Re: [musl] [PATCH] fix mips syscall asm regressions and kernel bug

[Rich Felker <dalias@libc.org>]

On Wed, Mar 11, 2020 at 07:09:25PM -0400, Rich Felker wrote:
> >From dd656a205b5f1d0513f7c71c70c786af1e248304 Mon Sep 17 00:00:00 2001
> From: Rich Felker <dalias@aerifal.cx>
> Date: Wed, 11 Mar 2020 18:58:38 -0400
> Subject: [PATCH 3/4] restore mips syscall asm improvements from reverted
>  change
> 
> in addition to the wrong change to loading of r2, commit
> 604f8d3d8b08ee4f548de193050ef93a7753c2e0 removed some useless,
> redundant, and possibly undefined constraints. bring back the good
> parts.
> ---
>  arch/mips/syscall_arch.h | 31 ++++++++++++++++---------------
>  1 file changed, 16 insertions(+), 15 deletions(-)
> 
> diff --git a/arch/mips/syscall_arch.h b/arch/mips/syscall_arch.h
> index a3812188..0a5d6d8f 100644
> --- a/arch/mips/syscall_arch.h
> +++ b/arch/mips/syscall_arch.h
> @@ -21,7 +21,8 @@ static inline long __syscall0(long n)
>  	register long r2 __asm__("$2");
>  	__asm__ __volatile__ (
>  		"addu $2,$0,%2 ; syscall"
> -		: "=&r"(r2), "=r"(r7) : "ir"(n), "0"(r2), "1"(r7)
> +		: "=&r"(r2), "=r"(r7)
> +		: "ir"(n)
>  		: SYSCALL_CLOBBERLIST, "$8", "$9", "$10");
>  	return r7 ? -r2 : r2;
>  }

This part of the change (also present in the 64-bit versions) was
horribly wrong because of an ancient GCC bug: specific-register
bindings for outputs are not honored at all with earlyclobber. This is
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=87733

I think for now I can work around it by adding back the dummy input
constraint the old asm was using. This looks wrong but at least it
worked.

Rich