KVM: x86/xen: Add KVM_IRQ_ROUTING_XEN_EVTCHN and event channel delivery

This adds basic support for delivering 2 level event channels to a guest.

Initially, it only supports delivery via the IRQ routing table, triggered
by an eventfd. In order to do so, it has a kvm_xen_set_evtchn_fast()
function which will use the pre-mapped shared_info page if it already
exists and is still valid, while the slow path through the irqfd_inject
workqueue will remap the shared_info page if necessary.

It sets the bits in the shared_info page but not the vcpu_info; that is
deferred to __kvm_xen_has_interrupt() which raises the vector to the
appropriate vCPU.

Add a 'verbose' mode to xen_shinfo_test while adding test cases for this.

Signed-off-by: David Woodhouse <dwmw@amazon.co.uk>
Message-Id: <20211210163625.2886-5-dwmw2@infradead.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

4 files changed, 280 insertions, 6 deletions

diff --git a/arch/x86/kvm/irq_comm.c b/arch/x86/kvm/irq_comm.c
index 39ad02d6dc63..6e0dab04320e 100644
--- a/arch/x86/kvm/irq_comm.c
+++ b/arch/x86/kvm/irq_comm.c
@@ -24,6 +24,7 @@
 
 #include "hyperv.h"
 #include "x86.h"
+#include "xen.h"
 
 static int kvm_set_pic_irq(struct kvm_kernel_irq_routing_entry *e,
 			   struct kvm *kvm, int irq_source_id, int level,
@@ -175,6 +176,13 @@ int kvm_arch_set_irq_inatomic(struct kvm_kernel_irq_routing_entry *e,
 			return r;
 		break;
 
+#ifdef CONFIG_KVM_XEN
+	case KVM_IRQ_ROUTING_XEN_EVTCHN:
+		if (!level)
+			return -1;
+
+		return kvm_xen_set_evtchn_fast(e, kvm);
+#endif
 	default:
 		break;
 	}
@@ -310,6 +318,10 @@ int kvm_set_routing_entry(struct kvm *kvm,
 		e->hv_sint.vcpu = ue->u.hv_sint.vcpu;
 		e->hv_sint.sint = ue->u.hv_sint.sint;
 		break;
+#ifdef CONFIG_KVM_XEN
+	case KVM_IRQ_ROUTING_XEN_EVTCHN:
+		return kvm_xen_setup_evtchn(kvm, e, ue);
+#endif
 	default:
 		return -EINVAL;
 	}
diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index 42bde45a1bc2..3050601d5d73 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -4188,7 +4188,8 @@ int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
 	case KVM_CAP_XEN_HVM:
 		r = KVM_XEN_HVM_CONFIG_HYPERCALL_MSR |
 		    KVM_XEN_HVM_CONFIG_INTERCEPT_HCALL |
-		    KVM_XEN_HVM_CONFIG_SHARED_INFO;
+		    KVM_XEN_HVM_CONFIG_SHARED_INFO |
+		    KVM_XEN_HVM_CONFIG_EVTCHN_2LEVEL;
 		if (sched_info_on())
 			r |= KVM_XEN_HVM_CONFIG_RUNSTATE;
 		break;
diff --git a/arch/x86/kvm/xen.c b/arch/x86/kvm/xen.c
index da4bf2c6407f..ceddabd1f5c6 100644
--- a/arch/x86/kvm/xen.c
+++ b/arch/x86/kvm/xen.c
@@ -16,6 +16,7 @@
 #include <trace/events/kvm.h>
 #include <xen/interface/xen.h>
 #include <xen/interface/vcpu.h>
+#include <xen/interface/event_channel.h>
 
 #include "trace.h"
 
@@ -195,6 +196,8 @@ void kvm_xen_update_runstate_guest(struct kvm_vcpu *v, int state)
 
 int __kvm_xen_has_interrupt(struct kvm_vcpu *v)
 {
+	unsigned long evtchn_pending_sel = READ_ONCE(v->arch.xen.evtchn_pending_sel);
+	bool atomic = in_atomic() || !task_is_running(current);
 	int err;
 	u8 rc = 0;
 
@@ -204,6 +207,9 @@ int __kvm_xen_has_interrupt(struct kvm_vcpu *v)
 	 */
 	struct gfn_to_hva_cache *ghc = &v->arch.xen.vcpu_info_cache;
 	struct kvm_memslots *slots = kvm_memslots(v->kvm);
+	bool ghc_valid = slots->generation == ghc->generation &&
+		!kvm_is_error_hva(ghc->hva) && ghc->memslot;
+
 	unsigned int offset = offsetof(struct vcpu_info, evtchn_upcall_pending);
 
