xref: /openbmc/linux/lib/stackdepot.c (revision 160b8e75)
1 /*
2  * Generic stack depot for storing stack traces.
3  *
4  * Some debugging tools need to save stack traces of certain events which can
5  * be later presented to the user. For example, KASAN needs to safe alloc and
6  * free stacks for each object, but storing two stack traces per object
7  * requires too much memory (e.g. SLUB_DEBUG needs 256 bytes per object for
8  * that).
9  *
10  * Instead, stack depot maintains a hashtable of unique stacktraces. Since alloc
11  * and free stacks repeat a lot, we save about 100x space.
12  * Stacks are never removed from depot, so we store them contiguously one after
13  * another in a contiguos memory allocation.
14  *
15  * Author: Alexander Potapenko <glider@google.com>
16  * Copyright (C) 2016 Google, Inc.
17  *
18  * Based on code by Dmitry Chernenkov.
19  *
20  * This program is free software; you can redistribute it and/or
21  * modify it under the terms of the GNU General Public License
22  * version 2 as published by the Free Software Foundation.
23  *
24  * This program is distributed in the hope that it will be useful, but
25  * WITHOUT ANY WARRANTY; without even the implied warranty of
26  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
27  * General Public License for more details.
28  *
29  */
30 
31 #include <linux/gfp.h>
32 #include <linux/jhash.h>
33 #include <linux/kernel.h>
34 #include <linux/mm.h>
35 #include <linux/percpu.h>
36 #include <linux/printk.h>
37 #include <linux/slab.h>
38 #include <linux/stacktrace.h>
39 #include <linux/stackdepot.h>
40 #include <linux/string.h>
41 #include <linux/types.h>
42 
43 #define DEPOT_STACK_BITS (sizeof(depot_stack_handle_t) * 8)
44 
45 #define STACK_ALLOC_NULL_PROTECTION_BITS 1
46 #define STACK_ALLOC_ORDER 2 /* 'Slab' size order for stack depot, 4 pages */
47 #define STACK_ALLOC_SIZE (1LL << (PAGE_SHIFT + STACK_ALLOC_ORDER))
48 #define STACK_ALLOC_ALIGN 4
49 #define STACK_ALLOC_OFFSET_BITS (STACK_ALLOC_ORDER + PAGE_SHIFT - \
50 					STACK_ALLOC_ALIGN)
51 #define STACK_ALLOC_INDEX_BITS (DEPOT_STACK_BITS - \
52 		STACK_ALLOC_NULL_PROTECTION_BITS - STACK_ALLOC_OFFSET_BITS)
53 #define STACK_ALLOC_SLABS_CAP 8192
54 #define STACK_ALLOC_MAX_SLABS \
55 	(((1LL << (STACK_ALLOC_INDEX_BITS)) < STACK_ALLOC_SLABS_CAP) ? \
56 	 (1LL << (STACK_ALLOC_INDEX_BITS)) : STACK_ALLOC_SLABS_CAP)
57 
58 /* The compact structure to store the reference to stacks. */
59 union handle_parts {
60 	depot_stack_handle_t handle;
61 	struct {
62 		u32 slabindex : STACK_ALLOC_INDEX_BITS;
63 		u32 offset : STACK_ALLOC_OFFSET_BITS;
64 		u32 valid : STACK_ALLOC_NULL_PROTECTION_BITS;
65 	};
66 };
67 
68 struct stack_record {
69 	struct stack_record *next;	/* Link in the hashtable */
70 	u32 hash;			/* Hash in the hastable */
71 	u32 size;			/* Number of frames in the stack */
72 	union handle_parts handle;
73 	unsigned long entries[1];	/* Variable-sized array of entries. */
74 };
75 
76 static void *stack_slabs[STACK_ALLOC_MAX_SLABS];
77 
78 static int depot_index;
79 static int next_slab_inited;
80 static size_t depot_offset;
81 static DEFINE_SPINLOCK(depot_lock);
82 
83 static bool init_stack_slab(void **prealloc)
84 {
85 	if (!*prealloc)
86 		return false;
87 	/*
88 	 * This smp_load_acquire() pairs with smp_store_release() to
89 	 * |next_slab_inited| below and in depot_alloc_stack().
90 	 */
91 	if (smp_load_acquire(&next_slab_inited))
92 		return true;
93 	if (stack_slabs[depot_index] == NULL) {
94 		stack_slabs[depot_index] = *prealloc;
95 	} else {
96 		stack_slabs[depot_index + 1] = *prealloc;
97 		/*
98 		 * This smp_store_release pairs with smp_load_acquire() from
99 		 * |next_slab_inited| above and in depot_save_stack().
100 		 */
101 		smp_store_release(&next_slab_inited, 1);
102 	}
103 	*prealloc = NULL;
104 	return true;
105 }
106 
107 /* Allocation of a new stack in raw storage */
108 static struct stack_record *depot_alloc_stack(unsigned long *entries, int size,
109 		u32 hash, void **prealloc, gfp_t alloc_flags)
110 {
111 	int required_size = offsetof(struct stack_record, entries) +
112 		sizeof(unsigned long) * size;
113 	struct stack_record *stack;
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 depot_save_stack() 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, size * sizeof(unsigned long));
144 	depot_offset += required_size;
145 
146 	return stack;
147 }
148 
149 #define STACK_HASH_ORDER 20
150 #define STACK_HASH_SIZE (1L << STACK_HASH_ORDER)
151 #define STACK_HASH_MASK (STACK_HASH_SIZE - 1)
152 #define STACK_HASH_SEED 0x9747b28c
153 
154 static struct stack_record *stack_table[STACK_HASH_SIZE] = {
155 	[0 ...	STACK_HASH_SIZE - 1] = NULL
156 };
157 
158 /* Calculate hash for a stack */
159 static inline u32 hash_stack(unsigned long *entries, unsigned int size)
160 {
161 	return jhash2((u32 *)entries,
162 			       size * sizeof(unsigned long) / sizeof(u32),
163 			       STACK_HASH_SEED);
164 }
165 
166 /* Use our own, non-instrumented version of memcmp().
