KVM: nVMX: Do not load EOI-exitmap while running L2
authorLiran Alon <liran.alon@oracle.com>
Wed, 21 Mar 2018 00:50:31 +0000 (02:50 +0200)
committerPaolo Bonzini <pbonzini@redhat.com>
Wed, 21 Mar 2018 13:16:44 +0000 (14:16 +0100)
When L1 IOAPIC redirection-table is written, a request of
KVM_REQ_SCAN_IOAPIC is set on all vCPUs. This is done such that
all vCPUs will now recalc their IOAPIC handled vectors and load
it to their EOI-exitmap.

However, it could be that one of the vCPUs is currently running
L2. In this case, load_eoi_exitmap() will be called which would
write to vmcs02->eoi_exit_bitmap, which is wrong because
vmcs02->eoi_exit_bitmap should always be equal to
vmcs12->eoi_exit_bitmap. Furthermore, at this point
KVM_REQ_SCAN_IOAPIC was already consumed and therefore we will
never update vmcs01->eoi_exit_bitmap. This could lead to remote_irr
of some IOAPIC level-triggered entry to remain set forever.

Fix this issue by delaying the load of EOI-exitmap to when vCPU
is running L1.

One may wonder why not just delay entire KVM_REQ_SCAN_IOAPIC
processing to when vCPU is running L1. This is done in order to handle
correctly the case where LAPIC & IO-APIC of L1 is pass-throughed into
L2. In this case, vmcs12->virtual_interrupt_delivery should be 0. In
current nVMX implementation, that results in
vmcs02->virtual_interrupt_delivery to also be 0. Thus,
vmcs02->eoi_exit_bitmap is not used. Therefore, every L2 EOI cause
a #VMExit into L0 (either on MSR_WRITE to x2APIC MSR or
APIC_ACCESS/APIC_WRITE/EPT_MISCONFIG to APIC MMIO page).
In order for such L2 EOI to be broadcasted, if needed, from LAPIC
to IO-APIC, vcpu->arch.ioapic_handled_vectors must be updated
while L2 is running. Therefore, patch makes sure to delay only the
loading of EOI-exitmap but not the update of
vcpu->arch.ioapic_handled_vectors.

Reviewed-by: Arbel Moshe <arbel.moshe@oracle.com>
Reviewed-by: Krish Sadhukhan <krish.sadhukhan@oracle.com>
Signed-off-by: Liran Alon <liran.alon@oracle.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
arch/x86/include/asm/kvm_host.h
arch/x86/kvm/kvm_cache_regs.h
arch/x86/kvm/x86.c

index 8fb60287d987dd64be888a1dddbc61ed65c6e522..c72891dd7d784acad490cf1cff5116fc3bcb1b1d 100644 (file)
@@ -73,6 +73,7 @@
 #define KVM_REQ_HV_RESET               KVM_ARCH_REQ(20)
 #define KVM_REQ_HV_EXIT                        KVM_ARCH_REQ(21)
 #define KVM_REQ_HV_STIMER              KVM_ARCH_REQ(22)
+#define KVM_REQ_LOAD_EOI_EXITMAP       KVM_ARCH_REQ(23)
 
 #define CR0_RESERVED_BITS                                               \
        (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
@@ -498,6 +499,7 @@ struct kvm_vcpu_arch {
        u64 apic_base;
        struct kvm_lapic *apic;    /* kernel irqchip context */
        bool apicv_active;
+       bool load_eoi_exitmap_pending;
        DECLARE_BITMAP(ioapic_handled_vectors, 256);
        unsigned long apic_attention;
        int32_t apic_arb_prio;
index 58ba12071014daf31b59f74a137e6106b60d37c7..9619dcc2b32593422381445f34339bf11f7e6e95 100644 (file)
@@ -93,6 +93,11 @@ static inline void enter_guest_mode(struct kvm_vcpu *vcpu)
 static inline void leave_guest_mode(struct kvm_vcpu *vcpu)
 {
        vcpu->arch.hflags &= ~HF_GUEST_MASK;
+
+       if (vcpu->arch.load_eoi_exitmap_pending) {
+               vcpu->arch.load_eoi_exitmap_pending = false;
+               kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
+       }
 }
 
 static inline bool is_guest_mode(struct kvm_vcpu *vcpu)
index 26a8cd93be0c1353057733122946c02531242041..9d19e6a31ed4d302301f743b82c5b4b3c86bf481 100644 (file)
@@ -7110,8 +7110,6 @@ void kvm_make_scan_ioapic_request(struct kvm *kvm)
 
 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
 {
-       u64 eoi_exit_bitmap[4];
-
        if (!kvm_apic_hw_enabled(vcpu->arch.apic))
                return;
 
@@ -7124,6 +7122,20 @@ static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
                        kvm_x86_ops->sync_pir_to_irr(vcpu);
                kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
        }
+
+       if (is_guest_mode(vcpu))
+               vcpu->arch.load_eoi_exitmap_pending = true;
+       else
+               kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
+}
+
+static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
+{
+       u64 eoi_exit_bitmap[4];
+
+       if (!kvm_apic_hw_enabled(vcpu->arch.apic))
+               return;
+
        bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors,
                  vcpu_to_synic(vcpu)->vec_bitmap, 256);
        kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
@@ -7238,6 +7250,8 @@ static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
                }
                if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
                        vcpu_scan_ioapic(vcpu);
+               if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
+                       vcpu_load_eoi_exitmap(vcpu);
                if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
                        kvm_vcpu_reload_apic_access_page(vcpu);
                if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {