KVM: x86: make vapics_in_nmi_mode atomic
[pandora-kernel.git] / arch / x86 / kvm / i8254.c
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 #include <linux/workqueue.h>
38
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->pit_timer.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->pit_timer.timer);
126         elapsed = ps->pit_timer.period - ktime_to_ns(remaining);
127         elapsed = mod_64(elapsed, ps->pit_timer.period);
128
129         return elapsed;
130 }
131
132 static s64 kpit_elapsed(struct kvm *kvm, struct kvm_kpit_channel_state *c,
133                         int channel)
134 {
135         if (channel == 0)
136                 return __kpit_elapsed(kvm);
137
138         return ktime_to_ns(ktime_sub(ktime_get(), c->count_load_time));
139 }
140
141 static int pit_get_count(struct kvm *kvm, int channel)
142 {
143         struct kvm_kpit_channel_state *c =
144                 &kvm->arch.vpit->pit_state.channels[channel];
145         s64 d, t;
146         int counter;
147
148         WARN_ON(!mutex_is_locked(&kvm->arch.vpit->pit_state.lock));
149
150         t = kpit_elapsed(kvm, c, channel);
151         d = muldiv64(t, KVM_PIT_FREQ, NSEC_PER_SEC);
152
153         switch (c->mode) {
154         case 0:
155         case 1:
156         case 4:
157         case 5:
158                 counter = (c->count - d) & 0xffff;
159                 break;
160         case 3:
161                 /* XXX: may be incorrect for odd counts */
162                 counter = c->count - (mod_64((2 * d), c->count));
163                 break;
164         default:
165                 counter = c->count - mod_64(d, c->count);
166                 break;
167         }
168         return counter;
169 }
170
171 static int pit_get_out(struct kvm *kvm, int channel)
172 {
173         struct kvm_kpit_channel_state *c =
174                 &kvm->arch.vpit->pit_state.channels[channel];
175         s64 d, t;
176         int out;
177
178         WARN_ON(!mutex_is_locked(&kvm->arch.vpit->pit_state.lock));
179
180         t = kpit_elapsed(kvm, c, channel);
181         d = muldiv64(t, KVM_PIT_FREQ, NSEC_PER_SEC);
182
183         switch (c->mode) {
184         default:
185         case 0:
186                 out = (d >= c->count);
187                 break;
188         case 1:
189                 out = (d < c->count);
190                 break;
191         case 2:
192                 out = ((mod_64(d, c->count) == 0) && (d != 0));
193                 break;
194         case 3:
195                 out = (mod_64(d, c->count) < ((c->count + 1) >> 1));
196                 break;
197         case 4:
198         case 5:
199                 out = (d == c->count);
200                 break;
201         }
202
203         return out;
204 }
205
206 static void pit_latch_count(struct kvm *kvm, int channel)
207 {
208         struct kvm_kpit_channel_state *c =
209                 &kvm->arch.vpit->pit_state.channels[channel];
210
211         WARN_ON(!mutex_is_locked(&kvm->arch.vpit->pit_state.lock));
212
213         if (!c->count_latched) {
214                 c->latched_count = pit_get_count(kvm, channel);
215                 c->count_latched = c->rw_mode;
216         }
217 }
218
219 static void pit_latch_status(struct kvm *kvm, int channel)
220 {
221         struct kvm_kpit_channel_state *c =
222                 &kvm->arch.vpit->pit_state.channels[channel];
223
224         WARN_ON(!mutex_is_locked(&kvm->arch.vpit->pit_state.lock));
225
226         if (!c->status_latched) {
227                 /* TODO: Return NULL COUNT (bit 6). */
228                 c->status = ((pit_get_out(kvm, channel) << 7) |
229                                 (c->rw_mode << 4) |
230                                 (c->mode << 1) |
231                                 c->bcd);
232                 c->status_latched = 1;
233         }
234 }
235
236 static void kvm_pit_ack_irq(struct kvm_irq_ack_notifier *kian)
237 {
238         struct kvm_kpit_state *ps = container_of(kian, struct kvm_kpit_state,
239                                                  irq_ack_notifier);
240         int value;
241
242         spin_lock(&ps->inject_lock);
243         value = atomic_dec_return(&ps->pit_timer.pending);
244         if (value < 0)
245                 /* spurious acks can be generated if, for example, the
246                  * PIC is being reset.  Handle it gracefully here
247                  */
248                 atomic_inc(&ps->pit_timer.pending);
249         else if (value > 0)
250                 /* in this case, we had multiple outstanding pit interrupts
