Commit | Line | Data |
---|---|---|
e5a81b62 SR |
1 | /* |
2 | * Copyright (C) 2008 Steven Rostedt <srostedt@redhat.com> | |
3 | * | |
4 | */ | |
5 | #include <linux/stacktrace.h> | |
6 | #include <linux/kallsyms.h> | |
7 | #include <linux/seq_file.h> | |
8 | #include <linux/spinlock.h> | |
9 | #include <linux/uaccess.h> | |
10 | #include <linux/debugfs.h> | |
11 | #include <linux/ftrace.h> | |
12 | #include <linux/module.h> | |
f38f1d2a | 13 | #include <linux/sysctl.h> |
e5a81b62 SR |
14 | #include <linux/init.h> |
15 | #include <linux/fs.h> | |
16 | #include "trace.h" | |
17 | ||
18 | #define STACK_TRACE_ENTRIES 500 | |
19 | ||
1b6cced6 SR |
20 | static unsigned long stack_dump_trace[STACK_TRACE_ENTRIES+1] = |
21 | { [0 ... (STACK_TRACE_ENTRIES)] = ULONG_MAX }; | |
22 | static unsigned stack_dump_index[STACK_TRACE_ENTRIES]; | |
23 | ||
e5a81b62 SR |
24 | static struct stack_trace max_stack_trace = { |
25 | .max_entries = STACK_TRACE_ENTRIES, | |
26 | .entries = stack_dump_trace, | |
27 | }; | |
28 | ||
29 | static unsigned long max_stack_size; | |
445c8951 | 30 | static arch_spinlock_t max_stack_lock = |
edc35bd7 | 31 | (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED; |
e5a81b62 SR |
32 | |
33 | static int stack_trace_disabled __read_mostly; | |
34 | static DEFINE_PER_CPU(int, trace_active); | |
f38f1d2a SR |
35 | static DEFINE_MUTEX(stack_sysctl_mutex); |
36 | ||
37 | int stack_tracer_enabled; | |
38 | static int last_stack_tracer_enabled; | |
e5a81b62 SR |
39 | |
40 | static inline void check_stack(void) | |
41 | { | |
1b6cced6 SR |
42 | unsigned long this_size, flags; |
43 | unsigned long *p, *top, *start; | |
44 | int i; | |
e5a81b62 SR |
45 | |
46 | this_size = ((unsigned long)&this_size) & (THREAD_SIZE-1); | |
47 | this_size = THREAD_SIZE - this_size; | |
48 | ||
49 | if (this_size <= max_stack_size) | |
50 | return; | |
51 | ||
81520a1b SR |
52 | /* we do not handle interrupt stacks yet */ |
53 | if (!object_is_on_stack(&this_size)) | |
54 | return; | |
55 | ||
a5e25883 | 56 | local_irq_save(flags); |
0199c4e6 | 57 | arch_spin_lock(&max_stack_lock); |
e5a81b62 SR |
58 | |
59 | /* a race could have already updated it */ | |
60 | if (this_size <= max_stack_size) | |
61 | goto out; | |
62 | ||
63 | max_stack_size = this_size; | |
64 | ||
65 | max_stack_trace.nr_entries = 0; | |
1b6cced6 | 66 | max_stack_trace.skip = 3; |
e5a81b62 SR |
67 | |
68 | save_stack_trace(&max_stack_trace); | |
69 | ||
1b6cced6 SR |
70 | /* |
71 | * Now find where in the stack these are. | |
72 | */ | |
73 | i = 0; | |
74 | start = &this_size; | |
75 | top = (unsigned long *) | |
76 | (((unsigned long)start & ~(THREAD_SIZE-1)) + THREAD_SIZE); | |
77 | ||
78 | /* | |
79 | * Loop through all the entries. One of the entries may | |
80 | * for some reason be missed on the stack, so we may | |
81 | * have to account for them. If they are all there, this | |
82 | * loop will only happen once. This code only takes place | |
83 | * on a new max, so it is far from a fast path. | |
84 | */ | |
85 | while (i < max_stack_trace.nr_entries) { | |
0a37119d | 86 | int found = 0; |
