[PATCH can-next 0/3] can: rx-offload: performance improvements + function be be called from threaded IRQ

[PATCH can-next 0/3] can: rx-offload: performance improvements + function be be called from threaded IRQ

[Marc Kleine-Budde]

Hello,

adding a skb to skb_queue in rx_offoad requires to take a lock. By
adding an unlocked skb queue that is appended at the end of the ISR
only one lock needs to be taken.

Further the infamous "Non-RCU local softirq work is pending" error is
fixed by introducing a new can_rx_offload_threaded_irq_finish()
function to be used from threaded IRQ handlers.

regards,
Marc

[PATCH can-next 1/3] can: rx-offload: add skb queue for use during ISR

[Marc Kleine-Budde]

Adding a skb to the skb_queue in rx-offload requires to take a lock.

This commit avoids this by adding an unlocked skb queue that is
appended at the end of the ISR. Having one lock at the end of the ISR
should be OK as the HW is empty, not about to overflow.

Tested-by: Oleksij Rempel <o.rempel@pengutronix.de>
Co-developed-by: Kurt Van Dijck <dev.kurt@vandijck-laurijssen.be>
Signed-off-by: Kurt Van Dijck <dev.kurt@vandijck-laurijssen.be>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
---
 drivers/net/can/dev/rx-offload.c              | 67 +++++++++----------
 drivers/net/can/flexcan.c                     |  3 +
 drivers/net/can/m_can/m_can.c                 |  3 +
 .../net/can/spi/mcp251xfd/mcp251xfd-core.c    |  6 +-
 drivers/net/can/ti_hecc.c                     |  2 +
 include/linux/can/rx-offload.h                |  2 +
 6 files changed, 48 insertions(+), 35 deletions(-)

diff --git a/drivers/net/can/dev/rx-offload.c b/drivers/net/can/dev/rx-offload.c
index ab2c1543786c..d0bdb6db3a57 100644
--- a/drivers/net/can/dev/rx-offload.c
+++ b/drivers/net/can/dev/rx-offload.c
@@ -1,7 +1,7 @@
 // SPDX-License-Identifier: GPL-2.0-only
 /* Copyright (c) 2014      Protonic Holland,
  *                         David Jander
- * Copyright (C) 2014-2017 Pengutronix,
+ * Copyright (C) 2014-2021 Pengutronix,
  *                         Marc Kleine-Budde <kernel@pengutronix.de>
  */
 
@@ -174,10 +174,8 @@ can_rx_offload_offload_one(struct can_rx_offload *offload, unsigned int n)
 int can_rx_offload_irq_offload_timestamp(struct can_rx_offload *offload,
 					 u64 pending)
 {
-	struct sk_buff_head skb_queue;
 	unsigned int i;
-
-	__skb_queue_head_init(&skb_queue);
+	int received = 0;
 
 	for (i = offload->mb_first;
 	     can_rx_offload_le(offload, i, offload->mb_last);
@@ -191,26 +189,12 @@ int can_rx_offload_irq_offload_timestamp(struct can_rx_offload *offload,
 		if (IS_ERR_OR_NULL(skb))
 			continue;
 
-		__skb_queue_add_sort(&skb_queue, skb, can_rx_offload_compare);
-	}
-
-	if (!skb_queue_empty(&skb_queue)) {
-		unsigned long flags;
-		u32 queue_len;
-
-		spin_lock_irqsave(&offload->skb_queue.lock, flags);
-		skb_queue_splice_tail(&skb_queue, &offload->skb_queue);
-		spin_unlock_irqrestore(&offload->skb_queue.lock, flags);
-
-		queue_len = skb_queue_len(&offload->skb_queue);
-		if (queue_len > offload->skb_queue_len_max / 8)
-			netdev_dbg(offload->dev, "%s: queue_len=%d\n",
-				   __func__, queue_len);
-
-		can_rx_offload_schedule(offload);
+		__skb_queue_add_sort(&offload->skb_irq_queue, skb,
+				     can_rx_offload_compare);
+		received++;
 	}
 
