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/interrupt.h>
  25 #include <linux/sched/signal.h>
  26 #include <uapi/linux/sched/types.h>
  27 #include <linux/atomic.h>
  28 #include <linux/bitops.h>
  29 #include <linux/completion.h>
  30 #include <linux/moduleparam.h>
  31 #include <linux/percpu.h>
  32 #include <linux/notifier.h>
  33 #include <linux/reboot.h>
  34 #include <linux/freezer.h>
  35 #include <linux/cpu.h>
  36 #include <linux/delay.h>
  37 #include <linux/stat.h>
  38 #include <linux/srcu.h>
  39 #include <linux/slab.h>
  40 #include <linux/trace_clock.h>
  41 #include <asm/byteorder.h>
  42 #include <linux/torture.h>
  43 #include <linux/vmalloc.h>
  44 #include <linux/sched/debug.h>
  45 #include <linux/sched/sysctl.h>
  46 #include <linux/oom.h>
  47 #include <linux/tick.h>
  48 #include <linux/rcupdate_trace.h>
  49
  50 #include "rcu.h"
  51
  52 MODULE_LICENSE("GPL");
  53 MODULE_AUTHOR("Paul E. McKenney <paulmck@linux.ibm.com> and Josh Triplett <josh@joshtriplett.org>");
  54
  55 /* Bits for ->extendables field, extendables param, and related definitions. */
  56 #define RCUTORTURE_RDR_SHIFT     8      /* Put SRCU index in upper bits. */
  57 #define RCUTORTURE_RDR_MASK      ((1 << RCUTORTURE_RDR_SHIFT) - 1)
  58 #define RCUTORTURE_RDR_BH        0x01   /* Extend readers by disabling bh. */
  59 #define RCUTORTURE_RDR_IRQ       0x02   /*  ... disabling interrupts. */
  60 #define RCUTORTURE_RDR_PREEMPT   0x04   /*  ... disabling preemption. */
  61 #define RCUTORTURE_RDR_RBH       0x08   /*  ... rcu_read_lock_bh(). */
  62 #define RCUTORTURE_RDR_SCHED     0x10   /*  ... rcu_read_lock_sched(). */
  63 #define RCUTORTURE_RDR_RCU       0x20   /*  ... entering another RCU reader. */
  64 #define RCUTORTURE_RDR_NBITS     6      /* Number of bits defined above. */
  65 #define RCUTORTURE_MAX_EXTEND    \
  66         (RCUTORTURE_RDR_BH | RCUTORTURE_RDR_IRQ | RCUTORTURE_RDR_PREEMPT | \
  67          RCUTORTURE_RDR_RBH | RCUTORTURE_RDR_SCHED)
  68 #define RCUTORTURE_RDR_MAX_LOOPS 0x7    /* Maximum reader extensions. */
  69                                         /* Must be power of two minus one. */
  70 #define RCUTORTURE_RDR_MAX_SEGS (RCUTORTURE_RDR_MAX_LOOPS + 3)
  71
  72 torture_param(int, extendables, RCUTORTURE_MAX_EXTEND,
  73               "Extend readers by disabling bh (1), irqs (2), or preempt (4)");
  74 torture_param(int, fqs_duration, 0,
  75               "Duration of fqs bursts (us), 0 to disable");
  76 torture_param(int, fqs_holdoff, 0, "Holdoff time within fqs bursts (us)");
  77 torture_param(int, fqs_stutter, 3, "Wait time between fqs bursts (s)");
  78 torture_param(bool, fwd_progress, 1, "Test grace-period forward progress");
  79 torture_param(int, fwd_progress_div, 4, "Fraction of CPU stall to wait");
  80 torture_param(int, fwd_progress_holdoff, 60,
  81               "Time between forward-progress tests (s)");
  82 torture_param(bool, fwd_progress_need_resched, 1,
  83               "Hide cond_resched() behind need_resched()");
  84 torture_param(bool, gp_cond, false, "Use conditional/async GP wait primitives");
  85 torture_param(bool, gp_exp, false, "Use expedited GP wait primitives");
  86 torture_param(bool, gp_normal, false,
  87              "Use normal (non-expedited) GP wait primitives");
  88 torture_param(bool, gp_poll, false, "Use polling GP wait primitives");
  89 torture_param(bool, gp_sync, false, "Use synchronous GP wait primitives");
  90 torture_param(int, irqreader, 1, "Allow RCU readers from irq handlers");
  91 torture_param(int, leakpointer, 0, "Leak pointer dereferences from readers");
  92 torture_param(int, n_barrier_cbs, 0,
  93              "# of callbacks/kthreads for barrier testing");
  94 torture_param(int, nfakewriters, 4, "Number of RCU fake writer threads");
  95 torture_param(int, nreaders, -1, "Number of RCU reader threads");
  96 torture_param(int, object_debug, 0,
  97              "Enable debug-object double call_rcu() testing");
  98 torture_param(int, onoff_holdoff, 0, "Time after boot before CPU hotplugs (s)");
  99 torture_param(int, onoff_interval, 0,
 100              "Time between CPU hotplugs (jiffies), 0=disable");
 101 torture_param(int, nocbs_nthreads, 0, "Number of NOCB toggle threads, 0 to disable");
 102 torture_param(int, nocbs_toggle, 1000, "Time between toggling nocb state (ms)");
 103 torture_param(int, read_exit_delay, 13,
 104               "Delay between read-then-exit episodes (s)");
 105 torture_param(int, read_exit_burst, 16,
 106               "# of read-then-exit bursts per episode, zero to disable");
 107 torture_param(int, shuffle_interval, 3, "Number of seconds between shuffles");
 108 torture_param(int, shutdown_secs, 0, "Shutdown time (s), <= zero to disable.");
 109 torture_param(int, stall_cpu, 0, "Stall duration (s), zero to disable.");
 110 torture_param(int, stall_cpu_holdoff, 10,
 111              "Time to wait before starting stall (s).");
 112 torture_param(int, stall_cpu_irqsoff, 0, "Disable interrupts while stalling.");
 113 torture_param(int, stall_cpu_block, 0, "Sleep while stalling.");
 114 torture_param(int, stall_gp_kthread, 0,
 115               "Grace-period kthread stall duration (s).");
 116 torture_param(int, stat_interval, 60,
 117              "Number of seconds between stats printk()s");
 118 torture_param(int, stutter, 5, "Number of seconds to run/halt test");
 119 torture_param(int, test_boost, 1, "Test RCU prio boost: 0=no, 1=maybe, 2=yes.");
 120 torture_param(int, test_boost_duration, 4,
 121              "Duration of each boost test, seconds.");
 122 torture_param(int, test_boost_interval, 7,
 123              "Interval between boost tests, seconds.");
 124 torture_param(bool, test_no_idle_hz, true,
 125              "Test support for tickless idle CPUs");
 126 torture_param(int, verbose, 1,
 127              "Enable verbose debugging printk()s");
 128
 129 static char *torture_type = "rcu";
 130 module_param(torture_type, charp, 0444);
 131 MODULE_PARM_DESC(torture_type, "Type of RCU to torture (rcu, srcu, ...)");
 132
 133 static int nrealnocbers;
 134 static int nrealreaders;
 135 static struct task_struct *writer_task;
 136 static struct task_struct **fakewriter_tasks;
 137 static struct task_struct **reader_tasks;
 138 static struct task_struct **nocb_tasks;
 139 static struct task_struct *stats_task;
 140 static struct task_struct *fqs_task;
 141 static struct task_struct *boost_tasks[NR_CPUS];
 142 static struct task_struct *stall_task;
 143 static struct task_struct *fwd_prog_task;
 144 static struct task_struct **barrier_cbs_tasks;
 145 static struct task_struct *barrier_task;
 146 static struct task_struct *read_exit_task;
 147
 148 #define RCU_TORTURE_PIPE_LEN 10
 149
 150 // Mailbox-like structure to check RCU global memory ordering.
 151 struct rcu_torture_reader_check {
 152         unsigned long rtc_myloops;
 153         int rtc_chkrdr;
 154         unsigned long rtc_chkloops;
 155         int rtc_ready;
 156         struct rcu_torture_reader_check *rtc_assigner;
 157 } ____cacheline_internodealigned_in_smp;
 158
 159 // Update-side data structure used to check RCU readers.
 160 struct rcu_torture {
 161         struct rcu_head rtort_rcu;
 162         int rtort_pipe_count;
 163         struct list_head rtort_free;
 164         int rtort_mbtest;
 165         struct rcu_torture_reader_check *rtort_chkp;
 166 };
 167
 168 static LIST_HEAD(rcu_torture_freelist);
 169 static struct rcu_torture __rcu *rcu_torture_current;
 170 static unsigned long rcu_torture_current_version;
 171 static struct rcu_torture rcu_tortures[10 * RCU_TORTURE_PIPE_LEN];
 172 static DEFINE_SPINLOCK(rcu_torture_lock);
 173 static DEFINE_PER_CPU(long [RCU_TORTURE_PIPE_LEN + 1], rcu_torture_count);
 174 static DEFINE_PER_CPU(long [RCU_TORTURE_PIPE_LEN + 1], rcu_torture_batch);
 175 static atomic_t rcu_torture_wcount[RCU_TORTURE_PIPE_LEN + 1];
 176 static struct rcu_torture_reader_check *rcu_torture_reader_mbchk;
 177 static atomic_t n_rcu_torture_alloc;
 178 static atomic_t n_rcu_torture_alloc_fail;
 179 static atomic_t n_rcu_torture_free;
 180 static atomic_t n_rcu_torture_mberror;
 181 static atomic_t n_rcu_torture_mbchk_fail;
 182 static atomic_t n_rcu_torture_mbchk_tries;
 183 static atomic_t n_rcu_torture_error;
 184 static long n_rcu_torture_barrier_error;
 185 static long n_rcu_torture_boost_ktrerror;
 186 static long n_rcu_torture_boost_rterror;
 187 static long n_rcu_torture_boost_failure;
 188 static long n_rcu_torture_boosts;
 189 static atomic_long_t n_rcu_torture_timers;
 190 static long n_barrier_attempts;
 191 static long n_barrier_successes; /* did rcu_barrier test succeed? */
 192 static unsigned long n_read_exits;
 193 static struct list_head rcu_torture_removed;
 194 static unsigned long shutdown_jiffies;
 195 static unsigned long start_gp_seq;
 196 static atomic_long_t n_nocb_offload;
 197 static atomic_long_t n_nocb_deoffload;
 198
 199 static int rcu_torture_writer_state;
 200 #define RTWS_FIXED_DELAY        0
 201 #define RTWS_DELAY              1
 202 #define RTWS_REPLACE            2
 203 #define RTWS_DEF_FREE           3
 204 #define RTWS_EXP_SYNC           4
 205 #define RTWS_COND_GET           5
 206 #define RTWS_COND_SYNC          6
 207 #define RTWS_POLL_GET           7
 208 #define RTWS_POLL_WAIT          8
 209 #define RTWS_SYNC               9
 210 #define RTWS_STUTTER            10
 211 #define RTWS_STOPPING           11
 212 static const char * const rcu_torture_writer_state_names[] = {
 213         "RTWS_FIXED_DELAY",
 214         "RTWS_DELAY",
 215         "RTWS_REPLACE",
 216         "RTWS_DEF_FREE",
 217         "RTWS_EXP_SYNC",
 218         "RTWS_COND_GET",
 219         "RTWS_COND_SYNC",
 220         "RTWS_POLL_GET",
 221         "RTWS_POLL_WAIT",
 222         "RTWS_SYNC",
 223         "RTWS_STUTTER",
 224         "RTWS_STOPPING",
 225 };
 226
 227 /* Record reader segment types and duration for first failing read. */
 228 struct rt_read_seg {
 229         int rt_readstate;
 230         unsigned long rt_delay_jiffies;
 231         unsigned long rt_delay_ms;
 232         unsigned long rt_delay_us;
 233         bool rt_preempted;
 234 };
 235 static int err_segs_recorded;
 236 static struct rt_read_seg err_segs[RCUTORTURE_RDR_MAX_SEGS];
 237 static int rt_read_nsegs;
 238
 239 static const char *rcu_torture_writer_state_getname(void)
 240 {
 241         unsigned int i = READ_ONCE(rcu_torture_writer_state);
 242
 243         if (i >= ARRAY_SIZE(rcu_torture_writer_state_names))
 244                 return "???";
 245         return rcu_torture_writer_state_names[i];
 246 }
 247
 248 #if defined(CONFIG_RCU_BOOST) && !defined(CONFIG_HOTPLUG_CPU)
 249 #define rcu_can_boost() 1
 250 #else /* #if defined(CONFIG_RCU_BOOST) && !defined(CONFIG_HOTPLUG_CPU) */
 251 #define rcu_can_boost() 0
 252 #endif /* #else #if defined(CONFIG_RCU_BOOST) && !defined(CONFIG_HOTPLUG_CPU) */
 253
 254 #ifdef CONFIG_RCU_TRACE
 255 static u64 notrace rcu_trace_clock_local(void)
 256 {
 257         u64 ts = trace_clock_local();
 258
 259         (void)do_div(ts, NSEC_PER_USEC);
 260         return ts;
 261 }
 262 #else /* #ifdef CONFIG_RCU_TRACE */
 263 static u64 notrace rcu_trace_clock_local(void)
 264 {
 265         return 0ULL;
 266 }
 267 #endif /* #else #ifdef CONFIG_RCU_TRACE */
 268
 269 /*
 270  * Stop aggressive CPU-hog tests a bit before the end of the test in order
 271  * to avoid interfering with test shutdown.
 272  */
 273 static bool shutdown_time_arrived(void)
 274 {
 275         return shutdown_secs && time_after(jiffies, shutdown_jiffies - 30 * HZ);
 276 }
 277
 278 static unsigned long boost_starttime;   /* jiffies of next boost test start. */
 279 static DEFINE_MUTEX(boost_mutex);       /* protect setting boost_starttime */
 280                                         /*  and boost task create/destroy. */
 281 static atomic_t barrier_cbs_count;      /* Barrier callbacks registered. */
 282 static bool barrier_phase;              /* Test phase. */
 283 static atomic_t barrier_cbs_invoked;    /* Barrier callbacks invoked. */
 284 static wait_queue_head_t *barrier_cbs_wq; /* Coordinate barrier testing. */
 285 static DECLARE_WAIT_QUEUE_HEAD(barrier_wq);
 286
 287 static bool rcu_fwd_cb_nodelay;         /* Short rcu_torture_delay() delays. */
 288
 289 /*
 290  * Allocate an element from the rcu_tortures pool.
 291  */
 292 static struct rcu_torture *
 293 rcu_torture_alloc(void)
 294 {
 295         struct list_head *p;
 296
 297         spin_lock_bh(&rcu_torture_lock);
 298         if (list_empty(&rcu_torture_freelist)) {
 299                 atomic_inc(&n_rcu_torture_alloc_fail);
 300                 spin_unlock_bh(&rcu_torture_lock);
 301                 return NULL;
 302         }
 303         atomic_inc(&n_rcu_torture_alloc);
 304         p = rcu_torture_freelist.next;
 305         list_del_init(p);
 306         spin_unlock_bh(&rcu_torture_lock);
 307         return container_of(p, struct rcu_torture, rtort_free);
 308 }
 309
 310 /*
 311  * Free an element to the rcu_tortures pool.
 312  */
 313 static void
 314 rcu_torture_free(struct rcu_torture *p)
 315 {
 316         atomic_inc(&n_rcu_torture_free);
 317         spin_lock_bh(&rcu_torture_lock);
 318         list_add_tail(&p->rtort_free, &rcu_torture_freelist);
 319         spin_unlock_bh(&rcu_torture_lock);
 320 }
 321
 322 /*
 323  * Operations vector for selecting different types of tests.
 324  */
 325
 326 struct rcu_torture_ops {
 327         int ttype;
 328         void (*init)(void);
 329         void (*cleanup)(void);
 330         int (*readlock)(void);
 331         void (*read_delay)(struct torture_random_state *rrsp,
 332                            struct rt_read_seg *rtrsp);
 333         void (*readunlock)(int idx);
 334         unsigned long (*get_gp_seq)(void);
 335         unsigned long (*gp_diff)(unsigned long new, unsigned long old);
 336         void (*deferred_free)(struct rcu_torture *p);
 337         void (*sync)(void);
 338         void (*exp_sync)(void);
 339         unsigned long (*get_gp_state)(void);
 340         unsigned long (*start_gp_poll)(void);
 341         bool (*poll_gp_state)(unsigned long oldstate);
 342         void (*cond_sync)(unsigned long oldstate);
 343         call_rcu_func_t call;
 344         void (*cb_barrier)(void);
 345         void (*fqs)(void);
 346         void (*stats)(void);
 347         void (*gp_kthread_dbg)(void);
 348         int (*stall_dur)(void);
 349         int irq_capable;
 350         int can_boost;
 351         int extendables;
 352         int slow_gps;
 353         const char *name;
 354 };
 355
 356 static struct rcu_torture_ops *cur_ops;
 357
 358 /*
 359  * Definitions for rcu torture testing.
