x86, apicv: add virtual interrupt delivery support
[linux-3.10.git] / arch / x86 / kvm / lapic.c
1
2 /*
3  * Local APIC virtualization
4  *
5  * Copyright (C) 2006 Qumranet, Inc.
6  * Copyright (C) 2007 Novell
7  * Copyright (C) 2007 Intel
8  * Copyright 2009 Red Hat, Inc. and/or its affiliates.
9  *
10  * Authors:
11  *   Dor Laor <dor.laor@qumranet.com>
12  *   Gregory Haskins <ghaskins@novell.com>
13  *   Yaozu (Eddie) Dong <eddie.dong@intel.com>
14  *
15  * Based on Xen 3.1 code, Copyright (c) 2004, Intel Corporation.
16  *
17  * This work is licensed under the terms of the GNU GPL, version 2.  See
18  * the COPYING file in the top-level directory.
19  */
20
21 #include <linux/kvm_host.h>
22 #include <linux/kvm.h>
23 #include <linux/mm.h>
24 #include <linux/highmem.h>
25 #include <linux/smp.h>
26 #include <linux/hrtimer.h>
27 #include <linux/io.h>
28 #include <linux/module.h>
29 #include <linux/math64.h>
30 #include <linux/slab.h>
31 #include <asm/processor.h>
32 #include <asm/msr.h>
33 #include <asm/page.h>
34 #include <asm/current.h>
35 #include <asm/apicdef.h>
36 #include <linux/atomic.h>
37 #include <linux/jump_label.h>
38 #include "kvm_cache_regs.h"
39 #include "irq.h"
40 #include "trace.h"
41 #include "x86.h"
42 #include "cpuid.h"
43
44 #ifndef CONFIG_X86_64
45 #define mod_64(x, y) ((x) - (y) * div64_u64(x, y))
46 #else
47 #define mod_64(x, y) ((x) % (y))
48 #endif
49
50 #define PRId64 "d"
51 #define PRIx64 "llx"
52 #define PRIu64 "u"
53 #define PRIo64 "o"
54
55 #define APIC_BUS_CYCLE_NS 1
56
57 /* #define apic_debug(fmt,arg...) printk(KERN_WARNING fmt,##arg) */
58 #define apic_debug(fmt, arg...)
59
60 #define APIC_LVT_NUM                    6
61 /* 14 is the version for Xeon and Pentium 8.4.8*/
62 #define APIC_VERSION                    (0x14UL | ((APIC_LVT_NUM - 1) << 16))
63 #define LAPIC_MMIO_LENGTH               (1 << 12)
64 /* followed define is not in apicdef.h */
65 #define APIC_SHORT_MASK                 0xc0000
66 #define APIC_DEST_NOSHORT               0x0
67 #define APIC_DEST_MASK                  0x800
68 #define MAX_APIC_VECTOR                 256
69 #define APIC_VECTORS_PER_REG            32
70
71 #define VEC_POS(v) ((v) & (32 - 1))
72 #define REG_POS(v) (((v) >> 5) << 4)
73
74 static unsigned int min_timer_period_us = 500;
75 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
76
77 static inline void apic_set_reg(struct kvm_lapic *apic, int reg_off, u32 val)
78 {
79         *((u32 *) (apic->regs + reg_off)) = val;
80 }
81
82 static inline int apic_test_and_set_vector(int vec, void *bitmap)
83 {
84         return test_and_set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
85 }
86
87 static inline int apic_test_and_clear_vector(int vec, void *bitmap)
88 {
89         return test_and_clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
90 }
91
92 static inline int apic_test_vector(int vec, void *bitmap)
93 {
94         return test_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
95 }
96
97 static inline void apic_set_vector(int vec, void *bitmap)
98 {
99         set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
100 }
101
102 static inline void apic_clear_vector(int vec, void *bitmap)
103 {
104         clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
105 }
106
107 static inline int __apic_test_and_set_vector(int vec, void *bitmap)
108 {
109         return __test_and_set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
110 }
111
112 static inline int __apic_test_and_clear_vector(int vec, void *bitmap)
113 {
114         return __test_and_clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
115 }
116
117 struct static_key_deferred apic_hw_disabled __read_mostly;
118 struct static_key_deferred apic_sw_disabled __read_mostly;
119
120 static inline void apic_set_spiv(struct kvm_lapic *apic, u32 val)
121 {
122         if ((kvm_apic_get_reg(apic, APIC_SPIV) ^ val) & APIC_SPIV_APIC_ENABLED) {
123                 if (val & APIC_SPIV_APIC_ENABLED)
124                         static_key_slow_dec_deferred(&apic_sw_disabled);
125                 else
126                         static_key_slow_inc(&apic_sw_disabled.key);
127         }
128         apic_set_reg(apic, APIC_SPIV, val);
129 }
130
131 static inline int apic_enabled(struct kvm_lapic *apic)
132 {
133         return kvm_apic_sw_enabled(apic) &&     kvm_apic_hw_enabled(apic);
134 }
135
136 #define LVT_MASK        \
137         (APIC_LVT_MASKED | APIC_SEND_PENDING | APIC_VECTOR_MASK)
138
139 #define LINT_MASK       \
140         (LVT_MASK | APIC_MODE_MASK | APIC_INPUT_POLARITY | \
141          APIC_LVT_REMOTE_IRR | APIC_LVT_LEVEL_TRIGGER)
142
143 static inline int kvm_apic_id(struct kvm_lapic *apic)
144 {
145         return (kvm_apic_get_reg(apic, APIC_ID) >> 24) & 0xff;
146 }
147
148 void kvm_calculate_eoi_exitmap(struct kvm_vcpu *vcpu,
149                                 struct kvm_lapic_irq *irq,
150                                 u64 *eoi_exit_bitmap)
151 {
152         struct kvm_lapic **dst;
153         struct kvm_apic_map *map;
154         unsigned long bitmap = 1;
155         int i;
156
157         rcu_read_lock();
158         map = rcu_dereference(vcpu->kvm->arch.apic_map);
159
160         if (unlikely(!map)) {
161                 __set_bit(irq->vector, (unsigned long *)eoi_exit_bitmap);
162                 goto out;
163         }
164
165         if (irq->dest_mode == 0) { /* physical mode */
166                 if (irq->delivery_mode == APIC_DM_LOWEST ||
167                                 irq->dest_id == 0xff) {
168                         __set_bit(irq->vector,
169                                   (unsigned long *)eoi_exit_bitmap);
170                         goto out;
171                 }
172                 dst = &map->phys_map[irq->dest_id & 0xff];
173         } else {
174                 u32 mda = irq->dest_id << (32 - map->ldr_bits);
175
176                 dst = map->logical_map[apic_cluster_id(map, mda)];
177
178                 bitmap = apic_logical_id(map, mda);
179         }
180
181         for_each_set_bit(i, &bitmap, 16) {
182                 if (!dst[i])
183                         continue;
184                 if (dst[i]->vcpu == vcpu) {
185                         __set_bit(irq->vector,
186                                   (unsigned long *)eoi_exit_bitmap);
187                         break;
188                 }
189         }
190
191 out:
192         rcu_read_unlock();
193 }
194
195 static void recalculate_apic_map(struct kvm *kvm)
196 {
197         struct kvm_apic_map *new, *old = NULL;
198         struct kvm_vcpu *vcpu;
199         int i;
200
201         new = kzalloc(sizeof(struct kvm_apic_map), GFP_KERNEL);
202
203         mutex_lock(&kvm->arch.apic_map_lock);
204
205         if (!new)
206                 goto out;
207
208         new->ldr_bits = 8;
209         /* flat mode is default */
210         new->cid_shift = 8;
211         new->cid_mask = 0;
212         new->lid_mask = 0xff;
213
214         kvm_for_each_vcpu(i, vcpu, kvm) {
215                 struct kvm_lapic *apic = vcpu->arch.apic;
216                 u16 cid, lid;
217                 u32 ldr;
218
219                 if (!kvm_apic_present(vcpu))
220                         continue;
221
222                 /*
223                  * All APICs have to be configured in the same mode by an OS.
224                  * We take advatage of this while building logical id loockup
225                  * table. After reset APICs are in xapic/flat mode, so if we
226                  * find apic with different setting we assume this is the mode
227                  * OS wants all apics to be in; build lookup table accordingly.
