From: Adrien Mazarguil <adrien.mazarguil@6wind.com>
To: dev@dpdk.org
Subject: [dpdk-dev] [PATCH v2 6/7] mlx5: free buffers immediately after completion
Date: Mon, 22 Feb 2016 19:18:02 +0100 [thread overview]
Message-ID: <1456165083-28316-7-git-send-email-adrien.mazarguil@6wind.com> (raw)
In-Reply-To: <1456165083-28316-1-git-send-email-adrien.mazarguil@6wind.com>
From: Nelio Laranjeiro <nelio.laranjeiro@6wind.com>
This lowers the amount of cache misses.
Signed-off-by: Nelio Laranjeiro <nelio.laranjeiro@6wind.com>
---
drivers/net/mlx5/mlx5_rxtx.c | 35 ++++++++++++++++++++---------------
1 file changed, 20 insertions(+), 15 deletions(-)
diff --git a/drivers/net/mlx5/mlx5_rxtx.c b/drivers/net/mlx5/mlx5_rxtx.c
index b82017e..622ac17 100644
--- a/drivers/net/mlx5/mlx5_rxtx.c
+++ b/drivers/net/mlx5/mlx5_rxtx.c
@@ -84,6 +84,7 @@ txq_complete(struct txq *txq)
{
unsigned int elts_comp = txq->elts_comp;
unsigned int elts_tail = txq->elts_tail;
+ unsigned int elts_free = txq->elts_tail;
const unsigned int elts_n = txq->elts_n;
int wcs_n;
@@ -110,6 +111,25 @@ txq_complete(struct txq *txq)
elts_tail += wcs_n * txq->elts_comp_cd_init;
if (elts_tail >= elts_n)
elts_tail -= elts_n;
+
+ while (elts_free != elts_tail) {
+ struct txq_elt *elt = &(*txq->elts)[elts_free];
+ unsigned int elts_free_next =
+ (((elts_free + 1) == elts_n) ? 0 : elts_free + 1);
+ struct rte_mbuf *tmp = elt->buf;
+ struct txq_elt *elt_next = &(*txq->elts)[elts_free_next];
+
+ RTE_MBUF_PREFETCH_TO_FREE(elt_next->buf);
+ /* Faster than rte_pktmbuf_free(). */
+ do {
+ struct rte_mbuf *next = NEXT(tmp);
+
+ rte_pktmbuf_free_seg(tmp);
+ tmp = next;
+ } while (tmp != NULL);
+ elts_free = elts_free_next;
+ }
+
txq->elts_tail = elts_tail;
txq->elts_comp = elts_comp;
return 0;
@@ -464,7 +484,6 @@ mlx5_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n)
struct rte_mbuf *buf_next = pkts[i + 1];
unsigned int elts_head_next =
(((elts_head + 1) == elts_n) ? 0 : elts_head + 1);
- struct txq_elt *elt_next = &(*txq->elts)[elts_head_next];
struct txq_elt *elt = &(*txq->elts)[elts_head];
unsigned int segs = NB_SEGS(buf);
#ifdef MLX5_PMD_SOFT_COUNTERS
@@ -472,18 +491,6 @@ mlx5_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n)
#endif
uint32_t send_flags = 0;
- /* Clean up old buffer. */
- if (likely(elt->buf != NULL)) {
- struct rte_mbuf *tmp = elt->buf;
-
- /* Faster than rte_pktmbuf_free(). */
- do {
- struct rte_mbuf *next = NEXT(tmp);
-
- rte_pktmbuf_free_seg(tmp);
- tmp = next;
- } while (tmp != NULL);
- }
if (i + 1 < max)
rte_prefetch0(buf_next);
/* Request TX completion. */
@@ -517,7 +524,6 @@ mlx5_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n)
if (txq->priv->vf)
rte_prefetch0((volatile void *)
(uintptr_t)addr);
- RTE_MBUF_PREFETCH_TO_FREE(elt_next->buf);
/* Prefetch next buffer data. */
if (i + 1 < max) {
buf_next_addr =
@@ -568,7 +574,6 @@ mlx5_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n)
&sges);
if (ret.length == (unsigned int)-1)
goto stop;