 360  */
 361
 362 static int rcu_torture_read_lock(void) __acquires(RCU)
 363 {
 364         rcu_read_lock();
 365         return 0;
 366 }
 367
 368 static void
 369 rcu_read_delay(struct torture_random_state *rrsp, struct rt_read_seg *rtrsp)
 370 {
 371         unsigned long started;
 372         unsigned long completed;
 373         const unsigned long shortdelay_us = 200;
 374         unsigned long longdelay_ms = 300;
 375         unsigned long long ts;
 376
 377         /* We want a short delay sometimes to make a reader delay the grace
 378          * period, and we want a long delay occasionally to trigger
 379          * force_quiescent_state. */
 380
 381         if (!READ_ONCE(rcu_fwd_cb_nodelay) &&
 382             !(torture_random(rrsp) % (nrealreaders * 2000 * longdelay_ms))) {
 383                 started = cur_ops->get_gp_seq();
 384                 ts = rcu_trace_clock_local();
 385                 if (preempt_count() & (SOFTIRQ_MASK | HARDIRQ_MASK))
 386                         longdelay_ms = 5; /* Avoid triggering BH limits. */
 387                 mdelay(longdelay_ms);
 388                 rtrsp->rt_delay_ms = longdelay_ms;
 389                 completed = cur_ops->get_gp_seq();
 390                 do_trace_rcu_torture_read(cur_ops->name, NULL, ts,
 391                                           started, completed);
 392         }
 393         if (!(torture_random(rrsp) % (nrealreaders * 2 * shortdelay_us))) {
 394                 udelay(shortdelay_us);
 395                 rtrsp->rt_delay_us = shortdelay_us;
 396         }
 397         if (!preempt_count() &&
 398             !(torture_random(rrsp) % (nrealreaders * 500))) {
 399                 torture_preempt_schedule();  /* QS only if preemptible. */
 400                 rtrsp->rt_preempted = true;
 401         }
 402 }
 403
 404 static void rcu_torture_read_unlock(int idx) __releases(RCU)
 405 {
 406         rcu_read_unlock();
 407 }
 408
 409 /*
 410  * Update callback in the pipe.  This should be invoked after a grace period.
 411  */
 412 static bool
 413 rcu_torture_pipe_update_one(struct rcu_torture *rp)
 414 {
 415         int i;
 416         struct rcu_torture_reader_check *rtrcp = READ_ONCE(rp->rtort_chkp);
 417
 418         if (rtrcp) {
 419                 WRITE_ONCE(rp->rtort_chkp, NULL);
 420                 smp_store_release(&rtrcp->rtc_ready, 1); // Pair with smp_load_acquire().
 421         }
 422         i = READ_ONCE(rp->rtort_pipe_count);
 423         if (i > RCU_TORTURE_PIPE_LEN)
 424                 i = RCU_TORTURE_PIPE_LEN;
 425         atomic_inc(&rcu_torture_wcount[i]);
 426         WRITE_ONCE(rp->rtort_pipe_count, i + 1);
 427         if (rp->rtort_pipe_count >= RCU_TORTURE_PIPE_LEN) {
 428                 rp->rtort_mbtest = 0;
 429                 return true;
 430         }
 431         return false;
 432 }
 433
 434 /*
 435  * Update all callbacks in the pipe.  Suitable for synchronous grace-period
 436  * primitives.
 437  */
 438 static void
 439 rcu_torture_pipe_update(struct rcu_torture *old_rp)
 440 {
 441         struct rcu_torture *rp;
 442         struct rcu_torture *rp1;
 443
 444         if (old_rp)
 445                 list_add(&old_rp->rtort_free, &rcu_torture_removed);
 446         list_for_each_entry_safe(rp, rp1, &rcu_torture_removed, rtort_free) {
 447                 if (rcu_torture_pipe_update_one(rp)) {
 448                         list_del(&rp->rtort_free);
 449                         rcu_torture_free(rp);
 450                 }
 451         }
 452 }
 453
 454 static void
 455 rcu_torture_cb(struct rcu_head *p)
 456 {
 457         struct rcu_torture *rp = container_of(p, struct rcu_torture, rtort_rcu);
 458
 459         if (torture_must_stop_irq()) {
 460                 /* Test is ending, just drop callbacks on the floor. */
 461                 /* The next initialization will pick up the pieces. */
 462                 return;
 463         }
 464         if (rcu_torture_pipe_update_one(rp))
 465                 rcu_torture_free(rp);
 466         else
 467                 cur_ops->deferred_free(rp);
 468 }
 469
 470 static unsigned long rcu_no_completed(void)
 471 {
 472         return 0;
 473 }
 474
 475 static void rcu_torture_deferred_free(struct rcu_torture *p)
 476 {
 477         call_rcu(&p->rtort_rcu, rcu_torture_cb);
 478 }
 479
 480 static void rcu_sync_torture_init(void)
 481 {
 482         INIT_LIST_HEAD(&rcu_torture_removed);
 483 }
 484
 485 static struct rcu_torture_ops rcu_ops = {
 486         .ttype          = RCU_FLAVOR,
 487         .init           = rcu_sync_torture_init,
 488         .readlock       = rcu_torture_read_lock,
 489         .read_delay     = rcu_read_delay,
 490         .readunlock     = rcu_torture_read_unlock,
 491         .get_gp_seq     = rcu_get_gp_seq,
 492         .gp_diff        = rcu_seq_diff,
 493         .deferred_free  = rcu_torture_deferred_free,
 494         .sync           = synchronize_rcu,
 495         .exp_sync       = synchronize_rcu_expedited,
 496         .get_gp_state   = get_state_synchronize_rcu,
 497         .cond_sync      = cond_synchronize_rcu,
 498         .call           = call_rcu,
 499         .cb_barrier     = rcu_barrier,
 500         .fqs            = rcu_force_quiescent_state,
 501         .stats          = NULL,
 502         .gp_kthread_dbg = show_rcu_gp_kthreads,
 503         .stall_dur      = rcu_jiffies_till_stall_check,
 504         .irq_capable    = 1,
 505         .can_boost      = rcu_can_boost(),
 506         .extendables    = RCUTORTURE_MAX_EXTEND,
 507         .name           = "rcu"
 508 };
 509
 510 /*
 511  * Don't even think about trying any of these in real life!!!
 512  * The names includes "busted", and they really means it!
 513  * The only purpose of these functions is to provide a buggy RCU
 514  * implementation to make sure that rcutorture correctly emits
 515  * buggy-RCU error messages.
 516  */
 517 static void rcu_busted_torture_deferred_free(struct rcu_torture *p)
 518 {
 519         /* This is a deliberate bug for testing purposes only! */
 520         rcu_torture_cb(&p->rtort_rcu);
 521 }
 522
 523 static void synchronize_rcu_busted(void)
 524 {
 525         /* This is a deliberate bug for testing purposes only! */
 526 }
 527
 528 static void
 529 call_rcu_busted(struct rcu_head *head, rcu_callback_t func)
 530 {
 531         /* This is a deliberate bug for testing purposes only! */
 532         func(head);
 533 }
 534
 535 static struct rcu_torture_ops rcu_busted_ops = {
 536         .ttype          = INVALID_RCU_FLAVOR,
 537         .init           = rcu_sync_torture_init,
 538         .readlock       = rcu_torture_read_lock,
 539         .read_delay     = rcu_read_delay,  /* just reuse rcu's version. */
 540         .readunlock     = rcu_torture_read_unlock,
 541         .get_gp_seq     = rcu_no_completed,
 542         .deferred_free  = rcu_busted_torture_deferred_free,
 543         .sync           = synchronize_rcu_busted,
 544         .exp_sync       = synchronize_rcu_busted,
 545         .call           = call_rcu_busted,
 546         .cb_barrier     = NULL,
 547         .fqs            = NULL,
 548         .stats          = NULL,
 549         .irq_capable    = 1,
 550         .name           = "busted"
 551 };
 552
 553 /*
 554  * Definitions for srcu torture testing.
 555  */
 556
 557 DEFINE_STATIC_SRCU(srcu_ctl);
 558 static struct srcu_struct srcu_ctld;
 559 static struct srcu_struct *srcu_ctlp = &srcu_ctl;
 560
 561 static int srcu_torture_read_lock(void) __acquires(srcu_ctlp)
 562 {
 563         return srcu_read_lock(srcu_ctlp);
 564 }
 565
 566 static void
 567 srcu_read_delay(struct torture_random_state *rrsp, struct rt_read_seg *rtrsp)
 568 {
 569         long delay;
 570         const long uspertick = 1000000 / HZ;
 571         const long longdelay = 10;
 572
 573         /* We want there to be long-running readers, but not all the time. */
 574
 575         delay = torture_random(rrsp) %
 576                 (nrealreaders * 2 * longdelay * uspertick);
 577         if (!delay && in_task()) {
 578                 schedule_timeout_interruptible(longdelay);
 579                 rtrsp->rt_delay_jiffies = longdelay;
 580         } else {
 581                 rcu_read_delay(rrsp, rtrsp);
 582         }
 583 }
 584
 585 static void srcu_torture_read_unlock(int idx) __releases(srcu_ctlp)
 586 {
 587         srcu_read_unlock(srcu_ctlp, idx);
 588 }
 589
 590 static unsigned long srcu_torture_completed(void)
 591 {
 592         return srcu_batches_completed(srcu_ctlp);
 593 }
 594
 595 static void srcu_torture_deferred_free(struct rcu_torture *rp)
 596 {
 597         call_srcu(srcu_ctlp, &rp->rtort_rcu, rcu_torture_cb);
 598 }
 599
 600 static void srcu_torture_synchronize(void)
 601 {
 602         synchronize_srcu(srcu_ctlp);
 603 }
 604
 605 static unsigned long srcu_torture_get_gp_state(void)
 606 {
 607         return get_state_synchronize_srcu(srcu_ctlp);
 608 }
 609
 610 static unsigned long srcu_torture_start_gp_poll(void)
 611 {
 612         return start_poll_synchronize_srcu(srcu_ctlp);
 613 }
 614
 615 static bool srcu_torture_poll_gp_state(unsigned long oldstate)
 616 {
 617         return poll_state_synchronize_srcu(srcu_ctlp, oldstate);
 618 }
 619
 620 static void srcu_torture_call(struct rcu_head *head,
 621                               rcu_callback_t func)
 622 {
 623         call_srcu(srcu_ctlp, head, func);
 624 }
 625
 626 static void srcu_torture_barrier(void)
 627 {
 628         srcu_barrier(srcu_ctlp);
 629 }
 630
 631 static void srcu_torture_stats(void)
 632 {
 633         srcu_torture_stats_print(srcu_ctlp, torture_type, TORTURE_FLAG);
 634 }
 635
 636 static void srcu_torture_synchronize_expedited(void)
 637 {
 638         synchronize_srcu_expedited(srcu_ctlp);
 639 }
 640
 641 static struct rcu_torture_ops srcu_ops = {
 642         .ttype          = SRCU_FLAVOR,
 643         .init           = rcu_sync_torture_init,
 644         .readlock       = srcu_torture_read_lock,
 645         .read_delay     = srcu_read_delay,
 646         .readunlock     = srcu_torture_read_unlock,
 647         .get_gp_seq     = srcu_torture_completed,
 648         .deferred_free  = srcu_torture_deferred_free,
 649         .sync           = srcu_torture_synchronize,
 650         .exp_sync       = srcu_torture_synchronize_expedited,
 651         .get_gp_state   = srcu_torture_get_gp_state,
 652         .start_gp_poll  = srcu_torture_start_gp_poll,
 653         .poll_gp_state  = srcu_torture_poll_gp_state,
 654         .call           = srcu_torture_call,
 655         .cb_barrier     = srcu_torture_barrier,
 656         .stats          = srcu_torture_stats,
 657         .irq_capable    = 1,
 658         .name           = "srcu"
 659 };
 660
 661 static void srcu_torture_init(void)
 662 {
 663         rcu_sync_torture_init();
 664         WARN_ON(init_srcu_struct(&srcu_ctld));
 665         srcu_ctlp = &srcu_ctld;
 666 }
 667
 668 static void srcu_torture_cleanup(void)
 669 {
 670         cleanup_srcu_struct(&srcu_ctld);
 671         srcu_ctlp = &srcu_ctl; /* In case of a later rcutorture run. */
 672 }
 673
 674 /* As above, but dynamically allocated. */
 675 static struct rcu_torture_ops srcud_ops = {
 676         .ttype          = SRCU_FLAVOR,
 677         .init           = srcu_torture_init,
 678         .cleanup        = srcu_torture_cleanup,
 679         .readlock       = srcu_torture_read_lock,
 680         .read_delay     = srcu_read_delay,
 681         .readunlock     = srcu_torture_read_unlock,
 682         .get_gp_seq     = srcu_torture_completed,
 683         .deferred_free  = srcu_torture_deferred_free,
 684         .sync           = srcu_torture_synchronize,
 685         .exp_sync       = srcu_torture_synchronize_expedited,
 686         .call           = srcu_torture_call,
 687         .cb_barrier     = srcu_torture_barrier,
 688         .stats          = srcu_torture_stats,
 689         .irq_capable    = 1,
 690         .name           = "srcud"
 691 };
 692
 693 /* As above, but broken due to inappropriate reader extension. */
 694 static struct rcu_torture_ops busted_srcud_ops = {
 695         .ttype          = SRCU_FLAVOR,
 696         .init           = srcu_torture_init,
 697         .cleanup        = srcu_torture_cleanup,
 698         .readlock       = srcu_torture_read_lock,
 699         .read_delay     = rcu_read_delay,
 700         .readunlock     = srcu_torture_read_unlock,
 701         .get_gp_seq     = srcu_torture_completed,
 702         .deferred_free  = srcu_torture_deferred_free,
 703         .sync           = srcu_torture_synchronize,
 704         .exp_sync       = srcu_torture_synchronize_expedited,
 705         .call           = srcu_torture_call,
 706         .cb_barrier     = srcu_torture_barrier,
 707         .stats          = srcu_torture_stats,
 708         .irq_capable    = 1,
 709         .extendables    = RCUTORTURE_MAX_EXTEND,
 710         .name           = "busted_srcud"
 711 };
 712
 713 /*
 714  * Definitions for RCU-tasks torture testing.
 715  */
 716
 717 static int tasks_torture_read_lock(void)
 718 {
 719         return 0;
 720 }
 721
 722 static void tasks_torture_read_unlock(int idx)
 723 {
 724 }
 725
 726 static void rcu_tasks_torture_deferred_free(struct rcu_torture *p)
 727 {
 728         call_rcu_tasks(&p->rtort_rcu, rcu_torture_cb);
 729 }
 730
 731 static void synchronize_rcu_mult_test(void)
 732 {
 733         synchronize_rcu_mult(call_rcu_tasks, call_rcu);
 734 }
 735
 736 static struct rcu_torture_ops tasks_ops = {
 737         .ttype          = RCU_TASKS_FLAVOR,
 738         .init           = rcu_sync_torture_init,
 739         .readlock       = tasks_torture_read_lock,
 740         .read_delay     = rcu_read_delay,  /* just reuse rcu's version. */
 741         .readunlock     = tasks_torture_read_unlock,
 742         .get_gp_seq     = rcu_no_completed,
 743         .deferred_free  = rcu_tasks_torture_deferred_free,
 744         .sync           = synchronize_rcu_tasks,
 745         .exp_sync       = synchronize_rcu_mult_test,
 746         .call           = call_rcu_tasks,
 747         .cb_barrier     = rcu_barrier_tasks,
 748         .gp_kthread_dbg = show_rcu_tasks_classic_gp_kthread,
 749         .fqs            = NULL,
 750         .stats          = NULL,
 751         .irq_capable    = 1,
 752         .slow_gps       = 1,
 753         .name           = "tasks"
 754 };
 755
 756 /*
 757  * Definitions for trivial CONFIG_PREEMPT=n-only torture testing.
 758  * This implementation does not necessarily work well with CPU hotplug.
 759  */
 760
 761 static void synchronize_rcu_trivial(void)
 762 {
 763         int cpu;
 764
 765         for_each_online_cpu(cpu) {
 766                 rcutorture_sched_setaffinity(current->pid, cpumask_of(cpu));
 767                 WARN_ON_ONCE(raw_smp_processor_id() != cpu);
 768         }
 769 }
 770
 771 static int rcu_torture_read_lock_trivial(void) __acquires(RCU)
 772 {
 773         preempt_disable();
 774         return 0;
 775 }
 776
 777 static void rcu_torture_read_unlock_trivial(int idx) __releases(RCU)
 778 {
 779         preempt_enable();
 780 }
 781
 782 static struct rcu_torture_ops trivial_ops = {
 783         .ttype          = RCU_TRIVIAL_FLAVOR,
 784         .init           = rcu_sync_torture_init,
 785         .readlock       = rcu_torture_read_lock_trivial,
 786         .read_delay     = rcu_read_delay,  /* just reuse rcu's version. */
 787         .readunlock     = rcu_torture_read_unlock_trivial,
 788         .get_gp_seq     = rcu_no_completed,
 789         .sync           = synchronize_rcu_trivial,
 790         .exp_sync       = synchronize_rcu_trivial,
 791         .fqs            = NULL,
 792         .stats          = NULL,
 793         .irq_capable    = 1,
 794         .name           = "trivial"
 795 };
 796
 797 /*
 798  * Definitions for rude RCU-tasks torture testing.
 799  */
 800
 801 static void rcu_tasks_rude_torture_deferred_free(struct rcu_torture *p)
 802 {
 803         call_rcu_tasks_rude(&p->rtort_rcu, rcu_torture_cb);
 804 }
 805
 806 static struct rcu_torture_ops tasks_rude_ops = {
 807         .ttype          = RCU_TASKS_RUDE_FLAVOR,
 808         .init           = rcu_sync_torture_init,
 809         .readlock       = rcu_torture_read_lock_trivial,
 810         .read_delay     = rcu_read_delay,  /* just reuse rcu's version. */
 811         .readunlock     = rcu_torture_read_unlock_trivial,
 812         .get_gp_seq     = rcu_no_completed,
 813         .deferred_free  = rcu_tasks_rude_torture_deferred_free,
 814         .sync           = synchronize_rcu_tasks_rude,
 815         .exp_sync       = synchronize_rcu_tasks_rude,
 816         .call           = call_rcu_tasks_rude,
 817         .cb_barrier     = rcu_barrier_tasks_rude,
 818         .gp_kthread_dbg = show_rcu_tasks_rude_gp_kthread,
 819         .fqs            = NULL,
 820         .stats          = NULL,
 821         .irq_capable    = 1,
 822         .name           = "tasks-rude"
 823 };
 824
 825 /*
 826  * Definitions for tracing RCU-tasks torture testing.
 827  */
 828
 829 static int tasks_tracing_torture_read_lock(void)
 830 {
 831         rcu_read_lock_trace();
 832         return 0;
 833 }
 834
 835 static void tasks_tracing_torture_read_unlock(int idx)
 836 {
 837         rcu_read_unlock_trace();
 838 }
 839
 840 static void rcu_tasks_tracing_torture_deferred_free(struct rcu_torture *p)
 841 {
 842         call_rcu_tasks_trace(&p->rtort_rcu, rcu_torture_cb);
 843 }
 844
 845 static struct rcu_torture_ops tasks_tracing_ops = {
 846         .ttype          = RCU_TASKS_TRACING_FLAVOR,
 847         .init           = rcu_sync_torture_init,
 848         .readlock       = tasks_tracing_torture_read_lock,
 849         .read_delay     = srcu_read_delay,  /* just reuse srcu's version. */
 850         .readunlock     = tasks_tracing_torture_read_unlock,
 851         .get_gp_seq     = rcu_no_completed,
 852         .deferred_free  = rcu_tasks_tracing_torture_deferred_free,
 853         .sync           = synchronize_rcu_tasks_trace,
 854         .exp_sync       = synchronize_rcu_tasks_trace,
 855         .call           = call_rcu_tasks_trace,
 856         .cb_barrier     = rcu_barrier_tasks_trace,
 857         .gp_kthread_dbg = show_rcu_tasks_trace_gp_kthread,
 858         .fqs            = NULL,
 859         .stats          = NULL,
 860         .irq_capable    = 1,
 861         .slow_gps       = 1,
 862         .name           = "tasks-tracing"
 863 };
 864
 865 static unsigned long rcutorture_seq_diff(unsigned long new, unsigned long old)
 866 {
 867         if (!cur_ops->gp_diff)
 868                 return new - old;
 869         return cur_ops->gp_diff(new, old);
 870 }
 871
 872 static bool __maybe_unused torturing_tasks(void)
 873 {
 874         return cur_ops == &tasks_ops || cur_ops == &tasks_rude_ops;
 875 }
 876
 877 /*
 878  * RCU torture priority-boost testing.  Runs one real-time thread per
 879  * CPU for moderate bursts, repeatedly registering RCU callbacks and
 880  * spinning waiting for them to be invoked.  If a given callback takes
 881  * too long to be invoked, we assume that priority inversion has occurred.
 882  */
 883
 884 struct rcu_boost_inflight {
 885         struct rcu_head rcu;
 886         int inflight;
 887 };
 888
 889 static void rcu_torture_boost_cb(struct rcu_head *head)
 890 {
 891         struct rcu_boost_inflight *rbip =
 892                 container_of(head, struct rcu_boost_inflight, rcu);
 893
 894         /* Ensure RCU-core accesses precede clearing ->inflight */
 895         smp_store_release(&rbip->inflight, 0);
 896 }
 897
 898 static int old_rt_runtime = -1;
 899
 900 static void rcu_torture_disable_rt_throttle(void)
 901 {
 902         /*
 903          * Disable RT throttling so that rcutorture's boost threads don't get
 904          * throttled. Only possible if rcutorture is built-in otherwise the
 905          * user should manually do this by setting the sched_rt_period_us and
 906          * sched_rt_runtime sysctls.
 907          */
 908         if (!IS_BUILTIN(CONFIG_RCU_TORTURE_TEST) || old_rt_runtime != -1)
 909                 return;
 910
 911         old_rt_runtime = sysctl_sched_rt_runtime;
 912         sysctl_sched_rt_runtime = -1;
 913 }
 914
 915 static void rcu_torture_enable_rt_throttle(void)
 916 {
 917         if (!IS_BUILTIN(CONFIG_RCU_TORTURE_TEST) || old_rt_runtime == -1)
 918                 return;
 919
 920         sysctl_sched_rt_runtime = old_rt_runtime;
 921         old_rt_runtime = -1;
 922 }
 923
 924 static bool rcu_torture_boost_failed(unsigned long start, unsigned long end)
 925 {
 926         if (end - start > test_boost_duration * HZ - HZ / 2) {
 927                 VERBOSE_TOROUT_STRING("rcu_torture_boost boosting failed");
 928                 n_rcu_torture_boost_failure++;
 929
 930                 return true; /* failed */
 931         }
 932
 933         return false; /* passed */
 934 }
 935
 936 static int rcu_torture_boost(void *arg)
 937 {
 938         unsigned long call_rcu_time;
 939         unsigned long endtime;
 940         unsigned long oldstarttime;
 941         struct rcu_boost_inflight rbi = { .inflight = 0 };
 942
 943         VERBOSE_TOROUT_STRING("rcu_torture_boost started");
 944
 945         /* Set real-time priority. */
 946         sched_set_fifo_low(current);
 947
 948         init_rcu_head_on_stack(&rbi.rcu);
 949         /* Each pass through the following loop does one boost-test cycle. */
 950         do {
 951                 /* Track if the test failed already in this test interval? */
 952                 bool failed = false;
 953
 954                 /* Increment n_rcu_torture_boosts once per boost-test */
 955                 while (!kthread_should_stop()) {
 956                         if (mutex_trylock(&boost_mutex)) {
 957                                 n_rcu_torture_boosts++;
 958                                 mutex_unlock(&boost_mutex);
 959                                 break;
 960                         }
 961                         schedule_timeout_uninterruptible(1);
 962                 }
 963                 if (kthread_should_stop())
 964                         goto checkwait;
 965
 966                 /* Wait for the next test interval. */
 967                 oldstarttime = boost_starttime;
 968                 while (time_before(jiffies, oldstarttime)) {
 969                         schedule_timeout_interruptible(oldstarttime - jiffies);
 970                         if (stutter_wait("rcu_torture_boost"))
 971                                 sched_set_fifo_low(current);
 972                         if (torture_must_stop())
 973                                 goto checkwait;
 974                 }
 975
 976                 /* Do one boost-test interval. */
 977                 endtime = oldstarttime + test_boost_duration * HZ;
 978                 call_rcu_time = jiffies;
 979                 while (time_before(jiffies, endtime)) {
 980                         /* If we don't have a callback in flight, post one. */
 981                         if (!smp_load_acquire(&rbi.inflight)) {
 982                                 /* RCU core before ->inflight = 1. */
 983                                 smp_store_release(&rbi.inflight, 1);
 984                                 call_rcu(&rbi.rcu, rcu_torture_boost_cb);
 985                                 /* Check if the boost test failed */
 986                                 failed = failed ||
 987                                          rcu_torture_boost_failed(call_rcu_time,
 988                                                                  jiffies);
 989                                 call_rcu_time = jiffies;
 990                         }
 991                         if (stutter_wait("rcu_torture_boost"))
 992                                 sched_set_fifo_low(current);
 993                         if (torture_must_stop())
 994                                 goto checkwait;
 995                 }
 996
 997                 /*
 998                  * If boost never happened, then inflight will always be 1, in
 999                  * this case the boost check would never happen in the above
1000                  * loop so do another one here.
1001                  */
1002                 if (!failed && smp_load_acquire(&rbi.inflight))
1003                         rcu_torture_boost_failed(call_rcu_time, jiffies);
1004
1005                 /*
1006                  * Set the start time of the next test interval.
1007                  * Yes, this is vulnerable to long delays, but such
1008                  * delays simply cause a false negative for the next
1009                  * interval.  Besides, we are running at RT priority,
1010                  * so delays should be relatively rare.
1011                  */
1012                 while (oldstarttime == boost_starttime &&
1013                        !kthread_should_stop()) {
1014                         if (mutex_trylock(&boost_mutex)) {
1015                                 boost_starttime = jiffies +
1016                                                   test_boost_interval * HZ;
1017                                 mutex_unlock(&boost_mutex);
1018                                 break;
1019                         }
1020                         schedule_timeout_uninterruptible(1);
1021                 }
1022
1023                 /* Go do the stutter. */
1024 checkwait:      if (stutter_wait("rcu_torture_boost"))
1025                         sched_set_fifo_low(current);
1026         } while (!torture_must_stop());
1027
1028         /* Clean up and exit. */
1029         while (!kthread_should_stop() || smp_load_acquire(&rbi.inflight)) {
1030                 torture_shutdown_absorb("rcu_torture_boost");
1031                 schedule_timeout_uninterruptible(1);
1032         }
1033         destroy_rcu_head_on_stack(&rbi.rcu);
1034         torture_kthread_stopping("rcu_torture_boost");
1035         return 0;
1036 }
1037
1038 /*
1039  * RCU torture force-quiescent-state kthread.  Repeatedly induces
1040  * bursts of calls to force_quiescent_state(), increasing the probability
1041  * of occurrence of some important types of race conditions.
1042  */
1043 static int
1044 rcu_torture_fqs(void *arg)
1045 {
1046         unsigned long fqs_resume_time;
1047         int fqs_burst_remaining;
1048         int oldnice = task_nice(current);
1049
1050         VERBOSE_TOROUT_STRING("rcu_torture_fqs task started");
1051         do {
1052                 fqs_resume_time = jiffies + fqs_stutter * HZ;
1053                 while (time_before(jiffies, fqs_resume_time) &&
1054                        !kthread_should_stop()) {
1055                         schedule_timeout_interruptible(1);
1056                 }
1057                 fqs_burst_remaining = fqs_duration;
1058                 while (fqs_burst_remaining > 0 &&
1059                        !kthread_should_stop()) {
1060                         cur_ops->fqs();
1061                         udelay(fqs_holdoff);
1062                         fqs_burst_remaining -= fqs_holdoff;
1063                 }
1064                 if (stutter_wait("rcu_torture_fqs"))
1065                         sched_set_normal(current, oldnice);
1066         } while (!torture_must_stop());
1067         torture_kthread_stopping("rcu_torture_fqs");
1068         return 0;
1069 }
1070
1071 // Used by writers to randomly choose from the available grace-period
1072 // primitives.  The only purpose of the initialization is to size the array.
1073 static int synctype[] = { RTWS_DEF_FREE, RTWS_EXP_SYNC, RTWS_COND_GET, RTWS_POLL_GET, RTWS_SYNC };
1074 static int nsynctypes;
1075
1076 /*
1077  * Determine which grace-period primitives are available.
1078  */
1079 static void rcu_torture_write_types(void)
1080 {
1081         bool gp_cond1 = gp_cond, gp_exp1 = gp_exp, gp_normal1 = gp_normal;
1082         bool gp_poll1 = gp_poll, gp_sync1 = gp_sync;
1083
1084         /* Initialize synctype[] array.  If none set, take default. */
1085         if (!gp_cond1 && !gp_exp1 && !gp_normal1 && !gp_poll1 && !gp_sync1)
1086                 gp_cond1 = gp_exp1 = gp_normal1 = gp_poll1 = gp_sync1 = true;
1087         if (gp_cond1 && cur_ops->get_gp_state && cur_ops->cond_sync) {
1088                 synctype[nsynctypes++] = RTWS_COND_GET;
1089                 pr_info("%s: Testing conditional GPs.\n", __func__);
1090         } else if (gp_cond && (!cur_ops->get_gp_state || !cur_ops->cond_sync)) {
1091                 pr_alert("%s: gp_cond without primitives.\n", __func__);
1092         }
1093         if (gp_exp1 && cur_ops->exp_sync) {
1094                 synctype[nsynctypes++] = RTWS_EXP_SYNC;
1095                 pr_info("%s: Testing expedited GPs.\n", __func__);
1096         } else if (gp_exp && !cur_ops->exp_sync) {
1097                 pr_alert("%s: gp_exp without primitives.\n", __func__);
1098         }
1099         if (gp_normal1 && cur_ops->deferred_free) {
1100                 synctype[nsynctypes++] = RTWS_DEF_FREE;
1101                 pr_info("%s: Testing asynchronous GPs.\n", __func__);
1102         } else if (gp_normal && !cur_ops->deferred_free) {
1103                 pr_alert("%s: gp_normal without primitives.\n", __func__);
1104         }
1105         if (gp_poll1 && cur_ops->start_gp_poll && cur_ops->poll_gp_state) {
1106                 synctype[nsynctypes++] = RTWS_POLL_GET;
1107                 pr_info("%s: Testing polling GPs.\n", __func__);
1108         } else if (gp_poll && (!cur_ops->start_gp_poll || !cur_ops->poll_gp_state)) {
1109                 pr_alert("%s: gp_poll without primitives.\n", __func__);
1110         }
1111         if (gp_sync1 && cur_ops->sync) {
1112                 synctype[nsynctypes++] = RTWS_SYNC;
1113                 pr_info("%s: Testing normal GPs.\n", __func__);
1114         } else if (gp_sync && !cur_ops->sync) {
1115                 pr_alert("%s: gp_sync without primitives.\n", __func__);
1116         }
1117 }
1118
1119 /*
1120  * RCU torture writer kthread.  Repeatedly substitutes a new structure
1121  * for that pointed to by rcu_torture_current, freeing the old structure
1122  * after a series of grace periods (the "pipeline").
1123  */
1124 static int
1125 rcu_torture_writer(void *arg)
1126 {
1127         bool boot_ended;
1128         bool can_expedite = !rcu_gp_is_expedited() && !rcu_gp_is_normal();
1129         unsigned long cookie;
1130         int expediting = 0;
1131         unsigned long gp_snap;
1132         int i;
1133         int idx;
1134         int oldnice = task_nice(current);
1135         struct rcu_torture *rp;
1136         struct rcu_torture *old_rp;
1137         static DEFINE_TORTURE_RANDOM(rand);
1138         bool stutter_waited;
1139
1140         VERBOSE_TOROUT_STRING("rcu_torture_writer task started");
1141         if (!can_expedite)
1142                 pr_alert("%s" TORTURE_FLAG
1143                          " GP expediting controlled from boot/sysfs for %s.\n",
1144                          torture_type, cur_ops->name);
1145         if (WARN_ONCE(nsynctypes == 0,
1146                       "rcu_torture_writer: No update-side primitives.\n")) {
1147                 /*
1148                  * No updates primitives, so don't try updating.
1149                  * The resulting test won't be testing much, hence the
1150                  * above WARN_ONCE().
1151                  */
1152                 rcu_torture_writer_state = RTWS_STOPPING;
1153                 torture_kthread_stopping("rcu_torture_writer");
1154         }
1155
1156         do {
1157                 rcu_torture_writer_state = RTWS_FIXED_DELAY;
1158                 torture_hrtimeout_us(500, 1000, &rand);
1159                 rp = rcu_torture_alloc();
1160                 if (rp == NULL)
1161                         continue;
1162                 rp->rtort_pipe_count = 0;
1163                 rcu_torture_writer_state = RTWS_DELAY;
1164                 udelay(torture_random(&rand) & 0x3ff);
1165                 rcu_torture_writer_state = RTWS_REPLACE;
1166                 old_rp = rcu_dereference_check(rcu_torture_current,
1167                                                current == writer_task);
1168                 rp->rtort_mbtest = 1;
1169                 rcu_assign_pointer(rcu_torture_current, rp);
1170                 smp_wmb(); /* Mods to old_rp must follow rcu_assign_pointer() */
1171                 if (old_rp) {
1172                         i = old_rp->rtort_pipe_count;
1173                         if (i > RCU_TORTURE_PIPE_LEN)
1174                                 i = RCU_TORTURE_PIPE_LEN;
1175                         atomic_inc(&rcu_torture_wcount[i]);
1176                         WRITE_ONCE(old_rp->rtort_pipe_count,
1177                                    old_rp->rtort_pipe_count + 1);
1178                         if (cur_ops->get_gp_state && cur_ops->poll_gp_state) {
1179                                 idx = cur_ops->readlock();
1180                                 cookie = cur_ops->get_gp_state();
1181                                 WARN_ONCE(rcu_torture_writer_state != RTWS_DEF_FREE &&
1182                                           cur_ops->poll_gp_state(cookie),
1183                                           "%s: Cookie check 1 failed %s(%d) %lu->%lu\n",
1184                                           __func__,
1185                                           rcu_torture_writer_state_getname(),
1186                                           rcu_torture_writer_state,
1187                                           cookie, cur_ops->get_gp_state());
1188                                 cur_ops->readunlock(idx);
1189                         }
1190                         switch (synctype[torture_random(&rand) % nsynctypes]) {
1191                         case RTWS_DEF_FREE:
1192                                 rcu_torture_writer_state = RTWS_DEF_FREE;
1193                                 cur_ops->deferred_free(old_rp);
1194                                 break;
1195                         case RTWS_EXP_SYNC:
1196                                 rcu_torture_writer_state = RTWS_EXP_SYNC;
1197                                 cur_ops->exp_sync();
1198                                 rcu_torture_pipe_update(old_rp);
1199                                 break;
1200                         case RTWS_COND_GET:
1201                                 rcu_torture_writer_state = RTWS_COND_GET;
1202                                 gp_snap = cur_ops->get_gp_state();
1203                                 torture_hrtimeout_jiffies(torture_random(&rand) % 16, &rand);
1204                                 rcu_torture_writer_state = RTWS_COND_SYNC;
1205                                 cur_ops->cond_sync(gp_snap);
1206                                 rcu_torture_pipe_update(old_rp);
1207                                 break;
1208                         case RTWS_POLL_GET:
1209                                 rcu_torture_writer_state = RTWS_POLL_GET;
1210                                 gp_snap = cur_ops->start_gp_poll();
1211                                 rcu_torture_writer_state = RTWS_POLL_WAIT;
1212                                 while (!cur_ops->poll_gp_state(gp_snap))
1213                                         torture_hrtimeout_jiffies(torture_random(&rand) % 16,
1214                                                                   &rand);
1215                                 rcu_torture_pipe_update(old_rp);
1216                                 break;
1217                         case RTWS_SYNC:
1218                                 rcu_torture_writer_state = RTWS_SYNC;
1219                                 cur_ops->sync();
1220                                 rcu_torture_pipe_update(old_rp);
1221                                 break;
1222                         default:
1223                                 WARN_ON_ONCE(1);
1224                                 break;
1225                         }
1226                         if (cur_ops->get_gp_state && cur_ops->poll_gp_state)
1227                                 WARN_ONCE(rcu_torture_writer_state != RTWS_DEF_FREE &&
1228                                           !cur_ops->poll_gp_state(cookie),
1229                                           "%s: Cookie check 2 failed %s(%d) %lu->%lu\n",
1230                                           __func__,
1231                                           rcu_torture_writer_state_getname(),
1232                                           rcu_torture_writer_state,
1233                                           cookie, cur_ops->get_gp_state());
1234                 }
1235                 WRITE_ONCE(rcu_torture_current_version,
1236                            rcu_torture_current_version + 1);
1237                 /* Cycle through nesting levels of rcu_expedite_gp() calls. */
1238                 if (can_expedite &&
1239                     !(torture_random(&rand) & 0xff & (!!expediting - 1))) {
1240                         WARN_ON_ONCE(expediting == 0 && rcu_gp_is_expedited());
1241                         if (expediting >= 0)
1242                                 rcu_expedite_gp();
1243                         else
1244                                 rcu_unexpedite_gp();
1245                         if (++expediting > 3)
1246                                 expediting = -expediting;
1247                 } else if (!can_expedite) { /* Disabled during boot, recheck. */
1248                         can_expedite = !rcu_gp_is_expedited() &&
1249                                        !rcu_gp_is_normal();
1250                 }
1251                 rcu_torture_writer_state = RTWS_STUTTER;
1252                 boot_ended = rcu_inkernel_boot_has_ended();
1253                 stutter_waited = stutter_wait("rcu_torture_writer");
1254                 if (stutter_waited &&
1255                     !READ_ONCE(rcu_fwd_cb_nodelay) &&
1256                     !cur_ops->slow_gps &&
1257                     !torture_must_stop() &&
1258                     boot_ended)
1259                         for (i = 0; i < ARRAY_SIZE(rcu_tortures); i++)
1260                                 if (list_empty(&rcu_tortures[i].rtort_free) &&
1261                                     rcu_access_pointer(rcu_torture_current) !=
1262                                     &rcu_tortures[i]) {
1263                                         rcu_ftrace_dump(DUMP_ALL);
1264                                         WARN(1, "%s: rtort_pipe_count: %d\n", __func__, rcu_tortures[i].rtort_pipe_count);
1265                                 }
1266                 if (stutter_waited)
1267                         sched_set_normal(current, oldnice);
1268         } while (!torture_must_stop());
1269         rcu_torture_current = NULL;  // Let stats task know that we are done.
