[PATCH] MIPS: VDSO: Avoid including .got in dynamic segment

[PATCH] MIPS: VDSO: Avoid including .got in dynamic segment

[Nathan Chancellor]

After commit 2db1ec80dfd5 ("MIPS: VDSO: Fold MIPS_DISABLE_VDSO into
MIPS_GENERIC_GETTIMEOFDAY"), building ARCH=mips allnoconfig with LLVM=1
shows some warnings from llvm-readelf while checking the VDSO for
dynamic relocations:

  llvm-readelf: warning: 'arch/mips/vdso/vdso.so.dbg.raw': invalid PT_DYNAMIC size (0xa4)
  llvm-readelf: warning: 'arch/mips/vdso/vdso.so.dbg.raw': PT_DYNAMIC dynamic table is invalid: SHT_DYNAMIC will be used

The blamed commit alters the link order of objects into vdso.so.raw,
placing vgettimeofday.o after sigreturn.o. This ultimately results in
the .text section shrinking slightly in size, which in turn changes the
offset of the .dynamic section.

  -  [ 9] .text             PROGBITS        000002f0 0002f0 000930 00  AX  0   0 16
  -  [10] .dynamic          DYNAMIC         00000c20 000c20 000090 08   A  5   0  4
  +  [ 9] .text             PROGBITS        000002f0 0002f0 000924 00  AX  0   0 16
  +  [10] .dynamic          DYNAMIC         00000c14 000c14 000090 08   A  5   0  4

Changing the offset of the .dynamic section causes the dynamic segment
size to grow by the same amount, which triggers a warning in
llvm-readelf because PT_DYNAMIC's p_filesz (0xa4) is no longer a
multiple of its sh_entsize (8):

  -  DYNAMIC        0x000c20 0x00000c20 0x00000c20 0x00098 0x00098 R   0x10
  +  DYNAMIC        0x000c14 0x00000c14 0x00000c14 0x000a4 0x000a4 R   0x10

The size of the dynamic segment was already incorrect before the blamed
comment, as it should be 0x90 like the .dynamic section above (18
entries at 8 bytes per entry); it just so happens that 0x98 % 8 is 0,
whereas 0xa4 % 8 is 4, so there was no warning.

Looking at the section to segment mapping of the dynamic segment reveals
that it includes the .got section, as it is implicitly placed after
.dynamic by ld.lld's orphan section heuristics and inherits its segments
from the linker script.

  [ 9] .text             PROGBITS        000002f0 0002f0 000924 00  AX  0   0 16
  [10] .dynamic          DYNAMIC         00000c14 000c14 000090 08   A  5   0  4
  [11] .got              PROGBITS        00000cb0 000cb0 000008 00 WAp  0   0 16

  Section to Segment mapping:
   Segment Sections...
    00     .mips_abiflags
    01     .reginfo
    02     .mips_abiflags .reginfo .hash .dynsym .dynstr .gnu.version .gnu.version_d .note .text .dynamic .got
    03     .dynamic .got
    04     .note

Explicitly describe the .got section in the MIPS VDSO linker script
after .rodata, which switches back to the default text segment,
resulting in a dynamic segment that is the exact size of the .dynamic
section as expected with no other layout changes.

  -  DYNAMIC        0x000c14 0x00000c14 0x00000c14 0x000a4 0x000a4 R   0x10
  +  DYNAMIC        0x000c14 0x00000c14 0x00000c14 0x00090 0x00090 R   0x4

  -   03     .dynamic .got
  +   03     .dynamic

Closes: https://github.com/ClangBuiltLinux/linux/issues/2166
Fixes: 2db1ec80dfd5 ("MIPS: VDSO: Fold MIPS_DISABLE_VDSO into MIPS_GENERIC_GETTIMEOFDAY")
Signed-off-by: Nathan Chancellor <nathan@kernel.org>
---
The fixes tag feels a little strong since it seems like it has just been
luck up until this point that there has been no warning but I decided to
be conservative and include it regardless. Feel free to remove it if you
see fit.