-	return skb_queue_len(&skb_queue);
+	return received;
 }
 EXPORT_SYMBOL_GPL(can_rx_offload_irq_offload_timestamp);
 
@@ -226,13 +210,10 @@ int can_rx_offload_irq_offload_fifo(struct can_rx_offload *offload)
 		if (!skb)
 			break;
 
-		skb_queue_tail(&offload->skb_queue, skb);
+		__skb_queue_tail(&offload->skb_irq_queue, skb);
 		received++;
 	}
 
-	if (received)
-		can_rx_offload_schedule(offload);
-
 	return received;
 }
 EXPORT_SYMBOL_GPL(can_rx_offload_irq_offload_fifo);
@@ -241,7 +222,6 @@ int can_rx_offload_queue_sorted(struct can_rx_offload *offload,
 				struct sk_buff *skb, u32 timestamp)
 {
 	struct can_rx_offload_cb *cb;
-	unsigned long flags;
 
 	if (skb_queue_len(&offload->skb_queue) >
 	    offload->skb_queue_len_max) {
@@ -252,11 +232,8 @@ int can_rx_offload_queue_sorted(struct can_rx_offload *offload,
 	cb = can_rx_offload_get_cb(skb);
 	cb->timestamp = timestamp;
 
-	spin_lock_irqsave(&offload->skb_queue.lock, flags);
-	__skb_queue_add_sort(&offload->skb_queue, skb, can_rx_offload_compare);
-	spin_unlock_irqrestore(&offload->skb_queue.lock, flags);
-
-	can_rx_offload_schedule(offload);
+	__skb_queue_add_sort(&offload->skb_irq_queue, skb,
+			     can_rx_offload_compare);
 
 	return 0;
 }
@@ -295,13 +272,33 @@ int can_rx_offload_queue_tail(struct can_rx_offload *offload,
 		return -ENOBUFS;
 	}
 
-	skb_queue_tail(&offload->skb_queue, skb);
-	can_rx_offload_schedule(offload);
+	__skb_queue_tail(&offload->skb_irq_queue, skb);
 
 	return 0;
 }
 EXPORT_SYMBOL_GPL(can_rx_offload_queue_tail);
 
+void can_rx_offload_irq_finish(struct can_rx_offload *offload)
+{
+	unsigned long flags;
+	int queue_len;
+
+	if (skb_queue_empty_lockless(&offload->skb_irq_queue))
+		return;
+
+	spin_lock_irqsave(&offload->skb_queue.lock, flags);
+	skb_queue_splice_tail_init(&offload->skb_irq_queue, &offload->skb_queue);
+	spin_unlock_irqrestore(&offload->skb_queue.lock, flags);
+
+	queue_len = skb_queue_len(&offload->skb_queue);
+	if (queue_len > offload->skb_queue_len_max / 8)
+		netdev_dbg(offload->dev, "%s: queue_len=%d\n",
+			   __func__, queue_len);
+
+	can_rx_offload_schedule(offload);
+}
+EXPORT_SYMBOL_GPL(can_rx_offload_irq_finish);
+
 static int can_rx_offload_init_queue(struct net_device *dev,
 				     struct can_rx_offload *offload,
 				     unsigned int weight)
@@ -312,6 +309,7 @@ static int can_rx_offload_init_queue(struct net_device *dev,
 	offload->skb_queue_len_max = 2 << fls(weight);
 	offload->skb_queue_len_max *= 4;
 	skb_queue_head_init(&offload->skb_queue);
+	__skb_queue_head_init(&offload->skb_irq_queue);
 
 	netif_napi_add(dev, &offload->napi, can_rx_offload_napi_poll, weight);
 
