1 // SPDX-License-Identifier: GPL-2.0-only
3 * Generic stack depot for storing stack traces.
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
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.
16 * Author: Alexander Potapenko <glider@google.com>
17 * Copyright (C) 2016 Google, Inc.
19 * Based on code by Dmitry Chernenkov.
22 #include <linux/gfp.h>
23 #include <linux/jhash.h>
24 #include <linux/kernel.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>
37 #define DEPOT_STACK_BITS (sizeof(depot_stack_handle_t) * 8)
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 - \
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)
52 /* The compact structure to store the reference to stacks. */
54 depot_stack_handle_t handle;
56 u32 slabindex : STACK_ALLOC_INDEX_BITS;
57 u32 offset : STACK_ALLOC_OFFSET_BITS;
58 u32 valid : STACK_ALLOC_NULL_PROTECTION_BITS;
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. */
70 static bool __stack_depot_want_early_init __initdata = IS_ENABLED(CONFIG_STACKDEPOT_ALWAYS_INIT);
71 static bool __stack_depot_early_init_passed __initdata;
73 static void *stack_slabs[STACK_ALLOC_MAX_SLABS];
75 static int depot_index;
76 static int next_slab_inited;
77 static size_t depot_offset;
78 static DEFINE_RAW_SPINLOCK(depot_lock);
80 static bool init_stack_slab(void **prealloc)
85 * This smp_load_acquire() pairs with smp_store_release() to
86 * |next_slab_inited| below and in depot_alloc_stack().
88 if (smp_load_acquire(&next_slab_inited))
90 if (stack_slabs[depot_index] == NULL) {
91 stack_slabs[depot_index] = *prealloc;
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;
100 * This smp_store_release pairs with smp_load_acquire() from
101 * |next_slab_inited| above and in stack_depot_save().
103 smp_store_release(&next_slab_inited, 1);
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)
112 struct stack_record *stack;
113 size_t required_size = struct_size(stack, entries, size);
115 required_size = ALIGN(required_size, 1 << STACK_ALLOC_ALIGN);
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");
125 * smp_store_release() here pairs with smp_load_acquire() from
126 * |next_slab_inited| in stack_depot_save() and
129 if (depot_index + 1 < STACK_ALLOC_MAX_SLABS)
130 smp_store_release(&next_slab_inited, 0);
132 init_stack_slab(prealloc);
133 if (stack_slabs[depot_index] == NULL)
136 stack = stack_slabs[depot_index] + depot_offset;
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;
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
156 static unsigned int stack_hash_order;
157 static unsigned int stack_hash_mask;
159 static bool stack_depot_disable;
160 static struct stack_record **stack_table;
162 static int __init is_stack_depot_disabled(char *str)
166 ret = kstrtobool(str, &stack_depot_disable);
167 if (!ret && stack_depot_disable) {
168 pr_info("Stack Depot is disabled\n");
173 early_param("stack_depot_disable", is_stack_depot_disabled);
175 void __init stack_depot_want_early_init(void)
177 /* Too late to request early init now */
178 WARN_ON(__stack_depot_early_init_passed);
180 __stack_depot_want_early_init = true;
183 int __init stack_depot_early_init(void)
185 unsigned long entries = 0;
187 /* This is supposed to be called only once, from mm_init() */
188 if (WARN_ON(__stack_depot_early_init_passed))
191 __stack_depot_early_init_passed = true;
193 if (kasan_enabled() && !stack_hash_order)
194 stack_hash_order = STACK_HASH_ORDER_MAX;
196 if (!__stack_depot_want_early_init || stack_depot_disable)
199 if (stack_hash_order)
200 entries = 1UL << stack_hash_order;
201 stack_table = alloc_large_system_hash("stackdepot",
202 sizeof(struct stack_record *),
205 HASH_EARLY | HASH_ZERO,
208 1UL << STACK_HASH_ORDER_MIN,
209 1UL << STACK_HASH_ORDER_MAX);
212 pr_err("Stack Depot hash table allocation failed, disabling\n");
213 stack_depot_disable = true;
220 int stack_depot_init(void)
222 static DEFINE_MUTEX(stack_depot_init_mutex);
225 mutex_lock(&stack_depot_init_mutex);
226 if (!stack_depot_disable && !stack_table) {
227 unsigned long entries;
228 int scale = STACK_HASH_SCALE;
230 if (stack_hash_order) {
231 entries = 1UL << stack_hash_order;
233 entries = nr_free_buffer_pages();
234 entries = roundup_pow_of_two(entries);
236 if (scale > PAGE_SHIFT)
237 entries >>= (scale - PAGE_SHIFT);
239 entries <<= (PAGE_SHIFT - scale);
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;
247 pr_info("Stack Depot allocating hash table of %lu entries with kvcalloc\n",
249 stack_table = kvcalloc(entries, sizeof(struct stack_record *), GFP_KERNEL);
251 pr_err("Stack Depot hash table allocation failed, disabling\n");
252 stack_depot_disable = true;
255 stack_hash_mask = entries - 1;
257 mutex_unlock(&stack_depot_init_mutex);
260 EXPORT_SYMBOL_GPL(stack_depot_init);
262 /* Calculate hash for a stack */
263 static inline u32 hash_stack(unsigned long *entries, unsigned int size)
265 return jhash2((u32 *)entries,
266 array_size(size, sizeof(*entries)) / sizeof(u32),
270 /* Use our own, non-instrumented version of memcmp().
