[PATCH v2 6/7] KVM: arm64: Break the table entries using TLBI range instructions

[Raghavendra Rao Ananta]

On Wed, Mar 29, 2023 at 5:17 PM Oliver Upton <oliver.upton at linux.dev> wrote:
>
> nit: s/break/invalidate/g
>
> There is a rather important degree of nuance there. 'Break' as it
> relates to break-before-make implies that the PTE is made invalid and
> visible to hardware _before_ a subsequent invalidation. There will be
> systems that relax this requirement and also support TLBIRANGE.
>
> On Mon, Feb 06, 2023 at 05:23:39PM +0000, Raghavendra Rao Ananta wrote:
>
> Some nitpicking on the changelog:
>
> > Currently, when breaking up the stage-2 table entries, KVM
>
> 'breaking up stage-2 table entries' is rather ambiguous. Instead
> describe the operation taking place on the page tables (i.e. hugepage
> collapse).
>
> > would flush the entire VM's context using 'vmalls12e1is'
> > TLBI operation. One of the problematic situation is collapsing
> > table entries into a hugepage, specifically if the VM is
> > faulting on many hugepages (say after dirty-logging). This
> > creates a performance penality for the guest whose pages have
>
> typo: penalty
>
> > already been faulted earlier as they would have to refill their
> > TLBs again.
> >
> > Hence, if the system supports it, use __kvm_tlb_flush_range_vmid_ipa()
>
> > to flush only the range of pages governed by the table entry,
> > while leaving other TLB entries alone. An upcoming patch also
> > takes advantage of this when breaking up table entries during
> > the unmap operation.
>
> Language regarding an upcoming patch isn't necessary, as this one stands
> on its own (implements and uses a range-based invalidation helper).
>
> > Signed-off-by: Raghavendra Rao Ananta <rananta at google.com>
> > ---
> >  arch/arm64/kvm/hyp/pgtable.c | 23 ++++++++++++++++++++---
> >  1 file changed, 20 insertions(+), 3 deletions(-)
> >
> > diff --git a/arch/arm64/kvm/hyp/pgtable.c b/arch/arm64/kvm/hyp/pgtable.c
> > index b11cf2c618a6c..0858d1fa85d6b 100644
> > --- a/arch/arm64/kvm/hyp/pgtable.c
> > +++ b/arch/arm64/kvm/hyp/pgtable.c
> > @@ -686,6 +686,20 @@ static bool stage2_try_set_pte(const struct kvm_pgtable_visit_ctx *ctx, kvm_pte_
> >       return cmpxchg(ctx->ptep, ctx->old, new) == ctx->old;
> >  }
> >
> > +static void kvm_pgtable_stage2_flush_range(struct kvm_s2_mmu *mmu, u64 start, u64 end,
> > +                                             u32 level, u32 tlb_level)
> > +{
> > +     if (system_supports_tlb_range())
>
> You also check this in __kvm_tlb_flush_range(), ideally this should be
> done exactly once per call.
>
> > +             kvm_call_hyp(__kvm_tlb_flush_range_vmid_ipa, mmu, start, end, level, tlb_level);
> > +     else
> > +             /*
> > +              * Invalidate the whole stage-2, as we may have numerous leaf
> > +              * entries below us which would otherwise need invalidating
> > +              * individually.
> > +              */
> > +             kvm_call_hyp(__kvm_tlb_flush_vmid, mmu);
> > +}
> > +
> >  /**
> >   * stage2_try_break_pte() - Invalidates a pte according to the
> >   *                       'break-before-make' requirements of the
> > @@ -721,10 +735,13 @@ static bool stage2_try_break_pte(const struct kvm_pgtable_visit_ctx *ctx,
> >        * Perform the appropriate TLB invalidation based on the evicted pte
> >        * value (if any).
> >        */
> > -     if (kvm_pte_table(ctx->old, ctx->level))
> > -             kvm_call_hyp(__kvm_tlb_flush_vmid, mmu);
> > -     else if (kvm_pte_valid(ctx->old))
> > +     if (kvm_pte_table(ctx->old, ctx->level)) {
> > +             u64 end = ctx->addr + kvm_granule_size(ctx->level);
> > +
> > +             kvm_pgtable_stage2_flush_range(mmu, ctx->addr, end, ctx->level, 0);
> > +     } else if (kvm_pte_valid(ctx->old)) {
> >               kvm_call_hyp(__kvm_tlb_flush_vmid_ipa, mmu, ctx->addr, ctx->level);
> > +     }
> >
> >       if (stage2_pte_is_counted(ctx->old))
> >               mm_ops->put_page(ctx->ptep);
> > --
> > 2.39.1.519.gcb327c4b5f-goog
> >
> >
ACK on all of the comments. I'll address them in next revision.

Thank you.
Raghavendra
>
> --
> Thanks,
> Oliver