Merge tag 'trace-tools-v6.5' of git://git.kernel.org/pub/scm/linux/kernel/git/trace/linux-trace

Pull tracing tooling updates from Steven Rostedt:

 - Add cgroup support for rtla via the -C option

 - Add --house-keeping option that tells rtla where to place the
   housekeeping threads

 - Have rtla/timerlat have its own tracing instance instead of using the
   top level tracing instance that is the default for other tracing
   users to use

 - Add auto analysis to timerlat_hist

 - Have rtla start the tracers after creating the instances

 - Reduce rtla hwnoise down to 75% from 100% as it runs with preemption
   disabled and can cause system instability at 100%

 - Add support to run timerlat_top and timerlat_hist threads in
   user-space instead of just using the kernel tasks

 - Some minor clean ups and documentation changes

* tag 'trace-tools-v6.5' of git://git.kernel.org/pub/scm/linux/kernel/git/trace/linux-trace:
  Documentation: Add tools/rtla timerlat -u option documentation
  rtla/timerlat_hist: Add timerlat user-space support
  rtla/timerlat_top: Add timerlat user-space support
  rtla/hwnoise: Reduce runtime to 75%
  rtla: Start the tracers after creating all instances
  rtla/timerlat_hist: Add auto-analysis support
  rtla/timerlat: Give timerlat auto analysis its own instance
  rtla: Automatically move rtla to a house-keeping cpu
  rtla: Change monitored_cpus from char * to cpu_set_t
  rtla: Add --house-keeping option
  rtla: Add -C cgroup support

Documentation/tools/rtla/common_options.rst

@@ -2,6 +2,10 @@
 
         Set the osnoise tracer to run the sample threads in the cpu-list.
 
+**-H**, **--house-keeping** *cpu-list*
+
+        Run rtla control threads only on the given cpu-list.
+
 **-d**, **--duration** *time[s|m|h|d]*
 
         Set the duration of the session.
@@ -42,6 +46,10 @@
         - *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.
 
+**-C**, **--cgroup**\[*=cgroup*]
+
+        Set a *cgroup* to the tracer's threads. If the **-C** option is passed without arguments, the tracer's thread will inherit **rtla**'s *cgroup*. Otherwise, the threads will be placed on the *cgroup* passed to the option.
+
 **-h**, **--help**
 
         Print help menu.

Documentation/tools/rtla/common_timerlat_aa.rst

@@ -5,10 +5,3 @@
 **--no-aa**
 
         disable auto-analysis, reducing rtla timerlat cpu usage
-
-**--aa-only** *us*
-
-        Set stop tracing conditions and run without collecting and displaying statistics.
-        Print the auto-analysis if the system hits the stop tracing condition. This option
-        is useful to reduce rtla timerlat CPU, enabling the debug without the overhead of
-        collecting the statistics.

Documentation/tools/rtla/common_timerlat_options.rst

@@ -26,3 +26,10 @@
         Set the /dev/cpu_dma_latency to *us*, aiming to bound exit from idle latencies.
         *cyclictest* sets this value to *0* by default, use **--dma-latency** *0* to have
         similar results.
+
+**-u**, **--user-threads**
+
+        Set timerlat to run without a workload, and then dispatches user-space workloads
+        to wait on the timerlat_fd. Once the workload is awakes, it goes to sleep again
+        adding so the measurement for the kernel-to-user and user-to-kernel to the tracer
+        output.

Documentation/tools/rtla/rtla-timerlat-hist.rst

@@ -29,15 +29,18 @@ OPTIONS
 
 .. include:: common_options.rst
 
+.. include:: common_timerlat_aa.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 *timerlat* threads to run under
 *SCHED_DEADLINE* priority, with a *10us* runtime every *1ms* period. The
-*1ms* period is also passed to the *timerlat* tracer::
+*1ms* period is also passed to the *timerlat* tracer. Auto-analysis is disabled
+to reduce overhead ::
 
-  [root@alien ~]# timerlat hist -d 10m -c 0-4 -P d:100us:1ms -p 1ms
+  [root@alien ~]# timerlat hist -d 10m -c 0-4 -P d:100us:1ms -p 1ms --no-aa
   # RTLA timerlat histogram
   # Time unit is microseconds (us)
   # Duration:   0 00:10:00

Documentation/tools/rtla/rtla-timerlat-top.rst

@@ -32,6 +32,13 @@ OPTIONS
 
 .. include:: common_timerlat_aa.rst
 
+**--aa-only** *us*
+
+        Set stop tracing conditions and run without collecting and displaying statistics.
+        Print the auto-analysis if the system hits the stop tracing condition. This option
+        is useful to reduce rtla timerlat CPU, enabling the debug without the overhead of
+        collecting the statistics.
+
 EXAMPLE
 =======
 

tools/tracing/rtla/src/osnoise.c

@@ -841,6 +841,67 @@ static void osnoise_put_irq_disable(struct osnoise_context *context)
 	context->orig_opt_irq_disable = OSNOISE_OPTION_INIT_VAL;
 }
 
+static int osnoise_get_workload(struct osnoise_context *context)
+{
+	if (context->opt_workload != OSNOISE_OPTION_INIT_VAL)
+		return context->opt_workload;
+
+	if (context->orig_opt_workload != OSNOISE_OPTION_INIT_VAL)
+		return context->orig_opt_workload;
+
+	context->orig_opt_workload = osnoise_options_get_option("OSNOISE_WORKLOAD");
+
+	return context->orig_opt_workload;
+}
+
+int osnoise_set_workload(struct osnoise_context *context, bool onoff)
+{
+	int opt_workload = osnoise_get_workload(context);
+	int retval;
+
+	if (opt_workload == OSNOISE_OPTION_INIT_VAL)
+		return -1;
+
+	if (opt_workload == onoff)
+		return 0;
+
+	retval = osnoise_options_set_option("OSNOISE_WORKLOAD", onoff);
+	if (retval < 0)
+		return -1;
+
+	context->opt_workload = onoff;
+
+	return 0;
+}
+
+static void osnoise_restore_workload(struct osnoise_context *context)
+{
+	int retval;
+
+	if (context->orig_opt_workload == OSNOISE_OPTION_INIT_VAL)
+		return;
+
+	if (context->orig_opt_workload == context->opt_workload)
+		goto out_done;
+
+	retval = osnoise_options_set_option("OSNOISE_WORKLOAD", context->orig_opt_workload);
+	if (retval < 0)
+		err_msg("Could not restore original OSNOISE_WORKLOAD option\n");
+
+out_done:
+	context->orig_opt_workload = OSNOISE_OPTION_INIT_VAL;
+}
+
+static void osnoise_put_workload(struct osnoise_context *context)
+{
+	osnoise_restore_workload(context);
+
+	if (context->orig_opt_workload == OSNOISE_OPTION_INIT_VAL)
+		return;
+
+	context->orig_opt_workload = OSNOISE_OPTION_INIT_VAL;
+}
+
 /*
  * enable_osnoise - enable osnoise tracer in the trace_instance
  */
@@ -908,6 +969,9 @@ struct osnoise_context *osnoise_context_alloc(void)
 	context->orig_opt_irq_disable	= OSNOISE_OPTION_INIT_VAL;
 	context->opt_irq_disable	= OSNOISE_OPTION_INIT_VAL;
 
