net/mlx5e: Don't create devices during unload flow
authorDmytro Linkin <dlinkin@nvidia.com>
Fri, 14 May 2021 08:14:19 +0000 (11:14 +0300)
committerSaeed Mahameed <saeedm@nvidia.com>
Wed, 16 Jun 2021 22:36:47 +0000 (15:36 -0700)
commita5ae8fc9058e37437c8c1f82b3d412b4abd1b9e6
treeec9fc0e79e881a6a268aff1094bd9feb68e5bd3e
parent65fb7d109abe3a1a9f1c2d3ba7e1249bc978d5f0
net/mlx5e: Don't create devices during unload flow

Running devlink reload command for port in switchdev mode cause
resources to corrupt: driver can't release allocated EQ and reclaim
memory pages, because "rdma" auxiliary device had add CQs which blocks
EQ from deletion.
Erroneous sequence happens during reload-down phase, and is following:

1. detach device - suspends auxiliary devices which support it, destroys
   others. During this step "eth-rep" and "rdma-rep" are destroyed,
   "eth" - suspended.
2. disable SRIOV - moves device to legacy mode; as part of disablement -
   rescans drivers. This step adds "rdma" auxiliary device.
3. destroy EQ table - <failure>.

Driver shouldn't create any device during unload flows. To handle that
implement MLX5_PRIV_FLAGS_DETACH flag, set it on device detach and unset
on device attach. If flag is set do no-op on drivers rescan.

Fixes: a925b5e309c9 ("net/mlx5: Register mlx5 devices to auxiliary virtual bus")
Signed-off-by: Dmytro Linkin <dlinkin@nvidia.com>
Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
Reviewed-by: Roi Dayan <roid@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
drivers/net/ethernet/mellanox/mlx5/core/dev.c
include/linux/mlx5/driver.h