Merge tag 'usb-6.8-rc3' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/usb
[linux-block.git] / kernel / rcu / rcutorture.c
1 // SPDX-License-Identifier: GPL-2.0+
2 /*
3  * Read-Copy Update module-based torture test facility
4  *
5  * Copyright (C) IBM Corporation, 2005, 2006
6  *
7  * Authors: Paul E. McKenney <paulmck@linux.ibm.com>
8  *        Josh Triplett <josh@joshtriplett.org>
9  *
10  * See also:  Documentation/RCU/torture.rst
11  */
12
13 #define pr_fmt(fmt) fmt
14
15 #include <linux/types.h>
16 #include <linux/kernel.h>
17 #include <linux/init.h>
18 #include <linux/module.h>
19 #include <linux/kthread.h>
20 #include <linux/err.h>
21 #include <linux/spinlock.h>
22 #include <linux/smp.h>
23 #include <linux/rcupdate_wait.h>
24 #include <linux/rcu_notifier.h>
25 #include <linux/interrupt.h>
26 #include <linux/sched/signal.h>
27 #include <uapi/linux/sched/types.h>
28 #include <linux/atomic.h>
29 #include <linux/bitops.h>
30 #include <linux/completion.h>
31 #include <linux/moduleparam.h>
32 #include <linux/percpu.h>
33 #include <linux/notifier.h>
34 #include <linux/reboot.h>
35 #include <linux/freezer.h>
36 #include <linux/cpu.h>
37 #include <linux/delay.h>
38 #include <linux/stat.h>
39 #include <linux/srcu.h>
40 #include <linux/slab.h>
41 #include <linux/trace_clock.h>
42 #include <asm/byteorder.h>
43 #include <linux/torture.h>
44 #include <linux/vmalloc.h>
45 #include <linux/sched/debug.h>
46 #include <linux/sched/sysctl.h>
47 #include <linux/oom.h>
48 #include <linux/tick.h>
49 #include <linux/rcupdate_trace.h>
50 #include <linux/nmi.h>
51
52 #include "rcu.h"
53
54 MODULE_LICENSE("GPL");
55 MODULE_AUTHOR("Paul E. McKenney <paulmck@linux.ibm.com> and Josh Triplett <josh@joshtriplett.org>");
56
57 /* Bits for ->extendables field, extendables param, and related definitions. */
58 #define RCUTORTURE_RDR_SHIFT_1   8      /* Put SRCU index in upper bits. */
59 #define RCUTORTURE_RDR_MASK_1    (1 << RCUTORTURE_RDR_SHIFT_1)
60 #define RCUTORTURE_RDR_SHIFT_2   9      /* Put SRCU index in upper bits. */
61 #define RCUTORTURE_RDR_MASK_2    (1 << RCUTORTURE_RDR_SHIFT_2)
62 #define RCUTORTURE_RDR_BH        0x01   /* Extend readers by disabling bh. */
63 #define RCUTORTURE_RDR_IRQ       0x02   /*  ... disabling interrupts. */
64 #define RCUTORTURE_RDR_PREEMPT   0x04   /*  ... disabling preemption. */
65 #define RCUTORTURE_RDR_RBH       0x08   /*  ... rcu_read_lock_bh(). */
66 #define RCUTORTURE_RDR_SCHED     0x10   /*  ... rcu_read_lock_sched(). */
67 #define RCUTORTURE_RDR_RCU_1     0x20   /*  ... entering another RCU reader. */
68 #define RCUTORTURE_RDR_RCU_2     0x40   /*  ... entering another RCU reader. */
69 #define RCUTORTURE_RDR_NBITS     7      /* Number of bits defined above. */
70 #define RCUTORTURE_MAX_EXTEND    \
71         (RCUTORTURE_RDR_BH | RCUTORTURE_RDR_IRQ | RCUTORTURE_RDR_PREEMPT | \
72          RCUTORTURE_RDR_RBH | RCUTORTURE_RDR_SCHED)
73 #define RCUTORTURE_RDR_MAX_LOOPS 0x7    /* Maximum reader extensions. */
74                                         /* Must be power of two minus one. */
75 #define RCUTORTURE_RDR_MAX_SEGS (RCUTORTURE_RDR_MAX_LOOPS + 3)
76
77 torture_param(int, extendables, RCUTORTURE_MAX_EXTEND,
78               "Extend readers by disabling bh (1), irqs (2), or preempt (4)");
79 torture_param(int, fqs_duration, 0, "Duration of fqs bursts (us), 0 to disable");
80 torture_param(int, fqs_holdoff, 0, "Holdoff time within fqs bursts (us)");
81 torture_param(int, fqs_stutter, 3, "Wait time between fqs bursts (s)");
82 torture_param(int, fwd_progress, 1, "Number of grace-period forward progress tasks (0 to disable)");
83 torture_param(int, fwd_progress_div, 4, "Fraction of CPU stall to wait");
84 torture_param(int, fwd_progress_holdoff, 60, "Time between forward-progress tests (s)");
85 torture_param(bool, fwd_progress_need_resched, 1, "Hide cond_resched() behind need_resched()");
86 torture_param(bool, gp_cond, false, "Use conditional/async GP wait primitives");
87 torture_param(bool, gp_cond_exp, false, "Use conditional/async expedited GP wait primitives");
88 torture_param(bool, gp_cond_full, false, "Use conditional/async full-state GP wait primitives");
89 torture_param(bool, gp_cond_exp_full, false,
90                     "Use conditional/async full-stateexpedited GP wait primitives");
91 torture_param(bool, gp_exp, false, "Use expedited GP wait primitives");
92 torture_param(bool, gp_normal, false, "Use normal (non-expedited) GP wait primitives");
93 torture_param(bool, gp_poll, false, "Use polling GP wait primitives");
94 torture_param(bool, gp_poll_exp, false, "Use polling expedited GP wait primitives");
95 torture_param(bool, gp_poll_full, false, "Use polling full-state GP wait primitives");
96 torture_param(bool, gp_poll_exp_full, false, "Use polling full-state expedited GP wait primitives");
97 torture_param(bool, gp_sync, false, "Use synchronous GP wait primitives");
98 torture_param(int, irqreader, 1, "Allow RCU readers from irq handlers");
99 torture_param(int, leakpointer, 0, "Leak pointer dereferences from readers");
100 torture_param(int, n_barrier_cbs, 0, "# of callbacks/kthreads for barrier testing");
101 torture_param(int, nfakewriters, 4, "Number of RCU fake writer threads");
102 torture_param(int, nreaders, -1, "Number of RCU reader threads");
103 torture_param(int, object_debug, 0, "Enable debug-object double call_rcu() testing");
104 torture_param(int, onoff_holdoff, 0, "Time after boot before CPU hotplugs (s)");
105 torture_param(int, onoff_interval, 0, "Time between CPU hotplugs (jiffies), 0=disable");
106 torture_param(int, nocbs_nthreads, 0, "Number of NOCB toggle threads, 0 to disable");
107 torture_param(int, nocbs_toggle, 1000, "Time between toggling nocb state (ms)");
108 torture_param(int, read_exit_delay, 13, "Delay between read-then-exit episodes (s)");
109 torture_param(int, read_exit_burst, 16, "# of read-then-exit bursts per episode, zero to disable");
110 torture_param(int, shuffle_interval, 3, "Number of seconds between shuffles");
111 torture_param(int, shutdown_secs, 0, "Shutdown time (s), <= zero to disable.");
112 torture_param(int, stall_cpu, 0, "Stall duration (s), zero to disable.");
113 torture_param(int, stall_cpu_holdoff, 10, "Time to wait before starting stall (s).");
114 torture_param(bool, stall_no_softlockup, false, "Avoid softlockup warning during cpu stall.");
115 torture_param(int, stall_cpu_irqsoff, 0, "Disable interrupts while stalling.");
116 torture_param(int, stall_cpu_block, 0, "Sleep while stalling.");
117 torture_param(int, stall_gp_kthread, 0, "Grace-period kthread stall duration (s).");
118 torture_param(int, stat_interval, 60, "Number of seconds between stats printk()s");
119 torture_param(int, stutter, 5, "Number of seconds to run/halt test");
120 torture_param(int, test_boost, 1, "Test RCU prio boost: 0=no, 1=maybe, 2=yes.");
121 torture_param(int, test_boost_duration, 4, "Duration of each boost test, seconds.");
122 torture_param(int, test_boost_interval, 7, "Interval between boost tests, seconds.");
123 torture_param(int, test_nmis, 0, "End-test NMI tests, 0 to disable.");
124 torture_param(bool, test_no_idle_hz, true, "Test support for tickless idle CPUs");
125 torture_param(int, test_srcu_lockdep, 0, "Test specified SRCU deadlock scenario.");
126 torture_param(int, verbose, 1, "Enable verbose debugging printk()s");
127
128 static char *torture_type = "rcu";
129 module_param(torture_type, charp, 0444);
130 MODULE_PARM_DESC(torture_type, "Type of RCU to torture (rcu, srcu, ...)");
131
132 static int nrealnocbers;
133 static int nrealreaders;
134 static struct task_struct *writer_task;
135 static struct task_struct **fakewriter_tasks;
136 static struct task_struct **reader_tasks;
137 static struct task_struct **nocb_tasks;
138 static struct task_struct *stats_task;
139 static struct task_struct *fqs_task;
140 static struct task_struct *boost_tasks[NR_CPUS];
141 static struct task_struct *stall_task;
142 static struct task_struct **fwd_prog_tasks;
143 static struct task_struct **barrier_cbs_tasks;
144 static struct task_struct *barrier_task;
145 static struct task_struct *read_exit_task;
146
147 #define RCU_TORTURE_PIPE_LEN 10
148
149 // Mailbox-like structure to check RCU global memory ordering.
150 struct rcu_torture_reader_check {
151         unsigned long rtc_myloops;
152         int rtc_chkrdr;
153         unsigned long rtc_chkloops;
154         int rtc_ready;
155         struct rcu_torture_reader_check *rtc_assigner;
156 } ____cacheline_internodealigned_in_smp;
157
158 // Update-side data structure used to check RCU readers.
159 struct rcu_torture {
160         struct rcu_head rtort_rcu;
161         int rtort_pipe_count;
162         struct list_head rtort_free;
163         int rtort_mbtest;
164         struct rcu_torture_reader_check *rtort_chkp;
165 };
166
167 static LIST_HEAD(rcu_torture_freelist);
168 static struct rcu_torture __rcu *rcu_torture_current;
169 static unsigned long rcu_torture_current_version;
170 static struct rcu_torture rcu_tortures[10 * RCU_TORTURE_PIPE_LEN];
171 static DEFINE_SPINLOCK(rcu_torture_lock);
172 static DEFINE_PER_CPU(long [RCU_TORTURE_PIPE_LEN + 1], rcu_torture_count);
173 static DEFINE_PER_CPU(long [RCU_TORTURE_PIPE_LEN + 1], rcu_torture_batch);
174 static atomic_t rcu_torture_wcount[RCU_TORTURE_PIPE_LEN + 1];
175 static struct rcu_torture_reader_check *rcu_torture_reader_mbchk;
176 static atomic_t n_rcu_torture_alloc;
177 static atomic_t n_rcu_torture_alloc_fail;
178 static atomic_t n_rcu_torture_free;
179 static atomic_t n_rcu_torture_mberror;
180 static atomic_t n_rcu_torture_mbchk_fail;
181 static atomic_t n_rcu_torture_mbchk_tries;
182 static atomic_t n_rcu_torture_error;
183 static long n_rcu_torture_barrier_error;
184 static long n_rcu_torture_boost_ktrerror;
185 static long n_rcu_torture_boost_failure;
186 static long n_rcu_torture_boosts;
187 static atomic_long_t n_rcu_torture_timers;
188 static long n_barrier_attempts;
189 static long n_barrier_successes; /* did rcu_barrier test succeed? */
190 static unsigned long n_read_exits;
191 static struct list_head rcu_torture_removed;
192 static unsigned long shutdown_jiffies;
193 static unsigned long start_gp_seq;
194 static atomic_long_t n_nocb_offload;
195 static atomic_long_t n_nocb_deoffload;
196
197 static int rcu_torture_writer_state;
198 #define RTWS_FIXED_DELAY        0
199 #define RTWS_DELAY              1
200 #define RTWS_REPLACE            2
201 #define RTWS_DEF_FREE           3
202 #define RTWS_EXP_SYNC           4
203 #define RTWS_COND_GET           5
204 #define RTWS_COND_GET_FULL      6
205 #define RTWS_COND_GET_EXP       7
206 #define RTWS_COND_GET_EXP_FULL  8
207 #define RTWS_COND_SYNC          9
208 #define RTWS_COND_SYNC_FULL     10
209 #define RTWS_COND_SYNC_EXP      11
210 #define RTWS_COND_SYNC_EXP_FULL 12
211 #define RTWS_POLL_GET           13
212 #define RTWS_POLL_GET_FULL      14
213 #define RTWS_POLL_GET_EXP       15
214 #define RTWS_POLL_GET_EXP_FULL  16
215 #define RTWS_POLL_WAIT          17
216 #define RTWS_POLL_WAIT_FULL     18
217 #define RTWS_POLL_WAIT_EXP      19
218 #define RTWS_POLL_WAIT_EXP_FULL 20
219 #define RTWS_SYNC               21
220 #define RTWS_STUTTER            22
221 #define RTWS_STOPPING           23
222 static const char * const rcu_torture_writer_state_names[] = {
223         "RTWS_FIXED_DELAY",
224         "RTWS_DELAY",
225         "RTWS_REPLACE",
226         "RTWS_DEF_FREE",
227         "RTWS_EXP_SYNC",
228         "RTWS_COND_GET",
229         "RTWS_COND_GET_FULL",
230         "RTWS_COND_GET_EXP",
231         "RTWS_COND_GET_EXP_FULL",
232         "RTWS_COND_SYNC",
233         "RTWS_COND_SYNC_FULL",
234         "RTWS_COND_SYNC_EXP",
235         "RTWS_COND_SYNC_EXP_FULL",
236         "RTWS_POLL_GET",
237         "RTWS_POLL_GET_FULL",
238         "RTWS_POLL_GET_EXP",
239         "RTWS_POLL_GET_EXP_FULL",
240         "RTWS_POLL_WAIT",
241         "RTWS_POLL_WAIT_FULL",
242         "RTWS_POLL_WAIT_EXP",
243         "RTWS_POLL_WAIT_EXP_FULL",
244         "RTWS_SYNC",
245         "RTWS_STUTTER",
246         "RTWS_STOPPING",
247 };
248
249 /* Record reader segment types and duration for first failing read. */
250 struct rt_read_seg {
251         int rt_readstate;
252         unsigned long rt_delay_jiffies;
253         unsigned long rt_delay_ms;
254         unsigned long rt_delay_us;
255         bool rt_preempted;
256 };
257 static int err_segs_recorded;
258 static struct rt_read_seg err_segs[RCUTORTURE_RDR_MAX_SEGS];
259 static int rt_read_nsegs;
260
261 static const char *rcu_torture_writer_state_getname(void)
262 {
263         unsigned int i = READ_ONCE(rcu_torture_writer_state);
264
265         if (i >= ARRAY_SIZE(rcu_torture_writer_state_names))
266                 return "???";
267         return rcu_torture_writer_state_names[i];
268 }
269
270 #ifdef CONFIG_RCU_TRACE
271 static u64 notrace rcu_trace_clock_local(void)
272 {
273         u64 ts = trace_clock_local();
274
275         (void)do_div(ts, NSEC_PER_USEC);
276         return ts;
277 }
278 #else /* #ifdef CONFIG_RCU_TRACE */
279 static u64 notrace rcu_trace_clock_local(void)
280 {
281         return 0ULL;
282 }
283 #endif /* #else #ifdef CONFIG_RCU_TRACE */
284
285 /*
286  * Stop aggressive CPU-hog tests a bit before the end of the test in order
287  * to avoid interfering with test shutdown.
288  */
289 static bool shutdown_time_arrived(void)
290 {
291         return shutdown_secs && time_after(jiffies, shutdown_jiffies - 30 * HZ);
292 }
293
294 static unsigned long boost_starttime;   /* jiffies of next boost test start. */
295 static DEFINE_MUTEX(boost_mutex);       /* protect setting boost_starttime */
296                                         /*  and boost task create/destroy. */
297 static atomic_t barrier_cbs_count;      /* Barrier callbacks registered. */
298 static bool barrier_phase;              /* Test phase. */
299 static atomic_t barrier_cbs_invoked;    /* Barrier callbacks invoked. */
300 static wait_queue_head_t *barrier_cbs_wq; /* Coordinate barrier testing. */
301 static DECLARE_WAIT_QUEUE_HEAD(barrier_wq);
302
303 static atomic_t rcu_fwd_cb_nodelay;     /* Short rcu_torture_delay() delays. */
304
305 /*
306  * Allocate an element from the rcu_tortures pool.
307  */
308 static struct rcu_torture *
309 rcu_torture_alloc(void)
310 {
311         struct list_head *p;
312
313         spin_lock_bh(&rcu_torture_lock);
314         if (list_empty(&rcu_torture_freelist)) {
315                 atomic_inc(&n_rcu_torture_alloc_fail);
316                 spin_unlock_bh(&rcu_torture_lock);
317                 return NULL;
318         }
319         atomic_inc(&n_rcu_torture_alloc);
320         p = rcu_torture_freelist.next;
321         list_del_init(p);
322         spin_unlock_bh(&rcu_torture_lock);
323         return container_of(p, struct rcu_torture, rtort_free);
324 }
325
326 /*
327  * Free an element to the rcu_tortures pool.
328  */
329 static void
330 rcu_torture_free(struct rcu_torture *p)
331 {
332         atomic_inc(&n_rcu_torture_free);
333         spin_lock_bh(&rcu_torture_lock);
334         list_add_tail(&p->rtort_free, &rcu_torture_freelist);
335         spin_unlock_bh(&rcu_torture_lock);
336 }
337
338 /*
339  * Operations vector for selecting different types of tests.
340  */
341
342 struct rcu_torture_ops {
343         int ttype;
344         void (*init)(void);
345         void (*cleanup)(void);
346         int (*readlock)(void);
347         void (*read_delay)(struct torture_random_state *rrsp,
348                            struct rt_read_seg *rtrsp);
349         void (*readunlock)(int idx);
350         int (*readlock_held)(void);
351         unsigned long (*get_gp_seq)(void);
352         unsigned long (*gp_diff)(unsigned long new, unsigned long old);
353         void (*deferred_free)(struct rcu_torture *p);
354         void (*sync)(void);
355         void (*exp_sync)(void);
356         unsigned long (*get_gp_state_exp)(void);
357         unsigned long (*start_gp_poll_exp)(void);
358         void (*start_gp_poll_exp_full)(struct rcu_gp_oldstate *rgosp);
359         bool (*poll_gp_state_exp)(unsigned long oldstate);
360         void (*cond_sync_exp)(unsigned long oldstate);
361         void (*cond_sync_exp_full)(struct rcu_gp_oldstate *rgosp);
362         unsigned long (*get_comp_state)(void);
363         void (*get_comp_state_full)(struct rcu_gp_oldstate *rgosp);
364         bool (*same_gp_state)(unsigned long oldstate1, unsigned long oldstate2);
365         bool (*same_gp_state_full)(struct rcu_gp_oldstate *rgosp1, struct rcu_gp_oldstate *rgosp2);
366         unsigned long (*get_gp_state)(void);
367         void (*get_gp_state_full)(struct rcu_gp_oldstate *rgosp);
368         unsigned long (*get_gp_completed)(void);
369         void (*get_gp_completed_full)(struct rcu_gp_oldstate *rgosp);
370         unsigned long (*start_gp_poll)(void);
371         void (*start_gp_poll_full)(struct rcu_gp_oldstate *rgosp);
372         bool (*poll_gp_state)(unsigned long oldstate);
373         bool (*poll_gp_state_full)(struct rcu_gp_oldstate *rgosp);
374         bool (*poll_need_2gp)(bool poll, bool poll_full);
375         void (*cond_sync)(unsigned long oldstate);
376         void (*cond_sync_full)(struct rcu_gp_oldstate *rgosp);
377         call_rcu_func_t call;
378         void (*cb_barrier)(void);
379         void (*fqs)(void);
380         void (*stats)(void);
381         void (*gp_kthread_dbg)(void);
382         bool (*check_boost_failed)(unsigned long gp_state, int *cpup);
383         int (*stall_dur)(void);
384         long cbflood_max;
385         int irq_capable;
386         int can_boost;
387         int extendables;
388         int slow_gps;
389         int no_pi_lock;
390         const char *name;
391 };
392
393 static struct rcu_torture_ops *cur_ops;
394
395 /*
396  * Definitions for rcu torture testing.
397  */
398
399 static int torture_readlock_not_held(void)
400 {
401         return rcu_read_lock_bh_held() || rcu_read_lock_sched_held();
402 }
403
404 static int rcu_torture_read_lock(void)
405 {
406         rcu_read_lock();
407         return 0;
408 }
409
410 static void
411 rcu_read_delay(struct torture_random_state *rrsp, struct rt_read_seg *rtrsp)
412 {
413         unsigned long started;
414         unsigned long completed;
415         const unsigned long shortdelay_us = 200;
416         unsigned long longdelay_ms = 300;
417         unsigned long long ts;
418
419         /* We want a short delay sometimes to make a reader delay the grace
420          * period, and we want a long delay occasionally to trigger
421          * force_quiescent_state. */
422
423         if (!atomic_read(&rcu_fwd_cb_nodelay) &&
424             !(torture_random(rrsp) % (nrealreaders * 2000 * longdelay_ms))) {
425                 started = cur_ops->get_gp_seq();
426                 ts = rcu_trace_clock_local();
427                 if (preempt_count() & (SOFTIRQ_MASK | HARDIRQ_MASK))
428                         longdelay_ms = 5; /* Avoid triggering BH limits. */
429                 mdelay(longdelay_ms);
430                 rtrsp->rt_delay_ms = longdelay_ms;
431                 completed = cur_ops->get_gp_seq();
432                 do_trace_rcu_torture_read(cur_ops->name, NULL, ts,
433                                           started, completed);
434         }
435         if (!(torture_random(rrsp) % (nrealreaders * 2 * shortdelay_us))) {
436                 udelay(shortdelay_us);
437                 rtrsp->rt_delay_us = shortdelay_us;
438         }
439         if (!preempt_count() &&
440             !(torture_random(rrsp) % (nrealreaders * 500))) {
441                 torture_preempt_schedule();  /* QS only if preemptible. */
442                 rtrsp->rt_preempted = true;
443         }
444 }
445
446 static void rcu_torture_read_unlock(int idx)
447 {
448         rcu_read_unlock();
449 }
450
451 /*
452  * Update callback in the pipe.  This should be invoked after a grace period.
453  */
454 static bool
455 rcu_torture_pipe_update_one(struct rcu_torture *rp)
456 {
457         int i;
458         struct rcu_torture_reader_check *rtrcp = READ_ONCE(rp->rtort_chkp);
459
460         if (rtrcp) {
461                 WRITE_ONCE(rp->rtort_chkp, NULL);
462                 smp_store_release(&rtrcp->rtc_ready, 1); // Pair with smp_load_acquire().
463         }
464         i = READ_ONCE(rp->rtort_pipe_count);
465         if (i > RCU_TORTURE_PIPE_LEN)
466                 i = RCU_TORTURE_PIPE_LEN;
467         atomic_inc(&rcu_torture_wcount[i]);
468         WRITE_ONCE(rp->rtort_pipe_count, i + 1);
469         if (rp->rtort_pipe_count >= RCU_TORTURE_PIPE_LEN) {
470                 rp->rtort_mbtest = 0;
471                 return true;
472         }
473         return false;
474 }
475
476 /*
477  * Update all callbacks in the pipe.  Suitable for synchronous grace-period
478  * primitives.
