Adding upstream version 6.7.12.upstream/6.7.12

Signed-off-by: Daniel Baumann <daniel.baumann@progress-linux.org>

8 files changed, 228 insertions, 117 deletions

diff --git a/drivers/vfio/fsl-mc/vfio_fsl_mc_intr.c b/drivers/vfio/fsl-mc/vfio_fsl_mc_intr.c
index c51229fcc..82b2afa9b 100644
--- a/drivers/vfio/fsl-mc/vfio_fsl_mc_intr.c
+++ b/drivers/vfio/fsl-mc/vfio_fsl_mc_intr.c
@@ -54,7 +54,7 @@ static irqreturn_t vfio_fsl_mc_irq_handler(int irq_num, void *arg)
 {
 	struct vfio_fsl_mc_irq *mc_irq = (struct vfio_fsl_mc_irq *)arg;
 
-	eventfd_signal(mc_irq->trigger, 1);
+	eventfd_signal(mc_irq->trigger);
 	return IRQ_HANDLED;
 }
 
@@ -141,13 +141,14 @@ static int vfio_fsl_mc_set_irq_trigger(struct vfio_fsl_mc_device *vdev,
 	irq = &vdev->mc_irqs[index];
 
 	if (flags & VFIO_IRQ_SET_DATA_NONE) {
-		vfio_fsl_mc_irq_handler(hwirq, irq);
+		if (irq->trigger)
+			eventfd_signal(irq->trigger);
 
 	} else if (flags & VFIO_IRQ_SET_DATA_BOOL) {
 		u8 trigger = *(u8 *)data;
 
-		if (trigger)
-			vfio_fsl_mc_irq_handler(hwirq, irq);
+		if (trigger && irq->trigger)
+			eventfd_signal(irq->trigger);
 	}
 
 	return 0;
diff --git a/drivers/vfio/pci/pds/lm.c b/drivers/vfio/pci/pds/lm.c
index 79fe2e66b..6b94cc0bf 100644
--- a/drivers/vfio/pci/pds/lm.c
+++ b/drivers/vfio/pci/pds/lm.c
@@ -92,8 +92,10 @@ static void pds_vfio_put_lm_file(struct pds_vfio_lm_file *lm_file)
 {
 	mutex_lock(&lm_file->lock);
 
+	lm_file->disabled = true;
 	lm_file->size = 0;
 	lm_file->alloc_size = 0;
+	lm_file->filep->f_pos = 0;
 
 	/* Free scatter list of file pages */
 	sg_free_table(&lm_file->sg_table);
@@ -183,6 +185,12 @@ static ssize_t pds_vfio_save_read(struct file *filp, char __user *buf,
 	pos = &filp->f_pos;
 
 	mutex_lock(&lm_file->lock);
+
+	if (lm_file->disabled) {
+		done = -ENODEV;
+		goto out_unlock;
+	}
+
 	if (*pos > lm_file->size) {
 		done = -EINVAL;
 		goto out_unlock;
@@ -283,6 +291,11 @@ static ssize_t pds_vfio_restore_write(struct file *filp, const char __user *buf,
 
 	mutex_lock(&lm_file->lock);
 
+	if (lm_file->disabled) {
+		done = -ENODEV;
+		goto out_unlock;
+	}
+
 	while (len) {
 		size_t page_offset;
 		struct page *page;
diff --git a/drivers/vfio/pci/pds/lm.h b/drivers/vfio/pci/pds/lm.h
index 13be89319..9511b1afc 100644
--- a/drivers/vfio/pci/pds/lm.h
+++ b/drivers/vfio/pci/pds/lm.h
@@ -27,6 +27,7 @@ struct pds_vfio_lm_file {
 	struct scatterlist *last_offset_sg;	/* Iterator */
 	unsigned int sg_last_entry;
 	unsigned long last_offset;
+	bool disabled;
 };
 
