On Thu, Jan 21, 2021 at 12:59:55PM -0800, Samudrala, Sridhar wrote:

quoted

+                 mlx5_core.sf.4
+          (subfunction auxiliary device)
+                       /\
+                      /  \
+                     /    \
+                    /      \
+                   /        \
+      mlx5_core.eth.4     mlx5_core.rdma.4
+     (sf eth aux dev)     (sf rdma aux dev)
+         |                      |
+         |                      |
+      p0sf88                  mlx5_0
+     (sf netdev)          (sf rdma device)

This picture seems to indicate that when SF is activated, a sub
function auxiliary device is created 

Yes

and when a driver is bound to that sub function aux device and
probed, 2 additional auxiliary devices are created.  

More than two, but yes

Is this correct? Are all these auxiliary devices seen on the same
aux bus?  

Yes

Why do we need another sf eth aux device?

The first aux device represents the physical HW and mlx5_core binds to it,
the analog is like a pci_device.

The other aux devices represent the subsystem split of the mlx5 driver
- mlx5_core creates them and each subsystem in turn binds to the
mlx5_core driver. This already exists, and Intel will be doing this as
well whenever the RDMA driver is posted again..

Jason