[parisc-linux] Switch console error.

[Matthew Wilcox]

On Tue, Mar 12, 2002 at 06:42:06PM -0700, Grant Grundler wrote:
> Matthew Wilcox wrote:
> > The kernel has its own mappings for pages so it can take advantage of
> > 16MB pages.  These are not coherent with the pages that are mapped into
> > user space. 
> 
> And this would cause an HPMC?

Yes, indirectly, see below.

> If someone could decode the Gururaj's HPMC data ("ser pim" from BCH prompt)
> and post the results.  At worst, it would confirm this theory.

Yes, that would certainly help.  As I said, it is only a theory...

> Sorry, I'm not VM clueful enough to understand why this would be a
> problem on N-class and not other boxes.  I know coherency works differently
> here (Merced Bus is "central" vs Runway for most other boxes).
> Someplace to look for more clues?

It's not to do with the bus (AFAIK...).  It's to do with Stretch (the
memory controller).  We've got away with using non-equivalent aliases (see
the PA 2.0 book [1]) up until now, but it seems that Stretch doesn't let
us get away with it any more.  So we have to be sure to use equivalent
aliases when accessing a page.  It's been suggested to me that Stretch
has its own tags for a page and if it gets writes to different tags on
the same physical page, it generates an HPMC.

[1] I would have given a reference here, but
http://h21007.www2.hp.com/dspp/tech/tech_TechDocumentDetailPage_IDX/1,1701,959,00.html
gives me a Redirection Limit exceeded error... hopefully this will get
fixed soon.

-- 
Revolutions do not require corporate support.