Re: [tip:x86/mm] x86/boot/compressed/64: Describe the logic behind the LA57 check

[Linus Torvalds]

On Mon, Mar 12, 2018 at 10:06 AM, Andy Lutomirski <luto@xxxxxxxxxx> wrote:
>
> I'd be surprised if there's a noticeable performance hit on anything
> except the micro-est of benchmarks.  We're talking one extra
> intermediate paging structure cache entry in use, maybe a few data
> cache lines, and (wild guess) 0 extra cycles on a TLB miss in the
> normal case.  This is because the walks are almost never going to
> start at the root.

Probably. But VM people may disagree if they already have high TLB miss costs.

> The real hit will be the extra page table for every task.

.. and it's unclear how noticeable that might be. It's not like it's
per-thread, only per process, and very few people have so many
processes that a page per process matters.

But regardless, I think we're better off with a "wait and see" approach.

IOW, try to use 5-level whenever possible for now, and _if_ somebody
actually can show that 4-level page tables perform better or have some
other advantage, we can then try to be clever later when it's all
tested and it's just an optimization, not a "that code won't even run
normally and gets basically zero coverage".

                Linus