1 files changed, 106 insertions, 70 deletions

diff --git a/drivers/vfio/pci/vfio_pci_intrs.c b/drivers/vfio/pci/vfio_pci_intrs.c
index 869dce5f1..5b0b7fab3 100644
--- a/drivers/vfio/pci/vfio_pci_intrs.c
+++ b/drivers/vfio/pci/vfio_pci_intrs.c
@@ -29,15 +29,22 @@ static void vfio_send_intx_eventfd(void *opaque, void *unused)
 {
 	struct vfio_pci_device *vdev = opaque;
 
-	if (likely(is_intx(vdev) && !vdev->virq_disabled))
-		eventfd_signal(vdev->ctx[0].trigger, 1);
+	if (likely(is_intx(vdev) && !vdev->virq_disabled)) {
+		struct eventfd_ctx *trigger;
+
+		trigger = READ_ONCE(vdev->ctx[0].trigger);
+		if (likely(trigger))
+			eventfd_signal(trigger, 1);
+	}
 }
 
-void vfio_pci_intx_mask(struct vfio_pci_device *vdev)
+static void __vfio_pci_intx_mask(struct vfio_pci_device *vdev)
 {
 	struct pci_dev *pdev = vdev->pdev;
 	unsigned long flags;
 
+	lockdep_assert_held(&vdev->igate);
+
 	spin_lock_irqsave(&vdev->irqlock, flags);
 
 	/*
@@ -65,6 +72,13 @@ void vfio_pci_intx_mask(struct vfio_pci_device *vdev)
 	spin_unlock_irqrestore(&vdev->irqlock, flags);
 }
 
+void vfio_pci_intx_mask(struct vfio_pci_device *vdev)
+{
+	mutex_lock(&vdev->igate);
+	__vfio_pci_intx_mask(vdev);
+	mutex_unlock(&vdev->igate);
+}
+
 /*
  * If this is triggered by an eventfd, we can't call eventfd_signal
  * or else we'll deadlock on the eventfd wait queue.  Return >0 when
@@ -107,12 +121,21 @@ static int vfio_pci_intx_unmask_handler(void *opaque, void *unused)
 	return ret;
 }
 
-void vfio_pci_intx_unmask(struct vfio_pci_device *vdev)
+static void __vfio_pci_intx_unmask(struct vfio_pci_device *vdev)
 {
+	lockdep_assert_held(&vdev->igate);
+
 	if (vfio_pci_intx_unmask_handler(vdev, NULL) > 0)
 		vfio_send_intx_eventfd(vdev, NULL);
 }
 
+void vfio_pci_intx_unmask(struct vfio_pci_device *vdev)
+{
+	mutex_lock(&vdev->igate);
+	__vfio_pci_intx_unmask(vdev);
+	mutex_unlock(&vdev->igate);
+}
+
 static irqreturn_t vfio_intx_handler(int irq, void *dev_id)
 {
 	struct vfio_pci_device *vdev = dev_id;
@@ -139,95 +162,104 @@ static irqreturn_t vfio_intx_handler(int irq, void *dev_id)
 	return ret;
 }
 
-static int vfio_intx_enable(struct vfio_pci_device *vdev)
+static int vfio_intx_enable(struct vfio_pci_device *vdev,
+			    struct eventfd_ctx *trigger)
 {
+	struct pci_dev *pdev = vdev->pdev;
+	unsigned long irqflags;
+	char *name;
+	int ret;
+
 	if (!is_irq_none(vdev))
 		return -EINVAL;
 
-	if (!vdev->pdev->irq)
+	if (!pdev->irq)
 		return -ENODEV;
 
+	name = kasprintf(GFP_KERNEL, "vfio-intx(%s)", pci_name(pdev));
+	if (!name)
+		return -ENOMEM;
+
 	vdev->ctx = kzalloc(sizeof(struct vfio_pci_irq_ctx), GFP_KERNEL);
 	if (!vdev->ctx)
 		return -ENOMEM;
 
 	vdev->num_ctx = 1;
 
