nvme: fix Kconfig description for BLK_DEV_NVME_SCSI
[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
1b6cced6
SR
128 /*
129 * Now find where in the stack these are.
130 */
72ac426a 131 x = 0;
87889501 132 start = stack;
1b6cced6
SR
133 top = (unsigned long *)
134 (((unsigned long)start & ~(THREAD_SIZE-1)) + THREAD_SIZE);
135
136 /*
137 * Loop through all the entries. One of the entries may
138 * for some reason be missed on the stack, so we may
139 * have to account for them. If they are all there, this
140 * loop will only happen once. This code only takes place
141 * on a new max, so it is far from a fast path.
142 */
bb99d8cc 143 while (i < stack_trace_max.nr_entries) {
0a37119d 144 int found = 0;
1b6cced6 145
bb99d8cc 146 stack_trace_index[x] = this_size;
1b6cced6
SR
147 p = start;
148
bb99d8cc 149 for (; p < top && i < stack_trace_max.nr_entries; p++) {
72ac426a
SRRH
150 if (stack_dump_trace[i] == ULONG_MAX)
151 break;
1b6cced6 152 if (*p == stack_dump_trace[i]) {
72ac426a 153 stack_dump_trace[x] = stack_dump_trace[i++];
bb99d8cc 154 this_size = stack_trace_index[x++] =
1b6cced6 155 (top - p) * sizeof(unsigned long);
0a37119d 156 found = 1;
1b6cced6
SR
157 /* Start the search from here */
158 start = p + 1;
4df29712
SRRH
159 /*
160 * We do not want to show the overhead
161 * of the stack tracer stack in the
162 * max stack. If we haven't figured
163 * out what that is, then figure it out
164 * now.
165 */
72ac426a 166 if (unlikely(!tracer_frame)) {
4df29712
SRRH
167 tracer_frame = (p - stack) *
168 sizeof(unsigned long);
bb99d8cc 169 stack_trace_max_size -= tracer_frame;
4df29712 170 }
1b6cced6
SR
171 }
172 }
173
0a37119d
SR
174 if (!found)
175 i++;
1b6cced6
SR
176 }
177
bb99d8cc 178 stack_trace_max.nr_entries = x;
72ac426a
SRRH
179 for (; x < i; x++)
180 stack_dump_trace[x] = ULONG_MAX;
181
a70857e4 182 if (task_stack_end_corrupted(current)) {
bb99d8cc 183 stack_trace_print();
e3172181
MK
184 BUG();
185 }
186
e5a81b62 187 out:
a2d76290 188 rcu_irq_exit();
d332736d 189 arch_spin_unlock(&stack_trace_max_lock);
a5e25883 190 local_irq_restore(flags);
e5a81b62
SR
191}
192
193static void
a1e2e31d
SR
194stack_trace_call(unsigned long ip, unsigned long parent_ip,
195 struct ftrace_ops *op, struct pt_regs *pt_regs)
e5a81b62 196{
87889501 197 unsigned long stack;
5168ae50 198 int cpu;
e5a81b62 199
5168ae50 200 preempt_disable_notrace();
e5a81b62
SR
201
202 cpu = raw_smp_processor_id();
203 /* no atomic needed, we only modify this variable by this cpu */
204 if (per_cpu(trace_active, cpu)++ != 0)
205 goto out;
206
72ac426a 207 ip += MCOUNT_INSN_SIZE;
4df29712
SRRH
208
209 check_stack(ip, &stack);
e5a81b62
SR
210
211 out:
212 per_cpu(trace_active, cpu)--;
213 /* prevent recursion in schedule */
5168ae50 214 preempt_enable_notrace();
e5a81b62
SR
215}
216
217static struct ftrace_ops trace_ops __read_mostly =
218{
219 .func = stack_trace_call,
4740974a 220 .flags = FTRACE_OPS_FL_RECURSION_SAFE,
e5a81b62
SR
221};
222
223static ssize_t
224stack_max_size_read(struct file *filp, char __user *ubuf,
225 size_t count, loff_t *ppos)
226{
227 unsigned long *ptr = filp->private_data;
228 char buf[64];
229 int r;
230
231 r = snprintf(buf, sizeof(buf), "%ld\n", *ptr);
232 if (r > sizeof(buf))
233 r = sizeof(buf);
234 return simple_read_from_buffer(ubuf, count, ppos, buf, r);
235}
236
237static ssize_t
238stack_max_size_write(struct file *filp, const char __user *ubuf,
239 size_t count, loff_t *ppos)
240{
241 long *ptr = filp->private_data;
242 unsigned long val, flags;
e5a81b62 243 int ret;
4f48f8b7 244 int cpu;
e5a81b62 245
22fe9b54
PH
246 ret = kstrtoul_from_user(ubuf, count, 10, &val);
247 if (ret)
e5a81b62
SR
248 return ret;
249
a5e25883 250 local_irq_save(flags);
4f48f8b7
LJ
251
252 /*
253 * In case we trace inside arch_spin_lock() or after (NMI),
254 * we will cause circular lock, so we also need to increase
255 * the percpu trace_active here.
