Skip to contentSkip to conversation

VM: SVM: Honor KVM_MAX_VCPUS when AVIC is enabled

PATCH In review Sean Christopherson 3 participants 22 messages Aug 15, 2023 – Oct 5, 2023

Subsystems (5)
x86 KVM KVM x86 IOMMU +1 more subsystems
lore
Share & tools
lore

Sean Christopherson

author maintainer
#0 lore raw

cover VM: SVM: Honor KVM_MAX_VCPUS when AVIC is enabled

The only true functional change in this entire mess is to change KVM's handling of KVM_CREATE_VCPU when AVIC is enabled. Currently, KVM rejects vCPU creation if the vcpu_id is unaddressable, i.e. if it's larger than what is suppported by AVIC/x2AVIC hardware. That is a rather blatant violation of both KVM_CAP_MAX_VCPUS and KVM_CAP_MAX_VCPU_ID, as KVM will advertise a KVM_CAP_MAX_VCPUS as 1024 and KVM_CAP_MAX_VCPU_ID as 4096, but then reject vcpu_ids as low as 256 (AVIC).

To fix the problem, add yet another AVIC inhibit to disable AVIC if userspace creates unaddressable vCPUs. Alternatively, KVM could report different KVM_CAP_MAX_VCPUS and KVM_CAP_MAX_VCPU_ID values when AVIC is enabled, but IMO that path sets KVM up for failure, e.g. it would make it really hard for us to enable AVIC/x2AVIC by default, and we'd have to have to rework KVM selftests, which assume that KVM supports at least 512 vCPUs, e.g. recalc_apic_map_test fails when AVIC is enabled.

The bulk of this series is cleaning up related code, most of which is purely opportunistic, e.g. the many pointless PA masks, but some of which are functionally "necessary", for some definitions of necessary.

Lightly tested, and the IOMMU interaction is basically compile tested only. But this is firmly post-6.6 material, so no rush on anyone testing this (I wouldn't even care all that much if the darn selftests didn't fail).

Sean Christopherson (10):
KVM: SVM: Drop pointless masking of default APIC base when setting
V_APIC_BAR
KVM: SVM: Use AVIC_HPA_MASK when initializing vCPU's Physical ID entry
KVM: SVM: Drop pointless masking of kernel page pa's with "AVIC's" HPA
mask
KVM: SVM: Add helper to deduplicate code for getting AVIC backing page
KVM: SVM: Drop vcpu_svm's pointless avic_backing_page field iommu/amd: KVM: SVM: Use pi_desc_addr to derive ga_root_ptr
KVM: SVM: Inhibit AVIC if ID is too big instead of rejecting vCPU
creation
KVM: SVM: WARN if KVM attempts to create AVIC backing page with user
APIC
KVM: SVM: Drop redundant check in AVIC code on ID during vCPU creation
KVM: SVM: Rename "avic_physical_id_cache" to "avic_physical_id_entry"

 arch/x86/include/asm/kvm_host.h |  6 +++
 arch/x86/include/asm/svm.h      |  6 +--
 arch/x86/kvm/svm/avic.c         | 79 +++++++++++++++------------------
 arch/x86/kvm/svm/svm.h          |  6 +--
 drivers/iommu/amd/iommu.c       |  2 +-
 include/linux/amd-iommu.h       |  1 -
 6 files changed, 48 insertions(+), 52 deletions(-)

base-commit: 240f736891887939571854bd6d734b6c9291f22e

Signature · 1 line
2.41.0.694.ge786442a9b-goog
author maintainer
#1 lore raw

PATCH 1/10 KVM: SVM: Drop pointless masking of default APIC base when setting V_APIC_BAR

Drop VMCB_AVIC_APIC_BAR_MASK, it's just a regurgitation of the maximum theoretical 4KiB-aligned physical address, i.e. is not novel in any way, and its only usage is to mask the default APIC base, which is 4KiB aligned and (obviously) a legal physical address.

No functional change intended.

---
 arch/x86/include/asm/svm.h | 2 --
 arch/x86/kvm/svm/avic.c    | 2 +-
 2 files changed, 1 insertion(+), 3 deletions(-)
arch/x86/include/asm/svm.h +0 -2
index 72ebd5e4e975..1e70600e84f7 100644
@@ -257,8 +257,6 @@ struct __attribute__ ((__packed__)) vmcb_control_area {
 
 #define AVIC_DOORBELL_PHYSICAL_ID_MASK			GENMASK_ULL(11, 0)
 
-#define VMCB_AVIC_APIC_BAR_MASK				0xFFFFFFFFFF000ULL
-
 #define AVIC_UNACCEL_ACCESS_WRITE_MASK		1
 #define AVIC_UNACCEL_ACCESS_OFFSET_MASK		0xFF0
 #define AVIC_UNACCEL_ACCESS_VECTOR_MASK		0xFFFFFFFF
arch/x86/kvm/svm/avic.c +1 -1
index cfc8ab773025..7062164e4041 100644
@@ -251,7 +251,7 @@ void avic_init_vmcb(struct vcpu_svm *svm, struct vmcb *vmcb)
 	vmcb->control.avic_backing_page = bpa & AVIC_HPA_MASK;
 	vmcb->control.avic_logical_id = lpa & AVIC_HPA_MASK;
 	vmcb->control.avic_physical_id = ppa & AVIC_HPA_MASK;
-	vmcb->control.avic_vapic_bar = APIC_DEFAULT_PHYS_BASE & VMCB_AVIC_APIC_BAR_MASK;
+	vmcb->control.avic_vapic_bar = APIC_DEFAULT_PHYS_BASE;
 
 	if (kvm_apicv_activated(svm->vcpu.kvm))
 		avic_activate_vmcb(svm);
Signature · 1 line
2.41.0.694.ge786442a9b-goog
#2 lore raw

Re: [PATCH 01/10] KVM: SVM: Drop pointless masking of default APIC base when setting V_APIC_BAR

У вт, 2023-08-15 у 14:35 -0700, Sean Christopherson пише:

Quoted 18 lines

Drop VMCB_AVIC_APIC_BAR_MASK, it's just a regurgitation of the maximum theoretical 4KiB-aligned physical address, i.e. is not novel in any way, and its only usage is to mask the default APIC base, which is 4KiB aligned and (obviously) a legal physical address.

No functional change intended.

---
 arch/x86/include/asm/svm.h | 2 --
 arch/x86/kvm/svm/avic.c    | 2 +-
 2 files changed, 1 insertion(+), 3 deletions(-)
arch/x86/include/asm/svm.h +0 -0
index 72ebd5e4e975..1e70600e84f7 100644

arch/x86/include/asm/svm.h struct __attribute__ ((__packed__)) vmcb_control_area { · line 260

Quoted 4 lines
arch/x86/include/asm/svm.h +0 -1
@@ -257,8 +257,6 @@ struct __attribute__ ((__packed__)) vmcb_control_area {
 
 #define AVIC_DOORBELL_PHYSICAL_ID_MASK			GENMASK_ULL(11, 0)
 
-#define VMCB_AVIC_APIC_BAR_MASK				0xFFFFFFFFFF000ULL

While this mask is indeed not needed now because AVIC doesn't support non default APIC base, this mask will be needed in my upcoming nested AVIC support, because nested hypervisor can ask for any apic base it wishes for.

Quoted 9 lines
arch/x86/include/asm/svm.h +0 -1
-
 #define AVIC_UNACCEL_ACCESS_WRITE_MASK		1
 #define AVIC_UNACCEL_ACCESS_OFFSET_MASK		0xFF0
 #define AVIC_UNACCEL_ACCESS_VECTOR_MASK		0xFFFFFFFF
arch/x86/kvm/svm/avic.c +0 -0
index cfc8ab773025..7062164e4041 100644

arch/x86/kvm/svm/avic.c void avic_init_vmcb(struct vcpu_svm *svm, struct vmcb *vmcb) · line 254

Quoted 5 lines
arch/x86/kvm/svm/avic.c +1 -1
@@ -251,7 +251,7 @@ void avic_init_vmcb(struct vcpu_svm *svm, struct vmcb *vmcb)
 	vmcb->control.avic_backing_page = bpa & AVIC_HPA_MASK;
 	vmcb->control.avic_logical_id = lpa & AVIC_HPA_MASK;
 	vmcb->control.avic_physical_id = ppa & AVIC_HPA_MASK;
-	vmcb->control.avic_vapic_bar = APIC_DEFAULT_PHYS_BASE & VMCB_AVIC_APIC_BAR_MASK;
+	vmcb->control.avic_vapic_bar = APIC_DEFAULT_PHYS_BASE;

Here I agree that the '&' is functionally pointless, although I am not sure that removing it makes the code more readable.

Best regards,
Maxim Levitsky

Quoted 3 lines
arch/x86/kvm/svm/avic.c +0 -0
 	if (kvm_apicv_activated(svm->vcpu.kvm))
 		avic_activate_vmcb(svm);
author maintainer
#3 lore raw

PATCH 2/10 KVM: SVM: Use AVIC_HPA_MASK when initializing vCPU's Physical ID entry

In reply to Sean Christopherson

Use AVIC_HPA_MASK instead of AVIC_PHYSICAL_ID_ENTRY_BACKING_PAGE_MASK when initializing a vCPU's Physical ID table entry, the two masks are identical. Keep both #defines for now, along with a few new static asserts. A future change will clean up the entire mess (spoiler alert, the masks are pointless).

