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 <[email protected]>
   8  *        Josh Triplett <[email protected]>
   9  *
  10  * See also:  Documentation/RCU/torture.rst
  11  */
  12
  13 #define pr_fmt(fmt) fmt
  14
  15 #include <linux/types.h>
  16 #include <linux/kernel.h>
  17 #include <linux/init.h>
  18 #include <linux/module.h>
  19 #include <linux/kthread.h>
  20 #include <linux/err.h>
  21 #include <linux/spinlock.h>
  22 #include <linux/smp.h>
  23 #include <linux/rcupdate_wait.h>
  24 #include <linux/rcu_notifier.h>
  25 #include <linux/interrupt.h>
  26 #include <linux/sched/signal.h>
  27 #include <uapi/linux/sched/types.h>
  28 #include <linux/atomic.h>
  29 #include <linux/bitops.h>
  30 #include <linux/completion.h>
  31 #include <linux/moduleparam.h>
  32 #include <linux/percpu.h>
  33 #include <linux/notifier.h>
  34 #include <linux/reboot.h>
  35 #include <linux/freezer.h>
  36 #include <linux/cpu.h>
  37 #include <linux/delay.h>
  38 #include <linux/stat.h>
  39 #include <linux/srcu.h>
  40 #include <linux/slab.h>
  41 #include <linux/trace_clock.h>
  42 #include <asm/byteorder.h>
  43 #include <linux/torture.h>
  44 #include <linux/vmalloc.h>
  45 #include <linux/sched/debug.h>
  46 #include <linux/sched/sysctl.h>
  47 #include <linux/oom.h>
  48 #include <linux/tick.h>
  49 #include <linux/rcupdate_trace.h>
  50 #include <linux/nmi.h>
  51
  52 #include "rcu.h"
  53
  54 MODULE_DESCRIPTION("Read-Copy Update module-based torture test facility");
  55 MODULE_LICENSE("GPL");
  56 MODULE_AUTHOR("Paul E. McKenney <[email protected]> and Josh Triplett <[email protected]>");
  57
  58 /* Bits for ->extendables field, extendables param, and related definitions. */
  59 #define RCUTORTURE_RDR_SHIFT_1   8      /* Put SRCU index in upper bits. */
  60 #define RCUTORTURE_RDR_MASK_1    (0xff << RCUTORTURE_RDR_SHIFT_1)
  61 #define RCUTORTURE_RDR_SHIFT_2   16     /* Put SRCU index in upper bits. */
  62 #define RCUTORTURE_RDR_MASK_2    (0xff << RCUTORTURE_RDR_SHIFT_2)
  63 #define RCUTORTURE_RDR_BH        0x01   /* Extend readers by disabling bh. */
  64 #define RCUTORTURE_RDR_IRQ       0x02   /*  ... disabling interrupts. */
  65 #define RCUTORTURE_RDR_PREEMPT   0x04   /*  ... disabling preemption. */
  66 #define RCUTORTURE_RDR_RBH       0x08   /*  ... rcu_read_lock_bh(). */
  67 #define RCUTORTURE_RDR_SCHED     0x10   /*  ... rcu_read_lock_sched(). */
  68 #define RCUTORTURE_RDR_RCU_1     0x20   /*  ... entering another RCU reader. */
  69 #define RCUTORTURE_RDR_RCU_2     0x40   /*  ... entering another RCU reader. */
  70 #define RCUTORTURE_RDR_NBITS     7      /* Number of bits defined above. */
  71 #define RCUTORTURE_MAX_EXTEND    \
  72         (RCUTORTURE_RDR_BH | RCUTORTURE_RDR_IRQ | RCUTORTURE_RDR_PREEMPT | \
  73          RCUTORTURE_RDR_RBH | RCUTORTURE_RDR_SCHED)
  74 #define RCUTORTURE_RDR_ALLBITS  \
  75         (RCUTORTURE_MAX_EXTEND | RCUTORTURE_RDR_RCU_1 | RCUTORTURE_RDR_RCU_2 | \
  76          RCUTORTURE_RDR_MASK_1 | RCUTORTURE_RDR_MASK_2)
  77 #define RCUTORTURE_RDR_MAX_LOOPS 0x7    /* Maximum reader extensions. */
  78                                         /* Must be power of two minus one. */
  79 #define RCUTORTURE_RDR_MAX_SEGS (RCUTORTURE_RDR_MAX_LOOPS + 3)
  80
  81 torture_param(int, extendables, RCUTORTURE_MAX_EXTEND,
  82               "Extend readers by disabling bh (1), irqs (2), or preempt (4)");
  83 torture_param(int, fqs_duration, 0, "Duration of fqs bursts (us), 0 to disable");
  84 torture_param(int, fqs_holdoff, 0, "Holdoff time within fqs bursts (us)");
  85 torture_param(int, fqs_stutter, 3, "Wait time between fqs bursts (s)");
  86 torture_param(int, fwd_progress, 1, "Number of grace-period forward progress tasks (0 to disable)");
  87 torture_param(int, fwd_progress_div, 4, "Fraction of CPU stall to wait");
  88 torture_param(int, fwd_progress_holdoff, 60, "Time between forward-progress tests (s)");
  89 torture_param(bool, fwd_progress_need_resched, 1, "Hide cond_resched() behind need_resched()");
  90 torture_param(bool, gp_cond, false, "Use conditional/async GP wait primitives");
  91 torture_param(bool, gp_cond_exp, false, "Use conditional/async expedited GP wait primitives");
  92 torture_param(bool, gp_cond_full, false, "Use conditional/async full-state GP wait primitives");
  93 torture_param(bool, gp_cond_exp_full, false,
  94                     "Use conditional/async full-stateexpedited GP wait primitives");
  95 torture_param(bool, gp_exp, false, "Use expedited GP wait primitives");
  96 torture_param(bool, gp_normal, false, "Use normal (non-expedited) GP wait primitives");
  97 torture_param(bool, gp_poll, false, "Use polling GP wait primitives");
  98 torture_param(bool, gp_poll_exp, false, "Use polling expedited GP wait primitives");
  99 torture_param(bool, gp_poll_full, false, "Use polling full-state GP wait primitives");
 100 torture_param(bool, gp_poll_exp_full, false, "Use polling full-state expedited GP wait primitives");
 101 torture_param(bool, gp_sync, false, "Use synchronous GP wait primitives");
 102 torture_param(int, irqreader, 1, "Allow RCU readers from irq handlers");
 103 torture_param(int, leakpointer, 0, "Leak pointer dereferences from readers");
 104 torture_param(int, n_barrier_cbs, 0, "# of callbacks/kthreads for barrier testing");
 105 torture_param(int, nfakewriters, 4, "Number of RCU fake writer threads");
 106 torture_param(int, nreaders, -1, "Number of RCU reader threads");
 107 torture_param(int, object_debug, 0, "Enable debug-object double call_rcu() testing");
 108 torture_param(int, onoff_holdoff, 0, "Time after boot before CPU hotplugs (s)");
 109 torture_param(int, onoff_interval, 0, "Time between CPU hotplugs (jiffies), 0=disable");
 110 torture_param(int, nocbs_nthreads, 0, "Number of NOCB toggle threads, 0 to disable");
 111 torture_param(int, nocbs_toggle, 1000, "Time between toggling nocb state (ms)");
 112 torture_param(int, read_exit_delay, 13, "Delay between read-then-exit episodes (s)");
 113 torture_param(int, read_exit_burst, 16, "# of read-then-exit bursts per episode, zero to disable");
 114 torture_param(int, reader_flavor, 0x1, "Reader flavors to use, one per bit.");
 115 torture_param(int, shuffle_interval, 3, "Number of seconds between shuffles");
 116 torture_param(int, shutdown_secs, 0, "Shutdown time (s), <= zero to disable.");
 117 torture_param(int, stall_cpu, 0, "Stall duration (s), zero to disable.");
 118 torture_param(int, stall_cpu_holdoff, 10, "Time to wait before starting stall (s).");
 119 torture_param(bool, stall_no_softlockup, false, "Avoid softlockup warning during cpu stall.");
 120 torture_param(int, stall_cpu_irqsoff, 0, "Disable interrupts while stalling.");
 121 torture_param(int, stall_cpu_block, 0, "Sleep while stalling.");
 122 torture_param(int, stall_cpu_repeat, 0, "Number of additional stalls after the first one.");
 123 torture_param(int, stall_gp_kthread, 0, "Grace-period kthread stall duration (s).");
 124 torture_param(int, stat_interval, 60, "Number of seconds between stats printk()s");
 125 torture_param(int, stutter, 5, "Number of seconds to run/halt test");
 126 torture_param(int, test_boost, 1, "Test RCU prio boost: 0=no, 1=maybe, 2=yes.");
 127 torture_param(int, test_boost_duration, 4, "Duration of each boost test, seconds.");
 128 torture_param(int, test_boost_interval, 7, "Interval between boost tests, seconds.");
 129 torture_param(int, test_nmis, 0, "End-test NMI tests, 0 to disable.");
 130 torture_param(bool, test_no_idle_hz, true, "Test support for tickless idle CPUs");
 131 torture_param(int, test_srcu_lockdep, 0, "Test specified SRCU deadlock scenario.");
 132 torture_param(int, verbose, 1, "Enable verbose debugging printk()s");
 133
 134 static char *torture_type = "rcu";
 135 module_param(torture_type, charp, 0444);
 136 MODULE_PARM_DESC(torture_type, "Type of RCU to torture (rcu, srcu, ...)");
 137
 138 static int nrealnocbers;
 139 static int nrealreaders;
 140 static struct task_struct *writer_task;
 141 static struct task_struct **fakewriter_tasks;
 142 static struct task_struct **reader_tasks;
 143 static struct task_struct **nocb_tasks;
 144 static struct task_struct *stats_task;
 145 static struct task_struct *fqs_task;
 146 static struct task_struct *boost_tasks[NR_CPUS];
 147 static struct task_struct *stall_task;
 148 static struct task_struct **fwd_prog_tasks;
 149 static struct task_struct **barrier_cbs_tasks;
 150 static struct task_struct *barrier_task;
 151 static struct task_struct *read_exit_task;
 152
 153 #define RCU_TORTURE_PIPE_LEN 10
 154
 155 // Mailbox-like structure to check RCU global memory ordering.
 156 struct rcu_torture_reader_check {
 157         unsigned long rtc_myloops;
 158         int rtc_chkrdr;
 159         unsigned long rtc_chkloops;
 160         int rtc_ready;
 161         struct rcu_torture_reader_check *rtc_assigner;
 162 } ____cacheline_internodealigned_in_smp;
 163
 164 // Update-side data structure used to check RCU readers.
 165 struct rcu_torture {
 166         struct rcu_head rtort_rcu;
 167         int rtort_pipe_count;
 168         struct list_head rtort_free;
 169         int rtort_mbtest;
 170         struct rcu_torture_reader_check *rtort_chkp;
 171 };
 172
 173 static LIST_HEAD(rcu_torture_freelist);
 174 static struct rcu_torture __rcu *rcu_torture_current;
 175 static unsigned long rcu_torture_current_version;
 176 static struct rcu_torture rcu_tortures[10 * RCU_TORTURE_PIPE_LEN];
 177 static DEFINE_SPINLOCK(rcu_torture_lock);
 178 static DEFINE_PER_CPU(long [RCU_TORTURE_PIPE_LEN + 1], rcu_torture_count);
 179 static DEFINE_PER_CPU(long [RCU_TORTURE_PIPE_LEN + 1], rcu_torture_batch);
 180 static atomic_t rcu_torture_wcount[RCU_TORTURE_PIPE_LEN + 1];
 181 static struct rcu_torture_reader_check *rcu_torture_reader_mbchk;
 182 static atomic_t n_rcu_torture_alloc;
 183 static atomic_t n_rcu_torture_alloc_fail;
 184 static atomic_t n_rcu_torture_free;
 185 static atomic_t n_rcu_torture_mberror;
 186 static atomic_t n_rcu_torture_mbchk_fail;
 187 static atomic_t n_rcu_torture_mbchk_tries;
 188 static atomic_t n_rcu_torture_error;
 189 static long n_rcu_torture_barrier_error;
 190 static long n_rcu_torture_boost_ktrerror;
 191 static long n_rcu_torture_boost_failure;
 192 static long n_rcu_torture_boosts;
 193 static atomic_long_t n_rcu_torture_timers;
 194 static long n_barrier_attempts;
 195 static long n_barrier_successes; /* did rcu_barrier test succeed? */
 196 static unsigned long n_read_exits;
 197 static struct list_head rcu_torture_removed;
 198 static unsigned long shutdown_jiffies;
 199 static unsigned long start_gp_seq;
 200 static atomic_long_t n_nocb_offload;
 201 static atomic_long_t n_nocb_deoffload;
 202
 203 static int rcu_torture_writer_state;
 204 #define RTWS_FIXED_DELAY        0
 205 #define RTWS_DELAY              1
 206 #define RTWS_REPLACE            2
 207 #define RTWS_DEF_FREE           3
 208 #define RTWS_EXP_SYNC           4
 209 #define RTWS_COND_GET           5
 210 #define RTWS_COND_GET_FULL      6
 211 #define RTWS_COND_GET_EXP       7
 212 #define RTWS_COND_GET_EXP_FULL  8
 213 #define RTWS_COND_SYNC          9
 214 #define RTWS_COND_SYNC_FULL     10
 215 #define RTWS_COND_SYNC_EXP      11
 216 #define RTWS_COND_SYNC_EXP_FULL 12
 217 #define RTWS_POLL_GET           13
 218 #define RTWS_POLL_GET_FULL      14
 219 #define RTWS_POLL_GET_EXP       15
 220 #define RTWS_POLL_GET_EXP_FULL  16
 221 #define RTWS_POLL_WAIT          17
 222 #define RTWS_POLL_WAIT_FULL     18
 223 #define RTWS_POLL_WAIT_EXP      19
 224 #define RTWS_POLL_WAIT_EXP_FULL 20
 225 #define RTWS_SYNC               21
 226 #define RTWS_STUTTER            22
 227 #define RTWS_STOPPING           23
 228 static const char * const rcu_torture_writer_state_names[] = {
 229         "RTWS_FIXED_DELAY",
 230         "RTWS_DELAY",
 231         "RTWS_REPLACE",
 232         "RTWS_DEF_FREE",
 233         "RTWS_EXP_SYNC",
 234         "RTWS_COND_GET",
 235         "RTWS_COND_GET_FULL",
 236         "RTWS_COND_GET_EXP",
 237         "RTWS_COND_GET_EXP_FULL",
 238         "RTWS_COND_SYNC",
 239         "RTWS_COND_SYNC_FULL",
 240         "RTWS_COND_SYNC_EXP",
 241         "RTWS_COND_SYNC_EXP_FULL",
 242         "RTWS_POLL_GET",
 243         "RTWS_POLL_GET_FULL",
 244         "RTWS_POLL_GET_EXP",
 245         "RTWS_POLL_GET_EXP_FULL",
 246         "RTWS_POLL_WAIT",
 247         "RTWS_POLL_WAIT_FULL",
 248         "RTWS_POLL_WAIT_EXP",
 249         "RTWS_POLL_WAIT_EXP_FULL",
 250         "RTWS_SYNC",
 251         "RTWS_STUTTER",
 252         "RTWS_STOPPING",
 253 };
 254
 255 /* Record reader segment types and duration for first failing read. */
 256 struct rt_read_seg {
 257         int rt_readstate;
 258         unsigned long rt_delay_jiffies;
 259         unsigned long rt_delay_ms;
 260         unsigned long rt_delay_us;
 261         bool rt_preempted;
 262 };
 263 static int err_segs_recorded;
 264 static struct rt_read_seg err_segs[RCUTORTURE_RDR_MAX_SEGS];
 265 static int rt_read_nsegs;
 266
 267 static const char *rcu_torture_writer_state_getname(void)
 268 {
 269         unsigned int i = READ_ONCE(rcu_torture_writer_state);
 270
 271         if (i >= ARRAY_SIZE(rcu_torture_writer_state_names))
 272                 return "???";
 273         return rcu_torture_writer_state_names[i];
 274 }
 275
 276 #ifdef CONFIG_RCU_TRACE
 277 static u64 notrace rcu_trace_clock_local(void)
 278 {
 279         u64 ts = trace_clock_local();
 280
 281         (void)do_div(ts, NSEC_PER_USEC);
 282         return ts;
 283 }
 284 #else /* #ifdef CONFIG_RCU_TRACE */
 285 static u64 notrace rcu_trace_clock_local(void)
 286 {
 287         return 0ULL;
 288 }
 289 #endif /* #else #ifdef CONFIG_RCU_TRACE */
 290
 291 /*
 292  * Stop aggressive CPU-hog tests a bit before the end of the test in order
 293  * to avoid interfering with test shutdown.
 294  */
 295 static bool shutdown_time_arrived(void)
 296 {
 297         return shutdown_secs && time_after(jiffies, shutdown_jiffies - 30 * HZ);
 298 }
 299
 300 static unsigned long boost_starttime;   /* jiffies of next boost test start. */
 301 static DEFINE_MUTEX(boost_mutex);       /* protect setting boost_starttime */
 302                                         /*  and boost task create/destroy. */
 303 static atomic_t barrier_cbs_count;      /* Barrier callbacks registered. */
 304 static bool barrier_phase;              /* Test phase. */
 305 static atomic_t barrier_cbs_invoked;    /* Barrier callbacks invoked. */
 306 static wait_queue_head_t *barrier_cbs_wq; /* Coordinate barrier testing. */
 307 static DECLARE_WAIT_QUEUE_HEAD(barrier_wq);
 308
 309 static atomic_t rcu_fwd_cb_nodelay;     /* Short rcu_torture_delay() delays. */
 310
 311 /*
 312  * Allocate an element from the rcu_tortures pool.
 313  */
 314 static struct rcu_torture *
 315 rcu_torture_alloc(void)
 316 {
 317         struct list_head *p;
 318
 319         spin_lock_bh(&rcu_torture_lock);
 320         if (list_empty(&rcu_torture_freelist)) {
 321                 atomic_inc(&n_rcu_torture_alloc_fail);
 322                 spin_unlock_bh(&rcu_torture_lock);
 323                 return NULL;
 324         }
 325         atomic_inc(&n_rcu_torture_alloc);
 326         p = rcu_torture_freelist.next;
 327         list_del_init(p);
 328         spin_unlock_bh(&rcu_torture_lock);
 329         return container_of(p, struct rcu_torture, rtort_free);
 330 }
 331
 332 /*
 333  * Free an element to the rcu_tortures pool.
 334  */
 335 static void
 336 rcu_torture_free(struct rcu_torture *p)
 337 {
 338         atomic_inc(&n_rcu_torture_free);
 339         spin_lock_bh(&rcu_torture_lock);
 340         list_add_tail(&p->rtort_free, &rcu_torture_freelist);
 341         spin_unlock_bh(&rcu_torture_lock);
 342 }
 343
 344 /*
 345  * Operations vector for selecting different types of tests.
 346  */
 347
 348 struct rcu_torture_ops {
 349         int ttype;
 350         void (*init)(void);
 351         void (*cleanup)(void);
 352         int (*readlock)(void);
 353         void (*read_delay)(struct torture_random_state *rrsp,
 354                            struct rt_read_seg *rtrsp);
 355         void (*readunlock)(int idx);
 356         int (*readlock_held)(void);
 357         unsigned long (*get_gp_seq)(void);
 358         unsigned long (*gp_diff)(unsigned long new, unsigned long old);
 359         void (*deferred_free)(struct rcu_torture *p);
 360         void (*sync)(void);
 361         void (*exp_sync)(void);
 362         unsigned long (*get_gp_state_exp)(void);
 363         unsigned long (*start_gp_poll_exp)(void);
 364         void (*start_gp_poll_exp_full)(struct rcu_gp_oldstate *rgosp);
 365         bool (*poll_gp_state_exp)(unsigned long oldstate);
 366         void (*cond_sync_exp)(unsigned long oldstate);
 367         void (*cond_sync_exp_full)(struct rcu_gp_oldstate *rgosp);
 368         unsigned long (*get_comp_state)(void);
 369         void (*get_comp_state_full)(struct rcu_gp_oldstate *rgosp);
 370         bool (*same_gp_state)(unsigned long oldstate1, unsigned long oldstate2);
 371         bool (*same_gp_state_full)(struct rcu_gp_oldstate *rgosp1, struct rcu_gp_oldstate *rgosp2);
 372         unsigned long (*get_gp_state)(void);
 373         void (*get_gp_state_full)(struct rcu_gp_oldstate *rgosp);
 374         unsigned long (*start_gp_poll)(void);
 375         void (*start_gp_poll_full)(struct rcu_gp_oldstate *rgosp);
 376         bool (*poll_gp_state)(unsigned long oldstate);
 377         bool (*poll_gp_state_full)(struct rcu_gp_oldstate *rgosp);
 378         bool (*poll_need_2gp)(bool poll, bool poll_full);
 379         void (*cond_sync)(unsigned long oldstate);
 380         void (*cond_sync_full)(struct rcu_gp_oldstate *rgosp);
 381         int poll_active;
 382         int poll_active_full;
 383         call_rcu_func_t call;
 384         void (*cb_barrier)(void);
 385         void (*fqs)(void);
 386         void (*stats)(void);
 387         void (*gp_kthread_dbg)(void);
 388         bool (*check_boost_failed)(unsigned long gp_state, int *cpup);
 389         int (*stall_dur)(void);
 390         void (*get_gp_data)(int *flags, unsigned long *gp_seq);
 391         void (*gp_slow_register)(atomic_t *rgssp);
 392         void (*gp_slow_unregister)(atomic_t *rgssp);
 393         long cbflood_max;
 394         int irq_capable;
 395         int can_boost;
 396         int extendables;
 397         int slow_gps;
 398         int no_pi_lock;
 399         int debug_objects;
 400         int start_poll_irqsoff;
 401         const char *name;
 402 };
 403
 404 static struct rcu_torture_ops *cur_ops;
 405
 406 /*
 407  * Definitions for rcu torture testing.
 408  */
 409
 410 static int torture_readlock_not_held(void)
 411 {
 412         return rcu_read_lock_bh_held() || rcu_read_lock_sched_held();
 413 }
 414
 415 static int rcu_torture_read_lock(void)
 416 {
 417         rcu_read_lock();
 418         return 0;
 419 }
 420
 421 static void
 422 rcu_read_delay(struct torture_random_state *rrsp, struct rt_read_seg *rtrsp)
 423 {
 424         unsigned long started;
 425         unsigned long completed;
 426         const unsigned long shortdelay_us = 200;
 427         unsigned long longdelay_ms = 300;
 428         unsigned long long ts;
 429
 430         /* We want a short delay sometimes to make a reader delay the grace
 431          * period, and we want a long delay occasionally to trigger
 432          * force_quiescent_state. */
 433
 434         if (!atomic_read(&rcu_fwd_cb_nodelay) &&
 435             !(torture_random(rrsp) % (nrealreaders * 2000 * longdelay_ms))) {
 436                 started = cur_ops->get_gp_seq();
 437                 ts = rcu_trace_clock_local();
 438                 if (preempt_count() & (SOFTIRQ_MASK | HARDIRQ_MASK))
 439                         longdelay_ms = 5; /* Avoid triggering BH limits. */
 440                 mdelay(longdelay_ms);
 441                 rtrsp->rt_delay_ms = longdelay_ms;
 442                 completed = cur_ops->get_gp_seq();
 443                 do_trace_rcu_torture_read(cur_ops->name, NULL, ts,
 444                                           started, completed);
 445         }
 446         if (!(torture_random(rrsp) % (nrealreaders * 2 * shortdelay_us))) {
 447                 udelay(shortdelay_us);
 448                 rtrsp->rt_delay_us = shortdelay_us;
 449         }
 450         if (!preempt_count() &&
 451             !(torture_random(rrsp) % (nrealreaders * 500))) {
 452                 torture_preempt_schedule();  /* QS only if preemptible. */
 453                 rtrsp->rt_preempted = true;
 454         }
 455 }
 456
 457 static void rcu_torture_read_unlock(int idx)
 458 {
 459         rcu_read_unlock();
 460 }
 461
 462 /*
 463  * Update callback in the pipe.  This should be invoked after a grace period.
 464  */
 465 static bool
 466 rcu_torture_pipe_update_one(struct rcu_torture *rp)
 467 {
 468         int i;
 469         struct rcu_torture_reader_check *rtrcp = READ_ONCE(rp->rtort_chkp);
 470
 471         if (rtrcp) {
 472                 WRITE_ONCE(rp->rtort_chkp, NULL);
 473                 smp_store_release(&rtrcp->rtc_ready, 1); // Pair with smp_load_acquire().
