Skip to contentSkip to conversation

SVM 5-level page table support

PATCH Superseded Wei Huang 5 participants 19 messages Aug 8, 2021 – Aug 10, 2021

Subsystems (3)

v2 of 3 Superseded Latest is v3 →

lore
Share & tools
lore

Wei Huang

author
#0 lore raw

v2 cover SVM 5-level page table support

This patch set adds 5-level page table support for AMD SVM. When the 5-level page table is enabled on host OS, the nested page table for guest VMs will use the same format as host OS (i.e. 5-level NPT). These patches were tested with various combination of different settings and test cases (nested/regular VMs, AMD64/i686 kernels, kvm-unit-tests, etc.)

v1->v2:
* Remove v1's arch-specific get_tdp_level() and add a new parameter, tdp_forced_root_level, to allow forced TDP level (Sean)
* Add additional comment on tdp_root table chaining trick and change the PML root table allocation code (Sean)
* Revise Patch 1's commit msg (Sean and Jim)

Thanks,
-Wei

Wei Huang (3):
KVM: x86: Allow CPU to force vendor-specific TDP level
KVM: x86: Handle the case of 5-level shadow page table
KVM: SVM: Add 5-level page table support for SVM

 arch/x86/include/asm/kvm_host.h |  6 ++--
 arch/x86/kvm/mmu/mmu.c          | 58 ++++++++++++++++++++++-----------
 arch/x86/kvm/svm/svm.c          | 13 ++++----
 arch/x86/kvm/vmx/vmx.c          |  3 +-
 4 files changed, 50 insertions(+), 30 deletions(-)
Signature · 1 line
2.31.1
author
#1 lore raw

PATCH v2 1/3 KVM: x86: Allow CPU to force vendor-specific TDP level

AMD future CPUs will require a 5-level NPT if host CR4.LA57 is set. To prevent kvm_mmu_get_tdp_level() from incorrectly changing NPT level on behalf of CPUs, add a new parameter in kvm_configure_mmu() to force a fixed TDP level.

---
 arch/x86/include/asm/kvm_host.h |  5 ++---
 arch/x86/kvm/mmu/mmu.c          | 10 ++++++++--
 arch/x86/kvm/svm/svm.c          |  4 +++-
 arch/x86/kvm/vmx/vmx.c          |  3 ++-
 4 files changed, 15 insertions(+), 7 deletions(-)
arch/x86/include/asm/kvm_host.h +2 -3
index 974cbfb1eefe..6d16f75cc8da 100644
@@ -723,7 +723,6 @@ struct kvm_vcpu_arch {
 
 	u64 reserved_gpa_bits;
 	int maxphyaddr;
-	int max_tdp_level;
 
 	/* emulate context */
 
@@ -1747,8 +1746,8 @@ void kvm_mmu_invalidate_gva(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
 void kvm_mmu_invpcid_gva(struct kvm_vcpu *vcpu, gva_t gva, unsigned long pcid);
 void kvm_mmu_new_pgd(struct kvm_vcpu *vcpu, gpa_t new_pgd);
 
-void kvm_configure_mmu(bool enable_tdp, int tdp_max_root_level,
-		       int tdp_huge_page_level);
+void kvm_configure_mmu(bool enable_tdp, int tdp_forced_root_level,
+		       int tdp_max_root_level, int tdp_huge_page_level);
 
 static inline u16 kvm_read_ldt(void)
 {
arch/x86/kvm/mmu/mmu.c +8 -2
index 66f7f5bc3482..c11ee4531f6d 100644
@@ -97,6 +97,7 @@ module_param_named(flush_on_reuse, force_flush_and_sync_on_reuse, bool, 0644);
 bool tdp_enabled = false;
 
 static int max_huge_page_level __read_mostly;
+static int tdp_root_level __read_mostly;
 static int max_tdp_level __read_mostly;
 
 enum {
@@ -4562,6 +4563,10 @@ static union kvm_mmu_role kvm_calc_mmu_role_common(struct kvm_vcpu *vcpu,
 
 static inline int kvm_mmu_get_tdp_level(struct kvm_vcpu *vcpu)
 {
+	/* tdp_root_level is architecture forced level, use it if nonzero */
+	if (tdp_root_level)
+		return tdp_root_level;
+
 	/* Use 5-level TDP if and only if it's useful/necessary. */
 	if (max_tdp_level == 5 && cpuid_maxphyaddr(vcpu) <= 48)
 		return 4;
@@ -5253,10 +5258,11 @@ void kvm_mmu_invpcid_gva(struct kvm_vcpu *vcpu, gva_t gva, unsigned long pcid)
 	 */
 }
 
-void kvm_configure_mmu(bool enable_tdp, int tdp_max_root_level,
-		       int tdp_huge_page_level)
+void kvm_configure_mmu(bool enable_tdp, int tdp_forced_root_level,
+		       int tdp_max_root_level, int tdp_huge_page_level)
 {
 	tdp_enabled = enable_tdp;
+	tdp_root_level = tdp_forced_root_level;
 	max_tdp_level = tdp_max_root_level;
 
 	/*
arch/x86/kvm/svm/svm.c +3 -1
index e8ccab50ebf6..f361d466e18e 100644
@@ -1015,7 +1015,9 @@ static __init int svm_hardware_setup(void)
 	if (!boot_cpu_has(X86_FEATURE_NPT))
 		npt_enabled = false;
 
