Linux Virtualization - Jul 2020 - [PATCH 0/5]

Hi all:

This series tries to support batched IOTLB updating vhost-vdpa.

Currently vhost-vdpa accepts userspace mapping via IOTLB API, and it
can only forward one mapping to IOMMU or device through IOMMU API or
dma_map(). Though set_map() is designed to have the capability to pass
an rbtree based mapping to vDPA device, it's still be called at least
once for each VHOST_IOTLB_UPDATE or VHOST_IOTLB_INVALIDATE. This is
because vhost-vdpa doesn't know the userspace start or stop then
updating.

So this patch introduces two flags as hints for vhost-vdpa to call
set_map() only when userspace finish a batch of IOTLB updating.

So instead of:

1) VHOST_IOTLB_UPDATE/VHOST_IOTLB_INVALIDATE -> set_map() (s)
2) VHOST_IOTLB_UPDATE/VHOST_IOTLB_INVALIDATE -> set_map() (s)
...
n) VHOST_IOTLB_UPDATE/VHOST_IOTLB_INVALIDATE -> set_map() (s)

With the help of hints, we do:

0) VHOST_IOTLB_BATCH_START

1) VHOST_IOTLB_UPDATE/INVALIDATE
...
n) VHOST_IOTLB_UPDATE/INVALIDATE

n+1) VHOST_IOTLB_BATCH_END -> set_map()

One one call of set_map() to vDPA device for a batch of IOTLB
mappings. So for the device that has its own DMA translation logic, it
can efficiently structure the memory mapping to get best performance.

Note, this only impacts the devices that want its own DMA
translation (less times of set_map() call). For other type of devices,
there's no changes in the behaviour.

Changes from RFCV1:

- tweak the comments per Michael's request

Jason Wang (5):
  vhost-vdpa: refine ioctl pre-processing
  vhost: generialize backend features setting/getting
  vhost-vdpa: support get/set backend features
  vhost-vdpa: support IOTLB batching hints
  vdpasim: support batch updating

 drivers/vdpa/vdpa_sim/vdpa_sim.c | 40 +++++++++++++++++++++--
 drivers/vhost/net.c              | 18 ++---------
 drivers/vhost/vdpa.c             | 55 ++++++++++++++++++++++++++------
 drivers/vhost/vhost.c            | 15 +++++++++
 drivers/vhost/vhost.h            |  2 ++
 include/uapi/linux/vhost.h       |  2 ++
 include/uapi/linux/vhost_types.h | 11 +++++++
 7 files changed, 114 insertions(+), 29 deletions(-)

-- 
2.20.1

Switch to use 'switch' to make the codes more easier to be extended.

Signed-off-by: Jason Wang <jasowang at redhat.com>
---
 drivers/vhost/vdpa.c | 9 +++++----
 1 file changed, 5 insertions(+), 4 deletions(-)

diff --git a/drivers/vhost/vdpa.c b/drivers/vhost/vdpa.c
index 77a0c9fb6cc3..db4c9cb44c61 100644
--- a/drivers/vhost/vdpa.c
+++ b/drivers/vhost/vdpa.c
@@ -353,15 +353,16 @@ static long vhost_vdpa_vring_ioctl(struct vhost_vdpa *v,
unsigned int cmd,
 	idx = array_index_nospec(idx, v->nvqs);
 	vq = &v->vqs[idx];
 
-	if (cmd == VHOST_VDPA_SET_VRING_ENABLE) {
+	switch (cmd) {
+	case VHOST_VDPA_SET_VRING_ENABLE:
 		if (copy_from_user(&s, argp, sizeof(s)))
 			return -EFAULT;
 		ops->set_vq_ready(vdpa, idx, s.num);
 		return 0;
-	}
-
-	if (cmd == VHOST_GET_VRING_BASE)
+	case VHOST_GET_VRING_BASE:
 		vq->last_avail_idx = ops->get_vq_state(v->vdpa, idx);
+		break;
+	}
 
 	r = vhost_vring_ioctl(&v->vdev, cmd, argp);
 	if (r)
-- 
2.20.1

Move the backend features setting/getting from net.c to vhost.c to be
reused by vhost-vdpa.