228                  */
229                 if (apic_x2apic_mode(apic)) {
230                         new->ldr_bits = 32;
231                         new->cid_shift = 16;
232                         new->cid_mask = new->lid_mask = 0xffff;
233                 } else if (kvm_apic_sw_enabled(apic) &&
234                                 !new->cid_mask /* flat mode */ &&
235                                 kvm_apic_get_reg(apic, APIC_DFR) == APIC_DFR_CLUSTER) {
236                         new->cid_shift = 4;
237                         new->cid_mask = 0xf;
238                         new->lid_mask = 0xf;
239                 }
240
241                 new->phys_map[kvm_apic_id(apic)] = apic;
242
243                 ldr = kvm_apic_get_reg(apic, APIC_LDR);
244                 cid = apic_cluster_id(new, ldr);
245                 lid = apic_logical_id(new, ldr);
246
247                 if (lid)
248                         new->logical_map[cid][ffs(lid) - 1] = apic;
249         }
250 out:
251         old = rcu_dereference_protected(kvm->arch.apic_map,
252                         lockdep_is_held(&kvm->arch.apic_map_lock));
253         rcu_assign_pointer(kvm->arch.apic_map, new);
254         mutex_unlock(&kvm->arch.apic_map_lock);
255
256         if (old)
257                 kfree_rcu(old, rcu);
258
259         kvm_ioapic_make_eoibitmap_request(kvm);
260 }
261
262 static inline void kvm_apic_set_id(struct kvm_lapic *apic, u8 id)
263 {
264         apic_set_reg(apic, APIC_ID, id << 24);
265         recalculate_apic_map(apic->vcpu->kvm);
266 }
267
268 static inline void kvm_apic_set_ldr(struct kvm_lapic *apic, u32 id)
269 {
270         apic_set_reg(apic, APIC_LDR, id);
271         recalculate_apic_map(apic->vcpu->kvm);
272 }
273
274 static inline int apic_lvt_enabled(struct kvm_lapic *apic, int lvt_type)
275 {
276         return !(kvm_apic_get_reg(apic, lvt_type) & APIC_LVT_MASKED);
277 }
278
279 static inline int apic_lvt_vector(struct kvm_lapic *apic, int lvt_type)
280 {
281         return kvm_apic_get_reg(apic, lvt_type) & APIC_VECTOR_MASK;
282 }
283
284 static inline int apic_lvtt_oneshot(struct kvm_lapic *apic)
285 {
286         return ((kvm_apic_get_reg(apic, APIC_LVTT) &
287                 apic->lapic_timer.timer_mode_mask) == APIC_LVT_TIMER_ONESHOT);
288 }
289
290 static inline int apic_lvtt_period(struct kvm_lapic *apic)
291 {
292         return ((kvm_apic_get_reg(apic, APIC_LVTT) &
293                 apic->lapic_timer.timer_mode_mask) == APIC_LVT_TIMER_PERIODIC);
294 }
295
296 static inline int apic_lvtt_tscdeadline(struct kvm_lapic *apic)
297 {
298         return ((kvm_apic_get_reg(apic, APIC_LVTT) &
299                 apic->lapic_timer.timer_mode_mask) ==
300                         APIC_LVT_TIMER_TSCDEADLINE);
301 }
302
303 static inline int apic_lvt_nmi_mode(u32 lvt_val)
304 {
305         return (lvt_val & (APIC_MODE_MASK | APIC_LVT_MASKED)) == APIC_DM_NMI;
306 }
307
308 void kvm_apic_set_version(struct kvm_vcpu *vcpu)
309 {
310         struct kvm_lapic *apic = vcpu->arch.apic;
311         struct kvm_cpuid_entry2 *feat;
312         u32 v = APIC_VERSION;
313
314         if (!kvm_vcpu_has_lapic(vcpu))
315                 return;
316
317         feat = kvm_find_cpuid_entry(apic->vcpu, 0x1, 0);
318         if (feat && (feat->ecx & (1 << (X86_FEATURE_X2APIC & 31))))
319                 v |= APIC_LVR_DIRECTED_EOI;
320         apic_set_reg(apic, APIC_LVR, v);
321 }
322
323 static const unsigned int apic_lvt_mask[APIC_LVT_NUM] = {
324         LVT_MASK ,      /* part LVTT mask, timer mode mask added at runtime */
325         LVT_MASK | APIC_MODE_MASK,      /* LVTTHMR */
326         LVT_MASK | APIC_MODE_MASK,      /* LVTPC */
327         LINT_MASK, LINT_MASK,   /* LVT0-1 */
328         LVT_MASK                /* LVTERR */
329 };
330
331 static int find_highest_vector(void *bitmap)
332 {
333         int vec;
334         u32 *reg;
335
336         for (vec = MAX_APIC_VECTOR - APIC_VECTORS_PER_REG;
337              vec >= 0; vec -= APIC_VECTORS_PER_REG) {
338                 reg = bitmap + REG_POS(vec);
339                 if (*reg)
340                         return fls(*reg) - 1 + vec;
341         }
342
343         return -1;
344 }
345
346 static u8 count_vectors(void *bitmap)
347 {
348         int vec;
349         u32 *reg;
350         u8 count = 0;
351
352         for (vec = 0; vec < MAX_APIC_VECTOR; vec += APIC_VECTORS_PER_REG) {
353                 reg = bitmap + REG_POS(vec);
354                 count += hweight32(*reg);
355         }
356
357         return count;
358 }
359
360 static inline int apic_test_and_set_irr(int vec, struct kvm_lapic *apic)
361 {
362         apic->irr_pending = true;
363         return apic_test_and_set_vector(vec, apic->regs + APIC_IRR);
364 }
365
366 static inline int apic_search_irr(struct kvm_lapic *apic)
367 {
368         return find_highest_vector(apic->regs + APIC_IRR);
369 }
370
371 static inline int apic_find_highest_irr(struct kvm_lapic *apic)
372 {
373         int result;
374
375         /*
376          * Note that irr_pending is just a hint. It will be always
377          * true with virtual interrupt delivery enabled.
378          */
379         if (!apic->irr_pending)
380                 return -1;
381
382         result = apic_search_irr(apic);
383         ASSERT(result == -1 || result >= 16);
384
385         return result;
386 }
387
388 static inline void apic_clear_irr(int vec, struct kvm_lapic *apic)
389 {
390         apic->irr_pending = false;
391         apic_clear_vector(vec, apic->regs + APIC_IRR);
392         if (apic_search_irr(apic) != -1)
393                 apic->irr_pending = true;
394 }
395
396 static inline void apic_set_isr(int vec, struct kvm_lapic *apic)
397 {
398         if (!__apic_test_and_set_vector(vec, apic->regs + APIC_ISR))
399                 ++apic->isr_count;
400         BUG_ON(apic->isr_count > MAX_APIC_VECTOR);
401         /*
402          * ISR (in service register) bit is set when injecting an interrupt.
403          * The highest vector is injected. Thus the latest bit set matches
404          * the highest bit in ISR.
405          */
406         apic->highest_isr_cache = vec;
407 }
408
409 static inline void apic_clear_isr(int vec, struct kvm_lapic *apic)
410 {
411         if (__apic_test_and_clear_vector(vec, apic->regs + APIC_ISR))
412                 --apic->isr_count;
413         BUG_ON(apic->isr_count < 0);
414         apic->highest_isr_cache = -1;
415 }
416
417 int kvm_lapic_find_highest_irr(struct kvm_vcpu *vcpu)
418 {
419         int highest_irr;
420
421         /* This may race with setting of irr in __apic_accept_irq() and
422          * value returned may be wrong, but kvm_vcpu_kick() in __apic_accept_irq
423          * will cause vmexit immediately and the value will be recalculated
424          * on the next vmentry.
