net: softnet_data: Make xmit per task.

Softirq is preemptible on PREEMPT_RT. Without a per-CPU lock in
local_bh_disable() there is no guarantee that only one device is
transmitting at a time.
With preemption and multiple senders it is possible that the per-CPU
`recursion' counter gets incremented by different threads and exceeds
XMIT_RECURSION_LIMIT leading to a false positive recursion alert.
The `more' member is subject to similar problems if set by one thread
for one driver and wrongly used by another driver within another thread.

Instead of adding a lock to protect the per-CPU variable it is simpler
to make xmit per-task. Sending and receiving skbs happens always
in thread context anyway.

Having a lock to protected the per-CPU counter would block/ serialize two
sending threads needlessly. It would also require a recursive lock to
ensure that the owner can increment the counter further.

Make the softnet_data.xmit a task_struct member on PREEMPT_RT. Add
needed wrapper.

Cc: Ben Segall <bsegall@google.com>
Cc: Daniel Bristot de Oliveira <bristot@redhat.com>
Cc: Dietmar Eggemann <dietmar.eggemann@arm.com>
Cc: Juri Lelli <juri.lelli@redhat.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Valentin Schneider <vschneid@redhat.com>
Cc: Vincent Guittot <vincent.guittot@linaro.org>
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Link: https://patch.msgid.link/20240620132727.660738-9-bigeasy@linutronix.de
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

5 files changed, 80 insertions, 12 deletions

diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index c83b390191d4..f6fc9066147d 100644
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -43,6 +43,7 @@
 
 #include <linux/netdev_features.h>
 #include <linux/neighbour.h>
+#include <linux/netdevice_xmit.h>
 #include <uapi/linux/netdevice.h>
 #include <uapi/linux/if_bonding.h>
 #include <uapi/linux/pkt_cls.h>
@@ -3223,13 +3224,7 @@ struct softnet_data {
 	struct sk_buff_head	xfrm_backlog;
 #endif
 	/* written and read only by owning cpu: */
-	struct {
-		u16 recursion;
-		u8  more;
-#ifdef CONFIG_NET_EGRESS
-		u8  skip_txqueue;
-#endif
-	} xmit;
+	struct netdev_xmit xmit;
 #ifdef CONFIG_RPS
 	/* input_queue_head should be written by cpu owning this struct,
 	 * and only read by other cpus. Worth using a cache line.
@@ -3257,10 +3252,18 @@ struct softnet_data {
 
 DECLARE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
 
+#ifndef CONFIG_PREEMPT_RT
 static inline int dev_recursion_level(void)
 {
 	return this_cpu_read(softnet_data.xmit.recursion);
 }
+#else
+static inline int dev_recursion_level(void)
+{
+	return current->net_xmit.recursion;
+}
+
+#endif
 
 void __netif_schedule(struct Qdisc *q);
 void netif_schedule_queue(struct netdev_queue *txq);
@@ -4872,18 +4875,35 @@ static inline ktime_t netdev_get_tstamp(struct net_device *dev,
 	return hwtstamps->hwtstamp;
 }
 
-static inline netdev_tx_t __netdev_start_xmit(const struct net_device_ops *ops,
-					      struct sk_buff *skb, struct net_device *dev,
-					      bool more)
+#ifndef CONFIG_PREEMPT_RT
+static inline void netdev_xmit_set_more(bool more)
 {
 	__this_cpu_write(softnet_data.xmit.more, more);
-	return ops->ndo_start_xmit(skb, dev);
 }
 
 static inline bool netdev_xmit_more(void)
 {
 	return __this_cpu_read(softnet_data.xmit.more);
 }
+#else
+static inline void netdev_xmit_set_more(bool more)
+{
+	current->net_xmit.more = more;
+}
+
+static inline bool netdev_xmit_more(void)
+{
+	return current->net_xmit.more;
+}
+#endif
+
+static inline netdev_tx_t __netdev_start_xmit(const struct net_device_ops *ops,
+					      struct sk_buff *skb, struct net_device *dev,
+					      bool more)
+{
+	netdev_xmit_set_more(more);
+	return ops->ndo_start_xmit(skb, dev);
+}
 
