xref: /openbmc/linux/arch/x86/kvm/lapic.c (revision 404e077a)
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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
19 
20 #include <linux/kvm_host.h>
21 #include <linux/kvm.h>
22 #include <linux/mm.h>
23 #include <linux/highmem.h>
24 #include <linux/smp.h>
25 #include <linux/hrtimer.h>
26 #include <linux/io.h>
27 #include <linux/export.h>
28 #include <linux/math64.h>
29 #include <linux/slab.h>
30 #include <asm/processor.h>
31 #include <asm/mce.h>
32 #include <asm/msr.h>
33 #include <asm/page.h>
34 #include <asm/current.h>
35 #include <asm/apicdef.h>
36 #include <asm/delay.h>
37 #include <linux/atomic.h>
38 #include <linux/jump_label.h>
39 #include "kvm_cache_regs.h"
40 #include "irq.h"
41 #include "ioapic.h"
42 #include "trace.h"
43 #include "x86.h"
44 #include "cpuid.h"
45 #include "hyperv.h"
46 #include "smm.h"
47 
48 #ifndef CONFIG_X86_64
49 #define mod_64(x, y) ((x) - (y) * div64_u64(x, y))
50 #else
51 #define mod_64(x, y) ((x) % (y))
52 #endif
53 
54 /* 14 is the version for Xeon and Pentium 8.4.8*/
55 #define APIC_VERSION			0x14UL
56 #define LAPIC_MMIO_LENGTH		(1 << 12)
57 /* followed define is not in apicdef.h */
58 #define MAX_APIC_VECTOR			256
59 #define APIC_VECTORS_PER_REG		32
60 
61 static bool lapic_timer_advance_dynamic __read_mostly;
62 #define LAPIC_TIMER_ADVANCE_ADJUST_MIN	100	/* clock cycles */
63 #define LAPIC_TIMER_ADVANCE_ADJUST_MAX	10000	/* clock cycles */
64 #define LAPIC_TIMER_ADVANCE_NS_INIT	1000
65 #define LAPIC_TIMER_ADVANCE_NS_MAX     5000
66 /* step-by-step approximation to mitigate fluctuation */
67 #define LAPIC_TIMER_ADVANCE_ADJUST_STEP 8
68 static int kvm_lapic_msr_read(struct kvm_lapic *apic, u32 reg, u64 *data);
69 static int kvm_lapic_msr_write(struct kvm_lapic *apic, u32 reg, u64 data);
70 
71 static inline void __kvm_lapic_set_reg(char *regs, int reg_off, u32 val)
72 {
73 	*((u32 *) (regs + reg_off)) = val;
74 }
75 
76 static inline void kvm_lapic_set_reg(struct kvm_lapic *apic, int reg_off, u32 val)
77 {
78 	__kvm_lapic_set_reg(apic->regs, reg_off, val);
79 }
80 
81 static __always_inline u64 __kvm_lapic_get_reg64(char *regs, int reg)
82 {
83 	BUILD_BUG_ON(reg != APIC_ICR);
84 	return *((u64 *) (regs + reg));
85 }
86 
87 static __always_inline u64 kvm_lapic_get_reg64(struct kvm_lapic *apic, int reg)
88 {
89 	return __kvm_lapic_get_reg64(apic->regs, reg);
90 }
91 
92 static __always_inline void __kvm_lapic_set_reg64(char *regs, int reg, u64 val)
93 {
94 	BUILD_BUG_ON(reg != APIC_ICR);
95 	*((u64 *) (regs + reg)) = val;
96 }
97 
98 static __always_inline void kvm_lapic_set_reg64(struct kvm_lapic *apic,
99 						int reg, u64 val)
100 {
101 	__kvm_lapic_set_reg64(apic->regs, reg, val);
102 }
103 
104 static inline int apic_test_vector(int vec, void *bitmap)
105 {
106 	return test_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
107 }
108 
109 bool kvm_apic_pending_eoi(struct kvm_vcpu *vcpu, int vector)
110 {
111 	struct kvm_lapic *apic = vcpu->arch.apic;
112 
113 	return apic_test_vector(vector, apic->regs + APIC_ISR) ||
114 		apic_test_vector(vector, apic->regs + APIC_IRR);
115 }
116 
117 static inline int __apic_test_and_set_vector(int vec, void *bitmap)
118 {
119 	return __test_and_set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
120 }
121 
122 static inline int __apic_test_and_clear_vector(int vec, void *bitmap)
123 {
124 	return __test_and_clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
125 }
126 
127 __read_mostly DEFINE_STATIC_KEY_DEFERRED_FALSE(apic_hw_disabled, HZ);
128 __read_mostly DEFINE_STATIC_KEY_DEFERRED_FALSE(apic_sw_disabled, HZ);
129 
130 static inline int apic_enabled(struct kvm_lapic *apic)
131 {
132 	return kvm_apic_sw_enabled(apic) &&	kvm_apic_hw_enabled(apic);
133 }
134 
135 #define LVT_MASK	\
136 	(APIC_LVT_MASKED | APIC_SEND_PENDING | APIC_VECTOR_MASK)
137 
138 #define LINT_MASK	\
139 	(LVT_MASK | APIC_MODE_MASK | APIC_INPUT_POLARITY | \
140 	 APIC_LVT_REMOTE_IRR | APIC_LVT_LEVEL_TRIGGER)
141 
142 static inline u32 kvm_x2apic_id(struct kvm_lapic *apic)
143 {
144 	return apic->vcpu->vcpu_id;
145 }
146 
147 static bool kvm_can_post_timer_interrupt(struct kvm_vcpu *vcpu)
148 {
149 	return pi_inject_timer && kvm_vcpu_apicv_active(vcpu) &&
150 		(kvm_mwait_in_guest(vcpu->kvm) || kvm_hlt_in_guest(vcpu->kvm));
151 }
152 
153 bool kvm_can_use_hv_timer(struct kvm_vcpu *vcpu)
154 {
155 	return kvm_x86_ops.set_hv_timer
156 	       && !(kvm_mwait_in_guest(vcpu->kvm) ||
157 		    kvm_can_post_timer_interrupt(vcpu));
158 }
159 
160 static bool kvm_use_posted_timer_interrupt(struct kvm_vcpu *vcpu)
161 {
162 	return kvm_can_post_timer_interrupt(vcpu) && vcpu->mode == IN_GUEST_MODE;
163 }
164 
165 static inline u32 kvm_apic_calc_x2apic_ldr(u32 id)
166 {
167 	return ((id >> 4) << 16) | (1 << (id & 0xf));
168 }
169 
170 static inline bool kvm_apic_map_get_logical_dest(struct kvm_apic_map *map,
171 		u32 dest_id, struct kvm_lapic ***cluster, u16 *mask) {
172 	switch (map->logical_mode) {
173 	case KVM_APIC_MODE_SW_DISABLED:
174 		/* Arbitrarily use the flat map so that @cluster isn't NULL. */
175 		*cluster = map->xapic_flat_map;
176 		*mask = 0;
177 		return true;
178 	case KVM_APIC_MODE_X2APIC: {
179 		u32 offset = (dest_id >> 16) * 16;
180 		u32 max_apic_id = map->max_apic_id;
181 
182 		if (offset <= max_apic_id) {
183 			u8 cluster_size = min(max_apic_id - offset + 1, 16U);
184 
185 			offset = array_index_nospec(offset, map->max_apic_id + 1);
186 			*cluster = &map->phys_map[offset];
187 			*mask = dest_id & (0xffff >> (16 - cluster_size));
188 		} else {
189 			*mask = 0;
190 		}
191 
192 		return true;
193 		}
194 	case KVM_APIC_MODE_XAPIC_FLAT:
195 		*cluster = map->xapic_flat_map;
196 		*mask = dest_id & 0xff;
197 		return true;
198 	case KVM_APIC_MODE_XAPIC_CLUSTER:
199 		*cluster = map->xapic_cluster_map[(dest_id >> 4) & 0xf];
200 		*mask = dest_id & 0xf;
201 		return true;
202 	case KVM_APIC_MODE_MAP_DISABLED:
203 		return false;
204 	default:
205 		WARN_ON_ONCE(1);
206 		return false;
207 	}
208 }
209 
210 static void kvm_apic_map_free(struct rcu_head *rcu)
211 {
212 	struct kvm_apic_map *map = container_of(rcu, struct kvm_apic_map, rcu);
213 
214 	kvfree(map);
215 }
216 
217 static int kvm_recalculate_phys_map(struct kvm_apic_map *new,
218 				    struct kvm_vcpu *vcpu,
219 				    bool *xapic_id_mismatch)
220 {
221 	struct kvm_lapic *apic = vcpu->arch.apic;
222 	u32 x2apic_id = kvm_x2apic_id(apic);
223 	u32 xapic_id = kvm_xapic_id(apic);
224 	u32 physical_id;
225 
226 	/*
227 	 * For simplicity, KVM always allocates enough space for all possible
228 	 * xAPIC IDs.  Yell, but don't kill the VM, as KVM can continue on
229 	 * without the optimized map.
230 	 */
231 	if (WARN_ON_ONCE(xapic_id > new->max_apic_id))
232 		return -EINVAL;
233 
234 	/*
235 	 * Bail if a vCPU was added and/or enabled its APIC between allocating
236 	 * the map and doing the actual calculations for the map.  Note, KVM
237 	 * hardcodes the x2APIC ID to vcpu_id, i.e. there's no TOCTOU bug if
238 	 * the compiler decides to reload x2apic_id after this check.
239 	 */
240 	if (x2apic_id > new->max_apic_id)
241 		return -E2BIG;
242 
243 	/*
244 	 * Deliberately truncate the vCPU ID when detecting a mismatched APIC
245 	 * ID to avoid false positives if the vCPU ID, i.e. x2APIC ID, is a
246 	 * 32-bit value.  Any unwanted aliasing due to truncation results will
247 	 * be detected below.
248 	 */
249 	if (!apic_x2apic_mode(apic) && xapic_id != (u8)vcpu->vcpu_id)
250 		*xapic_id_mismatch = true;
251 
252 	/*
253 	 * Apply KVM's hotplug hack if userspace has enable 32-bit APIC IDs.
254 	 * Allow sending events to vCPUs by their x2APIC ID even if the target
255 	 * vCPU is in legacy xAPIC mode, and silently ignore aliased xAPIC IDs
256 	 * (the x2APIC ID is truncated to 8 bits, causing IDs > 0xff to wrap
257 	 * and collide).
258 	 *
259 	 * Honor the architectural (and KVM's non-optimized) behavior if
260 	 * userspace has not enabled 32-bit x2APIC IDs.  Each APIC is supposed
261 	 * to process messages independently.  If multiple vCPUs have the same
262 	 * effective APIC ID, e.g. due to the x2APIC wrap or because the guest
263 	 * manually modified its xAPIC IDs, events targeting that ID are
264 	 * supposed to be recognized by all vCPUs with said ID.
265 	 */
266 	if (vcpu->kvm->arch.x2apic_format) {
267 		/* See also kvm_apic_match_physical_addr(). */
268 		if (apic_x2apic_mode(apic) || x2apic_id > 0xff)
269 			new->phys_map[x2apic_id] = apic;
270 
271 		if (!apic_x2apic_mode(apic) && !new->phys_map[xapic_id])
272 			new->phys_map[xapic_id] = apic;
273 	} else {
274 		/*
275 		 * Disable the optimized map if the physical APIC ID is already
276 		 * mapped, i.e. is aliased to multiple vCPUs.  The optimized
277 		 * map requires a strict 1:1 mapping between IDs and vCPUs.
278 		 */
279 		if (apic_x2apic_mode(apic))
280 			physical_id = x2apic_id;
281 		else
282 			physical_id = xapic_id;
283 
284 		if (new->phys_map[physical_id])
285 			return -EINVAL;
286 
287 		new->phys_map[physical_id] = apic;
288 	}
289 
290 	return 0;
291 }
292 
293 static void kvm_recalculate_logical_map(struct kvm_apic_map *new,
294 					struct kvm_vcpu *vcpu)
295 {
296 	struct kvm_lapic *apic = vcpu->arch.apic;
297 	enum kvm_apic_logical_mode logical_mode;
298 	struct kvm_lapic **cluster;
299 	u16 mask;
300 	u32 ldr;
301 
302 	if (new->logical_mode == KVM_APIC_MODE_MAP_DISABLED)
303 		return;
304 
305 	if (!kvm_apic_sw_enabled(apic))
306 		return;
307 
308 	ldr = kvm_lapic_get_reg(apic, APIC_LDR);
309 	if (!ldr)
310 		return;
311 
312 	if (apic_x2apic_mode(apic)) {
313 		logical_mode = KVM_APIC_MODE_X2APIC;
314 	} else {
315 		ldr = GET_APIC_LOGICAL_ID(ldr);
316 		if (kvm_lapic_get_reg(apic, APIC_DFR) == APIC_DFR_FLAT)
317 			logical_mode = KVM_APIC_MODE_XAPIC_FLAT;
318 		else
319 			logical_mode = KVM_APIC_MODE_XAPIC_CLUSTER;
320 	}
321 
322 	/*
323 	 * To optimize logical mode delivery, all software-enabled APICs must
324 	 * be configured for the same mode.
325 	 */
326 	if (new->logical_mode == KVM_APIC_MODE_SW_DISABLED) {
327 		new->logical_mode = logical_mode;
328 	} else if (new->logical_mode != logical_mode) {
329 		new->logical_mode = KVM_APIC_MODE_MAP_DISABLED;
330 		return;
331 	}
332 
333 	/*
334 	 * In x2APIC mode, the LDR is read-only and derived directly from the
335 	 * x2APIC ID, thus is guaranteed to be addressable.  KVM reuses
336 	 * kvm_apic_map.phys_map to optimize logical mode x2APIC interrupts by
337 	 * reversing the LDR calculation to get cluster of APICs, i.e. no
338 	 * additional work is required.
339 	 */
340 	if (apic_x2apic_mode(apic)) {
341 		WARN_ON_ONCE(ldr != kvm_apic_calc_x2apic_ldr(kvm_x2apic_id(apic)));
342 		return;
343 	}
344 
345 	if (WARN_ON_ONCE(!kvm_apic_map_get_logical_dest(new, ldr,
346 							&cluster, &mask))) {
347 		new->logical_mode = KVM_APIC_MODE_MAP_DISABLED;
348 		return;
349 	}
350 
351 	if (!mask)
352 		return;
353 
354 	ldr = ffs(mask) - 1;
355 	if (!is_power_of_2(mask) || cluster[ldr])
356 		new->logical_mode = KVM_APIC_MODE_MAP_DISABLED;
357 	else
358 		cluster[ldr] = apic;
359 }
360 
361 /*
362  * CLEAN -> DIRTY and UPDATE_IN_PROGRESS -> DIRTY changes happen without a lock.
363  *
364  * DIRTY -> UPDATE_IN_PROGRESS and UPDATE_IN_PROGRESS -> CLEAN happen with
365  * apic_map_lock_held.
366  */
367 enum {
368 	CLEAN,
369 	UPDATE_IN_PROGRESS,
370 	DIRTY
371 };
372 
373 void kvm_recalculate_apic_map(struct kvm *kvm)
374 {
375 	struct kvm_apic_map *new, *old = NULL;
376 	struct kvm_vcpu *vcpu;
377 	unsigned long i;
378 	u32 max_id = 255; /* enough space for any xAPIC ID */
379 	bool xapic_id_mismatch = false;
380 
381 	/* Read kvm->arch.apic_map_dirty before kvm->arch.apic_map.  */
382 	if (atomic_read_acquire(&kvm->arch.apic_map_dirty) == CLEAN)
383 		return;
384 
385 	WARN_ONCE(!irqchip_in_kernel(kvm),
386 		  "Dirty APIC map without an in-kernel local APIC");
387 
388 	mutex_lock(&kvm->arch.apic_map_lock);
389 	/*
390 	 * Read kvm->arch.apic_map_dirty before kvm->arch.apic_map
391 	 * (if clean) or the APIC registers (if dirty).
392 	 */
393 	if (atomic_cmpxchg_acquire(&kvm->arch.apic_map_dirty,
394 				   DIRTY, UPDATE_IN_PROGRESS) == CLEAN) {
395 		/* Someone else has updated the map. */
396 		mutex_unlock(&kvm->arch.apic_map_lock);
397 		return;
398 	}
399 
400 	kvm_for_each_vcpu(i, vcpu, kvm)
401 		if (kvm_apic_present(vcpu))
402 			max_id = max(max_id, kvm_x2apic_id(vcpu->arch.apic));
403 
404 	new = kvzalloc(sizeof(struct kvm_apic_map) +
405 	                   sizeof(struct kvm_lapic *) * ((u64)max_id + 1),
406 			   GFP_KERNEL_ACCOUNT);
407 
408 	if (!new)
409 		goto out;
410 
411 	new->max_apic_id = max_id;
412 	new->logical_mode = KVM_APIC_MODE_SW_DISABLED;
413 
414 	kvm_for_each_vcpu(i, vcpu, kvm) {
415 		if (!kvm_apic_present(vcpu))
416 			continue;
417 
418 		if (kvm_recalculate_phys_map(new, vcpu, &xapic_id_mismatch)) {
419 			kvfree(new);
420 			new = NULL;
421 			goto out;
422 		}
423 
424 		kvm_recalculate_logical_map(new, vcpu);
425 	}
426 out:
427 	/*
428 	 * The optimized map is effectively KVM's internal version of APICv,
429 	 * and all unwanted aliasing that results in disabling the optimized
430 	 * map also applies to APICv.
