lists.openwall.net   lists  /  announce  owl-users  owl-dev  john-users  john-dev  passwdqc-users  yescrypt  popa3d-users  /  oss-security  kernel-hardening  musl  sabotage  tlsify  passwords  /  crypt-dev  xvendor  /  Bugtraq  Full-Disclosure  linux-kernel  linux-netdev  linux-ext4  linux-hardening  linux-cve-announce  PHC 
Open Source and information security mailing list archives
 
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [thread-next>] [day] [month] [year] [list]
Message-Id: <2c4709ce46639f6abcad2f62f50b5101fbc524c5.1535625356.git.joabreu@synopsys.com>
Date:   Thu, 30 Aug 2018 11:37:45 +0100
From:   Jose Abreu <Jose.Abreu@...opsys.com>
To:     netdev@...r.kernel.org
Cc:     Jose Abreu <Jose.Abreu@...opsys.com>,
        Jerome Brunet <jbrunet@...libre.com>,
        Martin Blumenstingl <martin.blumenstingl@...glemail.com>,
        "David S. Miller" <davem@...emloft.net>,
        Joao Pinto <Joao.Pinto@...opsys.com>,
        Giuseppe Cavallaro <peppe.cavallaro@...com>,
        Alexandre Torgue <alexandre.torgue@...com>
Subject: [RFT net-next] net: stmmac: Rework coalesce timer and fix multi-queue races

[ As for now this is only for testing! ]

This follows David Miller advice and tries to fix coalesce timer in
multi-queue scenarios.

We are now using per-queue coalesce values and per-queue TX timer. This
assumes that tx_queues == rx_queues, which can not be necessarly true.
Official patch will need to have this fixed.

Coalesce timer default values was changed to 1ms and the coalesce frames
to 25.

Tested in B2B setup between XGMAC2 and GMAC5.

Signed-off-by: Jose Abreu <joabreu@...opsys.com>
Cc: Jerome Brunet <jbrunet@...libre.com>
Cc: Martin Blumenstingl <martin.blumenstingl@...glemail.com>
Cc: David S. Miller <davem@...emloft.net>
Cc: Joao Pinto <jpinto@...opsys.com>
Cc: Giuseppe Cavallaro <peppe.cavallaro@...com>
Cc: Alexandre Torgue <alexandre.torgue@...com>
---
 drivers/net/ethernet/stmicro/stmmac/common.h      |   4 +-
 drivers/net/ethernet/stmicro/stmmac/stmmac.h      |   6 +-
 drivers/net/ethernet/stmicro/stmmac/stmmac_main.c | 117 +++++++++++++---------
 3 files changed, 75 insertions(+), 52 deletions(-)

diff --git a/drivers/net/ethernet/stmicro/stmmac/common.h b/drivers/net/ethernet/stmicro/stmmac/common.h
index 1854f270ad66..b1b305f8f414 100644
--- a/drivers/net/ethernet/stmicro/stmmac/common.h
+++ b/drivers/net/ethernet/stmicro/stmmac/common.h
@@ -258,10 +258,10 @@ struct stmmac_safety_stats {
 #define MAX_DMA_RIWT		0xff
 #define MIN_DMA_RIWT		0x20
 /* Tx coalesce parameters */
-#define STMMAC_COAL_TX_TIMER	40000
+#define STMMAC_COAL_TX_TIMER	1000
 #define STMMAC_MAX_COAL_TX_TICK	100000
 #define STMMAC_TX_MAX_FRAMES	256
-#define STMMAC_TX_FRAMES	64
+#define STMMAC_TX_FRAMES	25
 
 /* Packets types */
 enum packets_types {
diff --git a/drivers/net/ethernet/stmicro/stmmac/stmmac.h b/drivers/net/ethernet/stmicro/stmmac/stmmac.h
index 76649adf8fb0..db42a5f03a5f 100644
--- a/drivers/net/ethernet/stmicro/stmmac/stmmac.h
+++ b/drivers/net/ethernet/stmicro/stmmac/stmmac.h
@@ -48,6 +48,9 @@ struct stmmac_tx_info {
 
