[PATCH v2 1/1] mempool: implement index-based per core cache

Wang, Haiyue haiyue.wang at intel.com
Sun Jan 23 08:13:50 CET 2022


> -----Original Message-----
> From: Dharmik Thakkar <dharmik.thakkar at arm.com>
> Sent: Thursday, January 13, 2022 13:37
> To: Olivier Matz <olivier.matz at 6wind.com>; Andrew Rybchenko <andrew.rybchenko at oktetlabs.ru>
> Cc: dev at dpdk.org; nd at arm.com; honnappa.nagarahalli at arm.com; ruifeng.wang at arm.com; Dharmik Thakkar
> <dharmik.thakkar at arm.com>
> Subject: [PATCH v2 1/1] mempool: implement index-based per core cache
> 
> Current mempool per core cache implementation stores pointers to mbufs
> On 64b architectures, each pointer consumes 8B
> This patch replaces it with index-based implementation,
> where in each buffer is addressed by (pool base address + index)
> It reduces the amount of memory/cache required for per core cache
> 
> L3Fwd performance testing reveals minor improvements in the cache
> performance (L1 and L2 misses reduced by 0.60%)
> with no change in throughput
> 
> Suggested-by: Honnappa Nagarahalli <honnappa.nagarahalli at arm.com>
> Signed-off-by: Dharmik Thakkar <dharmik.thakkar at arm.com>
> Reviewed-by: Ruifeng Wang <ruifeng.wang at arm.com>
> ---
>  lib/mempool/rte_mempool.h             | 150 +++++++++++++++++++++++++-
>  lib/mempool/rte_mempool_ops_default.c |   7 ++
>  2 files changed, 156 insertions(+), 1 deletion(-)
> 
> diff --git a/lib/mempool/rte_mempool.h b/lib/mempool/rte_mempool.h
> index 1e7a3c15273c..f2403fbc97a7 100644


> diff --git a/lib/mempool/rte_mempool_ops_default.c b/lib/mempool/rte_mempool_ops_default.c
> index 22fccf9d7619..3543cad9d4ce 100644
> --- a/lib/mempool/rte_mempool_ops_default.c
> +++ b/lib/mempool/rte_mempool_ops_default.c
> @@ -127,6 +127,13 @@ rte_mempool_op_populate_helper(struct rte_mempool *mp, unsigned int flags,
>  		obj = va + off;
>  		obj_cb(mp, obj_cb_arg, obj,
>  		       (iova == RTE_BAD_IOVA) ? RTE_BAD_IOVA : (iova + off));
> +#ifdef RTE_MEMPOOL_INDEX_BASED_LCORE_CACHE
> +		/* Store pool base value to calculate indices for index-based
> +		 * lcore cache implementation
> +		 */
> +		if (i == 0)
> +			mp->pool_base_value = obj;

This is wrong, the populate may run many times. ;-)

I tried bellow patch to run "rte_pktmbuf_pool_create(mbuf_pool_0, 1048575, 256, 0, 4096, 0)"

This is the debug message (also, your patch will make DPDK mempool not support > 4GB):

2bfffdb40 (from last debug line 'max') - 1b3fff240 (from first line 'base addr') = 10BFFE900

****mempool mbuf_pool_0 (size = 1048575, populated_size = 46952, elt_size = 4224): base addr = 0x1b3fff240, max = 0x0, diff = 18446744066394688960 (max_objs = 1048575)
****mempool mbuf_pool_0 (size = 1048575, populated_size = 297358, elt_size = 4224): base addr = 0x1c0000040, max = 0x0, diff = 18446744066193358784 (max_objs = 1001623)
****mempool mbuf_pool_0 (size = 1048575, populated_size = 547764, elt_size = 4224): base addr = 0x200000040, max = 0x0, diff = 18446744065119616960 (max_objs = 751217)
****mempool mbuf_pool_0 (size = 1048575, populated_size = 798170, elt_size = 4224): base addr = 0x240000040, max = 0x0, diff = 18446744064045875136 (max_objs = 500811)
****mempool mbuf_pool_0 (size = 1048575, populated_size = 1048575, elt_size = 4224): base addr = 0x280000040, max = 0x2bfffdb40, diff = 1073732352 (max_objs = 250405)

diff --git a/lib/mempool/rte_mempool_ops_default.c b/lib/mempool/rte_mempool_ops_default.c
index 22fccf9d76..854067cd43 100644
--- a/lib/mempool/rte_mempool_ops_default.c
+++ b/lib/mempool/rte_mempool_ops_default.c
@@ -99,6 +99,7 @@ rte_mempool_op_populate_helper(struct rte_mempool *mp, unsigned int flags,
        unsigned int i;
        void *obj;
        int ret;
+       void *pool_base_value = NULL, *pool_max_value = NULL;

        ret = rte_mempool_get_page_size(mp, &pg_sz);
        if (ret < 0)
@@ -128,9 +129,20 @@ rte_mempool_op_populate_helper(struct rte_mempool *mp, unsigned int flags,
                obj_cb(mp, obj_cb_arg, obj,
                       (iova == RTE_BAD_IOVA) ? RTE_BAD_IOVA : (iova + off));
                rte_mempool_ops_enqueue_bulk(mp, &obj, 1);
+               if (i == 0)
+                       pool_base_value = obj;
+               else if (i == (max_objs - 1))
+                       pool_max_value = obj;
                off += mp->elt_size + mp->trailer_size;
        }

+       printf("****mempool %s (size = %u, populated_size = %u, elt_size = %u): base addr = 0x%llx, max = 0x%llx, diff = %lu (max_objs = %u)\n",
+               mp->name, mp->size, mp->populated_size,
+               mp->elt_size,
+               (unsigned long long) pool_base_value,
+               (unsigned long long) pool_max_value,
+               RTE_PTR_DIFF(pool_max_value, pool_base_value), max_objs);
+
        return i;
 }


> +#endif
>  		rte_mempool_ops_enqueue_bulk(mp, &obj, 1);
>  		off += mp->elt_size + mp->trailer_size;
>  	}
> --
> 2.17.1



More information about the dev mailing list