1b6cced6 SR |
87 | |
88 | stack_dump_index[i] = this_size; | |
89 | p = start; | |
90 | ||
91 | for (; p < top && i < max_stack_trace.nr_entries; p++) { | |
92 | if (*p == stack_dump_trace[i]) { | |
93 | this_size = stack_dump_index[i++] = | |
94 | (top - p) * sizeof(unsigned long); | |
0a37119d | 95 | found = 1; |
1b6cced6 SR |
96 | /* Start the search from here */ |
97 | start = p + 1; | |
98 | } | |
99 | } | |
100 | ||
0a37119d SR |
101 | if (!found) |
102 | i++; | |
1b6cced6 SR |
103 | } |
104 | ||
e5a81b62 | 105 | out: |
0199c4e6 | 106 | arch_spin_unlock(&max_stack_lock); |
a5e25883 | 107 | local_irq_restore(flags); |
e5a81b62 SR |
108 | } |
109 | ||
110 | static void | |
111 | stack_trace_call(unsigned long ip, unsigned long parent_ip) | |
112 | { | |
5168ae50 | 113 | int cpu; |
e5a81b62 SR |
114 | |
115 | if (unlikely(!ftrace_enabled || stack_trace_disabled)) | |
116 | return; | |
117 | ||
5168ae50 | 118 | preempt_disable_notrace(); |
e5a81b62 SR |
119 | |
120 | cpu = raw_smp_processor_id(); | |
121 | /* no atomic needed, we only modify this variable by this cpu */ | |
122 | if (per_cpu(trace_active, cpu)++ != 0) | |
123 | goto out; | |
124 | ||
125 | check_stack(); | |
126 | ||
127 | out: | |
128 | per_cpu(trace_active, cpu)--; | |
129 | /* prevent recursion in schedule */ | |
5168ae50 | 130 | preempt_enable_notrace(); |
e5a81b62 SR |
131 | } |
132 | ||
133 | static struct ftrace_ops trace_ops __read_mostly = | |
134 | { | |
135 | .func = stack_trace_call, | |
136 | }; | |
137 | ||
138 | static ssize_t | |
139 | stack_max_size_read(struct file *filp, char __user *ubuf, | |
140 | size_t count, loff_t *ppos) | |
141 | { | |
142 | unsigned long *ptr = filp->private_data; | |
143 | char buf[64]; | |
144 | int r; | |
145 | ||
146 | r = snprintf(buf, sizeof(buf), "%ld\n", *ptr); | |
147 | if (r > sizeof(buf)) | |
148 | r = sizeof(buf); | |
149 | return simple_read_from_buffer(ubuf, count, ppos, buf, r); | |
150 | } | |
151 | ||
152 | static ssize_t | |
153 | stack_max_size_write(struct file *filp, const char __user *ubuf, | |
154 | size_t count, loff_t *ppos) | |
155 | { | |
156 | long *ptr = filp->private_data; | |
157 | unsigned long val, flags; | |
158 | char buf[64]; | |
159 | int ret; | |
4f48f8b7 | 160 | int cpu; |
e5a81b62 SR |
161 | |
162 | if (count >= sizeof(buf)) | |
163 | return -EINVAL; | |
164 | ||
165 | if (copy_from_user(&buf, ubuf, count)) | |
166 | return -EFAULT; | |
167 | ||
168 | buf[count] = 0; | |
169 | ||
170 | ret = strict_strtoul(buf, 10, &val); | |
171 | if (ret < 0) | |
172 | return ret; | |
173 | ||
a5e25883 | 174 | local_irq_save(flags); |
4f48f8b7 LJ |
175 | |
176 | /* | |
177 | * In case we trace inside arch_spin_lock() or after (NMI), | |
178 | * we will cause circular lock, so we also need to increase | |
179 | * the percpu trace_active here. | |
180 | */ | |
181 | cpu = smp_processor_id(); | |
182 | per_cpu(trace_active, cpu)++; | |
183 | ||
0199c4e6 | 184 | arch_spin_lock(&max_stack_lock); |
e5a81b62 | 185 | *ptr = val; |
0199c4e6 | 186 | arch_spin_unlock(&max_stack_lock); |
4f48f8b7 LJ |
187 | |
