Skip to contentSkip to conversation

GIC: Assign correct flow handler type in set_type callback

PATCH In review Abhijeet Dharmapurikar 2 participants 4 messages Dec 30, 2010 – Jan 2, 2011 Arm (32-bit)

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 GIC: Assign correct flow handler type in set_type callback

There are some interrupts that are true edge triggered in nature. If not marked IRQ_PENDING, when disabled, they will be lost.

Use the set_type callback to assign the correct flow type handler for shared peripheral interrupts.

---
This came to light when a edge triggered interrupt was supposed to wakeup the sytem. The flow handler was set to the default handle_level_irq. On the resume path the flow handler was invoked right after the I bit was cleared but before each individual interrupts were enabled. This made the handle_level_irq ignore the interrupt (mask_ack it) and it was lost. handle_edge_irq does the right thing by marking the interrupt as IRQ_PENDING and when the resume code gets to enabling each interrupt this interrupt is resent again.

 arch/arm/common/gic.c |    3 +++
 1 files changed, 3 insertions(+), 0 deletions(-)
arch/arm/common/gic.c +3 -0
index e6388dc..a83594a 100644
@@ -136,6 +136,9 @@ static int gic_set_type(unsigned int irq, unsigned int type)
 
 	spin_unlock(&irq_controller_lock);
 
+	if ((type & IRQ_TYPE_EDGE_RISING) && gicirq > 31)
+		__set_irq_handler_unlocked(irq, handle_edge_irq);
+
 	return 0;
 }
 
Signature · 3 lines
1.7.1
Sent by an employee of Qualcomm Innovation Center, Inc. Qualcomm Innovation
Center, Inc. is a member of Code Aurora Forum.
#1 lore raw
Quoted 6 lines · Abhijeet Dharmapurikar

There are some interrupts that are true edge triggered in nature. If not marked IRQ_PENDING, when disabled, they will be lost.

Use the set_type callback to assign the correct flow type handler for shared peripheral interrupts.

Quoted 6 lines

---
This came to light when a edge triggered interrupt was supposed to wakeup the sytem. The flow handler was set to the default handle_level_irq. On the resume path the flow handler was invoked right after the I bit was cleared but before each individual interrupts were enabled. This made the handle_level_irq ignore

Why does the flow handler hit when the interrupt is disabled? Have you set IRQF_NOSUSPEND on this interrupt?

Quoted 3 lines

the interrupt (mask_ack it) and it was lost. handle_edge_irq does the right thing by marking the interrupt as IRQ_PENDING and when the resume code gets to enabling each interrupt this interrupt is resent again.

#2 lore raw
Quoted 15 lines · Rabin Vincent
Quoted 6 lines · Abhijeet Dharmapurikar

There are some interrupts that are true edge triggered in nature. If not marked IRQ_PENDING, when disabled, they will be lost.

Use the set_type callback to assign the correct flow type handler for shared peripheral interrupts.

Quoted 6 lines

---
This came to light when a edge triggered interrupt was supposed to wakeup the sytem. The flow handler was set to the default handle_level_irq. On the resume path the flow handler was invoked right after the I bit was cleared but before each individual interrupts were enabled. This made the handle_level_irq ignore

Quoted 3 lines

Why does the flow handler hit when the interrupt is disabled? Have you set IRQF_NOSUSPEND on this interrupt?

Since GIC doesnt have disable callback it implements lazy disabling. The interrupt is only marked IRQ_DISABLED in the descriptor but is not masked in the GIC. Hence the interrupt flow handler is hit.

Now that I re-read the code setting IRQF_NO_SUSPEND would fix the issue. But shouldnt set_irq_wake() do something similar?

Do I need to request IRQF_NO_SUSPEND for all the interrupts that could possibly wakeup the system - seems a bit unnecessary. IMO the interrupt should not be disabled if it is marked IRQF_NO_SUPEND || IRQ_WAKEUP is set.

Abhijeet

#3 lore raw
Quoted 24 lines · Abhijeet Dharmapurikar
Quoted 19 lines · Rabin Vincent
Quoted 7 lines · Abhijeet Dharmapurikar

There are some interrupts that are true edge triggered in nature. If not marked IRQ_PENDING, when disabled, they will be lost.

Use the set_type callback to assign the correct flow type handler for shared peripheral interrupts.

Quoted 9 lines

---
This came to light when a edge triggered interrupt was supposed to wakeup the
sytem. The flow handler was set to the default handle_level_irq. On the resume
path the flow handler was invoked right after the I bit was cleared but before
each individual interrupts were enabled. This made the handle_level_irq ignore

Quoted 4 lines

Why does the flow handler hit when the interrupt is disabled? Have you set
IRQF_NOSUSPEND on this interrupt?

Quoted 5 lines

Since GIC doesnt have disable callback it implements lazy disabling. The interrupt is only marked IRQ_DISABLED in the descriptor but is not masked in the GIC. Hence the interrupt flow handler is hit.

Now that I re-read the code setting IRQF_NO_SUSPEND would fix the issue.

Let me correct myself, IRQF_NO_SUSPEND isnt exactly what I want to do. If an interrupt triggers after suspend_device_irqs() is run and before the I bit is set in arch_suspend_disable_irqs() the system will handle the interrupt but will not abort suspend - check_wakeup_irqs() will return 0.

IMO, using handle_edge_irq() for an edge triggered wakeup interrupt would be the right way to address the issue.