From: "Phil Yang (Arm Technology China)" <Phil.Yang@arm.com>
To: "Phil Yang (Arm Technology China)" <Phil.Yang@arm.com>,
"jgrajcia@cisco.com" <jgrajcia@cisco.com>,
"dev@dpdk.org" <dev@dpdk.org>
Cc: "thomas@monjalon.net" <thomas@monjalon.net>,
"jerinj@marvell.com" <jerinj@marvell.com>,
Honnappa Nagarahalli <Honnappa.Nagarahalli@arm.com>,
"damarion@cisco.com" <damarion@cisco.com>, nd <nd@arm.com>,
"Gavin Hu (Arm Technology China)" <Gavin.Hu@arm.com>,
nd <nd@arm.com>
Subject: Re: [dpdk-dev] [PATCH v1] net/memif: optimized with one-way barrier
Date: Mon, 26 Aug 2019 11:03:25 +0000 [thread overview]
Message-ID: <VE1PR08MB464091C924770F9E1757675DE9A10@VE1PR08MB4640.eurprd08.prod.outlook.com> (raw)
In-Reply-To: <1566817214-26599-1-git-send-email-phil.yang@arm.com>
+ Gavin
> -----Original Message-----
> From: dev <dev-bounces@dpdk.org> On Behalf Of Phil Yang
> Sent: Monday, August 26, 2019 7:00 PM
> To: jgrajcia@cisco.com; dev@dpdk.org
> Cc: thomas@monjalon.net; jerinj@marvell.com; Honnappa Nagarahalli
> <Honnappa.Nagarahalli@arm.com>; damarion@cisco.com; nd
> <nd@arm.com>
> Subject: [dpdk-dev] [PATCH v1] net/memif: optimized with one-way barrier
>
> Using 'rte_mb' to synchronize the shared ring head/tail between producer
> and consumer will stall the pipeline and damage performance on the weak
> memory model platforms, such like aarch64. Meanwhile update the shared
> ring head and tail are observable and ordered between CPUs on IA.
>
> Optimized this full barrier with the one-way barrier can improve the
> throughput. On aarch64 n1sdp server this patch make testpmd throughput
> boost 2.1%. On Intel E5-2640, testpmd got 3.98% performance gain.
>
> Signed-off-by: Phil Yang <phil.yang@arm.com>
> Reviewed-by: Gavin Hu <gavin.hu@arm.com>
> ---
> drivers/net/memif/memif.h | 4 +--
> drivers/net/memif/rte_eth_memif.c | 53 +++++++++++++++++++++--------
> ----------
> 2 files changed, 31 insertions(+), 26 deletions(-)
>
> diff --git a/drivers/net/memif/memif.h b/drivers/net/memif/memif.h
> index 3948b1f..a4d88c0 100644
> --- a/drivers/net/memif/memif.h
> +++ b/drivers/net/memif/memif.h
> @@ -169,9 +169,9 @@ typedef struct {
> uint32_t cookie; /**< MEMIF_COOKIE */
> uint16_t flags; /**< flags */
> #define MEMIF_RING_FLAG_MASK_INT 1 /**< disable interrupt
> mode */
> - volatile uint16_t head; /**< pointer to ring buffer
> head */
> + uint16_t head; /**< pointer to ring buffer head */
> MEMIF_CACHELINE_ALIGN_MARK(cacheline1);
> - volatile uint16_t tail; /**< pointer to ring buffer
> tail */
> + uint16_t tail; /**< pointer to ring buffer tail */
> MEMIF_CACHELINE_ALIGN_MARK(cacheline2);
> memif_desc_t desc[0]; /**< buffer descriptors */
> } memif_ring_t;
> diff --git a/drivers/net/memif/rte_eth_memif.c
> b/drivers/net/memif/rte_eth_memif.c
