2022-06-16 13:45:57

by Eli Cohen

[permalink] [raw]
Subject: [PATCH RFC 2/3] vdpa/mlx5: Support different address spaces for control and data

Partition virtqueues to two different address spaces: oce for control
virtqueue which is implemented in software, and one for data virtqueus.

Signed-off-by: Eli Cohen <[email protected]>
---
drivers/vdpa/mlx5/core/mlx5_vdpa.h | 11 ++++
drivers/vdpa/mlx5/net/mlx5_vnet.c | 101 +++++++++++++++++++++++++----
2 files changed, 101 insertions(+), 11 deletions(-)

diff --git a/drivers/vdpa/mlx5/core/mlx5_vdpa.h b/drivers/vdpa/mlx5/core/mlx5_vdpa.h
index 44104093163b..6af9fdbb86b7 100644
--- a/drivers/vdpa/mlx5/core/mlx5_vdpa.h
+++ b/drivers/vdpa/mlx5/core/mlx5_vdpa.h
@@ -70,6 +70,16 @@ struct mlx5_vdpa_wq_ent {
struct mlx5_vdpa_dev *mvdev;
};

+enum {
+ MLX5_VDPA_DATAVQ_GROUP,
+ MLX5_VDPA_CVQ_GROUP,
+ MLX5_VDPA_NUMVQ_GROUPS
+};
+
+enum {
+ MLX5_VDPA_NUM_AS = MLX5_VDPA_NUMVQ_GROUPS
+};
+
struct mlx5_vdpa_dev {
struct vdpa_device vdev;
struct mlx5_core_dev *mdev;
@@ -85,6 +95,7 @@ struct mlx5_vdpa_dev {
struct mlx5_vdpa_mr mr;
struct mlx5_control_vq cvq;
struct workqueue_struct *wq;
+ unsigned int group2asid[MLX5_VDPA_NUMVQ_GROUPS];
};

int mlx5_vdpa_alloc_pd(struct mlx5_vdpa_dev *dev, u32 *pdn, u16 uid);
diff --git a/drivers/vdpa/mlx5/net/mlx5_vnet.c b/drivers/vdpa/mlx5/net/mlx5_vnet.c
index ea4bc8a0cd25..34bd81cb697c 100644
--- a/drivers/vdpa/mlx5/net/mlx5_vnet.c
+++ b/drivers/vdpa/mlx5/net/mlx5_vnet.c
@@ -2125,9 +2125,14 @@ static u32 mlx5_vdpa_get_vq_align(struct vdpa_device *vdev)
return PAGE_SIZE;
}

-static u32 mlx5_vdpa_get_vq_group(struct vdpa_device *vdpa, u16 idx)
+static u32 mlx5_vdpa_get_vq_group(struct vdpa_device *vdev, u16 idx)
{
- return 0;
+ struct mlx5_vdpa_dev *mvdev = to_mvdev(vdev);
+
+ if (is_ctrl_vq_idx(mvdev, idx))
+ return MLX5_VDPA_CVQ_GROUP;
+
+ return MLX5_VDPA_DATAVQ_GROUP;
}

enum { MLX5_VIRTIO_NET_F_GUEST_CSUM = 1 << 9,
@@ -2541,6 +2546,15 @@ static void mlx5_vdpa_set_status(struct vdpa_device *vdev, u8 status)
up_write(&ndev->reslock);
}

+static void init_group_to_asid_map(struct mlx5_vdpa_dev *mvdev)
+{
+ int i;
+
+ /* default mapping all groups are mapped to asid 0 */
+ for (i = 0; i < MLX5_VDPA_NUMVQ_GROUPS; i++)
+ mvdev->group2asid[i] = 0;
+}
+
static int mlx5_vdpa_reset(struct vdpa_device *vdev)
{
struct mlx5_vdpa_dev *mvdev = to_mvdev(vdev);
@@ -2559,7 +2573,9 @@ static int mlx5_vdpa_reset(struct vdpa_device *vdev)
ndev->mvdev.cvq.completed_desc = 0;
memset(ndev->event_cbs, 0, sizeof(*ndev->event_cbs) * (mvdev->max_vqs + 1));
ndev->mvdev.actual_features = 0;
+ init_group_to_asid_map(mvdev);
++mvdev->generation;
+
if (MLX5_CAP_GEN(mvdev->mdev, umem_uid_0)) {
if (mlx5_vdpa_create_mr(mvdev, NULL))
mlx5_vdpa_warn(mvdev, "create MR failed\n");
@@ -2597,26 +2613,76 @@ static u32 mlx5_vdpa_get_generation(struct vdpa_device *vdev)
return mvdev->generation;
}

