xref: /openbmc/linux/arch/s390/kernel/vtime.c (revision f9834f18)
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/sched/cputime.h>
11 #include <linux/export.h>
12 #include <linux/kernel.h>
13 #include <linux/timex.h>
14 #include <linux/types.h>
15 #include <linux/time.h>
16 
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 u64 get_vtimer(void)
37 {
38 	u64 timer;
39 
40 	asm volatile("stpt %0" : "=Q" (timer));
41 	return timer;
42 }
43 
44 static inline void set_vtimer(u64 expires)
45 {
46 	u64 timer;
47 
48 	asm volatile(
49 		"	stpt	%0\n"	/* Store current cpu timer value */
50 		"	spt	%1"	/* Set new value imm. afterwards */
51 		: "=Q" (timer) : "Q" (expires));
52 	S390_lowcore.system_timer += S390_lowcore.last_update_timer - timer;
53 	S390_lowcore.last_update_timer = expires;
54 }
55 
56 static inline int virt_timer_forward(u64 elapsed)
57 {
58 	BUG_ON(!irqs_disabled());
59 
60 	if (list_empty(&virt_timer_list))
61 		return 0;
62 	elapsed = atomic64_add_return(elapsed, &virt_timer_elapsed);
63 	return elapsed >= atomic64_read(&virt_timer_current);
64 }
65 
66 static void update_mt_scaling(void)
67 {
68 	u64 cycles_new[8], *cycles_old;
69 	u64 delta, fac, mult, div;
70 	int i;
71 
72 	stcctm(MT_DIAG, smp_cpu_mtid + 1, cycles_new);
73 	cycles_old = this_cpu_ptr(mt_cycles);
74 	fac = 1;
75 	mult = div = 0;
76 	for (i = 0; i <= smp_cpu_mtid; i++) {
77 		delta = cycles_new[i] - cycles_old[i];
78 		div += delta;
79 		mult *= i + 1;
80 		mult += delta * fac;
81 		fac *= i + 1;
82 	}
83 	div *= fac;
84 	if (div > 0) {
85 		/* Update scaling factor */
86 		__this_cpu_write(mt_scaling_mult, mult);
87 		__this_cpu_write(mt_scaling_div, div);
88 		memcpy(cycles_old, cycles_new,
89 		       sizeof(u64) * (smp_cpu_mtid + 1));
90 	}
91 	__this_cpu_write(mt_scaling_jiffies, jiffies_64);
92 }
93 
94 static inline u64 update_tsk_timer(unsigned long *tsk_vtime, u64 new)
95 {
96 	u64 delta;
97 
98 	delta = new - *tsk_vtime;
99 	*tsk_vtime = new;
100 	return delta;
101 }
102 
103 
104 static inline u64 scale_vtime(u64 vtime)
105 {
106 	u64 mult = __this_cpu_read(mt_scaling_mult);
107 	u64 div = __this_cpu_read(mt_scaling_div);
108 
109 	if (smp_cpu_mtid)
110 		return vtime * mult / div;
111 	return vtime;
112 }
113 
114 static void account_system_index_scaled(struct task_struct *p, u64 cputime,
115 					enum cpu_usage_stat index)
116 {
117 	p->stimescaled += cputime_to_nsecs(scale_vtime(cputime));
118 	account_system_index_time(p, cputime_to_nsecs(cputime), index);
119 }
120 
121 /*
122  * Update process times based on virtual cpu times stored by entry.S
123  * to the lowcore fields user_timer, system_timer & steal_clock.
