Skip to contentSkip to conversation

pinctrl: Add pinctrl-s3c24xx driver

PATCH New Heiko Stübner 3 participants 10 messages Apr 9, 2013 – Apr 10, 2013

Subsystems (4)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 pinctrl: Add pinctrl-s3c24xx driver

The s3c24xx pins follow a similar pattern as the other Samsung SoCs and can therefore reuse the already introduced infrastructure.

The s3c24xx SoCs have one design oddity in that the first 4 external interrupts do not reside in the eint pending register but in the main interrupt controller instead. We solve this by forwarding the external interrupt from the main controller into the irq domain of the pin bank. The masking/acking of these interrupts is handled in the same way.

Furthermore the S3C2412/2413 SoCs contain another oddity in that they keep the same 4 eints in the main interrupt controller and eintpend register and requiring ack operations to happen in both. To solve this a ctrl_type enum is introduced which can keep the type of controller in the samsung_pin_ctrl struct for later retrieval.

The ctrl_type enum contains only S3C24XX and S3C2412 types, as the eint-speciality is currently the only use-case. But it can be expaned if other SoCs gain special handling requirements later on.

---
Depends on the s3c64xx pinctrl work from Tomasz Figa.

It also does not yet contain the pin-definitions for all s3c24xx SoCs, as I don't have datasheets for them.

Tested on a s3c2416 based board.

 .../bindings/pinctrl/samsung-pinctrl.txt           |    4 +
 drivers/gpio/gpio-samsung.c                        |    4 +
 drivers/pinctrl/Kconfig                            |    5 +
 drivers/pinctrl/Makefile                           |    1 +
 drivers/pinctrl/pinctrl-s3c24xx.c                  |  603 ++++++++++++++++++++
 drivers/pinctrl/pinctrl-samsung.c                  |   10 +
 drivers/pinctrl/pinctrl-samsung.h                  |   16 +
 7 files changed, 643 insertions(+), 0 deletions(-)
 create mode 100644 drivers/pinctrl/pinctrl-s3c24xx.c
Documentation/devicetree/bindings/pinctrl/samsung-pinctrl.txt +4 -0
index c70fca1..1d8fc3c 100644
@@ -7,6 +7,10 @@ on-chip controllers onto these pads.
 
 Required Properties:
 - compatible: should be one of the following.
+  - "samsung,s3c2413-pinctrl": for S3C64xx-compatible pin-controller,
+  - "samsung,s3c2416-pinctrl": for S3C64xx-compatible pin-controller,
+  - "samsung,s3c2440-pinctrl": for S3C64xx-compatible pin-controller,
+  - "samsung,s3c2450-pinctrl": for S3C64xx-compatible pin-controller,
   - "samsung,s3c64xx-pinctrl": for S3C64xx-compatible pin-controller,
   - "samsung,exynos4210-pinctrl": for Exynos4210 compatible pin-controller.
   - "samsung,exynos4x12-pinctrl": for Exynos4x12 compatible pin-controller.
drivers/gpio/gpio-samsung.c +4 -0
index dc06a6f..73017b9 100644
@@ -3026,6 +3026,10 @@ static __init int samsung_gpiolib_init(void)
 	*/
 	struct device_node *pctrl_np;
 	static const struct of_device_id exynos_pinctrl_ids[] = {
+		{ .compatible = "samsung,s3c2413-pinctrl", },
+		{ .compatible = "samsung,s3c2416-pinctrl", },
+		{ .compatible = "samsung,s3c2440-pinctrl", },
+		{ .compatible = "samsung,s3c2450-pinctrl", },
 		{ .compatible = "samsung,s3c64xx-pinctrl", },
 		{ .compatible = "samsung,exynos4210-pinctrl", },
 		{ .compatible = "samsung,exynos4x12-pinctrl", },
drivers/pinctrl/Kconfig +5 -0
index 7402ac9..58d73ac 100644
@@ -226,6 +226,11 @@ config PINCTRL_EXYNOS5440
 	select PINMUX
 	select PINCONF
 
+config PINCTRL_S3C24XX
+	bool "Samsung S3C24XX SoC pinctrl driver"
+	depends on ARCH_S3C24XX
+	select PINCTRL_SAMSUNG
+
 config PINCTRL_S3C64XX
 	bool "Samsung S3C64XX SoC pinctrl driver"
 	depends on ARCH_S3C64XX
drivers/pinctrl/Makefile +1 -0
index 21d34c2..1ccdfd8 100644
@@ -45,6 +45,7 @@ obj-$(CONFIG_PINCTRL_COH901)	+= pinctrl-coh901.o
 obj-$(CONFIG_PINCTRL_SAMSUNG)	+= pinctrl-samsung.o
 obj-$(CONFIG_PINCTRL_EXYNOS)	+= pinctrl-exynos.o
 obj-$(CONFIG_PINCTRL_EXYNOS5440)	+= pinctrl-exynos5440.o
+obj-$(CONFIG_PINCTRL_S3C24XX)	+= pinctrl-s3c24xx.o
 obj-$(CONFIG_PINCTRL_S3C64XX)	+= pinctrl-s3c64xx.o
 obj-$(CONFIG_PINCTRL_XWAY)	+= pinctrl-xway.o
 obj-$(CONFIG_PINCTRL_LANTIQ)	+= pinctrl-lantiq.o
drivers/pinctrl/pinctrl-s3c24xx.c new file +603 -0

603 lines not on this page · raw message

drivers/pinctrl/pinctrl-samsung.c +10 -0
index e836135..234c32f 100644
@@ -977,6 +977,16 @@ static const struct of_device_id samsung_pinctrl_dt_match[] = {
 	{ .compatible = "samsung,s3c64xx-pinctrl",
 		.data = s3c64xx_pin_ctrl },
 #endif
+#ifdef CONFIG_PINCTRL_S3C24XX
+	{ .compatible = "samsung,s3c2413-pinctrl",
+		.data = s3c2413_pin_ctrl },
+	{ .compatible = "samsung,s3c2416-pinctrl",
+		.data = s3c2416_pin_ctrl },
+	{ .compatible = "samsung,s3c2440-pinctrl",
+		.data = s3c2440_pin_ctrl },
+	{ .compatible = "samsung,s3c2450-pinctrl",
+		.data = s3c2450_pin_ctrl },
+#endif
 	{},
 };
 MODULE_DEVICE_TABLE(of, samsung_pinctrl_dt_match);
drivers/pinctrl/pinctrl-samsung.h +16 -0
index 7c7f9eb..8bd317c 100644
@@ -81,6 +81,16 @@ enum eint_type {
 	EINT_TYPE_WKUP_MUX,
 };
 
+/**
+ * enum ctrl_type - variances of a pin controller type
+ * @S3C24XX: generic s3c24xx pin handling
+ * @S3C2412: s3c2412-type eints, existing in the main and eint controllers
+ */
+enum ctrl_type {
+	S3C24XX,
+	S3C2412,
+};
+
 /* maximum length of a pin in pin descriptor (example: "gpa0-0") */
 #define PIN_NAME_LENGTH	10
 
@@ -164,6 +174,7 @@ struct samsung_pin_bank {
  * @eint_wkup_init: platform specific callback to setup the external wakeup
  *	interrupts for the controller.
  * @label: for debug information.
+ * @type: type of pin-controller
  */
 struct samsung_pin_ctrl {
 	struct samsung_pin_bank	*pin_banks;
@@ -185,6 +196,7 @@ struct samsung_pin_ctrl {
 	int		(*eint_gpio_init)(struct samsung_pinctrl_drv_data *);
 	int		(*eint_wkup_init)(struct samsung_pinctrl_drv_data *);
 	char		*label;
+	enum ctrl_type	type;
 };
 
 /**
@@ -246,5 +258,9 @@ extern struct samsung_pin_ctrl exynos4210_pin_ctrl[];
 extern struct samsung_pin_ctrl exynos4x12_pin_ctrl[];
 extern struct samsung_pin_ctrl exynos5250_pin_ctrl[];
 extern struct samsung_pin_ctrl s3c64xx_pin_ctrl[];
+extern struct samsung_pin_ctrl s3c2413_pin_ctrl[];
+extern struct samsung_pin_ctrl s3c2416_pin_ctrl[];
+extern struct samsung_pin_ctrl s3c2440_pin_ctrl[];
+extern struct samsung_pin_ctrl s3c2450_pin_ctrl[];
 
 #endif /* __PINCTRL_SAMSUNG_H */
Signature · 1 line
1.7.2.3
#1 lore raw

PATCH 1/1 pinctrl: Add pinctrl-s3c24xx driver

Quoted 16 lines · Heiko St?bner

The s3c24xx pins follow a similar pattern as the other Samsung SoCs and can therefore reuse the already introduced infrastructure.

The s3c24xx SoCs have one design oddity in that the first 4 external interrupts do not reside in the eint pending register but in the main interrupt controller instead. We solve this by forwarding the external interrupt from the main controller into the irq domain of the pin bank. The masking/acking of these interrupts is handled in the same way.

Furthermore the S3C2412/2413 SoCs contain another oddity in that they keep the same 4 eints in the main interrupt controller and eintpend register and requiring ack operations to happen in both. To solve this a ctrl_type enum is introduced which can keep the type of controller in the samsung_pin_ctrl struct for later retrieval.

Quoted 5 lines

The ctrl_type enum contains only S3C24XX and S3C2412 types, as the eint-speciality is currently the only use-case. But it can be expaned if other SoCs gain special handling requirements later on.

Looks good to me, need to implement more for other s3c24xx though.

Linus, if you want, please add:

Thanks.

- Kukjin

#2 lore raw

PATCH 1/1 pinctrl: Add pinctrl-s3c24xx driver

In reply to Heiko Stübner

Hi Heiko,

Basically looks good to me, but please see my inline comments about handling of EINT0-3.

drivers/pinctrl/Makefile obj-$(CONFIG_PINCTRL_COH901) += pinctrl-

Quoted 92 lines · On Wednesday 10 of April 2013 01:35:12 Heiko St?bner wrote:

The s3c24xx pins follow a similar pattern as the other Samsung SoCs and can therefore reuse the already introduced infrastructure.

The s3c24xx SoCs have one design oddity in that the first 4 external interrupts do not reside in the eint pending register but in the main interrupt controller instead. We solve this by forwarding the external interrupt from the main controller into the irq domain of the pin bank. The masking/acking of these interrupts is handled in the same way.

Furthermore the S3C2412/2413 SoCs contain another oddity in that they keep the same 4 eints in the main interrupt controller and eintpend register and requiring ack operations to happen in both. To solve this a ctrl_type enum is introduced which can keep the type of controller in the samsung_pin_ctrl struct for later retrieval.

The ctrl_type enum contains only S3C24XX and S3C2412 types, as the eint-speciality is currently the only use-case. But it can be expaned if other SoCs gain special handling requirements later on.

---
Depends on the s3c64xx pinctrl work from Tomasz Figa.

It also does not yet contain the pin-definitions for all s3c24xx SoCs, as I don't have datasheets for them.

Tested on a s3c2416 based board.

 .../bindings/pinctrl/samsung-pinctrl.txt           |    4 +
 drivers/gpio/gpio-samsung.c                        |    4 +
 drivers/pinctrl/Kconfig                            |    5 +
 drivers/pinctrl/Makefile                           |    1 +
 drivers/pinctrl/pinctrl-s3c24xx.c                  |  603
++++++++++++++++++++ drivers/pinctrl/pinctrl-samsung.c                 
|   10 +
 drivers/pinctrl/pinctrl-samsung.h                  |   16 +
 7 files changed, 643 insertions(+), 0 deletions(-)
 create mode 100644 drivers/pinctrl/pinctrl-s3c24xx.c

diff --git a/Documentation/devicetree/bindings/pinctrl/samsung-pinctrl.txt b/Documentation/devicetree/bindings/pinctrl/samsung-pinctrl.txt index c70fca1..1d8fc3c 100644

Documentation/devicetree/bindings/pinctrl/samsung-pinctrl.txt +4 -0
@@ -7,6 +7,10 @@ on-chip controllers onto these pads.
 
 Required Properties:
 - compatible: should be one of the following.
+  - "samsung,s3c2413-pinctrl": for S3C64xx-compatible pin-controller,
+  - "samsung,s3c2416-pinctrl": for S3C64xx-compatible pin-controller,
+  - "samsung,s3c2440-pinctrl": for S3C64xx-compatible pin-controller,
+  - "samsung,s3c2450-pinctrl": for S3C64xx-compatible pin-controller,
   - "samsung,s3c64xx-pinctrl": for S3C64xx-compatible pin-controller,
   - "samsung,exynos4210-pinctrl": for Exynos4210 compatible

pin-controller. - "samsung,exynos4x12-pinctrl": for Exynos4x12 compatible pin-controller. diff --git a/drivers/gpio/gpio-samsung.c b/drivers/gpio/gpio-samsung.c index dc06a6f..73017b9 100644

drivers/gpio/gpio-samsung.c +4 -0
@@ -3026,6 +3026,10 @@ static __init int samsung_gpiolib_init(void)
 	*/
 	struct device_node *pctrl_np;
 	static const struct of_device_id exynos_pinctrl_ids[] = {
+		{ .compatible = "samsung,s3c2413-pinctrl", },
+		{ .compatible = "samsung,s3c2416-pinctrl", },
+		{ .compatible = "samsung,s3c2440-pinctrl", },
+		{ .compatible = "samsung,s3c2450-pinctrl", },
 		{ .compatible = "samsung,s3c64xx-pinctrl", },
 		{ .compatible = "samsung,exynos4210-pinctrl", },
 		{ .compatible = "samsung,exynos4x12-pinctrl", },
drivers/pinctrl/Kconfig +5 -0
index 7402ac9..58d73ac 100644
@@ -226,6 +226,11 @@ config PINCTRL_EXYNOS5440
 	select PINMUX
 	select PINCONF
 
+config PINCTRL_S3C24XX
+	bool "Samsung S3C24XX SoC pinctrl driver"
+	depends on ARCH_S3C24XX
+	select PINCTRL_SAMSUNG
+
 config PINCTRL_S3C64XX
 	bool "Samsung S3C64XX SoC pinctrl driver"
 	depends on ARCH_S3C64XX
drivers/pinctrl/Makefile +0 -0
index 21d34c2..1ccdfd8 100644
@@ -45,6 +45,7 @@ obj-$(CONFIG_PINCTRL_COH901)	+= pinctrl-

coh901.o

Quoted 109 lines
drivers/pinctrl/Makefile +1 -0
 obj-$(CONFIG_PINCTRL_SAMSUNG)	+= pinctrl-samsung.o
 obj-$(CONFIG_PINCTRL_EXYNOS)	+= pinctrl-exynos.o
 obj-$(CONFIG_PINCTRL_EXYNOS5440)	+= pinctrl-exynos5440.o
+obj-$(CONFIG_PINCTRL_S3C24XX)	+= pinctrl-s3c24xx.o
 obj-$(CONFIG_PINCTRL_S3C64XX)	+= pinctrl-s3c64xx.o
 obj-$(CONFIG_PINCTRL_XWAY)	+= pinctrl-xway.o
 obj-$(CONFIG_PINCTRL_LANTIQ)	+= pinctrl-lantiq.o
(quoted diff) +0 -0

b/drivers/pinctrl/pinctrl-s3c24xx.c new file mode 100644 index 0000000..6b05519

drivers/pinctrl/pinctrl-s3c24xx.c new file +2 -0
@@ -0,0 +1,603 @@
+/*
+ * Exynos specific support for Samsung pinctrl/gpiolib driver with eint

support. + *
+ * Copyright (c) 2012 Samsung Electronics Co., Ltd.
+ * http://www.samsung.com
+ * Copyright (c) 2012 Linaro Ltd
+ * http://www.linaro.org
+ *
+ * Author: Thomas Abraham <[email protected]>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by + * the Free Software Foundation; either version 2 of the License, or + * (at your option) any later version.
+ *
+ * This file contains the Samsung Exynos specific information required by the + * the Samsung pinctrl/gpiolib driver. It also includes the implementation of + * external gpio and wakeup interrupt support.
+ */

drivers/pinctrl/pinctrl-s3c24xx.c new file +77 -0
+
+#include <linux/module.h>
+#include <linux/device.h>
+#include <linux/interrupt.h>
+#include <linux/irqdomain.h>
+#include <linux/irq.h>
+#include <linux/of_irq.h>
+#include <linux/io.h>
+#include <linux/slab.h>
+#include <linux/err.h>
+
+#include <asm/mach/irq.h>
+
+#include "pinctrl-samsung.h"
+
+#define NUM_EINT	24
+#define NUM_EINT_IRQ	6
+#define EINT_MAX_PER_GROUP	8
+
+#define EINTPEND_REG	0xa8
+#define EINTMASK_REG	0xa4
+
+#define EINT_GROUP(i)		((int)((i) / EINT_MAX_PER_GROUP))
+#define EINT_REG(i)		((EINT_GROUP(i) * 4) + 0x88)
+#define EINT_OFFS(i)		((i) % EINT_MAX_PER_GROUP * 4)
+
+#define EINT_LEVEL_LOW		0
+#define EINT_LEVEL_HIGH		1
+#define EINT_EDGE_FALLING	2
+#define EINT_EDGE_RISING	4
+#define EINT_EDGE_BOTH		6
+#define EINT_MASK		0xf
+
+static struct samsung_pin_bank_type bank_type_1bit = {
+	.fld_width = { 1, 1, },
+	.reg_offset = { 0x00, 0x04, },
+};
+
+static struct samsung_pin_bank_type bank_type_2bit = {
+	.fld_width = { 2, 1, 2, },
+	.reg_offset = { 0x00, 0x04, 0x08, },
+};
+
+#define PIN_BANK_A(pins, reg, id)		\
+	{						\
+		.type		= &bank_type_1bit,	\
+		.pctl_offset	= reg,			\
+		.nr_pins	= pins,			\
+		.eint_type	= EINT_TYPE_NONE,	\
+		.name		= id			\
+	}
+
+#define PIN_BANK_2BIT(pins, reg, id)		\
+	{						\
+		.type		= &bank_type_2bit,	\
+		.pctl_offset	= reg,			\
+		.nr_pins	= pins,			\
+		.eint_type	= EINT_TYPE_NONE,	\
+		.name		= id			\
+	}
+
+#define PIN_BANK_2BIT_EINTW(pins, reg, id, eoffs, emask)\
+	{						\
+		.type		= &bank_type_2bit,	\
+		.pctl_offset	= reg,			\
+		.nr_pins	= pins,			\
+		.eint_type	= EINT_TYPE_WKUP,	\
+		.eint_func	= 2,			\
+		.eint_mask	= emask,		\
+		.eint_offset	= eoffs,		\
+		.name		= id			\
+	}
+
+/**
+ * struct s3c24xx_eint_data: EINT common data
+ * @drvdata: pin controller driver data
+ * @domains: IRQ domains of particular EINT interrupts

drivers/pinctrl/pinctrl-s3c24xx.c line 106

Quoted 10 lines
drivers/pinctrl/pinctrl-s3c24xx.c new file +10 -0
+ */
+struct s3c24xx_eint_data {
+	struct samsung_pinctrl_drv_data *drvdata;
+	struct irq_domain *domains[NUM_EINT];
+	int parents[NUM_EINT_IRQ];
+};
+
+struct s3c24xx_eint_domain_data {
+	struct samsung_pin_bank *bank;
+	struct s3c24xx_eint_data *eint_data;

What about:

+ bool eint0_3_parent_only;

(or whatever name would be more appropriate), which would store the information about the s3c24xx-specific quirk in 24xx-specific data structure, without the need to add another field to the generic samsung_pinctrl_drv_data structure?

See my further comments on how I would see using this field in interrupt handling code.

Quoted 39 lines
drivers/pinctrl/pinctrl-s3c24xx.c new file +30 -0
+};
+
+static int s3c24xx_eint_get_trigger(unsigned int type)
+{
+	int trigger;
+
+	switch (type) {
+	case IRQ_TYPE_EDGE_RISING:
+		trigger = EINT_EDGE_RISING;
+		break;
+	case IRQ_TYPE_EDGE_FALLING:
+		trigger = EINT_EDGE_FALLING;
+		break;
+	case IRQ_TYPE_EDGE_BOTH:
+		trigger = EINT_EDGE_BOTH;
+		break;
+	case IRQ_TYPE_LEVEL_HIGH:
+		trigger = EINT_LEVEL_HIGH;
+		break;
+	case IRQ_TYPE_LEVEL_LOW:
+		trigger = EINT_LEVEL_LOW;
+		break;
+	default:
+		return -EINVAL;
+	}
+
+	return trigger;
+}
+
+static void s3c24xx_eint_set_handler(unsigned int irq, unsigned int

type) +{

drivers/pinctrl/pinctrl-s3c24xx.c new file +8 -0
+	/* Edge- and level-triggered interrupts need different handlers */
+	if (type & IRQ_TYPE_EDGE_BOTH)
+		__irq_set_handler_locked(irq, handle_edge_irq);
+	else
+		__irq_set_handler_locked(irq, handle_level_irq);
+}
+
+static void s3c24xx_eint_set_function(struct samsung_pinctrl_drv_data
Quoted 1 line

*d, + struct samsung_pin_bank *bank, int

pin)

Quoted 44 lines
drivers/pinctrl/pinctrl-s3c24xx.c new file +44 -0
+{
+	struct samsung_pin_bank_type *bank_type = bank->type;
+	unsigned long flags;
+	void __iomem *reg;
+	u8 shift;
+	u32 mask;
+	u32 val;
+
+	/* Make sure that pin is configured as interrupt */
+	reg = d->virt_base + bank->pctl_offset;
+	shift = pin * bank_type->fld_width[PINCFG_TYPE_FUNC];
+	mask = (1 << bank_type->fld_width[PINCFG_TYPE_FUNC]) - 1;
+
+	spin_lock_irqsave(&bank->slock, flags);
+
+	val = readl(reg);
+	val &= ~(mask << shift);
+	val |= bank->eint_func << shift;
+	writel(val, reg);
+
+	spin_unlock_irqrestore(&bank->slock, flags);
+}
+
+static int s3c24xx_eint_type(struct irq_data *data, unsigned int type)
+{
+	struct samsung_pin_bank *bank = irq_data_get_irq_chip_data(data);
+	struct samsung_pinctrl_drv_data *d = bank->drvdata;
+	int index = bank->eint_offset + data->hwirq;
+	void __iomem *reg;
+	int trigger;
+	u8 shift;
+	u32 val;
+
+	trigger = s3c24xx_eint_get_trigger(type);
+	if (trigger < 0) {
+		dev_err(d->dev, "unsupported external interrupt type\n");
+		return -EINVAL;
+	}
+
+	s3c24xx_eint_set_handler(data->irq, type);
+
+	/* Set up interrupt trigger */
+	reg = d->virt_base + EINT_REG(index);
+	shift = EINT_OFFS(index);

drivers/pinctrl/pinctrl-s3c24xx.c line 200

Quoted 10 lines
drivers/pinctrl/pinctrl-s3c24xx.c new file +10 -0
+
+	val = readl(reg);
+	val &= ~(EINT_MASK << shift);
+	val |= trigger << shift;
+	writel(val, reg);
+
+	s3c24xx_eint_set_function(d, bank, data->hwirq);
+
+	return 0;
+}

Now I would split the following 3 functions into two sets of 3 functions, one set for s3c2412 and other for remaining SoCs and make separate EINT0-3 IRQ chips for both cases.

Quoted 8 lines
drivers/pinctrl/pinctrl-s3c24xx.c new file +6 -0
+
+static void s3c24xx_eint0_3_ack(struct irq_data *data)
+{
+	struct samsung_pin_bank *bank = irq_data_get_irq_chip_data(data);
+	struct samsung_pinctrl_drv_data *d = bank->drvdata;
+	struct s3c24xx_eint_domain_data *ddata = bank->irq_domain-

host_data;

drivers/pinctrl/pinctrl-s3c24xx.c new file +1 -0
+	struct s3c24xx_eint_data *eint_data = ddata->eint_data;

drivers/pinctrl/pinctrl-s3c24xx.c line 217

Quoted 10 lines
drivers/pinctrl/pinctrl-s3c24xx.c new file +10 -0
+	int parent_irq = eint_data->parents[data->hwirq];
+	struct irq_chip *parent_chip = irq_get_chip(parent_irq);
+
+	if (d->ctrl->type == S3C2412) {
+		unsigned long bitval = 1UL << data->hwirq;
+		writel(bitval, d->virt_base + EINTPEND_REG);
+	}
+
+	if (parent_chip->irq_ack)
+		parent_chip->irq_ack(irq_get_irq_data(parent_irq));

Btw. Is this parent level acking really needed here?

Quoted 95 lines
drivers/pinctrl/pinctrl-s3c24xx.c new file +7 -0
+}
+
+static void s3c24xx_eint0_3_mask(struct irq_data *data)
+{
+	struct samsung_pin_bank *bank = irq_data_get_irq_chip_data(data);
+	struct samsung_pinctrl_drv_data *d = bank->drvdata;
+	struct s3c24xx_eint_domain_data *ddata = bank->irq_domain-

host_data;

drivers/pinctrl/pinctrl-s3c24xx.c new file +20 -0
+	struct s3c24xx_eint_data *eint_data = ddata->eint_data;
+	int parent_irq = eint_data->parents[data->hwirq];
+	struct irq_chip *parent_chip = irq_get_chip(parent_irq);
+
+	if (d->ctrl->type == S3C2412) {
+		unsigned long mask;
+		mask = readl(d->virt_base + EINTMASK_REG);
+		mask |= (1UL << data->hwirq);
+		writel(mask, d->virt_base + EINTMASK_REG);
+	}
+
+	if (parent_chip->irq_mask)
+		parent_chip->irq_mask(irq_get_irq_data(parent_irq));
+}
+
+static void s3c24xx_eint0_3_unmask(struct irq_data *data)
+{
+	struct samsung_pin_bank *bank = irq_data_get_irq_chip_data(data);
+	struct samsung_pinctrl_drv_data *d = bank->drvdata;
+	struct s3c24xx_eint_domain_data *ddata = bank->irq_domain-

host_data;

drivers/pinctrl/pinctrl-s3c24xx.c new file +65 -0
+	struct s3c24xx_eint_data *eint_data = ddata->eint_data;
+	int parent_irq = eint_data->parents[data->hwirq];
+	struct irq_chip *parent_chip = irq_get_chip(parent_irq);
+
+	if (d->ctrl->type == S3C2412) {
+		unsigned long mask;
+		mask = readl(d->virt_base + EINTMASK_REG);
+		mask &= ~(1UL << data->hwirq);
+		writel(mask, d->virt_base + EINTMASK_REG);
+	}
+
+	if (parent_chip->irq_unmask)
+		parent_chip->irq_unmask(irq_get_irq_data(parent_irq));
+}
+
+static struct irq_chip s3c24xx_eint0_3_chip = {
+	.name		= "s3c-eint0_3",
+	.irq_ack	= s3c24xx_eint0_3_ack,
+	.irq_mask	= s3c24xx_eint0_3_mask,
+	.irq_unmask	= s3c24xx_eint0_3_unmask,
+	.irq_set_type	= s3c24xx_eint_type,
+};
+
+static void s3c24xx_eint_ack(struct irq_data *data)
+{
+	struct samsung_pin_bank *bank = irq_data_get_irq_chip_data(data);
+	struct samsung_pinctrl_drv_data *d = bank->drvdata;
+	unsigned char index = bank->eint_offset + data->hwirq;
+
+	writel(1UL << index, d->virt_base + EINTPEND_REG);
+}
+
+static void s3c24xx_eint_mask(struct irq_data *data)
+{
+	struct samsung_pin_bank *bank = irq_data_get_irq_chip_data(data);
+	struct samsung_pinctrl_drv_data *d = bank->drvdata;
+	unsigned char index = bank->eint_offset + data->hwirq;
+	unsigned long mask;
+
+	mask = readl(d->virt_base + EINTMASK_REG);
+	mask |= (1UL << index);
+	writel(mask, d->virt_base + EINTMASK_REG);
+}
+
+static void s3c24xx_eint_unmask(struct irq_data *data)
+{
+	struct samsung_pin_bank *bank = irq_data_get_irq_chip_data(data);
+	struct samsung_pinctrl_drv_data *d = bank->drvdata;
+	unsigned char index = bank->eint_offset + data->hwirq;
+	unsigned long mask;
+
+	mask = readl(d->virt_base + EINTMASK_REG);
+	mask &= ~(1UL << index);
+	writel(mask, d->virt_base + EINTMASK_REG);
+}
+
+static struct irq_chip s3c24xx_eint_chip = {
+	.name		= "s3c-eint",
+	.irq_ack	= s3c24xx_eint_ack,
+	.irq_mask	= s3c24xx_eint_mask,
+	.irq_unmask	= s3c24xx_eint_unmask,
+	.irq_set_type	= s3c24xx_eint_type,
+};
+
+static void s3c24xx_demux_eint0_3(unsigned int irq, struct irq_desc

*desc) +{

drivers/pinctrl/pinctrl-s3c24xx.c line 314

Quoted 4 lines
drivers/pinctrl/pinctrl-s3c24xx.c new file +4 -0
+	struct irq_data *data = irq_desc_get_irq_data(desc);
+	struct s3c24xx_eint_data *eint_data = irq_get_handler_data(irq);
+	unsigned int virq;
+

Instead of acking the interrupt at parent chip from ack callback of EINT0_3 chip, I would rather use chained_irq_enter() here...

Quoted 3 lines
drivers/pinctrl/pinctrl-s3c24xx.c new file +2 -0
+	/* the first 4 eints have a simple 1 to 1 mapping */
+	virq = irq_linear_revmap(eint_data->domains[data->hwirq],

data->hwirq); + /* Something must be really wrong if an unmapped

EINT

drivers/pinctrl/pinctrl-s3c24xx.c line 322

Quoted 5 lines
drivers/pinctrl/pinctrl-s3c24xx.c new file +5 -0
+	 * was unmasked...
+	 */
+	BUG_ON(!virq);
+
+	generic_handle_irq(virq);

...and chained_irq_exit() here.

