On 09/05/2024 16:36, Dan Carpenter wrote:
Hello Yishai Hadas,
Commit 6de042240b0f ("vfio/mlx5: Let firmware knows upon leaving
PRE_COPY back to RUNNING") from Feb 5, 2024 (linux-next), leads to
the following Smatch static checker warning:
drivers/vfio/pci/mlx5/main.c:1164 mlx5vf_pci_step_device_state_locked()
error: uninitialized symbol 'state'.
drivers/vfio/pci/mlx5/main.c
1142 if ((cur == VFIO_DEVICE_STATE_PRE_COPY && new == VFIO_DEVICE_STATE_RUNNING) ||
1143 (cur == VFIO_DEVICE_STATE_PRE_COPY_P2P &&
1144 new == VFIO_DEVICE_STATE_RUNNING_P2P)) {
1145 struct mlx5_vf_migration_file *migf = mvdev->saving_migf;
1146 struct mlx5_vhca_data_buffer *buf;
1147 enum mlx5_vf_migf_state state;
^^^^^
1148 size_t size;
1149
1150 ret = mlx5vf_cmd_query_vhca_migration_state(mvdev, &size, NULL,
1151 MLX5VF_QUERY_INC | MLX5VF_QUERY_CLEANUP);
1152 if (ret)
1153 return ERR_PTR(ret);
1154 buf = mlx5vf_get_data_buffer(migf, size, DMA_FROM_DEVICE);
1155 if (IS_ERR(buf))
1156 return ERR_CAST(buf);
1157 /* pre_copy cleanup */
1158 ret = mlx5vf_cmd_save_vhca_state(mvdev, migf, buf, false, false);
1159 if (ret) {
1160 mlx5vf_put_data_buffer(buf);
1161 return ERR_PTR(ret);
1162 }
1163 mlx5vf_disable_fds(mvdev, &state);
^^^^^^
state is only set some of the time.
The 'state' will *always* be set in the above flow.
As we are in the source side of the migration we have a valid
saving_migf (see line 1145 above), as we pass in a non NULL pointer for
the state, it will be always filled inside.
We not just make mlx5vf_disable_fds()
return an error code?
mlx5vf_disable_fd() is a cleanup function that can't fail.
It just holds/sets the state of the migf following the completion of the
asynchronous SAVE command that was issued in line 1158.
So, it's a false alarm.
Thanks,
Yishai
--> 1164 return (state != MLX5_MIGF_STATE_ERROR) ? NULL : ERR_PTR(-EIO);
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Uninitialized.
1165 }
regards,
dan carpenter
