projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Merge branch 'drm-next-4.8' of git://people.freedesktop.org/~agd5f/linux into drm...
[linux-2.6-block.git] / arch / x86 / kvm / lapic.c
CommitLineData
97222cc8
ED		 1
2/*
3 * Local APIC virtualization
4 *
5 * Copyright (C) 2006 Qumranet, Inc.
6 * Copyright (C) 2007 Novell
7 * Copyright (C) 2007 Intel
9611c187 8 * Copyright 2009 Red Hat, Inc. and/or its affiliates.
97222cc8
ED		 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
edf88417 21#include <linux/kvm_host.h>
97222cc8
ED		 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>
1767e931 28#include <linux/export.h>
6f6d6a1a 29#include <linux/math64.h>
5a0e3ad6 30#include <linux/slab.h>
97222cc8
ED		 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>
d0659d94 36#include <asm/delay.h>
60063497 37#include <linux/atomic.h>
c5cc421b 38#include <linux/jump_label.h>
5fdbf976 39#include "kvm_cache_regs.h"
97222cc8 40#include "irq.h"
229456fc 41#include "trace.h"
fc61b800 42#include "x86.h"
00b27a3e 43#include "cpuid.h"
5c919412 44#include "hyperv.h"
97222cc8 45
b682b814
MT		 46#ifndef CONFIG_X86_64
47#define mod_64(x, y) ((x) - (y) * div64_u64(x, y))
48#else
49#define mod_64(x, y) ((x) % (y))
50#endif
51
97222cc8
ED		 52#define PRId64 "d"
53#define PRIx64 "llx"
54#define PRIu64 "u"
55#define PRIo64 "o"
56
57#define APIC_BUS_CYCLE_NS 1
58
59/* #define apic_debug(fmt,arg...) printk(KERN_WARNING fmt,##arg) */
60#define apic_debug(fmt, arg...)
61
97222cc8 62/* 14 is the version for Xeon and Pentium 8.4.8*/
1e6e2755 63#define APIC_VERSION (0x14UL | ((KVM_APIC_LVT_NUM - 1) << 16))
97222cc8
ED		 64#define LAPIC_MMIO_LENGTH (1 << 12)
65/* followed define is not in apicdef.h */
66#define APIC_SHORT_MASK 0xc0000
67#define APIC_DEST_NOSHORT 0x0
68#define APIC_DEST_MASK 0x800
69#define MAX_APIC_VECTOR 256
ecba9a52 70#define APIC_VECTORS_PER_REG 32
97222cc8 71
394457a9
NA		 72#define APIC_BROADCAST 0xFF
73#define X2APIC_BROADCAST 0xFFFFFFFFul
74
a0c9a822
MT		 75static inline int apic_test_vector(int vec, void *bitmap)
76{
77 return test_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
78}
79
10606919
YZ		 80bool kvm_apic_pending_eoi(struct kvm_vcpu *vcpu, int vector)
81{
82 struct kvm_lapic *apic = vcpu->arch.apic;
83
84 return apic_test_vector(vector, apic->regs + APIC_ISR) ||
85 apic_test_vector(vector, apic->regs + APIC_IRR);
86}
87
97222cc8
ED		 88static inline void apic_clear_vector(int vec, void *bitmap)
89{
90 clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
91}
92
8680b94b
MT		 93static inline int __apic_test_and_set_vector(int vec, void *bitmap)
94{
95 return __test_and_set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
96}
97
98static inline int __apic_test_and_clear_vector(int vec, void *bitmap)
99{
100 return __test_and_clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
101}
102
c5cc421b 103struct static_key_deferred apic_hw_disabled __read_mostly;
f8c1ea10
GN		 104struct static_key_deferred apic_sw_disabled __read_mostly;
105
97222cc8
ED		 106static inline int apic_enabled(struct kvm_lapic *apic)
107{
c48f1496 108 return kvm_apic_sw_enabled(apic) && kvm_apic_hw_enabled(apic);
54e9818f
GN		 109}
110
97222cc8
ED		 111#define LVT_MASK \
112 (APIC_LVT_MASKED | APIC_SEND_PENDING | APIC_VECTOR_MASK)
113
114#define LINT_MASK \
115 (LVT_MASK | APIC_MODE_MASK | APIC_INPUT_POLARITY | \
116 APIC_LVT_REMOTE_IRR | APIC_LVT_LEVEL_TRIGGER)
117
e45115b6
RK		 118static inline bool kvm_apic_map_get_logical_dest(struct kvm_apic_map *map,
119 u32 dest_id, struct kvm_lapic ***cluster, u16 *mask) {
120 switch (map->mode) {
121 case KVM_APIC_MODE_X2APIC: {
122 u32 offset = (dest_id >> 16) * 16;
0ca52e7b 123 u32 max_apic_id = map->max_apic_id;
e45115b6
RK		 124
125 if (offset <= max_apic_id) {
126 u8 cluster_size = min(max_apic_id - offset + 1, 16U);
127
128 *cluster = &map->phys_map[offset];
129 *mask = dest_id & (0xffff >> (16 - cluster_size));
130 } else {
131 *mask = 0;
132 }
3b5a5ffa 133
e45115b6
RK		 134 return true;
135 }
136 case KVM_APIC_MODE_XAPIC_FLAT:
137 *cluster = map->xapic_flat_map;
138 *mask = dest_id & 0xff;
139 return true;
140 case KVM_APIC_MODE_XAPIC_CLUSTER:
141 *cluster = map->xapic_cluster_map[dest_id >> 4];
142 *mask = dest_id & 0xf;
143 return true;
144 default:
145 /* Not optimized. */
146 return false;
147 }
3548a259
RK		 148}
149
af1bae54 150static void kvm_apic_map_free(struct rcu_head *rcu)
3b5a5ffa 151{
af1bae54 152 struct kvm_apic_map *map = container_of(rcu, struct kvm_apic_map, rcu);
3b5a5ffa 153
af1bae54 154 kvfree(map);
3b5a5ffa
RK		 155}
156
1e08ec4a
GN		 157static void recalculate_apic_map(struct kvm *kvm)
158{
159 struct kvm_apic_map *new, *old = NULL;
160 struct kvm_vcpu *vcpu;
161 int i;
0ca52e7b 162 u32 max_id = 255;
1e08ec4a
GN		 163
164 mutex_lock(&kvm->arch.apic_map_lock);
165
0ca52e7b
RK		 166 kvm_for_each_vcpu(i, vcpu, kvm)
167 if (kvm_apic_present(vcpu))
168 max_id = max(max_id, kvm_apic_id(vcpu->arch.apic));
169
af1bae54
RK		 170 new = kvm_kvzalloc(sizeof(struct kvm_apic_map) +
171 sizeof(struct kvm_lapic *) * ((u64)max_id + 1));
0ca52e7b 172
1e08ec4a
GN		 173 if (!new)
174 goto out;
175
0ca52e7b
RK		 176 new->max_apic_id = max_id;
177
173beedc
NA		 178 kvm_for_each_vcpu(i, vcpu, kvm) {
179 struct kvm_lapic *apic = vcpu->arch.apic;
e45115b6
RK		 180 struct kvm_lapic **cluster;
181 u16 mask;
25995e5b 182 u32 ldr, aid;
1e08ec4a 183
df04d1d1
RK		 184 if (!kvm_apic_present(vcpu))
185 continue;
186
25995e5b 187 aid = kvm_apic_id(apic);
dfb95954 188 ldr = kvm_lapic_get_reg(apic, APIC_LDR);
1e08ec4a 189
0ca52e7b 190 if (aid <= new->max_apic_id)
25995e5b 191 new->phys_map[aid] = apic;
3548a259 192
3b5a5ffa
RK		 193 if (apic_x2apic_mode(apic)) {
194 new->mode |= KVM_APIC_MODE_X2APIC;
195 } else if (ldr) {
196 ldr = GET_APIC_LOGICAL_ID(ldr);
dfb95954 197 if (kvm_lapic_get_reg(apic, APIC_DFR) == APIC_DFR_FLAT)
3b5a5ffa
RK		 198 new->mode |= KVM_APIC_MODE_XAPIC_FLAT;
199 else
200 new->mode |= KVM_APIC_MODE_XAPIC_CLUSTER;
201 }
202
e45115b6 203 if (!kvm_apic_map_get_logical_dest(new, ldr, &cluster, &mask))
3548a259
RK		 204 continue;
205
e45115b6
RK		 206 if (mask)
207 cluster[ffs(mask) - 1] = apic;
1e08ec4a
GN		 208 }
209out:
210 old = rcu_dereference_protected(kvm->arch.apic_map,
211 lockdep_is_held(&kvm->arch.apic_map_lock));
212 rcu_assign_pointer(kvm->arch.apic_map, new);
213 mutex_unlock(&kvm->arch.apic_map_lock);
214
215 if (old)
af1bae54 216 call_rcu(&old->rcu, kvm_apic_map_free);
c7c9c56c 217
b053b2ae 218 kvm_make_scan_ioapic_request(kvm);
1e08ec4a
GN		 219}
220
1e1b6c26
NA		 221static inline void apic_set_spiv(struct kvm_lapic *apic, u32 val)
222{
e462755c 223 bool enabled = val & APIC_SPIV_APIC_ENABLED;
1e1b6c26 224
1e6e2755 225 kvm_lapic_set_reg(apic, APIC_SPIV, val);
e462755c
RK		 226
227 if (enabled != apic->sw_enabled) {
228 apic->sw_enabled = enabled;
229 if (enabled) {
1e1b6c26
NA		 230 static_key_slow_dec_deferred(&apic_sw_disabled);
231 recalculate_apic_map(apic->vcpu->kvm);
232 } else
233 static_key_slow_inc(&apic_sw_disabled.key);
234 }
235}
236
a92e2543 237static inline void kvm_apic_set_xapic_id(struct kvm_lapic *apic, u8 id)
1e08ec4a 238{
1e6e2755 239 kvm_lapic_set_reg(apic, APIC_ID, id << 24);
1e08ec4a
GN		 240 recalculate_apic_map(apic->vcpu->kvm);
241}
242
243static inline void kvm_apic_set_ldr(struct kvm_lapic *apic, u32 id)
244{
1e6e2755 245 kvm_lapic_set_reg(apic, APIC_LDR, id);
1e08ec4a
GN		 246 recalculate_apic_map(apic->vcpu->kvm);
247}
248
a92e2543 249static inline void kvm_apic_set_x2apic_id(struct kvm_lapic *apic, u32 id)
257b9a5f
RK		 250{
251 u32 ldr = ((id >> 4) << 16) | (1 << (id & 0xf));
252
a92e2543 253 kvm_lapic_set_reg(apic, APIC_ID, id);
1e6e2755 254 kvm_lapic_set_reg(apic, APIC_LDR, ldr);
257b9a5f
RK		 255 recalculate_apic_map(apic->vcpu->kvm);
256}
257
97222cc8
ED		 258static inline int apic_lvt_enabled(struct kvm_lapic *apic, int lvt_type)
259{
dfb95954 260 return !(kvm_lapic_get_reg(apic, lvt_type) & APIC_LVT_MASKED);
97222cc8
ED		 261}
262
263static inline int apic_lvt_vector(struct kvm_lapic *apic, int lvt_type)
264{
dfb95954 265 return kvm_lapic_get_reg(apic, lvt_type) & APIC_VECTOR_MASK;
97222cc8
ED		 266}
267
a3e06bbe
LJ		 268static inline int apic_lvtt_oneshot(struct kvm_lapic *apic)
269{
f30ebc31 270 return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_ONESHOT;
a3e06bbe
LJ		 271}
272
97222cc8
ED		 273static inline int apic_lvtt_period(struct kvm_lapic *apic)
274{
f30ebc31 275 return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_PERIODIC;
a3e06bbe
LJ		 276}
277
278static inline int apic_lvtt_tscdeadline(struct kvm_lapic *apic)
279{
f30ebc31 280 return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_TSCDEADLINE;
97222cc8
ED		 281}
282
cc6e462c
JK		 283static inline int apic_lvt_nmi_mode(u32 lvt_val)
284{
285 return (lvt_val & (APIC_MODE_MASK | APIC_LVT_MASKED)) == APIC_DM_NMI;
286}
287
fc61b800
GN		 288void kvm_apic_set_version(struct kvm_vcpu *vcpu)
289{
290 struct kvm_lapic *apic = vcpu->arch.apic;
291 struct kvm_cpuid_entry2 *feat;
292 u32 v = APIC_VERSION;
293
bce87cce 294 if (!lapic_in_kernel(vcpu))
fc61b800
GN		 295 return;
296
297 feat = kvm_find_cpuid_entry(apic->vcpu, 0x1, 0);
298 if (feat && (feat->ecx & (1 << (X86_FEATURE_X2APIC & 31))))
299 v |= APIC_LVR_DIRECTED_EOI;
1e6e2755 300 kvm_lapic_set_reg(apic, APIC_LVR, v);
fc61b800
GN		 301}
302
1e6e2755 303static const unsigned int apic_lvt_mask[KVM_APIC_LVT_NUM] = {
a3e06bbe 304 LVT_MASK , /* part LVTT mask, timer mode mask added at runtime */
97222cc8
ED		 305 LVT_MASK | APIC_MODE_MASK, /* LVTTHMR */
306 LVT_MASK | APIC_MODE_MASK, /* LVTPC */
307 LINT_MASK, LINT_MASK, /* LVT0-1 */
308 LVT_MASK /* LVTERR */
309};
310
311static int find_highest_vector(void *bitmap)
312{
ecba9a52
TY		 313 int vec;
314 u32 *reg;
97222cc8 315
ecba9a52
TY		 316 for (vec = MAX_APIC_VECTOR - APIC_VECTORS_PER_REG;
317 vec >= 0; vec -= APIC_VECTORS_PER_REG) {
318 reg = bitmap + REG_POS(vec);
319 if (*reg)
320 return fls(*reg) - 1 + vec;
321 }
97222cc8 322
ecba9a52 323 return -1;
97222cc8
ED		 324}
325
8680b94b
MT		 326static u8 count_vectors(void *bitmap)
327{
ecba9a52
TY		 328 int vec;
329 u32 *reg;
8680b94b 330 u8 count = 0;
ecba9a52
TY		 331
332 for (vec = 0; vec < MAX_APIC_VECTOR; vec += APIC_VECTORS_PER_REG) {
333 reg = bitmap + REG_POS(vec);
334 count += hweight32(*reg);
335 }
336
8680b94b
MT		 337 return count;
338}
339
705699a1 340void __kvm_apic_update_irr(u32 *pir, void *regs)
a20ed54d
YZ		 341{
342 u32 i, pir_val;
a20ed54d
YZ		 343
344 for (i = 0; i <= 7; i++) {
345 pir_val = xchg(&pir[i], 0);
346 if (pir_val)
705699a1 347 *((u32 *)(regs + APIC_IRR + i * 0x10)) |= pir_val;
a20ed54d
YZ		 348 }
349}
705699a1
WV		 350EXPORT_SYMBOL_GPL(__kvm_apic_update_irr);
351
352void kvm_apic_update_irr(struct kvm_vcpu *vcpu, u32 *pir)
353{
354 struct kvm_lapic *apic = vcpu->arch.apic;
355
356 __kvm_apic_update_irr(pir, apic->regs);
c77f3fab
RK		 357
358 kvm_make_request(KVM_REQ_EVENT, vcpu);
705699a1 359}
a20ed54d
YZ		 360EXPORT_SYMBOL_GPL(kvm_apic_update_irr);
361
33e4c686 362static inline int apic_search_irr(struct kvm_lapic *apic)
97222cc8 363{
33e4c686 364 return find_highest_vector(apic->regs + APIC_IRR);
97222cc8
ED		 365}
366
367static inline int apic_find_highest_irr(struct kvm_lapic *apic)
368{
369 int result;
370
c7c9c56c
YZ		 371 /*
372 * Note that irr_pending is just a hint. It will be always
373 * true with virtual interrupt delivery enabled.
