It may happen that the serial IP block is performing some ongoing transmission (started at e.g. board_init()) when the serial "probe" is called.
As a result the serial port IP block is reset, so transmitted data is corrupted:
I2C: ready DRAM: 1 GiB jSS('HH��SL_SDHC: 04 rev 0x0
This patch prevents from this situation, by defining pre_probe() callback in which we wait till the TX buffer is empty (from previous transmission):
I2C: ready DRAM: 1 GiB ID: unit type 0x4 rev 0x0
All defined ->pending callbacks at ./drivers/serial are non blocking - just simple reading from registers and testing flags. Hence, it should be enough to not use any timeout from timer. One shall also note that we enable console very early - not all timers may be ready for work - adding timeout here would impose implicit dependency that timers are setup before serial.
Signed-off-by: Lukasz Majewski lukma@denx.de ---
drivers/serial/serial-uclass.c | 20 ++++++++++++++++++++ 1 file changed, 20 insertions(+)
diff --git a/drivers/serial/serial-uclass.c b/drivers/serial/serial-uclass.c index 2e5116f..5e6964d 100644 --- a/drivers/serial/serial-uclass.c +++ b/drivers/serial/serial-uclass.c @@ -420,10 +420,30 @@ static int serial_pre_remove(struct udevice *dev) return 0; }
+static int serial_pre_probe(struct udevice *dev) +{ + struct dm_serial_ops *ops = serial_get_ops(dev); + int ret = 0; + + /* + * Wait for any ongoing transmission to finish - for example + * from pre-relocation enabled UART + */ + if (ops && ops->pending) + do { + ret = ops->pending(dev, false); + if (ret < 0) + break; + } while (ret > 0); + + return ret; +} + UCLASS_DRIVER(serial) = { .id = UCLASS_SERIAL, .name = "serial", .flags = DM_UC_FLAG_SEQ_ALIAS, + .pre_probe = serial_pre_probe, .post_probe = serial_post_probe, .pre_remove = serial_pre_remove, .per_device_auto_alloc_size = sizeof(struct serial_dev_priv),