On Sun, Nov 20, 2005 at 05:23:16PM -0600, Bob Brose wrote:
> It seems like the adb gets hung. When it gets stuck, last time was on the "n"
> key, no other key makes it change, it keeps hammering out n's. When this
> happens, I can still access the machine via the ethernet. When I try a reboot
> (which appears to rely upon adb) the machine is unable to restart, it
> hangs at the point at which it's supposedly going to do a HW reboot.

Yes, that's a full ADB hang. All but the oldest models do reboot and
shutdown using commands to the ADB controller. The driver probably
got into a bad state and lost track of everything. About all you can
do at that point is a hardware reset using key combos or a programmer's
switch reset button.

> Now I tried a bunch of combinations of kernels and devices and made the
> following observations:
>
> With the b and c patches mentioned previously installed, if adb is
> DISABLED esp and sonic work ok. sg_verify suceeds on the whole disc.
>
> If adb is enabled, sonic works ok but esp gets the selection errors and
> fails the sg_verify randomly. ADB itself works with the keyboard but is
> very flakey.
>
> With ADB enabled, If the kernel is compiled with gcc2.95 on OSX, there are
> about 100 times as many esp select errors than there is if the same kernel
> is compiled with gcc 3.3.5 on the performa itself.
>
> If I enable the "alternate irq mapping" by adding the 575 to the
> table of supported machines in arch/m68k/mac/via.c I get MUCH better
> behavior out of ADB keyboard. It SEEMS like I can't make it fail. Sonic
> still works but esp throws many selection errors and fails right
> away on sg_verify. The modified alt irq mapping code:

This suggests to me that interrupts are getting lost. With the alternate
IRQ mapping, the priorities are all different, so things are a lot less
likely to overlap and cause problems. I'll take a look at a few parts
of your two kernels to see if they compile the interrupt code or the esp
driver in a different way that might explain problems like this. I'll
hopefully get back to you on this by the end of the week.

I suppose we should commit the patch to add this model to the alt irq
mapping list since it does seem to work and make things better.

Brad Boyer
flar@...

_______________________________________________
linux-mac68k mailing list
linux-mac68k@...
http://lists.purplehat.net/listinfo/linux-mac68k