xref: /openbmc/linux/kernel/kcsan/report.c (revision 9659281c)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * KCSAN reporting.
4  *
5  * Copyright (C) 2019, Google LLC.
6  */
7 
8 #include <linux/debug_locks.h>
9 #include <linux/delay.h>
10 #include <linux/jiffies.h>
11 #include <linux/kernel.h>
12 #include <linux/lockdep.h>
13 #include <linux/preempt.h>
14 #include <linux/printk.h>
15 #include <linux/sched.h>
16 #include <linux/spinlock.h>
17 #include <linux/stacktrace.h>
18 
19 #include "kcsan.h"
20 #include "encoding.h"
21 
22 /*
23  * Max. number of stack entries to show in the report.
24  */
25 #define NUM_STACK_ENTRIES 64
26 
27 /* Common access info. */
28 struct access_info {
29 	const volatile void	*ptr;
30 	size_t			size;
31 	int			access_type;
32 	int			task_pid;
33 	int			cpu_id;
34 };
35 
36 /*
37  * Other thread info: communicated from other racing thread to thread that set
38  * up the watchpoint, which then prints the complete report atomically.
39  */
40 struct other_info {
41 	struct access_info	ai;
42 	unsigned long		stack_entries[NUM_STACK_ENTRIES];
43 	int			num_stack_entries;
44 
45 	/*
46 	 * Optionally pass @current. Typically we do not need to pass @current
47 	 * via @other_info since just @task_pid is sufficient. Passing @current
48 	 * has additional overhead.
49 	 *
50 	 * To safely pass @current, we must either use get_task_struct/
51 	 * put_task_struct, or stall the thread that populated @other_info.
52 	 *
53 	 * We cannot rely on get_task_struct/put_task_struct in case
54 	 * release_report() races with a task being released, and would have to
55 	 * free it in release_report(). This may result in deadlock if we want
56 	 * to use KCSAN on the allocators.
57 	 *
58 	 * Since we also want to reliably print held locks for
59 	 * CONFIG_KCSAN_VERBOSE, the current implementation stalls the thread
60 	 * that populated @other_info until it has been consumed.
61 	 */
62 	struct task_struct	*task;
63 };
64 
65 /*
66  * To never block any producers of struct other_info, we need as many elements
67  * as we have watchpoints (upper bound on concurrent races to report).
68  */
69 static struct other_info other_infos[CONFIG_KCSAN_NUM_WATCHPOINTS + NUM_SLOTS-1];
70 
71 /*
72  * Information about reported races; used to rate limit reporting.
73  */
74 struct report_time {
75 	/*
76 	 * The last time the race was reported.
77 	 */
78 	unsigned long time;
79 
80 	/*
81 	 * The frames of the 2 threads; if only 1 thread is known, one frame
82 	 * will be 0.
83 	 */
84 	unsigned long frame1;
85 	unsigned long frame2;
86 };
87 
88 /*
89  * Since we also want to be able to debug allocators with KCSAN, to avoid
90  * deadlock, report_times cannot be dynamically resized with krealloc in
91  * rate_limit_report.
92  *
93  * Therefore, we use a fixed-size array, which at most will occupy a page. This
94  * still adequately rate limits reports, assuming that a) number of unique data
95  * races is not excessive, and b) occurrence of unique races within the
96  * same time window is limited.
97  */
98 #define REPORT_TIMES_MAX (PAGE_SIZE / sizeof(struct report_time))
99 #define REPORT_TIMES_SIZE                                                      \
100 	(CONFIG_KCSAN_REPORT_ONCE_IN_MS > REPORT_TIMES_MAX ?                   \
101 		 REPORT_TIMES_MAX :                                            \
102 		 CONFIG_KCSAN_REPORT_ONCE_IN_MS)
103 static struct report_time report_times[REPORT_TIMES_SIZE];
104 
105 /*
106  * Spinlock serializing report generation, and access to @other_infos. Although
107  * it could make sense to have a finer-grained locking story for @other_infos,
108  * report generation needs to be serialized either way, so not much is gained.
109  */
110 static DEFINE_RAW_SPINLOCK(report_lock);
111 
112 /*
113  * Checks if the race identified by thread frames frame1 and frame2 has
114  * been reported since (now - KCSAN_REPORT_ONCE_IN_MS).
