xref: /linux/arch/x86/kvm/x86.c (revision 4ab5a5d2a4a2289c2af07accbec7170ca5671f41)
1 /*
2  * Kernel-based Virtual Machine driver for Linux
3  *
4  * derived from drivers/kvm/kvm_main.c
5  *
6  * Copyright (C) 2006 Qumranet, Inc.
7  * Copyright (C) 2008 Qumranet, Inc.
8  * Copyright IBM Corporation, 2008
9  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
10  *
11  * Authors:
12  *   Avi Kivity   <avi@qumranet.com>
13  *   Yaniv Kamay  <yaniv@qumranet.com>
14  *   Amit Shah    <amit.shah@qumranet.com>
15  *   Ben-Ami Yassour <benami@il.ibm.com>
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 
22 #include <linux/kvm_host.h>
23 #include "irq.h"
24 #include "mmu.h"
25 #include "i8254.h"
26 #include "tss.h"
27 #include "kvm_cache_regs.h"
28 #include "x86.h"
29 #include "cpuid.h"
30 #include "pmu.h"
31 #include "hyperv.h"
32 
33 #include <linux/clocksource.h>
34 #include <linux/interrupt.h>
35 #include <linux/kvm.h>
36 #include <linux/fs.h>
37 #include <linux/vmalloc.h>
38 #include <linux/export.h>
39 #include <linux/moduleparam.h>
40 #include <linux/mman.h>
41 #include <linux/highmem.h>
42 #include <linux/iommu.h>
43 #include <linux/intel-iommu.h>
44 #include <linux/cpufreq.h>
45 #include <linux/user-return-notifier.h>
46 #include <linux/srcu.h>
47 #include <linux/slab.h>
48 #include <linux/perf_event.h>
49 #include <linux/uaccess.h>
50 #include <linux/hash.h>
51 #include <linux/pci.h>
52 #include <linux/timekeeper_internal.h>
53 #include <linux/pvclock_gtod.h>
54 #include <linux/kvm_irqfd.h>
55 #include <linux/irqbypass.h>
56 #include <linux/sched/stat.h>
57 #include <linux/mem_encrypt.h>
58 
59 #include <trace/events/kvm.h>
60 
61 #include <asm/debugreg.h>
62 #include <asm/msr.h>
63 #include <asm/desc.h>
64 #include <asm/mce.h>
65 #include <linux/kernel_stat.h>
66 #include <asm/fpu/internal.h> /* Ugh! */
67 #include <asm/pvclock.h>
68 #include <asm/div64.h>
69 #include <asm/irq_remapping.h>
70 #include <asm/mshyperv.h>
71 #include <asm/hypervisor.h>
72 
73 #define CREATE_TRACE_POINTS
74 #include "trace.h"
75 
76 #define MAX_IO_MSRS 256
77 #define KVM_MAX_MCE_BANKS 32
78 u64 __read_mostly kvm_mce_cap_supported = MCG_CTL_P | MCG_SER_P;
79 EXPORT_SYMBOL_GPL(kvm_mce_cap_supported);
80 
81 #define emul_to_vcpu(ctxt) \
82 	container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
83 
84 /* EFER defaults:
85  * - enable syscall per default because its emulated by KVM
86  * - enable LME and LMA per default on 64 bit KVM
87  */
88 #ifdef CONFIG_X86_64
89 static
90 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
91 #else
92 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
93 #endif
94 
95 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
96 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
97 
98 #define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \
99                                     KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
100 
101 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
102 static void process_nmi(struct kvm_vcpu *vcpu);
103 static void enter_smm(struct kvm_vcpu *vcpu);
104 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
105 static void store_regs(struct kvm_vcpu *vcpu);
106 static int sync_regs(struct kvm_vcpu *vcpu);
107 
108 struct kvm_x86_ops *kvm_x86_ops __read_mostly;
109 EXPORT_SYMBOL_GPL(kvm_x86_ops);
110 
111 static bool __read_mostly ignore_msrs = 0;
112 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
113 
114 static bool __read_mostly report_ignored_msrs = true;
115 module_param(report_ignored_msrs, bool, S_IRUGO | S_IWUSR);
116 
117 unsigned int min_timer_period_us = 200;
118 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
119 
120 static bool __read_mostly kvmclock_periodic_sync = true;
121 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
122 
123 bool __read_mostly kvm_has_tsc_control;
124 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
125 u32  __read_mostly kvm_max_guest_tsc_khz;
126 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
127 u8   __read_mostly kvm_tsc_scaling_ratio_frac_bits;
128 EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
129 u64  __read_mostly kvm_max_tsc_scaling_ratio;
130 EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
131 u64 __read_mostly kvm_default_tsc_scaling_ratio;
132 EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio);
133 
134 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
135 static u32 __read_mostly tsc_tolerance_ppm = 250;
136 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
137 
138 /* lapic timer advance (tscdeadline mode only) in nanoseconds */
139 unsigned int __read_mostly lapic_timer_advance_ns = 1000;
140 module_param(lapic_timer_advance_ns, uint, S_IRUGO | S_IWUSR);
141 EXPORT_SYMBOL_GPL(lapic_timer_advance_ns);
142 
143 static bool __read_mostly vector_hashing = true;
144 module_param(vector_hashing, bool, S_IRUGO);
145 
146 bool __read_mostly enable_vmware_backdoor = false;
147 module_param(enable_vmware_backdoor, bool, S_IRUGO);
148 EXPORT_SYMBOL_GPL(enable_vmware_backdoor);
149 
150 static bool __read_mostly force_emulation_prefix = false;
151 module_param(force_emulation_prefix, bool, S_IRUGO);
152 
153 #define KVM_NR_SHARED_MSRS 16
154 
155 struct kvm_shared_msrs_global {
156 	int nr;
157 	u32 msrs[KVM_NR_SHARED_MSRS];
158 };
159 
160 struct kvm_shared_msrs {
161 	struct user_return_notifier urn;
162 	bool registered;
163 	struct kvm_shared_msr_values {
164 		u64 host;
165 		u64 curr;
166 	} values[KVM_NR_SHARED_MSRS];
167 };
168 
169 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
170 static struct kvm_shared_msrs __percpu *shared_msrs;
171 
172 struct kvm_stats_debugfs_item debugfs_entries[] = {
173 	{ "pf_fixed", VCPU_STAT(pf_fixed) },
174 	{ "pf_guest", VCPU_STAT(pf_guest) },
175 	{ "tlb_flush", VCPU_STAT(tlb_flush) },
176 	{ "invlpg", VCPU_STAT(invlpg) },
177 	{ "exits", VCPU_STAT(exits) },
178 	{ "io_exits", VCPU_STAT(io_exits) },
179 	{ "mmio_exits", VCPU_STAT(mmio_exits) },
180 	{ "signal_exits", VCPU_STAT(signal_exits) },
181 	{ "irq_window", VCPU_STAT(irq_window_exits) },
182 	{ "nmi_window", VCPU_STAT(nmi_window_exits) },
183 	{ "halt_exits", VCPU_STAT(halt_exits) },
184 	{ "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
185 	{ "halt_attempted_poll", VCPU_STAT(halt_attempted_poll) },
186 	{ "halt_poll_invalid", VCPU_STAT(halt_poll_invalid) },
187 	{ "halt_wakeup", VCPU_STAT(halt_wakeup) },
188 	{ "hypercalls", VCPU_STAT(hypercalls) },
189 	{ "request_irq", VCPU_STAT(request_irq_exits) },
190 	{ "irq_exits", VCPU_STAT(irq_exits) },
191 	{ "host_state_reload", VCPU_STAT(host_state_reload) },
192 	{ "fpu_reload", VCPU_STAT(fpu_reload) },
193 	{ "insn_emulation", VCPU_STAT(insn_emulation) },
194 	{ "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
195 	{ "irq_injections", VCPU_STAT(irq_injections) },
196 	{ "nmi_injections", VCPU_STAT(nmi_injections) },
197 	{ "req_event", VCPU_STAT(req_event) },
198 	{ "l1d_flush", VCPU_STAT(l1d_flush) },
199 	{ "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
200 	{ "mmu_pte_write", VM_STAT(mmu_pte_write) },
201 	{ "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
202 	{ "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
203 	{ "mmu_flooded", VM_STAT(mmu_flooded) },
204 	{ "mmu_recycled", VM_STAT(mmu_recycled) },
205 	{ "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
206 	{ "mmu_unsync", VM_STAT(mmu_unsync) },
207 	{ "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
208 	{ "largepages", VM_STAT(lpages) },
209 	{ "max_mmu_page_hash_collisions",
210 		VM_STAT(max_mmu_page_hash_collisions) },
211 	{ NULL }
212 };
213 
214 u64 __read_mostly host_xcr0;
215 
216 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
217 
218 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
219 {
220 	int i;
221 	for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
222 		vcpu->arch.apf.gfns[i] = ~0;
223 }
224 
225 static void kvm_on_user_return(struct user_return_notifier *urn)
226 {
227 	unsigned slot;
228 	struct kvm_shared_msrs *locals
229 		= container_of(urn, struct kvm_shared_msrs, urn);
230 	struct kvm_shared_msr_values *values;
231 	unsigned long flags;
232 
233 	/*
234 	 * Disabling irqs at this point since the following code could be
235 	 * interrupted and executed through kvm_arch_hardware_disable()
236 	 */
237 	local_irq_save(flags);
238 	if (locals->registered) {
239 		locals->registered = false;
240 		user_return_notifier_unregister(urn);
241 	}
242 	local_irq_restore(flags);
243 	for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
244 		values = &locals->values[slot];
245 		if (values->host != values->curr) {
246 			wrmsrl(shared_msrs_global.msrs[slot], values->host);
247 			values->curr = values->host;
248 		}
249 	}
250 }
251 
252 static void shared_msr_update(unsigned slot, u32 msr)
253 {
254 	u64 value;
255 	unsigned int cpu = smp_processor_id();
256 	struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
257 
258 	/* only read, and nobody should modify it at this time,
259 	 * so don't need lock */
260 	if (slot >= shared_msrs_global.nr) {
261 		printk(KERN_ERR "kvm: invalid MSR slot!");
262 		return;
263 	}
264 	rdmsrl_safe(msr, &value);
265 	smsr->values[slot].host = value;
266 	smsr->values[slot].curr = value;
267 }
268 
269 void kvm_define_shared_msr(unsigned slot, u32 msr)
270 {
271 	BUG_ON(slot >= KVM_NR_SHARED_MSRS);
272 	shared_msrs_global.msrs[slot] = msr;
273 	if (slot >= shared_msrs_global.nr)
274 		shared_msrs_global.nr = slot + 1;
275 }
276 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
277 
278 static void kvm_shared_msr_cpu_online(void)
279 {
280 	unsigned i;
281 
282 	for (i = 0; i < shared_msrs_global.nr; ++i)
283 		shared_msr_update(i, shared_msrs_global.msrs[i]);
284 }
285 
286 int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
287 {
288 	unsigned int cpu = smp_processor_id();
289 	struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
290 	int err;
291 
292 	if (((value ^ smsr->values[slot].curr) & mask) == 0)
293 		return 0;
294 	smsr->values[slot].curr = value;
295 	err = wrmsrl_safe(shared_msrs_global.msrs[slot], value);
296 	if (err)
297 		return 1;
298 
299 	if (!smsr->registered) {
300 		smsr->urn.on_user_return = kvm_on_user_return;
301 		user_return_notifier_register(&smsr->urn);
302 		smsr->registered = true;
303 	}
304 	return 0;
305 }
306 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
307 
308 static void drop_user_return_notifiers(void)
309 {
310 	unsigned int cpu = smp_processor_id();
311 	struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
312 
313 	if (smsr->registered)
314 		kvm_on_user_return(&smsr->urn);
315 }
316 
317 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
318 {
319 	return vcpu->arch.apic_base;
320 }
321 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
322 
323 enum lapic_mode kvm_get_apic_mode(struct kvm_vcpu *vcpu)
324 {
325 	return kvm_apic_mode(kvm_get_apic_base(vcpu));
326 }
327 EXPORT_SYMBOL_GPL(kvm_get_apic_mode);
328 
329 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
330 {
331 	enum lapic_mode old_mode = kvm_get_apic_mode(vcpu);
332 	enum lapic_mode new_mode = kvm_apic_mode(msr_info->data);
333 	u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) | 0x2ff |
334 		(guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
335 
336 	if ((msr_info->data & reserved_bits) != 0 || new_mode == LAPIC_MODE_INVALID)
337 		return 1;
338 	if (!msr_info->host_initiated) {
339 		if (old_mode == LAPIC_MODE_X2APIC && new_mode == LAPIC_MODE_XAPIC)
340 			return 1;
341 		if (old_mode == LAPIC_MODE_DISABLED && new_mode == LAPIC_MODE_X2APIC)
342 			return 1;
343 	}
344 
345 	kvm_lapic_set_base(vcpu, msr_info->data);
346 	return 0;
347 }
348 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
349 
350 asmlinkage __visible void kvm_spurious_fault(void)
351 {
352 	/* Fault while not rebooting.  We want the trace. */
353 	BUG();
354 }
355 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
356 
357 #define EXCPT_BENIGN		0
358 #define EXCPT_CONTRIBUTORY	1
359 #define EXCPT_PF		2
360 
361 static int exception_class(int vector)
362 {
363 	switch (vector) {
364 	case PF_VECTOR:
365 		return EXCPT_PF;
366 	case DE_VECTOR:
367 	case TS_VECTOR:
368 	case NP_VECTOR:
369 	case SS_VECTOR:
370 	case GP_VECTOR:
371 		return EXCPT_CONTRIBUTORY;
372 	default:
373 		break;
374 	}
375 	return EXCPT_BENIGN;
376 }
377 
378 #define EXCPT_FAULT		0
379 #define EXCPT_TRAP		1
380 #define EXCPT_ABORT		2
381 #define EXCPT_INTERRUPT		3
382 
383 static int exception_type(int vector)
384 {
385 	unsigned int mask;
386 
387 	if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
388 		return EXCPT_INTERRUPT;
389 
390 	mask = 1 << vector;
391 
392 	/* #DB is trap, as instruction watchpoints are handled elsewhere */
393 	if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
394 		return EXCPT_TRAP;
395 
396 	if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
397 		return EXCPT_ABORT;
398 
399 	/* Reserved exceptions will result in fault */
400 	return EXCPT_FAULT;
401 }
402 
403 void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu)
404 {
405 	unsigned nr = vcpu->arch.exception.nr;
406 	bool has_payload = vcpu->arch.exception.has_payload;
407 	unsigned long payload = vcpu->arch.exception.payload;
408 
409 	if (!has_payload)
410 		return;
411 
412 	switch (nr) {
413 	case DB_VECTOR:
414 		/*
415 		 * "Certain debug exceptions may clear bit 0-3.  The
416 		 * remaining contents of the DR6 register are never
417 		 * cleared by the processor".
418 		 */
419 		vcpu->arch.dr6 &= ~DR_TRAP_BITS;
420 		/*
421 		 * DR6.RTM is set by all #DB exceptions that don't clear it.
422 		 */
423 		vcpu->arch.dr6 |= DR6_RTM;
424 		vcpu->arch.dr6 |= payload;
425 		/*
426 		 * Bit 16 should be set in the payload whenever the #DB
427 		 * exception should clear DR6.RTM. This makes the payload
428 		 * compatible with the pending debug exceptions under VMX.
429 		 * Though not currently documented in the SDM, this also
430 		 * makes the payload compatible with the exit qualification
431 		 * for #DB exceptions under VMX.
432 		 */
433 		vcpu->arch.dr6 ^= payload & DR6_RTM;
434 		break;
435 	case PF_VECTOR:
436 		vcpu->arch.cr2 = payload;
437 		break;
438 	}
439 
440 	vcpu->arch.exception.has_payload = false;
441 	vcpu->arch.exception.payload = 0;
442 }
443 EXPORT_SYMBOL_GPL(kvm_deliver_exception_payload);
444 
445 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
446 		unsigned nr, bool has_error, u32 error_code,
447 	        bool has_payload, unsigned long payload, bool reinject)
448 {
449 	u32 prev_nr;
450 	int class1, class2;
451 
452 	kvm_make_request(KVM_REQ_EVENT, vcpu);
453 
454 	if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) {
455 	queue:
456 		if (has_error && !is_protmode(vcpu))
457 			has_error = false;
458 		if (reinject) {
459 			/*
460 			 * On vmentry, vcpu->arch.exception.pending is only
461 			 * true if an event injection was blocked by
462 			 * nested_run_pending.  In that case, however,
463 			 * vcpu_enter_guest requests an immediate exit,
464 			 * and the guest shouldn't proceed far enough to
465 			 * need reinjection.
466 			 */
467 			WARN_ON_ONCE(vcpu->arch.exception.pending);
468 			vcpu->arch.exception.injected = true;
469 			if (WARN_ON_ONCE(has_payload)) {
470 				/*
471 				 * A reinjected event has already
472 				 * delivered its payload.
473 				 */
474 				has_payload = false;
475 				payload = 0;
476 			}
477 		} else {
478 			vcpu->arch.exception.pending = true;
479 			vcpu->arch.exception.injected = false;
480 		}
481 		vcpu->arch.exception.has_error_code = has_error;
482 		vcpu->arch.exception.nr = nr;
483 		vcpu->arch.exception.error_code = error_code;
484 		vcpu->arch.exception.has_payload = has_payload;
485 		vcpu->arch.exception.payload = payload;
486 		/*
487 		 * In guest mode, payload delivery should be deferred,
488 		 * so that the L1 hypervisor can intercept #PF before
489 		 * CR2 is modified (or intercept #DB before DR6 is
490 		 * modified under nVMX).  However, for ABI
491 		 * compatibility with KVM_GET_VCPU_EVENTS and
492 		 * KVM_SET_VCPU_EVENTS, we can't delay payload
493 		 * delivery unless userspace has enabled this
494 		 * functionality via the per-VM capability,
495 		 * KVM_CAP_EXCEPTION_PAYLOAD.
496 		 */
497 		if (!vcpu->kvm->arch.exception_payload_enabled ||
498 		    !is_guest_mode(vcpu))
499 			kvm_deliver_exception_payload(vcpu);
500 		return;
501 	}
502 
503 	/* to check exception */
504 	prev_nr = vcpu->arch.exception.nr;
505 	if (prev_nr == DF_VECTOR) {
506 		/* triple fault -> shutdown */
507 		kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
508 		return;
509 	}
510 	class1 = exception_class(prev_nr);
511 	class2 = exception_class(nr);
512 	if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
513 		|| (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
514 		/*
515 		 * Generate double fault per SDM Table 5-5.  Set
516 		 * exception.pending = true so that the double fault
517 		 * can trigger a nested vmexit.
518 		 */
519 		vcpu->arch.exception.pending = true;
520 		vcpu->arch.exception.injected = false;
521 		vcpu->arch.exception.has_error_code = true;
522 		vcpu->arch.exception.nr = DF_VECTOR;
523 		vcpu->arch.exception.error_code = 0;
524 		vcpu->arch.exception.has_payload = false;
525 		vcpu->arch.exception.payload = 0;
526 	} else
527 		/* replace previous exception with a new one in a hope
528 		   that instruction re-execution will regenerate lost
529 		   exception */
530 		goto queue;
531 }
532 
533 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
534 {
535 	kvm_multiple_exception(vcpu, nr, false, 0, false, 0, false);
536 }
537 EXPORT_SYMBOL_GPL(kvm_queue_exception);
538 
539 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
540 {
541 	kvm_multiple_exception(vcpu, nr, false, 0, false, 0, true);
542 }
543 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
544 
545 static void kvm_queue_exception_p(struct kvm_vcpu *vcpu, unsigned nr,
546 				  unsigned long payload)
547 {
548 	kvm_multiple_exception(vcpu, nr, false, 0, true, payload, false);
549 }
550 
551 static void kvm_queue_exception_e_p(struct kvm_vcpu *vcpu, unsigned nr,
552 				    u32 error_code, unsigned long payload)
553 {
554 	kvm_multiple_exception(vcpu, nr, true, error_code,
555 			       true, payload, false);
556 }
557 
558 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
559 {
560 	if (err)
561 		kvm_inject_gp(vcpu, 0);
562 	else
563 		return kvm_skip_emulated_instruction(vcpu);
564 
565 	return 1;
566 }
567 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
568 
569 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
570 {
571 	++vcpu->stat.pf_guest;
572 	vcpu->arch.exception.nested_apf =
573 		is_guest_mode(vcpu) && fault->async_page_fault;
574 	if (vcpu->arch.exception.nested_apf) {
575 		vcpu->arch.apf.nested_apf_token = fault->address;
576 		kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
577 	} else {
578 		kvm_queue_exception_e_p(vcpu, PF_VECTOR, fault->error_code,
579 					fault->address);
580 	}
581 }
582 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
583 
584 static bool kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
585 {
586 	if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
587 		vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
588 	else
589 		vcpu->arch.mmu->inject_page_fault(vcpu, fault);
590 
591 	return fault->nested_page_fault;
592 }
593 
594 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
595 {
596 	atomic_inc(&vcpu->arch.nmi_queued);
597 	kvm_make_request(KVM_REQ_NMI, vcpu);
598 }
599 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
600 
601 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
602 {
603 	kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, false);
604 }
605 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
606 
607 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
608 {
609 	kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, true);
610 }
611 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
612 
613 /*
614  * Checks if cpl <= required_cpl; if true, return true.  Otherwise queue
615  * a #GP and return false.
616  */
617 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
618 {
619 	if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
620 		return true;
621 	kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
622 	return false;
623 }
624 EXPORT_SYMBOL_GPL(kvm_require_cpl);
625 
626 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
627 {
628 	if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
629 		return true;
630 
631 	kvm_queue_exception(vcpu, UD_VECTOR);
632 	return false;
633 }
634 EXPORT_SYMBOL_GPL(kvm_require_dr);
635 
636 /*
637  * This function will be used to read from the physical memory of the currently
638  * running guest. The difference to kvm_vcpu_read_guest_page is that this function
639  * can read from guest physical or from the guest's guest physical memory.
640  */
641 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
642 			    gfn_t ngfn, void *data, int offset, int len,
643 			    u32 access)
644 {
645 	struct x86_exception exception;
646 	gfn_t real_gfn;
647 	gpa_t ngpa;
648 
649 	ngpa     = gfn_to_gpa(ngfn);
650 	real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
651 	if (real_gfn == UNMAPPED_GVA)
652 		return -EFAULT;
653 
654 	real_gfn = gpa_to_gfn(real_gfn);
655 
656 	return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
657 }
658 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
659 
660 static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
661 			       void *data, int offset, int len, u32 access)
662 {
663 	return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
664 				       data, offset, len, access);
665 }
666 
667 /*
668  * Load the pae pdptrs.  Return true is they are all valid.
669  */
670 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
671 {
672 	gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
673 	unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
674 	int i;
675 	int ret;
676 	u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
677 
678 	ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
679 				      offset * sizeof(u64), sizeof(pdpte),
680 				      PFERR_USER_MASK|PFERR_WRITE_MASK);
681 	if (ret < 0) {
682 		ret = 0;
683 		goto out;
684 	}
685 	for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
686 		if ((pdpte[i] & PT_PRESENT_MASK) &&
687 		    (pdpte[i] &
688 		     vcpu->arch.mmu->guest_rsvd_check.rsvd_bits_mask[0][2])) {
689 			ret = 0;
690 			goto out;
691 		}
692 	}
693 	ret = 1;
694 
695 	memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
696 	__set_bit(VCPU_EXREG_PDPTR,
697 		  (unsigned long *)&vcpu->arch.regs_avail);
698 	__set_bit(VCPU_EXREG_PDPTR,
699 		  (unsigned long *)&vcpu->arch.regs_dirty);
700 out:
701 
702 	return ret;
703 }
704 EXPORT_SYMBOL_GPL(load_pdptrs);
705 
706 bool pdptrs_changed(struct kvm_vcpu *vcpu)
707 {
708 	u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
709 	bool changed = true;
710 	int offset;
711 	gfn_t gfn;
712 	int r;
713 
714 	if (is_long_mode(vcpu) || !is_pae(vcpu) || !is_paging(vcpu))
715 		return false;
716 
717 	if (!test_bit(VCPU_EXREG_PDPTR,
718 		      (unsigned long *)&vcpu->arch.regs_avail))
719 		return true;
720 
721 	gfn = (kvm_read_cr3(vcpu) & 0xffffffe0ul) >> PAGE_SHIFT;
722 	offset = (kvm_read_cr3(vcpu) & 0xffffffe0ul) & (PAGE_SIZE - 1);
723 	r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
724 				       PFERR_USER_MASK | PFERR_WRITE_MASK);
725 	if (r < 0)
726 		goto out;
727 	changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
728 out:
729 
730 	return changed;
731 }
732 EXPORT_SYMBOL_GPL(pdptrs_changed);
733 
734 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
735 {
736 	unsigned long old_cr0 = kvm_read_cr0(vcpu);
737 	unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
738 
739 	cr0 |= X86_CR0_ET;
740 
741 #ifdef CONFIG_X86_64
742 	if (cr0 & 0xffffffff00000000UL)
743 		return 1;
744 #endif
745 
746 	cr0 &= ~CR0_RESERVED_BITS;
747 
748 	if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
749 		return 1;
750 
751 	if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
752 		return 1;
753 
754 	if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
755 #ifdef CONFIG_X86_64
756 		if ((vcpu->arch.efer & EFER_LME)) {
757 			int cs_db, cs_l;
758 
759 			if (!is_pae(vcpu))
760 				return 1;
761 			kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
762 			if (cs_l)
763 				return 1;
764 		} else
765 #endif
766 		if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
767 						 kvm_read_cr3(vcpu)))
768 			return 1;
769 	}
770 
771 	if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
772 		return 1;
773 
774 	kvm_x86_ops->set_cr0(vcpu, cr0);
775 
776 	if ((cr0 ^ old_cr0) & X86_CR0_PG) {
777 		kvm_clear_async_pf_completion_queue(vcpu);
778 		kvm_async_pf_hash_reset(vcpu);
779 	}
780 
781 	if ((cr0 ^ old_cr0) & update_bits)
782 		kvm_mmu_reset_context(vcpu);
783 
784 	if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
785 	    kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
786 	    !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
787 		kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
788 
789 	return 0;
790 }
791 EXPORT_SYMBOL_GPL(kvm_set_cr0);
792 
793 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
794 {
795 	(void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
796 }
797 EXPORT_SYMBOL_GPL(kvm_lmsw);
798 
799 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
800 {
801 	if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
802 			!vcpu->guest_xcr0_loaded) {
803 		/* kvm_set_xcr() also depends on this */
804 		if (vcpu->arch.xcr0 != host_xcr0)
805 			xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
806 		vcpu->guest_xcr0_loaded = 1;
807 	}
808 }
809 
810 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
811 {
812 	if (vcpu->guest_xcr0_loaded) {
813 		if (vcpu->arch.xcr0 != host_xcr0)
814 			xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
815 		vcpu->guest_xcr0_loaded = 0;
816 	}
817 }
818 
819 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
820 {
821 	u64 xcr0 = xcr;
822 	u64 old_xcr0 = vcpu->arch.xcr0;
823 	u64 valid_bits;
824 
825 	/* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now  */
826 	if (index != XCR_XFEATURE_ENABLED_MASK)
827 		return 1;
828 	if (!(xcr0 & XFEATURE_MASK_FP))
829 		return 1;
830 	if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
831 		return 1;
832 
833 	/*
834 	 * Do not allow the guest to set bits that we do not support
835 	 * saving.  However, xcr0 bit 0 is always set, even if the
836 	 * emulated CPU does not support XSAVE (see fx_init).
837 	 */
838 	valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
839 	if (xcr0 & ~valid_bits)
840 		return 1;
841 
842 	if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
843 	    (!(xcr0 & XFEATURE_MASK_BNDCSR)))
844 		return 1;
845 
846 	if (xcr0 & XFEATURE_MASK_AVX512) {
847 		if (!(xcr0 & XFEATURE_MASK_YMM))
848 			return 1;
849 		if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
850 			return 1;
851 	}
852 	vcpu->arch.xcr0 = xcr0;
853 
854 	if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
855 		kvm_update_cpuid(vcpu);
856 	return 0;
857 }
858 
859 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
860 {
861 	if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
862 	    __kvm_set_xcr(vcpu, index, xcr)) {
863 		kvm_inject_gp(vcpu, 0);
864 		return 1;
865 	}
866 	return 0;
867 }
868 EXPORT_SYMBOL_GPL(kvm_set_xcr);
869 
870 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
871 {
872 	unsigned long old_cr4 = kvm_read_cr4(vcpu);
873 	unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
874 				   X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE;
875 
876 	if (cr4 & CR4_RESERVED_BITS)
877 		return 1;
878 
879 	if (!guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) && (cr4 & X86_CR4_OSXSAVE))
880 		return 1;
881 
882 	if (!guest_cpuid_has(vcpu, X86_FEATURE_SMEP) && (cr4 & X86_CR4_SMEP))
883 		return 1;
884 
885 	if (!guest_cpuid_has(vcpu, X86_FEATURE_SMAP) && (cr4 & X86_CR4_SMAP))
886 		return 1;
887 
888 	if (!guest_cpuid_has(vcpu, X86_FEATURE_FSGSBASE) && (cr4 & X86_CR4_FSGSBASE))
889 		return 1;
890 
891 	if (!guest_cpuid_has(vcpu, X86_FEATURE_PKU) && (cr4 & X86_CR4_PKE))
892 		return 1;
893 
894 	if (!guest_cpuid_has(vcpu, X86_FEATURE_LA57) && (cr4 & X86_CR4_LA57))
895 		return 1;
896 
897 	if (!guest_cpuid_has(vcpu, X86_FEATURE_UMIP) && (cr4 & X86_CR4_UMIP))
898 		return 1;
899 
900 	if (is_long_mode(vcpu)) {
901 		if (!(cr4 & X86_CR4_PAE))
902 			return 1;
903 	} else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
904 		   && ((cr4 ^ old_cr4) & pdptr_bits)
905 		   && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
906 				   kvm_read_cr3(vcpu)))
907 		return 1;
908 
909 	if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
910 		if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
911 			return 1;
912 
913 		/* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
914 		if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
915 			return 1;
916 	}
917 
918 	if (kvm_x86_ops->set_cr4(vcpu, cr4))
919 		return 1;
920 
921 	if (((cr4 ^ old_cr4) & pdptr_bits) ||
922 	    (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
923 		kvm_mmu_reset_context(vcpu);
924 
925 	if ((cr4 ^ old_cr4) & (X86_CR4_OSXSAVE | X86_CR4_PKE))
926 		kvm_update_cpuid(vcpu);
927 
928 	return 0;
929 }
930 EXPORT_SYMBOL_GPL(kvm_set_cr4);
931 
932 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
933 {
934 	bool skip_tlb_flush = false;
935 #ifdef CONFIG_X86_64
936 	bool pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
937 
938 	if (pcid_enabled) {
939 		skip_tlb_flush = cr3 & X86_CR3_PCID_NOFLUSH;
940 		cr3 &= ~X86_CR3_PCID_NOFLUSH;
941 	}
942 #endif
943 
944 	if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
945 		if (!skip_tlb_flush) {
946 			kvm_mmu_sync_roots(vcpu);
947 			kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
948 		}
949 		return 0;
950 	}
951 
952 	if (is_long_mode(vcpu) &&
953 	    (cr3 & rsvd_bits(cpuid_maxphyaddr(vcpu), 63)))
954 		return 1;
955 	else if (is_pae(vcpu) && is_paging(vcpu) &&
956 		   !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
957 		return 1;
958 
959 	kvm_mmu_new_cr3(vcpu, cr3, skip_tlb_flush);
960 	vcpu->arch.cr3 = cr3;
961 	__set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
962 
963 	return 0;
964 }
965 EXPORT_SYMBOL_GPL(kvm_set_cr3);
966 
967 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
968 {
969 	if (cr8 & CR8_RESERVED_BITS)
970 		return 1;
971 	if (lapic_in_kernel(vcpu))
972 		kvm_lapic_set_tpr(vcpu, cr8);
973 	else
974 		vcpu->arch.cr8 = cr8;
975 	return 0;
976 }
977 EXPORT_SYMBOL_GPL(kvm_set_cr8);
978 
979 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
980 {
981 	if (lapic_in_kernel(vcpu))
982 		return kvm_lapic_get_cr8(vcpu);
983 	else
984 		return vcpu->arch.cr8;
985 }
986 EXPORT_SYMBOL_GPL(kvm_get_cr8);
987 
988 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
989 {
990 	int i;
991 
992 	if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
993 		for (i = 0; i < KVM_NR_DB_REGS; i++)
994 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
995 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
996 	}
997 }
998 
999 static void kvm_update_dr6(struct kvm_vcpu *vcpu)
1000 {
1001 	if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1002 		kvm_x86_ops->set_dr6(vcpu, vcpu->arch.dr6);
1003 }
1004 
1005 static void kvm_update_dr7(struct kvm_vcpu *vcpu)
1006 {
1007 	unsigned long dr7;
1008 
1009 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1010 		dr7 = vcpu->arch.guest_debug_dr7;
1011 	else
1012 		dr7 = vcpu->arch.dr7;
1013 	kvm_x86_ops->set_dr7(vcpu, dr7);
1014 	vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
1015 	if (dr7 & DR7_BP_EN_MASK)
1016 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
1017 }
1018 
1019 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
1020 {
1021 	u64 fixed = DR6_FIXED_1;
1022 
1023 	if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
1024 		fixed |= DR6_RTM;
1025 	return fixed;
1026 }
1027 
1028 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1029 {
1030 	switch (dr) {
1031 	case 0 ... 3:
1032 		vcpu->arch.db[dr] = val;
1033 		if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1034 			vcpu->arch.eff_db[dr] = val;
1035 		break;
1036 	case 4:
1037 		/* fall through */
1038 	case 6:
1039 		if (val & 0xffffffff00000000ULL)
1040 			return -1; /* #GP */
1041 		vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
1042 		kvm_update_dr6(vcpu);
1043 		break;
1044 	case 5:
1045 		/* fall through */
1046 	default: /* 7 */
1047 		if (val & 0xffffffff00000000ULL)
1048 			return -1; /* #GP */
1049 		vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
1050 		kvm_update_dr7(vcpu);
1051 		break;
1052 	}
1053 
1054 	return 0;
1055 }
1056 
1057 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1058 {
1059 	if (__kvm_set_dr(vcpu, dr, val)) {
1060 		kvm_inject_gp(vcpu, 0);
1061 		return 1;
1062 	}
1063 	return 0;
1064 }
1065 EXPORT_SYMBOL_GPL(kvm_set_dr);
1066 
1067 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
1068 {
1069 	switch (dr) {
1070 	case 0 ... 3:
1071 		*val = vcpu->arch.db[dr];
1072 		break;
1073 	case 4:
1074 		/* fall through */
1075 	case 6:
1076 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1077 			*val = vcpu->arch.dr6;
1078 		else
1079 			*val = kvm_x86_ops->get_dr6(vcpu);
1080 		break;
1081 	case 5:
1082 		/* fall through */
1083 	default: /* 7 */
1084 		*val = vcpu->arch.dr7;
1085 		break;
1086 	}
1087 	return 0;
1088 }
1089 EXPORT_SYMBOL_GPL(kvm_get_dr);
1090 
1091 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
1092 {
1093 	u32 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
1094 	u64 data;
1095 	int err;
1096 
1097 	err = kvm_pmu_rdpmc(vcpu, ecx, &data);
1098 	if (err)
1099 		return err;
1100 	kvm_register_write(vcpu, VCPU_REGS_RAX, (u32)data);
1101 	kvm_register_write(vcpu, VCPU_REGS_RDX, data >> 32);
1102 	return err;
1103 }
1104 EXPORT_SYMBOL_GPL(kvm_rdpmc);
1105 
1106 /*
1107  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1108  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1109  *
1110  * This list is modified at module load time to reflect the
1111  * capabilities of the host cpu. This capabilities test skips MSRs that are
1112  * kvm-specific. Those are put in emulated_msrs; filtering of emulated_msrs
1113  * may depend on host virtualization features rather than host cpu features.
