xref: /openbmc/linux/arch/x86/kvm/i8254.c (revision cd4d09ec)
1 /*
2  * 8253/8254 interval timer emulation
3  *
4  * Copyright (c) 2003-2004 Fabrice Bellard
5  * Copyright (c) 2006 Intel Corporation
6  * Copyright (c) 2007 Keir Fraser, XenSource Inc
7  * Copyright (c) 2008 Intel Corporation
8  * Copyright 2009 Red Hat, Inc. and/or its affiliates.
9  *
10  * Permission is hereby granted, free of charge, to any person obtaining a copy
11  * of this software and associated documentation files (the "Software"), to deal
12  * in the Software without restriction, including without limitation the rights
13  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
14  * copies of the Software, and to permit persons to whom the Software is
15  * furnished to do so, subject to the following conditions:
16  *
17  * The above copyright notice and this permission notice shall be included in
18  * all copies or substantial portions of the Software.
19  *
20  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
21  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
22  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
23  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
24  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
25  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
26  * THE SOFTWARE.
27  *
28  * Authors:
29  *   Sheng Yang <sheng.yang@intel.com>
30  *   Based on QEMU and Xen.
31  */
32 
33 #define pr_fmt(fmt) "pit: " fmt
34 
35 #include <linux/kvm_host.h>
36 #include <linux/slab.h>
37 
38 #include "ioapic.h"
39 #include "irq.h"
40 #include "i8254.h"
41 #include "x86.h"
42 
43 #ifndef CONFIG_X86_64
44 #define mod_64(x, y) ((x) - (y) * div64_u64(x, y))
45 #else
46 #define mod_64(x, y) ((x) % (y))
47 #endif
48 
49 #define RW_STATE_LSB 1
50 #define RW_STATE_MSB 2
51 #define RW_STATE_WORD0 3
52 #define RW_STATE_WORD1 4
53 
54 /* Compute with 96 bit intermediate result: (a*b)/c */
55 static u64 muldiv64(u64 a, u32 b, u32 c)
56 {
57 	union {
58 		u64 ll;
59 		struct {
60 			u32 low, high;
61 		} l;
62 	} u, res;
63 	u64 rl, rh;
64 
65 	u.ll = a;
66 	rl = (u64)u.l.low * (u64)b;
67 	rh = (u64)u.l.high * (u64)b;
68 	rh += (rl >> 32);
69 	res.l.high = div64_u64(rh, c);
70 	res.l.low = div64_u64(((mod_64(rh, c) << 32) + (rl & 0xffffffff)), c);
71 	return res.ll;
72 }
73 
74 static void pit_set_gate(struct kvm *kvm, int channel, u32 val)
75 {
76 	struct kvm_kpit_channel_state *c =
77 		&kvm->arch.vpit->pit_state.channels[channel];
78 
79 	WARN_ON(!mutex_is_locked(&kvm->arch.vpit->pit_state.lock));
80 
81 	switch (c->mode) {
82 	default:
83 	case 0:
84 	case 4:
85 		/* XXX: just disable/enable counting */
86 		break;
87 	case 1:
88 	case 2:
89 	case 3:
90 	case 5:
91 		/* Restart counting on rising edge. */
92 		if (c->gate < val)
93 			c->count_load_time = ktime_get();
94 		break;
95 	}
96 
97 	c->gate = val;
98 }
99 
100 static int pit_get_gate(struct kvm *kvm, int channel)
101 {
102 	WARN_ON(!mutex_is_locked(&kvm->arch.vpit->pit_state.lock));
103 
104 	return kvm->arch.vpit->pit_state.channels[channel].gate;
105 }
106 
107 static s64 __kpit_elapsed(struct kvm *kvm)
108 {
109 	s64 elapsed;
110 	ktime_t remaining;
111 	struct kvm_kpit_state *ps = &kvm->arch.vpit->pit_state;
112 
113 	if (!ps->period)
114 		return 0;
115 
116 	/*
117 	 * The Counter does not stop when it reaches zero. In
118 	 * Modes 0, 1, 4, and 5 the Counter ``wraps around'' to
119 	 * the highest count, either FFFF hex for binary counting
120 	 * or 9999 for BCD counting, and continues counting.
121 	 * Modes 2 and 3 are periodic; the Counter reloads
122 	 * itself with the initial count and continues counting
123 	 * from there.
