All of lore.kernel.org
 help / color / mirror / Atom feed
From: Roger Quadros <rogerq@kernel.org>
To: davem@davemloft.net, edumazet@google.com, kuba@kernel.org,
	pabeni@redhat.com, shuah@kernel.org, vladimir.oltean@nxp.com
Cc: s-vadapalli@ti.com, r-gunasekaran@ti.com, vigneshr@ti.com,
	srk@ti.com, horms@kernel.org, p-varis@ti.com,
	netdev@vger.kernel.org, linux-kernel@vger.kernel.org,
	linux-kselftest@vger.kernel.org, rogerq@kernel.org
Subject: [PATCH net-next v11 10/10] net: ethernet: ti: am65-cpsw: add sw tx/rx irq coalescing based on hrtimers
Date: Tue, 19 Dec 2023 12:58:05 +0200	[thread overview]
Message-ID: <20231219105805.80617-11-rogerq@kernel.org> (raw)
In-Reply-To: <20231219105805.80617-1-rogerq@kernel.org>

From: Grygorii Strashko <grygorii.strashko@ti.com>

Add SW IRQ coalescing based on hrtimers for TX and RX data path which
can be enabled by ethtool commands:

- RX coalescing
  ethtool -C eth1 rx-usecs 50

- TX coalescing can be enabled per TX queue

  - by default enables coalesing for TX0
  ethtool -C eth1 tx-usecs 50
  - configure TX0
  ethtool -Q eth0 queue_mask 1 --coalesce tx-usecs 100
  - configure TX1
  ethtool -Q eth0 queue_mask 2 --coalesce tx-usecs 100
  - configure TX0 and TX1
  ethtool -Q eth0 queue_mask 3 --coalesce tx-usecs 100 --coalesce tx-usecs 100

  show configuration for TX0 and TX1:
  ethtool -Q eth0 queue_mask 3 --show-coalesce

Comparing to gro_flush_timeout and napi_defer_hard_irqs, this patch
allows to enable IRQ coalesing for RX path separately.

Signed-off-by: Grygorii Strashko <grygorii.strashko@ti.com>
Signed-off-by: Roger Quadros <rogerq@kernel.org>
---
 drivers/net/ethernet/ti/am65-cpsw-ethtool.c | 79 +++++++++++++++++++++
 drivers/net/ethernet/ti/am65-cpsw-nuss.c    | 59 ++++++++++++---
 drivers/net/ethernet/ti/am65-cpsw-nuss.h    |  4 ++
 3 files changed, 134 insertions(+), 8 deletions(-)

Changelog:

v11: no change
v10: no change
v9: no change
v8: no change
v7: no change
v6: no change
v5: initial commit

diff --git a/drivers/net/ethernet/ti/am65-cpsw-ethtool.c b/drivers/net/ethernet/ti/am65-cpsw-ethtool.c
index 0a6134cdf39f..35fceba01ea4 100644
--- a/drivers/net/ethernet/ti/am65-cpsw-ethtool.c
+++ b/drivers/net/ethernet/ti/am65-cpsw-ethtool.c
@@ -904,6 +904,80 @@ static void am65_cpsw_get_mm_stats(struct net_device *ndev,
 	s->MACMergeHoldCount = readl(base + AM65_CPSW_STATN_IET_TX_HOLD);
 }
 
