[PATCH] smp: Fix scheduler helping protocol
Joel Sherrill
joel.sherrill at oarcorp.com
Fri Nov 21 16:09:53 UTC 2014
On 11/21/2014 10:07 AM, Sebastian Huber wrote:
> From: Luca Bonato <lohathe at gmail.com>
>
> New test case for smptests/smpmrsp01.
>
> It possible that a state change from SCHEDULER_SMP_NODE_READY to
> SCHEDULER_SMP_NODE_READY can happen.
Fix English.
How about a short description of the scenario this is addressing?
> ---
> cpukit/score/include/rtems/score/schedulerimpl.h | 28 ++--
> .../score/include/rtems/score/schedulersmpimpl.h | 25 ++-
> testsuites/smptests/smpmrsp01/init.c | 174 +++++++++++++++++++++
> testsuites/smptests/smpmrsp01/smpmrsp01.scn | 1 +
> 4 files changed, 210 insertions(+), 18 deletions(-)
>
> diff --git a/cpukit/score/include/rtems/score/schedulerimpl.h b/cpukit/score/include/rtems/score/schedulerimpl.h
> index 45a2f8d..b262b91 100644
> --- a/cpukit/score/include/rtems/score/schedulerimpl.h
> +++ b/cpukit/score/include/rtems/score/schedulerimpl.h
> @@ -1081,6 +1081,7 @@ RTEMS_INLINE_ROUTINE Thread_Control *_Scheduler_Release_idle_thread(
> */
> RTEMS_INLINE_ROUTINE bool _Scheduler_Block_node(
> Scheduler_Context *context,
> + Thread_Control *thread,
> Scheduler_Node *node,
> bool is_scheduled,
> Scheduler_Get_idle_thread get_idle_thread
> @@ -1088,25 +1089,24 @@ RTEMS_INLINE_ROUTINE bool _Scheduler_Block_node(
> {
> bool block;
> Thread_Control *old_user = _Scheduler_Node_get_user( node );
> - Thread_Control *new_user;
> + Thread_Control *new_user = NULL;
>
> _Scheduler_Thread_change_state( old_user, THREAD_SCHEDULER_BLOCKED );
>
> - if ( node->help_state == SCHEDULER_HELP_ACTIVE_RIVAL ) {
> - new_user = _Scheduler_Node_get_owner( node );
> -
> - _Assert( new_user != old_user );
> - _Scheduler_Node_set_user( node, new_user );
> - } else if (
> - node->help_state == SCHEDULER_HELP_ACTIVE_OWNER
> - && is_scheduled
> - ) {
> - new_user = _Scheduler_Use_idle_thread( context, node, get_idle_thread );
> - } else {
> - new_user = NULL;
> + if ( is_scheduled ) {
> + if ( node->help_state == SCHEDULER_HELP_ACTIVE_OWNER ) {
> + new_user = _Scheduler_Use_idle_thread( context, node, get_idle_thread );
> + } else if ( node->help_state == SCHEDULER_HELP_ACTIVE_RIVAL ) {
> + Thread_Control *owner = _Scheduler_Node_get_owner( node );
> +
> + if ( thread == old_user && owner != old_user ) {
> + new_user = owner;
> + _Scheduler_Node_set_user( node, new_user );
> + }
> + }
> }
>
> - if ( new_user != NULL && is_scheduled ) {
> + if ( new_user != NULL ) {
> Per_CPU_Control *cpu = _Thread_Get_CPU( old_user );
>
> _Scheduler_Thread_change_state( new_user, THREAD_SCHEDULER_SCHEDULED );
> diff --git a/cpukit/score/include/rtems/score/schedulersmpimpl.h b/cpukit/score/include/rtems/score/schedulersmpimpl.h
> index 156307d..0ddfce0 100644
> --- a/cpukit/score/include/rtems/score/schedulersmpimpl.h
> +++ b/cpukit/score/include/rtems/score/schedulersmpimpl.h
> @@ -793,13 +793,17 @@ static inline void _Scheduler_SMP_Block(
> {
> Scheduler_SMP_Node *node = _Scheduler_SMP_Thread_get_node( thread );
> bool is_scheduled = node->state == SCHEDULER_SMP_NODE_SCHEDULED;
> - bool block = _Scheduler_Block_node(
> + bool block;
> +
> + _Assert( is_scheduled || node->state == SCHEDULER_SMP_NODE_READY );
