xref: /openbmc/linux/arch/x86/kvm/i8254.c (revision f35e839a)
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 "irq.h"
39 #include "i8254.h"
40 
41 #ifndef CONFIG_X86_64
42 #define mod_64(x, y) ((x) - (y) * div64_u64(x, y))
43 #else
44 #define mod_64(x, y) ((x) % (y))
45 #endif
46 
47 #define RW_STATE_LSB 1
48 #define RW_STATE_MSB 2
49 #define RW_STATE_WORD0 3
50 #define RW_STATE_WORD1 4
51 
52 /* Compute with 96 bit intermediate result: (a*b)/c */
53 static u64 muldiv64(u64 a, u32 b, u32 c)
54 {
55 	union {
56 		u64 ll;
57 		struct {
58 			u32 low, high;
59 		} l;
60 	} u, res;
61 	u64 rl, rh;
62 
63 	u.ll = a;
64 	rl = (u64)u.l.low * (u64)b;
65 	rh = (u64)u.l.high * (u64)b;
66 	rh += (rl >> 32);
67 	res.l.high = div64_u64(rh, c);
68 	res.l.low = div64_u64(((mod_64(rh, c) << 32) + (rl & 0xffffffff)), c);
69 	return res.ll;
70 }
71 
72 static void pit_set_gate(struct kvm *kvm, int channel, u32 val)
73 {
74 	struct kvm_kpit_channel_state *c =
75 		&kvm->arch.vpit->pit_state.channels[channel];
76 
77 	WARN_ON(!mutex_is_locked(&kvm->arch.vpit->pit_state.lock));
78 
79 	switch (c->mode) {
80 	default:
81 	case 0:
82 	case 4:
83 		/* XXX: just disable/enable counting */
84 		break;
85 	case 1:
86 	case 2:
87 	case 3:
88 	case 5:
89 		/* Restart counting on rising edge. */
90 		if (c->gate < val)
91 			c->count_load_time = ktime_get();
92 		break;
93 	}
94 
95 	c->gate = val;
96 }
97 
98 static int pit_get_gate(struct kvm *kvm, int channel)
99 {
100 	WARN_ON(!mutex_is_locked(&kvm->arch.vpit->pit_state.lock));
101 
102 	return kvm->arch.vpit->pit_state.channels[channel].gate;
103 }
104 
105 static s64 __kpit_elapsed(struct kvm *kvm)
106 {
107 	s64 elapsed;
108 	ktime_t remaining;
109 	struct kvm_kpit_state *ps = &kvm->arch.vpit->pit_state;
110 
111 	if (!ps->period)
112 		return 0;
113 
114 	/*
115 	 * The Counter does not stop when it reaches zero. In
116 	 * Modes 0, 1, 4, and 5 the Counter ``wraps around'' to
117 	 * the highest count, either FFFF hex for binary counting
118 	 * or 9999 for BCD counting, and continues counting.
119 	 * Modes 2 and 3 are periodic; the Counter reloads
120 	 * itself with the initial count and continues counting
121 	 * from there.
