~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/arch/x86/kvm/lapic.c

Version: ~ [ linux-5.10-rc5 ] ~ [ linux-5.9.10 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.79 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.159 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.208 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.245 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.245 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.85 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 // SPDX-License-Identifier: GPL-2.0-only
  2 
  3 /*
  4  * Local APIC virtualization
  5  *
  6  * Copyright (C) 2006 Qumranet, Inc.
  7  * Copyright (C) 2007 Novell
  8  * Copyright (C) 2007 Intel
  9  * Copyright 2009 Red Hat, Inc. and/or its affiliates.
 10  *
 11  * Authors:
 12  *   Dor Laor <dor.laor@qumranet.com>
 13  *   Gregory Haskins <ghaskins@novell.com>
 14  *   Yaozu (Eddie) Dong <eddie.dong@intel.com>
 15  *
 16  * Based on Xen 3.1 code, Copyright (c) 2004, Intel Corporation.
 17  */
 18 
 19 #include <linux/kvm_host.h>
 20 #include <linux/kvm.h>
 21 #include <linux/mm.h>
 22 #include <linux/highmem.h>
 23 #include <linux/smp.h>
 24 #include <linux/hrtimer.h>
 25 #include <linux/io.h>
 26 #include <linux/export.h>
 27 #include <linux/math64.h>
 28 #include <linux/slab.h>
 29 #include <asm/processor.h>
 30 #include <asm/msr.h>
 31 #include <asm/page.h>
 32 #include <asm/current.h>
 33 #include <asm/apicdef.h>
 34 #include <asm/delay.h>
 35 #include <linux/atomic.h>
 36 #include <linux/jump_label.h>
 37 #include "kvm_cache_regs.h"
 38 #include "irq.h"
 39 #include "ioapic.h"
 40 #include "trace.h"
 41 #include "x86.h"
 42 #include "cpuid.h"
 43 #include "hyperv.h"
 44 
 45 #ifndef CONFIG_X86_64
 46 #define mod_64(x, y) ((x) - (y) * div64_u64(x, y))
 47 #else
 48 #define mod_64(x, y) ((x) % (y))
 49 #endif
 50 
 51 #define PRId64 "d"
 52 #define PRIx64 "llx"
 53 #define PRIu64 "u"
 54 #define PRIo64 "o"
 55 
 56 /* 14 is the version for Xeon and Pentium 8.4.8*/
 57 #define APIC_VERSION                    (0x14UL | ((KVM_APIC_LVT_NUM - 1) << 16))
 58 #define LAPIC_MMIO_LENGTH               (1 << 12)
 59 /* followed define is not in apicdef.h */
 60 #define MAX_APIC_VECTOR                 256
 61 #define APIC_VECTORS_PER_REG            32
 62 
 63 static bool lapic_timer_advance_dynamic __read_mostly;
 64 #define LAPIC_TIMER_ADVANCE_ADJUST_MIN  100     /* clock cycles */
 65 #define LAPIC_TIMER_ADVANCE_ADJUST_MAX  10000   /* clock cycles */
 66 #define LAPIC_TIMER_ADVANCE_NS_INIT     1000
 67 #define LAPIC_TIMER_ADVANCE_NS_MAX     5000
 68 /* step-by-step approximation to mitigate fluctuation */
 69 #define LAPIC_TIMER_ADVANCE_ADJUST_STEP 8
 70 
 71 static inline int apic_test_vector(int vec, void *bitmap)
 72 {
 73         return test_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
 74 }
 75 
 76 bool kvm_apic_pending_eoi(struct kvm_vcpu *vcpu, int vector)
 77 {
 78         struct kvm_lapic *apic = vcpu->arch.apic;
 79 
 80         return apic_test_vector(vector, apic->regs + APIC_ISR) ||
 81                 apic_test_vector(vector, apic->regs + APIC_IRR);
 82 }
 83 
 84 static inline int __apic_test_and_set_vector(int vec, void *bitmap)
 85 {
 86         return __test_and_set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
 87 }
 88 
 89 static inline int __apic_test_and_clear_vector(int vec, void *bitmap)
 90 {
 91         return __test_and_clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
 92 }
 93 
 94 struct static_key_deferred apic_hw_disabled __read_mostly;
 95 struct static_key_deferred apic_sw_disabled __read_mostly;
 96 
 97 static inline int apic_enabled(struct kvm_lapic *apic)
 98 {
 99         return kvm_apic_sw_enabled(apic) &&     kvm_apic_hw_enabled(apic);
100 }
101 
102 #define LVT_MASK        \
103         (APIC_LVT_MASKED | APIC_SEND_PENDING | APIC_VECTOR_MASK)
104 
105 #define LINT_MASK       \
106         (LVT_MASK | APIC_MODE_MASK | APIC_INPUT_POLARITY | \
107          APIC_LVT_REMOTE_IRR | APIC_LVT_LEVEL_TRIGGER)
108 
109 static inline u32 kvm_x2apic_id(struct kvm_lapic *apic)
110 {
111         return apic->vcpu->vcpu_id;
112 }
113 
114 static bool kvm_can_post_timer_interrupt(struct kvm_vcpu *vcpu)
115 {
116         return pi_inject_timer && kvm_vcpu_apicv_active(vcpu);
117 }
118 
119 bool kvm_can_use_hv_timer(struct kvm_vcpu *vcpu)
120 {
121         return kvm_x86_ops.set_hv_timer
122                && !(kvm_mwait_in_guest(vcpu->kvm) ||
123                     kvm_can_post_timer_interrupt(vcpu));
124 }
125 EXPORT_SYMBOL_GPL(kvm_can_use_hv_timer);
126 
127 static bool kvm_use_posted_timer_interrupt(struct kvm_vcpu *vcpu)
128 {
129         return kvm_can_post_timer_interrupt(vcpu) && vcpu->mode == IN_GUEST_MODE;
130 }
131 
132 static inline bool kvm_apic_map_get_logical_dest(struct kvm_apic_map *map,
133                 u32 dest_id, struct kvm_lapic ***cluster, u16 *mask) {
134         switch (map->mode) {
135         case KVM_APIC_MODE_X2APIC: {
136                 u32 offset = (dest_id >> 16) * 16;
137                 u32 max_apic_id = map->max_apic_id;
138 
139                 if (offset <= max_apic_id) {
140                         u8 cluster_size = min(max_apic_id - offset + 1, 16U);
141 
142                         offset = array_index_nospec(offset, map->max_apic_id + 1);
143                         *cluster = &map->phys_map[offset];
144                         *mask = dest_id & (0xffff >> (16 - cluster_size));
145                 } else {
146                         *mask = 0;
147                 }
148 
149                 return true;
150                 }
151         case KVM_APIC_MODE_XAPIC_FLAT:
152                 *cluster = map->xapic_flat_map;
153                 *mask = dest_id & 0xff;
154                 return true;
155         case KVM_APIC_MODE_XAPIC_CLUSTER:
156                 *cluster = map->xapic_cluster_map[(dest_id >> 4) & 0xf];
157                 *mask = dest_id & 0xf;
158                 return true;
159         default:
160                 /* Not optimized. */
161                 return false;
162         }
163 }
164 
165 static void kvm_apic_map_free(struct rcu_head *rcu)
166 {
167         struct kvm_apic_map *map = container_of(rcu, struct kvm_apic_map, rcu);
168 
169         kvfree(map);
170 }
171 
172 /*
173  * CLEAN -> DIRTY and UPDATE_IN_PROGRESS -> DIRTY changes happen without a lock.
174  *
175  * DIRTY -> UPDATE_IN_PROGRESS and UPDATE_IN_PROGRESS -> CLEAN happen with
176  * apic_map_lock_held.
177  */
178 enum {
179         CLEAN,
180         UPDATE_IN_PROGRESS,
181         DIRTY
182 };
183 
184 void kvm_recalculate_apic_map(struct kvm *kvm)
185 {
186         struct kvm_apic_map *new, *old = NULL;
187         struct kvm_vcpu *vcpu;
188         int i;
189         u32 max_id = 255; /* enough space for any xAPIC ID */
190 
191         /* Read kvm->arch.apic_map_dirty before kvm->arch.apic_map.  */
192         if (atomic_read_acquire(&kvm->arch.apic_map_dirty) == CLEAN)
193                 return;
194 
195         mutex_lock(&kvm->arch.apic_map_lock);
196         /*
197          * Read kvm->arch.apic_map_dirty before kvm->arch.apic_map
198          * (if clean) or the APIC registers (if dirty).
199          */
200         if (atomic_cmpxchg_acquire(&kvm->arch.apic_map_dirty,
201                                    DIRTY, UPDATE_IN_PROGRESS) == CLEAN) {
202                 /* Someone else has updated the map. */
203                 mutex_unlock(&kvm->arch.apic_map_lock);
204                 return;
205         }
206 
207         kvm_for_each_vcpu(i, vcpu, kvm)
208                 if (kvm_apic_present(vcpu))
209                         max_id = max(max_id, kvm_x2apic_id(vcpu->arch.apic));
210 
211         new = kvzalloc(sizeof(struct kvm_apic_map) +
212                            sizeof(struct kvm_lapic *) * ((u64)max_id + 1),
213                            GFP_KERNEL_ACCOUNT);
214 
215         if (!new)
216                 goto out;
217 
218         new->max_apic_id = max_id;
219 
220         kvm_for_each_vcpu(i, vcpu, kvm) {
221                 struct kvm_lapic *apic = vcpu->arch.apic;
222                 struct kvm_lapic **cluster;
223                 u16 mask;
224                 u32 ldr;
225                 u8 xapic_id;
226                 u32 x2apic_id;
227 
228                 if (!kvm_apic_present(vcpu))
229                         continue;
230 
231                 xapic_id = kvm_xapic_id(apic);
232                 x2apic_id = kvm_x2apic_id(apic);
233 
234                 /* Hotplug hack: see kvm_apic_match_physical_addr(), ... */
235                 if ((apic_x2apic_mode(apic) || x2apic_id > 0xff) &&
236                                 x2apic_id <= new->max_apic_id)
237                         new->phys_map[x2apic_id] = apic;
238                 /*
239                  * ... xAPIC ID of VCPUs with APIC ID > 0xff will wrap-around,
240                  * prevent them from masking VCPUs with APIC ID <= 0xff.
241                  */
242                 if (!apic_x2apic_mode(apic) && !new->phys_map[xapic_id])
243                         new->phys_map[xapic_id] = apic;
244 
245                 if (!kvm_apic_sw_enabled(apic))
246                         continue;
247 
248                 ldr = kvm_lapic_get_reg(apic, APIC_LDR);
249 
250                 if (apic_x2apic_mode(apic)) {
251                         new->mode |= KVM_APIC_MODE_X2APIC;
252                 } else if (ldr) {
253                         ldr = GET_APIC_LOGICAL_ID(ldr);
254                         if (kvm_lapic_get_reg(apic, APIC_DFR) == APIC_DFR_FLAT)
255                                 new->mode |= KVM_APIC_MODE_XAPIC_FLAT;
256                         else
257                                 new->mode |= KVM_APIC_MODE_XAPIC_CLUSTER;
258                 }
259 
260                 if (!kvm_apic_map_get_logical_dest(new, ldr, &cluster, &mask))
261                         continue;
262 
263                 if (mask)
264                         cluster[ffs(mask) - 1] = apic;
265         }
266 out:
267         old = rcu_dereference_protected(kvm->arch.apic_map,
268                         lockdep_is_held(&kvm->arch.apic_map_lock));
269         rcu_assign_pointer(kvm->arch.apic_map, new);
270         /*
271          * Write kvm->arch.apic_map before clearing apic->apic_map_dirty.
272          * If another update has come in, leave it DIRTY.
273          */
274         atomic_cmpxchg_release(&kvm->arch.apic_map_dirty,
275                                UPDATE_IN_PROGRESS, CLEAN);
276         mutex_unlock(&kvm->arch.apic_map_lock);
277 
278         if (old)
279                 call_rcu(&old->rcu, kvm_apic_map_free);
280 
281         kvm_make_scan_ioapic_request(kvm);
282 }
283 
284 static inline void apic_set_spiv(struct kvm_lapic *apic, u32 val)
285 {
286         bool enabled = val & APIC_SPIV_APIC_ENABLED;
287 
288         kvm_lapic_set_reg(apic, APIC_SPIV, val);
289 
290         if (enabled != apic->sw_enabled) {
291                 apic->sw_enabled = enabled;
292                 if (enabled)
293                         static_key_slow_dec_deferred(&apic_sw_disabled);
294                 else
295                         static_key_slow_inc(&apic_sw_disabled.key);
296 
297                 atomic_set_release(&apic->vcpu->kvm->arch.apic_map_dirty, DIRTY);
298         }
299 }
300 
301 static inline void kvm_apic_set_xapic_id(struct kvm_lapic *apic, u8 id)
302 {
303         kvm_lapic_set_reg(apic, APIC_ID, id << 24);
304         atomic_set_release(&apic->vcpu->kvm->arch.apic_map_dirty, DIRTY);
305 }
306 
307 static inline void kvm_apic_set_ldr(struct kvm_lapic *apic, u32 id)
308 {
309         kvm_lapic_set_reg(apic, APIC_LDR, id);
310         atomic_set_release(&apic->vcpu->kvm->arch.apic_map_dirty, DIRTY);
311 }
312 
313 static inline void kvm_apic_set_dfr(struct kvm_lapic *apic, u32 val)
314 {
315         kvm_lapic_set_reg(apic, APIC_DFR, val);
316         atomic_set_release(&apic->vcpu->kvm->arch.apic_map_dirty, DIRTY);
317 }
318 
319 static inline u32 kvm_apic_calc_x2apic_ldr(u32 id)
320 {
321         return ((id >> 4) << 16) | (1 << (id & 0xf));
322 }
323 
324 static inline void kvm_apic_set_x2apic_id(struct kvm_lapic *apic, u32 id)
325 {
326         u32 ldr = kvm_apic_calc_x2apic_ldr(id);
327 
328         WARN_ON_ONCE(id != apic->vcpu->vcpu_id);
329 
330         kvm_lapic_set_reg(apic, APIC_ID, id);
331         kvm_lapic_set_reg(apic, APIC_LDR, ldr);
332         atomic_set_release(&apic->vcpu->kvm->arch.apic_map_dirty, DIRTY);
333 }
334 
335 static inline int apic_lvt_enabled(struct kvm_lapic *apic, int lvt_type)
336 {
337         return !(kvm_lapic_get_reg(apic, lvt_type) & APIC_LVT_MASKED);
338 }
339 
340 static inline int apic_lvtt_oneshot(struct kvm_lapic *apic)
341 {
342         return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_ONESHOT;
343 }
344 
345 static inline int apic_lvtt_period(struct kvm_lapic *apic)
346 {
347         return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_PERIODIC;
348 }
349 
350 static inline int apic_lvtt_tscdeadline(struct kvm_lapic *apic)
351 {
352         return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_TSCDEADLINE;
353 }
354 
355 static inline int apic_lvt_nmi_mode(u32 lvt_val)
356 {
357         return (lvt_val & (APIC_MODE_MASK | APIC_LVT_MASKED)) == APIC_DM_NMI;
358 }
359 
360 void kvm_apic_set_version(struct kvm_vcpu *vcpu)
361 {
362         struct kvm_lapic *apic = vcpu->arch.apic;
363         u32 v = APIC_VERSION;
364 
365         if (!lapic_in_kernel(vcpu))
366                 return;
367 
368         /*
369          * KVM emulates 82093AA datasheet (with in-kernel IOAPIC implementation)
370          * which doesn't have EOI register; Some buggy OSes (e.g. Windows with
371          * Hyper-V role) disable EOI broadcast in lapic not checking for IOAPIC
372          * version first and level-triggered interrupts never get EOIed in
373          * IOAPIC.