Quoted 14 lines
drivers/pinctrl/pinctrl-s3c24xx.c new file +3 -0
+}
+
+static inline void s3c24xx_demux_eint(unsigned int irq, struct irq_desc

*desc, + u32 offset, u32 range)

drivers/pinctrl/pinctrl-s3c24xx.c new file +10 -0
+{
+	struct irq_chip *chip = irq_get_chip(irq);
+	struct s3c24xx_eint_data *data = irq_get_handler_data(irq);
+	struct samsung_pinctrl_drv_data *d = data->drvdata;
+	unsigned int pend, mask;
+
+	chained_irq_enter(chip, desc);
+
+	pend = readl(d->virt_base + EINTPEND_REG);
+	mask = readl(d->virt_base + EINTMASK_REG);

drivers/pinctrl/pinctrl-s3c24xx.c

Quoted 10 lines
drivers/pinctrl/pinctrl-s3c24xx.c new file +10 -0
+
+	pend &= ~mask;
+	pend &= range;
+
+	while (pend) {
+		unsigned int virq;
+
+		irq = __ffs(pend);
+		pend &= ~(1 << irq);
+		virq = irq_linear_revmap(data->domains[irq], irq - 

offset);

Quoted 30 lines
drivers/pinctrl/pinctrl-s3c24xx.c new file +12 -0
+		/* Something must be really wrong if an unmapped EINT
+		 * was unmasked...
+		 */
+		BUG_ON(!virq);
+
+		generic_handle_irq(virq);
+	}
+
+	chained_irq_exit(chip, desc);
+}
+
+static void s3c24xx_demux_eint4_7(unsigned int irq, struct irq_desc

*desc) +{

drivers/pinctrl/pinctrl-s3c24xx.c new file +4 -0
+	s3c24xx_demux_eint(irq, desc, 0, 0xf0);
+}
+
+static void s3c24xx_demux_eint8_23(unsigned int irq, struct irq_desc

*desc) +{

drivers/pinctrl/pinctrl-s3c24xx.c new file +12 -0
+	s3c24xx_demux_eint(irq, desc, 8, 0xffff00);
+}
+
+static irq_flow_handler_t s3c24xx_eint_handlers[NUM_EINT_IRQ] = {
+	s3c24xx_demux_eint0_3,
+	s3c24xx_demux_eint0_3,
+	s3c24xx_demux_eint0_3,
+	s3c24xx_demux_eint0_3,
+	s3c24xx_demux_eint4_7,
+	s3c24xx_demux_eint8_23,
+};
+

drivers/pinctrl/pinctrl-s3c24xx.c line 386

Quoted 10 lines
drivers/pinctrl/pinctrl-s3c24xx.c new file +10 -0
+static int s3c24xx_gpf_irq_map(struct irq_domain *h, unsigned int virq,
+					irq_hw_number_t hw)
+{
+	struct s3c24xx_eint_domain_data *ddata = h->host_data;
+	struct samsung_pin_bank *bank = ddata->bank;
+
+	if (!(bank->eint_mask & (1 << (bank->eint_offset + hw))))
+		return -EINVAL;
+
+	if (hw <= 3) {

Now here I would check ddata->eint0_3_parent_only and set the chip conditionally, either to s3c24xx or s3c2412 variant.

Quoted 17 lines
drivers/pinctrl/pinctrl-s3c24xx.c new file +17 -0
+		irq_set_chip_and_handler(virq, &s3c24xx_eint0_3_chip,
+					 handle_edge_irq);
+		irq_set_chip_data(virq, bank);
+	} else {
+		irq_set_chip_and_handler(virq, &s3c24xx_eint_chip,
+					 handle_edge_irq);
+		irq_set_chip_data(virq, bank);
+	}
+	set_irq_flags(virq, IRQF_VALID);
+	return 0;
+}
+
+static const struct irq_domain_ops s3c24xx_gpf_irq_ops = {
+	.map	= s3c24xx_gpf_irq_map,
+	.xlate	= irq_domain_xlate_twocell,
+};
+

drivers/pinctrl/pinctrl-s3c24xx.c

Quoted 10 lines
drivers/pinctrl/pinctrl-s3c24xx.c new file +10 -0
+static int s3c24xx_gpg_irq_map(struct irq_domain *h, unsigned int virq,
+					irq_hw_number_t hw)
+{
+	struct s3c24xx_eint_domain_data *ddata = h->host_data;
+	struct samsung_pin_bank *bank = ddata->bank;
+
+	if (!(bank->eint_mask & (1 << (bank->eint_offset + hw))))
+		return -EINVAL;
+
+	irq_set_chip_and_handler(virq, &s3c24xx_eint_chip, 

handle_edge_irq);

Quoted 27 lines
drivers/pinctrl/pinctrl-s3c24xx.c new file +27 -0
+	irq_set_chip_data(virq, bank);
+	set_irq_flags(virq, IRQF_VALID);
+	return 0;
+}
+
+static const struct irq_domain_ops s3c24xx_gpg_irq_ops = {
+	.map	= s3c24xx_gpg_irq_map,
+	.xlate	= irq_domain_xlate_twocell,
+};
+
+static const struct of_device_id s3c24xx_eint_irq_ids[] = {
+	{ .compatible = "samsung,s3c24xx-wakeup-eint", },
+	{ }
+};
+
+static int s3c24xx_eint_init(struct samsung_pinctrl_drv_data *d)
+{
+	struct device *dev = d->dev;
+	struct device_node *eint_np = NULL;
+	struct device_node *np;
+	struct samsung_pin_bank *bank;
+	struct s3c24xx_eint_data *eint_data;
+	const struct irq_domain_ops *ops;
+	unsigned int i;
+
+	for_each_child_of_node(dev->of_node, np) {
+		if (of_match_node(s3c24xx_eint_irq_ids, np)) {

drivers/pinctrl/pinctrl-s3c24xx.c

Quoted 10 lines
drivers/pinctrl/pinctrl-s3c24xx.c new file +10 -0
+			eint_np = np;
+			break;
+		}
+	}
+	if (!eint_np)
+		return -ENODEV;
+
+	eint_data = devm_kzalloc(dev, sizeof(*eint_data), GFP_KERNEL);
+	if (!eint_data) {
+		dev_err(dev, "could not allocate memory for wkup eint 

data\n");

drivers/pinctrl/pinctrl-s3c24xx.c

Quoted 10 lines
drivers/pinctrl/pinctrl-s3c24xx.c new file +10 -0
+		return -ENOMEM;
+	}
+	eint_data->drvdata = d;
+
+	for (i = 0; i < NUM_EINT_IRQ; ++i) {
+		unsigned int irq;
+
+		irq = irq_of_parse_and_map(eint_np, i);
+		if (!irq) {
+			dev_err(dev, "failed to get wakeup EINT IRQ %d\n", 

i);

Quoted 15 lines
drivers/pinctrl/pinctrl-s3c24xx.c new file +15 -0
+			return -ENXIO;
+		}
+
+		eint_data->parents[i] = irq;
+		irq_set_chained_handler(irq, s3c24xx_eint_handlers[i]);
+		irq_set_handler_data(irq, eint_data);
+	}
+
+	bank = d->ctrl->pin_banks;
+	for (i = 0; i < d->ctrl->nr_banks; ++i, ++bank) {
+		struct s3c24xx_eint_domain_data *ddata;
+		unsigned int mask;
+		unsigned int irq;
+		unsigned int pin;
+

drivers/pinctrl/pinctrl-s3c24xx.c line 485

Quoted 10 lines
drivers/pinctrl/pinctrl-s3c24xx.c new file +10 -0
+		if (bank->eint_type != EINT_TYPE_WKUP)
+			continue;
+
+		ddata = devm_kzalloc(dev, sizeof(*ddata), GFP_KERNEL);
+		if (!ddata) {
+			dev_err(dev, "failed to allocate domain data\n");
+			return -ENOMEM;
+		}
+		ddata->bank = bank;
+		ddata->eint_data = eint_data;

Here the eint0_3_parent_only field can be set based on compatible string of the EINT controller.

Quoted 35 lines
drivers/pinctrl/pinctrl-s3c24xx.c new file +35 -0
+
+		ops = (bank->eint_offset == 0) ? &s3c24xx_gpf_irq_ops
+					       : &s3c24xx_gpg_irq_ops;
+
+		bank->irq_domain = irq_domain_add_linear(bank->of_node,
+				bank->nr_pins, ops, ddata);
+		if (!bank->irq_domain) {
+			dev_err(dev, "wkup irq domain add failed\n");
+			return -ENXIO;
+		}
+
+		irq = bank->eint_offset;
+		mask = bank->eint_mask;
+		for (pin = 0; mask; ++pin, mask >>= 1) {
+			if (irq > NUM_EINT)
+				break;
+			if (!(mask & 1))
+				continue;
+			eint_data->domains[irq] = bank->irq_domain;
+			++irq;
+		}
+	}
+
+	return 0;
+}
+
+static struct samsung_pin_bank s3c2413_pin_banks[] = {
+	PIN_BANK_A(23, 0x000, "gpa"),
+	PIN_BANK_2BIT(11, 0x010, "gpb"),
+	PIN_BANK_2BIT(16, 0x020, "gpc"),
+	PIN_BANK_2BIT(16, 0x030, "gpd"),
+	PIN_BANK_2BIT(16, 0x040, "gpe"),
+	PIN_BANK_2BIT_EINTW(8, 0x050, "gpf", 0, 0xff),
+	PIN_BANK_2BIT_EINTW(16, 0x060, "gpg", 8, 0xffff00),
+	PIN_BANK_2BIT(11, 0x070, "gph"),

drivers/pinctrl/pinctrl-s3c24xx.c line 530

Quoted 10 lines
drivers/pinctrl/pinctrl-s3c24xx.c new file +10 -0
+	PIN_BANK_2BIT(13, 0x080, "gpj"),
+};
+
+struct samsung_pin_ctrl s3c2413_pin_ctrl[] = {
+	{
+		.pin_banks	= s3c2413_pin_banks,
+		.nr_banks	= ARRAY_SIZE(s3c2413_pin_banks),
+		.eint_wkup_init = s3c24xx_eint_init,
+		.label		= "S3C2413-GPIO",
+		.type		= S3C2412,

And then finally you shouldn't need this type field anymore.

Best regards,
Tomasz

Quoted 125 lines
drivers/pinctrl/pinctrl-s3c24xx.c new file +73 -0
+	},
+};
+
+static struct samsung_pin_bank s3c2416_pin_banks[] = {
+	PIN_BANK_A(27, 0x000, "gpa"),
+	PIN_BANK_2BIT(11, 0x010, "gpb"),
+	PIN_BANK_2BIT(16, 0x020, "gpc"),
+	PIN_BANK_2BIT(16, 0x030, "gpd"),
+	PIN_BANK_2BIT(16, 0x040, "gpe"),
+	PIN_BANK_2BIT_EINTW(8, 0x050, "gpf", 0, 0xff),
+	PIN_BANK_2BIT_EINTW(8, 0x060, "gpg", 8, 0xff00),
+	PIN_BANK_2BIT(15, 0x070, "gph"),
+	PIN_BANK_2BIT(16, 0x0e0, "gpk"),
+	PIN_BANK_2BIT(14, 0x0f0, "gpl"),
+	PIN_BANK_2BIT(2, 0x100, "gpm"),
+};
+
+struct samsung_pin_ctrl s3c2416_pin_ctrl[] = {
+	{
+		.pin_banks	= s3c2416_pin_banks,
+		.nr_banks	= ARRAY_SIZE(s3c2416_pin_banks),
+		.eint_wkup_init = s3c24xx_eint_init,
+		.label		= "S3C2416-GPIO",
+		.type		= S3C24XX,
+	},
+};
+
+static struct samsung_pin_bank s3c2440_pin_banks[] = {
+	PIN_BANK_A(25, 0x000, "gpa"),
+	PIN_BANK_2BIT(11, 0x010, "gpb"),
+	PIN_BANK_2BIT(16, 0x020, "gpc"),
+	PIN_BANK_2BIT(16, 0x030, "gpd"),
+	PIN_BANK_2BIT(16, 0x040, "gpe"),
+	PIN_BANK_2BIT_EINTW(8, 0x050, "gpf", 0, 0xff),
+	PIN_BANK_2BIT_EINTW(16, 0x060, "gpg", 8, 0xffff00),
+	PIN_BANK_2BIT(11, 0x070, "gph"),
+	PIN_BANK_2BIT(13, 0x0d0, "gpj"),
+};
+
+struct samsung_pin_ctrl s3c2440_pin_ctrl[] = {
+	{
+		.pin_banks	= s3c2440_pin_banks,
+		.nr_banks	= ARRAY_SIZE(s3c2440_pin_banks),
+		.eint_wkup_init = s3c24xx_eint_init,
+		.label		= "S3C2440-GPIO",
+		.type		= S3C24XX,
+	},
+};
+
+static struct samsung_pin_bank s3c2450_pin_banks[] = {
+	PIN_BANK_A(28, 0x000, "gpa"),
+	PIN_BANK_2BIT(11, 0x010, "gpb"),
+	PIN_BANK_2BIT(16, 0x020, "gpc"),
+	PIN_BANK_2BIT(16, 0x030, "gpd"),
+	PIN_BANK_2BIT(16, 0x040, "gpe"),
+	PIN_BANK_2BIT_EINTW(8, 0x050, "gpf", 0, 0xff),
+	PIN_BANK_2BIT_EINTW(16, 0x060, "gpg", 8, 0xffff00),
+	PIN_BANK_2BIT(15, 0x070, "gph"),
+	PIN_BANK_2BIT(16, 0x0d0, "gpj"),
+	PIN_BANK_2BIT(16, 0x0e0, "gpk"),
+	PIN_BANK_2BIT(15, 0x0f0, "gpl"),
+	PIN_BANK_2BIT(2, 0x100, "gpm"),
+};
+
+struct samsung_pin_ctrl s3c2450_pin_ctrl[] = {
+	{
+		.pin_banks	= s3c2450_pin_banks,
+		.nr_banks	= ARRAY_SIZE(s3c2450_pin_banks),
+		.eint_wkup_init = s3c24xx_eint_init,
+		.label		= "S3C2450-GPIO",
+		.type		= S3C24XX,
+	},
+};
(quoted diff) +0 -0

b/drivers/pinctrl/pinctrl-samsung.c index e836135..234c32f 100644

drivers/pinctrl/pinctrl-samsung.c +0 -0
@@ -977,6 +977,16 @@ static const struct of_device_id

samsung_pinctrl_dt_match[] = { { .compatible =
"samsung,s3c64xx-pinctrl",

drivers/pinctrl/pinctrl-samsung.c +10 -0
 		.data = s3c64xx_pin_ctrl },
 #endif
+#ifdef CONFIG_PINCTRL_S3C24XX
+	{ .compatible = "samsung,s3c2413-pinctrl",
+		.data = s3c2413_pin_ctrl },
+	{ .compatible = "samsung,s3c2416-pinctrl",
+		.data = s3c2416_pin_ctrl },
+	{ .compatible = "samsung,s3c2440-pinctrl",
+		.data = s3c2440_pin_ctrl },
+	{ .compatible = "samsung,s3c2450-pinctrl",
+		.data = s3c2450_pin_ctrl },
+#endif
 	{},
 };
 MODULE_DEVICE_TABLE(of, samsung_pinctrl_dt_match);
(quoted diff) +0 -0

b/drivers/pinctrl/pinctrl-samsung.h index 7c7f9eb..8bd317c 100644

drivers/pinctrl/pinctrl-samsung.h +4 -0
@@ -81,6 +81,16 @@ enum eint_type {
 	EINT_TYPE_WKUP_MUX,
 };
 
+/**
+ * enum ctrl_type - variances of a pin controller type
+ * @S3C24XX: generic s3c24xx pin handling
+ * @S3C2412: s3c2412-type eints, existing in the main and eint

controllers + */

drivers/pinctrl/pinctrl-samsung.h +5 -0
+enum ctrl_type {
+	S3C24XX,
+	S3C2412,
+};
+
 /* maximum length of a pin in pin descriptor (example: "gpa0-0") */
 #define PIN_NAME_LENGTH	10
@@ -164,6 +174,7 @@ struct samsung_pin_bank {
  * @eint_wkup_init: platform specific callback to setup the external

wakeup * interrupts for the controller.
* @label: for debug information.
+ * @type: type of pin-controller
*/
struct samsung_pin_ctrl {

drivers/pinctrl/pinctrl-samsung.h +0 -0
 	struct samsung_pin_bank	*pin_banks;

drivers/pinctrl/pinctrl-samsung.h struct samsung_pin_ctrl { · line 196

Quoted 1 line
drivers/pinctrl/pinctrl-samsung.h +0 -0
@@ -185,6 +196,7 @@ struct samsung_pin_ctrl {
 	int		(*eint_gpio_init)(struct samsung_pinctrl_drv_data 

*);

drivers/pinctrl/pinctrl-samsung.h struct samsung_pin_ctrl { · line 197

Quoted 1 line
drivers/pinctrl/pinctrl-samsung.h +0 -0
 	int		(*eint_wkup_init)(struct samsung_pinctrl_drv_data 

*);

Quoted 16 lines
drivers/pinctrl/pinctrl-samsung.h +1 -0
 	char		*label;
+	enum ctrl_type	type;
 };
 
 /**
@@ -246,5 +258,9 @@ extern struct samsung_pin_ctrl

exynos4210_pin_ctrl[]; extern struct samsung_pin_ctrl exynos4x12_pin_ctrl[];
extern struct samsung_pin_ctrl exynos5250_pin_ctrl[]; extern struct samsung_pin_ctrl s3c64xx_pin_ctrl[];

drivers/pinctrl/pinctrl-samsung.h +4 -0
+extern struct samsung_pin_ctrl s3c2413_pin_ctrl[];
+extern struct samsung_pin_ctrl s3c2416_pin_ctrl[];
+extern struct samsung_pin_ctrl s3c2440_pin_ctrl[];
+extern struct samsung_pin_ctrl s3c2450_pin_ctrl[];
 
 #endif /* __PINCTRL_SAMSUNG_H */
#3 lore raw

PATCH 1/1 pinctrl: Add pinctrl-s3c24xx driver

Hi Tomasz,

thanks for your comments, more inline.

