Re: [PATCH v1 4/4] gpio: gpio-aspeed-sgpio: Add AST2600 sgpio support

From: Andrew Jeffery
Date: Wed May 26 2021 - 21:27:14 EST


Hi Steven,

On Wed, 26 May 2021, at 19:16, Steven Lee wrote:
> AST2600 SoC has 2 SGPIO master interfaces one with 128 pins another one
> with 80 pins, and AST2500/AST2400 SoC has 1 SGPIO master interface that
> supports up to 80 pins.
> In the current driver design, the max number of sgpio pins is hardcoded
> in macro MAX_NR_HW_SGPIO and the value is 80.
> The patch makes the maximum gpio number *constraint of chip* comes from
> the dts. The property name is max-ngpios.
>
> Signed-off-by: Steven Lee <steven_lee@xxxxxxxxxxxxxx>
> ---
> drivers/gpio/gpio-aspeed-sgpio.c | 193 ++++++++++++++++++++-----------
> 1 file changed, 125 insertions(+), 68 deletions(-)
>
> diff --git a/drivers/gpio/gpio-aspeed-sgpio.c b/drivers/gpio/gpio-aspeed-sgpio.c
> index 64e54f8c30d2..f74bd7988b66 100644
> --- a/drivers/gpio/gpio-aspeed-sgpio.c
> +++ b/drivers/gpio/gpio-aspeed-sgpio.c
> @@ -17,37 +17,28 @@
> #include <linux/spinlock.h>
> #include <linux/string.h>
>
> -/*
> - * MAX_NR_HW_GPIO represents the number of actual hardware-supported GPIOs (ie,
> - * slots within the clocked serial GPIO data). Since each HW GPIO is both an
> - * input and an output, we provide MAX_NR_HW_GPIO * 2 lines on our gpiochip
> - * device.
> - *
> - * We use SGPIO_OUTPUT_OFFSET to define the split between the inputs and
> - * outputs; the inputs start at line 0, the outputs start at OUTPUT_OFFSET.
> - */
> -#define MAX_NR_HW_SGPIO 80
> -#define SGPIO_OUTPUT_OFFSET MAX_NR_HW_SGPIO

I expect you want to keep these as fall-back values for old devicetrees
that don't have your new properties.

> -
> #define ASPEED_SGPIO_CTRL 0x54
>
> -#define ASPEED_SGPIO_PINS_MASK GENMASK(9, 6)
> +#define ASPEED_SGPIO_PINS_MASK GENMASK(10, 6)

Is this valid on the 2400 and 2500? This might be better defined in
platform data so we don't go writing reserved bits on older hardware?

> #define ASPEED_SGPIO_CLK_DIV_MASK GENMASK(31, 16)
> #define ASPEED_SGPIO_ENABLE BIT(0)
>
> struct aspeed_sgpio {
> struct gpio_chip chip;
> + struct irq_chip intc;

My gut instinct is this could be in a separate patch.

> struct clk *pclk;
> spinlock_t lock;
> void __iomem *base;
> int irq;
> + int max_ngpios;
> int n_sgpio;
> };
>
> struct aspeed_sgpio_bank {
> - uint16_t val_regs;
> - uint16_t rdata_reg;
> - uint16_t irq_regs;
> + u16 val_regs;
> + u16 rdata_reg;
> + u16 irq_regs;
> + u16 tolerance_regs;

I suggest splitting this out into a cleanup patch.

> const char names[4][3];
> };
>
> @@ -63,19 +54,29 @@ static const struct aspeed_sgpio_bank
> aspeed_sgpio_banks[] = {
> .val_regs = 0x0000,
> .rdata_reg = 0x0070,
> .irq_regs = 0x0004,
> + .tolerance_regs = 0x0018,

Reset tolerance is supported on the 2500. Your patch subject claims its
for supporting the 2600, so I think the addition of reset tolerance
capabilities should be a separate patch.

By splitting out the different patches as requested, I think this patch
can just become about supporting max_ngpios, which will help focus the
discussion.

