Skip to contentSkip to conversation

gpio: add interrupt handling capability to max732x

PATCH In review Marc Zyngier 2 participants 5 messages Mar 16, 2010 – Mar 22, 2010 GPIO

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 gpio: add interrupt handling capability to max732x

Eric,

Following on Jebediah's remark last month that max732x didn't support interrupt handling, I gave it a shot and hacked something that is similar to what pca953x does.

It would be nice if someone (Jebediah?) could give it a shot, as it's been only compile-tested (need to find some time this WE to solder the chips to a board...).

Thanks,

        M.
Signature · 1 line
Who you jivin' with that Cosmik Debris?

From e1677d4ac8aa7365a4c7ea27348f9e8a14baaf42 Mon Sep 17 00:00:00 2001
From: Marc Zyngier <[email protected]>
Date: Mon, 15 Mar 2010 12:24:36 +0000
Subject: [PATCH] gpio: add interrupt handling capability to max732x

Most the GPIO expander supported by the max732x driver have interrupt generation capability by reporting changes on input pins through an INT# pin. This patch implements the irq_chip functionnality (edge detection only).

---
 drivers/gpio/Kconfig        |    7 +
 drivers/gpio/max732x.c      |  355 ++++++++++++++++++++++++++++++++++++++++---
 include/linux/i2c/max732x.h |    3 +
 3 files changed, 346 insertions(+), 19 deletions(-)
drivers/gpio/Kconfig +7 -0
index fee678f..1aa63a2 100644
@@ -139,6 +139,13 @@ config GPIO_MAX732X
 	  Board setup code must specify the model to use, and the start
 	  number for these GPIOs.
 
+config GPIO_MAX732X_IRQ
+	bool "Interrupt controller support for MAX732x"
+	depends on GPIO_MAX732X=y && GENERIC_HARDIRQS
+	help
+	  Say yes here to enable the max732x to be used as an interrupt
+	  controller. It requires the driver to be built in the kernel.
+
 config GPIO_PCA953X
 	tristate "PCA953x, PCA955x, TCA64xx, and MAX7310 I/O ports"
 	depends on I2C
drivers/gpio/max732x.c +336 -19

436 lines not on this page · raw message

include/linux/i2c/max732x.h +3 -0
index e103366..c04bac8 100644
@@ -7,6 +7,9 @@ struct max732x_platform_data {
 	/* number of the first GPIO */
 	unsigned	gpio_base;
 
+	/* interrupt base */
+	int		irq_base;
+
 	void		*context;	/* param to setup/teardown */
 
 	int		(*setup)(struct i2c_client *client,
Signature · 1 line
1.7.0.2
#1 lore raw

drivers/gpio/max732x.c static int max732x_gpio_direction_output(struct gpio_chip *gc, · line 443

Quoted 10 lines · On Tue, 16 Mar 2010 11:34:52 +0100 Marc Zyngier <[email protected]> wrote:
drivers/gpio/max732x.c +10 -0
+static irqreturn_t max732x_irq_handler(int irq, void *devid)
+{
+	struct max732x_chip *chip = devid;
+	uint8_t pending;
+	uint8_t level;
+
+	pending = max732x_irq_pending(chip);
+
+	if (!pending)
+		return IRQ_HANDLED;

Should be IRQ_NONE?

If this device was on a shared interrupt line and the interrupt was caused by some other device, and this handler is called first then the incorrect IRQ_HANDLED return could cause the irq core to do wrong things.

Quoted 9 lines
drivers/gpio/max732x.c +9 -0
+	do {
+		level = __ffs(pending);
+		handle_nested_irq(level + chip->irq_base);
+
+		pending &= ~(1 << level);
+	} while (pending);
+
+	return IRQ_HANDLED;
+}
#2 lore raw
Quoted 15 lines · On Fri, 19 Mar 2010 16:41:56 -0700 Andrew Morton <[email protected]>…

drivers/gpio/max732x.c static int max732x_gpio_direction_output(struct gpio_chip *gc, · line 443

Quoted 10 lines · On Tue, 16 Mar 2010 11:34:52 +0100 Marc Zyngier <[email protected]> wrote:
drivers/gpio/max732x.c +10 -0
+static irqreturn_t max732x_irq_handler(int irq, void *devid)
+{
+	struct max732x_chip *chip = devid;
+	uint8_t pending;
+	uint8_t level;
+
+	pending = max732x_irq_pending(chip);
+
+	if (!pending)
+		return IRQ_HANDLED;

Should be IRQ_NONE?

The main problem is that some of the supported expanders do not have a hardware interrupt mask, so the IRQ will fire on each input pin toggle, including the ones we don't care about (pca953x has the same problem).

Returning IRQ_NONE will cause the irq core to quickly disable the line. Not good either. I could add a test for the mask feature bit, and return IRQ_NONE only in that case.

