Re: [PATCH net-next v2 4/5] net: ravb: Do not apply RX checksum settings to hardware if the interface is down

[Biju Das]

Hi Claudiu,

On Sun, Feb 11, 2024 at 12:13 PM Biju Das <biju.das.au@xxxxxxxxx> wrote:
>
> Hi Sergey,
>
> On Sun, Feb 11, 2024 at 9:40 AM Sergei Shtylyov
> <sergei.shtylyov@xxxxxxxxx> wrote:
> >
> > On 2/11/24 11:56 AM, Biju Das wrote:
> >
> > >>>> From: Claudiu Beznea <claudiu.beznea.uj@xxxxxxxxxxxxxx>
> > >>>>
> > >>>> Do not apply the RX checksum settings to hardware if the interface is
> > >>>> down.

Gb eth supports both Rx/Tx Checksum

The intention is not to apply any hardware feature while the interface is done.
So please add a generic commit header and description.

Cheers,
Biju

> > >>>> In case runtime PM is enabled, and while the interface is down, the IP
> > >>>> will be in reset mode (as for some platforms disabling the clocks will
> > >>>> switch the IP to reset mode, which will lead to losing register
> > >>>> contents) and applying settings in reset mode is not an option.
> > >>>> Instead, cache the RX checksum settings and apply them in ravb_open()
> > >>>> through ravb_emac_init().
> > >>>> This has been solved by introducing pm_runtime_active() check. The
> > >>>> device runtime PM usage counter has been incremented to avoid
> > >>>> disabling the device clocks while the check is in progress (if any).
> > >>>>
> > >>>> Commit prepares for the addition of runtime PM.
> > >>>>
> > >>>> Signed-off-by: Claudiu Beznea <claudiu.beznea.uj@xxxxxxxxxxxxxx>
> > >>>
> > >>> Reviewed-by: Sergey Shtylyov <s.shtylyov@xxxxxx>
> > >>
> > >> This will do the same job, without code duplication right?
> > >>
> > >>> static int ravb_set_features(struct net_device *ndev,
> > >>>     netdev_features_t features)
> > >>> {
> > >>> struct ravb_private *priv = netdev_priv(ndev);
> > >>> struct device *dev = &priv->pdev->dev;
> > >>> const struct ravb_hw_info *info = priv->info;
> > >>>
> > >>> pm_runtime_get_noresume(dev);
> > >>> if (!pm_runtime_active(dev)) {
> > >>> pm_runtime_put_noidle(dev);
> > >>> ndev->features = features;
> > >>> return 0;
> > >>> }
> > >>>
> > >>> return info->set_feature(ndev, features);
> > >
> > >> We now leak the device reference by not calling pm_runtime_put_noidle()
> > >> after this statement...
> > >
> > > Oops. So this leak  can be fixed like [1]
> > >
> > >>  The approach seems sane though -- Claudiu, please consider following it.
> > >
> > > [1]
> > > static int ravb_set_features(struct net_device *ndev,
> > >     netdev_features_t features)
> > > {
> > > struct ravb_private *priv = netdev_priv(ndev);
> > > const struct ravb_hw_info *info = priv->info;
> > > struct device *dev = &priv->pdev->dev;
> > > bool pm_active;
> > >
> > > pm_runtime_get_noresume(dev);
> > > pm_active = pm_runtime_active(dev);
> > > pm_runtime_put_noidle(dev);
> >
> >    There is no point dropping the RPM reference before we access
> > the regs...
>
> I don't think there is an issue in accessing register by the usage of
> below API's
>
> pm_runtime_get_noresume:--- Bump up runtime PM usage counter of a device.
> pm_runtime_active:--- Check whether or not a device is runtime-active.
> pm_runtime_put_noidle:--Drop runtime PM usage counter of a device.
>
> Cheers,
> Biju