CET fix patches for nested guest
v4 cover CET fix patches for nested guest
This patch series is to fix a few issues found during patch review and testing on Linux, also including a patch to explictly disable CET support in nested guest over Hyper-V(s).
change in v4:
- Added guest CET cpuid check before sync vmcs02 CET state to vmcs12.
- Opportunistically added similar fix for MPX.
Yang Weijiang (3):
KVM: nVMX: Sync L2 guest CET states between L1/L2
KVM: nVMX: Set X86_CR4_CET in cr4_fixed1_bits if CET IBT is enabled
KVM: nVMX: Add CET entry/exit load bits to evmcs unsupported list
arch/x86/kvm/cpuid.c | 1 - arch/x86/kvm/vmx/evmcs.c | 4 ++-- arch/x86/kvm/vmx/evmcs.h | 6 ++++-- arch/x86/kvm/vmx/nested.c | 35 +++++++++++++++++++++++++++++++++-- arch/x86/kvm/vmx/vmx.c | 1 + arch/x86/kvm/vmx/vmx.h | 3 +++ 6 files changed, 43 insertions(+), 7 deletions(-)
Signature · 1 line
2.26.2
PATCH v4 1/3 KVM: nVMX: Sync L2 guest CET states between L1/L2
These fields are rarely updated by L1 QEMU/KVM, sync them when L1 is trying to read/write them and after they're changed. If CET guest entry-load bit is not set by L1 guest, migrate them to L2 manaully.
Opportunistically remove one blank line and add minor fix for MPX.
--- arch/x86/kvm/cpuid.c | 1 - arch/x86/kvm/vmx/nested.c | 35 +++++++++++++++++++++++++++++++++-- arch/x86/kvm/vmx/vmx.h | 3 +++ 3 files changed, 36 insertions(+), 3 deletions(-)
arch/x86/kvm/cpuid.c +0 -1
@@ -143,7 +143,6 @@ void kvm_update_cpuid_runtime(struct kvm_vcpu *vcpu) } vcpu->arch.guest_supported_xss = (((u64 )best->edx <<32 ) | best->ecx) & supported_xss; - }else { vcpu->arch.guest_supported_xss =0 ; }
arch/x86/kvm/vmx/nested.c +33 -2
@@ -2516,6 +2516,13 @@ static void prepare_vmcs02_rare(struct vcpu_vmx *vmx, struct vmcs12 *vmcs12)vmcs_write32 (VM_ENTRY_MSR_LOAD_COUNT , vmx->msr_autoload.guest.nr);set_cr4_guest_host_mask (vmx); + +if (kvm_cet_supported () && vmx->nested.nested_run_pending && + (vmcs12->vm_entry_controls &VM_ENTRY_LOAD_CET_STATE )) { +vmcs_writel (GUEST_SSP , vmcs12->guest_ssp); +vmcs_writel (GUEST_S_CET , vmcs12->guest_s_cet); +vmcs_writel (GUEST_INTR_SSP_TABLE , vmcs12->guest_ssp_tbl); + } }/*
@@ -2556,6 +2563,15 @@ static int prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,if (kvm_mpx_supported () && (!vmx->nested.nested_run_pending || !(vmcs12->vm_entry_controls &VM_ENTRY_LOAD_BNDCFGS )))vmcs_write64 (GUEST_BNDCFGS , vmx->nested.vmcs01_guest_bndcfgs); + +if (kvm_cet_supported () && (!vmx->nested.nested_run_pending || + !(vmcs12->vm_entry_controls &VM_ENTRY_LOAD_CET_STATE ))) { +vmcs_writel (GUEST_SSP , vmx->nested.vmcs01_guest_ssp); +vmcs_writel (GUEST_S_CET , vmx->nested.vmcs01_guest_s_cet); +vmcs_writel (GUEST_INTR_SSP_TABLE , + vmx->nested.vmcs01_guest_ssp_tbl); + } +vmx_set_rflags (vcpu, vmcs12->guest_rflags);/* EXCEPTION_BITMAP and CR0_GUEST_HOST_MASK should basically be the
@@ -3373,8 +3389,14 @@ enum nvmx_vmentry_status nested_vmx_enter_non_root_mode(struct kvm_vcpu *vcpu,if (!(vmcs12->vm_entry_controls &VM_ENTRY_LOAD_DEBUG_CONTROLS )) vmx->nested.vmcs01_debugctl =vmcs_read64 (GUEST_IA32_DEBUGCTL );if (kvm_mpx_supported () && - !(vmcs12->vm_entry_controls &VM_ENTRY_LOAD_BNDCFGS )) + !(vmcs12->vm_entry_controls &VM_ENTRY_LOAD_BNDCFGS )) vmx->nested.vmcs01_guest_bndcfgs =vmcs_read64 (GUEST_BNDCFGS ); +if (kvm_cet_supported () && + !(vmcs12->vm_entry_controls &VM_ENTRY_LOAD_CET_STATE )) { + vmx->nested.vmcs01_guest_ssp =vmcs_readl (GUEST_SSP ); + vmx->nested.vmcs01_guest_s_cet =vmcs_readl (GUEST_S_CET ); + vmx->nested.vmcs01_guest_ssp_tbl =vmcs_readl (GUEST_INTR_SSP_TABLE ); + }/* * Overwrite vmcs01.GUEST_CR3 with L1's CR3 if EPT is disabled *and*
@@ -4001,6 +4023,9 @@ static bool is_vmcs12_ext_field(unsigned long field)case GUEST_IDTR_BASE :case GUEST_PENDING_DBG_EXCEPTIONS :case GUEST_BNDCFGS : +case GUEST_SSP : +case GUEST_INTR_SSP_TABLE : +case GUEST_S_CET :return true;default :break ;
@@ -4050,8 +4075,14 @@ static void sync_vmcs02_to_vmcs12_rare(struct kvm_vcpu *vcpu, vmcs12->guest_idtr_base =vmcs_readl (GUEST_IDTR_BASE ); vmcs12->guest_pending_dbg_exceptions =vmcs_readl (GUEST_PENDING_DBG_EXCEPTIONS ); -if (kvm_mpx_supported ()) +if (kvm_mpx_supported () &&guest_cpuid_has (vcpu,X86_FEATURE_MPX )) vmcs12->guest_bndcfgs =vmcs_read64 (GUEST_BNDCFGS ); +if (kvm_cet_supported () && (guest_cpuid_has (vcpu,X86_FEATURE_SHSTK ) || +guest_cpuid_has (vcpu,X86_FEATURE_IBT ))) { + vmcs12->guest_ssp =vmcs_readl (GUEST_SSP ); + vmcs12->guest_s_cet =vmcs_readl (GUEST_S_CET ); + vmcs12->guest_ssp_tbl =vmcs_readl (GUEST_INTR_SSP_TABLE ); + } vmx->nested.need_sync_vmcs02_to_vmcs12_rare = false; }
arch/x86/kvm/vmx/vmx.h +3 -0
@@ -155,6 +155,9 @@ struct nested_vmx {/* to migrate it to L2 if VM_ENTRY_LOAD_DEBUG_CONTROLS is off */ u64 vmcs01_debugctl;u64 vmcs01_guest_bndcfgs; +u64 vmcs01_guest_ssp; +u64 vmcs01_guest_s_cet; +u64 vmcs01_guest_ssp_tbl;/* to migrate it to L1 if L2 writes to L1's CR8 directly */ int l1_tpr_threshold;
Signature · 1 line
2.26.2
Sean Christopherson
maintainerRe: [PATCH v4 1/3] KVM: nVMX: Sync L2 guest CET states between L1/L2
Quoted 19 lines · Yang Weijiang
These fields are rarely updated by L1 QEMU/KVM, sync them when L1 is trying to read/write them and after they're changed. If CET guest entry-load bit is not set by L1 guest, migrate them to L2 manaully.