479  */
480 static void
481 rcu_torture_pipe_update(struct rcu_torture *old_rp)
482 {
483         struct rcu_torture *rp;
484         struct rcu_torture *rp1;
485
486         if (old_rp)
487                 list_add(&old_rp->rtort_free, &rcu_torture_removed);
488         list_for_each_entry_safe(rp, rp1, &rcu_torture_removed, rtort_free) {
489                 if (rcu_torture_pipe_update_one(rp)) {
490                         list_del(&rp->rtort_free);
491                         rcu_torture_free(rp);
492                 }
493         }
494 }
495
496 static void
497 rcu_torture_cb(struct rcu_head *p)
498 {
499         struct rcu_torture *rp = container_of(p, struct rcu_torture, rtort_rcu);
500
501         if (torture_must_stop_irq()) {
502                 /* Test is ending, just drop callbacks on the floor. */
503                 /* The next initialization will pick up the pieces. */
504                 return;
505         }
506         if (rcu_torture_pipe_update_one(rp))
507                 rcu_torture_free(rp);
508         else
509                 cur_ops->deferred_free(rp);
510 }
511
512 static unsigned long rcu_no_completed(void)
513 {
514         return 0;
515 }
516
517 static void rcu_torture_deferred_free(struct rcu_torture *p)
518 {
519         call_rcu_hurry(&p->rtort_rcu, rcu_torture_cb);
520 }
521
522 static void rcu_sync_torture_init(void)
523 {
524         INIT_LIST_HEAD(&rcu_torture_removed);
525 }
526
527 static bool rcu_poll_need_2gp(bool poll, bool poll_full)
528 {
529         return poll;
530 }
531
532 static struct rcu_torture_ops rcu_ops = {
533         .ttype                  = RCU_FLAVOR,
534         .init                   = rcu_sync_torture_init,
535         .readlock               = rcu_torture_read_lock,
536         .read_delay             = rcu_read_delay,
537         .readunlock             = rcu_torture_read_unlock,
538         .readlock_held          = torture_readlock_not_held,
539         .get_gp_seq             = rcu_get_gp_seq,
540         .gp_diff                = rcu_seq_diff,
541         .deferred_free          = rcu_torture_deferred_free,
542         .sync                   = synchronize_rcu,
543         .exp_sync               = synchronize_rcu_expedited,
544         .same_gp_state          = same_state_synchronize_rcu,
545         .same_gp_state_full     = same_state_synchronize_rcu_full,
546         .get_comp_state         = get_completed_synchronize_rcu,
547         .get_comp_state_full    = get_completed_synchronize_rcu_full,
548         .get_gp_state           = get_state_synchronize_rcu,
549         .get_gp_state_full      = get_state_synchronize_rcu_full,
550         .get_gp_completed       = get_completed_synchronize_rcu,
551         .get_gp_completed_full  = get_completed_synchronize_rcu_full,
552         .start_gp_poll          = start_poll_synchronize_rcu,
553         .start_gp_poll_full     = start_poll_synchronize_rcu_full,
554         .poll_gp_state          = poll_state_synchronize_rcu,
555         .poll_gp_state_full     = poll_state_synchronize_rcu_full,
556         .poll_need_2gp          = rcu_poll_need_2gp,
557         .cond_sync              = cond_synchronize_rcu,
558         .cond_sync_full         = cond_synchronize_rcu_full,
559         .get_gp_state_exp       = get_state_synchronize_rcu,
560         .start_gp_poll_exp      = start_poll_synchronize_rcu_expedited,
561         .start_gp_poll_exp_full = start_poll_synchronize_rcu_expedited_full,
562         .poll_gp_state_exp      = poll_state_synchronize_rcu,
563         .cond_sync_exp          = cond_synchronize_rcu_expedited,
564         .call                   = call_rcu_hurry,
565         .cb_barrier             = rcu_barrier,
566         .fqs                    = rcu_force_quiescent_state,
567         .stats                  = NULL,
568         .gp_kthread_dbg         = show_rcu_gp_kthreads,
569         .check_boost_failed     = rcu_check_boost_fail,
570         .stall_dur              = rcu_jiffies_till_stall_check,
571         .irq_capable            = 1,
572         .can_boost              = IS_ENABLED(CONFIG_RCU_BOOST),
573         .extendables            = RCUTORTURE_MAX_EXTEND,
574         .name                   = "rcu"
575 };
576
577 /*
578  * Don't even think about trying any of these in real life!!!
579  * The names includes "busted", and they really means it!
580  * The only purpose of these functions is to provide a buggy RCU
581  * implementation to make sure that rcutorture correctly emits
582  * buggy-RCU error messages.
583  */
584 static void rcu_busted_torture_deferred_free(struct rcu_torture *p)
585 {
586         /* This is a deliberate bug for testing purposes only! */
587         rcu_torture_cb(&p->rtort_rcu);
588 }
589
590 static void synchronize_rcu_busted(void)
591 {
592         /* This is a deliberate bug for testing purposes only! */
593 }
594
595 static void
596 call_rcu_busted(struct rcu_head *head, rcu_callback_t func)
597 {
598         /* This is a deliberate bug for testing purposes only! */
599         func(head);
600 }
601
602 static struct rcu_torture_ops rcu_busted_ops = {
603         .ttype          = INVALID_RCU_FLAVOR,
604         .init           = rcu_sync_torture_init,
605         .readlock       = rcu_torture_read_lock,
606         .read_delay     = rcu_read_delay,  /* just reuse rcu's version. */
607         .readunlock     = rcu_torture_read_unlock,
608         .readlock_held  = torture_readlock_not_held,
609         .get_gp_seq     = rcu_no_completed,
610         .deferred_free  = rcu_busted_torture_deferred_free,
611         .sync           = synchronize_rcu_busted,
612         .exp_sync       = synchronize_rcu_busted,
613         .call           = call_rcu_busted,
614         .cb_barrier     = NULL,
615         .fqs            = NULL,
616         .stats          = NULL,
617         .irq_capable    = 1,
618         .name           = "busted"
619 };
620
621 /*
622  * Definitions for srcu torture testing.
623  */
624
625 DEFINE_STATIC_SRCU(srcu_ctl);
626 static struct srcu_struct srcu_ctld;
627 static struct srcu_struct *srcu_ctlp = &srcu_ctl;
628 static struct rcu_torture_ops srcud_ops;
629
630 static int srcu_torture_read_lock(void)
631 {
632         if (cur_ops == &srcud_ops)
633                 return srcu_read_lock_nmisafe(srcu_ctlp);
634         else
635                 return srcu_read_lock(srcu_ctlp);
636 }
637
638 static void
639 srcu_read_delay(struct torture_random_state *rrsp, struct rt_read_seg *rtrsp)
640 {
641         long delay;
642         const long uspertick = 1000000 / HZ;
643         const long longdelay = 10;
644
645         /* We want there to be long-running readers, but not all the time. */
646
647         delay = torture_random(rrsp) %
648                 (nrealreaders * 2 * longdelay * uspertick);
649         if (!delay && in_task()) {
650                 schedule_timeout_interruptible(longdelay);
651                 rtrsp->rt_delay_jiffies = longdelay;
652         } else {
653                 rcu_read_delay(rrsp, rtrsp);
654         }
655 }
656
657 static void srcu_torture_read_unlock(int idx)
658 {
659         if (cur_ops == &srcud_ops)
660                 srcu_read_unlock_nmisafe(srcu_ctlp, idx);
661         else
662                 srcu_read_unlock(srcu_ctlp, idx);
663 }
664
665 static int torture_srcu_read_lock_held(void)
666 {
667         return srcu_read_lock_held(srcu_ctlp);
668 }
669
670 static unsigned long srcu_torture_completed(void)
671 {
672         return srcu_batches_completed(srcu_ctlp);
673 }
674
675 static void srcu_torture_deferred_free(struct rcu_torture *rp)
676 {
677         call_srcu(srcu_ctlp, &rp->rtort_rcu, rcu_torture_cb);
678 }
679
680 static void srcu_torture_synchronize(void)
681 {
682         synchronize_srcu(srcu_ctlp);
683 }
684
685 static unsigned long srcu_torture_get_gp_state(void)
686 {
687         return get_state_synchronize_srcu(srcu_ctlp);
688 }
689
690 static unsigned long srcu_torture_start_gp_poll(void)
691 {
692         return start_poll_synchronize_srcu(srcu_ctlp);
693 }
694
695 static bool srcu_torture_poll_gp_state(unsigned long oldstate)
696 {
697         return poll_state_synchronize_srcu(srcu_ctlp, oldstate);
698 }
699
700 static void srcu_torture_call(struct rcu_head *head,
701                               rcu_callback_t func)
702 {
703         call_srcu(srcu_ctlp, head, func);
704 }
705
706 static void srcu_torture_barrier(void)
707 {
708         srcu_barrier(srcu_ctlp);
709 }
710
711 static void srcu_torture_stats(void)
712 {
713         srcu_torture_stats_print(srcu_ctlp, torture_type, TORTURE_FLAG);
714 }
715
716 static void srcu_torture_synchronize_expedited(void)
717 {
718         synchronize_srcu_expedited(srcu_ctlp);
719 }
720
721 static struct rcu_torture_ops srcu_ops = {
722         .ttype          = SRCU_FLAVOR,
723         .init           = rcu_sync_torture_init,
724         .readlock       = srcu_torture_read_lock,
725         .read_delay     = srcu_read_delay,
726         .readunlock     = srcu_torture_read_unlock,
727         .readlock_held  = torture_srcu_read_lock_held,
728         .get_gp_seq     = srcu_torture_completed,
729         .deferred_free  = srcu_torture_deferred_free,
730         .sync           = srcu_torture_synchronize,
731         .exp_sync       = srcu_torture_synchronize_expedited,
732         .get_gp_state   = srcu_torture_get_gp_state,
733         .start_gp_poll  = srcu_torture_start_gp_poll,
734         .poll_gp_state  = srcu_torture_poll_gp_state,
735         .call           = srcu_torture_call,
736         .cb_barrier     = srcu_torture_barrier,
737         .stats          = srcu_torture_stats,
738         .cbflood_max    = 50000,
739         .irq_capable    = 1,
740         .no_pi_lock     = IS_ENABLED(CONFIG_TINY_SRCU),
741         .name           = "srcu"
742 };
743
744 static void srcu_torture_init(void)
745 {
746         rcu_sync_torture_init();
747         WARN_ON(init_srcu_struct(&srcu_ctld));
748         srcu_ctlp = &srcu_ctld;
749 }
750
751 static void srcu_torture_cleanup(void)
752 {
753         cleanup_srcu_struct(&srcu_ctld);
754         srcu_ctlp = &srcu_ctl; /* In case of a later rcutorture run. */
755 }
756
757 /* As above, but dynamically allocated. */
758 static struct rcu_torture_ops srcud_ops = {
759         .ttype          = SRCU_FLAVOR,
760         .init           = srcu_torture_init,
761         .cleanup        = srcu_torture_cleanup,
762         .readlock       = srcu_torture_read_lock,
763         .read_delay     = srcu_read_delay,
764         .readunlock     = srcu_torture_read_unlock,
765         .readlock_held  = torture_srcu_read_lock_held,
766         .get_gp_seq     = srcu_torture_completed,
767         .deferred_free  = srcu_torture_deferred_free,
768         .sync           = srcu_torture_synchronize,
769         .exp_sync       = srcu_torture_synchronize_expedited,
770         .get_gp_state   = srcu_torture_get_gp_state,
771         .start_gp_poll  = srcu_torture_start_gp_poll,
772         .poll_gp_state  = srcu_torture_poll_gp_state,
773         .call           = srcu_torture_call,
774         .cb_barrier     = srcu_torture_barrier,
775         .stats          = srcu_torture_stats,
776         .cbflood_max    = 50000,
777         .irq_capable    = 1,
778         .no_pi_lock     = IS_ENABLED(CONFIG_TINY_SRCU),
779         .name           = "srcud"
780 };
781
782 /* As above, but broken due to inappropriate reader extension. */
783 static struct rcu_torture_ops busted_srcud_ops = {
784         .ttype          = SRCU_FLAVOR,
785         .init           = srcu_torture_init,
786         .cleanup        = srcu_torture_cleanup,
787         .readlock       = srcu_torture_read_lock,
788         .read_delay     = rcu_read_delay,
789         .readunlock     = srcu_torture_read_unlock,
790         .readlock_held  = torture_srcu_read_lock_held,
791         .get_gp_seq     = srcu_torture_completed,
792         .deferred_free  = srcu_torture_deferred_free,
793         .sync           = srcu_torture_synchronize,
794         .exp_sync       = srcu_torture_synchronize_expedited,
795         .call           = srcu_torture_call,
796         .cb_barrier     = srcu_torture_barrier,
797         .stats          = srcu_torture_stats,
798         .irq_capable    = 1,
799         .no_pi_lock     = IS_ENABLED(CONFIG_TINY_SRCU),
800         .extendables    = RCUTORTURE_MAX_EXTEND,
801         .name           = "busted_srcud"
802 };
803
804 /*
805  * Definitions for trivial CONFIG_PREEMPT=n-only torture testing.
806  * This implementation does not necessarily work well with CPU hotplug.
807  */
808
809 static void synchronize_rcu_trivial(void)
810 {
811         int cpu;
812
813         for_each_online_cpu(cpu) {
814                 torture_sched_setaffinity(current->pid, cpumask_of(cpu));
815                 WARN_ON_ONCE(raw_smp_processor_id() != cpu);
816         }
817 }
818
819 static int rcu_torture_read_lock_trivial(void)
820 {
821         preempt_disable();
822         return 0;
823 }
824
825 static void rcu_torture_read_unlock_trivial(int idx)
826 {
827         preempt_enable();
828 }
829
830 static struct rcu_torture_ops trivial_ops = {
831         .ttype          = RCU_TRIVIAL_FLAVOR,
832         .init           = rcu_sync_torture_init,
833         .readlock       = rcu_torture_read_lock_trivial,
834         .read_delay     = rcu_read_delay,  /* just reuse rcu's version. */
835         .readunlock     = rcu_torture_read_unlock_trivial,
836         .readlock_held  = torture_readlock_not_held,
837         .get_gp_seq     = rcu_no_completed,
838         .sync           = synchronize_rcu_trivial,
839         .exp_sync       = synchronize_rcu_trivial,
840         .fqs            = NULL,
841         .stats          = NULL,
842         .irq_capable    = 1,
843         .name           = "trivial"
844 };
845
846 #ifdef CONFIG_TASKS_RCU
847
848 /*
849  * Definitions for RCU-tasks torture testing.
850  */
851
852 static int tasks_torture_read_lock(void)
853 {
854         return 0;
855 }
856
857 static void tasks_torture_read_unlock(int idx)
858 {
859 }
860
861 static void rcu_tasks_torture_deferred_free(struct rcu_torture *p)
862 {
863         call_rcu_tasks(&p->rtort_rcu, rcu_torture_cb);
864 }
865
866 static void synchronize_rcu_mult_test(void)
867 {
868         synchronize_rcu_mult(call_rcu_tasks, call_rcu_hurry);
869 }
870
871 static struct rcu_torture_ops tasks_ops = {
872         .ttype          = RCU_TASKS_FLAVOR,
873         .init           = rcu_sync_torture_init,
874         .readlock       = tasks_torture_read_lock,
875         .read_delay     = rcu_read_delay,  /* just reuse rcu's version. */
876         .readunlock     = tasks_torture_read_unlock,
877         .get_gp_seq     = rcu_no_completed,
878         .deferred_free  = rcu_tasks_torture_deferred_free,
879         .sync           = synchronize_rcu_tasks,
880         .exp_sync       = synchronize_rcu_mult_test,
881         .call           = call_rcu_tasks,
882         .cb_barrier     = rcu_barrier_tasks,
883         .gp_kthread_dbg = show_rcu_tasks_classic_gp_kthread,
884         .fqs            = NULL,
885         .stats          = NULL,
886         .irq_capable    = 1,
887         .slow_gps       = 1,
888         .name           = "tasks"
889 };
890
891 #define TASKS_OPS &tasks_ops,
892
893 #else // #ifdef CONFIG_TASKS_RCU
894
895 #define TASKS_OPS
896
897 #endif // #else #ifdef CONFIG_TASKS_RCU
898
899
900 #ifdef CONFIG_TASKS_RUDE_RCU
901
902 /*
903  * Definitions for rude RCU-tasks torture testing.
904  */
905
906 static void rcu_tasks_rude_torture_deferred_free(struct rcu_torture *p)
907 {
908         call_rcu_tasks_rude(&p->rtort_rcu, rcu_torture_cb);
909 }
910
911 static struct rcu_torture_ops tasks_rude_ops = {
912         .ttype          = RCU_TASKS_RUDE_FLAVOR,
913         .init           = rcu_sync_torture_init,
914         .readlock       = rcu_torture_read_lock_trivial,
915         .read_delay     = rcu_read_delay,  /* just reuse rcu's version. */
916         .readunlock     = rcu_torture_read_unlock_trivial,
917         .get_gp_seq     = rcu_no_completed,
918         .deferred_free  = rcu_tasks_rude_torture_deferred_free,
919         .sync           = synchronize_rcu_tasks_rude,
920         .exp_sync       = synchronize_rcu_tasks_rude,
921         .call           = call_rcu_tasks_rude,
922         .cb_barrier     = rcu_barrier_tasks_rude,
923         .gp_kthread_dbg = show_rcu_tasks_rude_gp_kthread,
924         .cbflood_max    = 50000,
925         .fqs            = NULL,
926         .stats          = NULL,
927         .irq_capable    = 1,
928         .name           = "tasks-rude"
929 };
930
931 #define TASKS_RUDE_OPS &tasks_rude_ops,
932
933 #else // #ifdef CONFIG_TASKS_RUDE_RCU
934
935 #define TASKS_RUDE_OPS
936
937 #endif // #else #ifdef CONFIG_TASKS_RUDE_RCU
938
939
940 #ifdef CONFIG_TASKS_TRACE_RCU
941
942 /*
943  * Definitions for tracing RCU-tasks torture testing.
944  */
945
946 static int tasks_tracing_torture_read_lock(void)
947 {
948         rcu_read_lock_trace();
949         return 0;
950 }
951
952 static void tasks_tracing_torture_read_unlock(int idx)
953 {
954         rcu_read_unlock_trace();
955 }
956
957 static void rcu_tasks_tracing_torture_deferred_free(struct rcu_torture *p)
958 {
959         call_rcu_tasks_trace(&p->rtort_rcu, rcu_torture_cb);
960 }
961
962 static struct rcu_torture_ops tasks_tracing_ops = {
963         .ttype          = RCU_TASKS_TRACING_FLAVOR,
964         .init           = rcu_sync_torture_init,
965         .readlock       = tasks_tracing_torture_read_lock,
966         .read_delay     = srcu_read_delay,  /* just reuse srcu's version. */
967         .readunlock     = tasks_tracing_torture_read_unlock,
968         .readlock_held  = rcu_read_lock_trace_held,
969         .get_gp_seq     = rcu_no_completed,
970         .deferred_free  = rcu_tasks_tracing_torture_deferred_free,
971         .sync           = synchronize_rcu_tasks_trace,
972         .exp_sync       = synchronize_rcu_tasks_trace,
973         .call           = call_rcu_tasks_trace,
974         .cb_barrier     = rcu_barrier_tasks_trace,
975         .gp_kthread_dbg = show_rcu_tasks_trace_gp_kthread,
976         .cbflood_max    = 50000,
977         .fqs            = NULL,
978         .stats          = NULL,
979         .irq_capable    = 1,
980         .slow_gps       = 1,
981         .name           = "tasks-tracing"
982 };
983
984 #define TASKS_TRACING_OPS &tasks_tracing_ops,
985
986 #else // #ifdef CONFIG_TASKS_TRACE_RCU
987
988 #define TASKS_TRACING_OPS
989
990 #endif // #else #ifdef CONFIG_TASKS_TRACE_RCU
991
992
993 static unsigned long rcutorture_seq_diff(unsigned long new, unsigned long old)
994 {
995         if (!cur_ops->gp_diff)
996                 return new - old;
997         return cur_ops->gp_diff(new, old);
998 }
999
1000 /*
1001  * RCU torture priority-boost testing.  Runs one real-time thread per
1002  * CPU for moderate bursts, repeatedly starting grace periods and waiting
1003  * for them to complete.  If a given grace period takes too long, we assume
1004  * that priority inversion has occurred.
1005  */
1006
1007 static int old_rt_runtime = -1;
1008
1009 static void rcu_torture_disable_rt_throttle(void)
1010 {
1011         /*
1012          * Disable RT throttling so that rcutorture's boost threads don't get
1013          * throttled. Only possible if rcutorture is built-in otherwise the
1014          * user should manually do this by setting the sched_rt_period_us and
1015          * sched_rt_runtime sysctls.
1016          */
1017         if (!IS_BUILTIN(CONFIG_RCU_TORTURE_TEST) || old_rt_runtime != -1)
1018                 return;
1019
1020         old_rt_runtime = sysctl_sched_rt_runtime;
1021         sysctl_sched_rt_runtime = -1;
1022 }
1023
1024 static void rcu_torture_enable_rt_throttle(void)
1025 {
1026         if (!IS_BUILTIN(CONFIG_RCU_TORTURE_TEST) || old_rt_runtime == -1)
1027                 return;
1028
1029         sysctl_sched_rt_runtime = old_rt_runtime;
1030         old_rt_runtime = -1;
1031 }
1032
1033 static bool rcu_torture_boost_failed(unsigned long gp_state, unsigned long *start)
1034 {
1035         int cpu;
1036         static int dbg_done;
1037         unsigned long end = jiffies;
1038         bool gp_done;
1039         unsigned long j;
1040         static unsigned long last_persist;
1041         unsigned long lp;
1042         unsigned long mininterval = test_boost_duration * HZ - HZ / 2;
1043
1044         if (end - *start > mininterval) {
1045                 // Recheck after checking time to avoid false positives.
1046                 smp_mb(); // Time check before grace-period check.
1047                 if (cur_ops->poll_gp_state(gp_state))
1048                         return false; // passed, though perhaps just barely
1049                 if (cur_ops->check_boost_failed && !cur_ops->check_boost_failed(gp_state, &cpu)) {
1050                         // At most one persisted message per boost test.
1051                         j = jiffies;
1052                         lp = READ_ONCE(last_persist);
1053                         if (time_after(j, lp + mininterval) && cmpxchg(&last_persist, lp, j) == lp)
1054                                 pr_info("Boost inversion persisted: No QS from CPU %d\n", cpu);
1055                         return false; // passed on a technicality
1056                 }
1057                 VERBOSE_TOROUT_STRING("rcu_torture_boost boosting failed");
1058                 n_rcu_torture_boost_failure++;
1059                 if (!xchg(&dbg_done, 1) && cur_ops->gp_kthread_dbg) {
1060                         pr_info("Boost inversion thread ->rt_priority %u gp_state %lu jiffies %lu\n",
1061                                 current->rt_priority, gp_state, end - *start);
1062                         cur_ops->gp_kthread_dbg();
1063                         // Recheck after print to flag grace period ending during splat.
