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 static DEFINE_PER_CPU(u64, mt_cycles[32]);
  29 static DEFINE_PER_CPU(u64, mt_scaling_mult) = { 1 };
  30 static DEFINE_PER_CPU(u64, mt_scaling_div) = { 1 };
  31
  32 static inline u64 get_vtimer(void)
  33 {
  34         u64 timer;
  35
  36         asm volatile("stpt %0" : "=m" (timer));
  37         return timer;
  38 }
  39
  40 static inline void set_vtimer(u64 expires)
  41 {
  42         u64 timer;
  43
  44         asm volatile(
  45                 "       stpt    %0\n"   /* Store current cpu timer value */
  46                 "       spt     %1"     /* Set new value imm. afterwards */
  47                 : "=m" (timer) : "m" (expires));
  48         S390_lowcore.system_timer += S390_lowcore.last_update_timer - timer;
  49         S390_lowcore.last_update_timer = expires;
  50 }
  51
  52 static inline int virt_timer_forward(u64 elapsed)
  53 {
  54         BUG_ON(!irqs_disabled());
  55
  56         if (list_empty(&virt_timer_list))
  57                 return 0;
  58         elapsed = atomic64_add_return(elapsed, &virt_timer_elapsed);
  59         return elapsed >= atomic64_read(&virt_timer_current);
  60 }
  61
  62 /*
  63  * Update process times based on virtual cpu times stored by entry.S
  64  * to the lowcore fields user_timer, system_timer & steal_clock.
  65  */
  66 static int do_account_vtime(struct task_struct *tsk, int hardirq_offset)
  67 {
  68         struct thread_info *ti = task_thread_info(tsk);
  69         u64 timer, clock, user, system, steal;
  70         u64 user_scaled, system_scaled;
  71         int i;
  72
  73         timer = S390_lowcore.last_update_timer;
  74         clock = S390_lowcore.last_update_clock;
  75         asm volatile(
  76                 "       stpt    %0\n"   /* Store current cpu timer value */
  77 #ifdef CONFIG_HAVE_MARCH_Z9_109_FEATURES
  78                 "       stckf   %1"     /* Store current tod clock value */
  79 #else
  80                 "       stck    %1"     /* Store current tod clock value */
  81 #endif
  82                 : "=m" (S390_lowcore.last_update_timer),
  83                   "=m" (S390_lowcore.last_update_clock));
  84         S390_lowcore.system_timer += timer - S390_lowcore.last_update_timer;
  85         S390_lowcore.steal_timer += S390_lowcore.last_update_clock - clock;
  86
  87         /* Do MT utilization calculation */
  88         if (smp_cpu_mtid) {
  89                 u64 cycles_new[32], *cycles_old;
  90                 u64 delta, mult, div;
  91
  92                 cycles_old = this_cpu_ptr(mt_cycles);
  93                 if (stcctm5(smp_cpu_mtid + 1, cycles_new) < 2) {
  94                         mult = div = 0;
  95                         for (i = 0; i <= smp_cpu_mtid; i++) {
  96                                 delta = cycles_new[i] - cycles_old[i];
  97                                 mult += delta;
  98                                 div += (i + 1) * delta;
  99                         }
 100                         if (mult > 0) {
 101                                 /* Update scaling factor */
 102                                 __this_cpu_write(mt_scaling_mult, mult);
 103                                 __this_cpu_write(mt_scaling_div, div);
 104                                 memcpy(cycles_old, cycles_new,
 105                                        sizeof(u64) * (smp_cpu_mtid + 1));
 106                         }
 107                 }
 108         }
 109
 110         user = S390_lowcore.user_timer - ti->user_timer;
 111         S390_lowcore.steal_timer -= user;
 112         ti->user_timer = S390_lowcore.user_timer;
 113
 114         system = S390_lowcore.system_timer - ti->system_timer;
 115         S390_lowcore.steal_timer -= system;
 116         ti->system_timer = S390_lowcore.system_timer;
 117
 118         user_scaled = user;
 119         system_scaled = system;
 120         /* Do MT utilization scaling */
 121         if (smp_cpu_mtid) {
 122                 u64 mult = __this_cpu_read(mt_scaling_mult);
 123                 u64 div = __this_cpu_read(mt_scaling_div);
 124
 125                 user_scaled = (user_scaled * mult) / div;
 126                 system_scaled = (system_scaled * mult) / div;
 127         }
 128         account_user_time(tsk, user, user_scaled);
 129         account_system_time(tsk, hardirq_offset, system, system_scaled);
 130
 131         steal = S390_lowcore.steal_timer;
 132         if ((s64) steal > 0) {
 133                 S390_lowcore.steal_timer = 0;
 134                 account_steal_time(steal);
 135         }
 136
 137         return virt_timer_forward(user + system);
 138 }
 139
 140 void vtime_task_switch(struct task_struct *prev)
 141 {
 142         struct thread_info *ti;
 143
 144         do_account_vtime(prev, 0);
 145         ti = task_thread_info(prev);
 146         ti->user_timer = S390_lowcore.user_timer;
 147         ti->system_timer = S390_lowcore.system_timer;
 148         ti = task_thread_info(current);
 149         S390_lowcore.user_timer = ti->user_timer;
 150         S390_lowcore.system_timer = ti->system_timer;
 151 }
 152
 153 /*
 154  * In s390, accounting pending user time also implies
 155  * accounting system time in order to correctly compute
 156  * the stolen time accounting.
