Linux 4.4-rc8
[linux-2.6-block.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
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                 saved_mode = kvm->arch.vpit->pit_state.channels[0].mode;
424                 kvm->arch.vpit->pit_state.channels[0].mode = 0xff; /* disable timer */
425                 pit_load_count(kvm, channel, val);
426                 kvm->arch.vpit->pit_state.channels[0].mode = saved_mode;
427         } else {
428                 pit_load_count(kvm, channel, val);
429         }
430 }
431
432 static inline struct kvm_pit *dev_to_pit(struct kvm_io_device *dev)
433 {
434         return container_of(dev, struct kvm_pit, dev);
435 }
436
437 static inline struct kvm_pit *speaker_to_pit(struct kvm_io_device *dev)
438 {
439         return container_of(dev, struct kvm_pit, speaker_dev);
440 }
441
442 static inline int pit_in_range(gpa_t addr)
443 {
444         return ((addr >= KVM_PIT_BASE_ADDRESS) &&
445                 (addr < KVM_PIT_BASE_ADDRESS + KVM_PIT_MEM_LENGTH));
446 }
447
448 static int pit_ioport_write(struct kvm_vcpu *vcpu,
449                                 struct kvm_io_device *this,
450                             gpa_t addr, int len, const void *data)
451 {
452         struct kvm_pit *pit = dev_to_pit(this);
453         struct kvm_kpit_state *pit_state = &pit->pit_state;
454         struct kvm *kvm = pit->kvm;
455         int channel, access;
456         struct kvm_kpit_channel_state *s;
457         u32 val = *(u32 *) data;
458         if (!pit_in_range(addr))
459                 return -EOPNOTSUPP;
460
461         val  &= 0xff;
462         addr &= KVM_PIT_CHANNEL_MASK;
463
464         mutex_lock(&pit_state->lock);
465
466         if (val != 0)
467                 pr_debug("write addr is 0x%x, len is %d, val is 0x%x\n",
468                          (unsigned int)addr, len, val);
469
470         if (addr == 3) {
471                 channel = val >> 6;
472                 if (channel == 3) {
473                         /* Read-Back Command. */
474                         for (channel = 0; channel < 3; channel++) {
475                                 s = &pit_state->channels[channel];
476                                 if (val & (2 << channel)) {
477                                         if (!(val & 0x20))
478                                                 pit_latch_count(kvm, channel);
479                                         if (!(val & 0x10))
480                                                 pit_latch_status(kvm, channel);
481                                 }
482                         }
483                 } else {
484                         /* Select Counter <channel>. */
485                         s = &pit_state->channels[channel];
486                         access = (val >> 4) & KVM_PIT_CHANNEL_MASK;
487                         if (access == 0) {
488                                 pit_latch_count(kvm, channel);
489                         } else {
490                                 s->rw_mode = access;
491                                 s->read_state = access;
492                                 s->write_state = access;
493                                 s->mode = (val >> 1) & 7;
494                                 if (s->mode > 5)
495                                         s->mode -= 4;
496                                 s->bcd = val & 1;
497                         }
498                 }
499         } else {
500                 /* Write Count. */
501                 s = &pit_state->channels[addr];
502                 switch (s->write_state) {
503                 default:
504                 case RW_STATE_LSB:
505                         pit_load_count(kvm, addr, val);
506                         break;
507                 case RW_STATE_MSB:
508                         pit_load_count(kvm, addr, val << 8);
509                         break;
510                 case RW_STATE_WORD0:
511                         s->write_latch = val;
512                         s->write_state = RW_STATE_WORD1;
513                         break;
514                 case RW_STATE_WORD1:
515                         pit_load_count(kvm, addr, s->write_latch | (val << 8));
516                         s->write_state = RW_STATE_WORD0;
517                         break;
518                 }
519         }
520
521         mutex_unlock(&pit_state->lock);
522         return 0;
523 }
524
525 static int pit_ioport_read(struct kvm_vcpu *vcpu,
526                            struct kvm_io_device *this,
527                            gpa_t addr, int len, void *data)
528 {
529         struct kvm_pit *pit = dev_to_pit(this);
530         struct kvm_kpit_state *pit_state = &pit->pit_state;
531         struct kvm *kvm = pit->kvm;
532         int ret, count;
533         struct kvm_kpit_channel_state *s;
534         if (!pit_in_range(addr))