@@ -373,5 +371,6 @@ void can_rx_offload_del(struct can_rx_offload *offload)
 {
 	netif_napi_del(&offload->napi);
 	skb_queue_purge(&offload->skb_queue);
+	__skb_queue_purge(&offload->skb_irq_queue);
 }
 EXPORT_SYMBOL_GPL(can_rx_offload_del);
diff --git a/drivers/net/can/flexcan.c b/drivers/net/can/flexcan.c
index 57f3635ad8d7..d9dcf6a8412b 100644
--- a/drivers/net/can/flexcan.c
+++ b/drivers/net/can/flexcan.c
@@ -1198,6 +1198,9 @@ static irqreturn_t flexcan_irq(int irq, void *dev_id)
 		}
 	}
 
+	if (handled)
+		can_rx_offload_irq_finish(&priv->offload);
+
 	return handled;
 }
 
diff --git a/drivers/net/can/m_can/m_can.c b/drivers/net/can/m_can/m_can.c
index bba2a449ac70..18461982f7a1 100644
--- a/drivers/net/can/m_can/m_can.c
+++ b/drivers/net/can/m_can/m_can.c
@@ -1058,6 +1058,9 @@ static irqreturn_t m_can_isr(int irq, void *dev_id)
 		}
 	}
 
+	if (cdev->is_peripheral)
+		can_rx_offload_irq_finish(&cdev->offload);
+
 	return IRQ_HANDLED;
 }
 
diff --git a/drivers/net/can/spi/mcp251xfd/mcp251xfd-core.c b/drivers/net/can/spi/mcp251xfd/mcp251xfd-core.c
index 47c3f408a799..f3b267ec22e0 100644
--- a/drivers/net/can/spi/mcp251xfd/mcp251xfd-core.c
+++ b/drivers/net/can/spi/mcp251xfd/mcp251xfd-core.c
@@ -2195,8 +2195,10 @@ static irqreturn_t mcp251xfd_irq(int irq, void *dev_id)
 			FIELD_GET(MCP251XFD_REG_INT_IE_MASK,
 				  priv->regs_status.intf);
 
-		if (!(intf_pending))
+		if (!(intf_pending)) {
+			can_rx_offload_irq_finish(&priv->offload);
 			return handled;
+		}
 
 		/* Some interrupts must be ACKed in the
 		 * MCP251XFD_REG_INT register.
@@ -2296,6 +2298,8 @@ static irqreturn_t mcp251xfd_irq(int irq, void *dev_id)
 	} while (1);
 
  out_fail:
+	can_rx_offload_irq_finish(&priv->offload);
+
 	netdev_err(priv->ndev, "IRQ handler returned %d (intf=0x%08x).\n",
 		   err, priv->regs_status.intf);
 	mcp251xfd_dump(priv);
diff --git a/drivers/net/can/ti_hecc.c b/drivers/net/can/ti_hecc.c
index 73245d8836a9..353062ead98f 100644
--- a/drivers/net/can/ti_hecc.c
+++ b/drivers/net/can/ti_hecc.c
@@ -786,6 +786,8 @@ static irqreturn_t ti_hecc_interrupt(int irq, void *dev_id)
 		int_status = hecc_read(priv, HECC_CANGIF0);
 	}
 
+	can_rx_offload_irq_finish(&priv->offload);
+
 	return IRQ_HANDLED;
 }
 
diff --git a/include/linux/can/rx-offload.h b/include/linux/can/rx-offload.h
index 40882df7105e..d71c938e17d0 100644
--- a/include/linux/can/rx-offload.h
+++ b/include/linux/can/rx-offload.h
@@ -20,6 +20,7 @@ struct can_rx_offload {
 					bool drop);
 
 	struct sk_buff_head skb_queue;
+	struct sk_buff_head skb_irq_queue;
 	u32 skb_queue_len_max;
 
 	unsigned int mb_first;
@@ -48,6 +49,7 @@ unsigned int can_rx_offload_get_echo_skb(struct can_rx_offload *offload,
 					 unsigned int *frame_len_ptr);
 int can_rx_offload_queue_tail(struct can_rx_offload *offload,
 			      struct sk_buff *skb);
+void can_rx_offload_irq_finish(struct can_rx_offload *offload);
 void can_rx_offload_del(struct can_rx_offload *offload);
 void can_rx_offload_enable(struct can_rx_offload *offload);
 
-- 
2.30.2

[PATCH can-next 2/3] can: rx-offload: can_rx_offload_irq_finish(): directly call napi_schedule()

[Marc Kleine-Budde]

Instead of calling can_rx_offload_schedule() call napi_schedule()
directly. As this was the last use of can_rx_offload_schedule() remove
this helper function.