374          */
375         if (guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) &&
376             !ioapic_in_kernel(vcpu->kvm))
377                 v |= APIC_LVR_DIRECTED_EOI;
378         kvm_lapic_set_reg(apic, APIC_LVR, v);
379 }
380 
381 static const unsigned int apic_lvt_mask[KVM_APIC_LVT_NUM] = {
382         LVT_MASK ,      /* part LVTT mask, timer mode mask added at runtime */
383         LVT_MASK | APIC_MODE_MASK,      /* LVTTHMR */
384         LVT_MASK | APIC_MODE_MASK,      /* LVTPC */
385         LINT_MASK, LINT_MASK,   /* LVT0-1 */
386         LVT_MASK                /* LVTERR */
387 };
388 
389 static int find_highest_vector(void *bitmap)
390 {
391         int vec;
392         u32 *reg;
393 
394         for (vec = MAX_APIC_VECTOR - APIC_VECTORS_PER_REG;
395              vec >= 0; vec -= APIC_VECTORS_PER_REG) {
396                 reg = bitmap + REG_POS(vec);
397                 if (*reg)
398                         return __fls(*reg) + vec;
399         }
400 
401         return -1;
402 }
403 
404 static u8 count_vectors(void *bitmap)
405 {
406         int vec;
407         u32 *reg;
408         u8 count = 0;
409 
410         for (vec = 0; vec < MAX_APIC_VECTOR; vec += APIC_VECTORS_PER_REG) {
411                 reg = bitmap + REG_POS(vec);
412                 count += hweight32(*reg);
413         }
414 
415         return count;
416 }
417 
418 bool __kvm_apic_update_irr(u32 *pir, void *regs, int *max_irr)
419 {
420         u32 i, vec;
421         u32 pir_val, irr_val, prev_irr_val;
422         int max_updated_irr;
423 
424         max_updated_irr = -1;
425         *max_irr = -1;
426 
427         for (i = vec = 0; i <= 7; i++, vec += 32) {
428                 pir_val = READ_ONCE(pir[i]);
429                 irr_val = *((u32 *)(regs + APIC_IRR + i * 0x10));
430                 if (pir_val) {
431                         prev_irr_val = irr_val;
432                         irr_val |= xchg(&pir[i], 0);
433                         *((u32 *)(regs + APIC_IRR + i * 0x10)) = irr_val;
434                         if (prev_irr_val != irr_val) {
435                                 max_updated_irr =
436                                         __fls(irr_val ^ prev_irr_val) + vec;
437                         }
438                 }
439                 if (irr_val)
440                         *max_irr = __fls(irr_val) + vec;
441         }
442 
443         return ((max_updated_irr != -1) &&
444                 (max_updated_irr == *max_irr));
445 }
446 EXPORT_SYMBOL_GPL(__kvm_apic_update_irr);
447 
448 bool kvm_apic_update_irr(struct kvm_vcpu *vcpu, u32 *pir, int *max_irr)
449 {
450         struct kvm_lapic *apic = vcpu->arch.apic;
451 
452         return __kvm_apic_update_irr(pir, apic->regs, max_irr);
453 }
454 EXPORT_SYMBOL_GPL(kvm_apic_update_irr);
455 
456 static inline int apic_search_irr(struct kvm_lapic *apic)
457 {
458         return find_highest_vector(apic->regs + APIC_IRR);
459 }
460 
461 static inline int apic_find_highest_irr(struct kvm_lapic *apic)
462 {
463         int result;
464 
465         /*
466          * Note that irr_pending is just a hint. It will be always
467          * true with virtual interrupt delivery enabled.
468          */
469         if (!apic->irr_pending)
470                 return -1;
471 
472         result = apic_search_irr(apic);
473         ASSERT(result == -1 || result >= 16);
474 
475         return result;
476 }
477 
478 static inline void apic_clear_irr(int vec, struct kvm_lapic *apic)
479 {
480         struct kvm_vcpu *vcpu;
481 
482         vcpu = apic->vcpu;
483 
484         if (unlikely(vcpu->arch.apicv_active)) {
485                 /* need to update RVI */
486                 kvm_lapic_clear_vector(vec, apic->regs + APIC_IRR);
487                 kvm_x86_ops.hwapic_irr_update(vcpu,
488                                 apic_find_highest_irr(apic));
489         } else {
490                 apic->irr_pending = false;
491                 kvm_lapic_clear_vector(vec, apic->regs + APIC_IRR);
492                 if (apic_search_irr(apic) != -1)
493                         apic->irr_pending = true;
494         }
495 }
496 
497 void kvm_apic_clear_irr(struct kvm_vcpu *vcpu, int vec)
498 {
499         apic_clear_irr(vec, vcpu->arch.apic);
500 }
501 EXPORT_SYMBOL_GPL(kvm_apic_clear_irr);
502 
503 static inline void apic_set_isr(int vec, struct kvm_lapic *apic)
504 {
505         struct kvm_vcpu *vcpu;
506 
507         if (__apic_test_and_set_vector(vec, apic->regs + APIC_ISR))
508                 return;
509 
510         vcpu = apic->vcpu;
511 
512         /*
513          * With APIC virtualization enabled, all caching is disabled
514          * because the processor can modify ISR under the hood.  Instead
515          * just set SVI.
516          */
517         if (unlikely(vcpu->arch.apicv_active))
518                 kvm_x86_ops.hwapic_isr_update(vcpu, vec);
519         else {
520                 ++apic->isr_count;
521                 BUG_ON(apic->isr_count > MAX_APIC_VECTOR);
522                 /*
523                  * ISR (in service register) bit is set when injecting an interrupt.
524                  * The highest vector is injected. Thus the latest bit set matches
525                  * the highest bit in ISR.
526                  */
527                 apic->highest_isr_cache = vec;
528         }
529 }
530 
531 static inline int apic_find_highest_isr(struct kvm_lapic *apic)
532 {
533         int result;
534 
535         /*
536          * Note that isr_count is always 1, and highest_isr_cache
537          * is always -1, with APIC virtualization enabled.
538          */
539         if (!apic->isr_count)
540                 return -1;
541         if (likely(apic->highest_isr_cache != -1))
542                 return apic->highest_isr_cache;
543 
544         result = find_highest_vector(apic->regs + APIC_ISR);
545         ASSERT(result == -1 || result >= 16);
546 
547         return result;
548 }
549 
550 static inline void apic_clear_isr(int vec, struct kvm_lapic *apic)
551 {
552         struct kvm_vcpu *vcpu;
553         if (!__apic_test_and_clear_vector(vec, apic->regs + APIC_ISR))
554                 return;
555 
556         vcpu = apic->vcpu;
557 
558         /*
559          * We do get here for APIC virtualization enabled if the guest
560          * uses the Hyper-V APIC enlightenment.  In this case we may need
561          * to trigger a new interrupt delivery by writing the SVI field;
562          * on the other hand isr_count and highest_isr_cache are unused
563          * and must be left alone.
564          */
565         if (unlikely(vcpu->arch.apicv_active))
566                 kvm_x86_ops.hwapic_isr_update(vcpu,
567                                                apic_find_highest_isr(apic));
568         else {
569                 --apic->isr_count;
570                 BUG_ON(apic->isr_count < 0);
571                 apic->highest_isr_cache = -1;
572         }
573 }
574 
575 int kvm_lapic_find_highest_irr(struct kvm_vcpu *vcpu)
576 {
577         /* This may race with setting of irr in __apic_accept_irq() and
578          * value returned may be wrong, but kvm_vcpu_kick() in __apic_accept_irq
579          * will cause vmexit immediately and the value will be recalculated
580          * on the next vmentry.
581          */
582         return apic_find_highest_irr(vcpu->arch.apic);
583 }
584 EXPORT_SYMBOL_GPL(kvm_lapic_find_highest_irr);
585 
586 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
587                              int vector, int level, int trig_mode,
588                              struct dest_map *dest_map);
589 
590 int kvm_apic_set_irq(struct kvm_vcpu *vcpu, struct kvm_lapic_irq *irq,
591                      struct dest_map *dest_map)
592 {
593         struct kvm_lapic *apic = vcpu->arch.apic;
594 
595         return __apic_accept_irq(apic, irq->delivery_mode, irq->vector,
596                         irq->level, irq->trig_mode, dest_map);
597 }
598 
599 static int __pv_send_ipi(unsigned long *ipi_bitmap, struct kvm_apic_map *map,
600                          struct kvm_lapic_irq *irq, u32 min)
601 {
602         int i, count = 0;
603         struct kvm_vcpu *vcpu;
604 
605         if (min > map->max_apic_id)
606                 return 0;
607 
608         for_each_set_bit(i, ipi_bitmap,
609                 min((u32)BITS_PER_LONG, (map->max_apic_id - min + 1))) {
610                 if (map->phys_map[min + i]) {
611                         vcpu = map->phys_map[min + i]->vcpu;
612                         count += kvm_apic_set_irq(vcpu, irq, NULL);
613                 }
614         }
615 
616         return count;
617 }
618 
619 int kvm_pv_send_ipi(struct kvm *kvm, unsigned long ipi_bitmap_low,
620                     unsigned long ipi_bitmap_high, u32 min,
621                     unsigned long icr, int op_64_bit)
622 {
623         struct kvm_apic_map *map;
624         struct kvm_lapic_irq irq = {0};
625         int cluster_size = op_64_bit ? 64 : 32;
626         int count;
627 
628         if (icr & (APIC_DEST_MASK | APIC_SHORT_MASK))
629                 return -KVM_EINVAL;
630 
631         irq.vector = icr & APIC_VECTOR_MASK;
632         irq.delivery_mode = icr & APIC_MODE_MASK;
633         irq.level = (icr & APIC_INT_ASSERT) != 0;
634         irq.trig_mode = icr & APIC_INT_LEVELTRIG;
635 
636         rcu_read_lock();
637         map = rcu_dereference(kvm->arch.apic_map);
638 
639         count = -EOPNOTSUPP;
640         if (likely(map)) {
641                 count = __pv_send_ipi(&ipi_bitmap_low, map, &irq, min);
642                 min += cluster_size;
643                 count += __pv_send_ipi(&ipi_bitmap_high, map, &irq, min);
644         }
645 
646         rcu_read_unlock();
647         return count;
648 }
649 
650 static int pv_eoi_put_user(struct kvm_vcpu *vcpu, u8 val)
651 {
652 
653         return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, &val,
654                                       sizeof(val));
655 }
656 
657 static int pv_eoi_get_user(struct kvm_vcpu *vcpu, u8 *val)
658 {
659 
660         return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, val,
661                                       sizeof(*val));
662 }
663 
664 static inline bool pv_eoi_enabled(struct kvm_vcpu *vcpu)
665 {
666         return vcpu->arch.pv_eoi.msr_val & KVM_MSR_ENABLED;
667 }
668 
669 static bool pv_eoi_get_pending(struct kvm_vcpu *vcpu)
670 {
671         u8 val;
672         if (pv_eoi_get_user(vcpu, &val) < 0) {
673                 printk(KERN_WARNING "Can't read EOI MSR value: 0x%llx\n",
674                            (unsigned long long)vcpu->arch.pv_eoi.msr_val);
675                 return false;
676         }
677         return val & 0x1;
678 }
679 
680 static void pv_eoi_set_pending(struct kvm_vcpu *vcpu)
681 {
682         if (pv_eoi_put_user(vcpu, KVM_PV_EOI_ENABLED) < 0) {
683                 printk(KERN_WARNING "Can't set EOI MSR value: 0x%llx\n",
684                            (unsigned long long)vcpu->arch.pv_eoi.msr_val);
685                 return;
686         }
687         __set_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
688 }
689 
690 static void pv_eoi_clr_pending(struct kvm_vcpu *vcpu)
691 {
692         if (pv_eoi_put_user(vcpu, KVM_PV_EOI_DISABLED) < 0) {
693                 printk(KERN_WARNING "Can't clear EOI MSR value: 0x%llx\n",
694                            (unsigned long long)vcpu->arch.pv_eoi.msr_val);
695                 return;
696         }
697         __clear_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
698 }
699 
700 static int apic_has_interrupt_for_ppr(struct kvm_lapic *apic, u32 ppr)
701 {
702         int highest_irr;
703         if (apic->vcpu->arch.apicv_active)
704                 highest_irr = kvm_x86_ops.sync_pir_to_irr(apic->vcpu);
705         else
706                 highest_irr = apic_find_highest_irr(apic);
707         if (highest_irr == -1 || (highest_irr & 0xF0) <= ppr)
708                 return -1;
709         return highest_irr;
710 }
711 
712 static bool __apic_update_ppr(struct kvm_lapic *apic, u32 *new_ppr)
713 {
714         u32 tpr, isrv, ppr, old_ppr;
715         int isr;
716 
717         old_ppr = kvm_lapic_get_reg(apic, APIC_PROCPRI);
718         tpr = kvm_lapic_get_reg(apic, APIC_TASKPRI);
719         isr = apic_find_highest_isr(apic);
720         isrv = (isr != -1) ? isr : 0;
721 
722         if ((tpr & 0xf0) >= (isrv & 0xf0))
723                 ppr = tpr & 0xff;
724         else
725                 ppr = isrv & 0xf0;
726 
727         *new_ppr = ppr;
728         if (old_ppr != ppr)
729                 kvm_lapic_set_reg(apic, APIC_PROCPRI, ppr);
730 
731         return ppr < old_ppr;
732 }
733 
734 static void apic_update_ppr(struct kvm_lapic *apic)
735 {
736         u32 ppr;
737 
738         if (__apic_update_ppr(apic, &ppr) &&
739             apic_has_interrupt_for_ppr(apic, ppr) != -1)
740                 kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
741 }
742 
743 void kvm_apic_update_ppr(struct kvm_vcpu *vcpu)
744 {
745         apic_update_ppr(vcpu->arch.apic);
746 }
747 EXPORT_SYMBOL_GPL(kvm_apic_update_ppr);
748 
749 static void apic_set_tpr(struct kvm_lapic *apic, u32 tpr)
750 {
751         kvm_lapic_set_reg(apic, APIC_TASKPRI, tpr);
752         apic_update_ppr(apic);
753 }
754 
755 static bool kvm_apic_broadcast(struct kvm_lapic *apic, u32 mda)
756 {
757         return mda == (apic_x2apic_mode(apic) ?
