xref: /openbmc/linux/lib/stackdepot.c (revision 77423a62)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Generic stack depot for storing stack traces.
4  *
5  * Some debugging tools need to save stack traces of certain events which can
6  * be later presented to the user. For example, KASAN needs to safe alloc and
7  * free stacks for each object, but storing two stack traces per object
8  * requires too much memory (e.g. SLUB_DEBUG needs 256 bytes per object for
9  * that).
10  *
11  * Instead, stack depot maintains a hashtable of unique stacktraces. Since alloc
12  * and free stacks repeat a lot, we save about 100x space.
13  * Stacks are never removed from depot, so we store them contiguously one after
14  * another in a contiguous memory allocation.
15  *
16  * Author: Alexander Potapenko <glider@google.com>
17  * Copyright (C) 2016 Google, Inc.
18  *
19  * Based on code by Dmitry Chernenkov.
20  */
21 
22 #include <linux/gfp.h>
23 #include <linux/jhash.h>
24 #include <linux/kernel.h>
25 #include <linux/mm.h>
26 #include <linux/mutex.h>
27 #include <linux/percpu.h>
28 #include <linux/printk.h>
29 #include <linux/slab.h>
30 #include <linux/stacktrace.h>
31 #include <linux/stackdepot.h>
32 #include <linux/string.h>
33 #include <linux/types.h>
34 #include <linux/memblock.h>
35 #include <linux/kasan-enabled.h>
36 
37 #define DEPOT_STACK_BITS (sizeof(depot_stack_handle_t) * 8)
38 
39 #define STACK_ALLOC_NULL_PROTECTION_BITS 1
40 #define STACK_ALLOC_ORDER 2 /* 'Slab' size order for stack depot, 4 pages */
41 #define STACK_ALLOC_SIZE (1LL << (PAGE_SHIFT + STACK_ALLOC_ORDER))
42 #define STACK_ALLOC_ALIGN 4
43 #define STACK_ALLOC_OFFSET_BITS (STACK_ALLOC_ORDER + PAGE_SHIFT - \
44 					STACK_ALLOC_ALIGN)
45 #define STACK_ALLOC_INDEX_BITS (DEPOT_STACK_BITS - \
46 		STACK_ALLOC_NULL_PROTECTION_BITS - STACK_ALLOC_OFFSET_BITS)
47 #define STACK_ALLOC_SLABS_CAP 8192
48 #define STACK_ALLOC_MAX_SLABS \
49 	(((1LL << (STACK_ALLOC_INDEX_BITS)) < STACK_ALLOC_SLABS_CAP) ? \
50 	 (1LL << (STACK_ALLOC_INDEX_BITS)) : STACK_ALLOC_SLABS_CAP)
51 
52 /* The compact structure to store the reference to stacks. */
53 union handle_parts {
54 	depot_stack_handle_t handle;
55 	struct {
56 		u32 slabindex : STACK_ALLOC_INDEX_BITS;
57 		u32 offset : STACK_ALLOC_OFFSET_BITS;
58 		u32 valid : STACK_ALLOC_NULL_PROTECTION_BITS;
59 	};
60 };
61 
62 struct stack_record {
63 	struct stack_record *next;	/* Link in the hashtable */
64 	u32 hash;			/* Hash in the hastable */
65 	u32 size;			/* Number of frames in the stack */
66 	union handle_parts handle;
67 	unsigned long entries[];	/* Variable-sized array of entries. */
68 };
69 
70 static bool __stack_depot_want_early_init __initdata = IS_ENABLED(CONFIG_STACKDEPOT_ALWAYS_INIT);
71 static bool __stack_depot_early_init_passed __initdata;
72 
73 static void *stack_slabs[STACK_ALLOC_MAX_SLABS];
74 
75 static int depot_index;
76 static int next_slab_inited;
77 static size_t depot_offset;
78 static DEFINE_RAW_SPINLOCK(depot_lock);
79 
80 static bool init_stack_slab(void **prealloc)
81 {
82 	if (!*prealloc)
83 		return false;
84 	/*
85 	 * This smp_load_acquire() pairs with smp_store_release() to
86 	 * |next_slab_inited| below and in depot_alloc_stack().
