[PATCH 1/2] mfd: twl6030: fix lockdep recursion warning on setting wake IRQs

[PATCH 1/2] mfd: twl6030: fix lockdep recursion warning on setting wake IRQs

[Todd Poynor]

LOCKDEP explicitly sets all irq_desc locks as a single lock-class,
causing "possible recursive locking detected" when the TWL RTC
driver calls through enable_irq_wake to twl6030_irq_set_wake,
which recursively calls irq_set_irq_wake.  Although the
irq_desc and lock are different, LOCKDEP treats these as
equivalent, presumably due to problems that can be incurred
when locking more than one irq_desc, so best to avoid this.

Suspend/resume actions implemented as PM notifiers to avoid
touch the TWL core for this.

Signed-off-by: Todd Poynor <toddpoynor@google.com>
---

This patch applies on top of the patch from Santosh:
[PATCH] mfd: twl6030: Setup irq_wake infrastructure.

If PM notifiers aren't preferred then could change
twl-core to add suspend/resume callbacks, was trying to
make a lower-impact change.

This complexity assumes there are systems that include twl6030
and don't want its IRQs to act as wakeup.  Alternatively,
could just always enable wake for twl6030's IRQ.

 drivers/mfd/twl6030-irq.c |   46 +++++++++++++++++++++++++++++++++++++++++++-
 1 files changed, 44 insertions(+), 2 deletions(-)

diff --git a/drivers/mfd/twl6030-irq.c b/drivers/mfd/twl6030-irq.c
index 4477134..7d6553c 100644
--- a/drivers/mfd/twl6030-irq.c
+++ b/drivers/mfd/twl6030-irq.c
@@ -37,6 +37,7 @@
 #include <linux/kthread.h>
 #include <linux/i2c/twl.h>
 #include <linux/platform_device.h>
+#include <linux/suspend.h>
 
 #include "twl-core.h"
 
@@ -83,8 +84,42 @@ static int twl6030_interrupt_mapping[24] = {
 /*----------------------------------------------------------------------*/
 
 static unsigned twl6030_irq_base;
+static int twl_irq;
+static bool twl_irq_wake_enabled;
 
 static struct completion irq_event;
+static atomic_t twl6030_wakeirqs = ATOMIC_INIT(0);
+
+static int twl6030_irq_pm_notifier(struct notifier_block *notifier,
+				   unsigned long pm_event, void *unused)
+{
+	int chained_wakeups;
+
+	switch (pm_event) {
+	case PM_SUSPEND_PREPARE:
+		chained_wakeups = atomic_read(&twl6030_wakeirqs);
+
+		if (chained_wakeups && !twl_irq_wake_enabled) {
+			if (enable_irq_wake(twl_irq))
+				pr_err("twl6030 IRQ wake enable failed\n");
+			else
+				twl_irq_wake_enabled = true;
+		} else if (!chained_wakeups && twl_irq_wake_enabled) {
+			disable_irq_wake(twl_irq);
+			twl_irq_wake_enabled = false;
+		}
+
+		break;
+	default:
+		break;
+	}
+
+	return NOTIFY_DONE;
+}
+
+static struct notifier_block twl6030_irq_pm_notifier_block = {
+	.notifier_call = twl6030_irq_pm_notifier,
+};
 
 /*
  * This thread processes interrupts reported by the Primary Interrupt Handler.
@@ -189,9 +224,12 @@ static inline void activate_irq(int irq)
 
 int twl6030_irq_set_wake(struct irq_data *d, unsigned int on)
 {
-	int twl_irq = (int)irq_get_chip_data(d->irq);
+	if (on)
+		atomic_inc(&twl6030_wakeirqs);
+	else
+		atomic_dec(&twl6030_wakeirqs);
 
-	return irq_set_irq_wake(twl_irq, on);
+	return 0;
 }
 
 /*----------------------------------------------------------------------*/
@@ -353,6 +391,9 @@ int twl6030_init_irq(int irq_num, unsigned irq_base, unsigned irq_end)
 		pr_err("twl6030: could not claim irq%d: %d\n", irq_num, status);
 		goto fail_irq;
 	}
+
+	twl_irq = irq_num;
+	register_pm_notifier(&twl6030_irq_pm_notifier_block);
 	return status;
 fail_irq:
 	free_irq(irq_num, &irq_event);
@@ -365,6 +406,7 @@ fail_kthread:
 
 int twl6030_exit_irq(void)
 {
+	unregister_pm_notifier(&twl6030_irq_pm_notifier_block);
 
 	if (twl6030_irq_base) {
 		pr_err("twl6030: can't yet clean up IRQs?\n");
-- 
1.7.3.1

[PATCH 2/2] mfd: twl6030: Disable IRQ during suspend

[Todd Poynor]

Module IRQs may still be disabled by DPM at the time the TWL6030
ISR runs, causing handle_simple_irq() to silently do nothing.
This may result in missing TWL RTC alarm wakeups, for example,
since the RTC child module ISR is not called to ack the IRQ.