+	context->orig_opt_workload	= OSNOISE_OPTION_INIT_VAL;
+	context->opt_workload		= OSNOISE_OPTION_INIT_VAL;
+
 	osnoise_get_context(context);
 
 	return context;
@@ -935,6 +999,7 @@ void osnoise_put_context(struct osnoise_context *context)
 	osnoise_put_print_stack(context);
 	osnoise_put_tracing_thresh(context);
 	osnoise_put_irq_disable(context);
+	osnoise_put_workload(context);
 
 	free(context);
 }

tools/tracing/rtla/src/osnoise.h

@@ -42,6 +42,10 @@ struct osnoise_context {
 	/* -1 as init value because 0 is off */
 	int			orig_opt_irq_disable;
 	int			opt_irq_disable;
+
+	/* -1 as init value because 0 is off */
+	int			orig_opt_workload;
+	int			opt_workload;
 };
 
 /*
@@ -84,6 +88,7 @@ int osnoise_set_print_stack(struct osnoise_context *context,
 			    long long print_stack);
 
 int osnoise_set_irq_disable(struct osnoise_context *context, bool onoff);
+int osnoise_set_workload(struct osnoise_context *context, bool onoff);
 
 /*
  * osnoise_tool -  osnoise based tool definition.

tools/tracing/rtla/src/osnoise_hist.c

@@ -3,6 +3,7 @@
  * Copyright (C) 2021 Red Hat Inc, Daniel Bristot de Oliveira <bristot@kernel.org>
  */
 
+#define _GNU_SOURCE
 #include <getopt.h>
 #include <stdlib.h>
 #include <string.h>
@@ -11,14 +12,16 @@
 #include <errno.h>
 #include <stdio.h>
 #include <time.h>
+#include <sched.h>
 
 #include "utils.h"
 #include "osnoise.h"
 
 struct osnoise_hist_params {
 	char			*cpus;
-	char			*monitored_cpus;
+	cpu_set_t		monitored_cpus;
 	char			*trace_output;
+	char			*cgroup_name;
 	unsigned long long	runtime;
 	unsigned long long	period;
 	long long		threshold;
@@ -28,6 +31,9 @@ struct osnoise_hist_params {
 	int			duration;
 	int			set_sched;
 	int			output_divisor;
+	int			cgroup;
+	int			hk_cpus;
+	cpu_set_t		hk_cpu_set;
 	struct sched_attr	sched_param;
 	struct trace_events	*events;
 
@@ -268,7 +274,7 @@ static void osnoise_hist_header(struct osnoise_tool *tool)
 		trace_seq_printf(s, "Index");
 
 	for (cpu = 0; cpu < data->nr_cpus; cpu++) {
-		if (params->cpus && !params->monitored_cpus[cpu])
+		if (params->cpus && !CPU_ISSET(cpu, &params->monitored_cpus))
 			continue;
 
 		if (!data->hist[cpu].count)
@@ -299,7 +305,7 @@ osnoise_print_summary(struct osnoise_hist_params *params,
 		trace_seq_printf(trace->seq, "count:");
 
 	for (cpu = 0; cpu < data->nr_cpus; cpu++) {
-		if (params->cpus && !params->monitored_cpus[cpu])
+		if (params->cpus && !CPU_ISSET(cpu, &params->monitored_cpus))
 			continue;
 
 		if (!data->hist[cpu].count)
@@ -313,7 +319,7 @@ osnoise_print_summary(struct osnoise_hist_params *params,
 		trace_seq_printf(trace->seq, "min:  ");
 
 	for (cpu = 0; cpu < data->nr_cpus; cpu++) {
-		if (params->cpus && !params->monitored_cpus[cpu])
+		if (params->cpus && !CPU_ISSET(cpu, &params->monitored_cpus))
 			continue;
 
 		if (!data->hist[cpu].count)
@@ -328,7 +334,7 @@ osnoise_print_summary(struct osnoise_hist_params *params,
 		trace_seq_printf(trace->seq, "avg:  ");
 
 	for (cpu = 0; cpu < data->nr_cpus; cpu++) {
-		if (params->cpus && !params->monitored_cpus[cpu])
+		if (params->cpus && !CPU_ISSET(cpu, &params->monitored_cpus))
 			continue;
 
 		if (!data->hist[cpu].count)
@@ -346,7 +352,7 @@ osnoise_print_summary(struct osnoise_hist_params *params,
 		trace_seq_printf(trace->seq, "max:  ");
 
 	for (cpu = 0; cpu < data->nr_cpus; cpu++) {
-		if (params->cpus && !params->monitored_cpus[cpu])
+		if (params->cpus && !CPU_ISSET(cpu, &params->monitored_cpus))
 			continue;
 
 		if (!data->hist[cpu].count)
@@ -381,7 +387,7 @@ osnoise_print_stats(struct osnoise_hist_params *params, struct osnoise_tool *too
 					 bucket * data->bucket_size);
 
 		for (cpu = 0; cpu < data->nr_cpus; cpu++) {
-			if (params->cpus && !params->monitored_cpus[cpu])
+			if (params->cpus && !CPU_ISSET(cpu, &params->monitored_cpus))
 				continue;
 
 			if (!data->hist[cpu].count)
@@ -405,7 +411,7 @@ osnoise_print_stats(struct osnoise_hist_params *params, struct osnoise_tool *too
 		trace_seq_printf(trace->seq, "over: ");
 
 	for (cpu = 0; cpu < data->nr_cpus; cpu++) {
-		if (params->cpus && !params->monitored_cpus[cpu])
+		if (params->cpus && !CPU_ISSET(cpu, &params->monitored_cpus))
 			continue;
 