1064                         gp_done = cur_ops->poll_gp_state(gp_state);
1065                         pr_info("Boost inversion: GP %lu %s.\n", gp_state,
1066                                 gp_done ? "ended already" : "still pending");
1067
1068                 }
1069
1070                 return true; // failed
1071         } else if (cur_ops->check_boost_failed && !cur_ops->check_boost_failed(gp_state, NULL)) {
1072                 *start = jiffies;
1073         }
1074
1075         return false; // passed
1076 }
1077
1078 static int rcu_torture_boost(void *arg)
1079 {
1080         unsigned long endtime;
1081         unsigned long gp_state;
1082         unsigned long gp_state_time;
1083         unsigned long oldstarttime;
1084
1085         VERBOSE_TOROUT_STRING("rcu_torture_boost started");
1086
1087         /* Set real-time priority. */
1088         sched_set_fifo_low(current);
1089
1090         /* Each pass through the following loop does one boost-test cycle. */
1091         do {
1092                 bool failed = false; // Test failed already in this test interval
1093                 bool gp_initiated = false;
1094
1095                 if (kthread_should_stop())
1096                         goto checkwait;
1097
1098                 /* Wait for the next test interval. */
1099                 oldstarttime = READ_ONCE(boost_starttime);
1100                 while (time_before(jiffies, oldstarttime)) {
1101                         schedule_timeout_interruptible(oldstarttime - jiffies);
1102                         if (stutter_wait("rcu_torture_boost"))
1103                                 sched_set_fifo_low(current);
1104                         if (torture_must_stop())
1105                                 goto checkwait;
1106                 }
1107
1108                 // Do one boost-test interval.
1109                 endtime = oldstarttime + test_boost_duration * HZ;
1110                 while (time_before(jiffies, endtime)) {
1111                         // Has current GP gone too long?
1112                         if (gp_initiated && !failed && !cur_ops->poll_gp_state(gp_state))
1113                                 failed = rcu_torture_boost_failed(gp_state, &gp_state_time);
1114                         // If we don't have a grace period in flight, start one.
1115                         if (!gp_initiated || cur_ops->poll_gp_state(gp_state)) {
1116                                 gp_state = cur_ops->start_gp_poll();
1117                                 gp_initiated = true;
1118                                 gp_state_time = jiffies;
1119                         }
1120                         if (stutter_wait("rcu_torture_boost")) {
1121                                 sched_set_fifo_low(current);
1122                                 // If the grace period already ended,
1123                                 // we don't know when that happened, so
1124                                 // start over.
1125                                 if (cur_ops->poll_gp_state(gp_state))
1126                                         gp_initiated = false;
1127                         }
1128                         if (torture_must_stop())
1129                                 goto checkwait;
1130                 }
1131
1132                 // In case the grace period extended beyond the end of the loop.
1133                 if (gp_initiated && !failed && !cur_ops->poll_gp_state(gp_state))
1134                         rcu_torture_boost_failed(gp_state, &gp_state_time);
1135
1136                 /*
1137                  * Set the start time of the next test interval.
1138                  * Yes, this is vulnerable to long delays, but such
1139                  * delays simply cause a false negative for the next
1140                  * interval.  Besides, we are running at RT priority,
1141                  * so delays should be relatively rare.
1142                  */
1143                 while (oldstarttime == READ_ONCE(boost_starttime) && !kthread_should_stop()) {
1144                         if (mutex_trylock(&boost_mutex)) {
1145                                 if (oldstarttime == boost_starttime) {
1146                                         WRITE_ONCE(boost_starttime,
1147                                                    jiffies + test_boost_interval * HZ);
1148                                         n_rcu_torture_boosts++;
1149                                 }
1150                                 mutex_unlock(&boost_mutex);
1151                                 break;
1152                         }
1153                         schedule_timeout_uninterruptible(HZ / 20);
1154                 }
1155
1156                 /* Go do the stutter. */
1157 checkwait:      if (stutter_wait("rcu_torture_boost"))
1158                         sched_set_fifo_low(current);
1159         } while (!torture_must_stop());
1160
1161         /* Clean up and exit. */
1162         while (!kthread_should_stop()) {
1163                 torture_shutdown_absorb("rcu_torture_boost");
1164                 schedule_timeout_uninterruptible(HZ / 20);
1165         }
1166         torture_kthread_stopping("rcu_torture_boost");
1167         return 0;
1168 }
1169
1170 /*
1171  * RCU torture force-quiescent-state kthread.  Repeatedly induces
1172  * bursts of calls to force_quiescent_state(), increasing the probability
1173  * of occurrence of some important types of race conditions.
1174  */
1175 static int
1176 rcu_torture_fqs(void *arg)
1177 {
1178         unsigned long fqs_resume_time;
1179         int fqs_burst_remaining;
1180         int oldnice = task_nice(current);
1181
1182         VERBOSE_TOROUT_STRING("rcu_torture_fqs task started");
1183         do {
1184                 fqs_resume_time = jiffies + fqs_stutter * HZ;
1185                 while (time_before(jiffies, fqs_resume_time) &&
1186                        !kthread_should_stop()) {
1187                         schedule_timeout_interruptible(HZ / 20);
1188                 }
1189                 fqs_burst_remaining = fqs_duration;
1190                 while (fqs_burst_remaining > 0 &&
1191                        !kthread_should_stop()) {
1192                         cur_ops->fqs();
1193                         udelay(fqs_holdoff);
1194                         fqs_burst_remaining -= fqs_holdoff;
1195                 }
1196                 if (stutter_wait("rcu_torture_fqs"))
1197                         sched_set_normal(current, oldnice);
1198         } while (!torture_must_stop());
1199         torture_kthread_stopping("rcu_torture_fqs");
1200         return 0;
1201 }
1202
1203 // Used by writers to randomly choose from the available grace-period primitives.
1204 static int synctype[ARRAY_SIZE(rcu_torture_writer_state_names)] = { };
1205 static int nsynctypes;
1206
1207 /*
1208  * Determine which grace-period primitives are available.
1209  */
1210 static void rcu_torture_write_types(void)
1211 {
1212         bool gp_cond1 = gp_cond, gp_cond_exp1 = gp_cond_exp, gp_cond_full1 = gp_cond_full;
1213         bool gp_cond_exp_full1 = gp_cond_exp_full, gp_exp1 = gp_exp, gp_poll_exp1 = gp_poll_exp;
1214         bool gp_poll_exp_full1 = gp_poll_exp_full, gp_normal1 = gp_normal, gp_poll1 = gp_poll;
1215         bool gp_poll_full1 = gp_poll_full, gp_sync1 = gp_sync;
1216
1217         /* Initialize synctype[] array.  If none set, take default. */
1218         if (!gp_cond1 &&
1219             !gp_cond_exp1 &&
1220             !gp_cond_full1 &&
1221             !gp_cond_exp_full1 &&
1222             !gp_exp1 &&
1223             !gp_poll_exp1 &&
1224             !gp_poll_exp_full1 &&
1225             !gp_normal1 &&
1226             !gp_poll1 &&
1227             !gp_poll_full1 &&
1228             !gp_sync1) {
1229                 gp_cond1 = true;
1230                 gp_cond_exp1 = true;
1231                 gp_cond_full1 = true;
1232                 gp_cond_exp_full1 = true;
1233                 gp_exp1 = true;
1234                 gp_poll_exp1 = true;
1235                 gp_poll_exp_full1 = true;
1236                 gp_normal1 = true;
1237                 gp_poll1 = true;
1238                 gp_poll_full1 = true;
1239                 gp_sync1 = true;
1240         }
1241         if (gp_cond1 && cur_ops->get_gp_state && cur_ops->cond_sync) {
1242                 synctype[nsynctypes++] = RTWS_COND_GET;
1243                 pr_info("%s: Testing conditional GPs.\n", __func__);
1244         } else if (gp_cond && (!cur_ops->get_gp_state || !cur_ops->cond_sync)) {
1245                 pr_alert("%s: gp_cond without primitives.\n", __func__);
1246         }
1247         if (gp_cond_exp1 && cur_ops->get_gp_state_exp && cur_ops->cond_sync_exp) {
1248                 synctype[nsynctypes++] = RTWS_COND_GET_EXP;
1249                 pr_info("%s: Testing conditional expedited GPs.\n", __func__);
1250         } else if (gp_cond_exp && (!cur_ops->get_gp_state_exp || !cur_ops->cond_sync_exp)) {
1251                 pr_alert("%s: gp_cond_exp without primitives.\n", __func__);
1252         }
1253         if (gp_cond_full1 && cur_ops->get_gp_state && cur_ops->cond_sync_full) {
1254                 synctype[nsynctypes++] = RTWS_COND_GET_FULL;
1255                 pr_info("%s: Testing conditional full-state GPs.\n", __func__);
1256         } else if (gp_cond_full && (!cur_ops->get_gp_state || !cur_ops->cond_sync_full)) {
1257                 pr_alert("%s: gp_cond_full without primitives.\n", __func__);
1258         }
1259         if (gp_cond_exp_full1 && cur_ops->get_gp_state_exp && cur_ops->cond_sync_exp_full) {
1260                 synctype[nsynctypes++] = RTWS_COND_GET_EXP_FULL;
1261                 pr_info("%s: Testing conditional full-state expedited GPs.\n", __func__);
1262         } else if (gp_cond_exp_full &&
1263                    (!cur_ops->get_gp_state_exp || !cur_ops->cond_sync_exp_full)) {
1264                 pr_alert("%s: gp_cond_exp_full without primitives.\n", __func__);
1265         }
1266         if (gp_exp1 && cur_ops->exp_sync) {
1267                 synctype[nsynctypes++] = RTWS_EXP_SYNC;
1268                 pr_info("%s: Testing expedited GPs.\n", __func__);
1269         } else if (gp_exp && !cur_ops->exp_sync) {
1270                 pr_alert("%s: gp_exp without primitives.\n", __func__);
1271         }
1272         if (gp_normal1 && cur_ops->deferred_free) {
1273                 synctype[nsynctypes++] = RTWS_DEF_FREE;
1274                 pr_info("%s: Testing asynchronous GPs.\n", __func__);
1275         } else if (gp_normal && !cur_ops->deferred_free) {
1276                 pr_alert("%s: gp_normal without primitives.\n", __func__);
1277         }
1278         if (gp_poll1 && cur_ops->get_comp_state && cur_ops->same_gp_state &&
1279             cur_ops->start_gp_poll && cur_ops->poll_gp_state) {
1280                 synctype[nsynctypes++] = RTWS_POLL_GET;
1281                 pr_info("%s: Testing polling GPs.\n", __func__);
1282         } else if (gp_poll && (!cur_ops->start_gp_poll || !cur_ops->poll_gp_state)) {
1283                 pr_alert("%s: gp_poll without primitives.\n", __func__);
1284         }
1285         if (gp_poll_full1 && cur_ops->get_comp_state_full && cur_ops->same_gp_state_full
1286             && cur_ops->start_gp_poll_full && cur_ops->poll_gp_state_full) {
1287                 synctype[nsynctypes++] = RTWS_POLL_GET_FULL;
1288                 pr_info("%s: Testing polling full-state GPs.\n", __func__);
1289         } else if (gp_poll_full && (!cur_ops->start_gp_poll_full || !cur_ops->poll_gp_state_full)) {
1290                 pr_alert("%s: gp_poll_full without primitives.\n", __func__);
1291         }
1292         if (gp_poll_exp1 && cur_ops->start_gp_poll_exp && cur_ops->poll_gp_state_exp) {
1293                 synctype[nsynctypes++] = RTWS_POLL_GET_EXP;
1294                 pr_info("%s: Testing polling expedited GPs.\n", __func__);
1295         } else if (gp_poll_exp && (!cur_ops->start_gp_poll_exp || !cur_ops->poll_gp_state_exp)) {
1296                 pr_alert("%s: gp_poll_exp without primitives.\n", __func__);
1297         }
1298         if (gp_poll_exp_full1 && cur_ops->start_gp_poll_exp_full && cur_ops->poll_gp_state_full) {
1299                 synctype[nsynctypes++] = RTWS_POLL_GET_EXP_FULL;
1300                 pr_info("%s: Testing polling full-state expedited GPs.\n", __func__);
1301         } else if (gp_poll_exp_full &&
1302                    (!cur_ops->start_gp_poll_exp_full || !cur_ops->poll_gp_state_full)) {
1303                 pr_alert("%s: gp_poll_exp_full without primitives.\n", __func__);
1304         }
1305         if (gp_sync1 && cur_ops->sync) {
1306                 synctype[nsynctypes++] = RTWS_SYNC;
1307                 pr_info("%s: Testing normal GPs.\n", __func__);
1308         } else if (gp_sync && !cur_ops->sync) {
1309                 pr_alert("%s: gp_sync without primitives.\n", __func__);
1310         }
1311 }
1312
1313 /*
1314  * Do the specified rcu_torture_writer() synchronous grace period,
1315  * while also testing out the polled APIs.  Note well that the single-CPU
1316  * grace-period optimizations must be accounted for.
1317  */
1318 static void do_rtws_sync(struct torture_random_state *trsp, void (*sync)(void))
1319 {
1320         unsigned long cookie;
1321         struct rcu_gp_oldstate cookie_full;
1322         bool dopoll;
1323         bool dopoll_full;
1324         unsigned long r = torture_random(trsp);
1325
1326         dopoll = cur_ops->get_gp_state && cur_ops->poll_gp_state && !(r & 0x300);
1327         dopoll_full = cur_ops->get_gp_state_full && cur_ops->poll_gp_state_full && !(r & 0xc00);
1328         if (dopoll || dopoll_full)
1329                 cpus_read_lock();
1330         if (dopoll)
1331                 cookie = cur_ops->get_gp_state();
1332         if (dopoll_full)
1333                 cur_ops->get_gp_state_full(&cookie_full);
1334         if (cur_ops->poll_need_2gp && cur_ops->poll_need_2gp(dopoll, dopoll_full))
1335                 sync();
1336         sync();
1337         WARN_ONCE(dopoll && !cur_ops->poll_gp_state(cookie),
1338                   "%s: Cookie check 3 failed %pS() online %*pbl.",
1339                   __func__, sync, cpumask_pr_args(cpu_online_mask));
1340         WARN_ONCE(dopoll_full && !cur_ops->poll_gp_state_full(&cookie_full),
1341                   "%s: Cookie check 4 failed %pS() online %*pbl",
1342                   __func__, sync, cpumask_pr_args(cpu_online_mask));
1343         if (dopoll || dopoll_full)
1344                 cpus_read_unlock();
1345 }
1346
1347 /*
1348  * RCU torture writer kthread.  Repeatedly substitutes a new structure
1349  * for that pointed to by rcu_torture_current, freeing the old structure
1350  * after a series of grace periods (the "pipeline").
1351  */
1352 static int
1353 rcu_torture_writer(void *arg)
1354 {
1355         bool boot_ended;
1356         bool can_expedite = !rcu_gp_is_expedited() && !rcu_gp_is_normal();
1357         unsigned long cookie;
1358         struct rcu_gp_oldstate cookie_full;
1359         int expediting = 0;
1360         unsigned long gp_snap;
1361         unsigned long gp_snap1;
1362         struct rcu_gp_oldstate gp_snap_full;
1363         struct rcu_gp_oldstate gp_snap1_full;
1364         int i;
1365         int idx;
1366         int oldnice = task_nice(current);
1367         struct rcu_gp_oldstate rgo[NUM_ACTIVE_RCU_POLL_FULL_OLDSTATE];
1368         struct rcu_torture *rp;
1369         struct rcu_torture *old_rp;
1370         static DEFINE_TORTURE_RANDOM(rand);
1371         bool stutter_waited;
1372         unsigned long ulo[NUM_ACTIVE_RCU_POLL_OLDSTATE];
1373
1374         VERBOSE_TOROUT_STRING("rcu_torture_writer task started");
1375         if (!can_expedite)
1376                 pr_alert("%s" TORTURE_FLAG
1377                          " GP expediting controlled from boot/sysfs for %s.\n",
1378                          torture_type, cur_ops->name);
1379         if (WARN_ONCE(nsynctypes == 0,
1380                       "%s: No update-side primitives.\n", __func__)) {
1381                 /*
1382                  * No updates primitives, so don't try updating.
1383                  * The resulting test won't be testing much, hence the
1384                  * above WARN_ONCE().
1385                  */
1386                 rcu_torture_writer_state = RTWS_STOPPING;
1387                 torture_kthread_stopping("rcu_torture_writer");
1388                 return 0;
1389         }
1390
1391         do {
1392                 rcu_torture_writer_state = RTWS_FIXED_DELAY;
1393                 torture_hrtimeout_us(500, 1000, &rand);
1394                 rp = rcu_torture_alloc();
1395                 if (rp == NULL)
1396                         continue;
1397                 rp->rtort_pipe_count = 0;
1398                 rcu_torture_writer_state = RTWS_DELAY;
1399                 udelay(torture_random(&rand) & 0x3ff);
1400                 rcu_torture_writer_state = RTWS_REPLACE;
1401                 old_rp = rcu_dereference_check(rcu_torture_current,
1402                                                current == writer_task);
1403                 rp->rtort_mbtest = 1;
1404                 rcu_assign_pointer(rcu_torture_current, rp);
1405                 smp_wmb(); /* Mods to old_rp must follow rcu_assign_pointer() */
1406                 if (old_rp) {
1407                         i = old_rp->rtort_pipe_count;
1408                         if (i > RCU_TORTURE_PIPE_LEN)
1409                                 i = RCU_TORTURE_PIPE_LEN;
1410                         atomic_inc(&rcu_torture_wcount[i]);
1411                         WRITE_ONCE(old_rp->rtort_pipe_count,
1412                                    old_rp->rtort_pipe_count + 1);
1413
1414                         // Make sure readers block polled grace periods.
1415                         if (cur_ops->get_gp_state && cur_ops->poll_gp_state) {
1416                                 idx = cur_ops->readlock();
1417                                 cookie = cur_ops->get_gp_state();
1418                                 WARN_ONCE(cur_ops->poll_gp_state(cookie),
1419                                           "%s: Cookie check 1 failed %s(%d) %lu->%lu\n",
1420                                           __func__,
1421                                           rcu_torture_writer_state_getname(),
1422                                           rcu_torture_writer_state,
1423                                           cookie, cur_ops->get_gp_state());
1424                                 if (cur_ops->get_gp_completed) {
1425                                         cookie = cur_ops->get_gp_completed();
1426                                         WARN_ON_ONCE(!cur_ops->poll_gp_state(cookie));
1427                                 }
1428                                 cur_ops->readunlock(idx);
1429                         }
1430                         if (cur_ops->get_gp_state_full && cur_ops->poll_gp_state_full) {
1431                                 idx = cur_ops->readlock();
1432                                 cur_ops->get_gp_state_full(&cookie_full);
1433                                 WARN_ONCE(cur_ops->poll_gp_state_full(&cookie_full),
1434                                           "%s: Cookie check 5 failed %s(%d) online %*pbl\n",
1435                                           __func__,
1436                                           rcu_torture_writer_state_getname(),
1437                                           rcu_torture_writer_state,
1438                                           cpumask_pr_args(cpu_online_mask));
1439                                 if (cur_ops->get_gp_completed_full) {
1440                                         cur_ops->get_gp_completed_full(&cookie_full);
1441                                         WARN_ON_ONCE(!cur_ops->poll_gp_state_full(&cookie_full));
1442                                 }
1443                                 cur_ops->readunlock(idx);
1444                         }
1445                         switch (synctype[torture_random(&rand) % nsynctypes]) {
1446                         case RTWS_DEF_FREE:
1447                                 rcu_torture_writer_state = RTWS_DEF_FREE;
1448                                 cur_ops->deferred_free(old_rp);
1449                                 break;
1450                         case RTWS_EXP_SYNC:
1451                                 rcu_torture_writer_state = RTWS_EXP_SYNC;
1452                                 do_rtws_sync(&rand, cur_ops->exp_sync);
1453                                 rcu_torture_pipe_update(old_rp);
1454                                 break;
1455                         case RTWS_COND_GET:
1456                                 rcu_torture_writer_state = RTWS_COND_GET;
1457                                 gp_snap = cur_ops->get_gp_state();
1458                                 torture_hrtimeout_jiffies(torture_random(&rand) % 16, &rand);
1459                                 rcu_torture_writer_state = RTWS_COND_SYNC;
1460                                 cur_ops->cond_sync(gp_snap);
1461                                 rcu_torture_pipe_update(old_rp);
1462                                 break;
1463                         case RTWS_COND_GET_EXP:
1464                                 rcu_torture_writer_state = RTWS_COND_GET_EXP;
1465                                 gp_snap = cur_ops->get_gp_state_exp();
1466                                 torture_hrtimeout_jiffies(torture_random(&rand) % 16, &rand);
1467                                 rcu_torture_writer_state = RTWS_COND_SYNC_EXP;
1468                                 cur_ops->cond_sync_exp(gp_snap);
1469                                 rcu_torture_pipe_update(old_rp);
1470                                 break;
1471                         case RTWS_COND_GET_FULL:
1472                                 rcu_torture_writer_state = RTWS_COND_GET_FULL;
1473                                 cur_ops->get_gp_state_full(&gp_snap_full);
1474                                 torture_hrtimeout_jiffies(torture_random(&rand) % 16, &rand);
1475                                 rcu_torture_writer_state = RTWS_COND_SYNC_FULL;
1476                                 cur_ops->cond_sync_full(&gp_snap_full);
1477                                 rcu_torture_pipe_update(old_rp);
1478                                 break;
1479                         case RTWS_COND_GET_EXP_FULL:
1480                                 rcu_torture_writer_state = RTWS_COND_GET_EXP_FULL;
1481                                 cur_ops->get_gp_state_full(&gp_snap_full);
1482                                 torture_hrtimeout_jiffies(torture_random(&rand) % 16, &rand);
1483                                 rcu_torture_writer_state = RTWS_COND_SYNC_EXP_FULL;
1484                                 cur_ops->cond_sync_exp_full(&gp_snap_full);
1485                                 rcu_torture_pipe_update(old_rp);
1486                                 break;
1487                         case RTWS_POLL_GET:
1488                                 rcu_torture_writer_state = RTWS_POLL_GET;
1489                                 for (i = 0; i < ARRAY_SIZE(ulo); i++)
1490                                         ulo[i] = cur_ops->get_comp_state();
1491                                 gp_snap = cur_ops->start_gp_poll();
1492                                 rcu_torture_writer_state = RTWS_POLL_WAIT;
1493                                 while (!cur_ops->poll_gp_state(gp_snap)) {
1494                                         gp_snap1 = cur_ops->get_gp_state();
1495                                         for (i = 0; i < ARRAY_SIZE(ulo); i++)
1496                                                 if (cur_ops->poll_gp_state(ulo[i]) ||
1497                                                     cur_ops->same_gp_state(ulo[i], gp_snap1)) {
1498                                                         ulo[i] = gp_snap1;
1499                                                         break;
1500                                                 }
1501                                         WARN_ON_ONCE(i >= ARRAY_SIZE(ulo));
1502                                         torture_hrtimeout_jiffies(torture_random(&rand) % 16,
1503                                                                   &rand);
1504                                 }
1505                                 rcu_torture_pipe_update(old_rp);
1506                                 break;
1507                         case RTWS_POLL_GET_FULL:
1508                                 rcu_torture_writer_state = RTWS_POLL_GET_FULL;
1509                                 for (i = 0; i < ARRAY_SIZE(rgo); i++)
1510                                         cur_ops->get_comp_state_full(&rgo[i]);
1511                                 cur_ops->start_gp_poll_full(&gp_snap_full);
1512                                 rcu_torture_writer_state = RTWS_POLL_WAIT_FULL;
1513                                 while (!cur_ops->poll_gp_state_full(&gp_snap_full)) {
1514                                         cur_ops->get_gp_state_full(&gp_snap1_full);
1515                                         for (i = 0; i < ARRAY_SIZE(rgo); i++)
1516                                                 if (cur_ops->poll_gp_state_full(&rgo[i]) ||
1517                                                     cur_ops->same_gp_state_full(&rgo[i],
1518                                                                                 &gp_snap1_full)) {
1519                                                         rgo[i] = gp_snap1_full;
1520                                                         break;
1521                                                 }
1522                                         WARN_ON_ONCE(i >= ARRAY_SIZE(rgo));
1523                                         torture_hrtimeout_jiffies(torture_random(&rand) % 16,
1524                                                                   &rand);
1525                                 }
1526                                 rcu_torture_pipe_update(old_rp);
1527                                 break;
1528                         case RTWS_POLL_GET_EXP:
1529                                 rcu_torture_writer_state = RTWS_POLL_GET_EXP;
1530                                 gp_snap = cur_ops->start_gp_poll_exp();
1531                                 rcu_torture_writer_state = RTWS_POLL_WAIT_EXP;
1532                                 while (!cur_ops->poll_gp_state_exp(gp_snap))
1533                                         torture_hrtimeout_jiffies(torture_random(&rand) % 16,
1534                                                                   &rand);
1535                                 rcu_torture_pipe_update(old_rp);
1536                                 break;
1537                         case RTWS_POLL_GET_EXP_FULL:
1538                                 rcu_torture_writer_state = RTWS_POLL_GET_EXP_FULL;
1539                                 cur_ops->start_gp_poll_exp_full(&gp_snap_full);
1540                                 rcu_torture_writer_state = RTWS_POLL_WAIT_EXP_FULL;
1541                                 while (!cur_ops->poll_gp_state_full(&gp_snap_full))
1542                                         torture_hrtimeout_jiffies(torture_random(&rand) % 16,
1543                                                                   &rand);
1544                                 rcu_torture_pipe_update(old_rp);
1545                                 break;
1546                         case RTWS_SYNC:
1547                                 rcu_torture_writer_state = RTWS_SYNC;
1548                                 do_rtws_sync(&rand, cur_ops->sync);
1549                                 rcu_torture_pipe_update(old_rp);
1550                                 break;
1551                         default:
1552                                 WARN_ON_ONCE(1);
1553                                 break;
1554                         }
1555                 }
1556                 WRITE_ONCE(rcu_torture_current_version,
1557                            rcu_torture_current_version + 1);
1558                 /* Cycle through nesting levels of rcu_expedite_gp() calls. */
1559                 if (can_expedite &&
1560                     !(torture_random(&rand) & 0xff & (!!expediting - 1))) {
1561                         WARN_ON_ONCE(expediting == 0 && rcu_gp_is_expedited());
1562                         if (expediting >= 0)
1563                                 rcu_expedite_gp();
1564                         else
1565                                 rcu_unexpedite_gp();
1566                         if (++expediting > 3)
1567                                 expediting = -expediting;
1568                 } else if (!can_expedite) { /* Disabled during boot, recheck. */
1569                         can_expedite = !rcu_gp_is_expedited() &&
1570                                        !rcu_gp_is_normal();
1571                 }
1572                 rcu_torture_writer_state = RTWS_STUTTER;
1573                 boot_ended = rcu_inkernel_boot_has_ended();
1574                 stutter_waited = stutter_wait("rcu_torture_writer");
1575                 if (stutter_waited &&
1576                     !atomic_read(&rcu_fwd_cb_nodelay) &&
1577                     !cur_ops->slow_gps &&
1578                     !torture_must_stop() &&
1579                     boot_ended)
1580                         for (i = 0; i < ARRAY_SIZE(rcu_tortures); i++)
1581                                 if (list_empty(&rcu_tortures[i].rtort_free) &&
1582                                     rcu_access_pointer(rcu_torture_current) !=
1583                                     &rcu_tortures[i]) {
1584                                         tracing_off();
1585                                         show_rcu_gp_kthreads();
1586                                         WARN(1, "%s: rtort_pipe_count: %d\n", __func__, rcu_tortures[i].rtort_pipe_count);
1587                                         rcu_ftrace_dump(DUMP_ALL);
1588                                 }
1589                 if (stutter_waited)
1590                         sched_set_normal(current, oldnice);
1591         } while (!torture_must_stop());
1592         rcu_torture_current = NULL;  // Let stats task know that we are done.
