Subject: Re: X on TGA on NetBSD 1.6_RC1 on an XP1000
To: None <>
From: Roland Dowdeswell <>
List: port-alpha
Date: 08/20/2002 02:57:41
On 1029825726 seconds since the Beginning of the UNIX epoch wrote:
>On Mon, Aug 19, 2002 at 02:15:31AM -0400, Roland Dowdeswell wrote:
>> I have put my old PWS433a on top of a rack in a ``mission critical''
>> role and am left with only my XP1k with which to experiment.  I hadn't
>> really used the machine until now, and I loaded the latest 1.6 snapshot
>> on it and noticed that the X support for the PowerStorm 3d30 and 4d20
>> does not work.  I've been looking into why, and it seems that the
>> X server hasn't changed at all in a long time and it was all working
>> pretty recently on my PWS433a that is a few months ago.
>What are the symptoms ?

[ this is all with my PowerStorm 3d30 (TGA2/8bit) ]

Well, let's see.  I looked into this a couple of nights ago for a
couple of hours and basically found that the server was core dumping
on drawing the initial mesh pattern on the screen because it was
getting a segfault writing that last machine word of the screen.
At 1280x1024, this is right on the edge of the mmap(2), since the
rowsize is 2048 pixels and 1024 rows makes 2MB.

The mmap(2) is 4MB, but only the last 2MB are the frame buffer.
The first 2MB are register and ROM spaces.  So, at 1280x1024 the
end of the screen is right at the end of frame buffer memory.

I then hacked up the X server to consider the screen to be 1023
lines to avoid triggering the segfault, and with that the X server
runs properly but does not appear to change the screen with any of
its memory references (the ioctl(2)s to change the colour map still
work, though.)

Nothing has changed in the Xserver or the tga driver that should
make any real difference since the last time that I played with
this code which was in December when I got the 4d20 working.  And,
so I _think_ that this is a machine dependent problem---since I
haven't tried to use X on my XP1k yet.

The most likely suspect [so far] in my scanning through the problem
is perhaps bus_space_mmap(9) isn't working in the same way on the

I'm mainly asking for someone else's experiences on this on a
different machine, so that I can test my theory that it is specific
to my XP1k.  I do not have any other Alpha's in my place except
for one which I can't use for testing purposes.


    Roland Dowdeswell                      http://www.Imrryr.ORG/~elric/