 struct pds_vfio_pci_device;
diff --git a/drivers/vfio/pci/pds/vfio_dev.c b/drivers/vfio/pci/pds/vfio_dev.c
index 4c351c59d..a286ebcc7 100644
--- a/drivers/vfio/pci/pds/vfio_dev.c
+++ b/drivers/vfio/pci/pds/vfio_dev.c
@@ -32,9 +32,9 @@ again:
 	mutex_lock(&pds_vfio->reset_mutex);
 	if (pds_vfio->deferred_reset) {
 		pds_vfio->deferred_reset = false;
+		pds_vfio_put_restore_file(pds_vfio);
+		pds_vfio_put_save_file(pds_vfio);
 		if (pds_vfio->state == VFIO_DEVICE_STATE_ERROR) {
-			pds_vfio_put_restore_file(pds_vfio);
-			pds_vfio_put_save_file(pds_vfio);
 			pds_vfio_dirty_disable(pds_vfio, false);
 		}
 		pds_vfio->state = pds_vfio->deferred_reset_state;
diff --git a/drivers/vfio/pci/vfio_pci_core.c b/drivers/vfio/pci/vfio_pci_core.c
index 1929103ee..1cbc990d4 100644
--- a/drivers/vfio/pci/vfio_pci_core.c
+++ b/drivers/vfio/pci/vfio_pci_core.c
@@ -443,7 +443,7 @@ static int vfio_pci_core_runtime_resume(struct device *dev)
 	 */
 	down_write(&vdev->memory_lock);
 	if (vdev->pm_wake_eventfd_ctx) {
-		eventfd_signal(vdev->pm_wake_eventfd_ctx, 1);
+		eventfd_signal(vdev->pm_wake_eventfd_ctx);
 		__vfio_pci_runtime_pm_exit(vdev);
 	}
 	up_write(&vdev->memory_lock);
@@ -1883,7 +1883,7 @@ void vfio_pci_core_request(struct vfio_device *core_vdev, unsigned int count)
 			pci_notice_ratelimited(pdev,
 				"Relaying device request to user (#%u)\n",
 				count);
-		eventfd_signal(vdev->req_trigger, 1);
+		eventfd_signal(vdev->req_trigger);
 	} else if (count == 0) {
 		pci_warn(pdev,
 			"No device request channel registered, blocked until released by user\n");
@@ -2302,7 +2302,7 @@ pci_ers_result_t vfio_pci_core_aer_err_detected(struct pci_dev *pdev,
 	mutex_lock(&vdev->igate);
 
 	if (vdev->err_trigger)
-		eventfd_signal(vdev->err_trigger, 1);
+		eventfd_signal(vdev->err_trigger);
 
 	mutex_unlock(&vdev->igate);
 
diff --git a/drivers/vfio/pci/vfio_pci_intrs.c b/drivers/vfio/pci/vfio_pci_intrs.c
index cbb4bcbfb..fb5392b74 100644
--- a/drivers/vfio/pci/vfio_pci_intrs.c
+++ b/drivers/vfio/pci/vfio_pci_intrs.c
@@ -90,22 +90,28 @@ static void vfio_send_intx_eventfd(void *opaque, void *unused)
 
 	if (likely(is_intx(vdev) && !vdev->virq_disabled)) {
 		struct vfio_pci_irq_ctx *ctx;
+		struct eventfd_ctx *trigger;
 
 		ctx = vfio_irq_ctx_get(vdev, 0);
 		if (WARN_ON_ONCE(!ctx))
 			return;
-		eventfd_signal(ctx->trigger, 1);
+
+		trigger = READ_ONCE(ctx->trigger);
+		if (likely(trigger))
+			eventfd_signal(trigger);
 	}
 }
 
 /* Returns true if the INTx vfio_pci_irq_ctx.masked value is changed. */
-bool vfio_pci_intx_mask(struct vfio_pci_core_device *vdev)
+static bool __vfio_pci_intx_mask(struct vfio_pci_core_device *vdev)
 {
 	struct pci_dev *pdev = vdev->pdev;
 	struct vfio_pci_irq_ctx *ctx;
 	unsigned long flags;
 	bool masked_changed = false;
 