 	/* No need for compat handling here */
@@ -219,8 +225,7 @@ int __kvm_xen_has_interrupt(struct kvm_vcpu *v)
 	 * cache in kvm_read_guest_offset_cached(), but just uses
 	 * __get_user() instead. And falls back to the slow path.
 	 */
-	if (likely(slots->generation == ghc->generation &&
-		   !kvm_is_error_hva(ghc->hva) && ghc->memslot)) {
+	if (!evtchn_pending_sel && ghc_valid) {
 		/* Fast path */
 		pagefault_disable();
 		err = __get_user(rc, (u8 __user *)ghc->hva + offset);
@@ -239,11 +244,82 @@ int __kvm_xen_has_interrupt(struct kvm_vcpu *v)
 	 * and we'll end up getting called again from a context where we *can*
 	 * fault in the page and wait for it.
 	 */
-	if (in_atomic() || !task_is_running(current))
+	if (atomic)
 		return 1;
 
-	kvm_read_guest_offset_cached(v->kvm, ghc, &rc, offset,
-				     sizeof(rc));
+	if (!ghc_valid) {
+		err = kvm_gfn_to_hva_cache_init(v->kvm, ghc, ghc->gpa, ghc->len);
+		if (err || !ghc->memslot) {
+			/*
+			 * If this failed, userspace has screwed up the
+			 * vcpu_info mapping. No interrupts for you.
+			 */
+			return 0;
+		}
+	}
+
+	/*
+	 * Now we have a valid (protected by srcu) userspace HVA in
+	 * ghc->hva which points to the struct vcpu_info. If there
+	 * are any bits in the in-kernel evtchn_pending_sel then
+	 * we need to write those to the guest vcpu_info and set
+	 * its evtchn_upcall_pending flag. If there aren't any bits
+	 * to add, we only want to *check* evtchn_upcall_pending.
+	 */
+	if (evtchn_pending_sel) {
+		bool long_mode = v->kvm->arch.xen.long_mode;
+
+		if (!user_access_begin((void __user *)ghc->hva, sizeof(struct vcpu_info)))
+			return 0;
+
+		if (IS_ENABLED(CONFIG_64BIT) && long_mode) {
+			struct vcpu_info __user *vi = (void __user *)ghc->hva;
+
+			/* Attempt to set the evtchn_pending_sel bits in the
+			 * guest, and if that succeeds then clear the same
+			 * bits in the in-kernel version. */
+			asm volatile("1:\t" LOCK_PREFIX "orq %0, %1\n"
+				     "\tnotq %0\n"
+				     "\t" LOCK_PREFIX "andq %0, %2\n"
+				     "2:\n"
+				     "\t.section .fixup,\"ax\"\n"
+				     "3:\tjmp\t2b\n"
+				     "\t.previous\n"
+				     _ASM_EXTABLE_UA(1b, 3b)
+				     : "=r" (evtchn_pending_sel),
+				       "+m" (vi->evtchn_pending_sel),
+				       "+m" (v->arch.xen.evtchn_pending_sel)
+				     : "0" (evtchn_pending_sel));
+		} else {
+			struct compat_vcpu_info __user *vi = (void __user *)ghc->hva;
+			u32 evtchn_pending_sel32 = evtchn_pending_sel;
+
+			/* Attempt to set the evtchn_pending_sel bits in the
+			 * guest, and if that succeeds then clear the same
+			 * bits in the in-kernel version. */
+			asm volatile("1:\t" LOCK_PREFIX "orl %0, %1\n"
+				     "\tnotl %0\n"
+				     "\t" LOCK_PREFIX "andl %0, %2\n"
+				     "2:\n"
+				     "\t.section .fixup,\"ax\"\n"
+				     "3:\tjmp\t2b\n"
+				     "\t.previous\n"
+				     _ASM_EXTABLE_UA(1b, 3b)
+				     : "=r" (evtchn_pending_sel32),
+				       "+m" (vi->evtchn_pending_sel),
+				       "+m" (v->arch.xen.evtchn_pending_sel)
+				     : "0" (evtchn_pending_sel32));
+		}
+		rc = 1;
+		unsafe_put_user(rc, (u8 __user *)ghc->hva + offset, err);
+
+	err:
+		user_access_end();
+
+		mark_page_dirty_in_slot(v->kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
+	} else {
+		__get_user(rc, (u8 __user *)ghc->hva + offset);
+	}
 
