[PATCH 3/4] score: Add and use _Giant_Drop()

Gedare Bloom gedare at rtems.org
Wed Feb 19 15:14:30 UTC 2014


Does it make sense to put the Giant Lock in its own files?

On Wed, Feb 19, 2014 at 8:42 AM, Sebastian Huber
<sebastian.huber at embedded-brains.de> wrote:
> ---
>  cpukit/score/include/rtems/score/threaddispatch.h |   12 ++++++++++++
>  cpukit/score/src/smp.c                            |    8 ++++++++
>  cpukit/score/src/threaddispatchdisablelevel.c     |   13 +++++++++++++
>  3 files changed, 33 insertions(+), 0 deletions(-)
>
> diff --git a/cpukit/score/include/rtems/score/threaddispatch.h b/cpukit/score/include/rtems/score/threaddispatch.h
> index 2211662..4c2ecb0 100644
> --- a/cpukit/score/include/rtems/score/threaddispatch.h
> +++ b/cpukit/score/include/rtems/score/threaddispatch.h
> @@ -109,6 +109,18 @@ RTEMS_INLINE_ROUTINE void _Thread_Dispatch_initialization( void )
>    void _Giant_Release( void );
>
>    /**
> +   * @brief Releases the giant lock completely if held by the executing processor.
> +   *
> +   * The thread dispatch disable level is not altered by this function.
> +   *
> +   * The only use case for this operation is in
> +   * _SMP_Request_processors_to_shutdown().
> +   *
> +   * @param[in] self_cpu The current processor index.
> +   */
> +  void _Giant_Drop( uint32_t self_cpu );
> +
> +  /**
>     *  @brief Sets thread dispatch level to the value passed in.
>     *
>     * This routine sets thread dispatch level to the
> diff --git a/cpukit/score/src/smp.c b/cpukit/score/src/smp.c
> index 445a56a..41787e3 100644
> --- a/cpukit/score/src/smp.c
> +++ b/cpukit/score/src/smp.c
> @@ -115,6 +115,14 @@ void _SMP_Request_processors_to_shutdown( void )
>        _Per_CPU_Change_state( per_cpu, PER_CPU_STATE_SHUTDOWN );
>      }
>    }
> +
> +  /*
> +   * We have to drop the Giant lock here in order to give other processors the
> +   * opportunity to receive the inter-processor interrupts issued previously.
> +   * In case the executing thread still holds SMP locks, then other processors
> +   * already waiting for this SMP lock will spin forever.
> +   */
> +  _Giant_Drop( self );
>  }
>
>  void _SMP_Send_message( uint32_t cpu, uint32_t message )
> diff --git a/cpukit/score/src/threaddispatchdisablelevel.c b/cpukit/score/src/threaddispatchdisablelevel.c
> index a3dec99..ec5ac81 100644
> --- a/cpukit/score/src/threaddispatchdisablelevel.c
> +++ b/cpukit/score/src/threaddispatchdisablelevel.c
> @@ -57,6 +57,19 @@ static void _Giant_Do_release( void )
>    }
>  }
>
> +void _Giant_Drop( uint32_t self_cpu )
> +{
> +  Giant_Control *giant = &_Giant;
> +
> +  _Assert( _ISR_Get_level() != 0 );
> +
> +  if ( giant->owner_cpu == self_cpu ) {
> +    giant->nest_level = 0;
> +    giant->owner_cpu = NO_OWNER_CPU;
> +    _SMP_lock_Release( &giant->lock );
> +  }
> +}
> +
>  uint32_t _Thread_Dispatch_increment_disable_level( void )
>  {
>    ISR_Level isr_level;
> --
> 1.7.7
>
> _______________________________________________
> rtems-devel mailing list
> rtems-devel at rtems.org
> http://www.rtems.org/mailman/listinfo/rtems-devel



More information about the devel mailing list