Opportunistically remove one blank line and add minor fix for MPX.
Suggested-by Sean Christopherson <[email protected]>Signed-off-by Yang Weijiang <[email protected]>--- arch/x86/kvm/cpuid.c | 1 - arch/x86/kvm/vmx/nested.c | 35 +++++++++++++++++++++++++++++++++-- arch/x86/kvm/vmx/vmx.h | 3 +++ 3 files changed, 36 insertions(+), 3 deletions(-)arch/x86/kvm/cpuid.c +0
-0
arch/x86/kvm/cpuid.c void kvm_update_cpuid_runtime(struct kvm_vcpu *vcpu) · line 146
Quoted 4 lines
arch/x86/kvm/cpuid.c +0
-1@@ -143,7 +143,6 @@ void kvm_update_cpuid_runtime(struct kvm_vcpu *vcpu) } vcpu->arch.guest_supported_xss = (((u64 )best->edx <<32 ) | best->ecx) & supported_xss; -
Spurious whitespace deletion.
Quoted 22 lines
arch/x86/kvm/cpuid.c +0
-0}else { vcpu->arch.guest_supported_xss =0 ; }arch/x86/kvm/vmx/nested.c +7
-0@@ -2516,6 +2516,13 @@ static void prepare_vmcs02_rare(struct vcpu_vmx *vmx, struct vmcs12 *vmcs12)vmcs_write32 (VM_ENTRY_MSR_LOAD_COUNT , vmx->msr_autoload.guest.nr);set_cr4_guest_host_mask (vmx); + +if (kvm_cet_supported () && vmx->nested.nested_run_pending && + (vmcs12->vm_entry_controls &VM_ENTRY_LOAD_CET_STATE )) { +vmcs_writel (GUEST_SSP , vmcs12->guest_ssp); +vmcs_writel (GUEST_S_CET , vmcs12->guest_s_cet); +vmcs_writel (GUEST_INTR_SSP_TABLE , vmcs12->guest_ssp_tbl); + } }/*
arch/x86/kvm/vmx/nested.c static int prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12, · line 2568
Quoted 6 lines
arch/x86/kvm/vmx/nested.c +3
-0@@ -2556,6 +2563,15 @@ static int prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,if (kvm_mpx_supported () && (!vmx->nested.nested_run_pending || !(vmcs12->vm_entry_controls &VM_ENTRY_LOAD_BNDCFGS )))vmcs_write64 (GUEST_BNDCFGS , vmx->nested.vmcs01_guest_bndcfgs); + +if (kvm_cet_supported () && (!vmx->nested.nested_run_pending || + !(vmcs12->vm_entry_controls &VM_ENTRY_LOAD_CET_STATE ))) {
Not your code per se, since this pattern comes from BNDCFGS and DEBUGCTL, but I don't see how loading vmcs01 state in this combo is correct:
a. kvm_xxx_supported() == 1
b. nested_run_pending == false
c. vm_entry_controls.load_xxx_state == truenested_vmx_enter_non_root_mode() only snapshots vmcs01 if vm_entry_controls.load_xxx_state == false, which means the above combo is loading stale values (or more likely, zeros).
I _think_ nested_vmx_enter_non_root_mode() just needs to snapshot vmcs01 if nested_run_pending=false. For migration, if userspace restores MSRs after KVM_SET_NESTED_STATE, then what's done here is likely irrelevant. If userspace restores MSRs before nested state, then vmcs01 will hold the desired value since setting MSRs would have written the value into vmcs01.
I suspect no one has reported this issue because guests simply don't use MPX, and up until the recent LBR stuff, KVM effectively zeroed out DEBUGCTL for the guest.
arch/x86/kvm/vmx/nested.c +4 -3
@@ -3298,10 +3298,11 @@ enum nvmx_vmentry_status nested_vmx_enter_non_root_mode(struct kvm_vcpu *vcpu,if (likely (!evaluate_pending_interrupts) &&kvm_vcpu_apicv_active (vcpu)) evaluate_pending_interrupts |=vmx_has_apicv_interrupt (vcpu); -if (!(vmcs12->vm_entry_controls &VM_ENTRY_LOAD_DEBUG_CONTROLS )) +if (!vmx->nested.nested_run_pending || + !(vmcs12->vm_entry_controls &VM_ENTRY_LOAD_DEBUG_CONTROLS )) vmx->nested.vmcs01_debugctl =vmcs_read64 (GUEST_IA32_DEBUGCTL ); -if (kvm_mpx_supported () && - !(vmcs12->vm_entry_controls &VM_ENTRY_LOAD_BNDCFGS )) +if (kvm_mpx_supported () && (!vmx->nested.nested_run_pending || + !(vmcs12->vm_entry_controls &VM_ENTRY_LOAD_BNDCFGS ))) vmx->nested.vmcs01_guest_bndcfgs =vmcs_read64 (GUEST_BNDCFGS );/*
Side topic, all of this code is broken for SMM emulation. SMI+RSM don't do a full VM-Exit -> VM-Entry; the CPU forcefully exits non-root, but most state that is loaded from the VMCS is left untouched. It's the SMI handler's responsibility to not enable features, e.g. to not set CR4.CET. For sane use cases, this probably doesn't matter as vmcs12 will be configured to context switch state, but if L1 is doing anything out of the ordinary, SMI+RSM will corrupt state.
