KVM: cleanup io_device code
[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
38 #ifndef CONFIG_X86_64
39 #define mod_64(x, y) ((x) - (y) * div64_u64(x, y))
40 #else
41 #define mod_64(x, y) ((x) % (y))
42 #endif
43
44 #define PRId64 "d"
45 #define PRIx64 "llx"
46 #define PRIu64 "u"
47 #define PRIo64 "o"
48
49 #define APIC_BUS_CYCLE_NS 1
50
51 /* #define apic_debug(fmt,arg...) printk(KERN_WARNING fmt,##arg) */
52 #define apic_debug(fmt, arg...)
53
54 #define APIC_LVT_NUM                    6
55 /* 14 is the version for Xeon and Pentium 8.4.8*/
56 #define APIC_VERSION                    (0x14UL | ((APIC_LVT_NUM - 1) << 16))
57 #define LAPIC_MMIO_LENGTH               (1 << 12)
58 /* followed define is not in apicdef.h */
59 #define APIC_SHORT_MASK                 0xc0000
60 #define APIC_DEST_NOSHORT               0x0
61 #define APIC_DEST_MASK                  0x800
62 #define MAX_APIC_VECTOR                 256
63
64 #define VEC_POS(v) ((v) & (32 - 1))
65 #define REG_POS(v) (((v) >> 5) << 4)
66
67 static inline u32 apic_get_reg(struct kvm_lapic *apic, int reg_off)
68 {
69         return *((u32 *) (apic->regs + reg_off));
70 }
71
72 static inline void apic_set_reg(struct kvm_lapic *apic, int reg_off, u32 val)
73 {
74         *((u32 *) (apic->regs + reg_off)) = val;
75 }
76
77 static inline int apic_test_and_set_vector(int vec, void *bitmap)
78 {
79         return test_and_set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
80 }
81
82 static inline int apic_test_and_clear_vector(int vec, void *bitmap)
83 {
84         return test_and_clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
85 }
86
87 static inline void apic_set_vector(int vec, void *bitmap)
88 {
89         set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
90 }
91
92 static inline void apic_clear_vector(int vec, void *bitmap)
93 {
94         clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
95 }
96
97 static inline int apic_hw_enabled(struct kvm_lapic *apic)
98 {
99         return (apic)->vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE;
100 }
101
102 static inline int  apic_sw_enabled(struct kvm_lapic *apic)
103 {
104         return apic_get_reg(apic, APIC_SPIV) & APIC_SPIV_APIC_ENABLED;
105 }
106
107 static inline int apic_enabled(struct kvm_lapic *apic)
108 {
109         return apic_sw_enabled(apic) && apic_hw_enabled(apic);
110 }
111
112 #define LVT_MASK        \
113         (APIC_LVT_MASKED | APIC_SEND_PENDING | APIC_VECTOR_MASK)
114
115 #define LINT_MASK       \
116         (LVT_MASK | APIC_MODE_MASK | APIC_INPUT_POLARITY | \
117          APIC_LVT_REMOTE_IRR | APIC_LVT_LEVEL_TRIGGER)
118
119 static inline int kvm_apic_id(struct kvm_lapic *apic)
120 {
121         return (apic_get_reg(apic, APIC_ID) >> 24) & 0xff;
122 }
123
124 static inline int apic_lvt_enabled(struct kvm_lapic *apic, int lvt_type)
125 {
126         return !(apic_get_reg(apic, lvt_type) & APIC_LVT_MASKED);
127 }
128
129 static inline int apic_lvt_vector(struct kvm_lapic *apic, int lvt_type)
130 {
131         return apic_get_reg(apic, lvt_type) & APIC_VECTOR_MASK;
132 }
133
134 static inline int apic_lvtt_period(struct kvm_lapic *apic)
135 {
136         return apic_get_reg(apic, APIC_LVTT) & APIC_LVT_TIMER_PERIODIC;
137 }
138
139 static inline int apic_lvt_nmi_mode(u32 lvt_val)
140 {
141         return (lvt_val & (APIC_MODE_MASK | APIC_LVT_MASKED)) == APIC_DM_NMI;
142 }
143
144 static unsigned int apic_lvt_mask[APIC_LVT_NUM] = {
145         LVT_MASK | APIC_LVT_TIMER_PERIODIC,     /* LVTT */
146         LVT_MASK | APIC_MODE_MASK,      /* LVTTHMR */
147         LVT_MASK | APIC_MODE_MASK,      /* LVTPC */
148         LINT_MASK, LINT_MASK,   /* LVT0-1 */
149         LVT_MASK                /* LVTERR */
150 };
151
152 static int find_highest_vector(void *bitmap)
153 {
154         u32 *word = bitmap;
155         int word_offset = MAX_APIC_VECTOR >> 5;
156
157         while ((word_offset != 0) && (word[(--word_offset) << 2] == 0))
158                 continue;
159
160         if (likely(!word_offset && !word[0]))
161                 return -1;
162         else
163                 return fls(word[word_offset << 2]) - 1 + (word_offset << 5);
164 }
165
166 static inline int apic_test_and_set_irr(int vec, struct kvm_lapic *apic)
167 {
168         return apic_test_and_set_vector(vec, apic->regs + APIC_IRR);
169 }
170
171 static inline void apic_clear_irr(int vec, struct kvm_lapic *apic)
172 {
173         apic_clear_vector(vec, apic->regs + APIC_IRR);
174 }
175
176 static inline int apic_find_highest_irr(struct kvm_lapic *apic)
177 {
178         int result;
179
180         result = find_highest_vector(apic->regs + APIC_IRR);
181         ASSERT(result == -1 || result >= 16);
182
183         return result;
184 }
185
186 int kvm_lapic_find_highest_irr(struct kvm_vcpu *vcpu)
187 {
188         struct kvm_lapic *apic = vcpu->arch.apic;
189         int highest_irr;
190
191         if (!apic)
192                 return 0;
193         highest_irr = apic_find_highest_irr(apic);
194
195         return highest_irr;
196 }
197 EXPORT_SYMBOL_GPL(kvm_lapic_find_highest_irr);
198
199 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
200                              int vector, int level, int trig_mode);
201
202 int kvm_apic_set_irq(struct kvm_vcpu *vcpu, struct kvm_lapic_irq *irq)
203 {
204         struct kvm_lapic *apic = vcpu->arch.apic;
205
206         return __apic_accept_irq(apic, irq->delivery_mode, irq->vector,
207                         irq->level, irq->trig_mode);
208 }
209
210 static inline int apic_find_highest_isr(struct kvm_lapic *apic)
211 {
212         int result;
213
214         result = find_highest_vector(apic->regs + APIC_ISR);
215         ASSERT(result == -1 || result >= 16);
216
217         return result;
218 }
219
220 static void apic_update_ppr(struct kvm_lapic *apic)
221 {
222         u32 tpr, isrv, ppr;
223         int isr;
224
225         tpr = apic_get_reg(apic, APIC_TASKPRI);
226         isr = apic_find_highest_isr(apic);
227         isrv = (isr != -1) ? isr : 0;
228
229         if ((tpr & 0xf0) >= (isrv & 0xf0))
230                 ppr = tpr & 0xff;
231         else
232                 ppr = isrv & 0xf0;
233
234         apic_debug("vlapic %p, ppr 0x%x, isr 0x%x, isrv 0x%x",
235                    apic, ppr, isr, isrv);
236
237         apic_set_reg(apic, APIC_PROCPRI, ppr);
238 }
239
240 static void apic_set_tpr(struct kvm_lapic *apic, u32 tpr)
241 {
242         apic_set_reg(apic, APIC_TASKPRI, tpr);
243         apic_update_ppr(apic);
244 }
245
246 int kvm_apic_match_physical_addr(struct kvm_lapic *apic, u16 dest)
247 {
248         return dest == 0xff || kvm_apic_id(apic) == dest;
249 }
250
251 int kvm_apic_match_logical_addr(struct kvm_lapic *apic, u8 mda)
252 {
253         int result = 0;
254         u8 logical_id;
255
256         logical_id = GET_APIC_LOGICAL_ID(apic_get_reg(apic, APIC_LDR));
257
258         switch (apic_get_reg(apic, APIC_DFR)) {
259         case APIC_DFR_FLAT:
260                 if (logical_id & mda)
261                         result = 1;
262                 break;
263         case APIC_DFR_CLUSTER:
264                 if (((logical_id >> 4) == (mda >> 0x4))
265                     && (logical_id & mda & 0xf))
266                         result = 1;
267                 break;
268         default:
269                 printk(KERN_WARNING "Bad DFR vcpu %d: %08x\n",
270                        apic->vcpu->vcpu_id, apic_get_reg(apic, APIC_DFR));
271                 break;
272         }
273
274         return result;
275 }
276
277 int kvm_apic_match_dest(struct kvm_vcpu *vcpu, struct kvm_lapic *source,
278                            int short_hand, int dest, int dest_mode)
279 {
280         int result = 0;
281         struct kvm_lapic *target = vcpu->arch.apic;
282
283         apic_debug("target %p, source %p, dest 0x%x, "
284                    "dest_mode 0x%x, short_hand 0x%x\n",
285                    target, source, dest, dest_mode, short_hand);
286
287         ASSERT(!target);
288         switch (short_hand) {
289         case APIC_DEST_NOSHORT:
290                 if (dest_mode == 0)
291                         /* Physical mode. */
292                         result = kvm_apic_match_physical_addr(target, dest);
293                 else
294                         /* Logical mode. */
295                         result = kvm_apic_match_logical_addr(target, dest);
296                 break;
297         case APIC_DEST_SELF:
298                 result = (target == source);
299                 break;
300         case APIC_DEST_ALLINC:
301                 result = 1;
302                 break;
303         case APIC_DEST_ALLBUT:
304                 result = (target != source);
305                 break;
306         default:
307                 printk(KERN_WARNING "Bad dest shorthand value %x\n",
308                        short_hand);
309                 break;
310         }
311
312         return result;
313 }
314
315 /*
316  * Add a pending IRQ into lapic.
317  * Return 1 if successfully added and 0 if discarded.
318  */
319 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
320                              int vector, int level, int trig_mode)
321 {
322         int result = 0;
323         struct kvm_vcpu *vcpu = apic->vcpu;
324
325         switch (delivery_mode) {
326         case APIC_DM_LOWEST:
327                 vcpu->arch.apic_arb_prio++;
328         case APIC_DM_FIXED:
329                 /* FIXME add logic for vcpu on reset */
330                 if (unlikely(!apic_enabled(apic)))
331                         break;
332
333                 result = !apic_test_and_set_irr(vector, apic);
334                 if (!result) {
335                         if (trig_mode)
336                                 apic_debug("level trig mode repeatedly for "
337                                                 "vector %d", vector);
338                         break;
339                 }
340
341                 if (trig_mode) {
342                         apic_debug("level trig mode for vector %d", vector);
343                         apic_set_vector(vector, apic->regs + APIC_TMR);
344                 } else
345                         apic_clear_vector(vector, apic->regs + APIC_TMR);
346                 kvm_vcpu_kick(vcpu);
347                 break;
348
349         case APIC_DM_REMRD:
350                 printk(KERN_DEBUG "Ignoring delivery mode 3\n");
351                 break;
352
353         case APIC_DM_SMI:
354                 printk(KERN_DEBUG "Ignoring guest SMI\n");
355                 break;
356
357         case APIC_DM_NMI:
358                 result = 1;
359                 kvm_inject_nmi(vcpu);
360                 kvm_vcpu_kick(vcpu);
361                 break;
362
363         case APIC_DM_INIT:
364                 if (level) {
365                         result = 1;
366                         if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE)
367                                 printk(KERN_DEBUG
368                                        "INIT on a runnable vcpu %d\n",
369                                        vcpu->vcpu_id);
370                         vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
371                         kvm_vcpu_kick(vcpu);
372                 } else {
373                         apic_debug("Ignoring de-assert INIT to vcpu %d\n",
374                                    vcpu->vcpu_id);
375                 }
376                 break;
377
378         case APIC_DM_STARTUP:
379                 apic_debug("SIPI to vcpu %d vector 0x%02x\n",
380                            vcpu->vcpu_id, vector);
381                 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
382                         result = 1;
383                         vcpu->arch.sipi_vector = vector;
384                         vcpu->arch.mp_state = KVM_MP_STATE_SIPI_RECEIVED;
385                         kvm_vcpu_kick(vcpu);
386                 }
387                 break;
388
389         case APIC_DM_EXTINT:
390                 /*
391                  * Should only be called by kvm_apic_local_deliver() with LVT0,
392                  * before NMI watchdog was enabled. Already handled by
393                  * kvm_apic_accept_pic_intr().
394                  */
395                 break;
396
397         default:
398                 printk(KERN_ERR "TODO: unsupported delivery mode %x\n",
399                        delivery_mode);
400                 break;
401         }
402         return result;
403 }
404
405 int kvm_apic_compare_prio(struct kvm_vcpu *vcpu1, struct kvm_vcpu *vcpu2)
406 {
407         return vcpu1->arch.apic_arb_prio - vcpu2->arch.apic_arb_prio;
408 }
409
410 static void apic_set_eoi(struct kvm_lapic *apic)
411 {
412         int vector = apic_find_highest_isr(apic);
413         int trigger_mode;
414         /*
415          * Not every write EOI will has corresponding ISR,
416          * one example is when Kernel check timer on setup_IO_APIC
417          */
418         if (vector == -1)
419                 return;
420
421         apic_clear_vector(vector, apic->regs + APIC_ISR);
422         apic_update_ppr(apic);
423
424         if (apic_test_and_clear_vector(vector, apic->regs + APIC_TMR))
425                 trigger_mode = IOAPIC_LEVEL_TRIG;
426         else
427                 trigger_mode = IOAPIC_EDGE_TRIG;
428         kvm_ioapic_update_eoi(apic->vcpu->kvm, vector, trigger_mode);
429 }
430
431 static void apic_send_ipi(struct kvm_lapic *apic)
432 {
433         u32 icr_low = apic_get_reg(apic, APIC_ICR);
434         u32 icr_high = apic_get_reg(apic, APIC_ICR2);
435         struct kvm_lapic_irq irq;
436
437         irq.vector = icr_low & APIC_VECTOR_MASK;
438         irq.delivery_mode = icr_low & APIC_MODE_MASK;
439         irq.dest_mode = icr_low & APIC_DEST_MASK;
440         irq.level = icr_low & APIC_INT_ASSERT;
441         irq.trig_mode = icr_low & APIC_INT_LEVELTRIG;
442         irq.shorthand = icr_low & APIC_SHORT_MASK;
443         irq.dest_id = GET_APIC_DEST_FIELD(icr_high);
444
445         apic_debug("icr_high 0x%x, icr_low 0x%x, "
446                    "short_hand 0x%x, dest 0x%x, trig_mode 0x%x, level 0x%x, "
447                    "dest_mode 0x%x, delivery_mode 0x%x, vector 0x%x\n",
448                    icr_high, icr_low, irq.shorthand, irq.dest_id,
449                    irq.trig_mode, irq.level, irq.dest_mode, irq.delivery_mode,
450                    irq.vector);
451
452         kvm_irq_delivery_to_apic(apic->vcpu->kvm, apic, &irq);
453 }
454
455 static u32 apic_get_tmcct(struct kvm_lapic *apic)
456 {
457         ktime_t remaining;
458         s64 ns;
459         u32 tmcct;
460
461         ASSERT(apic != NULL);
462
463         /* if initial count is 0, current count should also be 0 */
464         if (apic_get_reg(apic, APIC_TMICT) == 0)
465                 return 0;
466
467         remaining = hrtimer_expires_remaining(&apic->lapic_timer.timer);
468         if (ktime_to_ns(remaining) < 0)
469                 remaining = ktime_set(0, 0);
470
471         ns = mod_64(ktime_to_ns(remaining), apic->lapic_timer.period);
472         tmcct = div64_u64(ns,
473                          (APIC_BUS_CYCLE_NS * apic->divide_count));
474
475         return tmcct;
476 }
477
478 static void __report_tpr_access(struct kvm_lapic *apic, bool write)
479 {
480         struct kvm_vcpu *vcpu = apic->vcpu;
481         struct kvm_run *run = vcpu->run;
482
483         set_bit(KVM_REQ_REPORT_TPR_ACCESS, &vcpu->requests);
484         run->tpr_access.rip = kvm_rip_read(vcpu);
485         run->tpr_access.is_write = write;
486 }
487
488 static inline void report_tpr_access(struct kvm_lapic *apic, bool write)
489 {
490         if (apic->vcpu->arch.tpr_access_reporting)
491                 __report_tpr_access(apic, write);
492 }
493
494 static u32 __apic_read(struct kvm_lapic *apic, unsigned int offset)
495 {
496         u32 val = 0;
497
498         KVMTRACE_1D(APIC_ACCESS, apic->vcpu, (u32)offset, handler);
499
500         if (offset >= LAPIC_MMIO_LENGTH)
501                 return 0;
502
503         switch (offset) {
504         case APIC_ARBPRI:
505                 printk(KERN_WARNING "Access APIC ARBPRI register "
506                        "which is for P6\n");
507                 break;
508
509         case APIC_TMCCT:        /* Timer CCR */
510                 val = apic_get_tmcct(apic);
511                 break;
512
513         case APIC_TASKPRI:
514                 report_tpr_access(apic, false);
515                 /* fall thru */
516         default:
517                 apic_update_ppr(apic);
518                 val = apic_get_reg(apic, offset);
519                 break;
520         }
521
522         return val;
523 }
524
525 static inline struct kvm_lapic *to_lapic(struct kvm_io_device *dev)
526 {
527         return container_of(dev, struct kvm_lapic, dev);
528 }
529
530 static void apic_mmio_read(struct kvm_io_device *this,
531                            gpa_t address, int len, void *data)
532 {
533         struct kvm_lapic *apic = to_lapic(this);
534         unsigned int offset = address - apic->base_address;
535         unsigned char alignment = offset & 0xf;
536         u32 result;
537
538         if ((alignment + len) > 4) {
539                 printk(KERN_ERR "KVM_APIC_READ: alignment error %lx %d",
540                        (unsigned long)address, len);
541                 return;
542         }
543         result = __apic_read(apic, offset & ~0xf);
544
545         switch (len) {
546         case 1:
547         case 2:
548         case 4:
549                 memcpy(data, (char *)&result + alignment, len);
550                 break;
551         default:
552                 printk(KERN_ERR "Local APIC read with len = %x, "
553                        "should be 1,2, or 4 instead\n", len);
554                 break;
555         }
556 }
557
558 static void update_divide_count(struct kvm_lapic *apic)
559 {
560         u32 tmp1, tmp2, tdcr;
561
562         tdcr = apic_get_reg(apic, APIC_TDCR);
563         tmp1 = tdcr & 0xf;
564         tmp2 = ((tmp1 & 0x3) | ((tmp1 & 0x8) >> 1)) + 1;
565         apic->divide_count = 0x1 << (tmp2 & 0x7);
566
567         apic_debug("timer divide count is 0x%x\n",
568                                    apic->divide_count);
569 }
570
571 static void start_apic_timer(struct kvm_lapic *apic)
572 {
573         ktime_t now = apic->lapic_timer.timer.base->get_time();
574
575         apic->lapic_timer.period = apic_get_reg(apic, APIC_TMICT) *
576                     APIC_BUS_CYCLE_NS * apic->divide_count;
577         atomic_set(&apic->lapic_timer.pending, 0);
578
579         if (!apic->lapic_timer.period)
580                 return;
581
582         hrtimer_start(&apic->lapic_timer.timer,
583                       ktime_add_ns(now, apic->lapic_timer.period),
584                       HRTIMER_MODE_ABS);
585
586         apic_debug("%s: bus cycle is %" PRId64 "ns, now 0x%016"
587                            PRIx64 ", "
588                            "timer initial count 0x%x, period %lldns, "
589                            "expire @ 0x%016" PRIx64 ".\n", __func__,
590                            APIC_BUS_CYCLE_NS, ktime_to_ns(now),
591                            apic_get_reg(apic, APIC_TMICT),
592                            apic->lapic_timer.period,
593                            ktime_to_ns(ktime_add_ns(now,
594                                         apic->lapic_timer.period)));
595 }
596
597 static void apic_manage_nmi_watchdog(struct kvm_lapic *apic, u32 lvt0_val)
598 {
599         int nmi_wd_enabled = apic_lvt_nmi_mode(apic_get_reg(apic, APIC_LVT0));
600
601         if (apic_lvt_nmi_mode(lvt0_val)) {
602                 if (!nmi_wd_enabled) {
603                         apic_debug("Receive NMI setting on APIC_LVT0 "
604                                    "for cpu %d\n", apic->vcpu->vcpu_id);
605                         apic->vcpu->kvm->arch.vapics_in_nmi_mode++;
606                 }
607         } else if (nmi_wd_enabled)
608                 apic->vcpu->kvm->arch.vapics_in_nmi_mode--;
609 }
610
611 static void apic_mmio_write(struct kvm_io_device *this,
612                             gpa_t address, int len, const void *data)
613 {
614         struct kvm_lapic *apic = to_lapic(this);
615         unsigned int offset = address - apic->base_address;
616         unsigned char alignment = offset & 0xf;
617         u32 val;
618
619         /*
620          * APIC register must be aligned on 128-bits boundary.