1114  */
1115 
1116 static u32 msrs_to_save[] = {
1117 	MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1118 	MSR_STAR,
1119 #ifdef CONFIG_X86_64
1120 	MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1121 #endif
1122 	MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
1123 	MSR_IA32_FEATURE_CONTROL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
1124 	MSR_IA32_SPEC_CTRL, MSR_IA32_ARCH_CAPABILITIES
1125 };
1126 
1127 static unsigned num_msrs_to_save;
1128 
1129 static u32 emulated_msrs[] = {
1130 	MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1131 	MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1132 	HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1133 	HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
1134 	HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
1135 	HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1136 	HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1137 	HV_X64_MSR_RESET,
1138 	HV_X64_MSR_VP_INDEX,
1139 	HV_X64_MSR_VP_RUNTIME,
1140 	HV_X64_MSR_SCONTROL,
1141 	HV_X64_MSR_STIMER0_CONFIG,
1142 	HV_X64_MSR_VP_ASSIST_PAGE,
1143 	HV_X64_MSR_REENLIGHTENMENT_CONTROL, HV_X64_MSR_TSC_EMULATION_CONTROL,
1144 	HV_X64_MSR_TSC_EMULATION_STATUS,
1145 
1146 	MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1147 	MSR_KVM_PV_EOI_EN,
1148 
1149 	MSR_IA32_TSC_ADJUST,
1150 	MSR_IA32_TSCDEADLINE,
1151 	MSR_IA32_MISC_ENABLE,
1152 	MSR_IA32_MCG_STATUS,
1153 	MSR_IA32_MCG_CTL,
1154 	MSR_IA32_MCG_EXT_CTL,
1155 	MSR_IA32_SMBASE,
1156 	MSR_SMI_COUNT,
1157 	MSR_PLATFORM_INFO,
1158 	MSR_MISC_FEATURES_ENABLES,
1159 	MSR_AMD64_VIRT_SPEC_CTRL,
1160 };
1161 
1162 static unsigned num_emulated_msrs;
1163 
1164 /*
1165  * List of msr numbers which are used to expose MSR-based features that
1166  * can be used by a hypervisor to validate requested CPU features.
1167  */
1168 static u32 msr_based_features[] = {
1169 	MSR_IA32_VMX_BASIC,
1170 	MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1171 	MSR_IA32_VMX_PINBASED_CTLS,
1172 	MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1173 	MSR_IA32_VMX_PROCBASED_CTLS,
1174 	MSR_IA32_VMX_TRUE_EXIT_CTLS,
1175 	MSR_IA32_VMX_EXIT_CTLS,
1176 	MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1177 	MSR_IA32_VMX_ENTRY_CTLS,
1178 	MSR_IA32_VMX_MISC,
1179 	MSR_IA32_VMX_CR0_FIXED0,
1180 	MSR_IA32_VMX_CR0_FIXED1,
1181 	MSR_IA32_VMX_CR4_FIXED0,
1182 	MSR_IA32_VMX_CR4_FIXED1,
1183 	MSR_IA32_VMX_VMCS_ENUM,
1184 	MSR_IA32_VMX_PROCBASED_CTLS2,
1185 	MSR_IA32_VMX_EPT_VPID_CAP,
1186 	MSR_IA32_VMX_VMFUNC,
1187 
1188 	MSR_F10H_DECFG,
1189 	MSR_IA32_UCODE_REV,
1190 	MSR_IA32_ARCH_CAPABILITIES,
1191 };
1192 
1193 static unsigned int num_msr_based_features;
1194 
1195 u64 kvm_get_arch_capabilities(void)
1196 {
1197 	u64 data;
1198 
1199 	rdmsrl_safe(MSR_IA32_ARCH_CAPABILITIES, &data);
1200 
1201 	/*
1202 	 * If we're doing cache flushes (either "always" or "cond")
1203 	 * we will do one whenever the guest does a vmlaunch/vmresume.
1204 	 * If an outer hypervisor is doing the cache flush for us
1205 	 * (VMENTER_L1D_FLUSH_NESTED_VM), we can safely pass that
1206 	 * capability to the guest too, and if EPT is disabled we're not
1207 	 * vulnerable.  Overall, only VMENTER_L1D_FLUSH_NEVER will
1208 	 * require a nested hypervisor to do a flush of its own.
1209 	 */
1210 	if (l1tf_vmx_mitigation != VMENTER_L1D_FLUSH_NEVER)
1211 		data |= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH;
1212 
1213 	return data;
1214 }
1215 EXPORT_SYMBOL_GPL(kvm_get_arch_capabilities);
1216 
1217 static int kvm_get_msr_feature(struct kvm_msr_entry *msr)
1218 {
1219 	switch (msr->index) {
1220 	case MSR_IA32_ARCH_CAPABILITIES:
1221 		msr->data = kvm_get_arch_capabilities();
1222 		break;
1223 	case MSR_IA32_UCODE_REV:
1224 		rdmsrl_safe(msr->index, &msr->data);
1225 		break;
1226 	default:
1227 		if (kvm_x86_ops->get_msr_feature(msr))
1228 			return 1;
1229 	}
1230 	return 0;
1231 }
1232 
1233 static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1234 {
1235 	struct kvm_msr_entry msr;
1236 	int r;
1237 
1238 	msr.index = index;
1239 	r = kvm_get_msr_feature(&msr);
1240 	if (r)
1241 		return r;
1242 
1243 	*data = msr.data;
1244 
1245 	return 0;
1246 }
1247 
1248 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1249 {
1250 	if (efer & efer_reserved_bits)
1251 		return false;
1252 
1253 	if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1254 			return false;
1255 
1256 	if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1257 			return false;
1258 
1259 	return true;
1260 }
1261 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1262 
1263 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
1264 {
1265 	u64 old_efer = vcpu->arch.efer;
1266 
1267 	if (!kvm_valid_efer(vcpu, efer))
1268 		return 1;
1269 
1270 	if (is_paging(vcpu)
1271 	    && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1272 		return 1;
1273 
1274 	efer &= ~EFER_LMA;
1275 	efer |= vcpu->arch.efer & EFER_LMA;
1276 
1277 	kvm_x86_ops->set_efer(vcpu, efer);
1278 
1279 	/* Update reserved bits */
1280 	if ((efer ^ old_efer) & EFER_NX)
1281 		kvm_mmu_reset_context(vcpu);
1282 
1283 	return 0;
1284 }
1285 
1286 void kvm_enable_efer_bits(u64 mask)
1287 {
1288        efer_reserved_bits &= ~mask;
1289 }
1290 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1291 
1292 /*
1293  * Writes msr value into into the appropriate "register".
1294  * Returns 0 on success, non-0 otherwise.
1295  * Assumes vcpu_load() was already called.
1296  */
1297 int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
1298 {
1299 	switch (msr->index) {
1300 	case MSR_FS_BASE:
1301 	case MSR_GS_BASE:
1302 	case MSR_KERNEL_GS_BASE:
1303 	case MSR_CSTAR:
1304 	case MSR_LSTAR:
1305 		if (is_noncanonical_address(msr->data, vcpu))
1306 			return 1;
1307 		break;
1308 	case MSR_IA32_SYSENTER_EIP:
1309 	case MSR_IA32_SYSENTER_ESP:
1310 		/*
1311 		 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1312 		 * non-canonical address is written on Intel but not on
1313 		 * AMD (which ignores the top 32-bits, because it does
1314 		 * not implement 64-bit SYSENTER).
1315 		 *
1316 		 * 64-bit code should hence be able to write a non-canonical
1317 		 * value on AMD.  Making the address canonical ensures that
1318 		 * vmentry does not fail on Intel after writing a non-canonical
1319 		 * value, and that something deterministic happens if the guest
1320 		 * invokes 64-bit SYSENTER.
1321 		 */
1322 		msr->data = get_canonical(msr->data, vcpu_virt_addr_bits(vcpu));
1323 	}
1324 	return kvm_x86_ops->set_msr(vcpu, msr);
1325 }
1326 EXPORT_SYMBOL_GPL(kvm_set_msr);
1327 
1328 /*
1329  * Adapt set_msr() to msr_io()'s calling convention
1330  */
1331 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1332 {
1333 	struct msr_data msr;
1334 	int r;
1335 
1336 	msr.index = index;
1337 	msr.host_initiated = true;
1338 	r = kvm_get_msr(vcpu, &msr);
1339 	if (r)
1340 		return r;
1341 
1342 	*data = msr.data;
1343 	return 0;
1344 }
1345 
1346 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1347 {
1348 	struct msr_data msr;
1349 
1350 	msr.data = *data;
1351 	msr.index = index;
1352 	msr.host_initiated = true;
1353 	return kvm_set_msr(vcpu, &msr);
1354 }
1355 
1356 #ifdef CONFIG_X86_64
1357 struct pvclock_gtod_data {
1358 	seqcount_t	seq;
1359 
1360 	struct { /* extract of a clocksource struct */
1361 		int vclock_mode;
1362 		u64	cycle_last;
1363 		u64	mask;
1364 		u32	mult;
1365 		u32	shift;
1366 	} clock;
1367 
1368 	u64		boot_ns;
1369 	u64		nsec_base;
1370 	u64		wall_time_sec;
1371 };
1372 
1373 static struct pvclock_gtod_data pvclock_gtod_data;
1374 
1375 static void update_pvclock_gtod(struct timekeeper *tk)
1376 {
1377 	struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1378 	u64 boot_ns;
1379 
1380 	boot_ns = ktime_to_ns(ktime_add(tk->tkr_mono.base, tk->offs_boot));
1381 
1382 	write_seqcount_begin(&vdata->seq);
1383 
1384 	/* copy pvclock gtod data */
1385 	vdata->clock.vclock_mode	= tk->tkr_mono.clock->archdata.vclock_mode;
1386 	vdata->clock.cycle_last		= tk->tkr_mono.cycle_last;
1387 	vdata->clock.mask		= tk->tkr_mono.mask;
1388 	vdata->clock.mult		= tk->tkr_mono.mult;
1389 	vdata->clock.shift		= tk->tkr_mono.shift;
1390 
1391 	vdata->boot_ns			= boot_ns;
1392 	vdata->nsec_base		= tk->tkr_mono.xtime_nsec;
1393 
1394 	vdata->wall_time_sec            = tk->xtime_sec;
1395 
1396 	write_seqcount_end(&vdata->seq);
1397 }
1398 #endif
1399 
1400 void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1401 {
1402 	/*
1403 	 * Note: KVM_REQ_PENDING_TIMER is implicitly checked in
1404 	 * vcpu_enter_guest.  This function is only called from
1405 	 * the physical CPU that is running vcpu.
1406 	 */
1407 	kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1408 }
1409 
1410 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1411 {
1412 	int version;
1413 	int r;
1414 	struct pvclock_wall_clock wc;
1415 	struct timespec64 boot;
1416 
1417 	if (!wall_clock)
1418 		return;
1419 
1420 	r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1421 	if (r)
1422 		return;
1423 
1424 	if (version & 1)
1425 		++version;  /* first time write, random junk */
1426 
1427 	++version;
1428 
1429 	if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
1430 		return;
1431 
1432 	/*
1433 	 * The guest calculates current wall clock time by adding
1434 	 * system time (updated by kvm_guest_time_update below) to the
1435 	 * wall clock specified here.  guest system time equals host
1436 	 * system time for us, thus we must fill in host boot time here.
1437 	 */
1438 	getboottime64(&boot);
1439 
1440 	if (kvm->arch.kvmclock_offset) {
1441 		struct timespec64 ts = ns_to_timespec64(kvm->arch.kvmclock_offset);
1442 		boot = timespec64_sub(boot, ts);
1443 	}
1444 	wc.sec = (u32)boot.tv_sec; /* overflow in 2106 guest time */
1445 	wc.nsec = boot.tv_nsec;
1446 	wc.version = version;
1447 
1448 	kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1449 
1450 	version++;
1451 	kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1452 }
1453 
1454 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1455 {
1456 	do_shl32_div32(dividend, divisor);
1457 	return dividend;
1458 }
1459 
1460 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
1461 			       s8 *pshift, u32 *pmultiplier)
1462 {
1463 	uint64_t scaled64;
1464 	int32_t  shift = 0;
1465 	uint64_t tps64;
1466 	uint32_t tps32;
1467 
1468 	tps64 = base_hz;
1469 	scaled64 = scaled_hz;
1470 	while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1471 		tps64 >>= 1;
1472 		shift--;
1473 	}
1474 
1475 	tps32 = (uint32_t)tps64;
1476 	while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1477 		if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1478 			scaled64 >>= 1;
1479 		else
1480 			tps32 <<= 1;
1481 		shift++;
1482 	}
1483 
1484 	*pshift = shift;
1485 	*pmultiplier = div_frac(scaled64, tps32);
1486 
1487 	pr_debug("%s: base_hz %llu => %llu, shift %d, mul %u\n",
1488 		 __func__, base_hz, scaled_hz, shift, *pmultiplier);
1489 }
1490 
1491 #ifdef CONFIG_X86_64
1492 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1493 #endif
1494 
1495 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1496 static unsigned long max_tsc_khz;
1497 
1498 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1499 {
1500 	u64 v = (u64)khz * (1000000 + ppm);
1501 	do_div(v, 1000000);
1502 	return v;
1503 }
1504 
1505 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
1506 {
1507 	u64 ratio;
1508 
1509 	/* Guest TSC same frequency as host TSC? */
1510 	if (!scale) {
1511 		vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1512 		return 0;
1513 	}
1514 
1515 	/* TSC scaling supported? */
1516 	if (!kvm_has_tsc_control) {
1517 		if (user_tsc_khz > tsc_khz) {
1518 			vcpu->arch.tsc_catchup = 1;
1519 			vcpu->arch.tsc_always_catchup = 1;
1520 			return 0;
1521 		} else {
1522 			WARN(1, "user requested TSC rate below hardware speed\n");
1523 			return -1;
1524 		}
1525 	}
1526 
1527 	/* TSC scaling required  - calculate ratio */
1528 	ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
1529 				user_tsc_khz, tsc_khz);
1530 
1531 	if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
1532 		WARN_ONCE(1, "Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
1533 			  user_tsc_khz);
1534 		return -1;
1535 	}
1536 
1537 	vcpu->arch.tsc_scaling_ratio = ratio;
1538 	return 0;
1539 }
1540 
1541 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
1542 {
1543 	u32 thresh_lo, thresh_hi;
1544 	int use_scaling = 0;
1545 
1546 	/* tsc_khz can be zero if TSC calibration fails */
1547 	if (user_tsc_khz == 0) {
1548 		/* set tsc_scaling_ratio to a safe value */
1549 		vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1550 		return -1;
1551 	}
1552 
1553 	/* Compute a scale to convert nanoseconds in TSC cycles */
1554 	kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
1555 			   &vcpu->arch.virtual_tsc_shift,
1556 			   &vcpu->arch.virtual_tsc_mult);
1557 	vcpu->arch.virtual_tsc_khz = user_tsc_khz;
1558 
1559 	/*
1560 	 * Compute the variation in TSC rate which is acceptable
1561 	 * within the range of tolerance and decide if the
1562 	 * rate being applied is within that bounds of the hardware
1563 	 * rate.  If so, no scaling or compensation need be done.
1564 	 */
1565 	thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1566 	thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1567 	if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
1568 		pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
1569 		use_scaling = 1;
1570 	}
1571 	return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
1572 }
1573 
1574 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1575 {
1576 	u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1577 				      vcpu->arch.virtual_tsc_mult,
1578 				      vcpu->arch.virtual_tsc_shift);
1579 	tsc += vcpu->arch.this_tsc_write;
1580 	return tsc;
1581 }
1582 
1583 static inline int gtod_is_based_on_tsc(int mode)
1584 {
1585 	return mode == VCLOCK_TSC || mode == VCLOCK_HVCLOCK;
1586 }
1587 
1588 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1589 {
1590 #ifdef CONFIG_X86_64
1591 	bool vcpus_matched;
1592 	struct kvm_arch *ka = &vcpu->kvm->arch;
1593 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1594 
1595 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1596 			 atomic_read(&vcpu->kvm->online_vcpus));
1597 
1598 	/*
1599 	 * Once the masterclock is enabled, always perform request in
1600 	 * order to update it.
1601 	 *
1602 	 * In order to enable masterclock, the host clocksource must be TSC
1603 	 * and the vcpus need to have matched TSCs.  When that happens,
1604 	 * perform request to enable masterclock.
1605 	 */
1606 	if (ka->use_master_clock ||
1607 	    (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched))
1608 		kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1609 
1610 	trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1611 			    atomic_read(&vcpu->kvm->online_vcpus),
1612 		            ka->use_master_clock, gtod->clock.vclock_mode);
1613 #endif
1614 }
1615 
1616 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1617 {
1618 	u64 curr_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1619 	vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1620 }
1621 
1622 /*
1623  * Multiply tsc by a fixed point number represented by ratio.
1624  *
1625  * The most significant 64-N bits (mult) of ratio represent the
1626  * integral part of the fixed point number; the remaining N bits
1627  * (frac) represent the fractional part, ie. ratio represents a fixed
1628  * point number (mult + frac * 2^(-N)).
1629  *
1630  * N equals to kvm_tsc_scaling_ratio_frac_bits.
1631  */
1632 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
1633 {
1634 	return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
1635 }
1636 
1637 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
1638 {
1639 	u64 _tsc = tsc;
1640 	u64 ratio = vcpu->arch.tsc_scaling_ratio;
1641 
1642 	if (ratio != kvm_default_tsc_scaling_ratio)
1643 		_tsc = __scale_tsc(ratio, tsc);
1644 
1645 	return _tsc;
1646 }
1647 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
1648 
1649 static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1650 {
1651 	u64 tsc;
1652 
1653 	tsc = kvm_scale_tsc(vcpu, rdtsc());
1654 
1655 	return target_tsc - tsc;
1656 }
1657 
1658 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
1659 {
1660 	u64 tsc_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1661 
1662 	return tsc_offset + kvm_scale_tsc(vcpu, host_tsc);
1663 }
1664 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
1665 
1666 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1667 {
1668 	vcpu->arch.tsc_offset = kvm_x86_ops->write_l1_tsc_offset(vcpu, offset);
1669 }
1670 
1671 static inline bool kvm_check_tsc_unstable(void)
1672 {
1673 #ifdef CONFIG_X86_64
1674 	/*
1675 	 * TSC is marked unstable when we're running on Hyper-V,
1676 	 * 'TSC page' clocksource is good.
1677 	 */
1678 	if (pvclock_gtod_data.clock.vclock_mode == VCLOCK_HVCLOCK)
1679 		return false;
1680 #endif
1681 	return check_tsc_unstable();
1682 }
1683 
1684 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1685 {
1686 	struct kvm *kvm = vcpu->kvm;
1687 	u64 offset, ns, elapsed;
1688 	unsigned long flags;
1689 	bool matched;
1690 	bool already_matched;
1691 	u64 data = msr->data;
1692 	bool synchronizing = false;
1693 
1694 	raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1695 	offset = kvm_compute_tsc_offset(vcpu, data);
1696 	ns = ktime_get_boot_ns();
1697 	elapsed = ns - kvm->arch.last_tsc_nsec;
1698 
1699 	if (vcpu->arch.virtual_tsc_khz) {
1700 		if (data == 0 && msr->host_initiated) {
1701 			/*
1702 			 * detection of vcpu initialization -- need to sync
1703 			 * with other vCPUs. This particularly helps to keep
1704 			 * kvm_clock stable after CPU hotplug
1705 			 */
1706 			synchronizing = true;
1707 		} else {
1708 			u64 tsc_exp = kvm->arch.last_tsc_write +
1709 						nsec_to_cycles(vcpu, elapsed);
1710 			u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
1711 			/*
1712 			 * Special case: TSC write with a small delta (1 second)
1713 			 * of virtual cycle time against real time is
1714 			 * interpreted as an attempt to synchronize the CPU.
1715 			 */
1716 			synchronizing = data < tsc_exp + tsc_hz &&
1717 					data + tsc_hz > tsc_exp;
1718 		}
1719 	}
1720 
1721 	/*
1722 	 * For a reliable TSC, we can match TSC offsets, and for an unstable
1723 	 * TSC, we add elapsed time in this computation.  We could let the
1724 	 * compensation code attempt to catch up if we fall behind, but
1725 	 * it's better to try to match offsets from the beginning.
1726          */
1727 	if (synchronizing &&
1728 	    vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1729 		if (!kvm_check_tsc_unstable()) {
1730 			offset = kvm->arch.cur_tsc_offset;
1731 			pr_debug("kvm: matched tsc offset for %llu\n", data);
1732 		} else {
1733 			u64 delta = nsec_to_cycles(vcpu, elapsed);
1734 			data += delta;
1735 			offset = kvm_compute_tsc_offset(vcpu, data);
1736 			pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1737 		}
1738 		matched = true;
1739 		already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
1740 	} else {
1741 		/*
1742 		 * We split periods of matched TSC writes into generations.
1743 		 * For each generation, we track the original measured
1744 		 * nanosecond time, offset, and write, so if TSCs are in
1745 		 * sync, we can match exact offset, and if not, we can match
1746 		 * exact software computation in compute_guest_tsc()
1747 		 *
1748 		 * These values are tracked in kvm->arch.cur_xxx variables.
1749 		 */
1750 		kvm->arch.cur_tsc_generation++;
1751 		kvm->arch.cur_tsc_nsec = ns;
1752 		kvm->arch.cur_tsc_write = data;
1753 		kvm->arch.cur_tsc_offset = offset;
1754 		matched = false;
1755 		pr_debug("kvm: new tsc generation %llu, clock %llu\n",
1756 			 kvm->arch.cur_tsc_generation, data);
1757 	}
1758 
1759 	/*
1760 	 * We also track th most recent recorded KHZ, write and time to
1761 	 * allow the matching interval to be extended at each write.
1762 	 */
1763 	kvm->arch.last_tsc_nsec = ns;
1764 	kvm->arch.last_tsc_write = data;
1765 	kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1766 
1767 	vcpu->arch.last_guest_tsc = data;
1768 
1769 	/* Keep track of which generation this VCPU has synchronized to */
1770 	vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1771 	vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1772 	vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1773 
1774 	if (!msr->host_initiated && guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST))
1775 		update_ia32_tsc_adjust_msr(vcpu, offset);
1776 
1777 	kvm_vcpu_write_tsc_offset(vcpu, offset);
1778 	raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1779 
1780 	spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1781 	if (!matched) {
1782 		kvm->arch.nr_vcpus_matched_tsc = 0;
1783 	} else if (!already_matched) {
1784 		kvm->arch.nr_vcpus_matched_tsc++;
1785 	}
1786 
1787 	kvm_track_tsc_matching(vcpu);
1788 	spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
1789 }
1790 
1791 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1792 
1793 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
1794 					   s64 adjustment)
1795 {
1796 	u64 tsc_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1797 	kvm_vcpu_write_tsc_offset(vcpu, tsc_offset + adjustment);
1798 }
1799 
1800 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
1801 {
1802 	if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
1803 		WARN_ON(adjustment < 0);
1804 	adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
1805 	adjust_tsc_offset_guest(vcpu, adjustment);
1806 }
1807 
1808 #ifdef CONFIG_X86_64
1809 
1810 static u64 read_tsc(void)
1811 {
1812 	u64 ret = (u64)rdtsc_ordered();
1813 	u64 last = pvclock_gtod_data.clock.cycle_last;
1814 
1815 	if (likely(ret >= last))
1816 		return ret;
1817 
1818 	/*
1819 	 * GCC likes to generate cmov here, but this branch is extremely
1820 	 * predictable (it's just a function of time and the likely is
1821 	 * very likely) and there's a data dependence, so force GCC
1822 	 * to generate a branch instead.  I don't barrier() because
1823 	 * we don't actually need a barrier, and if this function
1824 	 * ever gets inlined it will generate worse code.
1825 	 */
1826 	asm volatile ("");
1827 	return last;
1828 }
1829 
1830 static inline u64 vgettsc(u64 *tsc_timestamp, int *mode)
1831 {
1832 	long v;
1833 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1834 	u64 tsc_pg_val;
1835 
1836 	switch (gtod->clock.vclock_mode) {
1837 	case VCLOCK_HVCLOCK:
1838 		tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(),
1839 						  tsc_timestamp);
1840 		if (tsc_pg_val != U64_MAX) {
1841 			/* TSC page valid */
1842 			*mode = VCLOCK_HVCLOCK;
1843 			v = (tsc_pg_val - gtod->clock.cycle_last) &
1844 				gtod->clock.mask;
1845 		} else {
1846 			/* TSC page invalid */
1847 			*mode = VCLOCK_NONE;
1848 		}
1849 		break;
1850 	case VCLOCK_TSC:
1851 		*mode = VCLOCK_TSC;
1852 		*tsc_timestamp = read_tsc();
1853 		v = (*tsc_timestamp - gtod->clock.cycle_last) &
1854 			gtod->clock.mask;
1855 		break;
1856 	default:
1857 		*mode = VCLOCK_NONE;
1858 	}
1859 
1860 	if (*mode == VCLOCK_NONE)
1861 		*tsc_timestamp = v = 0;
1862 
1863 	return v * gtod->clock.mult;
1864 }
1865 
1866 static int do_monotonic_boot(s64 *t, u64 *tsc_timestamp)
1867 {
1868 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1869 	unsigned long seq;
1870 	int mode;
1871 	u64 ns;
1872 
1873 	do {
1874 		seq = read_seqcount_begin(&gtod->seq);
1875 		ns = gtod->nsec_base;
1876 		ns += vgettsc(tsc_timestamp, &mode);
1877 		ns >>= gtod->clock.shift;
1878 		ns += gtod->boot_ns;
1879 	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
1880 	*t = ns;
1881 
1882 	return mode;
1883 }
1884 
1885 static int do_realtime(struct timespec64 *ts, u64 *tsc_timestamp)
1886 {
1887 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1888 	unsigned long seq;
1889 	int mode;
1890 	u64 ns;
1891 
1892 	do {
1893 		seq = read_seqcount_begin(&gtod->seq);
1894 		ts->tv_sec = gtod->wall_time_sec;
1895 		ns = gtod->nsec_base;
1896 		ns += vgettsc(tsc_timestamp, &mode);
1897 		ns >>= gtod->clock.shift;
1898 	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
1899 
1900 	ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
1901 	ts->tv_nsec = ns;
1902 
1903 	return mode;
1904 }
1905 
1906 /* returns true if host is using TSC based clocksource */
1907 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp)
1908 {
1909 	/* checked again under seqlock below */
1910 	if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
1911 		return false;
1912 
1913 	return gtod_is_based_on_tsc(do_monotonic_boot(kernel_ns,
1914 						      tsc_timestamp));
1915 }
1916 
1917 /* returns true if host is using TSC based clocksource */
1918 static bool kvm_get_walltime_and_clockread(struct timespec64 *ts,
1919 					   u64 *tsc_timestamp)
1920 {
1921 	/* checked again under seqlock below */
1922 	if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
1923 		return false;
1924 
1925 	return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp));
1926 }
1927 #endif
1928 
1929 /*
1930  *
1931  * Assuming a stable TSC across physical CPUS, and a stable TSC
1932  * across virtual CPUs, the following condition is possible.
1933  * Each numbered line represents an event visible to both
1934  * CPUs at the next numbered event.
1935  *
1936  * "timespecX" represents host monotonic time. "tscX" represents
1937  * RDTSC value.
1938  *
1939  * 		VCPU0 on CPU0		|	VCPU1 on CPU1
1940  *
1941  * 1.  read timespec0,tsc0
1942  * 2.					| timespec1 = timespec0 + N
1943  * 					| tsc1 = tsc0 + M
1944  * 3. transition to guest		| transition to guest
1945  * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1946  * 5.				        | ret1 = timespec1 + (rdtsc - tsc1)
1947  * 				        | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1948  *
1949  * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1950  *
1951  * 	- ret0 < ret1
1952  *	- timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1953  *		...
1954  *	- 0 < N - M => M < N
1955  *
1956  * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1957  * always the case (the difference between two distinct xtime instances
1958  * might be smaller then the difference between corresponding TSC reads,
1959  * when updating guest vcpus pvclock areas).
1960  *
1961  * To avoid that problem, do not allow visibility of distinct
1962  * system_timestamp/tsc_timestamp values simultaneously: use a master
1963  * copy of host monotonic time values. Update that master copy
1964  * in lockstep.
1965  *
1966  * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
1967  *
1968  */
1969 
1970 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
1971 {
1972 #ifdef CONFIG_X86_64
1973 	struct kvm_arch *ka = &kvm->arch;
1974 	int vclock_mode;
1975 	bool host_tsc_clocksource, vcpus_matched;
1976 
1977 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1978 			atomic_read(&kvm->online_vcpus));
1979 
1980 	/*
1981 	 * If the host uses TSC clock, then passthrough TSC as stable
1982 	 * to the guest.
1983 	 */
1984 	host_tsc_clocksource = kvm_get_time_and_clockread(
1985 					&ka->master_kernel_ns,
1986 					&ka->master_cycle_now);
1987 
1988 	ka->use_master_clock = host_tsc_clocksource && vcpus_matched
1989 				&& !ka->backwards_tsc_observed
1990 				&& !ka->boot_vcpu_runs_old_kvmclock;
1991 
1992 	if (ka->use_master_clock)
1993 		atomic_set(&kvm_guest_has_master_clock, 1);
1994 
1995 	vclock_mode = pvclock_gtod_data.clock.vclock_mode;
1996 	trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
1997 					vcpus_matched);
1998 #endif
1999 }
2000 
2001 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
2002 {
2003 	kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
2004 }
2005 
2006 static void kvm_gen_update_masterclock(struct kvm *kvm)
2007 {
2008 #ifdef CONFIG_X86_64
2009 	int i;
2010 	struct kvm_vcpu *vcpu;
2011 	struct kvm_arch *ka = &kvm->arch;
2012 
2013 	spin_lock(&ka->pvclock_gtod_sync_lock);
2014 	kvm_make_mclock_inprogress_request(kvm);
2015 	/* no guest entries from this point */
2016 	pvclock_update_vm_gtod_copy(kvm);
2017 
2018 	kvm_for_each_vcpu(i, vcpu, kvm)
2019 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2020 
2021 	/* guest entries allowed */
2022 	kvm_for_each_vcpu(i, vcpu, kvm)
2023 		kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
2024 
2025 	spin_unlock(&ka->pvclock_gtod_sync_lock);
2026 #endif
2027 }
2028 
2029 u64 get_kvmclock_ns(struct kvm *kvm)
2030 {
2031 	struct kvm_arch *ka = &kvm->arch;
2032 	struct pvclock_vcpu_time_info hv_clock;
2033 	u64 ret;
2034 
2035 	spin_lock(&ka->pvclock_gtod_sync_lock);
2036 	if (!ka->use_master_clock) {
2037 		spin_unlock(&ka->pvclock_gtod_sync_lock);
2038 		return ktime_get_boot_ns() + ka->kvmclock_offset;
2039 	}
2040 
2041 	hv_clock.tsc_timestamp = ka->master_cycle_now;
2042 	hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
2043 	spin_unlock(&ka->pvclock_gtod_sync_lock);
2044 
2045 	/* both __this_cpu_read() and rdtsc() should be on the same cpu */
2046 	get_cpu();
2047 
2048 	if (__this_cpu_read(cpu_tsc_khz)) {
2049 		kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
2050 				   &hv_clock.tsc_shift,
2051 				   &hv_clock.tsc_to_system_mul);
2052 		ret = __pvclock_read_cycles(&hv_clock, rdtsc());
2053 	} else
2054 		ret = ktime_get_boot_ns() + ka->kvmclock_offset;
2055 
2056 	put_cpu();
2057 
2058 	return ret;
2059 }
2060 
2061 static void kvm_setup_pvclock_page(struct kvm_vcpu *v)
2062 {
2063 	struct kvm_vcpu_arch *vcpu = &v->arch;
2064 	struct pvclock_vcpu_time_info guest_hv_clock;
2065 
2066 	if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
2067 		&guest_hv_clock, sizeof(guest_hv_clock))))
2068 		return;
2069 
2070 	/* This VCPU is paused, but it's legal for a guest to read another
2071 	 * VCPU's kvmclock, so we really have to follow the specification where
2072 	 * it says that version is odd if data is being modified, and even after
2073 	 * it is consistent.
2074 	 *
2075 	 * Version field updates must be kept separate.  This is because
2076 	 * kvm_write_guest_cached might use a "rep movs" instruction, and
2077 	 * writes within a string instruction are weakly ordered.  So there
2078 	 * are three writes overall.
2079 	 *
2080 	 * As a small optimization, only write the version field in the first
2081 	 * and third write.  The vcpu->pv_time cache is still valid, because the
2082 	 * version field is the first in the struct.
2083 	 */
2084 	BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
2085 
2086 	if (guest_hv_clock.version & 1)
2087 		++guest_hv_clock.version;  /* first time write, random junk */
2088 
2089 	vcpu->hv_clock.version = guest_hv_clock.version + 1;
2090 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2091 				&vcpu->hv_clock,
2092 				sizeof(vcpu->hv_clock.version));
2093 
2094 	smp_wmb();
2095 
2096 	/* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
2097 	vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
2098 
2099 	if (vcpu->pvclock_set_guest_stopped_request) {
2100 		vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
2101 		vcpu->pvclock_set_guest_stopped_request = false;
2102 	}
2103 
2104 	trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
2105 
2106 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2107 				&vcpu->hv_clock,
2108 				sizeof(vcpu->hv_clock));
2109 
2110 	smp_wmb();
2111 
2112 	vcpu->hv_clock.version++;
2113 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2114 				&vcpu->hv_clock,
2115 				sizeof(vcpu->hv_clock.version));
2116 }
2117 
2118 static int kvm_guest_time_update(struct kvm_vcpu *v)
2119 {
2120 	unsigned long flags, tgt_tsc_khz;
2121 	struct kvm_vcpu_arch *vcpu = &v->arch;
2122 	struct kvm_arch *ka = &v->kvm->arch;
2123 	s64 kernel_ns;
2124 	u64 tsc_timestamp, host_tsc;
2125 	u8 pvclock_flags;
2126 	bool use_master_clock;
2127 
2128 	kernel_ns = 0;
2129 	host_tsc = 0;
2130 
2131 	/*
2132 	 * If the host uses TSC clock, then passthrough TSC as stable
2133 	 * to the guest.
2134 	 */
2135 	spin_lock(&ka->pvclock_gtod_sync_lock);
2136 	use_master_clock = ka->use_master_clock;
2137 	if (use_master_clock) {
2138 		host_tsc = ka->master_cycle_now;
2139 		kernel_ns = ka->master_kernel_ns;
2140 	}
2141 	spin_unlock(&ka->pvclock_gtod_sync_lock);
2142 
2143 	/* Keep irq disabled to prevent changes to the clock */
2144 	local_irq_save(flags);
2145 	tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
2146 	if (unlikely(tgt_tsc_khz == 0)) {
2147 		local_irq_restore(flags);
2148 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2149 		return 1;
2150 	}
2151 	if (!use_master_clock) {
2152 		host_tsc = rdtsc();
2153 		kernel_ns = ktime_get_boot_ns();
2154 	}
2155 
2156 	tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
2157 
2158 	/*
2159 	 * We may have to catch up the TSC to match elapsed wall clock
2160 	 * time for two reasons, even if kvmclock is used.
2161 	 *   1) CPU could have been running below the maximum TSC rate
2162 	 *   2) Broken TSC compensation resets the base at each VCPU
2163 	 *      entry to avoid unknown leaps of TSC even when running
2164 	 *      again on the same CPU.  This may cause apparent elapsed
2165 	 *      time to disappear, and the guest to stand still or run
2166 	 *	very slowly.
2167 	 */
2168 	if (vcpu->tsc_catchup) {
2169 		u64 tsc = compute_guest_tsc(v, kernel_ns);
2170 		if (tsc > tsc_timestamp) {
2171 			adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
2172 			tsc_timestamp = tsc;
2173 		}
2174 	}
2175 
2176 	local_irq_restore(flags);
2177 
2178 	/* With all the info we got, fill in the values */
2179 
2180 	if (kvm_has_tsc_control)
2181 		tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
2182 
2183 	if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
2184 		kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
2185 				   &vcpu->hv_clock.tsc_shift,
2186 				   &vcpu->hv_clock.tsc_to_system_mul);
2187 		vcpu->hw_tsc_khz = tgt_tsc_khz;
2188 	}
2189 
2190 	vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
2191 	vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
2192 	vcpu->last_guest_tsc = tsc_timestamp;
2193 
2194 	/* If the host uses TSC clocksource, then it is stable */
2195 	pvclock_flags = 0;
2196 	if (use_master_clock)
2197 		pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
2198 
2199 	vcpu->hv_clock.flags = pvclock_flags;
2200 
2201 	if (vcpu->pv_time_enabled)
2202 		kvm_setup_pvclock_page(v);
2203 	if (v == kvm_get_vcpu(v->kvm, 0))
2204 		kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
2205 	return 0;
2206 }
2207 
2208 /*
2209  * kvmclock updates which are isolated to a given vcpu, such as
2210  * vcpu->cpu migration, should not allow system_timestamp from
2211  * the rest of the vcpus to remain static. Otherwise ntp frequency
2212  * correction applies to one vcpu's system_timestamp but not
2213  * the others.
2214  *
2215  * So in those cases, request a kvmclock update for all vcpus.
2216  * We need to rate-limit these requests though, as they can
2217  * considerably slow guests that have a large number of vcpus.
2218  * The time for a remote vcpu to update its kvmclock is bound
2219  * by the delay we use to rate-limit the updates.
2220  */
2221 
2222 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
2223 
2224 static void kvmclock_update_fn(struct work_struct *work)
2225 {
2226 	int i;
2227 	struct delayed_work *dwork = to_delayed_work(work);
2228 	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2229 					   kvmclock_update_work);
2230 	struct kvm *kvm = container_of(ka, struct kvm, arch);
2231 	struct kvm_vcpu *vcpu;
2232 
2233 	kvm_for_each_vcpu(i, vcpu, kvm) {
2234 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2235 		kvm_vcpu_kick(vcpu);
2236 	}
2237 }
2238 
2239 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
2240 {
2241 	struct kvm *kvm = v->kvm;
2242 
2243 	kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2244 	schedule_delayed_work(&kvm->arch.kvmclock_update_work,
2245 					KVMCLOCK_UPDATE_DELAY);
2246 }
2247 
2248 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
2249 
2250 static void kvmclock_sync_fn(struct work_struct *work)
2251 {
2252 	struct delayed_work *dwork = to_delayed_work(work);
2253 	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2254 					   kvmclock_sync_work);
2255 	struct kvm *kvm = container_of(ka, struct kvm, arch);
2256 
2257 	if (!kvmclock_periodic_sync)
2258 		return;
2259 
2260 	schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
2261 	schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
2262 					KVMCLOCK_SYNC_PERIOD);
2263 }
2264 
2265 static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2266 {
2267 	u64 mcg_cap = vcpu->arch.mcg_cap;
2268 	unsigned bank_num = mcg_cap & 0xff;
2269 	u32 msr = msr_info->index;
2270 	u64 data = msr_info->data;
2271 
2272 	switch (msr) {
2273 	case MSR_IA32_MCG_STATUS:
2274 		vcpu->arch.mcg_status = data;
2275 		break;
2276 	case MSR_IA32_MCG_CTL:
2277 		if (!(mcg_cap & MCG_CTL_P) &&
2278 		    (data || !msr_info->host_initiated))
2279 			return 1;
2280 		if (data != 0 && data != ~(u64)0)
2281 			return 1;
2282 		vcpu->arch.mcg_ctl = data;
2283 		break;
2284 	default:
2285 		if (msr >= MSR_IA32_MC0_CTL &&
2286 		    msr < MSR_IA32_MCx_CTL(bank_num)) {
2287 			u32 offset = msr - MSR_IA32_MC0_CTL;
2288 			/* only 0 or all 1s can be written to IA32_MCi_CTL
2289 			 * some Linux kernels though clear bit 10 in bank 4 to
2290 			 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
2291 			 * this to avoid an uncatched #GP in the guest
2292 			 */
2293 			if ((offset & 0x3) == 0 &&
2294 			    data != 0 && (data | (1 << 10)) != ~(u64)0)
2295 				return -1;
2296 			if (!msr_info->host_initiated &&
2297 				(offset & 0x3) == 1 && data != 0)
2298 				return -1;
2299 			vcpu->arch.mce_banks[offset] = data;
2300 			break;
2301 		}
2302 		return 1;
2303 	}
2304 	return 0;
2305 }
2306 
2307 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
2308 {
2309 	struct kvm *kvm = vcpu->kvm;
2310 	int lm = is_long_mode(vcpu);
2311 	u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
2312 		: (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
2313 	u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
2314 		: kvm->arch.xen_hvm_config.blob_size_32;
2315 	u32 page_num = data & ~PAGE_MASK;
2316 	u64 page_addr = data & PAGE_MASK;
2317 	u8 *page;
2318 	int r;
2319 
2320 	r = -E2BIG;
2321 	if (page_num >= blob_size)
2322 		goto out;
2323 	r = -ENOMEM;
2324 	page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
2325 	if (IS_ERR(page)) {
2326 		r = PTR_ERR(page);
2327 		goto out;
2328 	}
2329 	if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
2330 		goto out_free;
2331 	r = 0;
2332 out_free:
2333 	kfree(page);
2334 out:
2335 	return r;
2336 }
2337 
2338 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
2339 {
2340 	gpa_t gpa = data & ~0x3f;
2341 
2342 	/* Bits 3:5 are reserved, Should be zero */
2343 	if (data & 0x38)
2344 		return 1;
2345 
2346 	vcpu->arch.apf.msr_val = data;
2347 
2348 	if (!(data & KVM_ASYNC_PF_ENABLED)) {
2349 		kvm_clear_async_pf_completion_queue(vcpu);
2350 		kvm_async_pf_hash_reset(vcpu);
2351 		return 0;
2352 	}
2353 
2354 	if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
2355 					sizeof(u32)))
2356 		return 1;
2357 
2358 	vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
2359 	vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
2360 	kvm_async_pf_wakeup_all(vcpu);
2361 	return 0;
2362 }
2363 
2364 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2365 {
2366 	vcpu->arch.pv_time_enabled = false;
2367 }
2368 
2369 static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu, bool invalidate_gpa)
2370 {
2371 	++vcpu->stat.tlb_flush;
2372 	kvm_x86_ops->tlb_flush(vcpu, invalidate_gpa);
2373 }
2374 
2375 static void record_steal_time(struct kvm_vcpu *vcpu)
2376 {
2377 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2378 		return;
2379 
2380 	if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2381 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
2382 		return;
2383 
2384 	/*
2385 	 * Doing a TLB flush here, on the guest's behalf, can avoid
2386 	 * expensive IPIs.
2387 	 */
2388 	if (xchg(&vcpu->arch.st.steal.preempted, 0) & KVM_VCPU_FLUSH_TLB)
2389 		kvm_vcpu_flush_tlb(vcpu, false);
2390 
2391 	if (vcpu->arch.st.steal.version & 1)
2392 		vcpu->arch.st.steal.version += 1;  /* first time write, random junk */
2393 
2394 	vcpu->arch.st.steal.version += 1;
2395 
2396 	kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2397 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2398 
2399 	smp_wmb();
2400 
2401 	vcpu->arch.st.steal.steal += current->sched_info.run_delay -
2402 		vcpu->arch.st.last_steal;
2403 	vcpu->arch.st.last_steal = current->sched_info.run_delay;
2404 
2405 	kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2406 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2407 
2408 	smp_wmb();
2409 
2410 	vcpu->arch.st.steal.version += 1;
2411 
2412 	kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2413 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2414 }
2415 
2416 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2417 {
2418 	bool pr = false;
2419 	u32 msr = msr_info->index;
2420 	u64 data = msr_info->data;
2421 
2422 	switch (msr) {
2423 	case MSR_AMD64_NB_CFG:
2424 	case MSR_IA32_UCODE_WRITE:
2425 	case MSR_VM_HSAVE_PA:
2426 	case MSR_AMD64_PATCH_LOADER:
2427 	case MSR_AMD64_BU_CFG2:
2428 	case MSR_AMD64_DC_CFG:
2429 	case MSR_F15H_EX_CFG:
2430 		break;
2431 
2432 	case MSR_IA32_UCODE_REV:
2433 		if (msr_info->host_initiated)
2434 			vcpu->arch.microcode_version = data;
2435 		break;
2436 	case MSR_EFER:
2437 		return set_efer(vcpu, data);
2438 	case MSR_K7_HWCR:
2439 		data &= ~(u64)0x40;	/* ignore flush filter disable */
2440 		data &= ~(u64)0x100;	/* ignore ignne emulation enable */
2441 		data &= ~(u64)0x8;	/* ignore TLB cache disable */
2442 		data &= ~(u64)0x40000;  /* ignore Mc status write enable */
2443 		if (data != 0) {
2444 			vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2445 				    data);
2446 			return 1;
2447 		}
2448 		break;
2449 	case MSR_FAM10H_MMIO_CONF_BASE:
2450 		if (data != 0) {
2451 			vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2452 				    "0x%llx\n", data);
2453 			return 1;
2454 		}
2455 		break;
2456 	case MSR_IA32_DEBUGCTLMSR:
2457 		if (!data) {
2458 			/* We support the non-activated case already */
2459 			break;
2460 		} else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2461 			/* Values other than LBR and BTF are vendor-specific,
2462 			   thus reserved and should throw a #GP */
2463 			return 1;
2464 		}
2465 		vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2466 			    __func__, data);
2467 		break;
2468 	case 0x200 ... 0x2ff:
2469 		return kvm_mtrr_set_msr(vcpu, msr, data);
2470 	case MSR_IA32_APICBASE:
2471 		return kvm_set_apic_base(vcpu, msr_info);
2472 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2473 		return kvm_x2apic_msr_write(vcpu, msr, data);
2474 	case MSR_IA32_TSCDEADLINE:
2475 		kvm_set_lapic_tscdeadline_msr(vcpu, data);
2476 		break;
2477 	case MSR_IA32_TSC_ADJUST:
2478 		if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
2479 			if (!msr_info->host_initiated) {
2480 				s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2481 				adjust_tsc_offset_guest(vcpu, adj);
2482 			}
2483 			vcpu->arch.ia32_tsc_adjust_msr = data;
2484 		}
2485 		break;
2486 	case MSR_IA32_MISC_ENABLE:
2487 		vcpu->arch.ia32_misc_enable_msr = data;
2488 		break;
2489 	case MSR_IA32_SMBASE:
2490 		if (!msr_info->host_initiated)
2491 			return 1;
2492 		vcpu->arch.smbase = data;
2493 		break;
2494 	case MSR_IA32_TSC:
2495 		kvm_write_tsc(vcpu, msr_info);
2496 		break;
2497 	case MSR_SMI_COUNT:
2498 		if (!msr_info->host_initiated)
2499 			return 1;
2500 		vcpu->arch.smi_count = data;
2501 		break;
2502 	case MSR_KVM_WALL_CLOCK_NEW:
2503 	case MSR_KVM_WALL_CLOCK:
2504 		vcpu->kvm->arch.wall_clock = data;
2505 		kvm_write_wall_clock(vcpu->kvm, data);
2506 		break;
2507 	case MSR_KVM_SYSTEM_TIME_NEW:
2508 	case MSR_KVM_SYSTEM_TIME: {
2509 		struct kvm_arch *ka = &vcpu->kvm->arch;
2510 
2511 		kvmclock_reset(vcpu);
2512 
2513 		if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2514 			bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2515 
2516 			if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2517 				kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2518 
2519 			ka->boot_vcpu_runs_old_kvmclock = tmp;
2520 		}
2521 
2522 		vcpu->arch.time = data;
2523 		kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2524 
2525 		/* we verify if the enable bit is set... */
2526 		if (!(data & 1))
2527 			break;
2528 
2529 		if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2530 		     &vcpu->arch.pv_time, data & ~1ULL,
2531 		     sizeof(struct pvclock_vcpu_time_info)))
2532 			vcpu->arch.pv_time_enabled = false;
2533 		else
2534 			vcpu->arch.pv_time_enabled = true;
2535 
2536 		break;
2537 	}
2538 	case MSR_KVM_ASYNC_PF_EN:
2539 		if (kvm_pv_enable_async_pf(vcpu, data))
2540 			return 1;
2541 		break;
2542 	case MSR_KVM_STEAL_TIME:
2543 
2544 		if (unlikely(!sched_info_on()))
2545 			return 1;
2546 
2547 		if (data & KVM_STEAL_RESERVED_MASK)
2548 			return 1;
2549 
2550 		if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
2551 						data & KVM_STEAL_VALID_BITS,
2552 						sizeof(struct kvm_steal_time)))
2553 			return 1;
2554 
2555 		vcpu->arch.st.msr_val = data;
2556 
2557 		if (!(data & KVM_MSR_ENABLED))
2558 			break;
2559 
2560 		kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2561 
2562 		break;
2563 	case MSR_KVM_PV_EOI_EN:
2564 		if (kvm_lapic_enable_pv_eoi(vcpu, data, sizeof(u8)))
2565 			return 1;
2566 		break;
2567 
2568 	case MSR_IA32_MCG_CTL:
2569 	case MSR_IA32_MCG_STATUS:
2570 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2571 		return set_msr_mce(vcpu, msr_info);
2572 
2573 	case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2574 	case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2575 		pr = true; /* fall through */
2576 	case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2577 	case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2578 		if (kvm_pmu_is_valid_msr(vcpu, msr))
2579 			return kvm_pmu_set_msr(vcpu, msr_info);
2580 
2581 		if (pr || data != 0)
2582 			vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2583 				    "0x%x data 0x%llx\n", msr, data);
2584 		break;
2585 	case MSR_K7_CLK_CTL:
2586 		/*
2587 		 * Ignore all writes to this no longer documented MSR.
2588 		 * Writes are only relevant for old K7 processors,
2589 		 * all pre-dating SVM, but a recommended workaround from
2590 		 * AMD for these chips. It is possible to specify the
2591 		 * affected processor models on the command line, hence
2592 		 * the need to ignore the workaround.
2593 		 */
2594 		break;
2595 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2596 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2597 	case HV_X64_MSR_CRASH_CTL:
2598 	case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2599 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
2600 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
2601 	case HV_X64_MSR_TSC_EMULATION_STATUS:
2602 		return kvm_hv_set_msr_common(vcpu, msr, data,
2603 					     msr_info->host_initiated);
2604 	case MSR_IA32_BBL_CR_CTL3:
2605 		/* Drop writes to this legacy MSR -- see rdmsr
2606 		 * counterpart for further detail.
2607 		 */
2608 		if (report_ignored_msrs)
2609 			vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
2610 				msr, data);
2611 		break;
2612 	case MSR_AMD64_OSVW_ID_LENGTH:
2613 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2614 			return 1;
2615 		vcpu->arch.osvw.length = data;
2616 		break;
2617 	case MSR_AMD64_OSVW_STATUS:
2618 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2619 			return 1;
2620 		vcpu->arch.osvw.status = data;
2621 		break;
2622 	case MSR_PLATFORM_INFO:
2623 		if (!msr_info->host_initiated ||
2624 		    (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
2625 		     cpuid_fault_enabled(vcpu)))
2626 			return 1;
2627 		vcpu->arch.msr_platform_info = data;
2628 		break;
2629 	case MSR_MISC_FEATURES_ENABLES:
2630 		if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
2631 		    (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
2632 		     !supports_cpuid_fault(vcpu)))
2633 			return 1;
2634 		vcpu->arch.msr_misc_features_enables = data;
2635 		break;
2636 	default:
2637 		if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2638 			return xen_hvm_config(vcpu, data);
2639 		if (kvm_pmu_is_valid_msr(vcpu, msr))
2640 			return kvm_pmu_set_msr(vcpu, msr_info);
2641 		if (!ignore_msrs) {
2642 			vcpu_debug_ratelimited(vcpu, "unhandled wrmsr: 0x%x data 0x%llx\n",
2643 				    msr, data);
2644 			return 1;
2645 		} else {
2646 			if (report_ignored_msrs)
2647 				vcpu_unimpl(vcpu,
2648 					"ignored wrmsr: 0x%x data 0x%llx\n",
2649 					msr, data);
2650 			break;
2651 		}
2652 	}
2653 	return 0;
2654 }
2655 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2656 
2657 
2658 /*
2659  * Reads an msr value (of 'msr_index') into 'pdata'.
2660  * Returns 0 on success, non-0 otherwise.
2661  * Assumes vcpu_load() was already called.
2662  */
2663 int kvm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
2664 {
2665 	return kvm_x86_ops->get_msr(vcpu, msr);
2666 }
2667 EXPORT_SYMBOL_GPL(kvm_get_msr);
2668 
2669 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
2670 {
2671 	u64 data;
2672 	u64 mcg_cap = vcpu->arch.mcg_cap;
2673 	unsigned bank_num = mcg_cap & 0xff;
2674 
2675 	switch (msr) {
2676 	case MSR_IA32_P5_MC_ADDR:
2677 	case MSR_IA32_P5_MC_TYPE:
2678 		data = 0;
2679 		break;
2680 	case MSR_IA32_MCG_CAP:
2681 		data = vcpu->arch.mcg_cap;
2682 		break;
2683 	case MSR_IA32_MCG_CTL:
2684 		if (!(mcg_cap & MCG_CTL_P) && !host)
2685 			return 1;
2686 		data = vcpu->arch.mcg_ctl;
2687 		break;
2688 	case MSR_IA32_MCG_STATUS:
2689 		data = vcpu->arch.mcg_status;
2690 		break;
2691 	default:
2692 		if (msr >= MSR_IA32_MC0_CTL &&
2693 		    msr < MSR_IA32_MCx_CTL(bank_num)) {
2694 			u32 offset = msr - MSR_IA32_MC0_CTL;
2695 			data = vcpu->arch.mce_banks[offset];
2696 			break;
2697 		}
2698 		return 1;
2699 	}
2700 	*pdata = data;
2701 	return 0;
2702 }
2703 
2704 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2705 {
2706 	switch (msr_info->index) {
2707 	case MSR_IA32_PLATFORM_ID:
2708 	case MSR_IA32_EBL_CR_POWERON:
2709 	case MSR_IA32_DEBUGCTLMSR:
2710 	case MSR_IA32_LASTBRANCHFROMIP:
2711 	case MSR_IA32_LASTBRANCHTOIP:
2712 	case MSR_IA32_LASTINTFROMIP:
2713 	case MSR_IA32_LASTINTTOIP:
2714 	case MSR_K8_SYSCFG:
2715 	case MSR_K8_TSEG_ADDR:
2716 	case MSR_K8_TSEG_MASK:
2717 	case MSR_K7_HWCR:
2718 	case MSR_VM_HSAVE_PA:
2719 	case MSR_K8_INT_PENDING_MSG:
2720 	case MSR_AMD64_NB_CFG:
2721 	case MSR_FAM10H_MMIO_CONF_BASE:
2722 	case MSR_AMD64_BU_CFG2:
2723 	case MSR_IA32_PERF_CTL:
2724 	case MSR_AMD64_DC_CFG:
2725 	case MSR_F15H_EX_CFG:
2726 		msr_info->data = 0;
2727 		break;
2728 	case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
2729 	case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2730 	case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2731 	case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2732 	case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2733 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2734 			return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2735 		msr_info->data = 0;
2736 		break;
2737 	case MSR_IA32_UCODE_REV:
2738 		msr_info->data = vcpu->arch.microcode_version;
2739 		break;
2740 	case MSR_IA32_TSC:
2741 		msr_info->data = kvm_scale_tsc(vcpu, rdtsc()) + vcpu->arch.tsc_offset;
2742 		break;
2743 	case MSR_MTRRcap:
2744 	case 0x200 ... 0x2ff:
2745 		return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
2746 	case 0xcd: /* fsb frequency */
2747 		msr_info->data = 3;
2748 		break;
2749 		/*
2750 		 * MSR_EBC_FREQUENCY_ID
2751 		 * Conservative value valid for even the basic CPU models.
2752 		 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2753 		 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2754 		 * and 266MHz for model 3, or 4. Set Core Clock
2755 		 * Frequency to System Bus Frequency Ratio to 1 (bits
2756 		 * 31:24) even though these are only valid for CPU
2757 		 * models > 2, however guests may end up dividing or
2758 		 * multiplying by zero otherwise.
2759 		 */
2760 	case MSR_EBC_FREQUENCY_ID:
2761 		msr_info->data = 1 << 24;
2762 		break;
2763 	case MSR_IA32_APICBASE:
2764 		msr_info->data = kvm_get_apic_base(vcpu);
2765 		break;
2766 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2767 		return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
2768 		break;
2769 	case MSR_IA32_TSCDEADLINE:
2770 		msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
2771 		break;
2772 	case MSR_IA32_TSC_ADJUST:
2773 		msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
2774 		break;
2775 	case MSR_IA32_MISC_ENABLE:
2776 		msr_info->data = vcpu->arch.ia32_misc_enable_msr;
2777 		break;
2778 	case MSR_IA32_SMBASE:
2779 		if (!msr_info->host_initiated)
2780 			return 1;
2781 		msr_info->data = vcpu->arch.smbase;
2782 		break;
2783 	case MSR_SMI_COUNT:
2784 		msr_info->data = vcpu->arch.smi_count;
2785 		break;
2786 	case MSR_IA32_PERF_STATUS:
2787 		/* TSC increment by tick */
2788 		msr_info->data = 1000ULL;
2789 		/* CPU multiplier */
2790 		msr_info->data |= (((uint64_t)4ULL) << 40);
2791 		break;
2792 	case MSR_EFER:
2793 		msr_info->data = vcpu->arch.efer;
2794 		break;
2795 	case MSR_KVM_WALL_CLOCK:
2796 	case MSR_KVM_WALL_CLOCK_NEW:
2797 		msr_info->data = vcpu->kvm->arch.wall_clock;
2798 		break;
2799 	case MSR_KVM_SYSTEM_TIME:
2800 	case MSR_KVM_SYSTEM_TIME_NEW:
2801 		msr_info->data = vcpu->arch.time;
2802 		break;
2803 	case MSR_KVM_ASYNC_PF_EN:
2804 		msr_info->data = vcpu->arch.apf.msr_val;
2805 		break;
2806 	case MSR_KVM_STEAL_TIME:
2807 		msr_info->data = vcpu->arch.st.msr_val;
2808 		break;
2809 	case MSR_KVM_PV_EOI_EN:
2810 		msr_info->data = vcpu->arch.pv_eoi.msr_val;
2811 		break;
2812 	case MSR_IA32_P5_MC_ADDR:
2813 	case MSR_IA32_P5_MC_TYPE:
2814 	case MSR_IA32_MCG_CAP:
2815 	case MSR_IA32_MCG_CTL:
2816 	case MSR_IA32_MCG_STATUS:
2817 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2818 		return get_msr_mce(vcpu, msr_info->index, &msr_info->data,
2819 				   msr_info->host_initiated);
2820 	case MSR_K7_CLK_CTL:
2821 		/*
2822 		 * Provide expected ramp-up count for K7. All other
2823 		 * are set to zero, indicating minimum divisors for
2824 		 * every field.
2825 		 *
2826 		 * This prevents guest kernels on AMD host with CPU
2827 		 * type 6, model 8 and higher from exploding due to
2828 		 * the rdmsr failing.
2829 		 */
2830 		msr_info->data = 0x20000000;
2831 		break;
2832 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2833 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2834 	case HV_X64_MSR_CRASH_CTL:
2835 	case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2836 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
2837 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
2838 	case HV_X64_MSR_TSC_EMULATION_STATUS:
2839 		return kvm_hv_get_msr_common(vcpu,
2840 					     msr_info->index, &msr_info->data,
2841 					     msr_info->host_initiated);
2842 		break;
2843 	case MSR_IA32_BBL_CR_CTL3:
2844 		/* This legacy MSR exists but isn't fully documented in current
2845 		 * silicon.  It is however accessed by winxp in very narrow
2846 		 * scenarios where it sets bit #19, itself documented as
2847 		 * a "reserved" bit.  Best effort attempt to source coherent
2848 		 * read data here should the balance of the register be
2849 		 * interpreted by the guest:
2850 		 *
2851 		 * L2 cache control register 3: 64GB range, 256KB size,
2852 		 * enabled, latency 0x1, configured
2853 		 */
2854 		msr_info->data = 0xbe702111;
2855 		break;
2856 	case MSR_AMD64_OSVW_ID_LENGTH:
2857 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2858 			return 1;
2859 		msr_info->data = vcpu->arch.osvw.length;
2860 		break;
2861 	case MSR_AMD64_OSVW_STATUS:
2862 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2863 			return 1;
2864 		msr_info->data = vcpu->arch.osvw.status;
2865 		break;
2866 	case MSR_PLATFORM_INFO:
2867 		if (!msr_info->host_initiated &&
2868 		    !vcpu->kvm->arch.guest_can_read_msr_platform_info)
2869 			return 1;
2870 		msr_info->data = vcpu->arch.msr_platform_info;
2871 		break;
2872 	case MSR_MISC_FEATURES_ENABLES:
2873 		msr_info->data = vcpu->arch.msr_misc_features_enables;
2874 		break;
2875 	default:
2876 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2877 			return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2878 		if (!ignore_msrs) {
2879 			vcpu_debug_ratelimited(vcpu, "unhandled rdmsr: 0x%x\n",
2880 					       msr_info->index);
2881 			return 1;
2882 		} else {
2883 			if (report_ignored_msrs)
2884 				vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n",
2885 					msr_info->index);
2886 			msr_info->data = 0;
2887 		}
2888 		break;
2889 	}
2890 	return 0;
2891 }
2892 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2893 
2894 /*
2895  * Read or write a bunch of msrs. All parameters are kernel addresses.
2896  *
2897  * @return number of msrs set successfully.
2898  */
2899 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2900 		    struct kvm_msr_entry *entries,
2901 		    int (*do_msr)(struct kvm_vcpu *vcpu,
2902 				  unsigned index, u64 *data))
2903 {
2904 	int i;
2905 
2906 	for (i = 0; i < msrs->nmsrs; ++i)
2907 		if (do_msr(vcpu, entries[i].index, &entries[i].data))
2908 			break;
2909 
2910 	return i;
2911 }
2912 
2913 /*
2914  * Read or write a bunch of msrs. Parameters are user addresses.
2915  *
2916  * @return number of msrs set successfully.
