On Tue, 11 Apr 2023, Herve Codina wrote:
> Hi Lee,
>
> On Wed, 5 Apr 2023 18:00:23 +0200
> Herve Codina <herve.codina@xxxxxxxxxxx> wrote:
>
> > Hi Lee,
> > On Wed, 5 Apr 2023 14:54:50 +0100
> > Lee Jones <lee@xxxxxxxxxx> wrote:
> >
> > > On Tue, 04 Apr 2023, Krzysztof Kozlowski wrote:
> > >
> > > > On 04/04/2023 10:07, Herve Codina wrote:
> > > >
> > > > >>> So, the structure I have in mind:
> > > > >>> - pef2256.c (MFD)
> > > > >>> implement and do the setup at probe()
> > > > >>> Add the children at probe():
> > > > >>> - pef2256-pinctrl (pinctrl) added using mfd_add_devices()
> > > > >>> - pef2256-codec (ASoC codec) added using devm_of_platform_populate()
> > > > >>>
> > > > >>> Lee, with this in mind, can the core pef2256.c be a MFD driver ?
> > > > >>
> > > > >> You do not use MFD here, so why do you want to keep it in MFD? If you
> > > > >> disagree, please tell me where is the MFD code in your patch?
> > > > >
> > > > > I don't want to absolutely use MFD.
> > > > > I just want to put my driver somewhere and I don't know the right location
> > > > > between MFD and Misc.
> > > > >
> > > > > Basically, the driver needs to do (little simplified and error path removed):
> > > > >
> > > > > static const struct mfd_cell pef2256_devs[] = {
> > > > > { .name = "lantiq-pef2256-pinctrl", },
> > > > > };
> > > > >
> > > > > static int pef2256_probe(struct platform_device *pdev)
> > > > > {
> > > > > struct pef2256 *pef2256;
> > > > > void __iomem *iomem;
> > > > > int ret;
> > > > > int irq;
> > > > >
> > > > > pef2256 = devm_kzalloc(&pdev->dev, sizeof(*pef2256), GFP_KERNEL);
> > > > > if (!pef2256)
> > > > > return -ENOMEM;
> > > > >
> > > > > pef2256->dev = &pdev->dev;
> > > > >
> > > > > iomem = devm_platform_ioremap_resource(pdev, 0);
> > > > >
> > > > > pef2256->regmap = devm_regmap_init_mmio(&pdev->dev, iomem,
> > > > > &pef2256_regmap_config);
> > > > >
> > > > > pef2256->mclk = devm_clk_get_enabled(&pdev->dev, "mclk");
> > > > > pef2256->sclkr = devm_clk_get_enabled(&pdev->dev, "sclkr");
> > > > > pef2256->sclkx = devm_clk_get_enabled(&pdev->dev, "sclkx");
> > > > >
> > > > > pef2256->reset_gpio = devm_gpiod_get_optional(&pdev->dev, "reset", GPIOD_OUT_LOW);
> > > > > if (pef2256->reset_gpio) {
> > > > > gpiod_set_value_cansleep(pef2256->reset_gpio, 1);
> > > > > udelay(10);
> > > > > gpiod_set_value_cansleep(pef2256->reset_gpio, 0);
> > > > > udelay(10);
> > > > > }
> > > > >
> > > > > pef2556_of_parse(pef2256, np);
> > > > >
> > > > > irq = platform_get_irq(pdev, 0);
> > > > > ret = devm_request_irq(pef2256->dev, irq, pef2256_irq_handler, 0, "pef2256", pef2256);
> > > > >
> > > > > platform_set_drvdata(pdev, pef2256);
> > > > >
> > > > > mfd_add_devices(pef2256->dev, PLATFORM_DEVID_NONE, pef2256_devs,
> > > > > ARRAY_SIZE(pef2256_devs), NULL, 0, NULL);
> > > >
> > > > Wait, now you use MFD framework, so the driver is suitable for MFD.
> > > > Before there was nothing like that in your code.
> > >
> > > Agree, the above is suitable for MFD, since it does all the things I
> > > said your last one did not. You *can* also use of_platform_populate()
> > > here, since you are *also* requesting and initialising shared resources.
> > > You cannot do *both* however.
> > >
> >
> > Thanks for having confirmed that this driver can be a MFD driver.
> >
> > Related to of_platform_populate() / mfd_add_devices(), I wanted to use both
> > because:
> > - the pinctrl part does not have (and does not need to have) a specific node
> > with a specific compatible property. In order to instantiate the related
> > driver mfd_add_devices() is the only way I know.
> > - the audio component nodes have a compatible string and several components
> > (ie several nodes) can be present. of_platform_populate() call seems the
> > simple way to instantiate them.
> >
> > Is there a way to use mfd_add_devices() in this case without the need to
> > count the audio component nodes in order to allocate as much mfd_cell as
> > nodes having a matching compatible property ? Is there an other API to do
> > it ?
> >
>
> I looked deeper for using mfd_add_devices() only and found an issue (related
> to my use-case).
>
> mfd_add_devices() calls mfd_add_device() and in this function we have:
> ---- 8< ----
> if (IS_ENABLED(CONFIG_OF) && parent->of_node && cell->of_compatible) {
> for_each_child_of_node(parent->of_node, np) {
> if (of_device_is_compatible(np, cell->of_compatible)) {
> /* Ignore 'disabled' devices error free */
> if (!of_device_is_available(np)) {
> of_node_put(np);
> ret = 0;
> goto fail_alias;
> }
>
> ret = mfd_match_of_node_to_dev(pdev, np, cell);
> if (ret == -EAGAIN)
> continue;
> of_node_put(np);
> if (ret)
> goto fail_alias;
>
> break;
> }
> }
>
> if (!pdev->dev.of_node)
> pr_warn("%s: Failed to locate of_node [id: %d]\n",
> cell->name, platform_id);
> }
> ---- 8< ----
>
> My issue is related to the loop.
> The loop is aborted if the device is not available (typically 'Disabled' in the
> DT node).
>
> My full pef2256 node can have several pef2256 audio subnodes with compatible
> set to "lantiq,pef2256-codec" and some of them can have 'status = "Disabled";'
> As soon as one device (with my expected compatible) is found 'Disabled', there
> is no way to look at other children.
>
> Having 'continue' in case of device not available in the loop to look at the
> next child will change the behavior of mfd_add_device().
> Do you think that looking at next child if we find a 'not available' child
> can be correct for mfd_add_device() ?
> This probably will have impacts on other MFD drivers.
Looks like a bug. Encountering one disabled cell should not prevent the
others from registering IMHO.
--
Lee Jones [李琼斯]
