Re: [PATCH] phy: phy_ethtool_ksettings_set: Don't discard phy_start_aneg's return

On 08.11.2021 15:25, Bastian Germann wrote:

Am 06.11.21 um 22:52 schrieb Heiner Kallweit:

quoted

On 05.11.2021 16:36, bage@linutronix.de wrote:

quoted

From: Bastian Germann <redacted>

Take the return of phy_start_aneg into account so that ethtool will handle
negotiation errors and not silently accept invalid input.

Fixes: 2d55173e71b0 ("phy: add generic function to support ksetting support")
Signed-off-by: Bastian Germann <redacted>
Reviewed-by: Benedikt Spranger <redacted>

In addition to what Andrew said already:

- This patch won't apply on net due to a4db9055fdb9.

Hi Heiner,

Thanks for your remarks. I would have gotten the first one right if the netdev
tree was documented in the MAINTAINERS file (T: ...) and not only in netdev-FAQ.

Maybe you want to address that.

I'd interpret the MAINTAINERS structure in a way that if no T: entry is given
the one from the next upper directory should be used. And for drivers/net/
the T: entries are available.

It's a typical case that a subsystem has a bigger number of drivers which
use the repo of the subsystem, and just few drivers have an own repo,
working with pull requests.
IMO having to add the subsystem T: entries to every driver entry would be
not needed overhead.

Not sure however which userspace tools use the T: entries and whether they are
smart enough to consider entry hierarchies.

Thanks,
Bastian

quoted

- Patch misses the "net" annotation.
- Prefix should be "net: phy:", not "phy:".

At least the formal aspects should have been covered by the internal review.

quoted

---
  drivers/net/phy/phy.c | 6 ++++--
  1 file changed, 4 insertions(+), 2 deletions(-)

diff --git a/drivers/net/phy/phy.c b/drivers/net/phy/phy.c
index a3bfb156c83d..f740b533abba 100644
--- a/drivers/net/phy/phy.c
+++ b/drivers/net/phy/phy.c

@@ -770,6 +770,8 @@ static int phy_poll_aneg_done(struct phy_device *phydev)

  int phy_ethtool_ksettings_set(struct phy_device *phydev,
                    const struct ethtool_link_ksettings *cmd)
  {
+    int ret = 0;

Why initializing ret?

quoted

+
      __ETHTOOL_DECLARE_LINK_MODE_MASK(advertising);
      u8 autoneg = cmd->base.autoneg;
      u8 duplex = cmd->base.duplex;

@@ -815,10 +817,10 @@ int phy_ethtool_ksettings_set(struct phy_device *phydev,

      phydev->mdix_ctrl = cmd->base.eth_tp_mdix_ctrl;
        /* Restart the PHY */
-    _phy_start_aneg(phydev);
+    ret = _phy_start_aneg(phydev);
        mutex_unlock(&phydev->lock);
-    return 0;
+    return ret;
  }
  EXPORT_SYMBOL(phy_ethtool_ksettings_set);