Re: [PATCH v2 3/4] x86/kvm/hyper-v: direct mode for synthetic timers

[Paolo Bonzini <pbonzini@redhat.com>]

On 26/11/18 16:47, Vitaly Kuznetsov wrote:
> Turns out Hyper-V on KVM (as of 2016) will only use synthetic timers
> if direct mode is available. With direct mode we notify the guest by
> asserting APIC irq instead of sending a SynIC message.
> 
> The implementation uses existing vec_bitmap for letting lapic code
> know that we're interested in the particular IRQ's EOI request. We assume
> that the same APIC irq won't be used by the guest for both direct mode
> stimer and as sint source (especially with AutoEOI semantics). It is
> unclear how things should be handled if that's not true.
> 
> Direct mode is also somewhat less expensive; in my testing
> stimer_send_msg() takes not less than 1500 cpu cycles and
> stimer_notify_direct() can usually be done in 300-400. WS2016 without
> Hyper-V, however, always sticks to non-direct version.
> 
> Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
> ---
> - Changes since v1: avoid open-coding stimer_mark_pending() in
>   kvm_hv_synic_send_eoi() [Paolo Bonzini]
> ---
>  arch/x86/kvm/hyperv.c    | 67 +++++++++++++++++++++++++++++++++++-----
>  arch/x86/kvm/trace.h     | 10 +++---
>  arch/x86/kvm/x86.c       |  1 +
>  include/uapi/linux/kvm.h |  1 +
>  4 files changed, 67 insertions(+), 12 deletions(-)
> 
> diff --git a/arch/x86/kvm/hyperv.c b/arch/x86/kvm/hyperv.c
> index eaec15c738df..9533133be566 100644
> --- a/arch/x86/kvm/hyperv.c
> +++ b/arch/x86/kvm/hyperv.c
> @@ -38,6 +38,9 @@
>  
>  #define KVM_HV_MAX_SPARSE_VCPU_SET_BITS DIV_ROUND_UP(KVM_MAX_VCPUS, 64)
>  
> +static void stimer_mark_pending(struct kvm_vcpu_hv_stimer *stimer,
> +				bool vcpu_kick);
> +
>  static inline u64 synic_read_sint(struct kvm_vcpu_hv_synic *synic, int sint)
>  {
>  	return atomic64_read(&synic->sint[sint]);
> @@ -53,8 +56,21 @@ static inline int synic_get_sint_vector(u64 sint_value)
>  static bool synic_has_vector_connected(struct kvm_vcpu_hv_synic *synic,
>  				      int vector)
>  {
> +	struct kvm_vcpu *vcpu = synic_to_vcpu(synic);
> +	struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
> +	struct kvm_vcpu_hv_stimer *stimer;
>  	int i;
>  
> +	for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++) {
> +		stimer = &hv_vcpu->stimer[i];
> +		if (stimer->config.enable && stimer->config.direct_mode &&
> +		    stimer->config.apic_vector == vector)
> +			return true;
> +	}
> +
> +	if (vector < HV_SYNIC_FIRST_VALID_VECTOR)
> +		return false;
> +
>  	for (i = 0; i < ARRAY_SIZE(synic->sint); i++) {
>  		if (synic_get_sint_vector(synic_read_sint(synic, i)) == vector)
>  			return true;
> @@ -80,14 +96,14 @@ static bool synic_has_vector_auto_eoi(struct kvm_vcpu_hv_synic *synic,
>  static void synic_update_vector(struct kvm_vcpu_hv_synic *synic,
>  				int vector)
>  {
> -	if (vector < HV_SYNIC_FIRST_VALID_VECTOR)
> -		return;
> -
>  	if (synic_has_vector_connected(synic, vector))
>  		__set_bit(vector, synic->vec_bitmap);
>  	else
