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/export.h> 11 #include <linux/kernel.h> 12 #include <linux/timex.h> 13 #include <linux/types.h> 14 #include <linux/time.h> 15 #include <asm/alternative.h> 16 #include <asm/cputime.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 void set_vtimer(u64 expires) 37 { 38 struct lowcore *lc = get_lowcore(); 39 u64 timer; 40 41 asm volatile( 42 " stpt %0\n" /* Store current cpu timer value */ 43 " spt %1" /* Set new value imm. afterwards */ 44 : "=Q" (timer) : "Q" (expires)); 45 lc->system_timer += lc->last_update_timer - timer; 46 lc->last_update_timer = expires; 47 } 48 49 static inline int virt_timer_forward(u64 elapsed) 50 { 51 lockdep_assert_irqs_disabled(); 52 if (list_empty(&virt_timer_list)) 53 return 0; 54 elapsed = atomic64_add_return(elapsed, &virt_timer_elapsed); 55 return elapsed >= atomic64_read(&virt_timer_current); 56 } 57 58 static void update_mt_scaling(void) 59 { 60 u64 cycles_new[8], *cycles_old; 61 u64 delta, fac, mult, div; 62 int i; 63 64 stcctm(MT_DIAG, smp_cpu_mtid + 1, cycles_new); 65 cycles_old = this_cpu_ptr(mt_cycles); 66 fac = 1; 67 mult = div = 0; 68 for (i = 0; i <= smp_cpu_mtid; i++) { 69 delta = cycles_new[i] - cycles_old[i]; 70 div += delta; 71 mult *= i + 1; 72 mult += delta * fac; 73 fac *= i + 1; 74 } 75 div *= fac; 76 if (div > 0) { 77 /* Update scaling factor */ 78 __this_cpu_write(mt_scaling_mult, mult); 79 __this_cpu_write(mt_scaling_div, div); 80 memcpy(cycles_old, cycles_new, 81 sizeof(u64) * (smp_cpu_mtid + 1)); 82 } 83 __this_cpu_write(mt_scaling_jiffies, jiffies_64); 84 } 85 86 static inline u64 update_tsk_timer(unsigned long *tsk_vtime, u64 new) 87 { 88 u64 delta; 89 90 delta = new - *tsk_vtime; 91 *tsk_vtime = new; 92 return delta; 93 } 94 95 96 static inline u64 scale_vtime(u64 vtime) 97 { 98 u64 mult = __this_cpu_read(mt_scaling_mult); 99 u64 div = __this_cpu_read(mt_scaling_div); 100 101 if (smp_cpu_mtid) 102 return vtime * mult / div; 103 return vtime; 104 } 105 106 static void account_system_index_scaled(struct task_struct *p, u64 cputime, 107 enum cpu_usage_stat index) 108 { 109 p->stimescaled += cputime_to_nsecs(scale_vtime(cputime)); 110 account_system_index_time(p, cputime_to_nsecs(cputime), index); 111 } 112 113 /* 114 * Update process times based on virtual cpu times stored by entry.S 115 * to the lowcore fields user_timer, system_timer & steal_clock. 116 */ 117 static int do_account_vtime(struct task_struct *tsk) 118 { 119 u64 timer, clock, user, guest, system, hardirq, softirq; 120 struct lowcore *lc = get_lowcore(); 121 122 timer = lc->last_update_timer; 123 clock = lc->last_update_clock; 124 asm volatile( 125 " stpt %0\n" /* Store current cpu timer value */ 126 " stckf %1" /* Store current tod clock value */ 127 : "=Q" (lc->last_update_timer), 128 "=Q" (lc->last_update_clock) 129 : : "cc"); 130 clock = lc->last_update_clock - clock; 131 timer -= lc->last_update_timer; 132 133 if (hardirq_count()) 134 lc->hardirq_timer += timer; 135 else 136 lc->system_timer += timer; 137 138 /* Update MT utilization calculation */ 139 if (smp_cpu_mtid && time_after64(jiffies_64, __this_cpu_read(mt_scaling_jiffies))) 140 update_mt_scaling(); 141 142 /* Calculate cputime delta */ 143 user = update_tsk_timer(&tsk->thread.user_timer, lc->user_timer); 144 guest = update_tsk_timer(&tsk->thread.guest_timer, lc->guest_timer); 145 system = update_tsk_timer(&tsk->thread.system_timer, lc->system_timer); 146 hardirq = update_tsk_timer(&tsk->thread.hardirq_timer, lc->hardirq_timer); 147 softirq = update_tsk_timer(&tsk->thread.softirq_timer, lc->softirq_timer); 148 lc->steal_timer += clock - user - guest - system - hardirq - softirq; 149 150 /* Push account value */ 151 if (user) { 152 account_user_time(tsk, cputime_to_nsecs(user)); 153 tsk->utimescaled += cputime_to_nsecs(scale_vtime(user)); 154 } 155 156 if (guest) { 157 account_guest_time(tsk, cputime_to_nsecs(guest)); 