[v3,16/17] KVM: x86/xen: handle PV spinlocks slowpath

Commit Message

David Woodhouse March 3, 2022, 3:41 p.m. UTC

From: Boris Ostrovsky <[e-mail]>

Add support for SCHEDOP_poll hypercall.

This implementation is optimized for polling for a single channel, which
is what Linux does. Polling for multiple channels is not especially
efficient (and has not been tested).

PV spinlocks slow path uses this hypercall, and explicitly crash if it's
not supported.

[ dwmw2: Rework to use kvm_vcpu_halt(), not supported for 32-bit guests ]
Signed-off-by: Boris Ostrovsky <[e-mail]>
Signed-off-by: David Woodhouse <[e-mail]>
---
 arch/x86/include/asm/kvm_host.h               |   3 +
 arch/x86/kvm/xen.c                            | 158 +++++++++++++++++-
 .../selftests/kvm/x86_64/xen_shinfo_test.c    |   6 +
 3 files changed, 165 insertions(+), 2 deletions(-)

Patch

diff --git a/arch/x86/include/asm/kvm_host.h b/arch/x86/include/asm/kvm_host.h
index 30f6fc549235..c7cf34adf3de 100644
--- a/arch/x86/include/asm/kvm_host.h
+++ b/arch/x86/include/asm/kvm_host.h
@@ -615,6 +615,8 @@  struct kvm_vcpu_xen {
 	u64 timer_expires; /* In guest epoch */
 	atomic_t timer_pending;
 	struct hrtimer timer;
+	int poll_evtchn;
+	struct timer_list poll_timer;
 };
 
 struct kvm_vcpu_arch {
@@ -1029,6 +1031,7 @@  struct kvm_xen {
 	u8 upcall_vector;
 	struct gfn_to_pfn_cache shinfo_cache;
 	struct idr evtchn_ports;
+	unsigned long poll_mask[BITS_TO_LONGS(KVM_MAX_VCPUS)];
 };
 
 enum kvm_irqchip_mode {
diff --git a/arch/x86/kvm/xen.c b/arch/x86/kvm/xen.c
index 3980cafa0b68..8c85a71aa8ca 100644
--- a/arch/x86/kvm/xen.c
+++ b/arch/x86/kvm/xen.c
@@ -10,6 +10,7 @@ 
 #include "xen.h"
 #include "lapic.h"
 #include "hyperv.h"
+#include "lapic.h"
 
 #include <linux/eventfd.h>
 #include <linux/kvm_host.h>
@@ -972,9 +973,146 @@  static int kvm_xen_hypercall_complete_userspace(struct kvm_vcpu *vcpu)
 	return kvm_xen_hypercall_set_result(vcpu, run->xen.u.hcall.result);
 }
 
