[PATCH] KVM: MMU: Flush guest tlb when reducing permissions on a pte
Avi Kivity [Sat, 6 Jan 2007 00:36:55 +0000 (16:36 -0800)]
If we reduce permissions on a pte, we must flush the cached copy of the pte
from the guest's tlb.

This is implemented at the moment by flushing the entire guest tlb, and can be
improved by flushing just the relevant virtual address, if it is known.

Signed-off-by: Avi Kivity <avi@qumranet.com>
Acked-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

drivers/kvm/mmu.c

index 7761089..2fc2528 100644 (file)
@@ -383,6 +383,7 @@ static void rmap_write_protect(struct kvm_vcpu *vcpu, u64 gfn)
                BUG_ON(!(*spte & PT_WRITABLE_MASK));
                rmap_printk("rmap_write_protect: spte %p %llx\n", spte, *spte);
                rmap_remove(vcpu, spte);
+               kvm_arch_ops->tlb_flush(vcpu);
                *spte &= ~(u64)PT_WRITABLE_MASK;
        }
 }
@@ -594,6 +595,7 @@ static void kvm_mmu_page_unlink_children(struct kvm_vcpu *vcpu,
                                rmap_remove(vcpu, &pt[i]);
                        pt[i] = 0;
                }
+               kvm_arch_ops->tlb_flush(vcpu);
                return;
        }
 
@@ -927,7 +929,10 @@ static inline void set_pte_common(struct kvm_vcpu *vcpu,
                        pgprintk("%s: found shadow page for %lx, marking ro\n",
                                 __FUNCTION__, gfn);
                        access_bits &= ~PT_WRITABLE_MASK;
-                       *shadow_pte &= ~PT_WRITABLE_MASK;
+                       if (is_writeble_pte(*shadow_pte)) {
+                                   *shadow_pte &= ~PT_WRITABLE_MASK;
+                                   kvm_arch_ops->tlb_flush(vcpu);
+                       }
                }
        }