+static int am65_cpsw_get_coalesce(struct net_device *ndev, struct ethtool_coalesce *coal,
+				  struct kernel_ethtool_coalesce *kernel_coal,
+				  struct netlink_ext_ack *extack)
+{
+	struct am65_cpsw_common *common = am65_ndev_to_common(ndev);
+	struct am65_cpsw_tx_chn *tx_chn;
+
+	tx_chn = &common->tx_chns[0];
+
+	coal->rx_coalesce_usecs = common->rx_pace_timeout / 1000;
+	coal->tx_coalesce_usecs = tx_chn->tx_pace_timeout / 1000;
+
+	return 0;
+}
+
+static int am65_cpsw_get_per_queue_coalesce(struct net_device *ndev, u32 queue,
+					    struct ethtool_coalesce *coal)
+{
+	struct am65_cpsw_common *common = am65_ndev_to_common(ndev);
+	struct am65_cpsw_tx_chn *tx_chn;
+
+	if (queue >= AM65_CPSW_MAX_TX_QUEUES)
+		return -EINVAL;
+
+	tx_chn = &common->tx_chns[queue];
+
+	coal->tx_coalesce_usecs = tx_chn->tx_pace_timeout / 1000;
+
+	return 0;
+}
+
+static int am65_cpsw_set_coalesce(struct net_device *ndev, struct ethtool_coalesce *coal,
+				  struct kernel_ethtool_coalesce *kernel_coal,
+				  struct netlink_ext_ack *extack)
+{
+	struct am65_cpsw_common *common = am65_ndev_to_common(ndev);
+	struct am65_cpsw_tx_chn *tx_chn;
+
+	tx_chn = &common->tx_chns[0];
+
+	if (coal->rx_coalesce_usecs && coal->rx_coalesce_usecs < 20)
+		return -EINVAL;
+
+	if (coal->tx_coalesce_usecs && coal->tx_coalesce_usecs < 20)
+		return -EINVAL;
+
+	common->rx_pace_timeout = coal->rx_coalesce_usecs * 1000;
+	tx_chn->tx_pace_timeout = coal->tx_coalesce_usecs * 1000;
+
+	return 0;
+}
+
+static int am65_cpsw_set_per_queue_coalesce(struct net_device *ndev, u32 queue,
+					    struct ethtool_coalesce *coal)
+{
+	struct am65_cpsw_common *common = am65_ndev_to_common(ndev);
+	struct am65_cpsw_tx_chn *tx_chn;
+
+	if (queue >= AM65_CPSW_MAX_TX_QUEUES)
+		return -EINVAL;
+
+	tx_chn = &common->tx_chns[queue];
+
+	if (coal->tx_coalesce_usecs && coal->tx_coalesce_usecs < 20) {
+		dev_info(common->dev, "defaulting to min value of 20us for tx-usecs for tx-%u\n",
+			 queue);
+		coal->tx_coalesce_usecs = 20;
+	}
+
+	tx_chn->tx_pace_timeout = coal->tx_coalesce_usecs * 1000;
+
+	return 0;
+}
+
 const struct ethtool_ops am65_cpsw_ethtool_ops_slave = {
 	.begin			= am65_cpsw_ethtool_op_begin,
 	.complete		= am65_cpsw_ethtool_op_complete,
@@ -922,6 +996,11 @@ const struct ethtool_ops am65_cpsw_ethtool_ops_slave = {
 	.get_ts_info		= am65_cpsw_get_ethtool_ts_info,
 	.get_priv_flags		= am65_cpsw_get_ethtool_priv_flags,
 	.set_priv_flags		= am65_cpsw_set_ethtool_priv_flags,
+	.supported_coalesce_params = ETHTOOL_COALESCE_USECS,
+	.get_coalesce           = am65_cpsw_get_coalesce,
+	.set_coalesce           = am65_cpsw_set_coalesce,
+	.get_per_queue_coalesce = am65_cpsw_get_per_queue_coalesce,
+	.set_per_queue_coalesce = am65_cpsw_set_per_queue_coalesce,
 
