[PATCH] powerpc: Fix stack overflow crash in resume_kernel when ftracing

[Steven Rostedt]

On Fri, 2013-06-14 at 00:31 +1000, Michael Ellerman wrote:
> On Thu, 2013-06-13 at 08:45 -0400, Steven Rostedt wrote:
> > On Thu, 2013-06-13 at 21:04 +1000, Michael Ellerman wrote:
> > 
> > > Although that should be sufficient to fix the bug, we also mark the
> > > runlatch routines as notrace. They are called very early in the
> > > exception entry and we are asking for trouble tracing them. They are
> > > also fairly uninteresting and tracing them just adds unnecessary
> > > overhead.
> > 
> > Note, I usually lean towards tracing everything that can be traced, and
> > only adding notrace to things that will actually cause a crash. If you
> > don't like them to be traced, you can always do:
> 
> Yeah fair enough. In this case I think we don't want to trace it.
> Although it doesn't cause a crash right now (at least after part 1 of
> this patch), it's being called at a time when things are fragile, and
> it's possible we could get bitten again some other way if the
> surrounding code changes.
> 

I wont push to trace it, as if it is fragile code, then it's best not to
risk it. Tracing is always a second class citizen in the kernel ;-)

> 
> > echo '*__ppc64_runlatch_*'
> > > /sys/kernel/debug/tracing/set_ftrace_notrace
> > 
> > and that will keep them from being traced. You can also add it to the
> > kernel command line with: ftrace_notrace=*__ppc64_runlatch_* which will
> > also disable them on boot up.
> > 
> > Also trace-cmd has:
> > 
> > trace-cmd record -p function_graph -n '*__ppc64_runlatch_*'
> > 
> > that will do the same thing.
> > 
> > Hmm, I should add a way to disable things that are usually considered
> > noise. Perhaps add something like:
> > 
> > 
> > FTRACE_DEFAULT_OFF(__ppc64_runlatch_on);
> > 
> > That adds the function to a different section that places it into
> > another file that keeps it from being traced, but can be enabled when
> > you want it to.
> 
> Yeah that would be cool.
> 
> Personally I'm just using shell scripts that poke the files in sysfs,
> and I'm often running on different boxes, so the more that just works
> without extra config by me the better.

Agreed, which is why I hope to get something like this in. Perhaps I'll
aim for 3.12. Then we can look at all the functions with "notrace" and
see which can be converted.

Thanks,

-- Steve