Virtualization - Jan 2017 - automatic IRQ affinity for virtio V2

Hi Michael, hi Jason,

This patches applies a few cleanups to the virtio PCI interrupt handling
code, and then converts the virtio PCI code to use the automatic MSI-X
vectors spreading, as well as using the information in virtio-blk
and virtio-scsi to automatically align the blk-mq queues to the MSI-X
vectors.

Changes since V1:
 - dropped the patches already merged for 4.10-rc
 - new patch to remove struct virtio_pci_vq_info
 - improve probe error handling by unwinding step by step
 - new patch to convert virtio-scsi

We don't really need struct virtio_pci_vq_info, as most field in there
are redundant:

 - the vq backpointer is not strictly neede to start with
 - the entry in the vqs list is not needed - the generic virtqueue already
   has list, we only need to check if it has a callback to get the same
   semantics
 - we can use a simple array to look up the MSI-X vec if needed.
 - That simple array now also duoble serves to replace the per_vq_vectors
   flag

Signed-off-by: Christoph Hellwig <hch at lst.de>
---
 drivers/virtio/virtio_pci_common.c | 117 +++++++++++--------------------------
 drivers/virtio/virtio_pci_common.h |  25 +-------
 drivers/virtio/virtio_pci_legacy.c |   6 +-
 drivers/virtio/virtio_pci_modern.c |   6 +-
 4 files changed, 39 insertions(+), 115 deletions(-)

diff --git a/drivers/virtio/virtio_pci_common.c
b/drivers/virtio/virtio_pci_common.c
index 186cbab..a3376731 100644
--- a/drivers/virtio/virtio_pci_common.c
+++ b/drivers/virtio/virtio_pci_common.c
@@ -62,16 +62,13 @@ static irqreturn_t vp_config_changed(int irq, void *opaque)
 static irqreturn_t vp_vring_interrupt(int irq, void *opaque)
 {
 	struct virtio_pci_device *vp_dev = opaque;
-	struct virtio_pci_vq_info *info;
 	irqreturn_t ret = IRQ_NONE;
-	unsigned long flags;
+	struct virtqueue *vq;
 
-	spin_lock_irqsave(&vp_dev->lock, flags);
-	list_for_each_entry(info, &vp_dev->virtqueues, node) {
-		if (vring_interrupt(irq, info->vq) == IRQ_HANDLED)
+	list_for_each_entry(vq, &vp_dev->vdev.vqs, list) {
+		if (vq->callback && vring_interrupt(irq, vq) == IRQ_HANDLED)
 			ret = IRQ_HANDLED;
 	}
-	spin_unlock_irqrestore(&vp_dev->lock, flags);
 
 	return ret;
 }
@@ -167,55 +164,6 @@ static int vp_request_msix_vectors(struct virtio_device
*vdev, int nvectors,
 	return err;
 }
 
-static struct virtqueue *vp_setup_vq(struct virtio_device *vdev, unsigned
index,
-				     void (*callback)(struct virtqueue *vq),
-				     const char *name,
-				     u16 msix_vec)
-{
-	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
-	struct virtio_pci_vq_info *info = kmalloc(sizeof *info, GFP_KERNEL);
-	struct virtqueue *vq;
-	unsigned long flags;
-
-	/* fill out our structure that represents an active queue */
-	if (!info)
-		return ERR_PTR(-ENOMEM);
-
-	vq = vp_dev->setup_vq(vp_dev, info, index, callback, name, msix_vec);
-	if (IS_ERR(vq))
-		goto out_info;
-
-	info->vq = vq;
-	if (callback) {
-		spin_lock_irqsave(&vp_dev->lock, flags);
-		list_add(&info->node, &vp_dev->virtqueues);
-		spin_unlock_irqrestore(&vp_dev->lock, flags);
-	} else {
-		INIT_LIST_HEAD(&info->node);
-	}
-
-	vp_dev->vqs[index] = info;
-	return vq;
-
-out_info:
-	kfree(info);
-	return vq;
-}
-
-static void vp_del_vq(struct virtqueue *vq)
-{
-	struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev);
-	struct virtio_pci_vq_info *info = vp_dev->vqs[vq->index];
-	unsigned long flags;
-
-	spin_lock_irqsave(&vp_dev->lock, flags);
-	list_del(&info->node);
-	spin_unlock_irqrestore(&vp_dev->lock, flags);
-
-	vp_dev->del_vq(info);
-	kfree(info);
-}
-
 /* the config->del_vqs() implementation */
 void vp_del_vqs(struct virtio_device *vdev)
 {
@@ -224,16 +172,15 @@ void vp_del_vqs(struct virtio_device *vdev)
 	int i;
 
 	list_for_each_entry_safe(vq, n, &vdev->vqs, list) {
-		if (vp_dev->per_vq_vectors) {
-			int v = vp_dev->vqs[vq->index]->msix_vector;
+		if (vp_dev->msix_vector_map) {
+			int v = vp_dev->msix_vector_map[vq->index];
 
 			if (v != VIRTIO_MSI_NO_VECTOR)
 				free_irq(pci_irq_vector(vp_dev->pci_dev, v),
 					vq);
 		}
-		vp_del_vq(vq);
+		vp_dev->del_vq(vq);
 	}
-	vp_dev->per_vq_vectors = false;
 
 	if (vp_dev->intx_enabled) {
 		free_irq(vp_dev->pci_dev->irq, vp_dev);
@@ -261,8 +208,8 @@ void vp_del_vqs(struct virtio_device *vdev)
 	vp_dev->msix_names = NULL;
 	kfree(vp_dev->msix_affinity_masks);
 	vp_dev->msix_affinity_masks = NULL;
-	kfree(vp_dev->vqs);
-	vp_dev->vqs = NULL;
+	kfree(vp_dev->msix_vector_map);
+	vp_dev->msix_vector_map = NULL;
 }
 
 static int vp_find_vqs_msix(struct virtio_device *vdev, unsigned nvqs,
@@ -275,10 +222,6 @@ static int vp_find_vqs_msix(struct virtio_device *vdev,
unsigned nvqs,
 	u16 msix_vec;
 	int i, err, nvectors, allocated_vectors;
 
-	vp_dev->vqs = kcalloc(nvqs, sizeof(*vp_dev->vqs), GFP_KERNEL);
-	if (!vp_dev->vqs)
-		return -ENOMEM;
-
 	if (per_vq_vectors) {
 		/* Best option: one for change interrupt, one per vq. */
 		nvectors = 1;
@@ -294,7 +237,13 @@ static int vp_find_vqs_msix(struct virtio_device *vdev,
unsigned nvqs,
 	if (err)
 		goto error_find;
 
-	vp_dev->per_vq_vectors = per_vq_vectors;
+	if (per_vq_vectors) {
+		vp_dev->msix_vector_map = kmalloc_array(nvqs,
+				sizeof(*vp_dev->msix_vector_map), GFP_KERNEL);
+		if (!vp_dev->msix_vector_map)
+			goto error_find;
+	}
+
 	allocated_vectors = vp_dev->msix_used_vectors;
 	for (i = 0; i < nvqs; ++i) {
 		if (!names[i]) {
@@ -304,19 +253,25 @@ static int vp_find_vqs_msix(struct virtio_device *vdev,
unsigned nvqs,
 
 		if (!callbacks[i])
 			msix_vec = VIRTIO_MSI_NO_VECTOR;
-		else if (vp_dev->per_vq_vectors)
+		else if (per_vq_vectors)
 			msix_vec = allocated_vectors++;
 		else
 			msix_vec = VP_MSIX_VQ_VECTOR;
-		vqs[i] = vp_setup_vq(vdev, i, callbacks[i], names[i], msix_vec);
+		vqs[i] = vp_dev->setup_vq(vp_dev, i, callbacks[i], names[i],
+				msix_vec);
 		if (IS_ERR(vqs[i])) {
 			err = PTR_ERR(vqs[i]);
 			goto error_find;
 		}
 
-		if (!vp_dev->per_vq_vectors || msix_vec == VIRTIO_MSI_NO_VECTOR)
+		if (!per_vq_vectors)
 			continue;
 
+		if (msix_vec == VIRTIO_MSI_NO_VECTOR) {
+			vp_dev->msix_vector_map[i] = VIRTIO_MSI_NO_VECTOR;
+			continue;
+		}
+
 		/* allocate per-vq irq if available and necessary */
 		snprintf(vp_dev->msix_names[msix_vec],
 			 sizeof *vp_dev->msix_names,
@@ -326,8 +281,12 @@ static int vp_find_vqs_msix(struct virtio_device *vdev,
unsigned nvqs,
 				  vring_interrupt, 0,
 				  vp_dev->msix_names[msix_vec],
 				  vqs[i]);
-		if (err)
+		if (err) {
+			/* don't free this irq on error */
+			vp_dev->msix_vector_map[i] = VIRTIO_MSI_NO_VECTOR;
 			goto error_find;
+		}
+		vp_dev->msix_vector_map[i] = msix_vec;
 	}
 	return 0;
 
@@ -343,23 +302,18 @@ static int vp_find_vqs_intx(struct virtio_device *vdev,
unsigned nvqs,
 	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 	int i, err;
 
-	vp_dev->vqs = kcalloc(nvqs, sizeof(*vp_dev->vqs), GFP_KERNEL);
-	if (!vp_dev->vqs)
-		return -ENOMEM;
-
 	err = request_irq(vp_dev->pci_dev->irq, vp_interrupt, IRQF_SHARED,
 			dev_name(&vdev->dev), vp_dev);
 	if (err)
 		goto out_del_vqs;
 
 	vp_dev->intx_enabled = 1;
-	vp_dev->per_vq_vectors = false;
 	for (i = 0; i < nvqs; ++i) {
 		if (!names[i]) {
 			vqs[i] = NULL;
 			continue;
 		}
-		vqs[i] = vp_setup_vq(vdev, i, callbacks[i], names[i],
+		vqs[i] = vp_dev->setup_vq(vp_dev, i, callbacks[i], names[i],
 				VIRTIO_MSI_NO_VECTOR);
 		if (IS_ERR(vqs[i])) {
 			err = PTR_ERR(vqs[i]);
@@ -409,16 +363,15 @@ int vp_set_vq_affinity(struct virtqueue *vq, int cpu)
 {
 	struct virtio_device *vdev = vq->vdev;
 	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
-	struct virtio_pci_vq_info *info = vp_dev->vqs[vq->index];
-	struct cpumask *mask;
-	unsigned int irq;
 
 	if (!vq->callback)
 		return -EINVAL;
 
 	if (vp_dev->msix_enabled) {
-		mask = vp_dev->msix_affinity_masks[info->msix_vector];
-		irq = pci_irq_vector(vp_dev->pci_dev, info->msix_vector);
+		int vec = vp_dev->msix_vector_map[vq->index];
+		struct cpumask *mask = vp_dev->msix_affinity_masks[vec];
+		unsigned int irq = pci_irq_vector(vp_dev->pci_dev, vec);
+
 		if (cpu == -1)
 			irq_set_affinity_hint(irq, NULL);
 		else {
@@ -498,8 +451,6 @@ static int virtio_pci_probe(struct pci_dev *pci_dev,
 	vp_dev->vdev.dev.parent = &pci_dev->dev;
 	vp_dev->vdev.dev.release = virtio_pci_release_dev;
 	vp_dev->pci_dev = pci_dev;
-	INIT_LIST_HEAD(&vp_dev->virtqueues);
-	spin_lock_init(&vp_dev->lock);
 
 	/* enable the device */
 	rc = pci_enable_device(pci_dev);
diff --git a/drivers/virtio/virtio_pci_common.h
b/drivers/virtio/virtio_pci_common.h
index b2f6662..2038887 100644
--- a/drivers/virtio/virtio_pci_common.h
+++ b/drivers/virtio/virtio_pci_common.h
@@ -31,17 +31,6 @@
 #include <linux/highmem.h>
 #include <linux/spinlock.h>
 
-struct virtio_pci_vq_info {
-	/* the actual virtqueue */
-	struct virtqueue *vq;
-
-	/* the list node for the virtqueues list */
-	struct list_head node;
-
-	/* MSI-X vector (or none) */
-	unsigned msix_vector;
-};
-
 /* Our device structure */
 struct virtio_pci_device {
 	struct virtio_device vdev;
@@ -75,13 +64,6 @@ struct virtio_pci_device {
 	/* the IO mapping for the PCI config space */
 	void __iomem *ioaddr;
 
-	/* a list of queues so we can dispatch IRQs */
-	spinlock_t lock;
-	struct list_head virtqueues;
-
-	/* array of all queues for house-keeping */
-	struct virtio_pci_vq_info **vqs;
-
 	/* MSI-X support */
 	int msix_enabled;
 	int intx_enabled;
@@ -94,16 +76,15 @@ struct virtio_pci_device {
 	/* Vectors allocated, excluding per-vq vectors if any */
 	unsigned msix_used_vectors;
 
