xref: /linux/kernel/rcu/rcutorture.c (revision f49040c7aaa5532a1f94355ef5073c49e6b32349)
1  // SPDX-License-Identifier: GPL-2.0+
2  /*
3   * Read-Copy Update module-based torture test facility
4   *
5   * Copyright (C) IBM Corporation, 2005, 2006
6   *
7   * Authors: Paul E. McKenney <paulmck@linux.ibm.com>
8   *	  Josh Triplett <josh@joshtriplett.org>
9   *
10   * See also:  Documentation/RCU/torture.rst
11   */
12  
13  #define pr_fmt(fmt) fmt
14  
15  #include <linux/types.h>
16  #include <linux/kernel.h>
17  #include <linux/init.h>
18  #include <linux/module.h>
19  #include <linux/kthread.h>
20  #include <linux/err.h>
21  #include <linux/spinlock.h>
22  #include <linux/smp.h>
23  #include <linux/rcupdate_wait.h>
24  #include <linux/rcu_notifier.h>
25  #include <linux/interrupt.h>
26  #include <linux/sched/signal.h>
27  #include <uapi/linux/sched/types.h>
28  #include <linux/atomic.h>
29  #include <linux/bitops.h>
30  #include <linux/completion.h>
31  #include <linux/moduleparam.h>
32  #include <linux/percpu.h>
33  #include <linux/notifier.h>
34  #include <linux/reboot.h>
35  #include <linux/freezer.h>
36  #include <linux/cpu.h>
37  #include <linux/delay.h>
38  #include <linux/stat.h>
39  #include <linux/srcu.h>
40  #include <linux/slab.h>
41  #include <linux/trace_clock.h>
42  #include <asm/byteorder.h>
43  #include <linux/torture.h>
44  #include <linux/vmalloc.h>
45  #include <linux/sched/debug.h>
46  #include <linux/sched/sysctl.h>
47  #include <linux/oom.h>
48  #include <linux/tick.h>
49  #include <linux/rcupdate_trace.h>
50  #include <linux/nmi.h>
51  
52  #include "rcu.h"
53  
54  MODULE_DESCRIPTION("Read-Copy Update module-based torture test facility");
55  MODULE_LICENSE("GPL");
56  MODULE_AUTHOR("Paul E. McKenney <paulmck@linux.ibm.com> and Josh Triplett <josh@joshtriplett.org>");
57  
58  /* Bits for ->extendables field, extendables param, and related definitions. */
59  #define RCUTORTURE_RDR_SHIFT_1	 8	/* Put SRCU index in upper bits. */
60  #define RCUTORTURE_RDR_MASK_1	 (0xff << RCUTORTURE_RDR_SHIFT_1)
61  #define RCUTORTURE_RDR_SHIFT_2	 16	/* Put SRCU index in upper bits. */
62  #define RCUTORTURE_RDR_MASK_2	 (0xff << RCUTORTURE_RDR_SHIFT_2)
63  #define RCUTORTURE_RDR_BH	 0x01	/* Extend readers by disabling bh. */
64  #define RCUTORTURE_RDR_IRQ	 0x02	/*  ... disabling interrupts. */
65  #define RCUTORTURE_RDR_PREEMPT	 0x04	/*  ... disabling preemption. */
66  #define RCUTORTURE_RDR_RBH	 0x08	/*  ... rcu_read_lock_bh(). */
67  #define RCUTORTURE_RDR_SCHED	 0x10	/*  ... rcu_read_lock_sched(). */
68  #define RCUTORTURE_RDR_RCU_1	 0x20	/*  ... entering another RCU reader. */
69  #define RCUTORTURE_RDR_RCU_2	 0x40	/*  ... entering another RCU reader. */
70  #define RCUTORTURE_RDR_NBITS	 7	/* Number of bits defined above. */
71  #define RCUTORTURE_MAX_EXTEND	 \
72  	(RCUTORTURE_RDR_BH | RCUTORTURE_RDR_IRQ | RCUTORTURE_RDR_PREEMPT | \
73  	 RCUTORTURE_RDR_RBH | RCUTORTURE_RDR_SCHED)
74  #define RCUTORTURE_RDR_ALLBITS	\
75  	(RCUTORTURE_MAX_EXTEND | RCUTORTURE_RDR_RCU_1 | RCUTORTURE_RDR_RCU_2 | \
76  	 RCUTORTURE_RDR_MASK_1 | RCUTORTURE_RDR_MASK_2)
77  #define RCUTORTURE_RDR_MAX_LOOPS 0x7	/* Maximum reader extensions. */
78  					/* Must be power of two minus one. */
79  #define RCUTORTURE_RDR_MAX_SEGS (RCUTORTURE_RDR_MAX_LOOPS + 3)
80  
81  torture_param(int, extendables, RCUTORTURE_MAX_EXTEND,
82  	      "Extend readers by disabling bh (1), irqs (2), or preempt (4)");
83  torture_param(int, fqs_duration, 0, "Duration of fqs bursts (us), 0 to disable");
84  torture_param(int, fqs_holdoff, 0, "Holdoff time within fqs bursts (us)");
85  torture_param(int, fqs_stutter, 3, "Wait time between fqs bursts (s)");
86  torture_param(int, fwd_progress, 1, "Number of grace-period forward progress tasks (0 to disable)");
87  torture_param(int, fwd_progress_div, 4, "Fraction of CPU stall to wait");
88  torture_param(int, fwd_progress_holdoff, 60, "Time between forward-progress tests (s)");
89  torture_param(bool, fwd_progress_need_resched, 1, "Hide cond_resched() behind need_resched()");
90  torture_param(bool, gp_cond, false, "Use conditional/async GP wait primitives");
91  torture_param(bool, gp_cond_exp, false, "Use conditional/async expedited GP wait primitives");
92  torture_param(bool, gp_cond_full, false, "Use conditional/async full-state GP wait primitives");
93  torture_param(bool, gp_cond_exp_full, false,
94  		    "Use conditional/async full-stateexpedited GP wait primitives");
95  torture_param(int, gp_cond_wi, 16 * USEC_PER_SEC / HZ,
96  		   "Wait interval for normal conditional grace periods, us (default 16 jiffies)");
97  torture_param(int, gp_cond_wi_exp, 128,
98  		   "Wait interval for expedited conditional grace periods, us (default 128 us)");
99  torture_param(bool, gp_exp, false, "Use expedited GP wait primitives");
100  torture_param(bool, gp_normal, false, "Use normal (non-expedited) GP wait primitives");
101  torture_param(bool, gp_poll, false, "Use polling GP wait primitives");
102  torture_param(bool, gp_poll_exp, false, "Use polling expedited GP wait primitives");
103  torture_param(bool, gp_poll_full, false, "Use polling full-state GP wait primitives");
104  torture_param(bool, gp_poll_exp_full, false, "Use polling full-state expedited GP wait primitives");
105  torture_param(int, gp_poll_wi, 16 * USEC_PER_SEC / HZ,
106  		   "Wait interval for normal polled grace periods, us (default 16 jiffies)");
107  torture_param(int, gp_poll_wi_exp, 128,
108  		   "Wait interval for expedited polled grace periods, us (default 128 us)");
109  torture_param(bool, gp_sync, false, "Use synchronous GP wait primitives");
110  torture_param(int, irqreader, 1, "Allow RCU readers from irq handlers");
111  torture_param(int, leakpointer, 0, "Leak pointer dereferences from readers");
112  torture_param(int, n_barrier_cbs, 0, "# of callbacks/kthreads for barrier testing");
113  torture_param(int, nfakewriters, 4, "Number of RCU fake writer threads");
114  torture_param(int, nreaders, -1, "Number of RCU reader threads");
115  torture_param(int, object_debug, 0, "Enable debug-object double call_rcu() testing");
116  torture_param(int, onoff_holdoff, 0, "Time after boot before CPU hotplugs (s)");
117  torture_param(int, onoff_interval, 0, "Time between CPU hotplugs (jiffies), 0=disable");
118  torture_param(int, nocbs_nthreads, 0, "Number of NOCB toggle threads, 0 to disable");
119  torture_param(int, nocbs_toggle, 1000, "Time between toggling nocb state (ms)");
120  torture_param(int, preempt_duration, 0, "Preemption duration (ms), zero to disable");
121  torture_param(int, preempt_interval, MSEC_PER_SEC, "Interval between preemptions (ms)");
122  torture_param(int, read_exit_delay, 13, "Delay between read-then-exit episodes (s)");
123  torture_param(int, read_exit_burst, 16, "# of read-then-exit bursts per episode, zero to disable");
124  torture_param(int, reader_flavor, SRCU_READ_FLAVOR_NORMAL, "Reader flavors to use, one per bit.");
125  torture_param(int, shuffle_interval, 3, "Number of seconds between shuffles");
126  torture_param(int, shutdown_secs, 0, "Shutdown time (s), <= zero to disable.");
127  torture_param(int, stall_cpu, 0, "Stall duration (s), zero to disable.");
128  torture_param(int, stall_cpu_holdoff, 10, "Time to wait before starting stall (s).");
129  torture_param(bool, stall_no_softlockup, false, "Avoid softlockup warning during cpu stall.");
130  torture_param(int, stall_cpu_irqsoff, 0, "Disable interrupts while stalling.");
131  torture_param(int, stall_cpu_block, 0, "Sleep while stalling.");
132  torture_param(int, stall_cpu_repeat, 0, "Number of additional stalls after the first one.");
133  torture_param(int, stall_gp_kthread, 0, "Grace-period kthread stall duration (s).");
134  torture_param(int, stat_interval, 60, "Number of seconds between stats printk()s");
135  torture_param(int, stutter, 5, "Number of seconds to run/halt test");
136  torture_param(int, test_boost, 1, "Test RCU prio boost: 0=no, 1=maybe, 2=yes.");
137  torture_param(int, test_boost_duration, 4, "Duration of each boost test, seconds.");
138  torture_param(int, test_boost_holdoff, 0, "Holdoff time from rcutorture start, seconds.");
139  torture_param(int, test_boost_interval, 7, "Interval between boost tests, seconds.");
140  torture_param(int, test_nmis, 0, "End-test NMI tests, 0 to disable.");
141  torture_param(bool, test_no_idle_hz, true, "Test support for tickless idle CPUs");
142  torture_param(int, test_srcu_lockdep, 0, "Test specified SRCU deadlock scenario.");
143  torture_param(int, verbose, 1, "Enable verbose debugging printk()s");
144  
145  static char *torture_type = "rcu";
146  module_param(torture_type, charp, 0444);
147  MODULE_PARM_DESC(torture_type, "Type of RCU to torture (rcu, srcu, ...)");
148  
149  static int nrealnocbers;
150  static int nrealreaders;
151  static int nrealfakewriters;
152  static struct task_struct *writer_task;
153  static struct task_struct **fakewriter_tasks;
154  static struct task_struct **reader_tasks;
155  static struct task_struct **nocb_tasks;
156  static struct task_struct *stats_task;
157  static struct task_struct *fqs_task;
158  static struct task_struct *boost_tasks[NR_CPUS];
159  static struct task_struct *stall_task;
160  static struct task_struct **fwd_prog_tasks;
161  static struct task_struct **barrier_cbs_tasks;
162  static struct task_struct *barrier_task;
163  static struct task_struct *read_exit_task;
164  static struct task_struct *preempt_task;
165  
166  #define RCU_TORTURE_PIPE_LEN 10
167  
168  // Mailbox-like structure to check RCU global memory ordering.
169  struct rcu_torture_reader_check {
170  	unsigned long rtc_myloops;
171  	int rtc_chkrdr;
172  	unsigned long rtc_chkloops;
173  	int rtc_ready;
174  	struct rcu_torture_reader_check *rtc_assigner;
175  } ____cacheline_internodealigned_in_smp;
176  
177  // Update-side data structure used to check RCU readers.
178  struct rcu_torture {
179  	struct rcu_head rtort_rcu;
180  	int rtort_pipe_count;
181  	struct list_head rtort_free;
182  	int rtort_mbtest;
183  	struct rcu_torture_reader_check *rtort_chkp;
184  };
185  
186  static LIST_HEAD(rcu_torture_freelist);
187  static struct rcu_torture __rcu *rcu_torture_current;
188  static unsigned long rcu_torture_current_version;
189  static struct rcu_torture rcu_tortures[10 * RCU_TORTURE_PIPE_LEN];
190  static DEFINE_SPINLOCK(rcu_torture_lock);
191  static DEFINE_PER_CPU(long [RCU_TORTURE_PIPE_LEN + 1], rcu_torture_count);
192  static DEFINE_PER_CPU(long [RCU_TORTURE_PIPE_LEN + 1], rcu_torture_batch);
193  static atomic_t rcu_torture_wcount[RCU_TORTURE_PIPE_LEN + 1];
194  static struct rcu_torture_reader_check *rcu_torture_reader_mbchk;
195  static atomic_t n_rcu_torture_alloc;
196  static atomic_t n_rcu_torture_alloc_fail;
197  static atomic_t n_rcu_torture_free;
198  static atomic_t n_rcu_torture_mberror;
199  static atomic_t n_rcu_torture_mbchk_fail;
200  static atomic_t n_rcu_torture_mbchk_tries;
201  static atomic_t n_rcu_torture_error;
202  static long n_rcu_torture_barrier_error;
203  static long n_rcu_torture_boost_ktrerror;
204  static long n_rcu_torture_boost_failure;
205  static long n_rcu_torture_boosts;
206  static atomic_long_t n_rcu_torture_timers;
207  static long n_barrier_attempts;
208  static long n_barrier_successes; /* did rcu_barrier test succeed? */
209  static unsigned long n_read_exits;
210  static struct list_head rcu_torture_removed;
211  static unsigned long shutdown_jiffies;
212  static unsigned long start_gp_seq;
213  static atomic_long_t n_nocb_offload;
214  static atomic_long_t n_nocb_deoffload;
215  
216  static int rcu_torture_writer_state;
217  #define RTWS_FIXED_DELAY	0
218  #define RTWS_DELAY		1
219  #define RTWS_REPLACE		2
220  #define RTWS_DEF_FREE		3
221  #define RTWS_EXP_SYNC		4
222  #define RTWS_COND_GET		5
223  #define RTWS_COND_GET_FULL	6
224  #define RTWS_COND_GET_EXP	7
225  #define RTWS_COND_GET_EXP_FULL	8
226  #define RTWS_COND_SYNC		9
227  #define RTWS_COND_SYNC_FULL	10
228  #define RTWS_COND_SYNC_EXP	11
229  #define RTWS_COND_SYNC_EXP_FULL	12
230  #define RTWS_POLL_GET		13
231  #define RTWS_POLL_GET_FULL	14
232  #define RTWS_POLL_GET_EXP	15
233  #define RTWS_POLL_GET_EXP_FULL	16
234  #define RTWS_POLL_WAIT		17
235  #define RTWS_POLL_WAIT_FULL	18
236  #define RTWS_POLL_WAIT_EXP	19
237  #define RTWS_POLL_WAIT_EXP_FULL	20
238  #define RTWS_SYNC		21
239  #define RTWS_STUTTER		22
240  #define RTWS_STOPPING		23
241  static const char * const rcu_torture_writer_state_names[] = {
242  	"RTWS_FIXED_DELAY",
243  	"RTWS_DELAY",
244  	"RTWS_REPLACE",
245  	"RTWS_DEF_FREE",
246  	"RTWS_EXP_SYNC",
247  	"RTWS_COND_GET",
248  	"RTWS_COND_GET_FULL",
249  	"RTWS_COND_GET_EXP",
250  	"RTWS_COND_GET_EXP_FULL",
251  	"RTWS_COND_SYNC",
252  	"RTWS_COND_SYNC_FULL",
253  	"RTWS_COND_SYNC_EXP",
254  	"RTWS_COND_SYNC_EXP_FULL",
255  	"RTWS_POLL_GET",
256  	"RTWS_POLL_GET_FULL",
257  	"RTWS_POLL_GET_EXP",
258  	"RTWS_POLL_GET_EXP_FULL",
259  	"RTWS_POLL_WAIT",
260  	"RTWS_POLL_WAIT_FULL",
261  	"RTWS_POLL_WAIT_EXP",
262  	"RTWS_POLL_WAIT_EXP_FULL",
263  	"RTWS_SYNC",
264  	"RTWS_STUTTER",
265  	"RTWS_STOPPING",
266  };
267  
268  /* Record reader segment types and duration for first failing read. */
269  struct rt_read_seg {
270  	int rt_readstate;
271  	unsigned long rt_delay_jiffies;
272  	unsigned long rt_delay_ms;
273  	unsigned long rt_delay_us;
274  	bool rt_preempted;
275  	int rt_cpu;
276  	int rt_end_cpu;
277  	unsigned long long rt_gp_seq;
278  	unsigned long long rt_gp_seq_end;
279  	u64 rt_ts;
280  };
281  static int err_segs_recorded;
282  static struct rt_read_seg err_segs[RCUTORTURE_RDR_MAX_SEGS];
283  static int rt_read_nsegs;
284  static int rt_read_preempted;
285  
rcu_torture_writer_state_getname(void)286  static const char *rcu_torture_writer_state_getname(void)
287  {
288  	unsigned int i = READ_ONCE(rcu_torture_writer_state);
289  
290  	if (i >= ARRAY_SIZE(rcu_torture_writer_state_names))
291  		return "???";
292  	return rcu_torture_writer_state_names[i];
293  }
294  
295  #ifdef CONFIG_RCU_TRACE
rcu_trace_clock_local(void)296  static u64 notrace rcu_trace_clock_local(void)
297  {
298  	u64 ts = trace_clock_local();
299  
300  	(void)do_div(ts, NSEC_PER_USEC);
301  	return ts;
302  }
303  #else /* #ifdef CONFIG_RCU_TRACE */
rcu_trace_clock_local(void)304  static u64 notrace rcu_trace_clock_local(void)
305  {
306  	return 0ULL;
307  }
308  #endif /* #else #ifdef CONFIG_RCU_TRACE */
309  
310  /*
311   * Stop aggressive CPU-hog tests a bit before the end of the test in order
312   * to avoid interfering with test shutdown.
313   */
shutdown_time_arrived(void)314  static bool shutdown_time_arrived(void)
315  {
316  	return shutdown_secs && time_after(jiffies, shutdown_jiffies - 30 * HZ);
317  }
318  
319  static unsigned long boost_starttime;	/* jiffies of next boost test start. */
320  static DEFINE_MUTEX(boost_mutex);	/* protect setting boost_starttime */
321  					/*  and boost task create/destroy. */
322  static atomic_t barrier_cbs_count;	/* Barrier callbacks registered. */
323  static bool barrier_phase;		/* Test phase. */
324  static atomic_t barrier_cbs_invoked;	/* Barrier callbacks invoked. */
325  static wait_queue_head_t *barrier_cbs_wq; /* Coordinate barrier testing. */
326  static DECLARE_WAIT_QUEUE_HEAD(barrier_wq);
327  
328  static atomic_t rcu_fwd_cb_nodelay;	/* Short rcu_torture_delay() delays. */
329  
330  /*
331   * Allocate an element from the rcu_tortures pool.
332   */
333  static struct rcu_torture *
rcu_torture_alloc(void)334  rcu_torture_alloc(void)
335  {
336  	struct list_head *p;
337  
338  	spin_lock_bh(&rcu_torture_lock);
339  	if (list_empty(&rcu_torture_freelist)) {
340  		atomic_inc(&n_rcu_torture_alloc_fail);
341  		spin_unlock_bh(&rcu_torture_lock);
342  		return NULL;
343  	}
344  	atomic_inc(&n_rcu_torture_alloc);
345  	p = rcu_torture_freelist.next;
346  	list_del_init(p);
347  	spin_unlock_bh(&rcu_torture_lock);
348  	return container_of(p, struct rcu_torture, rtort_free);
349  }
350  
351  /*
352   * Free an element to the rcu_tortures pool.
353   */
354  static void
rcu_torture_free(struct rcu_torture * p)355  rcu_torture_free(struct rcu_torture *p)
356  {
357  	atomic_inc(&n_rcu_torture_free);
358  	spin_lock_bh(&rcu_torture_lock);
359  	list_add_tail(&p->rtort_free, &rcu_torture_freelist);
360  	spin_unlock_bh(&rcu_torture_lock);
361  }
362  
363  /*
364   * Operations vector for selecting different types of tests.
365   */
366  
367  struct rcu_torture_ops {
368  	int ttype;
369  	void (*init)(void);
370  	void (*cleanup)(void);
371  	int (*readlock)(void);
372  	void (*read_delay)(struct torture_random_state *rrsp,
373  			   struct rt_read_seg *rtrsp);
374  	void (*readunlock)(int idx);
375  	int (*readlock_held)(void);   // lockdep.
376  	int (*readlock_nesting)(void); // actual nesting, if available, -1 if not.
377  	unsigned long (*get_gp_seq)(void);
378  	unsigned long (*gp_diff)(unsigned long new, unsigned long old);
379  	void (*deferred_free)(struct rcu_torture *p);
380  	void (*sync)(void);
381  	void (*exp_sync)(void);
382  	unsigned long (*get_gp_state_exp)(void);
383  	unsigned long (*start_gp_poll_exp)(void);
384  	void (*start_gp_poll_exp_full)(struct rcu_gp_oldstate *rgosp);
385  	bool (*poll_gp_state_exp)(unsigned long oldstate);
386  	void (*cond_sync_exp)(unsigned long oldstate);
387  	void (*cond_sync_exp_full)(struct rcu_gp_oldstate *rgosp);
388  	unsigned long (*get_comp_state)(void);
389  	void (*get_comp_state_full)(struct rcu_gp_oldstate *rgosp);
390  	bool (*same_gp_state)(unsigned long oldstate1, unsigned long oldstate2);
391  	bool (*same_gp_state_full)(struct rcu_gp_oldstate *rgosp1, struct rcu_gp_oldstate *rgosp2);
392  	unsigned long (*get_gp_state)(void);
393  	void (*get_gp_state_full)(struct rcu_gp_oldstate *rgosp);
394  	unsigned long (*start_gp_poll)(void);
395  	void (*start_gp_poll_full)(struct rcu_gp_oldstate *rgosp);
396  	bool (*poll_gp_state)(unsigned long oldstate);
397  	bool (*poll_gp_state_full)(struct rcu_gp_oldstate *rgosp);
398  	bool (*poll_need_2gp)(bool poll, bool poll_full);
399  	void (*cond_sync)(unsigned long oldstate);
400  	void (*cond_sync_full)(struct rcu_gp_oldstate *rgosp);
401  	int poll_active;
402  	int poll_active_full;
403  	call_rcu_func_t call;
404  	void (*cb_barrier)(void);
405  	void (*fqs)(void);
406  	void (*stats)(void);
407  	void (*gp_kthread_dbg)(void);
408  	bool (*check_boost_failed)(unsigned long gp_state, int *cpup);
409  	int (*stall_dur)(void);
410  	void (*get_gp_data)(int *flags, unsigned long *gp_seq);
411  	void (*gp_slow_register)(atomic_t *rgssp);
412  	void (*gp_slow_unregister)(atomic_t *rgssp);
413  	bool (*reader_blocked)(void);
414  	unsigned long long (*gather_gp_seqs)(void);
415  	void (*format_gp_seqs)(unsigned long long seqs, char *cp, size_t len);
416  	long cbflood_max;
417  	int irq_capable;
418  	int can_boost;
419  	int extendables;
420  	int slow_gps;
421  	int no_pi_lock;
422  	int debug_objects;
423  	int start_poll_irqsoff;
424  	const char *name;
425  };
426  
427  static struct rcu_torture_ops *cur_ops;
428  
429  /*
430   * Definitions for rcu torture testing.
431   */
432  
torture_readlock_not_held(void)433  static int torture_readlock_not_held(void)
434  {
435  	return rcu_read_lock_bh_held() || rcu_read_lock_sched_held();
436  }
437  
rcu_torture_read_lock(void)438  static int rcu_torture_read_lock(void)
439  {
440  	rcu_read_lock();
441  	return 0;
442  }
443  
444  static void
rcu_read_delay(struct torture_random_state * rrsp,struct rt_read_seg * rtrsp)445  rcu_read_delay(struct torture_random_state *rrsp, struct rt_read_seg *rtrsp)
446  {
447  	unsigned long started;
448  	unsigned long completed;
449  	const unsigned long shortdelay_us = 200;
450  	unsigned long longdelay_ms = 300;
451  	unsigned long long ts;
452  
453  	/* We want a short delay sometimes to make a reader delay the grace
454  	 * period, and we want a long delay occasionally to trigger
455  	 * force_quiescent_state. */
456  
457  	if (!atomic_read(&rcu_fwd_cb_nodelay) &&
458  	    !(torture_random(rrsp) % (nrealreaders * 2000 * longdelay_ms))) {
459  		started = cur_ops->get_gp_seq();
460  		ts = rcu_trace_clock_local();
461  		if (preempt_count() & (SOFTIRQ_MASK | HARDIRQ_MASK))
462  			longdelay_ms = 5; /* Avoid triggering BH limits. */
463  		mdelay(longdelay_ms);
464  		rtrsp->rt_delay_ms = longdelay_ms;
465  		completed = cur_ops->get_gp_seq();
466  		do_trace_rcu_torture_read(cur_ops->name, NULL, ts,
467  					  started, completed);
468  	}
469  	if (!(torture_random(rrsp) % (nrealreaders * 2 * shortdelay_us))) {
470  		udelay(shortdelay_us);
471  		rtrsp->rt_delay_us = shortdelay_us;
472  	}
473  	if (!preempt_count() &&
474  	    !(torture_random(rrsp) % (nrealreaders * 500)))
475  		torture_preempt_schedule();  /* QS only if preemptible. */
476  }
477  
rcu_torture_read_unlock(int idx)478  static void rcu_torture_read_unlock(int idx)
479  {
480  	rcu_read_unlock();
481  }
482  
rcu_torture_readlock_nesting(void)483  static int rcu_torture_readlock_nesting(void)
484  {
485  	if (IS_ENABLED(CONFIG_PREEMPT_RCU))
486  		return rcu_preempt_depth();
487  	if (IS_ENABLED(CONFIG_PREEMPT_COUNT))
488  		return (preempt_count() & PREEMPT_MASK);
489  	return -1;
490  }
491  
492  /*
493   * Update callback in the pipe.  This should be invoked after a grace period.
494   */
495  static bool
rcu_torture_pipe_update_one(struct rcu_torture * rp)496  rcu_torture_pipe_update_one(struct rcu_torture *rp)
497  {
498  	int i;
499  	struct rcu_torture_reader_check *rtrcp = READ_ONCE(rp->rtort_chkp);
500  
501  	if (rtrcp) {
502  		WRITE_ONCE(rp->rtort_chkp, NULL);
503  		smp_store_release(&rtrcp->rtc_ready, 1); // Pair with smp_load_acquire().
504  	}
505  	i = rp->rtort_pipe_count;
506  	if (i > RCU_TORTURE_PIPE_LEN)
507  		i = RCU_TORTURE_PIPE_LEN;
508  	atomic_inc(&rcu_torture_wcount[i]);
509  	WRITE_ONCE(rp->rtort_pipe_count, i + 1);
510  	ASSERT_EXCLUSIVE_WRITER(rp->rtort_pipe_count);
511  	if (i + 1 >= RCU_TORTURE_PIPE_LEN) {
512  		rp->rtort_mbtest = 0;
513  		return true;
514  	}
515  	return false;
516  }
517  
518  /*
519   * Update all callbacks in the pipe.  Suitable for synchronous grace-period
520   * primitives.