-static int mlx5_vdpa_set_map(struct vdpa_device *vdev, unsigned int asid,
- struct vhost_iotlb *iotlb)
+static u32 get_group(struct mlx5_vdpa_dev *mvdev, unsigned int asid)
+{
+ u32 group;
+
+ for (group = 0; group < MLX5_VDPA_NUMVQ_GROUPS; group++) {
+ if (mvdev->group2asid[group] == asid)
+ return group;
+ }
+ return -EINVAL;
+}
+
+static int set_map_control(struct mlx5_vdpa_dev *mvdev, struct vhost_iotlb *iotlb)
+{
+ u64 start = 0ULL, last = 0ULL - 1;
+ struct vhost_iotlb_map *map;
+ int err = 0;
+
+ spin_lock(&mvdev->cvq.iommu_lock);
+ vhost_iotlb_reset(mvdev->cvq.iotlb);
+
+ for (map = vhost_iotlb_itree_first(iotlb, start, last); map;
+ map = vhost_iotlb_itree_next(map, start, last)) {
+ err = vhost_iotlb_add_range(mvdev->cvq.iotlb, map->start,
+ map->last, map->addr, map->perm);
+ if (err)
+ goto out;
+ }
+
+out:
+ spin_unlock(&mvdev->cvq.iommu_lock);
+ return err;
+}
+
+static int set_map_data(struct mlx5_vdpa_dev *mvdev, struct vhost_iotlb *iotlb)
{
- struct mlx5_vdpa_dev *mvdev = to_mvdev(vdev);
- struct mlx5_vdpa_net *ndev = to_mlx5_vdpa_ndev(mvdev);
bool change_map;
int err;

- down_write(&ndev->reslock);
-
err = mlx5_vdpa_handle_set_map(mvdev, iotlb, &change_map);
if (err) {
mlx5_vdpa_warn(mvdev, "set map failed(%d)\n", err);
- goto err;
+ return err;
}

if (change_map)
err = mlx5_vdpa_change_map(mvdev, iotlb);

-err:
+ return err;
+}
+
+static int mlx5_vdpa_set_map(struct vdpa_device *vdev, unsigned int asid,
+ struct vhost_iotlb *iotlb)
+{
+ struct mlx5_vdpa_dev *mvdev = to_mvdev(vdev);
+ struct mlx5_vdpa_net *ndev = to_mlx5_vdpa_ndev(mvdev);
+ u32 group;
+ int err;
+
+ down_write(&ndev->reslock);
+ group = get_group(mvdev, asid);
+ switch (group) {
+ case MLX5_VDPA_DATAVQ_GROUP:
+ err = set_map_data(mvdev, iotlb);
+ break;
+ case MLX5_VDPA_CVQ_GROUP:
+ err = set_map_control(mvdev, iotlb);
+ break;
+ default:
+ err = -EINVAL;
+ }
up_write(&ndev->reslock);
return err;
}
@@ -2796,6 +2862,18 @@ static int mlx5_vdpa_suspend(struct vdpa_device *vdev, bool suspend)
return 0;
}

+static int mlx5_set_group_asid(struct vdpa_device *vdev, u32 group,
+ unsigned int asid)
+{
+ struct mlx5_vdpa_dev *mvdev = to_mvdev(vdev);
+
+ if (group >= MLX5_VDPA_NUMVQ_GROUPS)
+ return -EINVAL;
+
+ mvdev->group2asid[group] = asid;
+ return 0;
+}
+
static const struct vdpa_config_ops mlx5_vdpa_ops = {
.set_vq_address = mlx5_vdpa_set_vq_address,
.set_vq_num = mlx5_vdpa_set_vq_num,
@@ -2825,6 +2903,7 @@ static const struct vdpa_config_ops mlx5_vdpa_ops = {
.set_config = mlx5_vdpa_set_config,
.get_generation = mlx5_vdpa_get_generation,
.set_map = mlx5_vdpa_set_map,
+ .set_group_asid = mlx5_set_group_asid,
.free = mlx5_vdpa_free,
.suspend = mlx5_vdpa_suspend,
};
@@ -3047,7 +3126,7 @@ static int mlx5_vdpa_dev_add(struct vdpa_mgmt_dev *v_mdev, const char *name,
}

ndev = vdpa_alloc_device(struct mlx5_vdpa_net, mvdev.vdev, mdev->device, &mlx5_vdpa_ops,
- 1, 1, name, false);
+ MLX5_VDPA_NUMVQ_GROUPS, MLX5_VDPA_NUM_AS, name, false);
if (IS_ERR(ndev))
return PTR_ERR(ndev);

--
2.35.1


2022-06-20 08:53:20

by Jason Wang

[permalink] [raw]
Subject: Re: [PATCH RFC 2/3] vdpa/mlx5: Support different address spaces for control and data

On Thu, Jun 16, 2022 at 9:27 PM Eli Cohen <[email protected]> wrote:
>
> Partition virtqueues to two different address spaces: oce for control

Typo, should be "one"

> virtqueue which is implemented in software, and one for data virtqueus.

And should be "virtqueues".

Other than this.

