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 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 /*
  64  * Update process times based on virtual cpu times stored by entry.S
  65  * to the lowcore fields user_timer, system_timer & steal_clock.
  66  */
  67 static int do_account_vtime(struct task_struct *tsk, int hardirq_offset)
  68 {
  69         struct thread_info *ti = task_thread_info(tsk);
  70         u64 timer, clock, user, system, steal;
  71         u64 user_scaled, system_scaled;
  72         int i;
  73
  74         timer = S390_lowcore.last_update_timer;
  75         clock = S390_lowcore.last_update_clock;
  76         asm volatile(
  77                 "       stpt    %0\n"   /* Store current cpu timer value */
  78 #ifdef CONFIG_HAVE_MARCH_Z9_109_FEATURES
  79                 "       stckf   %1"     /* Store current tod clock value */
  80 #else
  81                 "       stck    %1"     /* Store current tod clock value */
  82 #endif
  83                 : "=m" (S390_lowcore.last_update_timer),
  84                   "=m" (S390_lowcore.last_update_clock));
  85         S390_lowcore.system_timer += timer - S390_lowcore.last_update_timer;
  86         S390_lowcore.steal_timer += S390_lowcore.last_update_clock - clock;
  87
  88         /* Do MT utilization calculation */
  89         if (smp_cpu_mtid &&
  90             time_after64(jiffies_64, __this_cpu_read(mt_scaling_jiffies))) {
  91                 u64 cycles_new[32], *cycles_old;
  92                 u64 delta, mult, div;
  93
  94                 cycles_old = this_cpu_ptr(mt_cycles);
  95                 if (stcctm5(smp_cpu_mtid + 1, cycles_new) < 2) {
  96                         mult = div = 0;
  97                         for (i = 0; i <= smp_cpu_mtid; i++) {
  98                                 delta = cycles_new[i] - cycles_old[i];
  99                                 mult += delta;
 100                                 div += (i + 1) * delta;
 101                         }
 102                         if (mult > 0) {
 103                                 /* Update scaling factor */
 104                                 __this_cpu_write(mt_scaling_mult, mult);
 105                                 __this_cpu_write(mt_scaling_div, div);
 106                                 memcpy(cycles_old, cycles_new,
 107                                        sizeof(u64) * (smp_cpu_mtid + 1));
 108                         }
 109                 }
 110                 __this_cpu_write(mt_scaling_jiffies, jiffies_64);
 111         }
 112
 113         user = S390_lowcore.user_timer - ti->user_timer;
 114         S390_lowcore.steal_timer -= user;
 115         ti->user_timer = S390_lowcore.user_timer;
 116
 117         system = S390_lowcore.system_timer - ti->system_timer;
 118         S390_lowcore.steal_timer -= system;
 119         ti->system_timer = S390_lowcore.system_timer;
 120
 121         user_scaled = user;
 122         system_scaled = system;
 123         /* Do MT utilization scaling */
 124         if (smp_cpu_mtid) {
 125                 u64 mult = __this_cpu_read(mt_scaling_mult);
 126                 u64 div = __this_cpu_read(mt_scaling_div);
 127
 128                 user_scaled = (user_scaled * mult) / div;
 129                 system_scaled = (system_scaled * mult) / div;
 130         }
 131         account_user_time(tsk, user, user_scaled);
 132         account_system_time(tsk, hardirq_offset, system, system_scaled);
 133
 134         steal = S390_lowcore.steal_timer;
 135         if ((s64) steal > 0) {
 136                 S390_lowcore.steal_timer = 0;
 137                 account_steal_time(steal);
 138         }
 139
 140         return virt_timer_forward(user + system);
 141 }
 142
 143 void vtime_task_switch(struct task_struct *prev)
 144 {
 145         struct thread_info *ti;
 146
 147         do_account_vtime(prev, 0);
 148         ti = task_thread_info(prev);
 149         ti->user_timer = S390_lowcore.user_timer;
 150         ti->system_timer = S390_lowcore.system_timer;
 151         ti = task_thread_info(current);
 152         S390_lowcore.user_timer = ti->user_timer;
 153         S390_lowcore.system_timer = ti->system_timer;
 154 }
 155
 156 /*
 157  * In s390, accounting pending user time also implies
 158  * accounting system time in order to correctly compute
 159  * the stolen time accounting.
