Re: [PATCH 3/25][V3] irq_flags / halt routines

[Glauber de Oliveira Costa <gcosta@redhat.com>]

Andi Kleen escreveu:
> On Wed, Aug 15, 2007 at 12:09:42PM -0300, Glauber de Oliveira Costa wrote:
> > Again, this is the code of such function:
> >
> > static inline int raw_irqs_disabled_flags(unsigned long flags)
> > {
> >         return !(flags & X86_EFLAGS_IF);
> > }
> > so all it is doing is getting a parameter (flags), and bitmasking it. It 
> > is not talking to any hypervisor. I can't see your point. Unless you are
> > arguing that it _should_ be talking to a hypervisor. Is that your point?
>
> vSMP is a hypervisor based architecture. For some reason that is not
> 100% clear to me, but Kiran or Shai can probably explain, it needs this 
> additional bit in EFLAGS when interrupts are disabled. That gives
> it some hints and then it goes somehow faster. That is clearly
> paravirtualization.
>
> Since paravirtops is designed to handle such hooks cleanly I request
> that you move vSMP over to it or work with the vSMP maintainers to 
> do that. Otherwise we have two different ways to do paravirtualization 
> which is wrong.

Thanks for the explanation, Andi. I understand it much better now, and 
agree with you.

As alternatives what we have now, we can either keep the paravirt_ops as 
it is now for the native case, just hooking the vsmp functions in place 
of the normal one, (there are just three ops anyway), refill the 
paravirt_ops entirely in somewhere like vsmp.c, or similar (or maybe 
even assigning paravirt_ops.fn = vsmp_fn on the fly, but early enough).

Maybe we could even make VSMP depend on PARAVIRT, to make it sure it is 
completely a paravirt client.

But as you could see, my knowledge of vsmp does not go that far, and I 
would really like to have input from the vsmp guys prior to touch 
anything here.

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/