Re: [dpdk-dev] [PATCH 3/4] hash: remove memory orderings from rw-lock lookup fns

[Jerin Jacob <jerin.jacob@caviumnetworks.com>]

-----Original Message-----
> Date: Fri, 9 Nov 2018 10:39:16 -0600
> From: Honnappa Nagarahalli <honnappa.nagarahalli@arm.com>
> To: bruce.richardson@intel.com, pablo.de.lara.guarch@intel.com
> CC: dev@dpdk.org, jerin.jacob@caviumnetworks.com, hemant.agrawal@nxp.com,
>  chaozhu@linux.vnet.ibm.com, yipeng1.wang@intel.com,
>  dharmik.thakkar@arm.com, gavin.hu@arm.com, honnappa.nagarahalli@arm.com,
>  nd@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.nagarahalli@arm.com
> 
> Suggested-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
> Signed-off-by: Honnappa Nagarahalli <honnappa.nagarahalli@arm.com>
> Reviewed-by: Ola Liljedahl <ola.liljedahl@arm.com>
> Reviewed-by: Gavin Hu <gavin.hu@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.