E.g. if L1 enables MPX in the guest, does not intercept L2 writes to BNDCFGS, and does not load BNDCFGS on VM-Entry, then SMI+RSM would corrupt BNDCFGS since the SMI "exit" would clear BNDCFGS, and the RSM "entry" would load zero. This is 100% contrived, and probably doesn't impact real world use cases, but it still bugs me :-)
Quoted 36 lines
arch/x86/kvm/vmx/nested.c +16
-1+vmcs_writel (GUEST_SSP , vmx->nested.vmcs01_guest_ssp); +vmcs_writel (GUEST_S_CET , vmx->nested.vmcs01_guest_s_cet); +vmcs_writel (GUEST_INTR_SSP_TABLE , + vmx->nested.vmcs01_guest_ssp_tbl); + } +vmx_set_rflags (vcpu, vmcs12->guest_rflags);/* EXCEPTION_BITMAP and CR0_GUEST_HOST_MASK should basically be the @@ -3373,8 +3389,14 @@ enum nvmx_vmentry_status nested_vmx_enter_non_root_mode(struct kvm_vcpu *vcpu,if (!(vmcs12->vm_entry_controls &VM_ENTRY_LOAD_DEBUG_CONTROLS )) vmx->nested.vmcs01_debugctl =vmcs_read64 (GUEST_IA32_DEBUGCTL );if (kvm_mpx_supported () && - !(vmcs12->vm_entry_controls &VM_ENTRY_LOAD_BNDCFGS )) + !(vmcs12->vm_entry_controls &VM_ENTRY_LOAD_BNDCFGS )) vmx->nested.vmcs01_guest_bndcfgs =vmcs_read64 (GUEST_BNDCFGS ); +if (kvm_cet_supported () && + !(vmcs12->vm_entry_controls &VM_ENTRY_LOAD_CET_STATE )) { + vmx->nested.vmcs01_guest_ssp =vmcs_readl (GUEST_SSP ); + vmx->nested.vmcs01_guest_s_cet =vmcs_readl (GUEST_S_CET ); + vmx->nested.vmcs01_guest_ssp_tbl =vmcs_readl (GUEST_INTR_SSP_TABLE ); + }/* * Overwrite vmcs01.GUEST_CR3 with L1's CR3 if EPT is disabled *and* @@ -4001,6 +4023,9 @@ static bool is_vmcs12_ext_field(unsigned long field)case GUEST_IDTR_BASE :case GUEST_PENDING_DBG_EXCEPTIONS :case GUEST_BNDCFGS : +case GUEST_SSP : +case GUEST_INTR_SSP_TABLE : +case GUEST_S_CET :return true;default :break ;
arch/x86/kvm/vmx/nested.c static void sync_vmcs02_to_vmcs12_rare(struct kvm_vcpu *vcpu, · line 4078
Quoted 5 lines
arch/x86/kvm/vmx/nested.c +1
-1@@ -4050,8 +4075,14 @@ static void sync_vmcs02_to_vmcs12_rare(struct kvm_vcpu *vcpu, vmcs12->guest_idtr_base =vmcs_readl (GUEST_IDTR_BASE ); vmcs12->guest_pending_dbg_exceptions =vmcs_readl (GUEST_PENDING_DBG_EXCEPTIONS ); -if (kvm_mpx_supported ()) +if (kvm_mpx_supported () &&guest_cpuid_has (vcpu,X86_FEATURE_MPX ))
Adding the CPUID check for MPX definitely needs to be a separate commit.
Quoted 27 lines
arch/x86/kvm/vmx/nested.c +6
-0vmcs12->guest_bndcfgs =vmcs_read64 (GUEST_BNDCFGS ); +if (kvm_cet_supported () && (guest_cpuid_has (vcpu,X86_FEATURE_SHSTK ) || +guest_cpuid_has (vcpu,X86_FEATURE_IBT ))) { + vmcs12->guest_ssp =vmcs_readl (GUEST_SSP ); + vmcs12->guest_s_cet =vmcs_readl (GUEST_S_CET ); + vmcs12->guest_ssp_tbl =vmcs_readl (GUEST_INTR_SSP_TABLE ); + } vmx->nested.need_sync_vmcs02_to_vmcs12_rare = false; }arch/x86/kvm/vmx/vmx.h +3
-0@@ -155,6 +155,9 @@ struct nested_vmx {/* to migrate it to L2 if VM_ENTRY_LOAD_DEBUG_CONTROLS is off */ u64 vmcs01_debugctl;u64 vmcs01_guest_bndcfgs; +u64 vmcs01_guest_ssp; +u64 vmcs01_guest_s_cet; +u64 vmcs01_guest_ssp_tbl;/* to migrate it to L1 if L2 writes to L1's CR8 directly */ int l1_tpr_threshold;Signature · 1 line
2.26.2
Re: [PATCH v4 1/3] KVM: nVMX: Sync L2 guest CET states between L1/L2
Quoted 15 lines · Sean Christopherson
Quoted 14 lines · Yang Weijiang
These fields are rarely updated by L1 QEMU/KVM, sync them when L1 is trying to read/write them and after they're changed. If CET guest entry-load bit is not set by L1 guest, migrate them to L2 manaully.
Opportunistically remove one blank line and add minor fix for MPX.
Suggested-by Sean Christopherson <[email protected]>Signed-off-by Yang Weijiang <[email protected]>--- arch/x86/kvm/cpuid.c | 1 - arch/x86/kvm/vmx/nested.c | 35 +++++++++++++++++++++++++++++++++-- arch/x86/kvm/vmx/vmx.h | 3 +++ 3 files changed, 36 insertions(+), 3 deletions(-)
Quoted 11 lines
arch/x86/kvm/cpuid.c void kvm_update_cpuid_runtime(struct kvm_vcpu *vcpu) · line 146
Quoted 9 lines
arch/x86/kvm/cpuid.c +0
-1@@ -143,7 +143,6 @@ void kvm_update_cpuid_runtime(struct kvm_vcpu *vcpu) } vcpu->arch.guest_supported_xss = (((u64 )best->edx <<32 ) | best->ecx) & supported_xss; -Spurious whitespace deletion.
Yes, Opportunistically did it as said in commit log :-D
Quoted 41 lines
Quoted 23 lines
arch/x86/kvm/cpuid.c +0
-0}else { vcpu->arch.guest_supported_xss =0 ; }arch/x86/kvm/vmx/nested.c +7
-0@@ -2516,6 +2516,13 @@ static void prepare_vmcs02_rare(struct vcpu_vmx *vmx, struct vmcs12 *vmcs12)vmcs_write32 (VM_ENTRY_MSR_LOAD_COUNT , vmx->msr_autoload.guest.nr);set_cr4_guest_host_mask (vmx); + +if (kvm_cet_supported () && vmx->nested.nested_run_pending && + (vmcs12->vm_entry_controls &VM_ENTRY_LOAD_CET_STATE )) { +vmcs_writel (GUEST_SSP , vmcs12->guest_ssp); +vmcs_writel (GUEST_S_CET , vmcs12->guest_s_cet); +vmcs_writel (GUEST_INTR_SSP_TABLE , vmcs12->guest_ssp_tbl); + } }/* arch/x86/kvm/vmx/nested.c static int prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12, · line 2568
Quoted 6 lines
arch/x86/kvm/vmx/nested.c +3
-0@@ -2556,6 +2563,15 @@ static int prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,if (kvm_mpx_supported () && (!vmx->nested.nested_run_pending || !(vmcs12->vm_entry_controls &VM_ENTRY_LOAD_BNDCFGS )))vmcs_write64 (GUEST_BNDCFGS , vmx->nested.vmcs01_guest_bndcfgs); + +if (kvm_cet_supported () && (!vmx->nested.nested_run_pending || + !(vmcs12->vm_entry_controls &VM_ENTRY_LOAD_CET_STATE ))) {Not your code per se, since this pattern comes from BNDCFGS and DEBUGCTL, but I don't see how loading vmcs01 state in this combo is correct:
a. kvm_xxx_supported() == 1 b. nested_run_pending == false c. vm_entry_controls.load_xxx_state == true
nested_vmx_enter_non_root_mode()only snapshots vmcs01 if vm_entry_controls.load_xxx_state == false, which means the above combo is loading stale values (or more likely, zeros).
