Re: Patch: hppa lazy FPU switching and QEMU bugs

[Nick Hudson <nick.hudson%gmx.co.uk@localhost>]

On 18/08/2026 23:05, Thor Lancelot Simon wrote:
> On Tue, Aug 18, 2026 at 04:26:21PM -0400, Thor Lancelot Simon wrote:
> > On Tue, Aug 18, 2026 at 06:09:20PM +0100, Nick Hudson wrote:
> > > On 18/08/2026 00:11, Thor Lancelot Simon wrote:
> > > > On Mon, Aug 17, 2026 at 06:27:43PM +0100, Nick Hudson wrote:
> > > > > It doesn't quite work - I saw an unexpected core dump and fp exception. Will investigate more later...
> > > >   From a user process, or of the kernel itself?
> > > Only userland problems.
> > >
> > > I'm happy to take a diff the just does eager.
> > I am still just barely gaining comfort with the architecture and the
> > asm syntax and I think that'd take a little more restructuring in
> > trap and locore to be what I'd call a clean finished result.  But
> > I don't mind poking at it later...
> >
> > I think I see the problem with eager mode.  I was testing it only
> > with the workload that caused me problems with lazy, namely,
> > integer multiplies!  So I was getting away with not properly
> > handling the FPU status register, which the lazy path kind of
> > does as a side effect (really, the lazy path is nothing but
> > side effects, depending how you squint at it).
> >
> > I'm fixing.  Will send another patch.
> Try the below - hopefully this is it.

[snip]

Actually setting ci_fpu_lwp to something other than NULL fixed it :)

I placed the diff I tested on real hw here:

    https://www.netbsd.org/~skrll/hppa.fp.diff

I tested with HPPA_FPU_EAGER and without.

I tried to test with qemu as well, but I kept getting scsi timeouts and gave up.