Loading...
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
74static inline void apic_set_reg(struct kvm_lapic *apic, int reg_off, u32 val)
75{
76 *((u32 *) (apic->regs + reg_off)) = val;
77}
78
79static inline int apic_test_vector(int vec, void *bitmap)
80{
81 return test_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
82}
83
84bool kvm_apic_pending_eoi(struct kvm_vcpu *vcpu, int vector)
85{
86 struct kvm_lapic *apic = vcpu->arch.apic;
87
88 return apic_test_vector(vector, apic->regs + APIC_ISR) ||
89 apic_test_vector(vector, apic->regs + APIC_IRR);
90}
91
92static inline void apic_set_vector(int vec, void *bitmap)
93{
94 set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
95}
96
97static inline void apic_clear_vector(int vec, void *bitmap)
98{
99 clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
100}
101
102static inline int __apic_test_and_set_vector(int vec, void *bitmap)
103{
104 return __test_and_set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
105}
106
107static inline int __apic_test_and_clear_vector(int vec, void *bitmap)
108{
109 return __test_and_clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
110}
111
112struct static_key_deferred apic_hw_disabled __read_mostly;
113struct static_key_deferred apic_sw_disabled __read_mostly;
114
115static inline void apic_set_spiv(struct kvm_lapic *apic, u32 val)
116{
117 if ((kvm_apic_get_reg(apic, APIC_SPIV) ^ val) & APIC_SPIV_APIC_ENABLED) {
118 if (val & APIC_SPIV_APIC_ENABLED)
119 static_key_slow_dec_deferred(&apic_sw_disabled);
120 else
121 static_key_slow_inc(&apic_sw_disabled.key);
122 }
123 apic_set_reg(apic, APIC_SPIV, val);
124}
125
126static inline int apic_enabled(struct kvm_lapic *apic)
127{
128 return kvm_apic_sw_enabled(apic) && kvm_apic_hw_enabled(apic);
129}
130
131#define LVT_MASK \
132 (APIC_LVT_MASKED | APIC_SEND_PENDING | APIC_VECTOR_MASK)
133
134#define LINT_MASK \
135 (LVT_MASK | APIC_MODE_MASK | APIC_INPUT_POLARITY | \
136 APIC_LVT_REMOTE_IRR | APIC_LVT_LEVEL_TRIGGER)
137
138static inline int kvm_apic_id(struct kvm_lapic *apic)
139{
140 return (kvm_apic_get_reg(apic, APIC_ID) >> 24) & 0xff;
141}
142
143#define KVM_X2APIC_CID_BITS 0
144
145static void recalculate_apic_map(struct kvm *kvm)
146{
147 struct kvm_apic_map *new, *old = NULL;
148 struct kvm_vcpu *vcpu;
149 int i;
150
151 new = kzalloc(sizeof(struct kvm_apic_map), GFP_KERNEL);
152
153 mutex_lock(&kvm->arch.apic_map_lock);
154
155 if (!new)
156 goto out;
157
158 new->ldr_bits = 8;
159 /* flat mode is default */
160 new->cid_shift = 8;
161 new->cid_mask = 0;
162 new->lid_mask = 0xff;
163
164 kvm_for_each_vcpu(i, vcpu, kvm) {
165 struct kvm_lapic *apic = vcpu->arch.apic;
166 u16 cid, lid;
167 u32 ldr;
168
169 if (!kvm_apic_present(vcpu))
170 continue;
171
172 /*
173 * All APICs have to be configured in the same mode by an OS.
174 * We take advatage of this while building logical id loockup
175 * table. After reset APICs are in xapic/flat mode, so if we
176 * find apic with different setting we assume this is the mode
177 * OS wants all apics to be in; build lookup table accordingly.
178 */
179 if (apic_x2apic_mode(apic)) {
180 new->ldr_bits = 32;
181 new->cid_shift = 16;
182 new->cid_mask = (1 << KVM_X2APIC_CID_BITS) - 1;
183 new->lid_mask = 0xffff;
184 } else if (kvm_apic_sw_enabled(apic) &&
185 !new->cid_mask /* flat mode */ &&
186 kvm_apic_get_reg(apic, APIC_DFR) == APIC_DFR_CLUSTER) {
187 new->cid_shift = 4;
188 new->cid_mask = 0xf;
189 new->lid_mask = 0xf;
190 }
191
192 new->phys_map[kvm_apic_id(apic)] = apic;
193
194 ldr = kvm_apic_get_reg(apic, APIC_LDR);
195 cid = apic_cluster_id(new, ldr);
196 lid = apic_logical_id(new, ldr);
197
198 if (lid)
199 new->logical_map[cid][ffs(lid) - 1] = apic;
200 }
201out:
202 old = rcu_dereference_protected(kvm->arch.apic_map,
203 lockdep_is_held(&kvm->arch.apic_map_lock));
204 rcu_assign_pointer(kvm->arch.apic_map, new);
205 mutex_unlock(&kvm->arch.apic_map_lock);
206
207 if (old)
208 kfree_rcu(old, rcu);
209
210 kvm_vcpu_request_scan_ioapic(kvm);
211}
212
213static inline void kvm_apic_set_id(struct kvm_lapic *apic, u8 id)
214{
215 apic_set_reg(apic, APIC_ID, id << 24);
216 recalculate_apic_map(apic->vcpu->kvm);
217}
218
219static inline void kvm_apic_set_ldr(struct kvm_lapic *apic, u32 id)
220{
221 apic_set_reg(apic, APIC_LDR, id);
222 recalculate_apic_map(apic->vcpu->kvm);
223}
224
225static inline int apic_lvt_enabled(struct kvm_lapic *apic, int lvt_type)
226{
227 return !(kvm_apic_get_reg(apic, lvt_type) & APIC_LVT_MASKED);
228}
229
230static inline int apic_lvt_vector(struct kvm_lapic *apic, int lvt_type)
231{
232 return kvm_apic_get_reg(apic, lvt_type) & APIC_VECTOR_MASK;
233}
234
235static inline int apic_lvtt_oneshot(struct kvm_lapic *apic)
236{
237 return ((kvm_apic_get_reg(apic, APIC_LVTT) &
238 apic->lapic_timer.timer_mode_mask) == APIC_LVT_TIMER_ONESHOT);
239}
240
241static inline int apic_lvtt_period(struct kvm_lapic *apic)
242{
243 return ((kvm_apic_get_reg(apic, APIC_LVTT) &
244 apic->lapic_timer.timer_mode_mask) == APIC_LVT_TIMER_PERIODIC);
245}
246
247static inline int apic_lvtt_tscdeadline(struct kvm_lapic *apic)
248{
249 return ((kvm_apic_get_reg(apic, APIC_LVTT) &
250 apic->lapic_timer.timer_mode_mask) ==
251 APIC_LVT_TIMER_TSCDEADLINE);
252}
253
254static inline int apic_lvt_nmi_mode(u32 lvt_val)
255{
256 return (lvt_val & (APIC_MODE_MASK | APIC_LVT_MASKED)) == APIC_DM_NMI;
257}
258
259void kvm_apic_set_version(struct kvm_vcpu *vcpu)
260{
261 struct kvm_lapic *apic = vcpu->arch.apic;
262 struct kvm_cpuid_entry2 *feat;
263 u32 v = APIC_VERSION;
264
265 if (!kvm_vcpu_has_lapic(vcpu))
266 return;
267
268 feat = kvm_find_cpuid_entry(apic->vcpu, 0x1, 0);
269 if (feat && (feat->ecx & (1 << (X86_FEATURE_X2APIC & 31))))
270 v |= APIC_LVR_DIRECTED_EOI;
271 apic_set_reg(apic, APIC_LVR, v);
272}
273
274static const unsigned int apic_lvt_mask[APIC_LVT_NUM] = {
275 LVT_MASK , /* part LVTT mask, timer mode mask added at runtime */
276 LVT_MASK | APIC_MODE_MASK, /* LVTTHMR */
277 LVT_MASK | APIC_MODE_MASK, /* LVTPC */
278 LINT_MASK, LINT_MASK, /* LVT0-1 */
279 LVT_MASK /* LVTERR */
280};
281
282static int find_highest_vector(void *bitmap)
283{
284 int vec;
285 u32 *reg;
286
287 for (vec = MAX_APIC_VECTOR - APIC_VECTORS_PER_REG;
288 vec >= 0; vec -= APIC_VECTORS_PER_REG) {
289 reg = bitmap + REG_POS(vec);
290 if (*reg)
291 return fls(*reg) - 1 + vec;
292 }
293
294 return -1;
295}
296
297static u8 count_vectors(void *bitmap)
298{
299 int vec;
300 u32 *reg;
301 u8 count = 0;
302
303 for (vec = 0; vec < MAX_APIC_VECTOR; vec += APIC_VECTORS_PER_REG) {
304 reg = bitmap + REG_POS(vec);
305 count += hweight32(*reg);
306 }
307
308 return count;
309}
310
311void kvm_apic_update_irr(struct kvm_vcpu *vcpu, u32 *pir)
312{
313 u32 i, pir_val;
314 struct kvm_lapic *apic = vcpu->arch.apic;
315
316 for (i = 0; i <= 7; i++) {
317 pir_val = xchg(&pir[i], 0);
318 if (pir_val)
319 *((u32 *)(apic->regs + APIC_IRR + i * 0x10)) |= pir_val;
320 }
321}
322EXPORT_SYMBOL_GPL(kvm_apic_update_irr);
323
324static inline void apic_set_irr(int vec, struct kvm_lapic *apic)
325{
326 apic->irr_pending = true;
327 apic_set_vector(vec, apic->regs + APIC_IRR);
328}
329
330static inline int apic_search_irr(struct kvm_lapic *apic)
331{
332 return find_highest_vector(apic->regs + APIC_IRR);
333}
334
335static inline int apic_find_highest_irr(struct kvm_lapic *apic)
336{
337 int result;
338
339 /*
340 * Note that irr_pending is just a hint. It will be always
341 * true with virtual interrupt delivery enabled.
342 */
343 if (!apic->irr_pending)
344 return -1;
345
346 kvm_x86_ops->sync_pir_to_irr(apic->vcpu);
347 result = apic_search_irr(apic);
348 ASSERT(result == -1 || result >= 16);
349
350 return result;
351}
352
353static inline void apic_clear_irr(int vec, struct kvm_lapic *apic)
354{
355 apic->irr_pending = false;
356 apic_clear_vector(vec, apic->regs + APIC_IRR);
357 if (apic_search_irr(apic) != -1)
358 apic->irr_pending = true;
359}
360
361static inline void apic_set_isr(int vec, struct kvm_lapic *apic)
362{
363 if (!__apic_test_and_set_vector(vec, apic->regs + APIC_ISR))
364 ++apic->isr_count;
365 BUG_ON(apic->isr_count > MAX_APIC_VECTOR);
366 /*
367 * ISR (in service register) bit is set when injecting an interrupt.
368 * The highest vector is injected. Thus the latest bit set matches
369 * the highest bit in ISR.
370 */
371 apic->highest_isr_cache = vec;
372}
373
374static inline void apic_clear_isr(int vec, struct kvm_lapic *apic)
375{
376 if (__apic_test_and_clear_vector(vec, apic->regs + APIC_ISR))
377 --apic->isr_count;
378 BUG_ON(apic->isr_count < 0);
379 apic->highest_isr_cache = -1;
380}
381
382int kvm_lapic_find_highest_irr(struct kvm_vcpu *vcpu)
383{
384 int highest_irr;
385
386 /* This may race with setting of irr in __apic_accept_irq() and
387 * value returned may be wrong, but kvm_vcpu_kick() in __apic_accept_irq
388 * will cause vmexit immediately and the value will be recalculated
389 * on the next vmentry.
390 */
391 if (!kvm_vcpu_has_lapic(vcpu))
392 return 0;
393 highest_irr = apic_find_highest_irr(vcpu->arch.apic);
394
395 return highest_irr;
396}
397
398static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
399 int vector, int level, int trig_mode,
400 unsigned long *dest_map);
401
402int kvm_apic_set_irq(struct kvm_vcpu *vcpu, struct kvm_lapic_irq *irq,
403 unsigned long *dest_map)
404{
405 struct kvm_lapic *apic = vcpu->arch.apic;
406
407 return __apic_accept_irq(apic, irq->delivery_mode, irq->vector,
408 irq->level, irq->trig_mode, dest_map);
409}
410
411static int pv_eoi_put_user(struct kvm_vcpu *vcpu, u8 val)
412{
413
414 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, &val,
415 sizeof(val));
416}
417
418static int pv_eoi_get_user(struct kvm_vcpu *vcpu, u8 *val)
419{
420
421 return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, val,
422 sizeof(*val));
423}
424
425static inline bool pv_eoi_enabled(struct kvm_vcpu *vcpu)
426{
427 return vcpu->arch.pv_eoi.msr_val & KVM_MSR_ENABLED;
428}
429
430static bool pv_eoi_get_pending(struct kvm_vcpu *vcpu)
431{
432 u8 val;
433 if (pv_eoi_get_user(vcpu, &val) < 0)
434 apic_debug("Can't read EOI MSR value: 0x%llx\n",
435 (unsigned long long)vcpu->arch.pv_eoi.msr_val);
436 return val & 0x1;
437}
438
439static void pv_eoi_set_pending(struct kvm_vcpu *vcpu)
440{
441 if (pv_eoi_put_user(vcpu, KVM_PV_EOI_ENABLED) < 0) {
442 apic_debug("Can't set EOI MSR value: 0x%llx\n",
443 (unsigned long long)vcpu->arch.pv_eoi.msr_val);
444 return;
445 }
446 __set_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
447}
448
449static void pv_eoi_clr_pending(struct kvm_vcpu *vcpu)
450{
451 if (pv_eoi_put_user(vcpu, KVM_PV_EOI_DISABLED) < 0) {
452 apic_debug("Can't clear EOI MSR value: 0x%llx\n",
453 (unsigned long long)vcpu->arch.pv_eoi.msr_val);
454 return;
455 }
456 __clear_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
457}
458
459static inline int apic_find_highest_isr(struct kvm_lapic *apic)
460{
461 int result;
462
463 /* Note that isr_count is always 1 with vid enabled */
464 if (!apic->isr_count)
465 return -1;
466 if (likely(apic->highest_isr_cache != -1))
467 return apic->highest_isr_cache;
468
469 result = find_highest_vector(apic->regs + APIC_ISR);
470 ASSERT(result == -1 || result >= 16);
471
472 return result;
473}
474
475void kvm_apic_update_tmr(struct kvm_vcpu *vcpu, u32 *tmr)
476{
477 struct kvm_lapic *apic = vcpu->arch.apic;
478 int i;
479
480 for (i = 0; i < 8; i++)
481 apic_set_reg(apic, APIC_TMR + 0x10 * i, tmr[i]);
482}
483
484static void apic_update_ppr(struct kvm_lapic *apic)
485{
486 u32 tpr, isrv, ppr, old_ppr;
487 int isr;
488
489 old_ppr = kvm_apic_get_reg(apic, APIC_PROCPRI);
490 tpr = kvm_apic_get_reg(apic, APIC_TASKPRI);
491 isr = apic_find_highest_isr(apic);
492 isrv = (isr != -1) ? isr : 0;
493
494 if ((tpr & 0xf0) >= (isrv & 0xf0))
495 ppr = tpr & 0xff;
496 else
497 ppr = isrv & 0xf0;
498
499 apic_debug("vlapic %p, ppr 0x%x, isr 0x%x, isrv 0x%x",
500 apic, ppr, isr, isrv);
501
502 if (old_ppr != ppr) {
503 apic_set_reg(apic, APIC_PROCPRI, ppr);
504 if (ppr < old_ppr)
505 kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
506 }
507}
508
509static void apic_set_tpr(struct kvm_lapic *apic, u32 tpr)
510{
511 apic_set_reg(apic, APIC_TASKPRI, tpr);
512 apic_update_ppr(apic);
513}
514
515int kvm_apic_match_physical_addr(struct kvm_lapic *apic, u16 dest)
516{
517 return dest == 0xff || kvm_apic_id(apic) == dest;
518}
519
520int kvm_apic_match_logical_addr(struct kvm_lapic *apic, u8 mda)
521{
522 int result = 0;
523 u32 logical_id;
524
525 if (apic_x2apic_mode(apic)) {
526 logical_id = kvm_apic_get_reg(apic, APIC_LDR);
527 return logical_id & mda;
528 }
529
530 logical_id = GET_APIC_LOGICAL_ID(kvm_apic_get_reg(apic, APIC_LDR));
531
532 switch (kvm_apic_get_reg(apic, APIC_DFR)) {
533 case APIC_DFR_FLAT:
534 if (logical_id & mda)
535 result = 1;
536 break;
537 case APIC_DFR_CLUSTER:
538 if (((logical_id >> 4) == (mda >> 0x4))
539 && (logical_id & mda & 0xf))
540 result = 1;
541 break;
542 default:
543 apic_debug("Bad DFR vcpu %d: %08x\n",
544 apic->vcpu->vcpu_id, kvm_apic_get_reg(apic, APIC_DFR));
545 break;
546 }
547
548 return result;
549}
550
551int kvm_apic_match_dest(struct kvm_vcpu *vcpu, struct kvm_lapic *source,
552 int short_hand, int dest, int dest_mode)
553{
554 int result = 0;
555 struct kvm_lapic *target = vcpu->arch.apic;
556
557 apic_debug("target %p, source %p, dest 0x%x, "
558 "dest_mode 0x%x, short_hand 0x%x\n",
559 target, source, dest, dest_mode, short_hand);
560
561 ASSERT(target);
562 switch (short_hand) {
563 case APIC_DEST_NOSHORT:
564 if (dest_mode == 0)
565 /* Physical mode. */
566 result = kvm_apic_match_physical_addr(target, dest);
567 else
568 /* Logical mode. */
569 result = kvm_apic_match_logical_addr(target, dest);
570 break;
571 case APIC_DEST_SELF:
572 result = (target == source);
573 break;
574 case APIC_DEST_ALLINC:
575 result = 1;
576 break;
577 case APIC_DEST_ALLBUT:
578 result = (target != source);
579 break;
580 default:
581 apic_debug("kvm: apic: Bad dest shorthand value %x\n",
582 short_hand);
583 break;
584 }
585
586 return result;
587}
588
589bool kvm_irq_delivery_to_apic_fast(struct kvm *kvm, struct kvm_lapic *src,
590 struct kvm_lapic_irq *irq, int *r, unsigned long *dest_map)
591{
592 struct kvm_apic_map *map;
593 unsigned long bitmap = 1;
594 struct kvm_lapic **dst;
595 int i;
596 bool ret = false;
597
598 *r = -1;
599
600 if (irq->shorthand == APIC_DEST_SELF) {
601 *r = kvm_apic_set_irq(src->vcpu, irq, dest_map);
602 return true;
603 }
604
605 if (irq->shorthand)
606 return false;
607
608 rcu_read_lock();
609 map = rcu_dereference(kvm->arch.apic_map);
610
611 if (!map)
612 goto out;
613
614 if (irq->dest_mode == 0) { /* physical mode */
615 if (irq->delivery_mode == APIC_DM_LOWEST ||
616 irq->dest_id == 0xff)
617 goto out;
618 dst = &map->phys_map[irq->dest_id & 0xff];
619 } else {
620 u32 mda = irq->dest_id << (32 - map->ldr_bits);
621
622 dst = map->logical_map[apic_cluster_id(map, mda)];
623
624 bitmap = apic_logical_id(map, mda);
625
626 if (irq->delivery_mode == APIC_DM_LOWEST) {
627 int l = -1;
628 for_each_set_bit(i, &bitmap, 16) {
629 if (!dst[i])
630 continue;
631 if (l < 0)
632 l = i;
633 else if (kvm_apic_compare_prio(dst[i]->vcpu, dst[l]->vcpu) < 0)
634 l = i;
635 }
636
637 bitmap = (l >= 0) ? 1 << l : 0;
638 }
639 }
640
641 for_each_set_bit(i, &bitmap, 16) {
642 if (!dst[i])
643 continue;
644 if (*r < 0)
645 *r = 0;
646 *r += kvm_apic_set_irq(dst[i]->vcpu, irq, dest_map);
647 }
648
649 ret = true;
650out:
651 rcu_read_unlock();
652 return ret;
653}
654
655/*
656 * Add a pending IRQ into lapic.