-	kvm_configure_mmu(npt_enabled, get_max_npt_level(), PG_LEVEL_1G);
+	/* Force VM NPT level equal to the host's max NPT level */
+	kvm_configure_mmu(npt_enabled, get_max_npt_level(),
+			  get_max_npt_level(), PG_LEVEL_1G);
 	pr_info("kvm: Nested Paging %sabled\n", npt_enabled ? "en" : "dis");
 
 	/* Note, SEV setup consumes npt_enabled. */
arch/x86/kvm/vmx/vmx.c +2 -1
index 927a552393b9..034e1397c7d5 100644
@@ -7803,7 +7803,8 @@ static __init int hardware_setup(void)
 		ept_lpage_level = PG_LEVEL_2M;
 	else
 		ept_lpage_level = PG_LEVEL_4K;
-	kvm_configure_mmu(enable_ept, vmx_get_max_tdp_level(), ept_lpage_level);
+	kvm_configure_mmu(enable_ept, 0, vmx_get_max_tdp_level(),
+			  ept_lpage_level);
 
 	/*
 	 * Only enable PML when hardware supports PML feature, and both EPT
Signature · 1 line
2.31.1
#2 lore raw

Re: [PATCH v2 1/3] KVM: x86: Allow CPU to force vendor-specific TDP level

Quoted 1 line · Wei Huang

AMD future CPUs will require a 5-level NPT if host CR4.LA57 is set.

Sorry, but why? NPT is not indexed by HVA.

Quoted 40 lines

To prevent kvm_mmu_get_tdp_level() from incorrectly changing NPT level on behalf of CPUs, add a new parameter in kvm_configure_mmu() to force a fixed TDP level.

---
 arch/x86/include/asm/kvm_host.h |  5 ++---
 arch/x86/kvm/mmu/mmu.c          | 10 ++++++++--
 arch/x86/kvm/svm/svm.c          |  4 +++-
 arch/x86/kvm/vmx/vmx.c          |  3 ++-
 4 files changed, 15 insertions(+), 7 deletions(-)
arch/x86/include/asm/kvm_host.h +2 -3
index 974cbfb1eefe..6d16f75cc8da 100644
@@ -723,7 +723,6 @@ struct kvm_vcpu_arch {
 
 	u64 reserved_gpa_bits;
 	int maxphyaddr;
-	int max_tdp_level;
 
 	/* emulate context */
 
@@ -1747,8 +1746,8 @@ void kvm_mmu_invalidate_gva(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
 void kvm_mmu_invpcid_gva(struct kvm_vcpu *vcpu, gva_t gva, unsigned long pcid);
 void kvm_mmu_new_pgd(struct kvm_vcpu *vcpu, gpa_t new_pgd);
 
-void kvm_configure_mmu(bool enable_tdp, int tdp_max_root_level,
-		       int tdp_huge_page_level);
+void kvm_configure_mmu(bool enable_tdp, int tdp_forced_root_level,
+		       int tdp_max_root_level, int tdp_huge_page_level);
 
 static inline u16 kvm_read_ldt(void)
 {
arch/x86/kvm/mmu/mmu.c +0 -0
index 66f7f5bc3482..c11ee4531f6d 100644

arch/x86/kvm/mmu/mmu.c module_param_named(flush_on_reuse, force_flush_and_sync_on_reuse, bool, 0644); · line 100

Quoted 4 lines
arch/x86/kvm/mmu/mmu.c +1 -0
@@ -97,6 +97,7 @@ module_param_named(flush_on_reuse, force_flush_and_sync_on_reuse, bool, 0644);
 bool tdp_enabled = false;
 
 static int max_huge_page_level __read_mostly;
+static int tdp_root_level __read_mostly;

I think this is a broken design - meaning KVM can only use 5-level or 4-level NPT for all VMs.

B.R.
Yu

Quoted 60 lines
arch/x86/kvm/mmu/mmu.c +7 -2
 static int max_tdp_level __read_mostly;
 
 enum {
@@ -4562,6 +4563,10 @@ static union kvm_mmu_role kvm_calc_mmu_role_common(struct kvm_vcpu *vcpu,
 
 static inline int kvm_mmu_get_tdp_level(struct kvm_vcpu *vcpu)
 {
+	/* tdp_root_level is architecture forced level, use it if nonzero */
+	if (tdp_root_level)
+		return tdp_root_level;
+
 	/* Use 5-level TDP if and only if it's useful/necessary. */
 	if (max_tdp_level == 5 && cpuid_maxphyaddr(vcpu) <= 48)
 		return 4;
@@ -5253,10 +5258,11 @@ void kvm_mmu_invpcid_gva(struct kvm_vcpu *vcpu, gva_t gva, unsigned long pcid)
 	 */
 }
 
-void kvm_configure_mmu(bool enable_tdp, int tdp_max_root_level,
-		       int tdp_huge_page_level)
+void kvm_configure_mmu(bool enable_tdp, int tdp_forced_root_level,
+		       int tdp_max_root_level, int tdp_huge_page_level)
 {
 	tdp_enabled = enable_tdp;
+	tdp_root_level = tdp_forced_root_level;
 	max_tdp_level = tdp_max_root_level;
 