374 */
33e4c686
GN		 375 if (!apic->irr_pending)
376 return -1;
377
d62caabb
AS		 378 if (apic->vcpu->arch.apicv_active)
379 kvm_x86_ops->sync_pir_to_irr(apic->vcpu);
33e4c686 380 result = apic_search_irr(apic);
97222cc8
ED		 381 ASSERT(result == -1 || result >= 16);
382
383 return result;
384}
385
33e4c686
GN		 386static inline void apic_clear_irr(int vec, struct kvm_lapic *apic)
387{
56cc2406
WL		 388 struct kvm_vcpu *vcpu;
389
390 vcpu = apic->vcpu;
391
d62caabb 392 if (unlikely(vcpu->arch.apicv_active)) {
56cc2406 393 /* try to update RVI */
f210f757 394 apic_clear_vector(vec, apic->regs + APIC_IRR);
56cc2406 395 kvm_make_request(KVM_REQ_EVENT, vcpu);
f210f757
NA		 396 } else {
397 apic->irr_pending = false;
398 apic_clear_vector(vec, apic->regs + APIC_IRR);
399 if (apic_search_irr(apic) != -1)
400 apic->irr_pending = true;
56cc2406 401 }
33e4c686
GN		 402}
403
8680b94b
MT		 404static inline void apic_set_isr(int vec, struct kvm_lapic *apic)
405{
56cc2406
WL		 406 struct kvm_vcpu *vcpu;
407
408 if (__apic_test_and_set_vector(vec, apic->regs + APIC_ISR))
409 return;
410
411 vcpu = apic->vcpu;
fc57ac2c 412
8680b94b 413 /*
56cc2406
WL		 414 * With APIC virtualization enabled, all caching is disabled
415 * because the processor can modify ISR under the hood. Instead
416 * just set SVI.
8680b94b 417 */
d62caabb 418 if (unlikely(vcpu->arch.apicv_active))
67c9dddc 419 kvm_x86_ops->hwapic_isr_update(vcpu, vec);
56cc2406
WL		 420 else {
421 ++apic->isr_count;
422 BUG_ON(apic->isr_count > MAX_APIC_VECTOR);
423 /*
424 * ISR (in service register) bit is set when injecting an interrupt.
425 * The highest vector is injected. Thus the latest bit set matches
426 * the highest bit in ISR.
427 */
428 apic->highest_isr_cache = vec;
429 }
8680b94b
MT		 430}
431
fc57ac2c
PB		 432static inline int apic_find_highest_isr(struct kvm_lapic *apic)
433{
434 int result;
435
436 /*
437 * Note that isr_count is always 1, and highest_isr_cache
438 * is always -1, with APIC virtualization enabled.
439 */
440 if (!apic->isr_count)
441 return -1;
442 if (likely(apic->highest_isr_cache != -1))
443 return apic->highest_isr_cache;
444
445 result = find_highest_vector(apic->regs + APIC_ISR);
446 ASSERT(result == -1 || result >= 16);
447
448 return result;
449}
450
8680b94b
MT		 451static inline void apic_clear_isr(int vec, struct kvm_lapic *apic)
452{
fc57ac2c
PB		 453 struct kvm_vcpu *vcpu;
454 if (!__apic_test_and_clear_vector(vec, apic->regs + APIC_ISR))
455 return;
456
457 vcpu = apic->vcpu;
458
459 /*
460 * We do get here for APIC virtualization enabled if the guest
461 * uses the Hyper-V APIC enlightenment. In this case we may need
462 * to trigger a new interrupt delivery by writing the SVI field;
463 * on the other hand isr_count and highest_isr_cache are unused
464 * and must be left alone.
465 */
d62caabb 466 if (unlikely(vcpu->arch.apicv_active))
67c9dddc 467 kvm_x86_ops->hwapic_isr_update(vcpu,
fc57ac2c
PB		 468 apic_find_highest_isr(apic));
469 else {
8680b94b 470 --apic->isr_count;
fc57ac2c
PB		 471 BUG_ON(apic->isr_count < 0);
472 apic->highest_isr_cache = -1;
473 }
8680b94b
MT		 474}
475
6e5d865c
YS		 476int kvm_lapic_find_highest_irr(struct kvm_vcpu *vcpu)
477{
33e4c686
GN		 478 /* This may race with setting of irr in __apic_accept_irq() and
479 * value returned may be wrong, but kvm_vcpu_kick() in __apic_accept_irq
480 * will cause vmexit immediately and the value will be recalculated
481 * on the next vmentry.
482 */
f8543d6a 483 return apic_find_highest_irr(vcpu->arch.apic);
6e5d865c 484}
6e5d865c 485
6da7e3f6 486static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
b4f2225c 487 int vector, int level, int trig_mode,
9e4aabe2 488 struct dest_map *dest_map);
6da7e3f6 489
b4f2225c 490int kvm_apic_set_irq(struct kvm_vcpu *vcpu, struct kvm_lapic_irq *irq,
9e4aabe2 491 struct dest_map *dest_map)
97222cc8 492{
ad312c7c 493 struct kvm_lapic *apic = vcpu->arch.apic;
8be5453f 494
58c2dde1 495 return __apic_accept_irq(apic, irq->delivery_mode, irq->vector,
b4f2225c 496 irq->level, irq->trig_mode, dest_map);
97222cc8
ED		 497}
498
ae7a2a3f
MT		 499static int pv_eoi_put_user(struct kvm_vcpu *vcpu, u8 val)
500{
501
502 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, &val,
503 sizeof(val));
504}
505
506static int pv_eoi_get_user(struct kvm_vcpu *vcpu, u8 *val)
507{
508
509 return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, val,
510 sizeof(*val));
511}
512
513static inline bool pv_eoi_enabled(struct kvm_vcpu *vcpu)
514{
515 return vcpu->arch.pv_eoi.msr_val & KVM_MSR_ENABLED;
516}
517
518static bool pv_eoi_get_pending(struct kvm_vcpu *vcpu)
519{
520 u8 val;
521 if (pv_eoi_get_user(vcpu, &val) < 0)
522 apic_debug("Can't read EOI MSR value: 0x%llx\n",
96893977 523 (unsigned long long)vcpu->arch.pv_eoi.msr_val);
ae7a2a3f
MT		 524 return val & 0x1;
525}
526
527static void pv_eoi_set_pending(struct kvm_vcpu *vcpu)
528{
529 if (pv_eoi_put_user(vcpu, KVM_PV_EOI_ENABLED) < 0) {
530 apic_debug("Can't set EOI MSR value: 0x%llx\n",
96893977 531 (unsigned long long)vcpu->arch.pv_eoi.msr_val);
ae7a2a3f
MT		 532 return;
533 }
534 __set_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
535}
536
537static void pv_eoi_clr_pending(struct kvm_vcpu *vcpu)
538{
539 if (pv_eoi_put_user(vcpu, KVM_PV_EOI_DISABLED) < 0) {
540 apic_debug("Can't clear EOI MSR value: 0x%llx\n",
96893977 541 (unsigned long long)vcpu->arch.pv_eoi.msr_val);
ae7a2a3f
MT		 542 return;
543 }
544 __clear_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
545}
546
97222cc8
ED		 547static void apic_update_ppr(struct kvm_lapic *apic)
548{
3842d135 549 u32 tpr, isrv, ppr, old_ppr;
97222cc8
ED		 550 int isr;
551
dfb95954
SS		 552 old_ppr = kvm_lapic_get_reg(apic, APIC_PROCPRI);
553 tpr = kvm_lapic_get_reg(apic, APIC_TASKPRI);
97222cc8
ED		 554 isr = apic_find_highest_isr(apic);
555 isrv = (isr != -1) ? isr : 0;
556
557 if ((tpr & 0xf0) >= (isrv & 0xf0))
558 ppr = tpr & 0xff;
559 else
560 ppr = isrv & 0xf0;
561
562 apic_debug("vlapic %p, ppr 0x%x, isr 0x%x, isrv 0x%x",
563 apic, ppr, isr, isrv);
564
3842d135 565 if (old_ppr != ppr) {
1e6e2755 566 kvm_lapic_set_reg(apic, APIC_PROCPRI, ppr);
83bcacb1
AK		 567 if (ppr < old_ppr)
568 kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
3842d135 569 }
97222cc8
ED		 570}
571
572static void apic_set_tpr(struct kvm_lapic *apic, u32 tpr)
573{
1e6e2755 574 kvm_lapic_set_reg(apic, APIC_TASKPRI, tpr);
97222cc8
ED		 575 apic_update_ppr(apic);
576}
577
03d2249e 578static bool kvm_apic_broadcast(struct kvm_lapic *apic, u32 mda)
394457a9 579{
03d2249e
RK		 580 if (apic_x2apic_mode(apic))
581 return mda == X2APIC_BROADCAST;
582
583 return GET_APIC_DEST_FIELD(mda) == APIC_BROADCAST;
394457a9
NA		 584}
585
03d2249e 586static bool kvm_apic_match_physical_addr(struct kvm_lapic *apic, u32 mda)
97222cc8 587{
03d2249e
RK		 588 if (kvm_apic_broadcast(apic, mda))
589 return true;
590
591 if (apic_x2apic_mode(apic))
592 return mda == kvm_apic_id(apic);
593
594 return mda == SET_APIC_DEST_FIELD(kvm_apic_id(apic));
97222cc8
ED		 595}
596
52c233a4 597static bool kvm_apic_match_logical_addr(struct kvm_lapic *apic, u32 mda)
97222cc8 598{
0105d1a5
GN		 599 u32 logical_id;
600
394457a9 601 if (kvm_apic_broadcast(apic, mda))
9368b567 602 return true;
394457a9 603
dfb95954 604 logical_id = kvm_lapic_get_reg(apic, APIC_LDR);
97222cc8 605
9368b567 606 if (apic_x2apic_mode(apic))
8a395363
RK		 607 return ((logical_id >> 16) == (mda >> 16))
608 && (logical_id & mda & 0xffff) != 0;
97222cc8 609
9368b567 610 logical_id = GET_APIC_LOGICAL_ID(logical_id);
03d2249e 611 mda = GET_APIC_DEST_FIELD(mda);
97222cc8 612
dfb95954 613 switch (kvm_lapic_get_reg(apic, APIC_DFR)) {
97222cc8 614 case APIC_DFR_FLAT:
9368b567 615 return (logical_id & mda) != 0;
97222cc8 616 case APIC_DFR_CLUSTER:
9368b567
RK		 617 return ((logical_id >> 4) == (mda >> 4))
618 && (logical_id & mda & 0xf) != 0;
97222cc8 619 default:
7712de87 620 apic_debug("Bad DFR vcpu %d: %08x\n",
dfb95954 621 apic->vcpu->vcpu_id, kvm_lapic_get_reg(apic, APIC_DFR));
9368b567 622 return false;
97222cc8 623 }
97222cc8
ED		 624}
625
c519265f
RK		 626/* The KVM local APIC implementation has two quirks:
627 *
628 * - the xAPIC MDA stores the destination at bits 24-31, while this
629 * is not true of struct kvm_lapic_irq's dest_id field. This is
630 * just a quirk in the API and is not problematic.
631 *
632 * - in-kernel IOAPIC messages have to be delivered directly to
633 * x2APIC, because the kernel does not support interrupt remapping.
634 * In order to support broadcast without interrupt remapping, x2APIC
635 * rewrites the destination of non-IPI messages from APIC_BROADCAST
636 * to X2APIC_BROADCAST.
