XF86_SVGA + Savage/IX on a thinkpad T20

[David Butts <david.butts@storability.com>]

Hello, all-

I recently set about getting XFree86 3.3.6 to recognize the S3 Savage/IX
in my T20, with mixed results.

I pulled down the Debian source package for xserver-svga and pulled down
and applied the fix-08-s3savage_ix+mx patches.  After replacing the
upstream tarball with the patched version, I ran ./debian/rules binary,
which ended up running out of space on an otherwise empty 1GB partition.
When I tried again about a week later, as part of my debugging attempt,
and used more fine-grained builds (e.g. source.unpack, patchapply,
build-libc6, etc), I got the same result, albeit with only 512 MB of
free space to squander.

Before the original ./debian/rules binary ran out of disk space, it did
manage to build an XF86_SVGA, which appeared to work well, and which I
was running happily for a couple of days until I discovered that it was
causing the entire system to hang (needed a power cycle to clear) after
20 minutes - 1 hour or so of user inactivity (but not, FWIW, needing
screen, disk or cpu inactivity.  It would hang in the middle of a long-
running task I had given it if I left it alone long enough).  The time
isn't consistent, but does appear to fall within that window (based on
/var/log/messages marks).

I have duplicated the problem with a 2.2.16 kernel, a 2.2.18pre9 kernel
(with and without sound, crypto and raid patches).  I may have done it
with the vanilla 2.2.17 kernel from the initial installation (w/out APM
support), but in retrospect, I was not at my most alert at the time, so
that particular result may not be valid, and I don't have the patience
right now to leave the laptop running long enough with all its features
missing to confirm it.  It also happened without the hardware cursor
enabled, since that was about the only input-related feature of the X
server that wasn't available in XF86_FBDev, which is stable, but _very_
slow by comparison (obviously).

So, my questions (finally) are these:

(1)	Has anybody else tried/failed/succeeded in getting a "real" X
	server running on a Savage/IX (preferably on a ThinkPad)?  If
	so what tricks did/didn't work?

(2)	Given that I currently have an un-stripped XF86_SVGA binary,
	are there any debugging steps I can take to try to figure out
	what is going wrong here?

(3)	What have I done wrong that caused X to use more than 1GB of
	space while it was building?  Is there a Right Way to compile
	just the X servers?

Any light you good folks can help shed on this would be very much
appreciated, since the FBDev server just isn't cutting it for me
(Mmm... scrolling in Mozilla).

Thanks-

David