Merge branch 'perf-core-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git...
[~shefty/rdma-dev.git] / kernel / trace / trace_stack.c
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
17 #include <asm/setup.h>
18
19 #include "trace.h"
20
21 #define STACK_TRACE_ENTRIES 500
22
23 static unsigned long stack_dump_trace[STACK_TRACE_ENTRIES+1] =
24          { [0 ... (STACK_TRACE_ENTRIES)] = ULONG_MAX };
25 static unsigned stack_dump_index[STACK_TRACE_ENTRIES];
26
27 static struct stack_trace max_stack_trace = {
28         .max_entries            = STACK_TRACE_ENTRIES,
29         .entries                = stack_dump_trace,
30 };
31
32 static unsigned long max_stack_size;
33 static arch_spinlock_t max_stack_lock =
34         (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
35
36 static int stack_trace_disabled __read_mostly;
37 static DEFINE_PER_CPU(int, trace_active);
38 static DEFINE_MUTEX(stack_sysctl_mutex);
39
40 int stack_tracer_enabled;
41 static int last_stack_tracer_enabled;
42
43 static inline void check_stack(void)
44 {
45         unsigned long this_size, flags;
46         unsigned long *p, *top, *start;
47         int i;
48
49         this_size = ((unsigned long)&this_size) & (THREAD_SIZE-1);
50         this_size = THREAD_SIZE - this_size;
51
52         if (this_size <= max_stack_size)
53                 return;
54
55         /* we do not handle interrupt stacks yet */
56         if (!object_is_on_stack(&this_size))
57                 return;
58
59         local_irq_save(flags);
60         arch_spin_lock(&max_stack_lock);
61
62         /* a race could have already updated it */
63         if (this_size <= max_stack_size)
64                 goto out;
65
66         max_stack_size = this_size;
67
68         max_stack_trace.nr_entries      = 0;
69         max_stack_trace.skip            = 3;
70
71         save_stack_trace(&max_stack_trace);
72
73         /*
74          * Now find where in the stack these are.
75          */
76         i = 0;
77         start = &this_size;
78         top = (unsigned long *)
79                 (((unsigned long)start & ~(THREAD_SIZE-1)) + THREAD_SIZE);
80
81         /*
82          * Loop through all the entries. One of the entries may
83          * for some reason be missed on the stack, so we may
84          * have to account for them. If they are all there, this
85          * loop will only happen once. This code only takes place
86          * on a new max, so it is far from a fast path.
87          */
88         while (i < max_stack_trace.nr_entries) {
89                 int found = 0;
90
91                 stack_dump_index[i] = this_size;
92                 p = start;
93
94                 for (; p < top && i < max_stack_trace.nr_entries; p++) {
95                         if (*p == stack_dump_trace[i]) {
96                                 this_size = stack_dump_index[i++] =
97                                         (top - p) * sizeof(unsigned long);
98                                 found = 1;
99                                 /* Start the search from here */
100                                 start = p + 1;
101                         }
102                 }
103
104                 if (!found)
105                         i++;
106         }
107
108  out:
109         arch_spin_unlock(&max_stack_lock);
110         local_irq_restore(flags);
111 }
112
113 static void
114 stack_trace_call(unsigned long ip, unsigned long parent_ip)
115 {
116         int cpu;
117
118         if (unlikely(!ftrace_enabled || stack_trace_disabled))
119                 return;
120
121         preempt_disable_notrace();
122
123         cpu = raw_smp_processor_id();
124         /* no atomic needed, we only modify this variable by this cpu */
125         if (per_cpu(trace_active, cpu)++ != 0)
126                 goto out;
127
128         check_stack();
129
130  out:
131         per_cpu(trace_active, cpu)--;
132         /* prevent recursion in schedule */
133         preempt_enable_notrace();
134 }
135
136 static struct ftrace_ops trace_ops __read_mostly =
137 {
138         .func = stack_trace_call,
139 };
140
141 static ssize_t
142 stack_max_size_read(struct file *filp, char __user *ubuf,
143                     size_t count, loff_t *ppos)
144 {
145         unsigned long *ptr = filp->private_data;
146         char buf[64];
147         int r;
148
149         r = snprintf(buf, sizeof(buf), "%ld\n", *ptr);
150         if (r > sizeof(buf))
151                 r = sizeof(buf);
152         return simple_read_from_buffer(ubuf, count, ppos, buf, r);
153 }
154
155 static ssize_t
156 stack_max_size_write(struct file *filp, const char __user *ubuf,
157                      size_t count, loff_t *ppos)
158 {
159         long *ptr = filp->private_data;
160         unsigned long val, flags;
161         int ret;
162         int cpu;
163
164         ret = kstrtoul_from_user(ubuf, count, 10, &val);
165         if (ret)
166                 return ret;
167
168         local_irq_save(flags);
169
170         /*
171          * In case we trace inside arch_spin_lock() or after (NMI),
172          * we will cause circular lock, so we also need to increase
173          * the percpu trace_active here.