Signed-off-by: Jason Wang <jasowang at redhat.com>
---
 drivers/vhost/net.c   | 18 ++----------------
 drivers/vhost/vhost.c | 15 +++++++++++++++
 drivers/vhost/vhost.h |  2 ++
 3 files changed, 19 insertions(+), 16 deletions(-)

diff --git a/drivers/vhost/net.c b/drivers/vhost/net.c
index 0b509be8d7b1..d88afe3f6060 100644
--- a/drivers/vhost/net.c
+++ b/drivers/vhost/net.c
@@ -1622,21 +1622,6 @@ static long vhost_net_reset_owner(struct vhost_net *n)
 	return err;
 }
 
-static int vhost_net_set_backend_features(struct vhost_net *n, u64 features)
-{
-	int i;
-
-	mutex_lock(&n->dev.mutex);
-	for (i = 0; i < VHOST_NET_VQ_MAX; ++i) {
-		mutex_lock(&n->vqs[i].vq.mutex);
-		n->vqs[i].vq.acked_backend_features = features;
-		mutex_unlock(&n->vqs[i].vq.mutex);
-	}
-	mutex_unlock(&n->dev.mutex);
-
-	return 0;
-}
-
 static int vhost_net_set_features(struct vhost_net *n, u64 features)
 {
 	size_t vhost_hlen, sock_hlen, hdr_len;
@@ -1737,7 +1722,8 @@ static long vhost_net_ioctl(struct file *f, unsigned int
ioctl,
 			return -EFAULT;
 		if (features & ~VHOST_NET_BACKEND_FEATURES)
 			return -EOPNOTSUPP;
-		return vhost_net_set_backend_features(n, features);
+		vhost_set_backend_features(&n->dev, features);
+		return 0;
 	case VHOST_RESET_OWNER:
 		return vhost_net_reset_owner(n);
 	case VHOST_SET_OWNER:
diff --git a/drivers/vhost/vhost.c b/drivers/vhost/vhost.c
index dc510dc2b1ef..ba61f499d420 100644
--- a/drivers/vhost/vhost.c
+++ b/drivers/vhost/vhost.c
@@ -2700,6 +2700,21 @@ struct vhost_msg_node *vhost_dequeue_msg(struct vhost_dev
*dev,
 }
 EXPORT_SYMBOL_GPL(vhost_dequeue_msg);
 
+void vhost_set_backend_features(struct vhost_dev *dev, u64 features)
+{
+	struct vhost_virtqueue *vq;
+	int i;
+
+	mutex_lock(&dev->mutex);
+	for (i = 0; i < dev->nvqs; ++i) {
+		vq = dev->vqs[i];
+		mutex_lock(&vq->mutex);
+		vq->acked_backend_features = features;
+		mutex_unlock(&vq->mutex);
+	}
+	mutex_unlock(&dev->mutex);
+}
+EXPORT_SYMBOL_GPL(vhost_set_backend_features);
 
 static int __init vhost_init(void)
 {
diff --git a/drivers/vhost/vhost.h b/drivers/vhost/vhost.h
index 264a2a2fae97..52753aecd82a 100644
--- a/drivers/vhost/vhost.h
+++ b/drivers/vhost/vhost.h
@@ -223,6 +223,8 @@ void vhost_enqueue_msg(struct vhost_dev *dev,
 		       struct vhost_msg_node *node);
 struct vhost_msg_node *vhost_dequeue_msg(struct vhost_dev *dev,
 					 struct list_head *head);
+void vhost_set_backend_features(struct vhost_dev *dev, u64 features);
+
 __poll_t vhost_chr_poll(struct file *file, struct vhost_dev *dev,
 			    poll_table *wait);
 ssize_t vhost_chr_read_iter(struct vhost_dev *dev, struct iov_iter *to,
-- 
2.20.1

This patch makes userspace can get and set backend features to
vhost-vdpa.