 474         }
 475         i = rp->rtort_pipe_count;
 476         if (i > RCU_TORTURE_PIPE_LEN)
 477                 i = RCU_TORTURE_PIPE_LEN;
 478         atomic_inc(&rcu_torture_wcount[i]);
 479         WRITE_ONCE(rp->rtort_pipe_count, i + 1);
 480         ASSERT_EXCLUSIVE_WRITER(rp->rtort_pipe_count);
 481         if (i + 1 >= RCU_TORTURE_PIPE_LEN) {
 482                 rp->rtort_mbtest = 0;
 483                 return true;
 484         }
 485         return false;
 486 }
 487
 488 /*
 489  * Update all callbacks in the pipe.  Suitable for synchronous grace-period
 490  * primitives.
 491  */
 492 static void
 493 rcu_torture_pipe_update(struct rcu_torture *old_rp)
 494 {
 495         struct rcu_torture *rp;
 496         struct rcu_torture *rp1;
 497
 498         if (old_rp)
 499                 list_add(&old_rp->rtort_free, &rcu_torture_removed);
 500         list_for_each_entry_safe(rp, rp1, &rcu_torture_removed, rtort_free) {
 501                 if (rcu_torture_pipe_update_one(rp)) {
 502                         list_del(&rp->rtort_free);
 503                         rcu_torture_free(rp);
 504                 }
 505         }
 506 }
 507
 508 static void
 509 rcu_torture_cb(struct rcu_head *p)
 510 {
 511         struct rcu_torture *rp = container_of(p, struct rcu_torture, rtort_rcu);
 512
 513         if (torture_must_stop_irq()) {
 514                 /* Test is ending, just drop callbacks on the floor. */
 515                 /* The next initialization will pick up the pieces. */
 516                 return;
 517         }
 518         if (rcu_torture_pipe_update_one(rp))
 519                 rcu_torture_free(rp);
 520         else
 521                 cur_ops->deferred_free(rp);
 522 }
 523
 524 static unsigned long rcu_no_completed(void)
 525 {
 526         return 0;
 527 }
 528
 529 static void rcu_torture_deferred_free(struct rcu_torture *p)
 530 {
 531         call_rcu_hurry(&p->rtort_rcu, rcu_torture_cb);
 532 }
 533
 534 static void rcu_sync_torture_init(void)
 535 {
 536         INIT_LIST_HEAD(&rcu_torture_removed);
 537 }
 538
 539 static bool rcu_poll_need_2gp(bool poll, bool poll_full)
 540 {
 541         return poll;
 542 }
 543
 544 static struct rcu_torture_ops rcu_ops = {
 545         .ttype                  = RCU_FLAVOR,
 546         .init                   = rcu_sync_torture_init,
 547         .readlock               = rcu_torture_read_lock,
 548         .read_delay             = rcu_read_delay,
 549         .readunlock             = rcu_torture_read_unlock,
 550         .readlock_held          = torture_readlock_not_held,
 551         .get_gp_seq             = rcu_get_gp_seq,
 552         .gp_diff                = rcu_seq_diff,
 553         .deferred_free          = rcu_torture_deferred_free,
 554         .sync                   = synchronize_rcu,
 555         .exp_sync               = synchronize_rcu_expedited,
 556         .same_gp_state          = same_state_synchronize_rcu,
 557         .same_gp_state_full     = same_state_synchronize_rcu_full,
 558         .get_comp_state         = get_completed_synchronize_rcu,
 559         .get_comp_state_full    = get_completed_synchronize_rcu_full,
 560         .get_gp_state           = get_state_synchronize_rcu,
 561         .get_gp_state_full      = get_state_synchronize_rcu_full,
 562         .start_gp_poll          = start_poll_synchronize_rcu,
 563         .start_gp_poll_full     = start_poll_synchronize_rcu_full,
 564         .poll_gp_state          = poll_state_synchronize_rcu,
 565         .poll_gp_state_full     = poll_state_synchronize_rcu_full,
 566         .poll_need_2gp          = rcu_poll_need_2gp,
 567         .cond_sync              = cond_synchronize_rcu,
 568         .cond_sync_full         = cond_synchronize_rcu_full,
 569         .poll_active            = NUM_ACTIVE_RCU_POLL_OLDSTATE,
 570         .poll_active_full       = NUM_ACTIVE_RCU_POLL_FULL_OLDSTATE,
 571         .get_gp_state_exp       = get_state_synchronize_rcu,
 572         .start_gp_poll_exp      = start_poll_synchronize_rcu_expedited,
 573         .start_gp_poll_exp_full = start_poll_synchronize_rcu_expedited_full,
 574         .poll_gp_state_exp      = poll_state_synchronize_rcu,
 575         .cond_sync_exp          = cond_synchronize_rcu_expedited,
 576         .call                   = call_rcu_hurry,
 577         .cb_barrier             = rcu_barrier,
 578         .fqs                    = rcu_force_quiescent_state,
 579         .gp_kthread_dbg         = show_rcu_gp_kthreads,
 580         .check_boost_failed     = rcu_check_boost_fail,
 581         .stall_dur              = rcu_jiffies_till_stall_check,
 582         .get_gp_data            = rcutorture_get_gp_data,
 583         .gp_slow_register       = rcu_gp_slow_register,
 584         .gp_slow_unregister     = rcu_gp_slow_unregister,
 585         .irq_capable            = 1,
 586         .can_boost              = IS_ENABLED(CONFIG_RCU_BOOST),
 587         .extendables            = RCUTORTURE_MAX_EXTEND,
 588         .debug_objects          = 1,
 589         .start_poll_irqsoff     = 1,
 590         .name                   = "rcu"
 591 };
 592
 593 /*
 594  * Don't even think about trying any of these in real life!!!
 595  * The names includes "busted", and they really means it!
 596  * The only purpose of these functions is to provide a buggy RCU
 597  * implementation to make sure that rcutorture correctly emits
 598  * buggy-RCU error messages.
 599  */
 600 static void rcu_busted_torture_deferred_free(struct rcu_torture *p)
 601 {
 602         /* This is a deliberate bug for testing purposes only! */
 603         rcu_torture_cb(&p->rtort_rcu);
 604 }
 605
 606 static void synchronize_rcu_busted(void)
 607 {
 608         /* This is a deliberate bug for testing purposes only! */
 609 }
 610
 611 static void
 612 call_rcu_busted(struct rcu_head *head, rcu_callback_t func)
 613 {
 614         /* This is a deliberate bug for testing purposes only! */
 615         func(head);
 616 }
 617
 618 static struct rcu_torture_ops rcu_busted_ops = {
 619         .ttype          = INVALID_RCU_FLAVOR,
 620         .init           = rcu_sync_torture_init,
 621         .readlock       = rcu_torture_read_lock,
 622         .read_delay     = rcu_read_delay,  /* just reuse rcu's version. */
 623         .readunlock     = rcu_torture_read_unlock,
 624         .readlock_held  = torture_readlock_not_held,
 625         .get_gp_seq     = rcu_no_completed,
 626         .deferred_free  = rcu_busted_torture_deferred_free,
 627         .sync           = synchronize_rcu_busted,
 628         .exp_sync       = synchronize_rcu_busted,
 629         .call           = call_rcu_busted,
 630         .irq_capable    = 1,
 631         .name           = "busted"
 632 };
 633
 634 /*
 635  * Definitions for srcu torture testing.
 636  */
 637
 638 DEFINE_STATIC_SRCU(srcu_ctl);
 639 static struct srcu_struct srcu_ctld;
 640 static struct srcu_struct *srcu_ctlp = &srcu_ctl;
 641 static struct rcu_torture_ops srcud_ops;
 642
 643 static void srcu_get_gp_data(int *flags, unsigned long *gp_seq)
 644 {
 645         srcutorture_get_gp_data(srcu_ctlp, flags, gp_seq);
 646 }
 647
 648 static int srcu_torture_read_lock(void)
 649 {
 650         int idx;
 651         int ret = 0;
 652
 653         if ((reader_flavor & 0x1) || !(reader_flavor & 0x7)) {
 654                 idx = srcu_read_lock(srcu_ctlp);
 655                 WARN_ON_ONCE(idx & ~0x1);
 656                 ret += idx;
 657         }
 658         if (reader_flavor & 0x2) {
 659                 idx = srcu_read_lock_nmisafe(srcu_ctlp);
 660                 WARN_ON_ONCE(idx & ~0x1);
 661                 ret += idx << 1;
 662         }
 663         if (reader_flavor & 0x4) {
 664                 idx = srcu_read_lock_lite(srcu_ctlp);
 665                 WARN_ON_ONCE(idx & ~0x1);
 666                 ret += idx << 2;
 667         }
 668         return ret;
 669 }
 670
 671 static void
 672 srcu_read_delay(struct torture_random_state *rrsp, struct rt_read_seg *rtrsp)
 673 {
 674         long delay;
 675         const long uspertick = 1000000 / HZ;
 676         const long longdelay = 10;
 677
 678         /* We want there to be long-running readers, but not all the time. */
 679
 680         delay = torture_random(rrsp) %
 681                 (nrealreaders * 2 * longdelay * uspertick);
 682         if (!delay && in_task()) {
 683                 schedule_timeout_interruptible(longdelay);
 684                 rtrsp->rt_delay_jiffies = longdelay;
 685         } else {
 686                 rcu_read_delay(rrsp, rtrsp);
 687         }
 688 }
 689
 690 static void srcu_torture_read_unlock(int idx)
 691 {
 692         WARN_ON_ONCE((reader_flavor && (idx & ~reader_flavor)) || (!reader_flavor && (idx & ~0x1)));
 693         if (reader_flavor & 0x4)
 694                 srcu_read_unlock_lite(srcu_ctlp, (idx & 0x4) >> 2);
 695         if (reader_flavor & 0x2)
 696                 srcu_read_unlock_nmisafe(srcu_ctlp, (idx & 0x2) >> 1);
 697         if ((reader_flavor & 0x1) || !(reader_flavor & 0x7))
 698                 srcu_read_unlock(srcu_ctlp, idx & 0x1);
 699 }
 700
 701 static int torture_srcu_read_lock_held(void)
 702 {
 703         return srcu_read_lock_held(srcu_ctlp);
 704 }
 705
 706 static unsigned long srcu_torture_completed(void)
 707 {
 708         return srcu_batches_completed(srcu_ctlp);
 709 }
 710
 711 static void srcu_torture_deferred_free(struct rcu_torture *rp)
 712 {
 713         call_srcu(srcu_ctlp, &rp->rtort_rcu, rcu_torture_cb);
 714 }
 715
 716 static void srcu_torture_synchronize(void)
 717 {
 718         synchronize_srcu(srcu_ctlp);
 719 }
 720
 721 static unsigned long srcu_torture_get_gp_state(void)
 722 {
 723         return get_state_synchronize_srcu(srcu_ctlp);
 724 }
 725
 726 static unsigned long srcu_torture_start_gp_poll(void)
 727 {
 728         return start_poll_synchronize_srcu(srcu_ctlp);
 729 }
 730
 731 static bool srcu_torture_poll_gp_state(unsigned long oldstate)
 732 {
 733         return poll_state_synchronize_srcu(srcu_ctlp, oldstate);
 734 }
 735
 736 static void srcu_torture_call(struct rcu_head *head,
 737                               rcu_callback_t func)
 738 {
 739         call_srcu(srcu_ctlp, head, func);
 740 }
 741
 742 static void srcu_torture_barrier(void)
 743 {
 744         srcu_barrier(srcu_ctlp);
 745 }
 746
 747 static void srcu_torture_stats(void)
 748 {
 749         srcu_torture_stats_print(srcu_ctlp, torture_type, TORTURE_FLAG);
 750 }
 751
 752 static void srcu_torture_synchronize_expedited(void)
 753 {
 754         synchronize_srcu_expedited(srcu_ctlp);
 755 }
 756
 757 static struct rcu_torture_ops srcu_ops = {
 758         .ttype          = SRCU_FLAVOR,
 759         .init           = rcu_sync_torture_init,
 760         .readlock       = srcu_torture_read_lock,
 761         .read_delay     = srcu_read_delay,
 762         .readunlock     = srcu_torture_read_unlock,
 763         .readlock_held  = torture_srcu_read_lock_held,
 764         .get_gp_seq     = srcu_torture_completed,
 765         .deferred_free  = srcu_torture_deferred_free,
 766         .sync           = srcu_torture_synchronize,
 767         .exp_sync       = srcu_torture_synchronize_expedited,
 768         .same_gp_state  = same_state_synchronize_srcu,
 769         .get_comp_state = get_completed_synchronize_srcu,
 770         .get_gp_state   = srcu_torture_get_gp_state,
 771         .start_gp_poll  = srcu_torture_start_gp_poll,
 772         .poll_gp_state  = srcu_torture_poll_gp_state,
 773         .poll_active    = NUM_ACTIVE_SRCU_POLL_OLDSTATE,
 774         .call           = srcu_torture_call,
 775         .cb_barrier     = srcu_torture_barrier,
 776         .stats          = srcu_torture_stats,
 777         .get_gp_data    = srcu_get_gp_data,
 778         .cbflood_max    = 50000,
 779         .irq_capable    = 1,
 780         .no_pi_lock     = IS_ENABLED(CONFIG_TINY_SRCU),
 781         .debug_objects  = 1,
 782         .name           = "srcu"
 783 };
 784
 785 static void srcu_torture_init(void)
 786 {
 787         rcu_sync_torture_init();
 788         WARN_ON(init_srcu_struct(&srcu_ctld));
 789         srcu_ctlp = &srcu_ctld;
 790 }
 791
 792 static void srcu_torture_cleanup(void)
 793 {
 794         cleanup_srcu_struct(&srcu_ctld);
 795         srcu_ctlp = &srcu_ctl; /* In case of a later rcutorture run. */
 796 }
 797
 798 /* As above, but dynamically allocated. */
 799 static struct rcu_torture_ops srcud_ops = {
 800         .ttype          = SRCU_FLAVOR,
 801         .init           = srcu_torture_init,
 802         .cleanup        = srcu_torture_cleanup,
 803         .readlock       = srcu_torture_read_lock,
 804         .read_delay     = srcu_read_delay,
 805         .readunlock     = srcu_torture_read_unlock,
 806         .readlock_held  = torture_srcu_read_lock_held,
 807         .get_gp_seq     = srcu_torture_completed,
 808         .deferred_free  = srcu_torture_deferred_free,
 809         .sync           = srcu_torture_synchronize,
 810         .exp_sync       = srcu_torture_synchronize_expedited,
 811         .same_gp_state  = same_state_synchronize_srcu,
 812         .get_comp_state = get_completed_synchronize_srcu,
 813         .get_gp_state   = srcu_torture_get_gp_state,
 814         .start_gp_poll  = srcu_torture_start_gp_poll,
 815         .poll_gp_state  = srcu_torture_poll_gp_state,
 816         .poll_active    = NUM_ACTIVE_SRCU_POLL_OLDSTATE,
 817         .call           = srcu_torture_call,
 818         .cb_barrier     = srcu_torture_barrier,
 819         .stats          = srcu_torture_stats,
 820         .get_gp_data    = srcu_get_gp_data,
 821         .cbflood_max    = 50000,
 822         .irq_capable    = 1,
 823         .no_pi_lock     = IS_ENABLED(CONFIG_TINY_SRCU),
 824         .debug_objects  = 1,
 825         .name           = "srcud"
 826 };
 827
 828 /* As above, but broken due to inappropriate reader extension. */
 829 static struct rcu_torture_ops busted_srcud_ops = {
 830         .ttype          = SRCU_FLAVOR,
 831         .init           = srcu_torture_init,
 832         .cleanup        = srcu_torture_cleanup,
 833         .readlock       = srcu_torture_read_lock,
 834         .read_delay     = rcu_read_delay,
 835         .readunlock     = srcu_torture_read_unlock,
 836         .readlock_held  = torture_srcu_read_lock_held,
 837         .get_gp_seq     = srcu_torture_completed,
 838         .deferred_free  = srcu_torture_deferred_free,
 839         .sync           = srcu_torture_synchronize,
 840         .exp_sync       = srcu_torture_synchronize_expedited,
 841         .call           = srcu_torture_call,
 842         .cb_barrier     = srcu_torture_barrier,
 843         .stats          = srcu_torture_stats,
 844         .irq_capable    = 1,
 845         .no_pi_lock     = IS_ENABLED(CONFIG_TINY_SRCU),
 846         .extendables    = RCUTORTURE_MAX_EXTEND,
 847         .name           = "busted_srcud"
 848 };
 849
 850 /*
 851  * Definitions for trivial CONFIG_PREEMPT=n-only torture testing.
 852  * This implementation does not necessarily work well with CPU hotplug.
 853  */
 854
 855 static void synchronize_rcu_trivial(void)
 856 {
 857         int cpu;
 858
 859         for_each_online_cpu(cpu) {
 860                 torture_sched_setaffinity(current->pid, cpumask_of(cpu));
 861                 WARN_ON_ONCE(raw_smp_processor_id() != cpu);
 862         }
 863 }
 864
 865 static int rcu_torture_read_lock_trivial(void)
 866 {
 867         preempt_disable();
 868         return 0;
 869 }
 870
 871 static void rcu_torture_read_unlock_trivial(int idx)
 872 {
 873         preempt_enable();
 874 }
 875
 876 static struct rcu_torture_ops trivial_ops = {
 877         .ttype          = RCU_TRIVIAL_FLAVOR,
 878         .init           = rcu_sync_torture_init,
 879         .readlock       = rcu_torture_read_lock_trivial,
 880         .read_delay     = rcu_read_delay,  /* just reuse rcu's version. */
 881         .readunlock     = rcu_torture_read_unlock_trivial,
 882         .readlock_held  = torture_readlock_not_held,
 883         .get_gp_seq     = rcu_no_completed,
 884         .sync           = synchronize_rcu_trivial,
 885         .exp_sync       = synchronize_rcu_trivial,
 886         .irq_capable    = 1,
 887         .name           = "trivial"
 888 };
 889
 890 #ifdef CONFIG_TASKS_RCU
 891
 892 /*
 893  * Definitions for RCU-tasks torture testing.
 894  */
 895
 896 static int tasks_torture_read_lock(void)
 897 {
 898         return 0;
 899 }
 900
 901 static void tasks_torture_read_unlock(int idx)
 902 {
 903 }
 904
 905 static void rcu_tasks_torture_deferred_free(struct rcu_torture *p)
 906 {
 907         call_rcu_tasks(&p->rtort_rcu, rcu_torture_cb);
 908 }
 909
 910 static void synchronize_rcu_mult_test(void)
 911 {
 912         synchronize_rcu_mult(call_rcu_tasks, call_rcu_hurry);
 913 }
 914
 915 static struct rcu_torture_ops tasks_ops = {
 916         .ttype          = RCU_TASKS_FLAVOR,
 917         .init           = rcu_sync_torture_init,
 918         .readlock       = tasks_torture_read_lock,
 919         .read_delay     = rcu_read_delay,  /* just reuse rcu's version. */
 920         .readunlock     = tasks_torture_read_unlock,
 921         .get_gp_seq     = rcu_no_completed,
 922         .deferred_free  = rcu_tasks_torture_deferred_free,
 923         .sync           = synchronize_rcu_tasks,
 924         .exp_sync       = synchronize_rcu_mult_test,
 925         .call           = call_rcu_tasks,
 926         .cb_barrier     = rcu_barrier_tasks,
 927         .gp_kthread_dbg = show_rcu_tasks_classic_gp_kthread,
 928         .get_gp_data    = rcu_tasks_get_gp_data,
 929         .irq_capable    = 1,
 930         .slow_gps       = 1,
 931         .name           = "tasks"
 932 };
 933
 934 #define TASKS_OPS &tasks_ops,
 935
 936 #else // #ifdef CONFIG_TASKS_RCU
 937
 938 #define TASKS_OPS
 939
 940 #endif // #else #ifdef CONFIG_TASKS_RCU
 941
 942
 943 #ifdef CONFIG_TASKS_RUDE_RCU
 944
 945 /*
 946  * Definitions for rude RCU-tasks torture testing.
 947  */
 948
 949 static struct rcu_torture_ops tasks_rude_ops = {
 950         .ttype          = RCU_TASKS_RUDE_FLAVOR,
 951         .init           = rcu_sync_torture_init,
 952         .readlock       = rcu_torture_read_lock_trivial,
 953         .read_delay     = rcu_read_delay,  /* just reuse rcu's version. */
 954         .readunlock     = rcu_torture_read_unlock_trivial,
 955         .get_gp_seq     = rcu_no_completed,
 956         .sync           = synchronize_rcu_tasks_rude,
 957         .exp_sync       = synchronize_rcu_tasks_rude,
 958         .gp_kthread_dbg = show_rcu_tasks_rude_gp_kthread,
 959         .get_gp_data    = rcu_tasks_rude_get_gp_data,
 960         .cbflood_max    = 50000,
 961         .irq_capable    = 1,
 962         .name           = "tasks-rude"
 963 };
 964
 965 #define TASKS_RUDE_OPS &tasks_rude_ops,
 966
 967 #else // #ifdef CONFIG_TASKS_RUDE_RCU
 968
 969 #define TASKS_RUDE_OPS
 970
 971 #endif // #else #ifdef CONFIG_TASKS_RUDE_RCU
 972
 973
 974 #ifdef CONFIG_TASKS_TRACE_RCU
 975
 976 /*
 977  * Definitions for tracing RCU-tasks torture testing.
 978  */
 979
 980 static int tasks_tracing_torture_read_lock(void)
 981 {
 982         rcu_read_lock_trace();
 983         return 0;
 984 }
 985
 986 static void tasks_tracing_torture_read_unlock(int idx)
 987 {
 988         rcu_read_unlock_trace();
 989 }
 990
 991 static void rcu_tasks_tracing_torture_deferred_free(struct rcu_torture *p)
 992 {
 993         call_rcu_tasks_trace(&p->rtort_rcu, rcu_torture_cb);
 994 }
 995
 996 static struct rcu_torture_ops tasks_tracing_ops = {
 997         .ttype          = RCU_TASKS_TRACING_FLAVOR,
 998         .init           = rcu_sync_torture_init,
 999         .readlock       = tasks_tracing_torture_read_lock,
1000         .read_delay     = srcu_read_delay,  /* just reuse srcu's version. */
1001         .readunlock     = tasks_tracing_torture_read_unlock,
1002         .readlock_held  = rcu_read_lock_trace_held,
1003         .get_gp_seq     = rcu_no_completed,
1004         .deferred_free  = rcu_tasks_tracing_torture_deferred_free,
1005         .sync           = synchronize_rcu_tasks_trace,
1006         .exp_sync       = synchronize_rcu_tasks_trace,
1007         .call           = call_rcu_tasks_trace,
1008         .cb_barrier     = rcu_barrier_tasks_trace,
1009         .gp_kthread_dbg = show_rcu_tasks_trace_gp_kthread,
1010         .get_gp_data    = rcu_tasks_trace_get_gp_data,
1011         .cbflood_max    = 50000,
1012         .irq_capable    = 1,
1013         .slow_gps       = 1,
1014         .name           = "tasks-tracing"
1015 };
1016
1017 #define TASKS_TRACING_OPS &tasks_tracing_ops,
1018
1019 #else // #ifdef CONFIG_TASKS_TRACE_RCU
1020
1021 #define TASKS_TRACING_OPS
1022
1023 #endif // #else #ifdef CONFIG_TASKS_TRACE_RCU
1024
1025
1026 static unsigned long rcutorture_seq_diff(unsigned long new, unsigned long old)
1027 {
1028         if (!cur_ops->gp_diff)
1029                 return new - old;
1030         return cur_ops->gp_diff(new, old);
1031 }
1032
1033 /*
1034  * RCU torture priority-boost testing.  Runs one real-time thread per
1035  * CPU for moderate bursts, repeatedly starting grace periods and waiting
1036  * for them to complete.  If a given grace period takes too long, we assume
1037  * that priority inversion has occurred.
1038  */
1039
1040 static int old_rt_runtime = -1;
1041
1042 static void rcu_torture_disable_rt_throttle(void)
1043 {
1044         /*
1045          * Disable RT throttling so that rcutorture's boost threads don't get
1046          * throttled. Only possible if rcutorture is built-in otherwise the
1047          * user should manually do this by setting the sched_rt_period_us and
1048          * sched_rt_runtime sysctls.