Acked-by: Jason Wang <[email protected]>

>
> Signed-off-by: Eli Cohen <[email protected]>
> ---
> drivers/vdpa/mlx5/core/mlx5_vdpa.h | 11 ++++
> drivers/vdpa/mlx5/net/mlx5_vnet.c | 101 +++++++++++++++++++++++++----
> 2 files changed, 101 insertions(+), 11 deletions(-)
>
> diff --git a/drivers/vdpa/mlx5/core/mlx5_vdpa.h b/drivers/vdpa/mlx5/core/mlx5_vdpa.h
> index 44104093163b..6af9fdbb86b7 100644
> --- a/drivers/vdpa/mlx5/core/mlx5_vdpa.h
> +++ b/drivers/vdpa/mlx5/core/mlx5_vdpa.h
> @@ -70,6 +70,16 @@ struct mlx5_vdpa_wq_ent {
> struct mlx5_vdpa_dev *mvdev;
> };
>
> +enum {
> + MLX5_VDPA_DATAVQ_GROUP,
> + MLX5_VDPA_CVQ_GROUP,
> + MLX5_VDPA_NUMVQ_GROUPS
> +};
> +
> +enum {
> + MLX5_VDPA_NUM_AS = MLX5_VDPA_NUMVQ_GROUPS
> +};
> +
> struct mlx5_vdpa_dev {
> struct vdpa_device vdev;
> struct mlx5_core_dev *mdev;
> @@ -85,6 +95,7 @@ struct mlx5_vdpa_dev {
> struct mlx5_vdpa_mr mr;
> struct mlx5_control_vq cvq;
> struct workqueue_struct *wq;
> + unsigned int group2asid[MLX5_VDPA_NUMVQ_GROUPS];
> };
>
> int mlx5_vdpa_alloc_pd(struct mlx5_vdpa_dev *dev, u32 *pdn, u16 uid);
> diff --git a/drivers/vdpa/mlx5/net/mlx5_vnet.c b/drivers/vdpa/mlx5/net/mlx5_vnet.c
> index ea4bc8a0cd25..34bd81cb697c 100644
> --- a/drivers/vdpa/mlx5/net/mlx5_vnet.c
> +++ b/drivers/vdpa/mlx5/net/mlx5_vnet.c
> @@ -2125,9 +2125,14 @@ static u32 mlx5_vdpa_get_vq_align(struct vdpa_device *vdev)
> return PAGE_SIZE;
> }
>
> -static u32 mlx5_vdpa_get_vq_group(struct vdpa_device *vdpa, u16 idx)
> +static u32 mlx5_vdpa_get_vq_group(struct vdpa_device *vdev, u16 idx)
> {
> - return 0;
> + struct mlx5_vdpa_dev *mvdev = to_mvdev(vdev);
> +
> + if (is_ctrl_vq_idx(mvdev, idx))
> + return MLX5_VDPA_CVQ_GROUP;
> +
> + return MLX5_VDPA_DATAVQ_GROUP;
> }
>
> enum { MLX5_VIRTIO_NET_F_GUEST_CSUM = 1 << 9,
> @@ -2541,6 +2546,15 @@ static void mlx5_vdpa_set_status(struct vdpa_device *vdev, u8 status)
> up_write(&ndev->reslock);
> }
>
> +static void init_group_to_asid_map(struct mlx5_vdpa_dev *mvdev)
> +{
> + int i;
> +
> + /* default mapping all groups are mapped to asid 0 */
> + for (i = 0; i < MLX5_VDPA_NUMVQ_GROUPS; i++)
> + mvdev->group2asid[i] = 0;
> +}
> +
> static int mlx5_vdpa_reset(struct vdpa_device *vdev)
> {
> struct mlx5_vdpa_dev *mvdev = to_mvdev(vdev);
> @@ -2559,7 +2573,9 @@ static int mlx5_vdpa_reset(struct vdpa_device *vdev)
> ndev->mvdev.cvq.completed_desc = 0;
> memset(ndev->event_cbs, 0, sizeof(*ndev->event_cbs) * (mvdev->max_vqs + 1));
> ndev->mvdev.actual_features = 0;
> + init_group_to_asid_map(mvdev);
> ++mvdev->generation;
> +
> if (MLX5_CAP_GEN(mvdev->mdev, umem_uid_0)) {
> if (mlx5_vdpa_create_mr(mvdev, NULL))
> mlx5_vdpa_warn(mvdev, "create MR failed\n");
> @@ -2597,26 +2613,76 @@ static u32 mlx5_vdpa_get_generation(struct vdpa_device *vdev)
> return mvdev->generation;
> }
>
> -static int mlx5_vdpa_set_map(struct vdpa_device *vdev, unsigned int asid,
> - struct vhost_iotlb *iotlb)
> +static u32 get_group(struct mlx5_vdpa_dev *mvdev, unsigned int asid)
> +{
> + u32 group;
> +
