Loading...
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 "xen.h"
45#include "cpuid.h"
46#include "hyperv.h"
47#include "smm.h"
48
49#ifndef CONFIG_X86_64
50#define mod_64(x, y) ((x) - (y) * div64_u64(x, y))
51#else
52#define mod_64(x, y) ((x) % (y))
53#endif
54
55/* 14 is the version for Xeon and Pentium 8.4.8*/
56#define APIC_VERSION 0x14UL
57#define LAPIC_MMIO_LENGTH (1 << 12)
58/* followed define is not in apicdef.h */
59#define MAX_APIC_VECTOR 256
60#define APIC_VECTORS_PER_REG 32
61
62/*
63 * Enable local APIC timer advancement (tscdeadline mode only) with adaptive
64 * tuning. When enabled, KVM programs the host timer event to fire early, i.e.
65 * before the deadline expires, to account for the delay between taking the
66 * VM-Exit (to inject the guest event) and the subsequent VM-Enter to resume
67 * the guest, i.e. so that the interrupt arrives in the guest with minimal
68 * latency relative to the deadline programmed by the guest.
69 */
70static bool lapic_timer_advance __read_mostly = true;
71module_param(lapic_timer_advance, bool, 0444);
72
73#define LAPIC_TIMER_ADVANCE_ADJUST_MIN 100 /* clock cycles */
74#define LAPIC_TIMER_ADVANCE_ADJUST_MAX 10000 /* clock cycles */
75#define LAPIC_TIMER_ADVANCE_NS_INIT 1000
76#define LAPIC_TIMER_ADVANCE_NS_MAX 5000
77/* step-by-step approximation to mitigate fluctuation */
78#define LAPIC_TIMER_ADVANCE_ADJUST_STEP 8
79static int kvm_lapic_msr_read(struct kvm_lapic *apic, u32 reg, u64 *data);
80static int kvm_lapic_msr_write(struct kvm_lapic *apic, u32 reg, u64 data);
81
82static inline void __kvm_lapic_set_reg(char *regs, int reg_off, u32 val)
83{
84 *((u32 *) (regs + reg_off)) = val;
85}
86
87static inline void kvm_lapic_set_reg(struct kvm_lapic *apic, int reg_off, u32 val)
88{
89 __kvm_lapic_set_reg(apic->regs, reg_off, val);
90}
91
92static __always_inline u64 __kvm_lapic_get_reg64(char *regs, int reg)
93{
94 BUILD_BUG_ON(reg != APIC_ICR);
95 return *((u64 *) (regs + reg));
96}
97
98static __always_inline u64 kvm_lapic_get_reg64(struct kvm_lapic *apic, int reg)
99{
100 return __kvm_lapic_get_reg64(apic->regs, reg);
101}
102
103static __always_inline void __kvm_lapic_set_reg64(char *regs, int reg, u64 val)
104{
105 BUILD_BUG_ON(reg != APIC_ICR);
106 *((u64 *) (regs + reg)) = val;
107}
108
109static __always_inline void kvm_lapic_set_reg64(struct kvm_lapic *apic,
110 int reg, u64 val)
111{
112 __kvm_lapic_set_reg64(apic->regs, reg, val);
113}
114
115static inline int apic_test_vector(int vec, void *bitmap)
116{
117 return test_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
118}
119
120bool kvm_apic_pending_eoi(struct kvm_vcpu *vcpu, int vector)
121{
122 struct kvm_lapic *apic = vcpu->arch.apic;
123
124 return apic_test_vector(vector, apic->regs + APIC_ISR) ||
125 apic_test_vector(vector, apic->regs + APIC_IRR);
126}
127
128static inline int __apic_test_and_set_vector(int vec, void *bitmap)
129{
130 return __test_and_set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
131}
132
133static inline int __apic_test_and_clear_vector(int vec, void *bitmap)
134{
135 return __test_and_clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
136}
137
138__read_mostly DEFINE_STATIC_KEY_FALSE(kvm_has_noapic_vcpu);
139EXPORT_SYMBOL_GPL(kvm_has_noapic_vcpu);
140
141__read_mostly DEFINE_STATIC_KEY_DEFERRED_FALSE(apic_hw_disabled, HZ);
142__read_mostly DEFINE_STATIC_KEY_DEFERRED_FALSE(apic_sw_disabled, HZ);
143
144static inline int apic_enabled(struct kvm_lapic *apic)
145{
146 return kvm_apic_sw_enabled(apic) && kvm_apic_hw_enabled(apic);
147}
148
149#define LVT_MASK \
150 (APIC_LVT_MASKED | APIC_SEND_PENDING | APIC_VECTOR_MASK)
151
152#define LINT_MASK \
153 (LVT_MASK | APIC_MODE_MASK | APIC_INPUT_POLARITY | \
154 APIC_LVT_REMOTE_IRR | APIC_LVT_LEVEL_TRIGGER)
155
156static inline u32 kvm_x2apic_id(struct kvm_lapic *apic)
157{
158 return apic->vcpu->vcpu_id;
159}
160
161static bool kvm_can_post_timer_interrupt(struct kvm_vcpu *vcpu)
162{
163 return pi_inject_timer && kvm_vcpu_apicv_active(vcpu) &&
164 (kvm_mwait_in_guest(vcpu->kvm) || kvm_hlt_in_guest(vcpu->kvm));
165}
166
167bool kvm_can_use_hv_timer(struct kvm_vcpu *vcpu)
168{
169 return kvm_x86_ops.set_hv_timer
170 && !(kvm_mwait_in_guest(vcpu->kvm) ||
171 kvm_can_post_timer_interrupt(vcpu));
172}
173
174static bool kvm_use_posted_timer_interrupt(struct kvm_vcpu *vcpu)
175{
176 return kvm_can_post_timer_interrupt(vcpu) && vcpu->mode == IN_GUEST_MODE;
177}
178
179static inline u32 kvm_apic_calc_x2apic_ldr(u32 id)
180{
181 return ((id >> 4) << 16) | (1 << (id & 0xf));
182}
183
184static inline bool kvm_apic_map_get_logical_dest(struct kvm_apic_map *map,
185 u32 dest_id, struct kvm_lapic ***cluster, u16 *mask) {
186 switch (map->logical_mode) {
187 case KVM_APIC_MODE_SW_DISABLED:
188 /* Arbitrarily use the flat map so that @cluster isn't NULL. */
189 *cluster = map->xapic_flat_map;
190 *mask = 0;
191 return true;
192 case KVM_APIC_MODE_X2APIC: {
193 u32 offset = (dest_id >> 16) * 16;
194 u32 max_apic_id = map->max_apic_id;
195
196 if (offset <= max_apic_id) {
197 u8 cluster_size = min(max_apic_id - offset + 1, 16U);
198
199 offset = array_index_nospec(offset, map->max_apic_id + 1);
200 *cluster = &map->phys_map[offset];
201 *mask = dest_id & (0xffff >> (16 - cluster_size));
202 } else {
203 *mask = 0;
204 }
205
206 return true;
207 }
208 case KVM_APIC_MODE_XAPIC_FLAT:
209 *cluster = map->xapic_flat_map;
210 *mask = dest_id & 0xff;
211 return true;
212 case KVM_APIC_MODE_XAPIC_CLUSTER:
213 *cluster = map->xapic_cluster_map[(dest_id >> 4) & 0xf];
214 *mask = dest_id & 0xf;
215 return true;
216 case KVM_APIC_MODE_MAP_DISABLED:
217 return false;
218 default:
219 WARN_ON_ONCE(1);
220 return false;
221 }
222}
223
224static void kvm_apic_map_free(struct rcu_head *rcu)
225{
226 struct kvm_apic_map *map = container_of(rcu, struct kvm_apic_map, rcu);
227
228 kvfree(map);
229}
230
231static int kvm_recalculate_phys_map(struct kvm_apic_map *new,
232 struct kvm_vcpu *vcpu,
233 bool *xapic_id_mismatch)
234{
235 struct kvm_lapic *apic = vcpu->arch.apic;
236 u32 x2apic_id = kvm_x2apic_id(apic);
237 u32 xapic_id = kvm_xapic_id(apic);
238 u32 physical_id;
239
240 /*
241 * For simplicity, KVM always allocates enough space for all possible
242 * xAPIC IDs. Yell, but don't kill the VM, as KVM can continue on
243 * without the optimized map.
244 */
245 if (WARN_ON_ONCE(xapic_id > new->max_apic_id))
246 return -EINVAL;
247
248 /*
249 * Bail if a vCPU was added and/or enabled its APIC between allocating
250 * the map and doing the actual calculations for the map. Note, KVM
251 * hardcodes the x2APIC ID to vcpu_id, i.e. there's no TOCTOU bug if
252 * the compiler decides to reload x2apic_id after this check.
253 */
254 if (x2apic_id > new->max_apic_id)
255 return -E2BIG;
256
257 /*
258 * Deliberately truncate the vCPU ID when detecting a mismatched APIC
259 * ID to avoid false positives if the vCPU ID, i.e. x2APIC ID, is a
260 * 32-bit value. Any unwanted aliasing due to truncation results will
261 * be detected below.
262 */
263 if (!apic_x2apic_mode(apic) && xapic_id != (u8)vcpu->vcpu_id)
264 *xapic_id_mismatch = true;
265
266 /*
267 * Apply KVM's hotplug hack if userspace has enable 32-bit APIC IDs.
268 * Allow sending events to vCPUs by their x2APIC ID even if the target
269 * vCPU is in legacy xAPIC mode, and silently ignore aliased xAPIC IDs
270 * (the x2APIC ID is truncated to 8 bits, causing IDs > 0xff to wrap
271 * and collide).
272 *
273 * Honor the architectural (and KVM's non-optimized) behavior if
274 * userspace has not enabled 32-bit x2APIC IDs. Each APIC is supposed
275 * to process messages independently. If multiple vCPUs have the same
276 * effective APIC ID, e.g. due to the x2APIC wrap or because the guest
277 * manually modified its xAPIC IDs, events targeting that ID are
278 * supposed to be recognized by all vCPUs with said ID.
279 */
280 if (vcpu->kvm->arch.x2apic_format) {
281 /* See also kvm_apic_match_physical_addr(). */
282 if (apic_x2apic_mode(apic) || x2apic_id > 0xff)
283 new->phys_map[x2apic_id] = apic;
284
285 if (!apic_x2apic_mode(apic) && !new->phys_map[xapic_id])
286 new->phys_map[xapic_id] = apic;
287 } else {
288 /*
289 * Disable the optimized map if the physical APIC ID is already
290 * mapped, i.e. is aliased to multiple vCPUs. The optimized
291 * map requires a strict 1:1 mapping between IDs and vCPUs.
292 */
293 if (apic_x2apic_mode(apic))
294 physical_id = x2apic_id;
295 else
296 physical_id = xapic_id;
297
298 if (new->phys_map[physical_id])
299 return -EINVAL;
300
301 new->phys_map[physical_id] = apic;
302 }
303
304 return 0;
305}
306
307static void kvm_recalculate_logical_map(struct kvm_apic_map *new,
308 struct kvm_vcpu *vcpu)
309{
310 struct kvm_lapic *apic = vcpu->arch.apic;
311 enum kvm_apic_logical_mode logical_mode;
312 struct kvm_lapic **cluster;
313 u16 mask;
314 u32 ldr;
315
316 if (new->logical_mode == KVM_APIC_MODE_MAP_DISABLED)
317 return;
318
319 if (!kvm_apic_sw_enabled(apic))
320 return;
321
322 ldr = kvm_lapic_get_reg(apic, APIC_LDR);
323 if (!ldr)
324 return;
325
326 if (apic_x2apic_mode(apic)) {
327 logical_mode = KVM_APIC_MODE_X2APIC;
328 } else {
329 ldr = GET_APIC_LOGICAL_ID(ldr);
330 if (kvm_lapic_get_reg(apic, APIC_DFR) == APIC_DFR_FLAT)
331 logical_mode = KVM_APIC_MODE_XAPIC_FLAT;
332 else
333 logical_mode = KVM_APIC_MODE_XAPIC_CLUSTER;
334 }
335
336 /*
337 * To optimize logical mode delivery, all software-enabled APICs must
338 * be configured for the same mode.
339 */
340 if (new->logical_mode == KVM_APIC_MODE_SW_DISABLED) {
341 new->logical_mode = logical_mode;
342 } else if (new->logical_mode != logical_mode) {
343 new->logical_mode = KVM_APIC_MODE_MAP_DISABLED;
344 return;
345 }
346
347 /*
348 * In x2APIC mode, the LDR is read-only and derived directly from the
349 * x2APIC ID, thus is guaranteed to be addressable. KVM reuses
350 * kvm_apic_map.phys_map to optimize logical mode x2APIC interrupts by
351 * reversing the LDR calculation to get cluster of APICs, i.e. no
352 * additional work is required.
353 */
354 if (apic_x2apic_mode(apic))
355 return;
356
357 if (WARN_ON_ONCE(!kvm_apic_map_get_logical_dest(new, ldr,
358 &cluster, &mask))) {
359 new->logical_mode = KVM_APIC_MODE_MAP_DISABLED;
360 return;
361 }
362
363 if (!mask)
364 return;
365
366 ldr = ffs(mask) - 1;
367 if (!is_power_of_2(mask) || cluster[ldr])
368 new->logical_mode = KVM_APIC_MODE_MAP_DISABLED;
369 else
370 cluster[ldr] = apic;
371}
372
373/*
374 * CLEAN -> DIRTY and UPDATE_IN_PROGRESS -> DIRTY changes happen without a lock.
375 *
376 * DIRTY -> UPDATE_IN_PROGRESS and UPDATE_IN_PROGRESS -> CLEAN happen with
377 * apic_map_lock_held.
378 */
379enum {
380 CLEAN,
381 UPDATE_IN_PROGRESS,
382 DIRTY
383};
384
385static void kvm_recalculate_apic_map(struct kvm *kvm)
386{
387 struct kvm_apic_map *new, *old = NULL;
388 struct kvm_vcpu *vcpu;
389 unsigned long i;
390 u32 max_id = 255; /* enough space for any xAPIC ID */
391 bool xapic_id_mismatch;
392 int r;
393
394 /* Read kvm->arch.apic_map_dirty before kvm->arch.apic_map. */
395 if (atomic_read_acquire(&kvm->arch.apic_map_dirty) == CLEAN)
396 return;
397
398 WARN_ONCE(!irqchip_in_kernel(kvm),
399 "Dirty APIC map without an in-kernel local APIC");
400
401 mutex_lock(&kvm->arch.apic_map_lock);
402
403retry:
404 /*
405 * Read kvm->arch.apic_map_dirty before kvm->arch.apic_map (if clean)
406 * or the APIC registers (if dirty). Note, on retry the map may have
407 * not yet been marked dirty by whatever task changed a vCPU's x2APIC
408 * ID, i.e. the map may still show up as in-progress. In that case
409 * this task still needs to retry and complete its calculation.
410 */
411 if (atomic_cmpxchg_acquire(&kvm->arch.apic_map_dirty,
412 DIRTY, UPDATE_IN_PROGRESS) == CLEAN) {
413 /* Someone else has updated the map. */
414 mutex_unlock(&kvm->arch.apic_map_lock);
415 return;
416 }
417
418 /*
419 * Reset the mismatch flag between attempts so that KVM does the right
420 * thing if a vCPU changes its xAPIC ID, but do NOT reset max_id, i.e.
421 * keep max_id strictly increasing. Disallowing max_id from shrinking
422 * ensures KVM won't get stuck in an infinite loop, e.g. if the vCPU
423 * with the highest x2APIC ID is toggling its APIC on and off.
424 */
425 xapic_id_mismatch = false;
426
427 kvm_for_each_vcpu(i, vcpu, kvm)
428 if (kvm_apic_present(vcpu))
429 max_id = max(max_id, kvm_x2apic_id(vcpu->arch.apic));
430
431 new = kvzalloc(sizeof(struct kvm_apic_map) +
432 sizeof(struct kvm_lapic *) * ((u64)max_id + 1),
433 GFP_KERNEL_ACCOUNT);
434
435 if (!new)
436 goto out;
437
438 new->max_apic_id = max_id;
439 new->logical_mode = KVM_APIC_MODE_SW_DISABLED;
440
441 kvm_for_each_vcpu(i, vcpu, kvm) {
442 if (!kvm_apic_present(vcpu))
443 continue;
444
445 r = kvm_recalculate_phys_map(new, vcpu, &xapic_id_mismatch);
446 if (r) {
447 kvfree(new);
448 new = NULL;
449 if (r == -E2BIG) {
450 cond_resched();
451 goto retry;
452 }
453
454 goto out;
455 }
456
457 kvm_recalculate_logical_map(new, vcpu);
458 }
459out:
460 /*
461 * The optimized map is effectively KVM's internal version of APICv,
462 * and all unwanted aliasing that results in disabling the optimized
463 * map also applies to APICv.
464 */
465 if (!new)
466 kvm_set_apicv_inhibit(kvm, APICV_INHIBIT_REASON_PHYSICAL_ID_ALIASED);
467 else
468 kvm_clear_apicv_inhibit(kvm, APICV_INHIBIT_REASON_PHYSICAL_ID_ALIASED);
469
470 if (!new || new->logical_mode == KVM_APIC_MODE_MAP_DISABLED)
471 kvm_set_apicv_inhibit(kvm, APICV_INHIBIT_REASON_LOGICAL_ID_ALIASED);
472 else
473 kvm_clear_apicv_inhibit(kvm, APICV_INHIBIT_REASON_LOGICAL_ID_ALIASED);
474
475 if (xapic_id_mismatch)
476 kvm_set_apicv_inhibit(kvm, APICV_INHIBIT_REASON_APIC_ID_MODIFIED);
477 else
478 kvm_clear_apicv_inhibit(kvm, APICV_INHIBIT_REASON_APIC_ID_MODIFIED);
479
480 old = rcu_dereference_protected(kvm->arch.apic_map,
481 lockdep_is_held(&kvm->arch.apic_map_lock));
482 rcu_assign_pointer(kvm->arch.apic_map, new);
483 /*
484 * Write kvm->arch.apic_map before clearing apic->apic_map_dirty.
485 * If another update has come in, leave it DIRTY.
486 */
487 atomic_cmpxchg_release(&kvm->arch.apic_map_dirty,
488 UPDATE_IN_PROGRESS, CLEAN);
489 mutex_unlock(&kvm->arch.apic_map_lock);
490
491 if (old)
492 call_rcu(&old->rcu, kvm_apic_map_free);
493
494 kvm_make_scan_ioapic_request(kvm);
495}
496
497static inline void apic_set_spiv(struct kvm_lapic *apic, u32 val)
498{
499 bool enabled = val & APIC_SPIV_APIC_ENABLED;
500
501 kvm_lapic_set_reg(apic, APIC_SPIV, val);
502
503 if (enabled != apic->sw_enabled) {
504 apic->sw_enabled = enabled;
505 if (enabled)
506 static_branch_slow_dec_deferred(&apic_sw_disabled);
507 else
508 static_branch_inc(&apic_sw_disabled.key);
509
510 atomic_set_release(&apic->vcpu->kvm->arch.apic_map_dirty, DIRTY);
511 }
512
513 /* Check if there are APF page ready requests pending */
514 if (enabled) {
515 kvm_make_request(KVM_REQ_APF_READY, apic->vcpu);
516 kvm_xen_sw_enable_lapic(apic->vcpu);
517 }
518}
519
520static inline void kvm_apic_set_xapic_id(struct kvm_lapic *apic, u8 id)
521{
522 kvm_lapic_set_reg(apic, APIC_ID, id << 24);
523 atomic_set_release(&apic->vcpu->kvm->arch.apic_map_dirty, DIRTY);
524}
525
526static inline void kvm_apic_set_ldr(struct kvm_lapic *apic, u32 id)
527{
528 kvm_lapic_set_reg(apic, APIC_LDR, id);
529 atomic_set_release(&apic->vcpu->kvm->arch.apic_map_dirty, DIRTY);
530}
531
532static inline void kvm_apic_set_dfr(struct kvm_lapic *apic, u32 val)
533{
534 kvm_lapic_set_reg(apic, APIC_DFR, val);
535 atomic_set_release(&apic->vcpu->kvm->arch.apic_map_dirty, DIRTY);
536}
537
538static inline void kvm_apic_set_x2apic_id(struct kvm_lapic *apic, u32 id)
539{
540 u32 ldr = kvm_apic_calc_x2apic_ldr(id);
541
542 WARN_ON_ONCE(id != apic->vcpu->vcpu_id);
543
544 kvm_lapic_set_reg(apic, APIC_ID, id);
545 kvm_lapic_set_reg(apic, APIC_LDR, ldr);
546 atomic_set_release(&apic->vcpu->kvm->arch.apic_map_dirty, DIRTY);
547}
548
549static inline int apic_lvt_enabled(struct kvm_lapic *apic, int lvt_type)
550{
551 return !(kvm_lapic_get_reg(apic, lvt_type) & APIC_LVT_MASKED);
552}
553
554static inline int apic_lvtt_oneshot(struct kvm_lapic *apic)
555{
556 return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_ONESHOT;
557}
558
559static inline int apic_lvtt_period(struct kvm_lapic *apic)
560{
561 return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_PERIODIC;
562}
563
564static inline int apic_lvtt_tscdeadline(struct kvm_lapic *apic)
565{
566 return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_TSCDEADLINE;
567}
568
569static inline int apic_lvt_nmi_mode(u32 lvt_val)
570{
571 return (lvt_val & (APIC_MODE_MASK | APIC_LVT_MASKED)) == APIC_DM_NMI;
572}
573
574static inline bool kvm_lapic_lvt_supported(struct kvm_lapic *apic, int lvt_index)
575{
576 return apic->nr_lvt_entries > lvt_index;
577}
578
579static inline int kvm_apic_calc_nr_lvt_entries(struct kvm_vcpu *vcpu)
580{
581 return KVM_APIC_MAX_NR_LVT_ENTRIES - !(vcpu->arch.mcg_cap & MCG_CMCI_P);
582}
583
584void kvm_apic_set_version(struct kvm_vcpu *vcpu)
585{
586 struct kvm_lapic *apic = vcpu->arch.apic;
587 u32 v = 0;
588
589 if (!lapic_in_kernel(vcpu))
590 return;
591
592 v = APIC_VERSION | ((apic->nr_lvt_entries - 1) << 16);
593
594 /*
595 * KVM emulates 82093AA datasheet (with in-kernel IOAPIC implementation)
596 * which doesn't have EOI register; Some buggy OSes (e.g. Windows with
597 * Hyper-V role) disable EOI broadcast in lapic not checking for IOAPIC
598 * version first and level-triggered interrupts never get EOIed in
599 * IOAPIC.
