Re: [PATCH v2 01/10] riscv: Implement cmpxchg32/64() using Zacas

[Alexandre Ghiti]

Hi Andrea,

On 27/06/2024 13:06, Andrea Parri wrote:
> > -#define __arch_cmpxchg(lr_sfx, sc_sfx, prepend, append, r, p, co, o, n)	\
> > +#define __arch_cmpxchg(lr_sfx, sc_cas_sfx, prepend, append, r, p, co, o, n)	\
> >   ({									\
> > +	__label__ zacas, end;						\
> >   	register unsigned int __rc;					\
> >   									\
> > +	if (IS_ENABLED(CONFIG_RISCV_ISA_ZACAS)) {			\
> > +		asm goto(ALTERNATIVE("nop", "j %[zacas]", 0,		\
> > +				     RISCV_ISA_EXT_ZACAS, 1)		\
> > +			 : : : : zacas);				\
> > +	}								\
> > +									\
> >   	__asm__ __volatile__ (						\
> >   		prepend							\
> >   		"0:	lr" lr_sfx " %0, %2\n"				\
> >   		"	bne  %0, %z3, 1f\n"				\
> > -		"	sc" sc_sfx " %1, %z4, %2\n"			\
> > +		"	sc" sc_cas_sfx " %1, %z4, %2\n"			\
> >   		"	bnez %1, 0b\n"					\
> >   		append							\
> >   		"1:\n"							\
> >   		: "=&r" (r), "=&r" (__rc), "+A" (*(p))			\
> >   		: "rJ" (co o), "rJ" (n)					\
> >   		: "memory");						\
> > +	goto end;							\
> > +									\
> > +zacas:									\
> > +	if (IS_ENABLED(CONFIG_RISCV_ISA_ZACAS)) {			\
> > +		__asm__ __volatile__ (					\
> > +			prepend						\
> > +			"	amocas" sc_cas_sfx " %0, %z2, %1\n"	\
> > +			append						\
> > +			: "+&r" (r), "+A" (*(p))			\
> > +			: "rJ" (n)					\
> > +			: "memory");					\
> > +	}								\
> Is this second IS_ENABLED(CONFIG_RISCV_ISA_ZACAS) check actually needed?
> (just wondering - no real objection)


To me yes, otherwise a toolchain without zacas support would fail to 
assemble the amocas instruction.


> > +end:;									\
> Why the semicolon?


That fixes a clang warning reported by Nathan here: 
https://lore.kernel.org/linux-riscv/20240528193110.GA2196855@thelio-3990X/


> >   })
> >   
> >   #define _arch_cmpxchg(ptr, old, new, sc_sfx, prepend, append)		\
> > @@ -156,7 +177,7 @@
> >   	__typeof__(ptr) __ptr = (ptr);					\
> >   	__typeof__(*(__ptr)) __old = (old);				\
> >   	__typeof__(*(__ptr)) __new = (new);				\
> > -	__typeof__(*(__ptr)) __ret;					\
> > +	__typeof__(*(__ptr)) __ret = (old);				\
> This is because the compiler doesn't realize __ret is actually
> initialized, right?  IAC, seems a bit unexpected to initialize
> with (old) (which indicates SUCCESS of the CMPXCHG operation);
> how about using (new) for the initialization of __ret instead?
> would (new) still work for you?


But amocas rd register must contain the expected old value in order to 
actually work right?


>    Andrea
>
> _______________________________________________
> linux-riscv mailing list
> linux-riscv@xxxxxxxxxxxxxxxxxxx
> http://lists.infradead.org/mailman/listinfo/linux-riscv