2917  */
2918 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2919 		  int (*do_msr)(struct kvm_vcpu *vcpu,
2920 				unsigned index, u64 *data),
2921 		  int writeback)
2922 {
2923 	struct kvm_msrs msrs;
2924 	struct kvm_msr_entry *entries;
2925 	int r, n;
2926 	unsigned size;
2927 
2928 	r = -EFAULT;
2929 	if (copy_from_user(&msrs, user_msrs, sizeof(msrs)))
2930 		goto out;
2931 
2932 	r = -E2BIG;
2933 	if (msrs.nmsrs >= MAX_IO_MSRS)
2934 		goto out;
2935 
2936 	size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2937 	entries = memdup_user(user_msrs->entries, size);
2938 	if (IS_ERR(entries)) {
2939 		r = PTR_ERR(entries);
2940 		goto out;
2941 	}
2942 
2943 	r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2944 	if (r < 0)
2945 		goto out_free;
2946 
2947 	r = -EFAULT;
2948 	if (writeback && copy_to_user(user_msrs->entries, entries, size))
2949 		goto out_free;
2950 
2951 	r = n;
2952 
2953 out_free:
2954 	kfree(entries);
2955 out:
2956 	return r;
2957 }
2958 
2959 static inline bool kvm_can_mwait_in_guest(void)
2960 {
2961 	return boot_cpu_has(X86_FEATURE_MWAIT) &&
2962 		!boot_cpu_has_bug(X86_BUG_MONITOR) &&
2963 		boot_cpu_has(X86_FEATURE_ARAT);
2964 }
2965 
2966 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
2967 {
2968 	int r = 0;
2969 
2970 	switch (ext) {
2971 	case KVM_CAP_IRQCHIP:
2972 	case KVM_CAP_HLT:
2973 	case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2974 	case KVM_CAP_SET_TSS_ADDR:
2975 	case KVM_CAP_EXT_CPUID:
2976 	case KVM_CAP_EXT_EMUL_CPUID:
2977 	case KVM_CAP_CLOCKSOURCE:
2978 	case KVM_CAP_PIT:
2979 	case KVM_CAP_NOP_IO_DELAY:
2980 	case KVM_CAP_MP_STATE:
2981 	case KVM_CAP_SYNC_MMU:
2982 	case KVM_CAP_USER_NMI:
2983 	case KVM_CAP_REINJECT_CONTROL:
2984 	case KVM_CAP_IRQ_INJECT_STATUS:
2985 	case KVM_CAP_IOEVENTFD:
2986 	case KVM_CAP_IOEVENTFD_NO_LENGTH:
2987 	case KVM_CAP_PIT2:
2988 	case KVM_CAP_PIT_STATE2:
2989 	case KVM_CAP_SET_IDENTITY_MAP_ADDR:
2990 	case KVM_CAP_XEN_HVM:
2991 	case KVM_CAP_VCPU_EVENTS:
2992 	case KVM_CAP_HYPERV:
2993 	case KVM_CAP_HYPERV_VAPIC:
2994 	case KVM_CAP_HYPERV_SPIN:
2995 	case KVM_CAP_HYPERV_SYNIC:
2996 	case KVM_CAP_HYPERV_SYNIC2:
2997 	case KVM_CAP_HYPERV_VP_INDEX:
2998 	case KVM_CAP_HYPERV_EVENTFD:
2999 	case KVM_CAP_HYPERV_TLBFLUSH:
3000 	case KVM_CAP_HYPERV_SEND_IPI:
3001 	case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
3002 	case KVM_CAP_PCI_SEGMENT:
3003 	case KVM_CAP_DEBUGREGS:
3004 	case KVM_CAP_X86_ROBUST_SINGLESTEP:
3005 	case KVM_CAP_XSAVE:
3006 	case KVM_CAP_ASYNC_PF:
3007 	case KVM_CAP_GET_TSC_KHZ:
3008 	case KVM_CAP_KVMCLOCK_CTRL:
3009 	case KVM_CAP_READONLY_MEM:
3010 	case KVM_CAP_HYPERV_TIME:
3011 	case KVM_CAP_IOAPIC_POLARITY_IGNORED:
3012 	case KVM_CAP_TSC_DEADLINE_TIMER:
3013 	case KVM_CAP_ENABLE_CAP_VM:
3014 	case KVM_CAP_DISABLE_QUIRKS:
3015 	case KVM_CAP_SET_BOOT_CPU_ID:
3016  	case KVM_CAP_SPLIT_IRQCHIP:
3017 	case KVM_CAP_IMMEDIATE_EXIT:
3018 	case KVM_CAP_GET_MSR_FEATURES:
3019 	case KVM_CAP_MSR_PLATFORM_INFO:
3020 	case KVM_CAP_EXCEPTION_PAYLOAD:
3021 		r = 1;
3022 		break;
3023 	case KVM_CAP_SYNC_REGS:
3024 		r = KVM_SYNC_X86_VALID_FIELDS;
3025 		break;
3026 	case KVM_CAP_ADJUST_CLOCK:
3027 		r = KVM_CLOCK_TSC_STABLE;
3028 		break;
3029 	case KVM_CAP_X86_DISABLE_EXITS:
3030 		r |=  KVM_X86_DISABLE_EXITS_HLT | KVM_X86_DISABLE_EXITS_PAUSE;
3031 		if(kvm_can_mwait_in_guest())
3032 			r |= KVM_X86_DISABLE_EXITS_MWAIT;
3033 		break;
3034 	case KVM_CAP_X86_SMM:
3035 		/* SMBASE is usually relocated above 1M on modern chipsets,
3036 		 * and SMM handlers might indeed rely on 4G segment limits,
3037 		 * so do not report SMM to be available if real mode is
3038 		 * emulated via vm86 mode.  Still, do not go to great lengths
3039 		 * to avoid userspace's usage of the feature, because it is a
3040 		 * fringe case that is not enabled except via specific settings
3041 		 * of the module parameters.
3042 		 */
3043 		r = kvm_x86_ops->has_emulated_msr(MSR_IA32_SMBASE);
3044 		break;
3045 	case KVM_CAP_VAPIC:
3046 		r = !kvm_x86_ops->cpu_has_accelerated_tpr();
3047 		break;
3048 	case KVM_CAP_NR_VCPUS:
3049 		r = KVM_SOFT_MAX_VCPUS;
3050 		break;
3051 	case KVM_CAP_MAX_VCPUS:
3052 		r = KVM_MAX_VCPUS;
3053 		break;
3054 	case KVM_CAP_NR_MEMSLOTS:
3055 		r = KVM_USER_MEM_SLOTS;
3056 		break;
3057 	case KVM_CAP_PV_MMU:	/* obsolete */
3058 		r = 0;
3059 		break;
3060 	case KVM_CAP_MCE:
3061 		r = KVM_MAX_MCE_BANKS;
3062 		break;
3063 	case KVM_CAP_XCRS:
3064 		r = boot_cpu_has(X86_FEATURE_XSAVE);
3065 		break;
3066 	case KVM_CAP_TSC_CONTROL:
3067 		r = kvm_has_tsc_control;
3068 		break;
3069 	case KVM_CAP_X2APIC_API:
3070 		r = KVM_X2APIC_API_VALID_FLAGS;
3071 		break;
3072 	case KVM_CAP_NESTED_STATE:
3073 		r = kvm_x86_ops->get_nested_state ?
3074 			kvm_x86_ops->get_nested_state(NULL, 0, 0) : 0;
3075 		break;
3076 	default:
3077 		break;
3078 	}
3079 	return r;
3080 
3081 }
3082 
3083 long kvm_arch_dev_ioctl(struct file *filp,
3084 			unsigned int ioctl, unsigned long arg)
3085 {
3086 	void __user *argp = (void __user *)arg;
3087 	long r;
3088 
3089 	switch (ioctl) {
3090 	case KVM_GET_MSR_INDEX_LIST: {
3091 		struct kvm_msr_list __user *user_msr_list = argp;
3092 		struct kvm_msr_list msr_list;
3093 		unsigned n;
3094 
3095 		r = -EFAULT;
3096 		if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3097 			goto out;
3098 		n = msr_list.nmsrs;
3099 		msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
3100 		if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3101 			goto out;
3102 		r = -E2BIG;
3103 		if (n < msr_list.nmsrs)
3104 			goto out;
3105 		r = -EFAULT;
3106 		if (copy_to_user(user_msr_list->indices, &msrs_to_save,
3107 				 num_msrs_to_save * sizeof(u32)))
3108 			goto out;
3109 		if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
3110 				 &emulated_msrs,
3111 				 num_emulated_msrs * sizeof(u32)))
3112 			goto out;
3113 		r = 0;
3114 		break;
3115 	}
3116 	case KVM_GET_SUPPORTED_CPUID:
3117 	case KVM_GET_EMULATED_CPUID: {
3118 		struct kvm_cpuid2 __user *cpuid_arg = argp;
3119 		struct kvm_cpuid2 cpuid;
3120 
3121 		r = -EFAULT;
3122 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3123 			goto out;
3124 
3125 		r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
3126 					    ioctl);
3127 		if (r)
3128 			goto out;
3129 
3130 		r = -EFAULT;
3131 		if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
3132 			goto out;
3133 		r = 0;
3134 		break;
3135 	}
3136 	case KVM_X86_GET_MCE_CAP_SUPPORTED: {
3137 		r = -EFAULT;
3138 		if (copy_to_user(argp, &kvm_mce_cap_supported,
3139 				 sizeof(kvm_mce_cap_supported)))
3140 			goto out;
3141 		r = 0;
3142 		break;
3143 	case KVM_GET_MSR_FEATURE_INDEX_LIST: {
3144 		struct kvm_msr_list __user *user_msr_list = argp;
3145 		struct kvm_msr_list msr_list;
3146 		unsigned int n;
3147 
3148 		r = -EFAULT;
3149 		if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3150 			goto out;
3151 		n = msr_list.nmsrs;
3152 		msr_list.nmsrs = num_msr_based_features;
3153 		if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3154 			goto out;
3155 		r = -E2BIG;
3156 		if (n < msr_list.nmsrs)
3157 			goto out;
3158 		r = -EFAULT;
3159 		if (copy_to_user(user_msr_list->indices, &msr_based_features,
3160 				 num_msr_based_features * sizeof(u32)))
3161 			goto out;
3162 		r = 0;
3163 		break;
3164 	}
3165 	case KVM_GET_MSRS:
3166 		r = msr_io(NULL, argp, do_get_msr_feature, 1);
3167 		break;
3168 	}
3169 	default:
3170 		r = -EINVAL;
3171 	}
3172 out:
3173 	return r;
3174 }
3175 
3176 static void wbinvd_ipi(void *garbage)
3177 {
3178 	wbinvd();
3179 }
3180 
3181 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
3182 {
3183 	return kvm_arch_has_noncoherent_dma(vcpu->kvm);
3184 }
3185 
3186 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
3187 {
3188 	/* Address WBINVD may be executed by guest */
3189 	if (need_emulate_wbinvd(vcpu)) {
3190 		if (kvm_x86_ops->has_wbinvd_exit())
3191 			cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
3192 		else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
3193 			smp_call_function_single(vcpu->cpu,
3194 					wbinvd_ipi, NULL, 1);
3195 	}
3196 
3197 	kvm_x86_ops->vcpu_load(vcpu, cpu);
3198 
3199 	/* Apply any externally detected TSC adjustments (due to suspend) */
3200 	if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
3201 		adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
3202 		vcpu->arch.tsc_offset_adjustment = 0;
3203 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3204 	}
3205 
3206 	if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) {
3207 		s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
3208 				rdtsc() - vcpu->arch.last_host_tsc;
3209 		if (tsc_delta < 0)
3210 			mark_tsc_unstable("KVM discovered backwards TSC");
3211 
3212 		if (kvm_check_tsc_unstable()) {
3213 			u64 offset = kvm_compute_tsc_offset(vcpu,
3214 						vcpu->arch.last_guest_tsc);
3215 			kvm_vcpu_write_tsc_offset(vcpu, offset);
3216 			vcpu->arch.tsc_catchup = 1;
3217 		}
3218 
3219 		if (kvm_lapic_hv_timer_in_use(vcpu))
3220 			kvm_lapic_restart_hv_timer(vcpu);
3221 
3222 		/*
3223 		 * On a host with synchronized TSC, there is no need to update
3224 		 * kvmclock on vcpu->cpu migration
3225 		 */
3226 		if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
3227 			kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
3228 		if (vcpu->cpu != cpu)
3229 			kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
3230 		vcpu->cpu = cpu;
3231 	}
3232 
3233 	kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3234 }
3235 
3236 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
3237 {
3238 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
3239 		return;
3240 
3241 	vcpu->arch.st.steal.preempted = KVM_VCPU_PREEMPTED;
3242 
3243 	kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.st.stime,
3244 			&vcpu->arch.st.steal.preempted,
3245 			offsetof(struct kvm_steal_time, preempted),
3246 			sizeof(vcpu->arch.st.steal.preempted));
3247 }
3248 
3249 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
3250 {
3251 	int idx;
3252 
3253 	if (vcpu->preempted)
3254 		vcpu->arch.preempted_in_kernel = !kvm_x86_ops->get_cpl(vcpu);
3255 
3256 	/*
3257 	 * Disable page faults because we're in atomic context here.
3258 	 * kvm_write_guest_offset_cached() would call might_fault()
3259 	 * that relies on pagefault_disable() to tell if there's a
3260 	 * bug. NOTE: the write to guest memory may not go through if
3261 	 * during postcopy live migration or if there's heavy guest
3262 	 * paging.
3263 	 */
3264 	pagefault_disable();
3265 	/*
3266 	 * kvm_memslots() will be called by
3267 	 * kvm_write_guest_offset_cached() so take the srcu lock.
3268 	 */
3269 	idx = srcu_read_lock(&vcpu->kvm->srcu);
3270 	kvm_steal_time_set_preempted(vcpu);
3271 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
3272 	pagefault_enable();
3273 	kvm_x86_ops->vcpu_put(vcpu);
3274 	vcpu->arch.last_host_tsc = rdtsc();
3275 	/*
3276 	 * If userspace has set any breakpoints or watchpoints, dr6 is restored
3277 	 * on every vmexit, but if not, we might have a stale dr6 from the
3278 	 * guest. do_debug expects dr6 to be cleared after it runs, do the same.
3279 	 */
3280 	set_debugreg(0, 6);
3281 }
3282 
3283 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
3284 				    struct kvm_lapic_state *s)
3285 {
3286 	if (vcpu->arch.apicv_active)
3287 		kvm_x86_ops->sync_pir_to_irr(vcpu);
3288 
3289 	return kvm_apic_get_state(vcpu, s);
3290 }
3291 
3292 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
3293 				    struct kvm_lapic_state *s)
3294 {
3295 	int r;
3296 
3297 	r = kvm_apic_set_state(vcpu, s);
3298 	if (r)
3299 		return r;
3300 	update_cr8_intercept(vcpu);
3301 
3302 	return 0;
3303 }
3304 
3305 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
3306 {
3307 	return (!lapic_in_kernel(vcpu) ||
3308 		kvm_apic_accept_pic_intr(vcpu));
3309 }
3310 
3311 /*
3312  * if userspace requested an interrupt window, check that the
3313  * interrupt window is open.
3314  *
3315  * No need to exit to userspace if we already have an interrupt queued.
3316  */
3317 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
3318 {
3319 	return kvm_arch_interrupt_allowed(vcpu) &&
3320 		!kvm_cpu_has_interrupt(vcpu) &&
3321 		!kvm_event_needs_reinjection(vcpu) &&
3322 		kvm_cpu_accept_dm_intr(vcpu);
3323 }
3324 
3325 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
3326 				    struct kvm_interrupt *irq)
3327 {
3328 	if (irq->irq >= KVM_NR_INTERRUPTS)
3329 		return -EINVAL;
3330 
3331 	if (!irqchip_in_kernel(vcpu->kvm)) {
3332 		kvm_queue_interrupt(vcpu, irq->irq, false);
3333 		kvm_make_request(KVM_REQ_EVENT, vcpu);
3334 		return 0;
3335 	}
3336 
3337 	/*
3338 	 * With in-kernel LAPIC, we only use this to inject EXTINT, so
3339 	 * fail for in-kernel 8259.
3340 	 */
3341 	if (pic_in_kernel(vcpu->kvm))
3342 		return -ENXIO;
3343 
3344 	if (vcpu->arch.pending_external_vector != -1)
3345 		return -EEXIST;
3346 
3347 	vcpu->arch.pending_external_vector = irq->irq;
3348 	kvm_make_request(KVM_REQ_EVENT, vcpu);
3349 	return 0;
3350 }
3351 
3352 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
3353 {
3354 	kvm_inject_nmi(vcpu);
3355 
3356 	return 0;
3357 }
3358 
3359 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
3360 {
3361 	kvm_make_request(KVM_REQ_SMI, vcpu);
3362 
3363 	return 0;
3364 }
3365 
3366 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
3367 					   struct kvm_tpr_access_ctl *tac)
3368 {
3369 	if (tac->flags)
3370 		return -EINVAL;
3371 	vcpu->arch.tpr_access_reporting = !!tac->enabled;
3372 	return 0;
3373 }
3374 
3375 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
3376 					u64 mcg_cap)
3377 {
3378 	int r;
3379 	unsigned bank_num = mcg_cap & 0xff, bank;
3380 
3381 	r = -EINVAL;
3382 	if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
3383 		goto out;
3384 	if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
3385 		goto out;
3386 	r = 0;
3387 	vcpu->arch.mcg_cap = mcg_cap;
3388 	/* Init IA32_MCG_CTL to all 1s */
3389 	if (mcg_cap & MCG_CTL_P)
3390 		vcpu->arch.mcg_ctl = ~(u64)0;
3391 	/* Init IA32_MCi_CTL to all 1s */
3392 	for (bank = 0; bank < bank_num; bank++)
3393 		vcpu->arch.mce_banks[bank*4] = ~(u64)0;
3394 
3395 	if (kvm_x86_ops->setup_mce)
3396 		kvm_x86_ops->setup_mce(vcpu);
3397 out:
3398 	return r;
3399 }
3400 
3401 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
3402 				      struct kvm_x86_mce *mce)
3403 {
3404 	u64 mcg_cap = vcpu->arch.mcg_cap;
3405 	unsigned bank_num = mcg_cap & 0xff;
3406 	u64 *banks = vcpu->arch.mce_banks;
3407 
3408 	if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
3409 		return -EINVAL;
3410 	/*
3411 	 * if IA32_MCG_CTL is not all 1s, the uncorrected error
3412 	 * reporting is disabled
3413 	 */
3414 	if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
3415 	    vcpu->arch.mcg_ctl != ~(u64)0)
3416 		return 0;
3417 	banks += 4 * mce->bank;
3418 	/*
3419 	 * if IA32_MCi_CTL is not all 1s, the uncorrected error
3420 	 * reporting is disabled for the bank
3421 	 */
3422 	if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
3423 		return 0;
3424 	if (mce->status & MCI_STATUS_UC) {
3425 		if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
3426 		    !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
3427 			kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3428 			return 0;
3429 		}
3430 		if (banks[1] & MCI_STATUS_VAL)
3431 			mce->status |= MCI_STATUS_OVER;
3432 		banks[2] = mce->addr;
3433 		banks[3] = mce->misc;
3434 		vcpu->arch.mcg_status = mce->mcg_status;
3435 		banks[1] = mce->status;
3436 		kvm_queue_exception(vcpu, MC_VECTOR);
3437 	} else if (!(banks[1] & MCI_STATUS_VAL)
3438 		   || !(banks[1] & MCI_STATUS_UC)) {
3439 		if (banks[1] & MCI_STATUS_VAL)
3440 			mce->status |= MCI_STATUS_OVER;
3441 		banks[2] = mce->addr;
3442 		banks[3] = mce->misc;
3443 		banks[1] = mce->status;
3444 	} else
3445 		banks[1] |= MCI_STATUS_OVER;
3446 	return 0;
3447 }
3448 
3449 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
3450 					       struct kvm_vcpu_events *events)
3451 {
3452 	process_nmi(vcpu);
3453 
3454 	/*
3455 	 * The API doesn't provide the instruction length for software
3456 	 * exceptions, so don't report them. As long as the guest RIP
3457 	 * isn't advanced, we should expect to encounter the exception
3458 	 * again.
3459 	 */
3460 	if (kvm_exception_is_soft(vcpu->arch.exception.nr)) {
3461 		events->exception.injected = 0;
3462 		events->exception.pending = 0;
3463 	} else {
3464 		events->exception.injected = vcpu->arch.exception.injected;
3465 		events->exception.pending = vcpu->arch.exception.pending;
3466 		/*
3467 		 * For ABI compatibility, deliberately conflate
3468 		 * pending and injected exceptions when
3469 		 * KVM_CAP_EXCEPTION_PAYLOAD isn't enabled.
3470 		 */
3471 		if (!vcpu->kvm->arch.exception_payload_enabled)
3472 			events->exception.injected |=
3473 				vcpu->arch.exception.pending;
3474 	}
3475 	events->exception.nr = vcpu->arch.exception.nr;
3476 	events->exception.has_error_code = vcpu->arch.exception.has_error_code;
3477 	events->exception.error_code = vcpu->arch.exception.error_code;
3478 	events->exception_has_payload = vcpu->arch.exception.has_payload;
3479 	events->exception_payload = vcpu->arch.exception.payload;
3480 
3481 	events->interrupt.injected =
3482 		vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft;
3483 	events->interrupt.nr = vcpu->arch.interrupt.nr;
3484 	events->interrupt.soft = 0;
3485 	events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
3486 
3487 	events->nmi.injected = vcpu->arch.nmi_injected;
3488 	events->nmi.pending = vcpu->arch.nmi_pending != 0;
3489 	events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
3490 	events->nmi.pad = 0;
3491 
3492 	events->sipi_vector = 0; /* never valid when reporting to user space */
3493 
3494 	events->smi.smm = is_smm(vcpu);
3495 	events->smi.pending = vcpu->arch.smi_pending;
3496 	events->smi.smm_inside_nmi =
3497 		!!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
3498 	events->smi.latched_init = kvm_lapic_latched_init(vcpu);
3499 
3500 	events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
3501 			 | KVM_VCPUEVENT_VALID_SHADOW
3502 			 | KVM_VCPUEVENT_VALID_SMM);
3503 	if (vcpu->kvm->arch.exception_payload_enabled)
3504 		events->flags |= KVM_VCPUEVENT_VALID_PAYLOAD;
3505 
3506 	memset(&events->reserved, 0, sizeof(events->reserved));
3507 }
3508 
3509 static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags);
3510 
3511 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
3512 					      struct kvm_vcpu_events *events)
3513 {
3514 	if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
3515 			      | KVM_VCPUEVENT_VALID_SIPI_VECTOR
3516 			      | KVM_VCPUEVENT_VALID_SHADOW
3517 			      | KVM_VCPUEVENT_VALID_SMM
3518 			      | KVM_VCPUEVENT_VALID_PAYLOAD))
3519 		return -EINVAL;
3520 
3521 	if (events->flags & KVM_VCPUEVENT_VALID_PAYLOAD) {
3522 		if (!vcpu->kvm->arch.exception_payload_enabled)
3523 			return -EINVAL;
3524 		if (events->exception.pending)
3525 			events->exception.injected = 0;
3526 		else
3527 			events->exception_has_payload = 0;
3528 	} else {
3529 		events->exception.pending = 0;
3530 		events->exception_has_payload = 0;
3531 	}
3532 
3533 	if ((events->exception.injected || events->exception.pending) &&
3534 	    (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
3535 		return -EINVAL;
3536 
3537 	/* INITs are latched while in SMM */
3538 	if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
3539 	    (events->smi.smm || events->smi.pending) &&
3540 	    vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
3541 		return -EINVAL;
3542 
3543 	process_nmi(vcpu);
3544 	vcpu->arch.exception.injected = events->exception.injected;
3545 	vcpu->arch.exception.pending = events->exception.pending;
3546 	vcpu->arch.exception.nr = events->exception.nr;
3547 	vcpu->arch.exception.has_error_code = events->exception.has_error_code;
3548 	vcpu->arch.exception.error_code = events->exception.error_code;
3549 	vcpu->arch.exception.has_payload = events->exception_has_payload;
3550 	vcpu->arch.exception.payload = events->exception_payload;
3551 
3552 	vcpu->arch.interrupt.injected = events->interrupt.injected;
3553 	vcpu->arch.interrupt.nr = events->interrupt.nr;
3554 	vcpu->arch.interrupt.soft = events->interrupt.soft;
3555 	if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
3556 		kvm_x86_ops->set_interrupt_shadow(vcpu,
3557 						  events->interrupt.shadow);
3558 
3559 	vcpu->arch.nmi_injected = events->nmi.injected;
3560 	if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
3561 		vcpu->arch.nmi_pending = events->nmi.pending;
3562 	kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
3563 
3564 	if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
3565 	    lapic_in_kernel(vcpu))
3566 		vcpu->arch.apic->sipi_vector = events->sipi_vector;
3567 
3568 	if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
3569 		u32 hflags = vcpu->arch.hflags;
3570 		if (events->smi.smm)
3571 			hflags |= HF_SMM_MASK;
3572 		else
3573 			hflags &= ~HF_SMM_MASK;
3574 		kvm_set_hflags(vcpu, hflags);
3575 
3576 		vcpu->arch.smi_pending = events->smi.pending;
3577 
3578 		if (events->smi.smm) {
3579 			if (events->smi.smm_inside_nmi)
3580 				vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
3581 			else
3582 				vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
3583 			if (lapic_in_kernel(vcpu)) {
3584 				if (events->smi.latched_init)
3585 					set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3586 				else
3587 					clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3588 			}
3589 		}
3590 	}
3591 
3592 	kvm_make_request(KVM_REQ_EVENT, vcpu);
3593 
3594 	return 0;
3595 }
3596 
3597 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3598 					     struct kvm_debugregs *dbgregs)
3599 {
3600 	unsigned long val;
3601 
3602 	memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
3603 	kvm_get_dr(vcpu, 6, &val);
3604 	dbgregs->dr6 = val;
3605 	dbgregs->dr7 = vcpu->arch.dr7;
3606 	dbgregs->flags = 0;
3607 	memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
3608 }
3609 
3610 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
3611 					    struct kvm_debugregs *dbgregs)
3612 {
3613 	if (dbgregs->flags)
3614 		return -EINVAL;
3615 
3616 	if (dbgregs->dr6 & ~0xffffffffull)
3617 		return -EINVAL;
3618 	if (dbgregs->dr7 & ~0xffffffffull)
3619 		return -EINVAL;
3620 
3621 	memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
3622 	kvm_update_dr0123(vcpu);
3623 	vcpu->arch.dr6 = dbgregs->dr6;
3624 	kvm_update_dr6(vcpu);
3625 	vcpu->arch.dr7 = dbgregs->dr7;
3626 	kvm_update_dr7(vcpu);
3627 
3628 	return 0;
3629 }
3630 
3631 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
3632 
3633 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
3634 {
3635 	struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
3636 	u64 xstate_bv = xsave->header.xfeatures;
3637 	u64 valid;
3638 
3639 	/*
3640 	 * Copy legacy XSAVE area, to avoid complications with CPUID
3641 	 * leaves 0 and 1 in the loop below.
3642 	 */
3643 	memcpy(dest, xsave, XSAVE_HDR_OFFSET);
3644 
3645 	/* Set XSTATE_BV */
3646 	xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
3647 	*(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
3648 
3649 	/*
3650 	 * Copy each region from the possibly compacted offset to the
3651 	 * non-compacted offset.
3652 	 */
3653 	valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3654 	while (valid) {
3655 		u64 feature = valid & -valid;
3656 		int index = fls64(feature) - 1;
3657 		void *src = get_xsave_addr(xsave, feature);
3658 
3659 		if (src) {
3660 			u32 size, offset, ecx, edx;
3661 			cpuid_count(XSTATE_CPUID, index,
3662 				    &size, &offset, &ecx, &edx);
3663 			if (feature == XFEATURE_MASK_PKRU)
3664 				memcpy(dest + offset, &vcpu->arch.pkru,
3665 				       sizeof(vcpu->arch.pkru));
3666 			else
3667 				memcpy(dest + offset, src, size);
3668 
3669 		}
3670 
3671 		valid -= feature;
3672 	}
3673 }
3674 
3675 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
3676 {
3677 	struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
3678 	u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
3679 	u64 valid;
3680 
3681 	/*
3682 	 * Copy legacy XSAVE area, to avoid complications with CPUID
3683 	 * leaves 0 and 1 in the loop below.
3684 	 */
3685 	memcpy(xsave, src, XSAVE_HDR_OFFSET);
3686 
3687 	/* Set XSTATE_BV and possibly XCOMP_BV.  */
3688 	xsave->header.xfeatures = xstate_bv;
3689 	if (boot_cpu_has(X86_FEATURE_XSAVES))
3690 		xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
3691 
3692 	/*
3693 	 * Copy each region from the non-compacted offset to the
3694 	 * possibly compacted offset.
3695 	 */
3696 	valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3697 	while (valid) {
3698 		u64 feature = valid & -valid;
3699 		int index = fls64(feature) - 1;
3700 		void *dest = get_xsave_addr(xsave, feature);
3701 
3702 		if (dest) {
3703 			u32 size, offset, ecx, edx;
3704 			cpuid_count(XSTATE_CPUID, index,
3705 				    &size, &offset, &ecx, &edx);
3706 			if (feature == XFEATURE_MASK_PKRU)
3707 				memcpy(&vcpu->arch.pkru, src + offset,
3708 				       sizeof(vcpu->arch.pkru));
3709 			else
3710 				memcpy(dest, src + offset, size);
3711 		}
3712 
3713 		valid -= feature;
3714 	}
3715 }
3716 
3717 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
3718 					 struct kvm_xsave *guest_xsave)
3719 {
3720 	if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3721 		memset(guest_xsave, 0, sizeof(struct kvm_xsave));
3722 		fill_xsave((u8 *) guest_xsave->region, vcpu);
3723 	} else {
3724 		memcpy(guest_xsave->region,
3725 			&vcpu->arch.guest_fpu.state.fxsave,
3726 			sizeof(struct fxregs_state));
3727 		*(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
3728 			XFEATURE_MASK_FPSSE;
3729 	}
3730 }
3731 
3732 #define XSAVE_MXCSR_OFFSET 24
3733 
3734 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
3735 					struct kvm_xsave *guest_xsave)
3736 {
3737 	u64 xstate_bv =
3738 		*(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
3739 	u32 mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)];
3740 
3741 	if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3742 		/*
3743 		 * Here we allow setting states that are not present in
3744 		 * CPUID leaf 0xD, index 0, EDX:EAX.  This is for compatibility
3745 		 * with old userspace.
3746 		 */
3747 		if (xstate_bv & ~kvm_supported_xcr0() ||
3748 			mxcsr & ~mxcsr_feature_mask)
3749 			return -EINVAL;
3750 		load_xsave(vcpu, (u8 *)guest_xsave->region);
3751 	} else {
3752 		if (xstate_bv & ~XFEATURE_MASK_FPSSE ||
3753 			mxcsr & ~mxcsr_feature_mask)
3754 			return -EINVAL;
3755 		memcpy(&vcpu->arch.guest_fpu.state.fxsave,
3756 			guest_xsave->region, sizeof(struct fxregs_state));
3757 	}
3758 	return 0;
3759 }
3760 
3761 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3762 					struct kvm_xcrs *guest_xcrs)
3763 {
3764 	if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
3765 		guest_xcrs->nr_xcrs = 0;
3766 		return;
3767 	}
3768 
3769 	guest_xcrs->nr_xcrs = 1;
3770 	guest_xcrs->flags = 0;
3771 	guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3772 	guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3773 }
3774 
3775 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3776 				       struct kvm_xcrs *guest_xcrs)
3777 {
3778 	int i, r = 0;
3779 
3780 	if (!boot_cpu_has(X86_FEATURE_XSAVE))
3781 		return -EINVAL;
3782 
3783 	if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
3784 		return -EINVAL;
3785 
3786 	for (i = 0; i < guest_xcrs->nr_xcrs; i++)
3787 		/* Only support XCR0 currently */
3788 		if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
3789 			r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
3790 				guest_xcrs->xcrs[i].value);
3791 			break;
3792 		}
3793 	if (r)
3794 		r = -EINVAL;
3795 	return r;
3796 }
3797 
3798 /*
3799  * kvm_set_guest_paused() indicates to the guest kernel that it has been
3800  * stopped by the hypervisor.  This function will be called from the host only.
3801  * EINVAL is returned when the host attempts to set the flag for a guest that
3802  * does not support pv clocks.