-	/* Whether we have vector per vq */
-	bool per_vq_vectors;
+	/* Map of per-VQ MSI-X vectors, may be NULL */
+	unsigned *msix_vector_map;
 
 	struct virtqueue *(*setup_vq)(struct virtio_pci_device *vp_dev,
-				      struct virtio_pci_vq_info *info,
 				      unsigned idx,
 				      void (*callback)(struct virtqueue *vq),
 				      const char *name,
 				      u16 msix_vec);
-	void (*del_vq)(struct virtio_pci_vq_info *info);
+	void (*del_vq)(struct virtqueue *vq);
 
 	u16 (*config_vector)(struct virtio_pci_device *vp_dev, u16 vector);
 };
diff --git a/drivers/virtio/virtio_pci_legacy.c
b/drivers/virtio/virtio_pci_legacy.c
index 6d9e517..47292da 100644
--- a/drivers/virtio/virtio_pci_legacy.c
+++ b/drivers/virtio/virtio_pci_legacy.c
@@ -112,7 +112,6 @@ static u16 vp_config_vector(struct virtio_pci_device
*vp_dev, u16 vector)
 }
 
 static struct virtqueue *setup_vq(struct virtio_pci_device *vp_dev,
-				  struct virtio_pci_vq_info *info,
 				  unsigned index,
 				  void (*callback)(struct virtqueue *vq),
 				  const char *name,
@@ -130,8 +129,6 @@ static struct virtqueue *setup_vq(struct virtio_pci_device
*vp_dev,
 	if (!num || ioread32(vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN))
 		return ERR_PTR(-ENOENT);
 
-	info->msix_vector = msix_vec;
-
 	/* create the vring */
 	vq = vring_create_virtqueue(index, num,
 				    VIRTIO_PCI_VRING_ALIGN, &vp_dev->vdev,
@@ -162,9 +159,8 @@ static struct virtqueue *setup_vq(struct virtio_pci_device
*vp_dev,
 	return ERR_PTR(err);
 }
 
-static void del_vq(struct virtio_pci_vq_info *info)
+static void del_vq(struct virtqueue *vq)
 {
-	struct virtqueue *vq = info->vq;
 	struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev);
 
 	iowrite16(vq->index, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_SEL);
diff --git a/drivers/virtio/virtio_pci_modern.c
b/drivers/virtio/virtio_pci_modern.c
index 4bf7ab3..00e6fc1 100644
--- a/drivers/virtio/virtio_pci_modern.c
+++ b/drivers/virtio/virtio_pci_modern.c
@@ -293,7 +293,6 @@ static u16 vp_config_vector(struct virtio_pci_device
*vp_dev, u16 vector)
 }
 
 static struct virtqueue *setup_vq(struct virtio_pci_device *vp_dev,
-				  struct virtio_pci_vq_info *info,
 				  unsigned index,
 				  void (*callback)(struct virtqueue *vq),
 				  const char *name,
@@ -323,8 +322,6 @@ static struct virtqueue *setup_vq(struct virtio_pci_device
*vp_dev,
 	/* get offset of notification word for this vq */
 	off = vp_ioread16(&cfg->queue_notify_off);
 
-	info->msix_vector = msix_vec;
-
 	/* create the vring */
 	vq = vring_create_virtqueue(index, num,
 				    SMP_CACHE_BYTES, &vp_dev->vdev,
@@ -409,9 +406,8 @@ static int vp_modern_find_vqs(struct virtio_device *vdev,
unsigned nvqs,
 	return 0;
 }
 
-static void del_vq(struct virtio_pci_vq_info *info)
+static void del_vq(struct virtqueue *vq)
 {
-	struct virtqueue *vq = info->vq;
 	struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev);
 
 	vp_iowrite16(vq->index, &vp_dev->common->queue_select);
-- 
2.1.4

This lets IRQ layer handle dispatching IRQs to separate handlers for the
case where we don't have per-VQ MSI-X vectors, and allows us to greatly
simplify the code based on the assumption that we always have interrupt
vector 0 (legacy INTx or config interrupt for MSI-X) available, and
any other interrupt is request/freed throught the VQ, even if the
actual interrupt line might be shared in some cases.

This allows removing a great deal of variables keeping track of the
interrupt state in struct virtio_pci_device, as we can now simply walk the
list of VQs and deal with per-VQ interrupt handlers there, and only treat
vector 0 special.

Additionally clean up the VQ allocation code to properly unwind on error
instead of having a single global cleanup label, which is error prone,
and in this case also leads to more code.

Signed-off-by: Christoph Hellwig <hch at lst.de>
---
 drivers/virtio/virtio_pci_common.c | 222 ++++++++++++++++---------------------
 drivers/virtio/virtio_pci_common.h |  16 +--
 2 files changed, 100 insertions(+), 138 deletions(-)

diff --git a/drivers/virtio/virtio_pci_common.c
b/drivers/virtio/virtio_pci_common.c
index a3376731..5880e86 100644
--- a/drivers/virtio/virtio_pci_common.c
+++ b/drivers/virtio/virtio_pci_common.c
@@ -33,10 +33,8 @@ void vp_synchronize_vectors(struct virtio_device *vdev)
 	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 	int i;
 
-	if (vp_dev->intx_enabled)
-		synchronize_irq(vp_dev->pci_dev->irq);
-
-	for (i = 0; i < vp_dev->msix_vectors; ++i)
+	synchronize_irq(pci_irq_vector(vp_dev->pci_dev, 0));
+	for (i = 1; i < vp_dev->msix_vectors; i++)
 		synchronize_irq(pci_irq_vector(vp_dev->pci_dev, i));
 }
 
@@ -99,77 +97,10 @@ static irqreturn_t vp_interrupt(int irq, void *opaque)
 	return vp_vring_interrupt(irq, opaque);
 }
 
-static int vp_request_msix_vectors(struct virtio_device *vdev, int nvectors,
-				   bool per_vq_vectors)
-{
-	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
-	const char *name = dev_name(&vp_dev->vdev.dev);
-	unsigned i, v;
-	int err = -ENOMEM;
-
-	vp_dev->msix_vectors = nvectors;
-
-	vp_dev->msix_names = kmalloc(nvectors * sizeof *vp_dev->msix_names,
-				     GFP_KERNEL);
-	if (!vp_dev->msix_names)
-		goto error;
-	vp_dev->msix_affinity_masks
-		= kzalloc(nvectors * sizeof *vp_dev->msix_affinity_masks,
-			  GFP_KERNEL);
-	if (!vp_dev->msix_affinity_masks)
-		goto error;
-	for (i = 0; i < nvectors; ++i)
-		if (!alloc_cpumask_var(&vp_dev->msix_affinity_masks[i],
-					GFP_KERNEL))
-			goto error;
-
-	err = pci_alloc_irq_vectors(vp_dev->pci_dev, nvectors, nvectors,
-			PCI_IRQ_MSIX);
-	if (err < 0)
-		goto error;
-	vp_dev->msix_enabled = 1;
-
-	/* Set the vector used for configuration */
-	v = vp_dev->msix_used_vectors;
-	snprintf(vp_dev->msix_names[v], sizeof *vp_dev->msix_names,
-		 "%s-config", name);
-	err = request_irq(pci_irq_vector(vp_dev->pci_dev, v),
-			  vp_config_changed, 0, vp_dev->msix_names[v],
-			  vp_dev);
-	if (err)
-		goto error;
-	++vp_dev->msix_used_vectors;
-
-	v = vp_dev->config_vector(vp_dev, v);
-	/* Verify we had enough resources to assign the vector */
-	if (v == VIRTIO_MSI_NO_VECTOR) {
-		err = -EBUSY;
-		goto error;
-	}
-
-	if (!per_vq_vectors) {
-		/* Shared vector for all VQs */
-		v = vp_dev->msix_used_vectors;
-		snprintf(vp_dev->msix_names[v], sizeof *vp_dev->msix_names,
-			 "%s-virtqueues", name);
-		err = request_irq(pci_irq_vector(vp_dev->pci_dev, v),
-				  vp_vring_interrupt, 0, vp_dev->msix_names[v],
-				  vp_dev);
-		if (err)
-			goto error;
-		++vp_dev->msix_used_vectors;
-	}
-	return 0;
-error:
-	return err;
-}
-
-/* the config->del_vqs() implementation */
-void vp_del_vqs(struct virtio_device *vdev)
+static void vp_remove_vqs(struct virtio_device *vdev)
 {
 	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 	struct virtqueue *vq, *n;
-	int i;
 
 	list_for_each_entry_safe(vq, n, &vdev->vqs, list) {
 		if (vp_dev->msix_vector_map) {
@@ -181,35 +112,33 @@ void vp_del_vqs(struct virtio_device *vdev)
 		}
 		vp_dev->del_vq(vq);
 	}
+}
 
-	if (vp_dev->intx_enabled) {
-		free_irq(vp_dev->pci_dev->irq, vp_dev);
-		vp_dev->intx_enabled = 0;
-	}
+/* the config->del_vqs() implementation */
+void vp_del_vqs(struct virtio_device *vdev)
+{
+	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
+	int i;
 
-	for (i = 0; i < vp_dev->msix_used_vectors; ++i)
-		free_irq(pci_irq_vector(vp_dev->pci_dev, i), vp_dev);
+	if (WARN_ON_ONCE(list_empty_careful(&vdev->vqs)))
+		return;
 
-	for (i = 0; i < vp_dev->msix_vectors; i++)
-		if (vp_dev->msix_affinity_masks[i])
-			free_cpumask_var(vp_dev->msix_affinity_masks[i]);
+	vp_remove_vqs(vdev);
 
 	if (vp_dev->msix_enabled) {
+		for (i = 0; i < vp_dev->msix_vectors; i++)
+			free_cpumask_var(vp_dev->msix_affinity_masks[i]);
+
 		/* Disable the vector used for configuration */
 		vp_dev->config_vector(vp_dev, VIRTIO_MSI_NO_VECTOR);
 
-		pci_free_irq_vectors(vp_dev->pci_dev);
-		vp_dev->msix_enabled = 0;
+		kfree(vp_dev->msix_affinity_masks);
+		kfree(vp_dev->msix_names);
+		kfree(vp_dev->msix_vector_map);
 	}
 
-	vp_dev->msix_vectors = 0;
-	vp_dev->msix_used_vectors = 0;
-	kfree(vp_dev->msix_names);
-	vp_dev->msix_names = NULL;
-	kfree(vp_dev->msix_affinity_masks);
-	vp_dev->msix_affinity_masks = NULL;
-	kfree(vp_dev->msix_vector_map);
-	vp_dev->msix_vector_map = NULL;
+	free_irq(pci_irq_vector(vp_dev->pci_dev, 0), vp_dev);
+	pci_free_irq_vectors(vp_dev->pci_dev);
 }
 
 static int vp_find_vqs_msix(struct virtio_device *vdev, unsigned nvqs,
@@ -219,8 +148,9 @@ static int vp_find_vqs_msix(struct virtio_device *vdev,
unsigned nvqs,
 			      bool per_vq_vectors)
 {
 	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
+	const char *name = dev_name(&vp_dev->vdev.dev);
+	int i, err = -ENOMEM, allocated_vectors, nvectors;
 	u16 msix_vec;
-	int i, err, nvectors, allocated_vectors;
 
 	if (per_vq_vectors) {
 		/* Best option: one for change interrupt, one per vq. */
@@ -233,65 +163,108 @@ static int vp_find_vqs_msix(struct virtio_device *vdev,
unsigned nvqs,
 		nvectors = 2;
 	}
 
-	err = vp_request_msix_vectors(vdev, nvectors, per_vq_vectors);
+	vp_dev->msix_vectors = nvectors;
+	vp_dev->msix_names = kmalloc_array(nvectors,
+			sizeof(*vp_dev->msix_names), GFP_KERNEL);
+	if (!vp_dev->msix_names)
+		return err;
+
+	vp_dev->msix_affinity_masks = kcalloc(nvectors,
+			sizeof(*vp_dev->msix_affinity_masks), GFP_KERNEL);
+	if (!vp_dev->msix_affinity_masks)
+		goto out_free_msix_names;
+
+	for (i = 0; i < nvectors; ++i) {
+		if (!alloc_cpumask_var(&vp_dev->msix_affinity_masks[i],
+				GFP_KERNEL))
+			goto out_free_msix_affinity_masks;
+	}
+
+	err = pci_alloc_irq_vectors(vp_dev->pci_dev, nvectors, nvectors,
+			PCI_IRQ_MSIX);
+	if (err < 0)
+		goto out_free_msix_affinity_masks;
+
+	/* Set the vector used for configuration */
+	snprintf(vp_dev->msix_names[0], sizeof(*vp_dev->msix_names),
+		 "%s-config", name);
+	err = request_irq(pci_irq_vector(vp_dev->pci_dev, 0), vp_config_changed,
+			0, vp_dev->msix_names[0], vp_dev);
 	if (err)
-		goto error_find;
+		goto out_free_irq_vectors;
 
-	if (per_vq_vectors) {
-		vp_dev->msix_vector_map = kmalloc_array(nvqs,
-				sizeof(*vp_dev->msix_vector_map), GFP_KERNEL);
-		if (!vp_dev->msix_vector_map)
-			goto error_find;
+	/* Verify we had enough resources to assign the vector */
+	if (vp_dev->config_vector(vp_dev, 0) == VIRTIO_MSI_NO_VECTOR) {
+		err = -EBUSY;
+		goto out_free_config_irq;
 	}
 
