KVM: Add irq ack notifier list
[linux-2.6.git] / arch / x86 / kvm / lapic.c
1
2 /*
3  * Local APIC virtualization
4  *
5  * Copyright (C) 2006 Qumranet, Inc.
6  * Copyright (C) 2007 Novell
7  * Copyright (C) 2007 Intel
8  *
9  * Authors:
10  *   Dor Laor <dor.laor@qumranet.com>
11  *   Gregory Haskins <ghaskins@novell.com>
12  *   Yaozu (Eddie) Dong <eddie.dong@intel.com>
13  *
14  * Based on Xen 3.1 code, Copyright (c) 2004, Intel Corporation.
15  *
16  * This work is licensed under the terms of the GNU GPL, version 2.  See
17  * the COPYING file in the top-level directory.
18  */
19
20 #include <linux/kvm_host.h>
21 #include <linux/kvm.h>
22 #include <linux/mm.h>
23 #include <linux/highmem.h>
24 #include <linux/smp.h>
25 #include <linux/hrtimer.h>
26 #include <linux/io.h>
27 #include <linux/module.h>
28 #include <linux/math64.h>
29 #include <asm/processor.h>
30 #include <asm/msr.h>
31 #include <asm/page.h>
32 #include <asm/current.h>
33 #include <asm/apicdef.h>
34 #include <asm/atomic.h>
35 #include "kvm_cache_regs.h"
36 #include "irq.h"
37
38 #define PRId64 "d"
39 #define PRIx64 "llx"
40 #define PRIu64 "u"
41 #define PRIo64 "o"
42
43 #define APIC_BUS_CYCLE_NS 1
44
45 /* #define apic_debug(fmt,arg...) printk(KERN_WARNING fmt,##arg) */
46 #define apic_debug(fmt, arg...)
47
48 #define APIC_LVT_NUM                    6
49 /* 14 is the version for Xeon and Pentium 8.4.8*/
50 #define APIC_VERSION                    (0x14UL | ((APIC_LVT_NUM - 1) << 16))
51 #define LAPIC_MMIO_LENGTH               (1 << 12)
52 /* followed define is not in apicdef.h */
53 #define APIC_SHORT_MASK                 0xc0000
54 #define APIC_DEST_NOSHORT               0x0
55 #define APIC_DEST_MASK                  0x800
56 #define MAX_APIC_VECTOR                 256
57
58 #define VEC_POS(v) ((v) & (32 - 1))
59 #define REG_POS(v) (((v) >> 5) << 4)
60
61 static inline u32 apic_get_reg(struct kvm_lapic *apic, int reg_off)
62 {
63         return *((u32 *) (apic->regs + reg_off));
64 }
65
66 static inline void apic_set_reg(struct kvm_lapic *apic, int reg_off, u32 val)
67 {
68         *((u32 *) (apic->regs + reg_off)) = val;
69 }
70
71 static inline int apic_test_and_set_vector(int vec, void *bitmap)
72 {
73         return test_and_set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
74 }
75
76 static inline int apic_test_and_clear_vector(int vec, void *bitmap)
77 {
78         return test_and_clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
79 }
80
81 static inline void apic_set_vector(int vec, void *bitmap)
82 {
83         set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
84 }
85
86 static inline void apic_clear_vector(int vec, void *bitmap)
87 {
88         clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
89 }
90
91 static inline int apic_hw_enabled(struct kvm_lapic *apic)
92 {
93         return (apic)->vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE;
94 }
95
96 static inline int  apic_sw_enabled(struct kvm_lapic *apic)
97 {
98         return apic_get_reg(apic, APIC_SPIV) & APIC_SPIV_APIC_ENABLED;
99 }
100
101 static inline int apic_enabled(struct kvm_lapic *apic)
102 {
103         return apic_sw_enabled(apic) && apic_hw_enabled(apic);
104 }
105
106 #define LVT_MASK        \
107         (APIC_LVT_MASKED | APIC_SEND_PENDING | APIC_VECTOR_MASK)
108
109 #define LINT_MASK       \
110         (LVT_MASK | APIC_MODE_MASK | APIC_INPUT_POLARITY | \
111          APIC_LVT_REMOTE_IRR | APIC_LVT_LEVEL_TRIGGER)
112
113 static inline int kvm_apic_id(struct kvm_lapic *apic)
114 {
115         return (apic_get_reg(apic, APIC_ID) >> 24) & 0xff;
116 }
117
118 static inline int apic_lvt_enabled(struct kvm_lapic *apic, int lvt_type)
119 {
120         return !(apic_get_reg(apic, lvt_type) & APIC_LVT_MASKED);
121 }
122
123 static inline int apic_lvt_vector(struct kvm_lapic *apic, int lvt_type)
124 {
125         return apic_get_reg(apic, lvt_type) & APIC_VECTOR_MASK;
126 }
127
128 static inline int apic_lvtt_period(struct kvm_lapic *apic)
129 {
130         return apic_get_reg(apic, APIC_LVTT) & APIC_LVT_TIMER_PERIODIC;
131 }
132
133 static unsigned int apic_lvt_mask[APIC_LVT_NUM] = {
134         LVT_MASK | APIC_LVT_TIMER_PERIODIC,     /* LVTT */
135         LVT_MASK | APIC_MODE_MASK,      /* LVTTHMR */
136         LVT_MASK | APIC_MODE_MASK,      /* LVTPC */
137         LINT_MASK, LINT_MASK,   /* LVT0-1 */
138         LVT_MASK                /* LVTERR */
139 };
140
141 static int find_highest_vector(void *bitmap)
142 {
143         u32 *word = bitmap;
144         int word_offset = MAX_APIC_VECTOR >> 5;
145
146         while ((word_offset != 0) && (word[(--word_offset) << 2] == 0))
147                 continue;
148
149         if (likely(!word_offset && !word[0]))
150                 return -1;
151         else
152                 return fls(word[word_offset << 2]) - 1 + (word_offset << 5);
153 }
154
155 static inline int apic_test_and_set_irr(int vec, struct kvm_lapic *apic)
156 {