758                         X2APIC_BROADCAST : APIC_BROADCAST);
759 }
760 
761 static bool kvm_apic_match_physical_addr(struct kvm_lapic *apic, u32 mda)
762 {
763         if (kvm_apic_broadcast(apic, mda))
764                 return true;
765 
766         if (apic_x2apic_mode(apic))
767                 return mda == kvm_x2apic_id(apic);
768 
769         /*
770          * Hotplug hack: Make LAPIC in xAPIC mode also accept interrupts as if
771          * it were in x2APIC mode.  Hotplugged VCPUs start in xAPIC mode and
772          * this allows unique addressing of VCPUs with APIC ID over 0xff.
773          * The 0xff condition is needed because writeable xAPIC ID.
774          */
775         if (kvm_x2apic_id(apic) > 0xff && mda == kvm_x2apic_id(apic))
776                 return true;
777 
778         return mda == kvm_xapic_id(apic);
779 }
780 
781 static bool kvm_apic_match_logical_addr(struct kvm_lapic *apic, u32 mda)
782 {
783         u32 logical_id;
784 
785         if (kvm_apic_broadcast(apic, mda))
786                 return true;
787 
788         logical_id = kvm_lapic_get_reg(apic, APIC_LDR);
789 
790         if (apic_x2apic_mode(apic))
791                 return ((logical_id >> 16) == (mda >> 16))
792                        && (logical_id & mda & 0xffff) != 0;
793 
794         logical_id = GET_APIC_LOGICAL_ID(logical_id);
795 
796         switch (kvm_lapic_get_reg(apic, APIC_DFR)) {
797         case APIC_DFR_FLAT:
798                 return (logical_id & mda) != 0;
799         case APIC_DFR_CLUSTER:
800                 return ((logical_id >> 4) == (mda >> 4))
801                        && (logical_id & mda & 0xf) != 0;
802         default:
803                 return false;
804         }
805 }
806 
807 /* The KVM local APIC implementation has two quirks:
808  *
809  *  - Real hardware delivers interrupts destined to x2APIC ID > 0xff to LAPICs
810  *    in xAPIC mode if the "destination & 0xff" matches its xAPIC ID.
811  *    KVM doesn't do that aliasing.
812  *
813  *  - in-kernel IOAPIC messages have to be delivered directly to
814  *    x2APIC, because the kernel does not support interrupt remapping.
815  *    In order to support broadcast without interrupt remapping, x2APIC
816  *    rewrites the destination of non-IPI messages from APIC_BROADCAST
817  *    to X2APIC_BROADCAST.
818  *
819  * The broadcast quirk can be disabled with KVM_CAP_X2APIC_API.  This is
820  * important when userspace wants to use x2APIC-format MSIs, because
821  * APIC_BROADCAST (0xff) is a legal route for "cluster 0, CPUs 0-7".
822  */
823 static u32 kvm_apic_mda(struct kvm_vcpu *vcpu, unsigned int dest_id,
824                 struct kvm_lapic *source, struct kvm_lapic *target)
825 {
826         bool ipi = source != NULL;
827 
828         if (!vcpu->kvm->arch.x2apic_broadcast_quirk_disabled &&
829             !ipi && dest_id == APIC_BROADCAST && apic_x2apic_mode(target))
830                 return X2APIC_BROADCAST;
831 
832         return dest_id;
833 }
834 
835 bool kvm_apic_match_dest(struct kvm_vcpu *vcpu, struct kvm_lapic *source,
836                            int shorthand, unsigned int dest, int dest_mode)
837 {
838         struct kvm_lapic *target = vcpu->arch.apic;
839         u32 mda = kvm_apic_mda(vcpu, dest, source, target);
840 
841         ASSERT(target);
842         switch (shorthand) {
843         case APIC_DEST_NOSHORT:
844                 if (dest_mode == APIC_DEST_PHYSICAL)
845                         return kvm_apic_match_physical_addr(target, mda);
846                 else
847                         return kvm_apic_match_logical_addr(target, mda);
848         case APIC_DEST_SELF:
849                 return target == source;
850         case APIC_DEST_ALLINC:
851                 return true;
852         case APIC_DEST_ALLBUT:
853                 return target != source;
854         default:
855                 return false;
856         }
857 }
858 EXPORT_SYMBOL_GPL(kvm_apic_match_dest);
859 
860 int kvm_vector_to_index(u32 vector, u32 dest_vcpus,
861                        const unsigned long *bitmap, u32 bitmap_size)
862 {
863         u32 mod;
864         int i, idx = -1;
865 
866         mod = vector % dest_vcpus;
867 
868         for (i = 0; i <= mod; i++) {
869                 idx = find_next_bit(bitmap, bitmap_size, idx + 1);
870                 BUG_ON(idx == bitmap_size);
871         }
872 
873         return idx;
874 }
875 
876 static void kvm_apic_disabled_lapic_found(struct kvm *kvm)
877 {
878         if (!kvm->arch.disabled_lapic_found) {
879                 kvm->arch.disabled_lapic_found = true;
880                 printk(KERN_INFO
881                        "Disabled LAPIC found during irq injection\n");
882         }
883 }
884 
885 static bool kvm_apic_is_broadcast_dest(struct kvm *kvm, struct kvm_lapic **src,
886                 struct kvm_lapic_irq *irq, struct kvm_apic_map *map)
887 {
888         if (kvm->arch.x2apic_broadcast_quirk_disabled) {
889                 if ((irq->dest_id == APIC_BROADCAST &&
890                                 map->mode != KVM_APIC_MODE_X2APIC))
891                         return true;
892                 if (irq->dest_id == X2APIC_BROADCAST)
893                         return true;
894         } else {
895                 bool x2apic_ipi = src && *src && apic_x2apic_mode(*src);
896                 if (irq->dest_id == (x2apic_ipi ?
897                                      X2APIC_BROADCAST : APIC_BROADCAST))
898                         return true;
899         }
900 
901         return false;
902 }
903 
904 /* Return true if the interrupt can be handled by using *bitmap as index mask
905  * for valid destinations in *dst array.
906  * Return false if kvm_apic_map_get_dest_lapic did nothing useful.
907  * Note: we may have zero kvm_lapic destinations when we return true, which
908  * means that the interrupt should be dropped.  In this case, *bitmap would be
909  * zero and *dst undefined.
910  */
911 static inline bool kvm_apic_map_get_dest_lapic(struct kvm *kvm,
912                 struct kvm_lapic **src, struct kvm_lapic_irq *irq,
913                 struct kvm_apic_map *map, struct kvm_lapic ***dst,
914                 unsigned long *bitmap)
915 {
916         int i, lowest;
917 
918         if (irq->shorthand == APIC_DEST_SELF && src) {
919                 *dst = src;
920                 *bitmap = 1;
921                 return true;
922         } else if (irq->shorthand)
923                 return false;
924 
925         if (!map || kvm_apic_is_broadcast_dest(kvm, src, irq, map))
926                 return false;
927 
928         if (irq->dest_mode == APIC_DEST_PHYSICAL) {
929                 if (irq->dest_id > map->max_apic_id) {
930                         *bitmap = 0;
931                 } else {
932                         u32 dest_id = array_index_nospec(irq->dest_id, map->max_apic_id + 1);
933                         *dst = &map->phys_map[dest_id];
934                         *bitmap = 1;
935                 }
936                 return true;
937         }
938 
939         *bitmap = 0;
940         if (!kvm_apic_map_get_logical_dest(map, irq->dest_id, dst,
941                                 (u16 *)bitmap))
942                 return false;
943 
944         if (!kvm_lowest_prio_delivery(irq))
945                 return true;
946 
947         if (!kvm_vector_hashing_enabled()) {
948                 lowest = -1;
949                 for_each_set_bit(i, bitmap, 16) {
950                         if (!(*dst)[i])
951                                 continue;
952                         if (lowest < 0)
953                                 lowest = i;
954                         else if (kvm_apic_compare_prio((*dst)[i]->vcpu,
955                                                 (*dst)[lowest]->vcpu) < 0)
956                                 lowest = i;
957                 }
958         } else {
959                 if (!*bitmap)
960                         return true;
961 
962                 lowest = kvm_vector_to_index(irq->vector, hweight16(*bitmap),
963                                 bitmap, 16);
964 
965                 if (!(*dst)[lowest]) {
966                         kvm_apic_disabled_lapic_found(kvm);
967                         *bitmap = 0;
968                         return true;
969                 }
970         }
971 
972         *bitmap = (lowest >= 0) ? 1 << lowest : 0;
973 
974         return true;
975 }
976 
977 bool kvm_irq_delivery_to_apic_fast(struct kvm *kvm, struct kvm_lapic *src,
978                 struct kvm_lapic_irq *irq, int *r, struct dest_map *dest_map)
979 {
980         struct kvm_apic_map *map;
981         unsigned long bitmap;
982         struct kvm_lapic **dst = NULL;
983         int i;
984         bool ret;
985 
986         *r = -1;
987 
988         if (irq->shorthand == APIC_DEST_SELF) {
989                 *r = kvm_apic_set_irq(src->vcpu, irq, dest_map);
990                 return true;
991         }
992 
993         rcu_read_lock();
994         map = rcu_dereference(kvm->arch.apic_map);
995 
996         ret = kvm_apic_map_get_dest_lapic(kvm, &src, irq, map, &dst, &bitmap);
997         if (ret) {
998                 *r = 0;
999                 for_each_set_bit(i, &bitmap, 16) {
1000                         if (!dst[i])
1001                                 continue;
1002                         *r += kvm_apic_set_irq(dst[i]->vcpu, irq, dest_map);
1003                 }
1004         }
1005 
1006         rcu_read_unlock();
1007         return ret;
1008 }
1009 
1010 /*
1011  * This routine tries to handle interrupts in posted mode, here is how
1012  * it deals with different cases:
1013  * - For single-destination interrupts, handle it in posted mode
1014  * - Else if vector hashing is enabled and it is a lowest-priority
1015  *   interrupt, handle it in posted mode and use the following mechanism
1016  *   to find the destination vCPU.
1017  *      1. For lowest-priority interrupts, store all the possible
1018  *         destination vCPUs in an array.
1019  *      2. Use "guest vector % max number of destination vCPUs" to find
1020  *         the right destination vCPU in the array for the lowest-priority
1021  *         interrupt.
1022  * - Otherwise, use remapped mode to inject the interrupt.
1023  */
1024 bool kvm_intr_is_single_vcpu_fast(struct kvm *kvm, struct kvm_lapic_irq *irq,
1025                         struct kvm_vcpu **dest_vcpu)
1026 {
1027         struct kvm_apic_map *map;
1028         unsigned long bitmap;
1029         struct kvm_lapic **dst = NULL;
1030         bool ret = false;
1031 
1032         if (irq->shorthand)
1033                 return false;
1034 
1035         rcu_read_lock();
1036         map = rcu_dereference(kvm->arch.apic_map);
1037 
1038         if (kvm_apic_map_get_dest_lapic(kvm, NULL, irq, map, &dst, &bitmap) &&
1039                         hweight16(bitmap) == 1) {
1040                 unsigned long i = find_first_bit(&bitmap, 16);
1041 
1042                 if (dst[i]) {
1043                         *dest_vcpu = dst[i]->vcpu;
1044                         ret = true;
1045                 }
1046         }
1047 
1048         rcu_read_unlock();
1049         return ret;
1050 }
1051 
1052 /*
1053  * Add a pending IRQ into lapic.
1054  * Return 1 if successfully added and 0 if discarded.
1055  */
1056 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
1057                              int vector, int level, int trig_mode,
1058                              struct dest_map *dest_map)
1059 {
1060         int result = 0;
1061         struct kvm_vcpu *vcpu = apic->vcpu;
1062 
1063         trace_kvm_apic_accept_irq(vcpu->vcpu_id, delivery_mode,
1064                                   trig_mode, vector);
1065         switch (delivery_mode) {
1066         case APIC_DM_LOWEST:
1067                 vcpu->arch.apic_arb_prio++;
1068                 fallthrough;
1069         case APIC_DM_FIXED:
1070                 if (unlikely(trig_mode && !level))
1071                         break;
1072 
1073                 /* FIXME add logic for vcpu on reset */
1074                 if (unlikely(!apic_enabled(apic)))
1075                         break;
1076 
1077                 result = 1;
1078 
1079                 if (dest_map) {
1080                         __set_bit(vcpu->vcpu_id, dest_map->map);
1081                         dest_map->vectors[vcpu->vcpu_id] = vector;
1082                 }
1083 
1084                 if (apic_test_vector(vector, apic->regs + APIC_TMR) != !!trig_mode) {
1085                         if (trig_mode)
1086                                 kvm_lapic_set_vector(vector,
1087                                                      apic->regs + APIC_TMR);
1088                         else
1089                                 kvm_lapic_clear_vector(vector,
1090                                                        apic->regs + APIC_TMR);
1091                 }
1092 
1093                 if (kvm_x86_ops.deliver_posted_interrupt(vcpu, vector)) {
1094                         kvm_lapic_set_irr(vector, apic);
1095                         kvm_make_request(KVM_REQ_EVENT, vcpu);
1096                         kvm_vcpu_kick(vcpu);
1097                 }
1098                 break;
1099 
1100         case APIC_DM_REMRD:
1101                 result = 1;
1102                 vcpu->arch.pv.pv_unhalted = 1;
1103                 kvm_make_request(KVM_REQ_EVENT, vcpu);
1104                 kvm_vcpu_kick(vcpu);
1105                 break;
1106 
1107         case APIC_DM_SMI:
1108                 result = 1;
1109                 kvm_make_request(KVM_REQ_SMI, vcpu);
1110                 kvm_vcpu_kick(vcpu);
1111                 break;
1112 
1113         case APIC_DM_NMI:
1114                 result = 1;
1115                 kvm_inject_nmi(vcpu);
1116                 kvm_vcpu_kick(vcpu);
1117                 break;
1118 
1119         case APIC_DM_INIT:
1120                 if (!trig_mode || level) {
1121                         result = 1;
1122                         /* assumes that there are only KVM_APIC_INIT/SIPI */
1123                         apic->pending_events = (1UL << KVM_APIC_INIT);
1124                         kvm_make_request(KVM_REQ_EVENT, vcpu);
1125                         kvm_vcpu_kick(vcpu);
1126                 }
1127                 break;
1128 
1129         case APIC_DM_STARTUP:
1130                 result = 1;
1131                 apic->sipi_vector = vector;
1132                 /* make sure sipi_vector is visible for the receiver */
1133                 smp_wmb();
1134                 set_bit(KVM_APIC_SIPI, &apic->pending_events);
1135                 kvm_make_request(KVM_REQ_EVENT, vcpu);
1136                 kvm_vcpu_kick(vcpu);
1137                 break;
1138 
1139         case APIC_DM_EXTINT:
1140                 /*
1141                  * Should only be called by kvm_apic_local_deliver() with LVT0,
1142                  * before NMI watchdog was enabled. Already handled by
1143                  * kvm_apic_accept_pic_intr().