251                  * that we needed to inject.  Reinject
252                  */
253                 queue_work(ps->pit->wq, &ps->pit->expired);
254         ps->irq_ack = 1;
255         spin_unlock(&ps->inject_lock);
256 }
257
258 void __kvm_migrate_pit_timer(struct kvm_vcpu *vcpu)
259 {
260         struct kvm_pit *pit = vcpu->kvm->arch.vpit;
261         struct hrtimer *timer;
262
263         if (!kvm_vcpu_is_bsp(vcpu) || !pit)
264                 return;
265
266         timer = &pit->pit_state.pit_timer.timer;
267         mutex_lock(&pit->pit_state.lock);
268         if (hrtimer_cancel(timer))
269                 hrtimer_start_expires(timer, HRTIMER_MODE_ABS);
270         mutex_unlock(&pit->pit_state.lock);
271 }
272
273 static void destroy_pit_timer(struct kvm_pit *pit)
274 {
275         hrtimer_cancel(&pit->pit_state.pit_timer.timer);
276         cancel_work_sync(&pit->expired);
277 }
278
279 static bool kpit_is_periodic(struct kvm_timer *ktimer)
280 {
281         struct kvm_kpit_state *ps = container_of(ktimer, struct kvm_kpit_state,
282                                                  pit_timer);
283         return ps->is_periodic;
284 }
285
286 static struct kvm_timer_ops kpit_ops = {
287         .is_periodic = kpit_is_periodic,
288 };
289
290 static void pit_do_work(struct work_struct *work)
291 {
292         struct kvm_pit *pit = container_of(work, struct kvm_pit, expired);
293         struct kvm *kvm = pit->kvm;
294         struct kvm_vcpu *vcpu;
295         int i;
296         struct kvm_kpit_state *ps = &pit->pit_state;
297         int inject = 0;
298
299         /* Try to inject pending interrupts when
300          * last one has been acked.
301          */
302         spin_lock(&ps->inject_lock);
303         if (ps->irq_ack) {
304                 ps->irq_ack = 0;
305                 inject = 1;
306         }
307         spin_unlock(&ps->inject_lock);
308         if (inject) {
309                 kvm_set_irq(kvm, kvm->arch.vpit->irq_source_id, 0, 1);
310                 kvm_set_irq(kvm, kvm->arch.vpit->irq_source_id, 0, 0);
311
312                 /*
313                  * Provides NMI watchdog support via Virtual Wire mode.
314                  * The route is: PIT -> PIC -> LVT0 in NMI mode.
315                  *
316                  * Note: Our Virtual Wire implementation is simplified, only
317                  * propagating PIT interrupts to all VCPUs when they have set
318                  * LVT0 to NMI delivery. Other PIC interrupts are just sent to
319                  * VCPU0, and only if its LVT0 is in EXTINT mode.
320                  */
321                 if (atomic_read(&kvm->arch.vapics_in_nmi_mode) > 0)
322                         kvm_for_each_vcpu(i, vcpu, kvm)
323                                 kvm_apic_nmi_wd_deliver(vcpu);
324         }
325 }
326
327 static enum hrtimer_restart pit_timer_fn(struct hrtimer *data)
328 {
329         struct kvm_timer *ktimer = container_of(data, struct kvm_timer, timer);
330         struct kvm_pit *pt = ktimer->kvm->arch.vpit;
331
332         if (ktimer->reinject || !atomic_read(&ktimer->pending)) {
333                 atomic_inc(&ktimer->pending);
334                 queue_work(pt->wq, &pt->expired);
335         }
336
337         if (ktimer->t_ops->is_periodic(ktimer)) {
338                 hrtimer_add_expires_ns(&ktimer->timer, ktimer->period);
339                 return HRTIMER_RESTART;
340         } else
341                 return HRTIMER_NORESTART;
342 }
343
344 static void create_pit_timer(struct kvm *kvm, u32 val, int is_period)
345 {
346         struct kvm_kpit_state *ps = &kvm->arch.vpit->pit_state;
347         struct kvm_timer *pt = &ps->pit_timer;
348         s64 interval;
349
350         if (!irqchip_in_kernel(kvm))
351                 return;
352
353         interval = muldiv64(val, NSEC_PER_SEC, KVM_PIT_FREQ);
354
355         pr_debug("create pit timer, interval is %llu nsec\n", interval);
356
357         /* TODO The new value only affected after the retriggered */
358         hrtimer_cancel(&pt->timer);
359         cancel_work_sync(&ps->pit->expired);
360         pt->period = interval;
361         ps->is_periodic = is_period;
362
363         pt->timer.function = pit_timer_fn;
364         pt->t_ops = &kpit_ops;
365         pt->kvm = ps->pit->kvm;
366
367         atomic_set(&pt->pending, 0);
368         ps->irq_ack = 1;
369
370         /*
371          * Do not allow the guest to program periodic timers with small
372          * interval, since the hrtimers are not throttled by the host
373          * scheduler.