122 	 */
123 	remaining = hrtimer_get_remaining(&ps->timer);
124 	elapsed = ps->period - ktime_to_ns(remaining);
125 
126 	return elapsed;
127 }
128 
129 static s64 kpit_elapsed(struct kvm *kvm, struct kvm_kpit_channel_state *c,
130 			int channel)
131 {
132 	if (channel == 0)
133 		return __kpit_elapsed(kvm);
134 
135 	return ktime_to_ns(ktime_sub(ktime_get(), c->count_load_time));
136 }
137 
138 static int pit_get_count(struct kvm *kvm, int channel)
139 {
140 	struct kvm_kpit_channel_state *c =
141 		&kvm->arch.vpit->pit_state.channels[channel];
142 	s64 d, t;
143 	int counter;
144 
145 	WARN_ON(!mutex_is_locked(&kvm->arch.vpit->pit_state.lock));
146 
147 	t = kpit_elapsed(kvm, c, channel);
148 	d = muldiv64(t, KVM_PIT_FREQ, NSEC_PER_SEC);
149 
150 	switch (c->mode) {
151 	case 0:
152 	case 1:
153 	case 4:
154 	case 5:
155 		counter = (c->count - d) & 0xffff;
156 		break;
157 	case 3:
158 		/* XXX: may be incorrect for odd counts */
159 		counter = c->count - (mod_64((2 * d), c->count));
160 		break;
161 	default:
162 		counter = c->count - mod_64(d, c->count);
163 		break;
164 	}
165 	return counter;
166 }
167 
168 static int pit_get_out(struct kvm *kvm, int channel)
169 {
170 	struct kvm_kpit_channel_state *c =
171 		&kvm->arch.vpit->pit_state.channels[channel];
172 	s64 d, t;
173 	int out;
174 
175 	WARN_ON(!mutex_is_locked(&kvm->arch.vpit->pit_state.lock));
176 
177 	t = kpit_elapsed(kvm, c, channel);
178 	d = muldiv64(t, KVM_PIT_FREQ, NSEC_PER_SEC);
179 
180 	switch (c->mode) {
181 	default:
182 	case 0:
183 		out = (d >= c->count);
184 		break;
185 	case 1:
186 		out = (d < c->count);
187 		break;
188 	case 2:
189 		out = ((mod_64(d, c->count) == 0) && (d != 0));
190 		break;
191 	case 3:
192 		out = (mod_64(d, c->count) < ((c->count + 1) >> 1));
193 		break;
194 	case 4:
195 	case 5:
196 		out = (d == c->count);
197 		break;
198 	}
199 
200 	return out;
201 }
202 
203 static void pit_latch_count(struct kvm *kvm, int channel)
204 {
205 	struct kvm_kpit_channel_state *c =
206 		&kvm->arch.vpit->pit_state.channels[channel];
207 
208 	WARN_ON(!mutex_is_locked(&kvm->arch.vpit->pit_state.lock));
209 
210 	if (!c->count_latched) {
211 		c->latched_count = pit_get_count(kvm, channel);
212 		c->count_latched = c->rw_mode;
213 	}
214 }
215 
216 static void pit_latch_status(struct kvm *kvm, int channel)
217 {
218 	struct kvm_kpit_channel_state *c =
219 		&kvm->arch.vpit->pit_state.channels[channel];
220 
221 	WARN_ON(!mutex_is_locked(&kvm->arch.vpit->pit_state.lock));
222 
223 	if (!c->status_latched) {
224 		/* TODO: Return NULL COUNT (bit 6). */
225 		c->status = ((pit_get_out(kvm, channel) << 7) |
226 				(c->rw_mode << 4) |
227 				(c->mode << 1) |
228 				c->bcd);
229 		c->status_latched = 1;
230 	}
231 }
232 
233 static void kvm_pit_ack_irq(struct kvm_irq_ack_notifier *kian)
234 {
235 	struct kvm_kpit_state *ps = container_of(kian, struct kvm_kpit_state,
236 						 irq_ack_notifier);
237 	int value;
238 
239 	spin_lock(&ps->inject_lock);
240 	value = atomic_dec_return(&ps->pending);
241 	if (value < 0)
242 		/* spurious acks can be generated if, for example, the
243 		 * PIC is being reset.  Handle it gracefully here
244 		 */
245 		atomic_inc(&ps->pending);
246 	else if (value > 0)
247 		/* in this case, we had multiple outstanding pit interrupts
248 		 * that we needed to inject.  Reinject
249 		 */
250 		queue_kthread_work(&ps->pit->worker, &ps->pit->expired);
251 	ps->irq_ack = 1;
252 	spin_unlock(&ps->inject_lock);
253 }
254 
255 void __kvm_migrate_pit_timer(struct kvm_vcpu *vcpu)
256 {
257 	struct kvm_pit *pit = vcpu->kvm->arch.vpit;
258 	struct hrtimer *timer;
259 
260 	if (!kvm_vcpu_is_bsp(vcpu) || !pit)
261 		return;
262 
263 	timer = &pit->pit_state.timer;
264 	if (hrtimer_cancel(timer))
265 		hrtimer_start_expires(timer, HRTIMER_MODE_ABS);
266 }
267 
268 static void destroy_pit_timer(struct kvm_pit *pit)
269 {
270 	hrtimer_cancel(&pit->pit_state.timer);
271 	flush_kthread_work(&pit->expired);
272 }
273 
274 static void pit_do_work(struct kthread_work *work)
275 {
276 	struct kvm_pit *pit = container_of(work, struct kvm_pit, expired);
277 	struct kvm *kvm = pit->kvm;
278 	struct kvm_vcpu *vcpu;
279 	int i;
280 	struct kvm_kpit_state *ps = &pit->pit_state;
281 	int inject = 0;
282 
283 	/* Try to inject pending interrupts when
284 	 * last one has been acked.
