2 * Copyright (C) 2008 Steven Rostedt <srostedt@redhat.com>
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>
10 #include <linux/ftrace.h>
11 #include <linux/module.h>
12 #include <linux/sysctl.h>
13 #include <linux/init.h>
15 #include <asm/setup.h>
19 #define STACK_TRACE_ENTRIES 500
21 static unsigned long stack_dump_trace[STACK_TRACE_ENTRIES+1] =
22 { [0 ... (STACK_TRACE_ENTRIES)] = ULONG_MAX };
23 static unsigned stack_dump_index[STACK_TRACE_ENTRIES];
26 * Reserve one entry for the passed in ip. This will allow
27 * us to remove most or all of the stack size overhead
28 * added by the stack tracer itself.
30 static struct stack_trace max_stack_trace = {
31 .max_entries = STACK_TRACE_ENTRIES - 1,
32 .entries = &stack_dump_trace[0],
35 static unsigned long max_stack_size;
36 static arch_spinlock_t max_stack_lock =
37 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
39 static DEFINE_PER_CPU(int, trace_active);
40 static DEFINE_MUTEX(stack_sysctl_mutex);
42 int stack_tracer_enabled;
43 static int last_stack_tracer_enabled;
45 static inline void print_max_stack(void)
50 pr_emerg(" Depth Size Location (%d entries)\n"
51 " ----- ---- --------\n",
52 max_stack_trace.nr_entries);
54 for (i = 0; i < max_stack_trace.nr_entries; i++) {
55 if (stack_dump_trace[i] == ULONG_MAX)
57 if (i+1 == max_stack_trace.nr_entries ||
58 stack_dump_trace[i+1] == ULONG_MAX)
59 size = stack_dump_index[i];
61 size = stack_dump_index[i] - stack_dump_index[i+1];
63 pr_emerg("%3ld) %8d %5d %pS\n", i, stack_dump_index[i],
64 size, (void *)stack_dump_trace[i]);
69 check_stack(unsigned long ip, unsigned long *stack)
71 unsigned long this_size, flags; unsigned long *p, *top, *start;
72 static int tracer_frame;
73 int frame_size = ACCESS_ONCE(tracer_frame);
76 this_size = ((unsigned long)stack) & (THREAD_SIZE-1);
77 this_size = THREAD_SIZE - this_size;
78 /* Remove the frame of the tracer */
79 this_size -= frame_size;
81 if (this_size <= max_stack_size)
84 /* we do not handle interrupt stacks yet */
85 if (!object_is_on_stack(stack))
88 local_irq_save(flags);
89 arch_spin_lock(&max_stack_lock);
92 * RCU may not be watching, make it see us.
93 * The stack trace code uses rcu_sched.
97 /* In case another CPU set the tracer_frame on us */
98 if (unlikely(!frame_size))
99 this_size -= tracer_frame;
101 /* a race could have already updated it */
102 if (this_size <= max_stack_size)
105 max_stack_size = this_size;
107 max_stack_trace.nr_entries = 0;
108 max_stack_trace.skip = 3;
110 save_stack_trace(&max_stack_trace);
112 /* Skip over the overhead of the stack tracer itself */
113 for (i = 0; i < max_stack_trace.nr_entries; i++) {
114 if (stack_dump_trace[i] == ip)
119 * Now find where in the stack these are.
123 top = (unsigned long *)
124 (((unsigned long)start & ~(THREAD_SIZE-1)) + THREAD_SIZE);
127 * Loop through all the entries. One of the entries may
128 * for some reason be missed on the stack, so we may
129 * have to account for them. If they are all there, this
130 * loop will only happen once. This code only takes place
131 * on a new max, so it is far from a fast path.
133 while (i < max_stack_trace.nr_entries) {
136 stack_dump_index[x] = this_size;
139 for (; p < top && i < max_stack_trace.nr_entries; p++) {
140 if (stack_dump_trace[i] == ULONG_MAX)
142 if (*p == stack_dump_trace[i]) {
143 stack_dump_trace[x] = stack_dump_trace[i++];
144 this_size = stack_dump_index[x++] =
145 (top - p) * sizeof(unsigned long);
147 /* Start the search from here */
150 * We do not want to show the overhead
151 * of the stack tracer stack in the
152 * max stack. If we haven't figured
153 * out what that is, then figure it out
156 if (unlikely(!tracer_frame)) {
157 tracer_frame = (p - stack) *
158 sizeof(unsigned long);
159 max_stack_size -= tracer_frame;
168 max_stack_trace.nr_entries = x;
170 stack_dump_trace[x] = ULONG_MAX;
172 if (task_stack_end_corrupted(current)) {
179 arch_spin_unlock(&max_stack_lock);
180 local_irq_restore(flags);
184 stack_trace_call(unsigned long ip, unsigned long parent_ip,
185 struct ftrace_ops *op, struct pt_regs *pt_regs)
190 preempt_disable_notrace();
192 cpu = raw_smp_processor_id();
193 /* no atomic needed, we only modify this variable by this cpu */
194 if (per_cpu(trace_active, cpu)++ != 0)
197 ip += MCOUNT_INSN_SIZE;
199 check_stack(ip, &stack);
202 per_cpu(trace_active, cpu)--;
203 /* prevent recursion in schedule */
204 preempt_enable_notrace();
207 static struct ftrace_ops trace_ops __read_mostly =
209 .func = stack_trace_call,
210 .flags = FTRACE_OPS_FL_RECURSION_SAFE,
214 stack_max_size_read(struct file *filp, char __user *ubuf,
215 size_t count, loff_t *ppos)
217 unsigned long *ptr = filp->private_data;
221 r = snprintf(buf, sizeof(buf), "%ld\n", *ptr);
224 return simple_read_from_buffer(ubuf, count, ppos, buf, r);
228 stack_max_size_write(struct file *filp, const char __user *ubuf,
229 size_t count, loff_t *ppos)
231 long *ptr = filp->private_data;
232 unsigned long val, flags;
236 ret = kstrtoul_from_user(ubuf, count, 10, &val);
240 local_irq_save(flags);
243 * In case we trace inside arch_spin_lock() or after (NMI),
244 * we will cause circular lock, so we also need to increase
245 * the percpu trace_active here.