272 * We actually don't care about the order, just the equality.
275 int stackdepot_memcmp(const unsigned long *u1, const unsigned long *u2,
278 for ( ; n-- ; u1++, u2++) {
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,
290 struct stack_record *found;
292 for (found = bucket; found; found = found->next) {
293 if (found->hash == hash &&
294 found->size == size &&
295 !stackdepot_memcmp(entries, found->entries, size))
302 * stack_depot_snprint - print stack entries from a depot into a buffer
304 * @handle: Stack depot handle which was returned from
305 * stack_depot_save().
306 * @buf: Pointer to the print buffer
308 * @size: Size of the print buffer
310 * @spaces: Number of leading spaces to print
312 * Return: Number of bytes printed.
314 int stack_depot_snprint(depot_stack_handle_t handle, char *buf, size_t size,
317 unsigned long *entries;
318 unsigned int nr_entries;
320 nr_entries = stack_depot_fetch(handle, &entries);
321 return nr_entries ? stack_trace_snprint(buf, size, entries, nr_entries,
324 EXPORT_SYMBOL_GPL(stack_depot_snprint);
327 * stack_depot_print - print stack entries from a depot
329 * @stack: Stack depot handle which was returned from
330 * stack_depot_save().
333 void stack_depot_print(depot_stack_handle_t stack)
335 unsigned long *entries;
336 unsigned int nr_entries;
338 nr_entries = stack_depot_fetch(stack, &entries);
340 stack_trace_print(entries, nr_entries, 0);
342 EXPORT_SYMBOL_GPL(stack_depot_print);
345 * stack_depot_fetch - Fetch stack entries from a depot
347 * @handle: Stack depot handle which was returned from
348 * stack_depot_save().
349 * @entries: Pointer to store the entries address
351 * Return: The number of trace entries for this depot.
353 unsigned int stack_depot_fetch(depot_stack_handle_t handle,
354 unsigned long **entries)
356 union handle_parts parts = { .handle = handle };
358 size_t offset = parts.offset << STACK_ALLOC_ALIGN;
359 struct stack_record *stack;
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);
370 slab = stack_slabs[parts.slabindex];
373 stack = slab + offset;
375 *entries = stack->entries;
378 EXPORT_SYMBOL_GPL(stack_depot_fetch);
381 * __stack_depot_save - Save a stack trace from an array
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)
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.
393 * If the stack trace in @entries is from an interrupt, only the portion up to
394 * interrupt entry is saved.
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).
401 * Return: The handle of the stack struct stored in depot, 0 on failure.
403 depot_stack_handle_t __stack_depot_save(unsigned long *entries,
404 unsigned int nr_entries,
405 gfp_t alloc_flags, bool can_alloc)
407 struct stack_record *found = NULL, **bucket;
408 depot_stack_handle_t retval = 0;
409 struct page *page = NULL;
410 void *prealloc = NULL;
415 * If this stack trace is from an interrupt, including anything before
416 * interrupt entry usually leads to unbounded stackdepot growth.
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.
422 nr_entries = filter_irq_stacks(entries, nr_entries);
424 if (unlikely(nr_entries == 0) || stack_depot_disable)
427 hash = hash_stack(entries, nr_entries);
428 bucket = &stack_table[hash & stack_hash_mask];
431 * Fast path: look the stack trace up without locking.
432 * The smp_load_acquire() here pairs with smp_store_release() to
435 found = find_stack(smp_load_acquire(bucket), entries,
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
445 * The smp_load_acquire() here pairs with smp_store_release() to
446 * |next_slab_inited| in depot_alloc_stack() and init_stack_slab().
448 if (unlikely(can_alloc && !smp_load_acquire(&next_slab_inited))) {
450 * Zero out zone modifiers, as we don't have specific zone
451 * requirements. Keep the flags related to allocation in atomic
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);
459 prealloc = page_address(page);
462 raw_spin_lock_irqsave(&depot_lock, flags);
464 found = find_stack(*bucket, entries, nr_entries, hash);
466 struct stack_record *new = depot_alloc_stack(entries, nr_entries, hash, &prealloc);
471 * This smp_store_release() pairs with
472 * smp_load_acquire() from |bucket| above.
474 smp_store_release(bucket, new);
477 } else if (prealloc) {
479 * We didn't need to store this stack trace, but let's keep
480 * the preallocated memory for the future.
482 WARN_ON(!init_stack_slab(&prealloc));
485 raw_spin_unlock_irqrestore(&depot_lock, flags);
488 /* Nobody used this memory, ok to free it. */
489 free_pages((unsigned long)prealloc, STACK_ALLOC_ORDER);
492 retval = found->handle.handle;
496 EXPORT_SYMBOL_GPL(__stack_depot_save);
499 * stack_depot_save - Save a stack trace from an array
501 * @entries: Pointer to storage array
502 * @nr_entries: Size of the storage array
503 * @alloc_flags: Allocation gfp flags
505 * Context: Contexts where allocations via alloc_pages() are allowed.
506 * See __stack_depot_save() for more details.
508 * Return: The handle of the stack struct stored in depot, 0 on failure.
510 depot_stack_handle_t stack_depot_save(unsigned long *entries,
511 unsigned int nr_entries,
514 return __stack_depot_save(entries, nr_entries, alloc_flags, true);
516 EXPORT_SYMBOL_GPL(stack_depot_save);