Report interrupt(s) that caused system wakeup*
Subsystems (5)
A message that is not in the archive (sent off-list or not yet received): [PATCH V5] Report interrupt(s) that caused a system wakeup
PATCH v5 1/1 Report interrupt(s) that caused system wakeup*
Hi Rafael,
Please accept this patch.
Alexandra Yates (1):
Report interrupt(s) that caused system wakeup
Documentation/ABI/testing/sysfs-power | 11 +++++++++++ drivers/base/power/wakeup.c | 31 +++++++++++++++++++++++++++++++ include/linux/irq.h | 7 +++++++ include/linux/suspend.h | 1 + kernel/irq/pm.c | 2 ++ kernel/power/main.c | 23 +++++++++++++++++++++++ 6 files changed, 75 insertions(+)
Signature · 1 line
1.9.1
PATCH v5 1/1 Report interrupt(s) that caused system wakeup
Add a new IRQ flag named IRQD_WAKEUP_TRIGGERED. This flag is set when a given IRQ fires after it has been armed for system wakeup. The flag is cleared before arming the IRQ for system wakeup during the next system suspend. This feature makes possible to check which IRQs might have woken the system up from sleep last time it was suspended.
Add a new sysfs attribute under /sys/power/ named:pm_last_wakeup_irqs when read, will return a list of IRQs with the new flag set. That will be useful for system wakeup diagnostics.
--- Documentation/ABI/testing/sysfs-power | 11 +++++++++++ drivers/base/power/wakeup.c | 31 +++++++++++++++++++++++++++++++ include/linux/irq.h | 7 +++++++ include/linux/suspend.h | 1 + kernel/irq/pm.c | 2 ++ kernel/power/main.c | 23 +++++++++++++++++++++++ 6 files changed, 75 insertions(+)
Documentation/ABI/testing/sysfs-power +11 -0
@@ -256,3 +256,14 @@ Description: Writing a "1" enables this printing while writing a "0" disables it. The default value is "0". Reading from this file will display the current value. + +What: /sys/power/pm_last_wakeup_irqs +Date: April 2015 +Contact: Alexandra Yates <[email protected]> +Description: + The /sys/power/pm_last_wakeup_irqs file allows user space + to identify and report the IRQs responsible for waking the + system up from sleep. The IRQD_WAKEUP_TRIGGERED flag is set and + reported when the given IRQ fires after it has been armed for + system wakeup. This output is useful for system wakeup + diagnostics.
drivers/base/power/wakeup.c +31 -0
@@ -870,6 +870,37 @@ void pm_wakeup_clear(void) pm_abort_suspend = false; } +#ifdef CONFIG_PM_SLEEP_DEBUG +/* +* pm_get_last_wakeup_irqs - Parses interrupts to identify which one +* caused the system to wake up from suspend-to-idle. +* @buf: keeps track of the irqs that casued the system to wakeup +*/ +ssize_t pm_get_last_wakeup_irqs (char *buf,size_t size) +{ +struct irq_desc *desc; +int irq; +char *str = buf; +char *end = buf + size; + +if (!pm_abort_suspend) +return 0 ; + +/* If pm_abort_suspend is not set, the previous suspend was aborted +* before arming the wakeup IRQs, so avoid printing stale information +* in that case. +*/ +for_each_irq_desc (irq, desc) +if (irqd_triggered_wakeup (&desc->irq_data)) + str +=scnprintf (str, end - str,"%d " , irq); + +if (str != buf) + str--; + +return (str - buf); +} +#endif /* CONFIG_PM_SLEEP_DEBUG */ +/** * pm_get_wakeup_count - Read the number of registered wakeup events. * @count: Address to store the value at.
