arch/s390/oprofile/init.c

   1 /*
   2  * S390 Version
   3  *   Copyright IBM Corp. 2002, 2011
   4  *   Author(s): Thomas Spatzier (tspat@de.ibm.com)
   5  *   Author(s): Mahesh Salgaonkar (mahesh@linux.vnet.ibm.com)
   6  *   Author(s): Heinz Graalfs (graalfs@linux.vnet.ibm.com)
   7  *   Author(s): Andreas Krebbel (krebbel@linux.vnet.ibm.com)
   8  *
   9  * @remark Copyright 2002-2011 OProfile authors
  10  */
  11
  12 #include <linux/oprofile.h>
  13 #include <linux/init.h>
  14 #include <linux/errno.h>
  15 #include <linux/fs.h>
  16 #include <linux/module.h>
  17 #include <asm/processor.h>
  18
  19 #include "../../../drivers/oprofile/oprof.h"
  20
  21 extern void s390_backtrace(struct pt_regs * const regs, unsigned int depth);
  22
  23 #ifdef CONFIG_64BIT
  24
  25 #include "hwsampler.h"
  26 #include "op_counter.h"
  27
  28 #define DEFAULT_INTERVAL        4127518
  29
  30 #define DEFAULT_SDBT_BLOCKS     1
  31 #define DEFAULT_SDB_BLOCKS      511
  32
  33 static unsigned long oprofile_hw_interval = DEFAULT_INTERVAL;
  34 static unsigned long oprofile_min_interval;
  35 static unsigned long oprofile_max_interval;
  36
  37 static unsigned long oprofile_sdbt_blocks = DEFAULT_SDBT_BLOCKS;
  38 static unsigned long oprofile_sdb_blocks = DEFAULT_SDB_BLOCKS;
  39
  40 static int hwsampler_enabled;
  41 static int hwsampler_running;   /* start_mutex must be held to change */
  42 static int hwsampler_available;
  43
  44 static struct oprofile_operations timer_ops;
  45
  46 struct op_counter_config counter_config;
  47
  48 enum __force_cpu_type {
  49         reserved = 0,           /* do not force */
  50         timer,
  51 };
  52 static int force_cpu_type;
  53
  54 static int set_cpu_type(const char *str, struct kernel_param *kp)
  55 {
  56         if (!strcmp(str, "timer")) {
  57                 force_cpu_type = timer;
  58                 printk(KERN_INFO "oprofile: forcing timer to be returned "
  59                                  "as cpu type\n");
  60         } else {
  61                 force_cpu_type = 0;
  62         }
  63
  64         return 0;
  65 }
  66 module_param_call(cpu_type, set_cpu_type, NULL, NULL, 0);
  67 MODULE_PARM_DESC(cpu_type, "Force legacy basic mode sampling"
  68                            "(report cpu_type \"timer\"");
  69
  70 static int oprofile_hwsampler_start(void)
  71 {
  72         int retval;
  73
  74         hwsampler_running = hwsampler_enabled;
  75
  76         if (!hwsampler_running)
  77                 return timer_ops.start();
  78
  79         retval = hwsampler_allocate(oprofile_sdbt_blocks, oprofile_sdb_blocks);
  80         if (retval)
  81                 return retval;
  82
  83         retval = hwsampler_start_all(oprofile_hw_interval);
  84         if (retval)
  85                 hwsampler_deallocate();
  86
  87         return retval;
  88 }
  89
  90 static void oprofile_hwsampler_stop(void)
  91 {
  92         if (!hwsampler_running) {
  93                 timer_ops.stop();
  94                 return;
  95         }
  96
  97         hwsampler_stop_all();
  98         hwsampler_deallocate();
  99         return;
 100 }
 101
 102 /*
 103  * File ops used for:
 104  * /dev/oprofile/0/enabled
 105  * /dev/oprofile/hwsampling/hwsampler  (cpu_type = timer)
 106  */
 107
 108 static ssize_t hwsampler_read(struct file *file, char __user *buf,
 109                 size_t count, loff_t *offset)
 110 {
 111         return oprofilefs_ulong_to_user(hwsampler_enabled, buf, count, offset);
 112 }
 113
 114 static ssize_t hwsampler_write(struct file *file, char const __user *buf,
 115                 size_t count, loff_t *offset)
 116 {
 117         unsigned long val;
 118         int retval;
 119
 120         if (*offset)
 121                 return -EINVAL;
 122
 123         retval = oprofilefs_ulong_from_user(&val, buf, count);
 124         if (retval <= 0)
 125                 return retval;
 126
 127         if (val != 0 && val != 1)
 128                 return -EINVAL;
 129
 130         if (oprofile_started)
 131                 /*
 132                  * save to do without locking as we set
 133                  * hwsampler_running in start() when start_mutex is
 134                  * held
 135                  */
 136                 return -EBUSY;
 137
 138         hwsampler_enabled = val;
 139
 140         return count;
 141 }
 142
 143 static const struct file_operations hwsampler_fops = {
 144         .read           = hwsampler_read,
 145         .write          = hwsampler_write,
 146 };
 147
 148 /*
 149  * File ops used for:
 150  * /dev/oprofile/0/count
 151  * /dev/oprofile/hwsampling/hw_interval  (cpu_type = timer)
 152  *
 153  * Make sure that the value is within the hardware range.