+	lockdep_assert_held(&vdev->igate);
+
 	spin_lock_irqsave(&vdev->irqlock, flags);
 
 	/*
@@ -143,6 +149,17 @@ out_unlock:
 	return masked_changed;
 }
 
+bool vfio_pci_intx_mask(struct vfio_pci_core_device *vdev)
+{
+	bool mask_changed;
+
+	mutex_lock(&vdev->igate);
+	mask_changed = __vfio_pci_intx_mask(vdev);
+	mutex_unlock(&vdev->igate);
+
+	return mask_changed;
+}
+
 /*
  * 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
@@ -194,12 +211,21 @@ out_unlock:
 	return ret;
 }
 
-void vfio_pci_intx_unmask(struct vfio_pci_core_device *vdev)
+static void __vfio_pci_intx_unmask(struct vfio_pci_core_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_core_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_core_device *vdev = dev_id;
@@ -231,97 +257,100 @@ static irqreturn_t vfio_intx_handler(int irq, void *dev_id)
 	return ret;
 }
 
-static int vfio_intx_enable(struct vfio_pci_core_device *vdev)
+static int vfio_intx_enable(struct vfio_pci_core_device *vdev,
+			    struct eventfd_ctx *trigger)
 {
+	struct pci_dev *pdev = vdev->pdev;
 	struct vfio_pci_irq_ctx *ctx;
+	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_ACCOUNT, "vfio-intx(%s)", pci_name(pdev));
+	if (!name)
+		return -ENOMEM;
+
 	ctx = vfio_irq_ctx_alloc(vdev, 0);
 	if (!ctx)
 		return -ENOMEM;
 
+	ctx->name = name;
+	ctx->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().
 	 */
 	ctx->masked = vdev->virq_disabled;
-	if (vdev->pci_2_3)
-		pci_intx(vdev->pdev, !ctx->masked);
+	if (vdev->pci_2_3) {
+		pci_intx(pdev, !ctx->masked);
+		irqflags = IRQF_SHARED;
+	} else {
+		irqflags = ctx->masked ? IRQF_NO_AUTOEN : 0;
+	}
 
 	vdev->irq_type = VFIO_PCI_INTX_IRQ_INDEX;
 
+	ret = request_irq(pdev->irq, vfio_intx_handler,
+			  irqflags, ctx->name, vdev);
+	if (ret) {
+		vdev->irq_type = VFIO_PCI_NUM_IRQS;
+		kfree(name);
+		vfio_irq_ctx_free(vdev, ctx, 0);
+		return ret;
+	}
+
 	return 0;
 }
 
-static int vfio_intx_set_signal(struct vfio_pci_core_device *vdev, int fd)
+static int vfio_intx_set_signal(struct vfio_pci_core_device *vdev,
+				struct eventfd_ctx *trigger)
 {
 	struct pci_dev *pdev = vdev->pdev;
-	unsigned long irqflags = IRQF_SHARED;
 	struct vfio_pci_irq_ctx *ctx;
-	struct eventfd_ctx *trigger;
-	unsigned long flags;
-	int ret;
+	struct eventfd_ctx *old;
 
 	ctx = vfio_irq_ctx_get(vdev, 0);
 	if (WARN_ON_ONCE(!ctx))
 		return -EINVAL;
 
-	if (ctx->trigger) {
-		free_irq(pdev->irq, vdev);
-		kfree(ctx->name);
-		eventfd_ctx_put(ctx->trigger);
-		ctx->trigger = NULL;
-	}
-
-	if (fd < 0) /* Disable only */
-		return 0;
+	old = ctx->trigger;
 
-	ctx->name = kasprintf(GFP_KERNEL_ACCOUNT, "vfio-intx(%s)",
-			      pci_name(pdev));
-	if (!ctx->name)
-		return -ENOMEM;
+	WRITE_ONCE(ctx->trigger, trigger);
 