I think this should go via timers/vdso with the blamed commit. I plan to
send a follow up series for 7.3 to add '--orphan-handling' to the MIPS
VDSO to avoid issues like this in the future but that can go via the
MIPS tree, as it is not really a fix and I will need to properly test
it.
---
 arch/mips/vdso/vdso.lds.S | 1 +
 1 file changed, 1 insertion(+)

diff --git a/arch/mips/vdso/vdso.lds.S b/arch/mips/vdso/vdso.lds.S
index 05badf3ae0ff..278ab6444e98 100644
--- a/arch/mips/vdso/vdso.lds.S
+++ b/arch/mips/vdso/vdso.lds.S
@@ -56,6 +56,7 @@ SECTIONS
 	.dynamic	: { *(.dynamic) }		:text :dynamic
 
 	.rodata		: { *(.rodata*) }		:text
+	.got		: { *(.got) }
 
 	_end = .;
 	PROVIDE(end = .);

---
base-commit: 13f6218e6fe79dc64aed76d738b765b45f62492b
change-id: 20260608-mips-vdso-fix-section-layout-262bc18d0c29

Best regards,
--  
Cheers,
Nathan

Re: [PATCH] MIPS: VDSO: Avoid including .got in dynamic segment

[Thomas Weißschuh]

On Tue, Jun 09, 2026 at 06:31:21PM -0700, Nathan Chancellor wrote:
> After commit 2db1ec80dfd5 ("MIPS: VDSO: Fold MIPS_DISABLE_VDSO into
> MIPS_GENERIC_GETTIMEOFDAY"), building ARCH=mips allnoconfig with LLVM=1
> shows some warnings from llvm-readelf while checking the VDSO for
> dynamic relocations:
> 
>   llvm-readelf: warning: 'arch/mips/vdso/vdso.so.dbg.raw': invalid PT_DYNAMIC size (0xa4)
>   llvm-readelf: warning: 'arch/mips/vdso/vdso.so.dbg.raw': PT_DYNAMIC dynamic table is invalid: SHT_DYNAMIC will be used
> 
> The blamed commit alters the link order of objects into vdso.so.raw,
> placing vgettimeofday.o after sigreturn.o. This ultimately results in
> the .text section shrinking slightly in size, which in turn changes the
> offset of the .dynamic section.
> 
>   -  [ 9] .text             PROGBITS        000002f0 0002f0 000930 00  AX  0   0 16
>   -  [10] .dynamic          DYNAMIC         00000c20 000c20 000090 08   A  5   0  4
>   +  [ 9] .text             PROGBITS        000002f0 0002f0 000924 00  AX  0   0 16
>   +  [10] .dynamic          DYNAMIC         00000c14 000c14 000090 08   A  5   0  4
> 
> Changing the offset of the .dynamic section causes the dynamic segment
> size to grow by the same amount, which triggers a warning in
> llvm-readelf because PT_DYNAMIC's p_filesz (0xa4) is no longer a
> multiple of its sh_entsize (8):
> 
>   -  DYNAMIC        0x000c20 0x00000c20 0x00000c20 0x00098 0x00098 R   0x10
>   +  DYNAMIC        0x000c14 0x00000c14 0x00000c14 0x000a4 0x000a4 R   0x10
> 
> The size of the dynamic segment was already incorrect before the blamed
> comment, as it should be 0x90 like the .dynamic section above (18
> entries at 8 bytes per entry); it just so happens that 0x98 % 8 is 0,
> whereas 0xa4 % 8 is 4, so there was no warning.
> 
> Looking at the section to segment mapping of the dynamic segment reveals
> that it includes the .got section, as it is implicitly placed after
> .dynamic by ld.lld's orphan section heuristics and inherits its segments
> from the linker script.
> 
>   [ 9] .text             PROGBITS        000002f0 0002f0 000924 00  AX  0   0 16
>   [10] .dynamic          DYNAMIC         00000c14 000c14 000090 08   A  5   0  4
>   [11] .got              PROGBITS        00000cb0 000cb0 000008 00 WAp  0   0 16
> 
>   Section to Segment mapping:
>    Segment Sections...
>     00     .mips_abiflags
>     01     .reginfo
>     02     .mips_abiflags .reginfo .hash .dynsym .dynstr .gnu.version .gnu.version_d .note .text .dynamic .got
>     03     .dynamic .got
>     04     .note
> 
> Explicitly describe the .got section in the MIPS VDSO linker script
> after .rodata, which switches back to the default text segment,
> resulting in a dynamic segment that is the exact size of the .dynamic
> section as expected with no other layout changes.
> 
>   -  DYNAMIC        0x000c14 0x00000c14 0x00000c14 0x000a4 0x000a4 R   0x10
>   +  DYNAMIC        0x000c14 0x00000c14 0x00000c14 0x00090 0x00090 R   0x4
> 
>   -   03     .dynamic .got
>   +   03     .dynamic
> 
> Closes: https://github.com/ClangBuiltLinux/linux/issues/2166
> Fixes: 2db1ec80dfd5 ("MIPS: VDSO: Fold MIPS_DISABLE_VDSO into MIPS_GENERIC_GETTIMEOFDAY")
> Signed-off-by: Nathan Chancellor <nathan@kernel.org>

Reviewed-by: Thomas Weißschuh <thomas.weissschuh@linutronix.de>

> ---
> The fixes tag feels a little strong since it seems like it has just been
> luck up until this point that there has been no warning but I decided to
> be conservative and include it regardless. Feel free to remove it if you
> see fit.