> + for (group = 0; group < MLX5_VDPA_NUMVQ_GROUPS; group++) {
> + if (mvdev->group2asid[group] == asid)
> + return group;
> + }
> + return -EINVAL;
> +}
> +
> +static int set_map_control(struct mlx5_vdpa_dev *mvdev, struct vhost_iotlb *iotlb)
> +{
> + u64 start = 0ULL, last = 0ULL - 1;
> + struct vhost_iotlb_map *map;
> + int err = 0;
> +
> + spin_lock(&mvdev->cvq.iommu_lock);
> + vhost_iotlb_reset(mvdev->cvq.iotlb);
> +
> + for (map = vhost_iotlb_itree_first(iotlb, start, last); map;
> + map = vhost_iotlb_itree_next(map, start, last)) {
> + err = vhost_iotlb_add_range(mvdev->cvq.iotlb, map->start,
> + map->last, map->addr, map->perm);
> + if (err)
> + goto out;
> + }
> +
> +out:
> + spin_unlock(&mvdev->cvq.iommu_lock);
> + return err;
> +}
> +
> +static int set_map_data(struct mlx5_vdpa_dev *mvdev, struct vhost_iotlb *iotlb)
> {
> - struct mlx5_vdpa_dev *mvdev = to_mvdev(vdev);
> - struct mlx5_vdpa_net *ndev = to_mlx5_vdpa_ndev(mvdev);
> bool change_map;
> int err;
>
> - down_write(&ndev->reslock);
> -
> err = mlx5_vdpa_handle_set_map(mvdev, iotlb, &change_map);
> if (err) {
> mlx5_vdpa_warn(mvdev, "set map failed(%d)\n", err);
> - goto err;
> + return err;
> }
>
> if (change_map)
> err = mlx5_vdpa_change_map(mvdev, iotlb);
>
> -err:
> + return err;
> +}
> +
> +static int mlx5_vdpa_set_map(struct vdpa_device *vdev, unsigned int asid,
> + struct vhost_iotlb *iotlb)
> +{
> + struct mlx5_vdpa_dev *mvdev = to_mvdev(vdev);
> + struct mlx5_vdpa_net *ndev = to_mlx5_vdpa_ndev(mvdev);
> + u32 group;
> + int err;
> +
> + down_write(&ndev->reslock);
> + group = get_group(mvdev, asid);
> + switch (group) {
> + case MLX5_VDPA_DATAVQ_GROUP:
> + err = set_map_data(mvdev, iotlb);
> + break;
> + case MLX5_VDPA_CVQ_GROUP:
> + err = set_map_control(mvdev, iotlb);
> + break;
> + default:
> + err = -EINVAL;
> + }
> up_write(&ndev->reslock);
> return err;
> }
> @@ -2796,6 +2862,18 @@ static int mlx5_vdpa_suspend(struct vdpa_device *vdev, bool suspend)
> return 0;
> }
>
> +static int mlx5_set_group_asid(struct vdpa_device *vdev, u32 group,
> + unsigned int asid)
> +{
> + struct mlx5_vdpa_dev *mvdev = to_mvdev(vdev);
> +
> + if (group >= MLX5_VDPA_NUMVQ_GROUPS)
> + return -EINVAL;
> +
> + mvdev->group2asid[group] = asid;
> + return 0;
> +}
> +
> static const struct vdpa_config_ops mlx5_vdpa_ops = {
> .set_vq_address = mlx5_vdpa_set_vq_address,
> .set_vq_num = mlx5_vdpa_set_vq_num,
> @@ -2825,6 +2903,7 @@ static const struct vdpa_config_ops mlx5_vdpa_ops = {
> .set_config = mlx5_vdpa_set_config,
> .get_generation = mlx5_vdpa_get_generation,
> .set_map = mlx5_vdpa_set_map,
> + .set_group_asid = mlx5_set_group_asid,
> .free = mlx5_vdpa_free,
> .suspend = mlx5_vdpa_suspend,
> };
> @@ -3047,7 +3126,7 @@ static int mlx5_vdpa_dev_add(struct vdpa_mgmt_dev *v_mdev, const char *name,
> }
>
> ndev = vdpa_alloc_device(struct mlx5_vdpa_net, mvdev.vdev, mdev->device, &mlx5_vdpa_ops,
> - 1, 1, name, false);
> + MLX5_VDPA_NUMVQ_GROUPS, MLX5_VDPA_NUM_AS, name, false);
> if (IS_ERR(ndev))
> return PTR_ERR(ndev);
>
> --
> 2.35.1
>

