[PATCH 1/2] gpio: tegra186: Force one interrupt per bank

[PATCH 1/2] gpio: tegra186: Force one interrupt per bank

[Thierry Reding]

From: Thierry Reding <treding@nvidia.com>

Newer chips support up to 8 interrupts per bank, which can be useful to
balance the load and decrease latency. However, it also required a very
complicated interrupt routing to be set up. To keep things simple for
now, ensure that a single interrupt per bank is enforced, even if all
possible interrupts are described in device tree.

Signed-off-by: Thierry Reding <treding@nvidia.com>
---
 drivers/gpio/gpio-tegra186.c | 68 ++++++++++++++++++++++++++++++++----
 1 file changed, 62 insertions(+), 6 deletions(-)

diff --git a/drivers/gpio/gpio-tegra186.c b/drivers/gpio/gpio-tegra186.c
index c99858f40a27..1bc4152e0275 100644
--- a/drivers/gpio/gpio-tegra186.c
+++ b/drivers/gpio/gpio-tegra186.c
@@ -81,6 +81,8 @@ struct tegra_gpio {
 	unsigned int *irq;
 
 	const struct tegra_gpio_soc *soc;
+	unsigned int num_irqs_per_bank;
+	unsigned int num_banks;
 
 	void __iomem *secure;
 	void __iomem *base;
@@ -594,6 +596,28 @@ static void tegra186_gpio_init_route_mapping(struct tegra_gpio *gpio)
 	}
 }
 
+static unsigned int tegra186_gpio_irqs_per_bank(struct tegra_gpio *gpio)
+{
+	struct device *dev = gpio->gpio.parent;
+
+	if (gpio->num_irq > gpio->num_banks) {
+		if (gpio->num_irq % gpio->num_banks != 0)
+			goto error;
+	}
+
+	if (gpio->num_irq < gpio->num_banks)
+		goto error;
+
+	gpio->num_irqs_per_bank = gpio->num_irq / gpio->num_banks;
+
+	return 0;
+
+error:
+	dev_err(dev, "invalid number of interrupts (%u) for %u banks\n",
+		gpio->num_irq, gpio->num_banks);
+	return -EINVAL;
+}
+
 static int tegra186_gpio_probe(struct platform_device *pdev)
 {
 	unsigned int i, j, offset;
@@ -608,7 +632,17 @@ static int tegra186_gpio_probe(struct platform_device *pdev)
 		return -ENOMEM;
 
 	gpio->soc = device_get_match_data(&pdev->dev);
+	gpio->gpio.label = gpio->soc->name;
+	gpio->gpio.parent = &pdev->dev;
+
+	/* count the number of banks in the controller */
+	for (i = 0; i < gpio->soc->num_ports; i++)
+		if (gpio->soc->ports[i].bank > gpio->num_banks)
+			gpio->num_banks = gpio->soc->ports[i].bank;
+
+	gpio->num_banks++;
 
+	/* get register apertures */
 	gpio->secure = devm_platform_ioremap_resource_byname(pdev, "security");
 	if (IS_ERR(gpio->secure)) {
 		gpio->secure = devm_platform_ioremap_resource(pdev, 0);
@@ -629,6 +663,10 @@ static int tegra186_gpio_probe(struct platform_device *pdev)
 
 	gpio->num_irq = err;
 
+	err = tegra186_gpio_irqs_per_bank(gpio);
+	if (err < 0)
+		return err;
+
 	gpio->irq = devm_kcalloc(&pdev->dev, gpio->num_irq, sizeof(*gpio->irq),
 				 GFP_KERNEL);
 	if (!gpio->irq)
@@ -642,9 +680,6 @@ static int tegra186_gpio_probe(struct platform_device *pdev)
 		gpio->irq[i] = err;
 	}
 
-	gpio->gpio.label = gpio->soc->name;
-	gpio->gpio.parent = &pdev->dev;
-
 	gpio->gpio.request = gpiochip_generic_request;
 	gpio->gpio.free = gpiochip_generic_free;
 	gpio->gpio.get_direction = tegra186_gpio_get_direction;
@@ -708,7 +743,30 @@ static int tegra186_gpio_probe(struct platform_device *pdev)
 	irq->parent_handler = tegra186_gpio_irq;
 	irq->parent_handler_data = gpio;
 	irq->num_parents = gpio->num_irq;
-	irq->parents = gpio->irq;
+
+	/*
+	 * To simplify things, use a single interrupt per bank for now. Some
+	 * chips support up to 8 interrupts per bank, which can be useful to
+	 * distribute the load and decrease the processing latency for GPIOs
+	 * but it also requires a more complicated interrupt routing than we
+	 * currently program.
+	 */
+	if (gpio->num_irqs_per_bank > 1) {
+		irq->parents = devm_kcalloc(&pdev->dev, gpio->num_banks,
+					    sizeof(*irq->parents), GFP_KERNEL);
+		if (!irq->parents)
+			return -ENOMEM;
+
+		for (i = 0; i < gpio->num_banks; i++)
+			irq->parents[i] = gpio->irq[i * gpio->num_irqs_per_bank];
+
+		irq->num_parents = gpio->num_banks;
+	} else {
+		irq->num_parents = gpio->num_irq;
+		irq->parents = gpio->irq;
+	}
+
+	tegra186_gpio_init_route_mapping(gpio);
 
 	np = of_find_matching_node(NULL, tegra186_pmc_of_match);
 	if (np) {
@@ -719,8 +777,6 @@ static int tegra186_gpio_probe(struct platform_device *pdev)
 			return -EPROBE_DEFER;
 	}
 
-	tegra186_gpio_init_route_mapping(gpio);
-
 	irq->map = devm_kcalloc(&pdev->dev, gpio->gpio.ngpio,
 				sizeof(*irq->map), GFP_KERNEL);
 	if (!irq->map)
-- 
2.33.0

[PATCH 2/2] gpio: tegra186: Support multiple interrupts per bank

[Thierry Reding]

From: Thierry Reding <treding@nvidia.com>

Tegra194 and later support more than a single interrupt per bank. This
is primarily useful for virtualization but can also be helpful for more
fine-grained CPU affinity control. To keep things simple for now, route
all pins to the first interrupt.

For backwards-compatibility, support old device trees that specify only
one interrupt per bank by counting the interrupts at probe time.

Signed-off-by: Thierry Reding <treding@nvidia.com>
---
 drivers/gpio/gpio-tegra186.c | 48 ++++++++++++++++++++++++++++++------
 1 file changed, 41 insertions(+), 7 deletions(-)

diff --git a/drivers/gpio/gpio-tegra186.c b/drivers/gpio/gpio-tegra186.c
index 1bc4152e0275..c026e7141e4e 100644
--- a/drivers/gpio/gpio-tegra186.c
+++ b/drivers/gpio/gpio-tegra186.c
@@ -69,6 +69,8 @@ struct tegra_gpio_soc {
 	const char *name;
 	unsigned int instance;
 
+	unsigned int num_irqs_per_bank;
+
 	const struct tegra186_pin_range *pin_ranges;
 	unsigned int num_pin_ranges;
 	const char *pinmux;
@@ -452,7 +454,7 @@ static void tegra186_gpio_irq(struct irq_desc *desc)
 	struct irq_domain *domain = gpio->gpio.irq.domain;
 	struct irq_chip *chip = irq_desc_get_chip(desc);
 	unsigned int parent = irq_desc_get_irq(desc);
-	unsigned int i, offset = 0;
+	unsigned int i, j, offset = 0;
 
 	chained_irq_enter(chip, desc);
 
@@ -465,7 +467,12 @@ static void tegra186_gpio_irq(struct irq_desc *desc)
 		base = gpio->base + port->bank * 0x1000 + port->port * 0x200;
 
