]> git.karo-electronics.de Git - karo-tx-linux.git/blob - kernel/trace/trace_stack.c
f2f02ff350d46897705062617f54578fd64eb7a4
[karo-tx-linux.git] / kernel / trace / trace_stack.c
1 /*
2  * Copyright (C) 2008 Steven Rostedt <srostedt@redhat.com>
3  *
4  */
5 #include <linux/sched/task_stack.h>
6 #include <linux/stacktrace.h>
7 #include <linux/kallsyms.h>
8 #include <linux/seq_file.h>
9 #include <linux/spinlock.h>
10 #include <linux/uaccess.h>
11 #include <linux/ftrace.h>
12 #include <linux/module.h>
13 #include <linux/sysctl.h>
14 #include <linux/init.h>
15
16 #include <asm/setup.h>
17
18 #include "trace.h"
19
20 static unsigned long stack_dump_trace[STACK_TRACE_ENTRIES+1] =
21          { [0 ... (STACK_TRACE_ENTRIES)] = ULONG_MAX };
22 unsigned stack_trace_index[STACK_TRACE_ENTRIES];
23
24 /*
25  * Reserve one entry for the passed in ip. This will allow
26  * us to remove most or all of the stack size overhead
27  * added by the stack tracer itself.
28  */
29 struct stack_trace stack_trace_max = {
30         .max_entries            = STACK_TRACE_ENTRIES - 1,
31         .entries                = &stack_dump_trace[0],
32 };
33
34 unsigned long stack_trace_max_size;
35 arch_spinlock_t stack_trace_max_lock =
36         (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
37
38 DEFINE_PER_CPU(int, disable_stack_tracer);
39 static DEFINE_MUTEX(stack_sysctl_mutex);
40
41 int stack_tracer_enabled;
42 static int last_stack_tracer_enabled;
43
44 void stack_trace_print(void)
45 {
46         long i;
47         int size;
48
49         pr_emerg("        Depth    Size   Location    (%d entries)\n"
50                            "        -----    ----   --------\n",
51                            stack_trace_max.nr_entries);
52
53         for (i = 0; i < stack_trace_max.nr_entries; i++) {
54                 if (stack_dump_trace[i] == ULONG_MAX)
55                         break;
56                 if (i+1 == stack_trace_max.nr_entries ||
57                                 stack_dump_trace[i+1] == ULONG_MAX)
58                         size = stack_trace_index[i];
59                 else
60                         size = stack_trace_index[i] - stack_trace_index[i+1];
61
62                 pr_emerg("%3ld) %8d   %5d   %pS\n", i, stack_trace_index[i],
63                                 size, (void *)stack_dump_trace[i]);
64         }
65 }
66
67 /*
68  * When arch-specific code overrides this function, the following
69  * data should be filled up, assuming stack_trace_max_lock is held to
70  * prevent concurrent updates.
71  *     stack_trace_index[]
72  *     stack_trace_max
73  *     stack_trace_max_size
74  */
75 void __weak
76 check_stack(unsigned long ip, unsigned long *stack)
77 {
78         unsigned long this_size, flags; unsigned long *p, *top, *start;
79         static int tracer_frame;
80         int frame_size = ACCESS_ONCE(tracer_frame);
81         int i, x;
82
83         this_size = ((unsigned long)stack) & (THREAD_SIZE-1);
84         this_size = THREAD_SIZE - this_size;
85         /* Remove the frame of the tracer */
86         this_size -= frame_size;
87
88         if (this_size <= stack_trace_max_size)
89                 return;
90
91         /* we do not handle interrupt stacks yet */
92         if (!object_is_on_stack(stack))
93                 return;
94
95         /* Can't do this from NMI context (can cause deadlocks) */
96         if (in_nmi())
97                 return;
98
99         local_irq_save(flags);
100         arch_spin_lock(&stack_trace_max_lock);
101
102         /*
103          * RCU may not be watching, make it see us.
104          * The stack trace code uses rcu_sched.
