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 /* Can't do this from NMI context (can cause deadlocks) */
92 local_irq_save(flags);
93 arch_spin_lock(&max_stack_lock);
96 * RCU may not be watching, make it see us.
97 * The stack trace code uses rcu_sched.
101 /* In case another CPU set the tracer_frame on us */
102 if (unlikely(!frame_size))
103 this_size -= tracer_frame;
105 /* a race could have already updated it */
106 if (this_size <= max_stack_size)
109 max_stack_size = this_size;
111 max_stack_trace.nr_entries = 0;
112 max_stack_trace.skip = 3;
114 save_stack_trace(&max_stack_trace);
116 /* Skip over the overhead of the stack tracer itself */
117 for (i = 0; i < max_stack_trace.nr_entries; i++) {
118 if (stack_dump_trace[i] == ip)
123 * Now find where in the stack these are.
127 top = (unsigned long *)
128 (((unsigned long)start & ~(THREAD_SIZE-1)) + THREAD_SIZE);
131 * Loop through all the entries. One of the entries may
132 * for some reason be missed on the stack, so we may
133 * have to account for them. If they are all there, this
134 * loop will only happen once. This code only takes place
135 * on a new max, so it is far from a fast path.
137 while (i < max_stack_trace.nr_entries) {
140 stack_dump_index[x] = this_size;
143 for (; p < top && i < max_stack_trace.nr_entries; p++) {
144 if (stack_dump_trace[i] == ULONG_MAX)
146 if (*p == stack_dump_trace[i]) {
147 stack_dump_trace[x] = stack_dump_trace[i++];
148 this_size = stack_dump_index[x++] =
149 (top - p) * sizeof(unsigned long);
151 /* Start the search from here */
154 * We do not want to show the overhead
155 * of the stack tracer stack in the
156 * max stack. If we haven't figured
157 * out what that is, then figure it out
160 if (unlikely(!tracer_frame)) {
161 tracer_frame = (p - stack) *
162 sizeof(unsigned long);
163 max_stack_size -= tracer_frame;
172 max_stack_trace.nr_entries = x;
174 stack_dump_trace[x] = ULONG_MAX;
176 if (task_stack_end_corrupted(current)) {
183 arch_spin_unlock(&max_stack_lock);
184 local_irq_restore(flags);
188 stack_trace_call(unsigned long ip, unsigned long parent_ip,
189 struct ftrace_ops *op, struct pt_regs *pt_regs)
194 preempt_disable_notrace();
196 cpu = raw_smp_processor_id();
197 /* no atomic needed, we only modify this variable by this cpu */
198 if (per_cpu(trace_active, cpu)++ != 0)
201 ip += MCOUNT_INSN_SIZE;
203 check_stack(ip, &stack);
206 per_cpu(trace_active, cpu)--;
207 /* prevent recursion in schedule */
208 preempt_enable_notrace();
211 static struct ftrace_ops trace_ops __read_mostly =
213 .func = stack_trace_call,
214 .flags = FTRACE_OPS_FL_RECURSION_SAFE,
218 stack_max_size_read(struct file *filp, char __user *ubuf,
219 size_t count, loff_t *ppos)
221 unsigned long *ptr = filp->private_data;
225 r = snprintf(buf, sizeof(buf), "%ld\n", *ptr);
228 return simple_read_from_buffer(ubuf, count, ppos, buf, r);
232 stack_max_size_write(struct file *filp, const char __user *ubuf,
233 size_t count, loff_t *ppos)
235 long *ptr = filp->private_data;
236 unsigned long val, flags;
240 ret = kstrtoul_from_user(ubuf, count, 10, &val);
244 local_irq_save(flags);
247 * In case we trace inside arch_spin_lock() or after (NMI),
248 * we will cause circular lock, so we also need to increase
249 * the percpu trace_active here.