124 	 */
125 	remaining = hrtimer_get_remaining(&ps->timer);
126 	elapsed = ps->period - ktime_to_ns(remaining);
127 
128 	return elapsed;
129 }
130 
131 static s64 kpit_elapsed(struct kvm *kvm, struct kvm_kpit_channel_state *c,
132 			int channel)
133 {
134 	if (channel == 0)
135 		return __kpit_elapsed(kvm);
136 
137 	return ktime_to_ns(ktime_sub(ktime_get(), c->count_load_time));
138 }
139 
140 static int pit_get_count(struct kvm *kvm, int channel)
141 {
142 	struct kvm_kpit_channel_state *c =
143 		&kvm->arch.vpit->pit_state.channels[channel];
144 	s64 d, t;
145 	int counter;
146 
147 	WARN_ON(!mutex_is_locked(&kvm->arch.vpit->pit_state.lock));
148 
149 	t = kpit_elapsed(kvm, c, channel);
150 	d = muldiv64(t, KVM_PIT_FREQ, NSEC_PER_SEC);
151 
152 	switch (c->mode) {
153 	case 0:
154 	case 1:
155 	case 4:
156 	case 5:
157 		counter = (c->count - d) & 0xffff;
158 		break;
159 	case 3:
160 		/* XXX: may be incorrect for odd counts */
161 		counter = c->count - (mod_64((2 * d), c->count));
162 		break;
163 	default:
164 		counter = c->count - mod_64(d, c->count);
165 		break;
166 	}
167 	return counter;
168 }
169 
170 static int pit_get_out(struct kvm *kvm, int channel)
171 {
172 	struct kvm_kpit_channel_state *c =
173 		&kvm->arch.vpit->pit_state.channels[channel];
174 	s64 d, t;
175 	int out;
176 
177 	WARN_ON(!mutex_is_locked(&kvm->arch.vpit->pit_state.lock));
178 
179 	t = kpit_elapsed(kvm, c, channel);
180 	d = muldiv64(t, KVM_PIT_FREQ, NSEC_PER_SEC);
181 
182 	switch (c->mode) {
183 	default:
184 	case 0:
185 		out = (d >= c->count);
186 		break;
187 	case 1:
188 		out = (d < c->count);
189 		break;
190 	case 2:
191 		out = ((mod_64(d, c->count) == 0) && (d != 0));
192 		break;
193 	case 3:
194 		out = (mod_64(d, c->count) < ((c->count + 1) >> 1));
195 		break;
196 	case 4:
197 	case 5:
198 		out = (d == c->count);
199 		break;
200 	}
201 
202 	return out;
203 }
204 
205 static void pit_latch_count(struct kvm *kvm, int channel)
206 {
207 	struct kvm_kpit_channel_state *c =
208 		&kvm->arch.vpit->pit_state.channels[channel];
209 
210 	WARN_ON(!mutex_is_locked(&kvm->arch.vpit->pit_state.lock));
211 
212 	if (!c->count_latched) {
213 		c->latched_count = pit_get_count(kvm, channel);
214 		c->count_latched = c->rw_mode;
215 	}
216 }
217 
218 static void pit_latch_status(struct kvm *kvm, int channel)
219 {
220 	struct kvm_kpit_channel_state *c =
221 		&kvm->arch.vpit->pit_state.channels[channel];
222 
223 	WARN_ON(!mutex_is_locked(&kvm->arch.vpit->pit_state.lock));
224 
225 	if (!c->status_latched) {
226 		/* TODO: Return NULL COUNT (bit 6). */
227 		c->status = ((pit_get_out(kvm, channel) << 7) |
228 				(c->rw_mode << 4) |
229 				(c->mode << 1) |
230 				c->bcd);
231 		c->status_latched = 1;
232 	}
233 }
234 
235 static void kvm_pit_ack_irq(struct kvm_irq_ack_notifier *kian)
236 {
237 	struct kvm_kpit_state *ps = container_of(kian, struct kvm_kpit_state,
238 						 irq_ack_notifier);
239 	int value;
240 
241 	spin_lock(&ps->inject_lock);
242 	value = atomic_dec_return(&ps->pending);
243 	if (value < 0)
244 		/* spurious acks can be generated if, for example, the
245 		 * PIC is being reset.  Handle it gracefully here
246 		 */
247 		atomic_inc(&ps->pending);
248 	else if (value > 0)
249 		/* in this case, we had multiple outstanding pit interrupts
250 		 * that we needed to inject.  Reinject
251 		 */
