xref: /openbmc/linux/lib/stackdepot.c (revision e2aa5e65)
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 
36 #define DEPOT_STACK_BITS (sizeof(depot_stack_handle_t) * 8)
37 
38 #define STACK_ALLOC_NULL_PROTECTION_BITS 1
39 #define STACK_ALLOC_ORDER 2 /* 'Slab' size order for stack depot, 4 pages */
40 #define STACK_ALLOC_SIZE (1LL << (PAGE_SHIFT + STACK_ALLOC_ORDER))
41 #define STACK_ALLOC_ALIGN 4
42 #define STACK_ALLOC_OFFSET_BITS (STACK_ALLOC_ORDER + PAGE_SHIFT - \
43 					STACK_ALLOC_ALIGN)
44 #define STACK_ALLOC_INDEX_BITS (DEPOT_STACK_BITS - \
45 		STACK_ALLOC_NULL_PROTECTION_BITS - STACK_ALLOC_OFFSET_BITS)
46 #define STACK_ALLOC_SLABS_CAP 8192
47 #define STACK_ALLOC_MAX_SLABS \
48 	(((1LL << (STACK_ALLOC_INDEX_BITS)) < STACK_ALLOC_SLABS_CAP) ? \
49 	 (1LL << (STACK_ALLOC_INDEX_BITS)) : STACK_ALLOC_SLABS_CAP)
50 
51 /* The compact structure to store the reference to stacks. */
52 union handle_parts {
53 	depot_stack_handle_t handle;
54 	struct {
55 		u32 slabindex : STACK_ALLOC_INDEX_BITS;
56 		u32 offset : STACK_ALLOC_OFFSET_BITS;
57 		u32 valid : STACK_ALLOC_NULL_PROTECTION_BITS;
58 	};
59 };
60 
61 struct stack_record {
62 	struct stack_record *next;	/* Link in the hashtable */
63 	u32 hash;			/* Hash in the hastable */
64 	u32 size;			/* Number of frames in the stack */
65 	union handle_parts handle;
66 	unsigned long entries[];	/* Variable-sized array of entries. */
67 };
68 
69 static void *stack_slabs[STACK_ALLOC_MAX_SLABS];
70 
71 static int depot_index;
72 static int next_slab_inited;
73 static size_t depot_offset;
74 static DEFINE_RAW_SPINLOCK(depot_lock);
75 
76 static bool init_stack_slab(void **prealloc)
77 {
78 	if (!*prealloc)
79 		return false;
80 	/*
81 	 * This smp_load_acquire() pairs with smp_store_release() to
82 	 * |next_slab_inited| below and in depot_alloc_stack().
83 	 */
84 	if (smp_load_acquire(&next_slab_inited))
85 		return true;
86 	if (stack_slabs[depot_index] == NULL) {
87 		stack_slabs[depot_index] = *prealloc;
88 		*prealloc = NULL;
89 	} else {
90 		/* If this is the last depot slab, do not touch the next one. */
91 		if (depot_index + 1 < STACK_ALLOC_MAX_SLABS) {
92 			stack_slabs[depot_index + 1] = *prealloc;
93 			*prealloc = NULL;
94 		}
95 		/*
96 		 * This smp_store_release pairs with smp_load_acquire() from
97 		 * |next_slab_inited| above and in stack_depot_save().
98 		 */
99 		smp_store_release(&next_slab_inited, 1);
100 	}
101 	return true;
102 }
103 
104 /* Allocation of a new stack in raw storage */
105 static struct stack_record *
106 depot_alloc_stack(unsigned long *entries, int size, u32 hash, void **prealloc)
107 {
108 	struct stack_record *stack;
109 	size_t required_size = struct_size(stack, entries, size);
110 
111 	required_size = ALIGN(required_size, 1 << STACK_ALLOC_ALIGN);
112 
113 	if (unlikely(depot_offset + required_size > STACK_ALLOC_SIZE)) {
114 		if (unlikely(depot_index + 1 >= STACK_ALLOC_MAX_SLABS)) {
115 			WARN_ONCE(1, "Stack depot reached limit capacity");
116 			return NULL;
117 		}
118 		depot_index++;
119 		depot_offset = 0;
120 		/*
121 		 * smp_store_release() here pairs with smp_load_acquire() from
122 		 * |next_slab_inited| in stack_depot_save() and
123 		 * init_stack_slab().