247 cpu = smp_processor_id();
248 per_cpu(trace_active, cpu)++;
250 arch_spin_lock(&max_stack_lock);
252 arch_spin_unlock(&max_stack_lock);
254 per_cpu(trace_active, cpu)--;
255 local_irq_restore(flags);
260 static const struct file_operations stack_max_size_fops = {
261 .open = tracing_open_generic,
262 .read = stack_max_size_read,
263 .write = stack_max_size_write,
264 .llseek = default_llseek,
268 __next(struct seq_file *m, loff_t *pos)
272 if (n > max_stack_trace.nr_entries || stack_dump_trace[n] == ULONG_MAX)
275 m->private = (void *)n;
280 t_next(struct seq_file *m, void *v, loff_t *pos)
283 return __next(m, pos);
286 static void *t_start(struct seq_file *m, loff_t *pos)
292 cpu = smp_processor_id();
293 per_cpu(trace_active, cpu)++;
295 arch_spin_lock(&max_stack_lock);
298 return SEQ_START_TOKEN;
300 return __next(m, pos);
303 static void t_stop(struct seq_file *m, void *p)
307 arch_spin_unlock(&max_stack_lock);
309 cpu = smp_processor_id();
310 per_cpu(trace_active, cpu)--;
315 static void trace_lookup_stack(struct seq_file *m, long i)
317 unsigned long addr = stack_dump_trace[i];
319 seq_printf(m, "%pS\n", (void *)addr);
322 static void print_disabled(struct seq_file *m)
325 "# Stack tracer disabled\n"
327 "# To enable the stack tracer, either add 'stacktrace' to the\n"
328 "# kernel command line\n"
329 "# or 'echo 1 > /proc/sys/kernel/stack_tracer_enabled'\n"
333 static int t_show(struct seq_file *m, void *v)
338 if (v == SEQ_START_TOKEN) {
339 seq_printf(m, " Depth Size Location"
341 " ----- ---- --------\n",
342 max_stack_trace.nr_entries);
344 if (!stack_tracer_enabled && !max_stack_size)
352 if (i >= max_stack_trace.nr_entries ||
353 stack_dump_trace[i] == ULONG_MAX)
356 if (i+1 == max_stack_trace.nr_entries ||
357 stack_dump_trace[i+1] == ULONG_MAX)
358 size = stack_dump_index[i];
360 size = stack_dump_index[i] - stack_dump_index[i+1];
362 seq_printf(m, "%3ld) %8d %5d ", i, stack_dump_index[i], size);
364 trace_lookup_stack(m, i);
369 static const struct seq_operations stack_trace_seq_ops = {
376 static int stack_trace_open(struct inode *inode, struct file *file)
378 return seq_open(file, &stack_trace_seq_ops);
381 static const struct file_operations stack_trace_fops = {
382 .open = stack_trace_open,
385 .release = seq_release,
389 stack_trace_filter_open(struct inode *inode, struct file *file)
391 return ftrace_regex_open(&trace_ops, FTRACE_ITER_FILTER,
395 static const struct file_operations stack_trace_filter_fops = {
396 .open = stack_trace_filter_open,
398 .write = ftrace_filter_write,
399 .llseek = tracing_lseek,
400 .release = ftrace_regex_release,
404 stack_trace_sysctl(struct ctl_table *table, int write,
405 void __user *buffer, size_t *lenp,
410 mutex_lock(&stack_sysctl_mutex);
412 ret = proc_dointvec(table, write, buffer, lenp, ppos);
415 (last_stack_tracer_enabled == !!stack_tracer_enabled))
418 last_stack_tracer_enabled = !!stack_tracer_enabled;
420 if (stack_tracer_enabled)
421 register_ftrace_function(&trace_ops);
423 unregister_ftrace_function(&trace_ops);
426 mutex_unlock(&stack_sysctl_mutex);
430 static char stack_trace_filter_buf[COMMAND_LINE_SIZE+1] __initdata;
432 static __init int enable_stacktrace(char *str)
434 if (strncmp(str, "_filter=", 8) == 0)
435 strncpy(stack_trace_filter_buf, str+8, COMMAND_LINE_SIZE);
437 stack_tracer_enabled = 1;
438 last_stack_tracer_enabled = 1;
441 __setup("stacktrace", enable_stacktrace);
443 static __init int stack_trace_init(void)
445 struct dentry *d_tracer;
447 d_tracer = tracing_init_dentry();
448 if (IS_ERR(d_tracer))
451 trace_create_file("stack_max_size", 0644, d_tracer,
452 &max_stack_size, &stack_max_size_fops);
454 trace_create_file("stack_trace", 0444, d_tracer,
455 NULL, &stack_trace_fops);
457 trace_create_file("stack_trace_filter", 0444, d_tracer,
458 NULL, &stack_trace_filter_fops);
460 if (stack_trace_filter_buf[0])
461 ftrace_set_early_filter(&trace_ops, stack_trace_filter_buf, 1);
463 if (stack_tracer_enabled)
464 register_ftrace_function(&trace_ops);
469 device_initcall(stack_trace_init);