105          */
106         rcu_irq_enter();
107
108         /* In case another CPU set the tracer_frame on us */
109         if (unlikely(!frame_size))
110                 this_size -= tracer_frame;
111
112         /* a race could have already updated it */
113         if (this_size <= stack_trace_max_size)
114                 goto out;
115
116         stack_trace_max_size = this_size;
117
118         stack_trace_max.nr_entries = 0;
119         stack_trace_max.skip = 3;
120
121         save_stack_trace(&stack_trace_max);
122
123         /* Skip over the overhead of the stack tracer itself */
124         for (i = 0; i < stack_trace_max.nr_entries; i++) {
125                 if (stack_dump_trace[i] == ip)
126                         break;
127         }
128
129         /*
130          * Some archs may not have the passed in ip in the dump.
131          * If that happens, we need to show everything.
132          */
133         if (i == stack_trace_max.nr_entries)
134                 i = 0;
135
136         /*
137          * Now find where in the stack these are.
138          */
139         x = 0;
140         start = stack;
141         top = (unsigned long *)
142                 (((unsigned long)start & ~(THREAD_SIZE-1)) + THREAD_SIZE);
143
144         /*
145          * Loop through all the entries. One of the entries may
146          * for some reason be missed on the stack, so we may
147          * have to account for them. If they are all there, this
148          * loop will only happen once. This code only takes place
149          * on a new max, so it is far from a fast path.
150          */
151         while (i < stack_trace_max.nr_entries) {
152                 int found = 0;
153
154                 stack_trace_index[x] = this_size;
155                 p = start;
156
157                 for (; p < top && i < stack_trace_max.nr_entries; p++) {
158                         if (stack_dump_trace[i] == ULONG_MAX)
159                                 break;
160                         /*
161                          * The READ_ONCE_NOCHECK is used to let KASAN know that
162                          * this is not a stack-out-of-bounds error.
163                          */
164                         if ((READ_ONCE_NOCHECK(*p)) == stack_dump_trace[i]) {
165                                 stack_dump_trace[x] = stack_dump_trace[i++];
166                                 this_size = stack_trace_index[x++] =
167                                         (top - p) * sizeof(unsigned long);
168                                 found = 1;
169                                 /* Start the search from here */
170                                 start = p + 1;
171                                 /*
172                                  * We do not want to show the overhead
173                                  * of the stack tracer stack in the
174                                  * max stack. If we haven't figured
175                                  * out what that is, then figure it out
176                                  * now.
177                                  */
178                                 if (unlikely(!tracer_frame)) {
179                                         tracer_frame = (p - stack) *
180                                                 sizeof(unsigned long);
181                                         stack_trace_max_size -= tracer_frame;
182                                 }
183                         }
184                 }
185
186                 if (!found)
187                         i++;
188         }
189
190         stack_trace_max.nr_entries = x;
191         for (; x < i; x++)
192                 stack_dump_trace[x] = ULONG_MAX;
193
194         if (task_stack_end_corrupted(current)) {
195                 stack_trace_print();
196                 BUG();
197         }
198
199  out:
200         rcu_irq_exit();
201         arch_spin_unlock(&stack_trace_max_lock);
202         local_irq_restore(flags);
203 }
204
205 static void
206 stack_trace_call(unsigned long ip, unsigned long parent_ip,
207                  struct ftrace_ops *op, struct pt_regs *pt_regs)
208 {