Quoted 5 lines
I _think_
nested_vmx_enter_non_root_mode()just needs to snapshot vmcs01 if nested_run_pending=false. For migration, if userspace restores MSRs afterKVM_SET_NESTED_STATE, then what's done here is likely irrelevant. If userspace restores MSRs before nested state, then vmcs01 will hold the desired value since setting MSRs would have written the value into vmcs01.
Then the code nested_vmx_enter_non_root_mode() would look like:
if (kvm_cet_supported() && !vmx->nested.nested_run_pending &&
!(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_CET_STATE)) {
...
}I have another concern now, if vm_entry_controls.load_cet_state == false, and L1 updated vmcs fields, so the latest states are in vmcs12, but they cannot be synced to vmcs02 because in prepare_vmcs02_rare():
if (kvm_cet_supported() && vmx->nested.nested_run_pending &&
(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_CET_STATE)) {
...
}so L2 got stale status. IMO, L1 guest sets vm_entry_controls.load_cet_state == false should be rare case. We event can igore this case :-)
Quoted 5 lines
I suspect no one has reported this issue because guests simply don't use MPX, and up until the recent LBR stuff, KVM effectively zeroed out DEBUGCTL for the guest.
So for MPX and DEBUGCTL, is it worth some separate fix patch?
Quoted 28 lines
arch/x86/kvm/vmx/nested.c +4
-3@@ -3298,10 +3298,11 @@ enum nvmx_vmentry_status nested_vmx_enter_non_root_mode(struct kvm_vcpu *vcpu,if (likely (!evaluate_pending_interrupts) &&kvm_vcpu_apicv_active (vcpu)) evaluate_pending_interrupts |=vmx_has_apicv_interrupt (vcpu); -if (!(vmcs12->vm_entry_controls &VM_ENTRY_LOAD_DEBUG_CONTROLS )) +if (!vmx->nested.nested_run_pending || + !(vmcs12->vm_entry_controls &VM_ENTRY_LOAD_DEBUG_CONTROLS )) vmx->nested.vmcs01_debugctl =vmcs_read64 (GUEST_IA32_DEBUGCTL ); -if (kvm_mpx_supported () && - !(vmcs12->vm_entry_controls &VM_ENTRY_LOAD_BNDCFGS )) +if (kvm_mpx_supported () && (!vmx->nested.nested_run_pending || + !(vmcs12->vm_entry_controls &VM_ENTRY_LOAD_BNDCFGS ))) vmx->nested.vmcs01_guest_bndcfgs =vmcs_read64 (GUEST_BNDCFGS );/* Side topic, all of this code is broken for SMM emulation. SMI+RSM don't do a full VM-Exit -> VM-Entry; the CPU forcefully exits non-root, but most state that is loaded from the VMCS is left untouched. It's the SMI handler's responsibility to not enable features, e.g. to not set CR4.CET. For sane use cases, this probably doesn't matter as vmcs12 will be configured to context switch state, but if L1 is doing anything out of the ordinary, SMI+RSM will corrupt state.
Quoted 5 lines
E.g. if L1 enables MPX in the guest, does not intercept L2 writes to BNDCFGS, and does not load BNDCFGS on VM-Entry, then SMI+RSM would corrupt BNDCFGS since the SMI "exit" would clear BNDCFGS, and the RSM "entry" would load zero. This is 100% contrived, and probably doesn't impact real world use cases, but it still bugs me :-)
Exactly, should it be fixed by separate patch or leave it as is?
Quoted 36 lines
Quoted 36 lines
arch/x86/kvm/vmx/nested.c +16
-1+vmcs_writel (GUEST_SSP , vmx->nested.vmcs01_guest_ssp); +vmcs_writel (GUEST_S_CET , vmx->nested.vmcs01_guest_s_cet); +vmcs_writel (GUEST_INTR_SSP_TABLE , + vmx->nested.vmcs01_guest_ssp_tbl); + } +vmx_set_rflags (vcpu, vmcs12->guest_rflags);/* EXCEPTION_BITMAP and CR0_GUEST_HOST_MASK should basically be the @@ -3373,8 +3389,14 @@ enum nvmx_vmentry_status nested_vmx_enter_non_root_mode(struct kvm_vcpu *vcpu,if (!(vmcs12->vm_entry_controls &VM_ENTRY_LOAD_DEBUG_CONTROLS )) vmx->nested.vmcs01_debugctl =vmcs_read64 (GUEST_IA32_DEBUGCTL );if (kvm_mpx_supported () && - !(vmcs12->vm_entry_controls &VM_ENTRY_LOAD_BNDCFGS )) + !(vmcs12->vm_entry_controls &VM_ENTRY_LOAD_BNDCFGS )) vmx->nested.vmcs01_guest_bndcfgs =vmcs_read64 (GUEST_BNDCFGS ); +if (kvm_cet_supported () && + !(vmcs12->vm_entry_controls &VM_ENTRY_LOAD_CET_STATE )) { + vmx->nested.vmcs01_guest_ssp =vmcs_readl (GUEST_SSP ); + vmx->nested.vmcs01_guest_s_cet =vmcs_readl (GUEST_S_CET ); + vmx->nested.vmcs01_guest_ssp_tbl =vmcs_readl (GUEST_INTR_SSP_TABLE ); + }/* * Overwrite vmcs01.GUEST_CR3 with L1's CR3 if EPT is disabled *and* @@ -4001,6 +4023,9 @@ static bool is_vmcs12_ext_field(unsigned long field)case GUEST_IDTR_BASE :case GUEST_PENDING_DBG_EXCEPTIONS :case GUEST_BNDCFGS : +case GUEST_SSP : +case GUEST_INTR_SSP_TABLE : +case GUEST_S_CET :return true;default :break ;
Quoted 8 lines
arch/x86/kvm/vmx/nested.c static void sync_vmcs02_to_vmcs12_rare(struct kvm_vcpu *vcpu, · line 4078
Quoted 6 lines
arch/x86/kvm/vmx/nested.c +1
-1@@ -4050,8 +4075,14 @@ static void sync_vmcs02_to_vmcs12_rare(struct kvm_vcpu *vcpu, vmcs12->guest_idtr_base =vmcs_readl (GUEST_IDTR_BASE ); vmcs12->guest_pending_dbg_exceptions =vmcs_readl (GUEST_PENDING_DBG_EXCEPTIONS ); -if (kvm_mpx_supported ()) +if (kvm_mpx_supported () &&guest_cpuid_has (vcpu,X86_FEATURE_MPX ))Adding the CPUID check for MPX definitely needs to be a separate commit.
Sure, will fix them by separate patch. Thanks for review!