431 	 */
432 	if (!new)
433 		kvm_set_apicv_inhibit(kvm, APICV_INHIBIT_REASON_PHYSICAL_ID_ALIASED);
434 	else
435 		kvm_clear_apicv_inhibit(kvm, APICV_INHIBIT_REASON_PHYSICAL_ID_ALIASED);
436 
437 	if (!new || new->logical_mode == KVM_APIC_MODE_MAP_DISABLED)
438 		kvm_set_apicv_inhibit(kvm, APICV_INHIBIT_REASON_LOGICAL_ID_ALIASED);
439 	else
440 		kvm_clear_apicv_inhibit(kvm, APICV_INHIBIT_REASON_LOGICAL_ID_ALIASED);
441 
442 	if (xapic_id_mismatch)
443 		kvm_set_apicv_inhibit(kvm, APICV_INHIBIT_REASON_APIC_ID_MODIFIED);
444 	else
445 		kvm_clear_apicv_inhibit(kvm, APICV_INHIBIT_REASON_APIC_ID_MODIFIED);
446 
447 	old = rcu_dereference_protected(kvm->arch.apic_map,
448 			lockdep_is_held(&kvm->arch.apic_map_lock));
449 	rcu_assign_pointer(kvm->arch.apic_map, new);
450 	/*
451 	 * Write kvm->arch.apic_map before clearing apic->apic_map_dirty.
452 	 * If another update has come in, leave it DIRTY.
453 	 */
454 	atomic_cmpxchg_release(&kvm->arch.apic_map_dirty,
455 			       UPDATE_IN_PROGRESS, CLEAN);
456 	mutex_unlock(&kvm->arch.apic_map_lock);
457 
458 	if (old)
459 		call_rcu(&old->rcu, kvm_apic_map_free);
460 
461 	kvm_make_scan_ioapic_request(kvm);
462 }
463 
464 static inline void apic_set_spiv(struct kvm_lapic *apic, u32 val)
465 {
466 	bool enabled = val & APIC_SPIV_APIC_ENABLED;
467 
468 	kvm_lapic_set_reg(apic, APIC_SPIV, val);
469 
470 	if (enabled != apic->sw_enabled) {
471 		apic->sw_enabled = enabled;
472 		if (enabled)
473 			static_branch_slow_dec_deferred(&apic_sw_disabled);
474 		else
475 			static_branch_inc(&apic_sw_disabled.key);
476 
477 		atomic_set_release(&apic->vcpu->kvm->arch.apic_map_dirty, DIRTY);
478 	}
479 
480 	/* Check if there are APF page ready requests pending */
481 	if (enabled)
482 		kvm_make_request(KVM_REQ_APF_READY, apic->vcpu);
483 }
484 
485 static inline void kvm_apic_set_xapic_id(struct kvm_lapic *apic, u8 id)
486 {
487 	kvm_lapic_set_reg(apic, APIC_ID, id << 24);
488 	atomic_set_release(&apic->vcpu->kvm->arch.apic_map_dirty, DIRTY);
489 }
490 
491 static inline void kvm_apic_set_ldr(struct kvm_lapic *apic, u32 id)
492 {
493 	kvm_lapic_set_reg(apic, APIC_LDR, id);
494 	atomic_set_release(&apic->vcpu->kvm->arch.apic_map_dirty, DIRTY);
495 }
496 
497 static inline void kvm_apic_set_dfr(struct kvm_lapic *apic, u32 val)
498 {
499 	kvm_lapic_set_reg(apic, APIC_DFR, val);
500 	atomic_set_release(&apic->vcpu->kvm->arch.apic_map_dirty, DIRTY);
501 }
502 
503 static inline void kvm_apic_set_x2apic_id(struct kvm_lapic *apic, u32 id)
504 {
505 	u32 ldr = kvm_apic_calc_x2apic_ldr(id);
506 
507 	WARN_ON_ONCE(id != apic->vcpu->vcpu_id);
508 
509 	kvm_lapic_set_reg(apic, APIC_ID, id);
510 	kvm_lapic_set_reg(apic, APIC_LDR, ldr);
511 	atomic_set_release(&apic->vcpu->kvm->arch.apic_map_dirty, DIRTY);
512 }
513 
514 static inline int apic_lvt_enabled(struct kvm_lapic *apic, int lvt_type)
515 {
516 	return !(kvm_lapic_get_reg(apic, lvt_type) & APIC_LVT_MASKED);
517 }
518 
519 static inline int apic_lvtt_oneshot(struct kvm_lapic *apic)
520 {
521 	return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_ONESHOT;
522 }
523 
524 static inline int apic_lvtt_period(struct kvm_lapic *apic)
525 {
526 	return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_PERIODIC;
527 }
528 
529 static inline int apic_lvtt_tscdeadline(struct kvm_lapic *apic)
530 {
531 	return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_TSCDEADLINE;
532 }
533 
534 static inline int apic_lvt_nmi_mode(u32 lvt_val)
535 {
536 	return (lvt_val & (APIC_MODE_MASK | APIC_LVT_MASKED)) == APIC_DM_NMI;
537 }
538 
539 static inline bool kvm_lapic_lvt_supported(struct kvm_lapic *apic, int lvt_index)
540 {
541 	return apic->nr_lvt_entries > lvt_index;
542 }
543 
544 static inline int kvm_apic_calc_nr_lvt_entries(struct kvm_vcpu *vcpu)
545 {
546 	return KVM_APIC_MAX_NR_LVT_ENTRIES - !(vcpu->arch.mcg_cap & MCG_CMCI_P);
547 }
548 
549 void kvm_apic_set_version(struct kvm_vcpu *vcpu)
550 {
551 	struct kvm_lapic *apic = vcpu->arch.apic;
552 	u32 v = 0;
553 
554 	if (!lapic_in_kernel(vcpu))
555 		return;
556 
557 	v = APIC_VERSION | ((apic->nr_lvt_entries - 1) << 16);
558 
559 	/*
560 	 * KVM emulates 82093AA datasheet (with in-kernel IOAPIC implementation)
561 	 * which doesn't have EOI register; Some buggy OSes (e.g. Windows with
562 	 * Hyper-V role) disable EOI broadcast in lapic not checking for IOAPIC
563 	 * version first and level-triggered interrupts never get EOIed in
564 	 * IOAPIC.
565 	 */
566 	if (guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) &&
567 	    !ioapic_in_kernel(vcpu->kvm))
568 		v |= APIC_LVR_DIRECTED_EOI;
569 	kvm_lapic_set_reg(apic, APIC_LVR, v);
570 }
571 
572 void kvm_apic_after_set_mcg_cap(struct kvm_vcpu *vcpu)
573 {
574 	int nr_lvt_entries = kvm_apic_calc_nr_lvt_entries(vcpu);
575 	struct kvm_lapic *apic = vcpu->arch.apic;
576 	int i;
577 
578 	if (!lapic_in_kernel(vcpu) || nr_lvt_entries == apic->nr_lvt_entries)
579 		return;
580 
581 	/* Initialize/mask any "new" LVT entries. */
582 	for (i = apic->nr_lvt_entries; i < nr_lvt_entries; i++)
583 		kvm_lapic_set_reg(apic, APIC_LVTx(i), APIC_LVT_MASKED);
584 
585 	apic->nr_lvt_entries = nr_lvt_entries;
586 
587 	/* The number of LVT entries is reflected in the version register. */
588 	kvm_apic_set_version(vcpu);
589 }
590 
591 static const unsigned int apic_lvt_mask[KVM_APIC_MAX_NR_LVT_ENTRIES] = {
592 	[LVT_TIMER] = LVT_MASK,      /* timer mode mask added at runtime */
593 	[LVT_THERMAL_MONITOR] = LVT_MASK | APIC_MODE_MASK,
594 	[LVT_PERFORMANCE_COUNTER] = LVT_MASK | APIC_MODE_MASK,
595 	[LVT_LINT0] = LINT_MASK,
596 	[LVT_LINT1] = LINT_MASK,
597 	[LVT_ERROR] = LVT_MASK,
598 	[LVT_CMCI] = LVT_MASK | APIC_MODE_MASK
599 };
600 
601 static int find_highest_vector(void *bitmap)
602 {
603 	int vec;
604 	u32 *reg;
605 
606 	for (vec = MAX_APIC_VECTOR - APIC_VECTORS_PER_REG;
607 	     vec >= 0; vec -= APIC_VECTORS_PER_REG) {
608 		reg = bitmap + REG_POS(vec);
609 		if (*reg)
610 			return __fls(*reg) + vec;
611 	}
612 
613 	return -1;
614 }
615 
616 static u8 count_vectors(void *bitmap)
617 {
618 	int vec;
619 	u32 *reg;
620 	u8 count = 0;
621 
622 	for (vec = 0; vec < MAX_APIC_VECTOR; vec += APIC_VECTORS_PER_REG) {
623 		reg = bitmap + REG_POS(vec);
624 		count += hweight32(*reg);
625 	}
626 
627 	return count;
628 }
629 
630 bool __kvm_apic_update_irr(u32 *pir, void *regs, int *max_irr)
631 {
632 	u32 i, vec;
633 	u32 pir_val, irr_val, prev_irr_val;
634 	int max_updated_irr;
635 
636 	max_updated_irr = -1;
637 	*max_irr = -1;
638 
639 	for (i = vec = 0; i <= 7; i++, vec += 32) {
640 		pir_val = READ_ONCE(pir[i]);
641 		irr_val = *((u32 *)(regs + APIC_IRR + i * 0x10));
642 		if (pir_val) {
643 			prev_irr_val = irr_val;
644 			irr_val |= xchg(&pir[i], 0);
645 			*((u32 *)(regs + APIC_IRR + i * 0x10)) = irr_val;
646 			if (prev_irr_val != irr_val) {
647 				max_updated_irr =
648 					__fls(irr_val ^ prev_irr_val) + vec;
649 			}
650 		}
651 		if (irr_val)
652 			*max_irr = __fls(irr_val) + vec;
653 	}
654 
655 	return ((max_updated_irr != -1) &&
656 		(max_updated_irr == *max_irr));
657 }
658 EXPORT_SYMBOL_GPL(__kvm_apic_update_irr);
659 
660 bool kvm_apic_update_irr(struct kvm_vcpu *vcpu, u32 *pir, int *max_irr)
661 {
662 	struct kvm_lapic *apic = vcpu->arch.apic;
663 
664 	return __kvm_apic_update_irr(pir, apic->regs, max_irr);
665 }
666 EXPORT_SYMBOL_GPL(kvm_apic_update_irr);
667 
668 static inline int apic_search_irr(struct kvm_lapic *apic)
669 {
670 	return find_highest_vector(apic->regs + APIC_IRR);
671 }
672 
673 static inline int apic_find_highest_irr(struct kvm_lapic *apic)
674 {
675 	int result;
676 
677 	/*
678 	 * Note that irr_pending is just a hint. It will be always
679 	 * true with virtual interrupt delivery enabled.
680 	 */
681 	if (!apic->irr_pending)
682 		return -1;
683 
684 	result = apic_search_irr(apic);
685 	ASSERT(result == -1 || result >= 16);
686 
687 	return result;
688 }
689 
690 static inline void apic_clear_irr(int vec, struct kvm_lapic *apic)
691 {
692 	if (unlikely(apic->apicv_active)) {
693 		/* need to update RVI */
694 		kvm_lapic_clear_vector(vec, apic->regs + APIC_IRR);
695 		static_call_cond(kvm_x86_hwapic_irr_update)(apic->vcpu,
696 							    apic_find_highest_irr(apic));
697 	} else {
698 		apic->irr_pending = false;
699 		kvm_lapic_clear_vector(vec, apic->regs + APIC_IRR);
700 		if (apic_search_irr(apic) != -1)
701 			apic->irr_pending = true;
702 	}
703 }
704 
705 void kvm_apic_clear_irr(struct kvm_vcpu *vcpu, int vec)
706 {
707 	apic_clear_irr(vec, vcpu->arch.apic);
708 }
709 EXPORT_SYMBOL_GPL(kvm_apic_clear_irr);
710 
711 static inline void apic_set_isr(int vec, struct kvm_lapic *apic)
712 {
713 	if (__apic_test_and_set_vector(vec, apic->regs + APIC_ISR))
714 		return;
715 
716 	/*
717 	 * With APIC virtualization enabled, all caching is disabled
718 	 * because the processor can modify ISR under the hood.  Instead
719 	 * just set SVI.
720 	 */
721 	if (unlikely(apic->apicv_active))
722 		static_call_cond(kvm_x86_hwapic_isr_update)(vec);
723 	else {
724 		++apic->isr_count;
725 		BUG_ON(apic->isr_count > MAX_APIC_VECTOR);
726 		/*
727 		 * ISR (in service register) bit is set when injecting an interrupt.
728 		 * The highest vector is injected. Thus the latest bit set matches
729 		 * the highest bit in ISR.
730 		 */
731 		apic->highest_isr_cache = vec;
732 	}
733 }
734 
735 static inline int apic_find_highest_isr(struct kvm_lapic *apic)
736 {
737 	int result;
738 
739 	/*
740 	 * Note that isr_count is always 1, and highest_isr_cache
741 	 * is always -1, with APIC virtualization enabled.
742 	 */
743 	if (!apic->isr_count)
744 		return -1;
745 	if (likely(apic->highest_isr_cache != -1))
746 		return apic->highest_isr_cache;
747 
748 	result = find_highest_vector(apic->regs + APIC_ISR);
749 	ASSERT(result == -1 || result >= 16);
750 
751 	return result;
752 }
753 
754 static inline void apic_clear_isr(int vec, struct kvm_lapic *apic)
755 {
756 	if (!__apic_test_and_clear_vector(vec, apic->regs + APIC_ISR))
757 		return;
758 
759 	/*
760 	 * We do get here for APIC virtualization enabled if the guest
761 	 * uses the Hyper-V APIC enlightenment.  In this case we may need
762 	 * to trigger a new interrupt delivery by writing the SVI field;
763 	 * on the other hand isr_count and highest_isr_cache are unused
764 	 * and must be left alone.
765 	 */
766 	if (unlikely(apic->apicv_active))
767 		static_call_cond(kvm_x86_hwapic_isr_update)(apic_find_highest_isr(apic));
768 	else {
769 		--apic->isr_count;
770 		BUG_ON(apic->isr_count < 0);
771 		apic->highest_isr_cache = -1;
772 	}
773 }
774 
775 int kvm_lapic_find_highest_irr(struct kvm_vcpu *vcpu)
776 {
777 	/* This may race with setting of irr in __apic_accept_irq() and
778 	 * value returned may be wrong, but kvm_vcpu_kick() in __apic_accept_irq
779 	 * will cause vmexit immediately and the value will be recalculated
780 	 * on the next vmentry.
781 	 */
782 	return apic_find_highest_irr(vcpu->arch.apic);
783 }
784 EXPORT_SYMBOL_GPL(kvm_lapic_find_highest_irr);
785 
786 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
787 			     int vector, int level, int trig_mode,
788 			     struct dest_map *dest_map);
789 
790 int kvm_apic_set_irq(struct kvm_vcpu *vcpu, struct kvm_lapic_irq *irq,
791 		     struct dest_map *dest_map)
792 {
793 	struct kvm_lapic *apic = vcpu->arch.apic;
794 
795 	return __apic_accept_irq(apic, irq->delivery_mode, irq->vector,
796 			irq->level, irq->trig_mode, dest_map);
797 }
798 
799 static int __pv_send_ipi(unsigned long *ipi_bitmap, struct kvm_apic_map *map,
800 			 struct kvm_lapic_irq *irq, u32 min)
801 {
802 	int i, count = 0;
803 	struct kvm_vcpu *vcpu;
804 
805 	if (min > map->max_apic_id)
806 		return 0;
807 
808 	for_each_set_bit(i, ipi_bitmap,
809 		min((u32)BITS_PER_LONG, (map->max_apic_id - min + 1))) {
810 		if (map->phys_map[min + i]) {
811 			vcpu = map->phys_map[min + i]->vcpu;
812 			count += kvm_apic_set_irq(vcpu, irq, NULL);
813 		}
814 	}
815 
816 	return count;
817 }
818 
819 int kvm_pv_send_ipi(struct kvm *kvm, unsigned long ipi_bitmap_low,
820 		    unsigned long ipi_bitmap_high, u32 min,
821 		    unsigned long icr, int op_64_bit)
822 {
823 	struct kvm_apic_map *map;
824 	struct kvm_lapic_irq irq = {0};
825 	int cluster_size = op_64_bit ? 64 : 32;
826 	int count;
827 
828 	if (icr & (APIC_DEST_MASK | APIC_SHORT_MASK))
829 		return -KVM_EINVAL;
830 
831 	irq.vector = icr & APIC_VECTOR_MASK;
832 	irq.delivery_mode = icr & APIC_MODE_MASK;
833 	irq.level = (icr & APIC_INT_ASSERT) != 0;
834 	irq.trig_mode = icr & APIC_INT_LEVELTRIG;
835 
836 	rcu_read_lock();
837 	map = rcu_dereference(kvm->arch.apic_map);
838 
839 	count = -EOPNOTSUPP;
840 	if (likely(map)) {
841 		count = __pv_send_ipi(&ipi_bitmap_low, map, &irq, min);
842 		min += cluster_size;
843 		count += __pv_send_ipi(&ipi_bitmap_high, map, &irq, min);
844 	}
845 
846 	rcu_read_unlock();
847 	return count;
848 }
849 
850 static int pv_eoi_put_user(struct kvm_vcpu *vcpu, u8 val)
851 {
852 
853 	return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, &val,
854 				      sizeof(val));
855 }
856 
857 static int pv_eoi_get_user(struct kvm_vcpu *vcpu, u8 *val)
858 {
859 
860 	return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, val,
861 				      sizeof(*val));
862 }
863 
864 static inline bool pv_eoi_enabled(struct kvm_vcpu *vcpu)
865 {
866 	return vcpu->arch.pv_eoi.msr_val & KVM_MSR_ENABLED;
867 }
868 
869 static void pv_eoi_set_pending(struct kvm_vcpu *vcpu)
870 {
871 	if (pv_eoi_put_user(vcpu, KVM_PV_EOI_ENABLED) < 0)
872 		return;
873 
874 	__set_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
875 }
876 
877 static bool pv_eoi_test_and_clr_pending(struct kvm_vcpu *vcpu)
878 {
879 	u8 val;
880 
881 	if (pv_eoi_get_user(vcpu, &val) < 0)
882 		return false;
883 
884 	val &= KVM_PV_EOI_ENABLED;
885 
886 	if (val && pv_eoi_put_user(vcpu, KVM_PV_EOI_DISABLED) < 0)
887 		return false;
888 
889 	/*
890 	 * Clear pending bit in any case: it will be set again on vmentry.
