Skip to contentSkip to conversation

RTLA: An interface for osnoise/timerlat tracers

PATCH Superseded Daniel Bristot de Oliveira 3 participants 32 messages Oct 21, 2021 – Oct 25, 2021

Subsystems (4)

v4 of 9 Superseded Latest is v9 →

lore
Share & tools
lore
#0 lore raw

v4 cover RTLA: An interface for osnoise/timerlat tracers

The rtla(1) is a meta-tool that includes a set of commands that aims to analyze the real-time properties of Linux. But instead of testing Linux as a black box, rtla leverages kernel tracing capabilities to provide precise information about the properties and root causes of unexpected results.

To start, it presents an interface to the osnoise and timerlat tracers. In the future, it will also serve as home to the rtsl [1] and other latency/noise tracers.

The first five patches are a re-send of [2] that enable multiple instances for osnoise/timerlat tracers. They are required to run the -T option - to save a trace with osnoise: events for debugging.

The next seven patches are rtla, rtla osnoise, and rtla timerlat.

The following patches are the man pages for the tools.

To compile rtla on fedora you need:

  $ git clone git://git.kernel.org/pub/scm/libs/libtrace/libtraceevent.git
  $ cd libtraceevent/
  $ make
  $ sudo make install
  $ cd ..
  $ git clone git://git.kernel.org/pub/scm/libs/libtrace/libtracefs.git
  $ cd libtracefs/
  $ make
  $ sudo make install
  $ cd ..
  $ sudo dnf install python3-docutils procps-devel
  $ cd $linux/tools/tracing/rtla/
  $ make
  $ sudo make install

This tool was be discussed at the RT-MC during LPC2021.

[1] rtsl: https://github.com/bristot/rtsl/
[2] https://lore.kernel.org/lkml/[email protected]/
[3] https://youtu.be/cZUzc0U1jJ4

Changes from v3:
- Add cross-compile support (Ahmed S. Darwish)
- Move documentation to Documentation/tools/rtla (Jonathan Corbet)
- Use .rst format for documentation (Jonathan Corbet)
- Use include option from .rst to group common parts of the documentation
- Makefile (main and doc) cleanups
Changes from v2:
- Fix the miss conception of the "size" for kernel histograms (Steven/Tom)
- Change the --skip-zeros to --with-zeros option as the former is better for humans (and the latter for computers to plot charts).
- A lot of checkpatch fixes for the user-space part.
Changes from v1:
- Fixes -t options on osnoise tracers (-t means --trace for all tools now)
- Fixes --bucket-size references (not --bucket_size)

Daniel Bristot de Oliveira (19):
trace/osnoise: Do not follow tracing_cpumask trace/osnoise: Split workload start from the tracer start trace/osnoise: Use start/stop_per_cpu_kthreads() on
osnoise_cpus_write()
trace/osnoise: Support a list of trace_array *tr trace/osnoise: Allow multiple instances of the same tracer
rtla: Real-Time Linux Analysis tool
rtla: Helper functions for rtla
rtla: Add osnoise tool
rtla/osnoise: Add osnoise top mode
rtla/osnoise: Add the hist mode
rtla: Add timerlat tool and timelart top mode rtla/timerlat: Add timerlat hist mode
rtla: Add Documentation
rtla: Add rtla osnoise man page
rtla: Add rtla osnoise top documentation
rtla: Add rtla osnoise hist documentation
rtla: Add rtla timerlat documentation
rtla: Add rtla timerlat top documentation
rtla: Add rtla timerlat hist documentation

 Documentation/tools/rtla/Makefile             |  41 +
 Documentation/tools/rtla/common_appendix.rst  |  12 +
 .../tools/rtla/common_hist_options.rst        |  23 +
 Documentation/tools/rtla/common_options.rst   |  24 +
 .../tools/rtla/common_osnoise_description.rst |   8 +
 .../tools/rtla/common_osnoise_options.rst     |  17 +
 .../rtla/common_timerlat_description.rst      |  10 +
 .../tools/rtla/common_timerlat_options.rst    |  16 +
 .../tools/rtla/common_top_options.rst         |   3 +
 .../tools/rtla/rtla-osnoise-hist.rst          |  66 ++
 Documentation/tools/rtla/rtla-osnoise-top.rst |  61 ++
 Documentation/tools/rtla/rtla-osnoise.rst     |  59 ++
 .../tools/rtla/rtla-timerlat-hist.rst         | 106 +++
 .../tools/rtla/rtla-timerlat-top.rst          | 145 +++
 Documentation/tools/rtla/rtla-timerlat.rst    |  57 ++
 Documentation/tools/rtla/rtla.rst             |  48 +
 kernel/trace/trace_osnoise.c                  | 410 ++++++---
 tools/tracing/rtla/Makefile                   | 102 +++
 tools/tracing/rtla/src/osnoise.c              | 837 ++++++++++++++++++
 tools/tracing/rtla/src/osnoise.h              |  87 ++
 tools/tracing/rtla/src/osnoise_hist.c         | 783 ++++++++++++++++
 tools/tracing/rtla/src/osnoise_top.c          | 567 ++++++++++++
 tools/tracing/rtla/src/rtla.c                 |  87 ++
 tools/tracing/rtla/src/timerlat.c             |  72 ++
 tools/tracing/rtla/src/timerlat.h             |   4 +
 tools/tracing/rtla/src/timerlat_hist.c        | 780 ++++++++++++++++
 tools/tracing/rtla/src/timerlat_top.c         | 581 ++++++++++++
 tools/tracing/rtla/src/trace.c                | 219 +++++
 tools/tracing/rtla/src/trace.h                |  27 +
 tools/tracing/rtla/src/utils.c                | 433 +++++++++
 tools/tracing/rtla/src/utils.h                |  56 ++
 31 files changed, 5620 insertions(+), 121 deletions(-)
 create mode 100644 Documentation/tools/rtla/Makefile
 create mode 100644 Documentation/tools/rtla/common_appendix.rst
 create mode 100644 Documentation/tools/rtla/common_hist_options.rst
 create mode 100644 Documentation/tools/rtla/common_options.rst
 create mode 100644 Documentation/tools/rtla/common_osnoise_description.rst
 create mode 100644 Documentation/tools/rtla/common_osnoise_options.rst
 create mode 100644 Documentation/tools/rtla/common_timerlat_description.rst
 create mode 100644 Documentation/tools/rtla/common_timerlat_options.rst
 create mode 100644 Documentation/tools/rtla/common_top_options.rst
 create mode 100644 Documentation/tools/rtla/rtla-osnoise-hist.rst
 create mode 100644 Documentation/tools/rtla/rtla-osnoise-top.rst
 create mode 100644 Documentation/tools/rtla/rtla-osnoise.rst
 create mode 100644 Documentation/tools/rtla/rtla-timerlat-hist.rst
 create mode 100644 Documentation/tools/rtla/rtla-timerlat-top.rst
 create mode 100644 Documentation/tools/rtla/rtla-timerlat.rst
 create mode 100644 Documentation/tools/rtla/rtla.rst
 create mode 100644 tools/tracing/rtla/Makefile
 create mode 100644 tools/tracing/rtla/src/osnoise.c
 create mode 100644 tools/tracing/rtla/src/osnoise.h
 create mode 100644 tools/tracing/rtla/src/osnoise_hist.c
 create mode 100644 tools/tracing/rtla/src/osnoise_top.c
 create mode 100644 tools/tracing/rtla/src/rtla.c
 create mode 100644 tools/tracing/rtla/src/timerlat.c
 create mode 100644 tools/tracing/rtla/src/timerlat.h
 create mode 100644 tools/tracing/rtla/src/timerlat_hist.c
 create mode 100644 tools/tracing/rtla/src/timerlat_top.c
 create mode 100644 tools/tracing/rtla/src/trace.c
 create mode 100644 tools/tracing/rtla/src/trace.h
 create mode 100644 tools/tracing/rtla/src/utils.c
 create mode 100644 tools/tracing/rtla/src/utils.h
Signature · 1 line
2.31.1
#1 lore raw

PATCH v4 1/19 trace/osnoise: Do not follow tracing_cpumask

In preparation to support multiple instances, decople the osnoise/timelat workload from instance specific tracing_cpumask.

Different instances can have conflicing cpumasks, making osnoise workload management needlessly complex. Osnoise already have its global cpu mask.

I also thought about using the first instance mask, but the "first" instance could be removed before the others.

This also fixes the problem that changing the tracing_mask was not re-starting the trace.

Cc Steven Rostedt <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Tom Zanussi <[email protected]>
Cc Masami Hiramatsu <[email protected]>
Cc Juri Lelli <[email protected]>
Cc Clark Williams <[email protected]>
Cc John Kacur <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Thomas Gleixner <[email protected]>
Cc Sebastian Andrzej Siewior <[email protected]>
Cc Daniel Bristot de Oliveira <[email protected]>
---
 kernel/trace/trace_osnoise.c | 25 +++++++------------------
 1 file changed, 7 insertions(+), 18 deletions(-)
kernel/trace/trace_osnoise.c +7 -18
index ce053619f289..7b1f8187764c 100644
@@ -1553,13 +1553,10 @@ static int start_per_cpu_kthreads(struct trace_array *tr)
 
 	cpus_read_lock();
 	/*
-	 * Run only on CPUs in which trace and osnoise are allowed to run.
+	 * Run only on online CPUs in which trace and osnoise are allowed to
+	 * run.
 	 */
-	cpumask_and(current_mask, tr->tracing_cpumask, &osnoise_cpumask);
-	/*
-	 * And the CPU is online.
-	 */
-	cpumask_and(current_mask, cpu_online_mask, current_mask);
+	cpumask_and(current_mask, cpu_online_mask, &osnoise_cpumask);
 
 	for_each_possible_cpu(cpu)
 		per_cpu(per_cpu_osnoise_var, cpu).kthread = NULL;
@@ -1580,10 +1577,8 @@ static int start_per_cpu_kthreads(struct trace_array *tr)
 #ifdef CONFIG_HOTPLUG_CPU
 static void osnoise_hotplug_workfn(struct work_struct *dummy)
 {
-	struct trace_array *tr = osnoise_trace;
 	unsigned int cpu = smp_processor_id();
 
-
 	mutex_lock(&trace_types_lock);
 
 	if (!osnoise_busy)
@@ -1595,9 +1590,6 @@ static void osnoise_hotplug_workfn(struct work_struct *dummy)
 	if (!cpumask_test_cpu(cpu, &osnoise_cpumask))
 		goto out_unlock;
 
-	if (!cpumask_test_cpu(cpu, tr->tracing_cpumask))
-		goto out_unlock;
-
 	start_kthread(cpu);
 
 out_unlock:
@@ -1700,13 +1692,10 @@ static void osnoise_tracer_stop(struct trace_array *tr);
  * interface to the osnoise trace. By default, it lists all  CPUs,
  * in this way, allowing osnoise threads to run on any online CPU
  * of the system. It serves to restrict the execution of osnoise to the
- * set of CPUs writing via this interface. Note that osnoise also
- * respects the "tracing_cpumask." Hence, osnoise threads will run only
- * on the set of CPUs allowed here AND on "tracing_cpumask." Why not
- * have just "tracing_cpumask?" Because the user might be interested
- * in tracing what is running on other CPUs. For instance, one might
- * run osnoise in one HT CPU while observing what is running on the
- * sibling HT CPU.
+ * set of CPUs writing via this interface. Why not use "tracing_cpumask"?
+ * Because the user might be interested in tracing what is running on
+ * other CPUs. For instance, one might run osnoise in one HT CPU
+ * while observing what is running on the sibling HT CPU.
  */
 static ssize_t
 osnoise_cpus_write(struct file *filp, const char __user *ubuf, size_t count,
Signature · 1 line
2.31.1
#4 lore raw

PATCH v4 2/19 trace/osnoise: Split workload start from the tracer start

In reply to Daniel Bristot de Oliveira

In preparation from supporting multiple trace instances, create workload start/stop specific functions.

No functional change.

Cc Steven Rostedt <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Tom Zanussi <[email protected]>
Cc Masami Hiramatsu <[email protected]>
Cc Juri Lelli <[email protected]>
Cc Clark Williams <[email protected]>
Cc John Kacur <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Thomas Gleixner <[email protected]>
Cc Sebastian Andrzej Siewior <[email protected]>
Cc Daniel Bristot de Oliveira <[email protected]>
---
 kernel/trace/trace_osnoise.c | 53 ++++++++++++++++++++++--------------
 1 file changed, 33 insertions(+), 20 deletions(-)
kernel/trace/trace_osnoise.c +33 -20
index 7b1f8187764c..0a04a9a4c85a 100644
@@ -1545,7 +1545,7 @@ static int start_kthread(unsigned int cpu)
  * This starts the kernel thread that will look for osnoise on many
  * cpus.
  */
-static int start_per_cpu_kthreads(struct trace_array *tr)
+static int start_per_cpu_kthreads(void)
 {
 	struct cpumask *current_mask = &save_cpumask;
 	int retval = 0;
@@ -1678,8 +1678,8 @@ osnoise_cpus_read(struct file *filp, char __user *ubuf, size_t count,
 	return count;
 }
 
-static void osnoise_tracer_start(struct trace_array *tr);
-static void osnoise_tracer_stop(struct trace_array *tr);
+static int osnoise_workload_start(void);
+static void osnoise_workload_stop(void);
 
 /*
  * osnoise_cpus_write - Write function for "cpus" entry
@@ -1701,7 +1701,6 @@ static ssize_t
 osnoise_cpus_write(struct file *filp, const char __user *ubuf, size_t count,
 		   loff_t *ppos)
 {
-	struct trace_array *tr = osnoise_trace;
 	cpumask_var_t osnoise_cpumask_new;
 	int running, err;
 	char buf[256];
@@ -1726,7 +1725,7 @@ osnoise_cpus_write(struct file *filp, const char __user *ubuf, size_t count,
 	mutex_lock(&trace_types_lock);
 	running = osnoise_busy;
 	if (running)
-		osnoise_tracer_stop(tr);
+		osnoise_workload_stop();
 
 	mutex_lock(&interface_lock);
 	/*
@@ -1740,7 +1739,7 @@ osnoise_cpus_write(struct file *filp, const char __user *ubuf, size_t count,
 	mutex_unlock(&interface_lock);
 
 	if (running)
-		osnoise_tracer_start(tr);
+		osnoise_workload_start();
 	mutex_unlock(&trace_types_lock);
 
 	free_cpumask_var(osnoise_cpumask_new);
@@ -1921,7 +1920,10 @@ static int osnoise_hook_events(void)
 	return -EINVAL;
 }
 
-static int __osnoise_tracer_start(struct trace_array *tr)
+/*
+ * osnoise_workload_start - start the workload and hook to events
+ */
+static int osnoise_workload_start(void)
 {
 	int retval;
 
@@ -1936,7 +1938,7 @@ static int __osnoise_tracer_start(struct trace_array *tr)
 	barrier();
 	trace_osnoise_callback_enabled = true;
 
-	retval = start_per_cpu_kthreads(tr);
+	retval = start_per_cpu_kthreads();
 	if (retval) {
 		unhook_irq_events();
 		return retval;
@@ -1947,6 +1949,26 @@ static int __osnoise_tracer_start(struct trace_array *tr)
 	return 0;
 }
 
+/*
+ * osnoise_workload_stop - stop the workload and unhook the events
+ */
+static void osnoise_workload_stop(void)
+{
+	if (!osnoise_busy)
+		return;
+
+	trace_osnoise_callback_enabled = false;
+	barrier();
+
+	stop_per_cpu_kthreads();
+
+	unhook_irq_events();
+	unhook_softirq_events();
+	unhook_thread_events();
+
+	osnoise_busy = false;
+}
+
 static void osnoise_tracer_start(struct trace_array *tr)
 {
 	int retval;
@@ -1954,7 +1976,7 @@ static void osnoise_tracer_start(struct trace_array *tr)
 	if (osnoise_busy)
 		return;
 
-	retval = __osnoise_tracer_start(tr);
+	retval = osnoise_workload_start();
 	if (retval)
 		pr_err(BANNER "Error starting osnoise tracer\n");
 
@@ -1965,16 +1987,7 @@ static void osnoise_tracer_stop(struct trace_array *tr)
 	if (!osnoise_busy)
 		return;
 
-	trace_osnoise_callback_enabled = false;
-	barrier();
-
-	stop_per_cpu_kthreads();
-
-	unhook_irq_events();
-	unhook_softirq_events();
-	unhook_thread_events();
-
-	osnoise_busy = false;
+	osnoise_workload_stop();
 }
 
 static int osnoise_tracer_init(struct trace_array *tr)
@@ -2017,7 +2030,7 @@ static void timerlat_tracer_start(struct trace_array *tr)
 
 	osnoise_data.timerlat_tracer = 1;
 
-	retval = __osnoise_tracer_start(tr);
+	retval = osnoise_workload_start();
 	if (retval)
 		goto out_err;
 
Signature · 1 line
2.31.1
#7 lore raw

PATCH v4 3/19 trace/osnoise: Use start/stop_per_cpu_kthreads() on osnoise_cpus_write()

In reply to Daniel Bristot de Oliveira

When writing a new CPU mask via osnoise/cpus, if the tracer is running, the workload is restarted to follow the new cpumask. The restart is currently done using osnoise_workload_start/stop(), which disables the workload *and* the instrumentation. However, disabling the instrumentation is not necessary.

Calling start/stop_per_cpu_kthreads() is enough to apply the new osnoise/cpus config.

No functional change.

Cc Steven Rostedt <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Tom Zanussi <[email protected]>
Cc Masami Hiramatsu <[email protected]>
Cc Juri Lelli <[email protected]>
Cc Clark Williams <[email protected]>
Cc John Kacur <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Thomas Gleixner <[email protected]>
Cc Sebastian Andrzej Siewior <[email protected]>
Cc Daniel Bristot de Oliveira <[email protected]>
---
 kernel/trace/trace_osnoise.c | 7 ++-----
 1 file changed, 2 insertions(+), 5 deletions(-)
kernel/trace/trace_osnoise.c +2 -5
index 0a04a9a4c85a..9b9e6bc889e9 100644
@@ -1678,9 +1678,6 @@ osnoise_cpus_read(struct file *filp, char __user *ubuf, size_t count,
 	return count;
 }
 
-static int osnoise_workload_start(void);
-static void osnoise_workload_stop(void);
-
 /*
  * osnoise_cpus_write - Write function for "cpus" entry
  * @filp: The active open file structure
@@ -1725,7 +1722,7 @@ osnoise_cpus_write(struct file *filp, const char __user *ubuf, size_t count,
 	mutex_lock(&trace_types_lock);
 	running = osnoise_busy;
 	if (running)
-		osnoise_workload_stop();
+		stop_per_cpu_kthreads();
 
 	mutex_lock(&interface_lock);
 	/*
@@ -1739,7 +1736,7 @@ osnoise_cpus_write(struct file *filp, const char __user *ubuf, size_t count,
 	mutex_unlock(&interface_lock);
 
 	if (running)
-		osnoise_workload_start();
+		start_per_cpu_kthreads();
 	mutex_unlock(&trace_types_lock);
 
 	free_cpumask_var(osnoise_cpumask_new);
Signature · 1 line
2.31.1
#8 lore raw

PATCH v4 4/19 trace/osnoise: Support a list of trace_array *tr

In reply to Daniel Bristot de Oliveira

osnoise/timerlat were built to run a single instance, and for this, a single variable is enough to store the current struct trace_array *tr with information about the tracing instance. This is done via the *osnoise_trace variable. A trace_array represents a trace instance.

In preparation to support multiple instances, replace the *osnoise_trace variable with an RCU protected list of instances.

The operations that refer to an instance now propagate to all elements of the list (all instances).

Also, replace the osnoise_busy variable with a check if the list has elements (busy).

No functional change is expected with this patch, i.e., only one instance is allowed yet.

