On Mon, 9 Nov 2026 at 14:34, Bin Meng <[email protected]> wrote: > > The GEM MMIO callbacks ignored the low address bits and access size. > Byte reads therefore always returned lane zero, causing guest driver > that reads the station address byte by byte to turn four unique MAC > addresses into the same value. Partial writes had the corresponding > lane-selection problem. > > Extract the requested little-endian read lane and merge partial writes > with the current register value. Preserve the existing read-clear and > write-clear behavior and dynamic queue pointer readback. > > With this change, reset the TX queue pointers only when a write covers > TX-enable and leaves it clear, and only start a PHY operation when a > write covers the OP/ST fields in the PHY maintenance register. > > Fixes: e9f186e514a7 ("cadence_gem: initial version of device model") > Signed-off-by: Bin Meng <[email protected]> > Reviewed-by: Philippe Mathieu-Daudé <[email protected]>
This device is used by both arm and riscv machines. Alistair, do you plan to take it via the riscv queue? thanks -- PMM