174          */
175         cpu = smp_processor_id();
176         per_cpu(trace_active, cpu)++;
177
178         arch_spin_lock(&max_stack_lock);
179         *ptr = val;
180         arch_spin_unlock(&max_stack_lock);
181
182         per_cpu(trace_active, cpu)--;
183         local_irq_restore(flags);
184
185         return count;
186 }
187
188 static const struct file_operations stack_max_size_fops = {
189         .open           = tracing_open_generic,
190         .read           = stack_max_size_read,
191         .write          = stack_max_size_write,
192         .llseek         = default_llseek,
193 };
194
195 static void *
196 __next(struct seq_file *m, loff_t *pos)
197 {
198         long n = *pos - 1;
199
200         if (n >= max_stack_trace.nr_entries || stack_dump_trace[n] == ULONG_MAX)
201                 return NULL;
202
203         m->private = (void *)n;
204         return &m->private;
205 }
206
207 static void *
208 t_next(struct seq_file *m, void *v, loff_t *pos)
209 {
210         (*pos)++;
211         return __next(m, pos);
212 }
213
214 static void *t_start(struct seq_file *m, loff_t *pos)
215 {
216         int cpu;
217
218         local_irq_disable();
219
220         cpu = smp_processor_id();
221         per_cpu(trace_active, cpu)++;
222
223         arch_spin_lock(&max_stack_lock);
224
225         if (*pos == 0)
226                 return SEQ_START_TOKEN;
227
228         return __next(m, pos);
229 }
230
231 static void t_stop(struct seq_file *m, void *p)
232 {
233         int cpu;
234
235         arch_spin_unlock(&max_stack_lock);
236
237         cpu = smp_processor_id();
238         per_cpu(trace_active, cpu)--;
239
240         local_irq_enable();
241 }
242
243 static int trace_lookup_stack(struct seq_file *m, long i)
244 {
245         unsigned long addr = stack_dump_trace[i];
246
247         return seq_printf(m, "%pS\n", (void *)addr);
248 }
249
250 static void print_disabled(struct seq_file *m)
251 {
252         seq_puts(m, "#\n"
253                  "#  Stack tracer disabled\n"
254                  "#\n"
255                  "# To enable the stack tracer, either add 'stacktrace' to the\n"
256                  "# kernel command line\n"
257                  "# or 'echo 1 > /proc/sys/kernel/stack_tracer_enabled'\n"
258                  "#\n");
259 }
260
261 static int t_show(struct seq_file *m, void *v)
262 {
263         long i;
264         int size;
265
266         if (v == SEQ_START_TOKEN) {
267                 seq_printf(m, "        Depth    Size   Location"
268                            "    (%d entries)\n"
269                            "        -----    ----   --------\n",
270                            max_stack_trace.nr_entries - 1);
271
272                 if (!stack_tracer_enabled && !max_stack_size)
273                         print_disabled(m);
274
275                 return 0;
276         }
277
278         i = *(long *)v;
279
280         if (i >= max_stack_trace.nr_entries ||
281             stack_dump_trace[i] == ULONG_MAX)
282                 return 0;
283
284         if (i+1 == max_stack_trace.nr_entries ||
285             stack_dump_trace[i+1] == ULONG_MAX)
286                 size = stack_dump_index[i];
287         else
288                 size = stack_dump_index[i] - stack_dump_index[i+1];
289