285 	 */
286 	spin_lock(&ps->inject_lock);
287 	if (ps->irq_ack) {
288 		ps->irq_ack = 0;
289 		inject = 1;
290 	}
291 	spin_unlock(&ps->inject_lock);
292 	if (inject) {
293 		kvm_set_irq(kvm, kvm->arch.vpit->irq_source_id, 0, 1, false);
294 		kvm_set_irq(kvm, kvm->arch.vpit->irq_source_id, 0, 0, false);
295 
296 		/*
297 		 * Provides NMI watchdog support via Virtual Wire mode.
298 		 * The route is: PIT -> PIC -> LVT0 in NMI mode.
299 		 *
300 		 * Note: Our Virtual Wire implementation is simplified, only
301 		 * propagating PIT interrupts to all VCPUs when they have set
302 		 * LVT0 to NMI delivery. Other PIC interrupts are just sent to
303 		 * VCPU0, and only if its LVT0 is in EXTINT mode.
304 		 */
305 		if (kvm->arch.vapics_in_nmi_mode > 0)
306 			kvm_for_each_vcpu(i, vcpu, kvm)
307 				kvm_apic_nmi_wd_deliver(vcpu);
308 	}
309 }
310 
311 static enum hrtimer_restart pit_timer_fn(struct hrtimer *data)
312 {
313 	struct kvm_kpit_state *ps = container_of(data, struct kvm_kpit_state, timer);
314 	struct kvm_pit *pt = ps->kvm->arch.vpit;
315 
316 	if (ps->reinject || !atomic_read(&ps->pending)) {
317 		atomic_inc(&ps->pending);
318 		queue_kthread_work(&pt->worker, &pt->expired);
319 	}
320 
321 	if (ps->is_periodic) {
322 		hrtimer_add_expires_ns(&ps->timer, ps->period);
323 		return HRTIMER_RESTART;
324 	} else
325 		return HRTIMER_NORESTART;
326 }
327 
328 static void create_pit_timer(struct kvm *kvm, u32 val, int is_period)
329 {
330 	struct kvm_kpit_state *ps = &kvm->arch.vpit->pit_state;
331 	s64 interval;
332 
333 	if (!irqchip_in_kernel(kvm) || ps->flags & KVM_PIT_FLAGS_HPET_LEGACY)
334 		return;
335 
336 	interval = muldiv64(val, NSEC_PER_SEC, KVM_PIT_FREQ);
337 
338 	pr_debug("create pit timer, interval is %llu nsec\n", interval);
339 
340 	/* TODO The new value only affected after the retriggered */
341 	hrtimer_cancel(&ps->timer);
342 	flush_kthread_work(&ps->pit->expired);
343 	ps->period = interval;
344 	ps->is_periodic = is_period;
345 
346 	ps->timer.function = pit_timer_fn;
347 	ps->kvm = ps->pit->kvm;
348 
349 	atomic_set(&ps->pending, 0);
350 	ps->irq_ack = 1;
351 
352 	hrtimer_start(&ps->timer, ktime_add_ns(ktime_get(), interval),
353 		      HRTIMER_MODE_ABS);
354 }
355 
356 static void pit_load_count(struct kvm *kvm, int channel, u32 val)
357 {
358 	struct kvm_kpit_state *ps = &kvm->arch.vpit->pit_state;
359 
360 	WARN_ON(!mutex_is_locked(&ps->lock));
361 
362 	pr_debug("load_count val is %d, channel is %d\n", val, channel);
363 
364 	/*
365 	 * The largest possible initial count is 0; this is equivalent
366 	 * to 216 for binary counting and 104 for BCD counting.