158 tsk->utimescaled += cputime_to_nsecs(scale_vtime(guest)); 159 } 160 161 if (system) 162 account_system_index_scaled(tsk, system, CPUTIME_SYSTEM); 163 if (hardirq) 164 account_system_index_scaled(tsk, hardirq, CPUTIME_IRQ); 165 if (softirq) 166 account_system_index_scaled(tsk, softirq, CPUTIME_SOFTIRQ); 167 168 return virt_timer_forward(user + guest + system + hardirq + softirq); 169 } 170 171 void vtime_task_switch(struct task_struct *prev) 172 { 173 struct lowcore *lc = get_lowcore(); 174 175 do_account_vtime(prev); 176 prev->thread.user_timer = lc->user_timer; 177 prev->thread.guest_timer = lc->guest_timer; 178 prev->thread.system_timer = lc->system_timer; 179 prev->thread.hardirq_timer = lc->hardirq_timer; 180 prev->thread.softirq_timer = lc->softirq_timer; 181 lc->user_timer = current->thread.user_timer; 182 lc->guest_timer = current->thread.guest_timer; 183 lc->system_timer = current->thread.system_timer; 184 lc->hardirq_timer = current->thread.hardirq_timer; 185 lc->softirq_timer = current->thread.softirq_timer; 186 } 187 188 /* 189 * In s390, accounting pending user time also implies 190 * accounting system time in order to correctly compute 191 * the stolen time accounting. 192 */ 193 void vtime_flush(struct task_struct *tsk) 194 { 195 struct lowcore *lc = get_lowcore(); 196 u64 steal, avg_steal; 197 198 if (do_account_vtime(tsk)) 199 virt_timer_expire(); 200 201 steal = lc->steal_timer; 202 avg_steal = lc->avg_steal_timer; 203 if ((s64) steal > 0) { 204 lc->steal_timer = 0; 205 account_steal_time(cputime_to_nsecs(steal)); 206 avg_steal += steal; 207 } 208 lc->avg_steal_timer = avg_steal / 2; 209 } 210 211 static u64 vtime_delta(void) 212 { 213 struct lowcore *lc = get_lowcore(); 214 u64 timer = lc->last_update_timer; 215 216 lc->last_update_timer = get_cpu_timer(); 217 return timer - lc->last_update_timer; 218 } 219 220 void vtime_account_kernel(struct task_struct *tsk) 221 { 222 struct lowcore *lc = get_lowcore(); 223 u64 delta = vtime_delta(); 224 225 if (tsk->flags & PF_VCPU) 226 lc->guest_timer += delta; 227 else 228 lc->system_timer += delta; 229 } 230 EXPORT_SYMBOL_GPL(vtime_account_kernel); 231 232 void vtime_account_softirq(struct task_struct *tsk) 233 { 234 get_lowcore()->softirq_timer += vtime_delta(); 235 } 236 237 void vtime_account_hardirq(struct task_struct *tsk) 238 { 239 get_lowcore()->hardirq_timer += vtime_delta(); 240 } 241 242 /* 243 * Sorted add to a list. List is linear searched until first bigger 244 * element is found. 245 */ 246 static void list_add_sorted(struct vtimer_list *timer, struct list_head *head) 247 { 248 struct vtimer_list *tmp; 249 250 list_for_each_entry(tmp, head, entry) { 251 if (tmp->expires > timer->expires) { 252 list_add_tail(&timer->entry, &tmp->entry); 253 return; 254 } 255 } 256 list_add_tail(&timer->entry, head); 257 } 258 259 /* 260 * Handler for expired virtual CPU timer. 261 */ 262 static void virt_timer_expire(void) 263 { 264 struct vtimer_list *timer, *tmp; 265 unsigned long elapsed; 266 LIST_HEAD(cb_list); 267 268 /* walk timer list, fire all expired timers */ 269 spin_lock(&virt_timer_lock); 270 elapsed = atomic64_read(&virt_timer_elapsed); 271 list_for_each_entry_safe(timer, tmp, &virt_timer_list, entry) { 272 if (timer->expires < elapsed) 273 /* move expired timer to the callback queue */ 274 list_move_tail(&timer->entry, &cb_list); 275 else 276 timer->expires -= elapsed; 277 } 278 if (!list_empty(&virt_timer_list)) { 279 timer = list_first_entry(&virt_timer_list, 280 struct vtimer_list, entry); 281 atomic64_set(&virt_timer_current, timer->expires); 282 } 283 atomic64_sub(elapsed, &virt_timer_elapsed); 284 spin_unlock(&virt_timer_lock); 285 286 /* Do callbacks and recharge periodic timers */ 287 list_for_each_entry_safe(timer, tmp, &cb_list, entry) { 288 list_del_init(&timer->entry); 289 timer->function(timer->data); 290 if (timer->interval) { 