patches for DPDK stable branches
 help / color / mirror / Atom feed
From: "Min Hu (Connor)" <humin29@huawei.com>
To: <dev@dpdk.org>
Cc: Huisong Li <lihuisong@huawei.com>, <stable@dpdk.org>,
	Min Hu <humin29@huawei.com>,
	Yisen Zhuang <yisen.zhuang@huawei.com>,
	Lijun Ou <oulijun@huawei.com>, Hao Chen <chenhao164@huawei.com>,
	"Wei Hu (Xavier)" <xavier.huwei@huawei.com>,
	Chunsong Feng <fengchunsong@huawei.com>,
	Ferruh Yigit <ferruh.yigit@intel.com>
Subject: [PATCH v2 2/7] net/hns3: fix MAC and queues HW statistics overflow
Date: Thu, 5 May 2022 20:27:02 +0800	[thread overview]
Message-ID: <20220505122707.61182-3-humin29@huawei.com> (raw)
In-Reply-To: <20220505122707.61182-1-humin29@huawei.com>

From: Huisong Li <lihuisong@huawei.com>

The MAC and queues statistics are 32-bit registers in hardware. If
hardware statistics are not obtained for a long time, these statistics will
be overflow. So PF and VF driver have to periodically obtain and save these
statistics. Since the periodical task and the stats API are in different
threads, we introduce a statistics lock to protect the statistics.

Fixes: 8839c5e202f3 ("net/hns3: support device stats")
Cc: stable@dpdk.org

Signed-off-by: Huisong Li <lihuisong@huawei.com>
Signed-off-by: Min Hu (Connor) <humin29@huawei.com>
---
 drivers/net/hns3/hns3_ethdev.c    |   6 +-
 drivers/net/hns3/hns3_ethdev.h    |   6 ++
 drivers/net/hns3/hns3_ethdev_vf.c |   6 +-
 drivers/net/hns3/hns3_stats.c     | 144 +++++++++++++++++++++---------
 drivers/net/hns3/hns3_stats.h     |   1 +
 5 files changed, 116 insertions(+), 47 deletions(-)

diff --git a/drivers/net/hns3/hns3_ethdev.c b/drivers/net/hns3/hns3_ethdev.c
index 5aed7046d8..1d9b19d83e 100644
--- a/drivers/net/hns3/hns3_ethdev.c
+++ b/drivers/net/hns3/hns3_ethdev.c
@@ -4364,10 +4364,12 @@ hns3_service_handler(void *param)
 	struct hns3_adapter *hns = eth_dev->data->dev_private;
 	struct hns3_hw *hw = &hns->hw;
 
-	if (!hns3_is_reset_pending(hns))
+	if (!hns3_is_reset_pending(hns)) {
 		hns3_update_linkstatus_and_event(hw, true);
-	else
+		hns3_update_hw_stats(hw);
+	} else {
 		hns3_warn(hw, "Cancel the query when reset is pending");
+	}
 
 	rte_eal_alarm_set(HNS3_SERVICE_INTERVAL, hns3_service_handler, eth_dev);
 }
diff --git a/drivers/net/hns3/hns3_ethdev.h b/drivers/net/hns3/hns3_ethdev.h
index 9a0fa09b57..56f2cdd2cd 100644
--- a/drivers/net/hns3/hns3_ethdev.h
+++ b/drivers/net/hns3/hns3_ethdev.h
@@ -503,6 +503,12 @@ struct hns3_hw {
 	uint32_t mac_stats_reg_num;
 	struct hns3_rx_missed_stats imissed_stats;
 	uint64_t oerror_stats;
+	/*
+	 * The lock is used to protect statistics update in stats APIs and
+	 * periodic task.
+	 */
+	rte_spinlock_t stats_lock;
+
 	uint32_t fw_version;
 	uint16_t pf_vf_if_version;  /* version of communication interface */
 
