Re: [PATCH net-next v4 02/15] net: dsa: vsc73xx: convert to PHYLINK

[Paweł Dembicki]

czw., 15 lut 2024 o 01:04 Vladimir Oltean <olteanv@xxxxxxxxx> napisał(a):
>
> On Wed, Feb 14, 2024 at 01:56:10PM +0100, Paweł Dembicki wrote:
> > śr., 14 lut 2024 o 00:19 Florian Fainelli <f.fainelli@xxxxxxxxx> napisał(a):
> > >
> > > On 2/13/24 14:03, Pawel Dembicki wrote:
> > > > This patch replaces the adjust_link api with the phylink apis that provide
> > > > equivalent functionality.
> > > >
> > > > The remaining functionality from the adjust_link is now covered in the
> > > > phylink_mac_link_* and phylink_mac_config.
> > > >
> > > > Removes:
> > > > .adjust_link
> > > > Adds:
> > > > .phylink_mac_config
> > > > .phylink_mac_link_up
> > > > .phylink_mac_link_down
> > >
> > > The implementation of phylink_mac_link_down() strictly mimics what had
> > > been done by adjust_link() in the phydev->link == 0 case, but it really
> > > makes me wonder whether some bits do not logically belong to
> > > phylink_mac_link_up(), like "Accept packets again" for instance.
> > >
> > > Are we certain there was not an assumption before that we would get
> > > adjust_link() called first with phydev->link = 0, and then phydev->link
> > > =1 and that this specific sequence would program things just the way we
> > > want?
> >
> > Yes, it was the simplest conversion possible, without any improvements.
> >
> > Some part is implementation of datasheet (description of ARBEMPTY register):
> >
> >         /* Discard packets */
> >         vsc73xx_update_bits(vsc, VSC73XX_BLOCK_ARBITER, 0,
> >                             VSC73XX_ARBDISC, BIT(port), BIT(port));
> >
> >         /* Wait until queue is empty */
> >         ret = read_poll_timeout(vsc73xx_read, err, err < 0 || (val & BIT(port)),
> >                                 1000, 10000, false, vsc, VSC73XX_BLOCK_ARBITER,
> >                                 0, VSC73XX_ARBEMPTY, &val);
> >         if (ret)
> >                 dev_err(vsc->dev,
> >                         "timeout waiting for block arbiter\n");
> >         else if (err < 0)
> >                 dev_err(vsc->dev, "error reading arbiter\n");
> >
> >         /* Put this port into reset */
> >         vsc73xx_write(vsc, VSC73XX_BLOCK_MAC, port, VSC73XX_MAC_CFG,
> >                       VSC73XX_MAC_CFG_RESET);
> >
> >
> > I agree that VSC73XX_ARBDISC should be moved to phylink_mac_link_up.
>
> FWIW, ocelot_phylink_mac_link_down() also calls ocelot_port_flush()
> which is more or less the same procedure for a different piece of hw.
>
> By re-reading the commit message of eb4733d7cffc ("net: dsa: felix:
> implement port flushing on .phylink_mac_link_down"), I can find a good
> reason to flush the port on link down and not on link up. With flow
> control enabled, packets would remain in the queue system until there's
> link again if not flushed there, otherwise.
>
> Paweł, maybe it is simply the case that you should move the procedure
> from the datasheet into a more clearly named sub-function?
>

I will try to do it more clearly.

> > Other things could be optimised and it needs more care. (eg. This
> > implementation doesn't disable phy when the interface goes down.) I
> > plan to tweak it after the driver becomes usable. Please let me know
> > if it should be fixed in this patch.
>
> What do you mean by disabling the PHY when the interface goes down,
> exactly? Down as in administratively down, aka "ip link set swp0 down",
> not when the link drops?
>

I should be more precise.

> That's a thing for the PHY driver to handle, by implementing .suspend()
> and .resume(), I guess?
>

Yes, exactly.

> What driver do the internal PHYs use?

It is a PHY Vitesse VSC7385 from vitesse.c.