 /* Frequently used values are kept adjacent for cache effect */
 struct stmmac_tx_queue {
+	u32 tx_count_frames;
+	int tx_timer_active;
+	struct timer_list txtimer;
 	u32 queue_index;
 	struct stmmac_priv *priv_data;
 	struct dma_extended_desc *dma_etx ____cacheline_aligned_in_smp;
@@ -109,15 +112,12 @@ struct stmmac_pps_cfg {
 
 struct stmmac_priv {
 	/* Frequently used values are kept adjacent for cache effect */
-	u32 tx_count_frames;
 	u32 tx_coal_frames;
 	u32 tx_coal_timer;
-	bool tx_timer_armed;
 
 	int tx_coalesce;
 	int hwts_tx_en;
 	bool tx_path_in_lpi_mode;
-	struct timer_list txtimer;
 	bool tso;
 
 	unsigned int dma_buf_sz;
diff --git a/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c b/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
index ff1ffb46198a..ae26a6e8608e 100644
--- a/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
+++ b/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
@@ -2034,7 +2034,6 @@ static void stmmac_dma_interrupt(struct stmmac_priv *priv)
 	u32 channels_to_check = tx_channel_count > rx_channel_count ?
 				tx_channel_count : rx_channel_count;
 	u32 chan;
-	bool poll_scheduled = false;
 	int status[max_t(u32, MTL_MAX_TX_QUEUES, MTL_MAX_RX_QUEUES)];
 
 	/* Make sure we never check beyond our status buffer. */
@@ -2058,7 +2057,6 @@ static void stmmac_dma_interrupt(struct stmmac_priv *priv)
 			if (likely(napi_schedule_prep(&rx_q->napi))) {
 				stmmac_disable_dma_irq(priv, priv->ioaddr, chan);
 				__napi_schedule(&rx_q->napi);
-				poll_scheduled = true;
 			}
 		}
 	}
@@ -2067,21 +2065,13 @@ static void stmmac_dma_interrupt(struct stmmac_priv *priv)
 	 * If we didn't schedule poll, see if any DMA channel (used by tx) has a
 	 * completed transmission, if so, call stmmac_poll (once).
 	 */
-	if (!poll_scheduled) {
-		for (chan = 0; chan < tx_channel_count; chan++) {
-			if (status[chan] & handle_tx) {
-				/* It doesn't matter what rx queue we choose
-				 * here. We use 0 since it always exists.
-				 */
-				struct stmmac_rx_queue *rx_q =
-					&priv->rx_queue[0];
+	for (chan = 0; chan < tx_channel_count; chan++) {
+		if (status[chan] & handle_tx) {
+			struct stmmac_rx_queue *rx_q = &priv->rx_queue[chan];
 
-				if (likely(napi_schedule_prep(&rx_q->napi))) {
-					stmmac_disable_dma_irq(priv,
-							priv->ioaddr, chan);
-					__napi_schedule(&rx_q->napi);
-				}
-				break;
+			if (likely(napi_schedule_prep(&rx_q->napi))) {
+				stmmac_disable_dma_irq(priv, priv->ioaddr, chan);
+				__napi_schedule(&rx_q->napi);
 			}
 		}
 	}
@@ -2241,13 +2231,18 @@ static int stmmac_init_dma_engine(struct stmmac_priv *priv)
  */
 static void stmmac_tx_timer(struct timer_list *t)
 {
-	struct stmmac_priv *priv = from_timer(priv, t, txtimer);
-	u32 tx_queues_count = priv->plat->tx_queues_to_use;
-	u32 queue;
+	struct stmmac_tx_queue *tx_q = from_timer(tx_q, t, txtimer);
+	struct stmmac_priv *priv = tx_q->priv_data;
+	struct napi_struct *napi;
+
+	napi = &priv->rx_queue[tx_q->queue_index].napi;
 
-	/* let's scan all the tx queues */
-	for (queue = 0; queue < tx_queues_count; queue++)
-		stmmac_tx_clean(priv, queue);
+	if (napi_schedule_prep(napi)) {
+		stmmac_disable_dma_irq(priv, priv->ioaddr, tx_q->queue_index);
+		__napi_schedule(napi);
+	}
+
+	tx_q->tx_timer_active = 0;
 }
 