 154  */
 155
 156 static ssize_t hw_interval_read(struct file *file, char __user *buf,
 157                                 size_t count, loff_t *offset)
 158 {
 159         return oprofilefs_ulong_to_user(oprofile_hw_interval, buf,
 160                                         count, offset);
 161 }
 162
 163 static ssize_t hw_interval_write(struct file *file, char const __user *buf,
 164                                  size_t count, loff_t *offset)
 165 {
 166         unsigned long val;
 167         int retval;
 168
 169         if (*offset)
 170                 return -EINVAL;
 171         retval = oprofilefs_ulong_from_user(&val, buf, count);
 172         if (retval <= 0)
 173                 return retval;
 174         if (val < oprofile_min_interval)
 175                 oprofile_hw_interval = oprofile_min_interval;
 176         else if (val > oprofile_max_interval)
 177                 oprofile_hw_interval = oprofile_max_interval;
 178         else
 179                 oprofile_hw_interval = val;
 180
 181         return count;
 182 }
 183
 184 static const struct file_operations hw_interval_fops = {
 185         .read           = hw_interval_read,
 186         .write          = hw_interval_write,
 187 };
 188
 189 /*
 190  * File ops used for:
 191  * /dev/oprofile/0/event
 192  * Only a single event with number 0 is supported with this counter.
 193  *
 194  * /dev/oprofile/0/unit_mask
 195  * This is a dummy file needed by the user space tools.
 196  * No value other than 0 is accepted or returned.
 197  */
 198
 199 static ssize_t hwsampler_zero_read(struct file *file, char __user *buf,
 200                                     size_t count, loff_t *offset)
 201 {
 202         return oprofilefs_ulong_to_user(0, buf, count, offset);
 203 }
 204
 205 static ssize_t hwsampler_zero_write(struct file *file, char const __user *buf,
 206                                      size_t count, loff_t *offset)
 207 {
 208         unsigned long val;
 209         int retval;
 210
 211         if (*offset)
 212                 return -EINVAL;
 213
 214         retval = oprofilefs_ulong_from_user(&val, buf, count);
 215         if (retval <= 0)
 216                 return retval;
 217         if (val != 0)
 218                 return -EINVAL;
 219         return count;
 220 }
 221
 222 static const struct file_operations zero_fops = {
 223         .read           = hwsampler_zero_read,
 224         .write          = hwsampler_zero_write,
 225 };
 226
 227 /* /dev/oprofile/0/kernel file ops.  */
 228
 229 static ssize_t hwsampler_kernel_read(struct file *file, char __user *buf,
 230                                      size_t count, loff_t *offset)
 231 {
 232         return oprofilefs_ulong_to_user(counter_config.kernel,
 233                                         buf, count, offset);
 234 }
 235
 236 static ssize_t hwsampler_kernel_write(struct file *file, char const __user *buf,
 237                                       size_t count, loff_t *offset)
 238 {
 239         unsigned long val;
 240         int retval;
 241
 242         if (*offset)
 243                 return -EINVAL;
 244
 245         retval = oprofilefs_ulong_from_user(&val, buf, count);
 246         if (retval <= 0)
 247                 return retval;
 248
 249         if (val != 0 && val != 1)
 250                 return -EINVAL;
 251
 252         counter_config.kernel = val;
 253
 254         return count;
 255 }
 256
 257 static const struct file_operations kernel_fops = {
 258         .read           = hwsampler_kernel_read,
 259         .write          = hwsampler_kernel_write,
 260 };
 261
 262 /* /dev/oprofile/0/user file ops. */
 263
 264 static ssize_t hwsampler_user_read(struct file *file, char __user *buf,
 265                                    size_t count, loff_t *offset)
 266 {
 267         return oprofilefs_ulong_to_user(counter_config.user,
 268                                         buf, count, offset);
 269 }
 270
 271 static ssize_t hwsampler_user_write(struct file *file, char const __user *buf,
 272                                     size_t count, loff_t *offset)
 273 {
 274         unsigned long val;
 275         int retval;
 276
 277         if (*offset)
 278                 return -EINVAL;
 279
 280         retval = oprofilefs_ulong_from_user(&val, buf, count);
 281         if (retval <= 0)
 282                 return retval;
 283
 284         if (val != 0 && val != 1)
 285                 return -EINVAL;
 286
 287         counter_config.user = val;
 288
 289         return count;
 290 }
 291
 292 static const struct file_operations user_fops = {
 293         .read           = hwsampler_user_read,
 294         .write          = hwsampler_user_write,
 295 };
 296
 297
 298 /*
 299  * File ops used for: /dev/oprofile/timer/enabled
 300  * The value always has to be the inverted value of hwsampler_enabled. So
 301  * no separate variable is created. That way we do not need locking.
