From: Balasubramanian Manoharan <[email protected]>

Signed-off-by: Balasubramanian Manoharan <[email protected]>
---
/** Email created from pull request 160 (bala-manoharan:api_sched_order_lock)
 ** https://github.com/Linaro/odp/pull/160
 ** Patch: https://github.com/Linaro/odp/pull/160.patch
 ** Base sha: 4eae04e80a634c17ac276bb06bce468cbe28cde0
 ** Merge commit sha: 0fdc4843f1c46de41bffe74270ffe252bdce1521
 **/
 platform/linux-generic/include/odp_schedule_if.h   |  1 +
 .../include/odp_schedule_scalable_ordered.h        |  1 +
 platform/linux-generic/odp_schedule.c              | 25 +++++++++++++++++++++-
 platform/linux-generic/odp_schedule_if.c           |  5 +++++
 platform/linux-generic/odp_schedule_iquery.c       | 24 ++++++++++++++++++++-
 platform/linux-generic/odp_schedule_scalable.c     | 19 ++++++++++++++++
 platform/linux-generic/odp_schedule_sp.c           |  8 ++++++-
 7 files changed, 80 insertions(+), 3 deletions(-)

diff --git a/platform/linux-generic/include/odp_schedule_if.h 
b/platform/linux-generic/include/odp_schedule_if.h
index 657993b1..b0db67ab 100644
--- a/platform/linux-generic/include/odp_schedule_if.h
+++ b/platform/linux-generic/include/odp_schedule_if.h
@@ -95,6 +95,7 @@ typedef struct {
                                   odp_schedule_group_info_t *);
        void (*schedule_order_lock)(unsigned);
        void (*schedule_order_unlock)(unsigned);
+       void (*schedule_order_unlock_lock)(unsigned);
 
 } schedule_api_t;
 
diff --git a/platform/linux-generic/include/odp_schedule_scalable_ordered.h 
b/platform/linux-generic/include/odp_schedule_scalable_ordered.h
index 1c365a2b..cdc2ab49 100644
--- a/platform/linux-generic/include/odp_schedule_scalable_ordered.h
+++ b/platform/linux-generic/include/odp_schedule_scalable_ordered.h
@@ -79,6 +79,7 @@ typedef struct reorder_window {
        uint32_t tail;
        uint32_t turn;
        uint32_t olock[CONFIG_QUEUE_MAX_ORD_LOCKS];
+       uint32_t lock_acquired;
        uint16_t lock_count;
        /* Reorder contexts in this window */
        reorder_context_t *ring[RWIN_SIZE];
diff --git a/platform/linux-generic/odp_schedule.c 
b/platform/linux-generic/odp_schedule.c
index 5b940762..35f69bd7 100644
--- a/platform/linux-generic/odp_schedule.c
+++ b/platform/linux-generic/odp_schedule.c
@@ -214,6 +214,8 @@ typedef struct {
        /* Array of ordered locks */
        odp_atomic_u64_t  lock[CONFIG_QUEUE_MAX_ORD_LOCKS];
 
+       odp_atomic_u64_t lock_acquired;
+
 } order_context_t ODP_ALIGNED_CACHE;
 
 typedef struct {
@@ -1103,6 +1105,7 @@ static void order_unlock(void)
 static void schedule_order_lock(unsigned lock_index)
 {
        odp_atomic_u64_t *ord_lock;
+       odp_atomic_u64_t *lock_acquired;
        uint32_t queue_index;
 
        queue_index = sched_local.ordered.src_queue;
@@ -1112,6 +1115,7 @@ static void schedule_order_lock(unsigned lock_index)
                   !sched_local.ordered.lock_called.u8[lock_index]);
 
        ord_lock = &sched->order[queue_index].lock[lock_index];
+       lock_acquired = &sched->order[queue_index].lock_acquired;
 
        /* Busy loop to synchronize ordered processing */
        while (1) {
@@ -1121,6 +1125,7 @@ static void schedule_order_lock(unsigned lock_index)
 
                if (lock_seq == sched_local.ordered.ctx) {
                        sched_local.ordered.lock_called.u8[lock_index] = 1;
+                       odp_atomic_store_rel_u64(lock_acquired, lock_index);
                        return;
                }
                odp_cpu_pause();
@@ -1130,6 +1135,7 @@ static void schedule_order_lock(unsigned lock_index)
 static void schedule_order_unlock(unsigned lock_index)
 {
        odp_atomic_u64_t *ord_lock;
+       odp_atomic_u64_t *lock_acquired;
        uint32_t queue_index;
 
        queue_index = sched_local.ordered.src_queue;
@@ -1138,12 +1144,28 @@ static void schedule_order_unlock(unsigned lock_index)
                   lock_index <= sched->queue[queue_index].order_lock_count);
 
        ord_lock = &sched->order[queue_index].lock[lock_index];
+       lock_acquired = &sched->order[queue_index].lock_acquired;
 
