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/export.h>
  10 #include <linux/kernel.h>
  11 #include <linux/timex.h>
  12 #include <linux/types.h>
  13 #include <linux/time.h>
  14
  15 #include <asm/cputime.h>
  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 /*
  94  * Update process times based on virtual cpu times stored by entry.S
  95  * to the lowcore fields user_timer, system_timer & steal_clock.
  96  */
  97 static int do_account_vtime(struct task_struct *tsk, int hardirq_offset)
  98 {
  99         u64 timer, clock, user, system, steal;
 100         u64 user_scaled, system_scaled;
 101
 102         timer = S390_lowcore.last_update_timer;
 103         clock = S390_lowcore.last_update_clock;
 104         asm volatile(
 105                 "       stpt    %0\n"   /* Store current cpu timer value */
 106 #ifdef CONFIG_HAVE_MARCH_Z9_109_FEATURES
 107                 "       stckf   %1"     /* Store current tod clock value */
 108 #else
 109                 "       stck    %1"     /* Store current tod clock value */
 110 #endif
 111                 : "=m" (S390_lowcore.last_update_timer),
 112                   "=m" (S390_lowcore.last_update_clock));
 113         S390_lowcore.system_timer += timer - S390_lowcore.last_update_timer;
 114         S390_lowcore.steal_timer += S390_lowcore.last_update_clock - clock;
 115
 116         /* Update MT utilization calculation */
 117         if (smp_cpu_mtid &&
 118             time_after64(jiffies_64, this_cpu_read(mt_scaling_jiffies)))
 119                 update_mt_scaling();
 120
 121         user = S390_lowcore.user_timer - tsk->thread.user_timer;
 122         S390_lowcore.steal_timer -= user;
 123         tsk->thread.user_timer = S390_lowcore.user_timer;
 124
 125         system = S390_lowcore.system_timer - tsk->thread.system_timer;
 126         S390_lowcore.steal_timer -= system;
 127         tsk->thread.system_timer = S390_lowcore.system_timer;
 128
 129         user_scaled = user;
 130         system_scaled = system;
 131         /* Do MT utilization scaling */
 132         if (smp_cpu_mtid) {
 133                 u64 mult = __this_cpu_read(mt_scaling_mult);
 134                 u64 div = __this_cpu_read(mt_scaling_div);
 135
 136                 user_scaled = (user_scaled * mult) / div;
 137                 system_scaled = (system_scaled * mult) / div;
 138         }
 139         account_user_time(tsk, user);
 140         tsk->utimescaled += user_scaled;
 141         account_system_time(tsk, hardirq_offset, system);
 142         tsk->stimescaled += system_scaled;
 143
 144         steal = S390_lowcore.steal_timer;
 145         if ((s64) steal > 0) {
 146                 S390_lowcore.steal_timer = 0;
 147                 account_steal_time(steal);
 148         }
 149
 150         return virt_timer_forward(user + system);
 151 }
 152
 153 void vtime_task_switch(struct task_struct *prev)
 154 {
 155         do_account_vtime(prev, 0);
 156         prev->thread.user_timer = S390_lowcore.user_timer;
 157         prev->thread.system_timer = S390_lowcore.system_timer;
 158         S390_lowcore.user_timer = current->thread.user_timer;
 159         S390_lowcore.system_timer = current->thread.system_timer;
 160 }
 161
 162 /*
 163  * In s390, accounting pending user time also implies
 164  * accounting system time in order to correctly compute
 165  * the stolen time accounting.
 166  */
 167 void vtime_account_user(struct task_struct *tsk)
 168 {
 169         if (do_account_vtime(tsk, HARDIRQ_OFFSET))
 170                 virt_timer_expire();
 171 }
 172
 173 /*
 174  * Update process times based on virtual cpu times stored by entry.S
 175  * to the lowcore fields user_timer, system_timer & steal_clock.
