trace_stack.c 10.3 KB
Newer Older
S
Steven Rostedt 已提交
1 2 3 4 5 6 7 8 9 10 11
/*
 * Copyright (C) 2008 Steven Rostedt <srostedt@redhat.com>
 *
 */
#include <linux/stacktrace.h>
#include <linux/kallsyms.h>
#include <linux/seq_file.h>
#include <linux/spinlock.h>
#include <linux/uaccess.h>
#include <linux/ftrace.h>
#include <linux/module.h>
12
#include <linux/sysctl.h>
S
Steven Rostedt 已提交
13
#include <linux/init.h>
14 15 16

#include <asm/setup.h>

S
Steven Rostedt 已提交
17 18
#include "trace.h"

S
Steven Rostedt 已提交
19 20
static unsigned long stack_dump_trace[STACK_TRACE_ENTRIES+1] =
	 { [0 ... (STACK_TRACE_ENTRIES)] = ULONG_MAX };
21
unsigned stack_trace_index[STACK_TRACE_ENTRIES];
S
Steven Rostedt 已提交
22

23 24 25 26 27
/*
 * Reserve one entry for the passed in ip. This will allow
 * us to remove most or all of the stack size overhead
 * added by the stack tracer itself.
 */
28
struct stack_trace stack_trace_max = {
29
	.max_entries		= STACK_TRACE_ENTRIES - 1,
30
	.entries		= &stack_dump_trace[0],
S
Steven Rostedt 已提交
31 32
};

33 34
unsigned long stack_trace_max_size;
arch_spinlock_t max_stack_lock =
35
	(arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
S
Steven Rostedt 已提交
36 37

static DEFINE_PER_CPU(int, trace_active);
38 39 40 41
static DEFINE_MUTEX(stack_sysctl_mutex);

int stack_tracer_enabled;
static int last_stack_tracer_enabled;
S
Steven Rostedt 已提交
42

43
void stack_trace_print(void)
44 45 46 47 48 49
{
	long i;
	int size;

	pr_emerg("        Depth    Size   Location    (%d entries)\n"
			   "        -----    ----   --------\n",
50
			   stack_trace_max.nr_entries);
51

52
	for (i = 0; i < stack_trace_max.nr_entries; i++) {
53 54
		if (stack_dump_trace[i] == ULONG_MAX)
			break;
55
		if (i+1 == stack_trace_max.nr_entries ||
56
				stack_dump_trace[i+1] == ULONG_MAX)
57
			size = stack_trace_index[i];
58
		else
59
			size = stack_trace_index[i] - stack_trace_index[i+1];
60

61
		pr_emerg("%3ld) %8d   %5d   %pS\n", i, stack_trace_index[i],
62 63 64 65
				size, (void *)stack_dump_trace[i]);
	}
}

66 67 68 69 70 71 72 73 74
/*
 * When arch-specific code overides this function, the following
 * data should be filled up, assuming max_stack_lock is held to
 * prevent concurrent updates.
 *     stack_trace_index[]
 *     stack_trace_max
 *     stack_trace_max_size
 */
void __weak
75
check_stack(unsigned long ip, unsigned long *stack)
S
Steven Rostedt 已提交
76
{
77
	unsigned long this_size, flags; unsigned long *p, *top, *start;
78 79
	static int tracer_frame;
	int frame_size = ACCESS_ONCE(tracer_frame);
80
	int i, x;
S
Steven Rostedt 已提交
81

82
	this_size = ((unsigned long)stack) & (THREAD_SIZE-1);
S
Steven Rostedt 已提交
83
	this_size = THREAD_SIZE - this_size;
84 85
	/* Remove the frame of the tracer */
	this_size -= frame_size;
S
Steven Rostedt 已提交
86

87
	if (this_size <= stack_trace_max_size)
S
Steven Rostedt 已提交
88 89
		return;

90
	/* we do not handle interrupt stacks yet */
91
	if (!object_is_on_stack(stack))
92 93
		return;

94
	local_irq_save(flags);
95
	arch_spin_lock(&max_stack_lock);
S
Steven Rostedt 已提交
96

97 98 99 100
	/* In case another CPU set the tracer_frame on us */
	if (unlikely(!frame_size))
		this_size -= tracer_frame;

S
Steven Rostedt 已提交
101
	/* a race could have already updated it */
102
	if (this_size <= stack_trace_max_size)
S
Steven Rostedt 已提交
103 104
		goto out;

105
	stack_trace_max_size = this_size;
S
Steven Rostedt 已提交
106

107 108
	stack_trace_max.nr_entries = 0;
	stack_trace_max.skip = 3;
S
Steven Rostedt 已提交
109