115  */
116 static bool rate_limit_report(unsigned long frame1, unsigned long frame2)
117 {
118 	struct report_time *use_entry = &report_times[0];
119 	unsigned long invalid_before;
120 	int i;
121 
122 	BUILD_BUG_ON(CONFIG_KCSAN_REPORT_ONCE_IN_MS != 0 && REPORT_TIMES_SIZE == 0);
123 
124 	if (CONFIG_KCSAN_REPORT_ONCE_IN_MS == 0)
125 		return false;
126 
127 	invalid_before = jiffies - msecs_to_jiffies(CONFIG_KCSAN_REPORT_ONCE_IN_MS);
128 
129 	/* Check if a matching race report exists. */
130 	for (i = 0; i < REPORT_TIMES_SIZE; ++i) {
131 		struct report_time *rt = &report_times[i];
132 
133 		/*
134 		 * Must always select an entry for use to store info as we
135 		 * cannot resize report_times; at the end of the scan, use_entry
136 		 * will be the oldest entry, which ideally also happened before
137 		 * KCSAN_REPORT_ONCE_IN_MS ago.
138 		 */
139 		if (time_before(rt->time, use_entry->time))
140 			use_entry = rt;
141 
142 		/*
143 		 * Initially, no need to check any further as this entry as well
144 		 * as following entries have never been used.
145 		 */
146 		if (rt->time == 0)
147 			break;
148 
149 		/* Check if entry expired. */
150 		if (time_before(rt->time, invalid_before))
151 			continue; /* before KCSAN_REPORT_ONCE_IN_MS ago */
152 
153 		/* Reported recently, check if race matches. */
154 		if ((rt->frame1 == frame1 && rt->frame2 == frame2) ||
155 		    (rt->frame1 == frame2 && rt->frame2 == frame1))
156 			return true;
157 	}
158 
159 	use_entry->time = jiffies;
160 	use_entry->frame1 = frame1;
161 	use_entry->frame2 = frame2;
162 	return false;
163 }
164 
165 /*
166  * Special rules to skip reporting.
167  */
168 static bool
169 skip_report(enum kcsan_value_change value_change, unsigned long top_frame)
170 {
171 	/* Should never get here if value_change==FALSE. */
172 	WARN_ON_ONCE(value_change == KCSAN_VALUE_CHANGE_FALSE);
173 
174 	/*
175 	 * The first call to skip_report always has value_change==TRUE, since we
176 	 * cannot know the value written of an instrumented access. For the 2nd
177 	 * call there are 6 cases with CONFIG_KCSAN_REPORT_VALUE_CHANGE_ONLY:
178 	 *
179 	 * 1. read watchpoint, conflicting write (value_change==TRUE): report;
180 	 * 2. read watchpoint, conflicting write (value_change==MAYBE): skip;
181 	 * 3. write watchpoint, conflicting write (value_change==TRUE): report;
182 	 * 4. write watchpoint, conflicting write (value_change==MAYBE): skip;
183 	 * 5. write watchpoint, conflicting read (value_change==MAYBE): skip;
184 	 * 6. write watchpoint, conflicting read (value_change==TRUE): report;
185 	 *
186 	 * Cases 1-4 are intuitive and expected; case 5 ensures we do not report
187 	 * data races where the write may have rewritten the same value; case 6
188 	 * is possible either if the size is larger than what we check value
189 	 * changes for or the access type is KCSAN_ACCESS_ASSERT.
190 	 */
191 	if (IS_ENABLED(CONFIG_KCSAN_REPORT_VALUE_CHANGE_ONLY) &&
192 	    value_change == KCSAN_VALUE_CHANGE_MAYBE) {
193 		/*
194 		 * The access is a write, but the data value did not change.
195 		 *
196 		 * We opt-out of this filter for certain functions at request of
197 		 * maintainers.