 	/*
arch/x86/kvm/svm/svm.c +3 -1
index e8ccab50ebf6..f361d466e18e 100644
@@ -1015,7 +1015,9 @@ static __init int svm_hardware_setup(void)
 	if (!boot_cpu_has(X86_FEATURE_NPT))
 		npt_enabled = false;
 
-	kvm_configure_mmu(npt_enabled, get_max_npt_level(), PG_LEVEL_1G);
+	/* Force VM NPT level equal to the host's max NPT level */
+	kvm_configure_mmu(npt_enabled, get_max_npt_level(),
+			  get_max_npt_level(), PG_LEVEL_1G);
 	pr_info("kvm: Nested Paging %sabled\n", npt_enabled ? "en" : "dis");
 
 	/* Note, SEV setup consumes npt_enabled. */
arch/x86/kvm/vmx/vmx.c +2 -1
index 927a552393b9..034e1397c7d5 100644
@@ -7803,7 +7803,8 @@ static __init int hardware_setup(void)
 		ept_lpage_level = PG_LEVEL_2M;
 	else
 		ept_lpage_level = PG_LEVEL_4K;
-	kvm_configure_mmu(enable_ept, vmx_get_max_tdp_level(), ept_lpage_level);
+	kvm_configure_mmu(enable_ept, 0, vmx_get_max_tdp_level(),
+			  ept_lpage_level);
 
 	/*
 	 * Only enable PML when hardware supports PML feature, and both EPT
Signature · 1 line
2.31.1
author
#3 lore raw

Re: [PATCH v2 1/3] KVM: x86: Allow CPU to force vendor-specific TDP level

Quoted 4 lines · Yu Zhang
Quoted 1 line · Wei Huang

AMD future CPUs will require a 5-level NPT if host CR4.LA57 is set.

Sorry, but why? NPT is not indexed by HVA.

NPT is not indexed by HVA - it is always indexed by GPA. What I meant is NPT page table level has to be the same as the host OS page table: if 5-level page table is enabled in host OS (CR4.LA57=1), guest NPT has to 5-level too.

Quoted 43 lines
Quoted 43 lines

To prevent kvm_mmu_get_tdp_level() from incorrectly changing NPT level on behalf of CPUs, add a new parameter in kvm_configure_mmu() to force a fixed TDP level.

---
  arch/x86/include/asm/kvm_host.h |  5 ++---
  arch/x86/kvm/mmu/mmu.c          | 10 ++++++++--
  arch/x86/kvm/svm/svm.c          |  4 +++-
  arch/x86/kvm/vmx/vmx.c          |  3 ++-
  4 files changed, 15 insertions(+), 7 deletions(-)
arch/x86/include/asm/kvm_host.h +2 -3
index 974cbfb1eefe..6d16f75cc8da 100644
@@ -723,7 +723,6 @@ struct kvm_vcpu_arch {
 
  	u64 reserved_gpa_bits;
  	int maxphyaddr;
-	int max_tdp_level;
 
  	/* emulate context */
 
@@ -1747,8 +1746,8 @@ void kvm_mmu_invalidate_gva(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
  void kvm_mmu_invpcid_gva(struct kvm_vcpu *vcpu, gva_t gva, unsigned long pcid);
  void kvm_mmu_new_pgd(struct kvm_vcpu *vcpu, gpa_t new_pgd);
 
-void kvm_configure_mmu(bool enable_tdp, int tdp_max_root_level,
-		       int tdp_huge_page_level);
+void kvm_configure_mmu(bool enable_tdp, int tdp_forced_root_level,
+		       int tdp_max_root_level, int tdp_huge_page_level);
 
  static inline u16 kvm_read_ldt(void)
  {
arch/x86/kvm/mmu/mmu.c +0 -0
index 66f7f5bc3482..c11ee4531f6d 100644
@@ -97,6 +97,7 @@ module_param_named(flush_on_reuse, force_flush_and_sync_on_reuse, bool, 0644);
  bool tdp_enabled = false;
 
Quoted 5 lines

arch/x86/kvm/mmu/mmu.c module_param_named(flush_on_reuse, force_flush_and_sync_on_reuse, bool, 0644); · line 100

Quoted 2 lines
arch/x86/kvm/mmu/mmu.c +1 -0
  static int max_huge_page_level __read_mostly;
+static int tdp_root_level __read_mostly;

I think this is a broken design - meaning KVM can only use 5-level or 4-level NPT for all VMs.

Broken normally means non-functional or buggy, which doesn't apply here. A good TLB design should be able to offset the potential overhead of 5-level page table for most cases.

Quoted 64 lines

B.R.
Yu

Quoted 60 lines
arch/x86/kvm/mmu/mmu.c +7 -2
  static int max_tdp_level __read_mostly;
 
  enum {
@@ -4562,6 +4563,10 @@ static union kvm_mmu_role kvm_calc_mmu_role_common(struct kvm_vcpu *vcpu,
 
  static inline int kvm_mmu_get_tdp_level(struct kvm_vcpu *vcpu)
  {
+	/* tdp_root_level is architecture forced level, use it if nonzero */
+	if (tdp_root_level)
+		return tdp_root_level;
+
  	/* Use 5-level TDP if and only if it's useful/necessary. */
  	if (max_tdp_level == 5 && cpuid_maxphyaddr(vcpu) <= 48)
  		return 4;
@@ -5253,10 +5258,11 @@ void kvm_mmu_invpcid_gva(struct kvm_vcpu *vcpu, gva_t gva, unsigned long pcid)
  	 */
  }
 
