2009/2/25 Michael Blizek <michi1@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx>:
> On 17:44 Wed 25 Feb , Denis Borisevich wrote:
>> Hi!
>> I've done some tests and here is what I figured out: the interrupts
>> assigned to the device controlled by the ahci driver keep going even
>> without my driver loaded and even with all my boards removed from
>> system physically. So it seems that everything ok.
>> But I have another question concerning interrupts:
>> Let's take the situation I described in my first and second posts - I
>> have a system with the PCI board which I develop driver for. This
>> board shares interrupt with the device controlled by ahci driver. The
>> question is why every interrupt I got on this line is interpreted by
>> my driver as the interrupt which belongs to it?
>
> If an interrupt is shared, the kernel does not know which device a raised
> interrupt belongs to. For this reason, it is delivered to every driver which
> has an interrupt handler for this interrupt.
>
>> Here is the code I use:
>>
>> === CODE BELOW ===
>>
>> to register ISR:
>> struct my_dev *board;
>> request_irq(board->irq, my_interrupt_handler, IRQF_SHARED, "my_module", board);
>>
>> ISR:
>> static irqreturn_t my_interrupt_handler(int irq, void *dev_id)
>> {
>> struct my_dev *board;
>> int i, retval = IRQ_NONE;
>>
>> // find out which board got an interrupt
>> for (i = 0; i < MAX_BOARDS; i++)
>> if (dev_id == &my_boards[i]) {
>> board = dev_id;
>> break;
>> }
>>
>> // Check if interrupt was issued by another device (in case of shared
>> // interrupt line).
>> if (i == MAX_BOARDS)
>> goto handled;
>> if (board == NULL)
>> goto handled;
>>
>> // if we are here then it's our interrupt
>> retval = IRQ_HANDLED;
>>
>> /* here goes some checks on the device registers
>> to know if there is interrupts pending in it.
>> these checks are sure to be right. If there's no
>> interrupt, then debug message "no interrupt" is
>> printed.
>> */
>> .......
>> /* here goes the interrupt handling, if the previous check
>> was successful
>> */
>> .......
>>
>> handled:
>> return retval;
>> }
>>
>> === CODE ABOVE ===
>>
>> So, what do I have now. When no program touches my device (it is
>> idling) and the ahci driver is loaded I have this "no interrupt"
>> message in dmesg output 20-30 times per second. When some application
>> uses my device I have the proper printouts (the kind of interrupt and
>> so on) and bunch of these "no interrupt". When I install my board into
>> another PCI slot (and get another interrupt line) I never see these
>> "no interrupt" messages, but only the ones concerning to the actual
>> interrupt handling in my driver. Which indicates that driver works as
>> intended. The questions is why the dev_id parameter of
>> my_interrupt_handler() holds the data which belongs to one of the
>> boards controlled by my driver (which is the reason why the check in
>> the very beginning of ISR routine is passed)?
>
> The dev_id is the what you passed it request_irq. You do not need the loop,
> because you will not be called if dev_id is not your device. The 2
> if...goto handled should never happen. What you need to do in the interrupt
> handler is to ask the device if it has raised the interrupt and do the
> interrupt handling if necessary. Something like this:
>
> static irqreturn_t my_interrupt_handler(int irq, void *dev_id)
> {
> struct my_dev *board = dev_id;
>
> int interrupts_pending = do_some_check;
>
> if (interrupts_pending == 0) {
> /* probably another device which is sharing the interrupt has
> * raised thi irq
> */
> return IRQ_NONE;
> }
>
> /* here goes the interrupt handling */
>
> return IRQ_HANDLED;
> }
>
> -Michi
>
> --
> programing a layer 3+4 network protocol for mesh networks
> see http://michaelblizek.twilightparadox.com
>
>
I see. Thanks for your help. Have you any idea why the ahci driver
behaves this way?
--
Denis
--
To unsubscribe from this list: send an email with
"unsubscribe kernelnewbies" to ecartis@xxxxxxxxxxxx
Please read the FAQ at http://kernelnewbies.org/FAQ
