KVM: Add save/restore supporting of in kernel PIT
[linux-2.6.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  *
9  * Permission is hereby granted, free of charge, to any person obtaining a copy
10  * of this software and associated documentation files (the "Software"), to deal
11  * in the Software without restriction, including without limitation the rights
12  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
13  * copies of the Software, and to permit persons to whom the Software is
14  * furnished to do so, subject to the following conditions:
15  *
16  * The above copyright notice and this permission notice shall be included in
17  * all copies or substantial portions of the Software.
18  *
19  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
20  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
21  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
22  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
23  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
24  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
25  * THE SOFTWARE.
26  *
27  * Authors:
28  *   Sheng Yang <sheng.yang@intel.com>
29  *   Based on QEMU and Xen.
30  */
31
32 #include <linux/kvm_host.h>
33
34 #include "irq.h"
35 #include "i8254.h"
36
37 #ifndef CONFIG_X86_64
38 #define mod_64(x, y) ((x) - (y) * div64_64(x, y))
39 #else
40 #define mod_64(x, y) ((x) % (y))
41 #endif
42
43 #define RW_STATE_LSB 1
44 #define RW_STATE_MSB 2
45 #define RW_STATE_WORD0 3
46 #define RW_STATE_WORD1 4
47
48 /* Compute with 96 bit intermediate result: (a*b)/c */
49 static u64 muldiv64(u64 a, u32 b, u32 c)
50 {
51         union {
52                 u64 ll;
53                 struct {
54                         u32 low, high;
55                 } l;
56         } u, res;
57         u64 rl, rh;
58
59         u.ll = a;
60         rl = (u64)u.l.low * (u64)b;
61         rh = (u64)u.l.high * (u64)b;
62         rh += (rl >> 32);
63         res.l.high = div64_64(rh, c);
64         res.l.low = div64_64(((mod_64(rh, c) << 32) + (rl & 0xffffffff)), c);
65         return res.ll;
66 }
67
68 static void pit_set_gate(struct kvm *kvm, int channel, u32 val)
69 {
70         struct kvm_kpit_channel_state *c =
71                 &kvm->arch.vpit->pit_state.channels[channel];
72
73         WARN_ON(!mutex_is_locked(&kvm->arch.vpit->pit_state.lock));
74
75         switch (c->mode) {
76         default:
77         case 0:
78         case 4:
79                 /* XXX: just disable/enable counting */
80                 break;
81         case 1:
82         case 2:
83         case 3:
84         case 5:
85                 /* Restart counting on rising edge. */
86                 if (c->gate < val)
87                         c->count_load_time = ktime_get();
88                 break;
89         }
90
91         c->gate = val;
92 }
93
94 int pit_get_gate(struct kvm *kvm, int channel)
95 {
96         WARN_ON(!mutex_is_locked(&kvm->arch.vpit->pit_state.lock));
97
98         return kvm->arch.vpit->pit_state.channels[channel].gate;
99 }
100
101 static int pit_get_count(struct kvm *kvm, int channel)
102 {
103         struct kvm_kpit_channel_state *c =
104                 &kvm->arch.vpit->pit_state.channels[channel];
105         s64 d, t;
106         int counter;
107
108         WARN_ON(!mutex_is_locked(&kvm->arch.vpit->pit_state.lock));
109
110         t = ktime_to_ns(ktime_sub(ktime_get(), c->count_load_time));
111         d = muldiv64(t, KVM_PIT_FREQ, NSEC_PER_SEC);
112
113         switch (c->mode) {
114         case 0:
115         case 1:
116         case 4:
117         case 5:
118                 counter = (c->count - d) & 0xffff;
119                 break;
120         case 3:
121                 /* XXX: may be incorrect for odd counts */
122                 counter = c->count - (mod_64((2 * d), c->count));
123                 break;
124         default:
125                 counter = c->count - mod_64(d, c->count);
126                 break;
127         }
128         return counter;
129 }
130
131 static int pit_get_out(struct kvm *kvm, int channel)
132 {
133         struct kvm_kpit_channel_state *c =
134                 &kvm->arch.vpit->pit_state.channels[channel];
135         s64 d, t;
136         int out;
137
138         WARN_ON(!mutex_is_locked(&kvm->arch.vpit->pit_state.lock));
139
