Skip to contentSkip to conversation

KVM: nVMX: Disallow access to vmcs12 fields that aren't supported by "hardware"

PATCH Superseded Sean Christopherson 3 participants 5 messages Dec 16, 2025 – Dec 17, 2025

Subsystems (3)

v1 of 4 Superseded Latest is v4 →

lore
Share & tools
lore

Sean Christopherson

author maintainer
#0 lore raw

PATCH 1/1 KVM: nVMX: Disallow access to vmcs12 fields that aren't supported by "hardware"

Disallow access (VMREAD/VMWRITE) to fields that the loaded incarnation of KVM doesn't support, e.g. due to lack of hardware support, as a middle ground between allowing access to any vmcs12 field defined by KVM (current behavior) and gating access based on the userspace-defined vCPU model (the most correct, but costly, implementation).

Disallowing access to unsupported fields helps a tiny bit in terms of closing the virtualization hole (see below), but the main motivation is to avoid having to weed out unsupported fields when synchronizing between vmcs12 and a shadow VMCS. Because shadow VMCS accesses are done via VMREAD and VMWRITE, KVM _must_ filter out unsupported fields (or eat VMREAD/VMWRITE failures), and filtering out just shadow VMCS fields is about the same amount of effort, and arguably much more confusing.

As a bonus, this also fixes a KVM-Unit-Test failure bug when running on _hardware_ without support for TSC Scaling, which fails with the same signature as the bug fixed by commit ba1f82456ba8 ("KVM: nVMX: Dynamically compute max VMCS index for vmcs12"):

FAIL: VMX_VMCS_ENUM.MAX_INDEX expected: 19, actual: 17

Dynamically computing the max VMCS index only resolved the issue where KVM was hardcoding max index, but for CPUs with TSC Scaling, that was "good enough".

---
 arch/x86/kvm/vmx/nested.c |  6 -----
 arch/x86/kvm/vmx/vmcs.h   |  8 ++++++
 arch/x86/kvm/vmx/vmcs12.c | 55 +++++++++++++++++++++++++++++++++++++--
 arch/x86/kvm/vmx/vmcs12.h |  8 ++++--
 arch/x86/kvm/vmx/vmx.c    |  2 ++
 5 files changed, 69 insertions(+), 10 deletions(-)
arch/x86/kvm/vmx/nested.c +0 -6
index 6137e5307d0f..9d8f84e3f2da 100644
@@ -7074,12 +7074,6 @@ void nested_vmx_set_vmcs_shadowing_bitmap(void)
 	}
 }
 
-/*
- * Indexing into the vmcs12 uses the VMCS encoding rotated left by 6.  Undo
- * that madness to get the encoding for comparison.
- */
-#define VMCS12_IDX_TO_ENC(idx) ((u16)(((u16)(idx) >> 6) | ((u16)(idx) << 10)))
-
 static u64 nested_vmx_calc_vmcs_enum_msr(void)
 {
 	/*
arch/x86/kvm/vmx/vmcs.h +8 -0
index b25625314658..98281e019e38 100644
@@ -11,7 +11,15 @@
 
 #include "capabilities.h"
 
+/*
+ * Indexing into the vmcs12 uses the VMCS encoding rotated left by 6 as a very
+ * rudimentary compression of the range of indices.  The compression ratio is
+ * good enough to allow KVM to use a (very sparsely populated) array without
+ * wasting too much memory, while the "algorithm" is fast enough to be used to
+ * lookup vmcs12 fields on-demand, e.g. for emulation.
+ */
 #define ROL16(val, n) ((u16)(((u16)(val) << (n)) | ((u16)(val) >> (16 - (n)))))
+#define VMCS12_IDX_TO_ENC(idx) ((u16)(((u16)(idx) >> 6) | ((u16)(idx) << 10)))
 
 struct vmcs_hdr {
 	u32 revision_id:31;
arch/x86/kvm/vmx/vmcs12.c +53 -2

64 lines not on this page · raw message

arch/x86/kvm/vmx/vmcs12.h +6 -2
index 4ad6b16525b9..e5905ba0bb42 100644
@@ -374,8 +374,12 @@ static inline void vmx_check_vmcs12_offsets(void)
 	CHECK_OFFSET(guest_pml_index, 996);
 }
 
