Re: [RFC PATCH v2 0/3] x86/mm/tlb: Defer TLB flushes with PTI
From: Dave Hansen
Date: Tue Aug 27 2019 - 14:17:35 EST
On 8/23/19 3:52 PM, Nadav Amit wrote:
> n_pages concurrent +deferred-pti change
> ------- ---------- ------------- ------
> 1 2119 1986 -6.7%
Is this implying that a single-page INVPCID is 133 cycles slower than
INVLPG? That seems a bit larger than I'd expect.