891 	 * While this might not be ideal from performance point of view,
892 	 * this makes sure pv eoi is only enabled when we know it's safe.
893 	 */
894 	__clear_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
895 
896 	return val;
897 }
898 
899 static int apic_has_interrupt_for_ppr(struct kvm_lapic *apic, u32 ppr)
900 {
901 	int highest_irr;
902 	if (kvm_x86_ops.sync_pir_to_irr)
903 		highest_irr = static_call(kvm_x86_sync_pir_to_irr)(apic->vcpu);
904 	else
905 		highest_irr = apic_find_highest_irr(apic);
906 	if (highest_irr == -1 || (highest_irr & 0xF0) <= ppr)
907 		return -1;
908 	return highest_irr;
909 }
910 
911 static bool __apic_update_ppr(struct kvm_lapic *apic, u32 *new_ppr)
912 {
913 	u32 tpr, isrv, ppr, old_ppr;
914 	int isr;
915 
916 	old_ppr = kvm_lapic_get_reg(apic, APIC_PROCPRI);
917 	tpr = kvm_lapic_get_reg(apic, APIC_TASKPRI);
918 	isr = apic_find_highest_isr(apic);
919 	isrv = (isr != -1) ? isr : 0;
920 
921 	if ((tpr & 0xf0) >= (isrv & 0xf0))
922 		ppr = tpr & 0xff;
923 	else
924 		ppr = isrv & 0xf0;
925 
926 	*new_ppr = ppr;
927 	if (old_ppr != ppr)
928 		kvm_lapic_set_reg(apic, APIC_PROCPRI, ppr);
929 
930 	return ppr < old_ppr;
931 }
932 
933 static void apic_update_ppr(struct kvm_lapic *apic)
934 {
935 	u32 ppr;
936 
937 	if (__apic_update_ppr(apic, &ppr) &&
938 	    apic_has_interrupt_for_ppr(apic, ppr) != -1)
939 		kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
940 }
941 
942 void kvm_apic_update_ppr(struct kvm_vcpu *vcpu)
943 {
944 	apic_update_ppr(vcpu->arch.apic);
945 }
946 EXPORT_SYMBOL_GPL(kvm_apic_update_ppr);
947 
948 static void apic_set_tpr(struct kvm_lapic *apic, u32 tpr)
949 {
950 	kvm_lapic_set_reg(apic, APIC_TASKPRI, tpr);
951 	apic_update_ppr(apic);
952 }
953 
954 static bool kvm_apic_broadcast(struct kvm_lapic *apic, u32 mda)
955 {
956 	return mda == (apic_x2apic_mode(apic) ?
957 			X2APIC_BROADCAST : APIC_BROADCAST);
958 }
959 
960 static bool kvm_apic_match_physical_addr(struct kvm_lapic *apic, u32 mda)
961 {
962 	if (kvm_apic_broadcast(apic, mda))
963 		return true;
964 
965 	/*
966 	 * Hotplug hack: Accept interrupts for vCPUs in xAPIC mode as if they
967 	 * were in x2APIC mode if the target APIC ID can't be encoded as an
968 	 * xAPIC ID.  This allows unique addressing of hotplugged vCPUs (which
969 	 * start in xAPIC mode) with an APIC ID that is unaddressable in xAPIC
970 	 * mode.  Match the x2APIC ID if and only if the target APIC ID can't
971 	 * be encoded in xAPIC to avoid spurious matches against a vCPU that
972 	 * changed its (addressable) xAPIC ID (which is writable).
973 	 */
974 	if (apic_x2apic_mode(apic) || mda > 0xff)
975 		return mda == kvm_x2apic_id(apic);
976 
977 	return mda == kvm_xapic_id(apic);
978 }
979 
980 static bool kvm_apic_match_logical_addr(struct kvm_lapic *apic, u32 mda)
981 {
982 	u32 logical_id;
983 
984 	if (kvm_apic_broadcast(apic, mda))
985 		return true;
986 
987 	logical_id = kvm_lapic_get_reg(apic, APIC_LDR);
988 
989 	if (apic_x2apic_mode(apic))
990 		return ((logical_id >> 16) == (mda >> 16))
991 		       && (logical_id & mda & 0xffff) != 0;
992 
993 	logical_id = GET_APIC_LOGICAL_ID(logical_id);
994 
995 	switch (kvm_lapic_get_reg(apic, APIC_DFR)) {
996 	case APIC_DFR_FLAT:
997 		return (logical_id & mda) != 0;
998 	case APIC_DFR_CLUSTER:
999 		return ((logical_id >> 4) == (mda >> 4))
1000 		       && (logical_id & mda & 0xf) != 0;
1001 	default:
1002 		return false;
1003 	}
1004 }
1005 
1006 /* The KVM local APIC implementation has two quirks:
1007  *
1008  *  - Real hardware delivers interrupts destined to x2APIC ID > 0xff to LAPICs
1009  *    in xAPIC mode if the "destination & 0xff" matches its xAPIC ID.
1010  *    KVM doesn't do that aliasing.
1011  *
1012  *  - in-kernel IOAPIC messages have to be delivered directly to
1013  *    x2APIC, because the kernel does not support interrupt remapping.
1014  *    In order to support broadcast without interrupt remapping, x2APIC
1015  *    rewrites the destination of non-IPI messages from APIC_BROADCAST
1016  *    to X2APIC_BROADCAST.
1017  *
1018  * The broadcast quirk can be disabled with KVM_CAP_X2APIC_API.  This is
1019  * important when userspace wants to use x2APIC-format MSIs, because
1020  * APIC_BROADCAST (0xff) is a legal route for "cluster 0, CPUs 0-7".
1021  */
1022 static u32 kvm_apic_mda(struct kvm_vcpu *vcpu, unsigned int dest_id,
1023 		struct kvm_lapic *source, struct kvm_lapic *target)
1024 {
1025 	bool ipi = source != NULL;
1026 
1027 	if (!vcpu->kvm->arch.x2apic_broadcast_quirk_disabled &&
1028 	    !ipi && dest_id == APIC_BROADCAST && apic_x2apic_mode(target))
1029 		return X2APIC_BROADCAST;
1030 
1031 	return dest_id;
1032 }
1033 
1034 bool kvm_apic_match_dest(struct kvm_vcpu *vcpu, struct kvm_lapic *source,
1035 			   int shorthand, unsigned int dest, int dest_mode)
1036 {
1037 	struct kvm_lapic *target = vcpu->arch.apic;
1038 	u32 mda = kvm_apic_mda(vcpu, dest, source, target);
1039 
1040 	ASSERT(target);
1041 	switch (shorthand) {
1042 	case APIC_DEST_NOSHORT:
1043 		if (dest_mode == APIC_DEST_PHYSICAL)
1044 			return kvm_apic_match_physical_addr(target, mda);
1045 		else
1046 			return kvm_apic_match_logical_addr(target, mda);
1047 	case APIC_DEST_SELF:
1048 		return target == source;
1049 	case APIC_DEST_ALLINC:
1050 		return true;
1051 	case APIC_DEST_ALLBUT:
1052 		return target != source;
1053 	default:
1054 		return false;
1055 	}
1056 }
1057 EXPORT_SYMBOL_GPL(kvm_apic_match_dest);
1058 
1059 int kvm_vector_to_index(u32 vector, u32 dest_vcpus,
1060 		       const unsigned long *bitmap, u32 bitmap_size)
1061 {
1062 	u32 mod;
1063 	int i, idx = -1;
1064 
1065 	mod = vector % dest_vcpus;
1066 
1067 	for (i = 0; i <= mod; i++) {
1068 		idx = find_next_bit(bitmap, bitmap_size, idx + 1);
1069 		BUG_ON(idx == bitmap_size);
1070 	}
1071 
1072 	return idx;
1073 }
1074 
1075 static void kvm_apic_disabled_lapic_found(struct kvm *kvm)
1076 {
1077 	if (!kvm->arch.disabled_lapic_found) {
1078 		kvm->arch.disabled_lapic_found = true;
1079 		pr_info("Disabled LAPIC found during irq injection\n");
1080 	}
1081 }
1082 
1083 static bool kvm_apic_is_broadcast_dest(struct kvm *kvm, struct kvm_lapic **src,
1084 		struct kvm_lapic_irq *irq, struct kvm_apic_map *map)
1085 {
1086 	if (kvm->arch.x2apic_broadcast_quirk_disabled) {
1087 		if ((irq->dest_id == APIC_BROADCAST &&
1088 		     map->logical_mode != KVM_APIC_MODE_X2APIC))
1089 			return true;
1090 		if (irq->dest_id == X2APIC_BROADCAST)
1091 			return true;
1092 	} else {
1093 		bool x2apic_ipi = src && *src && apic_x2apic_mode(*src);
1094 		if (irq->dest_id == (x2apic_ipi ?
1095 		                     X2APIC_BROADCAST : APIC_BROADCAST))
1096 			return true;
1097 	}
1098 
1099 	return false;
1100 }
1101 
1102 /* Return true if the interrupt can be handled by using *bitmap as index mask
1103  * for valid destinations in *dst array.
1104  * Return false if kvm_apic_map_get_dest_lapic did nothing useful.
1105  * Note: we may have zero kvm_lapic destinations when we return true, which
1106  * means that the interrupt should be dropped.  In this case, *bitmap would be
1107  * zero and *dst undefined.
1108  */
1109 static inline bool kvm_apic_map_get_dest_lapic(struct kvm *kvm,
1110 		struct kvm_lapic **src, struct kvm_lapic_irq *irq,
1111 		struct kvm_apic_map *map, struct kvm_lapic ***dst,
1112 		unsigned long *bitmap)
1113 {
1114 	int i, lowest;
1115 
1116 	if (irq->shorthand == APIC_DEST_SELF && src) {
1117 		*dst = src;
1118 		*bitmap = 1;
1119 		return true;
1120 	} else if (irq->shorthand)
1121 		return false;
1122 
1123 	if (!map || kvm_apic_is_broadcast_dest(kvm, src, irq, map))
1124 		return false;
1125 
1126 	if (irq->dest_mode == APIC_DEST_PHYSICAL) {
1127 		if (irq->dest_id > map->max_apic_id) {
1128 			*bitmap = 0;
1129 		} else {
1130 			u32 dest_id = array_index_nospec(irq->dest_id, map->max_apic_id + 1);
1131 			*dst = &map->phys_map[dest_id];
1132 			*bitmap = 1;
1133 		}
1134 		return true;
1135 	}
1136 
1137 	*bitmap = 0;
1138 	if (!kvm_apic_map_get_logical_dest(map, irq->dest_id, dst,
1139 				(u16 *)bitmap))
1140 		return false;
1141 
1142 	if (!kvm_lowest_prio_delivery(irq))
1143 		return true;
1144 
1145 	if (!kvm_vector_hashing_enabled()) {
1146 		lowest = -1;
1147 		for_each_set_bit(i, bitmap, 16) {
1148 			if (!(*dst)[i])
1149 				continue;
1150 			if (lowest < 0)
1151 				lowest = i;
1152 			else if (kvm_apic_compare_prio((*dst)[i]->vcpu,
1153 						(*dst)[lowest]->vcpu) < 0)
1154 				lowest = i;
1155 		}
1156 	} else {
1157 		if (!*bitmap)
1158 			return true;
1159 
1160 		lowest = kvm_vector_to_index(irq->vector, hweight16(*bitmap),
1161 				bitmap, 16);
1162 
1163 		if (!(*dst)[lowest]) {
1164 			kvm_apic_disabled_lapic_found(kvm);
1165 			*bitmap = 0;
1166 			return true;
1167 		}
1168 	}
1169 
1170 	*bitmap = (lowest >= 0) ? 1 << lowest : 0;
1171 
1172 	return true;
1173 }
1174 
1175 bool kvm_irq_delivery_to_apic_fast(struct kvm *kvm, struct kvm_lapic *src,
1176 		struct kvm_lapic_irq *irq, int *r, struct dest_map *dest_map)
1177 {
1178 	struct kvm_apic_map *map;
1179 	unsigned long bitmap;
1180 	struct kvm_lapic **dst = NULL;
1181 	int i;
1182 	bool ret;
1183 
1184 	*r = -1;
1185 
1186 	if (irq->shorthand == APIC_DEST_SELF) {
1187 		if (KVM_BUG_ON(!src, kvm)) {
1188 			*r = 0;
1189 			return true;
1190 		}
1191 		*r = kvm_apic_set_irq(src->vcpu, irq, dest_map);
1192 		return true;
1193 	}
1194 
1195 	rcu_read_lock();
1196 	map = rcu_dereference(kvm->arch.apic_map);
1197 
1198 	ret = kvm_apic_map_get_dest_lapic(kvm, &src, irq, map, &dst, &bitmap);
1199 	if (ret) {
1200 		*r = 0;
1201 		for_each_set_bit(i, &bitmap, 16) {
1202 			if (!dst[i])
1203 				continue;
1204 			*r += kvm_apic_set_irq(dst[i]->vcpu, irq, dest_map);
1205 		}
1206 	}
1207 
1208 	rcu_read_unlock();
1209 	return ret;
1210 }
1211 
1212 /*
1213  * This routine tries to handle interrupts in posted mode, here is how
1214  * it deals with different cases:
1215  * - For single-destination interrupts, handle it in posted mode
1216  * - Else if vector hashing is enabled and it is a lowest-priority
1217  *   interrupt, handle it in posted mode and use the following mechanism
1218  *   to find the destination vCPU.
1219  *	1. For lowest-priority interrupts, store all the possible
1220  *	   destination vCPUs in an array.
1221  *	2. Use "guest vector % max number of destination vCPUs" to find
1222  *	   the right destination vCPU in the array for the lowest-priority
1223  *	   interrupt.
1224  * - Otherwise, use remapped mode to inject the interrupt.
1225  */
1226 bool kvm_intr_is_single_vcpu_fast(struct kvm *kvm, struct kvm_lapic_irq *irq,
1227 			struct kvm_vcpu **dest_vcpu)
1228 {
1229 	struct kvm_apic_map *map;
1230 	unsigned long bitmap;
1231 	struct kvm_lapic **dst = NULL;
1232 	bool ret = false;
1233 
1234 	if (irq->shorthand)
1235 		return false;
1236 
1237 	rcu_read_lock();
1238 	map = rcu_dereference(kvm->arch.apic_map);
1239 
1240 	if (kvm_apic_map_get_dest_lapic(kvm, NULL, irq, map, &dst, &bitmap) &&
1241 			hweight16(bitmap) == 1) {
1242 		unsigned long i = find_first_bit(&bitmap, 16);
1243 
1244 		if (dst[i]) {
1245 			*dest_vcpu = dst[i]->vcpu;
1246 			ret = true;
1247 		}
1248 	}
1249 
1250 	rcu_read_unlock();
1251 	return ret;
1252 }
1253 
1254 /*
1255  * Add a pending IRQ into lapic.
1256  * Return 1 if successfully added and 0 if discarded.