 	return rc;
 }
@@ -740,3 +816,179 @@ int kvm_xen_hypercall(struct kvm_vcpu *vcpu)
 
 	return 0;
 }
+
+static inline int max_evtchn_port(struct kvm *kvm)
+{
+	if (IS_ENABLED(CONFIG_64BIT) && kvm->arch.xen.long_mode)
+		return EVTCHN_2L_NR_CHANNELS;
+	else
+		return COMPAT_EVTCHN_2L_NR_CHANNELS;
+}
+
+/*
+ * This follows the kvm_set_irq() API, so it returns:
+ *  < 0   Interrupt was ignored (masked or not delivered for other reasons)
+ *  = 0   Interrupt was coalesced (previous irq is still pending)
+ *  > 0   Number of CPUs interrupt was delivered to
+ */
+int kvm_xen_set_evtchn_fast(struct kvm_kernel_irq_routing_entry *e,
+			    struct kvm *kvm)
+{
+	struct gfn_to_pfn_cache *gpc = &kvm->arch.xen.shinfo_cache;
+	struct kvm_vcpu *vcpu;
+	unsigned long *pending_bits, *mask_bits;
+	unsigned long flags;
+	int port_word_bit;
+	bool kick_vcpu = false;
+	int idx;
+	int rc;
+
+	vcpu = kvm_get_vcpu_by_id(kvm, e->xen_evtchn.vcpu);
+	if (!vcpu)
+		return -1;
+
+	if (!vcpu->arch.xen.vcpu_info_set)
+		return -1;
+
+	if (e->xen_evtchn.port >= max_evtchn_port(kvm))
+		return -1;
+
+	rc = -EWOULDBLOCK;
+	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;
+
+	if (IS_ENABLED(CONFIG_64BIT) && kvm->arch.xen.long_mode) {
+		struct shared_info *shinfo = gpc->khva;
+		pending_bits = (unsigned long *)&shinfo->evtchn_pending;
+		mask_bits = (unsigned long *)&shinfo->evtchn_mask;
+		port_word_bit = e->xen_evtchn.port / 64;
+	} else {
+		struct compat_shared_info *shinfo = gpc->khva;
+		pending_bits = (unsigned long *)&shinfo->evtchn_pending;
+		mask_bits = (unsigned long *)&shinfo->evtchn_mask;
+		port_word_bit = e->xen_evtchn.port / 32;
+	}
+
+	/*
+	 * If this port wasn't already set, and if it isn't masked, then
+	 * we try to set the corresponding bit in the in-kernel shadow of
+	 * evtchn_pending_sel for the target vCPU. And if *that* wasn't
+	 * already set, then we kick the vCPU in question to write to the
+	 * *real* evtchn_pending_sel in its own guest vcpu_info struct.
+	 */
+	if (test_and_set_bit(e->xen_evtchn.port, pending_bits)) {
+		rc = 0; /* It was already raised */
+	} else if (test_bit(e->xen_evtchn.port, mask_bits)) {
+		rc = -1; /* Masked */
+	} else {
+		rc = 1; /* Delivered. But was the vCPU waking already? */
+		if (!test_and_set_bit(port_word_bit, &vcpu->arch.xen.evtchn_pending_sel))
+			kick_vcpu = true;
+	}
+
+ out_rcu:
+	srcu_read_unlock(&kvm->srcu, idx);
+	read_unlock_irqrestore(&gpc->lock, flags);
+
+	if (kick_vcpu) {
+		kvm_make_request(KVM_REQ_EVENT, vcpu);
+		kvm_vcpu_kick(vcpu);
+	}
+
+	return rc;
+}
+
+/* This is the version called from kvm_set_irq() as the .set function */
+static int evtchn_set_fn(struct kvm_kernel_irq_routing_entry *e, struct kvm *kvm,
+			 int irq_source_id, int level, bool line_status)
+{
+	bool mm_borrowed = false;
+	int rc;
+
+	if (!level)
+		return -1;
+
+	rc = kvm_xen_set_evtchn_fast(e, kvm);
+	if (rc != -EWOULDBLOCK)
+		return rc;
+
+	if (current->mm != kvm->mm) {
+		/*
+		 * If not on a thread which already belongs to this KVM,
+		 * we'd better be in the irqfd workqueue.
+		 */
+		if (WARN_ON_ONCE(current->mm))
