Re: Per-cpu IDT patch

On Sat, Jul 4, 2020 at 6:32 PM Manuel Bouyer <[email protected]> wrote:
>
> On Fri, Jul 03, 2020 at 02:11:43PM +0900, s ymgch wrote:
> > Hi all,
> >
> > I tried to make an IDT per-cpu. The patch is here:
> > http://www.netbsd.org/~yamaguchi/pcpu-idt.patch
> >
> > The summary of modifications is as follows.
> > - Add an IDT pointer and idt_allocmap into struct cpu_info
> > - Change prototype of cpu_init_idt()
> >    - void => struct cpu_info*
> >    - for loading an IDT address in struct cpu_info*
> > - Copy a part of IDT entries from cpu 0 at attaching a CPU
> >    - for the same entries, for example, exceptions, db, etc.
> >
> > This function can be enabled by PCPU_IDT option that is newly provided.
> >
> > Does this patch look good?
> > Please let me know if you have any comments about this.
>
> I guess idt_vec_init_cpu_md() is not used by XENPV at all ?
> If so I guess you can remove #ifndef XENPV here
> If is it used I think it can not work without setting the idt back to read-only
> anyway.

The function is used even by XENPV to initialize iv_idt for cpu0.
I guess the reason why it can be working without the setting in init_x86_64() is
that the same setting is done in set_idtgate().
I'm going to remove #ifndef XENPV by another commit if XEN3_DOM0 and
XEN3_DOMU kernel can work without it.

>
> Othervise I don't see problems at first glance. I tested an amd64 domU in
> PV, PVH and PVHVM mode.

Thank you very much for your test.

Best regards,
-- yamaguchi