[External] Re: [PATCH 1/3] Revert "riscv/efistub: Ensure GP-relative addressing is not used"

[yunhui cui]

Hi Ard,

On Wed, Mar 6, 2024 at 9:11 PM Ard Biesheuvel <ardb at kernel.org> wrote:
>
> On Wed, 6 Mar 2024 at 14:08, yunhui cui <cuiyunhui at bytedance.com> wrote:
> >
> > Hi Jan,
> >
> > On Wed, Mar 6, 2024 at 8:52 PM Jan Kiszka <jan.kiszka at siemens.com> wrote:
> > >
> > > On 06.03.24 09:56, Yunhui Cui wrote:
> > > > This reverts commit afb2a4fb84555ef9e61061f6ea63ed7087b295d5.
> > > >
> > >
> > > This comes without a reason - which is likely something around "will fix
> > > this properly later". But then you regress first and only fix
> > > afterwards. Can't that be done the other way around?
> >
> > Sorry, I don't quite understand what you mean. Can you help explain it
> > more clearly? Do you mean "delete mno-relax instead of revert
> > directly?"
> >
>
> You should order your patches in a way that does not create
> intermediate states (between 1-2 or between 2-3) where the original
> problem is being recreated.
>
> So in this case, you should
> a) fix the issue
> b) revert the existing patches in *opposite* order
Simply, I plan to remove "-mno-relax" and
"\|R_RISCV_$(BITS)\|R_RISCV_RELAX" in the third patch (fix patch).


> However, I don't think the EFI stub can use GP - please refer to my other reply.
The problem we encountered is that gcc 13 will optimize efistub using gp.

Thanks,
Yunhui