-	trigger = eventfd_ctx_fdget(fd);
-	if (IS_ERR(trigger)) {
-		kfree(ctx->name);
-		return PTR_ERR(trigger);
+	/* Releasing an old ctx requires synchronizing in-flight users */
+	if (old) {
+		synchronize_irq(pdev->irq);
+		vfio_virqfd_flush_thread(&ctx->unmask);
+		eventfd_ctx_put(old);
 	}
 
-	ctx->trigger = trigger;
-
-	if (!vdev->pci_2_3)
-		irqflags = 0;
-
-	ret = request_irq(pdev->irq, vfio_intx_handler,
-			  irqflags, ctx->name, vdev);
-	if (ret) {
-		ctx->trigger = NULL;
-		kfree(ctx->name);
-		eventfd_ctx_put(trigger);
-		return ret;
-	}
-
-	/*
-	 * INTx disable will stick across the new irq setup,
-	 * disable_irq won't.
-	 */
-	spin_lock_irqsave(&vdev->irqlock, flags);
-	if (!vdev->pci_2_3 && ctx->masked)
-		disable_irq_nosync(pdev->irq);
-	spin_unlock_irqrestore(&vdev->irqlock, flags);
-
 	return 0;
 }
 
 static void vfio_intx_disable(struct vfio_pci_core_device *vdev)
 {
+	struct pci_dev *pdev = vdev->pdev;
 	struct vfio_pci_irq_ctx *ctx;
 
 	ctx = vfio_irq_ctx_get(vdev, 0);
@@ -329,10 +358,13 @@ static void vfio_intx_disable(struct vfio_pci_core_device *vdev)
 	if (ctx) {
 		vfio_virqfd_disable(&ctx->unmask);
 		vfio_virqfd_disable(&ctx->mask);
+		free_irq(pdev->irq, vdev);
+		if (ctx->trigger)
+			eventfd_ctx_put(ctx->trigger);
+		kfree(ctx->name);
+		vfio_irq_ctx_free(vdev, ctx, 0);
 	}
-	vfio_intx_set_signal(vdev, -1);
 	vdev->irq_type = VFIO_PCI_NUM_IRQS;
-	vfio_irq_ctx_free(vdev, ctx, 0);
 }
 
 /*
@@ -342,7 +374,7 @@ static irqreturn_t vfio_msihandler(int irq, void *arg)
 {
 	struct eventfd_ctx *trigger = arg;
 
-	eventfd_signal(trigger, 1);
+	eventfd_signal(trigger);
 	return IRQ_HANDLED;
 }
 
@@ -560,11 +592,11 @@ static int vfio_pci_set_intx_unmask(struct vfio_pci_core_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) {
 		struct vfio_pci_irq_ctx *ctx = vfio_irq_ctx_get(vdev, 0);
 		int32_t fd = *(int32_t *)data;
@@ -591,11 +623,11 @@ static int vfio_pci_set_intx_mask(struct vfio_pci_core_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 */
 	}
@@ -616,19 +648,23 @@ static int vfio_pci_set_intx_trigger(struct vfio_pci_core_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;
 	}
@@ -689,11 +725,11 @@ static int vfio_pci_set_msi_trigger(struct vfio_pci_core_device *vdev,
 		if (!ctx)
 			continue;
 		if (flags & VFIO_IRQ_SET_DATA_NONE) {
-			eventfd_signal(ctx->trigger, 1);
+			eventfd_signal(ctx->trigger);
 		} else if (flags & VFIO_IRQ_SET_DATA_BOOL) {
 			uint8_t *bools = data;
 			if (bools[i - start])
-				eventfd_signal(ctx->trigger, 1);
+				eventfd_signal(ctx->trigger);
 		}
 	}
 	return 0;
@@ -707,7 +743,7 @@ static int vfio_pci_set_ctx_trigger_single(struct eventfd_ctx **ctx,
 	if (flags & VFIO_IRQ_SET_DATA_NONE) {
 		if (*ctx) {
 			if (count) {
-				eventfd_signal(*ctx, 1);
+				eventfd_signal(*ctx);
 			} else {
 				eventfd_ctx_put(*ctx);
 				*ctx = NULL;
@@ -722,7 +758,7 @@ static int vfio_pci_set_ctx_trigger_single(struct eventfd_ctx **ctx,
 