Cc Steven Rostedt <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Tom Zanussi <[email protected]>
Cc Masami Hiramatsu <[email protected]>
Cc Juri Lelli <[email protected]>
Cc Clark Williams <[email protected]>
Cc John Kacur <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Thomas Gleixner <[email protected]>
Cc Sebastian Andrzej Siewior <[email protected]>
Cc Daniel Bristot de Oliveira <[email protected]>
---
 kernel/trace/trace_osnoise.c | 253 +++++++++++++++++++++++++----------
 1 file changed, 184 insertions(+), 69 deletions(-)
kernel/trace/trace_osnoise.c +184 -69
index 9b9e6bc889e9..a6d8f514bd7c 100644
@@ -38,8 +38,6 @@
 #define CREATE_TRACE_POINTS
 #include <trace/events/osnoise.h>
 
-static struct trace_array	*osnoise_trace;
-
 /*
  * Default values.
  */
@@ -50,6 +48,73 @@ static struct trace_array	*osnoise_trace;
 #define DEFAULT_TIMERLAT_PERIOD	1000			/* 1ms */
 #define DEFAULT_TIMERLAT_PRIO	95			/* FIFO 95 */
 
+/*
+ * trace_array of the enabled osnoise/timerlat instances.
+ */
+struct osnoise_instance {
+	struct list_head	list;
+	struct trace_array	*tr;
+};
+struct list_head osnoise_instances;
+
+static bool osnoise_has_registered_instances(void)
+{
+	return !!list_first_or_null_rcu(&osnoise_instances,
+					struct osnoise_instance,
+					list);
+}
+
+/*
+ * osnoise_register_instance - register a new trace instance
+ *
+ * Register a trace_array *tr in the list of instances running
+ * osnoise/timerlat tracers.
+ */
+static int osnoise_register_instance(struct trace_array *tr)
+{
+	struct osnoise_instance *inst;
+
+	inst = kmalloc(sizeof(*inst), GFP_KERNEL);
+	if (!inst)
+		return -ENOMEM;
+
+	INIT_LIST_HEAD_RCU(&inst->list);
+	inst->tr = tr;
+	list_add_tail_rcu(&inst->list, &osnoise_instances);
+
+	return 0;
+}
+
+/*
+ *  osnoise_unregister_instance - unregister a registered trace instance
+ *
+ * Remove the trace_array *tr from the list of instances running
+ * osnoise/timerlat tracers.
+ */
+static void osnoise_unregister_instance(struct trace_array *tr)
+{
+	struct osnoise_instance *inst;
+	int found = 0;
+
+	rcu_read_lock();
+	list_for_each_entry_rcu(inst, &osnoise_instances, list) {
+		if (inst->tr == tr) {
+			list_del_rcu(&inst->list);
+			found = 1;
+			break;
+		}
+	}
+	rcu_read_unlock();
+
+	if (!found) {
+		WARN("osnoise unregister did not find tr %s\n", tr->name);
+		return;
+	}
+
+	synchronize_rcu();
+	kfree(inst);
+}
+
 /*
  * NMI runtime info.
  */
@@ -248,11 +313,6 @@ static struct osnoise_data {
 #endif
 };
 
-/*
- * Boolean variable used to inform that the tracer is currently sampling.
- */
-static bool osnoise_busy;
-
 #ifdef CONFIG_PREEMPT_RT
 /*
  * Print the osnoise header info.
@@ -315,19 +375,24 @@ static void print_osnoise_headers(struct seq_file *s)
  * osnoise_taint - report an osnoise error.
  */
 #define osnoise_taint(msg) ({							\
-	struct trace_array *tr = osnoise_trace;					\
+	struct osnoise_instance *inst;						\
+	struct trace_buffer *buffer;						\
 										\
-	trace_array_printk_buf(tr->array_buffer.buffer, _THIS_IP_, msg);	\
+	rcu_read_lock();							\
+	list_for_each_entry_rcu(inst, &osnoise_instances, list) {		\
+		buffer = inst->tr->array_buffer.buffer;				\
+		trace_array_printk_buf(buffer, _THIS_IP_, msg);			\
+	}									\
+	rcu_read_unlock();							\
 	osnoise_data.tainted = true;						\
 })
 
 /*
  * Record an osnoise_sample into the tracer buffer.
  */
-static void trace_osnoise_sample(struct osnoise_sample *sample)
+static void
+__trace_osnoise_sample(struct osnoise_sample *sample, struct trace_buffer *buffer)
 {
-	struct trace_array *tr = osnoise_trace;
-	struct trace_buffer *buffer = tr->array_buffer.buffer;
 	struct trace_event_call *call = &event_osnoise;
 	struct ring_buffer_event *event;
 	struct osnoise_entry *entry;
@@ -350,6 +415,22 @@ static void trace_osnoise_sample(struct osnoise_sample *sample)
 		trace_buffer_unlock_commit_nostack(buffer, event);
 }
 
+/*
+ * Record an osnoise_sample on all osnoise instances.
+ */
+static void trace_osnoise_sample(struct osnoise_sample *sample)
+{
+	struct osnoise_instance *inst;
+	struct trace_buffer *buffer;
+
+	rcu_read_lock();
+	list_for_each_entry_rcu(inst, &osnoise_instances, list) {
+		buffer = inst->tr->array_buffer.buffer;
+		__trace_osnoise_sample(sample, buffer);
+	}
+	rcu_read_unlock();
+}
+
 #ifdef CONFIG_TIMERLAT_TRACER
 /*
  * Print the timerlat header info.
@@ -387,14 +468,10 @@ static void print_timerlat_headers(struct seq_file *s)
 }
 #endif /* CONFIG_PREEMPT_RT */
 
-/*
- * Record an timerlat_sample into the tracer buffer.
- */
-static void trace_timerlat_sample(struct timerlat_sample *sample)
+static void
+__trace_timerlat_sample(struct timerlat_sample *sample, struct trace_buffer *buffer)
 {
-	struct trace_array *tr = osnoise_trace;
 	struct trace_event_call *call = &event_osnoise;
-	struct trace_buffer *buffer = tr->array_buffer.buffer;
 	struct ring_buffer_event *event;
 	struct timerlat_entry *entry;
 
@@ -411,6 +488,22 @@ static void trace_timerlat_sample(struct timerlat_sample *sample)
 		trace_buffer_unlock_commit_nostack(buffer, event);
 }
 
+/*
+ * Record an timerlat_sample into the tracer buffer.
+ */
+static void trace_timerlat_sample(struct timerlat_sample *sample)
+{
+	struct osnoise_instance *inst;
+	struct trace_buffer *buffer;
+
+	rcu_read_lock();
+	list_for_each_entry_rcu(inst, &osnoise_instances, list) {
+		buffer = inst->tr->array_buffer.buffer;
+		__trace_timerlat_sample(sample, buffer);
+	}
+	rcu_read_unlock();
+}
+
 #ifdef CONFIG_STACKTRACE
 
 #define	MAX_CALLS	256
@@ -450,29 +543,18 @@ static void timerlat_save_stack(int skip)
 	return;
 
 }
-/*
- * timerlat_dump_stack - dump a stack trace previously saved
- *
- * Dump a saved stack trace into the trace buffer.
- */
-static void timerlat_dump_stack(void)
+
+static void
+__timerlat_dump_stack(struct trace_buffer *buffer, struct trace_stack *fstack, unsigned int size)
 {
 	struct trace_event_call *call = &event_osnoise;
-	struct trace_array *tr = osnoise_trace;
-	struct trace_buffer *buffer = tr->array_buffer.buffer;
 	struct ring_buffer_event *event;
-	struct trace_stack *fstack;
 	struct stack_entry *entry;
-	unsigned int size;
-
-	preempt_disable_notrace();
-	fstack = this_cpu_ptr(&trace_stack);
-	size = fstack->stack_size;
 
 	event = trace_buffer_lock_reserve(buffer, TRACE_STACK, sizeof(*entry) + size,
 					  tracing_gen_ctx());
 	if (!event)
-		goto out;
+		return;
 
 	entry = ring_buffer_event_data(event);
 
@@ -481,8 +563,29 @@ static void timerlat_dump_stack(void)
 
 	if (!call_filter_check_discard(call, entry, buffer, event))
 		trace_buffer_unlock_commit_nostack(buffer, event);
+}
 
-out:
+/*
+ * timerlat_dump_stack - dump a stack trace previously saved
+ */
+static void timerlat_dump_stack(void)
+{
+	struct osnoise_instance *inst;
+	struct trace_buffer *buffer;
+	struct trace_stack *fstack;
+	unsigned int size;
+
+	preempt_disable_notrace();
+	fstack = this_cpu_ptr(&trace_stack);
+	size = fstack->stack_size;
+
+	rcu_read_lock();
+	list_for_each_entry_rcu(inst, &osnoise_instances, list) {
+		buffer = inst->tr->array_buffer.buffer;
+		__timerlat_dump_stack(buffer, fstack, size);
+
+	}
+	rcu_read_unlock();
 	preempt_enable_notrace();
 }
 #else
@@ -1077,12 +1180,37 @@ diff_osn_sample_stats(struct osnoise_variables *osn_var, struct osnoise_sample *
  */
 static __always_inline void osnoise_stop_tracing(void)
 {
-	struct trace_array *tr = osnoise_trace;
+	struct osnoise_instance *inst;
+	struct trace_array *tr;
+
+	rcu_read_lock();
+	list_for_each_entry_rcu(inst, &osnoise_instances, list) {
+		tr = inst->tr;
+		trace_array_printk_buf(tr->array_buffer.buffer, _THIS_IP_,
+				"stop tracing hit on cpu %d\n", smp_processor_id());
+
+		tracer_tracing_off(tr);
+	}
+	rcu_read_unlock();
+}
 
-	trace_array_printk_buf(tr->array_buffer.buffer, _THIS_IP_,
-			"stop tracing hit on cpu %d\n", smp_processor_id());
+/*
+ * notify_new_max_latency - Notify a new max latency via fsnotify interface.
+ */
+void notify_new_max_latency(u64 latency)
+{
+	struct osnoise_instance *inst;
+	struct trace_array *tr;
 
-	tracer_tracing_off(tr);
+	rcu_read_lock();
+	list_for_each_entry_rcu(inst, &osnoise_instances, list) {
+		tr = inst->tr;
+		if (tr->max_latency < latency) {
+			tr->max_latency = latency;
+			latency_fsnotify(tr);
+		}
+	}
+	rcu_read_unlock();
 }
 
 /*
@@ -1096,7 +1224,6 @@ static __always_inline void osnoise_stop_tracing(void)
 static int run_osnoise(void)
 {
 	struct osnoise_variables *osn_var = this_cpu_osn_var();
-	struct trace_array *tr = osnoise_trace;
 	u64 start, sample, last_sample;
 	u64 last_int_count, int_count;
 	s64 noise = 0, max_noise = 0;
@@ -1231,11 +1358,7 @@ static int run_osnoise(void)
 
 	trace_osnoise_sample(&s);
 
-	/* Keep a running maximum ever recorded osnoise "latency" */
-	if (max_noise > tr->max_latency) {
-		tr->max_latency = max_noise;
-		latency_fsnotify(tr);
-	}
+	notify_new_max_latency(max_noise);
 
 	if (osnoise_data.stop_tracing_total)
 		if (s.noise > osnoise_data.stop_tracing_total)
@@ -1293,7 +1416,6 @@ static int osnoise_main(void *data)
 static enum hrtimer_restart timerlat_irq(struct hrtimer *timer)
 {
 	struct osnoise_variables *osn_var = this_cpu_osn_var();
-	struct trace_array *tr = osnoise_trace;
 	struct timerlat_variables *tlat;
 	struct timerlat_sample s;
 	u64 now;
@@ -1364,11 +1486,7 @@ static enum hrtimer_restart timerlat_irq(struct hrtimer *timer)
 
 	trace_timerlat_sample(&s);
 
-	/* Keep a running maximum ever recorded os noise "latency" */
-	if (diff > tr->max_latency) {
-		tr->max_latency = diff;
-		latency_fsnotify(tr);
-	}
+	notify_new_max_latency(diff);
 
 	if (osnoise_data.stop_tracing)
 		if (time_to_us(diff) >= osnoise_data.stop_tracing)
@@ -1581,7 +1699,7 @@ static void osnoise_hotplug_workfn(struct work_struct *dummy)
 
 	mutex_lock(&trace_types_lock);
 
-	if (!osnoise_busy)
+	if (!osnoise_has_registered_instances())
 		goto out_unlock_trace;
 
 	mutex_lock(&interface_lock);
@@ -1716,11 +1834,10 @@ osnoise_cpus_write(struct file *filp, const char __user *ubuf, size_t count,
 		goto err_free;
 
 	/*
-	 * trace_types_lock is taken to avoid concurrency on start/stop
-	 * and osnoise_busy.
+	 * trace_types_lock is taken to avoid concurrency on start/stop.
 	 */
 	mutex_lock(&trace_types_lock);
-	running = osnoise_busy;
+	running = osnoise_has_registered_instances();
 	if (running)
 		stop_per_cpu_kthreads();
 
@@ -1941,8 +2058,6 @@ static int osnoise_workload_start(void)
 		return retval;
 	}
 
-	osnoise_busy = true;
-
 	return 0;
 }
 
@@ -1951,7 +2066,7 @@ static int osnoise_workload_start(void)
  */
 static void osnoise_workload_stop(void)
 {
-	if (!osnoise_busy)
+	if (osnoise_has_registered_instances())
 		return;
 
 	trace_osnoise_callback_enabled = false;
@@ -1962,28 +2077,28 @@ static void osnoise_workload_stop(void)
 	unhook_irq_events();
 	unhook_softirq_events();
 	unhook_thread_events();
-
-	osnoise_busy = false;
 }
 
 static void osnoise_tracer_start(struct trace_array *tr)
 {
 	int retval;
 
-	if (osnoise_busy)
+	if (osnoise_has_registered_instances())
 		return;
 
 	retval = osnoise_workload_start();
 	if (retval)
 		pr_err(BANNER "Error starting osnoise tracer\n");
 
+	osnoise_register_instance(tr);
 }
 
 static void osnoise_tracer_stop(struct trace_array *tr)
 {
-	if (!osnoise_busy)
+	if (!osnoise_has_registered_instances())
 		return;
 
+	osnoise_unregister_instance(tr);
 	osnoise_workload_stop();
 }
 
@@ -1991,14 +2106,12 @@ static int osnoise_tracer_init(struct trace_array *tr)
 {
 
 	/* Only allow one instance to enable this */
-	if (osnoise_busy)
+	if (osnoise_has_registered_instances())
 		return -EBUSY;
 
-	osnoise_trace = tr;
 	tr->max_latency = 0;
 
 	osnoise_tracer_start(tr);
-
 	return 0;
 }
 
@@ -2022,7 +2135,7 @@ static void timerlat_tracer_start(struct trace_array *tr)
 {
 	int retval;
 
-	if (osnoise_busy)
+	if (osnoise_has_registered_instances())
 		return;
 
 	osnoise_data.timerlat_tracer = 1;
@@ -2031,6 +2144,8 @@ static void timerlat_tracer_start(struct trace_array *tr)
 	if (retval)
 		goto out_err;
 
+	osnoise_register_instance(tr);
+
 	return;
 out_err:
 	pr_err(BANNER "Error starting timerlat tracer\n");
@@ -2040,7 +2155,7 @@ static void timerlat_tracer_stop(struct trace_array *tr)
 {
 	int cpu;
 
-	if (!osnoise_busy)
+	if (!osnoise_has_registered_instances())
 		return;
 
 	for_each_online_cpu(cpu)
@@ -2054,11 +2169,9 @@ static void timerlat_tracer_stop(struct trace_array *tr)
 static int timerlat_tracer_init(struct trace_array *tr)
 {
 	/* Only allow one instance to enable this */
-	if (osnoise_busy)
+	if (osnoise_has_registered_instances())
 		return -EBUSY;
 
-	osnoise_trace = tr;
-
 	tr->max_latency = 0;
 
 	timerlat_tracer_start(tr);
@@ -2105,6 +2218,8 @@ __init static int init_osnoise_tracer(void)
 #endif
 	osnoise_init_hotplug_support();
 
+	INIT_LIST_HEAD_RCU(&osnoise_instances);
+
 	init_tracefs();
 
 	return 0;
Signature · 1 line
2.31.1
#22 lore raw

PATCH v4 10/19 rtla/osnoise: Add the hist mode

In reply to Daniel Bristot de Oliveira

The rtla osnoise hist tool collects all osnoise:sample_threshold occurrence in a histogram, displaying the results in a user-friendly way. The tool also allows many configurations of the osnoise tracer and the collection of the tracer output.

Here is one example of the rtla osnoise hist tool output:
---------- %< ----------
[root@f34 ~]# rtla osnoise hist --bucket-size 10 --entries 100 -c 0-8 -d 1M -r 9000 -P F:1 # RTLA osnoise histogram
# Time unit is microseconds (us)
# Duration: 0 00:01:00
Index CPU-000 CPU-001 CPU-002 CPU-003 CPU-004 CPU-005 CPU-006 CPU-007 CPU-008

 0           430       434       352       455       440       463       467       436       484
 10           88        88        92       141       120       100       126       166       100
 20           19         7        12        22         8         8        13        13        16
 30            6         0         2         0         1         2         2         1         0
 50            0         0         0         0         0         0         1         0         0
 over:         0         0         0         0         0         0         0         0         0
 count:      543       529       458       618       569       573       609       616       600
 min:          0         0         0         0         0         0         0         0         0
 avg:          0         0         0         0         0         0         0         0         0

max: 30 20 30 20 30 30 50 30 20
---------- >% ----------

Running
- rtla osnoise hist --help

provides information about the available options.

Cc Steven Rostedt <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Tom Zanussi <[email protected]>
Cc Masami Hiramatsu <[email protected]>
Cc Juri Lelli <[email protected]>
Cc Clark Williams <[email protected]>
Cc John Kacur <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Thomas Gleixner <[email protected]>
Cc Sebastian Andrzej Siewior <[email protected]>
Cc Daniel Bristot de Oliveira <[email protected]>
---
 tools/tracing/rtla/src/osnoise.c      |   6 +-
 tools/tracing/rtla/src/osnoise.h      |   1 +
 tools/tracing/rtla/src/osnoise_hist.c | 783 ++++++++++++++++++++++++++
 3 files changed, 789 insertions(+), 1 deletion(-)
 create mode 100644 tools/tracing/rtla/src/osnoise_hist.c
tools/tracing/rtla/src/osnoise.c +5 -1
index b9866dfdda66..6c6358ebf66f 100644
@@ -790,7 +790,8 @@ static void osnoise_usage(void)
 		"  usage: [rtla] osnoise [MODE] ...",
 		"",
 		"  modes:",
-		"     top  - prints the summary from osnoise tracer",
+		"     top   - prints the summary from osnoise tracer",
+		"     hist  - prints a histogram of osnoise samples",
 		"",
 		"if no MODE is given, the top mode is called, passing the arguments",
 		NULL,
@@ -825,6 +826,9 @@ int osnoise_main(int argc, char *argv[])
 	} else if (strcmp(argv[1], "top") == 0) {
 		osnoise_top_main(argc-1, &argv[1]);
 		exit(0);
+	} else if (strcmp(argv[1], "hist") == 0) {
+		osnoise_hist_main(argc-1, &argv[1]);
+		exit(0);
 	}
 
 usage:
tools/tracing/rtla/src/osnoise.h +1 -0
index 4882ee275ea0..dbaad7a162a2 100644
@@ -82,5 +82,6 @@ void osnoise_destroy_tool(struct osnoise_tool *top);
 struct osnoise_tool *osnoise_init_tool(char *tool_name);
 struct osnoise_tool *osnoise_init_trace_tool(char *tracer);
 
+int osnoise_hist_main(int argc, char *argv[]);
 int osnoise_top_main(int argc, char **argv);
 int osnoise_main(int argc, char **argv);
tools/tracing/rtla/src/osnoise_hist.c new file +783 -0

783 lines not on this page · raw message

Signature · 1 line
2.31.1
#14 lore raw

PATCH v4 5/19 trace/osnoise: Allow multiple instances of the same tracer

In reply to Daniel Bristot de Oliveira

Allow more than one instance of the same tracer. The workload will start when the first trace_array (instance) is registered and will stop when the last instance is unregistered.

Osnoise and timerlat are still mutually exclusive because of the different behavior of the osnoise: tracepoints.

Cc Steven Rostedt <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Tom Zanussi <[email protected]>
Cc Masami Hiramatsu <[email protected]>
Cc Juri Lelli <[email protected]>
Cc Clark Williams <[email protected]>
Cc John Kacur <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Thomas Gleixner <[email protected]>
Cc Sebastian Andrzej Siewior <[email protected]>
Cc Daniel Bristot de Oliveira <[email protected]>
---
 kernel/trace/trace_osnoise.c | 106 ++++++++++++++++++++++++++---------
 1 file changed, 80 insertions(+), 26 deletions(-)
kernel/trace/trace_osnoise.c +80 -26
index a6d8f514bd7c..f913964f6861 100644
@@ -64,6 +64,24 @@ static bool osnoise_has_registered_instances(void)
 					list);
 }
 
+/*
+ * osnoise_instance_registered - check if a tr is already registered
+ */
+static int osnoise_instance_registered(struct trace_array *tr)
+{
+	struct osnoise_instance *inst;
+	int found = 0;
+
+	rcu_read_lock();
+	list_for_each_entry_rcu(inst, &osnoise_instances, list) {
+		if (inst->tr == tr)
+			found = 1;
+	}
+	rcu_read_unlock();
+
+	return found;
+}
+
 /*
  * osnoise_register_instance - register a new trace instance
  *
@@ -106,10 +124,8 @@ static void osnoise_unregister_instance(struct trace_array *tr)
 	}
 	rcu_read_unlock();
 
-	if (!found) {
-		WARN("osnoise unregister did not find tr %s\n", tr->name);
+	if (!found)
 		return;
-	}
 
 	synchronize_rcu();
 	kfree(inst);
@@ -1190,6 +1206,7 @@ static __always_inline void osnoise_stop_tracing(void)
 				"stop tracing hit on cpu %d\n", smp_processor_id());
 
 		tracer_tracing_off(tr);
+
 	}
 	rcu_read_unlock();
 }
@@ -2041,6 +2058,16 @@ static int osnoise_workload_start(void)
 {
 	int retval;
 
+	/*
+	 * Instances need to be registered after calling workload
+	 * start. Hence, if there is already an instance, the
+	 * workload was already registered. Otherwise, this
+	 * code is on the way to register the first instance,
+	 * and the workload will start.
+	 */
+	if (osnoise_has_registered_instances())
+		return 0;
+
 	osn_var_reset_all();
 
 	retval = osnoise_hook_events();
@@ -2066,6 +2093,13 @@ static int osnoise_workload_start(void)
  */
 static void osnoise_workload_stop(void)
 {
+	/*
+	 * Instances need to be unregistered before calling
+	 * stop. Hence, if there is a registered instance, more
+	 * than one instance is running, and the workload will not
+	 * yet stop. Otherwise, this code is on the way to disable
+	 * the last instance, and the workload can stop.
+	 */
 	if (osnoise_has_registered_instances())
 		return;
 
@@ -2083,7 +2117,11 @@ static void osnoise_tracer_start(struct trace_array *tr)
 {
 	int retval;
 
-	if (osnoise_has_registered_instances())
+	/*
+	 * If the instance is already registered, there is no need to
+	 * register it again.
+	 */
+	if (osnoise_instance_registered(tr))
 		return;
 
 	retval = osnoise_workload_start();
@@ -2095,18 +2133,17 @@ static void osnoise_tracer_start(struct trace_array *tr)
 
 static void osnoise_tracer_stop(struct trace_array *tr)
 {
-	if (!osnoise_has_registered_instances())
-		return;
-
 	osnoise_unregister_instance(tr);
 	osnoise_workload_stop();
 }
 
 static int osnoise_tracer_init(struct trace_array *tr)
 {
-
-	/* Only allow one instance to enable this */
-	if (osnoise_has_registered_instances())
+	/*
+	 * Only allow osnoise tracer if timerlat tracer is not running
+	 * already.
+	 */
+	if (osnoise_data.timerlat_tracer)
 		return -EBUSY;
 
 	tr->max_latency = 0;
@@ -2135,45 +2172,55 @@ static void timerlat_tracer_start(struct trace_array *tr)
 {
 	int retval;
 
-	if (osnoise_has_registered_instances())
+	/*
+	 * If the instance is already registered, there is no need to
+	 * register it again.
+	 */
+	if (osnoise_instance_registered(tr))
 		return;
 
-	osnoise_data.timerlat_tracer = 1;
-
 	retval = osnoise_workload_start();
 	if (retval)
-		goto out_err;
+		pr_err(BANNER "Error starting timerlat tracer\n");
 
 	osnoise_register_instance(tr);
 
 	return;
-out_err:
-	pr_err(BANNER "Error starting timerlat tracer\n");
 }
 
 static void timerlat_tracer_stop(struct trace_array *tr)
 {
 	int cpu;
 
-	if (!osnoise_has_registered_instances())
-		return;
-
-	for_each_online_cpu(cpu)
-		per_cpu(per_cpu_osnoise_var, cpu).sampling = 0;
+	osnoise_unregister_instance(tr);
 
-	osnoise_tracer_stop(tr);
+	/*
+	 * Instruct the threads to stop only if this is the last instance.
+	 */
+	if (!osnoise_has_registered_instances()) {
+		for_each_online_cpu(cpu)
+			per_cpu(per_cpu_osnoise_var, cpu).sampling = 0;
+	}
 
-	osnoise_data.timerlat_tracer = 0;
+	osnoise_workload_stop();
 }
 
 static int timerlat_tracer_init(struct trace_array *tr)
 {
-	/* Only allow one instance to enable this */
-	if (osnoise_has_registered_instances())
+	/*
+	 * Only allow timerlat tracer if osnoise tracer is not running already.
+	 */
+	if (osnoise_has_registered_instances() && !osnoise_data.timerlat_tracer)
 		return -EBUSY;
 
-	tr->max_latency = 0;
+	/*
+	 * If this is the first instance, set timerlat_tracer to block
+	 * osnoise tracer start.
+	 */
+	if (!osnoise_has_registered_instances())
+		osnoise_data.timerlat_tracer = 1;
 
+	tr->max_latency = 0;
 	timerlat_tracer_start(tr);
 
 	return 0;
@@ -2182,6 +2229,13 @@ static int timerlat_tracer_init(struct trace_array *tr)
 static void timerlat_tracer_reset(struct trace_array *tr)
 {
 	timerlat_tracer_stop(tr);
+
+	/*
+	 * If this is the last instance, reset timerlat_tracer allowing
+	 * osnoise to be started.
+	 */
+	if (!osnoise_has_registered_instances())
+		osnoise_data.timerlat_tracer = 0;
 }
 
 static struct tracer timerlat_tracer __read_mostly = {
Signature · 1 line
2.31.1
#28 lore raw

PATCH v4 16/19 rtla: Add rtla osnoise hist documentation

In reply to Daniel Bristot de Oliveira

Man page for rtla osnoise hist mode.

Cc Steven Rostedt <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Tom Zanussi <[email protected]>
Cc Masami Hiramatsu <[email protected]>
Cc Juri Lelli <[email protected]>
Cc Clark Williams <[email protected]>
Cc John Kacur <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Thomas Gleixner <[email protected]>
Cc Sebastian Andrzej Siewior <[email protected]>
Cc Daniel Bristot de Oliveira <[email protected]>
---
 .../tools/rtla/common_hist_options.rst        | 23 +++++++
 .../tools/rtla/rtla-osnoise-hist.rst          | 66 +++++++++++++++++++
 2 files changed, 89 insertions(+)
 create mode 100644 Documentation/tools/rtla/common_hist_options.rst
 create mode 100644 Documentation/tools/rtla/rtla-osnoise-hist.rst
Documentation/tools/rtla/common_hist_options.rst new file +23 -0
new file mode 100644
index 000000000000..0266cd08a6c9
@@ -0,0 +1,23 @@
+**-b**, **--bucket-size** *N*
+
+        Set the histogram bucket size (default *1*).
+
+**-e**, **--entries** *N*
+
+        Set the number of entries of the histogram (default 256).
+
+**--no-header**
+
+        Do not print header.
+
+**--no-summary**
+
+        Do not print summary.
+
+**--no-index**
+
+        Do not print index.
+
+**--with-zeros**
+
+        Print zero only entries.
Documentation/tools/rtla/rtla-osnoise-hist.rst new file +66 -0

66 lines not on this page · raw message

Signature · 1 line
2.31.1
#18 lore raw

PATCH v4 6/19 rtla: Real-Time Linux Analysis tool

In reply to Daniel Bristot de Oliveira

The rtla is a meta-tool that includes a set of commands that aims to analyze the real-time properties of Linux. But instead of testing Linux as a black box, rtla leverages kernel tracing capabilities to provide precise information about the properties and root causes of unexpected results.

rtla --help works and provide information about the available options.

This is just the "main" and the Makefile, no function yet.

Cc Steven Rostedt <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Tom Zanussi <[email protected]>
Cc Masami Hiramatsu <[email protected]>
Cc Juri Lelli <[email protected]>
Cc Clark Williams <[email protected]>
Cc John Kacur <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Thomas Gleixner <[email protected]>
Cc Sebastian Andrzej Siewior <[email protected]>
Cc Daniel Bristot de Oliveira <[email protected]>
---
 tools/tracing/rtla/Makefile   | 76 +++++++++++++++++++++++++++++++++++
 tools/tracing/rtla/src/rtla.c | 72 +++++++++++++++++++++++++++++++++
 2 files changed, 148 insertions(+)
 create mode 100644 tools/tracing/rtla/Makefile
 create mode 100644 tools/tracing/rtla/src/rtla.c
tools/tracing/rtla/Makefile new file +76 -0
new file mode 100644
index 000000000000..525e15b76156
@@ -0,0 +1,76 @@
+NAME	:=	rtla
+VERSION	:=	0.2
+
+# From libtracefs:
+# Makefiles suck: This macro sets a default value of $(2) for the
+# variable named by $(1), unless the variable has been set by
+# environment or command line. This is necessary for CC and AR
+# because make sets default values, so the simpler ?= approach
+# won't work as expected.
+define allow-override
+  $(if $(or $(findstring environment,$(origin $(1))),\
+            $(findstring command line,$(origin $(1)))),,\
+    $(eval $(1) = $(2)))
+endef
+
+# Allow setting CC and AR, or setting CROSS_COMPILE as a prefix.
+$(call allow-override,CC,$(CROSS_COMPILE)gcc)
+$(call allow-override,AR,$(CROSS_COMPILE)ar)
+$(call allow-override,STRIP,$(CROSS_COMPILE)strip)
+$(call allow-override,PKG_CONFIG,pkg-config)
+$(call allow-override,LD_SO_CONF_PATH,/etc/ld.so.conf.d/)
+$(call allow-override,LDCONFIG,ldconfig)
+
+INSTALL	=	install
+FOPTS	:=	-flto=auto -ffat-lto-objects -fexceptions -fstack-protector-strong \
+		-fasynchronous-unwind-tables -fstack-clash-protection
+WOPTS	:= 	-Wall -Werror=format-security -Wp,-D_FORTIFY_SOURCE=2 -Wp,-D_GLIBCXX_ASSERTIONS -Wno-maybe-uninitialized
+
+TRACEFS_HEADERS	:= $$($(PKG_CONFIG) --cflags libtracefs)
+
+CFLAGS	:=	-O -g -DVERSION=\"$(VERSION)\" $(FOPTS) $(MOPTS) $(WOPTS) $(TRACEFS_HEADERS)
+LDFLAGS	:=	-ggdb
+LIBS	:=	-ltracefs -ltraceevent -lpthread -lprocps
+
+SRC	:=	$(wildcard src/*.c)
+HDR	:=	$(wildcard src/*.h)
+OBJ	:=	$(SRC:.c=.o)
+DIRS	:=	src
+FILES	:=	Makefile
+CEXT	:=	bz2
+TARBALL	:=	$(NAME)-$(VERSION).tar.$(CEXT)
+TAROPTS	:=	-cvjf $(TARBALL)
+BINDIR	:=	/usr/bin
+DATADIR	:=	/usr/share
+DOCDIR	:=	$(DATADIR)/doc
+MANDIR	:=	$(DATADIR)/man
+LICDIR	:=	$(DATADIR)/licenses
+
+.PHONY:	all
+all:	rtla
+
+rtla: $(OBJ)
+	$(CC) -o rtla $(LDFLAGS) $(OBJ) $(LIBS)
+
+static: $(OBJ)
+	$(CC) -o rtla-static $(LDFLAGS) --static $(OBJ) $(LIBS) -lpthread -ldl
+
+.PHONY: install
+install:
+	$(INSTALL) -d -m 755 $(DESTDIR)$(BINDIR)
+	$(INSTALL) rtla -m 755 $(DESTDIR)$(BINDIR)
+	$(STRIP) $(DESTDIR)$(BINDIR)/rtla
+
+.PHONY: clean tarball
+clean:
+	@test ! -f rtla || rm rtla
+	@test ! -f rtla-static || rm rtla-static
+	@test ! -f src/rtla.o || rm src/rtla.o
+	@test ! -f $(TARBALL) || rm -f $(TARBALL)
+	@rm -rf *~ $(OBJ) *.tar.$(CEXT)
+
+tarball:  clean
+	rm -rf $(NAME)-$(VERSION) && mkdir $(NAME)-$(VERSION)
+	cp -r $(DIRS) $(FILES) $(NAME)-$(VERSION)
+	tar $(TAROPTS) --exclude='*~' $(NAME)-$(VERSION)
+	rm -rf $(NAME)-$(VERSION)
tools/tracing/rtla/src/rtla.c new file +72 -0

72 lines not on this page · raw message

Signature · 1 line
2.31.1
#19 lore raw

PATCH v4 7/19 rtla: Helper functions for rtla

In reply to Daniel Bristot de Oliveira

This is a set of utils and tracer helper functions. They are used by rtla mostly to parse config, display data and some trace operations that are not part of libtracefs (because they are only useful it for this case).

Cc Steven Rostedt <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Tom Zanussi <[email protected]>
Cc Masami Hiramatsu <[email protected]>
Cc Juri Lelli <[email protected]>
Cc Clark Williams <[email protected]>
Cc John Kacur <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Thomas Gleixner <[email protected]>
Cc Sebastian Andrzej Siewior <[email protected]>
Cc Daniel Bristot de Oliveira <[email protected]>
---
 tools/tracing/rtla/src/trace.c | 219 +++++++++++++++++
 tools/tracing/rtla/src/trace.h |  27 ++
 tools/tracing/rtla/src/utils.c | 433 +++++++++++++++++++++++++++++++++
 tools/tracing/rtla/src/utils.h |  56 +++++
 4 files changed, 735 insertions(+)
 create mode 100644 tools/tracing/rtla/src/trace.c
 create mode 100644 tools/tracing/rtla/src/trace.h
 create mode 100644 tools/tracing/rtla/src/utils.c
 create mode 100644 tools/tracing/rtla/src/utils.h
tools/tracing/rtla/src/trace.c new file +219 -0
new file mode 100644
index 000000000000..0f66e99fef0d
@@ -0,0 +1,219 @@
+// SPDX-License-Identifier: GPL-2.0
+#define _GNU_SOURCE
+#include <sys/sendfile.h>
+#include <tracefs.h>
+#include <signal.h>
+#include <stdlib.h>
+#include <unistd.h>
+#include <errno.h>
+
+#include "trace.h"
+#include "utils.h"
+
+/*
+ * enable_tracer_by_name - enable a tracer on the given instance
+ */
+int enable_tracer_by_name(struct tracefs_instance *inst, const char *tracer)
+{
+	enum tracefs_tracers t;
+	int retval;
+
+	t = TRACEFS_TRACER_CUSTOM;
+
+	debug_msg("enabling %s tracer\n", tracer);
+
+	retval = tracefs_tracer_set(inst, t, tracer);
+	if (retval < 0) {
+		if (errno == ENODEV)
+			err_msg("tracer %s not found!\n", tracer);
+
+		err_msg("failed to enable the tracer %s\n", tracer);
+		return -1;
+	}
+
+	return 0;
+}
+
+/*
+ * disable_tracer - set nop tracer to the insta
+ */
+void disable_tracer(struct tracefs_instance *inst)
+{
+	enum tracefs_tracers t = TRACEFS_TRACER_NOP;
+	int retval;
+
+	retval = tracefs_tracer_set(inst, t);
+	if (retval < 0)
+		err_msg("oops, error disabling tracer\n");
+}
+
+/*
+ * create_instance - create a trace instance with *instance_name
+ */
+struct tracefs_instance *create_instance(char *instance_name)
+{
+	return tracefs_instance_create(instance_name);
+}
+
+/*
+ * destroy_instance - remove a trace instance and free the data
+ */
+void destroy_instance(struct tracefs_instance *inst)
+{
+	tracefs_instance_destroy(inst);
+	tracefs_instance_free(inst);
+}
+
+/*
+ * save_trace_to_file - save the trace output of the instance to the file
+ */
+int save_trace_to_file(struct tracefs_instance *inst, const char *filename)
+{
+	const char *file = "trace";
+	mode_t mode = 0644;
+	char *buffer[4096];
+	int out_fd, in_fd;
+	int retval = -1;
+
+	in_fd = tracefs_instance_file_open(inst, file, O_RDONLY);
+	if (in_fd < 0) {
+		err_msg("Failed to open trace file\n");
+		return -1;
+	}
+
+	out_fd = creat(filename, mode);
+	if (out_fd < 0) {
+		err_msg("Failed to create output file %s\n", filename);
+		goto out_close;
+	}
+
+	do {
+		retval = read(in_fd, buffer, sizeof(buffer));
+		if (read <= 0)
+			goto out_close;
+
+		retval = write(out_fd, buffer, retval);
+		if (retval < 0)
+			goto out_close;
+	} while (retval > 0);
+
+	retval = 0;
+	close(out_fd);
+out_close:
+	close(in_fd);
+	return retval;
+}
+
+struct tracefs_instance *stop_inst;
+static void stop_trace_pipe(int sig)
+{
+	tracefs_trace_pipe_stop(stop_inst);
+}
+
+/*
+ * save_trace_pipe_to_file - save the trace output from a pipe to the file
+ *
+ * it works until receiving a signal to stop.
+ */
+int save_trace_pipe_to_file(struct tracefs_instance *inst, const char *filename)
+{
+	mode_t mode = 0644;
+	int retval;
+	int fd;
+
+	fd = creat(filename, mode);
+
+	stop_inst = inst;
+	signal(SIGINT, stop_trace_pipe);
+	retval = tracefs_trace_pipe_stream(fd, inst, SPLICE_F_NONBLOCK);
+	stop_inst = NULL;
+
+	close(fd);
+	return retval;
+}
+
+/*
+ * collect_registered_events - call the existing callback function for the event
+ *
+ * If an event has a registered callback function, call it.
+ * Otherwise, ignore the event.
+ */
+int
+collect_registered_events(struct tep_event *event, struct tep_record *record,
+			  int cpu, void *context)
+{
+	struct trace_instance *trace = context;
+	struct trace_seq *s = trace->seq;
+
+	if (!event->handler)
+		return 0;
+
+	event->handler(s, record, event, context);
+
+	return 0;
+}
+
+/*
+ * trace_instance_destroy - destroy and free a rtla trace instance
+ */
+void trace_instance_destroy(struct trace_instance *trace)
+{
+	if (trace->inst) {
+		disable_tracer(trace->inst);
+		destroy_instance(trace->inst);
+	}
+
+	if (trace->seq)
+		free(trace->seq);
+
+	if (trace->tep)
+		tep_free(trace->tep);
+}
+
+/*
+ * trace_instance_init - create an rtla trace instance
+ *
+ * It is more than the tracefs instance, as it contains other
+ * things required for the tracing, such as the local events and
+ * a seq file.
+ *
+ * Note that the trace instance is returned disabled. This allows
+ * the tool to apply some other configs, like setting priority
+ * to the kernel threads, before starting generating trace entries.
+ */
+int trace_instance_init(struct trace_instance *trace, char *tool_name)
+{
+	trace->seq = calloc(1, sizeof(*trace->seq));
+	if (!trace->seq)
+		goto out_err;
+
+	trace_seq_init(trace->seq);
+
+	trace->inst = create_instance(tool_name);
+	if (!trace->inst)
+		goto out_err;
+
+	trace->tep = tracefs_local_events(NULL);
+	if (!trace->tep)
+		goto out_err;
+
+	/*
+	 * Let the main enable the record after setting some other
+	 * things such as the priority of the tracer's threads.
+	 */
+	tracefs_trace_off(trace->inst);
+
+	return 0;
+
+out_err:
+	trace_instance_destroy(trace);
+	return 1;
+}
+
+/*
+ * trace_instance_start - start tracing a given rtla instance
+ */
+int trace_instance_start(struct trace_instance *trace)
+{
+	return tracefs_trace_on(trace->inst);
+}
tools/tracing/rtla/src/trace.h new file +27 -0
new file mode 100644
index 000000000000..7c3930909b70
@@ -0,0 +1,27 @@
+// SPDX-License-Identifier: GPL-2.0
+#include <tracefs.h>
+#include <stddef.h>
+
+struct trace_instance {
+	struct tracefs_instance		*inst;
+	struct tep_handle		*tep;
+	struct trace_seq		*seq;
+};
+
+int trace_instance_init(struct trace_instance *trace, char *tool_name);
+int trace_instance_start(struct trace_instance *trace);
+void trace_instance_destroy(struct trace_instance *trace);
+
+struct trace_seq *get_trace_seq(void);
+int enable_tracer_by_name(struct tracefs_instance *inst, const char *tracer);
+void disable_tracer(struct tracefs_instance *inst);
+
+int enable_osnoise(struct trace_instance *trace);
+int enable_timerlat(struct trace_instance *trace);
+
+struct tracefs_instance *create_instance(char *instance_name);
+void destroy_instance(struct tracefs_instance *inst);
+
+int save_trace_to_file(struct tracefs_instance *inst, const char *filename);
+int collect_registered_events(struct tep_event *tep, struct tep_record *record,
+			      int cpu, void *context);
tools/tracing/rtla/src/utils.c new file +433 -0

433 lines not on this page · raw message

tools/tracing/rtla/src/utils.h new file +56 -0

56 lines not on this page · raw message

Signature · 1 line
2.31.1
#27 lore raw

PATCH v4 15/19 rtla: Add rtla osnoise top documentation

In reply to Daniel Bristot de Oliveira

Man page for rtla osnoise top mode.

Cc Steven Rostedt <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Tom Zanussi <[email protected]>
Cc Masami Hiramatsu <[email protected]>
Cc Juri Lelli <[email protected]>
Cc Clark Williams <[email protected]>
Cc John Kacur <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Thomas Gleixner <[email protected]>
Cc Sebastian Andrzej Siewior <[email protected]>
Cc Daniel Bristot de Oliveira <[email protected]>
---
 Documentation/tools/rtla/common_options.rst   | 24 ++++++++
 .../tools/rtla/common_osnoise_options.rst     | 17 ++++++
 .../tools/rtla/common_top_options.rst         |  3 +
 Documentation/tools/rtla/rtla-osnoise-top.rst | 61 +++++++++++++++++++
 4 files changed, 105 insertions(+)
 create mode 100644 Documentation/tools/rtla/common_options.rst
 create mode 100644 Documentation/tools/rtla/common_osnoise_options.rst
 create mode 100644 Documentation/tools/rtla/common_top_options.rst
 create mode 100644 Documentation/tools/rtla/rtla-osnoise-top.rst
Documentation/tools/rtla/common_options.rst new file +24 -0
new file mode 100644
index 000000000000..ea6529667e7c
@@ -0,0 +1,24 @@
+**-c**, **--cpus** *cpu-list*
+
+        Set the osnoise tracer to run the sample threads in the cpu-list.
+
+**-d**, **--duration** *time[s|m|h|d]*
+
+        Set the duration of the session.
+
+**-t**, **--trace**\[*=file*]
+
+        Save the stopped trace to [*file|osnoise_trace.txt*].
+
+**-P**, **--priority** *o:prio|r:prio|f:prio|d:runtime:period*
+
+        Set scheduling parameters to the osnoise tracer threads, the format to set the priority are:
+
+        - *o:prio* - use SCHED_OTHER with *prio*;
+        - *r:prio* - use SCHED_RR with *prio*;
+        - *f:prio* - use SCHED_FIFO with *prio*;
+        - *d:runtime[us|ms|s]:period[us|ms|s]* - use SCHED_DEADLINE with *runtime* and *period* in nanoseconds.
+
+**-h**, **--help**
+
+        Print help menu.
Documentation/tools/rtla/common_osnoise_options.rst new file +17 -0
new file mode 100644
index 000000000000..d556883e4e26
@@ -0,0 +1,17 @@
+**-p**, **--period** *us*
+
+        Set the *osnoise* tracer period in microseconds.
+
+**-r**, **--runtime** *us*
+
+        Set the *osnoise* tracer runtime in microseconds.
+
+**-s**, **--stop** *us*
+
+        Stop the trace if a single sample is higher than the argument in microseconds.
+        If **-T** is set, it will also save the trace to the output.
+
+**-S**, **--stop-total** *us*
+
+        Stop the trace if the total sample is higher than the argument in microseconds.
+        If **-T** is set, it will also save the trace to the output.
Documentation/tools/rtla/common_top_options.rst new file +3 -0
new file mode 100644
index 000000000000..f48878938f84
@@ -0,0 +1,3 @@
+**-q**, **--quiet**
+
+        Print only a summary at the end of the session.
Documentation/tools/rtla/rtla-osnoise-top.rst new file +61 -0

61 lines not on this page · raw message

Signature · 1 line
2.31.1
#29 lore raw

PATCH v4 17/19 rtla: Add rtla timerlat documentation

In reply to Daniel Bristot de Oliveira

Man page for rtla timerlat tool.

Cc Steven Rostedt <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Tom Zanussi <[email protected]>
Cc Masami Hiramatsu <[email protected]>
Cc Juri Lelli <[email protected]>
Cc Clark Williams <[email protected]>
Cc John Kacur <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Thomas Gleixner <[email protected]>
Cc Sebastian Andrzej Siewior <[email protected]>
Cc Daniel Bristot de Oliveira <[email protected]>
---
 .../rtla/common_timerlat_description.rst      | 10 ++++
 Documentation/tools/rtla/rtla-timerlat.rst    | 57 +++++++++++++++++++
 2 files changed, 67 insertions(+)
 create mode 100644 Documentation/tools/rtla/common_timerlat_description.rst
 create mode 100644 Documentation/tools/rtla/rtla-timerlat.rst
Documentation/tools/rtla/common_timerlat_description.rst new file +10 -0
new file mode 100644
index 000000000000..321201cb8597
@@ -0,0 +1,10 @@
+The **rtla timerlat** tool is an interface for the *timerlat* tracer. The
+*timerlat* tracer dispatches a kernel thread per-cpu. These threads
+set a periodic timer to wake themselves up and go back to sleep. After
+the wakeup, they collect and generate useful information for the
+debugging of operating system timer latency.
+
+The *timerlat* tracer outputs information in two ways. It periodically
+prints the timer latency at the timer *IRQ* handler and the *Thread*
+handler. It also enable the trace of the most relevant information via
+**osnoise:** tracepoints.
Documentation/tools/rtla/rtla-timerlat.rst new file +57 -0

57 lines not on this page · raw message

Signature · 1 line
2.31.1
#24 lore raw

PATCH v4 12/19 rtla/timerlat: Add timerlat hist mode

In reply to Daniel Bristot de Oliveira

The rtla hist hist mode displays a histogram of each tracer event occurrence, both for IRQ and timer latencies. The tool also allows many configurations of the timerlat tracer and the collection of the tracer output.

Here is one example of the rtla timerlat hist mode output:
---------- %< ----------
[root@alien ~]# rtla timerlat hist -c 0-3 -d 1M # RTLA timerlat histogram
# Time unit is microseconds (us)
# Duration: 0 00:01:00
Index IRQ-000 Thr-000 IRQ-001 Thr-001 IRQ-002 Thr-002 IRQ-003 Thr-003
0 58572 0 59373 0 58691 0 58895 0
1 1422 57021 628 57241 1310 56160 1102 56805

 2             6      2931         0      2695         0      3567         4      3031
 3             1        40         0        53         0       260         0       142
 4             0         7         0         5         0         6         0        17
 5             0         2         0         5         0         7         0         4
 6             0         0         0         2         0         1         0         1
 8             0         0         0         0         0         0         0         1
 over:         0         0         0         0         0         0         0         0
 count:    60001     60001     60001     60001     60001     60001     60001     60001
 min:          0         1         0         1         0         1         0         1
 avg:          0         1         0         1         0         1         0         1
 max:          3         5         1         6         1         6         2         8

---------- >% ----------

Running
- rtla timerlat hist --help
provides information about the available options.

