Re: [PATCH 1/1] Speedfreq-SMI call clobbers ECX

[Stephan Diestelhorst <langer_mann@web.de>]

> On, March 5th 2008 16:35:20 Ingo Molnar wrote:
> > * Stephan Diestelhorst <langer_mann@web.de> wrote:
> > > @@ -184,7 +184,7 @@ static void speedstep_set_state (unsigne
> > >  		__asm__ __volatile__(
> > >  			"movl $0, %%edi\n"
> > >  			"out %%al, (%%dx)\n"
> > > -			: "=b" (new_state), "=D" (result)
> > > +			: "=b" (new_state), "=D" (result), "=c" (ecx_clobber)
> > >
> > >  			: "a" (command), "b" (function), "c" (state), "d"
> > >  			: (smi_port), "S" (0)
> > >
> > >  			);
> >
> > stupid suggestion: why not do a pusha/popa around those
> > instructions, to make sure everything is restored? This isnt a
> > fastpath and being conservative about SMI side-effects cannot
> > hurt

Stephan Diestelhorst wrote:
> That sounds like a sane thing to do to me. Should I provide a
> 'patch'? Or leave that (and the decision about it) to the
> maintainer?

H. Peter Anvin wrote: 
> You can't pusha/popa if you expect a result.  You can, of course,
> push and pop individual registers.
>
> It's also kind of odd to do "movl $0,%%edi" instead of just setting
> EDI as an input.

Whoops, HPA is correct, of course. Manually pushing / popping the 
registers is ugly, how about a larger clobber-list? Let the compiler 
figure out what it wants to save/restore. Only thing to worry about 
is EBP then.

Again, should I provide these patches? This thing just annoyed me for 
a while as I have been patching it in my personal kernels for too 
long.

Regards,
  Stephan

PS: I'm not on LKML, please CC me at your discretion.