riscv uses the .rela.dyn section to relocate the kernel at runtime but
that section is stripped from vmlinux. That prevents kexec to
successfully load vmlinux since it does not contain the relocations info
needed.

Fixes: 71d815bf5dfd ("kbuild: Strip runtime const RELA sections correctly")
Signed-off-by: Alexandre Ghiti <alexghiti@rivosinc.com>
---
 scripts/Makefile.lib | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/scripts/Makefile.lib b/scripts/Makefile.lib
index cad20f0e66ee..0a1f1e67a0ed 100644
--- a/scripts/Makefile.lib
+++ b/scripts/Makefile.lib
@@ -377,7 +377,7 @@ quiet_cmd_objcopy = OBJCOPY $@
 cmd_objcopy = $(OBJCOPY) $(OBJCOPYFLAGS) $(OBJCOPYFLAGS_$(@F)) $< $@
 
 quiet_cmd_strip_relocs = RSTRIP  $@
-cmd_strip_relocs = $(OBJCOPY) --remove-section='.rel*' $@
+cmd_strip_relocs = $(OBJCOPY) --remove-section='.rel*' --remove-section=!.rela.dyn $@
 
 # Gzip
 # ---------------------------------------------------------------------------
-- 
2.39.2

On Thu, 3 Apr 2025 at 16:42, Alexandre Ghiti <alexghiti@rivosinc.com> wrote:
>
> riscv uses the .rela.dyn section to relocate the kernel at runtime but
> that section is stripped from vmlinux. That prevents kexec to
> successfully load vmlinux since it does not contain the relocations info
> needed.
>

Maybe explain that .rela.dyn contains runtime relocations, which are
only emitted if they are actually needed - as opposed to the static
relocations that are not emitted as SHF_ALLOC sections, and are not
considered to be part of the runtime image in the first place. It
would be nice if we could use --remove-relocations= here, which only
removes static relocations, but unfortunately, llvm-objcopy does not
support this.

Also, I wonder if this should apply to all of .rel.dyn, .rela.dyn and
.relr.dyn, as they all carry runtime relocations.

Finally, I'd be curious to know why RISC-V relies on --emit-relocs in
the first place? Is the relocs check really needed? If not, it would
be a nice opportunity to get rid of Makefile.postlink entirely.

In any case, for this change, or a variation along the lines of what I
wrote above,

Acked-by: Ard Biesheuvel <ardb@kernel.org>


> Fixes: 71d815bf5dfd ("kbuild: Strip runtime const RELA sections correctly")
> Signed-off-by: Alexandre Ghiti <alexghiti@rivosinc.com>
> ---
>  scripts/Makefile.lib | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/scripts/Makefile.lib b/scripts/Makefile.lib
> index cad20f0e66ee..0a1f1e67a0ed 100644
> --- a/scripts/Makefile.lib
> +++ b/scripts/Makefile.lib
> @@ -377,7 +377,7 @@ quiet_cmd_objcopy = OBJCOPY $@
>  cmd_objcopy = $(OBJCOPY) $(OBJCOPYFLAGS) $(OBJCOPYFLAGS_$(@F)) $< $@
>
>  quiet_cmd_strip_relocs = RSTRIP  $@
> -cmd_strip_relocs = $(OBJCOPY) --remove-section='.rel*' $@
> +cmd_strip_relocs = $(OBJCOPY) --remove-section='.rel*' --remove-section=!.rela.dyn $@
>
>  # Gzip
>  # ---------------------------------------------------------------------------
> --
> 2.39.2
>

Hi Ard,

On 03/04/2025 17:11, Ard Biesheuvel wrote:
> On Thu, 3 Apr 2025 at 16:42, Alexandre Ghiti <alexghiti@rivosinc.com> wrote:
>> riscv uses the .rela.dyn section to relocate the kernel at runtime but
>> that section is stripped from vmlinux. That prevents kexec to
>> successfully load vmlinux since it does not contain the relocations info
>> needed.
>>
> Maybe explain that .rela.dyn contains runtime relocations, which are
> only emitted if they are actually needed - as opposed to the static
> relocations that are not emitted as SHF_ALLOC sections, and are not
> considered to be part of the runtime image in the first place.


Ok I'll do.


> It
> would be nice if we could use --remove-relocations= here, which only
> removes static relocations, but unfortunately, llvm-objcopy does not
> support this.
>
> Also, I wonder if this should apply to all of .rel.dyn, .rela.dyn and
> .relr.dyn, as they all carry runtime relocations.


Ok, I'll add them to the next version.


>
> Finally, I'd be curious to know why RISC-V relies on --emit-relocs in
> the first place? Is the relocs check really needed? If not, it would
> be a nice opportunity to get rid of Makefile.postlink entirely.


So I had to check and it happens that this was an issue with the 
toolchain, I should check if that still happens with newer ones.

commit 559d1e45a16dcf1542e430ea3dce9ab625be98d0
Author: Alexandre Ghiti <alexghiti@rivosinc.com>
Date:   Wed Mar 29 06:53:29 2023 +0200

     riscv: Use --emit-relocs in order to move .rela.dyn in init

     To circumvent an issue where placing the relocations inside the init
     sections produces empty relocations, use --emit-relocs. But to avoid
     carrying those relocations in vmlinux, use an intermediate
     vmlinux.relocs file which is a copy of vmlinux *before* stripping its
     relocations.

     Suggested-by: Björn Töpel <bjorn@kernel.org>
     Suggested-by: Nick Desaulniers <ndesaulniers@google.com>
     Signed-off-by: Alexandre Ghiti <alexghiti@rivosinc.com>
     Link: 
https://lore.kernel.org/r/20230329045329.64565-7-alexghiti@rivosinc.com
     Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>


>
> In any case, for this change, or a variation along the lines of what I
> wrote above,
>
> Acked-by: Ard Biesheuvel <ardb@kernel.org>


Thanks,

Alex


>
>
>> Fixes: 71d815bf5dfd ("kbuild: Strip runtime const RELA sections correctly")
>> Signed-off-by: Alexandre Ghiti <alexghiti@rivosinc.com>
>> ---
>>   scripts/Makefile.lib | 2 +-
>>   1 file changed, 1 insertion(+), 1 deletion(-)
>>
>> diff --git a/scripts/Makefile.lib b/scripts/Makefile.lib
>> index cad20f0e66ee..0a1f1e67a0ed 100644
>> --- a/scripts/Makefile.lib
>> +++ b/scripts/Makefile.lib
>> @@ -377,7 +377,7 @@ quiet_cmd_objcopy = OBJCOPY $@
>>   cmd_objcopy = $(OBJCOPY) $(OBJCOPYFLAGS) $(OBJCOPYFLAGS_$(@F)) $< $@
>>
>>   quiet_cmd_strip_relocs = RSTRIP  $@
>> -cmd_strip_relocs = $(OBJCOPY) --remove-section='.rel*' $@
>> +cmd_strip_relocs = $(OBJCOPY) --remove-section='.rel*' --remove-section=!.rela.dyn $@
>>
>>   # Gzip
>>   # ---------------------------------------------------------------------------
>> --
>> 2.39.2
>>
> _______________________________________________
> linux-riscv mailing list
> linux-riscv@lists.infradead.org
> http://lists.infradead.org/mailman/listinfo/linux-riscv