Disable the TWL6030 IRQ during suspend, enable it at DPM resume
time, at which time the child module IRQs will be re-enabled.

Signed-off-by: Todd Poynor <toddpoynor@google.com>
---
 drivers/mfd/twl6030-irq.c |    6 ++++++
 1 files changed, 6 insertions(+), 0 deletions(-)

diff --git a/drivers/mfd/twl6030-irq.c b/drivers/mfd/twl6030-irq.c
index 7d6553c..5c08ecd 100644
--- a/drivers/mfd/twl6030-irq.c
+++ b/drivers/mfd/twl6030-irq.c
@@ -109,7 +109,13 @@ static int twl6030_irq_pm_notifier(struct notifier_block *notifier,
 			twl_irq_wake_enabled = false;
 		}
 
+		disable_irq(twl_irq);
 		break;
+
+	case PM_POST_SUSPEND:
+		enable_irq(twl_irq);
+		break;
+
 	default:
 		break;
 	}
-- 
1.7.3.1

Re: [PATCH 1/2] mfd: twl6030: fix lockdep recursion warning on setting wake IRQs

[Santosh Shilimkar]

On Tuesday 27 September 2011 05:14 AM, Todd Poynor wrote:
> LOCKDEP explicitly sets all irq_desc locks as a single lock-class,
> causing "possible recursive locking detected" when the TWL RTC
> driver calls through enable_irq_wake to twl6030_irq_set_wake,
> which recursively calls irq_set_irq_wake.  Although the
> irq_desc and lock are different, LOCKDEP treats these as
> equivalent, presumably due to problems that can be incurred
> when locking more than one irq_desc, so best to avoid this.
> 
> Suspend/resume actions implemented as PM notifiers to avoid
> touch the TWL core for this.
> 
> Signed-off-by: Todd Poynor <toddpoynor@google.com>
> ---
> 
> This patch applies on top of the patch from Santosh:
> [PATCH] mfd: twl6030: Setup irq_wake infrastructure.
> 
> If PM notifiers aren't preferred then could change
> twl-core to add suspend/resume callbacks, was trying to
> make a lower-impact change.
> 
> This complexity assumes there are systems that include twl6030
> and don't want its IRQs to act as wakeup.  Alternatively,
> could just always enable wake for twl6030's IRQ.
> 
Both options would mostly end up with same lines
of code :)
The patch looks good to me .
Acked-by: Santosh Shilimkar <santosh.shilimkar@ti.com>	

Regards
Santosh

Re: [PATCH 2/2] mfd: twl6030: Disable IRQ during suspend

[Santosh Shilimkar]

On Tuesday 27 September 2011 05:14 AM, Todd Poynor wrote:
> Module IRQs may still be disabled by DPM at the time the TWL6030
> ISR runs, causing handle_simple_irq() to silently do nothing.
> This may result in missing TWL RTC alarm wakeups, for example,
> since the RTC child module ISR is not called to ack the IRQ.
> 
> Disable the TWL6030 IRQ during suspend, enable it at DPM resume
> time, at which time the child module IRQs will be re-enabled.
> 
> Signed-off-by: Todd Poynor <toddpoynor@google.com>
Acked-by: Santosh Shilimkar <santosh.shilimkar@ti.com>	

Regards
Santosh

Re: [PATCH 1/2] mfd: twl6030: fix lockdep recursion warning on setting wake IRQs

[Samuel Ortiz]

Hi Todd,

On Mon, Sep 26, 2011 at 04:44:23PM -0700, Todd Poynor wrote:
> LOCKDEP explicitly sets all irq_desc locks as a single lock-class,
> causing "possible recursive locking detected" when the TWL RTC
> driver calls through enable_irq_wake to twl6030_irq_set_wake,
> which recursively calls irq_set_irq_wake.  Although the
> irq_desc and lock are different, LOCKDEP treats these as
> equivalent, presumably due to problems that can be incurred
> when locking more than one irq_desc, so best to avoid this.
> 
> Suspend/resume actions implemented as PM notifiers to avoid
> touch the TWL core for this.
Thanks for that, I applied this patch.

Cheers,
Samuel.

-- 
Intel Open Source Technology Centre
http://oss.intel.com/

Re: [PATCH 2/2] mfd: twl6030: Disable IRQ during suspend

[Samuel Ortiz]

Hi Todd,

On Mon, Sep 26, 2011 at 04:44:24PM -0700, Todd Poynor wrote:
> Module IRQs may still be disabled by DPM at the time the TWL6030
> ISR runs, causing handle_simple_irq() to silently do nothing.
> This may result in missing TWL RTC alarm wakeups, for example,
> since the RTC child module ISR is not called to ack the IRQ.
> 
> Disable the TWL6030 IRQ during suspend, enable it at DPM resume
> time, at which time the child module IRQs will be re-enabled.
Applied as well, thanks.

Cheers,
Samuel.

-- 
Intel Open Source Technology Centre
http://oss.intel.com/