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 SCFTORTOUT(s, x...) \
42 pr_alert(SCFTORT_FLAG s, ## x)
44 #define VERBOSE_SCFTORTOUT(s, x...) \
45 do { if (verbose) pr_alert(SCFTORT_FLAG s, ## x); } while (0)
47 #define VERBOSE_SCFTORTOUT_ERRSTRING(s, x...) \
48 do { if (verbose) pr_alert(SCFTORT_FLAG "!!! " s, ## x); } while (0)
50 MODULE_LICENSE("GPL");
51 MODULE_AUTHOR("Paul E. McKenney <paulmck@kernel.org>");
53 // Wait until there are multiple CPUs before starting test.
54 torture_param(int, holdoff, IS_BUILTIN(CONFIG_SCF_TORTURE_TEST) ? 10 : 0,
55 "Holdoff time before test start (s)");
56 torture_param(int, longwait, 0, "Include ridiculously long waits? (seconds)");
57 torture_param(int, nthreads, -1, "# threads, defaults to -1 for all CPUs.");
58 torture_param(int, onoff_holdoff, 0, "Time after boot before CPU hotplugs (s)");
59 torture_param(int, onoff_interval, 0, "Time between CPU hotplugs (s), 0=disable");
60 torture_param(int, shutdown_secs, 0, "Shutdown time (ms), <= zero to disable.");
61 torture_param(int, stat_interval, 60, "Number of seconds between stats printk()s.");
62 torture_param(int, stutter, 5, "Number of jiffies to run/halt test, 0=disable");
63 torture_param(bool, use_cpus_read_lock, 0, "Use cpus_read_lock() to exclude CPU hotplug.");
64 torture_param(int, verbose, 0, "Enable verbose debugging printk()s");
65 torture_param(int, weight_resched, -1, "Testing weight for resched_cpu() operations.");
66 torture_param(int, weight_single, -1, "Testing weight for single-CPU no-wait operations.");
67 torture_param(int, weight_single_rpc, -1, "Testing weight for single-CPU RPC operations.");
68 torture_param(int, weight_single_wait, -1, "Testing weight for single-CPU operations.");
69 torture_param(int, weight_many, -1, "Testing weight for multi-CPU no-wait operations.");
70 torture_param(int, weight_many_wait, -1, "Testing weight for multi-CPU operations.");
71 torture_param(int, weight_all, -1, "Testing weight for all-CPU no-wait operations.");
72 torture_param(int, weight_all_wait, -1, "Testing weight for all-CPU operations.");
74 char *torture_type = "";
77 # define SCFTORT_SHUTDOWN 0
79 # define SCFTORT_SHUTDOWN 1
82 torture_param(bool, shutdown, SCFTORT_SHUTDOWN, "Shutdown at end of torture test.");
84 struct scf_statistics {
85 struct task_struct *task;
89 long long n_single_ofl;
90 long long n_single_rpc;
91 long long n_single_rpc_ofl;
92 long long n_single_wait;
93 long long n_single_wait_ofl;
95 long long n_many_wait;
100 static struct scf_statistics *scf_stats_p;
101 static struct task_struct *scf_torture_stats_task;
102 static DEFINE_PER_CPU(long long, scf_invoked_count);
104 // Data for random primitive selection
105 #define SCF_PRIM_RESCHED 0
106 #define SCF_PRIM_SINGLE 1
107 #define SCF_PRIM_SINGLE_RPC 2
108 #define SCF_PRIM_MANY 3
109 #define SCF_PRIM_ALL 4
110 #define SCF_NPRIMS 8 // Need wait and no-wait versions of each,
111 // except for SCF_PRIM_RESCHED and
112 // SCF_PRIM_SINGLE_RPC.
114 static char *scf_prim_name[] = {
116 "smp_call_function_single",
117 "smp_call_function_single_rpc",
118 "smp_call_function_many",
122 struct scf_selector {
123 unsigned long scfs_weight;
127 static struct scf_selector scf_sel_array[SCF_NPRIMS];
128 static int scf_sel_array_len;
129 static unsigned long scf_sel_totweight;
131 // Communicate between caller and handler.
135 int scfc_cpu; // -1 for not _single().
138 struct completion scfc_completion;
141 // Use to wait for all threads to start.