I have seen this issue independently from the commit blamed above.
It occurs at least on commit 2b7a25df823d ("Merge tag 'mm-nonmm-stable-2026-02-18-19-56'
of git://git.kernel.org/pub/scm/linux/kernel/git/akpm/mm").
If you want I can investigate more, but the Fixes above works for me, too.

> I think this should go via timers/vdso with the blamed commit. I plan to
> send a follow up series for 7.3 to add '--orphan-handling' to the MIPS
> VDSO to avoid issues like this in the future but that can go via the
> MIPS tree, as it is not really a fix and I will need to properly test
> it.
> ---
>  arch/mips/vdso/vdso.lds.S | 1 +
>  1 file changed, 1 insertion(+)
> 
> diff --git a/arch/mips/vdso/vdso.lds.S b/arch/mips/vdso/vdso.lds.S
> index 05badf3ae0ff..278ab6444e98 100644
> --- a/arch/mips/vdso/vdso.lds.S
> +++ b/arch/mips/vdso/vdso.lds.S
> @@ -56,6 +56,7 @@ SECTIONS
>  	.dynamic	: { *(.dynamic) }		:text :dynamic
>  
>  	.rodata		: { *(.rodata*) }		:text
> +	.got		: { *(.got) }
>  
>  	_end = .;
>  	PROVIDE(end = .);
> 
> ---
> base-commit: 13f6218e6fe79dc64aed76d738b765b45f62492b
> change-id: 20260608-mips-vdso-fix-section-layout-262bc18d0c29
> 
> Best regards,
> --  
> Cheers,
> Nathan
> 

Re: [PATCH] MIPS: VDSO: Avoid including .got in dynamic segment

[Nick Desaulniers]

On Tue, Jun 09, 2026 at 06:31:21PM -0700, Nathan Chancellor wrote:
> After commit 2db1ec80dfd5 ("MIPS: VDSO: Fold MIPS_DISABLE_VDSO into
> MIPS_GENERIC_GETTIMEOFDAY"), building ARCH=mips allnoconfig with LLVM=1
> shows some warnings from llvm-readelf while checking the VDSO for
> dynamic relocations:
> 
>   llvm-readelf: warning: 'arch/mips/vdso/vdso.so.dbg.raw': invalid PT_DYNAMIC size (0xa4)
>   llvm-readelf: warning: 'arch/mips/vdso/vdso.so.dbg.raw': PT_DYNAMIC dynamic table is invalid: SHT_DYNAMIC will be used
> 
> The blamed commit alters the link order of objects into vdso.so.raw,
> placing vgettimeofday.o after sigreturn.o. This ultimately results in
> the .text section shrinking slightly in size, which in turn changes the
> offset of the .dynamic section.
> 
>   -  [ 9] .text             PROGBITS        000002f0 0002f0 000930 00  AX  0   0 16
>   -  [10] .dynamic          DYNAMIC         00000c20 000c20 000090 08   A  5   0  4
>   +  [ 9] .text             PROGBITS        000002f0 0002f0 000924 00  AX  0   0 16
>   +  [10] .dynamic          DYNAMIC         00000c14 000c14 000090 08   A  5   0  4
> 
> Changing the offset of the .dynamic section causes the dynamic segment
> size to grow by the same amount, which triggers a warning in
> llvm-readelf because PT_DYNAMIC's p_filesz (0xa4) is no longer a
> multiple of its sh_entsize (8):
> 
>   -  DYNAMIC        0x000c20 0x00000c20 0x00000c20 0x00098 0x00098 R   0x10
>   +  DYNAMIC        0x000c14 0x00000c14 0x00000c14 0x000a4 0x000a4 R   0x10
> 
> The size of the dynamic segment was already incorrect before the blamed
> comment, as it should be 0x90 like the .dynamic section above (18
> entries at 8 bytes per entry); it just so happens that 0x98 % 8 is 0,
> whereas 0xa4 % 8 is 4, so there was no warning.
> 
> Looking at the section to segment mapping of the dynamic segment reveals
> that it includes the .got section, as it is implicitly placed after
> .dynamic by ld.lld's orphan section heuristics and inherits its segments
> from the linker script.
> 
>   [ 9] .text             PROGBITS        000002f0 0002f0 000924 00  AX  0   0 16
>   [10] .dynamic          DYNAMIC         00000c14 000c14 000090 08   A  5   0  4
>   [11] .got              PROGBITS        00000cb0 000cb0 000008 00 WAp  0   0 16
> 
>   Section to Segment mapping:
>    Segment Sections...
>     00     .mips_abiflags
>     01     .reginfo
>     02     .mips_abiflags .reginfo .hash .dynsym .dynstr .gnu.version .gnu.version_d .note .text .dynamic .got
>     03     .dynamic .got
>     04     .note
> 
> Explicitly describe the .got section in the MIPS VDSO linker script
> after .rodata, which switches back to the default text segment,
> resulting in a dynamic segment that is the exact size of the .dynamic
> section as expected with no other layout changes.
> 
>   -  DYNAMIC        0x000c14 0x00000c14 0x00000c14 0x000a4 0x000a4 R   0x10
>   +  DYNAMIC        0x000c14 0x00000c14 0x00000c14 0x00090 0x00090 R   0x4
> 
>   -   03     .dynamic .got
>   +   03     .dynamic
> 
> Closes: https://github.com/ClangBuiltLinux/linux/issues/2166
> Fixes: 2db1ec80dfd5 ("MIPS: VDSO: Fold MIPS_DISABLE_VDSO into MIPS_GENERIC_GETTIMEOFDAY")
> Signed-off-by: Nathan Chancellor <nathan@kernel.org>