 /**
@@ -2260,11 +2255,17 @@ static void stmmac_tx_timer(struct timer_list *t)
  */
 static void stmmac_init_tx_coalesce(struct stmmac_priv *priv)
 {
+	u32 tx_channel_count = priv->plat->tx_queues_to_use;
+	u32 chan;
+
 	priv->tx_coal_frames = STMMAC_TX_FRAMES;
 	priv->tx_coal_timer = STMMAC_COAL_TX_TIMER;
-	timer_setup(&priv->txtimer, stmmac_tx_timer, 0);
-	priv->txtimer.expires = STMMAC_COAL_TIMER(priv->tx_coal_timer);
-	add_timer(&priv->txtimer);
+
+	for (chan = 0; chan < tx_channel_count; chan++) {
+		struct stmmac_tx_queue *tx_q = &priv->tx_queue[chan];
+
+		timer_setup(&tx_q->txtimer, stmmac_tx_timer, 0);
+	}
 }
 
 static void stmmac_set_rings_length(struct stmmac_priv *priv)
@@ -2592,6 +2593,7 @@ static void stmmac_hw_teardown(struct net_device *dev)
 static int stmmac_open(struct net_device *dev)
 {
 	struct stmmac_priv *priv = netdev_priv(dev);
+	u32 chan;
 	int ret;
 
 	stmmac_check_ether_addr(priv);
@@ -2688,7 +2690,9 @@ static int stmmac_open(struct net_device *dev)
 	if (dev->phydev)
 		phy_stop(dev->phydev);
 
-	del_timer_sync(&priv->txtimer);
+	for (chan = 0; chan < priv->plat->tx_queues_to_use; chan++)
+		del_timer_sync(&priv->tx_queue[chan].txtimer);
+
 	stmmac_hw_teardown(dev);
 init_error:
 	free_dma_desc_resources(priv);
@@ -2708,6 +2712,7 @@ static int stmmac_open(struct net_device *dev)
 static int stmmac_release(struct net_device *dev)
 {
 	struct stmmac_priv *priv = netdev_priv(dev);
+	u32 chan;
 
 	if (priv->eee_enabled)
 		del_timer_sync(&priv->eee_ctrl_timer);
@@ -2722,7 +2727,8 @@ static int stmmac_release(struct net_device *dev)
 
 	stmmac_disable_all_queues(priv);
 
-	del_timer_sync(&priv->txtimer);
+	for (chan = 0; chan < priv->plat->tx_queues_to_use; chan++)
+		del_timer_sync(&priv->tx_queue[chan].txtimer);
 
 	/* Free the IRQ lines */
 	free_irq(dev->irq, dev);
@@ -2828,6 +2834,7 @@ static netdev_tx_t stmmac_tso_xmit(struct sk_buff *skb, struct net_device *dev)
 	int tmp_pay_len = 0;
 	u32 pay_len, mss;
 	u8 proto_hdr_len;
+	bool tx_ic;
 	int i;
 
 	tx_q = &priv->tx_queue[queue];
@@ -2936,12 +2943,17 @@ static netdev_tx_t stmmac_tso_xmit(struct sk_buff *skb, struct net_device *dev)
 	priv->xstats.tx_tso_nfrags += nfrags;
 
 	/* Manage tx mitigation */
-	priv->tx_count_frames += nfrags + 1;
-	if (likely(priv->tx_coal_frames > priv->tx_count_frames)) {
-		mod_timer(&priv->txtimer,
-			  STMMAC_COAL_TIMER(priv->tx_coal_timer));
-	} else {
-		priv->tx_count_frames = 0;
+	tx_q->tx_count_frames += nfrags + 1;
+	if (!priv->tx_coal_frames)
+		tx_ic = false;
+	else if ((nfrags + 1) > priv->tx_coal_frames)
+		tx_ic = true;
+	else if ((tx_q->tx_count_frames % priv->tx_coal_frames) < (nfrags + 1))
+		tx_ic = true;
+	else
+		tx_ic = false;
+
+	if (tx_ic) {
 		stmmac_set_tx_ic(priv, desc);
 		priv->xstats.tx_set_ic_bit++;
 	}
@@ -2994,6 +3006,12 @@ static netdev_tx_t stmmac_tso_xmit(struct sk_buff *skb, struct net_device *dev)
 
 	stmmac_set_tx_tail_ptr(priv, priv->ioaddr, tx_q->tx_tail_addr, queue);
 