 		trigger = *(uint8_t *)data;
 		if (trigger && *ctx)
-			eventfd_signal(*ctx, 1);
+			eventfd_signal(*ctx);
 
 		return 0;
 	} else if (flags & VFIO_IRQ_SET_DATA_EVENTFD) {
diff --git a/drivers/vfio/platform/vfio_platform_irq.c b/drivers/vfio/platform/vfio_platform_irq.c
index 665197cae..ef41ecef8 100644
--- a/drivers/vfio/platform/vfio_platform_irq.c
+++ b/drivers/vfio/platform/vfio_platform_irq.c
@@ -136,6 +136,16 @@ static int vfio_platform_set_irq_unmask(struct vfio_platform_device *vdev,
 	return 0;
 }
 
+/*
+ * The trigger eventfd is guaranteed valid in the interrupt path
+ * and protected by the igate mutex when triggered via ioctl.
+ */
+static void vfio_send_eventfd(struct vfio_platform_irq *irq_ctx)
+{
+	if (likely(irq_ctx->trigger))
+		eventfd_signal(irq_ctx->trigger);
+}
+
 static irqreturn_t vfio_automasked_irq_handler(int irq, void *dev_id)
 {
 	struct vfio_platform_irq *irq_ctx = dev_id;
@@ -155,7 +165,7 @@ static irqreturn_t vfio_automasked_irq_handler(int irq, void *dev_id)
 	spin_unlock_irqrestore(&irq_ctx->lock, flags);
 
 	if (ret == IRQ_HANDLED)
-		eventfd_signal(irq_ctx->trigger, 1);
+		vfio_send_eventfd(irq_ctx);
 
 	return ret;
 }
@@ -164,52 +174,40 @@ static irqreturn_t vfio_irq_handler(int irq, void *dev_id)
 {
 	struct vfio_platform_irq *irq_ctx = dev_id;
 
-	eventfd_signal(irq_ctx->trigger, 1);
+	vfio_send_eventfd(irq_ctx);
 
 	return IRQ_HANDLED;
 }
 
 static int vfio_set_trigger(struct vfio_platform_device *vdev, int index,
-			    int fd, irq_handler_t handler)
+			    int fd)
 {
 	struct vfio_platform_irq *irq = &vdev->irqs[index];
 	struct eventfd_ctx *trigger;
-	int ret;
 
 	if (irq->trigger) {
-		irq_clear_status_flags(irq->hwirq, IRQ_NOAUTOEN);
-		free_irq(irq->hwirq, irq);
-		kfree(irq->name);
+		disable_irq(irq->hwirq);
 		eventfd_ctx_put(irq->trigger);
 		irq->trigger = NULL;
 	}
 
 	if (fd < 0) /* Disable only */
 		return 0;
-	irq->name = kasprintf(GFP_KERNEL_ACCOUNT, "vfio-irq[%d](%s)",
-			      irq->hwirq, vdev->name);
-	if (!irq->name)
-		return -ENOMEM;
 
 	trigger = eventfd_ctx_fdget(fd);
-	if (IS_ERR(trigger)) {
-		kfree(irq->name);
+	if (IS_ERR(trigger))
 		return PTR_ERR(trigger);
-	}
 
 	irq->trigger = trigger;
 