124 		 */
125 		if (depot_index + 1 < STACK_ALLOC_MAX_SLABS)
126 			smp_store_release(&next_slab_inited, 0);
127 	}
128 	init_stack_slab(prealloc);
129 	if (stack_slabs[depot_index] == NULL)
130 		return NULL;
131 
132 	stack = stack_slabs[depot_index] + depot_offset;
133 
134 	stack->hash = hash;
135 	stack->size = size;
136 	stack->handle.slabindex = depot_index;
137 	stack->handle.offset = depot_offset >> STACK_ALLOC_ALIGN;
138 	stack->handle.valid = 1;
139 	memcpy(stack->entries, entries, flex_array_size(stack, entries, size));
140 	depot_offset += required_size;
141 
142 	return stack;
143 }
144 
145 #define STACK_HASH_SIZE (1L << CONFIG_STACK_HASH_ORDER)
146 #define STACK_HASH_MASK (STACK_HASH_SIZE - 1)
147 #define STACK_HASH_SEED 0x9747b28c
148 
149 static bool stack_depot_disable;
150 static struct stack_record **stack_table;
151 
152 static int __init is_stack_depot_disabled(char *str)
153 {
154 	int ret;
155 
156 	ret = kstrtobool(str, &stack_depot_disable);
157 	if (!ret && stack_depot_disable) {
158 		pr_info("Stack Depot is disabled\n");
159 		stack_table = NULL;
160 	}
161 	return 0;
162 }
163 early_param("stack_depot_disable", is_stack_depot_disabled);
164 
165 /*
166  * __ref because of memblock_alloc(), which will not be actually called after
167  * the __init code is gone, because at that point slab_is_available() is true
168  */
169 __ref int stack_depot_init(void)
170 {
171 	static DEFINE_MUTEX(stack_depot_init_mutex);
172 
173 	mutex_lock(&stack_depot_init_mutex);
174 	if (!stack_depot_disable && !stack_table) {
175 		size_t size = (STACK_HASH_SIZE * sizeof(struct stack_record *));
176 		int i;
177 
178 		if (slab_is_available()) {
179 			pr_info("Stack Depot allocating hash table with kvmalloc\n");
180 			stack_table = kvmalloc(size, GFP_KERNEL);
181 		} else {
182 			pr_info("Stack Depot allocating hash table with memblock_alloc\n");
183 			stack_table = memblock_alloc(size, SMP_CACHE_BYTES);
184 		}
185 		if (stack_table) {
186 			for (i = 0; i < STACK_HASH_SIZE;  i++)
187 				stack_table[i] = NULL;
188 		} else {
189 			pr_err("Stack Depot hash table allocation failed, disabling\n");
190 			stack_depot_disable = true;
191 			mutex_unlock(&stack_depot_init_mutex);
192 			return -ENOMEM;
193 		}
194 	}
195 	mutex_unlock(&stack_depot_init_mutex);
196 	return 0;
197 }
198 EXPORT_SYMBOL_GPL(stack_depot_init);
199 
200 /* Calculate hash for a stack */
201 static inline u32 hash_stack(unsigned long *entries, unsigned int size)
202 {
203 	return jhash2((u32 *)entries,
204 		      array_size(size,  sizeof(*entries)) / sizeof(u32),
205 		      STACK_HASH_SEED);
206 }
207 
208 /* Use our own, non-instrumented version of memcmp().
209  *
210  * We actually don't care about the order, just the equality.
211  */
212 static inline
213 int stackdepot_memcmp(const unsigned long *u1, const unsigned long *u2,
214 			unsigned int n)
215 {
216 	for ( ; n-- ; u1++, u2++) {
217 		if (*u1 != *u2)
218 			return 1;
219 	}
220 	return 0;
221 }
222 
223 /* Find a stack that is equal to the one stored in entries in the hash */
224 static inline struct stack_record *find_stack(struct stack_record *bucket,
225 					     unsigned long *entries, int size,
226 					     u32 hash)
227 {
228 	struct stack_record *found;
229 
230 	for (found = bucket; found; found = found->next) {
231 		if (found->hash == hash &&
232 		    found->size == size &&
233 		    !stackdepot_memcmp(entries, found->entries, size))
234 			return found;
235 	}
236 	return NULL;
237 }
238 
239 /**
240  * stack_depot_snprint - print stack entries from a depot into a buffer
241  *
242  * @handle:	Stack depot handle which was returned from
243  *		stack_depot_save().