 		if (!data->hist[cpu].count)
@@ -432,8 +438,8 @@ static void osnoise_hist_usage(char *usage)
 		"",
 		"  usage: rtla osnoise hist [-h] [-D] [-d s] [-a us] [-p us] [-r us] [-s us] [-S us] \\",
 		"	  [-T us] [-t[=file]] [-e sys[:event]] [--filter <filter>] [--trigger <trigger>] \\",
-		"	  [-c cpu-list] [-P priority] [-b N] [-E N] [--no-header] [--no-summary] [--no-index] \\",
-		"	  [--with-zeros]",
+		"	  [-c cpu-list] [-H cpu-list] [-P priority] [-b N] [-E N] [--no-header] [--no-summary] \\",
+		"	  [--no-index] [--with-zeros] [-C[=cgroup_name]]",
 		"",
 		"	  -h/--help: print this menu",
 		"	  -a/--auto: set automatic trace mode, stopping the session if argument in us sample is hit",
@@ -443,6 +449,8 @@ static void osnoise_hist_usage(char *usage)
 		"	  -S/--stop-total us: stop trace if the total sample is higher than the argument in us",
 		"	  -T/--threshold us: the minimum delta to be considered a noise",
 		"	  -c/--cpus cpu-list: list of cpus to run osnoise threads",
+		"	  -H/--house-keeping cpus: run rtla control threads only on the given cpus",
+		"	  -C/--cgroup[=cgroup_name]: set cgroup, if no cgroup_name is passed, the rtla's cgroup will be inherited",
 		"	  -d/--duration time[s|m|h|d]: duration of the session",
 		"	  -D/--debug: print debug info",
 		"	  -t/--trace[=file]: save the stopped trace to [file|osnoise_trace.txt]",
@@ -501,8 +509,10 @@ static struct osnoise_hist_params
 			{"bucket-size",		required_argument,	0, 'b'},
 			{"entries",		required_argument,	0, 'E'},
 			{"cpus",		required_argument,	0, 'c'},
+			{"cgroup",		optional_argument,	0, 'C'},
 			{"debug",		no_argument,		0, 'D'},
 			{"duration",		required_argument,	0, 'd'},
+			{"house-keeping",	required_argument,		0, 'H'},
 			{"help",		no_argument,		0, 'h'},
 			{"period",		required_argument,	0, 'p'},
 			{"priority",		required_argument,	0, 'P'},
@@ -524,7 +534,7 @@ static struct osnoise_hist_params
 		/* getopt_long stores the option index here. */
 		int option_index = 0;
 
-		c = getopt_long(argc, argv, "a:c:b:d:e:E:Dhp:P:r:s:S:t::T:01234:5:",
+		c = getopt_long(argc, argv, "a:c:C::b:d:e:E:DhH:p:P:r:s:S:t::T:01234:5:",
 				 long_options, &option_index);
 
 		/* detect the end of the options. */
@@ -549,11 +559,21 @@ static struct osnoise_hist_params
 				osnoise_hist_usage("Bucket size needs to be > 0 and <= 1000000\n");
 			break;
 		case 'c':
-			retval = parse_cpu_list(optarg, &params->monitored_cpus);
+			retval = parse_cpu_set(optarg, &params->monitored_cpus);
 			if (retval)
 				osnoise_hist_usage("\nInvalid -c cpu list\n");
 			params->cpus = optarg;
 			break;
+		case 'C':
+			params->cgroup = 1;
+			if (!optarg) {
+				/* will inherit this cgroup */
+				params->cgroup_name = NULL;
+			} else if (*optarg == '=') {
+				/* skip the = */
+				params->cgroup_name = ++optarg;
+			}
+			break;
 		case 'D':
 			config_debug = 1;
 			break;
@@ -583,6 +603,14 @@ static struct osnoise_hist_params
 		case '?':
 			osnoise_hist_usage(NULL);
 			break;
+		case 'H':
+			params->hk_cpus = 1;
+			retval = parse_cpu_set(optarg, &params->hk_cpu_set);
+			if (retval) {
+				err_msg("Error parsing house keeping CPUs\n");
+				exit(EXIT_FAILURE);
+			}
+			break;
 		case 'p':
 			params->period = get_llong_from_str(optarg);
 			if (params->period > 10000000)
@@ -718,6 +746,24 @@ osnoise_hist_apply_config(struct osnoise_tool *tool, struct osnoise_hist_params
 		}
 	}
 
+	if (params->hk_cpus) {
+		retval = sched_setaffinity(getpid(), sizeof(params->hk_cpu_set),
+					   &params->hk_cpu_set);
+		if (retval == -1) {
+			err_msg("Failed to set rtla to the house keeping CPUs\n");
+			goto out_err;
+		}
+	} else if (params->cpus) {
+		/*
+		 * Even if the user do not set a house-keeping CPU, try to
+		 * move rtla to a CPU set different to the one where the user
+		 * set the workload to run.
+		 *
+		 * No need to check results as this is an automatic attempt.
+		 */
+		auto_house_keeping(&params->monitored_cpus);
+	}
+
 	return 0;
 
 out_err:
@@ -816,7 +862,13 @@ int osnoise_hist_main(int argc, char *argv[])
 		}
 	}
 
-	trace_instance_start(trace);
+	if (params->cgroup) {
+		retval = set_comm_cgroup("timerlat/", params->cgroup_name);
+		if (!retval) {
+			err_msg("Failed to move threads to cgroup\n");
+			goto out_free;
+		}
+	}
 
 	if (params->trace_output) {
 		record = osnoise_init_trace_tool("osnoise");
@@ -831,9 +883,19 @@ int osnoise_hist_main(int argc, char *argv[])
 				goto out_hist;
 		}
 
-		trace_instance_start(&record->trace);
 	}
 
+	/*
+	 * Start the tracer here, after having set all instances.
+	 *
+	 * Let the trace instance start first for the case of hitting a stop
+	 * tracing while enabling other instances. The trace instance is the
+	 * one with most valuable information.
+	 */
+	if (params->trace_output)
+		trace_instance_start(&record->trace);
+	trace_instance_start(trace);
+
 	tool->start_time = time(NULL);
 	osnoise_hist_set_signals(params);
 

tools/tracing/rtla/src/osnoise_top.c

@@ -3,6 +3,7 @@
  * Copyright (C) 2021 Red Hat Inc, Daniel Bristot de Oliveira <bristot@kernel.org>
  */
 
+#define _GNU_SOURCE
 #include <getopt.h>
 #include <stdlib.h>
 #include <string.h>
@@ -10,6 +11,7 @@
 #include <unistd.h>
 #include <stdio.h>
 #include <time.h>
+#include <sched.h>
 