 	.get_link		= ethtool_op_get_link,
 	.get_link_ksettings	= am65_cpsw_get_link_ksettings,
diff --git a/drivers/net/ethernet/ti/am65-cpsw-nuss.c b/drivers/net/ethernet/ti/am65-cpsw-nuss.c
index 41e0046a52d5..faa0561e988e 100644
--- a/drivers/net/ethernet/ti/am65-cpsw-nuss.c
+++ b/drivers/net/ethernet/ti/am65-cpsw-nuss.c
@@ -596,8 +596,10 @@ static int am65_cpsw_nuss_common_stop(struct am65_cpsw_common *common)
 					msecs_to_jiffies(1000));
 	if (!i)
 		dev_err(common->dev, "tx timeout\n");
-	for (i = 0; i < common->tx_ch_num; i++)
+	for (i = 0; i < common->tx_ch_num; i++) {
 		napi_disable(&common->tx_chns[i].napi_tx);
+		hrtimer_cancel(&common->tx_chns[i].tx_hrtimer);
+	}
 
 	for (i = 0; i < common->tx_ch_num; i++) {
 		k3_udma_glue_reset_tx_chn(common->tx_chns[i].tx_chn,
@@ -616,6 +618,7 @@ static int am65_cpsw_nuss_common_stop(struct am65_cpsw_common *common)
 	}
 
 	napi_disable(&common->napi_rx);
+	hrtimer_cancel(&common->rx_hrtimer);
 
 	for (i = 0; i < AM65_CPSW_MAX_RX_FLOWS; i++)
 		k3_udma_glue_reset_rx_chn(common->rx_chns.rx_chn, i,
@@ -885,6 +888,15 @@ static int am65_cpsw_nuss_rx_packets(struct am65_cpsw_common *common,
 	return ret;
 }
 
+static enum hrtimer_restart am65_cpsw_nuss_rx_timer_callback(struct hrtimer *timer)
+{
+	struct am65_cpsw_common *common =
+			container_of(timer, struct am65_cpsw_common, rx_hrtimer);
+
+	enable_irq(common->rx_chns.irq);
+	return HRTIMER_NORESTART;
+}
+
 static int am65_cpsw_nuss_rx_poll(struct napi_struct *napi_rx, int budget)
 {
 	struct am65_cpsw_common *common = am65_cpsw_napi_to_common(napi_rx);
@@ -912,7 +924,13 @@ static int am65_cpsw_nuss_rx_poll(struct napi_struct *napi_rx, int budget)
 	if (num_rx < budget && napi_complete_done(napi_rx, num_rx)) {
 		if (common->rx_irq_disabled) {
 			common->rx_irq_disabled = false;
-			enable_irq(common->rx_chns.irq);
+			if (unlikely(common->rx_pace_timeout)) {
+				hrtimer_start(&common->rx_hrtimer,
+					      ns_to_ktime(common->rx_pace_timeout),
+					      HRTIMER_MODE_REL_PINNED);
+			} else {
+				enable_irq(common->rx_chns.irq);
+			}
 		}
 	}
 
@@ -968,7 +986,7 @@ static void am65_cpsw_nuss_tx_wake(struct am65_cpsw_tx_chn *tx_chn, struct net_d
 }
 
 static int am65_cpsw_nuss_tx_compl_packets(struct am65_cpsw_common *common,
-					   int chn, unsigned int budget)
+					   int chn, unsigned int budget, bool *tdown)
 {
 	struct device *dev = common->dev;
 	struct am65_cpsw_tx_chn *tx_chn;
@@ -991,6 +1009,7 @@ static int am65_cpsw_nuss_tx_compl_packets(struct am65_cpsw_common *common,
 		if (cppi5_desc_is_tdcm(desc_dma)) {
 			if (atomic_dec_and_test(&common->tdown_cnt))
 				complete(&common->tdown_complete);
+			*tdown = true;
 			break;
 		}
 
@@ -1013,7 +1032,7 @@ static int am65_cpsw_nuss_tx_compl_packets(struct am65_cpsw_common *common,
 }
 
 static int am65_cpsw_nuss_tx_compl_packets_2g(struct am65_cpsw_common *common,
-					      int chn, unsigned int budget)
+					      int chn, unsigned int budget, bool *tdown)
 {
 	struct device *dev = common->dev;
 	struct am65_cpsw_tx_chn *tx_chn;
@@ -1034,6 +1053,7 @@ static int am65_cpsw_nuss_tx_compl_packets_2g(struct am65_cpsw_common *common,
 		if (cppi5_desc_is_tdcm(desc_dma)) {
 			if (atomic_dec_and_test(&common->tdown_cnt))
 				complete(&common->tdown_complete);
+			*tdown = true;
 			break;
 		}
 