 157  */
 158 void vtime_account_user(struct task_struct *tsk)
 159 {
 160         if (do_account_vtime(tsk, HARDIRQ_OFFSET))
 161                 virt_timer_expire();
 162 }
 163
 164 /*
 165  * Update process times based on virtual cpu times stored by entry.S
 166  * to the lowcore fields user_timer, system_timer & steal_clock.
 167  */
 168 void vtime_account_irq_enter(struct task_struct *tsk)
 169 {
 170         struct thread_info *ti = task_thread_info(tsk);
 171         u64 timer, system, system_scaled;
 172
 173         timer = S390_lowcore.last_update_timer;
 174         S390_lowcore.last_update_timer = get_vtimer();
 175         S390_lowcore.system_timer += timer - S390_lowcore.last_update_timer;
 176
 177         system = S390_lowcore.system_timer - ti->system_timer;
 178         S390_lowcore.steal_timer -= system;
 179         ti->system_timer = S390_lowcore.system_timer;
 180         system_scaled = system;
 181         /* Do MT utilization scaling */
 182         if (smp_cpu_mtid) {
 183                 u64 mult = __this_cpu_read(mt_scaling_mult);
 184                 u64 div = __this_cpu_read(mt_scaling_div);
 185
 186                 system_scaled = (system_scaled * mult) / div;
 187         }
 188         account_system_time(tsk, 0, system, system_scaled);
 189
 190         virt_timer_forward(system);
 191 }
 192 EXPORT_SYMBOL_GPL(vtime_account_irq_enter);
 193
 194 void vtime_account_system(struct task_struct *tsk)
 195 __attribute__((alias("vtime_account_irq_enter")));
 196 EXPORT_SYMBOL_GPL(vtime_account_system);
 197
 198 /*
 199  * Sorted add to a list. List is linear searched until first bigger
 200  * element is found.
 201  */
 202 static void list_add_sorted(struct vtimer_list *timer, struct list_head *head)
 203 {
 204         struct vtimer_list *tmp;
 205
 206         list_for_each_entry(tmp, head, entry) {
 207                 if (tmp->expires > timer->expires) {
 208                         list_add_tail(&timer->entry, &tmp->entry);
 209                         return;
 210                 }
 211         }
 212         list_add_tail(&timer->entry, head);
 213 }
 214
 215 /*
 216  * Handler for expired virtual CPU timer.
 217  */
 218 static void virt_timer_expire(void)
 219 {
 220         struct vtimer_list *timer, *tmp;
 221         unsigned long elapsed;
 222         LIST_HEAD(cb_list);
 223
 224         /* walk timer list, fire all expired timers */
 225         spin_lock(&virt_timer_lock);
 226         elapsed = atomic64_read(&virt_timer_elapsed);
 227         list_for_each_entry_safe(timer, tmp, &virt_timer_list, entry) {
 228                 if (timer->expires < elapsed)
 229                         /* move expired timer to the callback queue */
 230                         list_move_tail(&timer->entry, &cb_list);
 231                 else
 232                         timer->expires -= elapsed;
 233         }
 234         if (!list_empty(&virt_timer_list)) {
 235                 timer = list_first_entry(&virt_timer_list,
 236                                          struct vtimer_list, entry);
 237                 atomic64_set(&virt_timer_current, timer->expires);
 238         }
 239         atomic64_sub(elapsed, &virt_timer_elapsed);
 240         spin_unlock(&virt_timer_lock);
 241
 242         /* Do callbacks and recharge periodic timers */
 243         list_for_each_entry_safe(timer, tmp, &cb_list, entry) {
 244                 list_del_init(&timer->entry);
 245                 timer->function(timer->data);
 246                 if (timer->interval) {
 247                         /* Recharge interval timer */
 248                         timer->expires = timer->interval +
 249                                 atomic64_read(&virt_timer_elapsed);
 250                         spin_lock(&virt_timer_lock);
 251                         list_add_sorted(timer, &virt_timer_list);
 252                         spin_unlock(&virt_timer_lock);
 253                 }
 254         }
 255 }
 256
 257 void init_virt_timer(struct vtimer_list *timer)
