]> pilppa.com Git - linux-2.6-omap-h63xx.git/commitdiff
KVM: MMU: Avoid unnecessary remote tlb flushes when guest updates a pte
authorAvi Kivity <avi@qumranet.com>
Wed, 21 Nov 2007 00:06:21 +0000 (02:06 +0200)
committerAvi Kivity <avi@qumranet.com>
Wed, 30 Jan 2008 15:53:10 +0000 (17:53 +0200)
If all we're doing is increasing permissions on a pte (typical for demand
paging), then there's not need to flush remote tlbs.  Worst case they'll
get a spurious page fault.

Signed-off-by: Avi Kivity <avi@qumranet.com>
drivers/kvm/mmu.c

index 101cd5377a89716377021baebe68733b14e5ca85..281dd5f9310c9b7a3e0fd8c4d29ee88501ae38f3 100644 (file)
@@ -134,6 +134,8 @@ static int dbg = 1;
 #define PT32_DIR_BASE_ADDR_MASK \
        (PAGE_MASK & ~((1ULL << (PAGE_SHIFT + PT32_LEVEL_BITS)) - 1))
 
+#define PT64_PERM_MASK (PT_PRESENT_MASK | PT_WRITABLE_MASK | PT_USER_MASK \
+                       | PT64_NX_MASK)
 
 #define PFERR_PRESENT_MASK (1U << 0)
 #define PFERR_WRITE_MASK (1U << 1)
@@ -1227,7 +1229,6 @@ static void mmu_pte_write_zap_pte(struct kvm_vcpu *vcpu,
                }
        }
        set_shadow_pte(spte, shadow_trap_nonpresent_pte);
-       kvm_flush_remote_tlbs(vcpu->kvm);
 }
 
 static void mmu_pte_write_new_pte(struct kvm_vcpu *vcpu,
@@ -1250,6 +1251,27 @@ static void mmu_pte_write_new_pte(struct kvm_vcpu *vcpu,
                                    offset_in_pte);
 }
 
+static bool need_remote_flush(u64 old, u64 new)
+{
+       if (!is_shadow_present_pte(old))
+               return false;
+       if (!is_shadow_present_pte(new))
+               return true;
+       if ((old ^ new) & PT64_BASE_ADDR_MASK)
+               return true;
+       old ^= PT64_NX_MASK;
+       new ^= PT64_NX_MASK;
+       return (old & ~new & PT64_PERM_MASK) != 0;
+}
+
+static void mmu_pte_write_flush_tlb(struct kvm_vcpu *vcpu, u64 old, u64 new)
+{
+       if (need_remote_flush(old, new))
+               kvm_flush_remote_tlbs(vcpu->kvm);
+       else
+               kvm_mmu_flush_tlb(vcpu);
+}
+
 static bool last_updated_pte_accessed(struct kvm_vcpu *vcpu)
 {
        u64 *spte = vcpu->last_pte_updated;
@@ -1265,6 +1287,7 @@ void kvm_mmu_pte_write(struct kvm_vcpu *vcpu, gpa_t gpa,
        struct hlist_node *node, *n;
        struct hlist_head *bucket;
        unsigned index;
+       u64 entry;
        u64 *spte;
        unsigned offset = offset_in_page(gpa);
        unsigned pte_size;
@@ -1335,9 +1358,11 @@ void kvm_mmu_pte_write(struct kvm_vcpu *vcpu, gpa_t gpa,
                }
                spte = &page->spt[page_offset / sizeof(*spte)];
                while (npte--) {
+                       entry = *spte;
                        mmu_pte_write_zap_pte(vcpu, page, spte);
                        mmu_pte_write_new_pte(vcpu, page, spte, new, bytes,
                                              page_offset & (pte_size - 1));
+                       mmu_pte_write_flush_tlb(vcpu, entry, *spte);
                        ++spte;
                }
        }