140         t = ktime_to_ns(ktime_sub(ktime_get(), c->count_load_time));
141         d = muldiv64(t, KVM_PIT_FREQ, NSEC_PER_SEC);
142
143         switch (c->mode) {
144         default:
145         case 0:
146                 out = (d >= c->count);
147                 break;
148         case 1:
149                 out = (d < c->count);
150                 break;
151         case 2:
152                 out = ((mod_64(d, c->count) == 0) && (d != 0));
153                 break;
154         case 3:
155                 out = (mod_64(d, c->count) < ((c->count + 1) >> 1));
156                 break;
157         case 4:
158         case 5:
159                 out = (d == c->count);
160                 break;
161         }
162
163         return out;
164 }
165
166 static void pit_latch_count(struct kvm *kvm, int channel)
167 {
168         struct kvm_kpit_channel_state *c =
169                 &kvm->arch.vpit->pit_state.channels[channel];
170
171         WARN_ON(!mutex_is_locked(&kvm->arch.vpit->pit_state.lock));
172
173         if (!c->count_latched) {
174                 c->latched_count = pit_get_count(kvm, channel);
175                 c->count_latched = c->rw_mode;
176         }
177 }
178
179 static void pit_latch_status(struct kvm *kvm, int channel)
180 {
181         struct kvm_kpit_channel_state *c =
182                 &kvm->arch.vpit->pit_state.channels[channel];
183
184         WARN_ON(!mutex_is_locked(&kvm->arch.vpit->pit_state.lock));
185
186         if (!c->status_latched) {
187                 /* TODO: Return NULL COUNT (bit 6). */
188                 c->status = ((pit_get_out(kvm, channel) << 7) |
189                                 (c->rw_mode << 4) |
190                                 (c->mode << 1) |
191                                 c->bcd);
192                 c->status_latched = 1;
193         }
194 }
195
196 int __pit_timer_fn(struct kvm_kpit_state *ps)
197 {
198         struct kvm_vcpu *vcpu0 = ps->pit->kvm->vcpus[0];
199         struct kvm_kpit_timer *pt = &ps->pit_timer;
200
201         atomic_inc(&pt->pending);
202         smp_mb__after_atomic_inc();
203         /* FIXME: handle case where the guest is in guest mode */
204         if (vcpu0 && waitqueue_active(&vcpu0->wq)) {
205                 vcpu0->arch.mp_state = VCPU_MP_STATE_RUNNABLE;
206                 wake_up_interruptible(&vcpu0->wq);
207         }
208
209         pt->timer.expires = ktime_add_ns(pt->timer.expires, pt->period);
210         pt->scheduled = ktime_to_ns(pt->timer.expires);
211
212         return (pt->period == 0 ? 0 : 1);
213 }
214
215 static enum hrtimer_restart pit_timer_fn(struct hrtimer *data)
216 {
217         struct kvm_kpit_state *ps;
218         int restart_timer = 0;
219
220         ps = container_of(data, struct kvm_kpit_state, pit_timer.timer);
221
222         restart_timer = __pit_timer_fn(ps);
223
224         if (restart_timer)
225                 return HRTIMER_RESTART;
226         else
227                 return HRTIMER_NORESTART;
228 }
229
230 static void destroy_pit_timer(struct kvm_kpit_timer *pt)
231 {
232         pr_debug("pit: execute del timer!\n");
233         hrtimer_cancel(&pt->timer);
234 }
235
236 static void create_pit_timer(struct kvm_kpit_timer *pt, u32 val, int is_period)
237 {
238         s64 interval;
239
240         interval = muldiv64(val, NSEC_PER_SEC, KVM_PIT_FREQ);
241
242         pr_debug("pit: create pit timer, interval is %llu nsec\n", interval);
243
244         /* TODO The new value only affected after the retriggered */
245         hrtimer_cancel(&pt->timer);
246         pt->period = (is_period == 0) ? 0 : interval;
247         pt->timer.function = pit_timer_fn;
248         atomic_set(&pt->pending, 0);
249
250         hrtimer_start(&pt->timer, ktime_add_ns(ktime_get(), interval),
251                       HRTIMER_MODE_ABS);
252 }
253
254 static void pit_load_count(struct kvm *kvm, int channel, u32 val)
255 {
256         struct kvm_kpit_state *ps = &kvm->arch.vpit->pit_state;
257
258         WARN_ON(!mutex_is_locked(&ps->lock));
259
260         pr_debug("pit: load_count val is %d, channel is %d\n", val, channel);
261
262         /*
263          * Though spec said the state of 8254 is undefined after power-up,
264          * seems some tricky OS like Windows XP depends on IRQ0 interrupt
265          * when booting up.