Am Mittwoch, 10. April 2013, 12:36:39 schrieb Tomasz Figa:

Quoted 8 lines

Hi Heiko,

Basically looks good to me, but please see my inline comments about handling of EINT0-3.

Quoted 2 lines · On Wednesday 10 of April 2013 01:35:12 Heiko St?bner wrote:

The s3c24xx pins follow a similar pattern as the other Samsung SoCs and can therefore reuse the already introduced infrastructure.

[...]

Quoted 14 lines

drivers/pinctrl/pinctrl-s3c24xx.c line 106

Quoted 9 lines
drivers/pinctrl/pinctrl-s3c24xx.c new file +9 -0
+struct s3c24xx_eint_data {
+	struct samsung_pinctrl_drv_data *drvdata;
+	struct irq_domain *domains[NUM_EINT];
+	int parents[NUM_EINT_IRQ];
+};
+
+struct s3c24xx_eint_domain_data {
+	struct samsung_pin_bank *bank;
+	struct s3c24xx_eint_data *eint_data;

What about:

(quoted diff) +1 -0
+	bool eint0_3_parent_only;
Quoted 7 lines

(or whatever name would be more appropriate), which would store the information about the s3c24xx-specific quirk in 24xx-specific data structure, without the need to add another field to the generic samsung_pinctrl_drv_data structure?

See my further comments on how I would see using this field in interrupt handling code.

ok, sounds good, especially gathering the type from the wakeup-int property

Quoted 1 line
Quoted 1 line

+};

[...]

Quoted 3 lines

Now I would split the following 3 functions into two sets of 3 functions, one set for s3c2412 and other for remaining SoCs and make separate EINT0-3 IRQ chips for both cases.

Not doing the decision every time, might bring some very slight speed improvements, so is probably the right way to go.

Quoted 9 lines
Quoted 9 lines
drivers/pinctrl/pinctrl-s3c24xx.c new file +6 -0
+
+static void s3c24xx_eint0_3_ack(struct irq_data *data)
+{
+	struct samsung_pin_bank *bank = irq_data_get_irq_chip_data(data);
+	struct samsung_pinctrl_drv_data *d = bank->drvdata;
+	struct s3c24xx_eint_domain_data *ddata = bank->irq_domain-

host_data;

Quoted 13 lines

drivers/pinctrl/pinctrl-s3c24xx.c line 217

Quoted 11 lines
drivers/pinctrl/pinctrl-s3c24xx.c new file +11 -0
+	struct s3c24xx_eint_data *eint_data = ddata->eint_data;
+	int parent_irq = eint_data->parents[data->hwirq];
+	struct irq_chip *parent_chip = irq_get_chip(parent_irq);
+
+	if (d->ctrl->type == S3C2412) {
+		unsigned long bitval = 1UL << data->hwirq;
+		writel(bitval, d->virt_base + EINTPEND_REG);
+	}
+
+	if (parent_chip->irq_ack)
+		parent_chip->irq_ack(irq_get_irq_data(parent_irq));

Btw. Is this parent level acking really needed here?

Depends. If using chained_irq_* of course not, but if the irq-handler should stay in charge of when to ack it might be better this way.

Generic s3c24xx SoCs need acking in the main controller only, while s3c2412 needs acking in both the main controller and eintpend.

Quoted 1 line
Quoted 1 line

+}

[...]

Quoted 16 lines

drivers/pinctrl/pinctrl-s3c24xx.c line 314

Quoted 6 lines

+static void s3c24xx_demux_eint0_3(unsigned int irq, struct irq_desc *desc) +{

drivers/pinctrl/pinctrl-s3c24xx.c new file +4 -0
+	struct irq_data *data = irq_desc_get_irq_data(desc);
+	struct s3c24xx_eint_data *eint_data = irq_get_handler_data(irq);
+	unsigned int virq;
+

Instead of acking the interrupt at parent chip from ack callback of EINT0_3 chip, I would rather use chained_irq_enter() here...

Quoted 3 lines
(quoted diff) +2 -0
+	/* the first 4 eints have a simple 1 to 1 mapping */
+	virq = irq_linear_revmap(eint_data->domains[data->hwirq],

data->hwirq); + /* Something must be really wrong if an unmapped

EINT

Quoted 7 lines

drivers/pinctrl/pinctrl-s3c24xx.c line 322

Quoted 5 lines
drivers/pinctrl/pinctrl-s3c24xx.c new file +5 -0
+	 * was unmasked...
+	 */
+	BUG_ON(!virq);
+
+	generic_handle_irq(virq);

...and chained_irq_exit() here.

If I understand it correctly, the way chained_irq_* works it would limit the eints to a level style handling. With the way it's currently the whole determination of when to ack,mask and unmask is completely for the real handler (edge or level) to decide, as the original interrupt gets completely forwarded into the irq-domain without further constraints.

So, after the change on regular s3c24xx SoCs when the real irq handler wants to ack the irq, it would be a no-op, as it would already have been acked by chained_irq_enter.

Masking might be even more interesting. Currently control is transfered completely to the pinctrl irq-domain, which then controls the masking of the interrupt thru the parent-calls - on regular s3c24xx the masking of these is also only done in the main controller.

When using chained_irq_* it also wants to mask the interrupt which might conflict with regular enable_irq/disable_irq calls being done for example in driver code.

So in short I agree with the earlier split of the irqchip, but would keep the irq operations themself centralized in the pinctrl driver, instead of using chained_irq_* functions.

Heiko

Quoted 230 lines
Quoted 14 lines
drivers/pinctrl/pinctrl-s3c24xx.c new file +3 -0
+}
+
+static inline void s3c24xx_demux_eint(unsigned int irq, struct irq_desc

*desc, + u32 offset, u32 range)

drivers/pinctrl/pinctrl-s3c24xx.c new file +10 -0
+{
+	struct irq_chip *chip = irq_get_chip(irq);
+	struct s3c24xx_eint_data *data = irq_get_handler_data(irq);
+	struct samsung_pinctrl_drv_data *d = data->drvdata;
+	unsigned int pend, mask;
+
+	chained_irq_enter(chip, desc);
+
+	pend = readl(d->virt_base + EINTPEND_REG);
+	mask = readl(d->virt_base + EINTMASK_REG);

drivers/pinctrl/pinctrl-s3c24xx.c

Quoted 10 lines
drivers/pinctrl/pinctrl-s3c24xx.c new file +10 -0
+
+	pend &= ~mask;
+	pend &= range;
+
+	while (pend) {
+		unsigned int virq;
+
+		irq = __ffs(pend);
+		pend &= ~(1 << irq);
+		virq = irq_linear_revmap(data->domains[irq], irq -

offset);

Quoted 30 lines
drivers/pinctrl/pinctrl-s3c24xx.c new file +12 -0
+		/* Something must be really wrong if an unmapped EINT
+		 * was unmasked...
+		 */
+		BUG_ON(!virq);
+
+		generic_handle_irq(virq);
+	}
+
+	chained_irq_exit(chip, desc);
+}
+
+static void s3c24xx_demux_eint4_7(unsigned int irq, struct irq_desc

*desc) +{

drivers/pinctrl/pinctrl-s3c24xx.c new file +4 -0
+	s3c24xx_demux_eint(irq, desc, 0, 0xf0);
+}
+
+static void s3c24xx_demux_eint8_23(unsigned int irq, struct irq_desc

*desc) +{

drivers/pinctrl/pinctrl-s3c24xx.c new file +12 -0
+	s3c24xx_demux_eint(irq, desc, 8, 0xffff00);
+}
+
+static irq_flow_handler_t s3c24xx_eint_handlers[NUM_EINT_IRQ] = {
+	s3c24xx_demux_eint0_3,
+	s3c24xx_demux_eint0_3,
+	s3c24xx_demux_eint0_3,
+	s3c24xx_demux_eint0_3,
+	s3c24xx_demux_eint4_7,
+	s3c24xx_demux_eint8_23,
+};
+

drivers/pinctrl/pinctrl-s3c24xx.c line 386

Quoted 10 lines
drivers/pinctrl/pinctrl-s3c24xx.c new file +10 -0
+static int s3c24xx_gpf_irq_map(struct irq_domain *h, unsigned int virq,
+					irq_hw_number_t hw)
+{
+	struct s3c24xx_eint_domain_data *ddata = h->host_data;
+	struct samsung_pin_bank *bank = ddata->bank;
+
+	if (!(bank->eint_mask & (1 << (bank->eint_offset + hw))))
+		return -EINVAL;
+
+	if (hw <= 3) {

Now here I would check ddata->eint0_3_parent_only and set the chip conditionally, either to s3c24xx or s3c2412 variant.

Quoted 17 lines
drivers/pinctrl/pinctrl-s3c24xx.c new file +17 -0
+		irq_set_chip_and_handler(virq, &s3c24xx_eint0_3_chip,
+					 handle_edge_irq);
+		irq_set_chip_data(virq, bank);
+	} else {
+		irq_set_chip_and_handler(virq, &s3c24xx_eint_chip,
+					 handle_edge_irq);
+		irq_set_chip_data(virq, bank);
+	}
+	set_irq_flags(virq, IRQF_VALID);
+	return 0;
+}
+
+static const struct irq_domain_ops s3c24xx_gpf_irq_ops = {
+	.map	= s3c24xx_gpf_irq_map,
+	.xlate	= irq_domain_xlate_twocell,
+};
+

drivers/pinctrl/pinctrl-s3c24xx.c

Quoted 10 lines
drivers/pinctrl/pinctrl-s3c24xx.c new file +10 -0
+static int s3c24xx_gpg_irq_map(struct irq_domain *h, unsigned int virq,
+					irq_hw_number_t hw)
+{
+	struct s3c24xx_eint_domain_data *ddata = h->host_data;
+	struct samsung_pin_bank *bank = ddata->bank;
+
+	if (!(bank->eint_mask & (1 << (bank->eint_offset + hw))))
+		return -EINVAL;
+
+	irq_set_chip_and_handler(virq, &s3c24xx_eint_chip,

handle_edge_irq);

Quoted 27 lines
drivers/pinctrl/pinctrl-s3c24xx.c new file +27 -0
+	irq_set_chip_data(virq, bank);
+	set_irq_flags(virq, IRQF_VALID);
+	return 0;
+}
+
+static const struct irq_domain_ops s3c24xx_gpg_irq_ops = {
+	.map	= s3c24xx_gpg_irq_map,
+	.xlate	= irq_domain_xlate_twocell,
+};
+
+static const struct of_device_id s3c24xx_eint_irq_ids[] = {
+	{ .compatible = "samsung,s3c24xx-wakeup-eint", },
+	{ }
+};
+
+static int s3c24xx_eint_init(struct samsung_pinctrl_drv_data *d)
+{
+	struct device *dev = d->dev;
+	struct device_node *eint_np = NULL;
+	struct device_node *np;
+	struct samsung_pin_bank *bank;
+	struct s3c24xx_eint_data *eint_data;
+	const struct irq_domain_ops *ops;
+	unsigned int i;
+
+	for_each_child_of_node(dev->of_node, np) {
+		if (of_match_node(s3c24xx_eint_irq_ids, np)) {

drivers/pinctrl/pinctrl-s3c24xx.c

Quoted 10 lines
drivers/pinctrl/pinctrl-s3c24xx.c new file +10 -0
+			eint_np = np;
+			break;
+		}
+	}
+	if (!eint_np)
+		return -ENODEV;
+
+	eint_data = devm_kzalloc(dev, sizeof(*eint_data), GFP_KERNEL);
+	if (!eint_data) {
+		dev_err(dev, "could not allocate memory for wkup eint

data\n");

drivers/pinctrl/pinctrl-s3c24xx.c

Quoted 10 lines
drivers/pinctrl/pinctrl-s3c24xx.c new file +10 -0
+		return -ENOMEM;
+	}
+	eint_data->drvdata = d;
+
+	for (i = 0; i < NUM_EINT_IRQ; ++i) {
+		unsigned int irq;
+
+		irq = irq_of_parse_and_map(eint_np, i);
+		if (!irq) {
+			dev_err(dev, "failed to get wakeup EINT IRQ %d\n",

i);

Quoted 15 lines
drivers/pinctrl/pinctrl-s3c24xx.c new file +15 -0
+			return -ENXIO;
+		}
+
+		eint_data->parents[i] = irq;
+		irq_set_chained_handler(irq, s3c24xx_eint_handlers[i]);
+		irq_set_handler_data(irq, eint_data);
+	}
+
+	bank = d->ctrl->pin_banks;
+	for (i = 0; i < d->ctrl->nr_banks; ++i, ++bank) {
+		struct s3c24xx_eint_domain_data *ddata;
+		unsigned int mask;
+		unsigned int irq;
+		unsigned int pin;
+

drivers/pinctrl/pinctrl-s3c24xx.c line 485

Quoted 10 lines
drivers/pinctrl/pinctrl-s3c24xx.c new file +10 -0
+		if (bank->eint_type != EINT_TYPE_WKUP)
+			continue;
+
+		ddata = devm_kzalloc(dev, sizeof(*ddata), GFP_KERNEL);
+		if (!ddata) {
+			dev_err(dev, "failed to allocate domain data\n");
+			return -ENOMEM;
+		}
+		ddata->bank = bank;
+		ddata->eint_data = eint_data;

Here the eint0_3_parent_only field can be set based on compatible string of the EINT controller.