Signed-off-by: Cindy Lu <lulu at redhat.com>
Signed-off-by: Jason Wang <jasowang at redhat.com>
---
 drivers/vhost/vdpa.c | 18 ++++++++++++++++++
 1 file changed, 18 insertions(+)

diff --git a/drivers/vhost/vdpa.c b/drivers/vhost/vdpa.c
index db4c9cb44c61..453057421f80 100644
--- a/drivers/vhost/vdpa.c
+++ b/drivers/vhost/vdpa.c
@@ -55,6 +55,10 @@ enum {
 		(1ULL << VIRTIO_NET_F_SPEED_DUPLEX),
 };
 
+enum {
+	VHOST_VDPA_BACKEND_FEATURES = (1ULL << VHOST_BACKEND_F_IOTLB_MSG_V2)
+};
+
 /* Currently, only network backend w/o multiqueue is supported. */
 #define VHOST_VDPA_VQ_MAX	2
 
@@ -340,6 +344,8 @@ static long vhost_vdpa_vring_ioctl(struct vhost_vdpa *v,
unsigned int cmd,
 	struct vdpa_callback cb;
 	struct vhost_virtqueue *vq;
 	struct vhost_vring_state s;
+	u64 __user *featurep = argp;
+	u64 features;
 	u32 idx;
 	long r;
 
@@ -362,6 +368,18 @@ static long vhost_vdpa_vring_ioctl(struct vhost_vdpa *v,
unsigned int cmd,
 	case VHOST_GET_VRING_BASE:
 		vq->last_avail_idx = ops->get_vq_state(v->vdpa, idx);
 		break;
+	case VHOST_GET_BACKEND_FEATURES:
+		features = VHOST_VDPA_BACKEND_FEATURES;
+		if (copy_to_user(featurep, &features, sizeof(features)))
+			return -EFAULT;
+		return 0;
+	case VHOST_SET_BACKEND_FEATURES:
+		if (copy_from_user(&features, featurep, sizeof(features)))
+			return -EFAULT;
+		if (features & ~VHOST_VDPA_BACKEND_FEATURES)
+			return -EOPNOTSUPP;
+		vhost_set_backend_features(&v->vdev, features);
+		return 0;
 	}
 
 	r = vhost_vring_ioctl(&v->vdev, cmd, argp);
-- 
2.20.1

This patches extend the vhost IOTLB API to accept batch updating hints
form userspace. When userspace wants update the device IOTLB in a
batch, it may do:

1) Write vhost_iotlb_msg with VHOST_IOTLB_BATCH_BEGIN flag
2) Perform a batch of IOTLB updating via VHOST_IOTLB_UPDATE/INVALIDATE
3) Write vhost_iotlb_msg with VHOST_IOTLB_BATCH_END flag

Vhost-vdpa may decide to batch the IOMMU/IOTLB updating in step 3 when
vDPA device support set_map() ops. This is useful for the vDPA device
that want to know all the mappings to tweak their own DMA translation
logic.

For vDPA device that doesn't require set_map(), no behavior changes.

This capability is advertised via VHOST_BACKEND_F_IOTLB_BATCH capability.

Signed-off-by: Jason Wang <jasowang at redhat.com>
---
 drivers/vhost/vdpa.c             | 30 +++++++++++++++++++++++-------
 include/uapi/linux/vhost.h       |  2 ++
 include/uapi/linux/vhost_types.h | 11 +++++++++++
 3 files changed, 36 insertions(+), 7 deletions(-)

diff --git a/drivers/vhost/vdpa.c b/drivers/vhost/vdpa.c
index 453057421f80..8f624bbafee7 100644
--- a/drivers/vhost/vdpa.c
+++ b/drivers/vhost/vdpa.c
@@ -56,7 +56,9 @@ enum {
 };
 
 enum {
-	VHOST_VDPA_BACKEND_FEATURES = (1ULL << VHOST_BACKEND_F_IOTLB_MSG_V2)
+	VHOST_VDPA_BACKEND_FEATURES +	(1ULL << VHOST_BACKEND_F_IOTLB_MSG_V2) |
+	(1ULL << VHOST_BACKEND_F_IOTLB_BATCH),
 };
 
 /* Currently, only network backend w/o multiqueue is supported. */
@@ -77,6 +79,7 @@ struct vhost_vdpa {
 	int virtio_id;
 	int minor;
 	struct eventfd_ctx *config_ctx;
+	int in_batch;
 };
 
 static DEFINE_IDA(vhost_vdpa_ida);
@@ -125,6 +128,7 @@ static void vhost_vdpa_reset(struct vhost_vdpa *v)
 	const struct vdpa_config_ops *ops = vdpa->config;
 
 	ops->set_status(vdpa, 0);
+	v->in_batch = 0;
 }
 
 static long vhost_vdpa_get_device_id(struct vhost_vdpa *v, u8 __user *argp)
@@ -540,9 +544,10 @@ static int vhost_vdpa_map(struct vhost_vdpa *v,
 
 	if (ops->dma_map)
 		r = ops->dma_map(vdpa, iova, size, pa, perm);
-	else if (ops->set_map)
-		r = ops->set_map(vdpa, dev->iotlb);
-	else
+	else if (ops->set_map) {
+		if (!v->in_batch)
+			r = ops->set_map(vdpa, dev->iotlb);
+	} else
 		r = iommu_map(v->domain, iova, pa, size,
 			      perm_to_iommu_flags(perm));
 
