KVM: MMU: Introduce drop_spte()
authorAvi Kivity <avi@redhat.com>
Sun, 6 Jun 2010 11:31:27 +0000 (14:31 +0300)
committerAvi Kivity <avi@redhat.com>
Mon, 2 Aug 2010 03:40:17 +0000 (06:40 +0300)
When we call rmap_remove(), we (almost) always immediately follow it by
an __set_spte() to a nonpresent pte.  Since we need to perform the two
operations atomically, to avoid losing the dirty and accessed bits, introduce
a helper drop_spte() and convert all call sites.

The operation is still nonatomic at this point.

Signed-off-by: Avi Kivity <avi@redhat.com>
arch/x86/kvm/mmu.c
arch/x86/kvm/paging_tmpl.h

index 70cdf6876b5f92397820bc7d9c65529333f413cc..1ad39cf70e18a6592b90c564cfe985f1e78650cd 100644 (file)
@@ -658,6 +658,12 @@ static void rmap_remove(struct kvm *kvm, u64 *spte)
        }
 }
 
+static void drop_spte(struct kvm *kvm, u64 *sptep, u64 new_spte)
+{
+       rmap_remove(kvm, sptep);
+       __set_spte(sptep, new_spte);
+}
+
 static u64 *rmap_next(struct kvm *kvm, unsigned long *rmapp, u64 *spte)
 {
        struct kvm_rmap_desc *desc;
@@ -722,9 +728,9 @@ static int rmap_write_protect(struct kvm *kvm, u64 gfn)
                        BUG_ON((*spte & (PT_PAGE_SIZE_MASK|PT_PRESENT_MASK)) != (PT_PAGE_SIZE_MASK|PT_PRESENT_MASK));
                        pgprintk("rmap_write_protect(large): spte %p %llx %lld\n", spte, *spte, gfn);
                        if (is_writable_pte(*spte)) {
-                               rmap_remove(kvm, spte);
+                               drop_spte(kvm, spte,
+                                         shadow_trap_nonpresent_pte);
                                --kvm->stat.lpages;
-                               __set_spte(spte, shadow_trap_nonpresent_pte);
                                spte = NULL;
                                write_protected = 1;
                        }
@@ -744,8 +750,7 @@ static int kvm_unmap_rmapp(struct kvm *kvm, unsigned long *rmapp,
        while ((spte = rmap_next(kvm, rmapp, NULL))) {
                BUG_ON(!(*spte & PT_PRESENT_MASK));
                rmap_printk("kvm_rmap_unmap_hva: spte %p %llx\n", spte, *spte);
-               rmap_remove(kvm, spte);
-               __set_spte(spte, shadow_trap_nonpresent_pte);
+               drop_spte(kvm, spte, shadow_trap_nonpresent_pte);
                need_tlb_flush = 1;
        }
        return need_tlb_flush;
@@ -767,8 +772,7 @@ static int kvm_set_pte_rmapp(struct kvm *kvm, unsigned long *rmapp,
                rmap_printk("kvm_set_pte_rmapp: spte %p %llx\n", spte, *spte);
                need_flush = 1;
                if (pte_write(*ptep)) {
-                       rmap_remove(kvm, spte);
-                       __set_spte(spte, shadow_trap_nonpresent_pte);
+                       drop_spte(kvm, spte, shadow_trap_nonpresent_pte);
                        spte = rmap_next(kvm, rmapp, NULL);
                } else {
                        new_spte = *spte &~ (PT64_BASE_ADDR_MASK);
@@ -1464,7 +1468,8 @@ static void kvm_mmu_page_unlink_children(struct kvm *kvm,
                        } else {
                                if (is_large_pte(ent))
                                        --kvm->stat.lpages;
-                               rmap_remove(kvm, &pt[i]);
+                               drop_spte(kvm, &pt[i],
+                                         shadow_trap_nonpresent_pte);
                        }
                }
                pt[i] = shadow_trap_nonpresent_pte;
@@ -1868,9 +1873,8 @@ static int set_spte(struct kvm_vcpu *vcpu, u64 *sptep,
                if (level > PT_PAGE_TABLE_LEVEL &&
                    has_wrprotected_page(vcpu->kvm, gfn, level)) {
                        ret = 1;
-                       rmap_remove(vcpu->kvm, sptep);
-                       spte = shadow_trap_nonpresent_pte;
-                       goto set_pte;
+                       drop_spte(vcpu->kvm, sptep, shadow_trap_nonpresent_pte);
+                       goto done;
                }
 