657 * Return 1 if successfully added and 0 if discarded.
658 */
659static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
660 int vector, int level, int trig_mode,
661 unsigned long *dest_map)
662{
663 int result = 0;
664 struct kvm_vcpu *vcpu = apic->vcpu;
665
666 switch (delivery_mode) {
667 case APIC_DM_LOWEST:
668 vcpu->arch.apic_arb_prio++;
669 case APIC_DM_FIXED:
670 /* FIXME add logic for vcpu on reset */
671 if (unlikely(!apic_enabled(apic)))
672 break;
673
674 result = 1;
675
676 if (dest_map)
677 __set_bit(vcpu->vcpu_id, dest_map);
678
679 if (kvm_x86_ops->deliver_posted_interrupt)
680 kvm_x86_ops->deliver_posted_interrupt(vcpu, vector);
681 else {
682 apic_set_irr(vector, apic);
683
684 kvm_make_request(KVM_REQ_EVENT, vcpu);
685 kvm_vcpu_kick(vcpu);
686 }
687 trace_kvm_apic_accept_irq(vcpu->vcpu_id, delivery_mode,
688 trig_mode, vector, false);
689 break;
690
691 case APIC_DM_REMRD:
692 result = 1;
693 vcpu->arch.pv.pv_unhalted = 1;
694 kvm_make_request(KVM_REQ_EVENT, vcpu);
695 kvm_vcpu_kick(vcpu);
696 break;
697
698 case APIC_DM_SMI:
699 apic_debug("Ignoring guest SMI\n");
700 break;
701
702 case APIC_DM_NMI:
703 result = 1;
704 kvm_inject_nmi(vcpu);
705 kvm_vcpu_kick(vcpu);
706 break;
707
708 case APIC_DM_INIT:
709 if (!trig_mode || level) {
710 result = 1;
711 /* assumes that there are only KVM_APIC_INIT/SIPI */
712 apic->pending_events = (1UL << KVM_APIC_INIT);
713 /* make sure pending_events is visible before sending
714 * the request */
715 smp_wmb();
716 kvm_make_request(KVM_REQ_EVENT, vcpu);
717 kvm_vcpu_kick(vcpu);
718 } else {
719 apic_debug("Ignoring de-assert INIT to vcpu %d\n",
720 vcpu->vcpu_id);
721 }
722 break;
723
724 case APIC_DM_STARTUP:
725 apic_debug("SIPI to vcpu %d vector 0x%02x\n",
726 vcpu->vcpu_id, vector);
727 result = 1;
728 apic->sipi_vector = vector;
729 /* make sure sipi_vector is visible for the receiver */
730 smp_wmb();
731 set_bit(KVM_APIC_SIPI, &apic->pending_events);
732 kvm_make_request(KVM_REQ_EVENT, vcpu);
733 kvm_vcpu_kick(vcpu);
734 break;
735
736 case APIC_DM_EXTINT:
737 /*
738 * Should only be called by kvm_apic_local_deliver() with LVT0,
739 * before NMI watchdog was enabled. Already handled by
740 * kvm_apic_accept_pic_intr().
741 */
742 break;
743
744 default:
745 printk(KERN_ERR "TODO: unsupported delivery mode %x\n",
746 delivery_mode);
747 break;
748 }
749 return result;
750}
751
752int kvm_apic_compare_prio(struct kvm_vcpu *vcpu1, struct kvm_vcpu *vcpu2)
753{
754 return vcpu1->arch.apic_arb_prio - vcpu2->arch.apic_arb_prio;
755}
756
757static void kvm_ioapic_send_eoi(struct kvm_lapic *apic, int vector)
758{
759 if (!(kvm_apic_get_reg(apic, APIC_SPIV) & APIC_SPIV_DIRECTED_EOI) &&
760 kvm_ioapic_handles_vector(apic->vcpu->kvm, vector)) {
761 int trigger_mode;
762 if (apic_test_vector(vector, apic->regs + APIC_TMR))
763 trigger_mode = IOAPIC_LEVEL_TRIG;
764 else
765 trigger_mode = IOAPIC_EDGE_TRIG;
766 kvm_ioapic_update_eoi(apic->vcpu, vector, trigger_mode);
767 }
768}
769
770static int apic_set_eoi(struct kvm_lapic *apic)
771{
772 int vector = apic_find_highest_isr(apic);
773
774 trace_kvm_eoi(apic, vector);
775
776 /*
777 * Not every write EOI will has corresponding ISR,
778 * one example is when Kernel check timer on setup_IO_APIC
779 */
780 if (vector == -1)
781 return vector;
782
783 apic_clear_isr(vector, apic);
784 apic_update_ppr(apic);
785
786 kvm_ioapic_send_eoi(apic, vector);
787 kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
788 return vector;
789}
790
791/*
792 * this interface assumes a trap-like exit, which has already finished
793 * desired side effect including vISR and vPPR update.
794 */
795void kvm_apic_set_eoi_accelerated(struct kvm_vcpu *vcpu, int vector)
796{
797 struct kvm_lapic *apic = vcpu->arch.apic;
798
799 trace_kvm_eoi(apic, vector);
800
801 kvm_ioapic_send_eoi(apic, vector);
802 kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
803}
804EXPORT_SYMBOL_GPL(kvm_apic_set_eoi_accelerated);
805
806static void apic_send_ipi(struct kvm_lapic *apic)
807{
808 u32 icr_low = kvm_apic_get_reg(apic, APIC_ICR);
809 u32 icr_high = kvm_apic_get_reg(apic, APIC_ICR2);
810 struct kvm_lapic_irq irq;
811
812 irq.vector = icr_low & APIC_VECTOR_MASK;
813 irq.delivery_mode = icr_low & APIC_MODE_MASK;
814 irq.dest_mode = icr_low & APIC_DEST_MASK;
815 irq.level = icr_low & APIC_INT_ASSERT;
816 irq.trig_mode = icr_low & APIC_INT_LEVELTRIG;
817 irq.shorthand = icr_low & APIC_SHORT_MASK;
818 if (apic_x2apic_mode(apic))
819 irq.dest_id = icr_high;
820 else
821 irq.dest_id = GET_APIC_DEST_FIELD(icr_high);
822
823 trace_kvm_apic_ipi(icr_low, irq.dest_id);
824
825 apic_debug("icr_high 0x%x, icr_low 0x%x, "
826 "short_hand 0x%x, dest 0x%x, trig_mode 0x%x, level 0x%x, "
827 "dest_mode 0x%x, delivery_mode 0x%x, vector 0x%x\n",
828 icr_high, icr_low, irq.shorthand, irq.dest_id,
829 irq.trig_mode, irq.level, irq.dest_mode, irq.delivery_mode,
830 irq.vector);
831
832 kvm_irq_delivery_to_apic(apic->vcpu->kvm, apic, &irq, NULL);
833}
834
835static u32 apic_get_tmcct(struct kvm_lapic *apic)
836{
837 ktime_t remaining;
838 s64 ns;
839 u32 tmcct;
840
841 ASSERT(apic != NULL);
842
843 /* if initial count is 0, current count should also be 0 */
844 if (kvm_apic_get_reg(apic, APIC_TMICT) == 0 ||
845 apic->lapic_timer.period == 0)
846 return 0;
847
848 remaining = hrtimer_get_remaining(&apic->lapic_timer.timer);
849 if (ktime_to_ns(remaining) < 0)
850 remaining = ktime_set(0, 0);
851
852 ns = mod_64(ktime_to_ns(remaining), apic->lapic_timer.period);
853 tmcct = div64_u64(ns,
854 (APIC_BUS_CYCLE_NS * apic->divide_count));
855
856 return tmcct;
857}
858
859static void __report_tpr_access(struct kvm_lapic *apic, bool write)
860{
861 struct kvm_vcpu *vcpu = apic->vcpu;
862 struct kvm_run *run = vcpu->run;
863
864 kvm_make_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu);
865 run->tpr_access.rip = kvm_rip_read(vcpu);
866 run->tpr_access.is_write = write;
867}
868
869static inline void report_tpr_access(struct kvm_lapic *apic, bool write)
870{
871 if (apic->vcpu->arch.tpr_access_reporting)
872 __report_tpr_access(apic, write);
873}
874
875static u32 __apic_read(struct kvm_lapic *apic, unsigned int offset)
876{
877 u32 val = 0;
878
879 if (offset >= LAPIC_MMIO_LENGTH)
880 return 0;
881
882 switch (offset) {
883 case APIC_ID:
884 if (apic_x2apic_mode(apic))
885 val = kvm_apic_id(apic);
886 else
887 val = kvm_apic_id(apic) << 24;
888 break;
889 case APIC_ARBPRI:
890 apic_debug("Access APIC ARBPRI register which is for P6\n");
891 break;
892
893 case APIC_TMCCT: /* Timer CCR */
894 if (apic_lvtt_tscdeadline(apic))
895 return 0;
896
897 val = apic_get_tmcct(apic);
898 break;
899 case APIC_PROCPRI:
900 apic_update_ppr(apic);
901 val = kvm_apic_get_reg(apic, offset);
902 break;
903 case APIC_TASKPRI:
904 report_tpr_access(apic, false);
905 /* fall thru */
906 default:
907 val = kvm_apic_get_reg(apic, offset);
908 break;
909 }
910
911 return val;
912}
913
914static inline struct kvm_lapic *to_lapic(struct kvm_io_device *dev)
915{
916 return container_of(dev, struct kvm_lapic, dev);
917}
918
919static int apic_reg_read(struct kvm_lapic *apic, u32 offset, int len,
920 void *data)
921{
922 unsigned char alignment = offset & 0xf;
923 u32 result;
924 /* this bitmask has a bit cleared for each reserved register */
925 static const u64 rmask = 0x43ff01ffffffe70cULL;
926
927 if ((alignment + len) > 4) {
928 apic_debug("KVM_APIC_READ: alignment error %x %d\n",
929 offset, len);
930 return 1;
931 }
932
933 if (offset > 0x3f0 || !(rmask & (1ULL << (offset >> 4)))) {
934 apic_debug("KVM_APIC_READ: read reserved register %x\n",
935 offset);
936 return 1;
937 }
938
939 result = __apic_read(apic, offset & ~0xf);
940
941 trace_kvm_apic_read(offset, result);
942
943 switch (len) {
944 case 1:
945 case 2:
946 case 4:
947 memcpy(data, (char *)&result + alignment, len);
948 break;
949 default:
950 printk(KERN_ERR "Local APIC read with len = %x, "
951 "should be 1,2, or 4 instead\n", len);
952 break;
953 }
954 return 0;
955}
956
957static int apic_mmio_in_range(struct kvm_lapic *apic, gpa_t addr)
958{
959 return kvm_apic_hw_enabled(apic) &&
960 addr >= apic->base_address &&
961 addr < apic->base_address + LAPIC_MMIO_LENGTH;
962}
963
964static int apic_mmio_read(struct kvm_io_device *this,
965 gpa_t address, int len, void *data)
966{
967 struct kvm_lapic *apic = to_lapic(this);
968 u32 offset = address - apic->base_address;
969
970 if (!apic_mmio_in_range(apic, address))
971 return -EOPNOTSUPP;
972
973 apic_reg_read(apic, offset, len, data);
974
975 return 0;
976}
977
978static void update_divide_count(struct kvm_lapic *apic)
979{
980 u32 tmp1, tmp2, tdcr;
981
982 tdcr = kvm_apic_get_reg(apic, APIC_TDCR);
983 tmp1 = tdcr & 0xf;
984 tmp2 = ((tmp1 & 0x3) | ((tmp1 & 0x8) >> 1)) + 1;
985 apic->divide_count = 0x1 << (tmp2 & 0x7);
986
987 apic_debug("timer divide count is 0x%x\n",
988 apic->divide_count);
989}
990
991static void start_apic_timer(struct kvm_lapic *apic)
992{
993 ktime_t now;
994 atomic_set(&apic->lapic_timer.pending, 0);
995
996 if (apic_lvtt_period(apic) || apic_lvtt_oneshot(apic)) {
997 /* lapic timer in oneshot or periodic mode */
998 now = apic->lapic_timer.timer.base->get_time();
999 apic->lapic_timer.period = (u64)kvm_apic_get_reg(apic, APIC_TMICT)
1000 * APIC_BUS_CYCLE_NS * apic->divide_count;
1001
1002 if (!apic->lapic_timer.period)
1003 return;
1004 /*
1005 * Do not allow the guest to program periodic timers with small
1006 * interval, since the hrtimers are not throttled by the host
1007 * scheduler.
1008 */
1009 if (apic_lvtt_period(apic)) {
1010 s64 min_period = min_timer_period_us * 1000LL;
1011
1012 if (apic->lapic_timer.period < min_period) {
1013 pr_info_ratelimited(
1014 "kvm: vcpu %i: requested %lld ns "
1015 "lapic timer period limited to %lld ns\n",
1016 apic->vcpu->vcpu_id,
1017 apic->lapic_timer.period, min_period);
1018 apic->lapic_timer.period = min_period;
1019 }
1020 }
1021
1022 hrtimer_start(&apic->lapic_timer.timer,
1023 ktime_add_ns(now, apic->lapic_timer.period),
1024 HRTIMER_MODE_ABS);
1025
1026 apic_debug("%s: bus cycle is %" PRId64 "ns, now 0x%016"
1027 PRIx64 ", "
1028 "timer initial count 0x%x, period %lldns, "
1029 "expire @ 0x%016" PRIx64 ".\n", __func__,
1030 APIC_BUS_CYCLE_NS, ktime_to_ns(now),
1031 kvm_apic_get_reg(apic, APIC_TMICT),
1032 apic->lapic_timer.period,
1033 ktime_to_ns(ktime_add_ns(now,
1034 apic->lapic_timer.period)));
1035 } else if (apic_lvtt_tscdeadline(apic)) {
1036 /* lapic timer in tsc deadline mode */
1037 u64 guest_tsc, tscdeadline = apic->lapic_timer.tscdeadline;
1038 u64 ns = 0;
1039 struct kvm_vcpu *vcpu = apic->vcpu;
1040 unsigned long this_tsc_khz = vcpu->arch.virtual_tsc_khz;
1041 unsigned long flags;
1042
1043 if (unlikely(!tscdeadline || !this_tsc_khz))
1044 return;
1045
1046 local_irq_save(flags);
1047
1048 now = apic->lapic_timer.timer.base->get_time();
1049 guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu, native_read_tsc());
1050 if (likely(tscdeadline > guest_tsc)) {
1051 ns = (tscdeadline - guest_tsc) * 1000000ULL;
1052 do_div(ns, this_tsc_khz);
1053 }
1054 hrtimer_start(&apic->lapic_timer.timer,
1055 ktime_add_ns(now, ns), HRTIMER_MODE_ABS);
1056
1057 local_irq_restore(flags);
1058 }
1059}
1060
1061static void apic_manage_nmi_watchdog(struct kvm_lapic *apic, u32 lvt0_val)
1062{
1063 int nmi_wd_enabled = apic_lvt_nmi_mode(kvm_apic_get_reg(apic, APIC_LVT0));
1064
1065 if (apic_lvt_nmi_mode(lvt0_val)) {
1066 if (!nmi_wd_enabled) {
1067 apic_debug("Receive NMI setting on APIC_LVT0 "
1068 "for cpu %d\n", apic->vcpu->vcpu_id);
1069 apic->vcpu->kvm->arch.vapics_in_nmi_mode++;
1070 }
1071 } else if (nmi_wd_enabled)
1072 apic->vcpu->kvm->arch.vapics_in_nmi_mode--;
1073}
1074
1075static int apic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val)
1076{
1077 int ret = 0;
1078
1079 trace_kvm_apic_write(reg, val);
1080
1081 switch (reg) {
1082 case APIC_ID: /* Local APIC ID */
1083 if (!apic_x2apic_mode(apic))
1084 kvm_apic_set_id(apic, val >> 24);
1085 else
1086 ret = 1;
1087 break;
1088
1089 case APIC_TASKPRI:
1090 report_tpr_access(apic, true);
1091 apic_set_tpr(apic, val & 0xff);
1092 break;
1093
1094 case APIC_EOI:
1095 apic_set_eoi(apic);
1096 break;
1097
1098 case APIC_LDR:
1099 if (!apic_x2apic_mode(apic))
1100 kvm_apic_set_ldr(apic, val & APIC_LDR_MASK);
1101 else
1102 ret = 1;
1103 break;
1104
1105 case APIC_DFR:
1106 if (!apic_x2apic_mode(apic)) {
1107 apic_set_reg(apic, APIC_DFR, val | 0x0FFFFFFF);
1108 recalculate_apic_map(apic->vcpu->kvm);
1109 } else
1110 ret = 1;
1111 break;
1112
1113 case APIC_SPIV: {
1114 u32 mask = 0x3ff;
1115 if (kvm_apic_get_reg(apic, APIC_LVR) & APIC_LVR_DIRECTED_EOI)
1116 mask |= APIC_SPIV_DIRECTED_EOI;
1117 apic_set_spiv(apic, val & mask);
1118 if (!(val & APIC_SPIV_APIC_ENABLED)) {
1119 int i;
1120 u32 lvt_val;
1121
1122 for (i = 0; i < APIC_LVT_NUM; i++) {
1123 lvt_val = kvm_apic_get_reg(apic,
1124 APIC_LVTT + 0x10 * i);
1125 apic_set_reg(apic, APIC_LVTT + 0x10 * i,
1126 lvt_val | APIC_LVT_MASKED);
1127 }
1128 atomic_set(&apic->lapic_timer.pending, 0);
1129
1130 }
1131 break;
1132 }
1133 case APIC_ICR:
1134 /* No delay here, so we always clear the pending bit */
1135 apic_set_reg(apic, APIC_ICR, val & ~(1 << 12));
1136 apic_send_ipi(apic);
1137 break;
1138
1139 case APIC_ICR2:
1140 if (!apic_x2apic_mode(apic))
1141 val &= 0xff000000;
1142 apic_set_reg(apic, APIC_ICR2, val);
1143 break;
1144
1145 case APIC_LVT0:
1146 apic_manage_nmi_watchdog(apic, val);
1147 case APIC_LVTTHMR:
1148 case APIC_LVTPC:
1149 case APIC_LVT1:
1150 case APIC_LVTERR:
1151 /* TODO: Check vector */
1152 if (!kvm_apic_sw_enabled(apic))
1153 val |= APIC_LVT_MASKED;
1154
1155 val &= apic_lvt_mask[(reg - APIC_LVTT) >> 4];
1156 apic_set_reg(apic, reg, val);
1157
1158 break;
1159
1160 case APIC_LVTT:
1161 if ((kvm_apic_get_reg(apic, APIC_LVTT) &
1162 apic->lapic_timer.timer_mode_mask) !=
1163 (val & apic->lapic_timer.timer_mode_mask))
1164 hrtimer_cancel(&apic->lapic_timer.timer);
1165
1166 if (!kvm_apic_sw_enabled(apic))
1167 val |= APIC_LVT_MASKED;
1168 val &= (apic_lvt_mask[0] | apic->lapic_timer.timer_mode_mask);
1169 apic_set_reg(apic, APIC_LVTT, val);
1170 break;
1171
1172 case APIC_TMICT:
1173 if (apic_lvtt_tscdeadline(apic))
1174 break;
1175
1176 hrtimer_cancel(&apic->lapic_timer.timer);
1177 apic_set_reg(apic, APIC_TMICT, val);
1178 start_apic_timer(apic);
1179 break;
1180
1181 case APIC_TDCR:
1182 if (val & 4)
1183 apic_debug("KVM_WRITE:TDCR %x\n", val);
1184 apic_set_reg(apic, APIC_TDCR, val);
1185 update_divide_count(apic);
1186 break;
1187
1188 case APIC_ESR:
1189 if (apic_x2apic_mode(apic) && val != 0) {
1190 apic_debug("KVM_WRITE:ESR not zero %x\n", val);
1191 ret = 1;
1192 }
1193 break;
1194
1195 case APIC_SELF_IPI:
1196 if (apic_x2apic_mode(apic)) {
1197 apic_reg_write(apic, APIC_ICR, 0x40000 | (val & 0xff));
1198 } else
1199 ret = 1;
1200 break;
1201 default:
1202 ret = 1;
1203 break;
1204 }
1205 if (ret)
1206 apic_debug("Local APIC Write to read-only register %x\n", reg);
1207 return ret;
1208}
1209
1210static int apic_mmio_write(struct kvm_io_device *this,
1211 gpa_t address, int len, const void *data)
1212{
1213 struct kvm_lapic *apic = to_lapic(this);
1214 unsigned int offset = address - apic->base_address;
1215 u32 val;
1216
1217 if (!apic_mmio_in_range(apic, address))
1218 return -EOPNOTSUPP;
1219
1220 /*
1221 * APIC register must be aligned on 128-bits boundary.
