xref: /linux/kernel/trace/trace_stack.c (revision 61a6976bf19a6cf5dfcf37c3536665b316f22d49)
1 /*
2  * Copyright (C) 2008 Steven Rostedt <srostedt@redhat.com>
3  *
4  */
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/debugfs.h>
11 #include <linux/ftrace.h>
12 #include <linux/module.h>
13 #include <linux/sysctl.h>
14 #include <linux/init.h>
15 #include <linux/fs.h>
16 #include "trace.h"
17 
18 #define STACK_TRACE_ENTRIES 500
19 
20 static unsigned long stack_dump_trace[STACK_TRACE_ENTRIES+1] =
21 	 { [0 ... (STACK_TRACE_ENTRIES)] = ULONG_MAX };
22 static unsigned stack_dump_index[STACK_TRACE_ENTRIES];
23 
24 static struct stack_trace max_stack_trace = {
25 	.max_entries		= STACK_TRACE_ENTRIES,
26 	.entries		= stack_dump_trace,
27 };
28 
29 static unsigned long max_stack_size;
30 static arch_spinlock_t max_stack_lock =
31 	(arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
32 
33 static int stack_trace_disabled __read_mostly;
34 static DEFINE_PER_CPU(int, trace_active);
35 static DEFINE_MUTEX(stack_sysctl_mutex);
36 
37 int stack_tracer_enabled;
38 static int last_stack_tracer_enabled;
39 
40 static inline void check_stack(void)
41 {
42 	unsigned long this_size, flags;
43 	unsigned long *p, *top, *start;
44 	int i;
45 
46 	this_size = ((unsigned long)&this_size) & (THREAD_SIZE-1);
47 	this_size = THREAD_SIZE - this_size;
48 
49 	if (this_size <= max_stack_size)
50 		return;
51 
52 	/* we do not handle interrupt stacks yet */
53 	if (!object_is_on_stack(&this_size))
54 		return;
55 
56 	local_irq_save(flags);
57 	arch_spin_lock(&max_stack_lock);
58 
59 	/* a race could have already updated it */
60 	if (this_size <= max_stack_size)
61 		goto out;
62 
63 	max_stack_size = this_size;
64 
65 	max_stack_trace.nr_entries	= 0;
66 	max_stack_trace.skip		= 3;
67 
68 	save_stack_trace(&max_stack_trace);
69 
70 	/*
71 	 * Now find where in the stack these are.
72 	 */
73 	i = 0;
74 	start = &this_size;
75 	top = (unsigned long *)
76 		(((unsigned long)start & ~(THREAD_SIZE-1)) + THREAD_SIZE);
77 
78 	/*
79 	 * Loop through all the entries. One of the entries may
80 	 * for some reason be missed on the stack, so we may
81 	 * have to account for them. If they are all there, this
82 	 * loop will only happen once. This code only takes place
83 	 * on a new max, so it is far from a fast path.
