[PATCH net-next 1/2] net: stmmac: Add PCS_LYNX as a dependency for the whole driver

[Geert Uytterhoeven]

Hi Russell,

On Tue, Jun 6, 2023 at 12:26 PM Russell King (Oracle)
<linux at armlinux.org.uk> wrote:
> On Tue, Jun 06, 2023 at 12:13:11PM +0200, Maxime Chevallier wrote:
> > On Tue, 6 Jun 2023 10:30:32 +0100
> > "Russell King (Oracle)" <linux at armlinux.org.uk> wrote:
> > > On Tue, Jun 06, 2023 at 10:29:20AM +0200, Geert Uytterhoeven wrote:
> > > > On Tue, 6 Jun 2023, Maxime Chevallier wrote:
> > > > > Although pcs_lynx is only used on dwmac_socfpga for now, the cleanup
> > > > > path is in the generic driver, and triggers build issues for other
> > > > > stmmac variants. Make sure we build pcs_lynx in all cases too, as for
> > > > > XPCS.
> > > >
> > > > That seems suboptimal to me, as it needlesly increases kernel size for
> > > > people who do not use dwmac_socfpga.  Hence I made an alternative patch:
> > > > https://lore.kernel/org/7b36ac43778b41831debd5c30b5b37d268512195.1686039915.git.geert+renesas@glider.be
> > >
> > > A better solution would be to re-architect the removal code so that
> > > whatever creates the PCS is also responsible for removing it.
> > >
> > > Also, dwmac_socfpga nees to be reorganised anyway, because it calls
> > > stmmac_dvr_probe() which then goes on to call register_netdev(),
> > > publishing the network device, and then after stmmac_dvr_probe(),
> > > further device setup is done. As the basic driver probe flow should
> > > be setup and then publish, the existing code structure violates that.
> > >
> >
> > I agree that this solution is definitely suboptimal, I wanted mostly to get it
> > fixed quickly as this breaks other stmmac variants.
> >
> > Do we still go on with the current patch (as Geert's has issues) and then
> > consider reworking dwmac_socfpga ?
>
> As Geert himself mentioned, passed on from Arnd:
>   As pointed out by Arnd, this doesn't work when PCS_LYNX is a loadable
>   module and STMMAC is built-in:
>   https://lore.kernel.org/r/11bd37e9-c62e-46ba-9456-8e3b353df28f@app.fastmail.com
>
> So Geert's solution will just get rid of the build error, but leave the
> Lynx PCS undestroyed. I take Geert's comment as a self-nack on his
> proposed patch.

Exactly...

> The changes are only in net-next at the moment, and we're at -rc5.
> There's probably about 2.5 weeks to get this sorted before the merge
> window opens.
>
> So, we currently have your suggestion. Here's mine as an immediate
> fix. This doesn't address all the points I've raised, but should
> resolve the immediate issue.
>
> Untested since I don't have the hardware... (the test build is
> running):
>
> diff --git a/drivers/net/ethernet/stmicro/stmmac/dwmac-socfpga.c b/drivers/net/ethernet/stmicro/stmmac/dwmac-socfpga.c
> index e399fccbafe5..239c7e9ed41d 100644
> --- a/drivers/net/ethernet/stmicro/stmmac/dwmac-socfpga.c
> +++ b/drivers/net/ethernet/stmicro/stmmac/dwmac-socfpga.c
> @@ -494,6 +494,17 @@ static int socfpga_dwmac_probe(struct platform_device *pdev)
>         return ret;
>  }
>
> +static void socfpga_dwmac_remove(struct platform_device *pdev)
> +{
> +       struct net_device *ndev = platform_get_drvdata(pdev);
> +       struct stmmac_priv *priv = netdev_priv(ndev);
> +       struct phylink_pcs *pcs = priv->hw->lynx_pcs;
> +
> +       stmmac_pltfr_remove(pdev);
> +
> +       lynx_pcs_destroy(pcs);
> +}
> +
>  #ifdef CONFIG_PM_SLEEP
>  static int socfpga_dwmac_resume(struct device *dev)
>  {
> @@ -565,7 +576,7 @@ MODULE_DEVICE_TABLE(of, socfpga_dwmac_match);
>
>  static struct platform_driver socfpga_dwmac_driver = {
>         .probe  = socfpga_dwmac_probe,
> -       .remove_new = stmmac_pltfr_remove,
> +       .remove_new = socfpga_dwmac_remove,
>         .driver = {
>                 .name           = "socfpga-dwmac",
>                 .pm             = &socfpga_dwmac_pm_ops,
> diff --git a/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c b/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
> index fa07b0d50b46..1801f8cc8413 100644
> --- a/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
> +++ b/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
> @@ -940,9 +940,6 @@ static struct phylink_pcs *stmmac_mac_select_pcs(struct phylink_config *config,
>         if (priv->hw->xpcs)
>                 return &priv->hw->xpcs->pcs;
>
> -       if (priv->hw->lynx_pcs)
> -               return priv->hw->lynx_pcs;
> -
>         return NULL;
>  }

I think the above hunk is wrong, and should be replaced by a removal
of the call to lynx_pcs_destroy()?

Gr{oetje,eeting}s,

                        Geert

-- 
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert at linux-m68k.org

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
                                -- Linus Torvalds