-void kvm_configure_mmu(bool enable_tdp, int tdp_max_root_level,
-		       int tdp_huge_page_level)
+void kvm_configure_mmu(bool enable_tdp, int tdp_forced_root_level,
+		       int tdp_max_root_level, int tdp_huge_page_level)
  {
  	tdp_enabled = enable_tdp;
+	tdp_root_level = tdp_forced_root_level;
  	max_tdp_level = tdp_max_root_level;
 
  	/*
arch/x86/kvm/svm/svm.c +3 -1
index e8ccab50ebf6..f361d466e18e 100644
@@ -1015,7 +1015,9 @@ static __init int svm_hardware_setup(void)
  	if (!boot_cpu_has(X86_FEATURE_NPT))
  		npt_enabled = false;
 
-	kvm_configure_mmu(npt_enabled, get_max_npt_level(), PG_LEVEL_1G);
+	/* Force VM NPT level equal to the host's max NPT level */
+	kvm_configure_mmu(npt_enabled, get_max_npt_level(),
+			  get_max_npt_level(), PG_LEVEL_1G);
  	pr_info("kvm: Nested Paging %sabled\n", npt_enabled ? "en" : "dis");
 
  	/* Note, SEV setup consumes npt_enabled. */
arch/x86/kvm/vmx/vmx.c +2 -1
index 927a552393b9..034e1397c7d5 100644
@@ -7803,7 +7803,8 @@ static __init int hardware_setup(void)
  		ept_lpage_level = PG_LEVEL_2M;
  	else
  		ept_lpage_level = PG_LEVEL_4K;
-	kvm_configure_mmu(enable_ept, vmx_get_max_tdp_level(), ept_lpage_level);
+	kvm_configure_mmu(enable_ept, 0, vmx_get_max_tdp_level(),
+			  ept_lpage_level);
 
  	/*
  	 * Only enable PML when hardware supports PML feature, and both EPT
Signature · 1 line
2.31.1
#4 lore raw

Re: [PATCH v2 1/3] KVM: x86: Allow CPU to force vendor-specific TDP level

Quoted 11 lines · Wei Huang
Quoted 4 lines · Yu Zhang
Quoted 1 line · Wei Huang

AMD future CPUs will require a 5-level NPT if host CR4.LA57 is set.

Sorry, but why? NPT is not indexed by HVA.

NPT is not indexed by HVA - it is always indexed by GPA. What I meant is NPT page table level has to be the same as the host OS page table: if 5-level page table is enabled in host OS (CR4.LA57=1), guest NPT has to 5-level too.

I know what you meant. But may I ask why?

B.R.
Yu

Quoted 105 lines
Quoted 35 lines
Quoted 35 lines

To prevent kvm_mmu_get_tdp_level() from incorrectly changing NPT level on behalf of CPUs, add a new parameter in kvm_configure_mmu() to force a fixed TDP level.

---
  arch/x86/include/asm/kvm_host.h |  5 ++---
  arch/x86/kvm/mmu/mmu.c          | 10 ++++++++--
  arch/x86/kvm/svm/svm.c          |  4 +++-
  arch/x86/kvm/vmx/vmx.c          |  3 ++-
  4 files changed, 15 insertions(+), 7 deletions(-)
arch/x86/include/asm/kvm_host.h +2 -3
index 974cbfb1eefe..6d16f75cc8da 100644
@@ -723,7 +723,6 @@ struct kvm_vcpu_arch {
  	u64 reserved_gpa_bits;
  	int maxphyaddr;
-	int max_tdp_level;
  	/* emulate context */
@@ -1747,8 +1746,8 @@ void kvm_mmu_invalidate_gva(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
  void kvm_mmu_invpcid_gva(struct kvm_vcpu *vcpu, gva_t gva, unsigned long pcid);
  void kvm_mmu_new_pgd(struct kvm_vcpu *vcpu, gpa_t new_pgd);
-void kvm_configure_mmu(bool enable_tdp, int tdp_max_root_level,
-		       int tdp_huge_page_level);
+void kvm_configure_mmu(bool enable_tdp, int tdp_forced_root_level,
+		       int tdp_max_root_level, int tdp_huge_page_level);
  static inline u16 kvm_read_ldt(void)
  {
arch/x86/kvm/mmu/mmu.c +0 -1
index 66f7f5bc3482..c11ee4531f6d 100644
--- a/arch/x86/kvm/mmu/mmu.c
Quoted 8 lines

arch/x86/kvm/mmu/mmu.c module_param_named(flush_on_reuse, force_flush_and_sync_on_reuse, bool, 0644); · line 99

Quoted 5 lines

+++ b/arch/x86/kvm/mmu/mmu.c

arch/x86/kvm/mmu/mmu.c +1 -0
@@ -97,6 +97,7 @@ module_param_named(flush_on_reuse, force_flush_and_sync_on_reuse, bool, 0644);
  bool tdp_enabled = false;
  static int max_huge_page_level __read_mostly;
+static int tdp_root_level __read_mostly;

I think this is a broken design - meaning KVM can only use 5-level or 4-level NPT for all VMs.

Broken normally means non-functional or buggy, which doesn't apply here. A good TLB design should be able to offset the potential overhead of 5-level page table for most cases.