1270         /* Reset expediting back to unexpedited. */
1271         if (expediting > 0)
1272                 expediting = -expediting;
1273         while (can_expedite && expediting++ < 0)
1274                 rcu_unexpedite_gp();
1275         WARN_ON_ONCE(can_expedite && rcu_gp_is_expedited());
1276         if (!can_expedite)
1277                 pr_alert("%s" TORTURE_FLAG
1278                          " Dynamic grace-period expediting was disabled.\n",
1279                          torture_type);
1280         rcu_torture_writer_state = RTWS_STOPPING;
1281         torture_kthread_stopping("rcu_torture_writer");
1282         return 0;
1283 }
1284
1285 /*
1286  * RCU torture fake writer kthread.  Repeatedly calls sync, with a random
1287  * delay between calls.
1288  */
1289 static int
1290 rcu_torture_fakewriter(void *arg)
1291 {
1292         unsigned long gp_snap;
1293         DEFINE_TORTURE_RANDOM(rand);
1294
1295         VERBOSE_TOROUT_STRING("rcu_torture_fakewriter task started");
1296         set_user_nice(current, MAX_NICE);
1297
1298         do {
1299                 torture_hrtimeout_jiffies(torture_random(&rand) % 10, &rand);
1300                 if (cur_ops->cb_barrier != NULL &&
1301                     torture_random(&rand) % (nfakewriters * 8) == 0) {
1302                         cur_ops->cb_barrier();
1303                 } else {
1304                         switch (synctype[torture_random(&rand) % nsynctypes]) {
1305                         case RTWS_DEF_FREE:
1306                                 break;
1307                         case RTWS_EXP_SYNC:
1308                                 cur_ops->exp_sync();
1309                                 break;
1310                         case RTWS_COND_GET:
1311                                 gp_snap = cur_ops->get_gp_state();
1312                                 torture_hrtimeout_jiffies(torture_random(&rand) % 16, &rand);
1313                                 cur_ops->cond_sync(gp_snap);
1314                                 break;
1315                         case RTWS_POLL_GET:
1316                                 gp_snap = cur_ops->start_gp_poll();
1317                                 while (!cur_ops->poll_gp_state(gp_snap)) {
1318                                         torture_hrtimeout_jiffies(torture_random(&rand) % 16,
1319                                                                   &rand);
1320                                 }
1321                                 break;
1322                         case RTWS_SYNC:
1323                                 cur_ops->sync();
1324                                 break;
1325                         default:
1326                                 WARN_ON_ONCE(1);
1327                                 break;
1328                         }
1329                 }
1330                 stutter_wait("rcu_torture_fakewriter");
1331         } while (!torture_must_stop());
1332
1333         torture_kthread_stopping("rcu_torture_fakewriter");
1334         return 0;
1335 }
1336
1337 static void rcu_torture_timer_cb(struct rcu_head *rhp)
1338 {
1339         kfree(rhp);
1340 }
1341
1342 // Set up and carry out testing of RCU's global memory ordering
1343 static void rcu_torture_reader_do_mbchk(long myid, struct rcu_torture *rtp,
1344                                         struct torture_random_state *trsp)
1345 {
1346         unsigned long loops;
1347         int noc = torture_num_online_cpus();
1348         int rdrchked;
1349         int rdrchker;
1350         struct rcu_torture_reader_check *rtrcp; // Me.
1351         struct rcu_torture_reader_check *rtrcp_assigner; // Assigned us to do checking.
1352         struct rcu_torture_reader_check *rtrcp_chked; // Reader being checked.
1353         struct rcu_torture_reader_check *rtrcp_chker; // Reader doing checking when not me.
1354
1355         if (myid < 0)
1356                 return; // Don't try this from timer handlers.
1357
1358         // Increment my counter.
1359         rtrcp = &rcu_torture_reader_mbchk[myid];
1360         WRITE_ONCE(rtrcp->rtc_myloops, rtrcp->rtc_myloops + 1);
1361
1362         // Attempt to assign someone else some checking work.
1363         rdrchked = torture_random(trsp) % nrealreaders;
1364         rtrcp_chked = &rcu_torture_reader_mbchk[rdrchked];
1365         rdrchker = torture_random(trsp) % nrealreaders;
1366         rtrcp_chker = &rcu_torture_reader_mbchk[rdrchker];
1367         if (rdrchked != myid && rdrchked != rdrchker && noc >= rdrchked && noc >= rdrchker &&
1368             smp_load_acquire(&rtrcp->rtc_chkrdr) < 0 && // Pairs with smp_store_release below.
1369             !READ_ONCE(rtp->rtort_chkp) &&
1370             !smp_load_acquire(&rtrcp_chker->rtc_assigner)) { // Pairs with smp_store_release below.
1371                 rtrcp->rtc_chkloops = READ_ONCE(rtrcp_chked->rtc_myloops);
1372                 WARN_ON_ONCE(rtrcp->rtc_chkrdr >= 0);
1373                 rtrcp->rtc_chkrdr = rdrchked;
1374                 WARN_ON_ONCE(rtrcp->rtc_ready); // This gets set after the grace period ends.
1375                 if (cmpxchg_relaxed(&rtrcp_chker->rtc_assigner, NULL, rtrcp) ||
1376                     cmpxchg_relaxed(&rtp->rtort_chkp, NULL, rtrcp))
1377                         (void)cmpxchg_relaxed(&rtrcp_chker->rtc_assigner, rtrcp, NULL); // Back out.
1378         }
1379
1380         // If assigned some completed work, do it!
1381         rtrcp_assigner = READ_ONCE(rtrcp->rtc_assigner);
1382         if (!rtrcp_assigner || !smp_load_acquire(&rtrcp_assigner->rtc_ready))
1383                 return; // No work or work not yet ready.
1384         rdrchked = rtrcp_assigner->rtc_chkrdr;
1385         if (WARN_ON_ONCE(rdrchked < 0))
1386                 return;
1387         rtrcp_chked = &rcu_torture_reader_mbchk[rdrchked];
1388         loops = READ_ONCE(rtrcp_chked->rtc_myloops);
1389         atomic_inc(&n_rcu_torture_mbchk_tries);
1390         if (ULONG_CMP_LT(loops, rtrcp_assigner->rtc_chkloops))
1391                 atomic_inc(&n_rcu_torture_mbchk_fail);
1392         rtrcp_assigner->rtc_chkloops = loops + ULONG_MAX / 2;
1393         rtrcp_assigner->rtc_ready = 0;
1394         smp_store_release(&rtrcp->rtc_assigner, NULL); // Someone else can assign us work.
1395         smp_store_release(&rtrcp_assigner->rtc_chkrdr, -1); // Assigner can again assign.
1396 }
1397
1398 /*
1399  * Do one extension of an RCU read-side critical section using the
1400  * current reader state in readstate (set to zero for initial entry
1401  * to extended critical section), set the new state as specified by
1402  * newstate (set to zero for final exit from extended critical section),
1403  * and random-number-generator state in trsp.  If this is neither the
1404  * beginning or end of the critical section and if there was actually a
1405  * change, do a ->read_delay().
1406  */
1407 static void rcutorture_one_extend(int *readstate, int newstate,
1408                                   struct torture_random_state *trsp,
1409                                   struct rt_read_seg *rtrsp)
1410 {
1411         unsigned long flags;
1412         int idxnew = -1;
1413         int idxold = *readstate;
1414         int statesnew = ~*readstate & newstate;
1415         int statesold = *readstate & ~newstate;
1416
1417         WARN_ON_ONCE(idxold < 0);
1418         WARN_ON_ONCE((idxold >> RCUTORTURE_RDR_SHIFT) > 1);
1419         rtrsp->rt_readstate = newstate;
1420
1421         /* First, put new protection in place to avoid critical-section gap. */
1422         if (statesnew & RCUTORTURE_RDR_BH)
1423                 local_bh_disable();
1424         if (statesnew & RCUTORTURE_RDR_IRQ)
1425                 local_irq_disable();
1426         if (statesnew & RCUTORTURE_RDR_PREEMPT)
1427                 preempt_disable();
1428         if (statesnew & RCUTORTURE_RDR_RBH)
1429                 rcu_read_lock_bh();
1430         if (statesnew & RCUTORTURE_RDR_SCHED)
1431                 rcu_read_lock_sched();
1432         if (statesnew & RCUTORTURE_RDR_RCU)
1433                 idxnew = cur_ops->readlock() << RCUTORTURE_RDR_SHIFT;
1434
1435         /* Next, remove old protection, irq first due to bh conflict. */
1436         if (statesold & RCUTORTURE_RDR_IRQ)
1437                 local_irq_enable();
1438         if (statesold & RCUTORTURE_RDR_BH)
1439                 local_bh_enable();
1440         if (statesold & RCUTORTURE_RDR_PREEMPT)
1441                 preempt_enable();
1442         if (statesold & RCUTORTURE_RDR_RBH)
1443                 rcu_read_unlock_bh();
1444         if (statesold & RCUTORTURE_RDR_SCHED)
1445                 rcu_read_unlock_sched();
1446         if (statesold & RCUTORTURE_RDR_RCU) {
1447                 bool lockit = !statesnew && !(torture_random(trsp) & 0xffff);
1448
1449                 if (lockit)
1450                         raw_spin_lock_irqsave(&current->pi_lock, flags);
1451                 cur_ops->readunlock(idxold >> RCUTORTURE_RDR_SHIFT);
1452                 if (lockit)
1453                         raw_spin_unlock_irqrestore(&current->pi_lock, flags);
1454         }
1455
1456         /* Delay if neither beginning nor end and there was a change. */
1457         if ((statesnew || statesold) && *readstate && newstate)
1458                 cur_ops->read_delay(trsp, rtrsp);
1459
1460         /* Update the reader state. */
1461         if (idxnew == -1)
1462                 idxnew = idxold & ~RCUTORTURE_RDR_MASK;
1463         WARN_ON_ONCE(idxnew < 0);
1464         WARN_ON_ONCE((idxnew >> RCUTORTURE_RDR_SHIFT) > 1);
1465         *readstate = idxnew | newstate;
1466         WARN_ON_ONCE((*readstate >> RCUTORTURE_RDR_SHIFT) < 0);
1467         WARN_ON_ONCE((*readstate >> RCUTORTURE_RDR_SHIFT) > 1);
1468 }
1469
1470 /* Return the biggest extendables mask given current RCU and boot parameters. */
1471 static int rcutorture_extend_mask_max(void)
1472 {
1473         int mask;
1474
1475         WARN_ON_ONCE(extendables & ~RCUTORTURE_MAX_EXTEND);
1476         mask = extendables & RCUTORTURE_MAX_EXTEND & cur_ops->extendables;
1477         mask = mask | RCUTORTURE_RDR_RCU;
1478         return mask;
1479 }
1480
1481 /* Return a random protection state mask, but with at least one bit set. */
1482 static int
1483 rcutorture_extend_mask(int oldmask, struct torture_random_state *trsp)
1484 {
1485         int mask = rcutorture_extend_mask_max();
1486         unsigned long randmask1 = torture_random(trsp) >> 8;
1487         unsigned long randmask2 = randmask1 >> 3;
1488
1489         WARN_ON_ONCE(mask >> RCUTORTURE_RDR_SHIFT);
1490         /* Mostly only one bit (need preemption!), sometimes lots of bits. */
1491         if (!(randmask1 & 0x7))
1492                 mask = mask & randmask2;
1493         else
1494                 mask = mask & (1 << (randmask2 % RCUTORTURE_RDR_NBITS));
1495         /* Can't enable bh w/irq disabled. */
1496         if ((mask & RCUTORTURE_RDR_IRQ) &&
1497             ((!(mask & RCUTORTURE_RDR_BH) && (oldmask & RCUTORTURE_RDR_BH)) ||
1498              (!(mask & RCUTORTURE_RDR_RBH) && (oldmask & RCUTORTURE_RDR_RBH))))
1499                 mask |= RCUTORTURE_RDR_BH | RCUTORTURE_RDR_RBH;
1500         return mask ?: RCUTORTURE_RDR_RCU;
1501 }
1502
1503 /*
1504  * Do a randomly selected number of extensions of an existing RCU read-side
1505  * critical section.
1506  */
1507 static struct rt_read_seg *
1508 rcutorture_loop_extend(int *readstate, struct torture_random_state *trsp,
1509                        struct rt_read_seg *rtrsp)
1510 {
1511         int i;
1512         int j;
1513         int mask = rcutorture_extend_mask_max();
1514
1515         WARN_ON_ONCE(!*readstate); /* -Existing- RCU read-side critsect! */
1516         if (!((mask - 1) & mask))
1517                 return rtrsp;  /* Current RCU reader not extendable. */
1518         /* Bias towards larger numbers of loops. */
1519         i = (torture_random(trsp) >> 3);
1520         i = ((i | (i >> 3)) & RCUTORTURE_RDR_MAX_LOOPS) + 1;
1521         for (j = 0; j < i; j++) {
1522                 mask = rcutorture_extend_mask(*readstate, trsp);
1523                 rcutorture_one_extend(readstate, mask, trsp, &rtrsp[j]);
1524         }
1525         return &rtrsp[j];
1526 }
1527
1528 /*
1529  * Do one read-side critical section, returning false if there was
1530  * no data to read.  Can be invoked both from process context and
1531  * from a timer handler.
1532  */
1533 static bool rcu_torture_one_read(struct torture_random_state *trsp, long myid)
1534 {
1535         unsigned long cookie;
1536         int i;
1537         unsigned long started;
1538         unsigned long completed;
1539         int newstate;
1540         struct rcu_torture *p;
1541         int pipe_count;
1542         int readstate = 0;
1543         struct rt_read_seg rtseg[RCUTORTURE_RDR_MAX_SEGS] = { { 0 } };
1544         struct rt_read_seg *rtrsp = &rtseg[0];
1545         struct rt_read_seg *rtrsp1;
1546         unsigned long long ts;
1547
1548         WARN_ON_ONCE(!rcu_is_watching());
1549         newstate = rcutorture_extend_mask(readstate, trsp);
1550         rcutorture_one_extend(&readstate, newstate, trsp, rtrsp++);
1551         if (cur_ops->get_gp_state && cur_ops->poll_gp_state)
1552                 cookie = cur_ops->get_gp_state();
1553         started = cur_ops->get_gp_seq();
1554         ts = rcu_trace_clock_local();
1555         p = rcu_dereference_check(rcu_torture_current,
1556                                   rcu_read_lock_bh_held() ||
1557                                   rcu_read_lock_sched_held() ||
1558                                   srcu_read_lock_held(srcu_ctlp) ||
1559                                   rcu_read_lock_trace_held() ||
1560                                   torturing_tasks());
1561         if (p == NULL) {
1562                 /* Wait for rcu_torture_writer to get underway */
1563                 rcutorture_one_extend(&readstate, 0, trsp, rtrsp);
1564                 return false;
1565         }
1566         if (p->rtort_mbtest == 0)
1567                 atomic_inc(&n_rcu_torture_mberror);
1568         rcu_torture_reader_do_mbchk(myid, p, trsp);
1569         rtrsp = rcutorture_loop_extend(&readstate, trsp, rtrsp);
1570         preempt_disable();
1571         pipe_count = READ_ONCE(p->rtort_pipe_count);
1572         if (pipe_count > RCU_TORTURE_PIPE_LEN) {
1573                 /* Should not happen, but... */
1574                 pipe_count = RCU_TORTURE_PIPE_LEN;
1575         }
1576         completed = cur_ops->get_gp_seq();
1577         if (pipe_count > 1) {
1578                 do_trace_rcu_torture_read(cur_ops->name, &p->rtort_rcu,
1579                                           ts, started, completed);
1580                 rcu_ftrace_dump(DUMP_ALL);
1581         }
1582         __this_cpu_inc(rcu_torture_count[pipe_count]);
1583         completed = rcutorture_seq_diff(completed, started);
1584         if (completed > RCU_TORTURE_PIPE_LEN) {
1585                 /* Should not happen, but... */
1586                 completed = RCU_TORTURE_PIPE_LEN;
1587         }
1588         __this_cpu_inc(rcu_torture_batch[completed]);
1589         preempt_enable();
1590         if (cur_ops->get_gp_state && cur_ops->poll_gp_state)
1591                 WARN_ONCE(cur_ops->poll_gp_state(cookie),
1592                           "%s: Cookie check 3 failed %s(%d) %lu->%lu\n",
1593                           __func__,
1594                           rcu_torture_writer_state_getname(),
1595                           rcu_torture_writer_state,
1596                           cookie, cur_ops->get_gp_state());
1597         rcutorture_one_extend(&readstate, 0, trsp, rtrsp);
1598         WARN_ON_ONCE(readstate & RCUTORTURE_RDR_MASK);
1599         // This next splat is expected behavior if leakpointer, especially
1600         // for CONFIG_RCU_STRICT_GRACE_PERIOD=y kernels.