@@ -1059,21 +1079,40 @@ static int am65_cpsw_nuss_tx_compl_packets_2g(struct am65_cpsw_common *common,
 	return num_tx;
 }
 
+static enum hrtimer_restart am65_cpsw_nuss_tx_timer_callback(struct hrtimer *timer)
+{
+	struct am65_cpsw_tx_chn *tx_chns =
+			container_of(timer, struct am65_cpsw_tx_chn, tx_hrtimer);
+
+	enable_irq(tx_chns->irq);
+	return HRTIMER_NORESTART;
+}
+
 static int am65_cpsw_nuss_tx_poll(struct napi_struct *napi_tx, int budget)
 {
 	struct am65_cpsw_tx_chn *tx_chn = am65_cpsw_napi_to_tx_chn(napi_tx);
+	bool tdown = false;
 	int num_tx;
 
 	if (AM65_CPSW_IS_CPSW2G(tx_chn->common))
-		num_tx = am65_cpsw_nuss_tx_compl_packets_2g(tx_chn->common, tx_chn->id, budget);
+		num_tx = am65_cpsw_nuss_tx_compl_packets_2g(tx_chn->common, tx_chn->id,
+							    budget, &tdown);
 	else
-		num_tx = am65_cpsw_nuss_tx_compl_packets(tx_chn->common, tx_chn->id, budget);
+		num_tx = am65_cpsw_nuss_tx_compl_packets(tx_chn->common,
+							 tx_chn->id, budget, &tdown);
 
 	if (num_tx >= budget)
 		return budget;
 
-	if (napi_complete_done(napi_tx, num_tx))
-		enable_irq(tx_chn->irq);
+	if (napi_complete_done(napi_tx, num_tx)) {
+		if (unlikely(tx_chn->tx_pace_timeout && !tdown)) {
+			hrtimer_start(&tx_chn->tx_hrtimer,
+				      ns_to_ktime(tx_chn->tx_pace_timeout),
+				      HRTIMER_MODE_REL_PINNED);
+		} else {
+			enable_irq(tx_chn->irq);
+		}
+	}
 
 	return 0;
 }
@@ -1705,6 +1744,8 @@ static int am65_cpsw_nuss_ndev_add_tx_napi(struct am65_cpsw_common *common)
 
 		netif_napi_add_tx(common->dma_ndev, &tx_chn->napi_tx,
 				  am65_cpsw_nuss_tx_poll);
+		hrtimer_init(&tx_chn->tx_hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
+		tx_chn->tx_hrtimer.function = &am65_cpsw_nuss_tx_timer_callback;
 
 		ret = devm_request_irq(dev, tx_chn->irq,
 				       am65_cpsw_nuss_tx_irq,
@@ -1930,6 +1971,8 @@ static int am65_cpsw_nuss_init_rx_chns(struct am65_cpsw_common *common)
 
 	netif_napi_add(common->dma_ndev, &common->napi_rx,
 		       am65_cpsw_nuss_rx_poll);
+	hrtimer_init(&common->rx_hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
+	common->rx_hrtimer.function = &am65_cpsw_nuss_rx_timer_callback;
 