Cc Steven Rostedt <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Tom Zanussi <[email protected]>
Cc Masami Hiramatsu <[email protected]>
Cc Juri Lelli <[email protected]>
Cc Clark Williams <[email protected]>
Cc John Kacur <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Thomas Gleixner <[email protected]>
Cc Sebastian Andrzej Siewior <[email protected]>
Cc Daniel Bristot de Oliveira <[email protected]>
---
 tools/tracing/rtla/src/timerlat.c      |   6 +-
 tools/tracing/rtla/src/timerlat.h      |   2 +-
 tools/tracing/rtla/src/timerlat_hist.c | 780 +++++++++++++++++++++++++
 3 files changed, 786 insertions(+), 2 deletions(-)
 create mode 100644 tools/tracing/rtla/src/timerlat_hist.c
tools/tracing/rtla/src/timerlat.c +5 -1
index 29a51c01b084..97abbf494fee 100644
@@ -25,7 +25,8 @@ static void timerlat_usage(void)
 		"  usage: [rtla] timerlat [MODE] ...",
 		"",
 		"  modes:",
-		"     top  - prints the summary from timerlat tracer",
+		"     top   - prints the summary from timerlat tracer",
+		"     hist  - prints a histogram of timer latencies",
 		"",
 		"if no MODE is given, the top mode is called, passing the arguments",
 		NULL,
@@ -60,6 +61,9 @@ int timerlat_main(int argc, char *argv[])
 	} else if (strcmp(argv[1], "top") == 0) {
 		timerlat_top_main(argc-1, &argv[1]);
 		exit(0);
+	} else if (strcmp(argv[1], "hist") == 0) {
+		timerlat_hist_main(argc-1, &argv[1]);
+		exit(0);
 	}
 
 usage:
tools/tracing/rtla/src/timerlat.h +1 -1
index 6f9d3925acd0..88561bfd14f3 100644
@@ -1,4 +1,4 @@
 // SPDX-License-Identifier: GPL-2.0
-
+int timerlat_hist_main(int argc, char *argv[]);
 int timerlat_top_main(int argc, char *argv[]);
 int timerlat_main(int argc, char *argv[]);
tools/tracing/rtla/src/timerlat_hist.c new file +780 -0

780 lines not on this page · raw message

Signature · 1 line
2.31.1
#30 lore raw

PATCH v4 18/19 rtla: Add rtla timerlat top documentation

In reply to Daniel Bristot de Oliveira

Man page for rtla timerlat top mode.

Cc Steven Rostedt <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Tom Zanussi <[email protected]>
Cc Masami Hiramatsu <[email protected]>
Cc Juri Lelli <[email protected]>
Cc Clark Williams <[email protected]>
Cc John Kacur <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Thomas Gleixner <[email protected]>
Cc Sebastian Andrzej Siewior <[email protected]>
Cc Daniel Bristot de Oliveira <[email protected]>
---
 .../tools/rtla/common_timerlat_options.rst    |  16 ++
 .../tools/rtla/rtla-timerlat-top.rst          | 145 ++++++++++++++++++
 2 files changed, 161 insertions(+)
 create mode 100644 Documentation/tools/rtla/common_timerlat_options.rst
 create mode 100644 Documentation/tools/rtla/rtla-timerlat-top.rst
Documentation/tools/rtla/common_timerlat_options.rst new file +16 -0
new file mode 100644
index 000000000000..e9c1bfd55d48
@@ -0,0 +1,16 @@
+**-p**, **--period** *us*
+
+        Set the *timerlat* tracer period in microseconds.
+
+**-i**, **--irq** *us*
+
+        Stop trace if the *IRQ* latency is higher than the argument in us.
+
+**-T**, **--thread** *us*
+
+        Stop trace if the *Thread* latency is higher than the argument in us.
+
+**-s**, **--stack** *us*
+
+        Save the stack trace at the *IRQ* if a *Thread* latency is higher than the
+        argument in us.
Documentation/tools/rtla/rtla-timerlat-top.rst new file +145 -0

145 lines not on this page · raw message

Signature · 1 line
2.31.1
#21 lore raw

PATCH v4 9/19 rtla/osnoise: Add osnoise top mode

In reply to Daniel Bristot de Oliveira

The rtla osnoise tool is an interface for the osnoise tracer. The osnoise tracer dispatches a kernel thread per-cpu. These threads read the time in a loop while with preemption, softirqs and IRQs enabled, thus allowing all the sources of osnoise during its execution. The osnoise threads take note of the entry and exit point of any source of interferences, increasing a per-cpu interference counter. The osnoise tracer also saves an interference counter for each source of interference.

The rtla osnoise top mode displays information about the periodic summary from the osnoise tracer.

One example of rtla osnoise top output is:

[root@alien ~]# rtla osnoise top -c 0-3 -d 1m -q -r 900000 -P F:1
Operating System Noise
duration: 0 00:01:00 | time is in us
CPU Period Runtime Noise % CPU Aval Max Noise Max Single HW NMI IRQ Softirq Thread

  0 #58         52200000         1031    99.99802          91           60           0            0        52285            0          101
  1 #59         53100000            5    99.99999           5            5           0            9        53122            0           18
  2 #59         53100000            7    99.99998           7            7           0            8        53115            0           18
  3 #59         53100000         8274    99.98441         277           23           0            9        53778            0          660

"rtla osnoise top --help" works and provide information about the available options.

Cc Steven Rostedt <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Tom Zanussi <[email protected]>
Cc Masami Hiramatsu <[email protected]>
Cc Juri Lelli <[email protected]>
Cc Clark Williams <[email protected]>
Cc John Kacur <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Thomas Gleixner <[email protected]>
Cc Sebastian Andrzej Siewior <[email protected]>
Cc Daniel Bristot de Oliveira <[email protected]>
---
 tools/tracing/rtla/src/osnoise.c     |  20 +-
 tools/tracing/rtla/src/osnoise.h     |   1 +
 tools/tracing/rtla/src/osnoise_top.c | 567 +++++++++++++++++++++++++++
 3 files changed, 587 insertions(+), 1 deletion(-)
 create mode 100644 tools/tracing/rtla/src/osnoise_top.c
tools/tracing/rtla/src/osnoise.c +19 -1
index fa3786c4aedc..b9866dfdda66 100644
@@ -790,7 +790,9 @@ static void osnoise_usage(void)
 		"  usage: [rtla] osnoise [MODE] ...",
 		"",
 		"  modes:",
+		"     top  - prints the summary from osnoise tracer",
 		"",
+		"if no MODE is given, the top mode is called, passing the arguments",
 		NULL,
 	};
 
@@ -801,12 +803,28 @@ static void osnoise_usage(void)
 
 int osnoise_main(int argc, char *argv[])
 {
-	if (argc <= 1)
+	if (argc == 0)
 		goto usage;
 
+	/*
+	 * if osnoise was called without any argument, run the
+	 * default cmdline.
+	 */
+	if (argc == 1) {
+		osnoise_top_main(argc, argv);
+		exit(0);
+	}
+
 	if ((strcmp(argv[1], "-h") == 0) || (strcmp(argv[1], "--help") == 0)) {
 		osnoise_usage();
 		exit(0);
+	} else if (strncmp(argv[1], "-", 1) == 0) {
+		/* the user skipped the tool, call the default one */
+		osnoise_top_main(argc, argv);
+		exit(0);
+	} else if (strcmp(argv[1], "top") == 0) {
+		osnoise_top_main(argc-1, &argv[1]);
+		exit(0);
 	}
 
 usage:
tools/tracing/rtla/src/osnoise.h +1 -0
index c8db59e7c337..4882ee275ea0 100644
@@ -82,4 +82,5 @@ void osnoise_destroy_tool(struct osnoise_tool *top);
 struct osnoise_tool *osnoise_init_tool(char *tool_name);
 struct osnoise_tool *osnoise_init_trace_tool(char *tracer);
 
+int osnoise_top_main(int argc, char **argv);
 int osnoise_main(int argc, char **argv);
tools/tracing/rtla/src/osnoise_top.c new file +567 -0

567 lines not on this page · raw message

Signature · 1 line
2.31.1
#23 lore raw

PATCH v4 11/19 rtla: Add timerlat tool and timelart top mode

In reply to Daniel Bristot de Oliveira

The rtla timerlat tool is an interface for the timerlat tracer. The timerlat tracer dispatches a kernel thread per-cpu. These threads set a periodic timer to wake themselves up and go back to sleep. After the wakeup, they collect and generate useful information for the debugging of operating system timer latency.

The timerlat tracer outputs information in two ways. It periodically prints the timer latency at the timer IRQ handler and the Thread handler. It also provides information for each noise via the osnoise tracepoints.

The rtla timerlat top mode displays a summary of the periodic output from the timerlat tracer.

Here is one example of the rtla timerlat tool output:
---------- %< ----------
[root@alien ~]# rtla timerlat top -c 0-3 -d 1m
Timer Latency

  0 00:01:00   |          IRQ Timer Latency (us)        |         Thread Timer Latency (us)
CPU COUNT      |      cur       min       avg       max |      cur       min       avg       max
  0 #60001     |        0         0         0         3 |        1         1         1         6
  1 #60001     |        0         0         0         3 |        2         1         1         5
  2 #60001     |        0         0         1         6 |        1         1         2         7
  3 #60001     |        0         0         0         7 |        1         1         1        11

---------- >% ----------

Running:

  # rtla timerlat --help
  # rtla timerlat top --help
provides information about the available options.
Cc Steven Rostedt <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Tom Zanussi <[email protected]>
Cc Masami Hiramatsu <[email protected]>
Cc Juri Lelli <[email protected]>
Cc Clark Williams <[email protected]>
Cc John Kacur <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Thomas Gleixner <[email protected]>
Cc Sebastian Andrzej Siewior <[email protected]>
Cc Daniel Bristot de Oliveira <[email protected]>
---
 tools/tracing/rtla/Makefile           |   2 +
 tools/tracing/rtla/src/rtla.c         |   5 +
 tools/tracing/rtla/src/timerlat.c     |  68 +++
 tools/tracing/rtla/src/timerlat.h     |   4 +
 tools/tracing/rtla/src/timerlat_top.c | 581 ++++++++++++++++++++++++++
 5 files changed, 660 insertions(+)
 create mode 100644 tools/tracing/rtla/src/timerlat.c
 create mode 100644 tools/tracing/rtla/src/timerlat.h
 create mode 100644 tools/tracing/rtla/src/timerlat_top.c
tools/tracing/rtla/Makefile +2 -0
index 5e93d000a821..3f8d5421ccc1 100644
@@ -62,6 +62,8 @@ install:
 	$(STRIP) $(DESTDIR)$(BINDIR)/rtla
 	@test ! -f $(DESTDIR)$(BINDIR)/osnoise || rm $(DESTDIR)$(BINDIR)/osnoise
 	ln -s $(DESTDIR)$(BINDIR)/rtla $(DESTDIR)$(BINDIR)/osnoise
+	@test ! -f $(DESTDIR)$(BINDIR)/timerlat || rm $(DESTDIR)$(BINDIR)/timerlat
+	ln -s $(DESTDIR)$(BINDIR)/rtla $(DESTDIR)$(BINDIR)/timerlat
 
 .PHONY: clean tarball
 clean:
tools/tracing/rtla/src/rtla.c +5 -0
index 669b9750b3b3..09bd21b8af81 100644
@@ -9,6 +9,7 @@
 #include <stdio.h>
 
 #include "osnoise.h"
+#include "timerlat.h"
 
 /*
  * rtla_usage - print rtla usage
@@ -25,6 +26,7 @@ static void rtla_usage(void)
 		"",
 		"  commands:",
 		"     osnoise  - gives information about the operating system noise (osnoise)",
+		"     timerlat - measures the timer irq and thread latency",
 		"",
 		NULL,
 	};
@@ -45,6 +47,9 @@ int run_command(int argc, char **argv, int start_position)
 	if (strcmp(argv[start_position], "osnoise") == 0) {
 		osnoise_main(argc-start_position, &argv[start_position]);
 		goto ran;
+	} else if (strcmp(argv[start_position], "timerlat") == 0) {
+		timerlat_main(argc-start_position, &argv[start_position]);
+		goto ran;
 	}
 
 	return 0;
tools/tracing/rtla/src/timerlat.c new file +68 -0

68 lines not on this page · raw message

tools/tracing/rtla/src/timerlat.h new file +4 -0
new file mode 100644
index 000000000000..6f9d3925acd0
@@ -0,0 +1,4 @@
+// SPDX-License-Identifier: GPL-2.0
+
+int timerlat_top_main(int argc, char *argv[]);
+int timerlat_main(int argc, char *argv[]);
tools/tracing/rtla/src/timerlat_top.c new file +581 -0

581 lines not on this page · raw message

Signature · 1 line
2.31.1
#25 lore raw

PATCH v4 13/19 rtla: Add Documentation

In reply to Daniel Bristot de Oliveira

Adds the basis for rtla documentation. This patch also includes the rtla(1) man page.

As suggested by Jonathan Corbet, we are placing these man pages at Documentation/tools/rtla, using rst format. It is not linked to the official documentation, though.

The Makefile is based on bpftool's Documentation one.

Cc Steven Rostedt <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Tom Zanussi <[email protected]>
Cc Masami Hiramatsu <[email protected]>
Cc Juri Lelli <[email protected]>
Cc Clark Williams <[email protected]>
Cc John Kacur <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Thomas Gleixner <[email protected]>
Cc Sebastian Andrzej Siewior <[email protected]>
Cc Daniel Bristot de Oliveira <[email protected]>
---
 Documentation/tools/rtla/Makefile            | 41 +++++++++++++++++
 Documentation/tools/rtla/common_appendix.rst | 12 +++++
 Documentation/tools/rtla/rtla.rst            | 48 ++++++++++++++++++++
 tools/tracing/rtla/Makefile                  | 30 ++++++++++--
 4 files changed, 127 insertions(+), 4 deletions(-)
 create mode 100644 Documentation/tools/rtla/Makefile
 create mode 100644 Documentation/tools/rtla/common_appendix.rst
 create mode 100644 Documentation/tools/rtla/rtla.rst
Documentation/tools/rtla/Makefile new file +41 -0
new file mode 100644
index 000000000000..9f2b84af1a6c
@@ -0,0 +1,41 @@
+# SPDX-License-Identifier: GPL-2.0-only
+# Based on bpftool's Documentation Makefile
+
+INSTALL		?= install
+RM		?= rm -f
+RMDIR		?= rmdir --ignore-fail-on-non-empty
+
+PREFIX		?= /usr/share
+MANDIR		?= $(PREFIX)/man
+MAN1DIR		= $(MANDIR)/man1
+
+MAN1_RST	= $(wildcard rtla*.rst)
+
+_DOC_MAN1	= $(patsubst %.rst,%.1,$(MAN1_RST))
+DOC_MAN1	= $(addprefix $(OUTPUT),$(_DOC_MAN1))
+
+RST2MAN_DEP	:= $(shell command -v rst2man 2>/dev/null)
+RST2MAN_OPTS	+= --verbose
+
+$(OUTPUT)%.1: %.rst
+ifndef RST2MAN_DEP
+	$(error "rst2man not found, but required to generate man pages")
+endif
+	rst2man $(RST2MAN_OPTS) $< > $@
+
+man1: $(DOC_MAN1)
+man: man1
+
+clean:
+	$(RM) $(DOC_MAN1)
+
+install: man
+	$(INSTALL) -d -m 755 $(DESTDIR)$(MAN1DIR)
+	$(INSTALL) -m 644 $(DOC_MAN1) $(DESTDIR)$(MAN1DIR)
+
+uninstall:
+	$(RM) $(addprefix $(DESTDIR)$(MAN1DIR)/,$(_DOC_MAN1))
+	$(RMDIR) $(DESTDIR)$(MAN1DIR)
+
+.PHONY: man man1 clean install uninstall
+.DEFAULT_GOAL := man
Documentation/tools/rtla/common_appendix.rst new file +12 -0
new file mode 100644
index 000000000000..b494084acccd
@@ -0,0 +1,12 @@
+REPORTING BUGS
+==============
+Report bugs to <[email protected]>
+
+LICENSE
+=======
+**rtla** is Free Software licensed under the GNU GPLv2
+
+COPYING
+=======
+Copyright \(C) 2021 Red Hat, Inc. Free use of this software is granted under
+the terms of the GNU Public License (GPL).
Documentation/tools/rtla/rtla.rst new file +48 -0

48 lines not on this page · raw message

tools/tracing/rtla/Makefile +26 -4
index 3f8d5421ccc1..4d1d158f006f 100644
@@ -45,18 +45,28 @@ DATADIR	:=	/usr/share
 DOCDIR	:=	$(DATADIR)/doc
 MANDIR	:=	$(DATADIR)/man
 LICDIR	:=	$(DATADIR)/licenses
+SRCTREE	:=	$(if $(BUILD_SRC),$(BUILD_SRC),$(CURDIR))
+
+# If running from the tarball, man pages are stored in the Documentation
+# dir. If running from the kernel source, man pages are stored in
+# Documentation/tools/rtla/.
+ifneq ($(wildcard Documentation/.*),)
+DOCSRC	=	Documentation/
+else
+DOCSRC	=	$(SRCTREE)/../../../Documentation/tools/rtla/
+endif
 
 .PHONY:	all
 all:	rtla
 
-rtla: $(OBJ)
+rtla: $(OBJ) doc
 	$(CC) -o rtla $(LDFLAGS) $(OBJ) $(LIBS)
 
 static: $(OBJ)
 	$(CC) -o rtla-static $(LDFLAGS) --static $(OBJ) $(LIBS) -lpthread -ldl
 
 .PHONY: install
-install:
+install: doc_install
 	$(INSTALL) -d -m 755 $(DESTDIR)$(BINDIR)
 	$(INSTALL) rtla -m 755 $(DESTDIR)$(BINDIR)
 	$(STRIP) $(DESTDIR)$(BINDIR)/rtla
@@ -66,15 +76,27 @@ install:
 	ln -s $(DESTDIR)$(BINDIR)/rtla $(DESTDIR)$(BINDIR)/timerlat
 
 .PHONY: clean tarball
-clean:
+clean: doc_clean
 	@test ! -f rtla || rm rtla
 	@test ! -f rtla-static || rm rtla-static
 	@test ! -f src/rtla.o || rm src/rtla.o
 	@test ! -f $(TARBALL) || rm -f $(TARBALL)
 	@rm -rf *~ $(OBJ) *.tar.$(CEXT)
 
-tarball:  clean
+tarball: clean
 	rm -rf $(NAME)-$(VERSION) && mkdir $(NAME)-$(VERSION)
 	cp -r $(DIRS) $(FILES) $(NAME)-$(VERSION)
+	mkdir $(NAME)-$(VERSION)/Documentation/
+	cp -rp $(SRCTREE)/../../../Documentation/tools/rtla/* $(NAME)-$(VERSION)/Documentation/
 	tar $(TAROPTS) --exclude='*~' $(NAME)-$(VERSION)
 	rm -rf $(NAME)-$(VERSION)
+
+.PHONY: doc doc_clean doc_install
+doc:
+	$(MAKE) -C $(DOCSRC)
+
+doc_clean:
+	$(MAKE) -C $(DOCSRC) clean
+
+doc_install:
+	$(MAKE) -C $(DOCSRC) install
Signature · 1 line
2.31.1
#20 lore raw

PATCH v4 8/19 rtla: Add osnoise tool

In reply to Daniel Bristot de Oliveira

The osnoise tool is the interface for the osnoise tracer. The osnoise tool will have multiple "modes" with different outputs. At this point, no mode is included.

The osnoise.c includes the osnoise_context abstraction. It serves to read-save-change-restore the default values from tracing/osnoise/ directory. When the context is deleted, the default values are restored.

It also includes some other helper functions for managing osnoise tracer sessions.

With these bits and pieces in place, we can start adding some functionality to rtla.

Cc Steven Rostedt <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Tom Zanussi <[email protected]>
Cc Masami Hiramatsu <[email protected]>
Cc Juri Lelli <[email protected]>
Cc Clark Williams <[email protected]>
Cc John Kacur <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Thomas Gleixner <[email protected]>
Cc Sebastian Andrzej Siewior <[email protected]>
Cc Daniel Bristot de Oliveira <[email protected]>
---
 tools/tracing/rtla/Makefile      |   2 +
 tools/tracing/rtla/src/osnoise.c | 815 +++++++++++++++++++++++++++++++
 tools/tracing/rtla/src/osnoise.h |  85 ++++
 tools/tracing/rtla/src/rtla.c    |  10 +
 4 files changed, 912 insertions(+)
 create mode 100644 tools/tracing/rtla/src/osnoise.c
 create mode 100644 tools/tracing/rtla/src/osnoise.h
tools/tracing/rtla/Makefile +2 -0
index 525e15b76156..5e93d000a821 100644
@@ -60,6 +60,8 @@ install:
 	$(INSTALL) -d -m 755 $(DESTDIR)$(BINDIR)
 	$(INSTALL) rtla -m 755 $(DESTDIR)$(BINDIR)
 	$(STRIP) $(DESTDIR)$(BINDIR)/rtla
+	@test ! -f $(DESTDIR)$(BINDIR)/osnoise || rm $(DESTDIR)$(BINDIR)/osnoise
+	ln -s $(DESTDIR)$(BINDIR)/rtla $(DESTDIR)$(BINDIR)/osnoise
 
 .PHONY: clean tarball
 clean:
tools/tracing/rtla/src/osnoise.c new file +815 -0

815 lines not on this page · raw message

tools/tracing/rtla/src/osnoise.h new file +85 -0

85 lines not on this page · raw message

tools/tracing/rtla/src/rtla.c +10 -0
index 5ae2664ed47d..669b9750b3b3 100644
@@ -8,6 +8,8 @@
 #include <string.h>
 #include <stdio.h>
 
+#include "osnoise.h"
+
 /*
  * rtla_usage - print rtla usage
  */
@@ -22,6 +24,7 @@ static void rtla_usage(void)
 		"  usage: rtla COMMAND ...",
 		"",
 		"  commands:",
+		"     osnoise  - gives information about the operating system noise (osnoise)",
 		"",
 		NULL,
 	};
@@ -39,7 +42,14 @@ static void rtla_usage(void)
  */
 int run_command(int argc, char **argv, int start_position)
 {
+	if (strcmp(argv[start_position], "osnoise") == 0) {
+		osnoise_main(argc-start_position, &argv[start_position]);
+		goto ran;
+	}
+
 	return 0;
+ran:
+	return 1;
 }
 
 int main(int argc, char *argv[])
Signature · 1 line
2.31.1
#26 lore raw

PATCH v4 14/19 rtla: Add rtla osnoise man page

In reply to Daniel Bristot de Oliveira

Man page for rtla osnoise command.

Cc Steven Rostedt <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Tom Zanussi <[email protected]>
Cc Masami Hiramatsu <[email protected]>
Cc Juri Lelli <[email protected]>
Cc Clark Williams <[email protected]>
Cc John Kacur <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Thomas Gleixner <[email protected]>
Cc Sebastian Andrzej Siewior <[email protected]>
Cc Daniel Bristot de Oliveira <[email protected]>
---
 .../tools/rtla/common_osnoise_description.rst |  8 +++
 Documentation/tools/rtla/rtla-osnoise.rst     | 59 +++++++++++++++++++
 2 files changed, 67 insertions(+)
 create mode 100644 Documentation/tools/rtla/common_osnoise_description.rst
 create mode 100644 Documentation/tools/rtla/rtla-osnoise.rst
Documentation/tools/rtla/common_osnoise_description.rst new file +8 -0
new file mode 100644
index 000000000000..8973c5df888f
@@ -0,0 +1,8 @@
+The **rtla osnoise** tool is an interface for the *osnoise* tracer. The
+*osnoise* tracer dispatches a kernel thread per-cpu. These threads read the
+time in a loop while with preemption, softirq and IRQs enabled, thus
+allowing all the sources of operating systme noise during its execution.
+The *osnoise*'s tracer threads take note of the delta between each time
+read, along with an interference counter of all sources of interference.
+At the end of each period, the *osnoise* tracer displays a summary of
+the results.
Documentation/tools/rtla/rtla-osnoise.rst new file +59 -0

59 lines not on this page · raw message

Signature · 1 line
2.31.1
#31 lore raw

PATCH v4 19/19 rtla: Add rtla timerlat hist documentation

In reply to Daniel Bristot de Oliveira

Man page for rtla timerlat hist mode.