 #include "osnoise.h"
 #include "utils.h"
@@ -24,8 +26,9 @@ enum osnoise_mode {
  */
 struct osnoise_top_params {
 	char			*cpus;
-	char			*monitored_cpus;
+	cpu_set_t		monitored_cpus;
 	char			*trace_output;
+	char			*cgroup_name;
 	unsigned long long	runtime;
 	unsigned long long	period;
 	long long		threshold;
@@ -35,6 +38,9 @@ struct osnoise_top_params {
 	int			duration;
 	int			quiet;
 	int			set_sched;
+	int			cgroup;
+	int			hk_cpus;
+	cpu_set_t		hk_cpu_set;
 	struct sched_attr	sched_param;
 	struct trace_events	*events;
 	enum osnoise_mode	mode;
@@ -257,7 +263,7 @@ osnoise_print_stats(struct osnoise_top_params *params, struct osnoise_tool *top)
 	osnoise_top_header(top);
 
 	for (i = 0; i < nr_cpus; i++) {
-		if (params->cpus && !params->monitored_cpus[i])
+		if (params->cpus && !CPU_ISSET(i, &params->monitored_cpus))
 			continue;
 		osnoise_top_print(top, i);
 	}
@@ -276,7 +282,7 @@ static void osnoise_top_usage(struct osnoise_top_params *params, char *usage)
 	static const char * const msg[] = {
 		" [-h] [-q] [-D] [-d s] [-a us] [-p us] [-r us] [-s us] [-S us] \\",
 		"	  [-T us] [-t[=file]] [-e sys[:event]] [--filter <filter>] [--trigger <trigger>] \\",
-		"	  [-c cpu-list] [-P priority]",
+		"	  [-c cpu-list] [-H cpu-list] [-P priority] [-C[=cgroup_name]]",
 		"",
 		"	  -h/--help: print this menu",
 		"	  -a/--auto: set automatic trace mode, stopping the session if argument in us sample is hit",
@@ -286,6 +292,8 @@ static void osnoise_top_usage(struct osnoise_top_params *params, char *usage)
 		"	  -S/--stop-total us: stop trace if the total sample is higher than the argument in us",
 		"	  -T/--threshold us: the minimum delta to be considered a noise",
 		"	  -c/--cpus cpu-list: list of cpus to run osnoise threads",
+		"	  -H/--house-keeping cpus: run rtla control threads only on the given cpus",
+		"	  -C/--cgroup[=cgroup_name]: set cgroup, if no cgroup_name is passed, the rtla's cgroup will be inherited",
 		"	  -d/--duration time[s|m|h|d]: duration of the session",
 		"	  -D/--debug: print debug info",
 		"	  -t/--trace[=file]: save the stopped trace to [file|osnoise_trace.txt]",
@@ -340,16 +348,24 @@ struct osnoise_top_params *osnoise_top_parse_args(int argc, char **argv)
 	if (!params)
 		exit(1);
 
-	if (strcmp(argv[0], "hwnoise") == 0)
+	if (strcmp(argv[0], "hwnoise") == 0) {
 		params->mode = MODE_HWNOISE;
+		/*
+		 * Reduce CPU usage for 75% to avoid killing the system.
+		 */
+		params->runtime = 750000;
+		params->period = 1000000;
+	}
 
 	while (1) {
 		static struct option long_options[] = {
 			{"auto",		required_argument,	0, 'a'},
 			{"cpus",		required_argument,	0, 'c'},
+			{"cgroup",		optional_argument,	0, 'C'},
 			{"debug",		no_argument,		0, 'D'},
 			{"duration",		required_argument,	0, 'd'},
 			{"event",		required_argument,	0, 'e'},
+			{"house-keeping",	required_argument,	0, 'H'},
 			{"help",		no_argument,		0, 'h'},
 			{"period",		required_argument,	0, 'p'},
 			{"priority",		required_argument,	0, 'P'},
@@ -367,7 +383,7 @@ struct osnoise_top_params *osnoise_top_parse_args(int argc, char **argv)
 		/* getopt_long stores the option index here. */
 		int option_index = 0;
 
-		c = getopt_long(argc, argv, "a:c:d:De:hp:P:qr:s:S:t::T:0:1:",
+		c = getopt_long(argc, argv, "a:c:C::d:De:hH:p:P:qr:s:S:t::T:0:1:",
 				 long_options, &option_index);
 
 		/* Detect the end of the options. */
@@ -387,11 +403,21 @@ struct osnoise_top_params *osnoise_top_parse_args(int argc, char **argv)
 
 			break;
 		case 'c':
-			retval = parse_cpu_list(optarg, &params->monitored_cpus);
+			retval = parse_cpu_set(optarg, &params->monitored_cpus);
 			if (retval)
 				osnoise_top_usage(params, "\nInvalid -c cpu list\n");
 			params->cpus = optarg;
 			break;
+		case 'C':
+			params->cgroup = 1;
+			if (!optarg) {
+				/* will inherit this cgroup */
+				params->cgroup_name = NULL;
+			} else if (*optarg == '=') {
+				/* skip the = */
+				params->cgroup_name = ++optarg;
+			}
+			break;
 		case 'D':
 			config_debug = 1;
 			break;
@@ -416,6 +442,14 @@ struct osnoise_top_params *osnoise_top_parse_args(int argc, char **argv)
 		case '?':
 			osnoise_top_usage(params, NULL);
 			break;
+		case 'H':
+			params->hk_cpus = 1;
+			retval = parse_cpu_set(optarg, &params->hk_cpu_set);
+			if (retval) {
+				err_msg("Error parsing house keeping CPUs\n");
+				exit(EXIT_FAILURE);
+			}
+			break;
 		case 'p':
 			params->period = get_llong_from_str(optarg);
 			if (params->period > 10000000)
@@ -547,6 +581,24 @@ osnoise_top_apply_config(struct osnoise_tool *tool, struct osnoise_top_params *p
 		}
 	}
 
+	if (params->hk_cpus) {
+		retval = sched_setaffinity(getpid(), sizeof(params->hk_cpu_set),
+					   &params->hk_cpu_set);
+		if (retval == -1) {
+			err_msg("Failed to set rtla to the house keeping CPUs\n");
+			goto out_err;
+		}
+	} else if (params->cpus) {
+		/*
+		 * Even if the user do not set a house-keeping CPU, try to
+		 * move rtla to a CPU set different to the one where the user
+		 * set the workload to run.
+		 *
+		 * No need to check results as this is an automatic attempt.
+		 */
+		auto_house_keeping(&params->monitored_cpus);
+	}
+
 	return 0;
 
 out_err:
@@ -643,7 +695,13 @@ int osnoise_top_main(int argc, char **argv)
 		}
 	}
 