367 	 */
368 	if (val == 0)
369 		val = 0x10000;
370 
371 	ps->channels[channel].count = val;
372 
373 	if (channel != 0) {
374 		ps->channels[channel].count_load_time = ktime_get();
375 		return;
376 	}
377 
378 	/* Two types of timer
379 	 * mode 1 is one shot, mode 2 is period, otherwise del timer */
380 	switch (ps->channels[0].mode) {
381 	case 0:
382 	case 1:
383         /* FIXME: enhance mode 4 precision */
384 	case 4:
385 		create_pit_timer(kvm, val, 0);
386 		break;
387 	case 2:
388 	case 3:
389 		create_pit_timer(kvm, val, 1);
390 		break;
391 	default:
392 		destroy_pit_timer(kvm->arch.vpit);
393 	}
394 }
395 
396 void kvm_pit_load_count(struct kvm *kvm, int channel, u32 val, int hpet_legacy_start)
397 {
398 	u8 saved_mode;
399 	if (hpet_legacy_start) {
400 		/* save existing mode for later reenablement */
401 		saved_mode = kvm->arch.vpit->pit_state.channels[0].mode;
402 		kvm->arch.vpit->pit_state.channels[0].mode = 0xff; /* disable timer */
403 		pit_load_count(kvm, channel, val);
404 		kvm->arch.vpit->pit_state.channels[0].mode = saved_mode;
405 	} else {
406 		pit_load_count(kvm, channel, val);
407 	}
408 }
409 
410 static inline struct kvm_pit *dev_to_pit(struct kvm_io_device *dev)
411 {
412 	return container_of(dev, struct kvm_pit, dev);
413 }
414 
415 static inline struct kvm_pit *speaker_to_pit(struct kvm_io_device *dev)
416 {
417 	return container_of(dev, struct kvm_pit, speaker_dev);
418 }
419 
420 static inline int pit_in_range(gpa_t addr)
421 {
422 	return ((addr >= KVM_PIT_BASE_ADDRESS) &&
423 		(addr < KVM_PIT_BASE_ADDRESS + KVM_PIT_MEM_LENGTH));
424 }
425 
426 static int pit_ioport_write(struct kvm_io_device *this,
427 			    gpa_t addr, int len, const void *data)
428 {
429 	struct kvm_pit *pit = dev_to_pit(this);
430 	struct kvm_kpit_state *pit_state = &pit->pit_state;
431 	struct kvm *kvm = pit->kvm;
432 	int channel, access;
433 	struct kvm_kpit_channel_state *s;
434 	u32 val = *(u32 *) data;
435 	if (!pit_in_range(addr))
436 		return -EOPNOTSUPP;
437 
438 	val  &= 0xff;
439 	addr &= KVM_PIT_CHANNEL_MASK;
440 
441 	mutex_lock(&pit_state->lock);
442 
443 	if (val != 0)
444 		pr_debug("write addr is 0x%x, len is %d, val is 0x%x\n",
445 			 (unsigned int)addr, len, val);
446 
447 	if (addr == 3) {
448 		channel = val >> 6;
449 		if (channel == 3) {
450 			/* Read-Back Command. */
451 			for (channel = 0; channel < 3; channel++) {
452 				s = &pit_state->channels[channel];
453 				if (val & (2 << channel)) {
454 					if (!(val & 0x20))
455 						pit_latch_count(kvm, channel);
456 					if (!(val & 0x10))
457 						pit_latch_status(kvm, channel);
458 				}
459 			}
460 		} else {
461 			/* Select Counter <channel>. */
462 			s = &pit_state->channels[channel];
463 			access = (val >> 4) & KVM_PIT_CHANNEL_MASK;
464 			if (access == 0) {
465 				pit_latch_count(kvm, channel);