>  		__clear_bit(vector, synic->vec_bitmap);
>  
> +	if (vector < HV_SYNIC_FIRST_VALID_VECTOR)
> +		return;
> +
>  	if (synic_has_vector_auto_eoi(synic, vector))
>  		__set_bit(vector, synic->auto_eoi_bitmap);
>  	else
> @@ -202,6 +218,7 @@ static void kvm_hv_notify_acked_sint(struct kvm_vcpu *vcpu, u32 sint)
>  	for (idx = 0; idx < ARRAY_SIZE(hv_vcpu->stimer); idx++) {
>  		stimer = &hv_vcpu->stimer[idx];
>  		if (stimer->msg_pending && stimer->config.enable &&
> +		    !stimer->config.direct_mode &&
>  		    stimer->config.sintx == sint) {
>  			set_bit(stimer->index,
>  				hv_vcpu->stimer_pending_bitmap);
> @@ -371,7 +388,9 @@ int kvm_hv_synic_set_irq(struct kvm *kvm, u32 vpidx, u32 sint)
>  
>  void kvm_hv_synic_send_eoi(struct kvm_vcpu *vcpu, int vector)
>  {
> +	struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
>  	struct kvm_vcpu_hv_synic *synic = vcpu_to_synic(vcpu);
> +	struct kvm_vcpu_hv_stimer *stimer;
>  	int i;
>  
>  	trace_kvm_hv_synic_send_eoi(vcpu->vcpu_id, vector);
> @@ -379,6 +398,14 @@ void kvm_hv_synic_send_eoi(struct kvm_vcpu *vcpu, int vector)
>  	for (i = 0; i < ARRAY_SIZE(synic->sint); i++)
>  		if (synic_get_sint_vector(synic_read_sint(synic, i)) == vector)
>  			kvm_hv_notify_acked_sint(vcpu, i);
> +
> +	for (i = 0; i < ARRAY_SIZE(hv_vcpu->stimer); i++) {
> +		stimer = &hv_vcpu->stimer[i];
> +		if (stimer->msg_pending && stimer->config.enable &&
> +		    stimer->config.direct_mode &&
> +		    stimer->config.apic_vector == vector)
> +			stimer_mark_pending(stimer, false);
> +	}
>  }
>  
>  static int kvm_hv_set_sint_gsi(struct kvm *kvm, u32 vpidx, u32 sint, int gsi)
> @@ -545,15 +572,25 @@ static int stimer_start(struct kvm_vcpu_hv_stimer *stimer)
>  static int stimer_set_config(struct kvm_vcpu_hv_stimer *stimer, u64 config,
>  			     bool host)
>  {
> -	union hv_stimer_config new_config = {.as_uint64 = config};
> +	struct kvm_vcpu *vcpu = stimer_to_vcpu(stimer);
> +	struct kvm_vcpu_hv *hv_vcpu = vcpu_to_hv_vcpu(vcpu);
> +	union hv_stimer_config new_config = {.as_uint64 = config},
> +		old_config = {.as_uint64 = stimer->config.as_uint64};
>  
>  	trace_kvm_hv_stimer_set_config(stimer_to_vcpu(stimer)->vcpu_id,
>  				       stimer->index, config, host);
>  
>  	stimer_cleanup(stimer);
> -	if (stimer->config.enable && new_config.sintx == 0)
> +	if (old_config.enable &&
> +	    !new_config.direct_mode && new_config.sintx == 0)
>  		new_config.enable = 0;
>  	stimer->config.as_uint64 = new_config.as_uint64;
> +
> +	if (old_config.direct_mode)
> +		synic_update_vector(&hv_vcpu->synic, old_config.apic_vector);
> +	if (new_config.direct_mode)
> +		synic_update_vector(&hv_vcpu->synic, new_config.apic_vector);
> +
>  	stimer_mark_pending(stimer, false);
>  	return 0;
>  }
> @@ -640,14 +677,28 @@ static int stimer_send_msg(struct kvm_vcpu_hv_stimer *stimer)
>  				 stimer->config.sintx, msg);
>  }
>  
> +static int stimer_notify_direct(struct kvm_vcpu_hv_stimer *stimer)
> +{
> +	struct kvm_vcpu *vcpu = stimer_to_vcpu(stimer);