-	trace_instance_start(trace);
+	if (params->cgroup) {
+		retval = set_comm_cgroup("osnoise/", params->cgroup_name);
+		if (!retval) {
+			err_msg("Failed to move threads to cgroup\n");
+			goto out_free;
+		}
+	}
 
 	if (params->trace_output) {
 		record = osnoise_init_trace_tool("osnoise");
@@ -657,9 +715,18 @@ int osnoise_top_main(int argc, char **argv)
 			if (retval)
 				goto out_top;
 		}
+	}
 
+	/*
+	 * Start the tracer here, after having set all instances.
+	 *
+	 * Let the trace instance start first for the case of hitting a stop
+	 * tracing while enabling other instances. The trace instance is the
+	 * one with most valuable information.
+	 */
+	if (params->trace_output)
 		trace_instance_start(&record->trace);
-	}
+	trace_instance_start(trace);
 
 	tool->start_time = time(NULL);
 	osnoise_top_set_signals(params);

tools/tracing/rtla/src/timerlat_aa.c

@@ -8,6 +8,7 @@
 #include "utils.h"
 #include "osnoise.h"
 #include "timerlat.h"
+#include <unistd.h>
 
 enum timelat_state {
 	TIMERLAT_INIT = 0,
@@ -233,7 +234,7 @@ static int timerlat_aa_thread_latency(struct timerlat_aa_data *taa_data,
  *
  * Returns 0 on success, -1 otherwise.
  */
-int timerlat_aa_handler(struct trace_seq *s, struct tep_record *record,
+static int timerlat_aa_handler(struct trace_seq *s, struct tep_record *record,
 			struct tep_event *event, void *context)
 {
 	struct timerlat_aa_context *taa_ctx = timerlat_aa_get_ctx();
@@ -665,6 +666,25 @@ static void timerlat_thread_analysis(struct timerlat_aa_data *taa_data, int cpu,
 		ns_to_usf(total));
 }
 
+static int timerlat_auto_analysis_collect_trace(struct timerlat_aa_context *taa_ctx)
+{
+	struct trace_instance *trace = &taa_ctx->tool->trace;
+	int retval;
+
+	retval = tracefs_iterate_raw_events(trace->tep,
+					    trace->inst,
+					    NULL,
+					    0,
+					    collect_registered_events,
+					    trace);
+		if (retval < 0) {
+			err_msg("Error iterating on events\n");
+			return 0;
+		}
+
+	return 1;
+}
+
 /**
  * timerlat_auto_analysis - Analyze the collected data
  */
@@ -677,6 +697,8 @@ void timerlat_auto_analysis(int irq_thresh, int thread_thresh)
 	struct tep_handle *tep;
 	int cpu;
 
+	timerlat_auto_analysis_collect_trace(taa_ctx);
+
 	/* bring stop tracing to the ns scale */
 	irq_thresh = irq_thresh * 1000;
 	thread_thresh = thread_thresh * 1000;
@@ -838,6 +860,10 @@ static int timerlat_aa_init_seqs(struct timerlat_aa_context *taa_ctx)
  */
 static void timerlat_aa_unregister_events(struct osnoise_tool *tool, int dump_tasks)
 {
+
+	tep_unregister_event_handler(tool->trace.tep, -1, "ftrace", "timerlat",
+				     timerlat_aa_handler, tool);
+
 	tracefs_event_disable(tool->trace.inst, "osnoise", NULL);
 
 	tep_unregister_event_handler(tool->trace.tep, -1, "osnoise", "nmi_noise",
@@ -875,6 +901,10 @@ static int timerlat_aa_register_events(struct osnoise_tool *tool, int dump_tasks
 {
 	int retval;
 
+	tep_register_event_handler(tool->trace.tep, -1, "ftrace", "timerlat",
+				timerlat_aa_handler, tool);
+
+
 	/*
 	 * register auto-analysis handlers.
 	 */
@@ -955,8 +985,9 @@ void timerlat_aa_destroy(void)
  *
  * Returns 0 on success, -1 otherwise.
  */
-int timerlat_aa_init(struct osnoise_tool *tool, int nr_cpus, int dump_tasks)
+int timerlat_aa_init(struct osnoise_tool *tool, int dump_tasks)
 {
+	int nr_cpus = sysconf(_SC_NPROCESSORS_CONF);
 	struct timerlat_aa_context *taa_ctx;
 	int retval;
 

tools/tracing/rtla/src/timerlat_aa.h

@@ -3,10 +3,7 @@
  * Copyright (C) 2023 Red Hat Inc, Daniel Bristot de Oliveira <bristot@kernel.org>
  */
 
-int timerlat_aa_init(struct osnoise_tool *tool, int nr_cpus, int dump_task);
+int timerlat_aa_init(struct osnoise_tool *tool, int dump_task);
 void timerlat_aa_destroy(void);
 
-int timerlat_aa_handler(struct trace_seq *s, struct tep_record *record,
-			struct tep_event *event, void *context);
-
 void timerlat_auto_analysis(int irq_thresh, int thread_thresh);