-	allocated_vectors = vp_dev->msix_used_vectors;
+	vp_dev->msix_vector_map = kmalloc_array(nvqs,
+			sizeof(*vp_dev->msix_vector_map), GFP_KERNEL);
+	if (!vp_dev->msix_vector_map)
+		goto out_disable_config_irq;
+
+	allocated_vectors = 1; /* vector 0 is the config interrupt */
 	for (i = 0; i < nvqs; ++i) {
 		if (!names[i]) {
 			vqs[i] = NULL;
 			continue;
 		}
 
-		if (!callbacks[i])
-			msix_vec = VIRTIO_MSI_NO_VECTOR;
-		else if (per_vq_vectors)
-			msix_vec = allocated_vectors++;
+		if (callbacks[i])
+			msix_vec = allocated_vectors;
 		else
-			msix_vec = VP_MSIX_VQ_VECTOR;
+			msix_vec = VIRTIO_MSI_NO_VECTOR;
+
 		vqs[i] = vp_dev->setup_vq(vp_dev, i, callbacks[i], names[i],
 				msix_vec);
 		if (IS_ERR(vqs[i])) {
 			err = PTR_ERR(vqs[i]);
-			goto error_find;
+			goto out_remove_vqs;
 		}
 
-		if (!per_vq_vectors)
-			continue;
-
 		if (msix_vec == VIRTIO_MSI_NO_VECTOR) {
 			vp_dev->msix_vector_map[i] = VIRTIO_MSI_NO_VECTOR;
 			continue;
 		}
 
-		/* allocate per-vq irq if available and necessary */
-		snprintf(vp_dev->msix_names[msix_vec],
-			 sizeof *vp_dev->msix_names,
-			 "%s-%s",
+		snprintf(vp_dev->msix_names[i + 1],
+			 sizeof(*vp_dev->msix_names), "%s-%s",
 			 dev_name(&vp_dev->vdev.dev), names[i]);
 		err = request_irq(pci_irq_vector(vp_dev->pci_dev, msix_vec),
-				  vring_interrupt, 0,
-				  vp_dev->msix_names[msix_vec],
-				  vqs[i]);
+				  vring_interrupt, IRQF_SHARED,
+				  vp_dev->msix_names[i + 1], vqs[i]);
 		if (err) {
 			/* don't free this irq on error */
 			vp_dev->msix_vector_map[i] = VIRTIO_MSI_NO_VECTOR;
-			goto error_find;
+			goto out_remove_vqs;
 		}
 		vp_dev->msix_vector_map[i] = msix_vec;
+
+		if (per_vq_vectors)
+			allocated_vectors++;
 	}
+
+	vp_dev->msix_enabled = 1;
 	return 0;
 
-error_find:
-	vp_del_vqs(vdev);
+out_remove_vqs:
+	vp_remove_vqs(vdev);
+	kfree(vp_dev->msix_vector_map);
+out_disable_config_irq:
+	vp_dev->config_vector(vp_dev, VIRTIO_MSI_NO_VECTOR);
+out_free_config_irq:
+	free_irq(pci_irq_vector(vp_dev->pci_dev, 0), vp_dev);
+out_free_irq_vectors:
+	pci_free_irq_vectors(vp_dev->pci_dev);
+out_free_msix_affinity_masks:
+	for (i = 0; i < nvectors; i++) {
+		if (vp_dev->msix_affinity_masks[i])
+			free_cpumask_var(vp_dev->msix_affinity_masks[i]);
+	}
+	kfree(vp_dev->msix_affinity_masks);
+out_free_msix_names:
+	kfree(vp_dev->msix_names);
 	return err;
 }
 
@@ -305,9 +278,8 @@ static int vp_find_vqs_intx(struct virtio_device *vdev,
unsigned nvqs,
 	err = request_irq(vp_dev->pci_dev->irq, vp_interrupt, IRQF_SHARED,
 			dev_name(&vdev->dev), vp_dev);
 	if (err)
-		goto out_del_vqs;
+		return err;
 
-	vp_dev->intx_enabled = 1;
 	for (i = 0; i < nvqs; ++i) {
 		if (!names[i]) {
 			vqs[i] = NULL;
@@ -317,13 +289,15 @@ static int vp_find_vqs_intx(struct virtio_device *vdev,
unsigned nvqs,
 				VIRTIO_MSI_NO_VECTOR);
 		if (IS_ERR(vqs[i])) {
 			err = PTR_ERR(vqs[i]);
-			goto out_del_vqs;
+			goto out_remove_vqs;
 		}
 	}
 
 	return 0;
-out_del_vqs:
-	vp_del_vqs(vdev);
+
+out_remove_vqs:
+	vp_remove_vqs(vdev);
+	free_irq(pci_irq_vector(vp_dev->pci_dev, 0), vp_dev);
 	return err;
 }
 
diff --git a/drivers/virtio/virtio_pci_common.h
b/drivers/virtio/virtio_pci_common.h
index 2038887..8559386 100644
--- a/drivers/virtio/virtio_pci_common.h
+++ b/drivers/virtio/virtio_pci_common.h
@@ -66,16 +66,12 @@ struct virtio_pci_device {
 
 	/* MSI-X support */
 	int msix_enabled;
-	int intx_enabled;
 	cpumask_var_t *msix_affinity_masks;
 	/* Name strings for interrupts. This size should be enough,
 	 * and I'm too lazy to allocate each name separately. */
 	char (*msix_names)[256];
-	/* Number of available vectors */
-	unsigned msix_vectors;
-	/* Vectors allocated, excluding per-vq vectors if any */
-	unsigned msix_used_vectors;
-
+	/* Total Number of MSI-X vectors (including per-VQ ones). */
+	int msix_vectors;
 	/* Map of per-VQ MSI-X vectors, may be NULL */
 	unsigned *msix_vector_map;
 
@@ -89,14 +85,6 @@ struct virtio_pci_device {
 	u16 (*config_vector)(struct virtio_pci_device *vp_dev, u16 vector);
 };
 
-/* Constants for MSI-X */
-/* Use first vector for configuration changes, second and the rest for
- * virtqueues Thus, we need at least 2 vectors for MSI. */
-enum {
-	VP_MSIX_CONFIG_VECTOR = 0,
-	VP_MSIX_VQ_VECTOR = 1,
-};
-
 /* Convert a generic virtio device to our structure */
 static struct virtio_pci_device *to_vp_device(struct virtio_device *vdev)
 {
-- 
2.1.4

Signed-off-by: Christoph Hellwig <hch at lst.de>
---
 drivers/virtio/virtio_pci_common.c | 5 ++---
 drivers/virtio/virtio_pci_common.h | 2 --
 drivers/virtio/virtio_pci_legacy.c | 2 +-
 drivers/virtio/virtio_pci_modern.c | 2 +-
 include/uapi/linux/virtio_pci.h    | 2 +-
 5 files changed, 5 insertions(+), 8 deletions(-)

diff --git a/drivers/virtio/virtio_pci_common.c
b/drivers/virtio/virtio_pci_common.c
index 5880e86..9c4ad7d3f 100644
--- a/drivers/virtio/virtio_pci_common.c
+++ b/drivers/virtio/virtio_pci_common.c
@@ -125,7 +125,7 @@ void vp_del_vqs(struct virtio_device *vdev)
 
 	vp_remove_vqs(vdev);
 
-	if (vp_dev->msix_enabled) {
+	if (vp_dev->pci_dev->msix_enabled) {
 		for (i = 0; i < vp_dev->msix_vectors; i++)
 			free_cpumask_var(vp_dev->msix_affinity_masks[i]);
 
@@ -245,7 +245,6 @@ static int vp_find_vqs_msix(struct virtio_device *vdev,
unsigned nvqs,
 			allocated_vectors++;
 	}
 
-	vp_dev->msix_enabled = 1;
 	return 0;
 
 out_remove_vqs:
@@ -341,7 +340,7 @@ int vp_set_vq_affinity(struct virtqueue *vq, int cpu)
 	if (!vq->callback)
 		return -EINVAL;
 
-	if (vp_dev->msix_enabled) {
+	if (vp_dev->pci_dev->msix_enabled) {
 		int vec = vp_dev->msix_vector_map[vq->index];
 		struct cpumask *mask = vp_dev->msix_affinity_masks[vec];
 		unsigned int irq = pci_irq_vector(vp_dev->pci_dev, vec);
diff --git a/drivers/virtio/virtio_pci_common.h
b/drivers/virtio/virtio_pci_common.h
index 8559386..217ca87 100644
--- a/drivers/virtio/virtio_pci_common.h
+++ b/drivers/virtio/virtio_pci_common.h
@@ -64,8 +64,6 @@ struct virtio_pci_device {
 	/* the IO mapping for the PCI config space */
 	void __iomem *ioaddr;
 
-	/* MSI-X support */
-	int msix_enabled;
 	cpumask_var_t *msix_affinity_masks;
 	/* Name strings for interrupts. This size should be enough,
 	 * and I'm too lazy to allocate each name separately. */
diff --git a/drivers/virtio/virtio_pci_legacy.c
b/drivers/virtio/virtio_pci_legacy.c
index 47292da..2ab6aee 100644
--- a/drivers/virtio/virtio_pci_legacy.c
+++ b/drivers/virtio/virtio_pci_legacy.c
@@ -165,7 +165,7 @@ static void del_vq(struct virtqueue *vq)
 
 	iowrite16(vq->index, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_SEL);
 
-	if (vp_dev->msix_enabled) {
+	if (vp_dev->pci_dev->msix_enabled) {
 		iowrite16(VIRTIO_MSI_NO_VECTOR,
 			  vp_dev->ioaddr + VIRTIO_MSI_QUEUE_VECTOR);
 		/* Flush the write out to device */
diff --git a/drivers/virtio/virtio_pci_modern.c
b/drivers/virtio/virtio_pci_modern.c
index 00e6fc1..e5ce310 100644
--- a/drivers/virtio/virtio_pci_modern.c
+++ b/drivers/virtio/virtio_pci_modern.c
@@ -412,7 +412,7 @@ static void del_vq(struct virtqueue *vq)
 
 	vp_iowrite16(vq->index, &vp_dev->common->queue_select);
 
-	if (vp_dev->msix_enabled) {
+	if (vp_dev->pci_dev->msix_enabled) {
 		vp_iowrite16(VIRTIO_MSI_NO_VECTOR,
 			     &vp_dev->common->queue_msix_vector);
 		/* Flush the write out to device */
diff --git a/include/uapi/linux/virtio_pci.h b/include/uapi/linux/virtio_pci.h
index 90007a1..15b4385 100644
--- a/include/uapi/linux/virtio_pci.h
+++ b/include/uapi/linux/virtio_pci.h
@@ -79,7 +79,7 @@
  * configuration space */
 #define VIRTIO_PCI_CONFIG_OFF(msix_enabled)	((msix_enabled) ? 24 : 20)
 /* Deprecated: please use VIRTIO_PCI_CONFIG_OFF instead */
-#define VIRTIO_PCI_CONFIG(dev)	VIRTIO_PCI_CONFIG_OFF((dev)->msix_enabled)
+#define VIRTIO_PCI_CONFIG(dev)
VIRTIO_PCI_CONFIG_OFF((dev)->pci_dev->msix_enabled)
 
 /* Virtio ABI version, this must match exactly */
 #define VIRTIO_PCI_ABI_VERSION		0
-- 
2.1.4

Try to grab the MSI-X vectors early and fall back to the shared one
before doing lots of allocations.

Signed-off-by: Christoph Hellwig <hch at lst.de>
---
 drivers/virtio/virtio_pci_common.c | 58 +++++++++++++++++++-------------------
 1 file changed, 29 insertions(+), 29 deletions(-)

diff --git a/drivers/virtio/virtio_pci_common.c
b/drivers/virtio/virtio_pci_common.c
index 9c4ad7d3f..74ff74c0 100644
--- a/drivers/virtio/virtio_pci_common.c
+++ b/drivers/virtio/virtio_pci_common.c
@@ -142,32 +142,39 @@ void vp_del_vqs(struct virtio_device *vdev)
 }
 
 static int vp_find_vqs_msix(struct virtio_device *vdev, unsigned nvqs,
-			      struct virtqueue *vqs[],
-			      vq_callback_t *callbacks[],
-			      const char * const names[],
-			      bool per_vq_vectors)
+		struct virtqueue *vqs[], vq_callback_t *callbacks[],
+		const char * const names[])
 {
 	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 	const char *name = dev_name(&vp_dev->vdev.dev);
 	int i, err = -ENOMEM, allocated_vectors, nvectors;
+	bool shared = false;
 	u16 msix_vec;
 
-	if (per_vq_vectors) {
-		/* Best option: one for change interrupt, one per vq. */
-		nvectors = 1;
-		for (i = 0; i < nvqs; ++i)
-			if (callbacks[i])
-				++nvectors;
-	} else {
-		/* Second best: one for change, shared for all vqs. */
+	nvectors = 1;
+	for (i = 0; i < nvqs; i++) {
+		if (callbacks[i])
+			nvectors++;
+	}
+
+	/* Try one vector per queue first. */
+	err = pci_alloc_irq_vectors(vp_dev->pci_dev, nvectors, nvectors,
+			PCI_IRQ_MSIX);
+	if (err < 0) {
+		/* Fallback to one vector for config, one shared for queues. */
+		shared = true;
 		nvectors = 2;
+		err = pci_alloc_irq_vectors(vp_dev->pci_dev, nvectors, nvectors,
+				PCI_IRQ_MSIX);
+		if (err < 0)
+			return err;
 	}
 
 	vp_dev->msix_vectors = nvectors;
 	vp_dev->msix_names = kmalloc_array(nvectors,
 			sizeof(*vp_dev->msix_names), GFP_KERNEL);
 	if (!vp_dev->msix_names)
-		return err;
+		goto out_free_irq_vectors;
 
 	vp_dev->msix_affinity_masks = kcalloc(nvectors,
 			sizeof(*vp_dev->msix_affinity_masks), GFP_KERNEL);
@@ -180,18 +187,13 @@ static int vp_find_vqs_msix(struct virtio_device *vdev,
unsigned nvqs,
 			goto out_free_msix_affinity_masks;
 	}
 
