Re: Xarm32VIDC from 1.6.1 very unstable

> In article <[email protected]> you write:
> >First, is still somebody actively maintaining the acorn32 port of NetBSD?
> 
> When in breaks for me, I try to fix it.  I occasionally look at PRs and try
> to fix them.  
> 
> >I installed the 1.6.1 release on a RiscPC/SA110/32 MB RAM and was very
> >impressed as it run out-of-the box, all hardware was detected and my
> >subjective impression of the performance is very good. Good work!
> >
> >However the X-server seems very unstable. As long as I use only the VIDC
> >console or only console via Xterm NetBSD runs stable like a rock. But
> >graphical interaction (resizing windows, opening windows, using a
> >webbrowser or a window manager) leads to a core dump of Xarm32VIDC sooner
> >or later.
> 
> Hmm.  Are you using a wscons-capable kernel?  I use Xarm32VIDC on an NC on a
> daily basis, and it's really quite solid (until a run the machine out of
> swap, whereupon the X server tends to get killed and restarted).
> 
> >- Is 1.6.1 and especially Xarm32VIDC ought to be stable?
> 
> I can't actually speak for 1.6.1, since my NC runs 1.6A at the moment (the
> standard 1.6 wscons kernels missed out opms, which was fatal to X, and I
> haven't got around to looking at 1.6.1.

I found 1.6.1 to be unstable on my RISC PC -- There were some kernel 
issues during the 1.6-1.6.1 transition that caused user-land instability, 
that I thought I'd managed to back out before the release occurred.  It 
may turn out that there is still some other underlying problem.... ;-(

Having said that, -current isn't much better.  Maybe it is a real bug in 
the X server...

> 
> >- Can netbsd cope with StrongARMs that suffer from the STM ^ bug?
> 
> I believe so.  As I understand it, this bug only bites when saving user-mode
> registers on entry to the kernel, and NetBSD's kernel entry sequence
> carefully avoids triggering the bug.

Processors suffering from that problem also suffer from a user-space ldm 
bug that affects returning from a function call if the ldm is the last 
instruction on a page.  There's a package in the package system 
(sysutils/fix4SA110rev2) that can be used to help to alleviate this, but 
I'm not sure if it's been updated for ELF images.

> 
> >- Is there an issue withloading coredumps into gdb? Whenever I tried this,
> >  gdb says it does not understand the dump's fileformat.
> 
> I'm sure I _have_ managed to load a coredump into gdb once, but I'm not sure
> it was anywhere near the 1.6 release.
> 

If you are having problems, you might find a build of the standard GNU 
release of GDB fixes your problems.

> >- I read about an in-kernel debugger ddb - how can it be activated?
> 
> Build a kernel with "options DDB" (I think all the release kernels have it). 
> The debugger will be entered on a panic, and you can drop into it using
> Ctrl+Alt+Esc (wscons kernels) or some other sequence I currently forget
> (non-wscons kernels).
> 

You will have problems if you are trying to use the console for X and you 
don't set up your system to have the "debug console" on a serial port, 
since I don't believe the current WSCONS code on acorn32 supports 
switching.

R.