ftracetest: Fix instance test to use proper shell command for pids
[linux-2.6-block.git] / kernel / trace / trace_stack.c
CommitLineData
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>
e5a81b62
SR
10#include <linux/ftrace.h>
11#include <linux/module.h>
f38f1d2a 12#include <linux/sysctl.h>
e5a81b62 13#include <linux/init.h>
762e1207
SR
14
15#include <asm/setup.h>
16
e5a81b62
SR
17#include "trace.h"
18
1b6cced6
SR
19static unsigned long stack_dump_trace[STACK_TRACE_ENTRIES+1] =
20 { [0 ... (STACK_TRACE_ENTRIES)] = ULONG_MAX };
bb99d8cc 21unsigned stack_trace_index[STACK_TRACE_ENTRIES];
1b6cced6 22
4df29712
SRRH
23/*
24 * Reserve one entry for the passed in ip. This will allow
25 * us to remove most or all of the stack size overhead
26 * added by the stack tracer itself.
27 */
bb99d8cc 28struct stack_trace stack_trace_max = {
4df29712 29 .max_entries = STACK_TRACE_ENTRIES - 1,
72ac426a 30 .entries = &stack_dump_trace[0],
e5a81b62
SR
31};
32
bb99d8cc 33unsigned long stack_trace_max_size;
d332736d 34arch_spinlock_t stack_trace_max_lock =
edc35bd7 35 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
e5a81b62 36
e5a81b62 37static DEFINE_PER_CPU(int, trace_active);
f38f1d2a
SR
38static DEFINE_MUTEX(stack_sysctl_mutex);
39
40int stack_tracer_enabled;
41static int last_stack_tracer_enabled;
e5a81b62 42
bb99d8cc 43void stack_trace_print(void)
e3172181
MK
44{
45 long i;
46 int size;
47
48 pr_emerg(" Depth Size Location (%d entries)\n"
49 " ----- ---- --------\n",
bb99d8cc 50 stack_trace_max.nr_entries);
e3172181 51
bb99d8cc 52 for (i = 0; i < stack_trace_max.nr_entries; i++) {
e3172181
MK
53 if (stack_dump_trace[i] == ULONG_MAX)
54 break;
bb99d8cc 55 if (i+1 == stack_trace_max.nr_entries ||
e3172181 56 stack_dump_trace[i+1] == ULONG_MAX)
bb99d8cc 57 size = stack_trace_index[i];
e3172181 58 else
bb99d8cc 59 size = stack_trace_index[i] - stack_trace_index[i+1];
e3172181 60
bb99d8cc 61 pr_emerg("%3ld) %8d %5d %pS\n", i, stack_trace_index[i],
e3172181
MK
62 size, (void *)stack_dump_trace[i]);
63 }
64}
65
bb99d8cc
AT
66/*
67 * When arch-specific code overides this function, the following
d332736d 68 * data should be filled up, assuming stack_trace_max_lock is held to
bb99d8cc
AT
69 * prevent concurrent updates.
70 * stack_trace_index[]
71 * stack_trace_max
72 * stack_trace_max_size
73 */
74void __weak
d4ecbfc4 75check_stack(unsigned long ip, unsigned long *stack)
e5a81b62 76{
e3172181 77 unsigned long this_size, flags; unsigned long *p, *top, *start;
4df29712
SRRH
78 static int tracer_frame;
79 int frame_size = ACCESS_ONCE(tracer_frame);
72ac426a 80 int i, x;
e5a81b62 81
87889501 82 this_size = ((unsigned long)stack) & (THREAD_SIZE-1);
e5a81b62 83 this_size = THREAD_SIZE - this_size;
4df29712
SRRH
84 /* Remove the frame of the tracer */
85 this_size -= frame_size;
e5a81b62 86
bb99d8cc 87 if (this_size <= stack_trace_max_size)
e5a81b62
SR
88 return;
89
81520a1b 90 /* we do not handle interrupt stacks yet */
87889501 91 if (!object_is_on_stack(stack))
81520a1b
SR
92 return;
93
1904be1b
SRRH
94 /* Can't do this from NMI context (can cause deadlocks) */
95 if (in_nmi())
96 return;
97
a5e25883 98 local_irq_save(flags);
d332736d 99 arch_spin_lock(&stack_trace_max_lock);
e5a81b62 100
a2d76290
SRRH
101 /*
102 * RCU may not be watching, make it see us.