Quoted 57 lines

B.R.
Yu

Quoted 53 lines
arch/x86/kvm/mmu/mmu.c +7 -2
  static int max_tdp_level __read_mostly;
  enum {
@@ -4562,6 +4563,10 @@ static union kvm_mmu_role kvm_calc_mmu_role_common(struct kvm_vcpu *vcpu,
  static inline int kvm_mmu_get_tdp_level(struct kvm_vcpu *vcpu)
  {
+	/* tdp_root_level is architecture forced level, use it if nonzero */
+	if (tdp_root_level)
+		return tdp_root_level;
+
  	/* Use 5-level TDP if and only if it's useful/necessary. */
  	if (max_tdp_level == 5 && cpuid_maxphyaddr(vcpu) <= 48)
  		return 4;
@@ -5253,10 +5258,11 @@ void kvm_mmu_invpcid_gva(struct kvm_vcpu *vcpu, gva_t gva, unsigned long pcid)
  	 */
  }
-void kvm_configure_mmu(bool enable_tdp, int tdp_max_root_level,
-		       int tdp_huge_page_level)
+void kvm_configure_mmu(bool enable_tdp, int tdp_forced_root_level,
+		       int tdp_max_root_level, int tdp_huge_page_level)
  {
  	tdp_enabled = enable_tdp;
+	tdp_root_level = tdp_forced_root_level;
  	max_tdp_level = tdp_max_root_level;
  	/*
arch/x86/kvm/svm/svm.c +3 -1
index e8ccab50ebf6..f361d466e18e 100644
@@ -1015,7 +1015,9 @@ static __init int svm_hardware_setup(void)
  	if (!boot_cpu_has(X86_FEATURE_NPT))
  		npt_enabled = false;
-	kvm_configure_mmu(npt_enabled, get_max_npt_level(), PG_LEVEL_1G);
+	/* Force VM NPT level equal to the host's max NPT level */
+	kvm_configure_mmu(npt_enabled, get_max_npt_level(),
+			  get_max_npt_level(), PG_LEVEL_1G);
  	pr_info("kvm: Nested Paging %sabled\n", npt_enabled ? "en" : "dis");
  	/* Note, SEV setup consumes npt_enabled. */
arch/x86/kvm/vmx/vmx.c +2 -1
index 927a552393b9..034e1397c7d5 100644
@@ -7803,7 +7803,8 @@ static __init int hardware_setup(void)
  		ept_lpage_level = PG_LEVEL_2M;
  	else
  		ept_lpage_level = PG_LEVEL_4K;
-	kvm_configure_mmu(enable_ept, vmx_get_max_tdp_level(), ept_lpage_level);
+	kvm_configure_mmu(enable_ept, 0, vmx_get_max_tdp_level(),
+			  ept_lpage_level);
  	/*
  	 * Only enable PML when hardware supports PML feature, and both EPT
Signature · 1 line
2.31.1
author
#5 lore raw

Re: [PATCH v2 1/3] KVM: x86: Allow CPU to force vendor-specific TDP level

Quoted 7 lines · Yu Zhang
Quoted 6 lines · Wei Huang
Quoted 3 lines · Yu Zhang
Quoted 1 line · Wei Huang

AMD future CPUs will require a 5-level NPT if host CR4.LA57 is set.

Quoted 7 lines
Quoted 5 lines
Quoted 1 line

Sorry, but why? NPT is not indexed by HVA.

NPT is not indexed by HVA - it is always indexed by GPA. What I meant is NPT page table level has to be the same as the host OS page table: if 5-level page table is enabled in host OS (CR4.LA57=1), guest NPT has to 5-level too.

I know what you meant. But may I ask why?

I don't have a good answer for it. From what I know, VMCB doesn't have a field to indicate guest page table level. As a result, hardware relies on host CR4 to infer NPT level.

Quoted 5 lines

B.R.
Yu

Quoted 1 line
#6 lore raw

Re: [PATCH v2 1/3] KVM: x86: Allow CPU to force vendor-specific TDP level

Quoted 16 lines · Wei Huang
Quoted 13 lines · Yu Zhang
Quoted 11 lines · Wei Huang
Quoted 4 lines · Yu Zhang
Quoted 1 line · Wei Huang

AMD future CPUs will require a 5-level NPT if host CR4.LA57 is set.

Sorry, but why? NPT is not indexed by HVA.

NPT is not indexed by HVA - it is always indexed by GPA. What I meant is NPT page table level has to be the same as the host OS page table: if 5-level page table is enabled in host OS (CR4.LA57=1), guest NPT has to 5-level too.

Quoted 5 lines
Quoted 1 line

I know what you meant. But may I ask why?

I don't have a good answer for it. From what I know, VMCB doesn't have a field to indicate guest page table level. As a result, hardware relies on host CR4 to infer NPT level.

I guess you mean not even in the N_CR3 field of VMCB?

Then it's not a broken design - it's a limitation of SVM. :)

B.R.
Yu

#7 lore raw

Re: [PATCH v2 1/3] KVM: x86: Allow CPU to force vendor-specific TDP level

Quoted 16 lines · Yu Zhang
Quoted 15 lines · Wei Huang
Quoted 13 lines · Yu Zhang
Quoted 11 lines · Wei Huang
Quoted 4 lines · Yu Zhang
Quoted 1 line · Wei Huang

AMD future CPUs will require a 5-level NPT if host CR4.LA57 is set.