2022-06-20 09:18:49

by Eugenio Perez Martin

[permalink] [raw]
Subject: Re: [PATCH RFC 2/3] vdpa/mlx5: Support different address spaces for control and data

On Thu, Jun 16, 2022 at 3:27 PM Eli Cohen <[email protected]> wrote:
>
> Partition virtqueues to two different address spaces: oce for control
> virtqueue which is implemented in software, and one for data virtqueus.
>
> Signed-off-by: Eli Cohen <[email protected]>
> ---
> drivers/vdpa/mlx5/core/mlx5_vdpa.h | 11 ++++
> drivers/vdpa/mlx5/net/mlx5_vnet.c | 101 +++++++++++++++++++++++++----
> 2 files changed, 101 insertions(+), 11 deletions(-)
>
> diff --git a/drivers/vdpa/mlx5/core/mlx5_vdpa.h b/drivers/vdpa/mlx5/core/mlx5_vdpa.h
> index 44104093163b..6af9fdbb86b7 100644
> --- a/drivers/vdpa/mlx5/core/mlx5_vdpa.h
> +++ b/drivers/vdpa/mlx5/core/mlx5_vdpa.h
> @@ -70,6 +70,16 @@ struct mlx5_vdpa_wq_ent {
> struct mlx5_vdpa_dev *mvdev;
> };
>
> +enum {
> + MLX5_VDPA_DATAVQ_GROUP,
> + MLX5_VDPA_CVQ_GROUP,
> + MLX5_VDPA_NUMVQ_GROUPS
> +};
> +
> +enum {
> + MLX5_VDPA_NUM_AS = MLX5_VDPA_NUMVQ_GROUPS
> +};
> +
> struct mlx5_vdpa_dev {
> struct vdpa_device vdev;
> struct mlx5_core_dev *mdev;
> @@ -85,6 +95,7 @@ struct mlx5_vdpa_dev {
> struct mlx5_vdpa_mr mr;
> struct mlx5_control_vq cvq;
> struct workqueue_struct *wq;
> + unsigned int group2asid[MLX5_VDPA_NUMVQ_GROUPS];
> };
>
> int mlx5_vdpa_alloc_pd(struct mlx5_vdpa_dev *dev, u32 *pdn, u16 uid);
> diff --git a/drivers/vdpa/mlx5/net/mlx5_vnet.c b/drivers/vdpa/mlx5/net/mlx5_vnet.c
> index ea4bc8a0cd25..34bd81cb697c 100644
> --- a/drivers/vdpa/mlx5/net/mlx5_vnet.c
> +++ b/drivers/vdpa/mlx5/net/mlx5_vnet.c
> @@ -2125,9 +2125,14 @@ static u32 mlx5_vdpa_get_vq_align(struct vdpa_device *vdev)
> return PAGE_SIZE;
> }
>
> -static u32 mlx5_vdpa_get_vq_group(struct vdpa_device *vdpa, u16 idx)
> +static u32 mlx5_vdpa_get_vq_group(struct vdpa_device *vdev, u16 idx)
> {
> - return 0;
> + struct mlx5_vdpa_dev *mvdev = to_mvdev(vdev);
> +
> + if (is_ctrl_vq_idx(mvdev, idx))
> + return MLX5_VDPA_CVQ_GROUP;
> +
> + return MLX5_VDPA_DATAVQ_GROUP;
> }
>
> enum { MLX5_VIRTIO_NET_F_GUEST_CSUM = 1 << 9,
> @@ -2541,6 +2546,15 @@ static void mlx5_vdpa_set_status(struct vdpa_device *vdev, u8 status)
> up_write(&ndev->reslock);
> }
>
> +static void init_group_to_asid_map(struct mlx5_vdpa_dev *mvdev)
> +{
> + int i;
> +
> + /* default mapping all groups are mapped to asid 0 */
> + for (i = 0; i < MLX5_VDPA_NUMVQ_GROUPS; i++)
> + mvdev->group2asid[i] = 0;
> +}
> +
> static int mlx5_vdpa_reset(struct vdpa_device *vdev)
> {
> struct mlx5_vdpa_dev *mvdev = to_mvdev(vdev);
> @@ -2559,7 +2573,9 @@ static int mlx5_vdpa_reset(struct vdpa_device *vdev)
> ndev->mvdev.cvq.completed_desc = 0;
> memset(ndev->event_cbs, 0, sizeof(*ndev->event_cbs) * (mvdev->max_vqs + 1));
> ndev->mvdev.actual_features = 0;
> + init_group_to_asid_map(mvdev);
> ++mvdev->generation;
> +
> if (MLX5_CAP_GEN(mvdev->mdev, umem_uid_0)) {
> if (mlx5_vdpa_create_mr(mvdev, NULL))
> mlx5_vdpa_warn(mvdev, "create MR failed\n");
> @@ -2597,26 +2613,76 @@ static u32 mlx5_vdpa_get_generation(struct vdpa_device *vdev)
> return mvdev->generation;
> }
>
> -static int mlx5_vdpa_set_map(struct vdpa_device *vdev, unsigned int asid,
> - struct vhost_iotlb *iotlb)
> +static u32 get_group(struct mlx5_vdpa_dev *mvdev, unsigned int asid)
> +{
> + u32 group;
> +
> + for (group = 0; group < MLX5_VDPA_NUMVQ_GROUPS; group++) {
> + if (mvdev->group2asid[group] == asid)
> + return group;
> + }
> + return -EINVAL;
> +}
> +
> +static int set_map_control(struct mlx5_vdpa_dev *mvdev, struct vhost_iotlb *iotlb)
> +{
> + u64 start = 0ULL, last = 0ULL - 1;
> + struct vhost_iotlb_map *map;
> + int err = 0;
> +
> + spin_lock(&mvdev->cvq.iommu_lock);
> + vhost_iotlb_reset(mvdev->cvq.iotlb);
> +
> + for (map = vhost_iotlb_itree_first(iotlb, start, last); map;
> + map = vhost_iotlb_itree_next(map, start, last)) {
> + err = vhost_iotlb_add_range(mvdev->cvq.iotlb, map->start,
> + map->last, map->addr, map->perm);
> + if (err)
> + goto out;
> + }
> +
> +out:
> + spin_unlock(&mvdev->cvq.iommu_lock);
> + return err;
> +}
> +
> +static int set_map_data(struct mlx5_vdpa_dev *mvdev, struct vhost_iotlb *iotlb)
> {
> - struct mlx5_vdpa_dev *mvdev = to_mvdev(vdev);
> - struct mlx5_vdpa_net *ndev = to_mlx5_vdpa_ndev(mvdev);
> bool change_map;
> int err;
>
> - down_write(&ndev->reslock);
> -
> err = mlx5_vdpa_handle_set_map(mvdev, iotlb, &change_map);
> if (err) {
> mlx5_vdpa_warn(mvdev, "set map failed(%d)\n", err);
> - goto err;
> + return err;
> }
>
> if (change_map)
> err = mlx5_vdpa_change_map(mvdev, iotlb);
>
> -err:
> + return err;
> +}
> +
> +static int mlx5_vdpa_set_map(struct vdpa_device *vdev, unsigned int asid,
> + struct vhost_iotlb *iotlb)
> +{
> + struct mlx5_vdpa_dev *mvdev = to_mvdev(vdev);
> + struct mlx5_vdpa_net *ndev = to_mlx5_vdpa_ndev(mvdev);
> + u32 group;
> + int err;
> +
> + down_write(&ndev->reslock);
> + group = get_group(mvdev, asid);
> + switch (group) {
> + case MLX5_VDPA_DATAVQ_GROUP:
> + err = set_map_data(mvdev, iotlb);
> + break;
> + case MLX5_VDPA_CVQ_GROUP:
> + err = set_map_control(mvdev, iotlb);
> + break;
> + default:
> + err = -EINVAL;
> + }