167  *
168  * We actually don't care about the order, just the equality.
169  */
170 static inline
171 int stackdepot_memcmp(const unsigned long *u1, const unsigned long *u2,
172 			unsigned int n)
173 {
174 	for ( ; n-- ; u1++, u2++) {
175 		if (*u1 != *u2)
176 			return 1;
177 	}
178 	return 0;
179 }
180 
181 /* Find a stack that is equal to the one stored in entries in the hash */
182 static inline struct stack_record *find_stack(struct stack_record *bucket,
183 					     unsigned long *entries, int size,
184 					     u32 hash)
185 {
186 	struct stack_record *found;
187 
188 	for (found = bucket; found; found = found->next) {
189 		if (found->hash == hash &&
190 		    found->size == size &&
191 		    !stackdepot_memcmp(entries, found->entries, size))
192 			return found;
193 	}
194 	return NULL;
195 }
196 
197 void depot_fetch_stack(depot_stack_handle_t handle, struct stack_trace *trace)
198 {
199 	union handle_parts parts = { .handle = handle };
200 	void *slab = stack_slabs[parts.slabindex];
201 	size_t offset = parts.offset << STACK_ALLOC_ALIGN;
202 	struct stack_record *stack = slab + offset;
203 
204 	trace->nr_entries = trace->max_entries = stack->size;
205 	trace->entries = stack->entries;
206 	trace->skip = 0;
207 }
208 EXPORT_SYMBOL_GPL(depot_fetch_stack);
209 
210 /**
211  * depot_save_stack - save stack in a stack depot.
212  * @trace - the stacktrace to save.
213  * @alloc_flags - flags for allocating additional memory if required.
214  *
215  * Returns the handle of the stack struct stored in depot.
216  */
217 depot_stack_handle_t depot_save_stack(struct stack_trace *trace,
218 				    gfp_t alloc_flags)
219 {
220 	u32 hash;
221 	depot_stack_handle_t retval = 0;
222 	struct stack_record *found = NULL, **bucket;
223 	unsigned long flags;
224 	struct page *page = NULL;
225 	void *prealloc = NULL;
226 
227 	if (unlikely(trace->nr_entries == 0))
228 		goto fast_exit;
229 
230 	hash = hash_stack(trace->entries, trace->nr_entries);
231 	bucket = &stack_table[hash & STACK_HASH_MASK];
232 
233 	/*
234 	 * Fast path: look the stack trace up without locking.
235 	 * The smp_load_acquire() here pairs with smp_store_release() to
236 	 * |bucket| below.
237 	 */
238 	found = find_stack(smp_load_acquire(bucket), trace->entries,
239 			   trace->nr_entries, hash);
240 	if (found)
241 		goto exit;
242 
243 	/*
244 	 * Check if the current or the next stack slab need to be initialized.
245 	 * If so, allocate the memory - we won't be able to do that under the
246 	 * lock.
247 	 *
248 	 * The smp_load_acquire() here pairs with smp_store_release() to
249 	 * |next_slab_inited| in depot_alloc_stack() and init_stack_slab().
250 	 */
251 	if (unlikely(!smp_load_acquire(&next_slab_inited))) {
252 		/*
253 		 * Zero out zone modifiers, as we don't have specific zone
254 		 * requirements. Keep the flags related to allocation in atomic
255 		 * contexts and I/O.
256 		 */
257 		alloc_flags &= ~GFP_ZONEMASK;
258 		alloc_flags &= (GFP_ATOMIC | GFP_KERNEL);
259 		alloc_flags |= __GFP_NOWARN;
260 		page = alloc_pages(alloc_flags, STACK_ALLOC_ORDER);
261 		if (page)
262 			prealloc = page_address(page);
263 	}
264 
265 	spin_lock_irqsave(&depot_lock, flags);
266 
267 	found = find_stack(*bucket, trace->entries, trace->nr_entries, hash);
268 	if (!found) {
269 		struct stack_record *new =
270 			depot_alloc_stack(trace->entries, trace->nr_entries,
271 					  hash, &prealloc, alloc_flags);
272 		if (new) {
273 			new->next = *bucket;
274 			/*
275 			 * This smp_store_release() pairs with
276 			 * smp_load_acquire() from |bucket| above.
277 			 */
278 			smp_store_release(bucket, new);
279 			found = new;
280 		}
281 	} else if (prealloc) {
282 		/*
283 		 * We didn't need to store this stack trace, but let's keep
284 		 * the preallocated memory for the future.
285 		 */
286 		WARN_ON(!init_stack_slab(&prealloc));
287 	}
288 
289 	spin_unlock_irqrestore(&depot_lock, flags);
290 exit:
291 	if (prealloc) {
292 		/* Nobody used this memory, ok to free it. */
293 		free_pages((unsigned long)prealloc, STACK_ALLOC_ORDER);
294 	}
295 	if (found)
296 		retval = found->handle.handle;
297 fast_exit:
298 	return retval;
299 }
300 EXPORT_SYMBOL_GPL(depot_save_stack);
301