1 // SPDX-License-Identifier: GPL-2.0+
3 // Torture test for smp_call_function() and friends.
5 // Copyright (C) Facebook, 2020.
7 // Author: Paul E. McKenney <paulmck@kernel.org>
9 #define pr_fmt(fmt) fmt
11 #include <linux/atomic.h>
12 #include <linux/bitops.h>
13 #include <linux/completion.h>
14 #include <linux/cpu.h>
15 #include <linux/delay.h>
16 #include <linux/err.h>
17 #include <linux/init.h>
18 #include <linux/interrupt.h>
19 #include <linux/kthread.h>
20 #include <linux/kernel.h>
22 #include <linux/module.h>
23 #include <linux/moduleparam.h>
24 #include <linux/notifier.h>
25 #include <linux/percpu.h>
26 #include <linux/rcupdate.h>
27 #include <linux/rcupdate_trace.h>
28 #include <linux/reboot.h>
29 #include <linux/sched.h>
30 #include <linux/spinlock.h>
31 #include <linux/smp.h>
32 #include <linux/stat.h>
33 #include <linux/srcu.h>
34 #include <linux/slab.h>
35 #include <linux/torture.h>
36 #include <linux/types.h>
38 #define SCFTORT_STRING "scftorture"
39 #define SCFTORT_FLAG SCFTORT_STRING ": "
41 #define VERBOSE_SCFTORTOUT(s, x...) \
42 do { if (verbose) pr_alert(SCFTORT_FLAG s "\n", ## x); } while (0)
44 #define SCFTORTOUT_ERRSTRING(s, x...) pr_alert(SCFTORT_FLAG "!!! " s "\n", ## x)
46 MODULE_LICENSE("GPL");
47 MODULE_AUTHOR("Paul E. McKenney <paulmck@kernel.org>");
49 // Wait until there are multiple CPUs before starting test.
50 torture_param(int, holdoff, IS_BUILTIN(CONFIG_SCF_TORTURE_TEST) ? 10 : 0,
51 "Holdoff time before test start (s)");
52 torture_param(int, longwait, 0, "Include ridiculously long waits? (seconds)");
53 torture_param(int, nthreads, -1, "# threads, defaults to -1 for all CPUs.");
54 torture_param(int, onoff_holdoff, 0, "Time after boot before CPU hotplugs (s)");
55 torture_param(int, onoff_interval, 0, "Time between CPU hotplugs (s), 0=disable");
56 torture_param(int, shutdown_secs, 0, "Shutdown time (ms), <= zero to disable.");
57 torture_param(int, stat_interval, 60, "Number of seconds between stats printk()s.");
58 torture_param(int, stutter, 5, "Number of jiffies to run/halt test, 0=disable");
59 torture_param(bool, use_cpus_read_lock, 0, "Use cpus_read_lock() to exclude CPU hotplug.");
60 torture_param(int, verbose, 0, "Enable verbose debugging printk()s");
61 torture_param(int, weight_resched, -1, "Testing weight for resched_cpu() operations.");
62 torture_param(int, weight_single, -1, "Testing weight for single-CPU no-wait operations.");
63 torture_param(int, weight_single_rpc, -1, "Testing weight for single-CPU RPC operations.");
64 torture_param(int, weight_single_wait, -1, "Testing weight for single-CPU operations.");
65 torture_param(int, weight_many, -1, "Testing weight for multi-CPU no-wait operations.");
66 torture_param(int, weight_many_wait, -1, "Testing weight for multi-CPU operations.");
67 torture_param(int, weight_all, -1, "Testing weight for all-CPU no-wait operations.");
68 torture_param(int, weight_all_wait, -1, "Testing weight for all-CPU operations.");
70 char *torture_type = "";
73 # define SCFTORT_SHUTDOWN 0
75 # define SCFTORT_SHUTDOWN 1
78 torture_param(bool, shutdown, SCFTORT_SHUTDOWN, "Shutdown at end of torture test.");
80 struct scf_statistics {
81 struct task_struct *task;
85 long long n_single_ofl;
86 long long n_single_rpc;
87 long long n_single_rpc_ofl;
88 long long n_single_wait;
89 long long n_single_wait_ofl;
91 long long n_many_wait;
96 static struct scf_statistics *scf_stats_p;
97 static struct task_struct *scf_torture_stats_task;
98 static DEFINE_PER_CPU(long long, scf_invoked_count);
100 // Data for random primitive selection
101 #define SCF_PRIM_RESCHED 0
102 #define SCF_PRIM_SINGLE 1
103 #define SCF_PRIM_SINGLE_RPC 2
104 #define SCF_PRIM_MANY 3
105 #define SCF_PRIM_ALL 4
106 #define SCF_NPRIMS 8 // Need wait and no-wait versions of each,
107 // except for SCF_PRIM_RESCHED and
108 // SCF_PRIM_SINGLE_RPC.
