All drivers that use queue management APIs already depend on the netdev
lock. Ultimately, we want to have most of the paths that work with
specific netdev to be rtnl_lock-free (ethtool mostly in particular).
Queue API currently has a much smaller API surface, so start with
rtnl_lock from it:

- add mutex to each dmabuf binding (to replace rtnl_lock)
- protect global net_devmem_dmabuf_bindings with a new (global) lock
- move netdev lock management to the callers of netdev_rx_queue_restart
  and drop rtnl_lock

Cc: Mina Almasry <almasrymina@google.com>

Stanislav Fomichev (4):
  net: create netdev_nl_sock to wrap bindings list
  net: protect net_devmem_dmabuf_bindings by new
    net_devmem_bindings_mutex
  net: add granular lock for the netdev netlink socket
  net: drop rtnl_lock for queue_mgmt operations

 Documentation/netlink/specs/netdev.yaml   |  4 +--
 drivers/net/ethernet/broadcom/bnxt/bnxt.c |  4 +--
 drivers/net/netdevsim/netdev.c            |  4 +--
 include/net/netdev_netlink.h              | 12 +++++++
 net/core/devmem.c                         | 32 ++++++++---------
 net/core/netdev-genl-gen.c                |  4 +--
 net/core/netdev-genl-gen.h                |  6 ++--
 net/core/netdev-genl.c                    | 42 ++++++++++-------------
 net/core/netdev_rx_queue.c                | 15 +++-----
 9 files changed, 63 insertions(+), 60 deletions(-)
 create mode 100644 include/net/netdev_netlink.h

-- 
2.48.1

On Fri,  7 Mar 2025 07:57:21 -0800 Stanislav Fomichev wrote:
> All drivers that use queue management APIs already depend on the netdev
> lock. Ultimately, we want to have most of the paths that work with
> specific netdev to be rtnl_lock-free (ethtool mostly in particular).
> Queue API currently has a much smaller API surface, so start with
> rtnl_lock from it:
> 
> - add mutex to each dmabuf binding (to replace rtnl_lock)
> - protect global net_devmem_dmabuf_bindings with a new (global) lock
> - move netdev lock management to the callers of netdev_rx_queue_restart
>   and drop rtnl_lock

One more note, looks like this silently conflicts with my:
https://lore.kernel.org/all/20250307183006.2312761-1-kuba@kernel.org/

You need to add:

#include <net/netdev_lock.h>

to net/core/netdev_rx_queue.c, otherwise the series together break 
the build.

On 03/07, Jakub Kicinski wrote:
> On Fri,  7 Mar 2025 07:57:21 -0800 Stanislav Fomichev wrote:
> > All drivers that use queue management APIs already depend on the netdev
> > lock. Ultimately, we want to have most of the paths that work with
> > specific netdev to be rtnl_lock-free (ethtool mostly in particular).
> > Queue API currently has a much smaller API surface, so start with
> > rtnl_lock from it:
> > 
> > - add mutex to each dmabuf binding (to replace rtnl_lock)
> > - protect global net_devmem_dmabuf_bindings with a new (global) lock
> > - move netdev lock management to the callers of netdev_rx_queue_restart
> >   and drop rtnl_lock
> 
> One more note, looks like this silently conflicts with my:
> https://lore.kernel.org/all/20250307183006.2312761-1-kuba@kernel.org/
> 
> You need to add:
> 
> #include <net/netdev_lock.h>
> 
> to net/core/netdev_rx_queue.c, otherwise the series together break 
> the build.

Noted, thanks!