On 7/10/20 4:35 PM, Andrew Lunn wrote: >> In principle there is nothing in this para-virtualization design that >> would preclude a quirky quad PHY from being accessed in a >> virtualization-safe mode. The main use case for PHY access in a VM is >> for detecting when the link went down. Worst case, the QEMU host-side >> driver could lie about the PHY ID, and could only expose the clause 22 >> subset of registers that could make it compatible with genphy. I don't >> think this changes the overall approach about how MDIO devices would be >> virtualized with QEMU. > > A more generic solution might be to fully virtualize the PHY. Let the > host kernel drive the PHY, and QEMU can use /sys/bus/mdio_bus/devices, > and uevents sent to user space. Ioana already added support for a PHY > not bound to a MAC in phylink. You would need to add a UAPI for > start/stop, and maybe a couple more operations, and probably export a > bit more information.
You would still need a struct device to bind to that PHY and I am not sure what device that might be since the userspace cannot provide one. > > This would then solve the quad PHY situation, and any other odd > setups. And all the VM would require is genphy, keeping it simple. > > Andrew > > How would the genphy driver work if there is no MDIO register map for it to access? This suggestion seems something in between a software PHY and hardware PHY. Also, it would work only on PHYs and not any other device accessible over an MDIO bus (so you couldn't assign a DSA switch to a VM). Coming back to a point that you made earlier, even as we speak, with an upstream kernel you can still have an userspace application accessing devices on a MDIO bus. Since the MDIO controller is memory mapped you just need some devmem commands wrapped-up in a nice script, no need for a vendor patch over an upstream kernel for this. Ioana