1593         /* Reset expediting back to unexpedited. */
1594         if (expediting > 0)
1595                 expediting = -expediting;
1596         while (can_expedite && expediting++ < 0)
1597                 rcu_unexpedite_gp();
1598         WARN_ON_ONCE(can_expedite && rcu_gp_is_expedited());
1599         if (!can_expedite)
1600                 pr_alert("%s" TORTURE_FLAG
1601                          " Dynamic grace-period expediting was disabled.\n",
1602                          torture_type);
1603         rcu_torture_writer_state = RTWS_STOPPING;
1604         torture_kthread_stopping("rcu_torture_writer");
1605         return 0;
1606 }
1607
1608 /*
1609  * RCU torture fake writer kthread.  Repeatedly calls sync, with a random
1610  * delay between calls.
1611  */
1612 static int
1613 rcu_torture_fakewriter(void *arg)
1614 {
1615         unsigned long gp_snap;
1616         struct rcu_gp_oldstate gp_snap_full;
1617         DEFINE_TORTURE_RANDOM(rand);
1618
1619         VERBOSE_TOROUT_STRING("rcu_torture_fakewriter task started");
1620         set_user_nice(current, MAX_NICE);
1621
1622         if (WARN_ONCE(nsynctypes == 0,
1623                       "%s: No update-side primitives.\n", __func__)) {
1624                 /*
1625                  * No updates primitives, so don't try updating.
1626                  * The resulting test won't be testing much, hence the
1627                  * above WARN_ONCE().
1628                  */
1629                 torture_kthread_stopping("rcu_torture_fakewriter");
1630                 return 0;
1631         }
1632
1633         do {
1634                 torture_hrtimeout_jiffies(torture_random(&rand) % 10, &rand);
1635                 if (cur_ops->cb_barrier != NULL &&
1636                     torture_random(&rand) % (nfakewriters * 8) == 0) {
1637                         cur_ops->cb_barrier();
1638                 } else {
1639                         switch (synctype[torture_random(&rand) % nsynctypes]) {
1640                         case RTWS_DEF_FREE:
1641                                 break;
1642                         case RTWS_EXP_SYNC:
1643                                 cur_ops->exp_sync();
1644                                 break;
1645                         case RTWS_COND_GET:
1646                                 gp_snap = cur_ops->get_gp_state();
1647                                 torture_hrtimeout_jiffies(torture_random(&rand) % 16, &rand);
1648                                 cur_ops->cond_sync(gp_snap);
1649                                 break;
1650                         case RTWS_COND_GET_EXP:
1651                                 gp_snap = cur_ops->get_gp_state_exp();
1652                                 torture_hrtimeout_jiffies(torture_random(&rand) % 16, &rand);
1653                                 cur_ops->cond_sync_exp(gp_snap);
1654                                 break;
1655                         case RTWS_COND_GET_FULL:
1656                                 cur_ops->get_gp_state_full(&gp_snap_full);
1657                                 torture_hrtimeout_jiffies(torture_random(&rand) % 16, &rand);
1658                                 cur_ops->cond_sync_full(&gp_snap_full);
1659                                 break;
1660                         case RTWS_COND_GET_EXP_FULL:
1661                                 cur_ops->get_gp_state_full(&gp_snap_full);
1662                                 torture_hrtimeout_jiffies(torture_random(&rand) % 16, &rand);
1663                                 cur_ops->cond_sync_exp_full(&gp_snap_full);
1664                                 break;
1665                         case RTWS_POLL_GET:
1666                                 gp_snap = cur_ops->start_gp_poll();
1667                                 while (!cur_ops->poll_gp_state(gp_snap)) {
1668                                         torture_hrtimeout_jiffies(torture_random(&rand) % 16,
1669                                                                   &rand);
1670                                 }
1671                                 break;
1672                         case RTWS_POLL_GET_FULL:
1673                                 cur_ops->start_gp_poll_full(&gp_snap_full);
1674                                 while (!cur_ops->poll_gp_state_full(&gp_snap_full)) {
1675                                         torture_hrtimeout_jiffies(torture_random(&rand) % 16,
1676                                                                   &rand);
1677                                 }
1678                                 break;
1679                         case RTWS_POLL_GET_EXP:
1680                                 gp_snap = cur_ops->start_gp_poll_exp();
1681                                 while (!cur_ops->poll_gp_state_exp(gp_snap)) {
1682                                         torture_hrtimeout_jiffies(torture_random(&rand) % 16,
1683                                                                   &rand);
1684                                 }
1685                                 break;
1686                         case RTWS_POLL_GET_EXP_FULL:
1687                                 cur_ops->start_gp_poll_exp_full(&gp_snap_full);
1688                                 while (!cur_ops->poll_gp_state_full(&gp_snap_full)) {
1689                                         torture_hrtimeout_jiffies(torture_random(&rand) % 16,
1690                                                                   &rand);
1691                                 }
1692                                 break;
1693                         case RTWS_SYNC:
1694                                 cur_ops->sync();
1695                                 break;
1696                         default:
1697                                 WARN_ON_ONCE(1);
1698                                 break;
1699                         }
1700                 }
1701                 stutter_wait("rcu_torture_fakewriter");
1702         } while (!torture_must_stop());
1703
1704         torture_kthread_stopping("rcu_torture_fakewriter");
1705         return 0;
1706 }
1707
1708 static void rcu_torture_timer_cb(struct rcu_head *rhp)
1709 {
1710         kfree(rhp);
1711 }
1712
1713 // Set up and carry out testing of RCU's global memory ordering
1714 static void rcu_torture_reader_do_mbchk(long myid, struct rcu_torture *rtp,
1715                                         struct torture_random_state *trsp)
1716 {
1717         unsigned long loops;
1718         int noc = torture_num_online_cpus();
1719         int rdrchked;
1720         int rdrchker;
1721         struct rcu_torture_reader_check *rtrcp; // Me.
1722         struct rcu_torture_reader_check *rtrcp_assigner; // Assigned us to do checking.
1723         struct rcu_torture_reader_check *rtrcp_chked; // Reader being checked.
1724         struct rcu_torture_reader_check *rtrcp_chker; // Reader doing checking when not me.
1725
1726         if (myid < 0)
1727                 return; // Don't try this from timer handlers.
1728
1729         // Increment my counter.
1730         rtrcp = &rcu_torture_reader_mbchk[myid];
1731         WRITE_ONCE(rtrcp->rtc_myloops, rtrcp->rtc_myloops + 1);
1732
1733         // Attempt to assign someone else some checking work.
1734         rdrchked = torture_random(trsp) % nrealreaders;
1735         rtrcp_chked = &rcu_torture_reader_mbchk[rdrchked];
1736         rdrchker = torture_random(trsp) % nrealreaders;
1737         rtrcp_chker = &rcu_torture_reader_mbchk[rdrchker];
1738         if (rdrchked != myid && rdrchked != rdrchker && noc >= rdrchked && noc >= rdrchker &&
1739             smp_load_acquire(&rtrcp->rtc_chkrdr) < 0 && // Pairs with smp_store_release below.
1740             !READ_ONCE(rtp->rtort_chkp) &&
1741             !smp_load_acquire(&rtrcp_chker->rtc_assigner)) { // Pairs with smp_store_release below.
1742                 rtrcp->rtc_chkloops = READ_ONCE(rtrcp_chked->rtc_myloops);
1743                 WARN_ON_ONCE(rtrcp->rtc_chkrdr >= 0);
1744                 rtrcp->rtc_chkrdr = rdrchked;
1745                 WARN_ON_ONCE(rtrcp->rtc_ready); // This gets set after the grace period ends.
1746                 if (cmpxchg_relaxed(&rtrcp_chker->rtc_assigner, NULL, rtrcp) ||
1747                     cmpxchg_relaxed(&rtp->rtort_chkp, NULL, rtrcp))
1748                         (void)cmpxchg_relaxed(&rtrcp_chker->rtc_assigner, rtrcp, NULL); // Back out.
1749         }
1750
1751         // If assigned some completed work, do it!
1752         rtrcp_assigner = READ_ONCE(rtrcp->rtc_assigner);
1753         if (!rtrcp_assigner || !smp_load_acquire(&rtrcp_assigner->rtc_ready))
1754                 return; // No work or work not yet ready.
1755         rdrchked = rtrcp_assigner->rtc_chkrdr;
1756         if (WARN_ON_ONCE(rdrchked < 0))
1757                 return;
1758         rtrcp_chked = &rcu_torture_reader_mbchk[rdrchked];
1759         loops = READ_ONCE(rtrcp_chked->rtc_myloops);
1760         atomic_inc(&n_rcu_torture_mbchk_tries);
1761         if (ULONG_CMP_LT(loops, rtrcp_assigner->rtc_chkloops))
1762                 atomic_inc(&n_rcu_torture_mbchk_fail);
1763         rtrcp_assigner->rtc_chkloops = loops + ULONG_MAX / 2;
1764         rtrcp_assigner->rtc_ready = 0;
1765         smp_store_release(&rtrcp->rtc_assigner, NULL); // Someone else can assign us work.
1766         smp_store_release(&rtrcp_assigner->rtc_chkrdr, -1); // Assigner can again assign.
1767 }
1768
1769 /*
1770  * Do one extension of an RCU read-side critical section using the
1771  * current reader state in readstate (set to zero for initial entry
1772  * to extended critical section), set the new state as specified by
1773  * newstate (set to zero for final exit from extended critical section),
1774  * and random-number-generator state in trsp.  If this is neither the
1775  * beginning or end of the critical section and if there was actually a
1776  * change, do a ->read_delay().
1777  */
1778 static void rcutorture_one_extend(int *readstate, int newstate,
1779                                   struct torture_random_state *trsp,
1780                                   struct rt_read_seg *rtrsp)
1781 {
1782         unsigned long flags;
1783         int idxnew1 = -1;
1784         int idxnew2 = -1;
1785         int idxold1 = *readstate;
1786         int idxold2 = idxold1;
1787         int statesnew = ~*readstate & newstate;
1788         int statesold = *readstate & ~newstate;
1789
1790         WARN_ON_ONCE(idxold2 < 0);
1791         WARN_ON_ONCE((idxold2 >> RCUTORTURE_RDR_SHIFT_2) > 1);
1792         rtrsp->rt_readstate = newstate;
1793
1794         /* First, put new protection in place to avoid critical-section gap. */
1795         if (statesnew & RCUTORTURE_RDR_BH)
1796                 local_bh_disable();
1797         if (statesnew & RCUTORTURE_RDR_RBH)
1798                 rcu_read_lock_bh();
1799         if (statesnew & RCUTORTURE_RDR_IRQ)
1800                 local_irq_disable();
1801         if (statesnew & RCUTORTURE_RDR_PREEMPT)
1802                 preempt_disable();
1803         if (statesnew & RCUTORTURE_RDR_SCHED)
1804                 rcu_read_lock_sched();
1805         if (statesnew & RCUTORTURE_RDR_RCU_1)
1806                 idxnew1 = (cur_ops->readlock() & 0x1) << RCUTORTURE_RDR_SHIFT_1;
1807         if (statesnew & RCUTORTURE_RDR_RCU_2)
1808                 idxnew2 = (cur_ops->readlock() & 0x1) << RCUTORTURE_RDR_SHIFT_2;
1809
1810         /*
1811          * Next, remove old protection, in decreasing order of strength
1812          * to avoid unlock paths that aren't safe in the stronger
1813          * context. Namely: BH can not be enabled with disabled interrupts.
1814          * Additionally PREEMPT_RT requires that BH is enabled in preemptible
1815          * context.
1816          */
1817         if (statesold & RCUTORTURE_RDR_IRQ)
1818                 local_irq_enable();
1819         if (statesold & RCUTORTURE_RDR_PREEMPT)
1820                 preempt_enable();
1821         if (statesold & RCUTORTURE_RDR_SCHED)
1822                 rcu_read_unlock_sched();
1823         if (statesold & RCUTORTURE_RDR_BH)
1824                 local_bh_enable();
1825         if (statesold & RCUTORTURE_RDR_RBH)
1826                 rcu_read_unlock_bh();
1827         if (statesold & RCUTORTURE_RDR_RCU_2) {
1828                 cur_ops->readunlock((idxold2 >> RCUTORTURE_RDR_SHIFT_2) & 0x1);
1829                 WARN_ON_ONCE(idxnew2 != -1);
1830                 idxold2 = 0;
1831         }
1832         if (statesold & RCUTORTURE_RDR_RCU_1) {
1833                 bool lockit;
1834
1835                 lockit = !cur_ops->no_pi_lock && !statesnew && !(torture_random(trsp) & 0xffff);
1836                 if (lockit)
1837                         raw_spin_lock_irqsave(&current->pi_lock, flags);
1838                 cur_ops->readunlock((idxold1 >> RCUTORTURE_RDR_SHIFT_1) & 0x1);
1839                 WARN_ON_ONCE(idxnew1 != -1);
1840                 idxold1 = 0;
1841                 if (lockit)
1842                         raw_spin_unlock_irqrestore(&current->pi_lock, flags);
1843         }
1844
1845         /* Delay if neither beginning nor end and there was a change. */
1846         if ((statesnew || statesold) && *readstate && newstate)
1847                 cur_ops->read_delay(trsp, rtrsp);
1848
1849         /* Update the reader state. */
1850         if (idxnew1 == -1)
1851                 idxnew1 = idxold1 & RCUTORTURE_RDR_MASK_1;
1852         WARN_ON_ONCE(idxnew1 < 0);
1853         if (WARN_ON_ONCE((idxnew1 >> RCUTORTURE_RDR_SHIFT_1) > 1))
1854                 pr_info("Unexpected idxnew1 value of %#x\n", idxnew1);
1855         if (idxnew2 == -1)
1856                 idxnew2 = idxold2 & RCUTORTURE_RDR_MASK_2;
1857         WARN_ON_ONCE(idxnew2 < 0);
1858         WARN_ON_ONCE((idxnew2 >> RCUTORTURE_RDR_SHIFT_2) > 1);
1859         *readstate = idxnew1 | idxnew2 | newstate;
1860         WARN_ON_ONCE(*readstate < 0);
1861         if (WARN_ON_ONCE((*readstate >> RCUTORTURE_RDR_SHIFT_2) > 1))
1862                 pr_info("Unexpected idxnew2 value of %#x\n", idxnew2);
1863 }
1864
1865 /* Return the biggest extendables mask given current RCU and boot parameters. */
1866 static int rcutorture_extend_mask_max(void)
1867 {
1868         int mask;
1869
1870         WARN_ON_ONCE(extendables & ~RCUTORTURE_MAX_EXTEND);
1871         mask = extendables & RCUTORTURE_MAX_EXTEND & cur_ops->extendables;
1872         mask = mask | RCUTORTURE_RDR_RCU_1 | RCUTORTURE_RDR_RCU_2;
1873         return mask;
1874 }
1875
1876 /* Return a random protection state mask, but with at least one bit set. */
1877 static int
1878 rcutorture_extend_mask(int oldmask, struct torture_random_state *trsp)
1879 {
1880         int mask = rcutorture_extend_mask_max();
1881         unsigned long randmask1 = torture_random(trsp);
1882         unsigned long randmask2 = randmask1 >> 3;
1883         unsigned long preempts = RCUTORTURE_RDR_PREEMPT | RCUTORTURE_RDR_SCHED;
1884         unsigned long preempts_irq = preempts | RCUTORTURE_RDR_IRQ;
1885         unsigned long bhs = RCUTORTURE_RDR_BH | RCUTORTURE_RDR_RBH;
1886
1887         WARN_ON_ONCE(mask >> RCUTORTURE_RDR_SHIFT_1);
1888         /* Mostly only one bit (need preemption!), sometimes lots of bits. */
1889         if (!(randmask1 & 0x7))
1890                 mask = mask & randmask2;
1891         else
1892                 mask = mask & (1 << (randmask2 % RCUTORTURE_RDR_NBITS));
1893
1894         // Can't have nested RCU reader without outer RCU reader.
1895         if (!(mask & RCUTORTURE_RDR_RCU_1) && (mask & RCUTORTURE_RDR_RCU_2)) {
1896                 if (oldmask & RCUTORTURE_RDR_RCU_1)
1897                         mask &= ~RCUTORTURE_RDR_RCU_2;
1898                 else
1899                         mask |= RCUTORTURE_RDR_RCU_1;
1900         }
1901
1902         /*
1903          * Can't enable bh w/irq disabled.
1904          */
1905         if (mask & RCUTORTURE_RDR_IRQ)
1906                 mask |= oldmask & bhs;
1907
1908         /*
1909          * Ideally these sequences would be detected in debug builds
1910          * (regardless of RT), but until then don't stop testing
1911          * them on non-RT.
1912          */
1913         if (IS_ENABLED(CONFIG_PREEMPT_RT)) {
1914                 /* Can't modify BH in atomic context */
1915                 if (oldmask & preempts_irq)
1916                         mask &= ~bhs;
1917                 if ((oldmask | mask) & preempts_irq)
1918                         mask |= oldmask & bhs;
1919         }
1920
1921         return mask ?: RCUTORTURE_RDR_RCU_1;
1922 }
1923
1924 /*
1925  * Do a randomly selected number of extensions of an existing RCU read-side
1926  * critical section.
1927  */
1928 static struct rt_read_seg *
1929 rcutorture_loop_extend(int *readstate, struct torture_random_state *trsp,
1930                        struct rt_read_seg *rtrsp)
1931 {
1932         int i;
1933         int j;
1934         int mask = rcutorture_extend_mask_max();
1935
1936         WARN_ON_ONCE(!*readstate); /* -Existing- RCU read-side critsect! */
1937         if (!((mask - 1) & mask))
1938                 return rtrsp;  /* Current RCU reader not extendable. */
1939         /* Bias towards larger numbers of loops. */
1940         i = torture_random(trsp);
1941         i = ((i | (i >> 3)) & RCUTORTURE_RDR_MAX_LOOPS) + 1;
1942         for (j = 0; j < i; j++) {
1943                 mask = rcutorture_extend_mask(*readstate, trsp);
1944                 rcutorture_one_extend(readstate, mask, trsp, &rtrsp[j]);
1945         }
1946         return &rtrsp[j];
1947 }
1948
1949 /*
1950  * Do one read-side critical section, returning false if there was
1951  * no data to read.  Can be invoked both from process context and
1952  * from a timer handler.