124  */
125 static int do_account_vtime(struct task_struct *tsk)
126 {
127 	u64 timer, clock, user, guest, system, hardirq, softirq;
128 
129 	timer = S390_lowcore.last_update_timer;
130 	clock = S390_lowcore.last_update_clock;
131 	asm volatile(
132 		"	stpt	%0\n"	/* Store current cpu timer value */
133 #ifdef CONFIG_HAVE_MARCH_Z9_109_FEATURES
134 		"	stckf	%1"	/* Store current tod clock value */
135 #else
136 		"	stck	%1"	/* Store current tod clock value */
137 #endif
138 		: "=Q" (S390_lowcore.last_update_timer),
139 		  "=Q" (S390_lowcore.last_update_clock));
140 	clock = S390_lowcore.last_update_clock - clock;
141 	timer -= S390_lowcore.last_update_timer;
142 
143 	if (hardirq_count())
144 		S390_lowcore.hardirq_timer += timer;
145 	else
146 		S390_lowcore.system_timer += timer;
147 
148 	/* Update MT utilization calculation */
149 	if (smp_cpu_mtid &&
150 	    time_after64(jiffies_64, this_cpu_read(mt_scaling_jiffies)))
151 		update_mt_scaling();
152 
153 	/* Calculate cputime delta */
154 	user = update_tsk_timer(&tsk->thread.user_timer,
155 				READ_ONCE(S390_lowcore.user_timer));
156 	guest = update_tsk_timer(&tsk->thread.guest_timer,
157 				 READ_ONCE(S390_lowcore.guest_timer));
158 	system = update_tsk_timer(&tsk->thread.system_timer,
159 				  READ_ONCE(S390_lowcore.system_timer));
160 	hardirq = update_tsk_timer(&tsk->thread.hardirq_timer,
161 				   READ_ONCE(S390_lowcore.hardirq_timer));
162 	softirq = update_tsk_timer(&tsk->thread.softirq_timer,
163 				   READ_ONCE(S390_lowcore.softirq_timer));
164 	S390_lowcore.steal_timer +=
165 		clock - user - guest - system - hardirq - softirq;
166 
167 	/* Push account value */
168 	if (user) {
169 		account_user_time(tsk, cputime_to_nsecs(user));
170 		tsk->utimescaled += cputime_to_nsecs(scale_vtime(user));
171 	}
172 
173 	if (guest) {
174 		account_guest_time(tsk, cputime_to_nsecs(guest));
175 		tsk->utimescaled += cputime_to_nsecs(scale_vtime(guest));
176 	}
177 
178 	if (system)
179 		account_system_index_scaled(tsk, system, CPUTIME_SYSTEM);
180 	if (hardirq)
181 		account_system_index_scaled(tsk, hardirq, CPUTIME_IRQ);
182 	if (softirq)
183 		account_system_index_scaled(tsk, softirq, CPUTIME_SOFTIRQ);
184 
185 	return virt_timer_forward(user + guest + system + hardirq + softirq);
186 }
187 
188 void vtime_task_switch(struct task_struct *prev)
189 {
190 	do_account_vtime(prev);
191 	prev->thread.user_timer = S390_lowcore.user_timer;
192 	prev->thread.guest_timer = S390_lowcore.guest_timer;
193 	prev->thread.system_timer = S390_lowcore.system_timer;
194 	prev->thread.hardirq_timer = S390_lowcore.hardirq_timer;
195 	prev->thread.softirq_timer = S390_lowcore.softirq_timer;
196 	S390_lowcore.user_timer = current->thread.user_timer;
197 	S390_lowcore.guest_timer = current->thread.guest_timer;
198 	S390_lowcore.system_timer = current->thread.system_timer;
199 	S390_lowcore.hardirq_timer = current->thread.hardirq_timer;
200 	S390_lowcore.softirq_timer = current->thread.softirq_timer;
201 }
202 
203 /*
204  * In s390, accounting pending user time also implies
205  * accounting system time in order to correctly compute
206  * the stolen time accounting.
207  */
208 void vtime_flush(struct task_struct *tsk)
209 {
210 	u64 steal, avg_steal;
211 
212 	if (do_account_vtime(tsk))
213 		virt_timer_expire();
214 
215 	steal = S390_lowcore.steal_timer;
216 	avg_steal = S390_lowcore.avg_steal_timer / 2;
217 	if ((s64) steal > 0) {
218 		S390_lowcore.steal_timer = 0;
219 		account_steal_time(steal);
220 		avg_steal += steal;
221 	}
222 	S390_lowcore.avg_steal_timer = avg_steal;
223 }
224 
225 /*
226  * Update process times based on virtual cpu times stored by entry.S
227  * to the lowcore fields user_timer, system_timer & steal_clock.