Cc Steven Rostedt <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Tom Zanussi <[email protected]>
Cc Masami Hiramatsu <[email protected]>
Cc Juri Lelli <[email protected]>
Cc Clark Williams <[email protected]>
Cc John Kacur <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Thomas Gleixner <[email protected]>
Cc Sebastian Andrzej Siewior <[email protected]>
Cc Daniel Bristot de Oliveira <[email protected]>
---
 .../tools/rtla/rtla-timerlat-hist.rst         | 106 ++++++++++++++++++
 1 file changed, 106 insertions(+)
 create mode 100644 Documentation/tools/rtla/rtla-timerlat-hist.rst
Documentation/tools/rtla/rtla-timerlat-hist.rst new file +106 -0
new file mode 100644
index 000000000000..ba47cf4f4667
@@ -0,0 +1,106 @@
+=====================
+rtla-timerlat-hist
+=====================
+------------------------------------------------
+Histograms of the operating system timer latency
+------------------------------------------------
+
+:Manual section: 1
+
+SYNOPSIS
+========
+**rtla timerlat hist** [*OPTIONS*] ...
+
+DESCRIPTION
+===========
+
+.. include:: common_timerlat_description.rst
+
+The **rtla timerlat hist** displays a histogram of each tracer event
+occurrence. This tool uses the periodic information, and the
+**osnoise:** tracepoints are enabled when using the **-T** option.
+
+OPTIONS
+=======
+
+.. include:: common_timerlat_options.rst
+
+.. include:: common_hist_options.rst
+
+.. include:: common_options.rst
+
+EXAMPLE
+=======
+In the example below, **rtla timerlat hist** is set to run for *10* minutes,
+in the cpus *0-4*, *skipping zero* only lines. Moreover, **rtla timerlat
+hist** will change the priority of the *timelat* threads to run under
+*SCHED_DEADLINE* priority, with a *10us* runtime every *1ms* period. The
+*1ms* period is also passed to the *timerlat* tracer::
+
+  [root@alien ~]# timerlat hist -d 10m -c 0-4 -P d:100us:1ms -p 1ms
+  # RTLA timerlat histogram
+  # Time unit is microseconds (us)
+  # Duration:   0 00:10:00
+  Index   IRQ-000   Thr-000   IRQ-001   Thr-001   IRQ-002   Thr-002   IRQ-003   Thr-003   IRQ-004   Thr-004
+  0        276489         0    206089         0    466018         0    481102         0    205546         0
+  1        318327     35487    388149     30024     94531     48382     83082     71078    388026     55730
+  2          3282    122584      4019    126527     28231    109012     23311     89309      4568     98739
+  3           940     11815       837      9863      6209     16227      6895     17196       910      9780
+  4           444     17287       424     11574      2097     38443      2169     36736       462     13476
+  5           206     43291       255     25581      1223    101908      1304    101137       236     28913
+  6           132    101501        96     64584       635    213774       757    215471        99     73453
+  7            74    169347        65    124758       350     57466       441     53639        69    148573
+  8            53     85183        31    156751       229      9052       306      9026        39    139907
+  9            22     10387        12     42762       161      2554       225      2689        19     26192
+  10           13      1898         8      5770       114      1247       128      1405        13      3772
+  11            9       560         9       924        71       686        76       765         8       713
+  12            4       256         2       360        50       411        64       474         3       278
+  13            2       167         2       172        43       256        53       350         4       180
+  14            1        88         1       116        15       198        42       223         0       115
+  15            2        63         3        94        11       139        20       150         0        58
+  16            2        37         0        56         5        78        10       102         0        39
+  17            0        18         0        28         4        57         8        80         0        15
+  18            0         8         0        17         2        50         6        56         0        12
+  19            0         9         0         5         0        19         0        48         0        18
+  20            0         4         0         8         0        11         2        27         0         4
+  21            0         2         0         3         1         9         1        18         0         6
+  22            0         1         0         3         1         7         0         3         0         5
+  23            0         2         0         4         0         2         0         7         0         2
+  24            0         2         0         2         1         3         0         3         0         5
+  25            0         0         0         1         0         1         0         1         0         3
+  26            0         1         0         0         0         2         0         2         0         0
+  27            0         0         0         3         0         1         0         0         0         1
+  28            0         0         0         3         0         0         0         1         0         0
+  29            0         0         0         2         0         2         0         1         0         3
+  30            0         1         0         0         0         0         0         0         0         0
+  31            0         1         0         0         0         0         0         2         0         2
+  32            0         0         0         1         0         2         0         0         0         0
+  33            0         0         0         2         0         0         0         0         0         1
+  34            0         0         0         0         0         0         0         0         0         2
+  35            0         1         0         1         0         0         0         0         0         1
+  36            0         1         0         0         0         1         0         1         0         0
+  37            0         0         0         1         0         0         0         0         0         0
+  40            0         0         0         0         0         1         0         1         0         0
+  41            0         0         0         0         0         0         0         0         0         1
+  42            0         0         0         0         0         0         0         0         0         1
+  44            0         0         0         0         0         1         0         0         0         0
+  46            0         0         0         0         0         0         0         1         0         0
+  47            0         0         0         0         0         0         0         0         0         1
+  50            0         0         0         0         0         0         0         0         0         1
+  54            0         0         0         1         0         0         0         0         0         0
+  58            0         0         0         1         0         0         0         0         0         0
+  over:         0         0         0         0         0         0         0         0         0         0
+  count:   600002    600002    600002    600002    600002    600002    600002    600002    600002    600002
+  min:          0         1         0         1         0         1         0         1         0         1
+  avg:          0         5         0         5         0         4         0         4         0         5
+  max:         16        36        15        58        24        44        21        46        13        50
+
+SEE ALSO
+========
+**rtla-timerlat**\(1), **rtla-timerlat-top**\(1)
+
+*timerlat* tracer documentation: <https://www.kernel.org/doc/html/latest/trace/timerlat-tracer.html>
+
+AUTHOR
+======
+Written by Daniel Bristot de Oliveira <[email protected]>
Signature · 1 line
2.31.1
maintainer
#2 lore raw

Re: [PATCH V4 01/19] trace/osnoise: Do not follow tracing_cpumask

In reply to Daniel Bristot de Oliveira
Quoted 1 line · On Thu, 21 Oct 2021 14:56:39 +0200 Daniel Bristot de Oliveira…

In preparation to support multiple instances, decople the

"decouple"

Quoted 3 lines

osnoise/timelat workload from instance specific tracing_cpumask.

Different instances can have conflicing cpumasks, making osnoise

"conflicting"

May I suggest a spell check for your commit logs? ;-)

Quoted 33 lines

workload management needlessly complex. Osnoise already have its global cpu mask.

I also thought about using the first instance mask, but the "first" instance could be removed before the others.

This also fixes the problem that changing the tracing_mask was not re-starting the trace.

Cc Steven Rostedt <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Tom Zanussi <[email protected]>
Cc Masami Hiramatsu <[email protected]>
Cc Juri Lelli <[email protected]>
Cc Clark Williams <[email protected]>
Cc John Kacur <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Thomas Gleixner <[email protected]>
Cc Sebastian Andrzej Siewior <[email protected]>
Cc Daniel Bristot de Oliveira <[email protected]>
---
 kernel/trace/trace_osnoise.c | 25 +++++++------------------
 1 file changed, 7 insertions(+), 18 deletions(-)
kernel/trace/trace_osnoise.c +0 -0
index ce053619f289..7b1f8187764c 100644

kernel/trace/trace_osnoise.c static int start_per_cpu_kthreads(struct trace_array *tr) · line 1556

Quoted 5 lines
kernel/trace/trace_osnoise.c +1 -1
@@ -1553,13 +1553,10 @@ static int start_per_cpu_kthreads(struct trace_array *tr)
 
 	cpus_read_lock();
 	/*
-	 * Run only on CPUs in which trace and osnoise are allowed to run.
+	 * Run only on online CPUs in which trace and osnoise are allowed to

which trace and osnise? I thought we were removing "trace"?

-- Steve

Quoted 50 lines
kernel/trace/trace_osnoise.c +6 -17
+	 * run.
 	 */
-	cpumask_and(current_mask, tr->tracing_cpumask, &osnoise_cpumask);
-	/*
-	 * And the CPU is online.
-	 */
-	cpumask_and(current_mask, cpu_online_mask, current_mask);
+	cpumask_and(current_mask, cpu_online_mask, &osnoise_cpumask);
 
 	for_each_possible_cpu(cpu)
 		per_cpu(per_cpu_osnoise_var, cpu).kthread = NULL;
@@ -1580,10 +1577,8 @@ static int start_per_cpu_kthreads(struct trace_array *tr)
 #ifdef CONFIG_HOTPLUG_CPU
 static void osnoise_hotplug_workfn(struct work_struct *dummy)
 {
-	struct trace_array *tr = osnoise_trace;
 	unsigned int cpu = smp_processor_id();
 
-
 	mutex_lock(&trace_types_lock);
 
 	if (!osnoise_busy)
@@ -1595,9 +1590,6 @@ static void osnoise_hotplug_workfn(struct work_struct *dummy)
 	if (!cpumask_test_cpu(cpu, &osnoise_cpumask))
 		goto out_unlock;
 
-	if (!cpumask_test_cpu(cpu, tr->tracing_cpumask))
-		goto out_unlock;
-
 	start_kthread(cpu);
 
 out_unlock:
@@ -1700,13 +1692,10 @@ static void osnoise_tracer_stop(struct trace_array *tr);
  * interface to the osnoise trace. By default, it lists all  CPUs,
  * in this way, allowing osnoise threads to run on any online CPU
  * of the system. It serves to restrict the execution of osnoise to the
- * set of CPUs writing via this interface. Note that osnoise also
- * respects the "tracing_cpumask." Hence, osnoise threads will run only
- * on the set of CPUs allowed here AND on "tracing_cpumask." Why not
- * have just "tracing_cpumask?" Because the user might be interested
- * in tracing what is running on other CPUs. For instance, one might
- * run osnoise in one HT CPU while observing what is running on the
- * sibling HT CPU.
+ * set of CPUs writing via this interface. Why not use "tracing_cpumask"?
+ * Because the user might be interested in tracing what is running on
+ * other CPUs. For instance, one might run osnoise in one HT CPU
+ * while observing what is running on the sibling HT CPU.
  */
 static ssize_t
 osnoise_cpus_write(struct file *filp, const char __user *ubuf, size_t count,
maintainer
#5 lore raw

Re: [PATCH V4 02/19] trace/osnoise: Split workload start from the tracer start

In reply to Daniel Bristot de Oliveira

kernel/trace/trace_osnoise.c static int __osnoise_tracer_start(struct trace_array *tr) · line 1960

Quoted 9 lines · On Thu, 21 Oct 2021 14:56:40 +0200 Daniel Bristot de Oliveira…
kernel/trace/trace_osnoise.c +9 -0
+/*
+ * osnoise_workload_stop - stop the workload and unhook the events
+ */
+static void osnoise_workload_stop(void)
+{
+	if (!osnoise_busy)
+		return;
+
+	trace_osnoise_callback_enabled = false;

I know this is just moving this code, but the original code had this issue too, but there should be a comment here to why we need the compiler barrier.

-- Steve

Quoted 11 lines
kernel/trace/trace_osnoise.c +11 -0
+	barrier();
+
+	stop_per_cpu_kthreads();
+
+	unhook_irq_events();
+	unhook_softirq_events();
+	unhook_thread_events();
+
+	osnoise_busy = false;
+}
+
maintainer
#9 lore raw

Re: [PATCH V4 04/19] trace/osnoise: Support a list of trace_array *tr

In reply to Daniel Bristot de Oliveira
Quoted 95 lines · On Thu, 21 Oct 2021 14:56:42 +0200 Daniel Bristot de Oliveira…

osnoise/timerlat were built to run a single instance, and for this, a single variable is enough to store the current struct trace_array *tr with information about the tracing instance. This is done via the *osnoise_trace variable. A trace_array represents a trace instance.

In preparation to support multiple instances, replace the *osnoise_trace variable with an RCU protected list of instances.

The operations that refer to an instance now propagate to all elements of the list (all instances).

Also, replace the osnoise_busy variable with a check if the list has elements (busy).

No functional change is expected with this patch, i.e., only one instance is allowed yet.

Cc Steven Rostedt <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Tom Zanussi <[email protected]>
Cc Masami Hiramatsu <[email protected]>
Cc Juri Lelli <[email protected]>
Cc Clark Williams <[email protected]>
Cc John Kacur <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Thomas Gleixner <[email protected]>
Cc Sebastian Andrzej Siewior <[email protected]>
Cc Daniel Bristot de Oliveira <[email protected]>
---
 kernel/trace/trace_osnoise.c | 253 +++++++++++++++++++++++++----------
 1 file changed, 184 insertions(+), 69 deletions(-)
kernel/trace/trace_osnoise.c +42 -2
index 9b9e6bc889e9..a6d8f514bd7c 100644
@@ -38,8 +38,6 @@
 #define CREATE_TRACE_POINTS
 #include <trace/events/osnoise.h>
 
-static struct trace_array	*osnoise_trace;
-
 /*
  * Default values.
  */
@@ -50,6 +48,73 @@ static struct trace_array	*osnoise_trace;
 #define DEFAULT_TIMERLAT_PERIOD	1000			/* 1ms */
 #define DEFAULT_TIMERLAT_PRIO	95			/* FIFO 95 */
 
+/*
+ * trace_array of the enabled osnoise/timerlat instances.
+ */
+struct osnoise_instance {
+	struct list_head	list;
+	struct trace_array	*tr;
+};
+struct list_head osnoise_instances;
+
+static bool osnoise_has_registered_instances(void)
+{
+	return !!list_first_or_null_rcu(&osnoise_instances,
+					struct osnoise_instance,
+					list);
+}
+
+/*
+ * osnoise_register_instance - register a new trace instance
+ *
+ * Register a trace_array *tr in the list of instances running
+ * osnoise/timerlat tracers.
+ */
+static int osnoise_register_instance(struct trace_array *tr)
+{
+	struct osnoise_instance *inst;
+
+	inst = kmalloc(sizeof(*inst), GFP_KERNEL);
+	if (!inst)
+		return -ENOMEM;
+
+	INIT_LIST_HEAD_RCU(&inst->list);
+	inst->tr = tr;
+	list_add_tail_rcu(&inst->list, &osnoise_instances);
+
+	return 0;
+}
+
+/*
+ *  osnoise_unregister_instance - unregister a registered trace instance
+ *
+ * Remove the trace_array *tr from the list of instances running
+ * osnoise/timerlat tracers.

kernel/trace/trace_osnoise.c static struct trace_array *osnoise_trace; · line 102

Quoted 10 lines
kernel/trace/trace_osnoise.c +10 -0
+ */
+static void osnoise_unregister_instance(struct trace_array *tr)
+{
+	struct osnoise_instance *inst;
+	int found = 0;
+
+	rcu_read_lock();
+	list_for_each_entry_rcu(inst, &osnoise_instances, list) {
+		if (inst->tr == tr) {
+			list_del_rcu(&inst->list);

Adding to the list requires more than RCU protection, and if this has that protection there's no reason for using rcu to protect here.

[ Added Paul to comment about this too ]

-- Steve

Quoted 411 lines
kernel/trace/trace_osnoise.c +132 -67
+			found = 1;
+			break;
+		}
+	}
+	rcu_read_unlock();
+
+	if (!found) {
+		WARN("osnoise unregister did not find tr %s\n", tr->name);
+		return;
+	}
+
+	synchronize_rcu();
+	kfree(inst);
+}
+
 /*
  * NMI runtime info.
  */
@@ -248,11 +313,6 @@ static struct osnoise_data {
 #endif
 };
 
-/*
- * Boolean variable used to inform that the tracer is currently sampling.
- */
-static bool osnoise_busy;
-
 #ifdef CONFIG_PREEMPT_RT
 /*
  * Print the osnoise header info.
@@ -315,19 +375,24 @@ static void print_osnoise_headers(struct seq_file *s)
  * osnoise_taint - report an osnoise error.
  */
 #define osnoise_taint(msg) ({							\
-	struct trace_array *tr = osnoise_trace;					\
+	struct osnoise_instance *inst;						\
+	struct trace_buffer *buffer;						\
 										\
-	trace_array_printk_buf(tr->array_buffer.buffer, _THIS_IP_, msg);	\
+	rcu_read_lock();							\
+	list_for_each_entry_rcu(inst, &osnoise_instances, list) {		\
+		buffer = inst->tr->array_buffer.buffer;				\
+		trace_array_printk_buf(buffer, _THIS_IP_, msg);			\
+	}									\
+	rcu_read_unlock();							\
 	osnoise_data.tainted = true;						\
 })
 
 /*
  * Record an osnoise_sample into the tracer buffer.
  */
-static void trace_osnoise_sample(struct osnoise_sample *sample)
+static void
+__trace_osnoise_sample(struct osnoise_sample *sample, struct trace_buffer *buffer)
 {
-	struct trace_array *tr = osnoise_trace;
-	struct trace_buffer *buffer = tr->array_buffer.buffer;
 	struct trace_event_call *call = &event_osnoise;
 	struct ring_buffer_event *event;
 	struct osnoise_entry *entry;
@@ -350,6 +415,22 @@ static void trace_osnoise_sample(struct osnoise_sample *sample)
 		trace_buffer_unlock_commit_nostack(buffer, event);
 }
 
+/*
+ * Record an osnoise_sample on all osnoise instances.
+ */
+static void trace_osnoise_sample(struct osnoise_sample *sample)
+{
+	struct osnoise_instance *inst;
+	struct trace_buffer *buffer;
+
+	rcu_read_lock();
+	list_for_each_entry_rcu(inst, &osnoise_instances, list) {
+		buffer = inst->tr->array_buffer.buffer;
+		__trace_osnoise_sample(sample, buffer);
+	}
+	rcu_read_unlock();
+}
+
 #ifdef CONFIG_TIMERLAT_TRACER
 /*
  * Print the timerlat header info.
@@ -387,14 +468,10 @@ static void print_timerlat_headers(struct seq_file *s)
 }
 #endif /* CONFIG_PREEMPT_RT */
 
-/*
- * Record an timerlat_sample into the tracer buffer.
- */
-static void trace_timerlat_sample(struct timerlat_sample *sample)
+static void
+__trace_timerlat_sample(struct timerlat_sample *sample, struct trace_buffer *buffer)
 {
-	struct trace_array *tr = osnoise_trace;
 	struct trace_event_call *call = &event_osnoise;
-	struct trace_buffer *buffer = tr->array_buffer.buffer;
 	struct ring_buffer_event *event;
 	struct timerlat_entry *entry;
 
@@ -411,6 +488,22 @@ static void trace_timerlat_sample(struct timerlat_sample *sample)
 		trace_buffer_unlock_commit_nostack(buffer, event);
 }
 
+/*
+ * Record an timerlat_sample into the tracer buffer.
+ */
+static void trace_timerlat_sample(struct timerlat_sample *sample)
+{
+	struct osnoise_instance *inst;
+	struct trace_buffer *buffer;
+
+	rcu_read_lock();
+	list_for_each_entry_rcu(inst, &osnoise_instances, list) {
+		buffer = inst->tr->array_buffer.buffer;
+		__trace_timerlat_sample(sample, buffer);
+	}
+	rcu_read_unlock();
+}
+
 #ifdef CONFIG_STACKTRACE
 
 #define	MAX_CALLS	256
@@ -450,29 +543,18 @@ static void timerlat_save_stack(int skip)
 	return;
 
 }
-/*
- * timerlat_dump_stack - dump a stack trace previously saved
- *
- * Dump a saved stack trace into the trace buffer.
- */
-static void timerlat_dump_stack(void)
+
+static void
+__timerlat_dump_stack(struct trace_buffer *buffer, struct trace_stack *fstack, unsigned int size)
 {
 	struct trace_event_call *call = &event_osnoise;
-	struct trace_array *tr = osnoise_trace;
-	struct trace_buffer *buffer = tr->array_buffer.buffer;
 	struct ring_buffer_event *event;
-	struct trace_stack *fstack;
 	struct stack_entry *entry;
-	unsigned int size;
-
-	preempt_disable_notrace();
-	fstack = this_cpu_ptr(&trace_stack);
-	size = fstack->stack_size;
 
 	event = trace_buffer_lock_reserve(buffer, TRACE_STACK, sizeof(*entry) + size,
 					  tracing_gen_ctx());
 	if (!event)
-		goto out;
+		return;
 
 	entry = ring_buffer_event_data(event);
 
@@ -481,8 +563,29 @@ static void timerlat_dump_stack(void)
 
 	if (!call_filter_check_discard(call, entry, buffer, event))
 		trace_buffer_unlock_commit_nostack(buffer, event);
+}
 
-out:
+/*
+ * timerlat_dump_stack - dump a stack trace previously saved
+ */
+static void timerlat_dump_stack(void)
+{
+	struct osnoise_instance *inst;
+	struct trace_buffer *buffer;
+	struct trace_stack *fstack;
+	unsigned int size;
+
+	preempt_disable_notrace();
+	fstack = this_cpu_ptr(&trace_stack);
+	size = fstack->stack_size;
+
+	rcu_read_lock();
+	list_for_each_entry_rcu(inst, &osnoise_instances, list) {
+		buffer = inst->tr->array_buffer.buffer;
+		__timerlat_dump_stack(buffer, fstack, size);
+
+	}
+	rcu_read_unlock();
 	preempt_enable_notrace();
 }
 #else
@@ -1077,12 +1180,37 @@ diff_osn_sample_stats(struct osnoise_variables *osn_var, struct osnoise_sample *
  */
 static __always_inline void osnoise_stop_tracing(void)
 {
-	struct trace_array *tr = osnoise_trace;
+	struct osnoise_instance *inst;
+	struct trace_array *tr;
+
+	rcu_read_lock();
+	list_for_each_entry_rcu(inst, &osnoise_instances, list) {
+		tr = inst->tr;
+		trace_array_printk_buf(tr->array_buffer.buffer, _THIS_IP_,
+				"stop tracing hit on cpu %d\n", smp_processor_id());
+
+		tracer_tracing_off(tr);
+	}
+	rcu_read_unlock();
+}
 
-	trace_array_printk_buf(tr->array_buffer.buffer, _THIS_IP_,
-			"stop tracing hit on cpu %d\n", smp_processor_id());
+/*
+ * notify_new_max_latency - Notify a new max latency via fsnotify interface.
+ */
+void notify_new_max_latency(u64 latency)
+{
+	struct osnoise_instance *inst;
+	struct trace_array *tr;
 
-	tracer_tracing_off(tr);
+	rcu_read_lock();
+	list_for_each_entry_rcu(inst, &osnoise_instances, list) {
+		tr = inst->tr;
+		if (tr->max_latency < latency) {
+			tr->max_latency = latency;
+			latency_fsnotify(tr);
+		}
+	}
+	rcu_read_unlock();
 }
 