Sorry, but why? NPT is not indexed by HVA.

NPT is not indexed by HVA - it is always indexed by GPA. What I meant is NPT page table level has to be the same as the host OS page table: if 5-level page table is enabled in host OS (CR4.LA57=1), guest NPT has to 5-level too.

Quoted 7 lines
Quoted 5 lines
Quoted 1 line

I know what you meant. But may I ask why?

I don't have a good answer for it. From what I know, VMCB doesn't have a field to indicate guest page table level. As a result, hardware relies on host CR4 to infer NPT level.

I guess you mean not even in the N_CR3 field of VMCB?

Correct, nCR3 is a basically a pure representation of a regular CR3.

Quoted 1 line

Then it's not a broken design - it's a limitation of SVM. :)

That's just a polite way of saying it's a broken design ;-)

Joking aside, NPT opted for a semblance of backwards compatibility at the cost of having to carry all the baggage that comes with a legacy design. Keeping the core functionality from IA32 paging presumably miminizes design and hardware costs, and required minimal enabling in hypervisors. The downside is that it's less flexible than EPT and has a few warts, e.g. shadowing NPT is gross because the host can't easily mirror L1's desired paging mode.

#8 lore raw

Re: [PATCH v2 1/3] KVM: x86: Allow CPU to force vendor-specific TDP level

Quoted 24 lines · Sean Christopherson
Quoted 22 lines · Yu Zhang
Quoted 20 lines · Wei Huang
Quoted 7 lines · Yu Zhang
Quoted 6 lines · Wei Huang
Quoted 3 lines · Yu Zhang
Quoted 1 line · Wei Huang

AMD future CPUs will require a 5-level NPT if host CR4.LA57 is set.

Quoted 7 lines
Quoted 5 lines
Quoted 1 line

Sorry, but why? NPT is not indexed by HVA.

NPT is not indexed by HVA - it is always indexed by GPA. What I meant is NPT page table level has to be the same as the host OS page table: if 5-level page table is enabled in host OS (CR4.LA57=1), guest NPT has to 5-level too.

I know what you meant. But may I ask why?

I don't have a good answer for it. From what I know, VMCB doesn't have a field to indicate guest page table level. As a result, hardware relies on host CR4 to infer NPT level.

Quoted 7 lines
Quoted 1 line

I guess you mean not even in the N_CR3 field of VMCB?

Correct, nCR3 is a basically a pure representation of a regular CR3.

Quoted 1 line

Then it's not a broken design - it's a limitation of SVM. :)

That's just a polite way of saying it's a broken design ;-)

Doesn't this break legacy type 2 hypervisors that don't know anything about 5-level NPT and don't have any control over whether or not the host uses 5-level paging?

Quoted 6 lines

Joking aside, NPT opted for a semblance of backwards compatibility at the cost of having to carry all the baggage that comes with a legacy design. Keeping the core functionality from IA32 paging presumably miminizes design and hardware costs, and required minimal enabling in hypervisors. The downside is that it's less flexible than EPT and has a few warts, e.g. shadowing NPT is gross because the host can't easily mirror L1's desired paging mode.

maintainer
#9 lore raw

Re: [PATCH v2 1/3] KVM: x86: Allow CPU to force vendor-specific TDP level

Quoted 3 lines · Jim Mattson

Doesn't this break legacy type 2 hypervisors that don't know anything about 5-level NPT and don't have any control over whether or not the host uses 5-level paging?

Yes, where "legacy" probably means "all released versions of all of them", including KVM. Host support for LA57 was merged in 4.13, while KVM started supporting 5-level page tables in EPT in 4.14 and even then just returned PT64_ROOT_LEVEL (i.e. 4) for the maximum NPT level.

So all Linux versions up to 5.13, which has "KVM: x86: Prevent KVM SVM from loading on kernels with 5-level paging", will break horribly. Better backport that patch to stable...

Paolo

#10 lore raw

Re: [PATCH v2 1/3] KVM: x86: Allow CPU to force vendor-specific TDP level

In reply to Sean Christopherson
Quoted 29 lines · Sean Christopherson
Quoted 16 lines · Yu Zhang
Quoted 15 lines · Wei Huang
Quoted 13 lines · Yu Zhang
Quoted 11 lines · Wei Huang
Quoted 4 lines · Yu Zhang
Quoted 1 line · Wei Huang

AMD future CPUs will require a 5-level NPT if host CR4.LA57 is set.

Sorry, but why? NPT is not indexed by HVA.

NPT is not indexed by HVA - it is always indexed by GPA. What I meant is NPT page table level has to be the same as the host OS page table: if 5-level page table is enabled in host OS (CR4.LA57=1), guest NPT has to 5-level too.

Quoted 7 lines
Quoted 5 lines
Quoted 1 line

I know what you meant. But may I ask why?

I don't have a good answer for it. From what I know, VMCB doesn't have a field to indicate guest page table level. As a result, hardware relies on host CR4 to infer NPT level.

I guess you mean not even in the N_CR3 field of VMCB?

Correct, nCR3 is a basically a pure representation of a regular CR3.

