[RFC PATCH 0/6] vdpa_sim: add support for user VA

[RFC PATCH 0/6] vdpa_sim: add support for user VA

[Stefano Garzarella]

This series adds support for the use of user virtual addresses in the
vDPA simulator devices.

The main reason for this change is to lift the pinning of all guest memory.
Especially with virtio devices implemented in software.

The next step would be to generalize the code in vdpa-sim to allow the
implementation of in-kernel software devices. Similar to vhost, but using vDPA
so we can reuse the same software stack (e.g. in QEMU) for both HW and SW
devices.

For example, we have never merged vhost-blk, and lately there has been interest.
So it would be nice to do it directly with vDPA to reuse the same code in the
VMM for both HW and SW vDPA block devices.

The main problem (addressed by this series) was due to the pinning of all
guest memory, which thus prevented the overcommit of guest memory.

There are still some TODOs to be fixed, but I would like to have your feedback
on this RFC.

Thanks,
Stefano

Note: this series is based on Linux v6.1 + couple of fixes (that I needed to
run libblkio tests) already posted but not yet merged.

Tree available here: https://gitlab.com/sgarzarella/linux/-/tree/vdpa-sim-use-va

Stefano Garzarella (6):
  vdpa: add bind_mm callback
  vhost-vdpa: use bind_mm device callback
  vringh: support VA with iotlb
  vdpa_sim: make devices agnostic for work management
  vdpa_sim: use kthread worker
  vdpa_sim: add support for user VA

 drivers/vdpa/vdpa_sim/vdpa_sim.h     |   7 +-
 include/linux/vdpa.h                 |   8 +
 include/linux/vringh.h               |   5 +-
 drivers/vdpa/mlx5/core/resources.c   |   3 +-
 drivers/vdpa/mlx5/net/mlx5_vnet.c    |   2 +-
 drivers/vdpa/vdpa_sim/vdpa_sim.c     | 132 +++++++++++++-
 drivers/vdpa/vdpa_sim/vdpa_sim_blk.c |   6 +-
 drivers/vdpa/vdpa_sim/vdpa_sim_net.c |   6 +-
 drivers/vhost/vdpa.c                 |  22 +++
 drivers/vhost/vringh.c               | 250 +++++++++++++++++++++------
 10 files changed, 370 insertions(+), 71 deletions(-)

-- 
2.38.1

_______________________________________________
Virtualization mailing list
Virtualization@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

[RFC PATCH 0/6] vdpa_sim: add support for user VA

[Stefano Garzarella]

This series adds support for the use of user virtual addresses in the
vDPA simulator devices.

The main reason for this change is to lift the pinning of all guest memory.
Especially with virtio devices implemented in software.

The next step would be to generalize the code in vdpa-sim to allow the
implementation of in-kernel software devices. Similar to vhost, but using vDPA
so we can reuse the same software stack (e.g. in QEMU) for both HW and SW
devices.

For example, we have never merged vhost-blk, and lately there has been interest.
So it would be nice to do it directly with vDPA to reuse the same code in the
VMM for both HW and SW vDPA block devices.

The main problem (addressed by this series) was due to the pinning of all
guest memory, which thus prevented the overcommit of guest memory.

There are still some TODOs to be fixed, but I would like to have your feedback
on this RFC.

Thanks,
Stefano

Note: this series is based on Linux v6.1 + couple of fixes (that I needed to
run libblkio tests) already posted but not yet merged.

Tree available here: https://gitlab.com/sgarzarella/linux/-/tree/vdpa-sim-use-va

Stefano Garzarella (6):
  vdpa: add bind_mm callback
  vhost-vdpa: use bind_mm device callback
  vringh: support VA with iotlb
  vdpa_sim: make devices agnostic for work management
  vdpa_sim: use kthread worker
  vdpa_sim: add support for user VA

 drivers/vdpa/vdpa_sim/vdpa_sim.h     |   7 +-
 include/linux/vdpa.h                 |   8 +
 include/linux/vringh.h               |   5 +-
 drivers/vdpa/mlx5/core/resources.c   |   3 +-
 drivers/vdpa/mlx5/net/mlx5_vnet.c    |   2 +-
 drivers/vdpa/vdpa_sim/vdpa_sim.c     | 132 +++++++++++++-
 drivers/vdpa/vdpa_sim/vdpa_sim_blk.c |   6 +-
 drivers/vdpa/vdpa_sim/vdpa_sim_net.c |   6 +-
 drivers/vhost/vdpa.c                 |  22 +++
 drivers/vhost/vringh.c               | 250 +++++++++++++++++++++------
 10 files changed, 370 insertions(+), 71 deletions(-)

-- 
2.38.1

[RFC PATCH 1/6] vdpa: add bind_mm callback

[Stefano Garzarella]

This new optional callback is used to bind the device to a specific
address space so the vDPA framework can use VA when this callback
is implemented.

Suggested-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
---
 include/linux/vdpa.h | 8 ++++++++
 1 file changed, 8 insertions(+)

diff --git a/include/linux/vdpa.h b/include/linux/vdpa.h
index 6d0f5e4e82c2..34388e21ef3f 100644
--- a/include/linux/vdpa.h
+++ b/include/linux/vdpa.h
@@ -282,6 +282,12 @@ struct vdpa_map_file {
  *				@iova: iova to be unmapped
  *				@size: size of the area
  *				Returns integer: success (0) or error (< 0)
+ * @bind_mm:			Bind the device to a specific address space
+ *				so the vDPA framework can use VA when this
+ *				callback is implemented. (optional)
+ *				@vdev: vdpa device
+ *				@mm: address space to bind
+ *				@owner: process that owns the address space
  * @free:			Free resources that belongs to vDPA (optional)
  *				@vdev: vdpa device
  */
@@ -341,6 +347,8 @@ struct vdpa_config_ops {
 			 u64 iova, u64 size);
 	int (*set_group_asid)(struct vdpa_device *vdev, unsigned int group,
 			      unsigned int asid);
+	int (*bind_mm)(struct vdpa_device *vdev, struct mm_struct *mm,
+		       struct task_struct *owner);
 
 	/* Free device resources */
 	void (*free)(struct vdpa_device *vdev);
-- 
2.38.1

_______________________________________________
Virtualization mailing list
Virtualization@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

[RFC PATCH 1/6] vdpa: add bind_mm callback

[Stefano Garzarella]

This new optional callback is used to bind the device to a specific
address space so the vDPA framework can use VA when this callback
is implemented.

Suggested-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
---
 include/linux/vdpa.h | 8 ++++++++
 1 file changed, 8 insertions(+)

diff --git a/include/linux/vdpa.h b/include/linux/vdpa.h
index 6d0f5e4e82c2..34388e21ef3f 100644
--- a/include/linux/vdpa.h
+++ b/include/linux/vdpa.h
@@ -282,6 +282,12 @@ struct vdpa_map_file {
  *				@iova: iova to be unmapped
  *				@size: size of the area
  *				Returns integer: success (0) or error (< 0)
+ * @bind_mm:			Bind the device to a specific address space
+ *				so the vDPA framework can use VA when this
+ *				callback is implemented. (optional)
+ *				@vdev: vdpa device
+ *				@mm: address space to bind
+ *				@owner: process that owns the address space
  * @free:			Free resources that belongs to vDPA (optional)
  *				@vdev: vdpa device
  */
@@ -341,6 +347,8 @@ struct vdpa_config_ops {
 			 u64 iova, u64 size);
 	int (*set_group_asid)(struct vdpa_device *vdev, unsigned int group,
 			      unsigned int asid);
+	int (*bind_mm)(struct vdpa_device *vdev, struct mm_struct *mm,
+		       struct task_struct *owner);
 
 	/* Free device resources */
 	void (*free)(struct vdpa_device *vdev);
-- 
2.38.1

[RFC PATCH 2/6] vhost-vdpa: use bind_mm device callback

[Stefano Garzarella]

When the user call VHOST_SET_OWNER ioctl and the vDPA device
has `use_va` set to true, let's call the bind_mm callback.

In this way we can bind the device to the user address space
and directly use the user VA.

Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
---
 drivers/vhost/vdpa.c | 22 ++++++++++++++++++++++
 1 file changed, 22 insertions(+)

diff --git a/drivers/vhost/vdpa.c b/drivers/vhost/vdpa.c
index b08e07fc7d1f..a775d1a52c77 100644
--- a/drivers/vhost/vdpa.c
+++ b/drivers/vhost/vdpa.c
@@ -219,6 +219,17 @@ static int vhost_vdpa_reset(struct vhost_vdpa *v)
 	return vdpa_reset(vdpa);
 }
 
+static long vhost_vdpa_bind_mm(struct vhost_vdpa *v)
+{
+	struct vdpa_device *vdpa = v->vdpa;
+	const struct vdpa_config_ops *ops = vdpa->config;
+
+	if (!vdpa->use_va || !ops->bind_mm)
+		return 0;
+
+	return ops->bind_mm(vdpa, v->vdev.mm, current);
+}
+
 static long vhost_vdpa_get_device_id(struct vhost_vdpa *v, u8 __user *argp)
 {
 	struct vdpa_device *vdpa = v->vdpa;
@@ -276,6 +287,10 @@ static long vhost_vdpa_set_status(struct vhost_vdpa *v, u8 __user *statusp)
 		ret = vdpa_reset(vdpa);
 		if (ret)
 			return ret;
+
+		ret = vhost_vdpa_bind_mm(v);
+		if (ret)
+			return ret;
 	} else
 		vdpa_set_status(vdpa, status);
 
@@ -679,6 +694,13 @@ static long vhost_vdpa_unlocked_ioctl(struct file *filep,
 		break;
 	default:
 		r = vhost_dev_ioctl(&v->vdev, cmd, argp);
+		if (!r && cmd == VHOST_SET_OWNER) {
+			r = vhost_vdpa_bind_mm(v);
+			if (r) {
+				vhost_dev_reset_owner(&v->vdev, NULL);
+				break;
+			}
+		}
 		if (r == -ENOIOCTLCMD)
 			r = vhost_vdpa_vring_ioctl(v, cmd, argp);
 		break;
-- 
2.38.1

_______________________________________________
Virtualization mailing list
Virtualization@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

[RFC PATCH 2/6] vhost-vdpa: use bind_mm device callback

[Stefano Garzarella]

When the user call VHOST_SET_OWNER ioctl and the vDPA device
has `use_va` set to true, let's call the bind_mm callback.

In this way we can bind the device to the user address space
and directly use the user VA.

Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
---
 drivers/vhost/vdpa.c | 22 ++++++++++++++++++++++
 1 file changed, 22 insertions(+)

diff --git a/drivers/vhost/vdpa.c b/drivers/vhost/vdpa.c
index b08e07fc7d1f..a775d1a52c77 100644
--- a/drivers/vhost/vdpa.c
+++ b/drivers/vhost/vdpa.c
@@ -219,6 +219,17 @@ static int vhost_vdpa_reset(struct vhost_vdpa *v)
 	return vdpa_reset(vdpa);
 }
 
+static long vhost_vdpa_bind_mm(struct vhost_vdpa *v)
+{
+	struct vdpa_device *vdpa = v->vdpa;
+	const struct vdpa_config_ops *ops = vdpa->config;
+
+	if (!vdpa->use_va || !ops->bind_mm)
+		return 0;
+
+	return ops->bind_mm(vdpa, v->vdev.mm, current);
+}
+
 static long vhost_vdpa_get_device_id(struct vhost_vdpa *v, u8 __user *argp)
 {
 	struct vdpa_device *vdpa = v->vdpa;
@@ -276,6 +287,10 @@ static long vhost_vdpa_set_status(struct vhost_vdpa *v, u8 __user *statusp)
 		ret = vdpa_reset(vdpa);
 		if (ret)
 			return ret;
+
+		ret = vhost_vdpa_bind_mm(v);
+		if (ret)
+			return ret;
 	} else
 		vdpa_set_status(vdpa, status);
 
@@ -679,6 +694,13 @@ static long vhost_vdpa_unlocked_ioctl(struct file *filep,
 		break;
 	default:
 		r = vhost_dev_ioctl(&v->vdev, cmd, argp);
+		if (!r && cmd == VHOST_SET_OWNER) {
+			r = vhost_vdpa_bind_mm(v);
+			if (r) {
+				vhost_dev_reset_owner(&v->vdev, NULL);
+				break;
+			}
+		}
 		if (r == -ENOIOCTLCMD)
 			r = vhost_vdpa_vring_ioctl(v, cmd, argp);
 		break;
-- 
2.38.1

[RFC PATCH 3/6] vringh: support VA with iotlb

[Stefano Garzarella]

vDPA supports the possibility to use user VA in the iotlb messages.
So, let's add support for user VA in vringh to use it in the vDPA
simulators.

Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
---
 include/linux/vringh.h             |   5 +-
 drivers/vdpa/mlx5/core/resources.c |   3 +-
 drivers/vdpa/mlx5/net/mlx5_vnet.c  |   2 +-
 drivers/vdpa/vdpa_sim/vdpa_sim.c   |   4 +-
 drivers/vhost/vringh.c             | 250 +++++++++++++++++++++++------
 5 files changed, 207 insertions(+), 57 deletions(-)

diff --git a/include/linux/vringh.h b/include/linux/vringh.h
index 212892cf9822..c70962f16b1f 100644
--- a/include/linux/vringh.h
+++ b/include/linux/vringh.h
@@ -32,6 +32,9 @@ struct vringh {
 	/* Can we get away with weak barriers? */
 	bool weak_barriers;
 
+	/* Use user's VA */
+	bool use_va;
+
 	/* Last available index we saw (ie. where we're up to). */
 	u16 last_avail_idx;
 
@@ -279,7 +282,7 @@ void vringh_set_iotlb(struct vringh *vrh, struct vhost_iotlb *iotlb,
 		      spinlock_t *iotlb_lock);
 
 int vringh_init_iotlb(struct vringh *vrh, u64 features,
-		      unsigned int num, bool weak_barriers,
+		      unsigned int num, bool weak_barriers, bool use_va,
 		      struct vring_desc *desc,
 		      struct vring_avail *avail,
 		      struct vring_used *used);
diff --git a/drivers/vdpa/mlx5/core/resources.c b/drivers/vdpa/mlx5/core/resources.c
index 9800f9bec225..e0bab3458b40 100644
--- a/drivers/vdpa/mlx5/core/resources.c
+++ b/drivers/vdpa/mlx5/core/resources.c
@@ -233,7 +233,8 @@ static int init_ctrl_vq(struct mlx5_vdpa_dev *mvdev)
 	if (!mvdev->cvq.iotlb)
 		return -ENOMEM;
 
-	vringh_set_iotlb(&mvdev->cvq.vring, mvdev->cvq.iotlb, &mvdev->cvq.iommu_lock);
+	vringh_set_iotlb(&mvdev->cvq.vring, mvdev->cvq.iotlb,
+			 &mvdev->cvq.iommu_lock);
 
 	return 0;
 }
diff --git a/drivers/vdpa/mlx5/net/mlx5_vnet.c b/drivers/vdpa/mlx5/net/mlx5_vnet.c
index 90913365def4..81ba0867e2c8 100644
--- a/drivers/vdpa/mlx5/net/mlx5_vnet.c
+++ b/drivers/vdpa/mlx5/net/mlx5_vnet.c
@@ -2504,7 +2504,7 @@ static int setup_cvq_vring(struct mlx5_vdpa_dev *mvdev)
 
 	if (mvdev->actual_features & BIT_ULL(VIRTIO_NET_F_CTRL_VQ))
 		err = vringh_init_iotlb(&cvq->vring, mvdev->actual_features,
-					MLX5_CVQ_MAX_ENT, false,
+					MLX5_CVQ_MAX_ENT, false, false,
 					(struct vring_desc *)(uintptr_t)cvq->desc_addr,
 					(struct vring_avail *)(uintptr_t)cvq->driver_addr,
 					(struct vring_used *)(uintptr_t)cvq->device_addr);
diff --git a/drivers/vdpa/vdpa_sim/vdpa_sim.c b/drivers/vdpa/vdpa_sim/vdpa_sim.c
index b20689f8fe89..2e0ee7280aa8 100644
--- a/drivers/vdpa/vdpa_sim/vdpa_sim.c
+++ b/drivers/vdpa/vdpa_sim/vdpa_sim.c
@@ -67,7 +67,7 @@ static void vdpasim_queue_ready(struct vdpasim *vdpasim, unsigned int idx)
 {
 	struct vdpasim_virtqueue *vq = &vdpasim->vqs[idx];
 
-	vringh_init_iotlb(&vq->vring, vdpasim->features, vq->num, false,
+	vringh_init_iotlb(&vq->vring, vdpasim->features, vq->num, false, false,
 			  (struct vring_desc *)(uintptr_t)vq->desc_addr,
 			  (struct vring_avail *)
 			  (uintptr_t)vq->driver_addr,
@@ -87,7 +87,7 @@ static void vdpasim_vq_reset(struct vdpasim *vdpasim,
 	vq->cb = NULL;
 	vq->private = NULL;
 	vringh_init_iotlb(&vq->vring, vdpasim->dev_attr.supported_features,
-			  VDPASIM_QUEUE_MAX, false, NULL, NULL, NULL);
+			  VDPASIM_QUEUE_MAX, false, false, NULL, NULL, NULL);
 
 	vq->vring.notify = NULL;
 }
diff --git a/drivers/vhost/vringh.c b/drivers/vhost/vringh.c
index 11f59dd06a74..c1f77dc93482 100644
--- a/drivers/vhost/vringh.c
+++ b/drivers/vhost/vringh.c
@@ -1094,15 +1094,99 @@ EXPORT_SYMBOL(vringh_need_notify_kern);
 
