Skip to contentSkip to conversation

net: make memory provider install / close paths more common

PATCH Applied Jakub Kicinski 3 participants 5 messages Apr 3, 2025 – Apr 4, 2025

Subsystems (3)
lore
Share & tools
lore

Jakub Kicinski

author maintainer
#0 lore raw

v2 cover net: make memory provider install / close paths more common

We seem to be fixing bugs in config path for devmem which also exist in the io_uring ZC path. Let's try to make the two paths more common, otherwise this is bound to keep happening.

Found by code inspection and compile tested only.

v2:
- [patch 1] add to commit msg
- [patch 1] fix arg naming in the header
- [patch 2] don't split the registration check, it may cause a race if we just bail on the registration state and not on the MP being present, as we drop and re-take the instance lock after setting reg_state
v1: https://lore.kernel.org/[email protected]

Jakub Kicinski (2):
net: move mp dev config validation to __net_mp_open_rxq()
net: avoid false positive warnings in __net_mp_close_rxq()

 include/net/page_pool/memory_provider.h |  6 +++
 net/core/devmem.c                       | 64 +++++++------------------
 net/core/netdev-genl.c                  |  6 ---
 net/core/netdev_rx_queue.c              | 53 +++++++++++++++-----
 4 files changed, 63 insertions(+), 66 deletions(-)
Signature · 1 line
2.49.0
author maintainer
#1 lore raw

PATCH v2 1/2 net: move mp dev config validation to __net_mp_open_rxq()

devmem code performs a number of safety checks to avoid having to reimplement all of them in the drivers. Move those to __net_mp_open_rxq() and reuse that function for binding to make sure that io_uring ZC also benefits from them.

While at it rename the queue ID variable to rxq_idx in __net_mp_open_rxq(), we touch most of the relevant lines.

The XArray insertion is reordered after the netdev_rx_queue_restart() call, otherwise we'd need to duplicate the queue index check or risk inserting an invalid pointer. The XArray allocation failures should be extremely rare.

Fixes 6e18ed929d3b ("net: add helpers for setting a memory provider on an rx queue")

---
v2:
- add to commit msg
- fix arg naming in the header (Stan)
v1: https://lore.kernel.org/[email protected]

---
 include/net/page_pool/memory_provider.h |  6 +++
 net/core/devmem.c                       | 52 ++++++-------------------
 net/core/netdev-genl.c                  |  6 ---
 net/core/netdev_rx_queue.c              | 49 +++++++++++++++++------
 4 files changed, 55 insertions(+), 58 deletions(-)
include/net/page_pool/memory_provider.h +6 -0
index b3e665897767..ada4f968960a 100644
@@ -6,6 +6,7 @@
 #include <net/page_pool/types.h>
 
 struct netdev_rx_queue;
+struct netlink_ext_ack;
 struct sk_buff;
 
 struct memory_provider_ops {
@@ -24,8 +25,13 @@ void net_mp_niov_clear_page_pool(struct net_iov *niov);
 
 int net_mp_open_rxq(struct net_device *dev, unsigned ifq_idx,
 		    struct pp_memory_provider_params *p);
+int __net_mp_open_rxq(struct net_device *dev, unsigned int rxq_idx,
+		      const struct pp_memory_provider_params *p,
+		      struct netlink_ext_ack *extack);
 void net_mp_close_rxq(struct net_device *dev, unsigned ifq_idx,
 		      struct pp_memory_provider_params *old_p);
+void __net_mp_close_rxq(struct net_device *dev, unsigned int rxq_idx,
+			const struct pp_memory_provider_params *old_p);
 
 /**
   * net_mp_netmem_place_in_cache() - give a netmem to a page pool
net/core/devmem.c +11 -41

75 lines not on this page · raw message

net/core/netdev-genl.c +0 -6
index 3afeaa8c5dc5..5d7af50fe702 100644
@@ -874,12 +874,6 @@ int netdev_nl_bind_rx_doit(struct sk_buff *skb, struct genl_info *info)
 		goto err_unlock;
 	}
 
-	if (dev_xdp_prog_count(netdev)) {
-		NL_SET_ERR_MSG(info->extack, "unable to bind dmabuf to device with XDP program attached");
-		err = -EEXIST;
-		goto err_unlock;
-	}
-
 	binding = net_devmem_bind_dmabuf(netdev, dmabuf_fd, info->extack);
 	if (IS_ERR(binding)) {
 		err = PTR_ERR(binding);
net/core/netdev_rx_queue.c +38 -11

86 lines not on this page · raw message

Signature · 1 line
2.49.0
author maintainer
#2 lore raw

PATCH v2 2/2 net: avoid false positive warnings in __net_mp_close_rxq()

In reply to Jakub Kicinski

Commit under Fixes solved the problem of spurious warnings when we uninstall an MP from a device while its down. The __net_mp_close_rxq() which is used by io_uring was not fixed. Move the fix over and reuse __net_mp_close_rxq() in the devmem path.

