KVM: convert custom marker based tracing to event traces
[safe/jmp/linux-2.6] / 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 #include "trace.h"
38
39 #ifndef CONFIG_X86_64
40 #define mod_64(x, y) ((x) - (y) * div64_u64(x, y))
41 #else
42 #define mod_64(x, y) ((x) % (y))
43 #endif
44
45 #define PRId64 "d"
46 #define PRIx64 "llx"
47 #define PRIu64 "u"
48 #define PRIo64 "o"
49
50 #define APIC_BUS_CYCLE_NS 1
51
52 /* #define apic_debug(fmt,arg...) printk(KERN_WARNING fmt,##arg) */
53 #define apic_debug(fmt, arg...)
54
55 #define APIC_LVT_NUM                    6
56 /* 14 is the version for Xeon and Pentium 8.4.8*/
57 #define APIC_VERSION                    (0x14UL | ((APIC_LVT_NUM - 1) << 16))
58 #define LAPIC_MMIO_LENGTH               (1 << 12)
59 /* followed define is not in apicdef.h */
60 #define APIC_SHORT_MASK                 0xc0000
61 #define APIC_DEST_NOSHORT               0x0
62 #define APIC_DEST_MASK                  0x800
63 #define MAX_APIC_VECTOR                 256
64
65 #define VEC_POS(v) ((v) & (32 - 1))
66 #define REG_POS(v) (((v) >> 5) << 4)
67
68 static inline u32 apic_get_reg(struct kvm_lapic *apic, int reg_off)
69 {
70         return *((u32 *) (apic->regs + reg_off));
71 }
72
73 static inline void apic_set_reg(struct kvm_lapic *apic, int reg_off, u32 val)
74 {
75         *((u32 *) (apic->regs + reg_off)) = val;
76 }
77
78 static inline int apic_test_and_set_vector(int vec, void *bitmap)
79 {
80         return test_and_set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
81 }
82
83 static inline int apic_test_and_clear_vector(int vec, void *bitmap)
84 {
85         return test_and_clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
86 }
87
88 static inline void apic_set_vector(int vec, void *bitmap)
89 {
90         set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
91 }
92
93 static inline void apic_clear_vector(int vec, void *bitmap)
94 {
95         clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
96 }
97
98 static inline int apic_hw_enabled(struct kvm_lapic *apic)
99 {
100         return (apic)->vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE;
101 }
102
103 static inline int  apic_sw_enabled(struct kvm_lapic *apic)
104 {
105         return apic_get_reg(apic, APIC_SPIV) & APIC_SPIV_APIC_ENABLED;
106 }
107
108 static inline int apic_enabled(struct kvm_lapic *apic)
109 {
110         return apic_sw_enabled(apic) && apic_hw_enabled(apic);
111 }
112
113 #define LVT_MASK        \
114         (APIC_LVT_MASKED | APIC_SEND_PENDING | APIC_VECTOR_MASK)
115
116 #define LINT_MASK       \
117         (LVT_MASK | APIC_MODE_MASK | APIC_INPUT_POLARITY | \
118          APIC_LVT_REMOTE_IRR | APIC_LVT_LEVEL_TRIGGER)
119
120 static inline int kvm_apic_id(struct kvm_lapic *apic)
121 {
122         return (apic_get_reg(apic, APIC_ID) >> 24) & 0xff;
123 }
124
125 static inline int apic_lvt_enabled(struct kvm_lapic *apic, int lvt_type)
126 {
127         return !(apic_get_reg(apic, lvt_type) & APIC_LVT_MASKED);
128 }
129
130 static inline int apic_lvt_vector(struct kvm_lapic *apic, int lvt_type)
131 {
132         return apic_get_reg(apic, lvt_type) & APIC_VECTOR_MASK;
133 }
134
135 static inline int apic_lvtt_period(struct kvm_lapic *apic)
136 {
137         return apic_get_reg(apic, APIC_LVTT) & APIC_LVT_TIMER_PERIODIC;
138 }
139
140 static inline int apic_lvt_nmi_mode(u32 lvt_val)
141 {
142         return (lvt_val & (APIC_MODE_MASK | APIC_LVT_MASKED)) == APIC_DM_NMI;
143 }
144
145 static unsigned int apic_lvt_mask[APIC_LVT_NUM] = {
146         LVT_MASK | APIC_LVT_TIMER_PERIODIC,     /* LVTT */
147         LVT_MASK | APIC_MODE_MASK,      /* LVTTHMR */
148         LVT_MASK | APIC_MODE_MASK,      /* LVTPC */
149         LINT_MASK, LINT_MASK,   /* LVT0-1 */
150         LVT_MASK                /* LVTERR */
151 };
152
153 static int find_highest_vector(void *bitmap)
154 {
155         u32 *word = bitmap;
156         int word_offset = MAX_APIC_VECTOR >> 5;
157
158         while ((word_offset != 0) && (word[(--word_offset) << 2] == 0))
159                 continue;
160
161         if (likely(!word_offset && !word[0]))
162                 return -1;
163         else
164                 return fls(word[word_offset << 2]) - 1 + (word_offset << 5);
165 }
166
167 static inline int apic_test_and_set_irr(int vec, struct kvm_lapic *apic)
168 {
169         apic->irr_pending = true;
170         return apic_test_and_set_vector(vec, apic->regs + APIC_IRR);
171 }
172
173 static inline int apic_search_irr(struct kvm_lapic *apic)
174 {
175         return find_highest_vector(apic->regs + APIC_IRR);
176 }
177
178 static inline int apic_find_highest_irr(struct kvm_lapic *apic)
179 {
180         int result;
181
182         if (!apic->irr_pending)
183                 return -1;
184
185         result = apic_search_irr(apic);
186         ASSERT(result == -1 || result >= 16);
187
188         return result;
189 }
190
191 static inline void apic_clear_irr(int vec, struct kvm_lapic *apic)
192 {
193         apic->irr_pending = false;
194         apic_clear_vector(vec, apic->regs + APIC_IRR);
195         if (apic_search_irr(apic) != -1)
196                 apic->irr_pending = true;
197 }
198
199 int kvm_lapic_find_highest_irr(struct kvm_vcpu *vcpu)
200 {
201         struct kvm_lapic *apic = vcpu->arch.apic;
202         int highest_irr;
203
204         /* This may race with setting of irr in __apic_accept_irq() and
205          * value returned may be wrong, but kvm_vcpu_kick() in __apic_accept_irq
206          * will cause vmexit immediately and the value will be recalculated
207          * on the next vmentry.