1601         WARN_ON_ONCE(leakpointer && READ_ONCE(p->rtort_pipe_count) > 1);
1602
1603         /* If error or close call, record the sequence of reader protections. */
1604         if ((pipe_count > 1 || completed > 1) && !xchg(&err_segs_recorded, 1)) {
1605                 i = 0;
1606                 for (rtrsp1 = &rtseg[0]; rtrsp1 < rtrsp; rtrsp1++)
1607                         err_segs[i++] = *rtrsp1;
1608                 rt_read_nsegs = i;
1609         }
1610
1611         return true;
1612 }
1613
1614 static DEFINE_TORTURE_RANDOM_PERCPU(rcu_torture_timer_rand);
1615
1616 /*
1617  * RCU torture reader from timer handler.  Dereferences rcu_torture_current,
1618  * incrementing the corresponding element of the pipeline array.  The
1619  * counter in the element should never be greater than 1, otherwise, the
1620  * RCU implementation is broken.
1621  */
1622 static void rcu_torture_timer(struct timer_list *unused)
1623 {
1624         atomic_long_inc(&n_rcu_torture_timers);
1625         (void)rcu_torture_one_read(this_cpu_ptr(&rcu_torture_timer_rand), -1);
1626
1627         /* Test call_rcu() invocation from interrupt handler. */
1628         if (cur_ops->call) {
1629                 struct rcu_head *rhp = kmalloc(sizeof(*rhp), GFP_NOWAIT);
1630
1631                 if (rhp)
1632                         cur_ops->call(rhp, rcu_torture_timer_cb);
1633         }
1634 }
1635
1636 /*
1637  * RCU torture reader kthread.  Repeatedly dereferences rcu_torture_current,
1638  * incrementing the corresponding element of the pipeline array.  The
1639  * counter in the element should never be greater than 1, otherwise, the
1640  * RCU implementation is broken.
1641  */
1642 static int
1643 rcu_torture_reader(void *arg)
1644 {
1645         unsigned long lastsleep = jiffies;
1646         long myid = (long)arg;
1647         int mynumonline = myid;
1648         DEFINE_TORTURE_RANDOM(rand);
1649         struct timer_list t;
1650
1651         VERBOSE_TOROUT_STRING("rcu_torture_reader task started");
1652         set_user_nice(current, MAX_NICE);
1653         if (irqreader && cur_ops->irq_capable)
1654                 timer_setup_on_stack(&t, rcu_torture_timer, 0);
1655         tick_dep_set_task(current, TICK_DEP_BIT_RCU);
1656         do {
1657                 if (irqreader && cur_ops->irq_capable) {
1658                         if (!timer_pending(&t))
1659                                 mod_timer(&t, jiffies + 1);
1660                 }
1661                 if (!rcu_torture_one_read(&rand, myid) && !torture_must_stop())
1662                         schedule_timeout_interruptible(HZ);
1663                 if (time_after(jiffies, lastsleep) && !torture_must_stop()) {
1664                         torture_hrtimeout_us(500, 1000, &rand);
1665                         lastsleep = jiffies + 10;
1666                 }
1667                 while (torture_num_online_cpus() < mynumonline && !torture_must_stop())
1668                         schedule_timeout_interruptible(HZ / 5);
1669                 stutter_wait("rcu_torture_reader");
1670         } while (!torture_must_stop());
1671         if (irqreader && cur_ops->irq_capable) {
1672                 del_timer_sync(&t);
1673                 destroy_timer_on_stack(&t);
1674         }
1675         tick_dep_clear_task(current, TICK_DEP_BIT_RCU);
1676         torture_kthread_stopping("rcu_torture_reader");
1677         return 0;
1678 }
1679
1680 /*
1681  * Randomly Toggle CPUs' callback-offload state.  This uses hrtimers to
1682  * increase race probabilities and fuzzes the interval between toggling.
1683  */
1684 static int rcu_nocb_toggle(void *arg)
1685 {
1686         int cpu;
1687         int maxcpu = -1;
1688         int oldnice = task_nice(current);
1689         long r;
1690         DEFINE_TORTURE_RANDOM(rand);
1691         ktime_t toggle_delay;
1692         unsigned long toggle_fuzz;
1693         ktime_t toggle_interval = ms_to_ktime(nocbs_toggle);
1694
1695         VERBOSE_TOROUT_STRING("rcu_nocb_toggle task started");
1696         while (!rcu_inkernel_boot_has_ended())
1697                 schedule_timeout_interruptible(HZ / 10);
1698         for_each_online_cpu(cpu)
1699                 maxcpu = cpu;
1700         WARN_ON(maxcpu < 0);
1701         if (toggle_interval > ULONG_MAX)
1702                 toggle_fuzz = ULONG_MAX >> 3;
1703         else
1704                 toggle_fuzz = toggle_interval >> 3;
1705         if (toggle_fuzz <= 0)
1706                 toggle_fuzz = NSEC_PER_USEC;
1707         do {
1708                 r = torture_random(&rand);
1709                 cpu = (r >> 4) % (maxcpu + 1);
1710                 if (r & 0x1) {
1711                         rcu_nocb_cpu_offload(cpu);
1712                         atomic_long_inc(&n_nocb_offload);
1713                 } else {
1714                         rcu_nocb_cpu_deoffload(cpu);
1715                         atomic_long_inc(&n_nocb_deoffload);
1716                 }
1717                 toggle_delay = torture_random(&rand) % toggle_fuzz + toggle_interval;
1718                 set_current_state(TASK_INTERRUPTIBLE);
1719                 schedule_hrtimeout(&toggle_delay, HRTIMER_MODE_REL);
1720                 if (stutter_wait("rcu_nocb_toggle"))
1721                         sched_set_normal(current, oldnice);
1722         } while (!torture_must_stop());
1723         torture_kthread_stopping("rcu_nocb_toggle");
1724         return 0;
1725 }
1726
1727 /*
1728  * Print torture statistics.  Caller must ensure that there is only
1729  * one call to this function at a given time!!!  This is normally
1730  * accomplished by relying on the module system to only have one copy
1731  * of the module loaded, and then by giving the rcu_torture_stats
1732  * kthread full control (or the init/cleanup functions when rcu_torture_stats
1733  * thread is not running).
1734  */
1735 static void
1736 rcu_torture_stats_print(void)
1737 {
1738         int cpu;
1739         int i;
1740         long pipesummary[RCU_TORTURE_PIPE_LEN + 1] = { 0 };
1741         long batchsummary[RCU_TORTURE_PIPE_LEN + 1] = { 0 };
1742         struct rcu_torture *rtcp;
1743         static unsigned long rtcv_snap = ULONG_MAX;
1744         static bool splatted;
1745         struct task_struct *wtp;
1746
1747         for_each_possible_cpu(cpu) {
1748                 for (i = 0; i < RCU_TORTURE_PIPE_LEN + 1; i++) {
1749                         pipesummary[i] += READ_ONCE(per_cpu(rcu_torture_count, cpu)[i]);
1750                         batchsummary[i] += READ_ONCE(per_cpu(rcu_torture_batch, cpu)[i]);
1751                 }
1752         }
1753         for (i = RCU_TORTURE_PIPE_LEN - 1; i >= 0; i--) {
1754                 if (pipesummary[i] != 0)
1755                         break;
1756         }
1757
1758         pr_alert("%s%s ", torture_type, TORTURE_FLAG);
1759         rtcp = rcu_access_pointer(rcu_torture_current);
1760         pr_cont("rtc: %p %s: %lu tfle: %d rta: %d rtaf: %d rtf: %d ",
1761                 rtcp,
1762                 rtcp && !rcu_stall_is_suppressed_at_boot() ? "ver" : "VER",
1763                 rcu_torture_current_version,
1764                 list_empty(&rcu_torture_freelist),
1765                 atomic_read(&n_rcu_torture_alloc),
1766                 atomic_read(&n_rcu_torture_alloc_fail),
1767                 atomic_read(&n_rcu_torture_free));
1768         pr_cont("rtmbe: %d rtmbkf: %d/%d rtbe: %ld rtbke: %ld rtbre: %ld ",
1769                 atomic_read(&n_rcu_torture_mberror),
1770                 atomic_read(&n_rcu_torture_mbchk_fail), atomic_read(&n_rcu_torture_mbchk_tries),
1771                 n_rcu_torture_barrier_error,
1772                 n_rcu_torture_boost_ktrerror,
1773                 n_rcu_torture_boost_rterror);
1774         pr_cont("rtbf: %ld rtb: %ld nt: %ld ",
1775                 n_rcu_torture_boost_failure,
1776                 n_rcu_torture_boosts,
1777                 atomic_long_read(&n_rcu_torture_timers));
1778         torture_onoff_stats();
1779         pr_cont("barrier: %ld/%ld:%ld ",
1780                 data_race(n_barrier_successes),
1781                 data_race(n_barrier_attempts),
1782                 data_race(n_rcu_torture_barrier_error));
1783         pr_cont("read-exits: %ld ", data_race(n_read_exits)); // Statistic.
1784         pr_cont("nocb-toggles: %ld:%ld\n",
1785                 atomic_long_read(&n_nocb_offload), atomic_long_read(&n_nocb_deoffload));
1786
1787         pr_alert("%s%s ", torture_type, TORTURE_FLAG);
1788         if (atomic_read(&n_rcu_torture_mberror) ||
1789             atomic_read(&n_rcu_torture_mbchk_fail) ||
1790             n_rcu_torture_barrier_error || n_rcu_torture_boost_ktrerror ||
1791             n_rcu_torture_boost_rterror || n_rcu_torture_boost_failure ||
1792             i > 1) {
1793                 pr_cont("%s", "!!! ");
1794                 atomic_inc(&n_rcu_torture_error);
1795                 WARN_ON_ONCE(atomic_read(&n_rcu_torture_mberror));
1796                 WARN_ON_ONCE(atomic_read(&n_rcu_torture_mbchk_fail));
1797                 WARN_ON_ONCE(n_rcu_torture_barrier_error);  // rcu_barrier()
1798                 WARN_ON_ONCE(n_rcu_torture_boost_ktrerror); // no boost kthread
1799                 WARN_ON_ONCE(n_rcu_torture_boost_rterror); // can't set RT prio
1800                 WARN_ON_ONCE(n_rcu_torture_boost_failure); // RCU boost failed
1801                 WARN_ON_ONCE(i > 1); // Too-short grace period
1802         }
1803         pr_cont("Reader Pipe: ");
1804         for (i = 0; i < RCU_TORTURE_PIPE_LEN + 1; i++)
1805                 pr_cont(" %ld", pipesummary[i]);
1806         pr_cont("\n");
1807
1808         pr_alert("%s%s ", torture_type, TORTURE_FLAG);
1809         pr_cont("Reader Batch: ");
1810         for (i = 0; i < RCU_TORTURE_PIPE_LEN + 1; i++)
1811                 pr_cont(" %ld", batchsummary[i]);
1812         pr_cont("\n");
1813
1814         pr_alert("%s%s ", torture_type, TORTURE_FLAG);
1815         pr_cont("Free-Block Circulation: ");
1816         for (i = 0; i < RCU_TORTURE_PIPE_LEN + 1; i++) {
1817                 pr_cont(" %d", atomic_read(&rcu_torture_wcount[i]));
1818         }
1819         pr_cont("\n");
1820
1821         if (cur_ops->stats)
1822                 cur_ops->stats();
1823         if (rtcv_snap == rcu_torture_current_version &&
1824             rcu_access_pointer(rcu_torture_current) &&
1825             !rcu_stall_is_suppressed()) {
1826                 int __maybe_unused flags = 0;
1827                 unsigned long __maybe_unused gp_seq = 0;
1828
1829                 rcutorture_get_gp_data(cur_ops->ttype,
1830                                        &flags, &gp_seq);
1831                 srcutorture_get_gp_data(cur_ops->ttype, srcu_ctlp,
1832                                         &flags, &gp_seq);
1833                 wtp = READ_ONCE(writer_task);
1834                 pr_alert("??? Writer stall state %s(%d) g%lu f%#x ->state %#lx cpu %d\n",
1835                          rcu_torture_writer_state_getname(),
1836                          rcu_torture_writer_state, gp_seq, flags,
1837                          wtp == NULL ? ~0UL : wtp->state,
1838                          wtp == NULL ? -1 : (int)task_cpu(wtp));
1839                 if (!splatted && wtp) {
1840                         sched_show_task(wtp);
1841                         splatted = true;
1842                 }
1843                 if (cur_ops->gp_kthread_dbg)
1844                         cur_ops->gp_kthread_dbg();
1845                 rcu_ftrace_dump(DUMP_ALL);
1846         }
1847         rtcv_snap = rcu_torture_current_version;
1848 }
1849
1850 /*
1851  * Periodically prints torture statistics, if periodic statistics printing
1852  * was specified via the stat_interval module parameter.
1853  */
1854 static int
1855 rcu_torture_stats(void *arg)
1856 {
1857         VERBOSE_TOROUT_STRING("rcu_torture_stats task started");
1858         do {
1859                 schedule_timeout_interruptible(stat_interval * HZ);
1860                 rcu_torture_stats_print();
1861                 torture_shutdown_absorb("rcu_torture_stats");
1862         } while (!torture_must_stop());
1863         torture_kthread_stopping("rcu_torture_stats");
1864         return 0;
1865 }
1866
1867 static void
1868 rcu_torture_print_module_parms(struct rcu_torture_ops *cur_ops, const char *tag)
1869 {
1870         pr_alert("%s" TORTURE_FLAG
1871                  "--- %s: nreaders=%d nfakewriters=%d "
1872                  "stat_interval=%d verbose=%d test_no_idle_hz=%d "
1873                  "shuffle_interval=%d stutter=%d irqreader=%d "
1874                  "fqs_duration=%d fqs_holdoff=%d fqs_stutter=%d "
1875                  "test_boost=%d/%d test_boost_interval=%d "
1876                  "test_boost_duration=%d shutdown_secs=%d "
1877                  "stall_cpu=%d stall_cpu_holdoff=%d stall_cpu_irqsoff=%d "
1878                  "stall_cpu_block=%d "
1879                  "n_barrier_cbs=%d "
1880                  "onoff_interval=%d onoff_holdoff=%d "
1881                  "read_exit_delay=%d read_exit_burst=%d "
1882                  "nocbs_nthreads=%d nocbs_toggle=%d\n",
1883                  torture_type, tag, nrealreaders, nfakewriters,
1884                  stat_interval, verbose, test_no_idle_hz, shuffle_interval,
1885                  stutter, irqreader, fqs_duration, fqs_holdoff, fqs_stutter,
1886                  test_boost, cur_ops->can_boost,
1887                  test_boost_interval, test_boost_duration, shutdown_secs,
1888                  stall_cpu, stall_cpu_holdoff, stall_cpu_irqsoff,
1889                  stall_cpu_block,
1890                  n_barrier_cbs,
1891                  onoff_interval, onoff_holdoff,
1892                  read_exit_delay, read_exit_burst,
1893                  nocbs_nthreads, nocbs_toggle);
1894 }
1895
1896 static int rcutorture_booster_cleanup(unsigned int cpu)
1897 {
1898         struct task_struct *t;
1899
1900         if (boost_tasks[cpu] == NULL)
1901                 return 0;
1902         mutex_lock(&boost_mutex);
1903         t = boost_tasks[cpu];
1904         boost_tasks[cpu] = NULL;
1905         rcu_torture_enable_rt_throttle();
1906         mutex_unlock(&boost_mutex);
1907
1908         /* This must be outside of the mutex, otherwise deadlock! */
1909         torture_stop_kthread(rcu_torture_boost, t);
1910         return 0;
1911 }
1912
1913 static int rcutorture_booster_init(unsigned int cpu)
1914 {
1915         int retval;
1916
1917         if (boost_tasks[cpu] != NULL)
1918                 return 0;  /* Already created, nothing more to do. */
1919
1920         /* Don't allow time recalculation while creating a new task. */
1921         mutex_lock(&boost_mutex);
1922         rcu_torture_disable_rt_throttle();
1923         VERBOSE_TOROUT_STRING("Creating rcu_torture_boost task");
1924         boost_tasks[cpu] = kthread_create_on_node(rcu_torture_boost, NULL,
1925                                                   cpu_to_node(cpu),
1926                                                   "rcu_torture_boost");
1927         if (IS_ERR(boost_tasks[cpu])) {
1928                 retval = PTR_ERR(boost_tasks[cpu]);
1929                 VERBOSE_TOROUT_STRING("rcu_torture_boost task create failed");
1930                 n_rcu_torture_boost_ktrerror++;
1931                 boost_tasks[cpu] = NULL;
1932                 mutex_unlock(&boost_mutex);
1933                 return retval;
1934         }
1935         kthread_bind(boost_tasks[cpu], cpu);
1936         wake_up_process(boost_tasks[cpu]);
1937         mutex_unlock(&boost_mutex);
1938         return 0;
1939 }
1940
1941 /*
1942  * CPU-stall kthread.  It waits as specified by stall_cpu_holdoff, then
1943  * induces a CPU stall for the time specified by stall_cpu.
1944  */
1945 static int rcu_torture_stall(void *args)
1946 {
1947         int idx;
1948         unsigned long stop_at;
1949
1950         VERBOSE_TOROUT_STRING("rcu_torture_stall task started");
1951         if (stall_cpu_holdoff > 0) {
1952                 VERBOSE_TOROUT_STRING("rcu_torture_stall begin holdoff");
1953                 schedule_timeout_interruptible(stall_cpu_holdoff * HZ);
1954                 VERBOSE_TOROUT_STRING("rcu_torture_stall end holdoff");
1955         }
1956         if (!kthread_should_stop() && stall_gp_kthread > 0) {
1957                 VERBOSE_TOROUT_STRING("rcu_torture_stall begin GP stall");
1958                 rcu_gp_set_torture_wait(stall_gp_kthread * HZ);
1959                 for (idx = 0; idx < stall_gp_kthread + 2; idx++) {
1960                         if (kthread_should_stop())
1961                                 break;
1962                         schedule_timeout_uninterruptible(HZ);
1963                 }
1964         }
1965         if (!kthread_should_stop() && stall_cpu > 0) {
1966                 VERBOSE_TOROUT_STRING("rcu_torture_stall begin CPU stall");
1967                 stop_at = ktime_get_seconds() + stall_cpu;
1968                 /* RCU CPU stall is expected behavior in following code. */
1969                 idx = cur_ops->readlock();
1970                 if (stall_cpu_irqsoff)
1971                         local_irq_disable();
1972                 else if (!stall_cpu_block)
1973                         preempt_disable();
1974                 pr_alert("rcu_torture_stall start on CPU %d.\n",
1975                          raw_smp_processor_id());
1976                 while (ULONG_CMP_LT((unsigned long)ktime_get_seconds(),
1977                                     stop_at))
1978                         if (stall_cpu_block)
1979                                 schedule_timeout_uninterruptible(HZ);
1980                 if (stall_cpu_irqsoff)
1981                         local_irq_enable();
1982                 else if (!stall_cpu_block)
1983                         preempt_enable();
1984                 cur_ops->readunlock(idx);
1985         }
1986         pr_alert("rcu_torture_stall end.\n");
1987         torture_shutdown_absorb("rcu_torture_stall");
1988         while (!kthread_should_stop())
1989                 schedule_timeout_interruptible(10 * HZ);
1990         return 0;
1991 }
1992
1993 /* Spawn CPU-stall kthread, if stall_cpu specified. */
1994 static int __init rcu_torture_stall_init(void)
1995 {
1996         if (stall_cpu <= 0 && stall_gp_kthread <= 0)
1997                 return 0;
1998         return torture_create_kthread(rcu_torture_stall, NULL, stall_task);
1999 }
2000
2001 /* State structure for forward-progress self-propagating RCU callback. */
2002 struct fwd_cb_state {
2003         struct rcu_head rh;
2004         int stop;
2005 };
2006
2007 /*
2008  * Forward-progress self-propagating RCU callback function.  Because
2009  * callbacks run from softirq, this function is an implicit RCU read-side
2010  * critical section.
2011  */
2012 static void rcu_torture_fwd_prog_cb(struct rcu_head *rhp)
2013 {
2014         struct fwd_cb_state *fcsp = container_of(rhp, struct fwd_cb_state, rh);
2015
2016         if (READ_ONCE(fcsp->stop)) {
2017                 WRITE_ONCE(fcsp->stop, 2);
2018                 return;
2019         }
2020         cur_ops->call(&fcsp->rh, rcu_torture_fwd_prog_cb);
2021 }
2022
2023 /* State for continuous-flood RCU callbacks. */
2024 struct rcu_fwd_cb {
2025         struct rcu_head rh;
2026         struct rcu_fwd_cb *rfc_next;
2027         struct rcu_fwd *rfc_rfp;
2028         int rfc_gps;
2029 };
2030
2031 #define MAX_FWD_CB_JIFFIES      (8 * HZ) /* Maximum CB test duration. */
2032 #define MIN_FWD_CB_LAUNDERS     3       /* This many CB invocations to count. */
2033 #define MIN_FWD_CBS_LAUNDERED   100     /* Number of counted CBs. */
2034 #define FWD_CBS_HIST_DIV        10      /* Histogram buckets/second. */
2035 #define N_LAUNDERS_HIST (2 * MAX_FWD_CB_JIFFIES / (HZ / FWD_CBS_HIST_DIV))
2036
2037 struct rcu_launder_hist {
2038         long n_launders;
2039         unsigned long launder_gp_seq;
2040 };
2041
2042 struct rcu_fwd {
2043         spinlock_t rcu_fwd_lock;
2044         struct rcu_fwd_cb *rcu_fwd_cb_head;
2045         struct rcu_fwd_cb **rcu_fwd_cb_tail;
2046         long n_launders_cb;
2047         unsigned long rcu_fwd_startat;
2048         struct rcu_launder_hist n_launders_hist[N_LAUNDERS_HIST];
2049         unsigned long rcu_launder_gp_seq_start;
2050 };
2051
2052 static DEFINE_MUTEX(rcu_fwd_mutex);
2053 static struct rcu_fwd *rcu_fwds;
2054 static bool rcu_fwd_emergency_stop;
2055
2056 static void rcu_torture_fwd_cb_hist(struct rcu_fwd *rfp)
2057 {
2058         unsigned long gps;
2059         unsigned long gps_old;
2060         int i;
2061         int j;
2062
2063         for (i = ARRAY_SIZE(rfp->n_launders_hist) - 1; i > 0; i--)
2064                 if (rfp->n_launders_hist[i].n_launders > 0)
2065                         break;
2066         pr_alert("%s: Callback-invocation histogram (duration %lu jiffies):",
2067                  __func__, jiffies - rfp->rcu_fwd_startat);
2068         gps_old = rfp->rcu_launder_gp_seq_start;
2069         for (j = 0; j <= i; j++) {
2070                 gps = rfp->n_launders_hist[j].launder_gp_seq;
2071                 pr_cont(" %ds/%d: %ld:%ld",
2072                         j + 1, FWD_CBS_HIST_DIV,
2073                         rfp->n_launders_hist[j].n_launders,
2074                         rcutorture_seq_diff(gps, gps_old));
2075                 gps_old = gps;
2076         }
2077         pr_cont("\n");
2078 }
2079
2080 /* Callback function for continuous-flood RCU callbacks. */
2081 static void rcu_torture_fwd_cb_cr(struct rcu_head *rhp)
2082 {
2083         unsigned long flags;
2084         int i;
2085         struct rcu_fwd_cb *rfcp = container_of(rhp, struct rcu_fwd_cb, rh);
2086         struct rcu_fwd_cb **rfcpp;
2087         struct rcu_fwd *rfp = rfcp->rfc_rfp;
2088
2089         rfcp->rfc_next = NULL;
2090         rfcp->rfc_gps++;
2091         spin_lock_irqsave(&rfp->rcu_fwd_lock, flags);
2092         rfcpp = rfp->rcu_fwd_cb_tail;
2093         rfp->rcu_fwd_cb_tail = &rfcp->rfc_next;
2094         WRITE_ONCE(*rfcpp, rfcp);
2095         WRITE_ONCE(rfp->n_launders_cb, rfp->n_launders_cb + 1);
2096         i = ((jiffies - rfp->rcu_fwd_startat) / (HZ / FWD_CBS_HIST_DIV));
2097         if (i >= ARRAY_SIZE(rfp->n_launders_hist))
2098                 i = ARRAY_SIZE(rfp->n_launders_hist) - 1;
2099         rfp->n_launders_hist[i].n_launders++;
2100         rfp->n_launders_hist[i].launder_gp_seq = cur_ops->get_gp_seq();
2101         spin_unlock_irqrestore(&rfp->rcu_fwd_lock, flags);
2102 }
2103
2104 // Give the scheduler a chance, even on nohz_full CPUs.
2105 static void rcu_torture_fwd_prog_cond_resched(unsigned long iter)
2106 {
2107         if (IS_ENABLED(CONFIG_PREEMPTION) && IS_ENABLED(CONFIG_NO_HZ_FULL)) {
2108                 // Real call_rcu() floods hit userspace, so emulate that.
2109                 if (need_resched() || (iter & 0xfff))
2110                         schedule();
2111                 return;
2112         }
2113         // No userspace emulation: CB invocation throttles call_rcu()
2114         cond_resched();
2115 }
2116
2117 /*
2118  * Free all callbacks on the rcu_fwd_cb_head list, either because the
2119  * test is over or because we hit an OOM event.
2120  */
2121 static unsigned long rcu_torture_fwd_prog_cbfree(struct rcu_fwd *rfp)
2122 {
2123         unsigned long flags;
2124         unsigned long freed = 0;
2125         struct rcu_fwd_cb *rfcp;
2126
2127         for (;;) {
2128                 spin_lock_irqsave(&rfp->rcu_fwd_lock, flags);
2129                 rfcp = rfp->rcu_fwd_cb_head;
2130                 if (!rfcp) {
2131                         spin_unlock_irqrestore(&rfp->rcu_fwd_lock, flags);
2132                         break;
2133                 }
2134                 rfp->rcu_fwd_cb_head = rfcp->rfc_next;
2135                 if (!rfp->rcu_fwd_cb_head)
2136                         rfp->rcu_fwd_cb_tail = &rfp->rcu_fwd_cb_head;
2137                 spin_unlock_irqrestore(&rfp->rcu_fwd_lock, flags);
2138                 kfree(rfcp);
2139                 freed++;
2140                 rcu_torture_fwd_prog_cond_resched(freed);
2141                 if (tick_nohz_full_enabled()) {
2142                         local_irq_save(flags);
2143                         rcu_momentary_dyntick_idle();
2144                         local_irq_restore(flags);
2145                 }
2146         }
2147         return freed;
2148 }
2149
2150 /* Carry out need_resched()/cond_resched() forward-progress testing. */
2151 static void rcu_torture_fwd_prog_nr(struct rcu_fwd *rfp,
2152                                     int *tested, int *tested_tries)
2153 {
2154         unsigned long cver;
2155         unsigned long dur;
2156         struct fwd_cb_state fcs;
2157         unsigned long gps;
2158         int idx;
2159         int sd;
2160         int sd4;
2161         bool selfpropcb = false;
2162         unsigned long stopat;
2163         static DEFINE_TORTURE_RANDOM(trs);
2164
2165         if (!cur_ops->sync)
2166                 return; // Cannot do need_resched() forward progress testing without ->sync.
2167         if (cur_ops->call && cur_ops->cb_barrier) {
2168                 init_rcu_head_on_stack(&fcs.rh);
2169                 selfpropcb = true;
2170         }
2171
2172         /* Tight loop containing cond_resched(). */
2173         WRITE_ONCE(rcu_fwd_cb_nodelay, true);
2174         cur_ops->sync(); /* Later readers see above write. */
2175         if  (selfpropcb) {
2176                 WRITE_ONCE(fcs.stop, 0);
2177                 cur_ops->call(&fcs.rh, rcu_torture_fwd_prog_cb);
2178         }
2179         cver = READ_ONCE(rcu_torture_current_version);
2180         gps = cur_ops->get_gp_seq();
2181         sd = cur_ops->stall_dur() + 1;
2182         sd4 = (sd + fwd_progress_div - 1) / fwd_progress_div;
2183         dur = sd4 + torture_random(&trs) % (sd - sd4);
2184         WRITE_ONCE(rfp->rcu_fwd_startat, jiffies);
2185         stopat = rfp->rcu_fwd_startat + dur;
2186         while (time_before(jiffies, stopat) &&
2187                !shutdown_time_arrived() &&
2188                !READ_ONCE(rcu_fwd_emergency_stop) && !torture_must_stop()) {
2189                 idx = cur_ops->readlock();
2190                 udelay(10);
2191                 cur_ops->readunlock(idx);
2192                 if (!fwd_progress_need_resched || need_resched())
2193                         cond_resched();
2194         }
2195         (*tested_tries)++;
2196         if (!time_before(jiffies, stopat) &&
2197             !shutdown_time_arrived() &&
2198             !READ_ONCE(rcu_fwd_emergency_stop) && !torture_must_stop()) {
2199                 (*tested)++;
2200                 cver = READ_ONCE(rcu_torture_current_version) - cver;
2201                 gps = rcutorture_seq_diff(cur_ops->get_gp_seq(), gps);
2202                 WARN_ON(!cver && gps < 2);
2203                 pr_alert("%s: Duration %ld cver %ld gps %ld\n", __func__, dur, cver, gps);
2204         }
2205         if (selfpropcb) {
2206                 WRITE_ONCE(fcs.stop, 1);
2207                 cur_ops->sync(); /* Wait for running CB to complete. */
2208                 cur_ops->cb_barrier(); /* Wait for queued callbacks. */
2209         }
2210
2211         if (selfpropcb) {
2212                 WARN_ON(READ_ONCE(fcs.stop) != 2);
2213                 destroy_rcu_head_on_stack(&fcs.rh);
2214         }
2215         schedule_timeout_uninterruptible(HZ / 10); /* Let kthreads recover. */
2216         WRITE_ONCE(rcu_fwd_cb_nodelay, false);
2217 }
2218
2219 /* Carry out call_rcu() forward-progress testing. */
2220 static void rcu_torture_fwd_prog_cr(struct rcu_fwd *rfp)
2221 {
2222         unsigned long cver;
2223         unsigned long flags;
2224         unsigned long gps;
2225         int i;
2226         long n_launders;
2227         long n_launders_cb_snap;
2228         long n_launders_sa;
2229         long n_max_cbs;
2230         long n_max_gps;
2231         struct rcu_fwd_cb *rfcp;
2232         struct rcu_fwd_cb *rfcpn;
2233         unsigned long stopat;
2234         unsigned long stoppedat;
2235
2236         if (READ_ONCE(rcu_fwd_emergency_stop))
2237                 return; /* Get out of the way quickly, no GP wait! */
2238         if (!cur_ops->call)
2239                 return; /* Can't do call_rcu() fwd prog without ->call. */
2240
2241         /* Loop continuously posting RCU callbacks. */
2242         WRITE_ONCE(rcu_fwd_cb_nodelay, true);
2243         cur_ops->sync(); /* Later readers see above write. */
2244         WRITE_ONCE(rfp->rcu_fwd_startat, jiffies);
2245         stopat = rfp->rcu_fwd_startat + MAX_FWD_CB_JIFFIES;
2246         n_launders = 0;
2247         rfp->n_launders_cb = 0; // Hoist initialization for multi-kthread
2248         n_launders_sa = 0;
2249         n_max_cbs = 0;
2250         n_max_gps = 0;
2251         for (i = 0; i < ARRAY_SIZE(rfp->n_launders_hist); i++)
2252                 rfp->n_launders_hist[i].n_launders = 0;
2253         cver = READ_ONCE(rcu_torture_current_version);
2254         gps = cur_ops->get_gp_seq();
2255         rfp->rcu_launder_gp_seq_start = gps;
2256         tick_dep_set_task(current, TICK_DEP_BIT_RCU);
2257         while (time_before(jiffies, stopat) &&
2258                !shutdown_time_arrived() &&
2259                !READ_ONCE(rcu_fwd_emergency_stop) && !torture_must_stop()) {
2260                 rfcp = READ_ONCE(rfp->rcu_fwd_cb_head);
2261                 rfcpn = NULL;
2262                 if (rfcp)
2263                         rfcpn = READ_ONCE(rfcp->rfc_next);
2264                 if (rfcpn) {
2265                         if (rfcp->rfc_gps >= MIN_FWD_CB_LAUNDERS &&
2266                             ++n_max_gps >= MIN_FWD_CBS_LAUNDERED)
2267                                 break;
2268                         rfp->rcu_fwd_cb_head = rfcpn;
2269                         n_launders++;
2270                         n_launders_sa++;
2271                 } else {
2272                         rfcp = kmalloc(sizeof(*rfcp), GFP_KERNEL);
2273                         if (WARN_ON_ONCE(!rfcp)) {
2274                                 schedule_timeout_interruptible(1);
2275                                 continue;
2276                         }
2277                         n_max_cbs++;
2278                         n_launders_sa = 0;
2279                         rfcp->rfc_gps = 0;
2280                         rfcp->rfc_rfp = rfp;
2281                 }
2282                 cur_ops->call(&rfcp->rh, rcu_torture_fwd_cb_cr);
2283                 rcu_torture_fwd_prog_cond_resched(n_launders + n_max_cbs);
2284                 if (tick_nohz_full_enabled()) {
2285                         local_irq_save(flags);
2286                         rcu_momentary_dyntick_idle();
2287                         local_irq_restore(flags);
2288                 }
2289         }
2290         stoppedat = jiffies;
2291         n_launders_cb_snap = READ_ONCE(rfp->n_launders_cb);
2292         cver = READ_ONCE(rcu_torture_current_version) - cver;
2293         gps = rcutorture_seq_diff(cur_ops->get_gp_seq(), gps);
2294         cur_ops->cb_barrier(); /* Wait for callbacks to be invoked. */
2295         (void)rcu_torture_fwd_prog_cbfree(rfp);
2296
2297         if (!torture_must_stop() && !READ_ONCE(rcu_fwd_emergency_stop) &&
2298             !shutdown_time_arrived()) {
2299                 WARN_ON(n_max_gps < MIN_FWD_CBS_LAUNDERED);
2300                 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",
2301                          __func__,
2302                          stoppedat - rfp->rcu_fwd_startat, jiffies - stoppedat,
2303                          n_launders + n_max_cbs - n_launders_cb_snap,
2304                          n_launders, n_launders_sa,
2305                          n_max_gps, n_max_cbs, cver, gps);
2306                 rcu_torture_fwd_cb_hist(rfp);
2307         }
2308         schedule_timeout_uninterruptible(HZ); /* Let CBs drain. */
2309         tick_dep_clear_task(current, TICK_DEP_BIT_RCU);
2310         WRITE_ONCE(rcu_fwd_cb_nodelay, false);
2311 }
2312
2313
2314 /*
2315  * OOM notifier, but this only prints diagnostic information for the
2316  * current forward-progress test.
