arch/s390/kernel/vtime.c

   1 /*
   2  *    Virtual cpu timer based timer functions.
   3  *
   4  *    Copyright IBM Corp. 2004, 2012
   5  *    Author(s): Jan Glauber <jan.glauber@de.ibm.com>
   6  */
   7
   8 #include <linux/kernel_stat.h>
   9 #include <linux/cputime.h>
  10 #include <linux/export.h>
  11 #include <linux/kernel.h>
  12 #include <linux/timex.h>
  13 #include <linux/types.h>
  14 #include <linux/time.h>
  15
  16 #include <asm/vtimer.h>
  17 #include <asm/vtime.h>
  18 #include <asm/cpu_mf.h>
  19 #include <asm/smp.h>
  20
  21 #include "entry.h"
  22
  23 static void virt_timer_expire(void);
  24
  25 static LIST_HEAD(virt_timer_list);
  26 static DEFINE_SPINLOCK(virt_timer_lock);
  27 static atomic64_t virt_timer_current;
  28 static atomic64_t virt_timer_elapsed;
  29
  30 DEFINE_PER_CPU(u64, mt_cycles[8]);
  31 static DEFINE_PER_CPU(u64, mt_scaling_mult) = { 1 };
  32 static DEFINE_PER_CPU(u64, mt_scaling_div) = { 1 };
  33 static DEFINE_PER_CPU(u64, mt_scaling_jiffies);
  34
  35 static inline u64 get_vtimer(void)
  36 {
  37         u64 timer;
  38
  39         asm volatile("stpt %0" : "=m" (timer));
  40         return timer;
  41 }
  42
  43 static inline void set_vtimer(u64 expires)
  44 {
  45         u64 timer;
  46
  47         asm volatile(
  48                 "       stpt    %0\n"   /* Store current cpu timer value */
  49                 "       spt     %1"     /* Set new value imm. afterwards */
  50                 : "=m" (timer) : "m" (expires));
  51         S390_lowcore.system_timer += S390_lowcore.last_update_timer - timer;
  52         S390_lowcore.last_update_timer = expires;
  53 }
  54
  55 static inline int virt_timer_forward(u64 elapsed)
  56 {
  57         BUG_ON(!irqs_disabled());
  58
  59         if (list_empty(&virt_timer_list))
  60                 return 0;
  61         elapsed = atomic64_add_return(elapsed, &virt_timer_elapsed);
  62         return elapsed >= atomic64_read(&virt_timer_current);
  63 }
  64
  65 static void update_mt_scaling(void)
  66 {
  67         u64 cycles_new[8], *cycles_old;
  68         u64 delta, fac, mult, div;
  69         int i;
  70
  71         stcctm5(smp_cpu_mtid + 1, cycles_new);
  72         cycles_old = this_cpu_ptr(mt_cycles);
  73         fac = 1;
  74         mult = div = 0;
  75         for (i = 0; i <= smp_cpu_mtid; i++) {
  76                 delta = cycles_new[i] - cycles_old[i];
  77                 div += delta;
  78                 mult *= i + 1;
  79                 mult += delta * fac;
  80                 fac *= i + 1;
  81         }
  82         div *= fac;
  83         if (div > 0) {
  84                 /* Update scaling factor */
  85                 __this_cpu_write(mt_scaling_mult, mult);
  86                 __this_cpu_write(mt_scaling_div, div);
  87                 memcpy(cycles_old, cycles_new,
  88                        sizeof(u64) * (smp_cpu_mtid + 1));
  89         }
  90         __this_cpu_write(mt_scaling_jiffies, jiffies_64);
  91 }
  92
  93 static inline u64 update_tsk_timer(unsigned long *tsk_vtime, u64 new)
  94 {
  95         u64 delta;
  96
  97         delta = new - *tsk_vtime;
  98         *tsk_vtime = new;
  99         return delta;
 100 }
 101
 102
 103 static inline u64 scale_vtime(u64 vtime)
 104 {
 105         u64 mult = __this_cpu_read(mt_scaling_mult);
 106         u64 div = __this_cpu_read(mt_scaling_div);
 107
 108         if (smp_cpu_mtid)
 109                 return vtime * mult / div;
 110         return vtime;
 111 }
 112
 113 static void account_system_index_scaled(struct task_struct *p,
 114                                         cputime_t cputime, cputime_t scaled,
 115                                         enum cpu_usage_stat index)
 116 {
 117         p->stimescaled += cputime_to_nsecs(scaled);
 118         account_system_index_time(p, cputime_to_nsecs(cputime), index);
 119 }
 120
 121 /*
 122  * Update process times based on virtual cpu times stored by entry.S
 123  * to the lowcore fields user_timer, system_timer & steal_clock.
 124  */
 125 static int do_account_vtime(struct task_struct *tsk)
 126 {
 127         u64 timer, clock, user, guest, system, hardirq, softirq, steal;
 128
 129         timer = S390_lowcore.last_update_timer;
 130         clock = S390_lowcore.last_update_clock;
 131         asm volatile(
 132                 "       stpt    %0\n"   /* Store current cpu timer value */
 133 #ifdef CONFIG_HAVE_MARCH_Z9_109_FEATURES
 134                 "       stckf   %1"     /* Store current tod clock value */
 135 #else
 136                 "       stck    %1"     /* Store current tod clock value */
 137 #endif
 138                 : "=m" (S390_lowcore.last_update_timer),
 139                   "=m" (S390_lowcore.last_update_clock));
 140         clock = S390_lowcore.last_update_clock - clock;
 141         timer -= S390_lowcore.last_update_timer;
 142
 143         if (hardirq_count())
 144                 S390_lowcore.hardirq_timer += timer;
 145         else
 146                 S390_lowcore.system_timer += timer;
 147
 148         /* Update MT utilization calculation */
 149         if (smp_cpu_mtid &&
 150             time_after64(jiffies_64, this_cpu_read(mt_scaling_jiffies)))
 151                 update_mt_scaling();
 152
 153         /* Calculate cputime delta */
 154         user = update_tsk_timer(&tsk->thread.user_timer,
 155                                 READ_ONCE(S390_lowcore.user_timer));
 156         guest = update_tsk_timer(&tsk->thread.guest_timer,
 157                                  READ_ONCE(S390_lowcore.guest_timer));
 158         system = update_tsk_timer(&tsk->thread.system_timer,
 159                                   READ_ONCE(S390_lowcore.system_timer));
 160         hardirq = update_tsk_timer(&tsk->thread.hardirq_timer,
 161                                    READ_ONCE(S390_lowcore.hardirq_timer));
 162         softirq = update_tsk_timer(&tsk->thread.softirq_timer,
 163                                    READ_ONCE(S390_lowcore.softirq_timer));
 164         S390_lowcore.steal_timer +=
 165                 clock - user - guest - system - hardirq - softirq;
 166
 167         /* Push account value */
 168         if (user) {
 169                 account_user_time(tsk, cputime_to_nsecs(user));
 170                 tsk->utimescaled += cputime_to_nsecs(scale_vtime(user));
 171         }
 172
 173         if (guest) {
 174                 account_guest_time(tsk, cputime_to_nsecs(guest));
 175                 tsk->utimescaled += cputime_to_nsecs(scale_vtime(guest));
 176         }
 177
 178         if (system)
 179                 account_system_index_scaled(tsk, system, scale_vtime(system),
 180                                             CPUTIME_SYSTEM);
 181         if (hardirq)
 182                 account_system_index_scaled(tsk, hardirq, scale_vtime(hardirq),
 183                                             CPUTIME_IRQ);
 184         if (softirq)
 185                 account_system_index_scaled(tsk, softirq, scale_vtime(softirq),
 186                                             CPUTIME_SOFTIRQ);
 187
 188         steal = S390_lowcore.steal_timer;
 189         if ((s64) steal > 0) {
 190                 S390_lowcore.steal_timer = 0;
 191                 account_steal_time(cputime_to_nsecs(steal));
 192         }
 193
 194         return virt_timer_forward(user + guest + system + hardirq + softirq);
 195 }
 196
 197 void vtime_task_switch(struct task_struct *prev)
 198 {
 199         do_account_vtime(prev);
 200         prev->thread.user_timer = S390_lowcore.user_timer;
 201         prev->thread.guest_timer = S390_lowcore.guest_timer;
 202         prev->thread.system_timer = S390_lowcore.system_timer;
 203         prev->thread.hardirq_timer = S390_lowcore.hardirq_timer;
 204         prev->thread.softirq_timer = S390_lowcore.softirq_timer;
 205         S390_lowcore.user_timer = current->thread.user_timer;
 206         S390_lowcore.guest_timer = current->thread.guest_timer;
 207         S390_lowcore.system_timer = current->thread.system_timer;
 208         S390_lowcore.hardirq_timer = current->thread.hardirq_timer;
 209         S390_lowcore.softirq_timer = current->thread.softirq_timer;
 210 }
 211
 212 /*
 213  * In s390, accounting pending user time also implies
 214  * accounting system time in order to correctly compute
 215  * the stolen time accounting.
 216  */
 217 void vtime_flush(struct task_struct *tsk)
 218 {
 219         if (do_account_vtime(tsk))
 220                 virt_timer_expire();
 221 }
 222
 223 /*
 224  * Update process times based on virtual cpu times stored by entry.S
 225  * to the lowcore fields user_timer, system_timer & steal_clock.
 226  */
 227 void vtime_account_irq_enter(struct task_struct *tsk)
 228 {
 229         u64 timer;
 230
 231         timer = S390_lowcore.last_update_timer;
 232         S390_lowcore.last_update_timer = get_vtimer();
 233         timer -= S390_lowcore.last_update_timer;
 234
 235         if ((tsk->flags & PF_VCPU) && (irq_count() == 0))
 236                 S390_lowcore.guest_timer += timer;
 237         else if (hardirq_count())
 238                 S390_lowcore.hardirq_timer += timer;
 239         else if (in_serving_softirq())
 240                 S390_lowcore.softirq_timer += timer;
 241         else
 242                 S390_lowcore.system_timer += timer;
 243
 244         virt_timer_forward(timer);
 245 }
 246 EXPORT_SYMBOL_GPL(vtime_account_irq_enter);
 247
 248 void vtime_account_system(struct task_struct *tsk)
 249 __attribute__((alias("vtime_account_irq_enter")));
 250 EXPORT_SYMBOL_GPL(vtime_account_system);
 251
 252 /*
 253  * Sorted add to a list. List is linear searched until first bigger
 254  * element is found.
 255  */
 256 static void list_add_sorted(struct vtimer_list *timer, struct list_head *head)
 257 {
 258         struct vtimer_list *tmp;
 259
 260         list_for_each_entry(tmp, head, entry) {
 261                 if (tmp->expires > timer->expires) {
 262                         list_add_tail(&timer->entry, &tmp->entry);
 263                         return;
 264                 }
 265         }
 266         list_add_tail(&timer->entry, head);
 267 }
 268
 269 /*
 270  * Handler for expired virtual CPU timer.
 271  */
 272 static void virt_timer_expire(void)
 273 {
 274         struct vtimer_list *timer, *tmp;
 275         unsigned long elapsed;
 276         LIST_HEAD(cb_list);
 277
 278         /* walk timer list, fire all expired timers */
 279         spin_lock(&virt_timer_lock);
 280         elapsed = atomic64_read(&virt_timer_elapsed);
 281         list_for_each_entry_safe(timer, tmp, &virt_timer_list, entry) {
 282                 if (timer->expires < elapsed)
 283                         /* move expired timer to the callback queue */
 284                         list_move_tail(&timer->entry, &cb_list);
 285                 else
 286                         timer->expires -= elapsed;
 287         }
 288         if (!list_empty(&virt_timer_list)) {
 289                 timer = list_first_entry(&virt_timer_list,
 290                                          struct vtimer_list, entry);
 291                 atomic64_set(&virt_timer_current, timer->expires);
 292         }
 293         atomic64_sub(elapsed, &virt_timer_elapsed);
 294         spin_unlock(&virt_timer_lock);
 295
 296         /* Do callbacks and recharge periodic timers */
 297         list_for_each_entry_safe(timer, tmp, &cb_list, entry) {
 298                 list_del_init(&timer->entry);
 299                 timer->function(timer->data);
 300                 if (timer->interval) {
 301                         /* Recharge interval timer */
 302                         timer->expires = timer->interval +
 303                                 atomic64_read(&virt_timer_elapsed);
 304                         spin_lock(&virt_timer_lock);
 305                         list_add_sorted(timer, &virt_timer_list);
 306                         spin_unlock(&virt_timer_lock);
 307                 }
 308         }
 309 }
 310
 311 void init_virt_timer(struct vtimer_list *timer)
 312 {
 313         timer->function = NULL;
 314         INIT_LIST_HEAD(&timer->entry);
 315 }
 316 EXPORT_SYMBOL(init_virt_timer);
 317
 318 static inline int vtimer_pending(struct vtimer_list *timer)
 319 {
 320         return !list_empty(&timer->entry);
 321 }
 322
 323 static void internal_add_vtimer(struct vtimer_list *timer)
 324 {
 325         if (list_empty(&virt_timer_list)) {
 326                 /* First timer, just program it. */
 327                 atomic64_set(&virt_timer_current, timer->expires);
 328                 atomic64_set(&virt_timer_elapsed, 0);
 329                 list_add(&timer->entry, &virt_timer_list);
 330         } else {
 331                 /* Update timer against current base. */
 332                 timer->expires += atomic64_read(&virt_timer_elapsed);
 333                 if (likely((s64) timer->expires <
 334                            (s64) atomic64_read(&virt_timer_current)))
 335                         /* The new timer expires before the current timer. */
 336                         atomic64_set(&virt_timer_current, timer->expires);
 337                 /* Insert new timer into the list. */
 338                 list_add_sorted(timer, &virt_timer_list);
 339         }
 340 }
 341
 342 static void __add_vtimer(struct vtimer_list *timer, int periodic)
 343 {
 344         unsigned long flags;
 345
 346         timer->interval = periodic ? timer->expires : 0;
 347         spin_lock_irqsave(&virt_timer_lock, flags);
 348         internal_add_vtimer(timer);
 349         spin_unlock_irqrestore(&virt_timer_lock, flags);
 350 }
 351
 352 /*
 353  * add_virt_timer - add a oneshot virtual CPU timer
 354  */
 355 void add_virt_timer(struct vtimer_list *timer)
 356 {
 357         __add_vtimer(timer, 0);
 358 }
 359 EXPORT_SYMBOL(add_virt_timer);
 360
 361 /*
 362  * add_virt_timer_int - add an interval virtual CPU timer
 363  */
 364 void add_virt_timer_periodic(struct vtimer_list *timer)
 365 {
 366         __add_vtimer(timer, 1);
 367 }
 368 EXPORT_SYMBOL(add_virt_timer_periodic);
 369
 370 static int __mod_vtimer(struct vtimer_list *timer, u64 expires, int periodic)
 371 {
 372         unsigned long flags;
 373         int rc;
 374
 375         BUG_ON(!timer->function);
 376
 377         if (timer->expires == expires && vtimer_pending(timer))
 378                 return 1;
 379         spin_lock_irqsave(&virt_timer_lock, flags);
 380         rc = vtimer_pending(timer);
 381         if (rc)
 382                 list_del_init(&timer->entry);
 383         timer->interval = periodic ? expires : 0;
 384         timer->expires = expires;
 385         internal_add_vtimer(timer);
 386         spin_unlock_irqrestore(&virt_timer_lock, flags);
 387         return rc;
 388 }
 389
 390 /*
 391  * returns whether it has modified a pending timer (1) or not (0)
 392  */
 393 int mod_virt_timer(struct vtimer_list *timer, u64 expires)
 394 {
 395         return __mod_vtimer(timer, expires, 0);
 396 }
 397 EXPORT_SYMBOL(mod_virt_timer);
 398
 399 /*
 400  * returns whether it has modified a pending timer (1) or not (0)
 401  */
 402 int mod_virt_timer_periodic(struct vtimer_list *timer, u64 expires)
 403 {
 404         return __mod_vtimer(timer, expires, 1);
 405 }
 406 EXPORT_SYMBOL(mod_virt_timer_periodic);
 407
 408 /*
 409  * Delete a virtual timer.
 410  *
 411  * returns whether the deleted timer was pending (1) or not (0)
 412  */
 413 int del_virt_timer(struct vtimer_list *timer)
 414 {
 415         unsigned long flags;
 416
 417         if (!vtimer_pending(timer))
 418                 return 0;
 419         spin_lock_irqsave(&virt_timer_lock, flags);
 420         list_del_init(&timer->entry);
 421         spin_unlock_irqrestore(&virt_timer_lock, flags);
 422         return 1;
 423 }
 424 EXPORT_SYMBOL(del_virt_timer);
 425
 426 /*
 427  * Start the virtual CPU timer on the current CPU.
 428  */
 429 void vtime_init(void)
 430 {
 431         /* set initial cpu timer */
 432         set_vtimer(VTIMER_MAX_SLICE);
 433         /* Setup initial MT scaling values */
 434         if (smp_cpu_mtid) {
 435                 __this_cpu_write(mt_scaling_jiffies, jiffies);
 436                 __this_cpu_write(mt_scaling_mult, 1);
 437                 __this_cpu_write(mt_scaling_div, 1);
 438                 stcctm5(smp_cpu_mtid + 1, this_cpu_ptr(mt_cycles));
 439         }
 440 }