Thanks for the patch!
Acked-by: Nick Desaulniers <ndesaulniers@google.com>

> ---
> The fixes tag feels a little strong since it seems like it has just been
> luck up until this point that there has been no warning but I decided to
> be conservative and include it regardless. Feel free to remove it if you
> see fit.
> 
> I think this should go via timers/vdso with the blamed commit. I plan to
> send a follow up series for 7.3 to add '--orphan-handling' to the MIPS
> VDSO to avoid issues like this in the future but that can go via the
> MIPS tree, as it is not really a fix and I will need to properly test
> it.
> ---
>  arch/mips/vdso/vdso.lds.S | 1 +
>  1 file changed, 1 insertion(+)
> 
> diff --git a/arch/mips/vdso/vdso.lds.S b/arch/mips/vdso/vdso.lds.S
> index 05badf3ae0ff..278ab6444e98 100644
> --- a/arch/mips/vdso/vdso.lds.S
> +++ b/arch/mips/vdso/vdso.lds.S
> @@ -56,6 +56,7 @@ SECTIONS
>  	.dynamic	: { *(.dynamic) }		:text :dynamic
>  
>  	.rodata		: { *(.rodata*) }		:text
> +	.got		: { *(.got) }
>  
>  	_end = .;
>  	PROVIDE(end = .);
> 
> ---
> base-commit: 13f6218e6fe79dc64aed76d738b765b45f62492b
> change-id: 20260608-mips-vdso-fix-section-layout-262bc18d0c29
> 
> Best regards,
> --  
> Cheers,
> Nathan
> 

Re: [PATCH] MIPS: VDSO: Avoid including .got in dynamic segment

[Nathan Chancellor]

On Wed, Jun 10, 2026 at 09:45:08AM +0200, Thomas Weißschuh wrote:
> On Tue, Jun 09, 2026 at 06:31:21PM -0700, Nathan Chancellor wrote:
> > The fixes tag feels a little strong since it seems like it has just been
> > luck up until this point that there has been no warning but I decided to
> > be conservative and include it regardless. Feel free to remove it if you
> > see fit.
> 
> I have seen this issue independently from the commit blamed above.
> It occurs at least on commit 2b7a25df823d ("Merge tag 'mm-nonmm-stable-2026-02-18-19-56'
> of git://git.kernel.org/pub/scm/linux/kernel/git/akpm/mm").
> If you want I can investigate more, but the Fixes above works for me, too.

Maybe it would be better to blame the original vDSO commit then?

Fixes: ebb5e78cc634 ("MIPS: Initial implementation of a VDSO")

I cannot build with LLVM back that far but something tells me that this
could happen back then too just because .got is not described there. I
don't have a strong opinion though, this does not appear to be a
functional issue since llvm-readelf just falls back to using .dynamic
directly IIUC. I guess tglx could make the call during application if he
is okay with that?

-- 
Cheers,
Nathan