Quoted 35 lines
drivers/pinctrl/pinctrl-s3c24xx.c new file +35 -0
+
+		ops = (bank->eint_offset == 0) ? &s3c24xx_gpf_irq_ops
+					       : &s3c24xx_gpg_irq_ops;
+
+		bank->irq_domain = irq_domain_add_linear(bank->of_node,
+				bank->nr_pins, ops, ddata);
+		if (!bank->irq_domain) {
+			dev_err(dev, "wkup irq domain add failed\n");
+			return -ENXIO;
+		}
+
+		irq = bank->eint_offset;
+		mask = bank->eint_mask;
+		for (pin = 0; mask; ++pin, mask >>= 1) {
+			if (irq > NUM_EINT)
+				break;
+			if (!(mask & 1))
+				continue;
+			eint_data->domains[irq] = bank->irq_domain;
+			++irq;
+		}
+	}
+
+	return 0;
+}
+
+static struct samsung_pin_bank s3c2413_pin_banks[] = {
+	PIN_BANK_A(23, 0x000, "gpa"),
+	PIN_BANK_2BIT(11, 0x010, "gpb"),
+	PIN_BANK_2BIT(16, 0x020, "gpc"),
+	PIN_BANK_2BIT(16, 0x030, "gpd"),
+	PIN_BANK_2BIT(16, 0x040, "gpe"),
+	PIN_BANK_2BIT_EINTW(8, 0x050, "gpf", 0, 0xff),
+	PIN_BANK_2BIT_EINTW(16, 0x060, "gpg", 8, 0xffff00),
+	PIN_BANK_2BIT(11, 0x070, "gph"),

drivers/pinctrl/pinctrl-s3c24xx.c line 530

Quoted 10 lines
drivers/pinctrl/pinctrl-s3c24xx.c new file +10 -0
+	PIN_BANK_2BIT(13, 0x080, "gpj"),
+};
+
+struct samsung_pin_ctrl s3c2413_pin_ctrl[] = {
+	{
+		.pin_banks	= s3c2413_pin_banks,
+		.nr_banks	= ARRAY_SIZE(s3c2413_pin_banks),
+		.eint_wkup_init = s3c24xx_eint_init,
+		.label		= "S3C2413-GPIO",
+		.type		= S3C2412,

And then finally you shouldn't need this type field anymore.

#4 lore raw

PATCH 1/1 pinctrl: Add pinctrl-s3c24xx driver

Quoted 125 lines · On Wednesday 10 of April 2013 14:20:22 Heiko St?bner wrote:

Hi Tomasz,

thanks for your comments, more inline.

Am Mittwoch, 10. April 2013, 12:36:39 schrieb Tomasz Figa:

Quoted 9 lines

Hi Heiko,

Basically looks good to me, but please see my inline comments about handling of EINT0-3.

Quoted 3 lines · On Wednesday 10 of April 2013 01:35:12 Heiko St?bner wrote:

The s3c24xx pins follow a similar pattern as the other Samsung SoCs and
can therefore reuse the already introduced infrastructure.

[...]

Quoted 14 lines

drivers/pinctrl/pinctrl-s3c24xx.c line 106

Quoted 9 lines
drivers/pinctrl/pinctrl-s3c24xx.c new file +9 -0
+struct s3c24xx_eint_data {
+	struct samsung_pinctrl_drv_data *drvdata;
+	struct irq_domain *domains[NUM_EINT];
+	int parents[NUM_EINT_IRQ];
+};
+
+struct s3c24xx_eint_domain_data {
+	struct samsung_pin_bank *bank;
+	struct s3c24xx_eint_data *eint_data;

What about:

(quoted diff) +1 -0
+	bool eint0_3_parent_only;
Quoted 7 lines

(or whatever name would be more appropriate), which would store the information about the s3c24xx-specific quirk in 24xx-specific data structure, without the need to add another field to the generic samsung_pinctrl_drv_data structure?

See my further comments on how I would see using this field in interrupt handling code.

ok, sounds good, especially gathering the type from the wakeup-int property

Quoted 1 line
Quoted 1 line

+};

[...]

Quoted 3 lines

Now I would split the following 3 functions into two sets of 3 functions, one set for s3c2412 and other for remaining SoCs and make separate EINT0-3 IRQ chips for both cases.

Not doing the decision every time, might bring some very slight speed improvements, so is probably the right way to go.

Quoted 9 lines
Quoted 9 lines
drivers/pinctrl/pinctrl-s3c24xx.c new file +6 -0
+
+static void s3c24xx_eint0_3_ack(struct irq_data *data)
+{
+	struct samsung_pin_bank *bank = irq_data_get_irq_chip_data(data);
+	struct samsung_pinctrl_drv_data *d = bank->drvdata;
+	struct s3c24xx_eint_domain_data *ddata = bank->irq_domain-

host_data;

Quoted 13 lines

drivers/pinctrl/pinctrl-s3c24xx.c line 217

Quoted 11 lines
drivers/pinctrl/pinctrl-s3c24xx.c new file +11 -0
+	struct s3c24xx_eint_data *eint_data = ddata->eint_data;
+	int parent_irq = eint_data->parents[data->hwirq];
+	struct irq_chip *parent_chip = irq_get_chip(parent_irq);
+
+	if (d->ctrl->type == S3C2412) {
+		unsigned long bitval = 1UL << data->hwirq;
+		writel(bitval, d->virt_base + EINTPEND_REG);
+	}
+
+	if (parent_chip->irq_ack)
+		parent_chip->irq_ack(irq_get_irq_data(parent_irq));

Btw. Is this parent level acking really needed here?

Depends. If using chained_irq_* of course not, but if the irq-handler should stay in charge of when to ack it might be better this way.

Generic s3c24xx SoCs need acking in the main controller only, while s3c2412 needs acking in both the main controller and eintpend.

Quoted 1 line
Quoted 1 line

+}

[...]

Quoted 16 lines

drivers/pinctrl/pinctrl-s3c24xx.c line 314

Quoted 6 lines

+static void s3c24xx_demux_eint0_3(unsigned int irq, struct irq_desc *desc) +{

drivers/pinctrl/pinctrl-s3c24xx.c new file +4 -0
+	struct irq_data *data = irq_desc_get_irq_data(desc);
+	struct s3c24xx_eint_data *eint_data = irq_get_handler_data(irq);
+	unsigned int virq;
+

Instead of acking the interrupt at parent chip from ack callback of EINT0_3 chip, I would rather use chained_irq_enter() here...

Quoted 3 lines
(quoted diff) +2 -0
+	/* the first 4 eints have a simple 1 to 1 mapping */
+	virq = irq_linear_revmap(eint_data->domains[data->hwirq],

data->hwirq); + /* Something must be really wrong if an unmapped

EINT

Quoted 7 lines

drivers/pinctrl/pinctrl-s3c24xx.c line 322

Quoted 5 lines
drivers/pinctrl/pinctrl-s3c24xx.c new file +5 -0
+	 * was unmasked...
+	 */
+	BUG_ON(!virq);
+
+	generic_handle_irq(virq);

...and chained_irq_exit() here.

If I understand it correctly, the way chained_irq_* works it would limit the eints to a level style handling. With the way it's currently the whole determination of when to ack,mask and unmask is completely for the real handler (edge or level) to decide, as the original interrupt gets completely forwarded into the irq-domain without further constraints.

So, after the change on regular s3c24xx SoCs when the real irq handler wants to ack the irq, it would be a no-op, as it would already have been acked by chained_irq_enter.

Masking might be even more interesting. Currently control is transfered completely to the pinctrl irq-domain, which then controls the masking of the interrupt thru the parent-calls - on regular s3c24xx the masking of these is also only done in the main controller.

Quoted 9 lines

When using chained_irq_* it also wants to mask the interrupt which might conflict with regular enable_irq/disable_irq calls being done for example in driver code.

So in short I agree with the earlier split of the irqchip, but would keep the irq operations themself centralized in the pinctrl driver, instead of using chained_irq_* functions.

Right, my solution wouldn't work properly in case of regular s3c24xx and edge triggered interrupts.

However I'm still wondering if it's OK to manually call parent chip operations in case of s3c2416. This would imply the same operation calling order as imposed by flow handler of the chained EINT (which can be handle_edge_irq), while the parent chip is probably level triggered, isn't it?

Best regards,
Tomasz

Quoted 7 lines

Heiko

Quoted 4 lines
Quoted 4 lines
drivers/pinctrl/pinctrl-s3c24xx.c new file +3 -0
+}
+
+static inline void s3c24xx_demux_eint(unsigned int irq, struct

irq_desc *desc, + u32 offset, u32

range)

Quoted 227 lines
Quoted 227 lines
Quoted 10 lines
drivers/pinctrl/pinctrl-s3c24xx.c new file +10 -0
+{
+	struct irq_chip *chip = irq_get_chip(irq);
+	struct s3c24xx_eint_data *data = irq_get_handler_data(irq);
+	struct samsung_pinctrl_drv_data *d = data->drvdata;
+	unsigned int pend, mask;
+
+	chained_irq_enter(chip, desc);
+
+	pend = readl(d->virt_base + EINTPEND_REG);
+	mask = readl(d->virt_base + EINTMASK_REG);

drivers/pinctrl/pinctrl-s3c24xx.c

Quoted 10 lines
drivers/pinctrl/pinctrl-s3c24xx.c new file +10 -0
+
+	pend &= ~mask;
+	pend &= range;
+
+	while (pend) {
+		unsigned int virq;
+
+		irq = __ffs(pend);
+		pend &= ~(1 << irq);
+		virq = irq_linear_revmap(data->domains[irq], irq -

offset);

Quoted 33 lines
drivers/pinctrl/pinctrl-s3c24xx.c new file +12 -0
+		/* Something must be really wrong if an unmapped EINT
+		 * was unmasked...
+		 */
+		BUG_ON(!virq);
+
+		generic_handle_irq(virq);
+	}
+
+	chained_irq_exit(chip, desc);
+}
+
+static void s3c24xx_demux_eint4_7(unsigned int irq, struct irq_desc

*desc) +{

drivers/pinctrl/pinctrl-s3c24xx.c new file +4 -0
+	s3c24xx_demux_eint(irq, desc, 0, 0xf0);
+}
+
+static void s3c24xx_demux_eint8_23(unsigned int irq, struct

irq_desc
*desc) +{

drivers/pinctrl/pinctrl-s3c24xx.c new file +13 -0
+	s3c24xx_demux_eint(irq, desc, 8, 0xffff00);
+}
+
+static irq_flow_handler_t s3c24xx_eint_handlers[NUM_EINT_IRQ] = {
+	s3c24xx_demux_eint0_3,
+	s3c24xx_demux_eint0_3,
+	s3c24xx_demux_eint0_3,
+	s3c24xx_demux_eint0_3,
+	s3c24xx_demux_eint4_7,
+	s3c24xx_demux_eint8_23,
+};
+
+static int s3c24xx_gpf_irq_map(struct irq_domain *h, unsigned int

virq, + irq_hw_number_t hw)

drivers/pinctrl/pinctrl-s3c24xx.c line 386

Quoted 8 lines
drivers/pinctrl/pinctrl-s3c24xx.c new file +8 -0
+{
+	struct s3c24xx_eint_domain_data *ddata = h->host_data;
+	struct samsung_pin_bank *bank = ddata->bank;
+
+	if (!(bank->eint_mask & (1 << (bank->eint_offset + hw))))
+		return -EINVAL;
+
+	if (hw <= 3) {

Now here I would check ddata->eint0_3_parent_only and set the chip conditionally, either to s3c24xx or s3c2412 variant.

Quoted 19 lines
drivers/pinctrl/pinctrl-s3c24xx.c new file +18 -0
+		irq_set_chip_and_handler(virq, &s3c24xx_eint0_3_chip,
+					 handle_edge_irq);
+		irq_set_chip_data(virq, bank);
+	} else {
+		irq_set_chip_and_handler(virq, &s3c24xx_eint_chip,
+					 handle_edge_irq);
+		irq_set_chip_data(virq, bank);
+	}
+	set_irq_flags(virq, IRQF_VALID);
+	return 0;
+}
+
+static const struct irq_domain_ops s3c24xx_gpf_irq_ops = {
+	.map	= s3c24xx_gpf_irq_map,
+	.xlate	= irq_domain_xlate_twocell,
+};
+
+static int s3c24xx_gpg_irq_map(struct irq_domain *h, unsigned int

virq, + irq_hw_number_t hw)

drivers/pinctrl/pinctrl-s3c24xx.c line 391

Quoted 8 lines
drivers/pinctrl/pinctrl-s3c24xx.c new file +8 -0
+{
+	struct s3c24xx_eint_domain_data *ddata = h->host_data;
+	struct samsung_pin_bank *bank = ddata->bank;
+
+	if (!(bank->eint_mask & (1 << (bank->eint_offset + hw))))
+		return -EINVAL;
+
+	irq_set_chip_and_handler(virq, &s3c24xx_eint_chip,

handle_edge_irq);

Quoted 27 lines
drivers/pinctrl/pinctrl-s3c24xx.c new file +27 -0
+	irq_set_chip_data(virq, bank);
+	set_irq_flags(virq, IRQF_VALID);
+	return 0;
+}
+
+static const struct irq_domain_ops s3c24xx_gpg_irq_ops = {
+	.map	= s3c24xx_gpg_irq_map,
+	.xlate	= irq_domain_xlate_twocell,
+};
+
+static const struct of_device_id s3c24xx_eint_irq_ids[] = {
+	{ .compatible = "samsung,s3c24xx-wakeup-eint", },
+	{ }
+};
+
+static int s3c24xx_eint_init(struct samsung_pinctrl_drv_data *d)
+{
+	struct device *dev = d->dev;
+	struct device_node *eint_np = NULL;
+	struct device_node *np;
+	struct samsung_pin_bank *bank;
+	struct s3c24xx_eint_data *eint_data;
+	const struct irq_domain_ops *ops;
+	unsigned int i;
+
+	for_each_child_of_node(dev->of_node, np) {
+		if (of_match_node(s3c24xx_eint_irq_ids, np)) {

drivers/pinctrl/pinctrl-s3c24xx.c

Quoted 10 lines
drivers/pinctrl/pinctrl-s3c24xx.c new file +10 -0
+			eint_np = np;
+			break;
+		}
+	}
+	if (!eint_np)
+		return -ENODEV;
+
+	eint_data = devm_kzalloc(dev, sizeof(*eint_data), GFP_KERNEL);
+	if (!eint_data) {
+		dev_err(dev, "could not allocate memory for wkup eint

data\n");

drivers/pinctrl/pinctrl-s3c24xx.c

Quoted 10 lines
drivers/pinctrl/pinctrl-s3c24xx.c new file +10 -0
+		return -ENOMEM;
+	}
+	eint_data->drvdata = d;
+
+	for (i = 0; i < NUM_EINT_IRQ; ++i) {
+		unsigned int irq;
+
+		irq = irq_of_parse_and_map(eint_np, i);
+		if (!irq) {
+			dev_err(dev, "failed to get wakeup EINT IRQ %d\n",

i);

Quoted 15 lines
drivers/pinctrl/pinctrl-s3c24xx.c new file +15 -0
+			return -ENXIO;
+		}
+
+		eint_data->parents[i] = irq;
+		irq_set_chained_handler(irq, s3c24xx_eint_handlers[i]);
+		irq_set_handler_data(irq, eint_data);
+	}
+
+	bank = d->ctrl->pin_banks;
+	for (i = 0; i < d->ctrl->nr_banks; ++i, ++bank) {
+		struct s3c24xx_eint_domain_data *ddata;
+		unsigned int mask;
+		unsigned int irq;
+		unsigned int pin;
+

drivers/pinctrl/pinctrl-s3c24xx.c line 485

Quoted 10 lines
drivers/pinctrl/pinctrl-s3c24xx.c new file +10 -0
+		if (bank->eint_type != EINT_TYPE_WKUP)
+			continue;
+
+		ddata = devm_kzalloc(dev, sizeof(*ddata), GFP_KERNEL);
+		if (!ddata) {
+			dev_err(dev, "failed to allocate domain data\n");
+			return -ENOMEM;
+		}
+		ddata->bank = bank;
+		ddata->eint_data = eint_data;

Here the eint0_3_parent_only field can be set based on compatible string of the EINT controller.