-	err = pci_alloc_irq_vectors(vp_dev->pci_dev, nvectors, nvectors,
-			PCI_IRQ_MSIX);
-	if (err < 0)
-		goto out_free_msix_affinity_masks;
-
 	/* Set the vector used for configuration */
 	snprintf(vp_dev->msix_names[0], sizeof(*vp_dev->msix_names),
 		 "%s-config", name);
 	err = request_irq(pci_irq_vector(vp_dev->pci_dev, 0), vp_config_changed,
 			0, vp_dev->msix_names[0], vp_dev);
 	if (err)
-		goto out_free_irq_vectors;
+		goto out_free_msix_affinity_masks;
 
 	/* Verify we had enough resources to assign the vector */
 	if (vp_dev->config_vector(vp_dev, 0) == VIRTIO_MSI_NO_VECTOR) {
@@ -241,7 +243,11 @@ static int vp_find_vqs_msix(struct virtio_device *vdev,
unsigned nvqs,
 		}
 		vp_dev->msix_vector_map[i] = msix_vec;
 
-		if (per_vq_vectors)
+		/*
+		 * Use a different vector for each queue if they are available,
+		 * else share the same vector for all VQs.
+		 */
+		if (!shared)
 			allocated_vectors++;
 	}
 
@@ -254,8 +260,6 @@ static int vp_find_vqs_msix(struct virtio_device *vdev,
unsigned nvqs,
 	vp_dev->config_vector(vp_dev, VIRTIO_MSI_NO_VECTOR);
 out_free_config_irq:
 	free_irq(pci_irq_vector(vp_dev->pci_dev, 0), vp_dev);
-out_free_irq_vectors:
-	pci_free_irq_vectors(vp_dev->pci_dev);
 out_free_msix_affinity_masks:
 	for (i = 0; i < nvectors; i++) {
 		if (vp_dev->msix_affinity_masks[i])
@@ -264,6 +268,8 @@ static int vp_find_vqs_msix(struct virtio_device *vdev,
unsigned nvqs,
 	kfree(vp_dev->msix_affinity_masks);
 out_free_msix_names:
 	kfree(vp_dev->msix_names);
+out_free_irq_vectors:
+	pci_free_irq_vectors(vp_dev->pci_dev);
 	return err;
 }
 
@@ -308,15 +314,9 @@ int vp_find_vqs(struct virtio_device *vdev, unsigned nvqs,
 {
 	int err;
 
-	/* Try MSI-X with one vector per queue. */
-	err = vp_find_vqs_msix(vdev, nvqs, vqs, callbacks, names, true);
-	if (!err)
-		return 0;
-	/* Fallback: MSI-X with one vector for config, one shared for queues. */
-	err = vp_find_vqs_msix(vdev, nvqs, vqs, callbacks, names, false);
+	err = vp_find_vqs_msix(vdev, nvqs, vqs, callbacks, names);
 	if (!err)
 		return 0;
-	/* Finally fall back to regular interrupts. */
 	return vp_find_vqs_intx(vdev, nvqs, vqs, callbacks, names);
 }
 
-- 
2.1.4

Add a struct irq_affinity pointer to the find_vqs methods, which if set
is used to tell the PCI layer to create the MSI-X vectors for our I/O
virtqueues with the proper affinity from the start.  Compared to after
the fact affinity hints this gives us an instantly working setup and
allows to allocate the irq descritors node-local and avoid interconnect
traffic.  Last but not least this will allow blk-mq queues are created
based on the interrupt affinity for storage drivers.

Signed-off-by: Christoph Hellwig <hch at lst.de>
---
 drivers/block/virtio_blk.c                 |  3 ++-
 drivers/char/virtio_console.c              |  2 +-
 drivers/crypto/virtio/virtio_crypto_core.c |  2 +-
 drivers/gpu/drm/virtio/virtgpu_kms.c       |  2 +-
 drivers/misc/mic/vop/vop_main.c            |  2 +-
 drivers/net/caif/caif_virtio.c             |  3 ++-
 drivers/net/virtio_net.c                   |  2 +-
 drivers/remoteproc/remoteproc_virtio.c     |  3 ++-
 drivers/rpmsg/virtio_rpmsg_bus.c           |  2 +-
 drivers/s390/virtio/kvm_virtio.c           |  3 ++-
 drivers/s390/virtio/virtio_ccw.c           |  3 ++-
 drivers/scsi/virtio_scsi.c                 |  3 ++-
 drivers/virtio/virtio_balloon.c            |  3 ++-
 drivers/virtio/virtio_input.c              |  3 ++-
 drivers/virtio/virtio_mmio.c               |  3 ++-
 drivers/virtio/virtio_pci_common.c         | 19 ++++++++++++-------
 drivers/virtio/virtio_pci_common.h         |  5 ++---
 drivers/virtio/virtio_pci_modern.c         |  7 +++----
 include/linux/virtio_config.h              |  9 +++++----
 net/vmw_vsock/virtio_transport.c           |  3 ++-
 20 files changed, 48 insertions(+), 34 deletions(-)

diff --git a/drivers/block/virtio_blk.c b/drivers/block/virtio_blk.c
index 10332c2..c54118b 100644
--- a/drivers/block/virtio_blk.c
+++ b/drivers/block/virtio_blk.c
@@ -411,7 +411,8 @@ static int init_vq(struct virtio_blk *vblk)
 	}
 
 	/* Discover virtqueues and write information to configuration.  */
-	err = vdev->config->find_vqs(vdev, num_vqs, vqs, callbacks, names);
+	err = vdev->config->find_vqs(vdev, num_vqs, vqs, callbacks, names,
+			NULL);
 	if (err)
 		goto out;
 
diff --git a/drivers/char/virtio_console.c b/drivers/char/virtio_console.c
index 17857be..6266c05 100644
--- a/drivers/char/virtio_console.c
+++ b/drivers/char/virtio_console.c
@@ -1939,7 +1939,7 @@ static int init_vqs(struct ports_device *portdev)
 	/* Find the queues. */
 	err = portdev->vdev->config->find_vqs(portdev->vdev, nr_queues,
vqs,
 					      io_callbacks,
-					      (const char **)io_names);
+					      (const char **)io_names, NULL);
 	if (err)
 		goto free;
 
diff --git a/drivers/crypto/virtio/virtio_crypto_core.c
b/drivers/crypto/virtio/virtio_crypto_core.c
index fe70ec8..0aa2f04 100644
--- a/drivers/crypto/virtio/virtio_crypto_core.c
+++ b/drivers/crypto/virtio/virtio_crypto_core.c
@@ -119,7 +119,7 @@ static int virtcrypto_find_vqs(struct virtio_crypto *vi)
 	}
 
 	ret = vi->vdev->config->find_vqs(vi->vdev, total_vqs, vqs,
callbacks,
-					 names);
+					 names, NULL);
 	if (ret)
 		goto err_find;
 
diff --git a/drivers/gpu/drm/virtio/virtgpu_kms.c
b/drivers/gpu/drm/virtio/virtgpu_kms.c
index 1235519..e975fa5 100644
--- a/drivers/gpu/drm/virtio/virtgpu_kms.c
+++ b/drivers/gpu/drm/virtio/virtgpu_kms.c
@@ -172,7 +172,7 @@ int virtio_gpu_driver_load(struct drm_device *dev, unsigned
long flags)
 		 vgdev->has_virgl_3d ? "enabled" : "not available");
 
 	ret = vgdev->vdev->config->find_vqs(vgdev->vdev, 2, vqs,
-					    callbacks, names);
+					    callbacks, names, NULL);
 	if (ret) {
 		DRM_ERROR("failed to find virt queues\n");
 		goto err_vqs;
diff --git a/drivers/misc/mic/vop/vop_main.c b/drivers/misc/mic/vop/vop_main.c
index 1a2b67f3..c2e29d7 100644
--- a/drivers/misc/mic/vop/vop_main.c
+++ b/drivers/misc/mic/vop/vop_main.c
@@ -374,7 +374,7 @@ static struct virtqueue *vop_find_vq(struct virtio_device
*dev,
 static int vop_find_vqs(struct virtio_device *dev, unsigned nvqs,
 			struct virtqueue *vqs[],
 			vq_callback_t *callbacks[],
-			const char * const names[])
+			const char * const names[], struct irq_affinity *desc)
 {
 	struct _vop_vdev *vdev = to_vopvdev(dev);
 	struct vop_device *vpdev = vdev->vpdev;
diff --git a/drivers/net/caif/caif_virtio.c b/drivers/net/caif/caif_virtio.c
index b306210..bc0eb47 100644
--- a/drivers/net/caif/caif_virtio.c
+++ b/drivers/net/caif/caif_virtio.c
@@ -679,7 +679,8 @@ static int cfv_probe(struct virtio_device *vdev)
 		goto err;
 
 	/* Get the TX virtio ring. This is a "guest side vring". */
-	err = vdev->config->find_vqs(vdev, 1, &cfv->vq_tx, &vq_cbs,
&names);
+	err = vdev->config->find_vqs(vdev, 1, &cfv->vq_tx, &vq_cbs,
&names,
+			NULL);
 	if (err)
 		goto err;
 
diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
index 4a10500..1a55926 100644
--- a/drivers/net/virtio_net.c
+++ b/drivers/net/virtio_net.c
@@ -1986,7 +1986,7 @@ static int virtnet_find_vqs(struct virtnet_info *vi)
 	}
 
 	ret = vi->vdev->config->find_vqs(vi->vdev, total_vqs, vqs,
callbacks,
-					 names);
+					 names, NULL);
 	if (ret)
 		goto err_find;
 
diff --git a/drivers/remoteproc/remoteproc_virtio.c
b/drivers/remoteproc/remoteproc_virtio.c
index 364411f..0142cc3 100644
--- a/drivers/remoteproc/remoteproc_virtio.c
+++ b/drivers/remoteproc/remoteproc_virtio.c
@@ -137,7 +137,8 @@ static void rproc_virtio_del_vqs(struct virtio_device *vdev)
 static int rproc_virtio_find_vqs(struct virtio_device *vdev, unsigned int nvqs,
 				 struct virtqueue *vqs[],
 				 vq_callback_t *callbacks[],
-				 const char * const names[])
+				 const char * const names[],
+				 struct irq_affinity *desc)
 {
 	int i, ret;
 
diff --git a/drivers/rpmsg/virtio_rpmsg_bus.c b/drivers/rpmsg/virtio_rpmsg_bus.c
index 3090b0d3..5e66e08 100644
--- a/drivers/rpmsg/virtio_rpmsg_bus.c
+++ b/drivers/rpmsg/virtio_rpmsg_bus.c
@@ -869,7 +869,7 @@ static int rpmsg_probe(struct virtio_device *vdev)
 	init_waitqueue_head(&vrp->sendq);
 
 	/* We expect two virtqueues, rx and tx (and in this order) */
-	err = vdev->config->find_vqs(vdev, 2, vqs, vq_cbs, names);
+	err = vdev->config->find_vqs(vdev, 2, vqs, vq_cbs, names, NULL);
 	if (err)
 		goto free_vrp;
 
diff --git a/drivers/s390/virtio/kvm_virtio.c b/drivers/s390/virtio/kvm_virtio.c
index 5e5c11f..2ce0b3e 100644
--- a/drivers/s390/virtio/kvm_virtio.c
+++ b/drivers/s390/virtio/kvm_virtio.c
@@ -255,7 +255,8 @@ static void kvm_del_vqs(struct virtio_device *vdev)
 static int kvm_find_vqs(struct virtio_device *vdev, unsigned nvqs,
 			struct virtqueue *vqs[],
 			vq_callback_t *callbacks[],
-			const char * const names[])
+			const char * const names[],
+			struct irq_affinity *desc)
 {
 	struct kvm_device *kdev = to_kvmdev(vdev);
 	int i;
diff --git a/drivers/s390/virtio/virtio_ccw.c b/drivers/s390/virtio/virtio_ccw.c
index 070c4da..304d3b3 100644
--- a/drivers/s390/virtio/virtio_ccw.c
+++ b/drivers/s390/virtio/virtio_ccw.c
@@ -628,7 +628,8 @@ static int virtio_ccw_register_adapter_ind(struct
virtio_ccw_device *vcdev,
 static int virtio_ccw_find_vqs(struct virtio_device *vdev, unsigned nvqs,
 			       struct virtqueue *vqs[],
 			       vq_callback_t *callbacks[],
-			       const char * const names[])
+			       const char * const names[],
+			       struct irq_affinity *desc)
 {
 	struct virtio_ccw_device *vcdev = to_vc_device(vdev);
 	unsigned long *indicatorp = NULL;
diff --git a/drivers/scsi/virtio_scsi.c b/drivers/scsi/virtio_scsi.c
index ec91bd0..32a1629 100644
--- a/drivers/scsi/virtio_scsi.c
+++ b/drivers/scsi/virtio_scsi.c
@@ -932,7 +932,8 @@ static int virtscsi_init(struct virtio_device *vdev,
 	}
 
