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 #include <asm/alternative.h>
  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         /* Use STORE CLOCK by default, STORE CLOCK FAST if available. */
 132         alternative_io("stpt %0\n .insn s,0xb2050000,%1\n",
 133                        "stpt %0\n .insn s,0xb27c0000,%1\n",
 134                        25,
 135                        ASM_OUTPUT2("=Q" (S390_lowcore.last_update_timer),
 136                                    "=Q" (S390_lowcore.last_update_clock)),
 137                        ASM_NO_INPUT_CLOBBER("cc"));
 138         clock = S390_lowcore.last_update_clock - clock;
 139         timer -= S390_lowcore.last_update_timer;
 140
 141         if (hardirq_count())
 142                 S390_lowcore.hardirq_timer += timer;
 143         else
 144                 S390_lowcore.system_timer += timer;
 145
 146         /* Update MT utilization calculation */
 147         if (smp_cpu_mtid &&
 148             time_after64(jiffies_64, this_cpu_read(mt_scaling_jiffies)))
 149                 update_mt_scaling();
 150
 151         /* Calculate cputime delta */
 152         user = update_tsk_timer(&tsk->thread.user_timer,
 153                                 READ_ONCE(S390_lowcore.user_timer));
 154         guest = update_tsk_timer(&tsk->thread.guest_timer,
 155                                  READ_ONCE(S390_lowcore.guest_timer));
 156         system = update_tsk_timer(&tsk->thread.system_timer,
 157                                   READ_ONCE(S390_lowcore.system_timer));
 158         hardirq = update_tsk_timer(&tsk->thread.hardirq_timer,
 159                                    READ_ONCE(S390_lowcore.hardirq_timer));
 160         softirq = update_tsk_timer(&tsk->thread.softirq_timer,
 161                                    READ_ONCE(S390_lowcore.softirq_timer));
 162         S390_lowcore.steal_timer +=
 163                 clock - user - guest - system - hardirq - softirq;
 164
 165         /* Push account value */
 166         if (user) {
 167                 account_user_time(tsk, cputime_to_nsecs(user));
 168                 tsk->utimescaled += cputime_to_nsecs(scale_vtime(user));
 169         }
 170
 171         if (guest) {
 172                 account_guest_time(tsk, cputime_to_nsecs(guest));
 173                 tsk->utimescaled += cputime_to_nsecs(scale_vtime(guest));
 174         }
 175
 176         if (system)
 177                 account_system_index_scaled(tsk, system, CPUTIME_SYSTEM);
 178         if (hardirq)
 179                 account_system_index_scaled(tsk, hardirq, CPUTIME_IRQ);
 180         if (softirq)
 181                 account_system_index_scaled(tsk, softirq, CPUTIME_SOFTIRQ);
 182
 183         return virt_timer_forward(user + guest + system + hardirq + softirq);
 184 }
 185
 186 void vtime_task_switch(struct task_struct *prev)
 187 {
 188         do_account_vtime(prev);
 189         prev->thread.user_timer = S390_lowcore.user_timer;
 190         prev->thread.guest_timer = S390_lowcore.guest_timer;
 191         prev->thread.system_timer = S390_lowcore.system_timer;
 192         prev->thread.hardirq_timer = S390_lowcore.hardirq_timer;
 193         prev->thread.softirq_timer = S390_lowcore.softirq_timer;
 194         S390_lowcore.user_timer = current->thread.user_timer;
 195         S390_lowcore.guest_timer = current->thread.guest_timer;
 196         S390_lowcore.system_timer = current->thread.system_timer;
 197         S390_lowcore.hardirq_timer = current->thread.hardirq_timer;
 198         S390_lowcore.softirq_timer = current->thread.softirq_timer;
 199 }
 200
 201 /*
 202  * In s390, accounting pending user time also implies
 203  * accounting system time in order to correctly compute
 204  * the stolen time accounting.
 205  */
 206 void vtime_flush(struct task_struct *tsk)
 207 {
 208         u64 steal, avg_steal;
 209
 210         if (do_account_vtime(tsk))
 211                 virt_timer_expire();
 212
 213         steal = S390_lowcore.steal_timer;
 214         avg_steal = S390_lowcore.avg_steal_timer / 2;
 215         if ((s64) steal > 0) {
 216                 S390_lowcore.steal_timer = 0;
 217                 account_steal_time(cputime_to_nsecs(steal));
 218                 avg_steal += steal;
 219         }
 220         S390_lowcore.avg_steal_timer = avg_steal;
 221 }
 222
 223 static u64 vtime_delta(void)
 224 {
 225         u64 timer = S390_lowcore.last_update_timer;
 226
 227         S390_lowcore.last_update_timer = get_vtimer();
 228
 229         return timer - S390_lowcore.last_update_timer;
 230 }
 231
 232 /*
 233  * Update process times based on virtual cpu times stored by entry.S
 234  * to the lowcore fields user_timer, system_timer & steal_clock.
