[PATCH v8 1/8] ppc64 (le): prepare for -mprofile-kernel

Michael Ellerman mpe at ellerman.id.au
Wed Feb 17 21:55:40 AEDT 2016


On Wed, 2016-02-10 at 17:21 +0100, Torsten Duwe wrote:

> The gcc switch -mprofile-kernel, available for ppc64 on gcc > 4.8.5,
> allows to call _mcount very early in the function, which low-level
> ASM code and code patching functions need to consider.
> Especially the link register and the parameter registers are still
> alive and not yet saved into a new stack frame.

...

> diff --git a/arch/powerpc/kernel/module_64.c b/arch/powerpc/kernel/module_64.c
> index ac64ffd..72a1a52 100644
> --- a/arch/powerpc/kernel/module_64.c
> +++ b/arch/powerpc/kernel/module_64.c
> @@ -476,17 +474,44 @@ static unsigned long stub_for_addr(Elf64_Shdr *sechdrs,
>  	return (unsigned long)&stubs[i];
>  }
>  
> +#ifdef CC_USING_MPROFILE_KERNEL
> +static int is_early_mcount_callsite(u32 *instruction)
> +{
> +	/* -mprofile-kernel sequence starting with
> +	 * mflr r0 and maybe std r0, LRSAVE(r1).
> +	 */
> +	if ((instruction[-3] == PPC_INST_MFLR &&
> +	     instruction[-2] == PPC_INST_STD_LR) ||
> +	    instruction[-2] == PPC_INST_MFLR) {
> +		/* Nothing to be done here, it's an _mcount
> +		 * call location and r2 will have to be
> +		 * restored in the _mcount function.
> +		 */
> +		return 1;
> +	}
> +	return 0;
> +}

So this logic to deal with the 2 vs 3 instruction version of the mcount
sequence is problematic.

On a kernel built with the 2 instruction version this will fault when the
function we're looking at is located at the beginning of a page. Because
instruction[-3] goes off the front of the mapping.

We can probably fix that. But it's still a bit dicey.

I'm wondering if we want to just say we only support the 2 instruction version.
Currently that means GCC 6 only, or a distro compiler with the backport of
e95d0248dace. But we could also ask GCC to backport it to 4.9 and 5.

Thoughts?

cheers



More information about the Linuxppc-dev mailing list