188 | per_cpu(trace_active, cpu)--; | |
a5e25883 | 189 | local_irq_restore(flags); |
e5a81b62 SR |
190 | |
191 | return count; | |
192 | } | |
193 | ||
f38f1d2a | 194 | static const struct file_operations stack_max_size_fops = { |
e5a81b62 SR |
195 | .open = tracing_open_generic, |
196 | .read = stack_max_size_read, | |
197 | .write = stack_max_size_write, | |
6038f373 | 198 | .llseek = default_llseek, |
e5a81b62 SR |
199 | }; |
200 | ||
201 | static void * | |
2fc5f0cf | 202 | __next(struct seq_file *m, loff_t *pos) |
e5a81b62 | 203 | { |
2fc5f0cf | 204 | long n = *pos - 1; |
e5a81b62 | 205 | |
2fc5f0cf | 206 | if (n >= max_stack_trace.nr_entries || stack_dump_trace[n] == ULONG_MAX) |
e5a81b62 SR |
207 | return NULL; |
208 | ||
2fc5f0cf | 209 | m->private = (void *)n; |
1b6cced6 | 210 | return &m->private; |
e5a81b62 SR |
211 | } |
212 | ||
2fc5f0cf LZ |
213 | static void * |
214 | t_next(struct seq_file *m, void *v, loff_t *pos) | |
e5a81b62 | 215 | { |
2fc5f0cf LZ |
216 | (*pos)++; |
217 | return __next(m, pos); | |
218 | } | |
e5a81b62 | 219 | |
2fc5f0cf LZ |
220 | static void *t_start(struct seq_file *m, loff_t *pos) |
221 | { | |
4f48f8b7 LJ |
222 | int cpu; |
223 | ||
e5a81b62 | 224 | local_irq_disable(); |
4f48f8b7 LJ |
225 | |
226 | cpu = smp_processor_id(); | |
227 | per_cpu(trace_active, cpu)++; | |
228 | ||
0199c4e6 | 229 | arch_spin_lock(&max_stack_lock); |
e5a81b62 | 230 | |
522a110b LW |
231 | if (*pos == 0) |
232 | return SEQ_START_TOKEN; | |
233 | ||
2fc5f0cf | 234 | return __next(m, pos); |
e5a81b62 SR |
235 | } |
236 | ||
237 | static void t_stop(struct seq_file *m, void *p) | |
238 | { | |
4f48f8b7 LJ |
239 | int cpu; |
240 | ||
0199c4e6 | 241 | arch_spin_unlock(&max_stack_lock); |
4f48f8b7 LJ |
242 | |
243 | cpu = smp_processor_id(); | |
244 | per_cpu(trace_active, cpu)--; | |
245 | ||
e5a81b62 SR |
246 | local_irq_enable(); |
247 | } | |
248 | ||
1b6cced6 | 249 | static int trace_lookup_stack(struct seq_file *m, long i) |
e5a81b62 | 250 | { |
1b6cced6 | 251 | unsigned long addr = stack_dump_trace[i]; |
e5a81b62 | 252 | |
151772db | 253 | return seq_printf(m, "%pS\n", (void *)addr); |
e5a81b62 SR |
254 | } |
255 | ||
e447e1df SR |
256 | static void print_disabled(struct seq_file *m) |
257 | { | |
258 | seq_puts(m, "#\n" | |
259 | "# Stack tracer disabled\n" | |
260 | "#\n" | |
261 | "# To enable the stack tracer, either add 'stacktrace' to the\n" | |
262 | "# kernel command line\n" | |
263 | "# or 'echo 1 > /proc/sys/kernel/stack_tracer_enabled'\n" | |
264 | "#\n"); | |
265 | } | |
266 | ||
e5a81b62 SR |
267 | static int t_show(struct seq_file *m, void *v) |
268 | { | |
522a110b | 269 | long i; |
1b6cced6 SR |
270 | int size; |
271 | ||
522a110b | 272 | if (v == SEQ_START_TOKEN) { |
eb1871f3 | 273 | seq_printf(m, " Depth Size Location" |
1b6cced6 | 274 | " (%d entries)\n" |
eb1871f3 | 275 | " ----- ---- --------\n", |
083a63b4 | 276 | max_stack_trace.nr_entries - 1); |
e447e1df SR |
277 | |
278 | if (!stack_tracer_enabled && !max_stack_size) | |
279 | print_disabled(m); | |
280 | ||
1b6cced6 SR |
281 | return 0; |
