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) ||
175 (!IS_ENABLED(CONFIG_KASAN) && atomic_read(&n_alloc_errs)))
177 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 ",
178 SCFTORT_FLAG, bangstr, isdone ? "VER" : "ver", invoked_count, scfs.n_resched,
179 scfs.n_single, scfs.n_single_wait, scfs.n_single_ofl, scfs.n_single_wait_ofl,
180 scfs.n_single_rpc, scfs.n_single_rpc_ofl,
181 scfs.n_many, scfs.n_many_wait, scfs.n_all, scfs.n_all_wait);
182 torture_onoff_stats();
183 pr_cont("ste: %d stnmie: %d stnmoe: %d staf: %d\n", atomic_read(&n_errs),
184 atomic_read(&n_mb_in_errs), atomic_read(&n_mb_out_errs),
185 atomic_read(&n_alloc_errs));
188 // Periodically prints torture statistics, if periodic statistics printing
189 // was specified via the stat_interval module parameter.
191 scf_torture_stats(void *arg)
193 VERBOSE_TOROUT_STRING("scf_torture_stats task started");
195 schedule_timeout_interruptible(stat_interval * HZ);
196 scf_torture_stats_print();
197 torture_shutdown_absorb("scf_torture_stats");
198 } while (!torture_must_stop());
199 torture_kthread_stopping("scf_torture_stats");
203 // Add a primitive to the scf_sel_array[].
204 static void scf_sel_add(unsigned long weight, int prim, bool wait)
206 struct scf_selector *scfsp = &scf_sel_array[scf_sel_array_len];
208 // If no weight, if array would overflow, if computing three-place
209 // percentages would overflow, or if the scf_prim_name[] array would
210 // overflow, don't bother. In the last three two cases, complain.
212 WARN_ON_ONCE(scf_sel_array_len >= ARRAY_SIZE(scf_sel_array)) ||
213 WARN_ON_ONCE(0 - 100000 * weight <= 100000 * scf_sel_totweight) ||
214 WARN_ON_ONCE(prim >= ARRAY_SIZE(scf_prim_name)))
216 scf_sel_totweight += weight;
217 scfsp->scfs_weight = scf_sel_totweight;
218 scfsp->scfs_prim = prim;
219 scfsp->scfs_wait = wait;
223 // Dump out weighting percentages for scf_prim_name[] array.
224 static void scf_sel_dump(void)
227 unsigned long oldw = 0;
228 struct scf_selector *scfsp;
231 for (i = 0; i < scf_sel_array_len; i++) {
232 scfsp = &scf_sel_array[i];
233 w = (scfsp->scfs_weight - oldw) * 100000 / scf_sel_totweight;
234 pr_info("%s: %3lu.%03lu %s(%s)\n", __func__, w / 1000, w % 1000,
235 scf_prim_name[scfsp->scfs_prim],
236 scfsp->scfs_wait ? "wait" : "nowait");
237 oldw = scfsp->scfs_weight;
241 // Randomly pick a primitive and wait/nowait, based on weightings.
242 static struct scf_selector *scf_sel_rand(struct torture_random_state *trsp)
245 unsigned long w = torture_random(trsp) % (scf_sel_totweight + 1);
247 for (i = 0; i < scf_sel_array_len; i++)
248 if (scf_sel_array[i].scfs_weight >= w)
249 return &scf_sel_array[i];
251 return &scf_sel_array[0];
254 // Update statistics and occasionally burn up mass quantities of CPU time,
255 // if told to do so via scftorture.longwait. Otherwise, occasionally burn
257 static void scf_handler(void *scfc_in)
261 unsigned long r = torture_random(this_cpu_ptr(&scf_torture_rand));
262 struct scf_check *scfcp = scfc_in;
265 WRITE_ONCE(scfcp->scfc_out, false); // For multiple receivers.
266 if (WARN_ON_ONCE(unlikely(!READ_ONCE(scfcp->scfc_in))))
267 atomic_inc(&n_mb_in_errs);
269 this_cpu_inc(scf_invoked_count);
280 udelay((r & 0xff) + 1);
283 r = r % longwait + 1;
284 for (i = 0; i < r; i++) {
285 for (j = 0; j < 1000; j++) {
291 if (unlikely(!scfcp))
293 if (scfcp->scfc_wait) {
294 WRITE_ONCE(scfcp->scfc_out, true);
296 complete(&scfcp->scfc_completion);
302 // As above, but check for correct CPU.