diff --git a/drivers/net/hns3/hns3_ethdev_vf.c b/drivers/net/hns3/hns3_ethdev_vf.c
index 9e9fdc4144..f641e0dc36 100644
--- a/drivers/net/hns3/hns3_ethdev_vf.c
+++ b/drivers/net/hns3/hns3_ethdev_vf.c
@@ -1337,10 +1337,12 @@ hns3vf_service_handler(void *param)
 	 * Before querying the link status, check whether there is a reset
 	 * pending, and if so, abandon the query.
 	 */
-	if (!hns3vf_is_reset_pending(hns))
+	if (!hns3vf_is_reset_pending(hns)) {
 		hns3vf_request_link_info(hw);
-	else
+		hns3_update_hw_stats(hw);
+	} else {
 		hns3_warn(hw, "Cancel the query when reset is pending");
+	}
 
 	rte_eal_alarm_set(HNS3VF_SERVICE_INTERVAL, hns3vf_service_handler,
 			  eth_dev);
diff --git a/drivers/net/hns3/hns3_stats.c b/drivers/net/hns3/hns3_stats.c
index e4a5dcf2f8..2799ff4432 100644
--- a/drivers/net/hns3/hns3_stats.c
+++ b/drivers/net/hns3/hns3_stats.c
@@ -584,6 +584,28 @@ hns3_update_oerror_stats(struct hns3_hw *hw, bool is_clear)
 	return 0;
 }
 
+static void
+hns3_rcb_rx_ring_stats_get(struct hns3_rx_queue *rxq,
+			   struct hns3_tqp_stats *stats)
+{
+	uint32_t cnt;
+
+	cnt = hns3_read_dev(rxq, HNS3_RING_RX_PKTNUM_RECORD_REG);
+	stats->rcb_rx_ring_pktnum_rcd += cnt;
+	stats->rcb_rx_ring_pktnum[rxq->queue_id] += cnt;
+}
+
+static void
+hns3_rcb_tx_ring_stats_get(struct hns3_tx_queue *txq,
+			   struct hns3_tqp_stats *stats)
+{
+	uint32_t cnt;
+
+	cnt = hns3_read_dev(txq, HNS3_RING_TX_PKTNUM_RECORD_REG);
+	stats->rcb_tx_ring_pktnum_rcd += cnt;
+	stats->rcb_tx_ring_pktnum[txq->queue_id] += cnt;
+}
+
 /*
  * Query tqp tx queue statistics ,opcode id: 0x0B03.
  * Query tqp rx queue statistics ,opcode id: 0x0B13.
@@ -604,16 +626,14 @@ hns3_stats_get(struct rte_eth_dev *eth_dev, struct rte_eth_stats *rte_stats)
 	struct hns3_tqp_stats *stats = &hw->tqp_stats;
 	struct hns3_rx_queue *rxq;
 	struct hns3_tx_queue *txq;
-	uint64_t cnt;
 	uint16_t i;
 	int ret;
 
 	/* Update imissed stats */
 	ret = hns3_update_imissed_stats(hw, false);
 	if (ret) {
-		hns3_err(hw, "update imissed stats failed, ret = %d",
-			 ret);
-		return ret;
+		hns3_err(hw, "update imissed stats failed, ret = %d", ret);
+		goto out;
 	}
 	rte_stats->imissed = imissed_stats->rpu_rx_drop_cnt +
 				imissed_stats->ssu_rx_drop_cnt;
@@ -624,15 +644,12 @@ hns3_stats_get(struct rte_eth_dev *eth_dev, struct rte_eth_stats *rte_stats)
 		if (rxq == NULL)
 			continue;
 
-		cnt = hns3_read_dev(rxq, HNS3_RING_RX_PKTNUM_RECORD_REG);
-		/*
-		 * Read hardware and software in adjacent positions to minimize
-		 * the timing variance.
-		 */
+		rte_spinlock_lock(&hw->stats_lock);
+		hns3_rcb_rx_ring_stats_get(rxq, stats);
+		rte_spinlock_unlock(&hw->stats_lock);
+
 		rte_stats->ierrors += rxq->err_stats.l2_errors +
 				      rxq->err_stats.pkt_len_errors;
-		stats->rcb_rx_ring_pktnum_rcd += cnt;
-		stats->rcb_rx_ring_pktnum[i] += cnt;
 		rte_stats->ibytes += rxq->basic_stats.bytes;
 	}
 
@@ -642,17 +659,16 @@ hns3_stats_get(struct rte_eth_dev *eth_dev, struct rte_eth_stats *rte_stats)
 		if (txq == NULL)
 			continue;
 