466 			} else {
467 				s->rw_mode = access;
468 				s->read_state = access;
469 				s->write_state = access;
470 				s->mode = (val >> 1) & 7;
471 				if (s->mode > 5)
472 					s->mode -= 4;
473 				s->bcd = val & 1;
474 			}
475 		}
476 	} else {
477 		/* Write Count. */
478 		s = &pit_state->channels[addr];
479 		switch (s->write_state) {
480 		default:
481 		case RW_STATE_LSB:
482 			pit_load_count(kvm, addr, val);
483 			break;
484 		case RW_STATE_MSB:
485 			pit_load_count(kvm, addr, val << 8);
486 			break;
487 		case RW_STATE_WORD0:
488 			s->write_latch = val;
489 			s->write_state = RW_STATE_WORD1;
490 			break;
491 		case RW_STATE_WORD1:
492 			pit_load_count(kvm, addr, s->write_latch | (val << 8));
493 			s->write_state = RW_STATE_WORD0;
494 			break;
495 		}
496 	}
497 
498 	mutex_unlock(&pit_state->lock);
499 	return 0;
500 }
501 
502 static int pit_ioport_read(struct kvm_io_device *this,
503 			   gpa_t addr, int len, void *data)
504 {
505 	struct kvm_pit *pit = dev_to_pit(this);
506 	struct kvm_kpit_state *pit_state = &pit->pit_state;
507 	struct kvm *kvm = pit->kvm;
508 	int ret, count;
509 	struct kvm_kpit_channel_state *s;
510 	if (!pit_in_range(addr))
511 		return -EOPNOTSUPP;
512 
513 	addr &= KVM_PIT_CHANNEL_MASK;
514 	if (addr == 3)
515 		return 0;
516 
517 	s = &pit_state->channels[addr];
518 
519 	mutex_lock(&pit_state->lock);
520 
521 	if (s->status_latched) {
522 		s->status_latched = 0;
523 		ret = s->status;
524 	} else if (s->count_latched) {
525 		switch (s->count_latched) {
526 		default:
527 		case RW_STATE_LSB:
528 			ret = s->latched_count & 0xff;
529 			s->count_latched = 0;
530 			break;
531 		case RW_STATE_MSB:
532 			ret = s->latched_count >> 8;
533 			s->count_latched = 0;
534 			break;
535 		case RW_STATE_WORD0:
536 			ret = s->latched_count & 0xff;
537 			s->count_latched = RW_STATE_MSB;
538 			break;
539 		}
540 	} else {
541 		switch (s->read_state) {
542 		default:
543 		case RW_STATE_LSB:
544 			count = pit_get_count(kvm, addr);
545 			ret = count & 0xff;
546 			break;
547 		case RW_STATE_MSB:
548 			count = pit_get_count(kvm, addr);
549 			ret = (count >> 8) & 0xff;
550 			break;
551 		case RW_STATE_WORD0:
552 			count = pit_get_count(kvm, addr);
553 			ret = count & 0xff;
554 			s->read_state = RW_STATE_WORD1;
555 			break;
556 		case RW_STATE_WORD1:
557 			count = pit_get_count(kvm, addr);
558 			ret = (count >> 8) & 0xff;
559 			s->read_state = RW_STATE_WORD0;
560 			break;
561 		}
562 	}
563 
564 	if (len > sizeof(ret))
565 		len = sizeof(ret);
566 	memcpy(data, (char *)&ret, len);
567 
568 	mutex_unlock(&pit_state->lock);
569 	return 0;
570 }
571 
572 static int speaker_ioport_write(struct kvm_io_device *this,
573 				gpa_t addr, int len, const void *data)
574 {
575 	struct kvm_pit *pit = speaker_to_pit(this);
576 	struct kvm_kpit_state *pit_state = &pit->pit_state;