290         seq_printf(m, "%3ld) %8d   %5d   ", i, stack_dump_index[i], size);
291
292         trace_lookup_stack(m, i);
293
294         return 0;
295 }
296
297 static const struct seq_operations stack_trace_seq_ops = {
298         .start          = t_start,
299         .next           = t_next,
300         .stop           = t_stop,
301         .show           = t_show,
302 };
303
304 static int stack_trace_open(struct inode *inode, struct file *file)
305 {
306         return seq_open(file, &stack_trace_seq_ops);
307 }
308
309 static const struct file_operations stack_trace_fops = {
310         .open           = stack_trace_open,
311         .read           = seq_read,
312         .llseek         = seq_lseek,
313         .release        = seq_release,
314 };
315
316 static int
317 stack_trace_filter_open(struct inode *inode, struct file *file)
318 {
319         return ftrace_regex_open(&trace_ops, FTRACE_ITER_FILTER,
320                                  inode, file);
321 }
322
323 static const struct file_operations stack_trace_filter_fops = {
324         .open = stack_trace_filter_open,
325         .read = seq_read,
326         .write = ftrace_filter_write,
327         .llseek = ftrace_regex_lseek,
328         .release = ftrace_regex_release,
329 };
330
331 int
332 stack_trace_sysctl(struct ctl_table *table, int write,
333                    void __user *buffer, size_t *lenp,
334                    loff_t *ppos)
335 {
336         int ret;
337
338         mutex_lock(&stack_sysctl_mutex);
339
340         ret = proc_dointvec(table, write, buffer, lenp, ppos);
341
342         if (ret || !write ||
343             (last_stack_tracer_enabled == !!stack_tracer_enabled))
344                 goto out;
345
346         last_stack_tracer_enabled = !!stack_tracer_enabled;
347
348         if (stack_tracer_enabled)
349                 register_ftrace_function(&trace_ops);
350         else
351                 unregister_ftrace_function(&trace_ops);
352
353  out:
354         mutex_unlock(&stack_sysctl_mutex);
355         return ret;
356 }
357
358 static char stack_trace_filter_buf[COMMAND_LINE_SIZE+1] __initdata;
359
360 static __init int enable_stacktrace(char *str)
361 {
362         if (strncmp(str, "_filter=", 8) == 0)
363                 strncpy(stack_trace_filter_buf, str+8, COMMAND_LINE_SIZE);
364
365         stack_tracer_enabled = 1;
366         last_stack_tracer_enabled = 1;
367         return 1;
368 }
369 __setup("stacktrace", enable_stacktrace);
370
371 static __init int stack_trace_init(void)
372 {
373         struct dentry *d_tracer;
374
375         d_tracer = tracing_init_dentry();
376
377         trace_create_file("stack_max_size", 0644, d_tracer,
378                         &max_stack_size, &stack_max_size_fops);
379
380         trace_create_file("stack_trace", 0444, d_tracer,
381                         NULL, &stack_trace_fops);
382
383         trace_create_file("stack_trace_filter", 0444, d_tracer,
384                         NULL, &stack_trace_filter_fops);
385
386         if (stack_trace_filter_buf[0])
387                 ftrace_set_early_filter(&trace_ops, stack_trace_filter_buf, 1);
388
389         if (stack_tracer_enabled)
390                 register_ftrace_function(&trace_ops);
391
392         return 0;
393 }
394
395 device_initcall(stack_trace_init);