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                 : : "cc");
 141         clock = S390_lowcore.last_update_clock - clock;
 142         timer -= S390_lowcore.last_update_timer;
 143
 144         if (hardirq_count())
 145                 S390_lowcore.hardirq_timer += timer;
 146         else
 147                 S390_lowcore.system_timer += timer;
 148
 149         /* Update MT utilization calculation */
 150         if (smp_cpu_mtid &&
 151             time_after64(jiffies_64, this_cpu_read(mt_scaling_jiffies)))
 152                 update_mt_scaling();
 153
 154         /* Calculate cputime delta */
 155         user = update_tsk_timer(&tsk->thread.user_timer,
 156                                 READ_ONCE(S390_lowcore.user_timer));
 157         guest = update_tsk_timer(&tsk->thread.guest_timer,
 158                                  READ_ONCE(S390_lowcore.guest_timer));
 159         system = update_tsk_timer(&tsk->thread.system_timer,
 160                                   READ_ONCE(S390_lowcore.system_timer));
 161         hardirq = update_tsk_timer(&tsk->thread.hardirq_timer,
 162                                    READ_ONCE(S390_lowcore.hardirq_timer));
 163         softirq = update_tsk_timer(&tsk->thread.softirq_timer,
 164                                    READ_ONCE(S390_lowcore.softirq_timer));
 165         S390_lowcore.steal_timer +=
 166                 clock - user - guest - system - hardirq - softirq;
 167
 168         /* Push account value */
 169         if (user) {
 170                 account_user_time(tsk, cputime_to_nsecs(user));
 171                 tsk->utimescaled += cputime_to_nsecs(scale_vtime(user));
 172         }
 173
 174         if (guest) {
 175                 account_guest_time(tsk, cputime_to_nsecs(guest));
 176                 tsk->utimescaled += cputime_to_nsecs(scale_vtime(guest));
 177         }
 178
 179         if (system)
 180                 account_system_index_scaled(tsk, system, CPUTIME_SYSTEM);
 181         if (hardirq)
 182                 account_system_index_scaled(tsk, hardirq, CPUTIME_IRQ);
 183         if (softirq)
 184                 account_system_index_scaled(tsk, softirq, CPUTIME_SOFTIRQ);
 185
 186         return virt_timer_forward(user + guest + system + hardirq + softirq);
 187 }
 188
 189 void vtime_task_switch(struct task_struct *prev)
 190 {
 191         do_account_vtime(prev);
 192         prev->thread.user_timer = S390_lowcore.user_timer;
 193         prev->thread.guest_timer = S390_lowcore.guest_timer;
 194         prev->thread.system_timer = S390_lowcore.system_timer;
 195         prev->thread.hardirq_timer = S390_lowcore.hardirq_timer;
 196         prev->thread.softirq_timer = S390_lowcore.softirq_timer;
 197         S390_lowcore.user_timer = current->thread.user_timer;
 198         S390_lowcore.guest_timer = current->thread.guest_timer;
 199         S390_lowcore.system_timer = current->thread.system_timer;
 200         S390_lowcore.hardirq_timer = current->thread.hardirq_timer;
 201         S390_lowcore.softirq_timer = current->thread.softirq_timer;
 202 }
 203
 204 /*
 205  * In s390, accounting pending user time also implies
 206  * accounting system time in order to correctly compute
 207  * the stolen time accounting.
 208  */
 209 void vtime_flush(struct task_struct *tsk)
 210 {
 211         u64 steal, avg_steal;
 212
 213         if (do_account_vtime(tsk))
 214                 virt_timer_expire();
 215
 216         steal = S390_lowcore.steal_timer;
 217         avg_steal = S390_lowcore.avg_steal_timer;
 218         if ((s64) steal > 0) {
 219                 S390_lowcore.steal_timer = 0;
 220                 account_steal_time(cputime_to_nsecs(steal));
 221                 avg_steal += steal;
 222         }
 223         S390_lowcore.avg_steal_timer = avg_steal / 2;
 224 }
 225
 226 /*
 227  * Update process times based on virtual cpu times stored by entry.S
 228  * to the lowcore fields user_timer, system_timer & steal_clock.
