Messages in this thread | | | Date | Thu, 14 May 2015 19:46:08 +0200 | From | Ingo Molnar <> | Subject | Re: [PATCH] x86/asm/entry/64: Use shorter MOVs from segmers registers |
| |
* Denys Vlasenko <dvlasenk@redhat.com> wrote:
> On 05/14/2015 07:08 PM, Linus Torvalds wrote: > > On Thu, May 14, 2015 at 10:05 AM, Linus Torvalds > > <torvalds@linux-foundation.org> wrote: > >> > >> I don't object to the patch, but did we actually confirm that it > >> always overwrites all of %ecx? > > > > Just to clarify: I don't object to the patch because the code > > doesn't actually end up *depending* on the high bits anyway, and > > does word-sized compares etc. And the instruction size and speed > > things I don't doubt. So it's just the commit message I wanted to > > check wrt that whole "always overwrites all of %ecx". Because > > older CPU's didn't necessarily (things like partial register > > writes are much less of an issue when you're in-order and stupid > > ;) > > This is 64-bit code, and all 64-bit CPUs zero-extend moves from > segment registers. As you said, in this particular code it wouldn't > matter anyway since subsequent code doesn't care about high bits of > %ecx...
Mind updating the changelog with all that information? It wasn't obvious to me either, as most of the mnemonics 'look' 32-bit.
I'd also say that a changelog is not complete, by definition, if Linus has to ask about it ;-)
Thanks,
Ingo
| |