600 */
601 if (guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) &&
602 !ioapic_in_kernel(vcpu->kvm))
603 v |= APIC_LVR_DIRECTED_EOI;
604 kvm_lapic_set_reg(apic, APIC_LVR, v);
605}
606
607void kvm_apic_after_set_mcg_cap(struct kvm_vcpu *vcpu)
608{
609 int nr_lvt_entries = kvm_apic_calc_nr_lvt_entries(vcpu);
610 struct kvm_lapic *apic = vcpu->arch.apic;
611 int i;
612
613 if (!lapic_in_kernel(vcpu) || nr_lvt_entries == apic->nr_lvt_entries)
614 return;
615
616 /* Initialize/mask any "new" LVT entries. */
617 for (i = apic->nr_lvt_entries; i < nr_lvt_entries; i++)
618 kvm_lapic_set_reg(apic, APIC_LVTx(i), APIC_LVT_MASKED);
619
620 apic->nr_lvt_entries = nr_lvt_entries;
621
622 /* The number of LVT entries is reflected in the version register. */
623 kvm_apic_set_version(vcpu);
624}
625
626static const unsigned int apic_lvt_mask[KVM_APIC_MAX_NR_LVT_ENTRIES] = {
627 [LVT_TIMER] = LVT_MASK, /* timer mode mask added at runtime */
628 [LVT_THERMAL_MONITOR] = LVT_MASK | APIC_MODE_MASK,
629 [LVT_PERFORMANCE_COUNTER] = LVT_MASK | APIC_MODE_MASK,
630 [LVT_LINT0] = LINT_MASK,
631 [LVT_LINT1] = LINT_MASK,
632 [LVT_ERROR] = LVT_MASK,
633 [LVT_CMCI] = LVT_MASK | APIC_MODE_MASK
634};
635
636static int find_highest_vector(void *bitmap)
637{
638 int vec;
639 u32 *reg;
640
641 for (vec = MAX_APIC_VECTOR - APIC_VECTORS_PER_REG;
642 vec >= 0; vec -= APIC_VECTORS_PER_REG) {
643 reg = bitmap + REG_POS(vec);
644 if (*reg)
645 return __fls(*reg) + vec;
646 }
647
648 return -1;
649}
650
651static u8 count_vectors(void *bitmap)
652{
653 int vec;
654 u32 *reg;
655 u8 count = 0;
656
657 for (vec = 0; vec < MAX_APIC_VECTOR; vec += APIC_VECTORS_PER_REG) {
658 reg = bitmap + REG_POS(vec);
659 count += hweight32(*reg);
660 }
661
662 return count;
663}
664
665bool __kvm_apic_update_irr(u32 *pir, void *regs, int *max_irr)
666{
667 u32 i, vec;
668 u32 pir_val, irr_val, prev_irr_val;
669 int max_updated_irr;
670
671 max_updated_irr = -1;
672 *max_irr = -1;
673
674 for (i = vec = 0; i <= 7; i++, vec += 32) {
675 u32 *p_irr = (u32 *)(regs + APIC_IRR + i * 0x10);
676
677 irr_val = *p_irr;
678 pir_val = READ_ONCE(pir[i]);
679
680 if (pir_val) {
681 pir_val = xchg(&pir[i], 0);
682
683 prev_irr_val = irr_val;
684 do {
685 irr_val = prev_irr_val | pir_val;
686 } while (prev_irr_val != irr_val &&
687 !try_cmpxchg(p_irr, &prev_irr_val, irr_val));
688
689 if (prev_irr_val != irr_val)
690 max_updated_irr = __fls(irr_val ^ prev_irr_val) + vec;
691 }
692 if (irr_val)
693 *max_irr = __fls(irr_val) + vec;
694 }
695
696 return ((max_updated_irr != -1) &&
697 (max_updated_irr == *max_irr));
698}
699EXPORT_SYMBOL_GPL(__kvm_apic_update_irr);
700
701bool kvm_apic_update_irr(struct kvm_vcpu *vcpu, u32 *pir, int *max_irr)
702{
703 struct kvm_lapic *apic = vcpu->arch.apic;
704 bool irr_updated = __kvm_apic_update_irr(pir, apic->regs, max_irr);
705
706 if (unlikely(!apic->apicv_active && irr_updated))
707 apic->irr_pending = true;
708 return irr_updated;
709}
710EXPORT_SYMBOL_GPL(kvm_apic_update_irr);
711
712static inline int apic_search_irr(struct kvm_lapic *apic)
713{
714 return find_highest_vector(apic->regs + APIC_IRR);
715}
716
717static inline int apic_find_highest_irr(struct kvm_lapic *apic)
718{
719 int result;
720
721 /*
722 * Note that irr_pending is just a hint. It will be always
723 * true with virtual interrupt delivery enabled.
724 */
725 if (!apic->irr_pending)
726 return -1;
727
728 result = apic_search_irr(apic);
729 ASSERT(result == -1 || result >= 16);
730
731 return result;
732}
733
734static inline void apic_clear_irr(int vec, struct kvm_lapic *apic)
735{
736 if (unlikely(apic->apicv_active)) {
737 /* need to update RVI */
738 kvm_lapic_clear_vector(vec, apic->regs + APIC_IRR);
739 kvm_x86_call(hwapic_irr_update)(apic->vcpu,
740 apic_find_highest_irr(apic));
741 } else {
742 apic->irr_pending = false;
743 kvm_lapic_clear_vector(vec, apic->regs + APIC_IRR);
744 if (apic_search_irr(apic) != -1)
745 apic->irr_pending = true;
746 }
747}
748
749void kvm_apic_clear_irr(struct kvm_vcpu *vcpu, int vec)
750{
751 apic_clear_irr(vec, vcpu->arch.apic);
752}
753EXPORT_SYMBOL_GPL(kvm_apic_clear_irr);
754
755static inline void apic_set_isr(int vec, struct kvm_lapic *apic)
756{
757 if (__apic_test_and_set_vector(vec, apic->regs + APIC_ISR))
758 return;
759
760 /*
761 * With APIC virtualization enabled, all caching is disabled
762 * because the processor can modify ISR under the hood. Instead
763 * just set SVI.
764 */
765 if (unlikely(apic->apicv_active))
766 kvm_x86_call(hwapic_isr_update)(apic->vcpu, vec);
767 else {
768 ++apic->isr_count;
769 BUG_ON(apic->isr_count > MAX_APIC_VECTOR);
770 /*
771 * ISR (in service register) bit is set when injecting an interrupt.
772 * The highest vector is injected. Thus the latest bit set matches
773 * the highest bit in ISR.
774 */
775 apic->highest_isr_cache = vec;
776 }
777}
778
779static inline int apic_find_highest_isr(struct kvm_lapic *apic)
780{
781 int result;
782
783 /*
784 * Note that isr_count is always 1, and highest_isr_cache
785 * is always -1, with APIC virtualization enabled.
786 */
787 if (!apic->isr_count)
788 return -1;
789 if (likely(apic->highest_isr_cache != -1))
790 return apic->highest_isr_cache;
791
792 result = find_highest_vector(apic->regs + APIC_ISR);
793 ASSERT(result == -1 || result >= 16);
794
795 return result;
796}
797
798static inline void apic_clear_isr(int vec, struct kvm_lapic *apic)
799{
800 if (!__apic_test_and_clear_vector(vec, apic->regs + APIC_ISR))
801 return;
802
803 /*
804 * We do get here for APIC virtualization enabled if the guest
805 * uses the Hyper-V APIC enlightenment. In this case we may need
806 * to trigger a new interrupt delivery by writing the SVI field;
807 * on the other hand isr_count and highest_isr_cache are unused
808 * and must be left alone.
809 */
810 if (unlikely(apic->apicv_active))
811 kvm_x86_call(hwapic_isr_update)(apic->vcpu, apic_find_highest_isr(apic));
812 else {
813 --apic->isr_count;
814 BUG_ON(apic->isr_count < 0);
815 apic->highest_isr_cache = -1;
816 }
817}
818
819void kvm_apic_update_hwapic_isr(struct kvm_vcpu *vcpu)
820{
821 struct kvm_lapic *apic = vcpu->arch.apic;
822
823 if (WARN_ON_ONCE(!lapic_in_kernel(vcpu)) || !apic->apicv_active)
824 return;
825
826 kvm_x86_call(hwapic_isr_update)(vcpu, apic_find_highest_isr(apic));
827}
828EXPORT_SYMBOL_GPL(kvm_apic_update_hwapic_isr);
829
830int kvm_lapic_find_highest_irr(struct kvm_vcpu *vcpu)
831{
832 /* This may race with setting of irr in __apic_accept_irq() and
833 * value returned may be wrong, but kvm_vcpu_kick() in __apic_accept_irq
834 * will cause vmexit immediately and the value will be recalculated
835 * on the next vmentry.
836 */
837 return apic_find_highest_irr(vcpu->arch.apic);
838}
839EXPORT_SYMBOL_GPL(kvm_lapic_find_highest_irr);
840
841static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
842 int vector, int level, int trig_mode,
843 struct dest_map *dest_map);
844
845int kvm_apic_set_irq(struct kvm_vcpu *vcpu, struct kvm_lapic_irq *irq,
846 struct dest_map *dest_map)
847{
848 struct kvm_lapic *apic = vcpu->arch.apic;
849
850 return __apic_accept_irq(apic, irq->delivery_mode, irq->vector,
851 irq->level, irq->trig_mode, dest_map);
852}
853
854static int __pv_send_ipi(unsigned long *ipi_bitmap, struct kvm_apic_map *map,
855 struct kvm_lapic_irq *irq, u32 min)
856{
857 int i, count = 0;
858 struct kvm_vcpu *vcpu;
859
860 if (min > map->max_apic_id)
861 return 0;
862
863 for_each_set_bit(i, ipi_bitmap,
864 min((u32)BITS_PER_LONG, (map->max_apic_id - min + 1))) {
865 if (map->phys_map[min + i]) {
866 vcpu = map->phys_map[min + i]->vcpu;
867 count += kvm_apic_set_irq(vcpu, irq, NULL);
868 }
869 }
870
871 return count;
872}
873
874int kvm_pv_send_ipi(struct kvm *kvm, unsigned long ipi_bitmap_low,
875 unsigned long ipi_bitmap_high, u32 min,
876 unsigned long icr, int op_64_bit)
877{
878 struct kvm_apic_map *map;
879 struct kvm_lapic_irq irq = {0};
880 int cluster_size = op_64_bit ? 64 : 32;
881 int count;
882
883 if (icr & (APIC_DEST_MASK | APIC_SHORT_MASK))
884 return -KVM_EINVAL;
885
886 irq.vector = icr & APIC_VECTOR_MASK;
887 irq.delivery_mode = icr & APIC_MODE_MASK;
888 irq.level = (icr & APIC_INT_ASSERT) != 0;
889 irq.trig_mode = icr & APIC_INT_LEVELTRIG;
890
891 rcu_read_lock();
892 map = rcu_dereference(kvm->arch.apic_map);
893
894 count = -EOPNOTSUPP;
895 if (likely(map)) {
896 count = __pv_send_ipi(&ipi_bitmap_low, map, &irq, min);
897 min += cluster_size;
898 count += __pv_send_ipi(&ipi_bitmap_high, map, &irq, min);
899 }
900
901 rcu_read_unlock();
902 return count;
903}
904
905static int pv_eoi_put_user(struct kvm_vcpu *vcpu, u8 val)
906{
907
908 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, &val,
909 sizeof(val));
910}
911
912static int pv_eoi_get_user(struct kvm_vcpu *vcpu, u8 *val)
913{
914
915 return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.pv_eoi.data, val,
916 sizeof(*val));
917}
918
919static inline bool pv_eoi_enabled(struct kvm_vcpu *vcpu)
920{
921 return vcpu->arch.pv_eoi.msr_val & KVM_MSR_ENABLED;
922}
923
924static void pv_eoi_set_pending(struct kvm_vcpu *vcpu)
925{
926 if (pv_eoi_put_user(vcpu, KVM_PV_EOI_ENABLED) < 0)
927 return;
928
929 __set_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
930}
931
932static bool pv_eoi_test_and_clr_pending(struct kvm_vcpu *vcpu)
933{
934 u8 val;
935
936 if (pv_eoi_get_user(vcpu, &val) < 0)
937 return false;
938
939 val &= KVM_PV_EOI_ENABLED;
940
941 if (val && pv_eoi_put_user(vcpu, KVM_PV_EOI_DISABLED) < 0)
942 return false;
943
944 /*
945 * Clear pending bit in any case: it will be set again on vmentry.
946 * While this might not be ideal from performance point of view,
947 * this makes sure pv eoi is only enabled when we know it's safe.
948 */
949 __clear_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention);
950
951 return val;
952}
953
954static int apic_has_interrupt_for_ppr(struct kvm_lapic *apic, u32 ppr)
955{
956 int highest_irr;
957 if (kvm_x86_ops.sync_pir_to_irr)
958 highest_irr = kvm_x86_call(sync_pir_to_irr)(apic->vcpu);
959 else
960 highest_irr = apic_find_highest_irr(apic);
961 if (highest_irr == -1 || (highest_irr & 0xF0) <= ppr)
962 return -1;
963 return highest_irr;
964}
965
966static bool __apic_update_ppr(struct kvm_lapic *apic, u32 *new_ppr)
967{
968 u32 tpr, isrv, ppr, old_ppr;
969 int isr;
970
971 old_ppr = kvm_lapic_get_reg(apic, APIC_PROCPRI);
972 tpr = kvm_lapic_get_reg(apic, APIC_TASKPRI);
973 isr = apic_find_highest_isr(apic);
974 isrv = (isr != -1) ? isr : 0;
975
976 if ((tpr & 0xf0) >= (isrv & 0xf0))
977 ppr = tpr & 0xff;
978 else
979 ppr = isrv & 0xf0;
980
981 *new_ppr = ppr;
982 if (old_ppr != ppr)
983 kvm_lapic_set_reg(apic, APIC_PROCPRI, ppr);
984
985 return ppr < old_ppr;
986}
987
988static void apic_update_ppr(struct kvm_lapic *apic)
989{
990 u32 ppr;
991
992 if (__apic_update_ppr(apic, &ppr) &&
993 apic_has_interrupt_for_ppr(apic, ppr) != -1)
994 kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
995}
996
997void kvm_apic_update_ppr(struct kvm_vcpu *vcpu)
998{
999 apic_update_ppr(vcpu->arch.apic);
1000}
1001EXPORT_SYMBOL_GPL(kvm_apic_update_ppr);
1002
1003static void apic_set_tpr(struct kvm_lapic *apic, u32 tpr)
1004{
1005 kvm_lapic_set_reg(apic, APIC_TASKPRI, tpr);
1006 apic_update_ppr(apic);
1007}
1008
1009static bool kvm_apic_broadcast(struct kvm_lapic *apic, u32 mda)
1010{
1011 return mda == (apic_x2apic_mode(apic) ?
1012 X2APIC_BROADCAST : APIC_BROADCAST);
1013}
1014
1015static bool kvm_apic_match_physical_addr(struct kvm_lapic *apic, u32 mda)
1016{
1017 if (kvm_apic_broadcast(apic, mda))
1018 return true;
1019
1020 /*
1021 * Hotplug hack: Accept interrupts for vCPUs in xAPIC mode as if they
1022 * were in x2APIC mode if the target APIC ID can't be encoded as an
1023 * xAPIC ID. This allows unique addressing of hotplugged vCPUs (which
1024 * start in xAPIC mode) with an APIC ID that is unaddressable in xAPIC
1025 * mode. Match the x2APIC ID if and only if the target APIC ID can't
1026 * be encoded in xAPIC to avoid spurious matches against a vCPU that
1027 * changed its (addressable) xAPIC ID (which is writable).
1028 */
1029 if (apic_x2apic_mode(apic) || mda > 0xff)
1030 return mda == kvm_x2apic_id(apic);
1031
1032 return mda == kvm_xapic_id(apic);
1033}
1034
1035static bool kvm_apic_match_logical_addr(struct kvm_lapic *apic, u32 mda)
1036{
1037 u32 logical_id;
1038
1039 if (kvm_apic_broadcast(apic, mda))
1040 return true;
1041
1042 logical_id = kvm_lapic_get_reg(apic, APIC_LDR);
1043
1044 if (apic_x2apic_mode(apic))
1045 return ((logical_id >> 16) == (mda >> 16))
1046 && (logical_id & mda & 0xffff) != 0;
1047
1048 logical_id = GET_APIC_LOGICAL_ID(logical_id);
1049
1050 switch (kvm_lapic_get_reg(apic, APIC_DFR)) {
1051 case APIC_DFR_FLAT:
1052 return (logical_id & mda) != 0;
1053 case APIC_DFR_CLUSTER:
1054 return ((logical_id >> 4) == (mda >> 4))
1055 && (logical_id & mda & 0xf) != 0;
1056 default:
1057 return false;
1058 }
1059}
1060
1061/* The KVM local APIC implementation has two quirks:
1062 *
1063 * - Real hardware delivers interrupts destined to x2APIC ID > 0xff to LAPICs
1064 * in xAPIC mode if the "destination & 0xff" matches its xAPIC ID.
1065 * KVM doesn't do that aliasing.
1066 *
1067 * - in-kernel IOAPIC messages have to be delivered directly to
1068 * x2APIC, because the kernel does not support interrupt remapping.
1069 * In order to support broadcast without interrupt remapping, x2APIC
1070 * rewrites the destination of non-IPI messages from APIC_BROADCAST
1071 * to X2APIC_BROADCAST.
1072 *
1073 * The broadcast quirk can be disabled with KVM_CAP_X2APIC_API. This is
1074 * important when userspace wants to use x2APIC-format MSIs, because
1075 * APIC_BROADCAST (0xff) is a legal route for "cluster 0, CPUs 0-7".
1076 */
1077static u32 kvm_apic_mda(struct kvm_vcpu *vcpu, unsigned int dest_id,
1078 struct kvm_lapic *source, struct kvm_lapic *target)
1079{
1080 bool ipi = source != NULL;
1081
1082 if (!vcpu->kvm->arch.x2apic_broadcast_quirk_disabled &&
1083 !ipi && dest_id == APIC_BROADCAST && apic_x2apic_mode(target))
1084 return X2APIC_BROADCAST;
1085
1086 return dest_id;
1087}
1088
1089bool kvm_apic_match_dest(struct kvm_vcpu *vcpu, struct kvm_lapic *source,
1090 int shorthand, unsigned int dest, int dest_mode)
1091{
1092 struct kvm_lapic *target = vcpu->arch.apic;
1093 u32 mda = kvm_apic_mda(vcpu, dest, source, target);
1094
1095 ASSERT(target);
1096 switch (shorthand) {
1097 case APIC_DEST_NOSHORT:
1098 if (dest_mode == APIC_DEST_PHYSICAL)
1099 return kvm_apic_match_physical_addr(target, mda);
1100 else
1101 return kvm_apic_match_logical_addr(target, mda);
1102 case APIC_DEST_SELF:
1103 return target == source;
1104 case APIC_DEST_ALLINC:
1105 return true;
1106 case APIC_DEST_ALLBUT:
1107 return target != source;
1108 default:
1109 return false;
1110 }
1111}
1112EXPORT_SYMBOL_GPL(kvm_apic_match_dest);
1113
1114int kvm_vector_to_index(u32 vector, u32 dest_vcpus,
1115 const unsigned long *bitmap, u32 bitmap_size)
1116{
1117 u32 mod;
1118 int i, idx = -1;
1119
1120 mod = vector % dest_vcpus;
1121
1122 for (i = 0; i <= mod; i++) {
1123 idx = find_next_bit(bitmap, bitmap_size, idx + 1);
1124 BUG_ON(idx == bitmap_size);
1125 }
1126
1127 return idx;
1128}
1129
1130static void kvm_apic_disabled_lapic_found(struct kvm *kvm)
1131{
1132 if (!kvm->arch.disabled_lapic_found) {
1133 kvm->arch.disabled_lapic_found = true;
1134 pr_info("Disabled LAPIC found during irq injection\n");
1135 }
1136}
1137
1138static bool kvm_apic_is_broadcast_dest(struct kvm *kvm, struct kvm_lapic **src,
1139 struct kvm_lapic_irq *irq, struct kvm_apic_map *map)
1140{
1141 if (kvm->arch.x2apic_broadcast_quirk_disabled) {
1142 if ((irq->dest_id == APIC_BROADCAST &&
1143 map->logical_mode != KVM_APIC_MODE_X2APIC))
1144 return true;
1145 if (irq->dest_id == X2APIC_BROADCAST)
1146 return true;
1147 } else {
1148 bool x2apic_ipi = src && *src && apic_x2apic_mode(*src);
1149 if (irq->dest_id == (x2apic_ipi ?
1150 X2APIC_BROADCAST : APIC_BROADCAST))
1151 return true;
1152 }
1153
1154 return false;
1155}
1156
1157/* Return true if the interrupt can be handled by using *bitmap as index mask
1158 * for valid destinations in *dst array.
1159 * Return false if kvm_apic_map_get_dest_lapic did nothing useful.
1160 * Note: we may have zero kvm_lapic destinations when we return true, which
1161 * means that the interrupt should be dropped. In this case, *bitmap would be
1162 * zero and *dst undefined.
1163 */
1164static inline bool kvm_apic_map_get_dest_lapic(struct kvm *kvm,
1165 struct kvm_lapic **src, struct kvm_lapic_irq *irq,
1166 struct kvm_apic_map *map, struct kvm_lapic ***dst,
1167 unsigned long *bitmap)
1168{
1169 int i, lowest;
1170
1171 if (irq->shorthand == APIC_DEST_SELF && src) {
1172 *dst = src;
1173 *bitmap = 1;
1174 return true;
1175 } else if (irq->shorthand)
1176 return false;
1177
1178 if (!map || kvm_apic_is_broadcast_dest(kvm, src, irq, map))
1179 return false;
1180
1181 if (irq->dest_mode == APIC_DEST_PHYSICAL) {
1182 if (irq->dest_id > map->max_apic_id) {
1183 *bitmap = 0;
1184 } else {
1185 u32 dest_id = array_index_nospec(irq->dest_id, map->max_apic_id + 1);
1186 *dst = &map->phys_map[dest_id];
1187 *bitmap = 1;
1188 }
1189 return true;
1190 }
1191
1192 *bitmap = 0;
1193 if (!kvm_apic_map_get_logical_dest(map, irq->dest_id, dst,
1194 (u16 *)bitmap))
1195 return false;
1196
1197 if (!kvm_lowest_prio_delivery(irq))
1198 return true;
1199
1200 if (!kvm_vector_hashing_enabled()) {
1201 lowest = -1;
1202 for_each_set_bit(i, bitmap, 16) {
1203 if (!(*dst)[i])
1204 continue;
1205 if (lowest < 0)
1206 lowest = i;
1207 else if (kvm_apic_compare_prio((*dst)[i]->vcpu,
1208 (*dst)[lowest]->vcpu) < 0)
1209 lowest = i;
1210 }
1211 } else {
1212 if (!*bitmap)
1213 return true;
1214
1215 lowest = kvm_vector_to_index(irq->vector, hweight16(*bitmap),
1216 bitmap, 16);
1217
1218 if (!(*dst)[lowest]) {
1219 kvm_apic_disabled_lapic_found(kvm);
1220 *bitmap = 0;
1221 return true;
1222 }
1223 }
1224
1225 *bitmap = (lowest >= 0) ? 1 << lowest : 0;
1226
1227 return true;
1228}
1229
1230bool kvm_irq_delivery_to_apic_fast(struct kvm *kvm, struct kvm_lapic *src,
1231 struct kvm_lapic_irq *irq, int *r, struct dest_map *dest_map)
1232{
1233 struct kvm_apic_map *map;
1234 unsigned long bitmap;
1235 struct kvm_lapic **dst = NULL;
1236 int i;
1237 bool ret;
1238
1239 *r = -1;
1240
1241 if (irq->shorthand == APIC_DEST_SELF) {
1242 if (KVM_BUG_ON(!src, kvm)) {
1243 *r = 0;
1244 return true;
1245 }
1246 *r = kvm_apic_set_irq(src->vcpu, irq, dest_map);
1247 return true;
1248 }
1249
1250 rcu_read_lock();
1251 map = rcu_dereference(kvm->arch.apic_map);
1252
1253 ret = kvm_apic_map_get_dest_lapic(kvm, &src, irq, map, &dst, &bitmap);
1254 if (ret) {
1255 *r = 0;
1256 for_each_set_bit(i, &bitmap, 16) {
1257 if (!dst[i])
1258 continue;
1259 *r += kvm_apic_set_irq(dst[i]->vcpu, irq, dest_map);
1260 }
1261 }
1262
1263 rcu_read_unlock();
1264 return ret;
1265}
1266
1267/*
1268 * This routine tries to handle interrupts in posted mode, here is how
1269 * it deals with different cases:
1270 * - For single-destination interrupts, handle it in posted mode
1271 * - Else if vector hashing is enabled and it is a lowest-priority
1272 * interrupt, handle it in posted mode and use the following mechanism
1273 * to find the destination vCPU.