- RTE_MBUF_PREFETCH_TO_FREE(elt_next->buf);
/* Put SG list into send queue. */
err = txq->send_pending_sg_list
(txq->qp,
--
2.1.4
next prev parent reply other threads:[~2016-02-22 18:18 UTC|newest]
Thread overview: 24+ messages / expand[flat|nested] mbox.gz Atom feed top
2016-01-29 10:32 [dpdk-dev] [PATCH 0/6] Performance optimizations for mlx5 Adrien Mazarguil
2016-01-29 10:32 ` [dpdk-dev] [PATCH 1/6] mlx5: prefetch next TX mbuf header and data Adrien Mazarguil
2016-01-29 10:32 ` [dpdk-dev] [PATCH 2/6] mlx5: reorder TX/RX queue structure Adrien Mazarguil
2016-01-29 10:32 ` [dpdk-dev] [PATCH 3/6] mlx5: remove one indirection level from RX/TX functions Adrien Mazarguil
2016-01-29 10:32 ` [dpdk-dev] [PATCH 4/6] mlx5: process offload flags only when requested Adrien Mazarguil
2016-01-29 10:32 ` [dpdk-dev] [PATCH 5/6] mlx5: avoid lkey retrieval for inlined packets Adrien Mazarguil
2016-01-29 10:32 ` [dpdk-dev] [PATCH 6/6] mlx5: free buffers immediately after completion Adrien Mazarguil
2016-02-22 18:17 ` [dpdk-dev] [PATCH v2 0/7] Performance optimizations for mlx5 and mlx4 Adrien Mazarguil
2016-02-22 18:17 ` [dpdk-dev] [PATCH v2 1/7] mlx5: prefetch next TX mbuf header and data Adrien Mazarguil
2016-02-22 18:17 ` [dpdk-dev] [PATCH v2 2/7] mlx5: reorder TX/RX queue structure Adrien Mazarguil
2016-02-22 18:17 ` [dpdk-dev] [PATCH v2 3/7] mlx5: remove one indirection level from RX/TX functions Adrien Mazarguil
2016-02-22 18:18 ` [dpdk-dev] [PATCH v2 4/7] mlx5: process offload flags only when requested Adrien Mazarguil
2016-02-22 18:18 ` [dpdk-dev] [PATCH v2 5/7] mlx5: avoid lkey retrieval for inlined packets Adrien Mazarguil
2016-02-22 18:18 ` Adrien Mazarguil [this message]
2016-02-22 18:18 ` [dpdk-dev] [PATCH v2 7/7] mlx: use aligned memory to register regions Adrien Mazarguil
2016-03-03 14:27 ` [dpdk-dev] [PATCH v3 0/7] Performance optimizations for mlx5 and mlx4 Adrien Mazarguil
2016-03-03 14:27 ` [dpdk-dev] [PATCH v3 1/7] mlx5: prefetch next TX mbuf header and data Adrien Mazarguil
2016-03-03 14:27 ` [dpdk-dev] [PATCH v3 2/7] mlx5: reorder TX/RX queue structure Adrien Mazarguil
2016-03-03 14:27 ` [dpdk-dev] [PATCH v3 3/7] mlx5: remove one indirection level from RX/TX functions Adrien Mazarguil
2016-03-03 14:27 ` [dpdk-dev] [PATCH v3 4/7] mlx5: process offload flags only when requested Adrien Mazarguil
2016-03-03 14:27 ` [dpdk-dev] [PATCH v3 5/7] mlx5: avoid lkey retrieval for inlined packets Adrien Mazarguil
2016-03-03 14:27 ` [dpdk-dev] [PATCH v3 6/7] mlx5: free buffers immediately after completion Adrien Mazarguil
2016-03-03 14:27 ` [dpdk-dev] [PATCH v3 7/7] mlx: use aligned memory to register regions Adrien Mazarguil
2016-03-09 16:28 ` [dpdk-dev] [PATCH v3 0/7] Performance optimizations for mlx5 and mlx4 Bruce Richardson
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=1456165083-28316-7-git-send-email-adrien.mazarguil@6wind.com \
--to=adrien.mazarguil@6wind.com \
--cc=dev@dpdk.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).