157         return apic_test_and_set_vector(vec, apic->regs + APIC_IRR);
158 }
159
160 static inline void apic_clear_irr(int vec, struct kvm_lapic *apic)
161 {
162         apic_clear_vector(vec, apic->regs + APIC_IRR);
163 }
164
165 static inline int apic_find_highest_irr(struct kvm_lapic *apic)
166 {
167         int result;
168
169         result = find_highest_vector(apic->regs + APIC_IRR);
170         ASSERT(result == -1 || result >= 16);
171
172         return result;
173 }
174
175 int kvm_lapic_find_highest_irr(struct kvm_vcpu *vcpu)
176 {
177         struct kvm_lapic *apic = vcpu->arch.apic;
178         int highest_irr;
179
180         if (!apic)
181                 return 0;
182         highest_irr = apic_find_highest_irr(apic);
183
184         return highest_irr;
185 }
186 EXPORT_SYMBOL_GPL(kvm_lapic_find_highest_irr);
187
188 int kvm_apic_set_irq(struct kvm_vcpu *vcpu, u8 vec, u8 trig)
189 {
190         struct kvm_lapic *apic = vcpu->arch.apic;
191
192         if (!apic_test_and_set_irr(vec, apic)) {
193                 /* a new pending irq is set in IRR */
194                 if (trig)
195                         apic_set_vector(vec, apic->regs + APIC_TMR);
196                 else
197                         apic_clear_vector(vec, apic->regs + APIC_TMR);
198                 kvm_vcpu_kick(apic->vcpu);
199                 return 1;
200         }
201         return 0;
202 }
203
204 static inline int apic_find_highest_isr(struct kvm_lapic *apic)
205 {
206         int result;
207
208         result = find_highest_vector(apic->regs + APIC_ISR);
209         ASSERT(result == -1 || result >= 16);
210
211         return result;
212 }
213
214 static void apic_update_ppr(struct kvm_lapic *apic)
215 {
216         u32 tpr, isrv, ppr;
217         int isr;
218
219         tpr = apic_get_reg(apic, APIC_TASKPRI);
220         isr = apic_find_highest_isr(apic);
221         isrv = (isr != -1) ? isr : 0;
222
223         if ((tpr & 0xf0) >= (isrv & 0xf0))
224                 ppr = tpr & 0xff;
225         else
226                 ppr = isrv & 0xf0;
227
228         apic_debug("vlapic %p, ppr 0x%x, isr 0x%x, isrv 0x%x",
229                    apic, ppr, isr, isrv);
230
231         apic_set_reg(apic, APIC_PROCPRI, ppr);
232 }
233
234 static void apic_set_tpr(struct kvm_lapic *apic, u32 tpr)
235 {
236         apic_set_reg(apic, APIC_TASKPRI, tpr);
237         apic_update_ppr(apic);
238 }
239
240 int kvm_apic_match_physical_addr(struct kvm_lapic *apic, u16 dest)
241 {
242         return kvm_apic_id(apic) == dest;
243 }
244
245 int kvm_apic_match_logical_addr(struct kvm_lapic *apic, u8 mda)
246 {
247         int result = 0;
248         u8 logical_id;
249
250         logical_id = GET_APIC_LOGICAL_ID(apic_get_reg(apic, APIC_LDR));
251
252         switch (apic_get_reg(apic, APIC_DFR)) {
253         case APIC_DFR_FLAT:
254                 if (logical_id & mda)
255                         result = 1;
256                 break;
257         case APIC_DFR_CLUSTER:
258                 if (((logical_id >> 4) == (mda >> 0x4))
259                     && (logical_id & mda & 0xf))
260                         result = 1;
261                 break;
262         default:
263                 printk(KERN_WARNING "Bad DFR vcpu %d: %08x\n",
264                        apic->vcpu->vcpu_id, apic_get_reg(apic, APIC_DFR));
265                 break;
266         }
267
268         return result;
269 }
270
271 static int apic_match_dest(struct kvm_vcpu *vcpu, struct kvm_lapic *source,
272                            int short_hand, int dest, int dest_mode)
273 {
274         int result = 0;
275         struct kvm_lapic *target = vcpu->arch.apic;
276
277         apic_debug("target %p, source %p, dest 0x%x, "
278                    "dest_mode 0x%x, short_hand 0x%x",
279                    target, source, dest, dest_mode, short_hand);
280
281         ASSERT(!target);
282         switch (short_hand) {
283         case APIC_DEST_NOSHORT:
284                 if (dest_mode == 0) {
285                         /* Physical mode. */
286                         if ((dest == 0xFF) || (dest == kvm_apic_id(target)))
287                                 result = 1;
288                 } else
289                         /* Logical mode. */
290                         result = kvm_apic_match_logical_addr(target, dest);
291                 break;
292         case APIC_DEST_SELF:
293                 if (target == source)
294                         result = 1;
295                 break;
296         case APIC_DEST_ALLINC:
297                 result = 1;
298                 break;
299         case APIC_DEST_ALLBUT:
300                 if (target != source)
301                         result = 1;
302                 break;
303         default:
304                 printk(KERN_WARNING "Bad dest shorthand value %x\n",
305                        short_hand);
306                 break;
307         }
308
309         return result;
310 }
311
312 /*
313  * Add a pending IRQ into lapic.
314  * Return 1 if successfully added and 0 if discarded.
315  */
316 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
317                              int vector, int level, int trig_mode)
318 {
319         int orig_irr, result = 0;
320         struct kvm_vcpu *vcpu = apic->vcpu;
321
322         switch (delivery_mode) {
323         case APIC_DM_FIXED:
324         case APIC_DM_LOWEST:
325                 /* FIXME add logic for vcpu on reset */
326                 if (unlikely(!apic_enabled(apic)))
327                         break;
328
329                 orig_irr = apic_test_and_set_irr(vector, apic);
330                 if (orig_irr && trig_mode) {
331                         apic_debug("level trig mode repeatedly for vector %d",
332                                    vector);
333                         break;
334                 }
335
336                 if (trig_mode) {
337                         apic_debug("level trig mode for vector %d", vector);
338                         apic_set_vector(vector, apic->regs + APIC_TMR);
339                 } else
340                         apic_clear_vector(vector, apic->regs + APIC_TMR);
341
342                 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE)
343                         kvm_vcpu_kick(vcpu);
344                 else if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED) {
345                         vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
346                         if (waitqueue_active(&vcpu->wq))
347                                 wake_up_interruptible(&vcpu->wq);
348                 }
349
350                 result = (orig_irr == 0);
351                 break;
352
353         case APIC_DM_REMRD:
354                 printk(KERN_DEBUG "Ignoring delivery mode 3\n");
355                 break;
356
357         case APIC_DM_SMI:
358                 printk(KERN_DEBUG "Ignoring guest SMI\n");
359                 break;
360
361         case APIC_DM_NMI:
362                 kvm_inject_nmi(vcpu);
363                 break;
364
365         case APIC_DM_INIT:
366                 if (level) {
367                         if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE)
368                                 printk(KERN_DEBUG
369                                        "INIT on a runnable vcpu %d\n",
370                                        vcpu->vcpu_id);
371                         vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
372                         kvm_vcpu_kick(vcpu);
373                 } else {
374                         printk(KERN_DEBUG
375                                "Ignoring de-assert INIT to vcpu %d\n",
376                                vcpu->vcpu_id);
377                 }
378
379                 break;
380
381         case APIC_DM_STARTUP:
382                 printk(KERN_DEBUG "SIPI to vcpu %d vector 0x%02x\n",
383                        vcpu->vcpu_id, vector);
384                 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
385                         vcpu->arch.sipi_vector = vector;
386                         vcpu->arch.mp_state = KVM_MP_STATE_SIPI_RECEIVED;
387                         if (waitqueue_active(&vcpu->wq))
388                                 wake_up_interruptible(&vcpu->wq);
389                 }
390                 break;
391
392         default:
393                 printk(KERN_ERR "TODO: unsupported delivery mode %x\n",
394                        delivery_mode);
395                 break;
396         }
397         return result;
398 }
399
400 static struct kvm_lapic *kvm_apic_round_robin(struct kvm *kvm, u8 vector,
401                                        unsigned long bitmap)
402 {
403         int last;
404         int next;
405         struct kvm_lapic *apic = NULL;
406
407         last = kvm->arch.round_robin_prev_vcpu;
408         next = last;
409
410         do {
411                 if (++next == KVM_MAX_VCPUS)
412                         next = 0;
413                 if (kvm->vcpus[next] == NULL || !test_bit(next, &bitmap))
414                         continue;
415                 apic = kvm->vcpus[next]->arch.apic;
416                 if (apic && apic_enabled(apic))
417                         break;
418                 apic = NULL;
419         } while (next != last);
420         kvm->arch.round_robin_prev_vcpu = next;
421
422         if (!apic)
423                 printk(KERN_DEBUG "vcpu not ready for apic_round_robin\n");
424
425         return apic;
426 }
427
428 struct kvm_vcpu *kvm_get_lowest_prio_vcpu(struct kvm *kvm, u8 vector,
429                 unsigned long bitmap)
430 {
431         struct kvm_lapic *apic;
432
433         apic = kvm_apic_round_robin(kvm, vector, bitmap);
434         if (apic)
435                 return apic->vcpu;
436         return NULL;
437 }
438
439 static void apic_set_eoi(struct kvm_lapic *apic)
440 {
441         int vector = apic_find_highest_isr(apic);
442
443         /*
444          * Not every write EOI will has corresponding ISR,
445          * one example is when Kernel check timer on setup_IO_APIC
446          */
447         if (vector == -1)
448                 return;
449