1953  */
1954 static bool rcu_torture_one_read(struct torture_random_state *trsp, long myid)
1955 {
1956         bool checkpolling = !(torture_random(trsp) & 0xfff);
1957         unsigned long cookie;
1958         struct rcu_gp_oldstate cookie_full;
1959         int i;
1960         unsigned long started;
1961         unsigned long completed;
1962         int newstate;
1963         struct rcu_torture *p;
1964         int pipe_count;
1965         int readstate = 0;
1966         struct rt_read_seg rtseg[RCUTORTURE_RDR_MAX_SEGS] = { { 0 } };
1967         struct rt_read_seg *rtrsp = &rtseg[0];
1968         struct rt_read_seg *rtrsp1;
1969         unsigned long long ts;
1970
1971         WARN_ON_ONCE(!rcu_is_watching());
1972         newstate = rcutorture_extend_mask(readstate, trsp);
1973         rcutorture_one_extend(&readstate, newstate, trsp, rtrsp++);
1974         if (checkpolling) {
1975                 if (cur_ops->get_gp_state && cur_ops->poll_gp_state)
1976                         cookie = cur_ops->get_gp_state();
1977                 if (cur_ops->get_gp_state_full && cur_ops->poll_gp_state_full)
1978                         cur_ops->get_gp_state_full(&cookie_full);
1979         }
1980         started = cur_ops->get_gp_seq();
1981         ts = rcu_trace_clock_local();
1982         p = rcu_dereference_check(rcu_torture_current,
1983                                   !cur_ops->readlock_held || cur_ops->readlock_held());
1984         if (p == NULL) {
1985                 /* Wait for rcu_torture_writer to get underway */
1986                 rcutorture_one_extend(&readstate, 0, trsp, rtrsp);
1987                 return false;
1988         }
1989         if (p->rtort_mbtest == 0)
1990                 atomic_inc(&n_rcu_torture_mberror);
1991         rcu_torture_reader_do_mbchk(myid, p, trsp);
1992         rtrsp = rcutorture_loop_extend(&readstate, trsp, rtrsp);
1993         preempt_disable();
1994         pipe_count = READ_ONCE(p->rtort_pipe_count);
1995         if (pipe_count > RCU_TORTURE_PIPE_LEN) {
1996                 /* Should not happen, but... */
1997                 pipe_count = RCU_TORTURE_PIPE_LEN;
1998         }
1999         completed = cur_ops->get_gp_seq();
2000         if (pipe_count > 1) {
2001                 do_trace_rcu_torture_read(cur_ops->name, &p->rtort_rcu,
2002                                           ts, started, completed);
2003                 rcu_ftrace_dump(DUMP_ALL);
2004         }
2005         __this_cpu_inc(rcu_torture_count[pipe_count]);
2006         completed = rcutorture_seq_diff(completed, started);
2007         if (completed > RCU_TORTURE_PIPE_LEN) {
2008                 /* Should not happen, but... */
2009                 completed = RCU_TORTURE_PIPE_LEN;
2010         }
2011         __this_cpu_inc(rcu_torture_batch[completed]);
2012         preempt_enable();
2013         if (checkpolling) {
2014                 if (cur_ops->get_gp_state && cur_ops->poll_gp_state)
2015                         WARN_ONCE(cur_ops->poll_gp_state(cookie),
2016                                   "%s: Cookie check 2 failed %s(%d) %lu->%lu\n",
2017                                   __func__,
2018                                   rcu_torture_writer_state_getname(),
2019                                   rcu_torture_writer_state,
2020                                   cookie, cur_ops->get_gp_state());
2021                 if (cur_ops->get_gp_state_full && cur_ops->poll_gp_state_full)
2022                         WARN_ONCE(cur_ops->poll_gp_state_full(&cookie_full),
2023                                   "%s: Cookie check 6 failed %s(%d) online %*pbl\n",
2024                                   __func__,
2025                                   rcu_torture_writer_state_getname(),
2026                                   rcu_torture_writer_state,
2027                                   cpumask_pr_args(cpu_online_mask));
2028         }
2029         rcutorture_one_extend(&readstate, 0, trsp, rtrsp);
2030         WARN_ON_ONCE(readstate);
2031         // This next splat is expected behavior if leakpointer, especially
2032         // for CONFIG_RCU_STRICT_GRACE_PERIOD=y kernels.
2033         WARN_ON_ONCE(leakpointer && READ_ONCE(p->rtort_pipe_count) > 1);
2034
2035         /* If error or close call, record the sequence of reader protections. */
2036         if ((pipe_count > 1 || completed > 1) && !xchg(&err_segs_recorded, 1)) {
2037                 i = 0;
2038                 for (rtrsp1 = &rtseg[0]; rtrsp1 < rtrsp; rtrsp1++)
2039                         err_segs[i++] = *rtrsp1;
2040                 rt_read_nsegs = i;
2041         }
2042
2043         return true;
2044 }
2045
2046 static DEFINE_TORTURE_RANDOM_PERCPU(rcu_torture_timer_rand);
2047
2048 /*
2049  * RCU torture reader from timer handler.  Dereferences rcu_torture_current,
2050  * incrementing the corresponding element of the pipeline array.  The
2051  * counter in the element should never be greater than 1, otherwise, the
2052  * RCU implementation is broken.
2053  */
2054 static void rcu_torture_timer(struct timer_list *unused)
2055 {
2056         atomic_long_inc(&n_rcu_torture_timers);
2057         (void)rcu_torture_one_read(this_cpu_ptr(&rcu_torture_timer_rand), -1);
2058
2059         /* Test call_rcu() invocation from interrupt handler. */
2060         if (cur_ops->call) {
2061                 struct rcu_head *rhp = kmalloc(sizeof(*rhp), GFP_NOWAIT);
2062
2063                 if (rhp)
2064                         cur_ops->call(rhp, rcu_torture_timer_cb);
2065         }
2066 }
2067
2068 /*
2069  * RCU torture reader kthread.  Repeatedly dereferences rcu_torture_current,
2070  * incrementing the corresponding element of the pipeline array.  The
2071  * counter in the element should never be greater than 1, otherwise, the
2072  * RCU implementation is broken.
2073  */
2074 static int
2075 rcu_torture_reader(void *arg)
2076 {
2077         unsigned long lastsleep = jiffies;
2078         long myid = (long)arg;
2079         int mynumonline = myid;
2080         DEFINE_TORTURE_RANDOM(rand);
2081         struct timer_list t;
2082
2083         VERBOSE_TOROUT_STRING("rcu_torture_reader task started");
2084         set_user_nice(current, MAX_NICE);
2085         if (irqreader && cur_ops->irq_capable)
2086                 timer_setup_on_stack(&t, rcu_torture_timer, 0);
2087         tick_dep_set_task(current, TICK_DEP_BIT_RCU);
2088         do {
2089                 if (irqreader && cur_ops->irq_capable) {
2090                         if (!timer_pending(&t))
2091                                 mod_timer(&t, jiffies + 1);
2092                 }
2093                 if (!rcu_torture_one_read(&rand, myid) && !torture_must_stop())
2094                         schedule_timeout_interruptible(HZ);
2095                 if (time_after(jiffies, lastsleep) && !torture_must_stop()) {
2096                         torture_hrtimeout_us(500, 1000, &rand);
2097                         lastsleep = jiffies + 10;
2098                 }
2099                 while (torture_num_online_cpus() < mynumonline && !torture_must_stop())
2100                         schedule_timeout_interruptible(HZ / 5);
2101                 stutter_wait("rcu_torture_reader");
2102         } while (!torture_must_stop());
2103         if (irqreader && cur_ops->irq_capable) {
2104                 del_timer_sync(&t);
2105                 destroy_timer_on_stack(&t);
2106         }
2107         tick_dep_clear_task(current, TICK_DEP_BIT_RCU);
2108         torture_kthread_stopping("rcu_torture_reader");
2109         return 0;
2110 }
2111
2112 /*
2113  * Randomly Toggle CPUs' callback-offload state.  This uses hrtimers to
2114  * increase race probabilities and fuzzes the interval between toggling.
2115  */
2116 static int rcu_nocb_toggle(void *arg)
2117 {
2118         int cpu;
2119         int maxcpu = -1;
2120         int oldnice = task_nice(current);
2121         long r;
2122         DEFINE_TORTURE_RANDOM(rand);
2123         ktime_t toggle_delay;
2124         unsigned long toggle_fuzz;
2125         ktime_t toggle_interval = ms_to_ktime(nocbs_toggle);
2126
2127         VERBOSE_TOROUT_STRING("rcu_nocb_toggle task started");
2128         while (!rcu_inkernel_boot_has_ended())
2129                 schedule_timeout_interruptible(HZ / 10);
2130         for_each_possible_cpu(cpu)
2131                 maxcpu = cpu;
2132         WARN_ON(maxcpu < 0);
2133         if (toggle_interval > ULONG_MAX)
2134                 toggle_fuzz = ULONG_MAX >> 3;
2135         else
2136                 toggle_fuzz = toggle_interval >> 3;
2137         if (toggle_fuzz <= 0)
2138                 toggle_fuzz = NSEC_PER_USEC;
2139         do {
2140                 r = torture_random(&rand);
2141                 cpu = (r >> 1) % (maxcpu + 1);
2142                 if (r & 0x1) {
2143                         rcu_nocb_cpu_offload(cpu);
2144                         atomic_long_inc(&n_nocb_offload);
2145                 } else {
2146                         rcu_nocb_cpu_deoffload(cpu);
2147                         atomic_long_inc(&n_nocb_deoffload);
2148                 }
2149                 toggle_delay = torture_random(&rand) % toggle_fuzz + toggle_interval;
2150                 set_current_state(TASK_INTERRUPTIBLE);
2151                 schedule_hrtimeout(&toggle_delay, HRTIMER_MODE_REL);
2152                 if (stutter_wait("rcu_nocb_toggle"))
2153                         sched_set_normal(current, oldnice);
2154         } while (!torture_must_stop());
2155         torture_kthread_stopping("rcu_nocb_toggle");
2156         return 0;
2157 }
2158
2159 /*
2160  * Print torture statistics.  Caller must ensure that there is only
2161  * one call to this function at a given time!!!  This is normally
2162  * accomplished by relying on the module system to only have one copy
2163  * of the module loaded, and then by giving the rcu_torture_stats
2164  * kthread full control (or the init/cleanup functions when rcu_torture_stats
2165  * thread is not running).
2166  */
2167 static void
2168 rcu_torture_stats_print(void)
2169 {
2170         int cpu;
2171         int i;
2172         long pipesummary[RCU_TORTURE_PIPE_LEN + 1] = { 0 };
2173         long batchsummary[RCU_TORTURE_PIPE_LEN + 1] = { 0 };
2174         struct rcu_torture *rtcp;
2175         static unsigned long rtcv_snap = ULONG_MAX;
2176         static bool splatted;
2177         struct task_struct *wtp;
2178
2179         for_each_possible_cpu(cpu) {
2180                 for (i = 0; i < RCU_TORTURE_PIPE_LEN + 1; i++) {
2181                         pipesummary[i] += READ_ONCE(per_cpu(rcu_torture_count, cpu)[i]);
2182                         batchsummary[i] += READ_ONCE(per_cpu(rcu_torture_batch, cpu)[i]);
2183                 }
2184         }
2185         for (i = RCU_TORTURE_PIPE_LEN; i >= 0; i--) {
2186                 if (pipesummary[i] != 0)
2187                         break;
2188         }
2189
2190         pr_alert("%s%s ", torture_type, TORTURE_FLAG);
2191         rtcp = rcu_access_pointer(rcu_torture_current);
2192         pr_cont("rtc: %p %s: %lu tfle: %d rta: %d rtaf: %d rtf: %d ",
2193                 rtcp,
2194                 rtcp && !rcu_stall_is_suppressed_at_boot() ? "ver" : "VER",
2195                 rcu_torture_current_version,
2196                 list_empty(&rcu_torture_freelist),
2197                 atomic_read(&n_rcu_torture_alloc),
2198                 atomic_read(&n_rcu_torture_alloc_fail),
2199                 atomic_read(&n_rcu_torture_free));
2200         pr_cont("rtmbe: %d rtmbkf: %d/%d rtbe: %ld rtbke: %ld ",
2201                 atomic_read(&n_rcu_torture_mberror),
2202                 atomic_read(&n_rcu_torture_mbchk_fail), atomic_read(&n_rcu_torture_mbchk_tries),
2203                 n_rcu_torture_barrier_error,
2204                 n_rcu_torture_boost_ktrerror);
2205         pr_cont("rtbf: %ld rtb: %ld nt: %ld ",
2206                 n_rcu_torture_boost_failure,
2207                 n_rcu_torture_boosts,
2208                 atomic_long_read(&n_rcu_torture_timers));
2209         torture_onoff_stats();
2210         pr_cont("barrier: %ld/%ld:%ld ",
2211                 data_race(n_barrier_successes),
2212                 data_race(n_barrier_attempts),
2213                 data_race(n_rcu_torture_barrier_error));
2214         pr_cont("read-exits: %ld ", data_race(n_read_exits)); // Statistic.
2215         pr_cont("nocb-toggles: %ld:%ld\n",
2216                 atomic_long_read(&n_nocb_offload), atomic_long_read(&n_nocb_deoffload));
2217
2218         pr_alert("%s%s ", torture_type, TORTURE_FLAG);
2219         if (atomic_read(&n_rcu_torture_mberror) ||
2220             atomic_read(&n_rcu_torture_mbchk_fail) ||
2221             n_rcu_torture_barrier_error || n_rcu_torture_boost_ktrerror ||
2222             n_rcu_torture_boost_failure || i > 1) {
2223                 pr_cont("%s", "!!! ");
2224                 atomic_inc(&n_rcu_torture_error);
2225                 WARN_ON_ONCE(atomic_read(&n_rcu_torture_mberror));
2226                 WARN_ON_ONCE(atomic_read(&n_rcu_torture_mbchk_fail));
2227                 WARN_ON_ONCE(n_rcu_torture_barrier_error);  // rcu_barrier()
2228                 WARN_ON_ONCE(n_rcu_torture_boost_ktrerror); // no boost kthread
2229                 WARN_ON_ONCE(n_rcu_torture_boost_failure); // boost failed (TIMER_SOFTIRQ RT prio?)
2230                 WARN_ON_ONCE(i > 1); // Too-short grace period
2231         }
2232         pr_cont("Reader Pipe: ");
2233         for (i = 0; i < RCU_TORTURE_PIPE_LEN + 1; i++)
2234                 pr_cont(" %ld", pipesummary[i]);
2235         pr_cont("\n");
2236
2237         pr_alert("%s%s ", torture_type, TORTURE_FLAG);
2238         pr_cont("Reader Batch: ");
2239         for (i = 0; i < RCU_TORTURE_PIPE_LEN + 1; i++)
2240                 pr_cont(" %ld", batchsummary[i]);
2241         pr_cont("\n");
2242
2243         pr_alert("%s%s ", torture_type, TORTURE_FLAG);
2244         pr_cont("Free-Block Circulation: ");
2245         for (i = 0; i < RCU_TORTURE_PIPE_LEN + 1; i++) {
2246                 pr_cont(" %d", atomic_read(&rcu_torture_wcount[i]));
2247         }
2248         pr_cont("\n");
2249
2250         if (cur_ops->stats)
2251                 cur_ops->stats();
2252         if (rtcv_snap == rcu_torture_current_version &&
2253             rcu_access_pointer(rcu_torture_current) &&
2254             !rcu_stall_is_suppressed()) {
2255                 int __maybe_unused flags = 0;
2256                 unsigned long __maybe_unused gp_seq = 0;
2257
2258                 rcutorture_get_gp_data(cur_ops->ttype,
2259                                        &flags, &gp_seq);
2260                 srcutorture_get_gp_data(cur_ops->ttype, srcu_ctlp,
2261                                         &flags, &gp_seq);
2262                 wtp = READ_ONCE(writer_task);
2263                 pr_alert("??? Writer stall state %s(%d) g%lu f%#x ->state %#x cpu %d\n",
2264                          rcu_torture_writer_state_getname(),
2265                          rcu_torture_writer_state, gp_seq, flags,
2266                          wtp == NULL ? ~0U : wtp->__state,
2267                          wtp == NULL ? -1 : (int)task_cpu(wtp));
2268                 if (!splatted && wtp) {
2269                         sched_show_task(wtp);
2270                         splatted = true;
2271                 }
2272                 if (cur_ops->gp_kthread_dbg)
2273                         cur_ops->gp_kthread_dbg();
2274                 rcu_ftrace_dump(DUMP_ALL);
2275         }
2276         rtcv_snap = rcu_torture_current_version;
2277 }
2278
2279 /*
2280  * Periodically prints torture statistics, if periodic statistics printing
2281  * was specified via the stat_interval module parameter.
2282  */
2283 static int
2284 rcu_torture_stats(void *arg)
2285 {
2286         VERBOSE_TOROUT_STRING("rcu_torture_stats task started");
2287         do {
2288                 schedule_timeout_interruptible(stat_interval * HZ);
2289                 rcu_torture_stats_print();
2290                 torture_shutdown_absorb("rcu_torture_stats");
2291         } while (!torture_must_stop());
2292         torture_kthread_stopping("rcu_torture_stats");
2293         return 0;
2294 }
2295
2296 /* Test mem_dump_obj() and friends.  */
2297 static void rcu_torture_mem_dump_obj(void)
2298 {
2299         struct rcu_head *rhp;
2300         struct kmem_cache *kcp;
2301         static int z;
2302
2303         kcp = kmem_cache_create("rcuscale", 136, 8, SLAB_STORE_USER, NULL);
2304         if (WARN_ON_ONCE(!kcp))
2305                 return;
2306         rhp = kmem_cache_alloc(kcp, GFP_KERNEL);
2307         if (WARN_ON_ONCE(!rhp)) {
2308                 kmem_cache_destroy(kcp);
2309                 return;
2310         }
2311         pr_alert("mem_dump_obj() slab test: rcu_torture_stats = %px, &rhp = %px, rhp = %px, &z = %px\n", stats_task, &rhp, rhp, &z);
2312         pr_alert("mem_dump_obj(ZERO_SIZE_PTR):");
2313         mem_dump_obj(ZERO_SIZE_PTR);
2314         pr_alert("mem_dump_obj(NULL):");
2315         mem_dump_obj(NULL);
2316         pr_alert("mem_dump_obj(%px):", &rhp);
2317         mem_dump_obj(&rhp);
2318         pr_alert("mem_dump_obj(%px):", rhp);
2319         mem_dump_obj(rhp);
2320         pr_alert("mem_dump_obj(%px):", &rhp->func);
2321         mem_dump_obj(&rhp->func);
2322         pr_alert("mem_dump_obj(%px):", &z);
2323         mem_dump_obj(&z);
2324         kmem_cache_free(kcp, rhp);
2325         kmem_cache_destroy(kcp);
2326         rhp = kmalloc(sizeof(*rhp), GFP_KERNEL);
2327         if (WARN_ON_ONCE(!rhp))
2328                 return;
2329         pr_alert("mem_dump_obj() kmalloc test: rcu_torture_stats = %px, &rhp = %px, rhp = %px\n", stats_task, &rhp, rhp);
2330         pr_alert("mem_dump_obj(kmalloc %px):", rhp);
2331         mem_dump_obj(rhp);
2332         pr_alert("mem_dump_obj(kmalloc %px):", &rhp->func);
2333         mem_dump_obj(&rhp->func);
2334         kfree(rhp);
2335         rhp = vmalloc(4096);
2336         if (WARN_ON_ONCE(!rhp))
2337                 return;
2338         pr_alert("mem_dump_obj() vmalloc test: rcu_torture_stats = %px, &rhp = %px, rhp = %px\n", stats_task, &rhp, rhp);
2339         pr_alert("mem_dump_obj(vmalloc %px):", rhp);
2340         mem_dump_obj(rhp);
2341         pr_alert("mem_dump_obj(vmalloc %px):", &rhp->func);
2342         mem_dump_obj(&rhp->func);
2343         vfree(rhp);
2344 }
2345
2346 static void
2347 rcu_torture_print_module_parms(struct rcu_torture_ops *cur_ops, const char *tag)
2348 {
2349         pr_alert("%s" TORTURE_FLAG
2350                  "--- %s: nreaders=%d nfakewriters=%d "
2351                  "stat_interval=%d verbose=%d test_no_idle_hz=%d "
2352                  "shuffle_interval=%d stutter=%d irqreader=%d "
2353                  "fqs_duration=%d fqs_holdoff=%d fqs_stutter=%d "
2354                  "test_boost=%d/%d test_boost_interval=%d "
2355                  "test_boost_duration=%d shutdown_secs=%d "
2356                  "stall_cpu=%d stall_cpu_holdoff=%d stall_cpu_irqsoff=%d "
2357                  "stall_cpu_block=%d "
2358                  "n_barrier_cbs=%d "
2359                  "onoff_interval=%d onoff_holdoff=%d "
2360                  "read_exit_delay=%d read_exit_burst=%d "
2361                  "nocbs_nthreads=%d nocbs_toggle=%d "
2362                  "test_nmis=%d\n",
2363                  torture_type, tag, nrealreaders, nfakewriters,
2364                  stat_interval, verbose, test_no_idle_hz, shuffle_interval,
2365                  stutter, irqreader, fqs_duration, fqs_holdoff, fqs_stutter,
2366                  test_boost, cur_ops->can_boost,
2367                  test_boost_interval, test_boost_duration, shutdown_secs,
2368                  stall_cpu, stall_cpu_holdoff, stall_cpu_irqsoff,
2369                  stall_cpu_block,
2370                  n_barrier_cbs,
2371                  onoff_interval, onoff_holdoff,
2372                  read_exit_delay, read_exit_burst,
2373                  nocbs_nthreads, nocbs_toggle,
2374                  test_nmis);
2375 }
2376
2377 static int rcutorture_booster_cleanup(unsigned int cpu)
2378 {
2379         struct task_struct *t;
2380
2381         if (boost_tasks[cpu] == NULL)
2382                 return 0;
2383         mutex_lock(&boost_mutex);
2384         t = boost_tasks[cpu];
2385         boost_tasks[cpu] = NULL;
2386         rcu_torture_enable_rt_throttle();
2387         mutex_unlock(&boost_mutex);
2388
2389         /* This must be outside of the mutex, otherwise deadlock! */
2390         torture_stop_kthread(rcu_torture_boost, t);
2391         return 0;
2392 }
2393
2394 static int rcutorture_booster_init(unsigned int cpu)
2395 {
2396         int retval;
2397
2398         if (boost_tasks[cpu] != NULL)
2399                 return 0;  /* Already created, nothing more to do. */
2400
2401         // Testing RCU priority boosting requires rcutorture do
2402         // some serious abuse.  Counter this by running ksoftirqd
2403         // at higher priority.
2404         if (IS_BUILTIN(CONFIG_RCU_TORTURE_TEST)) {
2405                 struct sched_param sp;
2406                 struct task_struct *t;
2407
2408                 t = per_cpu(ksoftirqd, cpu);
2409                 WARN_ON_ONCE(!t);
2410                 sp.sched_priority = 2;
2411                 sched_setscheduler_nocheck(t, SCHED_FIFO, &sp);
2412         }
2413
2414         /* Don't allow time recalculation while creating a new task. */
2415         mutex_lock(&boost_mutex);
2416         rcu_torture_disable_rt_throttle();
2417         VERBOSE_TOROUT_STRING("Creating rcu_torture_boost task");
2418         boost_tasks[cpu] = kthread_run_on_cpu(rcu_torture_boost, NULL,
2419                                               cpu, "rcu_torture_boost_%u");
2420         if (IS_ERR(boost_tasks[cpu])) {
2421                 retval = PTR_ERR(boost_tasks[cpu]);
2422                 VERBOSE_TOROUT_STRING("rcu_torture_boost task create failed");
2423                 n_rcu_torture_boost_ktrerror++;
2424                 boost_tasks[cpu] = NULL;
2425                 mutex_unlock(&boost_mutex);
2426                 return retval;
2427         }
2428         mutex_unlock(&boost_mutex);
2429         return 0;
2430 }
2431
2432 static int rcu_torture_stall_nf(struct notifier_block *nb, unsigned long v, void *ptr)
2433 {
2434         pr_info("%s: v=%lu, duration=%lu.\n", __func__, v, (unsigned long)ptr);
2435         return NOTIFY_OK;
2436 }
2437
2438 static struct notifier_block rcu_torture_stall_block = {
2439         .notifier_call = rcu_torture_stall_nf,
2440 };
2441
2442 /*
2443  * CPU-stall kthread.  It waits as specified by stall_cpu_holdoff, then
2444  * induces a CPU stall for the time specified by stall_cpu.