-		cnt = hns3_read_dev(txq, HNS3_RING_TX_PKTNUM_RECORD_REG);
-		stats->rcb_tx_ring_pktnum_rcd += cnt;
-		stats->rcb_tx_ring_pktnum[i] += cnt;
+		rte_spinlock_lock(&hw->stats_lock);
+		hns3_rcb_tx_ring_stats_get(txq, stats);
+		rte_spinlock_unlock(&hw->stats_lock);
 		rte_stats->obytes += txq->basic_stats.bytes;
 	}
 
 	ret = hns3_update_oerror_stats(hw, false);
 	if (ret) {
-		hns3_err(hw, "update oerror stats failed, ret = %d",
-			 ret);
-		return ret;
+		hns3_err(hw, "update oerror stats failed, ret = %d", ret);
+		goto out;
 	}
 	rte_stats->oerrors = hw->oerror_stats;
 
@@ -667,8 +683,8 @@ hns3_stats_get(struct rte_eth_dev *eth_dev, struct rte_eth_stats *rte_stats)
 	rte_stats->opackets  = stats->rcb_tx_ring_pktnum_rcd -
 		rte_stats->oerrors;
 	rte_stats->rx_nombuf = eth_dev->data->rx_mbuf_alloc_failed;
-
-	return 0;
+out:
+	return ret;
 }
 
 int
@@ -688,7 +704,7 @@ hns3_stats_reset(struct rte_eth_dev *eth_dev)
 	ret = hns3_update_imissed_stats(hw, true);
 	if (ret) {
 		hns3_err(hw, "clear imissed stats failed, ret = %d", ret);
-		return ret;
+		goto out;
 	}
 
 	/*
@@ -697,9 +713,8 @@ hns3_stats_reset(struct rte_eth_dev *eth_dev)
 	 */
 	ret = hns3_update_oerror_stats(hw, true);
 	if (ret) {
-		hns3_err(hw, "clear oerror stats failed, ret = %d",
-			 ret);
-		return ret;
+		hns3_err(hw, "clear oerror stats failed, ret = %d", ret);
+		goto out;
 	}
 
 	for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
@@ -717,6 +732,7 @@ hns3_stats_reset(struct rte_eth_dev *eth_dev)
 		if (rxq == NULL)
 			continue;
 
+		rte_spinlock_lock(&hw->stats_lock);
 		memset(&rxq->basic_stats, 0,
 				sizeof(struct hns3_rx_basic_stats));
 
@@ -724,6 +740,7 @@ hns3_stats_reset(struct rte_eth_dev *eth_dev)
 		(void)hns3_read_dev(rxq, HNS3_RING_RX_PKTNUM_RECORD_REG);
 		rxq->err_stats.pkt_len_errors = 0;
 		rxq->err_stats.l2_errors = 0;
+		rte_spinlock_unlock(&hw->stats_lock);
 	}
 
 	/* Clear all the stats of a txq in a loop to keep them synchronized */
@@ -732,16 +749,20 @@ hns3_stats_reset(struct rte_eth_dev *eth_dev)
 		if (txq == NULL)
 			continue;
 
+		rte_spinlock_lock(&hw->stats_lock);
 		memset(&txq->basic_stats, 0,
 				sizeof(struct hns3_tx_basic_stats));
 
 		/* This register is read-clear */
 		(void)hns3_read_dev(txq, HNS3_RING_TX_PKTNUM_RECORD_REG);
+		rte_spinlock_unlock(&hw->stats_lock);
 	}
 
+	rte_spinlock_lock(&hw->stats_lock);
 	hns3_tqp_stats_clear(hw);
-
-	return 0;
+	rte_spinlock_unlock(&hw->stats_lock);
+out:
+	return ret;
 }
 
 static int
@@ -908,7 +929,6 @@ hns3_rxq_basic_stats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
 	struct hns3_rx_basic_stats *rxq_stats;
 	struct hns3_rx_queue *rxq;
 	uint16_t i, j;
-	uint32_t cnt;
 	char *val;
 
 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
@@ -916,16 +936,10 @@ hns3_rxq_basic_stats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
 		if (rxq == NULL)
 			continue;
 