1144                  */
1145                 break;
1146 
1147         default:
1148                 printk(KERN_ERR "TODO: unsupported delivery mode %x\n",
1149                        delivery_mode);
1150                 break;
1151         }
1152         return result;
1153 }
1154 
1155 /*
1156  * This routine identifies the destination vcpus mask meant to receive the
1157  * IOAPIC interrupts. It either uses kvm_apic_map_get_dest_lapic() to find
1158  * out the destination vcpus array and set the bitmap or it traverses to
1159  * each available vcpu to identify the same.
1160  */
1161 void kvm_bitmap_or_dest_vcpus(struct kvm *kvm, struct kvm_lapic_irq *irq,
1162                               unsigned long *vcpu_bitmap)
1163 {
1164         struct kvm_lapic **dest_vcpu = NULL;
1165         struct kvm_lapic *src = NULL;
1166         struct kvm_apic_map *map;
1167         struct kvm_vcpu *vcpu;
1168         unsigned long bitmap;
1169         int i, vcpu_idx;
1170         bool ret;
1171 
1172         rcu_read_lock();
1173         map = rcu_dereference(kvm->arch.apic_map);
1174 
1175         ret = kvm_apic_map_get_dest_lapic(kvm, &src, irq, map, &dest_vcpu,
1176                                           &bitmap);
1177         if (ret) {
1178                 for_each_set_bit(i, &bitmap, 16) {
1179                         if (!dest_vcpu[i])
1180                                 continue;
1181                         vcpu_idx = dest_vcpu[i]->vcpu->vcpu_idx;
1182                         __set_bit(vcpu_idx, vcpu_bitmap);
1183                 }
1184         } else {
1185                 kvm_for_each_vcpu(i, vcpu, kvm) {
1186                         if (!kvm_apic_present(vcpu))
1187                                 continue;
1188                         if (!kvm_apic_match_dest(vcpu, NULL,
1189                                                  irq->shorthand,
1190                                                  irq->dest_id,
1191                                                  irq->dest_mode))
1192                                 continue;
1193                         __set_bit(i, vcpu_bitmap);
1194                 }
1195         }
1196         rcu_read_unlock();
1197 }
1198 
1199 int kvm_apic_compare_prio(struct kvm_vcpu *vcpu1, struct kvm_vcpu *vcpu2)
1200 {
1201         return vcpu1->arch.apic_arb_prio - vcpu2->arch.apic_arb_prio;
1202 }
1203 
1204 static bool kvm_ioapic_handles_vector(struct kvm_lapic *apic, int vector)
1205 {
1206         return test_bit(vector, apic->vcpu->arch.ioapic_handled_vectors);
1207 }
1208 
1209 static void kvm_ioapic_send_eoi(struct kvm_lapic *apic, int vector)
1210 {
1211         int trigger_mode;
1212 
1213         /* Eoi the ioapic only if the ioapic doesn't own the vector. */
1214         if (!kvm_ioapic_handles_vector(apic, vector))
1215                 return;
1216 
1217         /* Request a KVM exit to inform the userspace IOAPIC. */
1218         if (irqchip_split(apic->vcpu->kvm)) {
1219                 apic->vcpu->arch.pending_ioapic_eoi = vector;
1220                 kvm_make_request(KVM_REQ_IOAPIC_EOI_EXIT, apic->vcpu);
1221                 return;
1222         }
1223 
1224         if (apic_test_vector(vector, apic->regs + APIC_TMR))
1225                 trigger_mode = IOAPIC_LEVEL_TRIG;
1226         else
1227                 trigger_mode = IOAPIC_EDGE_TRIG;
1228 
1229         kvm_ioapic_update_eoi(apic->vcpu, vector, trigger_mode);
1230 }
1231 
1232 static int apic_set_eoi(struct kvm_lapic *apic)
1233 {
1234         int vector = apic_find_highest_isr(apic);
1235 
1236         trace_kvm_eoi(apic, vector);
1237 
1238         /*
1239          * Not every write EOI will has corresponding ISR,
1240          * one example is when Kernel check timer on setup_IO_APIC
1241          */
1242         if (vector == -1)
1243                 return vector;
1244 
1245         apic_clear_isr(vector, apic);
1246         apic_update_ppr(apic);
1247 
1248         if (test_bit(vector, vcpu_to_synic(apic->vcpu)->vec_bitmap))
1249                 kvm_hv_synic_send_eoi(apic->vcpu, vector);
1250 
1251         kvm_ioapic_send_eoi(apic, vector);
1252         kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
1253         return vector;
1254 }
1255 
1256 /*
1257  * this interface assumes a trap-like exit, which has already finished
1258  * desired side effect including vISR and vPPR update.
1259  */
1260 void kvm_apic_set_eoi_accelerated(struct kvm_vcpu *vcpu, int vector)
1261 {
1262         struct kvm_lapic *apic = vcpu->arch.apic;
1263 
1264         trace_kvm_eoi(apic, vector);
1265 
1266         kvm_ioapic_send_eoi(apic, vector);
1267         kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
1268 }
1269 EXPORT_SYMBOL_GPL(kvm_apic_set_eoi_accelerated);
1270 
1271 void kvm_apic_send_ipi(struct kvm_lapic *apic, u32 icr_low, u32 icr_high)
1272 {
1273         struct kvm_lapic_irq irq;
1274 
1275         irq.vector = icr_low & APIC_VECTOR_MASK;
1276         irq.delivery_mode = icr_low & APIC_MODE_MASK;
1277         irq.dest_mode = icr_low & APIC_DEST_MASK;
1278         irq.level = (icr_low & APIC_INT_ASSERT) != 0;
1279         irq.trig_mode = icr_low & APIC_INT_LEVELTRIG;
1280         irq.shorthand = icr_low & APIC_SHORT_MASK;
1281         irq.msi_redir_hint = false;
1282         if (apic_x2apic_mode(apic))
1283                 irq.dest_id = icr_high;
1284         else
1285                 irq.dest_id = GET_APIC_DEST_FIELD(icr_high);
1286 
1287         trace_kvm_apic_ipi(icr_low, irq.dest_id);
1288 
1289         kvm_irq_delivery_to_apic(apic->vcpu->kvm, apic, &irq, NULL);
1290 }
1291 
1292 static u32 apic_get_tmcct(struct kvm_lapic *apic)
1293 {
1294         ktime_t remaining, now;
1295         s64 ns;
1296         u32 tmcct;
1297 
1298         ASSERT(apic != NULL);
1299 
1300         /* if initial count is 0, current count should also be 0 */
1301         if (kvm_lapic_get_reg(apic, APIC_TMICT) == 0 ||
1302                 apic->lapic_timer.period == 0)
1303                 return 0;
1304 
1305         now = ktime_get();
1306         remaining = ktime_sub(apic->lapic_timer.target_expiration, now);
1307         if (ktime_to_ns(remaining) < 0)
1308                 remaining = 0;
1309 
1310         ns = mod_64(ktime_to_ns(remaining), apic->lapic_timer.period);
1311         tmcct = div64_u64(ns,
1312                          (APIC_BUS_CYCLE_NS * apic->divide_count));
1313 
1314         return tmcct;
1315 }
1316 
1317 static void __report_tpr_access(struct kvm_lapic *apic, bool write)
1318 {
1319         struct kvm_vcpu *vcpu = apic->vcpu;
1320         struct kvm_run *run = vcpu->run;
1321 
1322         kvm_make_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu);
1323         run->tpr_access.rip = kvm_rip_read(vcpu);
1324         run->tpr_access.is_write = write;
1325 }
1326 
1327 static inline void report_tpr_access(struct kvm_lapic *apic, bool write)
1328 {
1329         if (apic->vcpu->arch.tpr_access_reporting)
1330                 __report_tpr_access(apic, write);
1331 }
1332 
1333 static u32 __apic_read(struct kvm_lapic *apic, unsigned int offset)
1334 {
1335         u32 val = 0;
1336 
1337         if (offset >= LAPIC_MMIO_LENGTH)
1338                 return 0;
1339 
1340         switch (offset) {
1341         case APIC_ARBPRI:
1342                 break;
1343 
1344         case APIC_TMCCT:        /* Timer CCR */
1345                 if (apic_lvtt_tscdeadline(apic))
1346                         return 0;
1347 
1348                 val = apic_get_tmcct(apic);
1349                 break;
1350         case APIC_PROCPRI:
1351                 apic_update_ppr(apic);
1352                 val = kvm_lapic_get_reg(apic, offset);
1353                 break;
1354         case APIC_TASKPRI:
1355                 report_tpr_access(apic, false);
1356                 fallthrough;
1357         default:
1358                 val = kvm_lapic_get_reg(apic, offset);
1359                 break;
1360         }
1361 
1362         return val;
1363 }
1364 
1365 static inline struct kvm_lapic *to_lapic(struct kvm_io_device *dev)
1366 {
1367         return container_of(dev, struct kvm_lapic, dev);
1368 }
1369 
1370 #define APIC_REG_MASK(reg)      (1ull << ((reg) >> 4))
1371 #define APIC_REGS_MASK(first, count) \
1372         (APIC_REG_MASK(first) * ((1ull << (count)) - 1))
1373 
1374 int kvm_lapic_reg_read(struct kvm_lapic *apic, u32 offset, int len,
1375                 void *data)
1376 {
1377         unsigned char alignment = offset & 0xf;
1378         u32 result;
1379         /* this bitmask has a bit cleared for each reserved register */
1380         u64 valid_reg_mask =
1381                 APIC_REG_MASK(APIC_ID) |
1382                 APIC_REG_MASK(APIC_LVR) |
1383                 APIC_REG_MASK(APIC_TASKPRI) |
1384                 APIC_REG_MASK(APIC_PROCPRI) |
1385                 APIC_REG_MASK(APIC_LDR) |
1386                 APIC_REG_MASK(APIC_DFR) |
1387                 APIC_REG_MASK(APIC_SPIV) |
1388                 APIC_REGS_MASK(APIC_ISR, APIC_ISR_NR) |
1389                 APIC_REGS_MASK(APIC_TMR, APIC_ISR_NR) |
1390                 APIC_REGS_MASK(APIC_IRR, APIC_ISR_NR) |
1391                 APIC_REG_MASK(APIC_ESR) |
1392                 APIC_REG_MASK(APIC_ICR) |
1393                 APIC_REG_MASK(APIC_ICR2) |
1394                 APIC_REG_MASK(APIC_LVTT) |
1395                 APIC_REG_MASK(APIC_LVTTHMR) |
1396                 APIC_REG_MASK(APIC_LVTPC) |
1397                 APIC_REG_MASK(APIC_LVT0) |
1398                 APIC_REG_MASK(APIC_LVT1) |
1399                 APIC_REG_MASK(APIC_LVTERR) |
1400                 APIC_REG_MASK(APIC_TMICT) |
1401                 APIC_REG_MASK(APIC_TMCCT) |
1402                 APIC_REG_MASK(APIC_TDCR);
1403 
1404         /* ARBPRI is not valid on x2APIC */
1405         if (!apic_x2apic_mode(apic))
1406                 valid_reg_mask |= APIC_REG_MASK(APIC_ARBPRI);
1407 
1408         if (offset > 0x3f0 || !(valid_reg_mask & APIC_REG_MASK(offset)))
1409                 return 1;
1410 
1411         result = __apic_read(apic, offset & ~0xf);
1412 
1413         trace_kvm_apic_read(offset, result);
1414 
1415         switch (len) {
1416         case 1:
1417         case 2:
1418         case 4:
1419                 memcpy(data, (char *)&result + alignment, len);
1420                 break;
1421         default:
1422                 printk(KERN_ERR "Local APIC read with len = %x, "
1423                        "should be 1,2, or 4 instead\n", len);
1424                 break;
1425         }
1426         return 0;
1427 }
1428 EXPORT_SYMBOL_GPL(kvm_lapic_reg_read);
1429 
1430 static int apic_mmio_in_range(struct kvm_lapic *apic, gpa_t addr)
1431 {
1432         return addr >= apic->base_address &&
1433                 addr < apic->base_address + LAPIC_MMIO_LENGTH;
1434 }
1435 
1436 static int apic_mmio_read(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
1437                            gpa_t address, int len, void *data)
1438 {
1439         struct kvm_lapic *apic = to_lapic(this);
1440         u32 offset = address - apic->base_address;
1441 
1442         if (!apic_mmio_in_range(apic, address))
1443                 return -EOPNOTSUPP;
1444 
1445         if (!kvm_apic_hw_enabled(apic) || apic_x2apic_mode(apic)) {
1446                 if (!kvm_check_has_quirk(vcpu->kvm,
1447                                          KVM_X86_QUIRK_LAPIC_MMIO_HOLE))
1448                         return -EOPNOTSUPP;
1449 
1450                 memset(data, 0xff, len);
1451                 return 0;
1452         }
1453 
1454         kvm_lapic_reg_read(apic, offset, len, data);
1455 
1456         return 0;
1457 }
1458 
1459 static void update_divide_count(struct kvm_lapic *apic)
1460 {
1461         u32 tmp1, tmp2, tdcr;
1462 
1463         tdcr = kvm_lapic_get_reg(apic, APIC_TDCR);
1464         tmp1 = tdcr & 0xf;
1465         tmp2 = ((tmp1 & 0x3) | ((tmp1 & 0x8) >> 1)) + 1;
1466         apic->divide_count = 0x1 << (tmp2 & 0x7);
1467 }
1468 
1469 static void limit_periodic_timer_frequency(struct kvm_lapic *apic)
1470 {
1471         /*
1472          * Do not allow the guest to program periodic timers with small
1473          * interval, since the hrtimers are not throttled by the host
1474          * scheduler.
