> I think I have narrowed down the source of the SCSI phase problems but I
> don't think it makes any sense:
> 
> A fresh build (config + make depend;make) builds the same source without
> spurious SCSI phase problems. However, one to some small n number of builds
> make kernels that boot fine but exhibit spurious SCSI phase change problems
> right about when multiuser mode gets switched on. Any idea what the deal with
> this is? The development cycle is slow enough without having to make clean
> and recompile overnight for every test compile :)

So if you build an object file and use it immediately, you're fine. But if
you let it sit a while, you will get a kernel which has phase errors?

Something sounds wrong. Way wrong. That would mean the file system is getting
corrupted as time goes by. Or your make depend;make is perturbing the drive
to behave.

Boot into single user mode and do an fsck -f (for force). See if the fs
is ok. If it's marked clean, it will get pulled up even w/ errors. My
ppp problems arose when I had hidden fs damage. One time on shutdown,
stuff freaked out (it's only happened once in almost 2 years, so it's rare).
fsck -f helped.

Another option would be to keep some of the sick kernels around and see
if you can:

boot "sick" one, and get errors
boot "good" one and not get errors
boot "sick" one and get errors

The idea is to make sure that there's not some goofy way a make is getting
the drive to behave.

Or when you have a sick kernel, copy the compile dir, and make a new one.
Then do a binary diff (won't print how files differ, just if they do).

Weird.

Good luck!

Bill