xref: /linux/arch/s390/kernel/vtime.c (revision 63ba5b0fb4f54db256ec43b3062b2606b383055d)
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 	BUG_ON(!irqs_disabled());
52 
53 	if (list_empty(&virt_timer_list))
54 		return 0;
55 	elapsed = atomic64_add_return(elapsed, &virt_timer_elapsed);
56 	return elapsed >= atomic64_read(&virt_timer_current);
57 }
58 
59 static void update_mt_scaling(void)
60 {
61 	u64 cycles_new[8], *cycles_old;
62 	u64 delta, fac, mult, div;
63 	int i;
64 
65 	stcctm(MT_DIAG, smp_cpu_mtid + 1, cycles_new);
66 	cycles_old = this_cpu_ptr(mt_cycles);
67 	fac = 1;
68 	mult = div = 0;
69 	for (i = 0; i <= smp_cpu_mtid; i++) {
70 		delta = cycles_new[i] - cycles_old[i];
71 		div += delta;
72 		mult *= i + 1;
73 		mult += delta * fac;
74 		fac *= i + 1;
75 	}
76 	div *= fac;
77 	if (div > 0) {
78 		/* Update scaling factor */
79 		__this_cpu_write(mt_scaling_mult, mult);
80 		__this_cpu_write(mt_scaling_div, div);
81 		memcpy(cycles_old, cycles_new,
82 		       sizeof(u64) * (smp_cpu_mtid + 1));
83 	}
84 	__this_cpu_write(mt_scaling_jiffies, jiffies_64);
85 }
86 
87 static inline u64 update_tsk_timer(unsigned long *tsk_vtime, u64 new)
88 {
89 	u64 delta;
90 
91 	delta = new - *tsk_vtime;
92 	*tsk_vtime = new;
93 	return delta;
94 }
95 
96 
97 static inline u64 scale_vtime(u64 vtime)
98 {
99 	u64 mult = __this_cpu_read(mt_scaling_mult);
100 	u64 div = __this_cpu_read(mt_scaling_div);
101 
102 	if (smp_cpu_mtid)
103 		return vtime * mult / div;
104 	return vtime;
105 }
106 
107 static void account_system_index_scaled(struct task_struct *p, u64 cputime,
108 					enum cpu_usage_stat index)
109 {
110 	p->stimescaled += cputime_to_nsecs(scale_vtime(cputime));
111 	account_system_index_time(p, cputime_to_nsecs(cputime), index);
112 }
113 
114 /*
115  * Update process times based on virtual cpu times stored by entry.S
116  * to the lowcore fields user_timer, system_timer & steal_clock.
117  */
118 static int do_account_vtime(struct task_struct *tsk)
119 {
120 	u64 timer, clock, user, guest, system, hardirq, softirq;
121 	struct lowcore *lc = get_lowcore();
122 
123 	timer = lc->last_update_timer;
124 	clock = lc->last_update_clock;
125 	asm volatile(
126 		"	stpt	%0\n"	/* Store current cpu timer value */
127 		"	stckf	%1"	/* Store current tod clock value */
128 		: "=Q" (lc->last_update_timer),
129 		  "=Q" (lc->last_update_clock)
130 		: : "cc");
131 	clock = lc->last_update_clock - clock;
132 	timer -= lc->last_update_timer;
133 
134 	if (hardirq_count())
135 		lc->hardirq_timer += timer;
136 	else
137 		lc->system_timer += timer;
138 
139 	/* Update MT utilization calculation */
140 	if (smp_cpu_mtid &&
141 	    time_after64(jiffies_64, this_cpu_read(mt_scaling_jiffies)))
142 		update_mt_scaling();
143 
144 	/* Calculate cputime delta */
145 	user = update_tsk_timer(&tsk->thread.user_timer,
146 				READ_ONCE(lc->user_timer));
147 	guest = update_tsk_timer(&tsk->thread.guest_timer,
148 				 READ_ONCE(lc->guest_timer));
149 	system = update_tsk_timer(&tsk->thread.system_timer,
150 				  READ_ONCE(lc->system_timer));
151 	hardirq = update_tsk_timer(&tsk->thread.hardirq_timer,
152 				   READ_ONCE(lc->hardirq_timer));
153 	softirq = update_tsk_timer(&tsk->thread.softirq_timer,
154 				   READ_ONCE(lc->softirq_timer));
155 	lc->steal_timer +=
156 		clock - user - guest - system - hardirq - softirq;
157 
158 	/* Push account value */
159 	if (user) {
160 		account_user_time(tsk, cputime_to_nsecs(user));
161 		tsk->utimescaled += cputime_to_nsecs(scale_vtime(user));
162 	}
163 
164 	if (guest) {
165 		account_guest_time(tsk, cputime_to_nsecs(guest));
166 		tsk->utimescaled += cputime_to_nsecs(scale_vtime(guest));
167 	}
168 
169 	if (system)
170 		account_system_index_scaled(tsk, system, CPUTIME_SYSTEM);
171 	if (hardirq)
172 		account_system_index_scaled(tsk, hardirq, CPUTIME_IRQ);
173 	if (softirq)
174 		account_system_index_scaled(tsk, softirq, CPUTIME_SOFTIRQ);
175 
176 	return virt_timer_forward(user + guest + system + hardirq + softirq);
177 }
178 
179 void vtime_task_switch(struct task_struct *prev)
180 {
181 	struct lowcore *lc = get_lowcore();
182 
183 	do_account_vtime(prev);
184 	prev->thread.user_timer = lc->user_timer;
185 	prev->thread.guest_timer = lc->guest_timer;
186 	prev->thread.system_timer = lc->system_timer;
187 	prev->thread.hardirq_timer = lc->hardirq_timer;
188 	prev->thread.softirq_timer = lc->softirq_timer;
189 	lc->user_timer = current->thread.user_timer;
190 	lc->guest_timer = current->thread.guest_timer;
191 	lc->system_timer = current->thread.system_timer;
192 	lc->hardirq_timer = current->thread.hardirq_timer;
193 	lc->softirq_timer = current->thread.softirq_timer;
194 }
195 
196 /*
197  * In s390, accounting pending user time also implies
198  * accounting system time in order to correctly compute
199  * the stolen time accounting.