637 *
638 * The broadcast quirk can be disabled with KVM_CAP_X2APIC_API. This is
639 * important when userspace wants to use x2APIC-format MSIs, because
640 * APIC_BROADCAST (0xff) is a legal route for "cluster 0, CPUs 0-7".
03d2249e 641 */
c519265f
RK		 642static u32 kvm_apic_mda(struct kvm_vcpu *vcpu, unsigned int dest_id,
643 struct kvm_lapic *source, struct kvm_lapic *target)
03d2249e
RK		 644{
645 bool ipi = source != NULL;
646 bool x2apic_mda = apic_x2apic_mode(ipi ? source : target);
647
c519265f
RK		 648 if (!vcpu->kvm->arch.x2apic_broadcast_quirk_disabled &&
649 !ipi && dest_id == APIC_BROADCAST && x2apic_mda)
03d2249e
RK		 650 return X2APIC_BROADCAST;
651
652 return x2apic_mda ? dest_id : SET_APIC_DEST_FIELD(dest_id);
653}
654
52c233a4 655bool kvm_apic_match_dest(struct kvm_vcpu *vcpu, struct kvm_lapic *source,
394457a9 656 int short_hand, unsigned int dest, int dest_mode)
97222cc8 657{
ad312c7c 658 struct kvm_lapic *target = vcpu->arch.apic;
c519265f 659 u32 mda = kvm_apic_mda(vcpu, dest, source, target);
97222cc8
ED		 660
661 apic_debug("target %p, source %p, dest 0x%x, "
343f94fe 662 "dest_mode 0x%x, short_hand 0x%x\n",
97222cc8
ED		 663 target, source, dest, dest_mode, short_hand);
664
bd371396 665 ASSERT(target);
97222cc8
ED		 666 switch (short_hand) {
667 case APIC_DEST_NOSHORT:
3697f302 668 if (dest_mode == APIC_DEST_PHYSICAL)
03d2249e 669 return kvm_apic_match_physical_addr(target, mda);
343f94fe 670 else
03d2249e 671 return kvm_apic_match_logical_addr(target, mda);
97222cc8 672 case APIC_DEST_SELF:
9368b567 673 return target == source;
97222cc8 674 case APIC_DEST_ALLINC:
9368b567 675 return true;
97222cc8 676 case APIC_DEST_ALLBUT:
9368b567 677 return target != source;
97222cc8 678 default:
7712de87
JK		 679 apic_debug("kvm: apic: Bad dest shorthand value %x\n",
680 short_hand);
9368b567 681 return false;
97222cc8 682 }
97222cc8 683}
1e6e2755 684EXPORT_SYMBOL_GPL(kvm_apic_match_dest);
97222cc8 685
52004014
FW		 686int kvm_vector_to_index(u32 vector, u32 dest_vcpus,
687 const unsigned long *bitmap, u32 bitmap_size)
688{
689 u32 mod;
690 int i, idx = -1;
691
692 mod = vector % dest_vcpus;
693
694 for (i = 0; i <= mod; i++) {
695 idx = find_next_bit(bitmap, bitmap_size, idx + 1);
696 BUG_ON(idx == bitmap_size);
697 }
698
699 return idx;
700}
701
4efd805f
RK		 702static void kvm_apic_disabled_lapic_found(struct kvm *kvm)
703{
704 if (!kvm->arch.disabled_lapic_found) {
705 kvm->arch.disabled_lapic_found = true;
706 printk(KERN_INFO
707 "Disabled LAPIC found during irq injection\n");
708 }
709}
710
c519265f
RK		 711static bool kvm_apic_is_broadcast_dest(struct kvm *kvm, struct kvm_lapic **src,
712 struct kvm_lapic_irq *irq, struct kvm_apic_map *map)
1e08ec4a 713{
c519265f
RK		 714 if (kvm->arch.x2apic_broadcast_quirk_disabled) {
715 if ((irq->dest_id == APIC_BROADCAST &&
716 map->mode != KVM_APIC_MODE_X2APIC))
717 return true;
718 if (irq->dest_id == X2APIC_BROADCAST)
719 return true;
720 } else {
721 bool x2apic_ipi = src && *src && apic_x2apic_mode(*src);
722 if (irq->dest_id == (x2apic_ipi ?
723 X2APIC_BROADCAST : APIC_BROADCAST))
724 return true;
725 }
1e08ec4a 726
c519265f
RK		 727 return false;
728}
1e08ec4a 729
64aa47bf
RK		 730/* Return true if the interrupt can be handled by using *bitmap as index mask
731 * for valid destinations in *dst array.
732 * Return false if kvm_apic_map_get_dest_lapic did nothing useful.
733 * Note: we may have zero kvm_lapic destinations when we return true, which
734 * means that the interrupt should be dropped. In this case, *bitmap would be
735 * zero and *dst undefined.
736 */
737static inline bool kvm_apic_map_get_dest_lapic(struct kvm *kvm,
738 struct kvm_lapic **src, struct kvm_lapic_irq *irq,
739 struct kvm_apic_map *map, struct kvm_lapic ***dst,
740 unsigned long *bitmap)
741{
742 int i, lowest;
1e08ec4a 743
64aa47bf
RK		 744 if (irq->shorthand == APIC_DEST_SELF && src) {
745 *dst = src;
746 *bitmap = 1;
747 return true;
748 } else if (irq->shorthand)
1e08ec4a
GN		 749 return false;
750
c519265f 751 if (!map || kvm_apic_is_broadcast_dest(kvm, src, irq, map))
9ea369b0
RK		 752 return false;
753
64aa47bf 754 if (irq->dest_mode == APIC_DEST_PHYSICAL) {
0ca52e7b 755 if (irq->dest_id > map->max_apic_id) {
64aa47bf
RK		 756 *bitmap = 0;
757 } else {
758 *dst = &map->phys_map[irq->dest_id];
759 *bitmap = 1;
760 }
1e08ec4a 761 return true;
bea15428 762 }
698f9755 763
e45115b6
RK		 764 *bitmap = 0;
765 if (!kvm_apic_map_get_logical_dest(map, irq->dest_id, dst,
766 (u16 *)bitmap))
1e08ec4a 767 return false;
fa834e91 768
64aa47bf
RK		 769 if (!kvm_lowest_prio_delivery(irq))
770 return true;
3548a259 771
64aa47bf
RK		 772 if (!kvm_vector_hashing_enabled()) {
773 lowest = -1;
774 for_each_set_bit(i, bitmap, 16) {
775 if (!(*dst)[i])
776 continue;
777 if (lowest < 0)
778 lowest = i;
779 else if (kvm_apic_compare_prio((*dst)[i]->vcpu,
780 (*dst)[lowest]->vcpu) < 0)
781 lowest = i;
3548a259 782 }
64aa47bf
RK		 783 } else {
784 if (!*bitmap)
785 return true;
3548a259 786
64aa47bf
RK		 787 lowest = kvm_vector_to_index(irq->vector, hweight16(*bitmap),
788 bitmap, 16);
45c3094a 789
64aa47bf
RK		 790 if (!(*dst)[lowest]) {
791 kvm_apic_disabled_lapic_found(kvm);
792 *bitmap = 0;
793 return true;
794 }
795 }
1e08ec4a 796
64aa47bf 797 *bitmap = (lowest >= 0) ? 1 << lowest : 0;
1e08ec4a 798
64aa47bf
RK		 799 return true;
800}
52004014 801
64aa47bf
RK		 802bool kvm_irq_delivery_to_apic_fast(struct kvm *kvm, struct kvm_lapic *src,
803 struct kvm_lapic_irq *irq, int *r, struct dest_map *dest_map)
804{
805 struct kvm_apic_map *map;
806 unsigned long bitmap;
807 struct kvm_lapic **dst = NULL;
808 int i;
809 bool ret;
52004014 810
64aa47bf 811 *r = -1;
52004014 812
64aa47bf
RK		 813 if (irq->shorthand == APIC_DEST_SELF) {
814 *r = kvm_apic_set_irq(src->vcpu, irq, dest_map);
815 return true;
816 }
52004014 817
64aa47bf
RK		 818 rcu_read_lock();
819 map = rcu_dereference(kvm->arch.apic_map);
52004014 820
64aa47bf
RK		 821 ret = kvm_apic_map_get_dest_lapic(kvm, &src, irq, map, &dst, &bitmap);
822 if (ret)
823 for_each_set_bit(i, &bitmap, 16) {
824 if (!dst[i])
825 continue;
826 if (*r < 0)
827 *r = 0;
828 *r += kvm_apic_set_irq(dst[i]->vcpu, irq, dest_map);
1e08ec4a 829 }
1e08ec4a 830
1e08ec4a
GN		 831 rcu_read_unlock();
832 return ret;
833}
834
6228a0da
FW		 835/*
836 * This routine tries to handler interrupts in posted mode, here is how
837 * it deals with different cases:
838 * - For single-destination interrupts, handle it in posted mode
839 * - Else if vector hashing is enabled and it is a lowest-priority
840 * interrupt, handle it in posted mode and use the following mechanism
841 * to find the destinaiton vCPU.
842 * 1. For lowest-priority interrupts, store all the possible
843 * destination vCPUs in an array.
844 * 2. Use "guest vector % max number of destination vCPUs" to find
845 * the right destination vCPU in the array for the lowest-priority
846 * interrupt.
847 * - Otherwise, use remapped mode to inject the interrupt.
848 */
8feb4a04
FW		 849bool kvm_intr_is_single_vcpu_fast(struct kvm *kvm, struct kvm_lapic_irq *irq,
850 struct kvm_vcpu **dest_vcpu)
851{
852 struct kvm_apic_map *map;
64aa47bf
RK		 853 unsigned long bitmap;
854 struct kvm_lapic **dst = NULL;
8feb4a04 855 bool ret = false;
8feb4a04
FW		 856
857 if (irq->shorthand)
858 return false;
859
860 rcu_read_lock();
861 map = rcu_dereference(kvm->arch.apic_map);
862
64aa47bf
RK		 863 if (kvm_apic_map_get_dest_lapic(kvm, NULL, irq, map, &dst, &bitmap) &&
864 hweight16(bitmap) == 1) {
865 unsigned long i = find_first_bit(&bitmap, 16);
6228a0da 866
64aa47bf
RK		 867 if (dst[i]) {
868 *dest_vcpu = dst[i]->vcpu;
869 ret = true;
6228a0da 870 }
8feb4a04
FW		 871 }
872
8feb4a04
FW		 873 rcu_read_unlock();
874 return ret;
875}
876
97222cc8
ED		 877/*
878 * Add a pending IRQ into lapic.
879 * Return 1 if successfully added and 0 if discarded.
880 */
881static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
b4f2225c 882 int vector, int level, int trig_mode,
9e4aabe2 883 struct dest_map *dest_map)
97222cc8 884{
6da7e3f6 885 int result = 0;
c5ec1534 886 struct kvm_vcpu *vcpu = apic->vcpu;
97222cc8 887
a183b638
PB		 888 trace_kvm_apic_accept_irq(vcpu->vcpu_id, delivery_mode,
889 trig_mode, vector);
97222cc8 890 switch (delivery_mode) {
97222cc8 891 case APIC_DM_LOWEST:
e1035715
GN		 892 vcpu->arch.apic_arb_prio++;
893 case APIC_DM_FIXED:
bdaffe1d
PB		 894 if (unlikely(trig_mode && !level))
895 break;
896
97222cc8
ED		 897 /* FIXME add logic for vcpu on reset */
898 if (unlikely(!apic_enabled(apic)))
899 break;
900
11f5cc05
JK		 901 result = 1;
902
9daa5007 903 if (dest_map) {
9e4aabe2 904 __set_bit(vcpu->vcpu_id, dest_map->map);
9daa5007
JR		 905 dest_map->vectors[vcpu->vcpu_id] = vector;
906 }
a5d36f82 907
bdaffe1d
PB		 908 if (apic_test_vector(vector, apic->regs + APIC_TMR) != !!trig_mode) {
909 if (trig_mode)
1e6e2755 910 kvm_lapic_set_vector(vector, apic->regs + APIC_TMR);
bdaffe1d
PB		 911 else
912 apic_clear_vector(vector, apic->regs + APIC_TMR);
913 }
914
d62caabb 915 if (vcpu->arch.apicv_active)
5a71785d 916 kvm_x86_ops->deliver_posted_interrupt(vcpu, vector);
11f5cc05 917 else {
1e6e2755 918 kvm_lapic_set_irr(vector, apic);
5a71785d
YZ		 919
920 kvm_make_request(KVM_REQ_EVENT, vcpu);
921 kvm_vcpu_kick(vcpu);
922 }
97222cc8
ED		 923 break;
924
925 case APIC_DM_REMRD:
24d2166b
R		 926 result = 1;
927 vcpu->arch.pv.pv_unhalted = 1;
928 kvm_make_request(KVM_REQ_EVENT, vcpu);
929 kvm_vcpu_kick(vcpu);
97222cc8
ED		 930 break;
931
932 case APIC_DM_SMI:
64d60670
PB		 933 result = 1;
934 kvm_make_request(KVM_REQ_SMI, vcpu);
935 kvm_vcpu_kick(vcpu);
97222cc8 936 break;
3419ffc8 937
97222cc8 938 case APIC_DM_NMI:
6da7e3f6 939 result = 1;
3419ffc8 940 kvm_inject_nmi(vcpu);
26df99c6 941 kvm_vcpu_kick(vcpu);
97222cc8
ED		 942 break;
943
944 case APIC_DM_INIT:
a52315e1 945 if (!trig_mode || level) {
6da7e3f6 946 result = 1;
66450a21
JK		 947 /* assumes that there are only KVM_APIC_INIT/SIPI */
948 apic->pending_events = (1UL << KVM_APIC_INIT);
949 /* make sure pending_events is visible before sending
950 * the request */
951 smp_wmb();
3842d135 952 kvm_make_request(KVM_REQ_EVENT, vcpu);
c5ec1534
HQ		 953 kvm_vcpu_kick(vcpu);
954 } else {
1b10bf31
JK		 955 apic_debug("Ignoring de-assert INIT to vcpu %d\n",
956 vcpu->vcpu_id);
c5ec1534 957 }
97222cc8
ED		 958 break;
959
960 case APIC_DM_STARTUP:
1b10bf31
JK		 961 apic_debug("SIPI to vcpu %d vector 0x%02x\n",
962 vcpu->vcpu_id, vector);
66450a21
JK		 963 result = 1;
964 apic->sipi_vector = vector;
965 /* make sure sipi_vector is visible for the receiver */
966 smp_wmb();
967 set_bit(KVM_APIC_SIPI, &apic->pending_events);
968 kvm_make_request(KVM_REQ_EVENT, vcpu);
969 kvm_vcpu_kick(vcpu);
97222cc8
ED		 970 break;
971
23930f95
JK		 972 case APIC_DM_EXTINT:
973 /*
974 * Should only be called by kvm_apic_local_deliver() with LVT0,
975 * before NMI watchdog was enabled. Already handled by
976 * kvm_apic_accept_pic_intr().
