Merge branch 'sched-core-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
authorLinus Torvalds <torvalds@linux-foundation.org>
Mon, 31 Mar 2014 18:21:19 +0000 (11:21 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Mon, 31 Mar 2014 18:21:19 +0000 (11:21 -0700)
Pull scheduler changes from Ingo Molnar:
 "Bigger changes:

   - sched/idle restructuring: they are WIP preparation for deeper
     integration between the scheduler and idle state selection, by
     Nicolas Pitre.

   - add NUMA scheduling pseudo-interleaving, by Rik van Riel.

   - optimize cgroup context switches, by Peter Zijlstra.

   - RT scheduling enhancements, by Thomas Gleixner.

  The rest is smaller changes, non-urgnt fixes and cleanups"

* 'sched-core-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (68 commits)
  sched: Clean up the task_hot() function
  sched: Remove double calculation in fix_small_imbalance()
  sched: Fix broken setscheduler()
  sparc64, sched: Remove unused sparc64_multi_core
  sched: Remove unused mc_capable() and smt_capable()
  sched/numa: Move task_numa_free() to __put_task_struct()
  sched/fair: Fix endless loop in idle_balance()
  sched/core: Fix endless loop in pick_next_task()
  sched/fair: Push down check for high priority class task into idle_balance()
  sched/rt: Fix picking RT and DL tasks from empty queue
  trace: Replace hardcoding of 19 with MAX_NICE
  sched: Guarantee task priority in pick_next_task()
  sched/idle: Remove stale old file
  sched: Put rq's sched_avg under CONFIG_FAIR_GROUP_SCHED
  cpuidle/arm64: Remove redundant cpuidle_idle_call()
  cpuidle/powernv: Remove redundant cpuidle_idle_call()
  sched, nohz: Exclude isolated cores from load balancing
  sched: Fix select_task_rq_fair() description comments
  workqueue: Replace hardcoding of -20 and 19 with MIN_NICE and MAX_NICE
  sys: Replace hardcoding of -20 and 19 with MIN_NICE and MAX_NICE
  ...

1  2 
Documentation/sysctl/kernel.txt
arch/x86/include/asm/topology.h
kernel/Makefile
kernel/rcu/rcutorture.c

Simple merge
Simple merge
diff --cc kernel/Makefile
Simple merge
index f59d48597ddeff6cbfea56df85d672a46980fdcd,0000000000000000000000000000000000000000..bd30bc61bc05c273159541f260d047fa4747bc82
mode 100644,000000..100644
--- /dev/null
@@@ -1,1582 -1,0 +1,1582 @@@
-       set_user_nice(current, 19);
 +/*
 + * Read-Copy Update module-based torture test facility
 + *
 + * This program is free software; you can redistribute it and/or modify
 + * it under the terms of the GNU General Public License as published by
 + * the Free Software Foundation; either version 2 of the License, or
 + * (at your option) any later version.
 + *
 + * This program is distributed in the hope that it will be useful,
 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 + * GNU General Public License for more details.
 + *
 + * You should have received a copy of the GNU General Public License
 + * along with this program; if not, you can access it online at
 + * http://www.gnu.org/licenses/gpl-2.0.html.
 + *
 + * Copyright (C) IBM Corporation, 2005, 2006
 + *
 + * Authors: Paul E. McKenney <paulmck@us.ibm.com>
 + *      Josh Triplett <josh@freedesktop.org>
 + *
 + * See also:  Documentation/RCU/torture.txt
 + */
 +#include <linux/types.h>
 +#include <linux/kernel.h>
 +#include <linux/init.h>
 +#include <linux/module.h>
 +#include <linux/kthread.h>
 +#include <linux/err.h>
 +#include <linux/spinlock.h>
 +#include <linux/smp.h>
 +#include <linux/rcupdate.h>
 +#include <linux/interrupt.h>
 +#include <linux/sched.h>
 +#include <linux/atomic.h>
 +#include <linux/bitops.h>
 +#include <linux/completion.h>
 +#include <linux/moduleparam.h>
 +#include <linux/percpu.h>
 +#include <linux/notifier.h>
 +#include <linux/reboot.h>
 +#include <linux/freezer.h>
 +#include <linux/cpu.h>
 +#include <linux/delay.h>
 +#include <linux/stat.h>
 +#include <linux/srcu.h>
 +#include <linux/slab.h>
 +#include <linux/trace_clock.h>
 +#include <asm/byteorder.h>
 +#include <linux/torture.h>
 +
 +MODULE_LICENSE("GPL");
 +MODULE_AUTHOR("Paul E. McKenney <paulmck@us.ibm.com> and Josh Triplett <josh@freedesktop.org>");
 +
 +
 +torture_param(int, fqs_duration, 0,
 +            "Duration of fqs bursts (us), 0 to disable");
 +torture_param(int, fqs_holdoff, 0, "Holdoff time within fqs bursts (us)");
 +torture_param(int, fqs_stutter, 3, "Wait time between fqs bursts (s)");
 +torture_param(bool, gp_exp, false, "Use expedited GP wait primitives");
 +torture_param(bool, gp_normal, false,
 +           "Use normal (non-expedited) GP wait primitives");
 +torture_param(int, irqreader, 1, "Allow RCU readers from irq handlers");
 +torture_param(int, n_barrier_cbs, 0,
 +           "# of callbacks/kthreads for barrier testing");
 +torture_param(int, nfakewriters, 4, "Number of RCU fake writer threads");
 +torture_param(int, nreaders, -1, "Number of RCU reader threads");
 +torture_param(int, object_debug, 0,
 +           "Enable debug-object double call_rcu() testing");
 +torture_param(int, onoff_holdoff, 0, "Time after boot before CPU hotplugs (s)");
 +torture_param(int, onoff_interval, 0,
 +           "Time between CPU hotplugs (s), 0=disable");
 +torture_param(int, shuffle_interval, 3, "Number of seconds between shuffles");
 +torture_param(int, shutdown_secs, 0, "Shutdown time (s), <= zero to disable.");
 +torture_param(int, stall_cpu, 0, "Stall duration (s), zero to disable.");
 +torture_param(int, stall_cpu_holdoff, 10,
 +           "Time to wait before starting stall (s).");
 +torture_param(int, stat_interval, 60,
 +           "Number of seconds between stats printk()s");
 +torture_param(int, stutter, 5, "Number of seconds to run/halt test");
 +torture_param(int, test_boost, 1, "Test RCU prio boost: 0=no, 1=maybe, 2=yes.");
 +torture_param(int, test_boost_duration, 4,
 +           "Duration of each boost test, seconds.");
 +torture_param(int, test_boost_interval, 7,
 +           "Interval between boost tests, seconds.");
 +torture_param(bool, test_no_idle_hz, true,
 +           "Test support for tickless idle CPUs");
 +torture_param(bool, verbose, true,
 +           "Enable verbose debugging printk()s");
 +
 +static char *torture_type = "rcu";
 +module_param(torture_type, charp, 0444);
 +MODULE_PARM_DESC(torture_type, "Type of RCU to torture (rcu, rcu_bh, ...)");
 +
 +static int nrealreaders;
 +static struct task_struct *writer_task;
 +static struct task_struct **fakewriter_tasks;
 +static struct task_struct **reader_tasks;
 +static struct task_struct *stats_task;
 +static struct task_struct *fqs_task;
 +static struct task_struct *boost_tasks[NR_CPUS];
 +static struct task_struct *stall_task;
 +static struct task_struct **barrier_cbs_tasks;
 +static struct task_struct *barrier_task;
 +
 +#define RCU_TORTURE_PIPE_LEN 10
 +
 +struct rcu_torture {
 +      struct rcu_head rtort_rcu;
 +      int rtort_pipe_count;
 +      struct list_head rtort_free;
 +      int rtort_mbtest;
 +};
 +
 +static LIST_HEAD(rcu_torture_freelist);
 +static struct rcu_torture __rcu *rcu_torture_current;
 +static unsigned long rcu_torture_current_version;