@@ -559,9 +564,10 @@ static void vhost_vdpa_unmap(struct vhost_vdpa *v, u64
iova, u64 size)
 
 	if (ops->dma_map)
 		ops->dma_unmap(vdpa, iova, size);
-	else if (ops->set_map)
-		ops->set_map(vdpa, dev->iotlb);
-	else
+	else if (ops->set_map) {
+		if (!v->in_batch)
+			ops->set_map(vdpa, dev->iotlb);
+	} else
 		iommu_unmap(v->domain, iova, size);
 }
 
@@ -655,6 +661,8 @@ static int vhost_vdpa_process_iotlb_msg(struct vhost_dev
*dev,
 					struct vhost_iotlb_msg *msg)
 {
 	struct vhost_vdpa *v = container_of(dev, struct vhost_vdpa, vdev);
+	struct vdpa_device *vdpa = v->vdpa;
+	const struct vdpa_config_ops *ops = vdpa->config;
 	int r = 0;
 
 	r = vhost_dev_check_owner(dev);
@@ -668,6 +676,14 @@ static int vhost_vdpa_process_iotlb_msg(struct vhost_dev
*dev,
 	case VHOST_IOTLB_INVALIDATE:
 		vhost_vdpa_unmap(v, msg->iova, msg->size);
 		break;
+	case VHOST_IOTLB_BATCH_BEGIN:
+		v->in_batch = true;
+		break;
+	case VHOST_IOTLB_BATCH_END:
+		if (v->in_batch && ops->set_map)
+			ops->set_map(vdpa, dev->iotlb);
+		v->in_batch = false;
+		break;
 	default:
 		r = -EINVAL;
 		break;
diff --git a/include/uapi/linux/vhost.h b/include/uapi/linux/vhost.h
index 0c2349612e77..75232185324a 100644
--- a/include/uapi/linux/vhost.h
+++ b/include/uapi/linux/vhost.h
@@ -91,6 +91,8 @@
 
 /* Use message type V2 */
 #define VHOST_BACKEND_F_IOTLB_MSG_V2 0x1
+/* IOTLB can accept batching hints */
+#define VHOST_BACKEND_F_IOTLB_BATCH  0x2
 
 #define VHOST_SET_BACKEND_FEATURES _IOW(VHOST_VIRTIO, 0x25, __u64)
 #define VHOST_GET_BACKEND_FEATURES _IOR(VHOST_VIRTIO, 0x26, __u64)
diff --git a/include/uapi/linux/vhost_types.h b/include/uapi/linux/vhost_types.h
index 669457ce5c48..9a269a88a6ff 100644
--- a/include/uapi/linux/vhost_types.h
+++ b/include/uapi/linux/vhost_types.h
@@ -60,6 +60,17 @@ struct vhost_iotlb_msg {
 #define VHOST_IOTLB_UPDATE         2
 #define VHOST_IOTLB_INVALIDATE     3
 #define VHOST_IOTLB_ACCESS_FAIL    4
+/*
+ * VHOST_IOTLB_BATCH_BEGIN and VHOST_IOTLB_BATCH_END allow modifying
+ * multiple mappings in one go: beginning with
+ * VHOST_IOTLB_BATCH_BEGIN, followed by any number of
+ * VHOST_IOTLB_UPDATE messages, and ending with VHOST_IOTLB_BATCH_END.
+ * When one of these two values is used as the message type, the rest
+ * of the fields in the message are ignored. There's no guarantee that
+ * these changes take place automatically in the device.
+ */
+#define VHOST_IOTLB_BATCH_BEGIN    5
+#define VHOST_IOTLB_BATCH_END      6
 	__u8 type;
 };
 
-- 
2.20.1

The vDPA simulator support both set_map() and dma_map()/dma_unmap()
operations. But vhost-vdpa can only use one of them. So this patch
introduce a module parameter (batch_mapping) that let vpda_sim to
support only one of those dma operations. The batched mapping via
set_map() is enabled by default.

