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 };
 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         int ret;
 159         int cpu;
 160
 161         ret = kstrtoul_from_user(ubuf, count, 10, &val);
 162         if (ret)
 163                 return ret;
 164
 165         local_irq_save(flags);
 166
 167         /*
 168          * In case we trace inside arch_spin_lock() or after (NMI),
 169          * we will cause circular lock, so we also need to increase
 170          * the percpu trace_active here.
 171          */
 172         cpu = smp_processor_id();
 173         per_cpu(trace_active, cpu)++;
 174
 175         arch_spin_lock(&max_stack_lock);
 176         *ptr = val;
 177         arch_spin_unlock(&max_stack_lock);
 178
 179         per_cpu(trace_active, cpu)--;
 180         local_irq_restore(flags);
 181
 182         return count;
 183 }
 184
 185 static const struct file_operations stack_max_size_fops = {
 186         .open           = tracing_open_generic,
 187         .read           = stack_max_size_read,
 188         .write          = stack_max_size_write,
 189         .llseek         = default_llseek,
 190 };
 191
 192 static void *
 193 __next(struct seq_file *m, loff_t *pos)
 194 {
 195         long n = *pos - 1;
 196
 197         if (n >= max_stack_trace.nr_entries || stack_dump_trace[n] == ULONG_MAX)
 198                 return NULL;
 199
 200         m->private = (void *)n;
 201         return &m->private;
 202 }
 203
 204 static void *
 205 t_next(struct seq_file *m, void *v, loff_t *pos)
 206 {
 207         (*pos)++;
 208         return __next(m, pos);
 209 }
 210
 211 static void *t_start(struct seq_file *m, loff_t *pos)
 212 {
 213         int cpu;
 214
 215         local_irq_disable();
 216
 217         cpu = smp_processor_id();
 218         per_cpu(trace_active, cpu)++;
 219
 220         arch_spin_lock(&max_stack_lock);
 221
 222         if (*pos == 0)
 223                 return SEQ_START_TOKEN;
 224
 225         return __next(m, pos);
 226 }
 227
 228 static void t_stop(struct seq_file *m, void *p)
 229 {
 230         int cpu;
 231
 232         arch_spin_unlock(&max_stack_lock);
 233
 234         cpu = smp_processor_id();
 235         per_cpu(trace_active, cpu)--;
 236
 237         local_irq_enable();
 238 }
 239
 240 static int trace_lookup_stack(struct seq_file *m, long i)
 241 {
 242         unsigned long addr = stack_dump_trace[i];
 243
 244         return seq_printf(m, "%pS\n", (void *)addr);
 245 }
 246
 247 static void print_disabled(struct seq_file *m)
 248 {
 249         seq_puts(m, "#\n"
 250                  "#  Stack tracer disabled\n"
 251                  "#\n"
 252                  "# To enable the stack tracer, either add 'stacktrace' to the\n"
 253                  "# kernel command line\n"
 254                  "# or 'echo 1 > /proc/sys/kernel/stack_tracer_enabled'\n"
 255                  "#\n");
 256 }
 257
 258 static int t_show(struct seq_file *m, void *v)
 259 {
 260         long i;
 261         int size;
 262
 263         if (v == SEQ_START_TOKEN) {
 264                 seq_printf(m, "        Depth    Size   Location"
 265                            "    (%d entries)\n"
 266                            "        -----    ----   --------\n",
 267                            max_stack_trace.nr_entries - 1);
 268
 269                 if (!stack_tracer_enabled && !max_stack_size)
 270                         print_disabled(m);
 271
 272                 return 0;
 273         }
 274
 275         i = *(long *)v;
 276
 277         if (i >= max_stack_trace.nr_entries ||
 278             stack_dump_trace[i] == ULONG_MAX)
 279                 return 0;
 280
 281         if (i+1 == max_stack_trace.nr_entries ||
 282             stack_dump_trace[i+1] == ULONG_MAX)
 283                 size = stack_dump_index[i];
 284         else
 285                 size = stack_dump_index[i] - stack_dump_index[i+1];
 286
 287         seq_printf(m, "%3ld) %8d   %5d   ", i, stack_dump_index[i], size);
 288
 289         trace_lookup_stack(m, i);
 290
 291         return 0;
 292 }
 293
 294 static const struct seq_operations stack_trace_seq_ops = {
 295         .start          = t_start,
 296         .next           = t_next,
 297         .stop           = t_stop,
 298         .show           = t_show,
 299 };
 300
 301 static int stack_trace_open(struct inode *inode, struct file *file)
 302 {
 303         return seq_open(file, &stack_trace_seq_ops);
 304 }
 305
 306 static const struct file_operations stack_trace_fops = {
 307         .open           = stack_trace_open,
 308         .read           = seq_read,
 309         .llseek         = seq_lseek,
 310         .release        = seq_release,
 311 };
 312
 313 static int
 314 stack_trace_filter_open(struct inode *inode, struct file *file)
 315 {
 316         return ftrace_regex_open(&trace_ops, FTRACE_ITER_FILTER,
 317                                  inode, file);
 318 }
 319
 320 static const struct file_operations stack_trace_filter_fops = {
 321         .open = stack_trace_filter_open,
 322         .read = seq_read,
 323         .write = ftrace_filter_write,
 324         .llseek = ftrace_regex_lseek,
 325         .release = ftrace_regex_release,
 326 };
 327
 328 int
 329 stack_trace_sysctl(struct ctl_table *table, int write,
 330                    void __user *buffer, size_t *lenp,
 331                    loff_t *ppos)
 332 {
 333         int ret;
 334
 335         mutex_lock(&stack_sysctl_mutex);
 336
 337         ret = proc_dointvec(table, write, buffer, lenp, ppos);
 338
 339         if (ret || !write ||
 340             (last_stack_tracer_enabled == !!stack_tracer_enabled))
 341                 goto out;
 342
 343         last_stack_tracer_enabled = !!stack_tracer_enabled;
 344
 345         if (stack_tracer_enabled)
 346                 register_ftrace_function(&trace_ops);
 347         else
 348                 unregister_ftrace_function(&trace_ops);
 349
 350  out:
 351         mutex_unlock(&stack_sysctl_mutex);
 352         return ret;
 353 }
 354
 355 static __init int enable_stacktrace(char *str)
 356 {
 357         stack_tracer_enabled = 1;
 358         last_stack_tracer_enabled = 1;
 359         return 1;
 360 }
 361 __setup("stacktrace", enable_stacktrace);
 362
 363 static __init int stack_trace_init(void)
 364 {
 365         struct dentry *d_tracer;
 366
 367         d_tracer = tracing_init_dentry();
 368
 369         trace_create_file("stack_max_size", 0644, d_tracer,
 370                         &max_stack_size, &stack_max_size_fops);
 371
 372         trace_create_file("stack_trace", 0444, d_tracer,
 373                         NULL, &stack_trace_fops);
 374
 375         trace_create_file("stack_trace_filter", 0444, d_tracer,
 376                         NULL, &stack_trace_filter_fops);
 377
 378         if (stack_tracer_enabled)
 379                 register_ftrace_function(&trace_ops);
 380
 381         return 0;
 382 }
 383
 384 device_initcall(stack_trace_init);