On Tue, 16 Jan 2024 19:58:56 +0300 Maxim Kiselev bigunclemax@gmail.com wrote:
Hi Maxim,
вт, 16 янв. 2024 г. в 03:18, Andre Przywara andre.przywara@arm.com:
On Thu, 15 Jun 2023 00:44:06 +0300 Maxim Kiselev bigunclemax@gmail.com wrote:
Hi Maxim,
From: Maksim Kiselev bigunclemax@gmail.com
Based on dt-specs fixed-link doesn't require phy-handle to be used.
Do you have such a board?
Yes, I had a custom board with T113 SoC which has fixed phy eth.
So just to clarify: this board connected something like a switch directly to the MAC, just using RMII/RGMII? And then you have a fixed-phy node in the DT, detailing the speed and duplex parameters? Matching the fixed-phy node description in ethernet-controller.yaml? And drivers/net/phy/fixed.c takes care of parsing that?
And where is that written down? I don't see it explicitly mentioned as optional in ethernet-controller.yaml or in the DT spec PDF.
Sorry for that commit description. I just found the similar commit, that fixes the same problem for zynq_gem 3888c8d1979289efe685fe29276aed4d4b685975 ("net: zynq_gem: Add support for fixed-link phy") and copied the description from there.
The sun8i EMAC binding lists phy-handle as required, so that would need to be relaxed there then.
Oh, indeed. So it will require to send dt-binding changes to Linux first...
Therefore, I think we can live without fixed-phy support for sun8i EMAC. At least until some device with such a configuration appears on the market :)
Well, I am fine with that patch if it fixes a problem for you. It looks like requiring the phy-handle property is too strict, and just needs to be relaxed in the binding. If I see this correctly, the driver would still accept every valid DT, by today's and future bindings?
If you could just confirm my above assumptions, and maybe send a v2 with an amended commit message, I would be happy to merge that patch.
Cheers, Andre
Fix driver to only read phy related setting when phy-handle is found.
The patch itself looks fine, we already specify -1 as the default when the PHY DT node does not contain a reg property, so that looks like it would work.
Cheers, Andre
Signed-off-by: Maksim Kiselev bigunclemax@gmail.com
drivers/net/sun8i_emac.c | 7 ++----- 1 file changed, 2 insertions(+), 5 deletions(-)
diff --git a/drivers/net/sun8i_emac.c b/drivers/net/sun8i_emac.c index 04c3274fbe..0e339d69e0 100644 --- a/drivers/net/sun8i_emac.c +++ b/drivers/net/sun8i_emac.c @@ -834,11 +834,8 @@ static int sun8i_emac_eth_of_to_plat(struct udevice *dev) priv->use_internal_phy = false;
offset = fdtdec_lookup_phandle(gd->fdt_blob, node, "phy-handle");
if (offset < 0) {debug("%s: Cannot find PHY address\n", __func__);return -EINVAL;}priv->phyaddr = fdtdec_get_int(gd->fdt_blob, offset, "reg", -1);
if (offset >= 0)priv->phyaddr = fdtdec_get_int(gd->fdt_blob, offset, "reg", -1); pdata->phy_interface = dev_read_phy_mode(dev); debug("phy interface %d\n", pdata->phy_interface);Best wishes, Maksim