521   */
522  static void
rcu_torture_pipe_update(struct rcu_torture * old_rp)523  rcu_torture_pipe_update(struct rcu_torture *old_rp)
524  {
525  	struct rcu_torture *rp;
526  	struct rcu_torture *rp1;
527  
528  	if (old_rp)
529  		list_add(&old_rp->rtort_free, &rcu_torture_removed);
530  	list_for_each_entry_safe(rp, rp1, &rcu_torture_removed, rtort_free) {
531  		if (rcu_torture_pipe_update_one(rp)) {
532  			list_del(&rp->rtort_free);
533  			rcu_torture_free(rp);
534  		}
535  	}
536  }
537  
538  static void
rcu_torture_cb(struct rcu_head * p)539  rcu_torture_cb(struct rcu_head *p)
540  {
541  	struct rcu_torture *rp = container_of(p, struct rcu_torture, rtort_rcu);
542  
543  	if (torture_must_stop_irq()) {
544  		/* Test is ending, just drop callbacks on the floor. */
545  		/* The next initialization will pick up the pieces. */
546  		return;
547  	}
548  	if (rcu_torture_pipe_update_one(rp))
549  		rcu_torture_free(rp);
550  	else
551  		cur_ops->deferred_free(rp);
552  }
553  
rcu_no_completed(void)554  static unsigned long rcu_no_completed(void)
555  {
556  	return 0;
557  }
558  
rcu_torture_deferred_free(struct rcu_torture * p)559  static void rcu_torture_deferred_free(struct rcu_torture *p)
560  {
561  	call_rcu_hurry(&p->rtort_rcu, rcu_torture_cb);
562  }
563  
rcu_sync_torture_init(void)564  static void rcu_sync_torture_init(void)
565  {
566  	INIT_LIST_HEAD(&rcu_torture_removed);
567  }
568  
rcu_poll_need_2gp(bool poll,bool poll_full)569  static bool rcu_poll_need_2gp(bool poll, bool poll_full)
570  {
571  	return poll;
572  }
573  
574  static struct rcu_torture_ops rcu_ops = {
575  	.ttype			= RCU_FLAVOR,
576  	.init			= rcu_sync_torture_init,
577  	.readlock		= rcu_torture_read_lock,
578  	.read_delay		= rcu_read_delay,
579  	.readunlock		= rcu_torture_read_unlock,
580  	.readlock_held		= torture_readlock_not_held,
581  	.readlock_nesting	= rcu_torture_readlock_nesting,
582  	.get_gp_seq		= rcu_get_gp_seq,
583  	.gp_diff		= rcu_seq_diff,
584  	.deferred_free		= rcu_torture_deferred_free,
585  	.sync			= synchronize_rcu,
586  	.exp_sync		= synchronize_rcu_expedited,
587  	.same_gp_state		= same_state_synchronize_rcu,
588  	.same_gp_state_full	= same_state_synchronize_rcu_full,
589  	.get_comp_state		= get_completed_synchronize_rcu,
590  	.get_comp_state_full	= get_completed_synchronize_rcu_full,
591  	.get_gp_state		= get_state_synchronize_rcu,
592  	.get_gp_state_full	= get_state_synchronize_rcu_full,
593  	.start_gp_poll		= start_poll_synchronize_rcu,
594  	.start_gp_poll_full	= start_poll_synchronize_rcu_full,
595  	.poll_gp_state		= poll_state_synchronize_rcu,
596  	.poll_gp_state_full	= poll_state_synchronize_rcu_full,
597  	.poll_need_2gp		= rcu_poll_need_2gp,
598  	.cond_sync		= cond_synchronize_rcu,
599  	.cond_sync_full		= cond_synchronize_rcu_full,
600  	.poll_active		= NUM_ACTIVE_RCU_POLL_OLDSTATE,
601  	.poll_active_full	= NUM_ACTIVE_RCU_POLL_FULL_OLDSTATE,
602  	.get_gp_state_exp	= get_state_synchronize_rcu,
603  	.start_gp_poll_exp	= start_poll_synchronize_rcu_expedited,
604  	.start_gp_poll_exp_full	= start_poll_synchronize_rcu_expedited_full,
605  	.poll_gp_state_exp	= poll_state_synchronize_rcu,
606  	.cond_sync_exp		= cond_synchronize_rcu_expedited,
607  	.cond_sync_exp_full	= cond_synchronize_rcu_expedited_full,
608  	.call			= call_rcu_hurry,
609  	.cb_barrier		= rcu_barrier,
610  	.fqs			= rcu_force_quiescent_state,
611  	.gp_kthread_dbg		= show_rcu_gp_kthreads,
612  	.check_boost_failed	= rcu_check_boost_fail,
613  	.stall_dur		= rcu_jiffies_till_stall_check,
614  	.get_gp_data		= rcutorture_get_gp_data,
615  	.gp_slow_register	= rcu_gp_slow_register,
616  	.gp_slow_unregister	= rcu_gp_slow_unregister,
617  	.reader_blocked		= IS_ENABLED(CONFIG_RCU_TORTURE_TEST_LOG_CPU)
618  				  ? has_rcu_reader_blocked
619  				  : NULL,
620  	.gather_gp_seqs		= rcutorture_gather_gp_seqs,
621  	.format_gp_seqs		= rcutorture_format_gp_seqs,
622  	.irq_capable		= 1,
623  	.can_boost		= IS_ENABLED(CONFIG_RCU_BOOST),
624  	.extendables		= RCUTORTURE_MAX_EXTEND,
625  	.debug_objects		= 1,
626  	.start_poll_irqsoff	= 1,
627  	.name			= "rcu"
628  };
629  
630  /*
631   * Don't even think about trying any of these in real life!!!
632   * The names includes "busted", and they really means it!
633   * The only purpose of these functions is to provide a buggy RCU
634   * implementation to make sure that rcutorture correctly emits
635   * buggy-RCU error messages.
636   */
rcu_busted_torture_deferred_free(struct rcu_torture * p)637  static void rcu_busted_torture_deferred_free(struct rcu_torture *p)
638  {
639  	/* This is a deliberate bug for testing purposes only! */
640  	rcu_torture_cb(&p->rtort_rcu);
641  }
642  
synchronize_rcu_busted(void)643  static void synchronize_rcu_busted(void)
644  {
645  	/* This is a deliberate bug for testing purposes only! */
646  }
647  
648  static void
call_rcu_busted(struct rcu_head * head,rcu_callback_t func)649  call_rcu_busted(struct rcu_head *head, rcu_callback_t func)
650  {
651  	/* This is a deliberate bug for testing purposes only! */
652  	func(head);
653  }
654  
655  static struct rcu_torture_ops rcu_busted_ops = {
656  	.ttype		= INVALID_RCU_FLAVOR,
657  	.init		= rcu_sync_torture_init,
658  	.readlock	= rcu_torture_read_lock,
659  	.read_delay	= rcu_read_delay,  /* just reuse rcu's version. */
660  	.readunlock	= rcu_torture_read_unlock,
661  	.readlock_held	= torture_readlock_not_held,
662  	.get_gp_seq	= rcu_no_completed,
663  	.deferred_free	= rcu_busted_torture_deferred_free,
664  	.sync		= synchronize_rcu_busted,
665  	.exp_sync	= synchronize_rcu_busted,
666  	.call		= call_rcu_busted,
667  	.gather_gp_seqs	= rcutorture_gather_gp_seqs,
668  	.format_gp_seqs	= rcutorture_format_gp_seqs,
669  	.irq_capable	= 1,
670  	.extendables	= RCUTORTURE_MAX_EXTEND,
671  	.name		= "busted"
672  };
673  
674  /*
675   * Definitions for srcu torture testing.
676   */
677  
678  DEFINE_STATIC_SRCU(srcu_ctl);
679  static struct srcu_struct srcu_ctld;
680  static struct srcu_struct *srcu_ctlp = &srcu_ctl;
681  static struct rcu_torture_ops srcud_ops;
682  
srcu_get_gp_data(int * flags,unsigned long * gp_seq)683  static void srcu_get_gp_data(int *flags, unsigned long *gp_seq)
684  {
685  	srcutorture_get_gp_data(srcu_ctlp, flags, gp_seq);
686  }
687  
srcu_torture_read_lock(void)688  static int srcu_torture_read_lock(void)
689  {
690  	int idx;
691  	struct srcu_ctr __percpu *scp;
692  	int ret = 0;
693  
694  	WARN_ON_ONCE(reader_flavor & ~SRCU_READ_FLAVOR_ALL);
695  
696  	if ((reader_flavor & SRCU_READ_FLAVOR_NORMAL) || !(reader_flavor & SRCU_READ_FLAVOR_ALL)) {
697  		idx = srcu_read_lock(srcu_ctlp);
698  		WARN_ON_ONCE(idx & ~0x1);
699  		ret += idx;
700  	}
701  	if (reader_flavor & SRCU_READ_FLAVOR_NMI) {
702  		idx = srcu_read_lock_nmisafe(srcu_ctlp);
703  		WARN_ON_ONCE(idx & ~0x1);
704  		ret += idx << 1;
705  	}
706  	if (reader_flavor & SRCU_READ_FLAVOR_LITE) {
707  		idx = srcu_read_lock_lite(srcu_ctlp);
708  		WARN_ON_ONCE(idx & ~0x1);
709  		ret += idx << 2;
710  	}
711  	if (reader_flavor & SRCU_READ_FLAVOR_FAST) {
712  		scp = srcu_read_lock_fast(srcu_ctlp);
713  		idx = __srcu_ptr_to_ctr(srcu_ctlp, scp);
714  		WARN_ON_ONCE(idx & ~0x1);
715  		ret += idx << 3;
716  	}
717  	return ret;
718  }
719  
720  static void
srcu_read_delay(struct torture_random_state * rrsp,struct rt_read_seg * rtrsp)721  srcu_read_delay(struct torture_random_state *rrsp, struct rt_read_seg *rtrsp)
722  {
723  	long delay;
724  	const long uspertick = 1000000 / HZ;
725  	const long longdelay = 10;
726  
727  	/* We want there to be long-running readers, but not all the time. */
728  
729  	delay = torture_random(rrsp) %
730  		(nrealreaders * 2 * longdelay * uspertick);
731  	if (!delay && in_task()) {
732  		schedule_timeout_interruptible(longdelay);
733  		rtrsp->rt_delay_jiffies = longdelay;
734  	} else {
735  		rcu_read_delay(rrsp, rtrsp);
736  	}
737  }
738  
srcu_torture_read_unlock(int idx)739  static void srcu_torture_read_unlock(int idx)
740  {
741  	WARN_ON_ONCE((reader_flavor && (idx & ~reader_flavor)) || (!reader_flavor && (idx & ~0x1)));
742  	if (reader_flavor & SRCU_READ_FLAVOR_FAST)
743  		srcu_read_unlock_fast(srcu_ctlp, __srcu_ctr_to_ptr(srcu_ctlp, (idx & 0x8) >> 3));
744  	if (reader_flavor & SRCU_READ_FLAVOR_LITE)
745  		srcu_read_unlock_lite(srcu_ctlp, (idx & 0x4) >> 2);
746  	if (reader_flavor & SRCU_READ_FLAVOR_NMI)
747  		srcu_read_unlock_nmisafe(srcu_ctlp, (idx & 0x2) >> 1);
748  	if ((reader_flavor & SRCU_READ_FLAVOR_NORMAL) || !(reader_flavor & SRCU_READ_FLAVOR_ALL))
749  		srcu_read_unlock(srcu_ctlp, idx & 0x1);
750  }
751  
torture_srcu_read_lock_held(void)752  static int torture_srcu_read_lock_held(void)
753  {
754  	return srcu_read_lock_held(srcu_ctlp);
755  }
756  
srcu_torture_completed(void)757  static unsigned long srcu_torture_completed(void)
758  {
759  	return srcu_batches_completed(srcu_ctlp);
760  }
761  
srcu_torture_deferred_free(struct rcu_torture * rp)762  static void srcu_torture_deferred_free(struct rcu_torture *rp)
763  {
764  	call_srcu(srcu_ctlp, &rp->rtort_rcu, rcu_torture_cb);
765  }
766  
srcu_torture_synchronize(void)767  static void srcu_torture_synchronize(void)
768  {
769  	synchronize_srcu(srcu_ctlp);
770  }
771  
srcu_torture_get_gp_state(void)772  static unsigned long srcu_torture_get_gp_state(void)
773  {
774  	return get_state_synchronize_srcu(srcu_ctlp);
775  }
776  
srcu_torture_start_gp_poll(void)777  static unsigned long srcu_torture_start_gp_poll(void)
778  {
779  	return start_poll_synchronize_srcu(srcu_ctlp);
780  }
781  
srcu_torture_poll_gp_state(unsigned long oldstate)782  static bool srcu_torture_poll_gp_state(unsigned long oldstate)
783  {
784  	return poll_state_synchronize_srcu(srcu_ctlp, oldstate);
785  }
786  
srcu_torture_call(struct rcu_head * head,rcu_callback_t func)787  static void srcu_torture_call(struct rcu_head *head,
788  			      rcu_callback_t func)
789  {
790  	call_srcu(srcu_ctlp, head, func);
791  }
792  
srcu_torture_barrier(void)793  static void srcu_torture_barrier(void)
794  {
795  	srcu_barrier(srcu_ctlp);
796  }
797  
srcu_torture_stats(void)798  static void srcu_torture_stats(void)
799  {
800  	srcu_torture_stats_print(srcu_ctlp, torture_type, TORTURE_FLAG);
801  }
802  
srcu_torture_synchronize_expedited(void)803  static void srcu_torture_synchronize_expedited(void)
804  {
805  	synchronize_srcu_expedited(srcu_ctlp);
806  }
807  
808  static struct rcu_torture_ops srcu_ops = {
809  	.ttype		= SRCU_FLAVOR,
810  	.init		= rcu_sync_torture_init,
811  	.readlock	= srcu_torture_read_lock,
812  	.read_delay	= srcu_read_delay,
813  	.readunlock	= srcu_torture_read_unlock,
814  	.readlock_held	= torture_srcu_read_lock_held,
815  	.get_gp_seq	= srcu_torture_completed,
816  	.gp_diff	= rcu_seq_diff,
817  	.deferred_free	= srcu_torture_deferred_free,
818  	.sync		= srcu_torture_synchronize,
819  	.exp_sync	= srcu_torture_synchronize_expedited,
820  	.same_gp_state	= same_state_synchronize_srcu,
821  	.get_comp_state = get_completed_synchronize_srcu,
822  	.get_gp_state	= srcu_torture_get_gp_state,
823  	.start_gp_poll	= srcu_torture_start_gp_poll,
824  	.poll_gp_state	= srcu_torture_poll_gp_state,
825  	.poll_active	= NUM_ACTIVE_SRCU_POLL_OLDSTATE,
826  	.call		= srcu_torture_call,
827  	.cb_barrier	= srcu_torture_barrier,
828  	.stats		= srcu_torture_stats,
829  	.get_gp_data	= srcu_get_gp_data,
830  	.cbflood_max	= 50000,
831  	.irq_capable	= 1,
832  	.no_pi_lock	= IS_ENABLED(CONFIG_TINY_SRCU),
833  	.debug_objects	= 1,
834  	.name		= "srcu"
835  };
836  
srcu_torture_init(void)837  static void srcu_torture_init(void)
838  {
839  	rcu_sync_torture_init();
840  	WARN_ON(init_srcu_struct(&srcu_ctld));
841  	srcu_ctlp = &srcu_ctld;
842  }
843  
srcu_torture_cleanup(void)844  static void srcu_torture_cleanup(void)
845  {
846  	cleanup_srcu_struct(&srcu_ctld);
847  	srcu_ctlp = &srcu_ctl; /* In case of a later rcutorture run. */
848  }
849  
850  /* As above, but dynamically allocated. */
851  static struct rcu_torture_ops srcud_ops = {
852  	.ttype		= SRCU_FLAVOR,
853  	.init		= srcu_torture_init,
854  	.cleanup	= srcu_torture_cleanup,
855  	.readlock	= srcu_torture_read_lock,
856  	.read_delay	= srcu_read_delay,
857  	.readunlock	= srcu_torture_read_unlock,
858  	.readlock_held	= torture_srcu_read_lock_held,
859  	.get_gp_seq	= srcu_torture_completed,
860  	.gp_diff	= rcu_seq_diff,
861  	.deferred_free	= srcu_torture_deferred_free,
862  	.sync		= srcu_torture_synchronize,
863  	.exp_sync	= srcu_torture_synchronize_expedited,
864  	.same_gp_state	= same_state_synchronize_srcu,
865  	.get_comp_state = get_completed_synchronize_srcu,
866  	.get_gp_state	= srcu_torture_get_gp_state,
867  	.start_gp_poll	= srcu_torture_start_gp_poll,
868  	.poll_gp_state	= srcu_torture_poll_gp_state,
869  	.poll_active	= NUM_ACTIVE_SRCU_POLL_OLDSTATE,
870  	.call		= srcu_torture_call,
871  	.cb_barrier	= srcu_torture_barrier,
872  	.stats		= srcu_torture_stats,
873  	.get_gp_data	= srcu_get_gp_data,
874  	.cbflood_max	= 50000,
875  	.irq_capable	= 1,
876  	.no_pi_lock	= IS_ENABLED(CONFIG_TINY_SRCU),
877  	.debug_objects	= 1,
878  	.name		= "srcud"
879  };
880  
881  /* As above, but broken due to inappropriate reader extension. */
882  static struct rcu_torture_ops busted_srcud_ops = {
883  	.ttype		= SRCU_FLAVOR,
884  	.init		= srcu_torture_init,
885  	.cleanup	= srcu_torture_cleanup,
886  	.readlock	= srcu_torture_read_lock,
887  	.read_delay	= rcu_read_delay,
888  	.readunlock	= srcu_torture_read_unlock,
889  	.readlock_held	= torture_srcu_read_lock_held,
890  	.get_gp_seq	= srcu_torture_completed,
891  	.deferred_free	= srcu_torture_deferred_free,
892  	.sync		= srcu_torture_synchronize,
893  	.exp_sync	= srcu_torture_synchronize_expedited,
894  	.call		= srcu_torture_call,
895  	.cb_barrier	= srcu_torture_barrier,
896  	.stats		= srcu_torture_stats,
897  	.irq_capable	= 1,
898  	.no_pi_lock	= IS_ENABLED(CONFIG_TINY_SRCU),
899  	.extendables	= RCUTORTURE_MAX_EXTEND,
900  	.name		= "busted_srcud"
901  };
902  
903  /*
904   * Definitions for trivial CONFIG_PREEMPT=n-only torture testing.
905   * This implementation does not necessarily work well with CPU hotplug.
906   */
907  
synchronize_rcu_trivial(void)908  static void synchronize_rcu_trivial(void)
909  {
910  	int cpu;
911  
912  	for_each_online_cpu(cpu) {
913  		torture_sched_setaffinity(current->pid, cpumask_of(cpu), true);
914  		WARN_ON_ONCE(raw_smp_processor_id() != cpu);
915  	}
916  }
917  
rcu_torture_read_lock_trivial(void)918  static int rcu_torture_read_lock_trivial(void)
919  {
920  	preempt_disable();
921  	return 0;
922  }
923  
rcu_torture_read_unlock_trivial(int idx)924  static void rcu_torture_read_unlock_trivial(int idx)
925  {
926  	preempt_enable();
927  }
928  
929  static struct rcu_torture_ops trivial_ops = {
930  	.ttype		= RCU_TRIVIAL_FLAVOR,
931  	.init		= rcu_sync_torture_init,
932  	.readlock	= rcu_torture_read_lock_trivial,
933  	.read_delay	= rcu_read_delay,  /* just reuse rcu's version. */
934  	.readunlock	= rcu_torture_read_unlock_trivial,
935  	.readlock_held	= torture_readlock_not_held,
936  	.get_gp_seq	= rcu_no_completed,
937  	.sync		= synchronize_rcu_trivial,
938  	.exp_sync	= synchronize_rcu_trivial,
939  	.irq_capable	= 1,
940  	.name		= "trivial"
941  };
942  
943  #ifdef CONFIG_TASKS_RCU
944  
945  /*
946   * Definitions for RCU-tasks torture testing.
947   */
948  
tasks_torture_read_lock(void)949  static int tasks_torture_read_lock(void)
950  {
951  	return 0;
952  }
953  
tasks_torture_read_unlock(int idx)954  static void tasks_torture_read_unlock(int idx)
955  {
956  }
957  
rcu_tasks_torture_deferred_free(struct rcu_torture * p)958  static void rcu_tasks_torture_deferred_free(struct rcu_torture *p)
959  {
960  	call_rcu_tasks(&p->rtort_rcu, rcu_torture_cb);
961  }
962  
synchronize_rcu_mult_test(void)963  static void synchronize_rcu_mult_test(void)
964  {
965  	synchronize_rcu_mult(call_rcu_tasks, call_rcu_hurry);
966  }
967  
968  static struct rcu_torture_ops tasks_ops = {
969  	.ttype		= RCU_TASKS_FLAVOR,
970  	.init		= rcu_sync_torture_init,
971  	.readlock	= tasks_torture_read_lock,
972  	.read_delay	= rcu_read_delay,  /* just reuse rcu's version. */
973  	.readunlock	= tasks_torture_read_unlock,
974  	.get_gp_seq	= rcu_no_completed,
975  	.deferred_free	= rcu_tasks_torture_deferred_free,
976  	.sync		= synchronize_rcu_tasks,
977  	.exp_sync	= synchronize_rcu_mult_test,
978  	.call		= call_rcu_tasks,
979  	.cb_barrier	= rcu_barrier_tasks,
980  	.gp_kthread_dbg	= show_rcu_tasks_classic_gp_kthread,
981  	.get_gp_data	= rcu_tasks_get_gp_data,
982  	.irq_capable	= 1,
983  	.slow_gps	= 1,
984  	.name		= "tasks"
985  };
986  
987  #define TASKS_OPS &tasks_ops,
988  
989  #else // #ifdef CONFIG_TASKS_RCU
990  
991  #define TASKS_OPS
992  
993  #endif // #else #ifdef CONFIG_TASKS_RCU
994  
995  
996  #ifdef CONFIG_TASKS_RUDE_RCU
997  
998  /*
999   * Definitions for rude RCU-tasks torture testing.
1000   */
1001  
1002  static struct rcu_torture_ops tasks_rude_ops = {
1003  	.ttype		= RCU_TASKS_RUDE_FLAVOR,
1004  	.init		= rcu_sync_torture_init,
1005  	.readlock	= rcu_torture_read_lock_trivial,
1006  	.read_delay	= rcu_read_delay,  /* just reuse rcu's version. */
1007  	.readunlock	= rcu_torture_read_unlock_trivial,
1008  	.get_gp_seq	= rcu_no_completed,
1009  	.sync		= synchronize_rcu_tasks_rude,
1010  	.exp_sync	= synchronize_rcu_tasks_rude,
1011  	.gp_kthread_dbg	= show_rcu_tasks_rude_gp_kthread,
1012  	.get_gp_data	= rcu_tasks_rude_get_gp_data,
1013  	.cbflood_max	= 50000,
1014  	.irq_capable	= 1,
1015  	.name		= "tasks-rude"
1016  };
1017  
1018  #define TASKS_RUDE_OPS &tasks_rude_ops,
1019  
1020  #else // #ifdef CONFIG_TASKS_RUDE_RCU
1021  
1022  #define TASKS_RUDE_OPS
1023  
1024  #endif // #else #ifdef CONFIG_TASKS_RUDE_RCU
1025  
1026  
1027  #ifdef CONFIG_TASKS_TRACE_RCU
1028  
1029  /*
1030   * Definitions for tracing RCU-tasks torture testing.
1031   */
1032  
tasks_tracing_torture_read_lock(void)1033  static int tasks_tracing_torture_read_lock(void)
1034  {
1035  	rcu_read_lock_trace();
1036  	return 0;
1037  }
1038  
tasks_tracing_torture_read_unlock(int idx)1039  static void tasks_tracing_torture_read_unlock(int idx)
1040  {
1041  	rcu_read_unlock_trace();
1042  }
1043  
rcu_tasks_tracing_torture_deferred_free(struct rcu_torture * p)1044  static void rcu_tasks_tracing_torture_deferred_free(struct rcu_torture *p)
1045  {
1046  	call_rcu_tasks_trace(&p->rtort_rcu, rcu_torture_cb);
1047  }
1048  
1049  static struct rcu_torture_ops tasks_tracing_ops = {
1050  	.ttype		= RCU_TASKS_TRACING_FLAVOR,
1051  	.init		= rcu_sync_torture_init,
1052  	.readlock	= tasks_tracing_torture_read_lock,
1053  	.read_delay	= srcu_read_delay,  /* just reuse srcu's version. */
1054  	.readunlock	= tasks_tracing_torture_read_unlock,
1055  	.readlock_held	= rcu_read_lock_trace_held,
1056  	.get_gp_seq	= rcu_no_completed,
1057  	.deferred_free	= rcu_tasks_tracing_torture_deferred_free,
1058  	.sync		= synchronize_rcu_tasks_trace,
1059  	.exp_sync	= synchronize_rcu_tasks_trace,
1060  	.call		= call_rcu_tasks_trace,
1061  	.cb_barrier	= rcu_barrier_tasks_trace,
1062  	.gp_kthread_dbg	= show_rcu_tasks_trace_gp_kthread,
1063  	.get_gp_data    = rcu_tasks_trace_get_gp_data,
1064  	.cbflood_max	= 50000,
1065  	.irq_capable	= 1,
1066  	.slow_gps	= 1,
1067  	.name		= "tasks-tracing"
1068  };
1069  
1070  #define TASKS_TRACING_OPS &tasks_tracing_ops,
1071  
1072  #else // #ifdef CONFIG_TASKS_TRACE_RCU
1073  
1074  #define TASKS_TRACING_OPS
1075  
1076  #endif // #else #ifdef CONFIG_TASKS_TRACE_RCU
1077  
1078  
rcutorture_seq_diff(unsigned long new,unsigned long old)1079  static unsigned long rcutorture_seq_diff(unsigned long new, unsigned long old)
1080  {
1081  	if (!cur_ops->gp_diff)
1082  		return new - old;
1083  	return cur_ops->gp_diff(new, old);
1084  }
1085  
1086  /*
1087   * RCU torture priority-boost testing.  Runs one real-time thread per
1088   * CPU for moderate bursts, repeatedly starting grace periods and waiting
1089   * for them to complete.  If a given grace period takes too long, we assume
1090   * that priority inversion has occurred.
1091   */
1092  
1093  static int old_rt_runtime = -1;
1094  
rcu_torture_disable_rt_throttle(void)1095  static void rcu_torture_disable_rt_throttle(void)
1096  {
1097  	/*
1098  	 * Disable RT throttling so that rcutorture's boost threads don't get
1099  	 * throttled. Only possible if rcutorture is built-in otherwise the
1100  	 * user should manually do this by setting the sched_rt_period_us and
1101  	 * sched_rt_runtime sysctls.