 #if IS_REACHABLE(CONFIG_VHOST_IOTLB)
 
-static int iotlb_translate(const struct vringh *vrh,
-			   u64 addr, u64 len, u64 *translated,
-			   struct bio_vec iov[],
-			   int iov_size, u32 perm)
+static int iotlb_translate_va(const struct vringh *vrh,
+			      u64 addr, u64 len, u64 *translated,
+			      struct iovec iov[],
+			      int iov_size, u32 perm)
 {
 	struct vhost_iotlb_map *map;
 	struct vhost_iotlb *iotlb = vrh->iotlb;
+	u64 s = 0, last = addr + len - 1;
+	int ret = 0;
+
+	spin_lock(vrh->iotlb_lock);
+
+	while (len > s) {
+		u64 size;
+
+		if (unlikely(ret >= iov_size)) {
+			ret = -ENOBUFS;
+			break;
+		}
+
+		map = vhost_iotlb_itree_first(iotlb, addr, last);
+		if (!map || map->start > addr) {
+			ret = -EINVAL;
+			break;
+		} else if (!(map->perm & perm)) {
+			ret = -EPERM;
+			break;
+		}
+
+		size = map->size - addr + map->start;
+		iov[ret].iov_len = min(len - s, size);
+		iov[ret].iov_base = (void __user *)(unsigned long)
+				    (map->addr + addr - map->start);
+		s += size;
+		addr += size;
+		++ret;
+	}
+
+	spin_unlock(vrh->iotlb_lock);
+
+	if (translated)
+		*translated = min(len, s);
+
+	return ret;
+}
+
+static inline int copy_from_va(const struct vringh *vrh, void *dst, void *src,
+			       u64 len, u64 *translated)
+{
+	struct iovec iov[16];
+	struct iov_iter iter;
+	int ret;
+
+	ret = iotlb_translate_va(vrh, (u64)(uintptr_t)src, len, translated, iov,
+				 ARRAY_SIZE(iov), VHOST_MAP_RO);
+	if (ret == -ENOBUFS)
+		ret = ARRAY_SIZE(iov);
+	else if (ret < 0)
+		return ret;
+
+	iov_iter_init(&iter, READ, iov, ret, *translated);
+
+	return copy_from_iter(dst, *translated, &iter);
+}
+
+static inline int copy_to_va(const struct vringh *vrh, void *dst, void *src,
+			     u64 len, u64 *translated)
+{
+	struct iovec iov[16];
+	struct iov_iter iter;
+	int ret;
+
+	ret = iotlb_translate_va(vrh, (u64)(uintptr_t)dst, len, translated, iov,
+				 ARRAY_SIZE(iov), VHOST_MAP_WO);
+	if (ret == -ENOBUFS)
+		ret = ARRAY_SIZE(iov);
+	else if (ret < 0)
+		return ret;
+
+	iov_iter_init(&iter, WRITE, iov, ret, *translated);
+
+	return copy_to_iter(src, *translated, &iter);
+}
+
+static int iotlb_translate_pa(const struct vringh *vrh,
+			      u64 addr, u64 len, u64 *translated,
+			      struct bio_vec iov[],
+			      int iov_size, u32 perm)
+{
+	struct vhost_iotlb_map *map;
+	struct vhost_iotlb *iotlb = vrh->iotlb;
+	u64 s = 0, last = addr + len - 1;
 	int ret = 0;
-	u64 s = 0;
 
 	spin_lock(vrh->iotlb_lock);
 
@@ -1114,8 +1198,7 @@ static int iotlb_translate(const struct vringh *vrh,
 			break;
 		}
 
-		map = vhost_iotlb_itree_first(iotlb, addr,
-					      addr + len - 1);
+		map = vhost_iotlb_itree_first(iotlb, addr, last);
 		if (!map || map->start > addr) {
 			ret = -EINVAL;
 			break;
@@ -1143,28 +1226,61 @@ static int iotlb_translate(const struct vringh *vrh,
 	return ret;
 }
 
+static inline int copy_from_pa(const struct vringh *vrh, void *dst, void *src,
+			       u64 len, u64 *translated)
+{
+	struct bio_vec iov[16];
+	struct iov_iter iter;
+	int ret;
+
+	ret = iotlb_translate_pa(vrh, (u64)(uintptr_t)src, len, translated, iov,
+				 ARRAY_SIZE(iov), VHOST_MAP_RO);
+	if (ret == -ENOBUFS)
+		ret = ARRAY_SIZE(iov);
+	else if (ret < 0)
+		return ret;
+
+	iov_iter_bvec(&iter, READ, iov, ret, *translated);
+
+	return copy_from_iter(dst, *translated, &iter);
+}
+
+static inline int copy_to_pa(const struct vringh *vrh, void *dst, void *src,
+			     u64 len, u64 *translated)
+{
+	struct bio_vec iov[16];
+	struct iov_iter iter;
+	int ret;
+
+	ret = iotlb_translate_pa(vrh, (u64)(uintptr_t)dst, len, translated, iov,
+				 ARRAY_SIZE(iov), VHOST_MAP_WO);
+	if (ret == -ENOBUFS)
+		ret = ARRAY_SIZE(iov);
+	else if (ret < 0)
+		return ret;
+
+	iov_iter_bvec(&iter, WRITE, iov, ret, *translated);
+
+	return copy_to_iter(src, *translated, &iter);
+}
+
 static inline int copy_from_iotlb(const struct vringh *vrh, void *dst,
 				  void *src, size_t len)
 {
 	u64 total_translated = 0;
 
 	while (total_translated < len) {
-		struct bio_vec iov[16];
-		struct iov_iter iter;
 		u64 translated;
 		int ret;
 
-		ret = iotlb_translate(vrh, (u64)(uintptr_t)src,
-				      len - total_translated, &translated,
-				      iov, ARRAY_SIZE(iov), VHOST_MAP_RO);
-		if (ret == -ENOBUFS)
-			ret = ARRAY_SIZE(iov);
-		else if (ret < 0)
-			return ret;
-
-		iov_iter_bvec(&iter, READ, iov, ret, translated);
+		if (vrh->use_va) {
+			ret = copy_from_va(vrh, dst, src,
+					   len - total_translated, &translated);
+		} else {
+			ret = copy_from_pa(vrh, dst, src,
+					   len - total_translated, &translated);
+		}
 
-		ret = copy_from_iter(dst, translated, &iter);
 		if (ret < 0)
 			return ret;
 
@@ -1182,22 +1298,17 @@ static inline int copy_to_iotlb(const struct vringh *vrh, void *dst,
 	u64 total_translated = 0;
 
 	while (total_translated < len) {
-		struct bio_vec iov[16];
-		struct iov_iter iter;
 		u64 translated;
 		int ret;
 
-		ret = iotlb_translate(vrh, (u64)(uintptr_t)dst,
-				      len - total_translated, &translated,
-				      iov, ARRAY_SIZE(iov), VHOST_MAP_WO);
-		if (ret == -ENOBUFS)
-			ret = ARRAY_SIZE(iov);
-		else if (ret < 0)
-			return ret;
-
-		iov_iter_bvec(&iter, WRITE, iov, ret, translated);
+		if (vrh->use_va) {
+			ret = copy_to_va(vrh, dst, src,
+					 len - total_translated, &translated);
+		} else {
+			ret = copy_to_pa(vrh, dst, src,
+					 len - total_translated, &translated);
+		}
 
-		ret = copy_to_iter(src, translated, &iter);
 		if (ret < 0)
 			return ret;
 
@@ -1212,20 +1323,36 @@ static inline int copy_to_iotlb(const struct vringh *vrh, void *dst,
 static inline int getu16_iotlb(const struct vringh *vrh,
 			       u16 *val, const __virtio16 *p)
 {
-	struct bio_vec iov;
-	void *kaddr, *from;
 	int ret;
 
 	/* Atomic read is needed for getu16 */
-	ret = iotlb_translate(vrh, (u64)(uintptr_t)p, sizeof(*p), NULL,
-			      &iov, 1, VHOST_MAP_RO);
-	if (ret < 0)
-		return ret;
+	if (vrh->use_va) {
+		struct iovec iov;
+
+		ret = iotlb_translate_va(vrh, (u64)(uintptr_t)p, sizeof(*p),
+					 NULL, &iov, 1, VHOST_MAP_RO);
+		if (ret < 0)
+			return ret;
 
-	kaddr = kmap_atomic(iov.bv_page);
-	from = kaddr + iov.bv_offset;
-	*val = vringh16_to_cpu(vrh, READ_ONCE(*(__virtio16 *)from));
-	kunmap_atomic(kaddr);
+		ret = __get_user(*val, (__virtio16 *)iov.iov_base);
+		if (ret)
+			return ret;
+
+		*val = vringh16_to_cpu(vrh, *val);
+	} else {
+		struct bio_vec iov;
+		void *kaddr, *from;
+
+		ret = iotlb_translate_pa(vrh, (u64)(uintptr_t)p, sizeof(*p),
+					 NULL, &iov, 1, VHOST_MAP_RO);
+		if (ret < 0)
+			return ret;
+
+		kaddr = kmap_atomic(iov.bv_page);
+		from = kaddr + iov.bv_offset;
+		*val = vringh16_to_cpu(vrh, READ_ONCE(*(__virtio16 *)from));
+		kunmap_atomic(kaddr);
+	}
 
 	return 0;
 }
@@ -1233,20 +1360,36 @@ static inline int getu16_iotlb(const struct vringh *vrh,
 static inline int putu16_iotlb(const struct vringh *vrh,
 			       __virtio16 *p, u16 val)
 {
-	struct bio_vec iov;
-	void *kaddr, *to;
 	int ret;
 
 	/* Atomic write is needed for putu16 */
-	ret = iotlb_translate(vrh, (u64)(uintptr_t)p, sizeof(*p), NULL,
-			      &iov, 1, VHOST_MAP_WO);
-	if (ret < 0)
-		return ret;
+	if (vrh->use_va) {
+		struct iovec iov;
 
-	kaddr = kmap_atomic(iov.bv_page);
-	to = kaddr + iov.bv_offset;
-	WRITE_ONCE(*(__virtio16 *)to, cpu_to_vringh16(vrh, val));
-	kunmap_atomic(kaddr);
+		ret = iotlb_translate_va(vrh, (u64)(uintptr_t)p, sizeof(*p),
+					 NULL, &iov, 1, VHOST_MAP_RO);
+		if (ret < 0)
+			return ret;
+
+		val = cpu_to_vringh16(vrh, val);
+
+		ret = __put_user(val, (__virtio16 *)iov.iov_base);
+		if (ret)
+			return ret;
+	} else {
+		struct bio_vec iov;
+		void *kaddr, *to;
+
+		ret = iotlb_translate_pa(vrh, (u64)(uintptr_t)p, sizeof(*p), NULL,
+					 &iov, 1, VHOST_MAP_WO);
+		if (ret < 0)
+			return ret;
+
+		kaddr = kmap_atomic(iov.bv_page);
+		to = kaddr + iov.bv_offset;
+		WRITE_ONCE(*(__virtio16 *)to, cpu_to_vringh16(vrh, val));
+		kunmap_atomic(kaddr);
+	}
 
 	return 0;
 }
@@ -1308,6 +1451,7 @@ static inline int putused_iotlb(const struct vringh *vrh,
  * @features: the feature bits for this ring.
  * @num: the number of elements.
  * @weak_barriers: true if we only need memory barriers, not I/O.
+ * @use_va: true if IOTLB contains user VA
  * @desc: the userpace descriptor pointer.
  * @avail: the userpace avail pointer.
  * @used: the userpace used pointer.
@@ -1315,11 +1459,13 @@ static inline int putused_iotlb(const struct vringh *vrh,
  * Returns an error if num is invalid.
  */
 int vringh_init_iotlb(struct vringh *vrh, u64 features,
-		      unsigned int num, bool weak_barriers,
+		      unsigned int num, bool weak_barriers, bool use_va,
 		      struct vring_desc *desc,
 		      struct vring_avail *avail,
 		      struct vring_used *used)
 {
+	vrh->use_va = use_va;
+
 	return vringh_init_kern(vrh, features, num, weak_barriers,
 				desc, avail, used);
 }
-- 
2.38.1

_______________________________________________
Virtualization mailing list
Virtualization@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

[RFC PATCH 3/6] vringh: support VA with iotlb

[Stefano Garzarella]

vDPA supports the possibility to use user VA in the iotlb messages.
So, let's add support for user VA in vringh to use it in the vDPA
simulators.

Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
---
 include/linux/vringh.h             |   5 +-
 drivers/vdpa/mlx5/core/resources.c |   3 +-
 drivers/vdpa/mlx5/net/mlx5_vnet.c  |   2 +-
 drivers/vdpa/vdpa_sim/vdpa_sim.c   |   4 +-
 drivers/vhost/vringh.c             | 250 +++++++++++++++++++++++------
 5 files changed, 207 insertions(+), 57 deletions(-)

diff --git a/include/linux/vringh.h b/include/linux/vringh.h
index 212892cf9822..c70962f16b1f 100644
--- a/include/linux/vringh.h
+++ b/include/linux/vringh.h
@@ -32,6 +32,9 @@ struct vringh {
 	/* Can we get away with weak barriers? */
 	bool weak_barriers;
 
+	/* Use user's VA */
+	bool use_va;
+
 	/* Last available index we saw (ie. where we're up to). */
 	u16 last_avail_idx;
 
@@ -279,7 +282,7 @@ void vringh_set_iotlb(struct vringh *vrh, struct vhost_iotlb *iotlb,
 		      spinlock_t *iotlb_lock);
 
 int vringh_init_iotlb(struct vringh *vrh, u64 features,
-		      unsigned int num, bool weak_barriers,
+		      unsigned int num, bool weak_barriers, bool use_va,
 		      struct vring_desc *desc,
 		      struct vring_avail *avail,
 		      struct vring_used *used);
diff --git a/drivers/vdpa/mlx5/core/resources.c b/drivers/vdpa/mlx5/core/resources.c
index 9800f9bec225..e0bab3458b40 100644
--- a/drivers/vdpa/mlx5/core/resources.c
+++ b/drivers/vdpa/mlx5/core/resources.c
@@ -233,7 +233,8 @@ static int init_ctrl_vq(struct mlx5_vdpa_dev *mvdev)
 	if (!mvdev->cvq.iotlb)
 		return -ENOMEM;
 
-	vringh_set_iotlb(&mvdev->cvq.vring, mvdev->cvq.iotlb, &mvdev->cvq.iommu_lock);
+	vringh_set_iotlb(&mvdev->cvq.vring, mvdev->cvq.iotlb,
+			 &mvdev->cvq.iommu_lock);
 
 	return 0;
 }
diff --git a/drivers/vdpa/mlx5/net/mlx5_vnet.c b/drivers/vdpa/mlx5/net/mlx5_vnet.c
index 90913365def4..81ba0867e2c8 100644
--- a/drivers/vdpa/mlx5/net/mlx5_vnet.c
+++ b/drivers/vdpa/mlx5/net/mlx5_vnet.c
@@ -2504,7 +2504,7 @@ static int setup_cvq_vring(struct mlx5_vdpa_dev *mvdev)
 
 	if (mvdev->actual_features & BIT_ULL(VIRTIO_NET_F_CTRL_VQ))
 		err = vringh_init_iotlb(&cvq->vring, mvdev->actual_features,
-					MLX5_CVQ_MAX_ENT, false,
+					MLX5_CVQ_MAX_ENT, false, false,
 					(struct vring_desc *)(uintptr_t)cvq->desc_addr,
 					(struct vring_avail *)(uintptr_t)cvq->driver_addr,
 					(struct vring_used *)(uintptr_t)cvq->device_addr);
diff --git a/drivers/vdpa/vdpa_sim/vdpa_sim.c b/drivers/vdpa/vdpa_sim/vdpa_sim.c
index b20689f8fe89..2e0ee7280aa8 100644
--- a/drivers/vdpa/vdpa_sim/vdpa_sim.c
+++ b/drivers/vdpa/vdpa_sim/vdpa_sim.c
@@ -67,7 +67,7 @@ static void vdpasim_queue_ready(struct vdpasim *vdpasim, unsigned int idx)
 {
 	struct vdpasim_virtqueue *vq = &vdpasim->vqs[idx];
 
-	vringh_init_iotlb(&vq->vring, vdpasim->features, vq->num, false,
+	vringh_init_iotlb(&vq->vring, vdpasim->features, vq->num, false, false,
 			  (struct vring_desc *)(uintptr_t)vq->desc_addr,
 			  (struct vring_avail *)
 			  (uintptr_t)vq->driver_addr,
@@ -87,7 +87,7 @@ static void vdpasim_vq_reset(struct vdpasim *vdpasim,
 	vq->cb = NULL;
 	vq->private = NULL;
 	vringh_init_iotlb(&vq->vring, vdpasim->dev_attr.supported_features,
-			  VDPASIM_QUEUE_MAX, false, NULL, NULL, NULL);
+			  VDPASIM_QUEUE_MAX, false, false, NULL, NULL, NULL);
 
 	vq->vring.notify = NULL;
 }
diff --git a/drivers/vhost/vringh.c b/drivers/vhost/vringh.c
index 11f59dd06a74..c1f77dc93482 100644
--- a/drivers/vhost/vringh.c
+++ b/drivers/vhost/vringh.c
@@ -1094,15 +1094,99 @@ EXPORT_SYMBOL(vringh_need_notify_kern);
 
 #if IS_REACHABLE(CONFIG_VHOST_IOTLB)
 