256 */
257 cpu = smp_processor_id();
258 per_cpu(trace_active, cpu)++;
259
d332736d 260 arch_spin_lock(&stack_trace_max_lock);
e5a81b62 261 *ptr = val;
d332736d 262 arch_spin_unlock(&stack_trace_max_lock);
4f48f8b7
LJ
263
264 per_cpu(trace_active, cpu)--;
a5e25883 265 local_irq_restore(flags);
e5a81b62
SR
266
267 return count;
268}
269
f38f1d2a 270static const struct file_operations stack_max_size_fops = {
e5a81b62
SR
271 .open = tracing_open_generic,
272 .read = stack_max_size_read,
273 .write = stack_max_size_write,
6038f373 274 .llseek = default_llseek,
e5a81b62
SR
275};
276
277static void *
2fc5f0cf 278__next(struct seq_file *m, loff_t *pos)
e5a81b62 279{
2fc5f0cf 280 long n = *pos - 1;
e5a81b62 281
bb99d8cc 282 if (n > stack_trace_max.nr_entries || stack_dump_trace[n] == ULONG_MAX)
e5a81b62
SR
283 return NULL;
284
2fc5f0cf 285 m->private = (void *)n;
1b6cced6 286 return &m->private;
e5a81b62
SR
287}
288
2fc5f0cf
LZ
289static void *
290t_next(struct seq_file *m, void *v, loff_t *pos)
e5a81b62 291{
2fc5f0cf
LZ
292 (*pos)++;
293 return __next(m, pos);
294}
e5a81b62 295
2fc5f0cf
LZ
296static void *t_start(struct seq_file *m, loff_t *pos)
297{
4f48f8b7
LJ
298 int cpu;
299
e5a81b62 300 local_irq_disable();
4f48f8b7
LJ
301
302 cpu = smp_processor_id();
303 per_cpu(trace_active, cpu)++;
304
d332736d 305 arch_spin_lock(&stack_trace_max_lock);
e5a81b62 306
522a110b
LW
307 if (*pos == 0)
308 return SEQ_START_TOKEN;
309
2fc5f0cf 310 return __next(m, pos);
e5a81b62
SR
311}
312
313static void t_stop(struct seq_file *m, void *p)
314{
4f48f8b7
LJ
315 int cpu;
316
d332736d 317 arch_spin_unlock(&stack_trace_max_lock);
4f48f8b7
LJ
318
319 cpu = smp_processor_id();
320 per_cpu(trace_active, cpu)--;
321
e5a81b62
SR
322 local_irq_enable();
323}
324
962e3707 325static void trace_lookup_stack(struct seq_file *m, long i)
e5a81b62 326{
1b6cced6 327 unsigned long addr = stack_dump_trace[i];
e5a81b62 328
962e3707 329 seq_printf(m, "%pS\n", (void *)addr);
e5a81b62
SR
330}
331
e447e1df
SR
332static void print_disabled(struct seq_file *m)
333{
334 seq_puts(m, "#\n"
335 "# Stack tracer disabled\n"
336 "#\n"
337 "# To enable the stack tracer, either add 'stacktrace' to the\n"
338 "# kernel command line\n"
339 "# or 'echo 1 > /proc/sys/kernel/stack_tracer_enabled'\n"
340 "#\n");
341}
342
e5a81b62
SR
343static int t_show(struct seq_file *m, void *v)
344{
522a110b 345 long i;
1b6cced6
SR
346 int size;
347
522a110b 348 if (v == SEQ_START_TOKEN) {
eb1871f3 349 seq_printf(m, " Depth Size Location"
1b6cced6 350 " (%d entries)\n"
eb1871f3 351 " ----- ---- --------\n",
bb99d8cc 352 stack_trace_max.nr_entries);
e447e1df 353
bb99d8cc 354 if (!stack_tracer_enabled && !stack_trace_max_size)
e447e1df
SR
355 print_disabled(m);
356
1b6cced6
SR
357 return 0;
358 }
e5a81b62 359
522a110b
LW
360 i = *(long *)v;
361
bb99d8cc 362 if (i >= stack_trace_max.nr_entries ||
1b6cced6 363 stack_dump_trace[i] == ULONG_MAX)
e5a81b62
SR
364 return 0;
365
bb99d8cc 366 if (i+1 == stack_trace_max.nr_entries ||
1b6cced6 367 stack_dump_trace[i+1] == ULONG_MAX)
bb99d8cc 368 size = stack_trace_index[i];
1b6cced6 369 else
bb99d8cc 370 size = stack_trace_index[i] - stack_trace_index[i+1];