> index a59f809..b1c871e 100644
> --- a/drivers/net/memif/rte_eth_memif.c
> +++ b/drivers/net/memif/rte_eth_memif.c
> @@ -275,8 +275,14 @@ eth_memif_rx(void *queue, struct rte_mbuf **bufs,
> uint16_t nb_pkts)
> ring_size = 1 << mq->log2_ring_size;
> mask = ring_size - 1;
>
> - cur_slot = (type == MEMIF_RING_S2M) ? mq->last_head : mq-
> >last_tail;
> - last_slot = (type == MEMIF_RING_S2M) ? ring->head : ring->tail;
> + if (type == MEMIF_RING_S2M) {
> + cur_slot = mq->last_head;
> + last_slot = __atomic_load_n(&ring->head,
> __ATOMIC_ACQUIRE);
> + } else {
> + cur_slot = mq->last_tail;
> + last_slot = __atomic_load_n(&ring->tail,
> __ATOMIC_ACQUIRE);
> + }
> +
> if (cur_slot == last_slot)
> goto refill;
> n_slots = last_slot - cur_slot;
> @@ -344,8 +350,7 @@ eth_memif_rx(void *queue, struct rte_mbuf **bufs,
> uint16_t nb_pkts)
>
> no_free_bufs:
> if (type == MEMIF_RING_S2M) {
> - rte_mb();
> - ring->tail = cur_slot;
> + __atomic_store_n(&ring->tail, cur_slot,
> __ATOMIC_RELEASE);
> mq->last_head = cur_slot;
> } else {
> mq->last_tail = cur_slot;
> @@ -353,7 +358,7 @@ eth_memif_rx(void *queue, struct rte_mbuf **bufs,
> uint16_t nb_pkts)
>
> refill:
> if (type == MEMIF_RING_M2S) {
> - head = ring->head;
> + head = __atomic_load_n(&ring->head,
> __ATOMIC_ACQUIRE);
> n_slots = ring_size - head + mq->last_tail;
>
> while (n_slots--) {
> @@ -361,8 +366,7 @@ eth_memif_rx(void *queue, struct rte_mbuf **bufs,
> uint16_t nb_pkts)
> d0 = &ring->desc[s0];
> d0->length = pmd->run.pkt_buffer_size;
> }
> - rte_mb();
> - ring->head = head;
> + __atomic_store_n(&ring->head, head, __ATOMIC_RELEASE);
> }
>
> mq->n_pkts += n_rx_pkts;
> @@ -398,14 +402,16 @@ eth_memif_tx(void *queue, struct rte_mbuf
> **bufs, uint16_t nb_pkts)
> ring_size = 1 << mq->log2_ring_size;
> mask = ring_size - 1;
>
> - n_free = ring->tail - mq->last_tail;
> + n_free = __atomic_load_n(&ring->tail, __ATOMIC_ACQUIRE) - mq-
> >last_tail;
> mq->last_tail += n_free;
> - slot = (type == MEMIF_RING_S2M) ? ring->head : ring->tail;
>
> - if (type == MEMIF_RING_S2M)
> - n_free = ring_size - ring->head + mq->last_tail;
> - else
> - n_free = ring->head - ring->tail;
> + if (type == MEMIF_RING_S2M) {
> + slot = __atomic_load_n(&ring->head, __ATOMIC_ACQUIRE);
> + n_free = ring_size - slot + mq->last_tail;
> + } else {
> + slot = __atomic_load_n(&ring->tail, __ATOMIC_ACQUIRE);
> + n_free = __atomic_load_n(&ring->head,
> __ATOMIC_ACQUIRE) - slot;
> + }
>
> while (n_tx_pkts < nb_pkts && n_free) {
> mbuf_head = *bufs++;
> @@ -464,11 +470,10 @@ eth_memif_tx(void *queue, struct rte_mbuf
> **bufs, uint16_t nb_pkts)
> }
>
> no_free_slots:
> - rte_mb();
> if (type == MEMIF_RING_S2M)
> - ring->head = slot;
> + __atomic_store_n(&ring->head, slot, __ATOMIC_RELEASE);
> else
