From: Tomasz Duszynski <tdu@semihalf.com>
To: dev@dpdk.org
Cc: jck@semihalf.com, mw@semihalf.com, dima@marvell.com,
nsamsono@marvell.com, Jianbo.liu@arm.com
Subject: [dpdk-dev] [PATCH 5/5] net/mrvl: keep shadow txqs inside pmd txq
Date: Thu, 11 Jan 2018 16:35:43 +0100 [thread overview]
Message-ID: <1515684943-32506-6-git-send-email-tdu@semihalf.com> (raw)
In-Reply-To: <1515684943-32506-1-git-send-email-tdu@semihalf.com>
From: Natalie Samsonov <nsamsono@marvell.com>
Change shadow queues allocation from port/core to txq/core.
Use array of shadow queues (one per lcore) for each tx queue object to
avoid data corruption when few tx queues are handled by one lcore and
buffers that were not sent yet, can be released and used for receive.
Fixes: 0ddc9b8 ("net/mrvl: add net PMD skeleton")
Signed-off-by: Natalie Samsonov <nsamsono@marvell.com>
---
drivers/net/mrvl/mrvl_ethdev.c | 47 ++++++++++++++++++++++++------------------
1 file changed, 27 insertions(+), 20 deletions(-)
diff --git a/drivers/net/mrvl/mrvl_ethdev.c b/drivers/net/mrvl/mrvl_ethdev.c
index 7ce4df3..4294c56 100644
--- a/drivers/net/mrvl/mrvl_ethdev.c
+++ b/drivers/net/mrvl/mrvl_ethdev.c
@@ -150,22 +150,17 @@ struct mrvl_txq {
int queue_id;
int port_id;
uint64_t bytes_sent;
+ struct mrvl_shadow_txq shadow_txqs[RTE_MAX_LCORE];
};
-/*
- * Every tx queue should have dedicated shadow tx queue.
- *
- * Ports assigned by DPDK might not start at zero or be continuous so
- * as a workaround define shadow queues for each possible port so that
- * we eventually fit somewhere.
- */
-struct mrvl_shadow_txq shadow_txqs[RTE_MAX_ETHPORTS][RTE_MAX_LCORE];
-
static int mrvl_lcore_first;
static int mrvl_lcore_last;
static int mrvl_dev_num;
static int mrvl_fill_bpool(struct mrvl_rxq *rxq, int num);
+static inline void mrvl_free_sent_buffers(struct pp2_ppio *ppio,
+ struct pp2_hif *hif, unsigned int core_id,
+ struct mrvl_shadow_txq *sq, int qid, int force);
static inline int
mrvl_get_bpool_size(int pp2_id, int pool_id)
@@ -594,21 +589,32 @@ mrvl_flush_rx_queues(struct rte_eth_dev *dev)
static void
mrvl_flush_tx_shadow_queues(struct rte_eth_dev *dev)
{
- int i;
+ int i, j;
+ struct mrvl_txq *txq;
RTE_LOG(INFO, PMD, "Flushing tx shadow queues\n");
- for (i = 0; i < RTE_MAX_LCORE; i++) {
- struct mrvl_shadow_txq *sq =
- &shadow_txqs[dev->data->port_id][i];
+ for (i = 0; i < dev->data->nb_tx_queues; i++) {
+ txq = (struct mrvl_txq *)dev->data->tx_queues[i];
+
+ for (j = 0; j < RTE_MAX_LCORE; j++) {
+ struct mrvl_shadow_txq *sq;
+
+ if (!hifs[j])
+ continue;
- while (sq->tail != sq->head) {
- uint64_t addr = cookie_addr_high |
+ sq = &txq->shadow_txqs[j];
+ mrvl_free_sent_buffers(txq->priv->ppio,
+ hifs[j], j, sq, txq->queue_id, 1);
+ while (sq->tail != sq->head) {
+ uint64_t addr = cookie_addr_high |
sq->ent[sq->tail].buff.cookie;
- rte_pktmbuf_free((struct rte_mbuf *)addr);
- sq->tail = (sq->tail + 1) & MRVL_PP2_TX_SHADOWQ_MASK;
+ rte_pktmbuf_free(
+ (struct rte_mbuf *)addr);
+ sq->tail = (sq->tail + 1) &
+ MRVL_PP2_TX_SHADOWQ_MASK;
+ }
+ memset(sq, 0, sizeof(*sq));
}
-
- memset(sq, 0, sizeof(*sq));
}
}
@@ -1959,7 +1965,7 @@ static uint16_t
mrvl_tx_pkt_burst(void *txq, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
{
struct mrvl_txq *q = txq;
- struct mrvl_shadow_txq *sq = &shadow_txqs[q->port_id][rte_lcore_id()];
+ struct mrvl_shadow_txq *sq;
struct pp2_hif *hif;
struct pp2_ppio_desc descs[nb_pkts];
unsigned int core_id = rte_lcore_id();
@@ -1968,6 +1974,7 @@ mrvl_tx_pkt_burst(void *txq, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
uint64_t addr;
hif = mrvl_get_hif(q->priv, core_id);
+ sq = &q->shadow_txqs[core_id];
if (unlikely(!q->priv->ppio || !hif))
return 0;
--
2.7.4
next prev parent reply other threads:[~2018-01-11 15:36 UTC|newest]
Thread overview: 10+ messages / expand[flat|nested] mbox.gz Atom feed top
2018-01-11 15:35 [dpdk-dev] [PATCH 0/5] net/mrvl: fix recently found pmd issues Tomasz Duszynski
2018-01-11 15:35 ` [dpdk-dev] [PATCH 1/5] net/mrvl: fix multiple probe issue Tomasz Duszynski
2018-01-11 15:35 ` [dpdk-dev] [PATCH 2/5] net/mrvl: fix hif objects allocation Tomasz Duszynski
2018-01-12 19:13 ` Ferruh Yigit
2018-01-15 6:53 ` Tomasz Duszynski
2018-01-11 15:35 ` [dpdk-dev] [PATCH 3/5] net/mrvl: fix oversize bpool handling Tomasz Duszynski
2018-01-11 15:35 ` [dpdk-dev] [PATCH 4/5] net/mrvl: fix shadow queue tail and size calculations Tomasz Duszynski
2018-01-11 15:35 ` Tomasz Duszynski [this message]
2018-01-12 19:13 ` [dpdk-dev] [PATCH 0/5] net/mrvl: fix recently found pmd issues Ferruh Yigit
2018-01-15 7:35 ` Tomasz Duszynski
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=1515684943-32506-6-git-send-email-tdu@semihalf.com \
--to=tdu@semihalf.com \
--cc=Jianbo.liu@arm.com \
--cc=dev@dpdk.org \
--cc=dima@marvell.com \
--cc=jck@semihalf.com \
--cc=mw@semihalf.com \
--cc=nsamsono@marvell.com \
/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).