 235  */
 236 void vtime_account_kernel(struct task_struct *tsk)
 237 {
 238         u64 delta = vtime_delta();
 239
 240         if (tsk->flags & PF_VCPU)
 241                 S390_lowcore.guest_timer += delta;
 242         else
 243                 S390_lowcore.system_timer += delta;
 244
 245         virt_timer_forward(delta);
 246 }
 247 EXPORT_SYMBOL_GPL(vtime_account_kernel);
 248
 249 void vtime_account_softirq(struct task_struct *tsk)
 250 {
 251         u64 delta = vtime_delta();
 252
 253         S390_lowcore.softirq_timer += delta;
 254
 255         virt_timer_forward(delta);
 256 }
 257
 258 void vtime_account_hardirq(struct task_struct *tsk)
 259 {
 260         u64 delta = vtime_delta();
 261
 262         S390_lowcore.hardirq_timer += delta;
 263
 264         virt_timer_forward(delta);
 265 }
 266
 267 /*
 268  * Sorted add to a list. List is linear searched until first bigger
 269  * element is found.
 270  */
 271 static void list_add_sorted(struct vtimer_list *timer, struct list_head *head)
 272 {
 273         struct vtimer_list *tmp;
 274
 275         list_for_each_entry(tmp, head, entry) {
 276                 if (tmp->expires > timer->expires) {
 277                         list_add_tail(&timer->entry, &tmp->entry);
 278                         return;
 279                 }
 280         }
 281         list_add_tail(&timer->entry, head);
 282 }
 283
 284 /*
 285  * Handler for expired virtual CPU timer.
 286  */
 287 static void virt_timer_expire(void)
 288 {
 289         struct vtimer_list *timer, *tmp;
 290         unsigned long elapsed;
 291         LIST_HEAD(cb_list);
 292
 293         /* walk timer list, fire all expired timers */
 294         spin_lock(&virt_timer_lock);
 295         elapsed = atomic64_read(&virt_timer_elapsed);
 296         list_for_each_entry_safe(timer, tmp, &virt_timer_list, entry) {
 297                 if (timer->expires < elapsed)
 298                         /* move expired timer to the callback queue */
 299                         list_move_tail(&timer->entry, &cb_list);
 300                 else
 301                         timer->expires -= elapsed;
 302         }
 303         if (!list_empty(&virt_timer_list)) {
 304                 timer = list_first_entry(&virt_timer_list,
 305                                          struct vtimer_list, entry);
 306                 atomic64_set(&virt_timer_current, timer->expires);
 307         }
 308         atomic64_sub(elapsed, &virt_timer_elapsed);
 309         spin_unlock(&virt_timer_lock);
 310
 311         /* Do callbacks and recharge periodic timers */
 312         list_for_each_entry_safe(timer, tmp, &cb_list, entry) {
 313                 list_del_init(&timer->entry);
 314                 timer->function(timer->data);
 315                 if (timer->interval) {
 316                         /* Recharge interval timer */
 317                         timer->expires = timer->interval +
 318                                 atomic64_read(&virt_timer_elapsed);
 319                         spin_lock(&virt_timer_lock);
 320                         list_add_sorted(timer, &virt_timer_list);
 321                         spin_unlock(&virt_timer_lock);
 322                 }
 323         }
 324 }
 325
 326 void init_virt_timer(struct vtimer_list *timer)
 327 {
 328         timer->function = NULL;
 329         INIT_LIST_HEAD(&timer->entry);
 330 }
 331 EXPORT_SYMBOL(init_virt_timer);
 332
 333 static inline int vtimer_pending(struct vtimer_list *timer)
 334 {
 335         return !list_empty(&timer->entry);
 336 }
 337
 338 static void internal_add_vtimer(struct vtimer_list *timer)
 339 {
 340         if (list_empty(&virt_timer_list)) {
 341                 /* First timer, just program it. */
 342                 atomic64_set(&virt_timer_current, timer->expires);
