On 11/29/18 2:03 PM, Jeff Kirsher wrote: > On Thu, 2018-11-29 at 18:54 +0000, Steve Douthit wrote: >> Most dsa devices currently expect to get a pointer to a mii_bus from >> platform data of device tree information. >> >> The first patch exposes a mii_bus for ixgbe devices. >> >> Currently the ixgbe driver only allows accesses to the probed PHY >> address via ioctls. The second patch changes that behavior to allow >> access to all addresses. This should be useful for doing switch peeks & >> pokes from userspace for development, test, or if theres no dsa driver >> for a particular switch yet. >> >> I've tested the clause 22 portion of this code on a board that has an >> Intel C3xxx series SoC connected to a Marvell Peridot switch. I don't >> have any clause 45 devices or other ixgbe devices to test with. >> >> Stephen Douthit (2): >> ixgbe: register a mdiobus >> ixgbe: use mii_bus to handle MII related ioctls >> >> drivers/net/ethernet/intel/ixgbe/ixgbe.h | 2 + >> drivers/net/ethernet/intel/ixgbe/ixgbe_main.c | 35 ++-- >> drivers/net/ethernet/intel/ixgbe/ixgbe_phy.c | 192 ++++++++++++++++++ >> drivers/net/ethernet/intel/ixgbe/ixgbe_phy.h | 2 + >> 4 files changed, 216 insertions(+), 15 deletions(-) >> > > I will add these changes to my queue, please remember to send patches > against the drivers in drivers/net/ethernet/intel/* to > intel-wired-...@lists.osuosl.org mailing list as well, so that our > patchwork project can track the patches.
Sorry about that, I'll add it to the CC list for the next rev. I started adding another board to the platform driver that needs this code and found a bug. The new board has more lanes enabled via soft straps. That caused sysfs to complain because I ended up with duplicate bus IDs from this code: + /* Use the position of the device in the PCI hierarchy as the id */ + snprintf(bus->id, MII_BUS_ID_SIZE, "%s-mdio-%s", ixgbe_driver_name, + pci_name(pdev->bus->self)); Please drop this series for now. Thinking about this some more I probably need to ensure that only a single mii_bus is registered out of all the discovered x550em_a devices. More generally there's not always going to be a 1:1 mapping between MACs and MIIs on these devices and the code should handle that. I think for the next rev I'll only register a single mii_bus for the lowest numbered bus:device.function x500em_a device, and skip mii_bus setup entirely for other device IDs. Let me know if there's another/better way to tackle that.