Opportunisitically move the bitwise-OR of AVIC_PHYSICAL_ID_ENTRY_VALID_MASK outside of the call to __sme_set(), again to pave the way for code deduplication. __sme_set() is purely additive, i.e. ORing in the valid bit before or after the C-bit does not change the end result.

No functional change intended.

---
 arch/x86/include/asm/svm.h | 2 ++
 arch/x86/kvm/svm/avic.c    | 5 ++---
 2 files changed, 4 insertions(+), 3 deletions(-)
arch/x86/include/asm/svm.h +2 -0
index 1e70600e84f7..609c9b596399 100644
@@ -285,6 +285,8 @@ static_assert((AVIC_MAX_PHYSICAL_ID & AVIC_PHYSICAL_MAX_INDEX_MASK) == AVIC_MAX_
 static_assert((X2AVIC_MAX_PHYSICAL_ID & AVIC_PHYSICAL_MAX_INDEX_MASK) == X2AVIC_MAX_PHYSICAL_ID);
 
 #define AVIC_HPA_MASK	~((0xFFFULL << 52) | 0xFFF)
+static_assert(AVIC_PHYSICAL_ID_ENTRY_BACKING_PAGE_MASK == AVIC_HPA_MASK);
+static_assert(AVIC_PHYSICAL_ID_ENTRY_BACKING_PAGE_MASK == GENMASK_ULL(51, 12));
 
 #define SVM_SEV_FEAT_DEBUG_SWAP                        BIT(5)
 
arch/x86/kvm/svm/avic.c +2 -3
index 7062164e4041..442c58ef8158 100644
@@ -308,9 +308,8 @@ static int avic_init_backing_page(struct kvm_vcpu *vcpu)
 	if (!entry)
 		return -EINVAL;
 
-	new_entry = __sme_set((page_to_phys(svm->avic_backing_page) &
-			      AVIC_PHYSICAL_ID_ENTRY_BACKING_PAGE_MASK) |
-			      AVIC_PHYSICAL_ID_ENTRY_VALID_MASK);
+	new_entry = __sme_set(page_to_phys(svm->avic_backing_page) & AVIC_HPA_MASK) |
+		    AVIC_PHYSICAL_ID_ENTRY_VALID_MASK;
 	WRITE_ONCE(*entry, new_entry);
 
 	svm->avic_physical_id_cache = entry;
Signature · 1 line
2.41.0.694.ge786442a9b-goog
#4 lore raw

Re: [PATCH 02/10] KVM: SVM: Use AVIC_HPA_MASK when initializing vCPU's Physical ID entry

У вт, 2023-08-15 у 14:35 -0700, Sean Christopherson пише:

Quoted 2 lines

Use AVIC_HPA_MASK instead of AVIC_PHYSICAL_ID_ENTRY_BACKING_PAGE_MASK when initializing a vCPU's Physical ID table entry, the two masks are identical.

Masks are identical, but they refer to different things.
AVIC_HPA_MASK is about avic's fields in vmcb, while AVIC_PHYSICAL_ID_ENTRY_BACKING_PAGE_MASK is about the portion of the physical id entry which has the address.

They are the same currently but for documentation purposes I don't understand why removal of one of them makes sense.

IMHO, it's best to define them to the same value, e.g:

#define AVIC_HPA_MASK GENMASK_ULL(51, 12)

#define AVIC_PHYSICAL_ID_ENTRY_BACKING_PAGE_MASK GENMASK_ULL(51, 12)

Best regards,
Maxim Levitsky

Quoted 46 lines

Keep both #defines for now, along with a few new static asserts. A future change will clean up the entire mess (spoiler alert, the masks are pointless).

Opportunisitically move the bitwise-OR of AVIC_PHYSICAL_ID_ENTRY_VALID_MASK outside of the call to __sme_set(), again to pave the way for code deduplication. __sme_set() is purely additive, i.e. ORing in the valid bit before or after the C-bit does not change the end result.

No functional change intended.

---
 arch/x86/include/asm/svm.h | 2 ++
 arch/x86/kvm/svm/avic.c    | 5 ++---
 2 files changed, 4 insertions(+), 3 deletions(-)
arch/x86/include/asm/svm.h +2 -0
index 1e70600e84f7..609c9b596399 100644
@@ -285,6 +285,8 @@ static_assert((AVIC_MAX_PHYSICAL_ID & AVIC_PHYSICAL_MAX_INDEX_MASK) == AVIC_MAX_
 static_assert((X2AVIC_MAX_PHYSICAL_ID & AVIC_PHYSICAL_MAX_INDEX_MASK) == X2AVIC_MAX_PHYSICAL_ID);
 
 #define AVIC_HPA_MASK	~((0xFFFULL << 52) | 0xFFF)
+static_assert(AVIC_PHYSICAL_ID_ENTRY_BACKING_PAGE_MASK == AVIC_HPA_MASK);
+static_assert(AVIC_PHYSICAL_ID_ENTRY_BACKING_PAGE_MASK == GENMASK_ULL(51, 12));
 
 #define SVM_SEV_FEAT_DEBUG_SWAP                        BIT(5)
 
arch/x86/kvm/svm/avic.c +2 -3
index 7062164e4041..442c58ef8158 100644
@@ -308,9 +308,8 @@ static int avic_init_backing_page(struct kvm_vcpu *vcpu)
 	if (!entry)
 		return -EINVAL;
 
-	new_entry = __sme_set((page_to_phys(svm->avic_backing_page) &
-			      AVIC_PHYSICAL_ID_ENTRY_BACKING_PAGE_MASK) |
-			      AVIC_PHYSICAL_ID_ENTRY_VALID_MASK);
+	new_entry = __sme_set(page_to_phys(svm->avic_backing_page) & AVIC_HPA_MASK) |
+		    AVIC_PHYSICAL_ID_ENTRY_VALID_MASK;
 	WRITE_ONCE(*entry, new_entry);
 
 	svm->avic_physical_id_cache = entry;
author maintainer
#5 lore raw

PATCH 3/10 KVM: SVM: Drop pointless masking of kernel page pa's with "AVIC's" HPA mask

In reply to Sean Christopherson

Drop AVIC_HPA_MASK and all its users, the mask is just the 4KiB-aligned maximum theoretical physical address for x86-64 CPUs. All usage in KVM masks the result of page_to_phys(), which on x86-64 is guaranteed to be 4KiB aligned and a legal physical address; if either of those requirements doesn't hold true, KVM has far bigger problems.

The unnecessarily masking in avic_init_vmcb() also incorrectly assumes that SME's C-bit resides between bits 51:11; that holds true for current CPUs, but isn't required by AMD's architecture:

In some implementations, the bit used may be a physical address bit

Key word being "may".

Opportunistically use the GENMASK_ULL() version for AVIC_PHYSICAL_ID_ENTRY_BACKING_PAGE_MASK, which is far more readable than a set of repeating Fs. Keep the macro even though it's unused, and will likely never be used, as it helps visualize the layout of an entry.

---
 arch/x86/include/asm/svm.h |  6 +-----
 arch/x86/kvm/svm/avic.c    | 11 +++++------
 2 files changed, 6 insertions(+), 11 deletions(-)
arch/x86/include/asm/svm.h +1 -5
index 609c9b596399..df644ca3febe 100644
@@ -250,7 +250,7 @@ struct __attribute__ ((__packed__)) vmcb_control_area {
 #define AVIC_LOGICAL_ID_ENTRY_VALID_MASK		(1 << 31)
 
 #define AVIC_PHYSICAL_ID_ENTRY_HOST_PHYSICAL_ID_MASK	GENMASK_ULL(11, 0)
-#define AVIC_PHYSICAL_ID_ENTRY_BACKING_PAGE_MASK	(0xFFFFFFFFFFULL << 12)
+#define AVIC_PHYSICAL_ID_ENTRY_BACKING_PAGE_MASK	GENMASK_ULL(51, 12)
 #define AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK		(1ULL << 62)
 #define AVIC_PHYSICAL_ID_ENTRY_VALID_MASK		(1ULL << 63)
 #define AVIC_PHYSICAL_ID_TABLE_SIZE_MASK		(0xFFULL)
@@ -284,10 +284,6 @@ enum avic_ipi_failure_cause {
 static_assert((AVIC_MAX_PHYSICAL_ID & AVIC_PHYSICAL_MAX_INDEX_MASK) == AVIC_MAX_PHYSICAL_ID);
 static_assert((X2AVIC_MAX_PHYSICAL_ID & AVIC_PHYSICAL_MAX_INDEX_MASK) == X2AVIC_MAX_PHYSICAL_ID);
 
-#define AVIC_HPA_MASK	~((0xFFFULL << 52) | 0xFFF)
-static_assert(AVIC_PHYSICAL_ID_ENTRY_BACKING_PAGE_MASK == AVIC_HPA_MASK);
-static_assert(AVIC_PHYSICAL_ID_ENTRY_BACKING_PAGE_MASK == GENMASK_ULL(51, 12));
-
 #define SVM_SEV_FEAT_DEBUG_SWAP                        BIT(5)
 
 struct vmcb_seg {
arch/x86/kvm/svm/avic.c +5 -6
index 442c58ef8158..b8313f2d88fa 100644
@@ -248,9 +248,9 @@ void avic_init_vmcb(struct vcpu_svm *svm, struct vmcb *vmcb)
 	phys_addr_t lpa = __sme_set(page_to_phys(kvm_svm->avic_logical_id_table_page));
 	phys_addr_t ppa = __sme_set(page_to_phys(kvm_svm->avic_physical_id_table_page));
 
