arch/s390/kernel/vtime.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  *    Virtual cpu timer based timer functions.
   4  *
   5  *    Copyright IBM Corp. 2004, 2012
   6  *    Author(s): Jan Glauber <jan.glauber@de.ibm.com>
   7  */
   8
   9 #include <linux/kernel_stat.h>
  10 #include <linux/sched/cputime.h>
  11 #include <linux/export.h>
  12 #include <linux/kernel.h>
  13 #include <linux/timex.h>
  14 #include <linux/types.h>
  15 #include <linux/time.h>
  16
  17 #include <asm/vtimer.h>
  18 #include <asm/vtime.h>
  19 #include <asm/cpu_mf.h>
  20 #include <asm/smp.h>
  21
  22 #include "entry.h"
  23
  24 static void virt_timer_expire(void);
  25
  26 static LIST_HEAD(virt_timer_list);
  27 static DEFINE_SPINLOCK(virt_timer_lock);
  28 static atomic64_t virt_timer_current;
  29 static atomic64_t virt_timer_elapsed;
  30
  31 DEFINE_PER_CPU(u64, mt_cycles[8]);
  32 static DEFINE_PER_CPU(u64, mt_scaling_mult) = { 1 };
  33 static DEFINE_PER_CPU(u64, mt_scaling_div) = { 1 };
  34 static DEFINE_PER_CPU(u64, mt_scaling_jiffies);
  35
  36 static inline u64 get_vtimer(void)
  37 {
  38         u64 timer;
  39
  40         asm volatile("stpt %0" : "=Q" (timer));
  41         return timer;
  42 }
  43
  44 static inline void set_vtimer(u64 expires)
  45 {
  46         u64 timer;
  47
  48         asm volatile(
  49                 "       stpt    %0\n"   /* Store current cpu timer value */
  50                 "       spt     %1"     /* Set new value imm. afterwards */
  51                 : "=Q" (timer) : "Q" (expires));
  52         S390_lowcore.system_timer += S390_lowcore.last_update_timer - timer;
  53         S390_lowcore.last_update_timer = expires;
  54 }
  55
  56 static inline int virt_timer_forward(u64 elapsed)
  57 {
  58         BUG_ON(!irqs_disabled());
  59
  60         if (list_empty(&virt_timer_list))
  61                 return 0;
  62         elapsed = atomic64_add_return(elapsed, &virt_timer_elapsed);
  63         return elapsed >= atomic64_read(&virt_timer_current);
  64 }
  65
  66 static void update_mt_scaling(void)
  67 {
  68         u64 cycles_new[8], *cycles_old;
  69         u64 delta, fac, mult, div;
  70         int i;
  71
  72         stcctm(MT_DIAG, smp_cpu_mtid + 1, cycles_new);
  73         cycles_old = this_cpu_ptr(mt_cycles);
  74         fac = 1;
  75         mult = div = 0;
  76         for (i = 0; i <= smp_cpu_mtid; i++) {
  77                 delta = cycles_new[i] - cycles_old[i];
  78                 div += delta;
  79                 mult *= i + 1;
  80                 mult += delta * fac;
  81                 fac *= i + 1;
  82         }
  83         div *= fac;
  84         if (div > 0) {
  85                 /* Update scaling factor */
  86                 __this_cpu_write(mt_scaling_mult, mult);
  87                 __this_cpu_write(mt_scaling_div, div);
  88                 memcpy(cycles_old, cycles_new,
  89                        sizeof(u64) * (smp_cpu_mtid + 1));
  90         }
  91         __this_cpu_write(mt_scaling_jiffies, jiffies_64);
  92 }
  93
  94 static inline u64 update_tsk_timer(unsigned long *tsk_vtime, u64 new)
  95 {
  96         u64 delta;
  97
  98         delta = new - *tsk_vtime;
  99         *tsk_vtime = new;
 100         return delta;
 101 }
 102
 103
 104 static inline u64 scale_vtime(u64 vtime)
 105 {
 106         u64 mult = __this_cpu_read(mt_scaling_mult);
 107         u64 div = __this_cpu_read(mt_scaling_div);
 108
 109         if (smp_cpu_mtid)
 110                 return vtime * mult / div;
 111         return vtime;
 112 }
 113
 114 static void account_system_index_scaled(struct task_struct *p, u64 cputime,
 115                                         enum cpu_usage_stat index)
 116 {
 117         p->stimescaled += cputime_to_nsecs(scale_vtime(cputime));
 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;
 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                 : "=Q" (S390_lowcore.last_update_timer),
 139                   "=Q" (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, CPUTIME_SYSTEM);
 180         if (hardirq)
 181                 account_system_index_scaled(tsk, hardirq, CPUTIME_IRQ);
 182         if (softirq)
 183                 account_system_index_scaled(tsk, softirq, CPUTIME_SOFTIRQ);
 184
 185         return virt_timer_forward(user + guest + system + hardirq + softirq);
 186 }
 187
 188 void vtime_task_switch(struct task_struct *prev)
 189 {
 190         do_account_vtime(prev);
 191         prev->thread.user_timer = S390_lowcore.user_timer;
 192         prev->thread.guest_timer = S390_lowcore.guest_timer;
 193         prev->thread.system_timer = S390_lowcore.system_timer;
 194         prev->thread.hardirq_timer = S390_lowcore.hardirq_timer;
 195         prev->thread.softirq_timer = S390_lowcore.softirq_timer;
 196         S390_lowcore.user_timer = current->thread.user_timer;
 197         S390_lowcore.guest_timer = current->thread.guest_timer;
 198         S390_lowcore.system_timer = current->thread.system_timer;
 199         S390_lowcore.hardirq_timer = current->thread.hardirq_timer;
 200         S390_lowcore.softirq_timer = current->thread.softirq_timer;
 201 }
 202
 203 /*
 204  * In s390, accounting pending user time also implies
 205  * accounting system time in order to correctly compute
 206  * the stolen time accounting.
 207  */
 208 void vtime_flush(struct task_struct *tsk)
 209 {
 210         u64 steal, avg_steal;
 211
 212         if (do_account_vtime(tsk))
 213                 virt_timer_expire();
 214
 215         steal = S390_lowcore.steal_timer;
 216         avg_steal = S390_lowcore.avg_steal_timer / 2;
 217         if ((s64) steal > 0) {
 218                 S390_lowcore.steal_timer = 0;
 219                 account_steal_time(steal);
 220                 avg_steal += steal;
 221         }
 222         S390_lowcore.avg_steal_timer = avg_steal;
 223 }
 224
 225 static u64 vtime_delta(void)
 226 {
 227         u64 timer = S390_lowcore.last_update_timer;
 228
 229         S390_lowcore.last_update_timer = get_vtimer();
 230
 231         return timer - S390_lowcore.last_update_timer;
 232 }
 233
 234 /*
 235  * Update process times based on virtual cpu times stored by entry.S
 236  * to the lowcore fields user_timer, system_timer & steal_clock.
 237  */
 238 void vtime_account_kernel(struct task_struct *tsk)
 239 {
 240         u64 delta = vtime_delta();
 241
 242         if (tsk->flags & PF_VCPU)
 243                 S390_lowcore.guest_timer += delta;
 244         else
 245                 S390_lowcore.system_timer += delta;
 246
 247         virt_timer_forward(delta);
 248 }
 249 EXPORT_SYMBOL_GPL(vtime_account_kernel);
 250
 251 void vtime_account_softirq(struct task_struct *tsk)
 252 {
 253         u64 delta = vtime_delta();
 254
 255         S390_lowcore.softirq_timer += delta;
 256
 257         virt_timer_forward(delta);
 258 }
 259
 260 void vtime_account_hardirq(struct task_struct *tsk)
 261 {
 262         u64 delta = vtime_delta();
 263
 264         S390_lowcore.hardirq_timer += delta;
 265
 266         virt_timer_forward(delta);
 267 }
 268
 269 /*
 270  * Sorted add to a list. List is linear searched until first bigger
 271  * element is found.
 272  */
 273 static void list_add_sorted(struct vtimer_list *timer, struct list_head *head)
 274 {
 275         struct vtimer_list *tmp;
 276
 277         list_for_each_entry(tmp, head, entry) {
 278                 if (tmp->expires > timer->expires) {
 279                         list_add_tail(&timer->entry, &tmp->entry);
 280                         return;
 281                 }
 282         }
 283         list_add_tail(&timer->entry, head);
 284 }
 285
 286 /*
 287  * Handler for expired virtual CPU timer.
 288  */
 289 static void virt_timer_expire(void)
 290 {
 291         struct vtimer_list *timer, *tmp;
 292         unsigned long elapsed;
 293         LIST_HEAD(cb_list);
 294
 295         /* walk timer list, fire all expired timers */
 296         spin_lock(&virt_timer_lock);
 297         elapsed = atomic64_read(&virt_timer_elapsed);
 298         list_for_each_entry_safe(timer, tmp, &virt_timer_list, entry) {
 299                 if (timer->expires < elapsed)
 300                         /* move expired timer to the callback queue */
 301                         list_move_tail(&timer->entry, &cb_list);
 302                 else
 303                         timer->expires -= elapsed;
 304         }
 305         if (!list_empty(&virt_timer_list)) {
 306                 timer = list_first_entry(&virt_timer_list,
 307                                          struct vtimer_list, entry);
 308                 atomic64_set(&virt_timer_current, timer->expires);
 309         }
 310         atomic64_sub(elapsed, &virt_timer_elapsed);
 311         spin_unlock(&virt_timer_lock);
 312
 313         /* Do callbacks and recharge periodic timers */
 314         list_for_each_entry_safe(timer, tmp, &cb_list, entry) {
 315                 list_del_init(&timer->entry);
 316                 timer->function(timer->data);
 317                 if (timer->interval) {
 318                         /* Recharge interval timer */
 319                         timer->expires = timer->interval +
 320                                 atomic64_read(&virt_timer_elapsed);
 321                         spin_lock(&virt_timer_lock);
 322                         list_add_sorted(timer, &virt_timer_list);
 323                         spin_unlock(&virt_timer_lock);
 324                 }
 325         }
 326 }
 327
 328 void init_virt_timer(struct vtimer_list *timer)
 329 {
 330         timer->function = NULL;
 331         INIT_LIST_HEAD(&timer->entry);
 332 }
 333 EXPORT_SYMBOL(init_virt_timer);
 334
 335 static inline int vtimer_pending(struct vtimer_list *timer)
 336 {
 337         return !list_empty(&timer->entry);
 338 }
 339
 340 static void internal_add_vtimer(struct vtimer_list *timer)
 341 {
 342         if (list_empty(&virt_timer_list)) {
 343                 /* First timer, just program it. */
 344                 atomic64_set(&virt_timer_current, timer->expires);
 345                 atomic64_set(&virt_timer_elapsed, 0);
 346                 list_add(&timer->entry, &virt_timer_list);
 347         } else {
 348                 /* Update timer against current base. */
 349                 timer->expires += atomic64_read(&virt_timer_elapsed);
 350                 if (likely((s64) timer->expires <
 351                            (s64) atomic64_read(&virt_timer_current)))
 352                         /* The new timer expires before the current timer. */
 353                         atomic64_set(&virt_timer_current, timer->expires);
 354                 /* Insert new timer into the list. */
 355                 list_add_sorted(timer, &virt_timer_list);
 356         }
 357 }
 358
 359 static void __add_vtimer(struct vtimer_list *timer, int periodic)
 360 {
 361         unsigned long flags;
 362
 363         timer->interval = periodic ? timer->expires : 0;
 364         spin_lock_irqsave(&virt_timer_lock, flags);
 365         internal_add_vtimer(timer);
 366         spin_unlock_irqrestore(&virt_timer_lock, flags);
 367 }
 368
 369 /*
 370  * add_virt_timer - add a oneshot virtual CPU timer
 371  */
 372 void add_virt_timer(struct vtimer_list *timer)
 373 {
 374         __add_vtimer(timer, 0);
 375 }
 376 EXPORT_SYMBOL(add_virt_timer);
 377
 378 /*
 379  * add_virt_timer_int - add an interval virtual CPU timer
 380  */
 381 void add_virt_timer_periodic(struct vtimer_list *timer)
 382 {
 383         __add_vtimer(timer, 1);
 384 }
 385 EXPORT_SYMBOL(add_virt_timer_periodic);
 386
 387 static int __mod_vtimer(struct vtimer_list *timer, u64 expires, int periodic)
 388 {
 389         unsigned long flags;
 390         int rc;
 391
 392         BUG_ON(!timer->function);
 393
 394         if (timer->expires == expires && vtimer_pending(timer))
 395                 return 1;
 396         spin_lock_irqsave(&virt_timer_lock, flags);
 397         rc = vtimer_pending(timer);
 398         if (rc)
 399                 list_del_init(&timer->entry);
 400         timer->interval = periodic ? expires : 0;
 401         timer->expires = expires;
 402         internal_add_vtimer(timer);
 403         spin_unlock_irqrestore(&virt_timer_lock, flags);
 404         return rc;
 405 }
 406
 407 /*
 408  * returns whether it has modified a pending timer (1) or not (0)
 409  */
 410 int mod_virt_timer(struct vtimer_list *timer, u64 expires)
 411 {
 412         return __mod_vtimer(timer, expires, 0);
 413 }
 414 EXPORT_SYMBOL(mod_virt_timer);
 415
 416 /*
 417  * returns whether it has modified a pending timer (1) or not (0)
 418  */
 419 int mod_virt_timer_periodic(struct vtimer_list *timer, u64 expires)
 420 {
 421         return __mod_vtimer(timer, expires, 1);
 422 }
 423 EXPORT_SYMBOL(mod_virt_timer_periodic);
 424
 425 /*
 426  * Delete a virtual timer.
 427  *
 428  * returns whether the deleted timer was pending (1) or not (0)
 429  */
 430 int del_virt_timer(struct vtimer_list *timer)
 431 {
 432         unsigned long flags;
 433
 434         if (!vtimer_pending(timer))
 435                 return 0;
 436         spin_lock_irqsave(&virt_timer_lock, flags);
 437         list_del_init(&timer->entry);
 438         spin_unlock_irqrestore(&virt_timer_lock, flags);
 439         return 1;
 440 }
 441 EXPORT_SYMBOL(del_virt_timer);
 442
 443 /*
 444  * Start the virtual CPU timer on the current CPU.
 445  */
 446 void vtime_init(void)
 447 {
 448         /* set initial cpu timer */
 449         set_vtimer(VTIMER_MAX_SLICE);
 450         /* Setup initial MT scaling values */
 451         if (smp_cpu_mtid) {
 452                 __this_cpu_write(mt_scaling_jiffies, jiffies);
 453                 __this_cpu_write(mt_scaling_mult, 1);
 454                 __this_cpu_write(mt_scaling_div, 1);
 455                 stcctm(MT_DIAG, smp_cpu_mtid + 1, this_cpu_ptr(mt_cycles));
 456         }
 457 }