Quoted 28 lines
Quoted 28 lines
arch/x86/kvm/vmx/nested.c +6
-0vmcs12->guest_bndcfgs =vmcs_read64 (GUEST_BNDCFGS ); +if (kvm_cet_supported () && (guest_cpuid_has (vcpu,X86_FEATURE_SHSTK ) || +guest_cpuid_has (vcpu,X86_FEATURE_IBT ))) { + vmcs12->guest_ssp =vmcs_readl (GUEST_SSP ); + vmcs12->guest_s_cet =vmcs_readl (GUEST_S_CET ); + vmcs12->guest_ssp_tbl =vmcs_readl (GUEST_INTR_SSP_TABLE ); + } vmx->nested.need_sync_vmcs02_to_vmcs12_rare = false; }arch/x86/kvm/vmx/vmx.h +3
-0@@ -155,6 +155,9 @@ struct nested_vmx {/* to migrate it to L2 if VM_ENTRY_LOAD_DEBUG_CONTROLS is off */ u64 vmcs01_debugctl;u64 vmcs01_guest_bndcfgs; +u64 vmcs01_guest_ssp; +u64 vmcs01_guest_s_cet; +u64 vmcs01_guest_ssp_tbl;/* to migrate it to L1 if L2 writes to L1's CR8 directly */ int l1_tpr_threshold;Signature · 1 line
2.26.2
Re: [PATCH v4 1/3] KVM: nVMX: Sync L2 guest CET states between L1/L2
On Tue, Mar 16, 2021 at 05:03:47PM +0800, Yang Weijiang wrote:
Hi, Sean,
Could you respond my below rely? I'm not sure how to proceed, thanks!
Quoted 214 lines · Sean Christopherson
Quoted 15 lines · Yang Weijiang
Quoted 14 lines
These fields are rarely updated by L1 QEMU/KVM, sync them when L1 is trying to read/write them and after they're changed. If CET guest entry-load bit is not set by L1 guest, migrate them to L2 manaully.
Opportunistically remove one blank line and add minor fix for MPX.
Suggested-by Sean Christopherson <[email protected]>Signed-off-by Yang Weijiang <[email protected]>--- arch/x86/kvm/cpuid.c | 1 - arch/x86/kvm/vmx/nested.c | 35 +++++++++++++++++++++++++++++++++-- arch/x86/kvm/vmx/vmx.h | 3 +++ 3 files changed, 36 insertions(+), 3 deletions(-)Quoted 11 lines
arch/x86/kvm/cpuid.c void kvm_update_cpuid_runtime(struct kvm_vcpu *vcpu) · line 146
Quoted 9 lines
arch/x86/kvm/cpuid.c +0
-1@@ -143,7 +143,6 @@ void kvm_update_cpuid_runtime(struct kvm_vcpu *vcpu) } vcpu->arch.guest_supported_xss = (((u64 )best->edx <<32 ) | best->ecx) & supported_xss; -Spurious whitespace deletion.
Yes, Opportunistically did it as said in commit log :-D
Quoted 41 lines
Quoted 23 lines
arch/x86/kvm/cpuid.c +0
-0}else { vcpu->arch.guest_supported_xss =0 ; }arch/x86/kvm/vmx/nested.c +7
-0@@ -2516,6 +2516,13 @@ static void prepare_vmcs02_rare(struct vcpu_vmx *vmx, struct vmcs12 *vmcs12)vmcs_write32 (VM_ENTRY_MSR_LOAD_COUNT , vmx->msr_autoload.guest.nr);set_cr4_guest_host_mask (vmx); + +if (kvm_cet_supported () && vmx->nested.nested_run_pending && + (vmcs12->vm_entry_controls &VM_ENTRY_LOAD_CET_STATE )) { +vmcs_writel (GUEST_SSP , vmcs12->guest_ssp); +vmcs_writel (GUEST_S_CET , vmcs12->guest_s_cet); +vmcs_writel (GUEST_INTR_SSP_TABLE , vmcs12->guest_ssp_tbl); + } }/* arch/x86/kvm/vmx/nested.c static int prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12, · line 2568
Quoted 6 lines
arch/x86/kvm/vmx/nested.c +3
-0@@ -2556,6 +2563,15 @@ static int prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,if (kvm_mpx_supported () && (!vmx->nested.nested_run_pending || !(vmcs12->vm_entry_controls &VM_ENTRY_LOAD_BNDCFGS )))vmcs_write64 (GUEST_BNDCFGS , vmx->nested.vmcs01_guest_bndcfgs); + +if (kvm_cet_supported () && (!vmx->nested.nested_run_pending || + !(vmcs12->vm_entry_controls &VM_ENTRY_LOAD_CET_STATE ))) {Not your code per se, since this pattern comes from BNDCFGS and DEBUGCTL, but I don't see how loading vmcs01 state in this combo is correct:
a. kvm_xxx_supported() == 1 b. nested_run_pending == false c. vm_entry_controls.load_xxx_state == true
nested_vmx_enter_non_root_mode()only snapshots vmcs01 if vm_entry_controls.load_xxx_state == false, which means the above combo is loading stale values (or more likely, zeros).Quoted 5 lines
I _think_
nested_vmx_enter_non_root_mode()just needs to snapshot vmcs01 if nested_run_pending=false. For migration, if userspace restores MSRs afterKVM_SET_NESTED_STATE, then what's done here is likely irrelevant. If userspace restores MSRs before nested state, then vmcs01 will hold the desired value since setting MSRs would have written the value into vmcs01.Then the code
nested_vmx_enter_non_root_mode()would look like:if (
kvm_cet_supported()&& !vmx->nested.nested_run_pending &&!(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_CET_STATE)) { ... }I have another concern now, if vm_entry_controls.load_cet_state == false, and L1 updated vmcs fields, so the latest states are in vmcs12, but they cannot be synced to vmcs02 because in
prepare_vmcs02_rare():if (
kvm_cet_supported()&&vmx->nested.nested_run_pending&&(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_CET_STATE)) { ... }so L2 got stale status. IMO, L1 guest sets vm_entry_controls.load_cet_state == false should be rare case. We can even igore this case :-)
Quoted 5 lines
I suspect no one has reported this issue because guests simply don't use MPX, and up until the recent LBR stuff, KVM effectively zeroed out DEBUGCTL for the guest.
So for MPX and DEBUGCTL, is it worth some separate fix patch?
