[dpdk-stable] [PATCH 1/3] lib/hash: use ordered loads only if signature matches

[Wang, Yipeng1]

>-----Original Message-----
>From: Honnappa Nagarahalli [mailto:honnappa.nagarahalli at arm.com]
>Sent: Tuesday, June 25, 2019 2:15 PM
>To: Wang, Yipeng1 <yipeng1.wang at intel.com>; Gobriel, Sameh <sameh.gobriel at intel.com>; Richardson, Bruce
><bruce.richardson at intel.com>; De Lara Guarch, Pablo <pablo.de.lara.guarch at intel.com>; honnappa.nagarahalli at arm.com
>Cc: gavin.hu at arm.com; ruifeng.wang at arm.com; dev at dpdk.org; nd at arm.com; stable at dpdk.org
>Subject: [PATCH 1/3] lib/hash: use ordered loads only if signature matches
>
>Relaxed signature comparison is done first. Further ordered loads
>are done only if the signature matches. Any false positives are
>caught by the full key comparison.
[Wang, Yipeng] add: This commit improves lookup performance.
>
>Fixes: e605a1d36 ("hash: add lock-free r/w concurrency")
>Cc: stable at dpdk.org
>
>Signed-off-by: Honnappa Nagarahalli <honnappa.nagarahalli at arm.com>
>Reviewed-by: Gavin Hu <gavin.hu at arm.com>
>Tested-by: Ruifeng Wang <ruifeng.wang at arm.com>
>---
> lib/librte_hash/rte_cuckoo_hash.c | 35 ++++++++++++++++++-------------
> 1 file changed, 21 insertions(+), 14 deletions(-)
>
>diff --git a/lib/librte_hash/rte_cuckoo_hash.c b/lib/librte_hash/rte_cuckoo_hash.c
>index 953928f27..f37f6957d 100644
>--- a/lib/librte_hash/rte_cuckoo_hash.c
>+++ b/lib/librte_hash/rte_cuckoo_hash.c
>@@ -1188,22 +1188,29 @@ search_one_bucket_lf(const struct rte_hash *h, const void *key, uint16_t sig,
> 	struct rte_hash_key *k, *keys = h->key_store;
>
> 	for (i = 0; i < RTE_HASH_BUCKET_ENTRIES; i++) {
>-		key_idx = __atomic_load_n(&bkt->key_idx[i],
>+		/* Signature comparison is done before the acquire-load
>+		 * of the key index to achieve better performance.
>+		 * Any false positives will be caught in full comparison
>+		 * of the key.
>+		 */
[Wang, Yipeng] 
A bit more explanation would be helpful for future understanding of the code:
"Signature comparison is done ... performance.
Although this could accidentally cause the reader to read an old signature while the key_idx
is updated to a new key's,  any false positive will be .... of the key."

Otherwise:
Acked-by: Yipeng Wang <yipeng1.wang at intel.com>