84 	 */
85 	while (i < max_stack_trace.nr_entries) {
86 		int found = 0;
87 
88 		stack_dump_index[i] = this_size;
89 		p = start;
90 
91 		for (; p < top && i < max_stack_trace.nr_entries; p++) {
92 			if (*p == stack_dump_trace[i]) {
93 				this_size = stack_dump_index[i++] =
94 					(top - p) * sizeof(unsigned long);
95 				found = 1;
96 				/* Start the search from here */
97 				start = p + 1;
98 			}
99 		}
100 
101 		if (!found)
102 			i++;
103 	}
104 
105  out:
106 	arch_spin_unlock(&max_stack_lock);
107 	local_irq_restore(flags);
108 }
109 
110 static void
111 stack_trace_call(unsigned long ip, unsigned long parent_ip)
112 {
113 	int cpu;
114 
115 	if (unlikely(!ftrace_enabled || stack_trace_disabled))
116 		return;
117 
118 	preempt_disable_notrace();
119 
120 	cpu = raw_smp_processor_id();
121 	/* no atomic needed, we only modify this variable by this cpu */
122 	if (per_cpu(trace_active, cpu)++ != 0)
123 		goto out;
124 
125 	check_stack();
126 
127  out:
128 	per_cpu(trace_active, cpu)--;
129 	/* prevent recursion in schedule */
130 	preempt_enable_notrace();
131 }
132 
133 static struct ftrace_ops trace_ops __read_mostly =
134 {
135 	.func = stack_trace_call,
136 	.flags = FTRACE_OPS_FL_GLOBAL,
137 };
138 
139 static ssize_t
140 stack_max_size_read(struct file *filp, char __user *ubuf,
141 		    size_t count, loff_t *ppos)
142 {
143 	unsigned long *ptr = filp->private_data;
144 	char buf[64];
145 	int r;
146 
147 	r = snprintf(buf, sizeof(buf), "%ld\n", *ptr);
148 	if (r > sizeof(buf))
149 		r = sizeof(buf);
150 	return simple_read_from_buffer(ubuf, count, ppos, buf, r);
151 }
152 
153 static ssize_t
154 stack_max_size_write(struct file *filp, const char __user *ubuf,
155 		     size_t count, loff_t *ppos)
156 {
157 	long *ptr = filp->private_data;
158 	unsigned long val, flags;
159 	char buf[64];
160 	int ret;
161 	int cpu;
162 
163 	if (count >= sizeof(buf))
164 		return -EINVAL;
165 
166 	if (copy_from_user(&buf, ubuf, count))
167 		return -EFAULT;
168 
169 	buf[count] = 0;
170 
171 	ret = strict_strtoul(buf, 10, &val);
172 	if (ret < 0)
173 		return ret;
174 
175 	local_irq_save(flags);
176 
177 	/*
178 	 * In case we trace inside arch_spin_lock() or after (NMI),
179 	 * we will cause circular lock, so we also need to increase
180 	 * the percpu trace_active here.
181 	 */
182 	cpu = smp_processor_id();
183 	per_cpu(trace_active, cpu)++;
184 
185 	arch_spin_lock(&max_stack_lock);
186 	*ptr = val;
187 	arch_spin_unlock(&max_stack_lock);
188 
189 	per_cpu(trace_active, cpu)--;
190 	local_irq_restore(flags);
191 
192 	return count;
193 }
194 
195 static const struct file_operations stack_max_size_fops = {
196 	.open		= tracing_open_generic,
197 	.read		= stack_max_size_read,
198 	.write		= stack_max_size_write,
199 	.llseek		= default_llseek,
200 };
201 
202 static void *
203 __next(struct seq_file *m, loff_t *pos)
204 {
205 	long n = *pos - 1;
206 
207 	if (n >= max_stack_trace.nr_entries || stack_dump_trace[n] == ULONG_MAX)
208 		return NULL;
209 
210 	m->private = (void *)n;
211 	return &m->private;
212 }
213 
214 static void *
215 t_next(struct seq_file *m, void *v, loff_t *pos)
216 {
217 	(*pos)++;
218 	return __next(m, pos);
219 }
220 
221 static void *t_start(struct seq_file *m, loff_t *pos)
222 {
223 	int cpu;
224 
225 	local_irq_disable();
226 
227 	cpu = smp_processor_id();
228 	per_cpu(trace_active, cpu)++;
229 
230 	arch_spin_lock(&max_stack_lock);
231 
232 	if (*pos == 0)
233 		return SEQ_START_TOKEN;