374          */
375         if (ps->is_periodic) {
376                 s64 min_period = min_timer_period_us * 1000LL;
377
378                 if (pt->period < min_period) {
379                         pr_info_ratelimited(
380                             "kvm: requested %lld ns "
381                             "i8254 timer period limited to %lld ns\n",
382                             pt->period, min_period);
383                         pt->period = min_period;
384                 }
385         }
386
387         hrtimer_start(&pt->timer, ktime_add_ns(ktime_get(), interval),
388                       HRTIMER_MODE_ABS);
389 }
390
391 static void pit_load_count(struct kvm *kvm, int channel, u32 val)
392 {
393         struct kvm_kpit_state *ps = &kvm->arch.vpit->pit_state;
394
395         WARN_ON(!mutex_is_locked(&ps->lock));
396
397         pr_debug("load_count val is %d, channel is %d\n", val, channel);
398
399         /*
400          * The largest possible initial count is 0; this is equivalent
401          * to 216 for binary counting and 104 for BCD counting.
402          */
403         if (val == 0)
404                 val = 0x10000;
405
406         ps->channels[channel].count = val;
407
408         if (channel != 0) {
409                 ps->channels[channel].count_load_time = ktime_get();
410                 return;
411         }
412
413         /* Two types of timer
414          * mode 1 is one shot, mode 2 is period, otherwise del timer */
415         switch (ps->channels[0].mode) {
416         case 0:
417         case 1:
418         /* FIXME: enhance mode 4 precision */
419         case 4:
420                 if (!(ps->flags & KVM_PIT_FLAGS_HPET_LEGACY)) {
421                         create_pit_timer(kvm, val, 0);
422                 }
423                 break;
424         case 2:
425         case 3:
426                 if (!(ps->flags & KVM_PIT_FLAGS_HPET_LEGACY)){
427                         create_pit_timer(kvm, val, 1);
428                 }
429                 break;
430         default:
431                 destroy_pit_timer(kvm->arch.vpit);
432         }
433 }
434
435 void kvm_pit_load_count(struct kvm *kvm, int channel, u32 val, int hpet_legacy_start)
436 {
437         u8 saved_mode;
438         if (hpet_legacy_start) {
439                 /* save existing mode for later reenablement */
440                 saved_mode = kvm->arch.vpit->pit_state.channels[0].mode;
441                 kvm->arch.vpit->pit_state.channels[0].mode = 0xff; /* disable timer */
442                 pit_load_count(kvm, channel, val);
443                 kvm->arch.vpit->pit_state.channels[0].mode = saved_mode;
444         } else {
445                 pit_load_count(kvm, channel, val);
446         }
447 }
448
449 static inline struct kvm_pit *dev_to_pit(struct kvm_io_device *dev)
450 {
451         return container_of(dev, struct kvm_pit, dev);
452 }
453
454 static inline struct kvm_pit *speaker_to_pit(struct kvm_io_device *dev)
455 {
456         return container_of(dev, struct kvm_pit, speaker_dev);
457 }
458
459 static inline int pit_in_range(gpa_t addr)
460 {
461         return ((addr >= KVM_PIT_BASE_ADDRESS) &&
462                 (addr < KVM_PIT_BASE_ADDRESS + KVM_PIT_MEM_LENGTH));
463 }
464
465 static int pit_ioport_write(struct kvm_io_device *this,
466                             gpa_t addr, int len, const void *data)
467 {
468         struct kvm_pit *pit = dev_to_pit(this);
469         struct kvm_kpit_state *pit_state = &pit->pit_state;
470         struct kvm *kvm = pit->kvm;
471         int channel, access;
472         struct kvm_kpit_channel_state *s;
473         u32 val = *(u32 *) data;
474         if (!pit_in_range(addr))
475                 return -EOPNOTSUPP;
476
477         val  &= 0xff;
478         addr &= KVM_PIT_CHANNEL_MASK;
479
480         mutex_lock(&pit_state->lock);
481
482         if (val != 0)
483                 pr_debug("write addr is 0x%x, len is %d, val is 0x%x\n",
484                          (unsigned int)addr, len, val);
485
486         if (addr == 3) {
487                 channel = val >> 6;