282 | } | |
e5a81b62 | 283 | |
522a110b LW |
284 | i = *(long *)v; |
285 | ||
1b6cced6 SR |
286 | if (i >= max_stack_trace.nr_entries || |
287 | stack_dump_trace[i] == ULONG_MAX) | |
e5a81b62 SR |
288 | return 0; |
289 | ||
1b6cced6 SR |
290 | if (i+1 == max_stack_trace.nr_entries || |
291 | stack_dump_trace[i+1] == ULONG_MAX) | |
292 | size = stack_dump_index[i]; | |
293 | else | |
294 | size = stack_dump_index[i] - stack_dump_index[i+1]; | |
295 | ||
296 | seq_printf(m, "%3ld) %8d %5d ", i, stack_dump_index[i], size); | |
297 | ||
298 | trace_lookup_stack(m, i); | |
e5a81b62 SR |
299 | |
300 | return 0; | |
301 | } | |
302 | ||
f38f1d2a | 303 | static const struct seq_operations stack_trace_seq_ops = { |
e5a81b62 SR |
304 | .start = t_start, |
305 | .next = t_next, | |
306 | .stop = t_stop, | |
307 | .show = t_show, | |
308 | }; | |
309 | ||
310 | static int stack_trace_open(struct inode *inode, struct file *file) | |
311 | { | |
d8cc1ab7 | 312 | return seq_open(file, &stack_trace_seq_ops); |
e5a81b62 SR |
313 | } |
314 | ||
f38f1d2a | 315 | static const struct file_operations stack_trace_fops = { |
e5a81b62 SR |
316 | .open = stack_trace_open, |
317 | .read = seq_read, | |
318 | .llseek = seq_lseek, | |
d8cc1ab7 | 319 | .release = seq_release, |
e5a81b62 SR |
320 | }; |
321 | ||
f38f1d2a SR |
322 | int |
323 | stack_trace_sysctl(struct ctl_table *table, int write, | |
8d65af78 | 324 | void __user *buffer, size_t *lenp, |
f38f1d2a SR |
325 | loff_t *ppos) |
326 | { | |
327 | int ret; | |
328 | ||
329 | mutex_lock(&stack_sysctl_mutex); | |
330 | ||
8d65af78 | 331 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
f38f1d2a SR |
332 | |
333 | if (ret || !write || | |
a32c7765 | 334 | (last_stack_tracer_enabled == !!stack_tracer_enabled)) |
f38f1d2a SR |
335 | goto out; |
336 | ||
a32c7765 | 337 | last_stack_tracer_enabled = !!stack_tracer_enabled; |
f38f1d2a SR |
338 | |
339 | if (stack_tracer_enabled) | |
340 | register_ftrace_function(&trace_ops); | |
341 | else | |
342 | unregister_ftrace_function(&trace_ops); | |
343 | ||
344 | out: | |
345 | mutex_unlock(&stack_sysctl_mutex); | |
346 | return ret; | |
347 | } | |
348 | ||
f38f1d2a SR |
349 | static __init int enable_stacktrace(char *str) |
350 | { | |
e05a43b7 SR |
351 | stack_tracer_enabled = 1; |
352 | last_stack_tracer_enabled = 1; | |
f38f1d2a SR |
353 | return 1; |
354 | } | |
355 | __setup("stacktrace", enable_stacktrace); | |
356 | ||
e5a81b62 SR |
357 | static __init int stack_trace_init(void) |
358 | { | |
359 | struct dentry *d_tracer; | |
e5a81b62 SR |
360 | |
361 | d_tracer = tracing_init_dentry(); | |
362 | ||
5452af66 FW |
363 | trace_create_file("stack_max_size", 0644, d_tracer, |
364 | &max_stack_size, &stack_max_size_fops); | |
e5a81b62 | 365 | |
5452af66 FW |
366 | trace_create_file("stack_trace", 0444, d_tracer, |
367 | NULL, &stack_trace_fops); | |
e5a81b62 | 368 | |
e05a43b7 | 369 | if (stack_tracer_enabled) |
f38f1d2a | 370 | register_ftrace_function(&trace_ops); |
e5a81b62 SR |
371 | |
372 | return 0; | |
373 | } | |
374 | ||
375 | device_initcall(stack_trace_init); |