radeon driver: checkpoint

[Matthieu Herrb]

Benjamin Herrenschmidt wrote:
> On Sat, 2006-03-18 at 15:18 +0100, Matthieu Herrb wrote:
>>>> I've just check out ati-1-0-branch again. And it's back to usable.
>> For me the ati-1-0-branch is still worse than the driver from 6.9.
>> I'm using a Radeon 9200 card on an amd64 machine under OpenBSD.
>>
>> The 6.9 driver works in mergedfb mode, but not in traditional multi head 
>> mode (with one screen per head (this used to work in 6.8.2)):  it 
>> produces a corrupted display on the 2nd head, or just mirror from the 
>> 1st, depending on the options.
>>
>> With the current code from the ati-1-0-branch (copied to the monolithic 
>> tree and built from there, if it matters), my merdedfb setting, results 
>> on the 2nd screen always mirroring the 1st, although I specified 
>> otherwise in xorg.conf.
> 
> 
>> The dual-screen settings results in no video signal at all on the 2nd 
>> head, although the log file indicates that the 2nd monitor is probed 
>> correctly, and that it seems to setup up a correct mode for it.
> 
> This is with the latest CVS I suppose ?

Yes. ati-1-0-branch.

  I had a problem with Xinerama /
> second head that I fixed though it tended to cause a different issue.
> 
>> xorg.conf and log files for both configurations are there:
>> <http://www.herrb.com/mergedfb.xorg.conf>
>> <http://www.herrb.com/mergedfb.Xorg.0.log>
>>
>> <http://www.herrb.com/multi.xorg.conf>
>> <http://www.herrb.com/multi.Xorg.0.log>
> 
> The, the mergedfb log looks like everything is ok. Do you end up with a
> very large virtual area (sized like both screens) but instead of having
> the second screen showing what's on the "other" side, it mirrors and
> both are panning when the mouse hits the border ?

I don't get panning. The mouse "disappears" on the invisible other side.

> If yes, then that
> looks like something I've seen too, it looks like mergedfb doesn't
> always setup some of it's own stuff properly on startup.
> 
> Can you try adding another mergedfb mode in your MetaModes (for example
> a real clone mode, you can do that by using "+" instead of "-" as a
> separator between the two screen dimensions) and then try switching back
> and forth between the 2 mode pairs with xrandr ? Does that result in the
> inital mode starting to work properly once you switched out and back
> in ?

Yes, that works. After switching back from the real mirror mode, I get 
back to a correct display on the 2nd screen.

> I'm not totally sure what's up with mergedfb at this point there, the
> code is a bit hermetic to me, I'm still hoping that Alex will dig into
> that part of the problem :)
> 
> Regarding the Xinerama issue, I'm not sure what's up yet, best is if you
> could take register dumps of the working 6.8.2 setup vs. the non-working
> 6.9 setup...

Sure, what tool can I use to take those dumps ?

-- 
					Matthieu