> On Wed, 11 Nov 2015 12:13:59 +0100, Maarten Brock wrote: > > > We want the edge. > > > > I don't think you do. It's exactly this that breaks shared interrupts. > > Consider the following scenario. There are two SC16IS7xx's sharing their > > interrupt. Assume A to be handled first and B second. A character arrives > > at B which makes the /IRQ_B go low. The handler for A is run and it finds > > no flags in IIR so it ends. Then the handler for B is run and the > > character is retrieved. But in the meantime A also receives a character > > and makes /IRQ_A low. The /IRQ input of the interrupt controller never > > sees a second falling edge. When the handler for B is done, /IRQ_B will > > go inactive but the line stays low since /IRQ_A is still active. There is > > however nothing retriggering the handler of A. > > Right, I see your point now. The danger of using level-triggered IRQs > in general is obviously that if the line goes low permanently (HW > failure, chip multfunction) the entire system will die in IRQ storm. If you want to prevent system failure due to hardware failure as you describe here, the software should detect the storm and disable the interrupt. I don't have a clue if linux or your average interrupt controller is up to that task. If you really want maybe you could detect the average interrupt rate in the handler and set a threshold to it based on the baudrate. > > So how could a level interrupt help here? When /IRQ_B goes low first > > handler A and then handler B are executed as before. Due to the ONESHOT > > setting the interrupt is disabled in the interrupt controller until all > > threaded handlers have reenabled the interrupt. If at this point A still > > has an unhandled interrupt request the line is still low and the > > interrupt triggered. This leads to both handlers to be called again and > > the character in A retrieved. > > Yup. > > Small note - as I already said - ONESHOT seems to be only used by IRQ > core if we use devm_request_threaded_irq(), we should remove this flag > as it only confuses people. I think you have two choices here: 1. Remove ONESHOT and document that the driver does NOT support shared interrupts. 2. Keep ONESHOT, change to LEVEL and use devm_request_threaded_irq() properly. Maarten -- To unsubscribe from this list: send the line "unsubscribe linux-serial" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