+	vdev->ctx[0].name = name;
+	vdev->ctx[0].trigger = trigger;
+
 	/*
-	 * If the virtual interrupt is masked, restore it.  Devices
-	 * supporting DisINTx can be masked at the hardware level
-	 * here, non-PCI-2.3 devices will have to wait until the
-	 * interrupt is enabled.
+	 * Fill the initial masked state based on virq_disabled.  After
+	 * enable, changing the DisINTx bit in vconfig directly changes INTx
+	 * masking.  igate prevents races during setup, once running masked
+	 * is protected via irqlock.
+	 *
+	 * Devices supporting DisINTx also reflect the current mask state in
+	 * the physical DisINTx bit, which is not affected during IRQ setup.
+	 *
+	 * Devices without DisINTx support require an exclusive interrupt.
+	 * IRQ masking is performed at the IRQ chip.  Again, igate protects
+	 * against races during setup and IRQ handlers and irqfds are not
+	 * yet active, therefore masked is stable and can be used to
+	 * conditionally auto-enable the IRQ.
+	 *
+	 * irq_type must be stable while the IRQ handler is registered,
+	 * therefore it must be set before request_irq().
 	 */
 	vdev->ctx[0].masked = vdev->virq_disabled;
-	if (vdev->pci_2_3)
-		pci_intx(vdev->pdev, !vdev->ctx[0].masked);
+	if (vdev->pci_2_3) {
+		pci_intx(pdev, !vdev->ctx[0].masked);
+		irqflags = IRQF_SHARED;
+	} else {
+		irqflags = vdev->ctx[0].masked ? IRQF_NO_AUTOEN : 0;
+	}
 
 	vdev->irq_type = VFIO_PCI_INTX_IRQ_INDEX;
 
+	ret = request_irq(pdev->irq, vfio_intx_handler,
+			  irqflags, vdev->ctx[0].name, vdev);
+	if (ret) {
+		vdev->irq_type = VFIO_PCI_NUM_IRQS;
+		kfree(name);
+		vdev->num_ctx = 0;
+		kfree(vdev->ctx);
+		return ret;
+	}
+
 	return 0;
 }
 
-static int vfio_intx_set_signal(struct vfio_pci_device *vdev, int fd)
+static int vfio_intx_set_signal(struct vfio_pci_device *vdev,
+				struct eventfd_ctx *trigger)
 {
 	struct pci_dev *pdev = vdev->pdev;
-	unsigned long irqflags = IRQF_SHARED;
-	struct eventfd_ctx *trigger;
-	unsigned long flags;
-	int ret;
+	struct eventfd_ctx *old;
 
-	if (vdev->ctx[0].trigger) {
-		free_irq(pdev->irq, vdev);
-		kfree(vdev->ctx[0].name);
-		eventfd_ctx_put(vdev->ctx[0].trigger);
-		vdev->ctx[0].trigger = NULL;
-	}
-
-	if (fd < 0) /* Disable only */
-		return 0;
-
-	vdev->ctx[0].name = kasprintf(GFP_KERNEL, "vfio-intx(%s)",
-				      pci_name(pdev));
-	if (!vdev->ctx[0].name)
-		return -ENOMEM;
-
-	trigger = eventfd_ctx_fdget(fd);
-	if (IS_ERR(trigger)) {
-		kfree(vdev->ctx[0].name);
-		return PTR_ERR(trigger);
-	}
-
-	vdev->ctx[0].trigger = trigger;
+	old = vdev->ctx[0].trigger;
 
-	if (!vdev->pci_2_3)
-		irqflags = 0;
+	WRITE_ONCE(vdev->ctx[0].trigger, trigger);
 
-	ret = request_irq(pdev->irq, vfio_intx_handler,
-			  irqflags, vdev->ctx[0].name, vdev);
-	if (ret) {
-		vdev->ctx[0].trigger = NULL;
-		kfree(vdev->ctx[0].name);
-		eventfd_ctx_put(trigger);
-		return ret;
+	/* Releasing an old ctx requires synchronizing in-flight users */
+	if (old) {
+		synchronize_irq(pdev->irq);
+		vfio_virqfd_flush_thread(&vdev->ctx[0].unmask);
+		eventfd_ctx_put(old);
 	}
 