 229  */
 230 void vtime_account_irq_enter(struct task_struct *tsk)
 231 {
 232         u64 timer;
 233
 234         timer = S390_lowcore.last_update_timer;
 235         S390_lowcore.last_update_timer = get_vtimer();
 236         timer -= S390_lowcore.last_update_timer;
 237
 238         if ((tsk->flags & PF_VCPU) && (irq_count() == 0))
 239                 S390_lowcore.guest_timer += timer;
 240         else if (hardirq_count())
 241                 S390_lowcore.hardirq_timer += timer;
 242         else if (in_serving_softirq())
 243                 S390_lowcore.softirq_timer += timer;
 244         else
 245                 S390_lowcore.system_timer += timer;
 246
 247         virt_timer_forward(timer);
 248 }
 249 EXPORT_SYMBOL_GPL(vtime_account_irq_enter);
 250
 251 void vtime_account_system(struct task_struct *tsk)
 252 __attribute__((alias("vtime_account_irq_enter")));
 253 EXPORT_SYMBOL_GPL(vtime_account_system);
 254
 255 /*
 256  * Sorted add to a list. List is linear searched until first bigger
 257  * element is found.
 258  */
 259 static void list_add_sorted(struct vtimer_list *timer, struct list_head *head)
 260 {
 261         struct vtimer_list *tmp;
 262
 263         list_for_each_entry(tmp, head, entry) {
 264                 if (tmp->expires > timer->expires) {
 265                         list_add_tail(&timer->entry, &tmp->entry);
 266                         return;
 267                 }
 268         }
 269         list_add_tail(&timer->entry, head);
 270 }
 271
 272 /*
 273  * Handler for expired virtual CPU timer.
 274  */
 275 static void virt_timer_expire(void)
 276 {
 277         struct vtimer_list *timer, *tmp;
 278         unsigned long elapsed;
 279         LIST_HEAD(cb_list);
 280
 281         /* walk timer list, fire all expired timers */
 282         spin_lock(&virt_timer_lock);
 283         elapsed = atomic64_read(&virt_timer_elapsed);
 284         list_for_each_entry_safe(timer, tmp, &virt_timer_list, entry) {
 285                 if (timer->expires < elapsed)
 286                         /* move expired timer to the callback queue */
 287                         list_move_tail(&timer->entry, &cb_list);
 288                 else
 289                         timer->expires -= elapsed;
 290         }
 291         if (!list_empty(&virt_timer_list)) {
 292                 timer = list_first_entry(&virt_timer_list,
 293                                          struct vtimer_list, entry);
 294                 atomic64_set(&virt_timer_current, timer->expires);
 295         }
 296         atomic64_sub(elapsed, &virt_timer_elapsed);
 297         spin_unlock(&virt_timer_lock);
 298
 299         /* Do callbacks and recharge periodic timers */
 300         list_for_each_entry_safe(timer, tmp, &cb_list, entry) {
 301                 list_del_init(&timer->entry);
 302                 timer->function(timer->data);
 303                 if (timer->interval) {
 304                         /* Recharge interval timer */
 305                         timer->expires = timer->interval +
 306                                 atomic64_read(&virt_timer_elapsed);
 307                         spin_lock(&virt_timer_lock);
 308                         list_add_sorted(timer, &virt_timer_list);
 309                         spin_unlock(&virt_timer_lock);
 310                 }
 311         }
 312 }
 313
 314 void init_virt_timer(struct vtimer_list *timer)
 315 {
 316         timer->function = NULL;
 317         INIT_LIST_HEAD(&timer->entry);
 318 }
 319 EXPORT_SYMBOL(init_virt_timer);
 320
 321 static inline int vtimer_pending(struct vtimer_list *timer)
 322 {
 323         return !list_empty(&timer->entry);
 324 }
 325
 326 static void internal_add_vtimer(struct vtimer_list *timer)
 327 {
 328         if (list_empty(&virt_timer_list)) {
 329                 /* First timer, just program it. */