244  * @buf:	Pointer to the print buffer
245  *
246  * @size:	Size of the print buffer
247  *
248  * @spaces:	Number of leading spaces to print
249  *
250  * Return:	Number of bytes printed.
251  */
252 int stack_depot_snprint(depot_stack_handle_t handle, char *buf, size_t size,
253 		       int spaces)
254 {
255 	unsigned long *entries;
256 	unsigned int nr_entries;
257 
258 	nr_entries = stack_depot_fetch(handle, &entries);
259 	return nr_entries ? stack_trace_snprint(buf, size, entries, nr_entries,
260 						spaces) : 0;
261 }
262 EXPORT_SYMBOL_GPL(stack_depot_snprint);
263 
264 /**
265  * stack_depot_print - print stack entries from a depot
266  *
267  * @stack:		Stack depot handle which was returned from
268  *			stack_depot_save().
269  *
270  */
271 void stack_depot_print(depot_stack_handle_t stack)
272 {
273 	unsigned long *entries;
274 	unsigned int nr_entries;
275 
276 	nr_entries = stack_depot_fetch(stack, &entries);
277 	if (nr_entries > 0)
278 		stack_trace_print(entries, nr_entries, 0);
279 }
280 EXPORT_SYMBOL_GPL(stack_depot_print);
281 
282 /**
283  * stack_depot_fetch - Fetch stack entries from a depot
284  *
285  * @handle:		Stack depot handle which was returned from
286  *			stack_depot_save().
287  * @entries:		Pointer to store the entries address
288  *
289  * Return: The number of trace entries for this depot.
290  */
291 unsigned int stack_depot_fetch(depot_stack_handle_t handle,
292 			       unsigned long **entries)
293 {
294 	union handle_parts parts = { .handle = handle };
295 	void *slab;
296 	size_t offset = parts.offset << STACK_ALLOC_ALIGN;
297 	struct stack_record *stack;
298 
299 	*entries = NULL;
300 	if (!handle)
301 		return 0;
302 
303 	if (parts.slabindex > depot_index) {
304 		WARN(1, "slab index %d out of bounds (%d) for stack id %08x\n",
305 			parts.slabindex, depot_index, handle);
306 		return 0;
307 	}
308 	slab = stack_slabs[parts.slabindex];
309 	if (!slab)
310 		return 0;
311 	stack = slab + offset;
312 
313 	*entries = stack->entries;
314 	return stack->size;
315 }
316 EXPORT_SYMBOL_GPL(stack_depot_fetch);
317 
318 /**
319  * __stack_depot_save - Save a stack trace from an array
320  *
321  * @entries:		Pointer to storage array
322  * @nr_entries:		Size of the storage array
323  * @alloc_flags:	Allocation gfp flags
324  * @can_alloc:		Allocate stack slabs (increased chance of failure if false)
325  *
326  * Saves a stack trace from @entries array of size @nr_entries. If @can_alloc is
327  * %true, is allowed to replenish the stack slab pool in case no space is left
328  * (allocates using GFP flags of @alloc_flags). If @can_alloc is %false, avoids
329  * any allocations and will fail if no space is left to store the stack trace.
330  *
331  * If the stack trace in @entries is from an interrupt, only the portion up to
332  * interrupt entry is saved.
333  *
334  * Context: Any context, but setting @can_alloc to %false is required if
335  *          alloc_pages() cannot be used from the current context. Currently
336  *          this is the case from contexts where neither %GFP_ATOMIC nor
337  *          %GFP_NOWAIT can be used (NMI, raw_spin_lock).
338  *
339  * Return: The handle of the stack struct stored in depot, 0 on failure.
340  */
341 depot_stack_handle_t __stack_depot_save(unsigned long *entries,
342 					unsigned int nr_entries,
343 					gfp_t alloc_flags, bool can_alloc)
344 {
345 	struct stack_record *found = NULL, **bucket;
346 	depot_stack_handle_t retval = 0;
347 	struct page *page = NULL;
348 	void *prealloc = NULL;
349 	unsigned long flags;
350 	u32 hash;
351 
352 	/*
353 	 * If this stack trace is from an interrupt, including anything before
354 	 * interrupt entry usually leads to unbounded stackdepot growth.
