Re: [PATCH v5 08/13] usb: typec: tcpci_mt6370: Add MediaTek MT6370 tcpci driver

[Andy Shevchenko <andy.shevchenko@gmail.com>]

On Mon, Jul 18, 2022 at 10:08 AM ChiYuan Huang <u0084500@gmail.com> wrote:
> On Fri, Jul 15, 2022 at 03:10:42PM +0200, Andy Shevchenko wrote:
> > On Fri, Jul 15, 2022 at 1:28 PM ChiaEn Wu <peterwu.pub@gmail.com> wrote:

...

> > > This commit add support for the Type-C & Power Delivery controller in
> >
> > This commit add -> Add
> >
> Upper case? Or rewrite it as 'This commit is to add .....'?

Please, read this documentation [1] for better understanding. It
should clarify this and perhaps other possible questions.

[1]: https://www.kernel.org/doc/html/latest/process/submitting-patches.html#describe-your-changes

> > > MediaTek MT6370 IC.

...

> > > +       ret = devm_request_threaded_irq(dev, priv->irq, NULL,
> > > +                                       mt6370_irq_handler, IRQF_ONESHOT,
> > > +                                       dev_name(dev), priv);
> > > +       if (ret) {
> >
> > > +               tcpci_unregister_port(priv->tcpci);
> >
> > This is wrong.
> > You mixed devm_ with non-devm. Either drop devm_ *after* the first
> > non-devm_ call, or convert everything to be managed.
> >
> How about to add 'devm_add_action_or_reset' for tcpci_unregister_port?
> This will convert all as 'devm_' version.

I think it would work, that wrapper was designed to cover cases like this.

> > > +               return dev_err_probe(dev, ret, "Failed to allocate irq\n");
> > > +       }

...

> > > +static int mt6370_tcpc_remove(struct platform_device *pdev)
> > > +{
> > > +       struct mt6370_priv *priv = platform_get_drvdata(pdev);
> >
> > > +       disable_irq(priv->irq);
> >
> > Why?
> > An ugly workaround due to ordering issues in ->probe()?
> >
> Yes, due to the ordering in probe.
> 'bus remove' will be called before device resource releases.
>
> Like as you said, another way is to convert all as non-devm
> version after 'tcpci_unregister_port'.
>
> If to keep the original order, 'disable_irq' before
> 'tcpci_unregister_port' can make the flow more safe.
>
> Or you can think one case if irq triggers after
> 'tcpci_unregister_port'. Null pointer occurs.
>
> Anyway, in next revision, I'll convert all to be 'devm_' version.
> For this remove callback, only 'dev_pm_clear_wake_irq' and
> 'device_init_wakeup' will be kept.
>
> Is this better?

Sounds like a plan!

> > > +       tcpci_unregister_port(priv->tcpci);
> > > +       dev_pm_clear_wake_irq(&pdev->dev);
> > > +       device_init_wakeup(&pdev->dev, false);
> > > +
> > > +       return 0;
> > > +}

-- 
With Best Regards,
Andy Shevchenko