 176  */
 177 void vtime_account_irq_enter(struct task_struct *tsk)
 178 {
 179         u64 timer, system, system_scaled;
 180
 181         timer = S390_lowcore.last_update_timer;
 182         S390_lowcore.last_update_timer = get_vtimer();
 183         S390_lowcore.system_timer += timer - S390_lowcore.last_update_timer;
 184
 185         /* Update MT utilization calculation */
 186         if (smp_cpu_mtid &&
 187             time_after64(jiffies_64, this_cpu_read(mt_scaling_jiffies)))
 188                 update_mt_scaling();
 189
 190         system = S390_lowcore.system_timer - tsk->thread.system_timer;
 191         S390_lowcore.steal_timer -= system;
 192         tsk->thread.system_timer = S390_lowcore.system_timer;
 193         system_scaled = system;
 194         /* Do MT utilization scaling */
 195         if (smp_cpu_mtid) {
 196                 u64 mult = __this_cpu_read(mt_scaling_mult);
 197                 u64 div = __this_cpu_read(mt_scaling_div);
 198
 199                 system_scaled = (system_scaled * mult) / div;
 200         }
 201         account_system_time(tsk, 0, system);
 202         tsk->stimescaled += system_scaled;
 203
 204         virt_timer_forward(system);
 205 }
 206 EXPORT_SYMBOL_GPL(vtime_account_irq_enter);
 207
 208 void vtime_account_system(struct task_struct *tsk)
 209 __attribute__((alias("vtime_account_irq_enter")));
 210 EXPORT_SYMBOL_GPL(vtime_account_system);
 211
 212 /*
 213  * Sorted add to a list. List is linear searched until first bigger
 214  * element is found.
 215  */
 216 static void list_add_sorted(struct vtimer_list *timer, struct list_head *head)
 217 {
 218         struct vtimer_list *tmp;
 219
 220         list_for_each_entry(tmp, head, entry) {
 221                 if (tmp->expires > timer->expires) {
 222                         list_add_tail(&timer->entry, &tmp->entry);
 223                         return;
 224                 }
 225         }
 226         list_add_tail(&timer->entry, head);
 227 }
 228
 229 /*
 230  * Handler for expired virtual CPU timer.
 231  */
 232 static void virt_timer_expire(void)
 233 {
 234         struct vtimer_list *timer, *tmp;
 235         unsigned long elapsed;
 236         LIST_HEAD(cb_list);
 237
 238         /* walk timer list, fire all expired timers */
 239         spin_lock(&virt_timer_lock);
 240         elapsed = atomic64_read(&virt_timer_elapsed);
 241         list_for_each_entry_safe(timer, tmp, &virt_timer_list, entry) {
 242                 if (timer->expires < elapsed)
 243                         /* move expired timer to the callback queue */
 244                         list_move_tail(&timer->entry, &cb_list);
 245                 else
 246                         timer->expires -= elapsed;
 247         }
 248         if (!list_empty(&virt_timer_list)) {
 249                 timer = list_first_entry(&virt_timer_list,
 250                                          struct vtimer_list, entry);
 251                 atomic64_set(&virt_timer_current, timer->expires);
 252         }
 253         atomic64_sub(elapsed, &virt_timer_elapsed);
 254         spin_unlock(&virt_timer_lock);
 255
 256         /* Do callbacks and recharge periodic timers */
 257         list_for_each_entry_safe(timer, tmp, &cb_list, entry) {
 258                 list_del_init(&timer->entry);
 259                 timer->function(timer->data);
 260                 if (timer->interval) {
 261                         /* Recharge interval timer */
 262                         timer->expires = timer->interval +
 263                                 atomic64_read(&virt_timer_elapsed);
 264                         spin_lock(&virt_timer_lock);
 265                         list_add_sorted(timer, &virt_timer_list);
 266                         spin_unlock(&virt_timer_lock);
 267                 }
 268         }
 269 }
 270
 271 void init_virt_timer(struct vtimer_list *timer)
 272 {
 273         timer->function = NULL;
 274         INIT_LIST_HEAD(&timer->entry);
 275 }
 276 EXPORT_SYMBOL(init_virt_timer);
 277
 278 static inline int vtimer_pending(struct vtimer_list *timer)
 279 {