-extern const unsigned short vmcs12_field_offsets[];
-extern const unsigned int nr_vmcs12_fields;
+extern const __initconst u16 supported_vmcs12_field_offsets[];
+
+extern u16 vmcs12_field_offsets[] __ro_after_init;
+extern unsigned int nr_vmcs12_fields __ro_after_init;
+
+void __init nested_vmx_setup_vmcs12_fields(void);
 
 static inline short get_vmcs12_field_offset(unsigned long field)
 {
arch/x86/kvm/vmx/vmx.c +2 -0
index 6b96f7aea20b..e5ad3853f51d 100644
@@ -8670,6 +8670,8 @@ __init int vmx_hardware_setup(void)
 	 * can hide/show features based on kvm_cpu_cap_has().
 	 */
 	if (nested) {
+		nested_vmx_setup_vmcs12_fields();
+
 		nested_vmx_setup_ctls_msrs(&vmcs_config, vmx_capability.ept);
 
 		r = nested_vmx_hardware_setup(kvm_vmx_exit_handlers);

base-commit: 58e10b63777d0aebee2cf4e6c67e1a83e7edbe0f

Signature · 1 line
2.52.0.239.gd5f0c6e74e-goog
#1 lore raw

arch/x86/kvm/vmx/vmcs12.c line 12

Quoted 10 lines
arch/x86/kvm/vmx/vmcs12.c +1 -1
index 4233b5ca9461..78eca9399975 100644
@@ -9,7 +9,7 @@
 FIELD(number, name), \
 [ROL16(number##_HIGH, 6)] = VMCS12_OFFSET(name) + sizeof(u32)
 
-const unsigned short vmcs12_field_offsets[] = {
+const __initconst u16 supported_vmcs12_field_offsets[] = {

I initially misunderstood "supported" to mean the VMCS fields available at runtime. I'm unsure if it's necessary to make its meaning more explicit. E.g., prefix with kvm_?

Quoted 20 lines
arch/x86/kvm/vmx/vmcs12.c +12 -1
 FIELD(VIRTUAL_PROCESSOR_ID, virtual_processor_id),
 FIELD(POSTED_INTR_NV, posted_intr_nv),
 FIELD(GUEST_ES_SELECTOR, guest_es_selector),
@@ -158,4 +158,55 @@ const unsigned short vmcs12_field_offsets[] = {
 FIELD(HOST_SSP, host_ssp),
 FIELD(HOST_INTR_SSP_TABLE, host_ssp_tbl),
 };
-const unsigned int nr_vmcs12_fields = ARRAY_SIZE(vmcs12_field_offsets);
+
+u16 vmcs12_field_offsets[ARRAY_SIZE(supported_vmcs12_field_offsets)] __ro_after_init;
+unsigned int nr_vmcs12_fields __ro_after_init;
+
+#define VMCS12_CASE64(enc) case enc##_HIGH: case enc
+
+static __init bool cpu_has_vmcs12_field(unsigned int idx)
+{
+ switch (VMCS12_IDX_TO_ENC(idx)) {
+ case VIRTUAL_PROCESSOR_ID: return cpu_has_vmx_vpid();
+ case POSTED_INTR_NV: return cpu_has_vmx_posted_intr();
+ VMCS12_CASE64(TSC_MULTIPLIER): return cpu_has_vmx_tsc_scaling();

arch/x86/kvm/vmx/vmcs12.c const unsigned short vmcs12_field_offsets[] = { · line 182

