Re: [RFC net-next PATCH v2 2/2] net: dpaa2-mac: Add ACPI support for DPAA2 MAC driver

[Calvin Johnson]

On Sat, Apr 18, 2020 at 12:38:13PM +0100, Russell King - ARM Linux admin wrote:
> On Sat, Apr 18, 2020 at 04:24:32PM +0530, Calvin Johnson wrote:
> > Modify dpaa2_mac_connect() to support ACPI along with DT.
> > Modify dpaa2_mac_get_node() to get the dpmac fwnode from either
> > DT or ACPI.
> > Replace of_get_phy_mode with fwnode_get_phy_mode to get
> > phy-mode for a dpmac_node.
> > Define and use helper functions fwnode_phy_match() and
> > fwnode_phy_find_device() to find phy_dev that is later
> > connected to mac->phylink.
> > 
> > Signed-off-by: Calvin Johnson <calvin.johnson@xxxxxxxxxxx>
> > ---
> > 
> > Changes in v2:
> > - Major change following other network drivers supporting ACPI
> > - dropped v1 patches 1, 2, 4, 5 and 6 as they are no longer valid
> > - incorporated other v1 review comments
> > 
> >  .../net/ethernet/freescale/dpaa2/dpaa2-mac.c  | 122 ++++++++++++++----
> >  1 file changed, 94 insertions(+), 28 deletions(-)
> > 
> > diff --git a/drivers/net/ethernet/freescale/dpaa2/dpaa2-mac.c b/drivers/net/ethernet/freescale/dpaa2/dpaa2-mac.c
> > index 3ee236c5fc37..5a03da54a67f 100644
> > --- a/drivers/net/ethernet/freescale/dpaa2/dpaa2-mac.c
> > +++ b/drivers/net/ethernet/freescale/dpaa2/dpaa2-mac.c
> > @@ -3,6 +3,9 @@
> >  
> >  #include "dpaa2-eth.h"
> >  #include "dpaa2-mac.h"
> > +#include <linux/acpi.h>
> > +#include <linux/phy.h>
> > +#include <linux/phylink.h>
> 
> Why do you need linux/phy.h and linux/phylink.h here?  Please try
> building the driver without; you'll find they are already included
> via dpaa2-mac.h.

You are right. I'll remove them in v3

> > +static int fwnode_phy_match(struct device *dev, const void *phy_fwnode)
> > +{
> > +	return dev->fwnode == phy_fwnode;
> > +}
> > +
> > +static struct phy_device *fwnode_phy_find_device(struct fwnode_handle *phy_fwnode)
> > +{
> > +	struct device *d;
> > +	struct mdio_device *mdiodev;
> > +
> > +	if (!phy_fwnode)
> > +		return NULL;
> > +
> > +	d = bus_find_device(&mdio_bus_type, NULL, phy_fwnode, fwnode_phy_match);
> > +	if (d) {
> > +		mdiodev = to_mdio_device(d);
> > +		if (mdiodev->flags & MDIO_DEVICE_FLAG_PHY)
> > +			return to_phy_device(d);
> > +		put_device(d);
> > +	}
> > +
> > +	return NULL;
> > +}
> 
> This is groping around in the mdio bus implementation details; drivers
> must not do this layering violation.  Please propose an interface in
> the mdio code to do what you need.

I'll study and propose a solution.

