trace_stack.c 10.8 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12
/*
 * 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/debugfs.h>
#include <linux/ftrace.h>
#include <linux/module.h>
13
#include <linux/sysctl.h>
14 15
#include <linux/init.h>
#include <linux/fs.h>
16 17 18

#include <asm/setup.h>

19 20 21 22
#include "trace.h"

#define STACK_TRACE_ENTRIES 500

23
#ifdef CC_USING_FENTRY
24
# define fentry		1
25
#else
26
# define fentry		0
27 28
#endif

29 30 31 32
static unsigned long stack_dump_trace[STACK_TRACE_ENTRIES+1] =
	 { [0 ... (STACK_TRACE_ENTRIES)] = ULONG_MAX };
static unsigned stack_dump_index[STACK_TRACE_ENTRIES];

33 34 35 36 37
/*
 * 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.
 */
38
static struct stack_trace max_stack_trace = {
39 40
	.max_entries		= STACK_TRACE_ENTRIES - 1,
	.entries		= &stack_dump_trace[1],
41 42 43
};

static unsigned long max_stack_size;
44
static arch_spinlock_t max_stack_lock =
45
	(arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
46 47

static DEFINE_PER_CPU(int, trace_active);
48 49 50 51
static DEFINE_MUTEX(stack_sysctl_mutex);

int stack_tracer_enabled;
static int last_stack_tracer_enabled;
52

53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75
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]);
	}
}

76
static inline void
77
check_stack(unsigned long ip, unsigned long *stack)
78
{
79
	unsigned long this_size, flags; unsigned long *p, *top, *start;
80 81
	static int tracer_frame;
	int frame_size = ACCESS_ONCE(tracer_frame);
82
	int i;
83

84
	this_size = ((unsigned long)stack) & (THREAD_SIZE-1);
85
	this_size = THREAD_SIZE - this_size;
86 87
	/* Remove the frame of the tracer */
	this_size -= frame_size;
88 89 90 91

	if (this_size <= max_stack_size)
		return;

92
	/* we do not handle interrupt stacks yet */
93
	if (!object_is_on_stack(stack))
94 95
		return;

96
	local_irq_save(flags);
97
	arch_spin_lock(&max_stack_lock);
98

99 100 101 102
	/* In case another CPU set the tracer_frame on us */
	if (unlikely(!frame_size))
		this_size -= tracer_frame;

103 104 105 106 107 108
	/* a race could have already updated it */
	if (this_size <= max_stack_size)
		goto out;

	max_stack_size = this_size;

109 110 111 112 113 114
	max_stack_trace.nr_entries = 0;

	if (using_ftrace_ops_list_func())
		max_stack_trace.skip = 4;
	else
		max_stack_trace.skip = 3;
115 116 117

	save_stack_trace(&max_stack_trace);

118
	/*
119 120 121
	 * 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.
122
	 */
123 124
	stack_dump_trace[0] = ip;
	max_stack_trace.nr_entries++;
125

126 127 128 129
	/*
	 * Now find where in the stack these are.
	 */
	i = 0;
130
	start = stack;
131 132 133 134 135 136 137 138 139 140 141
	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) {
142
		int found = 0;
143 144 145 146 147 148 149 150

		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);
151
				found = 1;
152 153
				/* Start the search from here */
				start = p + 1;
154 155 156 157 158 159 160 161 162 163 164 165
				/*
				 * 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;
				}
166 167 168
			}
		}

169 170
		if (!found)
			i++;
171 172
	}

173
	if (*end_of_stack(current) != STACK_END_MAGIC) {
174 175 176 177
		print_max_stack();
		BUG();
	}

178
 out:
179
	arch_spin_unlock(&max_stack_lock);
180
	local_irq_restore(flags);
181 182 183
}

static void
184 185
stack_trace_call(unsigned long ip, unsigned long parent_ip,
		 struct ftrace_ops *op, struct pt_regs *pt_regs)
186
{
187
	unsigned long stack;
188
	int cpu;
189

190
	preempt_disable_notrace();
191 192 193 194 195 196

	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;

197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216
	/*
	 * 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);
217 218 219 220

 out:
	per_cpu(trace_active, cpu)--;
	/* prevent recursion in schedule */
221
	preempt_enable_notrace();
222 223 224 225 226
}

static struct ftrace_ops trace_ops __read_mostly =
{
	.func = stack_trace_call,
227
	.flags = FTRACE_OPS_FL_RECURSION_SAFE,
228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250
};

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;
251
	int cpu;
252

253 254
	ret = kstrtoul_from_user(ubuf, count, 10, &val);
	if (ret)
255 256
		return ret;

257
	local_irq_save(flags);
258 259 260 261 262 263 264 265 266

	/*
	 * 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)++;

267
	arch_spin_lock(&max_stack_lock);
268
	*ptr = val;
269
	arch_spin_unlock(&max_stack_lock);
270 271

	per_cpu(trace_active, cpu)--;
272
	local_irq_restore(flags);
273 274 275 276

	return count;
}

277
static const struct file_operations stack_max_size_fops = {
278 279 280
	.open		= tracing_open_generic,
	.read		= stack_max_size_read,
	.write		= stack_max_size_write,
281
	.llseek		= default_llseek,
282 283 284
};

static void *
285
__next(struct seq_file *m, loff_t *pos)
286
{
287
	long n = *pos - 1;
288

