[PATCH v2 06/15] rtems: Fix rtems_scheduler_remove_processor()

[Gedare Bloom]

On Fri, Jul 7, 2017 at 9:54 AM, Sebastian Huber
<sebastian.huber at embedded-brains.de> wrote:
> Account for the thread processor affinity and make sure that it is
> possible to allocate a processor to each thread dedicated to a scheduler
> instance.
>
> Update #3059.
> ---
>  cpukit/rtems/src/schedulerremoveprocessor.c | 12 ++++++++----
>  1 file changed, 8 insertions(+), 4 deletions(-)
>
> diff --git a/cpukit/rtems/src/schedulerremoveprocessor.c b/cpukit/rtems/src/schedulerremoveprocessor.c
> index 538366bb22..06acacd158 100644
> --- a/cpukit/rtems/src/schedulerremoveprocessor.c
> +++ b/cpukit/rtems/src/schedulerremoveprocessor.c
> @@ -45,7 +45,13 @@ static bool _Scheduler_Check_processor_removal(
>    _Thread_Wait_acquire( the_thread, &queue_context );
>    _Thread_State_acquire_critical( the_thread, &state_context );
>
> -  if ( _Thread_Scheduler_get_home( the_thread ) == iter_context->scheduler ) {
> +  if (
> +    _Thread_Scheduler_get_home( the_thread ) == iter_context->scheduler
> +      && !_Processor_mask_Has_overlap(
> +        &the_thread->Scheduler.Affinity,
> +        _Scheduler_Get_processors( iter_context->scheduler )
> +      )

It may be worth it to add a new function like
_Thread_Is_affine(Thread_Control *t, Processor_mask *m) that checks if
t is contained in m. This would be easier to interpret than the above
inverted use of _Has_overlap().

> +  ) {
>      iter_context->status = RTEMS_RESOURCE_IN_USE;
>    }
>
> @@ -103,9 +109,7 @@ rtems_status_code rtems_scheduler_remove_processor(
>    _Scheduler_Release_critical( scheduler, &lock_context );
>    _ISR_lock_ISR_enable( &lock_context );
>
> -  if ( processor_count == 0 ) {
> -    _Thread_Iterate( _Scheduler_Check_processor_removal, &iter_context );
> -  }
> +  _Thread_Iterate( _Scheduler_Check_processor_removal, &iter_context );
>
>    _ISR_lock_ISR_disable( &lock_context );
>    _Scheduler_Acquire_critical( scheduler, &lock_context );
> --
> 2.12.3
>
> _______________________________________________
> devel mailing list
> devel at rtems.org
> http://lists.rtems.org/mailman/listinfo/devel