Re: 10.0 crashes on leaving X

Hello,

On Mon, 22 Apr 2024 20:44:11 -0400
vom513 <[email protected]> wrote:

> Hello all,
> 
> I am testing 10.0 on an SS10, SMP (2x SM61), TGX+ (4mb) FB.
> 
> When the machine is running everything seems fine - nothing bad in dmesg etc.  
> 
> However upon shutdown - there is a kernel panic and the machine reboots (and has to fsck).

Hmm, I don't have an ss10, what I have here is an ss20 with 2x sm81 and
cg14/SX. Somewhere I have at lease one 4MB cg6 if that's where the
error comes from.

> 9.3 seems to run fine on this machine - and does not crash when exiting X.

Does it happen when exiting X, when rebooting from X, or powering off
from X? These are all separate cases that take different code paths.
If you can reproduce it without powering down then the panic output
should still be in the message buffer on reboot.

> I can try to take a video on shutdown and perhaps we can capture some
> info to give some clues.  Very difficult to capture this because it’s
> happening right before power off (assuming shutdown -p).  So nothing
> is getting logged other than the panic scrolling on the screen before
> the machine resets.

I'd try to reproduce it without powering down first.

> I also remember a post on port-sparc a ways back talking about
> fixes/enhancements to X / FB bits on sparc coming in 10 ?

Yes, cg14/SX gained proper 8bit acceleration, for people with 4MB cg14
who prefer higher resolutions over colour depth, some xrender
improvements, and cg6 got some attention as well.

> On that note - one thing I notice is that in 9.3 my FB is reported in dmesg
> as being 2 MB as opposed to 4 MB in 10.0.  My understanding is that
> the card is 4 MB hardware but the way it’s used it’s 2 MB in practice
> ? (I’m probably mangling that explanation).  Just caught my eye as a
> delta from 9.3 vs. 10.0.

CG6 with double buffering ( that is, (t)gx+ boards and possibly LX with
VSIMM ) divide the available memory into two independent framebuffers,
with bits to select which to read from, which to display and which to
write into ( can be both ).
We only ever used one of these buffers since the docs and available
driver code implied that there is only a 2MB window through which we
write into buffer A, B, or both. Turns out the window is bigger and we
can in fact access all framebuffer memory as buffer A, and so does the
blitter.
With that we now use all available video memory.

> Let me know if there’s anything specific I need to try - and if I
> should open a PR.

I dimly remember a thread about ss10s crashing on shutdown, with or
without X, from a while ago.
So, any chance you could try a -current kernel?

have fun
Michael