-	vmcb->control.avic_backing_page = bpa & AVIC_HPA_MASK;
-	vmcb->control.avic_logical_id = lpa & AVIC_HPA_MASK;
-	vmcb->control.avic_physical_id = ppa & AVIC_HPA_MASK;
+	vmcb->control.avic_backing_page = bpa;
+	vmcb->control.avic_logical_id = lpa;
+	vmcb->control.avic_physical_id = ppa;
 	vmcb->control.avic_vapic_bar = APIC_DEFAULT_PHYS_BASE;
 
 	if (kvm_apicv_activated(svm->vcpu.kvm))
@@ -308,7 +308,7 @@ static int avic_init_backing_page(struct kvm_vcpu *vcpu)
 	if (!entry)
 		return -EINVAL;
 
-	new_entry = __sme_set(page_to_phys(svm->avic_backing_page) & AVIC_HPA_MASK) |
+	new_entry = __sme_set(page_to_phys(svm->avic_backing_page)) |
 		    AVIC_PHYSICAL_ID_ENTRY_VALID_MASK;
 	WRITE_ONCE(*entry, new_entry);
 
@@ -917,8 +917,7 @@ int avic_pi_update_irte(struct kvm *kvm, unsigned int host_irq,
 			struct amd_iommu_pi_data pi;
 
 			/* Try to enable guest_mode in IRTE */
-			pi.base = __sme_set(page_to_phys(svm->avic_backing_page) &
-					    AVIC_HPA_MASK);
+			pi.base = __sme_set(page_to_phys(svm->avic_backing_page));
 			pi.ga_tag = AVIC_GATAG(to_kvm_svm(kvm)->avic_vm_id,
 						     svm->vcpu.vcpu_id);
 			pi.is_guest_mode = true;
Signature · 1 line
2.41.0.694.ge786442a9b-goog
#6 lore raw

Re: [PATCH 03/10] KVM: SVM: Drop pointless masking of kernel page pa's with "AVIC's" HPA mask

У вт, 2023-08-15 у 14:35 -0700, Sean Christopherson пише:

Quoted 2 lines

Drop AVIC_HPA_MASK and all its users, the mask is just the 4KiB-aligned maximum theoretical physical address for x86-64 CPUs.

Typo: you mean 'current max theoretical physical address' because it might increase the future again (Intel already did increase it, couple of times).

Quoted 4 lines

All usage in KVM
masks the result of page_to_phys(), which on x86-64 is guaranteed to be 4KiB aligned and a legal physical address; if either of those requirements doesn't hold true, KVM has far bigger problems.

I do think that a build time assert that max physical address is as expected by AVIC, is needed.

Consider this: intel/amd releases yet another extension of the max physical address, the kernel gets updated, but for CPUs which lack this extension the max physical address will still be 52 bit and should still be respected.

My CPU for example has 48 bits in max physical address, while the kernel thinks that max valid physical address is 52 bit.

I understand that this is a theoretical problem but at least a build time assert that kernel max physical address matches avic's should be done.

Especially if we consider the fact that this patch also fixes a purely theoretical problem as well.

I do agree that masking output of page_to_phys is pointless and even dangerous, but asserting that we are not trying to use address which is
outside of AVIC's capabilities is a good thing to have, instead of relying blindly on the kernel to do the right thing.

Quoted 84 lines

The unnecessarily masking in avic_init_vmcb() also incorrectly assumes that SME's C-bit resides between bits 51:11; that holds true for current CPUs, but isn't required by AMD's architecture:

In some implementations, the bit used may be a physical address bit

Key word being "may".

Opportunistically use the GENMASK_ULL() version for AVIC_PHYSICAL_ID_ENTRY_BACKING_PAGE_MASK, which is far more readable than a set of repeating Fs. Keep the macro even though it's unused, and will likely never be used, as it helps visualize the layout of an entry.

---
 arch/x86/include/asm/svm.h |  6 +-----
 arch/x86/kvm/svm/avic.c    | 11 +++++------
 2 files changed, 6 insertions(+), 11 deletions(-)
arch/x86/include/asm/svm.h +1 -5
index 609c9b596399..df644ca3febe 100644
@@ -250,7 +250,7 @@ struct __attribute__ ((__packed__)) vmcb_control_area {
 #define AVIC_LOGICAL_ID_ENTRY_VALID_MASK		(1 << 31)
 
 #define AVIC_PHYSICAL_ID_ENTRY_HOST_PHYSICAL_ID_MASK	GENMASK_ULL(11, 0)
-#define AVIC_PHYSICAL_ID_ENTRY_BACKING_PAGE_MASK	(0xFFFFFFFFFFULL << 12)
+#define AVIC_PHYSICAL_ID_ENTRY_BACKING_PAGE_MASK	GENMASK_ULL(51, 12)
 #define AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK		(1ULL << 62)
 #define AVIC_PHYSICAL_ID_ENTRY_VALID_MASK		(1ULL << 63)
 #define AVIC_PHYSICAL_ID_TABLE_SIZE_MASK		(0xFFULL)
@@ -284,10 +284,6 @@ enum avic_ipi_failure_cause {
 static_assert((AVIC_MAX_PHYSICAL_ID & AVIC_PHYSICAL_MAX_INDEX_MASK) == AVIC_MAX_PHYSICAL_ID);
 static_assert((X2AVIC_MAX_PHYSICAL_ID & AVIC_PHYSICAL_MAX_INDEX_MASK) == X2AVIC_MAX_PHYSICAL_ID);
 
-#define AVIC_HPA_MASK	~((0xFFFULL << 52) | 0xFFF)
-static_assert(AVIC_PHYSICAL_ID_ENTRY_BACKING_PAGE_MASK == AVIC_HPA_MASK);
-static_assert(AVIC_PHYSICAL_ID_ENTRY_BACKING_PAGE_MASK == GENMASK_ULL(51, 12));
-
 #define SVM_SEV_FEAT_DEBUG_SWAP                        BIT(5)
 
 struct vmcb_seg {
arch/x86/kvm/svm/avic.c +5 -6
index 442c58ef8158..b8313f2d88fa 100644
@@ -248,9 +248,9 @@ void avic_init_vmcb(struct vcpu_svm *svm, struct vmcb *vmcb)
 	phys_addr_t lpa = __sme_set(page_to_phys(kvm_svm->avic_logical_id_table_page));
 	phys_addr_t ppa = __sme_set(page_to_phys(kvm_svm->avic_physical_id_table_page));
 
-	vmcb->control.avic_backing_page = bpa & AVIC_HPA_MASK;
-	vmcb->control.avic_logical_id = lpa & AVIC_HPA_MASK;
-	vmcb->control.avic_physical_id = ppa & AVIC_HPA_MASK;
+	vmcb->control.avic_backing_page = bpa;
+	vmcb->control.avic_logical_id = lpa;
+	vmcb->control.avic_physical_id = ppa;
 	vmcb->control.avic_vapic_bar = APIC_DEFAULT_PHYS_BASE;
 
 	if (kvm_apicv_activated(svm->vcpu.kvm))
@@ -308,7 +308,7 @@ static int avic_init_backing_page(struct kvm_vcpu *vcpu)
 	if (!entry)
 		return -EINVAL;
 
-	new_entry = __sme_set(page_to_phys(svm->avic_backing_page) & AVIC_HPA_MASK) |
+	new_entry = __sme_set(page_to_phys(svm->avic_backing_page)) |
 		    AVIC_PHYSICAL_ID_ENTRY_VALID_MASK;
 	WRITE_ONCE(*entry, new_entry);
 
@@ -917,8 +917,7 @@ int avic_pi_update_irte(struct kvm *kvm, unsigned int host_irq,
 			struct amd_iommu_pi_data pi;
 
 			/* Try to enable guest_mode in IRTE */
-			pi.base = __sme_set(page_to_phys(svm->avic_backing_page) &
-					    AVIC_HPA_MASK);
+			pi.base = __sme_set(page_to_phys(svm->avic_backing_page));
 			pi.ga_tag = AVIC_GATAG(to_kvm_svm(kvm)->avic_vm_id,
 						     svm->vcpu.vcpu_id);
 			pi.is_guest_mode = true;
author maintainer
#7 lore raw

PATCH 4/10 KVM: SVM: Add helper to deduplicate code for getting AVIC backing page

In reply to Sean Christopherson

Add a helper to get the physical address of the AVIC backing page, both to deduplicate code and to prepare for getting the address directly from apic->regs, at which point it won't be all that obvious that the address in question is what SVM calls the AVIC backing page.

No functional change intended.