535                 return -EOPNOTSUPP;
536
537         addr &= KVM_PIT_CHANNEL_MASK;
538         if (addr == 3)
539                 return 0;
540
541         s = &pit_state->channels[addr];
542
543         mutex_lock(&pit_state->lock);
544
545         if (s->status_latched) {
546                 s->status_latched = 0;
547                 ret = s->status;
548         } else if (s->count_latched) {
549                 switch (s->count_latched) {
550                 default:
551                 case RW_STATE_LSB:
552                         ret = s->latched_count & 0xff;
553                         s->count_latched = 0;
554                         break;
555                 case RW_STATE_MSB:
556                         ret = s->latched_count >> 8;
557                         s->count_latched = 0;
558                         break;
559                 case RW_STATE_WORD0:
560                         ret = s->latched_count & 0xff;
561                         s->count_latched = RW_STATE_MSB;
562                         break;
563                 }
564         } else {
565                 switch (s->read_state) {
566                 default:
567                 case RW_STATE_LSB:
568                         count = pit_get_count(kvm, addr);
569                         ret = count & 0xff;
570                         break;
571                 case RW_STATE_MSB:
572                         count = pit_get_count(kvm, addr);
573                         ret = (count >> 8) & 0xff;
574                         break;
575                 case RW_STATE_WORD0:
576                         count = pit_get_count(kvm, addr);
577                         ret = count & 0xff;
578                         s->read_state = RW_STATE_WORD1;
579                         break;
580                 case RW_STATE_WORD1:
581                         count = pit_get_count(kvm, addr);
582                         ret = (count >> 8) & 0xff;
583                         s->read_state = RW_STATE_WORD0;
584                         break;
585                 }
586         }
587
588         if (len > sizeof(ret))
589                 len = sizeof(ret);
590         memcpy(data, (char *)&ret, len);
591
592         mutex_unlock(&pit_state->lock);
593         return 0;
594 }
595
596 static int speaker_ioport_write(struct kvm_vcpu *vcpu,
597                                 struct kvm_io_device *this,
598                                 gpa_t addr, int len, const void *data)
599 {
600         struct kvm_pit *pit = speaker_to_pit(this);
601         struct kvm_kpit_state *pit_state = &pit->pit_state;
602         struct kvm *kvm = pit->kvm;
603         u32 val = *(u32 *) data;
604         if (addr != KVM_SPEAKER_BASE_ADDRESS)
605                 return -EOPNOTSUPP;
606
607         mutex_lock(&pit_state->lock);
608         pit_state->speaker_data_on = (val >> 1) & 1;
609         pit_set_gate(kvm, 2, val & 1);
610         mutex_unlock(&pit_state->lock);
611         return 0;
612 }
613
614 static int speaker_ioport_read(struct kvm_vcpu *vcpu,
615                                    struct kvm_io_device *this,
616                                    gpa_t addr, int len, void *data)
617 {
618         struct kvm_pit *pit = speaker_to_pit(this);
619         struct kvm_kpit_state *pit_state = &pit->pit_state;
620         struct kvm *kvm = pit->kvm;
621         unsigned int refresh_clock;
622         int ret;
623         if (addr != KVM_SPEAKER_BASE_ADDRESS)
624                 return -EOPNOTSUPP;
625
626         /* Refresh clock toggles at about 15us. We approximate as 2^14ns. */
627         refresh_clock = ((unsigned int)ktime_to_ns(ktime_get()) >> 14) & 1;
628
629         mutex_lock(&pit_state->lock);
630         ret = ((pit_state->speaker_data_on << 1) | pit_get_gate(kvm, 2) |
631                 (pit_get_out(kvm, 2) << 5) | (refresh_clock << 4));
632         if (len > sizeof(ret))
633                 len = sizeof(ret);
634         memcpy(data, (char *)&ret, len);
635         mutex_unlock(&pit_state->lock);
636         return 0;
637 }
638
639 void kvm_pit_reset(struct kvm_pit *pit)
640 {
641         int i;
642         struct kvm_kpit_channel_state *c;
643
644         mutex_lock(&pit->pit_state.lock);
645         pit->pit_state.flags = 0;
646         for (i = 0; i < 3; i++) {
647                 c = &pit->pit_state.channels[i];
648                 c->mode = 0xff;
649                 c->gate = (i != 2);
650                 pit_load_count(pit->kvm, i, 0);
651         }
652         mutex_unlock(&pit->pit_state.lock);
653
654         atomic_set(&pit->pit_state.pending, 0);
655         pit->pit_state.irq_ack = 1;
656 }
657
658 static void pit_mask_notifer(struct kvm_irq_mask_notifier *kimn, bool mask)
659 {
660         struct kvm_pit *pit = container_of(kimn, struct kvm_pit, mask_notifier);
661