208          */
209         if (!apic)
210                 return 0;
211         highest_irr = apic_find_highest_irr(apic);
212
213         return highest_irr;
214 }
215
216 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
217                              int vector, int level, int trig_mode);
218
219 int kvm_apic_set_irq(struct kvm_vcpu *vcpu, struct kvm_lapic_irq *irq)
220 {
221         struct kvm_lapic *apic = vcpu->arch.apic;
222
223         return __apic_accept_irq(apic, irq->delivery_mode, irq->vector,
224                         irq->level, irq->trig_mode);
225 }
226
227 static inline int apic_find_highest_isr(struct kvm_lapic *apic)
228 {
229         int result;
230
231         result = find_highest_vector(apic->regs + APIC_ISR);
232         ASSERT(result == -1 || result >= 16);
233
234         return result;
235 }
236
237 static void apic_update_ppr(struct kvm_lapic *apic)
238 {
239         u32 tpr, isrv, ppr;
240         int isr;
241
242         tpr = apic_get_reg(apic, APIC_TASKPRI);
243         isr = apic_find_highest_isr(apic);
244         isrv = (isr != -1) ? isr : 0;
245
246         if ((tpr & 0xf0) >= (isrv & 0xf0))
247                 ppr = tpr & 0xff;
248         else
249                 ppr = isrv & 0xf0;
250
251         apic_debug("vlapic %p, ppr 0x%x, isr 0x%x, isrv 0x%x",
252                    apic, ppr, isr, isrv);
253
254         apic_set_reg(apic, APIC_PROCPRI, ppr);
255 }
256
257 static void apic_set_tpr(struct kvm_lapic *apic, u32 tpr)
258 {
259         apic_set_reg(apic, APIC_TASKPRI, tpr);
260         apic_update_ppr(apic);
261 }
262
263 int kvm_apic_match_physical_addr(struct kvm_lapic *apic, u16 dest)
264 {
265         return dest == 0xff || kvm_apic_id(apic) == dest;
266 }
267
268 int kvm_apic_match_logical_addr(struct kvm_lapic *apic, u8 mda)
269 {
270         int result = 0;
271         u8 logical_id;
272
273         logical_id = GET_APIC_LOGICAL_ID(apic_get_reg(apic, APIC_LDR));
274
275         switch (apic_get_reg(apic, APIC_DFR)) {
276         case APIC_DFR_FLAT:
277                 if (logical_id & mda)
278                         result = 1;
279                 break;
280         case APIC_DFR_CLUSTER:
281                 if (((logical_id >> 4) == (mda >> 0x4))
282                     && (logical_id & mda & 0xf))
283                         result = 1;
284                 break;
285         default:
286                 printk(KERN_WARNING "Bad DFR vcpu %d: %08x\n",
287                        apic->vcpu->vcpu_id, apic_get_reg(apic, APIC_DFR));
288                 break;
289         }
290
291         return result;
292 }
293
294 int kvm_apic_match_dest(struct kvm_vcpu *vcpu, struct kvm_lapic *source,
295                            int short_hand, int dest, int dest_mode)
296 {
297         int result = 0;
298         struct kvm_lapic *target = vcpu->arch.apic;
299
300         apic_debug("target %p, source %p, dest 0x%x, "
301                    "dest_mode 0x%x, short_hand 0x%x\n",
302                    target, source, dest, dest_mode, short_hand);
303
304         ASSERT(!target);
305         switch (short_hand) {
306         case APIC_DEST_NOSHORT:
307                 if (dest_mode == 0)
308                         /* Physical mode. */
309                         result = kvm_apic_match_physical_addr(target, dest);
310                 else
311                         /* Logical mode. */
312                         result = kvm_apic_match_logical_addr(target, dest);
313                 break;
314         case APIC_DEST_SELF:
315                 result = (target == source);
316                 break;
317         case APIC_DEST_ALLINC:
318                 result = 1;
319                 break;
320         case APIC_DEST_ALLBUT:
321                 result = (target != source);
322                 break;
323         default:
324                 printk(KERN_WARNING "Bad dest shorthand value %x\n",
325                        short_hand);
326                 break;
327         }
328
329         return result;
330 }
331
332 /*
333  * Add a pending IRQ into lapic.