1222 * 32/64/128 bits registers must be accessed thru 32 bits.
1223 * Refer SDM 8.4.1
1224 */
1225 if (len != 4 || (offset & 0xf)) {
1226 /* Don't shout loud, $infamous_os would cause only noise. */
1227 apic_debug("apic write: bad size=%d %lx\n", len, (long)address);
1228 return 0;
1229 }
1230
1231 val = *(u32*)data;
1232
1233 /* too common printing */
1234 if (offset != APIC_EOI)
1235 apic_debug("%s: offset 0x%x with length 0x%x, and value is "
1236 "0x%x\n", __func__, offset, len, val);
1237
1238 apic_reg_write(apic, offset & 0xff0, val);
1239
1240 return 0;
1241}
1242
1243void kvm_lapic_set_eoi(struct kvm_vcpu *vcpu)
1244{
1245 if (kvm_vcpu_has_lapic(vcpu))
1246 apic_reg_write(vcpu->arch.apic, APIC_EOI, 0);
1247}
1248EXPORT_SYMBOL_GPL(kvm_lapic_set_eoi);
1249
1250/* emulate APIC access in a trap manner */
1251void kvm_apic_write_nodecode(struct kvm_vcpu *vcpu, u32 offset)
1252{
1253 u32 val = 0;
1254
1255 /* hw has done the conditional check and inst decode */
1256 offset &= 0xff0;
1257
1258 apic_reg_read(vcpu->arch.apic, offset, 4, &val);
1259
1260 /* TODO: optimize to just emulate side effect w/o one more write */
1261 apic_reg_write(vcpu->arch.apic, offset, val);
1262}
1263EXPORT_SYMBOL_GPL(kvm_apic_write_nodecode);
1264
1265void kvm_free_lapic(struct kvm_vcpu *vcpu)
1266{
1267 struct kvm_lapic *apic = vcpu->arch.apic;
1268
1269 if (!vcpu->arch.apic)
1270 return;
1271
1272 hrtimer_cancel(&apic->lapic_timer.timer);
1273
1274 if (!(vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE))
1275 static_key_slow_dec_deferred(&apic_hw_disabled);
1276
1277 if (!(kvm_apic_get_reg(apic, APIC_SPIV) & APIC_SPIV_APIC_ENABLED))
1278 static_key_slow_dec_deferred(&apic_sw_disabled);
1279
1280 if (apic->regs)
1281 free_page((unsigned long)apic->regs);
1282
1283 kfree(apic);
1284}
1285
1286/*
1287 *----------------------------------------------------------------------
1288 * LAPIC interface
1289 *----------------------------------------------------------------------
1290 */
1291
1292u64 kvm_get_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu)
1293{
1294 struct kvm_lapic *apic = vcpu->arch.apic;
1295
1296 if (!kvm_vcpu_has_lapic(vcpu) || apic_lvtt_oneshot(apic) ||
1297 apic_lvtt_period(apic))
1298 return 0;
1299
1300 return apic->lapic_timer.tscdeadline;
1301}
1302
1303void kvm_set_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu, u64 data)
1304{
1305 struct kvm_lapic *apic = vcpu->arch.apic;
1306
1307 if (!kvm_vcpu_has_lapic(vcpu) || apic_lvtt_oneshot(apic) ||
1308 apic_lvtt_period(apic))
1309 return;
1310
1311 hrtimer_cancel(&apic->lapic_timer.timer);
1312 apic->lapic_timer.tscdeadline = data;
1313 start_apic_timer(apic);
1314}
1315
1316void kvm_lapic_set_tpr(struct kvm_vcpu *vcpu, unsigned long cr8)
1317{
1318 struct kvm_lapic *apic = vcpu->arch.apic;
1319
1320 if (!kvm_vcpu_has_lapic(vcpu))
1321 return;
1322
1323 apic_set_tpr(apic, ((cr8 & 0x0f) << 4)
1324 | (kvm_apic_get_reg(apic, APIC_TASKPRI) & 4));
1325}
1326
1327u64 kvm_lapic_get_cr8(struct kvm_vcpu *vcpu)
1328{
1329 u64 tpr;
1330
1331 if (!kvm_vcpu_has_lapic(vcpu))
1332 return 0;
1333
1334 tpr = (u64) kvm_apic_get_reg(vcpu->arch.apic, APIC_TASKPRI);
1335
1336 return (tpr & 0xf0) >> 4;
1337}
1338
1339void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
1340{
1341 u64 old_value = vcpu->arch.apic_base;
1342 struct kvm_lapic *apic = vcpu->arch.apic;
1343
1344 if (!apic) {
1345 value |= MSR_IA32_APICBASE_BSP;
1346 vcpu->arch.apic_base = value;
1347 return;
1348 }
1349
1350 if (!kvm_vcpu_is_bsp(apic->vcpu))
1351 value &= ~MSR_IA32_APICBASE_BSP;
1352 vcpu->arch.apic_base = value;
1353
1354 /* update jump label if enable bit changes */
1355 if ((old_value ^ value) & MSR_IA32_APICBASE_ENABLE) {
1356 if (value & MSR_IA32_APICBASE_ENABLE)
1357 static_key_slow_dec_deferred(&apic_hw_disabled);
1358 else
1359 static_key_slow_inc(&apic_hw_disabled.key);
1360 recalculate_apic_map(vcpu->kvm);
1361 }
1362
1363 if ((old_value ^ value) & X2APIC_ENABLE) {
1364 if (value & X2APIC_ENABLE) {
1365 u32 id = kvm_apic_id(apic);
1366 u32 ldr = ((id >> 4) << 16) | (1 << (id & 0xf));
1367 kvm_apic_set_ldr(apic, ldr);
1368 kvm_x86_ops->set_virtual_x2apic_mode(vcpu, true);
1369 } else
1370 kvm_x86_ops->set_virtual_x2apic_mode(vcpu, false);
1371 }
1372
1373 apic->base_address = apic->vcpu->arch.apic_base &
1374 MSR_IA32_APICBASE_BASE;
1375
1376 /* with FSB delivery interrupt, we can restart APIC functionality */
1377 apic_debug("apic base msr is 0x%016" PRIx64 ", and base address is "
1378 "0x%lx.\n", apic->vcpu->arch.apic_base, apic->base_address);
1379
1380}
1381
1382void kvm_lapic_reset(struct kvm_vcpu *vcpu)
1383{
1384 struct kvm_lapic *apic;
1385 int i;
1386
1387 apic_debug("%s\n", __func__);
1388
1389 ASSERT(vcpu);
1390 apic = vcpu->arch.apic;
1391 ASSERT(apic != NULL);
1392
1393 /* Stop the timer in case it's a reset to an active apic */
1394 hrtimer_cancel(&apic->lapic_timer.timer);
1395
1396 kvm_apic_set_id(apic, vcpu->vcpu_id);
1397 kvm_apic_set_version(apic->vcpu);
1398
1399 for (i = 0; i < APIC_LVT_NUM; i++)
1400 apic_set_reg(apic, APIC_LVTT + 0x10 * i, APIC_LVT_MASKED);
1401 apic_set_reg(apic, APIC_LVT0,
1402 SET_APIC_DELIVERY_MODE(0, APIC_MODE_EXTINT));
1403
1404 apic_set_reg(apic, APIC_DFR, 0xffffffffU);
1405 apic_set_spiv(apic, 0xff);
1406 apic_set_reg(apic, APIC_TASKPRI, 0);
1407 kvm_apic_set_ldr(apic, 0);
1408 apic_set_reg(apic, APIC_ESR, 0);
1409 apic_set_reg(apic, APIC_ICR, 0);
1410 apic_set_reg(apic, APIC_ICR2, 0);
1411 apic_set_reg(apic, APIC_TDCR, 0);
1412 apic_set_reg(apic, APIC_TMICT, 0);
1413 for (i = 0; i < 8; i++) {
1414 apic_set_reg(apic, APIC_IRR + 0x10 * i, 0);
1415 apic_set_reg(apic, APIC_ISR + 0x10 * i, 0);
1416 apic_set_reg(apic, APIC_TMR + 0x10 * i, 0);
1417 }
1418 apic->irr_pending = kvm_apic_vid_enabled(vcpu->kvm);
1419 apic->isr_count = kvm_apic_vid_enabled(vcpu->kvm);
1420 apic->highest_isr_cache = -1;
1421 update_divide_count(apic);
1422 atomic_set(&apic->lapic_timer.pending, 0);
1423 if (kvm_vcpu_is_bsp(vcpu))
1424 kvm_lapic_set_base(vcpu,
1425 vcpu->arch.apic_base | MSR_IA32_APICBASE_BSP);
1426 vcpu->arch.pv_eoi.msr_val = 0;
1427 apic_update_ppr(apic);
1428
1429 vcpu->arch.apic_arb_prio = 0;
1430 vcpu->arch.apic_attention = 0;
1431
1432 apic_debug(KERN_INFO "%s: vcpu=%p, id=%d, base_msr="
1433 "0x%016" PRIx64 ", base_address=0x%0lx.\n", __func__,
1434 vcpu, kvm_apic_id(apic),
1435 vcpu->arch.apic_base, apic->base_address);
1436}
1437
1438/*
1439 *----------------------------------------------------------------------
1440 * timer interface
1441 *----------------------------------------------------------------------
1442 */
1443
1444static bool lapic_is_periodic(struct kvm_lapic *apic)
1445{
1446 return apic_lvtt_period(apic);
1447}
1448
1449int apic_has_pending_timer(struct kvm_vcpu *vcpu)
1450{
1451 struct kvm_lapic *apic = vcpu->arch.apic;
1452
1453 if (kvm_vcpu_has_lapic(vcpu) && apic_enabled(apic) &&
1454 apic_lvt_enabled(apic, APIC_LVTT))
1455 return atomic_read(&apic->lapic_timer.pending);
1456
1457 return 0;
1458}
1459
1460int kvm_apic_local_deliver(struct kvm_lapic *apic, int lvt_type)
1461{
1462 u32 reg = kvm_apic_get_reg(apic, lvt_type);
1463 int vector, mode, trig_mode;
1464
1465 if (kvm_apic_hw_enabled(apic) && !(reg & APIC_LVT_MASKED)) {
1466 vector = reg & APIC_VECTOR_MASK;
1467 mode = reg & APIC_MODE_MASK;
1468 trig_mode = reg & APIC_LVT_LEVEL_TRIGGER;
1469 return __apic_accept_irq(apic, mode, vector, 1, trig_mode,
1470 NULL);
1471 }
1472 return 0;
1473}
1474
1475void kvm_apic_nmi_wd_deliver(struct kvm_vcpu *vcpu)
1476{
1477 struct kvm_lapic *apic = vcpu->arch.apic;
1478
1479 if (apic)
1480 kvm_apic_local_deliver(apic, APIC_LVT0);
1481}
1482
1483static const struct kvm_io_device_ops apic_mmio_ops = {
1484 .read = apic_mmio_read,
1485 .write = apic_mmio_write,
1486};
1487
1488static enum hrtimer_restart apic_timer_fn(struct hrtimer *data)
1489{
1490 struct kvm_timer *ktimer = container_of(data, struct kvm_timer, timer);
1491 struct kvm_lapic *apic = container_of(ktimer, struct kvm_lapic, lapic_timer);
1492 struct kvm_vcpu *vcpu = apic->vcpu;
1493 wait_queue_head_t *q = &vcpu->wq;
1494
1495 /*
1496 * There is a race window between reading and incrementing, but we do
1497 * not care about potentially losing timer events in the !reinject
1498 * case anyway. Note: KVM_REQ_PENDING_TIMER is implicitly checked
1499 * in vcpu_enter_guest.
1500 */
1501 if (!atomic_read(&ktimer->pending)) {
1502 atomic_inc(&ktimer->pending);
1503 /* FIXME: this code should not know anything about vcpus */
1504 kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1505 }
1506
1507 if (waitqueue_active(q))
1508 wake_up_interruptible(q);
1509
1510 if (lapic_is_periodic(apic)) {
1511 hrtimer_add_expires_ns(&ktimer->timer, ktimer->period);
1512 return HRTIMER_RESTART;
1513 } else
1514 return HRTIMER_NORESTART;
1515}
1516
1517int kvm_create_lapic(struct kvm_vcpu *vcpu)
1518{
1519 struct kvm_lapic *apic;
1520
1521 ASSERT(vcpu != NULL);
1522 apic_debug("apic_init %d\n", vcpu->vcpu_id);
1523
1524 apic = kzalloc(sizeof(*apic), GFP_KERNEL);
1525 if (!apic)
1526 goto nomem;
1527
1528 vcpu->arch.apic = apic;
1529
1530 apic->regs = (void *)get_zeroed_page(GFP_KERNEL);
1531 if (!apic->regs) {
1532 printk(KERN_ERR "malloc apic regs error for vcpu %x\n",
1533 vcpu->vcpu_id);
1534 goto nomem_free_apic;
1535 }
1536 apic->vcpu = vcpu;
1537
1538 hrtimer_init(&apic->lapic_timer.timer, CLOCK_MONOTONIC,
1539 HRTIMER_MODE_ABS);
1540 apic->lapic_timer.timer.function = apic_timer_fn;
1541
1542 /*
1543 * APIC is created enabled. This will prevent kvm_lapic_set_base from
1544 * thinking that APIC satet has changed.
1545 */
1546 vcpu->arch.apic_base = MSR_IA32_APICBASE_ENABLE;
1547 kvm_lapic_set_base(vcpu,
1548 APIC_DEFAULT_PHYS_BASE | MSR_IA32_APICBASE_ENABLE);
1549
1550 static_key_slow_inc(&apic_sw_disabled.key); /* sw disabled at reset */
1551 kvm_lapic_reset(vcpu);
1552 kvm_iodevice_init(&apic->dev, &apic_mmio_ops);
1553
1554 return 0;
1555nomem_free_apic:
1556 kfree(apic);
1557nomem:
1558 return -ENOMEM;
1559}
1560
1561int kvm_apic_has_interrupt(struct kvm_vcpu *vcpu)
1562{
1563 struct kvm_lapic *apic = vcpu->arch.apic;
1564 int highest_irr;
1565
1566 if (!kvm_vcpu_has_lapic(vcpu) || !apic_enabled(apic))
1567 return -1;
1568
1569 apic_update_ppr(apic);
1570 highest_irr = apic_find_highest_irr(apic);
1571 if ((highest_irr == -1) ||
1572 ((highest_irr & 0xF0) <= kvm_apic_get_reg(apic, APIC_PROCPRI)))
1573 return -1;
1574 return highest_irr;
1575}
1576
1577int kvm_apic_accept_pic_intr(struct kvm_vcpu *vcpu)
1578{
1579 u32 lvt0 = kvm_apic_get_reg(vcpu->arch.apic, APIC_LVT0);
1580 int r = 0;
1581
1582 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
1583 r = 1;
1584 if ((lvt0 & APIC_LVT_MASKED) == 0 &&
1585 GET_APIC_DELIVERY_MODE(lvt0) == APIC_MODE_EXTINT)
1586 r = 1;
1587 return r;
1588}
1589
1590void kvm_inject_apic_timer_irqs(struct kvm_vcpu *vcpu)
1591{
1592 struct kvm_lapic *apic = vcpu->arch.apic;
1593
1594 if (!kvm_vcpu_has_lapic(vcpu))
1595 return;
1596
1597 if (atomic_read(&apic->lapic_timer.pending) > 0) {
1598 kvm_apic_local_deliver(apic, APIC_LVTT);
1599 atomic_set(&apic->lapic_timer.pending, 0);
1600 }
1601}
1602
1603int kvm_get_apic_interrupt(struct kvm_vcpu *vcpu)
1604{
1605 int vector = kvm_apic_has_interrupt(vcpu);
1606 struct kvm_lapic *apic = vcpu->arch.apic;
1607
1608 if (vector == -1)
1609 return -1;
1610
1611 apic_set_isr(vector, apic);
1612 apic_update_ppr(apic);
1613 apic_clear_irr(vector, apic);
1614 return vector;
1615}
1616
1617void kvm_apic_post_state_restore(struct kvm_vcpu *vcpu,
1618 struct kvm_lapic_state *s)
1619{
1620 struct kvm_lapic *apic = vcpu->arch.apic;
1621
1622 kvm_lapic_set_base(vcpu, vcpu->arch.apic_base);
1623 /* set SPIV separately to get count of SW disabled APICs right */
1624 apic_set_spiv(apic, *((u32 *)(s->regs + APIC_SPIV)));
1625 memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
1626 /* call kvm_apic_set_id() to put apic into apic_map */
1627 kvm_apic_set_id(apic, kvm_apic_id(apic));
1628 kvm_apic_set_version(vcpu);
1629
1630 apic_update_ppr(apic);
1631 hrtimer_cancel(&apic->lapic_timer.timer);
1632 update_divide_count(apic);
1633 start_apic_timer(apic);
1634 apic->irr_pending = true;
1635 apic->isr_count = kvm_apic_vid_enabled(vcpu->kvm) ?
