Re: [PATCH net-next v5 07/13] net: ethtool: Introduce a command to list PHYs on an interface

[Jakub Kicinski]

On Thu, 21 Dec 2023 19:00:40 +0100 Maxime Chevallier wrote:
> As we have the ability to track the PHYs connected to a net_device
> through the link_topology, we can expose this list to userspace. This
> allows userspace to use these identifiers for phy-specific commands and
> take the decision of which PHY to target by knowing the link topology.
> 
> Add PHY_GET and PHY_DUMP, which can be a filtered DUMP operation to list
> devices on only one interface.
> 
> Signed-off-by: Maxime Chevallier <maxime.chevallier@xxxxxxxxxxx>

> diff --git a/Documentation/networking/ethtool-netlink.rst b/Documentation/networking/ethtool-netlink.rst
> index 3ca6c21e74af..97ff787a7dd8 100644
> --- a/Documentation/networking/ethtool-netlink.rst
> +++ b/Documentation/networking/ethtool-netlink.rst
> @@ -2011,6 +2011,49 @@ The attributes are propagated to the driver through the following structure:
>  .. kernel-doc:: include/linux/ethtool.h
>      :identifiers: ethtool_mm_cfg
>  
> +PHY_GET
> +=======
> +
> +Retrieve information about a given Ethernet PHY sitting on the link. As there
> +can be more than one PHY, the DUMP operation can be used to list the PHYs
> +present on a given interface, by passing an interface index or name in
> +the dump request
> +
> +Request contents:
> +
> +  ====================================  ======  ==========================
> +  ``ETHTOOL_A_PHY_HEADER``              nested  request header
> +  ====================================  ======  ==========================
> +
> +Kernel response contents:
> +
> +  ===================================== ======  ==========================
> +  ``ETHTOOL_A_PHY_HEADER``              nested  request header
> +  ``ETHTOOL_A_PHY_INDEX``               u32     the phy's unique index, that can

The fact that lines are longer than the ===== markings doesn't generate
warnings in htmldoc?

> +                                                be used for phy-specific requests
> +  ``ETHTOOL_A_PHY_DRVNAME``             string  the phy driver name
> +  ``ETHTOOL_A_PHY_NAME``                string  the phy device name
> +  ``ETHTOOL_A_PHY_UPSTREAM_TYPE``       u32     the type of device this phy is
> +                                                connected to
> +  ``ETHTOOL_A_PHY_UPSTREAM_PHY``        nested  if the phy is connected to another
> +                                                phy, this nest contains info on
> +                                                that connection
> +  ``ETHTOOL_A_PHY_DOWNSTREAM_SFP_NAME`` string  if the phy controls an sfp bus,
> +                                                the name of the sfp bus

Is upstream / downstream clear to everyone / from the spec.
I guess it's scoped to the netdev so upstream means "towards 
the netdev MAC"?

> +  ``ETHTOOL_A_PHY_ID``                  u32     the phy id if the phy is C22
> +  ===================================== ======  ==========================
> +
> +When ``ETHTOOL_A_PHY_UPSTREAM_TYPE`` is PHY_UPSTREAM_PHY, the PHY's parent is
> +another PHY. Information on the parent PHY will be set in the
> +``ETHTOOL_A_PHY_UPSTREAM_PHY`` nest, which has the following structure :
> +
> +  =================================== ======  ==========================
> +  ``ETHTOOL_A_PHY_UPSTREAM_INDEX``    u32     the PHY index of the upstream PHY
> +  ``ETHTOOL_A_PHY_UPSTREAM_SFP_NAME`` string  if this PHY is connected to it's
> +                                                parent PHY through an SFP bus, the
> +                                                name of this sfp bus
> +  =================================== ======  ==========================

Why is this a nest?

>  Request translation
>  ===================

> +enum {
> +	ETHTOOL_A_PHY_UNSPEC,
> +	ETHTOOL_A_PHY_HEADER,			/* nest - _A_HEADER_* */
> +	ETHTOOL_A_PHY_INDEX,			/* u32 */
> +	ETHTOOL_A_PHY_DRVNAME,			/* string */
> +	ETHTOOL_A_PHY_NAME,			/* string */
> +	ETHTOOL_A_PHY_UPSTREAM_TYPE,		/* u8 */

The Documentation say it's a u32 as it should be, AFAICT.
But code and some comments use u8.

