genirq: do not disable IRQ_WAKEUP marked irqs on suspend
A message that is not in the archive (sent off-list or not yet received): [email protected]
Quoted 7 lines · Thomas Gleixner
commit
0a0c5168df(PM: Introduce functions for suspending and resuming device interrupts) iterates through all interrupts and disables them on the hardware level. Some architectures have functionality implemented to mark an interrupt source as wakeup source for suspend, but the new power management code disables them unconditionally which breaks the resume on interrupt functionality.
Quoted 6 lines
The wakeup interrupts are marked in the status with the
IRQ_WAKEUPbit. Skip the disablement for those interrupts which have theIRQ_WAKEUPbit set.Signed-off-by Thomas Gleixner <[email protected]>
Hi Thomas,
I posted the same patch last month and lost the argument, original thread here:
http://lkml.org/lkml/2009/5/6/549
Kevin
Quoted 19 lines
kernel/irq/pm.c +2
-2@@ -29,7 +29,8 @@ void suspend_device_irqs(void)unsigned long flags;spin_lock_irqsave (&desc->lock, flags); -__disable_irq (desc, irq, true); +if (!(desc->status &IRQ_WAKEUP )) +__disable_irq (desc, irq, true);spin_unlock_irqrestore (&desc->lock, flags); } --To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
Quoted 16 lines · Kevin Hilman
Quoted 7 lines · Thomas Gleixner
commit
0a0c5168df(PM: Introduce functions for suspending and resuming device interrupts) iterates through all interrupts and disables them on the hardware level. Some architectures have functionality implemented to mark an interrupt source as wakeup source for suspend, but the new power management code disables them unconditionally which breaks the resume on interrupt functionality.Quoted 6 lines
The wakeup interrupts are marked in the status with the
IRQ_WAKEUPbit. Skip the disablement for those interrupts which have theIRQ_WAKEUPbit set.Signed-off-by Thomas Gleixner <[email protected]>
Quoted 6 lines
Hi Thomas,
I posted the same patch last month and lost the argument, original thread here:
Err no. Care to look at the difference ?
I missed the above discussion, but I'm revisiting the delayed disable issue.
Quoted 21 lines
Quoted 20 lines
kernel/irq/pm.c +2
-2@@ -29,7 +29,8 @@ void suspend_device_irqs(void)unsigned long flags;spin_lock_irqsave (&desc->lock, flags); -__disable_irq (desc, irq, true); +if (!(desc->status &IRQ_WAKEUP )) +__disable_irq (desc, irq, true);spin_unlock_irqrestore (&desc->lock, flags); } --To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
Quoted 17 lines · Thomas Gleixner
Quoted 16 lines · Kevin Hilman
Quoted 13 lines · Thomas Gleixner
commit
0a0c5168df(PM: Introduce functions for suspending and resuming device interrupts) iterates through all interrupts and disables them on the hardware level. Some architectures have functionality implemented to mark an interrupt source as wakeup source for suspend, but the new power management code disables them unconditionally which breaks the resume on interrupt functionality.The wakeup interrupts are marked in the status with the
IRQ_WAKEUPbit. Skip the disablement for those interrupts which have theIRQ_WAKEUPbit set.Signed-off-by Thomas Gleixner <[email protected]>
Quoted 8 lines
Quoted 6 lines
Hi Thomas,
I posted the same patch last month and lost the argument, original thread here:
Err no. Care to look at the difference ?
Oops, sent link to wrong patch. Here's the one solving the same problem:
http://marc.info/?l=linux-kernel&m=124148347804447&w=2
or
http://lkml.org/lkml/2009/5/4/448
Only difference is I did the checking outside of the lock, which is probably wrong. In any case, you'll be interested in the thread that follows.
Kevin
Quoted 24 lines
I missed the above discussion, but I'm revisiting the delayed disable issue.
