arch/s390/kernel/vtime.c

   1 /*
   2  *  arch/s390/kernel/vtime.c
   3  *    Virtual cpu timer based timer functions.
   4  *
   5  *  S390 version
   6  *    Copyright (C) 2004 IBM Deutschland Entwicklung GmbH, IBM Corporation
   7  *    Author(s): Jan Glauber <jan.glauber@de.ibm.com>
   8  */
   9
  10 #include <linux/module.h>
  11 #include <linux/kernel.h>
  12 #include <linux/time.h>
  13 #include <linux/delay.h>
  14 #include <linux/init.h>
  15 #include <linux/smp.h>
  16 #include <linux/types.h>
  17 #include <linux/timex.h>
  18 #include <linux/notifier.h>
  19 #include <linux/kernel_stat.h>
  20 #include <linux/rcupdate.h>
  21 #include <linux/posix-timers.h>
  22
  23 #include <asm/s390_ext.h>
  24 #include <asm/timer.h>
  25 #include <asm/irq_regs.h>
  26
  27 static ext_int_info_t ext_int_info_timer;
  28 static DEFINE_PER_CPU(struct vtimer_queue, virt_cpu_timer);
  29
  30 #ifdef CONFIG_VIRT_CPU_ACCOUNTING
  31 /*
  32  * Update process times based on virtual cpu times stored by entry.S
  33  * to the lowcore fields user_timer, system_timer & steal_clock.
  34  */
  35 void account_tick_vtime(struct task_struct *tsk)
  36 {
  37         cputime_t cputime;
  38         __u64 timer, clock;
  39         int rcu_user_flag;
  40
  41         timer = S390_lowcore.last_update_timer;
  42         clock = S390_lowcore.last_update_clock;
  43         asm volatile ("  STPT %0\n"    /* Store current cpu timer value */
  44                       "  STCK %1"      /* Store current tod clock value */
  45                       : "=m" (S390_lowcore.last_update_timer),
  46                         "=m" (S390_lowcore.last_update_clock) );
  47         S390_lowcore.system_timer += timer - S390_lowcore.last_update_timer;
  48         S390_lowcore.steal_clock += S390_lowcore.last_update_clock - clock;
  49
  50         cputime = S390_lowcore.user_timer >> 12;
  51         rcu_user_flag = cputime != 0;
  52         S390_lowcore.user_timer -= cputime << 12;
  53         S390_lowcore.steal_clock -= cputime << 12;
  54         account_user_time(tsk, cputime);
  55
  56         cputime =  S390_lowcore.system_timer >> 12;
  57         S390_lowcore.system_timer -= cputime << 12;
  58         S390_lowcore.steal_clock -= cputime << 12;
  59         account_system_time(tsk, HARDIRQ_OFFSET, cputime);
  60
  61         cputime = S390_lowcore.steal_clock;
  62         if ((__s64) cputime > 0) {
  63                 cputime >>= 12;
  64                 S390_lowcore.steal_clock -= cputime << 12;
  65                 account_steal_time(tsk, cputime);
  66         }
  67
  68         run_local_timers();
  69         if (rcu_pending(smp_processor_id()))
  70                 rcu_check_callbacks(smp_processor_id(), rcu_user_flag);
  71         scheduler_tick();
  72         run_posix_cpu_timers(tsk);
  73 }
  74
  75 /*
  76  * Update process times based on virtual cpu times stored by entry.S
  77  * to the lowcore fields user_timer, system_timer & steal_clock.
  78  */
  79 void account_vtime(struct task_struct *tsk)
  80 {
  81         cputime_t cputime;
  82         __u64 timer;
  83
  84         timer = S390_lowcore.last_update_timer;
  85         asm volatile ("  STPT %0"    /* Store current cpu timer value */
  86                       : "=m" (S390_lowcore.last_update_timer) );
  87         S390_lowcore.system_timer += timer - S390_lowcore.last_update_timer;
  88
  89         cputime = S390_lowcore.user_timer >> 12;
  90         S390_lowcore.user_timer -= cputime << 12;
  91         S390_lowcore.steal_clock -= cputime << 12;
  92         account_user_time(tsk, cputime);
  93
  94         cputime =  S390_lowcore.system_timer >> 12;
  95         S390_lowcore.system_timer -= cputime << 12;
  96         S390_lowcore.steal_clock -= cputime << 12;
  97         account_system_time(tsk, 0, cputime);
  98 }
  99
 100 /*
 101  * Update process times based on virtual cpu times stored by entry.S
 102  * to the lowcore fields user_timer, system_timer & steal_clock.
 103  */
 104 void account_system_vtime(struct task_struct *tsk)
 105 {
 106         cputime_t cputime;
 107         __u64 timer;
 108
 109         timer = S390_lowcore.last_update_timer;
 110         asm volatile ("  STPT %0"    /* Store current cpu timer value */
 111                       : "=m" (S390_lowcore.last_update_timer) );
 112         S390_lowcore.system_timer += timer - S390_lowcore.last_update_timer;
 113
 114         cputime =  S390_lowcore.system_timer >> 12;
 115         S390_lowcore.system_timer -= cputime << 12;
 116         S390_lowcore.steal_clock -= cputime << 12;
 117         account_system_time(tsk, 0, cputime);
 118 }
 119
 120 static inline void set_vtimer(__u64 expires)
 121 {
 122         __u64 timer;
 123
 124         asm volatile ("  STPT %0\n"  /* Store current cpu timer value */
 125                       "  SPT %1"     /* Set new value immediatly afterwards */
 126                       : "=m" (timer) : "m" (expires) );
 127         S390_lowcore.system_timer += S390_lowcore.last_update_timer - timer;
 128         S390_lowcore.last_update_timer = expires;
 129
 130         /* store expire time for this CPU timer */
 131         __get_cpu_var(virt_cpu_timer).to_expire = expires;
 132 }
 133 #else
 134 static inline void set_vtimer(__u64 expires)
 135 {
 136         S390_lowcore.last_update_timer = expires;
 137         asm volatile ("SPT %0" : : "m" (S390_lowcore.last_update_timer));
 138
 139         /* store expire time for this CPU timer */
 140         __get_cpu_var(virt_cpu_timer).to_expire = expires;
 141 }
 142 #endif
 143
 144 static void start_cpu_timer(void)
 145 {
 146         struct vtimer_queue *vt_list;
 147
 148         vt_list = &__get_cpu_var(virt_cpu_timer);
 149
 150         /* CPU timer interrupt is pending, don't reprogramm it */
 151         if (vt_list->idle & 1LL<<63)
 152                 return;
 153
 154         if (!list_empty(&vt_list->list))
 155                 set_vtimer(vt_list->idle);
 156 }
 157
 158 static void stop_cpu_timer(void)
 159 {
 160         struct vtimer_queue *vt_list;
 161
 162         vt_list = &__get_cpu_var(virt_cpu_timer);
 163
 164         /* nothing to do */
 165         if (list_empty(&vt_list->list)) {
 166                 vt_list->idle = VTIMER_MAX_SLICE;
 167                 goto fire;
 168         }
 169
 170         /* store the actual expire value */
 171         asm volatile ("STPT %0" : "=m" (vt_list->idle));
 172
 173         /*
 174          * If the CPU timer is negative we don't reprogramm
 175          * it because we will get instantly an interrupt.
 176          */
 177         if (vt_list->idle & 1LL<<63)
 178                 return;
 179
 180         vt_list->offset += vt_list->to_expire - vt_list->idle;
 181
 182         /*
 183          * We cannot halt the CPU timer, we just write a value that
 184          * nearly never expires (only after 71 years) and re-write
 185          * the stored expire value if we continue the timer
 186          */
 187  fire:
 188         set_vtimer(VTIMER_MAX_SLICE);
 189 }
 190
 191 /*
 192  * Sorted add to a list. List is linear searched until first bigger
 193  * element is found.
 194  */
 195 static void list_add_sorted(struct vtimer_list *timer, struct list_head *head)
 196 {
 197         struct vtimer_list *event;
 198
 199         list_for_each_entry(event, head, entry) {
 200                 if (event->expires > timer->expires) {
 201                         list_add_tail(&timer->entry, &event->entry);
 202                         return;
 203                 }
 204         }
 205         list_add_tail(&timer->entry, head);
 206 }
 207
 208 /*
 209  * Do the callback functions of expired vtimer events.
 210  * Called from within the interrupt handler.
 211  */
 212 static void do_callbacks(struct list_head *cb_list)
 213 {
 214         struct vtimer_queue *vt_list;
 215         struct vtimer_list *event, *tmp;
 216         void (*fn)(unsigned long);
 217         unsigned long data;
 218
 219         if (list_empty(cb_list))
 220                 return;
 221
 222         vt_list = &__get_cpu_var(virt_cpu_timer);
 223
 224         list_for_each_entry_safe(event, tmp, cb_list, entry) {
 225                 fn = event->function;
 226                 data = event->data;
 227                 fn(data);
 228
 229                 if (!event->interval)
 230                         /* delete one shot timer */
 231                         list_del_init(&event->entry);
 232                 else {
 233                         /* move interval timer back to list */
 234                         spin_lock(&vt_list->lock);
 235                         list_del_init(&event->entry);
 236                         list_add_sorted(event, &vt_list->list);
 237                         spin_unlock(&vt_list->lock);
 238                 }
 239         }
 240 }
 241
 242 /*
 243  * Handler for the virtual CPU timer.
 244  */
 245 static void do_cpu_timer_interrupt(__u16 error_code)
 246 {
 247         __u64 next, delta;
 248         struct vtimer_queue *vt_list;
 249         struct vtimer_list *event, *tmp;
 250         struct list_head *ptr;
 251         /* the callback queue */
 252         struct list_head cb_list;
 253
 254         INIT_LIST_HEAD(&cb_list);
 255         vt_list = &__get_cpu_var(virt_cpu_timer);
 256
 257         /* walk timer list, fire all expired events */
 258         spin_lock(&vt_list->lock);
 259
 260         if (vt_list->to_expire < VTIMER_MAX_SLICE)
 261                 vt_list->offset += vt_list->to_expire;
 262
 263         list_for_each_entry_safe(event, tmp, &vt_list->list, entry) {
 264                 if (event->expires > vt_list->offset)
 265                         /* found first unexpired event, leave */
 266                         break;
 267
 268                 /* re-charge interval timer, we have to add the offset */
 269                 if (event->interval)
 270                         event->expires = event->interval + vt_list->offset;
 271
 272                 /* move expired timer to the callback queue */
 273                 list_move_tail(&event->entry, &cb_list);
 274         }
 275         spin_unlock(&vt_list->lock);
 276         do_callbacks(&cb_list);
 277
 278         /* next event is first in list */
 279         spin_lock(&vt_list->lock);
 280         if (!list_empty(&vt_list->list)) {
 281                 ptr = vt_list->list.next;
 282                 event = list_entry(ptr, struct vtimer_list, entry);
 283                 next = event->expires - vt_list->offset;
 284
 285                 /* add the expired time from this interrupt handler
 286                  * and the callback functions
 287                  */
 288                 asm volatile ("STPT %0" : "=m" (delta));
 289                 delta = 0xffffffffffffffffLL - delta + 1;
 290                 vt_list->offset += delta;
 291                 next -= delta;
 292         } else {
 293                 vt_list->offset = 0;
 294                 next = VTIMER_MAX_SLICE;
 295         }
 296         spin_unlock(&vt_list->lock);
 297         set_vtimer(next);
 298 }
 299
 300 void init_virt_timer(struct vtimer_list *timer)
 301 {
 302         timer->function = NULL;
 303         INIT_LIST_HEAD(&timer->entry);
 304         spin_lock_init(&timer->lock);
 305 }
 306 EXPORT_SYMBOL(init_virt_timer);
 307
 308 static inline int vtimer_pending(struct vtimer_list *timer)
 309 {
 310         return (!list_empty(&timer->entry));
 311 }
 312
 313 /*
 314  * this function should only run on the specified CPU
 315  */
 316 static void internal_add_vtimer(struct vtimer_list *timer)
 317 {
 318         unsigned long flags;
 319         __u64 done;
 320         struct vtimer_list *event;
 321         struct vtimer_queue *vt_list;
 322
 323         vt_list = &per_cpu(virt_cpu_timer, timer->cpu);
 324         spin_lock_irqsave(&vt_list->lock, flags);
 325
 326         if (timer->cpu != smp_processor_id())
 327                 printk("internal_add_vtimer: BUG, running on wrong CPU");
 328
 329         /* if list is empty we only have to set the timer */
 330         if (list_empty(&vt_list->list)) {
 331                 /* reset the offset, this may happen if the last timer was
 332                  * just deleted by mod_virt_timer and the interrupt
 333                  * didn't happen until here
 334                  */
 335                 vt_list->offset = 0;
 336                 goto fire;
 337         }
 338
 339         /* save progress */
 340         asm volatile ("STPT %0" : "=m" (done));
 341
 342         /* calculate completed work */
 343         done = vt_list->to_expire - done + vt_list->offset;
 344         vt_list->offset = 0;
 345
 346         list_for_each_entry(event, &vt_list->list, entry)
 347                 event->expires -= done;
 348
 349  fire:
 350         list_add_sorted(timer, &vt_list->list);
 351
 352         /* get first element, which is the next vtimer slice */
 353         event = list_entry(vt_list->list.next, struct vtimer_list, entry);
 354
 355         set_vtimer(event->expires);
 356         spin_unlock_irqrestore(&vt_list->lock, flags);
 357         /* release CPU acquired in prepare_vtimer or mod_virt_timer() */
 358         put_cpu();
 359 }
 360
 361 static inline int prepare_vtimer(struct vtimer_list *timer)
 362 {
 363         if (!timer->function) {
 364                 printk("add_virt_timer: uninitialized timer\n");
 365                 return -EINVAL;
 366         }
 367
 368         if (!timer->expires || timer->expires > VTIMER_MAX_SLICE) {
 369                 printk("add_virt_timer: invalid timer expire value!\n");
 370                 return -EINVAL;
 371         }
 372
 373         if (vtimer_pending(timer)) {
 374                 printk("add_virt_timer: timer pending\n");
 375                 return -EBUSY;
 376         }
 377
 378         timer->cpu = get_cpu();
 379         return 0;
 380 }
 381
 382 /*
 383  * add_virt_timer - add an oneshot virtual CPU timer
 384  */
 385 void add_virt_timer(void *new)
 386 {
 387         struct vtimer_list *timer;
 388
 389         timer = (struct vtimer_list *)new;
 390
 391         if (prepare_vtimer(timer) < 0)
 392                 return;
 393
 394         timer->interval = 0;
 395         internal_add_vtimer(timer);
 396 }
 397 EXPORT_SYMBOL(add_virt_timer);
 398
 399 /*
 400  * add_virt_timer_int - add an interval virtual CPU timer
 401  */
 402 void add_virt_timer_periodic(void *new)
 403 {
 404         struct vtimer_list *timer;
 405
 406         timer = (struct vtimer_list *)new;
 407
 408         if (prepare_vtimer(timer) < 0)
 409                 return;
 410
 411         timer->interval = timer->expires;
 412         internal_add_vtimer(timer);
 413 }
 414 EXPORT_SYMBOL(add_virt_timer_periodic);
 415
 416 /*
 417  * If we change a pending timer the function must be called on the CPU
 418  * where the timer is running on, e.g. by smp_call_function_single()
 419  *
 420  * The original mod_timer adds the timer if it is not pending. For compatibility
 421  * we do the same. The timer will be added on the current CPU as a oneshot timer.
 422  *
 423  * returns whether it has modified a pending timer (1) or not (0)
 424  */
 425 int mod_virt_timer(struct vtimer_list *timer, __u64 expires)
 426 {
 427         struct vtimer_queue *vt_list;
 428         unsigned long flags;
 429         int cpu;
 430
 431         if (!timer->function) {
 432                 printk("mod_virt_timer: uninitialized timer\n");
 433                 return  -EINVAL;
 434         }
 435
 436         if (!expires || expires > VTIMER_MAX_SLICE) {
 437                 printk("mod_virt_timer: invalid expire range\n");
 438                 return -EINVAL;
 439         }
 440
 441         /*
 442          * This is a common optimization triggered by the
 443          * networking code - if the timer is re-modified
 444          * to be the same thing then just return:
 445          */
 446         if (timer->expires == expires && vtimer_pending(timer))
 447                 return 1;
 448
 449         cpu = get_cpu();
 450         vt_list = &per_cpu(virt_cpu_timer, cpu);
 451
 452         /* disable interrupts before test if timer is pending */
 453         spin_lock_irqsave(&vt_list->lock, flags);
 454
 455         /* if timer isn't pending add it on the current CPU */
 456         if (!vtimer_pending(timer)) {
 457                 spin_unlock_irqrestore(&vt_list->lock, flags);
 458                 /* we do not activate an interval timer with mod_virt_timer */
 459                 timer->interval = 0;
 460                 timer->expires = expires;
 461                 timer->cpu = cpu;
 462                 internal_add_vtimer(timer);
 463                 return 0;
 464         }
 465
 466         /* check if we run on the right CPU */
 467         if (timer->cpu != cpu) {
 468                 printk("mod_virt_timer: running on wrong CPU, check your code\n");
 469                 spin_unlock_irqrestore(&vt_list->lock, flags);
 470                 put_cpu();
 471                 return -EINVAL;
 472         }
 473
 474         list_del_init(&timer->entry);
 475         timer->expires = expires;
 476
 477         /* also change the interval if we have an interval timer */
 478         if (timer->interval)
 479                 timer->interval = expires;
 480
 481         /* the timer can't expire anymore so we can release the lock */
 482         spin_unlock_irqrestore(&vt_list->lock, flags);
 483         internal_add_vtimer(timer);
 484         return 1;
 485 }
 486 EXPORT_SYMBOL(mod_virt_timer);
 487
 488 /*
 489  * delete a virtual timer
 490  *
 491  * returns whether the deleted timer was pending (1) or not (0)
 492  */
 493 int del_virt_timer(struct vtimer_list *timer)
 494 {
 495         unsigned long flags;
 496         struct vtimer_queue *vt_list;
 497
 498         /* check if timer is pending */
 499         if (!vtimer_pending(timer))
 500                 return 0;
 501
 502         vt_list = &per_cpu(virt_cpu_timer, timer->cpu);
 503         spin_lock_irqsave(&vt_list->lock, flags);
 504
 505         /* we don't interrupt a running timer, just let it expire! */
 506         list_del_init(&timer->entry);
 507
 508         /* last timer removed */
 509         if (list_empty(&vt_list->list)) {
 510                 vt_list->to_expire = 0;
 511                 vt_list->offset = 0;
 512         }
 513
 514         spin_unlock_irqrestore(&vt_list->lock, flags);
 515         return 1;
 516 }
 517 EXPORT_SYMBOL(del_virt_timer);
 518
 519 /*
 520  * Start the virtual CPU timer on the current CPU.
 521  */
 522 void init_cpu_vtimer(void)
 523 {
 524         struct vtimer_queue *vt_list;
 525
 526         /* kick the virtual timer */
 527         S390_lowcore.exit_timer = VTIMER_MAX_SLICE;
 528         S390_lowcore.last_update_timer = VTIMER_MAX_SLICE;
 529         asm volatile ("SPT %0" : : "m" (S390_lowcore.last_update_timer));
 530         asm volatile ("STCK %0" : "=m" (S390_lowcore.last_update_clock));
 531
 532         /* enable cpu timer interrupts */
 533         __ctl_set_bit(0,10);
 534
 535         vt_list = &__get_cpu_var(virt_cpu_timer);
 536         INIT_LIST_HEAD(&vt_list->list);
 537         spin_lock_init(&vt_list->lock);
 538         vt_list->to_expire = 0;
 539         vt_list->offset = 0;
 540         vt_list->idle = 0;
 541
 542 }
 543
 544 static int vtimer_idle_notify(struct notifier_block *self,
 545                               unsigned long action, void *hcpu)
 546 {
 547         switch (action) {
 548         case S390_CPU_IDLE:
 549                 stop_cpu_timer();
 550                 break;
 551         case S390_CPU_NOT_IDLE:
 552                 start_cpu_timer();
 553                 break;
 554         }
 555         return NOTIFY_OK;
 556 }
 557
 558 static struct notifier_block vtimer_idle_nb = {
 559         .notifier_call = vtimer_idle_notify,
 560 };
 561
 562 void __init vtime_init(void)
 563 {
 564         /* request the cpu timer external interrupt */
 565         if (register_early_external_interrupt(0x1005, do_cpu_timer_interrupt,
 566                                               &ext_int_info_timer) != 0)
 567                 panic("Couldn't request external interrupt 0x1005");
 568
 569         if (register_idle_notifier(&vtimer_idle_nb))
 570                 panic("Couldn't register idle notifier");
 571
 572         /* Enable cpu timer interrupts on the boot cpu. */
 573         init_cpu_vtimer();
 574 }
 575