 343                 atomic64_set(&virt_timer_elapsed, 0);
 344                 list_add(&timer->entry, &virt_timer_list);
 345         } else {
 346                 /* Update timer against current base. */
 347                 timer->expires += atomic64_read(&virt_timer_elapsed);
 348                 if (likely((s64) timer->expires <
 349                            (s64) atomic64_read(&virt_timer_current)))
 350                         /* The new timer expires before the current timer. */
 351                         atomic64_set(&virt_timer_current, timer->expires);
 352                 /* Insert new timer into the list. */
 353                 list_add_sorted(timer, &virt_timer_list);
 354         }
 355 }
 356
 357 static void __add_vtimer(struct vtimer_list *timer, int periodic)
 358 {
 359         unsigned long flags;
 360
 361         timer->interval = periodic ? timer->expires : 0;
 362         spin_lock_irqsave(&virt_timer_lock, flags);
 363         internal_add_vtimer(timer);
 364         spin_unlock_irqrestore(&virt_timer_lock, flags);
 365 }
 366
 367 /*
 368  * add_virt_timer - add a oneshot virtual CPU timer
 369  */
 370 void add_virt_timer(struct vtimer_list *timer)
 371 {
 372         __add_vtimer(timer, 0);
 373 }
 374 EXPORT_SYMBOL(add_virt_timer);
 375
 376 /*
 377  * add_virt_timer_int - add an interval virtual CPU timer
 378  */
 379 void add_virt_timer_periodic(struct vtimer_list *timer)
 380 {
 381         __add_vtimer(timer, 1);
 382 }
 383 EXPORT_SYMBOL(add_virt_timer_periodic);
 384
 385 static int __mod_vtimer(struct vtimer_list *timer, u64 expires, int periodic)
 386 {
 387         unsigned long flags;
 388         int rc;
 389
 390         BUG_ON(!timer->function);
 391
 392         if (timer->expires == expires && vtimer_pending(timer))
 393                 return 1;
 394         spin_lock_irqsave(&virt_timer_lock, flags);
 395         rc = vtimer_pending(timer);
 396         if (rc)
 397                 list_del_init(&timer->entry);
 398         timer->interval = periodic ? expires : 0;
 399         timer->expires = expires;
 400         internal_add_vtimer(timer);
 401         spin_unlock_irqrestore(&virt_timer_lock, flags);
 402         return rc;
 403 }
 404
 405 /*
 406  * returns whether it has modified a pending timer (1) or not (0)
 407  */
 408 int mod_virt_timer(struct vtimer_list *timer, u64 expires)
 409 {
 410         return __mod_vtimer(timer, expires, 0);
 411 }
 412 EXPORT_SYMBOL(mod_virt_timer);
 413
 414 /*
 415  * returns whether it has modified a pending timer (1) or not (0)
 416  */
 417 int mod_virt_timer_periodic(struct vtimer_list *timer, u64 expires)
 418 {
 419         return __mod_vtimer(timer, expires, 1);
 420 }
 421 EXPORT_SYMBOL(mod_virt_timer_periodic);
 422
 423 /*
 424  * Delete a virtual timer.
 425  *
 426  * returns whether the deleted timer was pending (1) or not (0)
 427  */
 428 int del_virt_timer(struct vtimer_list *timer)
 429 {
 430         unsigned long flags;
 431
 432         if (!vtimer_pending(timer))
 433                 return 0;
 434         spin_lock_irqsave(&virt_timer_lock, flags);
 435         list_del_init(&timer->entry);
 436         spin_unlock_irqrestore(&virt_timer_lock, flags);
 437         return 1;
 438 }
 439 EXPORT_SYMBOL(del_virt_timer);
 440
 441 /*
 442  * Start the virtual CPU timer on the current CPU.
 443  */
 444 void vtime_init(void)
 445 {
 446         /* set initial cpu timer */
 447         set_vtimer(VTIMER_MAX_SLICE);
 448         /* Setup initial MT scaling values */
 449         if (smp_cpu_mtid) {
 450                 __this_cpu_write(mt_scaling_jiffies, jiffies);
 451                 __this_cpu_write(mt_scaling_mult, 1);
 452                 __this_cpu_write(mt_scaling_div, 1);
 453                 stcctm(MT_DIAG, smp_cpu_mtid + 1, this_cpu_ptr(mt_cycles));
 454         }
 455 }