334  * Return 1 if successfully added and 0 if discarded.
335  */
336 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
337                              int vector, int level, int trig_mode)
338 {
339         int result = 0;
340         struct kvm_vcpu *vcpu = apic->vcpu;
341
342         switch (delivery_mode) {
343         case APIC_DM_LOWEST:
344                 vcpu->arch.apic_arb_prio++;
345         case APIC_DM_FIXED:
346                 /* FIXME add logic for vcpu on reset */
347                 if (unlikely(!apic_enabled(apic)))
348                         break;
349
350                 result = !apic_test_and_set_irr(vector, apic);
351                 if (!result) {
352                         if (trig_mode)
353                                 apic_debug("level trig mode repeatedly for "
354                                                 "vector %d", vector);
355                         break;
356                 }
357
358                 if (trig_mode) {
359                         apic_debug("level trig mode for vector %d", vector);
360                         apic_set_vector(vector, apic->regs + APIC_TMR);
361                 } else
362                         apic_clear_vector(vector, apic->regs + APIC_TMR);
363                 kvm_vcpu_kick(vcpu);
364                 break;
365
366         case APIC_DM_REMRD:
367                 printk(KERN_DEBUG "Ignoring delivery mode 3\n");
368                 break;
369
370         case APIC_DM_SMI:
371                 printk(KERN_DEBUG "Ignoring guest SMI\n");
372                 break;
373
374         case APIC_DM_NMI:
375                 result = 1;
376                 kvm_inject_nmi(vcpu);
377                 kvm_vcpu_kick(vcpu);
378                 break;
379
380         case APIC_DM_INIT:
381                 if (level) {
382                         result = 1;
383                         if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE)
384                                 printk(KERN_DEBUG
385                                        "INIT on a runnable vcpu %d\n",
386                                        vcpu->vcpu_id);
387                         vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
388                         kvm_vcpu_kick(vcpu);
389                 } else {
390                         apic_debug("Ignoring de-assert INIT to vcpu %d\n",
391                                    vcpu->vcpu_id);
392                 }
393                 break;
394
395         case APIC_DM_STARTUP:
396                 apic_debug("SIPI to vcpu %d vector 0x%02x\n",
397                            vcpu->vcpu_id, vector);
398                 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
399                         result = 1;
400                         vcpu->arch.sipi_vector = vector;
401                         vcpu->arch.mp_state = KVM_MP_STATE_SIPI_RECEIVED;
402                         kvm_vcpu_kick(vcpu);
403                 }
404                 break;
405
406         case APIC_DM_EXTINT:
407                 /*
408                  * Should only be called by kvm_apic_local_deliver() with LVT0,
409                  * before NMI watchdog was enabled. Already handled by
410                  * kvm_apic_accept_pic_intr().
411                  */
412                 break;
413
414         default:
415                 printk(KERN_ERR "TODO: unsupported delivery mode %x\n",
416                        delivery_mode);
417                 break;
418         }
419         return result;
420 }
421
422 int kvm_apic_compare_prio(struct kvm_vcpu *vcpu1, struct kvm_vcpu *vcpu2)
423 {
424         return vcpu1->arch.apic_arb_prio - vcpu2->arch.apic_arb_prio;
425 }
426
427 static void apic_set_eoi(struct kvm_lapic *apic)
428 {
429         int vector = apic_find_highest_isr(apic);
430         int trigger_mode;
431         /*
432          * Not every write EOI will has corresponding ISR,
433          * one example is when Kernel check timer on setup_IO_APIC
434          */
435         if (vector == -1)
436                 return;
437
438         apic_clear_vector(vector, apic->regs + APIC_ISR);
439         apic_update_ppr(apic);
440
441         if (apic_test_and_clear_vector(vector, apic->regs + APIC_TMR))
442                 trigger_mode = IOAPIC_LEVEL_TRIG;
443         else
444                 trigger_mode = IOAPIC_EDGE_TRIG;
445         mutex_lock(&apic->vcpu->kvm->irq_lock);
446         kvm_ioapic_update_eoi(apic->vcpu->kvm, vector, trigger_mode);
447         mutex_unlock(&apic->vcpu->kvm->irq_lock);
448 }
449
450 static void apic_send_ipi(struct kvm_lapic *apic)
451 {
