Expose a new per-queue attribute, xsk, which indicates that a queue is
being used for AF_XDP. Update the documentation to more explicitly state
which queue types are linked.
Signed-off-by: Joe Damato <jdamato@fastly.com>
---
Documentation/netlink/specs/netdev.yaml | 10 +++++++++-
include/uapi/linux/netdev.h | 1 +
net/core/netdev-genl.c | 6 ++++++
tools/include/uapi/linux/netdev.h | 1 +
4 files changed, 17 insertions(+), 1 deletion(-)
diff --git a/Documentation/netlink/specs/netdev.yaml b/Documentation/netlink/specs/netdev.yaml
index cbb544bd6c84..7a72788cce03 100644
--- a/Documentation/netlink/specs/netdev.yaml
+++ b/Documentation/netlink/specs/netdev.yaml
@@ -286,6 +286,8 @@ attribute-sets:
-
name: type
doc: Queue type as rx, tx. Each queue type defines a separate ID space.
+ XDP TX queues allocated in the kernel are not linked to NAPIs and
+ thus not listed. AF_XDP queues have the xsk attribute set.
type: u32
enum: queue-type
-
@@ -296,7 +298,12 @@ attribute-sets:
name: dmabuf
doc: ID of the dmabuf attached to this queue, if any.
type: u32
-
+ -
+ name: xsk
+ doc: Non-zero for queues which are used for XSK (AF_XDP), 0 otherwise.
+ type: u32
+ checks:
+ max: 1
-
name: qstats
doc: |
@@ -637,6 +644,7 @@ operations:
- napi-id
- ifindex
- dmabuf
+ - xsk
dump:
request:
attributes:
diff --git a/include/uapi/linux/netdev.h b/include/uapi/linux/netdev.h
index e4be227d3ad6..4d2dcf4960ec 100644
--- a/include/uapi/linux/netdev.h
+++ b/include/uapi/linux/netdev.h
@@ -136,6 +136,7 @@ enum {
NETDEV_A_QUEUE_TYPE,
NETDEV_A_QUEUE_NAPI_ID,
NETDEV_A_QUEUE_DMABUF,
+ NETDEV_A_QUEUE_XSK,
__NETDEV_A_QUEUE_MAX,
NETDEV_A_QUEUE_MAX = (__NETDEV_A_QUEUE_MAX - 1)
diff --git a/net/core/netdev-genl.c b/net/core/netdev-genl.c
index 715f85c6b62e..964aebfcb079 100644
--- a/net/core/netdev-genl.c
+++ b/net/core/netdev-genl.c
@@ -394,12 +394,18 @@ netdev_nl_queue_fill_one(struct sk_buff *rsp, struct net_device *netdev,
nla_put_u32(rsp, NETDEV_A_QUEUE_DMABUF, binding->id))
goto nla_put_failure;
+ if (nla_put_u32(rsp, NETDEV_A_QUEUE_XSK, !!rxq->pool))
+ goto nla_put_failure;
+
break;
case NETDEV_QUEUE_TYPE_TX:
txq = netdev_get_tx_queue(netdev, q_idx);
if (txq->napi && nla_put_u32(rsp, NETDEV_A_QUEUE_NAPI_ID,
txq->napi->napi_id))
goto nla_put_failure;
+
+ if (nla_put_u32(rsp, NETDEV_A_QUEUE_XSK, !!txq->pool))
+ goto nla_put_failure;
}
genlmsg_end(rsp, hdr);
diff --git a/tools/include/uapi/linux/netdev.h b/tools/include/uapi/linux/netdev.h
index e4be227d3ad6..4d2dcf4960ec 100644
--- a/tools/include/uapi/linux/netdev.h
+++ b/tools/include/uapi/linux/netdev.h
@@ -136,6 +136,7 @@ enum {
NETDEV_A_QUEUE_TYPE,
NETDEV_A_QUEUE_NAPI_ID,
NETDEV_A_QUEUE_DMABUF,
+ NETDEV_A_QUEUE_XSK,
__NETDEV_A_QUEUE_MAX,
NETDEV_A_QUEUE_MAX = (__NETDEV_A_QUEUE_MAX - 1)
--
2.25.1On Wed, 29 Jan 2025 17:24:24 +0000 Joe Damato wrote: > Expose a new per-queue attribute, xsk, which indicates that a queue is > being used for AF_XDP. Update the documentation to more explicitly state > which queue types are linked. Let's do the same thing we did for io_uring queues? An empty nest: https://lore.kernel.org/all/20250116231704.2402455-6-dw@davidwei.uk/ At the protocol level nest is both smaller and more flexible. It's just 4B with zero length and a "this is a nest" flag. We can add attributes to it as we think of things to express.