355 	 *
356 	 * Because use of filter_irq_stacks() is a requirement to ensure
357 	 * stackdepot can efficiently deduplicate interrupt stacks, always
358 	 * filter_irq_stacks() to simplify all callers' use of stackdepot.
359 	 */
360 	nr_entries = filter_irq_stacks(entries, nr_entries);
361 
362 	if (unlikely(nr_entries == 0) || stack_depot_disable)
363 		goto fast_exit;
364 
365 	hash = hash_stack(entries, nr_entries);
366 	bucket = &stack_table[hash & STACK_HASH_MASK];
367 
368 	/*
369 	 * Fast path: look the stack trace up without locking.
370 	 * The smp_load_acquire() here pairs with smp_store_release() to
371 	 * |bucket| below.
372 	 */
373 	found = find_stack(smp_load_acquire(bucket), entries,
374 			   nr_entries, hash);
375 	if (found)
376 		goto exit;
377 
378 	/*
379 	 * Check if the current or the next stack slab need to be initialized.
380 	 * If so, allocate the memory - we won't be able to do that under the
381 	 * lock.
382 	 *
383 	 * The smp_load_acquire() here pairs with smp_store_release() to
384 	 * |next_slab_inited| in depot_alloc_stack() and init_stack_slab().
385 	 */
386 	if (unlikely(can_alloc && !smp_load_acquire(&next_slab_inited))) {
387 		/*
388 		 * Zero out zone modifiers, as we don't have specific zone
389 		 * requirements. Keep the flags related to allocation in atomic
390 		 * contexts and I/O.
391 		 */
392 		alloc_flags &= ~GFP_ZONEMASK;
393 		alloc_flags &= (GFP_ATOMIC | GFP_KERNEL);
394 		alloc_flags |= __GFP_NOWARN;
395 		page = alloc_pages(alloc_flags, STACK_ALLOC_ORDER);
396 		if (page)
397 			prealloc = page_address(page);
398 	}
399 
400 	raw_spin_lock_irqsave(&depot_lock, flags);
401 
402 	found = find_stack(*bucket, entries, nr_entries, hash);
403 	if (!found) {
404 		struct stack_record *new = depot_alloc_stack(entries, nr_entries, hash, &prealloc);
405 
406 		if (new) {
407 			new->next = *bucket;
408 			/*
409 			 * This smp_store_release() pairs with
410 			 * smp_load_acquire() from |bucket| above.
411 			 */
412 			smp_store_release(bucket, new);
413 			found = new;
414 		}
415 	} else if (prealloc) {
416 		/*
417 		 * We didn't need to store this stack trace, but let's keep
418 		 * the preallocated memory for the future.
419 		 */
420 		WARN_ON(!init_stack_slab(&prealloc));
421 	}
422 
423 	raw_spin_unlock_irqrestore(&depot_lock, flags);
424 exit:
425 	if (prealloc) {
426 		/* Nobody used this memory, ok to free it. */
427 		free_pages((unsigned long)prealloc, STACK_ALLOC_ORDER);
428 	}
429 	if (found)
430 		retval = found->handle.handle;
431 fast_exit:
432 	return retval;
433 }
434 EXPORT_SYMBOL_GPL(__stack_depot_save);
435 
436 /**
437  * stack_depot_save - Save a stack trace from an array
438  *
439  * @entries:		Pointer to storage array
440  * @nr_entries:		Size of the storage array
441  * @alloc_flags:	Allocation gfp flags
442  *
443  * Context: Contexts where allocations via alloc_pages() are allowed.
444  *          See __stack_depot_save() for more details.
445  *
446  * Return: The handle of the stack struct stored in depot, 0 on failure.
447  */
448 depot_stack_handle_t stack_depot_save(unsigned long *entries,
449 				      unsigned int nr_entries,
450 				      gfp_t alloc_flags)
451 {
452 	return __stack_depot_save(entries, nr_entries, alloc_flags, true);
453 }
454 EXPORT_SYMBOL_GPL(stack_depot_save);
455