452         u32 icr_low = apic_get_reg(apic, APIC_ICR);
453         u32 icr_high = apic_get_reg(apic, APIC_ICR2);
454         struct kvm_lapic_irq irq;
455
456         irq.vector = icr_low & APIC_VECTOR_MASK;
457         irq.delivery_mode = icr_low & APIC_MODE_MASK;
458         irq.dest_mode = icr_low & APIC_DEST_MASK;
459         irq.level = icr_low & APIC_INT_ASSERT;
460         irq.trig_mode = icr_low & APIC_INT_LEVELTRIG;
461         irq.shorthand = icr_low & APIC_SHORT_MASK;
462         irq.dest_id = GET_APIC_DEST_FIELD(icr_high);
463
464         apic_debug("icr_high 0x%x, icr_low 0x%x, "
465                    "short_hand 0x%x, dest 0x%x, trig_mode 0x%x, level 0x%x, "
466                    "dest_mode 0x%x, delivery_mode 0x%x, vector 0x%x\n",
467                    icr_high, icr_low, irq.shorthand, irq.dest_id,
468                    irq.trig_mode, irq.level, irq.dest_mode, irq.delivery_mode,
469                    irq.vector);
470
471         mutex_lock(&apic->vcpu->kvm->irq_lock);
472         kvm_irq_delivery_to_apic(apic->vcpu->kvm, apic, &irq);
473         mutex_unlock(&apic->vcpu->kvm->irq_lock);
474 }
475
476 static u32 apic_get_tmcct(struct kvm_lapic *apic)
477 {
478         ktime_t remaining;
479         s64 ns;
480         u32 tmcct;
481
482         ASSERT(apic != NULL);
483
484         /* if initial count is 0, current count should also be 0 */
485         if (apic_get_reg(apic, APIC_TMICT) == 0)
486                 return 0;
487
488         remaining = hrtimer_expires_remaining(&apic->lapic_timer.timer);
489         if (ktime_to_ns(remaining) < 0)
490                 remaining = ktime_set(0, 0);
491
492         ns = mod_64(ktime_to_ns(remaining), apic->lapic_timer.period);
493         tmcct = div64_u64(ns,
494                          (APIC_BUS_CYCLE_NS * apic->divide_count));
495
496         return tmcct;
497 }
498
499 static void __report_tpr_access(struct kvm_lapic *apic, bool write)
500 {
501         struct kvm_vcpu *vcpu = apic->vcpu;
502         struct kvm_run *run = vcpu->run;
503
504         set_bit(KVM_REQ_REPORT_TPR_ACCESS, &vcpu->requests);
505         run->tpr_access.rip = kvm_rip_read(vcpu);
506         run->tpr_access.is_write = write;
507 }
508
509 static inline void report_tpr_access(struct kvm_lapic *apic, bool write)
510 {
511         if (apic->vcpu->arch.tpr_access_reporting)
512                 __report_tpr_access(apic, write);
513 }
514
515 static u32 __apic_read(struct kvm_lapic *apic, unsigned int offset)
516 {
517         u32 val = 0;
518
519         if (offset >= LAPIC_MMIO_LENGTH)
520                 return 0;
521
522         switch (offset) {
523         case APIC_ARBPRI:
524                 printk(KERN_WARNING "Access APIC ARBPRI register "
525                        "which is for P6\n");
526                 break;
527
528         case APIC_TMCCT:        /* Timer CCR */
529                 val = apic_get_tmcct(apic);
530                 break;
531
532         case APIC_TASKPRI:
533                 report_tpr_access(apic, false);
534                 /* fall thru */
535         default:
536                 apic_update_ppr(apic);
537                 val = apic_get_reg(apic, offset);
538                 break;
539         }
540
541         return val;
542 }
543
544 static inline struct kvm_lapic *to_lapic(struct kvm_io_device *dev)
545 {
546         return container_of(dev, struct kvm_lapic, dev);
547 }
548
549 static void apic_mmio_read(struct kvm_io_device *this,
550                            gpa_t address, int len, void *data)
551 {
552         struct kvm_lapic *apic = to_lapic(this);
553         unsigned int offset = address - apic->base_address;
554         unsigned char alignment = offset & 0xf;
555         u32 result;
556
557         if ((alignment + len) > 4) {
558                 printk(KERN_ERR "KVM_APIC_READ: alignment error %lx %d",
559                        (unsigned long)address, len);
560                 return;
561         }
562         result = __apic_read(apic, offset & ~0xf);
563
564         trace_kvm_apic_read(offset, result);
565
566         switch (len) {
567         case 1:
568         case 2:
569         case 4:
570                 memcpy(data, (char *)&result + alignment, len);
571                 break;
572         default:
573                 printk(KERN_ERR "Local APIC read with len = %x, "
574                        "should be 1,2, or 4 instead\n", len);
575                 break;
576         }
577 }
578
579 static void update_divide_count(struct kvm_lapic *apic)
580 {
581         u32 tmp1, tmp2, tdcr;
582
583         tdcr = apic_get_reg(apic, APIC_TDCR);
584         tmp1 = tdcr & 0xf;