142 static atomic_t n_started;
143 static atomic_t n_errs;
144 static atomic_t n_mb_in_errs;
145 static atomic_t n_mb_out_errs;
146 static atomic_t n_alloc_errs;
148 static char *bangstr = "";
150 static DEFINE_TORTURE_RANDOM_PERCPU(scf_torture_rand);
152 extern void resched_cpu(int cpu); // An alternative IPI vector.
154 // Print torture statistics. Caller must ensure serialization.
155 static void scf_torture_stats_print(void)
159 long long invoked_count = 0;
160 bool isdone = READ_ONCE(scfdone);
161 struct scf_statistics scfs = {};
163 for_each_possible_cpu(cpu)
164 invoked_count += data_race(per_cpu(scf_invoked_count, cpu));
165 for (i = 0; i < nthreads; i++) {
166 scfs.n_resched += scf_stats_p[i].n_resched;
167 scfs.n_single += scf_stats_p[i].n_single;
168 scfs.n_single_ofl += scf_stats_p[i].n_single_ofl;
169 scfs.n_single_rpc += scf_stats_p[i].n_single_rpc;
170 scfs.n_single_wait += scf_stats_p[i].n_single_wait;
171 scfs.n_single_wait_ofl += scf_stats_p[i].n_single_wait_ofl;
172 scfs.n_many += scf_stats_p[i].n_many;
173 scfs.n_many_wait += scf_stats_p[i].n_many_wait;
174 scfs.n_all += scf_stats_p[i].n_all;
175 scfs.n_all_wait += scf_stats_p[i].n_all_wait;
177 if (atomic_read(&n_errs) || atomic_read(&n_mb_in_errs) ||
178 atomic_read(&n_mb_out_errs) || atomic_read(&n_alloc_errs))
180 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 ",
181 SCFTORT_FLAG, bangstr, isdone ? "VER" : "ver", invoked_count, scfs.n_resched,
182 scfs.n_single, scfs.n_single_wait, scfs.n_single_ofl, scfs.n_single_wait_ofl,
183 scfs.n_single_rpc, scfs.n_single_rpc_ofl,
184 scfs.n_many, scfs.n_many_wait, scfs.n_all, scfs.n_all_wait);
185 torture_onoff_stats();
186 pr_cont("ste: %d stnmie: %d stnmoe: %d staf: %d\n", atomic_read(&n_errs),
187 atomic_read(&n_mb_in_errs), atomic_read(&n_mb_out_errs),
188 atomic_read(&n_alloc_errs));
191 // Periodically prints torture statistics, if periodic statistics printing
192 // was specified via the stat_interval module parameter.
194 scf_torture_stats(void *arg)
196 VERBOSE_TOROUT_STRING("scf_torture_stats task started");
198 schedule_timeout_interruptible(stat_interval * HZ);
199 scf_torture_stats_print();
200 torture_shutdown_absorb("scf_torture_stats");
201 } while (!torture_must_stop());
202 torture_kthread_stopping("scf_torture_stats");
206 // Add a primitive to the scf_sel_array[].
207 static void scf_sel_add(unsigned long weight, int prim, bool wait)
209 struct scf_selector *scfsp = &scf_sel_array[scf_sel_array_len];
211 // If no weight, if array would overflow, if computing three-place
212 // percentages would overflow, or if the scf_prim_name[] array would
213 // overflow, don't bother. In the last three two cases, complain.
215 WARN_ON_ONCE(scf_sel_array_len >= ARRAY_SIZE(scf_sel_array)) ||
216 WARN_ON_ONCE(0 - 100000 * weight <= 100000 * scf_sel_totweight) ||
217 WARN_ON_ONCE(prim >= ARRAY_SIZE(scf_prim_name)))
219 scf_sel_totweight += weight;
220 scfsp->scfs_weight = scf_sel_totweight;
221 scfsp->scfs_prim = prim;
222 scfsp->scfs_wait = wait;
226 // Dump out weighting percentages for scf_prim_name[] array.
227 static void scf_sel_dump(void)
230 unsigned long oldw = 0;
231 struct scf_selector *scfsp;
234 for (i = 0; i < scf_sel_array_len; i++) {
235 scfsp = &scf_sel_array[i];
236 w = (scfsp->scfs_weight - oldw) * 100000 / scf_sel_totweight;
237 pr_info("%s: %3lu.%03lu %s(%s)\n", __func__, w / 1000, w % 1000,
238 scf_prim_name[scfsp->scfs_prim],
239 scfsp->scfs_wait ? "wait" : "nowait");
240 oldw = scfsp->scfs_weight;
244 // Randomly pick a primitive and wait/nowait, based on weightings.