289
	if (n >= max_stack_trace.nr_entries || stack_dump_trace[n] == ULONG_MAX)
290 291
		return NULL;

292
	m->private = (void *)n;
293
	return &m->private;
294 295
}

296 297
static void *
t_next(struct seq_file *m, void *v, loff_t *pos)
298
{
299 300 301
	(*pos)++;
	return __next(m, pos);
}
302

303 304
static void *t_start(struct seq_file *m, loff_t *pos)
{
305 306
	int cpu;

307
	local_irq_disable();
308 309 310 311

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

312
	arch_spin_lock(&max_stack_lock);
313

314 315 316
	if (*pos == 0)
		return SEQ_START_TOKEN;

317
	return __next(m, pos);
318 319 320 321
}

static void t_stop(struct seq_file *m, void *p)
{
322 323
	int cpu;

324
	arch_spin_unlock(&max_stack_lock);
325 326 327 328

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

329 330 331
	local_irq_enable();
}

332
static int trace_lookup_stack(struct seq_file *m, long i)
333
{
334
	unsigned long addr = stack_dump_trace[i];
335

336
	return seq_printf(m, "%pS\n", (void *)addr);
337 338
}

339 340 341 342 343 344 345 346 347 348 349
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");
}

350 351
static int t_show(struct seq_file *m, void *v)
{
352
	long i;
353 354
	int size;

355
	if (v == SEQ_START_TOKEN) {
356
		seq_printf(m, "        Depth    Size   Location"
357
			   "    (%d entries)\n"
358
			   "        -----    ----   --------\n",
359
			   max_stack_trace.nr_entries - 1);
360 361 362 363

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

364 365
		return 0;
	}
366

367 368
	i = *(long *)v;

369 370
	if (i >= max_stack_trace.nr_entries ||
	    stack_dump_trace[i] == ULONG_MAX)
371 372
		return 0;

373 374 375 376 377 378 379 380 381
	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);
382 383 384 385

	return 0;
}

386
static const struct seq_operations stack_trace_seq_ops = {
387 388 389 390 391 392 393 394
	.start		= t_start,
	.next		= t_next,
	.stop		= t_stop,
	.show		= t_show,
};

static int stack_trace_open(struct inode *inode, struct file *file)
{
395
	return seq_open(file, &stack_trace_seq_ops);
396 397
}

398
static const struct file_operations stack_trace_fops = {
399 400 401
	.open		= stack_trace_open,
	.read		= seq_read,
	.llseek		= seq_lseek,
402
	.release	= seq_release,
403 404
};

405 406 407 408 409 410 411 412 413 414 415
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,
416
	.llseek = tracing_lseek,
417 418 419
	.release = ftrace_regex_release,
};

420 421
int
stack_trace_sysctl(struct ctl_table *table, int write,
422
		   void __user *buffer, size_t *lenp,
423 424 425 426 427 428
		   loff_t *ppos)
{
	int ret;

	mutex_lock(&stack_sysctl_mutex);

429
	ret = proc_dointvec(table, write, buffer, lenp, ppos);
430 431

	if (ret || !write ||
432
	    (last_stack_tracer_enabled == !!stack_tracer_enabled))
433 434
		goto out;

435
	last_stack_tracer_enabled = !!stack_tracer_enabled;
436 437 438 439 440 441 442 443 444 445 446

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

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

447 448
static char stack_trace_filter_buf[COMMAND_LINE_SIZE+1] __initdata;

449 450
static __init int enable_stacktrace(char *str)
{
451 452 453
	if (strncmp(str, "_filter=", 8) == 0)
		strncpy(stack_trace_filter_buf, str+8, COMMAND_LINE_SIZE);

454 455
	stack_tracer_enabled = 1;
	last_stack_tracer_enabled = 1;
456 457 458 459
	return 1;
}
__setup("stacktrace", enable_stacktrace);

460 461 462 463 464
static __init int stack_trace_init(void)
{
	struct dentry *d_tracer;

	d_tracer = tracing_init_dentry();
465 466
	if (!d_tracer)
		return 0;
467

468 469
	trace_create_file("stack_max_size", 0644, d_tracer,
			&max_stack_size, &stack_max_size_fops);
470

471 472
	trace_create_file("stack_trace", 0444, d_tracer,
			NULL, &stack_trace_fops);
473

474 475 476
	trace_create_file("stack_trace_filter", 0444, d_tracer,
			NULL, &stack_trace_filter_fops);

477 478 479
	if (stack_trace_filter_buf[0])
		ftrace_set_early_filter(&trace_ops, stack_trace_filter_buf, 1);

480
	if (stack_tracer_enabled)
481
		register_ftrace_function(&trace_ops);
482 483 484 485 486

	return 0;
}

device_initcall(stack_trace_init);