266          * So here setting initialize rate for it, and not a specific number
267          */
268         if (val == 0)
269                 val = 0x10000;
270
271         ps->channels[channel].count_load_time = ktime_get();
272         ps->channels[channel].count = val;
273
274         if (channel != 0)
275                 return;
276
277         /* Two types of timer
278          * mode 1 is one shot, mode 2 is period, otherwise del timer */
279         switch (ps->channels[0].mode) {
280         case 1:
281                 create_pit_timer(&ps->pit_timer, val, 0);
282                 break;
283         case 2:
284                 create_pit_timer(&ps->pit_timer, val, 1);
285                 break;
286         default:
287                 destroy_pit_timer(&ps->pit_timer);
288         }
289 }
290
291 void kvm_pit_load_count(struct kvm *kvm, int channel, u32 val)
292 {
293         mutex_lock(&kvm->arch.vpit->pit_state.lock);
294         pit_load_count(kvm, channel, val);
295         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
296 }
297
298 static void pit_ioport_write(struct kvm_io_device *this,
299                              gpa_t addr, int len, const void *data)
300 {
301         struct kvm_pit *pit = (struct kvm_pit *)this->private;
302         struct kvm_kpit_state *pit_state = &pit->pit_state;
303         struct kvm *kvm = pit->kvm;
304         int channel, access;
305         struct kvm_kpit_channel_state *s;
306         u32 val = *(u32 *) data;
307
308         val  &= 0xff;
309         addr &= KVM_PIT_CHANNEL_MASK;
310
311         mutex_lock(&pit_state->lock);
312
313         if (val != 0)
314                 pr_debug("pit: write addr is 0x%x, len is %d, val is 0x%x\n",
315                           (unsigned int)addr, len, val);
316
317         if (addr == 3) {
318                 channel = val >> 6;
319                 if (channel == 3) {
320                         /* Read-Back Command. */
321                         for (channel = 0; channel < 3; channel++) {
322                                 s = &pit_state->channels[channel];
323                                 if (val & (2 << channel)) {
324                                         if (!(val & 0x20))
325                                                 pit_latch_count(kvm, channel);
326                                         if (!(val & 0x10))
327                                                 pit_latch_status(kvm, channel);
328                                 }
329                         }
330                 } else {
331                         /* Select Counter <channel>. */
332                         s = &pit_state->channels[channel];
333                         access = (val >> 4) & KVM_PIT_CHANNEL_MASK;
334                         if (access == 0) {
335                                 pit_latch_count(kvm, channel);
336                         } else {
337                                 s->rw_mode = access;
338                                 s->read_state = access;
339                                 s->write_state = access;
340                                 s->mode = (val >> 1) & 7;
341                                 if (s->mode > 5)
342                                         s->mode -= 4;
343                                 s->bcd = val & 1;
344                         }
345                 }
346         } else {
347                 /* Write Count. */
348                 s = &pit_state->channels[addr];
349                 switch (s->write_state) {
350                 default:
351                 case RW_STATE_LSB:
352                         pit_load_count(kvm, addr, val);
353                         break;
354                 case RW_STATE_MSB:
355                         pit_load_count(kvm, addr, val << 8);
356                         break;
357                 case RW_STATE_WORD0:
358                         s->write_latch = val;
359                         s->write_state = RW_STATE_WORD1;
360                         break;
361                 case RW_STATE_WORD1:
362                         pit_load_count(kvm, addr, s->write_latch | (val << 8));
363                         s->write_state = RW_STATE_WORD0;
364                         break;
365                 }
366         }
367
368         mutex_unlock(&pit_state->lock);