include/linux/irq.h +7 -0
@@ -190,6 +190,7 @@ struct irq_data {* IRQD_IRQ_MASKED - Masked state of the interrupt * IRQD_IRQ_INPROGRESS - In progress state of the interrupt * IRQD_WAKEUP_ARMED - Wakeup mode armed +* IRQD_WAKEUP_TRIGGERED - Triggered system wakeup */ enum {IRQD_TRIGGER_MASK =0xf ,
@@ -204,10 +205,16 @@ enum {IRQD_IRQ_MASKED = (1 <<17 ),IRQD_IRQ_INPROGRESS = (1 <<18 ),IRQD_WAKEUP_ARMED = (1 <<19 ), + IRQD_WAKEUP_TRIGGERED = (1 << 20), };#define __irqd_to_state (d) ((d)->common->state_use_accessors) +static inline bool irqd_triggered_wakeup (struct irq_data *d) +{ +return __irqd_to_state (d) &IRQD_WAKEUP_TRIGGERED ; +} +static inline bool irqd_is_setaffinity_pending (struct irq_data *d) {return __irqd_to_state (d) &IRQD_SETAFFINITY_PENDING ;
include/linux/suspend.h +1 -0
@@ -378,6 +378,7 @@ void restore_processor_state(void);/* kernel/power/main.c */ extern int register_pm_notifier (struct notifier_block *nb);extern int unregister_pm_notifier (struct notifier_block *nb); +extern ssize_t pm_get_last_wakeup_irqs (char *buf,size_t size);#define pm_notifier (fn, pri) { \ static struct notifier_block fn##_nb = \
kernel/irq/pm.c +2 -0
@@ -22,6 +22,7 @@ bool irq_pm_check_wakeup(struct irq_desc *desc) desc->depth++;irq_disable (desc);pm_system_wakeup (); +irqd_set (&desc->irq_data,IRQD_WAKEUP_TRIGGERED );return true; }return false;
@@ -73,6 +74,7 @@ static bool suspend_device_irq(struct irq_desc *desc, int irq)if (!desc->action || desc->no_suspend_depth)return false; +irqd_clear (&desc->irq_data,IRQD_WAKEUP_TRIGGERED );if (irqd_is_wakeup_set (&desc->irq_data)) {irqd_set (&desc->irq_data,IRQD_WAKEUP_ARMED );/*
kernel/power/main.c +23 -0
@@ -272,6 +272,28 @@ static inline void pm_print_times_init(void) { pm_print_times_enabled = !!initcall_debug; } + +static ssize_t pm_last_wakeup_irqs_show (struct kobject *kobj, +struct kobj_attribute *attr, +char *buf) +{ +size_t ret; + + ret =pm_get_last_wakeup_irqs (buf,PAGE_SIZE ); +if (ret) + buf[ret++] ='\n' ; + +return ret; +} + +static ssize_t pm_last_wakeup_irqs_store (struct kobject *kobj, +struct kobj_attribute *attr, +const char *buf,size_t n) +{ +return -EINVAL ; +} +power_attr (pm_last_wakeup_irqs); +#else /* !CONFIG_PM_SLEEP_DEBUG */ static inline void pm_print_times_init (void ) {}#endif /* CONFIG_PM_SLEEP_DEBUG */
@@ -604,6 +626,7 @@ static struct attribute * g[] = {#endif #ifdef CONFIG_PM_SLEEP_DEBUG &pm_print_times_attr.attr, + &pm_last_wakeup_irqs_attr.attr,#endif #endif #ifdef CONFIG_FREEZER
Signature · 1 line
1.9.1
Re: [PATCH v5] Report interrupt(s) that caused system wakeup
Quoted 7 lines · Alexandra Yates
Add a new IRQ flag named
IRQD_WAKEUP_TRIGGERED. This flag is set when a given IRQ fires after it has been armed for system wakeup. The flag is cleared before arming the IRQ for system wakeup during the next system suspend. This feature makes possible to check which IRQs might have woken the system up from sleep last time it was suspended.