Quoted 10 lines
arch/x86/kvm/vmx/vmcs12.c +10 -0
+ VMCS12_CASE64(VIRTUAL_APIC_PAGE_ADDR): return cpu_has_vmx_tpr_shadow();
+ VMCS12_CASE64(APIC_ACCESS_ADDR): return cpu_has_vmx_virtualize_apic_accesses();
+ VMCS12_CASE64(POSTED_INTR_DESC_ADDR): return cpu_has_vmx_posted_intr();
+ VMCS12_CASE64(VM_FUNCTION_CONTROL): return cpu_has_vmx_vmfunc();
+ VMCS12_CASE64(EPT_POINTER): return cpu_has_vmx_ept();
+ VMCS12_CASE64(EPTP_LIST_ADDRESS): return cpu_has_vmx_vmfunc();
+ VMCS12_CASE64(XSS_EXIT_BITMAP): return cpu_has_vmx_xsaves();
+ VMCS12_CASE64(ENCLS_EXITING_BITMAP): return cpu_has_vmx_encls_vmexit();
+ VMCS12_CASE64(GUEST_IA32_PERF_GLOBAL_CTRL): return cpu_has_load_perf_global_ctrl();
+ VMCS12_CASE64(HOST_IA32_PERF_GLOBAL_CTRL): return cpu_has_load_perf_global_ctrl();

Combine the above 2 cases?

arch/x86/kvm/vmx/vmcs12.c const unsigned short vmcs12_field_offsets[] = { · line 190

Quoted 8 lines
arch/x86/kvm/vmx/vmcs12.c +8 -0
+ case TPR_THRESHOLD: return cpu_has_vmx_tpr_shadow();
+ case SECONDARY_VM_EXEC_CONTROL: return cpu_has_secondary_exec_ctrls();
+ case GUEST_S_CET: return cpu_has_load_cet_ctrl();
+ case GUEST_SSP: return cpu_has_load_cet_ctrl();
+ case GUEST_INTR_SSP_TABLE: return cpu_has_load_cet_ctrl();
+ case HOST_S_CET: return cpu_has_load_cet_ctrl();
+ case HOST_SSP: return cpu_has_load_cet_ctrl();
+ case HOST_INTR_SSP_TABLE: return cpu_has_load_cet_ctrl();

Combine all CET cases?

author maintainer
#2 lore raw
Quoted 8 lines · Xin Li
Quoted 8 lines
arch/x86/kvm/vmx/vmcs12.c +0 -0
index 4233b5ca9461..78eca9399975 100644
@@ -9,7 +9,7 @@
 FIELD(number, name), \
 [ROL16(number##_HIGH, 6)] = VMCS12_OFFSET(name) + sizeof(u32)
 
Quoted 6 lines

arch/x86/kvm/vmx/vmcs12.c line 12

Quoted 2 lines
arch/x86/kvm/vmx/vmcs12.c +1 -1
-const unsigned short vmcs12_field_offsets[] = {
+const __initconst u16 supported_vmcs12_field_offsets[] = {

I initially misunderstood "supported" to mean the VMCS fields available at runtime. I'm unsure if it's necessary to make its meaning more explicit. E.g., prefix with kvm_?

Oh, good point. Ya, will do.

Quoted 42 lines
Quoted 20 lines
arch/x86/kvm/vmx/vmcs12.c +12 -1
 FIELD(VIRTUAL_PROCESSOR_ID, virtual_processor_id),
 FIELD(POSTED_INTR_NV, posted_intr_nv),
 FIELD(GUEST_ES_SELECTOR, guest_es_selector),
@@ -158,4 +158,55 @@ const unsigned short vmcs12_field_offsets[] = {
 FIELD(HOST_SSP, host_ssp),
 FIELD(HOST_INTR_SSP_TABLE, host_ssp_tbl),
 };
-const unsigned int nr_vmcs12_fields = ARRAY_SIZE(vmcs12_field_offsets);
+
+u16 vmcs12_field_offsets[ARRAY_SIZE(supported_vmcs12_field_offsets)] __ro_after_init;
+unsigned int nr_vmcs12_fields __ro_after_init;
+
+#define VMCS12_CASE64(enc) case enc##_HIGH: case enc
+
+static __init bool cpu_has_vmcs12_field(unsigned int idx)
+{
+ switch (VMCS12_IDX_TO_ENC(idx)) {
+ case VIRTUAL_PROCESSOR_ID: return cpu_has_vmx_vpid();
+ case POSTED_INTR_NV: return cpu_has_vmx_posted_intr();
+ VMCS12_CASE64(TSC_MULTIPLIER): return cpu_has_vmx_tsc_scaling();

arch/x86/kvm/vmx/vmcs12.c const unsigned short vmcs12_field_offsets[] = { · line 182

