net: axiemac: use a phandle to reference pcs_phy
PATCH 1/1 net: axiemac: use a phandle to reference pcs_phy
The of_mdio_find_device() would get a reference to mdio_device by a given device tree node. Thus, getting a reference to the internal PCS/PMA PHY by lp->phy_node is incorrect since "phy-handle" in the DT sould point to the external PHY. This incorrect use of "phy-hanlde" would cause a problem when the driver called phylink_of_phy_connect(), where it would use "phy-handle" in the DT to connect the external PHY. To fix it, we could add a phandle, "pcs-phy", in the DT so that it would point to the internal PHY. And the driver would get the correct address of PCS/PHY.
--- drivers/net/ethernet/xilinx/xilinx_axienet_main.c | 9 ++++++--- 1 file changed, 6 insertions(+), 3 deletions(-)
drivers/net/ethernet/xilinx/xilinx_axienet_main.c +6 -3
@@ -2073,12 +2073,15 @@ static int axienet_probe(struct platform_device *pdev) }if (lp->phy_mode ==PHY_INTERFACE_MODE_SGMII || lp->phy_mode ==PHY_INTERFACE_MODE_1000BASEX ) { -if (!lp->phy_node) { -dev_err (&pdev->dev,"phy-handle required for 1000BaseX/SGMII\n" ); + np =of_parse_phandle (pdev->dev.of_node,"pcs-phy" ,0 ); +if (!lp->phy_node || !np) { +dev_err (&pdev->dev,"phy-handle and pcs-phy are required for 1000BaseX/SGMII\n" ); +of_node_put (np); ret = -EINVAL ; goto cleanup_mdio; } - lp->pcs_phy =of_mdio_find_device (lp->phy_node); + lp->pcs_phy =of_mdio_find_device (np); +of_node_put (np);if (!lp->pcs_phy) { ret = -EPROBE_DEFER ; goto cleanup_mdio;
Signature · 1 line
2.34.1
Quoted 22 lines · Andy Chiu
The
of_mdio_find_device()would get a reference tomdio_deviceby a given device tree node. Thus, getting a reference to the internal PCS/PMA PHY bylp->phy_nodeis incorrect since "phy-handle" in the DT sould point to the external PHY. This incorrect use of "phy-hanlde" would cause a problem when the driver calledphylink_of_phy_connect(), where it would use "phy-handle" in the DT to connect the external PHY. To fix it, we could add a phandle, "pcs-phy", in the DT so that it would point to the internal PHY. And the driver would get the correct address of PCS/PHY.Fixes 1a02556086fc (net: axienet: Properly handle PCS/PMA PHY for 1000BaseX mode)Signed-off-by Andy Chiu <[email protected]>Reviewed-by Greentime Hu <[email protected]>--- drivers/net/ethernet/xilinx/xilinx_axienet_main.c | 9 ++++++--- 1 file changed, 6 insertions(+), 3 deletions(-)drivers/net/ethernet/xilinx/xilinx_axienet_main.c +0
-0
drivers/net/ethernet/xilinx/xilinx_axienet_main.c static int axienet_probe(struct platform_device *pdev) · line 2076
Quoted 6 lines
drivers/net/ethernet/xilinx/xilinx_axienet_main.c +1
-2@@ -2073,12 +2073,15 @@ static int axienet_probe(struct platform_device *pdev) }if (lp->phy_mode ==PHY_INTERFACE_MODE_SGMII || lp->phy_mode ==PHY_INTERFACE_MODE_1000BASEX ) { -if (!lp->phy_node) { -dev_err (&pdev->dev,"phy-handle required for 1000BaseX/SGMII\n" ); + np =of_parse_phandle (pdev->dev.of_node,"pcs-phy" ,0 );
Other drivers call this pcs-handle. Please be consistent with them.
You also should update the binding document with this new property.
Andrew
Jakub Kicinski
maintainerQuoted 1 line · On Wed, 16 Mar 2022 15:59:53 +0800 Andy Chiu wrote:
Fixes 1a02556086fc (net: axienet: Properly handle PCS/PMA PHY for 1000BaseX mode)
Please make sure to CC the maintainer of the driver. You also ate an "m" at the end of Robert's address.
Re: https://lore.kernel.org/all/[email protected]/
(looks like I was CCed with the wrong email):
I think we likely need something similar to this for the use case (I assume) you have, where there's an external SGMII PHY as well as the internal PCS/PMA PHY which both need to be configured. However, we (and possibly others) already have some cases where the core is connected to an SFP cage, phy-handle points to the internal PCS/PMA PHY, and the external PHY - if one exists at all - is not in the device tree because it would be on an SFP module.
Also, as Jakub mentioned, it looks like other drivers like dpaa2 use pcs-handle for this.
Possibly something like: in the 1000Base-X or SGMII modes, if pcs-handle is set, then use that as the PCS node, otherwise fall back to assuming phy-handle points to the PCS. It should not require that both are present, since even in the pure SGMII case, the PHY could still be supplied by an SFP module downstream.
Signature · 3 lines
Robert Hancock Senior Hardware Designer, Calian Advanced Technologies www.calian.com
Thanks for the kind help and suggestions.
I will submit the v2 patchset with an updated flow that supports both cases, and include a binding document. And use the correct mail address.
Regards,
Andy
Quoted 24 lines · Robert Hancock
Re: https://lore.kernel.org/all/[email protected]/
(looks like I was CCed with the wrong email):I think we likely need something similar to this for the use case (I assume) you have, where there's an external SGMII PHY as well as the internal PCS/PMA PHY which both need to be configured. However, we (and possibly others) already have some cases where the core is connected to an SFP cage, phy-handle points to the internal PCS/PMA PHY, and the external PHY - if one exists at all - is not in the device tree because it would be on an SFP module.
Also, as Jakub mentioned, it looks like other drivers like dpaa2 use pcs-handle for this.
Possibly something like: in the 1000Base-X or SGMII modes, if pcs-handle is set, then use that as the PCS node, otherwise fall back to assuming phy-handle points to the PCS. It should not require that both are present, since even in the pure SGMII case, the PHY could still be supplied by an SFP module downstream.
--
Robert Hancock
Senior Hardware Designer, Calian Advanced Technologies www.calian.com