 258 {
 259         timer->function = NULL;
 260         INIT_LIST_HEAD(&timer->entry);
 261 }
 262 EXPORT_SYMBOL(init_virt_timer);
 263
 264 static inline int vtimer_pending(struct vtimer_list *timer)
 265 {
 266         return !list_empty(&timer->entry);
 267 }
 268
 269 static void internal_add_vtimer(struct vtimer_list *timer)
 270 {
 271         if (list_empty(&virt_timer_list)) {
 272                 /* First timer, just program it. */
 273                 atomic64_set(&virt_timer_current, timer->expires);
 274                 atomic64_set(&virt_timer_elapsed, 0);
 275                 list_add(&timer->entry, &virt_timer_list);
 276         } else {
 277                 /* Update timer against current base. */
 278                 timer->expires += atomic64_read(&virt_timer_elapsed);
 279                 if (likely((s64) timer->expires <
 280                            (s64) atomic64_read(&virt_timer_current)))
 281                         /* The new timer expires before the current timer. */
 282                         atomic64_set(&virt_timer_current, timer->expires);
 283                 /* Insert new timer into the list. */
 284                 list_add_sorted(timer, &virt_timer_list);
 285         }
 286 }
 287
 288 static void __add_vtimer(struct vtimer_list *timer, int periodic)
 289 {
 290         unsigned long flags;
 291
 292         timer->interval = periodic ? timer->expires : 0;
 293         spin_lock_irqsave(&virt_timer_lock, flags);
 294         internal_add_vtimer(timer);
 295         spin_unlock_irqrestore(&virt_timer_lock, flags);
 296 }
 297
 298 /*
 299  * add_virt_timer - add an oneshot virtual CPU timer
 300  */
 301 void add_virt_timer(struct vtimer_list *timer)
 302 {
 303         __add_vtimer(timer, 0);
 304 }
 305 EXPORT_SYMBOL(add_virt_timer);
 306
 307 /*
 308  * add_virt_timer_int - add an interval virtual CPU timer
 309  */
 310 void add_virt_timer_periodic(struct vtimer_list *timer)
 311 {
 312         __add_vtimer(timer, 1);
 313 }
 314 EXPORT_SYMBOL(add_virt_timer_periodic);
 315
 316 static int __mod_vtimer(struct vtimer_list *timer, u64 expires, int periodic)
 317 {
 318         unsigned long flags;
 319         int rc;
 320
 321         BUG_ON(!timer->function);
 322
 323         if (timer->expires == expires && vtimer_pending(timer))
 324                 return 1;
 325         spin_lock_irqsave(&virt_timer_lock, flags);
 326         rc = vtimer_pending(timer);
 327         if (rc)
 328                 list_del_init(&timer->entry);
 329         timer->interval = periodic ? expires : 0;
 330         timer->expires = expires;
 331         internal_add_vtimer(timer);
 332         spin_unlock_irqrestore(&virt_timer_lock, flags);
 333         return rc;
 334 }
 335
 336 /*
 337  * returns whether it has modified a pending timer (1) or not (0)
 338  */
 339 int mod_virt_timer(struct vtimer_list *timer, u64 expires)
 340 {
 341         return __mod_vtimer(timer, expires, 0);
 342 }
 343 EXPORT_SYMBOL(mod_virt_timer);
 344
 345 /*
 346  * returns whether it has modified a pending timer (1) or not (0)
 347  */
 348 int mod_virt_timer_periodic(struct vtimer_list *timer, u64 expires)
 349 {
 350         return __mod_vtimer(timer, expires, 1);
 351 }
 352 EXPORT_SYMBOL(mod_virt_timer_periodic);
 353
 354 /*
 355  * Delete a virtual timer.
 356  *
 357  * returns whether the deleted timer was pending (1) or not (0)
 358  */
 359 int del_virt_timer(struct vtimer_list *timer)
 360 {
 361         unsigned long flags;
 362
 363         if (!vtimer_pending(timer))
 364                 return 0;
 365         spin_lock_irqsave(&virt_timer_lock, flags);
 366         list_del_init(&timer->entry);
 367         spin_unlock_irqrestore(&virt_timer_lock, flags);
 368         return 1;
 369 }
 370 EXPORT_SYMBOL(del_virt_timer);
 371
 372 /*
 373  * Start the virtual CPU timer on the current CPU.
 374  */
 375 void vtime_init(void)
 376 {
 377         /* set initial cpu timer */
 378         set_vtimer(VTIMER_MAX_SLICE);
 379 }