> .names = { "A", "B", "C", "D" },
> },
> {
> .val_regs = 0x001C,
> .rdata_reg = 0x0074,
> .irq_regs = 0x0020,
> + .tolerance_regs = 0x0034,
> .names = { "E", "F", "G", "H" },
> },
> {
> .val_regs = 0x0038,
> .rdata_reg = 0x0078,
> .irq_regs = 0x003C,
> - .names = { "I", "J" },
> + .tolerance_regs = 0x0050,
> + .names = { "I", "J", "K", "L" },
> + },
> + {
> + .val_regs = 0x0090,
> + .rdata_reg = 0x007C,
> + .irq_regs = 0x0094,
> + .tolerance_regs = 0x00A8,
> + .names = { "M", "N", "O", "P" },
> },
> };
>
> @@ -87,14 +88,14 @@ enum aspeed_sgpio_reg {
> reg_irq_type1,
> reg_irq_type2,
> reg_irq_status,
> + reg_tolerance,
> };
>
> -#define GPIO_VAL_VALUE 0x00
> -#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
> +#define GPIO_IRQ_OFFSET_ENABLE 0x00
> +#define GPIO_IRQ_OFFSET_TYPE0 0x04
> +#define GPIO_IRQ_OFFSET_TYPE1 0x08
> +#define GPIO_IRQ_OFFSET_TYPE2 0x0C
> +#define GPIO_IRQ_OFFSET_STATUS 0x10

I don't think this change is necessary. It looks like it produces a
bunch of noise in the diff.

>
> static void __iomem *bank_reg(struct aspeed_sgpio *gpio,
> const struct aspeed_sgpio_bank *bank,
> @@ -102,34 +103,37 @@ static void __iomem *bank_reg(struct aspeed_sgpio *gpio,
> {
> switch (reg) {
> case reg_val:
> - return gpio->base + bank->val_regs + GPIO_VAL_VALUE;
> + return gpio->base + bank->val_regs;
> case reg_rdata:
> return gpio->base + bank->rdata_reg;
> case reg_irq_enable:
> - return gpio->base + bank->irq_regs + GPIO_IRQ_ENABLE;
> + return gpio->base + bank->irq_regs + GPIO_IRQ_OFFSET_ENABLE;
> case reg_irq_type0:
> - return gpio->base + bank->irq_regs + GPIO_IRQ_TYPE0;
> + return gpio->base + bank->irq_regs + GPIO_IRQ_OFFSET_TYPE0;
> case reg_irq_type1:
> - return gpio->base + bank->irq_regs + GPIO_IRQ_TYPE1;
> + return gpio->base + bank->irq_regs + GPIO_IRQ_OFFSET_TYPE1;
> case reg_irq_type2:
> - return gpio->base + bank->irq_regs + GPIO_IRQ_TYPE2;
> + return gpio->base + bank->irq_regs + GPIO_IRQ_OFFSET_TYPE2;
> case reg_irq_status:
> - return gpio->base + bank->irq_regs + GPIO_IRQ_STATUS;
> + return gpio->base + bank->irq_regs + GPIO_IRQ_OFFSET_STATUS;
> + case reg_tolerance:
> + return gpio->base + bank->tolerance_regs;
> default:
> /* acturally if code runs to here, it's an error case */
> BUG();
> }
> }
>
> -#define GPIO_BANK(x) ((x % SGPIO_OUTPUT_OFFSET) >> 5)
> -#define GPIO_OFFSET(x) ((x % SGPIO_OUTPUT_OFFSET) & 0x1f)
> +#define GPIO_BANK(x) ((x) >> 5)
> +/* modulo 32 */
> +#define GPIO_OFFSET(x) ((x) & 0x1f)
> #define GPIO_BIT(x) BIT(GPIO_OFFSET(x))
>
> -static const struct aspeed_sgpio_bank *to_bank(unsigned int offset)
> +static const struct aspeed_sgpio_bank *to_bank(unsigned int offset,
> unsigned int max_ngpios)
> {
> unsigned int bank;
>
> - bank = GPIO_BANK(offset);
> + bank = GPIO_BANK(offset % max_ngpios);
>
> WARN_ON(bank >= ARRAY_SIZE(aspeed_sgpio_banks));
> return &aspeed_sgpio_banks[bank];
> @@ -139,18 +143,19 @@ static int aspeed_sgpio_init_valid_mask(struct
> gpio_chip *gc,
> unsigned long *valid_mask, unsigned int ngpios)
> {
> struct aspeed_sgpio *sgpio = gpiochip_get_data(gc);
> + int max_ngpios = sgpio->max_ngpios;
> int n = sgpio->n_sgpio;
> - int c = SGPIO_OUTPUT_OFFSET - n;
> + int c = max_ngpios - n;
>
> - WARN_ON(ngpios < MAX_NR_HW_SGPIO * 2);
> + WARN_ON(ngpios < max_ngpios * 2);
>
> /* input GPIOs in the lower range */
> bitmap_set(valid_mask, 0, n);
> bitmap_clear(valid_mask, n, c);
>
> - /* output GPIOS above SGPIO_OUTPUT_OFFSET */
> - bitmap_set(valid_mask, SGPIO_OUTPUT_OFFSET, n);
> - bitmap_clear(valid_mask, SGPIO_OUTPUT_OFFSET + n, c);
> + /* output GPIOS above max_ngpios */
> + bitmap_set(valid_mask, max_ngpios, n);
> + bitmap_clear(valid_mask, max_ngpios + n, c);
>
> return 0;
> }
> @@ -161,30 +166,30 @@ static void
> aspeed_sgpio_irq_init_valid_mask(struct gpio_chip *gc,
> struct aspeed_sgpio *sgpio = gpiochip_get_data(gc);
> int n = sgpio->n_sgpio;
>
> - WARN_ON(ngpios < MAX_NR_HW_SGPIO * 2);
> + WARN_ON(ngpios < sgpio->max_ngpios * 2);
>
> /* input GPIOs in the lower range */
> bitmap_set(valid_mask, 0, n);
> bitmap_clear(valid_mask, n, ngpios - n);
> }
>
> -static bool aspeed_sgpio_is_input(unsigned int offset)
> +static bool aspeed_sgpio_is_input(unsigned int offset, unsigned int max_ngpios)
> {
> - return offset < SGPIO_OUTPUT_OFFSET;
> + return offset < max_ngpios;
> }
>
> static int aspeed_sgpio_get(struct gpio_chip *gc, unsigned int offset)
> {
> + const struct aspeed_sgpio_bank *bank = to_bank(offset, gpio->max_ngpios);