Quoted 5 lines
Add a new sysfs attribute under
/sys/power/named:pm_last_wakeup_irqs when read, will return a list of IRQs with the new flag set. That will be useful for system wakeup diagnostics.Signed-off-by Alexandra Yates <[email protected]>
Thanks for the patch, but you've forgotten to CC the IRQ subsystem maintainer.
Hi Thomas, can you please have a look at the patch below and tell me if you have any concerns about it?
Rafael
Quoted 173 lines
--- Documentation/ABI/testing/sysfs-power | 11 +++++++++++ drivers/base/power/wakeup.c | 31 +++++++++++++++++++++++++++++++ include/linux/irq.h | 7 +++++++ include/linux/suspend.h | 1 + kernel/irq/pm.c | 2 ++ kernel/power/main.c | 23 +++++++++++++++++++++++ 6 files changed, 75 insertions(+)Documentation/ABI/testing/sysfs-power +11
-0@@ -256,3 +256,14 @@ Description: Writing a "1" enables this printing while writing a "0" disables it. The default value is "0". Reading from this file will display the current value. + +What: /sys/power/pm_last_wakeup_irqs +Date: April 2015 +Contact: Alexandra Yates <[email protected]> +Description: + The /sys/power/pm_last_wakeup_irqs file allows user space + to identify and report the IRQs responsible for waking the + system up from sleep. The IRQD_WAKEUP_TRIGGERED flag is set and + reported when the given IRQ fires after it has been armed for + system wakeup. This output is useful for system wakeup + diagnostics.drivers/base/power/wakeup.c +31
-0@@ -870,6 +870,37 @@ void pm_wakeup_clear(void) pm_abort_suspend = false; } +#ifdef CONFIG_PM_SLEEP_DEBUG +/* +* pm_get_last_wakeup_irqs - Parses interrupts to identify which one +* caused the system to wake up from suspend-to-idle. +* @buf: keeps track of the irqs that casued the system to wakeup +*/ +ssize_t pm_get_last_wakeup_irqs (char *buf,size_t size) +{ +struct irq_desc *desc; +int irq; +char *str = buf; +char *end = buf + size; + +if (!pm_abort_suspend) +return 0 ; + +/* If pm_abort_suspend is not set, the previous suspend was aborted +* before arming the wakeup IRQs, so avoid printing stale information +* in that case. +*/ +for_each_irq_desc (irq, desc) +if (irqd_triggered_wakeup (&desc->irq_data)) + str +=scnprintf (str, end - str,"%d " , irq); + +if (str != buf) + str--; + +return (str - buf); +} +#endif /* CONFIG_PM_SLEEP_DEBUG */ +/** * pm_get_wakeup_count - Read the number of registered wakeup events. * @count: Address to store the value at. include/linux/irq.h +7
-0@@ -190,6 +190,7 @@ struct irq_data {* IRQD_IRQ_MASKED - Masked state of the interrupt * IRQD_IRQ_INPROGRESS - In progress state of the interrupt * IRQD_WAKEUP_ARMED - Wakeup mode armed +* IRQD_WAKEUP_TRIGGERED - Triggered system wakeup */ enum {IRQD_TRIGGER_MASK =0xf ,@@ -204,10 +205,16 @@ enum {IRQD_IRQ_MASKED = (1 <<17 ),IRQD_IRQ_INPROGRESS = (1 <<18 ),IRQD_WAKEUP_ARMED = (1 <<19 ), + IRQD_WAKEUP_TRIGGERED = (1 << 20), };#define __irqd_to_state (d) ((d)->common->state_use_accessors) +static inline bool irqd_triggered_wakeup (struct irq_data *d) +{ +return __irqd_to_state (d) &IRQD_WAKEUP_TRIGGERED ; +} +static inline bool irqd_is_setaffinity_pending (struct irq_data *d) {return __irqd_to_state (d) &IRQD_SETAFFINITY_PENDING ;include/linux/suspend.h +1