 280         return !list_empty(&timer->entry);
 281 }
 282
 283 static void internal_add_vtimer(struct vtimer_list *timer)
 284 {
 285         if (list_empty(&virt_timer_list)) {
 286                 /* First timer, just program it. */
 287                 atomic64_set(&virt_timer_current, timer->expires);
 288                 atomic64_set(&virt_timer_elapsed, 0);
 289                 list_add(&timer->entry, &virt_timer_list);
 290         } else {
 291                 /* Update timer against current base. */
 292                 timer->expires += atomic64_read(&virt_timer_elapsed);
 293                 if (likely((s64) timer->expires <
 294                            (s64) atomic64_read(&virt_timer_current)))
 295                         /* The new timer expires before the current timer. */
 296                         atomic64_set(&virt_timer_current, timer->expires);
 297                 /* Insert new timer into the list. */
 298                 list_add_sorted(timer, &virt_timer_list);
 299         }
 300 }
 301
 302 static void __add_vtimer(struct vtimer_list *timer, int periodic)
 303 {
 304         unsigned long flags;
 305
 306         timer->interval = periodic ? timer->expires : 0;
 307         spin_lock_irqsave(&virt_timer_lock, flags);
 308         internal_add_vtimer(timer);
 309         spin_unlock_irqrestore(&virt_timer_lock, flags);
 310 }
 311
 312 /*
 313  * add_virt_timer - add an oneshot virtual CPU timer
 314  */
 315 void add_virt_timer(struct vtimer_list *timer)
 316 {
 317         __add_vtimer(timer, 0);
 318 }
 319 EXPORT_SYMBOL(add_virt_timer);
 320
 321 /*
 322  * add_virt_timer_int - add an interval virtual CPU timer
 323  */
 324 void add_virt_timer_periodic(struct vtimer_list *timer)
 325 {
 326         __add_vtimer(timer, 1);
 327 }
 328 EXPORT_SYMBOL(add_virt_timer_periodic);
 329
 330 static int __mod_vtimer(struct vtimer_list *timer, u64 expires, int periodic)
 331 {
 332         unsigned long flags;
 333         int rc;
 334
 335         BUG_ON(!timer->function);
 336
 337         if (timer->expires == expires && vtimer_pending(timer))
 338                 return 1;
 339         spin_lock_irqsave(&virt_timer_lock, flags);
 340         rc = vtimer_pending(timer);
 341         if (rc)
 342                 list_del_init(&timer->entry);
 343         timer->interval = periodic ? expires : 0;
 344         timer->expires = expires;
 345         internal_add_vtimer(timer);
 346         spin_unlock_irqrestore(&virt_timer_lock, flags);
 347         return rc;
 348 }
 349
 350 /*
 351  * returns whether it has modified a pending timer (1) or not (0)
 352  */
 353 int mod_virt_timer(struct vtimer_list *timer, u64 expires)
 354 {
 355         return __mod_vtimer(timer, expires, 0);
 356 }
 357 EXPORT_SYMBOL(mod_virt_timer);
 358
 359 /*
 360  * returns whether it has modified a pending timer (1) or not (0)
 361  */
 362 int mod_virt_timer_periodic(struct vtimer_list *timer, u64 expires)
 363 {
 364         return __mod_vtimer(timer, expires, 1);
 365 }
 366 EXPORT_SYMBOL(mod_virt_timer_periodic);
 367
 368 /*
 369  * Delete a virtual timer.
 370  *
 371  * returns whether the deleted timer was pending (1) or not (0)
 372  */
 373 int del_virt_timer(struct vtimer_list *timer)
 374 {
 375         unsigned long flags;
 376
 377         if (!vtimer_pending(timer))
 378                 return 0;
 379         spin_lock_irqsave(&virt_timer_lock, flags);
 380         list_del_init(&timer->entry);
 381         spin_unlock_irqrestore(&virt_timer_lock, flags);
 382         return 1;
 383 }
 384 EXPORT_SYMBOL(del_virt_timer);
 385
 386 /*
 387  * Start the virtual CPU timer on the current CPU.
 388  */
 389 void vtime_init(void)
 390 {
 391         /* set initial cpu timer */
 392         set_vtimer(VTIMER_MAX_SLICE);
 393         /* Setup initial MT scaling values */
 394         if (smp_cpu_mtid) {
 395                 __this_cpu_write(mt_scaling_jiffies, jiffies);
 396                 __this_cpu_write(mt_scaling_mult, 1);
 397                 __this_cpu_write(mt_scaling_div, 1);
 398                 stcctm5(smp_cpu_mtid + 1, this_cpu_ptr(mt_cycles));
 399         }
 400 }