| Message ID | E1ifS4S-000706-ON@rmk-PC.armlinux.org.uk |
|---|---|
| State | Accepted |
| Delegated to: | David Miller |
| Headers | show |
| Series | [net-next] net: phylink: propagate phy_attach_direct() return code | expand |
On Thu, 12 Dec 2019 17:16:12 +0000, Russell King wrote: > of_phy_attach() hides the return value of phy_attach_direct(), forcing > us to return a "generic" ENODEV error code that is indistinguishable > from the lack-of-phy-property case. > > Switch to using of_phy_find_device() to find the PHY device, and then > propagating any phy_attach_direct() error back to the caller. > > Link: https://lore.kernel.org/lkml/20191210113829.GT25745@shell.armlinux.org.uk > Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk> Applied thanks, the ref counting is not entirely obvious to a layman. In your reply to Milind you said he can immediately of_node_put() because the phy_dev is never deferenced in his code, but here it looks like it is actually - the reference used to be given up after attach is done, now its given up before attach_direct is called. But I don't know how the refcounting here works, so applied, and on the off chance the code is wrong follow up will be fine. > diff --git a/drivers/net/phy/phylink.c b/drivers/net/phy/phylink.c > index f7c660bf99d1..8d20cf3ba0b7 100644 > --- a/drivers/net/phy/phylink.c > +++ b/drivers/net/phy/phylink.c > @@ -859,14 +859,17 @@ int phylink_of_phy_connect(struct phylink *pl, struct device_node *dn, > return 0; > } > > - phy_dev = of_phy_attach(pl->netdev, phy_node, flags, > - pl->link_interface); > + phy_dev = of_phy_find_device(phy_node); > /* We're done with the phy_node handle */ > of_node_put(phy_node); > - > if (!phy_dev) > return -ENODEV; > > + ret = phy_attach_direct(pl->netdev, phy_dev, flags, > + pl->link_interface); > + if (ret) > + return ret; > + > ret = phylink_bringup_phy(pl, phy_dev, pl->link_config.interface); > if (ret) > phy_detach(phy_dev);
On Sat, Dec 14, 2019 at 08:27:45PM -0800, Jakub Kicinski wrote: > On Thu, 12 Dec 2019 17:16:12 +0000, Russell King wrote: > > of_phy_attach() hides the return value of phy_attach_direct(), forcing > > us to return a "generic" ENODEV error code that is indistinguishable > > from the lack-of-phy-property case. > > > > Switch to using of_phy_find_device() to find the PHY device, and then > > propagating any phy_attach_direct() error back to the caller. > > > > Link: https://lore.kernel.org/lkml/20191210113829.GT25745@shell.armlinux.org.uk > > Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk> > > Applied thanks, the ref counting is not entirely obvious to a layman. > In your reply to Milind you said he can immediately of_node_put() > because the phy_dev is never deferenced in his code, but here it looks > like it is actually - the reference used to be given up after attach is > done, now its given up before attach_direct is called. The refcount is on the device_node - once we've got the phy device itself (or failed to) we're not using the device_node anymore, so we can put it directly after the of_phy_find_device() call. > > But I don't know how the refcounting here works, so applied, and on the > off chance the code is wrong follow up will be fine. > > > diff --git a/drivers/net/phy/phylink.c b/drivers/net/phy/phylink.c > > index f7c660bf99d1..8d20cf3ba0b7 100644 > > --- a/drivers/net/phy/phylink.c > > +++ b/drivers/net/phy/phylink.c > > @@ -859,14 +859,17 @@ int phylink_of_phy_connect(struct phylink *pl, struct device_node *dn, > > return 0; > > } > > > > - phy_dev = of_phy_attach(pl->netdev, phy_node, flags, > > - pl->link_interface); > > + phy_dev = of_phy_find_device(phy_node); > > /* We're done with the phy_node handle */ > > of_node_put(phy_node); > > - > > if (!phy_dev) > > return -ENODEV; > > > > + ret = phy_attach_direct(pl->netdev, phy_dev, flags, > > + pl->link_interface); > > + if (ret) > > + return ret; > > + > > ret = phylink_bringup_phy(pl, phy_dev, pl->link_config.interface); > > if (ret) > > phy_detach(phy_dev); > >
diff --git a/drivers/net/phy/phylink.c b/drivers/net/phy/phylink.c index f7c660bf99d1..8d20cf3ba0b7 100644 --- a/drivers/net/phy/phylink.c +++ b/drivers/net/phy/phylink.c @@ -859,14 +859,17 @@ int phylink_of_phy_connect(struct phylink *pl, struct device_node *dn, return 0; } - phy_dev = of_phy_attach(pl->netdev, phy_node, flags, - pl->link_interface); + phy_dev = of_phy_find_device(phy_node); /* We're done with the phy_node handle */ of_node_put(phy_node); - if (!phy_dev) return -ENODEV; + ret = phy_attach_direct(pl->netdev, phy_dev, flags, + pl->link_interface); + if (ret) + return ret; + ret = phylink_bringup_phy(pl, phy_dev, pl->link_config.interface); if (ret) phy_detach(phy_dev);
of_phy_attach() hides the return value of phy_attach_direct(), forcing us to return a "generic" ENODEV error code that is indistinguishable from the lack-of-phy-property case. Switch to using of_phy_find_device() to find the PHY device, and then propagating any phy_attach_direct() error back to the caller. Link: https://lore.kernel.org/lkml/20191210113829.GT25745@shell.armlinux.org.uk Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk> --- drivers/net/phy/phylink.c | 9 ++++++--- 1 file changed, 6 insertions(+), 3 deletions(-)