425          */
426         if (!kvm_vcpu_has_lapic(vcpu))
427                 return 0;
428         highest_irr = apic_find_highest_irr(vcpu->arch.apic);
429
430         return highest_irr;
431 }
432
433 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
434                              int vector, int level, int trig_mode);
435
436 int kvm_apic_set_irq(struct kvm_vcpu *vcpu, struct kvm_lapic_irq *irq)
437 {
438         struct kvm_lapic *apic = vcpu->arch.apic;
439
440         return __apic_accept_irq(apic, irq->delivery_mode, irq->vector,
441                         irq->level, irq->trig_mode);
442 }
443
444 static int pv_eoi_put_user(struct kvm_vcpu *vcpu, u8 val)
445 {
446
447         return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, &val,
448                                       sizeof(val));
449 }
450
451 static int pv_eoi_get_user(struct kvm_vcpu *vcpu, u8 *val)
452 {
453
454         return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, val,
455                                       sizeof(*val));
456 }
457
458 static inline bool pv_eoi_enabled(struct kvm_vcpu *vcpu)
459 {
460         return vcpu->arch.pv_eoi.msr_val & KVM_MSR_ENABLED;
461 }
462
463 static bool pv_eoi_get_pending(struct kvm_vcpu *vcpu)
464 {
465         u8 val;
466         if (pv_eoi_get_user(vcpu, &val) < 0)
467                 apic_debug("Can't read EOI MSR value: 0x%llx\n",
468                            (unsigned long long)vcpi->arch.pv_eoi.msr_val);
469         return val & 0x1;
470 }
471
472 static void pv_eoi_set_pending(struct kvm_vcpu *vcpu)
473 {
474         if (pv_eoi_put_user(vcpu, KVM_PV_EOI_ENABLED) < 0) {
475                 apic_debug("Can't set EOI MSR value: 0x%llx\n",
476                            (unsigned long long)vcpi->arch.pv_eoi.msr_val);
477                 return;
478         }
479         __set_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
480 }
481
482 static void pv_eoi_clr_pending(struct kvm_vcpu *vcpu)
483 {
484         if (pv_eoi_put_user(vcpu, KVM_PV_EOI_DISABLED) < 0) {
485                 apic_debug("Can't clear EOI MSR value: 0x%llx\n",
486                            (unsigned long long)vcpi->arch.pv_eoi.msr_val);
487                 return;
488         }
489         __clear_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
490 }
491
492 static inline int apic_find_highest_isr(struct kvm_lapic *apic)
493 {
494         int result;
495
496         /* Note that isr_count is always 1 with vid enabled */
497         if (!apic->isr_count)
498                 return -1;
499         if (likely(apic->highest_isr_cache != -1))
500                 return apic->highest_isr_cache;
501
502         result = find_highest_vector(apic->regs + APIC_ISR);
503         ASSERT(result == -1 || result >= 16);
504
505         return result;
506 }
507
508 static void apic_update_ppr(struct kvm_lapic *apic)
509 {
510         u32 tpr, isrv, ppr, old_ppr;
511         int isr;
512
513         old_ppr = kvm_apic_get_reg(apic, APIC_PROCPRI);
514         tpr = kvm_apic_get_reg(apic, APIC_TASKPRI);
515         isr = apic_find_highest_isr(apic);
516         isrv = (isr != -1) ? isr : 0;
517
518         if ((tpr & 0xf0) >= (isrv & 0xf0))
519                 ppr = tpr & 0xff;
520         else
521                 ppr = isrv & 0xf0;
522
523         apic_debug("vlapic %p, ppr 0x%x, isr 0x%x, isrv 0x%x",
524                    apic, ppr, isr, isrv);
525
526         if (old_ppr != ppr) {
527                 apic_set_reg(apic, APIC_PROCPRI, ppr);
528                 if (ppr < old_ppr)
529                         kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
530         }
531 }
532
533 static void apic_set_tpr(struct kvm_lapic *apic, u32 tpr)
534 {
535         apic_set_reg(apic, APIC_TASKPRI, tpr);
536         apic_update_ppr(apic);
537 }
538
539 int kvm_apic_match_physical_addr(struct kvm_lapic *apic, u16 dest)
540 {
541         return dest == 0xff || kvm_apic_id(apic) == dest;
542 }
543
544 int kvm_apic_match_logical_addr(struct kvm_lapic *apic, u8 mda)
545 {
546         int result = 0;
547         u32 logical_id;
548
549         if (apic_x2apic_mode(apic)) {
550                 logical_id = kvm_apic_get_reg(apic, APIC_LDR);
551                 return logical_id & mda;
552         }
553
554         logical_id = GET_APIC_LOGICAL_ID(kvm_apic_get_reg(apic, APIC_LDR));
555
556         switch (kvm_apic_get_reg(apic, APIC_DFR)) {
557         case APIC_DFR_FLAT:
558                 if (logical_id & mda)
559                         result = 1;
560                 break;
561         case APIC_DFR_CLUSTER:
562                 if (((logical_id >> 4) == (mda >> 0x4))
563                     && (logical_id & mda & 0xf))
564                         result = 1;
565                 break;
566         default:
567                 apic_debug("Bad DFR vcpu %d: %08x\n",
568                            apic->vcpu->vcpu_id, kvm_apic_get_reg(apic, APIC_DFR));
569                 break;
570         }
571
572         return result;
573 }
574
575 int kvm_apic_match_dest(struct kvm_vcpu *vcpu, struct kvm_lapic *source,
576                            int short_hand, int dest, int dest_mode)
577 {
578         int result = 0;
579         struct kvm_lapic *target = vcpu->arch.apic;
580
581         apic_debug("target %p, source %p, dest 0x%x, "
582                    "dest_mode 0x%x, short_hand 0x%x\n",
583                    target, source, dest, dest_mode, short_hand);
584
585         ASSERT(target);
586         switch (short_hand) {
587         case APIC_DEST_NOSHORT:
588                 if (dest_mode == 0)
589                         /* Physical mode. */
590                         result = kvm_apic_match_physical_addr(target, dest);
591                 else
592                         /* Logical mode. */
593                         result = kvm_apic_match_logical_addr(target, dest);
594                 break;
595         case APIC_DEST_SELF:
596                 result = (target == source);
597                 break;
598         case APIC_DEST_ALLINC:
599                 result = 1;
600                 break;
601         case APIC_DEST_ALLBUT:
602                 result = (target != source);
603                 break;
604         default:
605                 apic_debug("kvm: apic: Bad dest shorthand value %x\n",
606                            short_hand);
607                 break;
608         }
609
610         return result;
611 }
612
613 bool kvm_irq_delivery_to_apic_fast(struct kvm *kvm, struct kvm_lapic *src,
614                 struct kvm_lapic_irq *irq, int *r)
615 {
616         struct kvm_apic_map *map;
617         unsigned long bitmap = 1;
618         struct kvm_lapic **dst;
619         int i;
620         bool ret = false;
621
622         *r = -1;
623
624         if (irq->shorthand == APIC_DEST_SELF) {
625                 *r = kvm_apic_set_irq(src->vcpu, irq);
626                 return true;
627         }
628
629         if (irq->shorthand)
630                 return false;
631
632         rcu_read_lock();
633         map = rcu_dereference(kvm->arch.apic_map);
634
635         if (!map)
636                 goto out;
637
638         if (irq->dest_mode == 0) { /* physical mode */
639                 if (irq->delivery_mode == APIC_DM_LOWEST ||
640                                 irq->dest_id == 0xff)
641                         goto out;
642                 dst = &map->phys_map[irq->dest_id & 0xff];
643         } else {
644                 u32 mda = irq->dest_id << (32 - map->ldr_bits);
645
646                 dst = map->logical_map[apic_cluster_id(map, mda)];
647
648                 bitmap = apic_logical_id(map, mda);
649
650                 if (irq->delivery_mode == APIC_DM_LOWEST) {
651                         int l = -1;
652                         for_each_set_bit(i, &bitmap, 16) {
653                                 if (!dst[i])
654                                         continue;
655                                 if (l < 0)
656                                         l = i;
657                                 else if (kvm_apic_compare_prio(dst[i]->vcpu, dst[l]->vcpu) < 0)
658                                         l = i;
659                         }
660
661                         bitmap = (l >= 0) ? 1 << l : 0;
662                 }
663         }
664
665         for_each_set_bit(i, &bitmap, 16) {
666                 if (!dst[i])
667                         continue;
668                 if (*r < 0)
669                         *r = 0;
670                 *r += kvm_apic_set_irq(dst[i]->vcpu, irq);
671         }
672
673         ret = true;
674 out:
675         rcu_read_unlock();
676         return ret;
677 }
678
679 /*
680  * Add a pending IRQ into lapic.
681  * Return 1 if successfully added and 0 if discarded.
682  */
683 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
684                              int vector, int level, int trig_mode)
685 {
686         int result = 0;
687         struct kvm_vcpu *vcpu = apic->vcpu;
688
689         switch (delivery_mode) {
690         case APIC_DM_LOWEST:
691                 vcpu->arch.apic_arb_prio++;
692         case APIC_DM_FIXED:
693                 /* FIXME add logic for vcpu on reset */
694                 if (unlikely(!apic_enabled(apic)))
695                         break;
696
697                 if (trig_mode) {
698                         apic_debug("level trig mode for vector %d", vector);
699                         apic_set_vector(vector, apic->regs + APIC_TMR);
700                 } else
701                         apic_clear_vector(vector, apic->regs + APIC_TMR);
702
703                 result = !apic_test_and_set_irr(vector, apic);
704                 trace_kvm_apic_accept_irq(vcpu->vcpu_id, delivery_mode,
705                                           trig_mode, vector, !result);
706                 if (!result) {
707                         if (trig_mode)
708                                 apic_debug("level trig mode repeatedly for "
709                                                 "vector %d", vector);
710                         break;
711                 }
712
713                 kvm_make_request(KVM_REQ_EVENT, vcpu);
714                 kvm_vcpu_kick(vcpu);
715                 break;
716
717         case APIC_DM_REMRD:
718                 apic_debug("Ignoring delivery mode 3\n");
719                 break;
720
721         case APIC_DM_SMI:
722                 apic_debug("Ignoring guest SMI\n");
723                 break;
724
725         case APIC_DM_NMI:
726                 result = 1;
727                 kvm_inject_nmi(vcpu);
728                 kvm_vcpu_kick(vcpu);
729                 break;
730
731         case APIC_DM_INIT:
732                 if (!trig_mode || level) {
733                         result = 1;
734                         vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
735                         kvm_make_request(KVM_REQ_EVENT, vcpu);
736                         kvm_vcpu_kick(vcpu);
737                 } else {
738                         apic_debug("Ignoring de-assert INIT to vcpu %d\n",
739                                    vcpu->vcpu_id);
740                 }
741                 break;
742
743         case APIC_DM_STARTUP:
744                 apic_debug("SIPI to vcpu %d vector 0x%02x\n",
745                            vcpu->vcpu_id, vector);
746                 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
747                         result = 1;
748                         vcpu->arch.sipi_vector = vector;
749                         vcpu->arch.mp_state = KVM_MP_STATE_SIPI_RECEIVED;
750                         kvm_make_request(KVM_REQ_EVENT, vcpu);
751                         kvm_vcpu_kick(vcpu);
752                 }
753                 break;
754
755         case APIC_DM_EXTINT:
756                 /*
757                  * Should only be called by kvm_apic_local_deliver() with LVT0,
758                  * before NMI watchdog was enabled. Already handled by
759                  * kvm_apic_accept_pic_intr().
760                  */
761                 break;
762
763         default:
764                 printk(KERN_ERR "TODO: unsupported delivery mode %x\n",
765                        delivery_mode);
766                 break;
767         }
768         return result;
769 }
770
771 int kvm_apic_compare_prio(struct kvm_vcpu *vcpu1, struct kvm_vcpu *vcpu2)
772 {
773         return vcpu1->arch.apic_arb_prio - vcpu2->arch.apic_arb_prio;
774 }
775
776 static void kvm_ioapic_send_eoi(struct kvm_lapic *apic, int vector)
777 {
778         if (!(kvm_apic_get_reg(apic, APIC_SPIV) & APIC_SPIV_DIRECTED_EOI) &&
779             kvm_ioapic_handles_vector(apic->vcpu->kvm, vector)) {
780                 int trigger_mode;
781                 if (apic_test_vector(vector, apic->regs + APIC_TMR))
782                         trigger_mode = IOAPIC_LEVEL_TRIG;
783                 else
784                         trigger_mode = IOAPIC_EDGE_TRIG;
785                 kvm_ioapic_update_eoi(apic->vcpu->kvm, vector, trigger_mode);
786         }
787 }
788
789 static int apic_set_eoi(struct kvm_lapic *apic)
790 {
791         int vector = apic_find_highest_isr(apic);
792
793         trace_kvm_eoi(apic, vector);
794
795         /*
796          * Not every write EOI will has corresponding ISR,
797          * one example is when Kernel check timer on setup_IO_APIC
798          */
799         if (vector == -1)
800                 return vector;
801
802         apic_clear_isr(vector, apic);
803         apic_update_ppr(apic);
804
805         kvm_ioapic_send_eoi(apic, vector);
806         kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
807         return vector;
808 }
809
810 /*
811  * this interface assumes a trap-like exit, which has already finished
812  * desired side effect including vISR and vPPR update.
813  */
814 void kvm_apic_set_eoi_accelerated(struct kvm_vcpu *vcpu, int vector)
815 {
816         struct kvm_lapic *apic = vcpu->arch.apic;
817
818         trace_kvm_eoi(apic, vector);
819
820         kvm_ioapic_send_eoi(apic, vector);
821         kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
822 }
823 EXPORT_SYMBOL_GPL(kvm_apic_set_eoi_accelerated);
824
825 static void apic_send_ipi(struct kvm_lapic *apic)
826 {
827         u32 icr_low = kvm_apic_get_reg(apic, APIC_ICR);
828         u32 icr_high = kvm_apic_get_reg(apic, APIC_ICR2);
829         struct kvm_lapic_irq irq;
830
831         irq.vector = icr_low & APIC_VECTOR_MASK;
832         irq.delivery_mode = icr_low & APIC_MODE_MASK;
833         irq.dest_mode = icr_low & APIC_DEST_MASK;
834         irq.level = icr_low & APIC_INT_ASSERT;
835         irq.trig_mode = icr_low & APIC_INT_LEVELTRIG;
836         irq.shorthand = icr_low & APIC_SHORT_MASK;
837         if (apic_x2apic_mode(apic))
838                 irq.dest_id = icr_high;
839         else
840                 irq.dest_id = GET_APIC_DEST_FIELD(icr_high);
841
842         trace_kvm_apic_ipi(icr_low, irq.dest_id);
843
844         apic_debug("icr_high 0x%x, icr_low 0x%x, "
845                    "short_hand 0x%x, dest 0x%x, trig_mode 0x%x, level 0x%x, "
846                    "dest_mode 0x%x, delivery_mode 0x%x, vector 0x%x\n",
847                    icr_high, icr_low, irq.shorthand, irq.dest_id,
848                    irq.trig_mode, irq.level, irq.dest_mode, irq.delivery_mode,
849                    irq.vector);
850
851         kvm_irq_delivery_to_apic(apic->vcpu->kvm, apic, &irq);
852 }
853
854 static u32 apic_get_tmcct(struct kvm_lapic *apic)
855 {
856         ktime_t remaining;
857         s64 ns;
858         u32 tmcct;
859
860         ASSERT(apic != NULL);
861
862         /* if initial count is 0, current count should also be 0 */
863         if (kvm_apic_get_reg(apic, APIC_TMICT) == 0)
864                 return 0;
865
866         remaining = hrtimer_get_remaining(&apic->lapic_timer.timer);
867         if (ktime_to_ns(remaining) < 0)
868                 remaining = ktime_set(0, 0);
869
870         ns = mod_64(ktime_to_ns(remaining), apic->lapic_timer.period);
871         tmcct = div64_u64(ns,
872                          (APIC_BUS_CYCLE_NS * apic->divide_count));
873
874         return tmcct;
875 }
876
877 static void __report_tpr_access(struct kvm_lapic *apic, bool write)
878 {
879         struct kvm_vcpu *vcpu = apic->vcpu;
880         struct kvm_run *run = vcpu->run;
881
882         kvm_make_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu);
883         run->tpr_access.rip = kvm_rip_read(vcpu);
884         run->tpr_access.is_write = write;
885 }
886
887 static inline void report_tpr_access(struct kvm_lapic *apic, bool write)
888 {
889         if (apic->vcpu->arch.tpr_access_reporting)
890                 __report_tpr_access(apic, write);
891 }
892
893 static u32 __apic_read(struct kvm_lapic *apic, unsigned int offset)
894 {
895         u32 val = 0;
896
897         if (offset >= LAPIC_MMIO_LENGTH)
898                 return 0;
899
900         switch (offset) {
901         case APIC_ID:
902                 if (apic_x2apic_mode(apic))
903                         val = kvm_apic_id(apic);
904                 else
905                         val = kvm_apic_id(apic) << 24;
906                 break;
907         case APIC_ARBPRI:
908                 apic_debug("Access APIC ARBPRI register which is for P6\n");
909                 break;
910
911         case APIC_TMCCT:        /* Timer CCR */
912                 if (apic_lvtt_tscdeadline(apic))
913                         return 0;
914
915                 val = apic_get_tmcct(apic);
916                 break;
917         case APIC_PROCPRI:
918                 apic_update_ppr(apic);
919                 val = kvm_apic_get_reg(apic, offset);
920                 break;
921         case APIC_TASKPRI:
922                 report_tpr_access(apic, false);
923                 /* fall thru */
924         default:
925                 val = kvm_apic_get_reg(apic, offset);
926                 break;
927         }
928
929         return val;
930 }
931
932 static inline struct kvm_lapic *to_lapic(struct kvm_io_device *dev)
933 {
934         return container_of(dev, struct kvm_lapic, dev);
935 }
936
937 static int apic_reg_read(struct kvm_lapic *apic, u32 offset, int len,
938                 void *data)
939 {
940         unsigned char alignment = offset & 0xf;
941         u32 result;
942         /* this bitmask has a bit cleared for each reserved register */
943         static const u64 rmask = 0x43ff01ffffffe70cULL;
944
945         if ((alignment + len) > 4) {
946                 apic_debug("KVM_APIC_READ: alignment error %x %d\n",
947                            offset, len);
948                 return 1;
949         }
950
951         if (offset > 0x3f0 || !(rmask & (1ULL << (offset >> 4)))) {
952                 apic_debug("KVM_APIC_READ: read reserved register %x\n",
953                            offset);
954                 return 1;
955         }
956
957         result = __apic_read(apic, offset & ~0xf);
958
959         trace_kvm_apic_read(offset, result);
960
961         switch (len) {
962         case 1:
963         case 2:
964         case 4:
965                 memcpy(data, (char *)&result + alignment, len);
966                 break;
967         default:
968                 printk(KERN_ERR "Local APIC read with len = %x, "
969                        "should be 1,2, or 4 instead\n", len);
970                 break;
971         }
972         return 0;
973 }
974
975 static int apic_mmio_in_range(struct kvm_lapic *apic, gpa_t addr)
976 {
977         return kvm_apic_hw_enabled(apic) &&
978             addr >= apic->base_address &&
979             addr < apic->base_address + LAPIC_MMIO_LENGTH;
980 }
981
982 static int apic_mmio_read(struct kvm_io_device *this,
983                            gpa_t address, int len, void *data)
984 {
985         struct kvm_lapic *apic = to_lapic(this);
986         u32 offset = address - apic->base_address;
987
988         if (!apic_mmio_in_range(apic, address))
989                 return -EOPNOTSUPP;
990
991         apic_reg_read(apic, offset, len, data);
992
993         return 0;
994 }
995
996 static void update_divide_count(struct kvm_lapic *apic)
997 {
998         u32 tmp1, tmp2, tdcr;
999
1000         tdcr = kvm_apic_get_reg(apic, APIC_TDCR);
1001         tmp1 = tdcr & 0xf;
1002         tmp2 = ((tmp1 & 0x3) | ((tmp1 & 0x8) >> 1)) + 1;
1003         apic->divide_count = 0x1 << (tmp2 & 0x7);
1004
1005         apic_debug("timer divide count is 0x%x\n",
1006                                    apic->divide_count);
1007 }
1008
1009 static void start_apic_timer(struct kvm_lapic *apic)
1010 {
1011         ktime_t now;
1012         atomic_set(&apic->lapic_timer.pending, 0);
1013
1014         if (apic_lvtt_period(apic) || apic_lvtt_oneshot(apic)) {
1015                 /* lapic timer in oneshot or periodic mode */
1016                 now = apic->lapic_timer.timer.base->get_time();
1017                 apic->lapic_timer.period = (u64)kvm_apic_get_reg(apic, APIC_TMICT)
1018                             * APIC_BUS_CYCLE_NS * apic->divide_count;
1019
1020                 if (!apic->lapic_timer.period)
1021                         return;
1022                 /*
1023                  * Do not allow the guest to program periodic timers with small
1024                  * interval, since the hrtimers are not throttled by the host
1025                  * scheduler.
1026                  */
1027                 if (apic_lvtt_period(apic)) {
1028                         s64 min_period = min_timer_period_us * 1000LL;
1029
1030                         if (apic->lapic_timer.period < min_period) {
1031                                 pr_info_ratelimited(
1032                                     "kvm: vcpu %i: requested %lld ns "
1033                                     "lapic timer period limited to %lld ns\n",
1034                                     apic->vcpu->vcpu_id,
1035                                     apic->lapic_timer.period, min_period);
1036                                 apic->lapic_timer.period = min_period;
1037                         }
1038                 }
1039
1040                 hrtimer_start(&apic->lapic_timer.timer,
1041                               ktime_add_ns(now, apic->lapic_timer.period),
1042                               HRTIMER_MODE_ABS);
1043
1044                 apic_debug("%s: bus cycle is %" PRId64 "ns, now 0x%016"
1045                            PRIx64 ", "
1046                            "timer initial count 0x%x, period %lldns, "
1047                            "expire @ 0x%016" PRIx64 ".\n", __func__,
1048                            APIC_BUS_CYCLE_NS, ktime_to_ns(now),
1049                            kvm_apic_get_reg(apic, APIC_TMICT),
1050                            apic->lapic_timer.period,
1051                            ktime_to_ns(ktime_add_ns(now,
1052                                         apic->lapic_timer.period)));
1053         } else if (apic_lvtt_tscdeadline(apic)) {
1054                 /* lapic timer in tsc deadline mode */
1055                 u64 guest_tsc, tscdeadline = apic->lapic_timer.tscdeadline;
1056                 u64 ns = 0;
1057                 struct kvm_vcpu *vcpu = apic->vcpu;
1058                 unsigned long this_tsc_khz = vcpu->arch.virtual_tsc_khz;
1059                 unsigned long flags;
1060
1061                 if (unlikely(!tscdeadline || !this_tsc_khz))
1062                         return;
1063
1064                 local_irq_save(flags);
1065
1066                 now = apic->lapic_timer.timer.base->get_time();
1067                 guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu, native_read_tsc());
1068                 if (likely(tscdeadline > guest_tsc)) {
1069                         ns = (tscdeadline - guest_tsc) * 1000000ULL;
1070                         do_div(ns, this_tsc_khz);
1071                 }
1072                 hrtimer_start(&apic->lapic_timer.timer,
1073                         ktime_add_ns(now, ns), HRTIMER_MODE_ABS);
1074
1075                 local_irq_restore(flags);
1076         }
1077 }
1078
1079 static void apic_manage_nmi_watchdog(struct kvm_lapic *apic, u32 lvt0_val)
1080 {
1081         int nmi_wd_enabled = apic_lvt_nmi_mode(kvm_apic_get_reg(apic, APIC_LVT0));
1082
1083         if (apic_lvt_nmi_mode(lvt0_val)) {
1084                 if (!nmi_wd_enabled) {
1085                         apic_debug("Receive NMI setting on APIC_LVT0 "
1086                                    "for cpu %d\n", apic->vcpu->vcpu_id);
1087                         apic->vcpu->kvm->arch.vapics_in_nmi_mode++;
1088                 }
1089         } else if (nmi_wd_enabled)
1090                 apic->vcpu->kvm->arch.vapics_in_nmi_mode--;
1091 }
1092
1093 static int apic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val)
1094 {
1095         int ret = 0;
1096
1097         trace_kvm_apic_write(reg, val);
1098
1099         switch (reg) {
1100         case APIC_ID:           /* Local APIC ID */
1101                 if (!apic_x2apic_mode(apic))
1102                         kvm_apic_set_id(apic, val >> 24);
1103                 else
1104                         ret = 1;
1105                 break;
1106
1107         case APIC_TASKPRI:
1108                 report_tpr_access(apic, true);
1109                 apic_set_tpr(apic, val & 0xff);
1110                 break;
1111
1112         case APIC_EOI:
1113                 apic_set_eoi(apic);
1114                 break;
1115
1116         case APIC_LDR:
1117                 if (!apic_x2apic_mode(apic))
1118                         kvm_apic_set_ldr(apic, val & APIC_LDR_MASK);
1119                 else
1120                         ret = 1;
1121                 break;
1122
1123         case APIC_DFR:
1124                 if (!apic_x2apic_mode(apic)) {
1125                         apic_set_reg(apic, APIC_DFR, val | 0x0FFFFFFF);
1126                         recalculate_apic_map(apic->vcpu->kvm);
1127                 } else
1128                         ret = 1;
1129                 break;
1130
1131         case APIC_SPIV: {
1132                 u32 mask = 0x3ff;
1133                 if (kvm_apic_get_reg(apic, APIC_LVR) & APIC_LVR_DIRECTED_EOI)
1134                         mask |= APIC_SPIV_DIRECTED_EOI;
1135                 apic_set_spiv(apic, val & mask);
1136                 if (!(val & APIC_SPIV_APIC_ENABLED)) {
1137                         int i;
1138                         u32 lvt_val;
1139
1140                         for (i = 0; i < APIC_LVT_NUM; i++) {
1141                                 lvt_val = kvm_apic_get_reg(apic,
1142                                                        APIC_LVTT + 0x10 * i);
1143                                 apic_set_reg(apic, APIC_LVTT + 0x10 * i,
1144                                              lvt_val | APIC_LVT_MASKED);
1145                         }
1146                         atomic_set(&apic->lapic_timer.pending, 0);
1147
1148                 }
1149                 break;
1150         }
1151         case APIC_ICR:
1152                 /* No delay here, so we always clear the pending bit */
1153                 apic_set_reg(apic, APIC_ICR, val & ~(1 << 12));
1154                 apic_send_ipi(apic);
1155                 break;
1156
1157         case APIC_ICR2:
1158                 if (!apic_x2apic_mode(apic))
1159                         val &= 0xff000000;
1160                 apic_set_reg(apic, APIC_ICR2, val);
1161                 break;
1162
1163         case APIC_LVT0:
1164                 apic_manage_nmi_watchdog(apic, val);
1165         case APIC_LVTTHMR:
1166         case APIC_LVTPC:
1167         case APIC_LVT1:
1168         case APIC_LVTERR:
1169                 /* TODO: Check vector */
1170                 if (!kvm_apic_sw_enabled(apic))
1171                         val |= APIC_LVT_MASKED;
1172
1173                 val &= apic_lvt_mask[(reg - APIC_LVTT) >> 4];
1174                 apic_set_reg(apic, reg, val);
1175
1176                 break;
1177
1178         case APIC_LVTT:
1179                 if ((kvm_apic_get_reg(apic, APIC_LVTT) &
1180                     apic->lapic_timer.timer_mode_mask) !=
1181                    (val & apic->lapic_timer.timer_mode_mask))
1182                         hrtimer_cancel(&apic->lapic_timer.timer);
1183
1184                 if (!kvm_apic_sw_enabled(apic))
1185                         val |= APIC_LVT_MASKED;
1186                 val &= (apic_lvt_mask[0] | apic->lapic_timer.timer_mode_mask);
1187                 apic_set_reg(apic, APIC_LVTT, val);
1188                 break;
1189
1190         case APIC_TMICT:
1191                 if (apic_lvtt_tscdeadline(apic))
1192                         break;
1193
1194                 hrtimer_cancel(&apic->lapic_timer.timer);
1195                 apic_set_reg(apic, APIC_TMICT, val);
1196                 start_apic_timer(apic);
1197                 break;
1198
1199         case APIC_TDCR:
1200                 if (val & 4)
1201                         apic_debug("KVM_WRITE:TDCR %x\n", val);
1202                 apic_set_reg(apic, APIC_TDCR, val);
1203                 update_divide_count(apic);
1204                 break;
1205
1206         case APIC_ESR:
1207                 if (apic_x2apic_mode(apic) && val != 0) {
1208                         apic_debug("KVM_WRITE:ESR not zero %x\n", val);
1209                         ret = 1;
1210                 }
1211                 break;
1212
1213         case APIC_SELF_IPI:
1214                 if (apic_x2apic_mode(apic)) {
1215                         apic_reg_write(apic, APIC_ICR, 0x40000 | (val & 0xff));
1216                 } else
1217                         ret = 1;
1218                 break;
1219         default:
1220                 ret = 1;
1221                 break;
1222         }
1223         if (ret)
1224                 apic_debug("Local APIC Write to read-only register %x\n", reg);
1225         return ret;
1226 }
1227
1228 static int apic_mmio_write(struct kvm_io_device *this,
1229                             gpa_t address, int len, const void *data)
1230 {
1231         struct kvm_lapic *apic = to_lapic(this);
1232         unsigned int offset = address - apic->base_address;
1233         u32 val;
1234
1235         if (!apic_mmio_in_range(apic, address))
1236                 return -EOPNOTSUPP;
1237
1238         /*
1239          * APIC register must be aligned on 128-bits boundary.
1240          * 32/64/128 bits registers must be accessed thru 32 bits.
1241          * Refer SDM 8.4.1
1242          */
1243         if (len != 4 || (offset & 0xf)) {
1244                 /* Don't shout loud, $infamous_os would cause only noise. */
1245                 apic_debug("apic write: bad size=%d %lx\n", len, (long)address);
1246                 return 0;
1247         }
1248
1249         val = *(u32*)data;
1250
1251         /* too common printing */
1252         if (offset != APIC_EOI)
1253                 apic_debug("%s: offset 0x%x with length 0x%x, and value is "
1254                            "0x%x\n", __func__, offset, len, val);
1255
1256         apic_reg_write(apic, offset & 0xff0, val);
1257
1258         return 0;
1259 }
1260
1261 void kvm_lapic_set_eoi(struct kvm_vcpu *vcpu)
1262 {
1263         if (kvm_vcpu_has_lapic(vcpu))
1264                 apic_reg_write(vcpu->arch.apic, APIC_EOI, 0);
1265 }
1266 EXPORT_SYMBOL_GPL(kvm_lapic_set_eoi);
1267
1268 /* emulate APIC access in a trap manner */
1269 void kvm_apic_write_nodecode(struct kvm_vcpu *vcpu, u32 offset)
1270 {
1271         u32 val = 0;
1272
1273         /* hw has done the conditional check and inst decode */
1274         offset &= 0xff0;
1275
1276         apic_reg_read(vcpu->arch.apic, offset, 4, &val);
1277
1278         /* TODO: optimize to just emulate side effect w/o one more write */
1279         apic_reg_write(vcpu->arch.apic, offset, val);
1280 }
1281 EXPORT_SYMBOL_GPL(kvm_apic_write_nodecode);
1282
1283 void kvm_free_lapic(struct kvm_vcpu *vcpu)
1284 {
1285         struct kvm_lapic *apic = vcpu->arch.apic;
1286
1287         if (!vcpu->arch.apic)
1288                 return;
1289
1290         hrtimer_cancel(&apic->lapic_timer.timer);
1291
1292         if (!(vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE))
1293                 static_key_slow_dec_deferred(&apic_hw_disabled);
1294
1295         if (!(kvm_apic_get_reg(apic, APIC_SPIV) & APIC_SPIV_APIC_ENABLED))
1296                 static_key_slow_dec_deferred(&apic_sw_disabled);
1297
1298         if (apic->regs)
1299                 free_page((unsigned long)apic->regs);
1300
1301         kfree(apic);
1302 }
1303
1304 /*
1305  *----------------------------------------------------------------------
1306  * LAPIC interface
1307  *----------------------------------------------------------------------
1308  */
1309
1310 u64 kvm_get_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu)
1311 {
1312         struct kvm_lapic *apic = vcpu->arch.apic;
1313
1314         if (!kvm_vcpu_has_lapic(vcpu) || apic_lvtt_oneshot(apic) ||
1315                         apic_lvtt_period(apic))
1316                 return 0;
1317
1318         return apic->lapic_timer.tscdeadline;
1319 }
1320
1321 void kvm_set_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu, u64 data)
1322 {
1323         struct kvm_lapic *apic = vcpu->arch.apic;
1324
1325         if (!kvm_vcpu_has_lapic(vcpu) || apic_lvtt_oneshot(apic) ||
1326                         apic_lvtt_period(apic))
1327                 return;
1328
1329         hrtimer_cancel(&apic->lapic_timer.timer);
1330         apic->lapic_timer.tscdeadline = data;
1331         start_apic_timer(apic);
1332 }
1333
1334 void kvm_lapic_set_tpr(struct kvm_vcpu *vcpu, unsigned long cr8)
1335 {
1336         struct kvm_lapic *apic = vcpu->arch.apic;
1337
1338         if (!kvm_vcpu_has_lapic(vcpu))
1339                 return;
1340
1341         apic_set_tpr(apic, ((cr8 & 0x0f) << 4)
1342                      | (kvm_apic_get_reg(apic, APIC_TASKPRI) & 4));
1343 }
1344
1345 u64 kvm_lapic_get_cr8(struct kvm_vcpu *vcpu)
1346 {
1347         u64 tpr;
1348
1349         if (!kvm_vcpu_has_lapic(vcpu))
1350                 return 0;
1351
1352         tpr = (u64) kvm_apic_get_reg(vcpu->arch.apic, APIC_TASKPRI);
1353
1354         return (tpr & 0xf0) >> 4;
1355 }
1356
1357 void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
1358 {
1359         u64 old_value = vcpu->arch.apic_base;
1360         struct kvm_lapic *apic = vcpu->arch.apic;
1361
1362         if (!apic) {
1363                 value |= MSR_IA32_APICBASE_BSP;
1364                 vcpu->arch.apic_base = value;
1365                 return;
1366         }
1367
1368         /* update jump label if enable bit changes */
1369         if ((vcpu->arch.apic_base ^ value) & MSR_IA32_APICBASE_ENABLE) {
1370                 if (value & MSR_IA32_APICBASE_ENABLE)
1371                         static_key_slow_dec_deferred(&apic_hw_disabled);
1372                 else
1373                         static_key_slow_inc(&apic_hw_disabled.key);
1374                 recalculate_apic_map(vcpu->kvm);
1375         }
1376
1377         if (!kvm_vcpu_is_bsp(apic->vcpu))
1378                 value &= ~MSR_IA32_APICBASE_BSP;
1379
1380         vcpu->arch.apic_base = value;
1381         if ((old_value ^ value) & X2APIC_ENABLE) {
1382                 if (value & X2APIC_ENABLE) {
1383                         u32 id = kvm_apic_id(apic);
1384                         u32 ldr = ((id >> 4) << 16) | (1 << (id & 0xf));
1385                         kvm_apic_set_ldr(apic, ldr);
1386                         kvm_x86_ops->set_virtual_x2apic_mode(vcpu, true);
1387                 } else
1388                         kvm_x86_ops->set_virtual_x2apic_mode(vcpu, false);
1389         }
1390
1391         apic->base_address = apic->vcpu->arch.apic_base &
1392                              MSR_IA32_APICBASE_BASE;
1393
1394         /* with FSB delivery interrupt, we can restart APIC functionality */
1395         apic_debug("apic base msr is 0x%016" PRIx64 ", and base address is "
1396                    "0x%lx.\n", apic->vcpu->arch.apic_base, apic->base_address);
1397
1398 }
1399
1400 void kvm_lapic_reset(struct kvm_vcpu *vcpu)
1401 {
1402         struct kvm_lapic *apic;
1403         int i;
1404
1405         apic_debug("%s\n", __func__);
1406
1407         ASSERT(vcpu);
1408         apic = vcpu->arch.apic;
1409         ASSERT(apic != NULL);
1410
1411         /* Stop the timer in case it's a reset to an active apic */
1412         hrtimer_cancel(&apic->lapic_timer.timer);
1413
1414         kvm_apic_set_id(apic, vcpu->vcpu_id);
1415         kvm_apic_set_version(apic->vcpu);
1416
1417         for (i = 0; i < APIC_LVT_NUM; i++)
1418                 apic_set_reg(apic, APIC_LVTT + 0x10 * i, APIC_LVT_MASKED);
1419         apic_set_reg(apic, APIC_LVT0,
1420                      SET_APIC_DELIVERY_MODE(0, APIC_MODE_EXTINT));
1421
1422         apic_set_reg(apic, APIC_DFR, 0xffffffffU);
1423         apic_set_spiv(apic, 0xff);
1424         apic_set_reg(apic, APIC_TASKPRI, 0);
1425         kvm_apic_set_ldr(apic, 0);
1426         apic_set_reg(apic, APIC_ESR, 0);
1427         apic_set_reg(apic, APIC_ICR, 0);
1428         apic_set_reg(apic, APIC_ICR2, 0);
1429         apic_set_reg(apic, APIC_TDCR, 0);
1430         apic_set_reg(apic, APIC_TMICT, 0);
1431         for (i = 0; i < 8; i++) {
1432                 apic_set_reg(apic, APIC_IRR + 0x10 * i, 0);
1433                 apic_set_reg(apic, APIC_ISR + 0x10 * i, 0);
1434                 apic_set_reg(apic, APIC_TMR + 0x10 * i, 0);
1435         }
1436         apic->irr_pending = kvm_apic_vid_enabled(vcpu->kvm);
1437         apic->isr_count = kvm_apic_vid_enabled(vcpu->kvm);
1438         apic->highest_isr_cache = -1;
1439         update_divide_count(apic);
1440         atomic_set(&apic->lapic_timer.pending, 0);
1441         if (kvm_vcpu_is_bsp(vcpu))
1442                 kvm_lapic_set_base(vcpu,
1443                                 vcpu->arch.apic_base | MSR_IA32_APICBASE_BSP);
1444         vcpu->arch.pv_eoi.msr_val = 0;
1445         apic_update_ppr(apic);
1446
1447         vcpu->arch.apic_arb_prio = 0;
1448         vcpu->arch.apic_attention = 0;
1449
1450         apic_debug(KERN_INFO "%s: vcpu=%p, id=%d, base_msr="
1451                    "0x%016" PRIx64 ", base_address=0x%0lx.\n", __func__,
1452                    vcpu, kvm_apic_id(apic),
1453                    vcpu->arch.apic_base, apic->base_address);
1454 }
1455
1456 /*
1457  *----------------------------------------------------------------------
1458  * timer interface
1459  *----------------------------------------------------------------------
1460  */
1461
1462 static bool lapic_is_periodic(struct kvm_lapic *apic)
1463 {
1464         return apic_lvtt_period(apic);
1465 }
1466
1467 int apic_has_pending_timer(struct kvm_vcpu *vcpu)
1468 {
1469         struct kvm_lapic *apic = vcpu->arch.apic;
1470
1471         if (kvm_vcpu_has_lapic(vcpu) && apic_enabled(apic) &&
1472                         apic_lvt_enabled(apic, APIC_LVTT))
1473                 return atomic_read(&apic->lapic_timer.pending);
1474
1475         return 0;
1476 }
1477
1478 int kvm_apic_local_deliver(struct kvm_lapic *apic, int lvt_type)
1479 {
1480         u32 reg = kvm_apic_get_reg(apic, lvt_type);
1481         int vector, mode, trig_mode;
1482
1483         if (kvm_apic_hw_enabled(apic) && !(reg & APIC_LVT_MASKED)) {
1484                 vector = reg & APIC_VECTOR_MASK;
1485                 mode = reg & APIC_MODE_MASK;
1486                 trig_mode = reg & APIC_LVT_LEVEL_TRIGGER;
1487                 return __apic_accept_irq(apic, mode, vector, 1, trig_mode);
1488         }
1489         return 0;
1490 }
1491
1492 void kvm_apic_nmi_wd_deliver(struct kvm_vcpu *vcpu)
1493 {
1494         struct kvm_lapic *apic = vcpu->arch.apic;
1495
1496         if (apic)
1497                 kvm_apic_local_deliver(apic, APIC_LVT0);
1498 }
1499
1500 static const struct kvm_io_device_ops apic_mmio_ops = {
1501         .read     = apic_mmio_read,
1502         .write    = apic_mmio_write,
1503 };
1504
1505 static enum hrtimer_restart apic_timer_fn(struct hrtimer *data)
1506 {
1507         struct kvm_timer *ktimer = container_of(data, struct kvm_timer, timer);
1508         struct kvm_lapic *apic = container_of(ktimer, struct kvm_lapic, lapic_timer);
1509         struct kvm_vcpu *vcpu = apic->vcpu;
1510         wait_queue_head_t *q = &vcpu->wq;
1511
1512         /*
1513          * There is a race window between reading and incrementing, but we do
1514          * not care about potentially losing timer events in the !reinject
1515          * case anyway. Note: KVM_REQ_PENDING_TIMER is implicitly checked
1516          * in vcpu_enter_guest.
1517          */
1518         if (!atomic_read(&ktimer->pending)) {
1519                 atomic_inc(&ktimer->pending);
1520                 /* FIXME: this code should not know anything about vcpus */
1521                 kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1522         }
1523
1524         if (waitqueue_active(q))
1525                 wake_up_interruptible(q);
1526
1527         if (lapic_is_periodic(apic)) {
1528                 hrtimer_add_expires_ns(&ktimer->timer, ktimer->period);
1529                 return HRTIMER_RESTART;
1530         } else
1531                 return HRTIMER_NORESTART;
1532 }
1533
1534 int kvm_create_lapic(struct kvm_vcpu *vcpu)
1535 {
1536         struct kvm_lapic *apic;
1537
1538         ASSERT(vcpu != NULL);
1539         apic_debug("apic_init %d\n", vcpu->vcpu_id);
1540
1541         apic = kzalloc(sizeof(*apic), GFP_KERNEL);
1542         if (!apic)
1543                 goto nomem;
1544
1545         vcpu->arch.apic = apic;
1546
1547         apic->regs = (void *)get_zeroed_page(GFP_KERNEL);
1548         if (!apic->regs) {
1549                 printk(KERN_ERR "malloc apic regs error for vcpu %x\n",
1550                        vcpu->vcpu_id);
1551                 goto nomem_free_apic;
1552         }
1553         apic->vcpu = vcpu;
1554
1555         hrtimer_init(&apic->lapic_timer.timer, CLOCK_MONOTONIC,
1556                      HRTIMER_MODE_ABS);
1557         apic->lapic_timer.timer.function = apic_timer_fn;
1558
1559         /*
1560          * APIC is created enabled. This will prevent kvm_lapic_set_base from
1561          * thinking that APIC satet has changed.
1562          */
1563         vcpu->arch.apic_base = MSR_IA32_APICBASE_ENABLE;
1564         kvm_lapic_set_base(vcpu,
1565                         APIC_DEFAULT_PHYS_BASE | MSR_IA32_APICBASE_ENABLE);
1566
1567         static_key_slow_inc(&apic_sw_disabled.key); /* sw disabled at reset */
1568         kvm_lapic_reset(vcpu);
1569         kvm_iodevice_init(&apic->dev, &apic_mmio_ops);
1570
1571         return 0;
1572 nomem_free_apic:
1573         kfree(apic);
1574 nomem:
1575         return -ENOMEM;
1576 }
1577
1578 int kvm_apic_has_interrupt(struct kvm_vcpu *vcpu)
1579 {
1580         struct kvm_lapic *apic = vcpu->arch.apic;
1581         int highest_irr;
1582
1583         if (!kvm_vcpu_has_lapic(vcpu) || !apic_enabled(apic))
1584                 return -1;
1585
1586         apic_update_ppr(apic);
1587         highest_irr = apic_find_highest_irr(apic);
1588         if ((highest_irr == -1) ||
1589             ((highest_irr & 0xF0) <= kvm_apic_get_reg(apic, APIC_PROCPRI)))
1590                 return -1;
1591         return highest_irr;
1592 }
1593
1594 int kvm_apic_accept_pic_intr(struct kvm_vcpu *vcpu)
1595 {
1596         u32 lvt0 = kvm_apic_get_reg(vcpu->arch.apic, APIC_LVT0);
1597         int r = 0;
1598
1599         if (!kvm_apic_hw_enabled(vcpu->arch.apic))
1600                 r = 1;
1601         if ((lvt0 & APIC_LVT_MASKED) == 0 &&
1602             GET_APIC_DELIVERY_MODE(lvt0) == APIC_MODE_EXTINT)
1603                 r = 1;
1604         return r;
1605 }
1606
1607 void kvm_inject_apic_timer_irqs(struct kvm_vcpu *vcpu)
1608 {
1609         struct kvm_lapic *apic = vcpu->arch.apic;
1610
1611         if (!kvm_vcpu_has_lapic(vcpu))
1612                 return;
1613
1614         if (atomic_read(&apic->lapic_timer.pending) > 0) {
1615                 if (kvm_apic_local_deliver(apic, APIC_LVTT))
1616                         atomic_dec(&apic->lapic_timer.pending);
1617         }
1618 }
1619
1620 int kvm_get_apic_interrupt(struct kvm_vcpu *vcpu)
1621 {
1622         int vector = kvm_apic_has_interrupt(vcpu);
1623         struct kvm_lapic *apic = vcpu->arch.apic;
1624
1625         if (vector == -1)
1626                 return -1;
1627
1628         apic_set_isr(vector, apic);
1629         apic_update_ppr(apic);
1630         apic_clear_irr(vector, apic);
1631         return vector;
1632 }
1633
1634 void kvm_apic_post_state_restore(struct kvm_vcpu *vcpu,
1635                 struct kvm_lapic_state *s)
1636 {
1637         struct kvm_lapic *apic = vcpu->arch.apic;
1638
1639         kvm_lapic_set_base(vcpu, vcpu->arch.apic_base);
1640         /* set SPIV separately to get count of SW disabled APICs right */
1641         apic_set_spiv(apic, *((u32 *)(s->regs + APIC_SPIV)));
1642         memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
1643         /* call kvm_apic_set_id() to put apic into apic_map */
1644         kvm_apic_set_id(apic, kvm_apic_id(apic));
1645         kvm_apic_set_version(vcpu);
1646
1647         apic_update_ppr(apic);
1648         hrtimer_cancel(&apic->lapic_timer.timer);
1649         update_divide_count(apic);
1650         start_apic_timer(apic);
1651         apic->irr_pending = true;
1652         apic->isr_count = kvm_apic_vid_enabled(vcpu->kvm) ?