1257  */
1258 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
1259 			     int vector, int level, int trig_mode,
1260 			     struct dest_map *dest_map)
1261 {
1262 	int result = 0;
1263 	struct kvm_vcpu *vcpu = apic->vcpu;
1264 
1265 	trace_kvm_apic_accept_irq(vcpu->vcpu_id, delivery_mode,
1266 				  trig_mode, vector);
1267 	switch (delivery_mode) {
1268 	case APIC_DM_LOWEST:
1269 		vcpu->arch.apic_arb_prio++;
1270 		fallthrough;
1271 	case APIC_DM_FIXED:
1272 		if (unlikely(trig_mode && !level))
1273 			break;
1274 
1275 		/* FIXME add logic for vcpu on reset */
1276 		if (unlikely(!apic_enabled(apic)))
1277 			break;
1278 
1279 		result = 1;
1280 
1281 		if (dest_map) {
1282 			__set_bit(vcpu->vcpu_id, dest_map->map);
1283 			dest_map->vectors[vcpu->vcpu_id] = vector;
1284 		}
1285 
1286 		if (apic_test_vector(vector, apic->regs + APIC_TMR) != !!trig_mode) {
1287 			if (trig_mode)
1288 				kvm_lapic_set_vector(vector,
1289 						     apic->regs + APIC_TMR);
1290 			else
1291 				kvm_lapic_clear_vector(vector,
1292 						       apic->regs + APIC_TMR);
1293 		}
1294 
1295 		static_call(kvm_x86_deliver_interrupt)(apic, delivery_mode,
1296 						       trig_mode, vector);
1297 		break;
1298 
1299 	case APIC_DM_REMRD:
1300 		result = 1;
1301 		vcpu->arch.pv.pv_unhalted = 1;
1302 		kvm_make_request(KVM_REQ_EVENT, vcpu);
1303 		kvm_vcpu_kick(vcpu);
1304 		break;
1305 
1306 	case APIC_DM_SMI:
1307 		if (!kvm_inject_smi(vcpu)) {
1308 			kvm_vcpu_kick(vcpu);
1309 			result = 1;
1310 		}
1311 		break;
1312 
1313 	case APIC_DM_NMI:
1314 		result = 1;
1315 		kvm_inject_nmi(vcpu);
1316 		kvm_vcpu_kick(vcpu);
1317 		break;
1318 
1319 	case APIC_DM_INIT:
1320 		if (!trig_mode || level) {
1321 			result = 1;
1322 			/* assumes that there are only KVM_APIC_INIT/SIPI */
1323 			apic->pending_events = (1UL << KVM_APIC_INIT);
1324 			kvm_make_request(KVM_REQ_EVENT, vcpu);
1325 			kvm_vcpu_kick(vcpu);
1326 		}
1327 		break;
1328 
1329 	case APIC_DM_STARTUP:
1330 		result = 1;
1331 		apic->sipi_vector = vector;
1332 		/* make sure sipi_vector is visible for the receiver */
1333 		smp_wmb();
1334 		set_bit(KVM_APIC_SIPI, &apic->pending_events);
1335 		kvm_make_request(KVM_REQ_EVENT, vcpu);
1336 		kvm_vcpu_kick(vcpu);
1337 		break;
1338 
1339 	case APIC_DM_EXTINT:
1340 		/*
1341 		 * Should only be called by kvm_apic_local_deliver() with LVT0,
1342 		 * before NMI watchdog was enabled. Already handled by
1343 		 * kvm_apic_accept_pic_intr().
1344 		 */
1345 		break;
1346 
1347 	default:
1348 		printk(KERN_ERR "TODO: unsupported delivery mode %x\n",
1349 		       delivery_mode);
1350 		break;
1351 	}
1352 	return result;
1353 }
1354 
1355 /*
1356  * This routine identifies the destination vcpus mask meant to receive the
1357  * IOAPIC interrupts. It either uses kvm_apic_map_get_dest_lapic() to find
1358  * out the destination vcpus array and set the bitmap or it traverses to
1359  * each available vcpu to identify the same.
1360  */
1361 void kvm_bitmap_or_dest_vcpus(struct kvm *kvm, struct kvm_lapic_irq *irq,
1362 			      unsigned long *vcpu_bitmap)
1363 {
1364 	struct kvm_lapic **dest_vcpu = NULL;
1365 	struct kvm_lapic *src = NULL;
1366 	struct kvm_apic_map *map;
1367 	struct kvm_vcpu *vcpu;
1368 	unsigned long bitmap, i;
1369 	int vcpu_idx;
1370 	bool ret;
1371 
1372 	rcu_read_lock();
1373 	map = rcu_dereference(kvm->arch.apic_map);
1374 
1375 	ret = kvm_apic_map_get_dest_lapic(kvm, &src, irq, map, &dest_vcpu,
1376 					  &bitmap);
1377 	if (ret) {
1378 		for_each_set_bit(i, &bitmap, 16) {
1379 			if (!dest_vcpu[i])
1380 				continue;
1381 			vcpu_idx = dest_vcpu[i]->vcpu->vcpu_idx;
1382 			__set_bit(vcpu_idx, vcpu_bitmap);
1383 		}
1384 	} else {
1385 		kvm_for_each_vcpu(i, vcpu, kvm) {
1386 			if (!kvm_apic_present(vcpu))
1387 				continue;
1388 			if (!kvm_apic_match_dest(vcpu, NULL,
1389 						 irq->shorthand,
1390 						 irq->dest_id,
1391 						 irq->dest_mode))
1392 				continue;
1393 			__set_bit(i, vcpu_bitmap);
1394 		}
1395 	}
1396 	rcu_read_unlock();
1397 }
1398 
1399 int kvm_apic_compare_prio(struct kvm_vcpu *vcpu1, struct kvm_vcpu *vcpu2)
1400 {
1401 	return vcpu1->arch.apic_arb_prio - vcpu2->arch.apic_arb_prio;
1402 }
1403 
1404 static bool kvm_ioapic_handles_vector(struct kvm_lapic *apic, int vector)
1405 {
1406 	return test_bit(vector, apic->vcpu->arch.ioapic_handled_vectors);
1407 }
1408 
1409 static void kvm_ioapic_send_eoi(struct kvm_lapic *apic, int vector)
1410 {
1411 	int trigger_mode;
1412 
1413 	/* Eoi the ioapic only if the ioapic doesn't own the vector. */
1414 	if (!kvm_ioapic_handles_vector(apic, vector))
1415 		return;
1416 
1417 	/* Request a KVM exit to inform the userspace IOAPIC. */
1418 	if (irqchip_split(apic->vcpu->kvm)) {
1419 		apic->vcpu->arch.pending_ioapic_eoi = vector;
1420 		kvm_make_request(KVM_REQ_IOAPIC_EOI_EXIT, apic->vcpu);
1421 		return;
1422 	}
1423 
1424 	if (apic_test_vector(vector, apic->regs + APIC_TMR))
1425 		trigger_mode = IOAPIC_LEVEL_TRIG;
1426 	else
1427 		trigger_mode = IOAPIC_EDGE_TRIG;
1428 
1429 	kvm_ioapic_update_eoi(apic->vcpu, vector, trigger_mode);
1430 }
1431 
1432 static int apic_set_eoi(struct kvm_lapic *apic)
1433 {
1434 	int vector = apic_find_highest_isr(apic);
1435 
1436 	trace_kvm_eoi(apic, vector);
1437 
1438 	/*
1439 	 * Not every write EOI will has corresponding ISR,
1440 	 * one example is when Kernel check timer on setup_IO_APIC
1441 	 */
1442 	if (vector == -1)
1443 		return vector;
1444 
1445 	apic_clear_isr(vector, apic);
1446 	apic_update_ppr(apic);
1447 
1448 	if (to_hv_vcpu(apic->vcpu) &&
1449 	    test_bit(vector, to_hv_synic(apic->vcpu)->vec_bitmap))
1450 		kvm_hv_synic_send_eoi(apic->vcpu, vector);
1451 
1452 	kvm_ioapic_send_eoi(apic, vector);
1453 	kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
1454 	return vector;
1455 }
1456 
1457 /*
1458  * this interface assumes a trap-like exit, which has already finished
1459  * desired side effect including vISR and vPPR update.
1460  */
1461 void kvm_apic_set_eoi_accelerated(struct kvm_vcpu *vcpu, int vector)
1462 {
1463 	struct kvm_lapic *apic = vcpu->arch.apic;
1464 
1465 	trace_kvm_eoi(apic, vector);
1466 
1467 	kvm_ioapic_send_eoi(apic, vector);
1468 	kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
1469 }
1470 EXPORT_SYMBOL_GPL(kvm_apic_set_eoi_accelerated);
1471 
1472 void kvm_apic_send_ipi(struct kvm_lapic *apic, u32 icr_low, u32 icr_high)
1473 {
1474 	struct kvm_lapic_irq irq;
1475 
1476 	/* KVM has no delay and should always clear the BUSY/PENDING flag. */
1477 	WARN_ON_ONCE(icr_low & APIC_ICR_BUSY);
1478 
1479 	irq.vector = icr_low & APIC_VECTOR_MASK;
1480 	irq.delivery_mode = icr_low & APIC_MODE_MASK;
1481 	irq.dest_mode = icr_low & APIC_DEST_MASK;
1482 	irq.level = (icr_low & APIC_INT_ASSERT) != 0;
1483 	irq.trig_mode = icr_low & APIC_INT_LEVELTRIG;
1484 	irq.shorthand = icr_low & APIC_SHORT_MASK;
1485 	irq.msi_redir_hint = false;
1486 	if (apic_x2apic_mode(apic))
1487 		irq.dest_id = icr_high;
1488 	else
1489 		irq.dest_id = GET_XAPIC_DEST_FIELD(icr_high);
1490 
1491 	trace_kvm_apic_ipi(icr_low, irq.dest_id);
1492 
1493 	kvm_irq_delivery_to_apic(apic->vcpu->kvm, apic, &irq, NULL);
1494 }
1495 EXPORT_SYMBOL_GPL(kvm_apic_send_ipi);
1496 
1497 static u32 apic_get_tmcct(struct kvm_lapic *apic)
1498 {
1499 	ktime_t remaining, now;
1500 	s64 ns;
1501 
1502 	ASSERT(apic != NULL);
1503 
1504 	/* if initial count is 0, current count should also be 0 */
1505 	if (kvm_lapic_get_reg(apic, APIC_TMICT) == 0 ||
1506 		apic->lapic_timer.period == 0)
1507 		return 0;
1508 
1509 	now = ktime_get();
1510 	remaining = ktime_sub(apic->lapic_timer.target_expiration, now);
1511 	if (ktime_to_ns(remaining) < 0)
1512 		remaining = 0;
1513 
1514 	ns = mod_64(ktime_to_ns(remaining), apic->lapic_timer.period);
1515 	return div64_u64(ns, (APIC_BUS_CYCLE_NS * apic->divide_count));
1516 }
1517 
1518 static void __report_tpr_access(struct kvm_lapic *apic, bool write)
1519 {
1520 	struct kvm_vcpu *vcpu = apic->vcpu;
1521 	struct kvm_run *run = vcpu->run;
1522 
1523 	kvm_make_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu);
1524 	run->tpr_access.rip = kvm_rip_read(vcpu);
1525 	run->tpr_access.is_write = write;
1526 }
1527 
1528 static inline void report_tpr_access(struct kvm_lapic *apic, bool write)
1529 {
1530 	if (apic->vcpu->arch.tpr_access_reporting)
1531 		__report_tpr_access(apic, write);
1532 }
1533 
1534 static u32 __apic_read(struct kvm_lapic *apic, unsigned int offset)
1535 {
1536 	u32 val = 0;
1537 
1538 	if (offset >= LAPIC_MMIO_LENGTH)
1539 		return 0;
1540 
1541 	switch (offset) {
1542 	case APIC_ARBPRI:
1543 		break;
1544 
1545 	case APIC_TMCCT:	/* Timer CCR */
1546 		if (apic_lvtt_tscdeadline(apic))
1547 			return 0;
1548 
1549 		val = apic_get_tmcct(apic);
1550 		break;
1551 	case APIC_PROCPRI:
1552 		apic_update_ppr(apic);
1553 		val = kvm_lapic_get_reg(apic, offset);
1554 		break;
1555 	case APIC_TASKPRI:
1556 		report_tpr_access(apic, false);
1557 		fallthrough;
1558 	default:
1559 		val = kvm_lapic_get_reg(apic, offset);
1560 		break;
1561 	}
1562 
1563 	return val;
1564 }
1565 
1566 static inline struct kvm_lapic *to_lapic(struct kvm_io_device *dev)
1567 {
1568 	return container_of(dev, struct kvm_lapic, dev);
1569 }
1570 
1571 #define APIC_REG_MASK(reg)	(1ull << ((reg) >> 4))
1572 #define APIC_REGS_MASK(first, count) \
1573 	(APIC_REG_MASK(first) * ((1ull << (count)) - 1))
1574 
1575 u64 kvm_lapic_readable_reg_mask(struct kvm_lapic *apic)
1576 {
1577 	/* Leave bits '0' for reserved and write-only registers. */
1578 	u64 valid_reg_mask =
1579 		APIC_REG_MASK(APIC_ID) |
1580 		APIC_REG_MASK(APIC_LVR) |
1581 		APIC_REG_MASK(APIC_TASKPRI) |
1582 		APIC_REG_MASK(APIC_PROCPRI) |
1583 		APIC_REG_MASK(APIC_LDR) |
1584 		APIC_REG_MASK(APIC_SPIV) |
1585 		APIC_REGS_MASK(APIC_ISR, APIC_ISR_NR) |
1586 		APIC_REGS_MASK(APIC_TMR, APIC_ISR_NR) |
1587 		APIC_REGS_MASK(APIC_IRR, APIC_ISR_NR) |
1588 		APIC_REG_MASK(APIC_ESR) |
1589 		APIC_REG_MASK(APIC_ICR) |
1590 		APIC_REG_MASK(APIC_LVTT) |
1591 		APIC_REG_MASK(APIC_LVTTHMR) |
1592 		APIC_REG_MASK(APIC_LVTPC) |
1593 		APIC_REG_MASK(APIC_LVT0) |
1594 		APIC_REG_MASK(APIC_LVT1) |
1595 		APIC_REG_MASK(APIC_LVTERR) |
1596 		APIC_REG_MASK(APIC_TMICT) |
1597 		APIC_REG_MASK(APIC_TMCCT) |
1598 		APIC_REG_MASK(APIC_TDCR);
1599 
1600 	if (kvm_lapic_lvt_supported(apic, LVT_CMCI))
1601 		valid_reg_mask |= APIC_REG_MASK(APIC_LVTCMCI);
1602 
1603 	/* ARBPRI, DFR, and ICR2 are not valid in x2APIC mode. */
1604 	if (!apic_x2apic_mode(apic))
1605 		valid_reg_mask |= APIC_REG_MASK(APIC_ARBPRI) |
1606 				  APIC_REG_MASK(APIC_DFR) |
1607 				  APIC_REG_MASK(APIC_ICR2);
1608 
1609 	return valid_reg_mask;
1610 }
1611 EXPORT_SYMBOL_GPL(kvm_lapic_readable_reg_mask);
1612 
1613 static int kvm_lapic_reg_read(struct kvm_lapic *apic, u32 offset, int len,
1614 			      void *data)
1615 {
1616 	unsigned char alignment = offset & 0xf;
1617 	u32 result;
1618 
1619 	/*
1620 	 * WARN if KVM reads ICR in x2APIC mode, as it's an 8-byte register in
1621 	 * x2APIC and needs to be manually handled by the caller.
1622 	 */
1623 	WARN_ON_ONCE(apic_x2apic_mode(apic) && offset == APIC_ICR);
1624 
1625 	if (alignment + len > 4)
1626 		return 1;
1627 
1628 	if (offset > 0x3f0 ||
1629 	    !(kvm_lapic_readable_reg_mask(apic) & APIC_REG_MASK(offset)))
1630 		return 1;
1631 
1632 	result = __apic_read(apic, offset & ~0xf);
1633 
1634 	trace_kvm_apic_read(offset, result);
1635 
1636 	switch (len) {
1637 	case 1:
1638 	case 2:
1639 	case 4:
1640 		memcpy(data, (char *)&result + alignment, len);
1641 		break;
1642 	default:
1643 		printk(KERN_ERR "Local APIC read with len = %x, "
1644 		       "should be 1,2, or 4 instead\n", len);
1645 		break;
1646 	}
1647 	return 0;
1648 }
1649 
1650 static int apic_mmio_in_range(struct kvm_lapic *apic, gpa_t addr)
1651 {
1652 	return addr >= apic->base_address &&
1653 		addr < apic->base_address + LAPIC_MMIO_LENGTH;
1654 }
1655 
1656 static int apic_mmio_read(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
1657 			   gpa_t address, int len, void *data)
1658 {
1659 	struct kvm_lapic *apic = to_lapic(this);
1660 	u32 offset = address - apic->base_address;
1661 
1662 	if (!apic_mmio_in_range(apic, address))
1663 		return -EOPNOTSUPP;
1664 
1665 	if (!kvm_apic_hw_enabled(apic) || apic_x2apic_mode(apic)) {
1666 		if (!kvm_check_has_quirk(vcpu->kvm,
1667 					 KVM_X86_QUIRK_LAPIC_MMIO_HOLE))
1668 			return -EOPNOTSUPP;
1669 
1670 		memset(data, 0xff, len);
1671 		return 0;
1672 	}
1673 
1674 	kvm_lapic_reg_read(apic, offset, len, data);
1675 
1676 	return 0;
1677 }
1678 
1679 static void update_divide_count(struct kvm_lapic *apic)
1680 {
1681 	u32 tmp1, tmp2, tdcr;
1682 
1683 	tdcr = kvm_lapic_get_reg(apic, APIC_TDCR);
1684 	tmp1 = tdcr & 0xf;
1685 	tmp2 = ((tmp1 & 0x3) | ((tmp1 & 0x8) >> 1)) + 1;
1686 	apic->divide_count = 0x1 << (tmp2 & 0x7);
1687 }
1688 
1689 static void limit_periodic_timer_frequency(struct kvm_lapic *apic)
1690 {
1691 	/*
1692 	 * Do not allow the guest to program periodic timers with small
1693 	 * interval, since the hrtimers are not throttled by the host
1694 	 * scheduler.