110 static char *scf_prim_name[] = {
112 "smp_call_function_single",
113 "smp_call_function_single_rpc",
114 "smp_call_function_many",
118 struct scf_selector {
119 unsigned long scfs_weight;
123 static struct scf_selector scf_sel_array[SCF_NPRIMS];
124 static int scf_sel_array_len;
125 static unsigned long scf_sel_totweight;
127 // Communicate between caller and handler.
131 int scfc_cpu; // -1 for not _single().
134 struct completion scfc_completion;
137 // Use to wait for all threads to start.
138 static atomic_t n_started;
139 static atomic_t n_errs;
140 static atomic_t n_mb_in_errs;
141 static atomic_t n_mb_out_errs;
142 static atomic_t n_alloc_errs;
144 static char *bangstr = "";
146 static DEFINE_TORTURE_RANDOM_PERCPU(scf_torture_rand);
148 extern void resched_cpu(int cpu); // An alternative IPI vector.
150 // Print torture statistics. Caller must ensure serialization.
151 static void scf_torture_stats_print(void)
155 long long invoked_count = 0;
156 bool isdone = READ_ONCE(scfdone);
157 struct scf_statistics scfs = {};
159 for_each_possible_cpu(cpu)
160 invoked_count += data_race(per_cpu(scf_invoked_count, cpu));
161 for (i = 0; i < nthreads; i++) {
162 scfs.n_resched += scf_stats_p[i].n_resched;
163 scfs.n_single += scf_stats_p[i].n_single;
164 scfs.n_single_ofl += scf_stats_p[i].n_single_ofl;
165 scfs.n_single_rpc += scf_stats_p[i].n_single_rpc;
166 scfs.n_single_wait += scf_stats_p[i].n_single_wait;
167 scfs.n_single_wait_ofl += scf_stats_p[i].n_single_wait_ofl;
168 scfs.n_many += scf_stats_p[i].n_many;
169 scfs.n_many_wait += scf_stats_p[i].n_many_wait;
170 scfs.n_all += scf_stats_p[i].n_all;
171 scfs.n_all_wait += scf_stats_p[i].n_all_wait;
173 if (atomic_read(&n_errs) || atomic_read(&n_mb_in_errs) ||
174 atomic_read(&n_mb_out_errs) || atomic_read(&n_alloc_errs))
176 pr_alert("%s %sscf_invoked_count %s: %lld resched: %lld single: %lld/%lld single_ofl: %lld/%lld single_rpc: %lld single_rpc_ofl: %lld many: %lld/%lld all: %lld/%lld ",
177 SCFTORT_FLAG, bangstr, isdone ? "VER" : "ver", invoked_count, scfs.n_resched,
178 scfs.n_single, scfs.n_single_wait, scfs.n_single_ofl, scfs.n_single_wait_ofl,
179 scfs.n_single_rpc, scfs.n_single_rpc_ofl,
180 scfs.n_many, scfs.n_many_wait, scfs.n_all, scfs.n_all_wait);
181 torture_onoff_stats();
182 pr_cont("ste: %d stnmie: %d stnmoe: %d staf: %d\n", atomic_read(&n_errs),
183 atomic_read(&n_mb_in_errs), atomic_read(&n_mb_out_errs),
184 atomic_read(&n_alloc_errs));
187 // Periodically prints torture statistics, if periodic statistics printing
188 // was specified via the stat_interval module parameter.