+	if (priv->tx_coal_timer && !tx_q->tx_timer_active) {
+		tx_q->tx_timer_active = 1;
+		mod_timer(&tx_q->txtimer,
+				STMMAC_COAL_TIMER(priv->tx_coal_timer));
+	}
+
 	return NETDEV_TX_OK;
 
 dma_map_err:
@@ -3024,6 +3042,7 @@ static netdev_tx_t stmmac_xmit(struct sk_buff *skb, struct net_device *dev)
 	struct stmmac_tx_queue *tx_q;
 	unsigned int enh_desc;
 	unsigned int des;
+	bool tx_ic;
 
 	tx_q = &priv->tx_queue[queue];
 
@@ -3146,17 +3165,19 @@ static netdev_tx_t stmmac_xmit(struct sk_buff *skb, struct net_device *dev)
 	 * This approach takes care about the fragments: desc is the first
 	 * element in case of no SG.
 	 */
-	priv->tx_count_frames += nfrags + 1;
-	if (likely(priv->tx_coal_frames > priv->tx_count_frames) &&
-	    !priv->tx_timer_armed) {
-		mod_timer(&priv->txtimer,
-			  STMMAC_COAL_TIMER(priv->tx_coal_timer));
-		priv->tx_timer_armed = true;
-	} else {
-		priv->tx_count_frames = 0;
+	tx_q->tx_count_frames += nfrags + 1;
+	if (!priv->tx_coal_frames)
+		tx_ic = false;
+	else if ((nfrags + 1) > priv->tx_coal_frames)
+		tx_ic = true;
+	else if ((tx_q->tx_count_frames % priv->tx_coal_frames) < (nfrags + 1))
+		tx_ic = true;
+	else
+		tx_ic = false;
+
+	if (tx_ic) {
 		stmmac_set_tx_ic(priv, desc);
 		priv->xstats.tx_set_ic_bit++;
-		priv->tx_timer_armed = false;
 	}
 
 	skb_tx_timestamp(skb);
@@ -3204,6 +3225,12 @@ static netdev_tx_t stmmac_xmit(struct sk_buff *skb, struct net_device *dev)
 	stmmac_enable_dma_transmission(priv, priv->ioaddr);
 	stmmac_set_tx_tail_ptr(priv, priv->ioaddr, tx_q->tx_tail_addr, queue);
 
+	if (priv->tx_coal_timer && !tx_q->tx_timer_active) {
+		tx_q->tx_timer_active = 1;
+		mod_timer(&tx_q->txtimer,
+				STMMAC_COAL_TIMER(priv->tx_coal_timer));
+	}
+
 	return NETDEV_TX_OK;
 
 dma_map_err:
@@ -3522,16 +3549,12 @@ static int stmmac_poll(struct napi_struct *napi, int budget)
 	struct stmmac_rx_queue *rx_q =
 		container_of(napi, struct stmmac_rx_queue, napi);
 	struct stmmac_priv *priv = rx_q->priv_data;
-	u32 tx_count = priv->plat->tx_queues_to_use;
 	u32 chan = rx_q->queue_index;
 	int work_done = 0;
-	u32 queue;
 
 	priv->xstats.napi_poll++;
 
-	/* check all the queues */
-	for (queue = 0; queue < tx_count; queue++)
-		stmmac_tx_clean(priv, queue);
+	stmmac_tx_clean(priv, chan);
 
 	work_done = stmmac_rx(priv, budget, rx_q->queue_index);
 	if (work_done < budget) {
-- 
2.7.4


Powered by blists - more mailing lists

Powered by Openwall GNU/*/Linux Powered by OpenVZ