Re: [RFC PATCH 01/11] phy: core add phy_set_netif_mode() api

On 10/09/2018 12:22 AM, Kishon Vijay Abraham I wrote:

Hi Grygorii,

On Tuesday 09 October 2018 05:19 AM, Grygorii Strashko wrote:

quoted

Add new API phy_set_netif_mode(struct phy *phy, phy_interface_t mode) and
new PHY operation callback .set_netif_mode() which intended to be implemnte
by PHY drivers which supports Network interrfaces mode selection. Both
accepts phy_interface_t vlaue as input parameter.

Cc: Kishon Vijay Abraham I <redacted>
Cc: Tony Lindgren <tony@atomide.com>
Signed-off-by: Grygorii Strashko <grygorii.strashko@ti.com>
---
  drivers/phy/phy-core.c  | 15 +++++++++++++++
  include/linux/phy/phy.h | 12 ++++++++++++
  2 files changed, 27 insertions(+)

diff --git a/drivers/phy/phy-core.c b/drivers/phy/phy-core.c
index 35fd38c..d9aba1a 100644
--- a/drivers/phy/phy-core.c
+++ b/drivers/phy/phy-core.c

@@ -377,6 +377,21 @@ int phy_set_mode(struct phy *phy, enum phy_mode mode)
  }
  EXPORT_SYMBOL_GPL(phy_set_mode);
  
+int phy_set_netif_mode(struct phy *phy, phy_interface_t mode)
+{
+	int ret;
+
+	if (!phy || !phy->ops->set_netif_mode)
+		return 0;
+
+	mutex_lock(&phy->mutex);
+	ret = phy->ops->set_netif_mode(phy, mode);
+	mutex_unlock(&phy->mutex);
+
+	return ret;
+}
+EXPORT_SYMBOL_GPL(phy_set_netif_mode);

We should try to add only generic PHY APIs and not subsystem specific APIs. In
this case I think phy_set_mode should suffice.

This is what I've had in mind first, but all my guts argued against it after I've tried:

diff --git a/include/linux/phy/phy.h b/include/linux/phy/phy.h
index bc73d2b..961b156 100644
--- a/include/linux/phy/phy.h
+++ b/include/linux/phy/phy.h

@@ -41,6 +41,14 @@ enum phy_mode {
        PHY_MODE_10GKR,
        PHY_MODE_UFS_HS_A,
        PHY_MODE_UFS_HS_B,
+       PHY_MODE_MODE_MII,
+       PHY_MODE_MODE_GMII,
+       PHY_MODE_MODE_SGMII,
+       PHY_MODE_MODE_RMII,
+       PHY_MODE_MODE_RGMII,
+       PHY_MODE_MODE_RGMII_ID,
+       PHY_MODE_MODE_RGMII_RXID,
+       PHY_MODE_MODE_RGMII_TXID,
 };

above introduces ugly constants duplication and required every network phy driver
to maintain conversation table  phy_interface_t -> enum phy_mode.
More over, if above change happens third time (first time PHY_MODE_SGMII/PHY_MODE_10GKR were added,
second - PHY_MODE_2500SGMII) it will never ends (there are ~15 more items only in phy_interface_t).
As result, enum phy_mode might became a un-maintainable monster.

So, as per above, and considering that Network subsystem is based on standards (phy_interface_t lists standard intf)
I've tried to add separate PHY API.

As an idea:
- seems it could be reasonable to introduce PHY_MODE_NETWORK (or PHY_MODE_ETHERNET) and
add generic phy_set_submode(struct phy *phy, long submode). 

So, single functional PHY device can just use phy_set_submode() and
multi-functional devices (like serdes which can be muxed between PCIe, USB, NET), can use:

phy_set_mode(PHY_MODE_ETHERNET)
phy_set_submode(X);

Any way, if you still agree to just add new above phy_modes - i'll redo patches.

-- 
regards,
-grygorii