1b6cced6 371
bb99d8cc 372 seq_printf(m, "%3ld) %8d %5d ", i, stack_trace_index[i], size);
1b6cced6
SR
373
374 trace_lookup_stack(m, i);
e5a81b62
SR
375
376 return 0;
377}
378
f38f1d2a 379static const struct seq_operations stack_trace_seq_ops = {
e5a81b62
SR
380 .start = t_start,
381 .next = t_next,
382 .stop = t_stop,
383 .show = t_show,
384};
385
386static int stack_trace_open(struct inode *inode, struct file *file)
387{
d8cc1ab7 388 return seq_open(file, &stack_trace_seq_ops);
e5a81b62
SR
389}
390
f38f1d2a 391static const struct file_operations stack_trace_fops = {
e5a81b62
SR
392 .open = stack_trace_open,
393 .read = seq_read,
394 .llseek = seq_lseek,
d8cc1ab7 395 .release = seq_release,
e5a81b62
SR
396};
397
d2d45c7a
SR
398static int
399stack_trace_filter_open(struct inode *inode, struct file *file)
400{
401 return ftrace_regex_open(&trace_ops, FTRACE_ITER_FILTER,
402 inode, file);
403}
404
405static const struct file_operations stack_trace_filter_fops = {
406 .open = stack_trace_filter_open,
407 .read = seq_read,
408 .write = ftrace_filter_write,
098c879e 409 .llseek = tracing_lseek,
d2d45c7a
SR
410 .release = ftrace_regex_release,
411};
412
f38f1d2a
SR
413int
414stack_trace_sysctl(struct ctl_table *table, int write,
8d65af78 415 void __user *buffer, size_t *lenp,
f38f1d2a
SR
416 loff_t *ppos)
417{
418 int ret;
419
420 mutex_lock(&stack_sysctl_mutex);
421
8d65af78 422 ret = proc_dointvec(table, write, buffer, lenp, ppos);
f38f1d2a
SR
423
424 if (ret || !write ||
a32c7765 425 (last_stack_tracer_enabled == !!stack_tracer_enabled))
f38f1d2a
SR
426 goto out;
427
a32c7765 428 last_stack_tracer_enabled = !!stack_tracer_enabled;
f38f1d2a
SR
429
430 if (stack_tracer_enabled)
431 register_ftrace_function(&trace_ops);
432 else
433 unregister_ftrace_function(&trace_ops);
434
435 out:
436 mutex_unlock(&stack_sysctl_mutex);
437 return ret;
438}
439
762e1207
SR
440static char stack_trace_filter_buf[COMMAND_LINE_SIZE+1] __initdata;
441
f38f1d2a
SR
442static __init int enable_stacktrace(char *str)
443{
762e1207
SR
444 if (strncmp(str, "_filter=", 8) == 0)
445 strncpy(stack_trace_filter_buf, str+8, COMMAND_LINE_SIZE);
446
e05a43b7
SR
447 stack_tracer_enabled = 1;
448 last_stack_tracer_enabled = 1;
f38f1d2a
SR
449 return 1;
450}
451__setup("stacktrace", enable_stacktrace);
452
e5a81b62
SR
453static __init int stack_trace_init(void)
454{
455 struct dentry *d_tracer;
e5a81b62
SR
456
457 d_tracer = tracing_init_dentry();
14a5ae40 458 if (IS_ERR(d_tracer))
ed6f1c99 459 return 0;
e5a81b62 460
5452af66 461 trace_create_file("stack_max_size", 0644, d_tracer,
bb99d8cc 462 &stack_trace_max_size, &stack_max_size_fops);
e5a81b62 463
5452af66
FW
464 trace_create_file("stack_trace", 0444, d_tracer,
465 NULL, &stack_trace_fops);
e5a81b62 466
d2d45c7a
SR
467 trace_create_file("stack_trace_filter", 0444, d_tracer,
468 NULL, &stack_trace_filter_fops);
469
762e1207
SR
470 if (stack_trace_filter_buf[0])
471 ftrace_set_early_filter(&trace_ops, stack_trace_filter_buf, 1);
472
e05a43b7 473 if (stack_tracer_enabled)
f38f1d2a 474 register_ftrace_function(&trace_ops);
e5a81b62
SR
475
476 return 0;
477}
478
479device_initcall(stack_trace_init);