1636 1 : count_vectors(apic->regs + APIC_ISR);
1637 apic->highest_isr_cache = -1;
1638 kvm_x86_ops->hwapic_isr_update(vcpu->kvm, apic_find_highest_isr(apic));
1639 kvm_make_request(KVM_REQ_EVENT, vcpu);
1640 kvm_rtc_eoi_tracking_restore_one(vcpu);
1641}
1642
1643void __kvm_migrate_apic_timer(struct kvm_vcpu *vcpu)
1644{
1645 struct hrtimer *timer;
1646
1647 if (!kvm_vcpu_has_lapic(vcpu))
1648 return;
1649
1650 timer = &vcpu->arch.apic->lapic_timer.timer;
1651 if (hrtimer_cancel(timer))
1652 hrtimer_start_expires(timer, HRTIMER_MODE_ABS);
1653}
1654
1655/*
1656 * apic_sync_pv_eoi_from_guest - called on vmexit or cancel interrupt
1657 *
1658 * Detect whether guest triggered PV EOI since the
1659 * last entry. If yes, set EOI on guests's behalf.
1660 * Clear PV EOI in guest memory in any case.
1661 */
1662static void apic_sync_pv_eoi_from_guest(struct kvm_vcpu *vcpu,
1663 struct kvm_lapic *apic)
1664{
1665 bool pending;
1666 int vector;
1667 /*
1668 * PV EOI state is derived from KVM_APIC_PV_EOI_PENDING in host
1669 * and KVM_PV_EOI_ENABLED in guest memory as follows:
1670 *
1671 * KVM_APIC_PV_EOI_PENDING is unset:
1672 * -> host disabled PV EOI.
1673 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is set:
1674 * -> host enabled PV EOI, guest did not execute EOI yet.
1675 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is unset:
1676 * -> host enabled PV EOI, guest executed EOI.
1677 */
1678 BUG_ON(!pv_eoi_enabled(vcpu));
1679 pending = pv_eoi_get_pending(vcpu);
1680 /*
1681 * Clear pending bit in any case: it will be set again on vmentry.
1682 * While this might not be ideal from performance point of view,
1683 * this makes sure pv eoi is only enabled when we know it's safe.
1684 */
1685 pv_eoi_clr_pending(vcpu);
1686 if (pending)
1687 return;
1688 vector = apic_set_eoi(apic);
1689 trace_kvm_pv_eoi(apic, vector);
1690}
1691
1692void kvm_lapic_sync_from_vapic(struct kvm_vcpu *vcpu)
1693{
1694 u32 data;
1695
1696 if (test_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention))
1697 apic_sync_pv_eoi_from_guest(vcpu, vcpu->arch.apic);
1698
1699 if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
1700 return;
1701
1702 kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
1703 sizeof(u32));
1704
1705 apic_set_tpr(vcpu->arch.apic, data & 0xff);
1706}
1707
1708/*
1709 * apic_sync_pv_eoi_to_guest - called before vmentry
1710 *
1711 * Detect whether it's safe to enable PV EOI and
1712 * if yes do so.
1713 */
1714static void apic_sync_pv_eoi_to_guest(struct kvm_vcpu *vcpu,
1715 struct kvm_lapic *apic)
1716{
1717 if (!pv_eoi_enabled(vcpu) ||
1718 /* IRR set or many bits in ISR: could be nested. */
1719 apic->irr_pending ||
1720 /* Cache not set: could be safe but we don't bother. */
1721 apic->highest_isr_cache == -1 ||
1722 /* Need EOI to update ioapic. */
1723 kvm_ioapic_handles_vector(vcpu->kvm, apic->highest_isr_cache)) {
1724 /*
1725 * PV EOI was disabled by apic_sync_pv_eoi_from_guest
1726 * so we need not do anything here.
1727 */
1728 return;
1729 }
1730
1731 pv_eoi_set_pending(apic->vcpu);
1732}
1733
1734void kvm_lapic_sync_to_vapic(struct kvm_vcpu *vcpu)
1735{
1736 u32 data, tpr;
1737 int max_irr, max_isr;
1738 struct kvm_lapic *apic = vcpu->arch.apic;
1739
1740 apic_sync_pv_eoi_to_guest(vcpu, apic);
1741
1742 if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
1743 return;
1744
1745 tpr = kvm_apic_get_reg(apic, APIC_TASKPRI) & 0xff;
1746 max_irr = apic_find_highest_irr(apic);
1747 if (max_irr < 0)
1748 max_irr = 0;
1749 max_isr = apic_find_highest_isr(apic);
1750 if (max_isr < 0)
1751 max_isr = 0;
1752 data = (tpr & 0xff) | ((max_isr & 0xf0) << 8) | (max_irr << 24);
1753
1754 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
1755 sizeof(u32));
1756}
1757
1758int kvm_lapic_set_vapic_addr(struct kvm_vcpu *vcpu, gpa_t vapic_addr)
1759{
1760 if (vapic_addr) {
1761 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
1762 &vcpu->arch.apic->vapic_cache,
1763 vapic_addr, sizeof(u32)))
1764 return -EINVAL;
1765 __set_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
1766 } else {
1767 __clear_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
1768 }
1769
1770 vcpu->arch.apic->vapic_addr = vapic_addr;
1771 return 0;
1772}
1773
1774int kvm_x2apic_msr_write(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1775{
1776 struct kvm_lapic *apic = vcpu->arch.apic;
1777 u32 reg = (msr - APIC_BASE_MSR) << 4;
1778
1779 if (!irqchip_in_kernel(vcpu->kvm) || !apic_x2apic_mode(apic))
1780 return 1;
1781
1782 /* if this is ICR write vector before command */
1783 if (msr == 0x830)
1784 apic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
1785 return apic_reg_write(apic, reg, (u32)data);
1786}
1787
1788int kvm_x2apic_msr_read(struct kvm_vcpu *vcpu, u32 msr, u64 *data)
1789{
1790 struct kvm_lapic *apic = vcpu->arch.apic;
1791 u32 reg = (msr - APIC_BASE_MSR) << 4, low, high = 0;
1792
1793 if (!irqchip_in_kernel(vcpu->kvm) || !apic_x2apic_mode(apic))
1794 return 1;
1795
1796 if (apic_reg_read(apic, reg, 4, &low))
1797 return 1;
1798 if (msr == 0x830)
1799 apic_reg_read(apic, APIC_ICR2, 4, &high);
1800
1801 *data = (((u64)high) << 32) | low;
1802
1803 return 0;
1804}
1805
1806int kvm_hv_vapic_msr_write(struct kvm_vcpu *vcpu, u32 reg, u64 data)
1807{
1808 struct kvm_lapic *apic = vcpu->arch.apic;
1809
1810 if (!kvm_vcpu_has_lapic(vcpu))
1811 return 1;
1812
1813 /* if this is ICR write vector before command */
1814 if (reg == APIC_ICR)
1815 apic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
1816 return apic_reg_write(apic, reg, (u32)data);
1817}
1818
1819int kvm_hv_vapic_msr_read(struct kvm_vcpu *vcpu, u32 reg, u64 *data)
1820{
1821 struct kvm_lapic *apic = vcpu->arch.apic;
1822 u32 low, high = 0;
1823
1824 if (!kvm_vcpu_has_lapic(vcpu))
1825 return 1;
1826
1827 if (apic_reg_read(apic, reg, 4, &low))
1828 return 1;
1829 if (reg == APIC_ICR)
1830 apic_reg_read(apic, APIC_ICR2, 4, &high);
1831
1832 *data = (((u64)high) << 32) | low;
1833
1834 return 0;
1835}
1836
1837int kvm_lapic_enable_pv_eoi(struct kvm_vcpu *vcpu, u64 data)
1838{
1839 u64 addr = data & ~KVM_MSR_ENABLED;
1840 if (!IS_ALIGNED(addr, 4))
1841 return 1;
1842
1843 vcpu->arch.pv_eoi.msr_val = data;
1844 if (!pv_eoi_enabled(vcpu))
1845 return 0;
1846 return kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.pv_eoi.data,
1847 addr, sizeof(u8));
1848}
1849
1850void kvm_apic_accept_events(struct kvm_vcpu *vcpu)
1851{
1852 struct kvm_lapic *apic = vcpu->arch.apic;
1853 unsigned int sipi_vector;
1854 unsigned long pe;
1855
1856 if (!kvm_vcpu_has_lapic(vcpu) || !apic->pending_events)
1857 return;
1858
1859 pe = xchg(&apic->pending_events, 0);
1860
1861 if (test_bit(KVM_APIC_INIT, &pe)) {
1862 kvm_lapic_reset(vcpu);
1863 kvm_vcpu_reset(vcpu);
1864 if (kvm_vcpu_is_bsp(apic->vcpu))
1865 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
1866 else
1867 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
1868 }
1869 if (test_bit(KVM_APIC_SIPI, &pe) &&
1870 vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
1871 /* evaluate pending_events before reading the vector */
1872 smp_rmb();
1873 sipi_vector = apic->sipi_vector;
1874 pr_debug("vcpu %d received sipi with vector # %x\n",
1875 vcpu->vcpu_id, sipi_vector);
1876 kvm_vcpu_deliver_sipi_vector(vcpu, sipi_vector);
1877 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
1878 }
1879}
1880
1881void kvm_lapic_init(void)
1882{
1883 /* do not patch jump label more than once per second */
1884 jump_label_rate_limit(&apic_hw_disabled, HZ);
1885 jump_label_rate_limit(&apic_sw_disabled, HZ);
1886}
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/export.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 <asm/delay.h>
37#include <linux/atomic.h>
38#include <linux/jump_label.h>
39#include "kvm_cache_regs.h"
40#include "irq.h"
41#include "trace.h"
42#include "x86.h"
43#include "cpuid.h"
44#include "hyperv.h"
45
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
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
62/* 14 is the version for Xeon and Pentium 8.4.8*/
63#define APIC_VERSION (0x14UL | ((KVM_APIC_LVT_NUM - 1) << 16))
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
70#define APIC_VECTORS_PER_REG 32
71
72#define APIC_BROADCAST 0xFF
73#define X2APIC_BROADCAST 0xFFFFFFFFul
74
75static inline int apic_test_vector(int vec, void *bitmap)
76{
77 return test_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
78}
79
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
88static inline void apic_clear_vector(int vec, void *bitmap)
89{
90 clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
91}
92
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
103struct static_key_deferred apic_hw_disabled __read_mostly;
104struct static_key_deferred apic_sw_disabled __read_mostly;
105
106static inline int apic_enabled(struct kvm_lapic *apic)
107{
108 return kvm_apic_sw_enabled(apic) && kvm_apic_hw_enabled(apic);
109}
110
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
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;
123 u32 max_apic_id = map->max_apic_id;
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 }
133
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) & 0xf];
142 *mask = dest_id & 0xf;
143 return true;
144 default:
145 /* Not optimized. */
146 return false;
147 }
148}
149
150static void kvm_apic_map_free(struct rcu_head *rcu)
151{
152 struct kvm_apic_map *map = container_of(rcu, struct kvm_apic_map, rcu);
153
154 kvfree(map);
155}
156
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;
162 u32 max_id = 255;
163
164 mutex_lock(&kvm->arch.apic_map_lock);
165
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
170 new = kvm_kvzalloc(sizeof(struct kvm_apic_map) +
171 sizeof(struct kvm_lapic *) * ((u64)max_id + 1));
172
173 if (!new)
174 goto out;
175
176 new->max_apic_id = max_id;
177
178 kvm_for_each_vcpu(i, vcpu, kvm) {
179 struct kvm_lapic *apic = vcpu->arch.apic;
180 struct kvm_lapic **cluster;
181 u16 mask;
182 u32 ldr, aid;
183
184 if (!kvm_apic_present(vcpu))
185 continue;
186
187 aid = kvm_apic_id(apic);
188 ldr = kvm_lapic_get_reg(apic, APIC_LDR);
189
190 if (aid <= new->max_apic_id)
191 new->phys_map[aid] = apic;
192
193 if (apic_x2apic_mode(apic)) {
194 new->mode |= KVM_APIC_MODE_X2APIC;
195 } else if (ldr) {
196 ldr = GET_APIC_LOGICAL_ID(ldr);
197 if (kvm_lapic_get_reg(apic, APIC_DFR) == APIC_DFR_FLAT)
198 new->mode |= KVM_APIC_MODE_XAPIC_FLAT;
199 else
200 new->mode |= KVM_APIC_MODE_XAPIC_CLUSTER;
201 }
202
203 if (!kvm_apic_map_get_logical_dest(new, ldr, &cluster, &mask))
204 continue;
205
206 if (mask)
207 cluster[ffs(mask) - 1] = apic;
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)
216 call_rcu(&old->rcu, kvm_apic_map_free);
217
218 kvm_make_scan_ioapic_request(kvm);
219}
220
221static inline void apic_set_spiv(struct kvm_lapic *apic, u32 val)
222{
223 bool enabled = val & APIC_SPIV_APIC_ENABLED;
224
225 kvm_lapic_set_reg(apic, APIC_SPIV, val);
226
227 if (enabled != apic->sw_enabled) {
228 apic->sw_enabled = enabled;
229 if (enabled) {
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
237static inline void kvm_apic_set_xapic_id(struct kvm_lapic *apic, u8 id)
238{
239 kvm_lapic_set_reg(apic, APIC_ID, id << 24);
240 recalculate_apic_map(apic->vcpu->kvm);
241}
242
243static inline void kvm_apic_set_ldr(struct kvm_lapic *apic, u32 id)
244{
245 kvm_lapic_set_reg(apic, APIC_LDR, id);
246 recalculate_apic_map(apic->vcpu->kvm);
247}
248
249static inline void kvm_apic_set_x2apic_id(struct kvm_lapic *apic, u32 id)
250{
251 u32 ldr = ((id >> 4) << 16) | (1 << (id & 0xf));
252
253 kvm_lapic_set_reg(apic, APIC_ID, id);
254 kvm_lapic_set_reg(apic, APIC_LDR, ldr);
255 recalculate_apic_map(apic->vcpu->kvm);
256}
257
258static inline int apic_lvt_enabled(struct kvm_lapic *apic, int lvt_type)
259{
260 return !(kvm_lapic_get_reg(apic, lvt_type) & APIC_LVT_MASKED);
261}
262
263static inline int apic_lvt_vector(struct kvm_lapic *apic, int lvt_type)
264{
265 return kvm_lapic_get_reg(apic, lvt_type) & APIC_VECTOR_MASK;
266}
267
268static inline int apic_lvtt_oneshot(struct kvm_lapic *apic)
269{
270 return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_ONESHOT;
271}
272
273static inline int apic_lvtt_period(struct kvm_lapic *apic)
274{
275 return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_PERIODIC;
276}
277
278static inline int apic_lvtt_tscdeadline(struct kvm_lapic *apic)
279{
280 return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_TSCDEADLINE;
281}
282
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
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
294 if (!lapic_in_kernel(vcpu))
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;
300 kvm_lapic_set_reg(apic, APIC_LVR, v);
301}
302
303static const unsigned int apic_lvt_mask[KVM_APIC_LVT_NUM] = {
304 LVT_MASK , /* part LVTT mask, timer mode mask added at runtime */
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{
313 int vec;
314 u32 *reg;
315
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 }
322
323 return -1;
324}
325
326static u8 count_vectors(void *bitmap)
327{
328 int vec;
329 u32 *reg;
330 u8 count = 0;
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
337 return count;
338}
339
340void __kvm_apic_update_irr(u32 *pir, void *regs)
341{
342 u32 i, pir_val;
343
344 for (i = 0; i <= 7; i++) {
345 pir_val = READ_ONCE(pir[i]);
346 if (pir_val) {
347 pir_val = xchg(&pir[i], 0);
348 *((u32 *)(regs + APIC_IRR + i * 0x10)) |= pir_val;
349 }
350 }
351}
352EXPORT_SYMBOL_GPL(__kvm_apic_update_irr);
353
354void kvm_apic_update_irr(struct kvm_vcpu *vcpu, u32 *pir)
355{
356 struct kvm_lapic *apic = vcpu->arch.apic;
357
358 __kvm_apic_update_irr(pir, apic->regs);
359
360 kvm_make_request(KVM_REQ_EVENT, vcpu);
361}
362EXPORT_SYMBOL_GPL(kvm_apic_update_irr);
363
364static inline int apic_search_irr(struct kvm_lapic *apic)
365{
366 return find_highest_vector(apic->regs + APIC_IRR);
367}
368
369static inline int apic_find_highest_irr(struct kvm_lapic *apic)
370{
371 int result;
372
373 /*
374 * Note that irr_pending is just a hint. It will be always
375 * true with virtual interrupt delivery enabled.
376 */
377 if (!apic->irr_pending)
378 return -1;
379
380 if (apic->vcpu->arch.apicv_active)
381 kvm_x86_ops->sync_pir_to_irr(apic->vcpu);
382 result = apic_search_irr(apic);
383 ASSERT(result == -1 || result >= 16);
384
385 return result;
386}
387
388static inline void apic_clear_irr(int vec, struct kvm_lapic *apic)
389{
390 struct kvm_vcpu *vcpu;
391
392 vcpu = apic->vcpu;
393
394 if (unlikely(vcpu->arch.apicv_active)) {
395 /* try to update RVI */
396 apic_clear_vector(vec, apic->regs + APIC_IRR);
397 kvm_make_request(KVM_REQ_EVENT, vcpu);
398 } else {
399 apic->irr_pending = false;
400 apic_clear_vector(vec, apic->regs + APIC_IRR);
401 if (apic_search_irr(apic) != -1)
402 apic->irr_pending = true;
403 }
404}
405
406static inline void apic_set_isr(int vec, struct kvm_lapic *apic)
407{
408 struct kvm_vcpu *vcpu;
409
410 if (__apic_test_and_set_vector(vec, apic->regs + APIC_ISR))
411 return;
412
413 vcpu = apic->vcpu;
414
415 /*
416 * With APIC virtualization enabled, all caching is disabled
417 * because the processor can modify ISR under the hood. Instead
418 * just set SVI.
419 */
420 if (unlikely(vcpu->arch.apicv_active))
421 kvm_x86_ops->hwapic_isr_update(vcpu, vec);
422 else {
423 ++apic->isr_count;
424 BUG_ON(apic->isr_count > MAX_APIC_VECTOR);
425 /*
426 * ISR (in service register) bit is set when injecting an interrupt.
427 * The highest vector is injected. Thus the latest bit set matches
428 * the highest bit in ISR.