> - ring->tail = slot;
> + __atomic_store_n(&ring->tail, slot, __ATOMIC_RELEASE);
>
> if ((ring->flags & MEMIF_RING_FLAG_MASK_INT) == 0) {
> a = 1;
> @@ -611,8 +616,8 @@ memif_init_rings(struct rte_eth_dev *dev)
>
> for (i = 0; i < pmd->run.num_s2m_rings; i++) {
> ring = memif_get_ring(pmd, proc_private,
> MEMIF_RING_S2M, i);
> - ring->head = 0;
> - ring->tail = 0;
> + __atomic_store_n(&ring->head, 0, __ATOMIC_RELAXED);
> + __atomic_store_n(&ring->tail, 0, __ATOMIC_RELAXED);
> ring->cookie = MEMIF_COOKIE;
> ring->flags = 0;
> for (j = 0; j < (1 << pmd->run.log2_ring_size); j++) {
> @@ -627,8 +632,8 @@ memif_init_rings(struct rte_eth_dev *dev)
>
> for (i = 0; i < pmd->run.num_m2s_rings; i++) {
> ring = memif_get_ring(pmd, proc_private,
> MEMIF_RING_M2S, i);
> - ring->head = 0;
> - ring->tail = 0;
> + __atomic_store_n(&ring->head, 0, __ATOMIC_RELAXED);
> + __atomic_store_n(&ring->tail, 0, __ATOMIC_RELAXED);
> ring->cookie = MEMIF_COOKIE;
> ring->flags = 0;
> for (j = 0; j < (1 << pmd->run.log2_ring_size); j++) {
> @@ -734,8 +739,8 @@ memif_connect(struct rte_eth_dev *dev)
> MIF_LOG(ERR, "Wrong ring");
> return -1;
> }
> - ring->head = 0;
> - ring->tail = 0;
> + __atomic_store_n(&ring->head, 0,
> __ATOMIC_RELAXED);
> + __atomic_store_n(&ring->tail, 0,
> __ATOMIC_RELAXED);
> mq->last_head = 0;
> mq->last_tail = 0;
> /* enable polling mode */
> @@ -750,8 +755,8 @@ memif_connect(struct rte_eth_dev *dev)
> MIF_LOG(ERR, "Wrong ring");
> return -1;
> }
> - ring->head = 0;
> - ring->tail = 0;
> + __atomic_store_n(&ring->head, 0,
> __ATOMIC_RELAXED);
> + __atomic_store_n(&ring->tail, 0,
> __ATOMIC_RELAXED);
> mq->last_head = 0;
> mq->last_tail = 0;
> /* enable polling mode */
> --
> 2.7.4
next prev parent reply other threads:[~2019-08-26 11:03 UTC|newest]
Thread overview: 9+ messages / expand[flat|nested] mbox.gz Atom feed top
2019-08-26 11:00 Phil Yang
2019-08-26 11:03 ` Phil Yang (Arm Technology China) [this message]
2019-10-08 11:05 ` Jakub Grajciar -X (jgrajcia - PANTHEON TECHNOLOGIES at Cisco)
2019-10-09 2:10 ` Phil Yang (Arm Technology China)
2019-10-09 11:17 ` Jakub Grajciar -X (jgrajcia - PANTHEON TECHNOLOGIES at Cisco)
2019-10-14 8:59 ` Ferruh Yigit
2019-10-09 2:04 ` [dpdk-dev] [PATCH v2] " Phil Yang
2019-10-09 11:14 ` Jakub Grajciar -X (jgrajcia - PANTHEON TECHNOLOGIES at Cisco)
2019-10-10 14:04 ` David Marchand
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=VE1PR08MB464091C924770F9E1757675DE9A10@VE1PR08MB4640.eurprd08.prod.outlook.com \
--to=phil.yang@arm.com \
--cc=Gavin.Hu@arm.com \
--cc=Honnappa.Nagarahalli@arm.com \
--cc=damarion@cisco.com \
--cc=dev@dpdk.org \
--cc=jerinj@marvell.com \
--cc=jgrajcia@cisco.com \
--cc=nd@arm.com \
--cc=thomas@monjalon.net \
/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).