---
 arch/x86/kvm/svm/avic.c | 14 +++++++++-----
 1 file changed, 9 insertions(+), 5 deletions(-)
arch/x86/kvm/svm/avic.c +9 -5
index b8313f2d88fa..954bdb45033b 100644
@@ -241,14 +241,18 @@ int avic_vm_init(struct kvm *kvm)
 	return err;
 }
 
+static phys_addr_t avic_get_backing_page_address(struct vcpu_svm *svm)
+{
+	return __sme_set(page_to_phys(svm->avic_backing_page));
+}
+
 void avic_init_vmcb(struct vcpu_svm *svm, struct vmcb *vmcb)
 {
 	struct kvm_svm *kvm_svm = to_kvm_svm(svm->vcpu.kvm);
-	phys_addr_t bpa = __sme_set(page_to_phys(svm->avic_backing_page));
 	phys_addr_t lpa = __sme_set(page_to_phys(kvm_svm->avic_logical_id_table_page));
 	phys_addr_t ppa = __sme_set(page_to_phys(kvm_svm->avic_physical_id_table_page));
 
-	vmcb->control.avic_backing_page = bpa;
+	vmcb->control.avic_backing_page = avic_get_backing_page_address(svm);
 	vmcb->control.avic_logical_id = lpa;
 	vmcb->control.avic_physical_id = ppa;
 	vmcb->control.avic_vapic_bar = APIC_DEFAULT_PHYS_BASE;
@@ -308,7 +312,7 @@ static int avic_init_backing_page(struct kvm_vcpu *vcpu)
 	if (!entry)
 		return -EINVAL;
 
-	new_entry = __sme_set(page_to_phys(svm->avic_backing_page)) |
+	new_entry = avic_get_backing_page_address(svm) |
 		    AVIC_PHYSICAL_ID_ENTRY_VALID_MASK;
 	WRITE_ONCE(*entry, new_entry);
 
@@ -859,7 +863,7 @@ get_pi_vcpu_info(struct kvm *kvm, struct kvm_kernel_irq_routing_entry *e,
 	pr_debug("SVM: %s: use GA mode for irq %u\n", __func__,
 		 irq.vector);
 	*svm = to_svm(vcpu);
-	vcpu_info->pi_desc_addr = __sme_set(page_to_phys((*svm)->avic_backing_page));
+	vcpu_info->pi_desc_addr = avic_get_backing_page_address(*svm);
 	vcpu_info->vector = irq.vector;
 
 	return 0;
@@ -917,7 +921,7 @@ int avic_pi_update_irte(struct kvm *kvm, unsigned int host_irq,
 			struct amd_iommu_pi_data pi;
 
 			/* Try to enable guest_mode in IRTE */
-			pi.base = __sme_set(page_to_phys(svm->avic_backing_page));
+			pi.base = avic_get_backing_page_address(svm);
 			pi.ga_tag = AVIC_GATAG(to_kvm_svm(kvm)->avic_vm_id,
 						     svm->vcpu.vcpu_id);
 			pi.is_guest_mode = true;
Signature · 1 line
2.41.0.694.ge786442a9b-goog
Reviewed-by on v1 4/10
#8 lore raw

Re: [PATCH 04/10] KVM: SVM: Add helper to deduplicate code for getting AVIC backing page

У вт, 2023-08-15 у 14:35 -0700, Sean Christopherson пише:

Quoted 56 lines

Add a helper to get the physical address of the AVIC backing page, both to deduplicate code and to prepare for getting the address directly from apic->regs, at which point it won't be all that obvious that the address in question is what SVM calls the AVIC backing page.

No functional change intended.

---
 arch/x86/kvm/svm/avic.c | 14 +++++++++-----
 1 file changed, 9 insertions(+), 5 deletions(-)
arch/x86/kvm/svm/avic.c +8 -4
index b8313f2d88fa..954bdb45033b 100644
@@ -241,14 +241,18 @@ int avic_vm_init(struct kvm *kvm)
 	return err;
 }
 
+static phys_addr_t avic_get_backing_page_address(struct vcpu_svm *svm)
+{
+	return __sme_set(page_to_phys(svm->avic_backing_page));
+}
+
 void avic_init_vmcb(struct vcpu_svm *svm, struct vmcb *vmcb)
 {
 	struct kvm_svm *kvm_svm = to_kvm_svm(svm->vcpu.kvm);
-	phys_addr_t bpa = __sme_set(page_to_phys(svm->avic_backing_page));
 	phys_addr_t lpa = __sme_set(page_to_phys(kvm_svm->avic_logical_id_table_page));
 	phys_addr_t ppa = __sme_set(page_to_phys(kvm_svm->avic_physical_id_table_page));
 
-	vmcb->control.avic_backing_page = bpa;
+	vmcb->control.avic_backing_page = avic_get_backing_page_address(svm);
 	vmcb->control.avic_logical_id = lpa;
 	vmcb->control.avic_physical_id = ppa;
 	vmcb->control.avic_vapic_bar = APIC_DEFAULT_PHYS_BASE;
@@ -308,7 +312,7 @@ static int avic_init_backing_page(struct kvm_vcpu *vcpu)
 	if (!entry)
 		return -EINVAL;
 
-	new_entry = __sme_set(page_to_phys(svm->avic_backing_page)) |
+	new_entry = avic_get_backing_page_address(svm) |
 		    AVIC_PHYSICAL_ID_ENTRY_VALID_MASK;
 	WRITE_ONCE(*entry, new_entry);
 
@@ -859,7 +863,7 @@ get_pi_vcpu_info(struct kvm *kvm, struct kvm_kernel_irq_routing_entry *e,
 	pr_debug("SVM: %s: use GA mode for irq %u\n", __func__,
 		 irq.vector);
 	*svm = to_svm(vcpu);
-	vcpu_info->pi_desc_addr = __sme_set(page_to_phys((*svm)->avic_backing_page));
+	vcpu_info->pi_desc_addr = avic_get_backing_page_address(*svm);
 	vcpu_info->vector = irq.vector;
 
 	return 0;

arch/x86/kvm/svm/avic.c int avic_pi_update_irte(struct kvm *kvm, unsigned int host_irq, · line 927

Quoted 8 lines
arch/x86/kvm/svm/avic.c +1 -1
@@ -917,7 +921,7 @@ int avic_pi_update_irte(struct kvm *kvm, unsigned int host_irq,
 			struct amd_iommu_pi_data pi;
 
 			/* Try to enable guest_mode in IRTE */
-			pi.base = __sme_set(page_to_phys(svm->avic_backing_page));
+			pi.base = avic_get_backing_page_address(svm);
 			pi.ga_tag = AVIC_GATAG(to_kvm_svm(kvm)->avic_vm_id,
 						     svm->vcpu.vcpu_id);
 			pi.is_guest_mode = true;

Best regards,
Maxim Levitsky

author maintainer
#9 lore raw

PATCH 5/10 KVM: SVM: Drop vcpu_svm's pointless avic_backing_page field

In reply to Sean Christopherson

Drop vcpu_svm's avic_backing_page pointer and instead grab the physical address of KVM's vAPIC page directly from the source. Getting a physical address from a kernel virtual address is not an expensive operation, and getting the physical address from a struct page is *more* expensive for CONFIG_SPARSEMEM=y kernels. Regardless, none of the paths that consume the address are hot paths, i.e. shaving cycles is not a priority.

Eliminating the "cache" means KVM doesn't have to worry about the cache being invalid, which will simplify a future fix when dealing with vCPU IDs that are too big.

---
 arch/x86/kvm/svm/avic.c | 4 +---
 arch/x86/kvm/svm/svm.h  | 1 -
 2 files changed, 1 insertion(+), 4 deletions(-)
arch/x86/kvm/svm/avic.c +1 -3
index 954bdb45033b..e49b682c8469 100644
@@ -243,7 +243,7 @@ int avic_vm_init(struct kvm *kvm)
 
 static phys_addr_t avic_get_backing_page_address(struct vcpu_svm *svm)
 {
-	return __sme_set(page_to_phys(svm->avic_backing_page));
+	return __sme_set(__pa(svm->vcpu.arch.apic->regs));
 }
 
 void avic_init_vmcb(struct vcpu_svm *svm, struct vmcb *vmcb)
@@ -305,8 +305,6 @@ static int avic_init_backing_page(struct kvm_vcpu *vcpu)
 			return ret;
 	}
 
-	svm->avic_backing_page = virt_to_page(vcpu->arch.apic->regs);
-
 	/* Setting AVIC backing page address in the phy APIC ID table */
 	entry = avic_get_physical_id_entry(vcpu, id);
 	if (!entry)
arch/x86/kvm/svm/svm.h +0 -1
index 2237230aad98..a9fde1bb85ee 100644
@@ -261,7 +261,6 @@ struct vcpu_svm {
 
 	u32 ldr_reg;
 	u32 dfr_reg;
-	struct page *avic_backing_page;
 	u64 *avic_physical_id_cache;
 
 	/*
Signature · 1 line
2.41.0.694.ge786442a9b-goog
#10 lore raw

Re: [PATCH 05/10] KVM: SVM: Drop vcpu_svm's pointless avic_backing_page field

У вт, 2023-08-15 у 14:35 -0700, Sean Christopherson пише:

Quoted 22 lines

Drop vcpu_svm's avic_backing_page pointer and instead grab the physical address of KVM's vAPIC page directly from the source. Getting a physical address from a kernel virtual address is not an expensive operation, and getting the physical address from a struct page is *more* expensive for CONFIG_SPARSEMEM=y kernels. Regardless, none of the paths that consume the address are hot paths, i.e. shaving cycles is not a priority.