1274 * 1. For lowest-priority interrupts, store all the possible
1275 * destination vCPUs in an array.
1276 * 2. Use "guest vector % max number of destination vCPUs" to find
1277 * the right destination vCPU in the array for the lowest-priority
1278 * interrupt.
1279 * - Otherwise, use remapped mode to inject the interrupt.
1280 */
1281bool kvm_intr_is_single_vcpu_fast(struct kvm *kvm, struct kvm_lapic_irq *irq,
1282 struct kvm_vcpu **dest_vcpu)
1283{
1284 struct kvm_apic_map *map;
1285 unsigned long bitmap;
1286 struct kvm_lapic **dst = NULL;
1287 bool ret = false;
1288
1289 if (irq->shorthand)
1290 return false;
1291
1292 rcu_read_lock();
1293 map = rcu_dereference(kvm->arch.apic_map);
1294
1295 if (kvm_apic_map_get_dest_lapic(kvm, NULL, irq, map, &dst, &bitmap) &&
1296 hweight16(bitmap) == 1) {
1297 unsigned long i = find_first_bit(&bitmap, 16);
1298
1299 if (dst[i]) {
1300 *dest_vcpu = dst[i]->vcpu;
1301 ret = true;
1302 }
1303 }
1304
1305 rcu_read_unlock();
1306 return ret;
1307}
1308
1309/*
1310 * Add a pending IRQ into lapic.
1311 * Return 1 if successfully added and 0 if discarded.
1312 */
1313static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
1314 int vector, int level, int trig_mode,
1315 struct dest_map *dest_map)
1316{
1317 int result = 0;
1318 struct kvm_vcpu *vcpu = apic->vcpu;
1319
1320 trace_kvm_apic_accept_irq(vcpu->vcpu_id, delivery_mode,
1321 trig_mode, vector);
1322 switch (delivery_mode) {
1323 case APIC_DM_LOWEST:
1324 vcpu->arch.apic_arb_prio++;
1325 fallthrough;
1326 case APIC_DM_FIXED:
1327 if (unlikely(trig_mode && !level))
1328 break;
1329
1330 /* FIXME add logic for vcpu on reset */
1331 if (unlikely(!apic_enabled(apic)))
1332 break;
1333
1334 result = 1;
1335
1336 if (dest_map) {
1337 __set_bit(vcpu->vcpu_id, dest_map->map);
1338 dest_map->vectors[vcpu->vcpu_id] = vector;
1339 }
1340
1341 if (apic_test_vector(vector, apic->regs + APIC_TMR) != !!trig_mode) {
1342 if (trig_mode)
1343 kvm_lapic_set_vector(vector,
1344 apic->regs + APIC_TMR);
1345 else
1346 kvm_lapic_clear_vector(vector,
1347 apic->regs + APIC_TMR);
1348 }
1349
1350 kvm_x86_call(deliver_interrupt)(apic, delivery_mode,
1351 trig_mode, vector);
1352 break;
1353
1354 case APIC_DM_REMRD:
1355 result = 1;
1356 vcpu->arch.pv.pv_unhalted = 1;
1357 kvm_make_request(KVM_REQ_EVENT, vcpu);
1358 kvm_vcpu_kick(vcpu);
1359 break;
1360
1361 case APIC_DM_SMI:
1362 if (!kvm_inject_smi(vcpu)) {
1363 kvm_vcpu_kick(vcpu);
1364 result = 1;
1365 }
1366 break;
1367
1368 case APIC_DM_NMI:
1369 result = 1;
1370 kvm_inject_nmi(vcpu);
1371 kvm_vcpu_kick(vcpu);
1372 break;
1373
1374 case APIC_DM_INIT:
1375 if (!trig_mode || level) {
1376 result = 1;
1377 /* assumes that there are only KVM_APIC_INIT/SIPI */
1378 apic->pending_events = (1UL << KVM_APIC_INIT);
1379 kvm_make_request(KVM_REQ_EVENT, vcpu);
1380 kvm_vcpu_kick(vcpu);
1381 }
1382 break;
1383
1384 case APIC_DM_STARTUP:
1385 result = 1;
1386 apic->sipi_vector = vector;
1387 /* make sure sipi_vector is visible for the receiver */
1388 smp_wmb();
1389 set_bit(KVM_APIC_SIPI, &apic->pending_events);
1390 kvm_make_request(KVM_REQ_EVENT, vcpu);
1391 kvm_vcpu_kick(vcpu);
1392 break;
1393
1394 case APIC_DM_EXTINT:
1395 /*
1396 * Should only be called by kvm_apic_local_deliver() with LVT0,
1397 * before NMI watchdog was enabled. Already handled by
1398 * kvm_apic_accept_pic_intr().
1399 */
1400 break;
1401
1402 default:
1403 printk(KERN_ERR "TODO: unsupported delivery mode %x\n",
1404 delivery_mode);
1405 break;
1406 }
1407 return result;
1408}
1409
1410/*
1411 * This routine identifies the destination vcpus mask meant to receive the
1412 * IOAPIC interrupts. It either uses kvm_apic_map_get_dest_lapic() to find
1413 * out the destination vcpus array and set the bitmap or it traverses to
1414 * each available vcpu to identify the same.
1415 */
1416void kvm_bitmap_or_dest_vcpus(struct kvm *kvm, struct kvm_lapic_irq *irq,
1417 unsigned long *vcpu_bitmap)
1418{
1419 struct kvm_lapic **dest_vcpu = NULL;
1420 struct kvm_lapic *src = NULL;
1421 struct kvm_apic_map *map;
1422 struct kvm_vcpu *vcpu;
1423 unsigned long bitmap, i;
1424 int vcpu_idx;
1425 bool ret;
1426
1427 rcu_read_lock();
1428 map = rcu_dereference(kvm->arch.apic_map);
1429
1430 ret = kvm_apic_map_get_dest_lapic(kvm, &src, irq, map, &dest_vcpu,
1431 &bitmap);
1432 if (ret) {
1433 for_each_set_bit(i, &bitmap, 16) {
1434 if (!dest_vcpu[i])
1435 continue;
1436 vcpu_idx = dest_vcpu[i]->vcpu->vcpu_idx;
1437 __set_bit(vcpu_idx, vcpu_bitmap);
1438 }
1439 } else {
1440 kvm_for_each_vcpu(i, vcpu, kvm) {
1441 if (!kvm_apic_present(vcpu))
1442 continue;
1443 if (!kvm_apic_match_dest(vcpu, NULL,
1444 irq->shorthand,
1445 irq->dest_id,
1446 irq->dest_mode))
1447 continue;
1448 __set_bit(i, vcpu_bitmap);
1449 }
1450 }
1451 rcu_read_unlock();
1452}
1453
1454int kvm_apic_compare_prio(struct kvm_vcpu *vcpu1, struct kvm_vcpu *vcpu2)
1455{
1456 return vcpu1->arch.apic_arb_prio - vcpu2->arch.apic_arb_prio;
1457}
1458
1459static bool kvm_ioapic_handles_vector(struct kvm_lapic *apic, int vector)
1460{
1461 return test_bit(vector, apic->vcpu->arch.ioapic_handled_vectors);
1462}
1463
1464static void kvm_ioapic_send_eoi(struct kvm_lapic *apic, int vector)
1465{
1466 int trigger_mode;
1467
1468 /* Eoi the ioapic only if the ioapic doesn't own the vector. */
1469 if (!kvm_ioapic_handles_vector(apic, vector))
1470 return;
1471
1472 /* Request a KVM exit to inform the userspace IOAPIC. */
1473 if (irqchip_split(apic->vcpu->kvm)) {
1474 apic->vcpu->arch.pending_ioapic_eoi = vector;
1475 kvm_make_request(KVM_REQ_IOAPIC_EOI_EXIT, apic->vcpu);
1476 return;
1477 }
1478
1479 if (apic_test_vector(vector, apic->regs + APIC_TMR))
1480 trigger_mode = IOAPIC_LEVEL_TRIG;
1481 else
1482 trigger_mode = IOAPIC_EDGE_TRIG;
1483
1484 kvm_ioapic_update_eoi(apic->vcpu, vector, trigger_mode);
1485}
1486
1487static int apic_set_eoi(struct kvm_lapic *apic)
1488{
1489 int vector = apic_find_highest_isr(apic);
1490
1491 trace_kvm_eoi(apic, vector);
1492
1493 /*
1494 * Not every write EOI will has corresponding ISR,
1495 * one example is when Kernel check timer on setup_IO_APIC
1496 */
1497 if (vector == -1)
1498 return vector;
1499
1500 apic_clear_isr(vector, apic);
1501 apic_update_ppr(apic);
1502
1503 if (kvm_hv_synic_has_vector(apic->vcpu, vector))
1504 kvm_hv_synic_send_eoi(apic->vcpu, vector);
1505
1506 kvm_ioapic_send_eoi(apic, vector);
1507 kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
1508 return vector;
1509}
1510
1511/*
1512 * this interface assumes a trap-like exit, which has already finished
1513 * desired side effect including vISR and vPPR update.
1514 */
1515void kvm_apic_set_eoi_accelerated(struct kvm_vcpu *vcpu, int vector)
1516{
1517 struct kvm_lapic *apic = vcpu->arch.apic;
1518
1519 trace_kvm_eoi(apic, vector);
1520
1521 kvm_ioapic_send_eoi(apic, vector);
1522 kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
1523}
1524EXPORT_SYMBOL_GPL(kvm_apic_set_eoi_accelerated);
1525
1526void kvm_apic_send_ipi(struct kvm_lapic *apic, u32 icr_low, u32 icr_high)
1527{
1528 struct kvm_lapic_irq irq;
1529
1530 /* KVM has no delay and should always clear the BUSY/PENDING flag. */
1531 WARN_ON_ONCE(icr_low & APIC_ICR_BUSY);
1532
1533 irq.vector = icr_low & APIC_VECTOR_MASK;
1534 irq.delivery_mode = icr_low & APIC_MODE_MASK;
1535 irq.dest_mode = icr_low & APIC_DEST_MASK;
1536 irq.level = (icr_low & APIC_INT_ASSERT) != 0;
1537 irq.trig_mode = icr_low & APIC_INT_LEVELTRIG;
1538 irq.shorthand = icr_low & APIC_SHORT_MASK;
1539 irq.msi_redir_hint = false;
1540 if (apic_x2apic_mode(apic))
1541 irq.dest_id = icr_high;
1542 else
1543 irq.dest_id = GET_XAPIC_DEST_FIELD(icr_high);
1544
1545 trace_kvm_apic_ipi(icr_low, irq.dest_id);
1546
1547 kvm_irq_delivery_to_apic(apic->vcpu->kvm, apic, &irq, NULL);
1548}
1549EXPORT_SYMBOL_GPL(kvm_apic_send_ipi);
1550
1551static u32 apic_get_tmcct(struct kvm_lapic *apic)
1552{
1553 ktime_t remaining, now;
1554 s64 ns;
1555
1556 ASSERT(apic != NULL);
1557
1558 /* if initial count is 0, current count should also be 0 */
1559 if (kvm_lapic_get_reg(apic, APIC_TMICT) == 0 ||
1560 apic->lapic_timer.period == 0)
1561 return 0;
1562
1563 now = ktime_get();
1564 remaining = ktime_sub(apic->lapic_timer.target_expiration, now);
1565 if (ktime_to_ns(remaining) < 0)
1566 remaining = 0;
1567
1568 ns = mod_64(ktime_to_ns(remaining), apic->lapic_timer.period);
1569 return div64_u64(ns, (apic->vcpu->kvm->arch.apic_bus_cycle_ns *
1570 apic->divide_count));
1571}
1572
1573static void __report_tpr_access(struct kvm_lapic *apic, bool write)
1574{
1575 struct kvm_vcpu *vcpu = apic->vcpu;
1576 struct kvm_run *run = vcpu->run;
1577
1578 kvm_make_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu);
1579 run->tpr_access.rip = kvm_rip_read(vcpu);
1580 run->tpr_access.is_write = write;
1581}
1582
1583static inline void report_tpr_access(struct kvm_lapic *apic, bool write)
1584{
1585 if (apic->vcpu->arch.tpr_access_reporting)
1586 __report_tpr_access(apic, write);
1587}
1588
1589static u32 __apic_read(struct kvm_lapic *apic, unsigned int offset)
1590{
1591 u32 val = 0;
1592
1593 if (offset >= LAPIC_MMIO_LENGTH)
1594 return 0;
1595
1596 switch (offset) {
1597 case APIC_ARBPRI:
1598 break;
1599
1600 case APIC_TMCCT: /* Timer CCR */
1601 if (apic_lvtt_tscdeadline(apic))
1602 return 0;
1603
1604 val = apic_get_tmcct(apic);
1605 break;
1606 case APIC_PROCPRI:
1607 apic_update_ppr(apic);
1608 val = kvm_lapic_get_reg(apic, offset);
1609 break;
1610 case APIC_TASKPRI:
1611 report_tpr_access(apic, false);
1612 fallthrough;
1613 default:
1614 val = kvm_lapic_get_reg(apic, offset);
1615 break;
1616 }
1617
1618 return val;
1619}
1620
1621static inline struct kvm_lapic *to_lapic(struct kvm_io_device *dev)
1622{
1623 return container_of(dev, struct kvm_lapic, dev);
1624}
1625
1626#define APIC_REG_MASK(reg) (1ull << ((reg) >> 4))
1627#define APIC_REGS_MASK(first, count) \
1628 (APIC_REG_MASK(first) * ((1ull << (count)) - 1))
1629
1630u64 kvm_lapic_readable_reg_mask(struct kvm_lapic *apic)
1631{
1632 /* Leave bits '0' for reserved and write-only registers. */
1633 u64 valid_reg_mask =
1634 APIC_REG_MASK(APIC_ID) |
1635 APIC_REG_MASK(APIC_LVR) |
1636 APIC_REG_MASK(APIC_TASKPRI) |
1637 APIC_REG_MASK(APIC_PROCPRI) |
1638 APIC_REG_MASK(APIC_LDR) |
1639 APIC_REG_MASK(APIC_SPIV) |
1640 APIC_REGS_MASK(APIC_ISR, APIC_ISR_NR) |
1641 APIC_REGS_MASK(APIC_TMR, APIC_ISR_NR) |
1642 APIC_REGS_MASK(APIC_IRR, APIC_ISR_NR) |
1643 APIC_REG_MASK(APIC_ESR) |
1644 APIC_REG_MASK(APIC_ICR) |
1645 APIC_REG_MASK(APIC_LVTT) |
1646 APIC_REG_MASK(APIC_LVTTHMR) |
1647 APIC_REG_MASK(APIC_LVTPC) |
1648 APIC_REG_MASK(APIC_LVT0) |
1649 APIC_REG_MASK(APIC_LVT1) |
1650 APIC_REG_MASK(APIC_LVTERR) |
1651 APIC_REG_MASK(APIC_TMICT) |
1652 APIC_REG_MASK(APIC_TMCCT) |
1653 APIC_REG_MASK(APIC_TDCR);
1654
1655 if (kvm_lapic_lvt_supported(apic, LVT_CMCI))
1656 valid_reg_mask |= APIC_REG_MASK(APIC_LVTCMCI);
1657
1658 /* ARBPRI, DFR, and ICR2 are not valid in x2APIC mode. */
1659 if (!apic_x2apic_mode(apic))
1660 valid_reg_mask |= APIC_REG_MASK(APIC_ARBPRI) |
1661 APIC_REG_MASK(APIC_DFR) |
1662 APIC_REG_MASK(APIC_ICR2);
1663
1664 return valid_reg_mask;
1665}
1666EXPORT_SYMBOL_GPL(kvm_lapic_readable_reg_mask);
1667
1668static int kvm_lapic_reg_read(struct kvm_lapic *apic, u32 offset, int len,
1669 void *data)
1670{
1671 unsigned char alignment = offset & 0xf;
1672 u32 result;
1673
1674 /*
1675 * WARN if KVM reads ICR in x2APIC mode, as it's an 8-byte register in
1676 * x2APIC and needs to be manually handled by the caller.
1677 */
1678 WARN_ON_ONCE(apic_x2apic_mode(apic) && offset == APIC_ICR);
1679
1680 if (alignment + len > 4)
1681 return 1;
1682
1683 if (offset > 0x3f0 ||
1684 !(kvm_lapic_readable_reg_mask(apic) & APIC_REG_MASK(offset)))
1685 return 1;
1686
1687 result = __apic_read(apic, offset & ~0xf);
1688
1689 trace_kvm_apic_read(offset, result);
1690
1691 switch (len) {
1692 case 1:
1693 case 2:
1694 case 4:
1695 memcpy(data, (char *)&result + alignment, len);
1696 break;
1697 default:
1698 printk(KERN_ERR "Local APIC read with len = %x, "
1699 "should be 1,2, or 4 instead\n", len);
1700 break;
1701 }
1702 return 0;
1703}
1704
1705static int apic_mmio_in_range(struct kvm_lapic *apic, gpa_t addr)
1706{
1707 return addr >= apic->base_address &&
1708 addr < apic->base_address + LAPIC_MMIO_LENGTH;
1709}
1710
1711static int apic_mmio_read(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
1712 gpa_t address, int len, void *data)
1713{
1714 struct kvm_lapic *apic = to_lapic(this);
1715 u32 offset = address - apic->base_address;
1716
1717 if (!apic_mmio_in_range(apic, address))
1718 return -EOPNOTSUPP;
1719
1720 if (!kvm_apic_hw_enabled(apic) || apic_x2apic_mode(apic)) {
1721 if (!kvm_check_has_quirk(vcpu->kvm,
1722 KVM_X86_QUIRK_LAPIC_MMIO_HOLE))
1723 return -EOPNOTSUPP;
1724
1725 memset(data, 0xff, len);
1726 return 0;
1727 }
1728
1729 kvm_lapic_reg_read(apic, offset, len, data);
1730
1731 return 0;
1732}
1733
1734static void update_divide_count(struct kvm_lapic *apic)
1735{
1736 u32 tmp1, tmp2, tdcr;
1737
1738 tdcr = kvm_lapic_get_reg(apic, APIC_TDCR);
1739 tmp1 = tdcr & 0xf;
1740 tmp2 = ((tmp1 & 0x3) | ((tmp1 & 0x8) >> 1)) + 1;
1741 apic->divide_count = 0x1 << (tmp2 & 0x7);
1742}
1743
1744static void limit_periodic_timer_frequency(struct kvm_lapic *apic)
1745{
1746 /*
1747 * Do not allow the guest to program periodic timers with small
1748 * interval, since the hrtimers are not throttled by the host
1749 * scheduler.
1750 */
1751 if (apic_lvtt_period(apic) && apic->lapic_timer.period) {
1752 s64 min_period = min_timer_period_us * 1000LL;
1753
1754 if (apic->lapic_timer.period < min_period) {
1755 pr_info_once(
1756 "vcpu %i: requested %lld ns "
1757 "lapic timer period limited to %lld ns\n",
1758 apic->vcpu->vcpu_id,
1759 apic->lapic_timer.period, min_period);
1760 apic->lapic_timer.period = min_period;
1761 }
1762 }
1763}
1764
1765static void cancel_hv_timer(struct kvm_lapic *apic);
1766
1767static void cancel_apic_timer(struct kvm_lapic *apic)
1768{
1769 hrtimer_cancel(&apic->lapic_timer.timer);
1770 preempt_disable();
1771 if (apic->lapic_timer.hv_timer_in_use)
1772 cancel_hv_timer(apic);
1773 preempt_enable();
1774 atomic_set(&apic->lapic_timer.pending, 0);
1775}
1776
1777static void apic_update_lvtt(struct kvm_lapic *apic)
1778{
1779 u32 timer_mode = kvm_lapic_get_reg(apic, APIC_LVTT) &
1780 apic->lapic_timer.timer_mode_mask;
1781
1782 if (apic->lapic_timer.timer_mode != timer_mode) {
1783 if (apic_lvtt_tscdeadline(apic) != (timer_mode ==
1784 APIC_LVT_TIMER_TSCDEADLINE)) {
1785 cancel_apic_timer(apic);
1786 kvm_lapic_set_reg(apic, APIC_TMICT, 0);
1787 apic->lapic_timer.period = 0;
1788 apic->lapic_timer.tscdeadline = 0;
1789 }
1790 apic->lapic_timer.timer_mode = timer_mode;
1791 limit_periodic_timer_frequency(apic);
1792 }
1793}
1794
1795/*
1796 * On APICv, this test will cause a busy wait
1797 * during a higher-priority task.
1798 */
1799
1800static bool lapic_timer_int_injected(struct kvm_vcpu *vcpu)
1801{
1802 struct kvm_lapic *apic = vcpu->arch.apic;
1803 u32 reg = kvm_lapic_get_reg(apic, APIC_LVTT);
1804
1805 if (kvm_apic_hw_enabled(apic)) {
1806 int vec = reg & APIC_VECTOR_MASK;
1807 void *bitmap = apic->regs + APIC_ISR;
1808
1809 if (apic->apicv_active)
1810 bitmap = apic->regs + APIC_IRR;
1811
1812 if (apic_test_vector(vec, bitmap))
1813 return true;
1814 }
1815 return false;
1816}
1817
1818static inline void __wait_lapic_expire(struct kvm_vcpu *vcpu, u64 guest_cycles)
1819{
1820 u64 timer_advance_ns = vcpu->arch.apic->lapic_timer.timer_advance_ns;
1821
1822 /*
1823 * If the guest TSC is running at a different ratio than the host, then
1824 * convert the delay to nanoseconds to achieve an accurate delay. Note
1825 * that __delay() uses delay_tsc whenever the hardware has TSC, thus
1826 * always for VMX enabled hardware.
