Skip to contentSkip to conversation

x86/kvm/mmu: fix switch between root and guest MMUs

PATCH In review Vitaly Kuznetsov 2 participants 3 messages Feb 22, 2019

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 x86/kvm/mmu: fix switch between root and guest MMUs

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")
---
 arch/x86/include/asm/kvm_host.h |  1 +
 arch/x86/kvm/mmu.c              | 17 +++++++++++++----
 2 files changed, 14 insertions(+), 4 deletions(-)
arch/x86/include/asm/kvm_host.h +1 -0
index be87f71ccaa5..180373360e34 100644
@@ -398,6 +398,7 @@ 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;
arch/x86/kvm/mmu.c +13 -4
index 90ba1a1755a4..f2d1d230d5b8 100644
@@ -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;
 }
@@ -5517,11 +5524,13 @@ 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;
Signature · 1 line
2.20.1
maintainer
#1 lore raw
Quoted 22 lines · Vitaly Kuznetsov

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).

Quoted 7 lines

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")

Is the bug latent until the other patch? Or are they completely separate issues?

Thanks,

Paolo

Quoted 102 lines
---
 arch/x86/include/asm/kvm_host.h |  1 +
 arch/x86/kvm/mmu.c              | 17 +++++++++++++----
 2 files changed, 14 insertions(+), 4 deletions(-)
arch/x86/include/asm/kvm_host.h +1 -0
index be87f71ccaa5..180373360e34 100644
@@ -398,6 +398,7 @@ 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;
arch/x86/kvm/mmu.c +13 -4
index 90ba1a1755a4..f2d1d230d5b8 100644
@@ -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;
 }
@@ -5517,11 +5524,13 @@ 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;
#2 lore raw
Quoted 28 lines · Paolo Bonzini
Quoted 27 lines · Vitaly Kuznetsov

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.

Quoted 6 lines
Quoted 2 lines
Fixes 14c07ad89f4d ("x86/kvm/mmu: introduce guest_mmu")

Is the bug latent until the other patch? Or are they completely separate issues?

This one was reported as https://bugs.launchpad.net/qemu/+bug/1813165
(and I still don't completely understand why it is only SMM which is broken - or is it?). While working on it I noticed that fast_cr3_switch() actually doesn't work (even after this patch when we stop putting incorrect values in the cache with cached_root_available()). So in the end they seem to be fairly independent.

Signature · 1 line
Vitaly