87 	 */
88 	if (smp_load_acquire(&next_slab_inited))
89 		return true;
90 	if (stack_slabs[depot_index] == NULL) {
91 		stack_slabs[depot_index] = *prealloc;
92 		*prealloc = NULL;
93 	} else {
94 		/* If this is the last depot slab, do not touch the next one. */
95 		if (depot_index + 1 < STACK_ALLOC_MAX_SLABS) {
96 			stack_slabs[depot_index + 1] = *prealloc;
97 			*prealloc = NULL;
98 		}
99 		/*
100 		 * This smp_store_release pairs with smp_load_acquire() from
101 		 * |next_slab_inited| above and in stack_depot_save().
102 		 */
103 		smp_store_release(&next_slab_inited, 1);
104 	}
105 	return true;
106 }
107 
108 /* Allocation of a new stack in raw storage */
109 static struct stack_record *
110 depot_alloc_stack(unsigned long *entries, int size, u32 hash, void **prealloc)
111 {
112 	struct stack_record *stack;
113 	size_t required_size = struct_size(stack, entries, size);
114 
115 	required_size = ALIGN(required_size, 1 << STACK_ALLOC_ALIGN);
116 
117 	if (unlikely(depot_offset + required_size > STACK_ALLOC_SIZE)) {
118 		if (unlikely(depot_index + 1 >= STACK_ALLOC_MAX_SLABS)) {
119 			WARN_ONCE(1, "Stack depot reached limit capacity");
120 			return NULL;
121 		}
122 		depot_index++;
123 		depot_offset = 0;
124 		/*
125 		 * smp_store_release() here pairs with smp_load_acquire() from
126 		 * |next_slab_inited| in stack_depot_save() and
127 		 * init_stack_slab().
128 		 */
129 		if (depot_index + 1 < STACK_ALLOC_MAX_SLABS)
130 			smp_store_release(&next_slab_inited, 0);
131 	}
132 	init_stack_slab(prealloc);
133 	if (stack_slabs[depot_index] == NULL)
134 		return NULL;
135 
136 	stack = stack_slabs[depot_index] + depot_offset;
137 
138 	stack->hash = hash;
139 	stack->size = size;
140 	stack->handle.slabindex = depot_index;
141 	stack->handle.offset = depot_offset >> STACK_ALLOC_ALIGN;
142 	stack->handle.valid = 1;
143 	memcpy(stack->entries, entries, flex_array_size(stack, entries, size));
144 	depot_offset += required_size;
145 
146 	return stack;
147 }
148 
149 /* one hash table bucket entry per 16kB of memory */
150 #define STACK_HASH_SCALE	14
151 /* limited between 4k and 1M buckets */
152 #define STACK_HASH_ORDER_MIN	12
153 #define STACK_HASH_ORDER_MAX	20
154 #define STACK_HASH_SEED 0x9747b28c
155 
156 static unsigned int stack_hash_order;
157 static unsigned int stack_hash_mask;
158 
159 static bool stack_depot_disable;
160 static struct stack_record **stack_table;
161 
162 static int __init is_stack_depot_disabled(char *str)
163 {
164 	int ret;
165 
166 	ret = kstrtobool(str, &stack_depot_disable);
167 	if (!ret && stack_depot_disable) {
168 		pr_info("Stack Depot is disabled\n");
169 		stack_table = NULL;
170 	}
171 	return 0;
172 }
173 early_param("stack_depot_disable", is_stack_depot_disabled);
174 
175 void __init stack_depot_want_early_init(void)
176 {
177 	/* Too late to request early init now */
178 	WARN_ON(__stack_depot_early_init_passed);
179 
180 	__stack_depot_want_early_init = true;
181 }
182 
183 int __init stack_depot_early_init(void)
184 {
185 	unsigned long entries = 0;
186 
187 	/* This is supposed to be called only once, from mm_init() */
188 	if (WARN_ON(__stack_depot_early_init_passed))
189 		return 0;
190 
191 	__stack_depot_early_init_passed = true;
192 
193 	if (kasan_enabled() && !stack_hash_order)
