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 DEFINE_PER_CPU(int, trace_active);
  37 static DEFINE_MUTEX(stack_sysctl_mutex);
  38
  39 int stack_tracer_enabled;
  40 static int last_stack_tracer_enabled;
  41
  42 static inline void check_stack(void)
  43 {
  44         unsigned long this_size, flags;
  45         unsigned long *p, *top, *start;
  46         int i;
  47
  48         this_size = ((unsigned long)&this_size) & (THREAD_SIZE-1);
  49         this_size = THREAD_SIZE - this_size;
  50
  51         if (this_size <= max_stack_size)
  52                 return;
  53
  54         /* we do not handle interrupt stacks yet */
  55         if (!object_is_on_stack(&this_size))
  56                 return;
  57
  58         local_irq_save(flags);
  59         arch_spin_lock(&max_stack_lock);
  60
  61         /* a race could have already updated it */
  62         if (this_size <= max_stack_size)
  63                 goto out;
  64
  65         max_stack_size = this_size;
  66
  67         max_stack_trace.nr_entries      = 0;
  68         max_stack_trace.skip            = 3;
  69
  70         save_stack_trace(&max_stack_trace);
  71
  72         /*
  73          * Now find where in the stack these are.
  74          */
  75         i = 0;
  76         start = &this_size;
  77         top = (unsigned long *)
  78                 (((unsigned long)start & ~(THREAD_SIZE-1)) + THREAD_SIZE);
  79
  80         /*
  81          * Loop through all the entries. One of the entries may
  82          * for some reason be missed on the stack, so we may
  83          * have to account for them. If they are all there, this
  84          * loop will only happen once. This code only takes place
  85          * on a new max, so it is far from a fast path.
  86          */
  87         while (i < max_stack_trace.nr_entries) {
  88                 int found = 0;
  89
  90                 stack_dump_index[i] = this_size;
  91                 p = start;
  92
  93                 for (; p < top && i < max_stack_trace.nr_entries; p++) {
  94                         if (*p == stack_dump_trace[i]) {
  95                                 this_size = stack_dump_index[i++] =
  96                                         (top - p) * sizeof(unsigned long);
  97                                 found = 1;
  98                                 /* Start the search from here */
  99                                 start = p + 1;
 100                         }
 101                 }
 102
 103                 if (!found)
 104                         i++;
 105         }
 106
 107  out:
 108         arch_spin_unlock(&max_stack_lock);
 109         local_irq_restore(flags);
 110 }
 111
 112 static void
 113 stack_trace_call(unsigned long ip, unsigned long parent_ip,
 114                  struct ftrace_ops *op, struct pt_regs *pt_regs)
 115 {
 116         int cpu;
 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_RECURSION_SAFE,
 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 static int
 315 stack_trace_filter_open(struct inode *inode, struct file *file)
 316 {
 317         return ftrace_regex_open(&trace_ops, FTRACE_ITER_FILTER,
 318                                  inode, file);
 319 }
 320
 321 static const struct file_operations stack_trace_filter_fops = {
 322         .open = stack_trace_filter_open,
 323         .read = seq_read,
 324         .write = ftrace_filter_write,
 325         .llseek = ftrace_filter_lseek,
 326         .release = ftrace_regex_release,
 327 };
 328
 329 int
 330 stack_trace_sysctl(struct ctl_table *table, int write,
 331                    void __user *buffer, size_t *lenp,
 332                    loff_t *ppos)
 333 {
 334         int ret;
 335
 336         mutex_lock(&stack_sysctl_mutex);
 337
 338         ret = proc_dointvec(table, write, buffer, lenp, ppos);
 339
 340         if (ret || !write ||
 341             (last_stack_tracer_enabled == !!stack_tracer_enabled))
 342                 goto out;
 343
 344         last_stack_tracer_enabled = !!stack_tracer_enabled;
 345
 346         if (stack_tracer_enabled)
 347                 register_ftrace_function(&trace_ops);
 348         else
 349                 unregister_ftrace_function(&trace_ops);
 350
 351  out:
 352         mutex_unlock(&stack_sysctl_mutex);
 353         return ret;
 354 }
 355
 356 static char stack_trace_filter_buf[COMMAND_LINE_SIZE+1] __initdata;
 357
 358 static __init int enable_stacktrace(char *str)
 359 {
 360         if (strncmp(str, "_filter=", 8) == 0)
 361                 strncpy(stack_trace_filter_buf, str+8, COMMAND_LINE_SIZE);
 362
 363         stack_tracer_enabled = 1;
 364         last_stack_tracer_enabled = 1;
 365         return 1;
 366 }
 367 __setup("stacktrace", enable_stacktrace);
 368
 369 static __init int stack_trace_init(void)
 370 {
 371         struct dentry *d_tracer;
 372
 373         d_tracer = tracing_init_dentry();
 374
 375         trace_create_file("stack_max_size", 0644, d_tracer,
 376                         &max_stack_size, &stack_max_size_fops);
 377
 378         trace_create_file("stack_trace", 0444, d_tracer,
 379                         NULL, &stack_trace_fops);
 380
 381         trace_create_file("stack_trace_filter", 0444, d_tracer,
 382                         NULL, &stack_trace_filter_fops);
 383
 384         if (stack_trace_filter_buf[0])
 385                 ftrace_set_early_filter(&trace_ops, stack_trace_filter_buf, 1);
 386
 387         if (stack_tracer_enabled)
 388                 register_ftrace_function(&trace_ops);
 389
 390         return 0;
 391 }
 392
 393 device_initcall(stack_trace_init);