3803  */
3804 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
3805 {
3806 	if (!vcpu->arch.pv_time_enabled)
3807 		return -EINVAL;
3808 	vcpu->arch.pvclock_set_guest_stopped_request = true;
3809 	kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3810 	return 0;
3811 }
3812 
3813 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
3814 				     struct kvm_enable_cap *cap)
3815 {
3816 	int r;
3817 	uint16_t vmcs_version;
3818 	void __user *user_ptr;
3819 
3820 	if (cap->flags)
3821 		return -EINVAL;
3822 
3823 	switch (cap->cap) {
3824 	case KVM_CAP_HYPERV_SYNIC2:
3825 		if (cap->args[0])
3826 			return -EINVAL;
3827 	case KVM_CAP_HYPERV_SYNIC:
3828 		if (!irqchip_in_kernel(vcpu->kvm))
3829 			return -EINVAL;
3830 		return kvm_hv_activate_synic(vcpu, cap->cap ==
3831 					     KVM_CAP_HYPERV_SYNIC2);
3832 	case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
3833 		r = kvm_x86_ops->nested_enable_evmcs(vcpu, &vmcs_version);
3834 		if (!r) {
3835 			user_ptr = (void __user *)(uintptr_t)cap->args[0];
3836 			if (copy_to_user(user_ptr, &vmcs_version,
3837 					 sizeof(vmcs_version)))
3838 				r = -EFAULT;
3839 		}
3840 		return r;
3841 
3842 	default:
3843 		return -EINVAL;
3844 	}
3845 }
3846 
3847 long kvm_arch_vcpu_ioctl(struct file *filp,
3848 			 unsigned int ioctl, unsigned long arg)
3849 {
3850 	struct kvm_vcpu *vcpu = filp->private_data;
3851 	void __user *argp = (void __user *)arg;
3852 	int r;
3853 	union {
3854 		struct kvm_lapic_state *lapic;
3855 		struct kvm_xsave *xsave;
3856 		struct kvm_xcrs *xcrs;
3857 		void *buffer;
3858 	} u;
3859 
3860 	vcpu_load(vcpu);
3861 
3862 	u.buffer = NULL;
3863 	switch (ioctl) {
3864 	case KVM_GET_LAPIC: {
3865 		r = -EINVAL;
3866 		if (!lapic_in_kernel(vcpu))
3867 			goto out;
3868 		u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
3869 
3870 		r = -ENOMEM;
3871 		if (!u.lapic)
3872 			goto out;
3873 		r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
3874 		if (r)
3875 			goto out;
3876 		r = -EFAULT;
3877 		if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
3878 			goto out;
3879 		r = 0;
3880 		break;
3881 	}
3882 	case KVM_SET_LAPIC: {
3883 		r = -EINVAL;
3884 		if (!lapic_in_kernel(vcpu))
3885 			goto out;
3886 		u.lapic = memdup_user(argp, sizeof(*u.lapic));
3887 		if (IS_ERR(u.lapic)) {
3888 			r = PTR_ERR(u.lapic);
3889 			goto out_nofree;
3890 		}
3891 
3892 		r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
3893 		break;
3894 	}
3895 	case KVM_INTERRUPT: {
3896 		struct kvm_interrupt irq;
3897 
3898 		r = -EFAULT;
3899 		if (copy_from_user(&irq, argp, sizeof(irq)))
3900 			goto out;
3901 		r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
3902 		break;
3903 	}
3904 	case KVM_NMI: {
3905 		r = kvm_vcpu_ioctl_nmi(vcpu);
3906 		break;
3907 	}
3908 	case KVM_SMI: {
3909 		r = kvm_vcpu_ioctl_smi(vcpu);
3910 		break;
3911 	}
3912 	case KVM_SET_CPUID: {
3913 		struct kvm_cpuid __user *cpuid_arg = argp;
3914 		struct kvm_cpuid cpuid;
3915 
3916 		r = -EFAULT;
3917 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3918 			goto out;
3919 		r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3920 		break;
3921 	}
3922 	case KVM_SET_CPUID2: {
3923 		struct kvm_cpuid2 __user *cpuid_arg = argp;
3924 		struct kvm_cpuid2 cpuid;
3925 
3926 		r = -EFAULT;
3927 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3928 			goto out;
3929 		r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
3930 					      cpuid_arg->entries);
3931 		break;
3932 	}
3933 	case KVM_GET_CPUID2: {
3934 		struct kvm_cpuid2 __user *cpuid_arg = argp;
3935 		struct kvm_cpuid2 cpuid;
3936 
3937 		r = -EFAULT;
3938 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3939 			goto out;
3940 		r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
3941 					      cpuid_arg->entries);
3942 		if (r)
3943 			goto out;
3944 		r = -EFAULT;
3945 		if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
3946 			goto out;
3947 		r = 0;
3948 		break;
3949 	}
3950 	case KVM_GET_MSRS: {
3951 		int idx = srcu_read_lock(&vcpu->kvm->srcu);
3952 		r = msr_io(vcpu, argp, do_get_msr, 1);
3953 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
3954 		break;
3955 	}
3956 	case KVM_SET_MSRS: {
3957 		int idx = srcu_read_lock(&vcpu->kvm->srcu);
3958 		r = msr_io(vcpu, argp, do_set_msr, 0);
3959 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
3960 		break;
3961 	}
3962 	case KVM_TPR_ACCESS_REPORTING: {
3963 		struct kvm_tpr_access_ctl tac;
3964 
3965 		r = -EFAULT;
3966 		if (copy_from_user(&tac, argp, sizeof(tac)))
3967 			goto out;
3968 		r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3969 		if (r)
3970 			goto out;
3971 		r = -EFAULT;
3972 		if (copy_to_user(argp, &tac, sizeof(tac)))
3973 			goto out;
3974 		r = 0;
3975 		break;
3976 	};
3977 	case KVM_SET_VAPIC_ADDR: {
3978 		struct kvm_vapic_addr va;
3979 		int idx;
3980 
3981 		r = -EINVAL;
3982 		if (!lapic_in_kernel(vcpu))
3983 			goto out;
3984 		r = -EFAULT;
3985 		if (copy_from_user(&va, argp, sizeof(va)))
3986 			goto out;
3987 		idx = srcu_read_lock(&vcpu->kvm->srcu);
3988 		r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
3989 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
3990 		break;
3991 	}
3992 	case KVM_X86_SETUP_MCE: {
3993 		u64 mcg_cap;
3994 
3995 		r = -EFAULT;
3996 		if (copy_from_user(&mcg_cap, argp, sizeof(mcg_cap)))
3997 			goto out;
3998 		r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3999 		break;
4000 	}
4001 	case KVM_X86_SET_MCE: {
4002 		struct kvm_x86_mce mce;
4003 
4004 		r = -EFAULT;
4005 		if (copy_from_user(&mce, argp, sizeof(mce)))
4006 			goto out;
4007 		r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
4008 		break;
4009 	}
4010 	case KVM_GET_VCPU_EVENTS: {
4011 		struct kvm_vcpu_events events;
4012 
4013 		kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
4014 
4015 		r = -EFAULT;
4016 		if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
4017 			break;
4018 		r = 0;
4019 		break;
4020 	}
4021 	case KVM_SET_VCPU_EVENTS: {
4022 		struct kvm_vcpu_events events;
4023 
4024 		r = -EFAULT;
4025 		if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
4026 			break;
4027 
4028 		r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
4029 		break;
4030 	}
4031 	case KVM_GET_DEBUGREGS: {
4032 		struct kvm_debugregs dbgregs;
4033 
4034 		kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
4035 
4036 		r = -EFAULT;
4037 		if (copy_to_user(argp, &dbgregs,
4038 				 sizeof(struct kvm_debugregs)))
4039 			break;
4040 		r = 0;
4041 		break;
4042 	}
4043 	case KVM_SET_DEBUGREGS: {
4044 		struct kvm_debugregs dbgregs;
4045 
4046 		r = -EFAULT;
4047 		if (copy_from_user(&dbgregs, argp,
4048 				   sizeof(struct kvm_debugregs)))
4049 			break;
4050 
4051 		r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
4052 		break;
4053 	}
4054 	case KVM_GET_XSAVE: {
4055 		u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
4056 		r = -ENOMEM;
4057 		if (!u.xsave)
4058 			break;
4059 
4060 		kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
4061 
4062 		r = -EFAULT;
4063 		if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
4064 			break;
4065 		r = 0;
4066 		break;
4067 	}
4068 	case KVM_SET_XSAVE: {
4069 		u.xsave = memdup_user(argp, sizeof(*u.xsave));
4070 		if (IS_ERR(u.xsave)) {
4071 			r = PTR_ERR(u.xsave);
4072 			goto out_nofree;
4073 		}
4074 
4075 		r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
4076 		break;
4077 	}
4078 	case KVM_GET_XCRS: {
4079 		u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
4080 		r = -ENOMEM;
4081 		if (!u.xcrs)
4082 			break;
4083 
4084 		kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
4085 
4086 		r = -EFAULT;
4087 		if (copy_to_user(argp, u.xcrs,
4088 				 sizeof(struct kvm_xcrs)))
4089 			break;
4090 		r = 0;
4091 		break;
4092 	}
4093 	case KVM_SET_XCRS: {
4094 		u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
4095 		if (IS_ERR(u.xcrs)) {
4096 			r = PTR_ERR(u.xcrs);
4097 			goto out_nofree;
4098 		}
4099 
4100 		r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
4101 		break;
4102 	}
4103 	case KVM_SET_TSC_KHZ: {
4104 		u32 user_tsc_khz;
4105 
4106 		r = -EINVAL;
4107 		user_tsc_khz = (u32)arg;
4108 
4109 		if (user_tsc_khz >= kvm_max_guest_tsc_khz)
4110 			goto out;
4111 
4112 		if (user_tsc_khz == 0)
4113 			user_tsc_khz = tsc_khz;
4114 
4115 		if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
4116 			r = 0;
4117 
4118 		goto out;
4119 	}
4120 	case KVM_GET_TSC_KHZ: {
4121 		r = vcpu->arch.virtual_tsc_khz;
4122 		goto out;
4123 	}
4124 	case KVM_KVMCLOCK_CTRL: {
4125 		r = kvm_set_guest_paused(vcpu);
4126 		goto out;
4127 	}
4128 	case KVM_ENABLE_CAP: {
4129 		struct kvm_enable_cap cap;
4130 
4131 		r = -EFAULT;
4132 		if (copy_from_user(&cap, argp, sizeof(cap)))
4133 			goto out;
4134 		r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
4135 		break;
4136 	}
4137 	case KVM_GET_NESTED_STATE: {
4138 		struct kvm_nested_state __user *user_kvm_nested_state = argp;
4139 		u32 user_data_size;
4140 
4141 		r = -EINVAL;
4142 		if (!kvm_x86_ops->get_nested_state)
4143 			break;
4144 
4145 		BUILD_BUG_ON(sizeof(user_data_size) != sizeof(user_kvm_nested_state->size));
4146 		r = -EFAULT;
4147 		if (get_user(user_data_size, &user_kvm_nested_state->size))
4148 			break;
4149 
4150 		r = kvm_x86_ops->get_nested_state(vcpu, user_kvm_nested_state,
4151 						  user_data_size);
4152 		if (r < 0)
4153 			break;
4154 
4155 		if (r > user_data_size) {
4156 			if (put_user(r, &user_kvm_nested_state->size))
4157 				r = -EFAULT;
4158 			else
4159 				r = -E2BIG;
4160 			break;
4161 		}
4162 
4163 		r = 0;
4164 		break;
4165 	}
4166 	case KVM_SET_NESTED_STATE: {
4167 		struct kvm_nested_state __user *user_kvm_nested_state = argp;
4168 		struct kvm_nested_state kvm_state;
4169 
4170 		r = -EINVAL;
4171 		if (!kvm_x86_ops->set_nested_state)
4172 			break;
4173 
4174 		r = -EFAULT;
4175 		if (copy_from_user(&kvm_state, user_kvm_nested_state, sizeof(kvm_state)))
4176 			break;
4177 
4178 		r = -EINVAL;
4179 		if (kvm_state.size < sizeof(kvm_state))
4180 			break;
4181 
4182 		if (kvm_state.flags &
4183 		    ~(KVM_STATE_NESTED_RUN_PENDING | KVM_STATE_NESTED_GUEST_MODE
4184 		      | KVM_STATE_NESTED_EVMCS))
4185 			break;
4186 
4187 		/* nested_run_pending implies guest_mode.  */
4188 		if ((kvm_state.flags & KVM_STATE_NESTED_RUN_PENDING)
4189 		    && !(kvm_state.flags & KVM_STATE_NESTED_GUEST_MODE))
4190 			break;
4191 
4192 		r = kvm_x86_ops->set_nested_state(vcpu, user_kvm_nested_state, &kvm_state);
4193 		break;
4194 	}
4195 	default:
4196 		r = -EINVAL;
4197 	}
4198 out:
4199 	kfree(u.buffer);
4200 out_nofree:
4201 	vcpu_put(vcpu);
4202 	return r;
4203 }
4204 
4205 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
4206 {
4207 	return VM_FAULT_SIGBUS;
4208 }
4209 
4210 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
4211 {
4212 	int ret;
4213 
4214 	if (addr > (unsigned int)(-3 * PAGE_SIZE))
4215 		return -EINVAL;
4216 	ret = kvm_x86_ops->set_tss_addr(kvm, addr);
4217 	return ret;
4218 }
4219 
4220 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
4221 					      u64 ident_addr)
4222 {
4223 	return kvm_x86_ops->set_identity_map_addr(kvm, ident_addr);
4224 }
4225 
4226 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
4227 					  u32 kvm_nr_mmu_pages)
4228 {
4229 	if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
4230 		return -EINVAL;
4231 
4232 	mutex_lock(&kvm->slots_lock);
4233 
4234 	kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
4235 	kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
4236 
4237 	mutex_unlock(&kvm->slots_lock);
4238 	return 0;
4239 }
4240 
4241 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
4242 {
4243 	return kvm->arch.n_max_mmu_pages;
4244 }
4245 
4246 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4247 {
4248 	struct kvm_pic *pic = kvm->arch.vpic;
4249 	int r;
4250 
4251 	r = 0;
4252 	switch (chip->chip_id) {
4253 	case KVM_IRQCHIP_PIC_MASTER:
4254 		memcpy(&chip->chip.pic, &pic->pics[0],
4255 			sizeof(struct kvm_pic_state));
4256 		break;
4257 	case KVM_IRQCHIP_PIC_SLAVE:
4258 		memcpy(&chip->chip.pic, &pic->pics[1],
4259 			sizeof(struct kvm_pic_state));
4260 		break;
4261 	case KVM_IRQCHIP_IOAPIC:
4262 		kvm_get_ioapic(kvm, &chip->chip.ioapic);
4263 		break;
4264 	default:
4265 		r = -EINVAL;
4266 		break;
4267 	}
4268 	return r;
4269 }
4270 
4271 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4272 {
4273 	struct kvm_pic *pic = kvm->arch.vpic;
4274 	int r;
4275 
4276 	r = 0;
4277 	switch (chip->chip_id) {
4278 	case KVM_IRQCHIP_PIC_MASTER:
4279 		spin_lock(&pic->lock);
4280 		memcpy(&pic->pics[0], &chip->chip.pic,
4281 			sizeof(struct kvm_pic_state));
4282 		spin_unlock(&pic->lock);
4283 		break;
4284 	case KVM_IRQCHIP_PIC_SLAVE:
4285 		spin_lock(&pic->lock);
4286 		memcpy(&pic->pics[1], &chip->chip.pic,
4287 			sizeof(struct kvm_pic_state));
4288 		spin_unlock(&pic->lock);
4289 		break;
4290 	case KVM_IRQCHIP_IOAPIC:
4291 		kvm_set_ioapic(kvm, &chip->chip.ioapic);
4292 		break;
4293 	default:
4294 		r = -EINVAL;
4295 		break;
4296 	}
4297 	kvm_pic_update_irq(pic);
4298 	return r;
4299 }
4300 
4301 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4302 {
4303 	struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
4304 
4305 	BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
4306 
4307 	mutex_lock(&kps->lock);
4308 	memcpy(ps, &kps->channels, sizeof(*ps));
4309 	mutex_unlock(&kps->lock);
4310 	return 0;
4311 }
4312 
4313 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4314 {
4315 	int i;
4316 	struct kvm_pit *pit = kvm->arch.vpit;
4317 
4318 	mutex_lock(&pit->pit_state.lock);
4319 	memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
4320 	for (i = 0; i < 3; i++)
4321 		kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
4322 	mutex_unlock(&pit->pit_state.lock);
4323 	return 0;
4324 }
4325 
4326 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4327 {
4328 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
4329 	memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
4330 		sizeof(ps->channels));
4331 	ps->flags = kvm->arch.vpit->pit_state.flags;
4332 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
4333 	memset(&ps->reserved, 0, sizeof(ps->reserved));
4334 	return 0;
4335 }
4336 
4337 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4338 {
4339 	int start = 0;
4340 	int i;
4341 	u32 prev_legacy, cur_legacy;
4342 	struct kvm_pit *pit = kvm->arch.vpit;
4343 
4344 	mutex_lock(&pit->pit_state.lock);
4345 	prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
4346 	cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
4347 	if (!prev_legacy && cur_legacy)
4348 		start = 1;
4349 	memcpy(&pit->pit_state.channels, &ps->channels,
4350 	       sizeof(pit->pit_state.channels));
4351 	pit->pit_state.flags = ps->flags;
4352 	for (i = 0; i < 3; i++)
4353 		kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
4354 				   start && i == 0);
4355 	mutex_unlock(&pit->pit_state.lock);
4356 	return 0;
4357 }
4358 
4359 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
4360 				 struct kvm_reinject_control *control)
4361 {
4362 	struct kvm_pit *pit = kvm->arch.vpit;
4363 
4364 	if (!pit)
4365 		return -ENXIO;
4366 
4367 	/* pit->pit_state.lock was overloaded to prevent userspace from getting
4368 	 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
4369 	 * ioctls in parallel.  Use a separate lock if that ioctl isn't rare.
4370 	 */
4371 	mutex_lock(&pit->pit_state.lock);
4372 	kvm_pit_set_reinject(pit, control->pit_reinject);
4373 	mutex_unlock(&pit->pit_state.lock);
4374 
4375 	return 0;
4376 }
4377 
4378 /**
4379  * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
4380  * @kvm: kvm instance
4381  * @log: slot id and address to which we copy the log
4382  *
4383  * Steps 1-4 below provide general overview of dirty page logging. See
4384  * kvm_get_dirty_log_protect() function description for additional details.
4385  *
4386  * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
4387  * always flush the TLB (step 4) even if previous step failed  and the dirty
4388  * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
4389  * does not preclude user space subsequent dirty log read. Flushing TLB ensures
4390  * writes will be marked dirty for next log read.
4391  *
4392  *   1. Take a snapshot of the bit and clear it if needed.
4393  *   2. Write protect the corresponding page.
4394  *   3. Copy the snapshot to the userspace.
4395  *   4. Flush TLB's if needed.
4396  */
4397 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
4398 {
4399 	bool is_dirty = false;
4400 	int r;
4401 
4402 	mutex_lock(&kvm->slots_lock);
4403 
4404 	/*
4405 	 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
4406 	 */
4407 	if (kvm_x86_ops->flush_log_dirty)
4408 		kvm_x86_ops->flush_log_dirty(kvm);
4409 
4410 	r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
4411 
4412 	/*
4413 	 * All the TLBs can be flushed out of mmu lock, see the comments in
4414 	 * kvm_mmu_slot_remove_write_access().
4415 	 */
4416 	lockdep_assert_held(&kvm->slots_lock);
4417 	if (is_dirty)
4418 		kvm_flush_remote_tlbs(kvm);
4419 
4420 	mutex_unlock(&kvm->slots_lock);
4421 	return r;
4422 }
4423 
4424 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
4425 			bool line_status)
4426 {
4427 	if (!irqchip_in_kernel(kvm))
4428 		return -ENXIO;
4429 
4430 	irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
4431 					irq_event->irq, irq_event->level,
4432 					line_status);
4433 	return 0;
4434 }
4435 
4436 static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
4437 				   struct kvm_enable_cap *cap)
4438 {
4439 	int r;
4440 
4441 	if (cap->flags)
4442 		return -EINVAL;
4443 
4444 	switch (cap->cap) {
4445 	case KVM_CAP_DISABLE_QUIRKS:
4446 		kvm->arch.disabled_quirks = cap->args[0];
4447 		r = 0;
4448 		break;
4449 	case KVM_CAP_SPLIT_IRQCHIP: {
4450 		mutex_lock(&kvm->lock);
4451 		r = -EINVAL;
4452 		if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
4453 			goto split_irqchip_unlock;
4454 		r = -EEXIST;
4455 		if (irqchip_in_kernel(kvm))
4456 			goto split_irqchip_unlock;
4457 		if (kvm->created_vcpus)
4458 			goto split_irqchip_unlock;
4459 		r = kvm_setup_empty_irq_routing(kvm);
4460 		if (r)
4461 			goto split_irqchip_unlock;
4462 		/* Pairs with irqchip_in_kernel. */
4463 		smp_wmb();
4464 		kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
4465 		kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
4466 		r = 0;
4467 split_irqchip_unlock:
4468 		mutex_unlock(&kvm->lock);
4469 		break;
4470 	}
4471 	case KVM_CAP_X2APIC_API:
4472 		r = -EINVAL;
4473 		if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
4474 			break;
4475 
4476 		if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
4477 			kvm->arch.x2apic_format = true;
4478 		if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
4479 			kvm->arch.x2apic_broadcast_quirk_disabled = true;
4480 
4481 		r = 0;
4482 		break;
4483 	case KVM_CAP_X86_DISABLE_EXITS:
4484 		r = -EINVAL;
4485 		if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
4486 			break;
4487 
4488 		if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
4489 			kvm_can_mwait_in_guest())
4490 			kvm->arch.mwait_in_guest = true;
4491 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_HLT)
4492 			kvm->arch.hlt_in_guest = true;
4493 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
4494 			kvm->arch.pause_in_guest = true;
4495 		r = 0;
4496 		break;
4497 	case KVM_CAP_MSR_PLATFORM_INFO:
4498 		kvm->arch.guest_can_read_msr_platform_info = cap->args[0];
4499 		r = 0;
4500 		break;
4501 	case KVM_CAP_EXCEPTION_PAYLOAD:
4502 		kvm->arch.exception_payload_enabled = cap->args[0];
4503 		r = 0;
4504 		break;
4505 	default:
4506 		r = -EINVAL;
4507 		break;
4508 	}
4509 	return r;
4510 }
4511 
4512 long kvm_arch_vm_ioctl(struct file *filp,
4513 		       unsigned int ioctl, unsigned long arg)
4514 {
4515 	struct kvm *kvm = filp->private_data;
4516 	void __user *argp = (void __user *)arg;
4517 	int r = -ENOTTY;
4518 	/*
4519 	 * This union makes it completely explicit to gcc-3.x
4520 	 * that these two variables' stack usage should be
4521 	 * combined, not added together.
4522 	 */
4523 	union {
4524 		struct kvm_pit_state ps;
4525 		struct kvm_pit_state2 ps2;
4526 		struct kvm_pit_config pit_config;
4527 	} u;
4528 
4529 	switch (ioctl) {
4530 	case KVM_SET_TSS_ADDR:
4531 		r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
4532 		break;
4533 	case KVM_SET_IDENTITY_MAP_ADDR: {
4534 		u64 ident_addr;
4535 
4536 		mutex_lock(&kvm->lock);
4537 		r = -EINVAL;
4538 		if (kvm->created_vcpus)
4539 			goto set_identity_unlock;
4540 		r = -EFAULT;
4541 		if (copy_from_user(&ident_addr, argp, sizeof(ident_addr)))
4542 			goto set_identity_unlock;
4543 		r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
4544 set_identity_unlock:
4545 		mutex_unlock(&kvm->lock);
4546 		break;
4547 	}
4548 	case KVM_SET_NR_MMU_PAGES:
4549 		r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
4550 		break;
4551 	case KVM_GET_NR_MMU_PAGES:
4552 		r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
4553 		break;
4554 	case KVM_CREATE_IRQCHIP: {
4555 		mutex_lock(&kvm->lock);
4556 
4557 		r = -EEXIST;
4558 		if (irqchip_in_kernel(kvm))
4559 			goto create_irqchip_unlock;
4560 
4561 		r = -EINVAL;
4562 		if (kvm->created_vcpus)
4563 			goto create_irqchip_unlock;
4564 
4565 		r = kvm_pic_init(kvm);
4566 		if (r)
4567 			goto create_irqchip_unlock;
4568 
4569 		r = kvm_ioapic_init(kvm);
4570 		if (r) {
4571 			kvm_pic_destroy(kvm);
4572 			goto create_irqchip_unlock;
4573 		}
4574 
4575 		r = kvm_setup_default_irq_routing(kvm);
4576 		if (r) {
4577 			kvm_ioapic_destroy(kvm);
4578 			kvm_pic_destroy(kvm);
4579 			goto create_irqchip_unlock;
4580 		}
4581 		/* Write kvm->irq_routing before enabling irqchip_in_kernel. */
4582 		smp_wmb();
4583 		kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
4584 	create_irqchip_unlock:
4585 		mutex_unlock(&kvm->lock);
4586 		break;
4587 	}
4588 	case KVM_CREATE_PIT:
4589 		u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
4590 		goto create_pit;
4591 	case KVM_CREATE_PIT2:
4592 		r = -EFAULT;
4593 		if (copy_from_user(&u.pit_config, argp,
4594 				   sizeof(struct kvm_pit_config)))
4595 			goto out;
4596 	create_pit:
4597 		mutex_lock(&kvm->lock);
4598 		r = -EEXIST;
4599 		if (kvm->arch.vpit)
4600 			goto create_pit_unlock;
4601 		r = -ENOMEM;
4602 		kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
4603 		if (kvm->arch.vpit)
4604 			r = 0;
4605 	create_pit_unlock:
4606 		mutex_unlock(&kvm->lock);
4607 		break;
4608 	case KVM_GET_IRQCHIP: {
4609 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4610 		struct kvm_irqchip *chip;
4611 
4612 		chip = memdup_user(argp, sizeof(*chip));
4613 		if (IS_ERR(chip)) {
4614 			r = PTR_ERR(chip);
4615 			goto out;
4616 		}
4617 
4618 		r = -ENXIO;
4619 		if (!irqchip_kernel(kvm))
4620 			goto get_irqchip_out;
4621 		r = kvm_vm_ioctl_get_irqchip(kvm, chip);
4622 		if (r)
4623 			goto get_irqchip_out;
4624 		r = -EFAULT;
4625 		if (copy_to_user(argp, chip, sizeof(*chip)))
4626 			goto get_irqchip_out;
4627 		r = 0;
4628 	get_irqchip_out:
4629 		kfree(chip);
4630 		break;
4631 	}
4632 	case KVM_SET_IRQCHIP: {
4633 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4634 		struct kvm_irqchip *chip;
4635 
4636 		chip = memdup_user(argp, sizeof(*chip));
4637 		if (IS_ERR(chip)) {
4638 			r = PTR_ERR(chip);
4639 			goto out;
4640 		}
4641 
4642 		r = -ENXIO;
4643 		if (!irqchip_kernel(kvm))
4644 			goto set_irqchip_out;
4645 		r = kvm_vm_ioctl_set_irqchip(kvm, chip);
4646 		if (r)
4647 			goto set_irqchip_out;
4648 		r = 0;
4649 	set_irqchip_out:
4650 		kfree(chip);
4651 		break;
4652 	}
4653 	case KVM_GET_PIT: {
4654 		r = -EFAULT;
4655 		if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
4656 			goto out;
4657 		r = -ENXIO;
4658 		if (!kvm->arch.vpit)
4659 			goto out;
4660 		r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
4661 		if (r)
4662 			goto out;
4663 		r = -EFAULT;
4664 		if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
4665 			goto out;
4666 		r = 0;
4667 		break;
4668 	}
4669 	case KVM_SET_PIT: {
4670 		r = -EFAULT;
4671 		if (copy_from_user(&u.ps, argp, sizeof(u.ps)))
4672 			goto out;
4673 		r = -ENXIO;
4674 		if (!kvm->arch.vpit)
4675 			goto out;
4676 		r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
4677 		break;
4678 	}
4679 	case KVM_GET_PIT2: {
4680 		r = -ENXIO;
4681 		if (!kvm->arch.vpit)
4682 			goto out;
4683 		r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
4684 		if (r)
4685 			goto out;
4686 		r = -EFAULT;
4687 		if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
4688 			goto out;
4689 		r = 0;
4690 		break;
4691 	}
4692 	case KVM_SET_PIT2: {
4693 		r = -EFAULT;
4694 		if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
4695 			goto out;
4696 		r = -ENXIO;
4697 		if (!kvm->arch.vpit)
4698 			goto out;
4699 		r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
4700 		break;
4701 	}
4702 	case KVM_REINJECT_CONTROL: {
4703 		struct kvm_reinject_control control;
4704 		r =  -EFAULT;
4705 		if (copy_from_user(&control, argp, sizeof(control)))
4706 			goto out;
4707 		r = kvm_vm_ioctl_reinject(kvm, &control);
4708 		break;
4709 	}
4710 	case KVM_SET_BOOT_CPU_ID:
4711 		r = 0;
4712 		mutex_lock(&kvm->lock);
4713 		if (kvm->created_vcpus)
4714 			r = -EBUSY;
4715 		else
4716 			kvm->arch.bsp_vcpu_id = arg;
4717 		mutex_unlock(&kvm->lock);
4718 		break;
4719 	case KVM_XEN_HVM_CONFIG: {
4720 		struct kvm_xen_hvm_config xhc;
4721 		r = -EFAULT;
4722 		if (copy_from_user(&xhc, argp, sizeof(xhc)))
4723 			goto out;
4724 		r = -EINVAL;
4725 		if (xhc.flags)
4726 			goto out;
4727 		memcpy(&kvm->arch.xen_hvm_config, &xhc, sizeof(xhc));
4728 		r = 0;
4729 		break;
4730 	}
4731 	case KVM_SET_CLOCK: {
4732 		struct kvm_clock_data user_ns;
4733 		u64 now_ns;
4734 
4735 		r = -EFAULT;
4736 		if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
4737 			goto out;
4738 
4739 		r = -EINVAL;
4740 		if (user_ns.flags)
4741 			goto out;
4742 
4743 		r = 0;
4744 		/*
4745 		 * TODO: userspace has to take care of races with VCPU_RUN, so
4746 		 * kvm_gen_update_masterclock() can be cut down to locked
4747 		 * pvclock_update_vm_gtod_copy().
4748 		 */
4749 		kvm_gen_update_masterclock(kvm);
4750 		now_ns = get_kvmclock_ns(kvm);
4751 		kvm->arch.kvmclock_offset += user_ns.clock - now_ns;
4752 		kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE);
4753 		break;
4754 	}
4755 	case KVM_GET_CLOCK: {
4756 		struct kvm_clock_data user_ns;
4757 		u64 now_ns;
4758 
4759 		now_ns = get_kvmclock_ns(kvm);
4760 		user_ns.clock = now_ns;
4761 		user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
4762 		memset(&user_ns.pad, 0, sizeof(user_ns.pad));
4763 
4764 		r = -EFAULT;
4765 		if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
4766 			goto out;
4767 		r = 0;
4768 		break;
4769 	}
4770 	case KVM_ENABLE_CAP: {
4771 		struct kvm_enable_cap cap;
4772 
4773 		r = -EFAULT;
4774 		if (copy_from_user(&cap, argp, sizeof(cap)))
4775 			goto out;
4776 		r = kvm_vm_ioctl_enable_cap(kvm, &cap);
4777 		break;
4778 	}
4779 	case KVM_MEMORY_ENCRYPT_OP: {
4780 		r = -ENOTTY;
4781 		if (kvm_x86_ops->mem_enc_op)
4782 			r = kvm_x86_ops->mem_enc_op(kvm, argp);
4783 		break;
4784 	}
4785 	case KVM_MEMORY_ENCRYPT_REG_REGION: {
4786 		struct kvm_enc_region region;
4787 
4788 		r = -EFAULT;
4789 		if (copy_from_user(&region, argp, sizeof(region)))
4790 			goto out;
4791 
4792 		r = -ENOTTY;
4793 		if (kvm_x86_ops->mem_enc_reg_region)
4794 			r = kvm_x86_ops->mem_enc_reg_region(kvm, &region);
4795 		break;
4796 	}
4797 	case KVM_MEMORY_ENCRYPT_UNREG_REGION: {
4798 		struct kvm_enc_region region;
4799 
4800 		r = -EFAULT;
4801 		if (copy_from_user(&region, argp, sizeof(region)))
4802 			goto out;
4803 
4804 		r = -ENOTTY;
4805 		if (kvm_x86_ops->mem_enc_unreg_region)
4806 			r = kvm_x86_ops->mem_enc_unreg_region(kvm, &region);
4807 		break;
4808 	}
4809 	case KVM_HYPERV_EVENTFD: {
4810 		struct kvm_hyperv_eventfd hvevfd;
4811 
4812 		r = -EFAULT;
4813 		if (copy_from_user(&hvevfd, argp, sizeof(hvevfd)))
4814 			goto out;
4815 		r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd);
4816 		break;
4817 	}
4818 	default:
4819 		r = -ENOTTY;
4820 	}
4821 out:
4822 	return r;
4823 }
4824 
4825 static void kvm_init_msr_list(void)
4826 {
4827 	u32 dummy[2];
4828 	unsigned i, j;
4829 
4830 	for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
4831 		if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
4832 			continue;
4833 
4834 		/*
4835 		 * Even MSRs that are valid in the host may not be exposed
4836 		 * to the guests in some cases.