1695 	 */
1696 	if (apic_lvtt_period(apic) && apic->lapic_timer.period) {
1697 		s64 min_period = min_timer_period_us * 1000LL;
1698 
1699 		if (apic->lapic_timer.period < min_period) {
1700 			pr_info_ratelimited(
1701 			    "vcpu %i: requested %lld ns "
1702 			    "lapic timer period limited to %lld ns\n",
1703 			    apic->vcpu->vcpu_id,
1704 			    apic->lapic_timer.period, min_period);
1705 			apic->lapic_timer.period = min_period;
1706 		}
1707 	}
1708 }
1709 
1710 static void cancel_hv_timer(struct kvm_lapic *apic);
1711 
1712 static void cancel_apic_timer(struct kvm_lapic *apic)
1713 {
1714 	hrtimer_cancel(&apic->lapic_timer.timer);
1715 	preempt_disable();
1716 	if (apic->lapic_timer.hv_timer_in_use)
1717 		cancel_hv_timer(apic);
1718 	preempt_enable();
1719 	atomic_set(&apic->lapic_timer.pending, 0);
1720 }
1721 
1722 static void apic_update_lvtt(struct kvm_lapic *apic)
1723 {
1724 	u32 timer_mode = kvm_lapic_get_reg(apic, APIC_LVTT) &
1725 			apic->lapic_timer.timer_mode_mask;
1726 
1727 	if (apic->lapic_timer.timer_mode != timer_mode) {
1728 		if (apic_lvtt_tscdeadline(apic) != (timer_mode ==
1729 				APIC_LVT_TIMER_TSCDEADLINE)) {
1730 			cancel_apic_timer(apic);
1731 			kvm_lapic_set_reg(apic, APIC_TMICT, 0);
1732 			apic->lapic_timer.period = 0;
1733 			apic->lapic_timer.tscdeadline = 0;
1734 		}
1735 		apic->lapic_timer.timer_mode = timer_mode;
1736 		limit_periodic_timer_frequency(apic);
1737 	}
1738 }
1739 
1740 /*
1741  * On APICv, this test will cause a busy wait
1742  * during a higher-priority task.
1743  */
1744 
1745 static bool lapic_timer_int_injected(struct kvm_vcpu *vcpu)
1746 {
1747 	struct kvm_lapic *apic = vcpu->arch.apic;
1748 	u32 reg = kvm_lapic_get_reg(apic, APIC_LVTT);
1749 
1750 	if (kvm_apic_hw_enabled(apic)) {
1751 		int vec = reg & APIC_VECTOR_MASK;
1752 		void *bitmap = apic->regs + APIC_ISR;
1753 
1754 		if (apic->apicv_active)
1755 			bitmap = apic->regs + APIC_IRR;
1756 
1757 		if (apic_test_vector(vec, bitmap))
1758 			return true;
1759 	}
1760 	return false;
1761 }
1762 
1763 static inline void __wait_lapic_expire(struct kvm_vcpu *vcpu, u64 guest_cycles)
1764 {
1765 	u64 timer_advance_ns = vcpu->arch.apic->lapic_timer.timer_advance_ns;
1766 
1767 	/*
1768 	 * If the guest TSC is running at a different ratio than the host, then
1769 	 * convert the delay to nanoseconds to achieve an accurate delay.  Note
1770 	 * that __delay() uses delay_tsc whenever the hardware has TSC, thus
1771 	 * always for VMX enabled hardware.
1772 	 */
1773 	if (vcpu->arch.tsc_scaling_ratio == kvm_caps.default_tsc_scaling_ratio) {
1774 		__delay(min(guest_cycles,
1775 			nsec_to_cycles(vcpu, timer_advance_ns)));
1776 	} else {
1777 		u64 delay_ns = guest_cycles * 1000000ULL;
1778 		do_div(delay_ns, vcpu->arch.virtual_tsc_khz);
1779 		ndelay(min_t(u32, delay_ns, timer_advance_ns));
1780 	}
1781 }
1782 
1783 static inline void adjust_lapic_timer_advance(struct kvm_vcpu *vcpu,
1784 					      s64 advance_expire_delta)
1785 {
1786 	struct kvm_lapic *apic = vcpu->arch.apic;
1787 	u32 timer_advance_ns = apic->lapic_timer.timer_advance_ns;
1788 	u64 ns;
1789 
1790 	/* Do not adjust for tiny fluctuations or large random spikes. */
1791 	if (abs(advance_expire_delta) > LAPIC_TIMER_ADVANCE_ADJUST_MAX ||
1792 	    abs(advance_expire_delta) < LAPIC_TIMER_ADVANCE_ADJUST_MIN)
1793 		return;
1794 
1795 	/* too early */
1796 	if (advance_expire_delta < 0) {
1797 		ns = -advance_expire_delta * 1000000ULL;
1798 		do_div(ns, vcpu->arch.virtual_tsc_khz);
1799 		timer_advance_ns -= ns/LAPIC_TIMER_ADVANCE_ADJUST_STEP;
1800 	} else {
1801 	/* too late */
1802 		ns = advance_expire_delta * 1000000ULL;
1803 		do_div(ns, vcpu->arch.virtual_tsc_khz);
1804 		timer_advance_ns += ns/LAPIC_TIMER_ADVANCE_ADJUST_STEP;
1805 	}
1806 
1807 	if (unlikely(timer_advance_ns > LAPIC_TIMER_ADVANCE_NS_MAX))
1808 		timer_advance_ns = LAPIC_TIMER_ADVANCE_NS_INIT;
1809 	apic->lapic_timer.timer_advance_ns = timer_advance_ns;
1810 }
1811 
1812 static void __kvm_wait_lapic_expire(struct kvm_vcpu *vcpu)
1813 {
1814 	struct kvm_lapic *apic = vcpu->arch.apic;
1815 	u64 guest_tsc, tsc_deadline;
1816 
1817 	tsc_deadline = apic->lapic_timer.expired_tscdeadline;
1818 	apic->lapic_timer.expired_tscdeadline = 0;
1819 	guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
1820 	trace_kvm_wait_lapic_expire(vcpu->vcpu_id, guest_tsc - tsc_deadline);
1821 
1822 	if (lapic_timer_advance_dynamic) {
1823 		adjust_lapic_timer_advance(vcpu, guest_tsc - tsc_deadline);
1824 		/*
1825 		 * If the timer fired early, reread the TSC to account for the
1826 		 * overhead of the above adjustment to avoid waiting longer
1827 		 * than is necessary.
1828 		 */
1829 		if (guest_tsc < tsc_deadline)
1830 			guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
1831 	}
1832 
1833 	if (guest_tsc < tsc_deadline)
1834 		__wait_lapic_expire(vcpu, tsc_deadline - guest_tsc);
1835 }
1836 
1837 void kvm_wait_lapic_expire(struct kvm_vcpu *vcpu)
1838 {
1839 	if (lapic_in_kernel(vcpu) &&
1840 	    vcpu->arch.apic->lapic_timer.expired_tscdeadline &&
1841 	    vcpu->arch.apic->lapic_timer.timer_advance_ns &&
1842 	    lapic_timer_int_injected(vcpu))
1843 		__kvm_wait_lapic_expire(vcpu);
1844 }
1845 EXPORT_SYMBOL_GPL(kvm_wait_lapic_expire);
1846 
1847 static void kvm_apic_inject_pending_timer_irqs(struct kvm_lapic *apic)
1848 {
1849 	struct kvm_timer *ktimer = &apic->lapic_timer;
1850 
1851 	kvm_apic_local_deliver(apic, APIC_LVTT);
1852 	if (apic_lvtt_tscdeadline(apic)) {
1853 		ktimer->tscdeadline = 0;
1854 	} else if (apic_lvtt_oneshot(apic)) {
1855 		ktimer->tscdeadline = 0;
1856 		ktimer->target_expiration = 0;
1857 	}
1858 }
1859 
1860 static void apic_timer_expired(struct kvm_lapic *apic, bool from_timer_fn)
1861 {
1862 	struct kvm_vcpu *vcpu = apic->vcpu;
1863 	struct kvm_timer *ktimer = &apic->lapic_timer;
1864 
1865 	if (atomic_read(&apic->lapic_timer.pending))
1866 		return;
1867 
1868 	if (apic_lvtt_tscdeadline(apic) || ktimer->hv_timer_in_use)
1869 		ktimer->expired_tscdeadline = ktimer->tscdeadline;
1870 
1871 	if (!from_timer_fn && apic->apicv_active) {
1872 		WARN_ON(kvm_get_running_vcpu() != vcpu);
1873 		kvm_apic_inject_pending_timer_irqs(apic);
1874 		return;
1875 	}
1876 
1877 	if (kvm_use_posted_timer_interrupt(apic->vcpu)) {
1878 		/*
1879 		 * Ensure the guest's timer has truly expired before posting an
1880 		 * interrupt.  Open code the relevant checks to avoid querying
1881 		 * lapic_timer_int_injected(), which will be false since the
1882 		 * interrupt isn't yet injected.  Waiting until after injecting
1883 		 * is not an option since that won't help a posted interrupt.
1884 		 */
1885 		if (vcpu->arch.apic->lapic_timer.expired_tscdeadline &&
1886 		    vcpu->arch.apic->lapic_timer.timer_advance_ns)
1887 			__kvm_wait_lapic_expire(vcpu);
1888 		kvm_apic_inject_pending_timer_irqs(apic);
1889 		return;
1890 	}
1891 
1892 	atomic_inc(&apic->lapic_timer.pending);
1893 	kvm_make_request(KVM_REQ_UNBLOCK, vcpu);
1894 	if (from_timer_fn)
1895 		kvm_vcpu_kick(vcpu);
1896 }
1897 
1898 static void start_sw_tscdeadline(struct kvm_lapic *apic)
1899 {
1900 	struct kvm_timer *ktimer = &apic->lapic_timer;
1901 	u64 guest_tsc, tscdeadline = ktimer->tscdeadline;
1902 	u64 ns = 0;
1903 	ktime_t expire;
1904 	struct kvm_vcpu *vcpu = apic->vcpu;
1905 	unsigned long this_tsc_khz = vcpu->arch.virtual_tsc_khz;
1906 	unsigned long flags;
1907 	ktime_t now;
1908 
1909 	if (unlikely(!tscdeadline || !this_tsc_khz))
1910 		return;
1911 
1912 	local_irq_save(flags);
1913 
1914 	now = ktime_get();
1915 	guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
1916 
1917 	ns = (tscdeadline - guest_tsc) * 1000000ULL;
1918 	do_div(ns, this_tsc_khz);
1919 
1920 	if (likely(tscdeadline > guest_tsc) &&
1921 	    likely(ns > apic->lapic_timer.timer_advance_ns)) {
1922 		expire = ktime_add_ns(now, ns);
1923 		expire = ktime_sub_ns(expire, ktimer->timer_advance_ns);
1924 		hrtimer_start(&ktimer->timer, expire, HRTIMER_MODE_ABS_HARD);
1925 	} else
1926 		apic_timer_expired(apic, false);
1927 
1928 	local_irq_restore(flags);
1929 }
1930 
1931 static inline u64 tmict_to_ns(struct kvm_lapic *apic, u32 tmict)
1932 {
1933 	return (u64)tmict * APIC_BUS_CYCLE_NS * (u64)apic->divide_count;
1934 }
1935 
1936 static void update_target_expiration(struct kvm_lapic *apic, uint32_t old_divisor)
1937 {
1938 	ktime_t now, remaining;
1939 	u64 ns_remaining_old, ns_remaining_new;
1940 
1941 	apic->lapic_timer.period =
1942 			tmict_to_ns(apic, kvm_lapic_get_reg(apic, APIC_TMICT));
1943 	limit_periodic_timer_frequency(apic);
1944 
1945 	now = ktime_get();
1946 	remaining = ktime_sub(apic->lapic_timer.target_expiration, now);
1947 	if (ktime_to_ns(remaining) < 0)
1948 		remaining = 0;
1949 
1950 	ns_remaining_old = ktime_to_ns(remaining);
1951 	ns_remaining_new = mul_u64_u32_div(ns_remaining_old,
1952 	                                   apic->divide_count, old_divisor);
1953 
1954 	apic->lapic_timer.tscdeadline +=
1955 		nsec_to_cycles(apic->vcpu, ns_remaining_new) -
1956 		nsec_to_cycles(apic->vcpu, ns_remaining_old);
1957 	apic->lapic_timer.target_expiration = ktime_add_ns(now, ns_remaining_new);
1958 }
1959 
1960 static bool set_target_expiration(struct kvm_lapic *apic, u32 count_reg)
1961 {
1962 	ktime_t now;
1963 	u64 tscl = rdtsc();
1964 	s64 deadline;
1965 
1966 	now = ktime_get();
1967 	apic->lapic_timer.period =
1968 			tmict_to_ns(apic, kvm_lapic_get_reg(apic, APIC_TMICT));
1969 
1970 	if (!apic->lapic_timer.period) {
1971 		apic->lapic_timer.tscdeadline = 0;
1972 		return false;
1973 	}
1974 
1975 	limit_periodic_timer_frequency(apic);
1976 	deadline = apic->lapic_timer.period;
1977 
1978 	if (apic_lvtt_period(apic) || apic_lvtt_oneshot(apic)) {
1979 		if (unlikely(count_reg != APIC_TMICT)) {
1980 			deadline = tmict_to_ns(apic,
1981 				     kvm_lapic_get_reg(apic, count_reg));
1982 			if (unlikely(deadline <= 0)) {
1983 				if (apic_lvtt_period(apic))
1984 					deadline = apic->lapic_timer.period;
1985 				else
1986 					deadline = 0;
1987 			}
1988 			else if (unlikely(deadline > apic->lapic_timer.period)) {
1989 				pr_info_ratelimited(
1990 				    "vcpu %i: requested lapic timer restore with "
1991 				    "starting count register %#x=%u (%lld ns) > initial count (%lld ns). "
1992 				    "Using initial count to start timer.\n",
1993 				    apic->vcpu->vcpu_id,
1994 				    count_reg,
1995 				    kvm_lapic_get_reg(apic, count_reg),
1996 				    deadline, apic->lapic_timer.period);
1997 				kvm_lapic_set_reg(apic, count_reg, 0);
1998 				deadline = apic->lapic_timer.period;
1999 			}
2000 		}
2001 	}
2002 
2003 	apic->lapic_timer.tscdeadline = kvm_read_l1_tsc(apic->vcpu, tscl) +
2004 		nsec_to_cycles(apic->vcpu, deadline);
2005 	apic->lapic_timer.target_expiration = ktime_add_ns(now, deadline);
2006 
2007 	return true;
2008 }
2009 
2010 static void advance_periodic_target_expiration(struct kvm_lapic *apic)
2011 {
2012 	ktime_t now = ktime_get();
2013 	u64 tscl = rdtsc();
2014 	ktime_t delta;
2015 
2016 	/*
2017 	 * Synchronize both deadlines to the same time source or
2018 	 * differences in the periods (caused by differences in the
2019 	 * underlying clocks or numerical approximation errors) will
2020 	 * cause the two to drift apart over time as the errors
2021 	 * accumulate.