 302  */
 303
 304 static ssize_t timer_enabled_read(struct file *file, char __user *buf,
 305                                   size_t count, loff_t *offset)
 306 {
 307         return oprofilefs_ulong_to_user(!hwsampler_enabled, buf, count, offset);
 308 }
 309
 310 static ssize_t timer_enabled_write(struct file *file, char const __user *buf,
 311                                    size_t count, loff_t *offset)
 312 {
 313         unsigned long val;
 314         int retval;
 315
 316         if (*offset)
 317                 return -EINVAL;
 318
 319         retval = oprofilefs_ulong_from_user(&val, buf, count);
 320         if (retval <= 0)
 321                 return retval;
 322
 323         if (val != 0 && val != 1)
 324                 return -EINVAL;
 325
 326         /* Timer cannot be disabled without having hardware sampling.  */
 327         if (val == 0 && !hwsampler_available)
 328                 return -EINVAL;
 329
 330         if (oprofile_started)
 331                 /*
 332                  * save to do without locking as we set
 333                  * hwsampler_running in start() when start_mutex is
 334                  * held
 335                  */
 336                 return -EBUSY;
 337
 338         hwsampler_enabled = !val;
 339
 340         return count;
 341 }
 342
 343 static const struct file_operations timer_enabled_fops = {
 344         .read           = timer_enabled_read,
 345         .write          = timer_enabled_write,
 346 };
 347
 348
 349 static int oprofile_create_hwsampling_files(struct super_block *sb,
 350                                             struct dentry *root)
 351 {
 352         struct dentry *dir;
 353
 354         dir = oprofilefs_mkdir(sb, root, "timer");
 355         if (!dir)
 356                 return -EINVAL;
 357
 358         oprofilefs_create_file(sb, dir, "enabled", &timer_enabled_fops);
 359
 360         if (!hwsampler_available)
 361                 return 0;
 362
 363         /* reinitialize default values */
 364         hwsampler_enabled = 1;
 365         counter_config.kernel = 1;
 366         counter_config.user = 1;
 367
 368         if (!force_cpu_type) {
 369                 /*
 370                  * Create the counter file system.  A single virtual
 371                  * counter is created which can be used to
 372                  * enable/disable hardware sampling dynamically from
 373                  * user space.  The user space will configure a single
 374                  * counter with a single event.  The value of 'event'
 375                  * and 'unit_mask' are not evaluated by the kernel code
 376                  * and can only be set to 0.
 377                  */
 378
 379                 dir = oprofilefs_mkdir(sb, root, "0");
 380                 if (!dir)
 381                         return -EINVAL;
 382
 383                 oprofilefs_create_file(sb, dir, "enabled", &hwsampler_fops);
 384                 oprofilefs_create_file(sb, dir, "event", &zero_fops);
 385                 oprofilefs_create_file(sb, dir, "count", &hw_interval_fops);
 386                 oprofilefs_create_file(sb, dir, "unit_mask", &zero_fops);
 387                 oprofilefs_create_file(sb, dir, "kernel", &kernel_fops);
 388                 oprofilefs_create_file(sb, dir, "user", &user_fops);
 389                 oprofilefs_create_ulong(sb, dir, "hw_sdbt_blocks",
 390                                         &oprofile_sdbt_blocks);
 391
 392         } else {
 393                 /*
 394                  * Hardware sampling can be used but the cpu_type is
 395                  * forced to timer in order to deal with legacy user
 396                  * space tools.  The /dev/oprofile/hwsampling fs is
 397                  * provided in that case.
