The barebox driver currently only ensures that the switch is out of reset, but doesn't actually trigger a reset pulse. The Linux driver, on the other hand, holds reset active for 10ms and then waits a whopping 100ms after reset deassertion. This seems excessive by a thousandfold for at least KSZ9893R[1], whose datasheet states: After the de-assertion of reset, it is recommended to wait a minimum of 100µs before starting to program the device through any interface. Therefore, let's pulse the reset like Linux does and give the switch 100 microseconds before continuing with the driver probe. [1]: DS00002420E-page 176 Signed-off-by: Ahmad Fatoum <a.fatoum@xxxxxxxxxxxxxx> --- drivers/net/ksz9477.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) diff --git a/drivers/net/ksz9477.c b/drivers/net/ksz9477.c index 1abea9d04017..5c5f4ec8d995 100644 --- a/drivers/net/ksz9477.c +++ b/drivers/net/ksz9477.c @@ -433,8 +433,10 @@ static int microchip_switch_probe(struct device *dev) if (IS_ERR(gpio)) { dev_warn(dev, "Failed to get 'reset' GPIO (ignored)\n"); } else if (gpio) { - mdelay(1); + gpiod_set_value(gpio, true); + mdelay(10); gpiod_set_value(gpio, false); + udelay(100); } ksz_reset_switch(dev->priv); -- 2.39.2