234 
235 	return __next(m, pos);
236 }
237 
238 static void t_stop(struct seq_file *m, void *p)
239 {
240 	int cpu;
241 
242 	arch_spin_unlock(&max_stack_lock);
243 
244 	cpu = smp_processor_id();
245 	per_cpu(trace_active, cpu)--;
246 
247 	local_irq_enable();
248 }
249 
250 static int trace_lookup_stack(struct seq_file *m, long i)
251 {
252 	unsigned long addr = stack_dump_trace[i];
253 
254 	return seq_printf(m, "%pS\n", (void *)addr);
255 }
256 
257 static void print_disabled(struct seq_file *m)
258 {
259 	seq_puts(m, "#\n"
260 		 "#  Stack tracer disabled\n"
261 		 "#\n"
262 		 "# To enable the stack tracer, either add 'stacktrace' to the\n"
263 		 "# kernel command line\n"
264 		 "# or 'echo 1 > /proc/sys/kernel/stack_tracer_enabled'\n"
265 		 "#\n");
266 }
267 
268 static int t_show(struct seq_file *m, void *v)
269 {
270 	long i;
271 	int size;
272 
273 	if (v == SEQ_START_TOKEN) {
274 		seq_printf(m, "        Depth    Size   Location"
275 			   "    (%d entries)\n"
276 			   "        -----    ----   --------\n",
277 			   max_stack_trace.nr_entries - 1);
278 
279 		if (!stack_tracer_enabled && !max_stack_size)
280 			print_disabled(m);
281 
282 		return 0;
283 	}
284 
285 	i = *(long *)v;
286 
287 	if (i >= max_stack_trace.nr_entries ||
288 	    stack_dump_trace[i] == ULONG_MAX)
289 		return 0;
290 
291 	if (i+1 == max_stack_trace.nr_entries ||
292 	    stack_dump_trace[i+1] == ULONG_MAX)
293 		size = stack_dump_index[i];
294 	else
295 		size = stack_dump_index[i] - stack_dump_index[i+1];
296 
297 	seq_printf(m, "%3ld) %8d   %5d   ", i, stack_dump_index[i], size);
298 
299 	trace_lookup_stack(m, i);
300 
301 	return 0;
302 }
303 
304 static const struct seq_operations stack_trace_seq_ops = {
305 	.start		= t_start,
306 	.next		= t_next,
307 	.stop		= t_stop,
308 	.show		= t_show,
309 };
310 
311 static int stack_trace_open(struct inode *inode, struct file *file)
312 {
313 	return seq_open(file, &stack_trace_seq_ops);
314 }
315 
316 static const struct file_operations stack_trace_fops = {
317 	.open		= stack_trace_open,
318 	.read		= seq_read,
319 	.llseek		= seq_lseek,
320 	.release	= seq_release,
321 };
322 
323 int
324 stack_trace_sysctl(struct ctl_table *table, int write,
325 		   void __user *buffer, size_t *lenp,
326 		   loff_t *ppos)
327 {
328 	int ret;
329 
330 	mutex_lock(&stack_sysctl_mutex);
331 
332 	ret = proc_dointvec(table, write, buffer, lenp, ppos);
333 
334 	if (ret || !write ||
335 	    (last_stack_tracer_enabled == !!stack_tracer_enabled))
336 		goto out;
337 
338 	last_stack_tracer_enabled = !!stack_tracer_enabled;
339 
340 	if (stack_tracer_enabled)
341 		register_ftrace_function(&trace_ops);
342 	else
343 		unregister_ftrace_function(&trace_ops);
344 
345  out:
346 	mutex_unlock(&stack_sysctl_mutex);
347 	return ret;
348 }
349 
350 static __init int enable_stacktrace(char *str)
351 {
352 	stack_tracer_enabled = 1;
353 	last_stack_tracer_enabled = 1;
354 	return 1;
355 }
356 __setup("stacktrace", enable_stacktrace);
357 
358 static __init int stack_trace_init(void)
359 {
360 	struct dentry *d_tracer;
361 
362 	d_tracer = tracing_init_dentry();
363 
364 	trace_create_file("stack_max_size", 0644, d_tracer,
365 			&max_stack_size, &stack_max_size_fops);
366 
367 	trace_create_file("stack_trace", 0444, d_tracer,
368 			NULL, &stack_trace_fops);
369 
370 	if (stack_tracer_enabled)
371 		register_ftrace_function(&trace_ops);
372 
373 	return 0;
374 }
375 
376 device_initcall(stack_trace_init);
377