-0@@ -378,6 +378,7 @@ void restore_processor_state(void);/* kernel/power/main.c */ extern int register_pm_notifier (struct notifier_block *nb);extern int unregister_pm_notifier (struct notifier_block *nb); +extern ssize_t pm_get_last_wakeup_irqs (char *buf,size_t size);#define pm_notifier (fn, pri) { \ static struct notifier_block fn##_nb = \kernel/irq/pm.c +2
-0@@ -22,6 +22,7 @@ bool irq_pm_check_wakeup(struct irq_desc *desc) desc->depth++;irq_disable (desc);pm_system_wakeup (); +irqd_set (&desc->irq_data,IRQD_WAKEUP_TRIGGERED );return true; }return false;@@ -73,6 +74,7 @@ static bool suspend_device_irq(struct irq_desc *desc, int irq)if (!desc->action || desc->no_suspend_depth)return false; +irqd_clear (&desc->irq_data,IRQD_WAKEUP_TRIGGERED );if (irqd_is_wakeup_set (&desc->irq_data)) {irqd_set (&desc->irq_data,IRQD_WAKEUP_ARMED );/* kernel/power/main.c +23
-0@@ -272,6 +272,28 @@ static inline void pm_print_times_init(void) { pm_print_times_enabled = !!initcall_debug; } + +static ssize_t pm_last_wakeup_irqs_show (struct kobject *kobj, +struct kobj_attribute *attr, +char *buf) +{ +size_t ret; + + ret =pm_get_last_wakeup_irqs (buf,PAGE_SIZE ); +if (ret) + buf[ret++] ='\n' ; + +return ret; +} + +static ssize_t pm_last_wakeup_irqs_store (struct kobject *kobj, +struct kobj_attribute *attr, +const char *buf,size_t n) +{ +return -EINVAL ; +} +power_attr (pm_last_wakeup_irqs); +#else /* !CONFIG_PM_SLEEP_DEBUG */ static inline void pm_print_times_init (void ) {}#endif /* CONFIG_PM_SLEEP_DEBUG */ @@ -604,6 +626,7 @@ static struct attribute * g[] = {#endif #ifdef CONFIG_PM_SLEEP_DEBUG &pm_print_times_attr.attr, + &pm_last_wakeup_irqs_attr.attr,#endif #endif #ifdef CONFIG_FREEZER
Re: [PATCH v5] Report interrupt(s) that caused system wakeup
Quoted 12 lines · Rafael J. Wysocki
Quoted 11 lines · Alexandra Yates
Add a new IRQ flag named
IRQD_WAKEUP_TRIGGERED. This flag is set when a given IRQ fires after it has been armed for system wakeup. The flag is cleared before arming the IRQ for system wakeup during the next system suspend. This feature makes possible to check which IRQs might have woken the system up from sleep last time it was suspended.Add a new sysfs attribute under
/sys/power/named:pm_last_wakeup_irqs when read, will return a list of IRQs with the new flag set. That will be useful for system wakeup diagnostics.
Quoted 6 lines
Quoted 1 line
Signed-off-by Alexandra Yates <[email protected]>Thanks for the patch, but you've forgotten to CC the IRQ subsystem maintainer.
Hi Thomas, can you please have a look at the patch below and tell me if you have any concerns about it?
Sure.
Quoted 23 lines
Quoted 23 lines
drivers/base/power/wakeup.c +15
-0@@ -870,6 +870,37 @@ void pm_wakeup_clear(void) pm_abort_suspend = false; } +#ifdef CONFIG_PM_SLEEP_DEBUG +/* +* pm_get_last_wakeup_irqs - Parses interrupts to identify which one +* caused the system to wake up from suspend-to-idle. +* @buf: keeps track of the irqs that casued the system to wakeup +*/ +ssize_t pm_get_last_wakeup_irqs (char *buf,size_t size) +{ +struct irq_desc *desc; +int irq; +char *str = buf; +char *end = buf + size; + +if (!pm_abort_suspend) +return 0 ;
Quoted 8 lines
drivers/base/power/wakeup.c void pm_wakeup_clear(void) · line 895
Quoted 8 lines
drivers/base/power/wakeup.c +8
-0+ +/* If pm_abort_suspend is not set, the previous suspend was aborted +* before arming the wakeup IRQs, so avoid printing stale information +* in that case. +*/ +for_each_irq_desc (irq, desc) +if (irqd_triggered_wakeup (&desc->irq_data)) + str +=scnprintf (str, end - str,"%d " , irq);
This iteration needs to be protected by irq_lock_sparse().
Quoted 14 lines
Quoted 14 lines
kernel/irq/pm.c +1
-0@@ -22,6 +22,7 @@ bool irq_pm_check_wakeup(struct irq_desc *desc) desc->depth++;irq_disable (desc);pm_system_wakeup (); +irqd_set (&desc->irq_data,IRQD_WAKEUP_TRIGGERED );return true; }return false;@@ -73,6 +74,7 @@ static bool suspend_device_irq(struct irq_desc *desc, int irq)if (!desc->action || desc->no_suspend_depth)return false;
Quoted 1 line
kernel/irq/pm.c static bool suspend_device_irq(struct irq_desc *desc, int irq) · line 77
Quoted 1 line
kernel/irq/pm.c +1
-0+irqd_clear (&desc->irq_data,IRQD_WAKEUP_TRIGGERED );
So that leaves a stale IRQD_WAKEUP_TRIGGERED bit in the following situation:
suspend() wakeup_triggered_by_irq(X) resume() ... close_device() free_irq(X) suspend() wakeup_triggered_by_irq(Y) resume()
So the next readout will show irqs X and Y being responsible for waking up the system.
I really have to ask why this bit fiddling is necessary at all. Isn't the really interesting information the first interrupt which triggered the resume? If we can reduce it to that information then this whole thing can be simplified into a few lines of code.
Thanks,
tglx
Re: [PATCH v5] Report interrupt(s) that caused system wakeup
Quoted 47 lines · Thomas Gleixner
Quoted 12 lines · Rafael J. Wysocki
Quoted 11 lines · Alexandra Yates
Add a new IRQ flag named
IRQD_WAKEUP_TRIGGERED. This flag is set when a given IRQ fires after it has been armed for system wakeup. The flag is cleared before arming the IRQ for system wakeup during the next system suspend. This feature makes possible to check which IRQs might have woken the system up from sleep last time it was suspended.Add a new sysfs attribute under
/sys/power/named:pm_last_wakeup_irqs when read, will return a list of IRQs with the new flag set. That will be useful for system wakeup diagnostics.Quoted 6 lines
Quoted 1 line
Signed-off-by Alexandra Yates <[email protected]>Thanks for the patch, but you've forgotten to CC the IRQ subsystem maintainer.
Hi Thomas, can you please have a look at the patch below and tell me if you have any concerns about it?
Sure.
Quoted 25 lines
Quoted 25 lines
drivers/base/power/wakeup.c +17
-0@@ -870,6 +870,37 @@ void pm_wakeup_clear(void) pm_abort_suspend = false; } +#ifdef CONFIG_PM_SLEEP_DEBUG +/* +* pm_get_last_wakeup_irqs - Parses interrupts to identify which one +* caused the system to wake up from suspend-to-idle. +* @buf: keeps track of the irqs that casued the system to wakeup +*/ +ssize_t pm_get_last_wakeup_irqs (char *buf,size_t size) +{ +struct irq_desc *desc; +int irq; +char *str = buf; +char *end = buf + size; + +if (!pm_abort_suspend) +return 0 ; + +/* If pm_abort_suspend is not set, the previous suspend was aborted
Quoted 8 lines
Quoted 6 lines
drivers/base/power/wakeup.c void pm_wakeup_clear(void) · line 895
Quoted 6 lines
drivers/base/power/wakeup.c +6
-0+* before arming the wakeup IRQs, so avoid printing stale information +* in that case. +*/ +for_each_irq_desc (irq, desc) +if (irqd_triggered_wakeup (&desc->irq_data)) + str +=scnprintf (str, end - str,"%d " , irq);This iteration needs to be protected by
irq_lock_sparse().
