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