103 * The stack trace code uses rcu_sched.
104 */
105 rcu_irq_enter();
106
4df29712
SRRH
107 /* In case another CPU set the tracer_frame on us */
108 if (unlikely(!frame_size))
109 this_size -= tracer_frame;
110
e5a81b62 111 /* a race could have already updated it */
bb99d8cc 112 if (this_size <= stack_trace_max_size)
e5a81b62
SR
113 goto out;
114
bb99d8cc 115 stack_trace_max_size = this_size;
e5a81b62 116
bb99d8cc
AT
117 stack_trace_max.nr_entries = 0;
118 stack_trace_max.skip = 3;
e5a81b62 119
bb99d8cc 120 save_stack_trace(&stack_trace_max);
e5a81b62 121
72ac426a 122 /* Skip over the overhead of the stack tracer itself */
bb99d8cc 123 for (i = 0; i < stack_trace_max.nr_entries; i++) {
72ac426a
SRRH
124 if (stack_dump_trace[i] == ip)
125 break;
126 }
d4ecbfc4 127
6ccd8371
SR
128 /*
129 * Some archs may not have the passed in ip in the dump.
130 * If that happens, we need to show everything.
131 */
132 if (i == stack_trace_max.nr_entries)
133 i = 0;
134
1b6cced6
SR
135 /*
136 * Now find where in the stack these are.
137 */
72ac426a 138 x = 0;
87889501 139 start = stack;
1b6cced6
SR
140 top = (unsigned long *)
141 (((unsigned long)start & ~(THREAD_SIZE-1)) + THREAD_SIZE);
142
143 /*
144 * Loop through all the entries. One of the entries may
145 * for some reason be missed on the stack, so we may
146 * have to account for them. If they are all there, this
147 * loop will only happen once. This code only takes place
148 * on a new max, so it is far from a fast path.
149 */
bb99d8cc 150 while (i < stack_trace_max.nr_entries) {
0a37119d 151 int found = 0;
1b6cced6 152
bb99d8cc 153 stack_trace_index[x] = this_size;
1b6cced6
SR
154 p = start;
155
bb99d8cc 156 for (; p < top && i < stack_trace_max.nr_entries; p++) {
72ac426a
SRRH
157 if (stack_dump_trace[i] == ULONG_MAX)
158 break;
1b6cced6 159 if (*p == stack_dump_trace[i]) {
72ac426a 160 stack_dump_trace[x] = stack_dump_trace[i++];
bb99d8cc 161 this_size = stack_trace_index[x++] =
1b6cced6 162 (top - p) * sizeof(unsigned long);
0a37119d 163 found = 1;
1b6cced6
SR
164 /* Start the search from here */
165 start = p + 1;
4df29712
SRRH
166 /*
167 * We do not want to show the overhead
168 * of the stack tracer stack in the
169 * max stack. If we haven't figured
170 * out what that is, then figure it out
171 * now.