450         apic_clear_vector(vector, apic->regs + APIC_ISR);
451         apic_update_ppr(apic);
452
453         if (apic_test_and_clear_vector(vector, apic->regs + APIC_TMR))
454                 kvm_ioapic_update_eoi(apic->vcpu->kvm, vector);
455 }
456
457 static void apic_send_ipi(struct kvm_lapic *apic)
458 {
459         u32 icr_low = apic_get_reg(apic, APIC_ICR);
460         u32 icr_high = apic_get_reg(apic, APIC_ICR2);
461
462         unsigned int dest = GET_APIC_DEST_FIELD(icr_high);
463         unsigned int short_hand = icr_low & APIC_SHORT_MASK;
464         unsigned int trig_mode = icr_low & APIC_INT_LEVELTRIG;
465         unsigned int level = icr_low & APIC_INT_ASSERT;
466         unsigned int dest_mode = icr_low & APIC_DEST_MASK;
467         unsigned int delivery_mode = icr_low & APIC_MODE_MASK;
468         unsigned int vector = icr_low & APIC_VECTOR_MASK;
469
470         struct kvm_vcpu *target;
471         struct kvm_vcpu *vcpu;
472         unsigned long lpr_map = 0;
473         int i;
474
475         apic_debug("icr_high 0x%x, icr_low 0x%x, "
476                    "short_hand 0x%x, dest 0x%x, trig_mode 0x%x, level 0x%x, "
477                    "dest_mode 0x%x, delivery_mode 0x%x, vector 0x%x\n",
478                    icr_high, icr_low, short_hand, dest,
479                    trig_mode, level, dest_mode, delivery_mode, vector);
480
481         for (i = 0; i < KVM_MAX_VCPUS; i++) {
482                 vcpu = apic->vcpu->kvm->vcpus[i];
483                 if (!vcpu)
484                         continue;
485
486                 if (vcpu->arch.apic &&
487                     apic_match_dest(vcpu, apic, short_hand, dest, dest_mode)) {
488                         if (delivery_mode == APIC_DM_LOWEST)
489                                 set_bit(vcpu->vcpu_id, &lpr_map);
490                         else
491                                 __apic_accept_irq(vcpu->arch.apic, delivery_mode,
492                                                   vector, level, trig_mode);
493                 }
494         }
495
496         if (delivery_mode == APIC_DM_LOWEST) {
497                 target = kvm_get_lowest_prio_vcpu(vcpu->kvm, vector, lpr_map);
498                 if (target != NULL)
499                         __apic_accept_irq(target->arch.apic, delivery_mode,
500                                           vector, level, trig_mode);
501         }
502 }
503
504 static u32 apic_get_tmcct(struct kvm_lapic *apic)
505 {
506         u64 counter_passed;
507         ktime_t passed, now;
508         u32 tmcct;
509
510         ASSERT(apic != NULL);
511
512         now = apic->timer.dev.base->get_time();
513         tmcct = apic_get_reg(apic, APIC_TMICT);
514
515         /* if initial count is 0, current count should also be 0 */
516         if (tmcct == 0)
517                 return 0;
518
519         if (unlikely(ktime_to_ns(now) <=
520                 ktime_to_ns(apic->timer.last_update))) {
521                 /* Wrap around */
522                 passed = ktime_add(( {
523                                     (ktime_t) {
524                                     .tv64 = KTIME_MAX -
525                                     (apic->timer.last_update).tv64}; }
526                                    ), now);
527                 apic_debug("time elapsed\n");
528         } else
529                 passed = ktime_sub(now, apic->timer.last_update);
530
531         counter_passed = div64_u64(ktime_to_ns(passed),
532                                    (APIC_BUS_CYCLE_NS * apic->timer.divide_count));
533
534         if (counter_passed > tmcct) {
535                 if (unlikely(!apic_lvtt_period(apic))) {
536                         /* one-shot timers stick at 0 until reset */
537                         tmcct = 0;
538                 } else {
539                         /*
540                          * periodic timers reset to APIC_TMICT when they
541                          * hit 0. The while loop simulates this happening N
542                          * times. (counter_passed %= tmcct) would also work,
543                          * but might be slower or not work on 32-bit??
544                          */
545                         while (counter_passed > tmcct)
546                                 counter_passed -= tmcct;
547                         tmcct -= counter_passed;
548                 }
549         } else {
550                 tmcct -= counter_passed;
551         }
552
553         return tmcct;
554 }
555
556 static void __report_tpr_access(struct kvm_lapic *apic, bool write)
557 {
558         struct kvm_vcpu *vcpu = apic->vcpu;
559         struct kvm_run *run = vcpu->run;
560
561         set_bit(KVM_REQ_REPORT_TPR_ACCESS, &vcpu->requests);
562         run->tpr_access.rip = kvm_rip_read(vcpu);
563         run->tpr_access.is_write = write;
564 }
565
566 static inline void report_tpr_access(struct kvm_lapic *apic, bool write)
567 {
568         if (apic->vcpu->arch.tpr_access_reporting)
569                 __report_tpr_access(apic, write);
570 }
571
572 static u32 __apic_read(struct kvm_lapic *apic, unsigned int offset)
573 {
574         u32 val = 0;
575