        ODP_ASSERT(sched_local.ordered.ctx == odp_atomic_load_u64(ord_lock));
 
+       odp_atomic_store_rel_u64(lock_acquired,
+                                sched->queue[queue_index].
+                                order_lock_count + 1);
        odp_atomic_store_rel_u64(ord_lock, sched_local.ordered.ctx + 1);
 }
 
+static void schedule_order_unlock_lock(unsigned lock_index)
+{
+       odp_atomic_u64_t *lock_acquired;
+       uint32_t queue_index;
+
+       queue_index = sched_local.ordered.src_queue;
+
+       lock_acquired = &sched->order[queue_index].lock_acquired;
+       schedule_order_unlock(odp_atomic_load_u64(lock_acquired));
+       schedule_order_lock(lock_index);
+}
+
 static void schedule_pause(void)
 {
        sched_local.pause = 1;
@@ -1429,5 +1451,6 @@ const schedule_api_t schedule_default_api = {
        .schedule_group_thrmask   = schedule_group_thrmask,
        .schedule_group_info      = schedule_group_info,
        .schedule_order_lock      = schedule_order_lock,
-       .schedule_order_unlock    = schedule_order_unlock
+       .schedule_order_unlock    = schedule_order_unlock,
+       .schedule_order_unlock_lock    = schedule_order_unlock_lock
 };
diff --git a/platform/linux-generic/odp_schedule_if.c 
b/platform/linux-generic/odp_schedule_if.c
index e56e3722..858c1949 100644
--- a/platform/linux-generic/odp_schedule_if.c
+++ b/platform/linux-generic/odp_schedule_if.c
@@ -129,3 +129,8 @@ void odp_schedule_order_unlock(unsigned lock_index)
 {
        return sched_api->schedule_order_unlock(lock_index);
 }
+
+void odp_schedule_order_unlock_lock(uint32_t lock_index)
+{
+       sched_api->schedule_order_unlock_lock(lock_index);
+}
diff --git a/platform/linux-generic/odp_schedule_iquery.c 
b/platform/linux-generic/odp_schedule_iquery.c
index b81e5dab..d8154358 100644
--- a/platform/linux-generic/odp_schedule_iquery.c
+++ b/platform/linux-generic/odp_schedule_iquery.c
@@ -135,6 +135,8 @@ typedef struct {
        /* Array of ordered locks */
        odp_atomic_u64_t  lock[CONFIG_QUEUE_MAX_ORD_LOCKS];
 
+       odp_atomic_u64_t lock_acquired;
+
 } order_context_t ODP_ALIGNED_CACHE;
 
 typedef struct {
@@ -1255,6 +1257,7 @@ static void order_unlock(void)
 static void schedule_order_lock(unsigned lock_index)
 {
        odp_atomic_u64_t *ord_lock;
+       odp_atomic_u64_t *lock_acquired;
        uint32_t queue_index;
 
        queue_index = thread_local.ordered.src_queue;
@@ -1264,6 +1267,7 @@ static void schedule_order_lock(unsigned lock_index)
                   !thread_local.ordered.lock_called.u8[lock_index]);
 
        ord_lock = &sched->order[queue_index].lock[lock_index];
+       lock_acquired = &sched->order[queue_index].lock_acquired;
 
        /* Busy loop to synchronize ordered processing */
        while (1) {
@@ -1273,6 +1277,7 @@ static void schedule_order_lock(unsigned lock_index)
 
                if (lock_seq == thread_local.ordered.ctx) {
                        thread_local.ordered.lock_called.u8[lock_index] = 1;
+                       odp_atomic_store_rel_u64(lock_acquired, lock_index);
                        return;
                }
                odp_cpu_pause();
@@ -1282,6 +1287,7 @@ static void schedule_order_lock(unsigned lock_index)
 static void schedule_order_unlock(unsigned lock_index)
 {
        odp_atomic_u64_t *ord_lock;
+       odp_atomic_u64_t *lock_acquired;
        uint32_t queue_index;
 
        queue_index = thread_local.ordered.src_queue;
@@ -1290,12 +1296,27 @@ static void schedule_order_unlock(unsigned lock_index)
                   lock_index <= sched->queues[queue_index].lock_count);
 
        ord_lock = &sched->order[queue_index].lock[lock_index];
+       lock_acquired = &sched->order[queue_index].lock_acquired;
 
