On Wed, Jul 20, 2016 at 7:58 AM, Andrew Jeffery <andrew@xxxxxxxx> wrote:
> diff --git a/arch/arm/mach-aspeed/Kconfig b/arch/arm/mach-aspeed/Kconfig
> index 25a0ae01429e..a52de9d3adfb 100644
> --- a/arch/arm/mach-aspeed/Kconfig
> +++ b/arch/arm/mach-aspeed/Kconfig
> @@ -6,6 +6,10 @@ menuconfig ARCH_ASPEED
> select ASPEED_WATCHDOG
> select MOXART_TIMER
> select PINCTRL
> + select GPIOLIB
> + select GPIO_ASPEED
Again this needs to be a separate patch to ARM SoC.
> + select GPIO_SYSFS
NAK over my dead body. I strongly discourage the use of the
GPIO sysfs, use the new character device, see
toos/gpio/* for examples.
> +config GPIO_ASPEED
> + bool "Aspeed GPIO support"
> + depends on (ARCH_ASPEED || COMPILE_TEST) && OF
> + select GENERIC_IRQ_CHIP
Why are you using GENERIC_IRQ_CHIP but not
GPIOLIB_IRQCHIP? Well I guess I may find out by
reading the code...
> + help
> + Say Y here to support Aspeed AST2400 and AST2500 GPIO controllers.
> +
> config GPIO_BCM_KONA
> bool "Broadcom Kona GPIO"
> depends on OF_GPIO && (ARCH_BCM_MOBILE || COMPILE_TEST)
> @@ -1072,7 +1079,6 @@ config GPIO_SODAVILLE
> select GENERIC_IRQ_CHIP
> help
> Say Y here to support Intel Sodaville GPIO.
> -
> endmenu
Drop this unrelated whitespace change.
> +#define GPIO_BANK(x) ((x) >> 5)
> +#define GPIO_OFFSET(x) ((x) & 0x1f)
> +#define GPIO_BIT(x) BIT(GPIO_OFFSET(x))
Clever, maybe needs some comments on how they work?
Or is it obvious from context?
> +#define GPIO_DATA 0x00
> +#define GPIO_DIR 0x04
> +
> +#define GPIO_IRQ_ENABLE 0x00
> +#define GPIO_IRQ_TYPE0 0x04
> +#define GPIO_IRQ_TYPE1 0x08
> +#define GPIO_IRQ_TYPE2 0x0c
> +#define GPIO_IRQ_STATUS 0x10
> +static inline struct aspeed_gpio *to_aspeed_gpio(struct gpio_chip *chip)
> +{
> + return container_of(chip, struct aspeed_gpio, chip);
> +}
NAK rewrite your code to use devm_gpiochip_add_data() and
then use gpiochip_get_data() inline in every function that
needs to get the state container from the gpiochip. Read
the code in the upstream kernel for *any* driver because I
think I changed this virtually everywhere.
(...)
> +static void __aspeed_gpio_irq_set_mask(struct irq_data *d, bool set)
Why __underscoring. It is just confusing, drop the underscores.
The function does set the mask.
> +{
> + const struct aspeed_gpio_bank *bank;
> + struct aspeed_gpio *gpio;
> + unsigned long flags;
> + u32 reg, bit;
> + void *addr;
> + int rc;
> +
> + rc = irqd_to_aspeed_gpio_data(d, &gpio, &bank, &bit);
> + if (rc)
> + return;
> +
> + addr = bank_irq_reg(gpio, bank, GPIO_IRQ_ENABLE);
> +
> + spin_lock_irqsave(&gpio->lock, flags);
> +
> + reg = ioread32(addr);
> + if (set)
> + reg |= bit;
> + else
> + reg &= bit;
> + iowrite32(reg, addr);
Hm, if this was done with regmap it would be regmap_update_bits()
simply ... maybe you should just throw a 32bit MMIO regmap over
the registers? (Not required, just an idea to simplify stuff...)
> +static int aspeed_gpio_set_type(struct irq_data *d, unsigned int type)
> +{
> + u32 type0, type1, type2, bit, reg;
> + const struct aspeed_gpio_bank *bank;
> + irq_flow_handler_t handler;
> + struct aspeed_gpio *gpio;
> + unsigned long flags;
> + void *addr;
> + int rc;
> +
> + rc = irqd_to_aspeed_gpio_data(d, &gpio, &bank, &bit);
> + if (rc)
> + return -EINVAL;
> +
> + type0 = type1 = type2 = 0;
Assign zero when declaring instead:
u32 type0 = 0;
u32 type1 = 0;
...
> +
> + switch (type & IRQ_TYPE_SENSE_MASK) {
> + case IRQ_TYPE_EDGE_BOTH:
> + type2 |= bit;
> + case IRQ_TYPE_EDGE_RISING:
> + type0 |= bit;
> + case IRQ_TYPE_EDGE_FALLING:
> + handler = handle_edge_irq;
> + break;
> + case IRQ_TYPE_LEVEL_HIGH:
> + type0 |= bit;
> + case IRQ_TYPE_LEVEL_LOW:
> + type1 |= bit;
> + handler = handle_level_irq;
> + break;
> + default:
> + return -EINVAL;
> + }
Very nice, this looks exactly as it should, handling the different
IRQs very nicely.
