KVM: Optimize searching for highest IRR
[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         apic->irr_pending = true;
169         return apic_test_and_set_vector(vec, apic->regs + APIC_IRR);
170 }
171
172 static inline int apic_search_irr(struct kvm_lapic *apic)
173 {
174         return find_highest_vector(apic->regs + APIC_IRR);
175 }
176
177 static inline int apic_find_highest_irr(struct kvm_lapic *apic)
178 {
179         int result;
180
181         if (!apic->irr_pending)
182                 return -1;
183
184         result = apic_search_irr(apic);
185         ASSERT(result == -1 || result >= 16);
186
187         return result;
188 }
189
190 static inline void apic_clear_irr(int vec, struct kvm_lapic *apic)
191 {
192         apic->irr_pending = false;
193         apic_clear_vector(vec, apic->regs + APIC_IRR);
194         if (apic_search_irr(apic) != -1)
195                 apic->irr_pending = true;
196 }
197
198 int kvm_lapic_find_highest_irr(struct kvm_vcpu *vcpu)
199 {
200         struct kvm_lapic *apic = vcpu->arch.apic;
201         int highest_irr;
202
203         /* This may race with setting of irr in __apic_accept_irq() and
204          * value returned may be wrong, but kvm_vcpu_kick() in __apic_accept_irq
205          * will cause vmexit immediately and the value will be recalculated
206          * on the next vmentry.
207          */
208         if (!apic)
209                 return 0;
210         highest_irr = apic_find_highest_irr(apic);
211
212         return highest_irr;
213 }
214
215 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
216                              int vector, int level, int trig_mode);
217
218 int kvm_apic_set_irq(struct kvm_vcpu *vcpu, struct kvm_lapic_irq *irq)
219 {
220         struct kvm_lapic *apic = vcpu->arch.apic;
221
222         return __apic_accept_irq(apic, irq->delivery_mode, irq->vector,
223                         irq->level, irq->trig_mode);
224 }
225
226 static inline int apic_find_highest_isr(struct kvm_lapic *apic)
227 {
228         int result;
229
230         result = find_highest_vector(apic->regs + APIC_ISR);
231         ASSERT(result == -1 || result >= 16);
232
233         return result;
234 }
235
236 static void apic_update_ppr(struct kvm_lapic *apic)
237 {
238         u32 tpr, isrv, ppr;
239         int isr;
240
241         tpr = apic_get_reg(apic, APIC_TASKPRI);
242         isr = apic_find_highest_isr(apic);
243         isrv = (isr != -1) ? isr : 0;
244
245         if ((tpr & 0xf0) >= (isrv & 0xf0))
246                 ppr = tpr & 0xff;
247         else
248                 ppr = isrv & 0xf0;
249
250         apic_debug("vlapic %p, ppr 0x%x, isr 0x%x, isrv 0x%x",
251                    apic, ppr, isr, isrv);
252
253         apic_set_reg(apic, APIC_PROCPRI, ppr);
254 }
255
256 static void apic_set_tpr(struct kvm_lapic *apic, u32 tpr)
257 {
258         apic_set_reg(apic, APIC_TASKPRI, tpr);
259         apic_update_ppr(apic);
260 }
261
262 int kvm_apic_match_physical_addr(struct kvm_lapic *apic, u16 dest)
263 {
264         return dest == 0xff || kvm_apic_id(apic) == dest;
265 }
266
267 int kvm_apic_match_logical_addr(struct kvm_lapic *apic, u8 mda)
268 {
269         int result = 0;
270         u8 logical_id;
271
272         logical_id = GET_APIC_LOGICAL_ID(apic_get_reg(apic, APIC_LDR));
273
274         switch (apic_get_reg(apic, APIC_DFR)) {
275         case APIC_DFR_FLAT:
276                 if (logical_id & mda)
277                         result = 1;
278                 break;
279         case APIC_DFR_CLUSTER:
280                 if (((logical_id >> 4) == (mda >> 0x4))
281                     && (logical_id & mda & 0xf))
282                         result = 1;
283                 break;
284         default:
285                 printk(KERN_WARNING "Bad DFR vcpu %d: %08x\n",
286                        apic->vcpu->vcpu_id, apic_get_reg(apic, APIC_DFR));
287                 break;
288         }
289
290         return result;
291 }
292
293 int kvm_apic_match_dest(struct kvm_vcpu *vcpu, struct kvm_lapic *source,
294                            int short_hand, int dest, int dest_mode)
295 {
296         int result = 0;
297         struct kvm_lapic *target = vcpu->arch.apic;
298
299         apic_debug("target %p, source %p, dest 0x%x, "
300                    "dest_mode 0x%x, short_hand 0x%x\n",
301                    target, source, dest, dest_mode, short_hand);
302
303         ASSERT(!target);
304         switch (short_hand) {
305         case APIC_DEST_NOSHORT:
306                 if (dest_mode == 0)
307                         /* Physical mode. */
308                         result = kvm_apic_match_physical_addr(target, dest);
309                 else
310                         /* Logical mode. */
311                         result = kvm_apic_match_logical_addr(target, dest);
312                 break;
313         case APIC_DEST_SELF:
314                 result = (target == source);
315                 break;
316         case APIC_DEST_ALLINC:
317                 result = 1;
318                 break;
319         case APIC_DEST_ALLBUT:
320                 result = (target != source);
321                 break;
322         default:
323                 printk(KERN_WARNING "Bad dest shorthand value %x\n",
324                        short_hand);
325                 break;
326         }
327
328         return result;
329 }
330
331 /*
332  * Add a pending IRQ into lapic.
333  * Return 1 if successfully added and 0 if discarded.
334  */
335 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
336                              int vector, int level, int trig_mode)
337 {
338         int result = 0;
339         struct kvm_vcpu *vcpu = apic->vcpu;
340
341         switch (delivery_mode) {
342         case APIC_DM_LOWEST:
343                 vcpu->arch.apic_arb_prio++;
344         case APIC_DM_FIXED:
345                 /* FIXME add logic for vcpu on reset */
346                 if (unlikely(!apic_enabled(apic)))
347                         break;
348
349                 result = !apic_test_and_set_irr(vector, apic);
350                 if (!result) {
351                         if (trig_mode)
352                                 apic_debug("level trig mode repeatedly for "
353                                                 "vector %d", vector);
354                         break;
355                 }
356
357                 if (trig_mode) {
358                         apic_debug("level trig mode for vector %d", vector);
359                         apic_set_vector(vector, apic->regs + APIC_TMR);
360                 } else
361                         apic_clear_vector(vector, apic->regs + APIC_TMR);
362                 kvm_vcpu_kick(vcpu);
363                 break;
364
365         case APIC_DM_REMRD:
366                 printk(KERN_DEBUG "Ignoring delivery mode 3\n");
367                 break;
368
369         case APIC_DM_SMI:
370                 printk(KERN_DEBUG "Ignoring guest SMI\n");
371                 break;
372
373         case APIC_DM_NMI:
374                 result = 1;
375                 kvm_inject_nmi(vcpu);
376                 kvm_vcpu_kick(vcpu);
377                 break;
378
379         case APIC_DM_INIT:
380                 if (level) {
381                         result = 1;
382                         if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE)
383                                 printk(KERN_DEBUG
384                                        "INIT on a runnable vcpu %d\n",
385                                        vcpu->vcpu_id);
386                         vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
387                         kvm_vcpu_kick(vcpu);
388                 } else {
389                         apic_debug("Ignoring de-assert INIT to vcpu %d\n",
390                                    vcpu->vcpu_id);
391                 }
392                 break;
393
394         case APIC_DM_STARTUP:
395                 apic_debug("SIPI to vcpu %d vector 0x%02x\n",
396                            vcpu->vcpu_id, vector);
397                 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
398                         result = 1;
399                         vcpu->arch.sipi_vector = vector;
400                         vcpu->arch.mp_state = KVM_MP_STATE_SIPI_RECEIVED;
401                         kvm_vcpu_kick(vcpu);
402                 }
403                 break;
404
405         case APIC_DM_EXTINT:
406                 /*
407                  * Should only be called by kvm_apic_local_deliver() with LVT0,
408                  * before NMI watchdog was enabled. Already handled by
409                  * kvm_apic_accept_pic_intr().
410                  */
411                 break;
412
413         default:
414                 printk(KERN_ERR "TODO: unsupported delivery mode %x\n",
415                        delivery_mode);
416                 break;
417         }
418         return result;
419 }
420
421 int kvm_apic_compare_prio(struct kvm_vcpu *vcpu1, struct kvm_vcpu *vcpu2)
422 {
423         return vcpu1->arch.apic_arb_prio - vcpu2->arch.apic_arb_prio;
424 }
425
426 static void apic_set_eoi(struct kvm_lapic *apic)
427 {
428         int vector = apic_find_highest_isr(apic);
429         int trigger_mode;
430         /*
431          * Not every write EOI will has corresponding ISR,
432          * one example is when Kernel check timer on setup_IO_APIC
433          */
434         if (vector == -1)
435                 return;
436
437         apic_clear_vector(vector, apic->regs + APIC_ISR);
438         apic_update_ppr(apic);
439
440         if (apic_test_and_clear_vector(vector, apic->regs + APIC_TMR))
441                 trigger_mode = IOAPIC_LEVEL_TRIG;
442         else
443                 trigger_mode = IOAPIC_EDGE_TRIG;
444         mutex_lock(&apic->vcpu->kvm->irq_lock);
445         kvm_ioapic_update_eoi(apic->vcpu->kvm, vector, trigger_mode);
446         mutex_unlock(&apic->vcpu->kvm->irq_lock);
447 }
448
449 static void apic_send_ipi(struct kvm_lapic *apic)
450 {
451         u32 icr_low = apic_get_reg(apic, APIC_ICR);
452         u32 icr_high = apic_get_reg(apic, APIC_ICR2);
453         struct kvm_lapic_irq irq;
454
455         irq.vector = icr_low & APIC_VECTOR_MASK;
456         irq.delivery_mode = icr_low & APIC_MODE_MASK;
457         irq.dest_mode = icr_low & APIC_DEST_MASK;
458         irq.level = icr_low & APIC_INT_ASSERT;
459         irq.trig_mode = icr_low & APIC_INT_LEVELTRIG;
460         irq.shorthand = icr_low & APIC_SHORT_MASK;
461         irq.dest_id = GET_APIC_DEST_FIELD(icr_high);
462
463         apic_debug("icr_high 0x%x, icr_low 0x%x, "
464                    "short_hand 0x%x, dest 0x%x, trig_mode 0x%x, level 0x%x, "
465                    "dest_mode 0x%x, delivery_mode 0x%x, vector 0x%x\n",
466                    icr_high, icr_low, irq.shorthand, irq.dest_id,
467                    irq.trig_mode, irq.level, irq.dest_mode, irq.delivery_mode,
468                    irq.vector);
469
470         mutex_lock(&apic->vcpu->kvm->irq_lock);
471         kvm_irq_delivery_to_apic(apic->vcpu->kvm, apic, &irq);
472         mutex_unlock(&apic->vcpu->kvm->irq_lock);
473 }
474
475 static u32 apic_get_tmcct(struct kvm_lapic *apic)
476 {
477         ktime_t remaining;
478         s64 ns;
479         u32 tmcct;
480
481         ASSERT(apic != NULL);
482
483         /* if initial count is 0, current count should also be 0 */
484         if (apic_get_reg(apic, APIC_TMICT) == 0)
485                 return 0;
486
487         remaining = hrtimer_expires_remaining(&apic->lapic_timer.timer);
488         if (ktime_to_ns(remaining) < 0)
489                 remaining = ktime_set(0, 0);
490
491         ns = mod_64(ktime_to_ns(remaining), apic->lapic_timer.period);
492         tmcct = div64_u64(ns,
493                          (APIC_BUS_CYCLE_NS * apic->divide_count));
494
495         return tmcct;
496 }
497
498 static void __report_tpr_access(struct kvm_lapic *apic, bool write)
499 {
500         struct kvm_vcpu *vcpu = apic->vcpu;
501         struct kvm_run *run = vcpu->run;
502
503         set_bit(KVM_REQ_REPORT_TPR_ACCESS, &vcpu->requests);
504         run->tpr_access.rip = kvm_rip_read(vcpu);
505         run->tpr_access.is_write = write;
506 }
507
508 static inline void report_tpr_access(struct kvm_lapic *apic, bool write)
509 {
510         if (apic->vcpu->arch.tpr_access_reporting)
511                 __report_tpr_access(apic, write);
512 }
513
514 static u32 __apic_read(struct kvm_lapic *apic, unsigned int offset)
515 {
516         u32 val = 0;
517
518         KVMTRACE_1D(APIC_ACCESS, apic->vcpu, (u32)offset, handler);
519
520         if (offset >= LAPIC_MMIO_LENGTH)
521                 return 0;
522
523         switch (offset) {
524         case APIC_ARBPRI:
525                 printk(KERN_WARNING "Access APIC ARBPRI register "
526                        "which is for P6\n");
527                 break;
528
529         case APIC_TMCCT:        /* Timer CCR */
530                 val = apic_get_tmcct(apic);
531                 break;
532
533         case APIC_TASKPRI:
534                 report_tpr_access(apic, false);
535                 /* fall thru */
536         default:
537                 apic_update_ppr(apic);
538                 val = apic_get_reg(apic, offset);
539                 break;
540         }
541
542         return val;
543 }
544
545 static inline struct kvm_lapic *to_lapic(struct kvm_io_device *dev)
546 {
547         return container_of(dev, struct kvm_lapic, dev);
548 }
549
550 static void apic_mmio_read(struct kvm_io_device *this,
551                            gpa_t address, int len, void *data)
552 {
553         struct kvm_lapic *apic = to_lapic(this);
554         unsigned int offset = address - apic->base_address;
555         unsigned char alignment = offset & 0xf;
556         u32 result;
557
558         if ((alignment + len) > 4) {
559                 printk(KERN_ERR "KVM_APIC_READ: alignment error %lx %d",
560                        (unsigned long)address, len);
561                 return;
562         }
563         result = __apic_read(apic, offset & ~0xf);
564
565         switch (len) {
566         case 1:
567         case 2:
568         case 4:
569                 memcpy(data, (char *)&result + alignment, len);
570                 break;
571         default:
572                 printk(KERN_ERR "Local APIC read with len = %x, "
573                        "should be 1,2, or 4 instead\n", len);
574                 break;
575         }
576 }
577
578 static void update_divide_count(struct kvm_lapic *apic)
579 {
580         u32 tmp1, tmp2, tdcr;
581
582         tdcr = apic_get_reg(apic, APIC_TDCR);
583         tmp1 = tdcr & 0xf;
584         tmp2 = ((tmp1 & 0x3) | ((tmp1 & 0x8) >> 1)) + 1;
585         apic->divide_count = 0x1 << (tmp2 & 0x7);
586
587         apic_debug("timer divide count is 0x%x\n",
588                                    apic->divide_count);
589 }
590
591 static void start_apic_timer(struct kvm_lapic *apic)
592 {
593         ktime_t now = apic->lapic_timer.timer.base->get_time();
594
595         apic->lapic_timer.period = apic_get_reg(apic, APIC_TMICT) *
596                     APIC_BUS_CYCLE_NS * apic->divide_count;
597         atomic_set(&apic->lapic_timer.pending, 0);
598
599         if (!apic->lapic_timer.period)
600                 return;
601
602         hrtimer_start(&apic->lapic_timer.timer,
603                       ktime_add_ns(now, apic->lapic_timer.period),
604                       HRTIMER_MODE_ABS);
605
606         apic_debug("%s: bus cycle is %" PRId64 "ns, now 0x%016"
607                            PRIx64 ", "
608                            "timer initial count 0x%x, period %lldns, "
609                            "expire @ 0x%016" PRIx64 ".\n", __func__,
610                            APIC_BUS_CYCLE_NS, ktime_to_ns(now),
611                            apic_get_reg(apic, APIC_TMICT),
612                            apic->lapic_timer.period,
613                            ktime_to_ns(ktime_add_ns(now,
614                                         apic->lapic_timer.period)));
615 }
616
617 static void apic_manage_nmi_watchdog(struct kvm_lapic *apic, u32 lvt0_val)
618 {
619         int nmi_wd_enabled = apic_lvt_nmi_mode(apic_get_reg(apic, APIC_LVT0));
620
621         if (apic_lvt_nmi_mode(lvt0_val)) {
622                 if (!nmi_wd_enabled) {
623                         apic_debug("Receive NMI setting on APIC_LVT0 "
624                                    "for cpu %d\n", apic->vcpu->vcpu_id);
625                         apic->vcpu->kvm->arch.vapics_in_nmi_mode++;
626                 }
627         } else if (nmi_wd_enabled)
628                 apic->vcpu->kvm->arch.vapics_in_nmi_mode--;
629 }
630
631 static void apic_mmio_write(struct kvm_io_device *this,
632                             gpa_t address, int len, const void *data)
633 {
634         struct kvm_lapic *apic = to_lapic(this);
635         unsigned int offset = address - apic->base_address;
636         unsigned char alignment = offset & 0xf;
637         u32 val;
638
639         /*
640          * APIC register must be aligned on 128-bits boundary.
641          * 32/64/128 bits registers must be accessed thru 32 bits.
642          * Refer SDM 8.4.1
643          */
644         if (len != 4 || alignment) {
645                 /* Don't shout loud, $infamous_os would cause only noise. */
646                 apic_debug("apic write: bad size=%d %lx\n",
647                            len, (long)address);
648                 return;
649         }
650
651         val = *(u32 *) data;
652
653         /* too common printing */
654         if (offset != APIC_EOI)
655                 apic_debug("%s: offset 0x%x with length 0x%x, and value is "
656                            "0x%x\n", __func__, offset, len, val);
657
658         offset &= 0xff0;
659
660         KVMTRACE_1D(APIC_ACCESS, apic->vcpu, (u32)offset, handler);
661
662         switch (offset) {
663         case APIC_ID:           /* Local APIC ID */
664                 apic_set_reg(apic, APIC_ID, val);
665                 break;
666
667         case APIC_TASKPRI:
668                 report_tpr_access(apic, true);
669                 apic_set_tpr(apic, val & 0xff);
670                 break;
671
672         case APIC_EOI:
673                 apic_set_eoi(apic);
674                 break;
675
676         case APIC_LDR:
677                 apic_set_reg(apic, APIC_LDR, val & APIC_LDR_MASK);
678                 break;
679
680         case APIC_DFR:
681                 apic_set_reg(apic, APIC_DFR, val | 0x0FFFFFFF);
682                 break;
683
684         case APIC_SPIV:
685                 apic_set_reg(apic, APIC_SPIV, val & 0x3ff);
686                 if (!(val & APIC_SPIV_APIC_ENABLED)) {
687                         int i;
688                         u32 lvt_val;
689
690                         for (i = 0; i < APIC_LVT_NUM; i++) {
691                                 lvt_val = apic_get_reg(apic,
692                                                        APIC_LVTT + 0x10 * i);
693                                 apic_set_reg(apic, APIC_LVTT + 0x10 * i,
694                                              lvt_val | APIC_LVT_MASKED);
695                         }
696                         atomic_set(&apic->lapic_timer.pending, 0);
697
698                 }
699                 break;
700
701         case APIC_ICR:
702                 /* No delay here, so we always clear the pending bit */
703                 apic_set_reg(apic, APIC_ICR, val & ~(1 << 12));
704                 apic_send_ipi(apic);
705                 break;
706
707         case APIC_ICR2:
708                 apic_set_reg(apic, APIC_ICR2, val & 0xff000000);
709                 break;
710
711         case APIC_LVT0:
712                 apic_manage_nmi_watchdog(apic, val);
713         case APIC_LVTT:
714         case APIC_LVTTHMR:
715         case APIC_LVTPC:
716         case APIC_LVT1:
717         case APIC_LVTERR:
718                 /* TODO: Check vector */
719                 if (!apic_sw_enabled(apic))
720                         val |= APIC_LVT_MASKED;
721
722                 val &= apic_lvt_mask[(offset - APIC_LVTT) >> 4];
723                 apic_set_reg(apic, offset, val);
724
725                 break;
726
727         case APIC_TMICT:
728                 hrtimer_cancel(&apic->lapic_timer.timer);
729                 apic_set_reg(apic, APIC_TMICT, val);
730                 start_apic_timer(apic);
731                 return;
732
733         case APIC_TDCR:
734                 if (val & 4)
735                         printk(KERN_ERR "KVM_WRITE:TDCR %x\n", val);
736                 apic_set_reg(apic, APIC_TDCR, val);
737                 update_divide_count(apic);
738                 break;
739
740         default:
741                 apic_debug("Local APIC Write to read-only register %x\n",
742                            offset);
743                 break;
744         }
745
746 }
747
748 static int apic_mmio_range(struct kvm_io_device *this, gpa_t addr,
749                            int len, int size)
750 {
751         struct kvm_lapic *apic = to_lapic(this);
752         int ret = 0;
753
754
755         if (apic_hw_enabled(apic) &&
756             (addr >= apic->base_address) &&
757             (addr < (apic->base_address + LAPIC_MMIO_LENGTH)))
758                 ret = 1;
759
760         return ret;
761 }
762
763 void kvm_free_lapic(struct kvm_vcpu *vcpu)
764 {
765         if (!vcpu->arch.apic)
766                 return;
767
768         hrtimer_cancel(&vcpu->arch.apic->lapic_timer.timer);
769
770         if (vcpu->arch.apic->regs_page)
771                 __free_page(vcpu->arch.apic->regs_page);
772
773         kfree(vcpu->arch.apic);
774 }
775
776 /*
777  *----------------------------------------------------------------------
778  * LAPIC interface
779  *----------------------------------------------------------------------
780  */
781
782 void kvm_lapic_set_tpr(struct kvm_vcpu *vcpu, unsigned long cr8)
783 {
784         struct kvm_lapic *apic = vcpu->arch.apic;
785
786         if (!apic)
787                 return;
788         apic_set_tpr(apic, ((cr8 & 0x0f) << 4)
789                      | (apic_get_reg(apic, APIC_TASKPRI) & 4));
790 }
791
792 u64 kvm_lapic_get_cr8(struct kvm_vcpu *vcpu)
793 {
794         struct kvm_lapic *apic = vcpu->arch.apic;
795         u64 tpr;
796
797         if (!apic)
798                 return 0;
799         tpr = (u64) apic_get_reg(apic, APIC_TASKPRI);
800
801         return (tpr & 0xf0) >> 4;
802 }
803
804 void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
805 {
806         struct kvm_lapic *apic = vcpu->arch.apic;
807
808         if (!apic) {
809                 value |= MSR_IA32_APICBASE_BSP;
810                 vcpu->arch.apic_base = value;
811                 return;
812         }
813
814         if (!kvm_vcpu_is_bsp(apic->vcpu))
815                 value &= ~MSR_IA32_APICBASE_BSP;
816
817         vcpu->arch.apic_base = value;
818         apic->base_address = apic->vcpu->arch.apic_base &
819                              MSR_IA32_APICBASE_BASE;
820
821         /* with FSB delivery interrupt, we can restart APIC functionality */
822         apic_debug("apic base msr is 0x%016" PRIx64 ", and base address is "
823                    "0x%lx.\n", apic->vcpu->arch.apic_base, apic->base_address);
824
825 }
826
827 void kvm_lapic_reset(struct kvm_vcpu *vcpu)
828 {
829         struct kvm_lapic *apic;
830         int i;
831
832         apic_debug("%s\n", __func__);
833
834         ASSERT(vcpu);
835         apic = vcpu->arch.apic;
836         ASSERT(apic != NULL);
837
838         /* Stop the timer in case it's a reset to an active apic */
839         hrtimer_cancel(&apic->lapic_timer.timer);
840
841         apic_set_reg(apic, APIC_ID, vcpu->vcpu_id << 24);
842         apic_set_reg(apic, APIC_LVR, APIC_VERSION);
843
844         for (i = 0; i < APIC_LVT_NUM; i++)
845                 apic_set_reg(apic, APIC_LVTT + 0x10 * i, APIC_LVT_MASKED);
846         apic_set_reg(apic, APIC_LVT0,
847                      SET_APIC_DELIVERY_MODE(0, APIC_MODE_EXTINT));
848
849         apic_set_reg(apic, APIC_DFR, 0xffffffffU);
850         apic_set_reg(apic, APIC_SPIV, 0xff);
851         apic_set_reg(apic, APIC_TASKPRI, 0);
852         apic_set_reg(apic, APIC_LDR, 0);
853         apic_set_reg(apic, APIC_ESR, 0);
854         apic_set_reg(apic, APIC_ICR, 0);
855         apic_set_reg(apic, APIC_ICR2, 0);
856         apic_set_reg(apic, APIC_TDCR, 0);
857         apic_set_reg(apic, APIC_TMICT, 0);
858         for (i = 0; i < 8; i++) {
859                 apic_set_reg(apic, APIC_IRR + 0x10 * i, 0);
860                 apic_set_reg(apic, APIC_ISR + 0x10 * i, 0);
861                 apic_set_reg(apic, APIC_TMR + 0x10 * i, 0);
862         }
863         apic->irr_pending = false;
864         update_divide_count(apic);
865         atomic_set(&apic->lapic_timer.pending, 0);
866         if (kvm_vcpu_is_bsp(vcpu))
867                 vcpu->arch.apic_base |= MSR_IA32_APICBASE_BSP;
868         apic_update_ppr(apic);
869
870         vcpu->arch.apic_arb_prio = 0;
871
872         apic_debug(KERN_INFO "%s: vcpu=%p, id=%d, base_msr="
873                    "0x%016" PRIx64 ", base_address=0x%0lx.\n", __func__,
874                    vcpu, kvm_apic_id(apic),
875                    vcpu->arch.apic_base, apic->base_address);
876 }
877
878 bool kvm_apic_present(struct kvm_vcpu *vcpu)
879 {
880         return vcpu->arch.apic && apic_hw_enabled(vcpu->arch.apic);
881 }
882
883 int kvm_lapic_enabled(struct kvm_vcpu *vcpu)
884 {
885         return kvm_apic_present(vcpu) && apic_sw_enabled(vcpu->arch.apic);
886 }
887
888 /*
889  *----------------------------------------------------------------------
890  * timer interface
891  *----------------------------------------------------------------------
892  */
893
894 static bool lapic_is_periodic(struct kvm_timer *ktimer)
895 {
896         struct kvm_lapic *apic = container_of(ktimer, struct kvm_lapic,
897                                               lapic_timer);
898         return apic_lvtt_period(apic);
899 }
900
901 int apic_has_pending_timer(struct kvm_vcpu *vcpu)
902 {
903         struct kvm_lapic *lapic = vcpu->arch.apic;
904
905         if (lapic && apic_enabled(lapic) && apic_lvt_enabled(lapic, APIC_LVTT))
906                 return atomic_read(&lapic->lapic_timer.pending);
907
908         return 0;
909 }
910
911 static int kvm_apic_local_deliver(struct kvm_lapic *apic, int lvt_type)
912 {
913         u32 reg = apic_get_reg(apic, lvt_type);
914         int vector, mode, trig_mode;
915
916         if (apic_hw_enabled(apic) && !(reg & APIC_LVT_MASKED)) {
917                 vector = reg & APIC_VECTOR_MASK;
918                 mode = reg & APIC_MODE_MASK;
919                 trig_mode = reg & APIC_LVT_LEVEL_TRIGGER;
920                 return __apic_accept_irq(apic, mode, vector, 1, trig_mode);
921         }
922         return 0;
923 }
924
925 void kvm_apic_nmi_wd_deliver(struct kvm_vcpu *vcpu)
926 {
927         struct kvm_lapic *apic = vcpu->arch.apic;
928
929         if (apic)
930                 kvm_apic_local_deliver(apic, APIC_LVT0);
931 }
932
933 static struct kvm_timer_ops lapic_timer_ops = {
934         .is_periodic = lapic_is_periodic,
935 };
936
937 static const struct kvm_io_device_ops apic_mmio_ops = {
938         .read     = apic_mmio_read,
939         .write    = apic_mmio_write,
940         .in_range = apic_mmio_range,
941 };
942
943 int kvm_create_lapic(struct kvm_vcpu *vcpu)
944 {
945         struct kvm_lapic *apic;
946
947         ASSERT(vcpu != NULL);
948         apic_debug("apic_init %d\n", vcpu->vcpu_id);
949
950         apic = kzalloc(sizeof(*apic), GFP_KERNEL);
951         if (!apic)
952                 goto nomem;
953
954         vcpu->arch.apic = apic;
955
956         apic->regs_page = alloc_page(GFP_KERNEL);
957         if (apic->regs_page == NULL) {
958                 printk(KERN_ERR "malloc apic regs error for vcpu %x\n",
959                        vcpu->vcpu_id);
960                 goto nomem_free_apic;
961         }
962         apic->regs = page_address(apic->regs_page);
963         memset(apic->regs, 0, PAGE_SIZE);
964         apic->vcpu = vcpu;
965
966         hrtimer_init(&apic->lapic_timer.timer, CLOCK_MONOTONIC,
967                      HRTIMER_MODE_ABS);
968         apic->lapic_timer.timer.function = kvm_timer_fn;
969         apic->lapic_timer.t_ops = &lapic_timer_ops;
970         apic->lapic_timer.kvm = vcpu->kvm;
971         apic->lapic_timer.vcpu = vcpu;
972
973         apic->base_address = APIC_DEFAULT_PHYS_BASE;
974         vcpu->arch.apic_base = APIC_DEFAULT_PHYS_BASE;
975
976         kvm_lapic_reset(vcpu);
977         kvm_iodevice_init(&apic->dev, &apic_mmio_ops);
978
979         return 0;
980 nomem_free_apic:
981         kfree(apic);
982 nomem:
983         return -ENOMEM;
984 }
985
986 int kvm_apic_has_interrupt(struct kvm_vcpu *vcpu)
987 {
988         struct kvm_lapic *apic = vcpu->arch.apic;
989         int highest_irr;
990
991         if (!apic || !apic_enabled(apic))
992                 return -1;
993
994         apic_update_ppr(apic);
995         highest_irr = apic_find_highest_irr(apic);
996         if ((highest_irr == -1) ||
