[ECOS] Memory mapping of PCI device memory - PC platform

[ECOS] Memory mapping of PCI device memory - PC platform

[Fabrice Gautier]

Hi,
	
In the PC platform:
The Memory model in eCos is a 4GB flat model right?

So there is nothing special to do with eCos when we map memory from a PCI
device, right ?

A+
-- 
Fabrice Gautier
fabrice_gautier@sdesigns.com 

Re: [ECOS] Memory mapping of PCI device memory - PC platform

[Nick Garnett]

Fabrice Gautier <Fabrice_Gautier@sdesigns.com> writes:

> Hi,
> 	
> In the PC platform:
> The Memory model in eCos is a 4GB flat model right?

Correct.

> 
> So there is nothing special to do with eCos when we map memory from a PCI
> device, right ?
> 

Not as far as I am aware.

-- 
Nick Garnett, eCos Kernel Architect
Red Hat, Cambridge, UK

Re: [ECOS] Memory mapping of PCI device memory - PC platform

[Andrew Lunn]

> > So there is nothing special to do with eCos when we map memory from a PCI
> > device, right ?
> > 
> 
> Not as far as I am aware.

When we do this for the ebsa285 you have to ensure the MMU is set up
correctly so that it does not cache or use a write buffer when access
the PCI memory. There is probably some equivelent setting required in
the MMU for the i386. Something to check anyway.

        Andrew

Re: [ECOS] Memory mapping of PCI device memory - PC platform

[Hugo Tyson]

andrew.lunn@ascom.ch (Andrew Lunn) writes:
> > > So there is nothing special to do with eCos when we map memory from a PCI
> > > device, right ?
> > 
> > Not as far as I am aware.
> 
> When we do this for the ebsa285 you have to ensure the MMU is set up
> correctly so that it does not cache or use a write buffer when access
> the PCI memory. There is probably some equivelent setting required in
> the MMU for the i386. Something to check anyway.

Correct.

BTW, for anoncvs fans, soon you'll be getting releases which automatically
allocate all "spare" memory to the heap (as in malloc()) on lots of
platforms.  You have to use the GUI MLT or edit nasty linker files to
prevent this.  The EBSA285 has an explicit and fairly hacky workaround, to
let PCI work,installed in the .ldi files.

We're working on a nicer solution, but for now beware on other platforms,
you may find your PCI area, for example, in memory that you *know* is not
being used, scribbled over by the heap, unless you take action to prevent
it, by editting the .ldi files or using the MLT.

Just in case, this might save someone some debug time...

	- Huge

[ECOS] POSIX API for eCOS ?

[Stuart Adams]

 How soon might we see a first cut of the
 Elix POSIX API for eCOS ???  

Thanks,
  Stuart

Re: [ECOS] POSIX API for eCOS ?

[Jonathan Larmour]

Stuart Adams wrote:
> 
>  How soon might we see a first cut of the
>  Elix POSIX API for eCOS ???

I'm working on updating anon CVS right now, but for various reasons I won't
have time to finish it this week.

Jifl
-- 
Red Hat, 35 Cambridge Place, Cambridge, UK. CB2 1NS  Tel: +44 (1223) 728762
"Plan to be spontaneous tomorrow."  ||  These opinions are all my own fault