> +	ETHTOOL_A_PHY_UPSTREAM,			/* nest - _A_PHY_UPSTREAM_* */
> +	ETHTOOL_A_PHY_DOWNSTREAM_SFP_NAME,	/* string */
> +	ETHTOOL_A_PHY_ID,			/* u32 */
> +
> +	/* add new constants above here */
> +	__ETHTOOL_A_PHY_CNT,
> +	ETHTOOL_A_PHY_MAX = (__ETHTOOL_A_PHY_CNT - 1)
> +};

> +++ b/net/ethtool/phy.c
> @@ -0,0 +1,306 @@
> +// SPDX-License-Identifier: GPL-2.0-only
> +/*
> + * Copyright 2023 Bootlin
> + *
> + */

Do you really need 4 lines for the copyright? :)


> +/* Caller holds rtnl */
> +static ssize_t
> +ethnl_phy_reply_size(const struct ethnl_req_info *req_base,
> +		     struct netlink_ext_ack *extack)
> +{
> +	struct phy_link_topology *topo;
> +	struct phy_device_node *pdn;
> +	struct phy_device *phydev;
> +	unsigned long index;
> +	size_t size;
> +
> +	ASSERT_RTNL();
> +
> +	topo = &req_base->dev->link_topo;
> +
> +	size = nla_total_size(0);

no comment on this one?

> +
> +	xa_for_each(&topo->phys, index, pdn) {

Why count all the PHYs, you only output one on doit, right?

> +		phydev = pdn->phy;
> +
> +		/* ETHTOOL_A_PHY_INDEX */
> +		size += nla_total_size(sizeof(u32));
> +
> +		/* ETHTOOL_A_DRVNAME */
> +		size += nla_total_size(strlen(phydev->drv->name) + 1);
> +
> +		/* ETHTOOL_A_NAME */
> +		size += nla_total_size(strlen(dev_name(&phydev->mdio.dev)) + 1);
> +
> +		/* ETHTOOL_A_PHY_UPSTREAM_TYPE */
> +		size += nla_total_size(sizeof(u8));
> +
> +		/* ETHTOOL_A_PHY_ID */
> +		size += nla_total_size(sizeof(u32));
> +
> +		if (phy_on_sfp(phydev)) {
> +			const char *upstream_sfp_name = sfp_get_name(pdn->parent_sfp_bus);
> +
> +			/* ETHTOOL_A_PHY_UPSTREAM_SFP_NAME */
> +			if (upstream_sfp_name)
> +				size += nla_total_size(strlen(upstream_sfp_name) + 1);
> +
> +			/* ETHTOOL_A_PHY_UPSTREAM_INDEX */
> +			size += nla_total_size(sizeof(u32));
> +		}
> +
> +		/* ETHTOOL_A_PHY_DOWNSTREAM_SFP_NAME */
> +		if (phydev->sfp_bus) {
> +			const char *sfp_name = sfp_get_name(phydev->sfp_bus);
> +
> +			if (sfp_name)
> +				size += nla_total_size(strlen(sfp_name) + 1);
> +		}
> +	}
> +
> +	return size;
> +}

> +static int ethnl_phy_parse_request(struct ethnl_req_info *req_base,
> +				   struct nlattr **tb)
> +{
> +	struct phy_link_topology *topo = &req_base->dev->link_topo;
> +	struct phy_req_info *req_info = PHY_REQINFO(req_base);
> +	struct phy_device_node *pdn;
> +
> +	if (!req_base->phydev)
> +		return 0;

The PHY INDEX should probably be a required attr, with 
GENL_REQ_ATTR_CHECK()? Without phydev being specified
what's the point?

> +	pdn = xa_load(&topo->phys, req_base->phydev->phyindex);
> +	memcpy(&req_info->pdn, pdn, sizeof(*pdn));
> +
> +	return 0;
> +}

> +int ethnl_phy_dumpit(struct sk_buff *skb, struct netlink_callback *cb)
> +{
> +	struct ethnl_phy_dump_ctx *ctx = (void *)cb->ctx;
> +	struct net *net = sock_net(skb->sk);
> +	unsigned long ifindex = 1;

This doesn't look right, if dump gets full you gotta pick up
when previous call left off.

> +	struct net_device *dev;
> +	int ret = 0;
> +
> +	rtnl_lock();
> +
> +	if (ctx->phy_req_info->base.dev) {
> +		ret = ethnl_phy_dump_one_dev(skb, ctx->phy_req_info->base.dev, cb);
> +		ethnl_parse_header_dev_put(&ctx->phy_req_info->base);
> +		ctx->phy_req_info->base.dev = NULL;
> +	} else {
> +		for_each_netdev_dump(net, dev, ifindex) {
> +			ret = ethnl_phy_dump_one_dev(skb, dev, cb);
> +			if (ret)
> +				break;
> +		}
> +	}
> +	rtnl_unlock();
> +
> +	if (ret == -EMSGSIZE && skb->len)
> +		return skb->len;
> +	return ret;
> +}
> +