Quoted 35 lines
drivers/pinctrl/pinctrl-s3c24xx.c new file +35 -0
+
+		ops = (bank->eint_offset == 0) ? &s3c24xx_gpf_irq_ops
+					       : &s3c24xx_gpg_irq_ops;
+
+		bank->irq_domain = irq_domain_add_linear(bank->of_node,
+				bank->nr_pins, ops, ddata);
+		if (!bank->irq_domain) {
+			dev_err(dev, "wkup irq domain add failed\n");
+			return -ENXIO;
+		}
+
+		irq = bank->eint_offset;
+		mask = bank->eint_mask;
+		for (pin = 0; mask; ++pin, mask >>= 1) {
+			if (irq > NUM_EINT)
+				break;
+			if (!(mask & 1))
+				continue;
+			eint_data->domains[irq] = bank->irq_domain;
+			++irq;
+		}
+	}
+
+	return 0;
+}
+
+static struct samsung_pin_bank s3c2413_pin_banks[] = {
+	PIN_BANK_A(23, 0x000, "gpa"),
+	PIN_BANK_2BIT(11, 0x010, "gpb"),
+	PIN_BANK_2BIT(16, 0x020, "gpc"),
+	PIN_BANK_2BIT(16, 0x030, "gpd"),
+	PIN_BANK_2BIT(16, 0x040, "gpe"),
+	PIN_BANK_2BIT_EINTW(8, 0x050, "gpf", 0, 0xff),
+	PIN_BANK_2BIT_EINTW(16, 0x060, "gpg", 8, 0xffff00),
+	PIN_BANK_2BIT(11, 0x070, "gph"),

drivers/pinctrl/pinctrl-s3c24xx.c line 530

Quoted 10 lines
drivers/pinctrl/pinctrl-s3c24xx.c new file +10 -0
+	PIN_BANK_2BIT(13, 0x080, "gpj"),
+};
+
+struct samsung_pin_ctrl s3c2413_pin_ctrl[] = {
+	{
+		.pin_banks	= s3c2413_pin_banks,
+		.nr_banks	= ARRAY_SIZE(s3c2413_pin_banks),
+		.eint_wkup_init = s3c24xx_eint_init,
+		.label		= "S3C2413-GPIO",
+		.type		= S3C2412,

And then finally you shouldn't need this type field anymore.

#5 lore raw

PATCH 1/1 pinctrl: Add pinctrl-s3c24xx driver

Am Mittwoch, 10. April 2013, 14:31:29 schrieb Tomasz Figa:

Quoted 136 lines · On Wednesday 10 of April 2013 14:20:22 Heiko St?bner wrote:
Quoted 126 lines

Hi Tomasz,

thanks for your comments, more inline.

Am Mittwoch, 10. April 2013, 12:36:39 schrieb Tomasz Figa:

Quoted 9 lines

Hi Heiko,

Basically looks good to me, but please see my inline comments about handling of EINT0-3.

Quoted 3 lines · On Wednesday 10 of April 2013 01:35:12 Heiko St?bner wrote:

The s3c24xx pins follow a similar pattern as the other Samsung SoCs and
can therefore reuse the already introduced infrastructure.

[...]

Quoted 14 lines

drivers/pinctrl/pinctrl-s3c24xx.c line 106

Quoted 9 lines
drivers/pinctrl/pinctrl-s3c24xx.c new file +9 -0
+struct s3c24xx_eint_data {
+	struct samsung_pinctrl_drv_data *drvdata;
+	struct irq_domain *domains[NUM_EINT];
+	int parents[NUM_EINT_IRQ];
+};
+
+struct s3c24xx_eint_domain_data {
+	struct samsung_pin_bank *bank;
+	struct s3c24xx_eint_data *eint_data;

What about:

(quoted diff) +1 -0
+	bool eint0_3_parent_only;
Quoted 7 lines

(or whatever name would be more appropriate), which would store the information about the s3c24xx-specific quirk in 24xx-specific data structure, without the need to add another field to the generic samsung_pinctrl_drv_data structure?

See my further comments on how I would see using this field in interrupt handling code.

ok, sounds good, especially gathering the type from the wakeup-int property

Quoted 1 line
Quoted 1 line

+};

[...]

Quoted 3 lines

Now I would split the following 3 functions into two sets of 3 functions, one set for s3c2412 and other for remaining SoCs and make separate EINT0-3 IRQ chips for both cases.

Not doing the decision every time, might bring some very slight speed improvements, so is probably the right way to go.

Quoted 9 lines
Quoted 9 lines
drivers/pinctrl/pinctrl-s3c24xx.c new file +6 -0
+
+static void s3c24xx_eint0_3_ack(struct irq_data *data)
+{
+	struct samsung_pin_bank *bank = irq_data_get_irq_chip_data(data);
+	struct samsung_pinctrl_drv_data *d = bank->drvdata;
+	struct s3c24xx_eint_domain_data *ddata = bank->irq_domain-

host_data;

Quoted 13 lines

drivers/pinctrl/pinctrl-s3c24xx.c line 217

Quoted 11 lines
drivers/pinctrl/pinctrl-s3c24xx.c new file +11 -0
+	struct s3c24xx_eint_data *eint_data = ddata->eint_data;
+	int parent_irq = eint_data->parents[data->hwirq];
+	struct irq_chip *parent_chip = irq_get_chip(parent_irq);
+
+	if (d->ctrl->type == S3C2412) {
+		unsigned long bitval = 1UL << data->hwirq;
+		writel(bitval, d->virt_base + EINTPEND_REG);
+	}
+
+	if (parent_chip->irq_ack)
+		parent_chip->irq_ack(irq_get_irq_data(parent_irq));

Btw. Is this parent level acking really needed here?

Depends. If using chained_irq_* of course not, but if the irq-handler should stay in charge of when to ack it might be better this way.

Generic s3c24xx SoCs need acking in the main controller only, while s3c2412 needs acking in both the main controller and eintpend.

Quoted 1 line
Quoted 1 line

+}

[...]

Quoted 16 lines

drivers/pinctrl/pinctrl-s3c24xx.c line 314

Quoted 6 lines

+static void s3c24xx_demux_eint0_3(unsigned int irq, struct irq_desc *desc) +{

drivers/pinctrl/pinctrl-s3c24xx.c new file +4 -0
+	struct irq_data *data = irq_desc_get_irq_data(desc);
+	struct s3c24xx_eint_data *eint_data = irq_get_handler_data(irq);
+	unsigned int virq;
+

Instead of acking the interrupt at parent chip from ack callback of EINT0_3 chip, I would rather use chained_irq_enter() here...

Quoted 3 lines
(quoted diff) +2 -0
+	/* the first 4 eints have a simple 1 to 1 mapping */
+	virq = irq_linear_revmap(eint_data->domains[data->hwirq],

data->hwirq); + /* Something must be really wrong if an unmapped

EINT

Quoted 7 lines

drivers/pinctrl/pinctrl-s3c24xx.c line 322

Quoted 5 lines
drivers/pinctrl/pinctrl-s3c24xx.c new file +5 -0
+	 * was unmasked...
+	 */
+	BUG_ON(!virq);
+
+	generic_handle_irq(virq);

...and chained_irq_exit() here.

If I understand it correctly, the way chained_irq_* works it would limit the eints to a level style handling. With the way it's currently the whole determination of when to ack,mask and unmask is completely for the real handler (edge or level) to decide, as the original interrupt gets completely forwarded into the irq-domain without further constraints.

So, after the change on regular s3c24xx SoCs when the real irq handler wants to ack the irq, it would be a no-op, as it would already have been acked by chained_irq_enter.

Masking might be even more interesting. Currently control is transfered completely to the pinctrl irq-domain, which then controls the masking of the interrupt thru the parent-calls - on regular s3c24xx the masking of these is also only done in the main controller.

Quoted 8 lines

When using chained_irq_* it also wants to mask the interrupt which might conflict with regular enable_irq/disable_irq calls being done for example in driver code.

So in short I agree with the earlier split of the irqchip, but would keep the irq operations themself centralized in the pinctrl driver, instead of using chained_irq_* functions.

Quoted 8 lines

Right, my solution wouldn't work properly in case of regular s3c24xx and edge triggered interrupts.

However I'm still wondering if it's OK to manually call parent chip operations in case of s3c2416. This would imply the same operation calling order as imposed by flow handler of the chained EINT (which can be handle_edge_irq), while the parent chip is probably level triggered, isn't it?

I think imposing the same calling order is the right way to go :-) .

The eints can be set hardware-wise to either level or edge triggered, but their flow handler is currently hard set to handle_edge_irq. I remember seeing a patch from the Openmoko guys somewhere indicating that this does not work at all with level irqs.

Only the main interrupts of the real demuxed eints (3_to_7 and 8_to...) are always level triggered, and therefore already use chained_irq...

In the current pinctrl driver, the parent eints do not have a flow handler of their own (as the chained handler replaces it). So all flow handling is done by the flowhandler of the irq in the pinctrl domain.

handle_IRQ(main_eint)
-> demux_eint0_3
-> generic_handle_irq(pinctrl_eint)
-> handle_edge_irq (or handle_level_irq)
-> call ack, mask or unmask of the pinctrl eint -> call ack, mask or unmask of the underlying main eint

Heiko

#6 lore raw

PATCH 1/1 pinctrl: Add pinctrl-s3c24xx driver

Quoted 109 lines · On Wednesday 10 of April 2013 15:45:48 Heiko St?bner wrote:

Am Mittwoch, 10. April 2013, 14:31:29 schrieb Tomasz Figa:

Quoted 69 lines · On Wednesday 10 of April 2013 14:20:22 Heiko St?bner wrote:
Quoted 68 lines

Hi Tomasz,

thanks for your comments, more inline.

Am Mittwoch, 10. April 2013, 12:36:39 schrieb Tomasz Figa:

Quoted 11 lines

Hi Heiko,

Basically looks good to me, but please see my inline comments about
handling of EINT0-3.

Quoted 4 lines · On Wednesday 10 of April 2013 01:35:12 Heiko St?bner wrote:

The s3c24xx pins follow a similar pattern as the other Samsung SoCs
and
can therefore reuse the already introduced infrastructure.

[...]

Quoted 14 lines

drivers/pinctrl/pinctrl-s3c24xx.c line 106

Quoted 9 lines
drivers/pinctrl/pinctrl-s3c24xx.c new file +9 -0
+struct s3c24xx_eint_data {
+	struct samsung_pinctrl_drv_data *drvdata;
+	struct irq_domain *domains[NUM_EINT];
+	int parents[NUM_EINT_IRQ];
+};
+
+struct s3c24xx_eint_domain_data {
+	struct samsung_pin_bank *bank;
+	struct s3c24xx_eint_data *eint_data;

What about:

(quoted diff) +1 -0
+	bool eint0_3_parent_only;
Quoted 8 lines

(or whatever name would be more appropriate), which would store the
information about the s3c24xx-specific quirk in 24xx-specific data structure, without the need to add another field to the generic samsung_pinctrl_drv_data structure?

See my further comments on how I would see using this field in interrupt handling code.

ok, sounds good, especially gathering the type from the wakeup-int property

Quoted 1 line
Quoted 1 line

+};

[...]

Quoted 4 lines

Now I would split the following 3 functions into two sets of 3 functions, one set for s3c2412 and other for remaining SoCs and make
separate EINT0-3 IRQ chips for both cases.

Not doing the decision every time, might bring some very slight speed
improvements, so is probably the right way to go.

Quoted 10 lines
Quoted 10 lines
drivers/pinctrl/pinctrl-s3c24xx.c new file +4 -0
+
+static void s3c24xx_eint0_3_ack(struct irq_data *data)
+{
+	struct samsung_pin_bank *bank =

irq_data_get_irq_chip_data(data);

(quoted diff) +2 -0
+	struct samsung_pinctrl_drv_data *d = bank->drvdata;
+	struct s3c24xx_eint_domain_data *ddata = bank->irq_domain-

host_data;

Quoted 11 lines
Quoted 11 lines
Quoted 11 lines

drivers/pinctrl/pinctrl-s3c24xx.c

Quoted 11 lines
drivers/pinctrl/pinctrl-s3c24xx.c new file +11 -0
+	struct s3c24xx_eint_data *eint_data = ddata->eint_data;
+	int parent_irq = eint_data->parents[data->hwirq];
+	struct irq_chip *parent_chip = irq_get_chip(parent_irq);
+
+	if (d->ctrl->type == S3C2412) {
+		unsigned long bitval = 1UL << data->hwirq;
+		writel(bitval, d->virt_base + EINTPEND_REG);
+	}
+
+	if (parent_chip->irq_ack)
+		parent_chip-

irq_ack(irq_get_irq_data(parent_irq));

Quoted 27 lines
Quoted 27 lines
Quoted 2 lines

Btw. Is this parent level acking really needed here?

Depends. If using chained_irq_* of course not, but if the irq-handler
should stay in charge of when to ack it might be better this way.

Generic s3c24xx SoCs need acking in the main controller only, while s3c2412 needs acking in both the main controller and eintpend.

Quoted 1 line
Quoted 1 line

+}

[...]

Quoted 13 lines

Review

Quoted 8 lines

+static void s3c24xx_demux_eint0_3(unsigned int irq, struct irq_desc
*desc) +{

(quoted diff) +2 -0
+	struct irq_data *data = irq_desc_get_irq_data(desc);
+	struct s3c24xx_eint_data *eint_data =

irq_get_handler_data(irq);

(quoted diff) +2 -0
+	unsigned int virq;
+

Instead of acking the interrupt at parent chip from ack callback of
EINT0_3 chip, I would rather use chained_irq_enter() here...

Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
(quoted diff) +2 -0
+	/* the first 4 eints have a simple 1 to 1 mapping */
+	virq = irq_linear_revmap(eint_data->domains[data->hwirq],

data->hwirq); + /* Something must be really wrong if an

unmapped

Quoted 48 lines
Quoted 39 lines
Quoted 30 lines
Quoted 10 lines

EINT

drivers/pinctrl/pinctrl-s3c24xx.c line 322

Quoted 5 lines
drivers/pinctrl/pinctrl-s3c24xx.c new file +5 -0
+	 * was unmasked...
+	 */
+	BUG_ON(!virq);
+
+	generic_handle_irq(virq);

...and chained_irq_exit() here.

If I understand it correctly, the way chained_irq_* works it would limit the eints to a level style handling. With the way it's currently the whole determination of when to ack,mask and unmask is completely for the real handler (edge or level) to decide, as the original interrupt gets completely forwarded into the irq-domain without further
constraints.

So, after the change on regular s3c24xx SoCs when the real irq handler
wants to ack the irq, it would be a no-op, as it would already have been acked by chained_irq_enter.

Masking might be even more interesting. Currently control is transfered
completely to the pinctrl irq-domain, which then controls the masking of the interrupt thru the parent-calls - on regular s3c24xx the masking of these is also only done in the main controller.

Quoted 8 lines

When using chained_irq_* it also wants to mask the interrupt which might conflict with regular enable_irq/disable_irq calls being done for example in driver code.

So in short I agree with the earlier split of the irqchip, but would keep the irq operations themself centralized in the pinctrl driver, instead of using chained_irq_* functions.