> 
> > +
> >  int dpaa2_mac_connect(struct dpaa2_mac *mac)
> >  {
> >  	struct fsl_mc_device *dpmac_dev = mac->mc_dev;
> >  	struct net_device *net_dev = mac->net_dev;
> > -	struct device_node *dpmac_node;
> > +	struct fwnode_handle *dpmac_node = NULL;
> > +	struct fwnode_reference_args args;
> > +	struct phy_device *phy_dev;
> >  	struct phylink *phylink;
> >  	struct dpmac_attr attr;
> > +	int status;
> >  	int err;
> >  
> >  	err = dpmac_open(mac->mc_io, 0, dpmac_dev->obj_desc.id,
> > @@ -251,7 +299,7 @@ int dpaa2_mac_connect(struct dpaa2_mac *mac)
> >  
> >  	mac->if_link_type = attr.link_type;
> >  
> > -	dpmac_node = dpaa2_mac_get_node(attr.id);
> > +	dpmac_node = dpaa2_mac_get_node(&mac->mc_dev->dev, attr.id);
> >  	if (!dpmac_node) {
> >  		netdev_err(net_dev, "No dpmac@%d node found.\n", attr.id);
> >  		err = -ENODEV;
> > @@ -269,7 +317,7 @@ int dpaa2_mac_connect(struct dpaa2_mac *mac)
> >  	 * error out if the interface mode requests them and there is no PHY
> >  	 * to act upon them
> >  	 */
> > -	if (of_phy_is_fixed_link(dpmac_node) &&
> > +	if (of_phy_is_fixed_link(to_of_node(dpmac_node)) &&
> >  	    (mac->if_mode == PHY_INTERFACE_MODE_RGMII_ID ||
> >  	     mac->if_mode == PHY_INTERFACE_MODE_RGMII_RXID ||
> >  	     mac->if_mode == PHY_INTERFACE_MODE_RGMII_TXID)) {
> > @@ -282,7 +330,7 @@ int dpaa2_mac_connect(struct dpaa2_mac *mac)
> >  	mac->phylink_config.type = PHYLINK_NETDEV;
> >  
> >  	phylink = phylink_create(&mac->phylink_config,
> > -				 of_fwnode_handle(dpmac_node), mac->if_mode,
> > +				 dpmac_node, mac->if_mode,
> >  				 &dpaa2_mac_phylink_ops);
> >  	if (IS_ERR(phylink)) {
> >  		err = PTR_ERR(phylink);
> > @@ -290,20 +338,38 @@ int dpaa2_mac_connect(struct dpaa2_mac *mac)
> >  	}
> >  	mac->phylink = phylink;
> >  
> > -	err = phylink_of_phy_connect(mac->phylink, dpmac_node, 0);
> > +	if (is_of_node(dpmac_node))
> > +		err = phylink_of_phy_connect(mac->phylink,
> > +					     to_of_node(dpmac_node), 0);
> > +	else if (is_acpi_node(dpmac_node)) {
> > +		status = acpi_node_get_property_reference(dpmac_node,
> > +							  "phy-handle",
> > +							  0, &args);
> > +		if (ACPI_FAILURE(status))
> > +			goto err_phylink_destroy;
> > +		phy_dev = fwnode_phy_find_device(args.fwnode);
> > +		if (!phy_dev)
> > +			goto err_phylink_destroy;
> > +
> > +		err = phylink_connect_phy(mac->phylink, phy_dev);
> > +		if (err)
> > +			phy_detach(phy_dev);
> 
> phy_detach() reverses the effect of phy_attach_direct().  This is not
> the correct cleanup function in this case, because the PHY hasn't been
> attached (and phylink_connect_phy() will clean up any effects it has
> on error.)  You only need to reverse the effect of your
> fwnode_phy_find_device(), which phy_detach() is inappropriate for.

Got it. I'll repair this part.

> 
> In any case, if this method of getting a PHY is accepted by ACPI folk,
> it could be moved into a helper in phylink_fwnode_phy_connect() - and
> that really needs to happen before a patch adding this functionality is
> acceptable.

There is already similar code in upstream kernel:
https://elixir.bootlin.com/linux/v5.7-rc2/source/drivers/net/ethernet/apm/xgene/xgene_enet_hw.c#L825
It makes sense to have a generic helper. Will create one.
Hope I can introduce it in the v3 patchset, ofcourse phylink_fwnode_phy_connect will be
defined in a patch before it is called. Let me know if it is not okay.

> 
> > +	}
> >  	if (err) {
> > -		netdev_err(net_dev, "phylink_of_phy_connect() = %d\n", err);
> > +		netdev_err(net_dev, "phylink_fwnode_phy_connect() = %d\n", err);
> 
> That's a very misleading change - there is no function named as per your
> new name.
My bad. Sorry. Will correct it.

Thanks
Calvin