KVM: x86/xen: Add support for vCPU runstate information

This is how Xen guests do steal time accounting. The hypervisor records
the amount of time spent in each of running/runnable/blocked/offline
states.

In the Xen accounting, a vCPU is still in state RUNSTATE_running while
in Xen for a hypercall or I/O trap, etc. Only if Xen explicitly schedules
does the state become RUNSTATE_blocked. In KVM this means that even when
the vCPU exits the kvm_run loop, the state remains RUNSTATE_running.

The VMM can explicitly set the vCPU to RUNSTATE_blocked by using the
KVM_XEN_VCPU_ATTR_TYPE_RUNSTATE_CURRENT attribute, and can also use
KVM_XEN_VCPU_ATTR_TYPE_RUNSTATE_ADJUST to retrospectively add a given
amount of time to the blocked state and subtract it from the running
state.

The state_entry_time corresponds to get_kvmclock_ns() at the time the
vCPU entered the current state, and the total times of all four states
should always add up to state_entry_time.

Co-developed-by: Joao Martins <joao.m.martins@oracle.com>
Signed-off-by: Joao Martins <joao.m.martins@oracle.com>
Signed-off-by: David Woodhouse <dwmw@amazon.co.uk>
Message-Id: <20210301125309.874953-2-dwmw2@infradead.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

3 files changed, 337 insertions, 2 deletions

diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index 4a5ce57b0bb2..868213ca4f98 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -2956,6 +2956,11 @@ static void record_steal_time(struct kvm_vcpu *vcpu)
 	struct kvm_host_map map;
 	struct kvm_steal_time *st;
 
+	if (kvm_xen_msr_enabled(vcpu->kvm)) {
+		kvm_xen_runstate_set_running(vcpu);
+		return;
+	}
+
 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
 		return;
 
@@ -3760,6 +3765,8 @@ int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
 		r = KVM_XEN_HVM_CONFIG_HYPERCALL_MSR |
 		    KVM_XEN_HVM_CONFIG_INTERCEPT_HCALL |
 		    KVM_XEN_HVM_CONFIG_SHARED_INFO;
+		if (sched_info_on())
+			r |= KVM_XEN_HVM_CONFIG_RUNSTATE;
 		break;
 #endif
 	case KVM_CAP_SYNC_REGS:
@@ -4039,7 +4046,11 @@ void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
 	if (vcpu->preempted && !vcpu->arch.guest_state_protected)
 		vcpu->arch.preempted_in_kernel = !static_call(kvm_x86_get_cpl)(vcpu);
 
-	kvm_steal_time_set_preempted(vcpu);
+	if (kvm_xen_msr_enabled(vcpu->kvm))
+		kvm_xen_runstate_set_preempted(vcpu);
+	else
+		kvm_steal_time_set_preempted(vcpu);
+
 	static_call(kvm_x86_vcpu_put)(vcpu);
 	vcpu->arch.last_host_tsc = rdtsc();
 	/*
diff --git a/arch/x86/kvm/xen.c b/arch/x86/kvm/xen.c
index 77b20ff09078..ae17250e1efe 100644
--- a/arch/x86/kvm/xen.c
+++ b/arch/x86/kvm/xen.c
@@ -11,9 +11,11 @@
 #include "hyperv.h"
 
 #include <linux/kvm_host.h>
+#include <linux/sched/stat.h>
 
 #include <trace/events/kvm.h>
 #include <xen/interface/xen.h>
+#include <xen/interface/vcpu.h>
 
 #include "trace.h"
 