194 		stack_hash_order = STACK_HASH_ORDER_MAX;
195 
196 	if (!__stack_depot_want_early_init || stack_depot_disable)
197 		return 0;
198 
199 	if (stack_hash_order)
200 		entries = 1UL <<  stack_hash_order;
201 	stack_table = alloc_large_system_hash("stackdepot",
202 						sizeof(struct stack_record *),
203 						entries,
204 						STACK_HASH_SCALE,
205 						HASH_EARLY | HASH_ZERO,
206 						NULL,
207 						&stack_hash_mask,
208 						1UL << STACK_HASH_ORDER_MIN,
209 						1UL << STACK_HASH_ORDER_MAX);
210 
211 	if (!stack_table) {
212 		pr_err("Stack Depot hash table allocation failed, disabling\n");
213 		stack_depot_disable = true;
214 		return -ENOMEM;
215 	}
216 
217 	return 0;
218 }
219 
220 int stack_depot_init(void)
221 {
222 	static DEFINE_MUTEX(stack_depot_init_mutex);
223 	int ret = 0;
224 
225 	mutex_lock(&stack_depot_init_mutex);
226 	if (!stack_depot_disable && !stack_table) {
227 		unsigned long entries;
228 		int scale = STACK_HASH_SCALE;
229 
230 		if (stack_hash_order) {
231 			entries = 1UL << stack_hash_order;
232 		} else {
233 			entries = nr_free_buffer_pages();
234 			entries = roundup_pow_of_two(entries);
235 
236 			if (scale > PAGE_SHIFT)
237 				entries >>= (scale - PAGE_SHIFT);
238 			else
239 				entries <<= (PAGE_SHIFT - scale);
240 		}
241 
242 		if (entries < 1UL << STACK_HASH_ORDER_MIN)
243 			entries = 1UL << STACK_HASH_ORDER_MIN;
244 		if (entries > 1UL << STACK_HASH_ORDER_MAX)
245 			entries = 1UL << STACK_HASH_ORDER_MAX;
246 
247 		pr_info("Stack Depot allocating hash table of %lu entries with kvcalloc\n",
248 				entries);
249 		stack_table = kvcalloc(entries, sizeof(struct stack_record *), GFP_KERNEL);
250 		if (!stack_table) {
251 			pr_err("Stack Depot hash table allocation failed, disabling\n");
252 			stack_depot_disable = true;
253 			ret = -ENOMEM;
254 		}
255 		stack_hash_mask = entries - 1;
256 	}
257 	mutex_unlock(&stack_depot_init_mutex);
258 	return ret;
259 }
260 EXPORT_SYMBOL_GPL(stack_depot_init);
261 
262 /* Calculate hash for a stack */
263 static inline u32 hash_stack(unsigned long *entries, unsigned int size)
264 {
265 	return jhash2((u32 *)entries,
266 		      array_size(size,  sizeof(*entries)) / sizeof(u32),
267 		      STACK_HASH_SEED);
268 }
269 
270 /* Use our own, non-instrumented version of memcmp().
271  *
272  * We actually don't care about the order, just the equality.
273  */
274 static inline
275 int stackdepot_memcmp(const unsigned long *u1, const unsigned long *u2,
276 			unsigned int n)
277 {
278 	for ( ; n-- ; u1++, u2++) {
279 		if (*u1 != *u2)
280 			return 1;
281 	}
282 	return 0;
283 }
284 
285 /* Find a stack that is equal to the one stored in entries in the hash */
286 static inline struct stack_record *find_stack(struct stack_record *bucket,
287 					     unsigned long *entries, int size,
288 					     u32 hash)
289 {
290 	struct stack_record *found;
291 
292 	for (found = bucket; found; found = found->next) {
293 		if (found->hash == hash &&
294 		    found->size == size &&
295 		    !stackdepot_memcmp(entries, found->entries, size))
296 			return found;
297 	}
298 	return NULL;
299 }
300 
301 /**
302  * stack_depot_snprint - print stack entries from a depot into a buffer
303  *
304  * @handle:	Stack depot handle which was returned from
305  *		stack_depot_save().