Quoted 21 lines
Quoted 20 lines
kernel/irq/pm.c +2
-2@@ -29,7 +29,8 @@ void suspend_device_irqs(void)unsigned long flags;spin_lock_irqsave (&desc->lock, flags); -__disable_irq (desc, irq, true); +if (!(desc->status &IRQ_WAKEUP )) +__disable_irq (desc, irq, true);spin_unlock_irqrestore (&desc->lock, flags); } --To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
Quoted 5 lines · Kevin Hilman
http://lkml.org/lkml/2009/5/4/448
Only difference is I did the checking outside of the lock, which is probably wrong. In any case, you'll be interested in the thread that follows.
Hmm, darn. That means that on hardware which has trouble with the delayed disable and therefor uses it's own chip->disable_irq() method the suspend logic is wreckaged.
But there is always a way to get broken hardware tamed. :)
suspend does:
__disable_irq(); status |= IRQ_SUSPENDED; chip->disable_irq();
resume does:
__enable_irq(); status &= ~IRQ_SUSPENDED; chip->enable_irq();
So
- set_irq_handler(handle_level_irq);
+ set_irq_handler(my_own_handler);
+my_own_handler()
+{
+ if (!(status & IRQ_SUSPENDED)) {
+ handle_level_irq();
+ } else {
+ mask_at_hardware_level(); + status |= IRQ_PENDING; + save_important_information();
+ }
+}
my_disable_irq()
{
+ if (!(status & IRQ_SUSPENDED))
mask_at_hardware_level(); }
my_enable_irq()
{
+ if (important_information_has_been_saved)
+ replay_what_happened();
+
unmask_at_hardware_level(); }
Ugly, but that might work somehow. Not sure about the replay part, but that can be deferred via some more hackery as well :)
Raphael, these delayed disable and the chip->irq_disable() override implications vs. suspend really need to be documented. The current comment of suspend_device_irqs() is bogus:
* During system-wide suspend or hibernation device interrupts need to be
* disabled at the chip level and this function is provided for this purpose. ^^^^^^^^^^^^^^^^^^^^^^^^^^
Thanks,
tglx
Quoted 10 lines · On Fri, 12 Jun 2009 21:52:46 +0200 (CEST) Thomas Gleixner <[email protected]>…
Quoted 5 lines · Kevin Hilman
http://lkml.org/lkml/2009/5/4/448
Only difference is I did the checking outside of the lock, which is probably wrong. In any case, you'll be interested in the thread that follows.
Hmm, darn. That means that on hardware which has trouble with the delayed disable and therefor uses it's own
chip->disable_irq()method the suspend logic is wreckaged.
Does this maen that your original patch is no longer applicable to mainline/-stable?
Quoted 52 lines
But there is always a way to get broken hardware tamed. :)
suspend does:
__disable_irq(); status |= IRQ_SUSPENDED; chip->disable_irq();resume does:
__enable_irq(); status &= ~IRQ_SUSPENDED; chip->enable_irq();So
(quoted diff) +11
-1- set_irq_handler(handle_level_irq); + set_irq_handler(my_own_handler); +my_own_handler() +{ + if (!(status & IRQ_SUSPENDED)) { + handle_level_irq(); + } else { + mask_at_hardware_level(); + status |= IRQ_PENDING; + save_important_information(); + } +}
my_disable_irq()
{(quoted diff) +1
-0+ if (!(status & IRQ_SUSPENDED))mask_at_hardware_level(); }
my_enable_irq()
{(quoted diff) +3
-0+ if (important_information_has_been_saved) + replay_what_happened(); + unmask_at_hardware_level();}
Ugly, but that might work somehow. Not sure about the replay part, but that can be deferred via some more hackery as well :)
Raphael, these delayed disable and the
chip->irq_disable()override implications vs. suspend really need to be documented. The current comment ofsuspend_device_irqs()is bogus:* During system-wide suspend or hibernation device interrupts need to be
* disabled at the chip level and this function is provided for this purpose. ^^^^^^^^^^^^^^^^^^^^^^^^^^
Quoted 10 lines · Andrew Morton
Quoted 7 lines · On Fri, 12 Jun 2009 21:52:46 +0200 (CEST) Thomas Gleixner <[email protected]>…
Quoted 5 lines · Kevin Hilman
http://lkml.org/lkml/2009/5/4/448
Only difference is I did the checking outside of the lock, which is probably wrong. In any case, you'll be interested in the thread that follows.
Quoted 6 lines
Quoted 3 lines
Hmm, darn. That means that on hardware which has trouble with the delayed disable and therefor uses it's own
chip->disable_irq()method the suspend logic is wreckaged.Does this maen that your original patch is no longer applicable to mainline/-stable?
I'd say so. There are good arguments for not doing this.
Quoted 52 lines
Quoted 52 lines
But there is always a way to get broken hardware tamed. :)
suspend does:
__disable_irq(); status |= IRQ_SUSPENDED; chip->disable_irq();resume does:
__enable_irq(); status &= ~IRQ_SUSPENDED; chip->enable_irq();So
(quoted diff) +11
-1- set_irq_handler(handle_level_irq); + set_irq_handler(my_own_handler); +my_own_handler() +{ + if (!(status & IRQ_SUSPENDED)) { + handle_level_irq(); + } else { + mask_at_hardware_level(); + status |= IRQ_PENDING; + save_important_information(); + } +}
my_disable_irq()
{(quoted diff) +1
-0+ if (!(status & IRQ_SUSPENDED))mask_at_hardware_level(); }
my_enable_irq()
{(quoted diff) +3
-0+ if (important_information_has_been_saved) + replay_what_happened(); + unmask_at_hardware_level();}
Ugly, but that might work somehow. Not sure about the replay part, but that can be deferred via some more hackery as well :)
Raphael, these delayed disable and the
chip->irq_disable()override implications vs. suspend really need to be documented. The current comment ofsuspend_device_irqs()is bogus:* During system-wide suspend or hibernation device interrupts need to be
* disabled at the chip level and this function is provided for this purpose. ^^^^^^^^^^^^^^^^^^^^^^^^^^
Quoted 53 lines · Thomas Gleixner
Quoted 5 lines · Kevin Hilman
http://lkml.org/lkml/2009/5/4/448
Only difference is I did the checking outside of the lock, which is probably wrong. In any case, you'll be interested in the thread that follows.
Hmm, darn. That means that on hardware which has trouble with the delayed disable and therefor uses it's own
chip->disable_irq()method the suspend logic is wreckaged.But there is always a way to get broken hardware tamed. :)
suspend does:
__disable_irq(); status |= IRQ_SUSPENDED; chip->disable_irq();resume does:
__enable_irq(); status &= ~IRQ_SUSPENDED; chip->enable_irq();So
(quoted diff) +11
-1- set_irq_handler(handle_level_irq); + set_irq_handler(my_own_handler); +my_own_handler() +{ + if (!(status & IRQ_SUSPENDED)) { + handle_level_irq(); + } else { + mask_at_hardware_level(); + status |= IRQ_PENDING; + save_important_information(); + } +}
my_disable_irq()
{(quoted diff) +1
-0+ if (!(status & IRQ_SUSPENDED))mask_at_hardware_level(); }
my_enable_irq()
{(quoted diff) +3
-0+ if (important_information_has_been_saved) + replay_what_happened(); + unmask_at_hardware_level();}
Quoted 5 lines
Ugly, but that might work somehow. Not sure about the replay part, but that can be deferred via some more hackery as well :)
Raphael, these delayed disable and the
chip->irq_disable()override implications vs. suspend really need to be documented.
Agreed, but can you please suggest what way would be the best?
Quoted 5 lines
The current comment of
suspend_device_irqs()is bogus:* During system-wide suspend or hibernation device interrupts need to be
* disabled at the chip level and this function is provided for this purpose. ^^^^^^^^^^^^^^^^^^^^^^^^^^
Yes, it is, sorry for that. I'll prepare a fix.
Best,
Rafael