Re: PAT and MTRRs

[Arjan van de Ven <arjan@infradead.org>]

On Sun, 26 Oct 2008 22:46:21 -0100
"Diego M. Vadell" <dvadell@linuxclusters.com.ar> wrote:

> Hello,
> 
>    I have 6 identical PCs (HPC cluster) with MTRR problems. In older
> kernels, I had to use "mem=3300M", or else, I would get a very slowly
> boot (as when you run out of MTRRs). 
> 
>    So I thought that PAT would make this lack of MTRRs problem go
> away, and upgraded to 2.6.26.6 and 2.6.27.2, but it didn't: I still
> get (from dmesg)
> 
> x86 PAT enabled: cpu 0, old 0x7040600070406, new 0x7010600070106
> WARNING: BIOS bug: CPU MTRRs don't cover all of memory, losing 704MB
> of RAM.
> 
>    Most probably, I understood wrong. I read lwn.net's article [1]
> about PAT several times, Documentation/x86/pat.txt , tried to use
> mtrr_chunk_size= and mtrr_gran_size= in various combinations (as
> discussed in this LKML thread [2]), but I still don't get it.
> 
>    So, what did I miss? Am I wrong thinking that PAT is a better MTRR
> (wrt setting the cache type of the RAM)? 
> 

PAT can't make memory cachable that the MTRR's have as uncachable.
What PAT *can* do is, within an MTRR, do fine grained mapping...


-- 
Arjan van de Ven 	Intel Open Source Technology Centre
For development, discussion and tips for power savings, 
visit http://www.lesswatts.org