RE: [RFC v5 net-next 4/5] rtnetlink: Add support for SyncE recovered clock configuration

-----Original Message-----
From: Jakub Kicinski <kuba@kernel.org>
Sent: Tuesday, October 26, 2021 11:33 PM
To: Machnikowski, Maciej <redacted>
Subject: Re: [RFC v5 net-next 4/5] rtnetlink: Add support for SyncE recovered
clock configuration

Please add a write up of how things fit together in Documentation/.
I'm sure reviewers and future users will appreciate that.

 
Sure! Documentation/networking/synce.rst would be the right place to add it?
Or is there any better place?

Some nits below.

On Tue, 26 Oct 2021 19:31:45 +0200 Maciej Machnikowski wrote:

quoted

Add support for RTNL messages for reading/configuring SyncE recovered
clocks.
The messages are:
RTM_GETRCLKRANGE: Reads the allowed pin index range for the

recovered

quoted

		  clock outputs. This can be aligned to PHY outputs or
		  to EEC inputs, whichever is better for a given
		  application

RTM_GETRCLKSTATE: Read the state of recovered pins that output

recovered

quoted

		  clock from a given port. The message will contain the
		  number of assigned clocks (IFLA_RCLK_STATE_COUNT) and
		  a N pin inexes in IFLA_RCLK_STATE_OUT_IDX

Do we need two separate calls for the gets?

I feel it's better to separate the "control plane" from the "information plane",
but if having less is better we can merge those 2. Then the GETRCLKSTATE would
return:
Number of active outputs
Output indexes
Max allowed output range
Min allowed output range

Since Min/Max are usually only needed once (and may require some FW
Interaction) I'd rather keep them separate.
 

quoted

RTM_SETRCLKSTATE: Sets the redirection of the recovered clock for
		  a given pin

quoted

+struct if_set_rclk_msg {
+	__u32 ifindex;
+	__u32 out_idx;
+	__u32 flags;

Why not break this out into separate attrs?

I think it would break the functionality - we need both the index and the
action (ena/dis in the flags) to know what to do.
 

quoted

+++ b/net/core/rtnetlink.c

@@ -5524,8 +5524,10 @@ static int rtnl_eec_state_get(struct sk_buff

*skb, struct nlmsghdr *nlh,

quoted

 	state = nlmsg_data(nlh);
 	dev = __dev_get_by_index(net, state->ifindex);
-	if (!dev)
+	if (!dev) {
+		NL_SET_ERR_MSG(extack, "unknown ifindex");
 		return -ENODEV;
+	}

 	nskb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
 	if (!nskb)

Belongs in previous patch?

True - will fix in next revision :)

Regards
Maciek