Re: Encrypted Message
Thomas Gleixner <[email protected]>
| Newsgroups | org.kernel.lore.historical-speck |
|---|---|
| Message-ID | <[email protected]> |
On Fri, 22 Feb 2019, speck for Linus Torvalds wrote: > On Thu, Feb 21, 2019 at 11:46 PM speck for Jon Masters > <[email protected]> wrote: > > > > Dunno if it's worth documenting that using a specifically valid segment > > is faster than a zero selector according to Intel. > > It probably is worth documenting, because it's so non-intuitive. > > In a sane world, Intel would have realized that > > (a) nobody uses verw for anything else > > (b) a zero selector doesn't need any LDT/GDT loads > > and optimized for that case. Preferably allowing a register operand too. > > So the fact that the verw has to have a memory op and is faster for a > non-zero segment descriptor is all kinds of crazy. Will add one. > And being crazy, a comment about it is worth it, since otherwise it > looks like _we_ are the crazy ones. Some time ago you stated at the kernel summit, that everyone in the room is crazy: https://quotes.yourdictionary.com/author/linus-torvalds/190350 Thanks, tglx