Fixes a70f891e0fa0 ("net: devmem: do not WARN conditionally after netdev_rx_queue_restart()")

---
v2:
- don't split the registration check, it may cause a race if we just bail on the registration state and not on the MP being present, as we drop and re-take the instance lock after setting reg_state
v1: https://lore.kernel.org/[email protected]

---
 net/core/devmem.c          | 12 +++++-------
 net/core/netdev_rx_queue.c |  4 +++-
 2 files changed, 8 insertions(+), 8 deletions(-)
net/core/devmem.c +5 -7
index f2ce3c2ebc97..6e27a47d0493 100644
@@ -116,21 +116,19 @@ void net_devmem_unbind_dmabuf(struct net_devmem_dmabuf_binding *binding)
 	struct netdev_rx_queue *rxq;
 	unsigned long xa_idx;
 	unsigned int rxq_idx;
-	int err;
 
 	if (binding->list.next)
 		list_del(&binding->list);
 
 	xa_for_each(&binding->bound_rxqs, xa_idx, rxq) {
-		WARN_ON(rxq->mp_params.mp_priv != binding);
-
-		rxq->mp_params.mp_priv = NULL;
-		rxq->mp_params.mp_ops = NULL;
+		const struct pp_memory_provider_params mp_params = {
+			.mp_priv	= binding,
+			.mp_ops		= &dmabuf_devmem_ops,
+		};
 
 		rxq_idx = get_netdev_rx_queue_index(rxq);
 
-		err = netdev_rx_queue_restart(binding->dev, rxq_idx);
-		WARN_ON(err && err != -ENETDOWN);
+		__net_mp_close_rxq(binding->dev, rxq_idx, &mp_params);
 	}
 
 	xa_erase(&net_devmem_dmabuf_bindings, binding->id);
net/core/netdev_rx_queue.c +3 -1
index 556b5393ec9f..d126f10197bf 100644
@@ -154,6 +154,7 @@ void __net_mp_close_rxq(struct net_device *dev, unsigned int ifq_idx,
 			const struct pp_memory_provider_params *old_p)
 {
 	struct netdev_rx_queue *rxq;
+	int err;
 
 	if (WARN_ON_ONCE(ifq_idx >= dev->real_num_rx_queues))
 		return;
@@ -173,7 +174,8 @@ void __net_mp_close_rxq(struct net_device *dev, unsigned int ifq_idx,
 
 	rxq->mp_params.mp_ops = NULL;
 	rxq->mp_params.mp_priv = NULL;
-	WARN_ON(netdev_rx_queue_restart(dev, ifq_idx));
+	err = netdev_rx_queue_restart(dev, ifq_idx);
+	WARN_ON(err && err != -ENETDOWN);
 }
 
 void net_mp_close_rxq(struct net_device *dev, unsigned ifq_idx,
Signature · 1 line
2.49.0
Reviewed-by on v2 2/2
#3 lore raw

Re: [PATCH net v2 2/2] net: avoid false positive warnings in __net_mp_close_rxq()

Quoted 9 lines · Jakub Kicinski

Commit under Fixes solved the problem of spurious warnings when we uninstall an MP from a device while its down. The __net_mp_close_rxq() which is used by io_uring was not fixed. Move the fix over and reuse __net_mp_close_rxq() in the devmem path.

Fixes a70f891e0fa0 ("net: devmem: do not WARN conditionally after netdev_rx_queue_restart()")
#4 lore raw
In reply to Jakub Kicinski
This series was applied to netdev/net.git (main) by Jakub Kicinski <[email protected]>:

Hello:

This series was applied to netdev/net.git (main)
by Jakub Kicinski <[email protected]>:

Quoted 15 lines · On Wed, 2 Apr 2025 18:34:03 -0700 you wrote:

We seem to be fixing bugs in config path for devmem which also exist in the io_uring ZC path. Let's try to make the two paths more common, otherwise this is bound to keep happening.

Found by code inspection and compile tested only.

v2:
- [patch 1] add to commit msg
- [patch 1] fix arg naming in the header
- [patch 2] don't split the registration check, it may cause a race if we just bail on the registration state and not on the MP being present, as we drop and re-take the instance lock after setting reg_state
v1: https://lore.kernel.org/[email protected]

Quoted 1 line

[...]

Here is the summary with links:
- [net,v2,1/2] net: move mp dev config validation to __net_mp_open_rxq()

    https://git.kernel.org/netdev/net/c/ec304b70d46b
  - [net,v2,2/2] net: avoid false positive warnings in __net_mp_close_rxq()
    https://git.kernel.org/netdev/net/c/34f71de3f548

You are awesome, thank you!

Signature · 2 lines
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html