4837 		 */
4838 		switch (msrs_to_save[i]) {
4839 		case MSR_IA32_BNDCFGS:
4840 			if (!kvm_mpx_supported())
4841 				continue;
4842 			break;
4843 		case MSR_TSC_AUX:
4844 			if (!kvm_x86_ops->rdtscp_supported())
4845 				continue;
4846 			break;
4847 		default:
4848 			break;
4849 		}
4850 
4851 		if (j < i)
4852 			msrs_to_save[j] = msrs_to_save[i];
4853 		j++;
4854 	}
4855 	num_msrs_to_save = j;
4856 
4857 	for (i = j = 0; i < ARRAY_SIZE(emulated_msrs); i++) {
4858 		if (!kvm_x86_ops->has_emulated_msr(emulated_msrs[i]))
4859 			continue;
4860 
4861 		if (j < i)
4862 			emulated_msrs[j] = emulated_msrs[i];
4863 		j++;
4864 	}
4865 	num_emulated_msrs = j;
4866 
4867 	for (i = j = 0; i < ARRAY_SIZE(msr_based_features); i++) {
4868 		struct kvm_msr_entry msr;
4869 
4870 		msr.index = msr_based_features[i];
4871 		if (kvm_get_msr_feature(&msr))
4872 			continue;
4873 
4874 		if (j < i)
4875 			msr_based_features[j] = msr_based_features[i];
4876 		j++;
4877 	}
4878 	num_msr_based_features = j;
4879 }
4880 
4881 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
4882 			   const void *v)
4883 {
4884 	int handled = 0;
4885 	int n;
4886 
4887 	do {
4888 		n = min(len, 8);
4889 		if (!(lapic_in_kernel(vcpu) &&
4890 		      !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
4891 		    && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
4892 			break;
4893 		handled += n;
4894 		addr += n;
4895 		len -= n;
4896 		v += n;
4897 	} while (len);
4898 
4899 	return handled;
4900 }
4901 
4902 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
4903 {
4904 	int handled = 0;
4905 	int n;
4906 
4907 	do {
4908 		n = min(len, 8);
4909 		if (!(lapic_in_kernel(vcpu) &&
4910 		      !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
4911 					 addr, n, v))
4912 		    && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
4913 			break;
4914 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
4915 		handled += n;
4916 		addr += n;
4917 		len -= n;
4918 		v += n;
4919 	} while (len);
4920 
4921 	return handled;
4922 }
4923 
4924 static void kvm_set_segment(struct kvm_vcpu *vcpu,
4925 			struct kvm_segment *var, int seg)
4926 {
4927 	kvm_x86_ops->set_segment(vcpu, var, seg);
4928 }
4929 
4930 void kvm_get_segment(struct kvm_vcpu *vcpu,
4931 		     struct kvm_segment *var, int seg)
4932 {
4933 	kvm_x86_ops->get_segment(vcpu, var, seg);
4934 }
4935 
4936 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
4937 			   struct x86_exception *exception)
4938 {
4939 	gpa_t t_gpa;
4940 
4941 	BUG_ON(!mmu_is_nested(vcpu));
4942 
4943 	/* NPT walks are always user-walks */
4944 	access |= PFERR_USER_MASK;
4945 	t_gpa  = vcpu->arch.mmu->gva_to_gpa(vcpu, gpa, access, exception);
4946 
4947 	return t_gpa;
4948 }
4949 
4950 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
4951 			      struct x86_exception *exception)
4952 {
4953 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4954 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4955 }
4956 
4957  gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
4958 				struct x86_exception *exception)
4959 {
4960 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4961 	access |= PFERR_FETCH_MASK;
4962 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4963 }
4964 
4965 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
4966 			       struct x86_exception *exception)
4967 {
4968 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4969 	access |= PFERR_WRITE_MASK;
4970 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4971 }
4972 
4973 /* uses this to access any guest's mapped memory without checking CPL */
4974 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
4975 				struct x86_exception *exception)
4976 {
4977 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
4978 }
4979 
4980 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
4981 				      struct kvm_vcpu *vcpu, u32 access,
4982 				      struct x86_exception *exception)
4983 {
4984 	void *data = val;
4985 	int r = X86EMUL_CONTINUE;
4986 
4987 	while (bytes) {
4988 		gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
4989 							    exception);
4990 		unsigned offset = addr & (PAGE_SIZE-1);
4991 		unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
4992 		int ret;
4993 
4994 		if (gpa == UNMAPPED_GVA)
4995 			return X86EMUL_PROPAGATE_FAULT;
4996 		ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
4997 					       offset, toread);
4998 		if (ret < 0) {
4999 			r = X86EMUL_IO_NEEDED;
5000 			goto out;
5001 		}
5002 
5003 		bytes -= toread;
5004 		data += toread;
5005 		addr += toread;
5006 	}
5007 out:
5008 	return r;
5009 }
5010 
5011 /* used for instruction fetching */
5012 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
5013 				gva_t addr, void *val, unsigned int bytes,
5014 				struct x86_exception *exception)
5015 {
5016 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5017 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5018 	unsigned offset;
5019 	int ret;
5020 
5021 	/* Inline kvm_read_guest_virt_helper for speed.  */
5022 	gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
5023 						    exception);
5024 	if (unlikely(gpa == UNMAPPED_GVA))
5025 		return X86EMUL_PROPAGATE_FAULT;
5026 
5027 	offset = addr & (PAGE_SIZE-1);
5028 	if (WARN_ON(offset + bytes > PAGE_SIZE))
5029 		bytes = (unsigned)PAGE_SIZE - offset;
5030 	ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
5031 				       offset, bytes);
5032 	if (unlikely(ret < 0))
5033 		return X86EMUL_IO_NEEDED;
5034 
5035 	return X86EMUL_CONTINUE;
5036 }
5037 
5038 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
5039 			       gva_t addr, void *val, unsigned int bytes,
5040 			       struct x86_exception *exception)
5041 {
5042 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5043 
5044 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
5045 					  exception);
5046 }
5047 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
5048 
5049 static int emulator_read_std(struct x86_emulate_ctxt *ctxt,
5050 			     gva_t addr, void *val, unsigned int bytes,
5051 			     struct x86_exception *exception, bool system)
5052 {
5053 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5054 	u32 access = 0;
5055 
5056 	if (!system && kvm_x86_ops->get_cpl(vcpu) == 3)
5057 		access |= PFERR_USER_MASK;
5058 
5059 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, exception);
5060 }
5061 
5062 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
5063 		unsigned long addr, void *val, unsigned int bytes)
5064 {
5065 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5066 	int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
5067 
5068 	return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
5069 }
5070 
5071 static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
5072 				      struct kvm_vcpu *vcpu, u32 access,
5073 				      struct x86_exception *exception)
5074 {
5075 	void *data = val;
5076 	int r = X86EMUL_CONTINUE;
5077 
5078 	while (bytes) {
5079 		gpa_t gpa =  vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
5080 							     access,
5081 							     exception);
5082 		unsigned offset = addr & (PAGE_SIZE-1);
5083 		unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
5084 		int ret;
5085 
5086 		if (gpa == UNMAPPED_GVA)
5087 			return X86EMUL_PROPAGATE_FAULT;
5088 		ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
5089 		if (ret < 0) {
5090 			r = X86EMUL_IO_NEEDED;
5091 			goto out;
5092 		}
5093 
5094 		bytes -= towrite;
5095 		data += towrite;
5096 		addr += towrite;
5097 	}
5098 out:
5099 	return r;
5100 }
5101 
5102 static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val,
5103 			      unsigned int bytes, struct x86_exception *exception,
5104 			      bool system)
5105 {
5106 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5107 	u32 access = PFERR_WRITE_MASK;
5108 
5109 	if (!system && kvm_x86_ops->get_cpl(vcpu) == 3)
5110 		access |= PFERR_USER_MASK;
5111 
5112 	return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
5113 					   access, exception);
5114 }
5115 
5116 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val,
5117 				unsigned int bytes, struct x86_exception *exception)
5118 {
5119 	/* kvm_write_guest_virt_system can pull in tons of pages. */
5120 	vcpu->arch.l1tf_flush_l1d = true;
5121 
5122 	return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
5123 					   PFERR_WRITE_MASK, exception);
5124 }
5125 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
5126 
5127 int handle_ud(struct kvm_vcpu *vcpu)
5128 {
5129 	int emul_type = EMULTYPE_TRAP_UD;
5130 	enum emulation_result er;
5131 	char sig[5]; /* ud2; .ascii "kvm" */
5132 	struct x86_exception e;
5133 
5134 	if (force_emulation_prefix &&
5135 	    kvm_read_guest_virt(vcpu, kvm_get_linear_rip(vcpu),
5136 				sig, sizeof(sig), &e) == 0 &&
5137 	    memcmp(sig, "\xf\xbkvm", sizeof(sig)) == 0) {
5138 		kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig));
5139 		emul_type = 0;
5140 	}
5141 
5142 	er = kvm_emulate_instruction(vcpu, emul_type);
5143 	if (er == EMULATE_USER_EXIT)
5144 		return 0;
5145 	if (er != EMULATE_DONE)
5146 		kvm_queue_exception(vcpu, UD_VECTOR);
5147 	return 1;
5148 }
5149 EXPORT_SYMBOL_GPL(handle_ud);
5150 
5151 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
5152 			    gpa_t gpa, bool write)
5153 {
5154 	/* For APIC access vmexit */
5155 	if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
5156 		return 1;
5157 
5158 	if (vcpu_match_mmio_gpa(vcpu, gpa)) {
5159 		trace_vcpu_match_mmio(gva, gpa, write, true);
5160 		return 1;
5161 	}
5162 
5163 	return 0;
5164 }
5165 
5166 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
5167 				gpa_t *gpa, struct x86_exception *exception,
5168 				bool write)
5169 {
5170 	u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
5171 		| (write ? PFERR_WRITE_MASK : 0);
5172 
5173 	/*
5174 	 * currently PKRU is only applied to ept enabled guest so
5175 	 * there is no pkey in EPT page table for L1 guest or EPT
5176 	 * shadow page table for L2 guest.
5177 	 */
5178 	if (vcpu_match_mmio_gva(vcpu, gva)
5179 	    && !permission_fault(vcpu, vcpu->arch.walk_mmu,
5180 				 vcpu->arch.access, 0, access)) {
5181 		*gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
5182 					(gva & (PAGE_SIZE - 1));
5183 		trace_vcpu_match_mmio(gva, *gpa, write, false);
5184 		return 1;
5185 	}
5186 
5187 	*gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5188 
5189 	if (*gpa == UNMAPPED_GVA)
5190 		return -1;
5191 
5192 	return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
5193 }
5194 
5195 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
5196 			const void *val, int bytes)
5197 {
5198 	int ret;
5199 
5200 	ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
5201 	if (ret < 0)
5202 		return 0;
5203 	kvm_page_track_write(vcpu, gpa, val, bytes);
5204 	return 1;
5205 }
5206 
5207 struct read_write_emulator_ops {
5208 	int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
5209 				  int bytes);
5210 	int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
5211 				  void *val, int bytes);
5212 	int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
5213 			       int bytes, void *val);
5214 	int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
5215 				    void *val, int bytes);
5216 	bool write;
5217 };
5218 
5219 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
5220 {
5221 	if (vcpu->mmio_read_completed) {
5222 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
5223 			       vcpu->mmio_fragments[0].gpa, val);
5224 		vcpu->mmio_read_completed = 0;
5225 		return 1;
5226 	}
5227 
5228 	return 0;
5229 }
5230 
5231 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
5232 			void *val, int bytes)
5233 {
5234 	return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
5235 }
5236 
5237 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
5238 			 void *val, int bytes)
5239 {
5240 	return emulator_write_phys(vcpu, gpa, val, bytes);
5241 }
5242 
5243 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
5244 {
5245 	trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
5246 	return vcpu_mmio_write(vcpu, gpa, bytes, val);
5247 }
5248 
5249 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
5250 			  void *val, int bytes)
5251 {
5252 	trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
5253 	return X86EMUL_IO_NEEDED;
5254 }
5255 
5256 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
5257 			   void *val, int bytes)
5258 {
5259 	struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
5260 
5261 	memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
5262 	return X86EMUL_CONTINUE;
5263 }
5264 
5265 static const struct read_write_emulator_ops read_emultor = {
5266 	.read_write_prepare = read_prepare,
5267 	.read_write_emulate = read_emulate,
5268 	.read_write_mmio = vcpu_mmio_read,
5269 	.read_write_exit_mmio = read_exit_mmio,
5270 };
5271 
5272 static const struct read_write_emulator_ops write_emultor = {
5273 	.read_write_emulate = write_emulate,
5274 	.read_write_mmio = write_mmio,
5275 	.read_write_exit_mmio = write_exit_mmio,
5276 	.write = true,
5277 };
5278 
5279 static int emulator_read_write_onepage(unsigned long addr, void *val,
5280 				       unsigned int bytes,
5281 				       struct x86_exception *exception,
5282 				       struct kvm_vcpu *vcpu,
5283 				       const struct read_write_emulator_ops *ops)
5284 {
5285 	gpa_t gpa;
5286 	int handled, ret;
5287 	bool write = ops->write;
5288 	struct kvm_mmio_fragment *frag;
5289 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5290 
5291 	/*
5292 	 * If the exit was due to a NPF we may already have a GPA.
5293 	 * If the GPA is present, use it to avoid the GVA to GPA table walk.
5294 	 * Note, this cannot be used on string operations since string
5295 	 * operation using rep will only have the initial GPA from the NPF
5296 	 * occurred.
5297 	 */
5298 	if (vcpu->arch.gpa_available &&
5299 	    emulator_can_use_gpa(ctxt) &&
5300 	    (addr & ~PAGE_MASK) == (vcpu->arch.gpa_val & ~PAGE_MASK)) {
5301 		gpa = vcpu->arch.gpa_val;
5302 		ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
5303 	} else {
5304 		ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
5305 		if (ret < 0)
5306 			return X86EMUL_PROPAGATE_FAULT;
5307 	}
5308 
5309 	if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
5310 		return X86EMUL_CONTINUE;
5311 
5312 	/*
5313 	 * Is this MMIO handled locally?
5314 	 */
5315 	handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
5316 	if (handled == bytes)
5317 		return X86EMUL_CONTINUE;
5318 
5319 	gpa += handled;
5320 	bytes -= handled;
5321 	val += handled;
5322 
5323 	WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
5324 	frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
5325 	frag->gpa = gpa;
5326 	frag->data = val;
5327 	frag->len = bytes;
5328 	return X86EMUL_CONTINUE;
5329 }
5330 
5331 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
5332 			unsigned long addr,
5333 			void *val, unsigned int bytes,
5334 			struct x86_exception *exception,
5335 			const struct read_write_emulator_ops *ops)
5336 {
5337 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5338 	gpa_t gpa;
5339 	int rc;
5340 
5341 	if (ops->read_write_prepare &&
5342 		  ops->read_write_prepare(vcpu, val, bytes))
5343 		return X86EMUL_CONTINUE;
5344 
5345 	vcpu->mmio_nr_fragments = 0;
5346 
5347 	/* Crossing a page boundary? */
5348 	if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
5349 		int now;
5350 
5351 		now = -addr & ~PAGE_MASK;
5352 		rc = emulator_read_write_onepage(addr, val, now, exception,
5353 						 vcpu, ops);
5354 
5355 		if (rc != X86EMUL_CONTINUE)
5356 			return rc;
5357 		addr += now;
5358 		if (ctxt->mode != X86EMUL_MODE_PROT64)
5359 			addr = (u32)addr;
5360 		val += now;
5361 		bytes -= now;
5362 	}
5363 
5364 	rc = emulator_read_write_onepage(addr, val, bytes, exception,
5365 					 vcpu, ops);
5366 	if (rc != X86EMUL_CONTINUE)
5367 		return rc;
5368 
5369 	if (!vcpu->mmio_nr_fragments)
5370 		return rc;
5371 
5372 	gpa = vcpu->mmio_fragments[0].gpa;
5373 
5374 	vcpu->mmio_needed = 1;
5375 	vcpu->mmio_cur_fragment = 0;
5376 
5377 	vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
5378 	vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
5379 	vcpu->run->exit_reason = KVM_EXIT_MMIO;
5380 	vcpu->run->mmio.phys_addr = gpa;
5381 
5382 	return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
5383 }
5384 
5385 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
5386 				  unsigned long addr,
5387 				  void *val,
5388 				  unsigned int bytes,
5389 				  struct x86_exception *exception)
5390 {
5391 	return emulator_read_write(ctxt, addr, val, bytes,
5392 				   exception, &read_emultor);
5393 }
5394 
5395 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
5396 			    unsigned long addr,
5397 			    const void *val,
5398 			    unsigned int bytes,
5399 			    struct x86_exception *exception)
5400 {
5401 	return emulator_read_write(ctxt, addr, (void *)val, bytes,
5402 				   exception, &write_emultor);
5403 }
5404 
5405 #define CMPXCHG_TYPE(t, ptr, old, new) \
5406 	(cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
5407 
5408 #ifdef CONFIG_X86_64
5409 #  define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
5410 #else
5411 #  define CMPXCHG64(ptr, old, new) \
5412 	(cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
5413 #endif
5414 
5415 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
5416 				     unsigned long addr,
5417 				     const void *old,
5418 				     const void *new,
5419 				     unsigned int bytes,
5420 				     struct x86_exception *exception)
5421 {
5422 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5423 	gpa_t gpa;
5424 	struct page *page;
5425 	char *kaddr;
5426 	bool exchanged;
5427 
5428 	/* guests cmpxchg8b have to be emulated atomically */
5429 	if (bytes > 8 || (bytes & (bytes - 1)))
5430 		goto emul_write;
5431 
5432 	gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
5433 
5434 	if (gpa == UNMAPPED_GVA ||
5435 	    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
5436 		goto emul_write;
5437 
5438 	if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
5439 		goto emul_write;
5440 
5441 	page = kvm_vcpu_gfn_to_page(vcpu, gpa >> PAGE_SHIFT);
5442 	if (is_error_page(page))
5443 		goto emul_write;
5444 
5445 	kaddr = kmap_atomic(page);
5446 	kaddr += offset_in_page(gpa);
5447 	switch (bytes) {
5448 	case 1:
5449 		exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
5450 		break;
5451 	case 2:
5452 		exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
5453 		break;
5454 	case 4:
5455 		exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
5456 		break;
5457 	case 8:
5458 		exchanged = CMPXCHG64(kaddr, old, new);
5459 		break;
5460 	default:
5461 		BUG();
5462 	}
5463 	kunmap_atomic(kaddr);
5464 	kvm_release_page_dirty(page);
5465 
5466 	if (!exchanged)
5467 		return X86EMUL_CMPXCHG_FAILED;
5468 
5469 	kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
5470 	kvm_page_track_write(vcpu, gpa, new, bytes);
5471 
5472 	return X86EMUL_CONTINUE;
5473 
5474 emul_write:
5475 	printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
5476 
5477 	return emulator_write_emulated(ctxt, addr, new, bytes, exception);
5478 }
5479 
5480 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
5481 {
5482 	int r = 0, i;
5483 
5484 	for (i = 0; i < vcpu->arch.pio.count; i++) {
5485 		if (vcpu->arch.pio.in)
5486 			r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
5487 					    vcpu->arch.pio.size, pd);
5488 		else
5489 			r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
5490 					     vcpu->arch.pio.port, vcpu->arch.pio.size,
5491 					     pd);
5492 		if (r)
5493 			break;
5494 		pd += vcpu->arch.pio.size;
5495 	}
5496 	return r;
5497 }
5498 
5499 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
5500 			       unsigned short port, void *val,
5501 			       unsigned int count, bool in)
5502 {
5503 	vcpu->arch.pio.port = port;
5504 	vcpu->arch.pio.in = in;
5505 	vcpu->arch.pio.count  = count;
5506 	vcpu->arch.pio.size = size;
5507 
5508 	if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
5509 		vcpu->arch.pio.count = 0;
5510 		return 1;
5511 	}
5512 
5513 	vcpu->run->exit_reason = KVM_EXIT_IO;
5514 	vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
5515 	vcpu->run->io.size = size;
5516 	vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
5517 	vcpu->run->io.count = count;
5518 	vcpu->run->io.port = port;
5519 
5520 	return 0;
5521 }
5522 
5523 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
5524 				    int size, unsigned short port, void *val,
5525 				    unsigned int count)
5526 {
5527 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5528 	int ret;
5529 
5530 	if (vcpu->arch.pio.count)
5531 		goto data_avail;
5532 
5533 	memset(vcpu->arch.pio_data, 0, size * count);
5534 
5535 	ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
5536 	if (ret) {
5537 data_avail:
5538 		memcpy(val, vcpu->arch.pio_data, size * count);
5539 		trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
5540 		vcpu->arch.pio.count = 0;
5541 		return 1;
5542 	}
5543 
5544 	return 0;
5545 }
5546 
5547 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
5548 				     int size, unsigned short port,
5549 				     const void *val, unsigned int count)
5550 {
5551 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5552 
5553 	memcpy(vcpu->arch.pio_data, val, size * count);
5554 	trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
5555 	return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
5556 }
5557 
5558 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
5559 {
5560 	return kvm_x86_ops->get_segment_base(vcpu, seg);
5561 }
5562 
5563 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
5564 {
5565 	kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
5566 }
5567 
5568 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
5569 {
5570 	if (!need_emulate_wbinvd(vcpu))
5571 		return X86EMUL_CONTINUE;
5572 
5573 	if (kvm_x86_ops->has_wbinvd_exit()) {
5574 		int cpu = get_cpu();
5575 
5576 		cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
5577 		smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
5578 				wbinvd_ipi, NULL, 1);
5579 		put_cpu();
5580 		cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
5581 	} else
5582 		wbinvd();
5583 	return X86EMUL_CONTINUE;
5584 }
5585 
5586 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
5587 {
5588 	kvm_emulate_wbinvd_noskip(vcpu);
5589 	return kvm_skip_emulated_instruction(vcpu);
5590 }
5591 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
5592 
5593 
5594 
5595 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
5596 {
5597 	kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
5598 }
5599 
5600 static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
5601 			   unsigned long *dest)
5602 {
5603 	return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
5604 }
5605 
5606 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
5607 			   unsigned long value)
5608 {
5609 
5610 	return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
5611 }
5612 
5613 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
5614 {
5615 	return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
5616 }
5617 
5618 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
5619 {
5620 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5621 	unsigned long value;
5622 
5623 	switch (cr) {
5624 	case 0:
5625 		value = kvm_read_cr0(vcpu);
5626 		break;
5627 	case 2:
5628 		value = vcpu->arch.cr2;
5629 		break;
5630 	case 3:
5631 		value = kvm_read_cr3(vcpu);
5632 		break;
5633 	case 4:
5634 		value = kvm_read_cr4(vcpu);
5635 		break;
5636 	case 8:
5637 		value = kvm_get_cr8(vcpu);
5638 		break;
5639 	default:
5640 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
5641 		return 0;
5642 	}
5643 
5644 	return value;
5645 }
5646 
5647 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
5648 {
5649 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5650 	int res = 0;
5651 
5652 	switch (cr) {
5653 	case 0:
5654 		res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
5655 		break;
5656 	case 2:
5657 		vcpu->arch.cr2 = val;
5658 		break;
5659 	case 3:
5660 		res = kvm_set_cr3(vcpu, val);
5661 		break;
5662 	case 4:
5663 		res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
5664 		break;
5665 	case 8:
5666 		res = kvm_set_cr8(vcpu, val);
5667 		break;
5668 	default:
5669 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
5670 		res = -1;
5671 	}
5672 
5673 	return res;
5674 }
5675 
5676 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
5677 {
5678 	return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
5679 }
5680 
5681 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5682 {
5683 	kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
5684 }
5685 
5686 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5687 {
5688 	kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
5689 }
5690 
5691 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5692 {
5693 	kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
5694 }
5695 
5696 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5697 {
5698 	kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
5699 }
5700 
5701 static unsigned long emulator_get_cached_segment_base(
5702 	struct x86_emulate_ctxt *ctxt, int seg)
5703 {
5704 	return get_segment_base(emul_to_vcpu(ctxt), seg);
5705 }
5706 
5707 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
5708 				 struct desc_struct *desc, u32 *base3,
5709 				 int seg)
5710 {
5711 	struct kvm_segment var;
5712 
5713 	kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
5714 	*selector = var.selector;
5715 
5716 	if (var.unusable) {
5717 		memset(desc, 0, sizeof(*desc));
5718 		if (base3)
5719 			*base3 = 0;
5720 		return false;
5721 	}
5722 
5723 	if (var.g)
5724 		var.limit >>= 12;
5725 	set_desc_limit(desc, var.limit);
5726 	set_desc_base(desc, (unsigned long)var.base);
5727 #ifdef CONFIG_X86_64
5728 	if (base3)
5729 		*base3 = var.base >> 32;
5730 #endif
5731 	desc->type = var.type;
5732 	desc->s = var.s;
5733 	desc->dpl = var.dpl;
5734 	desc->p = var.present;
5735 	desc->avl = var.avl;
5736 	desc->l = var.l;
5737 	desc->d = var.db;
5738 	desc->g = var.g;
5739 
5740 	return true;
5741 }
5742 
5743 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
5744 				 struct desc_struct *desc, u32 base3,
5745 				 int seg)
5746 {
5747 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5748 	struct kvm_segment var;
5749 
5750 	var.selector = selector;
5751 	var.base = get_desc_base(desc);
5752 #ifdef CONFIG_X86_64
5753 	var.base |= ((u64)base3) << 32;
5754 #endif
5755 	var.limit = get_desc_limit(desc);
5756 	if (desc->g)
5757 		var.limit = (var.limit << 12) | 0xfff;
5758 	var.type = desc->type;
5759 	var.dpl = desc->dpl;
5760 	var.db = desc->d;
5761 	var.s = desc->s;
5762 	var.l = desc->l;
5763 	var.g = desc->g;
5764 	var.avl = desc->avl;
5765 	var.present = desc->p;
5766 	var.unusable = !var.present;
5767 	var.padding = 0;
5768 
5769 	kvm_set_segment(vcpu, &var, seg);
5770 	return;
5771 }
5772 
5773 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
5774 			    u32 msr_index, u64 *pdata)
5775 {
5776 	struct msr_data msr;
5777 	int r;
5778 
5779 	msr.index = msr_index;
5780 	msr.host_initiated = false;
5781 	r = kvm_get_msr(emul_to_vcpu(ctxt), &msr);
5782 	if (r)
5783 		return r;
5784 
5785 	*pdata = msr.data;
5786 	return 0;
5787 }
5788 
5789 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
5790 			    u32 msr_index, u64 data)
5791 {
5792 	struct msr_data msr;
5793 
5794 	msr.data = data;
5795 	msr.index = msr_index;
5796 	msr.host_initiated = false;
5797 	return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
5798 }
5799 
5800 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
5801 {
5802 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5803 
5804 	return vcpu->arch.smbase;
5805 }
5806 
5807 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
5808 {
5809 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5810 
5811 	vcpu->arch.smbase = smbase;
5812 }
5813 
5814 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
5815 			      u32 pmc)
5816 {
5817 	return kvm_pmu_is_valid_msr_idx(emul_to_vcpu(ctxt), pmc);
5818 }
5819 
5820 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
5821 			     u32 pmc, u64 *pdata)
5822 {
5823 	return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
5824 }
5825 
5826 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
5827 {
5828 	emul_to_vcpu(ctxt)->arch.halt_request = 1;
5829 }
5830 
5831 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
5832 			      struct x86_instruction_info *info,
5833 			      enum x86_intercept_stage stage)
5834 {
5835 	return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
5836 }
5837 
5838 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
5839 			u32 *eax, u32 *ebx, u32 *ecx, u32 *edx, bool check_limit)
5840 {
5841 	return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, check_limit);
5842 }
5843 
5844 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
5845 {
5846 	return kvm_register_read(emul_to_vcpu(ctxt), reg);
5847 }
5848 
5849 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
5850 {
5851 	kvm_register_write(emul_to_vcpu(ctxt), reg, val);
5852 }
5853 
5854 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
5855 {
5856 	kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked);
5857 }
5858 
5859 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
5860 {
5861 	return emul_to_vcpu(ctxt)->arch.hflags;
5862 }
5863 
5864 static void emulator_set_hflags(struct x86_emulate_ctxt *ctxt, unsigned emul_flags)
5865 {
5866 	kvm_set_hflags(emul_to_vcpu(ctxt), emul_flags);
5867 }
5868 
5869 static int emulator_pre_leave_smm(struct x86_emulate_ctxt *ctxt, u64 smbase)
5870 {
5871 	return kvm_x86_ops->pre_leave_smm(emul_to_vcpu(ctxt), smbase);
5872 }
5873 
5874 static const struct x86_emulate_ops emulate_ops = {
5875 	.read_gpr            = emulator_read_gpr,
5876 	.write_gpr           = emulator_write_gpr,
5877 	.read_std            = emulator_read_std,
5878 	.write_std           = emulator_write_std,
5879 	.read_phys           = kvm_read_guest_phys_system,
5880 	.fetch               = kvm_fetch_guest_virt,
5881 	.read_emulated       = emulator_read_emulated,
5882 	.write_emulated      = emulator_write_emulated,
5883 	.cmpxchg_emulated    = emulator_cmpxchg_emulated,
5884 	.invlpg              = emulator_invlpg,
5885 	.pio_in_emulated     = emulator_pio_in_emulated,
5886 	.pio_out_emulated    = emulator_pio_out_emulated,
5887 	.get_segment         = emulator_get_segment,
5888 	.set_segment         = emulator_set_segment,
5889 	.get_cached_segment_base = emulator_get_cached_segment_base,
5890 	.get_gdt             = emulator_get_gdt,
5891 	.get_idt	     = emulator_get_idt,
5892 	.set_gdt             = emulator_set_gdt,
5893 	.set_idt	     = emulator_set_idt,
5894 	.get_cr              = emulator_get_cr,
5895 	.set_cr              = emulator_set_cr,
5896 	.cpl                 = emulator_get_cpl,
5897 	.get_dr              = emulator_get_dr,
5898 	.set_dr              = emulator_set_dr,
5899 	.get_smbase          = emulator_get_smbase,
5900 	.set_smbase          = emulator_set_smbase,
5901 	.set_msr             = emulator_set_msr,
5902 	.get_msr             = emulator_get_msr,
5903 	.check_pmc	     = emulator_check_pmc,
5904 	.read_pmc            = emulator_read_pmc,
5905 	.halt                = emulator_halt,
5906 	.wbinvd              = emulator_wbinvd,
5907 	.fix_hypercall       = emulator_fix_hypercall,
5908 	.intercept           = emulator_intercept,
5909 	.get_cpuid           = emulator_get_cpuid,
5910 	.set_nmi_mask        = emulator_set_nmi_mask,
5911 	.get_hflags          = emulator_get_hflags,
5912 	.set_hflags          = emulator_set_hflags,
5913 	.pre_leave_smm       = emulator_pre_leave_smm,
5914 };
5915 
5916 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
5917 {
5918 	u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
5919 	/*
5920 	 * an sti; sti; sequence only disable interrupts for the first
5921 	 * instruction. So, if the last instruction, be it emulated or
5922 	 * not, left the system with the INT_STI flag enabled, it
5923 	 * means that the last instruction is an sti. We should not
5924 	 * leave the flag on in this case. The same goes for mov ss
5925 	 */
5926 	if (int_shadow & mask)
5927 		mask = 0;
5928 	if (unlikely(int_shadow || mask)) {
5929 		kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
5930 		if (!mask)
5931 			kvm_make_request(KVM_REQ_EVENT, vcpu);
5932 	}
5933 }
5934 
5935 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
5936 {
5937 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5938 	if (ctxt->exception.vector == PF_VECTOR)
5939 		return kvm_propagate_fault(vcpu, &ctxt->exception);
5940 
5941 	if (ctxt->exception.error_code_valid)
5942 		kvm_queue_exception_e(vcpu, ctxt->exception.vector,
5943 				      ctxt->exception.error_code);
5944 	else
5945 		kvm_queue_exception(vcpu, ctxt->exception.vector);
5946 	return false;
5947 }
5948 
5949 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
5950 {
5951 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5952 	int cs_db, cs_l;
5953 
5954 	kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
5955 
5956 	ctxt->eflags = kvm_get_rflags(vcpu);
5957 	ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
5958 
5959 	ctxt->eip = kvm_rip_read(vcpu);
5960 	ctxt->mode = (!is_protmode(vcpu))		? X86EMUL_MODE_REAL :
5961 		     (ctxt->eflags & X86_EFLAGS_VM)	? X86EMUL_MODE_VM86 :
5962 		     (cs_l && is_long_mode(vcpu))	? X86EMUL_MODE_PROT64 :
5963 		     cs_db				? X86EMUL_MODE_PROT32 :
5964 							  X86EMUL_MODE_PROT16;
5965 	BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
5966 	BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
5967 	BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
5968 
5969 	init_decode_cache(ctxt);
5970 	vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5971 }
5972 
5973 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
5974 {
5975 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5976 	int ret;
5977 
5978 	init_emulate_ctxt(vcpu);
5979 
5980 	ctxt->op_bytes = 2;
5981 	ctxt->ad_bytes = 2;
5982 	ctxt->_eip = ctxt->eip + inc_eip;
5983 	ret = emulate_int_real(ctxt, irq);
5984 
5985 	if (ret != X86EMUL_CONTINUE)
5986 		return EMULATE_FAIL;
5987 
5988 	ctxt->eip = ctxt->_eip;
5989 	kvm_rip_write(vcpu, ctxt->eip);
5990 	kvm_set_rflags(vcpu, ctxt->eflags);
5991 
5992 	return EMULATE_DONE;
5993 }
5994 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
5995 
5996 static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type)
5997 {
5998 	int r = EMULATE_DONE;
5999 
6000 	++vcpu->stat.insn_emulation_fail;
6001 	trace_kvm_emulate_insn_failed(vcpu);
6002 
6003 	if (emulation_type & EMULTYPE_NO_UD_ON_FAIL)
6004 		return EMULATE_FAIL;
6005 
6006 	if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
6007 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6008 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6009 		vcpu->run->internal.ndata = 0;
6010 		r = EMULATE_USER_EXIT;
6011 	}
6012 
6013 	kvm_queue_exception(vcpu, UD_VECTOR);
6014 
6015 	return r;
6016 }
6017 
6018 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
6019 				  bool write_fault_to_shadow_pgtable,
6020 				  int emulation_type)
6021 {
6022 	gpa_t gpa = cr2;
6023 	kvm_pfn_t pfn;
6024 
6025 	if (!(emulation_type & EMULTYPE_ALLOW_RETRY))
6026 		return false;
6027 
6028 	if (WARN_ON_ONCE(is_guest_mode(vcpu)))
6029 		return false;
6030 
6031 	if (!vcpu->arch.mmu->direct_map) {
6032 		/*
6033 		 * Write permission should be allowed since only
6034 		 * write access need to be emulated.