@@ -61,6 +63,132 @@ out:
 	return ret;
 }
 
+static void kvm_xen_update_runstate(struct kvm_vcpu *v, int state)
+{
+	struct kvm_vcpu_xen *vx = &v->arch.xen;
+	u64 now = get_kvmclock_ns(v->kvm);
+	u64 delta_ns = now - vx->runstate_entry_time;
+	u64 run_delay = current->sched_info.run_delay;
+
+	if (unlikely(!vx->runstate_entry_time))
+		vx->current_runstate = RUNSTATE_offline;
+
+	/*
+	 * Time waiting for the scheduler isn't "stolen" if the
+	 * vCPU wasn't running anyway.
+	 */
+	if (vx->current_runstate == RUNSTATE_running) {
+		u64 steal_ns = run_delay - vx->last_steal;
+
+		delta_ns -= steal_ns;
+
+		vx->runstate_times[RUNSTATE_runnable] += steal_ns;
+	}
+	vx->last_steal = run_delay;
+
+	vx->runstate_times[vx->current_runstate] += delta_ns;
+	vx->current_runstate = state;
+	vx->runstate_entry_time = now;
+}
+
+void kvm_xen_update_runstate_guest(struct kvm_vcpu *v, int state)
+{
+	struct kvm_vcpu_xen *vx = &v->arch.xen;
+	uint64_t state_entry_time;
+	unsigned int offset;
+
+	kvm_xen_update_runstate(v, state);
+
+	if (!vx->runstate_set)
+		return;
+
+	BUILD_BUG_ON(sizeof(struct compat_vcpu_runstate_info) != 0x2c);
+
+	offset = offsetof(struct compat_vcpu_runstate_info, state_entry_time);
+#ifdef CONFIG_X86_64
+	/*
+	 * The only difference is alignment of uint64_t in 32-bit.
+	 * So the first field 'state' is accessed directly using
+	 * offsetof() (where its offset happens to be zero), while the
+	 * remaining fields which are all uint64_t, start at 'offset'
+	 * which we tweak here by adding 4.
+	 */
+	BUILD_BUG_ON(offsetof(struct vcpu_runstate_info, state_entry_time) !=
+		     offsetof(struct compat_vcpu_runstate_info, state_entry_time) + 4);
+	BUILD_BUG_ON(offsetof(struct vcpu_runstate_info, time) !=
+		     offsetof(struct compat_vcpu_runstate_info, time) + 4);
+
+	if (v->kvm->arch.xen.long_mode)
+		offset = offsetof(struct vcpu_runstate_info, state_entry_time);
+#endif
+	/*
+	 * First write the updated state_entry_time at the appropriate
+	 * location determined by 'offset'.
+	 */
+	state_entry_time = vx->runstate_entry_time;
+	state_entry_time |= XEN_RUNSTATE_UPDATE;
+
+	BUILD_BUG_ON(sizeof(((struct vcpu_runstate_info *)0)->state_entry_time) !=
+		     sizeof(state_entry_time));
+	BUILD_BUG_ON(sizeof(((struct compat_vcpu_runstate_info *)0)->state_entry_time) !=
+		     sizeof(state_entry_time));
+
+	if (kvm_write_guest_offset_cached(v->kvm, &v->arch.xen.runstate_cache,
+					  &state_entry_time, offset,
+					  sizeof(state_entry_time)))
+		return;
+	smp_wmb();
+
+	/*
+	 * Next, write the new runstate. This is in the *same* place
+	 * for 32-bit and 64-bit guests, asserted here for paranoia.
+	 */
+	BUILD_BUG_ON(offsetof(struct vcpu_runstate_info, state) !=
+		     offsetof(struct compat_vcpu_runstate_info, state));
+	BUILD_BUG_ON(sizeof(((struct vcpu_runstate_info *)0)->state) !=
+		     sizeof(vx->current_runstate));
+	BUILD_BUG_ON(sizeof(((struct compat_vcpu_runstate_info *)0)->state) !=
+		     sizeof(vx->current_runstate));
+
+	if (kvm_write_guest_offset_cached(v->kvm, &v->arch.xen.runstate_cache,
+					  &vx->current_runstate,
+					  offsetof(struct vcpu_runstate_info, state),
+					  sizeof(vx->current_runstate)))
+		return;
+
+	/*
+	 * Write the actual runstate times immediately after the
+	 * runstate_entry_time.
+	 */
+	BUILD_BUG_ON(offsetof(struct vcpu_runstate_info, state_entry_time) !=
+		     offsetof(struct vcpu_runstate_info, time) - sizeof(u64));
+	BUILD_BUG_ON(offsetof(struct compat_vcpu_runstate_info, state_entry_time) !=
+		     offsetof(struct compat_vcpu_runstate_info, time) - sizeof(u64));
+	BUILD_BUG_ON(sizeof(((struct vcpu_runstate_info *)0)->time) !=
+		     sizeof(((struct compat_vcpu_runstate_info *)0)->time));
+	BUILD_BUG_ON(sizeof(((struct vcpu_runstate_info *)0)->time) !=
+		     sizeof(vx->runstate_times));
+
+	if (kvm_write_guest_offset_cached(v->kvm, &v->arch.xen.runstate_cache,
+					  &vx->runstate_times[0],
+					  offset + sizeof(u64),
+					  sizeof(vx->runstate_times)))
+		return;
+
+	smp_wmb();
+
+	/*
+	 * Finally, clear the XEN_RUNSTATE_UPDATE bit in the guest's
+	 * runstate_entry_time field.
+	 */
+
+	state_entry_time &= ~XEN_RUNSTATE_UPDATE;
+	if (kvm_write_guest_offset_cached(v->kvm, &v->arch.xen.runstate_cache,
+					  &state_entry_time, offset,
+					  sizeof(state_entry_time)))
+		return;
+}
+
 int __kvm_xen_has_interrupt(struct kvm_vcpu *v)
 {
 	u8 rc = 0;
@@ -223,6 +351,121 @@ int kvm_xen_vcpu_set_attr(struct kvm_vcpu *vcpu, struct kvm_xen_vcpu_attr *data)
 		}
 		break;
 
