The following three patch series adds support to read vendor statistics
for a vdpa device.
The first patch is just a cleanup fix.
The second patch lays the ground to allow an upstream driver to provide
statistics in the form of an attribute name/attribute value pairs.
The third patch implements this for mlx5_vdpa which gives received
descriptors and completed descriptors information for all the
virtqueues.
v4 -> v5:
See inside the individual patches
Eli Cohen (3):
vdpa: Fix error logic in vdpa_nl_cmd_dev_get_doit
vdpa: Add support for querying vendor statistics
vdpa/mlx5: Add support for reading descriptor statistics
drivers/vdpa/mlx5/core/mlx5_vdpa.h | 2 +
drivers/vdpa/mlx5/net/mlx5_vnet.c | 165 +++++++++++++++++++++++++++++
drivers/vdpa/vdpa.c | 153 +++++++++++++++++++++++++-
include/linux/mlx5/mlx5_ifc.h | 1 +
include/linux/mlx5/mlx5_ifc_vdpa.h | 39 +++++++
include/linux/vdpa.h | 3 +
include/uapi/linux/vdpa.h | 6 ++
7 files changed, 365 insertions(+), 4 deletions(-)
--
2.35.1
Allows to read vendor statistics of a vdpa device. The specific
statistics data are received from the upstream driver in the form of an
(attribute name, attribute value) pairs.
An example of statistics for mlx5_vdpa device are:
received_desc - number of descriptors received by the virtqueue
completed_desc - number of descriptors completed by the virtqueue
A descriptor using indirect buffers is still counted as 1. In addition,
N chained descriptors are counted correctly N times as one would expect.
A new callback was added to vdpa_config_ops which provides the means for
the vdpa driver to return statistics results.
The interface allows for reading all the supported virtqueues, including
the control virtqueue if it exists.
Below are some examples taken from mlx5_vdpa which are introduced in the
following patch:
1. Read statistics for the virtqueue at index 1
$ vdpa dev vstats show vdpa-a qidx 1
vdpa-a:
queue_type tx queue_index 1 received_desc 3844836 completed_desc 3844836
2. Read statistics for the virtqueue at index 32
$ vdpa dev vstats show vdpa-a qidx 32
vdpa-a:
queue_type control_vq queue_index 32 received_desc 62 completed_desc 62
3. Read statisitics for the virtqueue at index 0 with json output
$ vdpa -j dev vstats show vdpa-a qidx 0
{"vstats":{"vdpa-a":{
"queue_type":"rx","queue_index":0,"name":"received_desc","value":417776,\
"name":"completed_desc","value":417548}}}
4. Read statistics for the virtqueue at index 0 with preety json output
$ vdpa -jp dev vstats show vdpa-a qidx 0
{
"vstats": {
"vdpa-a": {
"queue_type": "rx",
"queue_index": 0,
"name": "received_desc",
"value": 417776,
"name": "completed_desc",
"value": 417548
}
}
}
Signed-off-by: Eli Cohen <[email protected]>
---
v4 -> v5:
1. Remove unused macro definition
2. Avoid policy on queue index. Do validty check in the implementation.
3. Restrict to VIRTIO_ID_NET devices only.
4. Improve netlink error messages.
drivers/vdpa/vdpa.c | 140 ++++++++++++++++++++++++++++++++++++++
include/linux/vdpa.h | 3 +
include/uapi/linux/vdpa.h | 6 ++
3 files changed, 149 insertions(+)
diff --git a/drivers/vdpa/vdpa.c b/drivers/vdpa/vdpa.c
index fac89a0d8178..91f4c13c7c7c 100644
--- a/drivers/vdpa/vdpa.c
+++ b/drivers/vdpa/vdpa.c
@@ -914,6 +914,86 @@ vdpa_dev_config_fill(struct vdpa_device *vdev, struct sk_buff *msg, u32 portid,
return err;
}
+static int vdpa_fill_stats_rec(struct vdpa_device *vdev, struct sk_buff *msg,
+ struct genl_info *info, u32 index)
+{
+ int err;
+
+ err = vdev->config->get_vendor_vq_stats(vdev, index, msg, info->extack);
+ if (err)
+ return err;
+
+ if (nla_put_u32(msg, VDPA_ATTR_DEV_QUEUE_INDEX, index))
+ return -EMSGSIZE;
+
+ return 0;
+}
+
+static int vendor_stats_fill(struct vdpa_device *vdev, struct sk_buff *msg,
+ struct genl_info *info, u32 index)
+{
+ int err;
+
+ if (!vdev->config->get_vendor_vq_stats)
+ return -EOPNOTSUPP;
+
+ err = vdpa_fill_stats_rec(vdev, msg, info, index);
+ if (err)
+ return err;
+
+ return 0;
+}
+
+static int vdpa_dev_vendor_stats_fill(struct vdpa_device *vdev,
+ struct sk_buff *msg,
+ struct genl_info *info, u32 index)
+{
+ u32 device_id;
+ void *hdr;
+ int err;
+ u32 portid = info->snd_portid;
+ u32 seq = info->snd_seq;
+ u32 flags = 0;
+
+ hdr = genlmsg_put(msg, portid, seq, &vdpa_nl_family, flags,
+ VDPA_CMD_DEV_VSTATS_GET);
+ if (!hdr)
+ return -EMSGSIZE;
+
+ if (nla_put_string(msg, VDPA_ATTR_DEV_NAME, dev_name(&vdev->dev))) {
+ err = -EMSGSIZE;
+ goto undo_msg;
+ }
+
+ device_id = vdev->config->get_device_id(vdev);
+ if (nla_put_u32(msg, VDPA_ATTR_DEV_ID, device_id)) {
+ err = -EMSGSIZE;
+ goto undo_msg;
+ }
+
+ switch (device_id) {
+ case VIRTIO_ID_NET:
+ if (index > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX) {
+ NL_SET_ERR_MSG_MOD(info->extack, "queue index excceeds max value");
+ err = -ERANGE;
+ break;
+ }
+
+ err = vendor_stats_fill(vdev, msg, info, index);
+ break;
+ default:
+ err = -EOPNOTSUPP;
+ break;
+ }
+ genlmsg_end(msg, hdr);
+
+ return err;
+
+undo_msg:
+ genlmsg_cancel(msg, hdr);
+ return err;
+}
+
static int vdpa_nl_cmd_dev_config_get_doit(struct sk_buff *skb, struct genl_info *info)
{
struct vdpa_device *vdev;
@@ -995,6 +1075,60 @@ vdpa_nl_cmd_dev_config_get_dumpit(struct sk_buff *msg, struct netlink_callback *
return msg->len;
}
+static int vdpa_nl_cmd_dev_stats_get_doit(struct sk_buff *skb,
+ struct genl_info *info)
+{
+ struct vdpa_device *vdev;
+ struct sk_buff *msg;
+ const char *devname;
+ struct device *dev;
+ u32 index;
+ int err;
+
+ if (!info->attrs[VDPA_ATTR_DEV_NAME])
+ return -EINVAL;
+
+ if (!info->attrs[VDPA_ATTR_DEV_QUEUE_INDEX])
+ return -EINVAL;
+
+ devname = nla_data(info->attrs[VDPA_ATTR_DEV_NAME]);
+ msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
+ if (!msg)
+ return -ENOMEM;
+
+ index = nla_get_u32(info->attrs[VDPA_ATTR_DEV_QUEUE_INDEX]);
+ mutex_lock(&vdpa_dev_mutex);
