| Message ID | 20240214-mbly-gpio-v1-19-f88c0ccf372b@bootlin.com (mailing list archive) |
|---|---|
| State | Superseded |
| Headers | show |
| Series | Rework Nomadik GPIO to add Mobileye EyeQ5 support | expand |
On Mi, 2024-02-14 at 17:24 +0100, Théo Lebrun wrote: > Fetch a reference to the optional shared reset control and deassert it > if it exists. > > Optional because not all platforms that use this driver have a reset > attached to the reset block. Shared because some platforms that use the > reset (at least Mobileye EyeQ5) share the reset across banks. > > Do not keep a reference to the reset control as it is not needed > afterwards; the driver does not handle suspend, does not use runtime PM > and does not register a remove callback. I suppose you don't care that the reset is only ever deasserted once and never asserted again on this hardware, but for shared reset controls the expectation is that deassert/assert calls are balanced: https://docs.kernel.org/driver-api/reset.html?highlight=balanced#assertion-and-deassertion So maybe this warrants a comment in the code. Or do you mean to suppress unbind via suppress_bind_attrs to explain away any missing cleanup? > The operation is done in nmk_gpio_populate_chip(). This function is > called by either gpio-nomadik or pinctrl-nomadik, whoever comes first. > This is here for historic reasons and could probably be removed now; it > seems gpio-ranges enforces the ordering to be pinctrl-first. It is not > the topic of the present patch however. > > Signed-off-by: Théo Lebrun <theo.lebrun@bootlin.com> > --- > drivers/gpio/gpio-nomadik.c | 15 +++++++++++++++ > 1 file changed, 15 insertions(+) > > diff --git a/drivers/gpio/gpio-nomadik.c b/drivers/gpio/gpio-nomadik.c > index 21bb6d6363fc..b623c093b54d 100644 > --- a/drivers/gpio/gpio-nomadik.c > +++ b/drivers/gpio/gpio-nomadik.c > @@ -513,12 +513,14 @@ struct nmk_gpio_chip *nmk_gpio_populate_chip(struct device_node *np, > { > struct nmk_gpio_chip *nmk_chip; > struct platform_device *gpio_pdev; > + struct reset_control *reset; > struct gpio_chip *chip; > struct resource *res; > struct clk *clk; > void __iomem *base; > uintptr_t flags; > u32 id, ngpio; > + int ret; > > gpio_pdev = of_find_device_by_node(np); > if (!gpio_pdev) { > @@ -576,6 +578,19 @@ struct nmk_gpio_chip *nmk_gpio_populate_chip(struct device_node *np, > clk_prepare(clk); > nmk_chip->clk = clk; > > + reset = devm_reset_control_get_optional_shared(&gpio_pdev->dev, NULL); > + if (IS_ERR(reset)) { > + dev_err(&pdev->dev, "failed getting reset control: %ld\n", > + PTR_ERR(reset)); > + return ERR_CAST(reset); Consider using dev_err_probe() here. regards Philipp
Hello, On Thu Feb 15, 2024 at 11:19 AM CET, Philipp Zabel wrote: > On Mi, 2024-02-14 at 17:24 +0100, Théo Lebrun wrote: > > Fetch a reference to the optional shared reset control and deassert it > > if it exists. > > > > Optional because not all platforms that use this driver have a reset > > attached to the reset block. Shared because some platforms that use the > > reset (at least Mobileye EyeQ5) share the reset across banks. > > > > Do not keep a reference to the reset control as it is not needed > > afterwards; the driver does not handle suspend, does not use runtime PM > > and does not register a remove callback. > > I suppose you don't care that the reset is only ever deasserted once > and never asserted again on this hardware, but for shared reset > controls the expectation is that deassert/assert calls are balanced: > > https://docs.kernel.org/driver-api/reset.html?highlight=balanced#assertion-and-deassertion > > So maybe this warrants a comment in the code. Or do you mean to > suppress unbind via suppress_bind_attrs to explain away any missing > cleanup? Those resets are shared only across GPIO banks which have no reason to be