662         if (!mask) {
663                 atomic_set(&pit->pit_state.pending, 0);
664                 pit->pit_state.irq_ack = 1;
665         }
666 }
667
668 static const struct kvm_io_device_ops pit_dev_ops = {
669         .read     = pit_ioport_read,
670         .write    = pit_ioport_write,
671 };
672
673 static const struct kvm_io_device_ops speaker_dev_ops = {
674         .read     = speaker_ioport_read,
675         .write    = speaker_ioport_write,
676 };
677
678 /* Caller must hold slots_lock */
679 struct kvm_pit *kvm_create_pit(struct kvm *kvm, u32 flags)
680 {
681         struct kvm_pit *pit;
682         struct kvm_kpit_state *pit_state;
683         struct pid *pid;
684         pid_t pid_nr;
685         int ret;
686
687         pit = kzalloc(sizeof(struct kvm_pit), GFP_KERNEL);
688         if (!pit)
689                 return NULL;
690
691         pit->irq_source_id = kvm_request_irq_source_id(kvm);
692         if (pit->irq_source_id < 0) {
693                 kfree(pit);
694                 return NULL;
695         }
696
697         mutex_init(&pit->pit_state.lock);
698         mutex_lock(&pit->pit_state.lock);
699         spin_lock_init(&pit->pit_state.inject_lock);
700
701         pid = get_pid(task_tgid(current));
702         pid_nr = pid_vnr(pid);
703         put_pid(pid);
704
705         init_kthread_worker(&pit->worker);
706         pit->worker_task = kthread_run(kthread_worker_fn, &pit->worker,
707                                        "kvm-pit/%d", pid_nr);
708         if (IS_ERR(pit->worker_task)) {
709                 mutex_unlock(&pit->pit_state.lock);
710                 kvm_free_irq_source_id(kvm, pit->irq_source_id);
711                 kfree(pit);
712                 return NULL;
713         }
714         init_kthread_work(&pit->expired, pit_do_work);
715
716         kvm->arch.vpit = pit;
717         pit->kvm = kvm;
718
719         pit_state = &pit->pit_state;
720         pit_state->pit = pit;
721         hrtimer_init(&pit_state->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
722         pit_state->irq_ack_notifier.gsi = 0;
723         pit_state->irq_ack_notifier.irq_acked = kvm_pit_ack_irq;
724         kvm_register_irq_ack_notifier(kvm, &pit_state->irq_ack_notifier);
725         pit_state->reinject = true;
726         mutex_unlock(&pit->pit_state.lock);
727
728         kvm_pit_reset(pit);
729
730         pit->mask_notifier.func = pit_mask_notifer;
731         kvm_register_irq_mask_notifier(kvm, 0, &pit->mask_notifier);
732
733         kvm_iodevice_init(&pit->dev, &pit_dev_ops);
734         ret = kvm_io_bus_register_dev(kvm, KVM_PIO_BUS, KVM_PIT_BASE_ADDRESS,
735                                       KVM_PIT_MEM_LENGTH, &pit->dev);
736         if (ret < 0)
737                 goto fail;
738
739         if (flags & KVM_PIT_SPEAKER_DUMMY) {
740                 kvm_iodevice_init(&pit->speaker_dev, &speaker_dev_ops);
741                 ret = kvm_io_bus_register_dev(kvm, KVM_PIO_BUS,
742                                               KVM_SPEAKER_BASE_ADDRESS, 4,
743                                               &pit->speaker_dev);
744                 if (ret < 0)
745                         goto fail_unregister;
746         }
747
748         return pit;
749
750 fail_unregister:
751         kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS, &pit->dev);
752
753 fail:
754         kvm_unregister_irq_mask_notifier(kvm, 0, &pit->mask_notifier);
755         kvm_unregister_irq_ack_notifier(kvm, &pit_state->irq_ack_notifier);
756         kvm_free_irq_source_id(kvm, pit->irq_source_id);
757         kthread_stop(pit->worker_task);
758         kfree(pit);
759         return NULL;
760 }
761
762 void kvm_free_pit(struct kvm *kvm)
763 {
764         struct hrtimer *timer;
765
766         if (kvm->arch.vpit) {
767                 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS, &kvm->arch.vpit->dev);
768                 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
769                                               &kvm->arch.vpit->speaker_dev);
770                 kvm_unregister_irq_mask_notifier(kvm, 0,
771                                                &kvm->arch.vpit->mask_notifier);
772                 kvm_unregister_irq_ack_notifier(kvm,
773                                 &kvm->arch.vpit->pit_state.irq_ack_notifier);
774                 mutex_lock(&kvm->arch.vpit->pit_state.lock);
775                 timer = &kvm->arch.vpit->pit_state.timer;
776                 hrtimer_cancel(timer);
777                 flush_kthread_work(&kvm->arch.vpit->expired);
778                 kthread_stop(kvm->arch.vpit->worker_task);
779                 kvm_free_irq_source_id(kvm, kvm->arch.vpit->irq_source_id);
780                 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
781                 kfree(kvm->arch.vpit);
782         }
783 }