x86/apic: KVM: Use cpu_physical_id() to get APIC ID of running vCPU for AVIC
Sean Christopherson
maintainerPATCH 1/1 x86/apic: KVM: Use cpu_physical_id() to get APIC ID of running vCPU for AVIC
Use cpu_physical_id() instead of default_cpu_present_to_apicid() when getting the APIC ID of the pCPU on which a vCPU is running/loaded, as the kernel has gone way off the rails if a vCPU is loaded on a pCPU that has been physically removed from the system. Even if the impossible were to happen, the absolutely worst case scenario is that hardware will ring the AIVC doorbell on the wrong pCPU, i.e. a severely broken system will experience mild performance issues.
Kill off KVM's superfluous kvm_cpu_get_apicid() wrapper along with the for-KVM export of default_cpu_present_to_apicid(), as they existed purely for the wonky AVIC usage.
---
Tip tree folks, I'd like to take this through the kvm-x86 tree (in 7.3) for obvious reasons. I assume the odds of a conflict on the removal of EXPORT_SYMBOL_FOR_KVM() are tiny.
arch/x86/include/asm/kvm_host.h | 10 ---------- arch/x86/kernel/apic/apic_common.c | 1 - arch/x86/kvm/svm/avic.c | 6 +++--- 3 files changed, 3 insertions(+), 14 deletions(-)
arch/x86/include/asm/kvm_host.h +0 -10
@@ -2526,16 +2526,6 @@ static inline void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)kvm_x86_call (vcpu_unblocking)(vcpu); } -static inline int kvm_cpu_get_apicid (int mps_cpu) -{ -#ifdef CONFIG_X86_LOCAL_APIC -return default_cpu_present_to_apicid (mps_cpu); -#else -WARN_ON_ONCE (1 ); -return BAD_APICID ; -#endif -} -int memslot_rmap_alloc (struct kvm_memory_slot *slot,unsigned long npages);#define KVM_CLOCK_VALID_FLAGS \
arch/x86/kernel/apic/apic_common.c +0 -1
@@ -26,7 +26,6 @@ u32 default_cpu_present_to_apicid(int mps_cpu)else return BAD_APICID ; } -EXPORT_SYMBOL_FOR_KVM (default_cpu_present_to_apicid);/* * Set up the logical destination ID when the APIC operates in logical
arch/x86/kvm/svm/avic.c +3 -3
@@ -460,8 +460,8 @@ void avic_ring_doorbell(struct kvm_vcpu *vcpu)int cpu =READ_ONCE (vcpu->cpu);if (cpu !=get_cpu ()) { -wrmsrq (MSR_AMD64_SVM_AVIC_DOORBELL ,kvm_cpu_get_apicid (cpu)); -trace_kvm_avic_doorbell (vcpu->vcpu_id,kvm_cpu_get_apicid (cpu)); +wrmsrq (MSR_AMD64_SVM_AVIC_DOORBELL ,cpu_physical_id (cpu)); +trace_kvm_avic_doorbell (vcpu->vcpu_id,cpu_physical_id (cpu)); }put_cpu (); }
@@ -1013,7 +1013,7 @@ static void __avic_vcpu_load(struct kvm_vcpu *vcpu, int cpu, enum avic_vcpu_action action) {struct kvm_svm *kvm_svm =to_kvm_svm (vcpu->kvm); -int h_physical_id =kvm_cpu_get_apicid (cpu); +int h_physical_id =cpu_physical_id (cpu);struct vcpu_svm *svm =to_svm (vcpu);unsigned long flags;u64 entry;
base-commit: c1f7303302927f9cbf4efedf70f0512cde168c65
Signature · 1 line
2.54.0.1136.gdb2ca164c4-goog
Quoted 9 lines · Sean Christopherson
Use
cpu_physical_id()instead ofdefault_cpu_present_to_apicid()when getting the APIC ID of the pCPU on which a vCPU is running/loaded, as the kernel has gone way off the rails if a vCPU is loaded on a pCPU that has been physically removed from the system. Even if the impossible were to happen, the absolutely worst case scenario is that hardware will ring the AIVC doorbell on the wrong pCPU, i.e. a severely broken system will experience mild performance issues.
Quoted 7 lines
Kill off KVM's superfluous
kvm_cpu_get_apicid()wrapper along with the for-KVM export ofdefault_cpu_present_to_apicid(), as they existed purely for the wonky AVIC usage.Cc Kai Huang <[email protected]>Cc Yosry Ahmed <[email protected]>Signed-off-by Sean Christopherson <[email protected]>
Took me a bit to realize cpu_physical_id() is the APIC ID, but yeah this looks to be equivalent as long as the CPU is present, which as you described should be true as long as we didn't screw up. cpu_physical_id() seems to also be used by VMX anyway.
Quoted 8 lines · Sean Christopherson
Use
cpu_physical_id()instead ofdefault_cpu_present_to_apicid()when getting the APIC ID of the pCPU on which a vCPU is running/loaded, as the kernel has gone way off the rails if a vCPU is loaded on a pCPU that has been physically removed from the system. Even if the impossible were to happen, the absolutely worst case scenario is that hardware will ring the AIVC doorbell on the wrong pCPU, i.e. a severely broken system will experience mild performance issues.
Quoted 7 lines
Kill off KVM's superfluous
kvm_cpu_get_apicid()wrapper along with the for-KVM export ofdefault_cpu_present_to_apicid(), as they existed purely for the wonky AVIC usage.Cc Kai Huang <[email protected]>Cc Yosry Ahmed <[email protected]>Signed-off-by Sean Christopherson <[email protected]>
Quoted 4 lines · Sean Christopherson
Use
cpu_physical_id()instead ofdefault_cpu_present_to_apicid()when getting the APIC ID of the pCPU on which a vCPU is running/loaded, as the kernel has gone way off the rails if a vCPU is loaded on a pCPU that has been physically removed from the system.
Or, has been unloaded from a pCPU that is then soft-offlined and physically hot unplugged before that vCPU ever got a chance to run again!
Quoted 4 lines
Even if the impossible were to
happen, the absolutely worst case scenario is that hardware will ring the AIVC doorbell on the wrong pCPU, i.e. a severely broken system will experience mild performance issues.
Indeed, sending a doorbell to BAD_APICID doesn't seem to do anything bad, so I think we are ok with this.
Quoted 8 lines
Kill off KVM's superfluous
kvm_cpu_get_apicid()wrapper along with the for-KVM export ofdefault_cpu_present_to_apicid(), as they existed purely for the wonky AVIC usage.Cc Kai Huang <[email protected]>Cc Yosry Ahmed <[email protected]>Signed-off-by Sean Christopherson <[email protected]>
- Naveen
Queued, thanks.
Paolo