198 		 */
199 		char buf[64];
200 		int len = scnprintf(buf, sizeof(buf), "%ps", (void *)top_frame);
201 
202 		if (!strnstr(buf, "rcu_", len) &&
203 		    !strnstr(buf, "_rcu", len) &&
204 		    !strnstr(buf, "_srcu", len))
205 			return true;
206 	}
207 
208 	return kcsan_skip_report_debugfs(top_frame);
209 }
210 
211 static const char *get_access_type(int type)
212 {
213 	if (type & KCSAN_ACCESS_ASSERT) {
214 		if (type & KCSAN_ACCESS_SCOPED) {
215 			if (type & KCSAN_ACCESS_WRITE)
216 				return "assert no accesses (scoped)";
217 			else
218 				return "assert no writes (scoped)";
219 		} else {
220 			if (type & KCSAN_ACCESS_WRITE)
221 				return "assert no accesses";
222 			else
223 				return "assert no writes";
224 		}
225 	}
226 
227 	switch (type) {
228 	case 0:
229 		return "read";
230 	case KCSAN_ACCESS_ATOMIC:
231 		return "read (marked)";
232 	case KCSAN_ACCESS_WRITE:
233 		return "write";
234 	case KCSAN_ACCESS_WRITE | KCSAN_ACCESS_ATOMIC:
235 		return "write (marked)";
236 	case KCSAN_ACCESS_COMPOUND | KCSAN_ACCESS_WRITE:
237 		return "read-write";
238 	case KCSAN_ACCESS_COMPOUND | KCSAN_ACCESS_WRITE | KCSAN_ACCESS_ATOMIC:
239 		return "read-write (marked)";
240 	case KCSAN_ACCESS_SCOPED:
241 		return "read (scoped)";
242 	case KCSAN_ACCESS_SCOPED | KCSAN_ACCESS_ATOMIC:
243 		return "read (marked, scoped)";
244 	case KCSAN_ACCESS_SCOPED | KCSAN_ACCESS_WRITE:
245 		return "write (scoped)";
246 	case KCSAN_ACCESS_SCOPED | KCSAN_ACCESS_WRITE | KCSAN_ACCESS_ATOMIC:
247 		return "write (marked, scoped)";
248 	default:
249 		BUG();
250 	}
251 }
252 
253 static const char *get_bug_type(int type)
254 {
255 	return (type & KCSAN_ACCESS_ASSERT) != 0 ? "assert: race" : "data-race";
256 }
257 
258 /* Return thread description: in task or interrupt. */
259 static const char *get_thread_desc(int task_id)
260 {
261 	if (task_id != -1) {
262 		static char buf[32]; /* safe: protected by report_lock */
263 
264 		snprintf(buf, sizeof(buf), "task %i", task_id);
265 		return buf;
266 	}
267 	return "interrupt";
268 }
269 
270 /* Helper to skip KCSAN-related functions in stack-trace. */
271 static int get_stack_skipnr(const unsigned long stack_entries[], int num_entries)
272 {
273 	char buf[64];
274 	char *cur;
275 	int len, skip;
276 
277 	for (skip = 0; skip < num_entries; ++skip) {
278 		len = scnprintf(buf, sizeof(buf), "%ps", (void *)stack_entries[skip]);
279 
280 		/* Never show tsan_* or {read,write}_once_size. */
281 		if (strnstr(buf, "tsan_", len) ||
282 		    strnstr(buf, "_once_size", len))
283 			continue;
284 
285 		cur = strnstr(buf, "kcsan_", len);
286 		if (cur) {
287 			cur += strlen("kcsan_");
288 			if (!str_has_prefix(cur, "test"))
289 				continue; /* KCSAN runtime function. */
290 			/* KCSAN related test. */
291 		}
292 
293 		/*
294 		 * No match for runtime functions -- @skip entries to skip to
295 		 * get to first frame of interest.
296 		 */
297 		break;
298 	}
299 
300 	return skip;
301 }
302 
303 /* Compares symbolized strings of addr1 and addr2. */
304 static int sym_strcmp(void *addr1, void *addr2)
305 {
306 	char buf1[64];
307 	char buf2[64];
308 
309 	snprintf(buf1, sizeof(buf1), "%pS", addr1);
310 	snprintf(buf2, sizeof(buf2), "%pS", addr2);
311 
312 	return strncmp(buf1, buf2, sizeof(buf1));
313 }
314 
315 static void print_verbose_info(struct task_struct *task)
316 {
317 	if (!task)
318 		return;
319 
320 	/* Restore IRQ state trace for printing. */
321 	kcsan_restore_irqtrace(task);
322 
323 	pr_err("\n");
324 	debug_show_held_locks(task);
325 	print_irqtrace_events(task);
326 }
327 
328 static void print_report(enum kcsan_value_change value_change,
329 			 const struct access_info *ai,
330 			 const struct other_info *other_info,
331 			 u64 old, u64 new, u64 mask)
332 {
333 	unsigned long stack_entries[NUM_STACK_ENTRIES] = { 0 };
334 	int num_stack_entries = stack_trace_save(stack_entries, NUM_STACK_ENTRIES, 1);
335 	int skipnr = get_stack_skipnr(stack_entries, num_stack_entries);
336 	unsigned long this_frame = stack_entries[skipnr];
337 	unsigned long other_frame = 0;
338 	int other_skipnr = 0; /* silence uninit warnings */
339 
340 	/*
341 	 * Must check report filter rules before starting to print.
