Re: [PATCH rdma-next v1 06/11] RDMA/nldev: Add support to get status of all counters

On Wed, Sep 29, 2021 at 03:26:25PM +0300, Leon Romanovsky wrote:

On Tue, Sep 28, 2021 at 08:52:17AM -0300, Jason Gunthorpe wrote:

quoted

On Tue, Sep 28, 2021 at 05:12:39PM +0800, Mark Zhang wrote:

quoted

On 9/28/2021 1:30 AM, Jason Gunthorpe wrote:

quoted

On Wed, Sep 15, 2021 at 02:07:25AM +0300, Leon Romanovsky wrote:

quoted

+static int stat_get_doit_default_counter(struct sk_buff *skb,
+					 struct nlmsghdr *nlh,
+					 struct netlink_ext_ack *extack,
+					 struct nlattr *tb[])
+{
+	struct rdma_hw_stats *stats;
+	struct ib_device *device;
+	u32 index, port;
+	int ret;
+
+	if (!tb[RDMA_NLDEV_ATTR_DEV_INDEX] || !tb[RDMA_NLDEV_ATTR_PORT_INDEX])
+		return -EINVAL;
+
+	index = nla_get_u32(tb[RDMA_NLDEV_ATTR_DEV_INDEX]);
+	device = ib_device_get_by_index(sock_net(skb->sk), index);
+	if (!device)
+		return -EINVAL;
+
+	port = nla_get_u32(tb[RDMA_NLDEV_ATTR_PORT_INDEX]);
+	if (!rdma_is_port_valid(device, port)) {
+		ret = -EINVAL;
+		goto end;
+	}
+
+	stats = ib_get_hw_stats_port(device, port);
+	if (!stats) {
+		ret = -EINVAL;
+		goto end;
+	}
+
+	if (tb[RDMA_NLDEV_ATTR_STAT_HWCOUNTER_DYNAMIC])
+		ret = stat_get_doit_stats_list(skb, nlh, extack, tb,
+					       device, port, stats);
+	else
+		ret = stat_get_doit_stats_values(skb, nlh, extack, tb, device,
+						 port, stats);

This seems strange, why is the output of a get contingent on a ignored
input attribute? Shouldn't the HWCOUNTER_DYNAMIC just always be
emitted?

The CMD_STAT_GET is originally used to get the default hwcounter statistic
(the value of all hwstats), now we also want to use this command to get a
list of counters (just name and status), so kernel differentiates these 2
cases based on HWCOUNTER_DYNAMIC attr.

Don't do that, it is not how netlink works. Either the whole attribute
should be returned or you need a new get command

The netlink way is to be independent on returned parameter, if it not
supported, this parameter won't be available at all. This makes HWCOUNTER_DYNAMIC
to work exactly as netlink would do.

The issue is making the output dependent on the input:

 +	if (tb[RDMA_NLDEV_ATTR_STAT_HWCOUNTER_DYNAMIC])

Setting HWCOUNTER_DYNAMIC as an input flag to get the GET to return a
completely different output format is not netlinky

Either always return HWCOUNTER_DYNAMIC or make another query to get it

Jason