-	irq_set_status_flags(irq->hwirq, IRQ_NOAUTOEN);
-	ret = request_irq(irq->hwirq, handler, 0, irq->name, irq);
-	if (ret) {
-		kfree(irq->name);
-		eventfd_ctx_put(trigger);
-		irq->trigger = NULL;
-		return ret;
-	}
-
-	if (!irq->masked)
-		enable_irq(irq->hwirq);
+	/*
+	 * irq->masked effectively provides nested disables within the overall
+	 * enable relative to trigger.  Specifically request_irq() is called
+	 * with NO_AUTOEN, therefore the IRQ is initially disabled.  The user
+	 * may only further disable the IRQ with a MASK operations because
+	 * irq->masked is initially false.
+	 */
+	enable_irq(irq->hwirq);
 
 	return 0;
 }
@@ -228,7 +226,7 @@ static int vfio_platform_set_irq_trigger(struct vfio_platform_device *vdev,
 		handler = vfio_irq_handler;
 
 	if (!count && (flags & VFIO_IRQ_SET_DATA_NONE))
-		return vfio_set_trigger(vdev, index, -1, handler);
+		return vfio_set_trigger(vdev, index, -1);
 
 	if (start != 0 || count != 1)
 		return -EINVAL;
@@ -236,7 +234,7 @@ static int vfio_platform_set_irq_trigger(struct vfio_platform_device *vdev,
 	if (flags & VFIO_IRQ_SET_DATA_EVENTFD) {
 		int32_t fd = *(int32_t *)data;
 
-		return vfio_set_trigger(vdev, index, fd, handler);
+		return vfio_set_trigger(vdev, index, fd);
 	}
 
 	if (flags & VFIO_IRQ_SET_DATA_NONE) {
@@ -260,6 +258,14 @@ int vfio_platform_set_irqs_ioctl(struct vfio_platform_device *vdev,
 		    unsigned start, unsigned count, uint32_t flags,
 		    void *data) = NULL;
 
+	/*
+	 * For compatibility, errors from request_irq() are local to the
+	 * SET_IRQS path and reflected in the name pointer.  This allows,
+	 * for example, polling mode fallback for an exclusive IRQ failure.
+	 */
+	if (IS_ERR(vdev->irqs[index].name))
+		return PTR_ERR(vdev->irqs[index].name);
+
 	switch (flags & VFIO_IRQ_SET_ACTION_TYPE_MASK) {
 	case VFIO_IRQ_SET_ACTION_MASK:
 		func = vfio_platform_set_irq_mask;
@@ -280,7 +286,7 @@ int vfio_platform_set_irqs_ioctl(struct vfio_platform_device *vdev,
 
 int vfio_platform_irq_init(struct vfio_platform_device *vdev)
 {
-	int cnt = 0, i;
+	int cnt = 0, i, ret = 0;
 
 	while (vdev->get_irq(vdev, cnt) >= 0)
 		cnt++;
@@ -292,37 +298,70 @@ int vfio_platform_irq_init(struct vfio_platform_device *vdev)
 
 	for (i = 0; i < cnt; i++) {
 		int hwirq = vdev->get_irq(vdev, i);
+		irq_handler_t handler = vfio_irq_handler;
 
-		if (hwirq < 0)
+		if (hwirq < 0) {
+			ret = -EINVAL;
 			goto err;
+		}
 
 		spin_lock_init(&vdev->irqs[i].lock);
 
 		vdev->irqs[i].flags = VFIO_IRQ_INFO_EVENTFD;
 
-		if (irq_get_trigger_type(hwirq) & IRQ_TYPE_LEVEL_MASK)
+		if (irq_get_trigger_type(hwirq) & IRQ_TYPE_LEVEL_MASK) {
 			vdev->irqs[i].flags |= VFIO_IRQ_INFO_MASKABLE
 						| VFIO_IRQ_INFO_AUTOMASKED;
+			handler = vfio_automasked_irq_handler;
+		}
 