 330                 atomic64_set(&virt_timer_current, timer->expires);
 331                 atomic64_set(&virt_timer_elapsed, 0);
 332                 list_add(&timer->entry, &virt_timer_list);
 333         } else {
 334                 /* Update timer against current base. */
 335                 timer->expires += atomic64_read(&virt_timer_elapsed);
 336                 if (likely((s64) timer->expires <
 337                            (s64) atomic64_read(&virt_timer_current)))
 338                         /* The new timer expires before the current timer. */
 339                         atomic64_set(&virt_timer_current, timer->expires);
 340                 /* Insert new timer into the list. */
 341                 list_add_sorted(timer, &virt_timer_list);
 342         }
 343 }
 344
 345 static void __add_vtimer(struct vtimer_list *timer, int periodic)
 346 {
 347         unsigned long flags;
 348
 349         timer->interval = periodic ? timer->expires : 0;
 350         spin_lock_irqsave(&virt_timer_lock, flags);
 351         internal_add_vtimer(timer);
 352         spin_unlock_irqrestore(&virt_timer_lock, flags);
 353 }
 354
 355 /*
 356  * add_virt_timer - add a oneshot virtual CPU timer
 357  */
 358 void add_virt_timer(struct vtimer_list *timer)
 359 {
 360         __add_vtimer(timer, 0);
 361 }
 362 EXPORT_SYMBOL(add_virt_timer);
 363
 364 /*
 365  * add_virt_timer_int - add an interval virtual CPU timer
 366  */
 367 void add_virt_timer_periodic(struct vtimer_list *timer)
 368 {
 369         __add_vtimer(timer, 1);
 370 }
 371 EXPORT_SYMBOL(add_virt_timer_periodic);
 372
 373 static int __mod_vtimer(struct vtimer_list *timer, u64 expires, int periodic)
 374 {
 375         unsigned long flags;
 376         int rc;
 377
 378         BUG_ON(!timer->function);
 379
 380         if (timer->expires == expires && vtimer_pending(timer))
 381                 return 1;
 382         spin_lock_irqsave(&virt_timer_lock, flags);
 383         rc = vtimer_pending(timer);
 384         if (rc)
 385                 list_del_init(&timer->entry);
 386         timer->interval = periodic ? expires : 0;
 387         timer->expires = expires;
 388         internal_add_vtimer(timer);
 389         spin_unlock_irqrestore(&virt_timer_lock, flags);
 390         return rc;
 391 }
 392
 393 /*
 394  * returns whether it has modified a pending timer (1) or not (0)
 395  */
 396 int mod_virt_timer(struct vtimer_list *timer, u64 expires)
 397 {
 398         return __mod_vtimer(timer, expires, 0);
 399 }
 400 EXPORT_SYMBOL(mod_virt_timer);
 401
 402 /*
 403  * returns whether it has modified a pending timer (1) or not (0)
 404  */
 405 int mod_virt_timer_periodic(struct vtimer_list *timer, u64 expires)
 406 {
 407         return __mod_vtimer(timer, expires, 1);
 408 }
 409 EXPORT_SYMBOL(mod_virt_timer_periodic);
 410
 411 /*
 412  * Delete a virtual timer.
 413  *
 414  * returns whether the deleted timer was pending (1) or not (0)
 415  */
 416 int del_virt_timer(struct vtimer_list *timer)
 417 {
 418         unsigned long flags;
 419
 420         if (!vtimer_pending(timer))
 421                 return 0;
 422         spin_lock_irqsave(&virt_timer_lock, flags);
 423         list_del_init(&timer->entry);
 424         spin_unlock_irqrestore(&virt_timer_lock, flags);
 425         return 1;
 426 }
 427 EXPORT_SYMBOL(del_virt_timer);
 428
 429 /*
 430  * Start the virtual CPU timer on the current CPU.
 431  */
 432 void vtime_init(void)
 433 {
 434         /* set initial cpu timer */
 435         set_vtimer(VTIMER_MAX_SLICE);
 436         /* Setup initial MT scaling values */
 437         if (smp_cpu_mtid) {
 438                 __this_cpu_write(mt_scaling_jiffies, jiffies);
 439                 __this_cpu_write(mt_scaling_mult, 1);
 440                 __this_cpu_write(mt_scaling_div, 1);
 441                 stcctm(MT_DIAG, smp_cpu_mtid + 1, this_cpu_ptr(mt_cycles));
 442         }
 443 }