Le Mon, 16 Jun 2025 13:57:22 +0200,
Kory Maincent <kory.maincent@xxxxxxxxxxx> a écrit :
> Le Sat, 14 Jun 2025 12:18:43 -0700,
> Jakub Kicinski <kuba@xxxxxxxxxx> a écrit :
>
> > On Tue, 10 Jun 2025 10:11:36 +0200 Kory Maincent wrote:
> > > +static struct net_device *
> > > +pse_control_find_net_by_id(struct pse_controller_dev *pcdev, int id,
> > > + netdevice_tracker *tracker)
> > > +{
> > > + struct pse_control *psec, *next;
> > > +
> > > + mutex_lock(&pse_list_mutex);
> > > + list_for_each_entry_safe(psec, next, &pcdev->pse_control_head,
> > > list) {
> >
> > nit: _safe is not necessary here, the body of the if always exits after
> > dropping the lock
>
> Indeed, I will drop it.
>
> > Do you plan to add more callers for this function?
> > Maybe it's better if it returns the psec pointer with the refcount
> > elevated. Because it would be pretty neat if we could move the
> > ethnl_pse_send_ntf(netdev, notifs, &extack); that pse_isr() does
> > right after calling this function under the rtnl_lock.
> > I don't think calling ethnl_pse_send_ntf() may crash the kernel as is,
> > but it feels like a little bit of a trap to have ethtool code called
> > outside of any networking lock.
>
> Ok. My aim was to put the less amount of code inside the rtnl lock but if you
> prefer I will call ethnl_pse_send_ntf() with the lock acquired.
psec pointer is private to pse so we will have something like the following.
Is it ok for you ?
psec = pse_control_find_by_id(pcdev, i, &tracker);
rtnl_lock();
if (psec && psec->attached_phydev &&
psec->attached_phydev->attached_dev)
ethnl_pse_send_ntf(psec->attached_phydev->attached_dev, notifs,
&extack);
rtnl_unlock();
pse_control_put(psec);
--
Köry Maincent, Bootlin
Embedded Linux and kernel engineering
https://bootlin.com