Quoted 1 line

Then it's not a broken design - it's a limitation of SVM. :)

Quoted 8 lines

That's just a polite way of saying it's a broken design ;-)

Joking aside, NPT opted for a semblance of backwards compatibility at the cost of having to carry all the baggage that comes with a legacy design. Keeping the core functionality from IA32 paging presumably miminizes design and hardware costs, and required minimal enabling in hypervisors. The downside is that it's less flexible than EPT and has a few warts, e.g. shadowing NPT is gross because the host can't easily mirror L1's desired paging mode.

Thanks for your explaination, Sean. Everything has a cost, and now it's time to pay the price. :)

As to the max level, it is calculated in kvm_init(). Though I do not see any chance that host paging mode will be changed after kvm_init(), or any case that Linux uses different paging levels in different pCPUs, I am wondering, should we do something, e.g., to document this as an open?

About "host can't easily mirror L1's desired paging mode", could you please elaborate? Thanks!

B.R.
Yu

maintainer
#11 lore raw

Re: [PATCH v2 1/3] KVM: x86: Allow CPU to force vendor-specific TDP level

Quoted 2 lines · Yu Zhang

About "host can't easily mirror L1's desired paging mode", could you please elaborate? Thanks!

Shadow pgae tables in KVM will always have 3 levels on 32-bit machines and 4/5 levels on 64-bit machines. L1 instead might have any number of levels from 2 to 5 (though of course not more than the host has).

Therefore, when shadowing 32-bit NPT page tables, KVM has to add extra fixed levels on top of those that it's shadowing. See mmu_alloc_direct_roots for the code.

Paolo

#12 lore raw

Re: [PATCH v2 1/3] KVM: x86: Allow CPU to force vendor-specific TDP level

Quoted 7 lines · Paolo Bonzini
Quoted 2 lines · Yu Zhang

About "host can't easily mirror L1's desired paging mode", could you please elaborate? Thanks!

Shadow pgae tables in KVM will always have 3 levels on 32-bit machines and 4/5 levels on 64-bit machines. L1 instead might have any number of levels from 2 to 5 (though of course not more than the host has).

Thanks Paolo.

I guess it's because, unlike EPT which are with either 4 or 5 levels, NPT's level can range from 2 to 5, depending on the host paging mode...

Quoted 4 lines

Therefore, when shadowing 32-bit NPT page tables, KVM has to add extra fixed levels on top of those that it's shadowing. See mmu_alloc_direct_roots for the code.

So when shadowing NPTs(can be 2/3 levels, depending on the paging mode in L1), and if L0 Linux is running in 4/5 level mode, extra levels of paging structures is needed in the shadow NPT.

But shadow EPT does not have such annoyance. Is above understanding correct?

B.R.
Yu

maintainer
#13 lore raw

Re: [PATCH v2 1/3] KVM: x86: Allow CPU to force vendor-specific TDP level

Quoted 2 lines · Yu Zhang

I guess it's because, unlike EPT which are with either 4 or 5 levels, NPT's level can range from 2 to 5, depending on the host paging mode...

Yes, on Linux that will be one of 3/4/5 based on host paging mode, and it will apply to all N_CR3...

Quoted 1 line

But shadow EPT does not have such annoyance. Is above understanding correct?

... Right, because shadow EPT cannot have less than 4 levels, and it can always use 4 levels if that's what L1 uses.

Paolo

#14 lore raw

Re: [PATCH v2 1/3] KVM: x86: Allow CPU to force vendor-specific TDP level

Quoted 11 lines · Paolo Bonzini
Quoted 2 lines · Yu Zhang

I guess it's because, unlike EPT which are with either 4 or 5 levels, NPT's level can range from 2 to 5, depending on the host paging mode...

Yes, on Linux that will be one of 3/4/5 based on host paging mode, and it will apply to all N_CR3...

Quoted 1 line

But shadow EPT does not have such annoyance. Is above understanding correct?

... Right, because shadow EPT cannot have less than 4 levels, and it can always use 4 levels if that's what L1 uses.

Interesting. :) Thanks a lot for the explanation!

B.R.
Yu

author
#15 lore raw

PATCH v2 2/3 KVM: x86: Handle the case of 5-level shadow page table

In reply to Wei Huang

When the 5-level page table CPU flag is exposed, KVM code needs to handle this case by pointing mmu->root_hpa to a properly-constructed 5-level page table.

---
 arch/x86/include/asm/kvm_host.h |  1 +
 arch/x86/kvm/mmu/mmu.c          | 48 +++++++++++++++++++++------------
 2 files changed, 32 insertions(+), 17 deletions(-)
arch/x86/include/asm/kvm_host.h +1 -0
index 6d16f75cc8da..5d66a94ca428 100644
@@ -447,6 +447,7 @@ struct kvm_mmu {
 
 	u64 *pae_root;
 	u64 *pml4_root;
+	u64 *pml5_root;
 
 	/*
 	 * check zero bits on shadow page table entries, these
arch/x86/kvm/mmu/mmu.c +31 -17

97 lines not on this page · raw message

Signature · 1 line
2.31.1
#16 lore raw

Re: [PATCH v2 2/3] KVM: x86: Handle the case of 5-level shadow page table

Quoted 10 lines · Wei Huang
arch/x86/kvm/mmu/mmu.c +5 -1
@@ -3457,10 +3457,19 @@ static int mmu_alloc_shadow_roots(struct kvm_vcpu *vcpu)
 		mmu->pae_root[i] = root | pm_mask;
 	}
 
