On Wed, Nov 4, 2020 at 11:32 PM gregkh gregkh@linuxfoundation.org wrote:
On Wed, Nov 04, 2020 at 03:21:23PM -0800, Dan Williams wrote:
On Tue, Nov 3, 2020 at 7:45 AM Jason Gunthorpe jgg@ziepe.ca wrote: [..]
+MODULE_DEVICE_TABLE(auxiliary, mlx5v_id_table);
+static struct auxiliary_driver mlx5v_driver = {
.name = "vnet",.probe = mlx5v_probe,.remove = mlx5v_remove,.id_table = mlx5v_id_table,+};
It is hard to see from the diff, but when this patch is applied the vdpa module looks like I imagined things would look with the auxiliary bus. It is very similar in structure to a PCI driver with the probe() function cleanly registering with its subsystem. This is what I'd like to see from the new Intel RDMA driver.
Greg, I think this patch is the best clean usage example.
I've looked over this series and it has the right idea and parts. There is definitely more that can be done to improve mlx5 in this area, but this series is well scoped and cleans a good part of it.
Greg?
I know you alluded to going your own way if the auxiliary bus patches did not shape up soon, but it seems they have and the stakeholders have reached this consensus point.
Were there any additional changes you wanted to see happen? I'll go give the final set another once over, but David has been diligently fixing up all the declared major issues so I expect to find at most minor incremental fixups.
This is in my to-review pile, along with a load of other stuff at the moment: $ ~/bin/mdfrm -c ~/mail/todo/ 1709 messages in /home/gregkh/mail/todo/
So give me a chance. There is no rush on my side for this given the huge delays that have happened here on the authorship side many times in the past :)
Sure, I was more looking to confirm that it's worth continuing to polish this set given your mention of possibly going a different direction.
If you can review it, or anyone else, that is always most appreciated.
Thanks, will do.