429 */
430 apic->highest_isr_cache = vec;
431 }
432}
433
434static inline int apic_find_highest_isr(struct kvm_lapic *apic)
435{
436 int result;
437
438 /*
439 * Note that isr_count is always 1, and highest_isr_cache
440 * is always -1, with APIC virtualization enabled.
441 */
442 if (!apic->isr_count)
443 return -1;
444 if (likely(apic->highest_isr_cache != -1))
445 return apic->highest_isr_cache;
446
447 result = find_highest_vector(apic->regs + APIC_ISR);
448 ASSERT(result == -1 || result >= 16);
449
450 return result;
451}
452
453static inline void apic_clear_isr(int vec, struct kvm_lapic *apic)
454{
455 struct kvm_vcpu *vcpu;
456 if (!__apic_test_and_clear_vector(vec, apic->regs + APIC_ISR))
457 return;
458
459 vcpu = apic->vcpu;
460
461 /*
462 * We do get here for APIC virtualization enabled if the guest
463 * uses the Hyper-V APIC enlightenment. In this case we may need
464 * to trigger a new interrupt delivery by writing the SVI field;
465 * on the other hand isr_count and highest_isr_cache are unused
466 * and must be left alone.
467 */
468 if (unlikely(vcpu->arch.apicv_active))
469 kvm_x86_ops->hwapic_isr_update(vcpu,
470 apic_find_highest_isr(apic));
471 else {
472 --apic->isr_count;
473 BUG_ON(apic->isr_count < 0);
474 apic->highest_isr_cache = -1;
475 }
476}
477
478int kvm_lapic_find_highest_irr(struct kvm_vcpu *vcpu)
479{
480 /* This may race with setting of irr in __apic_accept_irq() and
481 * value returned may be wrong, but kvm_vcpu_kick() in __apic_accept_irq
482 * will cause vmexit immediately and the value will be recalculated
483 * on the next vmentry.
484 */
485 return apic_find_highest_irr(vcpu->arch.apic);
486}
487
488static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
489 int vector, int level, int trig_mode,
490 struct dest_map *dest_map);
491
492int kvm_apic_set_irq(struct kvm_vcpu *vcpu, struct kvm_lapic_irq *irq,
493 struct dest_map *dest_map)
494{
495 struct kvm_lapic *apic = vcpu->arch.apic;
496
497 return __apic_accept_irq(apic, irq->delivery_mode, irq->vector,
498 irq->level, irq->trig_mode, dest_map);
499}
500
501static int pv_eoi_put_user(struct kvm_vcpu *vcpu, u8 val)
502{
503
504 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, &val,
505 sizeof(val));
506}
507
508static int pv_eoi_get_user(struct kvm_vcpu *vcpu, u8 *val)
509{
510
511 return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, val,
512 sizeof(*val));
513}
514
515static inline bool pv_eoi_enabled(struct kvm_vcpu *vcpu)
516{
517 return vcpu->arch.pv_eoi.msr_val & KVM_MSR_ENABLED;
518}
519
520static bool pv_eoi_get_pending(struct kvm_vcpu *vcpu)
521{
522 u8 val;
523 if (pv_eoi_get_user(vcpu, &val) < 0)
524 apic_debug("Can't read EOI MSR value: 0x%llx\n",
525 (unsigned long long)vcpu->arch.pv_eoi.msr_val);
526 return val & 0x1;
527}
528
529static void pv_eoi_set_pending(struct kvm_vcpu *vcpu)
530{
531 if (pv_eoi_put_user(vcpu, KVM_PV_EOI_ENABLED) < 0) {
532 apic_debug("Can't set EOI MSR value: 0x%llx\n",
533 (unsigned long long)vcpu->arch.pv_eoi.msr_val);
534 return;
535 }
536 __set_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
537}
538
539static void pv_eoi_clr_pending(struct kvm_vcpu *vcpu)
540{
541 if (pv_eoi_put_user(vcpu, KVM_PV_EOI_DISABLED) < 0) {
542 apic_debug("Can't clear EOI MSR value: 0x%llx\n",
543 (unsigned long long)vcpu->arch.pv_eoi.msr_val);
544 return;
545 }
546 __clear_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
547}
548
549static void apic_update_ppr(struct kvm_lapic *apic)
550{
551 u32 tpr, isrv, ppr, old_ppr;
552 int isr;
553
554 old_ppr = kvm_lapic_get_reg(apic, APIC_PROCPRI);
555 tpr = kvm_lapic_get_reg(apic, APIC_TASKPRI);
556 isr = apic_find_highest_isr(apic);
557 isrv = (isr != -1) ? isr : 0;
558
559 if ((tpr & 0xf0) >= (isrv & 0xf0))
560 ppr = tpr & 0xff;
561 else
562 ppr = isrv & 0xf0;
563
564 apic_debug("vlapic %p, ppr 0x%x, isr 0x%x, isrv 0x%x",
565 apic, ppr, isr, isrv);
566
567 if (old_ppr != ppr) {
568 kvm_lapic_set_reg(apic, APIC_PROCPRI, ppr);
569 if (ppr < old_ppr)
570 kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
571 }
572}
573
574static void apic_set_tpr(struct kvm_lapic *apic, u32 tpr)
575{
576 kvm_lapic_set_reg(apic, APIC_TASKPRI, tpr);
577 apic_update_ppr(apic);
578}
579
580static bool kvm_apic_broadcast(struct kvm_lapic *apic, u32 mda)
581{
582 if (apic_x2apic_mode(apic))
583 return mda == X2APIC_BROADCAST;
584
585 return GET_APIC_DEST_FIELD(mda) == APIC_BROADCAST;
586}
587
588static bool kvm_apic_match_physical_addr(struct kvm_lapic *apic, u32 mda)
589{
590 if (kvm_apic_broadcast(apic, mda))
591 return true;
592
593 if (apic_x2apic_mode(apic))
594 return mda == kvm_apic_id(apic);
595
596 return mda == SET_APIC_DEST_FIELD(kvm_apic_id(apic));
597}
598
599static bool kvm_apic_match_logical_addr(struct kvm_lapic *apic, u32 mda)
600{
601 u32 logical_id;
602
603 if (kvm_apic_broadcast(apic, mda))
604 return true;
605
606 logical_id = kvm_lapic_get_reg(apic, APIC_LDR);
607
608 if (apic_x2apic_mode(apic))
609 return ((logical_id >> 16) == (mda >> 16))
610 && (logical_id & mda & 0xffff) != 0;
611
612 logical_id = GET_APIC_LOGICAL_ID(logical_id);
613 mda = GET_APIC_DEST_FIELD(mda);
614
615 switch (kvm_lapic_get_reg(apic, APIC_DFR)) {
616 case APIC_DFR_FLAT:
617 return (logical_id & mda) != 0;
618 case APIC_DFR_CLUSTER:
619 return ((logical_id >> 4) == (mda >> 4))
620 && (logical_id & mda & 0xf) != 0;
621 default:
622 apic_debug("Bad DFR vcpu %d: %08x\n",
623 apic->vcpu->vcpu_id, kvm_lapic_get_reg(apic, APIC_DFR));
624 return false;
625 }
626}
627
628/* The KVM local APIC implementation has two quirks:
629 *
630 * - the xAPIC MDA stores the destination at bits 24-31, while this
631 * is not true of struct kvm_lapic_irq's dest_id field. This is
632 * just a quirk in the API and is not problematic.
633 *
634 * - in-kernel IOAPIC messages have to be delivered directly to
635 * x2APIC, because the kernel does not support interrupt remapping.
636 * In order to support broadcast without interrupt remapping, x2APIC
637 * rewrites the destination of non-IPI messages from APIC_BROADCAST
638 * to X2APIC_BROADCAST.
639 *
640 * The broadcast quirk can be disabled with KVM_CAP_X2APIC_API. This is
641 * important when userspace wants to use x2APIC-format MSIs, because
642 * APIC_BROADCAST (0xff) is a legal route for "cluster 0, CPUs 0-7".
643 */
644static u32 kvm_apic_mda(struct kvm_vcpu *vcpu, unsigned int dest_id,
645 struct kvm_lapic *source, struct kvm_lapic *target)
646{
647 bool ipi = source != NULL;
648 bool x2apic_mda = apic_x2apic_mode(ipi ? source : target);
649
650 if (!vcpu->kvm->arch.x2apic_broadcast_quirk_disabled &&
651 !ipi && dest_id == APIC_BROADCAST && x2apic_mda)
652 return X2APIC_BROADCAST;
653
654 return x2apic_mda ? dest_id : SET_APIC_DEST_FIELD(dest_id);
655}
656
657bool kvm_apic_match_dest(struct kvm_vcpu *vcpu, struct kvm_lapic *source,
658 int short_hand, unsigned int dest, int dest_mode)
659{
660 struct kvm_lapic *target = vcpu->arch.apic;
661 u32 mda = kvm_apic_mda(vcpu, dest, source, target);
662
663 apic_debug("target %p, source %p, dest 0x%x, "
664 "dest_mode 0x%x, short_hand 0x%x\n",
665 target, source, dest, dest_mode, short_hand);
666
667 ASSERT(target);
668 switch (short_hand) {
669 case APIC_DEST_NOSHORT:
670 if (dest_mode == APIC_DEST_PHYSICAL)
671 return kvm_apic_match_physical_addr(target, mda);
672 else
673 return kvm_apic_match_logical_addr(target, mda);
674 case APIC_DEST_SELF:
675 return target == source;
676 case APIC_DEST_ALLINC:
677 return true;
678 case APIC_DEST_ALLBUT:
679 return target != source;
680 default:
681 apic_debug("kvm: apic: Bad dest shorthand value %x\n",
682 short_hand);
683 return false;
684 }
685}
686EXPORT_SYMBOL_GPL(kvm_apic_match_dest);
687
688int kvm_vector_to_index(u32 vector, u32 dest_vcpus,
689 const unsigned long *bitmap, u32 bitmap_size)
690{
691 u32 mod;
692 int i, idx = -1;
693
694 mod = vector % dest_vcpus;
695
696 for (i = 0; i <= mod; i++) {
697 idx = find_next_bit(bitmap, bitmap_size, idx + 1);
698 BUG_ON(idx == bitmap_size);
699 }
700
701 return idx;
702}
703
704static void kvm_apic_disabled_lapic_found(struct kvm *kvm)
705{
706 if (!kvm->arch.disabled_lapic_found) {
707 kvm->arch.disabled_lapic_found = true;
708 printk(KERN_INFO
709 "Disabled LAPIC found during irq injection\n");
710 }
711}
712
713static bool kvm_apic_is_broadcast_dest(struct kvm *kvm, struct kvm_lapic **src,
714 struct kvm_lapic_irq *irq, struct kvm_apic_map *map)
715{
716 if (kvm->arch.x2apic_broadcast_quirk_disabled) {
717 if ((irq->dest_id == APIC_BROADCAST &&
718 map->mode != KVM_APIC_MODE_X2APIC))
719 return true;
720 if (irq->dest_id == X2APIC_BROADCAST)
721 return true;
722 } else {
723 bool x2apic_ipi = src && *src && apic_x2apic_mode(*src);
724 if (irq->dest_id == (x2apic_ipi ?
725 X2APIC_BROADCAST : APIC_BROADCAST))
726 return true;
727 }
728
729 return false;
730}
731
732/* Return true if the interrupt can be handled by using *bitmap as index mask
733 * for valid destinations in *dst array.
734 * Return false if kvm_apic_map_get_dest_lapic did nothing useful.
735 * Note: we may have zero kvm_lapic destinations when we return true, which
736 * means that the interrupt should be dropped. In this case, *bitmap would be
737 * zero and *dst undefined.
738 */
739static inline bool kvm_apic_map_get_dest_lapic(struct kvm *kvm,
740 struct kvm_lapic **src, struct kvm_lapic_irq *irq,
741 struct kvm_apic_map *map, struct kvm_lapic ***dst,
742 unsigned long *bitmap)
743{
744 int i, lowest;
745
746 if (irq->shorthand == APIC_DEST_SELF && src) {
747 *dst = src;
748 *bitmap = 1;
749 return true;
750 } else if (irq->shorthand)
751 return false;
752
753 if (!map || kvm_apic_is_broadcast_dest(kvm, src, irq, map))
754 return false;
755
756 if (irq->dest_mode == APIC_DEST_PHYSICAL) {
757 if (irq->dest_id > map->max_apic_id) {
758 *bitmap = 0;
759 } else {
760 *dst = &map->phys_map[irq->dest_id];
761 *bitmap = 1;
762 }
763 return true;
764 }
765
766 *bitmap = 0;
767 if (!kvm_apic_map_get_logical_dest(map, irq->dest_id, dst,
768 (u16 *)bitmap))
769 return false;
770
771 if (!kvm_lowest_prio_delivery(irq))
772 return true;
773
774 if (!kvm_vector_hashing_enabled()) {
775 lowest = -1;
776 for_each_set_bit(i, bitmap, 16) {
777 if (!(*dst)[i])
778 continue;
779 if (lowest < 0)
780 lowest = i;
781 else if (kvm_apic_compare_prio((*dst)[i]->vcpu,
782 (*dst)[lowest]->vcpu) < 0)
783 lowest = i;
784 }
785 } else {
786 if (!*bitmap)
787 return true;
788
789 lowest = kvm_vector_to_index(irq->vector, hweight16(*bitmap),
790 bitmap, 16);
791
792 if (!(*dst)[lowest]) {
793 kvm_apic_disabled_lapic_found(kvm);
794 *bitmap = 0;
795 return true;
796 }
797 }
798
799 *bitmap = (lowest >= 0) ? 1 << lowest : 0;
800
801 return true;
802}
803
804bool kvm_irq_delivery_to_apic_fast(struct kvm *kvm, struct kvm_lapic *src,
805 struct kvm_lapic_irq *irq, int *r, struct dest_map *dest_map)
806{
807 struct kvm_apic_map *map;
808 unsigned long bitmap;
809 struct kvm_lapic **dst = NULL;
810 int i;
811 bool ret;
812
813 *r = -1;
814
815 if (irq->shorthand == APIC_DEST_SELF) {
816 *r = kvm_apic_set_irq(src->vcpu, irq, dest_map);
817 return true;
818 }
819
820 rcu_read_lock();
821 map = rcu_dereference(kvm->arch.apic_map);
822
823 ret = kvm_apic_map_get_dest_lapic(kvm, &src, irq, map, &dst, &bitmap);
824 if (ret)
825 for_each_set_bit(i, &bitmap, 16) {
826 if (!dst[i])
827 continue;
828 if (*r < 0)
829 *r = 0;
830 *r += kvm_apic_set_irq(dst[i]->vcpu, irq, dest_map);
831 }
832
833 rcu_read_unlock();
834 return ret;
835}
836
837/*
838 * This routine tries to handler interrupts in posted mode, here is how
839 * it deals with different cases:
840 * - For single-destination interrupts, handle it in posted mode
841 * - Else if vector hashing is enabled and it is a lowest-priority
842 * interrupt, handle it in posted mode and use the following mechanism
843 * to find the destinaiton vCPU.
844 * 1. For lowest-priority interrupts, store all the possible
845 * destination vCPUs in an array.
846 * 2. Use "guest vector % max number of destination vCPUs" to find
847 * the right destination vCPU in the array for the lowest-priority
848 * interrupt.
849 * - Otherwise, use remapped mode to inject the interrupt.
850 */
851bool kvm_intr_is_single_vcpu_fast(struct kvm *kvm, struct kvm_lapic_irq *irq,
852 struct kvm_vcpu **dest_vcpu)
853{
854 struct kvm_apic_map *map;
855 unsigned long bitmap;
856 struct kvm_lapic **dst = NULL;
857 bool ret = false;
858
859 if (irq->shorthand)
860 return false;
861
862 rcu_read_lock();
863 map = rcu_dereference(kvm->arch.apic_map);
864
865 if (kvm_apic_map_get_dest_lapic(kvm, NULL, irq, map, &dst, &bitmap) &&
866 hweight16(bitmap) == 1) {
867 unsigned long i = find_first_bit(&bitmap, 16);
868
869 if (dst[i]) {
870 *dest_vcpu = dst[i]->vcpu;
871 ret = true;
872 }
873 }
874
875 rcu_read_unlock();
876 return ret;
877}
878
879/*
880 * Add a pending IRQ into lapic.
881 * Return 1 if successfully added and 0 if discarded.
882 */
883static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
884 int vector, int level, int trig_mode,
885 struct dest_map *dest_map)
886{
887 int result = 0;
888 struct kvm_vcpu *vcpu = apic->vcpu;
889
890 trace_kvm_apic_accept_irq(vcpu->vcpu_id, delivery_mode,
891 trig_mode, vector);
892 switch (delivery_mode) {
893 case APIC_DM_LOWEST:
894 vcpu->arch.apic_arb_prio++;
895 case APIC_DM_FIXED:
896 if (unlikely(trig_mode && !level))
897 break;
898
899 /* FIXME add logic for vcpu on reset */
900 if (unlikely(!apic_enabled(apic)))
901 break;
902
903 result = 1;
904
905 if (dest_map) {
906 __set_bit(vcpu->vcpu_id, dest_map->map);
907 dest_map->vectors[vcpu->vcpu_id] = vector;
908 }
909
910 if (apic_test_vector(vector, apic->regs + APIC_TMR) != !!trig_mode) {
911 if (trig_mode)
912 kvm_lapic_set_vector(vector, apic->regs + APIC_TMR);
913 else
914 apic_clear_vector(vector, apic->regs + APIC_TMR);
915 }
916
917 if (vcpu->arch.apicv_active)
918 kvm_x86_ops->deliver_posted_interrupt(vcpu, vector);
919 else {
920 kvm_lapic_set_irr(vector, apic);
921
922 kvm_make_request(KVM_REQ_EVENT, vcpu);
923 kvm_vcpu_kick(vcpu);
924 }
925 break;
926
927 case APIC_DM_REMRD:
928 result = 1;
929 vcpu->arch.pv.pv_unhalted = 1;
930 kvm_make_request(KVM_REQ_EVENT, vcpu);
931 kvm_vcpu_kick(vcpu);
932 break;
933
934 case APIC_DM_SMI:
935 result = 1;
936 kvm_make_request(KVM_REQ_SMI, vcpu);
937 kvm_vcpu_kick(vcpu);
938 break;
939
940 case APIC_DM_NMI:
941 result = 1;
942 kvm_inject_nmi(vcpu);
943 kvm_vcpu_kick(vcpu);
944 break;
945
946 case APIC_DM_INIT:
947 if (!trig_mode || level) {
948 result = 1;
949 /* assumes that there are only KVM_APIC_INIT/SIPI */
950 apic->pending_events = (1UL << KVM_APIC_INIT);
951 /* make sure pending_events is visible before sending
952 * the request */
953 smp_wmb();
954 kvm_make_request(KVM_REQ_EVENT, vcpu);
955 kvm_vcpu_kick(vcpu);
956 } else {
957 apic_debug("Ignoring de-assert INIT to vcpu %d\n",
958 vcpu->vcpu_id);
959 }
960 break;
961
962 case APIC_DM_STARTUP:
963 apic_debug("SIPI to vcpu %d vector 0x%02x\n",
964 vcpu->vcpu_id, vector);
965 result = 1;
966 apic->sipi_vector = vector;
967 /* make sure sipi_vector is visible for the receiver */
968 smp_wmb();
969 set_bit(KVM_APIC_SIPI, &apic->pending_events);
970 kvm_make_request(KVM_REQ_EVENT, vcpu);
971 kvm_vcpu_kick(vcpu);
972 break;
973
974 case APIC_DM_EXTINT:
975 /*
976 * Should only be called by kvm_apic_local_deliver() with LVT0,
977 * before NMI watchdog was enabled. Already handled by
978 * kvm_apic_accept_pic_intr().