997             ((highest_irr & 0xF0) <= apic_get_reg(apic, APIC_PROCPRI)))
998                 return -1;
999         return highest_irr;
1000 }
1001
1002 int kvm_apic_accept_pic_intr(struct kvm_vcpu *vcpu)
1003 {
1004         u32 lvt0 = apic_get_reg(vcpu->arch.apic, APIC_LVT0);
1005         int r = 0;
1006
1007         if (kvm_vcpu_is_bsp(vcpu)) {
1008                 if (!apic_hw_enabled(vcpu->arch.apic))
1009                         r = 1;
1010                 if ((lvt0 & APIC_LVT_MASKED) == 0 &&
1011                     GET_APIC_DELIVERY_MODE(lvt0) == APIC_MODE_EXTINT)
1012                         r = 1;
1013         }
1014         return r;
1015 }
1016
1017 void kvm_inject_apic_timer_irqs(struct kvm_vcpu *vcpu)
1018 {
1019         struct kvm_lapic *apic = vcpu->arch.apic;
1020
1021         if (apic && atomic_read(&apic->lapic_timer.pending) > 0) {
1022                 if (kvm_apic_local_deliver(apic, APIC_LVTT))
1023                         atomic_dec(&apic->lapic_timer.pending);
1024         }
1025 }
1026
1027 int kvm_get_apic_interrupt(struct kvm_vcpu *vcpu)
1028 {
1029         int vector = kvm_apic_has_interrupt(vcpu);
1030         struct kvm_lapic *apic = vcpu->arch.apic;
1031
1032         if (vector == -1)
1033                 return -1;
1034
1035         apic_set_vector(vector, apic->regs + APIC_ISR);
1036         apic_update_ppr(apic);
1037         apic_clear_irr(vector, apic);
1038         return vector;
1039 }
1040
1041 void kvm_apic_post_state_restore(struct kvm_vcpu *vcpu)
1042 {
1043         struct kvm_lapic *apic = vcpu->arch.apic;
1044
1045         apic->base_address = vcpu->arch.apic_base &
1046                              MSR_IA32_APICBASE_BASE;
1047         apic_set_reg(apic, APIC_LVR, APIC_VERSION);
1048         apic_update_ppr(apic);
1049         hrtimer_cancel(&apic->lapic_timer.timer);
1050         update_divide_count(apic);
1051         start_apic_timer(apic);
1052 }
1053
1054 void __kvm_migrate_apic_timer(struct kvm_vcpu *vcpu)
1055 {
1056         struct kvm_lapic *apic = vcpu->arch.apic;
1057         struct hrtimer *timer;
1058
1059         if (!apic)
1060                 return;
1061
1062         timer = &apic->lapic_timer.timer;
1063         if (hrtimer_cancel(timer))
1064                 hrtimer_start_expires(timer, HRTIMER_MODE_ABS);
1065 }
1066
1067 void kvm_lapic_sync_from_vapic(struct kvm_vcpu *vcpu)
1068 {
1069         u32 data;
1070         void *vapic;
1071
1072         if (!irqchip_in_kernel(vcpu->kvm) || !vcpu->arch.apic->vapic_addr)
1073                 return;
1074
1075         vapic = kmap_atomic(vcpu->arch.apic->vapic_page, KM_USER0);
1076         data = *(u32 *)(vapic + offset_in_page(vcpu->arch.apic->vapic_addr));
1077         kunmap_atomic(vapic, KM_USER0);
1078
1079         apic_set_tpr(vcpu->arch.apic, data & 0xff);
1080 }
1081
1082 void kvm_lapic_sync_to_vapic(struct kvm_vcpu *vcpu)
1083 {
1084         u32 data, tpr;
1085         int max_irr, max_isr;
1086         struct kvm_lapic *apic;
1087         void *vapic;
1088
1089         if (!irqchip_in_kernel(vcpu->kvm) || !vcpu->arch.apic->vapic_addr)
1090                 return;
1091
1092         apic = vcpu->arch.apic;
1093         tpr = apic_get_reg(apic, APIC_TASKPRI) & 0xff;
1094         max_irr = apic_find_highest_irr(apic);
1095         if (max_irr < 0)
1096                 max_irr = 0;
1097         max_isr = apic_find_highest_isr(apic);
1098         if (max_isr < 0)
1099                 max_isr = 0;
1100         data = (tpr & 0xff) | ((max_isr & 0xf0) << 8) | (max_irr << 24);
1101
1102         vapic = kmap_atomic(vcpu->arch.apic->vapic_page, KM_USER0);
1103         *(u32 *)(vapic + offset_in_page(vcpu->arch.apic->vapic_addr)) = data;
1104         kunmap_atomic(vapic, KM_USER0);
1105 }
1106
1107 void kvm_lapic_set_vapic_addr(struct kvm_vcpu *vcpu, gpa_t vapic_addr)
1108 {
1109         if (!irqchip_in_kernel(vcpu->kvm))
1110                 return;
1111
1112         vcpu->arch.apic->vapic_addr = vapic_addr;
1113 }