585         tmp2 = ((tmp1 & 0x3) | ((tmp1 & 0x8) >> 1)) + 1;
586         apic->divide_count = 0x1 << (tmp2 & 0x7);
587
588         apic_debug("timer divide count is 0x%x\n",
589                                    apic->divide_count);
590 }
591
592 static void start_apic_timer(struct kvm_lapic *apic)
593 {
594         ktime_t now = apic->lapic_timer.timer.base->get_time();
595
596         apic->lapic_timer.period = apic_get_reg(apic, APIC_TMICT) *
597                     APIC_BUS_CYCLE_NS * apic->divide_count;
598         atomic_set(&apic->lapic_timer.pending, 0);
599
600         if (!apic->lapic_timer.period)
601                 return;
602
603         hrtimer_start(&apic->lapic_timer.timer,
604                       ktime_add_ns(now, apic->lapic_timer.period),
605                       HRTIMER_MODE_ABS);
606
607         apic_debug("%s: bus cycle is %" PRId64 "ns, now 0x%016"
608                            PRIx64 ", "
609                            "timer initial count 0x%x, period %lldns, "
610                            "expire @ 0x%016" PRIx64 ".\n", __func__,
611                            APIC_BUS_CYCLE_NS, ktime_to_ns(now),
612                            apic_get_reg(apic, APIC_TMICT),
613                            apic->lapic_timer.period,
614                            ktime_to_ns(ktime_add_ns(now,
615                                         apic->lapic_timer.period)));
616 }
617
618 static void apic_manage_nmi_watchdog(struct kvm_lapic *apic, u32 lvt0_val)
619 {
620         int nmi_wd_enabled = apic_lvt_nmi_mode(apic_get_reg(apic, APIC_LVT0));
621
622         if (apic_lvt_nmi_mode(lvt0_val)) {
623                 if (!nmi_wd_enabled) {
624                         apic_debug("Receive NMI setting on APIC_LVT0 "
625                                    "for cpu %d\n", apic->vcpu->vcpu_id);
626                         apic->vcpu->kvm->arch.vapics_in_nmi_mode++;
627                 }
628         } else if (nmi_wd_enabled)
629                 apic->vcpu->kvm->arch.vapics_in_nmi_mode--;
630 }
631
632 static void apic_mmio_write(struct kvm_io_device *this,
633                             gpa_t address, int len, const void *data)
634 {
635         struct kvm_lapic *apic = to_lapic(this);
636         unsigned int offset = address - apic->base_address;
637         unsigned char alignment = offset & 0xf;
638         u32 val;
639
640         /*
641          * APIC register must be aligned on 128-bits boundary.
642          * 32/64/128 bits registers must be accessed thru 32 bits.
643          * Refer SDM 8.4.1
644          */
645         if (len != 4 || alignment) {
646                 /* Don't shout loud, $infamous_os would cause only noise. */
647                 apic_debug("apic write: bad size=%d %lx\n",
648                            len, (long)address);
649                 return;
650         }
651
652         val = *(u32 *) data;
653
654         /* too common printing */
655         if (offset != APIC_EOI)
656                 apic_debug("%s: offset 0x%x with length 0x%x, and value is "
657                            "0x%x\n", __func__, offset, len, val);
658
659         offset &= 0xff0;
660
661         trace_kvm_apic_write(offset, val);
662
663         switch (offset) {
664         case APIC_ID:           /* Local APIC ID */
665                 apic_set_reg(apic, APIC_ID, val);
666                 break;
667
668         case APIC_TASKPRI:
669                 report_tpr_access(apic, true);
670                 apic_set_tpr(apic, val & 0xff);
671                 break;
672
673         case APIC_EOI:
674                 apic_set_eoi(apic);
675                 break;
676
677         case APIC_LDR:
678                 apic_set_reg(apic, APIC_LDR, val & APIC_LDR_MASK);
679                 break;
680
681         case APIC_DFR:
682                 apic_set_reg(apic, APIC_DFR, val | 0x0FFFFFFF);
683                 break;
684
685         case APIC_SPIV:
686                 apic_set_reg(apic, APIC_SPIV, val & 0x3ff);
687                 if (!(val & APIC_SPIV_APIC_ENABLED)) {
688                         int i;
689                         u32 lvt_val;
690
691                         for (i = 0; i < APIC_LVT_NUM; i++) {
692                                 lvt_val = apic_get_reg(apic,
693                                                        APIC_LVTT + 0x10 * i);
694                                 apic_set_reg(apic, APIC_LVTT + 0x10 * i,
695                                              lvt_val | APIC_LVT_MASKED);
696                         }
697                         atomic_set(&apic->lapic_timer.pending, 0);
698
699                 }
700                 break;
