[RFC PATCH 6/7] gpio: omap: move pm runtime in irq_chip.irq_bus_lock/sync_unlock

[RFC PATCH 6/7] gpio: omap: move pm runtime in irq_chip.irq_bus_lock/sync_unlock

[Grygorii Strashko]

The PM runtime API can't be used in atomic contex on -RT even if
it's configured as irqsafe. As result, below error report can
be seen when PM runtime API called from IRQ chip's callbacks
irq_startup/irq_shutdown/irq_set_type, because they are
protected by RAW spinlock:

BUG: sleeping function called from invalid context at kernel/locking/rtmutex.c:917
in_atomic(): 1, irqs_disabled(): 128, pid: 96, name: insmod
3 locks held by insmod/96:
 #0:  (&dev->mutex){......}, at: [<c04752c8>] __driver_attach+0x54/0xa0
 #1:  (&dev->mutex){......}, at: [<c04752d4>] __driver_attach+0x60/0xa0
 #2:  (class){......}, at: [<c00a408c>] __irq_get_desc_lock+0x60/0xa4
irq event stamp: 1834
hardirqs last  enabled at (1833): [<c06ab2a4>] _raw_spin_unlock_irqrestore+0x88/0x90
hardirqs last disabled at (1834): [<c06ab068>] _raw_spin_lock_irqsave+0x2c/0x64
softirqs last  enabled at (0): [<c003d220>] copy_process.part.52+0x410/0x19d8
softirqs last disabled at (0): [<  (null)>]   (null)
Preemption disabled at:[<  (null)>]   (null)

CPU: 1 PID: 96 Comm: insmod Tainted: G        W  O    4.1.3-rt3-00618-g57e2387-dirty #184
Hardware name: Generic DRA74X (Flattened Device Tree)
[<c00190f4>] (unwind_backtrace) from [<c0014734>] (show_stack+0x20/0x24)
[<c0014734>] (show_stack) from [<c06a62ec>] (dump_stack+0x88/0xdc)
[<c06a62ec>] (dump_stack) from [<c006ca44>] (___might_sleep+0x198/0x2a8)
[<c006ca44>] (___might_sleep) from [<c06ab6d4>] (rt_spin_lock+0x30/0x70)
[<c06ab6d4>] (rt_spin_lock) from [<c04815ac>] (__pm_runtime_resume+0x68/0xa4)
[<c04815ac>] (__pm_runtime_resume) from [<c04123f4>] (omap_gpio_irq_type+0x188/0x1d8)
[<c04123f4>] (omap_gpio_irq_type) from [<c00a64e4>] (__irq_set_trigger+0x68/0x130)
[<c00a64e4>] (__irq_set_trigger) from [<c00a7bc4>] (irq_set_irq_type+0x44/0x6c)
[<c00a7bc4>] (irq_set_irq_type) from [<c00abbf8>] (irq_create_of_mapping+0x120/0x174)
[<c00abbf8>] (irq_create_of_mapping) from [<c0577b74>] (of_irq_get+0x48/0x58)
[<c0577b74>] (of_irq_get) from [<c0540a14>] (i2c_device_probe+0x54/0x15c)
[<c0540a14>] (i2c_device_probe) from [<c04750dc>] (driver_probe_device+0x184/0x2c8)
[<c04750dc>] (driver_probe_device) from [<c0475310>] (__driver_attach+0x9c/0xa0)
[<c0475310>] (__driver_attach) from [<c0473238>] (bus_for_each_dev+0x7c/0xb0)
[<c0473238>] (bus_for_each_dev) from [<c0474af4>] (driver_attach+0x28/0x30)
[<c0474af4>] (driver_attach) from [<c0474760>] (bus_add_driver+0x154/0x200)
[<c0474760>] (bus_add_driver) from [<c0476348>] (driver_register+0x88/0x108)
[<c0476348>] (driver_register) from [<c0541600>] (i2c_register_driver+0x3c/0x90)
[<c0541600>] (i2c_register_driver) from [<bf003018>] (pcf857x_init+0x18/0x24 [gpio_pcf857x])
[<bf003018>] (pcf857x_init [gpio_pcf857x]) from [<c000998c>] (do_one_initcall+0x128/0x1e8)
[<c000998c>] (do_one_initcall) from [<c06a4220>] (do_init_module+0x6c/0x1bc)
[<c06a4220>] (do_init_module) from [<c00dd0c8>] (load_module+0x18e8/0x21c4)
[<c00dd0c8>] (load_module) from [<c00ddaa0>] (SyS_init_module+0xfc/0x158)
[<c00ddaa0>] (SyS_init_module) from [<c000ff40>] (ret_fast_syscall+0x0/0x54)

The IRQ chip interface defines only two callbacks which are executed in
non-atomic contex - irq_bus_lock/irq_bus_sync_unlock, so lets move
PM runtime calls there.

Cc: <linux-rt-users@vger.kernel.org>
Signed-off-by: Grygorii Strashko <grygorii.strashko@ti.com>
---
 drivers/gpio/gpio-omap.c | 25 +++++++++++++++----------
 1 file changed, 15 insertions(+), 10 deletions(-)

diff --git a/drivers/gpio/gpio-omap.c b/drivers/gpio/gpio-omap.c
index 2ae0d47..c33595e 100644
--- a/drivers/gpio/gpio-omap.c
+++ b/drivers/gpio/gpio-omap.c
@@ -496,9 +496,6 @@ static int omap_gpio_irq_type(struct irq_data *d, unsigned type)
 		(type & (IRQ_TYPE_LEVEL_LOW|IRQ_TYPE_LEVEL_HIGH)))
 		return -EINVAL;
 
-	if (!BANK_USED(bank))
-		pm_runtime_get_sync(bank->dev);
-
 	raw_spin_lock_irqsave(&bank->lock, flags);
 	retval = omap_set_gpio_triggering(bank, offset, type);
 	if (retval) {
@@ -521,8 +518,6 @@ static int omap_gpio_irq_type(struct irq_data *d, unsigned type)
 	return 0;
 
 error:
-	if (!BANK_USED(bank))
-		pm_runtime_put(bank->dev);
 	return retval;
 }
 
@@ -797,9 +792,6 @@ static unsigned int omap_gpio_irq_startup(struct irq_data *d)
 	unsigned long flags;
 	unsigned offset = d->hwirq;
 
-	if (!BANK_USED(bank))
-		pm_runtime_get_sync(bank->dev);
-
 	raw_spin_lock_irqsave(&bank->lock, flags);
 
 	if (!LINE_USED(bank->mod_usage, offset))
@@ -815,8 +807,6 @@ static unsigned int omap_gpio_irq_startup(struct irq_data *d)
 	return 0;
 err:
 	raw_spin_unlock_irqrestore(&bank->lock, flags);
-	if (!BANK_USED(bank))
-		pm_runtime_put(bank->dev);
 	return -EINVAL;
 }
 
@@ -835,6 +825,19 @@ static void omap_gpio_irq_shutdown(struct irq_data *d)
 		omap_clear_gpio_debounce(bank, offset);
 	omap_disable_gpio_module(bank, offset);
 	raw_spin_unlock_irqrestore(&bank->lock, flags);
+}
+
+static void omap_gpio_irq_bus_lock(struct irq_data *data)
+{
+	struct gpio_bank *bank = omap_irq_data_get_bank(data);
+
+	if (!BANK_USED(bank))
+		pm_runtime_get_sync(bank->dev);
+}
+
+static void gpio_irq_bus_sync_unlock(struct irq_data *data)
+{
+	struct gpio_bank *bank = omap_irq_data_get_bank(data);
 
 	/*
 	 * If this is the last IRQ to be freed in the bank,
@@ -1181,6 +1184,8 @@ static int omap_gpio_probe(struct platform_device *pdev)
 	irqc->irq_unmask = omap_gpio_unmask_irq,
 	irqc->irq_set_type = omap_gpio_irq_type,
 	irqc->irq_set_wake = omap_gpio_wake_enable,
+	irqc->irq_bus_lock = omap_gpio_irq_bus_lock,
+	irqc->irq_bus_sync_unlock = gpio_irq_bus_sync_unlock,
 	irqc->name = dev_name(&pdev->dev);
 
 	bank->irq = platform_get_irq(pdev, 0);
-- 
2.5.0

[RFC PATCH 7/7] gpio: omap: convert to use generic irq handler

[Grygorii Strashko]

This patch converts TI OMAP GPIO driver to use generic irq handler
instead of chained IRQ handler. This way OMAP GPIO driver will be
compatible with RT kernel where it will be forced thread IRQ handler
while in non-RT kernel it still will be executed in HW IRQ context.
As part of this change the IRQ wakeup configuration is applied to
GPIO Bank IRQ as it now will be under control of IRQ PM Core during
suspend.