577 	struct kvm *kvm = pit->kvm;
578 	u32 val = *(u32 *) data;
579 	if (addr != KVM_SPEAKER_BASE_ADDRESS)
580 		return -EOPNOTSUPP;
581 
582 	mutex_lock(&pit_state->lock);
583 	pit_state->speaker_data_on = (val >> 1) & 1;
584 	pit_set_gate(kvm, 2, val & 1);
585 	mutex_unlock(&pit_state->lock);
586 	return 0;
587 }
588 
589 static int speaker_ioport_read(struct kvm_io_device *this,
590 			       gpa_t addr, int len, void *data)
591 {
592 	struct kvm_pit *pit = speaker_to_pit(this);
593 	struct kvm_kpit_state *pit_state = &pit->pit_state;
594 	struct kvm *kvm = pit->kvm;
595 	unsigned int refresh_clock;
596 	int ret;
597 	if (addr != KVM_SPEAKER_BASE_ADDRESS)
598 		return -EOPNOTSUPP;
599 
600 	/* Refresh clock toggles at about 15us. We approximate as 2^14ns. */
601 	refresh_clock = ((unsigned int)ktime_to_ns(ktime_get()) >> 14) & 1;
602 
603 	mutex_lock(&pit_state->lock);
604 	ret = ((pit_state->speaker_data_on << 1) | pit_get_gate(kvm, 2) |
605 		(pit_get_out(kvm, 2) << 5) | (refresh_clock << 4));
606 	if (len > sizeof(ret))
607 		len = sizeof(ret);
608 	memcpy(data, (char *)&ret, len);
609 	mutex_unlock(&pit_state->lock);
610 	return 0;
611 }
612 
613 void kvm_pit_reset(struct kvm_pit *pit)
614 {
615 	int i;
616 	struct kvm_kpit_channel_state *c;
617 
618 	mutex_lock(&pit->pit_state.lock);
619 	pit->pit_state.flags = 0;
620 	for (i = 0; i < 3; i++) {
621 		c = &pit->pit_state.channels[i];
622 		c->mode = 0xff;
623 		c->gate = (i != 2);
624 		pit_load_count(pit->kvm, i, 0);
625 	}
626 	mutex_unlock(&pit->pit_state.lock);
627 
628 	atomic_set(&pit->pit_state.pending, 0);
629 	pit->pit_state.irq_ack = 1;
630 }
631 
632 static void pit_mask_notifer(struct kvm_irq_mask_notifier *kimn, bool mask)
633 {
634 	struct kvm_pit *pit = container_of(kimn, struct kvm_pit, mask_notifier);
635 
636 	if (!mask) {
637 		atomic_set(&pit->pit_state.pending, 0);
638 		pit->pit_state.irq_ack = 1;
639 	}
640 }
641 
642 static const struct kvm_io_device_ops pit_dev_ops = {
643 	.read     = pit_ioport_read,
644 	.write    = pit_ioport_write,
645 };
646 
647 static const struct kvm_io_device_ops speaker_dev_ops = {
648 	.read     = speaker_ioport_read,
649 	.write    = speaker_ioport_write,
650 };
651 
652 /* Caller must hold slots_lock */
653 struct kvm_pit *kvm_create_pit(struct kvm *kvm, u32 flags)
654 {
655 	struct kvm_pit *pit;
656 	struct kvm_kpit_state *pit_state;
657 	struct pid *pid;
658 	pid_t pid_nr;
659 	int ret;
660 
661 	pit = kzalloc(sizeof(struct kvm_pit), GFP_KERNEL);
662 	if (!pit)
663 		return NULL;
664 
665 	pit->irq_source_id = kvm_request_irq_source_id(kvm);
666 	if (pit->irq_source_id < 0) {
667 		kfree(pit);
668 		return NULL;
669 	}
670 
671 	mutex_init(&pit->pit_state.lock);
672 	mutex_lock(&pit->pit_state.lock);
673 	spin_lock_init(&pit->pit_state.inject_lock);
674 
675 	pid = get_pid(task_tgid(current));