1827 */
1828 if (vcpu->arch.tsc_scaling_ratio == kvm_caps.default_tsc_scaling_ratio) {
1829 __delay(min(guest_cycles,
1830 nsec_to_cycles(vcpu, timer_advance_ns)));
1831 } else {
1832 u64 delay_ns = guest_cycles * 1000000ULL;
1833 do_div(delay_ns, vcpu->arch.virtual_tsc_khz);
1834 ndelay(min_t(u32, delay_ns, timer_advance_ns));
1835 }
1836}
1837
1838static inline void adjust_lapic_timer_advance(struct kvm_vcpu *vcpu,
1839 s64 advance_expire_delta)
1840{
1841 struct kvm_lapic *apic = vcpu->arch.apic;
1842 u32 timer_advance_ns = apic->lapic_timer.timer_advance_ns;
1843 u64 ns;
1844
1845 /* Do not adjust for tiny fluctuations or large random spikes. */
1846 if (abs(advance_expire_delta) > LAPIC_TIMER_ADVANCE_ADJUST_MAX ||
1847 abs(advance_expire_delta) < LAPIC_TIMER_ADVANCE_ADJUST_MIN)
1848 return;
1849
1850 /* too early */
1851 if (advance_expire_delta < 0) {
1852 ns = -advance_expire_delta * 1000000ULL;
1853 do_div(ns, vcpu->arch.virtual_tsc_khz);
1854 timer_advance_ns -= ns/LAPIC_TIMER_ADVANCE_ADJUST_STEP;
1855 } else {
1856 /* too late */
1857 ns = advance_expire_delta * 1000000ULL;
1858 do_div(ns, vcpu->arch.virtual_tsc_khz);
1859 timer_advance_ns += ns/LAPIC_TIMER_ADVANCE_ADJUST_STEP;
1860 }
1861
1862 if (unlikely(timer_advance_ns > LAPIC_TIMER_ADVANCE_NS_MAX))
1863 timer_advance_ns = LAPIC_TIMER_ADVANCE_NS_INIT;
1864 apic->lapic_timer.timer_advance_ns = timer_advance_ns;
1865}
1866
1867static void __kvm_wait_lapic_expire(struct kvm_vcpu *vcpu)
1868{
1869 struct kvm_lapic *apic = vcpu->arch.apic;
1870 u64 guest_tsc, tsc_deadline;
1871
1872 tsc_deadline = apic->lapic_timer.expired_tscdeadline;
1873 apic->lapic_timer.expired_tscdeadline = 0;
1874 guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
1875 trace_kvm_wait_lapic_expire(vcpu->vcpu_id, guest_tsc - tsc_deadline);
1876
1877 adjust_lapic_timer_advance(vcpu, guest_tsc - tsc_deadline);
1878
1879 /*
1880 * If the timer fired early, reread the TSC to account for the overhead
1881 * of the above adjustment to avoid waiting longer than is necessary.
1882 */
1883 if (guest_tsc < tsc_deadline)
1884 guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
1885
1886 if (guest_tsc < tsc_deadline)
1887 __wait_lapic_expire(vcpu, tsc_deadline - guest_tsc);
1888}
1889
1890void kvm_wait_lapic_expire(struct kvm_vcpu *vcpu)
1891{
1892 if (lapic_in_kernel(vcpu) &&
1893 vcpu->arch.apic->lapic_timer.expired_tscdeadline &&
1894 vcpu->arch.apic->lapic_timer.timer_advance_ns &&
1895 lapic_timer_int_injected(vcpu))
1896 __kvm_wait_lapic_expire(vcpu);
1897}
1898EXPORT_SYMBOL_GPL(kvm_wait_lapic_expire);
1899
1900static void kvm_apic_inject_pending_timer_irqs(struct kvm_lapic *apic)
1901{
1902 struct kvm_timer *ktimer = &apic->lapic_timer;
1903
1904 kvm_apic_local_deliver(apic, APIC_LVTT);
1905 if (apic_lvtt_tscdeadline(apic)) {
1906 ktimer->tscdeadline = 0;
1907 } else if (apic_lvtt_oneshot(apic)) {
1908 ktimer->tscdeadline = 0;
1909 ktimer->target_expiration = 0;
1910 }
1911}
1912
1913static void apic_timer_expired(struct kvm_lapic *apic, bool from_timer_fn)
1914{
1915 struct kvm_vcpu *vcpu = apic->vcpu;
1916 struct kvm_timer *ktimer = &apic->lapic_timer;
1917
1918 if (atomic_read(&apic->lapic_timer.pending))
1919 return;
1920
1921 if (apic_lvtt_tscdeadline(apic) || ktimer->hv_timer_in_use)
1922 ktimer->expired_tscdeadline = ktimer->tscdeadline;
1923
1924 if (!from_timer_fn && apic->apicv_active) {
1925 WARN_ON(kvm_get_running_vcpu() != vcpu);
1926 kvm_apic_inject_pending_timer_irqs(apic);
1927 return;
1928 }
1929
1930 if (kvm_use_posted_timer_interrupt(apic->vcpu)) {
1931 /*
1932 * Ensure the guest's timer has truly expired before posting an
1933 * interrupt. Open code the relevant checks to avoid querying
1934 * lapic_timer_int_injected(), which will be false since the
1935 * interrupt isn't yet injected. Waiting until after injecting
1936 * is not an option since that won't help a posted interrupt.
1937 */
1938 if (vcpu->arch.apic->lapic_timer.expired_tscdeadline &&
1939 vcpu->arch.apic->lapic_timer.timer_advance_ns)
1940 __kvm_wait_lapic_expire(vcpu);
1941 kvm_apic_inject_pending_timer_irqs(apic);
1942 return;
1943 }
1944
1945 atomic_inc(&apic->lapic_timer.pending);
1946 kvm_make_request(KVM_REQ_UNBLOCK, vcpu);
1947 if (from_timer_fn)
1948 kvm_vcpu_kick(vcpu);
1949}
1950
1951static void start_sw_tscdeadline(struct kvm_lapic *apic)
1952{
1953 struct kvm_timer *ktimer = &apic->lapic_timer;
1954 u64 guest_tsc, tscdeadline = ktimer->tscdeadline;
1955 u64 ns = 0;
1956 ktime_t expire;
1957 struct kvm_vcpu *vcpu = apic->vcpu;
1958 u32 this_tsc_khz = vcpu->arch.virtual_tsc_khz;
1959 unsigned long flags;
1960 ktime_t now;
1961
1962 if (unlikely(!tscdeadline || !this_tsc_khz))
1963 return;
1964
1965 local_irq_save(flags);
1966
1967 now = ktime_get();
1968 guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
1969
1970 ns = (tscdeadline - guest_tsc) * 1000000ULL;
1971 do_div(ns, this_tsc_khz);
1972
1973 if (likely(tscdeadline > guest_tsc) &&
1974 likely(ns > apic->lapic_timer.timer_advance_ns)) {
1975 expire = ktime_add_ns(now, ns);
1976 expire = ktime_sub_ns(expire, ktimer->timer_advance_ns);
1977 hrtimer_start(&ktimer->timer, expire, HRTIMER_MODE_ABS_HARD);
1978 } else
1979 apic_timer_expired(apic, false);
1980
1981 local_irq_restore(flags);
1982}
1983
1984static inline u64 tmict_to_ns(struct kvm_lapic *apic, u32 tmict)
1985{
1986 return (u64)tmict * apic->vcpu->kvm->arch.apic_bus_cycle_ns *
1987 (u64)apic->divide_count;
1988}
1989
1990static void update_target_expiration(struct kvm_lapic *apic, uint32_t old_divisor)
1991{
1992 ktime_t now, remaining;
1993 u64 ns_remaining_old, ns_remaining_new;
1994
1995 apic->lapic_timer.period =
1996 tmict_to_ns(apic, kvm_lapic_get_reg(apic, APIC_TMICT));
1997 limit_periodic_timer_frequency(apic);
1998
1999 now = ktime_get();
2000 remaining = ktime_sub(apic->lapic_timer.target_expiration, now);
2001 if (ktime_to_ns(remaining) < 0)
2002 remaining = 0;
2003
2004 ns_remaining_old = ktime_to_ns(remaining);
2005 ns_remaining_new = mul_u64_u32_div(ns_remaining_old,
2006 apic->divide_count, old_divisor);
2007
2008 apic->lapic_timer.tscdeadline +=
2009 nsec_to_cycles(apic->vcpu, ns_remaining_new) -
2010 nsec_to_cycles(apic->vcpu, ns_remaining_old);
2011 apic->lapic_timer.target_expiration = ktime_add_ns(now, ns_remaining_new);
2012}
2013
2014static bool set_target_expiration(struct kvm_lapic *apic, u32 count_reg)
2015{
2016 ktime_t now;
2017 u64 tscl = rdtsc();
2018 s64 deadline;
2019
2020 now = ktime_get();
2021 apic->lapic_timer.period =
2022 tmict_to_ns(apic, kvm_lapic_get_reg(apic, APIC_TMICT));
2023
2024 if (!apic->lapic_timer.period) {
2025 apic->lapic_timer.tscdeadline = 0;
2026 return false;
2027 }
2028
2029 limit_periodic_timer_frequency(apic);
2030 deadline = apic->lapic_timer.period;
2031
2032 if (apic_lvtt_period(apic) || apic_lvtt_oneshot(apic)) {
2033 if (unlikely(count_reg != APIC_TMICT)) {
2034 deadline = tmict_to_ns(apic,
2035 kvm_lapic_get_reg(apic, count_reg));
2036 if (unlikely(deadline <= 0)) {
2037 if (apic_lvtt_period(apic))
2038 deadline = apic->lapic_timer.period;
2039 else
2040 deadline = 0;
2041 }
2042 else if (unlikely(deadline > apic->lapic_timer.period)) {
2043 pr_info_ratelimited(
2044 "vcpu %i: requested lapic timer restore with "
2045 "starting count register %#x=%u (%lld ns) > initial count (%lld ns). "
2046 "Using initial count to start timer.\n",
2047 apic->vcpu->vcpu_id,
2048 count_reg,
2049 kvm_lapic_get_reg(apic, count_reg),
2050 deadline, apic->lapic_timer.period);
2051 kvm_lapic_set_reg(apic, count_reg, 0);
2052 deadline = apic->lapic_timer.period;
2053 }
2054 }
2055 }
2056
2057 apic->lapic_timer.tscdeadline = kvm_read_l1_tsc(apic->vcpu, tscl) +
2058 nsec_to_cycles(apic->vcpu, deadline);
2059 apic->lapic_timer.target_expiration = ktime_add_ns(now, deadline);
2060
2061 return true;
2062}
2063
2064static void advance_periodic_target_expiration(struct kvm_lapic *apic)
2065{
2066 ktime_t now = ktime_get();
2067 u64 tscl = rdtsc();
2068 ktime_t delta;
2069
2070 /*
2071 * Synchronize both deadlines to the same time source or
2072 * differences in the periods (caused by differences in the
2073 * underlying clocks or numerical approximation errors) will
2074 * cause the two to drift apart over time as the errors
2075 * accumulate.
2076 */
2077 apic->lapic_timer.target_expiration =
2078 ktime_add_ns(apic->lapic_timer.target_expiration,
2079 apic->lapic_timer.period);
2080 delta = ktime_sub(apic->lapic_timer.target_expiration, now);
2081 apic->lapic_timer.tscdeadline = kvm_read_l1_tsc(apic->vcpu, tscl) +
2082 nsec_to_cycles(apic->vcpu, delta);
2083}
2084
2085static void start_sw_period(struct kvm_lapic *apic)
2086{
2087 if (!apic->lapic_timer.period)
2088 return;
2089
2090 if (ktime_after(ktime_get(),
2091 apic->lapic_timer.target_expiration)) {
2092 apic_timer_expired(apic, false);
2093
2094 if (apic_lvtt_oneshot(apic))
2095 return;
2096
2097 advance_periodic_target_expiration(apic);
2098 }
2099
2100 hrtimer_start(&apic->lapic_timer.timer,
2101 apic->lapic_timer.target_expiration,
2102 HRTIMER_MODE_ABS_HARD);
2103}
2104
2105bool kvm_lapic_hv_timer_in_use(struct kvm_vcpu *vcpu)
2106{
2107 if (!lapic_in_kernel(vcpu))
2108 return false;
2109
2110 return vcpu->arch.apic->lapic_timer.hv_timer_in_use;
2111}
2112
2113static void cancel_hv_timer(struct kvm_lapic *apic)
2114{
2115 WARN_ON(preemptible());
2116 WARN_ON(!apic->lapic_timer.hv_timer_in_use);
2117 kvm_x86_call(cancel_hv_timer)(apic->vcpu);
2118 apic->lapic_timer.hv_timer_in_use = false;
2119}
2120
2121static bool start_hv_timer(struct kvm_lapic *apic)
2122{
2123 struct kvm_timer *ktimer = &apic->lapic_timer;
2124 struct kvm_vcpu *vcpu = apic->vcpu;
2125 bool expired;
2126
2127 WARN_ON(preemptible());
2128 if (!kvm_can_use_hv_timer(vcpu))
2129 return false;
2130
2131 if (!ktimer->tscdeadline)
2132 return false;
2133
2134 if (kvm_x86_call(set_hv_timer)(vcpu, ktimer->tscdeadline, &expired))
2135 return false;
2136
2137 ktimer->hv_timer_in_use = true;
2138 hrtimer_cancel(&ktimer->timer);
2139
2140 /*
2141 * To simplify handling the periodic timer, leave the hv timer running
2142 * even if the deadline timer has expired, i.e. rely on the resulting
2143 * VM-Exit to recompute the periodic timer's target expiration.
2144 */
2145 if (!apic_lvtt_period(apic)) {
2146 /*
2147 * Cancel the hv timer if the sw timer fired while the hv timer
2148 * was being programmed, or if the hv timer itself expired.
2149 */
2150 if (atomic_read(&ktimer->pending)) {
2151 cancel_hv_timer(apic);
2152 } else if (expired) {
2153 apic_timer_expired(apic, false);
2154 cancel_hv_timer(apic);
2155 }
2156 }
2157
2158 trace_kvm_hv_timer_state(vcpu->vcpu_id, ktimer->hv_timer_in_use);
2159
2160 return true;
2161}
2162
2163static void start_sw_timer(struct kvm_lapic *apic)
2164{
2165 struct kvm_timer *ktimer = &apic->lapic_timer;
2166
2167 WARN_ON(preemptible());
2168 if (apic->lapic_timer.hv_timer_in_use)
2169 cancel_hv_timer(apic);
2170 if (!apic_lvtt_period(apic) && atomic_read(&ktimer->pending))
2171 return;
2172
2173 if (apic_lvtt_period(apic) || apic_lvtt_oneshot(apic))
2174 start_sw_period(apic);
2175 else if (apic_lvtt_tscdeadline(apic))
2176 start_sw_tscdeadline(apic);
2177 trace_kvm_hv_timer_state(apic->vcpu->vcpu_id, false);
2178}
2179
2180static void restart_apic_timer(struct kvm_lapic *apic)
2181{
2182 preempt_disable();
2183
2184 if (!apic_lvtt_period(apic) && atomic_read(&apic->lapic_timer.pending))
2185 goto out;
2186
2187 if (!start_hv_timer(apic))
2188 start_sw_timer(apic);
2189out:
2190 preempt_enable();
2191}
2192
2193void kvm_lapic_expired_hv_timer(struct kvm_vcpu *vcpu)
2194{
2195 struct kvm_lapic *apic = vcpu->arch.apic;
2196
2197 preempt_disable();
2198 /* If the preempt notifier has already run, it also called apic_timer_expired */
2199 if (!apic->lapic_timer.hv_timer_in_use)
2200 goto out;
2201 WARN_ON(kvm_vcpu_is_blocking(vcpu));
2202 apic_timer_expired(apic, false);
2203 cancel_hv_timer(apic);
2204
2205 if (apic_lvtt_period(apic) && apic->lapic_timer.period) {
2206 advance_periodic_target_expiration(apic);
2207 restart_apic_timer(apic);
2208 }
2209out:
2210 preempt_enable();
2211}
2212EXPORT_SYMBOL_GPL(kvm_lapic_expired_hv_timer);
2213
2214void kvm_lapic_switch_to_hv_timer(struct kvm_vcpu *vcpu)
2215{
2216 restart_apic_timer(vcpu->arch.apic);
2217}
2218
2219void kvm_lapic_switch_to_sw_timer(struct kvm_vcpu *vcpu)
2220{
2221 struct kvm_lapic *apic = vcpu->arch.apic;
2222
2223 preempt_disable();
2224 /* Possibly the TSC deadline timer is not enabled yet */
2225 if (apic->lapic_timer.hv_timer_in_use)
2226 start_sw_timer(apic);
2227 preempt_enable();
2228}
2229
2230void kvm_lapic_restart_hv_timer(struct kvm_vcpu *vcpu)
2231{
2232 struct kvm_lapic *apic = vcpu->arch.apic;
2233
2234 WARN_ON(!apic->lapic_timer.hv_timer_in_use);
2235 restart_apic_timer(apic);
2236}
2237
2238static void __start_apic_timer(struct kvm_lapic *apic, u32 count_reg)
2239{
2240 atomic_set(&apic->lapic_timer.pending, 0);
2241
2242 if ((apic_lvtt_period(apic) || apic_lvtt_oneshot(apic))
2243 && !set_target_expiration(apic, count_reg))
2244 return;
2245
2246 restart_apic_timer(apic);
2247}
2248
2249static void start_apic_timer(struct kvm_lapic *apic)
2250{
2251 __start_apic_timer(apic, APIC_TMICT);
2252}
2253
2254static void apic_manage_nmi_watchdog(struct kvm_lapic *apic, u32 lvt0_val)
2255{
2256 bool lvt0_in_nmi_mode = apic_lvt_nmi_mode(lvt0_val);
2257
2258 if (apic->lvt0_in_nmi_mode != lvt0_in_nmi_mode) {
2259 apic->lvt0_in_nmi_mode = lvt0_in_nmi_mode;
2260 if (lvt0_in_nmi_mode) {
2261 atomic_inc(&apic->vcpu->kvm->arch.vapics_in_nmi_mode);
2262 } else
2263 atomic_dec(&apic->vcpu->kvm->arch.vapics_in_nmi_mode);
2264 }
2265}
2266
2267static int get_lvt_index(u32 reg)
2268{
2269 if (reg == APIC_LVTCMCI)
2270 return LVT_CMCI;
2271 if (reg < APIC_LVTT || reg > APIC_LVTERR)
2272 return -1;
2273 return array_index_nospec(
2274 (reg - APIC_LVTT) >> 4, KVM_APIC_MAX_NR_LVT_ENTRIES);
2275}
2276
2277static int kvm_lapic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val)
2278{
2279 int ret = 0;
2280
2281 trace_kvm_apic_write(reg, val);
2282
2283 switch (reg) {
2284 case APIC_ID: /* Local APIC ID */
2285 if (!apic_x2apic_mode(apic)) {
2286 kvm_apic_set_xapic_id(apic, val >> 24);
2287 } else {
2288 ret = 1;
2289 }
2290 break;
2291
2292 case APIC_TASKPRI:
2293 report_tpr_access(apic, true);
2294 apic_set_tpr(apic, val & 0xff);
2295 break;
2296
2297 case APIC_EOI:
2298 apic_set_eoi(apic);
2299 break;
2300
2301 case APIC_LDR:
2302 if (!apic_x2apic_mode(apic))
2303 kvm_apic_set_ldr(apic, val & APIC_LDR_MASK);
2304 else
2305 ret = 1;
2306 break;
2307
2308 case APIC_DFR:
2309 if (!apic_x2apic_mode(apic))
2310 kvm_apic_set_dfr(apic, val | 0x0FFFFFFF);
2311 else
2312 ret = 1;
2313 break;
2314
2315 case APIC_SPIV: {
2316 u32 mask = 0x3ff;
2317 if (kvm_lapic_get_reg(apic, APIC_LVR) & APIC_LVR_DIRECTED_EOI)
2318 mask |= APIC_SPIV_DIRECTED_EOI;
2319 apic_set_spiv(apic, val & mask);
2320 if (!(val & APIC_SPIV_APIC_ENABLED)) {
2321 int i;
2322
2323 for (i = 0; i < apic->nr_lvt_entries; i++) {
2324 kvm_lapic_set_reg(apic, APIC_LVTx(i),
2325 kvm_lapic_get_reg(apic, APIC_LVTx(i)) | APIC_LVT_MASKED);
2326 }
2327 apic_update_lvtt(apic);
2328 atomic_set(&apic->lapic_timer.pending, 0);
2329
2330 }
2331 break;
2332 }
2333 case APIC_ICR:
2334 WARN_ON_ONCE(apic_x2apic_mode(apic));
2335
2336 /* No delay here, so we always clear the pending bit */
2337 val &= ~APIC_ICR_BUSY;
2338 kvm_apic_send_ipi(apic, val, kvm_lapic_get_reg(apic, APIC_ICR2));
2339 kvm_lapic_set_reg(apic, APIC_ICR, val);
2340 break;
2341 case APIC_ICR2:
2342 if (apic_x2apic_mode(apic))
2343 ret = 1;
2344 else
2345 kvm_lapic_set_reg(apic, APIC_ICR2, val & 0xff000000);
2346 break;
2347
2348 case APIC_LVT0:
2349 apic_manage_nmi_watchdog(apic, val);
2350 fallthrough;
2351 case APIC_LVTTHMR:
2352 case APIC_LVTPC:
2353 case APIC_LVT1:
2354 case APIC_LVTERR:
2355 case APIC_LVTCMCI: {
2356 u32 index = get_lvt_index(reg);
2357 if (!kvm_lapic_lvt_supported(apic, index)) {
2358 ret = 1;
2359 break;
2360 }
2361 if (!kvm_apic_sw_enabled(apic))
2362 val |= APIC_LVT_MASKED;
2363 val &= apic_lvt_mask[index];
2364 kvm_lapic_set_reg(apic, reg, val);
2365 break;
2366 }
2367
2368 case APIC_LVTT:
2369 if (!kvm_apic_sw_enabled(apic))
2370 val |= APIC_LVT_MASKED;
2371 val &= (apic_lvt_mask[0] | apic->lapic_timer.timer_mode_mask);
2372 kvm_lapic_set_reg(apic, APIC_LVTT, val);
2373 apic_update_lvtt(apic);
2374 break;
2375
2376 case APIC_TMICT:
2377 if (apic_lvtt_tscdeadline(apic))
2378 break;
2379
2380 cancel_apic_timer(apic);
2381 kvm_lapic_set_reg(apic, APIC_TMICT, val);
2382 start_apic_timer(apic);
2383 break;
2384
2385 case APIC_TDCR: {
2386 uint32_t old_divisor = apic->divide_count;
2387
2388 kvm_lapic_set_reg(apic, APIC_TDCR, val & 0xb);
2389 update_divide_count(apic);
2390 if (apic->divide_count != old_divisor &&
2391 apic->lapic_timer.period) {
2392 hrtimer_cancel(&apic->lapic_timer.timer);
2393 update_target_expiration(apic, old_divisor);
2394 restart_apic_timer(apic);
2395 }
2396 break;
2397 }
2398 case APIC_ESR:
2399 if (apic_x2apic_mode(apic) && val != 0)
2400 ret = 1;
2401 break;
2402
2403 case APIC_SELF_IPI:
2404 /*
2405 * Self-IPI exists only when x2APIC is enabled. Bits 7:0 hold
2406 * the vector, everything else is reserved.