369 }
370
371 static void pit_ioport_read(struct kvm_io_device *this,
372                             gpa_t addr, int len, void *data)
373 {
374         struct kvm_pit *pit = (struct kvm_pit *)this->private;
375         struct kvm_kpit_state *pit_state = &pit->pit_state;
376         struct kvm *kvm = pit->kvm;
377         int ret, count;
378         struct kvm_kpit_channel_state *s;
379
380         addr &= KVM_PIT_CHANNEL_MASK;
381         s = &pit_state->channels[addr];
382
383         mutex_lock(&pit_state->lock);
384
385         if (s->status_latched) {
386                 s->status_latched = 0;
387                 ret = s->status;
388         } else if (s->count_latched) {
389                 switch (s->count_latched) {
390                 default:
391                 case RW_STATE_LSB:
392                         ret = s->latched_count & 0xff;
393                         s->count_latched = 0;
394                         break;
395                 case RW_STATE_MSB:
396                         ret = s->latched_count >> 8;
397                         s->count_latched = 0;
398                         break;
399                 case RW_STATE_WORD0:
400                         ret = s->latched_count & 0xff;
401                         s->count_latched = RW_STATE_MSB;
402                         break;
403                 }
404         } else {
405                 switch (s->read_state) {
406                 default:
407                 case RW_STATE_LSB:
408                         count = pit_get_count(kvm, addr);
409                         ret = count & 0xff;
410                         break;
411                 case RW_STATE_MSB:
412                         count = pit_get_count(kvm, addr);
413                         ret = (count >> 8) & 0xff;
414                         break;
415                 case RW_STATE_WORD0:
416                         count = pit_get_count(kvm, addr);
417                         ret = count & 0xff;
418                         s->read_state = RW_STATE_WORD1;
419                         break;
420                 case RW_STATE_WORD1:
421                         count = pit_get_count(kvm, addr);
422                         ret = (count >> 8) & 0xff;
423                         s->read_state = RW_STATE_WORD0;
424                         break;
425                 }
426         }
427
428         if (len > sizeof(ret))
429                 len = sizeof(ret);
430         memcpy(data, (char *)&ret, len);
431
432         mutex_unlock(&pit_state->lock);
433 }
434
435 static int pit_in_range(struct kvm_io_device *this, gpa_t addr)
436 {
437         return ((addr >= KVM_PIT_BASE_ADDRESS) &&
438                 (addr < KVM_PIT_BASE_ADDRESS + KVM_PIT_MEM_LENGTH));
439 }
440
441 static void speaker_ioport_write(struct kvm_io_device *this,
442                                  gpa_t addr, int len, const void *data)
443 {
444         struct kvm_pit *pit = (struct kvm_pit *)this->private;
445         struct kvm_kpit_state *pit_state = &pit->pit_state;
446         struct kvm *kvm = pit->kvm;
447         u32 val = *(u32 *) data;
448
449         mutex_lock(&pit_state->lock);
450         pit_state->speaker_data_on = (val >> 1) & 1;
451         pit_set_gate(kvm, 2, val & 1);
452         mutex_unlock(&pit_state->lock);
453 }
454
455 static void speaker_ioport_read(struct kvm_io_device *this,
456                                 gpa_t addr, int len, void *data)
457 {
458         struct kvm_pit *pit = (struct kvm_pit *)this->private;
459         struct kvm_kpit_state *pit_state = &pit->pit_state;
460         struct kvm *kvm = pit->kvm;
461         unsigned int refresh_clock;
462         int ret;
463
464         /* Refresh clock toggles at about 15us. We approximate as 2^14ns. */
465         refresh_clock = ((unsigned int)ktime_to_ns(ktime_get()) >> 14) & 1;
466
467         mutex_lock(&pit_state->lock);
468         ret = ((pit_state->speaker_data_on << 1) | pit_get_gate(kvm, 2) |
469                 (pit_get_out(kvm, 2) << 5) | (refresh_clock << 4));
470         if (len > sizeof(ret))
471                 len = sizeof(ret);
472         memcpy(data, (char *)&ret, len);
473         mutex_unlock(&pit_state->lock);
474 }
475
476 static int speaker_in_range(struct kvm_io_device *this, gpa_t addr)
477 {