-	/*
-	 * INTx disable will stick across the new irq setup,
-	 * disable_irq won't.
-	 */
-	spin_lock_irqsave(&vdev->irqlock, flags);
-	if (!vdev->pci_2_3 && vdev->ctx[0].masked)
-		disable_irq_nosync(pdev->irq);
-	spin_unlock_irqrestore(&vdev->irqlock, flags);
-
 	return 0;
 }
 
 static void vfio_intx_disable(struct vfio_pci_device *vdev)
 {
+	struct pci_dev *pdev = vdev->pdev;
+
 	vfio_virqfd_disable(&vdev->ctx[0].unmask);
 	vfio_virqfd_disable(&vdev->ctx[0].mask);
-	vfio_intx_set_signal(vdev, -1);
+	free_irq(pdev->irq, vdev);
+	if (vdev->ctx[0].trigger)
+		eventfd_ctx_put(vdev->ctx[0].trigger);
+	kfree(vdev->ctx[0].name);
 	vdev->irq_type = VFIO_PCI_NUM_IRQS;
 	vdev->num_ctx = 0;
 	kfree(vdev->ctx);
@@ -425,11 +457,11 @@ static int vfio_pci_set_intx_unmask(struct vfio_pci_device *vdev,
 		return -EINVAL;
 
 	if (flags & VFIO_IRQ_SET_DATA_NONE) {
-		vfio_pci_intx_unmask(vdev);
+		__vfio_pci_intx_unmask(vdev);
 	} else if (flags & VFIO_IRQ_SET_DATA_BOOL) {
 		uint8_t unmask = *(uint8_t *)data;
 		if (unmask)
-			vfio_pci_intx_unmask(vdev);
+			__vfio_pci_intx_unmask(vdev);
 	} else if (flags & VFIO_IRQ_SET_DATA_EVENTFD) {
 		int32_t fd = *(int32_t *)data;
 		if (fd >= 0)
@@ -452,11 +484,11 @@ static int vfio_pci_set_intx_mask(struct vfio_pci_device *vdev,
 		return -EINVAL;
 
 	if (flags & VFIO_IRQ_SET_DATA_NONE) {
-		vfio_pci_intx_mask(vdev);
+		__vfio_pci_intx_mask(vdev);
 	} else if (flags & VFIO_IRQ_SET_DATA_BOOL) {
 		uint8_t mask = *(uint8_t *)data;
 		if (mask)
-			vfio_pci_intx_mask(vdev);
+			__vfio_pci_intx_mask(vdev);
 	} else if (flags & VFIO_IRQ_SET_DATA_EVENTFD) {
 		return -ENOTTY; /* XXX implement me */
 	}
@@ -477,19 +509,23 @@ static int vfio_pci_set_intx_trigger(struct vfio_pci_device *vdev,
 		return -EINVAL;
 
 	if (flags & VFIO_IRQ_SET_DATA_EVENTFD) {
+		struct eventfd_ctx *trigger = NULL;
 		int32_t fd = *(int32_t *)data;
 		int ret;
 
-		if (is_intx(vdev))
-			return vfio_intx_set_signal(vdev, fd);
+		if (fd >= 0) {
+			trigger = eventfd_ctx_fdget(fd);
+			if (IS_ERR(trigger))
+				return PTR_ERR(trigger);
+		}
 
-		ret = vfio_intx_enable(vdev);
-		if (ret)
-			return ret;
+		if (is_intx(vdev))
+			ret = vfio_intx_set_signal(vdev, trigger);
+		else
+			ret = vfio_intx_enable(vdev, trigger);
 
-		ret = vfio_intx_set_signal(vdev, fd);
-		if (ret)
-			vfio_intx_disable(vdev);
+		if (ret && trigger)
+			eventfd_ctx_put(trigger);
 
 		return ret;
 	}