KVM: MMU: invalidate and flush on spte small->large page size change
[safe/jmp/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 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, 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         int cpu;
161
162         if (count >= sizeof(buf))
163                 return -EINVAL;
164
165         if (copy_from_user(&buf, ubuf, count))
166                 return -EFAULT;
167
168         buf[count] = 0;
169
170         ret = strict_strtoul(buf, 10, &val);
171         if (ret < 0)
172                 return ret;
173
174         local_irq_save(flags);
175
176         /*
177          * In case we trace inside arch_spin_lock() or after (NMI),
178          * we will cause circular lock, so we also need to increase
179          * the percpu trace_active here.
180          */
181         cpu = smp_processor_id();
182         per_cpu(trace_active, cpu)++;
183
184         arch_spin_lock(&max_stack_lock);
185         *ptr = val;
186         arch_spin_unlock(&max_stack_lock);
187
188         per_cpu(trace_active, cpu)--;
189         local_irq_restore(flags);
190
191         return count;
192 }
193
194 static const struct file_operations stack_max_size_fops = {
195         .open           = tracing_open_generic,
196         .read           = stack_max_size_read,
197         .write          = stack_max_size_write,
198 };
199
200 static void *
201 __next(struct seq_file *m, loff_t *pos)
202 {
203         long n = *pos - 1;
204
205         if (n >= max_stack_trace.nr_entries || stack_dump_trace[n] == ULONG_MAX)
206                 return NULL;
207
208         m->private = (void *)n;
209         return &m->private;
210 }
211
212 static void *
213 t_next(struct seq_file *m, void *v, loff_t *pos)
214 {
215         (*pos)++;
216         return __next(m, pos);
217 }
218
219 static void *t_start(struct seq_file *m, loff_t *pos)
220 {
221         int cpu;
222
223         local_irq_disable();
224
225         cpu = smp_processor_id();
226         per_cpu(trace_active, cpu)++;
227
228         arch_spin_lock(&max_stack_lock);
229
230         if (*pos == 0)
231                 return SEQ_START_TOKEN;
232
233         return __next(m, pos);
234 }
235
236 static void t_stop(struct seq_file *m, void *p)
237 {
238         int cpu;
239
240         arch_spin_unlock(&max_stack_lock);
241
242         cpu = smp_processor_id();
243         per_cpu(trace_active, cpu)--;
244
245         local_irq_enable();
246 }
247
248 static int trace_lookup_stack(struct seq_file *m, long i)
249 {
250         unsigned long addr = stack_dump_trace[i];
251
252         return seq_printf(m, "%pF\n", (void *)addr);
253 }
254
255 static void print_disabled(struct seq_file *m)
256 {
257         seq_puts(m, "#\n"
258                  "#  Stack tracer disabled\n"
259                  "#\n"
260                  "# To enable the stack tracer, either add 'stacktrace' to the\n"
261                  "# kernel command line\n"
262                  "# or 'echo 1 > /proc/sys/kernel/stack_tracer_enabled'\n"
263                  "#\n");
264 }
265
266 static int t_show(struct seq_file *m, void *v)
267 {
268         long i;
269         int size;
270
271         if (v == SEQ_START_TOKEN) {
272                 seq_printf(m, "        Depth    Size   Location"
273                            "    (%d entries)\n"
274                            "        -----    ----   --------\n",
275                            max_stack_trace.nr_entries - 1);
276
277                 if (!stack_tracer_enabled && !max_stack_size)
278                         print_disabled(m);
279
280                 return 0;
281         }
282
283         i = *(long *)v;
284
285         if (i >= max_stack_trace.nr_entries ||
286             stack_dump_trace[i] == ULONG_MAX)
287                 return 0;
288
289         if (i+1 == max_stack_trace.nr_entries ||
290             stack_dump_trace[i+1] == ULONG_MAX)
291                 size = stack_dump_index[i];
292         else
293                 size = stack_dump_index[i] - stack_dump_index[i+1];
294
295         seq_printf(m, "%3ld) %8d   %5d   ", i, stack_dump_index[i], size);
296
297         trace_lookup_stack(m, i);
298
299         return 0;
300 }
301
302 static const struct seq_operations stack_trace_seq_ops = {
303         .start          = t_start,
304         .next           = t_next,
305         .stop           = t_stop,
306         .show           = t_show,
307 };
308
309 static int stack_trace_open(struct inode *inode, struct file *file)
310 {
311         return seq_open(file, &stack_trace_seq_ops);
312 }
313
314 static const struct file_operations stack_trace_fops = {
315         .open           = stack_trace_open,
316         .read           = seq_read,
317         .llseek         = seq_lseek,
318         .release        = seq_release,
319 };
320
321 int
322 stack_trace_sysctl(struct ctl_table *table, int write,
323                    void __user *buffer, size_t *lenp,
324                    loff_t *ppos)
325 {
326         int ret;
327
328         mutex_lock(&stack_sysctl_mutex);
329
330         ret = proc_dointvec(table, write, buffer, lenp, ppos);
331
332         if (ret || !write ||
333             (last_stack_tracer_enabled == !!stack_tracer_enabled))
334                 goto out;
335
336         last_stack_tracer_enabled = !!stack_tracer_enabled;
337
338         if (stack_tracer_enabled)
339                 register_ftrace_function(&trace_ops);
340         else
341                 unregister_ftrace_function(&trace_ops);
342
343  out:
344         mutex_unlock(&stack_sysctl_mutex);
345         return ret;
346 }
347
348 static __init int enable_stacktrace(char *str)
349 {
350         stack_tracer_enabled = 1;
351         last_stack_tracer_enabled = 1;
352         return 1;
353 }
354 __setup("stacktrace", enable_stacktrace);
355
356 static __init int stack_trace_init(void)
357 {
358         struct dentry *d_tracer;
359
360         d_tracer = tracing_init_dentry();
361
362         trace_create_file("stack_max_size", 0644, d_tracer,
363                         &max_stack_size, &stack_max_size_fops);
364
365         trace_create_file("stack_trace", 0444, d_tracer,
366                         NULL, &stack_trace_fops);
367
368         if (stack_tracer_enabled)
369                 register_ftrace_function(&trace_ops);
370
371         return 0;
372 }
373
374 device_initcall(stack_trace_init);