Tested-by: Oleksij Rempel <o.rempel@pengutronix.de>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
---
 drivers/net/can/dev/rx-offload.c | 2 +-
 include/linux/can/rx-offload.h   | 5 -----
 2 files changed, 1 insertion(+), 6 deletions(-)

diff --git a/drivers/net/can/dev/rx-offload.c b/drivers/net/can/dev/rx-offload.c
index d0bdb6db3a57..82ade3aa5c13 100644
--- a/drivers/net/can/dev/rx-offload.c
+++ b/drivers/net/can/dev/rx-offload.c
@@ -295,7 +295,7 @@ void can_rx_offload_irq_finish(struct can_rx_offload *offload)
 		netdev_dbg(offload->dev, "%s: queue_len=%d\n",
 			   __func__, queue_len);
 
-	can_rx_offload_schedule(offload);
+	napi_schedule(&offload->napi);
 }
 EXPORT_SYMBOL_GPL(can_rx_offload_irq_finish);
 
diff --git a/include/linux/can/rx-offload.h b/include/linux/can/rx-offload.h
index d71c938e17d0..516f64df0ebc 100644
--- a/include/linux/can/rx-offload.h
+++ b/include/linux/can/rx-offload.h
@@ -53,11 +53,6 @@ void can_rx_offload_irq_finish(struct can_rx_offload *offload);
 void can_rx_offload_del(struct can_rx_offload *offload);
 void can_rx_offload_enable(struct can_rx_offload *offload);
 
-static inline void can_rx_offload_schedule(struct can_rx_offload *offload)
-{
-	napi_schedule(&offload->napi);
-}
-
 static inline void can_rx_offload_disable(struct can_rx_offload *offload)
 {
 	napi_disable(&offload->napi);
-- 
2.30.2

[PATCH can-next 3/3] can: rx-offload: can_rx_offload_threaded_irq_finish(): add new function to be called from threaded interrupt

[Marc Kleine-Budde]

After reading all CAN frames from the controller in the IRQ handler
and storing them into a skb_queue, the driver calls napi_schedule().
In the napi poll function the skb from the skb_queue are then pushed
into the networking stack.

However if napi_schedule() is called from a threaded IRQ handler this
triggers the following error:

| NOHZ tick-stop error: Non-RCU local softirq work is pending, handler #08!!!

To avoid this, create a new rx-offload
function (can_rx_offload_threaded_irq_finish()) with a call to
local_bh_disable()/local_bh_enable() around the napi_schedule() call.

Convert all drivers that call can_rx_offload_irq_finish() from
threaded IRQ context to can_rx_offload_threaded_irq_finish().

Suggested-by: Daniel Glöckner <dg@emlix.com>
Tested-by: Oleksij Rempel <o.rempel@pengutronix.de>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
---
 drivers/net/can/dev/rx-offload.c              | 23 +++++++++++++++++++
 drivers/net/can/m_can/m_can.c                 |  2 +-
 .../net/can/spi/mcp251xfd/mcp251xfd-core.c    |  4 ++--
 include/linux/can/rx-offload.h                |  1 +
 4 files changed, 27 insertions(+), 3 deletions(-)

diff --git a/drivers/net/can/dev/rx-offload.c b/drivers/net/can/dev/rx-offload.c
index 82ade3aa5c13..37b0cc65237b 100644
--- a/drivers/net/can/dev/rx-offload.c
+++ b/drivers/net/can/dev/rx-offload.c
@@ -299,6 +299,29 @@ void can_rx_offload_irq_finish(struct can_rx_offload *offload)
 }
 EXPORT_SYMBOL_GPL(can_rx_offload_irq_finish);
 