576         KVMTRACE_1D(APIC_ACCESS, apic->vcpu, (u32)offset, handler);
577
578         if (offset >= LAPIC_MMIO_LENGTH)
579                 return 0;
580
581         switch (offset) {
582         case APIC_ARBPRI:
583                 printk(KERN_WARNING "Access APIC ARBPRI register "
584                        "which is for P6\n");
585                 break;
586
587         case APIC_TMCCT:        /* Timer CCR */
588                 val = apic_get_tmcct(apic);
589                 break;
590
591         case APIC_TASKPRI:
592                 report_tpr_access(apic, false);
593                 /* fall thru */
594         default:
595                 apic_update_ppr(apic);
596                 val = apic_get_reg(apic, offset);
597                 break;
598         }
599
600         return val;
601 }
602
603 static void apic_mmio_read(struct kvm_io_device *this,
604                            gpa_t address, int len, void *data)
605 {
606         struct kvm_lapic *apic = (struct kvm_lapic *)this->private;
607         unsigned int offset = address - apic->base_address;
608         unsigned char alignment = offset & 0xf;
609         u32 result;
610
611         if ((alignment + len) > 4) {
612                 printk(KERN_ERR "KVM_APIC_READ: alignment error %lx %d",
613                        (unsigned long)address, len);
614                 return;
615         }
616         result = __apic_read(apic, offset & ~0xf);
617
618         switch (len) {
619         case 1:
620         case 2:
621         case 4:
622                 memcpy(data, (char *)&result + alignment, len);
623                 break;
624         default:
625                 printk(KERN_ERR "Local APIC read with len = %x, "
626                        "should be 1,2, or 4 instead\n", len);
627                 break;
628         }
629 }
630
631 static void update_divide_count(struct kvm_lapic *apic)
632 {
633         u32 tmp1, tmp2, tdcr;
634
635         tdcr = apic_get_reg(apic, APIC_TDCR);
636         tmp1 = tdcr & 0xf;
637         tmp2 = ((tmp1 & 0x3) | ((tmp1 & 0x8) >> 1)) + 1;
638         apic->timer.divide_count = 0x1 << (tmp2 & 0x7);
639
640         apic_debug("timer divide count is 0x%x\n",
641                                    apic->timer.divide_count);
642 }
643
644 static void start_apic_timer(struct kvm_lapic *apic)
645 {
646         ktime_t now = apic->timer.dev.base->get_time();
647
648         apic->timer.last_update = now;
649
650         apic->timer.period = apic_get_reg(apic, APIC_TMICT) *
651                     APIC_BUS_CYCLE_NS * apic->timer.divide_count;
652         atomic_set(&apic->timer.pending, 0);
653
654         if (!apic->timer.period)
655                 return;
656
657         hrtimer_start(&apic->timer.dev,
658                       ktime_add_ns(now, apic->timer.period),
659                       HRTIMER_MODE_ABS);
660
661         apic_debug("%s: bus cycle is %" PRId64 "ns, now 0x%016"
662                            PRIx64 ", "
663                            "timer initial count 0x%x, period %lldns, "
664                            "expire @ 0x%016" PRIx64 ".\n", __func__,
665                            APIC_BUS_CYCLE_NS, ktime_to_ns(now),
666                            apic_get_reg(apic, APIC_TMICT),
667                            apic->timer.period,
668                            ktime_to_ns(ktime_add_ns(now,
669                                         apic->timer.period)));
670 }
671
672 static void apic_mmio_write(struct kvm_io_device *this,
673                             gpa_t address, int len, const void *data)
674 {
675         struct kvm_lapic *apic = (struct kvm_lapic *)this->private;
676         unsigned int offset = address - apic->base_address;
677         unsigned char alignment = offset & 0xf;
678         u32 val;
679
680         /*
681          * APIC register must be aligned on 128-bits boundary.
682          * 32/64/128 bits registers must be accessed thru 32 bits.
683          * Refer SDM 8.4.1
684          */
685         if (len != 4 || alignment) {
686                 if (printk_ratelimit())
687                         printk(KERN_ERR "apic write: bad size=%d %lx\n",
688                                len, (long)address);
689                 return;
690         }
691
692         val = *(u32 *) data;
693
694         /* too common printing */
695         if (offset != APIC_EOI)
696                 apic_debug("%s: offset 0x%x with length 0x%x, and value is "
697                            "0x%x\n", __func__, offset, len, val);
698
699         offset &= 0xff0;
700
701         KVMTRACE_1D(APIC_ACCESS, apic->vcpu, (u32)offset, handler);
702
703         switch (offset) {
704         case APIC_ID:           /* Local APIC ID */
705                 apic_set_reg(apic, APIC_ID, val);
706                 break;
707
708         case APIC_TASKPRI:
709                 report_tpr_access(apic, true);
710                 apic_set_tpr(apic, val & 0xff);
711                 break;
712
713         case APIC_EOI:
714                 apic_set_eoi(apic);
715                 break;
716
717         case APIC_LDR:
718                 apic_set_reg(apic, APIC_LDR, val & APIC_LDR_MASK);
719                 break;
720