+ dev = bus_find_device(&vdpa_bus, NULL, devname, vdpa_name_match);
+ if (!dev) {
+ NL_SET_ERR_MSG_MOD(info->extack, "device not found");
+ err = -ENODEV;
+ goto dev_err;
+ }
+ vdev = container_of(dev, struct vdpa_device, dev);
+ if (!vdev->mdev) {
+ NL_SET_ERR_MSG_MOD(info->extack, "unmanaged vdpa device");
+ err = -EINVAL;
+ goto mdev_err;
+ }
+ err = vdpa_dev_vendor_stats_fill(vdev, msg, info, index);
+ if (err)
+ goto mdev_err;
+
+ err = genlmsg_reply(msg, info);
+
+ put_device(dev);
+ mutex_unlock(&vdpa_dev_mutex);
+
+ return err;
+
+mdev_err:
+ put_device(dev);
+dev_err:
+ nlmsg_free(msg);
+ mutex_unlock(&vdpa_dev_mutex);
+ return err;
+}
+
static const struct nla_policy vdpa_nl_policy[VDPA_ATTR_MAX + 1] = {
[VDPA_ATTR_MGMTDEV_BUS_NAME] = { .type = NLA_NUL_STRING },
[VDPA_ATTR_MGMTDEV_DEV_NAME] = { .type = NLA_STRING },
@@ -1035,6 +1169,12 @@ static const struct genl_ops vdpa_nl_ops[] = {
.doit = vdpa_nl_cmd_dev_config_get_doit,
.dumpit = vdpa_nl_cmd_dev_config_get_dumpit,
},
+ {
+ .cmd = VDPA_CMD_DEV_VSTATS_GET,
+ .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
+ .doit = vdpa_nl_cmd_dev_stats_get_doit,
+ .flags = GENL_ADMIN_PERM,
+ },
};
static struct genl_family vdpa_nl_family __ro_after_init = {
diff --git a/include/linux/vdpa.h b/include/linux/vdpa.h
index 8943a209202e..2ae8443331e1 100644
--- a/include/linux/vdpa.h
+++ b/include/linux/vdpa.h
@@ -276,6 +276,9 @@ struct vdpa_config_ops {
const struct vdpa_vq_state *state);
int (*get_vq_state)(struct vdpa_device *vdev, u16 idx,
struct vdpa_vq_state *state);
+ int (*get_vendor_vq_stats)(struct vdpa_device *vdev, u16 idx,
+ struct sk_buff *msg,
+ struct netlink_ext_ack *extack);
struct vdpa_notification_area
(*get_vq_notification)(struct vdpa_device *vdev, u16 idx);
/* vq irq is not expected to be changed once DRIVER_OK is set */
diff --git a/include/uapi/linux/vdpa.h b/include/uapi/linux/vdpa.h
index 1061d8d2d09d..25c55cab3d7c 100644
--- a/include/uapi/linux/vdpa.h
+++ b/include/uapi/linux/vdpa.h
@@ -18,6 +18,7 @@ enum vdpa_command {
VDPA_CMD_DEV_DEL,
VDPA_CMD_DEV_GET, /* can dump */
VDPA_CMD_DEV_CONFIG_GET, /* can dump */
+ VDPA_CMD_DEV_VSTATS_GET,
};
enum vdpa_attr {
@@ -46,6 +47,11 @@ enum vdpa_attr {
VDPA_ATTR_DEV_NEGOTIATED_FEATURES, /* u64 */
VDPA_ATTR_DEV_MGMTDEV_MAX_VQS, /* u32 */
VDPA_ATTR_DEV_SUPPORTED_FEATURES, /* u64 */
+
+ VDPA_ATTR_DEV_QUEUE_INDEX, /* u32 */
+ VDPA_ATTR_DEV_VENDOR_ATTR_NAME, /* string */
+ VDPA_ATTR_DEV_VENDOR_ATTR_VALUE, /* u64 */
+
/* new attributes must be added above here */
VDPA_ATTR_MAX,
};
--
2.35.1
On 5/10/2022 4:27 AM, Eli Cohen wrote:
> Allows to read vendor statistics of a vdpa device. The specific
> statistics data are received from the upstream driver in the form of an
> (attribute name, attribute value) pairs.
>
> An example of statistics for mlx5_vdpa device are:
>
> received_desc - number of descriptors received by the virtqueue
> completed_desc - number of descriptors completed by the virtqueue
>
> A descriptor using indirect buffers is still counted as 1. In addition,
> N chained descriptors are counted correctly N times as one would expect.