1475          */
1476         if (apic_lvtt_period(apic) && apic->lapic_timer.period) {
1477                 s64 min_period = min_timer_period_us * 1000LL;
1478 
1479                 if (apic->lapic_timer.period < min_period) {
1480                         pr_info_ratelimited(
1481                             "kvm: vcpu %i: requested %lld ns "
1482                             "lapic timer period limited to %lld ns\n",
1483                             apic->vcpu->vcpu_id,
1484                             apic->lapic_timer.period, min_period);
1485                         apic->lapic_timer.period = min_period;
1486                 }
1487         }
1488 }
1489 
1490 static void cancel_hv_timer(struct kvm_lapic *apic);
1491 
1492 static void apic_update_lvtt(struct kvm_lapic *apic)
1493 {
1494         u32 timer_mode = kvm_lapic_get_reg(apic, APIC_LVTT) &
1495                         apic->lapic_timer.timer_mode_mask;
1496 
1497         if (apic->lapic_timer.timer_mode != timer_mode) {
1498                 if (apic_lvtt_tscdeadline(apic) != (timer_mode ==
1499                                 APIC_LVT_TIMER_TSCDEADLINE)) {
1500                         hrtimer_cancel(&apic->lapic_timer.timer);
1501                         preempt_disable();
1502                         if (apic->lapic_timer.hv_timer_in_use)
1503                                 cancel_hv_timer(apic);
1504                         preempt_enable();
1505                         kvm_lapic_set_reg(apic, APIC_TMICT, 0);
1506                         apic->lapic_timer.period = 0;
1507                         apic->lapic_timer.tscdeadline = 0;
1508                 }
1509                 apic->lapic_timer.timer_mode = timer_mode;
1510                 limit_periodic_timer_frequency(apic);
1511         }
1512 }
1513 
1514 /*
1515  * On APICv, this test will cause a busy wait
1516  * during a higher-priority task.
1517  */
1518 
1519 static bool lapic_timer_int_injected(struct kvm_vcpu *vcpu)
1520 {
1521         struct kvm_lapic *apic = vcpu->arch.apic;
1522         u32 reg = kvm_lapic_get_reg(apic, APIC_LVTT);
1523 
1524         if (kvm_apic_hw_enabled(apic)) {
1525                 int vec = reg & APIC_VECTOR_MASK;
1526                 void *bitmap = apic->regs + APIC_ISR;
1527 
1528                 if (vcpu->arch.apicv_active)
1529                         bitmap = apic->regs + APIC_IRR;
1530 
1531                 if (apic_test_vector(vec, bitmap))
1532                         return true;
1533         }
1534         return false;
1535 }
1536 
1537 static inline void __wait_lapic_expire(struct kvm_vcpu *vcpu, u64 guest_cycles)
1538 {
1539         u64 timer_advance_ns = vcpu->arch.apic->lapic_timer.timer_advance_ns;
1540 
1541         /*
1542          * If the guest TSC is running at a different ratio than the host, then
1543          * convert the delay to nanoseconds to achieve an accurate delay.  Note
1544          * that __delay() uses delay_tsc whenever the hardware has TSC, thus
1545          * always for VMX enabled hardware.
1546          */
1547         if (vcpu->arch.tsc_scaling_ratio == kvm_default_tsc_scaling_ratio) {
1548                 __delay(min(guest_cycles,
1549                         nsec_to_cycles(vcpu, timer_advance_ns)));
1550         } else {
1551                 u64 delay_ns = guest_cycles * 1000000ULL;
1552                 do_div(delay_ns, vcpu->arch.virtual_tsc_khz);
1553                 ndelay(min_t(u32, delay_ns, timer_advance_ns));
1554         }
1555 }
1556 
1557 static inline void adjust_lapic_timer_advance(struct kvm_vcpu *vcpu,
1558                                               s64 advance_expire_delta)
1559 {
1560         struct kvm_lapic *apic = vcpu->arch.apic;
1561         u32 timer_advance_ns = apic->lapic_timer.timer_advance_ns;
1562         u64 ns;
1563 
1564         /* Do not adjust for tiny fluctuations or large random spikes. */
1565         if (abs(advance_expire_delta) > LAPIC_TIMER_ADVANCE_ADJUST_MAX ||
1566             abs(advance_expire_delta) < LAPIC_TIMER_ADVANCE_ADJUST_MIN)
1567                 return;
1568 
1569         /* too early */
1570         if (advance_expire_delta < 0) {
1571                 ns = -advance_expire_delta * 1000000ULL;
1572                 do_div(ns, vcpu->arch.virtual_tsc_khz);
1573                 timer_advance_ns -= ns/LAPIC_TIMER_ADVANCE_ADJUST_STEP;
1574         } else {
1575         /* too late */
1576                 ns = advance_expire_delta * 1000000ULL;
1577                 do_div(ns, vcpu->arch.virtual_tsc_khz);
1578                 timer_advance_ns += ns/LAPIC_TIMER_ADVANCE_ADJUST_STEP;
1579         }
1580 
1581         if (unlikely(timer_advance_ns > LAPIC_TIMER_ADVANCE_NS_MAX))
1582                 timer_advance_ns = LAPIC_TIMER_ADVANCE_NS_INIT;
1583         apic->lapic_timer.timer_advance_ns = timer_advance_ns;
1584 }
1585 
1586 static void __kvm_wait_lapic_expire(struct kvm_vcpu *vcpu)
1587 {
1588         struct kvm_lapic *apic = vcpu->arch.apic;
1589         u64 guest_tsc, tsc_deadline;
1590 
1591         tsc_deadline = apic->lapic_timer.expired_tscdeadline;
1592         apic->lapic_timer.expired_tscdeadline = 0;
1593         guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
1594         apic->lapic_timer.advance_expire_delta = guest_tsc - tsc_deadline;
1595 
1596         if (guest_tsc < tsc_deadline)
1597                 __wait_lapic_expire(vcpu, tsc_deadline - guest_tsc);
1598 
1599         if (lapic_timer_advance_dynamic)
1600                 adjust_lapic_timer_advance(vcpu, apic->lapic_timer.advance_expire_delta);
1601 }
1602 
1603 void kvm_wait_lapic_expire(struct kvm_vcpu *vcpu)
1604 {
1605         if (lapic_in_kernel(vcpu) &&
1606             vcpu->arch.apic->lapic_timer.expired_tscdeadline &&
1607             vcpu->arch.apic->lapic_timer.timer_advance_ns &&
1608             lapic_timer_int_injected(vcpu))
1609                 __kvm_wait_lapic_expire(vcpu);
1610 }
1611 EXPORT_SYMBOL_GPL(kvm_wait_lapic_expire);
1612 
1613 static void kvm_apic_inject_pending_timer_irqs(struct kvm_lapic *apic)
1614 {
1615         struct kvm_timer *ktimer = &apic->lapic_timer;
1616 
1617         kvm_apic_local_deliver(apic, APIC_LVTT);
1618         if (apic_lvtt_tscdeadline(apic)) {
1619                 ktimer->tscdeadline = 0;
1620         } else if (apic_lvtt_oneshot(apic)) {
1621                 ktimer->tscdeadline = 0;
1622                 ktimer->target_expiration = 0;
1623         }
1624 }
1625 
1626 static void apic_timer_expired(struct kvm_lapic *apic, bool from_timer_fn)
1627 {
1628         struct kvm_vcpu *vcpu = apic->vcpu;
1629         struct kvm_timer *ktimer = &apic->lapic_timer;
1630 
1631         if (atomic_read(&apic->lapic_timer.pending))
1632                 return;
1633 
1634         if (apic_lvtt_tscdeadline(apic) || ktimer->hv_timer_in_use)
1635                 ktimer->expired_tscdeadline = ktimer->tscdeadline;
1636 
1637         if (!from_timer_fn && vcpu->arch.apicv_active) {
1638                 WARN_ON(kvm_get_running_vcpu() != vcpu);
1639                 kvm_apic_inject_pending_timer_irqs(apic);
1640                 return;
1641         }
1642 
1643         if (kvm_use_posted_timer_interrupt(apic->vcpu)) {
1644                 kvm_wait_lapic_expire(vcpu);
1645                 kvm_apic_inject_pending_timer_irqs(apic);
1646                 return;
1647         }
1648 
1649         atomic_inc(&apic->lapic_timer.pending);
1650         kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1651         if (from_timer_fn)
1652                 kvm_vcpu_kick(vcpu);
1653 }
1654 
1655 static void start_sw_tscdeadline(struct kvm_lapic *apic)
1656 {
1657         struct kvm_timer *ktimer = &apic->lapic_timer;
1658         u64 guest_tsc, tscdeadline = ktimer->tscdeadline;
1659         u64 ns = 0;
1660         ktime_t expire;
1661         struct kvm_vcpu *vcpu = apic->vcpu;
1662         unsigned long this_tsc_khz = vcpu->arch.virtual_tsc_khz;
1663         unsigned long flags;
1664         ktime_t now;
1665 
1666         if (unlikely(!tscdeadline || !this_tsc_khz))
1667                 return;
1668 
1669         local_irq_save(flags);
1670 
1671         now = ktime_get();
1672         guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
1673 
1674         ns = (tscdeadline - guest_tsc) * 1000000ULL;
1675         do_div(ns, this_tsc_khz);
1676 
1677         if (likely(tscdeadline > guest_tsc) &&
1678             likely(ns > apic->lapic_timer.timer_advance_ns)) {
1679                 expire = ktime_add_ns(now, ns);
1680                 expire = ktime_sub_ns(expire, ktimer->timer_advance_ns);
1681                 hrtimer_start(&ktimer->timer, expire, HRTIMER_MODE_ABS_HARD);
1682         } else
1683                 apic_timer_expired(apic, false);
1684 
1685         local_irq_restore(flags);
1686 }
1687 
1688 static inline u64 tmict_to_ns(struct kvm_lapic *apic, u32 tmict)
1689 {
1690         return (u64)tmict * APIC_BUS_CYCLE_NS * (u64)apic->divide_count;
1691 }
1692 
1693 static void update_target_expiration(struct kvm_lapic *apic, uint32_t old_divisor)
1694 {
1695         ktime_t now, remaining;
1696         u64 ns_remaining_old, ns_remaining_new;
1697 
1698         apic->lapic_timer.period =
1699                         tmict_to_ns(apic, kvm_lapic_get_reg(apic, APIC_TMICT));
1700         limit_periodic_timer_frequency(apic);
1701 
1702         now = ktime_get();
1703         remaining = ktime_sub(apic->lapic_timer.target_expiration, now);
1704         if (ktime_to_ns(remaining) < 0)
1705                 remaining = 0;
1706 
1707         ns_remaining_old = ktime_to_ns(remaining);
1708         ns_remaining_new = mul_u64_u32_div(ns_remaining_old,
1709                                            apic->divide_count, old_divisor);
1710 
1711         apic->lapic_timer.tscdeadline +=
1712                 nsec_to_cycles(apic->vcpu, ns_remaining_new) -
1713                 nsec_to_cycles(apic->vcpu, ns_remaining_old);
1714         apic->lapic_timer.target_expiration = ktime_add_ns(now, ns_remaining_new);
1715 }
1716 
1717 static bool set_target_expiration(struct kvm_lapic *apic, u32 count_reg)
1718 {
1719         ktime_t now;
1720         u64 tscl = rdtsc();
1721         s64 deadline;
1722 
1723         now = ktime_get();
1724         apic->lapic_timer.period =
1725                         tmict_to_ns(apic, kvm_lapic_get_reg(apic, APIC_TMICT));
1726 
1727         if (!apic->lapic_timer.period) {
1728                 apic->lapic_timer.tscdeadline = 0;
1729                 return false;
1730         }
1731 
1732         limit_periodic_timer_frequency(apic);
1733         deadline = apic->lapic_timer.period;
1734 
1735         if (apic_lvtt_period(apic) || apic_lvtt_oneshot(apic)) {
1736                 if (unlikely(count_reg != APIC_TMICT)) {
1737                         deadline = tmict_to_ns(apic,
1738                                      kvm_lapic_get_reg(apic, count_reg));
1739                         if (unlikely(deadline <= 0))
1740                                 deadline = apic->lapic_timer.period;
1741                         else if (unlikely(deadline > apic->lapic_timer.period)) {
1742                                 pr_info_ratelimited(
1743                                     "kvm: vcpu %i: requested lapic timer restore with "
1744                                     "starting count register %#x=%u (%lld ns) > initial count (%lld ns). "
1745                                     "Using initial count to start timer.\n",
1746                                     apic->vcpu->vcpu_id,
1747                                     count_reg,
1748                                     kvm_lapic_get_reg(apic, count_reg),
1749                                     deadline, apic->lapic_timer.period);
1750                                 kvm_lapic_set_reg(apic, count_reg, 0);
1751                                 deadline = apic->lapic_timer.period;
1752                         }
1753                 }
1754         }
1755 
1756         apic->lapic_timer.tscdeadline = kvm_read_l1_tsc(apic->vcpu, tscl) +
1757                 nsec_to_cycles(apic->vcpu, deadline);
1758         apic->lapic_timer.target_expiration = ktime_add_ns(now, deadline);
1759 
1760         return true;
1761 }
1762 
1763 static void advance_periodic_target_expiration(struct kvm_lapic *apic)
1764 {
1765         ktime_t now = ktime_get();
1766         u64 tscl = rdtsc();
1767         ktime_t delta;
1768 
1769         /*
1770          * Synchronize both deadlines to the same time source or
1771          * differences in the periods (caused by differences in the
1772          * underlying clocks or numerical approximation errors) will
1773          * cause the two to drift apart over time as the errors
1774          * accumulate.
1775          */
1776         apic->lapic_timer.target_expiration =
1777                 ktime_add_ns(apic->lapic_timer.target_expiration,
1778                                 apic->lapic_timer.period);
1779         delta = ktime_sub(apic->lapic_timer.target_expiration, now);
1780         apic->lapic_timer.tscdeadline = kvm_read_l1_tsc(apic->vcpu, tscl) +
1781                 nsec_to_cycles(apic->vcpu, delta);
1782 }
1783 
1784 static void start_sw_period(struct kvm_lapic *apic)
1785 {
1786         if (!apic->lapic_timer.period)
1787                 return;
1788 
1789         if (ktime_after(ktime_get(),
1790                         apic->lapic_timer.target_expiration)) {
1791                 apic_timer_expired(apic, false);
1792 
1793                 if (apic_lvtt_oneshot(apic))
1794                         return;
1795 
1796                 advance_periodic_target_expiration(apic);
1797         }
1798 
1799         hrtimer_start(&apic->lapic_timer.timer,
1800                 apic->lapic_timer.target_expiration,
1801                 HRTIMER_MODE_ABS_HARD);
1802 }
1803 
1804 bool kvm_lapic_hv_timer_in_use(struct kvm_vcpu *vcpu)
1805 {
1806         if (!lapic_in_kernel(vcpu))
1807                 return false;
1808 
1809         return vcpu->arch.apic->lapic_timer.hv_timer_in_use;
1810 }
1811 EXPORT_SYMBOL_GPL(kvm_lapic_hv_timer_in_use);
1812 
1813 static void cancel_hv_timer(struct kvm_lapic *apic)
1814 {
1815         WARN_ON(preemptible());
1816         WARN_ON(!apic->lapic_timer.hv_timer_in_use);
1817         kvm_x86_ops.cancel_hv_timer(apic->vcpu);
1818         apic->lapic_timer.hv_timer_in_use = false;
1819 }
1820 
1821 static bool start_hv_timer(struct kvm_lapic *apic)
1822 {
1823         struct kvm_timer *ktimer = &apic->lapic_timer;
1824         struct kvm_vcpu *vcpu = apic->vcpu;
1825         bool expired;
1826 
1827         WARN_ON(preemptible());
1828         if (!kvm_can_use_hv_timer(vcpu))
1829                 return false;
1830 
1831         if (!ktimer->tscdeadline)
1832                 return false;
1833 
1834         if (kvm_x86_ops.set_hv_timer(vcpu, ktimer->tscdeadline, &expired))
1835                 return false;
1836 
1837         ktimer->hv_timer_in_use = true;
1838         hrtimer_cancel(&ktimer->timer);
1839 
1840         /*
1841          * To simplify handling the periodic timer, leave the hv timer running
1842          * even if the deadline timer has expired, i.e. rely on the resulting
1843          * VM-Exit to recompute the periodic timer's target expiration.