 	/* Discover virtqueues and write information to configuration.  */
-	err = vdev->config->find_vqs(vdev, num_vqs, vqs, callbacks, names);
+	err = vdev->config->find_vqs(vdev, num_vqs, vqs, callbacks, names,
+			NULL);
 	if (err)
 		goto out;
 
diff --git a/drivers/virtio/virtio_balloon.c b/drivers/virtio/virtio_balloon.c
index 181793f..36c9c8f 100644
--- a/drivers/virtio/virtio_balloon.c
+++ b/drivers/virtio/virtio_balloon.c
@@ -413,7 +413,8 @@ static int init_vqs(struct virtio_balloon *vb)
 	 * optionally stat.
 	 */
 	nvqs = virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_STATS_VQ) ? 3 : 2;
-	err = vb->vdev->config->find_vqs(vb->vdev, nvqs, vqs, callbacks,
names);
+	err = vb->vdev->config->find_vqs(vb->vdev, nvqs, vqs, callbacks,
names,
+			NULL);
 	if (err)
 		return err;
 
diff --git a/drivers/virtio/virtio_input.c b/drivers/virtio/virtio_input.c
index 350a2a5..79f1293 100644
--- a/drivers/virtio/virtio_input.c
+++ b/drivers/virtio/virtio_input.c
@@ -173,7 +173,8 @@ static int virtinput_init_vqs(struct virtio_input *vi)
 	static const char * const names[] = { "events", "status"
};
 	int err;
 
-	err = vi->vdev->config->find_vqs(vi->vdev, 2, vqs, cbs, names);
+	err = vi->vdev->config->find_vqs(vi->vdev, 2, vqs, cbs, names,
+			NULL);
 	if (err)
 		return err;
 	vi->evt = vqs[0];
diff --git a/drivers/virtio/virtio_mmio.c b/drivers/virtio/virtio_mmio.c
index d47a2fc..98adb10 100644
--- a/drivers/virtio/virtio_mmio.c
+++ b/drivers/virtio/virtio_mmio.c
@@ -445,7 +445,8 @@ static struct virtqueue *vm_setup_vq(struct virtio_device
*vdev, unsigned index,
 static int vm_find_vqs(struct virtio_device *vdev, unsigned nvqs,
 		       struct virtqueue *vqs[],
 		       vq_callback_t *callbacks[],
-		       const char * const names[])
+		       const char * const names[],
+		       struct irq_affinity *desc)
 {
 	struct virtio_mmio_device *vm_dev = to_virtio_mmio_device(vdev);
 	unsigned int irq = platform_get_irq(vm_dev->pdev, 0);
diff --git a/drivers/virtio/virtio_pci_common.c
b/drivers/virtio/virtio_pci_common.c
index 74ff74c0..c244212 100644
--- a/drivers/virtio/virtio_pci_common.c
+++ b/drivers/virtio/virtio_pci_common.c
@@ -143,14 +143,20 @@ void vp_del_vqs(struct virtio_device *vdev)
 
 static int vp_find_vqs_msix(struct virtio_device *vdev, unsigned nvqs,
 		struct virtqueue *vqs[], vq_callback_t *callbacks[],
-		const char * const names[])
+		const char * const names[], struct irq_affinity *desc)
 {
 	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 	const char *name = dev_name(&vp_dev->vdev.dev);
 	int i, err = -ENOMEM, allocated_vectors, nvectors;
+	unsigned flags = PCI_IRQ_MSIX;
 	bool shared = false;
 	u16 msix_vec;
 
+	if (desc) {
+		flags |= PCI_IRQ_AFFINITY;
+		desc->pre_vectors++; /* virtio config vector */
+	}
+
 	nvectors = 1;
 	for (i = 0; i < nvqs; i++) {
 		if (callbacks[i])
@@ -158,8 +164,8 @@ static int vp_find_vqs_msix(struct virtio_device *vdev,
unsigned nvqs,
 	}
 
 	/* Try one vector per queue first. */
-	err = pci_alloc_irq_vectors(vp_dev->pci_dev, nvectors, nvectors,
-			PCI_IRQ_MSIX);
+	err = pci_alloc_irq_vectors_affinity(vp_dev->pci_dev, nvectors,
+			nvectors, flags, desc);
 	if (err < 0) {
 		/* Fallback to one vector for config, one shared for queues. */
 		shared = true;
@@ -308,13 +314,12 @@ static int vp_find_vqs_intx(struct virtio_device *vdev,
unsigned nvqs,
 
 /* the config->find_vqs() implementation */
 int vp_find_vqs(struct virtio_device *vdev, unsigned nvqs,
-		struct virtqueue *vqs[],
-		vq_callback_t *callbacks[],
-		const char * const names[])
+		struct virtqueue *vqs[], vq_callback_t *callbacks[],
+		const char * const names[], struct irq_affinity *desc)
 {
 	int err;
 
-	err = vp_find_vqs_msix(vdev, nvqs, vqs, callbacks, names);
+	err = vp_find_vqs_msix(vdev, nvqs, vqs, callbacks, names, desc);
 	if (!err)
 		return 0;
 	return vp_find_vqs_intx(vdev, nvqs, vqs, callbacks, names);
diff --git a/drivers/virtio/virtio_pci_common.h
b/drivers/virtio/virtio_pci_common.h
index 217ca87..a6ad9ec 100644
--- a/drivers/virtio/virtio_pci_common.h
+++ b/drivers/virtio/virtio_pci_common.h
@@ -97,9 +97,8 @@ bool vp_notify(struct virtqueue *vq);
 void vp_del_vqs(struct virtio_device *vdev);
 /* the config->find_vqs() implementation */
 int vp_find_vqs(struct virtio_device *vdev, unsigned nvqs,
-		       struct virtqueue *vqs[],
-		       vq_callback_t *callbacks[],
-		       const char * const names[]);
+		struct virtqueue *vqs[], vq_callback_t *callbacks[],
+		const char * const names[], struct irq_affinity *desc);
 const char *vp_bus_name(struct virtio_device *vdev);
 
 /* Setup the affinity for a virtqueue:
diff --git a/drivers/virtio/virtio_pci_modern.c
b/drivers/virtio/virtio_pci_modern.c
index e5ce310..a7a0981 100644
--- a/drivers/virtio/virtio_pci_modern.c
+++ b/drivers/virtio/virtio_pci_modern.c
@@ -384,13 +384,12 @@ static struct virtqueue *setup_vq(struct virtio_pci_device
*vp_dev,
 }
 
 static int vp_modern_find_vqs(struct virtio_device *vdev, unsigned nvqs,
-			      struct virtqueue *vqs[],
-			      vq_callback_t *callbacks[],
-			      const char * const names[])
+		struct virtqueue *vqs[], vq_callback_t *callbacks[],
+		const char * const names[], struct irq_affinity *desc)
 {
 	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 	struct virtqueue *vq;
-	int rc = vp_find_vqs(vdev, nvqs, vqs, callbacks, names);
+	int rc = vp_find_vqs(vdev, nvqs, vqs, callbacks, names, desc);
 
 	if (rc)
 		return rc;
diff --git a/include/linux/virtio_config.h b/include/linux/virtio_config.h
index 26c155b..2ebe506 100644
--- a/include/linux/virtio_config.h
+++ b/include/linux/virtio_config.h
@@ -7,6 +7,8 @@
 #include <linux/virtio_byteorder.h>
 #include <uapi/linux/virtio_config.h>
 
+struct irq_affinity;
+
 /**
  * virtio_config_ops - operations for configuring a virtio device
  * @get: read the value of a configuration field
@@ -68,9 +70,8 @@ struct virtio_config_ops {
 	void (*set_status)(struct virtio_device *vdev, u8 status);
 	void (*reset)(struct virtio_device *vdev);
 	int (*find_vqs)(struct virtio_device *, unsigned nvqs,
-			struct virtqueue *vqs[],
-			vq_callback_t *callbacks[],
-			const char * const names[]);
+			struct virtqueue *vqs[], vq_callback_t *callbacks[],
+			const char * const names[], struct irq_affinity *desc);
 	void (*del_vqs)(struct virtio_device *);
 	u64 (*get_features)(struct virtio_device *vdev);
 	int (*finalize_features)(struct virtio_device *vdev);
@@ -169,7 +170,7 @@ struct virtqueue *virtio_find_single_vq(struct virtio_device
*vdev,
 	vq_callback_t *callbacks[] = { c };
 	const char *names[] = { n };
 	struct virtqueue *vq;
-	int err = vdev->config->find_vqs(vdev, 1, &vq, callbacks, names);
+	int err = vdev->config->find_vqs(vdev, 1, &vq, callbacks, names,
NULL);
 	if (err < 0)
 		return ERR_PTR(err);
 	return vq;
diff --git a/net/vmw_vsock/virtio_transport.c b/net/vmw_vsock/virtio_transport.c
index 6788264..9d24c0e 100644
--- a/net/vmw_vsock/virtio_transport.c
+++ b/net/vmw_vsock/virtio_transport.c
@@ -532,7 +532,8 @@ static int virtio_vsock_probe(struct virtio_device *vdev)
 	vsock->vdev = vdev;
 
 	ret = vsock->vdev->config->find_vqs(vsock->vdev, VSOCK_VQ_MAX,
-					    vsock->vqs, callbacks, names);
+					    vsock->vqs, callbacks, names,
+					    NULL);
 	if (ret < 0)
 		goto out;
 
-- 
2.1.4

This basically passed up the pci_irq_get_affinity information through
virtio through an optional get_vq_affinity method.  It is only implemented
by the PCI backend for now, and only when we use per-virtqueue IRQs.

Signed-off-by: Christoph Hellwig <hch at lst.de>
---
 drivers/virtio/virtio_pci_common.c | 11 +++++++++++
 drivers/virtio/virtio_pci_common.h |  2 ++
 drivers/virtio/virtio_pci_legacy.c |  1 +
 drivers/virtio/virtio_pci_modern.c |  2 ++
 include/linux/virtio_config.h      |  3 +++
 5 files changed, 19 insertions(+)

diff --git a/drivers/virtio/virtio_pci_common.c
b/drivers/virtio/virtio_pci_common.c
index c244212..31ba259 100644
--- a/drivers/virtio/virtio_pci_common.c
+++ b/drivers/virtio/virtio_pci_common.c
@@ -361,6 +361,17 @@ int vp_set_vq_affinity(struct virtqueue *vq, int cpu)
 	return 0;
 }
 
+const struct cpumask *vp_get_vq_affinity(struct virtio_device *vdev, int index)
+{
+	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
+	unsigned int *map = vp_dev->msix_vector_map;
+
+	if (!map || map[index] == VIRTIO_MSI_NO_VECTOR)
+		return NULL;
+
+	return pci_irq_get_affinity(vp_dev->pci_dev, map[index]);
+}
+
 #ifdef CONFIG_PM_SLEEP
 static int virtio_pci_freeze(struct device *dev)
 {
diff --git a/drivers/virtio/virtio_pci_common.h
b/drivers/virtio/virtio_pci_common.h
index a6ad9ec..ac8c9d7 100644
--- a/drivers/virtio/virtio_pci_common.h
+++ b/drivers/virtio/virtio_pci_common.h
@@ -108,6 +108,8 @@ const char *vp_bus_name(struct virtio_device *vdev);
  */
 int vp_set_vq_affinity(struct virtqueue *vq, int cpu);
 
+const struct cpumask *vp_get_vq_affinity(struct virtio_device *vdev, int
index);
+
 #if IS_ENABLED(CONFIG_VIRTIO_PCI_LEGACY)
 int virtio_pci_legacy_probe(struct virtio_pci_device *);
 void virtio_pci_legacy_remove(struct virtio_pci_device *);
diff --git a/drivers/virtio/virtio_pci_legacy.c
b/drivers/virtio/virtio_pci_legacy.c
index 2ab6aee..f7362c5 100644
--- a/drivers/virtio/virtio_pci_legacy.c
+++ b/drivers/virtio/virtio_pci_legacy.c
@@ -190,6 +190,7 @@ static const struct virtio_config_ops virtio_pci_config_ops
= {
 	.finalize_features = vp_finalize_features,
 	.bus_name	= vp_bus_name,
 	.set_vq_affinity = vp_set_vq_affinity,
+	.get_vq_affinity = vp_get_vq_affinity,
 };
 
 /* the PCI probing function */
diff --git a/drivers/virtio/virtio_pci_modern.c
b/drivers/virtio/virtio_pci_modern.c
index a7a0981..7bc3004 100644
--- a/drivers/virtio/virtio_pci_modern.c
+++ b/drivers/virtio/virtio_pci_modern.c
@@ -437,6 +437,7 @@ static const struct virtio_config_ops
virtio_pci_config_nodev_ops = {
 	.finalize_features = vp_finalize_features,
 	.bus_name	= vp_bus_name,
 	.set_vq_affinity = vp_set_vq_affinity,
+	.get_vq_affinity = vp_get_vq_affinity,
 };
 
 static const struct virtio_config_ops virtio_pci_config_ops = {
@@ -452,6 +453,7 @@ static const struct virtio_config_ops virtio_pci_config_ops
= {
 	.finalize_features = vp_finalize_features,
 	.bus_name	= vp_bus_name,
 	.set_vq_affinity = vp_set_vq_affinity,
+	.get_vq_affinity = vp_get_vq_affinity,
 };
 
 /**
diff --git a/include/linux/virtio_config.h b/include/linux/virtio_config.h
index 2ebe506..8355bab 100644
--- a/include/linux/virtio_config.h
+++ b/include/linux/virtio_config.h
@@ -58,6 +58,7 @@ struct irq_affinity;
  *      This returns a pointer to the bus name a la pci_name from which
  *      the caller can then copy.
  * @set_vq_affinity: set the affinity for a virtqueue.
+ * @get_vq_affinity: get the affinity for a virtqueue (optional).
  */
 typedef void vq_callback_t(struct virtqueue *);
 struct virtio_config_ops {
@@ -77,6 +78,8 @@ struct virtio_config_ops {
 	int (*finalize_features)(struct virtio_device *vdev);
 	const char *(*bus_name)(struct virtio_device *vdev);
 	int (*set_vq_affinity)(struct virtqueue *vq, int cpu);
+	const struct cpumask *(*get_vq_affinity)(struct virtio_device *vdev,
+			int index);
 };
 
 /* If driver didn't advertise the feature, it will never appear. */
-- 
2.1.4

Similar to the PCI version, just calling into virtio instead.