There are also additional benefits:
 - on-RT kernel there will be no complains any more about PM runtime usage
   in atomic context  "BUG: sleeping function called from invalid context";
 - GPIO bank IRQs will appear in /proc/interrupts and its usage statistic
    will be  visible;
 - GPIO bank IRQs could be configured through IRQ proc_fs interface and,
   as result, could be a part of IRQ balancing process if needed;
 - GPIO bank IRQs will be under control of IRQ PM Core during
   suspend to RAM.

Disadvantage:
 - additional runtime overhed as call chain till
   omap_gpio_irq_handler() will be longer now
 - necessity to use wa_lock in omap_gpio_irq_handler() to W/A warning
   in handle_irq_event_percpu()
   WARNING: CPU: 1 PID: 35 at kernel/irq/handle.c:149 handle_irq_event_percpu+0x51c/0x638()

This patch doesn't fully follows recommendations provided by Sebastian
Andrzej Siewior [1], because It's required to go through and check all
GPIO IRQ pin states as fast as possible and pass control to handle_level_irq
or handle_edge_irq. handle_level_irq or handle_edge_irq will perform actions
specific for IRQ triggering type and wakeup corresponding registered
threaded IRQ handler (at least it's expected to be threaded).
IRQs can be lost if handle_nested_irq() will be used, because excecution
time of some pin specific GPIO IRQ handler can be very significant and
require accessing ext. devices (I2C).

Idea of such kind reworking was also discussed in [2].

[1] http://www.spinics.net/lists/linux-omap/msg120665.html
[2] http://www.spinics.net/lists/linux-omap/msg119516.html

Cc: <linux-rt-users@vger.kernel.org>
Signed-off-by: Grygorii Strashko <grygorii.strashko@ti.com>
---
 drivers/gpio/gpio-omap.c | 55 ++++++++++++++++++++++++------------------------
 1 file changed, 27 insertions(+), 28 deletions(-)

diff --git a/drivers/gpio/gpio-omap.c b/drivers/gpio/gpio-omap.c
index c33595e..0ef0180 100644
--- a/drivers/gpio/gpio-omap.c
+++ b/drivers/gpio/gpio-omap.c
@@ -59,6 +59,7 @@ struct gpio_bank {
 	u32 level_mask;
 	u32 toggle_mask;
 	raw_spinlock_t lock;
+	raw_spinlock_t wa_lock;
 	struct gpio_chip chip;
 	struct clk *dbck;
 	u32 mod_usage;
@@ -649,8 +650,13 @@ static int omap_gpio_wake_enable(struct irq_data *d, unsigned int enable)
 {
 	struct gpio_bank *bank = omap_irq_data_get_bank(d);
 	unsigned offset = d->hwirq;
+	int ret;
+
+	ret = omap_set_gpio_wakeup(bank, offset, enable);
+	if (!ret)
+		ret = irq_set_irq_wake(bank->irq, enable);
 
-	return omap_set_gpio_wakeup(bank, offset, enable);
+	return ret;
 }
 
 static int omap_gpio_request(struct gpio_chip *chip, unsigned offset)
@@ -704,26 +710,21 @@ static void omap_gpio_free(struct gpio_chip *chip, unsigned offset)
  * line's interrupt handler has been run, we may miss some nested
  * interrupts.
  */
-static void omap_gpio_irq_handler(unsigned int irq, struct irq_desc *desc)
+static irqreturn_t omap_gpio_irq_handler(int irq, void *gpiobank)
 {
 	void __iomem *isr_reg = NULL;
 	u32 isr;
 	unsigned int bit;
-	struct gpio_bank *bank;
-	int unmasked = 0;
-	struct irq_chip *irqchip = irq_desc_get_chip(desc);
-	struct gpio_chip *chip = irq_desc_get_handler_data(desc);
+	struct gpio_bank *bank = gpiobank;
+	unsigned long wa_lock_flags;
 	unsigned long lock_flags;
 
-	chained_irq_enter(irqchip, desc);
-
-	bank = container_of(chip, struct gpio_bank, chip);
 	isr_reg = bank->base + bank->regs->irqstatus;
-	pm_runtime_get_sync(bank->dev);
-
 	if (WARN_ON(!isr_reg))
 		goto exit;
 
+	pm_runtime_get_sync(bank->dev);
+
 	while (1) {
 		u32 isr_saved, level_mask = 0;
 		u32 enabled;
@@ -745,13 +746,6 @@ static void omap_gpio_irq_handler(unsigned int irq, struct irq_desc *desc)
 
 		raw_spin_unlock_irqrestore(&bank->lock, lock_flags);
 
