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