2317  */
2318 static int rcutorture_oom_notify(struct notifier_block *self,
2319                                  unsigned long notused, void *nfreed)
2320 {
2321         struct rcu_fwd *rfp;
2322
2323         mutex_lock(&rcu_fwd_mutex);
2324         rfp = rcu_fwds;
2325         if (!rfp) {
2326                 mutex_unlock(&rcu_fwd_mutex);
2327                 return NOTIFY_OK;
2328         }
2329         WARN(1, "%s invoked upon OOM during forward-progress testing.\n",
2330              __func__);
2331         rcu_torture_fwd_cb_hist(rfp);
2332         rcu_fwd_progress_check(1 + (jiffies - READ_ONCE(rfp->rcu_fwd_startat)) / 2);
2333         WRITE_ONCE(rcu_fwd_emergency_stop, true);
2334         smp_mb(); /* Emergency stop before free and wait to avoid hangs. */
2335         pr_info("%s: Freed %lu RCU callbacks.\n",
2336                 __func__, rcu_torture_fwd_prog_cbfree(rfp));
2337         rcu_barrier();
2338         pr_info("%s: Freed %lu RCU callbacks.\n",
2339                 __func__, rcu_torture_fwd_prog_cbfree(rfp));
2340         rcu_barrier();
2341         pr_info("%s: Freed %lu RCU callbacks.\n",
2342                 __func__, rcu_torture_fwd_prog_cbfree(rfp));
2343         smp_mb(); /* Frees before return to avoid redoing OOM. */
2344         (*(unsigned long *)nfreed)++; /* Forward progress CBs freed! */
2345         pr_info("%s returning after OOM processing.\n", __func__);
2346         mutex_unlock(&rcu_fwd_mutex);
2347         return NOTIFY_OK;
2348 }
2349
2350 static struct notifier_block rcutorture_oom_nb = {
2351         .notifier_call = rcutorture_oom_notify
2352 };
2353
2354 /* Carry out grace-period forward-progress testing. */
2355 static int rcu_torture_fwd_prog(void *args)
2356 {
2357         int oldnice = task_nice(current);
2358         struct rcu_fwd *rfp = args;
2359         int tested = 0;
2360         int tested_tries = 0;
2361
2362         VERBOSE_TOROUT_STRING("rcu_torture_fwd_progress task started");
2363         rcu_bind_current_to_nocb();
2364         if (!IS_ENABLED(CONFIG_SMP) || !IS_ENABLED(CONFIG_RCU_BOOST))
2365                 set_user_nice(current, MAX_NICE);
2366         do {
2367                 schedule_timeout_interruptible(fwd_progress_holdoff * HZ);
2368                 WRITE_ONCE(rcu_fwd_emergency_stop, false);
2369                 if (!IS_ENABLED(CONFIG_TINY_RCU) ||
2370                     rcu_inkernel_boot_has_ended())
2371                         rcu_torture_fwd_prog_nr(rfp, &tested, &tested_tries);
2372                 if (rcu_inkernel_boot_has_ended())
2373                         rcu_torture_fwd_prog_cr(rfp);
2374
2375                 /* Avoid slow periods, better to test when busy. */
2376                 if (stutter_wait("rcu_torture_fwd_prog"))
2377                         sched_set_normal(current, oldnice);
2378         } while (!torture_must_stop());
2379         /* Short runs might not contain a valid forward-progress attempt. */
2380         WARN_ON(!tested && tested_tries >= 5);
2381         pr_alert("%s: tested %d tested_tries %d\n", __func__, tested, tested_tries);
2382         torture_kthread_stopping("rcu_torture_fwd_prog");
2383         return 0;
2384 }
2385
2386 /* If forward-progress checking is requested and feasible, spawn the thread. */
2387 static int __init rcu_torture_fwd_prog_init(void)
2388 {
2389         struct rcu_fwd *rfp;
2390
2391         if (!fwd_progress)
2392                 return 0; /* Not requested, so don't do it. */
2393         if ((!cur_ops->sync && !cur_ops->call) ||
2394             !cur_ops->stall_dur || cur_ops->stall_dur() <= 0 || cur_ops == &rcu_busted_ops) {
2395                 VERBOSE_TOROUT_STRING("rcu_torture_fwd_prog_init: Disabled, unsupported by RCU flavor under test");
2396                 return 0;
2397         }
2398         if (stall_cpu > 0) {
2399                 VERBOSE_TOROUT_STRING("rcu_torture_fwd_prog_init: Disabled, conflicts with CPU-stall testing");
2400                 if (IS_MODULE(CONFIG_RCU_TORTURE_TESTS))
2401                         return -EINVAL; /* In module, can fail back to user. */
2402                 WARN_ON(1); /* Make sure rcutorture notices conflict. */
2403                 return 0;
2404         }
2405         if (fwd_progress_holdoff <= 0)
2406                 fwd_progress_holdoff = 1;
2407         if (fwd_progress_div <= 0)
2408                 fwd_progress_div = 4;
2409         rfp = kzalloc(sizeof(*rfp), GFP_KERNEL);
2410         if (!rfp)
2411                 return -ENOMEM;
2412         spin_lock_init(&rfp->rcu_fwd_lock);
2413         rfp->rcu_fwd_cb_tail = &rfp->rcu_fwd_cb_head;
2414         mutex_lock(&rcu_fwd_mutex);
2415         rcu_fwds = rfp;
2416         mutex_unlock(&rcu_fwd_mutex);
2417         register_oom_notifier(&rcutorture_oom_nb);
2418         return torture_create_kthread(rcu_torture_fwd_prog, rfp, fwd_prog_task);
2419 }
2420
2421 static void rcu_torture_fwd_prog_cleanup(void)
2422 {
2423         struct rcu_fwd *rfp;
2424
2425         torture_stop_kthread(rcu_torture_fwd_prog, fwd_prog_task);
2426         rfp = rcu_fwds;
2427         mutex_lock(&rcu_fwd_mutex);
2428         rcu_fwds = NULL;
2429         mutex_unlock(&rcu_fwd_mutex);
2430         unregister_oom_notifier(&rcutorture_oom_nb);
2431         kfree(rfp);
2432 }
2433
2434 /* Callback function for RCU barrier testing. */
2435 static void rcu_torture_barrier_cbf(struct rcu_head *rcu)
2436 {
2437         atomic_inc(&barrier_cbs_invoked);
2438 }
2439
2440 /* IPI handler to get callback posted on desired CPU, if online. */
2441 static void rcu_torture_barrier1cb(void *rcu_void)
2442 {
2443         struct rcu_head *rhp = rcu_void;
2444
2445         cur_ops->call(rhp, rcu_torture_barrier_cbf);
2446 }
2447
2448 /* kthread function to register callbacks used to test RCU barriers. */
2449 static int rcu_torture_barrier_cbs(void *arg)
2450 {
2451         long myid = (long)arg;
2452         bool lastphase = false;
2453         bool newphase;
2454         struct rcu_head rcu;
2455
2456         init_rcu_head_on_stack(&rcu);
2457         VERBOSE_TOROUT_STRING("rcu_torture_barrier_cbs task started");
2458         set_user_nice(current, MAX_NICE);
2459         do {
2460                 wait_event(barrier_cbs_wq[myid],
2461                            (newphase =
2462                             smp_load_acquire(&barrier_phase)) != lastphase ||
2463                            torture_must_stop());
2464                 lastphase = newphase;
2465                 if (torture_must_stop())
2466                         break;
2467                 /*
2468                  * The above smp_load_acquire() ensures barrier_phase load
2469                  * is ordered before the following ->call().
2470                  */
2471                 if (smp_call_function_single(myid, rcu_torture_barrier1cb,
2472                                              &rcu, 1)) {
2473                         // IPI failed, so use direct call from current CPU.
2474                         cur_ops->call(&rcu, rcu_torture_barrier_cbf);
2475                 }
2476                 if (atomic_dec_and_test(&barrier_cbs_count))
2477                         wake_up(&barrier_wq);
2478         } while (!torture_must_stop());
2479         if (cur_ops->cb_barrier != NULL)
2480                 cur_ops->cb_barrier();
2481         destroy_rcu_head_on_stack(&rcu);
2482         torture_kthread_stopping("rcu_torture_barrier_cbs");
2483         return 0;
2484 }
2485
2486 /* kthread function to drive and coordinate RCU barrier testing. */
2487 static int rcu_torture_barrier(void *arg)
2488 {
2489         int i;
2490
2491         VERBOSE_TOROUT_STRING("rcu_torture_barrier task starting");
2492         do {
2493                 atomic_set(&barrier_cbs_invoked, 0);
2494                 atomic_set(&barrier_cbs_count, n_barrier_cbs);
2495                 /* Ensure barrier_phase ordered after prior assignments. */
2496                 smp_store_release(&barrier_phase, !barrier_phase);
2497                 for (i = 0; i < n_barrier_cbs; i++)
2498                         wake_up(&barrier_cbs_wq[i]);
2499                 wait_event(barrier_wq,
2500                            atomic_read(&barrier_cbs_count) == 0 ||
2501                            torture_must_stop());
2502                 if (torture_must_stop())
2503                         break;
2504                 n_barrier_attempts++;
2505                 cur_ops->cb_barrier(); /* Implies smp_mb() for wait_event(). */
2506                 if (atomic_read(&barrier_cbs_invoked) != n_barrier_cbs) {
2507                         n_rcu_torture_barrier_error++;
2508                         pr_err("barrier_cbs_invoked = %d, n_barrier_cbs = %d\n",
2509                                atomic_read(&barrier_cbs_invoked),
2510                                n_barrier_cbs);
2511                         WARN_ON(1);
2512                         // Wait manually for the remaining callbacks
2513                         i = 0;
2514                         do {
2515                                 if (WARN_ON(i++ > HZ))
2516                                         i = INT_MIN;
2517                                 schedule_timeout_interruptible(1);
2518                                 cur_ops->cb_barrier();
2519                         } while (atomic_read(&barrier_cbs_invoked) !=
2520                                  n_barrier_cbs &&
2521                                  !torture_must_stop());
2522                         smp_mb(); // Can't trust ordering if broken.
2523                         if (!torture_must_stop())
2524                                 pr_err("Recovered: barrier_cbs_invoked = %d\n",
2525                                        atomic_read(&barrier_cbs_invoked));
2526                 } else {
2527                         n_barrier_successes++;
2528                 }
2529                 schedule_timeout_interruptible(HZ / 10);
2530         } while (!torture_must_stop());
2531         torture_kthread_stopping("rcu_torture_barrier");
2532         return 0;
2533 }
2534
2535 /* Initialize RCU barrier testing. */
2536 static int rcu_torture_barrier_init(void)
2537 {
2538         int i;
2539         int ret;
2540
2541         if (n_barrier_cbs <= 0)
2542                 return 0;
2543         if (cur_ops->call == NULL || cur_ops->cb_barrier == NULL) {
2544                 pr_alert("%s" TORTURE_FLAG
2545                          " Call or barrier ops missing for %s,\n",
2546                          torture_type, cur_ops->name);
2547                 pr_alert("%s" TORTURE_FLAG
2548                          " RCU barrier testing omitted from run.\n",
2549                          torture_type);
2550                 return 0;
2551         }
2552         atomic_set(&barrier_cbs_count, 0);
2553         atomic_set(&barrier_cbs_invoked, 0);
2554         barrier_cbs_tasks =
2555                 kcalloc(n_barrier_cbs, sizeof(barrier_cbs_tasks[0]),
2556                         GFP_KERNEL);
2557         barrier_cbs_wq =
2558                 kcalloc(n_barrier_cbs, sizeof(barrier_cbs_wq[0]), GFP_KERNEL);
2559         if (barrier_cbs_tasks == NULL || !barrier_cbs_wq)
2560                 return -ENOMEM;
2561         for (i = 0; i < n_barrier_cbs; i++) {
2562                 init_waitqueue_head(&barrier_cbs_wq[i]);
2563                 ret = torture_create_kthread(rcu_torture_barrier_cbs,
2564                                              (void *)(long)i,
2565                                              barrier_cbs_tasks[i]);
2566                 if (ret)
2567                         return ret;
2568         }
2569         return torture_create_kthread(rcu_torture_barrier, NULL, barrier_task);
2570 }
2571
2572 /* Clean up after RCU barrier testing. */
2573 static void rcu_torture_barrier_cleanup(void)
2574 {
2575         int i;
2576
2577         torture_stop_kthread(rcu_torture_barrier, barrier_task);
2578         if (barrier_cbs_tasks != NULL) {
2579                 for (i = 0; i < n_barrier_cbs; i++)
2580                         torture_stop_kthread(rcu_torture_barrier_cbs,
2581                                              barrier_cbs_tasks[i]);
2582                 kfree(barrier_cbs_tasks);
2583                 barrier_cbs_tasks = NULL;
2584         }
2585         if (barrier_cbs_wq != NULL) {
2586                 kfree(barrier_cbs_wq);
2587                 barrier_cbs_wq = NULL;
2588         }
2589 }
2590
2591 static bool rcu_torture_can_boost(void)
2592 {
2593         static int boost_warn_once;
2594         int prio;
2595
2596         if (!(test_boost == 1 && cur_ops->can_boost) && test_boost != 2)
2597                 return false;
2598
2599         prio = rcu_get_gp_kthreads_prio();
2600         if (!prio)
2601                 return false;
2602
2603         if (prio < 2) {
2604                 if (boost_warn_once  == 1)
2605                         return false;
2606
2607                 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);
2608                 boost_warn_once = 1;
2609                 return false;
2610         }
2611
2612         return true;
2613 }
2614
2615 static bool read_exit_child_stop;
2616 static bool read_exit_child_stopped;
2617 static wait_queue_head_t read_exit_wq;
2618
2619 // Child kthread which just does an rcutorture reader and exits.
2620 static int rcu_torture_read_exit_child(void *trsp_in)
2621 {
2622         struct torture_random_state *trsp = trsp_in;
2623
2624         set_user_nice(current, MAX_NICE);
2625         // Minimize time between reading and exiting.
2626         while (!kthread_should_stop())
2627                 schedule_timeout_uninterruptible(1);
2628         (void)rcu_torture_one_read(trsp, -1);
2629         return 0;
2630 }
2631
2632 // Parent kthread which creates and destroys read-exit child kthreads.
2633 static int rcu_torture_read_exit(void *unused)
2634 {
2635         int count = 0;
2636         bool errexit = false;
2637         int i;
2638         struct task_struct *tsp;
2639         DEFINE_TORTURE_RANDOM(trs);
2640
2641         // Allocate and initialize.
2642         set_user_nice(current, MAX_NICE);
2643         VERBOSE_TOROUT_STRING("rcu_torture_read_exit: Start of test");
2644
2645         // Each pass through this loop does one read-exit episode.
2646         do {
2647                 if (++count > read_exit_burst) {
2648                         VERBOSE_TOROUT_STRING("rcu_torture_read_exit: End of episode");
2649                         rcu_barrier(); // Wait for task_struct free, avoid OOM.
2650                         for (i = 0; i < read_exit_delay; i++) {
2651                                 schedule_timeout_uninterruptible(HZ);
2652                                 if (READ_ONCE(read_exit_child_stop))
2653                                         break;
2654                         }
2655                         if (!READ_ONCE(read_exit_child_stop))
2656                                 VERBOSE_TOROUT_STRING("rcu_torture_read_exit: Start of episode");
2657                         count = 0;
2658                 }
2659                 if (READ_ONCE(read_exit_child_stop))
2660                         break;
2661                 // Spawn child.
2662                 tsp = kthread_run(rcu_torture_read_exit_child,
2663                                      &trs, "%s",
2664                                      "rcu_torture_read_exit_child");
2665                 if (IS_ERR(tsp)) {
2666                         VERBOSE_TOROUT_ERRSTRING("out of memory");
2667                         errexit = true;
2668                         tsp = NULL;
2669                         break;
2670                 }
2671                 cond_resched();
2672                 kthread_stop(tsp);
2673                 n_read_exits ++;
2674                 stutter_wait("rcu_torture_read_exit");
2675         } while (!errexit && !READ_ONCE(read_exit_child_stop));
2676
2677         // Clean up and exit.
2678         smp_store_release(&read_exit_child_stopped, true); // After reaping.
2679         smp_mb(); // Store before wakeup.
2680         wake_up(&read_exit_wq);
2681         while (!torture_must_stop())
2682                 schedule_timeout_uninterruptible(1);
2683         torture_kthread_stopping("rcu_torture_read_exit");
2684         return 0;
2685 }
2686
2687 static int rcu_torture_read_exit_init(void)
2688 {
2689         if (read_exit_burst <= 0)
2690                 return -EINVAL;
2691         init_waitqueue_head(&read_exit_wq);
2692         read_exit_child_stop = false;
2693         read_exit_child_stopped = false;
2694         return torture_create_kthread(rcu_torture_read_exit, NULL,
2695                                       read_exit_task);
2696 }
2697
2698 static void rcu_torture_read_exit_cleanup(void)
2699 {
2700         if (!read_exit_task)
2701                 return;
2702         WRITE_ONCE(read_exit_child_stop, true);
2703         smp_mb(); // Above write before wait.