 +static struct rcu_torture rcu_tortures[10 * RCU_TORTURE_PIPE_LEN];
 +static DEFINE_SPINLOCK(rcu_torture_lock);
 +static DEFINE_PER_CPU(long [RCU_TORTURE_PIPE_LEN + 1],
 +                    rcu_torture_count) = { 0 };
 +static DEFINE_PER_CPU(long [RCU_TORTURE_PIPE_LEN + 1],
 +                    rcu_torture_batch) = { 0 };
 +static atomic_t rcu_torture_wcount[RCU_TORTURE_PIPE_LEN + 1];
 +static atomic_t n_rcu_torture_alloc;
 +static atomic_t n_rcu_torture_alloc_fail;
 +static atomic_t n_rcu_torture_free;
 +static atomic_t n_rcu_torture_mberror;
 +static atomic_t n_rcu_torture_error;
 +static long n_rcu_torture_barrier_error;
 +static long n_rcu_torture_boost_ktrerror;
 +static long n_rcu_torture_boost_rterror;
 +static long n_rcu_torture_boost_failure;
 +static long n_rcu_torture_boosts;
 +static long n_rcu_torture_timers;
 +static long n_barrier_attempts;
 +static long n_barrier_successes;
 +static struct list_head rcu_torture_removed;
 +
 +#if defined(MODULE) || defined(CONFIG_RCU_TORTURE_TEST_RUNNABLE)
 +#define RCUTORTURE_RUNNABLE_INIT 1
 +#else
 +#define RCUTORTURE_RUNNABLE_INIT 0
 +#endif
 +int rcutorture_runnable = RCUTORTURE_RUNNABLE_INIT;
 +module_param(rcutorture_runnable, int, 0444);
 +MODULE_PARM_DESC(rcutorture_runnable, "Start rcutorture at boot");
 +
 +#if defined(CONFIG_RCU_BOOST) && !defined(CONFIG_HOTPLUG_CPU)
 +#define rcu_can_boost() 1
 +#else /* #if defined(CONFIG_RCU_BOOST) && !defined(CONFIG_HOTPLUG_CPU) */
 +#define rcu_can_boost() 0
 +#endif /* #else #if defined(CONFIG_RCU_BOOST) && !defined(CONFIG_HOTPLUG_CPU) */
 +
 +#ifdef CONFIG_RCU_TRACE
 +static u64 notrace rcu_trace_clock_local(void)
 +{
 +      u64 ts = trace_clock_local();
 +      unsigned long __maybe_unused ts_rem = do_div(ts, NSEC_PER_USEC);
 +      return ts;
 +}
 +#else /* #ifdef CONFIG_RCU_TRACE */
 +static u64 notrace rcu_trace_clock_local(void)
 +{
 +      return 0ULL;
 +}
 +#endif /* #else #ifdef CONFIG_RCU_TRACE */
 +
 +static unsigned long boost_starttime; /* jiffies of next boost test start. */
 +DEFINE_MUTEX(boost_mutex);            /* protect setting boost_starttime */
 +                                      /*  and boost task create/destroy. */
 +static atomic_t barrier_cbs_count;    /* Barrier callbacks registered. */
 +static bool barrier_phase;            /* Test phase. */
 +static atomic_t barrier_cbs_invoked;  /* Barrier callbacks invoked. */
 +static wait_queue_head_t *barrier_cbs_wq; /* Coordinate barrier testing. */
 +static DECLARE_WAIT_QUEUE_HEAD(barrier_wq);
 +
 +/*
 + * Allocate an element from the rcu_tortures pool.
 + */
 +static struct rcu_torture *
 +rcu_torture_alloc(void)
 +{
 +      struct list_head *p;
 +
 +      spin_lock_bh(&rcu_torture_lock);
 +      if (list_empty(&rcu_torture_freelist)) {
 +              atomic_inc(&n_rcu_torture_alloc_fail);
 +              spin_unlock_bh(&rcu_torture_lock);
 +              return NULL;
 +      }
 +      atomic_inc(&n_rcu_torture_alloc);
 +      p = rcu_torture_freelist.next;
 +      list_del_init(p);
 +      spin_unlock_bh(&rcu_torture_lock);
 +      return container_of(p, struct rcu_torture, rtort_free);
 +}
 +
 +/*
 + * Free an element to the rcu_tortures pool.
 + */
 +static void
 +rcu_torture_free(struct rcu_torture *p)
 +{
 +      atomic_inc(&n_rcu_torture_free);
 +      spin_lock_bh(&rcu_torture_lock);
 +      list_add_tail(&p->rtort_free, &rcu_torture_freelist);
 +      spin_unlock_bh(&rcu_torture_lock);
 +}
 +
 +/*
 + * Operations vector for selecting different types of tests.
 + */
 +
 +struct rcu_torture_ops {
 +      void (*init)(void);
 +      int (*readlock)(void);
 +      void (*read_delay)(struct torture_random_state *rrsp);
 +      void (*readunlock)(int idx);
 +      int (*completed)(void);
 +      void (*deferred_free)(struct rcu_torture *p);
 +      void (*sync)(void);
 +      void (*exp_sync)(void);
 +      void (*call)(struct rcu_head *head, void (*func)(struct rcu_head *rcu));
 +      void (*cb_barrier)(void);
 +      void (*fqs)(void);
 +      void (*stats)(char *page);
 +      int irq_capable;
 +      int can_boost;
 +      const char *name;
 +};
 +
 +static struct rcu_torture_ops *cur_ops;
 +
 +/*
 + * Definitions for rcu torture testing.
 + */
 +
 +static int rcu_torture_read_lock(void) __acquires(RCU)
 +{
 +      rcu_read_lock();
 +      return 0;
 +}
 +
 +static void rcu_read_delay(struct torture_random_state *rrsp)
 +{
 +      const unsigned long shortdelay_us = 200;
 +      const unsigned long longdelay_ms = 50;
 +
 +      /* We want a short delay sometimes to make a reader delay the grace
 +       * period, and we want a long delay occasionally to trigger
 +       * force_quiescent_state. */
 +
 +      if (!(torture_random(rrsp) % (nrealreaders * 2000 * longdelay_ms)))
 +              mdelay(longdelay_ms);
 +      if (!(torture_random(rrsp) % (nrealreaders * 2 * shortdelay_us)))
 +              udelay(shortdelay_us);
 +#ifdef CONFIG_PREEMPT
 +      if (!preempt_count() &&
 +          !(torture_random(rrsp) % (nrealreaders * 20000)))
 +              preempt_schedule();  /* No QS if preempt_disable() in effect */
 +#endif
 +}
 +
 +static void rcu_torture_read_unlock(int idx) __releases(RCU)
 +{
 +      rcu_read_unlock();
 +}
 +
 +static int rcu_torture_completed(void)
 +{
 +      return rcu_batches_completed();
 +}
 +
 +static void
 +rcu_torture_cb(struct rcu_head *p)
 +{
 +      int i;
 +      struct rcu_torture *rp = container_of(p, struct rcu_torture, rtort_rcu);
 +
 +      if (torture_must_stop_irq()) {
 +              /* Test is ending, just drop callbacks on the floor. */
 +              /* The next initialization will pick up the pieces. */
 +              return;
 +      }
 +      i = rp->rtort_pipe_count;
 +      if (i > RCU_TORTURE_PIPE_LEN)
 +              i = RCU_TORTURE_PIPE_LEN;
 +      atomic_inc(&rcu_torture_wcount[i]);
 +      if (++rp->rtort_pipe_count >= RCU_TORTURE_PIPE_LEN) {
 +              rp->rtort_mbtest = 0;
 +              rcu_torture_free(rp);
 +      } else {
 +              cur_ops->deferred_free(rp);
 +      }
 +}
 +
 +static int rcu_no_completed(void)
 +{
 +      return 0;
 +}
 +
 +static void rcu_torture_deferred_free(struct rcu_torture *p)
 +{
 +      call_rcu(&p->rtort_rcu, rcu_torture_cb);
 +}
 +
 +static void rcu_sync_torture_init(void)
 +{
 +      INIT_LIST_HEAD(&rcu_torture_removed);
 +}
 +
 +static struct rcu_torture_ops rcu_ops = {
 +      .init           = rcu_sync_torture_init,
 +      .readlock       = rcu_torture_read_lock,
 +      .read_delay     = rcu_read_delay,
 +      .readunlock     = rcu_torture_read_unlock,
 +      .completed      = rcu_torture_completed,
 +      .deferred_free  = rcu_torture_deferred_free,
 +      .sync           = synchronize_rcu,
 +      .exp_sync       = synchronize_rcu_expedited,
 +      .call           = call_rcu,
 +      .cb_barrier     = rcu_barrier,
 +      .fqs            = rcu_force_quiescent_state,
 +      .stats          = NULL,
 +      .irq_capable    = 1,
 +      .can_boost      = rcu_can_boost(),
 +      .name           = "rcu"
 +};
 +
 +/*
 + * Definitions for rcu_bh torture testing.