Eliminating the "cache" means KVM doesn't have to worry about the cache being invalid, which will simplify a future fix when dealing with vCPU IDs that are too big.

---
 arch/x86/kvm/svm/avic.c | 4 +---
 arch/x86/kvm/svm/svm.h  | 1 -
 2 files changed, 1 insertion(+), 4 deletions(-)
arch/x86/kvm/svm/avic.c +0 -0
index 954bdb45033b..e49b682c8469 100644

arch/x86/kvm/svm/avic.c int avic_vm_init(struct kvm *kvm) · line 246

Quoted 5 lines
arch/x86/kvm/svm/avic.c +1 -1
@@ -243,7 +243,7 @@ int avic_vm_init(struct kvm *kvm)
 
 static phys_addr_t avic_get_backing_page_address(struct vcpu_svm *svm)
 {
-	return __sme_set(page_to_phys(svm->avic_backing_page));
+	return __sme_set(__pa(svm->vcpu.arch.apic->regs));

I overall agree with this patch however the old code was safer:

svm->avic_backing_page is set to physical address of the apic registers only in the avic_init_backing_page() and after checking the vcpu->arch.apic->regs != NULL and now in theory NULL vcpu->arch.apic->regs are not checked.

I know that you later add a patch which adds a similar warning, but I prefer that you fold it with this patch.

Quoted 17 lines
arch/x86/kvm/svm/avic.c +0 -2
 }
 
 void avic_init_vmcb(struct vcpu_svm *svm, struct vmcb *vmcb)
@@ -305,8 +305,6 @@ static int avic_init_backing_page(struct kvm_vcpu *vcpu)
 			return ret;
 	}
 
-	svm->avic_backing_page = virt_to_page(vcpu->arch.apic->regs);
-
 	/* Setting AVIC backing page address in the phy APIC ID table */
 	entry = avic_get_physical_id_entry(vcpu, id);
 	if (!entry)
arch/x86/kvm/svm/svm.h +0 -0
index 2237230aad98..a9fde1bb85ee 100644

arch/x86/kvm/svm/svm.h struct vcpu_svm { · line 266

Quoted 7 lines
arch/x86/kvm/svm/svm.h +0 -1
@@ -261,7 +261,6 @@ struct vcpu_svm {
 
 	u32 ldr_reg;
 	u32 dfr_reg;
-	struct page *avic_backing_page;
 	u64 *avic_physical_id_cache;
 
 	/*

Best regards,
Maxim Levitsky

author maintainer
#11 lore raw

PATCH 6/10 iommu/amd: KVM: SVM: Use pi_desc_addr to derive ga_root_ptr

In reply to Sean Christopherson

Use vcpu_data.pi_desc_addr instead of amd_iommu_pi_data.base to get the GA root pointer. KVM is the only source of amd_iommu_pi_data.base, and KVM's one and only path for writing amd_iommu_pi_data.base computes the exact same value for vcpu_data.pi_desc_addr and amd_iommu_pi_data.base, and fills amd_iommu_pi_data.base if and only if vcpu_data.pi_desc_addr is valid, i.e. amd_iommu_pi_data.base is fully redundant.

---
 arch/x86/kvm/svm/avic.c   | 1 -
 drivers/iommu/amd/iommu.c | 2 +-
 include/linux/amd-iommu.h | 1 -
 3 files changed, 1 insertion(+), 3 deletions(-)
arch/x86/kvm/svm/avic.c +0 -1
index e49b682c8469..bd81e3517838 100644
@@ -919,7 +919,6 @@ int avic_pi_update_irte(struct kvm *kvm, unsigned int host_irq,
 			struct amd_iommu_pi_data pi;
 
 			/* Try to enable guest_mode in IRTE */
-			pi.base = avic_get_backing_page_address(svm);
 			pi.ga_tag = AVIC_GATAG(to_kvm_svm(kvm)->avic_vm_id,
 						     svm->vcpu.vcpu_id);
 			pi.is_guest_mode = true;
drivers/iommu/amd/iommu.c +1 -1
index c3b58a8389b9..9814df73b9a7 100644
@@ -3622,7 +3622,7 @@ static int amd_ir_set_vcpu_affinity(struct irq_data *data, void *vcpu_info)
 
 	pi_data->prev_ga_tag = ir_data->cached_ga_tag;
 	if (pi_data->is_guest_mode) {
-		ir_data->ga_root_ptr = (pi_data->base >> 12);
+		ir_data->ga_root_ptr = (vcpu_pi_info->pi_desc_addr >> 12);
 		ir_data->ga_vector = vcpu_pi_info->vector;
 		ir_data->ga_tag = pi_data->ga_tag;
 		ret = amd_iommu_activate_guest_mode(ir_data);
include/linux/amd-iommu.h +0 -1
index 953e6f12fa1c..30d19ad0e8a9 100644
@@ -20,7 +20,6 @@ struct amd_iommu;
 struct amd_iommu_pi_data {
 	u32 ga_tag;
 	u32 prev_ga_tag;
-	u64 base;
 	bool is_guest_mode;
 	struct vcpu_data *vcpu_data;
 	void *ir_data;
Signature · 1 line
2.41.0.694.ge786442a9b-goog
Reviewed-by on v1 6/10
#12 lore raw

Re: [PATCH 06/10] iommu/amd: KVM: SVM: Use pi_desc_addr to derive ga_root_ptr

Quoted 8 lines · Sean Christopherson

Use vcpu_data.pi_desc_addr instead of amd_iommu_pi_data.base to get the GA root pointer. KVM is the only source of amd_iommu_pi_data.base, and KVM's one and only path for writing amd_iommu_pi_data.base computes the exact same value for vcpu_data.pi_desc_addr and amd_iommu_pi_data.base, and fills amd_iommu_pi_data.base if and only if vcpu_data.pi_desc_addr is valid, i.e. amd_iommu_pi_data.base is fully redundant.

Nice cleanup,

Quoted 43 lines
---
 arch/x86/kvm/svm/avic.c   | 1 -
 drivers/iommu/amd/iommu.c | 2 +-
 include/linux/amd-iommu.h | 1 -
 3 files changed, 1 insertion(+), 3 deletions(-)
arch/x86/kvm/svm/avic.c +0 -1
index e49b682c8469..bd81e3517838 100644
@@ -919,7 +919,6 @@ int avic_pi_update_irte(struct kvm *kvm, unsigned int host_irq,
 			struct amd_iommu_pi_data pi;
 
 			/* Try to enable guest_mode in IRTE */
-			pi.base = avic_get_backing_page_address(svm);
 			pi.ga_tag = AVIC_GATAG(to_kvm_svm(kvm)->avic_vm_id,
 						     svm->vcpu.vcpu_id);
 			pi.is_guest_mode = true;
drivers/iommu/amd/iommu.c +1 -1
index c3b58a8389b9..9814df73b9a7 100644
@@ -3622,7 +3622,7 @@ static int amd_ir_set_vcpu_affinity(struct irq_data *data, void *vcpu_info)
 
 	pi_data->prev_ga_tag = ir_data->cached_ga_tag;
 	if (pi_data->is_guest_mode) {
-		ir_data->ga_root_ptr = (pi_data->base >> 12);
+		ir_data->ga_root_ptr = (vcpu_pi_info->pi_desc_addr >> 12);
 		ir_data->ga_vector = vcpu_pi_info->vector;
 		ir_data->ga_tag = pi_data->ga_tag;
 		ret = amd_iommu_activate_guest_mode(ir_data);
include/linux/amd-iommu.h +0 -1
index 953e6f12fa1c..30d19ad0e8a9 100644
@@ -20,7 +20,6 @@ struct amd_iommu;
 struct amd_iommu_pi_data {
 	u32 ga_tag;
 	u32 prev_ga_tag;
-	u64 base;
 	bool is_guest_mode;
 	struct vcpu_data *vcpu_data;
 	void *ir_data;
#13 lore raw

Re: [PATCH 06/10] iommu/amd: KVM: SVM: Use pi_desc_addr to derive ga_root_ptr

In reply to Sean Christopherson

У вт, 2023-08-15 у 14:35 -0700, Sean Christopherson пише:

Quoted 44 lines

Use vcpu_data.pi_desc_addr instead of amd_iommu_pi_data.base to get the GA root pointer. KVM is the only source of amd_iommu_pi_data.base, and KVM's one and only path for writing amd_iommu_pi_data.base computes the exact same value for vcpu_data.pi_desc_addr and amd_iommu_pi_data.base, and fills amd_iommu_pi_data.base if and only if vcpu_data.pi_desc_addr is valid, i.e. amd_iommu_pi_data.base is fully redundant.