190 scf_torture_stats(void *arg)
192 VERBOSE_TOROUT_STRING("scf_torture_stats task started");
194 schedule_timeout_interruptible(stat_interval * HZ);
195 scf_torture_stats_print();
196 torture_shutdown_absorb("scf_torture_stats");
197 } while (!torture_must_stop());
198 torture_kthread_stopping("scf_torture_stats");
202 // Add a primitive to the scf_sel_array[].
203 static void scf_sel_add(unsigned long weight, int prim, bool wait)
205 struct scf_selector *scfsp = &scf_sel_array[scf_sel_array_len];
207 // If no weight, if array would overflow, if computing three-place
208 // percentages would overflow, or if the scf_prim_name[] array would
209 // overflow, don't bother. In the last three two cases, complain.
211 WARN_ON_ONCE(scf_sel_array_len >= ARRAY_SIZE(scf_sel_array)) ||
212 WARN_ON_ONCE(0 - 100000 * weight <= 100000 * scf_sel_totweight) ||
213 WARN_ON_ONCE(prim >= ARRAY_SIZE(scf_prim_name)))
215 scf_sel_totweight += weight;
216 scfsp->scfs_weight = scf_sel_totweight;
217 scfsp->scfs_prim = prim;
218 scfsp->scfs_wait = wait;
222 // Dump out weighting percentages for scf_prim_name[] array.
223 static void scf_sel_dump(void)
226 unsigned long oldw = 0;
227 struct scf_selector *scfsp;
230 for (i = 0; i < scf_sel_array_len; i++) {
231 scfsp = &scf_sel_array[i];
232 w = (scfsp->scfs_weight - oldw) * 100000 / scf_sel_totweight;
233 pr_info("%s: %3lu.%03lu %s(%s)\n", __func__, w / 1000, w % 1000,
234 scf_prim_name[scfsp->scfs_prim],
235 scfsp->scfs_wait ? "wait" : "nowait");
236 oldw = scfsp->scfs_weight;
240 // Randomly pick a primitive and wait/nowait, based on weightings.
241 static struct scf_selector *scf_sel_rand(struct torture_random_state *trsp)
244 unsigned long w = torture_random(trsp) % (scf_sel_totweight + 1);
246 for (i = 0; i < scf_sel_array_len; i++)
247 if (scf_sel_array[i].scfs_weight >= w)
248 return &scf_sel_array[i];
250 return &scf_sel_array[0];
253 // Update statistics and occasionally burn up mass quantities of CPU time,
254 // if told to do so via scftorture.longwait. Otherwise, occasionally burn
256 static void scf_handler(void *scfc_in)
260 unsigned long r = torture_random(this_cpu_ptr(&scf_torture_rand));
261 struct scf_check *scfcp = scfc_in;
264 WRITE_ONCE(scfcp->scfc_out, false); // For multiple receivers.
265 if (WARN_ON_ONCE(unlikely(!READ_ONCE(scfcp->scfc_in))))
266 atomic_inc(&n_mb_in_errs);
268 this_cpu_inc(scf_invoked_count);
278 udelay((r & 0xff) + 1);
281 r = r % longwait + 1;
282 for (i = 0; i < r; i++) {
283 for (j = 0; j < 1000; j++) {
289 if (unlikely(!scfcp))
291 if (scfcp->scfc_wait) {
292 WRITE_ONCE(scfcp->scfc_out, true);
294 complete(&scfcp->scfc_completion);
300 // As above, but check for correct CPU.
301 static void scf_handler_1(void *scfc_in)
303 struct scf_check *scfcp = scfc_in;
305 if (likely(scfcp) && WARN_ONCE(smp_processor_id() != scfcp->scfc_cpu, "%s: Wanted CPU %d got CPU %d\n", __func__, scfcp->scfc_cpu, smp_processor_id())) {
311 // Randomly do an smp_call_function*() invocation.