1102  	 */
1103  	if (!IS_BUILTIN(CONFIG_RCU_TORTURE_TEST) || old_rt_runtime != -1)
1104  		return;
1105  
1106  	old_rt_runtime = sysctl_sched_rt_runtime;
1107  	sysctl_sched_rt_runtime = -1;
1108  }
1109  
rcu_torture_enable_rt_throttle(void)1110  static void rcu_torture_enable_rt_throttle(void)
1111  {
1112  	if (!IS_BUILTIN(CONFIG_RCU_TORTURE_TEST) || old_rt_runtime == -1)
1113  		return;
1114  
1115  	sysctl_sched_rt_runtime = old_rt_runtime;
1116  	old_rt_runtime = -1;
1117  }
1118  
rcu_torture_boost_failed(unsigned long gp_state,unsigned long * start)1119  static bool rcu_torture_boost_failed(unsigned long gp_state, unsigned long *start)
1120  {
1121  	int cpu;
1122  	static int dbg_done;
1123  	unsigned long end = jiffies;
1124  	bool gp_done;
1125  	unsigned long j;
1126  	static unsigned long last_persist;
1127  	unsigned long lp;
1128  	unsigned long mininterval = test_boost_duration * HZ - HZ / 2;
1129  
1130  	if (end - *start > mininterval) {
1131  		// Recheck after checking time to avoid false positives.
1132  		smp_mb(); // Time check before grace-period check.
1133  		if (cur_ops->poll_gp_state(gp_state))
1134  			return false; // passed, though perhaps just barely
1135  		if (cur_ops->check_boost_failed && !cur_ops->check_boost_failed(gp_state, &cpu)) {
1136  			// At most one persisted message per boost test.
1137  			j = jiffies;
1138  			lp = READ_ONCE(last_persist);
1139  			if (time_after(j, lp + mininterval) &&
1140  			    cmpxchg(&last_persist, lp, j) == lp) {
1141  				if (cpu < 0)
1142  					pr_info("Boost inversion persisted: QS from all CPUs\n");
1143  				else
1144  					pr_info("Boost inversion persisted: No QS from CPU %d\n", cpu);
1145  			}
1146  			return false; // passed on a technicality
1147  		}
1148  		VERBOSE_TOROUT_STRING("rcu_torture_boost boosting failed");
1149  		n_rcu_torture_boost_failure++;
1150  		if (!xchg(&dbg_done, 1) && cur_ops->gp_kthread_dbg) {
1151  			pr_info("Boost inversion thread ->rt_priority %u gp_state %lu jiffies %lu\n",
1152  				current->rt_priority, gp_state, end - *start);
1153  			cur_ops->gp_kthread_dbg();
1154  			// Recheck after print to flag grace period ending during splat.
1155  			gp_done = cur_ops->poll_gp_state(gp_state);
1156  			pr_info("Boost inversion: GP %lu %s.\n", gp_state,
1157  				gp_done ? "ended already" : "still pending");
1158  
1159  		}
1160  
1161  		return true; // failed
1162  	} else if (cur_ops->check_boost_failed && !cur_ops->check_boost_failed(gp_state, NULL)) {
1163  		*start = jiffies;
1164  	}
1165  
1166  	return false; // passed
1167  }
1168  
rcu_torture_boost(void * arg)1169  static int rcu_torture_boost(void *arg)
1170  {
1171  	unsigned long endtime;
1172  	unsigned long gp_state;
1173  	unsigned long gp_state_time;
1174  	unsigned long oldstarttime;
1175  	unsigned long booststarttime = get_torture_init_jiffies() + test_boost_holdoff * HZ;
1176  
1177  	if (test_boost_holdoff <= 0 || time_after(jiffies, booststarttime)) {
1178  		VERBOSE_TOROUT_STRING("rcu_torture_boost started");
1179  	} else {
1180  		VERBOSE_TOROUT_STRING("rcu_torture_boost started holdoff period");
1181  		while (time_before(jiffies, booststarttime)) {
1182  			schedule_timeout_idle(HZ);
1183  			if (kthread_should_stop())
1184  				goto cleanup;
1185  		}
1186  		VERBOSE_TOROUT_STRING("rcu_torture_boost finished holdoff period");
1187  	}
1188  
1189  	/* Set real-time priority. */
1190  	sched_set_fifo_low(current);
1191  
1192  	/* Each pass through the following loop does one boost-test cycle. */
1193  	do {
1194  		bool failed = false; // Test failed already in this test interval
1195  		bool gp_initiated = false;
1196  
1197  		if (kthread_should_stop())
1198  			goto checkwait;
1199  
1200  		/* Wait for the next test interval. */
1201  		oldstarttime = READ_ONCE(boost_starttime);
1202  		while (time_before(jiffies, oldstarttime)) {
1203  			schedule_timeout_interruptible(oldstarttime - jiffies);
1204  			if (stutter_wait("rcu_torture_boost"))
1205  				sched_set_fifo_low(current);
1206  			if (torture_must_stop())
1207  				goto checkwait;
1208  		}
1209  
1210  		// Do one boost-test interval.
1211  		endtime = oldstarttime + test_boost_duration * HZ;
1212  		while (time_before(jiffies, endtime)) {
1213  			// Has current GP gone too long?
1214  			if (gp_initiated && !failed && !cur_ops->poll_gp_state(gp_state))
1215  				failed = rcu_torture_boost_failed(gp_state, &gp_state_time);
1216  			// If we don't have a grace period in flight, start one.
1217  			if (!gp_initiated || cur_ops->poll_gp_state(gp_state)) {
1218  				gp_state = cur_ops->start_gp_poll();
1219  				gp_initiated = true;
1220  				gp_state_time = jiffies;
1221  			}
1222  			if (stutter_wait("rcu_torture_boost")) {
1223  				sched_set_fifo_low(current);
1224  				// If the grace period already ended,
1225  				// we don't know when that happened, so
1226  				// start over.
1227  				if (cur_ops->poll_gp_state(gp_state))
1228  					gp_initiated = false;
1229  			}
1230  			if (torture_must_stop())
1231  				goto checkwait;
1232  		}
1233  
1234  		// In case the grace period extended beyond the end of the loop.
1235  		if (gp_initiated && !failed && !cur_ops->poll_gp_state(gp_state))
1236  			rcu_torture_boost_failed(gp_state, &gp_state_time);
1237  
1238  		/*
1239  		 * Set the start time of the next test interval.
1240  		 * Yes, this is vulnerable to long delays, but such
1241  		 * delays simply cause a false negative for the next
1242  		 * interval.  Besides, we are running at RT priority,
1243  		 * so delays should be relatively rare.
1244  		 */
1245  		while (oldstarttime == READ_ONCE(boost_starttime) && !kthread_should_stop()) {
1246  			if (mutex_trylock(&boost_mutex)) {
1247  				if (oldstarttime == boost_starttime) {
1248  					WRITE_ONCE(boost_starttime,
1249  						   jiffies + test_boost_interval * HZ);
1250  					n_rcu_torture_boosts++;
1251  				}
1252  				mutex_unlock(&boost_mutex);
1253  				break;
1254  			}
1255  			schedule_timeout_uninterruptible(HZ / 20);
1256  		}
1257  
1258  		/* Go do the stutter. */
1259  checkwait:	if (stutter_wait("rcu_torture_boost"))
1260  			sched_set_fifo_low(current);
1261  	} while (!torture_must_stop());
1262  
1263  cleanup:
1264  	/* Clean up and exit. */
1265  	while (!kthread_should_stop()) {
1266  		torture_shutdown_absorb("rcu_torture_boost");
1267  		schedule_timeout_uninterruptible(HZ / 20);
1268  	}
1269  	torture_kthread_stopping("rcu_torture_boost");
1270  	return 0;
1271  }
1272  
1273  /*
1274   * RCU torture force-quiescent-state kthread.  Repeatedly induces
1275   * bursts of calls to force_quiescent_state(), increasing the probability
1276   * of occurrence of some important types of race conditions.
1277   */
1278  static int
rcu_torture_fqs(void * arg)1279  rcu_torture_fqs(void *arg)
1280  {
1281  	unsigned long fqs_resume_time;
1282  	int fqs_burst_remaining;
1283  	int oldnice = task_nice(current);
1284  
1285  	VERBOSE_TOROUT_STRING("rcu_torture_fqs task started");
1286  	do {
1287  		fqs_resume_time = jiffies + fqs_stutter * HZ;
1288  		while (time_before(jiffies, fqs_resume_time) &&
1289  		       !kthread_should_stop()) {
1290  			schedule_timeout_interruptible(HZ / 20);
1291  		}
1292  		fqs_burst_remaining = fqs_duration;
1293  		while (fqs_burst_remaining > 0 &&
1294  		       !kthread_should_stop()) {
1295  			cur_ops->fqs();
1296  			udelay(fqs_holdoff);
1297  			fqs_burst_remaining -= fqs_holdoff;
1298  		}
1299  		if (stutter_wait("rcu_torture_fqs"))
1300  			sched_set_normal(current, oldnice);
1301  	} while (!torture_must_stop());
1302  	torture_kthread_stopping("rcu_torture_fqs");
1303  	return 0;
1304  }
1305  
1306  // Used by writers to randomly choose from the available grace-period primitives.
1307  static int synctype[ARRAY_SIZE(rcu_torture_writer_state_names)] = { };
1308  static int nsynctypes;
1309  
1310  /*
1311   * Determine which grace-period primitives are available.
1312   */
rcu_torture_write_types(void)1313  static void rcu_torture_write_types(void)
1314  {
1315  	bool gp_cond1 = gp_cond, gp_cond_exp1 = gp_cond_exp, gp_cond_full1 = gp_cond_full;
1316  	bool gp_cond_exp_full1 = gp_cond_exp_full, gp_exp1 = gp_exp, gp_poll_exp1 = gp_poll_exp;
1317  	bool gp_poll_exp_full1 = gp_poll_exp_full, gp_normal1 = gp_normal, gp_poll1 = gp_poll;
1318  	bool gp_poll_full1 = gp_poll_full, gp_sync1 = gp_sync;
1319  
1320  	/* Initialize synctype[] array.  If none set, take default. */
1321  	if (!gp_cond1 &&
1322  	    !gp_cond_exp1 &&
1323  	    !gp_cond_full1 &&
1324  	    !gp_cond_exp_full1 &&
1325  	    !gp_exp1 &&
1326  	    !gp_poll_exp1 &&
1327  	    !gp_poll_exp_full1 &&
1328  	    !gp_normal1 &&
1329  	    !gp_poll1 &&
1330  	    !gp_poll_full1 &&
1331  	    !gp_sync1) {
1332  		gp_cond1 = true;
1333  		gp_cond_exp1 = true;
1334  		gp_cond_full1 = true;
1335  		gp_cond_exp_full1 = true;
1336  		gp_exp1 = true;
1337  		gp_poll_exp1 = true;
1338  		gp_poll_exp_full1 = true;
1339  		gp_normal1 = true;
1340  		gp_poll1 = true;
1341  		gp_poll_full1 = true;
1342  		gp_sync1 = true;
1343  	}
1344  	if (gp_cond1 && cur_ops->get_gp_state && cur_ops->cond_sync) {
1345  		synctype[nsynctypes++] = RTWS_COND_GET;
1346  		pr_info("%s: Testing conditional GPs.\n", __func__);
1347  	} else if (gp_cond && (!cur_ops->get_gp_state || !cur_ops->cond_sync)) {
1348  		pr_alert("%s: gp_cond without primitives.\n", __func__);
1349  	}
1350  	if (gp_cond_exp1 && cur_ops->get_gp_state_exp && cur_ops->cond_sync_exp) {
1351  		synctype[nsynctypes++] = RTWS_COND_GET_EXP;
1352  		pr_info("%s: Testing conditional expedited GPs.\n", __func__);
1353  	} else if (gp_cond_exp && (!cur_ops->get_gp_state_exp || !cur_ops->cond_sync_exp)) {
1354  		pr_alert("%s: gp_cond_exp without primitives.\n", __func__);
1355  	}
1356  	if (gp_cond_full1 && cur_ops->get_gp_state && cur_ops->cond_sync_full) {
1357  		synctype[nsynctypes++] = RTWS_COND_GET_FULL;
1358  		pr_info("%s: Testing conditional full-state GPs.\n", __func__);
1359  	} else if (gp_cond_full && (!cur_ops->get_gp_state || !cur_ops->cond_sync_full)) {
1360  		pr_alert("%s: gp_cond_full without primitives.\n", __func__);
1361  	}
1362  	if (gp_cond_exp_full1 && cur_ops->get_gp_state_exp && cur_ops->cond_sync_exp_full) {
1363  		synctype[nsynctypes++] = RTWS_COND_GET_EXP_FULL;
1364  		pr_info("%s: Testing conditional full-state expedited GPs.\n", __func__);
1365  	} else if (gp_cond_exp_full &&
1366  		   (!cur_ops->get_gp_state_exp || !cur_ops->cond_sync_exp_full)) {
1367  		pr_alert("%s: gp_cond_exp_full without primitives.\n", __func__);
1368  	}
1369  	if (gp_exp1 && cur_ops->exp_sync) {
1370  		synctype[nsynctypes++] = RTWS_EXP_SYNC;
1371  		pr_info("%s: Testing expedited GPs.\n", __func__);
1372  	} else if (gp_exp && !cur_ops->exp_sync) {
1373  		pr_alert("%s: gp_exp without primitives.\n", __func__);
1374  	}
1375  	if (gp_normal1 && cur_ops->deferred_free) {
1376  		synctype[nsynctypes++] = RTWS_DEF_FREE;
1377  		pr_info("%s: Testing asynchronous GPs.\n", __func__);
1378  	} else if (gp_normal && !cur_ops->deferred_free) {
1379  		pr_alert("%s: gp_normal without primitives.\n", __func__);
1380  	}
1381  	if (gp_poll1 && cur_ops->get_comp_state && cur_ops->same_gp_state &&
1382  	    cur_ops->start_gp_poll && cur_ops->poll_gp_state) {
1383  		synctype[nsynctypes++] = RTWS_POLL_GET;
1384  		pr_info("%s: Testing polling GPs.\n", __func__);
1385  	} else if (gp_poll && (!cur_ops->start_gp_poll || !cur_ops->poll_gp_state)) {
1386  		pr_alert("%s: gp_poll without primitives.\n", __func__);
1387  	}
1388  	if (gp_poll_full1 && cur_ops->get_comp_state_full && cur_ops->same_gp_state_full
1389  	    && cur_ops->start_gp_poll_full && cur_ops->poll_gp_state_full) {
1390  		synctype[nsynctypes++] = RTWS_POLL_GET_FULL;
1391  		pr_info("%s: Testing polling full-state GPs.\n", __func__);
1392  	} else if (gp_poll_full && (!cur_ops->start_gp_poll_full || !cur_ops->poll_gp_state_full)) {
1393  		pr_alert("%s: gp_poll_full without primitives.\n", __func__);
1394  	}
1395  	if (gp_poll_exp1 && cur_ops->start_gp_poll_exp && cur_ops->poll_gp_state_exp) {
1396  		synctype[nsynctypes++] = RTWS_POLL_GET_EXP;
1397  		pr_info("%s: Testing polling expedited GPs.\n", __func__);
1398  	} else if (gp_poll_exp && (!cur_ops->start_gp_poll_exp || !cur_ops->poll_gp_state_exp)) {
1399  		pr_alert("%s: gp_poll_exp without primitives.\n", __func__);
1400  	}
1401  	if (gp_poll_exp_full1 && cur_ops->start_gp_poll_exp_full && cur_ops->poll_gp_state_full) {
1402  		synctype[nsynctypes++] = RTWS_POLL_GET_EXP_FULL;
1403  		pr_info("%s: Testing polling full-state expedited GPs.\n", __func__);
1404  	} else if (gp_poll_exp_full &&
1405  		   (!cur_ops->start_gp_poll_exp_full || !cur_ops->poll_gp_state_full)) {
1406  		pr_alert("%s: gp_poll_exp_full without primitives.\n", __func__);
1407  	}
1408  	if (gp_sync1 && cur_ops->sync) {
1409  		synctype[nsynctypes++] = RTWS_SYNC;
1410  		pr_info("%s: Testing normal GPs.\n", __func__);
1411  	} else if (gp_sync && !cur_ops->sync) {
1412  		pr_alert("%s: gp_sync without primitives.\n", __func__);
1413  	}
1414  	pr_alert("%s: Testing %d update types.\n", __func__, nsynctypes);
1415  	pr_info("%s: gp_cond_wi %d gp_cond_wi_exp %d gp_poll_wi %d gp_poll_wi_exp %d\n", __func__, gp_cond_wi, gp_cond_wi_exp, gp_poll_wi, gp_poll_wi_exp);
1416  }
1417  
1418  /*
1419   * Do the specified rcu_torture_writer() synchronous grace period,
1420   * while also testing out the polled APIs.  Note well that the single-CPU
1421   * grace-period optimizations must be accounted for.
1422   */
do_rtws_sync(struct torture_random_state * trsp,void (* sync)(void))1423  static void do_rtws_sync(struct torture_random_state *trsp, void (*sync)(void))
1424  {
1425  	unsigned long cookie;
1426  	struct rcu_gp_oldstate cookie_full;
1427  	bool dopoll;
1428  	bool dopoll_full;
1429  	unsigned long r = torture_random(trsp);
1430  
1431  	dopoll = cur_ops->get_gp_state && cur_ops->poll_gp_state && !(r & 0x300);
1432  	dopoll_full = cur_ops->get_gp_state_full && cur_ops->poll_gp_state_full && !(r & 0xc00);
1433  	if (dopoll || dopoll_full)
1434  		cpus_read_lock();
1435  	if (dopoll)
1436  		cookie = cur_ops->get_gp_state();
1437  	if (dopoll_full)
1438  		cur_ops->get_gp_state_full(&cookie_full);
1439  	if (cur_ops->poll_need_2gp && cur_ops->poll_need_2gp(dopoll, dopoll_full))
1440  		sync();
1441  	sync();
1442  	WARN_ONCE(dopoll && !cur_ops->poll_gp_state(cookie),
1443  		  "%s: Cookie check 3 failed %pS() online %*pbl.",
1444  		  __func__, sync, cpumask_pr_args(cpu_online_mask));
1445  	WARN_ONCE(dopoll_full && !cur_ops->poll_gp_state_full(&cookie_full),
1446  		  "%s: Cookie check 4 failed %pS() online %*pbl",
1447  		  __func__, sync, cpumask_pr_args(cpu_online_mask));
1448  	if (dopoll || dopoll_full)
1449  		cpus_read_unlock();
1450  }
1451  
1452  /*
1453   * RCU torture writer kthread.  Repeatedly substitutes a new structure
1454   * for that pointed to by rcu_torture_current, freeing the old structure
1455   * after a series of grace periods (the "pipeline").
1456   */
1457  static int
rcu_torture_writer(void * arg)1458  rcu_torture_writer(void *arg)
1459  {
1460  	bool boot_ended;
1461  	bool can_expedite = !rcu_gp_is_expedited() && !rcu_gp_is_normal();
1462  	unsigned long cookie;
1463  	struct rcu_gp_oldstate cookie_full;
1464  	int expediting = 0;
1465  	unsigned long gp_snap;
1466  	unsigned long gp_snap1;
1467  	struct rcu_gp_oldstate gp_snap_full;
1468  	struct rcu_gp_oldstate gp_snap1_full;
1469  	int i;
1470  	int idx;
1471  	int oldnice = task_nice(current);
1472  	struct rcu_gp_oldstate *rgo = NULL;
1473  	int rgo_size = 0;
1474  	struct rcu_torture *rp;
1475  	struct rcu_torture *old_rp;
1476  	static DEFINE_TORTURE_RANDOM(rand);
1477  	unsigned long stallsdone = jiffies;
1478  	bool stutter_waited;
1479  	unsigned long *ulo = NULL;
1480  	int ulo_size = 0;
1481  
1482  	// If a new stall test is added, this must be adjusted.
1483  	if (stall_cpu_holdoff + stall_gp_kthread + stall_cpu)
1484  		stallsdone += (stall_cpu_holdoff + stall_gp_kthread + stall_cpu + 60) *
1485  			      HZ * (stall_cpu_repeat + 1);
1486  	VERBOSE_TOROUT_STRING("rcu_torture_writer task started");
1487  	if (!can_expedite)
1488  		pr_alert("%s" TORTURE_FLAG
1489  			 " GP expediting controlled from boot/sysfs for %s.\n",
1490  			 torture_type, cur_ops->name);
1491  	if (WARN_ONCE(nsynctypes == 0,
1492  		      "%s: No update-side primitives.\n", __func__)) {
1493  		/*
1494  		 * No updates primitives, so don't try updating.
1495  		 * The resulting test won't be testing much, hence the
1496  		 * above WARN_ONCE().
1497  		 */
1498  		rcu_torture_writer_state = RTWS_STOPPING;
1499  		torture_kthread_stopping("rcu_torture_writer");
1500  		return 0;
1501  	}
1502  	if (cur_ops->poll_active > 0) {
1503  		ulo = kzalloc(cur_ops->poll_active * sizeof(ulo[0]), GFP_KERNEL);
1504  		if (!WARN_ON(!ulo))
1505  			ulo_size = cur_ops->poll_active;
1506  	}
1507  	if (cur_ops->poll_active_full > 0) {
1508  		rgo = kzalloc(cur_ops->poll_active_full * sizeof(rgo[0]), GFP_KERNEL);
1509  		if (!WARN_ON(!rgo))
1510  			rgo_size = cur_ops->poll_active_full;
1511  	}
1512  
1513  	do {
1514  		rcu_torture_writer_state = RTWS_FIXED_DELAY;
1515  		torture_hrtimeout_us(500, 1000, &rand);
1516  		rp = rcu_torture_alloc();
1517  		if (rp == NULL)
1518  			continue;
1519  		rp->rtort_pipe_count = 0;
1520  		ASSERT_EXCLUSIVE_WRITER(rp->rtort_pipe_count);
1521  		rcu_torture_writer_state = RTWS_DELAY;
1522  		udelay(torture_random(&rand) & 0x3ff);
1523  		rcu_torture_writer_state = RTWS_REPLACE;
1524  		old_rp = rcu_dereference_check(rcu_torture_current,
1525  					       current == writer_task);
1526  		rp->rtort_mbtest = 1;
1527  		rcu_assign_pointer(rcu_torture_current, rp);
1528  		smp_wmb(); /* Mods to old_rp must follow rcu_assign_pointer() */
1529  		if (old_rp) {
1530  			i = old_rp->rtort_pipe_count;
1531  			if (i > RCU_TORTURE_PIPE_LEN)
1532  				i = RCU_TORTURE_PIPE_LEN;
1533  			atomic_inc(&rcu_torture_wcount[i]);
1534  			WRITE_ONCE(old_rp->rtort_pipe_count,
1535  				   old_rp->rtort_pipe_count + 1);
1536  			ASSERT_EXCLUSIVE_WRITER(old_rp->rtort_pipe_count);
1537  
1538  			// Make sure readers block polled grace periods.
