On Mon, Jul 08, 2024 at 12:44:43PM +0200, Krzysztof Kozlowski wrote:
> On 08/07/2024 05:16, JieGan wrote:
> >
> >>
> >>> +
> >>> + drvdata->base = devm_ioremap(dev, res->start, resource_size(res));
> >>
> >> Use proper wrapper for this two.
> > Replaced by:
> > res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
> > drvdata->base = devm_ioremap_resource(dev, res);
>
> Why?
>
> Use the wrapper.
Sorry, I misunderstood the "wrapper" before.
Just wrapped drvdata->base by void __iomem *base;
>
>
> ...
>
> >>> +
> >>> +static struct platform_driver ccu_driver = {
> >>> + .probe = ccu_probe,
> >>> + .remove = ccu_remove,
> >>> + .driver = {
> >>> + .name = "coresight-ccu",
> >>> + .of_match_table = ccu_match,
> >>> + .suppress_bind_attrs = true,
> >>
> >> Why?
> > Sorry, I dont get the point here.
>
> You do not get the point why I am asking "why?"?
>
> Why do you need it?
>
> > We dont need automatic bind/unbind, so the suppress_bind_attrs sets to true.
>
> But I need it...
>
> > We need configure some settings before we register the device.
>
> Hm, is this expected for coresight devices?
The coresight device cannot be unbinded independently. As we known,
The coresight devices collaborate to form a path, from source, link, to sink.
If an unexpected unbinding occurs for a coresight device that is part of the path,
it will disrupt the entire path.
I think it's the reason why the coresight device driver does not need automatic bind/unbind.
Here is the previous discussion for suppress_bind_attrs:
https://lore.kernel.org/all/1453753248-1716-1-git-send-email-mathieu.poirier@xxxxxxxxxx/#r
>
> Best regards,
> Krzysztof
>
Thanks,
Jie
