[commit] r5897 - trunk/src/cpu/via/car

[repository service]

Author: stepan
Date: Fri Oct  1 14:24:57 2010
New Revision: 5897
URL: https://tracker.coreboot.org/trac/coreboot/changeset/5897

Log:
fix VIA C7 code.
Signed-off-by: Stefan Reinauer 
Acked-by: Stefan Reinauer

Modified:
   trunk/src/cpu/via/car/cache_as_ram.inc

Modified: trunk/src/cpu/via/car/cache_as_ram.inc
==============================================================================
--- trunk/src/cpu/via/car/cache_as_ram.inc	Fri Oct  1 13:34:05 2010	(r5896)
+++ trunk/src/cpu/via/car/cache_as_ram.inc	Fri Oct  1 14:24:57 2010	(r5897)
@@ -100,7 +100,8 @@
     */
    movl	$MTRRphysBase_MSR(1), %ecx
    xorl	%edx, %edx
-	movl	$(REAL_XIP_ROM_BASE | MTRR_TYPE_WRBACK), %eax
+	movl	$REAL_XIP_ROM_BASE, %eax
+	orl	$MTRR_TYPE_WRBACK, %eax
    wrmsr

movl	$MTRRphysMask_MSR(1), %ecx
@@ -243,7 +244,8 @@
    /* Cache XIP_ROM_BASE-SIZE to speedup coreboot code. */
    movl	$MTRRphysBase_MSR(3), %ecx
    xorl	%edx, %edx
-	movl	$(REAL_XIP_ROM_BASE | MTRR_TYPE_WRBACK), %eax
+	movl	$REAL_XIP_ROM_BASE, %eax
+	orl	$MTRR_TYPE_WRBACK, %eax
    wrmsr

movl	$MTRRphysMask_MSR(3), %ecx

[Warren Turkal]

Stefan,

Is there a technical reason to split the logical "or" considering that
the values constants that are known at compile time? It seems like the
additional orl could be avoided by how the code was before?

BTW, I am asking to learn, not because I think it's wrong.

Thanks,
wt

On Fri, Oct 1, 2010 at 5:24 AM, repository service  wrote:
> Author: stepan
> Date: Fri Oct  1 14:24:57 2010
> New Revision: 5897
> URL: https://tracker.coreboot.org/trac/coreboot/changeset/5897
>
> Log:
> fix VIA C7 code.
> Signed-off-by: Stefan Reinauer 
> Acked-by: Stefan Reinauer
>
> Modified:
>   trunk/src/cpu/via/car/cache_as_ram.inc
> Modified: trunk/src/cpu/via/car/cache_as_ram.inc
>
> --- trunk/src/cpu/via/car/cache_as_ram.inc      Fri Oct  1 13:34:05 2010        (r5896)
> +++ trunk/src/cpu/via/car/cache_as_ram.inc      Fri Oct  1 14:24:57 2010        (r5897)
> @@ -100,7 +100,8 @@
>         */
>        movl    $MTRRphysBase_MSR(1), %ecx
>        xorl    %edx, %edx
> movl    $(REAL_XIP_ROM_BASE | MTRR_TYPE_WRBACK), %eax
> movl    $REAL_XIP_ROM_BASE, %eax
> orl     $MTRR_TYPE_WRBACK, %eax
>
> wrmsr
>
> movl    $MTRRphysMask_MSR(1), %ecx
> @@ -243,7 +244,8 @@
>        /* Cache XIP_ROM_BASE-SIZE to speedup coreboot code. */
>        movl    $MTRRphysBase_MSR(3), %ecx
>        xorl    %edx, %edx
> movl    $(REAL_XIP_ROM_BASE | MTRR_TYPE_WRBACK), %eax
> movl    $REAL_XIP_ROM_BASE, %eax
> orl     $MTRR_TYPE_WRBACK, %eax
>
> wrmsr
>
> movl    $MTRRphysMask_MSR(3), %ecx

[Patrick Georgi]

Am 02.10.2010 11:33, schrieb Warren Turkal:
> Stefan,
>
> Is there a technical reason to split the logical "or" considering that
> the values constants that are known at compile time? It seems like the
> additional orl could be avoided by how the code was before?
>
> BTW, I am asking to learn, not because I think it's wrong.

With the bootblock, it's not known at compile time - it's resolved at
link time:
- first it's compiled
- then linked at a harmless address (maybe 0? not sure)
- then the size is determined, and a new offset is calculated
- then it's linked again, at the new offset

Patrick