Re: [PATCH v4 07/10] clk: starfive: Add StarFive JH7110 Video-Output clock driver

[Conor Dooley]

On Wed, Apr 12, 2023 at 09:04:08PM -0700, Stephen Boyd wrote:
> Quoting Xingyu Wu (2023-04-11 23:15:26)
> > On 2023/4/12 2:33, Stephen Boyd wrote:
> > > Quoting Xingyu Wu (2023-04-11 06:55:55)

> > >> +       if (ret < 0)
> > >> +               return dev_err_probe(priv->dev, ret, "failed to turn on power\n");
> > >> +
> > >> +       ret = jh7110_vout_top_crg_init(priv, top);
> > >> +       if (ret)
> > >> +               goto err_clk;
> > >> +
> > >> +       top->base = priv->base;
> > >> +       dev_set_drvdata(priv->dev, (void *)(&top->base));
> > > 
> > > See comment later about setting this to 'top' instead. Casting away
> > > iomem markings is not good hygiene.
> > 
> > JH7110 resets as the auxiliary device of clocks use the same iomem as the clocks
> > and the iomem will be got by dev_get_drvdata() in the 7110 reset drivers when registering reset.
> > So I follow the basic 7110 reset driver and also set the iomem not top_crg struct.
> 
> Oh I totally missed that this is how it's been done for the other
> starfive driver. It's still not good hygiene to stash the iomem pointer
> that way because the iomem marking is lost and has to be recovered. Can
> you make a wrapper struct, either for the adev or to pass in struct
> device::platform_data?

FWIW, this did come up on an earlier version of the other starfive
driver:
https://lore.kernel.org/linux-clk/e0d8f9ba-5bf4-d7dd-5110-20d4196556f9@xxxxxxxxxxxxxxxx/

I probably should've pushed Hal to use a struct, but evidently I didn't
reply to his final message there, so apologies for that!

Attachment: signature.asc
Description: PGP signature