On Wed, Nov 30, 2016 at 2:16 AM, Olliver Schinagl oliver@schinagl.nl wrote:
Hey Simon,
On 29-11-16 22:41, Simon Glass wrote:
Hi Oliver,
On 28 November 2016 at 03:38, Olliver Schinagl oliver@schinagl.nl wrote:
On 27-11-16 18:02, Simon Glass wrote:
Hi,
On 25 November 2016 at 08:38, Olliver Schinagl oliver@schinagl.nl wrote:
Add the read_rom_hwaddr net_op hook so that it can be called from boards to read the mac from a ROM chip.
Signed-off-by: Olliver Schinagl oliver@schinagl.nl
drivers/net/designware.c | 16 ++++++++++++++++ 1 file changed, 16 insertions(+)
diff --git a/drivers/net/designware.c b/drivers/net/designware.c index 9e6d726..3f2f67c 100644 --- a/drivers/net/designware.c +++ b/drivers/net/designware.c @@ -230,6 +230,21 @@ static int _dw_write_hwaddr(struct dw_eth_dev *priv, u8 *mac_id) return 0; }
+__weak int dw_board_read_rom_hwaddr(unsigned char *enetaddr, int id) +{
return -ENOSYS;+}
+static int designware_eth_read_rom_hwaddr(struct udevice *dev) +{
struct eth_pdata *pdata = dev_get_platdata(dev);if (!dev)return -ENOSYS;return dw_board_read_rom_hwaddr(pdata->enetaddr, dev->seq);+}
- static void dw_adjust_link(struct eth_mac_regs *mac_p, struct phy_device *phydev) {
@@ -685,6 +700,7 @@ static const struct eth_ops designware_eth_ops = { .free_pkt = designware_eth_free_pkt, .stop = designware_eth_stop, .write_hwaddr = designware_eth_write_hwaddr,
.read_rom_hwaddr = designware_eth_read_rom_hwaddr,};
static int designware_eth_ofdata_to_platdata(struct udevice *dev)
You should not call board code from a driver. But since this is a sunxi driver, why not move all the code that reads the serial number into this file?
Hey Simon,
unless I missunderstand, this is how Joe suggested in a while ago, and how it has been implemented in a few other drivers. Can you elaborate a bit more?
Yes...drivers must not call into board-specific code, nor have board-specific #defines. This limits their usefulness for other boards.
Hmm, well as I said, I just followed Joe's suggestion with his example. also isn't this exactly how the zynq does it as well?
Sorry for misleading you. Simon has since convinced me that making a separate board-specific driver that leverages the core driver's code is a cleaner approach, and now what I recommend.
Adding hooks like this (particularly with the word 'board' in the name) should be avoided.
We end up with bidirectional coupling between the board and the driver. The board should use the driver but not the other way around. I understand that you are trying to get around this by using a weak function, but with driver model I'm really trying hard to avoid weak functions. They normally indicate an ad-hoc API and can generally be avoided with a bit more design thought.
If you have a standard way of reading the serial number which is supported by all sunxi boards, then you should be able to add your changes to the sunxi Ethernet driver (which uses designware.c?). Then you can leave the designware.c code alone and avoid adding a hook.
Well yes and no. We use designware, but also sunxi_emac, and some sdio_realtek that does not have a driver yet. But in essence, this is somewhat what I do in this patch I guess. I have the weak driver specific function in the sunxi code.
But I think I'm starting to understand your solution and will read up on the rockchip patches and rewrite this bit.
In a sense you end up subclassing the designware driver.
Also see this series which deals with a similar problem with rockchip:
http://lists.denx.de/pipermail/u-boot/2016-November/274256.html
Regards, Simon
U-Boot mailing list U-Boot@lists.denx.de http://lists.denx.de/mailman/listinfo/u-boot