209         unsigned long stack;
210
211         preempt_disable_notrace();
212
213         /* no atomic needed, we only modify this variable by this cpu */
214         __this_cpu_inc(disable_stack_tracer);
215         if (__this_cpu_read(disable_stack_tracer) != 1)
216                 goto out;
217
218         ip += MCOUNT_INSN_SIZE;
219
220         check_stack(ip, &stack);
221
222  out:
223         __this_cpu_dec(disable_stack_tracer);
224         /* prevent recursion in schedule */
225         preempt_enable_notrace();
226 }
227
228 static struct ftrace_ops trace_ops __read_mostly =
229 {
230         .func = stack_trace_call,
231         .flags = FTRACE_OPS_FL_RECURSION_SAFE,
232 };
233
234 static ssize_t
235 stack_max_size_read(struct file *filp, char __user *ubuf,
236                     size_t count, loff_t *ppos)
237 {
238         unsigned long *ptr = filp->private_data;
239         char buf[64];
240         int r;
241
242         r = snprintf(buf, sizeof(buf), "%ld\n", *ptr);
243         if (r > sizeof(buf))
244                 r = sizeof(buf);
245         return simple_read_from_buffer(ubuf, count, ppos, buf, r);
246 }
247
248 static ssize_t
249 stack_max_size_write(struct file *filp, const char __user *ubuf,
250                      size_t count, loff_t *ppos)
251 {
252         long *ptr = filp->private_data;
253         unsigned long val, flags;
254         int ret;
255
256         ret = kstrtoul_from_user(ubuf, count, 10, &val);
257         if (ret)
258                 return ret;
259
260         local_irq_save(flags);
261
262         /*
263          * In case we trace inside arch_spin_lock() or after (NMI),
264          * we will cause circular lock, so we also need to increase
265          * the percpu disable_stack_tracer here.
266          */
267         __this_cpu_inc(disable_stack_tracer);
268
269         arch_spin_lock(&stack_trace_max_lock);
270         *ptr = val;
271         arch_spin_unlock(&stack_trace_max_lock);
272
273         __this_cpu_dec(disable_stack_tracer);
274         local_irq_restore(flags);
275
276         return count;
277 }
278
279 static const struct file_operations stack_max_size_fops = {
280         .open           = tracing_open_generic,
281         .read           = stack_max_size_read,
282         .write          = stack_max_size_write,
283         .llseek         = default_llseek,
284 };
285
286 static void *
287 __next(struct seq_file *m, loff_t *pos)
288 {
289         long n = *pos - 1;
290
291         if (n > stack_trace_max.nr_entries || stack_dump_trace[n] == ULONG_MAX)
292                 return NULL;
293
294         m->private = (void *)n;
295         return &m->private;
296 }
297
298 static void *
299 t_next(struct seq_file *m, void *v, loff_t *pos)
300 {
301         (*pos)++;
302         return __next(m, pos);
303 }
304
305 static void *t_start(struct seq_file *m, loff_t *pos)
306 {
307         local_irq_disable();
308
309         __this_cpu_inc(disable_stack_tracer);
310
311         arch_spin_lock(&stack_trace_max_lock);
312
313         if (*pos == 0)
314                 return SEQ_START_TOKEN;
315
316         return __next(m, pos);
317 }
318
319 static void t_stop(struct seq_file *m, void *p)
320 {
321         arch_spin_unlock(&stack_trace_max_lock);
322
323         __this_cpu_dec(disable_stack_tracer);
324
325         local_irq_enable();
326 }
327
328 static void trace_lookup_stack(struct seq_file *m, long i)
329 {
330         unsigned long addr = stack_dump_trace[i];
331
332         seq_printf(m, "%pS\n", (void *)addr);
333 }
334
335 static void print_disabled(struct seq_file *m)
336 {
337         seq_puts(m, "#\n"
338                  "#  Stack tracer disabled\n"
339                  "#\n"
340                  "# To enable the stack tracer, either add 'stacktrace' to the\n"
341                  "# kernel command line\n"
342                  "# or 'echo 1 > /proc/sys/kernel/stack_tracer_enabled'\n"
343                  "#\n");
344 }
345
346 static int t_show(struct seq_file *m, void *v)
347 {
348         long i;
349         int size;
350
351         if (v == SEQ_START_TOKEN) {
352                 seq_printf(m, "        Depth    Size   Location"
