Re: [patch 3/5] [PREEMPT_RT] Changing interrupt handlers from running in thread to hardirq and back runtime.

[Esben Nielsen <nielsen.esben@googlemail.com>]

On Sat, 3 Jun 2006, Steven Rostedt wrote:

> On Fri, 2006-06-02 at 23:23 +0100, Esben Nielsen wrote:
>> Makes it possible for the e100 ethernet driver to have it's interrupt handler
>> in both hard-irq and threaded context under PREEMPT_RT.
>>
>> Index: linux-2.6.16-rt23.spin_mutex/drivers/net/e100.c
>> ===================================================================
>> --- linux-2.6.16-rt23.spin_mutex.orig/drivers/net/e100.c
>> +++ linux-2.6.16-rt23.spin_mutex/drivers/net/e100.c
>> @@ -530,7 +530,7 @@ struct nic {
>>   	enum ru_state ru_running;
>>
>>   	spinlock_t cb_lock			____cacheline_aligned;
>> -	spinlock_t cmd_lock;
>> +	spin_mutex_t cmd_lock;
>>   	struct csr __iomem *csr;
>>   	enum scb_cmd_lo cuc_cmd;
>>   	unsigned int cbs_avail;
>> @@ -1950,6 +1950,30 @@ static int e100_rx_alloc_list(struct nic
>>   	return 0;
>>   }
>>
>> +static int e100_change_context(int irq, void *dev_id,
>> +			       enum change_context_cmd cmd)
>> +{
>> +	struct net_device *netdev = dev_id;
>> +	struct nic *nic = netdev_priv(netdev);
>> +
>> +	switch(cmd)
>> +	{
>> +	case IRQ_TO_HARDIRQ:
>> +		if(!spin_mutexes_can_spin())
>> +			return -ENOSYS;
>> +
>> +		spin_mutex_to_spin(&nic->cmd_lock);
>> +		break;
>> +	case IRQ_CAN_THREAD:
>> +		/* Ok - return 0 */
>> +		break;
>
> Why even bother with the IRQ_CAN_THREAD.  If this would be anything
> other than OK, then we shouldn't be using that request_irq2 (yuck!) call
> in the first place.
>
Just for the sake of generality. Nothing else. It would be very unlikely, 
as you say, that it wouldn't return 0.

> -- Steve

Esben