488                 if (channel == 3) {
489                         /* Read-Back Command. */
490                         for (channel = 0; channel < 3; channel++) {
491                                 s = &pit_state->channels[channel];
492                                 if (val & (2 << channel)) {
493                                         if (!(val & 0x20))
494                                                 pit_latch_count(kvm, channel);
495                                         if (!(val & 0x10))
496                                                 pit_latch_status(kvm, channel);
497                                 }
498                         }
499                 } else {
500                         /* Select Counter <channel>. */
501                         s = &pit_state->channels[channel];
502                         access = (val >> 4) & KVM_PIT_CHANNEL_MASK;
503                         if (access == 0) {
504                                 pit_latch_count(kvm, channel);
505                         } else {
506                                 s->rw_mode = access;
507                                 s->read_state = access;
508                                 s->write_state = access;
509                                 s->mode = (val >> 1) & 7;
510                                 if (s->mode > 5)
511                                         s->mode -= 4;
512                                 s->bcd = val & 1;
513                         }
514                 }
515         } else {
516                 /* Write Count. */
517                 s = &pit_state->channels[addr];
518                 switch (s->write_state) {
519                 default:
520                 case RW_STATE_LSB:
521                         pit_load_count(kvm, addr, val);
522                         break;
523                 case RW_STATE_MSB:
524                         pit_load_count(kvm, addr, val << 8);
525                         break;
526                 case RW_STATE_WORD0:
527                         s->write_latch = val;
528                         s->write_state = RW_STATE_WORD1;
529                         break;
530                 case RW_STATE_WORD1:
531                         pit_load_count(kvm, addr, s->write_latch | (val << 8));
532                         s->write_state = RW_STATE_WORD0;
533                         break;
534                 }
535         }
536
537         mutex_unlock(&pit_state->lock);
538         return 0;
539 }
540
541 static int pit_ioport_read(struct kvm_io_device *this,
542                            gpa_t addr, int len, void *data)
543 {
544         struct kvm_pit *pit = dev_to_pit(this);
545         struct kvm_kpit_state *pit_state = &pit->pit_state;
546         struct kvm *kvm = pit->kvm;
547         int ret, count;
548         struct kvm_kpit_channel_state *s;
549         if (!pit_in_range(addr))
550                 return -EOPNOTSUPP;
551
552         addr &= KVM_PIT_CHANNEL_MASK;
553         if (addr == 3)
554                 return 0;
555
556         s = &pit_state->channels[addr];
557
558         mutex_lock(&pit_state->lock);
559
560         if (s->status_latched) {
561                 s->status_latched = 0;
562                 ret = s->status;
563         } else if (s->count_latched) {
564                 switch (s->count_latched) {
565                 default:
566                 case RW_STATE_LSB:
567                         ret = s->latched_count & 0xff;
568                         s->count_latched = 0;
569                         break;
570                 case RW_STATE_MSB:
571                         ret = s->latched_count >> 8;
572                         s->count_latched = 0;
573                         break;
574                 case RW_STATE_WORD0:
575                         ret = s->latched_count & 0xff;
576                         s->count_latched = RW_STATE_MSB;
577                         break;
578                 }
579         } else {
580                 switch (s->read_state) {
581                 default:
582                 case RW_STATE_LSB:
583                         count = pit_get_count(kvm, addr);
584                         ret = count & 0xff;
585                         break;
586                 case RW_STATE_MSB:
587                         count = pit_get_count(kvm, addr);
588                         ret = (count >> 8) & 0xff;
589                         break;
590                 case RW_STATE_WORD0:
591                         count = pit_get_count(kvm, addr);
592                         ret = count & 0xff;
593                         s->read_state = RW_STATE_WORD1;
594                         break;
595                 case RW_STATE_WORD1:
596                         count = pit_get_count(kvm, addr);
597                         ret = (count >> 8) & 0xff;
598                         s->read_state = RW_STATE_WORD0;
599                         break;
600                 }
601         }
602
603         if (len > sizeof(ret))
604                 len = sizeof(ret);
605         memcpy(data, (char *)&ret, len);
606
607         mutex_unlock(&pit_state->lock);
608         return 0;
609 }
610
611 static int speaker_ioport_write(struct kvm_io_device *this,
612                                 gpa_t addr, int len, const void *data)
613 {
614         struct kvm_pit *pit = speaker_to_pit(this);
615         struct kvm_kpit_state *pit_state = &pit->pit_state;
616         struct kvm *kvm = pit->kvm;
617         u32 val = *(u32 *) data;
618         if (addr != KVM_SPEAKER_BASE_ADDRESS)
619                 return -EOPNOTSUPP;
620
621         mutex_lock(&pit_state->lock);
622         pit_state->speaker_data_on = (val >> 1) & 1;
623         pit_set_gate(kvm, 2, val & 1);
624         mutex_unlock(&pit_state->lock);
625         return 0;
626 }
627
628 static int speaker_ioport_read(struct kvm_io_device *this,
629                                gpa_t addr, int len, void *data)
630 {
631         struct kvm_pit *pit = speaker_to_pit(this);
632         struct kvm_kpit_state *pit_state = &pit->pit_state;
633         struct kvm *kvm = pit->kvm;
634         unsigned int refresh_clock;
635         int ret;
636         if (addr != KVM_SPEAKER_BASE_ADDRESS)
637                 return -EOPNOTSUPP;
638
639         /* Refresh clock toggles at about 15us. We approximate as 2^14ns. */
640         refresh_clock = ((unsigned int)ktime_to_ns(ktime_get()) >> 14) & 1;
641
642         mutex_lock(&pit_state->lock);
643         ret = ((pit_state->speaker_data_on << 1) | pit_get_gate(kvm, 2) |
644                 (pit_get_out(kvm, 2) << 5) | (refresh_clock << 4));
645         if (len > sizeof(ret))
646                 len = sizeof(ret);
647         memcpy(data, (char *)&ret, len);
648         mutex_unlock(&pit_state->lock);
649         return 0;
650 }
651
652 void kvm_pit_reset(struct kvm_pit *pit)
653 {
654         int i;
655         struct kvm_kpit_channel_state *c;
656
657         mutex_lock(&pit->pit_state.lock);
658         pit->pit_state.flags = 0;
659         for (i = 0; i < 3; i++) {
660                 c = &pit->pit_state.channels[i];
661                 c->mode = 0xff;
662                 c->gate = (i != 2);
663                 pit_load_count(pit->kvm, i, 0);
664         }
665         mutex_unlock(&pit->pit_state.lock);
666
667         atomic_set(&pit->pit_state.pit_timer.pending, 0);
668         pit->pit_state.irq_ack = 1;
669 }
670
671 static void pit_mask_notifer(struct kvm_irq_mask_notifier *kimn, bool mask)
672 {
673         struct kvm_pit *pit = container_of(kimn, struct kvm_pit, mask_notifier);
674
675         if (!mask) {
676                 atomic_set(&pit->pit_state.pit_timer.pending, 0);
677                 pit->pit_state.irq_ack = 1;
678         }
679 }
680
681 static const struct kvm_io_device_ops pit_dev_ops = {
682         .read     = pit_ioport_read,
683         .write    = pit_ioport_write,
684 };
685
686 static const struct kvm_io_device_ops speaker_dev_ops = {
687         .read     = speaker_ioport_read,
688         .write    = speaker_ioport_write,
689 };
690
691 /* Caller must hold slots_lock */