Right.
Quoted 30 lines
Quoted 14 lines
Quoted 14 lines
kernel/irq/pm.c +1
-0@@ -22,6 +22,7 @@ bool irq_pm_check_wakeup(struct irq_desc *desc) desc->depth++;irq_disable (desc);pm_system_wakeup (); +irqd_set (&desc->irq_data,IRQD_WAKEUP_TRIGGERED );return true; }return false;@@ -73,6 +74,7 @@ static bool suspend_device_irq(struct irq_desc *desc, int irq)if (!desc->action || desc->no_suspend_depth)return false;Quoted 1 line
kernel/irq/pm.c static bool suspend_device_irq(struct irq_desc *desc, int irq) · line 77
Quoted 1 line
kernel/irq/pm.c +1
-0+irqd_clear (&desc->irq_data,IRQD_WAKEUP_TRIGGERED );So that leaves a stale
IRQD_WAKEUP_TRIGGEREDbit in the following situation:suspend() wakeup_triggered_by_irq(X) resume() ... close_device() free_irq(X) suspend() wakeup_triggered_by_irq(Y) resume()
Quoted 2 lines
So the next readout will show irqs X and Y being responsible for waking up the system.
Right.
Quoted 4 lines
I really have to ask why this bit fiddling is necessary at all. Isn't the really interesting information the first interrupt which triggered the resume? If we can reduce it to that information then this whole thing can be simplified into a few lines of code.
The original point was that if two wakeup interrupts happened at the same time, it would be kind of moot which one was the "real" wakeup, but now that I think about it, reporting the first one should be enough to catch suprious wakeups anyway.
Thanks,
Rafael
Re: [PATCH v5] Report interrupt(s) that caused system wakeup
Quoted 4 lines · Rafael J. Wysocki
The original point was that if two wakeup interrupts happened at the same time, it would be kind of moot which one was the "real" wakeup, but now that I think about it, reporting the first one should be enough to catch suprious wakeups anyway.
So we can simply do the following:
kernel/irq/pm.c +1 -1
@@ -21,7 +21,7 @@ bool irq_pm_check_wakeup(struct irq_desc *desc) desc->istate |=IRQS_SUSPENDED |IRQS_PENDING ; desc->depth++;irq_disable (desc); -pm_system_wakeup (); +pm_system_irq_wakeup (irq_desc_get_irq (desc));return true; }return false;
and manage the storage in the PM code.
Thanks,
tglx
Rafael J. Wysocki
maintainerRe: [PATCH v5] Report interrupt(s) that caused system wakeup
Quoted 22 lines · Thomas Gleixner
Quoted 4 lines · Rafael J. Wysocki
The original point was that if two wakeup interrupts happened at the same time, it would be kind of moot which one was the "real" wakeup, but now that I think about it, reporting the first one should be enough to catch suprious wakeups anyway.
So we can simply do the following:
kernel/irq/pm.c +1
-1@@ -21,7 +21,7 @@ bool irq_pm_check_wakeup(struct irq_desc *desc) desc->istate |=IRQS_SUSPENDED |IRQS_PENDING ; desc->depth++;irq_disable (desc); -pm_system_wakeup (); +pm_system_irq_wakeup (irq_desc_get_irq (desc));return true; }return false;
Quoted 1 line
and manage the storage in the PM code.
Right.
That's exactly the approach we're going to follow. :-)
Thanks,
Rafael