> +
> + block = _Scheduler_Block_node(
> context,
> + thread,
> &node->Base,
> is_scheduled,
> _Scheduler_SMP_Get_idle_thread
> );
> -
> if ( block ) {
> _Scheduler_SMP_Node_change_state( node, SCHEDULER_SMP_NODE_BLOCKED );
>
> @@ -838,9 +842,22 @@ static inline Thread_Control *_Scheduler_SMP_Unblock(
> Thread_Control *needs_help;
>
> if ( unblock ) {
> - _Scheduler_SMP_Node_change_state( node, SCHEDULER_SMP_NODE_READY );
> + if ( node->state != SCHEDULER_SMP_NODE_READY ) {
> + _Scheduler_SMP_Node_change_state( node, SCHEDULER_SMP_NODE_READY );
> +
> + needs_help = ( *enqueue_fifo )( context, &node->Base, thread );
> + } else {
> + _Assert( node->state == SCHEDULER_SMP_NODE_READY );
> + _Assert( node->Base.idle == NULL );
>
> - needs_help = ( *enqueue_fifo )( context, &node->Base, thread );
> + if ( node->Base.accepts_help == thread ) {
> + _Assert( node->Base.help_state == SCHEDULER_HELP_ACTIVE_OWNER );
> + needs_help = thread;
> + } else {
> + _Assert( node->Base.help_state == SCHEDULER_HELP_ACTIVE_RIVAL );
> + needs_help = NULL;
> + }
> + }
> } else {
> needs_help = NULL;
> }
> diff --git a/testsuites/smptests/smpmrsp01/init.c b/testsuites/smptests/smpmrsp01/init.c
> index bfa5d98..f6a98e2 100644
> --- a/testsuites/smptests/smpmrsp01/init.c
> +++ b/testsuites/smptests/smpmrsp01/init.c
> @@ -54,6 +54,7 @@ typedef struct {
> typedef struct {
> rtems_id main_task_id;
> rtems_id migration_task_id;
> + rtems_id high_task_id;
> rtems_id counting_sem_id;
> rtems_id mrsp_ids[MRSP_COUNT];
> rtems_id scheduler_ids[CPU_COUNT];
> @@ -66,6 +67,7 @@ typedef struct {
> SMP_lock_Control switch_lock;
> size_t switch_index;
> switch_event switch_events[32];
> + volatile bool run;
> } test_context;
>
> static test_context test_instance = {
> @@ -728,6 +730,177 @@ static void run_task(rtems_task_argument arg)
> }
> }
>
> +static void ready_unlock_worker(rtems_task_argument arg)
> +{
> + test_context *ctx = &test_instance;
> + rtems_status_code sc;
> + SMP_barrier_State barrier_state = SMP_BARRIER_STATE_INITIALIZER;
> +
> + assert_prio(RTEMS_SELF, 4);
> +
> + /* Obtain (F) */
> + barrier(ctx, &barrier_state);
> +
> + sc = rtems_semaphore_obtain(ctx->mrsp_ids[0], RTEMS_WAIT, RTEMS_NO_TIMEOUT);
> + rtems_test_assert(sc == RTEMS_SUCCESSFUL);
> +
> + sc = rtems_semaphore_release(ctx->mrsp_ids[0]);
> + rtems_test_assert(sc == RTEMS_SUCCESSFUL);
> +
> + assert_prio(RTEMS_SELF, 4);
> +
> + /* Done (G) */
> + barrier(ctx, &barrier_state);
> +
> + rtems_task_suspend(RTEMS_SELF);
> + rtems_test_assert(0);
> +}
> +
> +static void ready_unblock_timer(rtems_id timer_id, void *arg)
> +{
> + test_context *ctx = arg;
> + rtems_status_code sc;
> +
> + sc = rtems_task_start(
> + ctx->high_task_id,
> + run_task,
> + (rtems_task_argument) &ctx->run
> + );
> + rtems_test_assert(sc == RTEMS_SUCCESSFUL);
> +
> + sc = rtems_task_suspend(ctx->high_task_id);
> + rtems_test_assert(sc == RTEMS_SUCCESSFUL);
> +
> + sc = rtems_task_resume(ctx->high_task_id);
> + rtems_test_assert(sc == RTEMS_SUCCESSFUL);
> +
> + /*
> + * At this point the scheduler node of the main thread is in the
> + * SCHEDULER_SMP_NODE_READY state and a _Scheduler_SMP_Unblock() operation is
> + * performed.