2407 */
2408 if (!apic_x2apic_mode(apic) || (val & ~APIC_VECTOR_MASK))
2409 ret = 1;
2410 else
2411 kvm_apic_send_ipi(apic, APIC_DEST_SELF | val, 0);
2412 break;
2413 default:
2414 ret = 1;
2415 break;
2416 }
2417
2418 /*
2419 * Recalculate APIC maps if necessary, e.g. if the software enable bit
2420 * was toggled, the APIC ID changed, etc... The maps are marked dirty
2421 * on relevant changes, i.e. this is a nop for most writes.
2422 */
2423 kvm_recalculate_apic_map(apic->vcpu->kvm);
2424
2425 return ret;
2426}
2427
2428static int apic_mmio_write(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
2429 gpa_t address, int len, const void *data)
2430{
2431 struct kvm_lapic *apic = to_lapic(this);
2432 unsigned int offset = address - apic->base_address;
2433 u32 val;
2434
2435 if (!apic_mmio_in_range(apic, address))
2436 return -EOPNOTSUPP;
2437
2438 if (!kvm_apic_hw_enabled(apic) || apic_x2apic_mode(apic)) {
2439 if (!kvm_check_has_quirk(vcpu->kvm,
2440 KVM_X86_QUIRK_LAPIC_MMIO_HOLE))
2441 return -EOPNOTSUPP;
2442
2443 return 0;
2444 }
2445
2446 /*
2447 * APIC register must be aligned on 128-bits boundary.
2448 * 32/64/128 bits registers must be accessed thru 32 bits.
2449 * Refer SDM 8.4.1
2450 */
2451 if (len != 4 || (offset & 0xf))
2452 return 0;
2453
2454 val = *(u32*)data;
2455
2456 kvm_lapic_reg_write(apic, offset & 0xff0, val);
2457
2458 return 0;
2459}
2460
2461void kvm_lapic_set_eoi(struct kvm_vcpu *vcpu)
2462{
2463 kvm_lapic_reg_write(vcpu->arch.apic, APIC_EOI, 0);
2464}
2465EXPORT_SYMBOL_GPL(kvm_lapic_set_eoi);
2466
2467#define X2APIC_ICR_RESERVED_BITS (GENMASK_ULL(31, 20) | GENMASK_ULL(17, 16) | BIT(13))
2468
2469int kvm_x2apic_icr_write(struct kvm_lapic *apic, u64 data)
2470{
2471 if (data & X2APIC_ICR_RESERVED_BITS)
2472 return 1;
2473
2474 /*
2475 * The BUSY bit is reserved on both Intel and AMD in x2APIC mode, but
2476 * only AMD requires it to be zero, Intel essentially just ignores the
2477 * bit. And if IPI virtualization (Intel) or x2AVIC (AMD) is enabled,
2478 * the CPU performs the reserved bits checks, i.e. the underlying CPU
2479 * behavior will "win". Arbitrarily clear the BUSY bit, as there is no
2480 * sane way to provide consistent behavior with respect to hardware.
2481 */
2482 data &= ~APIC_ICR_BUSY;
2483
2484 kvm_apic_send_ipi(apic, (u32)data, (u32)(data >> 32));
2485 if (kvm_x86_ops.x2apic_icr_is_split) {
2486 kvm_lapic_set_reg(apic, APIC_ICR, data);
2487 kvm_lapic_set_reg(apic, APIC_ICR2, data >> 32);
2488 } else {
2489 kvm_lapic_set_reg64(apic, APIC_ICR, data);
2490 }
2491 trace_kvm_apic_write(APIC_ICR, data);
2492 return 0;
2493}
2494
2495static u64 kvm_x2apic_icr_read(struct kvm_lapic *apic)
2496{
2497 if (kvm_x86_ops.x2apic_icr_is_split)
2498 return (u64)kvm_lapic_get_reg(apic, APIC_ICR) |
2499 (u64)kvm_lapic_get_reg(apic, APIC_ICR2) << 32;
2500
2501 return kvm_lapic_get_reg64(apic, APIC_ICR);
2502}
2503
2504/* emulate APIC access in a trap manner */
2505void kvm_apic_write_nodecode(struct kvm_vcpu *vcpu, u32 offset)
2506{
2507 struct kvm_lapic *apic = vcpu->arch.apic;
2508
2509 /*
2510 * ICR is a single 64-bit register when x2APIC is enabled, all others
2511 * registers hold 32-bit values. For legacy xAPIC, ICR writes need to
2512 * go down the common path to get the upper half from ICR2.
2513 *
2514 * Note, using the write helpers may incur an unnecessary write to the
2515 * virtual APIC state, but KVM needs to conditionally modify the value
2516 * in certain cases, e.g. to clear the ICR busy bit. The cost of extra
2517 * conditional branches is likely a wash relative to the cost of the
2518 * maybe-unecessary write, and both are in the noise anyways.
2519 */
2520 if (apic_x2apic_mode(apic) && offset == APIC_ICR)
2521 WARN_ON_ONCE(kvm_x2apic_icr_write(apic, kvm_x2apic_icr_read(apic)));
2522 else
2523 kvm_lapic_reg_write(apic, offset, kvm_lapic_get_reg(apic, offset));
2524}
2525EXPORT_SYMBOL_GPL(kvm_apic_write_nodecode);
2526
2527void kvm_free_lapic(struct kvm_vcpu *vcpu)
2528{
2529 struct kvm_lapic *apic = vcpu->arch.apic;
2530
2531 if (!vcpu->arch.apic) {
2532 static_branch_dec(&kvm_has_noapic_vcpu);
2533 return;
2534 }
2535
2536 hrtimer_cancel(&apic->lapic_timer.timer);
2537
2538 if (!(vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE))
2539 static_branch_slow_dec_deferred(&apic_hw_disabled);
2540
2541 if (!apic->sw_enabled)
2542 static_branch_slow_dec_deferred(&apic_sw_disabled);
2543
2544 if (apic->regs)
2545 free_page((unsigned long)apic->regs);
2546
2547 kfree(apic);
2548}
2549
2550/*
2551 *----------------------------------------------------------------------
2552 * LAPIC interface
2553 *----------------------------------------------------------------------
2554 */
2555u64 kvm_get_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu)
2556{
2557 struct kvm_lapic *apic = vcpu->arch.apic;
2558
2559 if (!kvm_apic_present(vcpu) || !apic_lvtt_tscdeadline(apic))
2560 return 0;
2561
2562 return apic->lapic_timer.tscdeadline;
2563}
2564
2565void kvm_set_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu, u64 data)
2566{
2567 struct kvm_lapic *apic = vcpu->arch.apic;
2568
2569 if (!kvm_apic_present(vcpu) || !apic_lvtt_tscdeadline(apic))
2570 return;
2571
2572 hrtimer_cancel(&apic->lapic_timer.timer);
2573 apic->lapic_timer.tscdeadline = data;
2574 start_apic_timer(apic);
2575}
2576
2577void kvm_lapic_set_tpr(struct kvm_vcpu *vcpu, unsigned long cr8)
2578{
2579 apic_set_tpr(vcpu->arch.apic, (cr8 & 0x0f) << 4);
2580}
2581
2582u64 kvm_lapic_get_cr8(struct kvm_vcpu *vcpu)
2583{
2584 u64 tpr;
2585
2586 tpr = (u64) kvm_lapic_get_reg(vcpu->arch.apic, APIC_TASKPRI);
2587
2588 return (tpr & 0xf0) >> 4;
2589}
2590
2591static void __kvm_apic_set_base(struct kvm_vcpu *vcpu, u64 value)
2592{
2593 u64 old_value = vcpu->arch.apic_base;
2594 struct kvm_lapic *apic = vcpu->arch.apic;
2595
2596 vcpu->arch.apic_base = value;
2597
2598 if ((old_value ^ value) & MSR_IA32_APICBASE_ENABLE)
2599 kvm_update_cpuid_runtime(vcpu);
2600
2601 if (!apic)
2602 return;
2603
2604 /* update jump label if enable bit changes */
2605 if ((old_value ^ value) & MSR_IA32_APICBASE_ENABLE) {
2606 if (value & MSR_IA32_APICBASE_ENABLE) {
2607 kvm_apic_set_xapic_id(apic, vcpu->vcpu_id);
2608 static_branch_slow_dec_deferred(&apic_hw_disabled);
2609 /* Check if there are APF page ready requests pending */
2610 kvm_make_request(KVM_REQ_APF_READY, vcpu);
2611 } else {
2612 static_branch_inc(&apic_hw_disabled.key);
2613 atomic_set_release(&apic->vcpu->kvm->arch.apic_map_dirty, DIRTY);
2614 }
2615 }
2616
2617 if ((old_value ^ value) & X2APIC_ENABLE) {
2618 if (value & X2APIC_ENABLE)
2619 kvm_apic_set_x2apic_id(apic, vcpu->vcpu_id);
2620 else if (value & MSR_IA32_APICBASE_ENABLE)
2621 kvm_apic_set_xapic_id(apic, vcpu->vcpu_id);
2622 }
2623
2624 if ((old_value ^ value) & (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE)) {
2625 kvm_make_request(KVM_REQ_APICV_UPDATE, vcpu);
2626 kvm_x86_call(set_virtual_apic_mode)(vcpu);
2627 }
2628
2629 apic->base_address = apic->vcpu->arch.apic_base &
2630 MSR_IA32_APICBASE_BASE;
2631
2632 if ((value & MSR_IA32_APICBASE_ENABLE) &&
2633 apic->base_address != APIC_DEFAULT_PHYS_BASE) {
2634 kvm_set_apicv_inhibit(apic->vcpu->kvm,
2635 APICV_INHIBIT_REASON_APIC_BASE_MODIFIED);
2636 }
2637}
2638
2639int kvm_apic_set_base(struct kvm_vcpu *vcpu, u64 value, bool host_initiated)
2640{
2641 enum lapic_mode old_mode = kvm_get_apic_mode(vcpu);
2642 enum lapic_mode new_mode = kvm_apic_mode(value);
2643
2644 if (vcpu->arch.apic_base == value)
2645 return 0;
2646
2647 u64 reserved_bits = kvm_vcpu_reserved_gpa_bits_raw(vcpu) | 0x2ff |
2648 (guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
2649
2650 if ((value & reserved_bits) != 0 || new_mode == LAPIC_MODE_INVALID)
2651 return 1;
2652 if (!host_initiated) {
2653 if (old_mode == LAPIC_MODE_X2APIC && new_mode == LAPIC_MODE_XAPIC)
2654 return 1;
2655 if (old_mode == LAPIC_MODE_DISABLED && new_mode == LAPIC_MODE_X2APIC)
2656 return 1;
2657 }
2658
2659 __kvm_apic_set_base(vcpu, value);
2660 kvm_recalculate_apic_map(vcpu->kvm);
2661 return 0;
2662}
2663
2664void kvm_apic_update_apicv(struct kvm_vcpu *vcpu)
2665{
2666 struct kvm_lapic *apic = vcpu->arch.apic;
2667
2668 /*
2669 * When APICv is enabled, KVM must always search the IRR for a pending
2670 * IRQ, as other vCPUs and devices can set IRR bits even if the vCPU
2671 * isn't running. If APICv is disabled, KVM _should_ search the IRR
2672 * for a pending IRQ. But KVM currently doesn't ensure *all* hardware,
2673 * e.g. CPUs and IOMMUs, has seen the change in state, i.e. searching
2674 * the IRR at this time could race with IRQ delivery from hardware that
2675 * still sees APICv as being enabled.
2676 *
2677 * FIXME: Ensure other vCPUs and devices observe the change in APICv
2678 * state prior to updating KVM's metadata caches, so that KVM
2679 * can safely search the IRR and set irr_pending accordingly.
2680 */
2681 apic->irr_pending = true;
2682
2683 if (apic->apicv_active)
2684 apic->isr_count = 1;
2685 else
2686 apic->isr_count = count_vectors(apic->regs + APIC_ISR);
2687
2688 apic->highest_isr_cache = -1;
2689}
2690
2691int kvm_alloc_apic_access_page(struct kvm *kvm)
2692{
2693 void __user *hva;
2694 int ret = 0;
2695
2696 mutex_lock(&kvm->slots_lock);
2697 if (kvm->arch.apic_access_memslot_enabled ||
2698 kvm->arch.apic_access_memslot_inhibited)
2699 goto out;
2700
2701 hva = __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
2702 APIC_DEFAULT_PHYS_BASE, PAGE_SIZE);
2703 if (IS_ERR(hva)) {
2704 ret = PTR_ERR(hva);
2705 goto out;
2706 }
2707
2708 kvm->arch.apic_access_memslot_enabled = true;
2709out:
2710 mutex_unlock(&kvm->slots_lock);
2711 return ret;
2712}
2713EXPORT_SYMBOL_GPL(kvm_alloc_apic_access_page);
2714
2715void kvm_inhibit_apic_access_page(struct kvm_vcpu *vcpu)
2716{
2717 struct kvm *kvm = vcpu->kvm;
2718
2719 if (!kvm->arch.apic_access_memslot_enabled)
2720 return;
2721
2722 kvm_vcpu_srcu_read_unlock(vcpu);
2723
2724 mutex_lock(&kvm->slots_lock);
2725
2726 if (kvm->arch.apic_access_memslot_enabled) {
2727 __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT, 0, 0);
2728 /*
2729 * Clear "enabled" after the memslot is deleted so that a
2730 * different vCPU doesn't get a false negative when checking
2731 * the flag out of slots_lock. No additional memory barrier is
2732 * needed as modifying memslots requires waiting other vCPUs to
2733 * drop SRCU (see above), and false positives are ok as the
2734 * flag is rechecked after acquiring slots_lock.
2735 */
2736 kvm->arch.apic_access_memslot_enabled = false;
2737
2738 /*
2739 * Mark the memslot as inhibited to prevent reallocating the
2740 * memslot during vCPU creation, e.g. if a vCPU is hotplugged.
2741 */
2742 kvm->arch.apic_access_memslot_inhibited = true;
2743 }
2744
2745 mutex_unlock(&kvm->slots_lock);
2746
2747 kvm_vcpu_srcu_read_lock(vcpu);
2748}
2749
2750void kvm_lapic_reset(struct kvm_vcpu *vcpu, bool init_event)
2751{
2752 struct kvm_lapic *apic = vcpu->arch.apic;
2753 u64 msr_val;
2754 int i;
2755
2756 kvm_x86_call(apicv_pre_state_restore)(vcpu);
2757
2758 if (!init_event) {
2759 msr_val = APIC_DEFAULT_PHYS_BASE | MSR_IA32_APICBASE_ENABLE;
2760 if (kvm_vcpu_is_reset_bsp(vcpu))
2761 msr_val |= MSR_IA32_APICBASE_BSP;
2762
2763 /*
2764 * Use the inner helper to avoid an extra recalcuation of the
2765 * optimized APIC map if some other task has dirtied the map.
2766 * The recalculation needed for this vCPU will be done after
2767 * all APIC state has been initialized (see below).
2768 */
2769 __kvm_apic_set_base(vcpu, msr_val);
2770 }
2771
2772 if (!apic)
2773 return;
2774
2775 /* Stop the timer in case it's a reset to an active apic */
2776 hrtimer_cancel(&apic->lapic_timer.timer);
2777
2778 /* The xAPIC ID is set at RESET even if the APIC was already enabled. */
2779 if (!init_event)
2780 kvm_apic_set_xapic_id(apic, vcpu->vcpu_id);
2781 kvm_apic_set_version(apic->vcpu);
2782
2783 for (i = 0; i < apic->nr_lvt_entries; i++)
2784 kvm_lapic_set_reg(apic, APIC_LVTx(i), APIC_LVT_MASKED);
2785 apic_update_lvtt(apic);
2786 if (kvm_vcpu_is_reset_bsp(vcpu) &&
2787 kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_LINT0_REENABLED))
2788 kvm_lapic_set_reg(apic, APIC_LVT0,
2789 SET_APIC_DELIVERY_MODE(0, APIC_MODE_EXTINT));
2790 apic_manage_nmi_watchdog(apic, kvm_lapic_get_reg(apic, APIC_LVT0));
2791
2792 kvm_apic_set_dfr(apic, 0xffffffffU);
2793 apic_set_spiv(apic, 0xff);
2794 kvm_lapic_set_reg(apic, APIC_TASKPRI, 0);
2795 if (!apic_x2apic_mode(apic))
2796 kvm_apic_set_ldr(apic, 0);
2797 kvm_lapic_set_reg(apic, APIC_ESR, 0);
2798 if (!apic_x2apic_mode(apic)) {
2799 kvm_lapic_set_reg(apic, APIC_ICR, 0);
2800 kvm_lapic_set_reg(apic, APIC_ICR2, 0);
2801 } else {
2802 kvm_lapic_set_reg64(apic, APIC_ICR, 0);
2803 }
2804 kvm_lapic_set_reg(apic, APIC_TDCR, 0);
2805 kvm_lapic_set_reg(apic, APIC_TMICT, 0);
2806 for (i = 0; i < 8; i++) {
2807 kvm_lapic_set_reg(apic, APIC_IRR + 0x10 * i, 0);
2808 kvm_lapic_set_reg(apic, APIC_ISR + 0x10 * i, 0);
2809 kvm_lapic_set_reg(apic, APIC_TMR + 0x10 * i, 0);
2810 }
2811 kvm_apic_update_apicv(vcpu);
2812 update_divide_count(apic);
2813 atomic_set(&apic->lapic_timer.pending, 0);
2814
2815 vcpu->arch.pv_eoi.msr_val = 0;
2816 apic_update_ppr(apic);
2817 if (apic->apicv_active) {
2818 kvm_x86_call(apicv_post_state_restore)(vcpu);
2819 kvm_x86_call(hwapic_irr_update)(vcpu, -1);
2820 kvm_x86_call(hwapic_isr_update)(vcpu, -1);
2821 }
2822
2823 vcpu->arch.apic_arb_prio = 0;
2824 vcpu->arch.apic_attention = 0;
2825
2826 kvm_recalculate_apic_map(vcpu->kvm);
2827}
2828
2829/*
2830 *----------------------------------------------------------------------
2831 * timer interface
2832 *----------------------------------------------------------------------
2833 */
2834
2835static bool lapic_is_periodic(struct kvm_lapic *apic)
2836{
2837 return apic_lvtt_period(apic);
2838}
2839
2840int apic_has_pending_timer(struct kvm_vcpu *vcpu)
2841{
2842 struct kvm_lapic *apic = vcpu->arch.apic;
2843
2844 if (apic_enabled(apic) && apic_lvt_enabled(apic, APIC_LVTT))
2845 return atomic_read(&apic->lapic_timer.pending);
2846
2847 return 0;
2848}
2849
2850int kvm_apic_local_deliver(struct kvm_lapic *apic, int lvt_type)
2851{
2852 u32 reg = kvm_lapic_get_reg(apic, lvt_type);
2853 int vector, mode, trig_mode;
2854 int r;
2855
2856 if (kvm_apic_hw_enabled(apic) && !(reg & APIC_LVT_MASKED)) {
2857 vector = reg & APIC_VECTOR_MASK;
2858 mode = reg & APIC_MODE_MASK;
2859 trig_mode = reg & APIC_LVT_LEVEL_TRIGGER;
2860
2861 r = __apic_accept_irq(apic, mode, vector, 1, trig_mode, NULL);
2862 if (r && lvt_type == APIC_LVTPC &&
2863 guest_cpuid_is_intel_compatible(apic->vcpu))
2864 kvm_lapic_set_reg(apic, APIC_LVTPC, reg | APIC_LVT_MASKED);
2865 return r;
2866 }
2867 return 0;
2868}
2869
2870void kvm_apic_nmi_wd_deliver(struct kvm_vcpu *vcpu)
2871{
2872 struct kvm_lapic *apic = vcpu->arch.apic;
2873
2874 if (apic)
2875 kvm_apic_local_deliver(apic, APIC_LVT0);
2876}
2877
2878static const struct kvm_io_device_ops apic_mmio_ops = {
2879 .read = apic_mmio_read,
2880 .write = apic_mmio_write,
2881};
2882
2883static enum hrtimer_restart apic_timer_fn(struct hrtimer *data)
2884{
2885 struct kvm_timer *ktimer = container_of(data, struct kvm_timer, timer);
2886 struct kvm_lapic *apic = container_of(ktimer, struct kvm_lapic, lapic_timer);
2887
2888 apic_timer_expired(apic, true);
2889
2890 if (lapic_is_periodic(apic)) {
2891 advance_periodic_target_expiration(apic);
2892 hrtimer_add_expires_ns(&ktimer->timer, ktimer->period);
2893 return HRTIMER_RESTART;
2894 } else
2895 return HRTIMER_NORESTART;
2896}
2897
2898int kvm_create_lapic(struct kvm_vcpu *vcpu)
2899{
2900 struct kvm_lapic *apic;
2901
2902 ASSERT(vcpu != NULL);
2903
2904 if (!irqchip_in_kernel(vcpu->kvm)) {
2905 static_branch_inc(&kvm_has_noapic_vcpu);
2906 return 0;
2907 }
2908
2909 apic = kzalloc(sizeof(*apic), GFP_KERNEL_ACCOUNT);
2910 if (!apic)
2911 goto nomem;
2912
2913 vcpu->arch.apic = apic;
2914
2915 if (kvm_x86_ops.alloc_apic_backing_page)
2916 apic->regs = kvm_x86_call(alloc_apic_backing_page)(vcpu);
2917 else
2918 apic->regs = (void *)get_zeroed_page(GFP_KERNEL_ACCOUNT);
2919 if (!apic->regs) {
2920 printk(KERN_ERR "malloc apic regs error for vcpu %x\n",
2921 vcpu->vcpu_id);
2922 goto nomem_free_apic;
2923 }
2924 apic->vcpu = vcpu;
2925
2926 apic->nr_lvt_entries = kvm_apic_calc_nr_lvt_entries(vcpu);
2927
2928 hrtimer_init(&apic->lapic_timer.timer, CLOCK_MONOTONIC,
2929 HRTIMER_MODE_ABS_HARD);
2930 apic->lapic_timer.timer.function = apic_timer_fn;
2931 if (lapic_timer_advance)
2932 apic->lapic_timer.timer_advance_ns = LAPIC_TIMER_ADVANCE_NS_INIT;
2933
2934 /*
2935 * Stuff the APIC ENABLE bit in lieu of temporarily incrementing
2936 * apic_hw_disabled; the full RESET value is set by kvm_lapic_reset().
2937 */
2938 vcpu->arch.apic_base = MSR_IA32_APICBASE_ENABLE;
2939 static_branch_inc(&apic_sw_disabled.key); /* sw disabled at reset */
2940 kvm_iodevice_init(&apic->dev, &apic_mmio_ops);
2941
2942 /*
2943 * Defer evaluating inhibits until the vCPU is first run, as this vCPU
2944 * will not get notified of any changes until this vCPU is visible to
2945 * other vCPUs (marked online and added to the set of vCPUs).
