Re: 2.6.33: ftrace triggers soft lockup

[Steven Rostedt <rostedt@goodmis.org>]

On Mon, 2010-03-15 at 09:46 +0800, Li Zefan wrote:
> Steven Rostedt wrote:
> > On Fri, 2010-03-12 at 21:07 -0500, Steven Rostedt wrote:
> >> On Tue, 2010-03-09 at 11:38 +0800, Li Zefan wrote:
> >>
> >>> The segfault is from userspace. This is a segfault I just got:
> >>>
> >>> ftrace_stack_ma[14989]: segfault at fe2e ip 00000000004474c0 sp 00007fffb97e38c0 error 4 in bash[400000+d1000]
> >> Li and Lai,
> >>
> >> Thanks a lot for reporting all this, and writing the ftrace_stress_test.
> >> I've fixed all the bugs that it has reported (to my knowledge) and after
> >> a little more testing I will soon be sending a patch series out.
> > 
> > Li,
> > 
> > I ran your test for 13 hours on three different machines. I did not hit
> > any more errors. Although I see a lockdep warning (not a locking problem
> > but the WARN_ON in fork.c is triggering). I'll look into that on Monday.
> > 
> 
> As I said in the offline mail, this warning is caused by global trace_clock,
> and I sent a fix sometime ago:
> 
> http://lkml.org/lkml/2009/12/21/4

Ah, I was waiting for a final Acked-by: from Peter Zijlstra.

Peter,

Should I pull in that patch? I don't see anything wrong with converting
a raw_local_irq_save() to a local_irq_save() in the tracing code. I'll
apply it locally and see if it fixes the warn on that I see.

Note, even the function tracer uses straight local_irq_save(), so I
would think that the clock code could too.

-- Steve

> 
> > Please take the new patches that Ingo just pushed to Linus, if you find
> > any more bugs, please let me know right away.
> > 
> 
> Thanks for the fixes!
>