Re: [patch 1/2] x86, fpu: split FPU state from task struct - v3

[Suresh Siddha <suresh.b.siddha@intel.com>]

On Tue, Mar 04, 2008 at 11:28:04AM +0100, Ingo Molnar wrote:
> 
> * Suresh Siddha <suresh.b.siddha@intel.com> wrote:
> 
> > Split the FPU save area from the task struct. This allows easy 
> > migration of FPU context, and it's generally cleaner. It also allows 
> > the following two optimizations:
> 
> hm, i didnt get a reply from you to:
> 
>     http://lkml.org/lkml/2008/3/3/47
> 
> I'm uneasy to apply patches without knowing whether previous problems 
> are fixed. In the changelog there's a mention of:
> 
>   v3: Fixed the non-atomic calling sequence in atomic context.
> 
> is that the same bug?

Yes. Sorry for not pointing out explicitly. Essentially 32bit kernel with
CONFIG_PREEMPT is calling math_state_restore() with interrupts disabled.
Handled this by enabling/disabling the interrupts around the blocking call.
with small code changes ensuring that it is safe to enable/disable interrupts
at this point.

Do you want a v4 or separate patch for handling the kmem_cache_alloc() failure

thanks,
suresh