[PATCH 02/29] pinctrl: mvebu: new driver for Orion platforms

[Thomas Petazzoni]

Sebastian,

Thanks for your review!

On Mon, 14 Apr 2014 11:15:17 +0200, Sebastian Hesselbarth wrote:

> > +* Marvell Orion 88f5182
> > +
> > +name          pins     functions
> > +================================================================================
> > +mpp0          0        pci(rstout), pci(req2), gpio
> 
> Shouldn't the above pcie(rstout)?
> 
> The one available datasheet I have for 5182 names the function
> PEX_RST_OUTn, which is PCIexpress then.

Indeed, fixed.

> > +mpp1          1        gpio, pci(gnt2)
> > +mpp2          2        gpio, pci(req3), pci(pme)
> 
> pinctrl-mvebu isn't that good with identically named functions.
> You need to rename one of the "pci" functions to e.g. "pci-1".

Done.

> > +mpp3          3        gpio, pci(gnt3)
> > +mpp4          4        gpio, pci(req4), bootnand(re), sata0(prsnt)
> > +mpp5          5        gpio, pci(gnt4), bootnand(we), sata1(prsnt)
> > +mpp6          6        gpio, pci(req5), nand(re0), sata0(act)
> > +mpp7          7        gpio, pci(gnt5), nand(we0), sata1(act)
> > +mpp8          8        gpio, ge(col)
> > +mpp9          9        gpio, ge(rxerr)
> > +mpp10         10       gpio, ge(crs)
> > +mpp11         11       gpio, ge(txerr)
> > +mpp12         12       gpio, ge(txd4), nand(re1), sata0(ledprsnt)
> > +mpp13         13       gpio, ge(txd5), nand(we1), sata1(ledprsnt)
> > +mpp14         14       gpio, ge(txd6), nand(re2), sata0(ledact)
> > +mpp15         15       gpio, ge(txd7), nand(we2), sata1(ledact)
> 
> Four "led" prefixes above should be removed.

I don't agree, because there would then be no difference between
sata0(act) and sata0(ledact), even if in the datasheet, their
description is different:

 * SATA 0 active indication (for which I've used "sata0(act)")
 * SATA 0 presence LED indication (Active Low) (for which I've used
   "sata0(ledact)")

Do you have another suggestion?


> > +static struct mvebu_mpp_ctrl orion_mpp_controls[] = {
> > +	MPP_FUNC_CTRL(0, 19, NULL, orion_mpp_ctrl),
> > +};
> > +
> > +static struct pinctrl_gpio_range mv88f5181l_gpio_ranges[] = {
> > +	MPP_GPIO_RANGE(0, 0, 0, 16),
> > +};
> > +
> > +static struct pinctrl_gpio_range mv88f5182_gpio_ranges[] = {
> > +	MPP_GPIO_RANGE(0, 0, 0, 19),
> > +};
> > +
> > +static struct pinctrl_gpio_range mv88f5281_gpio_ranges[] = {
> > +	MPP_GPIO_RANGE(0, 0, 0, 16),
> > +};
> 
> mv88f5181l_gpio_ranges == mv88f5281_gpio_ranges.
> 
> You can possibly join them to mv88f5x81_gpio ranges, but I have
> no strong opinion about it.

I prefer to have them all for each SoC, for consistency.


> > +static int orion_pinctrl_probe(struct platform_device *pdev)
> > +{
> > +	const struct of_device_id *match =
> > +		of_match_device(orion_pinctrl_of_match, &pdev->dev);
> > +	struct resource *res;
> > +
> > +	pdev->dev.platform_data = (void*) match->data;
> 
> Useless (void *) cast?

No: there is the same cast in pinctrl-dove.c and pinctrl-kirkwood.c.
The reason is a bit ugly: match->data is "const" but
pdev->dev.platform_data is not. Certainly something to fix at some
point, but probably not as part of this patch series, since Dove and
Kirkwood are already doing the bad thing :)

Thomas
-- 
Thomas Petazzoni, CTO, Free Electrons
Embedded Linux, Kernel and Android engineering
http://free-electrons.com