Quoted 28 lines
arch/x86/kvm/vmx/nested.c +4
-3@@ -3298,10 +3298,11 @@ enum nvmx_vmentry_status nested_vmx_enter_non_root_mode(struct kvm_vcpu *vcpu,if (likely (!evaluate_pending_interrupts) &&kvm_vcpu_apicv_active (vcpu)) evaluate_pending_interrupts |=vmx_has_apicv_interrupt (vcpu); -if (!(vmcs12->vm_entry_controls &VM_ENTRY_LOAD_DEBUG_CONTROLS )) +if (!vmx->nested.nested_run_pending || + !(vmcs12->vm_entry_controls &VM_ENTRY_LOAD_DEBUG_CONTROLS )) vmx->nested.vmcs01_debugctl =vmcs_read64 (GUEST_IA32_DEBUGCTL ); -if (kvm_mpx_supported () && - !(vmcs12->vm_entry_controls &VM_ENTRY_LOAD_BNDCFGS )) +if (kvm_mpx_supported () && (!vmx->nested.nested_run_pending || + !(vmcs12->vm_entry_controls &VM_ENTRY_LOAD_BNDCFGS ))) vmx->nested.vmcs01_guest_bndcfgs =vmcs_read64 (GUEST_BNDCFGS );/* Side topic, all of this code is broken for SMM emulation. SMI+RSM don't do a full VM-Exit -> VM-Entry; the CPU forcefully exits non-root, but most state that is loaded from the VMCS is left untouched. It's the SMI handler's responsibility to not enable features, e.g. to not set CR4.CET. For sane use cases, this probably doesn't matter as vmcs12 will be configured to context switch state, but if L1 is doing anything out of the ordinary, SMI+RSM will corrupt state.
Quoted 5 lines
E.g. if L1 enables MPX in the guest, does not intercept L2 writes to BNDCFGS, and does not load BNDCFGS on VM-Entry, then SMI+RSM would corrupt BNDCFGS since the SMI "exit" would clear BNDCFGS, and the RSM "entry" would load zero. This is 100% contrived, and probably doesn't impact real world use cases, but it still bugs me :-)
Exactly, should it be fixed by separate patch or leave it as is?
Quoted 36 lines
Quoted 36 lines
arch/x86/kvm/vmx/nested.c +16
-1+vmcs_writel (GUEST_SSP , vmx->nested.vmcs01_guest_ssp); +vmcs_writel (GUEST_S_CET , vmx->nested.vmcs01_guest_s_cet); +vmcs_writel (GUEST_INTR_SSP_TABLE , + vmx->nested.vmcs01_guest_ssp_tbl); + } +vmx_set_rflags (vcpu, vmcs12->guest_rflags);/* EXCEPTION_BITMAP and CR0_GUEST_HOST_MASK should basically be the @@ -3373,8 +3389,14 @@ enum nvmx_vmentry_status nested_vmx_enter_non_root_mode(struct kvm_vcpu *vcpu,if (!(vmcs12->vm_entry_controls &VM_ENTRY_LOAD_DEBUG_CONTROLS )) vmx->nested.vmcs01_debugctl =vmcs_read64 (GUEST_IA32_DEBUGCTL );if (kvm_mpx_supported () && - !(vmcs12->vm_entry_controls &VM_ENTRY_LOAD_BNDCFGS )) + !(vmcs12->vm_entry_controls &VM_ENTRY_LOAD_BNDCFGS )) vmx->nested.vmcs01_guest_bndcfgs =vmcs_read64 (GUEST_BNDCFGS ); +if (kvm_cet_supported () && + !(vmcs12->vm_entry_controls &VM_ENTRY_LOAD_CET_STATE )) { + vmx->nested.vmcs01_guest_ssp =vmcs_readl (GUEST_SSP ); + vmx->nested.vmcs01_guest_s_cet =vmcs_readl (GUEST_S_CET ); + vmx->nested.vmcs01_guest_ssp_tbl =vmcs_readl (GUEST_INTR_SSP_TABLE ); + }/* * Overwrite vmcs01.GUEST_CR3 with L1's CR3 if EPT is disabled *and* @@ -4001,6 +4023,9 @@ static bool is_vmcs12_ext_field(unsigned long field)case GUEST_IDTR_BASE :case GUEST_PENDING_DBG_EXCEPTIONS :case GUEST_BNDCFGS : +case GUEST_SSP : +case GUEST_INTR_SSP_TABLE : +case GUEST_S_CET :return true;default :break ;Quoted 8 lines
arch/x86/kvm/vmx/nested.c static void sync_vmcs02_to_vmcs12_rare(struct kvm_vcpu *vcpu, · line 4078
Quoted 6 lines
arch/x86/kvm/vmx/nested.c +1
-1@@ -4050,8 +4075,14 @@ static void sync_vmcs02_to_vmcs12_rare(struct kvm_vcpu *vcpu, vmcs12->guest_idtr_base =vmcs_readl (GUEST_IDTR_BASE ); vmcs12->guest_pending_dbg_exceptions =vmcs_readl (GUEST_PENDING_DBG_EXCEPTIONS ); -if (kvm_mpx_supported ()) +if (kvm_mpx_supported () &&guest_cpuid_has (vcpu,X86_FEATURE_MPX ))Adding the CPUID check for MPX definitely needs to be a separate commit.
Sure, will fix them by separate patch. Thanks for review!
Quoted 28 lines
Quoted 28 lines
arch/x86/kvm/vmx/nested.c +6
-0vmcs12->guest_bndcfgs =vmcs_read64 (GUEST_BNDCFGS ); +if (kvm_cet_supported () && (guest_cpuid_has (vcpu,X86_FEATURE_SHSTK ) || +guest_cpuid_has (vcpu,X86_FEATURE_IBT ))) { + vmcs12->guest_ssp =vmcs_readl (GUEST_SSP ); + vmcs12->guest_s_cet =vmcs_readl (GUEST_S_CET ); + vmcs12->guest_ssp_tbl =vmcs_readl (GUEST_INTR_SSP_TABLE ); + } vmx->nested.need_sync_vmcs02_to_vmcs12_rare = false; }arch/x86/kvm/vmx/vmx.h +3
-0@@ -155,6 +155,9 @@ struct nested_vmx {/* to migrate it to L2 if VM_ENTRY_LOAD_DEBUG_CONTROLS is off */ u64 vmcs01_debugctl;u64 vmcs01_guest_bndcfgs; +u64 vmcs01_guest_ssp; +u64 vmcs01_guest_s_cet; +u64 vmcs01_guest_ssp_tbl;/* to migrate it to L1 if L2 writes to L1's CR8 directly */ int l1_tpr_threshold;Signature · 1 line
2.26.2
Sean Christopherson
maintainerRe: [PATCH v4 1/3] KVM: nVMX: Sync L2 guest CET states between L1/L2
Quoted 7 lines · Yang Weijiang
On Tue, Mar 16, 2021 at 05:03:47PM +0800, Yang Weijiang wrote:
Hi, Sean,
Could you respond my below rely? I'm not sure how to proceed, thanks!Quoted 2 lines · Sean Christopherson
Quoted 1 line · Yang Weijiang
...
