On 2017-06-01 00:33, Stephen Boyd wrote:
> On 05/30, Kiran Gunda wrote:
>> From: Abhijeet Dharmapurikar <[email protected]>
>>
>> The current code uses handle_level_irq flow handler even if the
>> trigger type of the interrupt is edge. This can lead to missing
>> of an edge transition that happens when the interrupt is being
>> handled. The level flow handler masks the interrupt while it is
>> being handled, so if an edge transition happens at that time,
>> that edge is lost.
>>
>> Use an edge flow handler for edge type interrupts which ensures
>> that the interrupt stays enabled while being handled - at least
>> until it triggers at which point the flow handler sets the
>> IRQF_PENDING flag and only then masks the interrupt. That
>> IRQF_PENDING state indicates an edge transition happened while
>> the interrupt was being handled and the handler is called again.
>>
>> Signed-off-by: Abhijeet Dharmapurikar <[email protected]>
>> Signed-off-by: Kiran Gunda <[email protected]>
>
> Do we have any edge interrupts in the tree right now? At least
> RTC seems to be using edge... This should go back to stable with
> a Fixes tag.
>
Yes. A bunch of internal drivers use edge interrupts.
>> ---
>> drivers/spmi/spmi-pmic-arb.c | 6 ++++++
>> 1 file changed, 6 insertions(+)
>>
>> diff --git a/drivers/spmi/spmi-pmic-arb.c
>> b/drivers/spmi/spmi-pmic-arb.c
>> index 1d23df0..ad34491 100644
>> --- a/drivers/spmi/spmi-pmic-arb.c
>> +++ b/drivers/spmi/spmi-pmic-arb.c
>> @@ -625,6 +625,12 @@ static int qpnpint_irq_set_type(struct irq_data
>> *d, unsigned int flow_type)
>> }
>>
>> qpnpint_spmi_write(d, QPNPINT_REG_SET_TYPE, &type, sizeof(type));
>> +
>> + if (flow_type & IRQ_TYPE_EDGE_BOTH)
>
> IRQ_TYPE_EDGE_BOTH doesn't seem appropriate to use here. We're
> really just testing to see if the type is an edge type, not if
> it's BOTH edges.
>
Ok. Agree. Will fix in the follow up patch.
>> + irq_set_handler_locked(d, handle_edge_irq);
>> + else
>> + irq_set_handler_locked(d, handle_level_irq);
>> +
>
> And we already have code that does that check:
>
> if (flow_type & (IRQF_TRIGGER_RISING | IRQF_TRIGGER_FALLING))
>
> so just put the irq_set_handler_locked() calls in those if
> statements please.
Sure. Will fix in the follow up patch.