1539  			if (cur_ops->get_gp_state && cur_ops->poll_gp_state) {
1540  				idx = cur_ops->readlock();
1541  				cookie = cur_ops->get_gp_state();
1542  				WARN_ONCE(cur_ops->poll_gp_state(cookie),
1543  					  "%s: Cookie check 1 failed %s(%d) %lu->%lu\n",
1544  					  __func__,
1545  					  rcu_torture_writer_state_getname(),
1546  					  rcu_torture_writer_state,
1547  					  cookie, cur_ops->get_gp_state());
1548  				if (cur_ops->get_comp_state) {
1549  					cookie = cur_ops->get_comp_state();
1550  					WARN_ON_ONCE(!cur_ops->poll_gp_state(cookie));
1551  				}
1552  				cur_ops->readunlock(idx);
1553  			}
1554  			if (cur_ops->get_gp_state_full && cur_ops->poll_gp_state_full) {
1555  				idx = cur_ops->readlock();
1556  				cur_ops->get_gp_state_full(&cookie_full);
1557  				WARN_ONCE(cur_ops->poll_gp_state_full(&cookie_full),
1558  					  "%s: Cookie check 5 failed %s(%d) online %*pbl\n",
1559  					  __func__,
1560  					  rcu_torture_writer_state_getname(),
1561  					  rcu_torture_writer_state,
1562  					  cpumask_pr_args(cpu_online_mask));
1563  				if (cur_ops->get_comp_state_full) {
1564  					cur_ops->get_comp_state_full(&cookie_full);
1565  					WARN_ON_ONCE(!cur_ops->poll_gp_state_full(&cookie_full));
1566  				}
1567  				cur_ops->readunlock(idx);
1568  			}
1569  			switch (synctype[torture_random(&rand) % nsynctypes]) {
1570  			case RTWS_DEF_FREE:
1571  				rcu_torture_writer_state = RTWS_DEF_FREE;
1572  				cur_ops->deferred_free(old_rp);
1573  				break;
1574  			case RTWS_EXP_SYNC:
1575  				rcu_torture_writer_state = RTWS_EXP_SYNC;
1576  				do_rtws_sync(&rand, cur_ops->exp_sync);
1577  				rcu_torture_pipe_update(old_rp);
1578  				break;
1579  			case RTWS_COND_GET:
1580  				rcu_torture_writer_state = RTWS_COND_GET;
1581  				gp_snap = cur_ops->get_gp_state();
1582  				torture_hrtimeout_us(torture_random(&rand) % gp_cond_wi,
1583  						     1000, &rand);
1584  				rcu_torture_writer_state = RTWS_COND_SYNC;
1585  				cur_ops->cond_sync(gp_snap);
1586  				rcu_torture_pipe_update(old_rp);
1587  				break;
1588  			case RTWS_COND_GET_EXP:
1589  				rcu_torture_writer_state = RTWS_COND_GET_EXP;
1590  				gp_snap = cur_ops->get_gp_state_exp();
1591  				torture_hrtimeout_us(torture_random(&rand) % gp_cond_wi_exp,
1592  						     1000, &rand);
1593  				rcu_torture_writer_state = RTWS_COND_SYNC_EXP;
1594  				cur_ops->cond_sync_exp(gp_snap);
1595  				rcu_torture_pipe_update(old_rp);
1596  				break;
1597  			case RTWS_COND_GET_FULL:
1598  				rcu_torture_writer_state = RTWS_COND_GET_FULL;
1599  				cur_ops->get_gp_state_full(&gp_snap_full);
1600  				torture_hrtimeout_us(torture_random(&rand) % gp_cond_wi,
1601  						     1000, &rand);
1602  				rcu_torture_writer_state = RTWS_COND_SYNC_FULL;
1603  				cur_ops->cond_sync_full(&gp_snap_full);
1604  				rcu_torture_pipe_update(old_rp);
1605  				break;
1606  			case RTWS_COND_GET_EXP_FULL:
1607  				rcu_torture_writer_state = RTWS_COND_GET_EXP_FULL;
1608  				cur_ops->get_gp_state_full(&gp_snap_full);
1609  				torture_hrtimeout_us(torture_random(&rand) % gp_cond_wi_exp,
1610  						     1000, &rand);
1611  				rcu_torture_writer_state = RTWS_COND_SYNC_EXP_FULL;
1612  				cur_ops->cond_sync_exp_full(&gp_snap_full);
1613  				rcu_torture_pipe_update(old_rp);
1614  				break;
1615  			case RTWS_POLL_GET:
1616  				rcu_torture_writer_state = RTWS_POLL_GET;
1617  				for (i = 0; i < ulo_size; i++)
1618  					ulo[i] = cur_ops->get_comp_state();
1619  				gp_snap = cur_ops->start_gp_poll();
1620  				rcu_torture_writer_state = RTWS_POLL_WAIT;
1621  				while (!cur_ops->poll_gp_state(gp_snap)) {
1622  					gp_snap1 = cur_ops->get_gp_state();
1623  					for (i = 0; i < ulo_size; i++)
1624  						if (cur_ops->poll_gp_state(ulo[i]) ||
1625  						    cur_ops->same_gp_state(ulo[i], gp_snap1)) {
1626  							ulo[i] = gp_snap1;
1627  							break;
1628  						}
1629  					WARN_ON_ONCE(ulo_size > 0 && i >= ulo_size);
1630  					torture_hrtimeout_us(torture_random(&rand) % gp_poll_wi,
1631  							     1000, &rand);
1632  				}
1633  				rcu_torture_pipe_update(old_rp);
1634  				break;
1635  			case RTWS_POLL_GET_FULL:
1636  				rcu_torture_writer_state = RTWS_POLL_GET_FULL;
1637  				for (i = 0; i < rgo_size; i++)
1638  					cur_ops->get_comp_state_full(&rgo[i]);
1639  				cur_ops->start_gp_poll_full(&gp_snap_full);
1640  				rcu_torture_writer_state = RTWS_POLL_WAIT_FULL;
1641  				while (!cur_ops->poll_gp_state_full(&gp_snap_full)) {
1642  					cur_ops->get_gp_state_full(&gp_snap1_full);
1643  					for (i = 0; i < rgo_size; i++)
1644  						if (cur_ops->poll_gp_state_full(&rgo[i]) ||
1645  						    cur_ops->same_gp_state_full(&rgo[i],
1646  										&gp_snap1_full)) {
1647  							rgo[i] = gp_snap1_full;
1648  							break;
1649  						}
1650  					WARN_ON_ONCE(rgo_size > 0 && i >= rgo_size);
1651  					torture_hrtimeout_us(torture_random(&rand) % gp_poll_wi,
1652  							     1000, &rand);
1653  				}
1654  				rcu_torture_pipe_update(old_rp);
1655  				break;
1656  			case RTWS_POLL_GET_EXP:
1657  				rcu_torture_writer_state = RTWS_POLL_GET_EXP;
1658  				gp_snap = cur_ops->start_gp_poll_exp();
1659  				rcu_torture_writer_state = RTWS_POLL_WAIT_EXP;
1660  				while (!cur_ops->poll_gp_state_exp(gp_snap))
1661  					torture_hrtimeout_us(torture_random(&rand) % gp_poll_wi_exp,
1662  							     1000, &rand);
1663  				rcu_torture_pipe_update(old_rp);
1664  				break;
1665  			case RTWS_POLL_GET_EXP_FULL:
1666  				rcu_torture_writer_state = RTWS_POLL_GET_EXP_FULL;
1667  				cur_ops->start_gp_poll_exp_full(&gp_snap_full);
1668  				rcu_torture_writer_state = RTWS_POLL_WAIT_EXP_FULL;
1669  				while (!cur_ops->poll_gp_state_full(&gp_snap_full))
1670  					torture_hrtimeout_us(torture_random(&rand) % gp_poll_wi_exp,
1671  							     1000, &rand);
1672  				rcu_torture_pipe_update(old_rp);
1673  				break;
1674  			case RTWS_SYNC:
1675  				rcu_torture_writer_state = RTWS_SYNC;
1676  				do_rtws_sync(&rand, cur_ops->sync);
1677  				rcu_torture_pipe_update(old_rp);
1678  				break;
1679  			default:
1680  				WARN_ON_ONCE(1);
1681  				break;
1682  			}
1683  		}
1684  		WRITE_ONCE(rcu_torture_current_version,
1685  			   rcu_torture_current_version + 1);
1686  		/* Cycle through nesting levels of rcu_expedite_gp() calls. */
1687  		if (can_expedite &&
1688  		    !(torture_random(&rand) & 0xff & (!!expediting - 1))) {
1689  			WARN_ON_ONCE(expediting == 0 && rcu_gp_is_expedited());
1690  			if (expediting >= 0)
1691  				rcu_expedite_gp();
1692  			else
1693  				rcu_unexpedite_gp();
1694  			if (++expediting > 3)
1695  				expediting = -expediting;
1696  		} else if (!can_expedite) { /* Disabled during boot, recheck. */
1697  			can_expedite = !rcu_gp_is_expedited() &&
1698  				       !rcu_gp_is_normal();
1699  		}
1700  		rcu_torture_writer_state = RTWS_STUTTER;
1701  		boot_ended = rcu_inkernel_boot_has_ended();
1702  		stutter_waited = stutter_wait("rcu_torture_writer");
1703  		if (stutter_waited &&
1704  		    !atomic_read(&rcu_fwd_cb_nodelay) &&
1705  		    !cur_ops->slow_gps &&
1706  		    !torture_must_stop() &&
1707  		    boot_ended &&
1708  		    time_after(jiffies, stallsdone))
1709  			for (i = 0; i < ARRAY_SIZE(rcu_tortures); i++)
1710  				if (list_empty(&rcu_tortures[i].rtort_free) &&
1711  				    rcu_access_pointer(rcu_torture_current) != &rcu_tortures[i]) {
1712  					tracing_off();
1713  					if (cur_ops->gp_kthread_dbg)
1714  						cur_ops->gp_kthread_dbg();
1715  					WARN(1, "%s: rtort_pipe_count: %d\n", __func__, rcu_tortures[i].rtort_pipe_count);
1716  					rcu_ftrace_dump(DUMP_ALL);
1717  				}
1718  		if (stutter_waited)
1719  			sched_set_normal(current, oldnice);
1720  	} while (!torture_must_stop());
1721  	rcu_torture_current = NULL;  // Let stats task know that we are done.
1722  	/* Reset expediting back to unexpedited. */
1723  	if (expediting > 0)
1724  		expediting = -expediting;
1725  	while (can_expedite && expediting++ < 0)
1726  		rcu_unexpedite_gp();
1727  	WARN_ON_ONCE(can_expedite && rcu_gp_is_expedited());
1728  	if (!can_expedite)
1729  		pr_alert("%s" TORTURE_FLAG
1730  			 " Dynamic grace-period expediting was disabled.\n",
1731  			 torture_type);
1732  	kfree(ulo);
1733  	kfree(rgo);
1734  	rcu_torture_writer_state = RTWS_STOPPING;
1735  	torture_kthread_stopping("rcu_torture_writer");
1736  	return 0;
1737  }
1738  
1739  /*
1740   * RCU torture fake writer kthread.  Repeatedly calls sync, with a random
1741   * delay between calls.
1742   */
1743  static int
rcu_torture_fakewriter(void * arg)1744  rcu_torture_fakewriter(void *arg)
1745  {
1746  	unsigned long gp_snap;
1747  	struct rcu_gp_oldstate gp_snap_full;
1748  	DEFINE_TORTURE_RANDOM(rand);
1749  
1750  	VERBOSE_TOROUT_STRING("rcu_torture_fakewriter task started");
1751  	set_user_nice(current, MAX_NICE);
1752  
1753  	if (WARN_ONCE(nsynctypes == 0,
1754  		      "%s: No update-side primitives.\n", __func__)) {
1755  		/*
1756  		 * No updates primitives, so don't try updating.
1757  		 * The resulting test won't be testing much, hence the
1758  		 * above WARN_ONCE().
1759  		 */
1760  		torture_kthread_stopping("rcu_torture_fakewriter");
1761  		return 0;
1762  	}
1763  
1764  	do {
1765  		torture_hrtimeout_jiffies(torture_random(&rand) % 10, &rand);
1766  		if (cur_ops->cb_barrier != NULL &&
1767  		    torture_random(&rand) % (nrealfakewriters * 8) == 0) {
1768  			cur_ops->cb_barrier();
1769  		} else {
1770  			switch (synctype[torture_random(&rand) % nsynctypes]) {
1771  			case RTWS_DEF_FREE:
1772  				break;
1773  			case RTWS_EXP_SYNC:
1774  				cur_ops->exp_sync();
1775  				break;
1776  			case RTWS_COND_GET:
1777  				gp_snap = cur_ops->get_gp_state();
1778  				torture_hrtimeout_jiffies(torture_random(&rand) % 16, &rand);
1779  				cur_ops->cond_sync(gp_snap);
1780  				break;
1781  			case RTWS_COND_GET_EXP:
1782  				gp_snap = cur_ops->get_gp_state_exp();
1783  				torture_hrtimeout_jiffies(torture_random(&rand) % 16, &rand);
1784  				cur_ops->cond_sync_exp(gp_snap);
1785  				break;
1786  			case RTWS_COND_GET_FULL:
1787  				cur_ops->get_gp_state_full(&gp_snap_full);
1788  				torture_hrtimeout_jiffies(torture_random(&rand) % 16, &rand);
1789  				cur_ops->cond_sync_full(&gp_snap_full);
1790  				break;
1791  			case RTWS_COND_GET_EXP_FULL:
1792  				cur_ops->get_gp_state_full(&gp_snap_full);
1793  				torture_hrtimeout_jiffies(torture_random(&rand) % 16, &rand);
1794  				cur_ops->cond_sync_exp_full(&gp_snap_full);
1795  				break;
1796  			case RTWS_POLL_GET:
1797  				if (cur_ops->start_poll_irqsoff)
1798  					local_irq_disable();
1799  				gp_snap = cur_ops->start_gp_poll();
1800  				if (cur_ops->start_poll_irqsoff)
1801  					local_irq_enable();
1802  				while (!cur_ops->poll_gp_state(gp_snap)) {
1803  					torture_hrtimeout_jiffies(torture_random(&rand) % 16,
1804  								  &rand);
1805  				}
1806  				break;
1807  			case RTWS_POLL_GET_FULL:
1808  				if (cur_ops->start_poll_irqsoff)
1809  					local_irq_disable();
1810  				cur_ops->start_gp_poll_full(&gp_snap_full);
1811  				if (cur_ops->start_poll_irqsoff)
1812  					local_irq_enable();
1813  				while (!cur_ops->poll_gp_state_full(&gp_snap_full)) {
1814  					torture_hrtimeout_jiffies(torture_random(&rand) % 16,
1815  								  &rand);
1816  				}
1817  				break;
1818  			case RTWS_POLL_GET_EXP:
1819  				gp_snap = cur_ops->start_gp_poll_exp();
1820  				while (!cur_ops->poll_gp_state_exp(gp_snap)) {
1821  					torture_hrtimeout_jiffies(torture_random(&rand) % 16,
1822  								  &rand);
1823  				}
1824  				break;
1825  			case RTWS_POLL_GET_EXP_FULL:
1826  				cur_ops->start_gp_poll_exp_full(&gp_snap_full);
1827  				while (!cur_ops->poll_gp_state_full(&gp_snap_full)) {
1828  					torture_hrtimeout_jiffies(torture_random(&rand) % 16,
1829  								  &rand);
1830  				}
1831  				break;
1832  			case RTWS_SYNC:
1833  				cur_ops->sync();
1834  				break;
1835  			default:
1836  				WARN_ON_ONCE(1);
1837  				break;
1838  			}
1839  		}
1840  		stutter_wait("rcu_torture_fakewriter");
1841  	} while (!torture_must_stop());
1842  
1843  	torture_kthread_stopping("rcu_torture_fakewriter");
1844  	return 0;
1845  }
1846  
rcu_torture_timer_cb(struct rcu_head * rhp)1847  static void rcu_torture_timer_cb(struct rcu_head *rhp)
1848  {
1849  	kfree(rhp);
1850  }
1851  
1852  // Set up and carry out testing of RCU's global memory ordering
rcu_torture_reader_do_mbchk(long myid,struct rcu_torture * rtp,struct torture_random_state * trsp)1853  static void rcu_torture_reader_do_mbchk(long myid, struct rcu_torture *rtp,
1854  					struct torture_random_state *trsp)
1855  {
1856  	unsigned long loops;
1857  	int noc = torture_num_online_cpus();
1858  	int rdrchked;
1859  	int rdrchker;
1860  	struct rcu_torture_reader_check *rtrcp; // Me.
1861  	struct rcu_torture_reader_check *rtrcp_assigner; // Assigned us to do checking.
1862  	struct rcu_torture_reader_check *rtrcp_chked; // Reader being checked.
1863  	struct rcu_torture_reader_check *rtrcp_chker; // Reader doing checking when not me.
1864  
1865  	if (myid < 0)
1866  		return; // Don't try this from timer handlers.
1867  
1868  	// Increment my counter.
1869  	rtrcp = &rcu_torture_reader_mbchk[myid];
1870  	WRITE_ONCE(rtrcp->rtc_myloops, rtrcp->rtc_myloops + 1);
1871  
1872  	// Attempt to assign someone else some checking work.
1873  	rdrchked = torture_random(trsp) % nrealreaders;
1874  	rtrcp_chked = &rcu_torture_reader_mbchk[rdrchked];
1875  	rdrchker = torture_random(trsp) % nrealreaders;
1876  	rtrcp_chker = &rcu_torture_reader_mbchk[rdrchker];
1877  	if (rdrchked != myid && rdrchked != rdrchker && noc >= rdrchked && noc >= rdrchker &&
1878  	    smp_load_acquire(&rtrcp->rtc_chkrdr) < 0 && // Pairs with smp_store_release below.
1879  	    !READ_ONCE(rtp->rtort_chkp) &&
1880  	    !smp_load_acquire(&rtrcp_chker->rtc_assigner)) { // Pairs with smp_store_release below.
1881  		rtrcp->rtc_chkloops = READ_ONCE(rtrcp_chked->rtc_myloops);
1882  		WARN_ON_ONCE(rtrcp->rtc_chkrdr >= 0);
1883  		rtrcp->rtc_chkrdr = rdrchked;
1884  		WARN_ON_ONCE(rtrcp->rtc_ready); // This gets set after the grace period ends.
1885  		if (cmpxchg_relaxed(&rtrcp_chker->rtc_assigner, NULL, rtrcp) ||
1886  		    cmpxchg_relaxed(&rtp->rtort_chkp, NULL, rtrcp))
1887  			(void)cmpxchg_relaxed(&rtrcp_chker->rtc_assigner, rtrcp, NULL); // Back out.
1888  	}
1889  
1890  	// If assigned some completed work, do it!
1891  	rtrcp_assigner = READ_ONCE(rtrcp->rtc_assigner);
1892  	if (!rtrcp_assigner || !smp_load_acquire(&rtrcp_assigner->rtc_ready))
1893  		return; // No work or work not yet ready.
1894  	rdrchked = rtrcp_assigner->rtc_chkrdr;
1895  	if (WARN_ON_ONCE(rdrchked < 0))
1896  		return;
1897  	rtrcp_chked = &rcu_torture_reader_mbchk[rdrchked];
1898  	loops = READ_ONCE(rtrcp_chked->rtc_myloops);
1899  	atomic_inc(&n_rcu_torture_mbchk_tries);
1900  	if (ULONG_CMP_LT(loops, rtrcp_assigner->rtc_chkloops))
1901  		atomic_inc(&n_rcu_torture_mbchk_fail);
1902  	rtrcp_assigner->rtc_chkloops = loops + ULONG_MAX / 2;
1903  	rtrcp_assigner->rtc_ready = 0;
1904  	smp_store_release(&rtrcp->rtc_assigner, NULL); // Someone else can assign us work.
1905  	smp_store_release(&rtrcp_assigner->rtc_chkrdr, -1); // Assigner can again assign.
1906  }
1907  
1908  // Verify the specified RCUTORTURE_RDR* state.
1909  #define ROEC_ARGS "%s %s: Current %#x  To add %#x  To remove %#x  preempt_count() %#x\n", __func__, s, curstate, new, old, preempt_count()
rcutorture_one_extend_check(char * s,int curstate,int new,int old,bool insoftirq)1910  static void rcutorture_one_extend_check(char *s, int curstate, int new, int old, bool insoftirq)
1911  {
1912  	int mask;
1913  
1914  	if (!IS_ENABLED(CONFIG_RCU_TORTURE_TEST_CHK_RDR_STATE))
1915  		return;
1916  
1917  	WARN_ONCE(!(curstate & RCUTORTURE_RDR_IRQ) && irqs_disabled(), ROEC_ARGS);
1918  	WARN_ONCE((curstate & RCUTORTURE_RDR_IRQ) && !irqs_disabled(), ROEC_ARGS);
1919  
1920  	// If CONFIG_PREEMPT_COUNT=n, further checks are unreliable.
1921  	if (!IS_ENABLED(CONFIG_PREEMPT_COUNT))
1922  		return;
1923  
1924  	WARN_ONCE((curstate & (RCUTORTURE_RDR_BH | RCUTORTURE_RDR_RBH)) &&
1925  		  !(preempt_count() & SOFTIRQ_MASK), ROEC_ARGS);
1926  	WARN_ONCE((curstate & (RCUTORTURE_RDR_PREEMPT | RCUTORTURE_RDR_SCHED)) &&
1927  		  !(preempt_count() & PREEMPT_MASK), ROEC_ARGS);
1928  	WARN_ONCE(cur_ops->readlock_nesting &&
1929  		  (curstate & (RCUTORTURE_RDR_RCU_1 | RCUTORTURE_RDR_RCU_2)) &&
1930  		  cur_ops->readlock_nesting() == 0, ROEC_ARGS);
1931  
1932  	// Timer handlers have all sorts of stuff disabled, so ignore
1933  	// unintended disabling.
1934  	if (insoftirq)
1935  		return;
1936  
1937  	WARN_ONCE(cur_ops->extendables &&
1938  		  !(curstate & (RCUTORTURE_RDR_BH | RCUTORTURE_RDR_RBH)) &&
1939  		  (preempt_count() & SOFTIRQ_MASK), ROEC_ARGS);
1940  
1941  	/*
1942  	 * non-preemptible RCU in a preemptible kernel uses preempt_disable()
1943  	 * as rcu_read_lock().
1944  	 */
1945  	mask = RCUTORTURE_RDR_PREEMPT | RCUTORTURE_RDR_SCHED;
1946  	if (!IS_ENABLED(CONFIG_PREEMPT_RCU))
1947  		mask |= RCUTORTURE_RDR_RCU_1 | RCUTORTURE_RDR_RCU_2;
1948  
1949  	WARN_ONCE(cur_ops->extendables && !(curstate & mask) &&
1950  		  (preempt_count() & PREEMPT_MASK), ROEC_ARGS);
1951  
1952  	/*
1953  	 * non-preemptible RCU in a preemptible kernel uses "preempt_count() &
1954  	 * PREEMPT_MASK" as ->readlock_nesting().
1955  	 */
1956  	mask = RCUTORTURE_RDR_RCU_1 | RCUTORTURE_RDR_RCU_2;
1957  	if (!IS_ENABLED(CONFIG_PREEMPT_RCU))
1958  		mask |= RCUTORTURE_RDR_PREEMPT | RCUTORTURE_RDR_SCHED;
1959  
1960  	WARN_ONCE(cur_ops->readlock_nesting && !(curstate & mask) &&
1961  		  cur_ops->readlock_nesting() > 0, ROEC_ARGS);
1962  }
1963  
1964  /*
1965   * Do one extension of an RCU read-side critical section using the
1966   * current reader state in readstate (set to zero for initial entry
1967   * to extended critical section), set the new state as specified by
1968   * newstate (set to zero for final exit from extended critical section),
1969   * and random-number-generator state in trsp.  If this is neither the
1970   * beginning or end of the critical section and if there was actually a
1971   * change, do a ->read_delay().
1972   */
rcutorture_one_extend(int * readstate,int newstate,bool insoftirq,struct torture_random_state * trsp,struct rt_read_seg * rtrsp)1973  static void rcutorture_one_extend(int *readstate, int newstate, bool insoftirq,
1974  				  struct torture_random_state *trsp,
1975  				  struct rt_read_seg *rtrsp)
1976  {
1977  	bool first;
1978  	unsigned long flags;
1979  	int idxnew1 = -1;
1980  	int idxnew2 = -1;
1981  	int idxold1 = *readstate;
1982  	int idxold2 = idxold1;
1983  	int statesnew = ~*readstate & newstate;
1984  	int statesold = *readstate & ~newstate;
1985  
1986  	first = idxold1 == 0;
1987  	WARN_ON_ONCE(idxold2 < 0);
1988  	WARN_ON_ONCE(idxold2 & ~RCUTORTURE_RDR_ALLBITS);
1989  	rcutorture_one_extend_check("before change", idxold1, statesnew, statesold, insoftirq);
1990  	rtrsp->rt_readstate = newstate;
1991  
1992  	/* First, put new protection in place to avoid critical-section gap. */
1993  	if (statesnew & RCUTORTURE_RDR_BH)
1994  		local_bh_disable();
1995  	if (statesnew & RCUTORTURE_RDR_RBH)
1996  		rcu_read_lock_bh();
1997  	if (statesnew & RCUTORTURE_RDR_IRQ)
1998  		local_irq_disable();
1999  	if (statesnew & RCUTORTURE_RDR_PREEMPT)
2000  		preempt_disable();
2001  	if (statesnew & RCUTORTURE_RDR_SCHED)
2002  		rcu_read_lock_sched();
2003  	if (statesnew & RCUTORTURE_RDR_RCU_1)
2004  		idxnew1 = (cur_ops->readlock() << RCUTORTURE_RDR_SHIFT_1) & RCUTORTURE_RDR_MASK_1;
2005  	if (statesnew & RCUTORTURE_RDR_RCU_2)
2006  		idxnew2 = (cur_ops->readlock() << RCUTORTURE_RDR_SHIFT_2) & RCUTORTURE_RDR_MASK_2;
2007  
2008  	// Complain unless both the old and the new protection is in place.
2009  	rcutorture_one_extend_check("during change",
2010  				    idxold1 | statesnew, statesnew, statesold, insoftirq);
2011  
2012  	// Sample CPU under both sets of protections to reduce confusion.
2013  	if (IS_ENABLED(CONFIG_RCU_TORTURE_TEST_LOG_CPU)) {
2014  		int cpu = raw_smp_processor_id();
2015  		rtrsp->rt_cpu = cpu;
2016  		if (!first) {
2017  			rtrsp[-1].rt_end_cpu = cpu;
2018  			if (cur_ops->reader_blocked)
2019  				rtrsp[-1].rt_preempted = cur_ops->reader_blocked();
2020  		}
2021  	}
2022  	// Sample grace-period sequence number, as good a place as any.
2023  	if (IS_ENABLED(CONFIG_RCU_TORTURE_TEST_LOG_GP) && cur_ops->gather_gp_seqs) {
2024  		rtrsp->rt_gp_seq = cur_ops->gather_gp_seqs();
2025  		rtrsp->rt_ts = ktime_get_mono_fast_ns();
2026  		if (!first)
2027  			rtrsp[-1].rt_gp_seq_end = rtrsp->rt_gp_seq;
2028  	}
2029  
2030  	/*
2031  	 * Next, remove old protection, in decreasing order of strength
2032  	 * to avoid unlock paths that aren't safe in the stronger
2033  	 * context. Namely: BH can not be enabled with disabled interrupts.
2034  	 * Additionally PREEMPT_RT requires that BH is enabled in preemptible
2035  	 * context.
2036  	 */
2037  	if (statesold & RCUTORTURE_RDR_IRQ)
2038  		local_irq_enable();
2039  	if (statesold & RCUTORTURE_RDR_PREEMPT)
2040  		preempt_enable();
2041  	if (statesold & RCUTORTURE_RDR_SCHED)
2042  		rcu_read_unlock_sched();
2043  	if (statesold & RCUTORTURE_RDR_BH)
2044  		local_bh_enable();
2045  	if (statesold & RCUTORTURE_RDR_RBH)
2046  		rcu_read_unlock_bh();
2047  	if (statesold & RCUTORTURE_RDR_RCU_2) {
2048  		cur_ops->readunlock((idxold2 & RCUTORTURE_RDR_MASK_2) >> RCUTORTURE_RDR_SHIFT_2);
2049  		WARN_ON_ONCE(idxnew2 != -1);
2050  		idxold2 = 0;
2051  	}
2052  	if (statesold & RCUTORTURE_RDR_RCU_1) {
2053  		bool lockit;
2054  
2055  		lockit = !cur_ops->no_pi_lock && !statesnew && !(torture_random(trsp) & 0xffff);
2056  		if (lockit)
2057  			raw_spin_lock_irqsave(&current->pi_lock, flags);
2058  		cur_ops->readunlock((idxold1 & RCUTORTURE_RDR_MASK_1) >> RCUTORTURE_RDR_SHIFT_1);
2059  		WARN_ON_ONCE(idxnew1 != -1);
2060  		idxold1 = 0;
2061  		if (lockit)
2062  			raw_spin_unlock_irqrestore(&current->pi_lock, flags);
2063  	}
2064  
2065  	/* Delay if neither beginning nor end and there was a change. */
2066  	if ((statesnew || statesold) && *readstate && newstate)
2067  		cur_ops->read_delay(trsp, rtrsp);
2068  
2069  	/* Update the reader state. */
2070  	if (idxnew1 == -1)
2071  		idxnew1 = idxold1 & RCUTORTURE_RDR_MASK_1;
2072  	WARN_ON_ONCE(idxnew1 < 0);
2073  	if (idxnew2 == -1)
2074  		idxnew2 = idxold2 & RCUTORTURE_RDR_MASK_2;
2075  	WARN_ON_ONCE(idxnew2 < 0);
2076  	*readstate = idxnew1 | idxnew2 | newstate;
2077  	WARN_ON_ONCE(*readstate < 0);
2078  	if (WARN_ON_ONCE(*readstate & ~RCUTORTURE_RDR_ALLBITS))
2079  		pr_info("Unexpected readstate value of %#x\n", *readstate);
2080  	rcutorture_one_extend_check("after change", *readstate, statesnew, statesold, insoftirq);
2081  }
2082  
2083  /* Return the biggest extendables mask given current RCU and boot parameters. */
rcutorture_extend_mask_max(void)2084  static int rcutorture_extend_mask_max(void)
2085  {
2086  	int mask;
2087  
2088  	WARN_ON_ONCE(extendables & ~RCUTORTURE_MAX_EXTEND);
2089  	mask = extendables & RCUTORTURE_MAX_EXTEND & cur_ops->extendables;
2090  	mask = mask | RCUTORTURE_RDR_RCU_1 | RCUTORTURE_RDR_RCU_2;
2091  	return mask;
2092  }
2093  
2094  /* Return a random protection state mask, but with at least one bit set. */
2095  static int
rcutorture_extend_mask(int oldmask,struct torture_random_state * trsp)2096  rcutorture_extend_mask(int oldmask, struct torture_random_state *trsp)
2097  {
2098  	int mask = rcutorture_extend_mask_max();
2099  	unsigned long randmask1 = torture_random(trsp);
2100  	unsigned long randmask2 = randmask1 >> 3;
2101  	unsigned long preempts = RCUTORTURE_RDR_PREEMPT | RCUTORTURE_RDR_SCHED;
2102  	unsigned long preempts_irq = preempts | RCUTORTURE_RDR_IRQ;
2103  	unsigned long bhs = RCUTORTURE_RDR_BH | RCUTORTURE_RDR_RBH;
2104  
2105  	WARN_ON_ONCE(mask >> RCUTORTURE_RDR_SHIFT_1);  // Can't have reader idx bits.
2106  	/* Mostly only one bit (need preemption!), sometimes lots of bits. */
2107  	if (!(randmask1 & 0x7))
2108  		mask = mask & randmask2;
2109  	else
2110  		mask = mask & (1 << (randmask2 % RCUTORTURE_RDR_NBITS));
2111  
2112  	// Can't have nested RCU reader without outer RCU reader.