721         case APIC_DFR:
722                 apic_set_reg(apic, APIC_DFR, val | 0x0FFFFFFF);
723                 break;
724
725         case APIC_SPIV:
726                 apic_set_reg(apic, APIC_SPIV, val & 0x3ff);
727                 if (!(val & APIC_SPIV_APIC_ENABLED)) {
728                         int i;
729                         u32 lvt_val;
730
731                         for (i = 0; i < APIC_LVT_NUM; i++) {
732                                 lvt_val = apic_get_reg(apic,
733                                                        APIC_LVTT + 0x10 * i);
734                                 apic_set_reg(apic, APIC_LVTT + 0x10 * i,
735                                              lvt_val | APIC_LVT_MASKED);
736                         }
737                         atomic_set(&apic->timer.pending, 0);
738
739                 }
740                 break;
741
742         case APIC_ICR:
743                 /* No delay here, so we always clear the pending bit */
744                 apic_set_reg(apic, APIC_ICR, val & ~(1 << 12));
745                 apic_send_ipi(apic);
746                 break;
747
748         case APIC_ICR2:
749                 apic_set_reg(apic, APIC_ICR2, val & 0xff000000);
750                 break;
751
752         case APIC_LVTT:
753         case APIC_LVTTHMR:
754         case APIC_LVTPC:
755         case APIC_LVT0:
756         case APIC_LVT1:
757         case APIC_LVTERR:
758                 /* TODO: Check vector */
759                 if (!apic_sw_enabled(apic))
760                         val |= APIC_LVT_MASKED;
761
762                 val &= apic_lvt_mask[(offset - APIC_LVTT) >> 4];
763                 apic_set_reg(apic, offset, val);
764
765                 break;
766
767         case APIC_TMICT:
768                 hrtimer_cancel(&apic->timer.dev);
769                 apic_set_reg(apic, APIC_TMICT, val);
770                 start_apic_timer(apic);
771                 return;
772
773         case APIC_TDCR:
774                 if (val & 4)
775                         printk(KERN_ERR "KVM_WRITE:TDCR %x\n", val);
776                 apic_set_reg(apic, APIC_TDCR, val);
777                 update_divide_count(apic);
778                 break;
779
780         default:
781                 apic_debug("Local APIC Write to read-only register %x\n",
782                            offset);
783                 break;
784         }
785
786 }
787
788 static int apic_mmio_range(struct kvm_io_device *this, gpa_t addr,
789                            int len, int size)
790 {
791         struct kvm_lapic *apic = (struct kvm_lapic *)this->private;
792         int ret = 0;
793
794
795         if (apic_hw_enabled(apic) &&
796             (addr >= apic->base_address) &&
797             (addr < (apic->base_address + LAPIC_MMIO_LENGTH)))
798                 ret = 1;
799
800         return ret;
801 }
802
803 void kvm_free_lapic(struct kvm_vcpu *vcpu)
804 {
805         if (!vcpu->arch.apic)
806                 return;
807
808         hrtimer_cancel(&vcpu->arch.apic->timer.dev);
809
810         if (vcpu->arch.apic->regs_page)
811                 __free_page(vcpu->arch.apic->regs_page);
812
813         kfree(vcpu->arch.apic);
814 }
815
816 /*
817  *----------------------------------------------------------------------
818  * LAPIC interface
819  *----------------------------------------------------------------------
820  */
821
822 void kvm_lapic_set_tpr(struct kvm_vcpu *vcpu, unsigned long cr8)
823 {
824         struct kvm_lapic *apic = vcpu->arch.apic;
825
826         if (!apic)
827                 return;
828         apic_set_tpr(apic, ((cr8 & 0x0f) << 4)
829                      | (apic_get_reg(apic, APIC_TASKPRI) & 4));
830 }
831 EXPORT_SYMBOL_GPL(kvm_lapic_set_tpr);
832
833 u64 kvm_lapic_get_cr8(struct kvm_vcpu *vcpu)
834 {
835         struct kvm_lapic *apic = vcpu->arch.apic;
836         u64 tpr;
837
838         if (!apic)
839                 return 0;
840         tpr = (u64) apic_get_reg(apic, APIC_TASKPRI);
841
842         return (tpr & 0xf0) >> 4;
843 }
844 EXPORT_SYMBOL_GPL(kvm_lapic_get_cr8);
845
846 void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
847 {
848         struct kvm_lapic *apic = vcpu->arch.apic;
849
850         if (!apic) {
851                 value |= MSR_IA32_APICBASE_BSP;
852                 vcpu->arch.apic_base = value;
853                 return;
854         }
855         if (apic->vcpu->vcpu_id)
856                 value &= ~MSR_IA32_APICBASE_BSP;
857
858         vcpu->arch.apic_base = value;
859         apic->base_address = apic->vcpu->arch.apic_base &
860                              MSR_IA32_APICBASE_BASE;
861
862         /* with FSB delivery interrupt, we can restart APIC functionality */
863         apic_debug("apic base msr is 0x%016" PRIx64 ", and base address is "
864                    "0x%lx.\n", apic->vcpu->arch.apic_base, apic->base_address);
865
866 }
867
868 u64 kvm_lapic_get_base(struct kvm_vcpu *vcpu)
869 {
870         return vcpu->arch.apic_base;
871 }
872 EXPORT_SYMBOL_GPL(kvm_lapic_get_base);
873
874 void kvm_lapic_reset(struct kvm_vcpu *vcpu)
875 {
876         struct kvm_lapic *apic;
877         int i;
878
879         apic_debug("%s\n", __func__);
