On Wed, May 10, 2023 at 01:30:29PM +0800, Chao Gao wrote:
> On Tue, May 09, 2023 at 09:50:06PM +0800, Yan Zhao wrote:
> >Add a helper to indicate that the kvm_zap_gfn_range() request is to update
> >memory type.
> >
> >Then the zap can be avoided in cases:
> >1. TDP is not enabled.
> >2. EPT is not enabled.
> >
> >This is because only memory type of EPT leaf entries are subjected to
> >change when noncoherent DMA/guest CR0.CD/guest MTRR settings change.
> >
> >Signed-off-by: Yan Zhao <yan.y.zhao@xxxxxxxxx>
> >---
> > arch/x86/kvm/mmu.h | 1 +
> > arch/x86/kvm/mmu/mmu.c | 16 ++++++++++++++++
> > 2 files changed, 17 insertions(+)
> >
> >diff --git a/arch/x86/kvm/mmu.h b/arch/x86/kvm/mmu.h
> >index 92d5a1924fc1..a04577afbc71 100644
> >--- a/arch/x86/kvm/mmu.h
> >+++ b/arch/x86/kvm/mmu.h
> >@@ -236,6 +236,7 @@ static inline u8 permission_fault(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
> > }
> >
> > void kvm_zap_gfn_range(struct kvm *kvm, gfn_t gfn_start, gfn_t gfn_end);
> >+void kvm_zap_gfn_for_memtype(struct kvm *kvm, gfn_t gfn_start, gfn_t gfn_end);
> >
> > int kvm_arch_write_log_dirty(struct kvm_vcpu *vcpu);
> >
> >diff --git a/arch/x86/kvm/mmu/mmu.c b/arch/x86/kvm/mmu/mmu.c
> >index c8961f45e3b1..2706754794d1 100644
> >--- a/arch/x86/kvm/mmu/mmu.c
> >+++ b/arch/x86/kvm/mmu/mmu.c
> >@@ -6272,6 +6272,22 @@ static bool kvm_rmap_zap_gfn_range(struct kvm *kvm, gfn_t gfn_start, gfn_t gfn_e
> > return flush;
> > }
> >
> >+/*
> >+ * Invalidate (zap) TDP SPTEs that cover GFNs from gfn_start and up to gfn_end
> >+ * (not including it) for reason of memory type being updated.
> >+ */
> >+void kvm_zap_gfn_for_memtype(struct kvm *kvm, gfn_t gfn_start, gfn_t gfn_end)
> >+{
> >+ /* Currently only memory type of EPT leaf entries are affected by
> >+ * guest CR0.CD and guest MTRR.
> >+ * So skip invalidation (zap) in other cases
> >+ */
> >+ if (!shadow_memtype_mask)
>
> Do you need to check kvm_arch_has_noncoherent_dma()? if yes, maybe just extract
> the first if() statement and its associated comment from kvm_tdp_page_fault()?
>
This kvm_zap_gfn_for_memtype() helper will be called in
kvm_arch_unregister_noncoherent_dma() as well when noncoherent_dma_count
goes from 1 to 0.
So, checking kvm_arch_has_noncoherent_dma() is not desired.
> >+ return;
> >+
> >+ kvm_zap_gfn_range(kvm, gpa_to_gfn(0), gpa_to_gfn(~0ULL));
>
> This should be:
>
> kvm_zap_gfn_range(kvm, start, end);
>
Oops. sorry about this mistake. Will fix it in next version.
> >+}
> >+
> > /*
> > * Invalidate (zap) SPTEs that cover GFNs from gfn_start and up to gfn_end
> > * (not including it)
> >--
> >2.17.1
> >