2946 *
2947 * Opportunistically mark APICv active as VMX in particularly is highly
2948 * unlikely to have inhibits. Ignore the current per-VM APICv state so
2949 * that vCPU creation is guaranteed to run with a deterministic value,
2950 * the request will ensure the vCPU gets the correct state before VM-Entry.
2951 */
2952 if (enable_apicv) {
2953 apic->apicv_active = true;
2954 kvm_make_request(KVM_REQ_APICV_UPDATE, vcpu);
2955 }
2956
2957 return 0;
2958nomem_free_apic:
2959 kfree(apic);
2960 vcpu->arch.apic = NULL;
2961nomem:
2962 return -ENOMEM;
2963}
2964
2965int kvm_apic_has_interrupt(struct kvm_vcpu *vcpu)
2966{
2967 struct kvm_lapic *apic = vcpu->arch.apic;
2968 u32 ppr;
2969
2970 if (!kvm_apic_present(vcpu))
2971 return -1;
2972
2973 __apic_update_ppr(apic, &ppr);
2974 return apic_has_interrupt_for_ppr(apic, ppr);
2975}
2976EXPORT_SYMBOL_GPL(kvm_apic_has_interrupt);
2977
2978int kvm_apic_accept_pic_intr(struct kvm_vcpu *vcpu)
2979{
2980 u32 lvt0 = kvm_lapic_get_reg(vcpu->arch.apic, APIC_LVT0);
2981
2982 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
2983 return 1;
2984 if ((lvt0 & APIC_LVT_MASKED) == 0 &&
2985 GET_APIC_DELIVERY_MODE(lvt0) == APIC_MODE_EXTINT)
2986 return 1;
2987 return 0;
2988}
2989
2990void kvm_inject_apic_timer_irqs(struct kvm_vcpu *vcpu)
2991{
2992 struct kvm_lapic *apic = vcpu->arch.apic;
2993
2994 if (atomic_read(&apic->lapic_timer.pending) > 0) {
2995 kvm_apic_inject_pending_timer_irqs(apic);
2996 atomic_set(&apic->lapic_timer.pending, 0);
2997 }
2998}
2999
3000void kvm_apic_ack_interrupt(struct kvm_vcpu *vcpu, int vector)
3001{
3002 struct kvm_lapic *apic = vcpu->arch.apic;
3003 u32 ppr;
3004
3005 if (WARN_ON_ONCE(vector < 0 || !apic))
3006 return;
3007
3008 /*
3009 * We get here even with APIC virtualization enabled, if doing
3010 * nested virtualization and L1 runs with the "acknowledge interrupt
3011 * on exit" mode. Then we cannot inject the interrupt via RVI,
3012 * because the process would deliver it through the IDT.
3013 */
3014
3015 apic_clear_irr(vector, apic);
3016 if (kvm_hv_synic_auto_eoi_set(vcpu, vector)) {
3017 /*
3018 * For auto-EOI interrupts, there might be another pending
3019 * interrupt above PPR, so check whether to raise another
3020 * KVM_REQ_EVENT.
3021 */
3022 apic_update_ppr(apic);
3023 } else {
3024 /*
3025 * For normal interrupts, PPR has been raised and there cannot
3026 * be a higher-priority pending interrupt---except if there was
3027 * a concurrent interrupt injection, but that would have
3028 * triggered KVM_REQ_EVENT already.
3029 */
3030 apic_set_isr(vector, apic);
3031 __apic_update_ppr(apic, &ppr);
3032 }
3033
3034}
3035EXPORT_SYMBOL_GPL(kvm_apic_ack_interrupt);
3036
3037static int kvm_apic_state_fixup(struct kvm_vcpu *vcpu,
3038 struct kvm_lapic_state *s, bool set)
3039{
3040 if (apic_x2apic_mode(vcpu->arch.apic)) {
3041 u32 x2apic_id = kvm_x2apic_id(vcpu->arch.apic);
3042 u32 *id = (u32 *)(s->regs + APIC_ID);
3043 u32 *ldr = (u32 *)(s->regs + APIC_LDR);
3044 u64 icr;
3045
3046 if (vcpu->kvm->arch.x2apic_format) {
3047 if (*id != x2apic_id)
3048 return -EINVAL;
3049 } else {
3050 /*
3051 * Ignore the userspace value when setting APIC state.
3052 * KVM's model is that the x2APIC ID is readonly, e.g.
3053 * KVM only supports delivering interrupts to KVM's
3054 * version of the x2APIC ID. However, for backwards
3055 * compatibility, don't reject attempts to set a
3056 * mismatched ID for userspace that hasn't opted into
3057 * x2apic_format.
3058 */
3059 if (set)
3060 *id = x2apic_id;
3061 else
3062 *id = x2apic_id << 24;
3063 }
3064
3065 /*
3066 * In x2APIC mode, the LDR is fixed and based on the id. And
3067 * if the ICR is _not_ split, ICR is internally a single 64-bit
3068 * register, but needs to be split to ICR+ICR2 in userspace for
3069 * backwards compatibility.
3070 */
3071 if (set)
3072 *ldr = kvm_apic_calc_x2apic_ldr(x2apic_id);
3073
3074 if (!kvm_x86_ops.x2apic_icr_is_split) {
3075 if (set) {
3076 icr = __kvm_lapic_get_reg(s->regs, APIC_ICR) |
3077 (u64)__kvm_lapic_get_reg(s->regs, APIC_ICR2) << 32;
3078 __kvm_lapic_set_reg64(s->regs, APIC_ICR, icr);
3079 } else {
3080 icr = __kvm_lapic_get_reg64(s->regs, APIC_ICR);
3081 __kvm_lapic_set_reg(s->regs, APIC_ICR2, icr >> 32);
3082 }
3083 }
3084 }
3085
3086 return 0;
3087}
3088
3089int kvm_apic_get_state(struct kvm_vcpu *vcpu, struct kvm_lapic_state *s)
3090{
3091 memcpy(s->regs, vcpu->arch.apic->regs, sizeof(*s));
3092
3093 /*
3094 * Get calculated timer current count for remaining timer period (if
3095 * any) and store it in the returned register set.
3096 */
3097 __kvm_lapic_set_reg(s->regs, APIC_TMCCT,
3098 __apic_read(vcpu->arch.apic, APIC_TMCCT));
3099
3100 return kvm_apic_state_fixup(vcpu, s, false);
3101}
3102
3103int kvm_apic_set_state(struct kvm_vcpu *vcpu, struct kvm_lapic_state *s)
3104{
3105 struct kvm_lapic *apic = vcpu->arch.apic;
3106 int r;
3107
3108 kvm_x86_call(apicv_pre_state_restore)(vcpu);
3109
3110 /* set SPIV separately to get count of SW disabled APICs right */
3111 apic_set_spiv(apic, *((u32 *)(s->regs + APIC_SPIV)));
3112
3113 r = kvm_apic_state_fixup(vcpu, s, true);
3114 if (r) {
3115 kvm_recalculate_apic_map(vcpu->kvm);
3116 return r;
3117 }
3118 memcpy(vcpu->arch.apic->regs, s->regs, sizeof(*s));
3119
3120 atomic_set_release(&apic->vcpu->kvm->arch.apic_map_dirty, DIRTY);
3121 kvm_recalculate_apic_map(vcpu->kvm);
3122 kvm_apic_set_version(vcpu);
3123
3124 apic_update_ppr(apic);
3125 cancel_apic_timer(apic);
3126 apic->lapic_timer.expired_tscdeadline = 0;
3127 apic_update_lvtt(apic);
3128 apic_manage_nmi_watchdog(apic, kvm_lapic_get_reg(apic, APIC_LVT0));
3129 update_divide_count(apic);
3130 __start_apic_timer(apic, APIC_TMCCT);
3131 kvm_lapic_set_reg(apic, APIC_TMCCT, 0);
3132 kvm_apic_update_apicv(vcpu);
3133 if (apic->apicv_active) {
3134 kvm_x86_call(apicv_post_state_restore)(vcpu);
3135 kvm_x86_call(hwapic_irr_update)(vcpu, apic_find_highest_irr(apic));
3136 kvm_x86_call(hwapic_isr_update)(vcpu, apic_find_highest_isr(apic));
3137 }
3138 kvm_make_request(KVM_REQ_EVENT, vcpu);
3139 if (ioapic_in_kernel(vcpu->kvm))
3140 kvm_rtc_eoi_tracking_restore_one(vcpu);
3141
3142 vcpu->arch.apic_arb_prio = 0;
3143
3144 return 0;
3145}
3146
3147void __kvm_migrate_apic_timer(struct kvm_vcpu *vcpu)
3148{
3149 struct hrtimer *timer;
3150
3151 if (!lapic_in_kernel(vcpu) ||
3152 kvm_can_post_timer_interrupt(vcpu))
3153 return;
3154
3155 timer = &vcpu->arch.apic->lapic_timer.timer;
3156 if (hrtimer_cancel(timer))
3157 hrtimer_start_expires(timer, HRTIMER_MODE_ABS_HARD);
3158}
3159
3160/*
3161 * apic_sync_pv_eoi_from_guest - called on vmexit or cancel interrupt
3162 *
3163 * Detect whether guest triggered PV EOI since the
3164 * last entry. If yes, set EOI on guests's behalf.
3165 * Clear PV EOI in guest memory in any case.
3166 */
3167static void apic_sync_pv_eoi_from_guest(struct kvm_vcpu *vcpu,
3168 struct kvm_lapic *apic)
3169{
3170 int vector;
3171 /*
3172 * PV EOI state is derived from KVM_APIC_PV_EOI_PENDING in host
3173 * and KVM_PV_EOI_ENABLED in guest memory as follows:
3174 *
3175 * KVM_APIC_PV_EOI_PENDING is unset:
3176 * -> host disabled PV EOI.
3177 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is set:
3178 * -> host enabled PV EOI, guest did not execute EOI yet.
3179 * KVM_APIC_PV_EOI_PENDING is set, KVM_PV_EOI_ENABLED is unset:
3180 * -> host enabled PV EOI, guest executed EOI.
3181 */
3182 BUG_ON(!pv_eoi_enabled(vcpu));
3183
3184 if (pv_eoi_test_and_clr_pending(vcpu))
3185 return;
3186 vector = apic_set_eoi(apic);
3187 trace_kvm_pv_eoi(apic, vector);
3188}
3189
3190void kvm_lapic_sync_from_vapic(struct kvm_vcpu *vcpu)
3191{
3192 u32 data;
3193
3194 if (test_bit(KVM_APIC_PV_EOI_PENDING, &vcpu->arch.apic_attention))
3195 apic_sync_pv_eoi_from_guest(vcpu, vcpu->arch.apic);
3196
3197 if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
3198 return;
3199
3200 if (kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
3201 sizeof(u32)))
3202 return;
3203
3204 apic_set_tpr(vcpu->arch.apic, data & 0xff);
3205}
3206
3207/*
3208 * apic_sync_pv_eoi_to_guest - called before vmentry
3209 *
3210 * Detect whether it's safe to enable PV EOI and
3211 * if yes do so.
3212 */
3213static void apic_sync_pv_eoi_to_guest(struct kvm_vcpu *vcpu,
3214 struct kvm_lapic *apic)
3215{
3216 if (!pv_eoi_enabled(vcpu) ||
3217 /* IRR set or many bits in ISR: could be nested. */
3218 apic->irr_pending ||
3219 /* Cache not set: could be safe but we don't bother. */
3220 apic->highest_isr_cache == -1 ||
3221 /* Need EOI to update ioapic. */
3222 kvm_ioapic_handles_vector(apic, apic->highest_isr_cache)) {
3223 /*
3224 * PV EOI was disabled by apic_sync_pv_eoi_from_guest
3225 * so we need not do anything here.
3226 */
3227 return;
3228 }
3229
3230 pv_eoi_set_pending(apic->vcpu);
3231}
3232
3233void kvm_lapic_sync_to_vapic(struct kvm_vcpu *vcpu)
3234{
3235 u32 data, tpr;
3236 int max_irr, max_isr;
3237 struct kvm_lapic *apic = vcpu->arch.apic;
3238
3239 apic_sync_pv_eoi_to_guest(vcpu, apic);
3240
3241 if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
3242 return;
3243
3244 tpr = kvm_lapic_get_reg(apic, APIC_TASKPRI) & 0xff;
3245 max_irr = apic_find_highest_irr(apic);
3246 if (max_irr < 0)
3247 max_irr = 0;
3248 max_isr = apic_find_highest_isr(apic);
3249 if (max_isr < 0)
3250 max_isr = 0;
3251 data = (tpr & 0xff) | ((max_isr & 0xf0) << 8) | (max_irr << 24);
3252
3253 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apic->vapic_cache, &data,
3254 sizeof(u32));
3255}
3256
3257int kvm_lapic_set_vapic_addr(struct kvm_vcpu *vcpu, gpa_t vapic_addr)
3258{
3259 if (vapic_addr) {
3260 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
3261 &vcpu->arch.apic->vapic_cache,
3262 vapic_addr, sizeof(u32)))
3263 return -EINVAL;
3264 __set_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
3265 } else {
3266 __clear_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
3267 }
3268
3269 vcpu->arch.apic->vapic_addr = vapic_addr;
3270 return 0;
3271}
3272
3273static int kvm_lapic_msr_read(struct kvm_lapic *apic, u32 reg, u64 *data)
3274{
3275 u32 low;
3276
3277 if (reg == APIC_ICR) {
3278 *data = kvm_x2apic_icr_read(apic);
3279 return 0;
3280 }
3281
3282 if (kvm_lapic_reg_read(apic, reg, 4, &low))
3283 return 1;
3284
3285 *data = low;
3286
3287 return 0;
3288}
3289
3290static int kvm_lapic_msr_write(struct kvm_lapic *apic, u32 reg, u64 data)
3291{
3292 /*
3293 * ICR is a 64-bit register in x2APIC mode (and Hyper-V PV vAPIC) and
3294 * can be written as such, all other registers remain accessible only
3295 * through 32-bit reads/writes.
3296 */
3297 if (reg == APIC_ICR)
3298 return kvm_x2apic_icr_write(apic, data);
3299
3300 /* Bits 63:32 are reserved in all other registers. */
3301 if (data >> 32)
3302 return 1;
3303
3304 return kvm_lapic_reg_write(apic, reg, (u32)data);
3305}
3306
3307int kvm_x2apic_msr_write(struct kvm_vcpu *vcpu, u32 msr, u64 data)
3308{
3309 struct kvm_lapic *apic = vcpu->arch.apic;
3310 u32 reg = (msr - APIC_BASE_MSR) << 4;
3311
3312 if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(apic))
3313 return 1;
3314
3315 return kvm_lapic_msr_write(apic, reg, data);
3316}
3317
3318int kvm_x2apic_msr_read(struct kvm_vcpu *vcpu, u32 msr, u64 *data)
3319{
3320 struct kvm_lapic *apic = vcpu->arch.apic;
3321 u32 reg = (msr - APIC_BASE_MSR) << 4;
3322
3323 if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(apic))
3324 return 1;
3325
3326 return kvm_lapic_msr_read(apic, reg, data);
3327}
3328
3329int kvm_hv_vapic_msr_write(struct kvm_vcpu *vcpu, u32 reg, u64 data)
3330{
3331 if (!lapic_in_kernel(vcpu))
3332 return 1;
3333
3334 return kvm_lapic_msr_write(vcpu->arch.apic, reg, data);
3335}
3336
3337int kvm_hv_vapic_msr_read(struct kvm_vcpu *vcpu, u32 reg, u64 *data)
3338{
3339 if (!lapic_in_kernel(vcpu))
3340 return 1;
3341
3342 return kvm_lapic_msr_read(vcpu->arch.apic, reg, data);
3343}
3344
3345int kvm_lapic_set_pv_eoi(struct kvm_vcpu *vcpu, u64 data, unsigned long len)
3346{
3347 u64 addr = data & ~KVM_MSR_ENABLED;
3348 struct gfn_to_hva_cache *ghc = &vcpu->arch.pv_eoi.data;
3349 unsigned long new_len;
3350 int ret;
3351
3352 if (!IS_ALIGNED(addr, 4))
3353 return 1;
3354
3355 if (data & KVM_MSR_ENABLED) {
3356 if (addr == ghc->gpa && len <= ghc->len)
3357 new_len = ghc->len;
3358 else
3359 new_len = len;
3360
3361 ret = kvm_gfn_to_hva_cache_init(vcpu->kvm, ghc, addr, new_len);
3362 if (ret)
3363 return ret;
3364 }
3365
3366 vcpu->arch.pv_eoi.msr_val = data;
3367
3368 return 0;
3369}
3370
3371int kvm_apic_accept_events(struct kvm_vcpu *vcpu)
3372{
3373 struct kvm_lapic *apic = vcpu->arch.apic;
3374 u8 sipi_vector;
3375 int r;
3376
3377 if (!kvm_apic_has_pending_init_or_sipi(vcpu))
3378 return 0;
3379
3380 if (is_guest_mode(vcpu)) {
3381 r = kvm_check_nested_events(vcpu);
3382 if (r < 0)
3383 return r == -EBUSY ? 0 : r;
3384 /*
3385 * Continue processing INIT/SIPI even if a nested VM-Exit
3386 * occurred, e.g. pending SIPIs should be dropped if INIT+SIPI
3387 * are blocked as a result of transitioning to VMX root mode.
3388 */
3389 }
3390
3391 /*
3392 * INITs are blocked while CPU is in specific states (SMM, VMX root
3393 * mode, SVM with GIF=0), while SIPIs are dropped if the CPU isn't in
3394 * wait-for-SIPI (WFS).
3395 */
3396 if (!kvm_apic_init_sipi_allowed(vcpu)) {
3397 WARN_ON_ONCE(vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED);
3398 clear_bit(KVM_APIC_SIPI, &apic->pending_events);
3399 return 0;
3400 }
3401
3402 if (test_and_clear_bit(KVM_APIC_INIT, &apic->pending_events)) {
3403 kvm_vcpu_reset(vcpu, true);
3404 if (kvm_vcpu_is_bsp(apic->vcpu))
3405 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
3406 else
3407 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
3408 }
3409 if (test_and_clear_bit(KVM_APIC_SIPI, &apic->pending_events)) {
3410 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
3411 /* evaluate pending_events before reading the vector */
3412 smp_rmb();
3413 sipi_vector = apic->sipi_vector;
3414 kvm_x86_call(vcpu_deliver_sipi_vector)(vcpu,
3415 sipi_vector);
3416 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
3417 }
3418 }
3419 return 0;
3420}
3421
3422void kvm_lapic_exit(void)
3423{
3424 static_key_deferred_flush(&apic_hw_disabled);
3425 WARN_ON(static_branch_unlikely(&apic_hw_disabled.key));
3426 static_key_deferred_flush(&apic_sw_disabled);
3427 WARN_ON(static_branch_unlikely(&apic_sw_disabled.key));
3428}
1
2/*
3 * Local APIC virtualization
4 *
5 * Copyright (C) 2006 Qumranet, Inc.
6 * Copyright (C) 2007 Novell
7 * Copyright (C) 2007 Intel
8 * Copyright 2009 Red Hat, Inc. and/or its affiliates.
9 *
10 * Authors:
11 * Dor Laor <dor.laor@qumranet.com>
12 * Gregory Haskins <ghaskins@novell.com>
13 * Yaozu (Eddie) Dong <eddie.dong@intel.com>
14 *
15 * Based on Xen 3.1 code, Copyright (c) 2004, Intel Corporation.
16 *
17 * This work is licensed under the terms of the GNU GPL, version 2. See
18 * the COPYING file in the top-level directory.
19 */
20
21#include <linux/kvm_host.h>
22#include <linux/kvm.h>
23#include <linux/mm.h>
24#include <linux/highmem.h>
25#include <linux/smp.h>
26#include <linux/hrtimer.h>
27#include <linux/io.h>
28#include <linux/module.h>
29#include <linux/math64.h>
30#include <linux/slab.h>
31#include <asm/processor.h>
32#include <asm/msr.h>
33#include <asm/page.h>
34#include <asm/current.h>
35#include <asm/apicdef.h>
36#include <linux/atomic.h>
37#include "kvm_cache_regs.h"
38#include "irq.h"
39#include "trace.h"
40#include "x86.h"
41#include "cpuid.h"
42
43#ifndef CONFIG_X86_64
44#define mod_64(x, y) ((x) - (y) * div64_u64(x, y))
45#else
46#define mod_64(x, y) ((x) % (y))
47#endif
48
49#define PRId64 "d"
50#define PRIx64 "llx"
51#define PRIu64 "u"
52#define PRIo64 "o"
53
54#define APIC_BUS_CYCLE_NS 1
55
56/* #define apic_debug(fmt,arg...) printk(KERN_WARNING fmt,##arg) */
57#define apic_debug(fmt, arg...)