unbind-able. I'll add a comment and disable unbind. > > The operation is done in nmk_gpio_populate_chip(). This function is > > called by either gpio-nomadik or pinctrl-nomadik, whoever comes first. > > This is here for historic reasons and could probably be removed now; it > > seems gpio-ranges enforces the ordering to be pinctrl-first. It is not > > the topic of the present patch however. > > > > Signed-off-by: Théo Lebrun <theo.lebrun@bootlin.com> > > --- > > drivers/gpio/gpio-nomadik.c | 15 +++++++++++++++ > > 1 file changed, 15 insertions(+) > > > > diff --git a/drivers/gpio/gpio-nomadik.c b/drivers/gpio/gpio-nomadik.c > > index 21bb6d6363fc..b623c093b54d 100644 > > --- a/drivers/gpio/gpio-nomadik.c > > +++ b/drivers/gpio/gpio-nomadik.c > > @@ -513,12 +513,14 @@ struct nmk_gpio_chip *nmk_gpio_populate_chip(struct device_node *np, > > { > > struct nmk_gpio_chip *nmk_chip; > > struct platform_device *gpio_pdev; > > + struct reset_control *reset; > > struct gpio_chip *chip; > > struct resource *res; > > struct clk *clk; > > void __iomem *base; > > uintptr_t flags; > > u32 id, ngpio; > > + int ret; > > > > gpio_pdev = of_find_device_by_node(np); > > if (!gpio_pdev) { > > @@ -576,6 +578,19 @@ struct nmk_gpio_chip *nmk_gpio_populate_chip(struct device_node *np, > > clk_prepare(clk); > > nmk_chip->clk = clk; > > > > + reset = devm_reset_control_get_optional_shared(&gpio_pdev->dev, NULL); > > + if (IS_ERR(reset)) { > > + dev_err(&pdev->dev, "failed getting reset control: %ld\n", > > + PTR_ERR(reset)); > > + return ERR_CAST(reset); > > Consider using dev_err_probe() here. Makes sense, will do so for next revision. Thanks for the review Philipp, -- Théo Lebrun, Bootlin Embedded Linux and Kernel engineering https://bootlin.com
diff --git a/drivers/gpio/gpio-nomadik.c b/drivers/gpio/gpio-nomadik.c index 21bb6d6363fc..b623c093b54d 100644 --- a/drivers/gpio/gpio-nomadik.c +++ b/drivers/gpio/gpio-nomadik.c @@ -513,12 +513,14 @@ struct nmk_gpio_chip *nmk_gpio_populate_chip(struct device_node *np, { struct nmk_gpio_chip *nmk_chip; struct platform_device *gpio_pdev; + struct reset_control *reset; struct gpio_chip *chip; struct resource *res; struct clk *clk; void __iomem *base; uintptr_t flags; u32 id, ngpio; + int ret; gpio_pdev = of_find_device_by_node(np); if (!gpio_pdev) { @@ -576,6 +578,19 @@ struct nmk_gpio_chip *nmk_gpio_populate_chip(struct device_node *np, clk_prepare(clk); nmk_chip->clk = clk; + reset = devm_reset_control_get_optional_shared(&gpio_pdev->dev, NULL); + if (IS_ERR(reset)) { + dev_err(&pdev->dev, "failed getting reset control: %ld\n", + PTR_ERR(reset)); + return ERR_CAST(reset); + } + + ret = reset_control_deassert(reset); + if (ret) { + dev_err(&pdev->dev, "failed reset deassert: %d\n", ret); + return ERR_PTR(ret); + } + #ifdef CONFIG_PINCTRL_NOMADIK BUG_ON(nmk_chip->bank >= ARRAY_SIZE(nmk_gpio_chips)); nmk_gpio_chips[id] = nmk_chip;
Fetch a reference to the optional shared reset control and deassert it if it exists. Optional because not all platforms that use this driver have a reset attached to the reset block. Shared because some platforms that use the reset (at least Mobileye EyeQ5) share the reset across banks. Do not keep a reference to the reset control as it is not needed afterwards; the driver does not handle suspend, does not use runtime PM and does not register a remove callback. The operation is done in nmk_gpio_populate_chip(). This function is called by either gpio-nomadik or pinctrl-nomadik, whoever comes first. This is here for historic reasons and could probably be removed now; it seems gpio-ranges enforces the ordering to be pinctrl-first. It is not the topic of the present patch however. Signed-off-by: Théo Lebrun <theo.lebrun@bootlin.com> --- drivers/gpio/gpio-nomadik.c | 15 +++++++++++++++ 1 file changed, 15 insertions(+)