-static bool kvm_xen_hcall_sched_op(struct kvm_vcpu *vcpu, int cmd, u64 param, u64 *r)
+static bool wait_pending_event(struct kvm_vcpu *vcpu, int nr_ports,
+			       evtchn_port_t *ports)
+{
+	struct kvm *kvm = vcpu->kvm;
+	struct gfn_to_pfn_cache *gpc = &kvm->arch.xen.shinfo_cache;
+	unsigned long *pending_bits;
+	unsigned long flags;
+	bool ret = true;
+	int idx, i;
+
+	read_lock_irqsave(&gpc->lock, flags);
+	idx = srcu_read_lock(&kvm->srcu);
+	if (!kvm_gfn_to_pfn_cache_check(kvm, gpc, gpc->gpa, PAGE_SIZE))
+		goto out_rcu;
+
+	ret = false;
+	if (IS_ENABLED(CONFIG_64BIT) && kvm->arch.xen.long_mode) {
+		struct shared_info *shinfo = gpc->khva;
+		pending_bits = (unsigned long *)&shinfo->evtchn_pending;
+	} else {
+		struct compat_shared_info *shinfo = gpc->khva;
+		pending_bits = (unsigned long *)&shinfo->evtchn_pending;
+	}
+
+	for (i = 0; i < nr_ports; i++) {
+		if (test_bit(ports[i], pending_bits)) {
+			ret = true;
+			break;
+		}
+	}
+
+ out_rcu:
+	srcu_read_unlock(&kvm->srcu, idx);
+	read_unlock_irqrestore(&gpc->lock, flags);
+
+	return ret;
+}
+
+static bool kvm_xen_schedop_poll(struct kvm_vcpu *vcpu, bool longmode,
+				 u64 param, u64 *r)
+{
+	int idx, i;
+	struct sched_poll sched_poll;
+	evtchn_port_t port, *ports;
+	gpa_t gpa;
+
+	if (!longmode || !lapic_in_kernel(vcpu) ||
+	    !(vcpu->kvm->arch.xen_hvm_config.flags & KVM_XEN_HVM_CONFIG_EVTCHN_SEND))
+		return false;
+
+	idx = srcu_read_lock(&vcpu->kvm->srcu);
+	gpa = kvm_mmu_gva_to_gpa_system(vcpu, param, NULL);
+	srcu_read_unlock(&vcpu->kvm->srcu, idx);
+
+	if (!gpa || kvm_vcpu_read_guest(vcpu, gpa, &sched_poll,
+					sizeof(sched_poll))) {
+		*r = -EFAULT;
+		return true;
+	}
+
+	if (unlikely(sched_poll.nr_ports > 1)) {
+		/* Xen (unofficially) limits number of pollers to 128 */
+		if (sched_poll.nr_ports > 128) {
+			*r = -EINVAL;
+			return true;
+		}
+
+		ports = kmalloc_array(sched_poll.nr_ports,
+				      sizeof(*ports), GFP_KERNEL);
+		if (!ports) {
+			*r = -ENOMEM;
+			return true;
+		}
+	} else
+		ports = &port;
+
+	for (i = 0; i < sched_poll.nr_ports; i++) {
+		idx = srcu_read_lock(&vcpu->kvm->srcu);
+		gpa = kvm_mmu_gva_to_gpa_system(vcpu,
+						(gva_t)(sched_poll.ports + i),
+						NULL);
+		srcu_read_unlock(&vcpu->kvm->srcu, idx);
+
+		if (!gpa || kvm_vcpu_read_guest(vcpu, gpa,
+						&ports[i], sizeof(port))) {
+			*r = -EFAULT;
+			goto out;
+		}
+	}
+
+	if (sched_poll.nr_ports == 1)
+		vcpu->arch.xen.poll_evtchn = port;
+	else
+		vcpu->arch.xen.poll_evtchn = -1;
+
+	set_bit(kvm_vcpu_get_idx(vcpu), vcpu->kvm->arch.xen.poll_mask);
+
+	if (!wait_pending_event(vcpu, sched_poll.nr_ports, ports)) {
+		vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
+
+		if (sched_poll.timeout)
+			mod_timer(&vcpu->arch.xen.poll_timer,
+				  jiffies + nsecs_to_jiffies(sched_poll.timeout));
+
+		kvm_vcpu_halt(vcpu);
+
+		if (sched_poll.timeout)
+			del_timer(&vcpu->arch.xen.poll_timer);
+
+		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
+		kvm_clear_request(KVM_REQ_UNHALT, vcpu);
+	}
+
+	vcpu->arch.xen.poll_evtchn = 0;
+	*r = 0;
+out:
+	/* Really, this is only needed in case of timeout */
+	clear_bit(kvm_vcpu_get_idx(vcpu), vcpu->kvm->arch.xen.poll_mask);
+
+	if (unlikely(sched_poll.nr_ports > 1))
+		kfree(ports);
+	return true;
+}
+
+static void cancel_evtchn_poll(struct timer_list *t)
+{
+	struct kvm_vcpu *vcpu = from_timer(vcpu, t, arch.xen.poll_timer);
+
+	kvm_make_request(KVM_REQ_UNBLOCK, vcpu);
+	kvm_vcpu_kick(vcpu);
+}
+
+static bool kvm_xen_hcall_sched_op(struct kvm_vcpu *vcpu, bool longmode,
+				   int cmd, u64 param, u64 *r)
 {
 	switch (cmd) {
+	case SCHEDOP_poll:
+		if (kvm_xen_schedop_poll(vcpu, longmode, param, r))
+			return true;
+		fallthrough;
 	case SCHEDOP_yield:
 		kvm_vcpu_on_spin(vcpu, true);
 		*r = 0;
@@ -1139,7 +1277,8 @@  int kvm_xen_hypercall(struct kvm_vcpu *vcpu)
 			handled = kvm_xen_hcall_evtchn_send(vcpu, params[1], &r);
 		break;
 	case __HYPERVISOR_sched_op:
-		handled = kvm_xen_hcall_sched_op(vcpu, params[0], params[1], &r);
+		handled = kvm_xen_hcall_sched_op(vcpu, longmode, params[0],
+						 params[1], &r);
 		break;
 	case __HYPERVISOR_vcpu_op:
 		handled = kvm_xen_hcall_vcpu_op(vcpu, longmode, params[0], params[1],
@@ -1186,6 +1325,17 @@  static inline int max_evtchn_port(struct kvm *kvm)
 		return COMPAT_EVTCHN_2L_NR_CHANNELS;
 }
 