tools/tracing/rtla/src/timerlat_u.c

+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Copyright (C) 2023 Red Hat Inc, Daniel Bristot de Oliveira <bristot@kernel.org>
+ */
+
+#define _GNU_SOURCE
+#include <sched.h>
+#include <fcntl.h>
+#include <stdlib.h>
+#include <unistd.h>
+#include <stdio.h>
+#include <errno.h>
+#include <string.h>
+#include <tracefs.h>
+#include <pthread.h>
+#include <sys/wait.h>
+#include <sys/prctl.h>
+
+#include "utils.h"
+#include "timerlat_u.h"
+
+/*
+ * This is the user-space main for the tool timerlatu/ threads.
+ *
+ * It is as simple as this:
+ *  - set affinity
+ *  - set priority
+ *  - open tracer fd
+ *  - spin
+ *  - close
+ */
+static int timerlat_u_main(int cpu, struct timerlat_u_params *params)
+{
+	struct sched_param sp = { .sched_priority = 95 };
+	char buffer[1024];
+	int timerlat_fd;
+	cpu_set_t set;
+	int retval;
+
+	/*
+	 * This all is only setting up the tool.
+	 */
+	CPU_ZERO(&set);
+	CPU_SET(cpu, &set);
+
+	retval = sched_setaffinity(gettid(), sizeof(set), &set);
+	if (retval == -1) {
+		err_msg("Error setting user thread affinity\n");
+		exit(1);
+	}
+
+	if (!params->sched_param) {
+		retval = sched_setscheduler(0, SCHED_FIFO, &sp);
+		if (retval < 0) {
+			err_msg("Error setting timerlat u default priority: %s\n", strerror(errno));
+			exit(1);
+		}
+	} else {
+		retval = __set_sched_attr(getpid(), params->sched_param);
+		if (retval) {
+			/* __set_sched_attr prints an error message, so */
+			exit(0);
+		}
+	}
+
+	if (params->cgroup_name) {
+		retval = set_pid_cgroup(gettid(), params->cgroup_name);
+		if (!retval) {
+			err_msg("Error setting timerlat u cgroup pid\n");
+			pthread_exit(&retval);
+		}
+	}
+
+	/*
+	 * This is the tool's loop. If you want to use as base for your own tool...
+	 * go ahead.
+	 */
+	snprintf(buffer, sizeof(buffer), "osnoise/per_cpu/cpu%d/timerlat_fd", cpu);
+
+	timerlat_fd = tracefs_instance_file_open(NULL, buffer, O_RDONLY);
+	if (timerlat_fd < 0) {
+		err_msg("Error opening %s:%s\n", buffer, strerror(errno));
+		exit(1);
+	}
+
+	debug_msg("User-space timerlat pid %d on cpu %d\n", gettid(), cpu);
+
+	/* add should continue with a signal handler */
+	while (true) {
+		retval = read(timerlat_fd, buffer, 1024);
+		if (retval < 0)
+			break;
+	}
+
+	close(timerlat_fd);
+
+	debug_msg("Leaving timerlat pid %d on cpu %d\n", gettid(), cpu);
+	exit(0);
+}
+
+/*
+ * timerlat_u_send_kill - send a kill signal for all processes
+ *
+ * Return the number of processes that received the kill.
+ */
+static int timerlat_u_send_kill(pid_t *procs, int nr_cpus)
+{
+	int killed = 0;
+	int i, retval;
+
+	for (i = 0; i < nr_cpus; i++) {
+		if (!procs[i])
+			continue;
+		retval = kill(procs[i], SIGKILL);
+		if (!retval)
+			killed++;
+		else
+			err_msg("Error killing child process %d\n", procs[i]);
+	}
+
+	return killed;
+}
+
+/**
+ * timerlat_u_dispatcher - dispatch one timerlatu/ process per monitored CPU
+ *
+ * This is a thread main that will fork one new process for each monitored
+ * CPU. It will wait for:
+ *
+ *  - rtla to tell to kill the child processes
+ *  - some child process to die, and the cleanup all the processes
+ *
+ * whichever comes first.
+ *
+ */
+void *timerlat_u_dispatcher(void *data)
+{
+	int nr_cpus = sysconf(_SC_NPROCESSORS_CONF);
+	struct timerlat_u_params *params = data;
+	char proc_name[128];
+	int procs_count = 0;
+	int retval = 1;
+	pid_t *procs;
+	int wstatus;
+	pid_t pid;
+	int i;
+
+	debug_msg("Dispatching timerlat u procs\n");
+
+	procs = calloc(nr_cpus, sizeof(pid_t));
+	if (!procs)
+		pthread_exit(&retval);
+
+	for (i = 0; i < nr_cpus; i++) {
+		if (params->set && !CPU_ISSET(i, params->set))
+			continue;
+
+		pid = fork();
+
+		/* child */
+		if (!pid) {
+
+			/*
+			 * rename the process
+			 */
+			snprintf(proc_name, sizeof(proc_name), "timerlatu/%d", i);
+			pthread_setname_np(pthread_self(), proc_name);
+			prctl(PR_SET_NAME, (unsigned long)proc_name, 0, 0, 0);
+
+			timerlat_u_main(i, params);
+			/* timerlat_u_main should exit()! Anyways... */
+			pthread_exit(&retval);
+		}
+
+		/* parent */
+		if (pid == -1) {
+			timerlat_u_send_kill(procs, nr_cpus);
+			debug_msg("Failed to create child processes");
+			pthread_exit(&retval);
+		}
+
+		procs_count++;
+		procs[i] = pid;
+	}
+
+	while (params->should_run) {
+		/* check if processes died */
+		pid = waitpid(-1, &wstatus, WNOHANG);
+		if (pid != 0) {
+			for (i = 0; i < nr_cpus; i++) {
+				if (procs[i] == pid) {
+					procs[i] = 0;
+					procs_count--;
+				}
+			}
+			break;
+		}
+
+		sleep(1);
+	}
+
+	timerlat_u_send_kill(procs, nr_cpus);
+
+	while (procs_count) {
+		pid = waitpid(-1, &wstatus, 0);
+		if (pid == -1) {
+			err_msg("Failed to monitor child processes");
+			pthread_exit(&retval);
+		}
+		for (i = 0; i < nr_cpus; i++) {
+			if (procs[i] == pid) {
+				procs[i] = 0;
+				procs_count--;
+			}
+		}
+	}
+
+	params->stopped_running = 1;
+
+	free(procs);
+	retval = 0;
+	pthread_exit(&retval);
+
+}

tools/tracing/rtla/src/timerlat_u.h

+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Copyright (C) 2023 Red Hat Inc, Daniel Bristot de Oliveira <bristot@kernel.org>
+ */
+
+struct timerlat_u_params {
+	/* timerlat -> timerlat_u: user-space threads can keep running */
+	int should_run;
+	/* timerlat_u -> timerlat: all timerlat_u threads left, no reason to continue */
+	int stopped_running;
+
+	/* threads config */
+	cpu_set_t *set;
+	char *cgroup_name;
+	struct sched_attr *sched_param;
+};
+
+void *timerlat_u_dispatcher(void *data);

tools/tracing/rtla/src/utils.c

@@ -3,6 +3,7 @@
  * Copyright (C) 2021 Red Hat Inc, Daniel Bristot de Oliveira <bristot@kernel.org>
  */
 
+#define _GNU_SOURCE
 #include <dirent.h>
 #include <stdarg.h>
 #include <stdlib.h>
@@ -88,27 +89,24 @@ void get_duration(time_t start_time, char *output, int output_size)
 }
 