2022 	 */
2023 	apic->lapic_timer.target_expiration =
2024 		ktime_add_ns(apic->lapic_timer.target_expiration,
2025 				apic->lapic_timer.period);
2026 	delta = ktime_sub(apic->lapic_timer.target_expiration, now);
2027 	apic->lapic_timer.tscdeadline = kvm_read_l1_tsc(apic->vcpu, tscl) +
2028 		nsec_to_cycles(apic->vcpu, delta);
2029 }
2030 
2031 static void start_sw_period(struct kvm_lapic *apic)
2032 {
2033 	if (!apic->lapic_timer.period)
2034 		return;
2035 
2036 	if (ktime_after(ktime_get(),
2037 			apic->lapic_timer.target_expiration)) {
2038 		apic_timer_expired(apic, false);
2039 
2040 		if (apic_lvtt_oneshot(apic))
2041 			return;
2042 
2043 		advance_periodic_target_expiration(apic);
2044 	}
2045 
2046 	hrtimer_start(&apic->lapic_timer.timer,
2047 		apic->lapic_timer.target_expiration,
2048 		HRTIMER_MODE_ABS_HARD);
2049 }
2050 
2051 bool kvm_lapic_hv_timer_in_use(struct kvm_vcpu *vcpu)
2052 {
2053 	if (!lapic_in_kernel(vcpu))
2054 		return false;
2055 
2056 	return vcpu->arch.apic->lapic_timer.hv_timer_in_use;
2057 }
2058 
2059 static void cancel_hv_timer(struct kvm_lapic *apic)
2060 {
2061 	WARN_ON(preemptible());
2062 	WARN_ON(!apic->lapic_timer.hv_timer_in_use);
2063 	static_call(kvm_x86_cancel_hv_timer)(apic->vcpu);
2064 	apic->lapic_timer.hv_timer_in_use = false;
2065 }
2066 
2067 static bool start_hv_timer(struct kvm_lapic *apic)
2068 {
2069 	struct kvm_timer *ktimer = &apic->lapic_timer;
2070 	struct kvm_vcpu *vcpu = apic->vcpu;
2071 	bool expired;
2072 
2073 	WARN_ON(preemptible());
2074 	if (!kvm_can_use_hv_timer(vcpu))
2075 		return false;
2076 
2077 	if (!ktimer->tscdeadline)
2078 		return false;
2079 
2080 	if (static_call(kvm_x86_set_hv_timer)(vcpu, ktimer->tscdeadline, &expired))
2081 		return false;
2082 
2083 	ktimer->hv_timer_in_use = true;
2084 	hrtimer_cancel(&ktimer->timer);
2085 
2086 	/*
2087 	 * To simplify handling the periodic timer, leave the hv timer running
2088 	 * even if the deadline timer has expired, i.e. rely on the resulting
2089 	 * VM-Exit to recompute the periodic timer's target expiration.
2090 	 */
2091 	if (!apic_lvtt_period(apic)) {
2092 		/*
2093 		 * Cancel the hv timer if the sw timer fired while the hv timer
2094 		 * was being programmed, or if the hv timer itself expired.
2095 		 */
2096 		if (atomic_read(&ktimer->pending)) {
2097 			cancel_hv_timer(apic);
2098 		} else if (expired) {
2099 			apic_timer_expired(apic, false);
2100 			cancel_hv_timer(apic);
2101 		}
2102 	}
2103 
2104 	trace_kvm_hv_timer_state(vcpu->vcpu_id, ktimer->hv_timer_in_use);
2105 
2106 	return true;
2107 }
2108 
2109 static void start_sw_timer(struct kvm_lapic *apic)
2110 {
2111 	struct kvm_timer *ktimer = &apic->lapic_timer;
2112 
2113 	WARN_ON(preemptible());
2114 	if (apic->lapic_timer.hv_timer_in_use)
2115 		cancel_hv_timer(apic);
2116 	if (!apic_lvtt_period(apic) && atomic_read(&ktimer->pending))
2117 		return;
2118 
2119 	if (apic_lvtt_period(apic) || apic_lvtt_oneshot(apic))
2120 		start_sw_period(apic);
2121 	else if (apic_lvtt_tscdeadline(apic))
2122 		start_sw_tscdeadline(apic);
2123 	trace_kvm_hv_timer_state(apic->vcpu->vcpu_id, false);
2124 }
2125 
2126 static void restart_apic_timer(struct kvm_lapic *apic)
2127 {
2128 	preempt_disable();
2129 
2130 	if (!apic_lvtt_period(apic) && atomic_read(&apic->lapic_timer.pending))
2131 		goto out;
2132 
2133 	if (!start_hv_timer(apic))
2134 		start_sw_timer(apic);
2135 out:
2136 	preempt_enable();
2137 }
2138 
2139 void kvm_lapic_expired_hv_timer(struct kvm_vcpu *vcpu)
2140 {
2141 	struct kvm_lapic *apic = vcpu->arch.apic;
2142 
2143 	preempt_disable();
2144 	/* If the preempt notifier has already run, it also called apic_timer_expired */
2145 	if (!apic->lapic_timer.hv_timer_in_use)
2146 		goto out;
2147 	WARN_ON(kvm_vcpu_is_blocking(vcpu));
2148 	apic_timer_expired(apic, false);
2149 	cancel_hv_timer(apic);
2150 
2151 	if (apic_lvtt_period(apic) && apic->lapic_timer.period) {
2152 		advance_periodic_target_expiration(apic);
2153 		restart_apic_timer(apic);
2154 	}
2155 out:
2156 	preempt_enable();
2157 }
2158 EXPORT_SYMBOL_GPL(kvm_lapic_expired_hv_timer);
2159 
2160 void kvm_lapic_switch_to_hv_timer(struct kvm_vcpu *vcpu)
2161 {
2162 	restart_apic_timer(vcpu->arch.apic);
2163 }
2164 
2165 void kvm_lapic_switch_to_sw_timer(struct kvm_vcpu *vcpu)
2166 {
2167 	struct kvm_lapic *apic = vcpu->arch.apic;
2168 
2169 	preempt_disable();
2170 	/* Possibly the TSC deadline timer is not enabled yet */
2171 	if (apic->lapic_timer.hv_timer_in_use)
2172 		start_sw_timer(apic);
2173 	preempt_enable();
2174 }
2175 
2176 void kvm_lapic_restart_hv_timer(struct kvm_vcpu *vcpu)
2177 {
2178 	struct kvm_lapic *apic = vcpu->arch.apic;
2179 
2180 	WARN_ON(!apic->lapic_timer.hv_timer_in_use);
2181 	restart_apic_timer(apic);
2182 }
2183 
2184 static void __start_apic_timer(struct kvm_lapic *apic, u32 count_reg)
2185 {
2186 	atomic_set(&apic->lapic_timer.pending, 0);
2187 
2188 	if ((apic_lvtt_period(apic) || apic_lvtt_oneshot(apic))
2189 	    && !set_target_expiration(apic, count_reg))
2190 		return;
2191 
2192 	restart_apic_timer(apic);
2193 }
2194 
2195 static void start_apic_timer(struct kvm_lapic *apic)
2196 {
2197 	__start_apic_timer(apic, APIC_TMICT);
2198 }
2199 
2200 static void apic_manage_nmi_watchdog(struct kvm_lapic *apic, u32 lvt0_val)
2201 {
2202 	bool lvt0_in_nmi_mode = apic_lvt_nmi_mode(lvt0_val);
2203 
2204 	if (apic->lvt0_in_nmi_mode != lvt0_in_nmi_mode) {
2205 		apic->lvt0_in_nmi_mode = lvt0_in_nmi_mode;
2206 		if (lvt0_in_nmi_mode) {
2207 			atomic_inc(&apic->vcpu->kvm->arch.vapics_in_nmi_mode);
2208 		} else
2209 			atomic_dec(&apic->vcpu->kvm->arch.vapics_in_nmi_mode);
2210 	}
2211 }
2212 
2213 static int get_lvt_index(u32 reg)
2214 {
2215 	if (reg == APIC_LVTCMCI)
2216 		return LVT_CMCI;
2217 	if (reg < APIC_LVTT || reg > APIC_LVTERR)
2218 		return -1;
2219 	return array_index_nospec(
2220 			(reg - APIC_LVTT) >> 4, KVM_APIC_MAX_NR_LVT_ENTRIES);
2221 }
2222 
2223 static int kvm_lapic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val)
2224 {
2225 	int ret = 0;
2226 
2227 	trace_kvm_apic_write(reg, val);
2228 
2229 	switch (reg) {
2230 	case APIC_ID:		/* Local APIC ID */
2231 		if (!apic_x2apic_mode(apic)) {
2232 			kvm_apic_set_xapic_id(apic, val >> 24);
2233 		} else {
2234 			ret = 1;
2235 		}
2236 		break;
2237 
2238 	case APIC_TASKPRI:
2239 		report_tpr_access(apic, true);
2240 		apic_set_tpr(apic, val & 0xff);
2241 		break;
2242 
2243 	case APIC_EOI:
2244 		apic_set_eoi(apic);
2245 		break;
2246 
2247 	case APIC_LDR:
2248 		if (!apic_x2apic_mode(apic))
2249 			kvm_apic_set_ldr(apic, val & APIC_LDR_MASK);
2250 		else
2251 			ret = 1;
2252 		break;
2253 
2254 	case APIC_DFR:
2255 		if (!apic_x2apic_mode(apic))
2256 			kvm_apic_set_dfr(apic, val | 0x0FFFFFFF);
2257 		else
2258 			ret = 1;
2259 		break;
2260 
2261 	case APIC_SPIV: {
2262 		u32 mask = 0x3ff;
2263 		if (kvm_lapic_get_reg(apic, APIC_LVR) & APIC_LVR_DIRECTED_EOI)
2264 			mask |= APIC_SPIV_DIRECTED_EOI;
2265 		apic_set_spiv(apic, val & mask);
2266 		if (!(val & APIC_SPIV_APIC_ENABLED)) {
2267 			int i;
2268 
2269 			for (i = 0; i < apic->nr_lvt_entries; i++) {
2270 				kvm_lapic_set_reg(apic, APIC_LVTx(i),
2271 					kvm_lapic_get_reg(apic, APIC_LVTx(i)) | APIC_LVT_MASKED);
2272 			}
2273 			apic_update_lvtt(apic);
2274 			atomic_set(&apic->lapic_timer.pending, 0);
2275 
2276 		}
2277 		break;
2278 	}
2279 	case APIC_ICR:
2280 		WARN_ON_ONCE(apic_x2apic_mode(apic));
2281 
2282 		/* No delay here, so we always clear the pending bit */
2283 		val &= ~APIC_ICR_BUSY;
2284 		kvm_apic_send_ipi(apic, val, kvm_lapic_get_reg(apic, APIC_ICR2));
2285 		kvm_lapic_set_reg(apic, APIC_ICR, val);
2286 		break;
2287 	case APIC_ICR2:
2288 		if (apic_x2apic_mode(apic))
2289 			ret = 1;
2290 		else
2291 			kvm_lapic_set_reg(apic, APIC_ICR2, val & 0xff000000);
2292 		break;
2293 
2294 	case APIC_LVT0:
2295 		apic_manage_nmi_watchdog(apic, val);
2296 		fallthrough;
2297 	case APIC_LVTTHMR:
2298 	case APIC_LVTPC:
2299 	case APIC_LVT1:
2300 	case APIC_LVTERR:
2301 	case APIC_LVTCMCI: {
2302 		u32 index = get_lvt_index(reg);
2303 		if (!kvm_lapic_lvt_supported(apic, index)) {
2304 			ret = 1;
2305 			break;
2306 		}
2307 		if (!kvm_apic_sw_enabled(apic))
2308 			val |= APIC_LVT_MASKED;
2309 		val &= apic_lvt_mask[index];
2310 		kvm_lapic_set_reg(apic, reg, val);
2311 		break;
2312 	}
2313 
2314 	case APIC_LVTT:
2315 		if (!kvm_apic_sw_enabled(apic))
2316 			val |= APIC_LVT_MASKED;
2317 		val &= (apic_lvt_mask[0] | apic->lapic_timer.timer_mode_mask);
2318 		kvm_lapic_set_reg(apic, APIC_LVTT, val);
2319 		apic_update_lvtt(apic);
2320 		break;
2321 
2322 	case APIC_TMICT:
2323 		if (apic_lvtt_tscdeadline(apic))
2324 			break;
2325 
2326 		cancel_apic_timer(apic);
2327 		kvm_lapic_set_reg(apic, APIC_TMICT, val);
2328 		start_apic_timer(apic);
2329 		break;
2330 
2331 	case APIC_TDCR: {
2332 		uint32_t old_divisor = apic->divide_count;
2333 
2334 		kvm_lapic_set_reg(apic, APIC_TDCR, val & 0xb);
2335 		update_divide_count(apic);
2336 		if (apic->divide_count != old_divisor &&
2337 				apic->lapic_timer.period) {
2338 			hrtimer_cancel(&apic->lapic_timer.timer);
2339 			update_target_expiration(apic, old_divisor);
2340 			restart_apic_timer(apic);
2341 		}
2342 		break;
2343 	}
2344 	case APIC_ESR:
2345 		if (apic_x2apic_mode(apic) && val != 0)
2346 			ret = 1;
2347 		break;
2348 
2349 	case APIC_SELF_IPI:
2350 		/*
2351 		 * Self-IPI exists only when x2APIC is enabled.  Bits 7:0 hold
2352 		 * the vector, everything else is reserved.
2353 		 */
2354 		if (!apic_x2apic_mode(apic) || (val & ~APIC_VECTOR_MASK))
2355 			ret = 1;
2356 		else
2357 			kvm_apic_send_ipi(apic, APIC_DEST_SELF | val, 0);
2358 		break;
2359 	default:
2360 		ret = 1;
2361 		break;
2362 	}
2363 
2364 	/*
2365 	 * Recalculate APIC maps if necessary, e.g. if the software enable bit
2366 	 * was toggled, the APIC ID changed, etc...   The maps are marked dirty
2367 	 * on relevant changes, i.e. this is a nop for most writes.
2368 	 */
2369 	kvm_recalculate_apic_map(apic->vcpu->kvm);
2370 
2371 	return ret;
2372 }
2373 
2374 static int apic_mmio_write(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
2375 			    gpa_t address, int len, const void *data)
2376 {
2377 	struct kvm_lapic *apic = to_lapic(this);
2378 	unsigned int offset = address - apic->base_address;
2379 	u32 val;
2380 
2381 	if (!apic_mmio_in_range(apic, address))
2382 		return -EOPNOTSUPP;
2383 
2384 	if (!kvm_apic_hw_enabled(apic) || apic_x2apic_mode(apic)) {
2385 		if (!kvm_check_has_quirk(vcpu->kvm,
2386 					 KVM_X86_QUIRK_LAPIC_MMIO_HOLE))
2387 			return -EOPNOTSUPP;
2388 
2389 		return 0;
2390 	}
2391 
2392 	/*
2393 	 * APIC register must be aligned on 128-bits boundary.
2394 	 * 32/64/128 bits registers must be accessed thru 32 bits.
2395 	 * Refer SDM 8.4.1
2396 	 */
2397 	if (len != 4 || (offset & 0xf))
2398 		return 0;
2399 
2400 	val = *(u32*)data;
2401 
2402 	kvm_lapic_reg_write(apic, offset & 0xff0, val);
2403 
2404 	return 0;
2405 }
2406 
2407 void kvm_lapic_set_eoi(struct kvm_vcpu *vcpu)
2408 {
2409 	kvm_lapic_reg_write(vcpu->arch.apic, APIC_EOI, 0);
2410 }
2411 EXPORT_SYMBOL_GPL(kvm_lapic_set_eoi);
2412 
2413 /* emulate APIC access in a trap manner */
2414 void kvm_apic_write_nodecode(struct kvm_vcpu *vcpu, u32 offset)
2415 {
2416 	struct kvm_lapic *apic = vcpu->arch.apic;
2417 	u64 val;
2418 
2419 	/*
2420 	 * ICR is a single 64-bit register when x2APIC is enabled.  For legacy
2421 	 * xAPIC, ICR writes need to go down the common (slightly slower) path
2422 	 * to get the upper half from ICR2.
