On 03/25/2014 12:54 PM, Wojciech Domski wrote:
> W dniu 25.03.2014 12:36, Gilles Chanteperdrix pisze:
>> On 03/25/2014 12:28 PM, Wojciech Domski wrote:
>>> W dniu 25.03.2014 12:16, Gilles Chanteperdrix pisze:
>>>> On 03/25/2014 11:26 AM, Wojciech Domski wrote:
>>>>> W dniu 24.03.2014 13:19, Gilles Chanteperdrix pisze:
>>>>>> The solution which would be acceptable is not to have busy waits, except
>>>>>> for very short durations. But for instance transferring a byte on I2C
>>>>>> takes around 20us at 400 kHz, a 20us masking section is unacceptable.
>>>>>> rtdm_task_busy_sleep, as the name suggests is a busy wait loop, so, no,
>>>>>> it is not acceptable either.
>>>>>>
>>>>>> Use a thread or a timer to reschedule while you wait for the end of the
>>>>>> transfer, instead of busy waiting.
>>>>> Dear Gilles,
>>>>>
>>>>> As you mentioned before the driver has few places like this:
>>>>>
>>>>> while(!FLAG1);
>>>>> while(!FLAG2);
>>>>>
>>>>> Would a solution of creating a separate task for this purpose be ok?
>>>>>
>>>>> task()
>>>>> {
>>>>> while(!FLAG1);
>>>>> while(!FLAG2);
>>>>> }
>>>> I was rather thinking about rtdm_task_sleep() instead.
>>>>
>>>>> In the place of those loops a piece of code responsible for creating and
>>>>> joining task would be put instead:
>>>>>
>>>>> rtdm_task_init();
>>>>> rtdm_task_join_nrt();
>>>> This will not work for code currently running in interrupt handlers. An
>>>> interrupt handler can not call rtdm_task_join_nrt(). You will rather
>>>> have to wake up the thread, not reenabling the interrupt at the end of
>>>> the handler, and reenable the interrupt in the thread when the job is
>>>> done. Or alternatively, fire a timer instead of waking up a thread.
>>>>
>>>>
>>> Dear Gilles,
>>>
>>> If what I have understood correctly, basically what you mean is changing:
>>>
>>> while(!FLAG) ;
>>>
>>> into
>>>
>>> while(!FLAG)
>>> rtdm_task_sleep();
>>>
>>> Xenomai's documentation that rtdm_task_sleep() can be always
>>> rescheduled. In my opinion
>>> this solution is sufficient and meets Xenomai requirements.
>> This solution will not work if the busy loop was in an irq handler. You
>> can not call rtdm_task_sleep from the context of an irq handler.
>>
>>
> Ok, now I get your point.
>
> However, what do you mean exactly by rescheduling while waiting? Do you
> mean something like
> putting all irq services into another thread and only in irq handler
> notify the separate thread
> to do the job accordingly?
Well, I do not know exactly, I have not looked at the details of the
driver, you are supposed to do that. But you may be able to split the
interrupt code in two parts: a first part that will run in interrupt
mode, and a second part that will run as a thread. The irq handler would
wake the thread up at the end of the first part, and return without
reenabling the interrupt at interrupt controller level. The thread would
execute, sleep during the I2C transfers as needed, and reenable the
interrupt before going back to sleep.
--
Gilles.
_______________________________________________
Xenomai mailing list
[email protected]
http://www.xenomai.org/mailman/listinfo/xenomai