1844          */
1845         if (!apic_lvtt_period(apic)) {
1846                 /*
1847                  * Cancel the hv timer if the sw timer fired while the hv timer
1848                  * was being programmed, or if the hv timer itself expired.
1849                  */
1850                 if (atomic_read(&ktimer->pending)) {
1851                         cancel_hv_timer(apic);
1852                 } else if (expired) {
1853                         apic_timer_expired(apic, false);
1854                         cancel_hv_timer(apic);
1855                 }
1856         }
1857 
1858         trace_kvm_hv_timer_state(vcpu->vcpu_id, ktimer->hv_timer_in_use);
1859 
1860         return true;
1861 }
1862 
1863 static void start_sw_timer(struct kvm_lapic *apic)
1864 {
1865         struct kvm_timer *ktimer = &apic->lapic_timer;
1866 
1867         WARN_ON(preemptible());
1868         if (apic->lapic_timer.hv_timer_in_use)
1869                 cancel_hv_timer(apic);
1870         if (!apic_lvtt_period(apic) && atomic_read(&ktimer->pending))
1871                 return;
1872 
1873         if (apic_lvtt_period(apic) || apic_lvtt_oneshot(apic))
1874                 start_sw_period(apic);
1875         else if (apic_lvtt_tscdeadline(apic))
1876                 start_sw_tscdeadline(apic);
1877         trace_kvm_hv_timer_state(apic->vcpu->vcpu_id, false);
1878 }
1879 
1880 static void restart_apic_timer(struct kvm_lapic *apic)
1881 {
1882         preempt_disable();
1883 
1884         if (!apic_lvtt_period(apic) && atomic_read(&apic->lapic_timer.pending))
1885                 goto out;
1886 
1887         if (!start_hv_timer(apic))
1888                 start_sw_timer(apic);
1889 out:
1890         preempt_enable();
1891 }
1892 
1893 void kvm_lapic_expired_hv_timer(struct kvm_vcpu *vcpu)
1894 {
1895         struct kvm_lapic *apic = vcpu->arch.apic;
1896 
1897         preempt_disable();
1898         /* If the preempt notifier has already run, it also called apic_timer_expired */
1899         if (!apic->lapic_timer.hv_timer_in_use)
1900                 goto out;
1901         WARN_ON(rcuwait_active(&vcpu->wait));
1902         cancel_hv_timer(apic);
1903         apic_timer_expired(apic, false);
1904 
1905         if (apic_lvtt_period(apic) && apic->lapic_timer.period) {
1906                 advance_periodic_target_expiration(apic);
1907                 restart_apic_timer(apic);
1908         }
1909 out:
1910         preempt_enable();
1911 }
1912 EXPORT_SYMBOL_GPL(kvm_lapic_expired_hv_timer);
1913 
1914 void kvm_lapic_switch_to_hv_timer(struct kvm_vcpu *vcpu)
1915 {
1916         restart_apic_timer(vcpu->arch.apic);
1917 }
1918 EXPORT_SYMBOL_GPL(kvm_lapic_switch_to_hv_timer);
1919 
1920 void kvm_lapic_switch_to_sw_timer(struct kvm_vcpu *vcpu)
1921 {
1922         struct kvm_lapic *apic = vcpu->arch.apic;
1923 
1924         preempt_disable();
1925         /* Possibly the TSC deadline timer is not enabled yet */
1926         if (apic->lapic_timer.hv_timer_in_use)
1927                 start_sw_timer(apic);
1928         preempt_enable();
1929 }
1930 EXPORT_SYMBOL_GPL(kvm_lapic_switch_to_sw_timer);
1931 
1932 void kvm_lapic_restart_hv_timer(struct kvm_vcpu *vcpu)
1933 {
1934         struct kvm_lapic *apic = vcpu->arch.apic;
1935 
1936         WARN_ON(!apic->lapic_timer.hv_timer_in_use);
1937         restart_apic_timer(apic);
1938 }
1939 
1940 static void __start_apic_timer(struct kvm_lapic *apic, u32 count_reg)
1941 {
1942         atomic_set(&apic->lapic_timer.pending, 0);
1943 
1944         if ((apic_lvtt_period(apic) || apic_lvtt_oneshot(apic))
1945             && !set_target_expiration(apic, count_reg))
1946                 return;
1947 
1948         restart_apic_timer(apic);
1949 }
1950 
1951 static void start_apic_timer(struct kvm_lapic *apic)
1952 {
1953         __start_apic_timer(apic, APIC_TMICT);
1954 }
1955 
1956 static void apic_manage_nmi_watchdog(struct kvm_lapic *apic, u32 lvt0_val)
1957 {
1958         bool lvt0_in_nmi_mode = apic_lvt_nmi_mode(lvt0_val);
1959 
1960         if (apic->lvt0_in_nmi_mode != lvt0_in_nmi_mode) {
1961                 apic->lvt0_in_nmi_mode = lvt0_in_nmi_mode;
1962                 if (lvt0_in_nmi_mode) {
1963                         atomic_inc(&apic->vcpu->kvm->arch.vapics_in_nmi_mode);
1964                 } else
1965                         atomic_dec(&apic->vcpu->kvm->arch.vapics_in_nmi_mode);
1966         }
1967 }
1968 
1969 int kvm_lapic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val)
1970 {
1971         int ret = 0;
1972 
1973         trace_kvm_apic_write(reg, val);
1974 
1975         switch (reg) {
1976         case APIC_ID:           /* Local APIC ID */
1977                 if (!apic_x2apic_mode(apic))
1978                         kvm_apic_set_xapic_id(apic, val >> 24);
1979                 else
1980                         ret = 1;
1981                 break;
1982 
1983         case APIC_TASKPRI:
1984                 report_tpr_access(apic, true);
1985                 apic_set_tpr(apic, val & 0xff);
1986                 break;
1987 
1988         case APIC_EOI:
1989                 apic_set_eoi(apic);
1990                 break;
1991 
1992         case APIC_LDR:
1993                 if (!apic_x2apic_mode(apic))
1994                         kvm_apic_set_ldr(apic, val & APIC_LDR_MASK);
1995                 else
1996                         ret = 1;
1997                 break;
1998 
1999         case APIC_DFR:
2000                 if (!apic_x2apic_mode(apic))
2001                         kvm_apic_set_dfr(apic, val | 0x0FFFFFFF);
2002                 else
2003                         ret = 1;
2004                 break;
2005 
2006         case APIC_SPIV: {
2007                 u32 mask = 0x3ff;
2008                 if (kvm_lapic_get_reg(apic, APIC_LVR) & APIC_LVR_DIRECTED_EOI)
2009                         mask |= APIC_SPIV_DIRECTED_EOI;
2010                 apic_set_spiv(apic, val & mask);
2011                 if (!(val & APIC_SPIV_APIC_ENABLED)) {
2012                         int i;
2013                         u32 lvt_val;
2014 
2015                         for (i = 0; i < KVM_APIC_LVT_NUM; i++) {
2016                                 lvt_val = kvm_lapic_get_reg(apic,
2017                                                        APIC_LVTT + 0x10 * i);
2018                                 kvm_lapic_set_reg(apic, APIC_LVTT + 0x10 * i,
2019                                              lvt_val | APIC_LVT_MASKED);
2020                         }
2021                         apic_update_lvtt(apic);
2022                         atomic_set(&apic->lapic_timer.pending, 0);
2023 
2024                 }
2025                 break;
2026         }
2027         case APIC_ICR:
2028                 /* No delay here, so we always clear the pending bit */
2029                 val &= ~(1 << 12);
2030                 kvm_apic_send_ipi(apic, val, kvm_lapic_get_reg(apic, APIC_ICR2));
2031                 kvm_lapic_set_reg(apic, APIC_ICR, val);
2032                 break;
2033 
2034         case APIC_ICR2:
2035                 if (!apic_x2apic_mode(apic))
2036                         val &= 0xff000000;
2037                 kvm_lapic_set_reg(apic, APIC_ICR2, val);
2038                 break;
2039 
2040         case APIC_LVT0:
2041                 apic_manage_nmi_watchdog(apic, val);
2042                 fallthrough;
2043         case APIC_LVTTHMR:
2044         case APIC_LVTPC:
2045         case APIC_LVT1:
2046         case APIC_LVTERR: {
2047                 /* TODO: Check vector */
2048                 size_t size;
2049                 u32 index;
2050 
2051                 if (!kvm_apic_sw_enabled(apic))
2052                         val |= APIC_LVT_MASKED;
2053                 size = ARRAY_SIZE(apic_lvt_mask);
2054                 index = array_index_nospec(
2055                                 (reg - APIC_LVTT) >> 4, size);
2056                 val &= apic_lvt_mask[index];
2057                 kvm_lapic_set_reg(apic, reg, val);
2058                 break;
2059         }
2060 
2061         case APIC_LVTT:
2062                 if (!kvm_apic_sw_enabled(apic))
2063                         val |= APIC_LVT_MASKED;
2064                 val &= (apic_lvt_mask[0] | apic->lapic_timer.timer_mode_mask);
2065                 kvm_lapic_set_reg(apic, APIC_LVTT, val);
2066                 apic_update_lvtt(apic);
2067                 break;
2068 
2069         case APIC_TMICT:
2070                 if (apic_lvtt_tscdeadline(apic))
2071                         break;
2072 
2073                 hrtimer_cancel(&apic->lapic_timer.timer);
2074                 kvm_lapic_set_reg(apic, APIC_TMICT, val);
2075                 start_apic_timer(apic);
2076                 break;
2077 
2078         case APIC_TDCR: {
2079                 uint32_t old_divisor = apic->divide_count;
2080 
2081                 kvm_lapic_set_reg(apic, APIC_TDCR, val & 0xb);
2082                 update_divide_count(apic);
2083                 if (apic->divide_count != old_divisor &&
2084                                 apic->lapic_timer.period) {
2085                         hrtimer_cancel(&apic->lapic_timer.timer);
2086                         update_target_expiration(apic, old_divisor);
2087                         restart_apic_timer(apic);
2088                 }
2089                 break;
2090         }
2091         case APIC_ESR:
2092                 if (apic_x2apic_mode(apic) && val != 0)
2093                         ret = 1;
2094                 break;
2095 
2096         case APIC_SELF_IPI:
2097                 if (apic_x2apic_mode(apic)) {
2098                         kvm_lapic_reg_write(apic, APIC_ICR,
2099                                             APIC_DEST_SELF | (val & APIC_VECTOR_MASK));
2100                 } else
2101                         ret = 1;
2102                 break;
2103         default:
2104                 ret = 1;
2105                 break;
2106         }
2107 
2108         kvm_recalculate_apic_map(apic->vcpu->kvm);
2109 
2110         return ret;
2111 }
2112 EXPORT_SYMBOL_GPL(kvm_lapic_reg_write);
2113 
2114 static int apic_mmio_write(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
2115                             gpa_t address, int len, const void *data)
2116 {
2117         struct kvm_lapic *apic = to_lapic(this);
2118         unsigned int offset = address - apic->base_address;
2119         u32 val;
2120 
2121         if (!apic_mmio_in_range(apic, address))
2122                 return -EOPNOTSUPP;
2123 
2124         if (!kvm_apic_hw_enabled(apic) || apic_x2apic_mode(apic)) {
2125                 if (!kvm_check_has_quirk(vcpu->kvm,
2126                                          KVM_X86_QUIRK_LAPIC_MMIO_HOLE))
2127                         return -EOPNOTSUPP;
2128 
2129                 return 0;
2130         }
2131 
2132         /*
2133          * APIC register must be aligned on 128-bits boundary.
2134          * 32/64/128 bits registers must be accessed thru 32 bits.