2423 	 */
2424 	if (apic_x2apic_mode(apic) && offset == APIC_ICR) {
2425 		val = kvm_lapic_get_reg64(apic, APIC_ICR);
2426 		kvm_apic_send_ipi(apic, (u32)val, (u32)(val >> 32));
2427 		trace_kvm_apic_write(APIC_ICR, val);
2428 	} else {
2429 		/* TODO: optimize to just emulate side effect w/o one more write */
2430 		val = kvm_lapic_get_reg(apic, offset);
2431 		kvm_lapic_reg_write(apic, offset, (u32)val);
2432 	}
2433 }
2434 EXPORT_SYMBOL_GPL(kvm_apic_write_nodecode);
2435 
2436 void kvm_free_lapic(struct kvm_vcpu *vcpu)
2437 {
2438 	struct kvm_lapic *apic = vcpu->arch.apic;
2439 
2440 	if (!vcpu->arch.apic)
2441 		return;
2442 
2443 	hrtimer_cancel(&apic->lapic_timer.timer);
2444 
2445 	if (!(vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE))
2446 		static_branch_slow_dec_deferred(&apic_hw_disabled);
2447 
2448 	if (!apic->sw_enabled)
2449 		static_branch_slow_dec_deferred(&apic_sw_disabled);
2450 
2451 	if (apic->regs)
2452 		free_page((unsigned long)apic->regs);
2453 
2454 	kfree(apic);
2455 }
2456 
2457 /*
2458  *----------------------------------------------------------------------
2459  * LAPIC interface
2460  *----------------------------------------------------------------------
2461  */
2462 u64 kvm_get_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu)
2463 {
2464 	struct kvm_lapic *apic = vcpu->arch.apic;
2465 
2466 	if (!kvm_apic_present(vcpu) || !apic_lvtt_tscdeadline(apic))
2467 		return 0;
2468 
2469 	return apic->lapic_timer.tscdeadline;
2470 }
2471 
2472 void kvm_set_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu, u64 data)
2473 {
2474 	struct kvm_lapic *apic = vcpu->arch.apic;
2475 
2476 	if (!kvm_apic_present(vcpu) || !apic_lvtt_tscdeadline(apic))
2477 		return;
2478 
2479 	hrtimer_cancel(&apic->lapic_timer.timer);
2480 	apic->lapic_timer.tscdeadline = data;
2481 	start_apic_timer(apic);
2482 }
2483 
2484 void kvm_lapic_set_tpr(struct kvm_vcpu *vcpu, unsigned long cr8)
2485 {
2486 	apic_set_tpr(vcpu->arch.apic, (cr8 & 0x0f) << 4);
2487 }
2488 
2489 u64 kvm_lapic_get_cr8(struct kvm_vcpu *vcpu)
2490 {
2491 	u64 tpr;
2492 
2493 	tpr = (u64) kvm_lapic_get_reg(vcpu->arch.apic, APIC_TASKPRI);
2494 
2495 	return (tpr & 0xf0) >> 4;
2496 }
2497 
2498 void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
2499 {
2500 	u64 old_value = vcpu->arch.apic_base;
2501 	struct kvm_lapic *apic = vcpu->arch.apic;
2502 
2503 	vcpu->arch.apic_base = value;
2504 
2505 	if ((old_value ^ value) & MSR_IA32_APICBASE_ENABLE)
2506 		kvm_update_cpuid_runtime(vcpu);
2507 
2508 	if (!apic)
2509 		return;
2510 
2511 	/* update jump label if enable bit changes */
2512 	if ((old_value ^ value) & MSR_IA32_APICBASE_ENABLE) {
2513 		if (value & MSR_IA32_APICBASE_ENABLE) {
2514 			kvm_apic_set_xapic_id(apic, vcpu->vcpu_id);
2515 			static_branch_slow_dec_deferred(&apic_hw_disabled);
2516 			/* Check if there are APF page ready requests pending */
2517 			kvm_make_request(KVM_REQ_APF_READY, vcpu);
2518 		} else {
2519 			static_branch_inc(&apic_hw_disabled.key);
2520 			atomic_set_release(&apic->vcpu->kvm->arch.apic_map_dirty, DIRTY);
2521 		}
2522 	}
2523 
2524 	if ((old_value ^ value) & X2APIC_ENABLE) {
2525 		if (value & X2APIC_ENABLE)
2526 			kvm_apic_set_x2apic_id(apic, vcpu->vcpu_id);
2527 		else if (value & MSR_IA32_APICBASE_ENABLE)
2528 			kvm_apic_set_xapic_id(apic, vcpu->vcpu_id);
2529 	}
2530 
2531 	if ((old_value ^ value) & (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE)) {
2532 		kvm_make_request(KVM_REQ_APICV_UPDATE, vcpu);
2533 		static_call_cond(kvm_x86_set_virtual_apic_mode)(vcpu);
2534 	}
2535 
2536 	apic->base_address = apic->vcpu->arch.apic_base &
2537 			     MSR_IA32_APICBASE_BASE;
2538 
2539 	if ((value & MSR_IA32_APICBASE_ENABLE) &&
2540 	     apic->base_address != APIC_DEFAULT_PHYS_BASE) {
2541 		kvm_set_apicv_inhibit(apic->vcpu->kvm,
2542 				      APICV_INHIBIT_REASON_APIC_BASE_MODIFIED);
2543 	}
2544 }
2545 
2546 void kvm_apic_update_apicv(struct kvm_vcpu *vcpu)
2547 {
2548 	struct kvm_lapic *apic = vcpu->arch.apic;
2549 
2550 	if (apic->apicv_active) {
2551 		/* irr_pending is always true when apicv is activated. */
2552 		apic->irr_pending = true;
2553 		apic->isr_count = 1;
2554 	} else {
2555 		/*
2556 		 * Don't clear irr_pending, searching the IRR can race with
2557 		 * updates from the CPU as APICv is still active from hardware's
2558 		 * perspective.  The flag will be cleared as appropriate when
2559 		 * KVM injects the interrupt.
2560 		 */
2561 		apic->isr_count = count_vectors(apic->regs + APIC_ISR);
2562 	}
2563 	apic->highest_isr_cache = -1;
2564 }
2565 
2566 int kvm_alloc_apic_access_page(struct kvm *kvm)
2567 {
2568 	struct page *page;
2569 	void __user *hva;
2570 	int ret = 0;
2571 
2572 	mutex_lock(&kvm->slots_lock);
2573 	if (kvm->arch.apic_access_memslot_enabled ||
2574 	    kvm->arch.apic_access_memslot_inhibited)
2575 		goto out;
2576 
2577 	hva = __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
2578 				      APIC_DEFAULT_PHYS_BASE, PAGE_SIZE);
2579 	if (IS_ERR(hva)) {
2580 		ret = PTR_ERR(hva);
2581 		goto out;
2582 	}
2583 
2584 	page = gfn_to_page(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
2585 	if (is_error_page(page)) {
2586 		ret = -EFAULT;
2587 		goto out;
2588 	}
2589 
2590 	/*
2591 	 * Do not pin the page in memory, so that memory hot-unplug
2592 	 * is able to migrate it.
2593 	 */
2594 	put_page(page);
2595 	kvm->arch.apic_access_memslot_enabled = true;
2596 out:
2597 	mutex_unlock(&kvm->slots_lock);
2598 	return ret;
2599 }
2600 EXPORT_SYMBOL_GPL(kvm_alloc_apic_access_page);
2601 
2602 void kvm_inhibit_apic_access_page(struct kvm_vcpu *vcpu)
2603 {
2604 	struct kvm *kvm = vcpu->kvm;
2605 
2606 	if (!kvm->arch.apic_access_memslot_enabled)
2607 		return;
2608 
2609 	kvm_vcpu_srcu_read_unlock(vcpu);
2610 
2611 	mutex_lock(&kvm->slots_lock);
2612 
2613 	if (kvm->arch.apic_access_memslot_enabled) {
2614 		__x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT, 0, 0);
2615 		/*
2616 		 * Clear "enabled" after the memslot is deleted so that a
2617 		 * different vCPU doesn't get a false negative when checking
2618 		 * the flag out of slots_lock.  No additional memory barrier is
2619 		 * needed as modifying memslots requires waiting other vCPUs to
2620 		 * drop SRCU (see above), and false positives are ok as the
2621 		 * flag is rechecked after acquiring slots_lock.
2622 		 */
2623 		kvm->arch.apic_access_memslot_enabled = false;
2624 
2625 		/*
2626 		 * Mark the memslot as inhibited to prevent reallocating the
2627 		 * memslot during vCPU creation, e.g. if a vCPU is hotplugged.
2628 		 */
2629 		kvm->arch.apic_access_memslot_inhibited = true;
2630 	}
2631 
2632 	mutex_unlock(&kvm->slots_lock);
2633 
2634 	kvm_vcpu_srcu_read_lock(vcpu);
2635 }
2636 
2637 void kvm_lapic_reset(struct kvm_vcpu *vcpu, bool init_event)
2638 {
2639 	struct kvm_lapic *apic = vcpu->arch.apic;
2640 	u64 msr_val;
2641 	int i;
2642 
2643 	if (!init_event) {
2644 		msr_val = APIC_DEFAULT_PHYS_BASE | MSR_IA32_APICBASE_ENABLE;
2645 		if (kvm_vcpu_is_reset_bsp(vcpu))
2646 			msr_val |= MSR_IA32_APICBASE_BSP;
2647 		kvm_lapic_set_base(vcpu, msr_val);
2648 	}
2649 
2650 	if (!apic)
2651 		return;
2652 
2653 	/* Stop the timer in case it's a reset to an active apic */
2654 	hrtimer_cancel(&apic->lapic_timer.timer);
2655 
2656 	/* The xAPIC ID is set at RESET even if the APIC was already enabled. */
2657 	if (!init_event)
2658 		kvm_apic_set_xapic_id(apic, vcpu->vcpu_id);
2659 	kvm_apic_set_version(apic->vcpu);
2660 
2661 	for (i = 0; i < apic->nr_lvt_entries; i++)
2662 		kvm_lapic_set_reg(apic, APIC_LVTx(i), APIC_LVT_MASKED);
2663 	apic_update_lvtt(apic);
2664 	if (kvm_vcpu_is_reset_bsp(vcpu) &&
2665 	    kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_LINT0_REENABLED))
2666 		kvm_lapic_set_reg(apic, APIC_LVT0,
2667 			     SET_APIC_DELIVERY_MODE(0, APIC_MODE_EXTINT));
2668 	apic_manage_nmi_watchdog(apic, kvm_lapic_get_reg(apic, APIC_LVT0));
2669 
2670 	kvm_apic_set_dfr(apic, 0xffffffffU);
2671 	apic_set_spiv(apic, 0xff);
2672 	kvm_lapic_set_reg(apic, APIC_TASKPRI, 0);
2673 	if (!apic_x2apic_mode(apic))
2674 		kvm_apic_set_ldr(apic, 0);
2675 	kvm_lapic_set_reg(apic, APIC_ESR, 0);
2676 	if (!apic_x2apic_mode(apic)) {
2677 		kvm_lapic_set_reg(apic, APIC_ICR, 0);
2678 		kvm_lapic_set_reg(apic, APIC_ICR2, 0);
2679 	} else {
2680 		kvm_lapic_set_reg64(apic, APIC_ICR, 0);
2681 	}
2682 	kvm_lapic_set_reg(apic, APIC_TDCR, 0);
2683 	kvm_lapic_set_reg(apic, APIC_TMICT, 0);
2684 	for (i = 0; i < 8; i++) {
2685 		kvm_lapic_set_reg(apic, APIC_IRR + 0x10 * i, 0);
2686 		kvm_lapic_set_reg(apic, APIC_ISR + 0x10 * i, 0);
2687 		kvm_lapic_set_reg(apic, APIC_TMR + 0x10 * i, 0);
2688 	}
2689 	kvm_apic_update_apicv(vcpu);
2690 	update_divide_count(apic);
2691 	atomic_set(&apic->lapic_timer.pending, 0);
2692 
2693 	vcpu->arch.pv_eoi.msr_val = 0;
2694 	apic_update_ppr(apic);
2695 	if (apic->apicv_active) {
2696 		static_call_cond(kvm_x86_apicv_post_state_restore)(vcpu);
2697 		static_call_cond(kvm_x86_hwapic_irr_update)(vcpu, -1);
2698 		static_call_cond(kvm_x86_hwapic_isr_update)(-1);
2699 	}
2700 
2701 	vcpu->arch.apic_arb_prio = 0;
2702 	vcpu->arch.apic_attention = 0;
2703 
2704 	kvm_recalculate_apic_map(vcpu->kvm);
2705 }
2706 
2707 /*
2708  *----------------------------------------------------------------------
2709  * timer interface
2710  *----------------------------------------------------------------------
2711  */
2712 
2713 static bool lapic_is_periodic(struct kvm_lapic *apic)
2714 {
2715 	return apic_lvtt_period(apic);
2716 }
2717 
2718 int apic_has_pending_timer(struct kvm_vcpu *vcpu)
2719 {
2720 	struct kvm_lapic *apic = vcpu->arch.apic;
2721 
2722 	if (apic_enabled(apic) && apic_lvt_enabled(apic, APIC_LVTT))
2723 		return atomic_read(&apic->lapic_timer.pending);
2724 
2725 	return 0;
2726 }
2727 
2728 int kvm_apic_local_deliver(struct kvm_lapic *apic, int lvt_type)
2729 {
2730 	u32 reg = kvm_lapic_get_reg(apic, lvt_type);
2731 	int vector, mode, trig_mode;
2732 
2733 	if (kvm_apic_hw_enabled(apic) && !(reg & APIC_LVT_MASKED)) {
2734 		vector = reg & APIC_VECTOR_MASK;
2735 		mode = reg & APIC_MODE_MASK;
2736 		trig_mode = reg & APIC_LVT_LEVEL_TRIGGER;
2737 		return __apic_accept_irq(apic, mode, vector, 1, trig_mode,
2738 					NULL);
2739 	}
2740 	return 0;
2741 }
2742 
2743 void kvm_apic_nmi_wd_deliver(struct kvm_vcpu *vcpu)
2744 {
2745 	struct kvm_lapic *apic = vcpu->arch.apic;
2746 
2747 	if (apic)
2748 		kvm_apic_local_deliver(apic, APIC_LVT0);
2749 }
2750 
2751 static const struct kvm_io_device_ops apic_mmio_ops = {
2752 	.read     = apic_mmio_read,
2753 	.write    = apic_mmio_write,
2754 };
2755 
2756 static enum hrtimer_restart apic_timer_fn(struct hrtimer *data)
2757 {
2758 	struct kvm_timer *ktimer = container_of(data, struct kvm_timer, timer);
2759 	struct kvm_lapic *apic = container_of(ktimer, struct kvm_lapic, lapic_timer);
2760 
2761 	apic_timer_expired(apic, true);
2762 
2763 	if (lapic_is_periodic(apic)) {
2764 		advance_periodic_target_expiration(apic);
2765 		hrtimer_add_expires_ns(&ktimer->timer, ktimer->period);
2766 		return HRTIMER_RESTART;
2767 	} else
2768 		return HRTIMER_NORESTART;
2769 }
2770 
2771 int kvm_create_lapic(struct kvm_vcpu *vcpu, int timer_advance_ns)
2772 {
2773 	struct kvm_lapic *apic;
2774 
2775 	ASSERT(vcpu != NULL);
2776 
2777 	apic = kzalloc(sizeof(*apic), GFP_KERNEL_ACCOUNT);
2778 	if (!apic)
2779 		goto nomem;
2780 
2781 	vcpu->arch.apic = apic;
2782 
2783 	apic->regs = (void *)get_zeroed_page(GFP_KERNEL_ACCOUNT);
2784 	if (!apic->regs) {
2785 		printk(KERN_ERR "malloc apic regs error for vcpu %x\n",
2786 		       vcpu->vcpu_id);
2787 		goto nomem_free_apic;
2788 	}
2789 	apic->vcpu = vcpu;
2790 
2791 	apic->nr_lvt_entries = kvm_apic_calc_nr_lvt_entries(vcpu);
2792 
2793 	hrtimer_init(&apic->lapic_timer.timer, CLOCK_MONOTONIC,
2794 		     HRTIMER_MODE_ABS_HARD);
2795 	apic->lapic_timer.timer.function = apic_timer_fn;
2796 	if (timer_advance_ns == -1) {
2797 		apic->lapic_timer.timer_advance_ns = LAPIC_TIMER_ADVANCE_NS_INIT;
2798 		lapic_timer_advance_dynamic = true;
2799 	} else {
2800 		apic->lapic_timer.timer_advance_ns = timer_advance_ns;
2801 		lapic_timer_advance_dynamic = false;
2802 	}
2803 
2804 	/*
2805 	 * Stuff the APIC ENABLE bit in lieu of temporarily incrementing
2806 	 * apic_hw_disabled; the full RESET value is set by kvm_lapic_reset().
2807 	 */
2808 	vcpu->arch.apic_base = MSR_IA32_APICBASE_ENABLE;
2809 	static_branch_inc(&apic_sw_disabled.key); /* sw disabled at reset */
2810 	kvm_iodevice_init(&apic->dev, &apic_mmio_ops);
2811 
2812 	return 0;
2813 nomem_free_apic:
2814 	kfree(apic);
2815 	vcpu->arch.apic = NULL;
2816 nomem:
2817 	return -ENOMEM;
2818 }
2819 
2820 int kvm_apic_has_interrupt(struct kvm_vcpu *vcpu)
2821 {
2822 	struct kvm_lapic *apic = vcpu->arch.apic;
2823 	u32 ppr;
2824 
2825 	if (!kvm_apic_present(vcpu))
2826 		return -1;
2827 
2828 	__apic_update_ppr(apic, &ppr);
2829 	return apic_has_interrupt_for_ppr(apic, ppr);
2830 }
2831 EXPORT_SYMBOL_GPL(kvm_apic_has_interrupt);
2832 
2833 int kvm_apic_accept_pic_intr(struct kvm_vcpu *vcpu)
2834 {
2835 	u32 lvt0 = kvm_lapic_get_reg(vcpu->arch.apic, APIC_LVT0);
2836 
2837 	if (!kvm_apic_hw_enabled(vcpu->arch.apic))
2838 		return 1;
2839 	if ((lvt0 & APIC_LVT_MASKED) == 0 &&
2840 	    GET_APIC_DELIVERY_MODE(lvt0) == APIC_MODE_EXTINT)
2841 		return 1;
2842 	return 0;
2843 }
2844 
2845 void kvm_inject_apic_timer_irqs(struct kvm_vcpu *vcpu)
2846 {
2847 	struct kvm_lapic *apic = vcpu->arch.apic;
2848 
2849 	if (atomic_read(&apic->lapic_timer.pending) > 0) {
2850 		kvm_apic_inject_pending_timer_irqs(apic);
2851 		atomic_set(&apic->lapic_timer.pending, 0);
2852 	}
2853 }
2854 
2855 int kvm_get_apic_interrupt(struct kvm_vcpu *vcpu)
2856 {
2857 	int vector = kvm_apic_has_interrupt(vcpu);
2858 	struct kvm_lapic *apic = vcpu->arch.apic;
2859 	u32 ppr;
2860 
2861 	if (vector == -1)
2862 		return -1;
2863 
2864 	/*
2865 	 * We get here even with APIC virtualization enabled, if doing
2866 	 * nested virtualization and L1 runs with the "acknowledge interrupt
2867 	 * on exit" mode.  Then we cannot inject the interrupt via RVI,
2868 	 * because the process would deliver it through the IDT.