> +	struct kvm_lapic_irq irq = {
> +		.delivery_mode = APIC_DM_FIXED,
> +		.vector = stimer->config.apic_vector
> +	};
> +
> +	return !kvm_apic_set_irq(vcpu, &irq, NULL);
> +}
> +
>  static void stimer_expiration(struct kvm_vcpu_hv_stimer *stimer)
>  {
> -	int r;
> +	int r, direct = stimer->config.direct_mode;
>  
>  	stimer->msg_pending = true;
> -	r = stimer_send_msg(stimer);
> +	if (!direct)
> +		r = stimer_send_msg(stimer);
> +	else
> +		r = stimer_notify_direct(stimer);
>  	trace_kvm_hv_stimer_expiration(stimer_to_vcpu(stimer)->vcpu_id,
> -				       stimer->index, r);
> +				       stimer->index, direct, r);
>  	if (!r) {
>  		stimer->msg_pending = false;
>  		if (!(stimer->config.periodic))
> diff --git a/arch/x86/kvm/trace.h b/arch/x86/kvm/trace.h
> index 0659465a745c..705f40ae2532 100644
> --- a/arch/x86/kvm/trace.h
> +++ b/arch/x86/kvm/trace.h
> @@ -1254,24 +1254,26 @@ TRACE_EVENT(kvm_hv_stimer_callback,
>   * Tracepoint for stimer_expiration.
>   */
>  TRACE_EVENT(kvm_hv_stimer_expiration,
> -	TP_PROTO(int vcpu_id, int timer_index, int msg_send_result),
> -	TP_ARGS(vcpu_id, timer_index, msg_send_result),
> +	TP_PROTO(int vcpu_id, int timer_index, int direct, int msg_send_result),
> +	TP_ARGS(vcpu_id, timer_index, direct, msg_send_result),
>  
>  	TP_STRUCT__entry(
>  		__field(int, vcpu_id)
>  		__field(int, timer_index)
> +		__field(int, direct)
>  		__field(int, msg_send_result)
>  	),
>  
>  	TP_fast_assign(
>  		__entry->vcpu_id = vcpu_id;
>  		__entry->timer_index = timer_index;
> +		__entry->direct = direct;
>  		__entry->msg_send_result = msg_send_result;
>  	),
>  
> -	TP_printk("vcpu_id %d timer %d msg send result %d",
> +	TP_printk("vcpu_id %d timer %d direct %d send result %d",
>  		  __entry->vcpu_id, __entry->timer_index,
> -		  __entry->msg_send_result)
> +		  __entry->direct, __entry->msg_send_result)
>  );
>  
>  /*
> diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
> index 5cd5647120f2..b21b5ceb8d26 100644
> --- a/arch/x86/kvm/x86.c
> +++ b/arch/x86/kvm/x86.c
> @@ -2997,6 +2997,7 @@ int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
>  	case KVM_CAP_HYPERV_TLBFLUSH:
>  	case KVM_CAP_HYPERV_SEND_IPI:
>  	case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
> +	case KVM_CAP_HYPERV_STIMER_DIRECT:
>  	case KVM_CAP_PCI_SEGMENT:
>  	case KVM_CAP_DEBUGREGS:
>  	case KVM_CAP_X86_ROBUST_SINGLESTEP:
> diff --git a/include/uapi/linux/kvm.h b/include/uapi/linux/kvm.h
> index 2b7a652c9fa4..b8da14cee8e5 100644
> --- a/include/uapi/linux/kvm.h
> +++ b/include/uapi/linux/kvm.h
> @@ -975,6 +975,7 @@ struct kvm_ppc_resize_hpt {
>  #define KVM_CAP_HYPERV_ENLIGHTENED_VMCS 163
>  #define KVM_CAP_EXCEPTION_PAYLOAD 164
>  #define KVM_CAP_ARM_VM_IPA_SIZE 165
> +#define KVM_CAP_HYPERV_STIMER_DIRECT 166

I wonder if all these capabilities shouldn't be replaced by a single
KVM_GET_HYPERV_SUPPORTED_CPUID ioctl, or something like that.  If you
can do it for 4.21, before this one cap is crystallized into userspace
API, that would be great. :)

Paolo