245 static struct scf_selector *scf_sel_rand(struct torture_random_state *trsp)
248 unsigned long w = torture_random(trsp) % (scf_sel_totweight + 1);
250 for (i = 0; i < scf_sel_array_len; i++)
251 if (scf_sel_array[i].scfs_weight >= w)
252 return &scf_sel_array[i];
254 return &scf_sel_array[0];
257 // Update statistics and occasionally burn up mass quantities of CPU time,
258 // if told to do so via scftorture.longwait. Otherwise, occasionally burn
260 static void scf_handler(void *scfc_in)
264 unsigned long r = torture_random(this_cpu_ptr(&scf_torture_rand));
265 struct scf_check *scfcp = scfc_in;
268 WRITE_ONCE(scfcp->scfc_out, false); // For multiple receivers.
269 if (WARN_ON_ONCE(unlikely(!READ_ONCE(scfcp->scfc_in))))
270 atomic_inc(&n_mb_in_errs);
272 this_cpu_inc(scf_invoked_count);
282 udelay((r & 0xff) + 1);
285 r = r % longwait + 1;
286 for (i = 0; i < r; i++) {
287 for (j = 0; j < 1000; j++) {
293 if (unlikely(!scfcp))
295 if (scfcp->scfc_wait) {
296 WRITE_ONCE(scfcp->scfc_out, true);
298 complete(&scfcp->scfc_completion);
304 // As above, but check for correct CPU.
305 static void scf_handler_1(void *scfc_in)
307 struct scf_check *scfcp = scfc_in;
309 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())) {
315 // Randomly do an smp_call_function*() invocation.
316 static void scftorture_invoke_one(struct scf_statistics *scfp, struct torture_random_state *trsp)
320 struct scf_check *scfcp = NULL;
321 struct scf_selector *scfsp = scf_sel_rand(trsp);
323 if (use_cpus_read_lock)
327 if (scfsp->scfs_prim == SCF_PRIM_SINGLE || scfsp->scfs_wait) {
328 scfcp = kmalloc(sizeof(*scfcp), GFP_ATOMIC);
329 if (WARN_ON_ONCE(!scfcp)) {
330 atomic_inc(&n_alloc_errs);
332 scfcp->scfc_cpu = -1;
333 scfcp->scfc_wait = scfsp->scfs_wait;
334 scfcp->scfc_out = false;
335 scfcp->scfc_rpc = false;
338 switch (scfsp->scfs_prim) {
339 case SCF_PRIM_RESCHED:
340 if (IS_BUILTIN(CONFIG_SCF_TORTURE_TEST)) {
341 cpu = torture_random(trsp) % nr_cpu_ids;
346 case SCF_PRIM_SINGLE:
347 cpu = torture_random(trsp) % nr_cpu_ids;
348 if (scfsp->scfs_wait)
349 scfp->n_single_wait++;
353 scfcp->scfc_cpu = cpu;
354 barrier(); // Prevent race-reduction compiler optimizations.
355 scfcp->scfc_in = true;
357 ret = smp_call_function_single(cpu, scf_handler_1, (void *)scfcp, scfsp->scfs_wait);
359 if (scfsp->scfs_wait)
360 scfp->n_single_wait_ofl++;
362 scfp->n_single_ofl++;
367 case SCF_PRIM_SINGLE_RPC:
370 cpu = torture_random(trsp) % nr_cpu_ids;
371 scfp->n_single_rpc++;
372 scfcp->scfc_cpu = cpu;
373 scfcp->scfc_wait = true;
374 init_completion(&scfcp->scfc_completion);
375 scfcp->scfc_rpc = true;
376 barrier(); // Prevent race-reduction compiler optimizations.
377 scfcp->scfc_in = true;
378 ret = smp_call_function_single(cpu, scf_handler_1, (void *)scfcp, 0);
380 if (use_cpus_read_lock)
384 wait_for_completion(&scfcp->scfc_completion);
385 if (use_cpus_read_lock)
390 scfp->n_single_rpc_ofl++;
396 if (scfsp->scfs_wait)
401 barrier(); // Prevent race-reduction compiler optimizations.