2113  	if (!(mask & RCUTORTURE_RDR_RCU_1) && (mask & RCUTORTURE_RDR_RCU_2)) {
2114  		if (oldmask & RCUTORTURE_RDR_RCU_1)
2115  			mask &= ~RCUTORTURE_RDR_RCU_2;
2116  		else
2117  			mask |= RCUTORTURE_RDR_RCU_1;
2118  	}
2119  
2120  	/*
2121  	 * Can't enable bh w/irq disabled.
2122  	 */
2123  	if (mask & RCUTORTURE_RDR_IRQ)
2124  		mask |= oldmask & bhs;
2125  
2126  	/*
2127  	 * Ideally these sequences would be detected in debug builds
2128  	 * (regardless of RT), but until then don't stop testing
2129  	 * them on non-RT.
2130  	 */
2131  	if (IS_ENABLED(CONFIG_PREEMPT_RT)) {
2132  		/* Can't modify BH in atomic context */
2133  		if (oldmask & preempts_irq)
2134  			mask &= ~bhs;
2135  		if ((oldmask | mask) & preempts_irq)
2136  			mask |= oldmask & bhs;
2137  	}
2138  
2139  	return mask ?: RCUTORTURE_RDR_RCU_1;
2140  }
2141  
2142  /*
2143   * Do a randomly selected number of extensions of an existing RCU read-side
2144   * critical section.
2145   */
2146  static struct rt_read_seg *
rcutorture_loop_extend(int * readstate,bool insoftirq,struct torture_random_state * trsp,struct rt_read_seg * rtrsp)2147  rcutorture_loop_extend(int *readstate, bool insoftirq, struct torture_random_state *trsp,
2148  		       struct rt_read_seg *rtrsp)
2149  {
2150  	int i;
2151  	int j;
2152  	int mask = rcutorture_extend_mask_max();
2153  
2154  	WARN_ON_ONCE(!*readstate); /* -Existing- RCU read-side critsect! */
2155  	if (!((mask - 1) & mask))
2156  		return rtrsp;  /* Current RCU reader not extendable. */
2157  	/* Bias towards larger numbers of loops. */
2158  	i = torture_random(trsp);
2159  	i = ((i | (i >> 3)) & RCUTORTURE_RDR_MAX_LOOPS) + 1;
2160  	for (j = 0; j < i; j++) {
2161  		mask = rcutorture_extend_mask(*readstate, trsp);
2162  		rcutorture_one_extend(readstate, mask, insoftirq, trsp, &rtrsp[j]);
2163  	}
2164  	return &rtrsp[j];
2165  }
2166  
2167  /*
2168   * Do one read-side critical section, returning false if there was
2169   * no data to read.  Can be invoked both from process context and
2170   * from a timer handler.
2171   */
rcu_torture_one_read(struct torture_random_state * trsp,long myid)2172  static bool rcu_torture_one_read(struct torture_random_state *trsp, long myid)
2173  {
2174  	bool checkpolling = !(torture_random(trsp) & 0xfff);
2175  	unsigned long cookie;
2176  	struct rcu_gp_oldstate cookie_full;
2177  	int i;
2178  	unsigned long started;
2179  	unsigned long completed;
2180  	int newstate;
2181  	struct rcu_torture *p;
2182  	int pipe_count;
2183  	bool preempted = false;
2184  	int readstate = 0;
2185  	struct rt_read_seg rtseg[RCUTORTURE_RDR_MAX_SEGS] = { { 0 } };
2186  	struct rt_read_seg *rtrsp = &rtseg[0];
2187  	struct rt_read_seg *rtrsp1;
2188  	unsigned long long ts;
2189  
2190  	WARN_ON_ONCE(!rcu_is_watching());
2191  	newstate = rcutorture_extend_mask(readstate, trsp);
2192  	rcutorture_one_extend(&readstate, newstate, myid < 0, trsp, rtrsp++);
2193  	if (checkpolling) {
2194  		if (cur_ops->get_gp_state && cur_ops->poll_gp_state)
2195  			cookie = cur_ops->get_gp_state();
2196  		if (cur_ops->get_gp_state_full && cur_ops->poll_gp_state_full)
2197  			cur_ops->get_gp_state_full(&cookie_full);
2198  	}
2199  	started = cur_ops->get_gp_seq();
2200  	ts = rcu_trace_clock_local();
2201  	p = rcu_dereference_check(rcu_torture_current,
2202  				  !cur_ops->readlock_held || cur_ops->readlock_held());
2203  	if (p == NULL) {
2204  		/* Wait for rcu_torture_writer to get underway */
2205  		rcutorture_one_extend(&readstate, 0, myid < 0, trsp, rtrsp);
2206  		return false;
2207  	}
2208  	if (p->rtort_mbtest == 0)
2209  		atomic_inc(&n_rcu_torture_mberror);
2210  	rcu_torture_reader_do_mbchk(myid, p, trsp);
2211  	rtrsp = rcutorture_loop_extend(&readstate, myid < 0, trsp, rtrsp);
2212  	preempt_disable();
2213  	pipe_count = READ_ONCE(p->rtort_pipe_count);
2214  	if (pipe_count > RCU_TORTURE_PIPE_LEN) {
2215  		// Should not happen in a correct RCU implementation,
2216  		// happens quite often for torture_type=busted.
2217  		pipe_count = RCU_TORTURE_PIPE_LEN;
2218  	}
2219  	completed = cur_ops->get_gp_seq();
2220  	if (pipe_count > 1) {
2221  		do_trace_rcu_torture_read(cur_ops->name, &p->rtort_rcu,
2222  					  ts, started, completed);
2223  		rcu_ftrace_dump(DUMP_ALL);
2224  	}
2225  	__this_cpu_inc(rcu_torture_count[pipe_count]);
2226  	completed = rcutorture_seq_diff(completed, started);
2227  	if (completed > RCU_TORTURE_PIPE_LEN) {
2228  		/* Should not happen, but... */
2229  		completed = RCU_TORTURE_PIPE_LEN;
2230  	}
2231  	__this_cpu_inc(rcu_torture_batch[completed]);
2232  	preempt_enable();
2233  	if (checkpolling) {
2234  		if (cur_ops->get_gp_state && cur_ops->poll_gp_state)
2235  			WARN_ONCE(cur_ops->poll_gp_state(cookie),
2236  				  "%s: Cookie check 2 failed %s(%d) %lu->%lu\n",
2237  				  __func__,
2238  				  rcu_torture_writer_state_getname(),
2239  				  rcu_torture_writer_state,
2240  				  cookie, cur_ops->get_gp_state());
2241  		if (cur_ops->get_gp_state_full && cur_ops->poll_gp_state_full)
2242  			WARN_ONCE(cur_ops->poll_gp_state_full(&cookie_full),
2243  				  "%s: Cookie check 6 failed %s(%d) online %*pbl\n",
2244  				  __func__,
2245  				  rcu_torture_writer_state_getname(),
2246  				  rcu_torture_writer_state,
2247  				  cpumask_pr_args(cpu_online_mask));
2248  	}
2249  	if (cur_ops->reader_blocked)
2250  		preempted = cur_ops->reader_blocked();
2251  	rcutorture_one_extend(&readstate, 0, myid < 0, trsp, rtrsp);
2252  	WARN_ON_ONCE(readstate);
2253  	// This next splat is expected behavior if leakpointer, especially
2254  	// for CONFIG_RCU_STRICT_GRACE_PERIOD=y kernels.
2255  	WARN_ON_ONCE(leakpointer && READ_ONCE(p->rtort_pipe_count) > 1);
2256  
2257  	/* If error or close call, record the sequence of reader protections. */
2258  	if ((pipe_count > 1 || completed > 1) && !xchg(&err_segs_recorded, 1)) {
2259  		i = 0;
2260  		for (rtrsp1 = &rtseg[0]; rtrsp1 < rtrsp; rtrsp1++)
2261  			err_segs[i++] = *rtrsp1;
2262  		rt_read_nsegs = i;
2263  		rt_read_preempted = preempted;
2264  	}
2265  
2266  	return true;
2267  }
2268  
2269  static DEFINE_TORTURE_RANDOM_PERCPU(rcu_torture_timer_rand);
2270  
2271  /*
2272   * RCU torture reader from timer handler.  Dereferences rcu_torture_current,
2273   * incrementing the corresponding element of the pipeline array.  The
2274   * counter in the element should never be greater than 1, otherwise, the
2275   * RCU implementation is broken.
2276   */
rcu_torture_timer(struct timer_list * unused)2277  static void rcu_torture_timer(struct timer_list *unused)
2278  {
2279  	atomic_long_inc(&n_rcu_torture_timers);
2280  	(void)rcu_torture_one_read(this_cpu_ptr(&rcu_torture_timer_rand), -1);
2281  
2282  	/* Test call_rcu() invocation from interrupt handler. */
2283  	if (cur_ops->call) {
2284  		struct rcu_head *rhp = kmalloc(sizeof(*rhp), GFP_NOWAIT);
2285  
2286  		if (rhp)
2287  			cur_ops->call(rhp, rcu_torture_timer_cb);
2288  	}
2289  }
2290  
2291  /*
2292   * RCU torture reader kthread.  Repeatedly dereferences rcu_torture_current,
2293   * incrementing the corresponding element of the pipeline array.  The
2294   * counter in the element should never be greater than 1, otherwise, the
2295   * RCU implementation is broken.
2296   */
2297  static int
rcu_torture_reader(void * arg)2298  rcu_torture_reader(void *arg)
2299  {
2300  	unsigned long lastsleep = jiffies;
2301  	long myid = (long)arg;
2302  	int mynumonline = myid;
2303  	DEFINE_TORTURE_RANDOM(rand);
2304  	struct timer_list t;
2305  
2306  	VERBOSE_TOROUT_STRING("rcu_torture_reader task started");
2307  	set_user_nice(current, MAX_NICE);
2308  	if (irqreader && cur_ops->irq_capable)
2309  		timer_setup_on_stack(&t, rcu_torture_timer, 0);
2310  	tick_dep_set_task(current, TICK_DEP_BIT_RCU);
2311  	do {
2312  		if (irqreader && cur_ops->irq_capable) {
2313  			if (!timer_pending(&t))
2314  				mod_timer(&t, jiffies + 1);
2315  		}
2316  		if (!rcu_torture_one_read(&rand, myid) && !torture_must_stop())
2317  			schedule_timeout_interruptible(HZ);
2318  		if (time_after(jiffies, lastsleep) && !torture_must_stop()) {
2319  			torture_hrtimeout_us(500, 1000, &rand);
2320  			lastsleep = jiffies + 10;
2321  		}
2322  		while (torture_num_online_cpus() < mynumonline && !torture_must_stop())
2323  			schedule_timeout_interruptible(HZ / 5);
2324  		stutter_wait("rcu_torture_reader");
2325  	} while (!torture_must_stop());
2326  	if (irqreader && cur_ops->irq_capable) {
2327  		del_timer_sync(&t);
2328  		destroy_timer_on_stack(&t);
2329  	}
2330  	tick_dep_clear_task(current, TICK_DEP_BIT_RCU);
2331  	torture_kthread_stopping("rcu_torture_reader");
2332  	return 0;
2333  }
2334  
2335  /*
2336   * Randomly Toggle CPUs' callback-offload state.  This uses hrtimers to
2337   * increase race probabilities and fuzzes the interval between toggling.
2338   */
rcu_nocb_toggle(void * arg)2339  static int rcu_nocb_toggle(void *arg)
2340  {
2341  	int cpu;
2342  	int maxcpu = -1;
2343  	int oldnice = task_nice(current);
2344  	long r;
2345  	DEFINE_TORTURE_RANDOM(rand);
2346  	ktime_t toggle_delay;
2347  	unsigned long toggle_fuzz;
2348  	ktime_t toggle_interval = ms_to_ktime(nocbs_toggle);
2349  
2350  	VERBOSE_TOROUT_STRING("rcu_nocb_toggle task started");
2351  	while (!rcu_inkernel_boot_has_ended())
2352  		schedule_timeout_interruptible(HZ / 10);
2353  	for_each_possible_cpu(cpu)
2354  		maxcpu = cpu;
2355  	WARN_ON(maxcpu < 0);
2356  	if (toggle_interval > ULONG_MAX)
2357  		toggle_fuzz = ULONG_MAX >> 3;
2358  	else
2359  		toggle_fuzz = toggle_interval >> 3;
2360  	if (toggle_fuzz <= 0)
2361  		toggle_fuzz = NSEC_PER_USEC;
2362  	do {
2363  		r = torture_random(&rand);
2364  		cpu = (r >> 1) % (maxcpu + 1);
2365  		if (r & 0x1) {
2366  			rcu_nocb_cpu_offload(cpu);
2367  			atomic_long_inc(&n_nocb_offload);
2368  		} else {
2369  			rcu_nocb_cpu_deoffload(cpu);
2370  			atomic_long_inc(&n_nocb_deoffload);
2371  		}
2372  		toggle_delay = torture_random(&rand) % toggle_fuzz + toggle_interval;
2373  		set_current_state(TASK_INTERRUPTIBLE);
2374  		schedule_hrtimeout(&toggle_delay, HRTIMER_MODE_REL);
2375  		if (stutter_wait("rcu_nocb_toggle"))
2376  			sched_set_normal(current, oldnice);
2377  	} while (!torture_must_stop());
2378  	torture_kthread_stopping("rcu_nocb_toggle");
2379  	return 0;
2380  }
2381  
2382  /*
2383   * Print torture statistics.  Caller must ensure that there is only
2384   * one call to this function at a given time!!!  This is normally
2385   * accomplished by relying on the module system to only have one copy
2386   * of the module loaded, and then by giving the rcu_torture_stats
2387   * kthread full control (or the init/cleanup functions when rcu_torture_stats
2388   * thread is not running).
2389   */
2390  static void
rcu_torture_stats_print(void)2391  rcu_torture_stats_print(void)
2392  {
2393  	int cpu;
2394  	int i;
2395  	long pipesummary[RCU_TORTURE_PIPE_LEN + 1] = { 0 };
2396  	long batchsummary[RCU_TORTURE_PIPE_LEN + 1] = { 0 };
2397  	struct rcu_torture *rtcp;
2398  	static unsigned long rtcv_snap = ULONG_MAX;
2399  	static bool splatted;
2400  	struct task_struct *wtp;
2401  
2402  	for_each_possible_cpu(cpu) {
2403  		for (i = 0; i < RCU_TORTURE_PIPE_LEN + 1; i++) {
2404  			pipesummary[i] += READ_ONCE(per_cpu(rcu_torture_count, cpu)[i]);
2405  			batchsummary[i] += READ_ONCE(per_cpu(rcu_torture_batch, cpu)[i]);
2406  		}
2407  	}
2408  	for (i = RCU_TORTURE_PIPE_LEN; i >= 0; i--) {
2409  		if (pipesummary[i] != 0)
2410  			break;
2411  	}
2412  
2413  	pr_alert("%s%s ", torture_type, TORTURE_FLAG);
2414  	rtcp = rcu_access_pointer(rcu_torture_current);
2415  	pr_cont("rtc: %p %s: %lu tfle: %d rta: %d rtaf: %d rtf: %d ",
2416  		rtcp,
2417  		rtcp && !rcu_stall_is_suppressed_at_boot() ? "ver" : "VER",
2418  		rcu_torture_current_version,
2419  		list_empty(&rcu_torture_freelist),
2420  		atomic_read(&n_rcu_torture_alloc),
2421  		atomic_read(&n_rcu_torture_alloc_fail),
2422  		atomic_read(&n_rcu_torture_free));
2423  	pr_cont("rtmbe: %d rtmbkf: %d/%d rtbe: %ld rtbke: %ld ",
2424  		atomic_read(&n_rcu_torture_mberror),
2425  		atomic_read(&n_rcu_torture_mbchk_fail), atomic_read(&n_rcu_torture_mbchk_tries),
2426  		n_rcu_torture_barrier_error,
2427  		n_rcu_torture_boost_ktrerror);
2428  	pr_cont("rtbf: %ld rtb: %ld nt: %ld ",
2429  		n_rcu_torture_boost_failure,
2430  		n_rcu_torture_boosts,
2431  		atomic_long_read(&n_rcu_torture_timers));
2432  	torture_onoff_stats();
2433  	pr_cont("barrier: %ld/%ld:%ld ",
2434  		data_race(n_barrier_successes),
2435  		data_race(n_barrier_attempts),
2436  		data_race(n_rcu_torture_barrier_error));
2437  	pr_cont("read-exits: %ld ", data_race(n_read_exits)); // Statistic.
2438  	pr_cont("nocb-toggles: %ld:%ld\n",
2439  		atomic_long_read(&n_nocb_offload), atomic_long_read(&n_nocb_deoffload));
2440  
2441  	pr_alert("%s%s ", torture_type, TORTURE_FLAG);
2442  	if (atomic_read(&n_rcu_torture_mberror) ||
2443  	    atomic_read(&n_rcu_torture_mbchk_fail) ||
2444  	    n_rcu_torture_barrier_error || n_rcu_torture_boost_ktrerror ||
2445  	    n_rcu_torture_boost_failure || i > 1) {
2446  		pr_cont("%s", "!!! ");
2447  		atomic_inc(&n_rcu_torture_error);
2448  		WARN_ON_ONCE(atomic_read(&n_rcu_torture_mberror));
2449  		WARN_ON_ONCE(atomic_read(&n_rcu_torture_mbchk_fail));
2450  		WARN_ON_ONCE(n_rcu_torture_barrier_error);  // rcu_barrier()
2451  		WARN_ON_ONCE(n_rcu_torture_boost_ktrerror); // no boost kthread
2452  		WARN_ON_ONCE(n_rcu_torture_boost_failure); // boost failed (TIMER_SOFTIRQ RT prio?)
2453  		WARN_ON_ONCE(i > 1); // Too-short grace period
2454  	}
2455  	pr_cont("Reader Pipe: ");
2456  	for (i = 0; i < RCU_TORTURE_PIPE_LEN + 1; i++)
2457  		pr_cont(" %ld", pipesummary[i]);
2458  	pr_cont("\n");
2459  
2460  	pr_alert("%s%s ", torture_type, TORTURE_FLAG);
2461  	pr_cont("Reader Batch: ");
2462  	for (i = 0; i < RCU_TORTURE_PIPE_LEN + 1; i++)
2463  		pr_cont(" %ld", batchsummary[i]);
2464  	pr_cont("\n");
2465  
2466  	pr_alert("%s%s ", torture_type, TORTURE_FLAG);
2467  	pr_cont("Free-Block Circulation: ");
2468  	for (i = 0; i < RCU_TORTURE_PIPE_LEN + 1; i++) {
2469  		pr_cont(" %d", atomic_read(&rcu_torture_wcount[i]));
2470  	}
2471  	pr_cont("\n");
2472  
2473  	if (cur_ops->stats)
2474  		cur_ops->stats();
2475  	if (rtcv_snap == rcu_torture_current_version &&
2476  	    rcu_access_pointer(rcu_torture_current) &&
2477  	    !rcu_stall_is_suppressed()) {
2478  		int __maybe_unused flags = 0;
2479  		unsigned long __maybe_unused gp_seq = 0;
2480  
2481  		if (cur_ops->get_gp_data)
2482  			cur_ops->get_gp_data(&flags, &gp_seq);
2483  		wtp = READ_ONCE(writer_task);
2484  		pr_alert("??? Writer stall state %s(%d) g%lu f%#x ->state %#x cpu %d\n",
2485  			 rcu_torture_writer_state_getname(),
2486  			 rcu_torture_writer_state, gp_seq, flags,
2487  			 wtp == NULL ? ~0U : wtp->__state,
2488  			 wtp == NULL ? -1 : (int)task_cpu(wtp));
2489  		if (!splatted && wtp) {
2490  			sched_show_task(wtp);
2491  			splatted = true;
2492  		}
2493  		if (cur_ops->gp_kthread_dbg)
2494  			cur_ops->gp_kthread_dbg();
2495  		rcu_ftrace_dump(DUMP_ALL);
2496  	}
2497  	rtcv_snap = rcu_torture_current_version;
2498  }
2499  
2500  /*
2501   * Periodically prints torture statistics, if periodic statistics printing
2502   * was specified via the stat_interval module parameter.
2503   */
2504  static int
rcu_torture_stats(void * arg)2505  rcu_torture_stats(void *arg)
2506  {
2507  	VERBOSE_TOROUT_STRING("rcu_torture_stats task started");
2508  	do {
2509  		schedule_timeout_interruptible(stat_interval * HZ);
2510  		rcu_torture_stats_print();
2511  		torture_shutdown_absorb("rcu_torture_stats");
2512  	} while (!torture_must_stop());
2513  	torture_kthread_stopping("rcu_torture_stats");
2514  	return 0;
2515  }
2516  
2517  /* Test mem_dump_obj() and friends.  */
rcu_torture_mem_dump_obj(void)2518  static void rcu_torture_mem_dump_obj(void)
2519  {
2520  	struct rcu_head *rhp;
2521  	struct kmem_cache *kcp;
2522  	static int z;
2523  
2524  	kcp = kmem_cache_create("rcuscale", 136, 8, SLAB_STORE_USER, NULL);
2525  	if (WARN_ON_ONCE(!kcp))
2526  		return;
2527  	rhp = kmem_cache_alloc(kcp, GFP_KERNEL);
2528  	if (WARN_ON_ONCE(!rhp)) {
2529  		kmem_cache_destroy(kcp);
2530  		return;
2531  	}
2532  	pr_alert("mem_dump_obj() slab test: rcu_torture_stats = %px, &rhp = %px, rhp = %px, &z = %px\n", stats_task, &rhp, rhp, &z);
2533  	pr_alert("mem_dump_obj(ZERO_SIZE_PTR):");
2534  	mem_dump_obj(ZERO_SIZE_PTR);
2535  	pr_alert("mem_dump_obj(NULL):");
2536  	mem_dump_obj(NULL);
2537  	pr_alert("mem_dump_obj(%px):", &rhp);
2538  	mem_dump_obj(&rhp);
2539  	pr_alert("mem_dump_obj(%px):", rhp);
2540  	mem_dump_obj(rhp);
2541  	pr_alert("mem_dump_obj(%px):", &rhp->func);
2542  	mem_dump_obj(&rhp->func);
2543  	pr_alert("mem_dump_obj(%px):", &z);
2544  	mem_dump_obj(&z);
2545  	kmem_cache_free(kcp, rhp);
2546  	kmem_cache_destroy(kcp);
2547  	rhp = kmalloc(sizeof(*rhp), GFP_KERNEL);
2548  	if (WARN_ON_ONCE(!rhp))
2549  		return;
2550  	pr_alert("mem_dump_obj() kmalloc test: rcu_torture_stats = %px, &rhp = %px, rhp = %px\n", stats_task, &rhp, rhp);
2551  	pr_alert("mem_dump_obj(kmalloc %px):", rhp);
2552  	mem_dump_obj(rhp);
2553  	pr_alert("mem_dump_obj(kmalloc %px):", &rhp->func);
2554  	mem_dump_obj(&rhp->func);
2555  	kfree(rhp);
2556  	rhp = vmalloc(4096);
2557  	if (WARN_ON_ONCE(!rhp))
2558  		return;
2559  	pr_alert("mem_dump_obj() vmalloc test: rcu_torture_stats = %px, &rhp = %px, rhp = %px\n", stats_task, &rhp, rhp);
2560  	pr_alert("mem_dump_obj(vmalloc %px):", rhp);
2561  	mem_dump_obj(rhp);
2562  	pr_alert("mem_dump_obj(vmalloc %px):", &rhp->func);
2563  	mem_dump_obj(&rhp->func);
2564  	vfree(rhp);
2565  }
2566  
2567  static void
rcu_torture_print_module_parms(struct rcu_torture_ops * cur_ops,const char * tag)2568  rcu_torture_print_module_parms(struct rcu_torture_ops *cur_ops, const char *tag)
2569  {
2570  	pr_alert("%s" TORTURE_FLAG
2571  		 "--- %s: nreaders=%d nfakewriters=%d "
2572  		 "stat_interval=%d verbose=%d test_no_idle_hz=%d "
2573  		 "shuffle_interval=%d stutter=%d irqreader=%d "
2574  		 "fqs_duration=%d fqs_holdoff=%d fqs_stutter=%d "
2575  		 "test_boost=%d/%d test_boost_interval=%d "
2576  		 "test_boost_duration=%d test_boost_holdoff=%d shutdown_secs=%d "
2577  		 "stall_cpu=%d stall_cpu_holdoff=%d stall_cpu_irqsoff=%d "
2578  		 "stall_cpu_block=%d stall_cpu_repeat=%d "
2579  		 "n_barrier_cbs=%d "
2580  		 "onoff_interval=%d onoff_holdoff=%d "
2581  		 "read_exit_delay=%d read_exit_burst=%d "
2582  		 "reader_flavor=%x "
2583  		 "nocbs_nthreads=%d nocbs_toggle=%d "
2584  		 "test_nmis=%d "
2585  		 "preempt_duration=%d preempt_interval=%d\n",
2586  		 torture_type, tag, nrealreaders, nrealfakewriters,
2587  		 stat_interval, verbose, test_no_idle_hz, shuffle_interval,
2588  		 stutter, irqreader, fqs_duration, fqs_holdoff, fqs_stutter,
2589  		 test_boost, cur_ops->can_boost,
2590  		 test_boost_interval, test_boost_duration, test_boost_holdoff, shutdown_secs,
2591  		 stall_cpu, stall_cpu_holdoff, stall_cpu_irqsoff,
2592  		 stall_cpu_block, stall_cpu_repeat,
2593  		 n_barrier_cbs,
2594  		 onoff_interval, onoff_holdoff,
2595  		 read_exit_delay, read_exit_burst,
2596  		 reader_flavor,
2597  		 nocbs_nthreads, nocbs_toggle,
2598  		 test_nmis,
2599  		 preempt_duration, preempt_interval);
2600  }
2601  
rcutorture_booster_cleanup(unsigned int cpu)2602  static int rcutorture_booster_cleanup(unsigned int cpu)
2603  {
2604  	struct task_struct *t;
2605  
2606  	if (boost_tasks[cpu] == NULL)
2607  		return 0;
2608  	mutex_lock(&boost_mutex);
2609  	t = boost_tasks[cpu];
2610  	boost_tasks[cpu] = NULL;
2611  	rcu_torture_enable_rt_throttle();
2612  	mutex_unlock(&boost_mutex);
2613  
2614  	/* This must be outside of the mutex, otherwise deadlock! */
2615  	torture_stop_kthread(rcu_torture_boost, t);
2616  	return 0;
2617  }
2618  
rcutorture_booster_init(unsigned int cpu)2619  static int rcutorture_booster_init(unsigned int cpu)
2620  {
2621  	int retval;
2622  
2623  	if (boost_tasks[cpu] != NULL)
2624  		return 0;  /* Already created, nothing more to do. */
2625  
2626  	// Testing RCU priority boosting requires rcutorture do
2627  	// some serious abuse.  Counter this by running ksoftirqd
2628  	// at higher priority.