Quoted 10 lines
arch/x86/kvm/vmx/vmcs12.c +10 -0
+ VMCS12_CASE64(VIRTUAL_APIC_PAGE_ADDR): return cpu_has_vmx_tpr_shadow();
+ VMCS12_CASE64(APIC_ACCESS_ADDR): return cpu_has_vmx_virtualize_apic_accesses();
+ VMCS12_CASE64(POSTED_INTR_DESC_ADDR): return cpu_has_vmx_posted_intr();
+ VMCS12_CASE64(VM_FUNCTION_CONTROL): return cpu_has_vmx_vmfunc();
+ VMCS12_CASE64(EPT_POINTER): return cpu_has_vmx_ept();
+ VMCS12_CASE64(EPTP_LIST_ADDRESS): return cpu_has_vmx_vmfunc();
+ VMCS12_CASE64(XSS_EXIT_BITMAP): return cpu_has_vmx_xsaves();
+ VMCS12_CASE64(ENCLS_EXITING_BITMAP): return cpu_has_vmx_encls_vmexit();
+ VMCS12_CASE64(GUEST_IA32_PERF_GLOBAL_CTRL): return cpu_has_load_perf_global_ctrl();
+ VMCS12_CASE64(HOST_IA32_PERF_GLOBAL_CTRL): return cpu_has_load_perf_global_ctrl();

Combine the above 2 cases?

arch/x86/kvm/vmx/vmcs12.c const unsigned short vmcs12_field_offsets[] = { · line 190

Quoted 8 lines
arch/x86/kvm/vmx/vmcs12.c +8 -0
+ case TPR_THRESHOLD: return cpu_has_vmx_tpr_shadow();
+ case SECONDARY_VM_EXEC_CONTROL: return cpu_has_secondary_exec_ctrls();
+ case GUEST_S_CET: return cpu_has_load_cet_ctrl();
+ case GUEST_SSP: return cpu_has_load_cet_ctrl();
+ case GUEST_INTR_SSP_TABLE: return cpu_has_load_cet_ctrl();
+ case HOST_S_CET: return cpu_has_load_cet_ctrl();
+ case HOST_SSP: return cpu_has_load_cet_ctrl();
+ case HOST_INTR_SSP_TABLE: return cpu_has_load_cet_ctrl();
Quoted 1 line

Combine all CET cases?

Yeah, will do. I was on the fence as to whether it would be a net positive to combine them.

Thanks!

#3 lore raw
In reply to Sean Christopherson
Quoted 14 lines
arch/x86/kvm/vmx/vmcs12.c +14 -0
+static __init bool cpu_has_vmcs12_field(unsigned int idx)
+{
+	switch (VMCS12_IDX_TO_ENC(idx)) {
+	case VIRTUAL_PROCESSOR_ID: return cpu_has_vmx_vpid();
+	case POSTED_INTR_NV: return cpu_has_vmx_posted_intr();
+	VMCS12_CASE64(TSC_MULTIPLIER): return cpu_has_vmx_tsc_scaling();
+	VMCS12_CASE64(VIRTUAL_APIC_PAGE_ADDR): return cpu_has_vmx_tpr_shadow();
+	VMCS12_CASE64(APIC_ACCESS_ADDR): return cpu_has_vmx_virtualize_apic_accesses();
+	VMCS12_CASE64(POSTED_INTR_DESC_ADDR): return cpu_has_vmx_posted_intr();
+	VMCS12_CASE64(VM_FUNCTION_CONTROL): return cpu_has_vmx_vmfunc();
+	VMCS12_CASE64(EPT_POINTER): return cpu_has_vmx_ept();
+	VMCS12_CASE64(EPTP_LIST_ADDRESS): return cpu_has_vmx_vmfunc();
+	VMCS12_CASE64(XSS_EXIT_BITMAP): return cpu_has_vmx_xsaves();
+	VMCS12_CASE64(ENCLS_EXITING_BITMAP): return cpu_has_vmx_encls_vmexit();

arch/x86/kvm/vmx/vmcs12.c const unsigned short vmcs12_field_offsets[] = { · line 190