342 	 */
343 	if (skip_report(KCSAN_VALUE_CHANGE_TRUE, stack_entries[skipnr]))
344 		return;
345 
346 	if (other_info) {
347 		other_skipnr = get_stack_skipnr(other_info->stack_entries,
348 						other_info->num_stack_entries);
349 		other_frame = other_info->stack_entries[other_skipnr];
350 
351 		/* @value_change is only known for the other thread */
352 		if (skip_report(value_change, other_frame))
353 			return;
354 	}
355 
356 	if (rate_limit_report(this_frame, other_frame))
357 		return;
358 
359 	/* Print report header. */
360 	pr_err("==================================================================\n");
361 	if (other_info) {
362 		int cmp;
363 
364 		/*
365 		 * Order functions lexographically for consistent bug titles.
366 		 * Do not print offset of functions to keep title short.
367 		 */
368 		cmp = sym_strcmp((void *)other_frame, (void *)this_frame);
369 		pr_err("BUG: KCSAN: %s in %ps / %ps\n",
370 		       get_bug_type(ai->access_type | other_info->ai.access_type),
371 		       (void *)(cmp < 0 ? other_frame : this_frame),
372 		       (void *)(cmp < 0 ? this_frame : other_frame));
373 	} else {
374 		pr_err("BUG: KCSAN: %s in %pS\n", get_bug_type(ai->access_type),
375 		       (void *)this_frame);
376 	}
377 
378 	pr_err("\n");
379 
380 	/* Print information about the racing accesses. */
381 	if (other_info) {
382 		pr_err("%s to 0x%px of %zu bytes by %s on cpu %i:\n",
383 		       get_access_type(other_info->ai.access_type), other_info->ai.ptr,
384 		       other_info->ai.size, get_thread_desc(other_info->ai.task_pid),
385 		       other_info->ai.cpu_id);
386 
387 		/* Print the other thread's stack trace. */
388 		stack_trace_print(other_info->stack_entries + other_skipnr,
389 				  other_info->num_stack_entries - other_skipnr,
390 				  0);
391 
392 		if (IS_ENABLED(CONFIG_KCSAN_VERBOSE))
393 			print_verbose_info(other_info->task);
394 
395 		pr_err("\n");
396 		pr_err("%s to 0x%px of %zu bytes by %s on cpu %i:\n",
397 		       get_access_type(ai->access_type), ai->ptr, ai->size,
398 		       get_thread_desc(ai->task_pid), ai->cpu_id);
399 	} else {
400 		pr_err("race at unknown origin, with %s to 0x%px of %zu bytes by %s on cpu %i:\n",
401 		       get_access_type(ai->access_type), ai->ptr, ai->size,
402 		       get_thread_desc(ai->task_pid), ai->cpu_id);
403 	}
404 	/* Print stack trace of this thread. */
405 	stack_trace_print(stack_entries + skipnr, num_stack_entries - skipnr,
406 			  0);
407 
408 	if (IS_ENABLED(CONFIG_KCSAN_VERBOSE))
409 		print_verbose_info(current);
410 
411 	/* Print observed value change. */
412 	if (ai->size <= 8) {
413 		int hex_len = ai->size * 2;
414 		u64 diff = old ^ new;
415 
416 		if (mask)
417 			diff &= mask;
418 		if (diff) {
419 			pr_err("\n");
420 			pr_err("value changed: 0x%0*llx -> 0x%0*llx\n",
421 			       hex_len, old, hex_len, new);
422 			if (mask) {
423 				pr_err(" bits changed: 0x%0*llx with mask 0x%0*llx\n",
424 				       hex_len, diff, hex_len, mask);
425 			}
426 		}
427 	}
428 
429 	/* Print report footer. */
430 	pr_err("\n");
431 	pr_err("Reported by Kernel Concurrency Sanitizer on:\n");
432 	dump_stack_print_info(KERN_DEFAULT);
433 	pr_err("==================================================================\n");
434 
435 	if (panic_on_warn)
436 		panic("panic_on_warn set ...\n");
437 }
438 
439 static void release_report(unsigned long *flags, struct other_info *other_info)
440 {
441 	/*
442 	 * Use size to denote valid/invalid, since KCSAN entirely ignores
443 	 * 0-sized accesses.