 /*
@@ -1096,7 +1224,6 @@ static __always_inline void osnoise_stop_tracing(void)
 static int run_osnoise(void)
 {
 	struct osnoise_variables *osn_var = this_cpu_osn_var();
-	struct trace_array *tr = osnoise_trace;
 	u64 start, sample, last_sample;
 	u64 last_int_count, int_count;
 	s64 noise = 0, max_noise = 0;
@@ -1231,11 +1358,7 @@ static int run_osnoise(void)
 
 	trace_osnoise_sample(&s);
 
-	/* Keep a running maximum ever recorded osnoise "latency" */
-	if (max_noise > tr->max_latency) {
-		tr->max_latency = max_noise;
-		latency_fsnotify(tr);
-	}
+	notify_new_max_latency(max_noise);
 
 	if (osnoise_data.stop_tracing_total)
 		if (s.noise > osnoise_data.stop_tracing_total)
@@ -1293,7 +1416,6 @@ static int osnoise_main(void *data)
 static enum hrtimer_restart timerlat_irq(struct hrtimer *timer)
 {
 	struct osnoise_variables *osn_var = this_cpu_osn_var();
-	struct trace_array *tr = osnoise_trace;
 	struct timerlat_variables *tlat;
 	struct timerlat_sample s;
 	u64 now;
@@ -1364,11 +1486,7 @@ static enum hrtimer_restart timerlat_irq(struct hrtimer *timer)
 
 	trace_timerlat_sample(&s);
 
-	/* Keep a running maximum ever recorded os noise "latency" */
-	if (diff > tr->max_latency) {
-		tr->max_latency = diff;
-		latency_fsnotify(tr);
-	}
+	notify_new_max_latency(diff);
 
 	if (osnoise_data.stop_tracing)
 		if (time_to_us(diff) >= osnoise_data.stop_tracing)
@@ -1581,7 +1699,7 @@ static void osnoise_hotplug_workfn(struct work_struct *dummy)
 
 	mutex_lock(&trace_types_lock);
 
-	if (!osnoise_busy)
+	if (!osnoise_has_registered_instances())
 		goto out_unlock_trace;
 
 	mutex_lock(&interface_lock);
@@ -1716,11 +1834,10 @@ osnoise_cpus_write(struct file *filp, const char __user *ubuf, size_t count,
 		goto err_free;
 
 	/*
-	 * trace_types_lock is taken to avoid concurrency on start/stop
-	 * and osnoise_busy.
+	 * trace_types_lock is taken to avoid concurrency on start/stop.
 	 */
 	mutex_lock(&trace_types_lock);
-	running = osnoise_busy;
+	running = osnoise_has_registered_instances();
 	if (running)
 		stop_per_cpu_kthreads();
 
@@ -1941,8 +2058,6 @@ static int osnoise_workload_start(void)
 		return retval;
 	}
 
-	osnoise_busy = true;
-
 	return 0;
 }
 
@@ -1951,7 +2066,7 @@ static int osnoise_workload_start(void)
  */
 static void osnoise_workload_stop(void)
 {
-	if (!osnoise_busy)
+	if (osnoise_has_registered_instances())
 		return;
 
 	trace_osnoise_callback_enabled = false;
@@ -1962,28 +2077,28 @@ static void osnoise_workload_stop(void)
 	unhook_irq_events();
 	unhook_softirq_events();
 	unhook_thread_events();
-
-	osnoise_busy = false;
 }
 
 static void osnoise_tracer_start(struct trace_array *tr)
 {
 	int retval;
 
-	if (osnoise_busy)
+	if (osnoise_has_registered_instances())
 		return;
 
 	retval = osnoise_workload_start();
 	if (retval)
 		pr_err(BANNER "Error starting osnoise tracer\n");
 
+	osnoise_register_instance(tr);
 }
 
 static void osnoise_tracer_stop(struct trace_array *tr)
 {
-	if (!osnoise_busy)
+	if (!osnoise_has_registered_instances())
 		return;
 
+	osnoise_unregister_instance(tr);
 	osnoise_workload_stop();
 }
 
@@ -1991,14 +2106,12 @@ static int osnoise_tracer_init(struct trace_array *tr)
 {
 
 	/* Only allow one instance to enable this */
-	if (osnoise_busy)
+	if (osnoise_has_registered_instances())
 		return -EBUSY;
 
-	osnoise_trace = tr;
 	tr->max_latency = 0;
 
 	osnoise_tracer_start(tr);
-
 	return 0;
 }
 
@@ -2022,7 +2135,7 @@ static void timerlat_tracer_start(struct trace_array *tr)
 {
 	int retval;
 
-	if (osnoise_busy)
+	if (osnoise_has_registered_instances())
 		return;
 
 	osnoise_data.timerlat_tracer = 1;
@@ -2031,6 +2144,8 @@ static void timerlat_tracer_start(struct trace_array *tr)
 	if (retval)
 		goto out_err;
 
+	osnoise_register_instance(tr);
+
 	return;
 out_err:
 	pr_err(BANNER "Error starting timerlat tracer\n");
@@ -2040,7 +2155,7 @@ static void timerlat_tracer_stop(struct trace_array *tr)
 {
 	int cpu;
 
-	if (!osnoise_busy)
+	if (!osnoise_has_registered_instances())
 		return;
 
 	for_each_online_cpu(cpu)
@@ -2054,11 +2169,9 @@ static void timerlat_tracer_stop(struct trace_array *tr)
 static int timerlat_tracer_init(struct trace_array *tr)
 {
 	/* Only allow one instance to enable this */
-	if (osnoise_busy)
+	if (osnoise_has_registered_instances())
 		return -EBUSY;
 
-	osnoise_trace = tr;
-
 	tr->max_latency = 0;
 
 	timerlat_tracer_start(tr);
@@ -2105,6 +2218,8 @@ __init static int init_osnoise_tracer(void)
 #endif
 	osnoise_init_hotplug_support();
 
+	INIT_LIST_HEAD_RCU(&osnoise_instances);
+
 	init_tracefs();
 
 	return 0;
maintainer
#15 lore raw

Re: [PATCH V4 05/19] trace/osnoise: Allow multiple instances of the same tracer

In reply to Daniel Bristot de Oliveira
Quoted 8 lines · On Thu, 21 Oct 2021 14:56:43 +0200 Daniel Bristot de Oliveira…

Allow more than one instance of the same tracer. The workload will start when the first trace_array (instance) is registered and will stop when the last instance is unregistered.

Osnoise and timerlat are still mutually exclusive because of the different behavior of the osnoise: tracepoints.

I don't understand what the above means, or how it is relative to this commit?

-- Steve

#10 lore raw

Re: [PATCH V4 04/19] trace/osnoise: Support a list of trace_array *tr

In reply to Steven Rostedt
Quoted 112 lines · Steven Rostedt
Quoted 95 lines · On Thu, 21 Oct 2021 14:56:42 +0200 Daniel Bristot de Oliveira…

osnoise/timerlat were built to run a single instance, and for this, a single variable is enough to store the current struct trace_array *tr with information about the tracing instance. This is done via the *osnoise_trace variable. A trace_array represents a trace instance.

In preparation to support multiple instances, replace the *osnoise_trace variable with an RCU protected list of instances.

The operations that refer to an instance now propagate to all elements of the list (all instances).

Also, replace the osnoise_busy variable with a check if the list has elements (busy).

No functional change is expected with this patch, i.e., only one instance is allowed yet.

Cc Steven Rostedt <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Tom Zanussi <[email protected]>
Cc Masami Hiramatsu <[email protected]>
Cc Juri Lelli <[email protected]>
Cc Clark Williams <[email protected]>
Cc John Kacur <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Thomas Gleixner <[email protected]>
Cc Sebastian Andrzej Siewior <[email protected]>
Cc Daniel Bristot de Oliveira <[email protected]>
---
 kernel/trace/trace_osnoise.c | 253 +++++++++++++++++++++++++----------
 1 file changed, 184 insertions(+), 69 deletions(-)
kernel/trace/trace_osnoise.c +42 -2
index 9b9e6bc889e9..a6d8f514bd7c 100644
@@ -38,8 +38,6 @@
 #define CREATE_TRACE_POINTS
 #include <trace/events/osnoise.h>
 
-static struct trace_array	*osnoise_trace;
-
 /*
  * Default values.
  */
@@ -50,6 +48,73 @@ static struct trace_array	*osnoise_trace;
 #define DEFAULT_TIMERLAT_PERIOD	1000			/* 1ms */
 #define DEFAULT_TIMERLAT_PRIO	95			/* FIFO 95 */
 
+/*
+ * trace_array of the enabled osnoise/timerlat instances.
+ */
+struct osnoise_instance {
+	struct list_head	list;
+	struct trace_array	*tr;
+};
+struct list_head osnoise_instances;
+
+static bool osnoise_has_registered_instances(void)
+{
+	return !!list_first_or_null_rcu(&osnoise_instances,
+					struct osnoise_instance,
+					list);
+}
+
+/*
+ * osnoise_register_instance - register a new trace instance
+ *
+ * Register a trace_array *tr in the list of instances running
+ * osnoise/timerlat tracers.
+ */
+static int osnoise_register_instance(struct trace_array *tr)
+{
+	struct osnoise_instance *inst;
+
+	inst = kmalloc(sizeof(*inst), GFP_KERNEL);
+	if (!inst)
+		return -ENOMEM;
+
+	INIT_LIST_HEAD_RCU(&inst->list);
+	inst->tr = tr;
+	list_add_tail_rcu(&inst->list, &osnoise_instances);
+
+	return 0;
+}
+
+/*
+ *  osnoise_unregister_instance - unregister a registered trace instance
+ *
+ * Remove the trace_array *tr from the list of instances running
+ * osnoise/timerlat tracers.

kernel/trace/trace_osnoise.c static struct trace_array *osnoise_trace; · line 102

Quoted 10 lines
kernel/trace/trace_osnoise.c +10 -0
+ */
+static void osnoise_unregister_instance(struct trace_array *tr)
+{
+	struct osnoise_instance *inst;
+	int found = 0;
+
+	rcu_read_lock();
+	list_for_each_entry_rcu(inst, &osnoise_instances, list) {
+		if (inst->tr == tr) {
+			list_del_rcu(&inst->list);

Adding to the list requires more than RCU protection, and if this has that protection there's no reason for using rcu to protect here.

Quoted 1 line

[ Added Paul to comment about this too ]

Steve has it right. To see this, suppose that there are two concurrent calls to osnoise_unregister_instance() on elements that are adjacent in the list. The result will of course be a corrupted list. All RCU does is coordinate between readers on the one hand and updaters on the other. There must be some non-RCU synchronization between updaters, be this a lock, some lockless synchronization, there being only one task permitted to carry out updates, or whatever.

So what prevents concurrent calls to osnoise_unregister_instance()? I am not seeing anything preventing this is osnoise_tracer_stop(). Is there some appropriate serialization in the callers of the osnoise_tracer_stop() function?

Thanx, Paul

Quoted 412 lines
Quoted 411 lines
kernel/trace/trace_osnoise.c +132 -67
+			found = 1;
+			break;
+		}
+	}
+	rcu_read_unlock();
+
+	if (!found) {
+		WARN("osnoise unregister did not find tr %s\n", tr->name);
+		return;
+	}
+
+	synchronize_rcu();
+	kfree(inst);
+}
+
 /*
  * NMI runtime info.
  */
@@ -248,11 +313,6 @@ static struct osnoise_data {
 #endif
 };
 
-/*
- * Boolean variable used to inform that the tracer is currently sampling.
- */
-static bool osnoise_busy;
-
 #ifdef CONFIG_PREEMPT_RT
 /*
  * Print the osnoise header info.
@@ -315,19 +375,24 @@ static void print_osnoise_headers(struct seq_file *s)
  * osnoise_taint - report an osnoise error.
  */
 #define osnoise_taint(msg) ({							\
-	struct trace_array *tr = osnoise_trace;					\
+	struct osnoise_instance *inst;						\
+	struct trace_buffer *buffer;						\
 										\
-	trace_array_printk_buf(tr->array_buffer.buffer, _THIS_IP_, msg);	\
+	rcu_read_lock();							\
+	list_for_each_entry_rcu(inst, &osnoise_instances, list) {		\
+		buffer = inst->tr->array_buffer.buffer;				\
+		trace_array_printk_buf(buffer, _THIS_IP_, msg);			\
+	}									\
+	rcu_read_unlock();							\
 	osnoise_data.tainted = true;						\
 })
 
 /*
  * Record an osnoise_sample into the tracer buffer.
  */
-static void trace_osnoise_sample(struct osnoise_sample *sample)
+static void
+__trace_osnoise_sample(struct osnoise_sample *sample, struct trace_buffer *buffer)
 {
-	struct trace_array *tr = osnoise_trace;
-	struct trace_buffer *buffer = tr->array_buffer.buffer;
 	struct trace_event_call *call = &event_osnoise;
 	struct ring_buffer_event *event;
 	struct osnoise_entry *entry;
@@ -350,6 +415,22 @@ static void trace_osnoise_sample(struct osnoise_sample *sample)
 		trace_buffer_unlock_commit_nostack(buffer, event);
 }
 
+/*
+ * Record an osnoise_sample on all osnoise instances.
+ */
+static void trace_osnoise_sample(struct osnoise_sample *sample)
+{
+	struct osnoise_instance *inst;
+	struct trace_buffer *buffer;
+
+	rcu_read_lock();
+	list_for_each_entry_rcu(inst, &osnoise_instances, list) {
+		buffer = inst->tr->array_buffer.buffer;
+		__trace_osnoise_sample(sample, buffer);
+	}
+	rcu_read_unlock();
+}
+
 #ifdef CONFIG_TIMERLAT_TRACER
 /*
  * Print the timerlat header info.
@@ -387,14 +468,10 @@ static void print_timerlat_headers(struct seq_file *s)
 }
 #endif /* CONFIG_PREEMPT_RT */
 
-/*
- * Record an timerlat_sample into the tracer buffer.
- */
-static void trace_timerlat_sample(struct timerlat_sample *sample)
+static void
+__trace_timerlat_sample(struct timerlat_sample *sample, struct trace_buffer *buffer)
 {
-	struct trace_array *tr = osnoise_trace;
 	struct trace_event_call *call = &event_osnoise;
-	struct trace_buffer *buffer = tr->array_buffer.buffer;
 	struct ring_buffer_event *event;
 	struct timerlat_entry *entry;
 
@@ -411,6 +488,22 @@ static void trace_timerlat_sample(struct timerlat_sample *sample)
 		trace_buffer_unlock_commit_nostack(buffer, event);
 }
 
+/*
+ * Record an timerlat_sample into the tracer buffer.
+ */
+static void trace_timerlat_sample(struct timerlat_sample *sample)
+{
+	struct osnoise_instance *inst;
+	struct trace_buffer *buffer;
+
+	rcu_read_lock();
+	list_for_each_entry_rcu(inst, &osnoise_instances, list) {
+		buffer = inst->tr->array_buffer.buffer;
+		__trace_timerlat_sample(sample, buffer);
+	}
+	rcu_read_unlock();
+}
+
 #ifdef CONFIG_STACKTRACE
 
 #define	MAX_CALLS	256
@@ -450,29 +543,18 @@ static void timerlat_save_stack(int skip)
 	return;
 
 }
-/*
- * timerlat_dump_stack - dump a stack trace previously saved
- *
- * Dump a saved stack trace into the trace buffer.
- */
-static void timerlat_dump_stack(void)
+
+static void
+__timerlat_dump_stack(struct trace_buffer *buffer, struct trace_stack *fstack, unsigned int size)
 {
 	struct trace_event_call *call = &event_osnoise;
-	struct trace_array *tr = osnoise_trace;
-	struct trace_buffer *buffer = tr->array_buffer.buffer;
 	struct ring_buffer_event *event;
-	struct trace_stack *fstack;
 	struct stack_entry *entry;
-	unsigned int size;
-
-	preempt_disable_notrace();
-	fstack = this_cpu_ptr(&trace_stack);
-	size = fstack->stack_size;
 
 	event = trace_buffer_lock_reserve(buffer, TRACE_STACK, sizeof(*entry) + size,
 					  tracing_gen_ctx());
 	if (!event)
-		goto out;
+		return;
 
 	entry = ring_buffer_event_data(event);
 
@@ -481,8 +563,29 @@ static void timerlat_dump_stack(void)
 
 	if (!call_filter_check_discard(call, entry, buffer, event))
 		trace_buffer_unlock_commit_nostack(buffer, event);
+}
 
-out:
+/*
+ * timerlat_dump_stack - dump a stack trace previously saved
+ */
+static void timerlat_dump_stack(void)
+{
+	struct osnoise_instance *inst;
+	struct trace_buffer *buffer;
+	struct trace_stack *fstack;
+	unsigned int size;
+
+	preempt_disable_notrace();
+	fstack = this_cpu_ptr(&trace_stack);
+	size = fstack->stack_size;
+
+	rcu_read_lock();
+	list_for_each_entry_rcu(inst, &osnoise_instances, list) {
+		buffer = inst->tr->array_buffer.buffer;
+		__timerlat_dump_stack(buffer, fstack, size);
+
+	}
+	rcu_read_unlock();
 	preempt_enable_notrace();
 }
 #else
@@ -1077,12 +1180,37 @@ diff_osn_sample_stats(struct osnoise_variables *osn_var, struct osnoise_sample *
  */
 static __always_inline void osnoise_stop_tracing(void)
 {
-	struct trace_array *tr = osnoise_trace;
+	struct osnoise_instance *inst;
+	struct trace_array *tr;
+
+	rcu_read_lock();
+	list_for_each_entry_rcu(inst, &osnoise_instances, list) {
+		tr = inst->tr;
+		trace_array_printk_buf(tr->array_buffer.buffer, _THIS_IP_,
+				"stop tracing hit on cpu %d\n", smp_processor_id());
+
+		tracer_tracing_off(tr);
+	}
+	rcu_read_unlock();
+}
 
-	trace_array_printk_buf(tr->array_buffer.buffer, _THIS_IP_,
-			"stop tracing hit on cpu %d\n", smp_processor_id());
+/*
+ * notify_new_max_latency - Notify a new max latency via fsnotify interface.
+ */
+void notify_new_max_latency(u64 latency)
+{
+	struct osnoise_instance *inst;
+	struct trace_array *tr;
 
-	tracer_tracing_off(tr);
+	rcu_read_lock();
+	list_for_each_entry_rcu(inst, &osnoise_instances, list) {
+		tr = inst->tr;
+		if (tr->max_latency < latency) {
+			tr->max_latency = latency;
+			latency_fsnotify(tr);
+		}
+	}
+	rcu_read_unlock();
 }
 
 /*
@@ -1096,7 +1224,6 @@ static __always_inline void osnoise_stop_tracing(void)
 static int run_osnoise(void)
 {
 	struct osnoise_variables *osn_var = this_cpu_osn_var();
-	struct trace_array *tr = osnoise_trace;
 	u64 start, sample, last_sample;
 	u64 last_int_count, int_count;
 	s64 noise = 0, max_noise = 0;
@@ -1231,11 +1358,7 @@ static int run_osnoise(void)
 
 	trace_osnoise_sample(&s);
 
-	/* Keep a running maximum ever recorded osnoise "latency" */
-	if (max_noise > tr->max_latency) {
-		tr->max_latency = max_noise;
-		latency_fsnotify(tr);
-	}
+	notify_new_max_latency(max_noise);
 
 	if (osnoise_data.stop_tracing_total)
 		if (s.noise > osnoise_data.stop_tracing_total)
@@ -1293,7 +1416,6 @@ static int osnoise_main(void *data)
 static enum hrtimer_restart timerlat_irq(struct hrtimer *timer)
 {
 	struct osnoise_variables *osn_var = this_cpu_osn_var();
-	struct trace_array *tr = osnoise_trace;
 	struct timerlat_variables *tlat;
 	struct timerlat_sample s;
 	u64 now;
@@ -1364,11 +1486,7 @@ static enum hrtimer_restart timerlat_irq(struct hrtimer *timer)
 
 	trace_timerlat_sample(&s);
 
-	/* Keep a running maximum ever recorded os noise "latency" */
-	if (diff > tr->max_latency) {
-		tr->max_latency = diff;
-		latency_fsnotify(tr);
-	}
+	notify_new_max_latency(diff);
 
 	if (osnoise_data.stop_tracing)
 		if (time_to_us(diff) >= osnoise_data.stop_tracing)
@@ -1581,7 +1699,7 @@ static void osnoise_hotplug_workfn(struct work_struct *dummy)
 
 	mutex_lock(&trace_types_lock);
 
-	if (!osnoise_busy)
+	if (!osnoise_has_registered_instances())
 		goto out_unlock_trace;
 
 	mutex_lock(&interface_lock);
@@ -1716,11 +1834,10 @@ osnoise_cpus_write(struct file *filp, const char __user *ubuf, size_t count,
 		goto err_free;
 
 	/*
-	 * trace_types_lock is taken to avoid concurrency on start/stop
-	 * and osnoise_busy.
+	 * trace_types_lock is taken to avoid concurrency on start/stop.
 	 */
 	mutex_lock(&trace_types_lock);
-	running = osnoise_busy;
+	running = osnoise_has_registered_instances();
 	if (running)
 		stop_per_cpu_kthreads();
 
@@ -1941,8 +2058,6 @@ static int osnoise_workload_start(void)
 		return retval;
 	}
 
-	osnoise_busy = true;
-
 	return 0;
 }
 
@@ -1951,7 +2066,7 @@ static int osnoise_workload_start(void)
  */
 static void osnoise_workload_stop(void)
 {
-	if (!osnoise_busy)
+	if (osnoise_has_registered_instances())
 		return;
 
 	trace_osnoise_callback_enabled = false;
@@ -1962,28 +2077,28 @@ static void osnoise_workload_stop(void)
 	unhook_irq_events();
 	unhook_softirq_events();
 	unhook_thread_events();
-
-	osnoise_busy = false;
 }
 
 static void osnoise_tracer_start(struct trace_array *tr)
 {
 	int retval;
 
-	if (osnoise_busy)
+	if (osnoise_has_registered_instances())
 		return;
 
 	retval = osnoise_workload_start();
 	if (retval)
 		pr_err(BANNER "Error starting osnoise tracer\n");
 
+	osnoise_register_instance(tr);
 }
 
 static void osnoise_tracer_stop(struct trace_array *tr)
 {
-	if (!osnoise_busy)
+	if (!osnoise_has_registered_instances())
 		return;
 
+	osnoise_unregister_instance(tr);
 	osnoise_workload_stop();
 }
 
@@ -1991,14 +2106,12 @@ static int osnoise_tracer_init(struct trace_array *tr)
 {
 
 	/* Only allow one instance to enable this */
-	if (osnoise_busy)
+	if (osnoise_has_registered_instances())
 		return -EBUSY;
 
-	osnoise_trace = tr;
 	tr->max_latency = 0;
 
 	osnoise_tracer_start(tr);
-
 	return 0;
 }
 
@@ -2022,7 +2135,7 @@ static void timerlat_tracer_start(struct trace_array *tr)
 {
 	int retval;
 
-	if (osnoise_busy)
+	if (osnoise_has_registered_instances())
 		return;
 
 	osnoise_data.timerlat_tracer = 1;
@@ -2031,6 +2144,8 @@ static void timerlat_tracer_start(struct trace_array *tr)
 	if (retval)
 		goto out_err;
 
+	osnoise_register_instance(tr);
+
 	return;
 out_err:
 	pr_err(BANNER "Error starting timerlat tracer\n");
@@ -2040,7 +2155,7 @@ static void timerlat_tracer_stop(struct trace_array *tr)
 {
 	int cpu;
 
-	if (!osnoise_busy)
+	if (!osnoise_has_registered_instances())
 		return;
 
 	for_each_online_cpu(cpu)
@@ -2054,11 +2169,9 @@ static void timerlat_tracer_stop(struct trace_array *tr)
 static int timerlat_tracer_init(struct trace_array *tr)
 {
 	/* Only allow one instance to enable this */
-	if (osnoise_busy)
+	if (osnoise_has_registered_instances())
 		return -EBUSY;
 
-	osnoise_trace = tr;
-
 	tr->max_latency = 0;
 
 	timerlat_tracer_start(tr);
@@ -2105,6 +2218,8 @@ __init static int init_osnoise_tracer(void)
 #endif
 	osnoise_init_hotplug_support();
 
+	INIT_LIST_HEAD_RCU(&osnoise_instances);
+
 	init_tracefs();
 
 	return 0;
#3 lore raw

Re: [PATCH V4 01/19] trace/osnoise: Do not follow tracing_cpumask

In reply to Steven Rostedt
Quoted 7 lines · Steven Rostedt
Quoted 1 line · On Thu, 21 Oct 2021 14:56:39 +0200 Daniel Bristot de Oliveira…

In preparation to support multiple instances, decople the

"decouple"

Quoted 7 lines
Quoted 3 lines

osnoise/timelat workload from instance specific tracing_cpumask.