> + */
> + sc = rtems_event_transient_send(ctx->main_task_id);
> + rtems_test_assert(sc == RTEMS_SUCCESSFUL);
> +
> + sc = rtems_task_suspend(ctx->high_task_id);
> + rtems_test_assert(sc == RTEMS_SUCCESSFUL);
> +}
> +
> +static void test_mrsp_ready_unblock(test_context *ctx)
> +{
> + rtems_status_code sc;
> + rtems_id timer_id;
> + SMP_barrier_State barrier_state = SMP_BARRIER_STATE_INITIALIZER;
> +
> + puts("test MrsP ready unblock");
> +
> + ctx->run = false;
> +
> + change_prio(RTEMS_SELF, 4);
> +
> + sc = rtems_semaphore_create(
> + rtems_build_name(' ', ' ', ' ', 'A'),
> + 1,
> + RTEMS_MULTIPROCESSOR_RESOURCE_SHARING
> + | RTEMS_BINARY_SEMAPHORE,
> + 3,
> + &ctx->mrsp_ids[0]
> + );
> + rtems_test_assert(sc == RTEMS_SUCCESSFUL);
> +
> + assert_prio(RTEMS_SELF, 4);
> +
> + sc = rtems_task_create(
> + rtems_build_name('H', 'I', 'G', 'H'),
> + 2,
> + RTEMS_MINIMUM_STACK_SIZE,
> + RTEMS_DEFAULT_MODES,
> + RTEMS_DEFAULT_ATTRIBUTES,
> + &ctx->high_task_id
> + );
> + rtems_test_assert(sc == RTEMS_SUCCESSFUL);
> +
> + sc = rtems_task_create(
> + rtems_build_name('W', 'O', 'R', 'K'),
> + 4,
> + RTEMS_MINIMUM_STACK_SIZE,
> + RTEMS_DEFAULT_MODES,
> + RTEMS_DEFAULT_ATTRIBUTES,
> + &ctx->worker_ids[0]
> + );
> + rtems_test_assert(sc == RTEMS_SUCCESSFUL);
> +
> + sc = rtems_task_set_scheduler(ctx->worker_ids[0], ctx->scheduler_ids[1]);
> + rtems_test_assert(sc == RTEMS_SUCCESSFUL);
> +
> + sc = rtems_timer_create(rtems_build_name('T', 'I', 'M', 'R'), &timer_id);
> + rtems_test_assert(sc == RTEMS_SUCCESSFUL);
> +
> + sc = rtems_semaphore_obtain(ctx->mrsp_ids[0], RTEMS_WAIT, RTEMS_NO_TIMEOUT);
> + rtems_test_assert(sc == RTEMS_SUCCESSFUL);
> +
> + assert_prio(RTEMS_SELF, 3);
> +
> + sc = rtems_timer_fire_after(timer_id, 2, ready_unblock_timer, ctx);
> + rtems_test_assert(sc == RTEMS_SUCCESSFUL);
> +
> + sc = rtems_event_transient_receive(RTEMS_WAIT, RTEMS_NO_TIMEOUT);
> + rtems_test_assert(sc == RTEMS_SUCCESSFUL);
> +
> + rtems_test_assert(!ctx->run);
> +
> + sc = rtems_task_start(ctx->worker_ids[0], ready_unlock_worker, 0);
> + rtems_test_assert(sc == RTEMS_SUCCESSFUL);
> +
> + /* Worker obtain (F) */
> + barrier(ctx, &barrier_state);
> +
> + sc = rtems_task_wake_after(2);
> + rtems_test_assert(sc == RTEMS_SUCCESSFUL);
> +
> + sc = rtems_task_suspend(ctx->worker_ids[0]);
> + rtems_test_assert(sc == RTEMS_SUCCESSFUL);
> +
