Michael Kelly, le mer. 21 janv. 2026 13:15:12 +0000, a ecrit:
> > On 13/01/2026 18:19, Samuel Thibault wrote:
> > > which makes sure to put the content of save_data on the stack instead of
> > > in xmm0, while calling mach_msg().
> > >
> > > So we have something that clobbers xmm0 somehow, and this is only a dumb
> > > workaround that only avoids the issue in the notable but still
> > > particular case of blocking RPCs.
> >
> > This description might also apply to the similar situation that I am
> > seeing with the stress-ng tests. The asm generated from the stress-ng
> > code in function stress_vm_incdec makes use of a number of %xmm
> > registers.
>
> Has this issue with signal handling been resolved?
No, and I don't know if it's really related to signal handling actually,
it could also be user/user or user/kernel context switch management in
mach.
> I've seen some fixes in glibc (eg. hurd/x86_64/htl/pt-machdep.c)
Which fix are you thinking about? The only related fix is
e539a269990dac3ff4d2432c0eb6966a5ee4f274
("hurd: Fix sigreturn clobbering some xmm registers")
which should help, but does not completely fix the issue.
Samuel