Quoted 10 lines
arch/x86/kvm/vmx/vmcs12.c +10 -0
+	VMCS12_CASE64(GUEST_IA32_PERF_GLOBAL_CTRL): return cpu_has_load_perf_global_ctrl();
+	VMCS12_CASE64(HOST_IA32_PERF_GLOBAL_CTRL): return cpu_has_load_perf_global_ctrl();
+	case TPR_THRESHOLD: return cpu_has_vmx_tpr_shadow();
+	case SECONDARY_VM_EXEC_CONTROL: return cpu_has_secondary_exec_ctrls();
+	case GUEST_S_CET: return cpu_has_load_cet_ctrl();
+	case GUEST_SSP: return cpu_has_load_cet_ctrl();
+	case GUEST_INTR_SSP_TABLE: return cpu_has_load_cet_ctrl();
+	case HOST_S_CET: return cpu_has_load_cet_ctrl();
+	case HOST_SSP: return cpu_has_load_cet_ctrl();
+	case HOST_INTR_SSP_TABLE: return cpu_has_load_cet_ctrl();

Most fields here are not shadowed, e.g., CET-related fields. So, the plan is that new fields should be added here regardless of whether they are shadowed or not, right?

And GUEST_INTR_STATUS is missing here. It depends on APICv and is handled explicitly in init_vmcs_shadow_fields().

Quoted 27 lines
arch/x86/kvm/vmx/vmcs12.c +22 -0
+
+	/* KVM always emulates PML and the VMX preemption timer in software. */
+	case GUEST_PML_INDEX:
+	case VMX_PREEMPTION_TIMER_VALUE:
+	default:
+		return true;
+	}
+}
+
+void __init nested_vmx_setup_vmcs12_fields(void)
+{
+	unsigned int i;
+
+	for (i = 0; i < ARRAY_SIZE(supported_vmcs12_field_offsets); i++) {
+		if (!supported_vmcs12_field_offsets[i] ||
+		    !cpu_has_vmcs12_field(i))
+			continue;
+
+		vmcs12_field_offsets[i] = supported_vmcs12_field_offsets[i];
+		nr_vmcs12_fields = i + 1;
+	}
+}
arch/x86/kvm/vmx/vmcs12.h +0 -0
index 4ad6b16525b9..e5905ba0bb42 100644

arch/x86/kvm/vmx/vmcs12.h static inline void vmx_check_vmcs12_offsets(void) · line 377

Quoted 6 lines
arch/x86/kvm/vmx/vmcs12.h +1 -2
@@ -374,8 +374,12 @@ static inline void vmx_check_vmcs12_offsets(void)
 	CHECK_OFFSET(guest_pml_index, 996);
 }
 
-extern const unsigned short vmcs12_field_offsets[];
-extern const unsigned int nr_vmcs12_fields;
+extern const __initconst u16 supported_vmcs12_field_offsets[];

No need to extern supported_vmcs12_field_offsets since it's only used in vmcs12.c.

