xref: /linux/tools/testing/selftests/sched_ext/rt_stall.c (revision 69050f8d6d075dc01af7a5f2f550a8067510366f)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (c) 2025 NVIDIA Corporation.
4  */
5 #define _GNU_SOURCE
6 #include <stdio.h>
7 #include <stdlib.h>
8 #include <unistd.h>
9 #include <sched.h>
10 #include <sys/prctl.h>
11 #include <sys/types.h>
12 #include <sys/wait.h>
13 #include <time.h>
14 #include <linux/sched.h>
15 #include <signal.h>
16 #include <bpf/bpf.h>
17 #include <scx/common.h>
18 #include <unistd.h>
19 #include "rt_stall.bpf.skel.h"
20 #include "scx_test.h"
21 #include "../kselftest.h"
22 
23 #define CORE_ID		0	/* CPU to pin tasks to */
24 #define RUN_TIME        5	/* How long to run the test in seconds */
25 
26 /* Simple busy-wait function for test tasks */
27 static void process_func(void)
28 {
29 	while (1) {
30 		/* Busy wait */
31 		for (volatile unsigned long i = 0; i < 10000000UL; i++)
32 			;
33 	}
34 }
35 
36 /* Set CPU affinity to a specific core */
37 static void set_affinity(int cpu)
38 {
39 	cpu_set_t mask;
40 
41 	CPU_ZERO(&mask);
42 	CPU_SET(cpu, &mask);
43 	if (sched_setaffinity(0, sizeof(mask), &mask) != 0) {
44 		perror("sched_setaffinity");
45 		exit(EXIT_FAILURE);
46 	}
47 }
48 
49 /* Set task scheduling policy and priority */
50 static void set_sched(int policy, int priority)
51 {
52 	struct sched_param param;
53 
54 	param.sched_priority = priority;
55 	if (sched_setscheduler(0, policy, &param) != 0) {
56 		perror("sched_setscheduler");
57 		exit(EXIT_FAILURE);
58 	}
59 }
60 
61 /* Get process runtime from /proc/<pid>/stat */
62 static float get_process_runtime(int pid)
63 {
64 	char path[256];
65 	FILE *file;
66 	long utime, stime;
67 	int fields;
68 
69 	snprintf(path, sizeof(path), "/proc/%d/stat", pid);
70 	file = fopen(path, "r");
71 	if (file == NULL) {
72 		perror("Failed to open stat file");
73 		return -1;
74 	}
75 
76 	/* Skip the first 13 fields and read the 14th and 15th */
77 	fields = fscanf(file,
78 			"%*d %*s %*c %*d %*d %*d %*d %*d %*u %*u %*u %*u %*u %lu %lu",
79 			&utime, &stime);
80 	fclose(file);
81 
82 	if (fields != 2) {
83 		fprintf(stderr, "Failed to read stat file\n");
84 		return -1;
85 	}
86 
87 	/* Calculate the total time spent in the process */
88 	long total_time = utime + stime;
89 	long ticks_per_second = sysconf(_SC_CLK_TCK);
90 	float runtime_seconds = total_time * 1.0 / ticks_per_second;
91 
92 	return runtime_seconds;
93 }
94 
95 static enum scx_test_status setup(void **ctx)
96 {
97 	struct rt_stall *skel;
98 
99 	skel = rt_stall__open();
100 	SCX_FAIL_IF(!skel, "Failed to open");
101 	SCX_ENUM_INIT(skel);
102 	SCX_FAIL_IF(rt_stall__load(skel), "Failed to load skel");
103 
104 	*ctx = skel;
105 
106 	return SCX_TEST_PASS;
107 }
108 
109 static bool sched_stress_test(bool is_ext)
110 {
111 	/*
112 	 * We're expecting the EXT task to get around 5% of CPU time when
113 	 * competing with the RT task (small 1% fluctuations are expected).
114 	 *
115 	 * However, the EXT task should get at least 4% of the CPU to prove
116 	 * that the EXT deadline server is working correctly. A percentage
117 	 * less than 4% indicates a bug where RT tasks can potentially
118 	 * stall SCHED_EXT tasks, causing the test to fail.
