Subject: RE: PXA250 Xscale and page tables
To: None <kyle.unice@L-3com.com, colin@cook-tech.com, scw@netbsd.org,>
From: Gary Thorpe <gathorpe79@yahoo.com>
List: tech-kern
Date: 02/03/2003 14:00:39
 --- kyle.unice@L-3com.com wrote: > In the file
sys/arch/arm/arm32/pmap.c function pmap_vac_me_user() the
> caching atributes for a process are turned off.  The program can be a
> simple
> "hello world" without any process space page sharing and still the
> caching
> attributes for the process pages are disabled.   
> 
> My guess would be that the pages should be marked as cacheable until
> they
> are determined to be shared and then at that point in time they would
> be
> marked "uncached".
> 
> Kyle
> 

Would it be smarter to mark the pages uncached ONLY when one of the
mappings allows writing? Or is it correct to do it as you suggest? From
what I have read on cpu cache designs, ones based on virtual addresses
are almost useless for modern OSes... would anyone know of why they
would be designed this way despite cache coherancy problems the OS must
face?



______________________________________________________________________ 
Post your free ad now! http://personals.yahoo.ca