On Tue, Aug 24, 2021 at 10:13:34AM +0800, Mark Zhang wrote:

On 8/24/2021 3:44 AM, Jason Gunthorpe wrote:

quoted

On Wed, Aug 18, 2021 at 02:24:28PM +0300, Mark Zhang wrote:

quoted

diff --git a/include/uapi/rdma/rdma_netlink.h b/include/uapi/rdma/rdma_netlink.h
index 79e6ca87d2e0..57f39d8fe434 100644
+++ b/include/uapi/rdma/rdma_netlink.h

@@ -557,6 +557,8 @@ enum rdma_nldev_attr {
  	RDMA_NLDEV_ATTR_STAT_OPCOUNTER_ENTRY,	/* nested table */
  	RDMA_NLDEV_ATTR_STAT_OPCOUNTER_ENTRY_NAME,	/* string */
  	RDMA_NLDEV_ATTR_STAT_OPCOUNTER_ENTRY_VALUE,	/* u64 */
+	RDMA_NLDEV_ATTR_STAT_OP_MODE_LIST,		/* u8 */
+	RDMA_NLDEV_ATTR_STAT_OP_MODE_LIST_SUPPORTED,	/* u8 */

See, here - shouldn't manipulation of MODE_LIST be done by a normal
RDMA_NLDEV_CMD_STAT_SET with the new MODE_LIST array? This doesn't seem
netlinky at all..

Both of them are flags and this is a "get" operation; "MODE_LIST" asks
kernel to return currently enabled op-counters, "MODE_LIST_SUPPORTED" asks
kernel to return supported op-counters. Maybe the macro name are not good?

The marcors are fine, the protocol is just a bit wonky. The ADD/REMOVE
idea should only be used on top level objects, but this is a nested
sub so you should be using SET to manipulate it and it should provide
the entire current list, not a add/remove type operation.

Jason