Re: [PATCH net-next 1/2] devlink: Break parameter notification sequence to be before/after unload/load driver

[Jiri Pirko]

Thu, Jul 29, 2021 at 01:35:55PM CEST, leon@xxxxxxxxxx wrote:
>On Thu, Jul 29, 2021 at 01:22:58PM +0200, Jiri Pirko wrote:
>> Thu, Jul 29, 2021 at 10:15:25AM CEST, leon@xxxxxxxxxx wrote:
>> >From: Leon Romanovsky <leonro@xxxxxxxxxx>
>
><...>
>
>> >diff --git a/net/core/devlink.c b/net/core/devlink.c
>> >index b596a971b473..54e2a0375539 100644
>> >--- a/net/core/devlink.c
>> >+++ b/net/core/devlink.c
>> >@@ -3801,8 +3801,9 @@ static void devlink_param_notify(struct devlink *devlink,
>> > 				 struct devlink_param_item *param_item,
>> > 				 enum devlink_command cmd);
>> > 
>> >-static void devlink_reload_netns_change(struct devlink *devlink,
>> >-					struct net *dest_net)
>> >+static void devlink_params_notify(struct devlink *devlink, struct net *dest_net,
>> 
>> Please name it differently. This function notifies not only the params,
>> but the devlink instance itself as well.
>
>I'm open for suggestion. What did you have in mind?

devlink_ns_change_notify?

>
>> 
>> 
>> >+				  struct net *curr_net,
>> >+				  enum devlink_command cmd)
>> > {
>> > 	struct devlink_param_item *param_item;
>> > 
>> >@@ -3812,17 +3813,17 @@ static void devlink_reload_netns_change(struct devlink *devlink,
>> > 	 * reload process so the notifications are generated separatelly.
>> > 	 */
>> > 
>> >-	list_for_each_entry(param_item, &devlink->param_list, list)
>> >-		devlink_param_notify(devlink, 0, param_item,
>> >-				     DEVLINK_CMD_PARAM_DEL);
>> >-	devlink_notify(devlink, DEVLINK_CMD_DEL);
>> >+	if (!dest_net || net_eq(dest_net, curr_net))
>> >+		return;
>> > 
>> >-	__devlink_net_set(devlink, dest_net);
>> >+	if (cmd == DEVLINK_CMD_PARAM_NEW)
>> >+		devlink_notify(devlink, DEVLINK_CMD_NEW);
>> 
>> This is quite odd. According to PARAMS cmd you decife devlink CMD.
>> 
>> Just have bool arg which would help you select both
>> DEVLINK_CMD_PARAM_NEW/DEL and DEVLINK_CMD_NEW/DEL
>
>The patch is quite misleading, but the final result looks neat:
>
>   3847 static void devlink_params_notify(struct devlink *devlink, struct net *dest_net,
>   3848                                   struct net *curr_net,
>   3849                                   enum devlink_command cmd)
>   3850 {
>   3851         struct devlink_param_item *param_item;
>   3852 
>   3853         /* Userspace needs to be notified about devlink objects
>   3854          * removed from original and entering new network namespace.
>   3855          * The rest of the devlink objects are re-created during
>   3856          * reload process so the notifications are generated separatelly.
>   3857          */
>   3858 
>   3859         if (!dest_net || net_eq(dest_net, curr_net))
>   3860                 return;
>   3861 
>   3862         if (cmd == DEVLINK_CMD_PARAM_NEW)
>   3863                 devlink_notify(devlink, DEVLINK_CMD_NEW);

Nothing misleading here. This is exactly what I'm pointing out...



>   3864 
>   3865         list_for_each_entry(param_item, &devlink->param_list, list)
>   3866                 devlink_param_notify(devlink, 0, param_item, cmd);
>   3867 
>   3868         if (cmd == DEVLINK_CMD_PARAM_DEL)
>   3869                 devlink_notify(devlink, DEVLINK_CMD_DEL);
>   3870 }
>
>
>So as you can see in line 3866, we anyway will need to provide "cmd", so
>do you suggest to add extra two bool variables to the function signature
>to avoid "cmd == DEVLINK_CMD_PARAM_NEW" and "cmd == DEVLINK_CMD_PARAM_DEL" ifs?
>
>Thanks