Re: [PATCH] dmtimer posting

[Tony Lindgren <tony@atomide.com>]

* Woodruff, Richard <r-woodruff2@ti.com> [080331 15:18]:
> > > > Attached is a rev-4 with the cpu_relax() added to while loops as
> > > > suggested by Ladislav, and CTRL register defines moved to the
> right
> > > > location.
> > >
> > > I confirm this version of the patch works on 3430 with latest
> kernel.
> > 
> > I'll push the rev-4 version of the patch today. The second patch needs
> > a bit more work still.
> 
> Great.  
> 
> I've been testing on a few systems with a functional variant of the 1st
> + 2nd patch back ported to older TI CDP kernels.  So far I've only ran
> into positive side effects.  No noticeable load, pass normal functional
> tests, and still able to hit all low power states.
> 
> In that variant I just added a load_start() function in addition to the
> load function and converted all the in tree callers to use this as they
> all called load() then start() in order anyway.  This left the API
> behavior the same but optimized all current users.

Can you send a patch against l-o tree for the load_start()? That might
be optimized enough :)

Tony


> Regards,
> Richard W.
> 
> I did get my ETM adaptor board back so I have been able to verify
> directly.  ETM is cool in that you can get (depending on filtering)
> almost a 60 second complete capture of instruction execution during some
> use case.