 static inline netdev_tx_t netdev_start_xmit(struct sk_buff *skb, struct net_device *dev,
 					    struct netdev_queue *txq, bool more)
diff --git a/include/linux/netdevice_xmit.h b/include/linux/netdevice_xmit.h
new file mode 100644
index 000000000000..38325e070296
--- /dev/null
+++ b/include/linux/netdevice_xmit.h
@@ -0,0 +1,13 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+#ifndef _LINUX_NETDEVICE_XMIT_H
+#define _LINUX_NETDEVICE_XMIT_H
+
+struct netdev_xmit {
+	u16 recursion;
+	u8  more;
+#ifdef CONFIG_NET_EGRESS
+	u8  skip_txqueue;
+#endif
+};
+
+#endif
diff --git a/include/linux/sched.h b/include/linux/sched.h
index 61591ac6eab6..5187486c2522 100644
--- a/include/linux/sched.h
+++ b/include/linux/sched.h
@@ -36,6 +36,7 @@
 #include <linux/signal_types.h>
 #include <linux/syscall_user_dispatch_types.h>
 #include <linux/mm_types_task.h>
+#include <linux/netdevice_xmit.h>
 #include <linux/task_io_accounting.h>
 #include <linux/posix-timers_types.h>
 #include <linux/restart_block.h>
@@ -975,7 +976,9 @@ struct task_struct {
 	/* delay due to memory thrashing */
 	unsigned                        in_thrashing:1;
 #endif
-
+#ifdef CONFIG_PREEMPT_RT
+	struct netdev_xmit		net_xmit;
+#endif
 	unsigned long			atomic_flags; /* Flags requiring atomic access. */
 
 	struct restart_block		restart_block;
diff --git a/net/core/dev.c b/net/core/dev.c
index 093d82bf0e28..95b9e4cc1767 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -3940,6 +3940,7 @@ netdev_tx_queue_mapping(struct net_device *dev, struct sk_buff *skb)
 	return netdev_get_tx_queue(dev, netdev_cap_txqueue(dev, qm));
 }
 
+#ifndef CONFIG_PREEMPT_RT
 static bool netdev_xmit_txqueue_skipped(void)
 {
 	return __this_cpu_read(softnet_data.xmit.skip_txqueue);
@@ -3950,6 +3951,19 @@ void netdev_xmit_skip_txqueue(bool skip)
 	__this_cpu_write(softnet_data.xmit.skip_txqueue, skip);
 }
 EXPORT_SYMBOL_GPL(netdev_xmit_skip_txqueue);
+
+#else
+static bool netdev_xmit_txqueue_skipped(void)
+{
+	return current->net_xmit.skip_txqueue;
+}
+
+void netdev_xmit_skip_txqueue(bool skip)
+{
+	current->net_xmit.skip_txqueue = skip;
+}
+EXPORT_SYMBOL_GPL(netdev_xmit_skip_txqueue);
+#endif
 #endif /* CONFIG_NET_EGRESS */
 
 #ifdef CONFIG_NET_XGRESS
diff --git a/net/core/dev.h b/net/core/dev.h
index 58f88d28bc99..5654325c5b71 100644
--- a/net/core/dev.h
+++ b/net/core/dev.h
@@ -150,6 +150,8 @@ struct napi_struct *napi_by_id(unsigned int napi_id);
 void kick_defer_list_purge(struct softnet_data *sd, unsigned int cpu);
 
 #define XMIT_RECURSION_LIMIT	8
+
+#ifndef CONFIG_PREEMPT_RT
 static inline bool dev_xmit_recursion(void)
 {
 	return unlikely(__this_cpu_read(softnet_data.xmit.recursion) >
@@ -165,6 +167,22 @@ static inline void dev_xmit_recursion_dec(void)
 {
 	__this_cpu_dec(softnet_data.xmit.recursion);
 }
+#else
+static inline bool dev_xmit_recursion(void)
+{
+	return unlikely(current->net_xmit.recursion > XMIT_RECURSION_LIMIT);
+}
+
+static inline void dev_xmit_recursion_inc(void)
+{
+	current->net_xmit.recursion++;
+}
+
+static inline void dev_xmit_recursion_dec(void)
+{
+	current->net_xmit.recursion--;
+}
+#endif
 
 int dev_set_hwtstamp_phylib(struct net_device *dev,
 			    struct kernel_hwtstamp_config *cfg,