252 		queue_kthread_work(&ps->pit->worker, &ps->pit->expired);
253 	ps->irq_ack = 1;
254 	spin_unlock(&ps->inject_lock);
255 }
256 
257 void __kvm_migrate_pit_timer(struct kvm_vcpu *vcpu)
258 {
259 	struct kvm_pit *pit = vcpu->kvm->arch.vpit;
260 	struct hrtimer *timer;
261 
262 	if (!kvm_vcpu_is_bsp(vcpu) || !pit)
263 		return;
264 
265 	timer = &pit->pit_state.timer;
266 	mutex_lock(&pit->pit_state.lock);
267 	if (hrtimer_cancel(timer))
268 		hrtimer_start_expires(timer, HRTIMER_MODE_ABS);
269 	mutex_unlock(&pit->pit_state.lock);
270 }
271 
272 static void destroy_pit_timer(struct kvm_pit *pit)
273 {
274 	hrtimer_cancel(&pit->pit_state.timer);
275 	flush_kthread_work(&pit->expired);
276 }
277 
278 static void pit_do_work(struct kthread_work *work)
279 {
280 	struct kvm_pit *pit = container_of(work, struct kvm_pit, expired);
281 	struct kvm *kvm = pit->kvm;
282 	struct kvm_vcpu *vcpu;
283 	int i;
284 	struct kvm_kpit_state *ps = &pit->pit_state;
285 	int inject = 0;
286 
287 	/* Try to inject pending interrupts when
288 	 * last one has been acked.
289 	 */
290 	spin_lock(&ps->inject_lock);
291 	if (ps->irq_ack) {
292 		ps->irq_ack = 0;
293 		inject = 1;
294 	}
295 	spin_unlock(&ps->inject_lock);
296 	if (inject) {
297 		kvm_set_irq(kvm, kvm->arch.vpit->irq_source_id, 0, 1, false);
298 		kvm_set_irq(kvm, kvm->arch.vpit->irq_source_id, 0, 0, false);
299 
300 		/*
301 		 * Provides NMI watchdog support via Virtual Wire mode.
302 		 * The route is: PIT -> PIC -> LVT0 in NMI mode.
303 		 *
304 		 * Note: Our Virtual Wire implementation is simplified, only
305 		 * propagating PIT interrupts to all VCPUs when they have set
306 		 * LVT0 to NMI delivery. Other PIC interrupts are just sent to
307 		 * VCPU0, and only if its LVT0 is in EXTINT mode.
308 		 */
309 		if (atomic_read(&kvm->arch.vapics_in_nmi_mode) > 0)
310 			kvm_for_each_vcpu(i, vcpu, kvm)
311 				kvm_apic_nmi_wd_deliver(vcpu);
312 	}
313 }
314 
315 static enum hrtimer_restart pit_timer_fn(struct hrtimer *data)
316 {
317 	struct kvm_kpit_state *ps = container_of(data, struct kvm_kpit_state, timer);
318 	struct kvm_pit *pt = ps->kvm->arch.vpit;
319 
320 	if (ps->reinject || !atomic_read(&ps->pending)) {
321 		atomic_inc(&ps->pending);
322 		queue_kthread_work(&pt->worker, &pt->expired);
323 	}
324 
325 	if (ps->is_periodic) {
326 		hrtimer_add_expires_ns(&ps->timer, ps->period);
327 		return HRTIMER_RESTART;
328 	} else
329 		return HRTIMER_NORESTART;
330 }
331 
332 static void create_pit_timer(struct kvm *kvm, u32 val, int is_period)
333 {
334 	struct kvm_kpit_state *ps = &kvm->arch.vpit->pit_state;
335 	s64 interval;
336 
337 	if (!ioapic_in_kernel(kvm) ||
338 	    ps->flags & KVM_PIT_FLAGS_HPET_LEGACY)
339 		return;
340 
341 	interval = muldiv64(val, NSEC_PER_SEC, KVM_PIT_FREQ);
342 
343 	pr_debug("create pit timer, interval is %llu nsec\n", interval);
344 
345 	/* TODO The new value only affected after the retriggered */
346 	hrtimer_cancel(&ps->timer);
347 	flush_kthread_work(&ps->pit->expired);
348 	ps->period = interval;
349 	ps->is_periodic = is_period;
350 
351 	ps->timer.function = pit_timer_fn;
352 	ps->kvm = ps->pit->kvm;
353 
354 	atomic_set(&ps->pending, 0);
355 	ps->irq_ack = 1;
356 
357 	/*
358 	 * Do not allow the guest to program periodic timers with small
359 	 * interval, since the hrtimers are not throttled by the host
360 	 * scheduler.
