Hypertransport Speed

[Stefan Reinauer]

Hi,

I would like to check the applied patch into LinuxBIOS CVS if nobody
happens to disagree loudly:

1) hypertransport clocking

This patch allows to disable the speed cuts during hypertransport
   setup using cmos variables "amdk8_1GHz" and "amd8131_800MHz". 
   I've tried them on hardware which worked perfectly fine
   with the higher speed links for both devices (K8 and 8131).
   Since it is disabled per default and needs cmos and compile time 
   activation, it will not break anything.

Affected files:
    src/config/Options.lb
    src/devices/hypertransport.c
    src/northbridge/amd/amdk8/coherent_ht.c
    src/northbridge/amd/amdk8/incoherent_ht.c

2) ram init

This patch allows to use 8x (and probably 16x) dimms with LinuxBIOS
   on K8.

Affected files:
     src/northbridge/amd/amdk8/raminit.c

3) Debugging
   This patch will print the pci vendor/device id in print_pci_devices()
   which makes determining the early bus structure a lot easier.

It also only prints the first 128 bytes of SPDROM during 
   dump_spd_registers() since they are defined to be 128byte.

Affected files:
    src/northbridge/amd/amdk8/debug.c

Stefan

[Greg Lindahl]

On Wed, Jan 26, 2005 at 11:17:10AM +0100, Stefan Reinauer wrote:
> hypertransport clocking
>
> This patch allows to disable the speed cuts during hypertransport
> setup using cmos variables "amdk8_1GHz" and "amd8131_800MHz". 
> I've tried them on hardware which worked perfectly fine
> with the higher speed links for both devices (K8 and 8131).
> Since it is disabled per default and needs cmos and compile time 
> activation, it will not break anything.

You mean you tested for a short time and didn't see the AMD 8131 bug,
and so you are going to run production at the higher speed?

This is not so smart. The bug in the 8131 HT core is rare, but big
clusters will see it fairly frequently.

-- greg

[Stefan Reinauer]

* Greg Lindahl  [050126 22:48]:
> You mean you tested for a short time and didn't see the AMD 8131 bug,
> and so you are going to run production at the higher speed?
>
> This is not so smart. The bug in the 8131 HT core is rare, but big
> clusters will see it fairly frequently.

So you have experiences with the occurence of this bug? I'd like to hear
about it, since all I really know is in the public revision guide. I
could not reproduce this on modern systems yet.

Stefan