[Xen-devel] Re: [patch 13/26] Xen-paravirt_ops: Consistently wrap paravirt ops callsites to make them patchable

[Matt Mackall <mpm@xxxxxxxxxxx>]

On Tue, Mar 20, 2007 at 03:08:19PM -0800, Zachary Amsden wrote:
> Matt Mackall wrote:
> >I don't know that you need an xchg there. If you're still on the same
> >CPU, it should all be nice and causal even across an interrupt handler.
> >So it could be:
> >
> >   pda.intr_mask = 0; /* intr_pending can't get set after this */
> >  
> 
> Why not?  Oh, I see.  intr_mask is inverted form of EFLAGS_IF.

It's not even that. There are two things that can happen:

case 1:

  intr_mask = 1;
            <interrupt occurs and is deferred>
  intr_mask = 0;
  /* intr_pending is already set and CLI is in effect */
  if(intr_pending)

case 2:

  intr_mask = 1;
  intr_mask = 0;
            <interrupt occurs and is processed>
  /* intr_pending remains cleared */
  if(intr_pending)

As this is all about local interrupts, it's all on a single CPU and
out of order issues aren't visible..
 
> >(This would actually need a C barrier, but I'll ignore that as this'd
> >end up being asm...)

..unless the compiler is doing the reordering, of course.

> >But other interesting things could happen. If we never did a real CLI
> >and we get preempted and switched to another CPU between clearing
> >intr_mask and checking intr_pending, we get a little confused. 
> 
> I think Jeremy's idea was to have interrupt handlers leave interrupts 
> disabled on exit if pda.intr_mask was set.  In which case, they would 
> bypass all work and we could never get preempted.

I was actually worrying about the case where the interrupt came in
"late". But I don't think it's a problem there either.

> I don't think leaving 
> hardware interrupts disabled for such a long time is good though.

It can only be worse than the current situation by the amount of time
it takes to defer an interrupt once. On average, it'll be a lot
better as most critical sections are -tiny-.

-- 
Mathematics is the supreme nostalgia of our time.

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel