wake_up & wait_event are counting events ?

[michi1@michaelblizek.twilightparadox.com (michi1 at michaelblizek.twilightparadox.com)]

Hi!

On 15:14 Thu 19 Jan     , Ran Shalit wrote:
> Hello,
> 
> I am trying to correctly register interrupt in kernel for user interface:
> 
> irq handler
> =========
> 
> static irqreturn_t irq_handler(int irq, void *dev_id)
> {
>    struct elbit_irq_dev *elbit_irq = &elbit_irq_devices[0];
>     printk("irq in\n");
> 
>  <<==== clear interrupt in fpga  here or after the call to wake_up,
> (at the bottom of routine below ) ?
> 
>     atomic_set(&elbit_irq->flag, 1);
>     wake_up_interruptible(&elbit_irq->pollw);
> 
> 
> 
>     return IRQ_HANDLED;
> }
> 
> ioctl for waiting on interrupts
> ======================
> 
> static long elbit_device_ioctl( struct file *filp, /* ditto */
>          unsigned int ioctl_num, /* number and param for ioctl */
>          unsigned long ioctl_param)
> {
>     struct elbit_irq_dev *elbit_irq = &elbit_irq_devices[0];
> 
>     switch (ioctl_num) {
>     case IOCTL_WAIT_IRQ:
>         atomic_set(&elbit_irq->flag, 0);
>         wait_event_interruptible(elbit_irq->pollw,
> atomic_read(&elbit_irq->flag) != 0) ;
>         break;
> 
> 
>     default:
>         break;
>     }
>     return 0;
> }
> 
> my questions:
> 1. does wake_up and wait_event are countable, i.e. if there are 2
> interrupts for example before wait_event is called, does it mean it
> wake_event will not sleep on 2 consecutive calls ?

No, wait_event will sleep if the test "atomic_read(&elbit_irq->flag) != 0"
says so. In your code there is a race condition:

1) interrupt handler sets flag to 1
2) ioctl sets it to 0
==> wait_event_interruptible sleeps even tough it probably should not

You may want to replace the atomic_set() with atomic_inc() and atomic_dec() if
you want to wake up once every time the interrupt handler gets executed.

> 2. should we put the clear of interrupts in fpga in the interrupt
> before or after the wake_up ?

I do not think that this makes any difference.

> can we put them instead in the userspace handler (IOCTL_WAIT_IRQ) instead
> of clearing the interrupt in the interrupt handler ?

I do not think that this will work.

	-Michi
-- 
programing a layer 3+4 network protocol for mesh networks
see http://michaelblizek.twilightparadox.com