361 	 */
362 	if (ps->is_periodic) {
363 		s64 min_period = min_timer_period_us * 1000LL;
364 
365 		if (ps->period < min_period) {
366 			pr_info_ratelimited(
367 			    "kvm: requested %lld ns "
368 			    "i8254 timer period limited to %lld ns\n",
369 			    ps->period, min_period);
370 			ps->period = min_period;
371 		}
372 	}
373 
374 	hrtimer_start(&ps->timer, ktime_add_ns(ktime_get(), interval),
375 		      HRTIMER_MODE_ABS);
376 }
377 
378 static void pit_load_count(struct kvm *kvm, int channel, u32 val)
379 {
380 	struct kvm_kpit_state *ps = &kvm->arch.vpit->pit_state;
381 
382 	WARN_ON(!mutex_is_locked(&ps->lock));
383 
384 	pr_debug("load_count val is %d, channel is %d\n", val, channel);
385 
386 	/*
387 	 * The largest possible initial count is 0; this is equivalent
388 	 * to 216 for binary counting and 104 for BCD counting.
389 	 */
390 	if (val == 0)
391 		val = 0x10000;
392 
393 	ps->channels[channel].count = val;
394 
395 	if (channel != 0) {
396 		ps->channels[channel].count_load_time = ktime_get();
397 		return;
398 	}
399 
400 	/* Two types of timer
401 	 * mode 1 is one shot, mode 2 is period, otherwise del timer */
402 	switch (ps->channels[0].mode) {
403 	case 0:
404 	case 1:
405         /* FIXME: enhance mode 4 precision */
406 	case 4:
407 		create_pit_timer(kvm, val, 0);
408 		break;
409 	case 2:
410 	case 3:
411 		create_pit_timer(kvm, val, 1);
412 		break;
413 	default:
414 		destroy_pit_timer(kvm->arch.vpit);
415 	}
416 }
417 
418 void kvm_pit_load_count(struct kvm *kvm, int channel, u32 val, int hpet_legacy_start)
419 {
420 	u8 saved_mode;
421 	if (hpet_legacy_start) {
422 		/* save existing mode for later reenablement */
423 		WARN_ON(channel != 0);
424 		saved_mode = kvm->arch.vpit->pit_state.channels[0].mode;
425 		kvm->arch.vpit->pit_state.channels[0].mode = 0xff; /* disable timer */
426 		pit_load_count(kvm, channel, val);
427 		kvm->arch.vpit->pit_state.channels[0].mode = saved_mode;
428 	} else {
429 		pit_load_count(kvm, channel, val);
430 	}
431 }
432 
433 static inline struct kvm_pit *dev_to_pit(struct kvm_io_device *dev)
434 {
435 	return container_of(dev, struct kvm_pit, dev);
436 }
437 
438 static inline struct kvm_pit *speaker_to_pit(struct kvm_io_device *dev)
439 {
440 	return container_of(dev, struct kvm_pit, speaker_dev);
441 }
442 
443 static inline int pit_in_range(gpa_t addr)
444 {
445 	return ((addr >= KVM_PIT_BASE_ADDRESS) &&
446 		(addr < KVM_PIT_BASE_ADDRESS + KVM_PIT_MEM_LENGTH));
447 }
448 
449 static int pit_ioport_write(struct kvm_vcpu *vcpu,
450 				struct kvm_io_device *this,
451 			    gpa_t addr, int len, const void *data)
452 {
453 	struct kvm_pit *pit = dev_to_pit(this);
454 	struct kvm_kpit_state *pit_state = &pit->pit_state;
455 	struct kvm *kvm = pit->kvm;
456 	int channel, access;
457 	struct kvm_kpit_channel_state *s;
458 	u32 val = *(u32 *) data;
459 	if (!pit_in_range(addr))
460 		return -EOPNOTSUPP;
461 
462 	val  &= 0xff;
463 	addr &= KVM_PIT_CHANNEL_MASK;
464 
465 	mutex_lock(&pit_state->lock);
466 
467 	if (val != 0)
468 		pr_debug("write addr is 0x%x, len is %d, val is 0x%x\n",