979 */
980 break;
981
982 default:
983 printk(KERN_ERR "TODO: unsupported delivery mode %x\n",
984 delivery_mode);
985 break;
986 }
987 return result;
988}
989
990int kvm_apic_compare_prio(struct kvm_vcpu *vcpu1, struct kvm_vcpu *vcpu2)
991{
992 return vcpu1->arch.apic_arb_prio - vcpu2->arch.apic_arb_prio;
993}
994
995static bool kvm_ioapic_handles_vector(struct kvm_lapic *apic, int vector)
996{
997 return test_bit(vector, apic->vcpu->arch.ioapic_handled_vectors);
998}
999
1000static void kvm_ioapic_send_eoi(struct kvm_lapic *apic, int vector)
1001{
1002 int trigger_mode;
1003
1004 /* Eoi the ioapic only if the ioapic doesn't own the vector. */
1005 if (!kvm_ioapic_handles_vector(apic, vector))
1006 return;
1007
1008 /* Request a KVM exit to inform the userspace IOAPIC. */
1009 if (irqchip_split(apic->vcpu->kvm)) {
1010 apic->vcpu->arch.pending_ioapic_eoi = vector;
1011 kvm_make_request(KVM_REQ_IOAPIC_EOI_EXIT, apic->vcpu);
1012 return;
1013 }
1014
1015 if (apic_test_vector(vector, apic->regs + APIC_TMR))
1016 trigger_mode = IOAPIC_LEVEL_TRIG;
1017 else
1018 trigger_mode = IOAPIC_EDGE_TRIG;
1019
1020 kvm_ioapic_update_eoi(apic->vcpu, vector, trigger_mode);
1021}
1022
1023static int apic_set_eoi(struct kvm_lapic *apic)
1024{
1025 int vector = apic_find_highest_isr(apic);
1026
1027 trace_kvm_eoi(apic, vector);
1028
1029 /*
1030 * Not every write EOI will has corresponding ISR,
1031 * one example is when Kernel check timer on setup_IO_APIC
1032 */
1033 if (vector == -1)
1034 return vector;
1035
1036 apic_clear_isr(vector, apic);
1037 apic_update_ppr(apic);
1038
1039 if (test_bit(vector, vcpu_to_synic(apic->vcpu)->vec_bitmap))
1040 kvm_hv_synic_send_eoi(apic->vcpu, vector);
1041
1042 kvm_ioapic_send_eoi(apic, vector);
1043 kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
1044 return vector;
1045}
1046
1047/*
1048 * this interface assumes a trap-like exit, which has already finished
1049 * desired side effect including vISR and vPPR update.
1050 */
1051void kvm_apic_set_eoi_accelerated(struct kvm_vcpu *vcpu, int vector)
1052{
1053 struct kvm_lapic *apic = vcpu->arch.apic;
1054
1055 trace_kvm_eoi(apic, vector);
1056
1057 kvm_ioapic_send_eoi(apic, vector);
1058 kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
1059}
1060EXPORT_SYMBOL_GPL(kvm_apic_set_eoi_accelerated);
1061
1062static void apic_send_ipi(struct kvm_lapic *apic)
1063{
1064 u32 icr_low = kvm_lapic_get_reg(apic, APIC_ICR);
1065 u32 icr_high = kvm_lapic_get_reg(apic, APIC_ICR2);
1066 struct kvm_lapic_irq irq;
1067
1068 irq.vector = icr_low & APIC_VECTOR_MASK;
1069 irq.delivery_mode = icr_low & APIC_MODE_MASK;
1070 irq.dest_mode = icr_low & APIC_DEST_MASK;
1071 irq.level = (icr_low & APIC_INT_ASSERT) != 0;
1072 irq.trig_mode = icr_low & APIC_INT_LEVELTRIG;
1073 irq.shorthand = icr_low & APIC_SHORT_MASK;
1074 irq.msi_redir_hint = false;
1075 if (apic_x2apic_mode(apic))
1076 irq.dest_id = icr_high;
1077 else
1078 irq.dest_id = GET_APIC_DEST_FIELD(icr_high);
1079
1080 trace_kvm_apic_ipi(icr_low, irq.dest_id);
1081
1082 apic_debug("icr_high 0x%x, icr_low 0x%x, "
1083 "short_hand 0x%x, dest 0x%x, trig_mode 0x%x, level 0x%x, "
1084 "dest_mode 0x%x, delivery_mode 0x%x, vector 0x%x, "
1085 "msi_redir_hint 0x%x\n",
1086 icr_high, icr_low, irq.shorthand, irq.dest_id,
1087 irq.trig_mode, irq.level, irq.dest_mode, irq.delivery_mode,
1088 irq.vector, irq.msi_redir_hint);
1089
1090 kvm_irq_delivery_to_apic(apic->vcpu->kvm, apic, &irq, NULL);
1091}
1092
1093static u32 apic_get_tmcct(struct kvm_lapic *apic)
1094{
1095 ktime_t remaining, now;
1096 s64 ns;
1097 u32 tmcct;
1098
1099 ASSERT(apic != NULL);
1100
1101 /* if initial count is 0, current count should also be 0 */
1102 if (kvm_lapic_get_reg(apic, APIC_TMICT) == 0 ||
1103 apic->lapic_timer.period == 0)
1104 return 0;
1105
1106 now = ktime_get();
1107 remaining = ktime_sub(apic->lapic_timer.target_expiration, now);
1108 if (ktime_to_ns(remaining) < 0)
1109 remaining = 0;
1110
1111 ns = mod_64(ktime_to_ns(remaining), apic->lapic_timer.period);
1112 tmcct = div64_u64(ns,
1113 (APIC_BUS_CYCLE_NS * apic->divide_count));
1114
1115 return tmcct;
1116}
1117
1118static void __report_tpr_access(struct kvm_lapic *apic, bool write)
1119{
1120 struct kvm_vcpu *vcpu = apic->vcpu;
1121 struct kvm_run *run = vcpu->run;
1122
1123 kvm_make_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu);
1124 run->tpr_access.rip = kvm_rip_read(vcpu);
1125 run->tpr_access.is_write = write;
1126}
1127
1128static inline void report_tpr_access(struct kvm_lapic *apic, bool write)
1129{
1130 if (apic->vcpu->arch.tpr_access_reporting)
1131 __report_tpr_access(apic, write);
1132}
1133
1134static u32 __apic_read(struct kvm_lapic *apic, unsigned int offset)
1135{
1136 u32 val = 0;
1137
1138 if (offset >= LAPIC_MMIO_LENGTH)
1139 return 0;
1140
1141 switch (offset) {
1142 case APIC_ARBPRI:
1143 apic_debug("Access APIC ARBPRI register which is for P6\n");
1144 break;
1145
1146 case APIC_TMCCT: /* Timer CCR */
1147 if (apic_lvtt_tscdeadline(apic))
1148 return 0;
1149
1150 val = apic_get_tmcct(apic);
1151 break;
1152 case APIC_PROCPRI:
1153 apic_update_ppr(apic);
1154 val = kvm_lapic_get_reg(apic, offset);
1155 break;
1156 case APIC_TASKPRI:
1157 report_tpr_access(apic, false);
1158 /* fall thru */
1159 default:
1160 val = kvm_lapic_get_reg(apic, offset);
1161 break;
1162 }
1163
1164 return val;
1165}
1166
1167static inline struct kvm_lapic *to_lapic(struct kvm_io_device *dev)
1168{
1169 return container_of(dev, struct kvm_lapic, dev);
1170}
1171
1172int kvm_lapic_reg_read(struct kvm_lapic *apic, u32 offset, int len,
1173 void *data)
1174{
1175 unsigned char alignment = offset & 0xf;
1176 u32 result;
1177 /* this bitmask has a bit cleared for each reserved register */
1178 static const u64 rmask = 0x43ff01ffffffe70cULL;
1179
1180 if ((alignment + len) > 4) {
1181 apic_debug("KVM_APIC_READ: alignment error %x %d\n",
1182 offset, len);
1183 return 1;
1184 }
1185
1186 if (offset > 0x3f0 || !(rmask & (1ULL << (offset >> 4)))) {
1187 apic_debug("KVM_APIC_READ: read reserved register %x\n",
1188 offset);
1189 return 1;
1190 }
1191
1192 result = __apic_read(apic, offset & ~0xf);
1193
1194 trace_kvm_apic_read(offset, result);
1195
1196 switch (len) {
1197 case 1:
1198 case 2:
1199 case 4:
1200 memcpy(data, (char *)&result + alignment, len);
1201 break;
1202 default:
1203 printk(KERN_ERR "Local APIC read with len = %x, "
1204 "should be 1,2, or 4 instead\n", len);
1205 break;
1206 }
1207 return 0;
1208}
1209EXPORT_SYMBOL_GPL(kvm_lapic_reg_read);
1210
1211static int apic_mmio_in_range(struct kvm_lapic *apic, gpa_t addr)
1212{
1213 return kvm_apic_hw_enabled(apic) &&
1214 addr >= apic->base_address &&
1215 addr < apic->base_address + LAPIC_MMIO_LENGTH;
1216}
1217
1218static int apic_mmio_read(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
1219 gpa_t address, int len, void *data)
1220{
1221 struct kvm_lapic *apic = to_lapic(this);
1222 u32 offset = address - apic->base_address;
1223
1224 if (!apic_mmio_in_range(apic, address))
1225 return -EOPNOTSUPP;
1226
1227 kvm_lapic_reg_read(apic, offset, len, data);
1228
1229 return 0;
1230}
1231
1232static void update_divide_count(struct kvm_lapic *apic)
1233{
1234 u32 tmp1, tmp2, tdcr;
1235
1236 tdcr = kvm_lapic_get_reg(apic, APIC_TDCR);
1237 tmp1 = tdcr & 0xf;
1238 tmp2 = ((tmp1 & 0x3) | ((tmp1 & 0x8) >> 1)) + 1;
1239 apic->divide_count = 0x1 << (tmp2 & 0x7);
1240
1241 apic_debug("timer divide count is 0x%x\n",
1242 apic->divide_count);
1243}
1244
1245static void apic_update_lvtt(struct kvm_lapic *apic)
1246{
1247 u32 timer_mode = kvm_lapic_get_reg(apic, APIC_LVTT) &
1248 apic->lapic_timer.timer_mode_mask;
1249
1250 if (apic->lapic_timer.timer_mode != timer_mode) {
1251 apic->lapic_timer.timer_mode = timer_mode;
1252 hrtimer_cancel(&apic->lapic_timer.timer);
1253 }
1254}
1255
1256static void apic_timer_expired(struct kvm_lapic *apic)
1257{
1258 struct kvm_vcpu *vcpu = apic->vcpu;
1259 struct swait_queue_head *q = &vcpu->wq;
1260 struct kvm_timer *ktimer = &apic->lapic_timer;
1261
1262 if (atomic_read(&apic->lapic_timer.pending))
1263 return;
1264
1265 atomic_inc(&apic->lapic_timer.pending);
1266 kvm_set_pending_timer(vcpu);
1267
1268 if (swait_active(q))
1269 swake_up(q);
1270
1271 if (apic_lvtt_tscdeadline(apic))
1272 ktimer->expired_tscdeadline = ktimer->tscdeadline;
1273}
1274
1275/*
1276 * On APICv, this test will cause a busy wait
1277 * during a higher-priority task.
1278 */
1279
1280static bool lapic_timer_int_injected(struct kvm_vcpu *vcpu)
1281{
1282 struct kvm_lapic *apic = vcpu->arch.apic;
1283 u32 reg = kvm_lapic_get_reg(apic, APIC_LVTT);
1284
1285 if (kvm_apic_hw_enabled(apic)) {
1286 int vec = reg & APIC_VECTOR_MASK;
1287 void *bitmap = apic->regs + APIC_ISR;
1288
1289 if (vcpu->arch.apicv_active)
1290 bitmap = apic->regs + APIC_IRR;
1291
1292 if (apic_test_vector(vec, bitmap))
1293 return true;
1294 }
1295 return false;
1296}
1297
1298void wait_lapic_expire(struct kvm_vcpu *vcpu)
1299{
1300 struct kvm_lapic *apic = vcpu->arch.apic;
1301 u64 guest_tsc, tsc_deadline;
1302
1303 if (!lapic_in_kernel(vcpu))
1304 return;
1305
1306 if (apic->lapic_timer.expired_tscdeadline == 0)
1307 return;
1308
1309 if (!lapic_timer_int_injected(vcpu))
1310 return;
1311
1312 tsc_deadline = apic->lapic_timer.expired_tscdeadline;
1313 apic->lapic_timer.expired_tscdeadline = 0;
1314 guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
1315 trace_kvm_wait_lapic_expire(vcpu->vcpu_id, guest_tsc - tsc_deadline);
1316
1317 /* __delay is delay_tsc whenever the hardware has TSC, thus always. */
1318 if (guest_tsc < tsc_deadline)
1319 __delay(min(tsc_deadline - guest_tsc,
1320 nsec_to_cycles(vcpu, lapic_timer_advance_ns)));
1321}
1322
1323static void start_sw_tscdeadline(struct kvm_lapic *apic)
1324{
1325 u64 guest_tsc, tscdeadline = apic->lapic_timer.tscdeadline;
1326 u64 ns = 0;
1327 ktime_t expire;
1328 struct kvm_vcpu *vcpu = apic->vcpu;
1329 unsigned long this_tsc_khz = vcpu->arch.virtual_tsc_khz;
1330 unsigned long flags;
1331 ktime_t now;
1332
1333 if (unlikely(!tscdeadline || !this_tsc_khz))
1334 return;
1335
1336 local_irq_save(flags);
1337
1338 now = ktime_get();
1339 guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
1340 if (likely(tscdeadline > guest_tsc)) {
1341 ns = (tscdeadline - guest_tsc) * 1000000ULL;
1342 do_div(ns, this_tsc_khz);
1343 expire = ktime_add_ns(now, ns);
1344 expire = ktime_sub_ns(expire, lapic_timer_advance_ns);
1345 hrtimer_start(&apic->lapic_timer.timer,
1346 expire, HRTIMER_MODE_ABS_PINNED);
1347 } else
1348 apic_timer_expired(apic);
1349
1350 local_irq_restore(flags);
1351}
1352
1353static void start_sw_period(struct kvm_lapic *apic)
1354{
1355 if (!apic->lapic_timer.period)
1356 return;
1357
1358 if (apic_lvtt_oneshot(apic) &&
1359 ktime_after(ktime_get(),
1360 apic->lapic_timer.target_expiration)) {
1361 apic_timer_expired(apic);
1362 return;
1363 }
1364
1365 hrtimer_start(&apic->lapic_timer.timer,
1366 apic->lapic_timer.target_expiration,
1367 HRTIMER_MODE_ABS_PINNED);
1368}
1369
1370static bool set_target_expiration(struct kvm_lapic *apic)
1371{
1372 ktime_t now;
1373 u64 tscl = rdtsc();
1374
1375 now = ktime_get();
1376 apic->lapic_timer.period = (u64)kvm_lapic_get_reg(apic, APIC_TMICT)
1377 * APIC_BUS_CYCLE_NS * apic->divide_count;
1378
1379 if (!apic->lapic_timer.period)
1380 return false;
1381
1382 /*
1383 * Do not allow the guest to program periodic timers with small
1384 * interval, since the hrtimers are not throttled by the host
1385 * scheduler.