701
702         case APIC_ICR:
703                 /* No delay here, so we always clear the pending bit */
704                 apic_set_reg(apic, APIC_ICR, val & ~(1 << 12));
705                 apic_send_ipi(apic);
706                 break;
707
708         case APIC_ICR2:
709                 apic_set_reg(apic, APIC_ICR2, val & 0xff000000);
710                 break;
711
712         case APIC_LVT0:
713                 apic_manage_nmi_watchdog(apic, val);
714         case APIC_LVTT:
715         case APIC_LVTTHMR:
716         case APIC_LVTPC:
717         case APIC_LVT1:
718         case APIC_LVTERR:
719                 /* TODO: Check vector */
720                 if (!apic_sw_enabled(apic))
721                         val |= APIC_LVT_MASKED;
722
723                 val &= apic_lvt_mask[(offset - APIC_LVTT) >> 4];
724                 apic_set_reg(apic, offset, val);
725
726                 break;
727
728         case APIC_TMICT:
729                 hrtimer_cancel(&apic->lapic_timer.timer);
730                 apic_set_reg(apic, APIC_TMICT, val);
731                 start_apic_timer(apic);
732                 return;
733
734         case APIC_TDCR:
735                 if (val & 4)
736                         printk(KERN_ERR "KVM_WRITE:TDCR %x\n", val);
737                 apic_set_reg(apic, APIC_TDCR, val);
738                 update_divide_count(apic);
739                 break;
740
741         default:
742                 apic_debug("Local APIC Write to read-only register %x\n",
743                            offset);
744                 break;
745         }
746
747 }
748
749 static int apic_mmio_range(struct kvm_io_device *this, gpa_t addr,
750                            int len, int size)
751 {
752         struct kvm_lapic *apic = to_lapic(this);
753         int ret = 0;
754
755
756         if (apic_hw_enabled(apic) &&
757             (addr >= apic->base_address) &&
758             (addr < (apic->base_address + LAPIC_MMIO_LENGTH)))
759                 ret = 1;
760
761         return ret;
762 }
763
764 void kvm_free_lapic(struct kvm_vcpu *vcpu)
765 {
766         if (!vcpu->arch.apic)
767                 return;
768
769         hrtimer_cancel(&vcpu->arch.apic->lapic_timer.timer);
770
771         if (vcpu->arch.apic->regs_page)
772                 __free_page(vcpu->arch.apic->regs_page);
773
774         kfree(vcpu->arch.apic);
775 }
776
777 /*
778  *----------------------------------------------------------------------
779  * LAPIC interface
780  *----------------------------------------------------------------------
781  */
782
783 void kvm_lapic_set_tpr(struct kvm_vcpu *vcpu, unsigned long cr8)
784 {
785         struct kvm_lapic *apic = vcpu->arch.apic;
786
787         if (!apic)
788                 return;
789         apic_set_tpr(apic, ((cr8 & 0x0f) << 4)
790                      | (apic_get_reg(apic, APIC_TASKPRI) & 4));
791 }
792
793 u64 kvm_lapic_get_cr8(struct kvm_vcpu *vcpu)
794 {
795         struct kvm_lapic *apic = vcpu->arch.apic;
796         u64 tpr;
797
798         if (!apic)
799                 return 0;
800         tpr = (u64) apic_get_reg(apic, APIC_TASKPRI);
801
802         return (tpr & 0xf0) >> 4;
803 }
804
805 void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
806 {
807         struct kvm_lapic *apic = vcpu->arch.apic;
808
809         if (!apic) {
810                 value |= MSR_IA32_APICBASE_BSP;
811                 vcpu->arch.apic_base = value;
812                 return;
813         }
814
815         if (!kvm_vcpu_is_bsp(apic->vcpu))
816                 value &= ~MSR_IA32_APICBASE_BSP;
817
818         vcpu->arch.apic_base = value;
819         apic->base_address = apic->vcpu->arch.apic_base &
820                              MSR_IA32_APICBASE_BASE;
821
822         /* with FSB delivery interrupt, we can restart APIC functionality */
823         apic_debug("apic base msr is 0x%016" PRIx64 ", and base address is "
824                    "0x%lx.\n", apic->vcpu->arch.apic_base, apic->base_address);
825
826 }
827
828 void kvm_lapic_reset(struct kvm_vcpu *vcpu)
829 {
830         struct kvm_lapic *apic;
831         int i;
832
833         apic_debug("%s\n", __func__);
834
835         ASSERT(vcpu);
836         apic = vcpu->arch.apic;
837         ASSERT(apic != NULL);
838
839         /* Stop the timer in case it's a reset to an active apic */
840         hrtimer_cancel(&apic->lapic_timer.timer);
841
842         apic_set_reg(apic, APIC_ID, vcpu->vcpu_id << 24);
843         apic_set_reg(apic, APIC_LVR, APIC_VERSION);
844
845         for (i = 0; i < APIC_LVT_NUM; i++)
846                 apic_set_reg(apic, APIC_LVTT + 0x10 * i, APIC_LVT_MASKED);