2629  	if (IS_BUILTIN(CONFIG_RCU_TORTURE_TEST)) {
2630  		struct sched_param sp;
2631  		struct task_struct *t;
2632  
2633  		t = per_cpu(ksoftirqd, cpu);
2634  		WARN_ON_ONCE(!t);
2635  		sp.sched_priority = 2;
2636  		sched_setscheduler_nocheck(t, SCHED_FIFO, &sp);
2637  #ifdef CONFIG_IRQ_FORCED_THREADING
2638  		if (force_irqthreads()) {
2639  			t = per_cpu(ktimerd, cpu);
2640  			WARN_ON_ONCE(!t);
2641  			sp.sched_priority = 2;
2642  			sched_setscheduler_nocheck(t, SCHED_FIFO, &sp);
2643  		}
2644  #endif
2645  	}
2646  
2647  	/* Don't allow time recalculation while creating a new task. */
2648  	mutex_lock(&boost_mutex);
2649  	rcu_torture_disable_rt_throttle();
2650  	VERBOSE_TOROUT_STRING("Creating rcu_torture_boost task");
2651  	boost_tasks[cpu] = kthread_run_on_cpu(rcu_torture_boost, NULL,
2652  					      cpu, "rcu_torture_boost_%u");
2653  	if (IS_ERR(boost_tasks[cpu])) {
2654  		retval = PTR_ERR(boost_tasks[cpu]);
2655  		VERBOSE_TOROUT_STRING("rcu_torture_boost task create failed");
2656  		n_rcu_torture_boost_ktrerror++;
2657  		boost_tasks[cpu] = NULL;
2658  		mutex_unlock(&boost_mutex);
2659  		return retval;
2660  	}
2661  	mutex_unlock(&boost_mutex);
2662  	return 0;
2663  }
2664  
rcu_torture_stall_nf(struct notifier_block * nb,unsigned long v,void * ptr)2665  static int rcu_torture_stall_nf(struct notifier_block *nb, unsigned long v, void *ptr)
2666  {
2667  	pr_info("%s: v=%lu, duration=%lu.\n", __func__, v, (unsigned long)ptr);
2668  	return NOTIFY_OK;
2669  }
2670  
2671  static struct notifier_block rcu_torture_stall_block = {
2672  	.notifier_call = rcu_torture_stall_nf,
2673  };
2674  
2675  /*
2676   * CPU-stall kthread.  It waits as specified by stall_cpu_holdoff, then
2677   * induces a CPU stall for the time specified by stall_cpu.  If a new
2678   * stall test is added, stallsdone in rcu_torture_writer() must be adjusted.
2679   */
rcu_torture_stall_one(int rep,int irqsoff)2680  static void rcu_torture_stall_one(int rep, int irqsoff)
2681  {
2682  	int idx;
2683  	unsigned long stop_at;
2684  
2685  	if (stall_cpu_holdoff > 0) {
2686  		VERBOSE_TOROUT_STRING("rcu_torture_stall begin holdoff");
2687  		schedule_timeout_interruptible(stall_cpu_holdoff * HZ);
2688  		VERBOSE_TOROUT_STRING("rcu_torture_stall end holdoff");
2689  	}
2690  	if (!kthread_should_stop() && stall_gp_kthread > 0) {
2691  		VERBOSE_TOROUT_STRING("rcu_torture_stall begin GP stall");
2692  		rcu_gp_set_torture_wait(stall_gp_kthread * HZ);
2693  		for (idx = 0; idx < stall_gp_kthread + 2; idx++) {
2694  			if (kthread_should_stop())
2695  				break;
2696  			schedule_timeout_uninterruptible(HZ);
2697  		}
2698  	}
2699  	if (!kthread_should_stop() && stall_cpu > 0) {
2700  		VERBOSE_TOROUT_STRING("rcu_torture_stall begin CPU stall");
2701  		stop_at = ktime_get_seconds() + stall_cpu;
2702  		/* RCU CPU stall is expected behavior in following code. */
2703  		idx = cur_ops->readlock();
2704  		if (irqsoff)
2705  			local_irq_disable();
2706  		else if (!stall_cpu_block)
2707  			preempt_disable();
2708  		pr_alert("%s start stall episode %d on CPU %d.\n",
2709  			  __func__, rep + 1, raw_smp_processor_id());
2710  		while (ULONG_CMP_LT((unsigned long)ktime_get_seconds(), stop_at) &&
2711  		       !kthread_should_stop())
2712  			if (stall_cpu_block) {
2713  #ifdef CONFIG_PREEMPTION
2714  				preempt_schedule();
2715  #else
2716  				schedule_timeout_uninterruptible(HZ);
2717  #endif
2718  			} else if (stall_no_softlockup) {
2719  				touch_softlockup_watchdog();
2720  			}
2721  		if (irqsoff)
2722  			local_irq_enable();
2723  		else if (!stall_cpu_block)
2724  			preempt_enable();
2725  		cur_ops->readunlock(idx);
2726  	}
2727  }
2728  
2729  /*
2730   * CPU-stall kthread.  Invokes rcu_torture_stall_one() once, and then as many
2731   * additional times as specified by the stall_cpu_repeat module parameter.
2732   * Note that stall_cpu_irqsoff is ignored on the second and subsequent
2733   * stall.
2734   */
rcu_torture_stall(void * args)2735  static int rcu_torture_stall(void *args)
2736  {
2737  	int i;
2738  	int repeat = stall_cpu_repeat;
2739  	int ret;
2740  
2741  	VERBOSE_TOROUT_STRING("rcu_torture_stall task started");
2742  	if (repeat < 0) {
2743  		repeat = 0;
2744  		WARN_ON_ONCE(IS_BUILTIN(CONFIG_RCU_TORTURE_TEST));
2745  	}
2746  	if (rcu_cpu_stall_notifiers) {
2747  		ret = rcu_stall_chain_notifier_register(&rcu_torture_stall_block);
2748  		if (ret)
2749  			pr_info("%s: rcu_stall_chain_notifier_register() returned %d, %sexpected.\n",
2750  				__func__, ret, !IS_ENABLED(CONFIG_RCU_STALL_COMMON) ? "un" : "");
2751  	}
2752  	for (i = 0; i <= repeat; i++) {
2753  		if (kthread_should_stop())
2754  			break;
2755  		rcu_torture_stall_one(i, i == 0 ? stall_cpu_irqsoff : 0);
2756  	}
2757  	pr_alert("%s end.\n", __func__);
2758  	if (rcu_cpu_stall_notifiers && !ret) {
2759  		ret = rcu_stall_chain_notifier_unregister(&rcu_torture_stall_block);
2760  		if (ret)
2761  			pr_info("%s: rcu_stall_chain_notifier_unregister() returned %d.\n", __func__, ret);
2762  	}
2763  	torture_shutdown_absorb("rcu_torture_stall");
2764  	while (!kthread_should_stop())
2765  		schedule_timeout_interruptible(10 * HZ);
2766  	return 0;
2767  }
2768  
2769  /* Spawn CPU-stall kthread, if stall_cpu specified. */
rcu_torture_stall_init(void)2770  static int __init rcu_torture_stall_init(void)
2771  {
2772  	if (stall_cpu <= 0 && stall_gp_kthread <= 0)
2773  		return 0;
2774  	return torture_create_kthread(rcu_torture_stall, NULL, stall_task);
2775  }
2776  
2777  /* State structure for forward-progress self-propagating RCU callback. */
2778  struct fwd_cb_state {
2779  	struct rcu_head rh;
2780  	int stop;
2781  };
2782  
2783  /*
2784   * Forward-progress self-propagating RCU callback function.  Because
2785   * callbacks run from softirq, this function is an implicit RCU read-side
2786   * critical section.
2787   */
rcu_torture_fwd_prog_cb(struct rcu_head * rhp)2788  static void rcu_torture_fwd_prog_cb(struct rcu_head *rhp)
2789  {
2790  	struct fwd_cb_state *fcsp = container_of(rhp, struct fwd_cb_state, rh);
2791  
2792  	if (READ_ONCE(fcsp->stop)) {
2793  		WRITE_ONCE(fcsp->stop, 2);
2794  		return;
2795  	}
2796  	cur_ops->call(&fcsp->rh, rcu_torture_fwd_prog_cb);
2797  }
2798  
2799  /* State for continuous-flood RCU callbacks. */
2800  struct rcu_fwd_cb {
2801  	struct rcu_head rh;
2802  	struct rcu_fwd_cb *rfc_next;
2803  	struct rcu_fwd *rfc_rfp;
2804  	int rfc_gps;
2805  };
2806  
2807  #define MAX_FWD_CB_JIFFIES	(8 * HZ) /* Maximum CB test duration. */
2808  #define MIN_FWD_CB_LAUNDERS	3	/* This many CB invocations to count. */
2809  #define MIN_FWD_CBS_LAUNDERED	100	/* Number of counted CBs. */
2810  #define FWD_CBS_HIST_DIV	10	/* Histogram buckets/second. */
2811  #define N_LAUNDERS_HIST (2 * MAX_FWD_CB_JIFFIES / (HZ / FWD_CBS_HIST_DIV))
2812  
2813  struct rcu_launder_hist {
2814  	long n_launders;
2815  	unsigned long launder_gp_seq;
2816  };
2817  
2818  struct rcu_fwd {
2819  	spinlock_t rcu_fwd_lock;
2820  	struct rcu_fwd_cb *rcu_fwd_cb_head;
2821  	struct rcu_fwd_cb **rcu_fwd_cb_tail;
2822  	long n_launders_cb;
2823  	unsigned long rcu_fwd_startat;
2824  	struct rcu_launder_hist n_launders_hist[N_LAUNDERS_HIST];
2825  	unsigned long rcu_launder_gp_seq_start;
2826  	int rcu_fwd_id;
2827  };
2828  
2829  static DEFINE_MUTEX(rcu_fwd_mutex);
2830  static struct rcu_fwd *rcu_fwds;
2831  static unsigned long rcu_fwd_seq;
2832  static atomic_long_t rcu_fwd_max_cbs;
2833  static bool rcu_fwd_emergency_stop;
2834  
rcu_torture_fwd_cb_hist(struct rcu_fwd * rfp)2835  static void rcu_torture_fwd_cb_hist(struct rcu_fwd *rfp)
2836  {
2837  	unsigned long gps;
2838  	unsigned long gps_old;
2839  	int i;
2840  	int j;
2841  
2842  	for (i = ARRAY_SIZE(rfp->n_launders_hist) - 1; i > 0; i--)
2843  		if (rfp->n_launders_hist[i].n_launders > 0)
2844  			break;
2845  	pr_alert("%s: Callback-invocation histogram %d (duration %lu jiffies):",
2846  		 __func__, rfp->rcu_fwd_id, jiffies - rfp->rcu_fwd_startat);
2847  	gps_old = rfp->rcu_launder_gp_seq_start;
2848  	for (j = 0; j <= i; j++) {
2849  		gps = rfp->n_launders_hist[j].launder_gp_seq;
2850  		pr_cont(" %ds/%d: %ld:%ld",
2851  			j + 1, FWD_CBS_HIST_DIV,
2852  			rfp->n_launders_hist[j].n_launders,
2853  			rcutorture_seq_diff(gps, gps_old));
2854  		gps_old = gps;
2855  	}
2856  	pr_cont("\n");
2857  }
2858  
2859  /* Callback function for continuous-flood RCU callbacks. */
rcu_torture_fwd_cb_cr(struct rcu_head * rhp)2860  static void rcu_torture_fwd_cb_cr(struct rcu_head *rhp)
2861  {
2862  	unsigned long flags;
2863  	int i;
2864  	struct rcu_fwd_cb *rfcp = container_of(rhp, struct rcu_fwd_cb, rh);
2865  	struct rcu_fwd_cb **rfcpp;
2866  	struct rcu_fwd *rfp = rfcp->rfc_rfp;
2867  
2868  	rfcp->rfc_next = NULL;
2869  	rfcp->rfc_gps++;
2870  	spin_lock_irqsave(&rfp->rcu_fwd_lock, flags);
2871  	rfcpp = rfp->rcu_fwd_cb_tail;
2872  	rfp->rcu_fwd_cb_tail = &rfcp->rfc_next;
2873  	smp_store_release(rfcpp, rfcp);
2874  	WRITE_ONCE(rfp->n_launders_cb, rfp->n_launders_cb + 1);
2875  	i = ((jiffies - rfp->rcu_fwd_startat) / (HZ / FWD_CBS_HIST_DIV));
2876  	if (i >= ARRAY_SIZE(rfp->n_launders_hist))
2877  		i = ARRAY_SIZE(rfp->n_launders_hist) - 1;
2878  	rfp->n_launders_hist[i].n_launders++;
2879  	rfp->n_launders_hist[i].launder_gp_seq = cur_ops->get_gp_seq();
2880  	spin_unlock_irqrestore(&rfp->rcu_fwd_lock, flags);
2881  }
2882  
2883  // Give the scheduler a chance, even on nohz_full CPUs.
rcu_torture_fwd_prog_cond_resched(unsigned long iter)2884  static void rcu_torture_fwd_prog_cond_resched(unsigned long iter)
2885  {
2886  	if (IS_ENABLED(CONFIG_PREEMPTION) && IS_ENABLED(CONFIG_NO_HZ_FULL)) {
2887  		// Real call_rcu() floods hit userspace, so emulate that.
2888  		if (need_resched() || (iter & 0xfff))
2889  			schedule();
2890  		return;
2891  	}
2892  	// No userspace emulation: CB invocation throttles call_rcu()
2893  	cond_resched();
2894  }
2895  
2896  /*
2897   * Free all callbacks on the rcu_fwd_cb_head list, either because the
2898   * test is over or because we hit an OOM event.
2899   */
rcu_torture_fwd_prog_cbfree(struct rcu_fwd * rfp)2900  static unsigned long rcu_torture_fwd_prog_cbfree(struct rcu_fwd *rfp)
2901  {
2902  	unsigned long flags;
2903  	unsigned long freed = 0;
2904  	struct rcu_fwd_cb *rfcp;
2905  
2906  	for (;;) {
2907  		spin_lock_irqsave(&rfp->rcu_fwd_lock, flags);
2908  		rfcp = rfp->rcu_fwd_cb_head;
2909  		if (!rfcp) {
2910  			spin_unlock_irqrestore(&rfp->rcu_fwd_lock, flags);
2911  			break;
2912  		}
2913  		rfp->rcu_fwd_cb_head = rfcp->rfc_next;
2914  		if (!rfp->rcu_fwd_cb_head)
2915  			rfp->rcu_fwd_cb_tail = &rfp->rcu_fwd_cb_head;
2916  		spin_unlock_irqrestore(&rfp->rcu_fwd_lock, flags);
2917  		kfree(rfcp);
2918  		freed++;
2919  		rcu_torture_fwd_prog_cond_resched(freed);
2920  		if (tick_nohz_full_enabled()) {
2921  			local_irq_save(flags);
2922  			rcu_momentary_eqs();
2923  			local_irq_restore(flags);
2924  		}
2925  	}
2926  	return freed;
2927  }
2928  
2929  /* Carry out need_resched()/cond_resched() forward-progress testing. */
rcu_torture_fwd_prog_nr(struct rcu_fwd * rfp,int * tested,int * tested_tries)2930  static void rcu_torture_fwd_prog_nr(struct rcu_fwd *rfp,
2931  				    int *tested, int *tested_tries)
2932  {
2933  	unsigned long cver;
2934  	unsigned long dur;
2935  	struct fwd_cb_state fcs;
2936  	unsigned long gps;
2937  	int idx;
2938  	int sd;
2939  	int sd4;
2940  	bool selfpropcb = false;
2941  	unsigned long stopat;
2942  	static DEFINE_TORTURE_RANDOM(trs);
2943  
2944  	pr_alert("%s: Starting forward-progress test %d\n", __func__, rfp->rcu_fwd_id);
2945  	if (!cur_ops->sync)
2946  		return; // Cannot do need_resched() forward progress testing without ->sync.
2947  	if (cur_ops->call && cur_ops->cb_barrier) {
2948  		init_rcu_head_on_stack(&fcs.rh);
2949  		selfpropcb = true;
2950  	}
2951  
2952  	/* Tight loop containing cond_resched(). */
2953  	atomic_inc(&rcu_fwd_cb_nodelay);
2954  	cur_ops->sync(); /* Later readers see above write. */
2955  	if  (selfpropcb) {
2956  		WRITE_ONCE(fcs.stop, 0);
2957  		cur_ops->call(&fcs.rh, rcu_torture_fwd_prog_cb);
2958  	}
2959  	cver = READ_ONCE(rcu_torture_current_version);
2960  	gps = cur_ops->get_gp_seq();
2961  	sd = cur_ops->stall_dur() + 1;
2962  	sd4 = (sd + fwd_progress_div - 1) / fwd_progress_div;
2963  	dur = sd4 + torture_random(&trs) % (sd - sd4);
2964  	WRITE_ONCE(rfp->rcu_fwd_startat, jiffies);
2965  	stopat = rfp->rcu_fwd_startat + dur;
2966  	while (time_before(jiffies, stopat) &&
2967  	       !shutdown_time_arrived() &&
2968  	       !READ_ONCE(rcu_fwd_emergency_stop) && !torture_must_stop()) {
2969  		idx = cur_ops->readlock();
2970  		udelay(10);
2971  		cur_ops->readunlock(idx);
2972  		if (!fwd_progress_need_resched || need_resched())
2973  			cond_resched();
2974  	}
2975  	(*tested_tries)++;
2976  	if (!time_before(jiffies, stopat) &&
2977  	    !shutdown_time_arrived() &&
2978  	    !READ_ONCE(rcu_fwd_emergency_stop) && !torture_must_stop()) {
2979  		(*tested)++;
2980  		cver = READ_ONCE(rcu_torture_current_version) - cver;
2981  		gps = rcutorture_seq_diff(cur_ops->get_gp_seq(), gps);
2982  		WARN_ON(!cver && gps < 2);
2983  		pr_alert("%s: %d Duration %ld cver %ld gps %ld\n", __func__,
2984  			 rfp->rcu_fwd_id, dur, cver, gps);
2985  	}
2986  	if (selfpropcb) {
2987  		WRITE_ONCE(fcs.stop, 1);
2988  		cur_ops->sync(); /* Wait for running CB to complete. */
2989  		pr_alert("%s: Waiting for CBs: %pS() %d\n", __func__, cur_ops->cb_barrier, rfp->rcu_fwd_id);
2990  		cur_ops->cb_barrier(); /* Wait for queued callbacks. */
2991  	}
2992  
2993  	if (selfpropcb) {
2994  		WARN_ON(READ_ONCE(fcs.stop) != 2);
2995  		destroy_rcu_head_on_stack(&fcs.rh);
2996  	}
2997  	schedule_timeout_uninterruptible(HZ / 10); /* Let kthreads recover. */
2998  	atomic_dec(&rcu_fwd_cb_nodelay);
2999  }
3000  
3001  /* Carry out call_rcu() forward-progress testing. */
rcu_torture_fwd_prog_cr(struct rcu_fwd * rfp)3002  static void rcu_torture_fwd_prog_cr(struct rcu_fwd *rfp)
3003  {
3004  	unsigned long cver;
3005  	unsigned long flags;
3006  	unsigned long gps;
3007  	int i;
3008  	long n_launders;
3009  	long n_launders_cb_snap;
3010  	long n_launders_sa;
3011  	long n_max_cbs;
3012  	long n_max_gps;
3013  	struct rcu_fwd_cb *rfcp;
3014  	struct rcu_fwd_cb *rfcpn;
3015  	unsigned long stopat;
3016  	unsigned long stoppedat;
3017  
3018  	pr_alert("%s: Starting forward-progress test %d\n", __func__, rfp->rcu_fwd_id);
3019  	if (READ_ONCE(rcu_fwd_emergency_stop))
3020  		return; /* Get out of the way quickly, no GP wait! */
3021  	if (!cur_ops->call)
3022  		return; /* Can't do call_rcu() fwd prog without ->call. */
3023  
3024  	/* Loop continuously posting RCU callbacks. */
3025  	atomic_inc(&rcu_fwd_cb_nodelay);
3026  	cur_ops->sync(); /* Later readers see above write. */
3027  	WRITE_ONCE(rfp->rcu_fwd_startat, jiffies);
3028  	stopat = rfp->rcu_fwd_startat + MAX_FWD_CB_JIFFIES;
3029  	n_launders = 0;
3030  	rfp->n_launders_cb = 0; // Hoist initialization for multi-kthread
3031  	n_launders_sa = 0;
3032  	n_max_cbs = 0;
3033  	n_max_gps = 0;
3034  	for (i = 0; i < ARRAY_SIZE(rfp->n_launders_hist); i++)
3035  		rfp->n_launders_hist[i].n_launders = 0;
3036  	cver = READ_ONCE(rcu_torture_current_version);
3037  	gps = cur_ops->get_gp_seq();
3038  	rfp->rcu_launder_gp_seq_start = gps;
3039  	tick_dep_set_task(current, TICK_DEP_BIT_RCU);
3040  	while (time_before(jiffies, stopat) &&
3041  	       !shutdown_time_arrived() &&
3042  	       !READ_ONCE(rcu_fwd_emergency_stop) && !torture_must_stop()) {
3043  		rfcp = READ_ONCE(rfp->rcu_fwd_cb_head);
3044  		rfcpn = NULL;
3045  		if (rfcp)
3046  			rfcpn = READ_ONCE(rfcp->rfc_next);
3047  		if (rfcpn) {
3048  			if (rfcp->rfc_gps >= MIN_FWD_CB_LAUNDERS &&
3049  			    ++n_max_gps >= MIN_FWD_CBS_LAUNDERED)
3050  				break;
3051  			rfp->rcu_fwd_cb_head = rfcpn;
3052  			n_launders++;
3053  			n_launders_sa++;
3054  		} else if (!cur_ops->cbflood_max || cur_ops->cbflood_max > n_max_cbs) {
3055  			rfcp = kmalloc(sizeof(*rfcp), GFP_KERNEL);
3056  			if (WARN_ON_ONCE(!rfcp)) {
3057  				schedule_timeout_interruptible(1);
3058  				continue;
3059  			}
3060  			n_max_cbs++;
3061  			n_launders_sa = 0;
3062  			rfcp->rfc_gps = 0;
3063  			rfcp->rfc_rfp = rfp;
3064  		} else {
3065  			rfcp = NULL;
3066  		}
3067  		if (rfcp)
3068  			cur_ops->call(&rfcp->rh, rcu_torture_fwd_cb_cr);
3069  		rcu_torture_fwd_prog_cond_resched(n_launders + n_max_cbs);
3070  		if (tick_nohz_full_enabled()) {
3071  			local_irq_save(flags);
3072  			rcu_momentary_eqs();
3073  			local_irq_restore(flags);
3074  		}
3075  	}
3076  	stoppedat = jiffies;
3077  	n_launders_cb_snap = READ_ONCE(rfp->n_launders_cb);
3078  	cver = READ_ONCE(rcu_torture_current_version) - cver;
3079  	gps = rcutorture_seq_diff(cur_ops->get_gp_seq(), gps);
3080  	pr_alert("%s: Waiting for CBs: %pS() %d\n", __func__, cur_ops->cb_barrier, rfp->rcu_fwd_id);
3081  	cur_ops->cb_barrier(); /* Wait for callbacks to be invoked. */
3082  	(void)rcu_torture_fwd_prog_cbfree(rfp);
3083  
3084  	if (!torture_must_stop() && !READ_ONCE(rcu_fwd_emergency_stop) &&
3085  	    !shutdown_time_arrived()) {
3086  		if (WARN_ON(n_max_gps < MIN_FWD_CBS_LAUNDERED) && cur_ops->gp_kthread_dbg)
3087  			cur_ops->gp_kthread_dbg();
3088  		pr_alert("%s Duration %lu barrier: %lu pending %ld n_launders: %ld n_launders_sa: %ld n_max_gps: %ld n_max_cbs: %ld cver %ld gps %ld #online %u\n",
3089  			 __func__,
3090  			 stoppedat - rfp->rcu_fwd_startat, jiffies - stoppedat,
3091  			 n_launders + n_max_cbs - n_launders_cb_snap,
3092  			 n_launders, n_launders_sa,
3093  			 n_max_gps, n_max_cbs, cver, gps, num_online_cpus());
3094  		atomic_long_add(n_max_cbs, &rcu_fwd_max_cbs);
3095  		mutex_lock(&rcu_fwd_mutex); // Serialize histograms.
3096  		rcu_torture_fwd_cb_hist(rfp);
3097  		mutex_unlock(&rcu_fwd_mutex);
3098  	}
3099  	schedule_timeout_uninterruptible(HZ); /* Let CBs drain. */
3100  	tick_dep_clear_task(current, TICK_DEP_BIT_RCU);
3101  	atomic_dec(&rcu_fwd_cb_nodelay);
3102  }
3103  
3104  
3105  /*
3106   * OOM notifier, but this only prints diagnostic information for the
3107   * current forward-progress test.
