Re: Report on NetBSD-CURRENT and `hpcmips` on MobilePro 780

On Mon, 18 Dec 2023 at 13:25, Alexander Bochmann <[email protected]> wrote:

> ...on 2023-12-14 21:27:43, David Brownlee wrote:
>
>  > Foo... So it would be back to bisecting builds to work out where
>  > regressions were introduced...
>
> I'm really not sure if it's worth the effort with the low number
> of people interested, and the increasingly brittle hardware
> platforms. At least with the z50, it also doesn't seem possible
> to automate testing - I usually need to use the reset button or
> even remove the battery to get the system back into Windows CE
> after booting a test kernel.
>

It's down to the intersection of people with hardware and interest.
NetBSD/vax has been languishing for a few years, able to run mostly but due
to compiler bugs not able to build many packages or even build itself
natively. Thanks essentially to Kalvis Duckmanton NetBSD-10/vax will be
finally able to build itself natively again. It can happen :)

Also, as more people write retro emulators that can massively help the
debug cycle.


> It certainly would be nice to find out what the problem with
> segfaulting binaries is nevertheless... I mean, the kernel itself
> doesn't seem to be affected by that, even though it's built with
> the same compiler - but I think it uses floating point emulation?
> That's why I used to assume it's something with the FPU code
> somehow...
>

It should be possible to build a soft-float userland which avoids FPU
entirely. If the relevant issues go away then that isolates it to an FPU
issue

David