Re: UNS: Re: netbsd-5.1 cpu probe/attach for 11/7xx?

[Johnny Billquist <bqt%softjar.se@localhost>]

On 2011-06-27 13.18, Brad Parker wrote:
> Johnny Billquist wrote:
> > Older VAXen don't really have a CPU attach function. There can only be
> > one CPU, and it is implicitly already attached.
> I don't think that is how the code is working; at least in the abstract.
> There are cpu attach routines for
> all the 7xx cpu's, but none of them are being called. At the very least
> they print a small amount of information
> which is interestng to see -
>
> I'm proposing a simple fix to make that code (which once worked) work
> again.

Ok. I had to check things through a little more to refresh my memory.
Yes, there is a cpu_attach function for all cpus. But as far as I can 
tell, it is also called for all CPUs.

It's done in autoconf.c at line 186:

        if (dep_call->cpu_attach_cpu != NULL)
                (*dep_call->cpu_attach_cpu)(self);

Not that this attach function do much. On a 11/780, it prints some CPU 
information, and enables the FPA if it exists (but that is a pretty good 
thing to do :) ).

Sos it's not really called through the normal autoconf paths using the 
config file.

> > The system starts by trying to figure out what general type of system
> > it is (checking the SID), and then just start initializing pmap, and
> > all other stuff from there.
> >
> > Check locore.c, which have the initial CPU type setup code.
> Yes, thanks. I'm pretty familiar with locore.c and how it id's the cpu.
>
> I also might argue that the 11/730 cpu is directly connected to the unibus.

Well, the 11/730 also have a Unibus map, and the memory does not sit on 
the Unibus. And the Unibus exists within a specific memory range in the 
CPU addressing. Other addresses goes to other places. So I'd definitely 
say no - the CPU does not sit directly on the Unibus.
(Heck, the Unibus can only address 256Kbyte, that would be rather severe...)

> So the practical question is - fix the broken cpu probe code? other
> architectures have similar
> cpu probe code, even in UP cases where the bus structure is simple. It
> seems to make sense to
> fix it, or if not, remove it. The code in locore.c simply id's the cpu,
> it does not do any cpu specific
> work or print anything helpful.

True. locore sets up the CPU information. autoconf then calls the actual 
attach function. Not through some probing, but directly, as the CPU is 
already known to exist, and we already know which type.

        Johnny