 + */
 +
 +static int rcu_bh_torture_read_lock(void) __acquires(RCU_BH)
 +{
 +      rcu_read_lock_bh();
 +      return 0;
 +}
 +
 +static void rcu_bh_torture_read_unlock(int idx) __releases(RCU_BH)
 +{
 +      rcu_read_unlock_bh();
 +}
 +
 +static int rcu_bh_torture_completed(void)
 +{
 +      return rcu_batches_completed_bh();
 +}
 +
 +static void rcu_bh_torture_deferred_free(struct rcu_torture *p)
 +{
 +      call_rcu_bh(&p->rtort_rcu, rcu_torture_cb);
 +}
 +
 +static struct rcu_torture_ops rcu_bh_ops = {
 +      .init           = rcu_sync_torture_init,
 +      .readlock       = rcu_bh_torture_read_lock,
 +      .read_delay     = rcu_read_delay,  /* just reuse rcu's version. */
 +      .readunlock     = rcu_bh_torture_read_unlock,
 +      .completed      = rcu_bh_torture_completed,
 +      .deferred_free  = rcu_bh_torture_deferred_free,
 +      .sync           = synchronize_rcu_bh,
 +      .exp_sync       = synchronize_rcu_bh_expedited,
 +      .call           = call_rcu_bh,
 +      .cb_barrier     = rcu_barrier_bh,
 +      .fqs            = rcu_bh_force_quiescent_state,
 +      .stats          = NULL,
 +      .irq_capable    = 1,
 +      .name           = "rcu_bh"
 +};
 +
 +/*
 + * Don't even think about trying any of these in real life!!!
 + * The names includes "busted", and they really means it!
 + * The only purpose of these functions is to provide a buggy RCU
 + * implementation to make sure that rcutorture correctly emits
 + * buggy-RCU error messages.
 + */
 +static void rcu_busted_torture_deferred_free(struct rcu_torture *p)
 +{
 +      /* This is a deliberate bug for testing purposes only! */
 +      rcu_torture_cb(&p->rtort_rcu);
 +}
 +
 +static void synchronize_rcu_busted(void)
 +{
 +      /* This is a deliberate bug for testing purposes only! */
 +}
 +
 +static void
 +call_rcu_busted(struct rcu_head *head, void (*func)(struct rcu_head *rcu))
 +{
 +      /* This is a deliberate bug for testing purposes only! */
 +      func(head);
 +}
 +
 +static struct rcu_torture_ops rcu_busted_ops = {
 +      .init           = rcu_sync_torture_init,
 +      .readlock       = rcu_torture_read_lock,
 +      .read_delay     = rcu_read_delay,  /* just reuse rcu's version. */
 +      .readunlock     = rcu_torture_read_unlock,
 +      .completed      = rcu_no_completed,
 +      .deferred_free  = rcu_busted_torture_deferred_free,
 +      .sync           = synchronize_rcu_busted,
 +      .exp_sync       = synchronize_rcu_busted,
 +      .call           = call_rcu_busted,
 +      .cb_barrier     = NULL,
 +      .fqs            = NULL,
 +      .stats          = NULL,
 +      .irq_capable    = 1,
 +      .name           = "rcu_busted"
 +};
 +
 +/*
 + * Definitions for srcu torture testing.
 + */
 +
 +DEFINE_STATIC_SRCU(srcu_ctl);
 +
 +static int srcu_torture_read_lock(void) __acquires(&srcu_ctl)
 +{
 +      return srcu_read_lock(&srcu_ctl);
 +}
 +
 +static void srcu_read_delay(struct torture_random_state *rrsp)
 +{
 +      long delay;
 +      const long uspertick = 1000000 / HZ;
 +      const long longdelay = 10;
 +
 +      /* We want there to be long-running readers, but not all the time. */
 +
 +      delay = torture_random(rrsp) %
 +              (nrealreaders * 2 * longdelay * uspertick);
 +      if (!delay)
 +              schedule_timeout_interruptible(longdelay);
 +      else
 +              rcu_read_delay(rrsp);
 +}
 +
 +static void srcu_torture_read_unlock(int idx) __releases(&srcu_ctl)
 +{
 +      srcu_read_unlock(&srcu_ctl, idx);
 +}
 +
 +static int srcu_torture_completed(void)
 +{
 +      return srcu_batches_completed(&srcu_ctl);
 +}
 +
 +static void srcu_torture_deferred_free(struct rcu_torture *rp)
 +{
 +      call_srcu(&srcu_ctl, &rp->rtort_rcu, rcu_torture_cb);
 +}
 +
 +static void srcu_torture_synchronize(void)
 +{
 +      synchronize_srcu(&srcu_ctl);
 +}
 +
 +static void srcu_torture_call(struct rcu_head *head,
 +                            void (*func)(struct rcu_head *head))
 +{
 +      call_srcu(&srcu_ctl, head, func);
 +}
 +
 +static void srcu_torture_barrier(void)
 +{
 +      srcu_barrier(&srcu_ctl);
 +}
 +
 +static void srcu_torture_stats(char *page)
 +{
 +      int cpu;
 +      int idx = srcu_ctl.completed & 0x1;
 +
 +      page += sprintf(page, "%s%s per-CPU(idx=%d):",
 +                     torture_type, TORTURE_FLAG, idx);
 +      for_each_possible_cpu(cpu) {
 +              page += sprintf(page, " %d(%lu,%lu)", cpu,
 +                             per_cpu_ptr(srcu_ctl.per_cpu_ref, cpu)->c[!idx],
 +                             per_cpu_ptr(srcu_ctl.per_cpu_ref, cpu)->c[idx]);
 +      }
 +      sprintf(page, "\n");
 +}
 +
 +static void srcu_torture_synchronize_expedited(void)
 +{
 +      synchronize_srcu_expedited(&srcu_ctl);
 +}
 +
 +static struct rcu_torture_ops srcu_ops = {
 +      .init           = rcu_sync_torture_init,
 +      .readlock       = srcu_torture_read_lock,
 +      .read_delay     = srcu_read_delay,
 +      .readunlock     = srcu_torture_read_unlock,
 +      .completed      = srcu_torture_completed,
 +      .deferred_free  = srcu_torture_deferred_free,
 +      .sync           = srcu_torture_synchronize,
 +      .exp_sync       = srcu_torture_synchronize_expedited,
 +      .call           = srcu_torture_call,
 +      .cb_barrier     = srcu_torture_barrier,
 +      .stats          = srcu_torture_stats,
 +      .name           = "srcu"
 +};
 +
 +/*
 + * Definitions for sched torture testing.
 + */
 +
 +static int sched_torture_read_lock(void)
 +{
 +      preempt_disable();
 +      return 0;
 +}
 +
 +static void sched_torture_read_unlock(int idx)
 +{
 +      preempt_enable();
 +}
 +
 +static void rcu_sched_torture_deferred_free(struct rcu_torture *p)
 +{
 +      call_rcu_sched(&p->rtort_rcu, rcu_torture_cb);
 +}
 +
 +static struct rcu_torture_ops sched_ops = {
 +      .init           = rcu_sync_torture_init,
 +      .readlock       = sched_torture_read_lock,
 +      .read_delay     = rcu_read_delay,  /* just reuse rcu's version. */
 +      .readunlock     = sched_torture_read_unlock,
 +      .completed      = rcu_no_completed,
 +      .deferred_free  = rcu_sched_torture_deferred_free,
 +      .sync           = synchronize_sched,
 +      .exp_sync       = synchronize_sched_expedited,
 +      .call           = call_rcu_sched,
 +      .cb_barrier     = rcu_barrier_sched,
 +      .fqs            = rcu_sched_force_quiescent_state,
 +      .stats          = NULL,
 +      .irq_capable    = 1,
 +      .name           = "sched"
 +};
 +
 +/*
 + * RCU torture priority-boost testing.  Runs one real-time thread per
 + * CPU for moderate bursts, repeatedly registering RCU callbacks and
 + * spinning waiting for them to be invoked.  If a given callback takes
 + * too long to be invoked, we assume that priority inversion has occurred.