977 */
978 break;
979
97222cc8
ED		 980 default:
981 printk(KERN_ERR "TODO: unsupported delivery mode %x\n",
982 delivery_mode);
983 break;
984 }
985 return result;
986}
987
e1035715 988int kvm_apic_compare_prio(struct kvm_vcpu *vcpu1, struct kvm_vcpu *vcpu2)
8be5453f 989{
e1035715 990 return vcpu1->arch.apic_arb_prio - vcpu2->arch.apic_arb_prio;
8be5453f
ZX		 991}
992
3bb345f3
PB		 993static bool kvm_ioapic_handles_vector(struct kvm_lapic *apic, int vector)
994{
6308630b 995 return test_bit(vector, apic->vcpu->arch.ioapic_handled_vectors);
3bb345f3
PB		 996}
997
c7c9c56c
YZ		 998static void kvm_ioapic_send_eoi(struct kvm_lapic *apic, int vector)
999{
7543a635
SR		 1000 int trigger_mode;
1001
1002 /* Eoi the ioapic only if the ioapic doesn't own the vector. */
1003 if (!kvm_ioapic_handles_vector(apic, vector))
1004 return;
3bb345f3 1005
7543a635
SR		 1006 /* Request a KVM exit to inform the userspace IOAPIC. */
1007 if (irqchip_split(apic->vcpu->kvm)) {
1008 apic->vcpu->arch.pending_ioapic_eoi = vector;
1009 kvm_make_request(KVM_REQ_IOAPIC_EOI_EXIT, apic->vcpu);
1010 return;
c7c9c56c 1011 }
7543a635
SR		 1012
1013 if (apic_test_vector(vector, apic->regs + APIC_TMR))
1014 trigger_mode = IOAPIC_LEVEL_TRIG;
1015 else
1016 trigger_mode = IOAPIC_EDGE_TRIG;
1017
1018 kvm_ioapic_update_eoi(apic->vcpu, vector, trigger_mode);
c7c9c56c
YZ		 1019}
1020
ae7a2a3f 1021static int apic_set_eoi(struct kvm_lapic *apic)
97222cc8
ED		 1022{
1023 int vector = apic_find_highest_isr(apic);
ae7a2a3f
MT		 1024
1025 trace_kvm_eoi(apic, vector);
1026
97222cc8
ED		 1027 /*
1028 * Not every write EOI will has corresponding ISR,
1029 * one example is when Kernel check timer on setup_IO_APIC
1030 */
1031 if (vector == -1)
ae7a2a3f 1032 return vector;
97222cc8 1033
8680b94b 1034 apic_clear_isr(vector, apic);
97222cc8
ED		 1035 apic_update_ppr(apic);
1036
5c919412
AS		 1037 if (test_bit(vector, vcpu_to_synic(apic->vcpu)->vec_bitmap))
1038 kvm_hv_synic_send_eoi(apic->vcpu, vector);
1039
c7c9c56c 1040 kvm_ioapic_send_eoi(apic, vector);
3842d135 1041 kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
ae7a2a3f 1042 return vector;
97222cc8
ED		 1043}
1044
c7c9c56c
YZ		 1045/*
1046 * this interface assumes a trap-like exit, which has already finished
1047 * desired side effect including vISR and vPPR update.
1048 */
1049void kvm_apic_set_eoi_accelerated(struct kvm_vcpu *vcpu, int vector)
1050{
1051 struct kvm_lapic *apic = vcpu->arch.apic;
1052
1053 trace_kvm_eoi(apic, vector);
1054
1055 kvm_ioapic_send_eoi(apic, vector);
1056 kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
1057}
1058EXPORT_SYMBOL_GPL(kvm_apic_set_eoi_accelerated);
1059
97222cc8
ED		 1060static void apic_send_ipi(struct kvm_lapic *apic)
1061{
dfb95954
SS		 1062 u32 icr_low = kvm_lapic_get_reg(apic, APIC_ICR);
1063 u32 icr_high = kvm_lapic_get_reg(apic, APIC_ICR2);
58c2dde1 1064 struct kvm_lapic_irq irq;
97222cc8 1065
58c2dde1
GN		 1066 irq.vector = icr_low & APIC_VECTOR_MASK;
1067 irq.delivery_mode = icr_low & APIC_MODE_MASK;
1068 irq.dest_mode = icr_low & APIC_DEST_MASK;
b7cb2231 1069 irq.level = (icr_low & APIC_INT_ASSERT) != 0;
58c2dde1
GN		 1070 irq.trig_mode = icr_low & APIC_INT_LEVELTRIG;
1071 irq.shorthand = icr_low & APIC_SHORT_MASK;
93bbf0b8 1072 irq.msi_redir_hint = false;
0105d1a5
GN		 1073 if (apic_x2apic_mode(apic))
1074 irq.dest_id = icr_high;
1075 else
1076 irq.dest_id = GET_APIC_DEST_FIELD(icr_high);
97222cc8 1077
1000ff8d
GN		 1078 trace_kvm_apic_ipi(icr_low, irq.dest_id);
1079
97222cc8
ED		 1080 apic_debug("icr_high 0x%x, icr_low 0x%x, "
1081 "short_hand 0x%x, dest 0x%x, trig_mode 0x%x, level 0x%x, "
93bbf0b8
JS		 1082 "dest_mode 0x%x, delivery_mode 0x%x, vector 0x%x, "
1083 "msi_redir_hint 0x%x\n",
9b5843dd 1084 icr_high, icr_low, irq.shorthand, irq.dest_id,
58c2dde1 1085 irq.trig_mode, irq.level, irq.dest_mode, irq.delivery_mode,
93bbf0b8 1086 irq.vector, irq.msi_redir_hint);
58c2dde1 1087
b4f2225c 1088 kvm_irq_delivery_to_apic(apic->vcpu->kvm, apic, &irq, NULL);
97222cc8
ED		 1089}
1090
1091static u32 apic_get_tmcct(struct kvm_lapic *apic)
1092{
b682b814
MT		 1093 ktime_t remaining;
1094 s64 ns;
9da8f4e8 1095 u32 tmcct;
97222cc8
ED		 1096
1097 ASSERT(apic != NULL);
1098
9da8f4e8 1099 /* if initial count is 0, current count should also be 0 */
dfb95954 1100 if (kvm_lapic_get_reg(apic, APIC_TMICT) == 0 ||
b963a22e 1101 apic->lapic_timer.period == 0)
9da8f4e8
KP		 1102 return 0;
1103
ace15464 1104 remaining = hrtimer_get_remaining(&apic->lapic_timer.timer);
b682b814
MT		 1105 if (ktime_to_ns(remaining) < 0)
1106 remaining = ktime_set(0, 0);
1107
d3c7b77d
MT		 1108 ns = mod_64(ktime_to_ns(remaining), apic->lapic_timer.period);
1109 tmcct = div64_u64(ns,
1110 (APIC_BUS_CYCLE_NS * apic->divide_count));
97222cc8
ED		 1111
1112 return tmcct;
1113}
1114
b209749f
AK		 1115static void __report_tpr_access(struct kvm_lapic *apic, bool write)
1116{
1117 struct kvm_vcpu *vcpu = apic->vcpu;
1118 struct kvm_run *run = vcpu->run;
1119
a8eeb04a 1120 kvm_make_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu);
5fdbf976 1121 run->tpr_access.rip = kvm_rip_read(vcpu);
b209749f
AK		 1122 run->tpr_access.is_write = write;
1123}
1124
1125static inline void report_tpr_access(struct kvm_lapic *apic, bool write)
1126{
1127 if (apic->vcpu->arch.tpr_access_reporting)
1128 __report_tpr_access(apic, write);
1129}
1130
97222cc8
ED		 1131static u32 __apic_read(struct kvm_lapic *apic, unsigned int offset)
1132{
1133 u32 val = 0;
1134
1135 if (offset >= LAPIC_MMIO_LENGTH)
1136 return 0;
1137
1138 switch (offset) {
1139 case APIC_ARBPRI:
7712de87 1140 apic_debug("Access APIC ARBPRI register which is for P6\n");
97222cc8
ED		 1141 break;
1142
1143 case APIC_TMCCT: /* Timer CCR */
a3e06bbe
LJ		 1144 if (apic_lvtt_tscdeadline(apic))
1145 return 0;
1146
97222cc8
ED		 1147 val = apic_get_tmcct(apic);
1148 break;
4a4541a4
AK		 1149 case APIC_PROCPRI:
1150 apic_update_ppr(apic);
dfb95954 1151 val = kvm_lapic_get_reg(apic, offset);
4a4541a4 1152 break;
b209749f
AK		 1153 case APIC_TASKPRI:
1154 report_tpr_access(apic, false);
1155 /* fall thru */
97222cc8 1156 default:
dfb95954 1157 val = kvm_lapic_get_reg(apic, offset);
97222cc8
ED		 1158 break;
1159 }
1160
1161 return val;
1162}
1163
d76685c4
GH		 1164static inline struct kvm_lapic *to_lapic(struct kvm_io_device *dev)
1165{
1166 return container_of(dev, struct kvm_lapic, dev);
1167}
1168
1e6e2755 1169int kvm_lapic_reg_read(struct kvm_lapic *apic, u32 offset, int len,
0105d1a5 1170 void *data)
97222cc8 1171{
97222cc8
ED		 1172 unsigned char alignment = offset & 0xf;
1173 u32 result;
d5b0b5b1 1174 /* this bitmask has a bit cleared for each reserved register */
0105d1a5 1175 static const u64 rmask = 0x43ff01ffffffe70cULL;
97222cc8
ED		 1176
1177 if ((alignment + len) > 4) {
4088bb3c
GN		 1178 apic_debug("KVM_APIC_READ: alignment error %x %d\n",
1179 offset, len);
0105d1a5 1180 return 1;
97222cc8 1181 }
0105d1a5
GN		 1182
1183 if (offset > 0x3f0 || !(rmask & (1ULL << (offset >> 4)))) {
4088bb3c
GN		 1184 apic_debug("KVM_APIC_READ: read reserved register %x\n",
1185 offset);
0105d1a5
GN		 1186 return 1;
1187 }
1188
97222cc8
ED		 1189 result = __apic_read(apic, offset & ~0xf);
1190
229456fc
MT		 1191 trace_kvm_apic_read(offset, result);
1192
97222cc8
ED		 1193 switch (len) {
1194 case 1:
1195 case 2:
1196 case 4:
1197 memcpy(data, (char *)&result + alignment, len);
1198 break;
1199 default:
1200 printk(KERN_ERR "Local APIC read with len = %x, "
1201 "should be 1,2, or 4 instead\n", len);
1202 break;
1203 }
bda9020e 1204 return 0;
97222cc8 1205}
1e6e2755 1206EXPORT_SYMBOL_GPL(kvm_lapic_reg_read);
97222cc8 1207
0105d1a5
GN		 1208static int apic_mmio_in_range(struct kvm_lapic *apic, gpa_t addr)
1209{
c48f1496 1210 return kvm_apic_hw_enabled(apic) &&
0105d1a5
GN		 1211 addr >= apic->base_address &&
1212 addr < apic->base_address + LAPIC_MMIO_LENGTH;
1213}
1214
e32edf4f 1215static int apic_mmio_read(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
0105d1a5
GN		 1216 gpa_t address, int len, void *data)
1217{
1218 struct kvm_lapic *apic = to_lapic(this);
1219 u32 offset = address - apic->base_address;
1220
1221 if (!apic_mmio_in_range(apic, address))
1222 return -EOPNOTSUPP;
1223
1e6e2755 1224 kvm_lapic_reg_read(apic, offset, len, data);
0105d1a5
GN		 1225
1226 return 0;
1227}
1228
97222cc8
ED		 1229static void update_divide_count(struct kvm_lapic *apic)
1230{
1231 u32 tmp1, tmp2, tdcr;
1232
dfb95954 1233 tdcr = kvm_lapic_get_reg(apic, APIC_TDCR);
97222cc8
ED		 1234 tmp1 = tdcr & 0xf;
1235 tmp2 = ((tmp1 & 0x3) | ((tmp1 & 0x8) >> 1)) + 1;
d3c7b77d 1236 apic->divide_count = 0x1 << (tmp2 & 0x7);
97222cc8
ED		 1237
1238 apic_debug("timer divide count is 0x%x\n",
9b5843dd 1239 apic->divide_count);
97222cc8
ED		 1240}
1241
b6ac0695
RK		 1242static void apic_update_lvtt(struct kvm_lapic *apic)
1243{
dfb95954 1244 u32 timer_mode = kvm_lapic_get_reg(apic, APIC_LVTT) &
b6ac0695
RK		 1245 apic->lapic_timer.timer_mode_mask;
1246
1247 if (apic->lapic_timer.timer_mode != timer_mode) {
1248 apic->lapic_timer.timer_mode = timer_mode;
1249 hrtimer_cancel(&apic->lapic_timer.timer);
1250 }
1251}
1252
5d87db71
RK		 1253static void apic_timer_expired(struct kvm_lapic *apic)
1254{
1255 struct kvm_vcpu *vcpu = apic->vcpu;
8577370f 1256 struct swait_queue_head *q = &vcpu->wq;
d0659d94 1257 struct kvm_timer *ktimer = &apic->lapic_timer;
5d87db71 1258
5d87db71
RK		 1259 if (atomic_read(&apic->lapic_timer.pending))
1260 return;
1261
1262 atomic_inc(&apic->lapic_timer.pending);
bab5bb39 1263 kvm_set_pending_timer(vcpu);
5d87db71 1264
8577370f
MT		 1265 if (swait_active(q))
1266 swake_up(q);
d0659d94
MT		 1267
1268 if (apic_lvtt_tscdeadline(apic))
1269 ktimer->expired_tscdeadline = ktimer->tscdeadline;
1270}
1271
1272/*
1273 * On APICv, this test will cause a busy wait
1274 * during a higher-priority task.
