On Mon, 6 Jul 2026 15:49:21 +0200, [email protected] wrote: > >>> + > >>> +#define __HAVE_ARCH_MEMCPY_NT 1 > >>> +/* > >>> + * Reuse the existing x86 flushcache backend as the nt copy primitive. > >>> + * Callers pair it with memcpy_nt_drain() when later stores must be > >>> + * ordered after the copy. > >>> + */ > >>> +static __always_inline void memcpy_nt(void *dst, const void *src, size_t > >>> cnt) > >>> +{ > >>> + memcpy_flushcache(dst, src, cnt); > >>> +} > >> > >> In particular if you end up with a single function (that jsut requires a > >> write > >> memory barrier afterwards), please make this > >> > >> #define memcpy_nt memcpy_nt > >> > >> instead. > > > > Thanks for the review. > > > > Just to make sure I understood this correctly: do you mean that, rather > > than using a separate __HAVE_ARCH_MEMCPY_NT feature macro, you would > > prefer the usual self-macro override pattern, with > > > > #define memcpy_nt memcpy_nt > > > > in arch/x86/include/asm/string_64.h and the generic fallback kept under > > #ifndef memcpy_nt in include/linux/string.h? > > Yes, that's what Linus suggested as the preferred way for single functions > when > it came up in different context recently.
Thanks for confirming. I will switch v6 to that self-macro override pattern for memcpy_nt(), and I used the same style for memcpy_nt_drain(). Thanks, Zhe