On Wed, Jan 29, 2025 at 05:52:24PM -0800, Jakub Kicinski wrote:
> On Wed, 29 Jan 2025 17:24:24 +0000 Joe Damato wrote:
> > Expose a new per-queue attribute, xsk, which indicates that a queue is
> > being used for AF_XDP. Update the documentation to more explicitly state
> > which queue types are linked.
>
> Let's do the same thing we did for io_uring queues? An empty nest:
> https://lore.kernel.org/all/20250116231704.2402455-6-dw@davidwei.uk/
>
> At the protocol level nest is both smaller and more flexible.
> It's just 4B with zero length and a "this is a nest" flag.
> We can add attributes to it as we think of things to express.
I got a thing working locally, but just to make sure I'm
following... you are saying that the attribute will exist (but have
nothing in it) when the queue has a pool, and when !q->pool the
attribute will not exist?
For example:
[{'id': 0, 'ifindex': 5, 'napi-id': 8266, 'type': 'rx', 'xsk': {}},
{'id': 1, 'ifindex': 5, 'napi-id': 8267, 'type': 'rx'},
...
Is that what you are thinking?
Completely fine with me as I haven't read enough of the xsk code to
really have a good sense of what attributes might be helpful to
expose at this point.
On Thu, 30 Jan 2025 13:06:47 -0500 Joe Damato wrote:
> On Wed, Jan 29, 2025 at 05:52:24PM -0800, Jakub Kicinski wrote:
> > On Wed, 29 Jan 2025 17:24:24 +0000 Joe Damato wrote:
> > > Expose a new per-queue attribute, xsk, which indicates that a queue is
> > > being used for AF_XDP. Update the documentation to more explicitly state
> > > which queue types are linked.
> >
> > Let's do the same thing we did for io_uring queues? An empty nest:
> > https://lore.kernel.org/all/20250116231704.2402455-6-dw@davidwei.uk/
> >
> > At the protocol level nest is both smaller and more flexible.
> > It's just 4B with zero length and a "this is a nest" flag.
> > We can add attributes to it as we think of things to express.
>
> I got a thing working locally, but just to make sure I'm
> following... you are saying that the attribute will exist (but have
> nothing in it) when the queue has a pool, and when !q->pool the
> attribute will not exist?
>
> For example:
>
> [{'id': 0, 'ifindex': 5, 'napi-id': 8266, 'type': 'rx', 'xsk': {}},
> {'id': 1, 'ifindex': 5, 'napi-id': 8267, 'type': 'rx'},
> ...
>
> Is that what you are thinking?
Yup! That's it.
> Completely fine with me as I haven't read enough of the xsk code to
> really have a good sense of what attributes might be helpful to
> expose at this point.
On Wed, Jan 29, 2025 at 05:52:24PM -0800, Jakub Kicinski wrote: > On Wed, 29 Jan 2025 17:24:24 +0000 Joe Damato wrote: > > Expose a new per-queue attribute, xsk, which indicates that a queue is > > being used for AF_XDP. Update the documentation to more explicitly state > > which queue types are linked. > > Let's do the same thing we did for io_uring queues? An empty nest: > https://lore.kernel.org/all/20250116231704.2402455-6-dw@davidwei.uk/ > > At the protocol level nest is both smaller and more flexible. > It's just 4B with zero length and a "this is a nest" flag. > We can add attributes to it as we think of things to express. Thanks for the pointer; will do.
© 2016 - 2026 Red Hat, Inc.