+	case KVM_XEN_VCPU_ATTR_TYPE_RUNSTATE_ADDR:
+		if (!sched_info_on()) {
+			r = -EOPNOTSUPP;
+			break;
+		}
+		if (data->u.gpa == GPA_INVALID) {
+			vcpu->arch.xen.runstate_set = false;
+			r = 0;
+			break;
+		}
+
+		r = kvm_gfn_to_hva_cache_init(vcpu->kvm,
+					      &vcpu->arch.xen.runstate_cache,
+					      data->u.gpa,
+					      sizeof(struct vcpu_runstate_info));
+		if (!r) {
+			vcpu->arch.xen.runstate_set = true;
+		}
+		break;
+
+	case KVM_XEN_VCPU_ATTR_TYPE_RUNSTATE_CURRENT:
+		if (!sched_info_on()) {
+			r = -EOPNOTSUPP;
+			break;
+		}
+		if (data->u.runstate.state > RUNSTATE_offline) {
+			r = -EINVAL;
+			break;
+		}
+
+		kvm_xen_update_runstate(vcpu, data->u.runstate.state);
+		r = 0;
+		break;
+
+	case KVM_XEN_VCPU_ATTR_TYPE_RUNSTATE_DATA:
+		if (!sched_info_on()) {
+			r = -EOPNOTSUPP;
+			break;
+		}
+		if (data->u.runstate.state > RUNSTATE_offline) {
+			r = -EINVAL;
+			break;
+		}
+		if (data->u.runstate.state_entry_time !=
+		    (data->u.runstate.time_running +
+		     data->u.runstate.time_runnable +
+		     data->u.runstate.time_blocked +
+		     data->u.runstate.time_offline)) {
+			r = -EINVAL;
+			break;
+		}
+		if (get_kvmclock_ns(vcpu->kvm) <
+		    data->u.runstate.state_entry_time) {
+			r = -EINVAL;
+			break;
+		}
+
+		vcpu->arch.xen.current_runstate = data->u.runstate.state;
+		vcpu->arch.xen.runstate_entry_time =
+			data->u.runstate.state_entry_time;
+		vcpu->arch.xen.runstate_times[RUNSTATE_running] =
+			data->u.runstate.time_running;
+		vcpu->arch.xen.runstate_times[RUNSTATE_runnable] =
+			data->u.runstate.time_runnable;
+		vcpu->arch.xen.runstate_times[RUNSTATE_blocked] =
+			data->u.runstate.time_blocked;
+		vcpu->arch.xen.runstate_times[RUNSTATE_offline] =
+			data->u.runstate.time_offline;
+		vcpu->arch.xen.last_steal = current->sched_info.run_delay;
+		r = 0;
+		break;
+
+	case KVM_XEN_VCPU_ATTR_TYPE_RUNSTATE_ADJUST:
+		if (!sched_info_on()) {
+			r = -EOPNOTSUPP;
+			break;
+		}
+		if (data->u.runstate.state > RUNSTATE_offline &&
+		    data->u.runstate.state != (u64)-1) {
+			r = -EINVAL;
+			break;
+		}
+		/* The adjustment must add up */
+		if (data->u.runstate.state_entry_time !=
+		    (data->u.runstate.time_running +
+		     data->u.runstate.time_runnable +
+		     data->u.runstate.time_blocked +
+		     data->u.runstate.time_offline)) {
+			r = -EINVAL;
+			break;
+		}
+
+		if (get_kvmclock_ns(vcpu->kvm) <
+		    (vcpu->arch.xen.runstate_entry_time +
+		     data->u.runstate.state_entry_time)) {
+			r = -EINVAL;
+			break;
+		}
+
+		vcpu->arch.xen.runstate_entry_time +=
+			data->u.runstate.state_entry_time;
+		vcpu->arch.xen.runstate_times[RUNSTATE_running] +=
+			data->u.runstate.time_running;
+		vcpu->arch.xen.runstate_times[RUNSTATE_runnable] +=
+			data->u.runstate.time_runnable;
+		vcpu->arch.xen.runstate_times[RUNSTATE_blocked] +=
+			data->u.runstate.time_blocked;
+		vcpu->arch.xen.runstate_times[RUNSTATE_offline] +=
+			data->u.runstate.time_offline;
+
+		if (data->u.runstate.state <= RUNSTATE_offline)
+			kvm_xen_update_runstate(vcpu, data->u.runstate.state);
+		r = 0;
+		break;
+
 	default:
 		break;
 	}
@@ -255,6 +498,49 @@ int kvm_xen_vcpu_get_attr(struct kvm_vcpu *vcpu, struct kvm_xen_vcpu_attr *data)
 		r = 0;
 		break;
 
