XGI and Xorg driver source for xgi cards

[pcpa at mandriva.com.br]

Quoting Ian Romanick <idr at us.ibm.com>:

> |  Recently I received a xgi card, the sources from xgi, and some
> | documentation (3 pdf files).
>
> Which card?  Is it one of the older XG40 / XG45 cards (Volari V3XE,
> Volari V5, Volari V8, etc.) or is it one of the newer XP10 / XG47 cards
> (Volari 8300)?  I only ask because the each use a different driver.  The
> XG40 cards use xf86-video-xgi (which you mention below), and the XP10
> cards use xf86-video-xgixp.

  The card I have is a Volari Z7. I had it and some documentation
for only a few days (usually I work with OEMs with Brasil team, but
it is a France project, so I just got a card sent to me after I
asked for one, but I think this is not the card used in the
"final solution" ...)

> |  After some time looking at the code, and assuming that besides
> | based on an earlier Xorg version, I assumed the code from xgi is
> | newer. But now I have my doubts.
>
> The xf86-video-xgi driver is basically a hacked up version of
> xf86-video-sis.  It removes support for older SiS chips and adds support
> for the XG40.

  I noticed that :-)

> |  While "studying" the code, I ended up starting to "redo" what
> | Ian Romanick apparently did more then one year ago in
> | xf86-driver-xgi (based on logs), i.e. I changed it to use only
> | xgi_<foo>.h instead of a mix with <foo>.h, and did significant
> | changes to use only one internal data structure or function from
> | the different, and somehow coexisting implementations, etc.
> |
> |  I don't have any xgi contact, so I would like to know, most
> | likely from Ian Romanick, what is the current state of the xgi
> | driver.
>
> I haven't touched xf86-video-xgi in a long time, and I don't have any
> plans to pick it up again.  I'm trying to get xf86-video-xgixp in shape
> because IBM is going to ship a system (Google "Bimini IBM") with a
> Volari 8300.  If you intend to work on xf86-video-xgi, you should start
> from the code in GIT, *not* the code that XGI provides.

  After I sent the first email, I hacked xf86-video-xgixp and the xgi
kernel module to make it "recognize" the card I have. Apparently it
locks if using mmio, and the video ram detection seens buggy, but I
believe it's "batch buffer" is very similar to the one used by the card
you are using. Anyway, I will try a bit more, hopefully I can manage
to make it work... Btw I have reported in bugzilla that xgixp has
the unresolved symbols drmFenceDestroy.

  I agree that the start point should be the code in git, not the code
xgi provides, but it looks like that to be able to check for any changes
since you last worked on it, I pretty much would need to remake all the
changes you already did (I am using indent with same options in both
trees to figure out the diffs...)

Thanks again,
Paulo