[PATCH v4 14/16] net: stmmac: dwmac-s32: add basic NXP S32G/S32R glue driver

[Jan Petrous]

On Fri, Nov 01, 2024 at 04:40:50PM +0100, Krzysztof Kozlowski wrote:
> On 31/10/2024 18:24, Jan Petrous wrote:
> > On Thu, Oct 31, 2024 at 06:16:46PM +0100, Jan Petrous wrote:
> >> On Thu, Oct 31, 2024 at 04:44:45PM +0100, Krzysztof Kozlowski wrote:
> >>> On 31/10/2024 15:43, Jan Petrous wrote:
> >>>> On Tue, Oct 29, 2024 at 08:13:40AM +0100, Krzysztof Kozlowski wrote:
> >>>>> On Mon, Oct 28, 2024 at 09:24:56PM +0100, Jan Petrous (OSS) wrote:
> >>>>>> +	plat->init = s32_gmac_init;
> >>>>>> +	plat->exit = s32_gmac_exit;
> >>>>>> +	plat->fix_mac_speed = s32_fix_mac_speed;
> >>>>>> +
> >>>>>> +	plat->bsp_priv = gmac;
> >>>>>> +
> >>>>>> +	return stmmac_pltfr_probe(pdev, plat, &res);
> >>>>>> +}
> >>>>>> +
> >>>>>> +static const struct of_device_id s32_dwmac_match[] = {
> >>>>>> +	{ .compatible = "nxp,s32g2-dwmac" },
> >>>>>> +	{ .compatible = "nxp,s32g3-dwmac" },
> >>>>>> +	{ .compatible = "nxp,s32r-dwmac" },
> >>>>>
> >>>>> Why do you need three same entries?
> >>>>>
> >>>>
> >>>> We have three different SoCs and in v3 review you told me
> >>>> to return all back:
> >>>> https://patchwork.kernel.org/comment/26067257/
> >>>
> >>> It was about binding, not driver.
> >>>
> >>> I also asked there: use proper fallback and compatibility. Both comments
> >>> of course affect your driver, but why choosing only first part?
> >>>
> >>
> >> Does it mean I should remove first two (G2/G3) members from match array
> >> and use "nxp,s32r-dwmac" as fallback for G2/G3? And similarly change
> >> the bindings to:
> >>
> >>   compatible:
> >>     oneOf:
> >>       - const: nxp,s32r-dwmac
> >>       - items:
> >> 	  - enum:
> >> 	      - nxp,s32g2-dwmac
> >> 	      - nxp,s32g3-dwmac
> >>           - const: nxp,s32r-dwmac
> >>
> >> And add here, into the driver, those members back when some device
> >> specific feature will be needed? Am I understand your hints right?
> >>
> > 
> > Sorry, it's not correct. This way I'm not able to detect S32R which is
> > the only one with higher speed.
> > 
> > Then I could use the G2 as fallback I think, Ie.:
> > 
> >   compatible:
> >     oneOf:
> >       - const: nxp,s32g2-dwmac
> >       - items:
> > 	  - enum:
> >               - nxp,s32g3-dwmac
> >               - nxp,s32r-dwmac
> >            - const: nxp,s32g2-dwmac
> 
> I don't understand. In both cases you can 'detect r', if by this you
> meant match and bind. I don't care which one is the fallback, but if one
> does not work it points to different issues with your code.
> 

I see. I will use the last variant (with s32g2 as fallback) in v5.

BR.
/Jan