Re: [RFC][PATCH 21/22] x86,word-at-a-time: Remove .fixup usage

[Peter Zijlstra]

On Thu, Nov 04, 2021 at 04:33:10PM -0700, Josh Poimboeuf wrote:
> On Thu, Nov 04, 2021 at 05:47:50PM +0100, Peter Zijlstra wrote:
> > XXX: I'm not really happy with this patch
> >
> >  static inline unsigned long load_unaligned_zeropad(const void *addr)
> >  {
> > -	unsigned long ret, dummy;
> > +	unsigned long ret;
> > +
> > +	asm("1:\tmov (%0),%0\n"
> > +	    "2:\n"
> > +	    _ASM_EXTABLE_TYPE_REG(1b, 2b, EX_TYPE_LOAD_UNALIGNED, %0)
> > +	    : "=&r" (ret)
> > +	    : "0" ((unsigned long)addr));
> >  
> > -	asm(
> > -		"1:\tmov %2,%0\n"
> > -		"2:\n"
> > -		".section .fixup,\"ax\"\n"
> > -		"3:\t"
> > -		"lea %2,%1\n\t"
> > -		"and %3,%1\n\t"
> > -		"mov (%1),%0\n\t"
> > -		"leal %2,%%ecx\n\t"
> > -		"andl %4,%%ecx\n\t"
> > -		"shll $3,%%ecx\n\t"
> > -		"shr %%cl,%0\n\t"
> > -		"jmp 2b\n"
> > -		".previous\n"
> > -		_ASM_EXTABLE(1b, 3b)
> > -		:"=&r" (ret),"=&c" (dummy)
> > -		:"m" (*(unsigned long *)addr),
> > -		 "i" (-sizeof(unsigned long)),
> > -		 "i" (sizeof(unsigned long)-1));
> >  	return ret;
> >  }
> 
> Yeah, it hurts code generation and I guess it's a hot path.
> 
> Maybe put the fixup code in the function.  In case of
> CONFIG_CC_HAS_ASM_GOTO_OUTPUT, it could be at a label at the end of the
> function.  Otherwise it'd have to be inline.

Ooh, yes. If we use ASM_GOTO_OUTPUT then it'll work just right, and the
legacy code will get worse, but I can live with that (everybody is
already building their kernels with gcc-11/clang anyway :-))