 		/* skip ports that are not associated with this bank */
-		if (parent != gpio->irq[port->bank])
+		for (j = 0; j < gpio->num_irqs_per_bank; j++) {
+			if (parent == gpio->irq[port->bank * gpio->num_irqs_per_bank + j])
+				break;
+		}
+
+		if (j == gpio->num_irqs_per_bank)
 			goto skip;
 
 		value = readl(base + TEGRA186_GPIO_INTERRUPT_STATUS(1));
@@ -567,6 +574,7 @@ static const struct of_device_id tegra186_pmc_of_match[] = {
 
 static void tegra186_gpio_init_route_mapping(struct tegra_gpio *gpio)
 {
+	struct device *dev = gpio->gpio.parent;
 	unsigned int i, j;
 	u32 value;
 
@@ -585,12 +593,30 @@ static void tegra186_gpio_init_route_mapping(struct tegra_gpio *gpio)
 		 */
 		if ((value & TEGRA186_GPIO_CTL_SCR_SEC_REN) == 0 &&
 		    (value & TEGRA186_GPIO_CTL_SCR_SEC_WEN) == 0) {
-			for (j = 0; j < 8; j++) {
+			/*
+			 * On Tegra194 and later, each pin can be routed to one or more
+			 * interrupts.
+			 */
+			for (j = 0; j < gpio->num_irqs_per_bank; j++) {
+				dev_dbg(dev, "programming default interrupt routing for port %s\n",
+					port->name);
+
 				offset = TEGRA186_GPIO_INT_ROUTE_MAPPING(p, j);
 
-				value = readl(base + offset);
-				value = BIT(port->pins) - 1;
-				writel(value, base + offset);
+				/*
+				 * By default we only want to route GPIO pins to IRQ 0. This works
+				 * only under the assumption that we're running as the host kernel
+				 * and hence all GPIO pins are owned by Linux.
+				 *
+				 * For cases where Linux is the guest OS, the hypervisor will have
+				 * to configure the interrupt routing and pass only the valid
+				 * interrupts via device tree.
+				 */
+				if (j == 0) {
+					value = readl(base + offset);
+					value = BIT(port->pins) - 1;
+					writel(value, base + offset);
+				}
 			}
 		}
 	}
@@ -610,6 +636,9 @@ static unsigned int tegra186_gpio_irqs_per_bank(struct tegra_gpio *gpio)
 
 	gpio->num_irqs_per_bank = gpio->num_irq / gpio->num_banks;
 
+	if (gpio->num_irqs_per_bank > gpio->soc->num_irqs_per_bank)
+		goto error;
+
 	return 0;
 
 error:
@@ -766,7 +795,8 @@ static int tegra186_gpio_probe(struct platform_device *pdev)
 		irq->parents = gpio->irq;
 	}
 
-	tegra186_gpio_init_route_mapping(gpio);
+	if (gpio->soc->num_irqs_per_bank > 1)
+		tegra186_gpio_init_route_mapping(gpio);
 
 	np = of_find_matching_node(NULL, tegra186_pmc_of_match);
 	if (np) {
@@ -833,6 +863,7 @@ static const struct tegra_gpio_soc tegra186_main_soc = {
 	.ports = tegra186_main_ports,
 	.name = "tegra186-gpio",
 	.instance = 0,
+	.num_irqs_per_bank = 1,
 };
 
 #define TEGRA186_AON_GPIO_PORT(_name, _bank, _port, _pins)	\
@@ -859,6 +890,7 @@ static const struct tegra_gpio_soc tegra186_aon_soc = {
 	.ports = tegra186_aon_ports,
 	.name = "tegra186-gpio-aon",
 	.instance = 1,
+	.num_irqs_per_bank = 1,
 };
 