444 	 */
445 	other_info->ai.size = 0;
446 	raw_spin_unlock_irqrestore(&report_lock, *flags);
447 }
448 
449 /*
450  * Sets @other_info->task and awaits consumption of @other_info.
451  *
452  * Precondition: report_lock is held.
453  * Postcondition: report_lock is held.
454  */
455 static void set_other_info_task_blocking(unsigned long *flags,
456 					 const struct access_info *ai,
457 					 struct other_info *other_info)
458 {
459 	/*
460 	 * We may be instrumenting a code-path where current->state is already
461 	 * something other than TASK_RUNNING.
462 	 */
463 	const bool is_running = task_is_running(current);
464 	/*
465 	 * To avoid deadlock in case we are in an interrupt here and this is a
466 	 * race with a task on the same CPU (KCSAN_INTERRUPT_WATCHER), provide a
467 	 * timeout to ensure this works in all contexts.
468 	 *
469 	 * Await approximately the worst case delay of the reporting thread (if
470 	 * we are not interrupted).
471 	 */
472 	int timeout = max(kcsan_udelay_task, kcsan_udelay_interrupt);
473 
474 	other_info->task = current;
475 	do {
476 		if (is_running) {
477 			/*
478 			 * Let lockdep know the real task is sleeping, to print
479 			 * the held locks (recall we turned lockdep off, so
480 			 * locking/unlocking @report_lock won't be recorded).
481 			 */
482 			set_current_state(TASK_UNINTERRUPTIBLE);
483 		}
484 		raw_spin_unlock_irqrestore(&report_lock, *flags);
485 		/*
486 		 * We cannot call schedule() since we also cannot reliably
487 		 * determine if sleeping here is permitted -- see in_atomic().
488 		 */
489 
490 		udelay(1);
491 		raw_spin_lock_irqsave(&report_lock, *flags);
492 		if (timeout-- < 0) {
493 			/*
494 			 * Abort. Reset @other_info->task to NULL, since it
495 			 * appears the other thread is still going to consume
496 			 * it. It will result in no verbose info printed for
497 			 * this task.
498 			 */
499 			other_info->task = NULL;
500 			break;
501 		}
502 		/*
503 		 * If invalid, or @ptr nor @current matches, then @other_info
504 		 * has been consumed and we may continue. If not, retry.
505 		 */
506 	} while (other_info->ai.size && other_info->ai.ptr == ai->ptr &&
507 		 other_info->task == current);
508 	if (is_running)
509 		set_current_state(TASK_RUNNING);
510 }
511 
512 /* Populate @other_info; requires that the provided @other_info not in use. */
513 static void prepare_report_producer(unsigned long *flags,
514 				    const struct access_info *ai,
515 				    struct other_info *other_info)
516 {
517 	raw_spin_lock_irqsave(&report_lock, *flags);
518 
519 	/*
520 	 * The same @other_infos entry cannot be used concurrently, because
521 	 * there is a one-to-one mapping to watchpoint slots (@watchpoints in
522 	 * core.c), and a watchpoint is only released for reuse after reporting
523 	 * is done by the consumer of @other_info. Therefore, it is impossible
524 	 * for another concurrent prepare_report_producer() to set the same
525 	 * @other_info, and are guaranteed exclusivity for the @other_infos
526 	 * entry pointed to by @other_info.
527 	 *
528 	 * To check this property holds, size should never be non-zero here,
529 	 * because every consumer of struct other_info resets size to 0 in
530 	 * release_report().
