i830 raw mode driver

[Keith Whitwell]

Dave Airlie wrote:
> Hi,
> 
> Recently as many people know I've been working on an i810 driver that
> doesn't require the BIOS to do modesetting, I've gotten this mostly
> working on the chipsets I have (i865, i915GM and i945G) and I think it
> is as far as I can take it considering
> 
> a) I've delivered what my customer required which was CRT and DVI
> modesetting on an i915GM.
> b) I've no direct access to documentation and I spend a lot of time
> going around in circles breaking things on one chipset and then
> "fixing" them on another.
> 
> I'm just wondering how people feel it should be taken forward, with
> respect to merging my code into the official tree?
> 
> My code is in a git tree at
> http://people.freedesktop.org/~airlied/git/xf86-video-i810/.git
> 
> and the latest patch vs the point I started developing is at
> http://www.skynet.ie/~airlied/patches/i810/i810_raw_mode.diff
> 
> Now the code is ugly in a lot of places, using multiple pipes isn't
> going to work at all,  the relation between output and pipes isn't
> existing too well, the raw pll calculations may not always give a
> valid value (the i9xx ones just pissed me off too much in the end, I
> cannot get the p1/p2 relationship straight - someone with docs
> could..)
> 
> The SDVO code is designed to only talk to one type of SDVO chip, the
> Chrontel, again I implemented as much of the sDVO protocol as I could
> reverse engineer in the time I had, a lot of it is guesswork and I
> apologies for some of the ugly looking crap in PreSetMode.
> 
> All my code is controlled by an Option in the xorg.conf so it can be
> switched off quite easily.

Dave,

Alan's on holidays at the moment, so I can't answer directly for him,
but to say that it sounds like from your description that you've got
some reservations about having it go straight into the trunk.

While it's probably tempting fate to talk about future plans, addressing
native modesetting is something that we think that we'll be able to
make a proper go of, supported by docs, etc - but not for at least a
month or two.

Until then, do you think the code's in a state that could be merged but
disabled by default (ie an option to turn it on, rather than one to turn
it off)?  Or is it too raw for that?

Is this something you're thinking of for 7.1, or to hold over for the 
release after that?

I'd hate to see people who would have been ok with the current 
modesetting mechanism ending up having to muck around with xorg.conf 
files to try and get back to a working setup.  That seems to argue for a 
merged but disabled-by-default approach.

Anyway, Alan will be back next week, so maybe he can give a more 
informed opinion at that point...

Keith