On Fri, 28 Oct 2022 14:18:11 +0200 Andrew Lunn wrote:
> > @@ -67,6 +67,7 @@ static void phy_link_down(struct phy_device *phydev)
> > {
> > phydev->phy_link_change(phydev, false);
> > phy_led_trigger_change_speed(phydev);
> > + WRITE_ONCE(phydev->link_down_events, phydev->link_down_events + 1);
>
> I'm not sure the WRITE_ONCE adds much value. Many systems using PHYLIB
> are 32 bit, and i don't think WRITE_ONCE will make that 64 bit write
> atomic on 32 bit systems. And as Florian pointed out, you have bigger
> problems if you manged to overflow a u32 into a u64.
>
> > @@ -723,6 +724,8 @@ struct phy_device {
> >
> > int pma_extable;
> >
> > + unsigned int link_down_events;
>
> And here is unsigned int, not u64? Or u32? It would be good to be
> consistent.
I made it 32b on the phylib side so that WRITE_ONCE() was sufficient
to ensure atomic writes.
Do you have a preference for it being 64b vs 32b at the uAPI level?
I was leaning slightly towards making both 32b in v3..
> > --- a/include/uapi/linux/ethtool_netlink.h
> > +++ b/include/uapi/linux/ethtool_netlink.h
> > @@ -262,6 +262,8 @@ enum {
> > ETHTOOL_A_LINKSTATE_SQI_MAX, /* u32 */
> > ETHTOOL_A_LINKSTATE_EXT_STATE, /* u8 */
> > ETHTOOL_A_LINKSTATE_EXT_SUBSTATE, /* u8 */
> > + ETHTOOL_A_LINKSTATE_PAD,
> > + ETHTOOL_A_LINKSTATE_EXT_DOWN_CNT, /* u64 */
>
> What is the PAD for?
64b values have to be padded in netlink to ensure alignment.