119 	 */
120 	const float expected_min_ratio = 0.04; /* 4% */
121 	const char *class_str = is_ext ? "EXT" : "FAIR";
122 
123 	float ext_runtime, rt_runtime, actual_ratio;
124 	int ext_pid, rt_pid;
125 
126 	ksft_print_header();
127 	ksft_set_plan(1);
128 
129 	/* Create and set up a EXT task */
130 	ext_pid = fork();
131 	if (ext_pid == 0) {
132 		set_affinity(CORE_ID);
133 		process_func();
134 		exit(0);
135 	} else if (ext_pid < 0) {
136 		perror("fork task");
137 		ksft_exit_fail();
138 	}
139 
140 	/* Create an RT task */
141 	rt_pid = fork();
142 	if (rt_pid == 0) {
143 		set_affinity(CORE_ID);
144 		set_sched(SCHED_FIFO, 50);
145 		process_func();
146 		exit(0);
147 	} else if (rt_pid < 0) {
148 		perror("fork for RT task");
149 		ksft_exit_fail();
150 	}
151 
152 	/* Let the processes run for the specified time */
153 	sleep(RUN_TIME);
154 
155 	/* Get runtime for the EXT task */
156 	ext_runtime = get_process_runtime(ext_pid);
157 	if (ext_runtime == -1)
158 		ksft_exit_fail_msg("Error getting runtime for %s task (PID %d)\n",
159 				   class_str, ext_pid);
160 	ksft_print_msg("Runtime of %s task (PID %d) is %f seconds\n",
161 		       class_str, ext_pid, ext_runtime);
162 
163 	/* Get runtime for the RT task */
164 	rt_runtime = get_process_runtime(rt_pid);
165 	if (rt_runtime == -1)
166 		ksft_exit_fail_msg("Error getting runtime for RT task (PID %d)\n", rt_pid);
167 	ksft_print_msg("Runtime of RT task (PID %d) is %f seconds\n", rt_pid, rt_runtime);
168 
169 	/* Kill the processes */
170 	kill(ext_pid, SIGKILL);
171 	kill(rt_pid, SIGKILL);
172 	waitpid(ext_pid, NULL, 0);
173 	waitpid(rt_pid, NULL, 0);
174 
175 	/* Verify that the scx task got enough runtime */
176 	actual_ratio = ext_runtime / (ext_runtime + rt_runtime);
177 	ksft_print_msg("%s task got %.2f%% of total runtime\n",
178 		       class_str, actual_ratio * 100);
179 
180 	if (actual_ratio >= expected_min_ratio) {
181 		ksft_test_result_pass("PASS: %s task got more than %.2f%% of runtime\n",
182 				      class_str, expected_min_ratio * 100);
183 		return true;
184 	}
185 	ksft_test_result_fail("FAIL: %s task got less than %.2f%% of runtime\n",
186 			      class_str, expected_min_ratio * 100);
187 	return false;
188 }
189 
190 static enum scx_test_status run(void *ctx)
191 {
192 	struct rt_stall *skel = ctx;
193 	struct bpf_link *link = NULL;
194 	bool res;
195 	int i;
196 
197 	/*
198 	 * Test if the dl_server is working both with and without the
199 	 * sched_ext scheduler attached.
200 	 *
201 	 * This ensures all the scenarios are covered:
202 	 *   - fair_server stop -> ext_server start
203 	 *   - ext_server stop -> fair_server stop
204 	 */
205 	for (i = 0; i < 4; i++) {
206 		bool is_ext = i % 2;
207 
208 		if (is_ext) {
209 			memset(&skel->data->uei, 0, sizeof(skel->data->uei));
210 			link = bpf_map__attach_struct_ops(skel->maps.rt_stall_ops);
211 			SCX_FAIL_IF(!link, "Failed to attach scheduler");
212 		}
213 		res = sched_stress_test(is_ext);
214 		if (is_ext) {
215 			SCX_EQ(skel->data->uei.kind, EXIT_KIND(SCX_EXIT_NONE));
216 			bpf_link__destroy(link);
217 		}
218 
219 		if (!res)
220 			ksft_exit_fail();
221 	}
222 
223 	return SCX_TEST_PASS;
224 }
225 
226 static void cleanup(void *ctx)
227 {
228 	struct rt_stall *skel = ctx;
229 
230 	rt_stall__destroy(skel);
231 }
232 
233 struct scx_test rt_stall = {
234 	.name = "rt_stall",
235 	.description = "Verify that RT tasks cannot stall SCHED_EXT tasks",
236 	.setup = setup,
237 	.run = run,
238 	.cleanup = cleanup,
239 };
240 REGISTER_SCX_TEST(&rt_stall)
241