On Thu, 6 Sep 2018 12:49:39 +0200
Peter Oberparleiter <oberpar@xxxxxxxxxxxxx> wrote:
> On 28.08.2018 02:43, Masami Hiramatsu wrote:
> > I recently added a gcov profiling for ftrace, following Documentation/dev-tools/gcov.rst.
> > 6b7dca401cb1 ("tracing: Allow gcov profiling on only ftrace subsystem")
> >
> > and it caused may orphan section warnings/errors on arm and powerpc64.
> > We also found same error happens when CONFIG_GCOV_PROFILE_ALL=y.
> > So I guess GCOV kernel subsystem broken in some environment.
>
> I can confirm that these issues are unrelated to your commit which only
> made the existing problems visible by enabling gcov-kernel profiling for
> allyesconfig builds. Both problems already exist in v4.18 and likely
> earlier versions.
>
> Problem 1: link failure on arm
> (https://lkml.org/lkml/2018/8/24/345)
>
> The root cause appears to be that the arm linker script only partially
> discards unneeded sections generated by the compiler when gcov-kernel
> profiling is enabled.
>
> Problem 2: linker warnings on powerpc
> (https://lkml.org/lkml/2018/8/24/72)
>
> CONFIG_LD_DEAD_CODE_DATA_ELIMINATION=y adds compiler flag
> -fdata-sections. This option causes GCC to create separate data sections
> for data objects, including those generated by GCC internally for gcov
> profiling. Since the name of these internal data objects starts with a .
> (.LPBX0, .LPBX1), the resulting section name starts with "data..". As
> section names starting with "data.." are used for specific purposes in
> the Linux kernel, the linker script does not automatically handle them,
> resulting in the "orphan section" linker warnings.
Good catch! I didn't know that option.
>
> I've attached a quick fix that should address both problems. I'd
> appreciate if this patch could get some testing before I post proper fix
> patches.
Hmm, I'm still not able to reproduce it on powerpc cross build even with
CONFIG_LD_DEAD_CODE_DATA_ELIMINATION=y ... So, sorry I couldn't test this patch.
>
> -- >8 --
> diff --git a/arch/arm/kernel/vmlinux.lds.h b/arch/arm/kernel/vmlinux.lds.h
> index ae5fdff18406..2ca33277a28b 100644
> --- a/arch/arm/kernel/vmlinux.lds.h
> +++ b/arch/arm/kernel/vmlinux.lds.h
> @@ -48,6 +48,7 @@
>
> #define ARM_DISCARD \
> *(.ARM.exidx.exit.text) \
> + *(.ARM.exidx.text.exit) \
BTW, why would we need this?
Thanks,
> *(.ARM.extab.exit.text) \
> ARM_CPU_DISCARD(*(.ARM.exidx.cpuexit.text)) \
> ARM_CPU_DISCARD(*(.ARM.extab.cpuexit.text)) \
> diff --git a/include/asm-generic/vmlinux.lds.h b/include/asm-generic/vmlinux.lds.h
> index 7b75ff6e2fce..5cf3b90c5592 100644
> --- a/include/asm-generic/vmlinux.lds.h
> +++ b/include/asm-generic/vmlinux.lds.h
> @@ -68,7 +68,7 @@
> */
> #ifdef CONFIG_LD_DEAD_CODE_DATA_ELIMINATION
> #define TEXT_MAIN .text .text.[0-9a-zA-Z_]*
> -#define DATA_MAIN .data .data.[0-9a-zA-Z_]*
> +#define DATA_MAIN .data .data.[0-9a-zA-Z_]* .data..LPBX*
> #define SDATA_MAIN .sdata .sdata.[0-9a-zA-Z_]*
> #define RODATA_MAIN .rodata .rodata.[0-9a-zA-Z_]*
> #define BSS_MAIN .bss .bss.[0-9a-zA-Z_]*
> @@ -613,7 +613,7 @@
>
> #define EXIT_DATA \
> *(.exit.data .exit.data.*) \
> - *(.fini_array) \
> + *(.fini_array .fini_array.*) \
> *(.dtors) \
> MEM_DISCARD(exit.data*) \
> MEM_DISCARD(exit.rodata*)
> --
> Peter Oberparleiter
> Linux on Z Development - IBM Germany
>
--
Masami Hiramatsu <mhiramat@xxxxxxxxxx>