1275 */
1276
1277static bool lapic_timer_int_injected(struct kvm_vcpu *vcpu)
1278{
1279 struct kvm_lapic *apic = vcpu->arch.apic;
dfb95954 1280 u32 reg = kvm_lapic_get_reg(apic, APIC_LVTT);
d0659d94
MT		 1281
1282 if (kvm_apic_hw_enabled(apic)) {
1283 int vec = reg & APIC_VECTOR_MASK;
f9339860 1284 void *bitmap = apic->regs + APIC_ISR;
d0659d94 1285
d62caabb 1286 if (vcpu->arch.apicv_active)
f9339860
MT		 1287 bitmap = apic->regs + APIC_IRR;
1288
1289 if (apic_test_vector(vec, bitmap))
1290 return true;
d0659d94
MT		 1291 }
1292 return false;
1293}
1294
1295void wait_lapic_expire(struct kvm_vcpu *vcpu)
1296{
1297 struct kvm_lapic *apic = vcpu->arch.apic;
1298 u64 guest_tsc, tsc_deadline;
1299
bce87cce 1300 if (!lapic_in_kernel(vcpu))
d0659d94
MT		 1301 return;
1302
1303 if (apic->lapic_timer.expired_tscdeadline == 0)
1304 return;
1305
1306 if (!lapic_timer_int_injected(vcpu))
1307 return;
1308
1309 tsc_deadline = apic->lapic_timer.expired_tscdeadline;
1310 apic->lapic_timer.expired_tscdeadline = 0;
4ba76538 1311 guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
6c19b753 1312 trace_kvm_wait_lapic_expire(vcpu->vcpu_id, guest_tsc - tsc_deadline);
d0659d94
MT		 1313
1314 /* __delay is delay_tsc whenever the hardware has TSC, thus always. */
1315 if (guest_tsc < tsc_deadline)
b606f189
MT		 1316 __delay(min(tsc_deadline - guest_tsc,
1317 nsec_to_cycles(vcpu, lapic_timer_advance_ns)));
5d87db71
RK		 1318}
1319
53f9eedf
YJ		 1320static void start_sw_tscdeadline(struct kvm_lapic *apic)
1321{
1322 u64 guest_tsc, tscdeadline = apic->lapic_timer.tscdeadline;
1323 u64 ns = 0;
1324 ktime_t expire;
1325 struct kvm_vcpu *vcpu = apic->vcpu;
1326 unsigned long this_tsc_khz = vcpu->arch.virtual_tsc_khz;
1327 unsigned long flags;
1328 ktime_t now;
1329
1330 if (unlikely(!tscdeadline || !this_tsc_khz))
1331 return;
1332
1333 local_irq_save(flags);
1334
1335 now = apic->lapic_timer.timer.base->get_time();
1336 guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
1337 if (likely(tscdeadline > guest_tsc)) {
1338 ns = (tscdeadline - guest_tsc) * 1000000ULL;
1339 do_div(ns, this_tsc_khz);
1340 expire = ktime_add_ns(now, ns);
1341 expire = ktime_sub_ns(expire, lapic_timer_advance_ns);
1342 hrtimer_start(&apic->lapic_timer.timer,
1343 expire, HRTIMER_MODE_ABS_PINNED);
1344 } else
1345 apic_timer_expired(apic);
1346
1347 local_irq_restore(flags);
1348}
1349
ce7a058a
YJ		 1350bool kvm_lapic_hv_timer_in_use(struct kvm_vcpu *vcpu)
1351{
91005300
WL		 1352 if (!lapic_in_kernel(vcpu))
1353 return false;
1354
ce7a058a
YJ		 1355 return vcpu->arch.apic->lapic_timer.hv_timer_in_use;
1356}
1357EXPORT_SYMBOL_GPL(kvm_lapic_hv_timer_in_use);
1358
bd97ad0e
WL		 1359static void cancel_hv_tscdeadline(struct kvm_lapic *apic)
1360{
1361 kvm_x86_ops->cancel_hv_timer(apic->vcpu);
1362 apic->lapic_timer.hv_timer_in_use = false;
1363}
1364
ce7a058a
YJ		 1365void kvm_lapic_expired_hv_timer(struct kvm_vcpu *vcpu)
1366{
1367 struct kvm_lapic *apic = vcpu->arch.apic;
1368
1369 WARN_ON(!apic->lapic_timer.hv_timer_in_use);
1370 WARN_ON(swait_active(&vcpu->wq));
bd97ad0e 1371 cancel_hv_tscdeadline(apic);
ce7a058a
YJ		 1372 apic_timer_expired(apic);
1373}
1374EXPORT_SYMBOL_GPL(kvm_lapic_expired_hv_timer);
1375
196f20ca
WL		 1376static bool start_hv_tscdeadline(struct kvm_lapic *apic)
1377{
1378 u64 tscdeadline = apic->lapic_timer.tscdeadline;
1379
1380 if (atomic_read(&apic->lapic_timer.pending) ||
1381 kvm_x86_ops->set_hv_timer(apic->vcpu, tscdeadline)) {
1382 if (apic->lapic_timer.hv_timer_in_use)
1383 cancel_hv_tscdeadline(apic);
1384 } else {
1385 apic->lapic_timer.hv_timer_in_use = true;
1386 hrtimer_cancel(&apic->lapic_timer.timer);
1387
1388 /* In case the sw timer triggered in the window */
1389 if (atomic_read(&apic->lapic_timer.pending))
1390 cancel_hv_tscdeadline(apic);
1391 }
1392 trace_kvm_hv_timer_state(apic->vcpu->vcpu_id,
1393 apic->lapic_timer.hv_timer_in_use);
1394 return apic->lapic_timer.hv_timer_in_use;
1395}
1396
ce7a058a
YJ		 1397void kvm_lapic_switch_to_hv_timer(struct kvm_vcpu *vcpu)
1398{
1399 struct kvm_lapic *apic = vcpu->arch.apic;
1400
1401 WARN_ON(apic->lapic_timer.hv_timer_in_use);
1402
196f20ca
WL		 1403 if (apic_lvtt_tscdeadline(apic))
1404 start_hv_tscdeadline(apic);
ce7a058a
YJ		 1405}
1406EXPORT_SYMBOL_GPL(kvm_lapic_switch_to_hv_timer);
1407
1408void kvm_lapic_switch_to_sw_timer(struct kvm_vcpu *vcpu)
1409{
1410 struct kvm_lapic *apic = vcpu->arch.apic;
1411
1412 /* Possibly the TSC deadline timer is not enabled yet */
1413 if (!apic->lapic_timer.hv_timer_in_use)
1414 return;
1415
bd97ad0e 1416 cancel_hv_tscdeadline(apic);
ce7a058a
YJ		 1417
1418 if (atomic_read(&apic->lapic_timer.pending))
1419 return;
1420
1421 start_sw_tscdeadline(apic);
1422}
1423EXPORT_SYMBOL_GPL(kvm_lapic_switch_to_sw_timer);
1424
97222cc8
ED		 1425static void start_apic_timer(struct kvm_lapic *apic)
1426{
a3e06bbe 1427 ktime_t now;
d0659d94 1428
d3c7b77d 1429 atomic_set(&apic->lapic_timer.pending, 0);
0b975a3c 1430
a3e06bbe 1431 if (apic_lvtt_period(apic) || apic_lvtt_oneshot(apic)) {
d5b0b5b1 1432 /* lapic timer in oneshot or periodic mode */
a3e06bbe 1433 now = apic->lapic_timer.timer.base->get_time();
dfb95954 1434 apic->lapic_timer.period = (u64)kvm_lapic_get_reg(apic, APIC_TMICT)
a3e06bbe
LJ		 1435 * APIC_BUS_CYCLE_NS * apic->divide_count;
1436
1437 if (!apic->lapic_timer.period)
1438 return;
1439 /*
1440 * Do not allow the guest to program periodic timers with small
1441 * interval, since the hrtimers are not throttled by the host
1442 * scheduler.
1443 */
1444 if (apic_lvtt_period(apic)) {
1445 s64 min_period = min_timer_period_us * 1000LL;
1446
1447 if (apic->lapic_timer.period < min_period) {
1448 pr_info_ratelimited(
1449 "kvm: vcpu %i: requested %lld ns "
1450 "lapic timer period limited to %lld ns\n",
1451 apic->vcpu->vcpu_id,
1452 apic->lapic_timer.period, min_period);
1453 apic->lapic_timer.period = min_period;
1454 }
9bc5791d 1455 }
0b975a3c 1456
a3e06bbe
LJ		 1457 hrtimer_start(&apic->lapic_timer.timer,
1458 ktime_add_ns(now, apic->lapic_timer.period),
61abdbe0 1459 HRTIMER_MODE_ABS_PINNED);
97222cc8 1460
a3e06bbe 1461 apic_debug("%s: bus cycle is %" PRId64 "ns, now 0x%016"
97222cc8
ED		 1462 PRIx64 ", "
1463 "timer initial count 0x%x, period %lldns, "
b8688d51 1464 "expire @ 0x%016" PRIx64 ".\n", __func__,
97222cc8 1465 APIC_BUS_CYCLE_NS, ktime_to_ns(now),
dfb95954 1466 kvm_lapic_get_reg(apic, APIC_TMICT),
d3c7b77d 1467 apic->lapic_timer.period,
97222cc8 1468 ktime_to_ns(ktime_add_ns(now,
d3c7b77d 1469 apic->lapic_timer.period)));
a3e06bbe 1470 } else if (apic_lvtt_tscdeadline(apic)) {
196f20ca 1471 if (!(kvm_x86_ops->set_hv_timer && start_hv_tscdeadline(apic)))
ce7a058a 1472 start_sw_tscdeadline(apic);
a3e06bbe 1473 }
97222cc8
ED		 1474}
1475
cc6e462c
JK		 1476static void apic_manage_nmi_watchdog(struct kvm_lapic *apic, u32 lvt0_val)
1477{
59fd1323 1478 bool lvt0_in_nmi_mode = apic_lvt_nmi_mode(lvt0_val);
cc6e462c 1479
59fd1323
RK		 1480 if (apic->lvt0_in_nmi_mode != lvt0_in_nmi_mode) {
1481 apic->lvt0_in_nmi_mode = lvt0_in_nmi_mode;
1482 if (lvt0_in_nmi_mode) {
cc6e462c
JK		 1483 apic_debug("Receive NMI setting on APIC_LVT0 "
1484 "for cpu %d\n", apic->vcpu->vcpu_id);
42720138 1485 atomic_inc(&apic->vcpu->kvm->arch.vapics_in_nmi_mode);
59fd1323
RK		 1486 } else
1487 atomic_dec(&apic->vcpu->kvm->arch.vapics_in_nmi_mode);
1488 }
cc6e462c
JK		 1489}
1490
1e6e2755 1491int kvm_lapic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val)
97222cc8 1492{
0105d1a5 1493 int ret = 0;
97222cc8 1494
0105d1a5 1495 trace_kvm_apic_write(reg, val);
97222cc8 1496
0105d1a5 1497 switch (reg) {
97222cc8 1498 case APIC_ID: /* Local APIC ID */
0105d1a5 1499 if (!apic_x2apic_mode(apic))
a92e2543 1500 kvm_apic_set_xapic_id(apic, val >> 24);
0105d1a5
GN		 1501 else
1502 ret = 1;
97222cc8
ED		 1503 break;
1504
1505 case APIC_TASKPRI:
b209749f 1506 report_tpr_access(apic, true);
97222cc8
ED		 1507 apic_set_tpr(apic, val & 0xff);
1508 break;
1509
1510 case APIC_EOI:
1511 apic_set_eoi(apic);
1512 break;
1513
1514 case APIC_LDR:
0105d1a5 1515 if (!apic_x2apic_mode(apic))
1e08ec4a 1516 kvm_apic_set_ldr(apic, val & APIC_LDR_MASK);
0105d1a5
GN		 1517 else
1518 ret = 1;
97222cc8
ED		 1519 break;
1520
1521 case APIC_DFR:
1e08ec4a 1522 if (!apic_x2apic_mode(apic)) {
1e6e2755 1523 kvm_lapic_set_reg(apic, APIC_DFR, val | 0x0FFFFFFF);
1e08ec4a
GN		 1524 recalculate_apic_map(apic->vcpu->kvm);
1525 } else
0105d1a5 1526 ret = 1;
97222cc8
ED		 1527 break;
1528
fc61b800
GN		 1529 case APIC_SPIV: {
1530 u32 mask = 0x3ff;
dfb95954 1531 if (kvm_lapic_get_reg(apic, APIC_LVR) & APIC_LVR_DIRECTED_EOI)
fc61b800 1532 mask |= APIC_SPIV_DIRECTED_EOI;
f8c1ea10 1533 apic_set_spiv(apic, val & mask);
97222cc8
ED		 1534 if (!(val & APIC_SPIV_APIC_ENABLED)) {
1535 int i;
1536 u32 lvt_val;
1537
1e6e2755 1538 for (i = 0; i < KVM_APIC_LVT_NUM; i++) {
dfb95954 1539 lvt_val = kvm_lapic_get_reg(apic,
97222cc8 1540 APIC_LVTT + 0x10 * i);
1e6e2755 1541 kvm_lapic_set_reg(apic, APIC_LVTT + 0x10 * i,
97222cc8
ED		 1542 lvt_val | APIC_LVT_MASKED);
1543 }
b6ac0695 1544 apic_update_lvtt(apic);
d3c7b77d 1545 atomic_set(&apic->lapic_timer.pending, 0);
97222cc8