402 scfcp->scfc_in = true;
404 smp_call_function_many(cpu_online_mask, scf_handler, scfcp, scfsp->scfs_wait);
407 if (scfsp->scfs_wait)
412 barrier(); // Prevent race-reduction compiler optimizations.
413 scfcp->scfc_in = true;
415 smp_call_function(scf_handler, scfcp, scfsp->scfs_wait);
420 scfcp->scfc_out = true;
422 if (scfcp && scfsp->scfs_wait) {
423 if (WARN_ON_ONCE((num_online_cpus() > 1 || scfsp->scfs_prim == SCF_PRIM_SINGLE) &&
425 pr_warn("%s: Memory-ordering failure, scfs_prim: %d.\n", __func__, scfsp->scfs_prim);
426 atomic_inc(&n_mb_out_errs); // Leak rather than trash!
430 barrier(); // Prevent race-reduction compiler optimizations.
432 if (use_cpus_read_lock)
436 if (!(torture_random(trsp) & 0xfff))
437 schedule_timeout_uninterruptible(1);
440 // SCF test kthread. Repeatedly does calls to members of the
441 // smp_call_function() family of functions.
442 static int scftorture_invoker(void *arg)
446 DEFINE_TORTURE_RANDOM(rand);
447 struct scf_statistics *scfp = (struct scf_statistics *)arg;
448 bool was_offline = false;
450 VERBOSE_SCFTORTOUT("scftorture_invoker %d: task started", scfp->cpu);
451 cpu = scfp->cpu % nr_cpu_ids;
452 WARN_ON_ONCE(set_cpus_allowed_ptr(current, cpumask_of(cpu)));
453 set_user_nice(current, MAX_NICE);
455 schedule_timeout_interruptible(holdoff * HZ);
457 VERBOSE_SCFTORTOUT("scftorture_invoker %d: Waiting for all SCF torturers from cpu %d", scfp->cpu, raw_smp_processor_id());
459 // Make sure that the CPU is affinitized appropriately during testing.
460 curcpu = raw_smp_processor_id();
461 WARN_ONCE(curcpu != scfp->cpu % nr_cpu_ids,
462 "%s: Wanted CPU %d, running on %d, nr_cpu_ids = %d\n",
463 __func__, scfp->cpu, curcpu, nr_cpu_ids);
465 if (!atomic_dec_return(&n_started))
466 while (atomic_read_acquire(&n_started)) {
467 if (torture_must_stop()) {
468 VERBOSE_SCFTORTOUT("scftorture_invoker %d ended before starting", scfp->cpu);
471 schedule_timeout_uninterruptible(1);
474 VERBOSE_SCFTORTOUT("scftorture_invoker %d started", scfp->cpu);
477 scftorture_invoke_one(scfp, &rand);
478 while (cpu_is_offline(cpu) && !torture_must_stop()) {
479 schedule_timeout_interruptible(HZ / 5);
483 set_cpus_allowed_ptr(current, cpumask_of(cpu));
487 stutter_wait("scftorture_invoker");
488 } while (!torture_must_stop());
490 VERBOSE_SCFTORTOUT("scftorture_invoker %d ended", scfp->cpu);
492 torture_kthread_stopping("scftorture_invoker");
497 scftorture_print_module_parms(const char *tag)
499 pr_alert(SCFTORT_FLAG
500 "--- %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,
501 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);
504 static void scf_cleanup_handler(void *unused)
508 static void scf_torture_cleanup(void)
512 if (torture_cleanup_begin())
515 WRITE_ONCE(scfdone, true);
516 if (nthreads && scf_stats_p)
517 for (i = 0; i < nthreads; i++)
518 torture_stop_kthread("scftorture_invoker", scf_stats_p[i].task);
521 smp_call_function(scf_cleanup_handler, NULL, 0);
522 torture_stop_kthread(scf_torture_stats, scf_torture_stats_task);
523 scf_torture_stats_print(); // -After- the stats thread is stopped!
524 kfree(scf_stats_p); // -After- the last stats print has completed!