This shouldn't be a switch, but to check the asid assigned to the
different vqs individually.

In the current qemu version with no ASID support, all vq groups (data
and cvq) are assigned to asid 0 at the device reset. In this case,
emulated cvq also needs to receive the mappings, because guest's CVQ
commands will go from the guest's ASID directly.

Thanks!

> up_write(&ndev->reslock);
> return err;
> }
> @@ -2796,6 +2862,18 @@ static int mlx5_vdpa_suspend(struct vdpa_device *vdev, bool suspend)
> return 0;
> }
>
> +static int mlx5_set_group_asid(struct vdpa_device *vdev, u32 group,
> + unsigned int asid)
> +{
> + struct mlx5_vdpa_dev *mvdev = to_mvdev(vdev);
> +
> + if (group >= MLX5_VDPA_NUMVQ_GROUPS)
> + return -EINVAL;
> +
> + mvdev->group2asid[group] = asid;
> + return 0;
> +}
> +
> static const struct vdpa_config_ops mlx5_vdpa_ops = {
> .set_vq_address = mlx5_vdpa_set_vq_address,
> .set_vq_num = mlx5_vdpa_set_vq_num,
> @@ -2825,6 +2903,7 @@ static const struct vdpa_config_ops mlx5_vdpa_ops = {
> .set_config = mlx5_vdpa_set_config,
> .get_generation = mlx5_vdpa_get_generation,
> .set_map = mlx5_vdpa_set_map,
> + .set_group_asid = mlx5_set_group_asid,
> .free = mlx5_vdpa_free,
> .suspend = mlx5_vdpa_suspend,
> };
> @@ -3047,7 +3126,7 @@ static int mlx5_vdpa_dev_add(struct vdpa_mgmt_dev *v_mdev, const char *name,
> }
>
> ndev = vdpa_alloc_device(struct mlx5_vdpa_net, mvdev.vdev, mdev->device, &mlx5_vdpa_ops,
> - 1, 1, name, false);
> + MLX5_VDPA_NUMVQ_GROUPS, MLX5_VDPA_NUM_AS, name, false);
> if (IS_ERR(ndev))
> return PTR_ERR(ndev);
>
> --
> 2.35.1
>

2022-06-20 10:05:00

by Jason Wang

[permalink] [raw]
Subject: Re: [PATCH RFC 2/3] vdpa/mlx5: Support different address spaces for control and data