+			return -EINVAL;
+
+		kthread_use_mm(kvm->mm);
+		mm_borrowed = true;
+	}
+
+	/*
+	 * For the irqfd workqueue, using the main kvm->lock mutex is
+	 * fine since this function is invoked from kvm_set_irq() with
+	 * no other lock held, no srcu. In future if it will be called
+	 * directly from a vCPU thread (e.g. on hypercall for an IPI)
+	 * then it may need to switch to using a leaf-node mutex for
+	 * serializing the shared_info mapping.
+	 */
+	mutex_lock(&kvm->lock);
+
+	/*
+	 * It is theoretically possible for the page to be unmapped
+	 * and the MMU notifier to invalidate the shared_info before
+	 * we even get to use it. In that case, this looks like an
+	 * infinite loop. It was tempting to do it via the userspace
+	 * HVA instead... but that just *hides* the fact that it's
+	 * an infinite loop, because if a fault occurs and it waits
+	 * for the page to come back, it can *still* immediately
+	 * fault and have to wait again, repeatedly.
+	 *
+	 * Conversely, the page could also have been reinstated by
+	 * another thread before we even obtain the mutex above, so
+	 * check again *first* before remapping it.
+	 */
+	do {
+		struct gfn_to_pfn_cache *gpc = &kvm->arch.xen.shinfo_cache;
+		int idx;
+
+		rc = kvm_xen_set_evtchn_fast(e, kvm);
+		if (rc != -EWOULDBLOCK)
+			break;
+
+		idx = srcu_read_lock(&kvm->srcu);
+		rc = kvm_gfn_to_pfn_cache_refresh(kvm, gpc, gpc->gpa,
+						  PAGE_SIZE, false);
+		srcu_read_unlock(&kvm->srcu, idx);
+	} while(!rc);
+
+	mutex_unlock(&kvm->lock);
+
+	if (mm_borrowed)
+		kthread_unuse_mm(kvm->mm);
+
+	return rc;
+}
+
+int kvm_xen_setup_evtchn(struct kvm *kvm,
+			 struct kvm_kernel_irq_routing_entry *e,
+			 const struct kvm_irq_routing_entry *ue)
+
+{
+	if (ue->u.xen_evtchn.port >= max_evtchn_port(kvm))
+		return -EINVAL;
+
+	/* We only support 2 level event channels for now */
+	if (ue->u.xen_evtchn.priority != KVM_IRQ_ROUTING_XEN_EVTCHN_PRIO_2LEVEL)
+		return -EINVAL;
+
+	e->xen_evtchn.port = ue->u.xen_evtchn.port;
+	e->xen_evtchn.vcpu = ue->u.xen_evtchn.vcpu;
+	e->xen_evtchn.priority = ue->u.xen_evtchn.priority;
+	e->set = evtchn_set_fn;
+
+	return 0;
+}
diff --git a/arch/x86/kvm/xen.h b/arch/x86/kvm/xen.h
index cc0cf5f37450..adbcc9ed59db 100644
--- a/arch/x86/kvm/xen.h
+++ b/arch/x86/kvm/xen.h
@@ -24,6 +24,12 @@ int kvm_xen_hvm_config(struct kvm *kvm, struct kvm_xen_hvm_config *xhc);
 void kvm_xen_init_vm(struct kvm *kvm);
 void kvm_xen_destroy_vm(struct kvm *kvm);
 
+int kvm_xen_set_evtchn_fast(struct kvm_kernel_irq_routing_entry *e,
+			    struct kvm *kvm);
+int kvm_xen_setup_evtchn(struct kvm *kvm,
+			 struct kvm_kernel_irq_routing_entry *e,
+			 const struct kvm_irq_routing_entry *ue);
+
 static inline bool kvm_xen_msr_enabled(struct kvm *kvm)
 {
 	return static_branch_unlikely(&kvm_xen_enabled.key) &&
@@ -134,6 +140,9 @@ struct compat_shared_info {
 	struct compat_arch_shared_info arch;
 };
 
+#define COMPAT_EVTCHN_2L_NR_CHANNELS (8 *				\
+				      sizeof_field(struct compat_shared_info, \
+						   evtchn_pending))
 struct compat_vcpu_runstate_info {
     int state;
     uint64_t state_entry_time;