200  */
201 void vtime_flush(struct task_struct *tsk)
202 {
203 	struct lowcore *lc = get_lowcore();
204 	u64 steal, avg_steal;
205 
206 	if (do_account_vtime(tsk))
207 		virt_timer_expire();
208 
209 	steal = lc->steal_timer;
210 	avg_steal = lc->avg_steal_timer;
211 	if ((s64) steal > 0) {
212 		lc->steal_timer = 0;
213 		account_steal_time(cputime_to_nsecs(steal));
214 		avg_steal += steal;
215 	}
216 	lc->avg_steal_timer = avg_steal / 2;
217 }
218 
219 static u64 vtime_delta(void)
220 {
221 	struct lowcore *lc = get_lowcore();
222 	u64 timer = lc->last_update_timer;
223 
224 	lc->last_update_timer = get_cpu_timer();
225 	return timer - lc->last_update_timer;
226 }
227 
228 /*
229  * Update process times based on virtual cpu times stored by entry.S
230  * to the lowcore fields user_timer, system_timer & steal_clock.
231  */
232 void vtime_account_kernel(struct task_struct *tsk)
233 {
234 	struct lowcore *lc = get_lowcore();
235 	u64 delta = vtime_delta();
236 
237 	if (tsk->flags & PF_VCPU)
238 		lc->guest_timer += delta;
239 	else
240 		lc->system_timer += delta;
241 
242 	virt_timer_forward(delta);
243 }
244 EXPORT_SYMBOL_GPL(vtime_account_kernel);
245 
246 void vtime_account_softirq(struct task_struct *tsk)
247 {
248 	u64 delta = vtime_delta();
249 
250 	get_lowcore()->softirq_timer += delta;
251 
252 	virt_timer_forward(delta);
253 }
254 
255 void vtime_account_hardirq(struct task_struct *tsk)
256 {
257 	u64 delta = vtime_delta();
258 
259 	get_lowcore()->hardirq_timer += delta;
260 
261 	virt_timer_forward(delta);
262 }
263 
264 /*
265  * Sorted add to a list. List is linear searched until first bigger
266  * element is found.
267  */
268 static void list_add_sorted(struct vtimer_list *timer, struct list_head *head)
269 {
270 	struct vtimer_list *tmp;
271 
272 	list_for_each_entry(tmp, head, entry) {
273 		if (tmp->expires > timer->expires) {
274 			list_add_tail(&timer->entry, &tmp->entry);
275 			return;
276 		}
277 	}
278 	list_add_tail(&timer->entry, head);
279 }
280 
281 /*
282  * Handler for expired virtual CPU timer.