312 static void scftorture_invoke_one(struct scf_statistics *scfp, struct torture_random_state *trsp)
316 struct scf_check *scfcp = NULL;
317 struct scf_selector *scfsp = scf_sel_rand(trsp);
319 if (use_cpus_read_lock)
323 if (scfsp->scfs_prim == SCF_PRIM_SINGLE || scfsp->scfs_wait) {
324 scfcp = kmalloc(sizeof(*scfcp), GFP_ATOMIC);
325 if (WARN_ON_ONCE(!scfcp)) {
326 atomic_inc(&n_alloc_errs);
328 scfcp->scfc_cpu = -1;
329 scfcp->scfc_wait = scfsp->scfs_wait;
330 scfcp->scfc_out = false;
331 scfcp->scfc_rpc = false;
334 switch (scfsp->scfs_prim) {
335 case SCF_PRIM_RESCHED:
336 if (IS_BUILTIN(CONFIG_SCF_TORTURE_TEST)) {
337 cpu = torture_random(trsp) % nr_cpu_ids;
340 this_cpu_inc(scf_invoked_count);
343 case SCF_PRIM_SINGLE:
344 cpu = torture_random(trsp) % nr_cpu_ids;
345 if (scfsp->scfs_wait)
346 scfp->n_single_wait++;
350 scfcp->scfc_cpu = cpu;
351 barrier(); // Prevent race-reduction compiler optimizations.
352 scfcp->scfc_in = true;
354 ret = smp_call_function_single(cpu, scf_handler_1, (void *)scfcp, scfsp->scfs_wait);
356 if (scfsp->scfs_wait)
357 scfp->n_single_wait_ofl++;
359 scfp->n_single_ofl++;
364 case SCF_PRIM_SINGLE_RPC:
367 cpu = torture_random(trsp) % nr_cpu_ids;
368 scfp->n_single_rpc++;
369 scfcp->scfc_cpu = cpu;
370 scfcp->scfc_wait = true;
371 init_completion(&scfcp->scfc_completion);
372 scfcp->scfc_rpc = true;
373 barrier(); // Prevent race-reduction compiler optimizations.
374 scfcp->scfc_in = true;
375 ret = smp_call_function_single(cpu, scf_handler_1, (void *)scfcp, 0);
377 if (use_cpus_read_lock)
381 wait_for_completion(&scfcp->scfc_completion);
382 if (use_cpus_read_lock)
387 scfp->n_single_rpc_ofl++;
393 if (scfsp->scfs_wait)
398 barrier(); // Prevent race-reduction compiler optimizations.
399 scfcp->scfc_in = true;
401 smp_call_function_many(cpu_online_mask, scf_handler, scfcp, scfsp->scfs_wait);
404 if (scfsp->scfs_wait)
409 barrier(); // Prevent race-reduction compiler optimizations.
410 scfcp->scfc_in = true;
412 smp_call_function(scf_handler, scfcp, scfsp->scfs_wait);
417 scfcp->scfc_out = true;
419 if (scfcp && scfsp->scfs_wait) {
420 if (WARN_ON_ONCE((num_online_cpus() > 1 || scfsp->scfs_prim == SCF_PRIM_SINGLE) &&
422 pr_warn("%s: Memory-ordering failure, scfs_prim: %d.\n", __func__, scfsp->scfs_prim);
423 atomic_inc(&n_mb_out_errs); // Leak rather than trash!
427 barrier(); // Prevent race-reduction compiler optimizations.
429 if (use_cpus_read_lock)
433 if (!(torture_random(trsp) & 0xfff))
434 schedule_timeout_uninterruptible(1);
437 // SCF test kthread. Repeatedly does calls to members of the
438 // smp_call_function() family of functions.
439 static int scftorture_invoker(void *arg)
443 DEFINE_TORTURE_RANDOM(rand);
444 struct scf_statistics *scfp = (struct scf_statistics *)arg;
445 bool was_offline = false;
447 VERBOSE_SCFTORTOUT("scftorture_invoker %d: task started", scfp->cpu);
448 cpu = scfp->cpu % nr_cpu_ids;
449 WARN_ON_ONCE(set_cpus_allowed_ptr(current, cpumask_of(cpu)));
450 set_user_nice(current, MAX_NICE);
452 schedule_timeout_interruptible(holdoff * HZ);
454 VERBOSE_SCFTORTOUT("scftorture_invoker %d: Waiting for all SCF torturers from cpu %d", scfp->cpu, raw_smp_processor_id());
456 // Make sure that the CPU is affinitized appropriately during testing.