58
59#define APIC_LVT_NUM 6
60/* 14 is the version for Xeon and Pentium 8.4.8*/
61#define APIC_VERSION (0x14UL | ((APIC_LVT_NUM - 1) << 16))
62#define LAPIC_MMIO_LENGTH (1 << 12)
63/* followed define is not in apicdef.h */
64#define APIC_SHORT_MASK 0xc0000
65#define APIC_DEST_NOSHORT 0x0
66#define APIC_DEST_MASK 0x800
67#define MAX_APIC_VECTOR 256
68
69#define VEC_POS(v) ((v) & (32 - 1))
70#define REG_POS(v) (((v) >> 5) << 4)
71
72static unsigned int min_timer_period_us = 500;
73module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
74
75static inline u32 apic_get_reg(struct kvm_lapic *apic, int reg_off)
76{
77 return *((u32 *) (apic->regs + reg_off));
78}
79
80static inline void apic_set_reg(struct kvm_lapic *apic, int reg_off, u32 val)
81{
82 *((u32 *) (apic->regs + reg_off)) = val;
83}
84
85static inline int apic_test_and_set_vector(int vec, void *bitmap)
86{
87 return test_and_set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
88}
89
90static inline int apic_test_and_clear_vector(int vec, void *bitmap)
91{
92 return test_and_clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
93}
94
95static inline int apic_test_vector(int vec, void *bitmap)
96{
97 return test_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
98}
99
100static inline void apic_set_vector(int vec, void *bitmap)
101{
102 set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
103}
104
105static inline void apic_clear_vector(int vec, void *bitmap)
106{
107 clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
108}
109
110static inline int apic_hw_enabled(struct kvm_lapic *apic)
111{
112 return (apic)->vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE;
113}
114
115static inline int apic_sw_enabled(struct kvm_lapic *apic)
116{
117 return apic_get_reg(apic, APIC_SPIV) & APIC_SPIV_APIC_ENABLED;
118}
119
120static inline int apic_enabled(struct kvm_lapic *apic)
121{
122 return apic_sw_enabled(apic) && apic_hw_enabled(apic);
123}
124
125#define LVT_MASK \
126 (APIC_LVT_MASKED | APIC_SEND_PENDING | APIC_VECTOR_MASK)
127
128#define LINT_MASK \
129 (LVT_MASK | APIC_MODE_MASK | APIC_INPUT_POLARITY | \
130 APIC_LVT_REMOTE_IRR | APIC_LVT_LEVEL_TRIGGER)
131
132static inline int kvm_apic_id(struct kvm_lapic *apic)
133{
134 return (apic_get_reg(apic, APIC_ID) >> 24) & 0xff;
135}
136
137static inline int apic_lvt_enabled(struct kvm_lapic *apic, int lvt_type)
138{
139 return !(apic_get_reg(apic, lvt_type) & APIC_LVT_MASKED);
140}
141
142static inline int apic_lvt_vector(struct kvm_lapic *apic, int lvt_type)
143{
144 return apic_get_reg(apic, lvt_type) & APIC_VECTOR_MASK;
145}
146
147static inline int apic_lvtt_oneshot(struct kvm_lapic *apic)
148{
149 return ((apic_get_reg(apic, APIC_LVTT) &
150 apic->lapic_timer.timer_mode_mask) == APIC_LVT_TIMER_ONESHOT);
151}
152
153static inline int apic_lvtt_period(struct kvm_lapic *apic)
154{
155 return ((apic_get_reg(apic, APIC_LVTT) &
156 apic->lapic_timer.timer_mode_mask) == APIC_LVT_TIMER_PERIODIC);
157}
158
159static inline int apic_lvtt_tscdeadline(struct kvm_lapic *apic)
160{
161 return ((apic_get_reg(apic, APIC_LVTT) &
162 apic->lapic_timer.timer_mode_mask) ==
163 APIC_LVT_TIMER_TSCDEADLINE);
164}
165
166static inline int apic_lvt_nmi_mode(u32 lvt_val)
167{
168 return (lvt_val & (APIC_MODE_MASK | APIC_LVT_MASKED)) == APIC_DM_NMI;
169}
170
171void kvm_apic_set_version(struct kvm_vcpu *vcpu)
172{
173 struct kvm_lapic *apic = vcpu->arch.apic;
174 struct kvm_cpuid_entry2 *feat;
175 u32 v = APIC_VERSION;
176
177 if (!irqchip_in_kernel(vcpu->kvm))
178 return;
179
180 feat = kvm_find_cpuid_entry(apic->vcpu, 0x1, 0);
181 if (feat && (feat->ecx & (1 << (X86_FEATURE_X2APIC & 31))))
182 v |= APIC_LVR_DIRECTED_EOI;
183 apic_set_reg(apic, APIC_LVR, v);
184}
185
186static inline int apic_x2apic_mode(struct kvm_lapic *apic)
187{
188 return apic->vcpu->arch.apic_base & X2APIC_ENABLE;
189}
190
191static unsigned int apic_lvt_mask[APIC_LVT_NUM] = {
192 LVT_MASK , /* part LVTT mask, timer mode mask added at runtime */
193 LVT_MASK | APIC_MODE_MASK, /* LVTTHMR */
194 LVT_MASK | APIC_MODE_MASK, /* LVTPC */
195 LINT_MASK, LINT_MASK, /* LVT0-1 */
196 LVT_MASK /* LVTERR */
197};
198
199static int find_highest_vector(void *bitmap)
200{
201 u32 *word = bitmap;
202 int word_offset = MAX_APIC_VECTOR >> 5;
203
204 while ((word_offset != 0) && (word[(--word_offset) << 2] == 0))
205 continue;
206
207 if (likely(!word_offset && !word[0]))
208 return -1;
209 else
210 return fls(word[word_offset << 2]) - 1 + (word_offset << 5);
211}
212
213static inline int apic_test_and_set_irr(int vec, struct kvm_lapic *apic)
214{
215 apic->irr_pending = true;
216 return apic_test_and_set_vector(vec, apic->regs + APIC_IRR);
217}
218
219static inline int apic_search_irr(struct kvm_lapic *apic)
220{
221 return find_highest_vector(apic->regs + APIC_IRR);
222}
223
224static inline int apic_find_highest_irr(struct kvm_lapic *apic)
225{
226 int result;
227
228 if (!apic->irr_pending)
229 return -1;
230
231 result = apic_search_irr(apic);
232 ASSERT(result == -1 || result >= 16);
233
234 return result;
235}
236
237static inline void apic_clear_irr(int vec, struct kvm_lapic *apic)
238{
239 apic->irr_pending = false;
240 apic_clear_vector(vec, apic->regs + APIC_IRR);
241 if (apic_search_irr(apic) != -1)
242 apic->irr_pending = true;
243}
244
245int kvm_lapic_find_highest_irr(struct kvm_vcpu *vcpu)
246{
247 struct kvm_lapic *apic = vcpu->arch.apic;
248 int highest_irr;
249
250 /* This may race with setting of irr in __apic_accept_irq() and
251 * value returned may be wrong, but kvm_vcpu_kick() in __apic_accept_irq
252 * will cause vmexit immediately and the value will be recalculated
253 * on the next vmentry.
254 */
255 if (!apic)
256 return 0;
257 highest_irr = apic_find_highest_irr(apic);
258
259 return highest_irr;
260}
261
262static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
263 int vector, int level, int trig_mode);
264
265int kvm_apic_set_irq(struct kvm_vcpu *vcpu, struct kvm_lapic_irq *irq)
266{
267 struct kvm_lapic *apic = vcpu->arch.apic;
268
269 return __apic_accept_irq(apic, irq->delivery_mode, irq->vector,
270 irq->level, irq->trig_mode);
271}
272
273static inline int apic_find_highest_isr(struct kvm_lapic *apic)
274{
275 int result;
276
277 result = find_highest_vector(apic->regs + APIC_ISR);
278 ASSERT(result == -1 || result >= 16);
279
280 return result;
281}
282
283static void apic_update_ppr(struct kvm_lapic *apic)
284{
285 u32 tpr, isrv, ppr, old_ppr;
286 int isr;
287
288 old_ppr = apic_get_reg(apic, APIC_PROCPRI);
289 tpr = apic_get_reg(apic, APIC_TASKPRI);
290 isr = apic_find_highest_isr(apic);
291 isrv = (isr != -1) ? isr : 0;
292
293 if ((tpr & 0xf0) >= (isrv & 0xf0))
294 ppr = tpr & 0xff;
295 else
296 ppr = isrv & 0xf0;
297
298 apic_debug("vlapic %p, ppr 0x%x, isr 0x%x, isrv 0x%x",
299 apic, ppr, isr, isrv);
300
301 if (old_ppr != ppr) {
302 apic_set_reg(apic, APIC_PROCPRI, ppr);
303 if (ppr < old_ppr)
304 kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
305 }
306}
307
308static void apic_set_tpr(struct kvm_lapic *apic, u32 tpr)
309{
310 apic_set_reg(apic, APIC_TASKPRI, tpr);
311 apic_update_ppr(apic);
312}
313
314int kvm_apic_match_physical_addr(struct kvm_lapic *apic, u16 dest)
315{
316 return dest == 0xff || kvm_apic_id(apic) == dest;
317}
318
319int kvm_apic_match_logical_addr(struct kvm_lapic *apic, u8 mda)
320{
321 int result = 0;
322 u32 logical_id;
323
324 if (apic_x2apic_mode(apic)) {
325 logical_id = apic_get_reg(apic, APIC_LDR);
326 return logical_id & mda;
327 }
328
329 logical_id = GET_APIC_LOGICAL_ID(apic_get_reg(apic, APIC_LDR));
330
331 switch (apic_get_reg(apic, APIC_DFR)) {
332 case APIC_DFR_FLAT:
333 if (logical_id & mda)
334 result = 1;
335 break;
336 case APIC_DFR_CLUSTER:
337 if (((logical_id >> 4) == (mda >> 0x4))
338 && (logical_id & mda & 0xf))
339 result = 1;
340 break;
341 default:
342 apic_debug("Bad DFR vcpu %d: %08x\n",
343 apic->vcpu->vcpu_id, apic_get_reg(apic, APIC_DFR));
344 break;
345 }
346
347 return result;
348}
349
350int kvm_apic_match_dest(struct kvm_vcpu *vcpu, struct kvm_lapic *source,
351 int short_hand, int dest, int dest_mode)
352{
353 int result = 0;
354 struct kvm_lapic *target = vcpu->arch.apic;
355
356 apic_debug("target %p, source %p, dest 0x%x, "
357 "dest_mode 0x%x, short_hand 0x%x\n",
358 target, source, dest, dest_mode, short_hand);
359
360 ASSERT(target);
361 switch (short_hand) {
362 case APIC_DEST_NOSHORT:
363 if (dest_mode == 0)
364 /* Physical mode. */
365 result = kvm_apic_match_physical_addr(target, dest);
366 else
367 /* Logical mode. */
368 result = kvm_apic_match_logical_addr(target, dest);
369 break;
370 case APIC_DEST_SELF:
371 result = (target == source);
372 break;
373 case APIC_DEST_ALLINC:
374 result = 1;
375 break;
376 case APIC_DEST_ALLBUT:
377 result = (target != source);
378 break;
379 default:
380 apic_debug("kvm: apic: Bad dest shorthand value %x\n",
381 short_hand);
382 break;
383 }
384
385 return result;
386}
387
388/*
389 * Add a pending IRQ into lapic.
390 * Return 1 if successfully added and 0 if discarded.
391 */
392static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
393 int vector, int level, int trig_mode)
394{
395 int result = 0;
396 struct kvm_vcpu *vcpu = apic->vcpu;
397
398 switch (delivery_mode) {
399 case APIC_DM_LOWEST:
400 vcpu->arch.apic_arb_prio++;
401 case APIC_DM_FIXED:
402 /* FIXME add logic for vcpu on reset */
403 if (unlikely(!apic_enabled(apic)))
404 break;
405
406 if (trig_mode) {
407 apic_debug("level trig mode for vector %d", vector);
408 apic_set_vector(vector, apic->regs + APIC_TMR);
409 } else
410 apic_clear_vector(vector, apic->regs + APIC_TMR);
411
412 result = !apic_test_and_set_irr(vector, apic);
413 trace_kvm_apic_accept_irq(vcpu->vcpu_id, delivery_mode,
414 trig_mode, vector, !result);
415 if (!result) {
416 if (trig_mode)
417 apic_debug("level trig mode repeatedly for "
418 "vector %d", vector);
419 break;
420 }
421
422 kvm_make_request(KVM_REQ_EVENT, vcpu);
423 kvm_vcpu_kick(vcpu);
424 break;
425
426 case APIC_DM_REMRD:
427 apic_debug("Ignoring delivery mode 3\n");
428 break;
429
430 case APIC_DM_SMI:
431 apic_debug("Ignoring guest SMI\n");
432 break;
433
434 case APIC_DM_NMI:
435 result = 1;
436 kvm_inject_nmi(vcpu);
437 kvm_vcpu_kick(vcpu);
438 break;
439
440 case APIC_DM_INIT:
441 if (!trig_mode || level) {
442 result = 1;
443 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
444 kvm_make_request(KVM_REQ_EVENT, vcpu);
445 kvm_vcpu_kick(vcpu);
446 } else {
447 apic_debug("Ignoring de-assert INIT to vcpu %d\n",
448 vcpu->vcpu_id);
449 }
450 break;
451
452 case APIC_DM_STARTUP:
453 apic_debug("SIPI to vcpu %d vector 0x%02x\n",
454 vcpu->vcpu_id, vector);
455 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
456 result = 1;
457 vcpu->arch.sipi_vector = vector;
458 vcpu->arch.mp_state = KVM_MP_STATE_SIPI_RECEIVED;
459 kvm_make_request(KVM_REQ_EVENT, vcpu);
460 kvm_vcpu_kick(vcpu);
461 }
462 break;
463
464 case APIC_DM_EXTINT:
465 /*
466 * Should only be called by kvm_apic_local_deliver() with LVT0,
467 * before NMI watchdog was enabled. Already handled by
468 * kvm_apic_accept_pic_intr().
469 */
470 break;
471
472 default:
473 printk(KERN_ERR "TODO: unsupported delivery mode %x\n",
474 delivery_mode);
475 break;
476 }
477 return result;
478}
479
480int kvm_apic_compare_prio(struct kvm_vcpu *vcpu1, struct kvm_vcpu *vcpu2)
481{
482 return vcpu1->arch.apic_arb_prio - vcpu2->arch.apic_arb_prio;
483}
484
485static void apic_set_eoi(struct kvm_lapic *apic)
486{
487 int vector = apic_find_highest_isr(apic);
488 /*
489 * Not every write EOI will has corresponding ISR,
490 * one example is when Kernel check timer on setup_IO_APIC
491 */
492 if (vector == -1)
493 return;
494
495 apic_clear_vector(vector, apic->regs + APIC_ISR);
496 apic_update_ppr(apic);
497
498 if (!(apic_get_reg(apic, APIC_SPIV) & APIC_SPIV_DIRECTED_EOI) &&
499 kvm_ioapic_handles_vector(apic->vcpu->kvm, vector)) {
500 int trigger_mode;
501 if (apic_test_vector(vector, apic->regs + APIC_TMR))
502 trigger_mode = IOAPIC_LEVEL_TRIG;
503 else
504 trigger_mode = IOAPIC_EDGE_TRIG;
505 kvm_ioapic_update_eoi(apic->vcpu->kvm, vector, trigger_mode);
506 }
507 kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
508}
509
510static void apic_send_ipi(struct kvm_lapic *apic)
511{
512 u32 icr_low = apic_get_reg(apic, APIC_ICR);
513 u32 icr_high = apic_get_reg(apic, APIC_ICR2);
514 struct kvm_lapic_irq irq;
515
516 irq.vector = icr_low & APIC_VECTOR_MASK;
517 irq.delivery_mode = icr_low & APIC_MODE_MASK;
518 irq.dest_mode = icr_low & APIC_DEST_MASK;
519 irq.level = icr_low & APIC_INT_ASSERT;
520 irq.trig_mode = icr_low & APIC_INT_LEVELTRIG;
521 irq.shorthand = icr_low & APIC_SHORT_MASK;
522 if (apic_x2apic_mode(apic))
523 irq.dest_id = icr_high;
524 else
525 irq.dest_id = GET_APIC_DEST_FIELD(icr_high);
526
527 trace_kvm_apic_ipi(icr_low, irq.dest_id);
528
529 apic_debug("icr_high 0x%x, icr_low 0x%x, "
530 "short_hand 0x%x, dest 0x%x, trig_mode 0x%x, level 0x%x, "
531 "dest_mode 0x%x, delivery_mode 0x%x, vector 0x%x\n",
532 icr_high, icr_low, irq.shorthand, irq.dest_id,
533 irq.trig_mode, irq.level, irq.dest_mode, irq.delivery_mode,
534 irq.vector);
535
536 kvm_irq_delivery_to_apic(apic->vcpu->kvm, apic, &irq);
537}
538
539static u32 apic_get_tmcct(struct kvm_lapic *apic)
540{
541 ktime_t remaining;
542 s64 ns;
543 u32 tmcct;
544
545 ASSERT(apic != NULL);
546
547 /* if initial count is 0, current count should also be 0 */
548 if (apic_get_reg(apic, APIC_TMICT) == 0)
549 return 0;
550
551 remaining = hrtimer_get_remaining(&apic->lapic_timer.timer);
552 if (ktime_to_ns(remaining) < 0)
553 remaining = ktime_set(0, 0);
554
555 ns = mod_64(ktime_to_ns(remaining), apic->lapic_timer.period);
556 tmcct = div64_u64(ns,
557 (APIC_BUS_CYCLE_NS * apic->divide_count));
558
559 return tmcct;
560}
561
562static void __report_tpr_access(struct kvm_lapic *apic, bool write)
563{
564 struct kvm_vcpu *vcpu = apic->vcpu;
565 struct kvm_run *run = vcpu->run;
566
567 kvm_make_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu);
568 run->tpr_access.rip = kvm_rip_read(vcpu);
569 run->tpr_access.is_write = write;
570}
571
572static inline void report_tpr_access(struct kvm_lapic *apic, bool write)
573{
574 if (apic->vcpu->arch.tpr_access_reporting)
575 __report_tpr_access(apic, write);
576}
577
578static u32 __apic_read(struct kvm_lapic *apic, unsigned int offset)
579{
580 u32 val = 0;
581
582 if (offset >= LAPIC_MMIO_LENGTH)
583 return 0;
584
585 switch (offset) {
586 case APIC_ID:
587 if (apic_x2apic_mode(apic))
588 val = kvm_apic_id(apic);
589 else
590 val = kvm_apic_id(apic) << 24;
591 break;
592 case APIC_ARBPRI:
593 apic_debug("Access APIC ARBPRI register which is for P6\n");
594 break;
595
596 case APIC_TMCCT: /* Timer CCR */
597 if (apic_lvtt_tscdeadline(apic))
598 return 0;
599
600 val = apic_get_tmcct(apic);
601 break;
602
603 case APIC_TASKPRI:
604 report_tpr_access(apic, false);
605 /* fall thru */
606 default:
607 apic_update_ppr(apic);
608 val = apic_get_reg(apic, offset);
609 break;
610 }
611
612 return val;
613}
614
615static inline struct kvm_lapic *to_lapic(struct kvm_io_device *dev)
616{
617 return container_of(dev, struct kvm_lapic, dev);
618}
619
620static int apic_reg_read(struct kvm_lapic *apic, u32 offset, int len,
621 void *data)
622{
623 unsigned char alignment = offset & 0xf;
624 u32 result;
625 /* this bitmask has a bit cleared for each reserver register */
626 static const u64 rmask = 0x43ff01ffffffe70cULL;
627
628 if ((alignment + len) > 4) {
629 apic_debug("KVM_APIC_READ: alignment error %x %d\n",
630 offset, len);
631 return 1;
632 }
633
634 if (offset > 0x3f0 || !(rmask & (1ULL << (offset >> 4)))) {
635 apic_debug("KVM_APIC_READ: read reserved register %x\n",
636 offset);
637 return 1;
638 }
639
640 result = __apic_read(apic, offset & ~0xf);
641
642 trace_kvm_apic_read(offset, result);
643
644 switch (len) {
645 case 1:
646 case 2:
647 case 4:
648 memcpy(data, (char *)&result + alignment, len);
649 break;
650 default:
651 printk(KERN_ERR "Local APIC read with len = %x, "
652 "should be 1,2, or 4 instead\n", len);
653 break;
654 }
655 return 0;
656}
657
658static int apic_mmio_in_range(struct kvm_lapic *apic, gpa_t addr)
659{
660 return apic_hw_enabled(apic) &&
661 addr >= apic->base_address &&
662 addr < apic->base_address + LAPIC_MMIO_LENGTH;
663}
664
665static int apic_mmio_read(struct kvm_io_device *this,
666 gpa_t address, int len, void *data)
667{
668 struct kvm_lapic *apic = to_lapic(this);
669 u32 offset = address - apic->base_address;
670
671 if (!apic_mmio_in_range(apic, address))
672 return -EOPNOTSUPP;
673
674 apic_reg_read(apic, offset, len, data);
675
676 return 0;
677}
678
679static void update_divide_count(struct kvm_lapic *apic)
680{
681 u32 tmp1, tmp2, tdcr;
682
683 tdcr = apic_get_reg(apic, APIC_TDCR);
684 tmp1 = tdcr & 0xf;
685 tmp2 = ((tmp1 & 0x3) | ((tmp1 & 0x8) >> 1)) + 1;
686 apic->divide_count = 0x1 << (tmp2 & 0x7);
687
688 apic_debug("timer divide count is 0x%x\n",
689 apic->divide_count);
690}
691
692static void start_apic_timer(struct kvm_lapic *apic)
693{
694 ktime_t now;
695 atomic_set(&apic->lapic_timer.pending, 0);
696
697 if (apic_lvtt_period(apic) || apic_lvtt_oneshot(apic)) {
698 /* lapic timer in oneshot or peroidic mode */
699 now = apic->lapic_timer.timer.base->get_time();
700 apic->lapic_timer.period = (u64)apic_get_reg(apic, APIC_TMICT)
701 * APIC_BUS_CYCLE_NS * apic->divide_count;
702
703 if (!apic->lapic_timer.period)
704 return;
705 /*
706 * Do not allow the guest to program periodic timers with small
707 * interval, since the hrtimers are not throttled by the host
708 * scheduler.
