Hello,

> > Since there's no L2CR register, either it's always on or configured by
> > Open Firmware. It's outside what the kernel can do.
> 
> ok, so I should just get the cache.. wonder if I can check that somehow.
There are a few memory-bound benchmarks in pkgsrc/benchmarks, you should see a significant drop with datasets bigger than your cache if it's active.

> I get whatever the default for a GENERIC kernel is. Anyway this is not
> the failures I had when I had sync problems...
> Also, it was a problem of a drive not disabling sync if it was requested
> on the other box. Here I hope everything is normal.
Well, do you see the probe(...) messages? If not then the kernel didn't negotiate sync. GENERIC should default to async (well, by kernels don't but those you built should, at least if you left the flags for mesh on 0xffff ).
A problem could be that your disk tries to negotiate sync on its own - could you try another drive which doesn't do that? I doubt it's sync-related but you never know until you try.

have fun
Michael