ED		 1546
1547 }
1548 break;
fc61b800 1549 }
97222cc8
ED		 1550 case APIC_ICR:
1551 /* No delay here, so we always clear the pending bit */
1e6e2755 1552 kvm_lapic_set_reg(apic, APIC_ICR, val & ~(1 << 12));
97222cc8
ED		 1553 apic_send_ipi(apic);
1554 break;
1555
1556 case APIC_ICR2:
0105d1a5
GN		 1557 if (!apic_x2apic_mode(apic))
1558 val &= 0xff000000;
1e6e2755 1559 kvm_lapic_set_reg(apic, APIC_ICR2, val);
97222cc8
ED		 1560 break;
1561
23930f95 1562 case APIC_LVT0:
cc6e462c 1563 apic_manage_nmi_watchdog(apic, val);
97222cc8
ED		 1564 case APIC_LVTTHMR:
1565 case APIC_LVTPC:
97222cc8
ED		 1566 case APIC_LVT1:
1567 case APIC_LVTERR:
1568 /* TODO: Check vector */
c48f1496 1569 if (!kvm_apic_sw_enabled(apic))
97222cc8
ED		 1570 val |= APIC_LVT_MASKED;
1571
0105d1a5 1572 val &= apic_lvt_mask[(reg - APIC_LVTT) >> 4];
1e6e2755 1573 kvm_lapic_set_reg(apic, reg, val);
97222cc8
ED		 1574
1575 break;
1576
b6ac0695 1577 case APIC_LVTT:
c48f1496 1578 if (!kvm_apic_sw_enabled(apic))
a3e06bbe
LJ		 1579 val |= APIC_LVT_MASKED;
1580 val &= (apic_lvt_mask[0] | apic->lapic_timer.timer_mode_mask);
1e6e2755 1581 kvm_lapic_set_reg(apic, APIC_LVTT, val);
b6ac0695 1582 apic_update_lvtt(apic);
a3e06bbe
LJ		 1583 break;
1584
97222cc8 1585 case APIC_TMICT:
a3e06bbe
LJ		 1586 if (apic_lvtt_tscdeadline(apic))
1587 break;
1588
d3c7b77d 1589 hrtimer_cancel(&apic->lapic_timer.timer);
1e6e2755 1590 kvm_lapic_set_reg(apic, APIC_TMICT, val);
97222cc8 1591 start_apic_timer(apic);
0105d1a5 1592 break;
97222cc8
ED		 1593
1594 case APIC_TDCR:
1595 if (val & 4)
7712de87 1596 apic_debug("KVM_WRITE:TDCR %x\n", val);
1e6e2755 1597 kvm_lapic_set_reg(apic, APIC_TDCR, val);
97222cc8
ED		 1598 update_divide_count(apic);
1599 break;
1600
0105d1a5
GN		 1601 case APIC_ESR:
1602 if (apic_x2apic_mode(apic) && val != 0) {
7712de87 1603 apic_debug("KVM_WRITE:ESR not zero %x\n", val);
0105d1a5
GN		 1604 ret = 1;
1605 }
1606 break;
1607
1608 case APIC_SELF_IPI:
1609 if (apic_x2apic_mode(apic)) {
1e6e2755 1610 kvm_lapic_reg_write(apic, APIC_ICR, 0x40000 | (val & 0xff));
0105d1a5
GN		 1611 } else
1612 ret = 1;
1613 break;
97222cc8 1614 default:
0105d1a5 1615 ret = 1;
97222cc8
ED		 1616 break;
1617 }
0105d1a5
GN		 1618 if (ret)
1619 apic_debug("Local APIC Write to read-only register %x\n", reg);
1620 return ret;
1621}
1e6e2755 1622EXPORT_SYMBOL_GPL(kvm_lapic_reg_write);
0105d1a5 1623
e32edf4f 1624static int apic_mmio_write(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
0105d1a5
GN		 1625 gpa_t address, int len, const void *data)
1626{
1627 struct kvm_lapic *apic = to_lapic(this);
1628 unsigned int offset = address - apic->base_address;
1629 u32 val;
1630
1631 if (!apic_mmio_in_range(apic, address))
1632 return -EOPNOTSUPP;
1633
1634 /*
1635 * APIC register must be aligned on 128-bits boundary.
1636 * 32/64/128 bits registers must be accessed thru 32 bits.
1637 * Refer SDM 8.4.1
1638 */
1639 if (len != 4 || (offset & 0xf)) {
1640 /* Don't shout loud, $infamous_os would cause only noise. */
1641 apic_debug("apic write: bad size=%d %lx\n", len, (long)address);
756975bb 1642 return 0;
0105d1a5
GN		 1643 }
1644
1645 val = *(u32*)data;
1646
1647 /* too common printing */
1648 if (offset != APIC_EOI)
1649 apic_debug("%s: offset 0x%x with length 0x%x, and value is "
1650 "0x%x\n", __func__, offset, len, val);
1651
1e6e2755 1652 kvm_lapic_reg_write(apic, offset & 0xff0, val);
0105d1a5 1653
bda9020e 1654 return 0;
97222cc8
ED		 1655}
1656
58fbbf26
KT		 1657void kvm_lapic_set_eoi(struct kvm_vcpu *vcpu)
1658{
1e6e2755 1659 kvm_lapic_reg_write(vcpu->arch.apic, APIC_EOI, 0);
58fbbf26
KT		 1660}
1661EXPORT_SYMBOL_GPL(kvm_lapic_set_eoi);
1662
83d4c286
YZ		 1663/* emulate APIC access in a trap manner */
1664void kvm_apic_write_nodecode(struct kvm_vcpu *vcpu, u32 offset)
1665{
1666 u32 val = 0;
1667
1668 /* hw has done the conditional check and inst decode */
1669 offset &= 0xff0;
1670
1e6e2755 1671 kvm_lapic_reg_read(vcpu->arch.apic, offset, 4, &val);
83d4c286
YZ		 1672
1673 /* TODO: optimize to just emulate side effect w/o one more write */
1e6e2755 1674 kvm_lapic_reg_write(vcpu->arch.apic, offset, val);
83d4c286
YZ		 1675}
1676EXPORT_SYMBOL_GPL(kvm_apic_write_nodecode);
1677
d589444e 1678void kvm_free_lapic(struct kvm_vcpu *vcpu)
97222cc8 1679{
f8c1ea10
GN		 1680 struct kvm_lapic *apic = vcpu->arch.apic;
1681
ad312c7c 1682 if (!vcpu->arch.apic)
97222cc8
ED		 1683 return;
1684
f8c1ea10 1685 hrtimer_cancel(&apic->lapic_timer.timer);
97222cc8 1686
c5cc421b
GN		 1687 if (!(vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE))
1688 static_key_slow_dec_deferred(&apic_hw_disabled);
1689
e462755c 1690 if (!apic->sw_enabled)
f8c1ea10 1691 static_key_slow_dec_deferred(&apic_sw_disabled);
97222cc8 1692
f8c1ea10
GN		 1693 if (apic->regs)
1694 free_page((unsigned long)apic->regs);
1695
1696 kfree(apic);
97222cc8
ED		 1697}
1698
1699/*
1700 *----------------------------------------------------------------------
1701 * LAPIC interface
1702 *----------------------------------------------------------------------
1703 */
1704
a3e06bbe
LJ		 1705u64 kvm_get_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu)
1706{
1707 struct kvm_lapic *apic = vcpu->arch.apic;
a3e06bbe 1708
bce87cce 1709 if (!lapic_in_kernel(vcpu) || apic_lvtt_oneshot(apic) ||
54e9818f 1710 apic_lvtt_period(apic))
a3e06bbe
LJ		 1711 return 0;
1712
1713 return apic->lapic_timer.tscdeadline;
1714}
1715
1716void kvm_set_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu, u64 data)
1717{
1718 struct kvm_lapic *apic = vcpu->arch.apic;
a3e06bbe 1719
bce87cce 1720 if (!lapic_in_kernel(vcpu) || apic_lvtt_oneshot(apic) ||
54e9818f 1721 apic_lvtt_period(apic))
a3e06bbe
LJ		 1722 return;
1723
1724 hrtimer_cancel(&apic->lapic_timer.timer);
1725 apic->lapic_timer.tscdeadline = data;
1726 start_apic_timer(apic);
1727}
1728
97222cc8
ED		 1729void kvm_lapic_set_tpr(struct kvm_vcpu *vcpu, unsigned long cr8)
1730{
ad312c7c 1731 struct kvm_lapic *apic = vcpu->arch.apic;
97222cc8 1732
b93463aa 1733 apic_set_tpr(apic, ((cr8 & 0x0f) << 4)
dfb95954 1734 | (kvm_lapic_get_reg(apic, APIC_TASKPRI) & 4));
97222cc8
ED		 1735}
1736
1737u64 kvm_lapic_get_cr8(struct kvm_vcpu *vcpu)
1738{
97222cc8
ED		 1739 u64 tpr;
1740
dfb95954 1741 tpr = (u64) kvm_lapic_get_reg(vcpu->arch.apic, APIC_TASKPRI);
97222cc8
ED		 1742
1743 return (tpr & 0xf0) >> 4;
1744}
1745
1746void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
1747{
8d14695f 1748 u64 old_value = vcpu->arch.apic_base;
ad312c7c 1749 struct kvm_lapic *apic = vcpu->arch.apic;
97222cc8
ED		 1750
1751 if (!apic) {
1752 value |= MSR_IA32_APICBASE_BSP;
ad312c7c 1753 vcpu->arch.apic_base = value;
97222cc8
ED		 1754 return;
1755 }
c5af89b6 1756
e66d2ae7
JK		 1757 vcpu->arch.apic_base = value;
1758
c5cc421b 1759 /* update jump label if enable bit changes */
0dce7cd6 1760 if ((old_value ^ value) & MSR_IA32_APICBASE_ENABLE) {
49bd29ba
RK		 1761 if (value & MSR_IA32_APICBASE_ENABLE) {
1762 kvm_apic_set_xapic_id(apic, vcpu->vcpu_id);
c5cc421b 1763 static_key_slow_dec_deferred(&apic_hw_disabled);
49bd29ba 1764 } else
c5cc421b 1765 static_key_slow_inc(&apic_hw_disabled.key);
1e08ec4a 1766 recalculate_apic_map(vcpu->kvm);
c5cc421b
GN		 1767 }
1768
8d14695f
YZ		 1769 if ((old_value ^ value) & X2APIC_ENABLE) {
1770 if (value & X2APIC_ENABLE) {
257b9a5f 1771 kvm_apic_set_x2apic_id(apic, vcpu->vcpu_id);
8d14695f
YZ		 1772 kvm_x86_ops->set_virtual_x2apic_mode(vcpu, true);
1773 } else
1774 kvm_x86_ops->set_virtual_x2apic_mode(vcpu, false);
0105d1a5 1775 }
8d14695f 1776
ad312c7c 1777 apic->base_address = apic->vcpu->arch.apic_base &
97222cc8
ED		 1778 MSR_IA32_APICBASE_BASE;
1779
db324fe6
NA		 1780 if ((value & MSR_IA32_APICBASE_ENABLE) &&
1781 apic->base_address != APIC_DEFAULT_PHYS_BASE)
1782 pr_warn_once("APIC base relocation is unsupported by KVM");
1783
97222cc8
ED		 1784 /* with FSB delivery interrupt, we can restart APIC functionality */
1785 apic_debug("apic base msr is 0x%016" PRIx64 ", and base address is "
ad312c7c 1786 "0x%lx.\n", apic->vcpu->arch.apic_base, apic->base_address);
97222cc8
ED		 1787
1788}
1789
d28bc9dd 1790void kvm_lapic_reset(struct kvm_vcpu *vcpu, bool init_event)
97222cc8
ED		 1791{
1792 struct kvm_lapic *apic;
1793 int i;
1794
b8688d51 1795 apic_debug("%s\n", __func__);
97222cc8
ED		 1796
1797 ASSERT(vcpu);
ad312c7c 1798 apic = vcpu->arch.apic;
97222cc8
ED		 1799 ASSERT(apic != NULL);
1800
1801 /* Stop the timer in case it's a reset to an active apic */
d3c7b77d 1802 hrtimer_cancel(&apic->lapic_timer.timer);
97222cc8 1803
4d8e772b
RK		 1804 if (!init_event) {
1805 kvm_lapic_set_base(vcpu, APIC_DEFAULT_PHYS_BASE |
1806 MSR_IA32_APICBASE_ENABLE);
a92e2543 1807 kvm_apic_set_xapic_id(apic, vcpu->vcpu_id);
4d8e772b 1808 }
fc61b800 1809 kvm_apic_set_version(apic->vcpu);
97222cc8 1810
1e6e2755
SS		 1811 for (i = 0; i < KVM_APIC_LVT_NUM; i++)
1812 kvm_lapic_set_reg(apic, APIC_LVTT + 0x10 * i, APIC_LVT_MASKED);
b6ac0695 1813 apic_update_lvtt(apic);
0da029ed 1814 if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_LINT0_REENABLED))
1e6e2755 1815 kvm_lapic_set_reg(apic, APIC_LVT0,
90de4a18 1816 SET_APIC_DELIVERY_MODE(0, APIC_MODE_EXTINT));
