> + 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: \
> + __asm__ __volatile__ ( \
> + prepend \
> + " amocas" sc_cas_sfx " %0, %z2, %1\n" \
> + append \
> + : "+&r" (r), "+A" (*(p)) \
> + : "rJ" (n) \
> + : "memory"); \
With this, a cmpxchg32() will result in something like
amocas.w.rl a5,a4,(s1)
fence rw,rw
(cf. my remarks in patch #4); this will/should provide enough sync,
but you might want to try the alternative and currently more common
mapping for "fully-ordered AMO sequences", aka
amocas.w.aqrl a5,a4,(s1)
Similarly for cmpxchg64 and other sizes.
Andrea