531 	 */
532 	WARN_ON(other_info->ai.size);
533 
534 	other_info->ai = *ai;
535 	other_info->num_stack_entries = stack_trace_save(other_info->stack_entries, NUM_STACK_ENTRIES, 2);
536 
537 	if (IS_ENABLED(CONFIG_KCSAN_VERBOSE))
538 		set_other_info_task_blocking(flags, ai, other_info);
539 
540 	raw_spin_unlock_irqrestore(&report_lock, *flags);
541 }
542 
543 /* Awaits producer to fill @other_info and then returns. */
544 static bool prepare_report_consumer(unsigned long *flags,
545 				    const struct access_info *ai,
546 				    struct other_info *other_info)
547 {
548 
549 	raw_spin_lock_irqsave(&report_lock, *flags);
550 	while (!other_info->ai.size) { /* Await valid @other_info. */
551 		raw_spin_unlock_irqrestore(&report_lock, *flags);
552 		cpu_relax();
553 		raw_spin_lock_irqsave(&report_lock, *flags);
554 	}
555 
556 	/* Should always have a matching access based on watchpoint encoding. */
557 	if (WARN_ON(!matching_access((unsigned long)other_info->ai.ptr & WATCHPOINT_ADDR_MASK, other_info->ai.size,
558 				     (unsigned long)ai->ptr & WATCHPOINT_ADDR_MASK, ai->size)))
559 		goto discard;
560 
561 	if (!matching_access((unsigned long)other_info->ai.ptr, other_info->ai.size,
562 			     (unsigned long)ai->ptr, ai->size)) {
563 		/*
564 		 * If the actual accesses to not match, this was a false
565 		 * positive due to watchpoint encoding.
566 		 */
567 		atomic_long_inc(&kcsan_counters[KCSAN_COUNTER_ENCODING_FALSE_POSITIVES]);
568 		goto discard;
569 	}
570 
571 	return true;
572 
573 discard:
574 	release_report(flags, other_info);
575 	return false;
576 }
577 
578 static struct access_info prepare_access_info(const volatile void *ptr, size_t size,
579 					      int access_type)
580 {
581 	return (struct access_info) {
582 		.ptr		= ptr,
583 		.size		= size,
584 		.access_type	= access_type,
585 		.task_pid	= in_task() ? task_pid_nr(current) : -1,
586 		.cpu_id		= raw_smp_processor_id()
587 	};
588 }
589 
590 void kcsan_report_set_info(const volatile void *ptr, size_t size, int access_type,
591 			   int watchpoint_idx)
592 {
593 	const struct access_info ai = prepare_access_info(ptr, size, access_type);
594 	unsigned long flags;
595 
596 	kcsan_disable_current();
597 	lockdep_off(); /* See kcsan_report_known_origin(). */
598 
599 	prepare_report_producer(&flags, &ai, &other_infos[watchpoint_idx]);
600 
601 	lockdep_on();
602 	kcsan_enable_current();
603 }
604 
605 void kcsan_report_known_origin(const volatile void *ptr, size_t size, int access_type,
606 			       enum kcsan_value_change value_change, int watchpoint_idx,
607 			       u64 old, u64 new, u64 mask)
608 {
609 	const struct access_info ai = prepare_access_info(ptr, size, access_type);
610 	struct other_info *other_info = &other_infos[watchpoint_idx];
611 	unsigned long flags = 0;
612 
613 	kcsan_disable_current();
614 	/*
615 	 * Because we may generate reports when we're in scheduler code, the use
616 	 * of printk() could deadlock. Until such time that all printing code
617 	 * called in print_report() is scheduler-safe, accept the risk, and just
618 	 * get our message out. As such, also disable lockdep to hide the
619 	 * warning, and avoid disabling lockdep for the rest of the kernel.
620 	 */
621 	lockdep_off();
622 
623 	if (!prepare_report_consumer(&flags, &ai, other_info))
624 		goto out;
625 	/*
626 	 * Never report if value_change is FALSE, only when it is
627 	 * either TRUE or MAYBE. In case of MAYBE, further filtering may
628 	 * be done once we know the full stack trace in print_report().
629 	 */
630 	if (value_change != KCSAN_VALUE_CHANGE_FALSE)
631 		print_report(value_change, &ai, other_info, old, new, mask);
632 
633 	release_report(&flags, other_info);
634 out:
635 	lockdep_on();
636 	kcsan_enable_current();
637 }
638 
639 void kcsan_report_unknown_origin(const volatile void *ptr, size_t size, int access_type,
640 				 u64 old, u64 new, u64 mask)
641 {
642 	const struct access_info ai = prepare_access_info(ptr, size, access_type);
643 	unsigned long flags;
644 
645 	kcsan_disable_current();
646 	lockdep_off(); /* See kcsan_report_known_origin(). */
647 
648 	raw_spin_lock_irqsave(&report_lock, flags);
649 	print_report(KCSAN_VALUE_CHANGE_TRUE, &ai, NULL, old, new, mask);
650 	raw_spin_unlock_irqrestore(&report_lock, flags);
651 
652 	lockdep_on();
653 	kcsan_enable_current();
654 }
655