6035 		 */
6036 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
6037 
6038 		/*
6039 		 * If the mapping is invalid in guest, let cpu retry
6040 		 * it to generate fault.
6041 		 */
6042 		if (gpa == UNMAPPED_GVA)
6043 			return true;
6044 	}
6045 
6046 	/*
6047 	 * Do not retry the unhandleable instruction if it faults on the
6048 	 * readonly host memory, otherwise it will goto a infinite loop:
6049 	 * retry instruction -> write #PF -> emulation fail -> retry
6050 	 * instruction -> ...
6051 	 */
6052 	pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
6053 
6054 	/*
6055 	 * If the instruction failed on the error pfn, it can not be fixed,
6056 	 * report the error to userspace.
6057 	 */
6058 	if (is_error_noslot_pfn(pfn))
6059 		return false;
6060 
6061 	kvm_release_pfn_clean(pfn);
6062 
6063 	/* The instructions are well-emulated on direct mmu. */
6064 	if (vcpu->arch.mmu->direct_map) {
6065 		unsigned int indirect_shadow_pages;
6066 
6067 		spin_lock(&vcpu->kvm->mmu_lock);
6068 		indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
6069 		spin_unlock(&vcpu->kvm->mmu_lock);
6070 
6071 		if (indirect_shadow_pages)
6072 			kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6073 
6074 		return true;
6075 	}
6076 
6077 	/*
6078 	 * if emulation was due to access to shadowed page table
6079 	 * and it failed try to unshadow page and re-enter the
6080 	 * guest to let CPU execute the instruction.
6081 	 */
6082 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6083 
6084 	/*
6085 	 * If the access faults on its page table, it can not
6086 	 * be fixed by unprotecting shadow page and it should
6087 	 * be reported to userspace.
6088 	 */
6089 	return !write_fault_to_shadow_pgtable;
6090 }
6091 
6092 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
6093 			      unsigned long cr2,  int emulation_type)
6094 {
6095 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6096 	unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
6097 
6098 	last_retry_eip = vcpu->arch.last_retry_eip;
6099 	last_retry_addr = vcpu->arch.last_retry_addr;
6100 
6101 	/*
6102 	 * If the emulation is caused by #PF and it is non-page_table
6103 	 * writing instruction, it means the VM-EXIT is caused by shadow
6104 	 * page protected, we can zap the shadow page and retry this
6105 	 * instruction directly.
6106 	 *
6107 	 * Note: if the guest uses a non-page-table modifying instruction
6108 	 * on the PDE that points to the instruction, then we will unmap
6109 	 * the instruction and go to an infinite loop. So, we cache the
6110 	 * last retried eip and the last fault address, if we meet the eip
6111 	 * and the address again, we can break out of the potential infinite
6112 	 * loop.
6113 	 */
6114 	vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
6115 
6116 	if (!(emulation_type & EMULTYPE_ALLOW_RETRY))
6117 		return false;
6118 
6119 	if (WARN_ON_ONCE(is_guest_mode(vcpu)))
6120 		return false;
6121 
6122 	if (x86_page_table_writing_insn(ctxt))
6123 		return false;
6124 
6125 	if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
6126 		return false;
6127 
6128 	vcpu->arch.last_retry_eip = ctxt->eip;
6129 	vcpu->arch.last_retry_addr = cr2;
6130 
6131 	if (!vcpu->arch.mmu->direct_map)
6132 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
6133 
6134 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6135 
6136 	return true;
6137 }
6138 
6139 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
6140 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
6141 
6142 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
6143 {
6144 	if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
6145 		/* This is a good place to trace that we are exiting SMM.  */
6146 		trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
6147 
6148 		/* Process a latched INIT or SMI, if any.  */
6149 		kvm_make_request(KVM_REQ_EVENT, vcpu);
6150 	}
6151 
6152 	kvm_mmu_reset_context(vcpu);
6153 }
6154 
6155 static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags)
6156 {
6157 	unsigned changed = vcpu->arch.hflags ^ emul_flags;
6158 
6159 	vcpu->arch.hflags = emul_flags;
6160 
6161 	if (changed & HF_SMM_MASK)
6162 		kvm_smm_changed(vcpu);
6163 }
6164 
6165 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
6166 				unsigned long *db)
6167 {
6168 	u32 dr6 = 0;
6169 	int i;
6170 	u32 enable, rwlen;
6171 
6172 	enable = dr7;
6173 	rwlen = dr7 >> 16;
6174 	for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
6175 		if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
6176 			dr6 |= (1 << i);
6177 	return dr6;
6178 }
6179 
6180 static void kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu, int *r)
6181 {
6182 	struct kvm_run *kvm_run = vcpu->run;
6183 
6184 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
6185 		kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 | DR6_RTM;
6186 		kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
6187 		kvm_run->debug.arch.exception = DB_VECTOR;
6188 		kvm_run->exit_reason = KVM_EXIT_DEBUG;
6189 		*r = EMULATE_USER_EXIT;
6190 	} else {
6191 		kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BS);
6192 	}
6193 }
6194 
6195 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
6196 {
6197 	unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
6198 	int r = EMULATE_DONE;
6199 
6200 	kvm_x86_ops->skip_emulated_instruction(vcpu);
6201 
6202 	/*
6203 	 * rflags is the old, "raw" value of the flags.  The new value has
6204 	 * not been saved yet.
6205 	 *
6206 	 * This is correct even for TF set by the guest, because "the
6207 	 * processor will not generate this exception after the instruction
6208 	 * that sets the TF flag".
6209 	 */
6210 	if (unlikely(rflags & X86_EFLAGS_TF))
6211 		kvm_vcpu_do_singlestep(vcpu, &r);
6212 	return r == EMULATE_DONE;
6213 }
6214 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
6215 
6216 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
6217 {
6218 	if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
6219 	    (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
6220 		struct kvm_run *kvm_run = vcpu->run;
6221 		unsigned long eip = kvm_get_linear_rip(vcpu);
6222 		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
6223 					   vcpu->arch.guest_debug_dr7,
6224 					   vcpu->arch.eff_db);
6225 
6226 		if (dr6 != 0) {
6227 			kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
6228 			kvm_run->debug.arch.pc = eip;
6229 			kvm_run->debug.arch.exception = DB_VECTOR;
6230 			kvm_run->exit_reason = KVM_EXIT_DEBUG;
6231 			*r = EMULATE_USER_EXIT;
6232 			return true;
6233 		}
6234 	}
6235 
6236 	if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
6237 	    !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
6238 		unsigned long eip = kvm_get_linear_rip(vcpu);
6239 		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
6240 					   vcpu->arch.dr7,
6241 					   vcpu->arch.db);
6242 
6243 		if (dr6 != 0) {
6244 			vcpu->arch.dr6 &= ~15;
6245 			vcpu->arch.dr6 |= dr6 | DR6_RTM;
6246 			kvm_queue_exception(vcpu, DB_VECTOR);
6247 			*r = EMULATE_DONE;
6248 			return true;
6249 		}
6250 	}
6251 
6252 	return false;
6253 }
6254 
6255 static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
6256 {
6257 	switch (ctxt->opcode_len) {
6258 	case 1:
6259 		switch (ctxt->b) {
6260 		case 0xe4:	/* IN */
6261 		case 0xe5:
6262 		case 0xec:
6263 		case 0xed:
6264 		case 0xe6:	/* OUT */
6265 		case 0xe7:
6266 		case 0xee:
6267 		case 0xef:
6268 		case 0x6c:	/* INS */
6269 		case 0x6d:
6270 		case 0x6e:	/* OUTS */
6271 		case 0x6f:
6272 			return true;
6273 		}
6274 		break;
6275 	case 2:
6276 		switch (ctxt->b) {
6277 		case 0x33:	/* RDPMC */
6278 			return true;
6279 		}
6280 		break;
6281 	}
6282 
6283 	return false;
6284 }
6285 
6286 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
6287 			    unsigned long cr2,
6288 			    int emulation_type,
6289 			    void *insn,
6290 			    int insn_len)
6291 {
6292 	int r;
6293 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6294 	bool writeback = true;
6295 	bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
6296 
6297 	vcpu->arch.l1tf_flush_l1d = true;
6298 
6299 	/*
6300 	 * Clear write_fault_to_shadow_pgtable here to ensure it is
6301 	 * never reused.
6302 	 */
6303 	vcpu->arch.write_fault_to_shadow_pgtable = false;
6304 	kvm_clear_exception_queue(vcpu);
6305 
6306 	if (!(emulation_type & EMULTYPE_NO_DECODE)) {
6307 		init_emulate_ctxt(vcpu);
6308 
6309 		/*
6310 		 * We will reenter on the same instruction since
6311 		 * we do not set complete_userspace_io.  This does not
6312 		 * handle watchpoints yet, those would be handled in
6313 		 * the emulate_ops.
6314 		 */
6315 		if (!(emulation_type & EMULTYPE_SKIP) &&
6316 		    kvm_vcpu_check_breakpoint(vcpu, &r))
6317 			return r;
6318 
6319 		ctxt->interruptibility = 0;
6320 		ctxt->have_exception = false;
6321 		ctxt->exception.vector = -1;
6322 		ctxt->perm_ok = false;
6323 
6324 		ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
6325 
6326 		r = x86_decode_insn(ctxt, insn, insn_len);
6327 
6328 		trace_kvm_emulate_insn_start(vcpu);
6329 		++vcpu->stat.insn_emulation;
6330 		if (r != EMULATION_OK)  {
6331 			if (emulation_type & EMULTYPE_TRAP_UD)
6332 				return EMULATE_FAIL;
6333 			if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
6334 						emulation_type))
6335 				return EMULATE_DONE;
6336 			if (ctxt->have_exception && inject_emulated_exception(vcpu))
6337 				return EMULATE_DONE;
6338 			if (emulation_type & EMULTYPE_SKIP)
6339 				return EMULATE_FAIL;
6340 			return handle_emulation_failure(vcpu, emulation_type);
6341 		}
6342 	}
6343 
6344 	if ((emulation_type & EMULTYPE_VMWARE) &&
6345 	    !is_vmware_backdoor_opcode(ctxt))
6346 		return EMULATE_FAIL;
6347 
6348 	if (emulation_type & EMULTYPE_SKIP) {
6349 		kvm_rip_write(vcpu, ctxt->_eip);
6350 		if (ctxt->eflags & X86_EFLAGS_RF)
6351 			kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
6352 		return EMULATE_DONE;
6353 	}
6354 
6355 	if (retry_instruction(ctxt, cr2, emulation_type))
6356 		return EMULATE_DONE;
6357 
6358 	/* this is needed for vmware backdoor interface to work since it
6359 	   changes registers values  during IO operation */
6360 	if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
6361 		vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
6362 		emulator_invalidate_register_cache(ctxt);
6363 	}
6364 
6365 restart:
6366 	/* Save the faulting GPA (cr2) in the address field */
6367 	ctxt->exception.address = cr2;
6368 
6369 	r = x86_emulate_insn(ctxt);
6370 
6371 	if (r == EMULATION_INTERCEPTED)
6372 		return EMULATE_DONE;
6373 
6374 	if (r == EMULATION_FAILED) {
6375 		if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
6376 					emulation_type))
6377 			return EMULATE_DONE;
6378 
6379 		return handle_emulation_failure(vcpu, emulation_type);
6380 	}
6381 
6382 	if (ctxt->have_exception) {
6383 		r = EMULATE_DONE;
6384 		if (inject_emulated_exception(vcpu))
6385 			return r;
6386 	} else if (vcpu->arch.pio.count) {
6387 		if (!vcpu->arch.pio.in) {
6388 			/* FIXME: return into emulator if single-stepping.  */
6389 			vcpu->arch.pio.count = 0;
6390 		} else {
6391 			writeback = false;
6392 			vcpu->arch.complete_userspace_io = complete_emulated_pio;
6393 		}
6394 		r = EMULATE_USER_EXIT;
6395 	} else if (vcpu->mmio_needed) {
6396 		if (!vcpu->mmio_is_write)
6397 			writeback = false;
6398 		r = EMULATE_USER_EXIT;
6399 		vcpu->arch.complete_userspace_io = complete_emulated_mmio;
6400 	} else if (r == EMULATION_RESTART)
6401 		goto restart;
6402 	else
6403 		r = EMULATE_DONE;
6404 
6405 	if (writeback) {
6406 		unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
6407 		toggle_interruptibility(vcpu, ctxt->interruptibility);
6408 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6409 		kvm_rip_write(vcpu, ctxt->eip);
6410 		if (r == EMULATE_DONE &&
6411 		    (ctxt->tf || (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)))
6412 			kvm_vcpu_do_singlestep(vcpu, &r);
6413 		if (!ctxt->have_exception ||
6414 		    exception_type(ctxt->exception.vector) == EXCPT_TRAP)
6415 			__kvm_set_rflags(vcpu, ctxt->eflags);
6416 
6417 		/*
6418 		 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
6419 		 * do nothing, and it will be requested again as soon as
6420 		 * the shadow expires.  But we still need to check here,
6421 		 * because POPF has no interrupt shadow.
6422 		 */
6423 		if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
6424 			kvm_make_request(KVM_REQ_EVENT, vcpu);
6425 	} else
6426 		vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
6427 
6428 	return r;
6429 }
6430 
6431 int kvm_emulate_instruction(struct kvm_vcpu *vcpu, int emulation_type)
6432 {
6433 	return x86_emulate_instruction(vcpu, 0, emulation_type, NULL, 0);
6434 }
6435 EXPORT_SYMBOL_GPL(kvm_emulate_instruction);
6436 
6437 int kvm_emulate_instruction_from_buffer(struct kvm_vcpu *vcpu,
6438 					void *insn, int insn_len)
6439 {
6440 	return x86_emulate_instruction(vcpu, 0, 0, insn, insn_len);
6441 }
6442 EXPORT_SYMBOL_GPL(kvm_emulate_instruction_from_buffer);
6443 
6444 static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size,
6445 			    unsigned short port)
6446 {
6447 	unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
6448 	int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
6449 					    size, port, &val, 1);
6450 	/* do not return to emulator after return from userspace */
6451 	vcpu->arch.pio.count = 0;
6452 	return ret;
6453 }
6454 
6455 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
6456 {
6457 	unsigned long val;
6458 
6459 	/* We should only ever be called with arch.pio.count equal to 1 */
6460 	BUG_ON(vcpu->arch.pio.count != 1);
6461 
6462 	/* For size less than 4 we merge, else we zero extend */
6463 	val = (vcpu->arch.pio.size < 4) ? kvm_register_read(vcpu, VCPU_REGS_RAX)
6464 					: 0;
6465 
6466 	/*
6467 	 * Since vcpu->arch.pio.count == 1 let emulator_pio_in_emulated perform
6468 	 * the copy and tracing
6469 	 */
6470 	emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, vcpu->arch.pio.size,
6471 				 vcpu->arch.pio.port, &val, 1);
6472 	kvm_register_write(vcpu, VCPU_REGS_RAX, val);
6473 
6474 	return 1;
6475 }
6476 
6477 static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size,
6478 			   unsigned short port)
6479 {
6480 	unsigned long val;
6481 	int ret;
6482 
6483 	/* For size less than 4 we merge, else we zero extend */
6484 	val = (size < 4) ? kvm_register_read(vcpu, VCPU_REGS_RAX) : 0;
6485 
6486 	ret = emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, size, port,
6487 				       &val, 1);
6488 	if (ret) {
6489 		kvm_register_write(vcpu, VCPU_REGS_RAX, val);
6490 		return ret;
6491 	}
6492 
6493 	vcpu->arch.complete_userspace_io = complete_fast_pio_in;
6494 
6495 	return 0;
6496 }
6497 
6498 int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in)
6499 {
6500 	int ret = kvm_skip_emulated_instruction(vcpu);
6501 
6502 	/*
6503 	 * TODO: we might be squashing a KVM_GUESTDBG_SINGLESTEP-triggered
6504 	 * KVM_EXIT_DEBUG here.
6505 	 */
6506 	if (in)
6507 		return kvm_fast_pio_in(vcpu, size, port) && ret;
6508 	else
6509 		return kvm_fast_pio_out(vcpu, size, port) && ret;
6510 }
6511 EXPORT_SYMBOL_GPL(kvm_fast_pio);
6512 
6513 static int kvmclock_cpu_down_prep(unsigned int cpu)
6514 {
6515 	__this_cpu_write(cpu_tsc_khz, 0);
6516 	return 0;
6517 }
6518 
6519 static void tsc_khz_changed(void *data)
6520 {
6521 	struct cpufreq_freqs *freq = data;
6522 	unsigned long khz = 0;
6523 
6524 	if (data)
6525 		khz = freq->new;
6526 	else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
6527 		khz = cpufreq_quick_get(raw_smp_processor_id());
6528 	if (!khz)
6529 		khz = tsc_khz;
6530 	__this_cpu_write(cpu_tsc_khz, khz);
6531 }
6532 
6533 #ifdef CONFIG_X86_64
6534 static void kvm_hyperv_tsc_notifier(void)
6535 {
6536 	struct kvm *kvm;
6537 	struct kvm_vcpu *vcpu;
6538 	int cpu;
6539 
6540 	spin_lock(&kvm_lock);
6541 	list_for_each_entry(kvm, &vm_list, vm_list)
6542 		kvm_make_mclock_inprogress_request(kvm);
6543 
6544 	hyperv_stop_tsc_emulation();
6545 
6546 	/* TSC frequency always matches when on Hyper-V */
6547 	for_each_present_cpu(cpu)
6548 		per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
6549 	kvm_max_guest_tsc_khz = tsc_khz;
6550 
6551 	list_for_each_entry(kvm, &vm_list, vm_list) {
6552 		struct kvm_arch *ka = &kvm->arch;
6553 
6554 		spin_lock(&ka->pvclock_gtod_sync_lock);
6555 
6556 		pvclock_update_vm_gtod_copy(kvm);
6557 
6558 		kvm_for_each_vcpu(cpu, vcpu, kvm)
6559 			kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6560 
6561 		kvm_for_each_vcpu(cpu, vcpu, kvm)
6562 			kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
6563 
6564 		spin_unlock(&ka->pvclock_gtod_sync_lock);
6565 	}
6566 	spin_unlock(&kvm_lock);
6567 }
6568 #endif
6569 
6570 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
6571 				     void *data)
6572 {
6573 	struct cpufreq_freqs *freq = data;
6574 	struct kvm *kvm;
6575 	struct kvm_vcpu *vcpu;
6576 	int i, send_ipi = 0;
6577 
6578 	/*
6579 	 * We allow guests to temporarily run on slowing clocks,
6580 	 * provided we notify them after, or to run on accelerating
6581 	 * clocks, provided we notify them before.  Thus time never
6582 	 * goes backwards.
6583 	 *
6584 	 * However, we have a problem.  We can't atomically update
6585 	 * the frequency of a given CPU from this function; it is
6586 	 * merely a notifier, which can be called from any CPU.
6587 	 * Changing the TSC frequency at arbitrary points in time
6588 	 * requires a recomputation of local variables related to
6589 	 * the TSC for each VCPU.  We must flag these local variables
6590 	 * to be updated and be sure the update takes place with the
6591 	 * new frequency before any guests proceed.
6592 	 *
6593 	 * Unfortunately, the combination of hotplug CPU and frequency
6594 	 * change creates an intractable locking scenario; the order
6595 	 * of when these callouts happen is undefined with respect to
6596 	 * CPU hotplug, and they can race with each other.  As such,
6597 	 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
6598 	 * undefined; you can actually have a CPU frequency change take
6599 	 * place in between the computation of X and the setting of the
6600 	 * variable.  To protect against this problem, all updates of
6601 	 * the per_cpu tsc_khz variable are done in an interrupt
6602 	 * protected IPI, and all callers wishing to update the value
6603 	 * must wait for a synchronous IPI to complete (which is trivial
6604 	 * if the caller is on the CPU already).  This establishes the
6605 	 * necessary total order on variable updates.
6606 	 *
6607 	 * Note that because a guest time update may take place
6608 	 * anytime after the setting of the VCPU's request bit, the
6609 	 * correct TSC value must be set before the request.  However,
6610 	 * to ensure the update actually makes it to any guest which
6611 	 * starts running in hardware virtualization between the set
6612 	 * and the acquisition of the spinlock, we must also ping the
6613 	 * CPU after setting the request bit.
6614 	 *
6615 	 */
6616 
6617 	if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
6618 		return 0;
6619 	if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
6620 		return 0;
6621 
6622 	smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
6623 
6624 	spin_lock(&kvm_lock);
6625 	list_for_each_entry(kvm, &vm_list, vm_list) {
6626 		kvm_for_each_vcpu(i, vcpu, kvm) {
6627 			if (vcpu->cpu != freq->cpu)
6628 				continue;
6629 			kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6630 			if (vcpu->cpu != smp_processor_id())
6631 				send_ipi = 1;
6632 		}
6633 	}
6634 	spin_unlock(&kvm_lock);
6635 
6636 	if (freq->old < freq->new && send_ipi) {
6637 		/*
6638 		 * We upscale the frequency.  Must make the guest
6639 		 * doesn't see old kvmclock values while running with
6640 		 * the new frequency, otherwise we risk the guest sees
6641 		 * time go backwards.
6642 		 *
6643 		 * In case we update the frequency for another cpu
6644 		 * (which might be in guest context) send an interrupt
6645 		 * to kick the cpu out of guest context.  Next time
6646 		 * guest context is entered kvmclock will be updated,
6647 		 * so the guest will not see stale values.
6648 		 */
6649 		smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
6650 	}
6651 	return 0;
6652 }
6653 
6654 static struct notifier_block kvmclock_cpufreq_notifier_block = {
6655 	.notifier_call  = kvmclock_cpufreq_notifier
6656 };
6657 
6658 static int kvmclock_cpu_online(unsigned int cpu)
6659 {
6660 	tsc_khz_changed(NULL);
6661 	return 0;
6662 }
6663 
6664 static void kvm_timer_init(void)
6665 {
6666 	max_tsc_khz = tsc_khz;
6667 
6668 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
6669 #ifdef CONFIG_CPU_FREQ
6670 		struct cpufreq_policy policy;
6671 		int cpu;
6672 
6673 		memset(&policy, 0, sizeof(policy));
6674 		cpu = get_cpu();
6675 		cpufreq_get_policy(&policy, cpu);
6676 		if (policy.cpuinfo.max_freq)
6677 			max_tsc_khz = policy.cpuinfo.max_freq;
6678 		put_cpu();
6679 #endif
6680 		cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
6681 					  CPUFREQ_TRANSITION_NOTIFIER);
6682 	}
6683 	pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
6684 
6685 	cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
6686 			  kvmclock_cpu_online, kvmclock_cpu_down_prep);
6687 }
6688 
6689 DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
6690 EXPORT_PER_CPU_SYMBOL_GPL(current_vcpu);
6691 
6692 int kvm_is_in_guest(void)
6693 {
6694 	return __this_cpu_read(current_vcpu) != NULL;
6695 }
6696 
6697 static int kvm_is_user_mode(void)
6698 {
6699 	int user_mode = 3;
6700 
6701 	if (__this_cpu_read(current_vcpu))
6702 		user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
6703 
6704 	return user_mode != 0;
6705 }
6706 
6707 static unsigned long kvm_get_guest_ip(void)
6708 {
6709 	unsigned long ip = 0;
6710 
6711 	if (__this_cpu_read(current_vcpu))
6712 		ip = kvm_rip_read(__this_cpu_read(current_vcpu));
6713 
6714 	return ip;
6715 }
6716 
6717 static struct perf_guest_info_callbacks kvm_guest_cbs = {
6718 	.is_in_guest		= kvm_is_in_guest,
6719 	.is_user_mode		= kvm_is_user_mode,
6720 	.get_guest_ip		= kvm_get_guest_ip,
6721 };
6722 
6723 static void kvm_set_mmio_spte_mask(void)
6724 {
6725 	u64 mask;
6726 	int maxphyaddr = boot_cpu_data.x86_phys_bits;
6727 
6728 	/*
6729 	 * Set the reserved bits and the present bit of an paging-structure
6730 	 * entry to generate page fault with PFER.RSV = 1.
6731 	 */
6732 
6733 	/*
6734 	 * Mask the uppermost physical address bit, which would be reserved as
6735 	 * long as the supported physical address width is less than 52.
6736 	 */
6737 	mask = 1ull << 51;
6738 
6739 	/* Set the present bit. */
6740 	mask |= 1ull;
6741 
6742 	/*
6743 	 * If reserved bit is not supported, clear the present bit to disable
6744 	 * mmio page fault.
6745 	 */
6746 	if (IS_ENABLED(CONFIG_X86_64) && maxphyaddr == 52)
6747 		mask &= ~1ull;
6748 
6749 	kvm_mmu_set_mmio_spte_mask(mask, mask);
6750 }
6751 
6752 #ifdef CONFIG_X86_64
6753 static void pvclock_gtod_update_fn(struct work_struct *work)
6754 {
6755 	struct kvm *kvm;
6756 
6757 	struct kvm_vcpu *vcpu;
6758 	int i;
6759 
6760 	spin_lock(&kvm_lock);
6761 	list_for_each_entry(kvm, &vm_list, vm_list)
6762 		kvm_for_each_vcpu(i, vcpu, kvm)
6763 			kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
6764 	atomic_set(&kvm_guest_has_master_clock, 0);
6765 	spin_unlock(&kvm_lock);
6766 }
6767 
6768 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
6769 
6770 /*
6771  * Notification about pvclock gtod data update.
6772  */
6773 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
6774 			       void *priv)
6775 {
6776 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
6777 	struct timekeeper *tk = priv;
6778 
6779 	update_pvclock_gtod(tk);
6780 
6781 	/* disable master clock if host does not trust, or does not
6782 	 * use, TSC based clocksource.
6783 	 */
6784 	if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) &&
6785 	    atomic_read(&kvm_guest_has_master_clock) != 0)
6786 		queue_work(system_long_wq, &pvclock_gtod_work);
6787 
6788 	return 0;
6789 }
6790 
6791 static struct notifier_block pvclock_gtod_notifier = {
6792 	.notifier_call = pvclock_gtod_notify,
6793 };
6794 #endif
6795 
6796 int kvm_arch_init(void *opaque)
6797 {
6798 	int r;
6799 	struct kvm_x86_ops *ops = opaque;
6800 
6801 	if (kvm_x86_ops) {
6802 		printk(KERN_ERR "kvm: already loaded the other module\n");
6803 		r = -EEXIST;
6804 		goto out;
6805 	}
6806 
6807 	if (!ops->cpu_has_kvm_support()) {
6808 		printk(KERN_ERR "kvm: no hardware support\n");
6809 		r = -EOPNOTSUPP;
6810 		goto out;
6811 	}
6812 	if (ops->disabled_by_bios()) {
6813 		printk(KERN_ERR "kvm: disabled by bios\n");
6814 		r = -EOPNOTSUPP;
6815 		goto out;
6816 	}
6817 
6818 	r = -ENOMEM;
6819 	shared_msrs = alloc_percpu(struct kvm_shared_msrs);
6820 	if (!shared_msrs) {
6821 		printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
6822 		goto out;
6823 	}
6824 
6825 	r = kvm_mmu_module_init();
6826 	if (r)
6827 		goto out_free_percpu;
6828 
6829 	kvm_set_mmio_spte_mask();
6830 
6831 	kvm_x86_ops = ops;
6832 
6833 	kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
6834 			PT_DIRTY_MASK, PT64_NX_MASK, 0,
6835 			PT_PRESENT_MASK, 0, sme_me_mask);
6836 	kvm_timer_init();
6837 
6838 	perf_register_guest_info_callbacks(&kvm_guest_cbs);
6839 
6840 	if (boot_cpu_has(X86_FEATURE_XSAVE))
6841 		host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
6842 
6843 	kvm_lapic_init();
6844 #ifdef CONFIG_X86_64
6845 	pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
6846 
6847 	if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
6848 		set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
6849 #endif
6850 
6851 	return 0;
6852 
6853 out_free_percpu:
6854 	free_percpu(shared_msrs);
6855 out:
6856 	return r;
6857 }
6858 
6859 void kvm_arch_exit(void)
6860 {
6861 #ifdef CONFIG_X86_64
6862 	if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
6863 		clear_hv_tscchange_cb();
6864 #endif
6865 	kvm_lapic_exit();
6866 	perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
6867 
6868 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
6869 		cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
6870 					    CPUFREQ_TRANSITION_NOTIFIER);
6871 	cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
6872 #ifdef CONFIG_X86_64
6873 	pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
6874 #endif
6875 	kvm_x86_ops = NULL;
6876 	kvm_mmu_module_exit();
6877 	free_percpu(shared_msrs);
6878 }
6879 
6880 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
6881 {
6882 	++vcpu->stat.halt_exits;
6883 	if (lapic_in_kernel(vcpu)) {
6884 		vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
6885 		return 1;
6886 	} else {
6887 		vcpu->run->exit_reason = KVM_EXIT_HLT;
6888 		return 0;
6889 	}
6890 }
6891 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
6892 
6893 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
6894 {
6895 	int ret = kvm_skip_emulated_instruction(vcpu);
6896 	/*
6897 	 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
6898 	 * KVM_EXIT_DEBUG here.
6899 	 */
6900 	return kvm_vcpu_halt(vcpu) && ret;
6901 }
6902 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
6903 
6904 #ifdef CONFIG_X86_64
6905 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
6906 			        unsigned long clock_type)
6907 {
6908 	struct kvm_clock_pairing clock_pairing;
6909 	struct timespec64 ts;
6910 	u64 cycle;
6911 	int ret;
6912 
6913 	if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
6914 		return -KVM_EOPNOTSUPP;
6915 
6916 	if (kvm_get_walltime_and_clockread(&ts, &cycle) == false)
6917 		return -KVM_EOPNOTSUPP;
6918 
6919 	clock_pairing.sec = ts.tv_sec;
6920 	clock_pairing.nsec = ts.tv_nsec;
6921 	clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
6922 	clock_pairing.flags = 0;
6923 	memset(&clock_pairing.pad, 0, sizeof(clock_pairing.pad));
6924 
6925 	ret = 0;
6926 	if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
6927 			    sizeof(struct kvm_clock_pairing)))
6928 		ret = -KVM_EFAULT;
6929 
6930 	return ret;
6931 }
6932 #endif
6933 
6934 /*
6935  * kvm_pv_kick_cpu_op:  Kick a vcpu.
6936  *
6937  * @apicid - apicid of vcpu to be kicked.
6938  */
6939 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
6940 {
6941 	struct kvm_lapic_irq lapic_irq;
6942 
6943 	lapic_irq.shorthand = 0;
6944 	lapic_irq.dest_mode = 0;
6945 	lapic_irq.level = 0;
6946 	lapic_irq.dest_id = apicid;
6947 	lapic_irq.msi_redir_hint = false;
6948 
6949 	lapic_irq.delivery_mode = APIC_DM_REMRD;
6950 	kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
6951 }
6952 
6953 void kvm_vcpu_deactivate_apicv(struct kvm_vcpu *vcpu)
6954 {
6955 	vcpu->arch.apicv_active = false;
6956 	kvm_x86_ops->refresh_apicv_exec_ctrl(vcpu);
6957 }
6958 
6959 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
6960 {
6961 	unsigned long nr, a0, a1, a2, a3, ret;
6962 	int op_64_bit;
6963 
6964 	if (kvm_hv_hypercall_enabled(vcpu->kvm))
6965 		return kvm_hv_hypercall(vcpu);
6966 
6967 	nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
6968 	a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
6969 	a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
6970 	a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
6971 	a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
6972 
6973 	trace_kvm_hypercall(nr, a0, a1, a2, a3);
6974 
6975 	op_64_bit = is_64_bit_mode(vcpu);
6976 	if (!op_64_bit) {
6977 		nr &= 0xFFFFFFFF;
6978 		a0 &= 0xFFFFFFFF;
6979 		a1 &= 0xFFFFFFFF;
6980 		a2 &= 0xFFFFFFFF;
6981 		a3 &= 0xFFFFFFFF;
6982 	}
6983 
6984 	if (kvm_x86_ops->get_cpl(vcpu) != 0) {
6985 		ret = -KVM_EPERM;
6986 		goto out;
6987 	}
6988 
6989 	switch (nr) {
6990 	case KVM_HC_VAPIC_POLL_IRQ:
6991 		ret = 0;
6992 		break;
6993 	case KVM_HC_KICK_CPU:
6994 		kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
6995 		ret = 0;
6996 		break;
6997 #ifdef CONFIG_X86_64
6998 	case KVM_HC_CLOCK_PAIRING:
6999 		ret = kvm_pv_clock_pairing(vcpu, a0, a1);
7000 		break;
7001 	case KVM_HC_SEND_IPI:
7002 		ret = kvm_pv_send_ipi(vcpu->kvm, a0, a1, a2, a3, op_64_bit);
7003 		break;
7004 #endif
7005 	default:
7006 		ret = -KVM_ENOSYS;
7007 		break;
7008 	}
7009 out:
7010 	if (!op_64_bit)
7011 		ret = (u32)ret;
7012 	kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
7013 
7014 	++vcpu->stat.hypercalls;
7015 	return kvm_skip_emulated_instruction(vcpu);
7016 }
7017 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
7018 
7019 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
7020 {
7021 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7022 	char instruction[3];
7023 	unsigned long rip = kvm_rip_read(vcpu);
7024 
7025 	kvm_x86_ops->patch_hypercall(vcpu, instruction);
7026 
7027 	return emulator_write_emulated(ctxt, rip, instruction, 3,
7028 		&ctxt->exception);
7029 }
7030 
7031 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
7032 {
7033 	return vcpu->run->request_interrupt_window &&
7034 		likely(!pic_in_kernel(vcpu->kvm));
7035 }
7036 
7037 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
7038 {
7039 	struct kvm_run *kvm_run = vcpu->run;
7040 
7041 	kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
7042 	kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
7043 	kvm_run->cr8 = kvm_get_cr8(vcpu);
7044 	kvm_run->apic_base = kvm_get_apic_base(vcpu);
7045 	kvm_run->ready_for_interrupt_injection =
7046 		pic_in_kernel(vcpu->kvm) ||
7047 		kvm_vcpu_ready_for_interrupt_injection(vcpu);
7048 }
7049 
7050 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
7051 {
7052 	int max_irr, tpr;
7053 
7054 	if (!kvm_x86_ops->update_cr8_intercept)
7055 		return;
7056 
7057 	if (!lapic_in_kernel(vcpu))
7058 		return;
7059 
7060 	if (vcpu->arch.apicv_active)
7061 		return;
7062 
7063 	if (!vcpu->arch.apic->vapic_addr)
7064 		max_irr = kvm_lapic_find_highest_irr(vcpu);
7065 	else
7066 		max_irr = -1;
7067 
7068 	if (max_irr != -1)
7069 		max_irr >>= 4;
7070 
7071 	tpr = kvm_lapic_get_cr8(vcpu);
7072 
7073 	kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
7074 }
7075 
7076 static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
7077 {
7078 	int r;
7079 
7080 	/* try to reinject previous events if any */
7081 
7082 	if (vcpu->arch.exception.injected)
7083 		kvm_x86_ops->queue_exception(vcpu);
7084 	/*
7085 	 * Do not inject an NMI or interrupt if there is a pending
7086 	 * exception.  Exceptions and interrupts are recognized at
7087 	 * instruction boundaries, i.e. the start of an instruction.