2445  */
2446 static int rcu_torture_stall(void *args)
2447 {
2448         int idx;
2449         int ret;
2450         unsigned long stop_at;
2451
2452         VERBOSE_TOROUT_STRING("rcu_torture_stall task started");
2453         if (rcu_cpu_stall_notifiers) {
2454                 ret = rcu_stall_chain_notifier_register(&rcu_torture_stall_block);
2455                 if (ret)
2456                         pr_info("%s: rcu_stall_chain_notifier_register() returned %d, %sexpected.\n",
2457                                 __func__, ret, !IS_ENABLED(CONFIG_RCU_STALL_COMMON) ? "un" : "");
2458         }
2459         if (stall_cpu_holdoff > 0) {
2460                 VERBOSE_TOROUT_STRING("rcu_torture_stall begin holdoff");
2461                 schedule_timeout_interruptible(stall_cpu_holdoff * HZ);
2462                 VERBOSE_TOROUT_STRING("rcu_torture_stall end holdoff");
2463         }
2464         if (!kthread_should_stop() && stall_gp_kthread > 0) {
2465                 VERBOSE_TOROUT_STRING("rcu_torture_stall begin GP stall");
2466                 rcu_gp_set_torture_wait(stall_gp_kthread * HZ);
2467                 for (idx = 0; idx < stall_gp_kthread + 2; idx++) {
2468                         if (kthread_should_stop())
2469                                 break;
2470                         schedule_timeout_uninterruptible(HZ);
2471                 }
2472         }
2473         if (!kthread_should_stop() && stall_cpu > 0) {
2474                 VERBOSE_TOROUT_STRING("rcu_torture_stall begin CPU stall");
2475                 stop_at = ktime_get_seconds() + stall_cpu;
2476                 /* RCU CPU stall is expected behavior in following code. */
2477                 idx = cur_ops->readlock();
2478                 if (stall_cpu_irqsoff)
2479                         local_irq_disable();
2480                 else if (!stall_cpu_block)
2481                         preempt_disable();
2482                 pr_alert("%s start on CPU %d.\n",
2483                           __func__, raw_smp_processor_id());
2484                 while (ULONG_CMP_LT((unsigned long)ktime_get_seconds(),
2485                                     stop_at))
2486                         if (stall_cpu_block) {
2487 #ifdef CONFIG_PREEMPTION
2488                                 preempt_schedule();
2489 #else
2490                                 schedule_timeout_uninterruptible(HZ);
2491 #endif
2492                         } else if (stall_no_softlockup) {
2493                                 touch_softlockup_watchdog();
2494                         }
2495                 if (stall_cpu_irqsoff)
2496                         local_irq_enable();
2497                 else if (!stall_cpu_block)
2498                         preempt_enable();
2499                 cur_ops->readunlock(idx);
2500         }
2501         pr_alert("%s end.\n", __func__);
2502         if (rcu_cpu_stall_notifiers && !ret) {
2503                 ret = rcu_stall_chain_notifier_unregister(&rcu_torture_stall_block);
2504                 if (ret)
2505                         pr_info("%s: rcu_stall_chain_notifier_unregister() returned %d.\n", __func__, ret);
2506         }
2507         torture_shutdown_absorb("rcu_torture_stall");
2508         while (!kthread_should_stop())
2509                 schedule_timeout_interruptible(10 * HZ);
2510         return 0;
2511 }
2512
2513 /* Spawn CPU-stall kthread, if stall_cpu specified. */
2514 static int __init rcu_torture_stall_init(void)
2515 {
2516         if (stall_cpu <= 0 && stall_gp_kthread <= 0)
2517                 return 0;
2518         return torture_create_kthread(rcu_torture_stall, NULL, stall_task);
2519 }
2520
2521 /* State structure for forward-progress self-propagating RCU callback. */
2522 struct fwd_cb_state {
2523         struct rcu_head rh;
2524         int stop;
2525 };
2526
2527 /*
2528  * Forward-progress self-propagating RCU callback function.  Because
2529  * callbacks run from softirq, this function is an implicit RCU read-side
2530  * critical section.
2531  */
2532 static void rcu_torture_fwd_prog_cb(struct rcu_head *rhp)
2533 {
2534         struct fwd_cb_state *fcsp = container_of(rhp, struct fwd_cb_state, rh);
2535
2536         if (READ_ONCE(fcsp->stop)) {
2537                 WRITE_ONCE(fcsp->stop, 2);
2538                 return;
2539         }
2540         cur_ops->call(&fcsp->rh, rcu_torture_fwd_prog_cb);
2541 }
2542
2543 /* State for continuous-flood RCU callbacks. */
2544 struct rcu_fwd_cb {
2545         struct rcu_head rh;
2546         struct rcu_fwd_cb *rfc_next;
2547         struct rcu_fwd *rfc_rfp;
2548         int rfc_gps;
2549 };
2550
2551 #define MAX_FWD_CB_JIFFIES      (8 * HZ) /* Maximum CB test duration. */
2552 #define MIN_FWD_CB_LAUNDERS     3       /* This many CB invocations to count. */
2553 #define MIN_FWD_CBS_LAUNDERED   100     /* Number of counted CBs. */
2554 #define FWD_CBS_HIST_DIV        10      /* Histogram buckets/second. */
2555 #define N_LAUNDERS_HIST (2 * MAX_FWD_CB_JIFFIES / (HZ / FWD_CBS_HIST_DIV))
2556
2557 struct rcu_launder_hist {
2558         long n_launders;
2559         unsigned long launder_gp_seq;
2560 };
2561
2562 struct rcu_fwd {
2563         spinlock_t rcu_fwd_lock;
2564         struct rcu_fwd_cb *rcu_fwd_cb_head;
2565         struct rcu_fwd_cb **rcu_fwd_cb_tail;
2566         long n_launders_cb;
2567         unsigned long rcu_fwd_startat;
2568         struct rcu_launder_hist n_launders_hist[N_LAUNDERS_HIST];
2569         unsigned long rcu_launder_gp_seq_start;
2570         int rcu_fwd_id;
2571 };
2572
2573 static DEFINE_MUTEX(rcu_fwd_mutex);
2574 static struct rcu_fwd *rcu_fwds;
2575 static unsigned long rcu_fwd_seq;
2576 static atomic_long_t rcu_fwd_max_cbs;
2577 static bool rcu_fwd_emergency_stop;
2578
2579 static void rcu_torture_fwd_cb_hist(struct rcu_fwd *rfp)
2580 {
2581         unsigned long gps;
2582         unsigned long gps_old;
2583         int i;
2584         int j;
2585
2586         for (i = ARRAY_SIZE(rfp->n_launders_hist) - 1; i > 0; i--)
2587                 if (rfp->n_launders_hist[i].n_launders > 0)
2588                         break;
2589         pr_alert("%s: Callback-invocation histogram %d (duration %lu jiffies):",
2590                  __func__, rfp->rcu_fwd_id, jiffies - rfp->rcu_fwd_startat);
2591         gps_old = rfp->rcu_launder_gp_seq_start;
2592         for (j = 0; j <= i; j++) {
2593                 gps = rfp->n_launders_hist[j].launder_gp_seq;
2594                 pr_cont(" %ds/%d: %ld:%ld",
2595                         j + 1, FWD_CBS_HIST_DIV,
2596                         rfp->n_launders_hist[j].n_launders,
2597                         rcutorture_seq_diff(gps, gps_old));
2598                 gps_old = gps;
2599         }
2600         pr_cont("\n");
2601 }
2602
2603 /* Callback function for continuous-flood RCU callbacks. */
2604 static void rcu_torture_fwd_cb_cr(struct rcu_head *rhp)
2605 {
2606         unsigned long flags;
2607         int i;
2608         struct rcu_fwd_cb *rfcp = container_of(rhp, struct rcu_fwd_cb, rh);
2609         struct rcu_fwd_cb **rfcpp;
2610         struct rcu_fwd *rfp = rfcp->rfc_rfp;
2611
2612         rfcp->rfc_next = NULL;
2613         rfcp->rfc_gps++;
2614         spin_lock_irqsave(&rfp->rcu_fwd_lock, flags);
2615         rfcpp = rfp->rcu_fwd_cb_tail;
2616         rfp->rcu_fwd_cb_tail = &rfcp->rfc_next;
2617         WRITE_ONCE(*rfcpp, rfcp);
2618         WRITE_ONCE(rfp->n_launders_cb, rfp->n_launders_cb + 1);
2619         i = ((jiffies - rfp->rcu_fwd_startat) / (HZ / FWD_CBS_HIST_DIV));
2620         if (i >= ARRAY_SIZE(rfp->n_launders_hist))
2621                 i = ARRAY_SIZE(rfp->n_launders_hist) - 1;
2622         rfp->n_launders_hist[i].n_launders++;
2623         rfp->n_launders_hist[i].launder_gp_seq = cur_ops->get_gp_seq();
2624         spin_unlock_irqrestore(&rfp->rcu_fwd_lock, flags);
2625 }
2626
2627 // Give the scheduler a chance, even on nohz_full CPUs.
2628 static void rcu_torture_fwd_prog_cond_resched(unsigned long iter)
2629 {
2630         if (IS_ENABLED(CONFIG_PREEMPTION) && IS_ENABLED(CONFIG_NO_HZ_FULL)) {
2631                 // Real call_rcu() floods hit userspace, so emulate that.
2632                 if (need_resched() || (iter & 0xfff))
2633                         schedule();
2634                 return;
2635         }
2636         // No userspace emulation: CB invocation throttles call_rcu()
2637         cond_resched();
2638 }
2639
2640 /*
2641  * Free all callbacks on the rcu_fwd_cb_head list, either because the
2642  * test is over or because we hit an OOM event.
2643  */
2644 static unsigned long rcu_torture_fwd_prog_cbfree(struct rcu_fwd *rfp)
2645 {
2646         unsigned long flags;
2647         unsigned long freed = 0;
2648         struct rcu_fwd_cb *rfcp;
2649
2650         for (;;) {
2651                 spin_lock_irqsave(&rfp->rcu_fwd_lock, flags);
2652                 rfcp = rfp->rcu_fwd_cb_head;
2653                 if (!rfcp) {
2654                         spin_unlock_irqrestore(&rfp->rcu_fwd_lock, flags);
2655                         break;
2656                 }
2657                 rfp->rcu_fwd_cb_head = rfcp->rfc_next;
2658                 if (!rfp->rcu_fwd_cb_head)
2659                         rfp->rcu_fwd_cb_tail = &rfp->rcu_fwd_cb_head;
2660                 spin_unlock_irqrestore(&rfp->rcu_fwd_lock, flags);
2661                 kfree(rfcp);
2662                 freed++;
2663                 rcu_torture_fwd_prog_cond_resched(freed);
2664                 if (tick_nohz_full_enabled()) {
2665                         local_irq_save(flags);
2666                         rcu_momentary_dyntick_idle();
2667                         local_irq_restore(flags);
2668                 }
2669         }
2670         return freed;
2671 }
2672
2673 /* Carry out need_resched()/cond_resched() forward-progress testing. */
2674 static void rcu_torture_fwd_prog_nr(struct rcu_fwd *rfp,
2675                                     int *tested, int *tested_tries)
2676 {
2677         unsigned long cver;
2678         unsigned long dur;
2679         struct fwd_cb_state fcs;
2680         unsigned long gps;
2681         int idx;
2682         int sd;
2683         int sd4;
2684         bool selfpropcb = false;
2685         unsigned long stopat;
2686         static DEFINE_TORTURE_RANDOM(trs);
2687
2688         pr_alert("%s: Starting forward-progress test %d\n", __func__, rfp->rcu_fwd_id);
2689         if (!cur_ops->sync)
2690                 return; // Cannot do need_resched() forward progress testing without ->sync.
2691         if (cur_ops->call && cur_ops->cb_barrier) {
2692                 init_rcu_head_on_stack(&fcs.rh);
2693                 selfpropcb = true;
2694         }
2695
2696         /* Tight loop containing cond_resched(). */
2697         atomic_inc(&rcu_fwd_cb_nodelay);
2698         cur_ops->sync(); /* Later readers see above write. */
2699         if  (selfpropcb) {
2700                 WRITE_ONCE(fcs.stop, 0);
2701                 cur_ops->call(&fcs.rh, rcu_torture_fwd_prog_cb);
2702         }
2703         cver = READ_ONCE(rcu_torture_current_version);
2704         gps = cur_ops->get_gp_seq();
2705         sd = cur_ops->stall_dur() + 1;
2706         sd4 = (sd + fwd_progress_div - 1) / fwd_progress_div;
2707         dur = sd4 + torture_random(&trs) % (sd - sd4);
2708         WRITE_ONCE(rfp->rcu_fwd_startat, jiffies);
2709         stopat = rfp->rcu_fwd_startat + dur;
2710         while (time_before(jiffies, stopat) &&
2711                !shutdown_time_arrived() &&
2712                !READ_ONCE(rcu_fwd_emergency_stop) && !torture_must_stop()) {
2713                 idx = cur_ops->readlock();
2714                 udelay(10);
2715                 cur_ops->readunlock(idx);
2716                 if (!fwd_progress_need_resched || need_resched())
2717                         cond_resched();
2718         }
2719         (*tested_tries)++;
2720         if (!time_before(jiffies, stopat) &&
2721             !shutdown_time_arrived() &&
2722             !READ_ONCE(rcu_fwd_emergency_stop) && !torture_must_stop()) {
2723                 (*tested)++;
2724                 cver = READ_ONCE(rcu_torture_current_version) - cver;
2725                 gps = rcutorture_seq_diff(cur_ops->get_gp_seq(), gps);
2726                 WARN_ON(!cver && gps < 2);
2727                 pr_alert("%s: %d Duration %ld cver %ld gps %ld\n", __func__,
2728                          rfp->rcu_fwd_id, dur, cver, gps);
2729         }
2730         if (selfpropcb) {
2731                 WRITE_ONCE(fcs.stop, 1);
2732                 cur_ops->sync(); /* Wait for running CB to complete. */
2733                 pr_alert("%s: Waiting for CBs: %pS() %d\n", __func__, cur_ops->cb_barrier, rfp->rcu_fwd_id);
2734                 cur_ops->cb_barrier(); /* Wait for queued callbacks. */
2735         }
2736
2737         if (selfpropcb) {
2738                 WARN_ON(READ_ONCE(fcs.stop) != 2);
2739                 destroy_rcu_head_on_stack(&fcs.rh);
2740         }
2741         schedule_timeout_uninterruptible(HZ / 10); /* Let kthreads recover. */
2742         atomic_dec(&rcu_fwd_cb_nodelay);
2743 }
2744
2745 /* Carry out call_rcu() forward-progress testing. */
2746 static void rcu_torture_fwd_prog_cr(struct rcu_fwd *rfp)
2747 {
2748         unsigned long cver;
2749         unsigned long flags;
2750         unsigned long gps;
2751         int i;
2752         long n_launders;
2753         long n_launders_cb_snap;
2754         long n_launders_sa;
2755         long n_max_cbs;
2756         long n_max_gps;
2757         struct rcu_fwd_cb *rfcp;
2758         struct rcu_fwd_cb *rfcpn;
2759         unsigned long stopat;
2760         unsigned long stoppedat;
2761
2762         pr_alert("%s: Starting forward-progress test %d\n", __func__, rfp->rcu_fwd_id);
2763         if (READ_ONCE(rcu_fwd_emergency_stop))
2764                 return; /* Get out of the way quickly, no GP wait! */
2765         if (!cur_ops->call)
2766                 return; /* Can't do call_rcu() fwd prog without ->call. */
2767
2768         /* Loop continuously posting RCU callbacks. */
2769         atomic_inc(&rcu_fwd_cb_nodelay);
2770         cur_ops->sync(); /* Later readers see above write. */
2771         WRITE_ONCE(rfp->rcu_fwd_startat, jiffies);
2772         stopat = rfp->rcu_fwd_startat + MAX_FWD_CB_JIFFIES;
2773         n_launders = 0;
2774         rfp->n_launders_cb = 0; // Hoist initialization for multi-kthread
2775         n_launders_sa = 0;
2776         n_max_cbs = 0;
2777         n_max_gps = 0;
2778         for (i = 0; i < ARRAY_SIZE(rfp->n_launders_hist); i++)
2779                 rfp->n_launders_hist[i].n_launders = 0;
2780         cver = READ_ONCE(rcu_torture_current_version);
2781         gps = cur_ops->get_gp_seq();
2782         rfp->rcu_launder_gp_seq_start = gps;
2783         tick_dep_set_task(current, TICK_DEP_BIT_RCU);
2784         while (time_before(jiffies, stopat) &&
2785                !shutdown_time_arrived() &&
2786                !READ_ONCE(rcu_fwd_emergency_stop) && !torture_must_stop()) {
2787                 rfcp = READ_ONCE(rfp->rcu_fwd_cb_head);
2788                 rfcpn = NULL;
2789                 if (rfcp)
2790                         rfcpn = READ_ONCE(rfcp->rfc_next);
2791                 if (rfcpn) {
2792                         if (rfcp->rfc_gps >= MIN_FWD_CB_LAUNDERS &&
2793                             ++n_max_gps >= MIN_FWD_CBS_LAUNDERED)
2794                                 break;
2795                         rfp->rcu_fwd_cb_head = rfcpn;
2796                         n_launders++;
2797                         n_launders_sa++;
2798                 } else if (!cur_ops->cbflood_max || cur_ops->cbflood_max > n_max_cbs) {
2799                         rfcp = kmalloc(sizeof(*rfcp), GFP_KERNEL);
2800                         if (WARN_ON_ONCE(!rfcp)) {
2801                                 schedule_timeout_interruptible(1);
2802                                 continue;
2803                         }
2804                         n_max_cbs++;
2805                         n_launders_sa = 0;
2806                         rfcp->rfc_gps = 0;
2807                         rfcp->rfc_rfp = rfp;
2808                 } else {
2809                         rfcp = NULL;
2810                 }
2811                 if (rfcp)
2812                         cur_ops->call(&rfcp->rh, rcu_torture_fwd_cb_cr);
2813                 rcu_torture_fwd_prog_cond_resched(n_launders + n_max_cbs);
2814                 if (tick_nohz_full_enabled()) {
2815                         local_irq_save(flags);
2816                         rcu_momentary_dyntick_idle();
2817                         local_irq_restore(flags);
2818                 }
2819         }
2820         stoppedat = jiffies;
2821         n_launders_cb_snap = READ_ONCE(rfp->n_launders_cb);
2822         cver = READ_ONCE(rcu_torture_current_version) - cver;
2823         gps = rcutorture_seq_diff(cur_ops->get_gp_seq(), gps);
2824         pr_alert("%s: Waiting for CBs: %pS() %d\n", __func__, cur_ops->cb_barrier, rfp->rcu_fwd_id);
2825         cur_ops->cb_barrier(); /* Wait for callbacks to be invoked. */
2826         (void)rcu_torture_fwd_prog_cbfree(rfp);
2827
2828         if (!torture_must_stop() && !READ_ONCE(rcu_fwd_emergency_stop) &&
2829             !shutdown_time_arrived()) {
2830                 WARN_ON(n_max_gps < MIN_FWD_CBS_LAUNDERED);
2831                 pr_alert("%s Duration %lu barrier: %lu pending %ld n_launders: %ld n_launders_sa: %ld n_max_gps: %ld n_max_cbs: %ld cver %ld gps %ld\n",
2832                          __func__,
2833                          stoppedat - rfp->rcu_fwd_startat, jiffies - stoppedat,
2834                          n_launders + n_max_cbs - n_launders_cb_snap,
2835                          n_launders, n_launders_sa,
2836                          n_max_gps, n_max_cbs, cver, gps);
2837                 atomic_long_add(n_max_cbs, &rcu_fwd_max_cbs);
2838                 mutex_lock(&rcu_fwd_mutex); // Serialize histograms.
2839                 rcu_torture_fwd_cb_hist(rfp);
2840                 mutex_unlock(&rcu_fwd_mutex);
2841         }
2842         schedule_timeout_uninterruptible(HZ); /* Let CBs drain. */
2843         tick_dep_clear_task(current, TICK_DEP_BIT_RCU);
2844         atomic_dec(&rcu_fwd_cb_nodelay);
2845 }
2846
2847
2848 /*
2849  * OOM notifier, but this only prints diagnostic information for the
2850  * current forward-progress test.