 	ret = devm_request_irq(dev, rx_chn->irq,
 			       am65_cpsw_nuss_rx_irq,
diff --git a/drivers/net/ethernet/ti/am65-cpsw-nuss.h b/drivers/net/ethernet/ti/am65-cpsw-nuss.h
index 1e4a045057fc..7da0492dc091 100644
--- a/drivers/net/ethernet/ti/am65-cpsw-nuss.h
+++ b/drivers/net/ethernet/ti/am65-cpsw-nuss.h
@@ -75,6 +75,8 @@ struct am65_cpsw_tx_chn {
 	struct k3_cppi_desc_pool *desc_pool;
 	struct k3_udma_glue_tx_channel *tx_chn;
 	spinlock_t lock; /* protect TX rings in multi-port mode */
+	struct hrtimer tx_hrtimer;
+	unsigned long tx_pace_timeout;
 	int irq;
 	u32 id;
 	u32 descs_num;
@@ -138,6 +140,8 @@ struct am65_cpsw_common {
 	struct napi_struct	napi_rx;
 
 	bool			rx_irq_disabled;
+	struct hrtimer		rx_hrtimer;
+	unsigned long		rx_pace_timeout;
 
 	u32			nuss_ver;
 	u32			cpsw_ver;
-- 
2.34.1


  parent reply	other threads:[~2023-12-19 10:58 UTC|newest]

Thread overview: 14+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2023-12-19 10:57 [PATCH net-next v11 00/10] net: ethernet: am65-cpsw: Add mqprio, frame preemption & coalescing Roger Quadros
2023-12-19 10:57 ` [PATCH net-next v11 01/10] selftests: forwarding: ethtool_mm: support devices with higher rx-min-frag-size Roger Quadros
2023-12-21 15:25   ` Vladimir Oltean
2023-12-19 10:57 ` [PATCH net-next v11 02/10] selftests: forwarding: ethtool_mm: fall back to aggregate if device does not report pMAC stats Roger Quadros
2023-12-21 15:25   ` Vladimir Oltean
2023-12-19 10:57 ` [PATCH net-next v11 03/10] net: ethernet: am65-cpsw: Build am65-cpsw-qos only if required Roger Quadros
2023-12-19 10:57 ` [PATCH net-next v11 04/10] net: ethernet: am65-cpsw: Rename TI_AM65_CPSW_TAS to TI_AM65_CPSW_QOS Roger Quadros
2023-12-19 10:58 ` [PATCH net-next v11 05/10] net: ethernet: am65-cpsw: cleanup TAPRIO handling Roger Quadros
2023-12-19 10:58 ` [PATCH net-next v11 06/10] net: ethernet: ti: am65-cpsw: Move code to avoid forward declaration Roger Quadros
2023-12-19 10:58 ` [PATCH net-next v11 07/10] net: ethernet: am65-cpsw: Move register definitions to header file Roger Quadros
2023-12-19 10:58 ` [PATCH net-next v11 08/10] net: ethernet: ti: am65-cpsw: add mqprio qdisc offload in channel mode Roger Quadros
2023-12-19 10:58 ` [PATCH net-next v11 09/10] net: ethernet: ti: am65-cpsw-qos: Add Frame Preemption MAC Merge support Roger Quadros
2023-12-19 10:58 ` Roger Quadros [this message]
2023-12-23  1:10 ` [PATCH net-next v11 00/10] net: ethernet: am65-cpsw: Add mqprio, frame preemption & coalescing patchwork-bot+netdevbpf

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=20231219105805.80617-11-rogerq@kernel.org \
    --to=rogerq@kernel.org \
    --cc=davem@davemloft.net \
    --cc=edumazet@google.com \
    --cc=horms@kernel.org \
    --cc=kuba@kernel.org \
    --cc=linux-kernel@vger.kernel.org \
    --cc=linux-kselftest@vger.kernel.org \
    --cc=netdev@vger.kernel.org \
    --cc=p-varis@ti.com \
    --cc=pabeni@redhat.com \
    --cc=r-gunasekaran@ti.com \
    --cc=s-vadapalli@ti.com \
    --cc=shuah@kernel.org \
    --cc=srk@ti.com \
    --cc=vigneshr@ti.com \
    --cc=vladimir.oltean@nxp.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 an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.