OSDN Git Service

x86/kvm/mmu: fix switch between root and guest MMUs
authorVitaly Kuznetsov <vkuznets@redhat.com>
Fri, 22 Feb 2019 16:45:01 +0000 (17:45 +0100)
committerPaolo Bonzini <pbonzini@redhat.com>
Fri, 22 Feb 2019 18:24:48 +0000 (19:24 +0100)
Commit 14c07ad89f4d ("x86/kvm/mmu: introduce guest_mmu") brought one subtle
change: previously, when switching back from L2 to L1, we were resetting
MMU hooks (like mmu->get_cr3()) in kvm_init_mmu() called from
nested_vmx_load_cr3() and now we do that in nested_ept_uninit_mmu_context()
when we re-target vcpu->arch.mmu pointer.
The change itself looks logical: if nested_ept_init_mmu_context() changes
something than nested_ept_uninit_mmu_context() restores it back. There is,
however, one thing: the following call chain:

 nested_vmx_load_cr3()
  kvm_mmu_new_cr3()
    __kvm_mmu_new_cr3()
      fast_cr3_switch()
        cached_root_available()

now happens with MMU hooks pointing to the new MMU (root MMU in our case)
while previously it was happening with the old one. cached_root_available()
tries to stash current root but it is incorrect to read current CR3 with
mmu->get_cr3(), we need to use old_mmu->get_cr3() which in case we're
switching from L2 to L1 is guest_mmu. (BTW, in shadow page tables case this
is a non-issue because we don't switch MMU).

While we could've tried to guess that we're switching between MMUs and call
the right ->get_cr3() from cached_root_available() this seems to be overly
complicated. Instead, just stash the corresponding CR3 when setting
root_hpa and make cached_root_available() use the stashed value.

Fixes: 14c07ad89f4d ("x86/kvm/mmu: introduce guest_mmu")
Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Cc: stable@vger.kernel.org
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
arch/x86/include/asm/kvm_host.h
arch/x86/kvm/mmu.c

index 4660ce9..593e17b 100644 (file)
@@ -397,6 +397,7 @@ struct kvm_mmu {
        void (*update_pte)(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp,
                           u64 *spte, const void *pte);
        hpa_t root_hpa;
+       gpa_t root_cr3;
        union kvm_mmu_role mmu_role;
        u8 root_level;
        u8 shadow_root_level;
index da9c423..6e62ed3 100644 (file)
@@ -3555,6 +3555,7 @@ void kvm_mmu_free_roots(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
                                                           &invalid_list);
                        mmu->root_hpa = INVALID_PAGE;
                }
+               mmu->root_cr3 = 0;
        }
 
        kvm_mmu_commit_zap_page(vcpu->kvm, &invalid_list);
@@ -3610,6 +3611,7 @@ static int mmu_alloc_direct_roots(struct kvm_vcpu *vcpu)
                vcpu->arch.mmu->root_hpa = __pa(vcpu->arch.mmu->pae_root);
        } else
                BUG();
+       vcpu->arch.mmu->root_cr3 = vcpu->arch.mmu->get_cr3(vcpu);
 
        return 0;
 }
@@ -3618,10 +3620,11 @@ static int mmu_alloc_shadow_roots(struct kvm_vcpu *vcpu)
 {
        struct kvm_mmu_page *sp;
        u64 pdptr, pm_mask;
-       gfn_t root_gfn;
+       gfn_t root_gfn, root_cr3;
        int i;
 
-       root_gfn = vcpu->arch.mmu->get_cr3(vcpu) >> PAGE_SHIFT;
+       root_cr3 = vcpu->arch.mmu->get_cr3(vcpu);
+       root_gfn = root_cr3 >> PAGE_SHIFT;
 
        if (mmu_check_root(vcpu, root_gfn))
                return 1;
@@ -3646,7 +3649,7 @@ static int mmu_alloc_shadow_roots(struct kvm_vcpu *vcpu)
                ++sp->root_count;
                spin_unlock(&vcpu->kvm->mmu_lock);
                vcpu->arch.mmu->root_hpa = root;
-               return 0;
+               goto set_root_cr3;
        }
 
        /*
@@ -3712,6 +3715,9 @@ static int mmu_alloc_shadow_roots(struct kvm_vcpu *vcpu)
                vcpu->arch.mmu->root_hpa = __pa(vcpu->arch.mmu->lm_root);
        }
 
+set_root_cr3:
+       vcpu->arch.mmu->root_cr3 = root_cr3;
+
        return 0;
 }
 
@@ -4163,7 +4169,7 @@ static bool cached_root_available(struct kvm_vcpu *vcpu, gpa_t new_cr3,
        struct kvm_mmu_root_info root;
        struct kvm_mmu *mmu = vcpu->arch.mmu;
 
-       root.cr3 = mmu->get_cr3(vcpu);
+       root.cr3 = mmu->root_cr3;
        root.hpa = mmu->root_hpa;
 
        for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++) {
@@ -4176,6 +4182,7 @@ static bool cached_root_available(struct kvm_vcpu *vcpu, gpa_t new_cr3,
        }
 
        mmu->root_hpa = root.hpa;
+       mmu->root_cr3 = root.cr3;
 
        return i < KVM_MMU_NUM_PREV_ROOTS;
 }
@@ -5516,11 +5523,13 @@ int kvm_mmu_create(struct kvm_vcpu *vcpu)
        vcpu->arch.walk_mmu = &vcpu->arch.root_mmu;
 
        vcpu->arch.root_mmu.root_hpa = INVALID_PAGE;
+       vcpu->arch.root_mmu.root_cr3 = 0;
        vcpu->arch.root_mmu.translate_gpa = translate_gpa;
        for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++)
                vcpu->arch.root_mmu.prev_roots[i] = KVM_MMU_ROOT_INFO_INVALID;
 
        vcpu->arch.guest_mmu.root_hpa = INVALID_PAGE;
+       vcpu->arch.guest_mmu.root_cr3 = 0;
        vcpu->arch.guest_mmu.translate_gpa = translate_gpa;
        for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++)
                vcpu->arch.guest_mmu.prev_roots[i] = KVM_MMU_ROOT_INFO_INVALID;