1386 */
1387 if (apic_lvtt_period(apic)) {
1388 s64 min_period = min_timer_period_us * 1000LL;
1389
1390 if (apic->lapic_timer.period < min_period) {
1391 pr_info_ratelimited(
1392 "kvm: vcpu %i: requested %lld ns "
1393 "lapic timer period limited to %lld ns\n",
1394 apic->vcpu->vcpu_id,
1395 apic->lapic_timer.period, min_period);
1396 apic->lapic_timer.period = min_period;
1397 }
1398 }
1399
1400 apic_debug("%s: bus cycle is %" PRId64 "ns, now 0x%016"
1401 PRIx64 ", "
1402 "timer initial count 0x%x, period %lldns, "
1403 "expire @ 0x%016" PRIx64 ".\n", __func__,
1404 APIC_BUS_CYCLE_NS, ktime_to_ns(now),
1405 kvm_lapic_get_reg(apic, APIC_TMICT),
1406 apic->lapic_timer.period,
1407 ktime_to_ns(ktime_add_ns(now,
1408 apic->lapic_timer.period)));
1409
1410 apic->lapic_timer.tscdeadline = kvm_read_l1_tsc(apic->vcpu, tscl) +
1411 nsec_to_cycles(apic->vcpu, apic->lapic_timer.period);
1412 apic->lapic_timer.target_expiration = ktime_add_ns(now, apic->lapic_timer.period);
1413
1414 return true;
1415}
1416
1417static void advance_periodic_target_expiration(struct kvm_lapic *apic)
1418{
1419 apic->lapic_timer.tscdeadline +=
1420 nsec_to_cycles(apic->vcpu, apic->lapic_timer.period);
1421 apic->lapic_timer.target_expiration =
1422 ktime_add_ns(apic->lapic_timer.target_expiration,
1423 apic->lapic_timer.period);
1424}
1425
1426bool kvm_lapic_hv_timer_in_use(struct kvm_vcpu *vcpu)
1427{
1428 if (!lapic_in_kernel(vcpu))
1429 return false;
1430
1431 return vcpu->arch.apic->lapic_timer.hv_timer_in_use;
1432}
1433EXPORT_SYMBOL_GPL(kvm_lapic_hv_timer_in_use);
1434
1435static void cancel_hv_timer(struct kvm_lapic *apic)
1436{
1437 kvm_x86_ops->cancel_hv_timer(apic->vcpu);
1438 apic->lapic_timer.hv_timer_in_use = false;
1439}
1440
1441static bool start_hv_timer(struct kvm_lapic *apic)
1442{
1443 u64 tscdeadline = apic->lapic_timer.tscdeadline;
1444
1445 if ((atomic_read(&apic->lapic_timer.pending) &&
1446 !apic_lvtt_period(apic)) ||
1447 kvm_x86_ops->set_hv_timer(apic->vcpu, tscdeadline)) {
1448 if (apic->lapic_timer.hv_timer_in_use)
1449 cancel_hv_timer(apic);
1450 } else {
1451 apic->lapic_timer.hv_timer_in_use = true;
1452 hrtimer_cancel(&apic->lapic_timer.timer);
1453
1454 /* In case the sw timer triggered in the window */
1455 if (atomic_read(&apic->lapic_timer.pending) &&
1456 !apic_lvtt_period(apic))
1457 cancel_hv_timer(apic);
1458 }
1459 trace_kvm_hv_timer_state(apic->vcpu->vcpu_id,
1460 apic->lapic_timer.hv_timer_in_use);
1461 return apic->lapic_timer.hv_timer_in_use;
1462}
1463
1464void kvm_lapic_expired_hv_timer(struct kvm_vcpu *vcpu)
1465{
1466 struct kvm_lapic *apic = vcpu->arch.apic;
1467
1468 WARN_ON(!apic->lapic_timer.hv_timer_in_use);
1469 WARN_ON(swait_active(&vcpu->wq));
1470 cancel_hv_timer(apic);
1471 apic_timer_expired(apic);
1472
1473 if (apic_lvtt_period(apic) && apic->lapic_timer.period) {
1474 advance_periodic_target_expiration(apic);
1475 if (!start_hv_timer(apic))
1476 start_sw_period(apic);
1477 }
1478}
1479EXPORT_SYMBOL_GPL(kvm_lapic_expired_hv_timer);
1480
1481void kvm_lapic_switch_to_hv_timer(struct kvm_vcpu *vcpu)
1482{
1483 struct kvm_lapic *apic = vcpu->arch.apic;
1484
1485 WARN_ON(apic->lapic_timer.hv_timer_in_use);
1486
1487 start_hv_timer(apic);
1488}
1489EXPORT_SYMBOL_GPL(kvm_lapic_switch_to_hv_timer);
1490
1491void kvm_lapic_switch_to_sw_timer(struct kvm_vcpu *vcpu)
1492{
1493 struct kvm_lapic *apic = vcpu->arch.apic;
1494
1495 /* Possibly the TSC deadline timer is not enabled yet */
1496 if (!apic->lapic_timer.hv_timer_in_use)
1497 return;
1498
1499 cancel_hv_timer(apic);
1500
1501 if (atomic_read(&apic->lapic_timer.pending))
1502 return;
1503
1504 if (apic_lvtt_period(apic) || apic_lvtt_oneshot(apic))
1505 start_sw_period(apic);
1506 else if (apic_lvtt_tscdeadline(apic))
1507 start_sw_tscdeadline(apic);
1508}
1509EXPORT_SYMBOL_GPL(kvm_lapic_switch_to_sw_timer);
1510
1511static void start_apic_timer(struct kvm_lapic *apic)
1512{
1513 atomic_set(&apic->lapic_timer.pending, 0);
1514
1515 if (apic_lvtt_period(apic) || apic_lvtt_oneshot(apic)) {
1516 if (set_target_expiration(apic) &&
1517 !(kvm_x86_ops->set_hv_timer && start_hv_timer(apic)))
1518 start_sw_period(apic);
1519 } else if (apic_lvtt_tscdeadline(apic)) {
1520 if (!(kvm_x86_ops->set_hv_timer && start_hv_timer(apic)))
1521 start_sw_tscdeadline(apic);
1522 }
1523}
1524
1525static void apic_manage_nmi_watchdog(struct kvm_lapic *apic, u32 lvt0_val)
1526{
1527 bool lvt0_in_nmi_mode = apic_lvt_nmi_mode(lvt0_val);
1528
1529 if (apic->lvt0_in_nmi_mode != lvt0_in_nmi_mode) {
1530 apic->lvt0_in_nmi_mode = lvt0_in_nmi_mode;
1531 if (lvt0_in_nmi_mode) {
1532 apic_debug("Receive NMI setting on APIC_LVT0 "
1533 "for cpu %d\n", apic->vcpu->vcpu_id);
1534 atomic_inc(&apic->vcpu->kvm->arch.vapics_in_nmi_mode);
1535 } else
1536 atomic_dec(&apic->vcpu->kvm->arch.vapics_in_nmi_mode);
1537 }
1538}
1539
1540int kvm_lapic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val)
1541{
1542 int ret = 0;
1543
1544 trace_kvm_apic_write(reg, val);
1545
1546 switch (reg) {
1547 case APIC_ID: /* Local APIC ID */
1548 if (!apic_x2apic_mode(apic))
1549 kvm_apic_set_xapic_id(apic, val >> 24);
1550 else
1551 ret = 1;
1552 break;
1553
1554 case APIC_TASKPRI:
1555 report_tpr_access(apic, true);
1556 apic_set_tpr(apic, val & 0xff);
1557 break;
1558
1559 case APIC_EOI:
1560 apic_set_eoi(apic);
1561 break;
1562
1563 case APIC_LDR:
1564 if (!apic_x2apic_mode(apic))
1565 kvm_apic_set_ldr(apic, val & APIC_LDR_MASK);
1566 else
1567 ret = 1;
1568 break;
1569
1570 case APIC_DFR:
1571 if (!apic_x2apic_mode(apic)) {
1572 kvm_lapic_set_reg(apic, APIC_DFR, val | 0x0FFFFFFF);
1573 recalculate_apic_map(apic->vcpu->kvm);
1574 } else
1575 ret = 1;
1576 break;
1577
1578 case APIC_SPIV: {
1579 u32 mask = 0x3ff;
1580 if (kvm_lapic_get_reg(apic, APIC_LVR) & APIC_LVR_DIRECTED_EOI)
1581 mask |= APIC_SPIV_DIRECTED_EOI;
1582 apic_set_spiv(apic, val & mask);
1583 if (!(val & APIC_SPIV_APIC_ENABLED)) {
1584 int i;
1585 u32 lvt_val;
1586
1587 for (i = 0; i < KVM_APIC_LVT_NUM; i++) {
1588 lvt_val = kvm_lapic_get_reg(apic,
1589 APIC_LVTT + 0x10 * i);
1590 kvm_lapic_set_reg(apic, APIC_LVTT + 0x10 * i,
1591 lvt_val | APIC_LVT_MASKED);
1592 }
1593 apic_update_lvtt(apic);
1594 atomic_set(&apic->lapic_timer.pending, 0);
1595
1596 }
1597 break;
1598 }
1599 case APIC_ICR:
1600 /* No delay here, so we always clear the pending bit */
1601 kvm_lapic_set_reg(apic, APIC_ICR, val & ~(1 << 12));
1602 apic_send_ipi(apic);
1603 break;
1604
1605 case APIC_ICR2:
1606 if (!apic_x2apic_mode(apic))
1607 val &= 0xff000000;
1608 kvm_lapic_set_reg(apic, APIC_ICR2, val);
1609 break;
1610
1611 case APIC_LVT0:
1612 apic_manage_nmi_watchdog(apic, val);
1613 case APIC_LVTTHMR:
1614 case APIC_LVTPC:
1615 case APIC_LVT1:
1616 case APIC_LVTERR:
1617 /* TODO: Check vector */
1618 if (!kvm_apic_sw_enabled(apic))
1619 val |= APIC_LVT_MASKED;
1620
1621 val &= apic_lvt_mask[(reg - APIC_LVTT) >> 4];
1622 kvm_lapic_set_reg(apic, reg, val);
1623
1624 break;
1625
1626 case APIC_LVTT:
1627 if (!kvm_apic_sw_enabled(apic))
1628 val |= APIC_LVT_MASKED;
1629 val &= (apic_lvt_mask[0] | apic->lapic_timer.timer_mode_mask);
1630 kvm_lapic_set_reg(apic, APIC_LVTT, val);
1631 apic_update_lvtt(apic);
1632 break;
1633
1634 case APIC_TMICT:
1635 if (apic_lvtt_tscdeadline(apic))
1636 break;
1637
1638 hrtimer_cancel(&apic->lapic_timer.timer);
1639 kvm_lapic_set_reg(apic, APIC_TMICT, val);
1640 start_apic_timer(apic);
1641 break;
1642
1643 case APIC_TDCR:
1644 if (val & 4)
1645 apic_debug("KVM_WRITE:TDCR %x\n", val);
1646 kvm_lapic_set_reg(apic, APIC_TDCR, val);
1647 update_divide_count(apic);
1648 break;
1649
1650 case APIC_ESR:
1651 if (apic_x2apic_mode(apic) && val != 0) {
1652 apic_debug("KVM_WRITE:ESR not zero %x\n", val);
1653 ret = 1;
1654 }
1655 break;
1656
1657 case APIC_SELF_IPI:
1658 if (apic_x2apic_mode(apic)) {
1659 kvm_lapic_reg_write(apic, APIC_ICR, 0x40000 | (val & 0xff));
1660 } else
1661 ret = 1;
1662 break;
1663 default:
1664 ret = 1;
1665 break;
1666 }
1667 if (ret)
1668 apic_debug("Local APIC Write to read-only register %x\n", reg);
1669 return ret;
1670}
1671EXPORT_SYMBOL_GPL(kvm_lapic_reg_write);
1672
1673static int apic_mmio_write(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
1674 gpa_t address, int len, const void *data)
1675{
1676 struct kvm_lapic *apic = to_lapic(this);
1677 unsigned int offset = address - apic->base_address;
1678 u32 val;
1679
1680 if (!apic_mmio_in_range(apic, address))
1681 return -EOPNOTSUPP;
1682
1683 /*
1684 * APIC register must be aligned on 128-bits boundary.
1685 * 32/64/128 bits registers must be accessed thru 32 bits.
1686 * Refer SDM 8.4.1
1687 */
1688 if (len != 4 || (offset & 0xf)) {
1689 /* Don't shout loud, $infamous_os would cause only noise. */
1690 apic_debug("apic write: bad size=%d %lx\n", len, (long)address);
1691 return 0;
1692 }
1693
1694 val = *(u32*)data;
1695
1696 /* too common printing */
1697 if (offset != APIC_EOI)
1698 apic_debug("%s: offset 0x%x with length 0x%x, and value is "
1699 "0x%x\n", __func__, offset, len, val);
1700
1701 kvm_lapic_reg_write(apic, offset & 0xff0, val);
1702
1703 return 0;
1704}
1705
1706void kvm_lapic_set_eoi(struct kvm_vcpu *vcpu)
1707{
1708 kvm_lapic_reg_write(vcpu->arch.apic, APIC_EOI, 0);
1709}
1710EXPORT_SYMBOL_GPL(kvm_lapic_set_eoi);
1711
1712/* emulate APIC access in a trap manner */
1713void kvm_apic_write_nodecode(struct kvm_vcpu *vcpu, u32 offset)
1714{
1715 u32 val = 0;
1716
1717 /* hw has done the conditional check and inst decode */
1718 offset &= 0xff0;
1719
1720 kvm_lapic_reg_read(vcpu->arch.apic, offset, 4, &val);
1721
1722 /* TODO: optimize to just emulate side effect w/o one more write */
1723 kvm_lapic_reg_write(vcpu->arch.apic, offset, val);
1724}
1725EXPORT_SYMBOL_GPL(kvm_apic_write_nodecode);
1726
1727void kvm_free_lapic(struct kvm_vcpu *vcpu)
1728{
1729 struct kvm_lapic *apic = vcpu->arch.apic;
1730
1731 if (!vcpu->arch.apic)
1732 return;
1733
1734 hrtimer_cancel(&apic->lapic_timer.timer);
1735
1736 if (!(vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE))
1737 static_key_slow_dec_deferred(&apic_hw_disabled);
1738
1739 if (!apic->sw_enabled)
1740 static_key_slow_dec_deferred(&apic_sw_disabled);
1741
1742 if (apic->regs)
1743 free_page((unsigned long)apic->regs);
1744
1745 kfree(apic);
1746}
1747
1748/*
1749 *----------------------------------------------------------------------
1750 * LAPIC interface
1751 *----------------------------------------------------------------------
1752 */
1753u64 kvm_get_lapic_target_expiration_tsc(struct kvm_vcpu *vcpu)
1754{
1755 struct kvm_lapic *apic = vcpu->arch.apic;
1756
1757 if (!lapic_in_kernel(vcpu))
1758 return 0;
1759
1760 return apic->lapic_timer.tscdeadline;
1761}
1762
1763u64 kvm_get_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu)
1764{
1765 struct kvm_lapic *apic = vcpu->arch.apic;
1766
1767 if (!lapic_in_kernel(vcpu) ||
1768 !apic_lvtt_tscdeadline(apic))
1769 return 0;
1770
1771 return apic->lapic_timer.tscdeadline;
1772}
1773
1774void kvm_set_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu, u64 data)
1775{
1776 struct kvm_lapic *apic = vcpu->arch.apic;
1777
1778 if (!lapic_in_kernel(vcpu) || apic_lvtt_oneshot(apic) ||
1779 apic_lvtt_period(apic))
1780 return;
1781
1782 hrtimer_cancel(&apic->lapic_timer.timer);
1783 apic->lapic_timer.tscdeadline = data;
1784 start_apic_timer(apic);
1785}
1786
1787void kvm_lapic_set_tpr(struct kvm_vcpu *vcpu, unsigned long cr8)
1788{
1789 struct kvm_lapic *apic = vcpu->arch.apic;
1790
1791 apic_set_tpr(apic, ((cr8 & 0x0f) << 4)
1792 | (kvm_lapic_get_reg(apic, APIC_TASKPRI) & 4));
1793}
1794
1795u64 kvm_lapic_get_cr8(struct kvm_vcpu *vcpu)
1796{
1797 u64 tpr;
1798
1799 tpr = (u64) kvm_lapic_get_reg(vcpu->arch.apic, APIC_TASKPRI);
1800
1801 return (tpr & 0xf0) >> 4;
1802}
1803
1804void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
1805{
1806 u64 old_value = vcpu->arch.apic_base;
1807 struct kvm_lapic *apic = vcpu->arch.apic;
1808
1809 if (!apic)
1810 value |= MSR_IA32_APICBASE_BSP;
1811
1812 vcpu->arch.apic_base = value;
1813
1814 if ((old_value ^ value) & MSR_IA32_APICBASE_ENABLE)
1815 kvm_update_cpuid(vcpu);
1816
1817 if (!apic)
1818 return;
1819
1820 /* update jump label if enable bit changes */
1821 if ((old_value ^ value) & MSR_IA32_APICBASE_ENABLE) {
1822 if (value & MSR_IA32_APICBASE_ENABLE) {
1823 kvm_apic_set_xapic_id(apic, vcpu->vcpu_id);
1824 static_key_slow_dec_deferred(&apic_hw_disabled);
1825 } else {
1826 static_key_slow_inc(&apic_hw_disabled.key);
1827 recalculate_apic_map(vcpu->kvm);
1828 }
1829 }
1830
1831 if ((old_value ^ value) & X2APIC_ENABLE) {
1832 if (value & X2APIC_ENABLE) {
1833 kvm_apic_set_x2apic_id(apic, vcpu->vcpu_id);
1834 kvm_x86_ops->set_virtual_x2apic_mode(vcpu, true);
1835 } else
1836 kvm_x86_ops->set_virtual_x2apic_mode(vcpu, false);
1837 }
1838
1839 apic->base_address = apic->vcpu->arch.apic_base &
1840 MSR_IA32_APICBASE_BASE;
1841
1842 if ((value & MSR_IA32_APICBASE_ENABLE) &&
1843 apic->base_address != APIC_DEFAULT_PHYS_BASE)
1844 pr_warn_once("APIC base relocation is unsupported by KVM");
1845
1846 /* with FSB delivery interrupt, we can restart APIC functionality */
1847 apic_debug("apic base msr is 0x%016" PRIx64 ", and base address is "
1848 "0x%lx.\n", apic->vcpu->arch.apic_base, apic->base_address);
1849
1850}
1851
1852void kvm_lapic_reset(struct kvm_vcpu *vcpu, bool init_event)
1853{
1854 struct kvm_lapic *apic;
1855 int i;
1856
1857 apic_debug("%s\n", __func__);
1858
1859 ASSERT(vcpu);
1860 apic = vcpu->arch.apic;
1861 ASSERT(apic != NULL);
1862
1863 /* Stop the timer in case it's a reset to an active apic */
1864 hrtimer_cancel(&apic->lapic_timer.timer);
1865
1866 if (!init_event) {
1867 kvm_lapic_set_base(vcpu, APIC_DEFAULT_PHYS_BASE |
1868 MSR_IA32_APICBASE_ENABLE);
1869 kvm_apic_set_xapic_id(apic, vcpu->vcpu_id);
1870 }
1871 kvm_apic_set_version(apic->vcpu);
1872
1873 for (i = 0; i < KVM_APIC_LVT_NUM; i++)
1874 kvm_lapic_set_reg(apic, APIC_LVTT + 0x10 * i, APIC_LVT_MASKED);
1875 apic_update_lvtt(apic);
1876 if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_LINT0_REENABLED))
1877 kvm_lapic_set_reg(apic, APIC_LVT0,
1878 SET_APIC_DELIVERY_MODE(0, APIC_MODE_EXTINT));
1879 apic_manage_nmi_watchdog(apic, kvm_lapic_get_reg(apic, APIC_LVT0));
1880
1881 kvm_lapic_set_reg(apic, APIC_DFR, 0xffffffffU);
1882 apic_set_spiv(apic, 0xff);
1883 kvm_lapic_set_reg(apic, APIC_TASKPRI, 0);
1884 if (!apic_x2apic_mode(apic))
1885 kvm_apic_set_ldr(apic, 0);
1886 kvm_lapic_set_reg(apic, APIC_ESR, 0);
1887 kvm_lapic_set_reg(apic, APIC_ICR, 0);
1888 kvm_lapic_set_reg(apic, APIC_ICR2, 0);
1889 kvm_lapic_set_reg(apic, APIC_TDCR, 0);
1890 kvm_lapic_set_reg(apic, APIC_TMICT, 0);
1891 for (i = 0; i < 8; i++) {
1892 kvm_lapic_set_reg(apic, APIC_IRR + 0x10 * i, 0);
1893 kvm_lapic_set_reg(apic, APIC_ISR + 0x10 * i, 0);
1894 kvm_lapic_set_reg(apic, APIC_TMR + 0x10 * i, 0);
1895 }
1896 apic->irr_pending = vcpu->arch.apicv_active;
1897 apic->isr_count = vcpu->arch.apicv_active ? 1 : 0;
1898 apic->highest_isr_cache = -1;
1899 update_divide_count(apic);
1900 atomic_set(&apic->lapic_timer.pending, 0);
1901 if (kvm_vcpu_is_bsp(vcpu))
1902 kvm_lapic_set_base(vcpu,
1903 vcpu->arch.apic_base | MSR_IA32_APICBASE_BSP);
1904 vcpu->arch.pv_eoi.msr_val = 0;
1905 apic_update_ppr(apic);
1906
1907 vcpu->arch.apic_arb_prio = 0;
1908 vcpu->arch.apic_attention = 0;
1909
1910 apic_debug("%s: vcpu=%p, id=%d, base_msr="
1911 "0x%016" PRIx64 ", base_address=0x%0lx.\n", __func__,
1912 vcpu, kvm_apic_id(apic),
1913 vcpu->arch.apic_base, apic->base_address);
1914}
1915
1916/*
1917 *----------------------------------------------------------------------
1918 * timer interface
1919 *----------------------------------------------------------------------
1920 */
1921
1922static bool lapic_is_periodic(struct kvm_lapic *apic)
1923{
1924 return apic_lvtt_period(apic);
1925}
1926
1927int apic_has_pending_timer(struct kvm_vcpu *vcpu)
1928{
1929 struct kvm_lapic *apic = vcpu->arch.apic;
1930
1931 if (apic_enabled(apic) && apic_lvt_enabled(apic, APIC_LVTT))
1932 return atomic_read(&apic->lapic_timer.pending);
1933
1934 return 0;
1935}
1936
1937int kvm_apic_local_deliver(struct kvm_lapic *apic, int lvt_type)
1938{
1939 u32 reg = kvm_lapic_get_reg(apic, lvt_type);
1940 int vector, mode, trig_mode;
1941
1942 if (kvm_apic_hw_enabled(apic) && !(reg & APIC_LVT_MASKED)) {
1943 vector = reg & APIC_VECTOR_MASK;
1944 mode = reg & APIC_MODE_MASK;
1945 trig_mode = reg & APIC_LVT_LEVEL_TRIGGER;
1946 return __apic_accept_irq(apic, mode, vector, 1, trig_mode,
1947 NULL);
1948 }
1949 return 0;
1950}
1951
1952void kvm_apic_nmi_wd_deliver(struct kvm_vcpu *vcpu)
1953{
1954 struct kvm_lapic *apic = vcpu->arch.apic;
1955
1956 if (apic)
1957 kvm_apic_local_deliver(apic, APIC_LVT0);
1958}
1959
1960static const struct kvm_io_device_ops apic_mmio_ops = {
1961 .read = apic_mmio_read,
1962 .write = apic_mmio_write,
1963};
1964
1965static enum hrtimer_restart apic_timer_fn(struct hrtimer *data)
1966{
1967 struct kvm_timer *ktimer = container_of(data, struct kvm_timer, timer);
1968 struct kvm_lapic *apic = container_of(ktimer, struct kvm_lapic, lapic_timer);
1969
1970 apic_timer_expired(apic);
1971
1972 if (lapic_is_periodic(apic)) {
1973 advance_periodic_target_expiration(apic);
1974 hrtimer_add_expires_ns(&ktimer->timer, ktimer->period);
1975 return HRTIMER_RESTART;
1976 } else
1977 return HRTIMER_NORESTART;
1978}
1979
1980int kvm_create_lapic(struct kvm_vcpu *vcpu)
1981{
1982 struct kvm_lapic *apic;
1983
1984 ASSERT(vcpu != NULL);
1985 apic_debug("apic_init %d\n", vcpu->vcpu_id);
1986
1987 apic = kzalloc(sizeof(*apic), GFP_KERNEL);
1988 if (!apic)
1989 goto nomem;
1990
1991 vcpu->arch.apic = apic;
1992
1993 apic->regs = (void *)get_zeroed_page(GFP_KERNEL);
1994 if (!apic->regs) {
1995 printk(KERN_ERR "malloc apic regs error for vcpu %x\n",
1996 vcpu->vcpu_id);
1997 goto nomem_free_apic;
1998 }
1999 apic->vcpu = vcpu;
2000
2001 hrtimer_init(&apic->lapic_timer.timer, CLOCK_MONOTONIC,
2002 HRTIMER_MODE_ABS_PINNED);
2003 apic->lapic_timer.timer.function = apic_timer_fn;
2004
2005 /*
2006 * APIC is created enabled. This will prevent kvm_lapic_set_base from
2007 * thinking that APIC satet has changed.