>
> A new callback was added to vdpa_config_ops which provides the means for
> the vdpa driver to return statistics results.
>
> The interface allows for reading all the supported virtqueues, including
> the control virtqueue if it exists.
>
> Below are some examples taken from mlx5_vdpa which are introduced in the
> following patch:
>
> 1. Read statistics for the virtqueue at index 1
>
> $ vdpa dev vstats show vdpa-a qidx 1
> vdpa-a:
> queue_type tx queue_index 1 received_desc 3844836 completed_desc 3844836
>
> 2. Read statistics for the virtqueue at index 32
> $ vdpa dev vstats show vdpa-a qidx 32
> vdpa-a:
> queue_type control_vq queue_index 32 received_desc 62 completed_desc 62
>
> 3. Read statisitics for the virtqueue at index 0 with json output
> $ vdpa -j dev vstats show vdpa-a qidx 0
> {"vstats":{"vdpa-a":{
> "queue_type":"rx","queue_index":0,"name":"received_desc","value":417776,\
> "name":"completed_desc","value":417548}}}
>
> 4. Read statistics for the virtqueue at index 0 with preety json output
> $ vdpa -jp dev vstats show vdpa-a qidx 0
> {
> "vstats": {
> "vdpa-a": {
>
> "queue_type": "rx",
> "queue_index": 0,
> "name": "received_desc",
> "value": 417776,
> "name": "completed_desc",
> "value": 417548
The [IETF RFC 8259] "The JavaScript Object Notation (JSON) Data
Interchange Format" states:
... The names within an object SHOULD be unique.
:
:
:
An object whose names are all unique is interoperable in the sense
that all software implementations receiving that object will agree on
the name-value mappings. When the names within an object are not
unique, the behavior of software that receives such an object is
unpredictable. Many implementations report the last name/value pair
only. Other implementations report an error or fail to parse the
object, and some implementations report all of the name/value pairs,
including duplicates.
that said, duplicate keys in a JSON object are really not recommended,
which have the potential of breaking a lot of json parser tools. Are you
going to revise the example, and implement what Parav [1] or me [2]
suggested earlier? What worries me is that the new output would
ultimately affect the placement of attributes in the kernel side. And
looking at the userspace implementation, it doesn't seem like the paring
and the order of "name" and "value" keys are ensured: for e.g. name1
name2 value2, value1 value2 name2, name1 name2 value1 value2 should all
be invalid. A buggy vendor driver implementation which doesn't pay
attention to this could easily break script-able JSON output. IMHO it'd
be good to have clearly structured and self-describing attribute
placement in the kernel rather than have userspace validate all invalid
combinations.
[1]
https://lore.kernel.org/virtualization/PH0PR12MB5481F6D44451F01814470112DC089@PH0PR12MB5481.namprd12.prod.outlook.com/
[2]
https://lore.kernel.org/virtualization/[email protected]/
Thanks,
-Siwei
> }
> }
> }
>
> Signed-off-by: Eli Cohen <[email protected]>
> ---
> v4 -> v5:
> 1. Remove unused macro definition
> 2. Avoid policy on queue index. Do validty check in the implementation.
> 3. Restrict to VIRTIO_ID_NET devices only.
> 4. Improve netlink error messages.