353                            "    (%d entries)\n"
354                            "        -----    ----   --------\n",
355                            stack_trace_max.nr_entries);
356
357                 if (!stack_tracer_enabled && !stack_trace_max_size)
358                         print_disabled(m);
359
360                 return 0;
361         }
362
363         i = *(long *)v;
364
365         if (i >= stack_trace_max.nr_entries ||
366             stack_dump_trace[i] == ULONG_MAX)
367                 return 0;
368
369         if (i+1 == stack_trace_max.nr_entries ||
370             stack_dump_trace[i+1] == ULONG_MAX)
371                 size = stack_trace_index[i];
372         else
373                 size = stack_trace_index[i] - stack_trace_index[i+1];
374
375         seq_printf(m, "%3ld) %8d   %5d   ", i, stack_trace_index[i], size);
376
377         trace_lookup_stack(m, i);
378
379         return 0;
380 }
381
382 static const struct seq_operations stack_trace_seq_ops = {
383         .start          = t_start,
384         .next           = t_next,
385         .stop           = t_stop,
386         .show           = t_show,
387 };
388
389 static int stack_trace_open(struct inode *inode, struct file *file)
390 {
391         return seq_open(file, &stack_trace_seq_ops);
392 }
393
394 static const struct file_operations stack_trace_fops = {
395         .open           = stack_trace_open,
396         .read           = seq_read,
397         .llseek         = seq_lseek,
398         .release        = seq_release,
399 };
400
401 static int
402 stack_trace_filter_open(struct inode *inode, struct file *file)
403 {
404         return ftrace_regex_open(&trace_ops, FTRACE_ITER_FILTER,
405                                  inode, file);
406 }
407
408 static const struct file_operations stack_trace_filter_fops = {
409         .open = stack_trace_filter_open,
410         .read = seq_read,
411         .write = ftrace_filter_write,
412         .llseek = tracing_lseek,
413         .release = ftrace_regex_release,
414 };
415
416 int
417 stack_trace_sysctl(struct ctl_table *table, int write,
418                    void __user *buffer, size_t *lenp,
419                    loff_t *ppos)
420 {
421         int ret;
422
423         mutex_lock(&stack_sysctl_mutex);
424
425         ret = proc_dointvec(table, write, buffer, lenp, ppos);
426
427         if (ret || !write ||
428             (last_stack_tracer_enabled == !!stack_tracer_enabled))
429                 goto out;
430
431         last_stack_tracer_enabled = !!stack_tracer_enabled;
432
433         if (stack_tracer_enabled)
434                 register_ftrace_function(&trace_ops);
435         else
436                 unregister_ftrace_function(&trace_ops);
437
438  out:
439         mutex_unlock(&stack_sysctl_mutex);
440         return ret;
441 }
442
443 static char stack_trace_filter_buf[COMMAND_LINE_SIZE+1] __initdata;
444
445 static __init int enable_stacktrace(char *str)
446 {
447         if (strncmp(str, "_filter=", 8) == 0)
448                 strncpy(stack_trace_filter_buf, str+8, COMMAND_LINE_SIZE);
449
450         stack_tracer_enabled = 1;
451         last_stack_tracer_enabled = 1;
452         return 1;
453 }
454 __setup("stacktrace", enable_stacktrace);
455
456 static __init int stack_trace_init(void)
457 {
458         struct dentry *d_tracer;
459
460         d_tracer = tracing_init_dentry();
461         if (IS_ERR(d_tracer))
462                 return 0;
463
464         trace_create_file("stack_max_size", 0644, d_tracer,
465                         &stack_trace_max_size, &stack_max_size_fops);
466
467         trace_create_file("stack_trace", 0444, d_tracer,
468                         NULL, &stack_trace_fops);
469
470         trace_create_file("stack_trace_filter", 0444, d_tracer,
471                         NULL, &stack_trace_filter_fops);
472
473         if (stack_trace_filter_buf[0])
474                 ftrace_set_early_filter(&trace_ops, stack_trace_filter_buf, 1);
475
476         if (stack_tracer_enabled)
477                 register_ftrace_function(&trace_ops);
478
479         return 0;
480 }
481
482 device_initcall(stack_trace_init);