Signed-off-by: Christoph Hellwig <hch at lst.de>
---
 block/Kconfig                 |  5 ++++
 block/Makefile                |  1 +
 block/blk-mq-virtio.c         | 54 +++++++++++++++++++++++++++++++++++++++++++
 include/linux/blk-mq-virtio.h | 10 ++++++++
 4 files changed, 70 insertions(+)
 create mode 100644 block/blk-mq-virtio.c
 create mode 100644 include/linux/blk-mq-virtio.h

diff --git a/block/Kconfig b/block/Kconfig
index 8bf114a..3523b4f 100644
--- a/block/Kconfig
+++ b/block/Kconfig
@@ -165,4 +165,9 @@ config BLK_MQ_PCI
 	depends on BLOCK && PCI
 	default y
 
+config BLK_MQ_VIRTIO
+	bool
+	depends on BLOCK && VIRTIO
+	default y
+
 source block/Kconfig.iosched
diff --git a/block/Makefile b/block/Makefile
index a827f98..6069194 100644
--- a/block/Makefile
+++ b/block/Makefile
@@ -23,5 +23,6 @@ obj-$(CONFIG_BLOCK_COMPAT)	+= compat_ioctl.o
 obj-$(CONFIG_BLK_CMDLINE_PARSER)	+= cmdline-parser.o
 obj-$(CONFIG_BLK_DEV_INTEGRITY) += bio-integrity.o blk-integrity.o t10-pi.o
 obj-$(CONFIG_BLK_MQ_PCI)	+= blk-mq-pci.o
+obj-$(CONFIG_BLK_MQ_VIRTIO)	+= blk-mq-virtio.o
 obj-$(CONFIG_BLK_DEV_ZONED)	+= blk-zoned.o
 obj-$(CONFIG_BLK_WBT)		+= blk-wbt.o
diff --git a/block/blk-mq-virtio.c b/block/blk-mq-virtio.c
new file mode 100644
index 0000000..c3afbca
--- /dev/null
+++ b/block/blk-mq-virtio.c
@@ -0,0 +1,54 @@
+/*
+ * Copyright (c) 2016 Christoph Hellwig.
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms and conditions of the GNU General Public License,
+ * version 2, as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
+ * more details.
+ */
+#include <linux/device.h>
+#include <linux/blk-mq.h>
+#include <linux/blk-mq-virtio.h>
+#include <linux/virtio_config.h>
+#include <linux/module.h>
+#include "blk-mq.h"
+
+/**
+ * blk_mq_virtio_map_queues - provide a default queue mapping for virtio device
+ * @set:	tagset to provide the mapping for
+ * @vdev:	virtio device associated with @set.
+ * @first_vec:	first interrupt vectors to use for queues (usually 0)
+ *
+ * This function assumes the virtio device @vdev has at least as many available
+ * interrupt vetors as @set has queues.  It will then queuery the vector
+ * corresponding to each queue for it's affinity mask and built queue
mapping
+ * that maps a queue to the CPUs that have irq affinity for the corresponding
+ * vector.
+ */
+int blk_mq_virtio_map_queues(struct blk_mq_tag_set *set,
+		struct virtio_device *vdev, int first_vec)
+{
+	const struct cpumask *mask;
+	unsigned int queue, cpu;
+
+	if (!vdev->config->get_vq_affinity)
+		goto fallback;
+
+	for (queue = 0; queue < set->nr_hw_queues; queue++) {
+		mask = vdev->config->get_vq_affinity(vdev, first_vec + queue);
+		if (!mask)
+			goto fallback;
+
+		for_each_cpu(cpu, mask)
+			set->mq_map[cpu] = queue;
+	}
+
+	return 0;
+fallback:
+	return blk_mq_map_queues(set);
+}
+EXPORT_SYMBOL_GPL(blk_mq_virtio_map_queues);
diff --git a/include/linux/blk-mq-virtio.h b/include/linux/blk-mq-virtio.h
new file mode 100644
index 0000000..b1ef6e1
--- /dev/null
+++ b/include/linux/blk-mq-virtio.h
@@ -0,0 +1,10 @@
+#ifndef _LINUX_BLK_MQ_VIRTIO_H
+#define _LINUX_BLK_MQ_VIRTIO_H
+
+struct blk_mq_tag_set;
+struct virtio_device;
+
+int blk_mq_virtio_map_queues(struct blk_mq_tag_set *set,
+		struct virtio_device *vdev, int first_vec);
+
+#endif /* _LINUX_BLK_MQ_VIRTIO_H */
-- 
2.1.4

Use automatic IRQ affinity assignment in the virtio layer if available,
and build the blk-mq queues based on it.

Signed-off-by: Christoph Hellwig <hch at lst.de>
---
 drivers/block/virtio_blk.c | 13 ++++++++++++-
 1 file changed, 12 insertions(+), 1 deletion(-)

diff --git a/drivers/block/virtio_blk.c b/drivers/block/virtio_blk.c
index c54118b..1028dfe 100644
--- a/drivers/block/virtio_blk.c
+++ b/drivers/block/virtio_blk.c
@@ -5,6 +5,7 @@
 #include <linux/hdreg.h>
 #include <linux/module.h>
 #include <linux/mutex.h>
+#include <linux/interrupt.h>
 #include <linux/virtio.h>
 #include <linux/virtio_blk.h>
 #include <linux/scatterlist.h>
@@ -12,6 +13,7 @@
 #include <scsi/scsi_cmnd.h>
 #include <linux/idr.h>
 #include <linux/blk-mq.h>
+#include <linux/blk-mq-virtio.h>
 #include <linux/numa.h>
 
 #define PART_BITS 4
@@ -385,6 +387,7 @@ static int init_vq(struct virtio_blk *vblk)
 	struct virtqueue **vqs;
 	unsigned short num_vqs;
 	struct virtio_device *vdev = vblk->vdev;
+	struct irq_affinity desc = { 0, };
 
 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_MQ,
 				   struct virtio_blk_config, num_queues,
@@ -412,7 +415,7 @@ static int init_vq(struct virtio_blk *vblk)
 
 	/* Discover virtqueues and write information to configuration.  */
 	err = vdev->config->find_vqs(vdev, num_vqs, vqs, callbacks, names,
-			NULL);
+			&desc);
 	if (err)
 		goto out;
 
@@ -543,10 +546,18 @@ static int virtblk_init_request(void *data, struct request
*rq,
 	return 0;
 }
 
+static int virtblk_map_queues(struct blk_mq_tag_set *set)
+{
+	struct virtio_blk *vblk = set->driver_data;
+
+	return blk_mq_virtio_map_queues(set, vblk->vdev, 0);
+}
+
 static struct blk_mq_ops virtio_mq_ops = {
 	.queue_rq	= virtio_queue_rq,
 	.complete	= virtblk_request_done,
 	.init_request	= virtblk_init_request,
+	.map_queues	= virtblk_map_queues,
 };
 
 static unsigned int virtblk_queue_depth;
-- 
2.1.4

Use automatic IRQ affinity assignment in the virtio layer if available,
and build the blk-mq queues based on it.

Signed-off-by: Christoph Hellwig <hch at lst.de>
---
 drivers/scsi/virtio_scsi.c | 126 +++++----------------------------------------
 include/linux/cpuhotplug.h |   1 -
 2 files changed, 12 insertions(+), 115 deletions(-)

diff --git a/drivers/scsi/virtio_scsi.c b/drivers/scsi/virtio_scsi.c
index 32a1629..5582c8b 100644
--- a/drivers/scsi/virtio_scsi.c
+++ b/drivers/scsi/virtio_scsi.c
@@ -18,6 +18,7 @@
 #include <linux/module.h>
 #include <linux/slab.h>
 #include <linux/mempool.h>
+#include <linux/interrupt.h>
 #include <linux/virtio.h>
 #include <linux/virtio_ids.h>
 #include <linux/virtio_config.h>
@@ -29,6 +30,7 @@
 #include <scsi/scsi_cmnd.h>
 #include <scsi/scsi_tcq.h>
 #include <linux/seqlock.h>
+#include <linux/blk-mq-virtio.h>
 
 #define VIRTIO_SCSI_MEMPOOL_SZ 64
 #define VIRTIO_SCSI_EVENT_LEN 8
@@ -108,7 +110,6 @@ struct virtio_scsi {
 	bool affinity_hint_set;
 
 	struct hlist_node node;
-	struct hlist_node node_dead;
 
 	/* Protected by event_vq lock */
 	bool stop_events;
@@ -118,7 +119,6 @@ struct virtio_scsi {
 	struct virtio_scsi_vq req_vqs[];
 };
 
-static enum cpuhp_state virtioscsi_online;
 static struct kmem_cache *virtscsi_cmd_cache;
 static mempool_t *virtscsi_cmd_pool;
 
@@ -757,6 +757,13 @@ static void virtscsi_target_destroy(struct scsi_target
*starget)
 	kfree(tgt);
 }
 
+static int virtscsi_map_queues(struct Scsi_Host *shost)
+{
+	struct virtio_scsi *vscsi = shost_priv(shost);
+
+	return blk_mq_virtio_map_queues(&shost->tag_set, vscsi->vdev, 2);
+}
+
 static struct scsi_host_template virtscsi_host_template_single = {
 	.module = THIS_MODULE,
 	.name = "Virtio SCSI HBA",
@@ -792,6 +799,7 @@ static struct scsi_host_template
virtscsi_host_template_multi = {
 	.use_clustering = ENABLE_CLUSTERING,
 	.target_alloc = virtscsi_target_alloc,
 	.target_destroy = virtscsi_target_destroy,
+	.map_queues = virtscsi_map_queues,
 	.track_queue_depth = 1,
 };
 
@@ -808,80 +816,6 @@ static struct scsi_host_template
virtscsi_host_template_multi = {
 		virtio_cwrite(vdev, struct virtio_scsi_config, fld, &__val); \
 	} while(0)
 
-static void __virtscsi_set_affinity(struct virtio_scsi *vscsi, bool affinity)
-{
-	int i;
-	int cpu;
-
-	/* In multiqueue mode, when the number of cpu is equal
-	 * to the number of request queues, we let the qeueues
-	 * to be private to one cpu by setting the affinity hint
-	 * to eliminate the contention.
-	 */
-	if ((vscsi->num_queues == 1 ||
-	     vscsi->num_queues != num_online_cpus()) && affinity) {
-		if (vscsi->affinity_hint_set)
-			affinity = false;
-		else
-			return;
-	}
-
-	if (affinity) {
-		i = 0;
-		for_each_online_cpu(cpu) {
-			virtqueue_set_affinity(vscsi->req_vqs[i].vq, cpu);
-			i++;
-		}
-
-		vscsi->affinity_hint_set = true;
-	} else {
-		for (i = 0; i < vscsi->num_queues; i++) {
-			if (!vscsi->req_vqs[i].vq)
-				continue;
-
-			virtqueue_set_affinity(vscsi->req_vqs[i].vq, -1);
-		}
-
-		vscsi->affinity_hint_set = false;
-	}
-}
-
-static void virtscsi_set_affinity(struct virtio_scsi *vscsi, bool affinity)
-{
-	get_online_cpus();
-	__virtscsi_set_affinity(vscsi, affinity);
-	put_online_cpus();
-}
-
-static int virtscsi_cpu_online(unsigned int cpu, struct hlist_node *node)
-{
-	struct virtio_scsi *vscsi = hlist_entry_safe(node, struct virtio_scsi,
-						     node);
-	__virtscsi_set_affinity(vscsi, true);
-	return 0;
-}
-
-static int virtscsi_cpu_notif_add(struct virtio_scsi *vi)
-{
-	int ret;
-
-	ret = cpuhp_state_add_instance(virtioscsi_online, &vi->node);
-	if (ret)
-		return ret;
-
-	ret = cpuhp_state_add_instance(CPUHP_VIRT_SCSI_DEAD, &vi->node_dead);
-	if (ret)
-		cpuhp_state_remove_instance(virtioscsi_online, &vi->node);
-	return ret;
-}
-
-static void virtscsi_cpu_notif_remove(struct virtio_scsi *vi)
-{
-	cpuhp_state_remove_instance_nocalls(virtioscsi_online, &vi->node);
-	cpuhp_state_remove_instance_nocalls(CPUHP_VIRT_SCSI_DEAD,
-					    &vi->node_dead);
-}
-
 static void virtscsi_init_vq(struct virtio_scsi_vq *virtscsi_vq,
 			     struct virtqueue *vq)
 {
@@ -891,14 +825,8 @@ static void virtscsi_init_vq(struct virtio_scsi_vq
*virtscsi_vq,
 
 static void virtscsi_remove_vqs(struct virtio_device *vdev)
 {
-	struct Scsi_Host *sh = virtio_scsi_host(vdev);
-	struct virtio_scsi *vscsi = shost_priv(sh);
-
-	virtscsi_set_affinity(vscsi, false);
-
 	/* Stop all the virtqueues. */
 	vdev->config->reset(vdev);
-
 	vdev->config->del_vqs(vdev);
 }
 