>
>
> drivers/vdpa/vdpa.c | 140 ++++++++++++++++++++++++++++++++++++++
> include/linux/vdpa.h | 3 +
> include/uapi/linux/vdpa.h | 6 ++
> 3 files changed, 149 insertions(+)
>
> diff --git a/drivers/vdpa/vdpa.c b/drivers/vdpa/vdpa.c
> index fac89a0d8178..91f4c13c7c7c 100644
> --- a/drivers/vdpa/vdpa.c
> +++ b/drivers/vdpa/vdpa.c
> @@ -914,6 +914,86 @@ vdpa_dev_config_fill(struct vdpa_device *vdev, struct sk_buff *msg, u32 portid,
> return err;
> }
>
> +static int vdpa_fill_stats_rec(struct vdpa_device *vdev, struct sk_buff *msg,
> + struct genl_info *info, u32 index)
> +{
> + int err;
> +
> + err = vdev->config->get_vendor_vq_stats(vdev, index, msg, info->extack);
> + if (err)
> + return err;
> +
> + if (nla_put_u32(msg, VDPA_ATTR_DEV_QUEUE_INDEX, index))
> + return -EMSGSIZE;
> +
> + return 0;
> +}
> +
> +static int vendor_stats_fill(struct vdpa_device *vdev, struct sk_buff *msg,
> + struct genl_info *info, u32 index)
> +{
> + int err;
> +
> + if (!vdev->config->get_vendor_vq_stats)
> + return -EOPNOTSUPP;
> +
> + err = vdpa_fill_stats_rec(vdev, msg, info, index);
> + if (err)
> + return err;
> +
> + return 0;
> +}
> +
> +static int vdpa_dev_vendor_stats_fill(struct vdpa_device *vdev,
> + struct sk_buff *msg,
> + struct genl_info *info, u32 index)
> +{
> + u32 device_id;
> + void *hdr;
> + int err;
> + u32 portid = info->snd_portid;
> + u32 seq = info->snd_seq;
> + u32 flags = 0;
> +
> + hdr = genlmsg_put(msg, portid, seq, &vdpa_nl_family, flags,
> + VDPA_CMD_DEV_VSTATS_GET);
> + if (!hdr)
> + return -EMSGSIZE;
> +
> + if (nla_put_string(msg, VDPA_ATTR_DEV_NAME, dev_name(&vdev->dev))) {
> + err = -EMSGSIZE;
> + goto undo_msg;
> + }
> +
> + device_id = vdev->config->get_device_id(vdev);
> + if (nla_put_u32(msg, VDPA_ATTR_DEV_ID, device_id)) {
> + err = -EMSGSIZE;
> + goto undo_msg;
> + }
> +
> + switch (device_id) {
> + case VIRTIO_ID_NET:
> + if (index > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX) {
> + NL_SET_ERR_MSG_MOD(info->extack, "queue index excceeds max value");
> + err = -ERANGE;
> + break;
> + }
> +
> + err = vendor_stats_fill(vdev, msg, info, index);
> + break;
> + default:
> + err = -EOPNOTSUPP;
> + break;
> + }
> + genlmsg_end(msg, hdr);
> +
> + return err;
> +
> +undo_msg:
> + genlmsg_cancel(msg, hdr);
> + return err;
> +}
> +
> static int vdpa_nl_cmd_dev_config_get_doit(struct sk_buff *skb, struct genl_info *info)
> {
> struct vdpa_device *vdev;
> @@ -995,6 +1075,60 @@ vdpa_nl_cmd_dev_config_get_dumpit(struct sk_buff *msg, struct netlink_callback *
> return msg->len;
> }
>
> +static int vdpa_nl_cmd_dev_stats_get_doit(struct sk_buff *skb,
> + struct genl_info *info)
> +{
> + struct vdpa_device *vdev;
> + struct sk_buff *msg;
> + const char *devname;
> + struct device *dev;
> + u32 index;
> + int err;
> +
> + if (!info->attrs[VDPA_ATTR_DEV_NAME])
> + return -EINVAL;
> +
> + if (!info->attrs[VDPA_ATTR_DEV_QUEUE_INDEX])
> + return -EINVAL;
> +