3108   */
rcutorture_oom_notify(struct notifier_block * self,unsigned long notused,void * nfreed)3109  static int rcutorture_oom_notify(struct notifier_block *self,
3110  				 unsigned long notused, void *nfreed)
3111  {
3112  	int i;
3113  	long ncbs;
3114  	struct rcu_fwd *rfp;
3115  
3116  	mutex_lock(&rcu_fwd_mutex);
3117  	rfp = rcu_fwds;
3118  	if (!rfp) {
3119  		mutex_unlock(&rcu_fwd_mutex);
3120  		return NOTIFY_OK;
3121  	}
3122  	WARN(1, "%s invoked upon OOM during forward-progress testing.\n",
3123  	     __func__);
3124  	for (i = 0; i < fwd_progress; i++) {
3125  		rcu_torture_fwd_cb_hist(&rfp[i]);
3126  		rcu_fwd_progress_check(1 + (jiffies - READ_ONCE(rfp[i].rcu_fwd_startat)) / 2);
3127  	}
3128  	WRITE_ONCE(rcu_fwd_emergency_stop, true);
3129  	smp_mb(); /* Emergency stop before free and wait to avoid hangs. */
3130  	ncbs = 0;
3131  	for (i = 0; i < fwd_progress; i++)
3132  		ncbs += rcu_torture_fwd_prog_cbfree(&rfp[i]);
3133  	pr_info("%s: Freed %lu RCU callbacks.\n", __func__, ncbs);
3134  	cur_ops->cb_barrier();
3135  	ncbs = 0;
3136  	for (i = 0; i < fwd_progress; i++)
3137  		ncbs += rcu_torture_fwd_prog_cbfree(&rfp[i]);
3138  	pr_info("%s: Freed %lu RCU callbacks.\n", __func__, ncbs);
3139  	cur_ops->cb_barrier();
3140  	ncbs = 0;
3141  	for (i = 0; i < fwd_progress; i++)
3142  		ncbs += rcu_torture_fwd_prog_cbfree(&rfp[i]);
3143  	pr_info("%s: Freed %lu RCU callbacks.\n", __func__, ncbs);
3144  	smp_mb(); /* Frees before return to avoid redoing OOM. */
3145  	(*(unsigned long *)nfreed)++; /* Forward progress CBs freed! */
3146  	pr_info("%s returning after OOM processing.\n", __func__);
3147  	mutex_unlock(&rcu_fwd_mutex);
3148  	return NOTIFY_OK;
3149  }
3150  
3151  static struct notifier_block rcutorture_oom_nb = {
3152  	.notifier_call = rcutorture_oom_notify
3153  };
3154  
3155  /* Carry out grace-period forward-progress testing. */
rcu_torture_fwd_prog(void * args)3156  static int rcu_torture_fwd_prog(void *args)
3157  {
3158  	bool firsttime = true;
3159  	long max_cbs;
3160  	int oldnice = task_nice(current);
3161  	unsigned long oldseq = READ_ONCE(rcu_fwd_seq);
3162  	struct rcu_fwd *rfp = args;
3163  	int tested = 0;
3164  	int tested_tries = 0;
3165  
3166  	VERBOSE_TOROUT_STRING("rcu_torture_fwd_progress task started");
3167  	rcu_bind_current_to_nocb();
3168  	if (!IS_ENABLED(CONFIG_SMP) || !IS_ENABLED(CONFIG_RCU_BOOST))
3169  		set_user_nice(current, MAX_NICE);
3170  	do {
3171  		if (!rfp->rcu_fwd_id) {
3172  			schedule_timeout_interruptible(fwd_progress_holdoff * HZ);
3173  			WRITE_ONCE(rcu_fwd_emergency_stop, false);
3174  			if (!firsttime) {
3175  				max_cbs = atomic_long_xchg(&rcu_fwd_max_cbs, 0);
3176  				pr_alert("%s n_max_cbs: %ld\n", __func__, max_cbs);
3177  			}
3178  			firsttime = false;
3179  			WRITE_ONCE(rcu_fwd_seq, rcu_fwd_seq + 1);
3180  		} else {
3181  			while (READ_ONCE(rcu_fwd_seq) == oldseq && !torture_must_stop())
3182  				schedule_timeout_interruptible(HZ / 20);
3183  			oldseq = READ_ONCE(rcu_fwd_seq);
3184  		}
3185  		pr_alert("%s: Starting forward-progress test %d\n", __func__, rfp->rcu_fwd_id);
3186  		if (rcu_inkernel_boot_has_ended() && torture_num_online_cpus() > rfp->rcu_fwd_id)
3187  			rcu_torture_fwd_prog_cr(rfp);
3188  		if ((cur_ops->stall_dur && cur_ops->stall_dur() > 0) &&
3189  		    (!IS_ENABLED(CONFIG_TINY_RCU) ||
3190  		     (rcu_inkernel_boot_has_ended() &&
3191  		      torture_num_online_cpus() > rfp->rcu_fwd_id)))
3192  			rcu_torture_fwd_prog_nr(rfp, &tested, &tested_tries);
3193  
3194  		/* Avoid slow periods, better to test when busy. */
3195  		if (stutter_wait("rcu_torture_fwd_prog"))
3196  			sched_set_normal(current, oldnice);
3197  	} while (!torture_must_stop());
3198  	/* Short runs might not contain a valid forward-progress attempt. */
3199  	if (!rfp->rcu_fwd_id) {
3200  		WARN_ON(!tested && tested_tries >= 5);
3201  		pr_alert("%s: tested %d tested_tries %d\n", __func__, tested, tested_tries);
3202  	}
3203  	torture_kthread_stopping("rcu_torture_fwd_prog");
3204  	return 0;
3205  }
3206  
3207  /* If forward-progress checking is requested and feasible, spawn the thread. */
rcu_torture_fwd_prog_init(void)3208  static int __init rcu_torture_fwd_prog_init(void)
3209  {
3210  	int i;
3211  	int ret = 0;
3212  	struct rcu_fwd *rfp;
3213  
3214  	if (!fwd_progress)
3215  		return 0; /* Not requested, so don't do it. */
3216  	if (fwd_progress >= nr_cpu_ids) {
3217  		VERBOSE_TOROUT_STRING("rcu_torture_fwd_prog_init: Limiting fwd_progress to # CPUs.\n");
3218  		fwd_progress = nr_cpu_ids;
3219  	} else if (fwd_progress < 0) {
3220  		fwd_progress = nr_cpu_ids;
3221  	}
3222  	if ((!cur_ops->sync && !cur_ops->call) ||
3223  	    (!cur_ops->cbflood_max && (!cur_ops->stall_dur || cur_ops->stall_dur() <= 0)) ||
3224  	    cur_ops == &rcu_busted_ops) {
3225  		VERBOSE_TOROUT_STRING("rcu_torture_fwd_prog_init: Disabled, unsupported by RCU flavor under test");
3226  		fwd_progress = 0;
3227  		return 0;
3228  	}
3229  	if (stall_cpu > 0 || (preempt_duration > 0 && IS_ENABLED(CONFIG_RCU_NOCB_CPU))) {
3230  		VERBOSE_TOROUT_STRING("rcu_torture_fwd_prog_init: Disabled, conflicts with CPU-stall and/or preemption testing");
3231  		fwd_progress = 0;
3232  		if (IS_MODULE(CONFIG_RCU_TORTURE_TEST))
3233  			return -EINVAL; /* In module, can fail back to user. */
3234  		WARN_ON(1); /* Make sure rcutorture scripting notices conflict. */
3235  		return 0;
3236  	}
3237  	if (fwd_progress_holdoff <= 0)
3238  		fwd_progress_holdoff = 1;
3239  	if (fwd_progress_div <= 0)
3240  		fwd_progress_div = 4;
3241  	rfp = kcalloc(fwd_progress, sizeof(*rfp), GFP_KERNEL);
3242  	fwd_prog_tasks = kcalloc(fwd_progress, sizeof(*fwd_prog_tasks), GFP_KERNEL);
3243  	if (!rfp || !fwd_prog_tasks) {
3244  		kfree(rfp);
3245  		kfree(fwd_prog_tasks);
3246  		fwd_prog_tasks = NULL;
3247  		fwd_progress = 0;
3248  		return -ENOMEM;
3249  	}
3250  	for (i = 0; i < fwd_progress; i++) {
3251  		spin_lock_init(&rfp[i].rcu_fwd_lock);
3252  		rfp[i].rcu_fwd_cb_tail = &rfp[i].rcu_fwd_cb_head;
3253  		rfp[i].rcu_fwd_id = i;
3254  	}
3255  	mutex_lock(&rcu_fwd_mutex);
3256  	rcu_fwds = rfp;
3257  	mutex_unlock(&rcu_fwd_mutex);
3258  	register_oom_notifier(&rcutorture_oom_nb);
3259  	for (i = 0; i < fwd_progress; i++) {
3260  		ret = torture_create_kthread(rcu_torture_fwd_prog, &rcu_fwds[i], fwd_prog_tasks[i]);
3261  		if (ret) {
3262  			fwd_progress = i;
3263  			return ret;
3264  		}
3265  	}
3266  	return 0;
3267  }
3268  
rcu_torture_fwd_prog_cleanup(void)3269  static void rcu_torture_fwd_prog_cleanup(void)
3270  {
3271  	int i;
3272  	struct rcu_fwd *rfp;
3273  
3274  	if (!rcu_fwds || !fwd_prog_tasks)
3275  		return;
3276  	for (i = 0; i < fwd_progress; i++)
3277  		torture_stop_kthread(rcu_torture_fwd_prog, fwd_prog_tasks[i]);
3278  	unregister_oom_notifier(&rcutorture_oom_nb);
3279  	mutex_lock(&rcu_fwd_mutex);
3280  	rfp = rcu_fwds;
3281  	rcu_fwds = NULL;
3282  	mutex_unlock(&rcu_fwd_mutex);
3283  	kfree(rfp);
3284  	kfree(fwd_prog_tasks);
3285  	fwd_prog_tasks = NULL;
3286  }
3287  
3288  /* Callback function for RCU barrier testing. */
rcu_torture_barrier_cbf(struct rcu_head * rcu)3289  static void rcu_torture_barrier_cbf(struct rcu_head *rcu)
3290  {
3291  	atomic_inc(&barrier_cbs_invoked);
3292  }
3293  
3294  /* IPI handler to get callback posted on desired CPU, if online. */
rcu_torture_barrier1cb(void * rcu_void)3295  static int rcu_torture_barrier1cb(void *rcu_void)
3296  {
3297  	struct rcu_head *rhp = rcu_void;
3298  
3299  	cur_ops->call(rhp, rcu_torture_barrier_cbf);
3300  	return 0;
3301  }
3302  
3303  /* kthread function to register callbacks used to test RCU barriers. */
rcu_torture_barrier_cbs(void * arg)3304  static int rcu_torture_barrier_cbs(void *arg)
3305  {
3306  	long myid = (long)arg;
3307  	bool lastphase = false;
3308  	bool newphase;
3309  	struct rcu_head rcu;
3310  
3311  	init_rcu_head_on_stack(&rcu);
3312  	VERBOSE_TOROUT_STRING("rcu_torture_barrier_cbs task started");
3313  	set_user_nice(current, MAX_NICE);
3314  	do {
3315  		wait_event(barrier_cbs_wq[myid],
3316  			   (newphase =
3317  			    smp_load_acquire(&barrier_phase)) != lastphase ||
3318  			   torture_must_stop());
3319  		lastphase = newphase;
3320  		if (torture_must_stop())
3321  			break;
3322  		/*
3323  		 * The above smp_load_acquire() ensures barrier_phase load
3324  		 * is ordered before the following ->call().
3325  		 */
3326  		if (smp_call_on_cpu(myid, rcu_torture_barrier1cb, &rcu, 1))
3327  			cur_ops->call(&rcu, rcu_torture_barrier_cbf);
3328  
3329  		if (atomic_dec_and_test(&barrier_cbs_count))
3330  			wake_up(&barrier_wq);
3331  	} while (!torture_must_stop());
3332  	if (cur_ops->cb_barrier != NULL)
3333  		cur_ops->cb_barrier();
3334  	destroy_rcu_head_on_stack(&rcu);
3335  	torture_kthread_stopping("rcu_torture_barrier_cbs");
3336  	return 0;
3337  }
3338  
3339  /* kthread function to drive and coordinate RCU barrier testing. */
rcu_torture_barrier(void * arg)3340  static int rcu_torture_barrier(void *arg)
3341  {
3342  	int i;
3343  
3344  	VERBOSE_TOROUT_STRING("rcu_torture_barrier task starting");
3345  	do {
3346  		atomic_set(&barrier_cbs_invoked, 0);
3347  		atomic_set(&barrier_cbs_count, n_barrier_cbs);
3348  		/* Ensure barrier_phase ordered after prior assignments. */
3349  		smp_store_release(&barrier_phase, !barrier_phase);
3350  		for (i = 0; i < n_barrier_cbs; i++)
3351  			wake_up(&barrier_cbs_wq[i]);
3352  		wait_event(barrier_wq,
3353  			   atomic_read(&barrier_cbs_count) == 0 ||
3354  			   torture_must_stop());
3355  		if (torture_must_stop())
3356  			break;
3357  		n_barrier_attempts++;
3358  		cur_ops->cb_barrier(); /* Implies smp_mb() for wait_event(). */
3359  		if (atomic_read(&barrier_cbs_invoked) != n_barrier_cbs) {
3360  			n_rcu_torture_barrier_error++;
3361  			pr_err("barrier_cbs_invoked = %d, n_barrier_cbs = %d\n",
3362  			       atomic_read(&barrier_cbs_invoked),
3363  			       n_barrier_cbs);
3364  			WARN_ON(1);
3365  			// Wait manually for the remaining callbacks
3366  			i = 0;
3367  			do {
3368  				if (WARN_ON(i++ > HZ))
3369  					i = INT_MIN;
3370  				schedule_timeout_interruptible(1);
3371  				cur_ops->cb_barrier();
3372  			} while (atomic_read(&barrier_cbs_invoked) !=
3373  				 n_barrier_cbs &&
3374  				 !torture_must_stop());
3375  			smp_mb(); // Can't trust ordering if broken.
3376  			if (!torture_must_stop())
3377  				pr_err("Recovered: barrier_cbs_invoked = %d\n",
3378  				       atomic_read(&barrier_cbs_invoked));
3379  		} else {
3380  			n_barrier_successes++;
3381  		}
3382  		schedule_timeout_interruptible(HZ / 10);
3383  	} while (!torture_must_stop());
3384  	torture_kthread_stopping("rcu_torture_barrier");
3385  	return 0;
3386  }
3387  
3388  /* Initialize RCU barrier testing. */
rcu_torture_barrier_init(void)3389  static int rcu_torture_barrier_init(void)
3390  {
3391  	int i;
3392  	int ret;
3393  
3394  	if (n_barrier_cbs <= 0)
3395  		return 0;
3396  	if (cur_ops->call == NULL || cur_ops->cb_barrier == NULL) {
3397  		pr_alert("%s" TORTURE_FLAG
3398  			 " Call or barrier ops missing for %s,\n",
3399  			 torture_type, cur_ops->name);
3400  		pr_alert("%s" TORTURE_FLAG
3401  			 " RCU barrier testing omitted from run.\n",
3402  			 torture_type);
3403  		return 0;
3404  	}
3405  	atomic_set(&barrier_cbs_count, 0);
3406  	atomic_set(&barrier_cbs_invoked, 0);
3407  	barrier_cbs_tasks =
3408  		kcalloc(n_barrier_cbs, sizeof(barrier_cbs_tasks[0]),
3409  			GFP_KERNEL);
3410  	barrier_cbs_wq =
3411  		kcalloc(n_barrier_cbs, sizeof(barrier_cbs_wq[0]), GFP_KERNEL);
3412  	if (barrier_cbs_tasks == NULL || !barrier_cbs_wq)
3413  		return -ENOMEM;
3414  	for (i = 0; i < n_barrier_cbs; i++) {
3415  		init_waitqueue_head(&barrier_cbs_wq[i]);
3416  		ret = torture_create_kthread(rcu_torture_barrier_cbs,
3417  					     (void *)(long)i,
3418  					     barrier_cbs_tasks[i]);
3419  		if (ret)
3420  			return ret;
3421  	}
3422  	return torture_create_kthread(rcu_torture_barrier, NULL, barrier_task);
3423  }
3424  
3425  /* Clean up after RCU barrier testing. */
rcu_torture_barrier_cleanup(void)3426  static void rcu_torture_barrier_cleanup(void)
3427  {
3428  	int i;
3429  
3430  	torture_stop_kthread(rcu_torture_barrier, barrier_task);
3431  	if (barrier_cbs_tasks != NULL) {
3432  		for (i = 0; i < n_barrier_cbs; i++)
3433  			torture_stop_kthread(rcu_torture_barrier_cbs,
3434  					     barrier_cbs_tasks[i]);
3435  		kfree(barrier_cbs_tasks);
3436  		barrier_cbs_tasks = NULL;
3437  	}
3438  	if (barrier_cbs_wq != NULL) {
3439  		kfree(barrier_cbs_wq);
3440  		barrier_cbs_wq = NULL;
3441  	}
3442  }
3443  
rcu_torture_can_boost(void)3444  static bool rcu_torture_can_boost(void)
3445  {
3446  	static int boost_warn_once;
3447  	int prio;
3448  
3449  	if (!(test_boost == 1 && cur_ops->can_boost) && test_boost != 2)
3450  		return false;
3451  	if (!cur_ops->start_gp_poll || !cur_ops->poll_gp_state)
3452  		return false;
3453  
3454  	prio = rcu_get_gp_kthreads_prio();
3455  	if (!prio)
3456  		return false;
3457  
3458  	if (prio < 2) {
3459  		if (boost_warn_once == 1)
3460  			return false;
3461  
3462  		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);
3463  		boost_warn_once = 1;
3464  		return false;
3465  	}
3466  
3467  	return true;
3468  }
3469  
3470  static bool read_exit_child_stop;
3471  static bool read_exit_child_stopped;
3472  static wait_queue_head_t read_exit_wq;
3473  
3474  // Child kthread which just does an rcutorture reader and exits.
rcu_torture_read_exit_child(void * trsp_in)3475  static int rcu_torture_read_exit_child(void *trsp_in)
3476  {
3477  	struct torture_random_state *trsp = trsp_in;
3478  
3479  	set_user_nice(current, MAX_NICE);
3480  	// Minimize time between reading and exiting.
3481  	while (!kthread_should_stop())
3482  		schedule_timeout_uninterruptible(HZ / 20);
3483  	(void)rcu_torture_one_read(trsp, -1);
3484  	return 0;
3485  }
3486  
3487  // Parent kthread which creates and destroys read-exit child kthreads.
rcu_torture_read_exit(void * unused)3488  static int rcu_torture_read_exit(void *unused)
3489  {
3490  	bool errexit = false;
3491  	int i;
3492  	struct task_struct *tsp;
3493  	DEFINE_TORTURE_RANDOM(trs);
3494  
3495  	// Allocate and initialize.
3496  	set_user_nice(current, MAX_NICE);
3497  	VERBOSE_TOROUT_STRING("rcu_torture_read_exit: Start of test");
3498  
3499  	// Each pass through this loop does one read-exit episode.
3500  	do {
3501  		VERBOSE_TOROUT_STRING("rcu_torture_read_exit: Start of episode");
3502  		for (i = 0; i < read_exit_burst; i++) {
3503  			if (READ_ONCE(read_exit_child_stop))
3504  				break;
3505  			stutter_wait("rcu_torture_read_exit");
3506  			// Spawn child.
3507  			tsp = kthread_run(rcu_torture_read_exit_child,
3508  					  &trs, "%s", "rcu_torture_read_exit_child");
3509  			if (IS_ERR(tsp)) {
3510  				TOROUT_ERRSTRING("out of memory");
3511  				errexit = true;
3512  				break;
3513  			}
3514  			cond_resched();
3515  			kthread_stop(tsp);
3516  			n_read_exits++;
3517  		}
3518  		VERBOSE_TOROUT_STRING("rcu_torture_read_exit: End of episode");
3519  		rcu_barrier(); // Wait for task_struct free, avoid OOM.
3520  		i = 0;
3521  		for (; !errexit && !READ_ONCE(read_exit_child_stop) && i < read_exit_delay; i++)
3522  			schedule_timeout_uninterruptible(HZ);
3523  	} while (!errexit && !READ_ONCE(read_exit_child_stop));
3524  
3525  	// Clean up and exit.
3526  	smp_store_release(&read_exit_child_stopped, true); // After reaping.
3527  	smp_mb(); // Store before wakeup.
3528  	wake_up(&read_exit_wq);
3529  	while (!torture_must_stop())
3530  		schedule_timeout_uninterruptible(HZ / 20);
3531  	torture_kthread_stopping("rcu_torture_read_exit");
3532  	return 0;
3533  }
3534  
rcu_torture_read_exit_init(void)3535  static int rcu_torture_read_exit_init(void)
3536  {
3537  	if (read_exit_burst <= 0)
3538  		return 0;
3539  	init_waitqueue_head(&read_exit_wq);
3540  	read_exit_child_stop = false;
3541  	read_exit_child_stopped = false;
3542  	return torture_create_kthread(rcu_torture_read_exit, NULL,
3543  				      read_exit_task);
3544  }
3545  
rcu_torture_read_exit_cleanup(void)3546  static void rcu_torture_read_exit_cleanup(void)
3547  {
3548  	if (!read_exit_task)
3549  		return;
3550  	WRITE_ONCE(read_exit_child_stop, true);
3551  	smp_mb(); // Above write before wait.
3552  	wait_event(read_exit_wq, smp_load_acquire(&read_exit_child_stopped));
3553  	torture_stop_kthread(rcutorture_read_exit, read_exit_task);
3554  }
3555  
rcutorture_test_nmis(int n)3556  static void rcutorture_test_nmis(int n)
3557  {
3558  #if IS_BUILTIN(CONFIG_RCU_TORTURE_TEST)
3559  	int cpu;
3560  	int dumpcpu;
3561  	int i;
3562  
3563  	for (i = 0; i < n; i++) {
3564  		preempt_disable();
3565  		cpu = smp_processor_id();
3566  		dumpcpu = cpu + 1;
3567  		if (dumpcpu >= nr_cpu_ids)
3568  			dumpcpu = 0;
3569  		pr_alert("%s: CPU %d invoking dump_cpu_task(%d)\n", __func__, cpu, dumpcpu);
3570  		dump_cpu_task(dumpcpu);
3571  		preempt_enable();
3572  		schedule_timeout_uninterruptible(15 * HZ);
3573  	}
3574  #else // #if IS_BUILTIN(CONFIG_RCU_TORTURE_TEST)
3575  	WARN_ONCE(n, "Non-zero rcutorture.test_nmis=%d permitted only when rcutorture is built in.\n", test_nmis);
3576  #endif // #else // #if IS_BUILTIN(CONFIG_RCU_TORTURE_TEST)
3577  }
3578  
3579  // Randomly preempt online CPUs.
rcu_torture_preempt(void * unused)3580  static int rcu_torture_preempt(void *unused)
3581  {
3582  	int cpu = -1;
3583  	DEFINE_TORTURE_RANDOM(rand);
3584  
3585  	schedule_timeout_idle(stall_cpu_holdoff);
3586  	do {
3587  		// Wait for preempt_interval ms with up to 100us fuzz.
3588  		torture_hrtimeout_ms(preempt_interval, 100, &rand);
3589  		// Select online CPU.
3590  		cpu = cpumask_next(cpu, cpu_online_mask);
3591  		if (cpu >= nr_cpu_ids)
3592  			cpu = cpumask_next(-1, cpu_online_mask);
3593  		WARN_ON_ONCE(cpu >= nr_cpu_ids);
3594  		// Move to that CPU, if can't do so, retry later.
3595  		if (torture_sched_setaffinity(current->pid, cpumask_of(cpu), false))
3596  			continue;
3597  		// Preempt at high-ish priority, then reset to normal.
3598  		sched_set_fifo(current);
3599  		torture_sched_setaffinity(current->pid, cpu_present_mask, true);
3600  		mdelay(preempt_duration);
3601  		sched_set_normal(current, 0);
3602  		stutter_wait("rcu_torture_preempt");
3603  	} while (!torture_must_stop());
3604  	torture_kthread_stopping("rcu_torture_preempt");
3605  	return 0;
3606  }
3607  
3608  static enum cpuhp_state rcutor_hp;
3609  
3610  static void
rcu_torture_cleanup(void)3611  rcu_torture_cleanup(void)
3612  {
3613  	int firsttime;
3614  	int flags = 0;
3615  	unsigned long gp_seq = 0;
3616  	int i;
3617  	int j;
3618  
3619  	if (torture_cleanup_begin()) {
3620  		if (cur_ops->cb_barrier != NULL) {
3621  			pr_info("%s: Invoking %pS().\n", __func__, cur_ops->cb_barrier);
3622  			cur_ops->cb_barrier();
3623  		}
3624  		if (cur_ops->gp_slow_unregister)
3625  			cur_ops->gp_slow_unregister(NULL);
3626  		return;
3627  	}
3628  	if (!cur_ops) {
3629  		torture_cleanup_end();
3630  		return;
3631  	}
3632  
3633  	rcutorture_test_nmis(test_nmis);
3634  
3635  	if (cur_ops->gp_kthread_dbg)
3636  		cur_ops->gp_kthread_dbg();
3637  	torture_stop_kthread(rcu_torture_preempt, preempt_task);
3638  	rcu_torture_read_exit_cleanup();
3639  	rcu_torture_barrier_cleanup();
3640  	rcu_torture_fwd_prog_cleanup();
3641  	torture_stop_kthread(rcu_torture_stall, stall_task);
3642  	torture_stop_kthread(rcu_torture_writer, writer_task);
3643  
3644  	if (nocb_tasks) {
3645  		for (i = 0; i < nrealnocbers; i++)
3646  			torture_stop_kthread(rcu_nocb_toggle, nocb_tasks[i]);
3647  		kfree(nocb_tasks);
3648  		nocb_tasks = NULL;
3649  	}
3650  
3651  	if (reader_tasks) {
3652  		for (i = 0; i < nrealreaders; i++)
3653  			torture_stop_kthread(rcu_torture_reader,
3654  					     reader_tasks[i]);
3655  		kfree(reader_tasks);
3656  		reader_tasks = NULL;
3657  	}
3658  	kfree(rcu_torture_reader_mbchk);
3659  	rcu_torture_reader_mbchk = NULL;
3660  
3661  	if (fakewriter_tasks) {
3662  		for (i = 0; i < nrealfakewriters; i++)
3663  			torture_stop_kthread(rcu_torture_fakewriter,
3664  					     fakewriter_tasks[i]);
3665  		kfree(fakewriter_tasks);
3666  		fakewriter_tasks = NULL;
3667  	}
3668  
3669  	if (cur_ops->get_gp_data)
3670  		cur_ops->get_gp_data(&flags, &gp_seq);
3671  	pr_alert("%s:  End-test grace-period state: g%ld f%#x total-gps=%ld\n",
3672  		 cur_ops->name, (long)gp_seq, flags,
3673  		 rcutorture_seq_diff(gp_seq, start_gp_seq));
3674  	torture_stop_kthread(rcu_torture_stats, stats_task);
3675  	torture_stop_kthread(rcu_torture_fqs, fqs_task);
3676  	if (rcu_torture_can_boost() && rcutor_hp >= 0)
3677  		cpuhp_remove_state(rcutor_hp);
3678  
3679  	/*
3680  	 * Wait for all RCU callbacks to fire, then do torture-type-specific
3681  	 * cleanup operations.
3682  	 */
3683  	if (cur_ops->cb_barrier != NULL) {
3684  		pr_info("%s: Invoking %pS().\n", __func__, cur_ops->cb_barrier);
3685  		cur_ops->cb_barrier();
3686  	}
3687  	if (cur_ops->cleanup != NULL)
3688  		cur_ops->cleanup();
3689  
3690  	rcu_torture_mem_dump_obj();
3691  
3692  	rcu_torture_stats_print();  /* -After- the stats thread is stopped! */
3693  
3694  	if (err_segs_recorded) {
3695  		pr_alert("Failure/close-call rcutorture reader segments:\n");
3696  		if (rt_read_nsegs == 0)
3697  			pr_alert("\t: No segments recorded!!!\n");
3698  		firsttime = 1;
3699  		for (i = 0; i < rt_read_nsegs; i++) {
3700  			if (IS_ENABLED(CONFIG_RCU_TORTURE_TEST_LOG_GP))
3701  				pr_alert("\t%lluus ", div64_u64(err_segs[i].rt_ts, 1000ULL));
3702  			else
3703  				pr_alert("\t");
3704  			pr_cont("%d: %#4x", i, err_segs[i].rt_readstate);
3705  			if (err_segs[i].rt_delay_jiffies != 0) {
3706  				pr_cont("%s%ldjiffies", firsttime ? "" : "+",
3707  					err_segs[i].rt_delay_jiffies);
3708  				firsttime = 0;
3709  			}
3710  			if (IS_ENABLED(CONFIG_RCU_TORTURE_TEST_LOG_CPU)) {
3711  				pr_cont(" CPU %2d", err_segs[i].rt_cpu);
3712  				if (err_segs[i].rt_cpu != err_segs[i].rt_end_cpu)
3713  					pr_cont("->%-2d", err_segs[i].rt_end_cpu);
3714  				else
3715  					pr_cont(" ...");
3716  			}
3717  			if (IS_ENABLED(CONFIG_RCU_TORTURE_TEST_LOG_GP) &&
3718  			    cur_ops->gather_gp_seqs && cur_ops->format_gp_seqs) {
3719  				char buf1[20+1];
3720  				char buf2[20+1];
3721  				char sepchar = '-';
3722  
3723  				cur_ops->format_gp_seqs(err_segs[i].rt_gp_seq,
3724  							buf1, ARRAY_SIZE(buf1));
3725  				cur_ops->format_gp_seqs(err_segs[i].rt_gp_seq_end,
3726  							buf2, ARRAY_SIZE(buf2));
3727  				if (err_segs[i].rt_gp_seq == err_segs[i].rt_gp_seq_end) {
3728  					if (buf2[0]) {
3729  						for (j = 0; buf2[j]; j++)
3730  							buf2[j] = '.';
3731  						if (j)
3732  							buf2[j - 1] = ' ';
3733  					}
3734  					sepchar = ' ';
3735  				}
3736  				pr_cont(" %s%c%s", buf1, sepchar, buf2);
3737  			}
3738  			if (err_segs[i].rt_delay_ms != 0) {
3739  				pr_cont(" %s%ldms", firsttime ? "" : "+",
3740  					err_segs[i].rt_delay_ms);
3741  				firsttime = 0;
3742  			}
3743  			if (err_segs[i].rt_delay_us != 0) {
3744  				pr_cont(" %s%ldus", firsttime ? "" : "+",
3745  					err_segs[i].rt_delay_us);
3746  				firsttime = 0;
3747  			}
3748  			pr_cont("%s", err_segs[i].rt_preempted ? " preempted" : "");
3749  			if (err_segs[i].rt_readstate & RCUTORTURE_RDR_BH)
3750  				pr_cont(" BH");
3751  			if (err_segs[i].rt_readstate & RCUTORTURE_RDR_IRQ)
3752  				pr_cont(" IRQ");
3753  			if (err_segs[i].rt_readstate & RCUTORTURE_RDR_PREEMPT)
3754  				pr_cont(" PREEMPT");
3755  			if (err_segs[i].rt_readstate & RCUTORTURE_RDR_RBH)
3756  				pr_cont(" RBH");
3757  			if (err_segs[i].rt_readstate & RCUTORTURE_RDR_SCHED)
3758  				pr_cont(" SCHED");
3759  			if (err_segs[i].rt_readstate & RCUTORTURE_RDR_RCU_1)
3760  				pr_cont(" RCU_1");
3761  			if (err_segs[i].rt_readstate & RCUTORTURE_RDR_RCU_2)
3762  				pr_cont(" RCU_2");
3763  			pr_cont("\n");
3764  
3765  		}
3766  		if (rt_read_preempted)
3767  			pr_alert("\tReader was preempted.\n");
3768  	}
3769  	if (atomic_read(&n_rcu_torture_error) || n_rcu_torture_barrier_error)
3770  		rcu_torture_print_module_parms(cur_ops, "End of test: FAILURE");
3771  	else if (torture_onoff_failures())
3772  		rcu_torture_print_module_parms(cur_ops,
3773  					       "End of test: RCU_HOTPLUG");
3774  	else
3775  		rcu_torture_print_module_parms(cur_ops, "End of test: SUCCESS");
3776  	torture_cleanup_end();
3777  	if (cur_ops->gp_slow_unregister)
3778  		cur_ops->gp_slow_unregister(NULL);
3779  }
3780  
rcu_torture_leak_cb(struct rcu_head * rhp)3781  static void rcu_torture_leak_cb(struct rcu_head *rhp)
3782  {
3783  }
3784  
rcu_torture_err_cb(struct rcu_head * rhp)3785  static void rcu_torture_err_cb(struct rcu_head *rhp)
3786  {
3787  	/*
3788  	 * This -might- happen due to race conditions, but is unlikely.