@@ -911,6 +839,7 @@ static int virtscsi_init(struct virtio_device *vdev,
 	vq_callback_t **callbacks;
 	const char **names;
 	struct virtqueue **vqs;
+	struct irq_affinity desc = { .pre_vectors = 2 };
 
 	num_vqs = vscsi->num_queues + VIRTIO_SCSI_VQ_BASE;
 	vqs = kmalloc(num_vqs * sizeof(struct virtqueue *), GFP_KERNEL);
@@ -933,7 +862,7 @@ static int virtscsi_init(struct virtio_device *vdev,
 
 	/* Discover virtqueues and write information to configuration.  */
 	err = vdev->config->find_vqs(vdev, num_vqs, vqs, callbacks, names,
-			NULL);
+			&desc);
 	if (err)
 		goto out;
 
@@ -999,10 +928,6 @@ static int virtscsi_probe(struct virtio_device *vdev)
 	if (err)
 		goto virtscsi_init_failed;
 
-	err = virtscsi_cpu_notif_add(vscsi);
-	if (err)
-		goto scsi_add_host_failed;
-
 	cmd_per_lun = virtscsi_config_get(vdev, cmd_per_lun) ?: 1;
 	shost->cmd_per_lun = min_t(u32, cmd_per_lun, shost->can_queue);
 	shost->max_sectors = virtscsi_config_get(vdev, max_sectors) ?: 0xFFFF;
@@ -1057,9 +982,6 @@ static void virtscsi_remove(struct virtio_device *vdev)
 		virtscsi_cancel_event_work(vscsi);
 
 	scsi_remove_host(shost);
-
-	virtscsi_cpu_notif_remove(vscsi);
-
 	virtscsi_remove_vqs(vdev);
 	scsi_host_put(shost);
 }
@@ -1067,10 +989,6 @@ static void virtscsi_remove(struct virtio_device *vdev)
 #ifdef CONFIG_PM_SLEEP
 static int virtscsi_freeze(struct virtio_device *vdev)
 {
-	struct Scsi_Host *sh = virtio_scsi_host(vdev);
-	struct virtio_scsi *vscsi = shost_priv(sh);
-
-	virtscsi_cpu_notif_remove(vscsi);
 	virtscsi_remove_vqs(vdev);
 	return 0;
 }
@@ -1085,11 +1003,6 @@ static int virtscsi_restore(struct virtio_device *vdev)
 	if (err)
 		return err;
 
-	err = virtscsi_cpu_notif_add(vscsi);
-	if (err) {
-		vdev->config->del_vqs(vdev);
-		return err;
-	}
 	virtio_device_ready(vdev);
 
 	if (virtio_has_feature(vdev, VIRTIO_SCSI_F_HOTPLUG))
@@ -1144,16 +1057,6 @@ static int __init init(void)
 		pr_err("mempool_create() for virtscsi_cmd_pool failed\n");
 		goto error;
 	}
-	ret = cpuhp_setup_state_multi(CPUHP_AP_ONLINE_DYN,
-				      "scsi/virtio:online",
-				      virtscsi_cpu_online, NULL);
-	if (ret < 0)
-		goto error;
-	virtioscsi_online = ret;
-	ret = cpuhp_setup_state_multi(CPUHP_VIRT_SCSI_DEAD,
"scsi/virtio:dead",
-				      NULL, virtscsi_cpu_online);
-	if (ret)
-		goto error;
 	ret = register_virtio_driver(&virtio_scsi_driver);
 	if (ret < 0)
 		goto error;
@@ -1169,17 +1072,12 @@ static int __init init(void)
 		kmem_cache_destroy(virtscsi_cmd_cache);
 		virtscsi_cmd_cache = NULL;
 	}
-	if (virtioscsi_online)
-		cpuhp_remove_multi_state(virtioscsi_online);
-	cpuhp_remove_multi_state(CPUHP_VIRT_SCSI_DEAD);
 	return ret;
 }
 
 static void __exit fini(void)
 {
 	unregister_virtio_driver(&virtio_scsi_driver);
-	cpuhp_remove_multi_state(virtioscsi_online);
-	cpuhp_remove_multi_state(CPUHP_VIRT_SCSI_DEAD);
 	mempool_destroy(virtscsi_cmd_pool);
 	kmem_cache_destroy(virtscsi_cmd_cache);
 }
diff --git a/include/linux/cpuhotplug.h b/include/linux/cpuhotplug.h
index d936a00..16d8bb9 100644
--- a/include/linux/cpuhotplug.h
+++ b/include/linux/cpuhotplug.h
@@ -28,7 +28,6 @@ enum cpuhp_state {
 	CPUHP_ARM_OMAP_WAKE_DEAD,
 	CPUHP_IRQ_POLL_DEAD,
 	CPUHP_BLOCK_SOFTIRQ_DEAD,
-	CPUHP_VIRT_SCSI_DEAD,
 	CPUHP_ACPI_CPUDRV_DEAD,
 	CPUHP_S390_PFAULT_DEAD,
 	CPUHP_BLK_MQ_DEAD,
-- 
2.1.4

On 2017?01?27? 16:16, Christoph Hellwig wrote:
> We don't really need struct virtio_pci_vq_info, as most field in there
> are redundant:
>
>   - the vq backpointer is not strictly neede to start with
>   - the entry in the vqs list is not needed - the generic virtqueue already
>     has list, we only need to check if it has a callback to get the same
>     semantics
>   - we can use a simple array to look up the MSI-X vec if needed.
>   - That simple array now also duoble serves to replace the per_vq_vectors
>     flag
>
> Signed-off-by: Christoph Hellwig <hch at lst.de>
> ---
>   drivers/virtio/virtio_pci_common.c | 117
+++++++++++--------------------------
>   drivers/virtio/virtio_pci_common.h |  25 +-------
>   drivers/virtio/virtio_pci_legacy.c |   6 +-
>   drivers/virtio/virtio_pci_modern.c |   6 +-
>   4 files changed, 39 insertions(+), 115 deletions(-)
>
> diff --git a/drivers/virtio/virtio_pci_common.c
b/drivers/virtio/virtio_pci_common.c
> index 186cbab..a3376731 100644
> --- a/drivers/virtio/virtio_pci_common.c
> +++ b/drivers/virtio/virtio_pci_common.c
> @@ -62,16 +62,13 @@ static irqreturn_t vp_config_changed(int irq, void
*opaque)
>   static irqreturn_t vp_vring_interrupt(int irq, void *opaque)
>   {
>   	struct virtio_pci_device *vp_dev = opaque;
> -	struct virtio_pci_vq_info *info;
>   	irqreturn_t ret = IRQ_NONE;
> -	unsigned long flags;
> +	struct virtqueue *vq;
>   
> -	spin_lock_irqsave(&vp_dev->lock, flags);
> -	list_for_each_entry(info, &vp_dev->virtqueues, node) {
> -		if (vring_interrupt(irq, info->vq) == IRQ_HANDLED)
> +	list_for_each_entry(vq, &vp_dev->vdev.vqs, list) {
> +		if (vq->callback && vring_interrupt(irq, vq) == IRQ_HANDLED)
The check of vq->callback seems redundant, we will check it soon in 
vring_interrupt().

Thanks

On 2017?01?27? 16:16, Christoph Hellwig wrote:
> +		snprintf(vp_dev->msix_names[i + 1],
> +			 sizeof(*vp_dev->msix_names), "%s-%s",
>   			 dev_name(&vp_dev->vdev.dev), names[i]);
>   		err = request_irq(pci_irq_vector(vp_dev->pci_dev, msix_vec),
> -				  vring_interrupt, 0,
> -				  vp_dev->msix_names[msix_vec],
> -				  vqs[i]);
> +				  vring_interrupt, IRQF_SHARED,
> +				  vp_dev->msix_names[i + 1], vqs[i]);
Do we need to check per_vq_vectors before dereferencing msix_names[i + 1] ?

Thanks

On 2017?01?27? 16:16, Christoph Hellwig wrote:
> Signed-off-by: Christoph Hellwig <hch at lst.de>
> ---
Reviewed-by: Jason Wang <jasowang at redhat.com>
>   drivers/virtio/virtio_pci_common.c | 5 ++---
>   drivers/virtio/virtio_pci_common.h | 2 --
>   drivers/virtio/virtio_pci_legacy.c | 2 +-
>   drivers/virtio/virtio_pci_modern.c | 2 +-
>   include/uapi/linux/virtio_pci.h    | 2 +-
>   5 files changed, 5 insertions(+), 8 deletions(-)
>
> diff --git a/drivers/virtio/virtio_pci_common.c
b/drivers/virtio/virtio_pci_common.c
> index 5880e86..9c4ad7d3f 100644
> --- a/drivers/virtio/virtio_pci_common.c
> +++ b/drivers/virtio/virtio_pci_common.c
> @@ -125,7 +125,7 @@ void vp_del_vqs(struct virtio_device *vdev)
>   
>   	vp_remove_vqs(vdev);
>   
> -	if (vp_dev->msix_enabled) {
> +	if (vp_dev->pci_dev->msix_enabled) {
>   		for (i = 0; i < vp_dev->msix_vectors; i++)
>   			free_cpumask_var(vp_dev->msix_affinity_masks[i]);
>   
> @@ -245,7 +245,6 @@ static int vp_find_vqs_msix(struct virtio_device *vdev,
unsigned nvqs,
>   			allocated_vectors++;
>   	}
>   
> -	vp_dev->msix_enabled = 1;
>   	return 0;
>   
>   out_remove_vqs:
> @@ -341,7 +340,7 @@ int vp_set_vq_affinity(struct virtqueue *vq, int cpu)
>   	if (!vq->callback)
>   		return -EINVAL;
>   
> -	if (vp_dev->msix_enabled) {
> +	if (vp_dev->pci_dev->msix_enabled) {
>   		int vec = vp_dev->msix_vector_map[vq->index];
>   		struct cpumask *mask = vp_dev->msix_affinity_masks[vec];
>   		unsigned int irq = pci_irq_vector(vp_dev->pci_dev, vec);
> diff --git a/drivers/virtio/virtio_pci_common.h
b/drivers/virtio/virtio_pci_common.h
> index 8559386..217ca87 100644
> --- a/drivers/virtio/virtio_pci_common.h
> +++ b/drivers/virtio/virtio_pci_common.h
> @@ -64,8 +64,6 @@ struct virtio_pci_device {
>   	/* the IO mapping for the PCI config space */
>   	void __iomem *ioaddr;
>   
> -	/* MSI-X support */
> -	int msix_enabled;
>   	cpumask_var_t *msix_affinity_masks;
>   	/* Name strings for interrupts. This size should be enough,
>   	 * and I'm too lazy to allocate each name separately. */
> diff --git a/drivers/virtio/virtio_pci_legacy.c
b/drivers/virtio/virtio_pci_legacy.c
> index 47292da..2ab6aee 100644
> --- a/drivers/virtio/virtio_pci_legacy.c
> +++ b/drivers/virtio/virtio_pci_legacy.c
> @@ -165,7 +165,7 @@ static void del_vq(struct virtqueue *vq)
>   
>   	iowrite16(vq->index, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_SEL);
>   
> -	if (vp_dev->msix_enabled) {
> +	if (vp_dev->pci_dev->msix_enabled) {
>   		iowrite16(VIRTIO_MSI_NO_VECTOR,
>   			  vp_dev->ioaddr + VIRTIO_MSI_QUEUE_VECTOR);
>   		/* Flush the write out to device */
> diff --git a/drivers/virtio/virtio_pci_modern.c
b/drivers/virtio/virtio_pci_modern.c
> index 00e6fc1..e5ce310 100644
> --- a/drivers/virtio/virtio_pci_modern.c
> +++ b/drivers/virtio/virtio_pci_modern.c
> @@ -412,7 +412,7 @@ static void del_vq(struct virtqueue *vq)
>   
>   	vp_iowrite16(vq->index, &vp_dev->common->queue_select);
>   
> -	if (vp_dev->msix_enabled) {
> +	if (vp_dev->pci_dev->msix_enabled) {
>   		vp_iowrite16(VIRTIO_MSI_NO_VECTOR,
>   			     &vp_dev->common->queue_msix_vector);
>   		/* Flush the write out to device */
> diff --git a/include/uapi/linux/virtio_pci.h
b/include/uapi/linux/virtio_pci.h
> index 90007a1..15b4385 100644
> --- a/include/uapi/linux/virtio_pci.h
> +++ b/include/uapi/linux/virtio_pci.h
> @@ -79,7 +79,7 @@
>    * configuration space */
>   #define VIRTIO_PCI_CONFIG_OFF(msix_enabled)	((msix_enabled) ? 24 : 20)
>   /* Deprecated: please use VIRTIO_PCI_CONFIG_OFF instead */
> -#define VIRTIO_PCI_CONFIG(dev)
VIRTIO_PCI_CONFIG_OFF((dev)->msix_enabled)
> +#define VIRTIO_PCI_CONFIG(dev)
VIRTIO_PCI_CONFIG_OFF((dev)->pci_dev->msix_enabled)
>   
>   /* Virtio ABI version, this must match exactly */
>   #define VIRTIO_PCI_ABI_VERSION		0