Quoted 3 lines

If this device was on a shared interrupt line and the interrupt was caused by some other device, and this handler is called first then the incorrect IRQ_HANDLED return could cause the irq core to do wrong things.

This is exactly why the irq handler is requested without the SHARED flag:

+ ret = request_threaded_irq(client->irq,

+					   NULL,
+					   max732x_irq_handler,
+					   IRQF_TRIGGER_FALLING | IRQF_ONESHOT,
+					   dev_name(&client->dev), chip);

The same mask feature bit could be used to allow shared interrupts with devices that support it.

Thanks,

	M.
Signature · 1 line
I'm the slime oozin' out from your TV set...
#3 lore raw
In reply to Andrew Morton
Quoted 16 lines · On Fri, 19 Mar 2010 16:41:56 -0700 Andrew Morton <[email protected]>…

Review

Quoted 10 lines · On Tue, 16 Mar 2010 11:34:52 +0100 Marc Zyngier <[email protected]> wrote:
(quoted diff) +10 -0
+static irqreturn_t max732x_irq_handler(int irq, void *devid)
+{
+	struct max732x_chip *chip = devid;
+	uint8_t pending;
+	uint8_t level;
+
+	pending = max732x_irq_pending(chip);
+
+	if (!pending)
+		return IRQ_HANDLED;

Should be IRQ_NONE?

Quoted 3 lines

If this device was on a shared interrupt line and the interrupt was caused by some other device, and this handler is called first then the incorrect IRQ_HANDLED return could cause the irq core to do wrong things.

Hi Andrew,

Would the following patch (against the original version) address your concerns?

Thanks,

	M.
Quoted 1 line

From a8d40580ad5fe23e0737bac5f8b04dfd8cb93fe5 Mon Sep 17 00:00:00 2001

From: Marc Zyngier <[email protected]>
Date: Mon, 22 Mar 2010 07:28:30 +0000
Subject: [PATCH] gpio: max732x: allow IRQF_SHARED when hardware mask is available

---
 drivers/gpio/max732x.c |   25 ++++++++++++++++++++++---
 1 files changed, 22 insertions(+), 3 deletions(-)
drivers/gpio/max732x.c +22 -3
index 647ae6f..07a9405 100644
@@ -293,6 +293,14 @@ static int max732x_readw(struct max732x_chip *chip, uint16_t *val)
 	return 0;
 }
 
+static inline int max732x_has_irqmask(struct max732x_chip *chip)
+{
+	if (chip->irq_features == INT_NO_MASK)
+		return 0;
+
+	return 1;
+}
+
 static void max732x_irq_update_mask(struct max732x_chip *chip)
 {
 	uint16_t msg;
@@ -302,7 +310,7 @@ static void max732x_irq_update_mask(struct max732x_chip *chip)
 
 	chip->irq_mask = chip->irq_mask_cur;
 
-	if (chip->irq_features == INT_NO_MASK)
+	if (!max732x_has_irqmask(chip))
 		return;
 
 	mutex_lock(&chip->lock);
@@ -439,8 +447,14 @@ static irqreturn_t max732x_irq_handler(int irq, void *devid)
 
 	pending = max732x_irq_pending(chip);
 
-	if (!pending)
+	if (!pending) {
+		/* if we have a hardware mask, then the interrupt must
+		 * come from another device */
+		if (max732x_has_irqmask(chip))
+			return IRQ_NONE;
+
 		return IRQ_HANDLED;
+	}
 
 	do {
 		level = __ffs(pending);
@@ -462,6 +476,7 @@ static int max732x_irq_setup(struct max732x_chip *chip,
 
 	if (pdata->irq_base && has_irq != INT_NONE) {
 		int lvl;
+		int irq_flags;
 
 		chip->irq_base = pdata->irq_base;
 		chip->irq_features = has_irq;
@@ -484,10 +499,14 @@ static int max732x_irq_setup(struct max732x_chip *chip,
 #endif
 		}
 
+		irq_flags = IRQF_TRIGGER_FALLING | IRQF_ONESHOT;
+		if (max732x_has_irqmask(chip))
+			irq_flags |= IRQF_SHARED;
+
 		ret = request_threaded_irq(client->irq,
 					   NULL,
 					   max732x_irq_handler,
-					   IRQF_TRIGGER_FALLING | IRQF_ONESHOT,
+					   irq_flags,
 					   dev_name(&client->dev), chip);
 		if (ret) {
 			dev_err(&client->dev, "failed to request irq %d\n",
Signature · 6 lines
1.7.0.2



-- 
I'm the slime oozin' out from your TV set...
#4 lore raw
Quoted 2 lines · On Mon, 22 Mar 2010 07:39:46 +0000 Marc Zyngier <[email protected]> wrote:

Would the following patch (against the original version) address your concerns?

Gee, I dunno. Is it necessary? I thought your earlier explanation covered everything OK.

I wouldn't say I had concerns really - I just poke people to verify that problems didn't sneak through. If you think it improves the driver, sure, add it?