1049          */
1050         if (!IS_BUILTIN(CONFIG_RCU_TORTURE_TEST) || old_rt_runtime != -1)
1051                 return;
1052
1053         old_rt_runtime = sysctl_sched_rt_runtime;
1054         sysctl_sched_rt_runtime = -1;
1055 }
1056
1057 static void rcu_torture_enable_rt_throttle(void)
1058 {
1059         if (!IS_BUILTIN(CONFIG_RCU_TORTURE_TEST) || old_rt_runtime == -1)
1060                 return;
1061
1062         sysctl_sched_rt_runtime = old_rt_runtime;
1063         old_rt_runtime = -1;
1064 }
1065
1066 static bool rcu_torture_boost_failed(unsigned long gp_state, unsigned long *start)
1067 {
1068         int cpu;
1069         static int dbg_done;
1070         unsigned long end = jiffies;
1071         bool gp_done;
1072         unsigned long j;
1073         static unsigned long last_persist;
1074         unsigned long lp;
1075         unsigned long mininterval = test_boost_duration * HZ - HZ / 2;
1076
1077         if (end - *start > mininterval) {
1078                 // Recheck after checking time to avoid false positives.
1079                 smp_mb(); // Time check before grace-period check.
1080                 if (cur_ops->poll_gp_state(gp_state))
1081                         return false; // passed, though perhaps just barely
1082                 if (cur_ops->check_boost_failed && !cur_ops->check_boost_failed(gp_state, &cpu)) {
1083                         // At most one persisted message per boost test.
1084                         j = jiffies;
1085                         lp = READ_ONCE(last_persist);
1086                         if (time_after(j, lp + mininterval) &&
1087                             cmpxchg(&last_persist, lp, j) == lp) {
1088                                 if (cpu < 0)
1089                                         pr_info("Boost inversion persisted: QS from all CPUs\n");
1090                                 else
1091                                         pr_info("Boost inversion persisted: No QS from CPU %d\n", cpu);
1092                         }
1093                         return false; // passed on a technicality
1094                 }
1095                 VERBOSE_TOROUT_STRING("rcu_torture_boost boosting failed");
1096                 n_rcu_torture_boost_failure++;
1097                 if (!xchg(&dbg_done, 1) && cur_ops->gp_kthread_dbg) {
1098                         pr_info("Boost inversion thread ->rt_priority %u gp_state %lu jiffies %lu\n",
1099                                 current->rt_priority, gp_state, end - *start);
1100                         cur_ops->gp_kthread_dbg();
1101                         // Recheck after print to flag grace period ending during splat.
1102                         gp_done = cur_ops->poll_gp_state(gp_state);
1103                         pr_info("Boost inversion: GP %lu %s.\n", gp_state,
1104                                 gp_done ? "ended already" : "still pending");
1105
1106                 }
1107
1108                 return true; // failed
1109         } else if (cur_ops->check_boost_failed && !cur_ops->check_boost_failed(gp_state, NULL)) {
1110                 *start = jiffies;
1111         }
1112
1113         return false; // passed
1114 }
1115
1116 static int rcu_torture_boost(void *arg)
1117 {
1118         unsigned long endtime;
1119         unsigned long gp_state;
1120         unsigned long gp_state_time;
1121         unsigned long oldstarttime;
1122
1123         VERBOSE_TOROUT_STRING("rcu_torture_boost started");
1124
1125         /* Set real-time priority. */
1126         sched_set_fifo_low(current);
1127
1128         /* Each pass through the following loop does one boost-test cycle. */
1129         do {
1130                 bool failed = false; // Test failed already in this test interval
1131                 bool gp_initiated = false;
1132
1133                 if (kthread_should_stop())
1134                         goto checkwait;
1135
1136                 /* Wait for the next test interval. */
1137                 oldstarttime = READ_ONCE(boost_starttime);
1138                 while (time_before(jiffies, oldstarttime)) {
1139                         schedule_timeout_interruptible(oldstarttime - jiffies);
1140                         if (stutter_wait("rcu_torture_boost"))
1141                                 sched_set_fifo_low(current);
1142                         if (torture_must_stop())
1143                                 goto checkwait;
1144                 }
1145
1146                 // Do one boost-test interval.
1147                 endtime = oldstarttime + test_boost_duration * HZ;
1148                 while (time_before(jiffies, endtime)) {
1149                         // Has current GP gone too long?
1150                         if (gp_initiated && !failed && !cur_ops->poll_gp_state(gp_state))
1151                                 failed = rcu_torture_boost_failed(gp_state, &gp_state_time);
1152                         // If we don't have a grace period in flight, start one.
1153                         if (!gp_initiated || cur_ops->poll_gp_state(gp_state)) {
1154                                 gp_state = cur_ops->start_gp_poll();
1155                                 gp_initiated = true;
1156                                 gp_state_time = jiffies;
1157                         }
1158                         if (stutter_wait("rcu_torture_boost")) {
1159                                 sched_set_fifo_low(current);
1160                                 // If the grace period already ended,
1161                                 // we don't know when that happened, so
1162                                 // start over.
1163                                 if (cur_ops->poll_gp_state(gp_state))
1164                                         gp_initiated = false;
1165                         }
1166                         if (torture_must_stop())
1167                                 goto checkwait;
1168                 }
1169
1170                 // In case the grace period extended beyond the end of the loop.
1171                 if (gp_initiated && !failed && !cur_ops->poll_gp_state(gp_state))
1172                         rcu_torture_boost_failed(gp_state, &gp_state_time);
1173
1174                 /*
1175                  * Set the start time of the next test interval.
1176                  * Yes, this is vulnerable to long delays, but such
1177                  * delays simply cause a false negative for the next
1178                  * interval.  Besides, we are running at RT priority,
1179                  * so delays should be relatively rare.
1180                  */
1181                 while (oldstarttime == READ_ONCE(boost_starttime) && !kthread_should_stop()) {
1182                         if (mutex_trylock(&boost_mutex)) {
1183                                 if (oldstarttime == boost_starttime) {
1184                                         WRITE_ONCE(boost_starttime,
1185                                                    jiffies + test_boost_interval * HZ);
1186                                         n_rcu_torture_boosts++;
1187                                 }
1188                                 mutex_unlock(&boost_mutex);
1189                                 break;
1190                         }
1191                         schedule_timeout_uninterruptible(HZ / 20);
1192                 }
1193
1194                 /* Go do the stutter. */
1195 checkwait:      if (stutter_wait("rcu_torture_boost"))
1196                         sched_set_fifo_low(current);
1197         } while (!torture_must_stop());
1198
1199         /* Clean up and exit. */
1200         while (!kthread_should_stop()) {
1201                 torture_shutdown_absorb("rcu_torture_boost");
1202                 schedule_timeout_uninterruptible(HZ / 20);
1203         }
1204         torture_kthread_stopping("rcu_torture_boost");
1205         return 0;
1206 }
1207
1208 /*
1209  * RCU torture force-quiescent-state kthread.  Repeatedly induces
1210  * bursts of calls to force_quiescent_state(), increasing the probability
1211  * of occurrence of some important types of race conditions.
1212  */
1213 static int
1214 rcu_torture_fqs(void *arg)
1215 {
1216         unsigned long fqs_resume_time;
1217         int fqs_burst_remaining;
1218         int oldnice = task_nice(current);
1219
1220         VERBOSE_TOROUT_STRING("rcu_torture_fqs task started");
1221         do {
1222                 fqs_resume_time = jiffies + fqs_stutter * HZ;
1223                 while (time_before(jiffies, fqs_resume_time) &&
1224                        !kthread_should_stop()) {
1225                         schedule_timeout_interruptible(HZ / 20);
1226                 }
1227                 fqs_burst_remaining = fqs_duration;
1228                 while (fqs_burst_remaining > 0 &&
1229                        !kthread_should_stop()) {
1230                         cur_ops->fqs();
1231                         udelay(fqs_holdoff);
1232                         fqs_burst_remaining -= fqs_holdoff;
1233                 }
1234                 if (stutter_wait("rcu_torture_fqs"))
1235                         sched_set_normal(current, oldnice);
1236         } while (!torture_must_stop());
1237         torture_kthread_stopping("rcu_torture_fqs");
1238         return 0;
1239 }
1240
1241 // Used by writers to randomly choose from the available grace-period primitives.
1242 static int synctype[ARRAY_SIZE(rcu_torture_writer_state_names)] = { };
1243 static int nsynctypes;
1244
1245 /*
1246  * Determine which grace-period primitives are available.
1247  */
1248 static void rcu_torture_write_types(void)
1249 {
1250         bool gp_cond1 = gp_cond, gp_cond_exp1 = gp_cond_exp, gp_cond_full1 = gp_cond_full;
1251         bool gp_cond_exp_full1 = gp_cond_exp_full, gp_exp1 = gp_exp, gp_poll_exp1 = gp_poll_exp;
1252         bool gp_poll_exp_full1 = gp_poll_exp_full, gp_normal1 = gp_normal, gp_poll1 = gp_poll;
1253         bool gp_poll_full1 = gp_poll_full, gp_sync1 = gp_sync;
1254
1255         /* Initialize synctype[] array.  If none set, take default. */
1256         if (!gp_cond1 &&
1257             !gp_cond_exp1 &&
1258             !gp_cond_full1 &&
1259             !gp_cond_exp_full1 &&
1260             !gp_exp1 &&
1261             !gp_poll_exp1 &&
1262             !gp_poll_exp_full1 &&
1263             !gp_normal1 &&
1264             !gp_poll1 &&
1265             !gp_poll_full1 &&
1266             !gp_sync1) {
1267                 gp_cond1 = true;
1268                 gp_cond_exp1 = true;
1269                 gp_cond_full1 = true;
1270                 gp_cond_exp_full1 = true;
1271                 gp_exp1 = true;
1272                 gp_poll_exp1 = true;
1273                 gp_poll_exp_full1 = true;
1274                 gp_normal1 = true;
1275                 gp_poll1 = true;
1276                 gp_poll_full1 = true;
1277                 gp_sync1 = true;
1278         }
1279         if (gp_cond1 && cur_ops->get_gp_state && cur_ops->cond_sync) {
1280                 synctype[nsynctypes++] = RTWS_COND_GET;
1281                 pr_info("%s: Testing conditional GPs.\n", __func__);
1282         } else if (gp_cond && (!cur_ops->get_gp_state || !cur_ops->cond_sync)) {
1283                 pr_alert("%s: gp_cond without primitives.\n", __func__);
1284         }
1285         if (gp_cond_exp1 && cur_ops->get_gp_state_exp && cur_ops->cond_sync_exp) {
1286                 synctype[nsynctypes++] = RTWS_COND_GET_EXP;
1287                 pr_info("%s: Testing conditional expedited GPs.\n", __func__);
1288         } else if (gp_cond_exp && (!cur_ops->get_gp_state_exp || !cur_ops->cond_sync_exp)) {
1289                 pr_alert("%s: gp_cond_exp without primitives.\n", __func__);
1290         }
1291         if (gp_cond_full1 && cur_ops->get_gp_state && cur_ops->cond_sync_full) {
1292                 synctype[nsynctypes++] = RTWS_COND_GET_FULL;
1293                 pr_info("%s: Testing conditional full-state GPs.\n", __func__);
1294         } else if (gp_cond_full && (!cur_ops->get_gp_state || !cur_ops->cond_sync_full)) {
1295                 pr_alert("%s: gp_cond_full without primitives.\n", __func__);
1296         }
1297         if (gp_cond_exp_full1 && cur_ops->get_gp_state_exp && cur_ops->cond_sync_exp_full) {
1298                 synctype[nsynctypes++] = RTWS_COND_GET_EXP_FULL;
1299                 pr_info("%s: Testing conditional full-state expedited GPs.\n", __func__);
1300         } else if (gp_cond_exp_full &&
1301                    (!cur_ops->get_gp_state_exp || !cur_ops->cond_sync_exp_full)) {
1302                 pr_alert("%s: gp_cond_exp_full without primitives.\n", __func__);
1303         }
1304         if (gp_exp1 && cur_ops->exp_sync) {
1305                 synctype[nsynctypes++] = RTWS_EXP_SYNC;
1306                 pr_info("%s: Testing expedited GPs.\n", __func__);
1307         } else if (gp_exp && !cur_ops->exp_sync) {
1308                 pr_alert("%s: gp_exp without primitives.\n", __func__);
1309         }
1310         if (gp_normal1 && cur_ops->deferred_free) {
1311                 synctype[nsynctypes++] = RTWS_DEF_FREE;
1312                 pr_info("%s: Testing asynchronous GPs.\n", __func__);
1313         } else if (gp_normal && !cur_ops->deferred_free) {
1314                 pr_alert("%s: gp_normal without primitives.\n", __func__);
1315         }
1316         if (gp_poll1 && cur_ops->get_comp_state && cur_ops->same_gp_state &&
1317             cur_ops->start_gp_poll && cur_ops->poll_gp_state) {
1318                 synctype[nsynctypes++] = RTWS_POLL_GET;
1319                 pr_info("%s: Testing polling GPs.\n", __func__);
1320         } else if (gp_poll && (!cur_ops->start_gp_poll || !cur_ops->poll_gp_state)) {
1321                 pr_alert("%s: gp_poll without primitives.\n", __func__);
1322         }
1323         if (gp_poll_full1 && cur_ops->get_comp_state_full && cur_ops->same_gp_state_full
1324             && cur_ops->start_gp_poll_full && cur_ops->poll_gp_state_full) {
1325                 synctype[nsynctypes++] = RTWS_POLL_GET_FULL;
1326                 pr_info("%s: Testing polling full-state GPs.\n", __func__);
1327         } else if (gp_poll_full && (!cur_ops->start_gp_poll_full || !cur_ops->poll_gp_state_full)) {
1328                 pr_alert("%s: gp_poll_full without primitives.\n", __func__);
1329         }
1330         if (gp_poll_exp1 && cur_ops->start_gp_poll_exp && cur_ops->poll_gp_state_exp) {
1331                 synctype[nsynctypes++] = RTWS_POLL_GET_EXP;
1332                 pr_info("%s: Testing polling expedited GPs.\n", __func__);
1333         } else if (gp_poll_exp && (!cur_ops->start_gp_poll_exp || !cur_ops->poll_gp_state_exp)) {
1334                 pr_alert("%s: gp_poll_exp without primitives.\n", __func__);
1335         }
1336         if (gp_poll_exp_full1 && cur_ops->start_gp_poll_exp_full && cur_ops->poll_gp_state_full) {
1337                 synctype[nsynctypes++] = RTWS_POLL_GET_EXP_FULL;
1338                 pr_info("%s: Testing polling full-state expedited GPs.\n", __func__);
1339         } else if (gp_poll_exp_full &&
1340                    (!cur_ops->start_gp_poll_exp_full || !cur_ops->poll_gp_state_full)) {
1341                 pr_alert("%s: gp_poll_exp_full without primitives.\n", __func__);
1342         }
1343         if (gp_sync1 && cur_ops->sync) {
1344                 synctype[nsynctypes++] = RTWS_SYNC;
1345                 pr_info("%s: Testing normal GPs.\n", __func__);
1346         } else if (gp_sync && !cur_ops->sync) {
1347                 pr_alert("%s: gp_sync without primitives.\n", __func__);
1348         }
1349         pr_alert("%s: Testing %d update types.\n", __func__, nsynctypes);
1350 }
1351
1352 /*
1353  * Do the specified rcu_torture_writer() synchronous grace period,
1354  * while also testing out the polled APIs.  Note well that the single-CPU
1355  * grace-period optimizations must be accounted for.
1356  */
1357 static void do_rtws_sync(struct torture_random_state *trsp, void (*sync)(void))
1358 {
1359         unsigned long cookie;
1360         struct rcu_gp_oldstate cookie_full;
1361         bool dopoll;
1362         bool dopoll_full;
1363         unsigned long r = torture_random(trsp);
1364
1365         dopoll = cur_ops->get_gp_state && cur_ops->poll_gp_state && !(r & 0x300);
1366         dopoll_full = cur_ops->get_gp_state_full && cur_ops->poll_gp_state_full && !(r & 0xc00);
1367         if (dopoll || dopoll_full)
1368                 cpus_read_lock();
1369         if (dopoll)
1370                 cookie = cur_ops->get_gp_state();
1371         if (dopoll_full)
1372                 cur_ops->get_gp_state_full(&cookie_full);
1373         if (cur_ops->poll_need_2gp && cur_ops->poll_need_2gp(dopoll, dopoll_full))
1374                 sync();
1375         sync();
1376         WARN_ONCE(dopoll && !cur_ops->poll_gp_state(cookie),
1377                   "%s: Cookie check 3 failed %pS() online %*pbl.",
1378                   __func__, sync, cpumask_pr_args(cpu_online_mask));
1379         WARN_ONCE(dopoll_full && !cur_ops->poll_gp_state_full(&cookie_full),
1380                   "%s: Cookie check 4 failed %pS() online %*pbl",
1381                   __func__, sync, cpumask_pr_args(cpu_online_mask));
1382         if (dopoll || dopoll_full)
1383                 cpus_read_unlock();
1384 }
1385
1386 /*
1387  * RCU torture writer kthread.  Repeatedly substitutes a new structure
1388  * for that pointed to by rcu_torture_current, freeing the old structure
1389  * after a series of grace periods (the "pipeline").
1390  */
1391 static int
1392 rcu_torture_writer(void *arg)
1393 {
1394         bool boot_ended;
1395         bool can_expedite = !rcu_gp_is_expedited() && !rcu_gp_is_normal();
1396         unsigned long cookie;
1397         struct rcu_gp_oldstate cookie_full;
1398         int expediting = 0;
1399         unsigned long gp_snap;
1400         unsigned long gp_snap1;
1401         struct rcu_gp_oldstate gp_snap_full;
1402         struct rcu_gp_oldstate gp_snap1_full;
1403         int i;
1404         int idx;
1405         int oldnice = task_nice(current);
1406         struct rcu_gp_oldstate *rgo = NULL;
1407         int rgo_size = 0;
1408         struct rcu_torture *rp;
1409         struct rcu_torture *old_rp;
1410         static DEFINE_TORTURE_RANDOM(rand);
1411         unsigned long stallsdone = jiffies;
1412         bool stutter_waited;
1413         unsigned long *ulo = NULL;
1414         int ulo_size = 0;
1415
1416         // If a new stall test is added, this must be adjusted.
1417         if (stall_cpu_holdoff + stall_gp_kthread + stall_cpu)
1418                 stallsdone += (stall_cpu_holdoff + stall_gp_kthread + stall_cpu + 60) *
1419                               HZ * (stall_cpu_repeat + 1);
1420         VERBOSE_TOROUT_STRING("rcu_torture_writer task started");
1421         if (!can_expedite)
1422                 pr_alert("%s" TORTURE_FLAG
1423                          " GP expediting controlled from boot/sysfs for %s.\n",
1424                          torture_type, cur_ops->name);
1425         if (WARN_ONCE(nsynctypes == 0,
1426                       "%s: No update-side primitives.\n", __func__)) {
1427                 /*
1428                  * No updates primitives, so don't try updating.
1429                  * The resulting test won't be testing much, hence the
1430                  * above WARN_ONCE().
1431                  */
1432                 rcu_torture_writer_state = RTWS_STOPPING;
1433                 torture_kthread_stopping("rcu_torture_writer");
1434                 return 0;
1435         }
1436         if (cur_ops->poll_active > 0) {
1437                 ulo = kzalloc(cur_ops->poll_active * sizeof(ulo[0]), GFP_KERNEL);
1438                 if (!WARN_ON(!ulo))
1439                         ulo_size = cur_ops->poll_active;
1440         }
1441         if (cur_ops->poll_active_full > 0) {
1442                 rgo = kzalloc(cur_ops->poll_active_full * sizeof(rgo[0]), GFP_KERNEL);
1443                 if (!WARN_ON(!rgo))
1444                         rgo_size = cur_ops->poll_active_full;
1445         }
1446
1447         do {
1448                 rcu_torture_writer_state = RTWS_FIXED_DELAY;
1449                 torture_hrtimeout_us(500, 1000, &rand);
1450                 rp = rcu_torture_alloc();
1451                 if (rp == NULL)
1452                         continue;
1453                 rp->rtort_pipe_count = 0;
1454                 ASSERT_EXCLUSIVE_WRITER(rp->rtort_pipe_count);
1455                 rcu_torture_writer_state = RTWS_DELAY;
1456                 udelay(torture_random(&rand) & 0x3ff);
1457                 rcu_torture_writer_state = RTWS_REPLACE;
1458                 old_rp = rcu_dereference_check(rcu_torture_current,
1459                                                current == writer_task);
1460                 rp->rtort_mbtest = 1;
1461                 rcu_assign_pointer(rcu_torture_current, rp);
1462                 smp_wmb(); /* Mods to old_rp must follow rcu_assign_pointer() */
1463                 if (old_rp) {
1464                         i = old_rp->rtort_pipe_count;
1465                         if (i > RCU_TORTURE_PIPE_LEN)
1466                                 i = RCU_TORTURE_PIPE_LEN;
1467                         atomic_inc(&rcu_torture_wcount[i]);
1468                         WRITE_ONCE(old_rp->rtort_pipe_count,
1469                                    old_rp->rtort_pipe_count + 1);
1470                         ASSERT_EXCLUSIVE_WRITER(old_rp->rtort_pipe_count);
1471
1472                         // Make sure readers block polled grace periods.
