xref: /openbmc/linux/arch/s390/kernel/vtime.c (revision d0e22329)
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" : "=m" (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 		: "=m" (timer) : "m" (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 	stcctm5(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, steal;
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 		: "=m" (S390_lowcore.last_update_timer),
139 		  "=m" (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 	steal = S390_lowcore.steal_timer;
186 	if ((s64) steal > 0) {
187 		S390_lowcore.steal_timer = 0;
188 		account_steal_time(cputime_to_nsecs(steal));
189 	}
190 
191 	return virt_timer_forward(user + guest + system + hardirq + softirq);
192 }
193 
194 void vtime_task_switch(struct task_struct *prev)
195 {
196 	do_account_vtime(prev);
197 	prev->thread.user_timer = S390_lowcore.user_timer;
198 	prev->thread.guest_timer = S390_lowcore.guest_timer;
199 	prev->thread.system_timer = S390_lowcore.system_timer;
200 	prev->thread.hardirq_timer = S390_lowcore.hardirq_timer;
201 	prev->thread.softirq_timer = S390_lowcore.softirq_timer;
202 	S390_lowcore.user_timer = current->thread.user_timer;
203 	S390_lowcore.guest_timer = current->thread.guest_timer;
204 	S390_lowcore.system_timer = current->thread.system_timer;
205 	S390_lowcore.hardirq_timer = current->thread.hardirq_timer;
206 	S390_lowcore.softirq_timer = current->thread.softirq_timer;
207 }
208 
209 /*
210  * In s390, accounting pending user time also implies
211  * accounting system time in order to correctly compute
212  * the stolen time accounting.
213  */
214 void vtime_flush(struct task_struct *tsk)
215 {
216 	if (do_account_vtime(tsk))
217 		virt_timer_expire();
218 }
219 
220 /*
221  * Update process times based on virtual cpu times stored by entry.S
222  * to the lowcore fields user_timer, system_timer & steal_clock.
223  */
224 void vtime_account_irq_enter(struct task_struct *tsk)
225 {
226 	u64 timer;
227 
228 	timer = S390_lowcore.last_update_timer;
229 	S390_lowcore.last_update_timer = get_vtimer();
230 	timer -= S390_lowcore.last_update_timer;
231 
232 	if ((tsk->flags & PF_VCPU) && (irq_count() == 0))
233 		S390_lowcore.guest_timer += timer;
234 	else if (hardirq_count())
235 		S390_lowcore.hardirq_timer += timer;
236 	else if (in_serving_softirq())
237 		S390_lowcore.softirq_timer += timer;
238 	else
239 		S390_lowcore.system_timer += timer;
240 
241 	virt_timer_forward(timer);
242 }
243 EXPORT_SYMBOL_GPL(vtime_account_irq_enter);
244 
245 void vtime_account_system(struct task_struct *tsk)
246 __attribute__((alias("vtime_account_irq_enter")));
247 EXPORT_SYMBOL_GPL(vtime_account_system);
248 
249 /*
250  * Sorted add to a list. List is linear searched until first bigger
251  * element is found.
252  */
253 static void list_add_sorted(struct vtimer_list *timer, struct list_head *head)
254 {
255 	struct vtimer_list *tmp;
256 
257 	list_for_each_entry(tmp, head, entry) {
258 		if (tmp->expires > timer->expires) {
259 			list_add_tail(&timer->entry, &tmp->entry);
260 			return;
261 		}
262 	}
263 	list_add_tail(&timer->entry, head);
264 }
265 
266 /*
267  * Handler for expired virtual CPU timer.