 /*
- * parse_cpu_list - parse a cpu_list filling a char vector with cpus set
+ * parse_cpu_set - parse a cpu_list filling cpu_set_t argument
  *
- * Receives a cpu list, like 1-3,5 (cpus 1, 2, 3, 5), and then set the char
- * in the monitored_cpus.
+ * Receives a cpu list, like 1-3,5 (cpus 1, 2, 3, 5), and then set
+ * filling cpu_set_t argument.
  *
- * XXX: convert to a bitmask.
+ * Returns 1 on success, 0 otherwise.
  */
-int parse_cpu_list(char *cpu_list, char **monitored_cpus)
+int parse_cpu_set(char *cpu_list, cpu_set_t *set)
 {
-	char *mon_cpus;
 	const char *p;
 	int end_cpu;
 	int nr_cpus;
 	int cpu;
 	int i;
 
-	nr_cpus = sysconf(_SC_NPROCESSORS_CONF);
+	CPU_ZERO(set);
 
-	mon_cpus = calloc(nr_cpus, sizeof(char));
-	if (!mon_cpus)
-		goto err;
+	nr_cpus = sysconf(_SC_NPROCESSORS_CONF);
 
 	for (p = cpu_list; *p; ) {
 		cpu = atoi(p);
@@ -128,12 +126,12 @@ int parse_cpu_list(char *cpu_list, char **monitored_cpus)
 			end_cpu = cpu;
 
 		if (cpu == end_cpu) {
-			debug_msg("cpu_list: adding cpu %d\n", cpu);
-			mon_cpus[cpu] = 1;
+			debug_msg("cpu_set: adding cpu %d\n", cpu);
+			CPU_SET(cpu, set);
 		} else {
 			for (i = cpu; i <= end_cpu; i++) {
-				debug_msg("cpu_list: adding cpu %d\n", i);
-				mon_cpus[i] = 1;
+				debug_msg("cpu_set: adding cpu %d\n", i);
+				CPU_SET(i, set);
 			}
 		}
 
@@ -141,12 +139,9 @@ int parse_cpu_list(char *cpu_list, char **monitored_cpus)
 			p++;
 	}
 
-	*monitored_cpus = mon_cpus;
-
 	return 0;
-
 err:
-	debug_msg("Error parsing the cpu list %s", cpu_list);
+	debug_msg("Error parsing the cpu set %s\n", cpu_list);
 	return 1;
 }
 
@@ -529,3 +524,296 @@ int set_cpu_dma_latency(int32_t latency)
 
