Re: [PATCH v2] tracing: Remove definition of trace_*_rcuidle()

[Steven Rostedt <rostedt@goodmis.org>]

On Tue, 3 Dec 2024 17:48:33 -0800
Guenter Roeck <linux@roeck-us.net> wrote:

> Hmm. If you say so. Note that powerpc has the same or a similar problem.
> 
> [    0.142039][    T0] RCU not watching for tracepoint
> [    0.142488][    T0]
> [    0.142659][    T0] =============================
> [    0.142755][    T0] WARNING: suspicious RCU usage
> [    0.142914][    T0] 6.13.0-rc1-00058-ge75ce84aa5d3 #1 Not tainted
> [    0.143082][    T0] -----------------------------
> [    0.143178][    T0] kernel/notifier.c:586 notify_die called but RCU thinks we're quiescent!
> 
> 
> [    0.152733][    T0] RCU not watching for tracepoint
> [    0.152770][    T0]
> [    0.152995][    T0] =============================
> [    0.153092][    T0] WARNING: suspicious RCU usage
> [    0.153187][    T0] 6.13.0-rc1-00058-ge75ce84aa5d3 #1 Not tainted
> [    0.153301][    T0] -----------------------------
> [    0.153394][    T0] include/linux/rcupdate.h:850 rcu_read_lock() used illegally while idle!
> 
> [    0.165396][    T0] RCU not watching for tracepoint
> [    0.165540][    T0]
> [    0.165712][    T0] =============================
> [    0.165811][    T0] WARNING: suspicious RCU usage
> [    0.165909][    T0] 6.13.0-rc1-00058-ge75ce84aa5d3 #1 Not tainted
> [    0.166026][    T0] -----------------------------
> [    0.166122][    T0] include/linux/rcupdate.h:878 rcu_read_unlock() used illegally while idle!
> 
> and many more.

Grumble. It's just that one file. I wonder if we could just do a hack like
this?

Paul?

diff --git a/kernel/trace/trace_preemptirq.c b/kernel/trace/trace_preemptirq.c
index 5c03633316a6..58098873efa9 100644
--- a/kernel/trace/trace_preemptirq.c
+++ b/kernel/trace/trace_preemptirq.c
@@ -10,11 +10,42 @@
 #include <linux/module.h>
 #include <linux/ftrace.h>
 #include <linux/kprobes.h>
+#include <linux/hardirq.h>
 #include "trace.h"
 
 #define CREATE_TRACE_POINTS
 #include <trace/events/preemptirq.h>
 
+/*
+ * Use regular trace points on architectures that implement noinstr
+ * tooling: these calls will only happen with RCU enabled, which can
+ * use a regular tracepoint.
+ *
+ * On older architectures, RCU may not be watching in idle. In that
+ * case, wake up RCU to watch while calling the tracepoint. These
+ * aren't NMI-safe - so exclude NMI contexts:
+ */
+#ifdef CONFIG_ARCH_WANTS_NO_INSTR
+#define trace(point, args)	trace_##point(args)
+#else
+#define trace(point, args)					\
+	do {							\
+		if (trace_##point##_enabled()) {		\
+			bool exit_rcu = false;			\
+			if (in_nmi())				\
+				break;				\
+			if (!IS_ENABLED(CONFIG_TINY_RCU) &&	\
+			    is_idle_task(current)) {		\
+				ct_irq_enter();			\
+				exit_rcu = true;		\
+			}					\
+			trace_##point(args);			\
+			if (exit_rcu)				\
+				ct_irq_exit();			\
+		}						\
+	} while (0)
+#endif
+
 #ifdef CONFIG_TRACE_IRQFLAGS
 /* Per-cpu variable to prevent redundant calls when IRQs already off */
 static DEFINE_PER_CPU(int, tracing_irq_cpu);
@@ -28,7 +59,7 @@ static DEFINE_PER_CPU(int, tracing_irq_cpu);
 void trace_hardirqs_on_prepare(void)
 {
 	if (this_cpu_read(tracing_irq_cpu)) {
-		trace_irq_enable(CALLER_ADDR0, CALLER_ADDR1);
+		trace(irq_enable, TP_ARGS(CALLER_ADDR0, CALLER_ADDR1));
 		tracer_hardirqs_on(CALLER_ADDR0, CALLER_ADDR1);
 		this_cpu_write(tracing_irq_cpu, 0);
 	}
@@ -39,7 +70,7 @@ NOKPROBE_SYMBOL(trace_hardirqs_on_prepare);
 void trace_hardirqs_on(void)
 {
 	if (this_cpu_read(tracing_irq_cpu)) {
-		trace_irq_enable(CALLER_ADDR0, CALLER_ADDR1);
+		trace(irq_enable, TP_ARGS(CALLER_ADDR0, CALLER_ADDR1));
 		tracer_hardirqs_on(CALLER_ADDR0, CALLER_ADDR1);
 		this_cpu_write(tracing_irq_cpu, 0);
 	}
@@ -61,7 +92,7 @@ void trace_hardirqs_off_finish(void)
 	if (!this_cpu_read(tracing_irq_cpu)) {
 		this_cpu_write(tracing_irq_cpu, 1);
 		tracer_hardirqs_off(CALLER_ADDR0, CALLER_ADDR1);
-		trace_irq_disable(CALLER_ADDR0, CALLER_ADDR1);
+		trace(irq_disable, TP_ARGS(CALLER_ADDR0, CALLER_ADDR1));
 	}
 
 }
@@ -75,7 +106,7 @@ void trace_hardirqs_off(void)
 	if (!this_cpu_read(tracing_irq_cpu)) {
 		this_cpu_write(tracing_irq_cpu, 1);
 		tracer_hardirqs_off(CALLER_ADDR0, CALLER_ADDR1);
-		trace_irq_disable(CALLER_ADDR0, CALLER_ADDR1);
+		trace(irq_disable, TP_ARGS(CALLER_ADDR0, CALLER_ADDR1));
 	}
 }
 EXPORT_SYMBOL(trace_hardirqs_off);
@@ -86,13 +117,13 @@ NOKPROBE_SYMBOL(trace_hardirqs_off);
 
 void trace_preempt_on(unsigned long a0, unsigned long a1)
 {
-	trace_preempt_enable(a0, a1);
+	trace(preempt_enable, TP_ARGS(a0, a1));
 	tracer_preempt_on(a0, a1);
 }
 
 void trace_preempt_off(unsigned long a0, unsigned long a1)
 {
-	trace_preempt_disable(a0, a1);
+	trace(preempt_disable, TP_ARGS(a0, a1));
 	tracer_preempt_off(a0, a1);
 }
 #endif


I tested this by forcing x86 to use this code, and it appeared to work.

-- Steve