+static void kvm_xen_check_poller(struct kvm_vcpu *vcpu, int port)
+{
+	int poll_evtchn = vcpu->arch.xen.poll_evtchn;
+
+	if ((poll_evtchn == port || poll_evtchn == -1) &&
+	    test_and_clear_bit(kvm_vcpu_get_idx(vcpu), vcpu->kvm->arch.xen.poll_mask)) {
+		kvm_make_request(KVM_REQ_UNBLOCK, vcpu);
+		kvm_vcpu_kick(vcpu);
+	}
+}
+
 /*
  * The return value from this function is propagated to kvm_set_irq() API,
  * so it returns:
@@ -1253,6 +1403,7 @@  int kvm_xen_set_evtchn_fast(struct kvm_xen_evtchn *xe, struct kvm *kvm)
 		rc = 0; /* It was already raised */
 	} else if (test_bit(xe->port, mask_bits)) {
 		rc = -ENOTCONN; /* Masked */
+		kvm_xen_check_poller(vcpu, xe->port);
 	} else {
 		rc = 1; /* Delivered to the bitmap in shared_info. */
 		/* Now switch to the vCPU's vcpu_info to set the index and pending_sel */
@@ -1683,6 +1834,8 @@  static bool kvm_xen_hcall_evtchn_send(struct kvm_vcpu *vcpu, u64 param, u64 *r)
 void kvm_xen_init_vcpu(struct kvm_vcpu *vcpu)
 {
 	vcpu->arch.xen.vcpu_id = vcpu->vcpu_idx;
+	vcpu->arch.xen.poll_evtchn = 0;
+	timer_setup(&vcpu->arch.xen.poll_timer, cancel_evtchn_poll, 0);
 }
 
 void kvm_xen_destroy_vcpu(struct kvm_vcpu *vcpu)
@@ -1696,6 +1849,7 @@  void kvm_xen_destroy_vcpu(struct kvm_vcpu *vcpu)
 				     &vcpu->arch.xen.vcpu_info_cache);
 	kvm_gfn_to_pfn_cache_destroy(vcpu->kvm,
 				     &vcpu->arch.xen.vcpu_time_info_cache);
+	del_timer_sync(&vcpu->arch.xen.poll_timer);
 }
 
 void kvm_xen_init_vm(struct kvm *kvm)
diff --git a/tools/testing/selftests/kvm/x86_64/xen_shinfo_test.c b/tools/testing/selftests/kvm/x86_64/xen_shinfo_test.c
index 865e17146815..376c611443cd 100644
--- a/tools/testing/selftests/kvm/x86_64/xen_shinfo_test.c
+++ b/tools/testing/selftests/kvm/x86_64/xen_shinfo_test.c
@@ -233,6 +233,12 @@  int main(int argc, char *argv[])
 		.flags = KVM_XEN_HVM_CONFIG_INTERCEPT_HCALL,
 		.msr = XEN_HYPERCALL_MSR,
 	};
+
+	/* Let the kernel know that we *will* use it for sending all
+	 * event channels, which lets it intercept SCHEDOP_poll */
+	if (xen_caps & KVM_XEN_HVM_CONFIG_EVTCHN_SEND)
+		hvmc.flags |= KVM_XEN_HVM_CONFIG_EVTCHN_SEND;
+
 	vm_ioctl(vm, KVM_XEN_HVM_CONFIG, &hvmc);
 
 	struct kvm_xen_hvm_attr lm = {