[edk2-devel] [PATCH] BaseTools/GenFw: Change opcode when converting ADR to ADRP

[Pedro Falcato]

On Fri, Oct 27, 2023 at 2:47 PM Ard Biesheuvel <ardb at kernel.org> wrote:
>
> Hello all,
>
> Apologies for the late response.
>
> On Fri, 27 Oct 2023 at 14:44, Jake Garver via groups.io
> <jake=nvidia.com at groups.io> wrote:
> >
> > Thanks for your response, Pedro.
> >
> > I chased this as a toolchain bug originally, but concluded that the ADR indeed works before GenFw rewrites it.  But I see your point regarding the relocation statement.
> >
> > As requested, below is the disassembled function along with relocations.  This was generated from the dll using "aarch64-linux-gnu-objdump -r -D".  The ADR in question is at 2ffc.
> >
>
> Can you double check the object file? I suspect this is a linker
> relaxation not a compiler issue.
>
> > This code was generated by the current version of the GCC 10.x toolchain on Ubuntu20.  So, if we're concluding this is a toolchain issue, then it's with a fairly "stock" toolchain.
> >
> > 0000000000002fec <fdt_path_offset>:
> >     2fec:   a9b97bfd stp   x29, x30, [sp, #-112]!
> >     2ff0:   910003fd mov   x29, sp
> >     2ff4:   a90363f7 stp   x23, x24, [sp, #48]
> >     2ff8:   aa0003f8 mov   x24, x0
> >     2ffc:   10020020 adr   x0, 7000 <_cont+0xe98>
> > 2ffc: R_AARCH64_ADR_GOT_PAGE  __stack_chk_guard
>
> The nasty thing with relying on --emit-relocs is that they get out of
> sync. R_AARCH64_ADR_GOT_PAGE is documented as applying to ADRP only,
> so this code is non-compliant one way or the other.

I was narrowing it down to this, but _ADR_GOT_PAGE does not seem to be
relaxable as ADRP -> ADR, per the ABI spec (see "Large GOT
Indirection". "PC-relative addressing" only applies to
_ADR_PREL_PG_HI21...)
Maybe it's just not a documented relaxation? Or does it relax as
_ADR_GOT_PAGE -> _ADR_PREL_PG_HI21 -> _ADR_PREL_LO21 without updating
internal relocation info?

>
> There are other relaxations that also confuse GenFw when the static
> relocs don't get updated accordingly.

Wonderful that you can confirm this is probably a linker --emit-relocs
issue. So, if this proves to be the case:
Acked-by: Pedro Falcato <pedro.falcato at gmail.com>

but please rewrite the commit message so that it's clear that this is
an --emit-relocs toolchain bug.

-- 
Pedro


-=-=-=-=-=-=-=-=-=-=-=-
Groups.io Links: You receive all messages sent to this group.
View/Reply Online (#110209): https://edk2.groups.io/g/devel/message/110209
Mute This Topic: https://groups.io/mt/102202314/1813853
Group Owner: devel+owner at edk2.groups.io
Unsubscribe: https://edk2.groups.io/g/devel/unsub [edk2-devel-archive at redhat.com]
-=-=-=-=-=-=-=-=-=-=-=-