251 cpu = smp_processor_id();
252 per_cpu(trace_active, cpu)++;
254 arch_spin_lock(&max_stack_lock);
256 arch_spin_unlock(&max_stack_lock);
258 per_cpu(trace_active, cpu)--;
259 local_irq_restore(flags);
264 static const struct file_operations stack_max_size_fops = {
265 .open = tracing_open_generic,
266 .read = stack_max_size_read,
267 .write = stack_max_size_write,
268 .llseek = default_llseek,
272 __next(struct seq_file *m, loff_t *pos)
276 if (n > max_stack_trace.nr_entries || stack_dump_trace[n] == ULONG_MAX)
279 m->private = (void *)n;
284 t_next(struct seq_file *m, void *v, loff_t *pos)
287 return __next(m, pos);
290 static void *t_start(struct seq_file *m, loff_t *pos)
296 cpu = smp_processor_id();
297 per_cpu(trace_active, cpu)++;
299 arch_spin_lock(&max_stack_lock);
302 return SEQ_START_TOKEN;
304 return __next(m, pos);
307 static void t_stop(struct seq_file *m, void *p)
311 arch_spin_unlock(&max_stack_lock);
313 cpu = smp_processor_id();
314 per_cpu(trace_active, cpu)--;
319 static void trace_lookup_stack(struct seq_file *m, long i)
321 unsigned long addr = stack_dump_trace[i];
323 seq_printf(m, "%pS\n", (void *)addr);
326 static void print_disabled(struct seq_file *m)
329 "# Stack tracer disabled\n"
331 "# To enable the stack tracer, either add 'stacktrace' to the\n"
332 "# kernel command line\n"
333 "# or 'echo 1 > /proc/sys/kernel/stack_tracer_enabled'\n"
337 static int t_show(struct seq_file *m, void *v)
342 if (v == SEQ_START_TOKEN) {
343 seq_printf(m, " Depth Size Location"
345 " ----- ---- --------\n",
346 max_stack_trace.nr_entries);
348 if (!stack_tracer_enabled && !max_stack_size)
356 if (i >= max_stack_trace.nr_entries ||
357 stack_dump_trace[i] == ULONG_MAX)
360 if (i+1 == max_stack_trace.nr_entries ||
361 stack_dump_trace[i+1] == ULONG_MAX)
362 size = stack_dump_index[i];
364 size = stack_dump_index[i] - stack_dump_index[i+1];
366 seq_printf(m, "%3ld) %8d %5d ", i, stack_dump_index[i], size);
368 trace_lookup_stack(m, i);
373 static const struct seq_operations stack_trace_seq_ops = {
380 static int stack_trace_open(struct inode *inode, struct file *file)
382 return seq_open(file, &stack_trace_seq_ops);
385 static const struct file_operations stack_trace_fops = {
386 .open = stack_trace_open,
389 .release = seq_release,
393 stack_trace_filter_open(struct inode *inode, struct file *file)
395 return ftrace_regex_open(&trace_ops, FTRACE_ITER_FILTER,
399 static const struct file_operations stack_trace_filter_fops = {
400 .open = stack_trace_filter_open,
402 .write = ftrace_filter_write,
403 .llseek = tracing_lseek,
404 .release = ftrace_regex_release,
408 stack_trace_sysctl(struct ctl_table *table, int write,
409 void __user *buffer, size_t *lenp,
414 mutex_lock(&stack_sysctl_mutex);
416 ret = proc_dointvec(table, write, buffer, lenp, ppos);
419 (last_stack_tracer_enabled == !!stack_tracer_enabled))
422 last_stack_tracer_enabled = !!stack_tracer_enabled;
424 if (stack_tracer_enabled)
425 register_ftrace_function(&trace_ops);
427 unregister_ftrace_function(&trace_ops);
430 mutex_unlock(&stack_sysctl_mutex);
434 static char stack_trace_filter_buf[COMMAND_LINE_SIZE+1] __initdata;
436 static __init int enable_stacktrace(char *str)
438 if (strncmp(str, "_filter=", 8) == 0)
439 strncpy(stack_trace_filter_buf, str+8, COMMAND_LINE_SIZE);
441 stack_tracer_enabled = 1;
442 last_stack_tracer_enabled = 1;
445 __setup("stacktrace", enable_stacktrace);
447 static __init int stack_trace_init(void)
449 struct dentry *d_tracer;
451 d_tracer = tracing_init_dentry();
452 if (IS_ERR(d_tracer))
455 trace_create_file("stack_max_size", 0644, d_tracer,
456 &max_stack_size, &stack_max_size_fops);
458 trace_create_file("stack_trace", 0444, d_tracer,
459 NULL, &stack_trace_fops);
461 trace_create_file("stack_trace_filter", 0444, d_tracer,
462 NULL, &stack_trace_filter_fops);
464 if (stack_trace_filter_buf[0])
465 ftrace_set_early_filter(&trace_ops, stack_trace_filter_buf, 1);
467 if (stack_tracer_enabled)
468 register_ftrace_function(&trace_ops);
473 device_initcall(stack_trace_init);