303 static void scf_handler_1(void *scfc_in)
305 struct scf_check *scfcp = scfc_in;
307 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())) {
313 // Randomly do an smp_call_function*() invocation.
314 static void scftorture_invoke_one(struct scf_statistics *scfp, struct torture_random_state *trsp)
316 bool allocfail = false;
319 struct scf_check *scfcp = NULL;
320 struct scf_selector *scfsp = scf_sel_rand(trsp);
322 if (use_cpus_read_lock)
326 if (scfsp->scfs_prim == SCF_PRIM_SINGLE || scfsp->scfs_wait) {
327 scfcp = kmalloc(sizeof(*scfcp), GFP_ATOMIC);
329 WARN_ON_ONCE(!IS_ENABLED(CONFIG_KASAN));
330 atomic_inc(&n_alloc_errs);
333 scfcp->scfc_cpu = -1;
334 scfcp->scfc_wait = scfsp->scfs_wait;
335 scfcp->scfc_out = false;
336 scfcp->scfc_rpc = false;
339 switch (scfsp->scfs_prim) {
340 case SCF_PRIM_RESCHED:
341 if (IS_BUILTIN(CONFIG_SCF_TORTURE_TEST)) {
342 cpu = torture_random(trsp) % nr_cpu_ids;
345 this_cpu_inc(scf_invoked_count);
348 case SCF_PRIM_SINGLE:
349 cpu = torture_random(trsp) % nr_cpu_ids;
350 if (scfsp->scfs_wait)
351 scfp->n_single_wait++;
355 scfcp->scfc_cpu = cpu;
356 barrier(); // Prevent race-reduction compiler optimizations.
357 scfcp->scfc_in = true;
359 ret = smp_call_function_single(cpu, scf_handler_1, (void *)scfcp, scfsp->scfs_wait);
361 if (scfsp->scfs_wait)
362 scfp->n_single_wait_ofl++;
364 scfp->n_single_ofl++;
369 case SCF_PRIM_SINGLE_RPC:
372 cpu = torture_random(trsp) % nr_cpu_ids;
373 scfp->n_single_rpc++;
374 scfcp->scfc_cpu = cpu;
375 scfcp->scfc_wait = true;
376 init_completion(&scfcp->scfc_completion);
377 scfcp->scfc_rpc = true;
378 barrier(); // Prevent race-reduction compiler optimizations.
379 scfcp->scfc_in = true;
380 ret = smp_call_function_single(cpu, scf_handler_1, (void *)scfcp, 0);
382 if (use_cpus_read_lock)
386 wait_for_completion(&scfcp->scfc_completion);
387 if (use_cpus_read_lock)
392 scfp->n_single_rpc_ofl++;
398 if (scfsp->scfs_wait)
403 barrier(); // Prevent race-reduction compiler optimizations.
404 scfcp->scfc_in = true;
406 smp_call_function_many(cpu_online_mask, scf_handler, scfcp, scfsp->scfs_wait);
409 if (scfsp->scfs_wait)
414 barrier(); // Prevent race-reduction compiler optimizations.
415 scfcp->scfc_in = true;
417 smp_call_function(scf_handler, scfcp, scfsp->scfs_wait);
422 scfcp->scfc_out = true;
424 if (scfcp && scfsp->scfs_wait) {
425 if (WARN_ON_ONCE((num_online_cpus() > 1 || scfsp->scfs_prim == SCF_PRIM_SINGLE) &&
427 pr_warn("%s: Memory-ordering failure, scfs_prim: %d.\n", __func__, scfsp->scfs_prim);
428 atomic_inc(&n_mb_out_errs); // Leak rather than trash!
432 barrier(); // Prevent race-reduction compiler optimizations.
434 if (use_cpus_read_lock)
439 schedule_timeout_idle((1 + longwait) * HZ); // Let no-wait handlers complete.