283  */
284 static void virt_timer_expire(void)
285 {
286 	struct vtimer_list *timer, *tmp;
287 	unsigned long elapsed;
288 	LIST_HEAD(cb_list);
289 
290 	/* walk timer list, fire all expired timers */
291 	spin_lock(&virt_timer_lock);
292 	elapsed = atomic64_read(&virt_timer_elapsed);
293 	list_for_each_entry_safe(timer, tmp, &virt_timer_list, entry) {
294 		if (timer->expires < elapsed)
295 			/* move expired timer to the callback queue */
296 			list_move_tail(&timer->entry, &cb_list);
297 		else
298 			timer->expires -= elapsed;
299 	}
300 	if (!list_empty(&virt_timer_list)) {
301 		timer = list_first_entry(&virt_timer_list,
302 					 struct vtimer_list, entry);
303 		atomic64_set(&virt_timer_current, timer->expires);
304 	}
305 	atomic64_sub(elapsed, &virt_timer_elapsed);
306 	spin_unlock(&virt_timer_lock);
307 
308 	/* Do callbacks and recharge periodic timers */
309 	list_for_each_entry_safe(timer, tmp, &cb_list, entry) {
310 		list_del_init(&timer->entry);
311 		timer->function(timer->data);
312 		if (timer->interval) {
313 			/* Recharge interval timer */
314 			timer->expires = timer->interval +
315 				atomic64_read(&virt_timer_elapsed);
316 			spin_lock(&virt_timer_lock);
317 			list_add_sorted(timer, &virt_timer_list);
318 			spin_unlock(&virt_timer_lock);
319 		}
320 	}
321 }
322 
323 void init_virt_timer(struct vtimer_list *timer)
324 {
325 	timer->function = NULL;
326 	INIT_LIST_HEAD(&timer->entry);
327 }
328 EXPORT_SYMBOL(init_virt_timer);
329 
330 static inline int vtimer_pending(struct vtimer_list *timer)
331 {
332 	return !list_empty(&timer->entry);
333 }
334 
335 static void internal_add_vtimer(struct vtimer_list *timer)
336 {
337 	if (list_empty(&virt_timer_list)) {
338 		/* First timer, just program it. */
339 		atomic64_set(&virt_timer_current, timer->expires);
340 		atomic64_set(&virt_timer_elapsed, 0);
341 		list_add(&timer->entry, &virt_timer_list);
342 	} else {
343 		/* Update timer against current base. */
344 		timer->expires += atomic64_read(&virt_timer_elapsed);
345 		if (likely((s64) timer->expires <
346 			   (s64) atomic64_read(&virt_timer_current)))
347 			/* The new timer expires before the current timer. */
348 			atomic64_set(&virt_timer_current, timer->expires);
349 		/* Insert new timer into the list. */
350 		list_add_sorted(timer, &virt_timer_list);
351 	}
352 }
353 
354 static void __add_vtimer(struct vtimer_list *timer, int periodic)
355 {
356 	unsigned long flags;
357 
358 	timer->interval = periodic ? timer->expires : 0;
359 	spin_lock_irqsave(&virt_timer_lock, flags);
360 	internal_add_vtimer(timer);
361 	spin_unlock_irqrestore(&virt_timer_lock, flags);
362 }
363 
364 /*
365  * add_virt_timer - add a oneshot virtual CPU timer
366  */
367 void add_virt_timer(struct vtimer_list *timer)
368 {
369 	__add_vtimer(timer, 0);
370 }
371 EXPORT_SYMBOL(add_virt_timer);
372 
373 /*
374  * add_virt_timer_int - add an interval virtual CPU timer
375  */
376 void add_virt_timer_periodic(struct vtimer_list *timer)
377 {
378 	__add_vtimer(timer, 1);
379 }
380 EXPORT_SYMBOL(add_virt_timer_periodic);
381 
382 static int __mod_vtimer(struct vtimer_list *timer, u64 expires, int periodic)
383 {
384 	unsigned long flags;
385 	int rc;
386 
387 	BUG_ON(!timer->function);
388 
389 	if (timer->expires == expires && vtimer_pending(timer))
390 		return 1;
391 	spin_lock_irqsave(&virt_timer_lock, flags);
392 	rc = vtimer_pending(timer);
393 	if (rc)
394 		list_del_init(&timer->entry);
395 	timer->interval = periodic ? expires : 0;
396 	timer->expires = expires;
397 	internal_add_vtimer(timer);
398 	spin_unlock_irqrestore(&virt_timer_lock, flags);
399 	return rc;
400 }
401 
402 /*
403  * returns whether it has modified a pending timer (1) or not (0)
404  */
405 int mod_virt_timer(struct vtimer_list *timer, u64 expires)
406 {
407 	return __mod_vtimer(timer, expires, 0);
408 }
409 EXPORT_SYMBOL(mod_virt_timer);
410 
411 /*
412  * returns whether it has modified a pending timer (1) or not (0)
413  */
414 int mod_virt_timer_periodic(struct vtimer_list *timer, u64 expires)
415 {
416 	return __mod_vtimer(timer, expires, 1);
417 }
418 EXPORT_SYMBOL(mod_virt_timer_periodic);
419 
420 /*
421  * Delete a virtual timer.
422  *
423  * returns whether the deleted timer was pending (1) or not (0)
424  */
425 int del_virt_timer(struct vtimer_list *timer)
426 {
427 	unsigned long flags;
428 
429 	if (!vtimer_pending(timer))
430 		return 0;
431 	spin_lock_irqsave(&virt_timer_lock, flags);
432 	list_del_init(&timer->entry);
433 	spin_unlock_irqrestore(&virt_timer_lock, flags);
434 	return 1;
435 }
436 EXPORT_SYMBOL(del_virt_timer);
437 
438 /*
439  * Start the virtual CPU timer on the current CPU.
440  */
441 void vtime_init(void)
442 {
443 	/* set initial cpu timer */
444 	set_vtimer(VTIMER_MAX_SLICE);
445 	/* Setup initial MT scaling values */
446 	if (smp_cpu_mtid) {
447 		__this_cpu_write(mt_scaling_jiffies, jiffies);
448 		__this_cpu_write(mt_scaling_mult, 1);
449 		__this_cpu_write(mt_scaling_div, 1);
450 		stcctm(MT_DIAG, smp_cpu_mtid + 1, this_cpu_ptr(mt_cycles));
451 	}
452 }
453