-static int iotlb_translate(const struct vringh *vrh,
-			   u64 addr, u64 len, u64 *translated,
-			   struct bio_vec iov[],
-			   int iov_size, u32 perm)
+static int iotlb_translate_va(const struct vringh *vrh,
+			      u64 addr, u64 len, u64 *translated,
+			      struct iovec iov[],
+			      int iov_size, u32 perm)
 {
 	struct vhost_iotlb_map *map;
 	struct vhost_iotlb *iotlb = vrh->iotlb;
+	u64 s = 0, last = addr + len - 1;
+	int ret = 0;
+
+	spin_lock(vrh->iotlb_lock);
+
+	while (len > s) {
+		u64 size;
+
+		if (unlikely(ret >= iov_size)) {
+			ret = -ENOBUFS;
+			break;
+		}
+
+		map = vhost_iotlb_itree_first(iotlb, addr, last);
+		if (!map || map->start > addr) {
+			ret = -EINVAL;
+			break;
+		} else if (!(map->perm & perm)) {
+			ret = -EPERM;
+			break;
+		}
+
+		size = map->size - addr + map->start;
+		iov[ret].iov_len = min(len - s, size);
+		iov[ret].iov_base = (void __user *)(unsigned long)
+				    (map->addr + addr - map->start);
+		s += size;
+		addr += size;
+		++ret;
+	}
+
+	spin_unlock(vrh->iotlb_lock);
+
+	if (translated)
+		*translated = min(len, s);
+
+	return ret;
+}
+
+static inline int copy_from_va(const struct vringh *vrh, void *dst, void *src,
+			       u64 len, u64 *translated)
+{
+	struct iovec iov[16];
+	struct iov_iter iter;
+	int ret;
+
+	ret = iotlb_translate_va(vrh, (u64)(uintptr_t)src, len, translated, iov,
+				 ARRAY_SIZE(iov), VHOST_MAP_RO);
+	if (ret == -ENOBUFS)
+		ret = ARRAY_SIZE(iov);
+	else if (ret < 0)
+		return ret;
+
+	iov_iter_init(&iter, READ, iov, ret, *translated);
+
+	return copy_from_iter(dst, *translated, &iter);
+}
+
+static inline int copy_to_va(const struct vringh *vrh, void *dst, void *src,
+			     u64 len, u64 *translated)
+{
+	struct iovec iov[16];
+	struct iov_iter iter;
+	int ret;
+
+	ret = iotlb_translate_va(vrh, (u64)(uintptr_t)dst, len, translated, iov,
+				 ARRAY_SIZE(iov), VHOST_MAP_WO);
+	if (ret == -ENOBUFS)
+		ret = ARRAY_SIZE(iov);
+	else if (ret < 0)
+		return ret;
+
+	iov_iter_init(&iter, WRITE, iov, ret, *translated);
+
+	return copy_to_iter(src, *translated, &iter);
+}
+
+static int iotlb_translate_pa(const struct vringh *vrh,
+			      u64 addr, u64 len, u64 *translated,
+			      struct bio_vec iov[],
+			      int iov_size, u32 perm)
+{
+	struct vhost_iotlb_map *map;
+	struct vhost_iotlb *iotlb = vrh->iotlb;
+	u64 s = 0, last = addr + len - 1;
 	int ret = 0;
-	u64 s = 0;
 
 	spin_lock(vrh->iotlb_lock);
 
@@ -1114,8 +1198,7 @@ static int iotlb_translate(const struct vringh *vrh,
 			break;
 		}
 
-		map = vhost_iotlb_itree_first(iotlb, addr,
-					      addr + len - 1);
+		map = vhost_iotlb_itree_first(iotlb, addr, last);
 		if (!map || map->start > addr) {
 			ret = -EINVAL;
 			break;
@@ -1143,28 +1226,61 @@ static int iotlb_translate(const struct vringh *vrh,
 	return ret;
 }
 
+static inline int copy_from_pa(const struct vringh *vrh, void *dst, void *src,
+			       u64 len, u64 *translated)
+{
+	struct bio_vec iov[16];
+	struct iov_iter iter;
+	int ret;
+
+	ret = iotlb_translate_pa(vrh, (u64)(uintptr_t)src, len, translated, iov,
+				 ARRAY_SIZE(iov), VHOST_MAP_RO);
+	if (ret == -ENOBUFS)
+		ret = ARRAY_SIZE(iov);
+	else if (ret < 0)
+		return ret;
+
+	iov_iter_bvec(&iter, READ, iov, ret, *translated);
+
+	return copy_from_iter(dst, *translated, &iter);
+}
+
+static inline int copy_to_pa(const struct vringh *vrh, void *dst, void *src,
+			     u64 len, u64 *translated)
+{
+	struct bio_vec iov[16];
+	struct iov_iter iter;
+	int ret;
+
+	ret = iotlb_translate_pa(vrh, (u64)(uintptr_t)dst, len, translated, iov,
+				 ARRAY_SIZE(iov), VHOST_MAP_WO);
+	if (ret == -ENOBUFS)
+		ret = ARRAY_SIZE(iov);
+	else if (ret < 0)
+		return ret;
+
+	iov_iter_bvec(&iter, WRITE, iov, ret, *translated);
+
+	return copy_to_iter(src, *translated, &iter);
+}
+
 static inline int copy_from_iotlb(const struct vringh *vrh, void *dst,
 				  void *src, size_t len)
 {
 	u64 total_translated = 0;
 
 	while (total_translated < len) {
-		struct bio_vec iov[16];
-		struct iov_iter iter;
 		u64 translated;
 		int ret;
 
-		ret = iotlb_translate(vrh, (u64)(uintptr_t)src,
-				      len - total_translated, &translated,
-				      iov, ARRAY_SIZE(iov), VHOST_MAP_RO);
-		if (ret == -ENOBUFS)
-			ret = ARRAY_SIZE(iov);
-		else if (ret < 0)
-			return ret;
-
-		iov_iter_bvec(&iter, READ, iov, ret, translated);
+		if (vrh->use_va) {
+			ret = copy_from_va(vrh, dst, src,
+					   len - total_translated, &translated);
+		} else {
+			ret = copy_from_pa(vrh, dst, src,
+					   len - total_translated, &translated);
+		}
 
-		ret = copy_from_iter(dst, translated, &iter);
 		if (ret < 0)
 			return ret;
 
@@ -1182,22 +1298,17 @@ static inline int copy_to_iotlb(const struct vringh *vrh, void *dst,
 	u64 total_translated = 0;
 
 	while (total_translated < len) {
-		struct bio_vec iov[16];
-		struct iov_iter iter;
 		u64 translated;
 		int ret;
 
-		ret = iotlb_translate(vrh, (u64)(uintptr_t)dst,
-				      len - total_translated, &translated,
-				      iov, ARRAY_SIZE(iov), VHOST_MAP_WO);
-		if (ret == -ENOBUFS)
-			ret = ARRAY_SIZE(iov);
-		else if (ret < 0)
-			return ret;
-
-		iov_iter_bvec(&iter, WRITE, iov, ret, translated);
+		if (vrh->use_va) {
+			ret = copy_to_va(vrh, dst, src,
+					 len - total_translated, &translated);
+		} else {
+			ret = copy_to_pa(vrh, dst, src,
+					 len - total_translated, &translated);
+		}
 
-		ret = copy_to_iter(src, translated, &iter);
 		if (ret < 0)
 			return ret;
 
@@ -1212,20 +1323,36 @@ static inline int copy_to_iotlb(const struct vringh *vrh, void *dst,
 static inline int getu16_iotlb(const struct vringh *vrh,
 			       u16 *val, const __virtio16 *p)
 {
-	struct bio_vec iov;
-	void *kaddr, *from;
 	int ret;
 
 	/* Atomic read is needed for getu16 */
-	ret = iotlb_translate(vrh, (u64)(uintptr_t)p, sizeof(*p), NULL,
-			      &iov, 1, VHOST_MAP_RO);
-	if (ret < 0)
-		return ret;
+	if (vrh->use_va) {
+		struct iovec iov;
+
+		ret = iotlb_translate_va(vrh, (u64)(uintptr_t)p, sizeof(*p),
+					 NULL, &iov, 1, VHOST_MAP_RO);
+		if (ret < 0)
+			return ret;
 
-	kaddr = kmap_atomic(iov.bv_page);
-	from = kaddr + iov.bv_offset;
-	*val = vringh16_to_cpu(vrh, READ_ONCE(*(__virtio16 *)from));
-	kunmap_atomic(kaddr);
+		ret = __get_user(*val, (__virtio16 *)iov.iov_base);
+		if (ret)
+			return ret;
+
+		*val = vringh16_to_cpu(vrh, *val);
+	} else {
+		struct bio_vec iov;
+		void *kaddr, *from;
+
+		ret = iotlb_translate_pa(vrh, (u64)(uintptr_t)p, sizeof(*p),
+					 NULL, &iov, 1, VHOST_MAP_RO);
+		if (ret < 0)
+			return ret;
+
+		kaddr = kmap_atomic(iov.bv_page);
+		from = kaddr + iov.bv_offset;
+		*val = vringh16_to_cpu(vrh, READ_ONCE(*(__virtio16 *)from));
+		kunmap_atomic(kaddr);
+	}
 
 	return 0;
 }
@@ -1233,20 +1360,36 @@ static inline int getu16_iotlb(const struct vringh *vrh,
 static inline int putu16_iotlb(const struct vringh *vrh,
 			       __virtio16 *p, u16 val)
 {
-	struct bio_vec iov;
-	void *kaddr, *to;
 	int ret;
 
 	/* Atomic write is needed for putu16 */
-	ret = iotlb_translate(vrh, (u64)(uintptr_t)p, sizeof(*p), NULL,
-			      &iov, 1, VHOST_MAP_WO);
-	if (ret < 0)
-		return ret;
+	if (vrh->use_va) {
+		struct iovec iov;
 
-	kaddr = kmap_atomic(iov.bv_page);
-	to = kaddr + iov.bv_offset;
-	WRITE_ONCE(*(__virtio16 *)to, cpu_to_vringh16(vrh, val));
-	kunmap_atomic(kaddr);
+		ret = iotlb_translate_va(vrh, (u64)(uintptr_t)p, sizeof(*p),
+					 NULL, &iov, 1, VHOST_MAP_RO);
+		if (ret < 0)
+			return ret;
+
+		val = cpu_to_vringh16(vrh, val);
+
+		ret = __put_user(val, (__virtio16 *)iov.iov_base);
+		if (ret)
+			return ret;
+	} else {
+		struct bio_vec iov;
+		void *kaddr, *to;
+
+		ret = iotlb_translate_pa(vrh, (u64)(uintptr_t)p, sizeof(*p), NULL,
+					 &iov, 1, VHOST_MAP_WO);
+		if (ret < 0)
+			return ret;
+
+		kaddr = kmap_atomic(iov.bv_page);
+		to = kaddr + iov.bv_offset;
+		WRITE_ONCE(*(__virtio16 *)to, cpu_to_vringh16(vrh, val));
+		kunmap_atomic(kaddr);
+	}
 
 	return 0;
 }
@@ -1308,6 +1451,7 @@ static inline int putused_iotlb(const struct vringh *vrh,
  * @features: the feature bits for this ring.
  * @num: the number of elements.
  * @weak_barriers: true if we only need memory barriers, not I/O.
+ * @use_va: true if IOTLB contains user VA
  * @desc: the userpace descriptor pointer.
  * @avail: the userpace avail pointer.
  * @used: the userpace used pointer.
@@ -1315,11 +1459,13 @@ static inline int putused_iotlb(const struct vringh *vrh,
  * Returns an error if num is invalid.
  */
 int vringh_init_iotlb(struct vringh *vrh, u64 features,
-		      unsigned int num, bool weak_barriers,
+		      unsigned int num, bool weak_barriers, bool use_va,
 		      struct vring_desc *desc,
 		      struct vring_avail *avail,
 		      struct vring_used *used)
 {
+	vrh->use_va = use_va;
+
 	return vringh_init_kern(vrh, features, num, weak_barriers,
 				desc, avail, used);
 }
-- 
2.38.1

[RFC PATCH 4/6] vdpa_sim: make devices agnostic for work management

[Stefano Garzarella]

Let's move work management inside the vdpa_sim core.
This way we can easily change how we manage the works, without
having to change the devices each time.

Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
---
 drivers/vdpa/vdpa_sim/vdpa_sim.h     |  3 ++-
 drivers/vdpa/vdpa_sim/vdpa_sim.c     | 17 +++++++++++++++--
 drivers/vdpa/vdpa_sim/vdpa_sim_blk.c |  6 ++----
 drivers/vdpa/vdpa_sim/vdpa_sim_net.c |  6 ++----
 4 files changed, 21 insertions(+), 11 deletions(-)

diff --git a/drivers/vdpa/vdpa_sim/vdpa_sim.h b/drivers/vdpa/vdpa_sim/vdpa_sim.h
index 0e78737dcc16..7e6dd366856f 100644
--- a/drivers/vdpa/vdpa_sim/vdpa_sim.h
+++ b/drivers/vdpa/vdpa_sim/vdpa_sim.h
@@ -44,7 +44,7 @@ struct vdpasim_dev_attr {
 	u32 ngroups;
 	u32 nas;
 
-	work_func_t work_fn;
+	void (*work_fn)(struct vdpasim *vdpasim);
 	void (*get_config)(struct vdpasim *vdpasim, void *config);
 	void (*set_config)(struct vdpasim *vdpasim, const void *config);
 };
@@ -73,6 +73,7 @@ struct vdpasim {
 
 struct vdpasim *vdpasim_create(struct vdpasim_dev_attr *attr,
 			       const struct vdpa_dev_set_config *config);
+void vdpasim_schedule_work(struct vdpasim *vdpasim);
 
 /* TODO: cross-endian support */
 static inline bool vdpasim_is_little_endian(struct vdpasim *vdpasim)
diff --git a/drivers/vdpa/vdpa_sim/vdpa_sim.c b/drivers/vdpa/vdpa_sim/vdpa_sim.c
index 2e0ee7280aa8..9bde33e38e27 100644
--- a/drivers/vdpa/vdpa_sim/vdpa_sim.c
+++ b/drivers/vdpa/vdpa_sim/vdpa_sim.c
@@ -245,6 +245,13 @@ static const struct dma_map_ops vdpasim_dma_ops = {
 static const struct vdpa_config_ops vdpasim_config_ops;
 static const struct vdpa_config_ops vdpasim_batch_config_ops;
 
+static void vdpasim_work_fn(struct work_struct *work)
+{
+	struct vdpasim *vdpasim = container_of(work, struct vdpasim, work);
+
+	vdpasim->dev_attr.work_fn(vdpasim);
+}
+
 struct vdpasim *vdpasim_create(struct vdpasim_dev_attr *dev_attr,
 			       const struct vdpa_dev_set_config *config)
 {
@@ -275,7 +282,7 @@ struct vdpasim *vdpasim_create(struct vdpasim_dev_attr *dev_attr,
 	}
 
 	vdpasim->dev_attr = *dev_attr;
-	INIT_WORK(&vdpasim->work, dev_attr->work_fn);
+	INIT_WORK(&vdpasim->work, vdpasim_work_fn);
 	spin_lock_init(&vdpasim->lock);
 	spin_lock_init(&vdpasim->iommu_lock);
 
@@ -329,6 +336,12 @@ struct vdpasim *vdpasim_create(struct vdpasim_dev_attr *dev_attr,
 }
 EXPORT_SYMBOL_GPL(vdpasim_create);
 
+void vdpasim_schedule_work(struct vdpasim *vdpasim)
+{
+	schedule_work(&vdpasim->work);
+}
+EXPORT_SYMBOL_GPL(vdpasim_schedule_work);
+
 static int vdpasim_set_vq_address(struct vdpa_device *vdpa, u16 idx,
 				  u64 desc_area, u64 driver_area,
 				  u64 device_area)
@@ -357,7 +370,7 @@ static void vdpasim_kick_vq(struct vdpa_device *vdpa, u16 idx)
 	struct vdpasim_virtqueue *vq = &vdpasim->vqs[idx];
 
 	if (vq->ready)
-		schedule_work(&vdpasim->work);
+		vdpasim_schedule_work(vdpasim);
 }
 
 static void vdpasim_set_vq_cb(struct vdpa_device *vdpa, u16 idx,
diff --git a/drivers/vdpa/vdpa_sim/vdpa_sim_blk.c b/drivers/vdpa/vdpa_sim/vdpa_sim_blk.c
index c6db1a1baf76..ae2309411acd 100644
--- a/drivers/vdpa/vdpa_sim/vdpa_sim_blk.c
+++ b/drivers/vdpa/vdpa_sim/vdpa_sim_blk.c
@@ -11,7 +11,6 @@
 #include <linux/module.h>
 #include <linux/device.h>
 #include <linux/kernel.h>
-#include <linux/sched.h>
 #include <linux/blkdev.h>
 #include <linux/vringh.h>
 #include <linux/vdpa.h>
@@ -286,9 +285,8 @@ static bool vdpasim_blk_handle_req(struct vdpasim *vdpasim,
 	return handled;
 }
 
-static void vdpasim_blk_work(struct work_struct *work)
+static void vdpasim_blk_work(struct vdpasim *vdpasim)
 {
-	struct vdpasim *vdpasim = container_of(work, struct vdpasim, work);
 	bool reschedule = false;
 	int i;
 
@@ -326,7 +324,7 @@ static void vdpasim_blk_work(struct work_struct *work)
 	spin_unlock(&vdpasim->lock);
 
 	if (reschedule)
-		schedule_work(&vdpasim->work);
+		vdpasim_schedule_work(vdpasim);
 }
 
 static void vdpasim_blk_get_config(struct vdpasim *vdpasim, void *config)
diff --git a/drivers/vdpa/vdpa_sim/vdpa_sim_net.c b/drivers/vdpa/vdpa_sim/vdpa_sim_net.c
index c3cb225ea469..a209df365158 100644
--- a/drivers/vdpa/vdpa_sim/vdpa_sim_net.c
+++ b/drivers/vdpa/vdpa_sim/vdpa_sim_net.c
@@ -11,7 +11,6 @@
 #include <linux/module.h>
 #include <linux/device.h>
 #include <linux/kernel.h>
-#include <linux/sched.h>
 #include <linux/etherdevice.h>
 #include <linux/vringh.h>
 #include <linux/vdpa.h>
@@ -143,9 +142,8 @@ static void vdpasim_handle_cvq(struct vdpasim *vdpasim)
 	}
 }
 
-static void vdpasim_net_work(struct work_struct *work)
+static void vdpasim_net_work(struct vdpasim *vdpasim)
 {
-	struct vdpasim *vdpasim = container_of(work, struct vdpasim, work);
 	struct vdpasim_virtqueue *txq = &vdpasim->vqs[1];
 	struct vdpasim_virtqueue *rxq = &vdpasim->vqs[0];
 	ssize_t read, write;
@@ -196,7 +194,7 @@ static void vdpasim_net_work(struct work_struct *work)
 		vdpasim_net_complete(rxq, write);
 
 		if (++pkts > 4) {
-			schedule_work(&vdpasim->work);
+			vdpasim_schedule_work(vdpasim);
 			goto out;
 		}
 	}
-- 
2.38.1

_______________________________________________
Virtualization mailing list
Virtualization@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

[RFC PATCH 4/6] vdpa_sim: make devices agnostic for work management

[Stefano Garzarella]

Let's move work management inside the vdpa_sim core.
This way we can easily change how we manage the works, without
having to change the devices each time.

Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
---
 drivers/vdpa/vdpa_sim/vdpa_sim.h     |  3 ++-
 drivers/vdpa/vdpa_sim/vdpa_sim.c     | 17 +++++++++++++++--
 drivers/vdpa/vdpa_sim/vdpa_sim_blk.c |  6 ++----
 drivers/vdpa/vdpa_sim/vdpa_sim_net.c |  6 ++----
 4 files changed, 21 insertions(+), 11 deletions(-)

diff --git a/drivers/vdpa/vdpa_sim/vdpa_sim.h b/drivers/vdpa/vdpa_sim/vdpa_sim.h
index 0e78737dcc16..7e6dd366856f 100644
--- a/drivers/vdpa/vdpa_sim/vdpa_sim.h
+++ b/drivers/vdpa/vdpa_sim/vdpa_sim.h
@@ -44,7 +44,7 @@ struct vdpasim_dev_attr {
 	u32 ngroups;
 	u32 nas;
 
-	work_func_t work_fn;
+	void (*work_fn)(struct vdpasim *vdpasim);
 	void (*get_config)(struct vdpasim *vdpasim, void *config);
 	void (*set_config)(struct vdpasim *vdpasim, const void *config);
 };
@@ -73,6 +73,7 @@ struct vdpasim {
 
 struct vdpasim *vdpasim_create(struct vdpasim_dev_attr *attr,
 			       const struct vdpa_dev_set_config *config);
+void vdpasim_schedule_work(struct vdpasim *vdpasim);
 
 /* TODO: cross-endian support */
 static inline bool vdpasim_is_little_endian(struct vdpasim *vdpasim)
diff --git a/drivers/vdpa/vdpa_sim/vdpa_sim.c b/drivers/vdpa/vdpa_sim/vdpa_sim.c
index 2e0ee7280aa8..9bde33e38e27 100644
--- a/drivers/vdpa/vdpa_sim/vdpa_sim.c
+++ b/drivers/vdpa/vdpa_sim/vdpa_sim.c
@@ -245,6 +245,13 @@ static const struct dma_map_ops vdpasim_dma_ops = {
 static const struct vdpa_config_ops vdpasim_config_ops;
 static const struct vdpa_config_ops vdpasim_batch_config_ops;
 
+static void vdpasim_work_fn(struct work_struct *work)
+{
+	struct vdpasim *vdpasim = container_of(work, struct vdpasim, work);
+
+	vdpasim->dev_attr.work_fn(vdpasim);
+}
+
 struct vdpasim *vdpasim_create(struct vdpasim_dev_attr *dev_attr,
 			       const struct vdpa_dev_set_config *config)
 {
@@ -275,7 +282,7 @@ struct vdpasim *vdpasim_create(struct vdpasim_dev_attr *dev_attr,
 	}
 
 	vdpasim->dev_attr = *dev_attr;
-	INIT_WORK(&vdpasim->work, dev_attr->work_fn);
+	INIT_WORK(&vdpasim->work, vdpasim_work_fn);
 	spin_lock_init(&vdpasim->lock);
 	spin_lock_init(&vdpasim->iommu_lock);
 
@@ -329,6 +336,12 @@ struct vdpasim *vdpasim_create(struct vdpasim_dev_attr *dev_attr,
 }
 EXPORT_SYMBOL_GPL(vdpasim_create);
 
+void vdpasim_schedule_work(struct vdpasim *vdpasim)
+{
+	schedule_work(&vdpasim->work);
+}
+EXPORT_SYMBOL_GPL(vdpasim_schedule_work);
+
 static int vdpasim_set_vq_address(struct vdpa_device *vdpa, u16 idx,
 				  u64 desc_area, u64 driver_area,
 				  u64 device_area)
@@ -357,7 +370,7 @@ static void vdpasim_kick_vq(struct vdpa_device *vdpa, u16 idx)
 	struct vdpasim_virtqueue *vq = &vdpasim->vqs[idx];
 
 	if (vq->ready)
-		schedule_work(&vdpasim->work);
+		vdpasim_schedule_work(vdpasim);
 }
 
 static void vdpasim_set_vq_cb(struct vdpa_device *vdpa, u16 idx,
diff --git a/drivers/vdpa/vdpa_sim/vdpa_sim_blk.c b/drivers/vdpa/vdpa_sim/vdpa_sim_blk.c
index c6db1a1baf76..ae2309411acd 100644
--- a/drivers/vdpa/vdpa_sim/vdpa_sim_blk.c
+++ b/drivers/vdpa/vdpa_sim/vdpa_sim_blk.c
@@ -11,7 +11,6 @@
 #include <linux/module.h>
 #include <linux/device.h>
 #include <linux/kernel.h>
-#include <linux/sched.h>
 #include <linux/blkdev.h>
 #include <linux/vringh.h>
 #include <linux/vdpa.h>
@@ -286,9 +285,8 @@ static bool vdpasim_blk_handle_req(struct vdpasim *vdpasim,
 	return handled;
 }
 
-static void vdpasim_blk_work(struct work_struct *work)
+static void vdpasim_blk_work(struct vdpasim *vdpasim)
 {
-	struct vdpasim *vdpasim = container_of(work, struct vdpasim, work);
 	bool reschedule = false;
 	int i;
 
@@ -326,7 +324,7 @@ static void vdpasim_blk_work(struct work_struct *work)
 	spin_unlock(&vdpasim->lock);
 
 	if (reschedule)
-		schedule_work(&vdpasim->work);
+		vdpasim_schedule_work(vdpasim);
 }
 
 static void vdpasim_blk_get_config(struct vdpasim *vdpasim, void *config)
diff --git a/drivers/vdpa/vdpa_sim/vdpa_sim_net.c b/drivers/vdpa/vdpa_sim/vdpa_sim_net.c
index c3cb225ea469..a209df365158 100644
--- a/drivers/vdpa/vdpa_sim/vdpa_sim_net.c
+++ b/drivers/vdpa/vdpa_sim/vdpa_sim_net.c
@@ -11,7 +11,6 @@
 #include <linux/module.h>
 #include <linux/device.h>
 #include <linux/kernel.h>
-#include <linux/sched.h>
 #include <linux/etherdevice.h>
 #include <linux/vringh.h>
 #include <linux/vdpa.h>
@@ -143,9 +142,8 @@ static void vdpasim_handle_cvq(struct vdpasim *vdpasim)
 	}
 }
 
-static void vdpasim_net_work(struct work_struct *work)
+static void vdpasim_net_work(struct vdpasim *vdpasim)
 {
-	struct vdpasim *vdpasim = container_of(work, struct vdpasim, work);
 	struct vdpasim_virtqueue *txq = &vdpasim->vqs[1];
 	struct vdpasim_virtqueue *rxq = &vdpasim->vqs[0];
 	ssize_t read, write;
@@ -196,7 +194,7 @@ static void vdpasim_net_work(struct work_struct *work)
 		vdpasim_net_complete(rxq, write);
 
 		if (++pkts > 4) {
-			schedule_work(&vdpasim->work);
+			vdpasim_schedule_work(vdpasim);
 			goto out;
 		}
 	}
-- 
2.38.1

[RFC PATCH 5/6] vdpa_sim: use kthread worker

[Stefano Garzarella]

Let's use our own kthread to run device jobs.
This allows us more flexibility, especially we can attach the kthread
to the user address space when vDPA uses user's VA.

Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
---
 drivers/vdpa/vdpa_sim/vdpa_sim.h |  3 ++-
 drivers/vdpa/vdpa_sim/vdpa_sim.c | 17 ++++++++++++-----
 2 files changed, 14 insertions(+), 6 deletions(-)

diff --git a/drivers/vdpa/vdpa_sim/vdpa_sim.h b/drivers/vdpa/vdpa_sim/vdpa_sim.h
index 7e6dd366856f..07ef53ea375e 100644
--- a/drivers/vdpa/vdpa_sim/vdpa_sim.h
+++ b/drivers/vdpa/vdpa_sim/vdpa_sim.h
@@ -53,7 +53,8 @@ struct vdpasim_dev_attr {
 struct vdpasim {
 	struct vdpa_device vdpa;
 	struct vdpasim_virtqueue *vqs;
-	struct work_struct work;
+	struct kthread_worker *worker;
+	struct kthread_work work;
 	struct vdpasim_dev_attr dev_attr;
 	/* spinlock to synchronize virtqueue state */
 	spinlock_t lock;
diff --git a/drivers/vdpa/vdpa_sim/vdpa_sim.c b/drivers/vdpa/vdpa_sim/vdpa_sim.c
index 9bde33e38e27..36a1d2e0a6ba 100644
--- a/drivers/vdpa/vdpa_sim/vdpa_sim.c
+++ b/drivers/vdpa/vdpa_sim/vdpa_sim.c
@@ -11,8 +11,8 @@
 #include <linux/module.h>
 #include <linux/device.h>
 #include <linux/kernel.h>
+#include <linux/kthread.h>
 #include <linux/slab.h>
-#include <linux/sched.h>
 #include <linux/dma-map-ops.h>
 #include <linux/vringh.h>
 #include <linux/vdpa.h>
@@ -245,7 +245,7 @@ static const struct dma_map_ops vdpasim_dma_ops = {
 static const struct vdpa_config_ops vdpasim_config_ops;
 static const struct vdpa_config_ops vdpasim_batch_config_ops;
 
-static void vdpasim_work_fn(struct work_struct *work)
+static void vdpasim_work_fn(struct kthread_work *work)
 {
 	struct vdpasim *vdpasim = container_of(work, struct vdpasim, work);
 
@@ -282,7 +282,13 @@ struct vdpasim *vdpasim_create(struct vdpasim_dev_attr *dev_attr,
 	}
 
 	vdpasim->dev_attr = *dev_attr;
-	INIT_WORK(&vdpasim->work, vdpasim_work_fn);
+
+	kthread_init_work(&vdpasim->work, vdpasim_work_fn);
+	vdpasim->worker = kthread_create_worker(0, "vDPA sim worker: %s",
+						dev_attr->name);
+	if (IS_ERR(vdpasim->worker))
+		goto err_iommu;
+
 	spin_lock_init(&vdpasim->lock);
 	spin_lock_init(&vdpasim->iommu_lock);
 
@@ -338,7 +344,7 @@ EXPORT_SYMBOL_GPL(vdpasim_create);
 
 void vdpasim_schedule_work(struct vdpasim *vdpasim)
 {
-	schedule_work(&vdpasim->work);
+	kthread_queue_work(vdpasim->worker, &vdpasim->work);
 }
 EXPORT_SYMBOL_GPL(vdpasim_schedule_work);
 
@@ -689,7 +695,8 @@ static void vdpasim_free(struct vdpa_device *vdpa)
 	struct vdpasim *vdpasim = vdpa_to_sim(vdpa);
 	int i;
 
-	cancel_work_sync(&vdpasim->work);
+	kthread_cancel_work_sync(&vdpasim->work);
+	kthread_destroy_worker(vdpasim->worker);
 
 	for (i = 0; i < vdpasim->dev_attr.nvqs; i++) {
 		vringh_kiov_cleanup(&vdpasim->vqs[i].out_iov);
-- 
2.38.1

_______________________________________________
Virtualization mailing list
Virtualization@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

[RFC PATCH 5/6] vdpa_sim: use kthread worker

[Stefano Garzarella]

Let's use our own kthread to run device jobs.
This allows us more flexibility, especially we can attach the kthread
to the user address space when vDPA uses user's VA.

Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
---
 drivers/vdpa/vdpa_sim/vdpa_sim.h |  3 ++-
 drivers/vdpa/vdpa_sim/vdpa_sim.c | 17 ++++++++++++-----
 2 files changed, 14 insertions(+), 6 deletions(-)

diff --git a/drivers/vdpa/vdpa_sim/vdpa_sim.h b/drivers/vdpa/vdpa_sim/vdpa_sim.h
index 7e6dd366856f..07ef53ea375e 100644
--- a/drivers/vdpa/vdpa_sim/vdpa_sim.h
+++ b/drivers/vdpa/vdpa_sim/vdpa_sim.h
@@ -53,7 +53,8 @@ struct vdpasim_dev_attr {
 struct vdpasim {
 	struct vdpa_device vdpa;
 	struct vdpasim_virtqueue *vqs;
-	struct work_struct work;
+	struct kthread_worker *worker;
+	struct kthread_work work;
 	struct vdpasim_dev_attr dev_attr;
 	/* spinlock to synchronize virtqueue state */
 	spinlock_t lock;
diff --git a/drivers/vdpa/vdpa_sim/vdpa_sim.c b/drivers/vdpa/vdpa_sim/vdpa_sim.c
index 9bde33e38e27..36a1d2e0a6ba 100644
--- a/drivers/vdpa/vdpa_sim/vdpa_sim.c
+++ b/drivers/vdpa/vdpa_sim/vdpa_sim.c
@@ -11,8 +11,8 @@
 #include <linux/module.h>
 #include <linux/device.h>
 #include <linux/kernel.h>
+#include <linux/kthread.h>
 #include <linux/slab.h>
-#include <linux/sched.h>
 #include <linux/dma-map-ops.h>
 #include <linux/vringh.h>
 #include <linux/vdpa.h>
@@ -245,7 +245,7 @@ static const struct dma_map_ops vdpasim_dma_ops = {
 static const struct vdpa_config_ops vdpasim_config_ops;
 static const struct vdpa_config_ops vdpasim_batch_config_ops;
 
-static void vdpasim_work_fn(struct work_struct *work)
+static void vdpasim_work_fn(struct kthread_work *work)
 {
 	struct vdpasim *vdpasim = container_of(work, struct vdpasim, work);
 
@@ -282,7 +282,13 @@ struct vdpasim *vdpasim_create(struct vdpasim_dev_attr *dev_attr,
 	}
 
 	vdpasim->dev_attr = *dev_attr;
-	INIT_WORK(&vdpasim->work, vdpasim_work_fn);
+
+	kthread_init_work(&vdpasim->work, vdpasim_work_fn);
+	vdpasim->worker = kthread_create_worker(0, "vDPA sim worker: %s",
+						dev_attr->name);
+	if (IS_ERR(vdpasim->worker))
+		goto err_iommu;
+
 	spin_lock_init(&vdpasim->lock);
 	spin_lock_init(&vdpasim->iommu_lock);
 
@@ -338,7 +344,7 @@ EXPORT_SYMBOL_GPL(vdpasim_create);
 
 void vdpasim_schedule_work(struct vdpasim *vdpasim)
 {
-	schedule_work(&vdpasim->work);
+	kthread_queue_work(vdpasim->worker, &vdpasim->work);
 }
 EXPORT_SYMBOL_GPL(vdpasim_schedule_work);
 
@@ -689,7 +695,8 @@ static void vdpasim_free(struct vdpa_device *vdpa)
 	struct vdpasim *vdpasim = vdpa_to_sim(vdpa);
 	int i;
 
-	cancel_work_sync(&vdpasim->work);
+	kthread_cancel_work_sync(&vdpasim->work);
+	kthread_destroy_worker(vdpasim->worker);
 
 	for (i = 0; i < vdpasim->dev_attr.nvqs; i++) {
 		vringh_kiov_cleanup(&vdpasim->vqs[i].out_iov);
-- 
2.38.1

[RFC PATCH 6/6] vdpa_sim: add support for user VA

[Stefano Garzarella]

The new "use_va" module parameter (default: false) is used in
vdpa_alloc_device() to inform the vDPA framework that the device
supports VA.

vringh is initialized to use VA only when "use_va" is true and the
user's mm has been bound. So, only when the bus supports user VA
(e.g. vhost-vdpa).

vdpasim_mm_work_fn work is used to attach the kthread to the user
address space when the .bind_mm callback is invoked, and to detach
it when the device is reset.

Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
---
 drivers/vdpa/vdpa_sim/vdpa_sim.h |   1 +
 drivers/vdpa/vdpa_sim/vdpa_sim.c | 104 ++++++++++++++++++++++++++++++-
 2 files changed, 103 insertions(+), 2 deletions(-)

diff --git a/drivers/vdpa/vdpa_sim/vdpa_sim.h b/drivers/vdpa/vdpa_sim/vdpa_sim.h
index 07ef53ea375e..1b010e5c0445 100644
--- a/drivers/vdpa/vdpa_sim/vdpa_sim.h
+++ b/drivers/vdpa/vdpa_sim/vdpa_sim.h
@@ -55,6 +55,7 @@ struct vdpasim {
 	struct vdpasim_virtqueue *vqs;
 	struct kthread_worker *worker;
 	struct kthread_work work;
+	struct mm_struct *mm_bound;
 	struct vdpasim_dev_attr dev_attr;
 	/* spinlock to synchronize virtqueue state */
 	spinlock_t lock;
diff --git a/drivers/vdpa/vdpa_sim/vdpa_sim.c b/drivers/vdpa/vdpa_sim/vdpa_sim.c
index 36a1d2e0a6ba..6e07cedef30c 100644
--- a/drivers/vdpa/vdpa_sim/vdpa_sim.c
+++ b/drivers/vdpa/vdpa_sim/vdpa_sim.c
@@ -36,10 +36,90 @@ module_param(max_iotlb_entries, int, 0444);
 MODULE_PARM_DESC(max_iotlb_entries,
 		 "Maximum number of iotlb entries for each address space. 0 means unlimited. (default: 2048)");
 
+static bool use_va;
+module_param(use_va, bool, 0444);
+MODULE_PARM_DESC(use_va, "Enable the device's ability to use VA");
+
 #define VDPASIM_QUEUE_ALIGN PAGE_SIZE
 #define VDPASIM_QUEUE_MAX 256
 #define VDPASIM_VENDOR_ID 0
 
+struct vdpasim_mm_work {
+	struct kthread_work work;
+	struct task_struct *owner;
+	struct mm_struct *mm;
+	bool bind;
+	int ret;
+};
+
+static void vdpasim_mm_work_fn(struct kthread_work *work)
+{
+	struct vdpasim_mm_work *mm_work =
+		container_of(work, struct vdpasim_mm_work, work);
+
+	mm_work->ret = 0;
+
+	if (mm_work->bind) {
+		kthread_use_mm(mm_work->mm);
+#if 0
+		if (mm_work->owner)
+			mm_work->ret = cgroup_attach_task_all(mm_work->owner,
+							      current);
+#endif
+	} else {
+#if 0
+		//TODO: check it
+		cgroup_release(current);
+#endif
+		kthread_unuse_mm(mm_work->mm);
+	}
+}
+
+static void vdpasim_worker_queue_mm(struct vdpasim *vdpasim,
+				    struct vdpasim_mm_work *mm_work)
+{
+	struct kthread_work *work = &mm_work->work;
+
+	kthread_init_work(work, vdpasim_mm_work_fn);
+	kthread_queue_work(vdpasim->worker, work);
+
+	spin_unlock(&vdpasim->lock);
+	kthread_flush_work(work);
+	spin_lock(&vdpasim->lock);
+}
+
+static int vdpasim_worker_bind_mm(struct vdpasim *vdpasim,
+				  struct mm_struct *new_mm,
+				  struct task_struct *owner)
+{
+	struct vdpasim_mm_work mm_work;
+
+	mm_work.owner = owner;
+	mm_work.mm = new_mm;
+	mm_work.bind = true;
+
+	vdpasim_worker_queue_mm(vdpasim, &mm_work);
+
+	if (!mm_work.ret)
+		vdpasim->mm_bound = new_mm;
+
+	return mm_work.ret;
+}
+
+static void vdpasim_worker_unbind_mm(struct vdpasim *vdpasim)
+{
+	struct vdpasim_mm_work mm_work;
+
+	if (!vdpasim->mm_bound)
+		return;
+
+	mm_work.mm = vdpasim->mm_bound;
+	mm_work.bind = false;
+
+	vdpasim_worker_queue_mm(vdpasim, &mm_work);
+
+	vdpasim->mm_bound = NULL;
+}
 static struct vdpasim *vdpa_to_sim(struct vdpa_device *vdpa)
 {
 	return container_of(vdpa, struct vdpasim, vdpa);
@@ -66,8 +146,10 @@ static void vdpasim_vq_notify(struct vringh *vring)
 static void vdpasim_queue_ready(struct vdpasim *vdpasim, unsigned int idx)
 {
 	struct vdpasim_virtqueue *vq = &vdpasim->vqs[idx];
+	bool va_enabled = use_va && vdpasim->mm_bound;
 
-	vringh_init_iotlb(&vq->vring, vdpasim->features, vq->num, false, false,
+	vringh_init_iotlb(&vq->vring, vdpasim->features, vq->num, false,
+			  va_enabled,
 			  (struct vring_desc *)(uintptr_t)vq->desc_addr,
 			  (struct vring_avail *)
 			  (uintptr_t)vq->driver_addr,
@@ -96,6 +178,9 @@ static void vdpasim_do_reset(struct vdpasim *vdpasim)
 {
 	int i;
 
+	//TODO: should we cancel the works?
+	vdpasim_worker_unbind_mm(vdpasim);
+
 	spin_lock(&vdpasim->iommu_lock);
 
 	for (i = 0; i < vdpasim->dev_attr.nvqs; i++) {
@@ -275,7 +360,7 @@ struct vdpasim *vdpasim_create(struct vdpasim_dev_attr *dev_attr,
 
 	vdpasim = vdpa_alloc_device(struct vdpasim, vdpa, NULL, ops,
 				    dev_attr->ngroups, dev_attr->nas,
-				    dev_attr->name, false);
+				    dev_attr->name, use_va);
 	if (IS_ERR(vdpasim)) {
 		ret = PTR_ERR(vdpasim);
 		goto err_alloc;
@@ -657,6 +742,19 @@ static int vdpasim_set_map(struct vdpa_device *vdpa, unsigned int asid,
 	return ret;
 }
 
+static int vdpasim_bind_mm(struct vdpa_device *vdpa, struct mm_struct *mm,
+			   struct task_struct *owner)
+{
+	struct vdpasim *vdpasim = vdpa_to_sim(vdpa);
+	int ret;
+
+	spin_lock(&vdpasim->lock);
+	ret = vdpasim_worker_bind_mm(vdpasim, mm, owner);
+	spin_unlock(&vdpasim->lock);
+
+	return ret;
+}
+
 static int vdpasim_dma_map(struct vdpa_device *vdpa, unsigned int asid,
 			   u64 iova, u64 size,
 			   u64 pa, u32 perm, void *opaque)
@@ -744,6 +842,7 @@ static const struct vdpa_config_ops vdpasim_config_ops = {
 	.set_group_asid         = vdpasim_set_group_asid,
 	.dma_map                = vdpasim_dma_map,
 	.dma_unmap              = vdpasim_dma_unmap,
+	.bind_mm		= vdpasim_bind_mm,
 	.free                   = vdpasim_free,
 };
 
@@ -776,6 +875,7 @@ static const struct vdpa_config_ops vdpasim_batch_config_ops = {
 	.get_iova_range         = vdpasim_get_iova_range,
 	.set_group_asid         = vdpasim_set_group_asid,
 	.set_map                = vdpasim_set_map,
+	.bind_mm		= vdpasim_bind_mm,
 	.free                   = vdpasim_free,
 };
 
-- 
2.38.1

_______________________________________________
Virtualization mailing list
Virtualization@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

[RFC PATCH 6/6] vdpa_sim: add support for user VA

[Stefano Garzarella]

The new "use_va" module parameter (default: false) is used in
vdpa_alloc_device() to inform the vDPA framework that the device
supports VA.

vringh is initialized to use VA only when "use_va" is true and the
user's mm has been bound. So, only when the bus supports user VA
(e.g. vhost-vdpa).

vdpasim_mm_work_fn work is used to attach the kthread to the user
address space when the .bind_mm callback is invoked, and to detach
it when the device is reset.

Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
---
 drivers/vdpa/vdpa_sim/vdpa_sim.h |   1 +
 drivers/vdpa/vdpa_sim/vdpa_sim.c | 104 ++++++++++++++++++++++++++++++-
 2 files changed, 103 insertions(+), 2 deletions(-)

diff --git a/drivers/vdpa/vdpa_sim/vdpa_sim.h b/drivers/vdpa/vdpa_sim/vdpa_sim.h
index 07ef53ea375e..1b010e5c0445 100644
--- a/drivers/vdpa/vdpa_sim/vdpa_sim.h
+++ b/drivers/vdpa/vdpa_sim/vdpa_sim.h
@@ -55,6 +55,7 @@ struct vdpasim {
 	struct vdpasim_virtqueue *vqs;
 	struct kthread_worker *worker;
 	struct kthread_work work;
+	struct mm_struct *mm_bound;
 	struct vdpasim_dev_attr dev_attr;
 	/* spinlock to synchronize virtqueue state */
 	spinlock_t lock;
diff --git a/drivers/vdpa/vdpa_sim/vdpa_sim.c b/drivers/vdpa/vdpa_sim/vdpa_sim.c
index 36a1d2e0a6ba..6e07cedef30c 100644
--- a/drivers/vdpa/vdpa_sim/vdpa_sim.c
+++ b/drivers/vdpa/vdpa_sim/vdpa_sim.c
@@ -36,10 +36,90 @@ module_param(max_iotlb_entries, int, 0444);
 MODULE_PARM_DESC(max_iotlb_entries,
 		 "Maximum number of iotlb entries for each address space. 0 means unlimited. (default: 2048)");
 
+static bool use_va;
+module_param(use_va, bool, 0444);
+MODULE_PARM_DESC(use_va, "Enable the device's ability to use VA");
+
 #define VDPASIM_QUEUE_ALIGN PAGE_SIZE
 #define VDPASIM_QUEUE_MAX 256
 #define VDPASIM_VENDOR_ID 0
 
+struct vdpasim_mm_work {
+	struct kthread_work work;
+	struct task_struct *owner;
+	struct mm_struct *mm;
+	bool bind;
+	int ret;
+};
+
+static void vdpasim_mm_work_fn(struct kthread_work *work)
+{
+	struct vdpasim_mm_work *mm_work =
+		container_of(work, struct vdpasim_mm_work, work);
+
+	mm_work->ret = 0;
+
+	if (mm_work->bind) {
+		kthread_use_mm(mm_work->mm);
+#if 0
+		if (mm_work->owner)
+			mm_work->ret = cgroup_attach_task_all(mm_work->owner,
+							      current);
+#endif
+	} else {
+#if 0
+		//TODO: check it
+		cgroup_release(current);
+#endif
+		kthread_unuse_mm(mm_work->mm);
+	}
+}
+
+static void vdpasim_worker_queue_mm(struct vdpasim *vdpasim,
+				    struct vdpasim_mm_work *mm_work)
+{
+	struct kthread_work *work = &mm_work->work;
+
+	kthread_init_work(work, vdpasim_mm_work_fn);
+	kthread_queue_work(vdpasim->worker, work);
+
+	spin_unlock(&vdpasim->lock);
+	kthread_flush_work(work);
+	spin_lock(&vdpasim->lock);
+}
+
+static int vdpasim_worker_bind_mm(struct vdpasim *vdpasim,
+				  struct mm_struct *new_mm,
+				  struct task_struct *owner)
+{
+	struct vdpasim_mm_work mm_work;
+
+	mm_work.owner = owner;
+	mm_work.mm = new_mm;
+	mm_work.bind = true;
+
+	vdpasim_worker_queue_mm(vdpasim, &mm_work);
+
+	if (!mm_work.ret)
+		vdpasim->mm_bound = new_mm;
+
+	return mm_work.ret;
+}
+
+static void vdpasim_worker_unbind_mm(struct vdpasim *vdpasim)
+{
+	struct vdpasim_mm_work mm_work;
+
+	if (!vdpasim->mm_bound)
+		return;
+
+	mm_work.mm = vdpasim->mm_bound;
+	mm_work.bind = false;
+
+	vdpasim_worker_queue_mm(vdpasim, &mm_work);
+
+	vdpasim->mm_bound = NULL;
+}
 static struct vdpasim *vdpa_to_sim(struct vdpa_device *vdpa)
 {
 	return container_of(vdpa, struct vdpasim, vdpa);
@@ -66,8 +146,10 @@ static void vdpasim_vq_notify(struct vringh *vring)
 static void vdpasim_queue_ready(struct vdpasim *vdpasim, unsigned int idx)
 {
 	struct vdpasim_virtqueue *vq = &vdpasim->vqs[idx];
+	bool va_enabled = use_va && vdpasim->mm_bound;
 
-	vringh_init_iotlb(&vq->vring, vdpasim->features, vq->num, false, false,
+	vringh_init_iotlb(&vq->vring, vdpasim->features, vq->num, false,
+			  va_enabled,
 			  (struct vring_desc *)(uintptr_t)vq->desc_addr,
 			  (struct vring_avail *)
 			  (uintptr_t)vq->driver_addr,
@@ -96,6 +178,9 @@ static void vdpasim_do_reset(struct vdpasim *vdpasim)
 {
 	int i;
 
+	//TODO: should we cancel the works?
+	vdpasim_worker_unbind_mm(vdpasim);
+
 	spin_lock(&vdpasim->iommu_lock);
 
 	for (i = 0; i < vdpasim->dev_attr.nvqs; i++) {
@@ -275,7 +360,7 @@ struct vdpasim *vdpasim_create(struct vdpasim_dev_attr *dev_attr,
 
 	vdpasim = vdpa_alloc_device(struct vdpasim, vdpa, NULL, ops,
 				    dev_attr->ngroups, dev_attr->nas,
-				    dev_attr->name, false);
+				    dev_attr->name, use_va);
 	if (IS_ERR(vdpasim)) {
 		ret = PTR_ERR(vdpasim);
 		goto err_alloc;
@@ -657,6 +742,19 @@ static int vdpasim_set_map(struct vdpa_device *vdpa, unsigned int asid,
 	return ret;
 }
 
+static int vdpasim_bind_mm(struct vdpa_device *vdpa, struct mm_struct *mm,
+			   struct task_struct *owner)
+{
+	struct vdpasim *vdpasim = vdpa_to_sim(vdpa);
+	int ret;
+
+	spin_lock(&vdpasim->lock);
+	ret = vdpasim_worker_bind_mm(vdpasim, mm, owner);
+	spin_unlock(&vdpasim->lock);
+
+	return ret;
+}
+
 static int vdpasim_dma_map(struct vdpa_device *vdpa, unsigned int asid,
 			   u64 iova, u64 size,
 			   u64 pa, u32 perm, void *opaque)
@@ -744,6 +842,7 @@ static const struct vdpa_config_ops vdpasim_config_ops = {
 	.set_group_asid         = vdpasim_set_group_asid,
 	.dma_map                = vdpasim_dma_map,
 	.dma_unmap              = vdpasim_dma_unmap,
+	.bind_mm		= vdpasim_bind_mm,
 	.free                   = vdpasim_free,
 };
 
@@ -776,6 +875,7 @@ static const struct vdpa_config_ops vdpasim_batch_config_ops = {
 	.get_iova_range         = vdpasim_get_iova_range,
 	.set_group_asid         = vdpasim_set_group_asid,
 	.set_map                = vdpasim_set_map,
+	.bind_mm		= vdpasim_bind_mm,
 	.free                   = vdpasim_free,
 };
 
-- 
2.38.1

Re: [RFC PATCH 1/6] vdpa: add bind_mm callback

[Jason Wang]

On Thu, Dec 15, 2022 at 12:30 AM Stefano Garzarella <sgarzare@redhat.com> wrote:
>
> This new optional callback is used to bind the device to a specific
> address space so the vDPA framework can use VA when this callback
> is implemented.
>
> Suggested-by: Jason Wang <jasowang@redhat.com>
> Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
> ---
>  include/linux/vdpa.h | 8 ++++++++
>  1 file changed, 8 insertions(+)
>
> diff --git a/include/linux/vdpa.h b/include/linux/vdpa.h
> index 6d0f5e4e82c2..34388e21ef3f 100644
> --- a/include/linux/vdpa.h
> +++ b/include/linux/vdpa.h
> @@ -282,6 +282,12 @@ struct vdpa_map_file {
>   *                             @iova: iova to be unmapped
>   *                             @size: size of the area
>   *                             Returns integer: success (0) or error (< 0)
> + * @bind_mm:                   Bind the device to a specific address space
> + *                             so the vDPA framework can use VA when this
> + *                             callback is implemented. (optional)
> + *                             @vdev: vdpa device
> + *                             @mm: address space to bind

Do we need an unbind or did a NULL mm mean unbind?

> + *                             @owner: process that owns the address space

Any reason we need the task_struct here?