-		/* if there is only edge sensitive GPIO pin interrupts
-		configured, we could unmask GPIO bank interrupt immediately */
-		if (!level_mask && !unmasked) {
-			unmasked = 1;
-			chained_irq_exit(irqchip, desc);
-		}
-
 		if (!isr)
 			break;
 
@@ -772,18 +766,18 @@ static void omap_gpio_irq_handler(unsigned int irq, struct irq_desc *desc)
 
 			raw_spin_unlock_irqrestore(&bank->lock, lock_flags);
 
+			raw_spin_lock_irqsave(&bank->wa_lock, wa_lock_flags);
+
 			generic_handle_irq(irq_find_mapping(bank->chip.irqdomain,
 							    bit));
+
+			raw_spin_unlock_irqrestore(&bank->wa_lock,
+						   wa_lock_flags);
 		}
 	}
-	/* if bank has any level sensitive GPIO pin interrupt
-	configured, we must unmask the bank interrupt only after
-	handler(s) are executed in order to avoid spurious bank
-	interrupt */
 exit:
-	if (!unmasked)
-		chained_irq_exit(irqchip, desc);
 	pm_runtime_put(bank->dev);
+	return IRQ_HANDLED;
 }
 
 static unsigned int omap_gpio_irq_startup(struct irq_data *d)
@@ -1133,7 +1127,7 @@ static int omap_gpio_chip_init(struct gpio_bank *bank, struct irq_chip *irqc)
 	}
 
 	ret = gpiochip_irqchip_add(&bank->chip, irqc,
-				   irq_base, omap_gpio_irq_handler,
+				   irq_base, handle_bad_irq,
 				   IRQ_TYPE_NONE);
 
 	if (ret) {
@@ -1142,10 +1136,14 @@ static int omap_gpio_chip_init(struct gpio_bank *bank, struct irq_chip *irqc)
 		return -ENODEV;
 	}
 
-	gpiochip_set_chained_irqchip(&bank->chip, irqc,
-				     bank->irq, omap_gpio_irq_handler);
+	gpiochip_set_chained_irqchip(&bank->chip, irqc, bank->irq, NULL);
 
-	return 0;
+	ret = devm_request_irq(bank->dev, bank->irq, omap_gpio_irq_handler,
+			       0, dev_name(bank->dev), bank);
+	if (ret)
+		gpiochip_remove(&bank->chip);
+
+	return ret;
 }
 
 static const struct of_device_id omap_gpio_match[];
@@ -1227,6 +1225,7 @@ static int omap_gpio_probe(struct platform_device *pdev)
 		bank->set_dataout = omap_set_gpio_dataout_mask;
 
 	raw_spin_lock_init(&bank->lock);
+	raw_spin_lock_init(&bank->wa_lock);
 
 	/* Static mapping, never released */
 	res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
-- 
2.5.0

Re: [RFC PATCH 6/7] gpio: omap: move pm runtime in irq_chip.irq_bus_lock/sync_unlock

[Linus Walleij]

On Tue, Aug 18, 2015 at 4:10 AM, Grygorii Strashko
<grygorii.strashko@ti.com> wrote:

> The PM runtime API can't be used in atomic contex on -RT even if
> it's configured as irqsafe. As result, below error report can
> be seen when PM runtime API called from IRQ chip's callbacks
> irq_startup/irq_shutdown/irq_set_type, because they are
> protected by RAW spinlock:

Grygorri I have a massive backlog of mail but if patch 6&7 are
still applicable, can you rebase and send me these for the v4.3-rc2+
tree?

Yours,
Linus Walleij

Re: [RFC PATCH 6/7] gpio: omap: move pm runtime in irq_chip.irq_bus_lock/sync_unlock

[Grygorii Strashko]

On 09/24/2015 03:28 PM, Linus Walleij wrote:
> On Tue, Aug 18, 2015 at 4:10 AM, Grygorii Strashko
> <grygorii.strashko@ti.com> wrote:
>
>> The PM runtime API can't be used in atomic contex on -RT even if
>> it's configured as irqsafe. As result, below error report can
>> be seen when PM runtime API called from IRQ chip's callbacks
>> irq_startup/irq_shutdown/irq_set_type, because they are
>> protected by RAW spinlock:
>
> Grygorri I have a massive backlog of mail but if patch 6&7 are
> still applicable, can you rebase and send me these for the v4.3-rc2+
> tree?
>

sure.

Thanks.

-- 
regards,
-grygorii