+	case KVM_XEN_VCPU_ATTR_TYPE_RUNSTATE_ADDR:
+		if (!sched_info_on()) {
+			r = -EOPNOTSUPP;
+			break;
+		}
+		if (vcpu->arch.xen.runstate_set) {
+			data->u.gpa = vcpu->arch.xen.runstate_cache.gpa;
+			r = 0;
+		}
+		break;
+
+	case KVM_XEN_VCPU_ATTR_TYPE_RUNSTATE_CURRENT:
+		if (!sched_info_on()) {
+			r = -EOPNOTSUPP;
+			break;
+		}
+		data->u.runstate.state = vcpu->arch.xen.current_runstate;
+		r = 0;
+		break;
+
+	case KVM_XEN_VCPU_ATTR_TYPE_RUNSTATE_DATA:
+		if (!sched_info_on()) {
+			r = -EOPNOTSUPP;
+			break;
+		}
+		data->u.runstate.state = vcpu->arch.xen.current_runstate;
+		data->u.runstate.state_entry_time =
+			vcpu->arch.xen.runstate_entry_time;
+		data->u.runstate.time_running =
+			vcpu->arch.xen.runstate_times[RUNSTATE_running];
+		data->u.runstate.time_runnable =
+			vcpu->arch.xen.runstate_times[RUNSTATE_runnable];
+		data->u.runstate.time_blocked =
+			vcpu->arch.xen.runstate_times[RUNSTATE_blocked];
+		data->u.runstate.time_offline =
+			vcpu->arch.xen.runstate_times[RUNSTATE_offline];
+		r = 0;
+		break;
+
+	case KVM_XEN_VCPU_ATTR_TYPE_RUNSTATE_ADJUST:
+		r = -EINVAL;
+		break;
+
 	default:
 		break;
 	}
diff --git a/arch/x86/kvm/xen.h b/arch/x86/kvm/xen.h
index 87eaf2be9549..463a7844a8ca 100644
--- a/arch/x86/kvm/xen.h
+++ b/arch/x86/kvm/xen.h
@@ -23,6 +23,12 @@ int kvm_xen_write_hypercall_page(struct kvm_vcpu *vcpu, u64 data);
 int kvm_xen_hvm_config(struct kvm *kvm, struct kvm_xen_hvm_config *xhc);
 void kvm_xen_destroy_vm(struct kvm *kvm);
 