Different instances can have conflicing cpumasks, making osnoise

"conflicting"

May I suggest a spell check for your commit logs? ;-)

Ooops, I will fix them. I use Grammarly on most of my texts, but sometimes I forget to re-check.

Quoted 34 lines
Quoted 34 lines

workload management needlessly complex. Osnoise already have its global cpu mask.

I also thought about using the first instance mask, but the "first" instance could be removed before the others.

This also fixes the problem that changing the tracing_mask was not re-starting the trace.

Cc Steven Rostedt <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Tom Zanussi <[email protected]>
Cc Masami Hiramatsu <[email protected]>
Cc Juri Lelli <[email protected]>
Cc Clark Williams <[email protected]>
Cc John Kacur <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Thomas Gleixner <[email protected]>
Cc Sebastian Andrzej Siewior <[email protected]>
Cc Daniel Bristot de Oliveira <[email protected]>
---
 kernel/trace/trace_osnoise.c | 25 +++++++------------------
 1 file changed, 7 insertions(+), 18 deletions(-)
kernel/trace/trace_osnoise.c +0 -0
index ce053619f289..7b1f8187764c 100644
@@ -1553,13 +1553,10 @@ static int start_per_cpu_kthreads(struct trace_array *tr)
 
Quoted 6 lines

kernel/trace/trace_osnoise.c static int start_per_cpu_kthreads(struct trace_array *tr) · line 1556

Quoted 4 lines
kernel/trace/trace_osnoise.c +1 -1
 	cpus_read_lock();
 	/*
-	 * Run only on CPUs in which trace and osnoise are allowed to run.
+	 * Run only on online CPUs in which trace and osnoise are allowed to

which trace and osnise? I thought we were removing "trace"?

You are correct, I will remove the trace from the comment.

Thanks!
-- Daniel

Quoted 53 lines

-- Steve

Quoted 50 lines
kernel/trace/trace_osnoise.c +6 -17
+	 * run.
 	 */
-	cpumask_and(current_mask, tr->tracing_cpumask, &osnoise_cpumask);
-	/*
-	 * And the CPU is online.
-	 */
-	cpumask_and(current_mask, cpu_online_mask, current_mask);
+	cpumask_and(current_mask, cpu_online_mask, &osnoise_cpumask);
 
 	for_each_possible_cpu(cpu)
 		per_cpu(per_cpu_osnoise_var, cpu).kthread = NULL;
@@ -1580,10 +1577,8 @@ static int start_per_cpu_kthreads(struct trace_array *tr)
 #ifdef CONFIG_HOTPLUG_CPU
 static void osnoise_hotplug_workfn(struct work_struct *dummy)
 {
-	struct trace_array *tr = osnoise_trace;
 	unsigned int cpu = smp_processor_id();
 
-
 	mutex_lock(&trace_types_lock);
 
 	if (!osnoise_busy)
@@ -1595,9 +1590,6 @@ static void osnoise_hotplug_workfn(struct work_struct *dummy)
 	if (!cpumask_test_cpu(cpu, &osnoise_cpumask))
 		goto out_unlock;
 
-	if (!cpumask_test_cpu(cpu, tr->tracing_cpumask))
-		goto out_unlock;
-
 	start_kthread(cpu);
 
 out_unlock:
@@ -1700,13 +1692,10 @@ static void osnoise_tracer_stop(struct trace_array *tr);
  * interface to the osnoise trace. By default, it lists all  CPUs,
  * in this way, allowing osnoise threads to run on any online CPU
  * of the system. It serves to restrict the execution of osnoise to the
- * set of CPUs writing via this interface. Note that osnoise also
- * respects the "tracing_cpumask." Hence, osnoise threads will run only
- * on the set of CPUs allowed here AND on "tracing_cpumask." Why not
- * have just "tracing_cpumask?" Because the user might be interested
- * in tracing what is running on other CPUs. For instance, one might
- * run osnoise in one HT CPU while observing what is running on the
- * sibling HT CPU.
+ * set of CPUs writing via this interface. Why not use "tracing_cpumask"?
+ * Because the user might be interested in tracing what is running on
+ * other CPUs. For instance, one might run osnoise in one HT CPU
+ * while observing what is running on the sibling HT CPU.
  */
 static ssize_t
 osnoise_cpus_write(struct file *filp, const char __user *ubuf, size_t count,
#6 lore raw

Re: [PATCH V4 02/19] trace/osnoise: Split workload start from the tracer start

In reply to Steven Rostedt
Quoted 16 lines · Steven Rostedt

kernel/trace/trace_osnoise.c static int __osnoise_tracer_start(struct trace_array *tr) · line 1960

Quoted 9 lines · On Thu, 21 Oct 2021 14:56:40 +0200 Daniel Bristot de Oliveira…
kernel/trace/trace_osnoise.c +9 -0
+/*
+ * osnoise_workload_stop - stop the workload and unhook the events
+ */
+static void osnoise_workload_stop(void)
+{
+	if (!osnoise_busy)
+		return;
+
+	trace_osnoise_callback_enabled = false;

I know this is just moving this code, but the original code had this issue too, but there should be a comment here to why we need the compiler barrier.

I will add a comment, like we have on hwlat.

-- Daniel

Quoted 14 lines

-- Steve

Quoted 11 lines
kernel/trace/trace_osnoise.c +11 -0
+	barrier();
+
+	stop_per_cpu_kthreads();
+
+	unhook_irq_events();
+	unhook_softirq_events();
+	unhook_thread_events();
+
+	osnoise_busy = false;
+}
+
#16 lore raw

Re: [PATCH V4 05/19] trace/osnoise: Allow multiple instances of the same tracer

In reply to Steven Rostedt
Quoted 9 lines · Steven Rostedt
Quoted 4 lines · On Thu, 21 Oct 2021 14:56:43 +0200 Daniel Bristot de Oliveira…

Allow more than one instance of the same tracer. The workload will start when the first trace_array (instance) is registered and will stop when the last instance is unregistered.

Quoted 5 lines
Quoted 2 lines

Osnoise and timerlat are still mutually exclusive because of the different behavior of the osnoise: tracepoints.

I don't understand what the above means, or how it is relative to this commit?

osnoise and timerlat tracers share a lot of code, including the osnoise: events.

However, the osnoise: events have different behavior depending on which tracer is enabled. So, while this patch allows the same tracer to run in multiple instances (e.g., two instances running osnoise), it still does not allow instances of timerlat and osnoise in parallel (e.g., one timerlat and osnoise).

Should I add the text above to the log?

It is possible to extend the current behavior to allow concurrent osnoise and timerlat tracers to run on different CPUs. For instance, osnoise on cpu 0,1 and timerlat on cpu 2,3. This is on my to-do list, but it is not necessary for now.

-- Daniel

Quoted 2 lines

-- Steve

#11 lore raw

Re: [PATCH V4 04/19] trace/osnoise: Support a list of trace_array *tr

In reply to Paul E. McKenney
Quoted 123 lines · Paul E. McKenney
Quoted 112 lines · Steven Rostedt
Quoted 95 lines · On Thu, 21 Oct 2021 14:56:42 +0200 Daniel Bristot de Oliveira…

osnoise/timerlat were built to run a single instance, and for this, a single variable is enough to store the current struct trace_array *tr with information about the tracing instance. This is done via the *osnoise_trace variable. A trace_array represents a trace instance.

In preparation to support multiple instances, replace the *osnoise_trace variable with an RCU protected list of instances.

The operations that refer to an instance now propagate to all elements of the list (all instances).

Also, replace the osnoise_busy variable with a check if the list has elements (busy).

No functional change is expected with this patch, i.e., only one instance is allowed yet.

Cc Steven Rostedt <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Tom Zanussi <[email protected]>
Cc Masami Hiramatsu <[email protected]>
Cc Juri Lelli <[email protected]>
Cc Clark Williams <[email protected]>
Cc John Kacur <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Thomas Gleixner <[email protected]>
Cc Sebastian Andrzej Siewior <[email protected]>
Cc Daniel Bristot de Oliveira <[email protected]>
---
 kernel/trace/trace_osnoise.c | 253 +++++++++++++++++++++++++----------
 1 file changed, 184 insertions(+), 69 deletions(-)
kernel/trace/trace_osnoise.c +42 -2
index 9b9e6bc889e9..a6d8f514bd7c 100644
@@ -38,8 +38,6 @@
 #define CREATE_TRACE_POINTS
 #include <trace/events/osnoise.h>
 
-static struct trace_array	*osnoise_trace;
-
 /*
  * Default values.
  */
@@ -50,6 +48,73 @@ static struct trace_array	*osnoise_trace;
 #define DEFAULT_TIMERLAT_PERIOD	1000			/* 1ms */
 #define DEFAULT_TIMERLAT_PRIO	95			/* FIFO 95 */
 
+/*
+ * trace_array of the enabled osnoise/timerlat instances.
+ */
+struct osnoise_instance {
+	struct list_head	list;
+	struct trace_array	*tr;
+};
+struct list_head osnoise_instances;
+
+static bool osnoise_has_registered_instances(void)
+{
+	return !!list_first_or_null_rcu(&osnoise_instances,
+					struct osnoise_instance,
+					list);
+}
+
+/*
+ * osnoise_register_instance - register a new trace instance
+ *
+ * Register a trace_array *tr in the list of instances running
+ * osnoise/timerlat tracers.
+ */
+static int osnoise_register_instance(struct trace_array *tr)
+{
+	struct osnoise_instance *inst;
+
+	inst = kmalloc(sizeof(*inst), GFP_KERNEL);
+	if (!inst)
+		return -ENOMEM;
+
+	INIT_LIST_HEAD_RCU(&inst->list);
+	inst->tr = tr;
+	list_add_tail_rcu(&inst->list, &osnoise_instances);
+
+	return 0;
+}
+
+/*
+ *  osnoise_unregister_instance - unregister a registered trace instance
+ *
+ * Remove the trace_array *tr from the list of instances running
+ * osnoise/timerlat tracers.

kernel/trace/trace_osnoise.c static struct trace_array *osnoise_trace; · line 102

Quoted 10 lines
kernel/trace/trace_osnoise.c +10 -0
+ */
+static void osnoise_unregister_instance(struct trace_array *tr)
+{
+	struct osnoise_instance *inst;
+	int found = 0;
+
+	rcu_read_lock();
+	list_for_each_entry_rcu(inst, &osnoise_instances, list) {
+		if (inst->tr == tr) {
+			list_del_rcu(&inst->list);

Adding to the list requires more than RCU protection, and if this has that protection there's no reason for using rcu to protect here.

Quoted 1 line

[ Added Paul to comment about this too ]

Steve has it right. To see this, suppose that there are two concurrent calls to osnoise_unregister_instance() on elements that are adjacent in the list. The result will of course be a corrupted list. All RCU does is coordinate between readers on the one hand and updaters on the other. There must be some non-RCU synchronization between updaters, be this a lock, some lockless synchronization, there being only one task permitted to carry out updates, or whatever.

Quoted 6 lines

So what prevents concurrent calls to osnoise_unregister_instance()? I am not seeing anything preventing this is osnoise_tracer_stop(). Is there some appropriate serialization in the callers of the osnoise_tracer_stop() function?

Thanx, Paul

Thanks Steven and Paul!

The start/stop/reset operations are called from trace "core" operations. So they are all protected/serialized by trace_types_lock (by trace/core).

So, it seems that the way to go is to remove the rcu_read_lock/unlock() from unregister, adding a lockdep check to see if trace_types_lock is help as a bonus?

-- Daniel

Quoted 413 lines
Quoted 413 lines
Quoted 412 lines
kernel/trace/trace_osnoise.c +132 -67
+			found = 1;
+			break;
+		}
+	}
+	rcu_read_unlock();
+
+	if (!found) {
+		WARN("osnoise unregister did not find tr %s\n", tr->name);
+		return;
+	}
+
+	synchronize_rcu();
+	kfree(inst);
+}
+
 /*
  * NMI runtime info.
  */
@@ -248,11 +313,6 @@ static struct osnoise_data {
 #endif
 };
 
-/*
- * Boolean variable used to inform that the tracer is currently sampling.
- */
-static bool osnoise_busy;
-
 #ifdef CONFIG_PREEMPT_RT
 /*
  * Print the osnoise header info.
@@ -315,19 +375,24 @@ static void print_osnoise_headers(struct seq_file *s)
  * osnoise_taint - report an osnoise error.
  */
 #define osnoise_taint(msg) ({							\
-	struct trace_array *tr = osnoise_trace;					\
+	struct osnoise_instance *inst;						\
+	struct trace_buffer *buffer;						\
 										\
-	trace_array_printk_buf(tr->array_buffer.buffer, _THIS_IP_, msg);	\
+	rcu_read_lock();							\
+	list_for_each_entry_rcu(inst, &osnoise_instances, list) {		\
+		buffer = inst->tr->array_buffer.buffer;				\
+		trace_array_printk_buf(buffer, _THIS_IP_, msg);			\
+	}									\
+	rcu_read_unlock();							\
 	osnoise_data.tainted = true;						\
 })
 
 /*
  * Record an osnoise_sample into the tracer buffer.
  */
-static void trace_osnoise_sample(struct osnoise_sample *sample)
+static void
+__trace_osnoise_sample(struct osnoise_sample *sample, struct trace_buffer *buffer)
 {
-	struct trace_array *tr = osnoise_trace;
-	struct trace_buffer *buffer = tr->array_buffer.buffer;
 	struct trace_event_call *call = &event_osnoise;
 	struct ring_buffer_event *event;
 	struct osnoise_entry *entry;
@@ -350,6 +415,22 @@ static void trace_osnoise_sample(struct osnoise_sample *sample)
 		trace_buffer_unlock_commit_nostack(buffer, event);
 }
 
+/*
+ * Record an osnoise_sample on all osnoise instances.
+ */
+static void trace_osnoise_sample(struct osnoise_sample *sample)
+{
+	struct osnoise_instance *inst;
+	struct trace_buffer *buffer;
+
+	rcu_read_lock();
+	list_for_each_entry_rcu(inst, &osnoise_instances, list) {
+		buffer = inst->tr->array_buffer.buffer;
+		__trace_osnoise_sample(sample, buffer);
+	}
+	rcu_read_unlock();
+}
+
 #ifdef CONFIG_TIMERLAT_TRACER
 /*
  * Print the timerlat header info.
@@ -387,14 +468,10 @@ static void print_timerlat_headers(struct seq_file *s)
 }
 #endif /* CONFIG_PREEMPT_RT */
 
-/*
- * Record an timerlat_sample into the tracer buffer.
- */
-static void trace_timerlat_sample(struct timerlat_sample *sample)
+static void
+__trace_timerlat_sample(struct timerlat_sample *sample, struct trace_buffer *buffer)
 {
-	struct trace_array *tr = osnoise_trace;
 	struct trace_event_call *call = &event_osnoise;
-	struct trace_buffer *buffer = tr->array_buffer.buffer;
 	struct ring_buffer_event *event;
 	struct timerlat_entry *entry;
 
@@ -411,6 +488,22 @@ static void trace_timerlat_sample(struct timerlat_sample *sample)
 		trace_buffer_unlock_commit_nostack(buffer, event);
 }
 
+/*
+ * Record an timerlat_sample into the tracer buffer.
+ */
+static void trace_timerlat_sample(struct timerlat_sample *sample)
+{
+	struct osnoise_instance *inst;
+	struct trace_buffer *buffer;
+
+	rcu_read_lock();
+	list_for_each_entry_rcu(inst, &osnoise_instances, list) {
+		buffer = inst->tr->array_buffer.buffer;
+		__trace_timerlat_sample(sample, buffer);
+	}
+	rcu_read_unlock();
+}
+
 #ifdef CONFIG_STACKTRACE
 
 #define	MAX_CALLS	256
@@ -450,29 +543,18 @@ static void timerlat_save_stack(int skip)
 	return;
 
 }
-/*
- * timerlat_dump_stack - dump a stack trace previously saved
- *
- * Dump a saved stack trace into the trace buffer.
- */
-static void timerlat_dump_stack(void)
+
+static void
+__timerlat_dump_stack(struct trace_buffer *buffer, struct trace_stack *fstack, unsigned int size)
 {
 	struct trace_event_call *call = &event_osnoise;
-	struct trace_array *tr = osnoise_trace;
-	struct trace_buffer *buffer = tr->array_buffer.buffer;
 	struct ring_buffer_event *event;
-	struct trace_stack *fstack;
 	struct stack_entry *entry;
-	unsigned int size;
-
-	preempt_disable_notrace();
-	fstack = this_cpu_ptr(&trace_stack);
-	size = fstack->stack_size;
 
 	event = trace_buffer_lock_reserve(buffer, TRACE_STACK, sizeof(*entry) + size,
 					  tracing_gen_ctx());
 	if (!event)
-		goto out;
+		return;
 
 	entry = ring_buffer_event_data(event);
 
@@ -481,8 +563,29 @@ static void timerlat_dump_stack(void)
 
 	if (!call_filter_check_discard(call, entry, buffer, event))
 		trace_buffer_unlock_commit_nostack(buffer, event);
+}
 
-out:
+/*
+ * timerlat_dump_stack - dump a stack trace previously saved
+ */
+static void timerlat_dump_stack(void)
+{
+	struct osnoise_instance *inst;
+	struct trace_buffer *buffer;
+	struct trace_stack *fstack;
+	unsigned int size;
+
+	preempt_disable_notrace();
+	fstack = this_cpu_ptr(&trace_stack);
+	size = fstack->stack_size;
+
+	rcu_read_lock();
+	list_for_each_entry_rcu(inst, &osnoise_instances, list) {
+		buffer = inst->tr->array_buffer.buffer;
+		__timerlat_dump_stack(buffer, fstack, size);
+
+	}
+	rcu_read_unlock();
 	preempt_enable_notrace();
 }
 #else
@@ -1077,12 +1180,37 @@ diff_osn_sample_stats(struct osnoise_variables *osn_var, struct osnoise_sample *
  */
 static __always_inline void osnoise_stop_tracing(void)
 {
-	struct trace_array *tr = osnoise_trace;
+	struct osnoise_instance *inst;
+	struct trace_array *tr;
+
+	rcu_read_lock();
+	list_for_each_entry_rcu(inst, &osnoise_instances, list) {
+		tr = inst->tr;
+		trace_array_printk_buf(tr->array_buffer.buffer, _THIS_IP_,
+				"stop tracing hit on cpu %d\n", smp_processor_id());
+
+		tracer_tracing_off(tr);
+	}
+	rcu_read_unlock();
+}
 
-	trace_array_printk_buf(tr->array_buffer.buffer, _THIS_IP_,
-			"stop tracing hit on cpu %d\n", smp_processor_id());
+/*
+ * notify_new_max_latency - Notify a new max latency via fsnotify interface.
+ */
+void notify_new_max_latency(u64 latency)
+{
+	struct osnoise_instance *inst;
+	struct trace_array *tr;
 
-	tracer_tracing_off(tr);
+	rcu_read_lock();
+	list_for_each_entry_rcu(inst, &osnoise_instances, list) {
+		tr = inst->tr;
+		if (tr->max_latency < latency) {
+			tr->max_latency = latency;
+			latency_fsnotify(tr);
+		}
+	}
+	rcu_read_unlock();
 }
 
 /*
@@ -1096,7 +1224,6 @@ static __always_inline void osnoise_stop_tracing(void)
 static int run_osnoise(void)
 {
 	struct osnoise_variables *osn_var = this_cpu_osn_var();
-	struct trace_array *tr = osnoise_trace;
 	u64 start, sample, last_sample;
 	u64 last_int_count, int_count;
 	s64 noise = 0, max_noise = 0;
@@ -1231,11 +1358,7 @@ static int run_osnoise(void)
 
 	trace_osnoise_sample(&s);
 
-	/* Keep a running maximum ever recorded osnoise "latency" */
-	if (max_noise > tr->max_latency) {
-		tr->max_latency = max_noise;
-		latency_fsnotify(tr);
-	}
+	notify_new_max_latency(max_noise);
 
 	if (osnoise_data.stop_tracing_total)
 		if (s.noise > osnoise_data.stop_tracing_total)
@@ -1293,7 +1416,6 @@ static int osnoise_main(void *data)
 static enum hrtimer_restart timerlat_irq(struct hrtimer *timer)
 {
 	struct osnoise_variables *osn_var = this_cpu_osn_var();
-	struct trace_array *tr = osnoise_trace;
 	struct timerlat_variables *tlat;
 	struct timerlat_sample s;
 	u64 now;
@@ -1364,11 +1486,7 @@ static enum hrtimer_restart timerlat_irq(struct hrtimer *timer)
 
 	trace_timerlat_sample(&s);
 
-	/* Keep a running maximum ever recorded os noise "latency" */
-	if (diff > tr->max_latency) {
-		tr->max_latency = diff;
-		latency_fsnotify(tr);
-	}
+	notify_new_max_latency(diff);
 
 	if (osnoise_data.stop_tracing)
 		if (time_to_us(diff) >= osnoise_data.stop_tracing)
@@ -1581,7 +1699,7 @@ static void osnoise_hotplug_workfn(struct work_struct *dummy)
 
 	mutex_lock(&trace_types_lock);
 
-	if (!osnoise_busy)
+	if (!osnoise_has_registered_instances())
 		goto out_unlock_trace;
 
 	mutex_lock(&interface_lock);
@@ -1716,11 +1834,10 @@ osnoise_cpus_write(struct file *filp, const char __user *ubuf, size_t count,
 		goto err_free;
 
 	/*
-	 * trace_types_lock is taken to avoid concurrency on start/stop
-	 * and osnoise_busy.
+	 * trace_types_lock is taken to avoid concurrency on start/stop.
 	 */
 	mutex_lock(&trace_types_lock);
-	running = osnoise_busy;
+	running = osnoise_has_registered_instances();
 	if (running)
 		stop_per_cpu_kthreads();
 
