trace_stack.c 10.7 KB
Newer Older
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>
13
#include <linux/init.h>
14 15 16

#include <asm/setup.h>

17 18 19 20
#include "trace.h"

#define STACK_TRACE_ENTRIES 500

21
#ifdef CC_USING_FENTRY
22
# define fentry		1
23
#else
24
# define fentry		0
25 26
#endif

27 28 29 30
static unsigned long stack_dump_trace[STACK_TRACE_ENTRIES+1] =
	 { [0 ... (STACK_TRACE_ENTRIES)] = ULONG_MAX };
static unsigned stack_dump_index[STACK_TRACE_ENTRIES];

31 32 33 34 35
/*
 * 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.
 */
36
static struct stack_trace max_stack_trace = {
37 38
	.max_entries		= STACK_TRACE_ENTRIES - 1,
	.entries		= &stack_dump_trace[1],
39 40 41
};

static unsigned long max_stack_size;
42
static arch_spinlock_t max_stack_lock =
43
	(arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
44 45

static DEFINE_PER_CPU(int, trace_active);
46 47 48 49
static DEFINE_MUTEX(stack_sysctl_mutex);

int stack_tracer_enabled;
static int last_stack_tracer_enabled;
50

51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73
static inline void print_max_stack(void)
{
	long i;
	int size;

	pr_emerg("        Depth    Size   Location    (%d entries)\n"
			   "        -----    ----   --------\n",
			   max_stack_trace.nr_entries - 1);

	for (i = 0; i < max_stack_trace.nr_entries; i++) {
		if (stack_dump_trace[i] == ULONG_MAX)
			break;
		if (i+1 == max_stack_trace.nr_entries ||
				stack_dump_trace[i+1] == ULONG_MAX)
			size = stack_dump_index[i];
		else
			size = stack_dump_index[i] - stack_dump_index[i+1];

		pr_emerg("%3ld) %8d   %5d   %pS\n", i, stack_dump_index[i],
				size, (void *)stack_dump_trace[i]);
	}
}

74
static inline void
75
check_stack(unsigned long ip, unsigned long *stack)
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;
81

82
	this_size = ((unsigned long)stack) & (THREAD_SIZE-1);
83
	this_size = THREAD_SIZE - this_size;
84 85
	/* Remove the frame of the tracer */
	this_size -= frame_size;
86 87 88 89

	if (this_size <= max_stack_size)
		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);
96

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

101 102 103 104 105 106
	/* a race could have already updated it */
	if (this_size <= max_stack_size)
		goto out;

	max_stack_size = this_size;

107 108 109 110 111 112
	max_stack_trace.nr_entries = 0;

	if (using_ftrace_ops_list_func())
		max_stack_trace.skip = 4;
	else
		max_stack_trace.skip = 3;
113 114 115

	save_stack_trace(&max_stack_trace);

116
	/*
117 118 119
	 * Add the passed in ip from the function tracer.
	 * Searching for this on the stack will skip over
	 * most of the overhead from the stack tracer itself.
120
	 */
121 122
	stack_dump_trace[0] = ip;
	max_stack_trace.nr_entries++;
123

124 125 126 127
	/*
	 * Now find where in the stack these are.
	 */
	i = 0;
128
	start = stack;
129 130 131 132 133 134 135 136 137 138 139
	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.
	 */
	while (i < max_stack_trace.nr_entries) {
140
		int found = 0;
141 142 143 144 145 146 147 148

		stack_dump_index[i] = this_size;
		p = start;

		for (; p < top && i < max_stack_trace.nr_entries; p++) {
			if (*p == stack_dump_trace[i]) {
				this_size = stack_dump_index[i++] =
					(top - p) * sizeof(unsigned long);
149
				found = 1;
150 151
				/* Start the search from here */
				start = p + 1;
152 153 154 155 156 157 158 159 160 161 162 163
				/*
				 * 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.
				 */
				if (unlikely(!tracer_frame) && i == 1) {
					tracer_frame = (p - stack) *
						sizeof(unsigned long);
					max_stack_size -= tracer_frame;
				}
164 165 166
			}
		}

