kvm: x86: hyperv: make VP_INDEX managed by userspace

Hyper-V identifies vCPUs by Virtual Processor Index, which can be
queried via HV_X64_MSR_VP_INDEX msr.  It is defined by the spec as a
sequential number which can't exceed the maximum number of vCPUs per VM.
APIC ids can be sparse and thus aren't a valid replacement for VP
indices.

Current KVM uses its internal vcpu index as VP_INDEX.  However, to make
it predictable and persistent across VM migrations, the userspace has to
control the value of VP_INDEX.

This patch achieves that, by storing vp_index explicitly on vcpu, and
allowing HV_X64_MSR_VP_INDEX to be set from the host side.  For
compatibility it's initialized to KVM vcpu index.  Also a few variables
are renamed to make clear distinction betweed this Hyper-V vp_index and
KVM vcpu_id (== APIC id).  Besides, a new capability,
KVM_CAP_HYPERV_VP_INDEX, is added to allow the userspace to skip
attempting msr writes where unsupported, to avoid spamming error logs.
Signed-off-by: Roman Kagan <rkagan@virtuozzo.com>
Signed-off-by: Radim Krčmář <rkrcmar@redhat.com>

Documentation/virtual/kvm/api.txt

@@ -4338,3 +4338,12 @@ This capability enables a newer version of Hyper-V Synthetic interrupt
 controller (SynIC).  The only difference with KVM_CAP_HYPERV_SYNIC is that KVM
 doesn't clear SynIC message and event flags pages when they are enabled by
 writing to the respective MSRs.
+
+8.12 KVM_CAP_HYPERV_VP_INDEX
+
+Architectures: x86
+
+This capability indicates that userspace can load HV_X64_MSR_VP_INDEX msr.  Its
+value is used to denote the target vcpu for a SynIC interrupt.  For
+compatibilty, KVM initializes this msr to KVM's internal vcpu index.  When this
+capability is absent, userspace can still query this msr's value.

arch/x86/include/asm/kvm_host.h

@@ -467,6 +467,7 @@ struct kvm_vcpu_hv_synic {
 
 /* Hyper-V per vcpu emulation context */
 struct kvm_vcpu_hv {
+	u32 vp_index;
 	u64 hv_vapic;
 	s64 runtime_offset;
 	struct kvm_vcpu_hv_synic synic;

arch/x86/kvm/hyperv.c

@@ -106,14 +106,27 @@ static int synic_set_sint(struct kvm_vcpu_hv_synic *synic, int sint,
 	return 0;
 }
 
-static struct kvm_vcpu_hv_synic *synic_get(struct kvm *kvm, u32 vcpu_id)
+static struct kvm_vcpu *get_vcpu_by_vpidx(struct kvm *kvm, u32 vpidx)
+{
+	struct kvm_vcpu *vcpu = NULL;
+	int i;
+
+	if (vpidx < KVM_MAX_VCPUS)
+		vcpu = kvm_get_vcpu(kvm, vpidx);
+	if (vcpu && vcpu_to_hv_vcpu(vcpu)->vp_index == vpidx)
+		return vcpu;
+	kvm_for_each_vcpu(i, vcpu, kvm)
+		if (vcpu_to_hv_vcpu(vcpu)->vp_index == vpidx)
+			return vcpu;
+	return NULL;
+}
+
+static struct kvm_vcpu_hv_synic *synic_get(struct kvm *kvm, u32 vpidx)
 {
 	struct kvm_vcpu *vcpu;
 	struct kvm_vcpu_hv_synic *synic;
 
-	if (vcpu_id >= atomic_read(&kvm->online_vcpus))
-		return NULL;
-	vcpu = kvm_get_vcpu(kvm, vcpu_id);
+	vcpu = get_vcpu_by_vpidx(kvm, vpidx);
 	if (!vcpu)
 		return NULL;
 	synic = vcpu_to_synic(vcpu);
@@ -320,11 +333,11 @@ static int synic_set_irq(struct kvm_vcpu_hv_synic *synic, u32 sint)
 	return ret;
 }
 
-int kvm_hv_synic_set_irq(struct kvm *kvm, u32 vcpu_id, u32 sint)
+int kvm_hv_synic_set_irq(struct kvm *kvm, u32 vpidx, u32 sint)
 {
 	struct kvm_vcpu_hv_synic *synic;
 
-	synic = synic_get(kvm, vcpu_id);
+	synic = synic_get(kvm, vpidx);
 	if (!synic)
 		return -EINVAL;
 