@@ -1941,8 +2058,6 @@ static int osnoise_workload_start(void)
 		return retval;
 	}
 
-	osnoise_busy = true;
-
 	return 0;
 }
 
@@ -1951,7 +2066,7 @@ static int osnoise_workload_start(void)
  */
 static void osnoise_workload_stop(void)
 {
-	if (!osnoise_busy)
+	if (osnoise_has_registered_instances())
 		return;
 
 	trace_osnoise_callback_enabled = false;
@@ -1962,28 +2077,28 @@ static void osnoise_workload_stop(void)
 	unhook_irq_events();
 	unhook_softirq_events();
 	unhook_thread_events();
-
-	osnoise_busy = false;
 }
 
 static void osnoise_tracer_start(struct trace_array *tr)
 {
 	int retval;
 
-	if (osnoise_busy)
+	if (osnoise_has_registered_instances())
 		return;
 
 	retval = osnoise_workload_start();
 	if (retval)
 		pr_err(BANNER "Error starting osnoise tracer\n");
 
+	osnoise_register_instance(tr);
 }
 
 static void osnoise_tracer_stop(struct trace_array *tr)
 {
-	if (!osnoise_busy)
+	if (!osnoise_has_registered_instances())
 		return;
 
+	osnoise_unregister_instance(tr);
 	osnoise_workload_stop();
 }
 
@@ -1991,14 +2106,12 @@ static int osnoise_tracer_init(struct trace_array *tr)
 {
 
 	/* Only allow one instance to enable this */
-	if (osnoise_busy)
+	if (osnoise_has_registered_instances())
 		return -EBUSY;
 
-	osnoise_trace = tr;
 	tr->max_latency = 0;
 
 	osnoise_tracer_start(tr);
-
 	return 0;
 }
 
@@ -2022,7 +2135,7 @@ static void timerlat_tracer_start(struct trace_array *tr)
 {
 	int retval;
 
-	if (osnoise_busy)
+	if (osnoise_has_registered_instances())
 		return;
 
 	osnoise_data.timerlat_tracer = 1;
@@ -2031,6 +2144,8 @@ static void timerlat_tracer_start(struct trace_array *tr)
 	if (retval)
 		goto out_err;
 
+	osnoise_register_instance(tr);
+
 	return;
 out_err:
 	pr_err(BANNER "Error starting timerlat tracer\n");
@@ -2040,7 +2155,7 @@ static void timerlat_tracer_stop(struct trace_array *tr)
 {
 	int cpu;
 
-	if (!osnoise_busy)
+	if (!osnoise_has_registered_instances())
 		return;
 
 	for_each_online_cpu(cpu)
@@ -2054,11 +2169,9 @@ static void timerlat_tracer_stop(struct trace_array *tr)
 static int timerlat_tracer_init(struct trace_array *tr)
 {
 	/* Only allow one instance to enable this */
-	if (osnoise_busy)
+	if (osnoise_has_registered_instances())
 		return -EBUSY;
 
-	osnoise_trace = tr;
-
 	tr->max_latency = 0;
 
 	timerlat_tracer_start(tr);
@@ -2105,6 +2218,8 @@ __init static int init_osnoise_tracer(void)
 #endif
 	osnoise_init_hotplug_support();
 
+	INIT_LIST_HEAD_RCU(&osnoise_instances);
+
 	init_tracefs();
 
 	return 0;
maintainer
#12 lore raw

Re: [PATCH V4 04/19] trace/osnoise: Support a list of trace_array *tr

Quoted 5 lines · On Mon, 25 Oct 2021 11:10:26 +0200 Daniel Bristot de Oliveira…

The start/stop/reset operations are called from trace "core" operations. So they are all protected/serialized by trace_types_lock (by trace/core).

So, it seems that the way to go is to remove the rcu_read_lock/unlock() from unregister, adding a lockdep check to see if trace_types_lock is help as a bonus?

Yes, I would recommend the lockdep assert added. This way it also documents what lock protects modification of the list.

Thanks,

-- Steve

#13 lore raw

Re: [PATCH V4 04/19] trace/osnoise: Support a list of trace_array *tr

In reply to Daniel Bristot de Oliveira
Quoted 131 lines · Daniel Bristot de Oliveira
Quoted 123 lines · Paul E. McKenney
Quoted 112 lines · Steven Rostedt
Quoted 95 lines · On Thu, 21 Oct 2021 14:56:42 +0200 Daniel Bristot de Oliveira…

osnoise/timerlat were built to run a single instance, and for this, a single variable is enough to store the current struct trace_array *tr with information about the tracing instance. This is done via the *osnoise_trace variable. A trace_array represents a trace instance.

In preparation to support multiple instances, replace the *osnoise_trace variable with an RCU protected list of instances.

The operations that refer to an instance now propagate to all elements of the list (all instances).

Also, replace the osnoise_busy variable with a check if the list has elements (busy).

No functional change is expected with this patch, i.e., only one instance is allowed yet.

Cc Steven Rostedt <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Tom Zanussi <[email protected]>
Cc Masami Hiramatsu <[email protected]>
Cc Juri Lelli <[email protected]>
Cc Clark Williams <[email protected]>
Cc John Kacur <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Thomas Gleixner <[email protected]>
Cc Sebastian Andrzej Siewior <[email protected]>
Cc Daniel Bristot de Oliveira <[email protected]>
---
 kernel/trace/trace_osnoise.c | 253 +++++++++++++++++++++++++----------
 1 file changed, 184 insertions(+), 69 deletions(-)
kernel/trace/trace_osnoise.c +42 -2
index 9b9e6bc889e9..a6d8f514bd7c 100644
@@ -38,8 +38,6 @@
 #define CREATE_TRACE_POINTS
 #include <trace/events/osnoise.h>
 
-static struct trace_array	*osnoise_trace;
-
 /*
  * Default values.
  */
@@ -50,6 +48,73 @@ static struct trace_array	*osnoise_trace;
 #define DEFAULT_TIMERLAT_PERIOD	1000			/* 1ms */
 #define DEFAULT_TIMERLAT_PRIO	95			/* FIFO 95 */
 
+/*
+ * trace_array of the enabled osnoise/timerlat instances.
+ */
+struct osnoise_instance {
+	struct list_head	list;
+	struct trace_array	*tr;
+};
+struct list_head osnoise_instances;
+
+static bool osnoise_has_registered_instances(void)
+{
+	return !!list_first_or_null_rcu(&osnoise_instances,
+					struct osnoise_instance,
+					list);
+}
+
+/*
+ * osnoise_register_instance - register a new trace instance
+ *
+ * Register a trace_array *tr in the list of instances running
+ * osnoise/timerlat tracers.
+ */
+static int osnoise_register_instance(struct trace_array *tr)
+{
+	struct osnoise_instance *inst;
+
+	inst = kmalloc(sizeof(*inst), GFP_KERNEL);
+	if (!inst)
+		return -ENOMEM;
+
+	INIT_LIST_HEAD_RCU(&inst->list);
+	inst->tr = tr;
+	list_add_tail_rcu(&inst->list, &osnoise_instances);
+
+	return 0;
+}
+
+/*
+ *  osnoise_unregister_instance - unregister a registered trace instance
+ *
+ * Remove the trace_array *tr from the list of instances running
+ * osnoise/timerlat tracers.

kernel/trace/trace_osnoise.c static struct trace_array *osnoise_trace; · line 102

Quoted 10 lines
kernel/trace/trace_osnoise.c +10 -0
+ */
+static void osnoise_unregister_instance(struct trace_array *tr)
+{
+	struct osnoise_instance *inst;
+	int found = 0;
+
+	rcu_read_lock();
+	list_for_each_entry_rcu(inst, &osnoise_instances, list) {
+		if (inst->tr == tr) {
+			list_del_rcu(&inst->list);

Adding to the list requires more than RCU protection, and if this has that protection there's no reason for using rcu to protect here.

Quoted 1 line

[ Added Paul to comment about this too ]

Steve has it right. To see this, suppose that there are two concurrent calls to osnoise_unregister_instance() on elements that are adjacent in the list. The result will of course be a corrupted list. All RCU does is coordinate between readers on the one hand and updaters on the other. There must be some non-RCU synchronization between updaters, be this a lock, some lockless synchronization, there being only one task permitted to carry out updates, or whatever.

Quoted 6 lines

So what prevents concurrent calls to osnoise_unregister_instance()? I am not seeing anything preventing this is osnoise_tracer_stop(). Is there some appropriate serialization in the callers of the osnoise_tracer_stop() function?

Thanx, Paul

Quoted 7 lines

Thanks Steven and Paul!

The start/stop/reset operations are called from trace "core" operations. So they are all protected/serialized by trace_types_lock (by trace/core).

So, it seems that the way to go is to remove the rcu_read_lock/unlock() from unregister, adding a lockdep check to see if trace_types_lock is help as a bonus?

Works for me!

Thanx, Paul

Quoted 416 lines

-- Daniel

Quoted 413 lines
Quoted 413 lines
Quoted 412 lines
kernel/trace/trace_osnoise.c +132 -67
+			found = 1;
+			break;
+		}
+	}
+	rcu_read_unlock();
+
+	if (!found) {
+		WARN("osnoise unregister did not find tr %s\n", tr->name);
+		return;
+	}
+
+	synchronize_rcu();
+	kfree(inst);
+}
+
 /*
  * NMI runtime info.
  */
@@ -248,11 +313,6 @@ static struct osnoise_data {
 #endif
 };
 
-/*
- * Boolean variable used to inform that the tracer is currently sampling.
- */
-static bool osnoise_busy;
-
 #ifdef CONFIG_PREEMPT_RT
 /*
  * Print the osnoise header info.
@@ -315,19 +375,24 @@ static void print_osnoise_headers(struct seq_file *s)
  * osnoise_taint - report an osnoise error.
  */
 #define osnoise_taint(msg) ({							\
-	struct trace_array *tr = osnoise_trace;					\
+	struct osnoise_instance *inst;						\
+	struct trace_buffer *buffer;						\
 										\
-	trace_array_printk_buf(tr->array_buffer.buffer, _THIS_IP_, msg);	\
+	rcu_read_lock();							\
+	list_for_each_entry_rcu(inst, &osnoise_instances, list) {		\
+		buffer = inst->tr->array_buffer.buffer;				\
+		trace_array_printk_buf(buffer, _THIS_IP_, msg);			\
+	}									\
+	rcu_read_unlock();							\
 	osnoise_data.tainted = true;						\
 })
 
 /*
  * Record an osnoise_sample into the tracer buffer.
  */
-static void trace_osnoise_sample(struct osnoise_sample *sample)
+static void
+__trace_osnoise_sample(struct osnoise_sample *sample, struct trace_buffer *buffer)
 {
-	struct trace_array *tr = osnoise_trace;
-	struct trace_buffer *buffer = tr->array_buffer.buffer;
 	struct trace_event_call *call = &event_osnoise;
 	struct ring_buffer_event *event;
 	struct osnoise_entry *entry;
@@ -350,6 +415,22 @@ static void trace_osnoise_sample(struct osnoise_sample *sample)
 		trace_buffer_unlock_commit_nostack(buffer, event);
 }
 
+/*
+ * Record an osnoise_sample on all osnoise instances.
+ */
+static void trace_osnoise_sample(struct osnoise_sample *sample)
+{
+	struct osnoise_instance *inst;
+	struct trace_buffer *buffer;
+
+	rcu_read_lock();
+	list_for_each_entry_rcu(inst, &osnoise_instances, list) {
+		buffer = inst->tr->array_buffer.buffer;
+		__trace_osnoise_sample(sample, buffer);
+	}
+	rcu_read_unlock();
+}
+
 #ifdef CONFIG_TIMERLAT_TRACER
 /*
  * Print the timerlat header info.
@@ -387,14 +468,10 @@ static void print_timerlat_headers(struct seq_file *s)
 }
 #endif /* CONFIG_PREEMPT_RT */
 
-/*
- * Record an timerlat_sample into the tracer buffer.
- */
-static void trace_timerlat_sample(struct timerlat_sample *sample)
+static void
+__trace_timerlat_sample(struct timerlat_sample *sample, struct trace_buffer *buffer)
 {
-	struct trace_array *tr = osnoise_trace;
 	struct trace_event_call *call = &event_osnoise;
-	struct trace_buffer *buffer = tr->array_buffer.buffer;
 	struct ring_buffer_event *event;
 	struct timerlat_entry *entry;
 
@@ -411,6 +488,22 @@ static void trace_timerlat_sample(struct timerlat_sample *sample)
 		trace_buffer_unlock_commit_nostack(buffer, event);
 }
 
+/*
+ * Record an timerlat_sample into the tracer buffer.
+ */
+static void trace_timerlat_sample(struct timerlat_sample *sample)
+{
+	struct osnoise_instance *inst;
+	struct trace_buffer *buffer;
+
+	rcu_read_lock();
+	list_for_each_entry_rcu(inst, &osnoise_instances, list) {
+		buffer = inst->tr->array_buffer.buffer;
+		__trace_timerlat_sample(sample, buffer);
+	}
+	rcu_read_unlock();
+}
+
 #ifdef CONFIG_STACKTRACE
 
 #define	MAX_CALLS	256
@@ -450,29 +543,18 @@ static void timerlat_save_stack(int skip)
 	return;
 
 }
-/*
- * timerlat_dump_stack - dump a stack trace previously saved
- *
- * Dump a saved stack trace into the trace buffer.
- */
-static void timerlat_dump_stack(void)
+
+static void
+__timerlat_dump_stack(struct trace_buffer *buffer, struct trace_stack *fstack, unsigned int size)
 {
 	struct trace_event_call *call = &event_osnoise;
-	struct trace_array *tr = osnoise_trace;
-	struct trace_buffer *buffer = tr->array_buffer.buffer;
 	struct ring_buffer_event *event;
-	struct trace_stack *fstack;
 	struct stack_entry *entry;
-	unsigned int size;
-
-	preempt_disable_notrace();
-	fstack = this_cpu_ptr(&trace_stack);
-	size = fstack->stack_size;
 
 	event = trace_buffer_lock_reserve(buffer, TRACE_STACK, sizeof(*entry) + size,
 					  tracing_gen_ctx());
 	if (!event)
-		goto out;
+		return;
 
 	entry = ring_buffer_event_data(event);
 
@@ -481,8 +563,29 @@ static void timerlat_dump_stack(void)
 
 	if (!call_filter_check_discard(call, entry, buffer, event))
 		trace_buffer_unlock_commit_nostack(buffer, event);
+}
 
-out:
+/*
+ * timerlat_dump_stack - dump a stack trace previously saved
+ */
+static void timerlat_dump_stack(void)
+{
+	struct osnoise_instance *inst;
+	struct trace_buffer *buffer;
+	struct trace_stack *fstack;
+	unsigned int size;
+
+	preempt_disable_notrace();
+	fstack = this_cpu_ptr(&trace_stack);
+	size = fstack->stack_size;
+
+	rcu_read_lock();
+	list_for_each_entry_rcu(inst, &osnoise_instances, list) {
+		buffer = inst->tr->array_buffer.buffer;
+		__timerlat_dump_stack(buffer, fstack, size);
+
+	}
+	rcu_read_unlock();
 	preempt_enable_notrace();
 }
 #else
@@ -1077,12 +1180,37 @@ diff_osn_sample_stats(struct osnoise_variables *osn_var, struct osnoise_sample *
  */
 static __always_inline void osnoise_stop_tracing(void)
 {
-	struct trace_array *tr = osnoise_trace;
+	struct osnoise_instance *inst;
+	struct trace_array *tr;
+
+	rcu_read_lock();
+	list_for_each_entry_rcu(inst, &osnoise_instances, list) {
+		tr = inst->tr;
+		trace_array_printk_buf(tr->array_buffer.buffer, _THIS_IP_,
+				"stop tracing hit on cpu %d\n", smp_processor_id());
+
+		tracer_tracing_off(tr);
+	}
+	rcu_read_unlock();
+}
 
-	trace_array_printk_buf(tr->array_buffer.buffer, _THIS_IP_,
-			"stop tracing hit on cpu %d\n", smp_processor_id());
+/*
+ * notify_new_max_latency - Notify a new max latency via fsnotify interface.
+ */
+void notify_new_max_latency(u64 latency)
+{
+	struct osnoise_instance *inst;
+	struct trace_array *tr;
 
-	tracer_tracing_off(tr);
+	rcu_read_lock();
+	list_for_each_entry_rcu(inst, &osnoise_instances, list) {
+		tr = inst->tr;
+		if (tr->max_latency < latency) {
+			tr->max_latency = latency;
+			latency_fsnotify(tr);
+		}
+	}
+	rcu_read_unlock();
 }
 
 /*
@@ -1096,7 +1224,6 @@ static __always_inline void osnoise_stop_tracing(void)
 static int run_osnoise(void)
 {
 	struct osnoise_variables *osn_var = this_cpu_osn_var();
-	struct trace_array *tr = osnoise_trace;
 	u64 start, sample, last_sample;
 	u64 last_int_count, int_count;
 	s64 noise = 0, max_noise = 0;
@@ -1231,11 +1358,7 @@ static int run_osnoise(void)
 
 	trace_osnoise_sample(&s);
 
-	/* Keep a running maximum ever recorded osnoise "latency" */
-	if (max_noise > tr->max_latency) {
-		tr->max_latency = max_noise;
-		latency_fsnotify(tr);
-	}
+	notify_new_max_latency(max_noise);
 
 	if (osnoise_data.stop_tracing_total)
 		if (s.noise > osnoise_data.stop_tracing_total)
@@ -1293,7 +1416,6 @@ static int osnoise_main(void *data)
 static enum hrtimer_restart timerlat_irq(struct hrtimer *timer)
 {
 	struct osnoise_variables *osn_var = this_cpu_osn_var();
-	struct trace_array *tr = osnoise_trace;
 	struct timerlat_variables *tlat;
 	struct timerlat_sample s;
 	u64 now;
@@ -1364,11 +1486,7 @@ static enum hrtimer_restart timerlat_irq(struct hrtimer *timer)
 
 	trace_timerlat_sample(&s);
 
-	/* Keep a running maximum ever recorded os noise "latency" */
-	if (diff > tr->max_latency) {
-		tr->max_latency = diff;
-		latency_fsnotify(tr);
-	}
+	notify_new_max_latency(diff);
 
 	if (osnoise_data.stop_tracing)
 		if (time_to_us(diff) >= osnoise_data.stop_tracing)
@@ -1581,7 +1699,7 @@ static void osnoise_hotplug_workfn(struct work_struct *dummy)
 
 	mutex_lock(&trace_types_lock);
 
-	if (!osnoise_busy)
+	if (!osnoise_has_registered_instances())
 		goto out_unlock_trace;
 
 	mutex_lock(&interface_lock);
@@ -1716,11 +1834,10 @@ osnoise_cpus_write(struct file *filp, const char __user *ubuf, size_t count,
 		goto err_free;
 
 	/*
-	 * trace_types_lock is taken to avoid concurrency on start/stop
-	 * and osnoise_busy.
+	 * trace_types_lock is taken to avoid concurrency on start/stop.
 	 */
 	mutex_lock(&trace_types_lock);
-	running = osnoise_busy;
+	running = osnoise_has_registered_instances();
 	if (running)
 		stop_per_cpu_kthreads();
 
@@ -1941,8 +2058,6 @@ static int osnoise_workload_start(void)
 		return retval;
 	}
 
-	osnoise_busy = true;
-
 	return 0;
 }
 
@@ -1951,7 +2066,7 @@ static int osnoise_workload_start(void)
  */
 static void osnoise_workload_stop(void)
 {
-	if (!osnoise_busy)
+	if (osnoise_has_registered_instances())
 		return;
 
 	trace_osnoise_callback_enabled = false;
@@ -1962,28 +2077,28 @@ static void osnoise_workload_stop(void)
 	unhook_irq_events();
 	unhook_softirq_events();
 	unhook_thread_events();
-
-	osnoise_busy = false;
 }
 
 static void osnoise_tracer_start(struct trace_array *tr)
 {
 	int retval;
 
-	if (osnoise_busy)
+	if (osnoise_has_registered_instances())
 		return;
 
 	retval = osnoise_workload_start();
 	if (retval)
 		pr_err(BANNER "Error starting osnoise tracer\n");
 
+	osnoise_register_instance(tr);
 }
 
 static void osnoise_tracer_stop(struct trace_array *tr)
 {
-	if (!osnoise_busy)
+	if (!osnoise_has_registered_instances())
 		return;
 
+	osnoise_unregister_instance(tr);
 	osnoise_workload_stop();
 }
 
@@ -1991,14 +2106,12 @@ static int osnoise_tracer_init(struct trace_array *tr)
 {
 
 	/* Only allow one instance to enable this */
-	if (osnoise_busy)
+	if (osnoise_has_registered_instances())
 		return -EBUSY;
 
-	osnoise_trace = tr;
 	tr->max_latency = 0;
 
 	osnoise_tracer_start(tr);
-
 	return 0;
 }
 
@@ -2022,7 +2135,7 @@ static void timerlat_tracer_start(struct trace_array *tr)
 {
 	int retval;
 
-	if (osnoise_busy)
+	if (osnoise_has_registered_instances())
 		return;
 
 	osnoise_data.timerlat_tracer = 1;
@@ -2031,6 +2144,8 @@ static void timerlat_tracer_start(struct trace_array *tr)
 	if (retval)
 		goto out_err;
 
+	osnoise_register_instance(tr);
+
 	return;
 out_err:
 	pr_err(BANNER "Error starting timerlat tracer\n");
@@ -2040,7 +2155,7 @@ static void timerlat_tracer_stop(struct trace_array *tr)
 {
 	int cpu;
 
-	if (!osnoise_busy)
+	if (!osnoise_has_registered_instances())
 		return;
 
 	for_each_online_cpu(cpu)
@@ -2054,11 +2169,9 @@ static void timerlat_tracer_stop(struct trace_array *tr)
 static int timerlat_tracer_init(struct trace_array *tr)
 {
 	/* Only allow one instance to enable this */
-	if (osnoise_busy)
+	if (osnoise_has_registered_instances())
 		return -EBUSY;
 
-	osnoise_trace = tr;
-
 	tr->max_latency = 0;
 
 	timerlat_tracer_start(tr);
@@ -2105,6 +2218,8 @@ __init static int init_osnoise_tracer(void)
 #endif
 	osnoise_init_hotplug_support();
 
+	INIT_LIST_HEAD_RCU(&osnoise_instances);
+
 	init_tracefs();
 
 	return 0;
maintainer
#17 lore raw

Re: [PATCH V4 05/19] trace/osnoise: Allow multiple instances of the same tracer

In reply to Daniel Bristot de Oliveira
Quoted 8 lines · On Mon, 25 Oct 2021 10:08:39 +0200 Daniel Bristot de Oliveira…

osnoise and timerlat tracers share a lot of code, including the osnoise: events.

However, the osnoise: events have different behavior depending on which tracer is enabled. So, while this patch allows the same tracer to run in multiple instances (e.g., two instances running osnoise), it still does not allow instances of timerlat and osnoise in parallel (e.g., one timerlat and osnoise).

Should I add the text above to the log?

Yes please.

-- Steve