-	if (mmu->shadow_root_level == PT64_ROOT_4LEVEL)
+	/*
+	 * Depending on the shadow_root_level, build the root_hpa table by
+	 * chaining either pml5->pml4->pae or pml4->pae.
+	 */
+	mmu->root_hpa = __pa(mmu->pae_root);

arch/x86/kvm/mmu/mmu.c static int mmu_alloc_shadow_roots(struct kvm_vcpu *vcpu) · line 3472

Quoted 10 lines
arch/x86/kvm/mmu/mmu.c +7 -2
+	if (mmu->shadow_root_level >= PT64_ROOT_4LEVEL) {
+		mmu->pml4_root[0] = mmu->root_hpa | pm_mask;
 		mmu->root_hpa = __pa(mmu->pml4_root);
-	else
-		mmu->root_hpa = __pa(mmu->pae_root);
+	}
+	if (mmu->shadow_root_level == PT64_ROOT_5LEVEL) {
+		mmu->pml5_root[0] = mmu->root_hpa | pm_mask;
+		mmu->root_hpa = __pa(mmu->pml5_root);
+	}

I still really dislike this approach, it requires visually connecting multiple statements to understand the chain. I don't see any advantage (the 6-level paging comment was 99.9% a joke) of rewriting root_hpa other than that's how it's done today.

In the future, please give reviewers ample opportunity to respond before sending a new version if there's disagreement, otherwise the conversation gets carried over into a different thread and loses the original context.

Quoted 3 lines
arch/x86/kvm/mmu/mmu.c +0 -0
 set_root_pgd:
 	mmu->root_pgd = root_pgd;
author
#17 lore raw

Re: [PATCH v2 2/3] KVM: x86: Handle the case of 5-level shadow page table

Quoted 17 lines · Sean Christopherson
Quoted 16 lines · Wei Huang
arch/x86/kvm/mmu/mmu.c +8 -3
@@ -3457,10 +3457,19 @@ static int mmu_alloc_shadow_roots(struct kvm_vcpu *vcpu)
 		mmu->pae_root[i] = root | pm_mask;
 	}
 
-	if (mmu->shadow_root_level == PT64_ROOT_4LEVEL)
+	/*
+	 * Depending on the shadow_root_level, build the root_hpa table by
+	 * chaining either pml5->pml4->pae or pml4->pae.
+	 */
+	mmu->root_hpa = __pa(mmu->pae_root);
+	if (mmu->shadow_root_level >= PT64_ROOT_4LEVEL) {
+		mmu->pml4_root[0] = mmu->root_hpa | pm_mask;
 		mmu->root_hpa = __pa(mmu->pml4_root);
-	else
-		mmu->root_hpa = __pa(mmu->pae_root);
+	}
Quoted 9 lines

arch/x86/kvm/mmu/mmu.c static int mmu_alloc_shadow_roots(struct kvm_vcpu *vcpu) · line 3472

Quoted 4 lines
arch/x86/kvm/mmu/mmu.c +4 -0
+	if (mmu->shadow_root_level == PT64_ROOT_5LEVEL) {
+		mmu->pml5_root[0] = mmu->root_hpa | pm_mask;
+		mmu->root_hpa = __pa(mmu->pml5_root);
+	}

I still really dislike this approach, it requires visually connecting multiple statements to understand the chain. I don't see any advantage (the 6-level paging comment was 99.9% a joke) of rewriting root_hpa other than that's how it's done today.

I can change this part in v3, unless different comments from other reviewers.

Quoted 7 lines

In the future, please give reviewers ample opportunity to respond before sending a new version if there's disagreement, otherwise the conversation gets carried over into a different thread and loses the original context.

Quoted 3 lines
arch/x86/kvm/mmu/mmu.c +0 -0
 set_root_pgd:
 	mmu->root_pgd = root_pgd;
author
#18 lore raw

PATCH v2 3/3 KVM: SVM: Add 5-level page table support for SVM

In reply to Wei Huang

When the 5-level page table is enabled on host OS, the nested page table for guest VMs must use 5-level as well. Update get_npt_level() function to reflect this requirement. In the meanwhile, remove the code that prevents kvm-amd driver from being loaded when 5-level page table is detected.

---
 arch/x86/kvm/svm/svm.c | 9 +++------
 1 file changed, 3 insertions(+), 6 deletions(-)
arch/x86/kvm/svm/svm.c +3 -6
index f361d466e18e..dfb864f2674b 100644
@@ -261,7 +261,9 @@ u32 svm_msrpm_offset(u32 msr)
 static int get_max_npt_level(void)
 {
 #ifdef CONFIG_X86_64
-	return PT64_ROOT_4LEVEL;
+	bool la57 = (cr4_read_shadow() & X86_CR4_LA57) != 0;
+
+	return la57 ? PT64_ROOT_5LEVEL : PT64_ROOT_4LEVEL;
 #else
 	return PT32E_ROOT_LEVEL;
 #endif
@@ -462,11 +464,6 @@ static int has_svm(void)
 		return 0;
 	}
 
-	if (pgtable_l5_enabled()) {
-		pr_info("KVM doesn't yet support 5-level paging on AMD SVM\n");
-		return 0;
-	}
-
 	return 1;
 }
 
Signature · 1 line
2.31.1