 160  */
 161 void vtime_account_user(struct task_struct *tsk)
 162 {
 163         if (do_account_vtime(tsk, HARDIRQ_OFFSET))
 164                 virt_timer_expire();
 165 }
 166
 167 /*
 168  * Update process times based on virtual cpu times stored by entry.S
 169  * to the lowcore fields user_timer, system_timer & steal_clock.
 170  */
 171 void vtime_account_irq_enter(struct task_struct *tsk)
 172 {
 173         struct thread_info *ti = task_thread_info(tsk);
 174         u64 timer, system, system_scaled;
 175
 176         timer = S390_lowcore.last_update_timer;
 177         S390_lowcore.last_update_timer = get_vtimer();
 178         S390_lowcore.system_timer += timer - S390_lowcore.last_update_timer;
 179
 180         system = S390_lowcore.system_timer - ti->system_timer;
 181         S390_lowcore.steal_timer -= system;
 182         ti->system_timer = S390_lowcore.system_timer;
 183         system_scaled = system;
 184         /* Do MT utilization scaling */
 185         if (smp_cpu_mtid) {
 186                 u64 mult = __this_cpu_read(mt_scaling_mult);
 187                 u64 div = __this_cpu_read(mt_scaling_div);
 188
 189                 system_scaled = (system_scaled * mult) / div;
 190         }
 191         account_system_time(tsk, 0, system, system_scaled);
 192
 193         virt_timer_forward(system);
 194 }
 195 EXPORT_SYMBOL_GPL(vtime_account_irq_enter);
 196
 197 void vtime_account_system(struct task_struct *tsk)
 198 __attribute__((alias("vtime_account_irq_enter")));
 199 EXPORT_SYMBOL_GPL(vtime_account_system);
 200
 201 /*
 202  * Sorted add to a list. List is linear searched until first bigger
 203  * element is found.
 204  */
 205 static void list_add_sorted(struct vtimer_list *timer, struct list_head *head)
 206 {
 207         struct vtimer_list *tmp;
 208
 209         list_for_each_entry(tmp, head, entry) {
 210                 if (tmp->expires > timer->expires) {
 211                         list_add_tail(&timer->entry, &tmp->entry);
 212                         return;
 213                 }
 214         }
 215         list_add_tail(&timer->entry, head);
 216 }
 217
 218 /*
 219  * Handler for expired virtual CPU timer.
 220  */
 221 static void virt_timer_expire(void)
 222 {
 223         struct vtimer_list *timer, *tmp;
 224         unsigned long elapsed;
 225         LIST_HEAD(cb_list);
 226
 227         /* walk timer list, fire all expired timers */
 228         spin_lock(&virt_timer_lock);
 229         elapsed = atomic64_read(&virt_timer_elapsed);
 230         list_for_each_entry_safe(timer, tmp, &virt_timer_list, entry) {
 231                 if (timer->expires < elapsed)
 232                         /* move expired timer to the callback queue */
 233                         list_move_tail(&timer->entry, &cb_list);
 234                 else
 235                         timer->expires -= elapsed;
 236         }
 237         if (!list_empty(&virt_timer_list)) {
 238                 timer = list_first_entry(&virt_timer_list,
 239                                          struct vtimer_list, entry);
 240                 atomic64_set(&virt_timer_current, timer->expires);
 241         }
 242         atomic64_sub(elapsed, &virt_timer_elapsed);
 243         spin_unlock(&virt_timer_lock);
 244
 245         /* Do callbacks and recharge periodic timers */
 246         list_for_each_entry_safe(timer, tmp, &cb_list, entry) {
 247                 list_del_init(&timer->entry);
 248                 timer->function(timer->data);
 249                 if (timer->interval) {
 250                         /* Recharge interval timer */
 251                         timer->expires = timer->interval +
 252                                 atomic64_read(&virt_timer_elapsed);
 253                         spin_lock(&virt_timer_lock);
 254                         list_add_sorted(timer, &virt_timer_list);
 255                         spin_unlock(&virt_timer_lock);
 256                 }
 257         }
 258 }
 259
 260 void init_virt_timer(struct vtimer_list *timer)
 261 {
 262         timer->function = NULL;
 263         INIT_LIST_HEAD(&timer->entry);
 264 }
 265 EXPORT_SYMBOL(init_virt_timer);
 266
