perm 489.00 - ffffffff802bc060 : dput 470.00 - ffffffff8026c366 : audit_syscall_entry
Ingo
| |