---
 arch/x86/kvm/svm/avic.c   | 1 -
 drivers/iommu/amd/iommu.c | 2 +-
 include/linux/amd-iommu.h | 1 -
 3 files changed, 1 insertion(+), 3 deletions(-)
arch/x86/kvm/svm/avic.c +0 -1
index e49b682c8469..bd81e3517838 100644
@@ -919,7 +919,6 @@ int avic_pi_update_irte(struct kvm *kvm, unsigned int host_irq,
 			struct amd_iommu_pi_data pi;
 
 			/* Try to enable guest_mode in IRTE */
-			pi.base = avic_get_backing_page_address(svm);
 			pi.ga_tag = AVIC_GATAG(to_kvm_svm(kvm)->avic_vm_id,
 						     svm->vcpu.vcpu_id);
 			pi.is_guest_mode = true;
drivers/iommu/amd/iommu.c +1 -1
index c3b58a8389b9..9814df73b9a7 100644
@@ -3622,7 +3622,7 @@ static int amd_ir_set_vcpu_affinity(struct irq_data *data, void *vcpu_info)
 
 	pi_data->prev_ga_tag = ir_data->cached_ga_tag;
 	if (pi_data->is_guest_mode) {
-		ir_data->ga_root_ptr = (pi_data->base >> 12);
+		ir_data->ga_root_ptr = (vcpu_pi_info->pi_desc_addr >> 12);
 		ir_data->ga_vector = vcpu_pi_info->vector;
 		ir_data->ga_tag = pi_data->ga_tag;
 		ret = amd_iommu_activate_guest_mode(ir_data);
include/linux/amd-iommu.h +0 -0
index 953e6f12fa1c..30d19ad0e8a9 100644

include/linux/amd-iommu.h struct amd_iommu; · line 23

Quoted 4 lines
include/linux/amd-iommu.h +0 -1
@@ -20,7 +20,6 @@ struct amd_iommu;
 struct amd_iommu_pi_data {
 	u32 ga_tag;
 	u32 prev_ga_tag;
-	u64 base;

I think that a comment describing where the base address of the apic register page is taken now is needed, because before your patch the 'amd_iommu_pi_data' was self documenting, since it had all the info about the PI remap entry, but now it doesn't.

include/linux/amd-iommu.h struct amd_iommu; · line 25

Quoted 3 lines
include/linux/amd-iommu.h +0 -0
 	bool is_guest_mode;
 	struct vcpu_data *vcpu_data;
 	void *ir_data;

Best regards,
Maxim Levitsky

author maintainer
#14 lore raw

PATCH 7/10 KVM: SVM: Inhibit AVIC if ID is too big instead of rejecting vCPU creation

In reply to Sean Christopherson

Inhibit AVIC with a new "ID too big" flag if userspace creates a vCPU with an ID that is too big, but otherwise allow vCPU creation to succeed. Rejecting KVM_CREATE_VCPU with EINVAL violates KVM's ABI as KVM advertises that the max vCPU ID is 4095, but disallows creating vCPUs with IDs bigger than 254 (AVIC) or 511 (x2AVIC).

Alternatively, KVM could advertise an accurate value depending on which AVIC mode is in use, but that wouldn't really solve the underlying problem, e.g. would be a breaking change if KVM were to ever try and enable AVIC or x2AVIC by default.

---
 arch/x86/include/asm/kvm_host.h |  6 ++++++
 arch/x86/kvm/svm/avic.c         | 16 ++++++++++++++--
 arch/x86/kvm/svm/svm.h          |  3 ++-
 3 files changed, 22 insertions(+), 3 deletions(-)
arch/x86/include/asm/kvm_host.h +6 -0
index 60d430b4650f..4c2d659a1269 100644
@@ -1243,6 +1243,12 @@ enum kvm_apicv_inhibit {
 	 * mapping between logical ID and vCPU.
 	 */
 	APICV_INHIBIT_REASON_LOGICAL_ID_ALIASED,
+
+	/*
+	 * AVIC is disabled because the vCPU's APIC ID is beyond the max
+	 * supported by AVIC/x2AVIC, i.e. the vCPU is unaddressable.
+	 */
+	APICV_INHIBIT_REASON_ID_TOO_BIG,
 };
 
 struct kvm_arch {
arch/x86/kvm/svm/avic.c +14 -2
index bd81e3517838..522feaa711b4 100644
@@ -284,9 +284,21 @@ static int avic_init_backing_page(struct kvm_vcpu *vcpu)
 	int id = vcpu->vcpu_id;
 	struct vcpu_svm *svm = to_svm(vcpu);
 
+	/*
+	 * Inhibit AVIC if the vCPU ID is bigger than what is supported by AVIC
+	 * hardware.  Do so immediately, i.e. don't defer the update via a
+	 * request, as avic_vcpu_load() expects to be called if and only if the
+	 * vCPU has fully initialized AVIC.  Bypass all of the helpers and just
+	 * clear apicv_active directly, the vCPU isn't reachable and the VMCB
+	 * isn't even initialized at this point, i.e. there is no possibility
+	 * of needing to deal with the n
+	 */
 	if ((!x2avic_enabled && id > AVIC_MAX_PHYSICAL_ID) ||
-	    (id > X2AVIC_MAX_PHYSICAL_ID))
-		return -EINVAL;
+	    (id > X2AVIC_MAX_PHYSICAL_ID)) {
+		kvm_set_apicv_inhibit(vcpu->kvm, APICV_INHIBIT_REASON_ID_TOO_BIG);
+		vcpu->arch.apic->apicv_active = false;
+		return 0;
+	}
 
 	if (!vcpu->arch.apic->regs)
 		return -EINVAL;
arch/x86/kvm/svm/svm.h +2 -1
index a9fde1bb85ee..8b798982e5d0 100644
@@ -632,7 +632,8 @@ extern struct kvm_x86_nested_ops svm_nested_ops;
 	BIT(APICV_INHIBIT_REASON_PHYSICAL_ID_ALIASED) |	\
 	BIT(APICV_INHIBIT_REASON_APIC_ID_MODIFIED) |	\
 	BIT(APICV_INHIBIT_REASON_APIC_BASE_MODIFIED) |	\
-	BIT(APICV_INHIBIT_REASON_LOGICAL_ID_ALIASED)	\
+	BIT(APICV_INHIBIT_REASON_LOGICAL_ID_ALIASED) |	\
+	BIT(APICV_INHIBIT_REASON_ID_TOO_BIG)		\
 )
 
 bool avic_hardware_setup(void);
Signature · 1 line
2.41.0.694.ge786442a9b-goog
#15 lore raw

Re: [PATCH 07/10] KVM: SVM: Inhibit AVIC if ID is too big instead of rejecting vCPU creation

У вт, 2023-08-15 у 14:35 -0700, Sean Christopherson пише:

Quoted 24 lines

Inhibit AVIC with a new "ID too big" flag if userspace creates a vCPU with an ID that is too big, but otherwise allow vCPU creation to succeed. Rejecting KVM_CREATE_VCPU with EINVAL violates KVM's ABI as KVM advertises that the max vCPU ID is 4095, but disallows creating vCPUs with IDs bigger than 254 (AVIC) or 511 (x2AVIC).

Alternatively, KVM could advertise an accurate value depending on which AVIC mode is in use, but that wouldn't really solve the underlying problem, e.g. would be a breaking change if KVM were to ever try and enable AVIC or x2AVIC by default.

---
 arch/x86/include/asm/kvm_host.h |  6 ++++++
 arch/x86/kvm/svm/avic.c         | 16 ++++++++++++++--
 arch/x86/kvm/svm/svm.h          |  3 ++-
 3 files changed, 22 insertions(+), 3 deletions(-)
arch/x86/include/asm/kvm_host.h +0 -0
index 60d430b4650f..4c2d659a1269 100644

arch/x86/include/asm/kvm_host.h enum kvm_apicv_inhibit { · line 1251

Quoted 9 lines
arch/x86/include/asm/kvm_host.h +6 -0
@@ -1243,6 +1243,12 @@ enum kvm_apicv_inhibit {
 	 * mapping between logical ID and vCPU.
 	 */
 	APICV_INHIBIT_REASON_LOGICAL_ID_ALIASED,
+
+	/*
+	 * AVIC is disabled because the vCPU's APIC ID is beyond the max
+	 * supported by AVIC/x2AVIC, i.e. the vCPU is unaddressable.
+	 */
+	APICV_INHIBIT_REASON_ID_TOO_BIG,

I prefer APICV_INHIBIT_REASON_PHYSICAL_ID_TOO_BIG

Quoted 9 lines
arch/x86/include/asm/kvm_host.h +0 -0
 };
 
 struct kvm_arch {
arch/x86/kvm/svm/avic.c +0 -0
index bd81e3517838..522feaa711b4 100644
@@ -284,9 +284,21 @@ static int avic_init_backing_page(struct kvm_vcpu *vcpu)
 	int id = vcpu->vcpu_id;

arch/x86/kvm/svm/avic.c static int avic_init_backing_page(struct kvm_vcpu *vcpu) · line 294

Quoted 10 lines
arch/x86/kvm/svm/avic.c +8 -0
 	struct vcpu_svm *svm = to_svm(vcpu);
 
+	/*
+	 * Inhibit AVIC if the vCPU ID is bigger than what is supported by AVIC
+	 * hardware.  Do so immediately, i.e. don't defer the update via a
+	 * request, as avic_vcpu_load() expects to be called if and only if the
+	 * vCPU has fully initialized AVIC.  Bypass all of the helpers and just
+	 * clear apicv_active directly, the vCPU isn't reachable and the VMCB
+	 * isn't even initialized at this point, i.e. there is no possibility
+	 * of needing to deal with the n

Which helpers do you bypass? I see a call to normal kvm_set_apicv_inhibit() as it should be.