 		vdev->irqs[i].count = 1;
 		vdev->irqs[i].hwirq = hwirq;
 		vdev->irqs[i].masked = false;
+		vdev->irqs[i].name = kasprintf(GFP_KERNEL_ACCOUNT,
+					       "vfio-irq[%d](%s)", hwirq,
+					       vdev->name);
+		if (!vdev->irqs[i].name) {
+			ret = -ENOMEM;
+			goto err;
+		}
+
+		ret = request_irq(hwirq, handler, IRQF_NO_AUTOEN,
+				  vdev->irqs[i].name, &vdev->irqs[i]);
+		if (ret) {
+			kfree(vdev->irqs[i].name);
+			vdev->irqs[i].name = ERR_PTR(ret);
+		}
 	}
 
 	vdev->num_irqs = cnt;
 
 	return 0;
 err:
+	for (--i; i >= 0; i--) {
+		if (!IS_ERR(vdev->irqs[i].name)) {
+			free_irq(vdev->irqs[i].hwirq, &vdev->irqs[i]);
+			kfree(vdev->irqs[i].name);
+		}
+	}
 	kfree(vdev->irqs);
-	return -EINVAL;
+	return ret;
 }
 
 void vfio_platform_irq_cleanup(struct vfio_platform_device *vdev)
 {
 	int i;
 
-	for (i = 0; i < vdev->num_irqs; i++)
-		vfio_set_trigger(vdev, i, -1, NULL);
+	for (i = 0; i < vdev->num_irqs; i++) {
+		vfio_virqfd_disable(&vdev->irqs[i].mask);
+		vfio_virqfd_disable(&vdev->irqs[i].unmask);
+		if (!IS_ERR(vdev->irqs[i].name)) {
+			free_irq(vdev->irqs[i].hwirq, &vdev->irqs[i]);
+			if (vdev->irqs[i].trigger)
+				eventfd_ctx_put(vdev->irqs[i].trigger);
+			kfree(vdev->irqs[i].name);
+		}
+	}
 
 	vdev->num_irqs = 0;
 	kfree(vdev->irqs);
diff --git a/drivers/vfio/virqfd.c b/drivers/vfio/virqfd.c
index 29c564b7a..532269133 100644
--- a/drivers/vfio/virqfd.c
+++ b/drivers/vfio/virqfd.c
@@ -101,6 +101,13 @@ static void virqfd_inject(struct work_struct *work)
 		virqfd->thread(virqfd->opaque, virqfd->data);
 }
 
+static void virqfd_flush_inject(struct work_struct *work)
+{
+	struct virqfd *virqfd = container_of(work, struct virqfd, flush_inject);
+
+	flush_work(&virqfd->inject);
+}
+
 int vfio_virqfd_enable(void *opaque,
 		       int (*handler)(void *, void *),
 		       void (*thread)(void *, void *),
@@ -124,6 +131,7 @@ int vfio_virqfd_enable(void *opaque,
 
 	INIT_WORK(&virqfd->shutdown, virqfd_shutdown);
 	INIT_WORK(&virqfd->inject, virqfd_inject);
+	INIT_WORK(&virqfd->flush_inject, virqfd_flush_inject);
 
 	irqfd = fdget(fd);
 	if (!irqfd.file) {
@@ -213,3 +221,16 @@ void vfio_virqfd_disable(struct virqfd **pvirqfd)
 	flush_workqueue(vfio_irqfd_cleanup_wq);
 }
 EXPORT_SYMBOL_GPL(vfio_virqfd_disable);
+
+void vfio_virqfd_flush_thread(struct virqfd **pvirqfd)
+{
+	unsigned long flags;
+
+	spin_lock_irqsave(&virqfd_lock, flags);
+	if (*pvirqfd && (*pvirqfd)->thread)
+		queue_work(vfio_irqfd_cleanup_wq, &(*pvirqfd)->flush_inject);
+	spin_unlock_irqrestore(&virqfd_lock, flags);
+
+	flush_workqueue(vfio_irqfd_cleanup_wq);
+}
+EXPORT_SYMBOL_GPL(vfio_virqfd_flush_thread);