440 else if (!(torture_random(trsp) & 0xfff))
441 schedule_timeout_uninterruptible(1);
444 // SCF test kthread. Repeatedly does calls to members of the
445 // smp_call_function() family of functions.
446 static int scftorture_invoker(void *arg)
450 DEFINE_TORTURE_RANDOM(rand);
451 struct scf_statistics *scfp = (struct scf_statistics *)arg;
452 bool was_offline = false;
454 VERBOSE_SCFTORTOUT("scftorture_invoker %d: task started", scfp->cpu);
455 cpu = scfp->cpu % nr_cpu_ids;
456 WARN_ON_ONCE(set_cpus_allowed_ptr(current, cpumask_of(cpu)));
457 set_user_nice(current, MAX_NICE);
459 schedule_timeout_interruptible(holdoff * HZ);
461 VERBOSE_SCFTORTOUT("scftorture_invoker %d: Waiting for all SCF torturers from cpu %d", scfp->cpu, raw_smp_processor_id());
463 // Make sure that the CPU is affinitized appropriately during testing.
464 curcpu = raw_smp_processor_id();
465 WARN_ONCE(curcpu != scfp->cpu % nr_cpu_ids,
466 "%s: Wanted CPU %d, running on %d, nr_cpu_ids = %d\n",
467 __func__, scfp->cpu, curcpu, nr_cpu_ids);
469 if (!atomic_dec_return(&n_started))
470 while (atomic_read_acquire(&n_started)) {
471 if (torture_must_stop()) {
472 VERBOSE_SCFTORTOUT("scftorture_invoker %d ended before starting", scfp->cpu);
475 schedule_timeout_uninterruptible(1);
478 VERBOSE_SCFTORTOUT("scftorture_invoker %d started", scfp->cpu);
481 scftorture_invoke_one(scfp, &rand);
482 while (cpu_is_offline(cpu) && !torture_must_stop()) {
483 schedule_timeout_interruptible(HZ / 5);
487 set_cpus_allowed_ptr(current, cpumask_of(cpu));
491 stutter_wait("scftorture_invoker");
492 } while (!torture_must_stop());
494 VERBOSE_SCFTORTOUT("scftorture_invoker %d ended", scfp->cpu);
496 torture_kthread_stopping("scftorture_invoker");
501 scftorture_print_module_parms(const char *tag)
503 pr_alert(SCFTORT_FLAG
504 "--- %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,
505 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);
508 static void scf_cleanup_handler(void *unused)
512 static void scf_torture_cleanup(void)
516 if (torture_cleanup_begin())
519 WRITE_ONCE(scfdone, true);
520 if (nthreads && scf_stats_p)
521 for (i = 0; i < nthreads; i++)
522 torture_stop_kthread("scftorture_invoker", scf_stats_p[i].task);
525 smp_call_function(scf_cleanup_handler, NULL, 0);
526 torture_stop_kthread(scf_torture_stats, scf_torture_stats_task);
527 scf_torture_stats_print(); // -After- the stats thread is stopped!
528 kfree(scf_stats_p); // -After- the last stats print has completed!
