rpmsg_virtio: rpmsg_deinit_vdev: Add check for empty endpoint list

[bentheredonethat]

There can exist scenario where the endpoint list in a rpmsg vdev can be empty of endpoint-created nodes but the original remains, check if that is the case to ensure the while loop can properly terminate.

[bentheredonethat]

CC @tnmysh

[wmamills]

@bentheredonethat can you describe the scenario that where this arises?

[bentheredonethat]

I see this as bug in rpmsg_deinit_vdev because the initial list head is created before rpsmg-create endpoint and this adds check for that initial list head node.

[bentheredonethat]

I took a further look. Because there can exist independtly the RPMsg vdev's ns_ept I believe that the alternative solution of moving the RPMsg vdev's list of endpoints to create endpoint is not valid.

Given that, I think keeping the logic as is (the existing patch) is valid.

CC @wmamills

[tnmysh]

LGTM.

[arnopo]

The point is not clear to me I cannot figure out how we can have "node->next == NULL" regarding metal_list_XXX macro
after metal_list_init(&rdev->endpoints)

    rdev->endpoints.prev = rdev->endpoints;
    rdev->endpoints.next = rdev->endpoints;

After rpmsg_register_endpoint(rdev, &rdev->ns_ept,, ...)

    rdev->endpoints.prev =  rdev->ns_ept->node;
    rdev->endpoints.next =  rdev->ns_ept->node;

    rdev->ns_ept->node.prev =  rdev->endpoints;
    rdev->ns_ept->node.next =  rdev->endpoints;

After metal_list_del(&ept->node)

    rdev->endpoints.prev =   rdev->endpoints;
    rdev->endpoints.next =   rdev->endpoints;

   rdev->ns_ept->node =  rdev->ns_ept->node;
   rdev->ns_ept->node =  rdev->ns_ept->node;

Please could you detail the sequence?

[bentheredonethat]

we can skip this after further inspection. will close