 267 static inline int vtimer_pending(struct vtimer_list *timer)
 268 {
 269         return !list_empty(&timer->entry);
 270 }
 271
 272 static void internal_add_vtimer(struct vtimer_list *timer)
 273 {
 274         if (list_empty(&virt_timer_list)) {
 275                 /* First timer, just program it. */
 276                 atomic64_set(&virt_timer_current, timer->expires);
 277                 atomic64_set(&virt_timer_elapsed, 0);
 278                 list_add(&timer->entry, &virt_timer_list);
 279         } else {
 280                 /* Update timer against current base. */
 281                 timer->expires += atomic64_read(&virt_timer_elapsed);
 282                 if (likely((s64) timer->expires <
 283                            (s64) atomic64_read(&virt_timer_current)))
 284                         /* The new timer expires before the current timer. */
 285                         atomic64_set(&virt_timer_current, timer->expires);
 286                 /* Insert new timer into the list. */
 287                 list_add_sorted(timer, &virt_timer_list);
 288         }
 289 }
 290
 291 static void __add_vtimer(struct vtimer_list *timer, int periodic)
 292 {
 293         unsigned long flags;
 294
 295         timer->interval = periodic ? timer->expires : 0;
 296         spin_lock_irqsave(&virt_timer_lock, flags);
 297         internal_add_vtimer(timer);
 298         spin_unlock_irqrestore(&virt_timer_lock, flags);
 299 }
 300
 301 /*
 302  * add_virt_timer - add an oneshot virtual CPU timer
 303  */
 304 void add_virt_timer(struct vtimer_list *timer)
 305 {
 306         __add_vtimer(timer, 0);
 307 }
 308 EXPORT_SYMBOL(add_virt_timer);
 309
 310 /*
 311  * add_virt_timer_int - add an interval virtual CPU timer
 312  */
 313 void add_virt_timer_periodic(struct vtimer_list *timer)
 314 {
 315         __add_vtimer(timer, 1);
 316 }
 317 EXPORT_SYMBOL(add_virt_timer_periodic);
 318
 319 static int __mod_vtimer(struct vtimer_list *timer, u64 expires, int periodic)
 320 {
 321         unsigned long flags;
 322         int rc;
 323
 324         BUG_ON(!timer->function);
 325
 326         if (timer->expires == expires && vtimer_pending(timer))
 327                 return 1;
 328         spin_lock_irqsave(&virt_timer_lock, flags);
 329         rc = vtimer_pending(timer);
 330         if (rc)
 331                 list_del_init(&timer->entry);
 332         timer->interval = periodic ? expires : 0;
 333         timer->expires = expires;
 334         internal_add_vtimer(timer);
 335         spin_unlock_irqrestore(&virt_timer_lock, flags);
 336         return rc;
 337 }
 338
 339 /*
 340  * returns whether it has modified a pending timer (1) or not (0)
 341  */
 342 int mod_virt_timer(struct vtimer_list *timer, u64 expires)
 343 {
 344         return __mod_vtimer(timer, expires, 0);
 345 }
 346 EXPORT_SYMBOL(mod_virt_timer);
 347
 348 /*
 349  * returns whether it has modified a pending timer (1) or not (0)
 350  */
 351 int mod_virt_timer_periodic(struct vtimer_list *timer, u64 expires)
 352 {
 353         return __mod_vtimer(timer, expires, 1);
 354 }
 355 EXPORT_SYMBOL(mod_virt_timer_periodic);
 356
 357 /*
 358  * Delete a virtual timer.
 359  *
 360  * returns whether the deleted timer was pending (1) or not (0)
 361  */
 362 int del_virt_timer(struct vtimer_list *timer)
 363 {
 364         unsigned long flags;
 365
 366         if (!vtimer_pending(timer))
 367                 return 0;
 368         spin_lock_irqsave(&virt_timer_lock, flags);
 369         list_del_init(&timer->entry);
 370         spin_unlock_irqrestore(&virt_timer_lock, flags);
 371         return 1;
 372 }
 373 EXPORT_SYMBOL(del_virt_timer);
 374
 375 /*
 376  * Start the virtual CPU timer on the current CPU.
 377  */
 378 void vtime_init(void)
 379 {
 380         /* set initial cpu timer */
 381         set_vtimer(VTIMER_MAX_SLICE);
 382         /* Setup initial MT scaling values */
 383         if (smp_cpu_mtid) {
 384                 __this_cpu_write(mt_scaling_jiffies, jiffies);
 385                 __this_cpu_write(mt_scaling_mult, 1);
 386                 __this_cpu_write(mt_scaling_div, 1);
 387                 stcctm5(smp_cpu_mtid + 1, this_cpu_ptr(mt_cycles));
 388         }
 389 }