 + */
 +
 +struct rcu_boost_inflight {
 +      struct rcu_head rcu;
 +      int inflight;
 +};
 +
 +static void rcu_torture_boost_cb(struct rcu_head *head)
 +{
 +      struct rcu_boost_inflight *rbip =
 +              container_of(head, struct rcu_boost_inflight, rcu);
 +
 +      smp_mb(); /* Ensure RCU-core accesses precede clearing ->inflight */
 +      rbip->inflight = 0;
 +}
 +
 +static int rcu_torture_boost(void *arg)
 +{
 +      unsigned long call_rcu_time;
 +      unsigned long endtime;
 +      unsigned long oldstarttime;
 +      struct rcu_boost_inflight rbi = { .inflight = 0 };
 +      struct sched_param sp;
 +
 +      VERBOSE_TOROUT_STRING("rcu_torture_boost started");
 +
 +      /* Set real-time priority. */
 +      sp.sched_priority = 1;
 +      if (sched_setscheduler(current, SCHED_FIFO, &sp) < 0) {
 +              VERBOSE_TOROUT_STRING("rcu_torture_boost RT prio failed!");
 +              n_rcu_torture_boost_rterror++;
 +      }
 +
 +      init_rcu_head_on_stack(&rbi.rcu);
 +      /* Each pass through the following loop does one boost-test cycle. */
 +      do {
 +              /* Wait for the next test interval. */
 +              oldstarttime = boost_starttime;
 +              while (ULONG_CMP_LT(jiffies, oldstarttime)) {
 +                      schedule_timeout_interruptible(oldstarttime - jiffies);
 +                      stutter_wait("rcu_torture_boost");
 +                      if (torture_must_stop())
 +                              goto checkwait;
 +              }
 +
 +              /* Do one boost-test interval. */
 +              endtime = oldstarttime + test_boost_duration * HZ;
 +              call_rcu_time = jiffies;
 +              while (ULONG_CMP_LT(jiffies, endtime)) {
 +                      /* If we don't have a callback in flight, post one. */
 +                      if (!rbi.inflight) {
 +                              smp_mb(); /* RCU core before ->inflight = 1. */
 +                              rbi.inflight = 1;
 +                              call_rcu(&rbi.rcu, rcu_torture_boost_cb);
 +                              if (jiffies - call_rcu_time >
 +                                       test_boost_duration * HZ - HZ / 2) {
 +                                      VERBOSE_TOROUT_STRING("rcu_torture_boost boosting failed");
 +                                      n_rcu_torture_boost_failure++;
 +                              }
 +                              call_rcu_time = jiffies;
 +                      }
 +                      cond_resched();
 +                      stutter_wait("rcu_torture_boost");
 +                      if (torture_must_stop())
 +                              goto checkwait;
 +              }
 +
 +              /*
 +               * Set the start time of the next test interval.
 +               * Yes, this is vulnerable to long delays, but such
 +               * delays simply cause a false negative for the next
 +               * interval.  Besides, we are running at RT priority,
 +               * so delays should be relatively rare.
 +               */
 +              while (oldstarttime == boost_starttime &&
 +                     !kthread_should_stop()) {
 +                      if (mutex_trylock(&boost_mutex)) {
 +                              boost_starttime = jiffies +
 +                                                test_boost_interval * HZ;
 +                              n_rcu_torture_boosts++;
 +                              mutex_unlock(&boost_mutex);
 +                              break;
 +                      }
 +                      schedule_timeout_uninterruptible(1);
 +              }
 +
 +              /* Go do the stutter. */
 +checkwait:    stutter_wait("rcu_torture_boost");
 +      } while (!torture_must_stop());
 +
 +      /* Clean up and exit. */
 +      while (!kthread_should_stop() || rbi.inflight) {
 +              torture_shutdown_absorb("rcu_torture_boost");
 +              schedule_timeout_uninterruptible(1);
 +      }
 +      smp_mb(); /* order accesses to ->inflight before stack-frame death. */
 +      destroy_rcu_head_on_stack(&rbi.rcu);
 +      torture_kthread_stopping("rcu_torture_boost");
 +      return 0;
 +}
 +
 +/*
 + * RCU torture force-quiescent-state kthread.  Repeatedly induces
 + * bursts of calls to force_quiescent_state(), increasing the probability
 + * of occurrence of some important types of race conditions.
 + */
 +static int
 +rcu_torture_fqs(void *arg)
 +{
 +      unsigned long fqs_resume_time;
 +      int fqs_burst_remaining;
 +
 +      VERBOSE_TOROUT_STRING("rcu_torture_fqs task started");
 +      do {
 +              fqs_resume_time = jiffies + fqs_stutter * HZ;
 +              while (ULONG_CMP_LT(jiffies, fqs_resume_time) &&
 +                     !kthread_should_stop()) {
 +                      schedule_timeout_interruptible(1);
 +              }
 +              fqs_burst_remaining = fqs_duration;
 +              while (fqs_burst_remaining > 0 &&
 +                     !kthread_should_stop()) {
 +                      cur_ops->fqs();
 +                      udelay(fqs_holdoff);
 +                      fqs_burst_remaining -= fqs_holdoff;
 +              }
 +              stutter_wait("rcu_torture_fqs");
 +      } while (!torture_must_stop());
 +      torture_kthread_stopping("rcu_torture_fqs");
 +      return 0;
 +}
 +
 +/*
 + * RCU torture writer kthread.  Repeatedly substitutes a new structure
 + * for that pointed to by rcu_torture_current, freeing the old structure
 + * after a series of grace periods (the "pipeline").
 + */
 +static int
 +rcu_torture_writer(void *arg)
 +{
 +      bool exp;
 +      int i;
 +      struct rcu_torture *rp;
 +      struct rcu_torture *rp1;
 +      struct rcu_torture *old_rp;
 +      static DEFINE_TORTURE_RANDOM(rand);
 +
 +      VERBOSE_TOROUT_STRING("rcu_torture_writer task started");
-       set_user_nice(current, 19);
++      set_user_nice(current, MAX_NICE);
 +
 +      do {
 +              schedule_timeout_uninterruptible(1);
 +              rp = rcu_torture_alloc();
 +              if (rp == NULL)
 +                      continue;
 +              rp->rtort_pipe_count = 0;
 +              udelay(torture_random(&rand) & 0x3ff);
 +              old_rp = rcu_dereference_check(rcu_torture_current,
 +                                             current == writer_task);
 +              rp->rtort_mbtest = 1;
 +              rcu_assign_pointer(rcu_torture_current, rp);
 +              smp_wmb(); /* Mods to old_rp must follow rcu_assign_pointer() */
 +              if (old_rp) {
 +                      i = old_rp->rtort_pipe_count;
 +                      if (i > RCU_TORTURE_PIPE_LEN)
 +                              i = RCU_TORTURE_PIPE_LEN;
 +                      atomic_inc(&rcu_torture_wcount[i]);
 +                      old_rp->rtort_pipe_count++;
 +                      if (gp_normal == gp_exp)
 +                              exp = !!(torture_random(&rand) & 0x80);
 +                      else
 +                              exp = gp_exp;
 +                      if (!exp) {
 +                              cur_ops->deferred_free(old_rp);
 +                      } else {
 +                              cur_ops->exp_sync();
 +                              list_add(&old_rp->rtort_free,
 +                                       &rcu_torture_removed);
 +                              list_for_each_entry_safe(rp, rp1,
 +                                                       &rcu_torture_removed,
 +                                                       rtort_free) {
 +                                      i = rp->rtort_pipe_count;
 +                                      if (i > RCU_TORTURE_PIPE_LEN)
 +                                              i = RCU_TORTURE_PIPE_LEN;
 +                                      atomic_inc(&rcu_torture_wcount[i]);
 +                                      if (++rp->rtort_pipe_count >=
 +                                          RCU_TORTURE_PIPE_LEN) {
 +                                              rp->rtort_mbtest = 0;
 +                                              list_del(&rp->rtort_free);
 +                                              rcu_torture_free(rp);
 +                                      }
 +                               }
 +                      }
 +              }
 +              rcutorture_record_progress(++rcu_torture_current_version);
 +              stutter_wait("rcu_torture_writer");
 +      } while (!torture_must_stop());
 +      torture_kthread_stopping("rcu_torture_writer");
 +      return 0;
 +}
 +
 +/*
 + * RCU torture fake writer kthread.  Repeatedly calls sync, with a random
 + * delay between calls.
 + */
 +static int
 +rcu_torture_fakewriter(void *arg)
 +{
 +      DEFINE_TORTURE_RANDOM(rand);
 +
 +      VERBOSE_TOROUT_STRING("rcu_torture_fakewriter task started");
-       set_user_nice(current, 19);
++      set_user_nice(current, MAX_NICE);
 +
 +      do {
 +              schedule_timeout_uninterruptible(1 + torture_random(&rand)%10);
 +              udelay(torture_random(&rand) & 0x3ff);
 +              if (cur_ops->cb_barrier != NULL &&
 +                  torture_random(&rand) % (nfakewriters * 8) == 0) {
 +                      cur_ops->cb_barrier();
 +              } else if (gp_normal == gp_exp) {
 +                      if (torture_random(&rand) & 0x80)
 +                              cur_ops->sync();
 +                      else
 +                              cur_ops->exp_sync();
 +              } else if (gp_normal) {
 +                      cur_ops->sync();
 +              } else {
 +                      cur_ops->exp_sync();
 +              }
 +              stutter_wait("rcu_torture_fakewriter");
 +      } while (!torture_must_stop());
 +
 +      torture_kthread_stopping("rcu_torture_fakewriter");
 +      return 0;
 +}
 +
 +void rcutorture_trace_dump(void)
 +{
 +      static atomic_t beenhere = ATOMIC_INIT(0);
 +
 +      if (atomic_read(&beenhere))
 +              return;
 +      if (atomic_xchg(&beenhere, 1) != 0)
 +              return;
 +      ftrace_dump(DUMP_ALL);
 +}
 +
 +/*
 + * RCU torture reader from timer handler.  Dereferences rcu_torture_current,
 + * incrementing the corresponding element of the pipeline array.  The
 + * counter in the element should never be greater than 1, otherwise, the
 + * RCU implementation is broken.
