Re: [PATCH v4 05/27] x86: Replace ist_enter() with nmi_enter()

[Peter Zijlstra]

On Mon, Feb 24, 2020 at 05:02:31PM -0500, Steven Rostedt wrote:

> The other is for the hwlat detector that measures the time it was in an
> NMI, as NMIs appear as a hardware latency too.

Yeah,.. I hate that one. But I ended up with this patch.

And yes, I know some of those notrace annotations are strictly
unnessecary due to Makefile crap, but having them is _SO_ much easier.

---
Subject: x86,tracing: Robustify ftrace_nmi_enter()
From: Peter Zijlstra <peterz@xxxxxxxxxxxxx>
Date: Mon Feb 24 23:40:29 CET 2020

  ftrace_nmi_enter()
     trace_hwlat_callback()
       trace_clock_local()
         sched_clock()
           paravirt_sched_clock()
           native_sched_clock()

All must not be traced or kprobed, it will be called from do_debug()
before the kprobe handler.

Signed-off-by: Peter Zijlstra (Intel) <peterz@xxxxxxxxxxxxx>
---
 arch/x86/include/asm/paravirt.h |    2 +-
 arch/x86/kernel/tsc.c           |    7 +++++--
 include/linux/ftrace_irq.h      |    4 ++--
 kernel/trace/trace_clock.c      |    2 ++
 kernel/trace/trace_hwlat.c      |    4 +++-
 5 files changed, 13 insertions(+), 6 deletions(-)

--- a/arch/x86/include/asm/paravirt.h
+++ b/arch/x86/include/asm/paravirt.h
@@ -17,7 +17,7 @@
 #include <linux/cpumask.h>
 #include <asm/frame.h>
 
-static inline unsigned long long paravirt_sched_clock(void)
+static __always_inline unsigned long long paravirt_sched_clock(void)
 {
 	return PVOP_CALL0(unsigned long long, time.sched_clock);
 }
--- a/arch/x86/kernel/tsc.c
+++ b/arch/x86/kernel/tsc.c
@@ -14,6 +14,7 @@
 #include <linux/percpu.h>
 #include <linux/timex.h>
 #include <linux/static_key.h>
+#include <linux/kprobes.h>
 
 #include <asm/hpet.h>
 #include <asm/timer.h>
@@ -207,7 +208,7 @@ static void __init cyc2ns_init_secondary
 /*
  * Scheduler clock - returns current time in nanosec units.
  */
-u64 native_sched_clock(void)
+notrace u64 native_sched_clock(void)
 {
 	if (static_branch_likely(&__use_tsc)) {
 		u64 tsc_now = rdtsc();
@@ -228,6 +229,7 @@ u64 native_sched_clock(void)
 	/* No locking but a rare wrong value is not a big deal: */
 	return (jiffies_64 - INITIAL_JIFFIES) * (1000000000 / HZ);
 }
+NOKPROBE_SYMBOL(native_sched_clock);
 
 /*
  * Generate a sched_clock if you already have a TSC value.
@@ -240,10 +242,11 @@ u64 native_sched_clock_from_tsc(u64 tsc)
 /* We need to define a real function for sched_clock, to override the
    weak default version */
 #ifdef CONFIG_PARAVIRT
-unsigned long long sched_clock(void)
+notrace unsigned long long sched_clock(void)
 {
 	return paravirt_sched_clock();
 }
+NOKPROBE_SYMBOL(sched_clock);
 
 bool using_native_sched_clock(void)
 {
--- a/include/linux/ftrace_irq.h
+++ b/include/linux/ftrace_irq.h
@@ -7,7 +7,7 @@ extern bool trace_hwlat_callback_enabled
 extern void trace_hwlat_callback(bool enter);
 #endif
 
-static inline void ftrace_nmi_enter(void)
+static __always_inline void ftrace_nmi_enter(void)
 {
 #ifdef CONFIG_HWLAT_TRACER
 	if (trace_hwlat_callback_enabled)
@@ -15,7 +15,7 @@ static inline void ftrace_nmi_enter(void
 #endif
 }
 
-static inline void ftrace_nmi_exit(void)
+static __always_inline void ftrace_nmi_exit(void)
 {
 #ifdef CONFIG_HWLAT_TRACER
 	if (trace_hwlat_callback_enabled)
--- a/kernel/trace/trace_clock.c
+++ b/kernel/trace/trace_clock.c
@@ -22,6 +22,7 @@
 #include <linux/sched/clock.h>
 #include <linux/ktime.h>
 #include <linux/trace_clock.h>
+#include <linux/kprobes.h>
 
 /*
  * trace_clock_local(): the simplest and least coherent tracing clock.
@@ -44,6 +45,7 @@ u64 notrace trace_clock_local(void)
 
 	return clock;
 }
+NOKPROBE_SYMBOL(trace_clock_local);
 EXPORT_SYMBOL_GPL(trace_clock_local);
 
 /*
--- a/kernel/trace/trace_hwlat.c
+++ b/kernel/trace/trace_hwlat.c
@@ -43,6 +43,7 @@
 #include <linux/cpumask.h>
 #include <linux/delay.h>
 #include <linux/sched/clock.h>
+#include <linux/kprobes.h>
 #include "trace.h"
 
 static struct trace_array	*hwlat_trace;
@@ -137,7 +138,7 @@ static void trace_hwlat_sample(struct hw
 #define init_time(a, b)	(a = b)
 #define time_u64(a)	a
 
-void trace_hwlat_callback(bool enter)
+notrace void trace_hwlat_callback(bool enter)
 {
 	if (smp_processor_id() != nmi_cpu)
 		return;
@@ -156,6 +157,7 @@ void trace_hwlat_callback(bool enter)
 	if (enter)
 		nmi_count++;
 }
+NOKPROBE_SYMBOL(trace_hwlat_callback);
 
 /**
  * get_sample - sample the CPU TSC and look for likely hardware latencies