* Ingo Molnar <mingo@xxxxxxx> wrote:
i'll add a new feature to debug this: when crashing on an assert and
tracing is enabled the trace leading up to the crash will be printed
to the console. [...]
the -R3 patch has this feature:
http://redhat.com/~mingo/voluntary-preempt/voluntary-preempt-2.6.9-rc1-bk4-R3
you can enable it via enabling CONFIG_LATENCY_TRACE and doing:
echo 3 > /proc/sys/kernel/trace_enabled
it's all automatic from this point on: tracing will happen nonstop and
any assert or crash that prints the process stack will also print the
last 100 trace entries. Sample output:
Call Trace:
[<c0160401>] sys_munmap+0x61/0x80
[<c010520d>] sysenter_past_esp+0x52/0x71
Last 100 trace entries:
00000001: zap_pmd_range+0xe/0x90 <= (unmap_page_range+0x55/0x80)
00000001: stop_trace+0x8/0x20 <= (bust_spinlocks+0x20/0x60)
00000001: bust_spinlocks+0xe/0x60 <= (die+0xbc/0x2a0)
[... 97 more trace entries ...]
Please capture and mail the first 'extended' oops that triggers
(secondary followup traces are probably just side-effects of the crash),
it could give us clues about where the bug is.
Ingo