1473                         if (cur_ops->get_gp_state && cur_ops->poll_gp_state) {
1474                                 idx = cur_ops->readlock();
1475                                 cookie = cur_ops->get_gp_state();
1476                                 WARN_ONCE(cur_ops->poll_gp_state(cookie),
1477                                           "%s: Cookie check 1 failed %s(%d) %lu->%lu\n",
1478                                           __func__,
1479                                           rcu_torture_writer_state_getname(),
1480                                           rcu_torture_writer_state,
1481                                           cookie, cur_ops->get_gp_state());
1482                                 if (cur_ops->get_comp_state) {
1483                                         cookie = cur_ops->get_comp_state();
1484                                         WARN_ON_ONCE(!cur_ops->poll_gp_state(cookie));
1485                                 }
1486                                 cur_ops->readunlock(idx);
1487                         }
1488                         if (cur_ops->get_gp_state_full && cur_ops->poll_gp_state_full) {
1489                                 idx = cur_ops->readlock();
1490                                 cur_ops->get_gp_state_full(&cookie_full);
1491                                 WARN_ONCE(cur_ops->poll_gp_state_full(&cookie_full),
1492                                           "%s: Cookie check 5 failed %s(%d) online %*pbl\n",
1493                                           __func__,
1494                                           rcu_torture_writer_state_getname(),
1495                                           rcu_torture_writer_state,
1496                                           cpumask_pr_args(cpu_online_mask));
1497                                 if (cur_ops->get_comp_state_full) {
1498                                         cur_ops->get_comp_state_full(&cookie_full);
1499                                         WARN_ON_ONCE(!cur_ops->poll_gp_state_full(&cookie_full));
1500                                 }
1501                                 cur_ops->readunlock(idx);
1502                         }
1503                         switch (synctype[torture_random(&rand) % nsynctypes]) {
1504                         case RTWS_DEF_FREE:
1505                                 rcu_torture_writer_state = RTWS_DEF_FREE;
1506                                 cur_ops->deferred_free(old_rp);
1507                                 break;
1508                         case RTWS_EXP_SYNC:
1509                                 rcu_torture_writer_state = RTWS_EXP_SYNC;
1510                                 do_rtws_sync(&rand, cur_ops->exp_sync);
1511                                 rcu_torture_pipe_update(old_rp);
1512                                 break;
1513                         case RTWS_COND_GET:
1514                                 rcu_torture_writer_state = RTWS_COND_GET;
1515                                 gp_snap = cur_ops->get_gp_state();
1516                                 torture_hrtimeout_jiffies(torture_random(&rand) % 16, &rand);
1517                                 rcu_torture_writer_state = RTWS_COND_SYNC;
1518                                 cur_ops->cond_sync(gp_snap);
1519                                 rcu_torture_pipe_update(old_rp);
1520                                 break;
1521                         case RTWS_COND_GET_EXP:
1522                                 rcu_torture_writer_state = RTWS_COND_GET_EXP;
1523                                 gp_snap = cur_ops->get_gp_state_exp();
1524                                 torture_hrtimeout_jiffies(torture_random(&rand) % 16, &rand);
1525                                 rcu_torture_writer_state = RTWS_COND_SYNC_EXP;
1526                                 cur_ops->cond_sync_exp(gp_snap);
1527                                 rcu_torture_pipe_update(old_rp);
1528                                 break;
1529                         case RTWS_COND_GET_FULL:
1530                                 rcu_torture_writer_state = RTWS_COND_GET_FULL;
1531                                 cur_ops->get_gp_state_full(&gp_snap_full);
1532                                 torture_hrtimeout_jiffies(torture_random(&rand) % 16, &rand);
1533                                 rcu_torture_writer_state = RTWS_COND_SYNC_FULL;
1534                                 cur_ops->cond_sync_full(&gp_snap_full);
1535                                 rcu_torture_pipe_update(old_rp);
1536                                 break;
1537                         case RTWS_COND_GET_EXP_FULL:
1538                                 rcu_torture_writer_state = RTWS_COND_GET_EXP_FULL;
1539                                 cur_ops->get_gp_state_full(&gp_snap_full);
1540                                 torture_hrtimeout_jiffies(torture_random(&rand) % 16, &rand);
1541                                 rcu_torture_writer_state = RTWS_COND_SYNC_EXP_FULL;
1542                                 cur_ops->cond_sync_exp_full(&gp_snap_full);
1543                                 rcu_torture_pipe_update(old_rp);
1544                                 break;
1545                         case RTWS_POLL_GET:
1546                                 rcu_torture_writer_state = RTWS_POLL_GET;
1547                                 for (i = 0; i < ulo_size; i++)
1548                                         ulo[i] = cur_ops->get_comp_state();
1549                                 gp_snap = cur_ops->start_gp_poll();
1550                                 rcu_torture_writer_state = RTWS_POLL_WAIT;
1551                                 while (!cur_ops->poll_gp_state(gp_snap)) {
1552                                         gp_snap1 = cur_ops->get_gp_state();
1553                                         for (i = 0; i < ulo_size; i++)
1554                                                 if (cur_ops->poll_gp_state(ulo[i]) ||
1555                                                     cur_ops->same_gp_state(ulo[i], gp_snap1)) {
1556                                                         ulo[i] = gp_snap1;
1557                                                         break;
1558                                                 }
1559                                         WARN_ON_ONCE(ulo_size > 0 && i >= ulo_size);
1560                                         torture_hrtimeout_jiffies(torture_random(&rand) % 16,
1561                                                                   &rand);
1562                                 }
1563                                 rcu_torture_pipe_update(old_rp);
1564                                 break;
1565                         case RTWS_POLL_GET_FULL:
1566                                 rcu_torture_writer_state = RTWS_POLL_GET_FULL;
1567                                 for (i = 0; i < rgo_size; i++)
1568                                         cur_ops->get_comp_state_full(&rgo[i]);
1569                                 cur_ops->start_gp_poll_full(&gp_snap_full);
1570                                 rcu_torture_writer_state = RTWS_POLL_WAIT_FULL;
1571                                 while (!cur_ops->poll_gp_state_full(&gp_snap_full)) {
1572                                         cur_ops->get_gp_state_full(&gp_snap1_full);
1573                                         for (i = 0; i < rgo_size; i++)
1574                                                 if (cur_ops->poll_gp_state_full(&rgo[i]) ||
1575                                                     cur_ops->same_gp_state_full(&rgo[i],
1576                                                                                 &gp_snap1_full)) {
1577                                                         rgo[i] = gp_snap1_full;
1578                                                         break;
1579                                                 }
1580                                         WARN_ON_ONCE(rgo_size > 0 && i >= rgo_size);
1581                                         torture_hrtimeout_jiffies(torture_random(&rand) % 16,
1582                                                                   &rand);
1583                                 }
1584                                 rcu_torture_pipe_update(old_rp);
1585                                 break;
1586                         case RTWS_POLL_GET_EXP:
1587                                 rcu_torture_writer_state = RTWS_POLL_GET_EXP;
1588                                 gp_snap = cur_ops->start_gp_poll_exp();
1589                                 rcu_torture_writer_state = RTWS_POLL_WAIT_EXP;
1590                                 while (!cur_ops->poll_gp_state_exp(gp_snap))
1591                                         torture_hrtimeout_jiffies(torture_random(&rand) % 16,
1592                                                                   &rand);
1593                                 rcu_torture_pipe_update(old_rp);
1594                                 break;
1595                         case RTWS_POLL_GET_EXP_FULL:
1596                                 rcu_torture_writer_state = RTWS_POLL_GET_EXP_FULL;
1597                                 cur_ops->start_gp_poll_exp_full(&gp_snap_full);
1598                                 rcu_torture_writer_state = RTWS_POLL_WAIT_EXP_FULL;
1599                                 while (!cur_ops->poll_gp_state_full(&gp_snap_full))
1600                                         torture_hrtimeout_jiffies(torture_random(&rand) % 16,
1601                                                                   &rand);
1602                                 rcu_torture_pipe_update(old_rp);
1603                                 break;
1604                         case RTWS_SYNC:
1605                                 rcu_torture_writer_state = RTWS_SYNC;
1606                                 do_rtws_sync(&rand, cur_ops->sync);
1607                                 rcu_torture_pipe_update(old_rp);
1608                                 break;
1609                         default:
1610                                 WARN_ON_ONCE(1);
1611                                 break;
1612                         }
1613                 }
1614                 WRITE_ONCE(rcu_torture_current_version,
1615                            rcu_torture_current_version + 1);
1616                 /* Cycle through nesting levels of rcu_expedite_gp() calls. */
1617                 if (can_expedite &&
1618                     !(torture_random(&rand) & 0xff & (!!expediting - 1))) {
1619                         WARN_ON_ONCE(expediting == 0 && rcu_gp_is_expedited());
1620                         if (expediting >= 0)
1621                                 rcu_expedite_gp();
1622                         else
1623                                 rcu_unexpedite_gp();
1624                         if (++expediting > 3)
1625                                 expediting = -expediting;
1626                 } else if (!can_expedite) { /* Disabled during boot, recheck. */
1627                         can_expedite = !rcu_gp_is_expedited() &&
1628                                        !rcu_gp_is_normal();
1629                 }
1630                 rcu_torture_writer_state = RTWS_STUTTER;
1631                 boot_ended = rcu_inkernel_boot_has_ended();
1632                 stutter_waited = stutter_wait("rcu_torture_writer");
1633                 if (stutter_waited &&
1634                     !atomic_read(&rcu_fwd_cb_nodelay) &&
1635                     !cur_ops->slow_gps &&
1636                     !torture_must_stop() &&
1637                     boot_ended &&
1638                     time_after(jiffies, stallsdone))
1639                         for (i = 0; i < ARRAY_SIZE(rcu_tortures); i++)
1640                                 if (list_empty(&rcu_tortures[i].rtort_free) &&
1641                                     rcu_access_pointer(rcu_torture_current) != &rcu_tortures[i]) {
1642                                         tracing_off();
1643                                         if (cur_ops->gp_kthread_dbg)
1644                                                 cur_ops->gp_kthread_dbg();
1645                                         WARN(1, "%s: rtort_pipe_count: %d\n", __func__, rcu_tortures[i].rtort_pipe_count);
1646                                         rcu_ftrace_dump(DUMP_ALL);
1647                                 }
1648                 if (stutter_waited)
1649                         sched_set_normal(current, oldnice);
1650         } while (!torture_must_stop());
1651         rcu_torture_current = NULL;  // Let stats task know that we are done.
1652         /* Reset expediting back to unexpedited. */
1653         if (expediting > 0)
1654                 expediting = -expediting;
1655         while (can_expedite && expediting++ < 0)
1656                 rcu_unexpedite_gp();
1657         WARN_ON_ONCE(can_expedite && rcu_gp_is_expedited());
1658         if (!can_expedite)
1659                 pr_alert("%s" TORTURE_FLAG
1660                          " Dynamic grace-period expediting was disabled.\n",
1661                          torture_type);
1662         kfree(ulo);
1663         kfree(rgo);
1664         rcu_torture_writer_state = RTWS_STOPPING;
1665         torture_kthread_stopping("rcu_torture_writer");
1666         return 0;
1667 }
1668
1669 /*
1670  * RCU torture fake writer kthread.  Repeatedly calls sync, with a random
1671  * delay between calls.
1672  */
1673 static int
1674 rcu_torture_fakewriter(void *arg)
1675 {
1676         unsigned long gp_snap;
1677         struct rcu_gp_oldstate gp_snap_full;
1678         DEFINE_TORTURE_RANDOM(rand);
1679
1680         VERBOSE_TOROUT_STRING("rcu_torture_fakewriter task started");
1681         set_user_nice(current, MAX_NICE);
1682
1683         if (WARN_ONCE(nsynctypes == 0,
1684                       "%s: No update-side primitives.\n", __func__)) {
1685                 /*
1686                  * No updates primitives, so don't try updating.
1687                  * The resulting test won't be testing much, hence the
1688                  * above WARN_ONCE().
1689                  */
1690                 torture_kthread_stopping("rcu_torture_fakewriter");
1691                 return 0;
1692         }
1693
1694         do {
1695                 torture_hrtimeout_jiffies(torture_random(&rand) % 10, &rand);
1696                 if (cur_ops->cb_barrier != NULL &&
1697                     torture_random(&rand) % (nfakewriters * 8) == 0) {
1698                         cur_ops->cb_barrier();
1699                 } else {
1700                         switch (synctype[torture_random(&rand) % nsynctypes]) {
1701                         case RTWS_DEF_FREE:
1702                                 break;
1703                         case RTWS_EXP_SYNC:
1704                                 cur_ops->exp_sync();
1705                                 break;
1706                         case RTWS_COND_GET:
1707                                 gp_snap = cur_ops->get_gp_state();
1708                                 torture_hrtimeout_jiffies(torture_random(&rand) % 16, &rand);
1709                                 cur_ops->cond_sync(gp_snap);
1710                                 break;
1711                         case RTWS_COND_GET_EXP:
1712                                 gp_snap = cur_ops->get_gp_state_exp();
1713                                 torture_hrtimeout_jiffies(torture_random(&rand) % 16, &rand);
1714                                 cur_ops->cond_sync_exp(gp_snap);
1715                                 break;
1716                         case RTWS_COND_GET_FULL:
1717                                 cur_ops->get_gp_state_full(&gp_snap_full);
1718                                 torture_hrtimeout_jiffies(torture_random(&rand) % 16, &rand);
1719                                 cur_ops->cond_sync_full(&gp_snap_full);
1720                                 break;
1721                         case RTWS_COND_GET_EXP_FULL:
1722                                 cur_ops->get_gp_state_full(&gp_snap_full);
1723                                 torture_hrtimeout_jiffies(torture_random(&rand) % 16, &rand);
1724                                 cur_ops->cond_sync_exp_full(&gp_snap_full);
1725                                 break;
1726                         case RTWS_POLL_GET:
1727                                 if (cur_ops->start_poll_irqsoff)
1728                                         local_irq_disable();
1729                                 gp_snap = cur_ops->start_gp_poll();
1730                                 if (cur_ops->start_poll_irqsoff)
1731                                         local_irq_enable();
1732                                 while (!cur_ops->poll_gp_state(gp_snap)) {
1733                                         torture_hrtimeout_jiffies(torture_random(&rand) % 16,
1734                                                                   &rand);
1735                                 }
1736                                 break;
1737                         case RTWS_POLL_GET_FULL:
1738                                 if (cur_ops->start_poll_irqsoff)
1739                                         local_irq_disable();
1740                                 cur_ops->start_gp_poll_full(&gp_snap_full);
1741                                 if (cur_ops->start_poll_irqsoff)
1742                                         local_irq_enable();
1743                                 while (!cur_ops->poll_gp_state_full(&gp_snap_full)) {
1744                                         torture_hrtimeout_jiffies(torture_random(&rand) % 16,
1745                                                                   &rand);
1746                                 }
1747                                 break;
1748                         case RTWS_POLL_GET_EXP:
1749                                 gp_snap = cur_ops->start_gp_poll_exp();
1750                                 while (!cur_ops->poll_gp_state_exp(gp_snap)) {
1751                                         torture_hrtimeout_jiffies(torture_random(&rand) % 16,
1752                                                                   &rand);
1753                                 }
1754                                 break;
1755                         case RTWS_POLL_GET_EXP_FULL:
1756                                 cur_ops->start_gp_poll_exp_full(&gp_snap_full);
1757                                 while (!cur_ops->poll_gp_state_full(&gp_snap_full)) {
1758                                         torture_hrtimeout_jiffies(torture_random(&rand) % 16,
1759                                                                   &rand);
1760                                 }
1761                                 break;
1762                         case RTWS_SYNC:
1763                                 cur_ops->sync();
1764                                 break;
1765                         default:
1766                                 WARN_ON_ONCE(1);
1767                                 break;
1768                         }
1769                 }
1770                 stutter_wait("rcu_torture_fakewriter");
1771         } while (!torture_must_stop());
1772
1773         torture_kthread_stopping("rcu_torture_fakewriter");
1774         return 0;
1775 }
1776
1777 static void rcu_torture_timer_cb(struct rcu_head *rhp)
1778 {
1779         kfree(rhp);
1780 }
1781
1782 // Set up and carry out testing of RCU's global memory ordering
1783 static void rcu_torture_reader_do_mbchk(long myid, struct rcu_torture *rtp,
1784                                         struct torture_random_state *trsp)
1785 {
1786         unsigned long loops;
1787         int noc = torture_num_online_cpus();
1788         int rdrchked;
1789         int rdrchker;
1790         struct rcu_torture_reader_check *rtrcp; // Me.
1791         struct rcu_torture_reader_check *rtrcp_assigner; // Assigned us to do checking.
1792         struct rcu_torture_reader_check *rtrcp_chked; // Reader being checked.
1793         struct rcu_torture_reader_check *rtrcp_chker; // Reader doing checking when not me.
1794
1795         if (myid < 0)
1796                 return; // Don't try this from timer handlers.
1797
1798         // Increment my counter.
1799         rtrcp = &rcu_torture_reader_mbchk[myid];
1800         WRITE_ONCE(rtrcp->rtc_myloops, rtrcp->rtc_myloops + 1);
1801
1802         // Attempt to assign someone else some checking work.
1803         rdrchked = torture_random(trsp) % nrealreaders;
1804         rtrcp_chked = &rcu_torture_reader_mbchk[rdrchked];
1805         rdrchker = torture_random(trsp) % nrealreaders;
1806         rtrcp_chker = &rcu_torture_reader_mbchk[rdrchker];
1807         if (rdrchked != myid && rdrchked != rdrchker && noc >= rdrchked && noc >= rdrchker &&
1808             smp_load_acquire(&rtrcp->rtc_chkrdr) < 0 && // Pairs with smp_store_release below.
1809             !READ_ONCE(rtp->rtort_chkp) &&
1810             !smp_load_acquire(&rtrcp_chker->rtc_assigner)) { // Pairs with smp_store_release below.
1811                 rtrcp->rtc_chkloops = READ_ONCE(rtrcp_chked->rtc_myloops);
1812                 WARN_ON_ONCE(rtrcp->rtc_chkrdr >= 0);
1813                 rtrcp->rtc_chkrdr = rdrchked;
1814                 WARN_ON_ONCE(rtrcp->rtc_ready); // This gets set after the grace period ends.
1815                 if (cmpxchg_relaxed(&rtrcp_chker->rtc_assigner, NULL, rtrcp) ||
1816                     cmpxchg_relaxed(&rtp->rtort_chkp, NULL, rtrcp))
1817                         (void)cmpxchg_relaxed(&rtrcp_chker->rtc_assigner, rtrcp, NULL); // Back out.
1818         }
1819
1820         // If assigned some completed work, do it!
1821         rtrcp_assigner = READ_ONCE(rtrcp->rtc_assigner);
1822         if (!rtrcp_assigner || !smp_load_acquire(&rtrcp_assigner->rtc_ready))
1823                 return; // No work or work not yet ready.
1824         rdrchked = rtrcp_assigner->rtc_chkrdr;
1825         if (WARN_ON_ONCE(rdrchked < 0))
1826                 return;
1827         rtrcp_chked = &rcu_torture_reader_mbchk[rdrchked];
1828         loops = READ_ONCE(rtrcp_chked->rtc_myloops);
1829         atomic_inc(&n_rcu_torture_mbchk_tries);
1830         if (ULONG_CMP_LT(loops, rtrcp_assigner->rtc_chkloops))
1831                 atomic_inc(&n_rcu_torture_mbchk_fail);
1832         rtrcp_assigner->rtc_chkloops = loops + ULONG_MAX / 2;
1833         rtrcp_assigner->rtc_ready = 0;
1834         smp_store_release(&rtrcp->rtc_assigner, NULL); // Someone else can assign us work.
1835         smp_store_release(&rtrcp_assigner->rtc_chkrdr, -1); // Assigner can again assign.
1836 }
1837
1838 /*
1839  * Do one extension of an RCU read-side critical section using the
1840  * current reader state in readstate (set to zero for initial entry
1841  * to extended critical section), set the new state as specified by
1842  * newstate (set to zero for final exit from extended critical section),
1843  * and random-number-generator state in trsp.  If this is neither the
1844  * beginning or end of the critical section and if there was actually a
1845  * change, do a ->read_delay().
1846  */
1847 static void rcutorture_one_extend(int *readstate, int newstate,
1848                                   struct torture_random_state *trsp,
1849                                   struct rt_read_seg *rtrsp)
1850 {
1851         unsigned long flags;
1852         int idxnew1 = -1;
1853         int idxnew2 = -1;
1854         int idxold1 = *readstate;
1855         int idxold2 = idxold1;
1856         int statesnew = ~*readstate & newstate;
1857         int statesold = *readstate & ~newstate;
1858
1859         WARN_ON_ONCE(idxold2 < 0);
1860         WARN_ON_ONCE(idxold2 & ~RCUTORTURE_RDR_ALLBITS);
1861         rtrsp->rt_readstate = newstate;
1862
1863         /* First, put new protection in place to avoid critical-section gap. */
1864         if (statesnew & RCUTORTURE_RDR_BH)
1865                 local_bh_disable();
1866         if (statesnew & RCUTORTURE_RDR_RBH)
1867                 rcu_read_lock_bh();
1868         if (statesnew & RCUTORTURE_RDR_IRQ)
1869                 local_irq_disable();
1870         if (statesnew & RCUTORTURE_RDR_PREEMPT)
1871                 preempt_disable();
1872         if (statesnew & RCUTORTURE_RDR_SCHED)
1873                 rcu_read_lock_sched();
1874         if (statesnew & RCUTORTURE_RDR_RCU_1)
1875                 idxnew1 = (cur_ops->readlock() << RCUTORTURE_RDR_SHIFT_1) & RCUTORTURE_RDR_MASK_1;
1876         if (statesnew & RCUTORTURE_RDR_RCU_2)
1877                 idxnew2 = (cur_ops->readlock() << RCUTORTURE_RDR_SHIFT_2) & RCUTORTURE_RDR_MASK_2;
1878
1879         /*
1880          * Next, remove old protection, in decreasing order of strength
1881          * to avoid unlock paths that aren't safe in the stronger
1882          * context. Namely: BH can not be enabled with disabled interrupts.
1883          * Additionally PREEMPT_RT requires that BH is enabled in preemptible
1884          * context.
1885          */
1886         if (statesold & RCUTORTURE_RDR_IRQ)
1887                 local_irq_enable();
1888         if (statesold & RCUTORTURE_RDR_PREEMPT)
1889                 preempt_enable();
1890         if (statesold & RCUTORTURE_RDR_SCHED)
1891                 rcu_read_unlock_sched();
1892         if (statesold & RCUTORTURE_RDR_BH)
1893                 local_bh_enable();
1894         if (statesold & RCUTORTURE_RDR_RBH)
1895                 rcu_read_unlock_bh();
1896         if (statesold & RCUTORTURE_RDR_RCU_2) {
1897                 cur_ops->readunlock((idxold2 & RCUTORTURE_RDR_MASK_2) >> RCUTORTURE_RDR_SHIFT_2);
1898                 WARN_ON_ONCE(idxnew2 != -1);
1899                 idxold2 = 0;
1900         }
1901         if (statesold & RCUTORTURE_RDR_RCU_1) {
1902                 bool lockit;
1903
1904                 lockit = !cur_ops->no_pi_lock && !statesnew && !(torture_random(trsp) & 0xffff);
1905                 if (lockit)
1906                         raw_spin_lock_irqsave(&current->pi_lock, flags);
1907                 cur_ops->readunlock((idxold1 & RCUTORTURE_RDR_MASK_1) >> RCUTORTURE_RDR_SHIFT_1);
1908                 WARN_ON_ONCE(idxnew1 != -1);
1909                 idxold1 = 0;
1910                 if (lockit)
1911                         raw_spin_unlock_irqrestore(&current->pi_lock, flags);
1912         }
1913
1914         /* Delay if neither beginning nor end and there was a change. */
1915         if ((statesnew || statesold) && *readstate && newstate)
1916                 cur_ops->read_delay(trsp, rtrsp);
1917
1918         /* Update the reader state. */
1919         if (idxnew1 == -1)
1920                 idxnew1 = idxold1 & RCUTORTURE_RDR_MASK_1;
1921         WARN_ON_ONCE(idxnew1 < 0);
1922         if (idxnew2 == -1)
1923                 idxnew2 = idxold2 & RCUTORTURE_RDR_MASK_2;
1924         WARN_ON_ONCE(idxnew2 < 0);
1925         *readstate = idxnew1 | idxnew2 | newstate;
1926         WARN_ON_ONCE(*readstate < 0);
1927         if (WARN_ON_ONCE(*readstate & ~RCUTORTURE_RDR_ALLBITS))
1928                 pr_info("Unexpected readstate value of %#x\n", *readstate);
1929 }
1930
1931 /* Return the biggest extendables mask given current RCU and boot parameters. */
1932 static int rcutorture_extend_mask_max(void)
1933 {
1934         int mask;
1935
1936         WARN_ON_ONCE(extendables & ~RCUTORTURE_MAX_EXTEND);
1937         mask = extendables & RCUTORTURE_MAX_EXTEND & cur_ops->extendables;
1938         mask = mask | RCUTORTURE_RDR_RCU_1 | RCUTORTURE_RDR_RCU_2;
1939         return mask;
1940 }
1941
1942 /* Return a random protection state mask, but with at least one bit set. */
1943 static int
1944 rcutorture_extend_mask(int oldmask, struct torture_random_state *trsp)
1945 {
1946         int mask = rcutorture_extend_mask_max();
1947         unsigned long randmask1 = torture_random(trsp);
1948         unsigned long randmask2 = randmask1 >> 3;
1949         unsigned long preempts = RCUTORTURE_RDR_PREEMPT | RCUTORTURE_RDR_SCHED;
1950         unsigned long preempts_irq = preempts | RCUTORTURE_RDR_IRQ;
1951         unsigned long bhs = RCUTORTURE_RDR_BH | RCUTORTURE_RDR_RBH;
1952
1953         WARN_ON_ONCE(mask >> RCUTORTURE_RDR_SHIFT_1);  // Can't have reader idx bits.
1954         /* Mostly only one bit (need preemption!), sometimes lots of bits. */
1955         if (!(randmask1 & 0x7))
1956                 mask = mask & randmask2;
1957         else
1958                 mask = mask & (1 << (randmask2 % RCUTORTURE_RDR_NBITS));
1959
1960         // Can't have nested RCU reader without outer RCU reader.
1961         if (!(mask & RCUTORTURE_RDR_RCU_1) && (mask & RCUTORTURE_RDR_RCU_2)) {
1962                 if (oldmask & RCUTORTURE_RDR_RCU_1)
1963                         mask &= ~RCUTORTURE_RDR_RCU_2;
1964                 else
1965                         mask |= RCUTORTURE_RDR_RCU_1;
1966         }
1967
1968         /*
1969          * Can't enable bh w/irq disabled.
1970          */
1971         if (mask & RCUTORTURE_RDR_IRQ)
1972                 mask |= oldmask & bhs;
1973
1974         /*
1975          * Ideally these sequences would be detected in debug builds
1976          * (regardless of RT), but until then don't stop testing
1977          * them on non-RT.
1978          */
1979         if (IS_ENABLED(CONFIG_PREEMPT_RT)) {
1980                 /* Can't modify BH in atomic context */
1981                 if (oldmask & preempts_irq)
1982                         mask &= ~bhs;
1983                 if ((oldmask | mask) & preempts_irq)
1984                         mask |= oldmask & bhs;
1985         }
1986
1987         return mask ?: RCUTORTURE_RDR_RCU_1;
1988 }
1989
1990 /*
1991  * Do a randomly selected number of extensions of an existing RCU read-side
1992  * critical section.