172 */
72ac426a 173 if (unlikely(!tracer_frame)) {
4df29712
SRRH
174 tracer_frame = (p - stack) *
175 sizeof(unsigned long);
bb99d8cc 176 stack_trace_max_size -= tracer_frame;
4df29712 177 }
1b6cced6
SR
178 }
179 }
180
0a37119d
SR
181 if (!found)
182 i++;
1b6cced6
SR
183 }
184
bb99d8cc 185 stack_trace_max.nr_entries = x;
72ac426a
SRRH
186 for (; x < i; x++)
187 stack_dump_trace[x] = ULONG_MAX;
188
a70857e4 189 if (task_stack_end_corrupted(current)) {
bb99d8cc 190 stack_trace_print();
e3172181
MK
191 BUG();
192 }
193
e5a81b62 194 out:
a2d76290 195 rcu_irq_exit();
d332736d 196 arch_spin_unlock(&stack_trace_max_lock);
a5e25883 197 local_irq_restore(flags);
e5a81b62
SR
198}
199
200static void
a1e2e31d
SR
201stack_trace_call(unsigned long ip, unsigned long parent_ip,
202 struct ftrace_ops *op, struct pt_regs *pt_regs)
e5a81b62 203{
87889501 204 unsigned long stack;
5168ae50 205 int cpu;
e5a81b62 206
5168ae50 207 preempt_disable_notrace();
e5a81b62
SR
208
209 cpu = raw_smp_processor_id();
210 /* no atomic needed, we only modify this variable by this cpu */
211 if (per_cpu(trace_active, cpu)++ != 0)
212 goto out;
213
72ac426a 214 ip += MCOUNT_INSN_SIZE;
4df29712
SRRH
215
216 check_stack(ip, &stack);
e5a81b62
SR
217
218 out:
219 per_cpu(trace_active, cpu)--;
220 /* prevent recursion in schedule */
5168ae50 221 preempt_enable_notrace();
e5a81b62
SR
222}
223
224static struct ftrace_ops trace_ops __read_mostly =
225{
226 .func = stack_trace_call,
4740974a 227 .flags = FTRACE_OPS_FL_RECURSION_SAFE,
e5a81b62
SR
228};
229
230static ssize_t
231stack_max_size_read(struct file *filp, char __user *ubuf,
232 size_t count, loff_t *ppos)
233{
234 unsigned long *ptr = filp->private_data;
235 char buf[64];
236 int r;
237
238 r = snprintf(buf, sizeof(buf), "%ld\n", *ptr);
239 if (r > sizeof(buf))
240 r = sizeof(buf);
241 return simple_read_from_buffer(ubuf, count, ppos, buf, r);
242}
243
244static ssize_t
245stack_max_size_write(struct file *filp, const char __user *ubuf,
246 size_t count, loff_t *ppos)
247{
248 long *ptr = filp->private_data;
249 unsigned long val, flags;
e5a81b62 250 int ret;
4f48f8b7 251 int cpu;
e5a81b62 252
22fe9b54
PH
253 ret = kstrtoul_from_user(ubuf, count, 10, &val);
254 if (ret)
e5a81b62
SR
255 return ret;
256
a5e25883 257 local_irq_save(flags);
4f48f8b7
LJ
258
259 /*
260 * In case we trace inside arch_spin_lock() or after (NMI),
261 * we will cause circular lock, so we also need to increase
262 * the percpu trace_active here.