2851  */
2852 static int rcutorture_oom_notify(struct notifier_block *self,
2853                                  unsigned long notused, void *nfreed)
2854 {
2855         int i;
2856         long ncbs;
2857         struct rcu_fwd *rfp;
2858
2859         mutex_lock(&rcu_fwd_mutex);
2860         rfp = rcu_fwds;
2861         if (!rfp) {
2862                 mutex_unlock(&rcu_fwd_mutex);
2863                 return NOTIFY_OK;
2864         }
2865         WARN(1, "%s invoked upon OOM during forward-progress testing.\n",
2866              __func__);
2867         for (i = 0; i < fwd_progress; i++) {
2868                 rcu_torture_fwd_cb_hist(&rfp[i]);
2869                 rcu_fwd_progress_check(1 + (jiffies - READ_ONCE(rfp[i].rcu_fwd_startat)) / 2);
2870         }
2871         WRITE_ONCE(rcu_fwd_emergency_stop, true);
2872         smp_mb(); /* Emergency stop before free and wait to avoid hangs. */
2873         ncbs = 0;
2874         for (i = 0; i < fwd_progress; i++)
2875                 ncbs += rcu_torture_fwd_prog_cbfree(&rfp[i]);
2876         pr_info("%s: Freed %lu RCU callbacks.\n", __func__, ncbs);
2877         cur_ops->cb_barrier();
2878         ncbs = 0;
2879         for (i = 0; i < fwd_progress; i++)
2880                 ncbs += rcu_torture_fwd_prog_cbfree(&rfp[i]);
2881         pr_info("%s: Freed %lu RCU callbacks.\n", __func__, ncbs);
2882         cur_ops->cb_barrier();
2883         ncbs = 0;
2884         for (i = 0; i < fwd_progress; i++)
2885                 ncbs += rcu_torture_fwd_prog_cbfree(&rfp[i]);
2886         pr_info("%s: Freed %lu RCU callbacks.\n", __func__, ncbs);
2887         smp_mb(); /* Frees before return to avoid redoing OOM. */
2888         (*(unsigned long *)nfreed)++; /* Forward progress CBs freed! */
2889         pr_info("%s returning after OOM processing.\n", __func__);
2890         mutex_unlock(&rcu_fwd_mutex);
2891         return NOTIFY_OK;
2892 }
2893
2894 static struct notifier_block rcutorture_oom_nb = {
2895         .notifier_call = rcutorture_oom_notify
2896 };
2897
2898 /* Carry out grace-period forward-progress testing. */
2899 static int rcu_torture_fwd_prog(void *args)
2900 {
2901         bool firsttime = true;
2902         long max_cbs;
2903         int oldnice = task_nice(current);
2904         unsigned long oldseq = READ_ONCE(rcu_fwd_seq);
2905         struct rcu_fwd *rfp = args;
2906         int tested = 0;
2907         int tested_tries = 0;
2908
2909         VERBOSE_TOROUT_STRING("rcu_torture_fwd_progress task started");
2910         rcu_bind_current_to_nocb();
2911         if (!IS_ENABLED(CONFIG_SMP) || !IS_ENABLED(CONFIG_RCU_BOOST))
2912                 set_user_nice(current, MAX_NICE);
2913         do {
2914                 if (!rfp->rcu_fwd_id) {
2915                         schedule_timeout_interruptible(fwd_progress_holdoff * HZ);
2916                         WRITE_ONCE(rcu_fwd_emergency_stop, false);
2917                         if (!firsttime) {
2918                                 max_cbs = atomic_long_xchg(&rcu_fwd_max_cbs, 0);
2919                                 pr_alert("%s n_max_cbs: %ld\n", __func__, max_cbs);
2920                         }
2921                         firsttime = false;
2922                         WRITE_ONCE(rcu_fwd_seq, rcu_fwd_seq + 1);
2923                 } else {
2924                         while (READ_ONCE(rcu_fwd_seq) == oldseq && !torture_must_stop())
2925                                 schedule_timeout_interruptible(HZ / 20);
2926                         oldseq = READ_ONCE(rcu_fwd_seq);
2927                 }
2928                 pr_alert("%s: Starting forward-progress test %d\n", __func__, rfp->rcu_fwd_id);
2929                 if (rcu_inkernel_boot_has_ended() && torture_num_online_cpus() > rfp->rcu_fwd_id)
2930                         rcu_torture_fwd_prog_cr(rfp);
2931                 if ((cur_ops->stall_dur && cur_ops->stall_dur() > 0) &&
2932                     (!IS_ENABLED(CONFIG_TINY_RCU) ||
2933                      (rcu_inkernel_boot_has_ended() &&
2934                       torture_num_online_cpus() > rfp->rcu_fwd_id)))
2935                         rcu_torture_fwd_prog_nr(rfp, &tested, &tested_tries);
2936
2937                 /* Avoid slow periods, better to test when busy. */
2938                 if (stutter_wait("rcu_torture_fwd_prog"))
2939                         sched_set_normal(current, oldnice);
2940         } while (!torture_must_stop());
2941         /* Short runs might not contain a valid forward-progress attempt. */
2942         if (!rfp->rcu_fwd_id) {
2943                 WARN_ON(!tested && tested_tries >= 5);
2944                 pr_alert("%s: tested %d tested_tries %d\n", __func__, tested, tested_tries);
2945         }
2946         torture_kthread_stopping("rcu_torture_fwd_prog");
2947         return 0;
2948 }
2949
2950 /* If forward-progress checking is requested and feasible, spawn the thread. */
2951 static int __init rcu_torture_fwd_prog_init(void)
2952 {
2953         int i;
2954         int ret = 0;
2955         struct rcu_fwd *rfp;
2956
2957         if (!fwd_progress)
2958                 return 0; /* Not requested, so don't do it. */
2959         if (fwd_progress >= nr_cpu_ids) {
2960                 VERBOSE_TOROUT_STRING("rcu_torture_fwd_prog_init: Limiting fwd_progress to # CPUs.\n");
2961                 fwd_progress = nr_cpu_ids;
2962         } else if (fwd_progress < 0) {
2963                 fwd_progress = nr_cpu_ids;
2964         }
2965         if ((!cur_ops->sync && !cur_ops->call) ||
2966             (!cur_ops->cbflood_max && (!cur_ops->stall_dur || cur_ops->stall_dur() <= 0)) ||
2967             cur_ops == &rcu_busted_ops) {
2968                 VERBOSE_TOROUT_STRING("rcu_torture_fwd_prog_init: Disabled, unsupported by RCU flavor under test");
2969                 fwd_progress = 0;
2970                 return 0;
2971         }
2972         if (stall_cpu > 0) {
2973                 VERBOSE_TOROUT_STRING("rcu_torture_fwd_prog_init: Disabled, conflicts with CPU-stall testing");
2974                 fwd_progress = 0;
2975                 if (IS_MODULE(CONFIG_RCU_TORTURE_TEST))
2976                         return -EINVAL; /* In module, can fail back to user. */
2977                 WARN_ON(1); /* Make sure rcutorture notices conflict. */
2978                 return 0;
2979         }
2980         if (fwd_progress_holdoff <= 0)
2981                 fwd_progress_holdoff = 1;
2982         if (fwd_progress_div <= 0)
2983                 fwd_progress_div = 4;
2984         rfp = kcalloc(fwd_progress, sizeof(*rfp), GFP_KERNEL);
2985         fwd_prog_tasks = kcalloc(fwd_progress, sizeof(*fwd_prog_tasks), GFP_KERNEL);
2986         if (!rfp || !fwd_prog_tasks) {
2987                 kfree(rfp);
2988                 kfree(fwd_prog_tasks);
2989                 fwd_prog_tasks = NULL;
2990                 fwd_progress = 0;
2991                 return -ENOMEM;
2992         }
2993         for (i = 0; i < fwd_progress; i++) {
2994                 spin_lock_init(&rfp[i].rcu_fwd_lock);
2995                 rfp[i].rcu_fwd_cb_tail = &rfp[i].rcu_fwd_cb_head;
2996                 rfp[i].rcu_fwd_id = i;
2997         }
2998         mutex_lock(&rcu_fwd_mutex);
2999         rcu_fwds = rfp;
3000         mutex_unlock(&rcu_fwd_mutex);
3001         register_oom_notifier(&rcutorture_oom_nb);
3002         for (i = 0; i < fwd_progress; i++) {
3003                 ret = torture_create_kthread(rcu_torture_fwd_prog, &rcu_fwds[i], fwd_prog_tasks[i]);
3004                 if (ret) {
3005                         fwd_progress = i;
3006                         return ret;
3007                 }
3008         }
3009         return 0;
3010 }
3011
3012 static void rcu_torture_fwd_prog_cleanup(void)
3013 {
3014         int i;
3015         struct rcu_fwd *rfp;
3016
3017         if (!rcu_fwds || !fwd_prog_tasks)
3018                 return;
3019         for (i = 0; i < fwd_progress; i++)
3020                 torture_stop_kthread(rcu_torture_fwd_prog, fwd_prog_tasks[i]);
3021         unregister_oom_notifier(&rcutorture_oom_nb);
3022         mutex_lock(&rcu_fwd_mutex);
3023         rfp = rcu_fwds;
3024         rcu_fwds = NULL;
3025         mutex_unlock(&rcu_fwd_mutex);
3026         kfree(rfp);
3027         kfree(fwd_prog_tasks);
3028         fwd_prog_tasks = NULL;
3029 }
3030
3031 /* Callback function for RCU barrier testing. */
3032 static void rcu_torture_barrier_cbf(struct rcu_head *rcu)
3033 {
3034         atomic_inc(&barrier_cbs_invoked);
3035 }
3036
3037 /* IPI handler to get callback posted on desired CPU, if online. */
3038 static void rcu_torture_barrier1cb(void *rcu_void)
3039 {
3040         struct rcu_head *rhp = rcu_void;
3041
3042         cur_ops->call(rhp, rcu_torture_barrier_cbf);
3043 }
3044
3045 /* kthread function to register callbacks used to test RCU barriers. */
3046 static int rcu_torture_barrier_cbs(void *arg)
3047 {
3048         long myid = (long)arg;
3049         bool lastphase = false;
3050         bool newphase;
3051         struct rcu_head rcu;
3052
3053         init_rcu_head_on_stack(&rcu);
3054         VERBOSE_TOROUT_STRING("rcu_torture_barrier_cbs task started");
3055         set_user_nice(current, MAX_NICE);
3056         do {
3057                 wait_event(barrier_cbs_wq[myid],
3058                            (newphase =
3059                             smp_load_acquire(&barrier_phase)) != lastphase ||
3060                            torture_must_stop());
3061                 lastphase = newphase;
3062                 if (torture_must_stop())
3063                         break;
3064                 /*
3065                  * The above smp_load_acquire() ensures barrier_phase load
3066                  * is ordered before the following ->call().
3067                  */
3068                 if (smp_call_function_single(myid, rcu_torture_barrier1cb,
3069                                              &rcu, 1)) {
3070                         // IPI failed, so use direct call from current CPU.
3071                         cur_ops->call(&rcu, rcu_torture_barrier_cbf);
3072                 }
3073                 if (atomic_dec_and_test(&barrier_cbs_count))
3074                         wake_up(&barrier_wq);
3075         } while (!torture_must_stop());
3076         if (cur_ops->cb_barrier != NULL)
3077                 cur_ops->cb_barrier();
3078         destroy_rcu_head_on_stack(&rcu);
3079         torture_kthread_stopping("rcu_torture_barrier_cbs");
3080         return 0;
3081 }
3082
3083 /* kthread function to drive and coordinate RCU barrier testing. */
3084 static int rcu_torture_barrier(void *arg)
3085 {
3086         int i;
3087
3088         VERBOSE_TOROUT_STRING("rcu_torture_barrier task starting");
3089         do {
3090                 atomic_set(&barrier_cbs_invoked, 0);
3091                 atomic_set(&barrier_cbs_count, n_barrier_cbs);
3092                 /* Ensure barrier_phase ordered after prior assignments. */
3093                 smp_store_release(&barrier_phase, !barrier_phase);
3094                 for (i = 0; i < n_barrier_cbs; i++)
3095                         wake_up(&barrier_cbs_wq[i]);
3096                 wait_event(barrier_wq,
3097                            atomic_read(&barrier_cbs_count) == 0 ||
3098                            torture_must_stop());
3099                 if (torture_must_stop())
3100                         break;
3101                 n_barrier_attempts++;
3102                 cur_ops->cb_barrier(); /* Implies smp_mb() for wait_event(). */
3103                 if (atomic_read(&barrier_cbs_invoked) != n_barrier_cbs) {
3104                         n_rcu_torture_barrier_error++;
3105                         pr_err("barrier_cbs_invoked = %d, n_barrier_cbs = %d\n",
3106                                atomic_read(&barrier_cbs_invoked),
3107                                n_barrier_cbs);
3108                         WARN_ON(1);
3109                         // Wait manually for the remaining callbacks
3110                         i = 0;
3111                         do {
3112                                 if (WARN_ON(i++ > HZ))
3113                                         i = INT_MIN;
3114                                 schedule_timeout_interruptible(1);
3115                                 cur_ops->cb_barrier();
3116                         } while (atomic_read(&barrier_cbs_invoked) !=
3117                                  n_barrier_cbs &&
3118                                  !torture_must_stop());
3119                         smp_mb(); // Can't trust ordering if broken.
3120                         if (!torture_must_stop())
3121                                 pr_err("Recovered: barrier_cbs_invoked = %d\n",
3122                                        atomic_read(&barrier_cbs_invoked));
3123                 } else {
3124                         n_barrier_successes++;
3125                 }
3126                 schedule_timeout_interruptible(HZ / 10);
3127         } while (!torture_must_stop());
3128         torture_kthread_stopping("rcu_torture_barrier");
3129         return 0;
3130 }
3131
3132 /* Initialize RCU barrier testing. */
3133 static int rcu_torture_barrier_init(void)
3134 {
3135         int i;
3136         int ret;
3137
3138         if (n_barrier_cbs <= 0)
3139                 return 0;
3140         if (cur_ops->call == NULL || cur_ops->cb_barrier == NULL) {
3141                 pr_alert("%s" TORTURE_FLAG
3142                          " Call or barrier ops missing for %s,\n",
3143                          torture_type, cur_ops->name);
3144                 pr_alert("%s" TORTURE_FLAG
3145                          " RCU barrier testing omitted from run.\n",
3146                          torture_type);
3147                 return 0;
3148         }
3149         atomic_set(&barrier_cbs_count, 0);
3150         atomic_set(&barrier_cbs_invoked, 0);
3151         barrier_cbs_tasks =
3152                 kcalloc(n_barrier_cbs, sizeof(barrier_cbs_tasks[0]),
3153                         GFP_KERNEL);
3154         barrier_cbs_wq =
3155                 kcalloc(n_barrier_cbs, sizeof(barrier_cbs_wq[0]), GFP_KERNEL);
3156         if (barrier_cbs_tasks == NULL || !barrier_cbs_wq)
3157                 return -ENOMEM;
3158         for (i = 0; i < n_barrier_cbs; i++) {
3159                 init_waitqueue_head(&barrier_cbs_wq[i]);
3160                 ret = torture_create_kthread(rcu_torture_barrier_cbs,
3161                                              (void *)(long)i,
3162                                              barrier_cbs_tasks[i]);
3163                 if (ret)
3164                         return ret;
3165         }
3166         return torture_create_kthread(rcu_torture_barrier, NULL, barrier_task);
3167 }
3168
3169 /* Clean up after RCU barrier testing. */
3170 static void rcu_torture_barrier_cleanup(void)
3171 {
3172         int i;
3173
3174         torture_stop_kthread(rcu_torture_barrier, barrier_task);
3175         if (barrier_cbs_tasks != NULL) {
3176                 for (i = 0; i < n_barrier_cbs; i++)
3177                         torture_stop_kthread(rcu_torture_barrier_cbs,
3178                                              barrier_cbs_tasks[i]);
3179                 kfree(barrier_cbs_tasks);
3180                 barrier_cbs_tasks = NULL;
3181         }
3182         if (barrier_cbs_wq != NULL) {
3183                 kfree(barrier_cbs_wq);
3184                 barrier_cbs_wq = NULL;
3185         }
3186 }
3187
3188 static bool rcu_torture_can_boost(void)
3189 {
3190         static int boost_warn_once;
3191         int prio;
3192
3193         if (!(test_boost == 1 && cur_ops->can_boost) && test_boost != 2)
3194                 return false;
3195         if (!cur_ops->start_gp_poll || !cur_ops->poll_gp_state)
3196                 return false;
3197
3198         prio = rcu_get_gp_kthreads_prio();
3199         if (!prio)
3200                 return false;
3201
3202         if (prio < 2) {
3203                 if (boost_warn_once == 1)
3204                         return false;
3205
3206                 pr_alert("%s: WARN: RCU kthread priority too low to test boosting.  Skipping RCU boost test. Try passing rcutree.kthread_prio > 1 on the kernel command line.\n", KBUILD_MODNAME);
3207                 boost_warn_once = 1;
3208                 return false;
3209         }
3210
3211         return true;
3212 }
3213
3214 static bool read_exit_child_stop;
3215 static bool read_exit_child_stopped;
3216 static wait_queue_head_t read_exit_wq;
3217
3218 // Child kthread which just does an rcutorture reader and exits.
3219 static int rcu_torture_read_exit_child(void *trsp_in)
3220 {
3221         struct torture_random_state *trsp = trsp_in;
3222
3223         set_user_nice(current, MAX_NICE);
3224         // Minimize time between reading and exiting.
3225         while (!kthread_should_stop())
3226                 schedule_timeout_uninterruptible(HZ / 20);
3227         (void)rcu_torture_one_read(trsp, -1);
3228         return 0;
3229 }
3230
3231 // Parent kthread which creates and destroys read-exit child kthreads.
3232 static int rcu_torture_read_exit(void *unused)
3233 {
3234         bool errexit = false;
3235         int i;
3236         struct task_struct *tsp;
3237         DEFINE_TORTURE_RANDOM(trs);
3238
3239         // Allocate and initialize.
3240         set_user_nice(current, MAX_NICE);
3241         VERBOSE_TOROUT_STRING("rcu_torture_read_exit: Start of test");
3242
3243         // Each pass through this loop does one read-exit episode.
3244         do {
3245                 VERBOSE_TOROUT_STRING("rcu_torture_read_exit: Start of episode");
3246                 for (i = 0; i < read_exit_burst; i++) {
3247                         if (READ_ONCE(read_exit_child_stop))
3248                                 break;
3249                         stutter_wait("rcu_torture_read_exit");
3250                         // Spawn child.
3251                         tsp = kthread_run(rcu_torture_read_exit_child,
3252                                           &trs, "%s", "rcu_torture_read_exit_child");
3253                         if (IS_ERR(tsp)) {
3254                                 TOROUT_ERRSTRING("out of memory");
3255                                 errexit = true;
3256                                 break;
3257                         }
3258                         cond_resched();
3259                         kthread_stop(tsp);
3260                         n_read_exits++;
3261                 }
3262                 VERBOSE_TOROUT_STRING("rcu_torture_read_exit: End of episode");
3263                 rcu_barrier(); // Wait for task_struct free, avoid OOM.
3264                 i = 0;
3265                 for (; !errexit && !READ_ONCE(read_exit_child_stop) && i < read_exit_delay; i++)
3266                         schedule_timeout_uninterruptible(HZ);
3267         } while (!errexit && !READ_ONCE(read_exit_child_stop));
3268
3269         // Clean up and exit.
3270         smp_store_release(&read_exit_child_stopped, true); // After reaping.
3271         smp_mb(); // Store before wakeup.
3272         wake_up(&read_exit_wq);
3273         while (!torture_must_stop())
3274                 schedule_timeout_uninterruptible(HZ / 20);
3275         torture_kthread_stopping("rcu_torture_read_exit");
3276         return 0;
3277 }
3278
3279 static int rcu_torture_read_exit_init(void)
3280 {
3281         if (read_exit_burst <= 0)
3282                 return 0;
3283         init_waitqueue_head(&read_exit_wq);
3284         read_exit_child_stop = false;
3285         read_exit_child_stopped = false;
3286         return torture_create_kthread(rcu_torture_read_exit, NULL,
3287                                       read_exit_task);
3288 }
3289
3290 static void rcu_torture_read_exit_cleanup(void)
3291 {
3292         if (!read_exit_task)
3293                 return;
3294         WRITE_ONCE(read_exit_child_stop, true);
3295         smp_mb(); // Above write before wait.
3296         wait_event(read_exit_wq, smp_load_acquire(&read_exit_child_stopped));
3297         torture_stop_kthread(rcutorture_read_exit, read_exit_task);
3298 }
3299
3300 static void rcutorture_test_nmis(int n)
3301 {
3302 #if IS_BUILTIN(CONFIG_RCU_TORTURE_TEST)
3303         int cpu;
3304         int dumpcpu;
3305         int i;
3306
3307         for (i = 0; i < n; i++) {
3308                 preempt_disable();
3309                 cpu = smp_processor_id();
3310                 dumpcpu = cpu + 1;
3311                 if (dumpcpu >= nr_cpu_ids)
3312                         dumpcpu = 0;
3313                 pr_alert("%s: CPU %d invoking dump_cpu_task(%d)\n", __func__, cpu, dumpcpu);
3314                 dump_cpu_task(dumpcpu);
3315                 preempt_enable();
3316                 schedule_timeout_uninterruptible(15 * HZ);
3317         }
3318 #else // #if IS_BUILTIN(CONFIG_RCU_TORTURE_TEST)
3319         WARN_ONCE(n, "Non-zero rcutorture.test_nmis=%d permitted only when rcutorture is built in.\n", test_nmis);
3320 #endif // #else // #if IS_BUILTIN(CONFIG_RCU_TORTURE_TEST)
3321 }
3322
3323 static enum cpuhp_state rcutor_hp;
3324
3325 static void
3326 rcu_torture_cleanup(void)
3327 {
3328         int firsttime;
3329         int flags = 0;
3330         unsigned long gp_seq = 0;
3331         int i;
3332
3333         if (torture_cleanup_begin()) {
3334                 if (cur_ops->cb_barrier != NULL) {
3335                         pr_info("%s: Invoking %pS().\n", __func__, cur_ops->cb_barrier);
3336                         cur_ops->cb_barrier();
3337                 }
3338                 rcu_gp_slow_unregister(NULL);
3339                 return;
3340         }
3341         if (!cur_ops) {
3342                 torture_cleanup_end();
3343                 rcu_gp_slow_unregister(NULL);
3344                 return;
3345         }
3346
3347         rcutorture_test_nmis(test_nmis);
3348
3349         if (cur_ops->gp_kthread_dbg)
3350                 cur_ops->gp_kthread_dbg();
3351         rcu_torture_read_exit_cleanup();
3352         rcu_torture_barrier_cleanup();
3353         rcu_torture_fwd_prog_cleanup();
3354         torture_stop_kthread(rcu_torture_stall, stall_task);
3355         torture_stop_kthread(rcu_torture_writer, writer_task);
3356
3357         if (nocb_tasks) {
3358                 for (i = 0; i < nrealnocbers; i++)
3359                         torture_stop_kthread(rcu_nocb_toggle, nocb_tasks[i]);
3360                 kfree(nocb_tasks);
3361                 nocb_tasks = NULL;
3362         }
3363
3364         if (reader_tasks) {
3365                 for (i = 0; i < nrealreaders; i++)
3366                         torture_stop_kthread(rcu_torture_reader,
3367                                              reader_tasks[i]);
3368                 kfree(reader_tasks);
3369                 reader_tasks = NULL;
3370         }
3371         kfree(rcu_torture_reader_mbchk);
3372         rcu_torture_reader_mbchk = NULL;
3373
3374         if (fakewriter_tasks) {
3375                 for (i = 0; i < nfakewriters; i++)
3376                         torture_stop_kthread(rcu_torture_fakewriter,
3377                                              fakewriter_tasks[i]);
3378                 kfree(fakewriter_tasks);
3379                 fakewriter_tasks = NULL;
3380         }
3381
3382         rcutorture_get_gp_data(cur_ops->ttype, &flags, &gp_seq);
3383         srcutorture_get_gp_data(cur_ops->ttype, srcu_ctlp, &flags, &gp_seq);
3384         pr_alert("%s:  End-test grace-period state: g%ld f%#x total-gps=%ld\n",
3385                  cur_ops->name, (long)gp_seq, flags,
3386                  rcutorture_seq_diff(gp_seq, start_gp_seq));
3387         torture_stop_kthread(rcu_torture_stats, stats_task);
3388         torture_stop_kthread(rcu_torture_fqs, fqs_task);
3389         if (rcu_torture_can_boost() && rcutor_hp >= 0)
3390                 cpuhp_remove_state(rcutor_hp);
3391
3392         /*
3393          * Wait for all RCU callbacks to fire, then do torture-type-specific
3394          * cleanup operations.
3395          */
3396         if (cur_ops->cb_barrier != NULL) {
3397                 pr_info("%s: Invoking %pS().\n", __func__, cur_ops->cb_barrier);
3398                 cur_ops->cb_barrier();
3399         }
3400         if (cur_ops->cleanup != NULL)
3401                 cur_ops->cleanup();
3402
3403         rcu_torture_mem_dump_obj();
3404
3405         rcu_torture_stats_print();  /* -After- the stats thread is stopped! */
3406
3407         if (err_segs_recorded) {
3408                 pr_alert("Failure/close-call rcutorture reader segments:\n");
3409                 if (rt_read_nsegs == 0)
3410                         pr_alert("\t: No segments recorded!!!\n");
3411                 firsttime = 1;
3412                 for (i = 0; i < rt_read_nsegs; i++) {
3413                         pr_alert("\t%d: %#x ", i, err_segs[i].rt_readstate);
3414                         if (err_segs[i].rt_delay_jiffies != 0) {
3415                                 pr_cont("%s%ldjiffies", firsttime ? "" : "+",
3416                                         err_segs[i].rt_delay_jiffies);
3417                                 firsttime = 0;
3418                         }
3419                         if (err_segs[i].rt_delay_ms != 0) {
3420                                 pr_cont("%s%ldms", firsttime ? "" : "+",
3421                                         err_segs[i].rt_delay_ms);
3422                                 firsttime = 0;
3423                         }
3424                         if (err_segs[i].rt_delay_us != 0) {
3425                                 pr_cont("%s%ldus", firsttime ? "" : "+",
3426                                         err_segs[i].rt_delay_us);
3427                                 firsttime = 0;
3428                         }
3429                         pr_cont("%s\n",
3430                                 err_segs[i].rt_preempted ? "preempted" : "");
3431
3432                 }
3433         }
3434         if (atomic_read(&n_rcu_torture_error) || n_rcu_torture_barrier_error)
3435                 rcu_torture_print_module_parms(cur_ops, "End of test: FAILURE");
3436         else if (torture_onoff_failures())
3437                 rcu_torture_print_module_parms(cur_ops,
3438                                                "End of test: RCU_HOTPLUG");
3439         else
3440                 rcu_torture_print_module_parms(cur_ops, "End of test: SUCCESS");
3441         torture_cleanup_end();
3442         rcu_gp_slow_unregister(&rcu_fwd_cb_nodelay);
3443 }
3444
3445 #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD
3446 static void rcu_torture_leak_cb(struct rcu_head *rhp)
3447 {
3448 }
3449
3450 static void rcu_torture_err_cb(struct rcu_head *rhp)
3451 {
3452         /*
3453          * This -might- happen due to race conditions, but is unlikely.
