Re: [PATCH] Use global TLB flushes in MTRR code

[Ingo Molnar <mingo@elte.hu>]

* Andi Kleen <andi@firstfloor.org> wrote:

> On Thu, Feb 07, 2008 at 08:13:37PM +0100, Ingo Molnar wrote:
> > 
> > * Andi Kleen <andi@firstfloor.org> wrote:
> > 
> > > [probably stable material too]
> > > 
> > > Use global TLB flushes in MTRR code
> > > 
> > > Obviously kernel mappings should be flushed here too.
> > 
> > no, your patch is not needed:
> 
> Yes you're right.  Thanks makes more sense.  The weird style fooled 
> me.
> 
> It seems to come from the pseudo code in the Intel manual, but I think 
> it would be still cleaner/more idiomatic to use two __flush_tlb_all() 
> and remove the explicit code. The only drawback would be some more cr4 
> accesses, which does not seem like a big issue.
> 
> Updated patch follows.

... and this patch of yours breaks MTRR setting subtly:

> -	/*  Save value of CR4 and clear Page Global Enable (bit 7)  */
> -	if ( cpu_has_pge ) {
> -		cr4 = read_cr4();
> -		write_cr4(cr4 & ~X86_CR4_PGE);
> -	}
> -
> -	/* Flush all TLBs via a mov %cr3, %reg; mov %reg, %cr3 */
> -	__flush_tlb();
> +	/* Flush all TLBs */
> +	__flush_tlb_all();

because it's not just an open-coded __tlb_flush_all(), it _disables PGE 
and keeps it so while the MTRR's are changed on all CPUs_.

Your patch adds __flush_tlb_all() which re-enables the PGE bit in cr4, 
see asm-x86/tlbflush.h:

        /* clear PGE */
        write_cr4(cr4 & ~X86_CR4_PGE);
        /* write old PGE again and flush TLBs */
        write_cr4(cr4);

so we'll keep PGE enabled during the MTRR setting - which changes 
behavior.

	Ingo