2869 	 */
2870 
2871 	apic_clear_irr(vector, apic);
2872 	if (to_hv_vcpu(vcpu) && test_bit(vector, to_hv_synic(vcpu)->auto_eoi_bitmap)) {
2873 		/*
2874 		 * For auto-EOI interrupts, there might be another pending
2875 		 * interrupt above PPR, so check whether to raise another
2876 		 * KVM_REQ_EVENT.
2877 		 */
2878 		apic_update_ppr(apic);
2879 	} else {
2880 		/*
2881 		 * For normal interrupts, PPR has been raised and there cannot
2882 		 * be a higher-priority pending interrupt---except if there was
2883 		 * a concurrent interrupt injection, but that would have
2884 		 * triggered KVM_REQ_EVENT already.
2885 		 */
2886 		apic_set_isr(vector, apic);
2887 		__apic_update_ppr(apic, &ppr);
2888 	}
2889 
2890 	return vector;
2891 }
2892 
2893 static int kvm_apic_state_fixup(struct kvm_vcpu *vcpu,
2894 		struct kvm_lapic_state *s, bool set)
2895 {
2896 	if (apic_x2apic_mode(vcpu->arch.apic)) {
2897 		u32 *id = (u32 *)(s->regs + APIC_ID);
2898 		u32 *ldr = (u32 *)(s->regs + APIC_LDR);
2899 		u64 icr;
2900 
2901 		if (vcpu->kvm->arch.x2apic_format) {
2902 			if (*id != vcpu->vcpu_id)
2903 				return -EINVAL;
2904 		} else {
2905 			if (set)
2906 				*id >>= 24;
2907 			else
2908 				*id <<= 24;
2909 		}
2910 
2911 		/*
2912 		 * In x2APIC mode, the LDR is fixed and based on the id.  And
2913 		 * ICR is internally a single 64-bit register, but needs to be
2914 		 * split to ICR+ICR2 in userspace for backwards compatibility.
2915 		 */
2916 		if (set) {
2917 			*ldr = kvm_apic_calc_x2apic_ldr(*id);
2918 
2919 			icr = __kvm_lapic_get_reg(s->regs, APIC_ICR) |
2920 			      (u64)__kvm_lapic_get_reg(s->regs, APIC_ICR2) << 32;
2921 			__kvm_lapic_set_reg64(s->regs, APIC_ICR, icr);
2922 		} else {
2923 			icr = __kvm_lapic_get_reg64(s->regs, APIC_ICR);
2924 			__kvm_lapic_set_reg(s->regs, APIC_ICR2, icr >> 32);
2925 		}
2926 	}
2927 
2928 	return 0;
2929 }
2930 
2931 int kvm_apic_get_state(struct kvm_vcpu *vcpu, struct kvm_lapic_state *s)
2932 {
2933 	memcpy(s->regs, vcpu->arch.apic->regs, sizeof(*s));
2934 
2935 	/*
2936 	 * Get calculated timer current count for remaining timer period (if
2937 	 * any) and store it in the returned register set.
2938 	 */
2939 	__kvm_lapic_set_reg(s->regs, APIC_TMCCT,
2940 			    __apic_read(vcpu->arch.apic, APIC_TMCCT));
2941 
2942 	return kvm_apic_state_fixup(vcpu, s, false);
2943 }
2944 
2945 int kvm_apic_set_state(struct kvm_vcpu *vcpu, struct kvm_lapic_state *s)
2946 {
2947 	struct kvm_lapic *apic = vcpu->arch.apic;
2948 	int r;
2949 
2950 	kvm_lapic_set_base(vcpu, vcpu->arch.apic_base);
2951 	/* set SPIV separately to get count of SW disabled APICs right */
2952 	apic_set_spiv(apic, *((u32 *)(s->regs + APIC_SPIV)));
2953 
2954 	r = kvm_apic_state_fixup(vcpu, s, true);
2955 	if (r) {
2956 		kvm_recalculate_apic_map(vcpu->kvm);
2957 		return r;
2958 	}
2959 	memcpy(vcpu->arch.apic->regs, s->regs, sizeof(*s));
2960 
2961 	atomic_set_release(&apic->vcpu->kvm->arch.apic_map_dirty, DIRTY);
2962 	kvm_recalculate_apic_map(vcpu->kvm);
2963 	kvm_apic_set_version(vcpu);
2964 
2965 	apic_update_ppr(apic);
2966 	cancel_apic_timer(apic);
2967 	apic->lapic_timer.expired_tscdeadline = 0;
2968 	apic_update_lvtt(apic);
2969 	apic_manage_nmi_watchdog(apic, kvm_lapic_get_reg(apic, APIC_LVT0));
2970 	update_divide_count(apic);
2971 	__start_apic_timer(apic, APIC_TMCCT);
2972 	kvm_lapic_set_reg(apic, APIC_TMCCT, 0);
2973 	kvm_apic_update_apicv(vcpu);
2974 	if (apic->apicv_active) {
2975 		static_call_cond(kvm_x86_apicv_post_state_restore)(vcpu);
2976 		static_call_cond(kvm_x86_hwapic_irr_update)(vcpu, apic_find_highest_irr(apic));
2977 		static_call_cond(kvm_x86_hwapic_isr_update)(apic_find_highest_isr(apic));
2978 	}
2979 	kvm_make_request(KVM_REQ_EVENT, vcpu);
2980 	if (ioapic_in_kernel(vcpu->kvm))
2981 		kvm_rtc_eoi_tracking_restore_one(vcpu);
2982 
2983 	vcpu->arch.apic_arb_prio = 0;
2984 
2985 	return 0;
2986 }
2987 
2988 void __kvm_migrate_apic_timer(struct kvm_vcpu *vcpu)
2989 {
2990 	struct hrtimer *timer;
2991 
2992 	if (!lapic_in_kernel(vcpu) ||
2993 		kvm_can_post_timer_interrupt(vcpu))
2994 		return;
2995 
2996 	timer = &vcpu->arch.apic->lapic_timer.timer;
2997 	if (hrtimer_cancel(timer))
2998 		hrtimer_start_expires(timer, HRTIMER_MODE_ABS_HARD);
2999 }
3000 
3001 /*
3002  * apic_sync_pv_eoi_from_guest - called on vmexit or cancel interrupt
3003  *
3004  * Detect whether guest triggered PV EOI since the
3005  * last entry. If yes, set EOI on guests's behalf.
3006  * Clear PV EOI in guest memory in any case.
3007  */
3008 static void apic_sync_pv_eoi_from_guest(struct kvm_vcpu *vcpu,
3009 					struct kvm_lapic *apic)
3010 {
3011 	int vector;
3012 	/*
3013 	 * PV EOI state is derived from KVM_APIC_PV_EOI_PENDING in host
3014 	 * and KVM_PV_EOI_ENABLED in guest memory as follows:
3015 	 *
3016 	 * KVM_APIC_PV_EOI_PENDING is unset:
3017 	 * 	-> host disabled PV EOI.
3018 	 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is set:
3019 	 * 	-> host enabled PV EOI, guest did not execute EOI yet.
3020 	 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is unset:
3021 	 * 	-> host enabled PV EOI, guest executed EOI.
3022 	 */
3023 	BUG_ON(!pv_eoi_enabled(vcpu));
3024 
3025 	if (pv_eoi_test_and_clr_pending(vcpu))
3026 		return;
3027 	vector = apic_set_eoi(apic);
3028 	trace_kvm_pv_eoi(apic, vector);
3029 }
3030 
3031 void kvm_lapic_sync_from_vapic(struct kvm_vcpu *vcpu)
3032 {
3033 	u32 data;
3034 
3035 	if (test_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention))
3036 		apic_sync_pv_eoi_from_guest(vcpu, vcpu->arch.apic);
3037 
3038 	if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
3039 		return;
3040 
3041 	if (kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
3042 				  sizeof(u32)))
3043 		return;
3044 
3045 	apic_set_tpr(vcpu->arch.apic, data & 0xff);
3046 }
3047 
3048 /*
3049  * apic_sync_pv_eoi_to_guest - called before vmentry
3050  *
3051  * Detect whether it's safe to enable PV EOI and
3052  * if yes do so.
3053  */
3054 static void apic_sync_pv_eoi_to_guest(struct kvm_vcpu *vcpu,
3055 					struct kvm_lapic *apic)
3056 {
3057 	if (!pv_eoi_enabled(vcpu) ||
3058 	    /* IRR set or many bits in ISR: could be nested. */
3059 	    apic->irr_pending ||
3060 	    /* Cache not set: could be safe but we don't bother. */
3061 	    apic->highest_isr_cache == -1 ||
3062 	    /* Need EOI to update ioapic. */
3063 	    kvm_ioapic_handles_vector(apic, apic->highest_isr_cache)) {
3064 		/*
3065 		 * PV EOI was disabled by apic_sync_pv_eoi_from_guest
3066 		 * so we need not do anything here.
3067 		 */
3068 		return;
3069 	}
3070 
3071 	pv_eoi_set_pending(apic->vcpu);
3072 }
3073 
3074 void kvm_lapic_sync_to_vapic(struct kvm_vcpu *vcpu)
3075 {
3076 	u32 data, tpr;
3077 	int max_irr, max_isr;
3078 	struct kvm_lapic *apic = vcpu->arch.apic;
3079 
3080 	apic_sync_pv_eoi_to_guest(vcpu, apic);
3081 
3082 	if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
3083 		return;
3084 
3085 	tpr = kvm_lapic_get_reg(apic, APIC_TASKPRI) & 0xff;
3086 	max_irr = apic_find_highest_irr(apic);
3087 	if (max_irr < 0)
3088 		max_irr = 0;
3089 	max_isr = apic_find_highest_isr(apic);
3090 	if (max_isr < 0)
3091 		max_isr = 0;
3092 	data = (tpr & 0xff) | ((max_isr & 0xf0) << 8) | (max_irr << 24);
3093 
3094 	kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
3095 				sizeof(u32));
3096 }
3097 
3098 int kvm_lapic_set_vapic_addr(struct kvm_vcpu *vcpu, gpa_t vapic_addr)
3099 {
3100 	if (vapic_addr) {
3101 		if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
3102 					&vcpu->arch.apic->vapic_cache,
3103 					vapic_addr, sizeof(u32)))
3104 			return -EINVAL;
3105 		__set_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
3106 	} else {
3107 		__clear_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
3108 	}
3109 
3110 	vcpu->arch.apic->vapic_addr = vapic_addr;
3111 	return 0;
3112 }
3113 
3114 int kvm_x2apic_icr_write(struct kvm_lapic *apic, u64 data)
3115 {
3116 	data &= ~APIC_ICR_BUSY;
3117 
3118 	kvm_apic_send_ipi(apic, (u32)data, (u32)(data >> 32));
3119 	kvm_lapic_set_reg64(apic, APIC_ICR, data);
3120 	trace_kvm_apic_write(APIC_ICR, data);
3121 	return 0;
3122 }
3123 
3124 static int kvm_lapic_msr_read(struct kvm_lapic *apic, u32 reg, u64 *data)
3125 {
3126 	u32 low;
3127 
3128 	if (reg == APIC_ICR) {
3129 		*data = kvm_lapic_get_reg64(apic, APIC_ICR);
3130 		return 0;
3131 	}
3132 
3133 	if (kvm_lapic_reg_read(apic, reg, 4, &low))
3134 		return 1;
3135 
3136 	*data = low;
3137 
3138 	return 0;
3139 }
3140 
3141 static int kvm_lapic_msr_write(struct kvm_lapic *apic, u32 reg, u64 data)
3142 {
3143 	/*
3144 	 * ICR is a 64-bit register in x2APIC mode (and Hyper-V PV vAPIC) and
3145 	 * can be written as such, all other registers remain accessible only
3146 	 * through 32-bit reads/writes.
3147 	 */
3148 	if (reg == APIC_ICR)
3149 		return kvm_x2apic_icr_write(apic, data);
3150 
3151 	/* Bits 63:32 are reserved in all other registers. */
3152 	if (data >> 32)
3153 		return 1;
3154 
3155 	return kvm_lapic_reg_write(apic, reg, (u32)data);
3156 }
3157 
3158 int kvm_x2apic_msr_write(struct kvm_vcpu *vcpu, u32 msr, u64 data)
3159 {
3160 	struct kvm_lapic *apic = vcpu->arch.apic;
3161 	u32 reg = (msr - APIC_BASE_MSR) << 4;
3162 
3163 	if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(apic))
3164 		return 1;
3165 
3166 	return kvm_lapic_msr_write(apic, reg, data);
3167 }
3168 
3169 int kvm_x2apic_msr_read(struct kvm_vcpu *vcpu, u32 msr, u64 *data)
3170 {
3171 	struct kvm_lapic *apic = vcpu->arch.apic;
3172 	u32 reg = (msr - APIC_BASE_MSR) << 4;
3173 
3174 	if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(apic))
3175 		return 1;
3176 
3177 	return kvm_lapic_msr_read(apic, reg, data);
3178 }
3179 
3180 int kvm_hv_vapic_msr_write(struct kvm_vcpu *vcpu, u32 reg, u64 data)
3181 {
3182 	if (!lapic_in_kernel(vcpu))
3183 		return 1;
3184 
3185 	return kvm_lapic_msr_write(vcpu->arch.apic, reg, data);
3186 }
3187 
3188 int kvm_hv_vapic_msr_read(struct kvm_vcpu *vcpu, u32 reg, u64 *data)
3189 {
3190 	if (!lapic_in_kernel(vcpu))
3191 		return 1;
3192 
3193 	return kvm_lapic_msr_read(vcpu->arch.apic, reg, data);
3194 }
3195 
3196 int kvm_lapic_set_pv_eoi(struct kvm_vcpu *vcpu, u64 data, unsigned long len)
3197 {
3198 	u64 addr = data & ~KVM_MSR_ENABLED;
3199 	struct gfn_to_hva_cache *ghc = &vcpu->arch.pv_eoi.data;
3200 	unsigned long new_len;
3201 	int ret;
3202 
3203 	if (!IS_ALIGNED(addr, 4))
3204 		return 1;
3205 
3206 	if (data & KVM_MSR_ENABLED) {
3207 		if (addr == ghc->gpa && len <= ghc->len)
3208 			new_len = ghc->len;
3209 		else
3210 			new_len = len;
3211 
3212 		ret = kvm_gfn_to_hva_cache_init(vcpu->kvm, ghc, addr, new_len);
3213 		if (ret)
3214 			return ret;
3215 	}
3216 
3217 	vcpu->arch.pv_eoi.msr_val = data;
3218 
3219 	return 0;
3220 }
3221 
3222 int kvm_apic_accept_events(struct kvm_vcpu *vcpu)
3223 {
3224 	struct kvm_lapic *apic = vcpu->arch.apic;
3225 	u8 sipi_vector;
3226 	int r;
3227 
3228 	if (!kvm_apic_has_pending_init_or_sipi(vcpu))
3229 		return 0;
3230 
3231 	if (is_guest_mode(vcpu)) {
3232 		r = kvm_check_nested_events(vcpu);
3233 		if (r < 0)
3234 			return r == -EBUSY ? 0 : r;
3235 		/*
3236 		 * Continue processing INIT/SIPI even if a nested VM-Exit
3237 		 * occurred, e.g. pending SIPIs should be dropped if INIT+SIPI
3238 		 * are blocked as a result of transitioning to VMX root mode.
3239 		 */
3240 	}
3241 
3242 	/*
3243 	 * INITs are blocked while CPU is in specific states (SMM, VMX root
3244 	 * mode, SVM with GIF=0), while SIPIs are dropped if the CPU isn't in
3245 	 * wait-for-SIPI (WFS).
3246 	 */
3247 	if (!kvm_apic_init_sipi_allowed(vcpu)) {
3248 		WARN_ON_ONCE(vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED);
3249 		clear_bit(KVM_APIC_SIPI, &apic->pending_events);
3250 		return 0;
3251 	}
3252 
3253 	if (test_and_clear_bit(KVM_APIC_INIT, &apic->pending_events)) {
3254 		kvm_vcpu_reset(vcpu, true);
3255 		if (kvm_vcpu_is_bsp(apic->vcpu))
3256 			vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
3257 		else
3258 			vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
3259 	}
3260 	if (test_and_clear_bit(KVM_APIC_SIPI, &apic->pending_events)) {
3261 		if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
3262 			/* evaluate pending_events before reading the vector */
3263 			smp_rmb();
3264 			sipi_vector = apic->sipi_vector;
3265 			static_call(kvm_x86_vcpu_deliver_sipi_vector)(vcpu, sipi_vector);
3266 			vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
3267 		}
3268 	}
3269 	return 0;
3270 }
3271 
3272 void kvm_lapic_exit(void)
3273 {
3274 	static_key_deferred_flush(&apic_hw_disabled);
3275 	WARN_ON(static_branch_unlikely(&apic_hw_disabled.key));
3276 	static_key_deferred_flush(&apic_sw_disabled);
3277 	WARN_ON(static_branch_unlikely(&apic_sw_disabled.key));
3278 }
3279