[PATCH net-next] net: dsa: microchip: ksz8795: Fix KSZ8794 port map again

[PATCH net-next] net: dsa: microchip: ksz8795: Fix KSZ8794 port map again

[Marek Vasut]

The KSZ8795 switch has 4 external ports {0,1,2,3} and 1 CPU port {4}, so
does the KSZ8765. The KSZ8794 seems to be repackaged KSZ8795 with different
ID and port 3 not routed out, however the port 3 registers are present in
the silicon, so the KSZ8794 switch has 3 external ports {0,1,2} and 1 CPU
port {4}. Currently the driver always uses the last port as CPU port, on
KSZ8795/KSZ8765 that is port 4 and that is OK, but on KSZ8794 that is port
3 and that is not OK, as it must also be port 4.

This patch adjusts the driver such that it always registers a switch with
5 ports total (4 external ports, 1 CPU port), always sets the CPU port to
switch port 4, and then configures the external port mask according to
the switch model -- 3 ports for KSZ8794 and 4 for KSZ8795/KSZ8765.

Fixes: 68a1b676db52 ("net: dsa: microchip: ksz8795: remove superfluous port_cnt assignment")
Fixes: 4ce2a984abd8 ("net: dsa: microchip: ksz8795: use phy_port_cnt where possible")
Fixes: 241ed719bc98 ("net: dsa: microchip: ksz8795: use port_cnt instead of TOTOAL_PORT_NUM")
Signed-off-by: Marek Vasut <marex@denx.de>
Cc: Andrew Lunn <andrew@lunn.ch>
Cc: Florian Fainelli <f.fainelli@gmail.com>
Cc: Michael Grzeschik <m.grzeschik@pengutronix.de>
Cc: Jakub Kicinski <kuba@kernel.org>
---
NOTE: In case of KSZ8794, the DT node port@4 is the CPU port and
      port@3 should not be defined.
---
 drivers/net/dsa/microchip/ksz8795.c | 30 +++++++++++++++++++----------
 1 file changed, 20 insertions(+), 10 deletions(-)

diff --git a/drivers/net/dsa/microchip/ksz8795.c b/drivers/net/dsa/microchip/ksz8795.c
index c973db101b72..a4570ba29c83 100644
--- a/drivers/net/dsa/microchip/ksz8795.c
+++ b/drivers/net/dsa/microchip/ksz8795.c
@@ -1187,6 +1187,20 @@ static const struct ksz_chip_data ksz8795_switch_chips[] = {
 		.port_cnt = 5,		/* total cpu and user ports */
 	},
 	{
+		/*
+		 * WARNING
+		 * =======
+		 * KSZ8794 is similar to KSZ8795, except the port map
+		 * contains a gap between external and CPU ports, the
+		 * port map is NOT continuous. The per-port register
+		 * map is shifted accordingly too, i.e. registers at
+		 * offset 0x40 are NOT used on KSZ8794 and they ARE
+		 * used on KSZ8795 for external port 3.
+		 *           external  cpu
+		 * KSZ8794   0,1,2      4
+		 * KSZ8795   0,1,2,3    4
+		 * KSZ8765   0,1,2,3    4
+		 */
 		.chip_id = 0x8794,
 		.dev_name = "KSZ8794",
 		.num_vlans = 4096,
@@ -1220,9 +1234,13 @@ static int ksz8795_switch_init(struct ksz_device *dev)
 			dev->num_vlans = chip->num_vlans;
 			dev->num_alus = chip->num_alus;
 			dev->num_statics = chip->num_statics;
-			dev->port_cnt = chip->port_cnt;
+			dev->port_cnt = fls(chip->cpu_ports);
+			dev->cpu_port = fls(chip->cpu_ports) - 1;
+			dev->phy_port_cnt = dev->port_cnt - 1;
 			dev->cpu_ports = chip->cpu_ports;
-
+			dev->host_mask = chip->cpu_ports;
+			dev->port_mask = (BIT(dev->phy_port_cnt) - 1) |
+					 chip->cpu_ports;
 			break;
 		}
 	}
@@ -1231,17 +1249,9 @@ static int ksz8795_switch_init(struct ksz_device *dev)
 	if (!dev->cpu_ports)
 		return -ENODEV;
 
-	dev->port_mask = BIT(dev->port_cnt) - 1;
-	dev->port_mask |= dev->host_mask;
-
 	dev->reg_mib_cnt = KSZ8795_COUNTER_NUM;
 	dev->mib_cnt = ARRAY_SIZE(mib_names);
 
-	dev->phy_port_cnt = dev->port_cnt - 1;
-
-	dev->cpu_port = dev->port_cnt - 1;
-	dev->host_mask = BIT(dev->cpu_port);
-
 	dev->ports = devm_kzalloc(dev->dev,
 				  dev->port_cnt * sizeof(struct ksz_port),
 				  GFP_KERNEL);
-- 
2.29.2

Re: [PATCH net-next] net: dsa: microchip: ksz8795: Fix KSZ8794 port map again

[Andrew Lunn]

On Tue, Jan 19, 2021 at 09:55:18PM +0100, Marek Vasut wrote:
> The KSZ8795 switch has 4 external ports {0,1,2,3} and 1 CPU port {4}, so
> does the KSZ8765. The KSZ8794 seems to be repackaged KSZ8795 with different
> ID and port 3 not routed out, however the port 3 registers are present in
> the silicon, so the KSZ8794 switch has 3 external ports {0,1,2} and 1 CPU
> port {4}. Currently the driver always uses the last port as CPU port, on
> KSZ8795/KSZ8765 that is port 4 and that is OK, but on KSZ8794 that is port
> 3 and that is not OK, as it must also be port 4.
> 
> This patch adjusts the driver such that it always registers a switch with
> 5 ports total (4 external ports, 1 CPU port), always sets the CPU port to
> switch port 4, and then configures the external port mask according to
> the switch model -- 3 ports for KSZ8794 and 4 for KSZ8795/KSZ8765.

Hi Marek

What appears to be missing is any checks for somebody trying to use
the 'somewhat non-existent' port.

Assuming all the drivers set dev->port_mask properly, i think you can
add a check to ksz_enable_port() and return -EINVAL if the requested
port is not a member of port_mask.

     Andrew