2135          * Refer SDM 8.4.1
2136          */
2137         if (len != 4 || (offset & 0xf))
2138                 return 0;
2139 
2140         val = *(u32*)data;
2141 
2142         kvm_lapic_reg_write(apic, offset & 0xff0, val);
2143 
2144         return 0;
2145 }
2146 
2147 void kvm_lapic_set_eoi(struct kvm_vcpu *vcpu)
2148 {
2149         kvm_lapic_reg_write(vcpu->arch.apic, APIC_EOI, 0);
2150 }
2151 EXPORT_SYMBOL_GPL(kvm_lapic_set_eoi);
2152 
2153 /* emulate APIC access in a trap manner */
2154 void kvm_apic_write_nodecode(struct kvm_vcpu *vcpu, u32 offset)
2155 {
2156         u32 val = 0;
2157 
2158         /* hw has done the conditional check and inst decode */
2159         offset &= 0xff0;
2160 
2161         kvm_lapic_reg_read(vcpu->arch.apic, offset, 4, &val);
2162 
2163         /* TODO: optimize to just emulate side effect w/o one more write */
2164         kvm_lapic_reg_write(vcpu->arch.apic, offset, val);
2165 }
2166 EXPORT_SYMBOL_GPL(kvm_apic_write_nodecode);
2167 
2168 void kvm_free_lapic(struct kvm_vcpu *vcpu)
2169 {
2170         struct kvm_lapic *apic = vcpu->arch.apic;
2171 
2172         if (!vcpu->arch.apic)
2173                 return;
2174 
2175         hrtimer_cancel(&apic->lapic_timer.timer);
2176 
2177         if (!(vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE))
2178                 static_key_slow_dec_deferred(&apic_hw_disabled);
2179 
2180         if (!apic->sw_enabled)
2181                 static_key_slow_dec_deferred(&apic_sw_disabled);
2182 
2183         if (apic->regs)
2184                 free_page((unsigned long)apic->regs);
2185 
2186         kfree(apic);
2187 }
2188 
2189 /*
2190  *----------------------------------------------------------------------
2191  * LAPIC interface
2192  *----------------------------------------------------------------------
2193  */
2194 u64 kvm_get_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu)
2195 {
2196         struct kvm_lapic *apic = vcpu->arch.apic;
2197 
2198         if (!kvm_apic_present(vcpu) || !apic_lvtt_tscdeadline(apic))
2199                 return 0;
2200 
2201         return apic->lapic_timer.tscdeadline;
2202 }
2203 
2204 void kvm_set_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu, u64 data)
2205 {
2206         struct kvm_lapic *apic = vcpu->arch.apic;
2207 
2208         if (!kvm_apic_present(vcpu) || !apic_lvtt_tscdeadline(apic))
2209                 return;
2210 
2211         hrtimer_cancel(&apic->lapic_timer.timer);
2212         apic->lapic_timer.tscdeadline = data;
2213         start_apic_timer(apic);
2214 }
2215 
2216 void kvm_lapic_set_tpr(struct kvm_vcpu *vcpu, unsigned long cr8)
2217 {
2218         struct kvm_lapic *apic = vcpu->arch.apic;
2219 
2220         apic_set_tpr(apic, ((cr8 & 0x0f) << 4)
2221                      | (kvm_lapic_get_reg(apic, APIC_TASKPRI) & 4));
2222 }
2223 
2224 u64 kvm_lapic_get_cr8(struct kvm_vcpu *vcpu)
2225 {
2226         u64 tpr;
2227 
2228         tpr = (u64) kvm_lapic_get_reg(vcpu->arch.apic, APIC_TASKPRI);
2229 
2230         return (tpr & 0xf0) >> 4;
2231 }
2232 
2233 void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
2234 {
2235         u64 old_value = vcpu->arch.apic_base;
2236         struct kvm_lapic *apic = vcpu->arch.apic;
2237 
2238         if (!apic)
2239                 value |= MSR_IA32_APICBASE_BSP;
2240 
2241         vcpu->arch.apic_base = value;
2242 
2243         if ((old_value ^ value) & MSR_IA32_APICBASE_ENABLE)
2244                 kvm_update_cpuid_runtime(vcpu);
2245 
2246         if (!apic)
2247                 return;
2248 
2249         /* update jump label if enable bit changes */
2250         if ((old_value ^ value) & MSR_IA32_APICBASE_ENABLE) {
2251                 if (value & MSR_IA32_APICBASE_ENABLE) {
2252                         kvm_apic_set_xapic_id(apic, vcpu->vcpu_id);
2253                         static_key_slow_dec_deferred(&apic_hw_disabled);
2254                 } else {
2255                         static_key_slow_inc(&apic_hw_disabled.key);
2256                         atomic_set_release(&apic->vcpu->kvm->arch.apic_map_dirty, DIRTY);
2257                 }
2258         }
2259 
2260         if (((old_value ^ value) & X2APIC_ENABLE) && (value & X2APIC_ENABLE))
2261                 kvm_apic_set_x2apic_id(apic, vcpu->vcpu_id);
2262 
2263         if ((old_value ^ value) & (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE))
2264                 kvm_x86_ops.set_virtual_apic_mode(vcpu);
2265 
2266         apic->base_address = apic->vcpu->arch.apic_base &
2267                              MSR_IA32_APICBASE_BASE;
2268 
2269         if ((value & MSR_IA32_APICBASE_ENABLE) &&
2270              apic->base_address != APIC_DEFAULT_PHYS_BASE)
2271                 pr_warn_once("APIC base relocation is unsupported by KVM");
2272 }
2273 
2274 void kvm_apic_update_apicv(struct kvm_vcpu *vcpu)
2275 {
2276         struct kvm_lapic *apic = vcpu->arch.apic;
2277 
2278         if (vcpu->arch.apicv_active) {
2279                 /* irr_pending is always true when apicv is activated. */
2280                 apic->irr_pending = true;
2281                 apic->isr_count = 1;
2282         } else {
2283                 apic->irr_pending = (apic_search_irr(apic) != -1);
2284                 apic->isr_count = count_vectors(apic->regs + APIC_ISR);
2285         }
2286 }
2287 EXPORT_SYMBOL_GPL(kvm_apic_update_apicv);
2288 
2289 void kvm_lapic_reset(struct kvm_vcpu *vcpu, bool init_event)
2290 {
2291         struct kvm_lapic *apic = vcpu->arch.apic;
2292         int i;
2293 
2294         if (!apic)
2295                 return;
2296 
2297         /* Stop the timer in case it's a reset to an active apic */
2298         hrtimer_cancel(&apic->lapic_timer.timer);
2299 
2300         if (!init_event) {
2301                 kvm_lapic_set_base(vcpu, APIC_DEFAULT_PHYS_BASE |
2302                                          MSR_IA32_APICBASE_ENABLE);
2303                 kvm_apic_set_xapic_id(apic, vcpu->vcpu_id);
2304         }
2305         kvm_apic_set_version(apic->vcpu);
2306 
2307         for (i = 0; i < KVM_APIC_LVT_NUM; i++)
2308                 kvm_lapic_set_reg(apic, APIC_LVTT + 0x10 * i, APIC_LVT_MASKED);
2309         apic_update_lvtt(apic);
2310         if (kvm_vcpu_is_reset_bsp(vcpu) &&
2311             kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_LINT0_REENABLED))
2312                 kvm_lapic_set_reg(apic, APIC_LVT0,
2313                              SET_APIC_DELIVERY_MODE(0, APIC_MODE_EXTINT));
2314         apic_manage_nmi_watchdog(apic, kvm_lapic_get_reg(apic, APIC_LVT0));
2315 
2316         kvm_apic_set_dfr(apic, 0xffffffffU);
2317         apic_set_spiv(apic, 0xff);
2318         kvm_lapic_set_reg(apic, APIC_TASKPRI, 0);
2319         if (!apic_x2apic_mode(apic))
2320                 kvm_apic_set_ldr(apic, 0);
2321         kvm_lapic_set_reg(apic, APIC_ESR, 0);
2322         kvm_lapic_set_reg(apic, APIC_ICR, 0);
2323         kvm_lapic_set_reg(apic, APIC_ICR2, 0);
2324         kvm_lapic_set_reg(apic, APIC_TDCR, 0);
2325         kvm_lapic_set_reg(apic, APIC_TMICT, 0);
2326         for (i = 0; i < 8; i++) {
2327                 kvm_lapic_set_reg(apic, APIC_IRR + 0x10 * i, 0);
2328                 kvm_lapic_set_reg(apic, APIC_ISR + 0x10 * i, 0);
2329                 kvm_lapic_set_reg(apic, APIC_TMR + 0x10 * i, 0);
2330         }
2331         kvm_apic_update_apicv(vcpu);
2332         apic->highest_isr_cache = -1;
2333         update_divide_count(apic);
2334         atomic_set(&apic->lapic_timer.pending, 0);
2335         if (kvm_vcpu_is_bsp(vcpu))
2336                 kvm_lapic_set_base(vcpu,
2337                                 vcpu->arch.apic_base | MSR_IA32_APICBASE_BSP);
2338         vcpu->arch.pv_eoi.msr_val = 0;
2339         apic_update_ppr(apic);
2340         if (vcpu->arch.apicv_active) {
2341                 kvm_x86_ops.apicv_post_state_restore(vcpu);
2342                 kvm_x86_ops.hwapic_irr_update(vcpu, -1);
2343                 kvm_x86_ops.hwapic_isr_update(vcpu, -1);
2344         }
2345 
2346         vcpu->arch.apic_arb_prio = 0;
2347         vcpu->arch.apic_attention = 0;
2348 
2349         kvm_recalculate_apic_map(vcpu->kvm);
2350 }
2351 
2352 /*
2353  *----------------------------------------------------------------------
2354  * timer interface
2355  *----------------------------------------------------------------------
2356  */
2357 
2358 static bool lapic_is_periodic(struct kvm_lapic *apic)
2359 {
2360         return apic_lvtt_period(apic);
2361 }
2362 
2363 int apic_has_pending_timer(struct kvm_vcpu *vcpu)
2364 {
2365         struct kvm_lapic *apic = vcpu->arch.apic;
2366 
2367         if (apic_enabled(apic) && apic_lvt_enabled(apic, APIC_LVTT))
2368                 return atomic_read(&apic->lapic_timer.pending);
2369 
2370         return 0;
2371 }
2372 
2373 int kvm_apic_local_deliver(struct kvm_lapic *apic, int lvt_type)
2374 {
2375         u32 reg = kvm_lapic_get_reg(apic, lvt_type);
2376         int vector, mode, trig_mode;
2377 
2378         if (kvm_apic_hw_enabled(apic) && !(reg & APIC_LVT_MASKED)) {
2379                 vector = reg & APIC_VECTOR_MASK;
2380                 mode = reg & APIC_MODE_MASK;
2381                 trig_mode = reg & APIC_LVT_LEVEL_TRIGGER;
2382                 return __apic_accept_irq(apic, mode, vector, 1, trig_mode,
2383                                         NULL);
2384         }
2385         return 0;
2386 }
2387 
2388 void kvm_apic_nmi_wd_deliver(struct kvm_vcpu *vcpu)
2389 {
2390         struct kvm_lapic *apic = vcpu->arch.apic;
2391 
2392         if (apic)
2393                 kvm_apic_local_deliver(apic, APIC_LVT0);
2394 }
2395 
2396 static const struct kvm_io_device_ops apic_mmio_ops = {
2397         .read     = apic_mmio_read,
2398         .write    = apic_mmio_write,
2399 };
2400 
2401 static enum hrtimer_restart apic_timer_fn(struct hrtimer *data)
2402 {
2403         struct kvm_timer *ktimer = container_of(data, struct kvm_timer, timer);
2404         struct kvm_lapic *apic = container_of(ktimer, struct kvm_lapic, lapic_timer);
2405 
2406         apic_timer_expired(apic, true);
2407 
2408         if (lapic_is_periodic(apic)) {
2409                 advance_periodic_target_expiration(apic);
2410                 hrtimer_add_expires_ns(&ktimer->timer, ktimer->period);
2411                 return HRTIMER_RESTART;
2412         } else
2413                 return HRTIMER_NORESTART;
2414 }
2415 
2416 int kvm_create_lapic(struct kvm_vcpu *vcpu, int timer_advance_ns)
2417 {
2418         struct kvm_lapic *apic;
2419 
2420         ASSERT(vcpu != NULL);
2421 
2422         apic = kzalloc(sizeof(*apic), GFP_KERNEL_ACCOUNT);
2423         if (!apic)
2424                 goto nomem;
2425 
2426         vcpu->arch.apic = apic;
2427 
2428         apic->regs = (void *)get_zeroed_page(GFP_KERNEL_ACCOUNT);
2429         if (!apic->regs) {
2430                 printk(KERN_ERR "malloc apic regs error for vcpu %x\n",
2431                        vcpu->vcpu_id);
2432                 goto nomem_free_apic;
2433         }
2434         apic->vcpu = vcpu;
2435 
2436         hrtimer_init(&apic->lapic_timer.timer, CLOCK_MONOTONIC,
2437                      HRTIMER_MODE_ABS_HARD);
2438         apic->lapic_timer.timer.function = apic_timer_fn;
2439         if (timer_advance_ns == -1) {
2440                 apic->lapic_timer.timer_advance_ns = LAPIC_TIMER_ADVANCE_NS_INIT;
2441                 lapic_timer_advance_dynamic = true;
2442         } else {
2443                 apic->lapic_timer.timer_advance_ns = timer_advance_ns;
2444                 lapic_timer_advance_dynamic = false;
2445         }
2446 
2447         /*
2448          * APIC is created enabled. This will prevent kvm_lapic_set_base from
2449          * thinking that APIC state has changed.
2450          */
2451         vcpu->arch.apic_base = MSR_IA32_APICBASE_ENABLE;
2452         static_key_slow_inc(&apic_sw_disabled.key); /* sw disabled at reset */
2453         kvm_iodevice_init(&apic->dev, &apic_mmio_ops);
2454 
2455         return 0;
2456 nomem_free_apic:
2457         kfree(apic);
2458         vcpu->arch.apic = NULL;
2459 nomem:
2460         return -ENOMEM;
2461 }
2462 
2463 int kvm_apic_has_interrupt(struct kvm_vcpu *vcpu)
2464 {
2465         struct kvm_lapic *apic = vcpu->arch.apic;
2466         u32 ppr;
2467 
2468         if (!kvm_apic_hw_enabled(apic))
2469                 return -1;
2470 
2471         __apic_update_ppr(apic, &ppr);
2472         return apic_has_interrupt_for_ppr(apic, ppr);
2473 }
2474 EXPORT_SYMBOL_GPL(kvm_apic_has_interrupt);
2475 
2476 int kvm_apic_accept_pic_intr(struct kvm_vcpu *vcpu)
2477 {
2478         u32 lvt0 = kvm_lapic_get_reg(vcpu->arch.apic, APIC_LVT0);
2479 
2480         if (!kvm_apic_hw_enabled(vcpu->arch.apic))
2481                 return 1;
2482         if ((lvt0 & APIC_LVT_MASKED) == 0 &&
2483             GET_APIC_DELIVERY_MODE(lvt0) == APIC_MODE_EXTINT)
2484                 return 1;
2485         return 0;
2486 }
2487 
2488 void kvm_inject_apic_timer_irqs(struct kvm_vcpu *vcpu)
2489 {
2490         struct kvm_lapic *apic = vcpu->arch.apic;
2491 
2492         if (atomic_read(&apic->lapic_timer.pending) > 0) {
2493                 kvm_apic_inject_pending_timer_irqs(apic);
2494                 atomic_set(&apic->lapic_timer.pending, 0);
2495         }
2496 }
2497 
2498 int kvm_get_apic_interrupt(struct kvm_vcpu *vcpu)
2499 {
2500         int vector = kvm_apic_has_interrupt(vcpu);
2501         struct kvm_lapic *apic = vcpu->arch.apic;
2502         u32 ppr;
2503 
2504         if (vector == -1)
2505                 return -1;
2506 
2507         /*
2508          * We get here even with APIC virtualization enabled, if doing
2509          * nested virtualization and L1 runs with the "acknowledge interrupt
2510          * on exit" mode.  Then we cannot inject the interrupt via RVI,
2511          * because the process would deliver it through the IDT.
2512          */
2513 
2514         apic_clear_irr(vector, apic);
2515         if (test_bit(vector, vcpu_to_synic(vcpu)->auto_eoi_bitmap)) {
2516                 /*
2517                  * For auto-EOI interrupts, there might be another pending
2518                  * interrupt above PPR, so check whether to raise another
2519                  * KVM_REQ_EVENT.