This references gpio before it's defined, as the build bot noted.

> struct aspeed_sgpio *gpio = gpiochip_get_data(gc);
> - const struct aspeed_sgpio_bank *bank = to_bank(offset);
> unsigned long flags;
> enum aspeed_sgpio_reg reg;
> int rc = 0;
>
> spin_lock_irqsave(&gpio->lock, flags);
>
> - reg = aspeed_sgpio_is_input(offset) ? reg_val : reg_rdata;
> - rc = !!(ioread32(bank_reg(gpio, bank, reg)) & GPIO_BIT(offset));
> + reg = aspeed_sgpio_is_input(offset, gpio->max_ngpios) ? reg_val :
> reg_rdata;

We should just pass gpio here (i.e. make aspeed_sgpio_is_input take a
'const struct aspeed_sgpio *' parameter), rather than open-coding
gpio->max_ngpios. This approach will make it easier to refactor the
implementation in the future (if necessary).

> + rc = !!(ioread32(bank_reg(gpio, bank, reg)) & GPIO_BIT(offset %
> gpio->max_ngpios));
>
> spin_unlock_irqrestore(&gpio->lock, flags);
>
> @@ -193,12 +198,12 @@ static int aspeed_sgpio_get(struct gpio_chip *gc,
> unsigned int offset)
>
> static int sgpio_set_value(struct gpio_chip *gc, unsigned int offset, int val)
> {
> + const struct aspeed_sgpio_bank *bank = to_bank(offset, gpio->max_ngpios);

References gpio before it's defined.