621          * 32/64/128 bits registers must be accessed thru 32 bits.
622          * Refer SDM 8.4.1
623          */
624         if (len != 4 || alignment) {
625                 /* Don't shout loud, $infamous_os would cause only noise. */
626                 apic_debug("apic write: bad size=%d %lx\n",
627                            len, (long)address);
628                 return;
629         }
630
631         val = *(u32 *) data;
632
633         /* too common printing */
634         if (offset != APIC_EOI)
635                 apic_debug("%s: offset 0x%x with length 0x%x, and value is "
636                            "0x%x\n", __func__, offset, len, val);
637
638         offset &= 0xff0;
639
640         KVMTRACE_1D(APIC_ACCESS, apic->vcpu, (u32)offset, handler);
641
642         switch (offset) {
643         case APIC_ID:           /* Local APIC ID */
644                 apic_set_reg(apic, APIC_ID, val);
645                 break;
646
647         case APIC_TASKPRI:
648                 report_tpr_access(apic, true);
649                 apic_set_tpr(apic, val & 0xff);
650                 break;
651
652         case APIC_EOI:
653                 apic_set_eoi(apic);
654                 break;
655
656         case APIC_LDR:
657                 apic_set_reg(apic, APIC_LDR, val & APIC_LDR_MASK);
658                 break;
659
660         case APIC_DFR:
661                 apic_set_reg(apic, APIC_DFR, val | 0x0FFFFFFF);
662                 break;
663
664         case APIC_SPIV:
665                 apic_set_reg(apic, APIC_SPIV, val & 0x3ff);
666                 if (!(val & APIC_SPIV_APIC_ENABLED)) {
667                         int i;
668                         u32 lvt_val;
669
670                         for (i = 0; i < APIC_LVT_NUM; i++) {
671                                 lvt_val = apic_get_reg(apic,
672                                                        APIC_LVTT + 0x10 * i);
673                                 apic_set_reg(apic, APIC_LVTT + 0x10 * i,
674                                              lvt_val | APIC_LVT_MASKED);
675                         }
676                         atomic_set(&apic->lapic_timer.pending, 0);
677
678                 }
679                 break;
680
681         case APIC_ICR:
682                 /* No delay here, so we always clear the pending bit */
683                 apic_set_reg(apic, APIC_ICR, val & ~(1 << 12));
684                 apic_send_ipi(apic);
685                 break;
686
687         case APIC_ICR2:
688                 apic_set_reg(apic, APIC_ICR2, val & 0xff000000);
689                 break;
690
691         case APIC_LVT0:
692                 apic_manage_nmi_watchdog(apic, val);
693         case APIC_LVTT:
694         case APIC_LVTTHMR:
695         case APIC_LVTPC:
696         case APIC_LVT1:
697         case APIC_LVTERR:
698                 /* TODO: Check vector */
699                 if (!apic_sw_enabled(apic))
700                         val |= APIC_LVT_MASKED;
701
702                 val &= apic_lvt_mask[(offset - APIC_LVTT) >> 4];
703                 apic_set_reg(apic, offset, val);
704
705                 break;
706
707         case APIC_TMICT:
708                 hrtimer_cancel(&apic->lapic_timer.timer);
709                 apic_set_reg(apic, APIC_TMICT, val);
710                 start_apic_timer(apic);
711                 return;
712
713         case APIC_TDCR:
714                 if (val & 4)
715                         printk(KERN_ERR "KVM_WRITE:TDCR %x\n", val);
716                 apic_set_reg(apic, APIC_TDCR, val);
717                 update_divide_count(apic);
718                 break;
719
720         default:
721                 apic_debug("Local APIC Write to read-only register %x\n",
722                            offset);
723                 break;
724         }
725
726 }
727
728 static int apic_mmio_range(struct kvm_io_device *this, gpa_t addr,
729                            int len, int size)
730 {
731         struct kvm_lapic *apic = to_lapic(this);
732         int ret = 0;
733
734
735         if (apic_hw_enabled(apic) &&
736             (addr >= apic->base_address) &&
737             (addr < (apic->base_address + LAPIC_MMIO_LENGTH)))
738                 ret = 1;
739
740         return ret;
741 }
742
743 void kvm_free_lapic(struct kvm_vcpu *vcpu)
744 {
745         if (!vcpu->arch.apic)
746                 return;
747
748         hrtimer_cancel(&vcpu->arch.apic->lapic_timer.timer);
749
750         if (vcpu->arch.apic->regs_page)
751                 __free_page(vcpu->arch.apic->regs_page);
752
753         kfree(vcpu->arch.apic);
754 }
755
756 /*
757  *----------------------------------------------------------------------
758  * LAPIC interface
759  *----------------------------------------------------------------------
760  */
761
762 void kvm_lapic_set_tpr(struct kvm_vcpu *vcpu, unsigned long cr8)
763 {
764         struct kvm_lapic *apic = vcpu->arch.apic;
765
766         if (!apic)
767                 return;
768         apic_set_tpr(apic, ((cr8 & 0x0f) << 4)
769                      | (apic_get_reg(apic, APIC_TASKPRI) & 4));
770 }
771 EXPORT_SYMBOL_GPL(kvm_lapic_set_tpr);
772
773 u64 kvm_lapic_get_cr8(struct kvm_vcpu *vcpu)
774 {
775         struct kvm_lapic *apic = vcpu->arch.apic;
776         u64 tpr;
777
778         if (!apic)
779                 return 0;
780         tpr = (u64) apic_get_reg(apic, APIC_TASKPRI);
781
782         return (tpr & 0xf0) >> 4;
783 }
784 EXPORT_SYMBOL_GPL(kvm_lapic_get_cr8);
785
786 void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
787 {
788         struct kvm_lapic *apic = vcpu->arch.apic;
789
790         if (!apic) {
791                 value |= MSR_IA32_APICBASE_BSP;
792                 vcpu->arch.apic_base = value;
793                 return;
794         }
795         if (apic->vcpu->vcpu_id)
796                 value &= ~MSR_IA32_APICBASE_BSP;
797
798         vcpu->arch.apic_base = value;
799         apic->base_address = apic->vcpu->arch.apic_base &
800                              MSR_IA32_APICBASE_BASE;
801
802         /* with FSB delivery interrupt, we can restart APIC functionality */
803         apic_debug("apic base msr is 0x%016" PRIx64 ", and base address is "
804                    "0x%lx.\n", apic->vcpu->arch.apic_base, apic->base_address);
805
806 }
807
808 u64 kvm_lapic_get_base(struct kvm_vcpu *vcpu)
809 {
810         return vcpu->arch.apic_base;
811 }
812 EXPORT_SYMBOL_GPL(kvm_lapic_get_base);
813
814 void kvm_lapic_reset(struct kvm_vcpu *vcpu)
815 {
816         struct kvm_lapic *apic;
817         int i;
818
819         apic_debug("%s\n", __func__);
820
821         ASSERT(vcpu);
822         apic = vcpu->arch.apic;
823         ASSERT(apic != NULL);
824
825         /* Stop the timer in case it's a reset to an active apic */
826         hrtimer_cancel(&apic->lapic_timer.timer);
827
828         apic_set_reg(apic, APIC_ID, vcpu->vcpu_id << 24);
829         apic_set_reg(apic, APIC_LVR, APIC_VERSION);
830
831         for (i = 0; i < APIC_LVT_NUM; i++)
832                 apic_set_reg(apic, APIC_LVTT + 0x10 * i, APIC_LVT_MASKED);
833         apic_set_reg(apic, APIC_LVT0,
834                      SET_APIC_DELIVERY_MODE(0, APIC_MODE_EXTINT));
835
836         apic_set_reg(apic, APIC_DFR, 0xffffffffU);
837         apic_set_reg(apic, APIC_SPIV, 0xff);
838         apic_set_reg(apic, APIC_TASKPRI, 0);
839         apic_set_reg(apic, APIC_LDR, 0);
840         apic_set_reg(apic, APIC_ESR, 0);
841         apic_set_reg(apic, APIC_ICR, 0);
842         apic_set_reg(apic, APIC_ICR2, 0);
843         apic_set_reg(apic, APIC_TDCR, 0);
844         apic_set_reg(apic, APIC_TMICT, 0);
845         for (i = 0; i < 8; i++) {
846                 apic_set_reg(apic, APIC_IRR + 0x10 * i, 0);
847                 apic_set_reg(apic, APIC_ISR + 0x10 * i, 0);
848                 apic_set_reg(apic, APIC_TMR + 0x10 * i, 0);
849         }
850         update_divide_count(apic);
851         atomic_set(&apic->lapic_timer.pending, 0);
852         if (vcpu->vcpu_id == 0)
853                 vcpu->arch.apic_base |= MSR_IA32_APICBASE_BSP;
854         apic_update_ppr(apic);
855
856         vcpu->arch.apic_arb_prio = 0;
857
858         apic_debug(KERN_INFO "%s: vcpu=%p, id=%d, base_msr="
859                    "0x%016" PRIx64 ", base_address=0x%0lx.\n", __func__,
860                    vcpu, kvm_apic_id(apic),
861                    vcpu->arch.apic_base, apic->base_address);
862 }
863 EXPORT_SYMBOL_GPL(kvm_lapic_reset);
864
865 bool kvm_apic_present(struct kvm_vcpu *vcpu)
866 {
867         return vcpu->arch.apic && apic_hw_enabled(vcpu->arch.apic);
868 }
869
870 int kvm_lapic_enabled(struct kvm_vcpu *vcpu)
871 {
872         return kvm_apic_present(vcpu) && apic_sw_enabled(vcpu->arch.apic);
873 }
874 EXPORT_SYMBOL_GPL(kvm_lapic_enabled);
875
876 /*
877  *----------------------------------------------------------------------
878  * timer interface
879  *----------------------------------------------------------------------
880  */
881
882 static bool lapic_is_periodic(struct kvm_timer *ktimer)
883 {
884         struct kvm_lapic *apic = container_of(ktimer, struct kvm_lapic,
885                                               lapic_timer);
886         return apic_lvtt_period(apic);
887 }
888
889 int apic_has_pending_timer(struct kvm_vcpu *vcpu)
890 {
891         struct kvm_lapic *lapic = vcpu->arch.apic;
892
893         if (lapic && apic_enabled(lapic) && apic_lvt_enabled(lapic, APIC_LVTT))
894                 return atomic_read(&lapic->lapic_timer.pending);
895
896         return 0;
897 }
898
899 static int kvm_apic_local_deliver(struct kvm_lapic *apic, int lvt_type)
900 {
901         u32 reg = apic_get_reg(apic, lvt_type);
902         int vector, mode, trig_mode;
903
904         if (apic_hw_enabled(apic) && !(reg & APIC_LVT_MASKED)) {
905                 vector = reg & APIC_VECTOR_MASK;
906                 mode = reg & APIC_MODE_MASK;
907                 trig_mode = reg & APIC_LVT_LEVEL_TRIGGER;
908                 return __apic_accept_irq(apic, mode, vector, 1, trig_mode);
909         }
910         return 0;
911 }
912
913 void kvm_apic_nmi_wd_deliver(struct kvm_vcpu *vcpu)
914 {
915         struct kvm_lapic *apic = vcpu->arch.apic;
916
917         if (apic)
918                 kvm_apic_local_deliver(apic, APIC_LVT0);
919 }
920
921 static struct kvm_timer_ops lapic_timer_ops = {
922         .is_periodic = lapic_is_periodic,
923 };
924
925 static const struct kvm_io_device_ops apic_mmio_ops = {
926         .read     = apic_mmio_read,
927         .write    = apic_mmio_write,
928         .in_range = apic_mmio_range,
929 };
930
931 int kvm_create_lapic(struct kvm_vcpu *vcpu)
932 {
933         struct kvm_lapic *apic;
934
935         ASSERT(vcpu != NULL);
936         apic_debug("apic_init %d\n", vcpu->vcpu_id);
937
938         apic = kzalloc(sizeof(*apic), GFP_KERNEL);
939         if (!apic)
940                 goto nomem;
941
942         vcpu->arch.apic = apic;
943
944         apic->regs_page = alloc_page(GFP_KERNEL);
945         if (apic->regs_page == NULL) {
946                 printk(KERN_ERR "malloc apic regs error for vcpu %x\n",
947                        vcpu->vcpu_id);
948                 goto nomem_free_apic;
949         }
950         apic->regs = page_address(apic->regs_page);
951         memset(apic->regs, 0, PAGE_SIZE);
952         apic->vcpu = vcpu;
953
954         hrtimer_init(&apic->lapic_timer.timer, CLOCK_MONOTONIC,
955                      HRTIMER_MODE_ABS);
956         apic->lapic_timer.timer.function = kvm_timer_fn;
957         apic->lapic_timer.t_ops = &lapic_timer_ops;
958         apic->lapic_timer.kvm = vcpu->kvm;
959         apic->lapic_timer.vcpu_id = vcpu->vcpu_id;
960
961         apic->base_address = APIC_DEFAULT_PHYS_BASE;
962         vcpu->arch.apic_base = APIC_DEFAULT_PHYS_BASE;
963
964         kvm_lapic_reset(vcpu);
965         kvm_iodevice_init(&apic->dev, &apic_mmio_ops);
966
967         return 0;
968 nomem_free_apic:
969         kfree(apic);
970 nomem:
971         return -ENOMEM;
972 }
973 EXPORT_SYMBOL_GPL(kvm_create_lapic);
974
975 int kvm_apic_has_interrupt(struct kvm_vcpu *vcpu)
976 {
977         struct kvm_lapic *apic = vcpu->arch.apic;
978         int highest_irr;
979
980         if (!apic || !apic_enabled(apic))
981                 return -1;
982
983         apic_update_ppr(apic);
984         highest_irr = apic_find_highest_irr(apic);
985         if ((highest_irr == -1) ||
986             ((highest_irr & 0xF0) <= apic_get_reg(apic, APIC_PROCPRI)))
987                 return -1;
988         return highest_irr;
989 }
990
991 int kvm_apic_accept_pic_intr(struct kvm_vcpu *vcpu)
992 {
993         u32 lvt0 = apic_get_reg(vcpu->arch.apic, APIC_LVT0);
994         int r = 0;
995
996         if (vcpu->vcpu_id == 0) {
997                 if (!apic_hw_enabled(vcpu->arch.apic))
998                         r = 1;
999                 if ((lvt0 & APIC_LVT_MASKED) == 0 &&
1000                     GET_APIC_DELIVERY_MODE(lvt0) == APIC_MODE_EXTINT)
1001                         r = 1;
1002         }
1003         return r;
1004 }
1005
1006 void kvm_inject_apic_timer_irqs(struct kvm_vcpu *vcpu)
1007 {
1008         struct kvm_lapic *apic = vcpu->arch.apic;
1009
1010         if (apic && atomic_read(&apic->lapic_timer.pending) > 0) {
1011                 if (kvm_apic_local_deliver(apic, APIC_LVTT))
1012                         atomic_dec(&apic->lapic_timer.pending);
1013         }
1014 }
1015
1016 int kvm_get_apic_interrupt(struct kvm_vcpu *vcpu)
1017 {
1018         int vector = kvm_apic_has_interrupt(vcpu);
1019         struct kvm_lapic *apic = vcpu->arch.apic;
1020
1021         if (vector == -1)
1022                 return -1;
1023
1024         apic_set_vector(vector, apic->regs + APIC_ISR);
1025         apic_update_ppr(apic);
1026         apic_clear_irr(vector, apic);
1027         return vector;
1028 }
1029
1030 void kvm_apic_post_state_restore(struct kvm_vcpu *vcpu)
1031 {
1032         struct kvm_lapic *apic = vcpu->arch.apic;
1033
1034         apic->base_address = vcpu->arch.apic_base &
1035                              MSR_IA32_APICBASE_BASE;
1036         apic_set_reg(apic, APIC_LVR, APIC_VERSION);
1037         apic_update_ppr(apic);
1038         hrtimer_cancel(&apic->lapic_timer.timer);
1039         update_divide_count(apic);
1040         start_apic_timer(apic);
1041 }
1042
1043 void __kvm_migrate_apic_timer(struct kvm_vcpu *vcpu)
1044 {
1045         struct kvm_lapic *apic = vcpu->arch.apic;
1046         struct hrtimer *timer;
1047
1048         if (!apic)
1049                 return;
1050
1051         timer = &apic->lapic_timer.timer;
1052         if (hrtimer_cancel(timer))
1053                 hrtimer_start_expires(timer, HRTIMER_MODE_ABS);
1054 }
1055
1056 void kvm_lapic_sync_from_vapic(struct kvm_vcpu *vcpu)
1057 {
1058         u32 data;
1059         void *vapic;
1060
1061         if (!irqchip_in_kernel(vcpu->kvm) || !vcpu->arch.apic->vapic_addr)
1062                 return;
1063
1064         vapic = kmap_atomic(vcpu->arch.apic->vapic_page, KM_USER0);
1065         data = *(u32 *)(vapic + offset_in_page(vcpu->arch.apic->vapic_addr));
1066         kunmap_atomic(vapic, KM_USER0);
1067
1068         apic_set_tpr(vcpu->arch.apic, data & 0xff);
1069 }
1070
1071 void kvm_lapic_sync_to_vapic(struct kvm_vcpu *vcpu)
1072 {
1073         u32 data, tpr;
1074         int max_irr, max_isr;
1075         struct kvm_lapic *apic;
1076         void *vapic;
1077
1078         if (!irqchip_in_kernel(vcpu->kvm) || !vcpu->arch.apic->vapic_addr)
1079                 return;
1080
1081         apic = vcpu->arch.apic;
1082         tpr = apic_get_reg(apic, APIC_TASKPRI) & 0xff;
1083         max_irr = apic_find_highest_irr(apic);
1084         if (max_irr < 0)
1085                 max_irr = 0;
1086         max_isr = apic_find_highest_isr(apic);
1087         if (max_isr < 0)
1088                 max_isr = 0;
1089         data = (tpr & 0xff) | ((max_isr & 0xf0) << 8) | (max_irr << 24);
1090
1091         vapic = kmap_atomic(vcpu->arch.apic->vapic_page, KM_USER0);
1092         *(u32 *)(vapic + offset_in_page(vcpu->arch.apic->vapic_addr)) = data;
1093         kunmap_atomic(vapic, KM_USER0);
1094 }
1095
1096 void kvm_lapic_set_vapic_addr(struct kvm_vcpu *vcpu, gpa_t vapic_addr)
1097 {
1098         if (!irqchip_in_kernel(vcpu->kvm))
1099                 return;
1100
1101         vcpu->arch.apic->vapic_addr = vapic_addr;
1102 }