[v11,2/2] net/i40e: replace put function
Checks
Commit Message
Integrated zero-copy put API in mempool cache in i40e PMD.
On Ampere Altra server, l3fwd single core's performance improves by 5%
with the new API
Signed-off-by: Kamalakshitha Aligeri <kamalakshitha.aligeri@arm.com>
Reviewed-by: Ruifeng Wang <ruifeng.wang@arm.com>
Reviewed-by: Feifei Wang <feifei.wang2@arm.com>
---
drivers/net/i40e/i40e_rxtx_vec_common.h | 27 ++++++++++++++++++++-----
1 file changed, 22 insertions(+), 5 deletions(-)
--
2.25.1
Comments
> From: Kamalakshitha Aligeri [mailto:kamalakshitha.aligeri@arm.com]
> Sent: Wednesday, 5 July 2023 20.03
>
> Integrated zero-copy put API in mempool cache in i40e PMD.
> On Ampere Altra server, l3fwd single core's performance improves by 5%
> with the new API
>
> Signed-off-by: Kamalakshitha Aligeri <kamalakshitha.aligeri@arm.com>
> Reviewed-by: Ruifeng Wang <ruifeng.wang@arm.com>
> Reviewed-by: Feifei Wang <feifei.wang2@arm.com>
> ---
> drivers/net/i40e/i40e_rxtx_vec_common.h | 27 ++++++++++++++++++++-----
> 1 file changed, 22 insertions(+), 5 deletions(-)
>
> diff --git a/drivers/net/i40e/i40e_rxtx_vec_common.h
> b/drivers/net/i40e/i40e_rxtx_vec_common.h
> index fe1a6ec75e..35cdb31b2e 100644
> --- a/drivers/net/i40e/i40e_rxtx_vec_common.h
> +++ b/drivers/net/i40e/i40e_rxtx_vec_common.h
> @@ -95,18 +95,35 @@ i40e_tx_free_bufs(struct i40e_tx_queue *txq)
>
> n = txq->tx_rs_thresh;
>
> - /* first buffer to free from S/W ring is at index
> - * tx_next_dd - (tx_rs_thresh-1)
> - */
> + /* first buffer to free from S/W ring is at index
> + * tx_next_dd - (tx_rs_thresh-1)
> + */
> txep = &txq->sw_ring[txq->tx_next_dd - (n - 1)];
>
> if (txq->offloads & RTE_ETH_TX_OFFLOAD_MBUF_FAST_FREE) {
> + struct rte_mempool *mp = txep[0].mbuf->pool;
> + struct rte_mempool_cache *cache =
> rte_mempool_default_cache(mp, rte_lcore_id());
> + void **cache_objs;
> +
> + if (unlikely(!cache))
> + goto fallback;
> +
> + cache_objs = rte_mempool_cache_zc_put_bulk(cache, mp, n);
> + if (unlikely(!cache_objs))
> + goto fallback;
> +
> for (i = 0; i < n; i++) {
> - free[i] = txep[i].mbuf;
> + cache_objs[i] = txep[i].mbuf;
> /* no need to reset txep[i].mbuf in vector path */
> }
> - rte_mempool_put_bulk(free[0]->pool, (void **)free, n);
> goto done;
> +
> +fallback:
> + for (i = 0; i < n; i++)
> + free[i] = txep[i].mbuf;
> + rte_mempool_generic_put(mp, (void **)free, n, cache);
Patchwork, when building with gcc-debug for FreeBSD13-64 and RHEL92-64, complains that "free" may be used uninitialized in rte_mempool_check_cookies() called via RTE_MEMPOOL_CHECK_COOKIES() from rte_mempool_generic_put() here. But "free" is initialized by the preceding loop, so I don't understand why GCC thinks it may be uninitialized.
Does adding curly braces to the loop help GCC understand that the free array is initialized, or what is the proper workaround?
> + goto done;
> +
> }
>
> m = rte_pktmbuf_prefree_seg(txep[0].mbuf);
> --
> 2.25.1
Acked-by: Morten Brørup <mb@smartsharesystems.com>
@@ -95,18 +95,35 @@ i40e_tx_free_bufs(struct i40e_tx_queue *txq)
n = txq->tx_rs_thresh;
- /* first buffer to free from S/W ring is at index
- * tx_next_dd - (tx_rs_thresh-1)
- */
+ /* first buffer to free from S/W ring is at index
+ * tx_next_dd - (tx_rs_thresh-1)
+ */
txep = &txq->sw_ring[txq->tx_next_dd - (n - 1)];
if (txq->offloads & RTE_ETH_TX_OFFLOAD_MBUF_FAST_FREE) {
+ struct rte_mempool *mp = txep[0].mbuf->pool;
+ struct rte_mempool_cache *cache = rte_mempool_default_cache(mp, rte_lcore_id());
+ void **cache_objs;
+
+ if (unlikely(!cache))
+ goto fallback;
+
+ cache_objs = rte_mempool_cache_zc_put_bulk(cache, mp, n);
+ if (unlikely(!cache_objs))
+ goto fallback;
+
for (i = 0; i < n; i++) {
- free[i] = txep[i].mbuf;
+ cache_objs[i] = txep[i].mbuf;
/* no need to reset txep[i].mbuf in vector path */
}
- rte_mempool_put_bulk(free[0]->pool, (void **)free, n);
goto done;
+
+fallback:
+ for (i = 0; i < n; i++)
+ free[i] = txep[i].mbuf;
+ rte_mempool_generic_put(mp, (void **)free, n, cache);
+ goto done;
+
}
m = rte_pktmbuf_prefree_seg(txep[0].mbuf);