263 */
264 cpu = smp_processor_id();
265 per_cpu(trace_active, cpu)++;
266
d332736d 267 arch_spin_lock(&stack_trace_max_lock);
e5a81b62 268 *ptr = val;
d332736d 269 arch_spin_unlock(&stack_trace_max_lock);
4f48f8b7
LJ
270
271 per_cpu(trace_active, cpu)--;
a5e25883 272 local_irq_restore(flags);
e5a81b62
SR
273
274 return count;
275}
276
f38f1d2a 277static const struct file_operations stack_max_size_fops = {
e5a81b62
SR
278 .open = tracing_open_generic,
279 .read = stack_max_size_read,
280 .write = stack_max_size_write,
6038f373 281 .llseek = default_llseek,
e5a81b62
SR
282};
283
284static void *
2fc5f0cf 285__next(struct seq_file *m, loff_t *pos)
e5a81b62 286{
2fc5f0cf 287 long n = *pos - 1;
e5a81b62 288
bb99d8cc 289 if (n > stack_trace_max.nr_entries || stack_dump_trace[n] == ULONG_MAX)
e5a81b62
SR
290 return NULL;
291
2fc5f0cf 292 m->private = (void *)n;
1b6cced6 293 return &m->private;
e5a81b62
SR
294}
295
2fc5f0cf
LZ
296static void *
297t_next(struct seq_file *m, void *v, loff_t *pos)
e5a81b62 298{
2fc5f0cf
LZ
299 (*pos)++;
300 return __next(m, pos);
301}
e5a81b62 302
2fc5f0cf
LZ
303static void *t_start(struct seq_file *m, loff_t *pos)
304{
4f48f8b7
LJ
305 int cpu;
306
e5a81b62 307 local_irq_disable();
4f48f8b7
LJ
308
309 cpu = smp_processor_id();
310 per_cpu(trace_active, cpu)++;
311
d332736d 312 arch_spin_lock(&stack_trace_max_lock);
e5a81b62 313
522a110b
LW
314 if (*pos == 0)
315 return SEQ_START_TOKEN;
316
2fc5f0cf 317 return __next(m, pos);
e5a81b62
SR
318}
319
320static void t_stop(struct seq_file *m, void *p)
321{
4f48f8b7
LJ
322 int cpu;
323
d332736d 324 arch_spin_unlock(&stack_trace_max_lock);
4f48f8b7
LJ
325
326 cpu = smp_processor_id();
327 per_cpu(trace_active, cpu)--;
328
e5a81b62
SR
329 local_irq_enable();
330}
331
962e3707 332static void trace_lookup_stack(struct seq_file *m, long i)
e5a81b62 333{
1b6cced6 334 unsigned long addr = stack_dump_trace[i];
e5a81b62 335
962e3707 336 seq_printf(m, "%pS\n", (void *)addr);
e5a81b62
SR
337}
338
e447e1df
SR
339static void print_disabled(struct seq_file *m)
340{
341 seq_puts(m, "#\n"
342 "# Stack tracer disabled\n"
343 "#\n"
344 "# To enable the stack tracer, either add 'stacktrace' to the\n"
345 "# kernel command line\n"
346 "# or 'echo 1 > /proc/sys/kernel/stack_tracer_enabled'\n"
347 "#\n");
348}
349
e5a81b62
SR
350static int t_show(struct seq_file *m, void *v)
351{
522a110b 352 long i;
1b6cced6
SR
353 int size;
354
522a110b 355 if (v == SEQ_START_TOKEN) {
eb1871f3 356 seq_printf(m, " Depth Size Location"
1b6cced6 357 " (%d entries)\n"
eb1871f3 358 " ----- ---- --------\n",
bb99d8cc 359 stack_trace_max.nr_entries);
e447e1df 360
bb99d8cc 361 if (!stack_tracer_enabled && !stack_trace_max_size)
e447e1df
SR
362 print_disabled(m);
363
1b6cced6
SR
364 return 0;
365 }
e5a81b62 366
522a110b
LW
367 i = *(long *)v;
368
bb99d8cc 369 if (i >= stack_trace_max.nr_entries ||
1b6cced6 370 stack_dump_trace[i] == ULONG_MAX)
e5a81b62
SR
371 return 0;
372
bb99d8cc 373 if (i+1 == stack_trace_max.nr_entries ||
1b6cced6 374 stack_dump_trace[i+1] == ULONG_MAX)
bb99d8cc 375 size = stack_trace_index[i];
1b6cced6 376 else
bb99d8cc 377 size = stack_trace_index[i] - stack_trace_index[i+1];