110
	save_stack_trace(&stack_trace_max);
S
Steven Rostedt 已提交
111

112
	/* Skip over the overhead of the stack tracer itself */
113
	for (i = 0; i < stack_trace_max.nr_entries; i++) {
114 115 116
		if (stack_dump_trace[i] == ip)
			break;
	}
117

S
Steven Rostedt 已提交
118 119 120
	/*
	 * Now find where in the stack these are.
	 */
121
	x = 0;
122
	start = stack;
S
Steven Rostedt 已提交
123 124 125 126 127 128 129 130 131 132
	top = (unsigned long *)
		(((unsigned long)start & ~(THREAD_SIZE-1)) + THREAD_SIZE);

	/*
	 * Loop through all the entries. One of the entries may
	 * for some reason be missed on the stack, so we may
	 * have to account for them. If they are all there, this
	 * loop will only happen once. This code only takes place
	 * on a new max, so it is far from a fast path.
	 */
133
	while (i < stack_trace_max.nr_entries) {
S
Steven Rostedt 已提交
134
		int found = 0;
S
Steven Rostedt 已提交
135

136
		stack_trace_index[x] = this_size;
S
Steven Rostedt 已提交
137 138
		p = start;

139
		for (; p < top && i < stack_trace_max.nr_entries; p++) {
140 141
			if (stack_dump_trace[i] == ULONG_MAX)
				break;
S
Steven Rostedt 已提交
142
			if (*p == stack_dump_trace[i]) {
143
				stack_dump_trace[x] = stack_dump_trace[i++];
144
				this_size = stack_trace_index[x++] =
S
Steven Rostedt 已提交
145
					(top - p) * sizeof(unsigned long);
S
Steven Rostedt 已提交
146
				found = 1;
S
Steven Rostedt 已提交
147 148
				/* Start the search from here */
				start = p + 1;
149 150 151 152 153 154 155
				/*
				 * We do not want to show the overhead
				 * of the stack tracer stack in the
				 * max stack. If we haven't figured
				 * out what that is, then figure it out
				 * now.
				 */
156
				if (unlikely(!tracer_frame)) {
157 158
					tracer_frame = (p - stack) *
						sizeof(unsigned long);
159
					stack_trace_max_size -= tracer_frame;
160
				}
S
Steven Rostedt 已提交
161 162 163
			}
		}

S
Steven Rostedt 已提交
164 165
		if (!found)
			i++;
S
Steven Rostedt 已提交
166 167
	}

168
	stack_trace_max.nr_entries = x;
169 170 171
	for (; x < i; x++)
		stack_dump_trace[x] = ULONG_MAX;

172
	if (task_stack_end_corrupted(current)) {
173
		stack_trace_print();
174 175 176
		BUG();
	}

S
Steven Rostedt 已提交
177
 out:
178
	arch_spin_unlock(&max_stack_lock);
179
	local_irq_restore(flags);
S
Steven Rostedt 已提交
180 181 182
}

static void
183 184
stack_trace_call(unsigned long ip, unsigned long parent_ip,
		 struct ftrace_ops *op, struct pt_regs *pt_regs)
S
Steven Rostedt 已提交
185
{
186
	unsigned long stack;
187
	int cpu;
S
Steven Rostedt 已提交
188

189
	preempt_disable_notrace();
S
Steven Rostedt 已提交
190 191 192 193 194 195

	cpu = raw_smp_processor_id();
	/* no atomic needed, we only modify this variable by this cpu */
	if (per_cpu(trace_active, cpu)++ != 0)
		goto out;

196
	ip += MCOUNT_INSN_SIZE;
197 198

	check_stack(ip, &stack);
S
Steven Rostedt 已提交
199 200 201 202

 out:
	per_cpu(trace_active, cpu)--;
	/* prevent recursion in schedule */
203
	preempt_enable_notrace();
S
Steven Rostedt 已提交
204 205 206 207 208
}

static struct ftrace_ops trace_ops __read_mostly =
{
	.func = stack_trace_call,
209
	.flags = FTRACE_OPS_FL_RECURSION_SAFE,
S
Steven Rostedt 已提交
210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232
};

static ssize_t
stack_max_size_read(struct file *filp, char __user *ubuf,
		    size_t count, loff_t *ppos)
{
	unsigned long *ptr = filp->private_data;
	char buf[64];
	int r;

	r = snprintf(buf, sizeof(buf), "%ld\n", *ptr);
	if (r > sizeof(buf))
		r = sizeof(buf);
	return simple_read_from_buffer(ubuf, count, ppos, buf, r);
}

static ssize_t
stack_max_size_write(struct file *filp, const char __user *ubuf,
		     size_t count, loff_t *ppos)
{
	long *ptr = filp->private_data;
	unsigned long val, flags;
	int ret;
233
	int cpu;
S
Steven Rostedt 已提交
234

