[PATCH] MIPS: Avoid dead code/data elimination when using recordmcount

[PATCH] MIPS: Avoid dead code/data elimination when using recordmcount

[Rosen Penev]

CONFIG_LD_DEAD_CODE_DATA_ELIMINATION builds with -ffunction-sections,
placing every function in its own section. recordmcount refuses to use
weak symbols as section bases, as a strong definition could override
them, so a weak function alone in its section, such as the generic
pcibios_fixup_bus() stub, breaks the build:

  Cannot find symbol for section 88: .text.pcibios_fixup_bus.
  drivers/pci/probe.o: failed

MIPS always records mcount locations with recordmcount, so this hits
any config with the function tracer and dead code elimination enabled,
e.g. malta_defconfig built with LLVM=1.

Only offer dead code elimination when recordmcount is not used, as
powerpc does since commit f8b2336f15f3 ("powerpc: Avoid dead code/data
elimination when using recordmcount"). In practice this makes it
unavailable together with the function tracer on MIPS.

Fixes: c0436b503591 ("MIPS: Enable dead code elimination")
Signed-off-by: Rosen Penev <rosenp@gmail.com>
---
 arch/mips/Kconfig | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/arch/mips/Kconfig b/arch/mips/Kconfig
index ddabe25d6bf7..f4f82bd75b66 100644
--- a/arch/mips/Kconfig
+++ b/arch/mips/Kconfig
@@ -79,7 +79,7 @@ config MIPS
 	select HAVE_IRQ_TIME_ACCOUNTING
 	select HAVE_KPROBES
 	select HAVE_KRETPROBES
-	select HAVE_LD_DEAD_CODE_DATA_ELIMINATION
+	select HAVE_LD_DEAD_CODE_DATA_ELIMINATION if !FTRACE_MCOUNT_USE_RECORDMCOUNT
 	select HAVE_MOD_ARCH_SPECIFIC
 	select HAVE_NMI
 	select HAVE_PAGE_SIZE_4KB if !CPU_LOONGSON2EF && !CPU_LOONGSON64
-- 
2.56.0

Re: [PATCH] MIPS: Avoid dead code/data elimination when using recordmcount

[Thomas Bogendoerfer]

On Sun, Oct 04, 2026 at 04:48:26PM -0700, Rosen Penev wrote:
> CONFIG_LD_DEAD_CODE_DATA_ELIMINATION builds with -ffunction-sections,
> placing every function in its own section. recordmcount refuses to use
> weak symbols as section bases, as a strong definition could override
> them, so a weak function alone in its section, such as the generic
> pcibios_fixup_bus() stub, breaks the build:
> 
>   Cannot find symbol for section 88: .text.pcibios_fixup_bus.
>   drivers/pci/probe.o: failed
> 
> MIPS always records mcount locations with recordmcount, so this hits
> any config with the function tracer and dead code elimination enabled,
> e.g. malta_defconfig built with LLVM=1.
> 
> Only offer dead code elimination when recordmcount is not used, as
> powerpc does since commit f8b2336f15f3 ("powerpc: Avoid dead code/data
> elimination when using recordmcount"). In practice this makes it
> unavailable together with the function tracer on MIPS.
> 
> Fixes: c0436b503591 ("MIPS: Enable dead code elimination")
> Signed-off-by: Rosen Penev <rosenp@gmail.com>
> ---
>  arch/mips/Kconfig | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/arch/mips/Kconfig b/arch/mips/Kconfig
> index ddabe25d6bf7..f4f82bd75b66 100644
> --- a/arch/mips/Kconfig
> +++ b/arch/mips/Kconfig
> @@ -79,7 +79,7 @@ config MIPS
>  	select HAVE_IRQ_TIME_ACCOUNTING
>  	select HAVE_KPROBES
>  	select HAVE_KRETPROBES
> -	select HAVE_LD_DEAD_CODE_DATA_ELIMINATION
> +	select HAVE_LD_DEAD_CODE_DATA_ELIMINATION if !FTRACE_MCOUNT_USE_RECORDMCOUNT
>  	select HAVE_MOD_ARCH_SPECIFIC
>  	select HAVE_NMI
>  	select HAVE_PAGE_SIZE_4KB if !CPU_LOONGSON2EF && !CPU_LOONGSON64
> -- 
> 2.56.0

applied to mips-fixes

Thomas.

-- 
Crap can work. Given enough thrust pigs will fly, but it's not necessarily a
good idea.                                                [ RFC1925, 2.3 ]