> + devname = nla_data(info->attrs[VDPA_ATTR_DEV_NAME]);
> + msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
> + if (!msg)
> + return -ENOMEM;
> +
> + index = nla_get_u32(info->attrs[VDPA_ATTR_DEV_QUEUE_INDEX]);
> + mutex_lock(&vdpa_dev_mutex);
> + dev = bus_find_device(&vdpa_bus, NULL, devname, vdpa_name_match);
> + if (!dev) {
> + NL_SET_ERR_MSG_MOD(info->extack, "device not found");
> + err = -ENODEV;
> + goto dev_err;
> + }
> + vdev = container_of(dev, struct vdpa_device, dev);
> + if (!vdev->mdev) {
> + NL_SET_ERR_MSG_MOD(info->extack, "unmanaged vdpa device");
> + err = -EINVAL;
> + goto mdev_err;
> + }
> + err = vdpa_dev_vendor_stats_fill(vdev, msg, info, index);
> + if (err)
> + goto mdev_err;
> +
> + err = genlmsg_reply(msg, info);
> +
> + put_device(dev);
> + mutex_unlock(&vdpa_dev_mutex);
> +
> + return err;
> +
> +mdev_err:
> + put_device(dev);
> +dev_err:
> + nlmsg_free(msg);
> + mutex_unlock(&vdpa_dev_mutex);
> + return err;
> +}
> +
> static const struct nla_policy vdpa_nl_policy[VDPA_ATTR_MAX + 1] = {
> [VDPA_ATTR_MGMTDEV_BUS_NAME] = { .type = NLA_NUL_STRING },
> [VDPA_ATTR_MGMTDEV_DEV_NAME] = { .type = NLA_STRING },
> @@ -1035,6 +1169,12 @@ static const struct genl_ops vdpa_nl_ops[] = {
> .doit = vdpa_nl_cmd_dev_config_get_doit,
> .dumpit = vdpa_nl_cmd_dev_config_get_dumpit,
> },
> + {
> + .cmd = VDPA_CMD_DEV_VSTATS_GET,
> + .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
> + .doit = vdpa_nl_cmd_dev_stats_get_doit,
> + .flags = GENL_ADMIN_PERM,
> + },
> };
>
> static struct genl_family vdpa_nl_family __ro_after_init = {
> diff --git a/include/linux/vdpa.h b/include/linux/vdpa.h
> index 8943a209202e..2ae8443331e1 100644
> --- a/include/linux/vdpa.h
> +++ b/include/linux/vdpa.h
> @@ -276,6 +276,9 @@ struct vdpa_config_ops {
> const struct vdpa_vq_state *state);
> int (*get_vq_state)(struct vdpa_device *vdev, u16 idx,
> struct vdpa_vq_state *state);
> + int (*get_vendor_vq_stats)(struct vdpa_device *vdev, u16 idx,
> + struct sk_buff *msg,
> + struct netlink_ext_ack *extack);
> struct vdpa_notification_area
> (*get_vq_notification)(struct vdpa_device *vdev, u16 idx);
> /* vq irq is not expected to be changed once DRIVER_OK is set */
> diff --git a/include/uapi/linux/vdpa.h b/include/uapi/linux/vdpa.h
> index 1061d8d2d09d..25c55cab3d7c 100644
> --- a/include/uapi/linux/vdpa.h
> +++ b/include/uapi/linux/vdpa.h
> @@ -18,6 +18,7 @@ enum vdpa_command {
> VDPA_CMD_DEV_DEL,
> VDPA_CMD_DEV_GET, /* can dump */
> VDPA_CMD_DEV_CONFIG_GET, /* can dump */
> + VDPA_CMD_DEV_VSTATS_GET,
> };
>
> enum vdpa_attr {
> @@ -46,6 +47,11 @@ enum vdpa_attr {
> VDPA_ATTR_DEV_NEGOTIATED_FEATURES, /* u64 */
> VDPA_ATTR_DEV_MGMTDEV_MAX_VQS, /* u32 */
> VDPA_ATTR_DEV_SUPPORTED_FEATURES, /* u64 */
> +
> + VDPA_ATTR_DEV_QUEUE_INDEX, /* u32 */
> + VDPA_ATTR_DEV_VENDOR_ATTR_NAME, /* string */
> + VDPA_ATTR_DEV_VENDOR_ATTR_VALUE, /* u64 */
> +
> /* new attributes must be added above here */
> VDPA_ATTR_MAX,
> };