167 168
		if (!found)
			i++;
169 170
	}

171
	if (task_stack_end_corrupted(current)) {
172 173 174 175
		print_max_stack();
		BUG();
	}

176
 out:
177
	arch_spin_unlock(&max_stack_lock);
178
	local_irq_restore(flags);
179 180 181
}

static void
182 183
stack_trace_call(unsigned long ip, unsigned long parent_ip,
		 struct ftrace_ops *op, struct pt_regs *pt_regs)
184
{
185
	unsigned long stack;
186
	int cpu;
187

188
	preempt_disable_notrace();
189 190 191 192 193 194

	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;

195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214
	/*
	 * When fentry is used, the traced function does not get
	 * its stack frame set up, and we lose the parent.
	 * The ip is pretty useless because the function tracer
	 * was called before that function set up its stack frame.
	 * In this case, we use the parent ip.
	 *
	 * By adding the return address of either the parent ip
	 * or the current ip we can disregard most of the stack usage
	 * caused by the stack tracer itself.
	 *
	 * The function tracer always reports the address of where the
	 * mcount call was, but the stack will hold the return address.
	 */
	if (fentry)
		ip = parent_ip;
	else
		ip += MCOUNT_INSN_SIZE;

	check_stack(ip, &stack);
215 216 217 218

 out:
	per_cpu(trace_active, cpu)--;
	/* prevent recursion in schedule */
219
	preempt_enable_notrace();
220 221 222 223 224
}

static struct ftrace_ops trace_ops __read_mostly =
{
	.func = stack_trace_call,
225
	.flags = FTRACE_OPS_FL_RECURSION_SAFE,
226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248
};

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;
249
	int cpu;
250

251 252
	ret = kstrtoul_from_user(ubuf, count, 10, &val);
	if (ret)
253 254
		return ret;

255
	local_irq_save(flags);
256 257 258 259 260 261 262 263 264

	/*
	 * 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)++;

265
	arch_spin_lock(&max_stack_lock);
266
	*ptr = val;
267
	arch_spin_unlock(&max_stack_lock);
268 269

	per_cpu(trace_active, cpu)--;
270
	local_irq_restore(flags);
271 272 273 274

	return count;
}

275
static const struct file_operations stack_max_size_fops = {
276 277 278
	.open		= tracing_open_generic,
	.read		= stack_max_size_read,
	.write		= stack_max_size_write,
279
	.llseek		= default_llseek,
280 281 282
};

static void *
283
__next(struct seq_file *m, loff_t *pos)
284
{
285
	long n = *pos - 1;
286

287
	if (n >= max_stack_trace.nr_entries || stack_dump_trace[n] == ULONG_MAX)
288 289
		return NULL;

290
	m->private = (void *)n;
291
	return &m->private;
292 293
}

294 295
static void *
t_next(struct seq_file *m, void *v, loff_t *pos)
296
{
297 298 299
	(*pos)++;
	return __next(m, pos);
}
300

301 302
static void *t_start(struct seq_file *m, loff_t *pos)
{
303 304
	int cpu;

305
	local_irq_disable();
306 307 308 309

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

310
	arch_spin_lock(&max_stack_lock);
311

312 313 314
	if (*pos == 0)
		return SEQ_START_TOKEN;

315
	return __next(m, pos);
316 317 318 319
}

static void t_stop(struct seq_file *m, void *p)
{
320 321
	int cpu;

322
	arch_spin_unlock(&max_stack_lock);
323 324 325 326

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

327 328 329
	local_irq_enable();
}

330
static void trace_lookup_stack(struct seq_file *m, long i)
331
{
332
	unsigned long addr = stack_dump_trace[i];
333

334
	seq_printf(m, "%pS\n", (void *)addr);
335 336
}

337 338 339 340 341 342 343 344 345 346 347
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");
}