1653                                 1 : count_vectors(apic->regs + APIC_ISR);
1654         apic->highest_isr_cache = -1;
1655         kvm_x86_ops->hwapic_isr_update(vcpu->kvm, apic_find_highest_isr(apic));
1656         kvm_make_request(KVM_REQ_EVENT, vcpu);
1657 }
1658
1659 void __kvm_migrate_apic_timer(struct kvm_vcpu *vcpu)
1660 {
1661         struct hrtimer *timer;
1662
1663         if (!kvm_vcpu_has_lapic(vcpu))
1664                 return;
1665
1666         timer = &vcpu->arch.apic->lapic_timer.timer;
1667         if (hrtimer_cancel(timer))
1668                 hrtimer_start_expires(timer, HRTIMER_MODE_ABS);
1669 }
1670
1671 /*
1672  * apic_sync_pv_eoi_from_guest - called on vmexit or cancel interrupt
1673  *
1674  * Detect whether guest triggered PV EOI since the
1675  * last entry. If yes, set EOI on guests's behalf.
1676  * Clear PV EOI in guest memory in any case.
1677  */
1678 static void apic_sync_pv_eoi_from_guest(struct kvm_vcpu *vcpu,
1679                                         struct kvm_lapic *apic)
1680 {
1681         bool pending;
1682         int vector;
1683         /*
1684          * PV EOI state is derived from KVM_APIC_PV_EOI_PENDING in host
1685          * and KVM_PV_EOI_ENABLED in guest memory as follows:
1686          *
1687          * KVM_APIC_PV_EOI_PENDING is unset:
1688          *      -> host disabled PV EOI.
1689          * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is set:
1690          *      -> host enabled PV EOI, guest did not execute EOI yet.
1691          * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is unset:
1692          *      -> host enabled PV EOI, guest executed EOI.
1693          */
1694         BUG_ON(!pv_eoi_enabled(vcpu));
1695         pending = pv_eoi_get_pending(vcpu);
1696         /*
1697          * Clear pending bit in any case: it will be set again on vmentry.
1698          * While this might not be ideal from performance point of view,
1699          * this makes sure pv eoi is only enabled when we know it's safe.
1700          */
1701         pv_eoi_clr_pending(vcpu);
1702         if (pending)
1703                 return;
1704         vector = apic_set_eoi(apic);
1705         trace_kvm_pv_eoi(apic, vector);
1706 }
1707
1708 void kvm_lapic_sync_from_vapic(struct kvm_vcpu *vcpu)
1709 {
1710         u32 data;
1711         void *vapic;
1712
1713         if (test_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention))
1714                 apic_sync_pv_eoi_from_guest(vcpu, vcpu->arch.apic);
1715
1716         if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
1717                 return;
1718
1719         vapic = kmap_atomic(vcpu->arch.apic->vapic_page);
1720         data = *(u32 *)(vapic + offset_in_page(vcpu->arch.apic->vapic_addr));
1721         kunmap_atomic(vapic);
1722
1723         apic_set_tpr(vcpu->arch.apic, data & 0xff);
1724 }
1725
1726 /*
1727  * apic_sync_pv_eoi_to_guest - called before vmentry
1728  *
1729  * Detect whether it's safe to enable PV EOI and
1730  * if yes do so.
1731  */
1732 static void apic_sync_pv_eoi_to_guest(struct kvm_vcpu *vcpu,
1733                                         struct kvm_lapic *apic)
1734 {
1735         if (!pv_eoi_enabled(vcpu) ||
1736             /* IRR set or many bits in ISR: could be nested. */
1737             apic->irr_pending ||
1738             /* Cache not set: could be safe but we don't bother. */
1739             apic->highest_isr_cache == -1 ||
1740             /* Need EOI to update ioapic. */
1741             kvm_ioapic_handles_vector(vcpu->kvm, apic->highest_isr_cache)) {
1742                 /*
1743                  * PV EOI was disabled by apic_sync_pv_eoi_from_guest
1744                  * so we need not do anything here.
1745                  */
1746                 return;
1747         }
1748
1749         pv_eoi_set_pending(apic->vcpu);
1750 }
1751
1752 void kvm_lapic_sync_to_vapic(struct kvm_vcpu *vcpu)
1753 {
1754         u32 data, tpr;
1755         int max_irr, max_isr;
1756         struct kvm_lapic *apic = vcpu->arch.apic;
1757         void *vapic;
1758
1759         apic_sync_pv_eoi_to_guest(vcpu, apic);
1760
1761         if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
1762                 return;
1763
1764         tpr = kvm_apic_get_reg(apic, APIC_TASKPRI) & 0xff;
1765         max_irr = apic_find_highest_irr(apic);
1766         if (max_irr < 0)
1767                 max_irr = 0;
1768         max_isr = apic_find_highest_isr(apic);
1769         if (max_isr < 0)
1770                 max_isr = 0;
1771         data = (tpr & 0xff) | ((max_isr & 0xf0) << 8) | (max_irr << 24);
1772
1773         vapic = kmap_atomic(vcpu->arch.apic->vapic_page);
1774         *(u32 *)(vapic + offset_in_page(vcpu->arch.apic->vapic_addr)) = data;
1775         kunmap_atomic(vapic);
1776 }
1777
1778 void kvm_lapic_set_vapic_addr(struct kvm_vcpu *vcpu, gpa_t vapic_addr)
1779 {
1780         vcpu->arch.apic->vapic_addr = vapic_addr;
1781         if (vapic_addr)
1782                 __set_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
1783         else
1784                 __clear_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
1785 }
1786
1787 int kvm_x2apic_msr_write(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1788 {
1789         struct kvm_lapic *apic = vcpu->arch.apic;
1790         u32 reg = (msr - APIC_BASE_MSR) << 4;
1791
1792         if (!irqchip_in_kernel(vcpu->kvm) || !apic_x2apic_mode(apic))
1793                 return 1;
1794
1795         /* if this is ICR write vector before command */
1796         if (msr == 0x830)
1797                 apic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
1798         return apic_reg_write(apic, reg, (u32)data);
1799 }
1800
1801 int kvm_x2apic_msr_read(struct kvm_vcpu *vcpu, u32 msr, u64 *data)
1802 {
1803         struct kvm_lapic *apic = vcpu->arch.apic;
1804         u32 reg = (msr - APIC_BASE_MSR) << 4, low, high = 0;
1805
1806         if (!irqchip_in_kernel(vcpu->kvm) || !apic_x2apic_mode(apic))
1807                 return 1;
1808
1809         if (apic_reg_read(apic, reg, 4, &low))
1810                 return 1;
1811         if (msr == 0x830)
1812                 apic_reg_read(apic, APIC_ICR2, 4, &high);
1813
1814         *data = (((u64)high) << 32) | low;
1815
1816         return 0;
1817 }
1818
1819 int kvm_hv_vapic_msr_write(struct kvm_vcpu *vcpu, u32 reg, u64 data)
1820 {
1821         struct kvm_lapic *apic = vcpu->arch.apic;
1822
1823         if (!kvm_vcpu_has_lapic(vcpu))
1824                 return 1;
1825
1826         /* if this is ICR write vector before command */
1827         if (reg == APIC_ICR)
1828                 apic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
1829         return apic_reg_write(apic, reg, (u32)data);
1830 }
1831
1832 int kvm_hv_vapic_msr_read(struct kvm_vcpu *vcpu, u32 reg, u64 *data)
1833 {
1834         struct kvm_lapic *apic = vcpu->arch.apic;
1835         u32 low, high = 0;
1836
1837         if (!kvm_vcpu_has_lapic(vcpu))
1838                 return 1;
1839
1840         if (apic_reg_read(apic, reg, 4, &low))
1841                 return 1;
1842         if (reg == APIC_ICR)
1843                 apic_reg_read(apic, APIC_ICR2, 4, &high);
1844
1845         *data = (((u64)high) << 32) | low;
1846
1847         return 0;
1848 }
1849
1850 int kvm_lapic_enable_pv_eoi(struct kvm_vcpu *vcpu, u64 data)
1851 {
1852         u64 addr = data & ~KVM_MSR_ENABLED;
1853         if (!IS_ALIGNED(addr, 4))
1854                 return 1;
1855
1856         vcpu->arch.pv_eoi.msr_val = data;
1857         if (!pv_eoi_enabled(vcpu))
1858                 return 0;
1859         return kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.pv_eoi.data,
1860                                          addr);
1861 }
1862
1863 void kvm_lapic_init(void)
1864 {
1865         /* do not patch jump label more than once per second */
1866         jump_label_rate_limit(&apic_hw_disabled, HZ);
1867         jump_label_rate_limit(&apic_sw_disabled, HZ);
1868 }