847         apic_set_reg(apic, APIC_LVT0,
848                      SET_APIC_DELIVERY_MODE(0, APIC_MODE_EXTINT));
849
850         apic_set_reg(apic, APIC_DFR, 0xffffffffU);
851         apic_set_reg(apic, APIC_SPIV, 0xff);
852         apic_set_reg(apic, APIC_TASKPRI, 0);
853         apic_set_reg(apic, APIC_LDR, 0);
854         apic_set_reg(apic, APIC_ESR, 0);
855         apic_set_reg(apic, APIC_ICR, 0);
856         apic_set_reg(apic, APIC_ICR2, 0);
857         apic_set_reg(apic, APIC_TDCR, 0);
858         apic_set_reg(apic, APIC_TMICT, 0);
859         for (i = 0; i < 8; i++) {
860                 apic_set_reg(apic, APIC_IRR + 0x10 * i, 0);
861                 apic_set_reg(apic, APIC_ISR + 0x10 * i, 0);
862                 apic_set_reg(apic, APIC_TMR + 0x10 * i, 0);
863         }
864         apic->irr_pending = false;
865         update_divide_count(apic);
866         atomic_set(&apic->lapic_timer.pending, 0);
867         if (kvm_vcpu_is_bsp(vcpu))
868                 vcpu->arch.apic_base |= MSR_IA32_APICBASE_BSP;
869         apic_update_ppr(apic);
870
871         vcpu->arch.apic_arb_prio = 0;
872
873         apic_debug(KERN_INFO "%s: vcpu=%p, id=%d, base_msr="
874                    "0x%016" PRIx64 ", base_address=0x%0lx.\n", __func__,
875                    vcpu, kvm_apic_id(apic),
876                    vcpu->arch.apic_base, apic->base_address);
877 }
878
879 bool kvm_apic_present(struct kvm_vcpu *vcpu)
880 {
881         return vcpu->arch.apic && apic_hw_enabled(vcpu->arch.apic);
882 }
883
884 int kvm_lapic_enabled(struct kvm_vcpu *vcpu)
885 {
886         return kvm_apic_present(vcpu) && apic_sw_enabled(vcpu->arch.apic);
887 }
888
889 /*
890  *----------------------------------------------------------------------
891  * timer interface
892  *----------------------------------------------------------------------
893  */
894
895 static bool lapic_is_periodic(struct kvm_timer *ktimer)
896 {
897         struct kvm_lapic *apic = container_of(ktimer, struct kvm_lapic,
898                                               lapic_timer);
899         return apic_lvtt_period(apic);
900 }
901
902 int apic_has_pending_timer(struct kvm_vcpu *vcpu)
903 {
904         struct kvm_lapic *lapic = vcpu->arch.apic;
905
906         if (lapic && apic_enabled(lapic) && apic_lvt_enabled(lapic, APIC_LVTT))
907                 return atomic_read(&lapic->lapic_timer.pending);
908
909         return 0;
910 }
911
912 static int kvm_apic_local_deliver(struct kvm_lapic *apic, int lvt_type)
913 {
914         u32 reg = apic_get_reg(apic, lvt_type);
915         int vector, mode, trig_mode;
916
917         if (apic_hw_enabled(apic) && !(reg & APIC_LVT_MASKED)) {
918                 vector = reg & APIC_VECTOR_MASK;
919                 mode = reg & APIC_MODE_MASK;
920                 trig_mode = reg & APIC_LVT_LEVEL_TRIGGER;
921                 return __apic_accept_irq(apic, mode, vector, 1, trig_mode);
922         }
923         return 0;
924 }
925
926 void kvm_apic_nmi_wd_deliver(struct kvm_vcpu *vcpu)
927 {
928         struct kvm_lapic *apic = vcpu->arch.apic;
929
930         if (apic)
931                 kvm_apic_local_deliver(apic, APIC_LVT0);
932 }
933
934 static struct kvm_timer_ops lapic_timer_ops = {
935         .is_periodic = lapic_is_periodic,
936 };
937
938 static const struct kvm_io_device_ops apic_mmio_ops = {
939         .read     = apic_mmio_read,
940         .write    = apic_mmio_write,
941         .in_range = apic_mmio_range,
942 };
943
944 int kvm_create_lapic(struct kvm_vcpu *vcpu)
945 {
946         struct kvm_lapic *apic;
947
948         ASSERT(vcpu != NULL);
949         apic_debug("apic_init %d\n", vcpu->vcpu_id);
950
951         apic = kzalloc(sizeof(*apic), GFP_KERNEL);
952         if (!apic)
953                 goto nomem;
954
955         vcpu->arch.apic = apic;
956
957         apic->regs_page = alloc_page(GFP_KERNEL);
958         if (apic->regs_page == NULL) {
959                 printk(KERN_ERR "malloc apic regs error for vcpu %x\n",
960                        vcpu->vcpu_id);
961                 goto nomem_free_apic;
962         }
963         apic->regs = page_address(apic->regs_page);
964         memset(apic->regs, 0, PAGE_SIZE);
965         apic->vcpu = vcpu;
966
967         hrtimer_init(&apic->lapic_timer.timer, CLOCK_MONOTONIC,
968                      HRTIMER_MODE_ABS);
969         apic->lapic_timer.timer.function = kvm_timer_fn;
970         apic->lapic_timer.t_ops = &lapic_timer_ops;
971         apic->lapic_timer.kvm = vcpu->kvm;
972         apic->lapic_timer.vcpu = vcpu;
973
974         apic->base_address = APIC_DEFAULT_PHYS_BASE;