457 curcpu = raw_smp_processor_id();
458 WARN_ONCE(curcpu != scfp->cpu % nr_cpu_ids,
459 "%s: Wanted CPU %d, running on %d, nr_cpu_ids = %d\n",
460 __func__, scfp->cpu, curcpu, nr_cpu_ids);
462 if (!atomic_dec_return(&n_started))
463 while (atomic_read_acquire(&n_started)) {
464 if (torture_must_stop()) {
465 VERBOSE_SCFTORTOUT("scftorture_invoker %d ended before starting", scfp->cpu);
468 schedule_timeout_uninterruptible(1);
471 VERBOSE_SCFTORTOUT("scftorture_invoker %d started", scfp->cpu);
474 scftorture_invoke_one(scfp, &rand);
475 while (cpu_is_offline(cpu) && !torture_must_stop()) {
476 schedule_timeout_interruptible(HZ / 5);
480 set_cpus_allowed_ptr(current, cpumask_of(cpu));
484 stutter_wait("scftorture_invoker");
485 } while (!torture_must_stop());
487 VERBOSE_SCFTORTOUT("scftorture_invoker %d ended", scfp->cpu);
489 torture_kthread_stopping("scftorture_invoker");
494 scftorture_print_module_parms(const char *tag)
496 pr_alert(SCFTORT_FLAG
497 "--- %s: verbose=%d holdoff=%d longwait=%d nthreads=%d onoff_holdoff=%d onoff_interval=%d shutdown_secs=%d stat_interval=%d stutter=%d use_cpus_read_lock=%d, weight_resched=%d, weight_single=%d, weight_single_rpc=%d, weight_single_wait=%d, weight_many=%d, weight_many_wait=%d, weight_all=%d, weight_all_wait=%d\n", tag,
498 verbose, holdoff, longwait, nthreads, onoff_holdoff, onoff_interval, shutdown, stat_interval, stutter, use_cpus_read_lock, weight_resched, weight_single, weight_single_rpc, weight_single_wait, weight_many, weight_many_wait, weight_all, weight_all_wait);
501 static void scf_cleanup_handler(void *unused)
505 static void scf_torture_cleanup(void)
509 if (torture_cleanup_begin())
512 WRITE_ONCE(scfdone, true);
513 if (nthreads && scf_stats_p)
514 for (i = 0; i < nthreads; i++)
515 torture_stop_kthread("scftorture_invoker", scf_stats_p[i].task);
518 smp_call_function(scf_cleanup_handler, NULL, 0);
519 torture_stop_kthread(scf_torture_stats, scf_torture_stats_task);
520 scf_torture_stats_print(); // -After- the stats thread is stopped!
521 kfree(scf_stats_p); // -After- the last stats print has completed!
524 if (atomic_read(&n_errs) || atomic_read(&n_mb_in_errs) || atomic_read(&n_mb_out_errs))
525 scftorture_print_module_parms("End of test: FAILURE");
526 else if (torture_onoff_failures())
527 scftorture_print_module_parms("End of test: LOCK_HOTPLUG");
529 scftorture_print_module_parms("End of test: SUCCESS");
532 torture_cleanup_end();
535 static int __init scf_torture_init(void)
539 unsigned long weight_resched1 = weight_resched;
540 unsigned long weight_single1 = weight_single;
541 unsigned long weight_single_rpc1 = weight_single_rpc;
542 unsigned long weight_single_wait1 = weight_single_wait;
543 unsigned long weight_many1 = weight_many;
544 unsigned long weight_many_wait1 = weight_many_wait;
545 unsigned long weight_all1 = weight_all;
546 unsigned long weight_all_wait1 = weight_all_wait;
548 if (!torture_init_begin(SCFTORT_STRING, verbose))
551 scftorture_print_module_parms("Start of test");
553 if (weight_resched <= 0 &&
