Merge branch 'x86-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
[~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 #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         int ret;
160         int cpu;
161
162         ret = kstrtoul_from_user(ubuf, count, 10, &val);
163         if (ret)
164                 return ret;
165
166         local_irq_save(flags);
167
168         /*
169          * In case we trace inside arch_spin_lock() or after (NMI),
170          * we will cause circular lock, so we also need to increase
171          * the percpu trace_active here.
172          */
173         cpu = smp_processor_id();
174         per_cpu(trace_active, cpu)++;
175
176         arch_spin_lock(&max_stack_lock);
177         *ptr = val;
178         arch_spin_unlock(&max_stack_lock);
179
180         per_cpu(trace_active, cpu)--;
181         local_irq_restore(flags);
182
183         return count;
184 }
185
186 static const struct file_operations stack_max_size_fops = {
187         .open           = tracing_open_generic,
188         .read           = stack_max_size_read,
189         .write          = stack_max_size_write,
190         .llseek         = default_llseek,
191 };
192
193 static void *
194 __next(struct seq_file *m, loff_t *pos)
195 {
196         long n = *pos - 1;
197
198         if (n >= max_stack_trace.nr_entries || stack_dump_trace[n] == ULONG_MAX)
199                 return NULL;
200
201         m->private = (void *)n;
202         return &m->private;
203 }
204
205 static void *
206 t_next(struct seq_file *m, void *v, loff_t *pos)
207 {
208         (*pos)++;
209         return __next(m, pos);
210 }
211
212 static void *t_start(struct seq_file *m, loff_t *pos)
213 {
214         int cpu;
215
216         local_irq_disable();
217
218         cpu = smp_processor_id();
219         per_cpu(trace_active, cpu)++;
220
221         arch_spin_lock(&max_stack_lock);
222
223         if (*pos == 0)
224                 return SEQ_START_TOKEN;
225
226         return __next(m, pos);
227 }
228
229 static void t_stop(struct seq_file *m, void *p)
230 {
231         int cpu;
232
233         arch_spin_unlock(&max_stack_lock);
234
235         cpu = smp_processor_id();
236         per_cpu(trace_active, cpu)--;
237
238         local_irq_enable();
239 }
240
241 static int trace_lookup_stack(struct seq_file *m, long i)
242 {
243         unsigned long addr = stack_dump_trace[i];
244
245         return seq_printf(m, "%pS\n", (void *)addr);
246 }
247
248 static void print_disabled(struct seq_file *m)
249 {
250         seq_puts(m, "#\n"
251                  "#  Stack tracer disabled\n"
252                  "#\n"
253                  "# To enable the stack tracer, either add 'stacktrace' to the\n"
254                  "# kernel command line\n"
255                  "# or 'echo 1 > /proc/sys/kernel/stack_tracer_enabled'\n"
256                  "#\n");
257 }
258
259 static int t_show(struct seq_file *m, void *v)
260 {
261         long i;
262         int size;
263
264         if (v == SEQ_START_TOKEN) {
265                 seq_printf(m, "        Depth    Size   Location"
266                            "    (%d entries)\n"
267                            "        -----    ----   --------\n",
268                            max_stack_trace.nr_entries - 1);
269
270                 if (!stack_tracer_enabled && !max_stack_size)
271                         print_disabled(m);
272
273                 return 0;
274         }
275
276         i = *(long *)v;
277
278         if (i >= max_stack_trace.nr_entries ||
279             stack_dump_trace[i] == ULONG_MAX)
280                 return 0;
281
282         if (i+1 == max_stack_trace.nr_entries ||
283             stack_dump_trace[i+1] == ULONG_MAX)
284                 size = stack_dump_index[i];
285         else
286                 size = stack_dump_index[i] - stack_dump_index[i+1];
287
288         seq_printf(m, "%3ld) %8d   %5d   ", i, stack_dump_index[i], size);
289
290         trace_lookup_stack(m, i);
291
292         return 0;
293 }
294
295 static const struct seq_operations stack_trace_seq_ops = {
296         .start          = t_start,
297         .next           = t_next,
298         .stop           = t_stop,
299         .show           = t_show,
300 };
301
302 static int stack_trace_open(struct inode *inode, struct file *file)
303 {
304         return seq_open(file, &stack_trace_seq_ops);
305 }
306
307 static const struct file_operations stack_trace_fops = {
308         .open           = stack_trace_open,
309         .read           = seq_read,
310         .llseek         = seq_lseek,
311         .release        = seq_release,
312 };
313
314 int
315 stack_trace_sysctl(struct ctl_table *table, int write,
316                    void __user *buffer, size_t *lenp,
317                    loff_t *ppos)
318 {
319         int ret;
320
321         mutex_lock(&stack_sysctl_mutex);
322
323         ret = proc_dointvec(table, write, buffer, lenp, ppos);
324
325         if (ret || !write ||
326             (last_stack_tracer_enabled == !!stack_tracer_enabled))
327                 goto out;
328
329         last_stack_tracer_enabled = !!stack_tracer_enabled;
330
331         if (stack_tracer_enabled)
332                 register_ftrace_function(&trace_ops);
333         else
334                 unregister_ftrace_function(&trace_ops);
335
336  out:
337         mutex_unlock(&stack_sysctl_mutex);
338         return ret;
339 }
340
341 static __init int enable_stacktrace(char *str)
342 {
343         stack_tracer_enabled = 1;
344         last_stack_tracer_enabled = 1;
345         return 1;
346 }
347 __setup("stacktrace", enable_stacktrace);
348
349 static __init int stack_trace_init(void)
350 {
351         struct dentry *d_tracer;
352
353         d_tracer = tracing_init_dentry();
354
355         trace_create_file("stack_max_size", 0644, d_tracer,
356                         &max_stack_size, &stack_max_size_fops);
357
358         trace_create_file("stack_trace", 0444, d_tracer,
359                         NULL, &stack_trace_fops);
360
361         if (stack_tracer_enabled)
362                 register_ftrace_function(&trace_ops);
363
364         return 0;
365 }
366
367 device_initcall(stack_trace_init);