235 236
	ret = kstrtoul_from_user(ubuf, count, 10, &val);
	if (ret)
S
Steven Rostedt 已提交
237 238
		return ret;

239
	local_irq_save(flags);
240 241 242 243 244 245 246 247 248

	/*
	 * In case we trace inside arch_spin_lock() or after (NMI),
	 * we will cause circular lock, so we also need to increase
	 * the percpu trace_active here.
	 */
	cpu = smp_processor_id();
	per_cpu(trace_active, cpu)++;

249
	arch_spin_lock(&max_stack_lock);
S
Steven Rostedt 已提交
250
	*ptr = val;
251
	arch_spin_unlock(&max_stack_lock);
252 253

	per_cpu(trace_active, cpu)--;
254
	local_irq_restore(flags);
S
Steven Rostedt 已提交
255 256 257 258

	return count;
}

259
static const struct file_operations stack_max_size_fops = {
S
Steven Rostedt 已提交
260 261 262
	.open		= tracing_open_generic,
	.read		= stack_max_size_read,
	.write		= stack_max_size_write,
263
	.llseek		= default_llseek,
S
Steven Rostedt 已提交
264 265 266
};

static void *
L
Li Zefan 已提交
267
__next(struct seq_file *m, loff_t *pos)
S
Steven Rostedt 已提交
268
{
L
Li Zefan 已提交
269
	long n = *pos - 1;
S
Steven Rostedt 已提交
270

271
	if (n > stack_trace_max.nr_entries || stack_dump_trace[n] == ULONG_MAX)
S
Steven Rostedt 已提交
272 273
		return NULL;

L
Li Zefan 已提交
274
	m->private = (void *)n;
S
Steven Rostedt 已提交
275
	return &m->private;
S
Steven Rostedt 已提交
276 277
}

L
Li Zefan 已提交
278 279
static void *
t_next(struct seq_file *m, void *v, loff_t *pos)
S
Steven Rostedt 已提交
280
{
L
Li Zefan 已提交
281 282 283
	(*pos)++;
	return __next(m, pos);
}
S
Steven Rostedt 已提交
284

L
Li Zefan 已提交
285 286
static void *t_start(struct seq_file *m, loff_t *pos)
{
287 288
	int cpu;

S
Steven Rostedt 已提交
289
	local_irq_disable();
290 291 292 293

	cpu = smp_processor_id();
	per_cpu(trace_active, cpu)++;

294
	arch_spin_lock(&max_stack_lock);
S
Steven Rostedt 已提交
295

296 297 298
	if (*pos == 0)
		return SEQ_START_TOKEN;

L
Li Zefan 已提交
299
	return __next(m, pos);
S
Steven Rostedt 已提交
300 301 302 303
}

static void t_stop(struct seq_file *m, void *p)
{
304 305
	int cpu;

306
	arch_spin_unlock(&max_stack_lock);
307 308 309 310

	cpu = smp_processor_id();
	per_cpu(trace_active, cpu)--;

S
Steven Rostedt 已提交
311 312 313
	local_irq_enable();
}

314
static void trace_lookup_stack(struct seq_file *m, long i)
S
Steven Rostedt 已提交
315
{
S
Steven Rostedt 已提交
316
	unsigned long addr = stack_dump_trace[i];
S
Steven Rostedt 已提交
317

318
	seq_printf(m, "%pS\n", (void *)addr);
S
Steven Rostedt 已提交
319 320
}

321 322 323 324 325 326 327 328 329 330 331
static void print_disabled(struct seq_file *m)
{
	seq_puts(m, "#\n"
		 "#  Stack tracer disabled\n"
		 "#\n"
		 "# To enable the stack tracer, either add 'stacktrace' to the\n"
		 "# kernel command line\n"
		 "# or 'echo 1 > /proc/sys/kernel/stack_tracer_enabled'\n"
		 "#\n");
}

S
Steven Rostedt 已提交
332 333
static int t_show(struct seq_file *m, void *v)
{
334
	long i;
S
Steven Rostedt 已提交
335 336
	int size;

337
	if (v == SEQ_START_TOKEN) {
338
		seq_printf(m, "        Depth    Size   Location"
S
Steven Rostedt 已提交
339
			   "    (%d entries)\n"
340
			   "        -----    ----   --------\n",
341
			   stack_trace_max.nr_entries);
342

343
		if (!stack_tracer_enabled && !stack_trace_max_size)
344 345
			print_disabled(m);

S
Steven Rostedt 已提交
346 347
		return 0;
	}
S
Steven Rostedt 已提交
348

349 350
	i = *(long *)v;