                spte |= PT_WRITABLE_MASK;
@@ -1902,6 +1906,7 @@ static int set_spte(struct kvm_vcpu *vcpu, u64 *sptep,
 
 set_pte:
        __set_spte(sptep, spte);
+done:
        return ret;
 }
 
@@ -1938,8 +1943,7 @@ static void mmu_set_spte(struct kvm_vcpu *vcpu, u64 *sptep,
                } else if (pfn != spte_to_pfn(*sptep)) {
                        pgprintk("hfn old %lx new %lx\n",
                                 spte_to_pfn(*sptep), pfn);
-                       rmap_remove(vcpu->kvm, sptep);
-                       __set_spte(sptep, shadow_trap_nonpresent_pte);
+                       drop_spte(vcpu->kvm, sptep, shadow_trap_nonpresent_pte);
                        kvm_flush_remote_tlbs(vcpu->kvm);
                } else
                        was_rmapped = 1;
@@ -2591,7 +2595,7 @@ static void mmu_pte_write_zap_pte(struct kvm_vcpu *vcpu,
        pte = *spte;
        if (is_shadow_present_pte(pte)) {
                if (is_last_spte(pte, sp->role.level))
-                       rmap_remove(vcpu->kvm, spte);
+                       drop_spte(vcpu->kvm, spte, shadow_trap_nonpresent_pte);
                else {
                        child = page_header(pte & PT64_BASE_ADDR_MASK);
                        mmu_page_remove_parent_pte(child, spte);
index 59e750c1a269765780f887c33880570dab23d5af..796a325c7e595fd705227f9c0f762e95a0072372 100644 (file)
@@ -353,8 +353,7 @@ static u64 *FNAME(fetch)(struct kvm_vcpu *vcpu, gva_t addr,
                }
 
                if (is_large_pte(*sptep)) {
-                       rmap_remove(vcpu->kvm, sptep);
-                       __set_spte(sptep, shadow_trap_nonpresent_pte);
+                       drop_spte(vcpu->kvm, sptep, shadow_trap_nonpresent_pte);
                        kvm_flush_remote_tlbs(vcpu->kvm);
                }
 
@@ -516,12 +515,13 @@ static void FNAME(invlpg)(struct kvm_vcpu *vcpu, gva_t gva)
                        pte_gpa += (sptep - sp->spt) * sizeof(pt_element_t);
 
                        if (is_shadow_present_pte(*sptep)) {
-                               rmap_remove(vcpu->kvm, sptep);
                                if (is_large_pte(*sptep))
                                        --vcpu->kvm->stat.lpages;
+                               drop_spte(vcpu->kvm, sptep,
+                                         shadow_trap_nonpresent_pte);
                                need_flush = 1;
-                       }
-                       __set_spte(sptep, shadow_trap_nonpresent_pte);
+                       } else
+                               __set_spte(sptep, shadow_trap_nonpresent_pte);
                        break;
                }
 
@@ -637,12 +637,11 @@ static int FNAME(sync_page)(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp,
                      !is_present_gpte(gpte) || !(gpte & PT_ACCESSED_MASK)) {
                        u64 nonpresent;
 
-                       rmap_remove(vcpu->kvm, &sp->spt[i]);
                        if (is_present_gpte(gpte) || !clear_unsync)
                                nonpresent = shadow_trap_nonpresent_pte;
                        else
                                nonpresent = shadow_notrap_nonpresent_pte;
-                       __set_spte(&sp->spt[i], nonpresent);
+                       drop_spte(vcpu->kvm, &sp->spt[i], nonpresent);
                        continue;
                }