The serial_check_stdout() has code to bind devices that are not marked to be bound before relocation. This code had a known bug in the past and should not be needed if targets correctly mark these devices as bound before relocation in the device tree or mark the appropriate device drivers using the DM_FLAG_PRE_RELOC flag.
Note that this commit will break many targets. Those targets will have need to have the appropriate serial device marked to be bound before relocation to make them work again.
Signed-off-by: Mark Kettenis kettenis@openbsd.org ---
Note: this is defenitely not material for 2022.04. I'm not even sure if this is a good idea. But if this does get accepted, it is probably wise for it to go onto the next branch to give maintainers enough time to fix any fallout.
drivers/serial/serial-uclass.c | 11 ----------- 1 file changed, 11 deletions(-)
diff --git a/drivers/serial/serial-uclass.c b/drivers/serial/serial-uclass.c index f30f352bd7..0da5fde429 100644 --- a/drivers/serial/serial-uclass.c +++ b/drivers/serial/serial-uclass.c @@ -60,17 +60,6 @@ static int serial_check_stdout(const void *blob, struct udevice **devp) if (!uclass_get_device_by_of_offset(UCLASS_SERIAL, node, devp)) return 0;
- /* - * If the console is not marked to be bound before relocation, bind it - * anyway. - */ - if (node > 0 && !lists_bind_fdt(gd->dm_root, offset_to_ofnode(node), - devp, NULL, false)) { - if (device_get_uclass_id(*devp) == UCLASS_SERIAL && - !device_probe(*devp)) - return 0; - } - return -ENODEV; }