975         vcpu->arch.apic_base = APIC_DEFAULT_PHYS_BASE;
976
977         kvm_lapic_reset(vcpu);
978         kvm_iodevice_init(&apic->dev, &apic_mmio_ops);
979
980         return 0;
981 nomem_free_apic:
982         kfree(apic);
983 nomem:
984         return -ENOMEM;
985 }
986
987 int kvm_apic_has_interrupt(struct kvm_vcpu *vcpu)
988 {
989         struct kvm_lapic *apic = vcpu->arch.apic;
990         int highest_irr;
991
992         if (!apic || !apic_enabled(apic))
993                 return -1;
994
995         apic_update_ppr(apic);
996         highest_irr = apic_find_highest_irr(apic);
997         if ((highest_irr == -1) ||
998             ((highest_irr & 0xF0) <= apic_get_reg(apic, APIC_PROCPRI)))
999                 return -1;
1000         return highest_irr;
1001 }
1002
1003 int kvm_apic_accept_pic_intr(struct kvm_vcpu *vcpu)
1004 {
1005         u32 lvt0 = apic_get_reg(vcpu->arch.apic, APIC_LVT0);
1006         int r = 0;
1007
1008         if (kvm_vcpu_is_bsp(vcpu)) {
1009                 if (!apic_hw_enabled(vcpu->arch.apic))
1010                         r = 1;
1011                 if ((lvt0 & APIC_LVT_MASKED) == 0 &&
1012                     GET_APIC_DELIVERY_MODE(lvt0) == APIC_MODE_EXTINT)
1013                         r = 1;
1014         }
1015         return r;
1016 }
1017
1018 void kvm_inject_apic_timer_irqs(struct kvm_vcpu *vcpu)
1019 {
1020         struct kvm_lapic *apic = vcpu->arch.apic;
1021
1022         if (apic && atomic_read(&apic->lapic_timer.pending) > 0) {
1023                 if (kvm_apic_local_deliver(apic, APIC_LVTT))
1024                         atomic_dec(&apic->lapic_timer.pending);
1025         }
1026 }
1027
1028 int kvm_get_apic_interrupt(struct kvm_vcpu *vcpu)
1029 {
1030         int vector = kvm_apic_has_interrupt(vcpu);
1031         struct kvm_lapic *apic = vcpu->arch.apic;
1032
1033         if (vector == -1)
1034                 return -1;
1035
1036         apic_set_vector(vector, apic->regs + APIC_ISR);
1037         apic_update_ppr(apic);
1038         apic_clear_irr(vector, apic);
1039         return vector;
1040 }
1041
1042 void kvm_apic_post_state_restore(struct kvm_vcpu *vcpu)
1043 {
1044         struct kvm_lapic *apic = vcpu->arch.apic;
1045
1046         apic->base_address = vcpu->arch.apic_base &
1047                              MSR_IA32_APICBASE_BASE;
1048         apic_set_reg(apic, APIC_LVR, APIC_VERSION);
1049         apic_update_ppr(apic);
1050         hrtimer_cancel(&apic->lapic_timer.timer);
1051         update_divide_count(apic);
1052         start_apic_timer(apic);
1053 }
1054
1055 void __kvm_migrate_apic_timer(struct kvm_vcpu *vcpu)
1056 {
1057         struct kvm_lapic *apic = vcpu->arch.apic;
1058         struct hrtimer *timer;
1059
1060         if (!apic)
1061                 return;
1062
1063         timer = &apic->lapic_timer.timer;
1064         if (hrtimer_cancel(timer))
1065                 hrtimer_start_expires(timer, HRTIMER_MODE_ABS);
1066 }
1067
1068 void kvm_lapic_sync_from_vapic(struct kvm_vcpu *vcpu)
1069 {
1070         u32 data;
1071         void *vapic;
1072
1073         if (!irqchip_in_kernel(vcpu->kvm) || !vcpu->arch.apic->vapic_addr)
1074                 return;
1075
1076         vapic = kmap_atomic(vcpu->arch.apic->vapic_page, KM_USER0);
1077         data = *(u32 *)(vapic + offset_in_page(vcpu->arch.apic->vapic_addr));
1078         kunmap_atomic(vapic, KM_USER0);
1079
1080         apic_set_tpr(vcpu->arch.apic, data & 0xff);
1081 }
1082
1083 void kvm_lapic_sync_to_vapic(struct kvm_vcpu *vcpu)
1084 {
1085         u32 data, tpr;
1086         int max_irr, max_isr;
1087         struct kvm_lapic *apic;
1088         void *vapic;
1089
1090         if (!irqchip_in_kernel(vcpu->kvm) || !vcpu->arch.apic->vapic_addr)
1091                 return;
1092
1093         apic = vcpu->arch.apic;
1094         tpr = apic_get_reg(apic, APIC_TASKPRI) & 0xff;
1095         max_irr = apic_find_highest_irr(apic);
1096         if (max_irr < 0)
1097                 max_irr = 0;
1098         max_isr = apic_find_highest_isr(apic);
1099         if (max_isr < 0)
1100                 max_isr = 0;
1101         data = (tpr & 0xff) | ((max_isr & 0xf0) << 8) | (max_irr << 24);
1102
1103         vapic = kmap_atomic(vcpu->arch.apic->vapic_page, KM_USER0);
1104         *(u32 *)(vapic + offset_in_page(vcpu->arch.apic->vapic_addr)) = data;
1105         kunmap_atomic(vapic, KM_USER0);
1106 }
1107
1108 void kvm_lapic_set_vapic_addr(struct kvm_vcpu *vcpu, gpa_t vapic_addr)
1109 {
1110         if (!irqchip_in_kernel(vcpu->kvm))
1111                 return;
1112
1113         vcpu->arch.apic->vapic_addr = vapic_addr;
1114 }