+static inline bool kvm_xen_msr_enabled(struct kvm *kvm)
+{
+	return static_branch_unlikely(&kvm_xen_enabled.key) &&
+		kvm->arch.xen_hvm_config.msr;
+}
+
 static inline bool kvm_xen_hypercall_enabled(struct kvm *kvm)
 {
 	return static_branch_unlikely(&kvm_xen_enabled.key) &&
@@ -48,6 +54,11 @@ static inline void kvm_xen_destroy_vm(struct kvm *kvm)
 {
 }
 
+static inline bool kvm_xen_msr_enabled(struct kvm *kvm)
+{
+	return false;
+}
+
 static inline bool kvm_xen_hypercall_enabled(struct kvm *kvm)
 {
 	return false;
@@ -61,10 +72,31 @@ static inline int kvm_xen_has_interrupt(struct kvm_vcpu *vcpu)
 
 int kvm_xen_hypercall(struct kvm_vcpu *vcpu);
 
-/* 32-bit compatibility definitions, also used natively in 32-bit build */
 #include <asm/pvclock-abi.h>
 #include <asm/xen/interface.h>
+#include <xen/interface/vcpu.h>
 
+void kvm_xen_update_runstate_guest(struct kvm_vcpu *vcpu, int state);
+
+static inline void kvm_xen_runstate_set_running(struct kvm_vcpu *vcpu)
+{
+	kvm_xen_update_runstate_guest(vcpu, RUNSTATE_running);
+}
+
+static inline void kvm_xen_runstate_set_preempted(struct kvm_vcpu *vcpu)
+{
+	/*
+	 * If the vCPU wasn't preempted but took a normal exit for
+	 * some reason (hypercalls, I/O, etc.), that is accounted as
+	 * still RUNSTATE_running, as the VMM is still operating on
+	 * behalf of the vCPU. Only if the VMM does actually block
+	 * does it need to enter RUNSTATE_blocked.
+	 */
+	if (vcpu->preempted)
+		kvm_xen_update_runstate_guest(vcpu, RUNSTATE_runnable);
+}
+
+/* 32-bit compatibility definitions, also used natively in 32-bit build */
 struct compat_arch_vcpu_info {
 	unsigned int cr2;
 	unsigned int pad[5];
@@ -97,4 +129,10 @@ struct compat_shared_info {
 	struct compat_arch_shared_info arch;
 };
 
+struct compat_vcpu_runstate_info {
+    int state;
+    uint64_t state_entry_time;
+    uint64_t time[4];
+} __attribute__((packed));
+
 #endif /* __ARCH_X86_KVM_XEN_H__ */