The ksz9477 is currently the only driver using dev_set_parent_priv() outside of the driver model. Also, there was no explanation in the commit adding ksz9477 driver and why dev_set_parent_priv() is required.
Actually there is a typo in ksz_mdio_probe() while retrieving the parent (switch@0) private data:
- priv->ksz = dev_get_parent_priv(dev->parent); + priv->ksz = dev_get_priv(dev->parent);
Printing the address of struct ksz_dsa_priv *priv allows to notice the slight difference:
ksz_i2c_probe: ksz_dsa_priv *priv 0xfdf45768 // address of the saved priv ksz_mdio_bind: ksz_dsa_priv *priv 0xfdf45798 // address returned by dev_get_parent_priv(dev->parent) ksz_mdio_bind: ksz_dsa_priv *priv 0xfdf45768 // address returned by dev_get_priv(dev->parent)
The ksz_mdio driver get the wrong data and without dev_set_parent_priv() the mdio driver fail to access the underlying bus.
While it doesn't cause any issue with I2C bus, it override the per-child data used by the SPI bus (struct spi_slave) and prevent further bus access (even with sspi command).
Signed-off-by: Romain Naour romain.naour@smile.fr --- Tested using the upcoming ksz9477 driver with spi support added later in this patch series. --- drivers/net/ksz9477.c | 4 +--- 1 file changed, 1 insertion(+), 3 deletions(-)
diff --git a/drivers/net/ksz9477.c b/drivers/net/ksz9477.c index 091e9689487..e67012a071a 100644 --- a/drivers/net/ksz9477.c +++ b/drivers/net/ksz9477.c @@ -276,7 +276,7 @@ static int ksz_mdio_probe(struct udevice *dev) struct ksz_mdio_priv *priv = dev_get_priv(dev);
dev_dbg(dev, "%s\n", __func__); - priv->ksz = dev_get_parent_priv(dev->parent); + priv->ksz = dev_get_priv(dev->parent);
return 0; } @@ -514,8 +514,6 @@ static int ksz_i2c_probe(struct udevice *dev) u8 data8; u32 id;
- dev_set_parent_priv(dev, priv); - ret = i2c_set_chip_offset_len(dev, 2); if (ret) { printf("i2c_set_chip_offset_len failed: %d\n", ret);