MMap function

[VALETTE Eric]

Aaron J. Grier wrote:
> On Thu, Mar 21, 2002 at 09:20:57AM +0100, VALETTE Eric wrote:
> 
> 
>>Aaron J. Grier wrote:
>>
>>>RTEMS has no virtual memory subsystem, and the filesystems are usually
>>>in-memory.  so no mmap() in RTEMS.
>>
>>This is not totally accurate. pc386 BSP has virtual memory turned on
>>(with of course a 1-1)  mapping. The reason is that there is no way
>>non-using the mmu to mark *only* portion of physical address space non
>>cacheable (think about PCI configuration space registers access).
> 
> 
> <slaps head> of course...  do other processors (thinking of powerPC and
> MIPS) do similar things with their MMUs when running RTEMS?
> 

I do not know MIPS architecture at all so I cannot say. For PPC, 
situation is a little bit different depending on models :
	1) On 6xx you have BATS registers that enable to control some big chunks 
of memory and they can be used if the memory map is not too complex (and 
in particlular if you do not need to map VME address space in addition 
to PCI, legacy IO, ...). I used them on motorola shared for performance 
reasons and also because I do not need to mess with Mmu. However, I 
think a more general implementation should use a conjunction of MMU and 
BAT for introducing flexibility in the address map,
	2) On 8xx the MMU is on as there is no BAT, so MMU is activated and you 
can control caching attributes,

Hope this helps,

-- 
    __
   /  `                          Eric Valette - Canon CRF
  /--   __  o _.                 Product Dev. Group Software Team Leader
(___, / (_(_(__                 Rue de la touche lambert
                                 35517 Cesson-Sevigne  Cedex
                                 FRANCE
Tel: +33 (0)2 99 87 68 91       Fax: +33 (0)2 99 84 11 30
E-mail: valette at crf.canon.fr    http://www.crf.canon.fr