dfb95954 1817 apic_manage_nmi_watchdog(apic, kvm_lapic_get_reg(apic, APIC_LVT0));
97222cc8 1818
1e6e2755 1819 kvm_lapic_set_reg(apic, APIC_DFR, 0xffffffffU);
f8c1ea10 1820 apic_set_spiv(apic, 0xff);
1e6e2755 1821 kvm_lapic_set_reg(apic, APIC_TASKPRI, 0);
c028dd6b
RK		 1822 if (!apic_x2apic_mode(apic))
1823 kvm_apic_set_ldr(apic, 0);
1e6e2755
SS		 1824 kvm_lapic_set_reg(apic, APIC_ESR, 0);
1825 kvm_lapic_set_reg(apic, APIC_ICR, 0);
1826 kvm_lapic_set_reg(apic, APIC_ICR2, 0);
1827 kvm_lapic_set_reg(apic, APIC_TDCR, 0);
1828 kvm_lapic_set_reg(apic, APIC_TMICT, 0);
97222cc8 1829 for (i = 0; i < 8; i++) {
1e6e2755
SS		 1830 kvm_lapic_set_reg(apic, APIC_IRR + 0x10 * i, 0);
1831 kvm_lapic_set_reg(apic, APIC_ISR + 0x10 * i, 0);
1832 kvm_lapic_set_reg(apic, APIC_TMR + 0x10 * i, 0);
97222cc8 1833 }
d62caabb
AS		 1834 apic->irr_pending = vcpu->arch.apicv_active;
1835 apic->isr_count = vcpu->arch.apicv_active ? 1 : 0;
8680b94b 1836 apic->highest_isr_cache = -1;
b33ac88b 1837 update_divide_count(apic);
d3c7b77d 1838 atomic_set(&apic->lapic_timer.pending, 0);
c5af89b6 1839 if (kvm_vcpu_is_bsp(vcpu))
5dbc8f3f
GN		 1840 kvm_lapic_set_base(vcpu,
1841 vcpu->arch.apic_base | MSR_IA32_APICBASE_BSP);
ae7a2a3f 1842 vcpu->arch.pv_eoi.msr_val = 0;
97222cc8
ED		 1843 apic_update_ppr(apic);
1844
e1035715 1845 vcpu->arch.apic_arb_prio = 0;
41383771 1846 vcpu->arch.apic_attention = 0;
e1035715 1847
98eff52a 1848 apic_debug("%s: vcpu=%p, id=%d, base_msr="
b8688d51 1849 "0x%016" PRIx64 ", base_address=0x%0lx.\n", __func__,
97222cc8 1850 vcpu, kvm_apic_id(apic),
ad312c7c 1851 vcpu->arch.apic_base, apic->base_address);
97222cc8
ED		 1852}
1853
97222cc8
ED		 1854/*
1855 *----------------------------------------------------------------------
1856 * timer interface
1857 *----------------------------------------------------------------------
1858 */
1b9778da 1859
2a6eac96 1860static bool lapic_is_periodic(struct kvm_lapic *apic)
97222cc8 1861{
d3c7b77d 1862 return apic_lvtt_period(apic);
97222cc8
ED		 1863}
1864
3d80840d
MT		 1865int apic_has_pending_timer(struct kvm_vcpu *vcpu)
1866{
54e9818f 1867 struct kvm_lapic *apic = vcpu->arch.apic;
3d80840d 1868
1e3161b4 1869 if (apic_enabled(apic) && apic_lvt_enabled(apic, APIC_LVTT))
54e9818f 1870 return atomic_read(&apic->lapic_timer.pending);
3d80840d
MT		 1871
1872 return 0;
1873}
1874
89342082 1875int kvm_apic_local_deliver(struct kvm_lapic *apic, int lvt_type)
1b9778da 1876{
dfb95954 1877 u32 reg = kvm_lapic_get_reg(apic, lvt_type);
23930f95 1878 int vector, mode, trig_mode;
23930f95 1879
c48f1496 1880 if (kvm_apic_hw_enabled(apic) && !(reg & APIC_LVT_MASKED)) {
23930f95
JK		 1881 vector = reg & APIC_VECTOR_MASK;
1882 mode = reg & APIC_MODE_MASK;
1883 trig_mode = reg & APIC_LVT_LEVEL_TRIGGER;
b4f2225c
YZ		 1884 return __apic_accept_irq(apic, mode, vector, 1, trig_mode,
1885 NULL);
23930f95
JK		 1886 }
1887 return 0;
1888}
1b9778da 1889
8fdb2351 1890void kvm_apic_nmi_wd_deliver(struct kvm_vcpu *vcpu)
23930f95 1891{
8fdb2351
JK		 1892 struct kvm_lapic *apic = vcpu->arch.apic;
1893
1894 if (apic)
1895 kvm_apic_local_deliver(apic, APIC_LVT0);
1b9778da
ED		 1896}
1897
d76685c4
GH		 1898static const struct kvm_io_device_ops apic_mmio_ops = {
1899 .read = apic_mmio_read,
1900 .write = apic_mmio_write,
d76685c4
GH		 1901};
1902
e9d90d47
AK		 1903static enum hrtimer_restart apic_timer_fn(struct hrtimer *data)
1904{
1905 struct kvm_timer *ktimer = container_of(data, struct kvm_timer, timer);
2a6eac96 1906 struct kvm_lapic *apic = container_of(ktimer, struct kvm_lapic, lapic_timer);
e9d90d47 1907
5d87db71 1908 apic_timer_expired(apic);
e9d90d47 1909
2a6eac96 1910 if (lapic_is_periodic(apic)) {
e9d90d47
AK		 1911 hrtimer_add_expires_ns(&ktimer->timer, ktimer->period);
1912 return HRTIMER_RESTART;
1913 } else
1914 return HRTIMER_NORESTART;
1915}
1916
97222cc8
ED		 1917int kvm_create_lapic(struct kvm_vcpu *vcpu)
1918{
1919 struct kvm_lapic *apic;
1920
1921 ASSERT(vcpu != NULL);
1922 apic_debug("apic_init %d\n", vcpu->vcpu_id);
1923
1924 apic = kzalloc(sizeof(*apic), GFP_KERNEL);
1925 if (!apic)
1926 goto nomem;
1927
ad312c7c 1928 vcpu->arch.apic = apic;
97222cc8 1929
afc20184
TY		 1930 apic->regs = (void *)get_zeroed_page(GFP_KERNEL);
1931 if (!apic->regs) {
97222cc8
ED		 1932 printk(KERN_ERR "malloc apic regs error for vcpu %x\n",
1933 vcpu->vcpu_id);
d589444e 1934 goto nomem_free_apic;
97222cc8 1935 }
97222cc8
ED		 1936 apic->vcpu = vcpu;
1937
d3c7b77d 1938 hrtimer_init(&apic->lapic_timer.timer, CLOCK_MONOTONIC,
61abdbe0 1939 HRTIMER_MODE_ABS_PINNED);
e9d90d47 1940 apic->lapic_timer.timer.function = apic_timer_fn;
d3c7b77d 1941
c5cc421b
GN		 1942 /*
1943 * APIC is created enabled. This will prevent kvm_lapic_set_base from
1944 * thinking that APIC satet has changed.
1945 */
1946 vcpu->arch.apic_base = MSR_IA32_APICBASE_ENABLE;
f8c1ea10 1947 static_key_slow_inc(&apic_sw_disabled.key); /* sw disabled at reset */
d28bc9dd 1948 kvm_lapic_reset(vcpu, false);
d76685c4 1949 kvm_iodevice_init(&apic->dev, &apic_mmio_ops);
97222cc8
ED		 1950
1951 return 0;
d589444e
RR		 1952nomem_free_apic:
1953 kfree(apic);
97222cc8 1954nomem:
97222cc8
ED		 1955 return -ENOMEM;
1956}
97222cc8
ED		 1957
1958int kvm_apic_has_interrupt(struct kvm_vcpu *vcpu)
1959{
ad312c7c 1960 struct kvm_lapic *apic = vcpu->arch.apic;
97222cc8
ED		 1961 int highest_irr;
1962
f8543d6a 1963 if (!apic_enabled(apic))
97222cc8
ED		 1964 return -1;
1965
6e5d865c 1966 apic_update_ppr(apic);
97222cc8
ED		 1967 highest_irr = apic_find_highest_irr(apic);
1968 if ((highest_irr == -1) ||
dfb95954 1969 ((highest_irr & 0xF0) <= kvm_lapic_get_reg(apic, APIC_PROCPRI)))
97222cc8
ED		 1970 return -1;
1971 return highest_irr;
1972}
1973
40487c68
QH		 1974int kvm_apic_accept_pic_intr(struct kvm_vcpu *vcpu)
1975{
dfb95954 1976 u32 lvt0 = kvm_lapic_get_reg(vcpu->arch.apic, APIC_LVT0);
40487c68
QH		 1977 int r = 0;
1978
c48f1496 1979 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
e7dca5c0
CL		 1980 r = 1;
1981 if ((lvt0 & APIC_LVT_MASKED) == 0 &&
1982 GET_APIC_DELIVERY_MODE(lvt0) == APIC_MODE_EXTINT)
1983 r = 1;
40487c68
QH		 1984 return r;
1985}
1986
1b9778da
ED		 1987void kvm_inject_apic_timer_irqs(struct kvm_vcpu *vcpu)
1988{
ad312c7c 1989 struct kvm_lapic *apic = vcpu->arch.apic;
1b9778da 1990
54e9818f 1991 if (atomic_read(&apic->lapic_timer.pending) > 0) {
f1ed0450 1992 kvm_apic_local_deliver(apic, APIC_LVTT);
fae0ba21
NA		 1993 if (apic_lvtt_tscdeadline(apic))
1994 apic->lapic_timer.tscdeadline = 0;
f1ed0450 1995 atomic_set(&apic->lapic_timer.pending, 0);
1b9778da
ED		 1996 }
1997}
1998
97222cc8
ED		 1999int kvm_get_apic_interrupt(struct kvm_vcpu *vcpu)
2000{
2001 int vector = kvm_apic_has_interrupt(vcpu);
ad312c7c 2002 struct kvm_lapic *apic = vcpu->arch.apic;
97222cc8
ED		 2003
2004 if (vector == -1)
2005 return -1;
2006
56cc2406
WL		 2007 /*
2008 * We get here even with APIC virtualization enabled, if doing
2009 * nested virtualization and L1 runs with the "acknowledge interrupt
2010 * on exit" mode. Then we cannot inject the interrupt via RVI,
2011 * because the process would deliver it through the IDT.
2012 */
2013
8680b94b 2014 apic_set_isr(vector, apic);
97222cc8
ED		 2015 apic_update_ppr(apic);
2016 apic_clear_irr(vector, apic);
5c919412
AS		 2017
2018 if (test_bit(vector, vcpu_to_synic(vcpu)->auto_eoi_bitmap)) {
2019 apic_clear_isr(vector, apic);
2020 apic_update_ppr(apic);
2021 }
2022
97222cc8
ED		 2023 return vector;
2024}
96ad2cc6 2025
a92e2543
RK		 2026static int kvm_apic_state_fixup(struct kvm_vcpu *vcpu,
2027 struct kvm_lapic_state *s, bool set)
2028{
2029 if (apic_x2apic_mode(vcpu->arch.apic)) {
2030 u32 *id = (u32 *)(s->regs + APIC_ID);
2031
37131313
RK		 2032 if (vcpu->kvm->arch.x2apic_format) {
2033 if (*id != vcpu->vcpu_id)
2034 return -EINVAL;
2035 } else {
2036 if (set)
2037 *id >>= 24;
2038 else
2039 *id <<= 24;
2040 }
a92e2543
RK		 2041 }
2042
2043 return 0;
2044}
2045
2046int kvm_apic_get_state(struct kvm_vcpu *vcpu, struct kvm_lapic_state *s)
2047{
2048 memcpy(s->regs, vcpu->arch.apic->regs, sizeof(*s));
2049 return kvm_apic_state_fixup(vcpu, s, false);
2050}
2051
2052int kvm_apic_set_state(struct kvm_vcpu *vcpu, struct kvm_lapic_state *s)
96ad2cc6 2053{
ad312c7c 2054 struct kvm_lapic *apic = vcpu->arch.apic;
a92e2543
RK		 2055 int r;
2056
96ad2cc6 2057
5dbc8f3f 2058 kvm_lapic_set_base(vcpu, vcpu->arch.apic_base);
64eb0620
GN		 2059 /* set SPIV separately to get count of SW disabled APICs right */
2060 apic_set_spiv(apic, *((u32 *)(s->regs + APIC_SPIV)));
a92e2543
RK		 2061
2062 r = kvm_apic_state_fixup(vcpu, s, true);
2063 if (r)
2064 return r;
64eb0620 2065 memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
a92e2543
RK		 2066
2067 recalculate_apic_map(vcpu->kvm);
fc61b800
GN		 2068 kvm_apic_set_version(vcpu);
2069
96ad2cc6 2070 apic_update_ppr(apic);
d3c7b77d 2071 hrtimer_cancel(&apic->lapic_timer.timer);
b6ac0695 2072 apic_update_lvtt(apic);
dfb95954 2073 apic_manage_nmi_watchdog(apic, kvm_lapic_get_reg(apic, APIC_LVT0));
96ad2cc6
ED		 2074 update_divide_count(apic);
2075 start_apic_timer(apic);
6e24a6ef 2076 apic->irr_pending = true;
d62caabb 2077 apic->isr_count = vcpu->arch.apicv_active ?