 398                  */
 399                 dir = oprofilefs_mkdir(sb, root, "hwsampling");
 400                 if (!dir)
 401                         return -EINVAL;
 402
 403                 oprofilefs_create_file(sb, dir, "hwsampler",
 404                                        &hwsampler_fops);
 405                 oprofilefs_create_file(sb, dir, "hw_interval",
 406                                        &hw_interval_fops);
 407                 oprofilefs_create_ro_ulong(sb, dir, "hw_min_interval",
 408                                            &oprofile_min_interval);
 409                 oprofilefs_create_ro_ulong(sb, dir, "hw_max_interval",
 410                                            &oprofile_max_interval);
 411                 oprofilefs_create_ulong(sb, dir, "hw_sdbt_blocks",
 412                                         &oprofile_sdbt_blocks);
 413         }
 414         return 0;
 415 }
 416
 417 static int oprofile_hwsampler_init(struct oprofile_operations *ops)
 418 {
 419         /*
 420          * Initialize the timer mode infrastructure as well in order
 421          * to be able to switch back dynamically.  oprofile_timer_init
 422          * is not supposed to fail.
 423          */
 424         if (oprofile_timer_init(ops))
 425                 BUG();
 426
 427         memcpy(&timer_ops, ops, sizeof(timer_ops));
 428         ops->create_files = oprofile_create_hwsampling_files;
 429
 430         /*
 431          * If the user space tools do not support newer cpu types,
 432          * the force_cpu_type module parameter
 433          * can be used to always return \"timer\" as cpu type.
 434          */
 435         if (force_cpu_type != timer) {
 436                 struct cpuid id;
 437
 438                 get_cpu_id (&id);
 439
 440                 switch (id.machine) {
 441                 case 0x2097: case 0x2098: ops->cpu_type = "s390/z10"; break;
 442                 case 0x2817: case 0x2818: ops->cpu_type = "s390/z196"; break;
 443                 default: return -ENODEV;
 444                 }
 445         }
 446
 447         if (hwsampler_setup())
 448                 return -ENODEV;
 449
 450         /*
 451          * Query the range for the sampling interval from the
 452          * hardware.
 453          */
 454         oprofile_min_interval = hwsampler_query_min_interval();
 455         if (oprofile_min_interval == 0)
 456                 return -ENODEV;
 457         oprofile_max_interval = hwsampler_query_max_interval();
 458         if (oprofile_max_interval == 0)
 459                 return -ENODEV;
 460
 461         /* The initial value should be sane */
 462         if (oprofile_hw_interval < oprofile_min_interval)
 463                 oprofile_hw_interval = oprofile_min_interval;
 464         if (oprofile_hw_interval > oprofile_max_interval)
 465                 oprofile_hw_interval = oprofile_max_interval;
 466
 467         printk(KERN_INFO "oprofile: System z hardware sampling "
 468                "facility found.\n");
 469
 470         ops->start = oprofile_hwsampler_start;
 471         ops->stop = oprofile_hwsampler_stop;
 472
 473         return 0;
 474 }
 475
 476 static void oprofile_hwsampler_exit(void)
 477 {
 478         hwsampler_shutdown();
 479 }
 480
 481 #endif /* CONFIG_64BIT */
 482
 483 int __init oprofile_arch_init(struct oprofile_operations *ops)
 484 {
 485         ops->backtrace = s390_backtrace;
 486
 487 #ifdef CONFIG_64BIT
 488
 489         /*
 490          * -ENODEV is not reported to the caller.  The module itself
 491          * will use the timer mode sampling as fallback and this is
 492          * always available.
 493          */
 494         hwsampler_available = oprofile_hwsampler_init(ops) == 0;
 495
 496         return 0;
 497 #else
 498         return -ENODEV;
 499 #endif
 500 }
 501
 502 void oprofile_arch_exit(void)
 503 {
 504 #ifdef CONFIG_64BIT
 505         oprofile_hwsampler_exit();
 506 #endif
 507 }