On Mon, Jun 20, 2022 at 4:58 PM Eugenio Perez Martin
<[email protected]> wrote:
>
> On Thu, Jun 16, 2022 at 3:27 PM Eli Cohen <[email protected]> wrote:
> >
> > Partition virtqueues to two different address spaces: oce for control
> > virtqueue which is implemented in software, and one for data virtqueus.
> >
> > Signed-off-by: Eli Cohen <[email protected]>
> > ---
> > drivers/vdpa/mlx5/core/mlx5_vdpa.h | 11 ++++
> > drivers/vdpa/mlx5/net/mlx5_vnet.c | 101 +++++++++++++++++++++++++----
> > 2 files changed, 101 insertions(+), 11 deletions(-)
> >
> > diff --git a/drivers/vdpa/mlx5/core/mlx5_vdpa.h b/drivers/vdpa/mlx5/core/mlx5_vdpa.h
> > index 44104093163b..6af9fdbb86b7 100644
> > --- a/drivers/vdpa/mlx5/core/mlx5_vdpa.h
> > +++ b/drivers/vdpa/mlx5/core/mlx5_vdpa.h
> > @@ -70,6 +70,16 @@ struct mlx5_vdpa_wq_ent {
> > struct mlx5_vdpa_dev *mvdev;
> > };
> >
> > +enum {
> > + MLX5_VDPA_DATAVQ_GROUP,
> > + MLX5_VDPA_CVQ_GROUP,
> > + MLX5_VDPA_NUMVQ_GROUPS
> > +};
> > +
> > +enum {
> > + MLX5_VDPA_NUM_AS = MLX5_VDPA_NUMVQ_GROUPS
> > +};
> > +
> > struct mlx5_vdpa_dev {
> > struct vdpa_device vdev;
> > struct mlx5_core_dev *mdev;
> > @@ -85,6 +95,7 @@ struct mlx5_vdpa_dev {
> > struct mlx5_vdpa_mr mr;
> > struct mlx5_control_vq cvq;
> > struct workqueue_struct *wq;
> > + unsigned int group2asid[MLX5_VDPA_NUMVQ_GROUPS];
> > };
> >
> > int mlx5_vdpa_alloc_pd(struct mlx5_vdpa_dev *dev, u32 *pdn, u16 uid);
> > diff --git a/drivers/vdpa/mlx5/net/mlx5_vnet.c b/drivers/vdpa/mlx5/net/mlx5_vnet.c
> > index ea4bc8a0cd25..34bd81cb697c 100644
> > --- a/drivers/vdpa/mlx5/net/mlx5_vnet.c
> > +++ b/drivers/vdpa/mlx5/net/mlx5_vnet.c
> > @@ -2125,9 +2125,14 @@ static u32 mlx5_vdpa_get_vq_align(struct vdpa_device *vdev)
> > return PAGE_SIZE;
> > }
> >
> > -static u32 mlx5_vdpa_get_vq_group(struct vdpa_device *vdpa, u16 idx)
> > +static u32 mlx5_vdpa_get_vq_group(struct vdpa_device *vdev, u16 idx)
> > {
> > - return 0;
> > + struct mlx5_vdpa_dev *mvdev = to_mvdev(vdev);
> > +
> > + if (is_ctrl_vq_idx(mvdev, idx))
> > + return MLX5_VDPA_CVQ_GROUP;
> > +
> > + return MLX5_VDPA_DATAVQ_GROUP;
> > }
> >
> > enum { MLX5_VIRTIO_NET_F_GUEST_CSUM = 1 << 9,
> > @@ -2541,6 +2546,15 @@ static void mlx5_vdpa_set_status(struct vdpa_device *vdev, u8 status)
> > up_write(&ndev->reslock);
> > }
> >
> > +static void init_group_to_asid_map(struct mlx5_vdpa_dev *mvdev)
> > +{
> > + int i;
> > +
> > + /* default mapping all groups are mapped to asid 0 */
> > + for (i = 0; i < MLX5_VDPA_NUMVQ_GROUPS; i++)
> > + mvdev->group2asid[i] = 0;
> > +}
> > +
> > static int mlx5_vdpa_reset(struct vdpa_device *vdev)
> > {
> > struct mlx5_vdpa_dev *mvdev = to_mvdev(vdev);
> > @@ -2559,7 +2573,9 @@ static int mlx5_vdpa_reset(struct vdpa_device *vdev)
> > ndev->mvdev.cvq.completed_desc = 0;
> > memset(ndev->event_cbs, 0, sizeof(*ndev->event_cbs) * (mvdev->max_vqs + 1));
> > ndev->mvdev.actual_features = 0;
> > + init_group_to_asid_map(mvdev);
> > ++mvdev->generation;
> > +
> > if (MLX5_CAP_GEN(mvdev->mdev, umem_uid_0)) {
> > if (mlx5_vdpa_create_mr(mvdev, NULL))
> > mlx5_vdpa_warn(mvdev, "create MR failed\n");
> > @@ -2597,26 +2613,76 @@ static u32 mlx5_vdpa_get_generation(struct vdpa_device *vdev)
> > return mvdev->generation;
> > }
> >
> > -static int mlx5_vdpa_set_map(struct vdpa_device *vdev, unsigned int asid,
> > - struct vhost_iotlb *iotlb)
> > +static u32 get_group(struct mlx5_vdpa_dev *mvdev, unsigned int asid)
> > +{
> > + u32 group;
> > +
> > + for (group = 0; group < MLX5_VDPA_NUMVQ_GROUPS; group++) {
> > + if (mvdev->group2asid[group] == asid)
> > + return group;
> > + }
> > + return -EINVAL;
> > +}
> > +
> > +static int set_map_control(struct mlx5_vdpa_dev *mvdev, struct vhost_iotlb *iotlb)
> > +{
> > + u64 start = 0ULL, last = 0ULL - 1;
> > + struct vhost_iotlb_map *map;
> > + int err = 0;
> > +
> > + spin_lock(&mvdev->cvq.iommu_lock);
> > + vhost_iotlb_reset(mvdev->cvq.iotlb);
> > +
> > + for (map = vhost_iotlb_itree_first(iotlb, start, last); map;
> > + map = vhost_iotlb_itree_next(map, start, last)) {
> > + err = vhost_iotlb_add_range(mvdev->cvq.iotlb, map->start,
> > + map->last, map->addr, map->perm);
> > + if (err)
> > + goto out;
> > + }
> > +
> > +out:
> > + spin_unlock(&mvdev->cvq.iommu_lock);
> > + return err;
> > +}
> > +
> > +static int set_map_data(struct mlx5_vdpa_dev *mvdev, struct vhost_iotlb *iotlb)
> > {
> > - struct mlx5_vdpa_dev *mvdev = to_mvdev(vdev);
> > - struct mlx5_vdpa_net *ndev = to_mlx5_vdpa_ndev(mvdev);
> > bool change_map;
> > int err;
> >
> > - down_write(&ndev->reslock);
> > -
> > err = mlx5_vdpa_handle_set_map(mvdev, iotlb, &change_map);
> > if (err) {
> > mlx5_vdpa_warn(mvdev, "set map failed(%d)\n", err);
> > - goto err;
> > + return err;
> > }
> >
> > if (change_map)
> > err = mlx5_vdpa_change_map(mvdev, iotlb);
> >
> > -err:
> > + return err;
> > +}
> > +
> > +static int mlx5_vdpa_set_map(struct vdpa_device *vdev, unsigned int asid,
> > + struct vhost_iotlb *iotlb)
> > +{
> > + struct mlx5_vdpa_dev *mvdev = to_mvdev(vdev);
> > + struct mlx5_vdpa_net *ndev = to_mlx5_vdpa_ndev(mvdev);
> > + u32 group;
> > + int err;
> > +
> > + down_write(&ndev->reslock);
> > + group = get_group(mvdev, asid);
> > + switch (group) {
> > + case MLX5_VDPA_DATAVQ_GROUP:
> > + err = set_map_data(mvdev, iotlb);
> > + break;
> > + case MLX5_VDPA_CVQ_GROUP:
> > + err = set_map_control(mvdev, iotlb);
> > + break;
> > + default:
> > + err = -EINVAL;
> > + }
>
> This shouldn't be a switch, but to check the asid assigned to the
> different vqs individually.
>
> In the current qemu version with no ASID support, all vq groups (data
> and cvq) are assigned to asid 0 at the device reset. In this case,
> emulated cvq also needs to receive the mappings, because guest's CVQ
> commands will go from the guest's ASID directly.