880
881         ASSERT(vcpu);
882         apic = vcpu->arch.apic;
883         ASSERT(apic != NULL);
884
885         /* Stop the timer in case it's a reset to an active apic */
886         hrtimer_cancel(&apic->timer.dev);
887
888         apic_set_reg(apic, APIC_ID, vcpu->vcpu_id << 24);
889         apic_set_reg(apic, APIC_LVR, APIC_VERSION);
890
891         for (i = 0; i < APIC_LVT_NUM; i++)
892                 apic_set_reg(apic, APIC_LVTT + 0x10 * i, APIC_LVT_MASKED);
893         apic_set_reg(apic, APIC_LVT0,
894                      SET_APIC_DELIVERY_MODE(0, APIC_MODE_EXTINT));
895
896         apic_set_reg(apic, APIC_DFR, 0xffffffffU);
897         apic_set_reg(apic, APIC_SPIV, 0xff);
898         apic_set_reg(apic, APIC_TASKPRI, 0);
899         apic_set_reg(apic, APIC_LDR, 0);
900         apic_set_reg(apic, APIC_ESR, 0);
901         apic_set_reg(apic, APIC_ICR, 0);
902         apic_set_reg(apic, APIC_ICR2, 0);
903         apic_set_reg(apic, APIC_TDCR, 0);
904         apic_set_reg(apic, APIC_TMICT, 0);
905         for (i = 0; i < 8; i++) {
906                 apic_set_reg(apic, APIC_IRR + 0x10 * i, 0);
907                 apic_set_reg(apic, APIC_ISR + 0x10 * i, 0);
908                 apic_set_reg(apic, APIC_TMR + 0x10 * i, 0);
909         }
910         update_divide_count(apic);
911         atomic_set(&apic->timer.pending, 0);
912         if (vcpu->vcpu_id == 0)
913                 vcpu->arch.apic_base |= MSR_IA32_APICBASE_BSP;
914         apic_update_ppr(apic);
915
916         apic_debug(KERN_INFO "%s: vcpu=%p, id=%d, base_msr="
917                    "0x%016" PRIx64 ", base_address=0x%0lx.\n", __func__,
918                    vcpu, kvm_apic_id(apic),
919                    vcpu->arch.apic_base, apic->base_address);
920 }
921 EXPORT_SYMBOL_GPL(kvm_lapic_reset);
922
923 int kvm_lapic_enabled(struct kvm_vcpu *vcpu)
924 {
925         struct kvm_lapic *apic = vcpu->arch.apic;
926         int ret = 0;
927
928         if (!apic)
929                 return 0;
930         ret = apic_enabled(apic);
931
932         return ret;
933 }
934 EXPORT_SYMBOL_GPL(kvm_lapic_enabled);
935
936 /*
937  *----------------------------------------------------------------------
938  * timer interface
939  *----------------------------------------------------------------------
940  */
941
942 /* TODO: make sure __apic_timer_fn runs in current pCPU */
943 static int __apic_timer_fn(struct kvm_lapic *apic)
944 {
945         int result = 0;
946         wait_queue_head_t *q = &apic->vcpu->wq;
947
948         if(!atomic_inc_and_test(&apic->timer.pending))
949                 set_bit(KVM_REQ_PENDING_TIMER, &apic->vcpu->requests);
950         if (waitqueue_active(q)) {
951                 apic->vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
952                 wake_up_interruptible(q);
953         }
954         if (apic_lvtt_period(apic)) {
955                 result = 1;
956                 apic->timer.dev.expires = ktime_add_ns(
957                                         apic->timer.dev.expires,
958                                         apic->timer.period);
959         }
960         return result;
961 }
962
963 int apic_has_pending_timer(struct kvm_vcpu *vcpu)
964 {
965         struct kvm_lapic *lapic = vcpu->arch.apic;
966
967         if (lapic && apic_enabled(lapic) && apic_lvt_enabled(lapic, APIC_LVTT))
968                 return atomic_read(&lapic->timer.pending);
969
970         return 0;
971 }
972
973 static int __inject_apic_timer_irq(struct kvm_lapic *apic)
974 {
975         int vector;
976
977         vector = apic_lvt_vector(apic, APIC_LVTT);
978         return __apic_accept_irq(apic, APIC_DM_FIXED, vector, 1, 0);
979 }
980
981 static enum hrtimer_restart apic_timer_fn(struct hrtimer *data)
982 {
983         struct kvm_lapic *apic;
984         int restart_timer = 0;
985
986         apic = container_of(data, struct kvm_lapic, timer.dev);
987
988         restart_timer = __apic_timer_fn(apic);
989
990         if (restart_timer)
991                 return HRTIMER_RESTART;
992         else
993                 return HRTIMER_NORESTART;
994 }
995
996 int kvm_create_lapic(struct kvm_vcpu *vcpu)
997 {
998         struct kvm_lapic *apic;
999
1000         ASSERT(vcpu != NULL);
1001         apic_debug("apic_init %d\n", vcpu->vcpu_id);
1002
1003         apic = kzalloc(sizeof(*apic), GFP_KERNEL);
1004         if (!apic)
1005                 goto nomem;
1006
1007         vcpu->arch.apic = apic;
1008
1009         apic->regs_page = alloc_page(GFP_KERNEL);
1010         if (apic->regs_page == NULL) {
1011                 printk(KERN_ERR "malloc apic regs error for vcpu %x\n",
1012                        vcpu->vcpu_id);
1013                 goto nomem_free_apic;
1014         }
1015         apic->regs = page_address(apic->regs_page);
1016         memset(apic->regs, 0, PAGE_SIZE);
1017         apic->vcpu = vcpu;
1018
1019         hrtimer_init(&apic->timer.dev, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
1020         apic->timer.dev.function = apic_timer_fn;
1021         apic->base_address = APIC_DEFAULT_PHYS_BASE;
1022         vcpu->arch.apic_base = APIC_DEFAULT_PHYS_BASE;
1023