554 weight_single <= 0 && weight_single_rpc <= 0 && weight_single_wait <= 0 &&
555 weight_many <= 0 && weight_many_wait <= 0 &&
556 weight_all <= 0 && weight_all_wait <= 0) {
557 weight_resched1 = weight_resched == 0 ? 0 : 2 * nr_cpu_ids;
558 weight_single1 = weight_single == 0 ? 0 : 2 * nr_cpu_ids;
559 weight_single_rpc1 = weight_single_rpc == 0 ? 0 : 2 * nr_cpu_ids;
560 weight_single_wait1 = weight_single_wait == 0 ? 0 : 2 * nr_cpu_ids;
561 weight_many1 = weight_many == 0 ? 0 : 2;
562 weight_many_wait1 = weight_many_wait == 0 ? 0 : 2;
563 weight_all1 = weight_all == 0 ? 0 : 1;
564 weight_all_wait1 = weight_all_wait == 0 ? 0 : 1;
566 if (weight_resched == -1)
568 if (weight_single == -1)
570 if (weight_single_rpc == -1)
571 weight_single_rpc1 = 0;
572 if (weight_single_wait == -1)
573 weight_single_wait1 = 0;
574 if (weight_many == -1)
576 if (weight_many_wait == -1)
577 weight_many_wait1 = 0;
578 if (weight_all == -1)
580 if (weight_all_wait == -1)
581 weight_all_wait1 = 0;
583 if (weight_resched1 == 0 && weight_single1 == 0 && weight_single_rpc1 == 0 &&
584 weight_single_wait1 == 0 && weight_many1 == 0 && weight_many_wait1 == 0 &&
585 weight_all1 == 0 && weight_all_wait1 == 0) {
586 SCFTORTOUT_ERRSTRING("all zero weights makes no sense");
590 if (IS_BUILTIN(CONFIG_SCF_TORTURE_TEST))
591 scf_sel_add(weight_resched1, SCF_PRIM_RESCHED, false);
592 else if (weight_resched1)
593 SCFTORTOUT_ERRSTRING("built as module, weight_resched ignored");
594 scf_sel_add(weight_single1, SCF_PRIM_SINGLE, false);
595 scf_sel_add(weight_single_rpc1, SCF_PRIM_SINGLE_RPC, true);
596 scf_sel_add(weight_single_wait1, SCF_PRIM_SINGLE, true);
597 scf_sel_add(weight_many1, SCF_PRIM_MANY, false);
598 scf_sel_add(weight_many_wait1, SCF_PRIM_MANY, true);
599 scf_sel_add(weight_all1, SCF_PRIM_ALL, false);
600 scf_sel_add(weight_all_wait1, SCF_PRIM_ALL, true);
603 if (onoff_interval > 0) {
604 firsterr = torture_onoff_init(onoff_holdoff * HZ, onoff_interval, NULL);
605 if (torture_init_error(firsterr))
608 if (shutdown_secs > 0) {
609 firsterr = torture_shutdown_init(shutdown_secs, scf_torture_cleanup);
610 if (torture_init_error(firsterr))
614 firsterr = torture_stutter_init(stutter, stutter);
615 if (torture_init_error(firsterr))
619 // Worker tasks invoking smp_call_function().
621 nthreads = num_online_cpus();
622 scf_stats_p = kcalloc(nthreads, sizeof(scf_stats_p[0]), GFP_KERNEL);
624 SCFTORTOUT_ERRSTRING("out of memory");
629 VERBOSE_SCFTORTOUT("Starting %d smp_call_function() threads", nthreads);
631 atomic_set(&n_started, nthreads);
632 for (i = 0; i < nthreads; i++) {
633 scf_stats_p[i].cpu = i;
634 firsterr = torture_create_kthread(scftorture_invoker, (void *)&scf_stats_p[i],
635 scf_stats_p[i].task);
636 if (torture_init_error(firsterr))
639 if (stat_interval > 0) {
640 firsterr = torture_create_kthread(scf_torture_stats, NULL, scf_torture_stats_task);
641 if (torture_init_error(firsterr))
650 scf_torture_cleanup();
652 WARN_ON(!IS_MODULE(CONFIG_SCF_TORTURE_TEST));
658 module_init(scf_torture_init);
659 module_exit(scf_torture_cleanup);