The commit 565b4824 ("devlink: change port event netdev notifier
from per-net to global") changed original per-net notifier to be
per-devlink instance. That fixed the issue of non-receiving events
of netdev uninit if that moved to a different namespace.
That worked fine in -net tree.

However, later on when commit ee75f1fc ("net/mlx5e: Create
separate devlink instance for ethernet auxiliary device") and
commit 72ed5d56 ("net/mlx5: Suspend auxiliary devices only in
case of PCI device suspend") were merged, a deadlock was introduced
when removing a namespace with devlink instance with another nested
instance.

Here there is the bad flow example resulting in deadlock with mlx5:
net_cleanup_work -> cleanup_net (takes down_read(&pernet_ops_rwsem) ->
devlink_pernet_pre_exit() -> devlink_reload() ->
mlx5_devlink_reload_down() -> mlx5_unload_one_devl_locked() ->
mlx5_detach_device() -> del_adev() -> mlx5e_remove() ->
mlx5e_destroy_devlink() -> devlink_free() ->
unregister_netdevice_notifier() (takes down_write(&pernet_ops_rwsem)

Steps to reproduce:
$ modprobe mlx5_core
$ ip netns add ns1
$ devlink dev reload pci/0000:08:00.0 netns ns1
$ ip netns del ns1

Resolve this by converting the notifier from per-devlink instance to
a static one registered during init phase and leaving it registered
forever. Use this notifier for all devlink port instances created
later on.

Note what a tree needs this fix only in case all of the cited fixes
commits are present.
Reported-by: Moshe Shemesh <moshe@nvidia.com>
Fixes: 565b4824 ("devlink: change port event netdev notifier from per-net to global")
Fixes: ee75f1fc ("net/mlx5e: Create separate devlink instance for ethernet auxiliary device")
Fixes: 72ed5d56 ("net/mlx5: Suspend auxiliary devices only in case of PCI device suspend")
Signed-off-by: Jiri Pirko <jiri@nvidia.com>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Link: https://lore.kernel.org/r/20230510144621.932017-1-jiri@resnulli.usSigned-off-by: Jakub Kicinski <kuba@kernel.org>