469 			 (unsigned int)addr, len, val);
470 
471 	if (addr == 3) {
472 		channel = val >> 6;
473 		if (channel == 3) {
474 			/* Read-Back Command. */
475 			for (channel = 0; channel < 3; channel++) {
476 				s = &pit_state->channels[channel];
477 				if (val & (2 << channel)) {
478 					if (!(val & 0x20))
479 						pit_latch_count(kvm, channel);
480 					if (!(val & 0x10))
481 						pit_latch_status(kvm, channel);
482 				}
483 			}
484 		} else {
485 			/* Select Counter <channel>. */
486 			s = &pit_state->channels[channel];
487 			access = (val >> 4) & KVM_PIT_CHANNEL_MASK;
488 			if (access == 0) {
489 				pit_latch_count(kvm, channel);
490 			} else {
491 				s->rw_mode = access;
492 				s->read_state = access;
493 				s->write_state = access;
494 				s->mode = (val >> 1) & 7;
495 				if (s->mode > 5)
496 					s->mode -= 4;
497 				s->bcd = val & 1;
498 			}
499 		}
500 	} else {
501 		/* Write Count. */
502 		s = &pit_state->channels[addr];
503 		switch (s->write_state) {
504 		default:
505 		case RW_STATE_LSB:
506 			pit_load_count(kvm, addr, val);
507 			break;
508 		case RW_STATE_MSB:
509 			pit_load_count(kvm, addr, val << 8);
510 			break;
511 		case RW_STATE_WORD0:
512 			s->write_latch = val;
513 			s->write_state = RW_STATE_WORD1;
514 			break;
515 		case RW_STATE_WORD1:
516 			pit_load_count(kvm, addr, s->write_latch | (val << 8));
517 			s->write_state = RW_STATE_WORD0;
518 			break;
519 		}
520 	}
521 
522 	mutex_unlock(&pit_state->lock);
523 	return 0;
524 }
525 
526 static int pit_ioport_read(struct kvm_vcpu *vcpu,
527 			   struct kvm_io_device *this,
528 			   gpa_t addr, int len, void *data)
529 {
530 	struct kvm_pit *pit = dev_to_pit(this);
531 	struct kvm_kpit_state *pit_state = &pit->pit_state;
532 	struct kvm *kvm = pit->kvm;
533 	int ret, count;
534 	struct kvm_kpit_channel_state *s;
535 	if (!pit_in_range(addr))
536 		return -EOPNOTSUPP;
537 
538 	addr &= KVM_PIT_CHANNEL_MASK;
539 	if (addr == 3)
540 		return 0;
541 
542 	s = &pit_state->channels[addr];
543 
544 	mutex_lock(&pit_state->lock);
545 
546 	if (s->status_latched) {
547 		s->status_latched = 0;
548 		ret = s->status;
549 	} else if (s->count_latched) {
550 		switch (s->count_latched) {
551 		default:
552 		case RW_STATE_LSB:
553 			ret = s->latched_count & 0xff;
554 			s->count_latched = 0;
555 			break;
556 		case RW_STATE_MSB:
557 			ret = s->latched_count >> 8;
558 			s->count_latched = 0;
559 			break;
560 		case RW_STATE_WORD0:
561 			ret = s->latched_count & 0xff;
562 			s->count_latched = RW_STATE_MSB;
563 			break;
564 		}
565 	} else {
566 		switch (s->read_state) {
567 		default:
568 		case RW_STATE_LSB:
569 			count = pit_get_count(kvm, addr);
570 			ret = count & 0xff;
571 			break;
572 		case RW_STATE_MSB:
573 			count = pit_get_count(kvm, addr);
574 			ret = (count >> 8) & 0xff;
575 			break;
576 		case RW_STATE_WORD0:
577 			count = pit_get_count(kvm, addr);
578 			ret = count & 0xff;
579 			s->read_state = RW_STATE_WORD1;
580 			break;
581 		case RW_STATE_WORD1:
582 			count = pit_get_count(kvm, addr);
583 			ret = (count >> 8) & 0xff;