+void can_rx_offload_threaded_irq_finish(struct can_rx_offload *offload)
+{
+	unsigned long flags;
+	int queue_len;
+
+	if (skb_queue_empty_lockless(&offload->skb_irq_queue))
+		return;
+
+	spin_lock_irqsave(&offload->skb_queue.lock, flags);
+	skb_queue_splice_tail_init(&offload->skb_irq_queue, &offload->skb_queue);
+	spin_unlock_irqrestore(&offload->skb_queue.lock, flags);
+
+	queue_len = skb_queue_len(&offload->skb_queue);
+	if (queue_len > offload->skb_queue_len_max / 8)
+		netdev_dbg(offload->dev, "%s: queue_len=%d\n",
+			   __func__, queue_len);
+
+	local_bh_disable();
+	napi_schedule(&offload->napi);
+	local_bh_enable();
+}
+EXPORT_SYMBOL_GPL(can_rx_offload_threaded_irq_finish);
+
 static int can_rx_offload_init_queue(struct net_device *dev,
 				     struct can_rx_offload *offload,
 				     unsigned int weight)
diff --git a/drivers/net/can/m_can/m_can.c b/drivers/net/can/m_can/m_can.c
index 18461982f7a1..317cdc98c539 100644
--- a/drivers/net/can/m_can/m_can.c
+++ b/drivers/net/can/m_can/m_can.c
@@ -1059,7 +1059,7 @@ static irqreturn_t m_can_isr(int irq, void *dev_id)
 	}
 
 	if (cdev->is_peripheral)
-		can_rx_offload_irq_finish(&cdev->offload);
+		can_rx_offload_threaded_irq_finish(&cdev->offload);
 
 	return IRQ_HANDLED;
 }
diff --git a/drivers/net/can/spi/mcp251xfd/mcp251xfd-core.c b/drivers/net/can/spi/mcp251xfd/mcp251xfd-core.c
index f3b267ec22e0..6962ab2749df 100644
--- a/drivers/net/can/spi/mcp251xfd/mcp251xfd-core.c
+++ b/drivers/net/can/spi/mcp251xfd/mcp251xfd-core.c
@@ -2196,7 +2196,7 @@ static irqreturn_t mcp251xfd_irq(int irq, void *dev_id)
 				  priv->regs_status.intf);
 
 		if (!(intf_pending)) {
-			can_rx_offload_irq_finish(&priv->offload);
+			can_rx_offload_threaded_irq_finish(&priv->offload);
 			return handled;
 		}
 
@@ -2298,7 +2298,7 @@ static irqreturn_t mcp251xfd_irq(int irq, void *dev_id)
 	} while (1);
 
  out_fail:
-	can_rx_offload_irq_finish(&priv->offload);
+	can_rx_offload_threaded_irq_finish(&priv->offload);
 
 	netdev_err(priv->ndev, "IRQ handler returned %d (intf=0x%08x).\n",
 		   err, priv->regs_status.intf);
diff --git a/include/linux/can/rx-offload.h b/include/linux/can/rx-offload.h
index 516f64df0ebc..c11477620403 100644
--- a/include/linux/can/rx-offload.h
+++ b/include/linux/can/rx-offload.h
@@ -50,6 +50,7 @@ unsigned int can_rx_offload_get_echo_skb(struct can_rx_offload *offload,
 int can_rx_offload_queue_tail(struct can_rx_offload *offload,
 			      struct sk_buff *skb);
 void can_rx_offload_irq_finish(struct can_rx_offload *offload);
+void can_rx_offload_threaded_irq_finish(struct can_rx_offload *offload);
 void can_rx_offload_del(struct can_rx_offload *offload);
 void can_rx_offload_enable(struct can_rx_offload *offload);
 
-- 
2.30.2