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