1b6cced6 378
bb99d8cc 379 seq_printf(m, "%3ld) %8d %5d ", i, stack_trace_index[i], size);
1b6cced6
SR
380
381 trace_lookup_stack(m, i);
e5a81b62
SR
382
383 return 0;
384}
385
f38f1d2a 386static const struct seq_operations stack_trace_seq_ops = {
e5a81b62
SR
387 .start = t_start,
388 .next = t_next,
389 .stop = t_stop,
390 .show = t_show,
391};
392
393static int stack_trace_open(struct inode *inode, struct file *file)
394{
d8cc1ab7 395 return seq_open(file, &stack_trace_seq_ops);
e5a81b62
SR
396}
397
f38f1d2a 398static const struct file_operations stack_trace_fops = {
e5a81b62
SR
399 .open = stack_trace_open,
400 .read = seq_read,
401 .llseek = seq_lseek,
d8cc1ab7 402 .release = seq_release,
e5a81b62
SR
403};
404
d2d45c7a
SR
405static int
406stack_trace_filter_open(struct inode *inode, struct file *file)
407{
408 return ftrace_regex_open(&trace_ops, FTRACE_ITER_FILTER,
409 inode, file);
410}
411
412static const struct file_operations stack_trace_filter_fops = {
413 .open = stack_trace_filter_open,
414 .read = seq_read,
415 .write = ftrace_filter_write,
098c879e 416 .llseek = tracing_lseek,
d2d45c7a
SR
417 .release = ftrace_regex_release,
418};
419
f38f1d2a
SR
420int
421stack_trace_sysctl(struct ctl_table *table, int write,
8d65af78 422 void __user *buffer, size_t *lenp,
f38f1d2a
SR
423 loff_t *ppos)
424{
425 int ret;
426
427 mutex_lock(&stack_sysctl_mutex);
428
8d65af78 429 ret = proc_dointvec(table, write, buffer, lenp, ppos);
f38f1d2a
SR
430
431 if (ret || !write ||
a32c7765 432 (last_stack_tracer_enabled == !!stack_tracer_enabled))
f38f1d2a
SR
433 goto out;
434
a32c7765 435 last_stack_tracer_enabled = !!stack_tracer_enabled;
f38f1d2a
SR
436
437 if (stack_tracer_enabled)
438 register_ftrace_function(&trace_ops);
439 else
440 unregister_ftrace_function(&trace_ops);
441
442 out:
443 mutex_unlock(&stack_sysctl_mutex);
444 return ret;
445}
446
762e1207
SR
447static char stack_trace_filter_buf[COMMAND_LINE_SIZE+1] __initdata;
448
f38f1d2a
SR
449static __init int enable_stacktrace(char *str)
450{
762e1207
SR
451 if (strncmp(str, "_filter=", 8) == 0)
452 strncpy(stack_trace_filter_buf, str+8, COMMAND_LINE_SIZE);
453
e05a43b7
SR
454 stack_tracer_enabled = 1;
455 last_stack_tracer_enabled = 1;
f38f1d2a
SR
456 return 1;
457}
458__setup("stacktrace", enable_stacktrace);
459
e5a81b62
SR
460static __init int stack_trace_init(void)
461{
462 struct dentry *d_tracer;
e5a81b62
SR
463
464 d_tracer = tracing_init_dentry();
14a5ae40 465 if (IS_ERR(d_tracer))
ed6f1c99 466 return 0;
e5a81b62 467
5452af66 468 trace_create_file("stack_max_size", 0644, d_tracer,
bb99d8cc 469 &stack_trace_max_size, &stack_max_size_fops);
e5a81b62 470
5452af66
FW
471 trace_create_file("stack_trace", 0444, d_tracer,
472 NULL, &stack_trace_fops);
e5a81b62 473
d2d45c7a
SR
474 trace_create_file("stack_trace_filter", 0444, d_tracer,
475 NULL, &stack_trace_filter_fops);
476
762e1207
SR
477 if (stack_trace_filter_buf[0])
478 ftrace_set_early_filter(&trace_ops, stack_trace_filter_buf, 1);
479
e05a43b7 480 if (stack_tracer_enabled)
f38f1d2a 481 register_ftrace_function(&trace_ops);
e5a81b62
SR
482
483 return 0;
484}
485
486device_initcall(stack_trace_init);