584 			s->read_state = RW_STATE_WORD0;
585 			break;
586 		}
587 	}
588 
589 	if (len > sizeof(ret))
590 		len = sizeof(ret);
591 	memcpy(data, (char *)&ret, len);
592 
593 	mutex_unlock(&pit_state->lock);
594 	return 0;
595 }
596 
597 static int speaker_ioport_write(struct kvm_vcpu *vcpu,
598 				struct kvm_io_device *this,
599 				gpa_t addr, int len, const void *data)
600 {
601 	struct kvm_pit *pit = speaker_to_pit(this);
602 	struct kvm_kpit_state *pit_state = &pit->pit_state;
603 	struct kvm *kvm = pit->kvm;
604 	u32 val = *(u32 *) data;
605 	if (addr != KVM_SPEAKER_BASE_ADDRESS)
606 		return -EOPNOTSUPP;
607 
608 	mutex_lock(&pit_state->lock);
609 	pit_state->speaker_data_on = (val >> 1) & 1;
610 	pit_set_gate(kvm, 2, val & 1);
611 	mutex_unlock(&pit_state->lock);
612 	return 0;
613 }
614 
615 static int speaker_ioport_read(struct kvm_vcpu *vcpu,
616 				   struct kvm_io_device *this,
617 				   gpa_t addr, int len, void *data)
618 {
619 	struct kvm_pit *pit = speaker_to_pit(this);
620 	struct kvm_kpit_state *pit_state = &pit->pit_state;
621 	struct kvm *kvm = pit->kvm;
622 	unsigned int refresh_clock;
623 	int ret;
624 	if (addr != KVM_SPEAKER_BASE_ADDRESS)
625 		return -EOPNOTSUPP;
626 
627 	/* Refresh clock toggles at about 15us. We approximate as 2^14ns. */
628 	refresh_clock = ((unsigned int)ktime_to_ns(ktime_get()) >> 14) & 1;
629 
630 	mutex_lock(&pit_state->lock);
631 	ret = ((pit_state->speaker_data_on << 1) | pit_get_gate(kvm, 2) |
632 		(pit_get_out(kvm, 2) << 5) | (refresh_clock << 4));
633 	if (len > sizeof(ret))
634 		len = sizeof(ret);
635 	memcpy(data, (char *)&ret, len);
636 	mutex_unlock(&pit_state->lock);
637 	return 0;
638 }
639 
640 void kvm_pit_reset(struct kvm_pit *pit)
641 {
642 	int i;
643 	struct kvm_kpit_channel_state *c;
644 
645 	mutex_lock(&pit->pit_state.lock);
646 	pit->pit_state.flags = 0;
647 	for (i = 0; i < 3; i++) {
648 		c = &pit->pit_state.channels[i];
649 		c->mode = 0xff;
650 		c->gate = (i != 2);
651 		pit_load_count(pit->kvm, i, 0);
652 	}
653 	mutex_unlock(&pit->pit_state.lock);
654 
655 	atomic_set(&pit->pit_state.pending, 0);
656 	pit->pit_state.irq_ack = 1;
657 }
658 
659 static void pit_mask_notifer(struct kvm_irq_mask_notifier *kimn, bool mask)
660 {
661 	struct kvm_pit *pit = container_of(kimn, struct kvm_pit, mask_notifier);
662 
663 	if (!mask) {
664 		atomic_set(&pit->pit_state.pending, 0);
665 		pit->pit_state.irq_ack = 1;
666 	}
667 }
668 
669 static const struct kvm_io_device_ops pit_dev_ops = {
670 	.read     = pit_ioport_read,
671 	.write    = pit_ioport_write,
672 };
673 
674 static const struct kvm_io_device_ops speaker_dev_ops = {
675 	.read     = speaker_ioport_read,
676 	.write    = speaker_ioport_write,
677 };
678 
679 /* Caller must hold slots_lock */
680 struct kvm_pit *kvm_create_pit(struct kvm *kvm, u32 flags)
681 {
682 	struct kvm_pit *pit;
683 	struct kvm_kpit_state *pit_state;
684 	struct pid *pid;
685 	pid_t pid_nr;
686 	int ret;
687 
688 	pit = kzalloc(sizeof(struct kvm_pit), GFP_KERNEL);
689 	if (!pit)
690 		return NULL;
691 
692 	pit->irq_source_id = kvm_request_irq_source_id(kvm);