Quoted 8 lines

Right, my solution wouldn't work properly in case of regular s3c24xx and edge triggered interrupts.

However I'm still wondering if it's OK to manually call parent chip operations in case of s3c2416. This would imply the same operation calling order as imposed by flow handler of the chained EINT (which can be handle_edge_irq), while the parent chip is probably level triggered, isn't it?

Quoted 6 lines

I think imposing the same calling order is the right way to go :-) .

The eints can be set hardware-wise to either level or edge triggered, but their flow handler is currently hard set to handle_edge_irq. I remember seeing a patch from the Openmoko guys somewhere indicating that this does not work at all with level irqs.

Sounds like something broken to me. Level-triggered interrupts need different handling than edge-triggered handling and so different flow handlers exist for them.

To make my doubts clear, this is how I'm seeing the hardware on 2412:

edge or level triggered level triggered

       |   ___________________   |   ________________
       |  |                   |  |  |                |
EINT0 --> | EINT0   EINT0PEND | --> | IRQ0   ARM_INT | --> to the ARM core
       |  |                   |  |  |                |
EINT1 --> | EINT1   EINT1PEND | --> | IRQ1           |
       |  |                   |  |  |                |
EINT2 --> | EINT2   EINT2PEND | --> | IRQ2           |
       |  |                   |  |  |                |
EINT3 --> | EINT3   EINT3PEND | --> | IRQ3           |
          |___________________|     |________________|

GPIO controller Interrupt controller

Now EINT0-EINT3 signals of the GPIO controller are directly driven with external interrupt signals, either edge or level triggered, but the interrupt status is latched in EINT0PEND-EINT3PEND bits inside the GPIO controller and state of those bits is exported through EINT0PEND-EINT3PEND signals to the interrupt controller's IRQ0-IRQ3 signals, which are now level-triggered, since the interrupt is active only when EINTxPEND bit is high.

Now, if my vision is wrong, and the hardware has simply EINT0-EINT3 pins routed to IRQ0-IRQ3 signals of the interrupt controller, then what you're saying is completely right. In this case just ignore my moaning. ;)

Best regards,
Tomasz

Quoted 16 lines

Only the main interrupts of the real demuxed eints (3_to_7 and 8_to...) are always level triggered, and therefore already use chained_irq...

In the current pinctrl driver, the parent eints do not have a flow handler of their own (as the chained handler replaces it). So all flow handling is done by the flowhandler of the irq in the pinctrl domain.

handle_IRQ(main_eint)
-> demux_eint0_3
-> generic_handle_irq(pinctrl_eint)
-> handle_edge_irq (or handle_level_irq)
-> call ack, mask or unmask of the pinctrl eint -> call ack, mask or unmask of the underlying main eint

Heiko

#7 lore raw

PATCH 1/1 pinctrl: Add pinctrl-s3c24xx driver

Am Mittwoch, 10. April 2013, 21:51:11 schrieb Tomasz Figa:

Quoted 201 lines · On Wednesday 10 of April 2013 15:45:48 Heiko St?bner wrote:
Quoted 110 lines

Am Mittwoch, 10. April 2013, 14:31:29 schrieb Tomasz Figa:

Quoted 69 lines · On Wednesday 10 of April 2013 14:20:22 Heiko St?bner wrote:
Quoted 68 lines

Hi Tomasz,

thanks for your comments, more inline.

Am Mittwoch, 10. April 2013, 12:36:39 schrieb Tomasz Figa:

Quoted 11 lines

Hi Heiko,

Basically looks good to me, but please see my inline comments about
handling of EINT0-3.

Quoted 4 lines · On Wednesday 10 of April 2013 01:35:12 Heiko St?bner wrote:

The s3c24xx pins follow a similar pattern as the other Samsung SoCs
and
can therefore reuse the already introduced infrastructure.

[...]

Quoted 14 lines

drivers/pinctrl/pinctrl-s3c24xx.c line 106

Quoted 9 lines
drivers/pinctrl/pinctrl-s3c24xx.c new file +9 -0
+struct s3c24xx_eint_data {
+	struct samsung_pinctrl_drv_data *drvdata;
+	struct irq_domain *domains[NUM_EINT];
+	int parents[NUM_EINT_IRQ];
+};
+
+struct s3c24xx_eint_domain_data {
+	struct samsung_pin_bank *bank;
+	struct s3c24xx_eint_data *eint_data;

What about:

(quoted diff) +1 -0
+	bool eint0_3_parent_only;
Quoted 8 lines

(or whatever name would be more appropriate), which would store the
information about the s3c24xx-specific quirk in 24xx-specific data structure, without the need to add another field to the generic samsung_pinctrl_drv_data structure?

See my further comments on how I would see using this field in interrupt handling code.

ok, sounds good, especially gathering the type from the wakeup-int property

Quoted 1 line
Quoted 1 line

+};

[...]

Quoted 4 lines

Now I would split the following 3 functions into two sets of 3 functions, one set for s3c2412 and other for remaining SoCs and make
separate EINT0-3 IRQ chips for both cases.

Not doing the decision every time, might bring some very slight speed
improvements, so is probably the right way to go.

Quoted 10 lines
Quoted 10 lines
drivers/pinctrl/pinctrl-s3c24xx.c new file +4 -0
+
+static void s3c24xx_eint0_3_ack(struct irq_data *data)
+{
+	struct samsung_pin_bank *bank =

irq_data_get_irq_chip_data(data);

(quoted diff) +2 -0
+	struct samsung_pinctrl_drv_data *d = bank->drvdata;
+	struct s3c24xx_eint_domain_data *ddata = bank->irq_domain-

host_data;

Quoted 11 lines
Quoted 11 lines
Quoted 11 lines

drivers/pinctrl/pinctrl-s3c24xx.c

Quoted 11 lines
drivers/pinctrl/pinctrl-s3c24xx.c new file +11 -0
+	struct s3c24xx_eint_data *eint_data = ddata->eint_data;
+	int parent_irq = eint_data->parents[data->hwirq];
+	struct irq_chip *parent_chip = irq_get_chip(parent_irq);
+
+	if (d->ctrl->type == S3C2412) {
+		unsigned long bitval = 1UL << data->hwirq;
+		writel(bitval, d->virt_base + EINTPEND_REG);
+	}
+
+	if (parent_chip->irq_ack)
+		parent_chip-

irq_ack(irq_get_irq_data(parent_irq));

Quoted 26 lines
Quoted 26 lines
Quoted 1 line

Btw. Is this parent level acking really needed here?

Depends. If using chained_irq_* of course not, but if the irq-handler
should stay in charge of when to ack it might be better this way.

Generic s3c24xx SoCs need acking in the main controller only, while s3c2412 needs acking in both the main controller and eintpend.

Quoted 1 line
Quoted 1 line

+}

[...]

Quoted 13 lines

Review

Quoted 8 lines

+static void s3c24xx_demux_eint0_3(unsigned int irq, struct irq_desc
*desc) +{

(quoted diff) +2 -0
+	struct irq_data *data = irq_desc_get_irq_data(desc);
+	struct s3c24xx_eint_data *eint_data =

irq_get_handler_data(irq);

(quoted diff) +2 -0
+	unsigned int virq;
+

Instead of acking the interrupt at parent chip from ack callback of
EINT0_3 chip, I would rather use chained_irq_enter() here...

Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
(quoted diff) +2 -0
+	/* the first 4 eints have a simple 1 to 1 mapping */
+	virq = irq_linear_revmap(eint_data->domains[data->hwirq],

data->hwirq); + /* Something must be really wrong if an

unmapped

Quoted 47 lines
Quoted 38 lines
Quoted 29 lines
Quoted 9 lines

EINT

drivers/pinctrl/pinctrl-s3c24xx.c line 322

Quoted 5 lines
drivers/pinctrl/pinctrl-s3c24xx.c new file +5 -0
+	 * was unmasked...
+	 */
+	BUG_ON(!virq);
+
+	generic_handle_irq(virq);

...and chained_irq_exit() here.

If I understand it correctly, the way chained_irq_* works it would limit the eints to a level style handling. With the way it's currently the whole determination of when to ack,mask and unmask is completely for the real handler (edge or level) to decide, as the original interrupt gets completely forwarded into the irq-domain without further
constraints.

So, after the change on regular s3c24xx SoCs when the real irq handler
wants to ack the irq, it would be a no-op, as it would already have been acked by chained_irq_enter.

Masking might be even more interesting. Currently control is transfered
completely to the pinctrl irq-domain, which then controls the masking of the interrupt thru the parent-calls - on regular s3c24xx the masking of these is also only done in the main controller.

Quoted 8 lines

When using chained_irq_* it also wants to mask the interrupt which might conflict with regular enable_irq/disable_irq calls being done for example in driver code.

So in short I agree with the earlier split of the irqchip, but would keep the irq operations themself centralized in the pinctrl driver, instead of using chained_irq_* functions.

Quoted 8 lines

Right, my solution wouldn't work properly in case of regular s3c24xx and edge triggered interrupts.

However I'm still wondering if it's OK to manually call parent chip operations in case of s3c2416. This would imply the same operation calling order as imposed by flow handler of the chained EINT (which can be handle_edge_irq), while the parent chip is probably level triggered, isn't it?

Quoted 6 lines

I think imposing the same calling order is the right way to go :-) .

The eints can be set hardware-wise to either level or edge triggered, but their flow handler is currently hard set to handle_edge_irq. I remember seeing a patch from the Openmoko guys somewhere indicating that this does not work at all with level irqs.

Sounds like something broken to me. Level-triggered interrupts need different handling than edge-triggered handling and so different flow handlers exist for them.

To make my doubts clear, this is how I'm seeing the hardware on 2412:

edge or level triggered level triggered

       |   ___________________   |   ________________

EINT0 --> | EINT0 EINT0PEND | --> | IRQ0 ARM_INT | --> to the ARM core

EINT1 --> | EINT1 EINT1PEND | --> | IRQ1 |

EINT2 --> | EINT2 EINT2PEND | --> | IRQ2 |

EINT3 --> | EINT3 EINT3PEND | --> | IRQ3 |

|___________________| |________________|

GPIO controller Interrupt controller

Now EINT0-EINT3 signals of the GPIO controller are directly driven with external interrupt signals, either edge or level triggered, but the interrupt status is latched in EINT0PEND-EINT3PEND bits inside the GPIO controller and state of those bits is exported through EINT0PEND-EINT3PEND signals to the interrupt controller's IRQ0-IRQ3 signals, which are now level-triggered, since the interrupt is active only when EINTxPEND bit is high.

Quoted 3 lines

Now, if my vision is wrong, and the hardware has simply EINT0-EINT3 pins routed to IRQ0-IRQ3 signals of the interrupt controller, then what you're saying is completely right. In this case just ignore my moaning. ;)

I know the s3c2412 only in theory myself and am just going with the old code :-) . So all the irq rework stuff I did was only done on a theoretical level for s3c2412, so I don't really know if the current irq code runs for sure there.

But when looking at the real old code that must have run sucessfully at some point in the past [0], the comment says

"the s3c2412 changes the behaviour of IRQ_EINT0 through IRQ_EINT3 by
having them turn up in both the INT* and the EINT* registers. Whilst both show the status, they both now need to be acked when the IRQs go off."

And then simply doing a

	__raw_writel(bitval, S3C2412_EINTPEND);
	__raw_writel(bitval, S3C2410_SRCPND);
	__raw_writel(bitval, S3C2410_INTPND);

there - which is essentially the same that happens in the current pinctrl ack code.

So if we assume the really old code worked at some point, I'm relativly sure the new one will too :-). I'll just exchange the mask ordering in s3c2412_eint0_3_mask, to match the ordering in the original function.

Heiko

[0]
https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/tree/arch/arm/mach-
s3c24xx/irq-s3c2412.c

#8 lore raw

PATCH 1/1 pinctrl: Add pinctrl-s3c24xx driver

Quoted 249 lines · On Wednesday 10 of April 2013 22:11:03 Heiko St?bner wrote:

Am Mittwoch, 10. April 2013, 21:51:11 schrieb Tomasz Figa:

Quoted 225 lines · On Wednesday 10 of April 2013 15:45:48 Heiko St?bner wrote:
Quoted 119 lines

Am Mittwoch, 10. April 2013, 14:31:29 schrieb Tomasz Figa:

Quoted 75 lines · On Wednesday 10 of April 2013 14:20:22 Heiko St?bner wrote:
Quoted 74 lines

Hi Tomasz,

thanks for your comments, more inline.

Am Mittwoch, 10. April 2013, 12:36:39 schrieb Tomasz Figa:

Quoted 12 lines

Hi Heiko,

Basically looks good to me, but please see my inline comments about
handling of EINT0-3.

Quoted 5 lines · On Wednesday 10 of April 2013 01:35:12 Heiko St?bner wrote:

The s3c24xx pins follow a similar pattern as the other Samsung
SoCs
and
can therefore reuse the already introduced infrastructure.

[...]

Quoted 23 lines

Review

Quoted 9 lines
(quoted diff) +9 -0
+struct s3c24xx_eint_data {
+	struct samsung_pinctrl_drv_data *drvdata;
+	struct irq_domain *domains[NUM_EINT];
+	int parents[NUM_EINT_IRQ];
+};
+
+struct s3c24xx_eint_domain_data {
+	struct samsung_pin_bank *bank;
+	struct s3c24xx_eint_data *eint_data;

What about:

(quoted diff) +1 -0
+	bool eint0_3_parent_only;

(or whatever name would be more appropriate), which would store
the
information about the s3c24xx-specific quirk in 24xx-specific data
structure, without the need to add another field to the generic
samsung_pinctrl_drv_data structure?

Quoted 2 lines

See my further comments on how I would see using this field in interrupt handling code.

ok, sounds good, especially gathering the type from the wakeup-int
property

Quoted 1 line
Quoted 1 line

+};

[...]

Quoted 5 lines

Now I would split the following 3 functions into two sets of 3 functions, one set for s3c2412 and other for remaining SoCs and
make
separate EINT0-3 IRQ chips for both cases.

Not doing the decision every time, might bring some very slight speed
improvements, so is probably the right way to go.

Quoted 10 lines
Quoted 10 lines
(quoted diff) +4 -0
+
+static void s3c24xx_eint0_3_ack(struct irq_data *data)
+{
+	struct samsung_pin_bank *bank =

irq_data_get_irq_chip_data(data);

(quoted diff) +2 -0
+	struct samsung_pinctrl_drv_data *d = bank->drvdata;
+	struct s3c24xx_eint_domain_data *ddata = bank->irq_domain-

host_data;

Quoted 11 lines
Quoted 11 lines
Quoted 11 lines

Review

Quoted 11 lines
(quoted diff) +11 -0
+	struct s3c24xx_eint_data *eint_data = ddata->eint_data;
+	int parent_irq = eint_data->parents[data->hwirq];
+	struct irq_chip *parent_chip = irq_get_chip(parent_irq);
+
+	if (d->ctrl->type == S3C2412) {
+		unsigned long bitval = 1UL << data->hwirq;
+		writel(bitval, d->virt_base + EINTPEND_REG);
+	}
+
+	if (parent_chip->irq_ack)
+		parent_chip-

irq_ack(irq_get_irq_data(parent_irq));

Quoted 29 lines
Quoted 29 lines
Quoted 1 line

Btw. Is this parent level acking really needed here?

Depends. If using chained_irq_* of course not, but if the irq-handler
should stay in charge of when to ack it might be better this way.

Generic s3c24xx SoCs need acking in the main controller only, while
s3c2412 needs acking in both the main controller and eintpend.