692 struct kvm_pit *kvm_create_pit(struct kvm *kvm, u32 flags)
693 {
694         struct kvm_pit *pit;
695         struct kvm_kpit_state *pit_state;
696         int ret;
697
698         pit = kzalloc(sizeof(struct kvm_pit), GFP_KERNEL);
699         if (!pit)
700                 return NULL;
701
702         pit->irq_source_id = kvm_request_irq_source_id(kvm);
703         if (pit->irq_source_id < 0) {
704                 kfree(pit);
705                 return NULL;
706         }
707
708         mutex_init(&pit->pit_state.lock);
709         mutex_lock(&pit->pit_state.lock);
710         spin_lock_init(&pit->pit_state.inject_lock);
711
712         pit->wq = create_singlethread_workqueue("kvm-pit-wq");
713         if (!pit->wq) {
714                 mutex_unlock(&pit->pit_state.lock);
715                 kvm_free_irq_source_id(kvm, pit->irq_source_id);
716                 kfree(pit);
717                 return NULL;
718         }
719         INIT_WORK(&pit->expired, pit_do_work);
720
721         kvm->arch.vpit = pit;
722         pit->kvm = kvm;
723
724         pit_state = &pit->pit_state;
725         pit_state->pit = pit;
726         hrtimer_init(&pit_state->pit_timer.timer,
727                      CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
728         pit_state->irq_ack_notifier.gsi = 0;
729         pit_state->irq_ack_notifier.irq_acked = kvm_pit_ack_irq;
730         kvm_register_irq_ack_notifier(kvm, &pit_state->irq_ack_notifier);
731         pit_state->pit_timer.reinject = true;
732         mutex_unlock(&pit->pit_state.lock);
733
734         kvm_pit_reset(pit);
735
736         pit->mask_notifier.func = pit_mask_notifer;
737         kvm_register_irq_mask_notifier(kvm, 0, &pit->mask_notifier);
738
739         kvm_iodevice_init(&pit->dev, &pit_dev_ops);
740         ret = kvm_io_bus_register_dev(kvm, KVM_PIO_BUS, KVM_PIT_BASE_ADDRESS,
741                                       KVM_PIT_MEM_LENGTH, &pit->dev);
742         if (ret < 0)
743                 goto fail;
744
745         if (flags & KVM_PIT_SPEAKER_DUMMY) {
746                 kvm_iodevice_init(&pit->speaker_dev, &speaker_dev_ops);
747                 ret = kvm_io_bus_register_dev(kvm, KVM_PIO_BUS,
748                                               KVM_SPEAKER_BASE_ADDRESS, 4,
749                                               &pit->speaker_dev);
750                 if (ret < 0)
751                         goto fail_unregister;
752         }
753
754         return pit;
755
756 fail_unregister:
757         kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS, &pit->dev);
758
759 fail:
760         kvm_unregister_irq_mask_notifier(kvm, 0, &pit->mask_notifier);
761         kvm_unregister_irq_ack_notifier(kvm, &pit_state->irq_ack_notifier);
762         kvm_free_irq_source_id(kvm, pit->irq_source_id);
763         destroy_workqueue(pit->wq);
764         kfree(pit);
765         return NULL;
766 }
767
768 void kvm_free_pit(struct kvm *kvm)
769 {
770         struct hrtimer *timer;
771
772         if (kvm->arch.vpit) {
773                 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS, &kvm->arch.vpit->dev);
774                 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
775                                               &kvm->arch.vpit->speaker_dev);
776                 kvm_unregister_irq_mask_notifier(kvm, 0,
777                                                &kvm->arch.vpit->mask_notifier);
778                 kvm_unregister_irq_ack_notifier(kvm,
779                                 &kvm->arch.vpit->pit_state.irq_ack_notifier);
780                 mutex_lock(&kvm->arch.vpit->pit_state.lock);
781                 timer = &kvm->arch.vpit->pit_state.pit_timer.timer;
782                 hrtimer_cancel(timer);
783                 cancel_work_sync(&kvm->arch.vpit->expired);
784                 kvm_free_irq_source_id(kvm, kvm->arch.vpit->irq_source_id);
785                 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
786                 destroy_workqueue(kvm->arch.vpit->wq);
787                 kfree(kvm->arch.vpit);
788         }
789 }