228  */
229 void vtime_account_irq_enter(struct task_struct *tsk)
230 {
231 	u64 timer;
232 
233 	timer = S390_lowcore.last_update_timer;
234 	S390_lowcore.last_update_timer = get_vtimer();
235 	timer -= S390_lowcore.last_update_timer;
236 
237 	if ((tsk->flags & PF_VCPU) && (irq_count() == 0))
238 		S390_lowcore.guest_timer += timer;
239 	else if (hardirq_count())
240 		S390_lowcore.hardirq_timer += timer;
241 	else if (in_serving_softirq())
242 		S390_lowcore.softirq_timer += timer;
243 	else
244 		S390_lowcore.system_timer += timer;
245 
246 	virt_timer_forward(timer);
247 }
248 EXPORT_SYMBOL_GPL(vtime_account_irq_enter);
249 
250 void vtime_account_kernel(struct task_struct *tsk)
251 __attribute__((alias("vtime_account_irq_enter")));
252 EXPORT_SYMBOL_GPL(vtime_account_kernel);
253 
254 /*
255  * Sorted add to a list. List is linear searched until first bigger
256  * element is found.
257  */
258 static void list_add_sorted(struct vtimer_list *timer, struct list_head *head)
259 {
260 	struct vtimer_list *tmp;
261 
262 	list_for_each_entry(tmp, head, entry) {
263 		if (tmp->expires > timer->expires) {
264 			list_add_tail(&timer->entry, &tmp->entry);
265 			return;
266 		}
267 	}
268 	list_add_tail(&timer->entry, head);
269 }
270 
271 /*
272  * Handler for expired virtual CPU timer.
273  */
274 static void virt_timer_expire(void)
275 {
276 	struct vtimer_list *timer, *tmp;
277 	unsigned long elapsed;
278 	LIST_HEAD(cb_list);
279 
280 	/* walk timer list, fire all expired timers */
281 	spin_lock(&virt_timer_lock);
282 	elapsed = atomic64_read(&virt_timer_elapsed);
283 	list_for_each_entry_safe(timer, tmp, &virt_timer_list, entry) {
284 		if (timer->expires < elapsed)
285 			/* move expired timer to the callback queue */
286 			list_move_tail(&timer->entry, &cb_list);
287 		else
288 			timer->expires -= elapsed;
289 	}
290 	if (!list_empty(&virt_timer_list)) {
291 		timer = list_first_entry(&virt_timer_list,
292 					 struct vtimer_list, entry);
293 		atomic64_set(&virt_timer_current, timer->expires);
294 	}
295 	atomic64_sub(elapsed, &virt_timer_elapsed);
296 	spin_unlock(&virt_timer_lock);
297 
298 	/* Do callbacks and recharge periodic timers */
299 	list_for_each_entry_safe(timer, tmp, &cb_list, entry) {
300 		list_del_init(&timer->entry);
301 		timer->function(timer->data);
302 		if (timer->interval) {
303 			/* Recharge interval timer */
304 			timer->expires = timer->interval +
305 				atomic64_read(&virt_timer_elapsed);
306 			spin_lock(&virt_timer_lock);
307 			list_add_sorted(timer, &virt_timer_list);
308 			spin_unlock(&virt_timer_lock);
309 		}
310 	}
311 }
312 
313 void init_virt_timer(struct vtimer_list *timer)
314 {
315 	timer->function = NULL;
316 	INIT_LIST_HEAD(&timer->entry);
317 }
318 EXPORT_SYMBOL(init_virt_timer);
319 
320 static inline int vtimer_pending(struct vtimer_list *timer)
321 {
322 	return !list_empty(&timer->entry);
323 }
324 
325 static void internal_add_vtimer(struct vtimer_list *timer)
326 {
327 	if (list_empty(&virt_timer_list)) {
328 		/* First timer, just program it. */
329 		atomic64_set(&virt_timer_current, timer->expires);
330 		atomic64_set(&virt_timer_elapsed, 0);
331 		list_add(&timer->entry, &virt_timer_list);