527 if (atomic_read(&n_errs) || atomic_read(&n_mb_in_errs) || atomic_read(&n_mb_out_errs))
528 scftorture_print_module_parms("End of test: FAILURE");
529 else if (torture_onoff_failures())
530 scftorture_print_module_parms("End of test: LOCK_HOTPLUG");
532 scftorture_print_module_parms("End of test: SUCCESS");
535 torture_cleanup_end();
538 static int __init scf_torture_init(void)
542 unsigned long weight_resched1 = weight_resched;
543 unsigned long weight_single1 = weight_single;
544 unsigned long weight_single_rpc1 = weight_single_rpc;
545 unsigned long weight_single_wait1 = weight_single_wait;
546 unsigned long weight_many1 = weight_many;
547 unsigned long weight_many_wait1 = weight_many_wait;
548 unsigned long weight_all1 = weight_all;
549 unsigned long weight_all_wait1 = weight_all_wait;
551 if (!torture_init_begin(SCFTORT_STRING, verbose))
554 scftorture_print_module_parms("Start of test");
556 if (weight_resched == -1 &&
557 weight_single == -1 && weight_single_rpc == -1 && weight_single_wait == -1 &&
558 weight_many == -1 && weight_many_wait == -1 &&
559 weight_all == -1 && weight_all_wait == -1) {
560 weight_resched1 = 2 * nr_cpu_ids;
561 weight_single1 = 2 * nr_cpu_ids;
562 weight_single_rpc1 = 2 * nr_cpu_ids;
563 weight_single_wait1 = 2 * nr_cpu_ids;
565 weight_many_wait1 = 2;
567 weight_all_wait1 = 1;
569 if (weight_resched == -1)
571 if (weight_single == -1)
573 if (weight_single_rpc == -1)
574 weight_single_rpc1 = 0;
575 if (weight_single_wait == -1)
576 weight_single_wait1 = 0;
577 if (weight_many == -1)
579 if (weight_many_wait == -1)
580 weight_many_wait1 = 0;
581 if (weight_all == -1)
583 if (weight_all_wait == -1)
584 weight_all_wait1 = 0;
586 if (weight_single1 == 0 && weight_single_rpc1 == 0 && weight_single_wait1 == 0 &&
587 weight_many1 == 0 && weight_many_wait1 == 0 &&
588 weight_all1 == 0 && weight_all_wait1 == 0) {
589 VERBOSE_SCFTORTOUT_ERRSTRING("all zero weights makes no sense");
593 if (IS_BUILTIN(CONFIG_SCF_TORTURE_TEST))
594 scf_sel_add(weight_resched1, SCF_PRIM_RESCHED, false);
595 else if (weight_resched1)
596 VERBOSE_SCFTORTOUT_ERRSTRING("built as module, weight_resched ignored");
597 scf_sel_add(weight_single1, SCF_PRIM_SINGLE, false);
598 scf_sel_add(weight_single_rpc1, SCF_PRIM_SINGLE_RPC, true);
599 scf_sel_add(weight_single_wait1, SCF_PRIM_SINGLE, true);
600 scf_sel_add(weight_many1, SCF_PRIM_MANY, false);
601 scf_sel_add(weight_many_wait1, SCF_PRIM_MANY, true);
602 scf_sel_add(weight_all1, SCF_PRIM_ALL, false);
603 scf_sel_add(weight_all_wait1, SCF_PRIM_ALL, true);
606 if (onoff_interval > 0) {
607 firsterr = torture_onoff_init(onoff_holdoff * HZ, onoff_interval, NULL);
611 if (shutdown_secs > 0) {
612 firsterr = torture_shutdown_init(shutdown_secs, scf_torture_cleanup);
617 firsterr = torture_stutter_init(stutter, stutter);
622 // Worker tasks invoking smp_call_function().
624 nthreads = num_online_cpus();
625 scf_stats_p = kcalloc(nthreads, sizeof(scf_stats_p[0]), GFP_KERNEL);
627 VERBOSE_SCFTORTOUT_ERRSTRING("out of memory");
632 VERBOSE_SCFTORTOUT("Starting %d smp_call_function() threads\n", nthreads);
634 atomic_set(&n_started, nthreads);
635 for (i = 0; i < nthreads; i++) {
636 scf_stats_p[i].cpu = i;
637 firsterr = torture_create_kthread(scftorture_invoker, (void *)&scf_stats_p[i],
638 scf_stats_p[i].task);
642 if (stat_interval > 0) {
643 firsterr = torture_create_kthread(scf_torture_stats, NULL, scf_torture_stats_task);
653 scf_torture_cleanup();
657 module_init(scf_torture_init);
658 module_exit(scf_torture_cleanup);