Re: mac68k built-in video (PR#58164)

Update:

On 6/4/24 12:12 PM, Stan Johnson wrote:
> Hi David,
> 
> On 6/4/24 9:56 AM, David Brownlee wrote:
>> ...
>>
>> If you have time it would be great to track down the point of breakage.
> 
> I'll see what I can find.

The video issue with the Mac IIci does not appear to be a regression in
the kernel. It seems to be a limitation of the NetBSD Booter.

> 
>>
>> Also the incorrect total memory reporting in 58164 is... curious, and
>> potentially indicative of something deeper. Would you have time to
>> build some test kernels to try to get more details out of that? :)
>>
>> David
>>
> 
> Yes, I can build some test kernels, but I would need advice on the
> config files to use, since I'm not a developer.

The incorrect memory reporting is likely caused by the what the NetBSD
Booter is reporting to the kernel. There may still be an issue with the
kernel, but it appears to be avoidable (see below).

> 
> In Problem Report 58164, I had at first installed memory the wrong way
> (64 MiB in Bank A and 16 MiB in Bank B). Some part of the kernel knew
> that I had roughly 80 MiB installed (as reported in /proc/meminfo and
> "available memory" in dmesg), but the kernel was also throwing away all
> the memory in Bank A (64 MiB) so another part of the kernel thought my
> total memory was only the 16 MiB that it saw in Bank B. The ideal
> configuration (in NetBSD and Linux) when using built-in video on the
> IIci is to have just 1 MiB in Bank A and 64 MiB in Bank B.
> 

I've done some additional testing. Even NetBSD 1.6.2 fails with 1 MiB in
Bank A and 64 MiB in Bank B, with the error that it can't work with the
memory mapping. However, putting 64 MiB in both banks works with NetBSD
1.6.2 (and 10.0) using Nubus video:

# dmesg
...
[     1.000000] NetBSD 10.0 (MACIICI) #0: Sat Jun  8 09:21:42 MDT 2024
...
[     1.000000] Apple Macintosh IIci  (68030)
[     1.000000] cpu: delay factor 266
[     1.000000] fpu: mc68882
[     1.000000] total memory = 128 MB
[     1.000000] avail memory = 122 MB
...

I don't have a IIsi, so I don't know whether the NetBSD kernel would
panic when a Nubus video card is used. The IIsi is limited to 1 MiB in
its Bank A (soldered in), but I don't know whether a "full" Bank A would
still be 64 MiB.

Anyway, having 64 MiB in both banks A and B is a good workaround to run
NetBSD on a IIci using an unaccelerated Nubus video card, and that also
works for Mac OS and Linux. And it's a good use for a Toby Frame buffer
card; those are still relatively inexpensive, though collectors are
closing in.

NetBSD 10.0 still ignores all the memory in Bank A if RBV is used:

# dmesg
...
[     1.000000] NetBSD 10.0 (MACIICI) #0: Sat Jun  8 09:21:42 MDT 2024
...
[     1.000000] Apple Macintosh IIci  (68030)
[     1.000000] cpu: delay factor 266
[     1.000000] fpu: mc68882
[     1.000000] total memory = 65536 KB
[     1.000000] avail memory = 122 MB
...


My disk layout has "/" within the first 1 GiB of the boot disk, with a
separate partition for "/usr". This is needed to ensure that the NetBSD
Booter can find the kernel within the first 1 GiB of the boot disk.

I'll be happy to test other configurations or updates to the NetBSD
Booter. I'm looking into the NetBSD Booter and Mkfs source code, but
it's not trivial, at least for me.

BTW, the Traditional Installation Method works great for installing
NetBSD 1.6.2 on a SCSI disk connected to a 400 MHz PowerBook Lombard
running Mac OS 9, taking just a fraction of the time that a IIci would
use for a comparable sysinst. So I look forward to helping get the
NetBSD Installer, NetBSD Booter, and Mkfs updated to support NetBSD 10.0.

-Stan