1993  */
1994 static struct rt_read_seg *
1995 rcutorture_loop_extend(int *readstate, struct torture_random_state *trsp,
1996                        struct rt_read_seg *rtrsp)
1997 {
1998         int i;
1999         int j;
2000         int mask = rcutorture_extend_mask_max();
2001
2002         WARN_ON_ONCE(!*readstate); /* -Existing- RCU read-side critsect! */
2003         if (!((mask - 1) & mask))
2004                 return rtrsp;  /* Current RCU reader not extendable. */
2005         /* Bias towards larger numbers of loops. */
2006         i = torture_random(trsp);
2007         i = ((i | (i >> 3)) & RCUTORTURE_RDR_MAX_LOOPS) + 1;
2008         for (j = 0; j < i; j++) {
2009                 mask = rcutorture_extend_mask(*readstate, trsp);
2010                 rcutorture_one_extend(readstate, mask, trsp, &rtrsp[j]);
2011         }
2012         return &rtrsp[j];
2013 }
2014
2015 /*
2016  * Do one read-side critical section, returning false if there was
2017  * no data to read.  Can be invoked both from process context and
2018  * from a timer handler.
2019  */
2020 static bool rcu_torture_one_read(struct torture_random_state *trsp, long myid)
2021 {
2022         bool checkpolling = !(torture_random(trsp) & 0xfff);
2023         unsigned long cookie;
2024         struct rcu_gp_oldstate cookie_full;
2025         int i;
2026         unsigned long started;
2027         unsigned long completed;
2028         int newstate;
2029         struct rcu_torture *p;
2030         int pipe_count;
2031         int readstate = 0;
2032         struct rt_read_seg rtseg[RCUTORTURE_RDR_MAX_SEGS] = { { 0 } };
2033         struct rt_read_seg *rtrsp = &rtseg[0];
2034         struct rt_read_seg *rtrsp1;
2035         unsigned long long ts;
2036
2037         WARN_ON_ONCE(!rcu_is_watching());
2038         newstate = rcutorture_extend_mask(readstate, trsp);
2039         rcutorture_one_extend(&readstate, newstate, trsp, rtrsp++);
2040         if (checkpolling) {
2041                 if (cur_ops->get_gp_state && cur_ops->poll_gp_state)
2042                         cookie = cur_ops->get_gp_state();
2043                 if (cur_ops->get_gp_state_full && cur_ops->poll_gp_state_full)
2044                         cur_ops->get_gp_state_full(&cookie_full);
2045         }
2046         started = cur_ops->get_gp_seq();
2047         ts = rcu_trace_clock_local();
2048         p = rcu_dereference_check(rcu_torture_current,
2049                                   !cur_ops->readlock_held || cur_ops->readlock_held());
2050         if (p == NULL) {
2051                 /* Wait for rcu_torture_writer to get underway */
2052                 rcutorture_one_extend(&readstate, 0, trsp, rtrsp);
2053                 return false;
2054         }
2055         if (p->rtort_mbtest == 0)
2056                 atomic_inc(&n_rcu_torture_mberror);
2057         rcu_torture_reader_do_mbchk(myid, p, trsp);
2058         rtrsp = rcutorture_loop_extend(&readstate, trsp, rtrsp);
2059         preempt_disable();
2060         pipe_count = READ_ONCE(p->rtort_pipe_count);
2061         if (pipe_count > RCU_TORTURE_PIPE_LEN) {
2062                 // Should not happen in a correct RCU implementation,
2063                 // happens quite often for torture_type=busted.
2064                 pipe_count = RCU_TORTURE_PIPE_LEN;
2065         }
2066         completed = cur_ops->get_gp_seq();
2067         if (pipe_count > 1) {
2068                 do_trace_rcu_torture_read(cur_ops->name, &p->rtort_rcu,
2069                                           ts, started, completed);
2070                 rcu_ftrace_dump(DUMP_ALL);
2071         }
2072         __this_cpu_inc(rcu_torture_count[pipe_count]);
2073         completed = rcutorture_seq_diff(completed, started);
2074         if (completed > RCU_TORTURE_PIPE_LEN) {
2075                 /* Should not happen, but... */
2076                 completed = RCU_TORTURE_PIPE_LEN;
2077         }
2078         __this_cpu_inc(rcu_torture_batch[completed]);
2079         preempt_enable();
2080         if (checkpolling) {
2081                 if (cur_ops->get_gp_state && cur_ops->poll_gp_state)
2082                         WARN_ONCE(cur_ops->poll_gp_state(cookie),
2083                                   "%s: Cookie check 2 failed %s(%d) %lu->%lu\n",
2084                                   __func__,
2085                                   rcu_torture_writer_state_getname(),
2086                                   rcu_torture_writer_state,
2087                                   cookie, cur_ops->get_gp_state());
2088                 if (cur_ops->get_gp_state_full && cur_ops->poll_gp_state_full)
2089                         WARN_ONCE(cur_ops->poll_gp_state_full(&cookie_full),
2090                                   "%s: Cookie check 6 failed %s(%d) online %*pbl\n",
2091                                   __func__,
2092                                   rcu_torture_writer_state_getname(),
2093                                   rcu_torture_writer_state,
2094                                   cpumask_pr_args(cpu_online_mask));
2095         }
2096         rcutorture_one_extend(&readstate, 0, trsp, rtrsp);
2097         WARN_ON_ONCE(readstate);
2098         // This next splat is expected behavior if leakpointer, especially
2099         // for CONFIG_RCU_STRICT_GRACE_PERIOD=y kernels.
2100         WARN_ON_ONCE(leakpointer && READ_ONCE(p->rtort_pipe_count) > 1);
2101
2102         /* If error or close call, record the sequence of reader protections. */
2103         if ((pipe_count > 1 || completed > 1) && !xchg(&err_segs_recorded, 1)) {
2104                 i = 0;
2105                 for (rtrsp1 = &rtseg[0]; rtrsp1 < rtrsp; rtrsp1++)
2106                         err_segs[i++] = *rtrsp1;
2107                 rt_read_nsegs = i;
2108         }
2109
2110         return true;
2111 }
2112
2113 static DEFINE_TORTURE_RANDOM_PERCPU(rcu_torture_timer_rand);
2114
2115 /*
2116  * RCU torture reader from timer handler.  Dereferences rcu_torture_current,
2117  * incrementing the corresponding element of the pipeline array.  The
2118  * counter in the element should never be greater than 1, otherwise, the
2119  * RCU implementation is broken.
2120  */
2121 static void rcu_torture_timer(struct timer_list *unused)
2122 {
2123         atomic_long_inc(&n_rcu_torture_timers);
2124         (void)rcu_torture_one_read(this_cpu_ptr(&rcu_torture_timer_rand), -1);
2125
2126         /* Test call_rcu() invocation from interrupt handler. */
2127         if (cur_ops->call) {
2128                 struct rcu_head *rhp = kmalloc(sizeof(*rhp), GFP_NOWAIT);
2129
2130                 if (rhp)
2131                         cur_ops->call(rhp, rcu_torture_timer_cb);
2132         }
2133 }
2134
2135 /*
2136  * RCU torture reader kthread.  Repeatedly dereferences rcu_torture_current,
2137  * incrementing the corresponding element of the pipeline array.  The
2138  * counter in the element should never be greater than 1, otherwise, the
2139  * RCU implementation is broken.
2140  */
2141 static int
2142 rcu_torture_reader(void *arg)
2143 {
2144         unsigned long lastsleep = jiffies;
2145         long myid = (long)arg;
2146         int mynumonline = myid;
2147         DEFINE_TORTURE_RANDOM(rand);
2148         struct timer_list t;
2149
2150         VERBOSE_TOROUT_STRING("rcu_torture_reader task started");
2151         set_user_nice(current, MAX_NICE);
2152         if (irqreader && cur_ops->irq_capable)
2153                 timer_setup_on_stack(&t, rcu_torture_timer, 0);
2154         tick_dep_set_task(current, TICK_DEP_BIT_RCU);
2155         do {
2156                 if (irqreader && cur_ops->irq_capable) {
2157                         if (!timer_pending(&t))
2158                                 mod_timer(&t, jiffies + 1);
2159                 }
2160                 if (!rcu_torture_one_read(&rand, myid) && !torture_must_stop())
2161                         schedule_timeout_interruptible(HZ);
2162                 if (time_after(jiffies, lastsleep) && !torture_must_stop()) {
2163                         torture_hrtimeout_us(500, 1000, &rand);
2164                         lastsleep = jiffies + 10;
2165                 }
2166                 while (torture_num_online_cpus() < mynumonline && !torture_must_stop())
2167                         schedule_timeout_interruptible(HZ / 5);
2168                 stutter_wait("rcu_torture_reader");
2169         } while (!torture_must_stop());
2170         if (irqreader && cur_ops->irq_capable) {
2171                 del_timer_sync(&t);
2172                 destroy_timer_on_stack(&t);
2173         }
2174         tick_dep_clear_task(current, TICK_DEP_BIT_RCU);
2175         torture_kthread_stopping("rcu_torture_reader");
2176         return 0;
2177 }
2178
2179 /*
2180  * Randomly Toggle CPUs' callback-offload state.  This uses hrtimers to
2181  * increase race probabilities and fuzzes the interval between toggling.
2182  */
2183 static int rcu_nocb_toggle(void *arg)
2184 {
2185         int cpu;
2186         int maxcpu = -1;
2187         int oldnice = task_nice(current);
2188         long r;
2189         DEFINE_TORTURE_RANDOM(rand);
2190         ktime_t toggle_delay;
2191         unsigned long toggle_fuzz;
2192         ktime_t toggle_interval = ms_to_ktime(nocbs_toggle);
2193
2194         VERBOSE_TOROUT_STRING("rcu_nocb_toggle task started");
2195         while (!rcu_inkernel_boot_has_ended())
2196                 schedule_timeout_interruptible(HZ / 10);
2197         for_each_possible_cpu(cpu)
2198                 maxcpu = cpu;
2199         WARN_ON(maxcpu < 0);
2200         if (toggle_interval > ULONG_MAX)
2201                 toggle_fuzz = ULONG_MAX >> 3;
2202         else
2203                 toggle_fuzz = toggle_interval >> 3;
2204         if (toggle_fuzz <= 0)
2205                 toggle_fuzz = NSEC_PER_USEC;
2206         do {
2207                 r = torture_random(&rand);
2208                 cpu = (r >> 1) % (maxcpu + 1);
2209                 if (r & 0x1) {
2210                         rcu_nocb_cpu_offload(cpu);
2211                         atomic_long_inc(&n_nocb_offload);
2212                 } else {
2213                         rcu_nocb_cpu_deoffload(cpu);
2214                         atomic_long_inc(&n_nocb_deoffload);
2215                 }
2216                 toggle_delay = torture_random(&rand) % toggle_fuzz + toggle_interval;
2217                 set_current_state(TASK_INTERRUPTIBLE);
2218                 schedule_hrtimeout(&toggle_delay, HRTIMER_MODE_REL);
2219                 if (stutter_wait("rcu_nocb_toggle"))
2220                         sched_set_normal(current, oldnice);
2221         } while (!torture_must_stop());
2222         torture_kthread_stopping("rcu_nocb_toggle");
2223         return 0;
2224 }
2225
2226 /*
2227  * Print torture statistics.  Caller must ensure that there is only
2228  * one call to this function at a given time!!!  This is normally
2229  * accomplished by relying on the module system to only have one copy
2230  * of the module loaded, and then by giving the rcu_torture_stats
2231  * kthread full control (or the init/cleanup functions when rcu_torture_stats
2232  * thread is not running).
2233  */
2234 static void
2235 rcu_torture_stats_print(void)
2236 {
2237         int cpu;
2238         int i;
2239         long pipesummary[RCU_TORTURE_PIPE_LEN + 1] = { 0 };
2240         long batchsummary[RCU_TORTURE_PIPE_LEN + 1] = { 0 };
2241         struct rcu_torture *rtcp;
2242         static unsigned long rtcv_snap = ULONG_MAX;
2243         static bool splatted;
2244         struct task_struct *wtp;
2245
2246         for_each_possible_cpu(cpu) {
2247                 for (i = 0; i < RCU_TORTURE_PIPE_LEN + 1; i++) {
2248                         pipesummary[i] += READ_ONCE(per_cpu(rcu_torture_count, cpu)[i]);
2249                         batchsummary[i] += READ_ONCE(per_cpu(rcu_torture_batch, cpu)[i]);
2250                 }
2251         }
2252         for (i = RCU_TORTURE_PIPE_LEN; i >= 0; i--) {
2253                 if (pipesummary[i] != 0)
2254                         break;
2255         }
2256
2257         pr_alert("%s%s ", torture_type, TORTURE_FLAG);
2258         rtcp = rcu_access_pointer(rcu_torture_current);
2259         pr_cont("rtc: %p %s: %lu tfle: %d rta: %d rtaf: %d rtf: %d ",
2260                 rtcp,
2261                 rtcp && !rcu_stall_is_suppressed_at_boot() ? "ver" : "VER",
2262                 rcu_torture_current_version,
2263                 list_empty(&rcu_torture_freelist),
2264                 atomic_read(&n_rcu_torture_alloc),
2265                 atomic_read(&n_rcu_torture_alloc_fail),
2266                 atomic_read(&n_rcu_torture_free));
2267         pr_cont("rtmbe: %d rtmbkf: %d/%d rtbe: %ld rtbke: %ld ",
2268                 atomic_read(&n_rcu_torture_mberror),
2269                 atomic_read(&n_rcu_torture_mbchk_fail), atomic_read(&n_rcu_torture_mbchk_tries),
2270                 n_rcu_torture_barrier_error,
2271                 n_rcu_torture_boost_ktrerror);
2272         pr_cont("rtbf: %ld rtb: %ld nt: %ld ",
2273                 n_rcu_torture_boost_failure,
2274                 n_rcu_torture_boosts,
2275                 atomic_long_read(&n_rcu_torture_timers));
2276         torture_onoff_stats();
2277         pr_cont("barrier: %ld/%ld:%ld ",
2278                 data_race(n_barrier_successes),
2279                 data_race(n_barrier_attempts),
2280                 data_race(n_rcu_torture_barrier_error));
2281         pr_cont("read-exits: %ld ", data_race(n_read_exits)); // Statistic.
2282         pr_cont("nocb-toggles: %ld:%ld\n",
2283                 atomic_long_read(&n_nocb_offload), atomic_long_read(&n_nocb_deoffload));
2284
2285         pr_alert("%s%s ", torture_type, TORTURE_FLAG);
2286         if (atomic_read(&n_rcu_torture_mberror) ||
2287             atomic_read(&n_rcu_torture_mbchk_fail) ||
2288             n_rcu_torture_barrier_error || n_rcu_torture_boost_ktrerror ||
2289             n_rcu_torture_boost_failure || i > 1) {
2290                 pr_cont("%s", "!!! ");
2291                 atomic_inc(&n_rcu_torture_error);
2292                 WARN_ON_ONCE(atomic_read(&n_rcu_torture_mberror));
2293                 WARN_ON_ONCE(atomic_read(&n_rcu_torture_mbchk_fail));
2294                 WARN_ON_ONCE(n_rcu_torture_barrier_error);  // rcu_barrier()
2295                 WARN_ON_ONCE(n_rcu_torture_boost_ktrerror); // no boost kthread
2296                 WARN_ON_ONCE(n_rcu_torture_boost_failure); // boost failed (TIMER_SOFTIRQ RT prio?)
2297                 WARN_ON_ONCE(i > 1); // Too-short grace period
2298         }
2299         pr_cont("Reader Pipe: ");
2300         for (i = 0; i < RCU_TORTURE_PIPE_LEN + 1; i++)
2301                 pr_cont(" %ld", pipesummary[i]);
2302         pr_cont("\n");
2303
2304         pr_alert("%s%s ", torture_type, TORTURE_FLAG);
2305         pr_cont("Reader Batch: ");
2306         for (i = 0; i < RCU_TORTURE_PIPE_LEN + 1; i++)
2307                 pr_cont(" %ld", batchsummary[i]);
2308         pr_cont("\n");
2309
2310         pr_alert("%s%s ", torture_type, TORTURE_FLAG);
2311         pr_cont("Free-Block Circulation: ");
2312         for (i = 0; i < RCU_TORTURE_PIPE_LEN + 1; i++) {
2313                 pr_cont(" %d", atomic_read(&rcu_torture_wcount[i]));
2314         }
2315         pr_cont("\n");
2316
2317         if (cur_ops->stats)
2318                 cur_ops->stats();
2319         if (rtcv_snap == rcu_torture_current_version &&
2320             rcu_access_pointer(rcu_torture_current) &&
2321             !rcu_stall_is_suppressed()) {
2322                 int __maybe_unused flags = 0;
2323                 unsigned long __maybe_unused gp_seq = 0;
2324
2325                 if (cur_ops->get_gp_data)
2326                         cur_ops->get_gp_data(&flags, &gp_seq);
2327                 wtp = READ_ONCE(writer_task);
2328                 pr_alert("??? Writer stall state %s(%d) g%lu f%#x ->state %#x cpu %d\n",
2329                          rcu_torture_writer_state_getname(),
2330                          rcu_torture_writer_state, gp_seq, flags,
2331                          wtp == NULL ? ~0U : wtp->__state,
2332                          wtp == NULL ? -1 : (int)task_cpu(wtp));
2333                 if (!splatted && wtp) {
2334                         sched_show_task(wtp);
2335                         splatted = true;
2336                 }
2337                 if (cur_ops->gp_kthread_dbg)
2338                         cur_ops->gp_kthread_dbg();
2339                 rcu_ftrace_dump(DUMP_ALL);
2340         }
2341         rtcv_snap = rcu_torture_current_version;
2342 }
2343
2344 /*
2345  * Periodically prints torture statistics, if periodic statistics printing
2346  * was specified via the stat_interval module parameter.
2347  */
2348 static int
2349 rcu_torture_stats(void *arg)
2350 {
2351         VERBOSE_TOROUT_STRING("rcu_torture_stats task started");
2352         do {
2353                 schedule_timeout_interruptible(stat_interval * HZ);
2354                 rcu_torture_stats_print();
2355                 torture_shutdown_absorb("rcu_torture_stats");
2356         } while (!torture_must_stop());
2357         torture_kthread_stopping("rcu_torture_stats");
2358         return 0;
2359 }
2360
2361 /* Test mem_dump_obj() and friends.  */
2362 static void rcu_torture_mem_dump_obj(void)
2363 {
2364         struct rcu_head *rhp;
2365         struct kmem_cache *kcp;
2366         static int z;
2367
2368         kcp = kmem_cache_create("rcuscale", 136, 8, SLAB_STORE_USER, NULL);
2369         if (WARN_ON_ONCE(!kcp))
2370                 return;
2371         rhp = kmem_cache_alloc(kcp, GFP_KERNEL);
2372         if (WARN_ON_ONCE(!rhp)) {
2373                 kmem_cache_destroy(kcp);
2374                 return;
2375         }
2376         pr_alert("mem_dump_obj() slab test: rcu_torture_stats = %px, &rhp = %px, rhp = %px, &z = %px\n", stats_task, &rhp, rhp, &z);
2377         pr_alert("mem_dump_obj(ZERO_SIZE_PTR):");
2378         mem_dump_obj(ZERO_SIZE_PTR);
2379         pr_alert("mem_dump_obj(NULL):");
2380         mem_dump_obj(NULL);
2381         pr_alert("mem_dump_obj(%px):", &rhp);
2382         mem_dump_obj(&rhp);
2383         pr_alert("mem_dump_obj(%px):", rhp);
2384         mem_dump_obj(rhp);
2385         pr_alert("mem_dump_obj(%px):", &rhp->func);
2386         mem_dump_obj(&rhp->func);
2387         pr_alert("mem_dump_obj(%px):", &z);
2388         mem_dump_obj(&z);
2389         kmem_cache_free(kcp, rhp);
2390         kmem_cache_destroy(kcp);
2391         rhp = kmalloc(sizeof(*rhp), GFP_KERNEL);
2392         if (WARN_ON_ONCE(!rhp))
2393                 return;
2394         pr_alert("mem_dump_obj() kmalloc test: rcu_torture_stats = %px, &rhp = %px, rhp = %px\n", stats_task, &rhp, rhp);
2395         pr_alert("mem_dump_obj(kmalloc %px):", rhp);
2396         mem_dump_obj(rhp);
2397         pr_alert("mem_dump_obj(kmalloc %px):", &rhp->func);
2398         mem_dump_obj(&rhp->func);
2399         kfree(rhp);
2400         rhp = vmalloc(4096);
2401         if (WARN_ON_ONCE(!rhp))
2402                 return;
2403         pr_alert("mem_dump_obj() vmalloc test: rcu_torture_stats = %px, &rhp = %px, rhp = %px\n", stats_task, &rhp, rhp);
2404         pr_alert("mem_dump_obj(vmalloc %px):", rhp);
2405         mem_dump_obj(rhp);
2406         pr_alert("mem_dump_obj(vmalloc %px):", &rhp->func);
2407         mem_dump_obj(&rhp->func);
2408         vfree(rhp);
2409 }
2410
2411 static void
2412 rcu_torture_print_module_parms(struct rcu_torture_ops *cur_ops, const char *tag)
2413 {
2414         pr_alert("%s" TORTURE_FLAG
2415                  "--- %s: nreaders=%d nfakewriters=%d "
2416                  "stat_interval=%d verbose=%d test_no_idle_hz=%d "
2417                  "shuffle_interval=%d stutter=%d irqreader=%d "
2418                  "fqs_duration=%d fqs_holdoff=%d fqs_stutter=%d "
2419                  "test_boost=%d/%d test_boost_interval=%d "
2420                  "test_boost_duration=%d shutdown_secs=%d "
2421                  "stall_cpu=%d stall_cpu_holdoff=%d stall_cpu_irqsoff=%d "
2422                  "stall_cpu_block=%d stall_cpu_repeat=%d "
2423                  "n_barrier_cbs=%d "
2424                  "onoff_interval=%d onoff_holdoff=%d "
2425                  "read_exit_delay=%d read_exit_burst=%d "
2426                  "reader_flavor=%x "
2427                  "nocbs_nthreads=%d nocbs_toggle=%d "
2428                  "test_nmis=%d\n",
2429                  torture_type, tag, nrealreaders, nfakewriters,
2430                  stat_interval, verbose, test_no_idle_hz, shuffle_interval,
2431                  stutter, irqreader, fqs_duration, fqs_holdoff, fqs_stutter,
2432                  test_boost, cur_ops->can_boost,
2433                  test_boost_interval, test_boost_duration, shutdown_secs,
2434                  stall_cpu, stall_cpu_holdoff, stall_cpu_irqsoff,
2435                  stall_cpu_block, stall_cpu_repeat,
2436                  n_barrier_cbs,
2437                  onoff_interval, onoff_holdoff,
2438                  read_exit_delay, read_exit_burst,
2439                  reader_flavor,
2440                  nocbs_nthreads, nocbs_toggle,
2441                  test_nmis);
2442 }
2443
2444 static int rcutorture_booster_cleanup(unsigned int cpu)
2445 {
2446         struct task_struct *t;
2447
2448         if (boost_tasks[cpu] == NULL)
2449                 return 0;
2450         mutex_lock(&boost_mutex);
2451         t = boost_tasks[cpu];
2452         boost_tasks[cpu] = NULL;
2453         rcu_torture_enable_rt_throttle();
2454         mutex_unlock(&boost_mutex);
2455
2456         /* This must be outside of the mutex, otherwise deadlock! */
2457         torture_stop_kthread(rcu_torture_boost, t);
2458         return 0;
2459 }
2460
2461 static int rcutorture_booster_init(unsigned int cpu)
2462 {
2463         int retval;
2464
2465         if (boost_tasks[cpu] != NULL)
2466                 return 0;  /* Already created, nothing more to do. */
2467
2468         // Testing RCU priority boosting requires rcutorture do
2469         // some serious abuse.  Counter this by running ksoftirqd
2470         // at higher priority.
2471         if (IS_BUILTIN(CONFIG_RCU_TORTURE_TEST)) {
2472                 struct sched_param sp;
2473                 struct task_struct *t;
2474
2475                 t = per_cpu(ksoftirqd, cpu);
2476                 WARN_ON_ONCE(!t);
2477                 sp.sched_priority = 2;
2478                 sched_setscheduler_nocheck(t, SCHED_FIFO, &sp);
2479 #ifdef CONFIG_IRQ_FORCED_THREADING
2480                 if (force_irqthreads()) {
2481                         t = per_cpu(ktimerd, cpu);
2482                         WARN_ON_ONCE(!t);
2483                         sp.sched_priority = 2;
2484                         sched_setscheduler_nocheck(t, SCHED_FIFO, &sp);
2485                 }
2486 #endif
2487         }
2488
2489         /* Don't allow time recalculation while creating a new task. */
2490         mutex_lock(&boost_mutex);
2491         rcu_torture_disable_rt_throttle();
2492         VERBOSE_TOROUT_STRING("Creating rcu_torture_boost task");
2493         boost_tasks[cpu] = kthread_run_on_cpu(rcu_torture_boost, NULL,
2494                                               cpu, "rcu_torture_boost_%u");
2495         if (IS_ERR(boost_tasks[cpu])) {
2496                 retval = PTR_ERR(boost_tasks[cpu]);
2497                 VERBOSE_TOROUT_STRING("rcu_torture_boost task create failed");
2498                 n_rcu_torture_boost_ktrerror++;
2499                 boost_tasks[cpu] = NULL;
2500                 mutex_unlock(&boost_mutex);
2501                 return retval;
2502         }
2503         mutex_unlock(&boost_mutex);
2504         return 0;
2505 }
2506
2507 static int rcu_torture_stall_nf(struct notifier_block *nb, unsigned long v, void *ptr)
2508 {
2509         pr_info("%s: v=%lu, duration=%lu.\n", __func__, v, (unsigned long)ptr);
2510         return NOTIFY_OK;
2511 }
2512
2513 static struct notifier_block rcu_torture_stall_block = {
2514         .notifier_call = rcu_torture_stall_nf,
2515 };
2516
2517 /*
2518  * CPU-stall kthread.  It waits as specified by stall_cpu_holdoff, then
2519  * induces a CPU stall for the time specified by stall_cpu.  If a new
2520  * stall test is added, stallsdone in rcu_torture_writer() must be adjusted.
2521  */
2522 static void rcu_torture_stall_one(int rep, int irqsoff)
2523 {
2524         int idx;
2525         unsigned long stop_at;
2526
2527         if (stall_cpu_holdoff > 0) {
2528                 VERBOSE_TOROUT_STRING("rcu_torture_stall begin holdoff");
2529                 schedule_timeout_interruptible(stall_cpu_holdoff * HZ);
2530                 VERBOSE_TOROUT_STRING("rcu_torture_stall end holdoff");
2531         }
2532         if (!kthread_should_stop() && stall_gp_kthread > 0) {
2533                 VERBOSE_TOROUT_STRING("rcu_torture_stall begin GP stall");
2534                 rcu_gp_set_torture_wait(stall_gp_kthread * HZ);
2535                 for (idx = 0; idx < stall_gp_kthread + 2; idx++) {
2536                         if (kthread_should_stop())
2537                                 break;
2538                         schedule_timeout_uninterruptible(HZ);
2539                 }
2540         }
2541         if (!kthread_should_stop() && stall_cpu > 0) {
2542                 VERBOSE_TOROUT_STRING("rcu_torture_stall begin CPU stall");
2543                 stop_at = ktime_get_seconds() + stall_cpu;
2544                 /* RCU CPU stall is expected behavior in following code. */
2545                 idx = cur_ops->readlock();
2546                 if (irqsoff)
2547                         local_irq_disable();
2548                 else if (!stall_cpu_block)
2549                         preempt_disable();
2550                 pr_alert("%s start stall episode %d on CPU %d.\n",
2551                           __func__, rep + 1, raw_smp_processor_id());
2552                 while (ULONG_CMP_LT((unsigned long)ktime_get_seconds(), stop_at) &&
2553                        !kthread_should_stop())
2554                         if (stall_cpu_block) {
2555 #ifdef CONFIG_PREEMPTION
2556                                 preempt_schedule();
2557 #else
2558                                 schedule_timeout_uninterruptible(HZ);
2559 #endif
2560                         } else if (stall_no_softlockup) {
2561                                 touch_softlockup_watchdog();
2562                         }
2563                 if (irqsoff)
2564                         local_irq_enable();
2565                 else if (!stall_cpu_block)
2566                         preempt_enable();
2567                 cur_ops->readunlock(idx);
2568         }
2569 }
2570
2571 /*
2572  * CPU-stall kthread.  Invokes rcu_torture_stall_one() once, and then as many
2573  * additional times as specified by the stall_cpu_repeat module parameter.
2574  * Note that stall_cpu_irqsoff is ignored on the second and subsequent
2575  * stall.
2576  */
2577 static int rcu_torture_stall(void *args)
2578 {
2579         int i;
2580         int repeat = stall_cpu_repeat;
2581         int ret;
2582
2583         VERBOSE_TOROUT_STRING("rcu_torture_stall task started");
2584         if (repeat < 0) {
2585                 repeat = 0;
2586                 WARN_ON_ONCE(IS_BUILTIN(CONFIG_RCU_TORTURE_TEST));
2587         }
2588         if (rcu_cpu_stall_notifiers) {
2589                 ret = rcu_stall_chain_notifier_register(&rcu_torture_stall_block);
2590                 if (ret)
2591                         pr_info("%s: rcu_stall_chain_notifier_register() returned %d, %sexpected.\n",
2592                                 __func__, ret, !IS_ENABLED(CONFIG_RCU_STALL_COMMON) ? "un" : "");
2593         }
2594         for (i = 0; i <= repeat; i++) {
2595                 if (kthread_should_stop())
2596                         break;
2597                 rcu_torture_stall_one(i, i == 0 ? stall_cpu_irqsoff : 0);
2598         }
2599         pr_alert("%s end.\n", __func__);
2600         if (rcu_cpu_stall_notifiers && !ret) {
2601                 ret = rcu_stall_chain_notifier_unregister(&rcu_torture_stall_block);
2602                 if (ret)
2603                         pr_info("%s: rcu_stall_chain_notifier_unregister() returned %d.\n", __func__, ret);
2604         }
2605         torture_shutdown_absorb("rcu_torture_stall");
2606         while (!kthread_should_stop())
2607                 schedule_timeout_interruptible(10 * HZ);
2608         return 0;
2609 }
2610
2611 /* Spawn CPU-stall kthread, if stall_cpu specified. */
2612 static int __init rcu_torture_stall_init(void)
2613 {
2614         if (stall_cpu <= 0 && stall_gp_kthread <= 0)
2615                 return 0;
2616         return torture_create_kthread(rcu_torture_stall, NULL, stall_task);
2617 }
2618
2619 /* State structure for forward-progress self-propagating RCU callback. */
2620 struct fwd_cb_state {
2621         struct rcu_head rh;
2622         int stop;
2623 };
2624
2625 /*
2626  * Forward-progress self-propagating RCU callback function.  Because
2627  * callbacks run from softirq, this function is an implicit RCU read-side
2628  * critical section.
2629  */
2630 static void rcu_torture_fwd_prog_cb(struct rcu_head *rhp)
2631 {
2632         struct fwd_cb_state *fcsp = container_of(rhp, struct fwd_cb_state, rh);
2633
2634         if (READ_ONCE(fcsp->stop)) {
2635                 WRITE_ONCE(fcsp->stop, 2);
2636                 return;
2637         }
2638         cur_ops->call(&fcsp->rh, rcu_torture_fwd_prog_cb);
2639 }
2640
2641 /* State for continuous-flood RCU callbacks. */
2642 struct rcu_fwd_cb {
2643         struct rcu_head rh;
2644         struct rcu_fwd_cb *rfc_next;
2645         struct rcu_fwd *rfc_rfp;
2646         int rfc_gps;
2647 };
2648
2649 #define MAX_FWD_CB_JIFFIES      (8 * HZ) /* Maximum CB test duration. */
2650 #define MIN_FWD_CB_LAUNDERS     3       /* This many CB invocations to count. */
2651 #define MIN_FWD_CBS_LAUNDERED   100     /* Number of counted CBs. */
2652 #define FWD_CBS_HIST_DIV        10      /* Histogram buckets/second. */
2653 #define N_LAUNDERS_HIST (2 * MAX_FWD_CB_JIFFIES / (HZ / FWD_CBS_HIST_DIV))
2654
2655 struct rcu_launder_hist {
2656         long n_launders;
2657         unsigned long launder_gp_seq;
2658 };
2659
2660 struct rcu_fwd {
2661         spinlock_t rcu_fwd_lock;
2662         struct rcu_fwd_cb *rcu_fwd_cb_head;
2663         struct rcu_fwd_cb **rcu_fwd_cb_tail;
2664         long n_launders_cb;
2665         unsigned long rcu_fwd_startat;
2666         struct rcu_launder_hist n_launders_hist[N_LAUNDERS_HIST];
2667         unsigned long rcu_launder_gp_seq_start;
2668         int rcu_fwd_id;
2669 };
2670
2671 static DEFINE_MUTEX(rcu_fwd_mutex);
2672 static struct rcu_fwd *rcu_fwds;
2673 static unsigned long rcu_fwd_seq;
2674 static atomic_long_t rcu_fwd_max_cbs;
2675 static bool rcu_fwd_emergency_stop;
2676
2677 static void rcu_torture_fwd_cb_hist(struct rcu_fwd *rfp)
2678 {
2679         unsigned long gps;
2680         unsigned long gps_old;
2681         int i;
2682         int j;
2683
2684         for (i = ARRAY_SIZE(rfp->n_launders_hist) - 1; i > 0; i--)
2685                 if (rfp->n_launders_hist[i].n_launders > 0)
2686                         break;
2687         pr_alert("%s: Callback-invocation histogram %d (duration %lu jiffies):",
2688                  __func__, rfp->rcu_fwd_id, jiffies - rfp->rcu_fwd_startat);
2689         gps_old = rfp->rcu_launder_gp_seq_start;
2690         for (j = 0; j <= i; j++) {
2691                 gps = rfp->n_launders_hist[j].launder_gp_seq;
2692                 pr_cont(" %ds/%d: %ld:%ld",
2693                         j + 1, FWD_CBS_HIST_DIV,
2694                         rfp->n_launders_hist[j].n_launders,
2695                         rcutorture_seq_diff(gps, gps_old));
2696                 gps_old = gps;
2697         }
2698         pr_cont("\n");
2699 }
2700
2701 /* Callback function for continuous-flood RCU callbacks. */
2702 static void rcu_torture_fwd_cb_cr(struct rcu_head *rhp)
2703 {
2704         unsigned long flags;
2705         int i;
2706         struct rcu_fwd_cb *rfcp = container_of(rhp, struct rcu_fwd_cb, rh);
2707         struct rcu_fwd_cb **rfcpp;
2708         struct rcu_fwd *rfp = rfcp->rfc_rfp;
2709
2710         rfcp->rfc_next = NULL;
2711         rfcp->rfc_gps++;
2712         spin_lock_irqsave(&rfp->rcu_fwd_lock, flags);
2713         rfcpp = rfp->rcu_fwd_cb_tail;
2714         rfp->rcu_fwd_cb_tail = &rfcp->rfc_next;
2715         smp_store_release(rfcpp, rfcp);
2716         WRITE_ONCE(rfp->n_launders_cb, rfp->n_launders_cb + 1);
2717         i = ((jiffies - rfp->rcu_fwd_startat) / (HZ / FWD_CBS_HIST_DIV));
2718         if (i >= ARRAY_SIZE(rfp->n_launders_hist))
2719                 i = ARRAY_SIZE(rfp->n_launders_hist) - 1;
2720         rfp->n_launders_hist[i].n_launders++;
2721         rfp->n_launders_hist[i].launder_gp_seq = cur_ops->get_gp_seq();
2722         spin_unlock_irqrestore(&rfp->rcu_fwd_lock, flags);
2723 }
2724
2725 // Give the scheduler a chance, even on nohz_full CPUs.
2726 static void rcu_torture_fwd_prog_cond_resched(unsigned long iter)
2727 {
2728         if (IS_ENABLED(CONFIG_PREEMPTION) && IS_ENABLED(CONFIG_NO_HZ_FULL)) {
2729                 // Real call_rcu() floods hit userspace, so emulate that.
2730                 if (need_resched() || (iter & 0xfff))
2731                         schedule();
2732                 return;
2733         }
2734         // No userspace emulation: CB invocation throttles call_rcu()
2735         cond_resched();
2736 }
2737
2738 /*
2739  * Free all callbacks on the rcu_fwd_cb_head list, either because the
2740  * test is over or because we hit an OOM event.
2741  */
2742 static unsigned long rcu_torture_fwd_prog_cbfree(struct rcu_fwd *rfp)
2743 {
2744         unsigned long flags;
2745         unsigned long freed = 0;
2746         struct rcu_fwd_cb *rfcp;
2747
2748         for (;;) {
2749                 spin_lock_irqsave(&rfp->rcu_fwd_lock, flags);
2750                 rfcp = rfp->rcu_fwd_cb_head;
2751                 if (!rfcp) {
2752                         spin_unlock_irqrestore(&rfp->rcu_fwd_lock, flags);
2753                         break;
2754                 }
2755                 rfp->rcu_fwd_cb_head = rfcp->rfc_next;
2756                 if (!rfp->rcu_fwd_cb_head)
2757                         rfp->rcu_fwd_cb_tail = &rfp->rcu_fwd_cb_head;
2758                 spin_unlock_irqrestore(&rfp->rcu_fwd_lock, flags);
2759                 kfree(rfcp);
2760                 freed++;
2761                 rcu_torture_fwd_prog_cond_resched(freed);
2762                 if (tick_nohz_full_enabled()) {
2763                         local_irq_save(flags);
2764                         rcu_momentary_eqs();
2765                         local_irq_restore(flags);
2766                 }
2767         }
2768         return freed;
2769 }
2770
2771 /* Carry out need_resched()/cond_resched() forward-progress testing. */
2772 static void rcu_torture_fwd_prog_nr(struct rcu_fwd *rfp,
2773                                     int *tested, int *tested_tries)
2774 {
2775         unsigned long cver;
2776         unsigned long dur;
2777         struct fwd_cb_state fcs;
2778         unsigned long gps;
2779         int idx;
2780         int sd;
2781         int sd4;
2782         bool selfpropcb = false;
2783         unsigned long stopat;
2784         static DEFINE_TORTURE_RANDOM(trs);
2785
2786         pr_alert("%s: Starting forward-progress test %d\n", __func__, rfp->rcu_fwd_id);
2787         if (!cur_ops->sync)
2788                 return; // Cannot do need_resched() forward progress testing without ->sync.
2789         if (cur_ops->call && cur_ops->cb_barrier) {
2790                 init_rcu_head_on_stack(&fcs.rh);
2791                 selfpropcb = true;
2792         }
2793
2794         /* Tight loop containing cond_resched(). */
2795         atomic_inc(&rcu_fwd_cb_nodelay);
2796         cur_ops->sync(); /* Later readers see above write. */
2797         if  (selfpropcb) {
2798                 WRITE_ONCE(fcs.stop, 0);
2799                 cur_ops->call(&fcs.rh, rcu_torture_fwd_prog_cb);
2800         }
2801         cver = READ_ONCE(rcu_torture_current_version);
2802         gps = cur_ops->get_gp_seq();
2803         sd = cur_ops->stall_dur() + 1;
2804         sd4 = (sd + fwd_progress_div - 1) / fwd_progress_div;
2805         dur = sd4 + torture_random(&trs) % (sd - sd4);
2806         WRITE_ONCE(rfp->rcu_fwd_startat, jiffies);
2807         stopat = rfp->rcu_fwd_startat + dur;
2808         while (time_before(jiffies, stopat) &&
2809                !shutdown_time_arrived() &&
2810                !READ_ONCE(rcu_fwd_emergency_stop) && !torture_must_stop()) {
2811                 idx = cur_ops->readlock();
2812                 udelay(10);
2813                 cur_ops->readunlock(idx);
2814                 if (!fwd_progress_need_resched || need_resched())
2815                         cond_resched();
2816         }
2817         (*tested_tries)++;
2818         if (!time_before(jiffies, stopat) &&
2819             !shutdown_time_arrived() &&
2820             !READ_ONCE(rcu_fwd_emergency_stop) && !torture_must_stop()) {
2821                 (*tested)++;
2822                 cver = READ_ONCE(rcu_torture_current_version) - cver;
2823                 gps = rcutorture_seq_diff(cur_ops->get_gp_seq(), gps);
2824                 WARN_ON(!cver && gps < 2);
2825                 pr_alert("%s: %d Duration %ld cver %ld gps %ld\n", __func__,
2826                          rfp->rcu_fwd_id, dur, cver, gps);
2827         }
2828         if (selfpropcb) {
2829                 WRITE_ONCE(fcs.stop, 1);
2830                 cur_ops->sync(); /* Wait for running CB to complete. */
2831                 pr_alert("%s: Waiting for CBs: %pS() %d\n", __func__, cur_ops->cb_barrier, rfp->rcu_fwd_id);
2832                 cur_ops->cb_barrier(); /* Wait for queued callbacks. */
2833         }
2834
2835         if (selfpropcb) {
2836                 WARN_ON(READ_ONCE(fcs.stop) != 2);
2837                 destroy_rcu_head_on_stack(&fcs.rh);
2838         }
2839         schedule_timeout_uninterruptible(HZ / 10); /* Let kthreads recover. */
2840         atomic_dec(&rcu_fwd_cb_nodelay);
2841 }
2842
2843 /* Carry out call_rcu() forward-progress testing. */
2844 static void rcu_torture_fwd_prog_cr(struct rcu_fwd *rfp)
2845 {
2846         unsigned long cver;
2847         unsigned long flags;
2848         unsigned long gps;
2849         int i;
2850         long n_launders;
2851         long n_launders_cb_snap;
2852         long n_launders_sa;
2853         long n_max_cbs;
2854         long n_max_gps;
2855         struct rcu_fwd_cb *rfcp;
2856         struct rcu_fwd_cb *rfcpn;
2857         unsigned long stopat;
2858         unsigned long stoppedat;
2859
2860         pr_alert("%s: Starting forward-progress test %d\n", __func__, rfp->rcu_fwd_id);
2861         if (READ_ONCE(rcu_fwd_emergency_stop))
2862                 return; /* Get out of the way quickly, no GP wait! */
2863         if (!cur_ops->call)
2864                 return; /* Can't do call_rcu() fwd prog without ->call. */
2865
2866         /* Loop continuously posting RCU callbacks. */
2867         atomic_inc(&rcu_fwd_cb_nodelay);
2868         cur_ops->sync(); /* Later readers see above write. */
2869         WRITE_ONCE(rfp->rcu_fwd_startat, jiffies);
2870         stopat = rfp->rcu_fwd_startat + MAX_FWD_CB_JIFFIES;
2871         n_launders = 0;
2872         rfp->n_launders_cb = 0; // Hoist initialization for multi-kthread
2873         n_launders_sa = 0;
2874         n_max_cbs = 0;
2875         n_max_gps = 0;
2876         for (i = 0; i < ARRAY_SIZE(rfp->n_launders_hist); i++)
2877                 rfp->n_launders_hist[i].n_launders = 0;
2878         cver = READ_ONCE(rcu_torture_current_version);
2879         gps = cur_ops->get_gp_seq();
2880         rfp->rcu_launder_gp_seq_start = gps;
2881         tick_dep_set_task(current, TICK_DEP_BIT_RCU);
2882         while (time_before(jiffies, stopat) &&
2883                !shutdown_time_arrived() &&
2884                !READ_ONCE(rcu_fwd_emergency_stop) && !torture_must_stop()) {
2885                 rfcp = READ_ONCE(rfp->rcu_fwd_cb_head);
2886                 rfcpn = NULL;
2887                 if (rfcp)
2888                         rfcpn = READ_ONCE(rfcp->rfc_next);
2889                 if (rfcpn) {
2890                         if (rfcp->rfc_gps >= MIN_FWD_CB_LAUNDERS &&
2891                             ++n_max_gps >= MIN_FWD_CBS_LAUNDERED)
2892                                 break;
2893                         rfp->rcu_fwd_cb_head = rfcpn;
2894                         n_launders++;
2895                         n_launders_sa++;
2896                 } else if (!cur_ops->cbflood_max || cur_ops->cbflood_max > n_max_cbs) {
2897                         rfcp = kmalloc(sizeof(*rfcp), GFP_KERNEL);
2898                         if (WARN_ON_ONCE(!rfcp)) {
2899                                 schedule_timeout_interruptible(1);
2900                                 continue;
2901                         }
2902                         n_max_cbs++;
2903                         n_launders_sa = 0;
2904                         rfcp->rfc_gps = 0;
2905                         rfcp->rfc_rfp = rfp;
2906                 } else {
2907                         rfcp = NULL;
2908                 }
2909                 if (rfcp)
2910                         cur_ops->call(&rfcp->rh, rcu_torture_fwd_cb_cr);
2911                 rcu_torture_fwd_prog_cond_resched(n_launders + n_max_cbs);
2912                 if (tick_nohz_full_enabled()) {
2913                         local_irq_save(flags);
2914                         rcu_momentary_eqs();
2915                         local_irq_restore(flags);
2916                 }
2917         }
2918         stoppedat = jiffies;
2919         n_launders_cb_snap = READ_ONCE(rfp->n_launders_cb);
2920         cver = READ_ONCE(rcu_torture_current_version) - cver;
2921         gps = rcutorture_seq_diff(cur_ops->get_gp_seq(), gps);
2922         pr_alert("%s: Waiting for CBs: %pS() %d\n", __func__, cur_ops->cb_barrier, rfp->rcu_fwd_id);
2923         cur_ops->cb_barrier(); /* Wait for callbacks to be invoked. */
2924         (void)rcu_torture_fwd_prog_cbfree(rfp);
2925
2926         if (!torture_must_stop() && !READ_ONCE(rcu_fwd_emergency_stop) &&
2927             !shutdown_time_arrived()) {
2928                 if (WARN_ON(n_max_gps < MIN_FWD_CBS_LAUNDERED) && cur_ops->gp_kthread_dbg)
2929                         cur_ops->gp_kthread_dbg();
2930                 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 #online %u\n",
2931                          __func__,
2932                          stoppedat - rfp->rcu_fwd_startat, jiffies - stoppedat,
2933                          n_launders + n_max_cbs - n_launders_cb_snap,
2934                          n_launders, n_launders_sa,
2935                          n_max_gps, n_max_cbs, cver, gps, num_online_cpus());
2936                 atomic_long_add(n_max_cbs, &rcu_fwd_max_cbs);
2937                 mutex_lock(&rcu_fwd_mutex); // Serialize histograms.
2938                 rcu_torture_fwd_cb_hist(rfp);
2939                 mutex_unlock(&rcu_fwd_mutex);
2940         }
2941         schedule_timeout_uninterruptible(HZ); /* Let CBs drain. */
2942         tick_dep_clear_task(current, TICK_DEP_BIT_RCU);
2943         atomic_dec(&rcu_fwd_cb_nodelay);
2944 }
2945
2946
2947 /*
2948  * OOM notifier, but this only prints diagnostic information for the
2949  * current forward-progress test.
2950  */
2951 static int rcutorture_oom_notify(struct notifier_block *self,
2952                                  unsigned long notused, void *nfreed)
2953 {
2954         int i;
2955         long ncbs;
2956         struct rcu_fwd *rfp;
2957
2958         mutex_lock(&rcu_fwd_mutex);
2959         rfp = rcu_fwds;
2960         if (!rfp) {
2961                 mutex_unlock(&rcu_fwd_mutex);
2962                 return NOTIFY_OK;
2963         }
2964         WARN(1, "%s invoked upon OOM during forward-progress testing.\n",
2965              __func__);
2966         for (i = 0; i < fwd_progress; i++) {
2967                 rcu_torture_fwd_cb_hist(&rfp[i]);
2968                 rcu_fwd_progress_check(1 + (jiffies - READ_ONCE(rfp[i].rcu_fwd_startat)) / 2);
2969         }
2970         WRITE_ONCE(rcu_fwd_emergency_stop, true);
2971         smp_mb(); /* Emergency stop before free and wait to avoid hangs. */
2972         ncbs = 0;
2973         for (i = 0; i < fwd_progress; i++)
2974                 ncbs += rcu_torture_fwd_prog_cbfree(&rfp[i]);
2975         pr_info("%s: Freed %lu RCU callbacks.\n", __func__, ncbs);
2976         cur_ops->cb_barrier();
2977         ncbs = 0;
2978         for (i = 0; i < fwd_progress; i++)
2979                 ncbs += rcu_torture_fwd_prog_cbfree(&rfp[i]);
2980         pr_info("%s: Freed %lu RCU callbacks.\n", __func__, ncbs);
2981         cur_ops->cb_barrier();
2982         ncbs = 0;
2983         for (i = 0; i < fwd_progress; i++)
2984                 ncbs += rcu_torture_fwd_prog_cbfree(&rfp[i]);
2985         pr_info("%s: Freed %lu RCU callbacks.\n", __func__, ncbs);
2986         smp_mb(); /* Frees before return to avoid redoing OOM. */
2987         (*(unsigned long *)nfreed)++; /* Forward progress CBs freed! */
2988         pr_info("%s returning after OOM processing.\n", __func__);
2989         mutex_unlock(&rcu_fwd_mutex);
2990         return NOTIFY_OK;
2991 }
2992
2993 static struct notifier_block rcutorture_oom_nb = {
2994         .notifier_call = rcutorture_oom_notify
2995 };
2996
2997 /* Carry out grace-period forward-progress testing. */
2998 static int rcu_torture_fwd_prog(void *args)
2999 {
3000         bool firsttime = true;
3001         long max_cbs;
3002         int oldnice = task_nice(current);
3003         unsigned long oldseq = READ_ONCE(rcu_fwd_seq);
3004         struct rcu_fwd *rfp = args;
3005         int tested = 0;
3006         int tested_tries = 0;
3007
3008         VERBOSE_TOROUT_STRING("rcu_torture_fwd_progress task started");
3009         rcu_bind_current_to_nocb();
3010         if (!IS_ENABLED(CONFIG_SMP) || !IS_ENABLED(CONFIG_RCU_BOOST))
3011                 set_user_nice(current, MAX_NICE);
3012         do {
3013                 if (!rfp->rcu_fwd_id) {
3014                         schedule_timeout_interruptible(fwd_progress_holdoff * HZ);
3015                         WRITE_ONCE(rcu_fwd_emergency_stop, false);
3016                         if (!firsttime) {
3017                                 max_cbs = atomic_long_xchg(&rcu_fwd_max_cbs, 0);
3018                                 pr_alert("%s n_max_cbs: %ld\n", __func__, max_cbs);
3019                         }
3020                         firsttime = false;
3021                         WRITE_ONCE(rcu_fwd_seq, rcu_fwd_seq + 1);
3022                 } else {
3023                         while (READ_ONCE(rcu_fwd_seq) == oldseq && !torture_must_stop())
3024                                 schedule_timeout_interruptible(HZ / 20);
3025                         oldseq = READ_ONCE(rcu_fwd_seq);
3026                 }
3027                 pr_alert("%s: Starting forward-progress test %d\n", __func__, rfp->rcu_fwd_id);
3028                 if (rcu_inkernel_boot_has_ended() && torture_num_online_cpus() > rfp->rcu_fwd_id)
3029                         rcu_torture_fwd_prog_cr(rfp);
3030                 if ((cur_ops->stall_dur && cur_ops->stall_dur() > 0) &&
3031                     (!IS_ENABLED(CONFIG_TINY_RCU) ||
3032                      (rcu_inkernel_boot_has_ended() &&
3033                       torture_num_online_cpus() > rfp->rcu_fwd_id)))
3034                         rcu_torture_fwd_prog_nr(rfp, &tested, &tested_tries);
3035
3036                 /* Avoid slow periods, better to test when busy. */
3037                 if (stutter_wait("rcu_torture_fwd_prog"))
3038                         sched_set_normal(current, oldnice);
3039         } while (!torture_must_stop());
3040         /* Short runs might not contain a valid forward-progress attempt. */
3041         if (!rfp->rcu_fwd_id) {
3042                 WARN_ON(!tested && tested_tries >= 5);
3043                 pr_alert("%s: tested %d tested_tries %d\n", __func__, tested, tested_tries);
3044         }
3045         torture_kthread_stopping("rcu_torture_fwd_prog");
3046         return 0;
3047 }
3048
3049 /* If forward-progress checking is requested and feasible, spawn the thread. */
3050 static int __init rcu_torture_fwd_prog_init(void)
3051 {
3052         int i;
3053         int ret = 0;
3054         struct rcu_fwd *rfp;
3055
3056         if (!fwd_progress)
3057                 return 0; /* Not requested, so don't do it. */
3058         if (fwd_progress >= nr_cpu_ids) {
3059                 VERBOSE_TOROUT_STRING("rcu_torture_fwd_prog_init: Limiting fwd_progress to # CPUs.\n");
3060                 fwd_progress = nr_cpu_ids;
3061         } else if (fwd_progress < 0) {
3062                 fwd_progress = nr_cpu_ids;
3063         }
3064         if ((!cur_ops->sync && !cur_ops->call) ||
3065             (!cur_ops->cbflood_max && (!cur_ops->stall_dur || cur_ops->stall_dur() <= 0)) ||
3066             cur_ops == &rcu_busted_ops) {
3067                 VERBOSE_TOROUT_STRING("rcu_torture_fwd_prog_init: Disabled, unsupported by RCU flavor under test");
3068                 fwd_progress = 0;
3069                 return 0;
3070         }
3071         if (stall_cpu > 0) {
3072                 VERBOSE_TOROUT_STRING("rcu_torture_fwd_prog_init: Disabled, conflicts with CPU-stall testing");
3073                 fwd_progress = 0;
3074                 if (IS_MODULE(CONFIG_RCU_TORTURE_TEST))
3075                         return -EINVAL; /* In module, can fail back to user. */
3076                 WARN_ON(1); /* Make sure rcutorture notices conflict. */
3077                 return 0;
3078         }
3079         if (fwd_progress_holdoff <= 0)
3080                 fwd_progress_holdoff = 1;
3081         if (fwd_progress_div <= 0)
3082                 fwd_progress_div = 4;
3083         rfp = kcalloc(fwd_progress, sizeof(*rfp), GFP_KERNEL);
3084         fwd_prog_tasks = kcalloc(fwd_progress, sizeof(*fwd_prog_tasks), GFP_KERNEL);
3085         if (!rfp || !fwd_prog_tasks) {
3086                 kfree(rfp);
3087                 kfree(fwd_prog_tasks);
3088                 fwd_prog_tasks = NULL;
3089                 fwd_progress = 0;
3090                 return -ENOMEM;
3091         }
3092         for (i = 0; i < fwd_progress; i++) {
3093                 spin_lock_init(&rfp[i].rcu_fwd_lock);
3094                 rfp[i].rcu_fwd_cb_tail = &rfp[i].rcu_fwd_cb_head;
3095                 rfp[i].rcu_fwd_id = i;
3096         }
3097         mutex_lock(&rcu_fwd_mutex);
3098         rcu_fwds = rfp;
3099         mutex_unlock(&rcu_fwd_mutex);
3100         register_oom_notifier(&rcutorture_oom_nb);
3101         for (i = 0; i < fwd_progress; i++) {
3102                 ret = torture_create_kthread(rcu_torture_fwd_prog, &rcu_fwds[i], fwd_prog_tasks[i]);
3103                 if (ret) {
3104                         fwd_progress = i;
3105                         return ret;
3106                 }
3107         }
3108         return 0;
3109 }
3110
3111 static void rcu_torture_fwd_prog_cleanup(void)
3112 {
3113         int i;
3114         struct rcu_fwd *rfp;
3115
3116         if (!rcu_fwds || !fwd_prog_tasks)
3117                 return;
3118         for (i = 0; i < fwd_progress; i++)
3119                 torture_stop_kthread(rcu_torture_fwd_prog, fwd_prog_tasks[i]);
3120         unregister_oom_notifier(&rcutorture_oom_nb);
3121         mutex_lock(&rcu_fwd_mutex);
3122         rfp = rcu_fwds;
3123         rcu_fwds = NULL;
3124         mutex_unlock(&rcu_fwd_mutex);
3125         kfree(rfp);
3126         kfree(fwd_prog_tasks);
3127         fwd_prog_tasks = NULL;
3128 }
3129
3130 /* Callback function for RCU barrier testing. */
3131 static void rcu_torture_barrier_cbf(struct rcu_head *rcu)
3132 {
3133         atomic_inc(&barrier_cbs_invoked);
3134 }
3135
3136 /* IPI handler to get callback posted on desired CPU, if online. */
3137 static int rcu_torture_barrier1cb(void *rcu_void)
3138 {
3139         struct rcu_head *rhp = rcu_void;
3140
3141         cur_ops->call(rhp, rcu_torture_barrier_cbf);
3142         return 0;
3143 }
3144
3145 /* kthread function to register callbacks used to test RCU barriers. */
3146 static int rcu_torture_barrier_cbs(void *arg)
3147 {
3148         long myid = (long)arg;
3149         bool lastphase = false;
3150         bool newphase;
3151         struct rcu_head rcu;
3152
3153         init_rcu_head_on_stack(&rcu);
3154         VERBOSE_TOROUT_STRING("rcu_torture_barrier_cbs task started");
3155         set_user_nice(current, MAX_NICE);
3156         do {
3157                 wait_event(barrier_cbs_wq[myid],
3158                            (newphase =
3159                             smp_load_acquire(&barrier_phase)) != lastphase ||
3160                            torture_must_stop());
3161                 lastphase = newphase;
3162                 if (torture_must_stop())
3163                         break;
3164                 /*
3165                  * The above smp_load_acquire() ensures barrier_phase load
3166                  * is ordered before the following ->call().
3167                  */
3168                 if (smp_call_on_cpu(myid, rcu_torture_barrier1cb, &rcu, 1))
3169                         cur_ops->call(&rcu, rcu_torture_barrier_cbf);
3170
3171                 if (atomic_dec_and_test(&barrier_cbs_count))
3172                         wake_up(&barrier_wq);
3173         } while (!torture_must_stop());
3174         if (cur_ops->cb_barrier != NULL)
3175                 cur_ops->cb_barrier();
3176         destroy_rcu_head_on_stack(&rcu);
3177         torture_kthread_stopping("rcu_torture_barrier_cbs");
3178         return 0;
3179 }
3180
3181 /* kthread function to drive and coordinate RCU barrier testing. */
3182 static int rcu_torture_barrier(void *arg)
3183 {
3184         int i;
3185
3186         VERBOSE_TOROUT_STRING("rcu_torture_barrier task starting");
3187         do {
3188                 atomic_set(&barrier_cbs_invoked, 0);
3189                 atomic_set(&barrier_cbs_count, n_barrier_cbs);
3190                 /* Ensure barrier_phase ordered after prior assignments. */
3191                 smp_store_release(&barrier_phase, !barrier_phase);
3192                 for (i = 0; i < n_barrier_cbs; i++)
3193                         wake_up(&barrier_cbs_wq[i]);
3194                 wait_event(barrier_wq,
3195                            atomic_read(&barrier_cbs_count) == 0 ||
3196                            torture_must_stop());
3197                 if (torture_must_stop())
3198                         break;
3199                 n_barrier_attempts++;
3200                 cur_ops->cb_barrier(); /* Implies smp_mb() for wait_event(). */
3201                 if (atomic_read(&barrier_cbs_invoked) != n_barrier_cbs) {
3202                         n_rcu_torture_barrier_error++;
3203                         pr_err("barrier_cbs_invoked = %d, n_barrier_cbs = %d\n",
3204                                atomic_read(&barrier_cbs_invoked),
3205                                n_barrier_cbs);
3206                         WARN_ON(1);
3207                         // Wait manually for the remaining callbacks
3208                         i = 0;
3209                         do {
3210                                 if (WARN_ON(i++ > HZ))
3211                                         i = INT_MIN;
3212                                 schedule_timeout_interruptible(1);
3213                                 cur_ops->cb_barrier();
3214                         } while (atomic_read(&barrier_cbs_invoked) !=
3215                                  n_barrier_cbs &&
3216                                  !torture_must_stop());
3217                         smp_mb(); // Can't trust ordering if broken.
3218                         if (!torture_must_stop())
3219                                 pr_err("Recovered: barrier_cbs_invoked = %d\n",
3220                                        atomic_read(&barrier_cbs_invoked));
3221                 } else {
3222                         n_barrier_successes++;
3223                 }
3224                 schedule_timeout_interruptible(HZ / 10);
3225         } while (!torture_must_stop());
3226         torture_kthread_stopping("rcu_torture_barrier");
3227         return 0;
3228 }
3229
3230 /* Initialize RCU barrier testing. */
3231 static int rcu_torture_barrier_init(void)
3232 {
3233         int i;
3234         int ret;
3235
3236         if (n_barrier_cbs <= 0)
3237                 return 0;
3238         if (cur_ops->call == NULL || cur_ops->cb_barrier == NULL) {
3239                 pr_alert("%s" TORTURE_FLAG
3240                          " Call or barrier ops missing for %s,\n",
3241                          torture_type, cur_ops->name);
3242                 pr_alert("%s" TORTURE_FLAG
3243                          " RCU barrier testing omitted from run.\n",
3244                          torture_type);
3245                 return 0;
3246         }
3247         atomic_set(&barrier_cbs_count, 0);
3248         atomic_set(&barrier_cbs_invoked, 0);
3249         barrier_cbs_tasks =
3250                 kcalloc(n_barrier_cbs, sizeof(barrier_cbs_tasks[0]),
3251                         GFP_KERNEL);
3252         barrier_cbs_wq =
3253                 kcalloc(n_barrier_cbs, sizeof(barrier_cbs_wq[0]), GFP_KERNEL);
3254         if (barrier_cbs_tasks == NULL || !barrier_cbs_wq)
3255                 return -ENOMEM;
3256         for (i = 0; i < n_barrier_cbs; i++) {
3257                 init_waitqueue_head(&barrier_cbs_wq[i]);
3258                 ret = torture_create_kthread(rcu_torture_barrier_cbs,
3259                                              (void *)(long)i,
3260                                              barrier_cbs_tasks[i]);
3261                 if (ret)
3262                         return ret;
3263         }
3264         return torture_create_kthread(rcu_torture_barrier, NULL, barrier_task);
3265 }
3266
3267 /* Clean up after RCU barrier testing. */
3268 static void rcu_torture_barrier_cleanup(void)
3269 {
3270         int i;
3271
3272         torture_stop_kthread(rcu_torture_barrier, barrier_task);
3273         if (barrier_cbs_tasks != NULL) {
3274                 for (i = 0; i < n_barrier_cbs; i++)
3275                         torture_stop_kthread(rcu_torture_barrier_cbs,
3276                                              barrier_cbs_tasks[i]);
3277                 kfree(barrier_cbs_tasks);
3278                 barrier_cbs_tasks = NULL;
3279         }
3280         if (barrier_cbs_wq != NULL) {
3281                 kfree(barrier_cbs_wq);
3282                 barrier_cbs_wq = NULL;
3283         }
3284 }
3285
3286 static bool rcu_torture_can_boost(void)
3287 {
3288         static int boost_warn_once;
3289         int prio;
3290
3291         if (!(test_boost == 1 && cur_ops->can_boost) && test_boost != 2)
3292                 return false;
3293         if (!cur_ops->start_gp_poll || !cur_ops->poll_gp_state)
3294                 return false;
3295
3296         prio = rcu_get_gp_kthreads_prio();
3297         if (!prio)
3298                 return false;
3299
3300         if (prio < 2) {
3301                 if (boost_warn_once == 1)
3302                         return false;
3303
3304                 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);
3305                 boost_warn_once = 1;
3306                 return false;
3307         }
3308
3309         return true;
3310 }
3311
3312 static bool read_exit_child_stop;
3313 static bool read_exit_child_stopped;
3314 static wait_queue_head_t read_exit_wq;
3315
3316 // Child kthread which just does an rcutorture reader and exits.
3317 static int rcu_torture_read_exit_child(void *trsp_in)
3318 {
3319         struct torture_random_state *trsp = trsp_in;
3320
3321         set_user_nice(current, MAX_NICE);
3322         // Minimize time between reading and exiting.
3323         while (!kthread_should_stop())
3324                 schedule_timeout_uninterruptible(HZ / 20);
3325         (void)rcu_torture_one_read(trsp, -1);
3326         return 0;
3327 }
3328
3329 // Parent kthread which creates and destroys read-exit child kthreads.
3330 static int rcu_torture_read_exit(void *unused)
3331 {
3332         bool errexit = false;
3333         int i;
3334         struct task_struct *tsp;
3335         DEFINE_TORTURE_RANDOM(trs);
3336
3337         // Allocate and initialize.
3338         set_user_nice(current, MAX_NICE);
3339         VERBOSE_TOROUT_STRING("rcu_torture_read_exit: Start of test");
3340
3341         // Each pass through this loop does one read-exit episode.
3342         do {
3343                 VERBOSE_TOROUT_STRING("rcu_torture_read_exit: Start of episode");
3344                 for (i = 0; i < read_exit_burst; i++) {
3345                         if (READ_ONCE(read_exit_child_stop))
3346                                 break;
3347                         stutter_wait("rcu_torture_read_exit");
3348                         // Spawn child.
3349                         tsp = kthread_run(rcu_torture_read_exit_child,
3350                                           &trs, "%s", "rcu_torture_read_exit_child");
3351                         if (IS_ERR(tsp)) {
3352                                 TOROUT_ERRSTRING("out of memory");
3353                                 errexit = true;
3354                                 break;
3355                         }
3356                         cond_resched();
3357                         kthread_stop(tsp);
3358                         n_read_exits++;
3359                 }
3360                 VERBOSE_TOROUT_STRING("rcu_torture_read_exit: End of episode");
3361                 rcu_barrier(); // Wait for task_struct free, avoid OOM.
3362                 i = 0;
3363                 for (; !errexit && !READ_ONCE(read_exit_child_stop) && i < read_exit_delay; i++)
3364                         schedule_timeout_uninterruptible(HZ);
3365         } while (!errexit && !READ_ONCE(read_exit_child_stop));
3366
3367         // Clean up and exit.
3368         smp_store_release(&read_exit_child_stopped, true); // After reaping.
3369         smp_mb(); // Store before wakeup.
3370         wake_up(&read_exit_wq);
3371         while (!torture_must_stop())
3372                 schedule_timeout_uninterruptible(HZ / 20);
3373         torture_kthread_stopping("rcu_torture_read_exit");
3374         return 0;
3375 }
3376
3377 static int rcu_torture_read_exit_init(void)
3378 {
3379         if (read_exit_burst <= 0)
3380                 return 0;
3381         init_waitqueue_head(&read_exit_wq);
3382         read_exit_child_stop = false;
3383         read_exit_child_stopped = false;
3384         return torture_create_kthread(rcu_torture_read_exit, NULL,
3385                                       read_exit_task);
3386 }
3387
3388 static void rcu_torture_read_exit_cleanup(void)
3389 {
3390         if (!read_exit_task)
3391                 return;
3392         WRITE_ONCE(read_exit_child_stop, true);
3393         smp_mb(); // Above write before wait.
3394         wait_event(read_exit_wq, smp_load_acquire(&read_exit_child_stopped));
3395         torture_stop_kthread(rcutorture_read_exit, read_exit_task);
3396 }
3397
3398 static void rcutorture_test_nmis(int n)
3399 {
3400 #if IS_BUILTIN(CONFIG_RCU_TORTURE_TEST)
3401         int cpu;
3402         int dumpcpu;
3403         int i;
3404
3405         for (i = 0; i < n; i++) {
3406                 preempt_disable();
3407                 cpu = smp_processor_id();
3408                 dumpcpu = cpu + 1;
3409                 if (dumpcpu >= nr_cpu_ids)
3410                         dumpcpu = 0;
3411                 pr_alert("%s: CPU %d invoking dump_cpu_task(%d)\n", __func__, cpu, dumpcpu);
3412                 dump_cpu_task(dumpcpu);
3413                 preempt_enable();
3414                 schedule_timeout_uninterruptible(15 * HZ);
3415         }
3416 #else // #if IS_BUILTIN(CONFIG_RCU_TORTURE_TEST)
3417         WARN_ONCE(n, "Non-zero rcutorture.test_nmis=%d permitted only when rcutorture is built in.\n", test_nmis);
3418 #endif // #else // #if IS_BUILTIN(CONFIG_RCU_TORTURE_TEST)
3419 }
3420
3421 static enum cpuhp_state rcutor_hp;
3422
3423 static void
3424 rcu_torture_cleanup(void)
3425 {
3426         int firsttime;
3427         int flags = 0;
3428         unsigned long gp_seq = 0;
3429         int i;
3430
3431         if (torture_cleanup_begin()) {
3432                 if (cur_ops->cb_barrier != NULL) {
3433                         pr_info("%s: Invoking %pS().\n", __func__, cur_ops->cb_barrier);
3434                         cur_ops->cb_barrier();
3435                 }
3436                 if (cur_ops->gp_slow_unregister)
3437                         cur_ops->gp_slow_unregister(NULL);
3438                 return;
3439         }
3440         if (!cur_ops) {
3441                 torture_cleanup_end();
3442                 return;
3443         }
3444
3445         rcutorture_test_nmis(test_nmis);
3446
3447         if (cur_ops->gp_kthread_dbg)
3448                 cur_ops->gp_kthread_dbg();
3449         rcu_torture_read_exit_cleanup();
3450         rcu_torture_barrier_cleanup();
3451         rcu_torture_fwd_prog_cleanup();
3452         torture_stop_kthread(rcu_torture_stall, stall_task);
3453         torture_stop_kthread(rcu_torture_writer, writer_task);
3454
3455         if (nocb_tasks) {
3456                 for (i = 0; i < nrealnocbers; i++)
3457                         torture_stop_kthread(rcu_nocb_toggle, nocb_tasks[i]);
3458                 kfree(nocb_tasks);
3459                 nocb_tasks = NULL;
3460         }
3461
3462         if (reader_tasks) {
3463                 for (i = 0; i < nrealreaders; i++)
3464                         torture_stop_kthread(rcu_torture_reader,
3465                                              reader_tasks[i]);
3466                 kfree(reader_tasks);
3467                 reader_tasks = NULL;
3468         }
3469         kfree(rcu_torture_reader_mbchk);
3470         rcu_torture_reader_mbchk = NULL;
3471
3472         if (fakewriter_tasks) {
3473                 for (i = 0; i < nfakewriters; i++)
3474                         torture_stop_kthread(rcu_torture_fakewriter,
3475                                              fakewriter_tasks[i]);
3476                 kfree(fakewriter_tasks);
3477                 fakewriter_tasks = NULL;
3478         }
3479
3480         if (cur_ops->get_gp_data)
3481                 cur_ops->get_gp_data(&flags, &gp_seq);
3482         pr_alert("%s:  End-test grace-period state: g%ld f%#x total-gps=%ld\n",
3483                  cur_ops->name, (long)gp_seq, flags,
3484                  rcutorture_seq_diff(gp_seq, start_gp_seq));
3485         torture_stop_kthread(rcu_torture_stats, stats_task);
3486         torture_stop_kthread(rcu_torture_fqs, fqs_task);
3487         if (rcu_torture_can_boost() && rcutor_hp >= 0)
3488                 cpuhp_remove_state(rcutor_hp);
3489
3490         /*
3491          * Wait for all RCU callbacks to fire, then do torture-type-specific
3492          * cleanup operations.
3493          */
3494         if (cur_ops->cb_barrier != NULL) {
3495                 pr_info("%s: Invoking %pS().\n", __func__, cur_ops->cb_barrier);
3496                 cur_ops->cb_barrier();
3497         }
3498         if (cur_ops->cleanup != NULL)
3499                 cur_ops->cleanup();
3500
3501         rcu_torture_mem_dump_obj();
3502
3503         rcu_torture_stats_print();  /* -After- the stats thread is stopped! */
3504
3505         if (err_segs_recorded) {
3506                 pr_alert("Failure/close-call rcutorture reader segments:\n");
3507                 if (rt_read_nsegs == 0)
3508                         pr_alert("\t: No segments recorded!!!\n");
3509                 firsttime = 1;
3510                 for (i = 0; i < rt_read_nsegs; i++) {
3511                         pr_alert("\t%d: %#x ", i, err_segs[i].rt_readstate);
3512                         if (err_segs[i].rt_delay_jiffies != 0) {
3513                                 pr_cont("%s%ldjiffies", firsttime ? "" : "+",
3514                                         err_segs[i].rt_delay_jiffies);
3515                                 firsttime = 0;
3516                         }
3517                         if (err_segs[i].rt_delay_ms != 0) {
3518                                 pr_cont("%s%ldms", firsttime ? "" : "+",
3519                                         err_segs[i].rt_delay_ms);
3520                                 firsttime = 0;
3521                         }
3522                         if (err_segs[i].rt_delay_us != 0) {
3523                                 pr_cont("%s%ldus", firsttime ? "" : "+",
3524                                         err_segs[i].rt_delay_us);
3525                                 firsttime = 0;
3526                         }
3527                         pr_cont("%s\n",
3528                                 err_segs[i].rt_preempted ? "preempted" : "");
3529
3530                 }
3531         }
3532         if (atomic_read(&n_rcu_torture_error) || n_rcu_torture_barrier_error)
3533                 rcu_torture_print_module_parms(cur_ops, "End of test: FAILURE");
3534         else if (torture_onoff_failures())
3535                 rcu_torture_print_module_parms(cur_ops,
3536                                                "End of test: RCU_HOTPLUG");
3537         else
3538                 rcu_torture_print_module_parms(cur_ops, "End of test: SUCCESS");
3539         torture_cleanup_end();
3540         if (cur_ops->gp_slow_unregister)
3541                 cur_ops->gp_slow_unregister(NULL);
3542 }
3543
3544 static void rcu_torture_leak_cb(struct rcu_head *rhp)
3545 {
3546 }
3547
3548 static void rcu_torture_err_cb(struct rcu_head *rhp)
3549 {
3550         /*
3551          * This -might- happen due to race conditions, but is unlikely.