@@ -343,11 +356,11 @@ void kvm_hv_synic_send_eoi(struct kvm_vcpu *vcpu, int vector)
 			kvm_hv_notify_acked_sint(vcpu, i);
 }
 
-static int kvm_hv_set_sint_gsi(struct kvm *kvm, u32 vcpu_id, u32 sint, int gsi)
+static int kvm_hv_set_sint_gsi(struct kvm *kvm, u32 vpidx, u32 sint, int gsi)
 {
 	struct kvm_vcpu_hv_synic *synic;
 
-	synic = synic_get(kvm, vcpu_id);
+	synic = synic_get(kvm, vpidx);
 	if (!synic)
 		return -EINVAL;
 
@@ -689,6 +702,13 @@ void kvm_hv_vcpu_init(struct kvm_vcpu *vcpu)
 		stimer_init(&hv_vcpu->stimer[i], i);
 }
 
+void kvm_hv_vcpu_postcreate(struct kvm_vcpu *vcpu)
+{
+	struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
+
+	hv_vcpu->vp_index = kvm_vcpu_get_idx(vcpu);
+}
+
 int kvm_hv_activate_synic(struct kvm_vcpu *vcpu, bool dont_zero_synic_pages)
 {
 	struct kvm_vcpu_hv_synic *synic = vcpu_to_synic(vcpu);
@@ -983,6 +1003,11 @@ static int kvm_hv_set_msr(struct kvm_vcpu *vcpu, u32 msr, u64 data, bool host)
 	struct kvm_vcpu_hv *hv = &vcpu->arch.hyperv;
 
 	switch (msr) {
+	case HV_X64_MSR_VP_INDEX:
+		if (!host)
+			return 1;
+		hv->vp_index = (u32)data;
+		break;
 	case HV_X64_MSR_APIC_ASSIST_PAGE: {
 		u64 gfn;
 		unsigned long addr;
@@ -1094,18 +1119,9 @@ static int kvm_hv_get_msr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
 	struct kvm_vcpu_hv *hv = &vcpu->arch.hyperv;
 
 	switch (msr) {
-	case HV_X64_MSR_VP_INDEX: {
-		int r;
-		struct kvm_vcpu *v;
-
-		kvm_for_each_vcpu(r, v, vcpu->kvm) {
-			if (v == vcpu) {
-				data = r;
-				break;
-			}
-		}
+	case HV_X64_MSR_VP_INDEX:
+		data = hv->vp_index;
 		break;
-	}
 	case HV_X64_MSR_EOI:
 		return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
 	case HV_X64_MSR_ICR:

arch/x86/kvm/hyperv.h

@@ -59,6 +59,7 @@ void kvm_hv_synic_send_eoi(struct kvm_vcpu *vcpu, int vector);
 int kvm_hv_activate_synic(struct kvm_vcpu *vcpu, bool dont_zero_synic_pages);
 
 void kvm_hv_vcpu_init(struct kvm_vcpu *vcpu);
+void kvm_hv_vcpu_postcreate(struct kvm_vcpu *vcpu);
 void kvm_hv_vcpu_uninit(struct kvm_vcpu *vcpu);
 
 static inline struct kvm_vcpu_hv_stimer *vcpu_to_stimer(struct kvm_vcpu *vcpu,

arch/x86/kvm/x86.c

@@ -2666,6 +2666,7 @@ int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
 	case KVM_CAP_HYPERV_SPIN:
 	case KVM_CAP_HYPERV_SYNIC:
 	case KVM_CAP_HYPERV_SYNIC2:
+	case KVM_CAP_HYPERV_VP_INDEX:
 	case KVM_CAP_PCI_SEGMENT:
 	case KVM_CAP_DEBUGREGS:
 	case KVM_CAP_X86_ROBUST_SINGLESTEP:
@@ -7688,6 +7689,8 @@ void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
 	struct msr_data msr;
 	struct kvm *kvm = vcpu->kvm;
 
+	kvm_hv_vcpu_postcreate(vcpu);
+
 	if (vcpu_load(vcpu))
 		return;
 	msr.data = 0x0;

include/uapi/linux/kvm.h

@@ -928,6 +928,7 @@ struct kvm_ppc_resize_hpt {
 #define KVM_CAP_PPC_FWNMI 146
 #define KVM_CAP_PPC_SMT_POSSIBLE 147
 #define KVM_CAP_HYPERV_SYNIC2 148
+#define KVM_CAP_HYPERV_VP_INDEX 149
 
 #ifdef KVM_CAP_IRQ_ROUTING