Re: Some fixes for m68k users (incl/ for FPU emulation)

[Romain Dolbeau <romain%dolbeau.org@localhost>]

Le dim. 1 nov. 2026 à 13:54, Jason Thorpe <thorpej%me.com@localhost> a écrit :
> It seems to me that you’re starting from the NetBSD 11 code base

Oups, sorry, should have said that; I used github's netbsd-11 tag as
the baseline, so I'd have something reasonably stable. That was also
used as the baseline for virt68k-smp, hence all the work done on
"pmap_motorola.c"

I haven't looked at pmap_68k so it might be OK already; please make
sure U0/U1 aren't used for internal purposes - they are exposed on
external pins so could have an impact on the hardware (even if in r
practice almost nobody used them), and they are needed to "properly"
support SMP on MC68040 (they are used to indicate the level of type of
snooping to use for the data in that page).

> It’ll take me a little more time to process those

There was no request to do anything about any of it in a hurry,
implied or otherwise! On my side, I was just mentioning (alleged)
issues the LLM I used encountered and "fixed" along the way [with
"fixed" meaning "the tests no longer fails", which isn't a very high
standard...]. FPU_EMULATE is a useful tool, but in real life probably
only used by people with soldered xC68LC040 that can't be upgraded to
full xC68040 and that aren't XC68LC040. Most MC68030 systems had a
socket for a MC68881/2. And as mentioned using soft-float userland is
the right option IMHO.

OTOH, FPU_EMULATE is quite useful to debug $4 frames on 'LC040 clones :-)

Cordially,

-- 
Romain Dolbeau