2704         wait_event(read_exit_wq, smp_load_acquire(&read_exit_child_stopped));
2705         torture_stop_kthread(rcutorture_read_exit, read_exit_task);
2706 }
2707
2708 static enum cpuhp_state rcutor_hp;
2709
2710 static void
2711 rcu_torture_cleanup(void)
2712 {
2713         int firsttime;
2714         int flags = 0;
2715         unsigned long gp_seq = 0;
2716         int i;
2717
2718         if (torture_cleanup_begin()) {
2719                 if (cur_ops->cb_barrier != NULL)
2720                         cur_ops->cb_barrier();
2721                 return;
2722         }
2723         if (!cur_ops) {
2724                 torture_cleanup_end();
2725                 return;
2726         }
2727
2728         if (cur_ops->gp_kthread_dbg)
2729                 cur_ops->gp_kthread_dbg();
2730         rcu_torture_read_exit_cleanup();
2731         rcu_torture_barrier_cleanup();
2732         rcu_torture_fwd_prog_cleanup();
2733         torture_stop_kthread(rcu_torture_stall, stall_task);
2734         torture_stop_kthread(rcu_torture_writer, writer_task);
2735
2736         if (nocb_tasks) {
2737                 for (i = 0; i < nrealnocbers; i++)
2738                         torture_stop_kthread(rcu_nocb_toggle, nocb_tasks[i]);
2739                 kfree(nocb_tasks);
2740                 nocb_tasks = NULL;
2741         }
2742
2743         if (reader_tasks) {
2744                 for (i = 0; i < nrealreaders; i++)
2745                         torture_stop_kthread(rcu_torture_reader,
2746                                              reader_tasks[i]);
2747                 kfree(reader_tasks);
2748                 reader_tasks = NULL;
2749         }
2750         kfree(rcu_torture_reader_mbchk);
2751         rcu_torture_reader_mbchk = NULL;
2752
2753         if (fakewriter_tasks) {
2754                 for (i = 0; i < nfakewriters; i++)
2755                         torture_stop_kthread(rcu_torture_fakewriter,
2756                                              fakewriter_tasks[i]);
2757                 kfree(fakewriter_tasks);
2758                 fakewriter_tasks = NULL;
2759         }
2760
2761         rcutorture_get_gp_data(cur_ops->ttype, &flags, &gp_seq);
2762         srcutorture_get_gp_data(cur_ops->ttype, srcu_ctlp, &flags, &gp_seq);
2763         pr_alert("%s:  End-test grace-period state: g%ld f%#x total-gps=%ld\n",
2764                  cur_ops->name, (long)gp_seq, flags,
2765                  rcutorture_seq_diff(gp_seq, start_gp_seq));
2766         torture_stop_kthread(rcu_torture_stats, stats_task);
2767         torture_stop_kthread(rcu_torture_fqs, fqs_task);
2768         if (rcu_torture_can_boost())
2769                 cpuhp_remove_state(rcutor_hp);
2770
2771         /*
2772          * Wait for all RCU callbacks to fire, then do torture-type-specific
2773          * cleanup operations.
2774          */
2775         if (cur_ops->cb_barrier != NULL)
2776                 cur_ops->cb_barrier();
2777         if (cur_ops->cleanup != NULL)
2778                 cur_ops->cleanup();
2779
2780         rcu_torture_stats_print();  /* -After- the stats thread is stopped! */
2781
2782         if (err_segs_recorded) {
2783                 pr_alert("Failure/close-call rcutorture reader segments:\n");
2784                 if (rt_read_nsegs == 0)
2785                         pr_alert("\t: No segments recorded!!!\n");
2786                 firsttime = 1;
2787                 for (i = 0; i < rt_read_nsegs; i++) {
2788                         pr_alert("\t%d: %#x ", i, err_segs[i].rt_readstate);
2789                         if (err_segs[i].rt_delay_jiffies != 0) {
2790                                 pr_cont("%s%ldjiffies", firsttime ? "" : "+",
2791                                         err_segs[i].rt_delay_jiffies);
2792                                 firsttime = 0;
2793                         }
2794                         if (err_segs[i].rt_delay_ms != 0) {
2795                                 pr_cont("%s%ldms", firsttime ? "" : "+",
2796                                         err_segs[i].rt_delay_ms);
2797                                 firsttime = 0;
2798                         }
2799                         if (err_segs[i].rt_delay_us != 0) {
2800                                 pr_cont("%s%ldus", firsttime ? "" : "+",
2801                                         err_segs[i].rt_delay_us);
2802                                 firsttime = 0;
2803                         }
2804                         pr_cont("%s\n",
2805                                 err_segs[i].rt_preempted ? "preempted" : "");
2806
2807                 }
2808         }
2809         if (atomic_read(&n_rcu_torture_error) || n_rcu_torture_barrier_error)
2810                 rcu_torture_print_module_parms(cur_ops, "End of test: FAILURE");
2811         else if (torture_onoff_failures())
2812                 rcu_torture_print_module_parms(cur_ops,
2813                                                "End of test: RCU_HOTPLUG");
2814         else
2815                 rcu_torture_print_module_parms(cur_ops, "End of test: SUCCESS");
2816         torture_cleanup_end();
2817 }
2818
2819 #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD
2820 static void rcu_torture_leak_cb(struct rcu_head *rhp)
2821 {
2822 }
2823
2824 static void rcu_torture_err_cb(struct rcu_head *rhp)
2825 {
2826         /*
2827          * This -might- happen due to race conditions, but is unlikely.
2828          * The scenario that leads to this happening is that the
2829          * first of the pair of duplicate callbacks is queued,
2830          * someone else starts a grace period that includes that
2831          * callback, then the second of the pair must wait for the
2832          * next grace period.  Unlikely, but can happen.  If it
2833          * does happen, the debug-objects subsystem won't have splatted.
2834          */
2835         pr_alert("%s: duplicated callback was invoked.\n", KBUILD_MODNAME);
2836 }
2837 #endif /* #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD */
2838
2839 /*
2840  * Verify that double-free causes debug-objects to complain, but only
2841  * if CONFIG_DEBUG_OBJECTS_RCU_HEAD=y.  Otherwise, say that the test
2842  * cannot be carried out.
2843  */
2844 static void rcu_test_debug_objects(void)
2845 {
2846 #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD
2847         struct rcu_head rh1;
2848         struct rcu_head rh2;
2849         struct rcu_head *rhp = kmalloc(sizeof(*rhp), GFP_KERNEL);
2850
2851         init_rcu_head_on_stack(&rh1);
2852         init_rcu_head_on_stack(&rh2);
2853         pr_alert("%s: WARN: Duplicate call_rcu() test starting.\n", KBUILD_MODNAME);
2854
2855         /* Try to queue the rh2 pair of callbacks for the same grace period. */
2856         preempt_disable(); /* Prevent preemption from interrupting test. */
2857         rcu_read_lock(); /* Make it impossible to finish a grace period. */
2858         call_rcu(&rh1, rcu_torture_leak_cb); /* Start grace period. */
2859         local_irq_disable(); /* Make it harder to start a new grace period. */
2860         call_rcu(&rh2, rcu_torture_leak_cb);
2861         call_rcu(&rh2, rcu_torture_err_cb); /* Duplicate callback. */
2862         if (rhp) {
2863                 call_rcu(rhp, rcu_torture_leak_cb);
2864                 call_rcu(rhp, rcu_torture_err_cb); /* Another duplicate callback. */
2865         }
2866         local_irq_enable();
2867         rcu_read_unlock();
2868         preempt_enable();
2869
2870         /* Wait for them all to get done so we can safely return. */
2871         rcu_barrier();
2872         pr_alert("%s: WARN: Duplicate call_rcu() test complete.\n", KBUILD_MODNAME);
2873         destroy_rcu_head_on_stack(&rh1);
2874         destroy_rcu_head_on_stack(&rh2);
2875 #else /* #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD */
2876         pr_alert("%s: !CONFIG_DEBUG_OBJECTS_RCU_HEAD, not testing duplicate call_rcu()\n", KBUILD_MODNAME);
2877 #endif /* #else #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD */
2878 }
2879
2880 static void rcutorture_sync(void)
2881 {
2882         static unsigned long n;
2883
2884         if (cur_ops->sync && !(++n & 0xfff))
2885                 cur_ops->sync();
2886 }
2887
2888 static int __init
2889 rcu_torture_init(void)
2890 {
2891         long i;
2892         int cpu;
2893         int firsterr = 0;
2894         int flags = 0;
2895         unsigned long gp_seq = 0;
2896         static struct rcu_torture_ops *torture_ops[] = {
2897                 &rcu_ops, &rcu_busted_ops, &srcu_ops, &srcud_ops,
2898                 &busted_srcud_ops, &tasks_ops, &tasks_rude_ops,
2899                 &tasks_tracing_ops, &trivial_ops,
2900         };
2901
2902         if (!torture_init_begin(torture_type, verbose))
2903                 return -EBUSY;
2904
2905         /* Process args and tell the world that the torturer is on the job. */
2906         for (i = 0; i < ARRAY_SIZE(torture_ops); i++) {
2907                 cur_ops = torture_ops[i];
2908                 if (strcmp(torture_type, cur_ops->name) == 0)
2909                         break;
2910         }
2911         if (i == ARRAY_SIZE(torture_ops)) {
2912                 pr_alert("rcu-torture: invalid torture type: \"%s\"\n",
2913                          torture_type);
2914                 pr_alert("rcu-torture types:");
2915                 for (i = 0; i < ARRAY_SIZE(torture_ops); i++)
2916                         pr_cont(" %s", torture_ops[i]->name);
2917                 pr_cont("\n");
2918                 firsterr = -EINVAL;
2919                 cur_ops = NULL;
2920                 goto unwind;
2921         }
2922         if (cur_ops->fqs == NULL && fqs_duration != 0) {
2923                 pr_alert("rcu-torture: ->fqs NULL and non-zero fqs_duration, fqs disabled.\n");
2924                 fqs_duration = 0;
2925         }
2926         if (cur_ops->init)
2927                 cur_ops->init();
2928
2929         if (nreaders >= 0) {
2930                 nrealreaders = nreaders;
2931         } else {
2932                 nrealreaders = num_online_cpus() - 2 - nreaders;
2933                 if (nrealreaders <= 0)
2934                         nrealreaders = 1;
2935         }
2936         rcu_torture_print_module_parms(cur_ops, "Start of test");
2937         rcutorture_get_gp_data(cur_ops->ttype, &flags, &gp_seq);
2938         srcutorture_get_gp_data(cur_ops->ttype, srcu_ctlp, &flags, &gp_seq);
2939         start_gp_seq = gp_seq;
2940         pr_alert("%s:  Start-test grace-period state: g%ld f%#x\n",
2941                  cur_ops->name, (long)gp_seq, flags);
2942
2943         /* Set up the freelist. */
2944
2945         INIT_LIST_HEAD(&rcu_torture_freelist);
2946         for (i = 0; i < ARRAY_SIZE(rcu_tortures); i++) {
2947                 rcu_tortures[i].rtort_mbtest = 0;
2948                 list_add_tail(&rcu_tortures[i].rtort_free,
2949                               &rcu_torture_freelist);
2950         }
2951
2952         /* Initialize the statistics so that each run gets its own numbers. */
2953
2954         rcu_torture_current = NULL;
2955         rcu_torture_current_version = 0;
2956         atomic_set(&n_rcu_torture_alloc, 0);
2957         atomic_set(&n_rcu_torture_alloc_fail, 0);
2958         atomic_set(&n_rcu_torture_free, 0);
2959         atomic_set(&n_rcu_torture_mberror, 0);
2960         atomic_set(&n_rcu_torture_mbchk_fail, 0);
2961         atomic_set(&n_rcu_torture_mbchk_tries, 0);
2962         atomic_set(&n_rcu_torture_error, 0);
2963         n_rcu_torture_barrier_error = 0;
2964         n_rcu_torture_boost_ktrerror = 0;
2965         n_rcu_torture_boost_rterror = 0;
2966         n_rcu_torture_boost_failure = 0;
2967         n_rcu_torture_boosts = 0;
2968         for (i = 0; i < RCU_TORTURE_PIPE_LEN + 1; i++)
2969                 atomic_set(&rcu_torture_wcount[i], 0);
2970         for_each_possible_cpu(cpu) {
2971                 for (i = 0; i < RCU_TORTURE_PIPE_LEN + 1; i++) {
2972                         per_cpu(rcu_torture_count, cpu)[i] = 0;
2973                         per_cpu(rcu_torture_batch, cpu)[i] = 0;
2974                 }
2975         }
2976         err_segs_recorded = 0;
2977         rt_read_nsegs = 0;
2978
2979         /* Start up the kthreads. */
2980
2981         rcu_torture_write_types();
2982         firsterr = torture_create_kthread(rcu_torture_writer, NULL,
2983                                           writer_task);
2984         if (firsterr)
2985                 goto unwind;
2986         if (nfakewriters > 0) {
2987                 fakewriter_tasks = kcalloc(nfakewriters,
2988                                            sizeof(fakewriter_tasks[0]),
2989                                            GFP_KERNEL);
2990                 if (fakewriter_tasks == NULL) {
2991                         VERBOSE_TOROUT_ERRSTRING("out of memory");
2992                         firsterr = -ENOMEM;
2993                         goto unwind;
2994                 }
2995         }
2996         for (i = 0; i < nfakewriters; i++) {
2997                 firsterr = torture_create_kthread(rcu_torture_fakewriter,
2998                                                   NULL, fakewriter_tasks[i]);
2999                 if (firsterr)
3000                         goto unwind;
3001         }
3002         reader_tasks = kcalloc(nrealreaders, sizeof(reader_tasks[0]),
3003                                GFP_KERNEL);
3004         rcu_torture_reader_mbchk = kcalloc(nrealreaders, sizeof(*rcu_torture_reader_mbchk),
3005                                            GFP_KERNEL);
3006         if (!reader_tasks || !rcu_torture_reader_mbchk) {
3007                 VERBOSE_TOROUT_ERRSTRING("out of memory");
3008                 firsterr = -ENOMEM;
3009                 goto unwind;
3010         }
3011         for (i = 0; i < nrealreaders; i++) {
3012                 rcu_torture_reader_mbchk[i].rtc_chkrdr = -1;
3013                 firsterr = torture_create_kthread(rcu_torture_reader, (void *)i,
3014                                                   reader_tasks[i]);
3015                 if (firsterr)
3016                         goto unwind;
3017         }
3018         nrealnocbers = nocbs_nthreads;
3019         if (WARN_ON(nrealnocbers < 0))
3020                 nrealnocbers = 1;
3021         if (WARN_ON(nocbs_toggle < 0))
3022                 nocbs_toggle = HZ;
3023         if (nrealnocbers > 0) {
3024                 nocb_tasks = kcalloc(nrealnocbers, sizeof(nocb_tasks[0]), GFP_KERNEL);
3025                 if (nocb_tasks == NULL) {
3026                         VERBOSE_TOROUT_ERRSTRING("out of memory");
3027                         firsterr = -ENOMEM;
3028                         goto unwind;
3029                 }
3030         } else {
3031                 nocb_tasks = NULL;
3032         }
3033         for (i = 0; i < nrealnocbers; i++) {
3034                 firsterr = torture_create_kthread(rcu_nocb_toggle, NULL, nocb_tasks[i]);
3035                 if (firsterr)
3036                         goto unwind;
3037         }
3038         if (stat_interval > 0) {
3039                 firsterr = torture_create_kthread(rcu_torture_stats, NULL,
3040                                                   stats_task);
3041                 if (firsterr)
3042                         goto unwind;
3043         }
3044         if (test_no_idle_hz && shuffle_interval > 0) {
3045                 firsterr = torture_shuffle_init(shuffle_interval * HZ);
3046                 if (firsterr)
3047                         goto unwind;
3048         }
3049         if (stutter < 0)
3050                 stutter = 0;
3051         if (stutter) {
3052                 int t;
3053
3054                 t = cur_ops->stall_dur ? cur_ops->stall_dur() : stutter * HZ;
3055                 firsterr = torture_stutter_init(stutter * HZ, t);
3056                 if (firsterr)
3057                         goto unwind;
3058         }
3059         if (fqs_duration < 0)
3060                 fqs_duration = 0;
3061         if (fqs_duration) {
3062                 /* Create the fqs thread */
3063                 firsterr = torture_create_kthread(rcu_torture_fqs, NULL,
3064                                                   fqs_task);
3065                 if (firsterr)
3066                         goto unwind;
3067         }
3068         if (test_boost_interval < 1)
3069                 test_boost_interval = 1;
3070         if (test_boost_duration < 2)
3071                 test_boost_duration = 2;
3072         if (rcu_torture_can_boost()) {
3073
3074                 boost_starttime = jiffies + test_boost_interval * HZ;
3075
3076                 firsterr = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "RCU_TORTURE",
3077                                              rcutorture_booster_init,
3078                                              rcutorture_booster_cleanup);
3079                 if (firsterr < 0)
3080                         goto unwind;
3081                 rcutor_hp = firsterr;
3082         }
3083         shutdown_jiffies = jiffies + shutdown_secs * HZ;
3084         firsterr = torture_shutdown_init(shutdown_secs, rcu_torture_cleanup);
3085         if (firsterr)
3086                 goto unwind;
3087         firsterr = torture_onoff_init(onoff_holdoff * HZ, onoff_interval,
3088                                       rcutorture_sync);
3089         if (firsterr)
3090                 goto unwind;
3091         firsterr = rcu_torture_stall_init();
3092         if (firsterr)
3093                 goto unwind;
3094         firsterr = rcu_torture_fwd_prog_init();
3095         if (firsterr)
3096                 goto unwind;
3097         firsterr = rcu_torture_barrier_init();
3098         if (firsterr)
3099                 goto unwind;
3100         firsterr = rcu_torture_read_exit_init();
3101         if (firsterr)
3102                 goto unwind;
3103         if (object_debug)
3104                 rcu_test_debug_objects();
3105         torture_init_end();
3106         return 0;
3107
3108 unwind:
3109         torture_init_end();
3110         rcu_torture_cleanup();
3111         if (shutdown_secs) {
3112                 WARN_ON(!IS_MODULE(CONFIG_RCU_TORTURE_TEST));
3113                 kernel_power_off();
3114         }
3115         return firsterr;
3116 }
3117
3118 module_init(rcu_torture_init);
3119 module_exit(rcu_torture_cleanup);