| Message ID | 20210508002920.19945-26-ansuelsmth@gmail.com (mailing list archive) |
|---|---|
| State | Superseded |
| Delegated to: | Netdev Maintainers |
| Headers | show |
| Series | [RFC,net-next,v4,01/28] net: mdio: ipq8064: clean whitespaces in define | expand |
| Context | Check | Description |
|---|---|---|
| netdev/apply | fail | Patch does not apply to net-next |
| netdev/tree_selection | success | Clearly marked for net-next |
On 5/7/2021 5:29 PM, Ansuel Smith wrote: > Some switch doesn't have a 1:1 map phy to port. Permit driver to provide > a custom phy_mii_mask so the internal mdiobus can correctly use the > provided phy reg as it can differ from the port reg. > The qca8k driver is provided as a first user of this function. Why not have qca8k be in charge of allocating its internal MDIO bus like what mv88e6xxx or bcm_sf2 do? That would allow you to do all sorts of customization there and you could skip having patches 23 and 24.
On Sat, May 08, 2021 at 08:52:03AM -0700, Florian Fainelli wrote: > > > On 5/7/2021 5:29 PM, Ansuel Smith wrote: > > Some switch doesn't have a 1:1 map phy to port. Permit driver to provide > > a custom phy_mii_mask so the internal mdiobus can correctly use the > > provided phy reg as it can differ from the port reg. > > The qca8k driver is provided as a first user of this function. > > Why not have qca8k be in charge of allocating its internal MDIO bus like > what mv88e6xxx or bcm_sf2 do? That would allow you to do all sorts of > customization there and you could skip having patches 23 and 24. Oh ok, I will implement the internal MDIO bus directly in the qca8k driver. Was thinking... Should I keep the extra mdio node and move the documentation to qca8k or should I handle all of that internally? To me it looks cleaner the direct definition in the devicetree than the port_to_phy function in the driver but as you can see extra checks are required to handle the new implementation and still supports the old one.
diff --git a/drivers/net/dsa/qca8k.c b/drivers/net/dsa/qca8k.c index 3d195fdd7ed5..3c3e05735b2d 100644 --- a/drivers/net/dsa/qca8k.c +++ b/drivers/net/dsa/qca8k.c @@ -1685,7 +1685,37 @@ qca8k_get_tag_protocol(struct dsa_switch *ds, int port, return DSA_TAG_PROTO_QCA; } +static u32 +qca8k_get_phys_mii_mask(struct dsa_switch *ds) +{ + struct device_node *mdio, *phy; + u32 reg, phy_mii_mask = 0; + int err; + + mdio = of_get_child_by_name(ds->dev->of_node, "mdio"); + if (mdio) { + for_each_available_child_of_node(mdio, phy) { + err = of_property_read_u32(phy, "reg", ®); + if (err) { + of_node_put(phy); + of_node_put(mdio); + return 0; + } + + phy_mii_mask |= BIT(reg); + } + + of_node_put(mdio); + return phy_mii_mask; + } + + /* Fallback to the lagacy mapping if mdio node is not found */ + dev_warn(ds->dev, "Using the legacy phys_mii_mapping. Consider updating the dts."); + return dsa_user_ports(ds); +} + static const struct dsa_switch_ops qca8k_switch_ops = { + .get_phys_mii_mask = qca8k_get_phys_mii_mask, .get_tag_protocol = qca8k_get_tag_protocol, .setup = qca8k_setup, .get_strings = qca8k_get_strings, diff --git a/include/net/dsa.h b/include/net/dsa.h index 83a933e563fe..4003ffc659a4 100644 --- a/include/net/dsa.h +++ b/include/net/dsa.h @@ -511,6 +511,13 @@ struct dsa_switch_ops { void (*teardown)(struct dsa_switch *ds); u32 (*get_phy_flags)(struct dsa_switch *ds, int port); + /* + * Provide a custom phys_mii_mask for the dsa slave mdiobus instead + * of relying on the dsa_user_ports. Not every switch has a 1:1 map + * port to PHY, hence the driver can provide their fixed mask. + */ + u32 (*get_phys_mii_mask)(struct dsa_switch *ds); + /* * Access to the switch's PHY registers. */ diff --git a/net/dsa/dsa2.c b/net/dsa/dsa2.c index 79adabe3e2a7..7eabd4d67849 100644 --- a/net/dsa/dsa2.c +++ b/net/dsa/dsa2.c @@ -682,8 +682,13 @@ static int dsa_switch_setup(struct dsa_switch *ds) * driver and before ops->setup() has run, since the switch drivers and * the slave MDIO bus driver rely on these values for probing PHY * devices or not + * Driver can provide his on mask as some switch doesn't have a 1:1 map + * phy to port. */ - ds->phys_mii_mask |= dsa_user_ports(ds); + if (ds->ops->get_phys_mii_mask) + ds->phys_mii_mask = ds->ops->get_phys_mii_mask(ds); + else + ds->phys_mii_mask |= dsa_user_ports(ds); /* Add the switch to devlink before calling setup, so that setup can * add dpipe tables
Some switch doesn't have a 1:1 map phy to port. Permit driver to provide a custom phy_mii_mask so the internal mdiobus can correctly use the provided phy reg as it can differ from the port reg. The qca8k driver is provided as a first user of this function. Signed-off-by: Ansuel Smith <ansuelsmth@gmail.com> --- drivers/net/dsa/qca8k.c | 30 ++++++++++++++++++++++++++++++ include/net/dsa.h | 7 +++++++ net/dsa/dsa2.c | 7 ++++++- 3 files changed, 43 insertions(+), 1 deletion(-)