Re: [patch 1/5] x86, bts: detect size of DS fields

[Ingo Molnar <mingo@elte.hu>]

* Metzger, Markus T <markus.t.metzger@intel.com> wrote:

> >-----Original Message-----
> >From: Ingo Molnar [mailto:mingo@elte.hu]
> >Sent: Tuesday, March 17, 2009 11:56 AM
> >To: Metzger, Markus T
> 
> 
> >the bigger problem is the runtime failure (boot warning) that i
> >reported. Hence i cannot apply any new patches yet until that is
> >resolved.
> 
> I found your email at lkml.org.

(i sent it to your intel.com address - perhaps it got dropped 
there somehow?)

> I think I found the problem:
>   ds_request_bts() needs to allocate a small amount of memory.
>   It uses GFP_KERNEL.
> 
> Hw-branch-tracer essentially does on_each_cpu(ds_request_bts()).
> 
> Since ds_request_bts() is meant to be called on the cpu that 
> should be traced, it should use GFP_ATOMIC to allow calls from 
> smp_call_function() and friends.
> 
> Would that be acceptable?

That's not really a good solution - GFP_ATOMIC is not a reliable 
form of allocation.

the other callsites are buggy too:

                smp_call_function_single(cpu, bts_trace_start_cpu, NULL, 1);

done under the bts_tracer_lock in addition to an atomic IPI 
context.

for_each_online_cpu() done under the mutex would be better i 
guess, plus you can allocate any memory before you do the SMP 
cross-call, and pass it to the IPI handler via the data 
parameter. (NULL in the sequence above)

	Ingo