Thanks

>   * @free:                      Free resources that belongs to vDPA (optional)
>   *                             @vdev: vdpa device
>   */
> @@ -341,6 +347,8 @@ struct vdpa_config_ops {
>                          u64 iova, u64 size);
>         int (*set_group_asid)(struct vdpa_device *vdev, unsigned int group,
>                               unsigned int asid);
> +       int (*bind_mm)(struct vdpa_device *vdev, struct mm_struct *mm,
> +                      struct task_struct *owner);
>
>         /* Free device resources */
>         void (*free)(struct vdpa_device *vdev);
> --
> 2.38.1
>

_______________________________________________
Virtualization mailing list
Virtualization@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

Re: [RFC PATCH 1/6] vdpa: add bind_mm callback

[Jason Wang]

On Thu, Dec 15, 2022 at 12:30 AM Stefano Garzarella <sgarzare@redhat.com> wrote:
>
> This new optional callback is used to bind the device to a specific
> address space so the vDPA framework can use VA when this callback
> is implemented.
>
> Suggested-by: Jason Wang <jasowang@redhat.com>
> Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
> ---
>  include/linux/vdpa.h | 8 ++++++++
>  1 file changed, 8 insertions(+)
>
> diff --git a/include/linux/vdpa.h b/include/linux/vdpa.h
> index 6d0f5e4e82c2..34388e21ef3f 100644
> --- a/include/linux/vdpa.h
> +++ b/include/linux/vdpa.h
> @@ -282,6 +282,12 @@ struct vdpa_map_file {
>   *                             @iova: iova to be unmapped
>   *                             @size: size of the area
>   *                             Returns integer: success (0) or error (< 0)
> + * @bind_mm:                   Bind the device to a specific address space
> + *                             so the vDPA framework can use VA when this
> + *                             callback is implemented. (optional)
> + *                             @vdev: vdpa device
> + *                             @mm: address space to bind

Do we need an unbind or did a NULL mm mean unbind?

> + *                             @owner: process that owns the address space

Any reason we need the task_struct here?

Thanks

>   * @free:                      Free resources that belongs to vDPA (optional)
>   *                             @vdev: vdpa device
>   */
> @@ -341,6 +347,8 @@ struct vdpa_config_ops {
>                          u64 iova, u64 size);
>         int (*set_group_asid)(struct vdpa_device *vdev, unsigned int group,
>                               unsigned int asid);
> +       int (*bind_mm)(struct vdpa_device *vdev, struct mm_struct *mm,
> +                      struct task_struct *owner);
>
>         /* Free device resources */
>         void (*free)(struct vdpa_device *vdev);
> --
> 2.38.1
>

Re: [RFC PATCH 6/6] vdpa_sim: add support for user VA

[Jason Wang]

On Thu, Dec 15, 2022 at 12:31 AM Stefano Garzarella <sgarzare@redhat.com> wrote:
>
> The new "use_va" module parameter (default: false) is used in
> vdpa_alloc_device() to inform the vDPA framework that the device
> supports VA.
>
> vringh is initialized to use VA only when "use_va" is true and the
> user's mm has been bound. So, only when the bus supports user VA
> (e.g. vhost-vdpa).
>
> vdpasim_mm_work_fn work is used to attach the kthread to the user
> address space when the .bind_mm callback is invoked, and to detach
> it when the device is reset.

One thing in my mind is that the current datapath is running under
spinlock which prevents us from using iov_iter (which may have page
faults).

We need to get rid of the spinlock first.

>
> Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
> ---
>  drivers/vdpa/vdpa_sim/vdpa_sim.h |   1 +
>  drivers/vdpa/vdpa_sim/vdpa_sim.c | 104 ++++++++++++++++++++++++++++++-
>  2 files changed, 103 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/vdpa/vdpa_sim/vdpa_sim.h b/drivers/vdpa/vdpa_sim/vdpa_sim.h
> index 07ef53ea375e..1b010e5c0445 100644
> --- a/drivers/vdpa/vdpa_sim/vdpa_sim.h
> +++ b/drivers/vdpa/vdpa_sim/vdpa_sim.h
> @@ -55,6 +55,7 @@ struct vdpasim {
>         struct vdpasim_virtqueue *vqs;
>         struct kthread_worker *worker;
>         struct kthread_work work;
> +       struct mm_struct *mm_bound;
>         struct vdpasim_dev_attr dev_attr;
>         /* spinlock to synchronize virtqueue state */
>         spinlock_t lock;
> diff --git a/drivers/vdpa/vdpa_sim/vdpa_sim.c b/drivers/vdpa/vdpa_sim/vdpa_sim.c
> index 36a1d2e0a6ba..6e07cedef30c 100644
> --- a/drivers/vdpa/vdpa_sim/vdpa_sim.c
> +++ b/drivers/vdpa/vdpa_sim/vdpa_sim.c
> @@ -36,10 +36,90 @@ module_param(max_iotlb_entries, int, 0444);
>  MODULE_PARM_DESC(max_iotlb_entries,
>                  "Maximum number of iotlb entries for each address space. 0 means unlimited. (default: 2048)");
>
> +static bool use_va;
> +module_param(use_va, bool, 0444);
> +MODULE_PARM_DESC(use_va, "Enable the device's ability to use VA");
> +
>  #define VDPASIM_QUEUE_ALIGN PAGE_SIZE
>  #define VDPASIM_QUEUE_MAX 256
>  #define VDPASIM_VENDOR_ID 0
>
> +struct vdpasim_mm_work {
> +       struct kthread_work work;
> +       struct task_struct *owner;
> +       struct mm_struct *mm;
> +       bool bind;
> +       int ret;
> +};
> +
> +static void vdpasim_mm_work_fn(struct kthread_work *work)
> +{
> +       struct vdpasim_mm_work *mm_work =
> +               container_of(work, struct vdpasim_mm_work, work);
> +
> +       mm_work->ret = 0;
> +
> +       if (mm_work->bind) {
> +               kthread_use_mm(mm_work->mm);
> +#if 0
> +               if (mm_work->owner)
> +                       mm_work->ret = cgroup_attach_task_all(mm_work->owner,
> +                                                             current);
> +#endif
> +       } else {
> +#if 0
> +               //TODO: check it
> +               cgroup_release(current);
> +#endif
> +               kthread_unuse_mm(mm_work->mm);
> +       }
> +}
> +
> +static void vdpasim_worker_queue_mm(struct vdpasim *vdpasim,
> +                                   struct vdpasim_mm_work *mm_work)
> +{
> +       struct kthread_work *work = &mm_work->work;
> +
> +       kthread_init_work(work, vdpasim_mm_work_fn);
> +       kthread_queue_work(vdpasim->worker, work);
> +
> +       spin_unlock(&vdpasim->lock);
> +       kthread_flush_work(work);
> +       spin_lock(&vdpasim->lock);
> +}
> +
> +static int vdpasim_worker_bind_mm(struct vdpasim *vdpasim,
> +                                 struct mm_struct *new_mm,
> +                                 struct task_struct *owner)
> +{
> +       struct vdpasim_mm_work mm_work;
> +
> +       mm_work.owner = owner;
> +       mm_work.mm = new_mm;
> +       mm_work.bind = true;
> +
> +       vdpasim_worker_queue_mm(vdpasim, &mm_work);
> +

Should we wait for the work to be finished?

> +       if (!mm_work.ret)
> +               vdpasim->mm_bound = new_mm;
> +
> +       return mm_work.ret;
> +}
> +
> +static void vdpasim_worker_unbind_mm(struct vdpasim *vdpasim)
> +{
> +       struct vdpasim_mm_work mm_work;
> +
> +       if (!vdpasim->mm_bound)
> +               return;
> +
> +       mm_work.mm = vdpasim->mm_bound;
> +       mm_work.bind = false;
> +
> +       vdpasim_worker_queue_mm(vdpasim, &mm_work);
> +
> +       vdpasim->mm_bound = NULL;
> +}
>  static struct vdpasim *vdpa_to_sim(struct vdpa_device *vdpa)
>  {
>         return container_of(vdpa, struct vdpasim, vdpa);
> @@ -66,8 +146,10 @@ static void vdpasim_vq_notify(struct vringh *vring)
>  static void vdpasim_queue_ready(struct vdpasim *vdpasim, unsigned int idx)
>  {
>         struct vdpasim_virtqueue *vq = &vdpasim->vqs[idx];
> +       bool va_enabled = use_va && vdpasim->mm_bound;
>
> -       vringh_init_iotlb(&vq->vring, vdpasim->features, vq->num, false, false,
> +       vringh_init_iotlb(&vq->vring, vdpasim->features, vq->num, false,
> +                         va_enabled,
>                           (struct vring_desc *)(uintptr_t)vq->desc_addr,
>                           (struct vring_avail *)
>                           (uintptr_t)vq->driver_addr,
> @@ -96,6 +178,9 @@ static void vdpasim_do_reset(struct vdpasim *vdpasim)
>  {
>         int i;
>
> +       //TODO: should we cancel the works?
> +       vdpasim_worker_unbind_mm(vdpasim);

We probably don't need this since it's the virtio level reset so we
need to keep the mm bound in this case. Otherwise we may break the
guest. It should be the responsibility of the driver to call
config_ops->unbind if it needs to do that.

Thanks


> +
>         spin_lock(&vdpasim->iommu_lock);
>
>         for (i = 0; i < vdpasim->dev_attr.nvqs; i++) {
> @@ -275,7 +360,7 @@ struct vdpasim *vdpasim_create(struct vdpasim_dev_attr *dev_attr,
>
>         vdpasim = vdpa_alloc_device(struct vdpasim, vdpa, NULL, ops,
>                                     dev_attr->ngroups, dev_attr->nas,
> -                                   dev_attr->name, false);
> +                                   dev_attr->name, use_va);
>         if (IS_ERR(vdpasim)) {
>                 ret = PTR_ERR(vdpasim);
>                 goto err_alloc;
> @@ -657,6 +742,19 @@ static int vdpasim_set_map(struct vdpa_device *vdpa, unsigned int asid,
>         return ret;
>  }
>
> +static int vdpasim_bind_mm(struct vdpa_device *vdpa, struct mm_struct *mm,
> +                          struct task_struct *owner)
> +{
> +       struct vdpasim *vdpasim = vdpa_to_sim(vdpa);
> +       int ret;
> +
> +       spin_lock(&vdpasim->lock);
> +       ret = vdpasim_worker_bind_mm(vdpasim, mm, owner);
> +       spin_unlock(&vdpasim->lock);
> +
> +       return ret;
> +}
> +
>  static int vdpasim_dma_map(struct vdpa_device *vdpa, unsigned int asid,
>                            u64 iova, u64 size,
>                            u64 pa, u32 perm, void *opaque)
> @@ -744,6 +842,7 @@ static const struct vdpa_config_ops vdpasim_config_ops = {
>         .set_group_asid         = vdpasim_set_group_asid,
>         .dma_map                = vdpasim_dma_map,
>         .dma_unmap              = vdpasim_dma_unmap,
> +       .bind_mm                = vdpasim_bind_mm,
>         .free                   = vdpasim_free,
>  };
>
> @@ -776,6 +875,7 @@ static const struct vdpa_config_ops vdpasim_batch_config_ops = {
>         .get_iova_range         = vdpasim_get_iova_range,
>         .set_group_asid         = vdpasim_set_group_asid,
>         .set_map                = vdpasim_set_map,
> +       .bind_mm                = vdpasim_bind_mm,
>         .free                   = vdpasim_free,
>  };
>
> --
> 2.38.1
>

Re: [RFC PATCH 6/6] vdpa_sim: add support for user VA

[Jason Wang]

On Thu, Dec 15, 2022 at 12:31 AM Stefano Garzarella <sgarzare@redhat.com> wrote:
>
> The new "use_va" module parameter (default: false) is used in
> vdpa_alloc_device() to inform the vDPA framework that the device
> supports VA.
>
> vringh is initialized to use VA only when "use_va" is true and the
> user's mm has been bound. So, only when the bus supports user VA
> (e.g. vhost-vdpa).
>
> vdpasim_mm_work_fn work is used to attach the kthread to the user
> address space when the .bind_mm callback is invoked, and to detach
> it when the device is reset.

One thing in my mind is that the current datapath is running under
spinlock which prevents us from using iov_iter (which may have page
faults).

We need to get rid of the spinlock first.

>
> Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
> ---
>  drivers/vdpa/vdpa_sim/vdpa_sim.h |   1 +
>  drivers/vdpa/vdpa_sim/vdpa_sim.c | 104 ++++++++++++++++++++++++++++++-
>  2 files changed, 103 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/vdpa/vdpa_sim/vdpa_sim.h b/drivers/vdpa/vdpa_sim/vdpa_sim.h
> index 07ef53ea375e..1b010e5c0445 100644
> --- a/drivers/vdpa/vdpa_sim/vdpa_sim.h
> +++ b/drivers/vdpa/vdpa_sim/vdpa_sim.h
> @@ -55,6 +55,7 @@ struct vdpasim {
>         struct vdpasim_virtqueue *vqs;
>         struct kthread_worker *worker;
>         struct kthread_work work;
> +       struct mm_struct *mm_bound;
>         struct vdpasim_dev_attr dev_attr;
>         /* spinlock to synchronize virtqueue state */
>         spinlock_t lock;
> diff --git a/drivers/vdpa/vdpa_sim/vdpa_sim.c b/drivers/vdpa/vdpa_sim/vdpa_sim.c
> index 36a1d2e0a6ba..6e07cedef30c 100644
> --- a/drivers/vdpa/vdpa_sim/vdpa_sim.c
> +++ b/drivers/vdpa/vdpa_sim/vdpa_sim.c
> @@ -36,10 +36,90 @@ module_param(max_iotlb_entries, int, 0444);
>  MODULE_PARM_DESC(max_iotlb_entries,
>                  "Maximum number of iotlb entries for each address space. 0 means unlimited. (default: 2048)");
>
> +static bool use_va;
> +module_param(use_va, bool, 0444);
> +MODULE_PARM_DESC(use_va, "Enable the device's ability to use VA");
> +
>  #define VDPASIM_QUEUE_ALIGN PAGE_SIZE
>  #define VDPASIM_QUEUE_MAX 256
>  #define VDPASIM_VENDOR_ID 0
>
> +struct vdpasim_mm_work {
> +       struct kthread_work work;
> +       struct task_struct *owner;
> +       struct mm_struct *mm;
> +       bool bind;
> +       int ret;
> +};
> +
> +static void vdpasim_mm_work_fn(struct kthread_work *work)
> +{
> +       struct vdpasim_mm_work *mm_work =
> +               container_of(work, struct vdpasim_mm_work, work);
> +
> +       mm_work->ret = 0;
> +
> +       if (mm_work->bind) {
> +               kthread_use_mm(mm_work->mm);
> +#if 0
> +               if (mm_work->owner)
> +                       mm_work->ret = cgroup_attach_task_all(mm_work->owner,
> +                                                             current);
> +#endif
> +       } else {
> +#if 0
> +               //TODO: check it
> +               cgroup_release(current);
> +#endif
> +               kthread_unuse_mm(mm_work->mm);
> +       }
> +}
> +
> +static void vdpasim_worker_queue_mm(struct vdpasim *vdpasim,
> +                                   struct vdpasim_mm_work *mm_work)
> +{
> +       struct kthread_work *work = &mm_work->work;
> +
> +       kthread_init_work(work, vdpasim_mm_work_fn);
> +       kthread_queue_work(vdpasim->worker, work);
> +
> +       spin_unlock(&vdpasim->lock);
> +       kthread_flush_work(work);
> +       spin_lock(&vdpasim->lock);
> +}
> +
> +static int vdpasim_worker_bind_mm(struct vdpasim *vdpasim,
> +                                 struct mm_struct *new_mm,
> +                                 struct task_struct *owner)
> +{
> +       struct vdpasim_mm_work mm_work;
> +
> +       mm_work.owner = owner;
> +       mm_work.mm = new_mm;
> +       mm_work.bind = true;
> +
> +       vdpasim_worker_queue_mm(vdpasim, &mm_work);
> +

Should we wait for the work to be finished?

> +       if (!mm_work.ret)
> +               vdpasim->mm_bound = new_mm;
> +
> +       return mm_work.ret;
> +}
> +
> +static void vdpasim_worker_unbind_mm(struct vdpasim *vdpasim)
> +{
> +       struct vdpasim_mm_work mm_work;
> +
> +       if (!vdpasim->mm_bound)
> +               return;
> +
> +       mm_work.mm = vdpasim->mm_bound;
> +       mm_work.bind = false;
> +
> +       vdpasim_worker_queue_mm(vdpasim, &mm_work);
> +
> +       vdpasim->mm_bound = NULL;
> +}
>  static struct vdpasim *vdpa_to_sim(struct vdpa_device *vdpa)
>  {
>         return container_of(vdpa, struct vdpasim, vdpa);
> @@ -66,8 +146,10 @@ static void vdpasim_vq_notify(struct vringh *vring)
>  static void vdpasim_queue_ready(struct vdpasim *vdpasim, unsigned int idx)
>  {
>         struct vdpasim_virtqueue *vq = &vdpasim->vqs[idx];
> +       bool va_enabled = use_va && vdpasim->mm_bound;
>
> -       vringh_init_iotlb(&vq->vring, vdpasim->features, vq->num, false, false,
> +       vringh_init_iotlb(&vq->vring, vdpasim->features, vq->num, false,
> +                         va_enabled,
>                           (struct vring_desc *)(uintptr_t)vq->desc_addr,
>                           (struct vring_avail *)
>                           (uintptr_t)vq->driver_addr,
> @@ -96,6 +178,9 @@ static void vdpasim_do_reset(struct vdpasim *vdpasim)
>  {
>         int i;
>
> +       //TODO: should we cancel the works?
> +       vdpasim_worker_unbind_mm(vdpasim);

We probably don't need this since it's the virtio level reset so we
need to keep the mm bound in this case. Otherwise we may break the
guest. It should be the responsibility of the driver to call
config_ops->unbind if it needs to do that.

