Resolve the corner case of releasing order for an order that still has events on the reorder queue. This also allows the reorder_complete() routine to be streamlined.
This patch resolves Bug https://bugs.linaro.org/show_bug.cgi?id=1879 Signed-off-by: Bill Fischofer <bill.fischo...@linaro.org> --- .../linux-generic/include/odp_queue_internal.h | 5 +- platform/linux-generic/odp_queue.c | 57 +++++++++++++++++----- 2 files changed, 48 insertions(+), 14 deletions(-) diff --git a/platform/linux-generic/include/odp_queue_internal.h b/platform/linux-generic/include/odp_queue_internal.h index 6120740..a70044b 100644 --- a/platform/linux-generic/include/odp_queue_internal.h +++ b/platform/linux-generic/include/odp_queue_internal.h @@ -335,8 +335,7 @@ static inline int reorder_deq(queue_entry_t *queue, static inline void reorder_complete(queue_entry_t *origin_qe, odp_buffer_hdr_t **reorder_buf_return, odp_buffer_hdr_t **placeholder_buf, - int placeholder_append, - int order_released) + int placeholder_append) { odp_buffer_hdr_t *reorder_buf = origin_qe->s.reorder_head; odp_buffer_hdr_t *next_buf; @@ -356,7 +355,7 @@ static inline void reorder_complete(queue_entry_t *origin_qe, reorder_buf = next_buf; order_release(origin_qe, 1); - } else if (!order_released && reorder_buf->flags.sustain) { + } else if (reorder_buf->flags.sustain) { reorder_buf = next_buf; } else { *reorder_buf_return = origin_qe->s.reorder_head; diff --git a/platform/linux-generic/odp_queue.c b/platform/linux-generic/odp_queue.c index 9cab9b2..a5e60d7 100644 --- a/platform/linux-generic/odp_queue.c +++ b/platform/linux-generic/odp_queue.c @@ -39,6 +39,11 @@ #include <string.h> +#define RESOLVE_ORDER 0 +#define SUSTAIN_ORDER 1 + +#define NOAPPEND 0 +#define APPEND 1 typedef struct queue_table_t { queue_entry_t queue[ODP_CONFIG_QUEUES]; @@ -521,8 +526,7 @@ int ordered_queue_enq(queue_entry_t *queue, if (sched && schedule_queue(queue)) ODP_ABORT("schedule_queue failed\n"); - reorder_complete(origin_qe, &reorder_buf, &placeholder_buf, - 1, 0); + reorder_complete(origin_qe, &reorder_buf, &placeholder_buf, APPEND); UNLOCK(&origin_qe->s.lock); if (reorder_buf) @@ -606,7 +610,8 @@ int odp_queue_enq_multi(odp_queue_t handle, const odp_event_t ev[], int num) for (i = 0; i < num; i++) buf_hdr[i] = odp_buf_to_hdr(odp_buffer_from_event(ev[i])); - return num == 0 ? 0 : queue->s.enqueue_multi(queue, buf_hdr, num, 1); + return num == 0 ? 0 : queue->s.enqueue_multi(queue, buf_hdr, + num, SUSTAIN_ORDER); } int odp_queue_enq(odp_queue_t handle, odp_event_t ev) @@ -620,7 +625,7 @@ int odp_queue_enq(odp_queue_t handle, odp_event_t ev) /* No chains via this entry */ buf_hdr->link = NULL; - return queue->s.enqueue(queue, buf_hdr, 1); + return queue->s.enqueue(queue, buf_hdr, SUSTAIN_ORDER); } int queue_enq_internal(odp_buffer_hdr_t *buf_hdr) @@ -660,7 +665,7 @@ odp_buffer_hdr_t *queue_deq(queue_entry_t *queue) buf_hdr->sync[i] = odp_atomic_fetch_inc_u64(&queue->s.sync_in[i]); } - buf_hdr->flags.sustain = 0; + buf_hdr->flags.sustain = SUSTAIN_ORDER; } else { buf_hdr->origin_qe = NULL; } @@ -713,7 +718,7 @@ int queue_deq_multi(queue_entry_t *queue, odp_buffer_hdr_t *buf_hdr[], int num) odp_atomic_fetch_inc_u64 (&queue->s.sync_in[j]); } - buf_hdr[i]->flags.sustain = 0; + buf_hdr[i]->flags.sustain = SUSTAIN_ORDER; } else { buf_hdr[i]->origin_qe = NULL; } @@ -879,7 +884,7 @@ int queue_pktout_enq(queue_entry_t *queue, odp_buffer_hdr_t *buf_hdr, order_release(origin_qe, release_count + placeholder_count); /* Now handle sends to other queues that are ready to go */ - reorder_complete(origin_qe, &reorder_buf, &placeholder_buf, 1, 0); + reorder_complete(origin_qe, &reorder_buf, &placeholder_buf, APPEND); /* We're fully done with the origin_qe at last */ UNLOCK(&origin_qe->s.lock); @@ -947,13 +952,43 @@ int release_order(queue_entry_t *origin_qe, uint64_t order, odp_buffer_t placeholder_buf; odp_buffer_hdr_t *placeholder_buf_hdr, *reorder_buf, *next_buf; - /* Must tlock the origin queue to process the release */ + /* Must lock the origin queue to process the release */ LOCK(&origin_qe->s.lock); - /* If we are in the order we can release immediately since there can - * be no confusion about intermediate elements + /* If we are in order we can release immediately since there can be no + * confusion about intermediate elements */ if (order <= origin_qe->s.order_out) { + reorder_buf = origin_qe->s.reorder_head; + + /* We're in order, however there may be one or more events on + * the reorder queue that are part of this order. If that is + * the case, remove them and let ordered_queue_enq() handle + * them and resolve the order for us. + */ + if (reorder_buf && reorder_buf->order == order) { + odp_buffer_hdr_t *reorder_head = reorder_buf; + + next_buf = reorder_buf->next; + + while (next_buf && next_buf->order == order) { + reorder_buf = next_buf; + next_buf = next_buf->next; + } + + origin_qe->s.reorder_head = reorder_buf->next; + reorder_buf->next = NULL; + + UNLOCK(&origin_qe->s.lock); + reorder_head->link = reorder_buf->next; + return ordered_queue_enq(reorder_head->target_qe, + reorder_head, RESOLVE_ORDER, + origin_qe, order); + } + + /* Reorder queue has no elements for this order, so it's safe + * to resolve order here + */ order_release(origin_qe, 1); /* Check if this release allows us to unblock waiters. At the @@ -965,7 +1000,7 @@ int release_order(queue_entry_t *origin_qe, uint64_t order, * element(s) on the reorder queue */ reorder_complete(origin_qe, &reorder_buf, - &placeholder_buf_hdr, 0, 1); + &placeholder_buf_hdr, NOAPPEND); /* Now safe to unlock */ UNLOCK(&origin_qe->s.lock); -- 2.1.4 _______________________________________________ lng-odp mailing list lng-odp@lists.linaro.org https://lists.linaro.org/mailman/listinfo/lng-odp