693 	if (pit->irq_source_id < 0) {
694 		kfree(pit);
695 		return NULL;
696 	}
697 
698 	mutex_init(&pit->pit_state.lock);
699 	mutex_lock(&pit->pit_state.lock);
700 	spin_lock_init(&pit->pit_state.inject_lock);
701 
702 	pid = get_pid(task_tgid(current));
703 	pid_nr = pid_vnr(pid);
704 	put_pid(pid);
705 
706 	init_kthread_worker(&pit->worker);
707 	pit->worker_task = kthread_run(kthread_worker_fn, &pit->worker,
708 				       "kvm-pit/%d", pid_nr);
709 	if (IS_ERR(pit->worker_task)) {
710 		mutex_unlock(&pit->pit_state.lock);
711 		kvm_free_irq_source_id(kvm, pit->irq_source_id);
712 		kfree(pit);
713 		return NULL;
714 	}
715 	init_kthread_work(&pit->expired, pit_do_work);
716 
717 	kvm->arch.vpit = pit;
718 	pit->kvm = kvm;
719 
720 	pit_state = &pit->pit_state;
721 	pit_state->pit = pit;
722 	hrtimer_init(&pit_state->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
723 	pit_state->irq_ack_notifier.gsi = 0;
724 	pit_state->irq_ack_notifier.irq_acked = kvm_pit_ack_irq;
725 	kvm_register_irq_ack_notifier(kvm, &pit_state->irq_ack_notifier);
726 	pit_state->reinject = true;
727 	mutex_unlock(&pit->pit_state.lock);
728 
729 	kvm_pit_reset(pit);
730 
731 	pit->mask_notifier.func = pit_mask_notifer;
732 	kvm_register_irq_mask_notifier(kvm, 0, &pit->mask_notifier);
733 
734 	kvm_iodevice_init(&pit->dev, &pit_dev_ops);
735 	ret = kvm_io_bus_register_dev(kvm, KVM_PIO_BUS, KVM_PIT_BASE_ADDRESS,
736 				      KVM_PIT_MEM_LENGTH, &pit->dev);
737 	if (ret < 0)
738 		goto fail;
739 
740 	if (flags & KVM_PIT_SPEAKER_DUMMY) {
741 		kvm_iodevice_init(&pit->speaker_dev, &speaker_dev_ops);
742 		ret = kvm_io_bus_register_dev(kvm, KVM_PIO_BUS,
743 					      KVM_SPEAKER_BASE_ADDRESS, 4,
744 					      &pit->speaker_dev);
745 		if (ret < 0)
746 			goto fail_unregister;
747 	}
748 
749 	return pit;
750 
751 fail_unregister:
752 	kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS, &pit->dev);
753 
754 fail:
755 	kvm_unregister_irq_mask_notifier(kvm, 0, &pit->mask_notifier);
756 	kvm_unregister_irq_ack_notifier(kvm, &pit_state->irq_ack_notifier);
757 	kvm_free_irq_source_id(kvm, pit->irq_source_id);
758 	kthread_stop(pit->worker_task);
759 	kfree(pit);
760 	return NULL;
761 }
762 
763 void kvm_free_pit(struct kvm *kvm)
764 {
765 	struct hrtimer *timer;
766 
767 	if (kvm->arch.vpit) {
768 		kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS, &kvm->arch.vpit->dev);
769 		kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
770 					      &kvm->arch.vpit->speaker_dev);
771 		kvm_unregister_irq_mask_notifier(kvm, 0,
772 					       &kvm->arch.vpit->mask_notifier);
773 		kvm_unregister_irq_ack_notifier(kvm,
774 				&kvm->arch.vpit->pit_state.irq_ack_notifier);
775 		mutex_lock(&kvm->arch.vpit->pit_state.lock);
776 		timer = &kvm->arch.vpit->pit_state.timer;
777 		hrtimer_cancel(timer);
778 		flush_kthread_work(&kvm->arch.vpit->expired);
779 		kthread_stop(kvm->arch.vpit->worker_task);
780 		kvm_free_irq_source_id(kvm, kvm->arch.vpit->irq_source_id);
781 		mutex_unlock(&kvm->arch.vpit->pit_state.lock);
782 		kfree(kvm->arch.vpit);
783 	}
784 }
785