1024         kvm_lapic_reset(vcpu);
1025         apic->dev.read = apic_mmio_read;
1026         apic->dev.write = apic_mmio_write;
1027         apic->dev.in_range = apic_mmio_range;
1028         apic->dev.private = apic;
1029
1030         return 0;
1031 nomem_free_apic:
1032         kfree(apic);
1033 nomem:
1034         return -ENOMEM;
1035 }
1036 EXPORT_SYMBOL_GPL(kvm_create_lapic);
1037
1038 int kvm_apic_has_interrupt(struct kvm_vcpu *vcpu)
1039 {
1040         struct kvm_lapic *apic = vcpu->arch.apic;
1041         int highest_irr;
1042
1043         if (!apic || !apic_enabled(apic))
1044                 return -1;
1045
1046         apic_update_ppr(apic);
1047         highest_irr = apic_find_highest_irr(apic);
1048         if ((highest_irr == -1) ||
1049             ((highest_irr & 0xF0) <= apic_get_reg(apic, APIC_PROCPRI)))
1050                 return -1;
1051         return highest_irr;
1052 }
1053
1054 int kvm_apic_accept_pic_intr(struct kvm_vcpu *vcpu)
1055 {
1056         u32 lvt0 = apic_get_reg(vcpu->arch.apic, APIC_LVT0);
1057         int r = 0;
1058
1059         if (vcpu->vcpu_id == 0) {
1060                 if (!apic_hw_enabled(vcpu->arch.apic))
1061                         r = 1;
1062                 if ((lvt0 & APIC_LVT_MASKED) == 0 &&
1063                     GET_APIC_DELIVERY_MODE(lvt0) == APIC_MODE_EXTINT)
1064                         r = 1;
1065         }
1066         return r;
1067 }
1068
1069 void kvm_inject_apic_timer_irqs(struct kvm_vcpu *vcpu)
1070 {
1071         struct kvm_lapic *apic = vcpu->arch.apic;
1072
1073         if (apic && apic_lvt_enabled(apic, APIC_LVTT) &&
1074                 atomic_read(&apic->timer.pending) > 0) {
1075                 if (__inject_apic_timer_irq(apic))
1076                         atomic_dec(&apic->timer.pending);
1077         }
1078 }
1079
1080 void kvm_apic_timer_intr_post(struct kvm_vcpu *vcpu, int vec)
1081 {
1082         struct kvm_lapic *apic = vcpu->arch.apic;
1083
1084         if (apic && apic_lvt_vector(apic, APIC_LVTT) == vec)
1085                 apic->timer.last_update = ktime_add_ns(
1086                                 apic->timer.last_update,
1087                                 apic->timer.period);
1088 }
1089
1090 int kvm_get_apic_interrupt(struct kvm_vcpu *vcpu)
1091 {
1092         int vector = kvm_apic_has_interrupt(vcpu);
1093         struct kvm_lapic *apic = vcpu->arch.apic;
1094
1095         if (vector == -1)
1096                 return -1;
1097
1098         apic_set_vector(vector, apic->regs + APIC_ISR);
1099         apic_update_ppr(apic);
1100         apic_clear_irr(vector, apic);
1101         return vector;
1102 }
1103
1104 void kvm_apic_post_state_restore(struct kvm_vcpu *vcpu)
1105 {
1106         struct kvm_lapic *apic = vcpu->arch.apic;
1107
1108         apic->base_address = vcpu->arch.apic_base &
1109                              MSR_IA32_APICBASE_BASE;
1110         apic_set_reg(apic, APIC_LVR, APIC_VERSION);
1111         apic_update_ppr(apic);
1112         hrtimer_cancel(&apic->timer.dev);
1113         update_divide_count(apic);
1114         start_apic_timer(apic);
1115 }
1116
1117 void __kvm_migrate_apic_timer(struct kvm_vcpu *vcpu)
1118 {
1119         struct kvm_lapic *apic = vcpu->arch.apic;
1120         struct hrtimer *timer;
1121
1122         if (!apic)
1123                 return;
1124
1125         timer = &apic->timer.dev;
1126         if (hrtimer_cancel(timer))
1127                 hrtimer_start(timer, timer->expires, HRTIMER_MODE_ABS);
1128 }
1129
1130 void kvm_lapic_sync_from_vapic(struct kvm_vcpu *vcpu)
1131 {
1132         u32 data;
1133         void *vapic;
1134
1135         if (!irqchip_in_kernel(vcpu->kvm) || !vcpu->arch.apic->vapic_addr)
1136                 return;
1137
1138         vapic = kmap_atomic(vcpu->arch.apic->vapic_page, KM_USER0);
1139         data = *(u32 *)(vapic + offset_in_page(vcpu->arch.apic->vapic_addr));
1140         kunmap_atomic(vapic, KM_USER0);
1141
1142         apic_set_tpr(vcpu->arch.apic, data & 0xff);
1143 }
1144
1145 void kvm_lapic_sync_to_vapic(struct kvm_vcpu *vcpu)
1146 {
1147         u32 data, tpr;
1148         int max_irr, max_isr;
1149         struct kvm_lapic *apic;
1150         void *vapic;
1151
1152         if (!irqchip_in_kernel(vcpu->kvm) || !vcpu->arch.apic->vapic_addr)
1153                 return;
1154
1155         apic = vcpu->arch.apic;
1156         tpr = apic_get_reg(apic, APIC_TASKPRI) & 0xff;
1157         max_irr = apic_find_highest_irr(apic);
1158         if (max_irr < 0)
1159                 max_irr = 0;
1160         max_isr = apic_find_highest_isr(apic);
1161         if (max_isr < 0)
1162                 max_isr = 0;
1163         data = (tpr & 0xff) | ((max_isr & 0xf0) << 8) | (max_irr << 24);
1164
1165         vapic = kmap_atomic(vcpu->arch.apic->vapic_page, KM_USER0);
1166         *(u32 *)(vapic + offset_in_page(vcpu->arch.apic->vapic_addr)) = data;
1167         kunmap_atomic(vapic, KM_USER0);
1168 }
1169
1170 void kvm_lapic_set_vapic_addr(struct kvm_vcpu *vcpu, gpa_t vapic_addr)
1171 {
1172         if (!irqchip_in_kernel(vcpu->kvm))
1173                 return;
1174
1175         vcpu->arch.apic->vapic_addr = vapic_addr;
1176 }