[LITMUS^RT] Problem with Litmus-RT plugin

[Evgeniy Veretennikov]

On 27.05.2014 12:23, Andrea Bastoni wrote:
> Hi Evgeniy,
>
> On Tue, May 27, 2014 at 9:33 AM, Evgeniy Veretennikov
> <evg.veretennikov at gmail.com> wrote:
>> Hello Andrea,
>>
>> I'm student of the Moscow Aviation Institute, and I develop Litmus-RT plugin
>> which schedules processes like ARINC 653 scheduler schedules partitions
>> (ARINC 653 is aviation standard). This is simple round-robin algorithm: each
>> process has predefined runtime inside predefined period. My plugin is based
>> on version 2012.2 as a demo plugin in this CreateAPluginTutorial:
>> https://wiki.litmus-rt.org/litmus/CreateAPluginTutorial
> Can I ask you why 2012.2 and not a more updated version?
Of course. Only reason is that demo plugin is for 2012.2, so I thought 
it would be better to have simple working example.
>
>> My plugin's function arinc_schedule(struct task_struct*) just chooses
>> task_struct* of process which should run now and returns this pointer. So, I
>> can successfully start my plugin ($ setsched ARINC), but when I start some
>> RT-process (for example, $ rtspin 10 100 10), Ubuntu GUI freezes and virtual
>> machine with Litmus uses 100% of all CPUs. I think, I should change process
>> state or some queues inside arinc_schedule() function, but I don't know what
>> exactly to do. I saw some queue-management in demo plugin, but I don't know
>> how to use process queues, because my algorithm is just round-robin, either
>> array, not a queue.
> The Litmus RT infrastructure allows you to define the logic to select
> the next task to be run, but is integrated in the linux-scheduler
> infrastructure.
>
> Although it should be relatively straightforward to define your own
> scheduling policy (either ARINC o P-EDF, or G-EDF), you should make
> sure you also understand the linux-scheduler infrastructure where
> Litmus is "plugged-in".
>
> To implement an A653 policy, I'd suggest you to start with a P-FP
> plugin, understand its implementation, and then hack-in the
> modification required to support your specific need. The data
> structure you use to manage your tasks (provided that you adequately
> protect them) are your specific plugin implementation details.
Ok, I had already tried to understood P-FP. Unfortunately, some details 
aren't clear for me. That's my skeleton of current schedule function's 
implementation:

static struct task_struct *arinc_schedule(struct task_struct *prev)
{
     struct task_struct *next = NULL;
     spin_lock_irqsave(...);
     /* here is schedule decision: next = some next task; do nothing 
with cpu state, process state, etc. */
     sched_state_task_picked();
     spin_unlock_irqrestore(...);
     return next;
}

But I can't start RT-processes with current code (see my previous 
letter). I think, I have to change process state, I tried this after 
schedule decision:

set_rt_flags(prev, RT_F_SLEEP);
set_rt_flags(next, RT_F_RUNNING);

But problem wasn't solved. Maybe I need to manipulate with rt_param 
structure? All what I need is to sleep prev task and to run next task.

Thanks for the quick reply.

Best regards,
Evgeniy Veretennikov
Student of the Moscow Aviation Institute