How to make graphical issue go away with 4GB of ram

[Denis 'GNUtoo' Carikli]

Hi,

me and Kyösti Mälkki(kmalkki on IRC) figured out how to workarround the
following problem:
If I installed 4GB of ram in the m4a785t-m the screen comes up,however
it's black.
If I only installed 2GB of ram it worked fine.

I made a patch(I'll send it right after that email) but it seem to break
graphics for the case where 2GB of ram is installed.

So I wonder what would be the correct fix(a fix that would work for
every cases).

Explanation of the patch that will be sent right after the mail:
"basek - mmio_basek;" should be 1048576 (1G):
According to Kyösti Mälkki the correct value for the sizek is 1048576.
So I looked for what could create that size and found that.
I'm not convinced that "basek - mmio_basek" is the right fix but at
least we're in the right direction.

Denis.

[PATCH] Fix black screen issue when 4GB of RAM is present.

[Denis 'GNUtoo' Carikli]

Thanks a lot to Kyösti Mälkki(kmalkki) for the help on #coreboot
  IRC channel on Freenode Servers.

Signed-off-by: Denis 'GNUtoo' Carikli 
---
 src/northbridge/amd/amdfam10/northbridge.c |   12 ++++++------
 1 files changed, 6 insertions(+), 6 deletions(-)

diff --git a/src/northbridge/amd/amdfam10/northbridge.c b/src/northbridge/amd/amdfam10/northbridge.c
index 3b20684..573c535 100644
--- a/src/northbridge/amd/amdfam10/northbridge.c
+++ b/src/northbridge/amd/amdfam10/northbridge.c
@@ -1028,6 +1028,11 @@ static void amdfam10_domain_set_resources(device_t dev)
    			unsigned pre_sizek;
    			pre_sizek = mmio_basek - basek;
    			if(pre_sizek>0) {
+#if CONFIG_GFXUMA
+					/* Deduct uma memory before reporting because
+					 * this is what the mtrr code expects */
+					pre_sizek -= uma_memory_size / 1024;
+#endif
    				ram_resource(dev, (idx | i), basek, pre_sizek);
    				idx += 0x10;
    				sizek -= pre_sizek;
@@ -1063,15 +1068,10 @@ static void amdfam10_domain_set_resources(device_t dev)
    		}
    		else {
    			basek = 4*1024*1024;
-				sizek -= (4*1024*1024 - mmio_basek);
+				sizek = basek - mmio_basek;
    		}
    	}

-#if CONFIG_GFXUMA
-		/* Deduct uma memory before reporting because
-		 * this is what the mtrr code expects */
-		sizek -= uma_memory_size / 1024;
-#endif
    	ram_resource(dev, (idx | i), basek, sizek);
    	idx += 0x10;
 #if CONFIG_WRITE_HIGH_TABLES

[Kyösti Mälkki]

On Sat, 2012-06-30 at 18:18 +0200, Denis 'GNUtoo' Carikli wrote:
> Hi,
>
> me and Kyösti Mälkki(kmalkki on IRC) figured out how to workarround the
> following problem:
> If I installed 4GB of ram in the m4a785t-m the screen comes up,however
> it's black.
> If I only installed 2GB of ram it worked fine.
>
> I made a patch(I'll send it right after that email) but it seem to break
> graphics for the case where 2GB of ram is installed.
>
> So I wonder what would be the correct fix(a fix that would work for
> every cases).
>
> Explanation of the patch that will be sent right after the mail:
> "basek - mmio_basek;" should be 1048576 (1G):
> According to Kyösti Mälkki the correct value for the sizek is 1048576.
> So I looked for what could create that size and found that.
> I'm not convinced that "basek - mmio_basek" is the right fix but at
> least we're in the right direction.

My explanation of the issue: uma_memory_size must reduce the size of the
last RAM resource _below_ 4GB. It now touches the last RAM resource,
even if located over 4GB.

Thanks,
KM

[Marc Jones]

On Sat, Jun 30, 2012 at 12:01 PM, Kyösti Mälkki  wrote:
> On Sat, 2012-06-30 at 18:18 +0200, Denis 'GNUtoo' Carikli wrote:
> > Hi,
> >
> > me and Kyösti Mälkki(kmalkki on IRC) figured out how to workarround the
> > following problem:
> > If I installed 4GB of ram in the m4a785t-m the screen comes up,however
> > it's black.
> > If I only installed 2GB of ram it worked fine.
> >
> > I made a patch(I'll send it right after that email) but it seem to break
> > graphics for the case where 2GB of ram is installed.
> >
> > So I wonder what would be the correct fix(a fix that would work for
> > every cases).
> >
> > Explanation of the patch that will be sent right after the mail:
> > "basek - mmio_basek;" should be 1048576 (1G):
> > According to Kyösti Mälkki the correct value for the sizek is 1048576.
> > So I looked for what could create that size and found that.
> > I'm not convinced that "basek - mmio_basek" is the right fix but at
> > least we're in the right direction.
>
> My explanation of the issue: uma_memory_size must reduce the size of the
> last RAM resource _below_ 4GB. It now touches the last RAM resource,
> even if located over 4GB.

Yes, The UMA graphics memory is allocated in the lower 4GB. I think
that should be happening in mainboard.c. See the AMD reference
platforms.

Regards,
Marc