Re: Serious breakage in latest PowerMac snapshot

[mycroft%mit.edu@localhost (Charles M. Hannum)]

Tsubai Masanari <tsubai%iri.co.jp@localhost> writes:

> >* The POWERMAC kernel hangs during configure(), after printing:
> >
> >  adb0 at obio0 offset 0x16000 irq 18
> 
> Please apply following patch.
> (I don't know 1ms is appropriate:-)
> 
> 
> *** adb_direct.c      Thu Mar  5 19:02:59 1998
> --- /a/src/sys/arch/powermac/dev/adb_direct.c Thu Mar  5 19:02:23 1998
> ***************
> *** 1256,1261 ****
> --- 1254,1261 ----
>   
>       adb_hw_setup();         /* init the VIA bits and hard reset ADB */
>   
> +     DELAY(1000);
> + 
>       /* send an ADB reset first */
>       adb_op_sync((Ptr) 0, (Ptr) 0, (Ptr) 0, 0x00);

That seems to fix the ADB hangs, as far as I can tell.

BTW, Frank and I noticed that at least part of the reason for my NFS
problems was that splsoftnet() was not blocking softclock().  I've
hacked on the interrupt handling code a little (and in particular
fixed this), and I'll send out diffs in a bit.

This still leaves a few of problems:

* When using the `POWERMAC' kernel, the serial ports just don't work
any more.  They seem to output, but they don't accept any input.

* The `OFCONS' kernel just doesn't talk to my console device any more.
This is pretty strange.  I'm not sure what changed that would cause
it.

* I still get `panic: receive 1' fairly often when using NFS.  I
believe this is due to bugs in the MACE driver.

* The MACE driver cannot handle back-to-back packets.  This causes NFS
with a large block size to lose, and causes abysmal TCP performance in
general.  I also get an occasional `mc0: receive FIFO overflow' and
`mcwatchdog: resetting chip'.

* With the BeBox executables in your disk image, I got an occasional
semi-random core dump during a `make build' that went away when I
reran the failed command.  This problem *seems* to have gone away when
I rebuilt the libraries, but I wouldn't say this is conclusive yet.
It may be that the PowerPC code was not correctly updated for Jason's
vfork(2) changes.

In short, the main problems are:

* The MACE driver is flaky.

* The console handling still needs work.

But, all told, I can get it to successfully boot if I sacrifice a few
chickens.  B-)