-----Original Message-----
> Date: Fri, 9 Nov 2018 10:39:16 -0600
> From: Honnappa Nagarahalli <honnappa.nagaraha...@arm.com>
> To: bruce.richard...@intel.com, pablo.de.lara.gua...@intel.com
> CC: dev@dpdk.org, jerin.ja...@caviumnetworks.com, hemant.agra...@nxp.com,
> chao...@linux.vnet.ibm.com, yipeng1.w...@intel.com,
> dharmik.thak...@arm.com, gavin...@arm.com, honnappa.nagaraha...@arm.com,
> n...@arm.com
> Subject: [PATCH 3/4] hash: remove memory orderings from rw-lock lookup fns
> X-Mailer: git-send-email 2.17.1
>
>
> Remove the memory orderings from lookup functions using
> rw-lock.
> This is an intermediate commit meant to ease the
> review process.
>
> Fixes: e605a1d36 ("hash: add lock-free r/w concurrency")
> Cc: honnappa.nagaraha...@arm.com
>
> Suggested-by: Jerin Jacob <jerin.ja...@caviumnetworks.com>
> Signed-off-by: Honnappa Nagarahalli <honnappa.nagaraha...@arm.com>
> Reviewed-by: Ola Liljedahl <ola.liljed...@arm.com>
> Reviewed-by: Gavin Hu <gavin...@arm.com>
> ---
> lib/librte_hash/rte_cuckoo_hash.c | 277 +++++++++++-------------------
> 1 file changed, 105 insertions(+), 172 deletions(-)
>
> diff --git a/lib/librte_hash/rte_cuckoo_hash.c
> b/lib/librte_hash/rte_cuckoo_hash.c
> index e6b84c6bc..9390dc5e4 100644
> --- a/lib/librte_hash/rte_cuckoo_hash.c
> +++ b/lib/librte_hash/rte_cuckoo_hash.c
> @@ -1135,27 +1135,22 @@ search_one_bucket(const struct rte_hash *h, const
> void *key, uint16_t sig,
> void **data, const struct rte_hash_bucket *bkt)
> {
> int i;
> - uint32_t key_idx;
> - void *pdata;
> 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],
> - __ATOMIC_ACQUIRE);
> - if (bkt->sig_current[i] == sig && key_idx != EMPTY_SLOT) {
> + if (bkt->sig_current[i] == sig &&
> + bkt->key_idx[i] != EMPTY_SLOT) {
> k = (struct rte_hash_key *) ((char *)keys +
> - key_idx * h->key_entry_size);
> - pdata = __atomic_load_n(&k->pdata,
> - __ATOMIC_ACQUIRE);
> + bkt->key_idx[i] * h->key_entry_size);
>
> if (rte_hash_cmp_eq(key, k->key, h) == 0) {
> if (data != NULL)
> - *data = pdata;
> + *data = k->pdata;
> /*
> * Return index where key is stored,
> * subtracting the first dummy index
> */
> - return key_idx - 1;
> + return bkt->key_idx[i] - 1;
> }
> }
> }
> @@ -1201,7 +1196,6 @@ __rte_hash_lookup_with_hash(const struct rte_hash *h,
> const void *key,
> {
> uint32_t prim_bucket_idx, sec_bucket_idx;
> struct rte_hash_bucket *bkt, *cur_bkt;
> - uint32_t cnt_b, cnt_a;
> int ret;
> uint16_t short_sig;
>
> @@ -1211,49 +1205,25 @@ __rte_hash_lookup_with_hash(const struct rte_hash *h,
> const void *key,
>
> __hash_rw_reader_lock(h);
>
> - do {
> - /* Load the table change counter before the lookup
> - * starts. Acquire semantics will make sure that
> - * loads in search_one_bucket are not hoisted.
> - */
> - cnt_b = __atomic_load_n(h->tbl_chng_cnt,
> - __ATOMIC_ACQUIRE);
> + /* Check if key is in primary location */
> + bkt = &h->buckets[prim_bucket_idx];
In original version, this bkt assignment is before to __hash_rw_reader_lock().
This causing performance issue in lookup 'hit' case.
Following change is fixing it.i.e bringing back to orginal version.
[master]83xx1.2[dpdk]# git diff
diff --git a/lib/librte_hash/rte_cuckoo_hash.c
b/lib/librte_hash/rte_cuckoo_hash.c
index 7e1a9ac96..bc8a55f0f 100644
--- a/lib/librte_hash/rte_cuckoo_hash.c
+++ b/lib/librte_hash/rte_cuckoo_hash.c
@@ -1204,10 +1204,11 @@ __rte_hash_lookup_with_hash_l(const struct
rte_hash *h, const void *key,
prim_bucket_idx = get_prim_bucket_index(h, sig);
sec_bucket_idx = get_alt_bucket_index(h, prim_bucket_idx,
short_sig);
- __hash_rw_reader_lock(h);
-
/* Check if key is in primary location */
bkt = &h->buckets[prim_bucket_idx];
+
+ __hash_rw_reader_lock(h);
+
ret = search_one_bucket_l(h, key, short_sig, data, bkt);
if (ret != -1) {
__hash_rw_reader_unlock(h);
Could you send the final version that needs to taken into tree.
i.e remove intermediate commits only for review purpose.
I can test it finally with that.