 #define TEGRA194_MAIN_GPIO_PORT(_name, _bank, _port, _pins)	\
@@ -910,6 +942,7 @@ static const struct tegra_gpio_soc tegra194_main_soc = {
 	.ports = tegra194_main_ports,
 	.name = "tegra194-gpio",
 	.instance = 0,
+	.num_irqs_per_bank = 8,
 	.num_pin_ranges = ARRAY_SIZE(tegra194_main_pin_ranges),
 	.pin_ranges = tegra194_main_pin_ranges,
 	.pinmux = "nvidia,tegra194-pinmux",
@@ -936,6 +969,7 @@ static const struct tegra_gpio_soc tegra194_aon_soc = {
 	.ports = tegra194_aon_ports,
 	.name = "tegra194-gpio-aon",
 	.instance = 1,
+	.num_irqs_per_bank = 8,
 };
 
 static const struct of_device_id tegra186_gpio_of_match[] = {
-- 
2.33.0

Re: [PATCH 1/2] gpio: tegra186: Force one interrupt per bank

[Linus Walleij]

On Fri, Sep 17, 2021 at 12:54 PM Thierry Reding
<thierry.reding@gmail.com> wrote:

> From: Thierry Reding <treding@nvidia.com>
>
> Newer chips support up to 8 interrupts per bank, which can be useful to
> balance the load and decrease latency. However, it also required a very
> complicated interrupt routing to be set up. To keep things simple for
> now, ensure that a single interrupt per bank is enforced, even if all
> possible interrupts are described in device tree.
>
> Signed-off-by: Thierry Reding <treding@nvidia.com>

Reviewed-by: Linus Walleij <linus.walleij@linaro.org>

Yours,
Linus Walleij

Re: [PATCH 2/2] gpio: tegra186: Support multiple interrupts per bank

[Linus Walleij]

On Fri, Sep 17, 2021 at 12:54 PM Thierry Reding
<thierry.reding@gmail.com> wrote:

> From: Thierry Reding <treding@nvidia.com>
>
> Tegra194 and later support more than a single interrupt per bank. This
> is primarily useful for virtualization but can also be helpful for more
> fine-grained CPU affinity control. To keep things simple for now, route
> all pins to the first interrupt.
>
> For backwards-compatibility, support old device trees that specify only
> one interrupt per bank by counting the interrupts at probe time.
>
> Signed-off-by: Thierry Reding <treding@nvidia.com>

Reviewed-by: Linus Walleij <linus.walleij@linaro.org>

Yours,
Linus Walleij

Re: [PATCH 1/2] gpio: tegra186: Force one interrupt per bank

[Bartosz Golaszewski]

On Fri, Sep 17, 2021 at 12:54 PM Thierry Reding
<thierry.reding@gmail.com> wrote:
>
> From: Thierry Reding <treding@nvidia.com>
>
> Newer chips support up to 8 interrupts per bank, which can be useful to
> balance the load and decrease latency. However, it also required a very
> complicated interrupt routing to be set up. To keep things simple for
> now, ensure that a single interrupt per bank is enforced, even if all
> possible interrupts are described in device tree.
>
> Signed-off-by: Thierry Reding <treding@nvidia.com>
> ---

Applied, thanks!

Bart

Re: [PATCH 2/2] gpio: tegra186: Support multiple interrupts per bank

[Bartosz Golaszewski]

On Fri, Sep 17, 2021 at 12:54 PM Thierry Reding
<thierry.reding@gmail.com> wrote:
>
> From: Thierry Reding <treding@nvidia.com>
>
> Tegra194 and later support more than a single interrupt per bank. This
> is primarily useful for virtualization but can also be helpful for more
> fine-grained CPU affinity control. To keep things simple for now, route
> all pins to the first interrupt.
>
> For backwards-compatibility, support old device trees that specify only
> one interrupt per bank by counting the interrupts at probe time.
>
> Signed-off-by: Thierry Reding <treding@nvidia.com>
> ---

Applied, thanks!

Bart