676 	pid_nr = pid_vnr(pid);
677 	put_pid(pid);
678 
679 	init_kthread_worker(&pit->worker);
680 	pit->worker_task = kthread_run(kthread_worker_fn, &pit->worker,
681 				       "kvm-pit/%d", pid_nr);
682 	if (IS_ERR(pit->worker_task)) {
683 		mutex_unlock(&pit->pit_state.lock);
684 		kvm_free_irq_source_id(kvm, pit->irq_source_id);
685 		kfree(pit);
686 		return NULL;
687 	}
688 	init_kthread_work(&pit->expired, pit_do_work);
689 
690 	kvm->arch.vpit = pit;
691 	pit->kvm = kvm;
692 
693 	pit_state = &pit->pit_state;
694 	pit_state->pit = pit;
695 	hrtimer_init(&pit_state->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
696 	pit_state->irq_ack_notifier.gsi = 0;
697 	pit_state->irq_ack_notifier.irq_acked = kvm_pit_ack_irq;
698 	kvm_register_irq_ack_notifier(kvm, &pit_state->irq_ack_notifier);
699 	pit_state->reinject = true;
700 	mutex_unlock(&pit->pit_state.lock);
701 
702 	kvm_pit_reset(pit);
703 
704 	pit->mask_notifier.func = pit_mask_notifer;
705 	kvm_register_irq_mask_notifier(kvm, 0, &pit->mask_notifier);
706 
707 	kvm_iodevice_init(&pit->dev, &pit_dev_ops);
708 	ret = kvm_io_bus_register_dev(kvm, KVM_PIO_BUS, KVM_PIT_BASE_ADDRESS,
709 				      KVM_PIT_MEM_LENGTH, &pit->dev);
710 	if (ret < 0)
711 		goto fail;
712 
713 	if (flags & KVM_PIT_SPEAKER_DUMMY) {
714 		kvm_iodevice_init(&pit->speaker_dev, &speaker_dev_ops);
715 		ret = kvm_io_bus_register_dev(kvm, KVM_PIO_BUS,
716 					      KVM_SPEAKER_BASE_ADDRESS, 4,
717 					      &pit->speaker_dev);
718 		if (ret < 0)
719 			goto fail_unregister;
720 	}
721 
722 	return pit;
723 
724 fail_unregister:
725 	kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS, &pit->dev);
726 
727 fail:
728 	kvm_unregister_irq_mask_notifier(kvm, 0, &pit->mask_notifier);
729 	kvm_unregister_irq_ack_notifier(kvm, &pit_state->irq_ack_notifier);
730 	kvm_free_irq_source_id(kvm, pit->irq_source_id);
731 	kthread_stop(pit->worker_task);
732 	kfree(pit);
733 	return NULL;
734 }
735 
736 void kvm_free_pit(struct kvm *kvm)
737 {
738 	struct hrtimer *timer;
739 
740 	if (kvm->arch.vpit) {
741 		kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS, &kvm->arch.vpit->dev);
742 		kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
743 					      &kvm->arch.vpit->speaker_dev);
744 		kvm_unregister_irq_mask_notifier(kvm, 0,
745 					       &kvm->arch.vpit->mask_notifier);
746 		kvm_unregister_irq_ack_notifier(kvm,
747 				&kvm->arch.vpit->pit_state.irq_ack_notifier);
748 		mutex_lock(&kvm->arch.vpit->pit_state.lock);
749 		timer = &kvm->arch.vpit->pit_state.timer;
750 		hrtimer_cancel(timer);
751 		flush_kthread_work(&kvm->arch.vpit->expired);
752 		kthread_stop(kvm->arch.vpit->worker_task);
753 		kvm_free_irq_source_id(kvm, kvm->arch.vpit->irq_source_id);
754 		mutex_unlock(&kvm->arch.vpit->pit_state.lock);
755 		kfree(kvm->arch.vpit);
756 	}
757 }
758