[PATCH 2/4] common/mlx5: release unused mempool entries

[Bing Zhao]

From: Roi Dayan <roid at nvidia.com>

When creating a new mempool but assigning a shared entries
from a different mempool need to release the newly unused
allocated entries. Fix it.

Fixes: 8947eebc999e ("common/mlx5: fix shared memory region ranges allocation")
Cc: bingz at nvidia.com

Signed-off-by: Roi Dayan <roid at nvidia.com>
Signed-off-by: Gregory Etelson <getelson at nvidia.com>
---
 drivers/common/mlx5/mlx5_common_mr.c | 6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/drivers/common/mlx5/mlx5_common_mr.c b/drivers/common/mlx5/mlx5_common_mr.c
index c41ffff2d5..8ed988dec9 100644
--- a/drivers/common/mlx5/mlx5_common_mr.c
+++ b/drivers/common/mlx5/mlx5_common_mr.c
@@ -1717,18 +1717,24 @@ mlx5_mr_mempool_register_primary(struct mlx5_mr_share_cache *share_cache,
 	 * hugepage can be shared across mempools that also fit in it.
 	 */
 	if (share_hugepage) {
+		struct mlx5_mempool_mr *gc_mrs = NULL;
+
 		rte_rwlock_write_lock(&share_cache->rwlock);
 		LIST_FOREACH(mpr, &share_cache->mempool_reg_list, next) {
 			if (mpr->mrs[0].pmd_mr.addr == (void *)ranges[0].start)
 				break;
 		}
 		if (mpr != NULL) {
+			/* Releasing MRs here can create a dead-lock on share_cache->rwlock */
+			gc_mrs = new_mpr->mrs;
 			new_mpr->mrs = mpr->mrs;
 			mlx5_mempool_reg_attach(new_mpr);
 			LIST_INSERT_HEAD(&share_cache->mempool_reg_list,
 					 new_mpr, next);
 		}
 		rte_rwlock_write_unlock(&share_cache->rwlock);
+		if (gc_mrs != NULL)
+			mlx5_free(gc_mrs);
 		if (mpr != NULL) {
 			DRV_LOG(DEBUG, "Shared MR %#x in PD %p for mempool %s with mempool %s",
 				mpr->mrs[0].pmd_mr.lkey, pd, mp->name,
-- 
2.34.1