351
	if (i >= stack_trace_max.nr_entries ||
S
Steven Rostedt 已提交
352
	    stack_dump_trace[i] == ULONG_MAX)
S
Steven Rostedt 已提交
353 354
		return 0;

355
	if (i+1 == stack_trace_max.nr_entries ||
S
Steven Rostedt 已提交
356
	    stack_dump_trace[i+1] == ULONG_MAX)
357
		size = stack_trace_index[i];
S
Steven Rostedt 已提交
358
	else
359
		size = stack_trace_index[i] - stack_trace_index[i+1];
S
Steven Rostedt 已提交
360

361
	seq_printf(m, "%3ld) %8d   %5d   ", i, stack_trace_index[i], size);
S
Steven Rostedt 已提交
362 363

	trace_lookup_stack(m, i);
S
Steven Rostedt 已提交
364 365 366 367

	return 0;
}

368
static const struct seq_operations stack_trace_seq_ops = {
S
Steven Rostedt 已提交
369 370 371 372 373 374 375 376
	.start		= t_start,
	.next		= t_next,
	.stop		= t_stop,
	.show		= t_show,
};

static int stack_trace_open(struct inode *inode, struct file *file)
{
L
Li Zefan 已提交
377
	return seq_open(file, &stack_trace_seq_ops);
S
Steven Rostedt 已提交
378 379
}

380
static const struct file_operations stack_trace_fops = {
S
Steven Rostedt 已提交
381 382 383
	.open		= stack_trace_open,
	.read		= seq_read,
	.llseek		= seq_lseek,
L
Li Zefan 已提交
384
	.release	= seq_release,
S
Steven Rostedt 已提交
385 386
};

387 388 389 390 391 392 393 394 395 396 397
static int
stack_trace_filter_open(struct inode *inode, struct file *file)
{
	return ftrace_regex_open(&trace_ops, FTRACE_ITER_FILTER,
				 inode, file);
}

static const struct file_operations stack_trace_filter_fops = {
	.open = stack_trace_filter_open,
	.read = seq_read,
	.write = ftrace_filter_write,
398
	.llseek = tracing_lseek,
399 400 401
	.release = ftrace_regex_release,
};

402 403
int
stack_trace_sysctl(struct ctl_table *table, int write,
404
		   void __user *buffer, size_t *lenp,
405 406 407 408 409 410
		   loff_t *ppos)
{
	int ret;

	mutex_lock(&stack_sysctl_mutex);

411
	ret = proc_dointvec(table, write, buffer, lenp, ppos);
412 413

	if (ret || !write ||
414
	    (last_stack_tracer_enabled == !!stack_tracer_enabled))
415 416
		goto out;

417
	last_stack_tracer_enabled = !!stack_tracer_enabled;
418 419 420 421 422 423 424 425 426 427 428

	if (stack_tracer_enabled)
		register_ftrace_function(&trace_ops);
	else
		unregister_ftrace_function(&trace_ops);

 out:
	mutex_unlock(&stack_sysctl_mutex);
	return ret;
}

429 430
static char stack_trace_filter_buf[COMMAND_LINE_SIZE+1] __initdata;

431 432
static __init int enable_stacktrace(char *str)
{
433 434 435
	if (strncmp(str, "_filter=", 8) == 0)
		strncpy(stack_trace_filter_buf, str+8, COMMAND_LINE_SIZE);

436 437
	stack_tracer_enabled = 1;
	last_stack_tracer_enabled = 1;
438 439 440 441
	return 1;
}
__setup("stacktrace", enable_stacktrace);

S
Steven Rostedt 已提交
442 443 444 445 446
static __init int stack_trace_init(void)
{
	struct dentry *d_tracer;

	d_tracer = tracing_init_dentry();
447
	if (IS_ERR(d_tracer))
448
		return 0;
S
Steven Rostedt 已提交
449

450
	trace_create_file("stack_max_size", 0644, d_tracer,
451
			&stack_trace_max_size, &stack_max_size_fops);
S
Steven Rostedt 已提交
452

453 454
	trace_create_file("stack_trace", 0444, d_tracer,
			NULL, &stack_trace_fops);
S
Steven Rostedt 已提交
455

456 457 458
	trace_create_file("stack_trace_filter", 0444, d_tracer,
			NULL, &stack_trace_filter_fops);

459 460 461
	if (stack_trace_filter_buf[0])
		ftrace_set_early_filter(&trace_ops, stack_trace_filter_buf, 1);

462
	if (stack_tracer_enabled)
463
		register_ftrace_function(&trace_ops);
S
Steven Rostedt 已提交
464 465 466 467 468

	return 0;
}

device_initcall(stack_trace_init);