3789  	 * The scenario that leads to this happening is that the
3790  	 * first of the pair of duplicate callbacks is queued,
3791  	 * someone else starts a grace period that includes that
3792  	 * callback, then the second of the pair must wait for the
3793  	 * next grace period.  Unlikely, but can happen.  If it
3794  	 * does happen, the debug-objects subsystem won't have splatted.
3795  	 */
3796  	pr_alert("%s: duplicated callback was invoked.\n", KBUILD_MODNAME);
3797  }
3798  
3799  /*
3800   * Verify that double-free causes debug-objects to complain, but only
3801   * if CONFIG_DEBUG_OBJECTS_RCU_HEAD=y.  Otherwise, say that the test
3802   * cannot be carried out.
3803   */
rcu_test_debug_objects(void)3804  static void rcu_test_debug_objects(void)
3805  {
3806  	struct rcu_head rh1;
3807  	struct rcu_head rh2;
3808  	int idx;
3809  
3810  	if (!IS_ENABLED(CONFIG_DEBUG_OBJECTS_RCU_HEAD)) {
3811  		pr_alert("%s: !CONFIG_DEBUG_OBJECTS_RCU_HEAD, not testing duplicate call_%s()\n",
3812  					KBUILD_MODNAME, cur_ops->name);
3813  		return;
3814  	}
3815  
3816  	if (WARN_ON_ONCE(cur_ops->debug_objects &&
3817  			(!cur_ops->call || !cur_ops->cb_barrier)))
3818  		return;
3819  
3820  	struct rcu_head *rhp = kmalloc(sizeof(*rhp), GFP_KERNEL);
3821  
3822  	init_rcu_head_on_stack(&rh1);
3823  	init_rcu_head_on_stack(&rh2);
3824  	pr_alert("%s: WARN: Duplicate call_%s() test starting.\n", KBUILD_MODNAME, cur_ops->name);
3825  
3826  	/* Try to queue the rh2 pair of callbacks for the same grace period. */
3827  	idx = cur_ops->readlock(); /* Make it impossible to finish a grace period. */
3828  	cur_ops->call(&rh1, rcu_torture_leak_cb); /* Start grace period. */
3829  	cur_ops->call(&rh2, rcu_torture_leak_cb);
3830  	cur_ops->call(&rh2, rcu_torture_err_cb); /* Duplicate callback. */
3831  	if (rhp) {
3832  		cur_ops->call(rhp, rcu_torture_leak_cb);
3833  		cur_ops->call(rhp, rcu_torture_err_cb); /* Another duplicate callback. */
3834  	}
3835  	cur_ops->readunlock(idx);
3836  
3837  	/* Wait for them all to get done so we can safely return. */
3838  	cur_ops->cb_barrier();
3839  	pr_alert("%s: WARN: Duplicate call_%s() test complete.\n", KBUILD_MODNAME, cur_ops->name);
3840  	destroy_rcu_head_on_stack(&rh1);
3841  	destroy_rcu_head_on_stack(&rh2);
3842  	kfree(rhp);
3843  }
3844  
rcutorture_sync(void)3845  static void rcutorture_sync(void)
3846  {
3847  	static unsigned long n;
3848  
3849  	if (cur_ops->sync && !(++n & 0xfff))
3850  		cur_ops->sync();
3851  }
3852  
3853  static DEFINE_MUTEX(mut0);
3854  static DEFINE_MUTEX(mut1);
3855  static DEFINE_MUTEX(mut2);
3856  static DEFINE_MUTEX(mut3);
3857  static DEFINE_MUTEX(mut4);
3858  static DEFINE_MUTEX(mut5);
3859  static DEFINE_MUTEX(mut6);
3860  static DEFINE_MUTEX(mut7);
3861  static DEFINE_MUTEX(mut8);
3862  static DEFINE_MUTEX(mut9);
3863  
3864  static DECLARE_RWSEM(rwsem0);
3865  static DECLARE_RWSEM(rwsem1);
3866  static DECLARE_RWSEM(rwsem2);
3867  static DECLARE_RWSEM(rwsem3);
3868  static DECLARE_RWSEM(rwsem4);
3869  static DECLARE_RWSEM(rwsem5);
3870  static DECLARE_RWSEM(rwsem6);
3871  static DECLARE_RWSEM(rwsem7);
3872  static DECLARE_RWSEM(rwsem8);
3873  static DECLARE_RWSEM(rwsem9);
3874  
3875  DEFINE_STATIC_SRCU(srcu0);
3876  DEFINE_STATIC_SRCU(srcu1);
3877  DEFINE_STATIC_SRCU(srcu2);
3878  DEFINE_STATIC_SRCU(srcu3);
3879  DEFINE_STATIC_SRCU(srcu4);
3880  DEFINE_STATIC_SRCU(srcu5);
3881  DEFINE_STATIC_SRCU(srcu6);
3882  DEFINE_STATIC_SRCU(srcu7);
3883  DEFINE_STATIC_SRCU(srcu8);
3884  DEFINE_STATIC_SRCU(srcu9);
3885  
srcu_lockdep_next(const char * f,const char * fl,const char * fs,const char * fu,int i,int cyclelen,int deadlock)3886  static int srcu_lockdep_next(const char *f, const char *fl, const char *fs, const char *fu, int i,
3887  			     int cyclelen, int deadlock)
3888  {
3889  	int j = i + 1;
3890  
3891  	if (j >= cyclelen)
3892  		j = deadlock ? 0 : -1;
3893  	if (j >= 0)
3894  		pr_info("%s: %s(%d), %s(%d), %s(%d)\n", f, fl, i, fs, j, fu, i);
3895  	else
3896  		pr_info("%s: %s(%d), %s(%d)\n", f, fl, i, fu, i);
3897  	return j;
3898  }
3899  
3900  // Test lockdep on SRCU-based deadlock scenarios.
rcu_torture_init_srcu_lockdep(void)3901  static void rcu_torture_init_srcu_lockdep(void)
3902  {
3903  	int cyclelen;
3904  	int deadlock;
3905  	bool err = false;
3906  	int i;
3907  	int j;
3908  	int idx;
3909  	struct mutex *muts[] = { &mut0, &mut1, &mut2, &mut3, &mut4,
3910  				 &mut5, &mut6, &mut7, &mut8, &mut9 };
3911  	struct rw_semaphore *rwsems[] = { &rwsem0, &rwsem1, &rwsem2, &rwsem3, &rwsem4,
3912  					  &rwsem5, &rwsem6, &rwsem7, &rwsem8, &rwsem9 };
3913  	struct srcu_struct *srcus[] = { &srcu0, &srcu1, &srcu2, &srcu3, &srcu4,
3914  					&srcu5, &srcu6, &srcu7, &srcu8, &srcu9 };
3915  	int testtype;
3916  
3917  	if (!test_srcu_lockdep)
3918  		return;
3919  
3920  	deadlock = test_srcu_lockdep / 1000;
3921  	testtype = (test_srcu_lockdep / 10) % 100;
3922  	cyclelen = test_srcu_lockdep % 10;
3923  	WARN_ON_ONCE(ARRAY_SIZE(muts) != ARRAY_SIZE(srcus));
3924  	if (WARN_ONCE(deadlock != !!deadlock,
3925  		      "%s: test_srcu_lockdep=%d and deadlock digit %d must be zero or one.\n",
3926  		      __func__, test_srcu_lockdep, deadlock))
3927  		err = true;
3928  	if (WARN_ONCE(cyclelen <= 0,
3929  		      "%s: test_srcu_lockdep=%d and cycle-length digit %d must be greater than zero.\n",
3930  		      __func__, test_srcu_lockdep, cyclelen))
3931  		err = true;
3932  	if (err)
3933  		goto err_out;
3934  
3935  	if (testtype == 0) {
3936  		pr_info("%s: test_srcu_lockdep = %05d: SRCU %d-way %sdeadlock.\n",
3937  			__func__, test_srcu_lockdep, cyclelen, deadlock ? "" : "non-");
3938  		if (deadlock && cyclelen == 1)
3939  			pr_info("%s: Expect hang.\n", __func__);
3940  		for (i = 0; i < cyclelen; i++) {
3941  			j = srcu_lockdep_next(__func__, "srcu_read_lock", "synchronize_srcu",
3942  					      "srcu_read_unlock", i, cyclelen, deadlock);
3943  			idx = srcu_read_lock(srcus[i]);
3944  			if (j >= 0)
3945  				synchronize_srcu(srcus[j]);
3946  			srcu_read_unlock(srcus[i], idx);
3947  		}
3948  		return;
3949  	}
3950  
3951  	if (testtype == 1) {
3952  		pr_info("%s: test_srcu_lockdep = %05d: SRCU/mutex %d-way %sdeadlock.\n",
3953  			__func__, test_srcu_lockdep, cyclelen, deadlock ? "" : "non-");
3954  		for (i = 0; i < cyclelen; i++) {
3955  			pr_info("%s: srcu_read_lock(%d), mutex_lock(%d), mutex_unlock(%d), srcu_read_unlock(%d)\n",
3956  				__func__, i, i, i, i);
3957  			idx = srcu_read_lock(srcus[i]);
3958  			mutex_lock(muts[i]);
3959  			mutex_unlock(muts[i]);
3960  			srcu_read_unlock(srcus[i], idx);
3961  
3962  			j = srcu_lockdep_next(__func__, "mutex_lock", "synchronize_srcu",
3963  					      "mutex_unlock", i, cyclelen, deadlock);
3964  			mutex_lock(muts[i]);
3965  			if (j >= 0)
3966  				synchronize_srcu(srcus[j]);
3967  			mutex_unlock(muts[i]);
3968  		}
3969  		return;
3970  	}
3971  
3972  	if (testtype == 2) {
3973  		pr_info("%s: test_srcu_lockdep = %05d: SRCU/rwsem %d-way %sdeadlock.\n",
3974  			__func__, test_srcu_lockdep, cyclelen, deadlock ? "" : "non-");
3975  		for (i = 0; i < cyclelen; i++) {
3976  			pr_info("%s: srcu_read_lock(%d), down_read(%d), up_read(%d), srcu_read_unlock(%d)\n",
3977  				__func__, i, i, i, i);
3978  			idx = srcu_read_lock(srcus[i]);
3979  			down_read(rwsems[i]);
3980  			up_read(rwsems[i]);
3981  			srcu_read_unlock(srcus[i], idx);
3982  
3983  			j = srcu_lockdep_next(__func__, "down_write", "synchronize_srcu",
3984  					      "up_write", i, cyclelen, deadlock);
3985  			down_write(rwsems[i]);
3986  			if (j >= 0)
3987  				synchronize_srcu(srcus[j]);
3988  			up_write(rwsems[i]);
3989  		}
3990  		return;
3991  	}
3992  
3993  #ifdef CONFIG_TASKS_TRACE_RCU
3994  	if (testtype == 3) {
3995  		pr_info("%s: test_srcu_lockdep = %05d: SRCU and Tasks Trace RCU %d-way %sdeadlock.\n",
3996  			__func__, test_srcu_lockdep, cyclelen, deadlock ? "" : "non-");
3997  		if (deadlock && cyclelen == 1)
3998  			pr_info("%s: Expect hang.\n", __func__);
3999  		for (i = 0; i < cyclelen; i++) {
4000  			char *fl = i == 0 ? "rcu_read_lock_trace" : "srcu_read_lock";
4001  			char *fs = i == cyclelen - 1 ? "synchronize_rcu_tasks_trace"
4002  						     : "synchronize_srcu";
4003  			char *fu = i == 0 ? "rcu_read_unlock_trace" : "srcu_read_unlock";
4004  
4005  			j = srcu_lockdep_next(__func__, fl, fs, fu, i, cyclelen, deadlock);
4006  			if (i == 0)
4007  				rcu_read_lock_trace();
4008  			else
4009  				idx = srcu_read_lock(srcus[i]);
4010  			if (j >= 0) {
4011  				if (i == cyclelen - 1)
4012  					synchronize_rcu_tasks_trace();
4013  				else
4014  					synchronize_srcu(srcus[j]);
4015  			}
4016  			if (i == 0)
4017  				rcu_read_unlock_trace();
4018  			else
4019  				srcu_read_unlock(srcus[i], idx);
4020  		}
4021  		return;
4022  	}
4023  #endif // #ifdef CONFIG_TASKS_TRACE_RCU
4024  
4025  err_out:
4026  	pr_info("%s: test_srcu_lockdep = %05d does nothing.\n", __func__, test_srcu_lockdep);
4027  	pr_info("%s: test_srcu_lockdep = DNNL.\n", __func__);
4028  	pr_info("%s: D: Deadlock if nonzero.\n", __func__);
4029  	pr_info("%s: NN: Test number, 0=SRCU, 1=SRCU/mutex, 2=SRCU/rwsem, 3=SRCU/Tasks Trace RCU.\n", __func__);
4030  	pr_info("%s: L: Cycle length.\n", __func__);
4031  	if (!IS_ENABLED(CONFIG_TASKS_TRACE_RCU))
4032  		pr_info("%s: NN=3 disallowed because kernel is built with CONFIG_TASKS_TRACE_RCU=n\n", __func__);
4033  }
4034  
4035  static int __init
rcu_torture_init(void)4036  rcu_torture_init(void)
4037  {
4038  	long i;
4039  	int cpu;
4040  	int firsterr = 0;
4041  	int flags = 0;
4042  	unsigned long gp_seq = 0;
4043  	static struct rcu_torture_ops *torture_ops[] = {
4044  		&rcu_ops, &rcu_busted_ops, &srcu_ops, &srcud_ops, &busted_srcud_ops,
4045  		TASKS_OPS TASKS_RUDE_OPS TASKS_TRACING_OPS
4046  		&trivial_ops,
4047  	};
4048  
4049  	if (!torture_init_begin(torture_type, verbose))
4050  		return -EBUSY;
4051  
4052  	/* Process args and tell the world that the torturer is on the job. */
4053  	for (i = 0; i < ARRAY_SIZE(torture_ops); i++) {
4054  		cur_ops = torture_ops[i];
4055  		if (strcmp(torture_type, cur_ops->name) == 0)
4056  			break;
4057  	}
4058  	if (i == ARRAY_SIZE(torture_ops)) {
4059  		pr_alert("rcu-torture: invalid torture type: \"%s\"\n",
4060  			 torture_type);
4061  		pr_alert("rcu-torture types:");
4062  		for (i = 0; i < ARRAY_SIZE(torture_ops); i++)
4063  			pr_cont(" %s", torture_ops[i]->name);
4064  		pr_cont("\n");
4065  		firsterr = -EINVAL;
4066  		cur_ops = NULL;
4067  		goto unwind;
4068  	}
4069  	if (cur_ops->fqs == NULL && fqs_duration != 0) {
4070  		pr_alert("rcu-torture: ->fqs NULL and non-zero fqs_duration, fqs disabled.\n");
4071  		fqs_duration = 0;
4072  	}
4073  	if (nocbs_nthreads != 0 && (cur_ops != &rcu_ops ||
4074  				    !IS_ENABLED(CONFIG_RCU_NOCB_CPU))) {
4075  		pr_alert("rcu-torture types: %s and CONFIG_RCU_NOCB_CPU=%d, nocb toggle disabled.\n",
4076  			 cur_ops->name, IS_ENABLED(CONFIG_RCU_NOCB_CPU));
4077  		nocbs_nthreads = 0;
4078  	}
4079  	if (cur_ops->init)
4080  		cur_ops->init();
4081  
4082  	rcu_torture_init_srcu_lockdep();
4083  
4084  	if (nfakewriters >= 0) {
4085  		nrealfakewriters = nfakewriters;
4086  	} else {
4087  		nrealfakewriters = num_online_cpus() - 2 - nfakewriters;
4088  		if (nrealfakewriters <= 0)
4089  			nrealfakewriters = 1;
4090  	}
4091  
4092  	if (nreaders >= 0) {
4093  		nrealreaders = nreaders;
4094  	} else {
4095  		nrealreaders = num_online_cpus() - 2 - nreaders;
4096  		if (nrealreaders <= 0)
4097  			nrealreaders = 1;
4098  	}
4099  	rcu_torture_print_module_parms(cur_ops, "Start of test");
4100  	if (cur_ops->get_gp_data)
4101  		cur_ops->get_gp_data(&flags, &gp_seq);
4102  	start_gp_seq = gp_seq;
4103  	pr_alert("%s:  Start-test grace-period state: g%ld f%#x\n",
4104  		 cur_ops->name, (long)gp_seq, flags);
4105  
4106  	/* Set up the freelist. */
4107  
4108  	INIT_LIST_HEAD(&rcu_torture_freelist);
4109  	for (i = 0; i < ARRAY_SIZE(rcu_tortures); i++) {
4110  		rcu_tortures[i].rtort_mbtest = 0;
4111  		list_add_tail(&rcu_tortures[i].rtort_free,
4112  			      &rcu_torture_freelist);
4113  	}
4114  
4115  	/* Initialize the statistics so that each run gets its own numbers. */
4116  
4117  	rcu_torture_current = NULL;
4118  	rcu_torture_current_version = 0;
4119  	atomic_set(&n_rcu_torture_alloc, 0);
4120  	atomic_set(&n_rcu_torture_alloc_fail, 0);
4121  	atomic_set(&n_rcu_torture_free, 0);
4122  	atomic_set(&n_rcu_torture_mberror, 0);
4123  	atomic_set(&n_rcu_torture_mbchk_fail, 0);
4124  	atomic_set(&n_rcu_torture_mbchk_tries, 0);
4125  	atomic_set(&n_rcu_torture_error, 0);
4126  	n_rcu_torture_barrier_error = 0;
4127  	n_rcu_torture_boost_ktrerror = 0;
4128  	n_rcu_torture_boost_failure = 0;
4129  	n_rcu_torture_boosts = 0;
4130  	for (i = 0; i < RCU_TORTURE_PIPE_LEN + 1; i++)
4131  		atomic_set(&rcu_torture_wcount[i], 0);
4132  	for_each_possible_cpu(cpu) {
4133  		for (i = 0; i < RCU_TORTURE_PIPE_LEN + 1; i++) {
4134  			per_cpu(rcu_torture_count, cpu)[i] = 0;
4135  			per_cpu(rcu_torture_batch, cpu)[i] = 0;
4136  		}
4137  	}
4138  	err_segs_recorded = 0;
4139  	rt_read_nsegs = 0;
4140  
4141  	/* Start up the kthreads. */
4142  
4143  	rcu_torture_write_types();
4144  	firsterr = torture_create_kthread(rcu_torture_writer, NULL,
4145  					  writer_task);
4146  	if (torture_init_error(firsterr))
4147  		goto unwind;
4148  
4149  	if (nrealfakewriters > 0) {
4150  		fakewriter_tasks = kcalloc(nrealfakewriters,
4151  					   sizeof(fakewriter_tasks[0]),
4152  					   GFP_KERNEL);
4153  		if (fakewriter_tasks == NULL) {
4154  			TOROUT_ERRSTRING("out of memory");
4155  			firsterr = -ENOMEM;
4156  			goto unwind;
4157  		}
4158  	}
4159  	for (i = 0; i < nrealfakewriters; i++) {
4160  		firsterr = torture_create_kthread(rcu_torture_fakewriter,
4161  						  NULL, fakewriter_tasks[i]);
4162  		if (torture_init_error(firsterr))
4163  			goto unwind;
4164  	}
4165  	reader_tasks = kcalloc(nrealreaders, sizeof(reader_tasks[0]),
4166  			       GFP_KERNEL);
4167  	rcu_torture_reader_mbchk = kcalloc(nrealreaders, sizeof(*rcu_torture_reader_mbchk),
4168  					   GFP_KERNEL);
4169  	if (!reader_tasks || !rcu_torture_reader_mbchk) {
4170  		TOROUT_ERRSTRING("out of memory");
4171  		firsterr = -ENOMEM;
4172  		goto unwind;
4173  	}
4174  	for (i = 0; i < nrealreaders; i++) {
4175  		rcu_torture_reader_mbchk[i].rtc_chkrdr = -1;
4176  		firsterr = torture_create_kthread(rcu_torture_reader, (void *)i,
4177  						  reader_tasks[i]);
4178  		if (torture_init_error(firsterr))
4179  			goto unwind;
4180  	}
4181  	nrealnocbers = nocbs_nthreads;
4182  	if (WARN_ON(nrealnocbers < 0))
4183  		nrealnocbers = 1;
4184  	if (WARN_ON(nocbs_toggle < 0))
4185  		nocbs_toggle = HZ;
4186  	if (nrealnocbers > 0) {
4187  		nocb_tasks = kcalloc(nrealnocbers, sizeof(nocb_tasks[0]), GFP_KERNEL);
4188  		if (nocb_tasks == NULL) {
4189  			TOROUT_ERRSTRING("out of memory");
4190  			firsterr = -ENOMEM;
4191  			goto unwind;
4192  		}
4193  	} else {
4194  		nocb_tasks = NULL;
4195  	}
4196  	for (i = 0; i < nrealnocbers; i++) {
4197  		firsterr = torture_create_kthread(rcu_nocb_toggle, NULL, nocb_tasks[i]);
4198  		if (torture_init_error(firsterr))
4199  			goto unwind;
4200  	}
4201  	if (stat_interval > 0) {
4202  		firsterr = torture_create_kthread(rcu_torture_stats, NULL,
4203  						  stats_task);
4204  		if (torture_init_error(firsterr))
4205  			goto unwind;
4206  	}
4207  	if (test_no_idle_hz && shuffle_interval > 0) {
4208  		firsterr = torture_shuffle_init(shuffle_interval * HZ);
4209  		if (torture_init_error(firsterr))
4210  			goto unwind;
4211  	}
4212  	if (stutter < 0)
4213  		stutter = 0;
4214  	if (stutter) {
4215  		int t;
4216  
4217  		t = cur_ops->stall_dur ? cur_ops->stall_dur() : stutter * HZ;
4218  		firsterr = torture_stutter_init(stutter * HZ, t);
4219  		if (torture_init_error(firsterr))
4220  			goto unwind;
4221  	}
4222  	if (fqs_duration < 0)
4223  		fqs_duration = 0;
4224  	if (fqs_holdoff < 0)
4225  		fqs_holdoff = 0;
4226  	if (fqs_duration && fqs_holdoff) {
4227  		/* Create the fqs thread */
4228  		firsterr = torture_create_kthread(rcu_torture_fqs, NULL,
4229  						  fqs_task);
4230  		if (torture_init_error(firsterr))
4231  			goto unwind;
4232  	}
4233  	if (test_boost_interval < 1)
4234  		test_boost_interval = 1;
4235  	if (test_boost_duration < 2)
4236  		test_boost_duration = 2;
4237  	if (rcu_torture_can_boost()) {
4238  
4239  		boost_starttime = jiffies + test_boost_interval * HZ;
4240  
4241  		firsterr = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "RCU_TORTURE",
4242  					     rcutorture_booster_init,
4243  					     rcutorture_booster_cleanup);
4244  		rcutor_hp = firsterr;
4245  		if (torture_init_error(firsterr))
4246  			goto unwind;
4247  	}
4248  	shutdown_jiffies = jiffies + shutdown_secs * HZ;
4249  	firsterr = torture_shutdown_init(shutdown_secs, rcu_torture_cleanup);
4250  	if (torture_init_error(firsterr))
4251  		goto unwind;
4252  	firsterr = torture_onoff_init(onoff_holdoff * HZ, onoff_interval,
4253  				      rcutorture_sync);
4254  	if (torture_init_error(firsterr))
4255  		goto unwind;
4256  	firsterr = rcu_torture_stall_init();
4257  	if (torture_init_error(firsterr))
4258  		goto unwind;
4259  	firsterr = rcu_torture_fwd_prog_init();
4260  	if (torture_init_error(firsterr))
4261  		goto unwind;
4262  	firsterr = rcu_torture_barrier_init();
4263  	if (torture_init_error(firsterr))
4264  		goto unwind;
4265  	firsterr = rcu_torture_read_exit_init();
4266  	if (torture_init_error(firsterr))
4267  		goto unwind;
4268  	if (preempt_duration > 0) {
4269  		firsterr = torture_create_kthread(rcu_torture_preempt, NULL, preempt_task);
4270  		if (torture_init_error(firsterr))
4271  			goto unwind;
4272  	}
4273  	if (object_debug)
4274  		rcu_test_debug_objects();
4275  	torture_init_end();
4276  	if (cur_ops->gp_slow_register && !WARN_ON_ONCE(!cur_ops->gp_slow_unregister))
4277  		cur_ops->gp_slow_register(&rcu_fwd_cb_nodelay);
4278  	return 0;
4279  
4280  unwind:
4281  	torture_init_end();
4282  	rcu_torture_cleanup();
4283  	if (shutdown_secs) {
4284  		WARN_ON(!IS_MODULE(CONFIG_RCU_TORTURE_TEST));
4285  		kernel_power_off();
4286  	}
4287  	return firsterr;
4288  }
4289  
4290  module_init(rcu_torture_init);
4291  module_exit(rcu_torture_cleanup);
4292