3454          * The scenario that leads to this happening is that the
3455          * first of the pair of duplicate callbacks is queued,
3456          * someone else starts a grace period that includes that
3457          * callback, then the second of the pair must wait for the
3458          * next grace period.  Unlikely, but can happen.  If it
3459          * does happen, the debug-objects subsystem won't have splatted.
3460          */
3461         pr_alert("%s: duplicated callback was invoked.\n", KBUILD_MODNAME);
3462 }
3463 #endif /* #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD */
3464
3465 /*
3466  * Verify that double-free causes debug-objects to complain, but only
3467  * if CONFIG_DEBUG_OBJECTS_RCU_HEAD=y.  Otherwise, say that the test
3468  * cannot be carried out.
3469  */
3470 static void rcu_test_debug_objects(void)
3471 {
3472 #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD
3473         struct rcu_head rh1;
3474         struct rcu_head rh2;
3475         struct rcu_head *rhp = kmalloc(sizeof(*rhp), GFP_KERNEL);
3476
3477         init_rcu_head_on_stack(&rh1);
3478         init_rcu_head_on_stack(&rh2);
3479         pr_alert("%s: WARN: Duplicate call_rcu() test starting.\n", KBUILD_MODNAME);
3480
3481         /* Try to queue the rh2 pair of callbacks for the same grace period. */
3482         preempt_disable(); /* Prevent preemption from interrupting test. */
3483         rcu_read_lock(); /* Make it impossible to finish a grace period. */
3484         call_rcu_hurry(&rh1, rcu_torture_leak_cb); /* Start grace period. */
3485         local_irq_disable(); /* Make it harder to start a new grace period. */
3486         call_rcu_hurry(&rh2, rcu_torture_leak_cb);
3487         call_rcu_hurry(&rh2, rcu_torture_err_cb); /* Duplicate callback. */
3488         if (rhp) {
3489                 call_rcu_hurry(rhp, rcu_torture_leak_cb);
3490                 call_rcu_hurry(rhp, rcu_torture_err_cb); /* Another duplicate callback. */
3491         }
3492         local_irq_enable();
3493         rcu_read_unlock();
3494         preempt_enable();
3495
3496         /* Wait for them all to get done so we can safely return. */
3497         rcu_barrier();
3498         pr_alert("%s: WARN: Duplicate call_rcu() test complete.\n", KBUILD_MODNAME);
3499         destroy_rcu_head_on_stack(&rh1);
3500         destroy_rcu_head_on_stack(&rh2);
3501         kfree(rhp);
3502 #else /* #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD */
3503         pr_alert("%s: !CONFIG_DEBUG_OBJECTS_RCU_HEAD, not testing duplicate call_rcu()\n", KBUILD_MODNAME);
3504 #endif /* #else #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD */
3505 }
3506
3507 static void rcutorture_sync(void)
3508 {
3509         static unsigned long n;
3510
3511         if (cur_ops->sync && !(++n & 0xfff))
3512                 cur_ops->sync();
3513 }
3514
3515 static DEFINE_MUTEX(mut0);
3516 static DEFINE_MUTEX(mut1);
3517 static DEFINE_MUTEX(mut2);
3518 static DEFINE_MUTEX(mut3);
3519 static DEFINE_MUTEX(mut4);
3520 static DEFINE_MUTEX(mut5);
3521 static DEFINE_MUTEX(mut6);
3522 static DEFINE_MUTEX(mut7);
3523 static DEFINE_MUTEX(mut8);
3524 static DEFINE_MUTEX(mut9);
3525
3526 static DECLARE_RWSEM(rwsem0);
3527 static DECLARE_RWSEM(rwsem1);
3528 static DECLARE_RWSEM(rwsem2);
3529 static DECLARE_RWSEM(rwsem3);
3530 static DECLARE_RWSEM(rwsem4);
3531 static DECLARE_RWSEM(rwsem5);
3532 static DECLARE_RWSEM(rwsem6);
3533 static DECLARE_RWSEM(rwsem7);
3534 static DECLARE_RWSEM(rwsem8);
3535 static DECLARE_RWSEM(rwsem9);
3536
3537 DEFINE_STATIC_SRCU(srcu0);
3538 DEFINE_STATIC_SRCU(srcu1);
3539 DEFINE_STATIC_SRCU(srcu2);
3540 DEFINE_STATIC_SRCU(srcu3);
3541 DEFINE_STATIC_SRCU(srcu4);
3542 DEFINE_STATIC_SRCU(srcu5);
3543 DEFINE_STATIC_SRCU(srcu6);
3544 DEFINE_STATIC_SRCU(srcu7);
3545 DEFINE_STATIC_SRCU(srcu8);
3546 DEFINE_STATIC_SRCU(srcu9);
3547
3548 static int srcu_lockdep_next(const char *f, const char *fl, const char *fs, const char *fu, int i,
3549                              int cyclelen, int deadlock)
3550 {
3551         int j = i + 1;
3552
3553         if (j >= cyclelen)
3554                 j = deadlock ? 0 : -1;
3555         if (j >= 0)
3556                 pr_info("%s: %s(%d), %s(%d), %s(%d)\n", f, fl, i, fs, j, fu, i);
3557         else
3558                 pr_info("%s: %s(%d), %s(%d)\n", f, fl, i, fu, i);
3559         return j;
3560 }
3561
3562 // Test lockdep on SRCU-based deadlock scenarios.
3563 static void rcu_torture_init_srcu_lockdep(void)
3564 {
3565         int cyclelen;
3566         int deadlock;
3567         bool err = false;
3568         int i;
3569         int j;
3570         int idx;
3571         struct mutex *muts[] = { &mut0, &mut1, &mut2, &mut3, &mut4,
3572                                  &mut5, &mut6, &mut7, &mut8, &mut9 };
3573         struct rw_semaphore *rwsems[] = { &rwsem0, &rwsem1, &rwsem2, &rwsem3, &rwsem4,
3574                                           &rwsem5, &rwsem6, &rwsem7, &rwsem8, &rwsem9 };
3575         struct srcu_struct *srcus[] = { &srcu0, &srcu1, &srcu2, &srcu3, &srcu4,
3576                                         &srcu5, &srcu6, &srcu7, &srcu8, &srcu9 };
3577         int testtype;
3578
3579         if (!test_srcu_lockdep)
3580                 return;
3581
3582         deadlock = test_srcu_lockdep / 1000;
3583         testtype = (test_srcu_lockdep / 10) % 100;
3584         cyclelen = test_srcu_lockdep % 10;
3585         WARN_ON_ONCE(ARRAY_SIZE(muts) != ARRAY_SIZE(srcus));
3586         if (WARN_ONCE(deadlock != !!deadlock,
3587                       "%s: test_srcu_lockdep=%d and deadlock digit %d must be zero or one.\n",
3588                       __func__, test_srcu_lockdep, deadlock))
3589                 err = true;
3590         if (WARN_ONCE(cyclelen <= 0,
3591                       "%s: test_srcu_lockdep=%d and cycle-length digit %d must be greater than zero.\n",
3592                       __func__, test_srcu_lockdep, cyclelen))
3593                 err = true;
3594         if (err)
3595                 goto err_out;
3596
3597         if (testtype == 0) {
3598                 pr_info("%s: test_srcu_lockdep = %05d: SRCU %d-way %sdeadlock.\n",
3599                         __func__, test_srcu_lockdep, cyclelen, deadlock ? "" : "non-");
3600                 if (deadlock && cyclelen == 1)
3601                         pr_info("%s: Expect hang.\n", __func__);
3602                 for (i = 0; i < cyclelen; i++) {
3603                         j = srcu_lockdep_next(__func__, "srcu_read_lock", "synchronize_srcu",
3604                                               "srcu_read_unlock", i, cyclelen, deadlock);
3605                         idx = srcu_read_lock(srcus[i]);
3606                         if (j >= 0)
3607                                 synchronize_srcu(srcus[j]);
3608                         srcu_read_unlock(srcus[i], idx);
3609                 }
3610                 return;
3611         }
3612
3613         if (testtype == 1) {
3614                 pr_info("%s: test_srcu_lockdep = %05d: SRCU/mutex %d-way %sdeadlock.\n",
3615                         __func__, test_srcu_lockdep, cyclelen, deadlock ? "" : "non-");
3616                 for (i = 0; i < cyclelen; i++) {
3617                         pr_info("%s: srcu_read_lock(%d), mutex_lock(%d), mutex_unlock(%d), srcu_read_unlock(%d)\n",
3618                                 __func__, i, i, i, i);
3619                         idx = srcu_read_lock(srcus[i]);
3620                         mutex_lock(muts[i]);
3621                         mutex_unlock(muts[i]);
3622                         srcu_read_unlock(srcus[i], idx);
3623
3624                         j = srcu_lockdep_next(__func__, "mutex_lock", "synchronize_srcu",
3625                                               "mutex_unlock", i, cyclelen, deadlock);
3626                         mutex_lock(muts[i]);
3627                         if (j >= 0)
3628                                 synchronize_srcu(srcus[j]);
3629                         mutex_unlock(muts[i]);
3630                 }
3631                 return;
3632         }
3633
3634         if (testtype == 2) {
3635                 pr_info("%s: test_srcu_lockdep = %05d: SRCU/rwsem %d-way %sdeadlock.\n",
3636                         __func__, test_srcu_lockdep, cyclelen, deadlock ? "" : "non-");
3637                 for (i = 0; i < cyclelen; i++) {
3638                         pr_info("%s: srcu_read_lock(%d), down_read(%d), up_read(%d), srcu_read_unlock(%d)\n",
3639                                 __func__, i, i, i, i);
3640                         idx = srcu_read_lock(srcus[i]);
3641                         down_read(rwsems[i]);
3642                         up_read(rwsems[i]);
3643                         srcu_read_unlock(srcus[i], idx);
3644
3645                         j = srcu_lockdep_next(__func__, "down_write", "synchronize_srcu",
3646                                               "up_write", i, cyclelen, deadlock);
3647                         down_write(rwsems[i]);
3648                         if (j >= 0)
3649                                 synchronize_srcu(srcus[j]);
3650                         up_write(rwsems[i]);
3651                 }
3652                 return;
3653         }
3654
3655 #ifdef CONFIG_TASKS_TRACE_RCU
3656         if (testtype == 3) {
3657                 pr_info("%s: test_srcu_lockdep = %05d: SRCU and Tasks Trace RCU %d-way %sdeadlock.\n",
3658                         __func__, test_srcu_lockdep, cyclelen, deadlock ? "" : "non-");
3659                 if (deadlock && cyclelen == 1)
3660                         pr_info("%s: Expect hang.\n", __func__);
3661                 for (i = 0; i < cyclelen; i++) {
3662                         char *fl = i == 0 ? "rcu_read_lock_trace" : "srcu_read_lock";
3663                         char *fs = i == cyclelen - 1 ? "synchronize_rcu_tasks_trace"
3664                                                      : "synchronize_srcu";
3665                         char *fu = i == 0 ? "rcu_read_unlock_trace" : "srcu_read_unlock";
3666
3667                         j = srcu_lockdep_next(__func__, fl, fs, fu, i, cyclelen, deadlock);
3668                         if (i == 0)
3669                                 rcu_read_lock_trace();
3670                         else
3671                                 idx = srcu_read_lock(srcus[i]);
3672                         if (j >= 0) {
3673                                 if (i == cyclelen - 1)
3674                                         synchronize_rcu_tasks_trace();
3675                                 else
3676                                         synchronize_srcu(srcus[j]);
3677                         }
3678                         if (i == 0)
3679                                 rcu_read_unlock_trace();
3680                         else
3681                                 srcu_read_unlock(srcus[i], idx);
3682                 }
3683                 return;
3684         }
3685 #endif // #ifdef CONFIG_TASKS_TRACE_RCU
3686
3687 err_out:
3688         pr_info("%s: test_srcu_lockdep = %05d does nothing.\n", __func__, test_srcu_lockdep);
3689         pr_info("%s: test_srcu_lockdep = DNNL.\n", __func__);
3690         pr_info("%s: D: Deadlock if nonzero.\n", __func__);
3691         pr_info("%s: NN: Test number, 0=SRCU, 1=SRCU/mutex, 2=SRCU/rwsem, 3=SRCU/Tasks Trace RCU.\n", __func__);
3692         pr_info("%s: L: Cycle length.\n", __func__);
3693         if (!IS_ENABLED(CONFIG_TASKS_TRACE_RCU))
3694                 pr_info("%s: NN=3 disallowed because kernel is built with CONFIG_TASKS_TRACE_RCU=n\n", __func__);
3695 }
3696
3697 static int __init
3698 rcu_torture_init(void)
3699 {
3700         long i;
3701         int cpu;
3702         int firsterr = 0;
3703         int flags = 0;
3704         unsigned long gp_seq = 0;
3705         static struct rcu_torture_ops *torture_ops[] = {
3706                 &rcu_ops, &rcu_busted_ops, &srcu_ops, &srcud_ops, &busted_srcud_ops,
3707                 TASKS_OPS TASKS_RUDE_OPS TASKS_TRACING_OPS
3708                 &trivial_ops,
3709         };
3710
3711         if (!torture_init_begin(torture_type, verbose))
3712                 return -EBUSY;
3713
3714         /* Process args and tell the world that the torturer is on the job. */
3715         for (i = 0; i < ARRAY_SIZE(torture_ops); i++) {
3716                 cur_ops = torture_ops[i];
3717                 if (strcmp(torture_type, cur_ops->name) == 0)
3718                         break;
3719         }
3720         if (i == ARRAY_SIZE(torture_ops)) {
3721                 pr_alert("rcu-torture: invalid torture type: \"%s\"\n",
3722                          torture_type);
3723                 pr_alert("rcu-torture types:");
3724                 for (i = 0; i < ARRAY_SIZE(torture_ops); i++)
3725                         pr_cont(" %s", torture_ops[i]->name);
3726                 pr_cont("\n");
3727                 firsterr = -EINVAL;
3728                 cur_ops = NULL;
3729                 goto unwind;
3730         }
3731         if (cur_ops->fqs == NULL && fqs_duration != 0) {
3732                 pr_alert("rcu-torture: ->fqs NULL and non-zero fqs_duration, fqs disabled.\n");
3733                 fqs_duration = 0;
3734         }
3735         if (nocbs_nthreads != 0 && (cur_ops != &rcu_ops ||
3736                                     !IS_ENABLED(CONFIG_RCU_NOCB_CPU))) {
3737                 pr_alert("rcu-torture types: %s and CONFIG_RCU_NOCB_CPU=%d, nocb toggle disabled.\n",
3738                          cur_ops->name, IS_ENABLED(CONFIG_RCU_NOCB_CPU));
3739                 nocbs_nthreads = 0;
3740         }
3741         if (cur_ops->init)
3742                 cur_ops->init();
3743
3744         rcu_torture_init_srcu_lockdep();
3745
3746         if (nreaders >= 0) {
3747                 nrealreaders = nreaders;
3748         } else {
3749                 nrealreaders = num_online_cpus() - 2 - nreaders;
3750                 if (nrealreaders <= 0)
3751                         nrealreaders = 1;
3752         }
3753         rcu_torture_print_module_parms(cur_ops, "Start of test");
3754         rcutorture_get_gp_data(cur_ops->ttype, &flags, &gp_seq);
3755         srcutorture_get_gp_data(cur_ops->ttype, srcu_ctlp, &flags, &gp_seq);
3756         start_gp_seq = gp_seq;
3757         pr_alert("%s:  Start-test grace-period state: g%ld f%#x\n",
3758                  cur_ops->name, (long)gp_seq, flags);
3759
3760         /* Set up the freelist. */
3761
3762         INIT_LIST_HEAD(&rcu_torture_freelist);
3763         for (i = 0; i < ARRAY_SIZE(rcu_tortures); i++) {
3764                 rcu_tortures[i].rtort_mbtest = 0;
3765                 list_add_tail(&rcu_tortures[i].rtort_free,
3766                               &rcu_torture_freelist);
3767         }
3768
3769         /* Initialize the statistics so that each run gets its own numbers. */
3770
3771         rcu_torture_current = NULL;
3772         rcu_torture_current_version = 0;
3773         atomic_set(&n_rcu_torture_alloc, 0);
3774         atomic_set(&n_rcu_torture_alloc_fail, 0);
3775         atomic_set(&n_rcu_torture_free, 0);
3776         atomic_set(&n_rcu_torture_mberror, 0);
3777         atomic_set(&n_rcu_torture_mbchk_fail, 0);
3778         atomic_set(&n_rcu_torture_mbchk_tries, 0);
3779         atomic_set(&n_rcu_torture_error, 0);
3780         n_rcu_torture_barrier_error = 0;
3781         n_rcu_torture_boost_ktrerror = 0;
3782         n_rcu_torture_boost_failure = 0;
3783         n_rcu_torture_boosts = 0;
3784         for (i = 0; i < RCU_TORTURE_PIPE_LEN + 1; i++)
3785                 atomic_set(&rcu_torture_wcount[i], 0);
3786         for_each_possible_cpu(cpu) {
3787                 for (i = 0; i < RCU_TORTURE_PIPE_LEN + 1; i++) {
3788                         per_cpu(rcu_torture_count, cpu)[i] = 0;
3789                         per_cpu(rcu_torture_batch, cpu)[i] = 0;
3790                 }
3791         }
3792         err_segs_recorded = 0;
3793         rt_read_nsegs = 0;
3794
3795         /* Start up the kthreads. */
3796
3797         rcu_torture_write_types();
3798         firsterr = torture_create_kthread(rcu_torture_writer, NULL,
3799                                           writer_task);
3800         if (torture_init_error(firsterr))
3801                 goto unwind;
3802         if (nfakewriters > 0) {
3803                 fakewriter_tasks = kcalloc(nfakewriters,
3804                                            sizeof(fakewriter_tasks[0]),
3805                                            GFP_KERNEL);
3806                 if (fakewriter_tasks == NULL) {
3807                         TOROUT_ERRSTRING("out of memory");
3808                         firsterr = -ENOMEM;
3809                         goto unwind;
3810                 }
3811         }
3812         for (i = 0; i < nfakewriters; i++) {
3813                 firsterr = torture_create_kthread(rcu_torture_fakewriter,
3814                                                   NULL, fakewriter_tasks[i]);
3815                 if (torture_init_error(firsterr))
3816                         goto unwind;
3817         }
3818         reader_tasks = kcalloc(nrealreaders, sizeof(reader_tasks[0]),
3819                                GFP_KERNEL);
3820         rcu_torture_reader_mbchk = kcalloc(nrealreaders, sizeof(*rcu_torture_reader_mbchk),
3821                                            GFP_KERNEL);
3822         if (!reader_tasks || !rcu_torture_reader_mbchk) {
3823                 TOROUT_ERRSTRING("out of memory");
3824                 firsterr = -ENOMEM;
3825                 goto unwind;
3826         }
3827         for (i = 0; i < nrealreaders; i++) {
3828                 rcu_torture_reader_mbchk[i].rtc_chkrdr = -1;
3829                 firsterr = torture_create_kthread(rcu_torture_reader, (void *)i,
3830                                                   reader_tasks[i]);
3831                 if (torture_init_error(firsterr))
3832                         goto unwind;
3833         }
3834         nrealnocbers = nocbs_nthreads;
3835         if (WARN_ON(nrealnocbers < 0))
3836                 nrealnocbers = 1;
3837         if (WARN_ON(nocbs_toggle < 0))
3838                 nocbs_toggle = HZ;
3839         if (nrealnocbers > 0) {
3840                 nocb_tasks = kcalloc(nrealnocbers, sizeof(nocb_tasks[0]), GFP_KERNEL);
3841                 if (nocb_tasks == NULL) {
3842                         TOROUT_ERRSTRING("out of memory");
3843                         firsterr = -ENOMEM;
3844                         goto unwind;
3845                 }
3846         } else {
3847                 nocb_tasks = NULL;
3848         }
3849         for (i = 0; i < nrealnocbers; i++) {
3850                 firsterr = torture_create_kthread(rcu_nocb_toggle, NULL, nocb_tasks[i]);
3851                 if (torture_init_error(firsterr))
3852                         goto unwind;
3853         }
3854         if (stat_interval > 0) {
3855                 firsterr = torture_create_kthread(rcu_torture_stats, NULL,
3856                                                   stats_task);
3857                 if (torture_init_error(firsterr))
3858                         goto unwind;
3859         }
3860         if (test_no_idle_hz && shuffle_interval > 0) {
3861                 firsterr = torture_shuffle_init(shuffle_interval * HZ);
3862                 if (torture_init_error(firsterr))
3863                         goto unwind;
3864         }
3865         if (stutter < 0)
3866                 stutter = 0;
3867         if (stutter) {
3868                 int t;
3869
3870                 t = cur_ops->stall_dur ? cur_ops->stall_dur() : stutter * HZ;
3871                 firsterr = torture_stutter_init(stutter * HZ, t);
3872                 if (torture_init_error(firsterr))
3873                         goto unwind;
3874         }
3875         if (fqs_duration < 0)
3876                 fqs_duration = 0;
3877         if (fqs_holdoff < 0)
3878                 fqs_holdoff = 0;
3879         if (fqs_duration && fqs_holdoff) {
3880                 /* Create the fqs thread */
3881                 firsterr = torture_create_kthread(rcu_torture_fqs, NULL,
3882                                                   fqs_task);
3883                 if (torture_init_error(firsterr))
3884                         goto unwind;
3885         }
3886         if (test_boost_interval < 1)
3887                 test_boost_interval = 1;
3888         if (test_boost_duration < 2)
3889                 test_boost_duration = 2;
3890         if (rcu_torture_can_boost()) {
3891
3892                 boost_starttime = jiffies + test_boost_interval * HZ;
3893
3894                 firsterr = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "RCU_TORTURE",
3895                                              rcutorture_booster_init,
3896                                              rcutorture_booster_cleanup);
3897                 rcutor_hp = firsterr;
3898                 if (torture_init_error(firsterr))
3899                         goto unwind;
3900         }
3901         shutdown_jiffies = jiffies + shutdown_secs * HZ;
3902         firsterr = torture_shutdown_init(shutdown_secs, rcu_torture_cleanup);
3903         if (torture_init_error(firsterr))
3904                 goto unwind;
3905         firsterr = torture_onoff_init(onoff_holdoff * HZ, onoff_interval,
3906                                       rcutorture_sync);
3907         if (torture_init_error(firsterr))
3908                 goto unwind;
3909         firsterr = rcu_torture_stall_init();
3910         if (torture_init_error(firsterr))
3911                 goto unwind;
3912         firsterr = rcu_torture_fwd_prog_init();
3913         if (torture_init_error(firsterr))
3914                 goto unwind;
3915         firsterr = rcu_torture_barrier_init();
3916         if (torture_init_error(firsterr))
3917                 goto unwind;
3918         firsterr = rcu_torture_read_exit_init();
3919         if (torture_init_error(firsterr))
3920                 goto unwind;
3921         if (object_debug)
3922                 rcu_test_debug_objects();
3923         torture_init_end();
3924         rcu_gp_slow_register(&rcu_fwd_cb_nodelay);
3925         return 0;
3926
3927 unwind:
3928         torture_init_end();
3929         rcu_torture_cleanup();
3930         if (shutdown_secs) {
3931                 WARN_ON(!IS_MODULE(CONFIG_RCU_TORTURE_TEST));
3932                 kernel_power_off();
3933         }
3934         return firsterr;
3935 }
3936
3937 module_init(rcu_torture_init);
3938 module_exit(rcu_torture_cleanup);