Merge branch 'kvm-updates/2.6.31' of git://git.kernel.org/pub/scm/virt/kvm/kvm
[linux-2.6] / 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 raw_spinlock_t max_stack_lock =
31         (raw_spinlock_t)__RAW_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         __raw_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         __raw_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, resched;
114
115         if (unlikely(!ftrace_enabled || stack_trace_disabled))
116                 return;
117
118         resched = ftrace_preempt_disable();
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         ftrace_preempt_enable(resched);
131 }
132
133 static struct ftrace_ops trace_ops __read_mostly =
134 {
135         .func = stack_trace_call,
136 };
137
138 static ssize_t
139 stack_max_size_read(struct file *filp, char __user *ubuf,
140                     size_t count, loff_t *ppos)
141 {
142         unsigned long *ptr = filp->private_data;
143         char buf[64];
144         int r;
145
146         r = snprintf(buf, sizeof(buf), "%ld\n", *ptr);
147         if (r > sizeof(buf))
148                 r = sizeof(buf);
149         return simple_read_from_buffer(ubuf, count, ppos, buf, r);
150 }
151
152 static ssize_t
153 stack_max_size_write(struct file *filp, const char __user *ubuf,
154                      size_t count, loff_t *ppos)
155 {
156         long *ptr = filp->private_data;
157         unsigned long val, flags;
158         char buf[64];
159         int ret;
160
161         if (count >= sizeof(buf))
162                 return -EINVAL;
163
164         if (copy_from_user(&buf, ubuf, count))
165                 return -EFAULT;
166
167         buf[count] = 0;
168
169         ret = strict_strtoul(buf, 10, &val);
170         if (ret < 0)
171                 return ret;
172
173         local_irq_save(flags);
174         __raw_spin_lock(&max_stack_lock);
175         *ptr = val;
176         __raw_spin_unlock(&max_stack_lock);
177         local_irq_restore(flags);
178
179         return count;
180 }
181
182 static const struct file_operations stack_max_size_fops = {
183         .open           = tracing_open_generic,
184         .read           = stack_max_size_read,
185         .write          = stack_max_size_write,
186 };
187
188 static void *
189 t_next(struct seq_file *m, void *v, loff_t *pos)
190 {
191         long i;
192
193         (*pos)++;
194
195         if (v == SEQ_START_TOKEN)
196                 i = 0;
197         else {
198                 i = *(long *)v;
199                 i++;
200         }
201
202         if (i >= max_stack_trace.nr_entries ||
203             stack_dump_trace[i] == ULONG_MAX)
204                 return NULL;
205
206         m->private = (void *)i;
207
208         return &m->private;
209 }
210
211 static void *t_start(struct seq_file *m, loff_t *pos)
212 {
213         void *t = SEQ_START_TOKEN;
214         loff_t l = 0;
215
216         local_irq_disable();
217         __raw_spin_lock(&max_stack_lock);
218
219         if (*pos == 0)
220                 return SEQ_START_TOKEN;
221
222         for (; t && l < *pos; t = t_next(m, t, &l))
223                 ;
224
225         return t;
226 }
227
228 static void t_stop(struct seq_file *m, void *p)
229 {
230         __raw_spin_unlock(&max_stack_lock);
231         local_irq_enable();
232 }
233
234 static int trace_lookup_stack(struct seq_file *m, long i)
235 {
236         unsigned long addr = stack_dump_trace[i];
237 #ifdef CONFIG_KALLSYMS
238         char str[KSYM_SYMBOL_LEN];
239
240         sprint_symbol(str, addr);
241
242         return seq_printf(m, "%s\n", str);
243 #else
244         return seq_printf(m, "%p\n", (void*)addr);
245 #endif
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         int ret;
305
306         ret = seq_open(file, &stack_trace_seq_ops);
307
308         return ret;
309 }
310
311 static const struct file_operations stack_trace_fops = {
312         .open           = stack_trace_open,
313         .read           = seq_read,
314         .llseek         = seq_lseek,
315 };
316
317 int
318 stack_trace_sysctl(struct ctl_table *table, int write,
319                    struct file *file, void __user *buffer, size_t *lenp,
320                    loff_t *ppos)
321 {
322         int ret;
323
324         mutex_lock(&stack_sysctl_mutex);
325
326         ret = proc_dointvec(table, write, file, buffer, lenp, ppos);
327
328         if (ret || !write ||
329             (last_stack_tracer_enabled == stack_tracer_enabled))
330                 goto out;
331
332         last_stack_tracer_enabled = stack_tracer_enabled;
333
334         if (stack_tracer_enabled)
335                 register_ftrace_function(&trace_ops);
336         else
337                 unregister_ftrace_function(&trace_ops);
338
339  out:
340         mutex_unlock(&stack_sysctl_mutex);
341         return ret;
342 }
343
344 static __init int enable_stacktrace(char *str)
345 {
346         stack_tracer_enabled = 1;
347         last_stack_tracer_enabled = 1;
348         return 1;
349 }
350 __setup("stacktrace", enable_stacktrace);
351
352 static __init int stack_trace_init(void)
353 {
354         struct dentry *d_tracer;
355
356         d_tracer = tracing_init_dentry();
357
358         trace_create_file("stack_max_size", 0644, d_tracer,
359                         &max_stack_size, &stack_max_size_fops);
360
361         trace_create_file("stack_trace", 0444, d_tracer,
362                         NULL, &stack_trace_fops);
363
364         if (stack_tracer_enabled)
365                 register_ftrace_function(&trace_ops);
366
367         return 0;
368 }
369
370 device_initcall(stack_trace_init);