306  * @buf:	Pointer to the print buffer
307  *
308  * @size:	Size of the print buffer
309  *
310  * @spaces:	Number of leading spaces to print
311  *
312  * Return:	Number of bytes printed.
313  */
314 int stack_depot_snprint(depot_stack_handle_t handle, char *buf, size_t size,
315 		       int spaces)
316 {
317 	unsigned long *entries;
318 	unsigned int nr_entries;
319 
320 	nr_entries = stack_depot_fetch(handle, &entries);
321 	return nr_entries ? stack_trace_snprint(buf, size, entries, nr_entries,
322 						spaces) : 0;
323 }
324 EXPORT_SYMBOL_GPL(stack_depot_snprint);
325 
326 /**
327  * stack_depot_print - print stack entries from a depot
328  *
329  * @stack:		Stack depot handle which was returned from
330  *			stack_depot_save().
331  *
332  */
333 void stack_depot_print(depot_stack_handle_t stack)
334 {
335 	unsigned long *entries;
336 	unsigned int nr_entries;
337 
338 	nr_entries = stack_depot_fetch(stack, &entries);
339 	if (nr_entries > 0)
340 		stack_trace_print(entries, nr_entries, 0);
341 }
342 EXPORT_SYMBOL_GPL(stack_depot_print);
343 
344 /**
345  * stack_depot_fetch - Fetch stack entries from a depot
346  *
347  * @handle:		Stack depot handle which was returned from
348  *			stack_depot_save().
349  * @entries:		Pointer to store the entries address
350  *
351  * Return: The number of trace entries for this depot.
352  */
353 unsigned int stack_depot_fetch(depot_stack_handle_t handle,
354 			       unsigned long **entries)
355 {
356 	union handle_parts parts = { .handle = handle };
357 	void *slab;
358 	size_t offset = parts.offset << STACK_ALLOC_ALIGN;
359 	struct stack_record *stack;
360 
361 	*entries = NULL;
362 	if (!handle)
363 		return 0;
364 
365 	if (parts.slabindex > depot_index) {
366 		WARN(1, "slab index %d out of bounds (%d) for stack id %08x\n",
367 			parts.slabindex, depot_index, handle);
368 		return 0;
369 	}
370 	slab = stack_slabs[parts.slabindex];
371 	if (!slab)
372 		return 0;
373 	stack = slab + offset;
374 
375 	*entries = stack->entries;
376 	return stack->size;
377 }
378 EXPORT_SYMBOL_GPL(stack_depot_fetch);
379 
380 /**
381  * __stack_depot_save - Save a stack trace from an array
382  *
383  * @entries:		Pointer to storage array
384  * @nr_entries:		Size of the storage array
385  * @alloc_flags:	Allocation gfp flags
386  * @can_alloc:		Allocate stack slabs (increased chance of failure if false)
387  *
388  * Saves a stack trace from @entries array of size @nr_entries. If @can_alloc is
389  * %true, is allowed to replenish the stack slab pool in case no space is left
390  * (allocates using GFP flags of @alloc_flags). If @can_alloc is %false, avoids
391  * any allocations and will fail if no space is left to store the stack trace.
392  *
393  * If the stack trace in @entries is from an interrupt, only the portion up to
394  * interrupt entry is saved.
395  *
396  * Context: Any context, but setting @can_alloc to %false is required if
397  *          alloc_pages() cannot be used from the current context. Currently
398  *          this is the case from contexts where neither %GFP_ATOMIC nor
399  *          %GFP_NOWAIT can be used (NMI, raw_spin_lock).
400  *
401  * Return: The handle of the stack struct stored in depot, 0 on failure.
402  */
403 depot_stack_handle_t __stack_depot_save(unsigned long *entries,
404 					unsigned int nr_entries,
405 					gfp_t alloc_flags, bool can_alloc)
406 {
407 	struct stack_record *found = NULL, **bucket;
408 	depot_stack_handle_t retval = 0;
409 	struct page *page = NULL;
410 	void *prealloc = NULL;
411 	unsigned long flags;
412 	u32 hash;
413 
414 	/*
415 	 * If this stack trace is from an interrupt, including anything before
416 	 * interrupt entry usually leads to unbounded stackdepot growth.