Ack.

Thanks

>
> Thanks!
>
> > up_write(&ndev->reslock);
> > return err;
> > }
> > @@ -2796,6 +2862,18 @@ static int mlx5_vdpa_suspend(struct vdpa_device *vdev, bool suspend)
> > return 0;
> > }
> >
> > +static int mlx5_set_group_asid(struct vdpa_device *vdev, u32 group,
> > + unsigned int asid)
> > +{
> > + struct mlx5_vdpa_dev *mvdev = to_mvdev(vdev);
> > +
> > + if (group >= MLX5_VDPA_NUMVQ_GROUPS)
> > + return -EINVAL;
> > +
> > + mvdev->group2asid[group] = asid;
> > + return 0;
> > +}
> > +
> > static const struct vdpa_config_ops mlx5_vdpa_ops = {
> > .set_vq_address = mlx5_vdpa_set_vq_address,
> > .set_vq_num = mlx5_vdpa_set_vq_num,
> > @@ -2825,6 +2903,7 @@ static const struct vdpa_config_ops mlx5_vdpa_ops = {
> > .set_config = mlx5_vdpa_set_config,
> > .get_generation = mlx5_vdpa_get_generation,
> > .set_map = mlx5_vdpa_set_map,
> > + .set_group_asid = mlx5_set_group_asid,
> > .free = mlx5_vdpa_free,
> > .suspend = mlx5_vdpa_suspend,
> > };
> > @@ -3047,7 +3126,7 @@ static int mlx5_vdpa_dev_add(struct vdpa_mgmt_dev *v_mdev, const char *name,
> > }
> >
> > ndev = vdpa_alloc_device(struct mlx5_vdpa_net, mvdev.vdev, mdev->device, &mlx5_vdpa_ops,
> > - 1, 1, name, false);
> > + MLX5_VDPA_NUMVQ_GROUPS, MLX5_VDPA_NUM_AS, name, false);
> > if (IS_ERR(ndev))
> > return PTR_ERR(ndev);
> >
> > --
> > 2.35.1
> >
>