531 if (atomic_read(&n_errs) || atomic_read(&n_mb_in_errs) || atomic_read(&n_mb_out_errs))
532 scftorture_print_module_parms("End of test: FAILURE");
533 else if (torture_onoff_failures())
534 scftorture_print_module_parms("End of test: LOCK_HOTPLUG");
536 scftorture_print_module_parms("End of test: SUCCESS");
539 torture_cleanup_end();
542 static int __init scf_torture_init(void)
546 unsigned long weight_resched1 = weight_resched;
547 unsigned long weight_single1 = weight_single;
548 unsigned long weight_single_rpc1 = weight_single_rpc;
549 unsigned long weight_single_wait1 = weight_single_wait;
550 unsigned long weight_many1 = weight_many;
551 unsigned long weight_many_wait1 = weight_many_wait;
552 unsigned long weight_all1 = weight_all;
553 unsigned long weight_all_wait1 = weight_all_wait;
555 if (!torture_init_begin(SCFTORT_STRING, verbose))
558 scftorture_print_module_parms("Start of test");
560 if (weight_resched <= 0 &&
561 weight_single <= 0 && weight_single_rpc <= 0 && weight_single_wait <= 0 &&
562 weight_many <= 0 && weight_many_wait <= 0 &&
563 weight_all <= 0 && weight_all_wait <= 0) {
564 weight_resched1 = weight_resched == 0 ? 0 : 2 * nr_cpu_ids;
565 weight_single1 = weight_single == 0 ? 0 : 2 * nr_cpu_ids;
566 weight_single_rpc1 = weight_single_rpc == 0 ? 0 : 2 * nr_cpu_ids;
567 weight_single_wait1 = weight_single_wait == 0 ? 0 : 2 * nr_cpu_ids;
568 weight_many1 = weight_many == 0 ? 0 : 2;
569 weight_many_wait1 = weight_many_wait == 0 ? 0 : 2;
570 weight_all1 = weight_all == 0 ? 0 : 1;
571 weight_all_wait1 = weight_all_wait == 0 ? 0 : 1;
573 if (weight_resched == -1)
575 if (weight_single == -1)
577 if (weight_single_rpc == -1)
578 weight_single_rpc1 = 0;
579 if (weight_single_wait == -1)
580 weight_single_wait1 = 0;
581 if (weight_many == -1)
583 if (weight_many_wait == -1)
584 weight_many_wait1 = 0;
585 if (weight_all == -1)
587 if (weight_all_wait == -1)
588 weight_all_wait1 = 0;
590 if (weight_resched1 == 0 && weight_single1 == 0 && weight_single_rpc1 == 0 &&
591 weight_single_wait1 == 0 && weight_many1 == 0 && weight_many_wait1 == 0 &&
592 weight_all1 == 0 && weight_all_wait1 == 0) {
593 SCFTORTOUT_ERRSTRING("all zero weights makes no sense");
597 if (IS_BUILTIN(CONFIG_SCF_TORTURE_TEST))
598 scf_sel_add(weight_resched1, SCF_PRIM_RESCHED, false);
599 else if (weight_resched1)
600 SCFTORTOUT_ERRSTRING("built as module, weight_resched ignored");
601 scf_sel_add(weight_single1, SCF_PRIM_SINGLE, false);
602 scf_sel_add(weight_single_rpc1, SCF_PRIM_SINGLE_RPC, true);
603 scf_sel_add(weight_single_wait1, SCF_PRIM_SINGLE, true);
604 scf_sel_add(weight_many1, SCF_PRIM_MANY, false);
605 scf_sel_add(weight_many_wait1, SCF_PRIM_MANY, true);
606 scf_sel_add(weight_all1, SCF_PRIM_ALL, false);
607 scf_sel_add(weight_all_wait1, SCF_PRIM_ALL, true);
610 if (onoff_interval > 0) {
611 firsterr = torture_onoff_init(onoff_holdoff * HZ, onoff_interval, NULL);
612 if (torture_init_error(firsterr))
615 if (shutdown_secs > 0) {
616 firsterr = torture_shutdown_init(shutdown_secs, scf_torture_cleanup);
617 if (torture_init_error(firsterr))
621 firsterr = torture_stutter_init(stutter, stutter);
622 if (torture_init_error(firsterr))
626 // Worker tasks invoking smp_call_function().
628 nthreads = num_online_cpus();
629 scf_stats_p = kcalloc(nthreads, sizeof(scf_stats_p[0]), GFP_KERNEL);
631 SCFTORTOUT_ERRSTRING("out of memory");
636 VERBOSE_SCFTORTOUT("Starting %d smp_call_function() threads", nthreads);
638 atomic_set(&n_started, nthreads);
639 for (i = 0; i < nthreads; i++) {
640 scf_stats_p[i].cpu = i;
641 firsterr = torture_create_kthread(scftorture_invoker, (void *)&scf_stats_p[i],
642 scf_stats_p[i].task);
643 if (torture_init_error(firsterr))
646 if (stat_interval > 0) {
647 firsterr = torture_create_kthread(scf_torture_stats, NULL, scf_torture_stats_task);
648 if (torture_init_error(firsterr))
657 scf_torture_cleanup();
659 WARN_ON(!IS_MODULE(CONFIG_SCF_TORTURE_TEST));
665 module_init(scf_torture_init);
666 module_exit(scf_torture_cleanup);