> struct aspeed_sgpio *gpio = gpiochip_get_data(gc);
> - const struct aspeed_sgpio_bank *bank = to_bank(offset);
> void __iomem *addr_r, *addr_w;
> u32 reg = 0;
>
> - if (aspeed_sgpio_is_input(offset))
> + if (aspeed_sgpio_is_input(offset, gpio->max_ngpios))
> return -EINVAL;
>
> /* Since this is an output, read the cached value from rdata, then
> @@ -209,9 +214,9 @@ static int sgpio_set_value(struct gpio_chip *gc,
> unsigned int offset, int val)
> reg = ioread32(addr_r);
>
> if (val)
> - reg |= GPIO_BIT(offset);
> + reg |= GPIO_BIT(offset % gpio->max_ngpios);

Pass gpio through GPIO_BIT() too.

> else
> - reg &= ~GPIO_BIT(offset);
> + reg &= ~GPIO_BIT(offset % gpio->max_ngpios);
>
> iowrite32(reg, addr_w);
>
> @@ -232,7 +237,9 @@ static void aspeed_sgpio_set(struct gpio_chip *gc,
> unsigned int offset, int val)
>
> static int aspeed_sgpio_dir_in(struct gpio_chip *gc, unsigned int offset)
> {
> - return aspeed_sgpio_is_input(offset) ? 0 : -EINVAL;
> + struct aspeed_sgpio *gpio = gpiochip_get_data(gc);
> +
> + return aspeed_sgpio_is_input(offset, gpio->max_ngpios) ? 0 : -EINVAL;
> }
>
> static int aspeed_sgpio_dir_out(struct gpio_chip *gc, unsigned int
> offset, int val)
> @@ -253,7 +260,9 @@ static int aspeed_sgpio_dir_out(struct gpio_chip
> *gc, unsigned int offset, int v
>
> static int aspeed_sgpio_get_direction(struct gpio_chip *gc, unsigned
> int offset)
> {
> - return !!aspeed_sgpio_is_input(offset);
> + struct aspeed_sgpio *gpio = gpiochip_get_data(gc);
> +
> + return !!aspeed_sgpio_is_input(offset, gpio->max_ngpios);
> }
>
> static void irqd_to_aspeed_sgpio_data(struct irq_data *d,
> @@ -268,8 +277,8 @@ static void irqd_to_aspeed_sgpio_data(struct irq_data *d,
> WARN_ON(!internal);
>
> *gpio = internal;
> - *bank = to_bank(*offset);
> - *bit = GPIO_BIT(*offset);
> + *bank = to_bank(*offset, internal->max_ngpios);

Make to_bank take a 'const struct aspeed_sgpio *'.

> + *bit = GPIO_BIT(*offset % internal->max_ngpios);
> }
>
> static void aspeed_sgpio_irq_ack(struct irq_data *d)
> @@ -412,14 +421,6 @@ static void aspeed_sgpio_irq_handler(struct irq_desc *desc)
> chained_irq_exit(ic, desc);
> }
>
> -static struct irq_chip aspeed_sgpio_irqchip = {
> - .name = "aspeed-sgpio",
> - .irq_ack = aspeed_sgpio_irq_ack,
> - .irq_mask = aspeed_sgpio_irq_mask,
> - .irq_unmask = aspeed_sgpio_irq_unmask,
> - .irq_set_type = aspeed_sgpio_set_type,
> -};
> -
> static int aspeed_sgpio_setup_irqs(struct aspeed_sgpio *gpio,
> struct platform_device *pdev)
> {
> @@ -442,8 +443,14 @@ static int aspeed_sgpio_setup_irqs(struct
> aspeed_sgpio *gpio,
> iowrite32(0xffffffff, bank_reg(gpio, bank, reg_irq_status));
> }
>
> + gpio->intc.name = dev_name(&pdev->dev);
> + gpio->intc.irq_ack = aspeed_sgpio_irq_ack;
> + gpio->intc.irq_mask = aspeed_sgpio_irq_mask;
> + gpio->intc.irq_unmask = aspeed_sgpio_irq_unmask;
> + gpio->intc.irq_set_type = aspeed_sgpio_set_type;
> +
> irq = &gpio->chip.irq;
> - irq->chip = &aspeed_sgpio_irqchip;
> + irq->chip = &gpio->intc;

As I said earlier, I think it would be best to split the IRQ changes
out into a separate patch ordered before this one.