 + */
 +static void rcu_torture_timer(unsigned long unused)
 +{
 +      int idx;
 +      int completed;
 +      int completed_end;
 +      static DEFINE_TORTURE_RANDOM(rand);
 +      static DEFINE_SPINLOCK(rand_lock);
 +      struct rcu_torture *p;
 +      int pipe_count;
 +      unsigned long long ts;
 +
 +      idx = cur_ops->readlock();
 +      completed = cur_ops->completed();
 +      ts = rcu_trace_clock_local();
 +      p = rcu_dereference_check(rcu_torture_current,
 +                                rcu_read_lock_bh_held() ||
 +                                rcu_read_lock_sched_held() ||
 +                                srcu_read_lock_held(&srcu_ctl));
 +      if (p == NULL) {
 +              /* Leave because rcu_torture_writer is not yet underway */
 +              cur_ops->readunlock(idx);
 +              return;
 +      }
 +      if (p->rtort_mbtest == 0)
 +              atomic_inc(&n_rcu_torture_mberror);
 +      spin_lock(&rand_lock);
 +      cur_ops->read_delay(&rand);
 +      n_rcu_torture_timers++;
 +      spin_unlock(&rand_lock);
 +      preempt_disable();
 +      pipe_count = p->rtort_pipe_count;
 +      if (pipe_count > RCU_TORTURE_PIPE_LEN) {
 +              /* Should not happen, but... */
 +              pipe_count = RCU_TORTURE_PIPE_LEN;
 +      }
 +      completed_end = cur_ops->completed();
 +      if (pipe_count > 1) {
 +              do_trace_rcu_torture_read(cur_ops->name, &p->rtort_rcu, ts,
 +                                        completed, completed_end);
 +              rcutorture_trace_dump();
 +      }
 +      __this_cpu_inc(rcu_torture_count[pipe_count]);
 +      completed = completed_end - completed;
 +      if (completed > RCU_TORTURE_PIPE_LEN) {
 +              /* Should not happen, but... */
 +              completed = RCU_TORTURE_PIPE_LEN;
 +      }
 +      __this_cpu_inc(rcu_torture_batch[completed]);
 +      preempt_enable();
 +      cur_ops->readunlock(idx);
 +}
 +
 +/*
 + * RCU torture reader kthread.  Repeatedly dereferences rcu_torture_current,
 + * incrementing the corresponding element of the pipeline array.  The
 + * counter in the element should never be greater than 1, otherwise, the
 + * RCU implementation is broken.
 + */
 +static int
 +rcu_torture_reader(void *arg)
 +{
 +      int completed;
 +      int completed_end;
 +      int idx;
 +      DEFINE_TORTURE_RANDOM(rand);
 +      struct rcu_torture *p;
 +      int pipe_count;
 +      struct timer_list t;
 +      unsigned long long ts;
 +
 +      VERBOSE_TOROUT_STRING("rcu_torture_reader task started");
-       set_user_nice(current, 19);
++      set_user_nice(current, MAX_NICE);
 +      if (irqreader && cur_ops->irq_capable)
 +              setup_timer_on_stack(&t, rcu_torture_timer, 0);
 +
 +      do {
 +              if (irqreader && cur_ops->irq_capable) {
 +                      if (!timer_pending(&t))
 +                              mod_timer(&t, jiffies + 1);
 +              }
 +              idx = cur_ops->readlock();
 +              completed = cur_ops->completed();
 +              ts = rcu_trace_clock_local();
 +              p = rcu_dereference_check(rcu_torture_current,
 +                                        rcu_read_lock_bh_held() ||
 +                                        rcu_read_lock_sched_held() ||
 +                                        srcu_read_lock_held(&srcu_ctl));
 +              if (p == NULL) {
 +                      /* Wait for rcu_torture_writer to get underway */
 +                      cur_ops->readunlock(idx);
 +                      schedule_timeout_interruptible(HZ);
 +                      continue;
 +              }
 +              if (p->rtort_mbtest == 0)
 +                      atomic_inc(&n_rcu_torture_mberror);
 +              cur_ops->read_delay(&rand);
 +              preempt_disable();
 +              pipe_count = p->rtort_pipe_count;
 +              if (pipe_count > RCU_TORTURE_PIPE_LEN) {
 +                      /* Should not happen, but... */
 +                      pipe_count = RCU_TORTURE_PIPE_LEN;
 +              }
 +              completed_end = cur_ops->completed();
 +              if (pipe_count > 1) {
 +                      do_trace_rcu_torture_read(cur_ops->name, &p->rtort_rcu,
 +                                                ts, completed, completed_end);
 +                      rcutorture_trace_dump();
 +              }
 +              __this_cpu_inc(rcu_torture_count[pipe_count]);
 +              completed = completed_end - completed;
 +              if (completed > RCU_TORTURE_PIPE_LEN) {
 +                      /* Should not happen, but... */
 +                      completed = RCU_TORTURE_PIPE_LEN;
 +              }
 +              __this_cpu_inc(rcu_torture_batch[completed]);
 +              preempt_enable();
 +              cur_ops->readunlock(idx);
 +              schedule();
 +              stutter_wait("rcu_torture_reader");
 +      } while (!torture_must_stop());
 +      if (irqreader && cur_ops->irq_capable)
 +              del_timer_sync(&t);
 +      torture_kthread_stopping("rcu_torture_reader");
 +      return 0;
 +}
 +
 +/*
 + * Create an RCU-torture statistics message in the specified buffer.
 + */
 +static void
 +rcu_torture_printk(char *page)
 +{
 +      int cpu;
 +      int i;
 +      long pipesummary[RCU_TORTURE_PIPE_LEN + 1] = { 0 };
 +      long batchsummary[RCU_TORTURE_PIPE_LEN + 1] = { 0 };
 +
 +      for_each_possible_cpu(cpu) {
 +              for (i = 0; i < RCU_TORTURE_PIPE_LEN + 1; i++) {
 +                      pipesummary[i] += per_cpu(rcu_torture_count, cpu)[i];
 +                      batchsummary[i] += per_cpu(rcu_torture_batch, cpu)[i];
 +              }
 +      }
 +      for (i = RCU_TORTURE_PIPE_LEN - 1; i >= 0; i--) {
 +              if (pipesummary[i] != 0)
 +                      break;
 +      }
 +      page += sprintf(page, "%s%s ", torture_type, TORTURE_FLAG);
 +      page += sprintf(page,
 +                     "rtc: %p ver: %lu tfle: %d rta: %d rtaf: %d rtf: %d ",
 +                     rcu_torture_current,
 +                     rcu_torture_current_version,
 +                     list_empty(&rcu_torture_freelist),
 +                     atomic_read(&n_rcu_torture_alloc),
 +                     atomic_read(&n_rcu_torture_alloc_fail),
 +                     atomic_read(&n_rcu_torture_free));
 +      page += sprintf(page, "rtmbe: %d rtbke: %ld rtbre: %ld ",
 +                     atomic_read(&n_rcu_torture_mberror),
 +                     n_rcu_torture_boost_ktrerror,
 +                     n_rcu_torture_boost_rterror);
 +      page += sprintf(page, "rtbf: %ld rtb: %ld nt: %ld ",
 +                     n_rcu_torture_boost_failure,
 +                     n_rcu_torture_boosts,
 +                     n_rcu_torture_timers);
 +      page = torture_onoff_stats(page);
 +      page += sprintf(page, "barrier: %ld/%ld:%ld",
 +                     n_barrier_successes,
 +                     n_barrier_attempts,
 +                     n_rcu_torture_barrier_error);
 +      page += sprintf(page, "\n%s%s ", torture_type, TORTURE_FLAG);
 +      if (atomic_read(&n_rcu_torture_mberror) != 0 ||
 +          n_rcu_torture_barrier_error != 0 ||
 +          n_rcu_torture_boost_ktrerror != 0 ||
 +          n_rcu_torture_boost_rterror != 0 ||
 +          n_rcu_torture_boost_failure != 0 ||
 +          i > 1) {
 +              page += sprintf(page, "!!! ");
 +              atomic_inc(&n_rcu_torture_error);
 +              WARN_ON_ONCE(1);
 +      }
 +      page += sprintf(page, "Reader Pipe: ");
 +      for (i = 0; i < RCU_TORTURE_PIPE_LEN + 1; i++)
 +              page += sprintf(page, " %ld", pipesummary[i]);
 +      page += sprintf(page, "\n%s%s ", torture_type, TORTURE_FLAG);
 +      page += sprintf(page, "Reader Batch: ");
 +      for (i = 0; i < RCU_TORTURE_PIPE_LEN + 1; i++)
 +              page += sprintf(page, " %ld", batchsummary[i]);
 +      page += sprintf(page, "\n%s%s ", torture_type, TORTURE_FLAG);
 +      page += sprintf(page, "Free-Block Circulation: ");
 +      for (i = 0; i < RCU_TORTURE_PIPE_LEN + 1; i++) {
 +              page += sprintf(page, " %d",
 +                             atomic_read(&rcu_torture_wcount[i]));
 +      }
 +      page += sprintf(page, "\n");
 +      if (cur_ops->stats)
 +              cur_ops->stats(page);
 +}
 +
 +/*
 + * Print torture statistics.  Caller must ensure that there is only
 + * one call to this function at a given time!!!  This is normally
 + * accomplished by relying on the module system to only have one copy
 + * of the module loaded, and then by giving the rcu_torture_stats
 + * kthread full control (or the init/cleanup functions when rcu_torture_stats
 + * thread is not running).