709 */
710 if (apic_lvtt_period(apic)) {
711 s64 min_period = min_timer_period_us * 1000LL;
712
713 if (apic->lapic_timer.period < min_period) {
714 pr_info_ratelimited(
715 "kvm: vcpu %i: requested %lld ns "
716 "lapic timer period limited to %lld ns\n",
717 apic->vcpu->vcpu_id,
718 apic->lapic_timer.period, min_period);
719 apic->lapic_timer.period = min_period;
720 }
721 }
722
723 hrtimer_start(&apic->lapic_timer.timer,
724 ktime_add_ns(now, apic->lapic_timer.period),
725 HRTIMER_MODE_ABS);
726
727 apic_debug("%s: bus cycle is %" PRId64 "ns, now 0x%016"
728 PRIx64 ", "
729 "timer initial count 0x%x, period %lldns, "
730 "expire @ 0x%016" PRIx64 ".\n", __func__,
731 APIC_BUS_CYCLE_NS, ktime_to_ns(now),
732 apic_get_reg(apic, APIC_TMICT),
733 apic->lapic_timer.period,
734 ktime_to_ns(ktime_add_ns(now,
735 apic->lapic_timer.period)));
736 } else if (apic_lvtt_tscdeadline(apic)) {
737 /* lapic timer in tsc deadline mode */
738 u64 guest_tsc, tscdeadline = apic->lapic_timer.tscdeadline;
739 u64 ns = 0;
740 struct kvm_vcpu *vcpu = apic->vcpu;
741 unsigned long this_tsc_khz = vcpu->arch.virtual_tsc_khz;
742 unsigned long flags;
743
744 if (unlikely(!tscdeadline || !this_tsc_khz))
745 return;
746
747 local_irq_save(flags);
748
749 now = apic->lapic_timer.timer.base->get_time();
750 guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu);
751 if (likely(tscdeadline > guest_tsc)) {
752 ns = (tscdeadline - guest_tsc) * 1000000ULL;
753 do_div(ns, this_tsc_khz);
754 }
755 hrtimer_start(&apic->lapic_timer.timer,
756 ktime_add_ns(now, ns), HRTIMER_MODE_ABS);
757
758 local_irq_restore(flags);
759 }
760}
761
762static void apic_manage_nmi_watchdog(struct kvm_lapic *apic, u32 lvt0_val)
763{
764 int nmi_wd_enabled = apic_lvt_nmi_mode(apic_get_reg(apic, APIC_LVT0));
765
766 if (apic_lvt_nmi_mode(lvt0_val)) {
767 if (!nmi_wd_enabled) {
768 apic_debug("Receive NMI setting on APIC_LVT0 "
769 "for cpu %d\n", apic->vcpu->vcpu_id);
770 apic->vcpu->kvm->arch.vapics_in_nmi_mode++;
771 }
772 } else if (nmi_wd_enabled)
773 apic->vcpu->kvm->arch.vapics_in_nmi_mode--;
774}
775
776static int apic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val)
777{
778 int ret = 0;
779
780 trace_kvm_apic_write(reg, val);
781
782 switch (reg) {
783 case APIC_ID: /* Local APIC ID */
784 if (!apic_x2apic_mode(apic))
785 apic_set_reg(apic, APIC_ID, val);
786 else
787 ret = 1;
788 break;
789
790 case APIC_TASKPRI:
791 report_tpr_access(apic, true);
792 apic_set_tpr(apic, val & 0xff);
793 break;
794
795 case APIC_EOI:
796 apic_set_eoi(apic);
797 break;
798
799 case APIC_LDR:
800 if (!apic_x2apic_mode(apic))
801 apic_set_reg(apic, APIC_LDR, val & APIC_LDR_MASK);
802 else
803 ret = 1;
804 break;
805
806 case APIC_DFR:
807 if (!apic_x2apic_mode(apic))
808 apic_set_reg(apic, APIC_DFR, val | 0x0FFFFFFF);
809 else
810 ret = 1;
811 break;
812
813 case APIC_SPIV: {
814 u32 mask = 0x3ff;
815 if (apic_get_reg(apic, APIC_LVR) & APIC_LVR_DIRECTED_EOI)
816 mask |= APIC_SPIV_DIRECTED_EOI;
817 apic_set_reg(apic, APIC_SPIV, val & mask);
818 if (!(val & APIC_SPIV_APIC_ENABLED)) {
819 int i;
820 u32 lvt_val;
821
822 for (i = 0; i < APIC_LVT_NUM; i++) {
823 lvt_val = apic_get_reg(apic,
824 APIC_LVTT + 0x10 * i);
825 apic_set_reg(apic, APIC_LVTT + 0x10 * i,
826 lvt_val | APIC_LVT_MASKED);
827 }
828 atomic_set(&apic->lapic_timer.pending, 0);
829
830 }
831 break;
832 }
833 case APIC_ICR:
834 /* No delay here, so we always clear the pending bit */
835 apic_set_reg(apic, APIC_ICR, val & ~(1 << 12));
836 apic_send_ipi(apic);
837 break;
838
839 case APIC_ICR2:
840 if (!apic_x2apic_mode(apic))
841 val &= 0xff000000;
842 apic_set_reg(apic, APIC_ICR2, val);
843 break;
844
845 case APIC_LVT0:
846 apic_manage_nmi_watchdog(apic, val);
847 case APIC_LVTTHMR:
848 case APIC_LVTPC:
849 case APIC_LVT1:
850 case APIC_LVTERR:
851 /* TODO: Check vector */
852 if (!apic_sw_enabled(apic))
853 val |= APIC_LVT_MASKED;
854
855 val &= apic_lvt_mask[(reg - APIC_LVTT) >> 4];
856 apic_set_reg(apic, reg, val);
857
858 break;
859
860 case APIC_LVTT:
861 if ((apic_get_reg(apic, APIC_LVTT) &
862 apic->lapic_timer.timer_mode_mask) !=
863 (val & apic->lapic_timer.timer_mode_mask))
864 hrtimer_cancel(&apic->lapic_timer.timer);
865
866 if (!apic_sw_enabled(apic))
867 val |= APIC_LVT_MASKED;
868 val &= (apic_lvt_mask[0] | apic->lapic_timer.timer_mode_mask);
869 apic_set_reg(apic, APIC_LVTT, val);
870 break;
871
872 case APIC_TMICT:
873 if (apic_lvtt_tscdeadline(apic))
874 break;
875
876 hrtimer_cancel(&apic->lapic_timer.timer);
877 apic_set_reg(apic, APIC_TMICT, val);
878 start_apic_timer(apic);
879 break;
880
881 case APIC_TDCR:
882 if (val & 4)
883 apic_debug("KVM_WRITE:TDCR %x\n", val);
884 apic_set_reg(apic, APIC_TDCR, val);
885 update_divide_count(apic);
886 break;
887
888 case APIC_ESR:
889 if (apic_x2apic_mode(apic) && val != 0) {
890 apic_debug("KVM_WRITE:ESR not zero %x\n", val);
891 ret = 1;
892 }
893 break;
894
895 case APIC_SELF_IPI:
896 if (apic_x2apic_mode(apic)) {
897 apic_reg_write(apic, APIC_ICR, 0x40000 | (val & 0xff));
898 } else
899 ret = 1;
900 break;
901 default:
902 ret = 1;
903 break;
904 }
905 if (ret)
906 apic_debug("Local APIC Write to read-only register %x\n", reg);
907 return ret;
908}
909
910static int apic_mmio_write(struct kvm_io_device *this,
911 gpa_t address, int len, const void *data)
912{
913 struct kvm_lapic *apic = to_lapic(this);
914 unsigned int offset = address - apic->base_address;
915 u32 val;
916
917 if (!apic_mmio_in_range(apic, address))
918 return -EOPNOTSUPP;
919
920 /*
921 * APIC register must be aligned on 128-bits boundary.
922 * 32/64/128 bits registers must be accessed thru 32 bits.
923 * Refer SDM 8.4.1
924 */
925 if (len != 4 || (offset & 0xf)) {
926 /* Don't shout loud, $infamous_os would cause only noise. */
927 apic_debug("apic write: bad size=%d %lx\n", len, (long)address);
928 return 0;
929 }
930
931 val = *(u32*)data;
932
933 /* too common printing */
934 if (offset != APIC_EOI)
935 apic_debug("%s: offset 0x%x with length 0x%x, and value is "
936 "0x%x\n", __func__, offset, len, val);
937
938 apic_reg_write(apic, offset & 0xff0, val);
939
940 return 0;
941}
942
943void kvm_lapic_set_eoi(struct kvm_vcpu *vcpu)
944{
945 struct kvm_lapic *apic = vcpu->arch.apic;
946
947 if (apic)
948 apic_reg_write(vcpu->arch.apic, APIC_EOI, 0);
949}
950EXPORT_SYMBOL_GPL(kvm_lapic_set_eoi);
951
952void kvm_free_lapic(struct kvm_vcpu *vcpu)
953{
954 if (!vcpu->arch.apic)
955 return;
956
957 hrtimer_cancel(&vcpu->arch.apic->lapic_timer.timer);
958
959 if (vcpu->arch.apic->regs)
960 free_page((unsigned long)vcpu->arch.apic->regs);
961
962 kfree(vcpu->arch.apic);
963}
964
965/*
966 *----------------------------------------------------------------------
967 * LAPIC interface
968 *----------------------------------------------------------------------
969 */
970
971u64 kvm_get_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu)
972{
973 struct kvm_lapic *apic = vcpu->arch.apic;
974 if (!apic)
975 return 0;
976
977 if (apic_lvtt_oneshot(apic) || apic_lvtt_period(apic))
978 return 0;
979
980 return apic->lapic_timer.tscdeadline;
981}
982
983void kvm_set_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu, u64 data)
984{
985 struct kvm_lapic *apic = vcpu->arch.apic;
986 if (!apic)
987 return;
988
989 if (apic_lvtt_oneshot(apic) || apic_lvtt_period(apic))
990 return;
991
992 hrtimer_cancel(&apic->lapic_timer.timer);
993 apic->lapic_timer.tscdeadline = data;
994 start_apic_timer(apic);
995}
996
997void kvm_lapic_set_tpr(struct kvm_vcpu *vcpu, unsigned long cr8)
998{
999 struct kvm_lapic *apic = vcpu->arch.apic;
1000
1001 if (!apic)
1002 return;
1003 apic_set_tpr(apic, ((cr8 & 0x0f) << 4)
1004 | (apic_get_reg(apic, APIC_TASKPRI) & 4));
1005}
1006
1007u64 kvm_lapic_get_cr8(struct kvm_vcpu *vcpu)
1008{
1009 struct kvm_lapic *apic = vcpu->arch.apic;
1010 u64 tpr;
1011
1012 if (!apic)
1013 return 0;
1014 tpr = (u64) apic_get_reg(apic, APIC_TASKPRI);
1015
1016 return (tpr & 0xf0) >> 4;
1017}
1018
1019void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
1020{
1021 struct kvm_lapic *apic = vcpu->arch.apic;
1022
1023 if (!apic) {
1024 value |= MSR_IA32_APICBASE_BSP;
1025 vcpu->arch.apic_base = value;
1026 return;
1027 }
1028
1029 if (!kvm_vcpu_is_bsp(apic->vcpu))
1030 value &= ~MSR_IA32_APICBASE_BSP;
1031
1032 vcpu->arch.apic_base = value;
1033 if (apic_x2apic_mode(apic)) {
1034 u32 id = kvm_apic_id(apic);
1035 u32 ldr = ((id & ~0xf) << 16) | (1 << (id & 0xf));
1036 apic_set_reg(apic, APIC_LDR, ldr);
1037 }
1038 apic->base_address = apic->vcpu->arch.apic_base &
1039 MSR_IA32_APICBASE_BASE;
1040
1041 /* with FSB delivery interrupt, we can restart APIC functionality */
1042 apic_debug("apic base msr is 0x%016" PRIx64 ", and base address is "
1043 "0x%lx.\n", apic->vcpu->arch.apic_base, apic->base_address);
1044
1045}
1046
1047void kvm_lapic_reset(struct kvm_vcpu *vcpu)
1048{
1049 struct kvm_lapic *apic;
1050 int i;
1051
1052 apic_debug("%s\n", __func__);
1053
1054 ASSERT(vcpu);
1055 apic = vcpu->arch.apic;
1056 ASSERT(apic != NULL);
1057
1058 /* Stop the timer in case it's a reset to an active apic */
1059 hrtimer_cancel(&apic->lapic_timer.timer);
1060
1061 apic_set_reg(apic, APIC_ID, vcpu->vcpu_id << 24);
1062 kvm_apic_set_version(apic->vcpu);
1063
1064 for (i = 0; i < APIC_LVT_NUM; i++)
1065 apic_set_reg(apic, APIC_LVTT + 0x10 * i, APIC_LVT_MASKED);
1066 apic_set_reg(apic, APIC_LVT0,
1067 SET_APIC_DELIVERY_MODE(0, APIC_MODE_EXTINT));
1068
1069 apic_set_reg(apic, APIC_DFR, 0xffffffffU);
1070 apic_set_reg(apic, APIC_SPIV, 0xff);
1071 apic_set_reg(apic, APIC_TASKPRI, 0);
1072 apic_set_reg(apic, APIC_LDR, 0);
1073 apic_set_reg(apic, APIC_ESR, 0);
1074 apic_set_reg(apic, APIC_ICR, 0);
1075 apic_set_reg(apic, APIC_ICR2, 0);
1076 apic_set_reg(apic, APIC_TDCR, 0);
1077 apic_set_reg(apic, APIC_TMICT, 0);
1078 for (i = 0; i < 8; i++) {
1079 apic_set_reg(apic, APIC_IRR + 0x10 * i, 0);
1080 apic_set_reg(apic, APIC_ISR + 0x10 * i, 0);
1081 apic_set_reg(apic, APIC_TMR + 0x10 * i, 0);
1082 }
1083 apic->irr_pending = false;
1084 update_divide_count(apic);
1085 atomic_set(&apic->lapic_timer.pending, 0);
1086 if (kvm_vcpu_is_bsp(vcpu))
1087 vcpu->arch.apic_base |= MSR_IA32_APICBASE_BSP;
1088 apic_update_ppr(apic);
1089
1090 vcpu->arch.apic_arb_prio = 0;
1091 vcpu->arch.apic_attention = 0;
1092
1093 apic_debug(KERN_INFO "%s: vcpu=%p, id=%d, base_msr="
1094 "0x%016" PRIx64 ", base_address=0x%0lx.\n", __func__,
1095 vcpu, kvm_apic_id(apic),
1096 vcpu->arch.apic_base, apic->base_address);
1097}
1098
1099bool kvm_apic_present(struct kvm_vcpu *vcpu)
1100{
1101 return vcpu->arch.apic && apic_hw_enabled(vcpu->arch.apic);
1102}
1103
1104int kvm_lapic_enabled(struct kvm_vcpu *vcpu)
1105{
1106 return kvm_apic_present(vcpu) && apic_sw_enabled(vcpu->arch.apic);
1107}
1108
1109/*
1110 *----------------------------------------------------------------------
1111 * timer interface
1112 *----------------------------------------------------------------------
1113 */
1114
1115static bool lapic_is_periodic(struct kvm_timer *ktimer)
1116{
1117 struct kvm_lapic *apic = container_of(ktimer, struct kvm_lapic,
1118 lapic_timer);
1119 return apic_lvtt_period(apic);
1120}
1121
1122int apic_has_pending_timer(struct kvm_vcpu *vcpu)
1123{
1124 struct kvm_lapic *lapic = vcpu->arch.apic;
1125
1126 if (lapic && apic_enabled(lapic) && apic_lvt_enabled(lapic, APIC_LVTT))
1127 return atomic_read(&lapic->lapic_timer.pending);
1128
1129 return 0;
1130}
1131
1132int kvm_apic_local_deliver(struct kvm_lapic *apic, int lvt_type)
1133{
1134 u32 reg = apic_get_reg(apic, lvt_type);
1135 int vector, mode, trig_mode;
1136
1137 if (apic_hw_enabled(apic) && !(reg & APIC_LVT_MASKED)) {
1138 vector = reg & APIC_VECTOR_MASK;
1139 mode = reg & APIC_MODE_MASK;
1140 trig_mode = reg & APIC_LVT_LEVEL_TRIGGER;
1141 return __apic_accept_irq(apic, mode, vector, 1, trig_mode);
1142 }
1143 return 0;
1144}
1145
1146void kvm_apic_nmi_wd_deliver(struct kvm_vcpu *vcpu)
1147{
1148 struct kvm_lapic *apic = vcpu->arch.apic;
1149
1150 if (apic)
1151 kvm_apic_local_deliver(apic, APIC_LVT0);
1152}
1153
1154static struct kvm_timer_ops lapic_timer_ops = {
1155 .is_periodic = lapic_is_periodic,
1156};
1157
1158static const struct kvm_io_device_ops apic_mmio_ops = {
1159 .read = apic_mmio_read,
1160 .write = apic_mmio_write,
1161};
1162
1163int kvm_create_lapic(struct kvm_vcpu *vcpu)
1164{
1165 struct kvm_lapic *apic;
1166
1167 ASSERT(vcpu != NULL);
1168 apic_debug("apic_init %d\n", vcpu->vcpu_id);
1169
1170 apic = kzalloc(sizeof(*apic), GFP_KERNEL);
1171 if (!apic)
1172 goto nomem;
1173
1174 vcpu->arch.apic = apic;
1175
1176 apic->regs = (void *)get_zeroed_page(GFP_KERNEL);
1177 if (!apic->regs) {
1178 printk(KERN_ERR "malloc apic regs error for vcpu %x\n",
1179 vcpu->vcpu_id);
1180 goto nomem_free_apic;
1181 }
1182 apic->vcpu = vcpu;
1183
1184 hrtimer_init(&apic->lapic_timer.timer, CLOCK_MONOTONIC,
1185 HRTIMER_MODE_ABS);
1186 apic->lapic_timer.timer.function = kvm_timer_fn;
1187 apic->lapic_timer.t_ops = &lapic_timer_ops;
1188 apic->lapic_timer.kvm = vcpu->kvm;
1189 apic->lapic_timer.vcpu = vcpu;
1190
1191 apic->base_address = APIC_DEFAULT_PHYS_BASE;
1192 vcpu->arch.apic_base = APIC_DEFAULT_PHYS_BASE;
1193
1194 kvm_lapic_reset(vcpu);
1195 kvm_iodevice_init(&apic->dev, &apic_mmio_ops);
1196
1197 return 0;
1198nomem_free_apic:
1199 kfree(apic);
1200nomem:
1201 return -ENOMEM;
1202}
1203
1204int kvm_apic_has_interrupt(struct kvm_vcpu *vcpu)
1205{
1206 struct kvm_lapic *apic = vcpu->arch.apic;
1207 int highest_irr;
1208
1209 if (!apic || !apic_enabled(apic))
1210 return -1;
1211
1212 apic_update_ppr(apic);
1213 highest_irr = apic_find_highest_irr(apic);
1214 if ((highest_irr == -1) ||
1215 ((highest_irr & 0xF0) <= apic_get_reg(apic, APIC_PROCPRI)))
1216 return -1;
1217 return highest_irr;
1218}
1219
1220int kvm_apic_accept_pic_intr(struct kvm_vcpu *vcpu)
1221{
1222 u32 lvt0 = apic_get_reg(vcpu->arch.apic, APIC_LVT0);
1223 int r = 0;
1224
1225 if (!apic_hw_enabled(vcpu->arch.apic))
1226 r = 1;
1227 if ((lvt0 & APIC_LVT_MASKED) == 0 &&
1228 GET_APIC_DELIVERY_MODE(lvt0) == APIC_MODE_EXTINT)
1229 r = 1;
1230 return r;
1231}
1232
1233void kvm_inject_apic_timer_irqs(struct kvm_vcpu *vcpu)
1234{
1235 struct kvm_lapic *apic = vcpu->arch.apic;
1236
1237 if (apic && atomic_read(&apic->lapic_timer.pending) > 0) {
1238 if (kvm_apic_local_deliver(apic, APIC_LVTT))
1239 atomic_dec(&apic->lapic_timer.pending);
1240 }
1241}
1242
1243int kvm_get_apic_interrupt(struct kvm_vcpu *vcpu)
1244{
1245 int vector = kvm_apic_has_interrupt(vcpu);
1246 struct kvm_lapic *apic = vcpu->arch.apic;
1247
1248 if (vector == -1)
1249 return -1;
1250
1251 apic_set_vector(vector, apic->regs + APIC_ISR);
1252 apic_update_ppr(apic);
1253 apic_clear_irr(vector, apic);
1254 return vector;
1255}
1256
1257void kvm_apic_post_state_restore(struct kvm_vcpu *vcpu)
1258{
1259 struct kvm_lapic *apic = vcpu->arch.apic;
1260
1261 apic->base_address = vcpu->arch.apic_base &
1262 MSR_IA32_APICBASE_BASE;
1263 kvm_apic_set_version(vcpu);
1264
1265 apic_update_ppr(apic);
1266 hrtimer_cancel(&apic->lapic_timer.timer);
1267 update_divide_count(apic);
1268 start_apic_timer(apic);
1269 apic->irr_pending = true;
1270 kvm_make_request(KVM_REQ_EVENT, vcpu);
1271}
1272
1273void __kvm_migrate_apic_timer(struct kvm_vcpu *vcpu)
1274{
1275 struct kvm_lapic *apic = vcpu->arch.apic;
1276 struct hrtimer *timer;
1277
1278 if (!apic)
1279 return;
1280
1281 timer = &apic->lapic_timer.timer;
1282 if (hrtimer_cancel(timer))
1283 hrtimer_start_expires(timer, HRTIMER_MODE_ABS);
1284}
1285
1286void kvm_lapic_sync_from_vapic(struct kvm_vcpu *vcpu)
1287{
1288 u32 data;
1289 void *vapic;
1290
1291 if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
1292 return;
1293
1294 vapic = kmap_atomic(vcpu->arch.apic->vapic_page);
1295 data = *(u32 *)(vapic + offset_in_page(vcpu->arch.apic->vapic_addr));
1296 kunmap_atomic(vapic);
1297
1298 apic_set_tpr(vcpu->arch.apic, data & 0xff);
1299}
1300
1301void kvm_lapic_sync_to_vapic(struct kvm_vcpu *vcpu)
1302{
1303 u32 data, tpr;
1304 int max_irr, max_isr;
1305 struct kvm_lapic *apic;
1306 void *vapic;
1307
1308 if (!test_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention))
1309 return;
1310
1311 apic = vcpu->arch.apic;
1312 tpr = apic_get_reg(apic, APIC_TASKPRI) & 0xff;
1313 max_irr = apic_find_highest_irr(apic);
1314 if (max_irr < 0)
1315 max_irr = 0;
1316 max_isr = apic_find_highest_isr(apic);
1317 if (max_isr < 0)
1318 max_isr = 0;
1319 data = (tpr & 0xff) | ((max_isr & 0xf0) << 8) | (max_irr << 24);
1320
1321 vapic = kmap_atomic(vcpu->arch.apic->vapic_page);
1322 *(u32 *)(vapic + offset_in_page(vcpu->arch.apic->vapic_addr)) = data;
1323 kunmap_atomic(vapic);
1324}
1325
1326void kvm_lapic_set_vapic_addr(struct kvm_vcpu *vcpu, gpa_t vapic_addr)
1327{
1328 vcpu->arch.apic->vapic_addr = vapic_addr;
1329 if (vapic_addr)
1330 __set_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
1331 else
1332 __clear_bit(KVM_APIC_CHECK_VAPIC, &vcpu->arch.apic_attention);
1333}
1334
1335int kvm_x2apic_msr_write(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1336{
1337 struct kvm_lapic *apic = vcpu->arch.apic;
1338 u32 reg = (msr - APIC_BASE_MSR) << 4;
1339
1340 if (!irqchip_in_kernel(vcpu->kvm) || !apic_x2apic_mode(apic))
1341 return 1;
1342
1343 /* if this is ICR write vector before command */
1344 if (msr == 0x830)
1345 apic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
1346 return apic_reg_write(apic, reg, (u32)data);
1347}
1348
1349int kvm_x2apic_msr_read(struct kvm_vcpu *vcpu, u32 msr, u64 *data)
1350{
1351 struct kvm_lapic *apic = vcpu->arch.apic;
1352 u32 reg = (msr - APIC_BASE_MSR) << 4, low, high = 0;
1353
1354 if (!irqchip_in_kernel(vcpu->kvm) || !apic_x2apic_mode(apic))
1355 return 1;
1356
1357 if (apic_reg_read(apic, reg, 4, &low))
1358 return 1;
1359 if (msr == 0x830)
1360 apic_reg_read(apic, APIC_ICR2, 4, &high);
1361
1362 *data = (((u64)high) << 32) | low;
1363
1364 return 0;
1365}
1366
1367int kvm_hv_vapic_msr_write(struct kvm_vcpu *vcpu, u32 reg, u64 data)
1368{
1369 struct kvm_lapic *apic = vcpu->arch.apic;
1370
1371 if (!irqchip_in_kernel(vcpu->kvm))
1372 return 1;
1373
1374 /* if this is ICR write vector before command */
1375 if (reg == APIC_ICR)
1376 apic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
1377 return apic_reg_write(apic, reg, (u32)data);
1378}
1379
1380int kvm_hv_vapic_msr_read(struct kvm_vcpu *vcpu, u32 reg, u64 *data)
1381{
1382 struct kvm_lapic *apic = vcpu->arch.apic;
1383 u32 low, high = 0;
1384
1385 if (!irqchip_in_kernel(vcpu->kvm))
1386 return 1;
1387
1388 if (apic_reg_read(apic, reg, 4, &low))
1389 return 1;
1390 if (reg == APIC_ICR)
1391 apic_reg_read(apic, APIC_ICR2, 4, &high);
1392
1393 *data = (((u64)high) << 32) | low;
1394
1395 return 0;
1396}