Signed-off-by: Jason Wang <jasowang at redhat.com>
---
 drivers/vdpa/vdpa_sim/vdpa_sim.c | 40 +++++++++++++++++++++++++++++---
 1 file changed, 37 insertions(+), 3 deletions(-)

diff --git a/drivers/vdpa/vdpa_sim/vdpa_sim.c b/drivers/vdpa/vdpa_sim/vdpa_sim.c
index c7334cc65bb2..a7a0962ed8a8 100644
--- a/drivers/vdpa/vdpa_sim/vdpa_sim.c
+++ b/drivers/vdpa/vdpa_sim/vdpa_sim.c
@@ -33,6 +33,10 @@
 #define DRV_DESC     "vDPA Device Simulator"
 #define DRV_LICENSE  "GPL v2"
 
+static int batch_mapping = 1;
+module_param(batch_mapping, int, 0444);
+MODULE_PARM_DESC(batch_mapping, "Batched mapping 1 -Enable; 0 -
Disable");
+
 struct vdpasim_virtqueue {
 	struct vringh vring;
 	struct vringh_kiov iov;
@@ -303,16 +307,22 @@ static const struct dma_map_ops vdpasim_dma_ops = {
 };
 
 static const struct vdpa_config_ops vdpasim_net_config_ops;
+static const struct vdpa_config_ops vdpasim_net_batch_config_ops;
 
 static struct vdpasim *vdpasim_create(void)
 {
+	const struct vdpa_config_ops *ops;
 	struct virtio_net_config *config;
 	struct vdpasim *vdpasim;
 	struct device *dev;
 	int ret = -ENOMEM;
 
-	vdpasim = vdpa_alloc_device(struct vdpasim, vdpa, NULL,
-				    &vdpasim_net_config_ops);
+	if (batch_mapping)
+		ops = &vdpasim_net_batch_config_ops;
+	else
+		ops = &vdpasim_net_config_ops;
+
+	vdpasim = vdpa_alloc_device(struct vdpasim, vdpa, NULL, ops);
 	if (!vdpasim)
 		goto err_alloc;
 
@@ -597,12 +607,36 @@ static const struct vdpa_config_ops vdpasim_net_config_ops
= {
 	.get_config             = vdpasim_get_config,
 	.set_config             = vdpasim_set_config,
 	.get_generation         = vdpasim_get_generation,
-	.set_map                = vdpasim_set_map,
 	.dma_map                = vdpasim_dma_map,
 	.dma_unmap              = vdpasim_dma_unmap,
 	.free                   = vdpasim_free,
 };
 
+static const struct vdpa_config_ops vdpasim_net_batch_config_ops = {
+	.set_vq_address         = vdpasim_set_vq_address,
+	.set_vq_num             = vdpasim_set_vq_num,
+	.kick_vq                = vdpasim_kick_vq,
+	.set_vq_cb              = vdpasim_set_vq_cb,
+	.set_vq_ready           = vdpasim_set_vq_ready,
+	.get_vq_ready           = vdpasim_get_vq_ready,
+	.set_vq_state           = vdpasim_set_vq_state,
+	.get_vq_state           = vdpasim_get_vq_state,
+	.get_vq_align           = vdpasim_get_vq_align,
+	.get_features           = vdpasim_get_features,
+	.set_features           = vdpasim_set_features,
+	.set_config_cb          = vdpasim_set_config_cb,
+	.get_vq_num_max         = vdpasim_get_vq_num_max,
+	.get_device_id          = vdpasim_get_device_id,
+	.get_vendor_id          = vdpasim_get_vendor_id,
+	.get_status             = vdpasim_get_status,
+	.set_status             = vdpasim_set_status,
+	.get_config             = vdpasim_get_config,
+	.set_config             = vdpasim_set_config,
+	.get_generation         = vdpasim_get_generation,
+	.set_map                = vdpasim_set_map,
+	.free                   = vdpasim_free,
+};
+
 static int __init vdpasim_dev_init(void)
 {
 	vdpasim_dev = vdpasim_create();
-- 
2.20.1