478         return (addr == KVM_SPEAKER_BASE_ADDRESS);
479 }
480
481 struct kvm_pit *kvm_create_pit(struct kvm *kvm)
482 {
483         int i;
484         struct kvm_pit *pit;
485         struct kvm_kpit_state *pit_state;
486         struct kvm_kpit_channel_state *c;
487
488         pit = kzalloc(sizeof(struct kvm_pit), GFP_KERNEL);
489         if (!pit)
490                 return NULL;
491
492         mutex_init(&pit->pit_state.lock);
493         mutex_lock(&pit->pit_state.lock);
494
495         /* Initialize PIO device */
496         pit->dev.read = pit_ioport_read;
497         pit->dev.write = pit_ioport_write;
498         pit->dev.in_range = pit_in_range;
499         pit->dev.private = pit;
500         kvm_io_bus_register_dev(&kvm->pio_bus, &pit->dev);
501
502         pit->speaker_dev.read = speaker_ioport_read;
503         pit->speaker_dev.write = speaker_ioport_write;
504         pit->speaker_dev.in_range = speaker_in_range;
505         pit->speaker_dev.private = pit;
506         kvm_io_bus_register_dev(&kvm->pio_bus, &pit->speaker_dev);
507
508         kvm->arch.vpit = pit;
509         pit->kvm = kvm;
510
511         pit_state = &pit->pit_state;
512         pit_state->pit = pit;
513         hrtimer_init(&pit_state->pit_timer.timer,
514                      CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
515         atomic_set(&pit_state->pit_timer.pending, 0);
516         for (i = 0; i < 3; i++) {
517                 c = &pit_state->channels[i];
518                 c->mode = 0xff;
519                 c->gate = (i != 2);
520                 pit_load_count(kvm, i, 0);
521         }
522
523         mutex_unlock(&pit->pit_state.lock);
524
525         pit->pit_state.inject_pending = 1;
526
527         return pit;
528 }
529
530 void kvm_free_pit(struct kvm *kvm)
531 {
532         struct hrtimer *timer;
533
534         if (kvm->arch.vpit) {
535                 mutex_lock(&kvm->arch.vpit->pit_state.lock);
536                 timer = &kvm->arch.vpit->pit_state.pit_timer.timer;
537                 hrtimer_cancel(timer);
538                 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
539                 kfree(kvm->arch.vpit);
540         }
541 }
542
543 void __inject_pit_timer_intr(struct kvm *kvm)
544 {
545         mutex_lock(&kvm->lock);
546         kvm_ioapic_set_irq(kvm->arch.vioapic, 0, 1);
547         kvm_ioapic_set_irq(kvm->arch.vioapic, 0, 0);
548         kvm_pic_set_irq(pic_irqchip(kvm), 0, 1);
549         kvm_pic_set_irq(pic_irqchip(kvm), 0, 0);
550         mutex_unlock(&kvm->lock);
551 }
552
553 void kvm_inject_pit_timer_irqs(struct kvm_vcpu *vcpu)
554 {
555         struct kvm_pit *pit = vcpu->kvm->arch.vpit;
556         struct kvm *kvm = vcpu->kvm;
557         struct kvm_kpit_state *ps;
558
559         if (vcpu && pit) {
560                 ps = &pit->pit_state;
561
562                 /* Try to inject pending interrupts when:
563                  * 1. Pending exists
564                  * 2. Last interrupt was accepted or waited for too long time*/
565                 if (atomic_read(&ps->pit_timer.pending) &&
566                     (ps->inject_pending ||
567                     (jiffies - ps->last_injected_time
568                                 >= KVM_MAX_PIT_INTR_INTERVAL))) {
569                         ps->inject_pending = 0;
570                         __inject_pit_timer_intr(kvm);
571                         ps->last_injected_time = jiffies;
572                 }
573         }
574 }
575
576 void kvm_pit_timer_intr_post(struct kvm_vcpu *vcpu, int vec)
577 {
578         struct kvm_arch *arch = &vcpu->kvm->arch;
579         struct kvm_kpit_state *ps;
580
581         if (vcpu && arch->vpit) {
582                 ps = &arch->vpit->pit_state;
583                 if (atomic_read(&ps->pit_timer.pending) &&
584                 (((arch->vpic->pics[0].imr & 1) == 0 &&
585                   arch->vpic->pics[0].irq_base == vec) ||
586                   (arch->vioapic->redirtbl[0].fields.vector == vec &&
587                   arch->vioapic->redirtbl[0].fields.mask != 1))) {
588                         ps->inject_pending = 1;
589                         atomic_dec(&ps->pit_timer.pending);
590                         ps->channels[0].count_load_time = ktime_get();
591                 }
592         }
593 }