c7c9c56c 2078 1 : count_vectors(apic->regs + APIC_ISR);
8680b94b 2079 apic->highest_isr_cache = -1;
d62caabb 2080 if (vcpu->arch.apicv_active) {
be8ca170
SS		 2081 if (kvm_x86_ops->apicv_post_state_restore)
2082 kvm_x86_ops->apicv_post_state_restore(vcpu);
4114c27d
WW		 2083 kvm_x86_ops->hwapic_irr_update(vcpu,
2084 apic_find_highest_irr(apic));
67c9dddc 2085 kvm_x86_ops->hwapic_isr_update(vcpu,
b4eef9b3 2086 apic_find_highest_isr(apic));
d62caabb 2087 }
3842d135 2088 kvm_make_request(KVM_REQ_EVENT, vcpu);
49df6397
SR		 2089 if (ioapic_in_kernel(vcpu->kvm))
2090 kvm_rtc_eoi_tracking_restore_one(vcpu);
0669a510
RK		 2091
2092 vcpu->arch.apic_arb_prio = 0;
a92e2543
RK		 2093
2094 return 0;
96ad2cc6 2095}
a3d7f85f 2096
2f52d58c 2097void __kvm_migrate_apic_timer(struct kvm_vcpu *vcpu)
a3d7f85f 2098{
a3d7f85f
ED		 2099 struct hrtimer *timer;
2100
bce87cce 2101 if (!lapic_in_kernel(vcpu))
a3d7f85f
ED		 2102 return;
2103
54e9818f 2104 timer = &vcpu->arch.apic->lapic_timer.timer;
a3d7f85f 2105 if (hrtimer_cancel(timer))
61abdbe0 2106 hrtimer_start_expires(timer, HRTIMER_MODE_ABS_PINNED);
a3d7f85f 2107}
b93463aa 2108
ae7a2a3f
MT		 2109/*
2110 * apic_sync_pv_eoi_from_guest - called on vmexit or cancel interrupt
2111 *
2112 * Detect whether guest triggered PV EOI since the
2113 * last entry. If yes, set EOI on guests's behalf.
2114 * Clear PV EOI in guest memory in any case.
2115 */
2116static void apic_sync_pv_eoi_from_guest(struct kvm_vcpu *vcpu,
2117 struct kvm_lapic *apic)
2118{
2119 bool pending;
2120 int vector;
2121 /*
2122 * PV EOI state is derived from KVM_APIC_PV_EOI_PENDING in host
2123 * and KVM_PV_EOI_ENABLED in guest memory as follows:
2124 *
2125 * KVM_APIC_PV_EOI_PENDING is unset:
2126 * -> host disabled PV EOI.
2127 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is set:
2128 * -> host enabled PV EOI, guest did not execute EOI yet.
2129 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is unset:
2130 * -> host enabled PV EOI, guest executed EOI.
2131 */
2132 BUG_ON(!pv_eoi_enabled(vcpu));
2133 pending = pv_eoi_get_pending(vcpu);
2134 /*
2135 * Clear pending bit in any case: it will be set again on vmentry.
2136 * While this might not be ideal from performance point of view,
2137 * this makes sure pv eoi is only enabled when we know it's safe.
2138 */
2139 pv_eoi_clr_pending(vcpu);
2140 if (pending)
2141 return;
2142 vector = apic_set_eoi(apic);
2143 trace_kvm_pv_eoi(apic, vector);
2144}
2145
b93463aa
AK		 2146void kvm_lapic_sync_from_vapic(struct kvm_vcpu *vcpu)
2147{
2148 u32 data;
b93463aa 2149
ae7a2a3f
MT		 2150 if (test_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention))
2151 apic_sync_pv_eoi_from_guest(vcpu, vcpu->arch.apic);
2152
41383771 2153 if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
b93463aa
AK		 2154 return;
2155
603242a8
NK		 2156 if (kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
2157 sizeof(u32)))
2158 return;
b93463aa
AK		 2159
2160 apic_set_tpr(vcpu->arch.apic, data & 0xff);
2161}
2162
ae7a2a3f
MT		 2163/*
2164 * apic_sync_pv_eoi_to_guest - called before vmentry
2165 *
2166 * Detect whether it's safe to enable PV EOI and
2167 * if yes do so.
2168 */
2169static void apic_sync_pv_eoi_to_guest(struct kvm_vcpu *vcpu,
2170 struct kvm_lapic *apic)
2171{
2172 if (!pv_eoi_enabled(vcpu) ||
2173 /* IRR set or many bits in ISR: could be nested. */
2174 apic->irr_pending ||
2175 /* Cache not set: could be safe but we don't bother. */
2176 apic->highest_isr_cache == -1 ||
2177 /* Need EOI to update ioapic. */
3bb345f3 2178 kvm_ioapic_handles_vector(apic, apic->highest_isr_cache)) {
ae7a2a3f
MT		 2179 /*
2180 * PV EOI was disabled by apic_sync_pv_eoi_from_guest
2181 * so we need not do anything here.
2182 */
2183 return;
2184 }
2185
2186 pv_eoi_set_pending(apic->vcpu);
2187}
2188
b93463aa
AK		 2189void kvm_lapic_sync_to_vapic(struct kvm_vcpu *vcpu)
2190{
2191 u32 data, tpr;
2192 int max_irr, max_isr;
ae7a2a3f 2193 struct kvm_lapic *apic = vcpu->arch.apic;
b93463aa 2194
ae7a2a3f
MT		 2195 apic_sync_pv_eoi_to_guest(vcpu, apic);
2196
41383771 2197 if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
b93463aa
AK		 2198 return;
2199
dfb95954 2200 tpr = kvm_lapic_get_reg(apic, APIC_TASKPRI) & 0xff;
b93463aa
AK		 2201 max_irr = apic_find_highest_irr(apic);
2202 if (max_irr < 0)
2203 max_irr = 0;
2204 max_isr = apic_find_highest_isr(apic);
2205 if (max_isr < 0)
2206 max_isr = 0;
2207 data = (tpr & 0xff) | ((max_isr & 0xf0) << 8) | (max_irr << 24);
2208
fda4e2e8
AH		 2209 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
2210 sizeof(u32));
b93463aa
AK		 2211}
2212
fda4e2e8 2213int kvm_lapic_set_vapic_addr(struct kvm_vcpu *vcpu, gpa_t vapic_addr)
b93463aa 2214{
fda4e2e8
AH		 2215 if (vapic_addr) {
2216 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2217 &vcpu->arch.apic->vapic_cache,
2218 vapic_addr, sizeof(u32)))
2219 return -EINVAL;
41383771 2220 __set_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
fda4e2e8 2221 } else {
41383771 2222 __clear_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
fda4e2e8
AH		 2223 }
2224
2225 vcpu->arch.apic->vapic_addr = vapic_addr;
2226 return 0;
b93463aa 2227}
0105d1a5
GN		 2228
2229int kvm_x2apic_msr_write(struct kvm_vcpu *vcpu, u32 msr, u64 data)
2230{
2231 struct kvm_lapic *apic = vcpu->arch.apic;
2232 u32 reg = (msr - APIC_BASE_MSR) << 4;
2233
35754c98 2234 if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(apic))
0105d1a5
GN		 2235 return 1;
2236
c69d3d9b
NA		 2237 if (reg == APIC_ICR2)
2238 return 1;
2239
0105d1a5 2240 /* if this is ICR write vector before command */
decdc283 2241 if (reg == APIC_ICR)
1e6e2755
SS		 2242 kvm_lapic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
2243 return kvm_lapic_reg_write(apic, reg, (u32)data);
0105d1a5
GN		 2244}
2245
2246int kvm_x2apic_msr_read(struct kvm_vcpu *vcpu, u32 msr, u64 *data)
2247{
2248 struct kvm_lapic *apic = vcpu->arch.apic;
2249 u32 reg = (msr - APIC_BASE_MSR) << 4, low, high = 0;
2250
35754c98 2251 if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(apic))
0105d1a5
GN		 2252 return 1;
2253
c69d3d9b
NA		 2254 if (reg == APIC_DFR || reg == APIC_ICR2) {
2255 apic_debug("KVM_APIC_READ: read x2apic reserved register %x\n",
2256 reg);
2257 return 1;
2258 }
2259
1e6e2755 2260 if (kvm_lapic_reg_read(apic, reg, 4, &low))
0105d1a5 2261 return 1;
decdc283 2262 if (reg == APIC_ICR)
1e6e2755 2263 kvm_lapic_reg_read(apic, APIC_ICR2, 4, &high);
0105d1a5
GN		 2264
2265 *data = (((u64)high) << 32) | low;
2266
2267 return 0;
2268}
10388a07
GN		 2269
2270int kvm_hv_vapic_msr_write(struct kvm_vcpu *vcpu, u32 reg, u64 data)
2271{
2272 struct kvm_lapic *apic = vcpu->arch.apic;
2273
bce87cce 2274 if (!lapic_in_kernel(vcpu))
10388a07
GN		 2275 return 1;
2276
2277 /* if this is ICR write vector before command */
2278 if (reg == APIC_ICR)
1e6e2755
SS		 2279 kvm_lapic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
2280 return kvm_lapic_reg_write(apic, reg, (u32)data);
10388a07
GN		 2281}
2282
2283int kvm_hv_vapic_msr_read(struct kvm_vcpu *vcpu, u32 reg, u64 *data)
2284{
2285 struct kvm_lapic *apic = vcpu->arch.apic;
2286 u32 low, high = 0;
2287
bce87cce 2288 if (!lapic_in_kernel(vcpu))
10388a07
GN		 2289 return 1;
2290
1e6e2755 2291 if (kvm_lapic_reg_read(apic, reg, 4, &low))
10388a07
GN		 2292 return 1;
2293 if (reg == APIC_ICR)
1e6e2755 2294 kvm_lapic_reg_read(apic, APIC_ICR2, 4, &high);
10388a07
GN		 2295
2296 *data = (((u64)high) << 32) | low;
2297
2298 return 0;
2299}
ae7a2a3f
MT		 2300
2301int kvm_lapic_enable_pv_eoi(struct kvm_vcpu *vcpu, u64 data)
2302{
2303 u64 addr = data & ~KVM_MSR_ENABLED;
2304 if (!IS_ALIGNED(addr, 4))
2305 return 1;
2306
2307 vcpu->arch.pv_eoi.msr_val = data;
2308 if (!pv_eoi_enabled(vcpu))
2309 return 0;
2310 return kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.pv_eoi.data,
8f964525 2311 addr, sizeof(u8));
ae7a2a3f 2312}
c5cc421b 2313
66450a21
JK		 2314void kvm_apic_accept_events(struct kvm_vcpu *vcpu)
2315{
2316 struct kvm_lapic *apic = vcpu->arch.apic;
2b4a273b 2317 u8 sipi_vector;
299018f4 2318 unsigned long pe;
66450a21 2319
bce87cce 2320 if (!lapic_in_kernel(vcpu) || !apic->pending_events)
66450a21
JK		 2321 return;
2322
cd7764fe
PB		 2323 /*
2324 * INITs are latched while in SMM. Because an SMM CPU cannot
2325 * be in KVM_MP_STATE_INIT_RECEIVED state, just eat SIPIs
2326 * and delay processing of INIT until the next RSM.
2327 */
2328 if (is_smm(vcpu)) {
2329 WARN_ON_ONCE(vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED);
2330 if (test_bit(KVM_APIC_SIPI, &apic->pending_events))
2331 clear_bit(KVM_APIC_SIPI, &apic->pending_events);
2332 return;
2333 }
299018f4 2334
cd7764fe 2335 pe = xchg(&apic->pending_events, 0);
299018f4 2336 if (test_bit(KVM_APIC_INIT, &pe)) {
d28bc9dd
NA		 2337 kvm_lapic_reset(vcpu, true);
2338 kvm_vcpu_reset(vcpu, true);
66450a21
JK		 2339 if (kvm_vcpu_is_bsp(apic->vcpu))
2340 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
2341 else
2342 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
2343 }
299018f4 2344 if (test_bit(KVM_APIC_SIPI, &pe) &&
66450a21
JK		 2345 vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
2346 /* evaluate pending_events before reading the vector */
2347 smp_rmb();
2348 sipi_vector = apic->sipi_vector;
98eff52a 2349 apic_debug("vcpu %d received sipi with vector # %x\n",
66450a21
JK		 2350 vcpu->vcpu_id, sipi_vector);
2351 kvm_vcpu_deliver_sipi_vector(vcpu, sipi_vector);
2352 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
2353 }
2354}
2355
c5cc421b
GN		 2356void kvm_lapic_init(void)
2357{
2358 /* do not patch jump label more than once per second */
2359 jump_label_rate_limit(&apic_hw_disabled, HZ);
f8c1ea10 2360 jump_label_rate_limit(&apic_sw_disabled, HZ);
c5cc421b 2361}
Linux 6.x block layer and io_uring tree(s)