[dpdk-dev] [PATCH v2 06/11] mempool/cnxk: add cn9k optimized mempool enqueue/dequeue

[Ashwin Sekhar T K <asekhar@marvell.com>]

Add Marvell CN9k mempool enqueue/dequeue. Marvell CN9k
supports burst dequeue which allows to dequeue up to 32
pointers using pipelined casp instructions.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
Signed-off-by: Ashwin Sekhar T K <asekhar@marvell.com>
---
 doc/guides/mempool/cnxk.rst             |  4 +++
 drivers/mempool/cnxk/cn9k_mempool_ops.c | 39 +++++++++++++++++++++++--
 2 files changed, 41 insertions(+), 2 deletions(-)

diff --git a/doc/guides/mempool/cnxk.rst b/doc/guides/mempool/cnxk.rst
index 907c19c841..f51532b101 100644
--- a/doc/guides/mempool/cnxk.rst
+++ b/doc/guides/mempool/cnxk.rst
@@ -21,6 +21,10 @@ CNXK NPA PMD supports:
 - Ethdev Rx buffer allocation in HW to save CPU cycles in the Rx path.
 - Ethdev Tx buffer recycling in HW to save CPU cycles in the Tx path.
 
+CN9k NPA supports:
+
+- Burst alloc of up to 32 pointers.
+
 Prerequisites and Compilation procedure
 ---------------------------------------
 
diff --git a/drivers/mempool/cnxk/cn9k_mempool_ops.c b/drivers/mempool/cnxk/cn9k_mempool_ops.c
index f5ac163af9..c0cdba640b 100644
--- a/drivers/mempool/cnxk/cn9k_mempool_ops.c
+++ b/drivers/mempool/cnxk/cn9k_mempool_ops.c
@@ -7,6 +7,41 @@
 #include "roc_api.h"
 #include "cnxk_mempool.h"
 
+static int __rte_hot
+cn9k_mempool_enq(struct rte_mempool *mp, void *const *obj_table, unsigned int n)
+{
+	/* Ensure mbuf init changes are written before the free pointers
+	 * are enqueued to the stack.
+	 */
+	rte_io_wmb();
+	roc_npa_aura_op_bulk_free(mp->pool_id, (const uint64_t *)obj_table, n,
+				  0);
+
+	return 0;
+}
+
+static inline int __rte_hot
+cn9k_mempool_deq(struct rte_mempool *mp, void **obj_table, unsigned int n)
+{
+	unsigned int count;
+
+	count = roc_npa_aura_op_bulk_alloc(mp->pool_id, (uint64_t *)obj_table,
+					   n, 0, 1);
+
+	if (unlikely(count != n)) {
+		/* If bulk alloc failed to allocate all pointers, try
+		 * allocating remaining pointers with the default alloc
+		 * with retry scheme.
+		 */
+		if (cnxk_mempool_deq(mp, &obj_table[count], n - count)) {
+			cn9k_mempool_enq(mp, obj_table, count);
+			return -ENOENT;
+		}
+	}
+
+	return 0;
+}
+
 static int
 cn9k_mempool_alloc(struct rte_mempool *mp)
 {
@@ -44,8 +79,8 @@ static struct rte_mempool_ops cn9k_mempool_ops = {
 	.name = "cn9k_mempool_ops",
 	.alloc = cn9k_mempool_alloc,
 	.free = cnxk_mempool_free,
-	.enqueue = cnxk_mempool_enq,
-	.dequeue = cnxk_mempool_deq,
+	.enqueue = cn9k_mempool_enq,
+	.dequeue = cn9k_mempool_deq,
 	.get_count = cnxk_mempool_get_count,
 	.calc_mem_size = cnxk_mempool_calc_mem_size,
 	.populate = cnxk_mempool_populate,
-- 
2.31.0