Hi Andrew, On 23-04-05, Andrew Lunn wrote: > On Wed, Apr 05, 2023 at 11:26:51AM +0200, Marco Felsch wrote: > > The current phy reset handling is broken in a way that it needs > > pre-running firmware to setup the phy initially. Since the very first > > step is to readout the PHYID1/2 registers before doing anything else. > > > > The whole dection logic will fall apart if the pre-running firmware > > don't setup the phy accordingly or the kernel boot resets GPIOs states > > or disables clocks. In such cases the PHYID1/2 read access will fail and > > so the whole detection will fail. > > > > I fixed this via this series, the fix will include a new kernel API > > called phy_device_atomic_register() which will do all necessary things > > and return a 'struct phy_device' on success. So setting up a phy and the > > phy state machine is more convenient. > > Please add a section explaining why the current API is broken beyond > repair. You need to justify adding a new call, rather than fixing the > existing code to just do what is necessary to allow the PHY to be > found. TIL from Florian that you use the cover-letter information in your merge commits. I will adapt the cover-letter accordingly and mention why this PR introduces a new API. Regards, Marco > > Andrew >