> irq->init_valid_mask = aspeed_sgpio_irq_init_valid_mask;
> irq->handler = handle_bad_irq;
> irq->default_type = IRQ_TYPE_NONE;
> @@ -466,9 +473,48 @@ static int aspeed_sgpio_setup_irqs(struct
> aspeed_sgpio *gpio,
> return 0;
> }
>
> +static int aspeed_sgpio_reset_tolerance(struct gpio_chip *chip,
> + unsigned int offset, bool enable)
> +{
> + struct aspeed_sgpio *gpio = gpiochip_get_data(chip);
> + unsigned long flags;
> + void __iomem *reg;
> + u32 val;
> +
> + reg = bank_reg(gpio, to_bank(offset, gpio->max_ngpios), reg_tolerance);
> +
> + spin_lock_irqsave(&gpio->lock, flags);
> +
> + val = readl(reg);
> +
> + if (enable)
> + val |= GPIO_BIT(offset % gpio->max_ngpios);
> + else
> + val &= ~GPIO_BIT(offset % gpio->max_ngpios);
> +
> + writel(val, reg);
> +
> + spin_unlock_irqrestore(&gpio->lock, flags);
> +
> + return 0;
> +}
> +
> +static int aspeed_sgpio_set_config(struct gpio_chip *chip, unsigned int offset,
> + unsigned long config)
> +{
> + unsigned long param = pinconf_to_config_param(config);
> + u32 arg = pinconf_to_config_argument(config);
> +
> + if (param == PIN_CONFIG_PERSIST_STATE)
> + return aspeed_sgpio_reset_tolerance(chip, offset, arg);
> + else
> + return -EOPNOTSUPP;
> +}
> +

Again, lets split reset tolerance out to a separate patch as well.

> static const struct of_device_id aspeed_sgpio_of_table[] = {
> - { .compatible = "aspeed,ast2400-sgpio" },
> - { .compatible = "aspeed,ast2500-sgpio" },
> + { .compatible = "aspeed,ast2400-sgpiom" },
> + { .compatible = "aspeed,ast2500-sgpiom" },
> + { .compatible = "aspeed,ast2600-sgpiom" },

No. You'll need to support both while we transition between the two as
the devicetree patch and the driver patch are separate, and this will
break bisectability when tracking down runtime issues. Also, the
devicetree patch should be ordered after the driver change for the same
reason.

You should continue to support the 'aspeed,ast2{4,5}00-sgpio' style
compatible here to handle old devicetrees.

So in summary, just add the 'aspeed,ast2{4,5,6}00-sgpiom' compatibles
for now, don't remove the 'aspeed,ast2{4,5}00-sgpio' compatibles.

> {}
> };
>
> @@ -476,8 +522,8 @@ MODULE_DEVICE_TABLE(of, aspeed_sgpio_of_table);
>
> static int __init aspeed_sgpio_probe(struct platform_device *pdev)
> {
> + u32 nr_gpios, sgpio_freq, sgpio_clk_div, max_ngpios;
> struct aspeed_sgpio *gpio;
> - u32 nr_gpios, sgpio_freq, sgpio_clk_div;
> int rc;
> unsigned long apb_freq;
>
> @@ -489,13 +535,24 @@ static int __init aspeed_sgpio_probe(struct
> platform_device *pdev)
> if (IS_ERR(gpio->base))
> return PTR_ERR(gpio->base);
>
> + rc = of_property_read_u32(pdev->dev.of_node, "max-ngpios",
> &max_ngpios);
> + if (rc < 0) {
> + dev_err(&pdev->dev, "Could not read max-ngpios property\n");
> + return -EINVAL;

I don't think this is right. You should just fall back to using the
hard-coded value.

Cheers,

Andrew