 	return fd;
 }
+
+#define _STR(x) #x
+#define STR(x) _STR(x)
+
+/*
+ * find_mount - find a the mount point of a given fs
+ *
+ * Returns 0 if mount is not found, otherwise return 1 and fill mp
+ * with the mount point.
+ */
+static const int find_mount(const char *fs, char *mp, int sizeof_mp)
+{
+	char mount_point[MAX_PATH];
+	char type[100];
+	int found;
+	FILE *fp;
+
+	fp = fopen("/proc/mounts", "r");
+	if (!fp)
+		return 0;
+
+	while (fscanf(fp, "%*s %" STR(MAX_PATH) "s %99s %*s %*d %*d\n",	mount_point, type) == 2) {
+		if (strcmp(type, fs) == 0) {
+			found = 1;
+			break;
+		}
+	}
+	fclose(fp);
+
+	if (!found)
+		return 0;
+
+	memset(mp, 0, sizeof_mp);
+	strncpy(mp, mount_point, sizeof_mp - 1);
+
+	debug_msg("Fs %s found at %s\n", fs, mp);
+	return 1;
+}
+
+/*
+ * get_self_cgroup - get the current thread cgroup path
+ *
+ * Parse /proc/$$/cgroup file to get the thread's cgroup. As an example of line to parse:
+ *
+ * 0::/user.slice/user-0.slice/session-3.scope'\n'
+ *
+ * This function is interested in the content after the second : and before the '\n'.
+ *
+ * Returns 1 if a string was found, 0 otherwise.
+ */
+static int get_self_cgroup(char *self_cg, int sizeof_self_cg)
+{
+	char path[MAX_PATH], *start;
+	int fd, retval;
+
+	snprintf(path, MAX_PATH, "/proc/%d/cgroup", getpid());
+
+	fd = open(path, O_RDONLY);
+	if (fd < 0)
+		return 0;
+
+	retval = read(fd, path, MAX_PATH);
+
+	close(fd);
+
+	if (retval <= 0)
+		return 0;
+
+	start = path;
+
+	start = strstr(start, ":");
+	if (!start)
+		return 0;
+
+	/* skip ":" */
+	start++;
+
+	start = strstr(start, ":");
+	if (!start)
+		return 0;
+
+	/* skip ":" */
+	start++;
+
+	if (strlen(start) >= sizeof_self_cg)
+		return 0;
+
+	snprintf(self_cg, sizeof_self_cg, "%s", start);
+
+	/* Swap '\n' with '\0' */
+	start = strstr(self_cg, "\n");
+
+	/* there must be '\n' */
+	if (!start)
+		return 0;
+
+	/* ok, it found a string after the second : and before the \n */
+	*start = '\0';
+
+	return 1;
+}
+
+/*
+ * set_comm_cgroup - Set cgroup to pid_t pid
+ *
+ * If cgroup argument is not NULL, the threads will move to the given cgroup.
+ * Otherwise, the cgroup of the calling, i.e., rtla, thread will be used.
+ *
+ * Supports cgroup v2.
+ *
+ * Returns 1 on success, 0 otherwise.
+ */
+int set_pid_cgroup(pid_t pid, const char *cgroup)
+{
+	char cgroup_path[MAX_PATH - strlen("/cgroup.procs")];
+	char cgroup_procs[MAX_PATH];
+	char pid_str[24];
+	int retval;
+	int cg_fd;
+
+	retval = find_mount("cgroup2", cgroup_path, sizeof(cgroup_path));
+	if (!retval) {
+		err_msg("Did not find cgroupv2 mount point\n");
+		return 0;
+	}
+
+	if (!cgroup) {
+		retval = get_self_cgroup(&cgroup_path[strlen(cgroup_path)],
+				sizeof(cgroup_path) - strlen(cgroup_path));
+		if (!retval) {
+			err_msg("Did not find self cgroup\n");
+			return 0;
+		}
+	} else {
+		snprintf(&cgroup_path[strlen(cgroup_path)],
+				sizeof(cgroup_path) - strlen(cgroup_path), "%s/", cgroup);
+	}
+
+	snprintf(cgroup_procs, MAX_PATH, "%s/cgroup.procs", cgroup_path);
+
+	debug_msg("Using cgroup path at: %s\n", cgroup_procs);
+
+	cg_fd = open(cgroup_procs, O_RDWR);
+	if (cg_fd < 0)
+		return 0;
+
+	snprintf(pid_str, sizeof(pid_str), "%d\n", pid);
+
+	retval = write(cg_fd, pid_str, strlen(pid_str));
+	if (retval < 0)
+		err_msg("Error setting cgroup attributes for pid:%s - %s\n",
+				pid_str, strerror(errno));
+	else
+		debug_msg("Set cgroup attributes for pid:%s\n", pid_str);
+
+	close(cg_fd);
+
+	return (retval >= 0);
+}
+
+/**
+ * set_comm_cgroup - Set cgroup to threads starting with char *comm_prefix
+ *
+ * If cgroup argument is not NULL, the threads will move to the given cgroup.
+ * Otherwise, the cgroup of the calling, i.e., rtla, thread will be used.
+ *
+ * Supports cgroup v2.
+ *
+ * Returns 1 on success, 0 otherwise.
+ */
+int set_comm_cgroup(const char *comm_prefix, const char *cgroup)
+{
+	char cgroup_path[MAX_PATH - strlen("/cgroup.procs")];
+	char cgroup_procs[MAX_PATH];
+	struct dirent *proc_entry;
+	DIR *procfs;
+	int retval;
+	int cg_fd;
+
+	if (strlen(comm_prefix) >= MAX_PATH) {
+		err_msg("Command prefix is too long: %d < strlen(%s)\n",
+			MAX_PATH, comm_prefix);
+		return 0;
+	}
+
+	retval = find_mount("cgroup2", cgroup_path, sizeof(cgroup_path));
+	if (!retval) {
+		err_msg("Did not find cgroupv2 mount point\n");
+		return 0;
+	}
+
+	if (!cgroup) {
+		retval = get_self_cgroup(&cgroup_path[strlen(cgroup_path)],
+				sizeof(cgroup_path) - strlen(cgroup_path));
+		if (!retval) {
+			err_msg("Did not find self cgroup\n");
+			return 0;
+		}
+	} else {
+		snprintf(&cgroup_path[strlen(cgroup_path)],
+				sizeof(cgroup_path) - strlen(cgroup_path), "%s/", cgroup);
+	}
+
+	snprintf(cgroup_procs, MAX_PATH, "%s/cgroup.procs", cgroup_path);
+
+	debug_msg("Using cgroup path at: %s\n", cgroup_procs);
+
+	cg_fd = open(cgroup_procs, O_RDWR);
+	if (cg_fd < 0)
+		return 0;
+
+	procfs = opendir("/proc");
+	if (!procfs) {
+		err_msg("Could not open procfs\n");
+		goto out_cg;
+	}
+
+	while ((proc_entry = readdir(procfs))) {
+
+		retval = procfs_is_workload_pid(comm_prefix, proc_entry);
+		if (!retval)
+			continue;
+
+		retval = write(cg_fd, proc_entry->d_name, strlen(proc_entry->d_name));
+		if (retval < 0) {
+			err_msg("Error setting cgroup attributes for pid:%s - %s\n",
+				proc_entry->d_name, strerror(errno));
+			goto out_procfs;
+		}
+
+		debug_msg("Set cgroup attributes for pid:%s\n", proc_entry->d_name);
+	}
+
+	closedir(procfs);
+	close(cg_fd);
+	return 1;
+
+out_procfs:
+	closedir(procfs);
+out_cg:
+	close(cg_fd);
+	return 0;
+}
+
+/**
+ * auto_house_keeping - Automatically move rtla out of measurement threads
+ *
+ * Try to move rtla away from the tracer, if possible.
+ *
+ * Returns 1 on success, 0 otherwise.
+ */
+int auto_house_keeping(cpu_set_t *monitored_cpus)
+{
+	cpu_set_t rtla_cpus, house_keeping_cpus;
+	int retval;
+
+	/* first get the CPUs in which rtla can actually run. */
+	retval = sched_getaffinity(getpid(), sizeof(rtla_cpus), &rtla_cpus);
+	if (retval == -1) {
+		debug_msg("Could not get rtla affinity, rtla might run with the threads!\n");
+		return 0;
+	}
+
+	/* then check if the existing setup is already good. */
+	CPU_AND(&house_keeping_cpus, &rtla_cpus, monitored_cpus);
+	if (!CPU_COUNT(&house_keeping_cpus)) {
+		debug_msg("rtla and the monitored CPUs do not share CPUs.");
+		debug_msg("Skipping auto house-keeping\n");
+		return 1;
+	}
+
+	/* remove the intersection */
+	CPU_XOR(&house_keeping_cpus, &rtla_cpus, monitored_cpus);
+
+	/* get only those that rtla can run */
+	CPU_AND(&house_keeping_cpus, &house_keeping_cpus, &rtla_cpus);
+
+	/* is there any cpu left? */
+	if (!CPU_COUNT(&house_keeping_cpus)) {
+		debug_msg("Could not find any CPU for auto house-keeping\n");
+		return 0;
+	}
+
+	retval = sched_setaffinity(getpid(), sizeof(house_keeping_cpus), &house_keeping_cpus);
+	if (retval == -1) {
+		debug_msg("Could not set affinity for auto house-keeping\n");
+		return 0;
+	}
+
+	debug_msg("rtla automatically moved to an auto house-keeping cpu set\n");
+
+	return 1;
+}

tools/tracing/rtla/src/utils.h

 // SPDX-License-Identifier: GPL-2.0
+
 #include <stdint.h>
 #include <time.h>
+#include <sched.h>
 
 /*
  * '18446744073709551615\0'
@@ -54,8 +56,13 @@ struct sched_attr {
 };
 
 int parse_prio(char *arg, struct sched_attr *sched_param);
+int parse_cpu_set(char *cpu_list, cpu_set_t *set);
+int __set_sched_attr(int pid, struct sched_attr *attr);
 int set_comm_sched_attr(const char *comm_prefix, struct sched_attr *attr);
+int set_comm_cgroup(const char *comm_prefix, const char *cgroup);
+int set_pid_cgroup(pid_t pid, const char *cgroup);
 int set_cpu_dma_latency(int32_t latency);
+int auto_house_keeping(cpu_set_t *monitored_cpus);
 
 #define ns_to_usf(x) (((double)x/1000))
 #define ns_to_per(total, part) ((part * 100) / (double)total)