        ODP_ASSERT(thread_local.ordered.ctx == odp_atomic_load_u64(ord_lock));
 
+       odp_atomic_store_rel_u64(lock_acquired,
+                                sched->queues[queue_index].lock_count + 1);
        odp_atomic_store_rel_u64(ord_lock, thread_local.ordered.ctx + 1);
 }
 
+static void schedule_order_unlock_lock(unsigned lock_index)
+{
+       uint32_t queue_index;
+       odp_atomic_u64_t *lock_acquired;
+
+       queue_index = thread_local.ordered.src_queue;
+
+       lock_acquired = &sched->order[queue_index].lock_acquired;
+       schedule_order_unlock(odp_atomic_load_u64(lock_acquired));
+       schedule_order_lock(lock_index);
+}
+
 static unsigned schedule_max_ordered_locks(void)
 {
        return CONFIG_QUEUE_MAX_ORD_LOCKS;
@@ -1368,7 +1389,8 @@ const schedule_api_t schedule_iquery_api = {
        .schedule_group_thrmask   = schedule_group_thrmask,
        .schedule_group_info      = schedule_group_info,
        .schedule_order_lock      = schedule_order_lock,
-       .schedule_order_unlock    = schedule_order_unlock
+       .schedule_order_unlock    = schedule_order_unlock,
+       .schedule_order_unlock_lock    = schedule_order_unlock_lock
 };
 
 static void thread_set_interest(sched_thread_local_t *thread,
diff --git a/platform/linux-generic/odp_schedule_scalable.c 
b/platform/linux-generic/odp_schedule_scalable.c
index 765326e8..0baedb06 100644
--- a/platform/linux-generic/odp_schedule_scalable.c
+++ b/platform/linux-generic/odp_schedule_scalable.c
@@ -1007,6 +1007,8 @@ static void schedule_order_lock(unsigned lock_index)
                       monitor32(&rctx->rwin->olock[lock_index],
                                 __ATOMIC_ACQUIRE) != rctx->sn)
                        doze();
+       atomic_store_release(&rctx->rwin->lock_acquired, lock_index, false);
+
        }
 }
 
@@ -1025,9 +1027,25 @@ static void schedule_order_unlock(unsigned lock_index)
        atomic_store_release(&rctx->rwin->olock[lock_index],
                             rctx->sn + 1,
                             /*readonly=*/false);
+       atomic_store_release(&rctx->rwin->lock_acquired,
+                            rctx->rwin->lock_count + 1, false);
        rctx->olock_flags |= 1U << lock_index;
 }
 
+static void schedule_order_unlock_lock(unsigned lock_index)
+{
+       struct reorder_context *rctx;
+
+       rctx = sched_ts->rctx;
+       if (odp_unlikely(rctx == NULL || rctx->rwin == NULL)) {
+               ODP_ERR("Invalid call to odp_schedule_order_unlock_lock\n");
+               return;
+       }
+       schedule_order_unlock(__atomic_load_n(&rctx->rwin->lock_acquired,
+                                             __ATOMIC_ACQUIRE));
+       schedule_order_lock(lock_index);
+}
+
 static void schedule_release_atomic(void)
 {
        sched_scalable_thread_state_t *ts;
@@ -1978,4 +1996,5 @@ const schedule_api_t schedule_scalable_api = {
        .schedule_group_info            = schedule_group_info,
        .schedule_order_lock            = schedule_order_lock,
        .schedule_order_unlock          = schedule_order_unlock,
+       .schedule_order_unlock_lock     = schedule_order_unlock_lock,
 };
diff --git a/platform/linux-generic/odp_schedule_sp.c 
b/platform/linux-generic/odp_schedule_sp.c
index 05241275..d4dfbcaf 100644
--- a/platform/linux-generic/odp_schedule_sp.c
+++ b/platform/linux-generic/odp_schedule_sp.c
@@ -819,6 +819,11 @@ static void schedule_order_unlock(unsigned lock_index)
        (void)lock_index;
 }
 
+static void schedule_order_unlock_lock(unsigned lock_index)
+{
+       (void)lock_index;
+}
+
 static void order_lock(void)
 {
 }
@@ -868,5 +873,6 @@ const schedule_api_t schedule_sp_api = {
        .schedule_group_thrmask   = schedule_group_thrmask,
        .schedule_group_info      = schedule_group_info,
        .schedule_order_lock      = schedule_order_lock,
-       .schedule_order_unlock    = schedule_order_unlock
+       .schedule_order_unlock    = schedule_order_unlock,
+       .schedule_order_unlock_lock     = schedule_order_unlock_lock
 };

Reply via email to