> + spin_lock_irqsave(&gpio->lock, flags);
> +
> + addr = bank_irq_reg(gpio, bank, GPIO_IRQ_TYPE0);
> + reg = ioread32(addr);
> + reg = (reg & ~bit) | type0;
> + iowrite32(reg, addr);
> +
> + addr = bank_irq_reg(gpio, bank, GPIO_IRQ_TYPE1);
> + reg = ioread32(addr);
> + reg = (reg & ~bit) | type1;
> + iowrite32(reg, addr);
> +
> + addr = bank_irq_reg(gpio, bank, GPIO_IRQ_TYPE2);
> + reg = ioread32(addr);
> + reg = (reg & ~bit) | type2;
> + iowrite32(reg, addr);
> +
> + spin_unlock_irqrestore(&gpio->lock, flags);
> +
> + irq_set_handler_locked(d, handler);
> +
> + return 0;
> +}
Overall very nice .set_type().
> +static void aspeed_gpio_irq_handler(struct irq_desc *desc)
> +{
> + struct aspeed_gpio *gpio = irq_desc_get_handler_data(desc);
> + struct irq_chip *chip = irq_desc_get_chip(desc);
> + unsigned int i, p, girq;
> + unsigned long reg;
> +
> + chained_irq_enter(chip, desc);
> +
> + for (i = 0; i < ARRAY_SIZE(aspeed_gpio_banks); i++) {
> + const struct aspeed_gpio_bank *bank = &aspeed_gpio_banks[i];
> +
> + reg = ioread32(bank_irq_reg(gpio, bank, GPIO_IRQ_STATUS));
> +
> + for_each_set_bit(p, ®, 32) {
> + girq = irq_find_mapping(gpio->irq_domain, i * 32 + p);
> + generic_handle_irq(girq);
> + }
> +
> + }
> +
> + chained_irq_exit(chip, desc);
> +}
This looks so generic so I think you should be using
GPIOLIB_IRQCHIP.
> +
> +static struct irq_chip aspeed_gpio_irqchip = {
> + .name = "aspeed-gpio",
> + .irq_ack = aspeed_gpio_irq_ack,
> + .irq_mask = aspeed_gpio_irq_mask,
> + .irq_unmask = aspeed_gpio_irq_unmask,
> + .irq_set_type = aspeed_gpio_set_type,
> +};
There is a missing .request/.release resources marking the lines
as irq to the GPIO core. But if you switch to using GPIOLIB_IRQCHIP
the core will handle all that for you.
> +static int aspeed_gpio_to_irq(struct gpio_chip *chip, unsigned int offset)
> +{
> + struct aspeed_gpio *gpio = to_aspeed_gpio(chip);
> +
> + return irq_find_mapping(gpio->irq_domain, offset);
> +}
> +
> +static void aspeed_gpio_setup_irqs(struct aspeed_gpio *gpio,
> + struct platform_device *pdev)
> +{
> + int i, irq;
> +
> + /* request our upstream IRQ */
> + gpio->irq = platform_get_irq(pdev, 0);
> + if (gpio->irq < 0)
> + return;
> +
> + /* establish our irq domain to provide IRQs for each extended bank */
> + gpio->irq_domain = irq_domain_add_linear(pdev->dev.of_node,
> + gpio->chip.ngpio, &irq_domain_simple_ops, NULL);
> + if (!gpio->irq_domain)
> + return;
> +
> + for (i = 0; i < gpio->chip.ngpio; i++) {
> + irq = irq_create_mapping(gpio->irq_domain, i);
> + irq_set_chip_data(irq, gpio);
> + irq_set_chip_and_handler(irq, &aspeed_gpio_irqchip,
> + handle_simple_irq);
> + irq_set_probe(irq);
> + }
> +
> + irq_set_chained_handler_and_data(gpio->irq,
> + aspeed_gpio_irq_handler, gpio);
> +}
Also all this goes away with GPILIB_IRQCHIP, so use it.
See e.g. drivers/gpio/gpio-pl061.c for an example of a similar
driver doing this.
> +static int __init aspeed_gpio_probe(struct platform_device *pdev)
> +{
> + struct aspeed_gpio *gpio;
> + struct resource *res;
> + int rc;
> +
> + gpio = devm_kzalloc(&pdev->dev, sizeof(*gpio), GFP_KERNEL);
> + if (!gpio)
> + return -ENOMEM;
> +
> + res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
> + if (!res)
> + return -ENXIO;
> +
> + gpio->base = devm_ioremap_resource(&pdev->dev, res);
> + if (!gpio->base)
> + return -ENOMEM;
> +
> + spin_lock_init(&gpio->lock);
> +
> + gpio->chip.ngpio = ARRAY_SIZE(aspeed_gpio_banks) * 32;
> +
> + gpio->chip.parent = &pdev->dev;
> + gpio->chip.direction_input = aspeed_gpio_dir_in;
> + gpio->chip.direction_output = aspeed_gpio_dir_out;
Please add gpio->chip.get_direction() to complete the picture.
> + gpio->chip.request = aspeed_gpio_request;
> + gpio->chip.free = aspeed_gpio_free;
> + gpio->chip.get = aspeed_gpio_get;
> + gpio->chip.set = aspeed_gpio_set;
> + gpio->chip.to_irq = aspeed_gpio_to_irq;
.to_irq goes away with GPILIB_IRQCHIP.
> + gpio->chip.label = dev_name(&pdev->dev);
> + gpio->chip.base = -1;
> +
> + platform_set_drvdata(pdev, gpio);
> +
> + rc = gpiochip_add(&gpio->chip);
> + if (rc < 0)
> + return rc;
Use devm_gpiochip_add_data(), then gpiochip_irqchip_add().
> +static int aspeed_gpio_remove(struct platform_device *pdev)
> +{
> + struct aspeed_gpio *gpio = platform_get_drvdata(pdev);
> +
> + gpiochip_remove(&gpio->chip);
> + return 0;
> +}
And then this is not even needed. (devm*)
Yours,
Linus Walleij
--
To unsubscribe from this list: send the line "unsubscribe devicetree" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