Note that userspace can add vCPUs at will so this can happen any time during the guest's lifetime so I don't think that this code can bypass anything.

Quoted 26 lines
arch/x86/kvm/svm/avic.c +6 -2
+	 */
 	if ((!x2avic_enabled && id > AVIC_MAX_PHYSICAL_ID) ||
-	    (id > X2AVIC_MAX_PHYSICAL_ID))
-		return -EINVAL;
+	    (id > X2AVIC_MAX_PHYSICAL_ID)) {
+		kvm_set_apicv_inhibit(vcpu->kvm, APICV_INHIBIT_REASON_ID_TOO_BIG);
+		vcpu->arch.apic->apicv_active = false;
+		return 0;
+	}
 
 	if (!vcpu->arch.apic->regs)
 		return -EINVAL;
arch/x86/kvm/svm/svm.h +2 -1
index a9fde1bb85ee..8b798982e5d0 100644
@@ -632,7 +632,8 @@ extern struct kvm_x86_nested_ops svm_nested_ops;
 	BIT(APICV_INHIBIT_REASON_PHYSICAL_ID_ALIASED) |	\
 	BIT(APICV_INHIBIT_REASON_APIC_ID_MODIFIED) |	\
 	BIT(APICV_INHIBIT_REASON_APIC_BASE_MODIFIED) |	\
-	BIT(APICV_INHIBIT_REASON_LOGICAL_ID_ALIASED)	\
+	BIT(APICV_INHIBIT_REASON_LOGICAL_ID_ALIASED) |	\
+	BIT(APICV_INHIBIT_REASON_ID_TOO_BIG)		\
 
 )
 
Quoted 1 line

bool avic_hardware_setup(void);

Best regards,
Maxim Levitsky

author maintainer
#16 lore raw

PATCH 8/10 KVM: SVM: WARN if KVM attempts to create AVIC backing page with user APIC

In reply to Sean Christopherson

WARN if KVM attempts to allocate a vCPU's AVIC backing page without an in-kernel local APIC. avic_init_vcpu() bails early if the APIC is not in-kernel, and KVM disallows enabling an in-kernel APIC after vCPUs have been created, i.e. it should be impossible to reach avic_init_backing_page() without the vAPIC being allocated.

---
 arch/x86/kvm/svm/avic.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
arch/x86/kvm/svm/avic.c +1 -1
index 522feaa711b4..3b2d00d9ca9b 100644
@@ -300,7 +300,7 @@ static int avic_init_backing_page(struct kvm_vcpu *vcpu)
 		return 0;
 	}
 
-	if (!vcpu->arch.apic->regs)
+	if (WARN_ON_ONCE(!vcpu->arch.apic->regs))
 		return -EINVAL;
 
 	if (kvm_apicv_activated(vcpu->kvm)) {
Signature · 1 line
2.41.0.694.ge786442a9b-goog
#17 lore raw

Re: [PATCH 08/10] KVM: SVM: WARN if KVM attempts to create AVIC backing page with user APIC

У вт, 2023-08-15 у 14:35 -0700, Sean Christopherson пише:

Quoted 16 lines

WARN if KVM attempts to allocate a vCPU's AVIC backing page without an in-kernel local APIC. avic_init_vcpu() bails early if the APIC is not in-kernel, and KVM disallows enabling an in-kernel APIC after vCPUs have been created, i.e. it should be impossible to reach avic_init_backing_page() without the vAPIC being allocated.

---
 arch/x86/kvm/svm/avic.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
arch/x86/kvm/svm/avic.c +0 -0
index 522feaa711b4..3b2d00d9ca9b 100644

arch/x86/kvm/svm/avic.c static int avic_init_backing_page(struct kvm_vcpu *vcpu) · line 306

Quoted 8 lines
arch/x86/kvm/svm/avic.c +1 -1
@@ -300,7 +300,7 @@ static int avic_init_backing_page(struct kvm_vcpu *vcpu)
 		return 0;
 	}
 
-	if (!vcpu->arch.apic->regs)
+	if (WARN_ON_ONCE(!vcpu->arch.apic->regs))
 		return -EINVAL;
 
 	if (kvm_apicv_activated(vcpu->kvm)) {

As I said I prefer this to be folded with patch that adds the avic_init_backing_page().

Best regards,
Maxim Levitsky

author maintainer
#18 lore raw

PATCH 9/10 KVM: SVM: Drop redundant check in AVIC code on ID during vCPU creation

In reply to Sean Christopherson

Drop avic_get_physical_id_entry()'s compatibility check on the incoming ID, as its sole caller, avic_init_backing_page(), performs the exact same check. Drop avic_get_physical_id_entry() entirely as the only remaining functionality is getting the address of the Physical ID table, and accessing the array without an immediate bounds check is kludgy.

No functional change intended.

---
 arch/x86/kvm/svm/avic.c | 28 ++++++----------------------
 1 file changed, 6 insertions(+), 22 deletions(-)
arch/x86/kvm/svm/avic.c +6 -22
index 3b2d00d9ca9b..6803e2d7bc22 100644
@@ -263,26 +263,12 @@ void avic_init_vmcb(struct vcpu_svm *svm, struct vmcb *vmcb)
 		avic_deactivate_vmcb(svm);
 }
 
-static u64 *avic_get_physical_id_entry(struct kvm_vcpu *vcpu,
-				       unsigned int index)
-{
-	u64 *avic_physical_id_table;
-	struct kvm_svm *kvm_svm = to_kvm_svm(vcpu->kvm);
-
-	if ((!x2avic_enabled && index > AVIC_MAX_PHYSICAL_ID) ||
-	    (index > X2AVIC_MAX_PHYSICAL_ID))
-		return NULL;
-
-	avic_physical_id_table = page_address(kvm_svm->avic_physical_id_table_page);
-
-	return &avic_physical_id_table[index];
-}
-
 static int avic_init_backing_page(struct kvm_vcpu *vcpu)
 {
-	u64 *entry, new_entry;
+	struct kvm_svm *kvm_svm = to_kvm_svm(vcpu->kvm);
+	struct vcpu_svm *svm = to_svm(vcpu);
+	u64 *table, new_entry;
 	int id = vcpu->vcpu_id;
-	struct vcpu_svm *svm = to_svm(vcpu);
 
 	/*
 	 * Inhibit AVIC if the vCPU ID is bigger than what is supported by AVIC
@@ -318,15 +304,13 @@ static int avic_init_backing_page(struct kvm_vcpu *vcpu)
 	}
 
 	/* Setting AVIC backing page address in the phy APIC ID table */
-	entry = avic_get_physical_id_entry(vcpu, id);
-	if (!entry)
-		return -EINVAL;
+	table = page_address(kvm_svm->avic_physical_id_table_page);
 
 	new_entry = avic_get_backing_page_address(svm) |
 		    AVIC_PHYSICAL_ID_ENTRY_VALID_MASK;
-	WRITE_ONCE(*entry, new_entry);
+	WRITE_ONCE(table[id], new_entry);
 
-	svm->avic_physical_id_cache = entry;
+	svm->avic_physical_id_cache = &table[id];
 
 	return 0;
 }
Signature · 1 line
2.41.0.694.ge786442a9b-goog
#19 lore raw

Re: [PATCH 09/10] KVM: SVM: Drop redundant check in AVIC code on ID during vCPU creation

У вт, 2023-08-15 у 14:35 -0700, Sean Christopherson пише:

Quoted 20 lines

Drop avic_get_physical_id_entry()'s compatibility check on the incoming ID, as its sole caller, avic_init_backing_page(), performs the exact same check. Drop avic_get_physical_id_entry() entirely as the only remaining functionality is getting the address of the Physical ID table, and accessing the array without an immediate bounds check is kludgy.

No functional change intended.

---
 arch/x86/kvm/svm/avic.c | 28 ++++++----------------------
 1 file changed, 6 insertions(+), 22 deletions(-)
arch/x86/kvm/svm/avic.c +0 -0
index 3b2d00d9ca9b..6803e2d7bc22 100644
@@ -263,26 +263,12 @@ void avic_init_vmcb(struct vcpu_svm *svm, struct vmcb *vmcb)
 		avic_deactivate_vmcb(svm);
 }

arch/x86/kvm/svm/avic.c void avic_init_vmcb(struct vcpu_svm *svm, struct vmcb *vmcb) · line 274

Quoted 10 lines
arch/x86/kvm/svm/avic.c +0 -9
 
-static u64 *avic_get_physical_id_entry(struct kvm_vcpu *vcpu,
-				       unsigned int index)
-{
-	u64 *avic_physical_id_table;
-	struct kvm_svm *kvm_svm = to_kvm_svm(vcpu->kvm);
-
-	if ((!x2avic_enabled && index > AVIC_MAX_PHYSICAL_ID) ||
-	    (index > X2AVIC_MAX_PHYSICAL_ID))
-		return NULL;

While removing this code doesn't introduce a bug, it does make it less safe, because new code just blindly trusts that vcpu_id will never be out of bounds of the physical id table.

Bugs happen and that can and will someday happen.

Quoted 19 lines
arch/x86/kvm/svm/avic.c +3 -8
-
-	avic_physical_id_table = page_address(kvm_svm->avic_physical_id_table_page);
-
-	return &avic_physical_id_table[index];
-}
-
 static int avic_init_backing_page(struct kvm_vcpu *vcpu)
 {
-	u64 *entry, new_entry;
+	struct kvm_svm *kvm_svm = to_kvm_svm(vcpu->kvm);
+	struct vcpu_svm *svm = to_svm(vcpu);
+	u64 *table, new_entry;
 	int id = vcpu->vcpu_id;
-	struct vcpu_svm *svm = to_svm(vcpu);
 
 	/*
 	 * Inhibit AVIC if the vCPU ID is bigger than what is supported by AVIC
@@ -318,15 +304,13 @@ static int avic_init_backing_page(struct kvm_vcpu *vcpu)
 	}

arch/x86/kvm/svm/avic.c static int avic_init_backing_page(struct kvm_vcpu *vcpu) · line 327

Quoted 10 lines
arch/x86/kvm/svm/avic.c +1 -4
 
 	/* Setting AVIC backing page address in the phy APIC ID table */
-	entry = avic_get_physical_id_entry(vcpu, id);
-	if (!entry)
-		return -EINVAL;
+	table = page_address(kvm_svm->avic_physical_id_table_page);
 
 	new_entry = avic_get_backing_page_address(svm) |
 		    AVIC_PHYSICAL_ID_ENTRY_VALID_MASK;
-	WRITE_ONCE(*entry, new_entry);

Here I prefer to at least have an assert that id is in bounds of a page (at least less than 512) so that a bug will not turn into a security issue by overflowing the buffer.

arch/x86/kvm/svm/avic.c static int avic_init_backing_page(struct kvm_vcpu *vcpu) · line 316

Quoted 7 lines
arch/x86/kvm/svm/avic.c +2 -1
+	WRITE_ONCE(table[id], new_entry);
 
-	svm->avic_physical_id_cache = entry;
+	svm->avic_physical_id_cache = &table[id];
 
 	return 0;
 }

Best regards,
Maxim Levitsky

author maintainer
#20 lore raw

PATCH 10/10 KVM: SVM: Rename "avic_physical_id_cache" to "avic_physical_id_entry"

In reply to Sean Christopherson

Rename the vCPU's pointer to its AVIC Physical ID entry from "cache" to "entry". While the field technically caches the result of the pointer calculation, it's all too easy to misinterpret the name and think that the field somehow caches the _data_ in the table.

No functional change intended.

---
 arch/x86/kvm/svm/avic.c | 10 +++++-----
 arch/x86/kvm/svm/svm.h  |  2 +-
 2 files changed, 6 insertions(+), 6 deletions(-)
arch/x86/kvm/svm/avic.c +5 -5
index 6803e2d7bc22..8d162ff83aa8 100644
@@ -310,7 +310,7 @@ static int avic_init_backing_page(struct kvm_vcpu *vcpu)
 		    AVIC_PHYSICAL_ID_ENTRY_VALID_MASK;
 	WRITE_ONCE(table[id], new_entry);
 
-	svm->avic_physical_id_cache = &table[id];
+	svm->avic_physical_id_entry = &table[id];
 
 	return 0;
 }
@@ -1028,14 +1028,14 @@ void avic_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 	if (kvm_vcpu_is_blocking(vcpu))
 		return;
 
-	entry = READ_ONCE(*(svm->avic_physical_id_cache));
+	entry = READ_ONCE(*(svm->avic_physical_id_entry));
 	WARN_ON_ONCE(entry & AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK);
 
 	entry &= ~AVIC_PHYSICAL_ID_ENTRY_HOST_PHYSICAL_ID_MASK;
 	entry |= (h_physical_id & AVIC_PHYSICAL_ID_ENTRY_HOST_PHYSICAL_ID_MASK);
 	entry |= AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK;
 
-	WRITE_ONCE(*(svm->avic_physical_id_cache), entry);
+	WRITE_ONCE(*(svm->avic_physical_id_entry), entry);
 	avic_update_iommu_vcpu_affinity(vcpu, h_physical_id, true);
 }
 
@@ -1046,7 +1046,7 @@ void avic_vcpu_put(struct kvm_vcpu *vcpu)
 
 	lockdep_assert_preemption_disabled();
 
-	entry = READ_ONCE(*(svm->avic_physical_id_cache));
+	entry = READ_ONCE(*(svm->avic_physical_id_entry));
 
 	/* Nothing to do if IsRunning == '0' due to vCPU blocking. */
 	if (!(entry & AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK))
@@ -1055,7 +1055,7 @@ void avic_vcpu_put(struct kvm_vcpu *vcpu)
 	avic_update_iommu_vcpu_affinity(vcpu, -1, 0);
 
 	entry &= ~AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK;
-	WRITE_ONCE(*(svm->avic_physical_id_cache), entry);
+	WRITE_ONCE(*(svm->avic_physical_id_entry), entry);
 }
 
 void avic_refresh_virtual_apic_mode(struct kvm_vcpu *vcpu)
arch/x86/kvm/svm/svm.h +1 -1
index 8b798982e5d0..4362048493d1 100644
@@ -261,7 +261,7 @@ struct vcpu_svm {
 
 	u32 ldr_reg;
 	u32 dfr_reg;
-	u64 *avic_physical_id_cache;
+	u64 *avic_physical_id_entry;
 
 	/*
 	 * Per-vcpu list of struct amd_svm_iommu_ir:
Signature · 1 line
2.41.0.694.ge786442a9b-goog
#21 lore raw

Re: [PATCH 10/10] KVM: SVM: Rename "avic_physical_id_cache" to "avic_physical_id_entry"

У вт, 2023-08-15 у 14:35 -0700, Sean Christopherson пише:

Quoted 4 lines

Rename the vCPU's pointer to its AVIC Physical ID entry from "cache" to "entry". While the field technically caches the result of the pointer calculation, it's all too easy to misinterpret the name and think that the field somehow caches the _data_ in the table.

I also strongly dislike the 'avic_physical_id_cache', but if you are refactoring it, IMHO the 'avic_physical_id_entry' is just as confusing since its a pointer to an entry and not the entry itself.

At least a comment to explain where this pointer points, or maybe (not sure) drop the avic_physical_id_cache completely and calculate it every time (I doubt that there is any perf loss due to this)

Best regards,
Maxim Levitsky

Quoted 70 lines

No functional change intended.

---
 arch/x86/kvm/svm/avic.c | 10 +++++-----
 arch/x86/kvm/svm/svm.h  |  2 +-
 2 files changed, 6 insertions(+), 6 deletions(-)
arch/x86/kvm/svm/avic.c +5 -5
index 6803e2d7bc22..8d162ff83aa8 100644
@@ -310,7 +310,7 @@ static int avic_init_backing_page(struct kvm_vcpu *vcpu)
 		    AVIC_PHYSICAL_ID_ENTRY_VALID_MASK;
 	WRITE_ONCE(table[id], new_entry);
 
-	svm->avic_physical_id_cache = &table[id];
+	svm->avic_physical_id_entry = &table[id];
 
 	return 0;
 }
@@ -1028,14 +1028,14 @@ void avic_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 	if (kvm_vcpu_is_blocking(vcpu))
 		return;
 
-	entry = READ_ONCE(*(svm->avic_physical_id_cache));
+	entry = READ_ONCE(*(svm->avic_physical_id_entry));
 	WARN_ON_ONCE(entry & AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK);
 
 	entry &= ~AVIC_PHYSICAL_ID_ENTRY_HOST_PHYSICAL_ID_MASK;
 	entry |= (h_physical_id & AVIC_PHYSICAL_ID_ENTRY_HOST_PHYSICAL_ID_MASK);
 	entry |= AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK;
 
-	WRITE_ONCE(*(svm->avic_physical_id_cache), entry);
+	WRITE_ONCE(*(svm->avic_physical_id_entry), entry);
 	avic_update_iommu_vcpu_affinity(vcpu, h_physical_id, true);
 }
 
@@ -1046,7 +1046,7 @@ void avic_vcpu_put(struct kvm_vcpu *vcpu)
 
 	lockdep_assert_preemption_disabled();
 
-	entry = READ_ONCE(*(svm->avic_physical_id_cache));
+	entry = READ_ONCE(*(svm->avic_physical_id_entry));
 
 	/* Nothing to do if IsRunning == '0' due to vCPU blocking. */
 	if (!(entry & AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK))
@@ -1055,7 +1055,7 @@ void avic_vcpu_put(struct kvm_vcpu *vcpu)
 	avic_update_iommu_vcpu_affinity(vcpu, -1, 0);
 
 	entry &= ~AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK;
-	WRITE_ONCE(*(svm->avic_physical_id_cache), entry);
+	WRITE_ONCE(*(svm->avic_physical_id_entry), entry);
 }
 
 void avic_refresh_virtual_apic_mode(struct kvm_vcpu *vcpu)
arch/x86/kvm/svm/svm.h +1 -1
index 8b798982e5d0..4362048493d1 100644
@@ -261,7 +261,7 @@ struct vcpu_svm {
 
 	u32 ldr_reg;
 	u32 dfr_reg;
-	u64 *avic_physical_id_cache;
+	u64 *avic_physical_id_entry;
 
 	/*
 	 * Per-vcpu list of struct amd_svm_iommu_ir: