On Fri, 2010-01-08 at 12:10 +0100, Philippe Gerum wrote:
> On Fri, 2010-01-08 at 11:59 +0100, Richard Cochran wrote:
> > On Thu, Dec 10, 2009 at 07:56:28AM +0100, Bernhard Pfund wrote:
> > > I'm currently working with a fairly new ppc development board (P2020RDB) 
> > > by 
> > > Freescale. The board's BSP went mainline with 2.6.32 and I'd like to 
> > > deploy an 
> > > I-pipe enabled DENX kernel for testing. Now, is there an I-pipe patch for 
> > > said 
> > > kernel on the release schedule anywhere soon?
> > 
> > I got ipipe-2.6.30 running the p2020ds with minimal effort. I can post
> > a diff if that might help you.
> > 
> > I should have my own p2020rdb soon...
> > 
> > I have noticed some unexplained freezes running xenomai programs on
> > the p2020ds in SMP mode, but I have not yet had time to look into
> > it. I hope we can work together to get SMP working reliably.
> I'm working on fixing the pipeline for 2.6.32/ppc64, which includes SMP
> mode. 32bit looks ok in UP mode, but I still need to validate SMP there.
> So it would be great if you could hammer that patch over your P2020 as
> well once it's ready. ETA, early next week. I'll keep you informed about
> this.

As expected, upgrading from 2.6.30 to 2.6.32 for ppc64 was a real pain.
Anyway, here is a patch which applies against mainline; it is
pretty large because it also includes DENX-originated bits, from v2.6.32
to DENX-v2.6.32.


This is expected to work over Xenomai 2.5.0. Tested over 52xx, 85xx, 4xx
for the 32bit support, pasemi for the 64bit part. I did not manage to
put my hands on a working ppc32/SMP board to give it a shot yet, so,
well, I wish you luck. This said, ppc64/SMP has been validated, so there
is hope.

Sidenote: ftrace may induce a massive stack consumption on ppc64 under
certain circumstances, causing kernel stack overflows. A way to work
around this should be to enable CONFIG_IRQSTACKS, but this won't work
with Xenomai over ppc64 yet, albeit this does work over ppc32. 

This patch is in a staging directory for now; waiting for some feedback
to go further.



Xenomai-core mailing list

Reply via email to