On Wed, May 25, 2022 at 01:11:17PM +0100, Mark Hemment wrote:
> A slight doubt here; comment says "less than a cachline", but the code
> is using 'ja' (jump if above) - so calls 'clear_user_original' for a
> 'len' less than or equal to 64.
> Not sure of the intended behaviour for 64 bytes here, but
> 'copy_user_enhanced_fast_string' uses the slow-method for lengths less
> than 64. So, should this be coded as;
> cmp $64,%rcx
> jb clear_user_original
> ?
Yeah, it probably doesn't matter whether you clear a cacheline the "old"
way or with some of the new ones. clear_user() performance matters only
in microbenchmarks, as I've come to realize.
But your suggestion simplifies the code so lemme do that.
Thx!
--
Regards/Gruss,
Boris.
https://people.kernel.org/tglx/notes-about-netiquette