Quoted 27 lines
arch/x86/kvm/vmx/vmcs12.h +5 -0
+
+extern u16 vmcs12_field_offsets[] __ro_after_init;
+extern unsigned int nr_vmcs12_fields __ro_after_init;
+
+void __init nested_vmx_setup_vmcs12_fields(void);
 
 static inline short get_vmcs12_field_offset(unsigned long field)
 {
arch/x86/kvm/vmx/vmx.c +2 -0
index 6b96f7aea20b..e5ad3853f51d 100644
@@ -8670,6 +8670,8 @@ __init int vmx_hardware_setup(void)
 	 * can hide/show features based on kvm_cpu_cap_has().
 	 */
 	if (nested) {
+		nested_vmx_setup_vmcs12_fields();
+
 		nested_vmx_setup_ctls_msrs(&vmcs_config, vmx_capability.ept);
 
 		r = nested_vmx_hardware_setup(kvm_vmx_exit_handlers);

base-commit: 58e10b63777d0aebee2cf4e6c67e1a83e7edbe0f

Signature · 1 line
2.52.0.239.gd5f0c6e74e-goog
author maintainer
#4 lore raw
Quoted 20 lines · Chao Gao
Quoted 20 lines
arch/x86/kvm/vmx/vmcs12.c +20 -0
+static __init bool cpu_has_vmcs12_field(unsigned int idx)
+{
+	switch (VMCS12_IDX_TO_ENC(idx)) {
+	case VIRTUAL_PROCESSOR_ID: return cpu_has_vmx_vpid();
+	case POSTED_INTR_NV: return cpu_has_vmx_posted_intr();
+	VMCS12_CASE64(TSC_MULTIPLIER): return cpu_has_vmx_tsc_scaling();
+	VMCS12_CASE64(VIRTUAL_APIC_PAGE_ADDR): return cpu_has_vmx_tpr_shadow();
+	VMCS12_CASE64(APIC_ACCESS_ADDR): return cpu_has_vmx_virtualize_apic_accesses();
+	VMCS12_CASE64(POSTED_INTR_DESC_ADDR): return cpu_has_vmx_posted_intr();
+	VMCS12_CASE64(VM_FUNCTION_CONTROL): return cpu_has_vmx_vmfunc();
+	VMCS12_CASE64(EPT_POINTER): return cpu_has_vmx_ept();
+	VMCS12_CASE64(EPTP_LIST_ADDRESS): return cpu_has_vmx_vmfunc();
+	VMCS12_CASE64(XSS_EXIT_BITMAP): return cpu_has_vmx_xsaves();
+	VMCS12_CASE64(ENCLS_EXITING_BITMAP): return cpu_has_vmx_encls_vmexit();
+	VMCS12_CASE64(GUEST_IA32_PERF_GLOBAL_CTRL): return cpu_has_load_perf_global_ctrl();
+	VMCS12_CASE64(HOST_IA32_PERF_GLOBAL_CTRL): return cpu_has_load_perf_global_ctrl();
+	case TPR_THRESHOLD: return cpu_has_vmx_tpr_shadow();
+	case SECONDARY_VM_EXEC_CONTROL: return cpu_has_secondary_exec_ctrls();
+	case GUEST_S_CET: return cpu_has_load_cet_ctrl();
+	case GUEST_SSP: return cpu_has_load_cet_ctrl();
Quoted 8 lines

arch/x86/kvm/vmx/vmcs12.c const unsigned short vmcs12_field_offsets[] = { · line 190

Quoted 4 lines
arch/x86/kvm/vmx/vmcs12.c +4 -0
+	case GUEST_INTR_SSP_TABLE: return cpu_has_load_cet_ctrl();
+	case HOST_S_CET: return cpu_has_load_cet_ctrl();
+	case HOST_SSP: return cpu_has_load_cet_ctrl();
+	case HOST_INTR_SSP_TABLE: return cpu_has_load_cet_ctrl();

Most fields here are not shadowed, e.g., CET-related fields. So, the plan is that new fields should be added here regardless of whether they are shadowed or not, right?

Yep. It'll be mildly annoying to keep up-to-date, but I hopefully having an "unconditional" rule will be less confusing than limiting the checks to fields that are allowed to hit the shadow VMCS.

Quoted 2 lines

And GUEST_INTR_STATUS is missing here. It depends on APICv and is handled explicitly in init_vmcs_shadow_fields().

Gah, I had that one on my todo list, but got sidetracked for a week and completely forgot about it.

Thank you!