On Tue, Aug 24, 2021, Lai Jiangshan wrote:
> From: Lai Jiangshan <laijs@xxxxxxxxxxxxxxxxx>
>
> __direct_pte_prefetch() skips prefetching the last range.
>
> The last range are often the whole range after the faulted spte when
> guest is touching huge-page-mapped(in guest view) memory forwardly
> which means prefetching them can reduce pagefault.
>
> Signed-off-by: Lai Jiangshan <laijs@xxxxxxxxxxxxxxxxx>
> ---
> arch/x86/kvm/mmu/mmu.c | 5 +++--
> 1 file changed, 3 insertions(+), 2 deletions(-)
>
> diff --git a/arch/x86/kvm/mmu/mmu.c b/arch/x86/kvm/mmu/mmu.c
> index e5932af6f11c..ac260e01e9d8 100644
> --- a/arch/x86/kvm/mmu/mmu.c
> +++ b/arch/x86/kvm/mmu/mmu.c
> @@ -2847,8 +2847,9 @@ static void __direct_pte_prefetch(struct kvm_vcpu *vcpu,
> i = (sptep - sp->spt) & ~(PTE_PREFETCH_NUM - 1);
> spte = sp->spt + i;
>
> - for (i = 0; i < PTE_PREFETCH_NUM; i++, spte++) {
> - if (is_shadow_present_pte(*spte) || spte == sptep) {
> + for (i = 0; i <= PTE_PREFETCH_NUM; i++, spte++) {
> + if (i == PTE_PREFETCH_NUM ||
> + is_shadow_present_pte(*spte) || spte == sptep) {
Heh, I posted a fix just a few days ago. I prefer having a separate call after
the loop. The "<= PTE_PREFETCH_NUM" is subtle, and a check at the ends avoids
a CMP+Jcc in the loop, though I highly doubt that actually affects performance.
https://lkml.kernel.org/r/20210818235615.2047588-1-seanjc@xxxxxxxxxx
> if (!start)
> continue;
> if (direct_pte_prefetch_many(vcpu, sp, start, spte) < 0)
> --
> 2.19.1.6.gb485710b
>