291 /* Recharge interval timer */ 292 timer->expires = timer->interval + 293 atomic64_read(&virt_timer_elapsed); 294 spin_lock(&virt_timer_lock); 295 list_add_sorted(timer, &virt_timer_list); 296 spin_unlock(&virt_timer_lock); 297 } 298 } 299 } 300 301 void init_virt_timer(struct vtimer_list *timer) 302 { 303 timer->function = NULL; 304 INIT_LIST_HEAD(&timer->entry); 305 } 306 EXPORT_SYMBOL(init_virt_timer); 307 308 static inline int vtimer_pending(struct vtimer_list *timer) 309 { 310 return !list_empty(&timer->entry); 311 } 312 313 static void internal_add_vtimer(struct vtimer_list *timer) 314 { 315 if (list_empty(&virt_timer_list)) { 316 /* First timer, just program it. */ 317 atomic64_set(&virt_timer_current, timer->expires); 318 atomic64_set(&virt_timer_elapsed, 0); 319 list_add(&timer->entry, &virt_timer_list); 320 } else { 321 /* Update timer against current base. */ 322 timer->expires += atomic64_read(&virt_timer_elapsed); 323 if (likely((s64) timer->expires < 324 (s64) atomic64_read(&virt_timer_current))) 325 /* The new timer expires before the current timer. */ 326 atomic64_set(&virt_timer_current, timer->expires); 327 /* Insert new timer into the list. */ 328 list_add_sorted(timer, &virt_timer_list); 329 } 330 } 331 332 static void __add_vtimer(struct vtimer_list *timer, int periodic) 333 { 334 unsigned long flags; 335 336 timer->interval = periodic ? timer->expires : 0; 337 spin_lock_irqsave(&virt_timer_lock, flags); 338 internal_add_vtimer(timer); 339 spin_unlock_irqrestore(&virt_timer_lock, flags); 340 } 341 342 /* 343 * add_virt_timer - add a oneshot virtual CPU timer 344 */ 345 void add_virt_timer(struct vtimer_list *timer) 346 { 347 __add_vtimer(timer, 0); 348 } 349 EXPORT_SYMBOL(add_virt_timer); 350 351 /* 352 * add_virt_timer_int - add an interval virtual CPU timer 353 */ 354 void add_virt_timer_periodic(struct vtimer_list *timer) 355 { 356 __add_vtimer(timer, 1); 357 } 358 EXPORT_SYMBOL(add_virt_timer_periodic); 359 360 static int __mod_vtimer(struct vtimer_list *timer, u64 expires, int periodic) 361 { 362 unsigned long flags; 363 int rc; 364 365 BUG_ON(!timer->function); 366 367 if (timer->expires == expires && vtimer_pending(timer)) 368 return 1; 369 spin_lock_irqsave(&virt_timer_lock, flags); 370 rc = vtimer_pending(timer); 371 if (rc) 372 list_del_init(&timer->entry); 373 timer->interval = periodic ? expires : 0; 374 timer->expires = expires; 375 internal_add_vtimer(timer); 376 spin_unlock_irqrestore(&virt_timer_lock, flags); 377 return rc; 378 } 379 380 /* 381 * returns whether it has modified a pending timer (1) or not (0) 382 */ 383 int mod_virt_timer(struct vtimer_list *timer, u64 expires) 384 { 385 return __mod_vtimer(timer, expires, 0); 386 } 387 EXPORT_SYMBOL(mod_virt_timer); 388 389 /* 390 * returns whether it has modified a pending timer (1) or not (0) 391 */ 392 int mod_virt_timer_periodic(struct vtimer_list *timer, u64 expires) 393 { 394 return __mod_vtimer(timer, expires, 1); 395 } 396 EXPORT_SYMBOL(mod_virt_timer_periodic); 397 398 /* 399 * Delete a virtual timer. 400 * 401 * returns whether the deleted timer was pending (1) or not (0) 402 */ 403 int del_virt_timer(struct vtimer_list *timer) 404 { 405 unsigned long flags; 406 407 if (!vtimer_pending(timer)) 408 return 0; 409 spin_lock_irqsave(&virt_timer_lock, flags); 410 list_del_init(&timer->entry); 411 spin_unlock_irqrestore(&virt_timer_lock, flags); 412 return 1; 413 } 414 EXPORT_SYMBOL(del_virt_timer); 415 416 /* 417 * Start the virtual CPU timer on the current CPU. 418 */ 419 void vtime_init(void) 420 { 421 /* set initial cpu timer */ 422 set_vtimer(VTIMER_MAX_SLICE); 423 /* Setup initial MT scaling values */ 424 if (smp_cpu_mtid) { 425 __this_cpu_write(mt_scaling_jiffies, jiffies); 426 __this_cpu_write(mt_scaling_mult, 1); 427 __this_cpu_write(mt_scaling_div, 1); 428 stcctm(MT_DIAG, smp_cpu_mtid + 1, this_cpu_ptr(mt_cycles)); 429 } 430 } 431