417 	 *
418 	 * Because use of filter_irq_stacks() is a requirement to ensure
419 	 * stackdepot can efficiently deduplicate interrupt stacks, always
420 	 * filter_irq_stacks() to simplify all callers' use of stackdepot.
421 	 */
422 	nr_entries = filter_irq_stacks(entries, nr_entries);
423 
424 	if (unlikely(nr_entries == 0) || stack_depot_disable)
425 		goto fast_exit;
426 
427 	hash = hash_stack(entries, nr_entries);
428 	bucket = &stack_table[hash & stack_hash_mask];
429 
430 	/*
431 	 * Fast path: look the stack trace up without locking.
432 	 * The smp_load_acquire() here pairs with smp_store_release() to
433 	 * |bucket| below.
434 	 */
435 	found = find_stack(smp_load_acquire(bucket), entries,
436 			   nr_entries, hash);
437 	if (found)
438 		goto exit;
439 
440 	/*
441 	 * Check if the current or the next stack slab need to be initialized.
442 	 * If so, allocate the memory - we won't be able to do that under the
443 	 * lock.
444 	 *
445 	 * The smp_load_acquire() here pairs with smp_store_release() to
446 	 * |next_slab_inited| in depot_alloc_stack() and init_stack_slab().
447 	 */
448 	if (unlikely(can_alloc && !smp_load_acquire(&next_slab_inited))) {
449 		/*
450 		 * Zero out zone modifiers, as we don't have specific zone
451 		 * requirements. Keep the flags related to allocation in atomic
452 		 * contexts and I/O.
453 		 */
454 		alloc_flags &= ~GFP_ZONEMASK;
455 		alloc_flags &= (GFP_ATOMIC | GFP_KERNEL);
456 		alloc_flags |= __GFP_NOWARN;
457 		page = alloc_pages(alloc_flags, STACK_ALLOC_ORDER);
458 		if (page)
459 			prealloc = page_address(page);
460 	}
461 
462 	raw_spin_lock_irqsave(&depot_lock, flags);
463 
464 	found = find_stack(*bucket, entries, nr_entries, hash);
465 	if (!found) {
466 		struct stack_record *new = depot_alloc_stack(entries, nr_entries, hash, &prealloc);
467 
468 		if (new) {
469 			new->next = *bucket;
470 			/*
471 			 * This smp_store_release() pairs with
472 			 * smp_load_acquire() from |bucket| above.
473 			 */
474 			smp_store_release(bucket, new);
475 			found = new;
476 		}
477 	} else if (prealloc) {
478 		/*
479 		 * We didn't need to store this stack trace, but let's keep
480 		 * the preallocated memory for the future.
481 		 */
482 		WARN_ON(!init_stack_slab(&prealloc));
483 	}
484 
485 	raw_spin_unlock_irqrestore(&depot_lock, flags);
486 exit:
487 	if (prealloc) {
488 		/* Nobody used this memory, ok to free it. */
489 		free_pages((unsigned long)prealloc, STACK_ALLOC_ORDER);
490 	}
491 	if (found)
492 		retval = found->handle.handle;
493 fast_exit:
494 	return retval;
495 }
496 EXPORT_SYMBOL_GPL(__stack_depot_save);
497 
498 /**
499  * stack_depot_save - Save a stack trace from an array
500  *
501  * @entries:		Pointer to storage array
502  * @nr_entries:		Size of the storage array
503  * @alloc_flags:	Allocation gfp flags
504  *
505  * Context: Contexts where allocations via alloc_pages() are allowed.
506  *          See __stack_depot_save() for more details.
507  *
508  * Return: The handle of the stack struct stored in depot, 0 on failure.
509  */
510 depot_stack_handle_t stack_depot_save(unsigned long *entries,
511 				      unsigned int nr_entries,
512 				      gfp_t alloc_flags)
513 {
514 	return __stack_depot_save(entries, nr_entries, alloc_flags, true);
515 }
516 EXPORT_SYMBOL_GPL(stack_depot_save);
517