332 	} else {
333 		/* Update timer against current base. */
334 		timer->expires += atomic64_read(&virt_timer_elapsed);
335 		if (likely((s64) timer->expires <
336 			   (s64) atomic64_read(&virt_timer_current)))
337 			/* The new timer expires before the current timer. */
338 			atomic64_set(&virt_timer_current, timer->expires);
339 		/* Insert new timer into the list. */
340 		list_add_sorted(timer, &virt_timer_list);
341 	}
342 }
343 
344 static void __add_vtimer(struct vtimer_list *timer, int periodic)
345 {
346 	unsigned long flags;
347 
348 	timer->interval = periodic ? timer->expires : 0;
349 	spin_lock_irqsave(&virt_timer_lock, flags);
350 	internal_add_vtimer(timer);
351 	spin_unlock_irqrestore(&virt_timer_lock, flags);
352 }
353 
354 /*
355  * add_virt_timer - add a oneshot virtual CPU timer
356  */
357 void add_virt_timer(struct vtimer_list *timer)
358 {
359 	__add_vtimer(timer, 0);
360 }
361 EXPORT_SYMBOL(add_virt_timer);
362 
363 /*
364  * add_virt_timer_int - add an interval virtual CPU timer
365  */
366 void add_virt_timer_periodic(struct vtimer_list *timer)
367 {
368 	__add_vtimer(timer, 1);
369 }
370 EXPORT_SYMBOL(add_virt_timer_periodic);
371 
372 static int __mod_vtimer(struct vtimer_list *timer, u64 expires, int periodic)
373 {
374 	unsigned long flags;
375 	int rc;
376 
377 	BUG_ON(!timer->function);
378 
379 	if (timer->expires == expires && vtimer_pending(timer))
380 		return 1;
381 	spin_lock_irqsave(&virt_timer_lock, flags);
382 	rc = vtimer_pending(timer);
383 	if (rc)
384 		list_del_init(&timer->entry);
385 	timer->interval = periodic ? expires : 0;
386 	timer->expires = expires;
387 	internal_add_vtimer(timer);
388 	spin_unlock_irqrestore(&virt_timer_lock, flags);
389 	return rc;
390 }
391 
392 /*
393  * returns whether it has modified a pending timer (1) or not (0)
394  */
395 int mod_virt_timer(struct vtimer_list *timer, u64 expires)
396 {
397 	return __mod_vtimer(timer, expires, 0);
398 }
399 EXPORT_SYMBOL(mod_virt_timer);
400 
401 /*
402  * returns whether it has modified a pending timer (1) or not (0)
403  */
404 int mod_virt_timer_periodic(struct vtimer_list *timer, u64 expires)
405 {
406 	return __mod_vtimer(timer, expires, 1);
407 }
408 EXPORT_SYMBOL(mod_virt_timer_periodic);
409 
410 /*
411  * Delete a virtual timer.
412  *
413  * returns whether the deleted timer was pending (1) or not (0)
414  */
415 int del_virt_timer(struct vtimer_list *timer)
416 {
417 	unsigned long flags;
418 
419 	if (!vtimer_pending(timer))
420 		return 0;
421 	spin_lock_irqsave(&virt_timer_lock, flags);
422 	list_del_init(&timer->entry);
423 	spin_unlock_irqrestore(&virt_timer_lock, flags);
424 	return 1;
425 }
426 EXPORT_SYMBOL(del_virt_timer);
427 
428 /*
429  * Start the virtual CPU timer on the current CPU.
430  */
431 void vtime_init(void)
432 {
433 	/* set initial cpu timer */
434 	set_vtimer(VTIMER_MAX_SLICE);
435 	/* Setup initial MT scaling values */
436 	if (smp_cpu_mtid) {
437 		__this_cpu_write(mt_scaling_jiffies, jiffies);
438 		__this_cpu_write(mt_scaling_mult, 1);
439 		__this_cpu_write(mt_scaling_div, 1);
440 		stcctm(MT_DIAG, smp_cpu_mtid + 1, this_cpu_ptr(mt_cycles));
441 	}
442 }
443