 + */
 +static void
 +rcu_torture_stats_print(void)
 +{
 +      int size = nr_cpu_ids * 200 + 8192;
 +      char *buf;
 +
 +      buf = kmalloc(size, GFP_KERNEL);
 +      if (!buf) {
 +              pr_err("rcu-torture: Out of memory, need: %d", size);
 +              return;
 +      }
 +      rcu_torture_printk(buf);
 +      pr_alert("%s", buf);
 +      kfree(buf);
 +}
 +
 +/*
 + * Periodically prints torture statistics, if periodic statistics printing
 + * was specified via the stat_interval module parameter.
 + */
 +static int
 +rcu_torture_stats(void *arg)
 +{
 +      VERBOSE_TOROUT_STRING("rcu_torture_stats task started");
 +      do {
 +              schedule_timeout_interruptible(stat_interval * HZ);
 +              rcu_torture_stats_print();
 +              torture_shutdown_absorb("rcu_torture_stats");
 +      } while (!torture_must_stop());
 +      torture_kthread_stopping("rcu_torture_stats");
 +      return 0;
 +}
 +
 +static inline void
 +rcu_torture_print_module_parms(struct rcu_torture_ops *cur_ops, const char *tag)
 +{
 +      pr_alert("%s" TORTURE_FLAG
 +               "--- %s: nreaders=%d nfakewriters=%d "
 +               "stat_interval=%d verbose=%d test_no_idle_hz=%d "
 +               "shuffle_interval=%d stutter=%d irqreader=%d "
 +               "fqs_duration=%d fqs_holdoff=%d fqs_stutter=%d "
 +               "test_boost=%d/%d test_boost_interval=%d "
 +               "test_boost_duration=%d shutdown_secs=%d "
 +               "stall_cpu=%d stall_cpu_holdoff=%d "
 +               "n_barrier_cbs=%d "
 +               "onoff_interval=%d onoff_holdoff=%d\n",
 +               torture_type, tag, nrealreaders, nfakewriters,
 +               stat_interval, verbose, test_no_idle_hz, shuffle_interval,
 +               stutter, irqreader, fqs_duration, fqs_holdoff, fqs_stutter,
 +               test_boost, cur_ops->can_boost,
 +               test_boost_interval, test_boost_duration, shutdown_secs,
 +               stall_cpu, stall_cpu_holdoff,
 +               n_barrier_cbs,
 +               onoff_interval, onoff_holdoff);
 +}
 +
 +static void rcutorture_booster_cleanup(int cpu)
 +{
 +      struct task_struct *t;
 +
 +      if (boost_tasks[cpu] == NULL)
 +              return;
 +      mutex_lock(&boost_mutex);
 +      t = boost_tasks[cpu];
 +      boost_tasks[cpu] = NULL;
 +      mutex_unlock(&boost_mutex);
 +
 +      /* This must be outside of the mutex, otherwise deadlock! */
 +      torture_stop_kthread(rcu_torture_boost, t);
 +}
 +
 +static int rcutorture_booster_init(int cpu)
 +{
 +      int retval;
 +
 +      if (boost_tasks[cpu] != NULL)
 +              return 0;  /* Already created, nothing more to do. */
 +
 +      /* Don't allow time recalculation while creating a new task. */
 +      mutex_lock(&boost_mutex);
 +      VERBOSE_TOROUT_STRING("Creating rcu_torture_boost task");
 +      boost_tasks[cpu] = kthread_create_on_node(rcu_torture_boost, NULL,
 +                                                cpu_to_node(cpu),
 +                                                "rcu_torture_boost");
 +      if (IS_ERR(boost_tasks[cpu])) {
 +              retval = PTR_ERR(boost_tasks[cpu]);
 +              VERBOSE_TOROUT_STRING("rcu_torture_boost task create failed");
 +              n_rcu_torture_boost_ktrerror++;
 +              boost_tasks[cpu] = NULL;
 +              mutex_unlock(&boost_mutex);
 +              return retval;
 +      }
 +      kthread_bind(boost_tasks[cpu], cpu);
 +      wake_up_process(boost_tasks[cpu]);
 +      mutex_unlock(&boost_mutex);
 +      return 0;
 +}
 +
 +/*
 + * CPU-stall kthread.  It waits as specified by stall_cpu_holdoff, then
 + * induces a CPU stall for the time specified by stall_cpu.