-		cnt = hns3_read_dev(rxq, HNS3_RING_RX_PKTNUM_RECORD_REG);
-		/*
-		 * Read hardware and software in adjacent positions to minimize
-		 * the time difference.
-		 */
+		hns3_rcb_rx_ring_stats_get(rxq, stats);
 		rxq_stats = &rxq->basic_stats;
 		rxq_stats->errors = rxq->err_stats.l2_errors +
 					rxq->err_stats.pkt_len_errors;
-		stats->rcb_rx_ring_pktnum_rcd += cnt;
-		stats->rcb_rx_ring_pktnum[i] += cnt;
 
 		/*
 		 * If HW statistics are reset by stats_reset, but a lot of
@@ -955,7 +969,6 @@ hns3_txq_basic_stats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
 	struct hns3_tx_basic_stats *txq_stats;
 	struct hns3_tx_queue *txq;
 	uint16_t i, j;
-	uint32_t cnt;
 	char *val;
 
 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
@@ -963,9 +976,7 @@ hns3_txq_basic_stats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
 		if (txq == NULL)
 			continue;
 
-		cnt = hns3_read_dev(txq, HNS3_RING_TX_PKTNUM_RECORD_REG);
-		stats->rcb_tx_ring_pktnum_rcd += cnt;
-		stats->rcb_tx_ring_pktnum[i] += cnt;
+		hns3_rcb_tx_ring_stats_get(txq, stats);
 
 		txq_stats = &txq->basic_stats;
 		txq_stats->packets = stats->rcb_tx_ring_pktnum[i];
@@ -1050,6 +1061,7 @@ hns3_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
 
 	count = 0;
 
+	rte_spinlock_lock(&hw->stats_lock);
 	hns3_tqp_basic_stats_get(dev, xstats, &count);
 
 	if (!hns->is_vf) {
@@ -1057,6 +1069,7 @@ hns3_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
 		ret = hns3_query_update_mac_stats(dev);
 		if (ret < 0) {
 			hns3_err(hw, "Update Mac stats fail : %d", ret);
+			rte_spinlock_unlock(&hw->stats_lock);
 			return ret;
 		}
 
@@ -1068,11 +1081,11 @@ hns3_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
 			count++;
 		}
 	}
+	rte_spinlock_unlock(&hw->stats_lock);
 
 	ret = hns3_update_imissed_stats(hw, false);
 	if (ret) {
-		hns3_err(hw, "update imissed stats failed, ret = %d",
-			 ret);
+		hns3_err(hw, "update imissed stats failed, ret = %d", ret);
 		return ret;
 	}
 
@@ -1101,8 +1114,10 @@ hns3_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
 		}
 	}
 
+	rte_spinlock_lock(&hw->stats_lock);
 	hns3_tqp_dfx_stats_get(dev, xstats, &count);
 	hns3_queue_stats_get(dev, xstats, &count);
+	rte_spinlock_unlock(&hw->stats_lock);
 
 	return count;
 }
@@ -1453,6 +1468,7 @@ int
 hns3_dev_xstats_reset(struct rte_eth_dev *dev)
 {
 	struct hns3_adapter *hns = dev->data->dev_private;
+	struct hns3_hw *hw = &hns->hw;
 	int ret;
 
 	/* Clear tqp stats */
@@ -1460,20 +1476,22 @@ hns3_dev_xstats_reset(struct rte_eth_dev *dev)
 	if (ret)
 		return ret;
 
+	rte_spinlock_lock(&hw->stats_lock);
 	hns3_tqp_dfx_stats_clear(dev);
 
 	/* Clear reset stats */
 	memset(&hns->hw.reset.stats, 0, sizeof(struct hns3_reset_stats));
 
 	if (hns->is_vf)
-		return 0;
+		goto out;
 
 	/* HW registers are cleared on read */
 	ret = hns3_mac_stats_reset(dev);
-	if (ret)
-		return ret;
 
-	return 0;
+out:
+	rte_spinlock_unlock(&hw->stats_lock);
+
+	return ret;
 }
 
 static int
@@ -1527,6 +1545,7 @@ hns3_stats_init(struct hns3_hw *hw)
 {
 	int ret;
 