268  */
269 static void virt_timer_expire(void)
270 {
271 	struct vtimer_list *timer, *tmp;
272 	unsigned long elapsed;
273 	LIST_HEAD(cb_list);
274 
275 	/* walk timer list, fire all expired timers */
276 	spin_lock(&virt_timer_lock);
277 	elapsed = atomic64_read(&virt_timer_elapsed);
278 	list_for_each_entry_safe(timer, tmp, &virt_timer_list, entry) {
279 		if (timer->expires < elapsed)
280 			/* move expired timer to the callback queue */
281 			list_move_tail(&timer->entry, &cb_list);
282 		else
283 			timer->expires -= elapsed;
284 	}
285 	if (!list_empty(&virt_timer_list)) {
286 		timer = list_first_entry(&virt_timer_list,
287 					 struct vtimer_list, entry);
288 		atomic64_set(&virt_timer_current, timer->expires);
289 	}
290 	atomic64_sub(elapsed, &virt_timer_elapsed);
291 	spin_unlock(&virt_timer_lock);
292 
293 	/* Do callbacks and recharge periodic timers */
294 	list_for_each_entry_safe(timer, tmp, &cb_list, entry) {
295 		list_del_init(&timer->entry);
296 		timer->function(timer->data);
297 		if (timer->interval) {
298 			/* Recharge interval timer */
299 			timer->expires = timer->interval +
300 				atomic64_read(&virt_timer_elapsed);
301 			spin_lock(&virt_timer_lock);
302 			list_add_sorted(timer, &virt_timer_list);
303 			spin_unlock(&virt_timer_lock);
304 		}
305 	}
306 }
307 
308 void init_virt_timer(struct vtimer_list *timer)
309 {
310 	timer->function = NULL;
311 	INIT_LIST_HEAD(&timer->entry);
312 }
313 EXPORT_SYMBOL(init_virt_timer);
314 
315 static inline int vtimer_pending(struct vtimer_list *timer)
316 {
317 	return !list_empty(&timer->entry);
318 }
319 
320 static void internal_add_vtimer(struct vtimer_list *timer)
321 {
322 	if (list_empty(&virt_timer_list)) {
323 		/* First timer, just program it. */
324 		atomic64_set(&virt_timer_current, timer->expires);
325 		atomic64_set(&virt_timer_elapsed, 0);
326 		list_add(&timer->entry, &virt_timer_list);
327 	} else {
328 		/* Update timer against current base. */
329 		timer->expires += atomic64_read(&virt_timer_elapsed);
330 		if (likely((s64) timer->expires <
331 			   (s64) atomic64_read(&virt_timer_current)))
332 			/* The new timer expires before the current timer. */
333 			atomic64_set(&virt_timer_current, timer->expires);
334 		/* Insert new timer into the list. */
335 		list_add_sorted(timer, &virt_timer_list);
336 	}
337 }
338 
339 static void __add_vtimer(struct vtimer_list *timer, int periodic)
340 {
341 	unsigned long flags;
342 
343 	timer->interval = periodic ? timer->expires : 0;
344 	spin_lock_irqsave(&virt_timer_lock, flags);
345 	internal_add_vtimer(timer);
346 	spin_unlock_irqrestore(&virt_timer_lock, flags);
347 }
348 
349 /*
350  * add_virt_timer - add a oneshot virtual CPU timer
351  */
352 void add_virt_timer(struct vtimer_list *timer)
353 {
354 	__add_vtimer(timer, 0);
355 }
356 EXPORT_SYMBOL(add_virt_timer);
357 
358 /*
359  * add_virt_timer_int - add an interval virtual CPU timer
360  */
361 void add_virt_timer_periodic(struct vtimer_list *timer)
362 {
363 	__add_vtimer(timer, 1);
364 }
365 EXPORT_SYMBOL(add_virt_timer_periodic);
366 
367 static int __mod_vtimer(struct vtimer_list *timer, u64 expires, int periodic)
368 {
369 	unsigned long flags;
370 	int rc;
371 
372 	BUG_ON(!timer->function);
373 
374 	if (timer->expires == expires && vtimer_pending(timer))
375 		return 1;
376 	spin_lock_irqsave(&virt_timer_lock, flags);
377 	rc = vtimer_pending(timer);
378 	if (rc)
379 		list_del_init(&timer->entry);
380 	timer->interval = periodic ? expires : 0;
381 	timer->expires = expires;
382 	internal_add_vtimer(timer);
383 	spin_unlock_irqrestore(&virt_timer_lock, flags);
384 	return rc;
385 }
386 
387 /*
388  * returns whether it has modified a pending timer (1) or not (0)
389  */
390 int mod_virt_timer(struct vtimer_list *timer, u64 expires)
391 {
392 	return __mod_vtimer(timer, expires, 0);
393 }
394 EXPORT_SYMBOL(mod_virt_timer);
395 
396 /*
397  * returns whether it has modified a pending timer (1) or not (0)
398  */
399 int mod_virt_timer_periodic(struct vtimer_list *timer, u64 expires)
400 {
401 	return __mod_vtimer(timer, expires, 1);
402 }
403 EXPORT_SYMBOL(mod_virt_timer_periodic);
404 
405 /*
406  * Delete a virtual timer.
407  *
408  * returns whether the deleted timer was pending (1) or not (0)
409  */
410 int del_virt_timer(struct vtimer_list *timer)
411 {
412 	unsigned long flags;
413 
414 	if (!vtimer_pending(timer))
415 		return 0;
416 	spin_lock_irqsave(&virt_timer_lock, flags);
417 	list_del_init(&timer->entry);
418 	spin_unlock_irqrestore(&virt_timer_lock, flags);
419 	return 1;
420 }
421 EXPORT_SYMBOL(del_virt_timer);
422 
423 /*
424  * Start the virtual CPU timer on the current CPU.
425  */
426 void vtime_init(void)
427 {
428 	/* set initial cpu timer */
429 	set_vtimer(VTIMER_MAX_SLICE);
430 	/* Setup initial MT scaling values */
431 	if (smp_cpu_mtid) {
432 		__this_cpu_write(mt_scaling_jiffies, jiffies);
433 		__this_cpu_write(mt_scaling_mult, 1);
434 		__this_cpu_write(mt_scaling_div, 1);
435 		stcctm5(smp_cpu_mtid + 1, this_cpu_ptr(mt_cycles));
436 	}
437 }
438