 + */
 +static int rcu_torture_stall(void *args)
 +{
 +      unsigned long stop_at;
 +
 +      VERBOSE_TOROUT_STRING("rcu_torture_stall task started");
 +      if (stall_cpu_holdoff > 0) {
 +              VERBOSE_TOROUT_STRING("rcu_torture_stall begin holdoff");
 +              schedule_timeout_interruptible(stall_cpu_holdoff * HZ);
 +              VERBOSE_TOROUT_STRING("rcu_torture_stall end holdoff");
 +      }
 +      if (!kthread_should_stop()) {
 +              stop_at = get_seconds() + stall_cpu;
 +              /* RCU CPU stall is expected behavior in following code. */
 +              pr_alert("rcu_torture_stall start.\n");
 +              rcu_read_lock();
 +              preempt_disable();
 +              while (ULONG_CMP_LT(get_seconds(), stop_at))
 +                      continue;  /* Induce RCU CPU stall warning. */
 +              preempt_enable();
 +              rcu_read_unlock();
 +              pr_alert("rcu_torture_stall end.\n");
 +      }
 +      torture_shutdown_absorb("rcu_torture_stall");
 +      while (!kthread_should_stop())
 +              schedule_timeout_interruptible(10 * HZ);
 +      return 0;
 +}
 +
 +/* Spawn CPU-stall kthread, if stall_cpu specified. */
 +static int __init rcu_torture_stall_init(void)
 +{
 +      if (stall_cpu <= 0)
 +              return 0;
 +      return torture_create_kthread(rcu_torture_stall, NULL, stall_task);
 +}
 +
 +/* Callback function for RCU barrier testing. */
 +void rcu_torture_barrier_cbf(struct rcu_head *rcu)
 +{
 +      atomic_inc(&barrier_cbs_invoked);
 +}
 +
 +/* kthread function to register callbacks used to test RCU barriers. */
 +static int rcu_torture_barrier_cbs(void *arg)
 +{
 +      long myid = (long)arg;
 +      bool lastphase = 0;
 +      bool newphase;
 +      struct rcu_head rcu;
 +
 +      init_rcu_head_on_stack(&rcu);
 +      VERBOSE_TOROUT_STRING("rcu_torture_barrier_cbs task started");
++      set_user_nice(current, MAX_NICE);
 +      do {
 +              wait_event(barrier_cbs_wq[myid],
 +                         (newphase =
 +                          ACCESS_ONCE(barrier_phase)) != lastphase ||
 +                         torture_must_stop());
 +              lastphase = newphase;
 +              smp_mb(); /* ensure barrier_phase load before ->call(). */
 +              if (torture_must_stop())
 +                      break;
 +              cur_ops->call(&rcu, rcu_torture_barrier_cbf);
 +              if (atomic_dec_and_test(&barrier_cbs_count))
 +                      wake_up(&barrier_wq);
 +      } while (!torture_must_stop());
 +      cur_ops->cb_barrier();
 +      destroy_rcu_head_on_stack(&rcu);
 +      torture_kthread_stopping("rcu_torture_barrier_cbs");
 +      return 0;
 +}
 +
 +/* kthread function to drive and coordinate RCU barrier testing. */
 +static int rcu_torture_barrier(void *arg)
 +{
 +      int i;
 +
 +      VERBOSE_TOROUT_STRING("rcu_torture_barrier task starting");
 +      do {
 +              atomic_set(&barrier_cbs_invoked, 0);
 +              atomic_set(&barrier_cbs_count, n_barrier_cbs);
 +              smp_mb(); /* Ensure barrier_phase after prior assignments. */
 +              barrier_phase = !barrier_phase;
 +              for (i = 0; i < n_barrier_cbs; i++)
 +                      wake_up(&barrier_cbs_wq[i]);
 +              wait_event(barrier_wq,
 +                         atomic_read(&barrier_cbs_count) == 0 ||
 +                         torture_must_stop());
 +              if (torture_must_stop())
 +                      break;
 +              n_barrier_attempts++;
 +              cur_ops->cb_barrier(); /* Implies smp_mb() for wait_event(). */
 +              if (atomic_read(&barrier_cbs_invoked) != n_barrier_cbs) {
 +                      n_rcu_torture_barrier_error++;
 +                      WARN_ON_ONCE(1);
 +              }
 +              n_barrier_successes++;
 +              schedule_timeout_interruptible(HZ / 10);
 +      } while (!torture_must_stop());
 +      torture_kthread_stopping("rcu_torture_barrier");
 +      return 0;
 +}
 +
 +/* Initialize RCU barrier testing. */
 +static int rcu_torture_barrier_init(void)
 +{
 +      int i;
 +      int ret;
 +
 +      if (n_barrier_cbs == 0)
 +              return 0;
 +      if (cur_ops->call == NULL || cur_ops->cb_barrier == NULL) {
 +              pr_alert("%s" TORTURE_FLAG
 +                       " Call or barrier ops missing for %s,\n",
 +                       torture_type, cur_ops->name);
 +              pr_alert("%s" TORTURE_FLAG
 +                       " RCU barrier testing omitted from run.\n",
 +                       torture_type);
 +              return 0;
 +      }
 +      atomic_set(&barrier_cbs_count, 0);
 +      atomic_set(&barrier_cbs_invoked, 0);
 +      barrier_cbs_tasks =
 +              kzalloc(n_barrier_cbs * sizeof(barrier_cbs_tasks[0]),
 +                      GFP_KERNEL);
 +      barrier_cbs_wq =
 +              kzalloc(n_barrier_cbs * sizeof(barrier_cbs_wq[0]),
 +                      GFP_KERNEL);
 +      if (barrier_cbs_tasks == NULL || !barrier_cbs_wq)
 +              return -ENOMEM;
 +      for (i = 0; i < n_barrier_cbs; i++) {
 +              init_waitqueue_head(&barrier_cbs_wq[i]);
 +              ret = torture_create_kthread(rcu_torture_barrier_cbs,
 +                                           (void *)(long)i,
 +                                           barrier_cbs_tasks[i]);
 +              if (ret)
 +                      return ret;
 +      }
 +      return torture_create_kthread(rcu_torture_barrier, NULL, barrier_task);
 +}
 +
 +/* Clean up after RCU barrier testing. */
 +static void rcu_torture_barrier_cleanup(void)
 +{
 +      int i;
 +
 +      torture_stop_kthread(rcu_torture_barrier, barrier_task);
 +      if (barrier_cbs_tasks != NULL) {
 +              for (i = 0; i < n_barrier_cbs; i++)
 +                      torture_stop_kthread(rcu_torture_barrier_cbs,
 +                                           barrier_cbs_tasks[i]);
 +              kfree(barrier_cbs_tasks);
 +              barrier_cbs_tasks = NULL;
 +      }
 +      if (barrier_cbs_wq != NULL) {
 +              kfree(barrier_cbs_wq);
 +              barrier_cbs_wq = NULL;
 +      }
 +}
 +
 +static int rcutorture_cpu_notify(struct notifier_block *self,
 +                               unsigned long action, void *hcpu)
 +{
 +      long cpu = (long)hcpu;
 +
 +      switch (action) {
 +      case CPU_ONLINE:
 +      case CPU_DOWN_FAILED:
 +              (void)rcutorture_booster_init(cpu);
 +              break;
 +      case CPU_DOWN_PREPARE:
 +              rcutorture_booster_cleanup(cpu);
 +              break;
 +      default:
 +              break;
 +      }
 +      return NOTIFY_OK;
 +}
 +
 +static struct notifier_block rcutorture_cpu_nb = {
 +      .notifier_call = rcutorture_cpu_notify,
 +};
 +
 +static void
 +rcu_torture_cleanup(void)
 +{
 +      int i;
 +
 +      rcutorture_record_test_transition();
 +      if (torture_cleanup()) {
 +              if (cur_ops->cb_barrier != NULL)
 +                      cur_ops->cb_barrier();
 +              return;
 +      }
 +
 +      rcu_torture_barrier_cleanup();
 +      torture_stop_kthread(rcu_torture_stall, stall_task);
 +      torture_stop_kthread(rcu_torture_writer, writer_task);
 +
 +      if (reader_tasks) {
 +              for (i = 0; i < nrealreaders; i++)
 +                      torture_stop_kthread(rcu_torture_reader,
 +                                           reader_tasks[i]);
 +              kfree(reader_tasks);
 +      }
 +      rcu_torture_current = NULL;
 +
 +      if (fakewriter_tasks) {
 +              for (i = 0; i < nfakewriters; i++) {
 +                      torture_stop_kthread(rcu_torture_fakewriter,
 +                                           fakewriter_tasks[i]);
 +              }
 +              kfree(fakewriter_tasks);
 +              fakewriter_tasks = NULL;
 +      }
 +
 +      torture_stop_kthread(rcu_torture_stats, stats_task);
 +      torture_stop_kthread(rcu_torture_fqs, fqs_task);
 +      if ((test_boost == 1 && cur_ops->can_boost) ||
 +          test_boost == 2) {
 +              unregister_cpu_notifier(&rcutorture_cpu_nb);
 +              for_each_possible_cpu(i)
 +                      rcutorture_booster_cleanup(i);
 +      }
 +
 +      /* Wait for all RCU callbacks to fire.  */
 +
 +      if (cur_ops->cb_barrier != NULL)
 +              cur_ops->cb_barrier();
 +
 +      rcu_torture_stats_print();  /* -After- the stats thread is stopped! */
 +
 +      if (atomic_read(&n_rcu_torture_error) || n_rcu_torture_barrier_error)
 +              rcu_torture_print_module_parms(cur_ops, "End of test: FAILURE");
 +      else if (torture_onoff_failures())
 +              rcu_torture_print_module_parms(cur_ops,
 +                                             "End of test: RCU_HOTPLUG");
 +      else
 +              rcu_torture_print_module_parms(cur_ops, "End of test: SUCCESS");
 +}
 +
 +#ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD
 +static void rcu_torture_leak_cb(struct rcu_head *rhp)
 +{
 +}
 +
 +static void rcu_torture_err_cb(struct rcu_head *rhp)
 +{
 +      /*
 +       * This -might- happen due to race conditions, but is unlikely.
 +       * The scenario that leads to this happening is that the
 +       * first of the pair of duplicate callbacks is queued,
 +       * someone else starts a grace period that includes that
 +       * callback, then the second of the pair must wait for the
 +       * next grace period.  Unlikely, but can happen.  If it
 +       * does happen, the debug-objects subsystem won't have splatted.
 +       */
 +      pr_alert("rcutorture: duplicated callback was invoked.\n");
 +}
 +#endif /* #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD */
 +
 +/*
 + * Verify that double-free causes debug-objects to complain, but only
 + * if CONFIG_DEBUG_OBJECTS_RCU_HEAD=y.  Otherwise, say that the test
 + * cannot be carried out.