2008 */
2009 vcpu->arch.apic_base = MSR_IA32_APICBASE_ENABLE;
2010 static_key_slow_inc(&apic_sw_disabled.key); /* sw disabled at reset */
2011 kvm_lapic_reset(vcpu, false);
2012 kvm_iodevice_init(&apic->dev, &apic_mmio_ops);
2013
2014 return 0;
2015nomem_free_apic:
2016 kfree(apic);
2017nomem:
2018 return -ENOMEM;
2019}
2020
2021int kvm_apic_has_interrupt(struct kvm_vcpu *vcpu)
2022{
2023 struct kvm_lapic *apic = vcpu->arch.apic;
2024 int highest_irr;
2025
2026 if (!apic_enabled(apic))
2027 return -1;
2028
2029 apic_update_ppr(apic);
2030 highest_irr = apic_find_highest_irr(apic);
2031 if ((highest_irr == -1) ||
2032 ((highest_irr & 0xF0) <= kvm_lapic_get_reg(apic, APIC_PROCPRI)))
2033 return -1;
2034 return highest_irr;
2035}
2036
2037int kvm_apic_accept_pic_intr(struct kvm_vcpu *vcpu)
2038{
2039 u32 lvt0 = kvm_lapic_get_reg(vcpu->arch.apic, APIC_LVT0);
2040 int r = 0;
2041
2042 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
2043 r = 1;
2044 if ((lvt0 & APIC_LVT_MASKED) == 0 &&
2045 GET_APIC_DELIVERY_MODE(lvt0) == APIC_MODE_EXTINT)
2046 r = 1;
2047 return r;
2048}
2049
2050void kvm_inject_apic_timer_irqs(struct kvm_vcpu *vcpu)
2051{
2052 struct kvm_lapic *apic = vcpu->arch.apic;
2053
2054 if (atomic_read(&apic->lapic_timer.pending) > 0) {
2055 kvm_apic_local_deliver(apic, APIC_LVTT);
2056 if (apic_lvtt_tscdeadline(apic))
2057 apic->lapic_timer.tscdeadline = 0;
2058 if (apic_lvtt_oneshot(apic)) {
2059 apic->lapic_timer.tscdeadline = 0;
2060 apic->lapic_timer.target_expiration = 0;
2061 }
2062 atomic_set(&apic->lapic_timer.pending, 0);
2063 }
2064}
2065
2066int kvm_get_apic_interrupt(struct kvm_vcpu *vcpu)
2067{
2068 int vector = kvm_apic_has_interrupt(vcpu);
2069 struct kvm_lapic *apic = vcpu->arch.apic;
2070
2071 if (vector == -1)
2072 return -1;
2073
2074 /*
2075 * We get here even with APIC virtualization enabled, if doing
2076 * nested virtualization and L1 runs with the "acknowledge interrupt
2077 * on exit" mode. Then we cannot inject the interrupt via RVI,
2078 * because the process would deliver it through the IDT.
2079 */
2080
2081 apic_set_isr(vector, apic);
2082 apic_update_ppr(apic);
2083 apic_clear_irr(vector, apic);
2084
2085 if (test_bit(vector, vcpu_to_synic(vcpu)->auto_eoi_bitmap)) {
2086 apic_clear_isr(vector, apic);
2087 apic_update_ppr(apic);
2088 }
2089
2090 return vector;
2091}
2092
2093static int kvm_apic_state_fixup(struct kvm_vcpu *vcpu,
2094 struct kvm_lapic_state *s, bool set)
2095{
2096 if (apic_x2apic_mode(vcpu->arch.apic)) {
2097 u32 *id = (u32 *)(s->regs + APIC_ID);
2098
2099 if (vcpu->kvm->arch.x2apic_format) {
2100 if (*id != vcpu->vcpu_id)
2101 return -EINVAL;
2102 } else {
2103 if (set)
2104 *id >>= 24;
2105 else
2106 *id <<= 24;
2107 }
2108 }
2109
2110 return 0;
2111}
2112
2113int kvm_apic_get_state(struct kvm_vcpu *vcpu, struct kvm_lapic_state *s)
2114{
2115 memcpy(s->regs, vcpu->arch.apic->regs, sizeof(*s));
2116 return kvm_apic_state_fixup(vcpu, s, false);
2117}
2118
2119int kvm_apic_set_state(struct kvm_vcpu *vcpu, struct kvm_lapic_state *s)
2120{
2121 struct kvm_lapic *apic = vcpu->arch.apic;
2122 int r;
2123
2124
2125 kvm_lapic_set_base(vcpu, vcpu->arch.apic_base);
2126 /* set SPIV separately to get count of SW disabled APICs right */
2127 apic_set_spiv(apic, *((u32 *)(s->regs + APIC_SPIV)));
2128
2129 r = kvm_apic_state_fixup(vcpu, s, true);
2130 if (r)
2131 return r;
2132 memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
2133
2134 recalculate_apic_map(vcpu->kvm);
2135 kvm_apic_set_version(vcpu);
2136
2137 apic_update_ppr(apic);
2138 hrtimer_cancel(&apic->lapic_timer.timer);
2139 apic_update_lvtt(apic);
2140 apic_manage_nmi_watchdog(apic, kvm_lapic_get_reg(apic, APIC_LVT0));
2141 update_divide_count(apic);
2142 start_apic_timer(apic);
2143 apic->irr_pending = true;
2144 apic->isr_count = vcpu->arch.apicv_active ?
2145 1 : count_vectors(apic->regs + APIC_ISR);
2146 apic->highest_isr_cache = -1;
2147 if (vcpu->arch.apicv_active) {
2148 if (kvm_x86_ops->apicv_post_state_restore)
2149 kvm_x86_ops->apicv_post_state_restore(vcpu);
2150 kvm_x86_ops->hwapic_irr_update(vcpu,
2151 apic_find_highest_irr(apic));
2152 kvm_x86_ops->hwapic_isr_update(vcpu,
2153 apic_find_highest_isr(apic));
2154 }
2155 kvm_make_request(KVM_REQ_EVENT, vcpu);
2156 if (ioapic_in_kernel(vcpu->kvm))
2157 kvm_rtc_eoi_tracking_restore_one(vcpu);
2158
2159 vcpu->arch.apic_arb_prio = 0;
2160
2161 return 0;
2162}
2163
2164void __kvm_migrate_apic_timer(struct kvm_vcpu *vcpu)
2165{
2166 struct hrtimer *timer;
2167
2168 if (!lapic_in_kernel(vcpu))
2169 return;
2170
2171 timer = &vcpu->arch.apic->lapic_timer.timer;
2172 if (hrtimer_cancel(timer))
2173 hrtimer_start_expires(timer, HRTIMER_MODE_ABS_PINNED);
2174}
2175
2176/*
2177 * apic_sync_pv_eoi_from_guest - called on vmexit or cancel interrupt
2178 *
2179 * Detect whether guest triggered PV EOI since the
2180 * last entry. If yes, set EOI on guests's behalf.
2181 * Clear PV EOI in guest memory in any case.
2182 */
2183static void apic_sync_pv_eoi_from_guest(struct kvm_vcpu *vcpu,
2184 struct kvm_lapic *apic)
2185{
2186 bool pending;
2187 int vector;
2188 /*
2189 * PV EOI state is derived from KVM_APIC_PV_EOI_PENDING in host
2190 * and KVM_PV_EOI_ENABLED in guest memory as follows:
2191 *
2192 * KVM_APIC_PV_EOI_PENDING is unset:
2193 * -> host disabled PV EOI.
2194 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is set:
2195 * -> host enabled PV EOI, guest did not execute EOI yet.
2196 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is unset:
2197 * -> host enabled PV EOI, guest executed EOI.
2198 */
2199 BUG_ON(!pv_eoi_enabled(vcpu));
2200 pending = pv_eoi_get_pending(vcpu);
2201 /*
2202 * Clear pending bit in any case: it will be set again on vmentry.
2203 * While this might not be ideal from performance point of view,
2204 * this makes sure pv eoi is only enabled when we know it's safe.
2205 */
2206 pv_eoi_clr_pending(vcpu);
2207 if (pending)
2208 return;
2209 vector = apic_set_eoi(apic);
2210 trace_kvm_pv_eoi(apic, vector);
2211}
2212
2213void kvm_lapic_sync_from_vapic(struct kvm_vcpu *vcpu)
2214{
2215 u32 data;
2216
2217 if (test_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention))
2218 apic_sync_pv_eoi_from_guest(vcpu, vcpu->arch.apic);
2219
2220 if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
2221 return;
2222
2223 if (kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
2224 sizeof(u32)))
2225 return;
2226
2227 apic_set_tpr(vcpu->arch.apic, data & 0xff);
2228}
2229
2230/*
2231 * apic_sync_pv_eoi_to_guest - called before vmentry
2232 *
2233 * Detect whether it's safe to enable PV EOI and
2234 * if yes do so.
2235 */
2236static void apic_sync_pv_eoi_to_guest(struct kvm_vcpu *vcpu,
2237 struct kvm_lapic *apic)
2238{
2239 if (!pv_eoi_enabled(vcpu) ||
2240 /* IRR set or many bits in ISR: could be nested. */
2241 apic->irr_pending ||
2242 /* Cache not set: could be safe but we don't bother. */
2243 apic->highest_isr_cache == -1 ||
2244 /* Need EOI to update ioapic. */
2245 kvm_ioapic_handles_vector(apic, apic->highest_isr_cache)) {
2246 /*
2247 * PV EOI was disabled by apic_sync_pv_eoi_from_guest
2248 * so we need not do anything here.
2249 */
2250 return;
2251 }
2252
2253 pv_eoi_set_pending(apic->vcpu);
2254}
2255
2256void kvm_lapic_sync_to_vapic(struct kvm_vcpu *vcpu)
2257{
2258 u32 data, tpr;
2259 int max_irr, max_isr;
2260 struct kvm_lapic *apic = vcpu->arch.apic;
2261
2262 apic_sync_pv_eoi_to_guest(vcpu, apic);
2263
2264 if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
2265 return;
2266
2267 tpr = kvm_lapic_get_reg(apic, APIC_TASKPRI) & 0xff;
2268 max_irr = apic_find_highest_irr(apic);
2269 if (max_irr < 0)
2270 max_irr = 0;
2271 max_isr = apic_find_highest_isr(apic);
2272 if (max_isr < 0)
2273 max_isr = 0;
2274 data = (tpr & 0xff) | ((max_isr & 0xf0) << 8) | (max_irr << 24);
2275
2276 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
2277 sizeof(u32));
2278}
2279
2280int kvm_lapic_set_vapic_addr(struct kvm_vcpu *vcpu, gpa_t vapic_addr)
2281{
2282 if (vapic_addr) {
2283 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2284 &vcpu->arch.apic->vapic_cache,
2285 vapic_addr, sizeof(u32)))
2286 return -EINVAL;
2287 __set_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
2288 } else {
2289 __clear_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
2290 }
2291
2292 vcpu->arch.apic->vapic_addr = vapic_addr;
2293 return 0;
2294}
2295
2296int kvm_x2apic_msr_write(struct kvm_vcpu *vcpu, u32 msr, u64 data)
2297{
2298 struct kvm_lapic *apic = vcpu->arch.apic;
2299 u32 reg = (msr - APIC_BASE_MSR) << 4;
2300
2301 if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(apic))
2302 return 1;
2303
2304 if (reg == APIC_ICR2)
2305 return 1;
2306
2307 /* if this is ICR write vector before command */
2308 if (reg == APIC_ICR)
2309 kvm_lapic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
2310 return kvm_lapic_reg_write(apic, reg, (u32)data);
2311}
2312
2313int kvm_x2apic_msr_read(struct kvm_vcpu *vcpu, u32 msr, u64 *data)
2314{
2315 struct kvm_lapic *apic = vcpu->arch.apic;
2316 u32 reg = (msr - APIC_BASE_MSR) << 4, low, high = 0;
2317
2318 if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(apic))
2319 return 1;
2320
2321 if (reg == APIC_DFR || reg == APIC_ICR2) {
2322 apic_debug("KVM_APIC_READ: read x2apic reserved register %x\n",
2323 reg);
2324 return 1;
2325 }
2326
2327 if (kvm_lapic_reg_read(apic, reg, 4, &low))
2328 return 1;
2329 if (reg == APIC_ICR)
2330 kvm_lapic_reg_read(apic, APIC_ICR2, 4, &high);
2331
2332 *data = (((u64)high) << 32) | low;
2333
2334 return 0;
2335}
2336
2337int kvm_hv_vapic_msr_write(struct kvm_vcpu *vcpu, u32 reg, u64 data)
2338{
2339 struct kvm_lapic *apic = vcpu->arch.apic;
2340
2341 if (!lapic_in_kernel(vcpu))
2342 return 1;
2343
2344 /* if this is ICR write vector before command */
2345 if (reg == APIC_ICR)
2346 kvm_lapic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
2347 return kvm_lapic_reg_write(apic, reg, (u32)data);
2348}
2349
2350int kvm_hv_vapic_msr_read(struct kvm_vcpu *vcpu, u32 reg, u64 *data)
2351{
2352 struct kvm_lapic *apic = vcpu->arch.apic;
2353 u32 low, high = 0;
2354
2355 if (!lapic_in_kernel(vcpu))
2356 return 1;
2357
2358 if (kvm_lapic_reg_read(apic, reg, 4, &low))
2359 return 1;
2360 if (reg == APIC_ICR)
2361 kvm_lapic_reg_read(apic, APIC_ICR2, 4, &high);
2362
2363 *data = (((u64)high) << 32) | low;
2364
2365 return 0;
2366}
2367
2368int kvm_lapic_enable_pv_eoi(struct kvm_vcpu *vcpu, u64 data)
2369{
2370 u64 addr = data & ~KVM_MSR_ENABLED;
2371 if (!IS_ALIGNED(addr, 4))
2372 return 1;
2373
2374 vcpu->arch.pv_eoi.msr_val = data;
2375 if (!pv_eoi_enabled(vcpu))
2376 return 0;
2377 return kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.pv_eoi.data,
2378 addr, sizeof(u8));
2379}
2380
2381void kvm_apic_accept_events(struct kvm_vcpu *vcpu)
2382{
2383 struct kvm_lapic *apic = vcpu->arch.apic;
2384 u8 sipi_vector;
2385 unsigned long pe;
2386
2387 if (!lapic_in_kernel(vcpu) || !apic->pending_events)
2388 return;
2389
2390 /*
2391 * INITs are latched while in SMM. Because an SMM CPU cannot
2392 * be in KVM_MP_STATE_INIT_RECEIVED state, just eat SIPIs
2393 * and delay processing of INIT until the next RSM.
2394 */
2395 if (is_smm(vcpu)) {
2396 WARN_ON_ONCE(vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED);
2397 if (test_bit(KVM_APIC_SIPI, &apic->pending_events))
2398 clear_bit(KVM_APIC_SIPI, &apic->pending_events);
2399 return;
2400 }
2401
2402 pe = xchg(&apic->pending_events, 0);
2403 if (test_bit(KVM_APIC_INIT, &pe)) {
2404 kvm_lapic_reset(vcpu, true);
2405 kvm_vcpu_reset(vcpu, true);
2406 if (kvm_vcpu_is_bsp(apic->vcpu))
2407 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
2408 else
2409 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
2410 }
2411 if (test_bit(KVM_APIC_SIPI, &pe) &&
2412 vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
2413 /* evaluate pending_events before reading the vector */
2414 smp_rmb();
2415 sipi_vector = apic->sipi_vector;
2416 apic_debug("vcpu %d received sipi with vector # %x\n",
2417 vcpu->vcpu_id, sipi_vector);
2418 kvm_vcpu_deliver_sipi_vector(vcpu, sipi_vector);
2419 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
2420 }
2421}
2422
2423void kvm_lapic_init(void)
2424{
2425 /* do not patch jump label more than once per second */
2426 jump_label_rate_limit(&apic_hw_disabled, HZ);
2427 jump_label_rate_limit(&apic_sw_disabled, HZ);
2428}
2429
2430void kvm_lapic_exit(void)
2431{
2432 static_key_deferred_flush(&apic_hw_disabled);
2433 static_key_deferred_flush(&apic_sw_disabled);
2434}