Re: NetBSD 5.0 and SPARCstation 5 not a lucky combination

[Bill Roman <netbsd%songdog.eskimo.com@localhost>]

On Sun, 2009-07-26 at 23:29 +0100, Andrew Smallshaw wrote:

> I have a very similar machine here (170MHz SS5 with CG6) and I
> found I had exactly the same problem. Didn't explore it too much
> because that was a machine that I _needed_ to get working, so I
> stopped fussing and went back to 4.0 which I knew works perfectly.
> I did notice a couple of things that may be useful though.
> 
> Firstly, just before the hang the last character output is slightly
> misaligned.  Not by a whole character cell - from memory the text
> was a pixel or two higher than it was supposed to be.  That suggests
> to me that possibly address arithmetic is going slightly awry
> somewhere.

Good eye!  I think I'm seeing something similar.

I tried another experiment with a slightly more interesting text pattern
than "XXXXX...".  Using a repeating "123456789.", with four lines of 60
characters, the first line is fine, the second line is garbled, and the
third and fourth lines are intact.  There's a shell prompt after this,
but no more response to keyboard input.

The garble is interesting:

1234569.123456789.123456789.9.12345678*

Notice missing digits near the beginning.  At "9.9.", the second "9."
and the remainder of the line look like they are displaced one pixel
down.  The "*" represents a black cursor rectangle; there is another
cursor after the shell prompt.

> The other thing I noticed was that it happens much more quickly if
> the mouse is not connected.  I am hesitant to state that there is
> a definite relationship between the two factors but it seemed to
> happen too consistently to be purely by chance.  This is using a
> type 4 keyboard and mouse BTW.

This is so quickly reproducible for me, I haven't tried the "no mouse"
test.

> The only other relevant factor I can think of is the screen resolution
> - mine is set to 1024x768x60.  That may be relevant if it is that
> addresses within the framebuffer are being miscalculated.  Are you
> using some other resolution Bill?

I've left it at the default (which I think is something like
1152x864x76).  I'm new to both SPARC and NetBSD, so I'm trying not to
mess with too many things at once.

Given that an extremely similar issue has been reported on another 170
MHz SS5, I'm starting to wonder if perhaps it's some sort of race
condition that's evident with the faster CPU.  In another response that
seems possibly relevant:

On Sun, 2009-07-26 at 19:40 -0400, der Mouse wrote:
> I _have_ had cg6s wedge, when software doesn't drain the command queue
> often enough. 

Are there any cg6 experts that would like to comment on this?