On 2017?01?27? 16:16, Christoph Hellwig wrote:
> Try to grab the MSI-X vectors early and fall back to the shared one
> before doing lots of allocations.
>
> Signed-off-by: Christoph Hellwig <hch at lst.de>
> ---
Reviewed-by: Jason Wang <jasowang at redhat.com>
>   drivers/virtio/virtio_pci_common.c | 58
+++++++++++++++++++-------------------
>   1 file changed, 29 insertions(+), 29 deletions(-)
>
> diff --git a/drivers/virtio/virtio_pci_common.c
b/drivers/virtio/virtio_pci_common.c
> index 9c4ad7d3f..74ff74c0 100644
> --- a/drivers/virtio/virtio_pci_common.c
> +++ b/drivers/virtio/virtio_pci_common.c
> @@ -142,32 +142,39 @@ void vp_del_vqs(struct virtio_device *vdev)
>   }
>   
>   static int vp_find_vqs_msix(struct virtio_device *vdev, unsigned nvqs,
> -			      struct virtqueue *vqs[],
> -			      vq_callback_t *callbacks[],
> -			      const char * const names[],
> -			      bool per_vq_vectors)
> +		struct virtqueue *vqs[], vq_callback_t *callbacks[],
> +		const char * const names[])
>   {
>   	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
>   	const char *name = dev_name(&vp_dev->vdev.dev);
>   	int i, err = -ENOMEM, allocated_vectors, nvectors;
> +	bool shared = false;
>   	u16 msix_vec;
>   
> -	if (per_vq_vectors) {
> -		/* Best option: one for change interrupt, one per vq. */
> -		nvectors = 1;
> -		for (i = 0; i < nvqs; ++i)
> -			if (callbacks[i])
> -				++nvectors;
> -	} else {
> -		/* Second best: one for change, shared for all vqs. */
> +	nvectors = 1;
> +	for (i = 0; i < nvqs; i++) {
> +		if (callbacks[i])
> +			nvectors++;
> +	}
> +
> +	/* Try one vector per queue first. */
> +	err = pci_alloc_irq_vectors(vp_dev->pci_dev, nvectors, nvectors,
> +			PCI_IRQ_MSIX);
> +	if (err < 0) {
> +		/* Fallback to one vector for config, one shared for queues. */
> +		shared = true;
>   		nvectors = 2;
> +		err = pci_alloc_irq_vectors(vp_dev->pci_dev, nvectors, nvectors,
> +				PCI_IRQ_MSIX);
> +		if (err < 0)
> +			return err;
>   	}
>   
>   	vp_dev->msix_vectors = nvectors;
>   	vp_dev->msix_names = kmalloc_array(nvectors,
>   			sizeof(*vp_dev->msix_names), GFP_KERNEL);
>   	if (!vp_dev->msix_names)
> -		return err;
> +		goto out_free_irq_vectors;
>   
>   	vp_dev->msix_affinity_masks = kcalloc(nvectors,
>   			sizeof(*vp_dev->msix_affinity_masks), GFP_KERNEL);
> @@ -180,18 +187,13 @@ static int vp_find_vqs_msix(struct virtio_device
*vdev, unsigned nvqs,
>   			goto out_free_msix_affinity_masks;
>   	}
>   
> -	err = pci_alloc_irq_vectors(vp_dev->pci_dev, nvectors, nvectors,
> -			PCI_IRQ_MSIX);
> -	if (err < 0)
> -		goto out_free_msix_affinity_masks;
> -
>   	/* Set the vector used for configuration */
>   	snprintf(vp_dev->msix_names[0], sizeof(*vp_dev->msix_names),
>   		 "%s-config", name);
>   	err = request_irq(pci_irq_vector(vp_dev->pci_dev, 0),
vp_config_changed,
>   			0, vp_dev->msix_names[0], vp_dev);
>   	if (err)
> -		goto out_free_irq_vectors;
> +		goto out_free_msix_affinity_masks;
>   
>   	/* Verify we had enough resources to assign the vector */
>   	if (vp_dev->config_vector(vp_dev, 0) == VIRTIO_MSI_NO_VECTOR) {
> @@ -241,7 +243,11 @@ static int vp_find_vqs_msix(struct virtio_device
*vdev, unsigned nvqs,
>   		}
>   		vp_dev->msix_vector_map[i] = msix_vec;
>   
> -		if (per_vq_vectors)
> +		/*
> +		 * Use a different vector for each queue if they are available,
> +		 * else share the same vector for all VQs.
> +		 */
> +		if (!shared)
>   			allocated_vectors++;
>   	}
>   
> @@ -254,8 +260,6 @@ static int vp_find_vqs_msix(struct virtio_device *vdev,
unsigned nvqs,
>   	vp_dev->config_vector(vp_dev, VIRTIO_MSI_NO_VECTOR);
>   out_free_config_irq:
>   	free_irq(pci_irq_vector(vp_dev->pci_dev, 0), vp_dev);
> -out_free_irq_vectors:
> -	pci_free_irq_vectors(vp_dev->pci_dev);
>   out_free_msix_affinity_masks:
>   	for (i = 0; i < nvectors; i++) {
>   		if (vp_dev->msix_affinity_masks[i])
> @@ -264,6 +268,8 @@ static int vp_find_vqs_msix(struct virtio_device *vdev,
unsigned nvqs,
>   	kfree(vp_dev->msix_affinity_masks);
>   out_free_msix_names:
>   	kfree(vp_dev->msix_names);
> +out_free_irq_vectors:
> +	pci_free_irq_vectors(vp_dev->pci_dev);
>   	return err;
>   }
>   
> @@ -308,15 +314,9 @@ int vp_find_vqs(struct virtio_device *vdev, unsigned
nvqs,
>   {
>   	int err;
>   
> -	/* Try MSI-X with one vector per queue. */
> -	err = vp_find_vqs_msix(vdev, nvqs, vqs, callbacks, names, true);
> -	if (!err)
> -		return 0;
> -	/* Fallback: MSI-X with one vector for config, one shared for queues. */
> -	err = vp_find_vqs_msix(vdev, nvqs, vqs, callbacks, names, false);
> +	err = vp_find_vqs_msix(vdev, nvqs, vqs, callbacks, names);
>   	if (!err)
>   		return 0;
> -	/* Finally fall back to regular interrupts. */
>   	return vp_find_vqs_intx(vdev, nvqs, vqs, callbacks, names);
>   }
>

On 2017?01?27? 16:16, Christoph Hellwig wrote:
> Add a struct irq_affinity pointer to the find_vqs methods, which if set
> is used to tell the PCI layer to create the MSI-X vectors for our I/O
> virtqueues with the proper affinity from the start.  Compared to after
> the fact affinity hints this gives us an instantly working setup and
> allows to allocate the irq descritors node-local and avoid interconnect
> traffic.  Last but not least this will allow blk-mq queues are created
> based on the interrupt affinity for storage drivers.
>
> Signed-off-by: Christoph Hellwig<hch at lst.de>
> ---
Reviewed-by: Jason Wang <jasowang at redhat.com>

On 2017?01?27? 16:16, Christoph Hellwig wrote:
> This basically passed up the pci_irq_get_affinity information through
> virtio through an optional get_vq_affinity method.  It is only implemented
> by the PCI backend for now, and only when we use per-virtqueue IRQs.
>
> Signed-off-by: Christoph Hellwig <hch at lst.de>
> ---
Reviewed-by: Jason Wang <jasowang at redhat.com>
>   drivers/virtio/virtio_pci_common.c | 11 +++++++++++
>   drivers/virtio/virtio_pci_common.h |  2 ++
>   drivers/virtio/virtio_pci_legacy.c |  1 +
>   drivers/virtio/virtio_pci_modern.c |  2 ++
>   include/linux/virtio_config.h      |  3 +++
>   5 files changed, 19 insertions(+)
>
> diff --git a/drivers/virtio/virtio_pci_common.c
b/drivers/virtio/virtio_pci_common.c
> index c244212..31ba259 100644
> --- a/drivers/virtio/virtio_pci_common.c
> +++ b/drivers/virtio/virtio_pci_common.c
> @@ -361,6 +361,17 @@ int vp_set_vq_affinity(struct virtqueue *vq, int cpu)
>   	return 0;
>   }
>   
> +const struct cpumask *vp_get_vq_affinity(struct virtio_device *vdev, int
index)
> +{
> +	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
> +	unsigned int *map = vp_dev->msix_vector_map;
> +
> +	if (!map || map[index] == VIRTIO_MSI_NO_VECTOR)
> +		return NULL;
> +
> +	return pci_irq_get_affinity(vp_dev->pci_dev, map[index]);
> +}
> +
>   #ifdef CONFIG_PM_SLEEP
>   static int virtio_pci_freeze(struct device *dev)
>   {
> diff --git a/drivers/virtio/virtio_pci_common.h
b/drivers/virtio/virtio_pci_common.h
> index a6ad9ec..ac8c9d7 100644
> --- a/drivers/virtio/virtio_pci_common.h
> +++ b/drivers/virtio/virtio_pci_common.h
> @@ -108,6 +108,8 @@ const char *vp_bus_name(struct virtio_device *vdev);
>    */
>   int vp_set_vq_affinity(struct virtqueue *vq, int cpu);
>   
> +const struct cpumask *vp_get_vq_affinity(struct virtio_device *vdev, int
index);
> +
>   #if IS_ENABLED(CONFIG_VIRTIO_PCI_LEGACY)
>   int virtio_pci_legacy_probe(struct virtio_pci_device *);
>   void virtio_pci_legacy_remove(struct virtio_pci_device *);
> diff --git a/drivers/virtio/virtio_pci_legacy.c
b/drivers/virtio/virtio_pci_legacy.c
> index 2ab6aee..f7362c5 100644
> --- a/drivers/virtio/virtio_pci_legacy.c
> +++ b/drivers/virtio/virtio_pci_legacy.c
> @@ -190,6 +190,7 @@ static const struct virtio_config_ops
virtio_pci_config_ops = {
>   	.finalize_features = vp_finalize_features,
>   	.bus_name	= vp_bus_name,
>   	.set_vq_affinity = vp_set_vq_affinity,
> +	.get_vq_affinity = vp_get_vq_affinity,
>   };
>   
>   /* the PCI probing function */
> diff --git a/drivers/virtio/virtio_pci_modern.c
b/drivers/virtio/virtio_pci_modern.c
> index a7a0981..7bc3004 100644
> --- a/drivers/virtio/virtio_pci_modern.c
> +++ b/drivers/virtio/virtio_pci_modern.c
> @@ -437,6 +437,7 @@ static const struct virtio_config_ops
virtio_pci_config_nodev_ops = {
>   	.finalize_features = vp_finalize_features,
>   	.bus_name	= vp_bus_name,
>   	.set_vq_affinity = vp_set_vq_affinity,
> +	.get_vq_affinity = vp_get_vq_affinity,
>   };
>   
>   static const struct virtio_config_ops virtio_pci_config_ops = {
> @@ -452,6 +453,7 @@ static const struct virtio_config_ops
virtio_pci_config_ops = {
>   	.finalize_features = vp_finalize_features,
>   	.bus_name	= vp_bus_name,
>   	.set_vq_affinity = vp_set_vq_affinity,
> +	.get_vq_affinity = vp_get_vq_affinity,
>   };
>   
>   /**
> diff --git a/include/linux/virtio_config.h b/include/linux/virtio_config.h
> index 2ebe506..8355bab 100644
> --- a/include/linux/virtio_config.h
> +++ b/include/linux/virtio_config.h
> @@ -58,6 +58,7 @@ struct irq_affinity;
>    *      This returns a pointer to the bus name a la pci_name from which
>    *      the caller can then copy.
>    * @set_vq_affinity: set the affinity for a virtqueue.
> + * @get_vq_affinity: get the affinity for a virtqueue (optional).
>    */
>   typedef void vq_callback_t(struct virtqueue *);
>   struct virtio_config_ops {
> @@ -77,6 +78,8 @@ struct virtio_config_ops {
>   	int (*finalize_features)(struct virtio_device *vdev);
>   	const char *(*bus_name)(struct virtio_device *vdev);
>   	int (*set_vq_affinity)(struct virtqueue *vq, int cpu);
> +	const struct cpumask *(*get_vq_affinity)(struct virtio_device *vdev,
> +			int index);
>   };
>   
>   /* If driver didn't advertise the feature, it will never appear. */