Quoted 36 lines
Quoted 36 lines
Quoted 19 lines
Review static int prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12, · line 2568
Quoted 7 lines
(quoted diff) +3
-0@@ -2556,6 +2563,15 @@ static int prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12, if (kvm_mpx_supported() && (!vmx->nested.nested_run_pending || !(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS))) vmcs_write64(GUEST_BNDCFGS, vmx->nested.vmcs01_guest_bndcfgs); + + if (kvm_cet_supported() && (!vmx->nested.nested_run_pending || + !(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_CET_STATE))) {Not your code per se, since this pattern comes from BNDCFGS and DEBUGCTL, but I don't see how loading vmcs01 state in this combo is correct:
a. kvm_xxx_supported() == 1 b. nested_run_pending == false c. vm_entry_controls.load_xxx_state == true
nested_vmx_enter_non_root_mode()only snapshots vmcs01 if vm_entry_controls.load_xxx_state == false, which means the above combo is loading stale values (or more likely, zeros).Quoted 5 lines
I _think_
nested_vmx_enter_non_root_mode()just needs to snapshot vmcs01 if nested_run_pending=false. For migration, if userspace restores MSRs afterKVM_SET_NESTED_STATE, then what's done here is likely irrelevant. If userspace restores MSRs before nested state, then vmcs01 will hold the desired value since setting MSRs would have written the value into vmcs01.Then the code
nested_vmx_enter_non_root_mode()would look like:if (
kvm_cet_supported()&& !vmx->nested.nested_run_pending &&!(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_CET_STATE)) { ... }I have another concern now, if vm_entry_controls.load_cet_state == false, and L1 updated vmcs fields, so the latest states are in vmcs12, but they cannot be synced to vmcs02 because in
prepare_vmcs02_rare():
Quoted 7 lines
Quoted 7 lines
if (
kvm_cet_supported()&&vmx->nested.nested_run_pending&&(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_CET_STATE)) { ... }so L2 got stale status. IMO, L1 guest sets vm_entry_controls.load_cet_state == false should be rare case. We can even igore this case :-)
Yes, that's an L1 bug if it expects L2 state to come from vmcs12 in that case. Architecturally, the vcms12 value won't be visible to L2 until L1 enables the VM-Entry control, at which point KVM would detect the refreshed vmcs12 and sync the "rare" fields.
Quoted 5 lines
Quoted 5 lines
Quoted 4 lines
I suspect no one has reported this issue because guests simply don't use MPX, and up until the recent LBR stuff, KVM effectively zeroed out DEBUGCTL for the guest.
So for MPX and DEBUGCTL, is it worth some separate fix patch?
Yes, assuming my analysis is correct. That doesn't necessarily need to be your responsibility, though patches are of course welcome :-)
Jim, Paolo, any thoughts?
Quoted 28 lines
Quoted 28 lines
Quoted 28 lines
arch/x86/kvm/vmx/nested.c +4
-3@@ -3298,10 +3298,11 @@ enum nvmx_vmentry_status nested_vmx_enter_non_root_mode(struct kvm_vcpu *vcpu,if (likely (!evaluate_pending_interrupts) &&kvm_vcpu_apicv_active (vcpu)) evaluate_pending_interrupts |=vmx_has_apicv_interrupt (vcpu); -if (!(vmcs12->vm_entry_controls &VM_ENTRY_LOAD_DEBUG_CONTROLS )) +if (!vmx->nested.nested_run_pending || + !(vmcs12->vm_entry_controls &VM_ENTRY_LOAD_DEBUG_CONTROLS )) vmx->nested.vmcs01_debugctl =vmcs_read64 (GUEST_IA32_DEBUGCTL ); -if (kvm_mpx_supported () && - !(vmcs12->vm_entry_controls &VM_ENTRY_LOAD_BNDCFGS )) +if (kvm_mpx_supported () && (!vmx->nested.nested_run_pending || + !(vmcs12->vm_entry_controls &VM_ENTRY_LOAD_BNDCFGS ))) vmx->nested.vmcs01_guest_bndcfgs =vmcs_read64 (GUEST_BNDCFGS );/* Side topic, all of this code is broken for SMM emulation. SMI+RSM don't do a full VM-Exit -> VM-Entry; the CPU forcefully exits non-root, but most state that is loaded from the VMCS is left untouched. It's the SMI handler's responsibility to not enable features, e.g. to not set CR4.CET. For sane use cases, this probably doesn't matter as vmcs12 will be configured to context switch state, but if L1 is doing anything out of the ordinary, SMI+RSM will corrupt state.
Quoted 7 lines
Quoted 7 lines
Quoted 5 lines
E.g. if L1 enables MPX in the guest, does not intercept L2 writes to BNDCFGS, and does not load BNDCFGS on VM-Entry, then SMI+RSM would corrupt BNDCFGS since the SMI "exit" would clear BNDCFGS, and the RSM "entry" would load zero. This is 100% contrived, and probably doesn't impact real world use cases, but it still bugs me :-)
Exactly, should it be fixed by separate patch or leave it as is?
Definitely leave it for now, properly fixing the SMI+RSM code goes far beyond basic CET support.
Re: [PATCH v4 1/3] KVM: nVMX: Sync L2 guest CET states between L1/L2
Quoted 52 lines · Sean Christopherson
Quoted 7 lines · Yang Weijiang
On Tue, Mar 16, 2021 at 05:03:47PM +0800, Yang Weijiang wrote:
Hi, Sean,
Could you respond my below rely? I'm not sure how to proceed, thanks!Quoted 2 lines · Sean Christopherson
Quoted 1 line · Yang Weijiang
...
Quoted 41 lines
Quoted 41 lines
Quoted 19 lines
Review static int prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12, · line 2568
Quoted 7 lines
(quoted diff) +3
-0@@ -2556,6 +2563,15 @@ static int prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12, if (kvm_mpx_supported() && (!vmx->nested.nested_run_pending || !(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS))) vmcs_write64(GUEST_BNDCFGS, vmx->nested.vmcs01_guest_bndcfgs); + + if (kvm_cet_supported() && (!vmx->nested.nested_run_pending || + !(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_CET_STATE))) {Not your code per se, since this pattern comes from BNDCFGS and DEBUGCTL, but I don't see how loading vmcs01 state in this combo is correct:
a. kvm_xxx_supported() == 1 b. nested_run_pending == false c. vm_entry_controls.load_xxx_state == true
nested_vmx_enter_non_root_mode()only snapshots vmcs01 if vm_entry_controls.load_xxx_state == false, which means the above combo is loading stale values (or more likely, zeros).Quoted 5 lines
I _think_
nested_vmx_enter_non_root_mode()just needs to snapshot vmcs01 if nested_run_pending=false. For migration, if userspace restores MSRs afterKVM_SET_NESTED_STATE, then what's done here is likely irrelevant. If userspace restores MSRs before nested state, then vmcs01 will hold the desired value since setting MSRs would have written the value into vmcs01.Then the code
nested_vmx_enter_non_root_mode()would look like:if (
kvm_cet_supported()&& !vmx->nested.nested_run_pending &&!(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_CET_STATE)) { ... }I have another concern now, if vm_entry_controls.load_cet_state == false, and L1 updated vmcs fields, so the latest states are in vmcs12, but they cannot be synced to vmcs02 because in
prepare_vmcs02_rare():if (
kvm_cet_supported()&&vmx->nested.nested_run_pending&&(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_CET_STATE)) { ... }
Quoted 7 lines
Quoted 2 lines
Quoted 2 lines
so L2 got stale status. IMO, L1 guest sets vm_entry_controls.load_cet_state == false should be rare case. We can even igore this case :-)
Yes, that's an L1 bug if it expects L2 state to come from vmcs12 in that case. Architecturally, the vcms12 value won't be visible to L2 until L1 enables the VM-Entry control, at which point KVM would detect the refreshed vmcs12 and sync the "rare" fields.