348 349
static int t_show(struct seq_file *m, void *v)
{
350
	long i;
351 352
	int size;

353
	if (v == SEQ_START_TOKEN) {
354
		seq_printf(m, "        Depth    Size   Location"
355
			   "    (%d entries)\n"
356
			   "        -----    ----   --------\n",
357
			   max_stack_trace.nr_entries - 1);
358 359 360 361

		if (!stack_tracer_enabled && !max_stack_size)
			print_disabled(m);

362 363
		return 0;
	}
364

365 366
	i = *(long *)v;

367 368
	if (i >= max_stack_trace.nr_entries ||
	    stack_dump_trace[i] == ULONG_MAX)
369 370
		return 0;

371 372 373 374 375 376 377 378 379
	if (i+1 == max_stack_trace.nr_entries ||
	    stack_dump_trace[i+1] == ULONG_MAX)
		size = stack_dump_index[i];
	else
		size = stack_dump_index[i] - stack_dump_index[i+1];

	seq_printf(m, "%3ld) %8d   %5d   ", i, stack_dump_index[i], size);

	trace_lookup_stack(m, i);
380 381 382 383

	return 0;
}

384
static const struct seq_operations stack_trace_seq_ops = {
385 386 387 388 389 390 391 392
	.start		= t_start,
	.next		= t_next,
	.stop		= t_stop,
	.show		= t_show,
};

static int stack_trace_open(struct inode *inode, struct file *file)
{
393
	return seq_open(file, &stack_trace_seq_ops);
394 395
}

396
static const struct file_operations stack_trace_fops = {
397 398 399
	.open		= stack_trace_open,
	.read		= seq_read,
	.llseek		= seq_lseek,
400
	.release	= seq_release,
401 402
};

403 404 405 406 407 408 409 410 411 412 413
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,
414
	.llseek = tracing_lseek,
415 416 417
	.release = ftrace_regex_release,
};

418 419
int
stack_trace_sysctl(struct ctl_table *table, int write,
420
		   void __user *buffer, size_t *lenp,
421 422 423 424 425 426
		   loff_t *ppos)
{
	int ret;

	mutex_lock(&stack_sysctl_mutex);

427
	ret = proc_dointvec(table, write, buffer, lenp, ppos);
428 429

	if (ret || !write ||
430
	    (last_stack_tracer_enabled == !!stack_tracer_enabled))
431 432
		goto out;

433
	last_stack_tracer_enabled = !!stack_tracer_enabled;
434 435 436 437 438 439 440 441 442 443 444

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

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

445 446
static char stack_trace_filter_buf[COMMAND_LINE_SIZE+1] __initdata;

447 448
static __init int enable_stacktrace(char *str)
{
449 450 451
	if (strncmp(str, "_filter=", 8) == 0)
		strncpy(stack_trace_filter_buf, str+8, COMMAND_LINE_SIZE);

452 453
	stack_tracer_enabled = 1;
	last_stack_tracer_enabled = 1;
454 455 456 457
	return 1;
}
__setup("stacktrace", enable_stacktrace);

458 459 460 461 462
static __init int stack_trace_init(void)
{
	struct dentry *d_tracer;

	d_tracer = tracing_init_dentry();
463
	if (IS_ERR(d_tracer))
464
		return 0;
465

466 467
	trace_create_file("stack_max_size", 0644, d_tracer,
			&max_stack_size, &stack_max_size_fops);
468

469 470
	trace_create_file("stack_trace", 0444, d_tracer,
			NULL, &stack_trace_fops);
471

472 473 474
	trace_create_file("stack_trace_filter", 0444, d_tracer,
			NULL, &stack_trace_filter_fops);

475 476 477
	if (stack_trace_filter_buf[0])
		ftrace_set_early_filter(&trace_ops, stack_trace_filter_buf, 1);

478
	if (stack_tracer_enabled)
479
		register_ftrace_function(&trace_ops);
480 481 482 483 484

	return 0;
}

device_initcall(stack_trace_init);