Thanks


> +
>         spin_lock(&vdpasim->iommu_lock);
>
>         for (i = 0; i < vdpasim->dev_attr.nvqs; i++) {
> @@ -275,7 +360,7 @@ struct vdpasim *vdpasim_create(struct vdpasim_dev_attr *dev_attr,
>
>         vdpasim = vdpa_alloc_device(struct vdpasim, vdpa, NULL, ops,
>                                     dev_attr->ngroups, dev_attr->nas,
> -                                   dev_attr->name, false);
> +                                   dev_attr->name, use_va);
>         if (IS_ERR(vdpasim)) {
>                 ret = PTR_ERR(vdpasim);
>                 goto err_alloc;
> @@ -657,6 +742,19 @@ static int vdpasim_set_map(struct vdpa_device *vdpa, unsigned int asid,
>         return ret;
>  }
>
> +static int vdpasim_bind_mm(struct vdpa_device *vdpa, struct mm_struct *mm,
> +                          struct task_struct *owner)
> +{
> +       struct vdpasim *vdpasim = vdpa_to_sim(vdpa);
> +       int ret;
> +
> +       spin_lock(&vdpasim->lock);
> +       ret = vdpasim_worker_bind_mm(vdpasim, mm, owner);
> +       spin_unlock(&vdpasim->lock);
> +
> +       return ret;
> +}
> +
>  static int vdpasim_dma_map(struct vdpa_device *vdpa, unsigned int asid,
>                            u64 iova, u64 size,
>                            u64 pa, u32 perm, void *opaque)
> @@ -744,6 +842,7 @@ static const struct vdpa_config_ops vdpasim_config_ops = {
>         .set_group_asid         = vdpasim_set_group_asid,
>         .dma_map                = vdpasim_dma_map,
>         .dma_unmap              = vdpasim_dma_unmap,
> +       .bind_mm                = vdpasim_bind_mm,
>         .free                   = vdpasim_free,
>  };
>
> @@ -776,6 +875,7 @@ static const struct vdpa_config_ops vdpasim_batch_config_ops = {
>         .get_iova_range         = vdpasim_get_iova_range,
>         .set_group_asid         = vdpasim_set_group_asid,
>         .set_map                = vdpasim_set_map,
> +       .bind_mm                = vdpasim_bind_mm,
>         .free                   = vdpasim_free,
>  };
>
> --
> 2.38.1
>

_______________________________________________
Virtualization mailing list
Virtualization@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

Re: [RFC PATCH 6/6] vdpa_sim: add support for user VA

[Stefano Garzarella]

On Fri, Dec 16, 2022 at 03:26:46PM +0800, Jason Wang wrote:
>On Thu, Dec 15, 2022 at 12:31 AM Stefano Garzarella <sgarzare@redhat.com> wrote:
>>
>> The new "use_va" module parameter (default: false) is used in
>> vdpa_alloc_device() to inform the vDPA framework that the device
>> supports VA.
>>
>> vringh is initialized to use VA only when "use_va" is true and the
>> user's mm has been bound. So, only when the bus supports user VA
>> (e.g. vhost-vdpa).
>>
>> vdpasim_mm_work_fn work is used to attach the kthread to the user
>> address space when the .bind_mm callback is invoked, and to detach
>> it when the device is reset.
>
>One thing in my mind is that the current datapath is running under
>spinlock which prevents us from using iov_iter (which may have page
>faults).
>
>We need to get rid of the spinlock first.

Right! I already have a patch for that since I used for the vdpa-blk 
software device in-kernel PoC where I had the same issue.

I'll add it to the series!

>
>>
>> Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
>> ---
>>  drivers/vdpa/vdpa_sim/vdpa_sim.h |   1 +
>>  drivers/vdpa/vdpa_sim/vdpa_sim.c | 104 ++++++++++++++++++++++++++++++-
>>  2 files changed, 103 insertions(+), 2 deletions(-)
>>
>> diff --git a/drivers/vdpa/vdpa_sim/vdpa_sim.h b/drivers/vdpa/vdpa_sim/vdpa_sim.h
>> index 07ef53ea375e..1b010e5c0445 100644
>> --- a/drivers/vdpa/vdpa_sim/vdpa_sim.h
>> +++ b/drivers/vdpa/vdpa_sim/vdpa_sim.h
>> @@ -55,6 +55,7 @@ struct vdpasim {
>>         struct vdpasim_virtqueue *vqs;
>>         struct kthread_worker *worker;
>>         struct kthread_work work;
>> +       struct mm_struct *mm_bound;
>>         struct vdpasim_dev_attr dev_attr;
>>         /* spinlock to synchronize virtqueue state */
>>         spinlock_t lock;
>> diff --git a/drivers/vdpa/vdpa_sim/vdpa_sim.c b/drivers/vdpa/vdpa_sim/vdpa_sim.c
>> index 36a1d2e0a6ba..6e07cedef30c 100644
>> --- a/drivers/vdpa/vdpa_sim/vdpa_sim.c
>> +++ b/drivers/vdpa/vdpa_sim/vdpa_sim.c
>> @@ -36,10 +36,90 @@ module_param(max_iotlb_entries, int, 0444);
>>  MODULE_PARM_DESC(max_iotlb_entries,
>>                  "Maximum number of iotlb entries for each address space. 0 means unlimited. (default: 2048)");
>>
>> +static bool use_va;
>> +module_param(use_va, bool, 0444);
>> +MODULE_PARM_DESC(use_va, "Enable the device's ability to use VA");
>> +
>>  #define VDPASIM_QUEUE_ALIGN PAGE_SIZE
>>  #define VDPASIM_QUEUE_MAX 256
>>  #define VDPASIM_VENDOR_ID 0
>>
>> +struct vdpasim_mm_work {
>> +       struct kthread_work work;
>> +       struct task_struct *owner;
>> +       struct mm_struct *mm;
>> +       bool bind;
>> +       int ret;
>> +};
>> +
>> +static void vdpasim_mm_work_fn(struct kthread_work *work)
>> +{
>> +       struct vdpasim_mm_work *mm_work =
>> +               container_of(work, struct vdpasim_mm_work, work);
>> +
>> +       mm_work->ret = 0;
>> +
>> +       if (mm_work->bind) {
>> +               kthread_use_mm(mm_work->mm);
>> +#if 0
>> +               if (mm_work->owner)
>> +                       mm_work->ret = cgroup_attach_task_all(mm_work->owner,
>> +                                                             current);
>> +#endif
>> +       } else {
>> +#if 0
>> +               //TODO: check it
>> +               cgroup_release(current);
>> +#endif
>> +               kthread_unuse_mm(mm_work->mm);
>> +       }
>> +}
>> +
>> +static void vdpasim_worker_queue_mm(struct vdpasim *vdpasim,
>> +                                   struct vdpasim_mm_work *mm_work)
>> +{
>> +       struct kthread_work *work = &mm_work->work;
>> +
>> +       kthread_init_work(work, vdpasim_mm_work_fn);
>> +       kthread_queue_work(vdpasim->worker, work);
>> +
>> +       spin_unlock(&vdpasim->lock);
>> +       kthread_flush_work(work);
>> +       spin_lock(&vdpasim->lock);
>> +}
>> +
>> +static int vdpasim_worker_bind_mm(struct vdpasim *vdpasim,
>> +                                 struct mm_struct *new_mm,
>> +                                 struct task_struct *owner)
>> +{
>> +       struct vdpasim_mm_work mm_work;
>> +
>> +       mm_work.owner = owner;
>> +       mm_work.mm = new_mm;
>> +       mm_work.bind = true;
>> +
>> +       vdpasim_worker_queue_mm(vdpasim, &mm_work);
>> +
>
>Should we wait for the work to be finished?

Yep, I'm waiting inside vdpasim_worker_queue_mm() calling 
kthread_flush_work().

If we will use mutex, I think we can avoid the lock release around that 
call.

>
>> +       if (!mm_work.ret)
>> +               vdpasim->mm_bound = new_mm;
>> +
>> +       return mm_work.ret;
>> +}
>> +
>> +static void vdpasim_worker_unbind_mm(struct vdpasim *vdpasim)
>> +{
>> +       struct vdpasim_mm_work mm_work;
>> +
>> +       if (!vdpasim->mm_bound)
>> +               return;
>> +
>> +       mm_work.mm = vdpasim->mm_bound;
>> +       mm_work.bind = false;
>> +
>> +       vdpasim_worker_queue_mm(vdpasim, &mm_work);
>> +
>> +       vdpasim->mm_bound = NULL;
>> +}
>>  static struct vdpasim *vdpa_to_sim(struct vdpa_device *vdpa)
>>  {
>>         return container_of(vdpa, struct vdpasim, vdpa);
>> @@ -66,8 +146,10 @@ static void vdpasim_vq_notify(struct vringh *vring)
>>  static void vdpasim_queue_ready(struct vdpasim *vdpasim, unsigned int idx)
>>  {
>>         struct vdpasim_virtqueue *vq = &vdpasim->vqs[idx];
>> +       bool va_enabled = use_va && vdpasim->mm_bound;
>>
>> -       vringh_init_iotlb(&vq->vring, vdpasim->features, vq->num, false, false,
>> +       vringh_init_iotlb(&vq->vring, vdpasim->features, vq->num, false,
>> +                         va_enabled,
>>                           (struct vring_desc *)(uintptr_t)vq->desc_addr,
>>                           (struct vring_avail *)
>>                           (uintptr_t)vq->driver_addr,
>> @@ -96,6 +178,9 @@ static void vdpasim_do_reset(struct vdpasim *vdpasim)
>>  {
>>         int i;
>>
>> +       //TODO: should we cancel the works?
>> +       vdpasim_worker_unbind_mm(vdpasim);
>
>We probably don't need this since it's the virtio level reset so we
>need to keep the mm bound in this case. Otherwise we may break the
>guest. It should be the responsibility of the driver to call
>config_ops->unbind if it needs to do that.

Got it, my biggest concern was when we go from a vhost-vdpa virtio-vdpa, 
but as you said, in vhost-vdpa I can call unbind before releasing the 
device.

Thanks,
Stefano

_______________________________________________
Virtualization mailing list
Virtualization@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

Re: [RFC PATCH 6/6] vdpa_sim: add support for user VA

[Stefano Garzarella]

On Fri, Dec 16, 2022 at 03:26:46PM +0800, Jason Wang wrote:
>On Thu, Dec 15, 2022 at 12:31 AM Stefano Garzarella <sgarzare@redhat.com> wrote:
>>
>> The new "use_va" module parameter (default: false) is used in
>> vdpa_alloc_device() to inform the vDPA framework that the device
>> supports VA.
>>
>> vringh is initialized to use VA only when "use_va" is true and the
>> user's mm has been bound. So, only when the bus supports user VA
>> (e.g. vhost-vdpa).
>>
>> vdpasim_mm_work_fn work is used to attach the kthread to the user
>> address space when the .bind_mm callback is invoked, and to detach
>> it when the device is reset.
>
>One thing in my mind is that the current datapath is running under
>spinlock which prevents us from using iov_iter (which may have page
>faults).
>
>We need to get rid of the spinlock first.

Right! I already have a patch for that since I used for the vdpa-blk 
software device in-kernel PoC where I had the same issue.

I'll add it to the series!

>
>>
>> Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
>> ---
>>  drivers/vdpa/vdpa_sim/vdpa_sim.h |   1 +
>>  drivers/vdpa/vdpa_sim/vdpa_sim.c | 104 ++++++++++++++++++++++++++++++-
>>  2 files changed, 103 insertions(+), 2 deletions(-)
>>
>> diff --git a/drivers/vdpa/vdpa_sim/vdpa_sim.h b/drivers/vdpa/vdpa_sim/vdpa_sim.h
>> index 07ef53ea375e..1b010e5c0445 100644
>> --- a/drivers/vdpa/vdpa_sim/vdpa_sim.h
>> +++ b/drivers/vdpa/vdpa_sim/vdpa_sim.h
>> @@ -55,6 +55,7 @@ struct vdpasim {
>>         struct vdpasim_virtqueue *vqs;
>>         struct kthread_worker *worker;
>>         struct kthread_work work;
>> +       struct mm_struct *mm_bound;
>>         struct vdpasim_dev_attr dev_attr;
>>         /* spinlock to synchronize virtqueue state */
>>         spinlock_t lock;
>> diff --git a/drivers/vdpa/vdpa_sim/vdpa_sim.c b/drivers/vdpa/vdpa_sim/vdpa_sim.c
>> index 36a1d2e0a6ba..6e07cedef30c 100644
>> --- a/drivers/vdpa/vdpa_sim/vdpa_sim.c
>> +++ b/drivers/vdpa/vdpa_sim/vdpa_sim.c
>> @@ -36,10 +36,90 @@ module_param(max_iotlb_entries, int, 0444);
>>  MODULE_PARM_DESC(max_iotlb_entries,
>>                  "Maximum number of iotlb entries for each address space. 0 means unlimited. (default: 2048)");
>>
>> +static bool use_va;
>> +module_param(use_va, bool, 0444);
>> +MODULE_PARM_DESC(use_va, "Enable the device's ability to use VA");
>> +
>>  #define VDPASIM_QUEUE_ALIGN PAGE_SIZE
>>  #define VDPASIM_QUEUE_MAX 256
>>  #define VDPASIM_VENDOR_ID 0
>>
>> +struct vdpasim_mm_work {
>> +       struct kthread_work work;
>> +       struct task_struct *owner;
>> +       struct mm_struct *mm;
>> +       bool bind;
>> +       int ret;
>> +};
>> +
>> +static void vdpasim_mm_work_fn(struct kthread_work *work)
>> +{
>> +       struct vdpasim_mm_work *mm_work =
>> +               container_of(work, struct vdpasim_mm_work, work);
>> +
>> +       mm_work->ret = 0;
>> +
>> +       if (mm_work->bind) {
>> +               kthread_use_mm(mm_work->mm);
>> +#if 0
>> +               if (mm_work->owner)
>> +                       mm_work->ret = cgroup_attach_task_all(mm_work->owner,
>> +                                                             current);
>> +#endif
>> +       } else {
>> +#if 0
>> +               //TODO: check it
>> +               cgroup_release(current);
>> +#endif
>> +               kthread_unuse_mm(mm_work->mm);
>> +       }
>> +}
>> +
>> +static void vdpasim_worker_queue_mm(struct vdpasim *vdpasim,
>> +                                   struct vdpasim_mm_work *mm_work)
>> +{
>> +       struct kthread_work *work = &mm_work->work;
>> +
>> +       kthread_init_work(work, vdpasim_mm_work_fn);
>> +       kthread_queue_work(vdpasim->worker, work);
>> +
>> +       spin_unlock(&vdpasim->lock);
>> +       kthread_flush_work(work);
>> +       spin_lock(&vdpasim->lock);
>> +}
>> +
>> +static int vdpasim_worker_bind_mm(struct vdpasim *vdpasim,
>> +                                 struct mm_struct *new_mm,
>> +                                 struct task_struct *owner)
>> +{
>> +       struct vdpasim_mm_work mm_work;
>> +
>> +       mm_work.owner = owner;
>> +       mm_work.mm = new_mm;
>> +       mm_work.bind = true;
>> +
>> +       vdpasim_worker_queue_mm(vdpasim, &mm_work);
>> +
>
>Should we wait for the work to be finished?

Yep, I'm waiting inside vdpasim_worker_queue_mm() calling 
kthread_flush_work().

If we will use mutex, I think we can avoid the lock release around that 
call.

>
>> +       if (!mm_work.ret)
>> +               vdpasim->mm_bound = new_mm;
>> +
>> +       return mm_work.ret;
>> +}
>> +
>> +static void vdpasim_worker_unbind_mm(struct vdpasim *vdpasim)
>> +{
>> +       struct vdpasim_mm_work mm_work;
>> +
>> +       if (!vdpasim->mm_bound)
>> +               return;
>> +
>> +       mm_work.mm = vdpasim->mm_bound;
>> +       mm_work.bind = false;
>> +
>> +       vdpasim_worker_queue_mm(vdpasim, &mm_work);
>> +
>> +       vdpasim->mm_bound = NULL;
>> +}
>>  static struct vdpasim *vdpa_to_sim(struct vdpa_device *vdpa)
>>  {
>>         return container_of(vdpa, struct vdpasim, vdpa);
>> @@ -66,8 +146,10 @@ static void vdpasim_vq_notify(struct vringh *vring)
>>  static void vdpasim_queue_ready(struct vdpasim *vdpasim, unsigned int idx)
>>  {
>>         struct vdpasim_virtqueue *vq = &vdpasim->vqs[idx];
>> +       bool va_enabled = use_va && vdpasim->mm_bound;
>>
>> -       vringh_init_iotlb(&vq->vring, vdpasim->features, vq->num, false, false,
>> +       vringh_init_iotlb(&vq->vring, vdpasim->features, vq->num, false,
>> +                         va_enabled,
>>                           (struct vring_desc *)(uintptr_t)vq->desc_addr,
>>                           (struct vring_avail *)
>>                           (uintptr_t)vq->driver_addr,
>> @@ -96,6 +178,9 @@ static void vdpasim_do_reset(struct vdpasim *vdpasim)
>>  {
>>         int i;
>>
>> +       //TODO: should we cancel the works?
>> +       vdpasim_worker_unbind_mm(vdpasim);
>
>We probably don't need this since it's the virtio level reset so we
>need to keep the mm bound in this case. Otherwise we may break the
>guest. It should be the responsibility of the driver to call
>config_ops->unbind if it needs to do that.

Got it, my biggest concern was when we go from a vhost-vdpa virtio-vdpa, 
but as you said, in vhost-vdpa I can call unbind before releasing the 
device.

