Re: [patch] sched: fix scheduling latencies in mtrr.c

[Ingo Molnar]

* Alan Cox <alan@xxxxxxxxxxxxxxxxxxx> wrote:

> On Maw, 2004-09-14 at 14:25, Ingo Molnar wrote:
> > caveat: one of the wbinvd() removals is correct i believe, but the other
> > one might be problematic. It doesnt seem logical to do a wbinvd() at
> > that point though ...
> 
> See the intel ppro manual volume 3 page 11-25. Its quite specific
> about the sequence, so unless anything changes with AMD or later
> processors the change seems to match the description.
> 
> IRQs are required to be off far longer than this sequence according to
> the docs however, and PGE is supposed to be cleared too.

the problem is, we've had IRQs on in all of the 2.6 kernels so any
argument about extra wbinvd brings the obvious question: 'why did it
work so far'?

the only wbinvd that makes sense is the one you quote:

> Step 11: set CD flag
> Step 12: wbinvd
> Step 13: set pge in CR4 if previously cleared

and i agree that the pge thing should be fixed too.

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