Quoted 1 line
Quoted 1 line

+}

[...]

Quoted 14 lines

Review

Quoted 8 lines

+static void s3c24xx_demux_eint0_3(unsigned int irq, struct irq_desc
*desc) +{

(quoted diff) +2 -0
+	struct irq_data *data = irq_desc_get_irq_data(desc);
+	struct s3c24xx_eint_data *eint_data =

irq_get_handler_data(irq);

(quoted diff) +2 -0
+	unsigned int virq;
+

Instead of acking the interrupt at parent chip from ack callback
of
EINT0_3 chip, I would rather use chained_irq_enter() here...

Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
(quoted diff) +2 -0
+	/* the first 4 eints have a simple 1 to 1 mapping */
+	virq = irq_linear_revmap(eint_data->domains[data->hwirq],

data->hwirq); + /* Something must be really wrong if an

unmapped

Quoted 59 lines
Quoted 51 lines
Quoted 41 lines
Quoted 9 lines

EINT

Review

Quoted 5 lines
(quoted diff) +5 -0
+	 * was unmasked...
+	 */
+	BUG_ON(!virq);
+
+	generic_handle_irq(virq);

...and chained_irq_exit() here.

If I understand it correctly, the way chained_irq_* works it would
limit the eints to a level style handling. With the way it's currently the whole determination of when to ack,mask and unmask is
completely for the real handler (edge or level) to decide, as the
original interrupt gets completely forwarded into the irq-domain without further
constraints.

So, after the change on regular s3c24xx SoCs when the real irq handler
wants to ack the irq, it would be a no-op, as it would already have
been acked by chained_irq_enter.

Masking might be even more interesting. Currently control is transfered
completely to the pinctrl irq-domain, which then controls the masking of the interrupt thru the parent-calls - on regular s3c24xx
the masking of these is also only done in the main controller.

When using chained_irq_* it also wants to mask the interrupt which
might conflict with regular enable_irq/disable_irq calls being done
for example in driver code.

Quoted 5 lines

So in short I agree with the earlier split of the irqchip, but would
keep the irq operations themself centralized in the pinctrl driver,
instead of using chained_irq_* functions.

Right, my solution wouldn't work properly in case of regular s3c24xx
and edge triggered interrupts.

Quoted 7 lines

However I'm still wondering if it's OK to manually call parent chip
operations in case of s3c2416. This would imply the same operation calling order as imposed by flow handler of the chained EINT (which
can be handle_edge_irq), while the parent chip is probably level triggered, isn't it?

Quoted 7 lines

I think imposing the same calling order is the right way to go :-) .

The eints can be set hardware-wise to either level or edge triggered,
but their flow handler is currently hard set to handle_edge_irq. I remember seeing a patch from the Openmoko guys somewhere indicating that this does not work at all with level irqs.

Sounds like something broken to me. Level-triggered interrupts need different handling than edge-triggered handling and so different flow handlers exist for them.

To make my doubts clear, this is how I'm seeing the hardware on 2412:
edge or level triggered level triggered

       |   ___________________   |   ________________

EINT0 --> | EINT0 EINT0PEND | --> | IRQ0 ARM_INT | --> to the ARM
core

EINT1 --> | EINT1 EINT1PEND | --> | IRQ1 |

EINT2 --> | EINT2 EINT2PEND | --> | IRQ2 |

EINT3 --> | EINT3 EINT3PEND | --> | IRQ3 |

          |___________________|     |________________|
          |
             GPIO controller       Interrupt controller

Now EINT0-EINT3 signals of the GPIO controller are directly driven with
external interrupt signals, either edge or level triggered, but the interrupt status is latched in EINT0PEND-EINT3PEND bits inside the GPIO
controller and state of those bits is exported through EINT0PEND-EINT3PEND signals to the interrupt controller's IRQ0-IRQ3 signals, which are now level-triggered, since the interrupt is active only when EINTxPEND bit is high.

Quoted 4 lines

Now, if my vision is wrong, and the hardware has simply EINT0-EINT3 pins routed to IRQ0-IRQ3 signals of the interrupt controller, then what you're saying is completely right. In this case just ignore my moaning. ;)

I know the s3c2412 only in theory myself and am just going with the old code
:-) . So all the irq rework stuff I did was only done on a theoretical :level
for s3c2412, so I don't really know if the current irq code runs for sure there.

But when looking at the real old code that must have run sucessfully at some point in the past [0], the comment says

"the s3c2412 changes the behaviour of IRQ_EINT0 through IRQ_EINT3 by
having them turn up in both the INT* and the EINT* registers. Whilst both show the status, they both now need to be acked when the IRQs go off."

And then simply doing a

	__raw_writel(bitval, S3C2412_EINTPEND);
	__raw_writel(bitval, S3C2410_SRCPND);
	__raw_writel(bitval, S3C2410_INTPND);
Quoted 6 lines

there - which is essentially the same that happens in the current pinctrl ack code.

So if we assume the really old code worked at some point, I'm relativly sure the new one will too :-). I'll just exchange the mask ordering in s3c2412_eint0_3_mask, to match the ordering in the original function.

Still, the old code seems to support only edge-triggered interrupts and this might be why it's working. I guess we won't know that without testing it on a S3C2412, though.

OK, it should be fine to keep the old behavior for now, but I think this is something that should be investigated in future and, if needed, fixed with another patch.

Btw. I wonder if I can get somewhere a board with S3C2412. Such board would be really useful for testing any of my patches touching plat-samsung and mach-s3c* on s3c24xx, as currently I can only test on s3c6410.

Best regards,
Tomasz

Quoted 7 lines

Heiko

[0]
https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/tree/arc
h/arm/mach- s3c24xx/irq-s3c2412.c

#9 lore raw

PATCH 1/1 pinctrl: Add pinctrl-s3c24xx driver

Am Mittwoch, 10. April 2013, 22:17:43 schrieb Tomasz Figa:

Quoted 263 lines · On Wednesday 10 of April 2013 22:11:03 Heiko St?bner wrote:
Quoted 255 lines

Am Mittwoch, 10. April 2013, 21:51:11 schrieb Tomasz Figa:

Quoted 225 lines · On Wednesday 10 of April 2013 15:45:48 Heiko St?bner wrote:
Quoted 119 lines

Am Mittwoch, 10. April 2013, 14:31:29 schrieb Tomasz Figa:

Quoted 75 lines · On Wednesday 10 of April 2013 14:20:22 Heiko St?bner wrote:
Quoted 74 lines

Hi Tomasz,

thanks for your comments, more inline.

Am Mittwoch, 10. April 2013, 12:36:39 schrieb Tomasz Figa:

Quoted 12 lines

Hi Heiko,

Basically looks good to me, but please see my inline comments about
handling of EINT0-3.

Quoted 5 lines · On Wednesday 10 of April 2013 01:35:12 Heiko St?bner wrote:

The s3c24xx pins follow a similar pattern as the other Samsung
SoCs
and
can therefore reuse the already introduced infrastructure.

[...]

Quoted 23 lines

Review

Quoted 9 lines
(quoted diff) +9 -0
+struct s3c24xx_eint_data {
+	struct samsung_pinctrl_drv_data *drvdata;
+	struct irq_domain *domains[NUM_EINT];
+	int parents[NUM_EINT_IRQ];
+};
+
+struct s3c24xx_eint_domain_data {
+	struct samsung_pin_bank *bank;
+	struct s3c24xx_eint_data *eint_data;

What about:

(quoted diff) +1 -0
+	bool eint0_3_parent_only;

(or whatever name would be more appropriate), which would store
the
information about the s3c24xx-specific quirk in 24xx-specific data
structure, without the need to add another field to the generic
samsung_pinctrl_drv_data structure?

Quoted 2 lines

See my further comments on how I would see using this field in interrupt handling code.

ok, sounds good, especially gathering the type from the wakeup-int
property

Quoted 1 line
Quoted 1 line

+};

[...]

Quoted 5 lines

Now I would split the following 3 functions into two sets of 3 functions, one set for s3c2412 and other for remaining SoCs and
make
separate EINT0-3 IRQ chips for both cases.

Not doing the decision every time, might bring some very slight speed
improvements, so is probably the right way to go.

Quoted 10 lines
Quoted 10 lines
(quoted diff) +4 -0
+
+static void s3c24xx_eint0_3_ack(struct irq_data *data)
+{
+	struct samsung_pin_bank *bank =

irq_data_get_irq_chip_data(data);

(quoted diff) +2 -0
+	struct samsung_pinctrl_drv_data *d = bank->drvdata;
+	struct s3c24xx_eint_domain_data *ddata = bank->irq_domain-

host_data;

Quoted 11 lines
Quoted 11 lines
Quoted 11 lines

Review

Quoted 11 lines
(quoted diff) +11 -0
+	struct s3c24xx_eint_data *eint_data = ddata->eint_data;
+	int parent_irq = eint_data->parents[data->hwirq];
+	struct irq_chip *parent_chip = irq_get_chip(parent_irq);
+
+	if (d->ctrl->type == S3C2412) {
+		unsigned long bitval = 1UL << data->hwirq;
+		writel(bitval, d->virt_base + EINTPEND_REG);
+	}
+
+	if (parent_chip->irq_ack)
+		parent_chip-

irq_ack(irq_get_irq_data(parent_irq));

Quoted 29 lines
Quoted 29 lines
Quoted 1 line

Btw. Is this parent level acking really needed here?

Depends. If using chained_irq_* of course not, but if the irq-handler
should stay in charge of when to ack it might be better this way.

Generic s3c24xx SoCs need acking in the main controller only, while
s3c2412 needs acking in both the main controller and eintpend.

Quoted 1 line
Quoted 1 line

+}

[...]

Quoted 14 lines

Review

Quoted 8 lines

+static void s3c24xx_demux_eint0_3(unsigned int irq, struct irq_desc
*desc) +{

(quoted diff) +2 -0
+	struct irq_data *data = irq_desc_get_irq_data(desc);
+	struct s3c24xx_eint_data *eint_data =

irq_get_handler_data(irq);

(quoted diff) +2 -0
+	unsigned int virq;
+

Instead of acking the interrupt at parent chip from ack callback
of
EINT0_3 chip, I would rather use chained_irq_enter() here...

Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
(quoted diff) +2 -0
+	/* the first 4 eints have a simple 1 to 1 mapping */
+	virq = irq_linear_revmap(eint_data->domains[data->hwirq],

data->hwirq); + /* Something must be really wrong if an

unmapped

Quoted 59 lines
Quoted 51 lines
Quoted 41 lines
Quoted 9 lines

EINT

Review

Quoted 5 lines
(quoted diff) +5 -0
+	 * was unmasked...
+	 */
+	BUG_ON(!virq);
+
+	generic_handle_irq(virq);

...and chained_irq_exit() here.

If I understand it correctly, the way chained_irq_* works it would
limit the eints to a level style handling. With the way it's currently the whole determination of when to ack,mask and unmask is
completely for the real handler (edge or level) to decide, as the
original interrupt gets completely forwarded into the irq-domain without further
constraints.

So, after the change on regular s3c24xx SoCs when the real irq handler
wants to ack the irq, it would be a no-op, as it would already have
been acked by chained_irq_enter.

Masking might be even more interesting. Currently control is transfered
completely to the pinctrl irq-domain, which then controls the masking of the interrupt thru the parent-calls - on regular s3c24xx
the masking of these is also only done in the main controller.

When using chained_irq_* it also wants to mask the interrupt which
might conflict with regular enable_irq/disable_irq calls being done
for example in driver code.

Quoted 5 lines

So in short I agree with the earlier split of the irqchip, but would
keep the irq operations themself centralized in the pinctrl driver,
instead of using chained_irq_* functions.

Right, my solution wouldn't work properly in case of regular s3c24xx
and edge triggered interrupts.

Quoted 7 lines

However I'm still wondering if it's OK to manually call parent chip
operations in case of s3c2416. This would imply the same operation calling order as imposed by flow handler of the chained EINT (which
can be handle_edge_irq), while the parent chip is probably level triggered, isn't it?

Quoted 7 lines

I think imposing the same calling order is the right way to go :-) .

The eints can be set hardware-wise to either level or edge triggered,
but their flow handler is currently hard set to handle_edge_irq. I remember seeing a patch from the Openmoko guys somewhere indicating that this does not work at all with level irqs.

Sounds like something broken to me. Level-triggered interrupts need different handling than edge-triggered handling and so different flow handlers exist for them.

To make my doubts clear, this is how I'm seeing the hardware on 2412:
edge or level triggered level triggered

       |   ___________________   |   ________________

EINT0 --> | EINT0 EINT0PEND | --> | IRQ0 ARM_INT | --> to the ARM
core

EINT1 --> | EINT1 EINT1PEND | --> | IRQ1 |

EINT2 --> | EINT2 EINT2PEND | --> | IRQ2 |

EINT3 --> | EINT3 EINT3PEND | --> | IRQ3 |

          |___________________|     |________________|
          |
             GPIO controller       Interrupt controller

Now EINT0-EINT3 signals of the GPIO controller are directly driven with
external interrupt signals, either edge or level triggered, but the interrupt status is latched in EINT0PEND-EINT3PEND bits inside the GPIO
controller and state of those bits is exported through EINT0PEND-EINT3PEND signals to the interrupt controller's IRQ0-IRQ3 signals, which are now level-triggered, since the interrupt is active only when EINTxPEND bit is high.

Quoted 4 lines

Now, if my vision is wrong, and the hardware has simply EINT0-EINT3 pins routed to IRQ0-IRQ3 signals of the interrupt controller, then what you're saying is completely right. In this case just ignore my moaning. ;)

I know the s3c2412 only in theory myself and am just going with the old code

:-) . So all the irq rework stuff I did was only done on a theoretical :level

for s3c2412, so I don't really know if the current irq code runs for sure there.

But when looking at the real old code that must have run sucessfully at some point in the past [0], the comment says

"the s3c2412 changes the behaviour of IRQ_EINT0 through IRQ_EINT3 by

having them turn up in both the INT* and the EINT* registers. Whilst both show the status, they both now need to be acked when the IRQs go off."

And then simply doing a

	__raw_writel(bitval, S3C2412_EINTPEND);
	__raw_writel(bitval, S3C2410_SRCPND);
	__raw_writel(bitval, S3C2410_INTPND);
Quoted 6 lines

there - which is essentially the same that happens in the current pinctrl ack code.

So if we assume the really old code worked at some point, I'm relativly sure the new one will too :-). I'll just exchange the mask ordering in s3c2412_eint0_3_mask, to match the ordering in the original function.

Quoted 7 lines

Still, the old code seems to support only edge-triggered interrupts and this might be why it's working. I guess we won't know that without testing it on a S3C2412, though.

OK, it should be fine to keep the old behavior for now, but I think this is something that should be investigated in future and, if needed, fixed with another patch.

Ok, I'll prepare another patch with the other proposed changes.

Quoted 3 lines

Btw. I wonder if I can get somewhere a board with S3C2412. Such board would be really useful for testing any of my patches touching plat-samsung and mach-s3c* on s3c24xx, as currently I can only test on s3c6410.

Personally I found the S3C2412-based Logitech Squeezebox Controller [0] quite intriguing some time ago. And if only s3c24xx instead of s3c2412 is mandatory, there should also still be quite a lot of s3c2416/2450 based ebook-readers around [1] :-) .

Heiko

[0] http://wiki.slimdevices.com/index.php/Squeezebox_Controller
[1] http://www.youtube.com/mmind81

Quoted 6 lines
Quoted 6 lines

Heiko

[0]
https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/tree/arc
h/arm/mach- s3c24xx/irq-s3c2412.c