 + */
 +static void rcu_test_debug_objects(void)
 +{
 +#ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD
 +      struct rcu_head rh1;
 +      struct rcu_head rh2;
 +
 +      init_rcu_head_on_stack(&rh1);
 +      init_rcu_head_on_stack(&rh2);
 +      pr_alert("rcutorture: WARN: Duplicate call_rcu() test starting.\n");
 +
 +      /* Try to queue the rh2 pair of callbacks for the same grace period. */
 +      preempt_disable(); /* Prevent preemption from interrupting test. */
 +      rcu_read_lock(); /* Make it impossible to finish a grace period. */
 +      call_rcu(&rh1, rcu_torture_leak_cb); /* Start grace period. */
 +      local_irq_disable(); /* Make it harder to start a new grace period. */
 +      call_rcu(&rh2, rcu_torture_leak_cb);
 +      call_rcu(&rh2, rcu_torture_err_cb); /* Duplicate callback. */
 +      local_irq_enable();
 +      rcu_read_unlock();
 +      preempt_enable();
 +
 +      /* Wait for them all to get done so we can safely return. */
 +      rcu_barrier();
 +      pr_alert("rcutorture: WARN: Duplicate call_rcu() test complete.\n");
 +      destroy_rcu_head_on_stack(&rh1);
 +      destroy_rcu_head_on_stack(&rh2);
 +#else /* #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD */
 +      pr_alert("rcutorture: !CONFIG_DEBUG_OBJECTS_RCU_HEAD, not testing duplicate call_rcu()\n");
 +#endif /* #else #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD */
 +}
 +
 +static int __init
 +rcu_torture_init(void)
 +{
 +      int i;
 +      int cpu;
 +      int firsterr = 0;
 +      static struct rcu_torture_ops *torture_ops[] = {
 +              &rcu_ops, &rcu_bh_ops, &rcu_busted_ops, &srcu_ops, &sched_ops,
 +      };
 +
 +      torture_init_begin(torture_type, verbose, &rcutorture_runnable);
 +
 +      /* Process args and tell the world that the torturer is on the job. */
 +      for (i = 0; i < ARRAY_SIZE(torture_ops); i++) {
 +              cur_ops = torture_ops[i];
 +              if (strcmp(torture_type, cur_ops->name) == 0)
 +                      break;
 +      }
 +      if (i == ARRAY_SIZE(torture_ops)) {
 +              pr_alert("rcu-torture: invalid torture type: \"%s\"\n",
 +                       torture_type);
 +              pr_alert("rcu-torture types:");
 +              for (i = 0; i < ARRAY_SIZE(torture_ops); i++)
 +                      pr_alert(" %s", torture_ops[i]->name);
 +              pr_alert("\n");
 +              torture_init_end();
 +              return -EINVAL;
 +      }
 +      if (cur_ops->fqs == NULL && fqs_duration != 0) {
 +              pr_alert("rcu-torture: ->fqs NULL and non-zero fqs_duration, fqs disabled.\n");
 +              fqs_duration = 0;
 +      }
 +      if (cur_ops->init)
 +              cur_ops->init(); /* no "goto unwind" prior to this point!!! */
 +
 +      if (nreaders >= 0)
 +              nrealreaders = nreaders;
 +      else
 +              nrealreaders = 2 * num_online_cpus();
 +      rcu_torture_print_module_parms(cur_ops, "Start of test");
 +
 +      /* Set up the freelist. */
 +
 +      INIT_LIST_HEAD(&rcu_torture_freelist);
 +      for (i = 0; i < ARRAY_SIZE(rcu_tortures); i++) {
 +              rcu_tortures[i].rtort_mbtest = 0;
 +              list_add_tail(&rcu_tortures[i].rtort_free,
 +                            &rcu_torture_freelist);
 +      }
 +
 +      /* Initialize the statistics so that each run gets its own numbers. */
 +
 +      rcu_torture_current = NULL;
 +      rcu_torture_current_version = 0;
 +      atomic_set(&n_rcu_torture_alloc, 0);
 +      atomic_set(&n_rcu_torture_alloc_fail, 0);
 +      atomic_set(&n_rcu_torture_free, 0);
 +      atomic_set(&n_rcu_torture_mberror, 0);
 +      atomic_set(&n_rcu_torture_error, 0);
 +      n_rcu_torture_barrier_error = 0;
 +      n_rcu_torture_boost_ktrerror = 0;
 +      n_rcu_torture_boost_rterror = 0;
 +      n_rcu_torture_boost_failure = 0;
 +      n_rcu_torture_boosts = 0;
 +      for (i = 0; i < RCU_TORTURE_PIPE_LEN + 1; i++)
 +              atomic_set(&rcu_torture_wcount[i], 0);
 +      for_each_possible_cpu(cpu) {
 +              for (i = 0; i < RCU_TORTURE_PIPE_LEN + 1; i++) {
 +                      per_cpu(rcu_torture_count, cpu)[i] = 0;
 +                      per_cpu(rcu_torture_batch, cpu)[i] = 0;
 +              }
 +      }
 +
 +      /* Start up the kthreads. */
 +
 +      firsterr = torture_create_kthread(rcu_torture_writer, NULL,
 +                                        writer_task);
 +      if (firsterr)
 +              goto unwind;
 +      fakewriter_tasks = kzalloc(nfakewriters * sizeof(fakewriter_tasks[0]),
 +                                 GFP_KERNEL);
 +      if (fakewriter_tasks == NULL) {
 +              VERBOSE_TOROUT_ERRSTRING("out of memory");
 +              firsterr = -ENOMEM;
 +              goto unwind;
 +      }
 +      for (i = 0; i < nfakewriters; i++) {
 +              firsterr = torture_create_kthread(rcu_torture_fakewriter,
 +                                                NULL, fakewriter_tasks[i]);
 +              if (firsterr)
 +                      goto unwind;
 +      }
 +      reader_tasks = kzalloc(nrealreaders * sizeof(reader_tasks[0]),
 +                             GFP_KERNEL);
 +      if (reader_tasks == NULL) {
 +              VERBOSE_TOROUT_ERRSTRING("out of memory");
 +              firsterr = -ENOMEM;
 +              goto unwind;
 +      }
 +      for (i = 0; i < nrealreaders; i++) {
 +              firsterr = torture_create_kthread(rcu_torture_reader, NULL,
 +                                                reader_tasks[i]);
 +              if (firsterr)
 +                      goto unwind;
 +      }
 +      if (stat_interval > 0) {
 +              firsterr = torture_create_kthread(rcu_torture_stats, NULL,
 +                                                stats_task);
 +              if (firsterr)
 +                      goto unwind;
 +      }
 +      if (test_no_idle_hz) {
 +              firsterr = torture_shuffle_init(shuffle_interval * HZ);
 +              if (firsterr)
 +                      goto unwind;
 +      }
 +      if (stutter < 0)
 +              stutter = 0;
 +      if (stutter) {
 +              firsterr = torture_stutter_init(stutter * HZ);
 +              if (firsterr)
 +                      goto unwind;
 +      }
 +      if (fqs_duration < 0)
 +              fqs_duration = 0;
 +      if (fqs_duration) {
 +              /* Create the fqs thread */
 +              torture_create_kthread(rcu_torture_fqs, NULL, fqs_task);
 +              if (firsterr)
 +                      goto unwind;
 +      }
 +      if (test_boost_interval < 1)
 +              test_boost_interval = 1;
 +      if (test_boost_duration < 2)
 +              test_boost_duration = 2;
 +      if ((test_boost == 1 && cur_ops->can_boost) ||
 +          test_boost == 2) {
 +
 +              boost_starttime = jiffies + test_boost_interval * HZ;
 +              register_cpu_notifier(&rcutorture_cpu_nb);
 +              for_each_possible_cpu(i) {
 +                      if (cpu_is_offline(i))
 +                              continue;  /* Heuristic: CPU can go offline. */
 +                      firsterr = rcutorture_booster_init(i);
 +                      if (firsterr)
 +                              goto unwind;
 +              }
 +      }
 +      firsterr = torture_shutdown_init(shutdown_secs, rcu_torture_cleanup);
 +      if (firsterr)
 +              goto unwind;
 +      firsterr = torture_onoff_init(onoff_holdoff * HZ, onoff_interval * HZ);
 +      if (firsterr)
 +              goto unwind;
 +      firsterr = rcu_torture_stall_init();
 +      if (firsterr)
 +              goto unwind;
 +      firsterr = rcu_torture_barrier_init();
 +      if (firsterr)
 +              goto unwind;
 +      if (object_debug)
 +              rcu_test_debug_objects();
 +      rcutorture_record_test_transition();
 +      torture_init_end();
 +      return 0;
 +
 +unwind:
 +      torture_init_end();
 +      rcu_torture_cleanup();
 +      return firsterr;
 +}
 +
 +module_init(rcu_torture_init);
 +module_exit(rcu_torture_cleanup);