> + sc = rtems_task_set_scheduler(ctx->high_task_id, ctx->scheduler_ids[1]);
> + rtems_test_assert(sc == RTEMS_SUCCESSFUL);
> +
> + sc = rtems_task_resume(ctx->high_task_id);
> + rtems_test_assert(sc == RTEMS_SUCCESSFUL);
> +
> + while (!ctx->run) {
> + /* Do noting */
> + }
> +
> + sc = rtems_task_resume(ctx->worker_ids[0]);
> + rtems_test_assert(sc == RTEMS_SUCCESSFUL);
> +
> + sc = rtems_task_suspend(ctx->high_task_id);
> + rtems_test_assert(sc == RTEMS_SUCCESSFUL);
> +
> + sc = rtems_semaphore_release(ctx->mrsp_ids[0]);
> + rtems_test_assert(sc == RTEMS_SUCCESSFUL);
> +
> + assert_prio(RTEMS_SELF, 4);
> +
> + /* Worker done (F) */
> + barrier(ctx, &barrier_state);
> +
> + sc = rtems_timer_delete(timer_id);
> + rtems_test_assert(sc == RTEMS_SUCCESSFUL);
> +
> + sc = rtems_task_delete(ctx->worker_ids[0]);
> + rtems_test_assert(sc == RTEMS_SUCCESSFUL);
> +
> + sc = rtems_task_delete(ctx->high_task_id);
> + rtems_test_assert(sc == RTEMS_SUCCESSFUL);
> +
> + sc = rtems_semaphore_delete(ctx->mrsp_ids[0]);
> + rtems_test_assert(sc == RTEMS_SUCCESSFUL);
> +
> + change_prio(RTEMS_SELF, 2);
> + rtems_test_assert(sc == RTEMS_SUCCESSFUL);
> +}
> +
> static void test_mrsp_obtain_and_sleep_and_release(test_context *ctx)
> {
> rtems_status_code sc;
> @@ -1232,6 +1405,7 @@ static void Init(rtems_task_argument arg)
> test_mrsp_unlock_order_error();
> test_mrsp_deadlock_error(ctx);
> test_mrsp_multiple_obtain();
> + test_mrsp_ready_unblock(ctx);
> test_mrsp_obtain_and_sleep_and_release(ctx);
> test_mrsp_obtain_and_release_with_help(ctx);
> test_mrsp_obtain_and_release(ctx);
> diff --git a/testsuites/smptests/smpmrsp01/smpmrsp01.scn b/testsuites/smptests/smpmrsp01/smpmrsp01.scn
> index 3315db1..2e1f32c 100644
> --- a/testsuites/smptests/smpmrsp01/smpmrsp01.scn
> +++ b/testsuites/smptests/smpmrsp01/smpmrsp01.scn
> @@ -5,6 +5,7 @@ test MrsP nested obtain error
> test MrsP unlock order error
> test MrsP deadlock error
> test MrsP multiple obtain
> +test MrsP ready unblock
> test MrsP obtain and sleep and release
> [0] MAIN -> RUN (prio 2, node RUN)
> [0] RUN -> MAIN (prio 1, node MAIN)
--
Joel Sherrill, Ph.D. Director of Research & Development
joel.sherrill at OARcorp.com On-Line Applications Research
Ask me about RTEMS: a free RTOS Huntsville AL 35805
Support Available (256) 722-9985
More information about the devel
mailing list