3552          * The scenario that leads to this happening is that the
3553          * first of the pair of duplicate callbacks is queued,
3554          * someone else starts a grace period that includes that
3555          * callback, then the second of the pair must wait for the
3556          * next grace period.  Unlikely, but can happen.  If it
3557          * does happen, the debug-objects subsystem won't have splatted.
3558          */
3559         pr_alert("%s: duplicated callback was invoked.\n", KBUILD_MODNAME);
3560 }
3561
3562 /*
3563  * Verify that double-free causes debug-objects to complain, but only
3564  * if CONFIG_DEBUG_OBJECTS_RCU_HEAD=y.  Otherwise, say that the test
3565  * cannot be carried out.
3566  */
3567 static void rcu_test_debug_objects(void)
3568 {
3569         struct rcu_head rh1;
3570         struct rcu_head rh2;
3571         int idx;
3572
3573         if (!IS_ENABLED(CONFIG_DEBUG_OBJECTS_RCU_HEAD)) {
3574                 pr_alert("%s: !CONFIG_DEBUG_OBJECTS_RCU_HEAD, not testing duplicate call_%s()\n",
3575                                         KBUILD_MODNAME, cur_ops->name);
3576                 return;
3577         }
3578
3579         if (WARN_ON_ONCE(cur_ops->debug_objects &&
3580                         (!cur_ops->call || !cur_ops->cb_barrier)))
3581                 return;
3582
3583         struct rcu_head *rhp = kmalloc(sizeof(*rhp), GFP_KERNEL);
3584
3585         init_rcu_head_on_stack(&rh1);
3586         init_rcu_head_on_stack(&rh2);
3587         pr_alert("%s: WARN: Duplicate call_%s() test starting.\n", KBUILD_MODNAME, cur_ops->name);
3588
3589         /* Try to queue the rh2 pair of callbacks for the same grace period. */
3590         idx = cur_ops->readlock(); /* Make it impossible to finish a grace period. */
3591         cur_ops->call(&rh1, rcu_torture_leak_cb); /* Start grace period. */
3592         cur_ops->call(&rh2, rcu_torture_leak_cb);
3593         cur_ops->call(&rh2, rcu_torture_err_cb); /* Duplicate callback. */
3594         if (rhp) {
3595                 cur_ops->call(rhp, rcu_torture_leak_cb);
3596                 cur_ops->call(rhp, rcu_torture_err_cb); /* Another duplicate callback. */
3597         }
3598         cur_ops->readunlock(idx);
3599
3600         /* Wait for them all to get done so we can safely return. */
3601         cur_ops->cb_barrier();
3602         pr_alert("%s: WARN: Duplicate call_%s() test complete.\n", KBUILD_MODNAME, cur_ops->name);
3603         destroy_rcu_head_on_stack(&rh1);
3604         destroy_rcu_head_on_stack(&rh2);
3605         kfree(rhp);
3606 }
3607
3608 static void rcutorture_sync(void)
3609 {
3610         static unsigned long n;
3611
3612         if (cur_ops->sync && !(++n & 0xfff))
3613                 cur_ops->sync();
3614 }
3615
3616 static DEFINE_MUTEX(mut0);
3617 static DEFINE_MUTEX(mut1);
3618 static DEFINE_MUTEX(mut2);
3619 static DEFINE_MUTEX(mut3);
3620 static DEFINE_MUTEX(mut4);
3621 static DEFINE_MUTEX(mut5);
3622 static DEFINE_MUTEX(mut6);
3623 static DEFINE_MUTEX(mut7);
3624 static DEFINE_MUTEX(mut8);
3625 static DEFINE_MUTEX(mut9);
3626
3627 static DECLARE_RWSEM(rwsem0);
3628 static DECLARE_RWSEM(rwsem1);
3629 static DECLARE_RWSEM(rwsem2);
3630 static DECLARE_RWSEM(rwsem3);
3631 static DECLARE_RWSEM(rwsem4);
3632 static DECLARE_RWSEM(rwsem5);
3633 static DECLARE_RWSEM(rwsem6);
3634 static DECLARE_RWSEM(rwsem7);
3635 static DECLARE_RWSEM(rwsem8);
3636 static DECLARE_RWSEM(rwsem9);
3637
3638 DEFINE_STATIC_SRCU(srcu0);
3639 DEFINE_STATIC_SRCU(srcu1);
3640 DEFINE_STATIC_SRCU(srcu2);
3641 DEFINE_STATIC_SRCU(srcu3);
3642 DEFINE_STATIC_SRCU(srcu4);
3643 DEFINE_STATIC_SRCU(srcu5);
3644 DEFINE_STATIC_SRCU(srcu6);
3645 DEFINE_STATIC_SRCU(srcu7);
3646 DEFINE_STATIC_SRCU(srcu8);
3647 DEFINE_STATIC_SRCU(srcu9);
3648
3649 static int srcu_lockdep_next(const char *f, const char *fl, const char *fs, const char *fu, int i,
3650                              int cyclelen, int deadlock)
3651 {
3652         int j = i + 1;
3653
3654         if (j >= cyclelen)
3655                 j = deadlock ? 0 : -1;
3656         if (j >= 0)
3657                 pr_info("%s: %s(%d), %s(%d), %s(%d)\n", f, fl, i, fs, j, fu, i);
3658         else
3659                 pr_info("%s: %s(%d), %s(%d)\n", f, fl, i, fu, i);
3660         return j;
3661 }
3662
3663 // Test lockdep on SRCU-based deadlock scenarios.
3664 static void rcu_torture_init_srcu_lockdep(void)
3665 {
3666         int cyclelen;
3667         int deadlock;
3668         bool err = false;
3669         int i;
3670         int j;
3671         int idx;
3672         struct mutex *muts[] = { &mut0, &mut1, &mut2, &mut3, &mut4,
3673                                  &mut5, &mut6, &mut7, &mut8, &mut9 };
3674         struct rw_semaphore *rwsems[] = { &rwsem0, &rwsem1, &rwsem2, &rwsem3, &rwsem4,
3675                                           &rwsem5, &rwsem6, &rwsem7, &rwsem8, &rwsem9 };
3676         struct srcu_struct *srcus[] = { &srcu0, &srcu1, &srcu2, &srcu3, &srcu4,
3677                                         &srcu5, &srcu6, &srcu7, &srcu8, &srcu9 };
3678         int testtype;
3679
3680         if (!test_srcu_lockdep)
3681                 return;
3682
3683         deadlock = test_srcu_lockdep / 1000;
3684         testtype = (test_srcu_lockdep / 10) % 100;
3685         cyclelen = test_srcu_lockdep % 10;
3686         WARN_ON_ONCE(ARRAY_SIZE(muts) != ARRAY_SIZE(srcus));
3687         if (WARN_ONCE(deadlock != !!deadlock,
3688                       "%s: test_srcu_lockdep=%d and deadlock digit %d must be zero or one.\n",
3689                       __func__, test_srcu_lockdep, deadlock))
3690                 err = true;
3691         if (WARN_ONCE(cyclelen <= 0,
3692                       "%s: test_srcu_lockdep=%d and cycle-length digit %d must be greater than zero.\n",
3693                       __func__, test_srcu_lockdep, cyclelen))
3694                 err = true;
3695         if (err)
3696                 goto err_out;
3697
3698         if (testtype == 0) {
3699                 pr_info("%s: test_srcu_lockdep = %05d: SRCU %d-way %sdeadlock.\n",
3700                         __func__, test_srcu_lockdep, cyclelen, deadlock ? "" : "non-");
3701                 if (deadlock && cyclelen == 1)
3702                         pr_info("%s: Expect hang.\n", __func__);
3703                 for (i = 0; i < cyclelen; i++) {
3704                         j = srcu_lockdep_next(__func__, "srcu_read_lock", "synchronize_srcu",
3705                                               "srcu_read_unlock", i, cyclelen, deadlock);
3706                         idx = srcu_read_lock(srcus[i]);
3707                         if (j >= 0)
3708                                 synchronize_srcu(srcus[j]);
3709                         srcu_read_unlock(srcus[i], idx);
3710                 }
3711                 return;
3712         }
3713
3714         if (testtype == 1) {
3715                 pr_info("%s: test_srcu_lockdep = %05d: SRCU/mutex %d-way %sdeadlock.\n",
3716                         __func__, test_srcu_lockdep, cyclelen, deadlock ? "" : "non-");
3717                 for (i = 0; i < cyclelen; i++) {
3718                         pr_info("%s: srcu_read_lock(%d), mutex_lock(%d), mutex_unlock(%d), srcu_read_unlock(%d)\n",
3719                                 __func__, i, i, i, i);
3720                         idx = srcu_read_lock(srcus[i]);
3721                         mutex_lock(muts[i]);
3722                         mutex_unlock(muts[i]);
3723                         srcu_read_unlock(srcus[i], idx);
3724
3725                         j = srcu_lockdep_next(__func__, "mutex_lock", "synchronize_srcu",
3726                                               "mutex_unlock", i, cyclelen, deadlock);
3727                         mutex_lock(muts[i]);
3728                         if (j >= 0)
3729                                 synchronize_srcu(srcus[j]);
3730                         mutex_unlock(muts[i]);
3731                 }
3732                 return;
3733         }
3734
3735         if (testtype == 2) {
3736                 pr_info("%s: test_srcu_lockdep = %05d: SRCU/rwsem %d-way %sdeadlock.\n",
3737                         __func__, test_srcu_lockdep, cyclelen, deadlock ? "" : "non-");
3738                 for (i = 0; i < cyclelen; i++) {
3739                         pr_info("%s: srcu_read_lock(%d), down_read(%d), up_read(%d), srcu_read_unlock(%d)\n",
3740                                 __func__, i, i, i, i);
3741                         idx = srcu_read_lock(srcus[i]);
3742                         down_read(rwsems[i]);
3743                         up_read(rwsems[i]);
3744                         srcu_read_unlock(srcus[i], idx);
3745
3746                         j = srcu_lockdep_next(__func__, "down_write", "synchronize_srcu",
3747                                               "up_write", i, cyclelen, deadlock);
3748                         down_write(rwsems[i]);
3749                         if (j >= 0)
3750                                 synchronize_srcu(srcus[j]);
3751                         up_write(rwsems[i]);
3752                 }
3753                 return;
3754         }
3755
3756 #ifdef CONFIG_TASKS_TRACE_RCU
3757         if (testtype == 3) {
3758                 pr_info("%s: test_srcu_lockdep = %05d: SRCU and Tasks Trace RCU %d-way %sdeadlock.\n",
3759                         __func__, test_srcu_lockdep, cyclelen, deadlock ? "" : "non-");
3760                 if (deadlock && cyclelen == 1)
3761                         pr_info("%s: Expect hang.\n", __func__);
3762                 for (i = 0; i < cyclelen; i++) {
3763                         char *fl = i == 0 ? "rcu_read_lock_trace" : "srcu_read_lock";
3764                         char *fs = i == cyclelen - 1 ? "synchronize_rcu_tasks_trace"
3765                                                      : "synchronize_srcu";
3766                         char *fu = i == 0 ? "rcu_read_unlock_trace" : "srcu_read_unlock";
3767
3768                         j = srcu_lockdep_next(__func__, fl, fs, fu, i, cyclelen, deadlock);
3769                         if (i == 0)
3770                                 rcu_read_lock_trace();
3771                         else
3772                                 idx = srcu_read_lock(srcus[i]);
3773                         if (j >= 0) {
3774                                 if (i == cyclelen - 1)
3775                                         synchronize_rcu_tasks_trace();
3776                                 else
3777                                         synchronize_srcu(srcus[j]);
3778                         }
3779                         if (i == 0)
3780                                 rcu_read_unlock_trace();
3781                         else
3782                                 srcu_read_unlock(srcus[i], idx);
3783                 }
3784                 return;
3785         }
3786 #endif // #ifdef CONFIG_TASKS_TRACE_RCU
3787
3788 err_out:
3789         pr_info("%s: test_srcu_lockdep = %05d does nothing.\n", __func__, test_srcu_lockdep);
3790         pr_info("%s: test_srcu_lockdep = DNNL.\n", __func__);
3791         pr_info("%s: D: Deadlock if nonzero.\n", __func__);
3792         pr_info("%s: NN: Test number, 0=SRCU, 1=SRCU/mutex, 2=SRCU/rwsem, 3=SRCU/Tasks Trace RCU.\n", __func__);
3793         pr_info("%s: L: Cycle length.\n", __func__);
3794         if (!IS_ENABLED(CONFIG_TASKS_TRACE_RCU))
3795                 pr_info("%s: NN=3 disallowed because kernel is built with CONFIG_TASKS_TRACE_RCU=n\n", __func__);
3796 }
3797
3798 static int __init
3799 rcu_torture_init(void)
3800 {
3801         long i;
3802         int cpu;
3803         int firsterr = 0;
3804         int flags = 0;
3805         unsigned long gp_seq = 0;
3806         static struct rcu_torture_ops *torture_ops[] = {
3807                 &rcu_ops, &rcu_busted_ops, &srcu_ops, &srcud_ops, &busted_srcud_ops,
3808                 TASKS_OPS TASKS_RUDE_OPS TASKS_TRACING_OPS
3809                 &trivial_ops,
3810         };
3811
3812         if (!torture_init_begin(torture_type, verbose))
3813                 return -EBUSY;
3814
3815         /* Process args and tell the world that the torturer is on the job. */
3816         for (i = 0; i < ARRAY_SIZE(torture_ops); i++) {
3817                 cur_ops = torture_ops[i];
3818                 if (strcmp(torture_type, cur_ops->name) == 0)
3819                         break;
3820         }
3821         if (i == ARRAY_SIZE(torture_ops)) {
3822                 pr_alert("rcu-torture: invalid torture type: \"%s\"\n",
3823                          torture_type);
3824                 pr_alert("rcu-torture types:");
3825                 for (i = 0; i < ARRAY_SIZE(torture_ops); i++)
3826                         pr_cont(" %s", torture_ops[i]->name);
3827                 pr_cont("\n");
3828                 firsterr = -EINVAL;
3829                 cur_ops = NULL;
3830                 goto unwind;
3831         }
3832         if (cur_ops->fqs == NULL && fqs_duration != 0) {
3833                 pr_alert("rcu-torture: ->fqs NULL and non-zero fqs_duration, fqs disabled.\n");
3834                 fqs_duration = 0;
3835         }
3836         if (nocbs_nthreads != 0 && (cur_ops != &rcu_ops ||
3837                                     !IS_ENABLED(CONFIG_RCU_NOCB_CPU))) {
3838                 pr_alert("rcu-torture types: %s and CONFIG_RCU_NOCB_CPU=%d, nocb toggle disabled.\n",
3839                          cur_ops->name, IS_ENABLED(CONFIG_RCU_NOCB_CPU));
3840                 nocbs_nthreads = 0;
3841         }
3842         if (cur_ops->init)
3843                 cur_ops->init();
3844
3845         rcu_torture_init_srcu_lockdep();
3846
3847         if (nreaders >= 0) {
3848                 nrealreaders = nreaders;
3849         } else {
3850                 nrealreaders = num_online_cpus() - 2 - nreaders;
3851                 if (nrealreaders <= 0)
3852                         nrealreaders = 1;
3853         }
3854         rcu_torture_print_module_parms(cur_ops, "Start of test");
3855         if (cur_ops->get_gp_data)
3856                 cur_ops->get_gp_data(&flags, &gp_seq);
3857         start_gp_seq = gp_seq;
3858         pr_alert("%s:  Start-test grace-period state: g%ld f%#x\n",
3859                  cur_ops->name, (long)gp_seq, flags);
3860
3861         /* Set up the freelist. */
3862
3863         INIT_LIST_HEAD(&rcu_torture_freelist);
3864         for (i = 0; i < ARRAY_SIZE(rcu_tortures); i++) {
3865                 rcu_tortures[i].rtort_mbtest = 0;
3866                 list_add_tail(&rcu_tortures[i].rtort_free,
3867                               &rcu_torture_freelist);
3868         }
3869
3870         /* Initialize the statistics so that each run gets its own numbers. */
3871
3872         rcu_torture_current = NULL;
3873         rcu_torture_current_version = 0;
3874         atomic_set(&n_rcu_torture_alloc, 0);
3875         atomic_set(&n_rcu_torture_alloc_fail, 0);
3876         atomic_set(&n_rcu_torture_free, 0);
3877         atomic_set(&n_rcu_torture_mberror, 0);
3878         atomic_set(&n_rcu_torture_mbchk_fail, 0);
3879         atomic_set(&n_rcu_torture_mbchk_tries, 0);
3880         atomic_set(&n_rcu_torture_error, 0);
3881         n_rcu_torture_barrier_error = 0;
3882         n_rcu_torture_boost_ktrerror = 0;
3883         n_rcu_torture_boost_failure = 0;
3884         n_rcu_torture_boosts = 0;
3885         for (i = 0; i < RCU_TORTURE_PIPE_LEN + 1; i++)
3886                 atomic_set(&rcu_torture_wcount[i], 0);
3887         for_each_possible_cpu(cpu) {
3888                 for (i = 0; i < RCU_TORTURE_PIPE_LEN + 1; i++) {
3889                         per_cpu(rcu_torture_count, cpu)[i] = 0;
3890                         per_cpu(rcu_torture_batch, cpu)[i] = 0;
3891                 }
3892         }
3893         err_segs_recorded = 0;
3894         rt_read_nsegs = 0;
3895
3896         /* Start up the kthreads. */
3897
3898         rcu_torture_write_types();
3899         firsterr = torture_create_kthread(rcu_torture_writer, NULL,
3900                                           writer_task);
3901         if (torture_init_error(firsterr))
3902                 goto unwind;
3903         if (nfakewriters > 0) {
3904                 fakewriter_tasks = kcalloc(nfakewriters,
3905                                            sizeof(fakewriter_tasks[0]),
3906                                            GFP_KERNEL);
3907                 if (fakewriter_tasks == NULL) {
3908                         TOROUT_ERRSTRING("out of memory");
3909                         firsterr = -ENOMEM;
3910                         goto unwind;
3911                 }
3912         }
3913         for (i = 0; i < nfakewriters; i++) {
3914                 firsterr = torture_create_kthread(rcu_torture_fakewriter,
3915                                                   NULL, fakewriter_tasks[i]);
3916                 if (torture_init_error(firsterr))
3917                         goto unwind;
3918         }
3919         reader_tasks = kcalloc(nrealreaders, sizeof(reader_tasks[0]),
3920                                GFP_KERNEL);
3921         rcu_torture_reader_mbchk = kcalloc(nrealreaders, sizeof(*rcu_torture_reader_mbchk),
3922                                            GFP_KERNEL);
3923         if (!reader_tasks || !rcu_torture_reader_mbchk) {
3924                 TOROUT_ERRSTRING("out of memory");
3925                 firsterr = -ENOMEM;
3926                 goto unwind;
3927         }
3928         for (i = 0; i < nrealreaders; i++) {
3929                 rcu_torture_reader_mbchk[i].rtc_chkrdr = -1;
3930                 firsterr = torture_create_kthread(rcu_torture_reader, (void *)i,
3931                                                   reader_tasks[i]);
3932                 if (torture_init_error(firsterr))
3933                         goto unwind;
3934         }
3935         nrealnocbers = nocbs_nthreads;
3936         if (WARN_ON(nrealnocbers < 0))
3937                 nrealnocbers = 1;
3938         if (WARN_ON(nocbs_toggle < 0))
3939                 nocbs_toggle = HZ;
3940         if (nrealnocbers > 0) {
3941                 nocb_tasks = kcalloc(nrealnocbers, sizeof(nocb_tasks[0]), GFP_KERNEL);
3942                 if (nocb_tasks == NULL) {
3943                         TOROUT_ERRSTRING("out of memory");
3944                         firsterr = -ENOMEM;
3945                         goto unwind;
3946                 }
3947         } else {
3948                 nocb_tasks = NULL;
3949         }
3950         for (i = 0; i < nrealnocbers; i++) {
3951                 firsterr = torture_create_kthread(rcu_nocb_toggle, NULL, nocb_tasks[i]);
3952                 if (torture_init_error(firsterr))
3953                         goto unwind;
3954         }
3955         if (stat_interval > 0) {
3956                 firsterr = torture_create_kthread(rcu_torture_stats, NULL,
3957                                                   stats_task);
3958                 if (torture_init_error(firsterr))
3959                         goto unwind;
3960         }
3961         if (test_no_idle_hz && shuffle_interval > 0) {
3962                 firsterr = torture_shuffle_init(shuffle_interval * HZ);
3963                 if (torture_init_error(firsterr))
3964                         goto unwind;
3965         }
3966         if (stutter < 0)
3967                 stutter = 0;
3968         if (stutter) {
3969                 int t;
3970
3971                 t = cur_ops->stall_dur ? cur_ops->stall_dur() : stutter * HZ;
3972                 firsterr = torture_stutter_init(stutter * HZ, t);
3973                 if (torture_init_error(firsterr))
3974                         goto unwind;
3975         }
3976         if (fqs_duration < 0)
3977                 fqs_duration = 0;
3978         if (fqs_holdoff < 0)
3979                 fqs_holdoff = 0;
3980         if (fqs_duration && fqs_holdoff) {
3981                 /* Create the fqs thread */
3982                 firsterr = torture_create_kthread(rcu_torture_fqs, NULL,
3983                                                   fqs_task);
3984                 if (torture_init_error(firsterr))
3985                         goto unwind;
3986         }
3987         if (test_boost_interval < 1)
3988                 test_boost_interval = 1;
3989         if (test_boost_duration < 2)
3990                 test_boost_duration = 2;
3991         if (rcu_torture_can_boost()) {
3992
3993                 boost_starttime = jiffies + test_boost_interval * HZ;
3994
3995                 firsterr = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "RCU_TORTURE",
3996                                              rcutorture_booster_init,
3997                                              rcutorture_booster_cleanup);
3998                 rcutor_hp = firsterr;
3999                 if (torture_init_error(firsterr))
4000                         goto unwind;
4001         }
4002         shutdown_jiffies = jiffies + shutdown_secs * HZ;
4003         firsterr = torture_shutdown_init(shutdown_secs, rcu_torture_cleanup);
4004         if (torture_init_error(firsterr))
4005                 goto unwind;
4006         firsterr = torture_onoff_init(onoff_holdoff * HZ, onoff_interval,
4007                                       rcutorture_sync);
4008         if (torture_init_error(firsterr))
4009                 goto unwind;
4010         firsterr = rcu_torture_stall_init();
4011         if (torture_init_error(firsterr))
4012                 goto unwind;
4013         firsterr = rcu_torture_fwd_prog_init();
4014         if (torture_init_error(firsterr))
4015                 goto unwind;
4016         firsterr = rcu_torture_barrier_init();
4017         if (torture_init_error(firsterr))
4018                 goto unwind;
4019         firsterr = rcu_torture_read_exit_init();
4020         if (torture_init_error(firsterr))
4021                 goto unwind;
4022         if (object_debug)
4023                 rcu_test_debug_objects();
4024         torture_init_end();
4025         if (cur_ops->gp_slow_register && !WARN_ON_ONCE(!cur_ops->gp_slow_unregister))
4026                 cur_ops->gp_slow_register(&rcu_fwd_cb_nodelay);
4027         return 0;
4028
4029 unwind:
4030         torture_init_end();
4031         rcu_torture_cleanup();
4032         if (shutdown_secs) {
4033                 WARN_ON(!IS_MODULE(CONFIG_RCU_TORTURE_TEST));
4034                 kernel_power_off();
4035         }
4036         return firsterr;
4037 }
4038
4039 module_init(rcu_torture_init);
4040 module_exit(rcu_torture_cleanup);