Thanks, Sean!
So I'll change code as below:
if (kvm_cet_supported() && !vmx->nested.nested_run_pending &&
!(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_CET_STATE)) {
...
}Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
I suspect no one has reported this issue because guests simply don't use MPX, and up until the recent LBR stuff, KVM effectively zeroed out DEBUGCTL for the guest.
Quoted 7 lines
Quoted 1 line
Quoted 1 line
So for MPX and DEBUGCTL, is it worth some separate fix patch?
Yes, assuming my analysis is correct. That doesn't necessarily need to be your responsibility, though patches are of course welcome :-)
Jim, Paolo, any thoughts?
OK, let me wait for Jim and Paolo's comments on this...
Quoted 36 lines
Quoted 28 lines
Quoted 28 lines
Quoted 28 lines
arch/x86/kvm/vmx/nested.c +4
-3@@ -3298,10 +3298,11 @@ enum nvmx_vmentry_status nested_vmx_enter_non_root_mode(struct kvm_vcpu *vcpu,if (likely (!evaluate_pending_interrupts) &&kvm_vcpu_apicv_active (vcpu)) evaluate_pending_interrupts |=vmx_has_apicv_interrupt (vcpu); -if (!(vmcs12->vm_entry_controls &VM_ENTRY_LOAD_DEBUG_CONTROLS )) +if (!vmx->nested.nested_run_pending || + !(vmcs12->vm_entry_controls &VM_ENTRY_LOAD_DEBUG_CONTROLS )) vmx->nested.vmcs01_debugctl =vmcs_read64 (GUEST_IA32_DEBUGCTL ); -if (kvm_mpx_supported () && - !(vmcs12->vm_entry_controls &VM_ENTRY_LOAD_BNDCFGS )) +if (kvm_mpx_supported () && (!vmx->nested.nested_run_pending || + !(vmcs12->vm_entry_controls &VM_ENTRY_LOAD_BNDCFGS ))) vmx->nested.vmcs01_guest_bndcfgs =vmcs_read64 (GUEST_BNDCFGS );/* Side topic, all of this code is broken for SMM emulation. SMI+RSM don't do a full VM-Exit -> VM-Entry; the CPU forcefully exits non-root, but most state that is loaded from the VMCS is left untouched. It's the SMI handler's responsibility to not enable features, e.g. to not set CR4.CET. For sane use cases, this probably doesn't matter as vmcs12 will be configured to context switch state, but if L1 is doing anything out of the ordinary, SMI+RSM will corrupt state.
Quoted 7 lines
Quoted 7 lines
Quoted 5 lines
E.g. if L1 enables MPX in the guest, does not intercept L2 writes to BNDCFGS, and does not load BNDCFGS on VM-Entry, then SMI+RSM would corrupt BNDCFGS since the SMI "exit" would clear BNDCFGS, and the RSM "entry" would load zero. This is 100% contrived, and probably doesn't impact real world use cases, but it still bugs me :-)
Exactly, should it be fixed by separate patch or leave it as is?
Quoted 2 lines
Definitely leave it for now, properly fixing the SMI+RSM code goes far beyond basic CET support.
Sure.
PATCH v4 2/3 KVM: nVMX: Set X86_CR4_CET in cr4_fixed1_bits if CET IBT is enabled
In reply to Yang WeijiangCET SHSTK and IBT are independently controlled by kernel, set X86_CR4_CET bit in cr4_fixed1_bits if either of them is enabled so that nested guest can enjoy the feature.
--- arch/x86/kvm/vmx/vmx.c | 1 + 1 file changed, 1 insertion(+)
arch/x86/kvm/vmx/vmx.c +1 -0
@@ -7267,6 +7267,7 @@ static void nested_vmx_cr_fixed1_bits_update(struct kvm_vcpu *vcpu)cr4_fixed1_update (X86_CR4_UMIP , ecx,feature_bit (UMIP ));cr4_fixed1_update (X86_CR4_LA57 , ecx,feature_bit (LA57 ));cr4_fixed1_update (X86_CR4_CET , ecx,feature_bit (SHSTK )); +cr4_fixed1_update (X86_CR4_CET , edx,feature_bit (IBT ));#undef cr4_fixed1_update }
Signature · 1 line
2.26.2
PATCH v4 3/3 KVM: nVMX: Add CET entry/exit load bits to evmcs unsupported list
In reply to Yang WeijiangNested guest doesn't support CET when KVM is running as an intermediate layer between two Hyper-Vs for now, so mask out related CET entry/exit load bits. Relevant enabling patches will be posted as a separate patch series.
--- arch/x86/kvm/vmx/evmcs.c | 4 ++-- arch/x86/kvm/vmx/evmcs.h | 6 ++++-- 2 files changed, 6 insertions(+), 4 deletions(-)
arch/x86/kvm/vmx/evmcs.c +2 -2
@@ -351,11 +351,11 @@ void nested_evmcs_filter_control_msr(u32 msr_index, u64 *pdata)switch (msr_index) {case MSR_IA32_VMX_EXIT_CTLS :case MSR_IA32_VMX_TRUE_EXIT_CTLS : - ctl_high &= ~VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL ; + ctl_high &= ~EVMCS1_UNSUPPORTED_VMEXIT_CTRL ;break ;case MSR_IA32_VMX_ENTRY_CTLS :case MSR_IA32_VMX_TRUE_ENTRY_CTLS : - ctl_high &= ~VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL ; + ctl_high &= ~EVMCS1_UNSUPPORTED_VMENTRY_CTRL ;break ;case MSR_IA32_VMX_PROCBASED_CTLS2 : ctl_high &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES ;
arch/x86/kvm/vmx/evmcs.h +4 -2
@@ -59,8 +59,10 @@ DECLARE_STATIC_KEY_FALSE(enable_evmcs); SECONDARY_EXEC_SHADOW_VMCS | \ SECONDARY_EXEC_TSC_SCALING | \ SECONDARY_EXEC_PAUSE_LOOP_EXITING) -#define EVMCS1_UNSUPPORTED_VMEXIT_CTRL (VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL) -#define EVMCS1_UNSUPPORTED_VMENTRY_CTRL (VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL) +#define EVMCS1_UNSUPPORTED_VMEXIT_CTRL (VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL | \ + VM_EXIT_LOAD_CET_STATE) +#define EVMCS1_UNSUPPORTED_VMENTRY_CTRL (VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL | \ + VM_ENTRY_LOAD_CET_STATE)#define EVMCS1_UNSUPPORTED_VMFUNC (VMX_VMFUNC_EPTP_SWITCHING)#if IS_ENABLED (CONFIG_HYPERV )
Signature · 1 line
2.26.2