Skip to contentSkip to conversation

x86: Sanitize apb timer interrupt handling

tip:irq/core Jacob Pan 2 participants 6 messages Oct 12, 2010 – Oct 13, 2010 1 message not in the archive

lore
Share & tools
lore

A message that is not in the archive (sent off-list or not yet received): [email protected]

Jacob Pan

author
#1 lore raw
Quoted 4 lines

x86: Sanitize apb timer interrupt handling

Disable the interrupt in CPU_DEAD where it belongs.

My main concern is the performance cost. The power management code for Moorestown system make use of the cpu hotplug code (disable_nonboot_cpus) but much more frequently. The system low power states are call S0 idle state (s0ix).

Leaving the irq enabled at the chip and desc level between S0ix states might give some performance benefit. That was my original thought. Will it cause problems?

Thanks,

Jacob

#2 lore raw
Quoted 4 lines · Jacob Pan
Quoted 4 lines

x86: Sanitize apb timer interrupt handling

Disable the interrupt in CPU_DEAD where it belongs.

Quoted 8 lines

My main concern is the performance cost. The power management code for Moorestown system make use of the cpu hotplug code (disable_nonboot_cpus) but much more frequently. The system low power states are call S0 idle state (s0ix).

Leaving the irq enabled at the chip and desc level between S0ix states might give some performance benefit. That was my original thought. Will it cause problems?

Errm. I merily moved it to the place where it should be. You do the disable/enable dance already today.

Thanks,

tglx

author
#3 lore raw

Thomas Gleixner Tue, 12 Oct 2010 23:28:19 +0200 (CEST)

Quoted 11 lines · Jacob Pan
Quoted 10 lines
Quoted 5 lines

x86: Sanitize apb timer interrupt handling

Disable the interrupt in CPU_DEAD where it belongs.

My main concern is the performance cost. The power management code for Moorestown system make use of the cpu hotplug code (disable_nonboot_cpus) but much more frequently. The system low power states are call S0 idle state (s0ix).

Quoted 7 lines
Quoted 3 lines

Leaving the irq enabled at the chip and desc level between S0ix states might give some performance benefit. That was my original thought. Will it cause problems?

Errm. I merily moved it to the place where it should be. You do the disable/enable dance already today.

I think I only do disable/enable at the timer HW level today during cpu hp notification, not calling disable_irq(). Am i missing something?

Thanks,

Jacob

#4 lore raw
Quoted 16 lines · Jacob Pan

Thomas Gleixner Tue, 12 Oct 2010 23:28:19 +0200 (CEST)

Quoted 15 lines · Jacob Pan
Quoted 13 lines
Quoted 5 lines

x86: Sanitize apb timer interrupt handling

Disable the interrupt in CPU_DEAD where it belongs.

My main concern is the performance cost. The power management code for Moorestown system make use of the cpu hotplug code (disable_nonboot_cpus) but much more frequently. The system low power states are call S0 idle state (s0ix).

Leaving the irq enabled at the chip and desc level between S0ix states might give some performance benefit. That was my original thought. Will it cause problems?

Quoted 5 lines
Quoted 3 lines

Errm. I merily moved it to the place where it should be. You do the disable/enable dance already today.

I think I only do disable/enable at the timer HW level today during cpu hp notification, not calling disable_irq(). Am i missing something?

apbt_setup_irq()
{
...

--->	disable_irq(adev->irq);
	desc->status |= IRQ_MOVE_PCNTXT;
	irq_set_affinity(adev->irq, cpumask_of(adev->cpu));
	/* APB timer irqs are set up as mp_irqs, timer is edge triggerred */
	set_irq_chip_and_handler_name(adev->irq, chip, handle_edge_irq, "edge");
--->	enable_irq(adev->irq);
	if (system_state == SYSTEM_BOOTING)

...
}

So that's a disable/enable pair on every cpu hotplug, right ?

Now I moved the disable to CPU_DEAD where it belongs and the enable stayed at the same place.

Thanks,

tglx

author
#5 lore raw

Thomas Gleixner Wed, 13 Oct 2010 00:26:08 +0200 (CEST)

Quoted 35 lines · Jacob Pan
Quoted 17 lines

Thomas Gleixner Tue, 12 Oct 2010 23:28:19 +0200 (CEST)

Quoted 15 lines · Jacob Pan
Quoted 13 lines
Quoted 5 lines

x86: Sanitize apb timer interrupt handling

Disable the interrupt in CPU_DEAD where it belongs.

My main concern is the performance cost. The power management code for Moorestown system make use of the cpu hotplug code (disable_nonboot_cpus) but much more frequently. The system low power states are call S0 idle state (s0ix).

Leaving the irq enabled at the chip and desc level between S0ix states might give some performance benefit. That was my original thought. Will it cause problems?

Quoted 5 lines
Quoted 3 lines

Errm. I merily moved it to the place where it should be. You do the disable/enable dance already today.

I think I only do disable/enable at the timer HW level today during cpu hp notification, not calling disable_irq(). Am i missing something?

apbt_setup_irq()
{
...

--->	disable_irq(adev->irq);
	desc->status |= IRQ_MOVE_PCNTXT;
	irq_set_affinity(adev->irq, cpumask_of(adev->cpu));
	/* APB timer irqs are set up as mp_irqs, timer is edge triggerred */
	set_irq_chip_and_handler_name(adev->irq, chip, handle_edge_irq, "edge");
--->	enable_irq(adev->irq);
	if (system_state == SYSTEM_BOOTING)
Quoted 7 lines

...
}

So that's a disable/enable pair on every cpu hotplug, right ?

Now I moved the disable to CPU_DEAD where it belongs and the enable stayed at the same place.

apbt_setup_irq() does not get called in cpu hotplug code, but CPU_DEAD notification will happen every time disable_nonboot_cpus() are called. Since Moorestown OSPM code uses disable_nonboot_cpus often, that is why i am concerned about the overhead of disable_irq().

author
#6 lore raw
In reply to Thomas Gleixner

Hi tglx,

You are right, i am already doing the enable/disable dance.

Thanks for pointing it out, patiently.

Jacob

Thomas Gleixner Wed, 13 Oct 2010 00:26:08 +0200 (CEST)

Quoted 38 lines · Jacob Pan
Quoted 17 lines

Thomas Gleixner Tue, 12 Oct 2010 23:28:19 +0200 (CEST)

Quoted 15 lines · Jacob Pan
Quoted 13 lines
Quoted 5 lines

x86: Sanitize apb timer interrupt handling

Disable the interrupt in CPU_DEAD where it belongs.

My main concern is the performance cost. The power management code for Moorestown system make use of the cpu hotplug code (disable_nonboot_cpus) but much more frequently. The system low power states are call S0 idle state (s0ix).

Leaving the irq enabled at the chip and desc level between S0ix states might give some performance benefit. That was my original thought. Will it cause problems?

Quoted 5 lines
Quoted 3 lines

Errm. I merily moved it to the place where it should be. You do the disable/enable dance already today.

I think I only do disable/enable at the timer HW level today during cpu hp notification, not calling disable_irq(). Am i missing something?

apbt_setup_irq()
{
...

--->	disable_irq(adev->irq);
	desc->status |= IRQ_MOVE_PCNTXT;
	irq_set_affinity(adev->irq, cpumask_of(adev->cpu));
	/* APB timer irqs are set up as mp_irqs, timer is edge triggerred */
	set_irq_chip_and_handler_name(adev->irq, chip, handle_edge_irq, "edge");
--->	enable_irq(adev->irq);
	if (system_state == SYSTEM_BOOTING)

...
}

Quoted 8 lines

So that's a disable/enable pair on every cpu hotplug, right ?

Now I moved the disable to CPU_DEAD where it belongs and the enable stayed at the same place.

Thanks,

tglx

yo