Re: [PATCH] Fix copy_user on x86_64

[Linus Torvalds <torvalds@linux-foundation.org>]

On Thu, 26 Jun 2008, Vitaly Mayatskikh wrote:
> 
> Seems reasonable. However, we still need specialized memset() routine,
> because, again, destination can fail. Thanks for the review, Linus!

Actually, the "zero at the end" case is only for copy_from_user() (at 
least it _should_ be), so for the clearing-at-end you should be able to 
use a regular memset().

But it's not a big deal either way. As long as we only get into the fixup 
routine at exception time, and handle all the common cases fast (ie do the 
32-byte unrolled thing etc optimally), the fixup routine can do everything 
a byte at a time with "get_user()" and "put_user()" etc. The "fault at 
copy_*_user()" case really isn't all that performance-sensitive, because 
it really happens essentially _never_.

(That's obviously why nobody even noticed how broken they were for 
essentially what must have been _years_. It's not just not a performance 
sensitive area, it's one that is entered so seldom that it's hard to ever 
hit any correctness issues either)

		Linus