Thanks,
Stefano

Re: [RFC PATCH 1/6] vdpa: add bind_mm callback

[Stefano Garzarella]

On Fri, Dec 16, 2022 at 02:37:45PM +0800, Jason Wang wrote:
>On Thu, Dec 15, 2022 at 12:30 AM Stefano Garzarella <sgarzare@redhat.com> wrote:
>>
>> This new optional callback is used to bind the device to a specific
>> address space so the vDPA framework can use VA when this callback
>> is implemented.
>>
>> Suggested-by: Jason Wang <jasowang@redhat.com>
>> Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
>> ---
>>  include/linux/vdpa.h | 8 ++++++++
>>  1 file changed, 8 insertions(+)
>>
>> diff --git a/include/linux/vdpa.h b/include/linux/vdpa.h
>> index 6d0f5e4e82c2..34388e21ef3f 100644
>> --- a/include/linux/vdpa.h
>> +++ b/include/linux/vdpa.h
>> @@ -282,6 +282,12 @@ struct vdpa_map_file {
>>   *                             @iova: iova to be unmapped
>>   *                             @size: size of the area
>>   *                             Returns integer: success (0) or error (< 0)
>> + * @bind_mm:                   Bind the device to a specific address space
>> + *                             so the vDPA framework can use VA when this
>> + *                             callback is implemented. (optional)
>> + *                             @vdev: vdpa device
>> + *                             @mm: address space to bind
>
>Do we need an unbind or did a NULL mm mean unbind?

Yep, your comment in patch 6 makes it necessary. I will add it!

>
>> + *                             @owner: process that owns the address space
>
>Any reason we need the task_struct here?

Mainly to attach to kthread to the process cgroups, but that part is 
still in TODO since I need to understand it better.

Maybe we can remove the task_struct here and use `current` directly in 
the callback.

Thanks,
Stefano

Re: [RFC PATCH 1/6] vdpa: add bind_mm callback

[Stefano Garzarella]

On Fri, Dec 16, 2022 at 02:37:45PM +0800, Jason Wang wrote:
>On Thu, Dec 15, 2022 at 12:30 AM Stefano Garzarella <sgarzare@redhat.com> wrote:
>>
>> This new optional callback is used to bind the device to a specific
>> address space so the vDPA framework can use VA when this callback
>> is implemented.
>>
>> Suggested-by: Jason Wang <jasowang@redhat.com>
>> Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
>> ---
>>  include/linux/vdpa.h | 8 ++++++++
>>  1 file changed, 8 insertions(+)
>>
>> diff --git a/include/linux/vdpa.h b/include/linux/vdpa.h
>> index 6d0f5e4e82c2..34388e21ef3f 100644
>> --- a/include/linux/vdpa.h
>> +++ b/include/linux/vdpa.h
>> @@ -282,6 +282,12 @@ struct vdpa_map_file {
>>   *                             @iova: iova to be unmapped
>>   *                             @size: size of the area
>>   *                             Returns integer: success (0) or error (< 0)
>> + * @bind_mm:                   Bind the device to a specific address space
>> + *                             so the vDPA framework can use VA when this
>> + *                             callback is implemented. (optional)
>> + *                             @vdev: vdpa device
>> + *                             @mm: address space to bind
>
>Do we need an unbind or did a NULL mm mean unbind?

Yep, your comment in patch 6 makes it necessary. I will add it!

>
>> + *                             @owner: process that owns the address space
>
>Any reason we need the task_struct here?

Mainly to attach to kthread to the process cgroups, but that part is 
still in TODO since I need to understand it better.

Maybe we can remove the task_struct here and use `current` directly in 
the callback.

Thanks,
Stefano

_______________________________________________
Virtualization mailing list
Virtualization@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

Re: [RFC PATCH 6/6] vdpa_sim: add support for user VA

[Eugenio Perez Martin]

On Wed, Dec 14, 2022 at 5:31 PM Stefano Garzarella <sgarzare@redhat.com> wrote:
>
> The new "use_va" module parameter (default: false) is used in

Why not true by default? I'd say it makes more sense for the simulator
to use va mode and only use pa for testing it.

> vdpa_alloc_device() to inform the vDPA framework that the device
> supports VA.
>
> vringh is initialized to use VA only when "use_va" is true and the
> user's mm has been bound. So, only when the bus supports user VA
> (e.g. vhost-vdpa).
>
> vdpasim_mm_work_fn work is used to attach the kthread to the user
> address space when the .bind_mm callback is invoked, and to detach
> it when the device is reset.
>
> Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
> ---
>  drivers/vdpa/vdpa_sim/vdpa_sim.h |   1 +
>  drivers/vdpa/vdpa_sim/vdpa_sim.c | 104 ++++++++++++++++++++++++++++++-
>  2 files changed, 103 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/vdpa/vdpa_sim/vdpa_sim.h b/drivers/vdpa/vdpa_sim/vdpa_sim.h
> index 07ef53ea375e..1b010e5c0445 100644
> --- a/drivers/vdpa/vdpa_sim/vdpa_sim.h
> +++ b/drivers/vdpa/vdpa_sim/vdpa_sim.h
> @@ -55,6 +55,7 @@ struct vdpasim {
>         struct vdpasim_virtqueue *vqs;
>         struct kthread_worker *worker;
>         struct kthread_work work;
> +       struct mm_struct *mm_bound;
>         struct vdpasim_dev_attr dev_attr;
>         /* spinlock to synchronize virtqueue state */
>         spinlock_t lock;
> diff --git a/drivers/vdpa/vdpa_sim/vdpa_sim.c b/drivers/vdpa/vdpa_sim/vdpa_sim.c
> index 36a1d2e0a6ba..6e07cedef30c 100644
> --- a/drivers/vdpa/vdpa_sim/vdpa_sim.c
> +++ b/drivers/vdpa/vdpa_sim/vdpa_sim.c
> @@ -36,10 +36,90 @@ module_param(max_iotlb_entries, int, 0444);
>  MODULE_PARM_DESC(max_iotlb_entries,
>                  "Maximum number of iotlb entries for each address space. 0 means unlimited. (default: 2048)");
>
> +static bool use_va;
> +module_param(use_va, bool, 0444);
> +MODULE_PARM_DESC(use_va, "Enable the device's ability to use VA");
> +
>  #define VDPASIM_QUEUE_ALIGN PAGE_SIZE
>  #define VDPASIM_QUEUE_MAX 256
>  #define VDPASIM_VENDOR_ID 0
>
> +struct vdpasim_mm_work {
> +       struct kthread_work work;
> +       struct task_struct *owner;
> +       struct mm_struct *mm;
> +       bool bind;
> +       int ret;
> +};
> +
> +static void vdpasim_mm_work_fn(struct kthread_work *work)
> +{
> +       struct vdpasim_mm_work *mm_work =
> +               container_of(work, struct vdpasim_mm_work, work);
> +
> +       mm_work->ret = 0;
> +
> +       if (mm_work->bind) {
> +               kthread_use_mm(mm_work->mm);
> +#if 0
> +               if (mm_work->owner)
> +                       mm_work->ret = cgroup_attach_task_all(mm_work->owner,
> +                                                             current);
> +#endif
> +       } else {
> +#if 0
> +               //TODO: check it
> +               cgroup_release(current);
> +#endif
> +               kthread_unuse_mm(mm_work->mm);
> +       }
> +}
> +
> +static void vdpasim_worker_queue_mm(struct vdpasim *vdpasim,
> +                                   struct vdpasim_mm_work *mm_work)
> +{
> +       struct kthread_work *work = &mm_work->work;
> +
> +       kthread_init_work(work, vdpasim_mm_work_fn);
> +       kthread_queue_work(vdpasim->worker, work);
> +
> +       spin_unlock(&vdpasim->lock);
> +       kthread_flush_work(work);
> +       spin_lock(&vdpasim->lock);
> +}
> +
> +static int vdpasim_worker_bind_mm(struct vdpasim *vdpasim,
> +                                 struct mm_struct *new_mm,
> +                                 struct task_struct *owner)
> +{
> +       struct vdpasim_mm_work mm_work;
> +
> +       mm_work.owner = owner;
> +       mm_work.mm = new_mm;
> +       mm_work.bind = true;
> +
> +       vdpasim_worker_queue_mm(vdpasim, &mm_work);
> +
> +       if (!mm_work.ret)
> +               vdpasim->mm_bound = new_mm;
> +
> +       return mm_work.ret;
> +}
> +
> +static void vdpasim_worker_unbind_mm(struct vdpasim *vdpasim)
> +{
> +       struct vdpasim_mm_work mm_work;
> +
> +       if (!vdpasim->mm_bound)
> +               return;
> +
> +       mm_work.mm = vdpasim->mm_bound;
> +       mm_work.bind = false;
> +
> +       vdpasim_worker_queue_mm(vdpasim, &mm_work);
> +
> +       vdpasim->mm_bound = NULL;
> +}
>  static struct vdpasim *vdpa_to_sim(struct vdpa_device *vdpa)
>  {
>         return container_of(vdpa, struct vdpasim, vdpa);
> @@ -66,8 +146,10 @@ static void vdpasim_vq_notify(struct vringh *vring)
>  static void vdpasim_queue_ready(struct vdpasim *vdpasim, unsigned int idx)
>  {
>         struct vdpasim_virtqueue *vq = &vdpasim->vqs[idx];
> +       bool va_enabled = use_va && vdpasim->mm_bound;
>
> -       vringh_init_iotlb(&vq->vring, vdpasim->features, vq->num, false, false,
> +       vringh_init_iotlb(&vq->vring, vdpasim->features, vq->num, false,
> +                         va_enabled,
>                           (struct vring_desc *)(uintptr_t)vq->desc_addr,
>                           (struct vring_avail *)
>                           (uintptr_t)vq->driver_addr,
> @@ -96,6 +178,9 @@ static void vdpasim_do_reset(struct vdpasim *vdpasim)
>  {
>         int i;
>
> +       //TODO: should we cancel the works?
> +       vdpasim_worker_unbind_mm(vdpasim);
> +
>         spin_lock(&vdpasim->iommu_lock);
>
>         for (i = 0; i < vdpasim->dev_attr.nvqs; i++) {
> @@ -275,7 +360,7 @@ struct vdpasim *vdpasim_create(struct vdpasim_dev_attr *dev_attr,
>
>         vdpasim = vdpa_alloc_device(struct vdpasim, vdpa, NULL, ops,
>                                     dev_attr->ngroups, dev_attr->nas,
> -                                   dev_attr->name, false);
> +                                   dev_attr->name, use_va);
>         if (IS_ERR(vdpasim)) {
>                 ret = PTR_ERR(vdpasim);
>                 goto err_alloc;
> @@ -657,6 +742,19 @@ static int vdpasim_set_map(struct vdpa_device *vdpa, unsigned int asid,
>         return ret;
>  }
>
> +static int vdpasim_bind_mm(struct vdpa_device *vdpa, struct mm_struct *mm,
> +                          struct task_struct *owner)
> +{
> +       struct vdpasim *vdpasim = vdpa_to_sim(vdpa);
> +       int ret;
> +
> +       spin_lock(&vdpasim->lock);
> +       ret = vdpasim_worker_bind_mm(vdpasim, mm, owner);
> +       spin_unlock(&vdpasim->lock);
> +
> +       return ret;
> +}
> +
>  static int vdpasim_dma_map(struct vdpa_device *vdpa, unsigned int asid,
>                            u64 iova, u64 size,
>                            u64 pa, u32 perm, void *opaque)
> @@ -744,6 +842,7 @@ static const struct vdpa_config_ops vdpasim_config_ops = {
>         .set_group_asid         = vdpasim_set_group_asid,
>         .dma_map                = vdpasim_dma_map,
>         .dma_unmap              = vdpasim_dma_unmap,
> +       .bind_mm                = vdpasim_bind_mm,
>         .free                   = vdpasim_free,
>  };
>
> @@ -776,6 +875,7 @@ static const struct vdpa_config_ops vdpasim_batch_config_ops = {
>         .get_iova_range         = vdpasim_get_iova_range,
>         .set_group_asid         = vdpasim_set_group_asid,
>         .set_map                = vdpasim_set_map,
> +       .bind_mm                = vdpasim_bind_mm,
>         .free                   = vdpasim_free,
>  };
>
> --
> 2.38.1
>

Re: [RFC PATCH 6/6] vdpa_sim: add support for user VA

[Stefano Garzarella]

On Wed, Dec 21, 2022 at 08:17:41AM +0100, Eugenio Perez Martin wrote:
>On Wed, Dec 14, 2022 at 5:31 PM Stefano Garzarella <sgarzare@redhat.com> wrote:
>>
>> The new "use_va" module parameter (default: false) is used in
>
>Why not true by default? I'd say it makes more sense for the simulator
>to use va mode and only use pa for testing it.

Yep, you are right. I'll change it in the version.
I initially left it at false because we usually use the simulator to 
test the paths that would be used for the real hardware.

Thanks,
Stefano

_______________________________________________
Virtualization mailing list
Virtualization@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

Re: [RFC PATCH 6/6] vdpa_sim: add support for user VA

[Stefano Garzarella]

On Wed, Dec 21, 2022 at 08:17:41AM +0100, Eugenio Perez Martin wrote:
>On Wed, Dec 14, 2022 at 5:31 PM Stefano Garzarella <sgarzare@redhat.com> wrote:
>>
>> The new "use_va" module parameter (default: false) is used in
>
>Why not true by default? I'd say it makes more sense for the simulator
>to use va mode and only use pa for testing it.

Yep, you are right. I'll change it in the version.
I initially left it at false because we usually use the simulator to 
test the paths that would be used for the real hardware.

Thanks,
Stefano

Re: [RFC PATCH 1/6] vdpa: add bind_mm callback

[Jason Wang]

在 2022/12/16 16:17, Stefano Garzarella 写道:
> On Fri, Dec 16, 2022 at 02:37:45PM +0800, Jason Wang wrote:
>> On Thu, Dec 15, 2022 at 12:30 AM Stefano Garzarella 
>> <sgarzare@redhat.com> wrote:
>>>
>>> This new optional callback is used to bind the device to a specific
>>> address space so the vDPA framework can use VA when this callback
>>> is implemented.
>>>
>>> Suggested-by: Jason Wang <jasowang@redhat.com>
>>> Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
>>> ---
>>>  include/linux/vdpa.h | 8 ++++++++
>>>  1 file changed, 8 insertions(+)
>>>
>>> diff --git a/include/linux/vdpa.h b/include/linux/vdpa.h
>>> index 6d0f5e4e82c2..34388e21ef3f 100644
>>> --- a/include/linux/vdpa.h
>>> +++ b/include/linux/vdpa.h
>>> @@ -282,6 +282,12 @@ struct vdpa_map_file {
>>>   *                             @iova: iova to be unmapped
>>>   *                             @size: size of the area
>>>   *                             Returns integer: success (0) or 
>>> error (< 0)
>>> + * @bind_mm:                   Bind the device to a specific 
>>> address space
>>> + *                             so the vDPA framework can use VA 
>>> when this
>>> + *                             callback is implemented. (optional)
>>> + *                             @vdev: vdpa device
>>> + *                             @mm: address space to bind
>>
>> Do we need an unbind or did a NULL mm mean unbind?
>
> Yep, your comment in patch 6 makes it necessary. I will add it!
>
>>
>>> + *                             @owner: process that owns the 
>>> address space
>>
>> Any reason we need the task_struct here?
>
> Mainly to attach to kthread to the process cgroups, but that part is 
> still in TODO since I need to understand it better.


Ok I see.


>
> Maybe we can remove the task_struct here and use `current` directly in 
> the callback.


Yes, it's easier to start without cgroup and we can add it on top.

Thanks


>
> Thanks,
> Stefano
>

Re: [RFC PATCH 1/6] vdpa: add bind_mm callback

[Jason Wang]

在 2022/12/16 16:17, Stefano Garzarella 写道:
> On Fri, Dec 16, 2022 at 02:37:45PM +0800, Jason Wang wrote:
>> On Thu, Dec 15, 2022 at 12:30 AM Stefano Garzarella 
>> <sgarzare@redhat.com> wrote:
>>>
>>> This new optional callback is used to bind the device to a specific
>>> address space so the vDPA framework can use VA when this callback
>>> is implemented.
>>>
>>> Suggested-by: Jason Wang <jasowang@redhat.com>
>>> Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
>>> ---
>>>  include/linux/vdpa.h | 8 ++++++++
>>>  1 file changed, 8 insertions(+)
>>>
>>> diff --git a/include/linux/vdpa.h b/include/linux/vdpa.h
>>> index 6d0f5e4e82c2..34388e21ef3f 100644
>>> --- a/include/linux/vdpa.h
>>> +++ b/include/linux/vdpa.h
>>> @@ -282,6 +282,12 @@ struct vdpa_map_file {
>>>   *                             @iova: iova to be unmapped
>>>   *                             @size: size of the area
>>>   *                             Returns integer: success (0) or 
>>> error (< 0)
>>> + * @bind_mm:                   Bind the device to a specific 
>>> address space
>>> + *                             so the vDPA framework can use VA 
>>> when this
>>> + *                             callback is implemented. (optional)
>>> + *                             @vdev: vdpa device
>>> + *                             @mm: address space to bind
>>
>> Do we need an unbind or did a NULL mm mean unbind?
>
> Yep, your comment in patch 6 makes it necessary. I will add it!
>
>>
>>> + *                             @owner: process that owns the 
>>> address space
>>
>> Any reason we need the task_struct here?
>
> Mainly to attach to kthread to the process cgroups, but that part is 
> still in TODO since I need to understand it better.


Ok I see.


>
> Maybe we can remove the task_struct here and use `current` directly in 
> the callback.


Yes, it's easier to start without cgroup and we can add it on top.

Thanks


>
> Thanks,
> Stefano
>

_______________________________________________
Virtualization mailing list
Virtualization@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/virtualization