+	rte_spinlock_init(&hw->stats_lock);
 	/* Hardware statistics of imissed registers cleared. */
 	ret = hns3_update_imissed_stats(hw, true);
 	if (ret) {
@@ -1542,3 +1561,42 @@ hns3_stats_uninit(struct hns3_hw *hw)
 {
 	hns3_tqp_stats_uninit(hw);
 }
+
+static void
+hns3_update_queues_stats(struct hns3_hw *hw)
+{
+	struct rte_eth_dev_data *data = hw->data;
+	struct hns3_rx_queue *rxq;
+	struct hns3_tx_queue *txq;
+	uint16_t i;
+
+	for (i = 0; i < data->nb_rx_queues; i++) {
+		rxq = data->rx_queues[i];
+		if (rxq != NULL)
+			hns3_rcb_rx_ring_stats_get(rxq, &hw->tqp_stats);
+	}
+
+	for (i = 0; i < data->nb_tx_queues; i++) {
+		txq = data->tx_queues[i];
+		if (txq != NULL)
+			hns3_rcb_tx_ring_stats_get(txq, &hw->tqp_stats);
+	}
+}
+
+/*
+ * Some hardware statistics registers are not 64-bit. If hardware statistics are
+ * not obtained for a long time, these statistics may be reversed. This function
+ * is used to update these hardware statistics in periodic task.
+ */
+void
+hns3_update_hw_stats(struct hns3_hw *hw)
+{
+	struct hns3_adapter *hns = HNS3_DEV_HW_TO_ADAPTER(hw);
+
+	rte_spinlock_lock(&hw->stats_lock);
+	if (!hns->is_vf)
+		hns3_update_mac_stats(hw);
+
+	hns3_update_queues_stats(hw);
+	rte_spinlock_unlock(&hw->stats_lock);
+}
diff --git a/drivers/net/hns3/hns3_stats.h b/drivers/net/hns3/hns3_stats.h
index e89dc97632..b5cd6188b4 100644
--- a/drivers/net/hns3/hns3_stats.h
+++ b/drivers/net/hns3/hns3_stats.h
@@ -164,5 +164,6 @@ int hns3_stats_reset(struct rte_eth_dev *dev);
 int hns3_stats_init(struct hns3_hw *hw);
 void hns3_stats_uninit(struct hns3_hw *hw);
 int hns3_query_mac_stats_reg_num(struct hns3_hw *hw);
+void hns3_update_hw_stats(struct hns3_hw *hw);
 
 #endif /* _HNS3_STATS_H_ */
-- 
2.33.0


  parent reply	other threads:[~2022-05-05 12:28 UTC|newest]

Thread overview: 7+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
     [not found] <20220406092240.52900-1-humin29@huawei.com>
     [not found] ` <20220505122707.61182-1-humin29@huawei.com>
2022-05-05 12:27   ` [PATCH v2 1/7] net/hns3: fix order of clearing imissed register in PF Min Hu (Connor)
2022-05-05 12:27   ` Min Hu (Connor) [this message]
2022-05-05 12:27   ` [PATCH v2 3/7] net/hns3: fix pseudo-sharing between threads Min Hu (Connor)
2022-05-05 12:27   ` [PATCH v2 4/7] net/hns3: fix more mbufs are freed when Tx done cleanup Min Hu (Connor)
2022-05-05 12:27   ` [PATCH v2 5/7] net/hns3: fix RSS disable Min Hu (Connor)
2022-05-05 12:27   ` [PATCH v2 6/7] net/hns3: fix undo rollback when update RSS hash Min Hu (Connor)
2022-05-05 12:27   ` [PATCH v2 7/7] net/hns3: remove redundant RSS tuple field Min Hu (Connor)

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=20220505122707.61182-3-humin29@huawei.com \
    --to=humin29@huawei.com \
    --cc=chenhao164@huawei.com \
    --cc=dev@dpdk.org \
    --cc=fengchunsong@huawei.com \
    --cc=ferruh.yigit@intel.com \
    --cc=lihuisong@huawei.com \
    --cc=oulijun@huawei.com \
    --cc=stable@dpdk.org \
    --cc=xavier.huwei@huawei.com \
    --cc=yisen.zhuang@huawei.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).