2520                  */
2521                 apic_update_ppr(apic);
2522         } else {
2523                 /*
2524                  * For normal interrupts, PPR has been raised and there cannot
2525                  * be a higher-priority pending interrupt---except if there was
2526                  * a concurrent interrupt injection, but that would have
2527                  * triggered KVM_REQ_EVENT already.
2528                  */
2529                 apic_set_isr(vector, apic);
2530                 __apic_update_ppr(apic, &ppr);
2531         }
2532 
2533         return vector;
2534 }
2535 
2536 static int kvm_apic_state_fixup(struct kvm_vcpu *vcpu,
2537                 struct kvm_lapic_state *s, bool set)
2538 {
2539         if (apic_x2apic_mode(vcpu->arch.apic)) {
2540                 u32 *id = (u32 *)(s->regs + APIC_ID);
2541                 u32 *ldr = (u32 *)(s->regs + APIC_LDR);
2542 
2543                 if (vcpu->kvm->arch.x2apic_format) {
2544                         if (*id != vcpu->vcpu_id)
2545                                 return -EINVAL;
2546                 } else {
2547                         if (set)
2548                                 *id >>= 24;
2549                         else
2550                                 *id <<= 24;
2551                 }
2552 
2553                 /* In x2APIC mode, the LDR is fixed and based on the id */
2554                 if (set)
2555                         *ldr = kvm_apic_calc_x2apic_ldr(*id);
2556         }
2557 
2558         return 0;
2559 }
2560 
2561 int kvm_apic_get_state(struct kvm_vcpu *vcpu, struct kvm_lapic_state *s)
2562 {
2563         memcpy(s->regs, vcpu->arch.apic->regs, sizeof(*s));
2564 
2565         /*
2566          * Get calculated timer current count for remaining timer period (if
2567          * any) and store it in the returned register set.
2568          */
2569         __kvm_lapic_set_reg(s->regs, APIC_TMCCT,
2570                             __apic_read(vcpu->arch.apic, APIC_TMCCT));
2571 
2572         return kvm_apic_state_fixup(vcpu, s, false);
2573 }
2574 
2575 int kvm_apic_set_state(struct kvm_vcpu *vcpu, struct kvm_lapic_state *s)
2576 {
2577         struct kvm_lapic *apic = vcpu->arch.apic;
2578         int r;
2579 
2580         kvm_lapic_set_base(vcpu, vcpu->arch.apic_base);
2581         /* set SPIV separately to get count of SW disabled APICs right */
2582         apic_set_spiv(apic, *((u32 *)(s->regs + APIC_SPIV)));
2583 
2584         r = kvm_apic_state_fixup(vcpu, s, true);
2585         if (r) {
2586                 kvm_recalculate_apic_map(vcpu->kvm);
2587                 return r;
2588         }
2589         memcpy(vcpu->arch.apic->regs, s->regs, sizeof(*s));
2590 
2591         atomic_set_release(&apic->vcpu->kvm->arch.apic_map_dirty, DIRTY);
2592         kvm_recalculate_apic_map(vcpu->kvm);
2593         kvm_apic_set_version(vcpu);
2594 
2595         apic_update_ppr(apic);
2596         hrtimer_cancel(&apic->lapic_timer.timer);
2597         apic_update_lvtt(apic);
2598         apic_manage_nmi_watchdog(apic, kvm_lapic_get_reg(apic, APIC_LVT0));
2599         update_divide_count(apic);
2600         __start_apic_timer(apic, APIC_TMCCT);
2601         kvm_apic_update_apicv(vcpu);
2602         apic->highest_isr_cache = -1;
2603         if (vcpu->arch.apicv_active) {
2604                 kvm_x86_ops.apicv_post_state_restore(vcpu);
2605                 kvm_x86_ops.hwapic_irr_update(vcpu,
2606                                 apic_find_highest_irr(apic));
2607                 kvm_x86_ops.hwapic_isr_update(vcpu,
2608                                 apic_find_highest_isr(apic));
2609         }
2610         kvm_make_request(KVM_REQ_EVENT, vcpu);
2611         if (ioapic_in_kernel(vcpu->kvm))
2612                 kvm_rtc_eoi_tracking_restore_one(vcpu);
2613 
2614         vcpu->arch.apic_arb_prio = 0;
2615 
2616         return 0;
2617 }
2618 
2619 void __kvm_migrate_apic_timer(struct kvm_vcpu *vcpu)
2620 {
2621         struct hrtimer *timer;
2622 
2623         if (!lapic_in_kernel(vcpu) ||
2624                 kvm_can_post_timer_interrupt(vcpu))
2625                 return;
2626 
2627         timer = &vcpu->arch.apic->lapic_timer.timer;
2628         if (hrtimer_cancel(timer))
2629                 hrtimer_start_expires(timer, HRTIMER_MODE_ABS_HARD);
2630 }
2631 
2632 /*
2633  * apic_sync_pv_eoi_from_guest - called on vmexit or cancel interrupt
2634  *
2635  * Detect whether guest triggered PV EOI since the
2636  * last entry. If yes, set EOI on guests's behalf.
2637  * Clear PV EOI in guest memory in any case.
2638  */
2639 static void apic_sync_pv_eoi_from_guest(struct kvm_vcpu *vcpu,
2640                                         struct kvm_lapic *apic)
2641 {
2642         bool pending;
2643         int vector;
2644         /*
2645          * PV EOI state is derived from KVM_APIC_PV_EOI_PENDING in host
2646          * and KVM_PV_EOI_ENABLED in guest memory as follows:
2647          *
2648          * KVM_APIC_PV_EOI_PENDING is unset:
2649          *      -> host disabled PV EOI.
2650          * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is set:
2651          *      -> host enabled PV EOI, guest did not execute EOI yet.
2652          * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is unset:
2653          *      -> host enabled PV EOI, guest executed EOI.
2654          */
2655         BUG_ON(!pv_eoi_enabled(vcpu));
2656         pending = pv_eoi_get_pending(vcpu);
2657         /*
2658          * Clear pending bit in any case: it will be set again on vmentry.
2659          * While this might not be ideal from performance point of view,
2660          * this makes sure pv eoi is only enabled when we know it's safe.
2661          */
2662         pv_eoi_clr_pending(vcpu);
2663         if (pending)
2664                 return;
2665         vector = apic_set_eoi(apic);
2666         trace_kvm_pv_eoi(apic, vector);
2667 }
2668 
2669 void kvm_lapic_sync_from_vapic(struct kvm_vcpu *vcpu)
2670 {
2671         u32 data;
2672 
2673         if (test_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention))
2674                 apic_sync_pv_eoi_from_guest(vcpu, vcpu->arch.apic);
2675 
2676         if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
2677                 return;
2678 
2679         if (kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
2680                                   sizeof(u32)))
2681                 return;
2682 
2683         apic_set_tpr(vcpu->arch.apic, data & 0xff);
2684 }
2685 
2686 /*
2687  * apic_sync_pv_eoi_to_guest - called before vmentry
2688  *
2689  * Detect whether it's safe to enable PV EOI and
2690  * if yes do so.
2691  */
2692 static void apic_sync_pv_eoi_to_guest(struct kvm_vcpu *vcpu,
2693                                         struct kvm_lapic *apic)
2694 {
2695         if (!pv_eoi_enabled(vcpu) ||
2696             /* IRR set or many bits in ISR: could be nested. */
2697             apic->irr_pending ||
2698             /* Cache not set: could be safe but we don't bother. */
2699             apic->highest_isr_cache == -1 ||
2700             /* Need EOI to update ioapic. */
2701             kvm_ioapic_handles_vector(apic, apic->highest_isr_cache)) {
2702                 /*
2703                  * PV EOI was disabled by apic_sync_pv_eoi_from_guest
2704                  * so we need not do anything here.
2705                  */
2706                 return;
2707         }
2708 
2709         pv_eoi_set_pending(apic->vcpu);
2710 }
2711 
2712 void kvm_lapic_sync_to_vapic(struct kvm_vcpu *vcpu)
2713 {
2714         u32 data, tpr;
2715         int max_irr, max_isr;
2716         struct kvm_lapic *apic = vcpu->arch.apic;
2717 
2718         apic_sync_pv_eoi_to_guest(vcpu, apic);
2719 
2720         if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
2721                 return;
2722 
2723         tpr = kvm_lapic_get_reg(apic, APIC_TASKPRI) & 0xff;
2724         max_irr = apic_find_highest_irr(apic);
2725         if (max_irr < 0)
2726                 max_irr = 0;
2727         max_isr = apic_find_highest_isr(apic);
2728         if (max_isr < 0)
2729                 max_isr = 0;
2730         data = (tpr & 0xff) | ((max_isr & 0xf0) << 8) | (max_irr << 24);
2731 
2732         kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
2733                                 sizeof(u32));
2734 }
2735 
2736 int kvm_lapic_set_vapic_addr(struct kvm_vcpu *vcpu, gpa_t vapic_addr)
2737 {
2738         if (vapic_addr) {
2739                 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2740                                         &vcpu->arch.apic->vapic_cache,
2741                                         vapic_addr, sizeof(u32)))
2742                         return -EINVAL;
2743                 __set_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
2744         } else {
2745                 __clear_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
2746         }
2747 
2748         vcpu->arch.apic->vapic_addr = vapic_addr;
2749         return 0;
2750 }
2751 
2752 int kvm_x2apic_msr_write(struct kvm_vcpu *vcpu, u32 msr, u64 data)
2753 {
2754         struct kvm_lapic *apic = vcpu->arch.apic;
2755         u32 reg = (msr - APIC_BASE_MSR) << 4;
2756 
2757         if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(apic))
2758                 return 1;
2759 
2760         if (reg == APIC_ICR2)
2761                 return 1;
2762 
2763         /* if this is ICR write vector before command */
2764         if (reg == APIC_ICR)
2765                 kvm_lapic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
2766         return kvm_lapic_reg_write(apic, reg, (u32)data);
2767 }
2768 
2769 int kvm_x2apic_msr_read(struct kvm_vcpu *vcpu, u32 msr, u64 *data)
2770 {
2771         struct kvm_lapic *apic = vcpu->arch.apic;
2772         u32 reg = (msr - APIC_BASE_MSR) << 4, low, high = 0;
2773 
2774         if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(apic))
2775                 return 1;
2776 
2777         if (reg == APIC_DFR || reg == APIC_ICR2)
2778                 return 1;
2779 
2780         if (kvm_lapic_reg_read(apic, reg, 4, &low))
2781                 return 1;
2782         if (reg == APIC_ICR)
2783                 kvm_lapic_reg_read(apic, APIC_ICR2, 4, &high);
2784 
2785         *data = (((u64)high) << 32) | low;
2786 
2787         return 0;
2788 }
2789 
2790 int kvm_hv_vapic_msr_write(struct kvm_vcpu *vcpu, u32 reg, u64 data)
2791 {
2792         struct kvm_lapic *apic = vcpu->arch.apic;
2793 
2794         if (!lapic_in_kernel(vcpu))
2795                 return 1;
2796 
2797         /* if this is ICR write vector before command */
2798         if (reg == APIC_ICR)
2799                 kvm_lapic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
2800         return kvm_lapic_reg_write(apic, reg, (u32)data);
2801 }
2802 
2803 int kvm_hv_vapic_msr_read(struct kvm_vcpu *vcpu, u32 reg, u64 *data)
2804 {
2805         struct kvm_lapic *apic = vcpu->arch.apic;
2806         u32 low, high = 0;
2807 
2808         if (!lapic_in_kernel(vcpu))
2809                 return 1;
2810 
2811         if (kvm_lapic_reg_read(apic, reg, 4, &low))
2812                 return 1;
2813         if (reg == APIC_ICR)
2814                 kvm_lapic_reg_read(apic, APIC_ICR2, 4, &high);
2815 
2816         *data = (((u64)high) << 32) | low;
2817 
2818         return 0;
2819 }
2820 
2821 int kvm_lapic_enable_pv_eoi(struct kvm_vcpu *vcpu, u64 data, unsigned long len)
2822 {
2823         u64 addr = data & ~KVM_MSR_ENABLED;
2824         struct gfn_to_hva_cache *ghc = &vcpu->arch.pv_eoi.data;
2825         unsigned long new_len;
2826 
2827         if (!IS_ALIGNED(addr, 4))
2828                 return 1;
2829 
2830         vcpu->arch.pv_eoi.msr_val = data;
2831         if (!pv_eoi_enabled(vcpu))
2832                 return 0;
2833 
2834         if (addr == ghc->gpa && len <= ghc->len)
2835                 new_len = ghc->len;
2836         else
2837                 new_len = len;
2838 
2839         return kvm_gfn_to_hva_cache_init(vcpu->kvm, ghc, addr, new_len);
2840 }
2841 
2842 void kvm_apic_accept_events(struct kvm_vcpu *vcpu)
2843 {
2844         struct kvm_lapic *apic = vcpu->arch.apic;
2845         u8 sipi_vector;
2846         unsigned long pe;
2847 
2848         if (!lapic_in_kernel(vcpu) || !apic->pending_events)
2849                 return;
2850 
2851         /*
2852          * INITs are latched while CPU is in specific states
2853          * (SMM, VMX non-root mode, SVM with GIF=0).
2854          * Because a CPU cannot be in these states immediately
2855          * after it has processed an INIT signal (and thus in
2856          * KVM_MP_STATE_INIT_RECEIVED state), just eat SIPIs
2857          * and leave the INIT pending.
2858          */
2859         if (kvm_vcpu_latch_init(vcpu)) {
2860                 WARN_ON_ONCE(vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED);
2861                 if (test_bit(KVM_APIC_SIPI, &apic->pending_events))
2862                         clear_bit(KVM_APIC_SIPI, &apic->pending_events);
2863                 return;
2864         }
2865 
2866         pe = xchg(&apic->pending_events, 0);
2867         if (test_bit(KVM_APIC_INIT, &pe)) {
2868                 kvm_vcpu_reset(vcpu, true);
2869                 if (kvm_vcpu_is_bsp(apic->vcpu))
2870                         vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
2871                 else
2872                         vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
2873         }
2874         if (test_bit(KVM_APIC_SIPI, &pe) &&
2875             vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
2876                 /* evaluate pending_events before reading the vector */
2877                 smp_rmb();
2878                 sipi_vector = apic->sipi_vector;
2879                 kvm_vcpu_deliver_sipi_vector(vcpu, sipi_vector);
2880                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
2881         }
2882 }
2883 
2884 void kvm_lapic_init(void)
2885 {
2886         /* do not patch jump label more than once per second */
2887         jump_label_rate_limit(&apic_hw_disabled, HZ);
2888         jump_label_rate_limit(&apic_sw_disabled, HZ);
2889 }
2890 
2891 void kvm_lapic_exit(void)
2892 {
2893         static_key_deferred_flush(&apic_hw_disabled);
2894         static_key_deferred_flush(&apic_sw_disabled);
2895 }
2896 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp