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> | |
38628078 | 16 | #include <linux/magic.h> |
762e1207 SR |
17 | |
18 | #include <asm/setup.h> | |
19 | ||
e5a81b62 SR |
20 | #include "trace.h" |
21 | ||
22 | #define STACK_TRACE_ENTRIES 500 | |
23 | ||
d4ecbfc4 | 24 | #ifdef CC_USING_FENTRY |
4df29712 | 25 | # define fentry 1 |
d4ecbfc4 | 26 | #else |
4df29712 | 27 | # define fentry 0 |
d4ecbfc4 SRRH |
28 | #endif |
29 | ||
1b6cced6 SR |
30 | static unsigned long stack_dump_trace[STACK_TRACE_ENTRIES+1] = |
31 | { [0 ... (STACK_TRACE_ENTRIES)] = ULONG_MAX }; | |
32 | static unsigned stack_dump_index[STACK_TRACE_ENTRIES]; | |
33 | ||
4df29712 SRRH |
34 | /* |
35 | * Reserve one entry for the passed in ip. This will allow | |
36 | * us to remove most or all of the stack size overhead | |
37 | * added by the stack tracer itself. | |
38 | */ | |
e5a81b62 | 39 | static struct stack_trace max_stack_trace = { |
4df29712 SRRH |
40 | .max_entries = STACK_TRACE_ENTRIES - 1, |
41 | .entries = &stack_dump_trace[1], | |
e5a81b62 SR |
42 | }; |
43 | ||
44 | static unsigned long max_stack_size; | |
445c8951 | 45 | static arch_spinlock_t max_stack_lock = |
edc35bd7 | 46 | (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED; |
e5a81b62 | 47 | |
e5a81b62 | 48 | static DEFINE_PER_CPU(int, trace_active); |
f38f1d2a SR |
49 | static DEFINE_MUTEX(stack_sysctl_mutex); |
50 | ||
51 | int stack_tracer_enabled; | |
52 | static int last_stack_tracer_enabled; | |
e5a81b62 | 53 | |
87889501 | 54 | static inline void |
d4ecbfc4 | 55 | check_stack(unsigned long ip, unsigned long *stack) |
e5a81b62 | 56 | { |
1b6cced6 SR |
57 | unsigned long this_size, flags; |
58 | unsigned long *p, *top, *start; | |
4df29712 SRRH |
59 | static int tracer_frame; |
60 | int frame_size = ACCESS_ONCE(tracer_frame); | |
1b6cced6 | 61 | int i; |
e5a81b62 | 62 | |
87889501 | 63 | this_size = ((unsigned long)stack) & (THREAD_SIZE-1); |
e5a81b62 | 64 | this_size = THREAD_SIZE - this_size; |
4df29712 SRRH |
65 | /* Remove the frame of the tracer */ |
66 | this_size -= frame_size; | |
e5a81b62 SR |
67 | |
68 | if (this_size <= max_stack_size) | |
69 | return; | |
70 | ||
81520a1b | 71 | /* we do not handle interrupt stacks yet */ |
87889501 | 72 | if (!object_is_on_stack(stack)) |
81520a1b SR |
73 | return; |
74 | ||
a5e25883 | 75 | local_irq_save(flags); |
0199c4e6 | 76 | arch_spin_lock(&max_stack_lock); |
e5a81b62 | 77 | |
4df29712 SRRH |
78 | /* In case another CPU set the tracer_frame on us */ |
79 | if (unlikely(!frame_size)) | |
80 | this_size -= tracer_frame; | |
81 | ||
e5a81b62 SR |
82 | /* a race could have already updated it */ |
83 | if (this_size <= max_stack_size) | |
84 | goto out; | |
85 | ||
86 | max_stack_size = this_size; | |
87 | ||
7eea4fce JW |
88 | max_stack_trace.nr_entries = 0; |
89 | ||
90 | if (using_ftrace_ops_list_func()) | |
91 | max_stack_trace.skip = 4; | |
92 | else | |
93 | max_stack_trace.skip = 3; | |
e5a81b62 SR |
94 | |
95 | save_stack_trace(&max_stack_trace); | |
96 | ||
d4ecbfc4 | 97 | /* |
4df29712 SRRH |
98 | * Add the passed in ip from the function tracer. |
99 | * Searching for this on the stack will skip over | |
100 | * most of the overhead from the stack tracer itself. | |
d4ecbfc4 | 101 | */ |
4df29712 SRRH |
102 | stack_dump_trace[0] = ip; |
103 | max_stack_trace.nr_entries++; | |
d4ecbfc4 | 104 | |
1b6cced6 SR |
105 | /* |
106 | * Now find where in the stack these are. | |
107 | */ | |
108 | i = 0; | |
87889501 | 109 | start = stack; |
1b6cced6 SR |
110 | top = (unsigned long *) |
111 | (((unsigned long)start & ~(THREAD_SIZE-1)) + THREAD_SIZE); | |
112 | ||
113 | /* | |
114 | * Loop through all the entries. One of the entries may | |
115 | * for some reason be missed on the stack, so we may | |
116 | * have to account for them. If they are all there, this | |
117 | * loop will only happen once. This code only takes place | |
118 | * on a new max, so it is far from a fast path. | |
119 | */ | |
120 | while (i < max_stack_trace.nr_entries) { | |
0a37119d | 121 | int found = 0; |
1b6cced6 SR |
122 | |
123 | stack_dump_index[i] = this_size; | |
124 | p = start; | |
125 | ||
126 | for (; p < top && i < max_stack_trace.nr_entries; p++) { | |
127 | if (*p == stack_dump_trace[i]) { | |
128 | this_size = stack_dump_index[i++] = | |
129 | (top - p) * sizeof(unsigned long); | |
0a37119d | 130 | found = 1; |
1b6cced6 SR |
131 | /* Start the search from here */ |
132 | start = p + 1; | |
4df29712 SRRH |
133 | /* |
134 | * We do not want to show the overhead | |
135 | * of the stack tracer stack in the | |
136 | * max stack. If we haven't figured | |
137 | * out what that is, then figure it out | |
138 | * now. | |
139 | */ | |
140 | if (unlikely(!tracer_frame) && i == 1) { | |
141 | tracer_frame = (p - stack) * | |
142 | sizeof(unsigned long); | |
143 | max_stack_size -= tracer_frame; | |
144 | } | |
1b6cced6 SR |
145 | } |
146 | } | |
147 | ||
0a37119d SR |
148 | if (!found) |
149 | i++; | |
1b6cced6 SR |
150 | } |
151 | ||
38628078 AT |
152 | BUG_ON(current != &init_task && |
153 | *(end_of_stack(current)) != STACK_END_MAGIC); | |
e5a81b62 | 154 | out: |
0199c4e6 | 155 | arch_spin_unlock(&max_stack_lock); |
a5e25883 | 156 | local_irq_restore(flags); |
e5a81b62 SR |
157 | } |
158 | ||
159 | static void | |
a1e2e31d SR |
160 | stack_trace_call(unsigned long ip, unsigned long parent_ip, |
161 | struct ftrace_ops *op, struct pt_regs *pt_regs) | |
e5a81b62 | 162 | { |
87889501 | 163 | unsigned long stack; |
5168ae50 | 164 | int cpu; |
e5a81b62 | 165 | |
5168ae50 | 166 | preempt_disable_notrace(); |
e5a81b62 SR |
167 | |
168 | cpu = raw_smp_processor_id(); | |
169 | /* no atomic needed, we only modify this variable by this cpu */ | |
170 | if (per_cpu(trace_active, cpu)++ != 0) | |
171 | goto out; | |
172 | ||
4df29712 SRRH |
173 | /* |
174 | * When fentry is used, the traced function does not get | |
175 | * its stack frame set up, and we lose the parent. | |
176 | * The ip is pretty useless because the function tracer | |
177 | * was called before that function set up its stack frame. | |
178 | * In this case, we use the parent ip. | |
179 | * | |
180 | * By adding the return address of either the parent ip | |
181 | * or the current ip we can disregard most of the stack usage | |
182 | * caused by the stack tracer itself. | |
183 | * | |
184 | * The function tracer always reports the address of where the | |
185 | * mcount call was, but the stack will hold the return address. | |
186 | */ | |
187 | if (fentry) | |
188 | ip = parent_ip; | |
189 | else | |
190 | ip += MCOUNT_INSN_SIZE; | |
191 | ||
192 | check_stack(ip, &stack); | |
e5a81b62 SR |
193 | |
194 | out: | |
195 | per_cpu(trace_active, cpu)--; | |
196 | /* prevent recursion in schedule */ | |
5168ae50 | 197 | preempt_enable_notrace(); |
e5a81b62 SR |
198 | } |
199 | ||
200 | static struct ftrace_ops trace_ops __read_mostly = | |
201 | { | |
202 | .func = stack_trace_call, | |
4740974a | 203 | .flags = FTRACE_OPS_FL_RECURSION_SAFE, |
e5a81b62 SR |
204 | }; |
205 | ||
206 | static ssize_t | |
207 | stack_max_size_read(struct file *filp, char __user *ubuf, | |
208 | size_t count, loff_t *ppos) | |
209 | { | |
210 | unsigned long *ptr = filp->private_data; | |
211 | char buf[64]; | |
212 | int r; | |
213 | ||
214 | r = snprintf(buf, sizeof(buf), "%ld\n", *ptr); | |
215 | if (r > sizeof(buf)) | |
216 | r = sizeof(buf); | |
217 | return simple_read_from_buffer(ubuf, count, ppos, buf, r); | |
218 | } | |
219 | ||
220 | static ssize_t | |
221 | stack_max_size_write(struct file *filp, const char __user *ubuf, | |
222 | size_t count, loff_t *ppos) | |
223 | { | |
224 | long *ptr = filp->private_data; | |
225 | unsigned long val, flags; | |
e5a81b62 | 226 | int ret; |
4f48f8b7 | 227 | int cpu; |
e5a81b62 | 228 | |
22fe9b54 PH |
229 | ret = kstrtoul_from_user(ubuf, count, 10, &val); |
230 | if (ret) | |
e5a81b62 SR |
231 | return ret; |
232 | ||
a5e25883 | 233 | local_irq_save(flags); |
4f48f8b7 LJ |
234 | |
235 | /* | |
236 | * In case we trace inside arch_spin_lock() or after (NMI), | |
237 | * we will cause circular lock, so we also need to increase | |
238 | * the percpu trace_active here. | |
239 | */ | |
240 | cpu = smp_processor_id(); | |
241 | per_cpu(trace_active, cpu)++; | |
242 | ||
0199c4e6 | 243 | arch_spin_lock(&max_stack_lock); |
e5a81b62 | 244 | *ptr = val; |
0199c4e6 | 245 | arch_spin_unlock(&max_stack_lock); |
4f48f8b7 LJ |
246 | |
247 | per_cpu(trace_active, cpu)--; | |
a5e25883 | 248 | local_irq_restore(flags); |
e5a81b62 SR |
249 | |
250 | return count; | |
251 | } | |
252 | ||
f38f1d2a | 253 | static const struct file_operations stack_max_size_fops = { |
e5a81b62 SR |
254 | .open = tracing_open_generic, |
255 | .read = stack_max_size_read, | |
256 | .write = stack_max_size_write, | |
6038f373 | 257 | .llseek = default_llseek, |
e5a81b62 SR |
258 | }; |
259 | ||
260 | static void * | |
2fc5f0cf | 261 | __next(struct seq_file *m, loff_t *pos) |
e5a81b62 | 262 | { |
2fc5f0cf | 263 | long n = *pos - 1; |
e5a81b62 | 264 | |
2fc5f0cf | 265 | if (n >= max_stack_trace.nr_entries || stack_dump_trace[n] == ULONG_MAX) |
e5a81b62 SR |
266 | return NULL; |
267 | ||
2fc5f0cf | 268 | m->private = (void *)n; |
1b6cced6 | 269 | return &m->private; |
e5a81b62 SR |
270 | } |
271 | ||
2fc5f0cf LZ |
272 | static void * |
273 | t_next(struct seq_file *m, void *v, loff_t *pos) | |
e5a81b62 | 274 | { |
2fc5f0cf LZ |
275 | (*pos)++; |
276 | return __next(m, pos); | |
277 | } | |
e5a81b62 | 278 | |
2fc5f0cf LZ |
279 | static void *t_start(struct seq_file *m, loff_t *pos) |
280 | { | |
4f48f8b7 LJ |
281 | int cpu; |
282 | ||
e5a81b62 | 283 | local_irq_disable(); |
4f48f8b7 LJ |
284 | |
285 | cpu = smp_processor_id(); | |
286 | per_cpu(trace_active, cpu)++; | |
287 | ||
0199c4e6 | 288 | arch_spin_lock(&max_stack_lock); |
e5a81b62 | 289 | |
522a110b LW |
290 | if (*pos == 0) |
291 | return SEQ_START_TOKEN; | |
292 | ||
2fc5f0cf | 293 | return __next(m, pos); |
e5a81b62 SR |
294 | } |
295 | ||
296 | static void t_stop(struct seq_file *m, void *p) | |
297 | { | |
4f48f8b7 LJ |
298 | int cpu; |
299 | ||
0199c4e6 | 300 | arch_spin_unlock(&max_stack_lock); |
4f48f8b7 LJ |
301 | |
302 | cpu = smp_processor_id(); | |
303 | per_cpu(trace_active, cpu)--; | |
304 | ||
e5a81b62 SR |
305 | local_irq_enable(); |
306 | } | |
307 | ||
1b6cced6 | 308 | static int trace_lookup_stack(struct seq_file *m, long i) |
e5a81b62 | 309 | { |
1b6cced6 | 310 | unsigned long addr = stack_dump_trace[i]; |
e5a81b62 | 311 | |
151772db | 312 | return seq_printf(m, "%pS\n", (void *)addr); |
e5a81b62 SR |
313 | } |
314 | ||
e447e1df SR |
315 | static void print_disabled(struct seq_file *m) |
316 | { | |
317 | seq_puts(m, "#\n" | |
318 | "# Stack tracer disabled\n" | |
319 | "#\n" | |
320 | "# To enable the stack tracer, either add 'stacktrace' to the\n" | |
321 | "# kernel command line\n" | |
322 | "# or 'echo 1 > /proc/sys/kernel/stack_tracer_enabled'\n" | |
323 | "#\n"); | |
324 | } | |
325 | ||
e5a81b62 SR |
326 | static int t_show(struct seq_file *m, void *v) |
327 | { | |
522a110b | 328 | long i; |
1b6cced6 SR |
329 | int size; |
330 | ||
522a110b | 331 | if (v == SEQ_START_TOKEN) { |
eb1871f3 | 332 | seq_printf(m, " Depth Size Location" |
1b6cced6 | 333 | " (%d entries)\n" |
eb1871f3 | 334 | " ----- ---- --------\n", |
083a63b4 | 335 | max_stack_trace.nr_entries - 1); |
e447e1df SR |
336 | |
337 | if (!stack_tracer_enabled && !max_stack_size) | |
338 | print_disabled(m); | |
339 | ||
1b6cced6 SR |
340 | return 0; |
341 | } | |
e5a81b62 | 342 | |
522a110b LW |
343 | i = *(long *)v; |
344 | ||
1b6cced6 SR |
345 | if (i >= max_stack_trace.nr_entries || |
346 | stack_dump_trace[i] == ULONG_MAX) | |
e5a81b62 SR |
347 | return 0; |
348 | ||
1b6cced6 SR |
349 | if (i+1 == max_stack_trace.nr_entries || |
350 | stack_dump_trace[i+1] == ULONG_MAX) | |
351 | size = stack_dump_index[i]; | |
352 | else | |
353 | size = stack_dump_index[i] - stack_dump_index[i+1]; | |
354 | ||
355 | seq_printf(m, "%3ld) %8d %5d ", i, stack_dump_index[i], size); | |
356 | ||
357 | trace_lookup_stack(m, i); | |
e5a81b62 SR |
358 | |
359 | return 0; | |
360 | } | |
361 | ||
f38f1d2a | 362 | static const struct seq_operations stack_trace_seq_ops = { |
e5a81b62 SR |
363 | .start = t_start, |
364 | .next = t_next, | |
365 | .stop = t_stop, | |
366 | .show = t_show, | |
367 | }; | |
368 | ||
369 | static int stack_trace_open(struct inode *inode, struct file *file) | |
370 | { | |
d8cc1ab7 | 371 | return seq_open(file, &stack_trace_seq_ops); |
e5a81b62 SR |
372 | } |
373 | ||
f38f1d2a | 374 | static const struct file_operations stack_trace_fops = { |
e5a81b62 SR |
375 | .open = stack_trace_open, |
376 | .read = seq_read, | |
377 | .llseek = seq_lseek, | |
d8cc1ab7 | 378 | .release = seq_release, |
e5a81b62 SR |
379 | }; |
380 | ||
d2d45c7a SR |
381 | static int |
382 | stack_trace_filter_open(struct inode *inode, struct file *file) | |
383 | { | |
384 | return ftrace_regex_open(&trace_ops, FTRACE_ITER_FILTER, | |
385 | inode, file); | |
386 | } | |
387 | ||
388 | static const struct file_operations stack_trace_filter_fops = { | |
389 | .open = stack_trace_filter_open, | |
390 | .read = seq_read, | |
391 | .write = ftrace_filter_write, | |
098c879e | 392 | .llseek = tracing_lseek, |
d2d45c7a SR |
393 | .release = ftrace_regex_release, |
394 | }; | |
395 | ||
f38f1d2a SR |
396 | int |
397 | stack_trace_sysctl(struct ctl_table *table, int write, | |
8d65af78 | 398 | void __user *buffer, size_t *lenp, |
f38f1d2a SR |
399 | loff_t *ppos) |
400 | { | |
401 | int ret; | |
402 | ||
403 | mutex_lock(&stack_sysctl_mutex); | |
404 | ||
8d65af78 | 405 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
f38f1d2a SR |
406 | |
407 | if (ret || !write || | |
a32c7765 | 408 | (last_stack_tracer_enabled == !!stack_tracer_enabled)) |
f38f1d2a SR |
409 | goto out; |
410 | ||
a32c7765 | 411 | last_stack_tracer_enabled = !!stack_tracer_enabled; |
f38f1d2a SR |
412 | |
413 | if (stack_tracer_enabled) | |
414 | register_ftrace_function(&trace_ops); | |
415 | else | |
416 | unregister_ftrace_function(&trace_ops); | |
417 | ||
418 | out: | |
419 | mutex_unlock(&stack_sysctl_mutex); | |
420 | return ret; | |
421 | } | |
422 | ||
762e1207 SR |
423 | static char stack_trace_filter_buf[COMMAND_LINE_SIZE+1] __initdata; |
424 | ||
f38f1d2a SR |
425 | static __init int enable_stacktrace(char *str) |
426 | { | |
762e1207 SR |
427 | if (strncmp(str, "_filter=", 8) == 0) |
428 | strncpy(stack_trace_filter_buf, str+8, COMMAND_LINE_SIZE); | |
429 | ||
e05a43b7 SR |
430 | stack_tracer_enabled = 1; |
431 | last_stack_tracer_enabled = 1; | |
f38f1d2a SR |
432 | return 1; |
433 | } | |
434 | __setup("stacktrace", enable_stacktrace); | |
435 | ||
e5a81b62 SR |
436 | static __init int stack_trace_init(void) |
437 | { | |
438 | struct dentry *d_tracer; | |
e5a81b62 SR |
439 | |
440 | d_tracer = tracing_init_dentry(); | |
ed6f1c99 NK |
441 | if (!d_tracer) |
442 | return 0; | |
e5a81b62 | 443 | |
5452af66 FW |
444 | trace_create_file("stack_max_size", 0644, d_tracer, |
445 | &max_stack_size, &stack_max_size_fops); | |
e5a81b62 | 446 | |
5452af66 FW |
447 | trace_create_file("stack_trace", 0444, d_tracer, |
448 | NULL, &stack_trace_fops); | |
e5a81b62 | 449 | |
d2d45c7a SR |
450 | trace_create_file("stack_trace_filter", 0444, d_tracer, |
451 | NULL, &stack_trace_filter_fops); | |
452 | ||
762e1207 SR |
453 | if (stack_trace_filter_buf[0]) |
454 | ftrace_set_early_filter(&trace_ops, stack_trace_filter_buf, 1); | |
455 | ||
e05a43b7 | 456 | if (stack_tracer_enabled) |
f38f1d2a | 457 | register_ftrace_function(&trace_ops); |
e5a81b62 SR |
458 | |
459 | return 0; | |
460 | } | |
461 | ||
462 | device_initcall(stack_trace_init); |