7088 	 * Trap-like exceptions, e.g. #DB, have higher priority than
7089 	 * NMIs and interrupts, i.e. traps are recognized before an
7090 	 * NMI/interrupt that's pending on the same instruction.
7091 	 * Fault-like exceptions, e.g. #GP and #PF, are the lowest
7092 	 * priority, but are only generated (pended) during instruction
7093 	 * execution, i.e. a pending fault-like exception means the
7094 	 * fault occurred on the *previous* instruction and must be
7095 	 * serviced prior to recognizing any new events in order to
7096 	 * fully complete the previous instruction.
7097 	 */
7098 	else if (!vcpu->arch.exception.pending) {
7099 		if (vcpu->arch.nmi_injected)
7100 			kvm_x86_ops->set_nmi(vcpu);
7101 		else if (vcpu->arch.interrupt.injected)
7102 			kvm_x86_ops->set_irq(vcpu);
7103 	}
7104 
7105 	/*
7106 	 * Call check_nested_events() even if we reinjected a previous event
7107 	 * in order for caller to determine if it should require immediate-exit
7108 	 * from L2 to L1 due to pending L1 events which require exit
7109 	 * from L2 to L1.
7110 	 */
7111 	if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
7112 		r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
7113 		if (r != 0)
7114 			return r;
7115 	}
7116 
7117 	/* try to inject new event if pending */
7118 	if (vcpu->arch.exception.pending) {
7119 		trace_kvm_inj_exception(vcpu->arch.exception.nr,
7120 					vcpu->arch.exception.has_error_code,
7121 					vcpu->arch.exception.error_code);
7122 
7123 		WARN_ON_ONCE(vcpu->arch.exception.injected);
7124 		vcpu->arch.exception.pending = false;
7125 		vcpu->arch.exception.injected = true;
7126 
7127 		if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
7128 			__kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
7129 					     X86_EFLAGS_RF);
7130 
7131 		if (vcpu->arch.exception.nr == DB_VECTOR) {
7132 			/*
7133 			 * This code assumes that nSVM doesn't use
7134 			 * check_nested_events(). If it does, the
7135 			 * DR6/DR7 changes should happen before L1
7136 			 * gets a #VMEXIT for an intercepted #DB in
7137 			 * L2.  (Under VMX, on the other hand, the
7138 			 * DR6/DR7 changes should not happen in the
7139 			 * event of a VM-exit to L1 for an intercepted
7140 			 * #DB in L2.)
7141 			 */
7142 			kvm_deliver_exception_payload(vcpu);
7143 			if (vcpu->arch.dr7 & DR7_GD) {
7144 				vcpu->arch.dr7 &= ~DR7_GD;
7145 				kvm_update_dr7(vcpu);
7146 			}
7147 		}
7148 
7149 		kvm_x86_ops->queue_exception(vcpu);
7150 	}
7151 
7152 	/* Don't consider new event if we re-injected an event */
7153 	if (kvm_event_needs_reinjection(vcpu))
7154 		return 0;
7155 
7156 	if (vcpu->arch.smi_pending && !is_smm(vcpu) &&
7157 	    kvm_x86_ops->smi_allowed(vcpu)) {
7158 		vcpu->arch.smi_pending = false;
7159 		++vcpu->arch.smi_count;
7160 		enter_smm(vcpu);
7161 	} else if (vcpu->arch.nmi_pending && kvm_x86_ops->nmi_allowed(vcpu)) {
7162 		--vcpu->arch.nmi_pending;
7163 		vcpu->arch.nmi_injected = true;
7164 		kvm_x86_ops->set_nmi(vcpu);
7165 	} else if (kvm_cpu_has_injectable_intr(vcpu)) {
7166 		/*
7167 		 * Because interrupts can be injected asynchronously, we are
7168 		 * calling check_nested_events again here to avoid a race condition.
7169 		 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
7170 		 * proposal and current concerns.  Perhaps we should be setting
7171 		 * KVM_REQ_EVENT only on certain events and not unconditionally?
7172 		 */
7173 		if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
7174 			r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
7175 			if (r != 0)
7176 				return r;
7177 		}
7178 		if (kvm_x86_ops->interrupt_allowed(vcpu)) {
7179 			kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
7180 					    false);
7181 			kvm_x86_ops->set_irq(vcpu);
7182 		}
7183 	}
7184 
7185 	return 0;
7186 }
7187 
7188 static void process_nmi(struct kvm_vcpu *vcpu)
7189 {
7190 	unsigned limit = 2;
7191 
7192 	/*
7193 	 * x86 is limited to one NMI running, and one NMI pending after it.
7194 	 * If an NMI is already in progress, limit further NMIs to just one.
7195 	 * Otherwise, allow two (and we'll inject the first one immediately).
7196 	 */
7197 	if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
7198 		limit = 1;
7199 
7200 	vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
7201 	vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
7202 	kvm_make_request(KVM_REQ_EVENT, vcpu);
7203 }
7204 
7205 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
7206 {
7207 	u32 flags = 0;
7208 	flags |= seg->g       << 23;
7209 	flags |= seg->db      << 22;
7210 	flags |= seg->l       << 21;
7211 	flags |= seg->avl     << 20;
7212 	flags |= seg->present << 15;
7213 	flags |= seg->dpl     << 13;
7214 	flags |= seg->s       << 12;
7215 	flags |= seg->type    << 8;
7216 	return flags;
7217 }
7218 
7219 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
7220 {
7221 	struct kvm_segment seg;
7222 	int offset;
7223 
7224 	kvm_get_segment(vcpu, &seg, n);
7225 	put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
7226 
7227 	if (n < 3)
7228 		offset = 0x7f84 + n * 12;
7229 	else
7230 		offset = 0x7f2c + (n - 3) * 12;
7231 
7232 	put_smstate(u32, buf, offset + 8, seg.base);
7233 	put_smstate(u32, buf, offset + 4, seg.limit);
7234 	put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
7235 }
7236 
7237 #ifdef CONFIG_X86_64
7238 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
7239 {
7240 	struct kvm_segment seg;
7241 	int offset;
7242 	u16 flags;
7243 
7244 	kvm_get_segment(vcpu, &seg, n);
7245 	offset = 0x7e00 + n * 16;
7246 
7247 	flags = enter_smm_get_segment_flags(&seg) >> 8;
7248 	put_smstate(u16, buf, offset, seg.selector);
7249 	put_smstate(u16, buf, offset + 2, flags);
7250 	put_smstate(u32, buf, offset + 4, seg.limit);
7251 	put_smstate(u64, buf, offset + 8, seg.base);
7252 }
7253 #endif
7254 
7255 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
7256 {
7257 	struct desc_ptr dt;
7258 	struct kvm_segment seg;
7259 	unsigned long val;
7260 	int i;
7261 
7262 	put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
7263 	put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
7264 	put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
7265 	put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
7266 
7267 	for (i = 0; i < 8; i++)
7268 		put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
7269 
7270 	kvm_get_dr(vcpu, 6, &val);
7271 	put_smstate(u32, buf, 0x7fcc, (u32)val);
7272 	kvm_get_dr(vcpu, 7, &val);
7273 	put_smstate(u32, buf, 0x7fc8, (u32)val);
7274 
7275 	kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
7276 	put_smstate(u32, buf, 0x7fc4, seg.selector);
7277 	put_smstate(u32, buf, 0x7f64, seg.base);
7278 	put_smstate(u32, buf, 0x7f60, seg.limit);
7279 	put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
7280 
7281 	kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
7282 	put_smstate(u32, buf, 0x7fc0, seg.selector);
7283 	put_smstate(u32, buf, 0x7f80, seg.base);
7284 	put_smstate(u32, buf, 0x7f7c, seg.limit);
7285 	put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
7286 
7287 	kvm_x86_ops->get_gdt(vcpu, &dt);
7288 	put_smstate(u32, buf, 0x7f74, dt.address);
7289 	put_smstate(u32, buf, 0x7f70, dt.size);
7290 
7291 	kvm_x86_ops->get_idt(vcpu, &dt);
7292 	put_smstate(u32, buf, 0x7f58, dt.address);
7293 	put_smstate(u32, buf, 0x7f54, dt.size);
7294 
7295 	for (i = 0; i < 6; i++)
7296 		enter_smm_save_seg_32(vcpu, buf, i);
7297 
7298 	put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
7299 
7300 	/* revision id */
7301 	put_smstate(u32, buf, 0x7efc, 0x00020000);
7302 	put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
7303 }
7304 
7305 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
7306 {
7307 #ifdef CONFIG_X86_64
7308 	struct desc_ptr dt;
7309 	struct kvm_segment seg;
7310 	unsigned long val;
7311 	int i;
7312 
7313 	for (i = 0; i < 16; i++)
7314 		put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
7315 
7316 	put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
7317 	put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
7318 
7319 	kvm_get_dr(vcpu, 6, &val);
7320 	put_smstate(u64, buf, 0x7f68, val);
7321 	kvm_get_dr(vcpu, 7, &val);
7322 	put_smstate(u64, buf, 0x7f60, val);
7323 
7324 	put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
7325 	put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
7326 	put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
7327 
7328 	put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
7329 
7330 	/* revision id */
7331 	put_smstate(u32, buf, 0x7efc, 0x00020064);
7332 
7333 	put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
7334 
7335 	kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
7336 	put_smstate(u16, buf, 0x7e90, seg.selector);
7337 	put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
7338 	put_smstate(u32, buf, 0x7e94, seg.limit);
7339 	put_smstate(u64, buf, 0x7e98, seg.base);
7340 
7341 	kvm_x86_ops->get_idt(vcpu, &dt);
7342 	put_smstate(u32, buf, 0x7e84, dt.size);
7343 	put_smstate(u64, buf, 0x7e88, dt.address);
7344 
7345 	kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
7346 	put_smstate(u16, buf, 0x7e70, seg.selector);
7347 	put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
7348 	put_smstate(u32, buf, 0x7e74, seg.limit);
7349 	put_smstate(u64, buf, 0x7e78, seg.base);
7350 
7351 	kvm_x86_ops->get_gdt(vcpu, &dt);
7352 	put_smstate(u32, buf, 0x7e64, dt.size);
7353 	put_smstate(u64, buf, 0x7e68, dt.address);
7354 
7355 	for (i = 0; i < 6; i++)
7356 		enter_smm_save_seg_64(vcpu, buf, i);
7357 #else
7358 	WARN_ON_ONCE(1);
7359 #endif
7360 }
7361 
7362 static void enter_smm(struct kvm_vcpu *vcpu)
7363 {
7364 	struct kvm_segment cs, ds;
7365 	struct desc_ptr dt;
7366 	char buf[512];
7367 	u32 cr0;
7368 
7369 	trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
7370 	memset(buf, 0, 512);
7371 	if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
7372 		enter_smm_save_state_64(vcpu, buf);
7373 	else
7374 		enter_smm_save_state_32(vcpu, buf);
7375 
7376 	/*
7377 	 * Give pre_enter_smm() a chance to make ISA-specific changes to the
7378 	 * vCPU state (e.g. leave guest mode) after we've saved the state into
7379 	 * the SMM state-save area.
7380 	 */
7381 	kvm_x86_ops->pre_enter_smm(vcpu, buf);
7382 
7383 	vcpu->arch.hflags |= HF_SMM_MASK;
7384 	kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
7385 
7386 	if (kvm_x86_ops->get_nmi_mask(vcpu))
7387 		vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
7388 	else
7389 		kvm_x86_ops->set_nmi_mask(vcpu, true);
7390 
7391 	kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
7392 	kvm_rip_write(vcpu, 0x8000);
7393 
7394 	cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
7395 	kvm_x86_ops->set_cr0(vcpu, cr0);
7396 	vcpu->arch.cr0 = cr0;
7397 
7398 	kvm_x86_ops->set_cr4(vcpu, 0);
7399 
7400 	/* Undocumented: IDT limit is set to zero on entry to SMM.  */
7401 	dt.address = dt.size = 0;
7402 	kvm_x86_ops->set_idt(vcpu, &dt);
7403 
7404 	__kvm_set_dr(vcpu, 7, DR7_FIXED_1);
7405 
7406 	cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
7407 	cs.base = vcpu->arch.smbase;
7408 
7409 	ds.selector = 0;
7410 	ds.base = 0;
7411 
7412 	cs.limit    = ds.limit = 0xffffffff;
7413 	cs.type     = ds.type = 0x3;
7414 	cs.dpl      = ds.dpl = 0;
7415 	cs.db       = ds.db = 0;
7416 	cs.s        = ds.s = 1;
7417 	cs.l        = ds.l = 0;
7418 	cs.g        = ds.g = 1;
7419 	cs.avl      = ds.avl = 0;
7420 	cs.present  = ds.present = 1;
7421 	cs.unusable = ds.unusable = 0;
7422 	cs.padding  = ds.padding = 0;
7423 
7424 	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
7425 	kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
7426 	kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
7427 	kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
7428 	kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
7429 	kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
7430 
7431 	if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
7432 		kvm_x86_ops->set_efer(vcpu, 0);
7433 
7434 	kvm_update_cpuid(vcpu);
7435 	kvm_mmu_reset_context(vcpu);
7436 }
7437 
7438 static void process_smi(struct kvm_vcpu *vcpu)
7439 {
7440 	vcpu->arch.smi_pending = true;
7441 	kvm_make_request(KVM_REQ_EVENT, vcpu);
7442 }
7443 
7444 void kvm_make_scan_ioapic_request(struct kvm *kvm)
7445 {
7446 	kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
7447 }
7448 
7449 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
7450 {
7451 	if (!kvm_apic_present(vcpu))
7452 		return;
7453 
7454 	bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
7455 
7456 	if (irqchip_split(vcpu->kvm))
7457 		kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
7458 	else {
7459 		if (vcpu->arch.apicv_active)
7460 			kvm_x86_ops->sync_pir_to_irr(vcpu);
7461 		if (ioapic_in_kernel(vcpu->kvm))
7462 			kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
7463 	}
7464 
7465 	if (is_guest_mode(vcpu))
7466 		vcpu->arch.load_eoi_exitmap_pending = true;
7467 	else
7468 		kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
7469 }
7470 
7471 static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
7472 {
7473 	u64 eoi_exit_bitmap[4];
7474 
7475 	if (!kvm_apic_hw_enabled(vcpu->arch.apic))
7476 		return;
7477 
7478 	bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors,
7479 		  vcpu_to_synic(vcpu)->vec_bitmap, 256);
7480 	kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
7481 }
7482 
7483 int kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
7484 		unsigned long start, unsigned long end,
7485 		bool blockable)
7486 {
7487 	unsigned long apic_address;
7488 
7489 	/*
7490 	 * The physical address of apic access page is stored in the VMCS.
7491 	 * Update it when it becomes invalid.
7492 	 */
7493 	apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
7494 	if (start <= apic_address && apic_address < end)
7495 		kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
7496 
7497 	return 0;
7498 }
7499 
7500 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
7501 {
7502 	struct page *page = NULL;
7503 
7504 	if (!lapic_in_kernel(vcpu))
7505 		return;
7506 
7507 	if (!kvm_x86_ops->set_apic_access_page_addr)
7508 		return;
7509 
7510 	page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
7511 	if (is_error_page(page))
7512 		return;
7513 	kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page));
7514 
7515 	/*
7516 	 * Do not pin apic access page in memory, the MMU notifier
7517 	 * will call us again if it is migrated or swapped out.
7518 	 */
7519 	put_page(page);
7520 }
7521 EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page);
7522 
7523 void __kvm_request_immediate_exit(struct kvm_vcpu *vcpu)
7524 {
7525 	smp_send_reschedule(vcpu->cpu);
7526 }
7527 EXPORT_SYMBOL_GPL(__kvm_request_immediate_exit);
7528 
7529 /*
7530  * Returns 1 to let vcpu_run() continue the guest execution loop without
7531  * exiting to the userspace.  Otherwise, the value will be returned to the
7532  * userspace.
7533  */
7534 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
7535 {
7536 	int r;
7537 	bool req_int_win =
7538 		dm_request_for_irq_injection(vcpu) &&
7539 		kvm_cpu_accept_dm_intr(vcpu);
7540 
7541 	bool req_immediate_exit = false;
7542 
7543 	if (kvm_request_pending(vcpu)) {
7544 		if (kvm_check_request(KVM_REQ_GET_VMCS12_PAGES, vcpu))
7545 			kvm_x86_ops->get_vmcs12_pages(vcpu);
7546 		if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
7547 			kvm_mmu_unload(vcpu);
7548 		if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
7549 			__kvm_migrate_timers(vcpu);
7550 		if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
7551 			kvm_gen_update_masterclock(vcpu->kvm);
7552 		if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
7553 			kvm_gen_kvmclock_update(vcpu);
7554 		if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
7555 			r = kvm_guest_time_update(vcpu);
7556 			if (unlikely(r))
7557 				goto out;
7558 		}
7559 		if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
7560 			kvm_mmu_sync_roots(vcpu);
7561 		if (kvm_check_request(KVM_REQ_LOAD_CR3, vcpu))
7562 			kvm_mmu_load_cr3(vcpu);
7563 		if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
7564 			kvm_vcpu_flush_tlb(vcpu, true);
7565 		if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
7566 			vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
7567 			r = 0;
7568 			goto out;
7569 		}
7570 		if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
7571 			vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
7572 			vcpu->mmio_needed = 0;
7573 			r = 0;
7574 			goto out;
7575 		}
7576 		if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
7577 			/* Page is swapped out. Do synthetic halt */
7578 			vcpu->arch.apf.halted = true;
7579 			r = 1;
7580 			goto out;
7581 		}
7582 		if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
7583 			record_steal_time(vcpu);
7584 		if (kvm_check_request(KVM_REQ_SMI, vcpu))
7585 			process_smi(vcpu);
7586 		if (kvm_check_request(KVM_REQ_NMI, vcpu))
7587 			process_nmi(vcpu);
7588 		if (kvm_check_request(KVM_REQ_PMU, vcpu))
7589 			kvm_pmu_handle_event(vcpu);
7590 		if (kvm_check_request(KVM_REQ_PMI, vcpu))
7591 			kvm_pmu_deliver_pmi(vcpu);
7592 		if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
7593 			BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
7594 			if (test_bit(vcpu->arch.pending_ioapic_eoi,
7595 				     vcpu->arch.ioapic_handled_vectors)) {
7596 				vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
7597 				vcpu->run->eoi.vector =
7598 						vcpu->arch.pending_ioapic_eoi;
7599 				r = 0;
7600 				goto out;
7601 			}
7602 		}
7603 		if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
7604 			vcpu_scan_ioapic(vcpu);
7605 		if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
7606 			vcpu_load_eoi_exitmap(vcpu);
7607 		if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
7608 			kvm_vcpu_reload_apic_access_page(vcpu);
7609 		if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
7610 			vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
7611 			vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
7612 			r = 0;
7613 			goto out;
7614 		}
7615 		if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
7616 			vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
7617 			vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
7618 			r = 0;
7619 			goto out;
7620 		}
7621 		if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
7622 			vcpu->run->exit_reason = KVM_EXIT_HYPERV;
7623 			vcpu->run->hyperv = vcpu->arch.hyperv.exit;
7624 			r = 0;
7625 			goto out;
7626 		}
7627 
7628 		/*
7629 		 * KVM_REQ_HV_STIMER has to be processed after
7630 		 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
7631 		 * depend on the guest clock being up-to-date
7632 		 */
7633 		if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
7634 			kvm_hv_process_stimers(vcpu);
7635 	}
7636 
7637 	if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
7638 		++vcpu->stat.req_event;
7639 		kvm_apic_accept_events(vcpu);
7640 		if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
7641 			r = 1;
7642 			goto out;
7643 		}
7644 
7645 		if (inject_pending_event(vcpu, req_int_win) != 0)
7646 			req_immediate_exit = true;
7647 		else {
7648 			/* Enable SMI/NMI/IRQ window open exits if needed.
7649 			 *
7650 			 * SMIs have three cases:
7651 			 * 1) They can be nested, and then there is nothing to
7652 			 *    do here because RSM will cause a vmexit anyway.
7653 			 * 2) There is an ISA-specific reason why SMI cannot be
7654 			 *    injected, and the moment when this changes can be
7655 			 *    intercepted.
7656 			 * 3) Or the SMI can be pending because
7657 			 *    inject_pending_event has completed the injection
7658 			 *    of an IRQ or NMI from the previous vmexit, and
7659 			 *    then we request an immediate exit to inject the
7660 			 *    SMI.
7661 			 */
7662 			if (vcpu->arch.smi_pending && !is_smm(vcpu))
7663 				if (!kvm_x86_ops->enable_smi_window(vcpu))
7664 					req_immediate_exit = true;
7665 			if (vcpu->arch.nmi_pending)
7666 				kvm_x86_ops->enable_nmi_window(vcpu);
7667 			if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
7668 				kvm_x86_ops->enable_irq_window(vcpu);
7669 			WARN_ON(vcpu->arch.exception.pending);
7670 		}
7671 
7672 		if (kvm_lapic_enabled(vcpu)) {
7673 			update_cr8_intercept(vcpu);
7674 			kvm_lapic_sync_to_vapic(vcpu);
7675 		}
7676 	}
7677 
7678 	r = kvm_mmu_reload(vcpu);
7679 	if (unlikely(r)) {
7680 		goto cancel_injection;
7681 	}
7682 
7683 	preempt_disable();
7684 
7685 	kvm_x86_ops->prepare_guest_switch(vcpu);
7686 
7687 	/*
7688 	 * Disable IRQs before setting IN_GUEST_MODE.  Posted interrupt
7689 	 * IPI are then delayed after guest entry, which ensures that they
7690 	 * result in virtual interrupt delivery.
7691 	 */
7692 	local_irq_disable();
7693 	vcpu->mode = IN_GUEST_MODE;
7694 
7695 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
7696 
7697 	/*
7698 	 * 1) We should set ->mode before checking ->requests.  Please see
7699 	 * the comment in kvm_vcpu_exiting_guest_mode().
7700 	 *
7701 	 * 2) For APICv, we should set ->mode before checking PIR.ON.  This
7702 	 * pairs with the memory barrier implicit in pi_test_and_set_on
7703 	 * (see vmx_deliver_posted_interrupt).
7704 	 *
7705 	 * 3) This also orders the write to mode from any reads to the page
7706 	 * tables done while the VCPU is running.  Please see the comment
7707 	 * in kvm_flush_remote_tlbs.
7708 	 */
7709 	smp_mb__after_srcu_read_unlock();
7710 
7711 	/*
7712 	 * This handles the case where a posted interrupt was
7713 	 * notified with kvm_vcpu_kick.
7714 	 */
7715 	if (kvm_lapic_enabled(vcpu) && vcpu->arch.apicv_active)
7716 		kvm_x86_ops->sync_pir_to_irr(vcpu);
7717 
7718 	if (vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu)
7719 	    || need_resched() || signal_pending(current)) {
7720 		vcpu->mode = OUTSIDE_GUEST_MODE;
7721 		smp_wmb();
7722 		local_irq_enable();
7723 		preempt_enable();
7724 		vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
7725 		r = 1;
7726 		goto cancel_injection;
7727 	}
7728 
7729 	kvm_load_guest_xcr0(vcpu);
7730 
7731 	if (req_immediate_exit) {
7732 		kvm_make_request(KVM_REQ_EVENT, vcpu);
7733 		kvm_x86_ops->request_immediate_exit(vcpu);
7734 	}
7735 
7736 	trace_kvm_entry(vcpu->vcpu_id);
7737 	if (lapic_timer_advance_ns)
7738 		wait_lapic_expire(vcpu);
7739 	guest_enter_irqoff();
7740 
7741 	if (unlikely(vcpu->arch.switch_db_regs)) {
7742 		set_debugreg(0, 7);
7743 		set_debugreg(vcpu->arch.eff_db[0], 0);
7744 		set_debugreg(vcpu->arch.eff_db[1], 1);
7745 		set_debugreg(vcpu->arch.eff_db[2], 2);
7746 		set_debugreg(vcpu->arch.eff_db[3], 3);
7747 		set_debugreg(vcpu->arch.dr6, 6);
7748 		vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
7749 	}
7750 
7751 	kvm_x86_ops->run(vcpu);
7752 
7753 	/*
7754 	 * Do this here before restoring debug registers on the host.  And
7755 	 * since we do this before handling the vmexit, a DR access vmexit
7756 	 * can (a) read the correct value of the debug registers, (b) set
7757 	 * KVM_DEBUGREG_WONT_EXIT again.
7758 	 */
7759 	if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
7760 		WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
7761 		kvm_x86_ops->sync_dirty_debug_regs(vcpu);
7762 		kvm_update_dr0123(vcpu);
7763 		kvm_update_dr6(vcpu);
7764 		kvm_update_dr7(vcpu);
7765 		vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
7766 	}
7767 
7768 	/*
7769 	 * If the guest has used debug registers, at least dr7
7770 	 * will be disabled while returning to the host.
7771 	 * If we don't have active breakpoints in the host, we don't
7772 	 * care about the messed up debug address registers. But if
7773 	 * we have some of them active, restore the old state.
7774 	 */
7775 	if (hw_breakpoint_active())
7776 		hw_breakpoint_restore();
7777 
7778 	vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
7779 
7780 	vcpu->mode = OUTSIDE_GUEST_MODE;
7781 	smp_wmb();
7782 
7783 	kvm_put_guest_xcr0(vcpu);
7784 
7785 	kvm_before_interrupt(vcpu);
7786 	kvm_x86_ops->handle_external_intr(vcpu);
7787 	kvm_after_interrupt(vcpu);
7788 
7789 	++vcpu->stat.exits;
7790 
7791 	guest_exit_irqoff();
7792 
7793 	local_irq_enable();
7794 	preempt_enable();
7795 
7796 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
7797 
7798 	/*
7799 	 * Profile KVM exit RIPs:
7800 	 */
7801 	if (unlikely(prof_on == KVM_PROFILING)) {
7802 		unsigned long rip = kvm_rip_read(vcpu);
7803 		profile_hit(KVM_PROFILING, (void *)rip);
7804 	}
7805 
7806 	if (unlikely(vcpu->arch.tsc_always_catchup))
7807 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7808 
7809 	if (vcpu->arch.apic_attention)
7810 		kvm_lapic_sync_from_vapic(vcpu);
7811 
7812 	vcpu->arch.gpa_available = false;
7813 	r = kvm_x86_ops->handle_exit(vcpu);
7814 	return r;
7815 
7816 cancel_injection:
7817 	kvm_x86_ops->cancel_injection(vcpu);
7818 	if (unlikely(vcpu->arch.apic_attention))
7819 		kvm_lapic_sync_from_vapic(vcpu);
7820 out:
7821 	return r;
7822 }
7823 
7824 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
7825 {
7826 	if (!kvm_arch_vcpu_runnable(vcpu) &&
7827 	    (!kvm_x86_ops->pre_block || kvm_x86_ops->pre_block(vcpu) == 0)) {
7828 		srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
7829 		kvm_vcpu_block(vcpu);
7830 		vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
7831 
7832 		if (kvm_x86_ops->post_block)
7833 			kvm_x86_ops->post_block(vcpu);
7834 
7835 		if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
7836 			return 1;
7837 	}
7838 
7839 	kvm_apic_accept_events(vcpu);
7840 	switch(vcpu->arch.mp_state) {
7841 	case KVM_MP_STATE_HALTED:
7842 		vcpu->arch.pv.pv_unhalted = false;
7843 		vcpu->arch.mp_state =
7844 			KVM_MP_STATE_RUNNABLE;
7845 	case KVM_MP_STATE_RUNNABLE:
7846 		vcpu->arch.apf.halted = false;
7847 		break;
7848 	case KVM_MP_STATE_INIT_RECEIVED:
7849 		break;
7850 	default:
7851 		return -EINTR;
7852 		break;
7853 	}
7854 	return 1;
7855 }
7856 
7857 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
7858 {
7859 	if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
7860 		kvm_x86_ops->check_nested_events(vcpu, false);
7861 
7862 	return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
7863 		!vcpu->arch.apf.halted);
7864 }
7865 
7866 static int vcpu_run(struct kvm_vcpu *vcpu)
7867 {
7868 	int r;
7869 	struct kvm *kvm = vcpu->kvm;
7870 
7871 	vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
7872 	vcpu->arch.l1tf_flush_l1d = true;
7873 
7874 	for (;;) {
7875 		if (kvm_vcpu_running(vcpu)) {
7876 			r = vcpu_enter_guest(vcpu);
7877 		} else {
7878 			r = vcpu_block(kvm, vcpu);
7879 		}
7880 
7881 		if (r <= 0)
7882 			break;
7883 
7884 		kvm_clear_request(KVM_REQ_PENDING_TIMER, vcpu);
7885 		if (kvm_cpu_has_pending_timer(vcpu))
7886 			kvm_inject_pending_timer_irqs(vcpu);
7887 
7888 		if (dm_request_for_irq_injection(vcpu) &&
7889 			kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
7890 			r = 0;
7891 			vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
7892 			++vcpu->stat.request_irq_exits;
7893 			break;
7894 		}
7895 
7896 		kvm_check_async_pf_completion(vcpu);
7897 
7898 		if (signal_pending(current)) {
7899 			r = -EINTR;
7900 			vcpu->run->exit_reason = KVM_EXIT_INTR;
7901 			++vcpu->stat.signal_exits;
7902 			break;
7903 		}
7904 		if (need_resched()) {
7905 			srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
7906 			cond_resched();
7907 			vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
7908 		}
7909 	}
7910 
7911 	srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
7912 
7913 	return r;
7914 }
7915 
7916 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
7917 {
7918 	int r;
7919 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
7920 	r = kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
7921 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
7922 	if (r != EMULATE_DONE)
7923 		return 0;
7924 	return 1;
7925 }
7926 
7927 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
7928 {
7929 	BUG_ON(!vcpu->arch.pio.count);
7930 
7931 	return complete_emulated_io(vcpu);
7932 }
7933 
7934 /*
7935  * Implements the following, as a state machine:
7936  *
7937  * read:
7938  *   for each fragment
7939  *     for each mmio piece in the fragment
7940  *       write gpa, len
7941  *       exit
7942  *       copy data
7943  *   execute insn
7944  *
7945  * write:
7946  *   for each fragment
7947  *     for each mmio piece in the fragment
7948  *       write gpa, len
7949  *       copy data
7950  *       exit
7951  */
7952 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
7953 {
7954 	struct kvm_run *run = vcpu->run;
7955 	struct kvm_mmio_fragment *frag;
7956 	unsigned len;
7957 
7958 	BUG_ON(!vcpu->mmio_needed);
7959 
7960 	/* Complete previous fragment */
7961 	frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
7962 	len = min(8u, frag->len);
7963 	if (!vcpu->mmio_is_write)
7964 		memcpy(frag->data, run->mmio.data, len);
7965 
7966 	if (frag->len <= 8) {
7967 		/* Switch to the next fragment. */
7968 		frag++;
7969 		vcpu->mmio_cur_fragment++;
7970 	} else {
7971 		/* Go forward to the next mmio piece. */
7972 		frag->data += len;
7973 		frag->gpa += len;
7974 		frag->len -= len;
7975 	}
7976 
7977 	if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
7978 		vcpu->mmio_needed = 0;
7979 
7980 		/* FIXME: return into emulator if single-stepping.  */
7981 		if (vcpu->mmio_is_write)
7982 			return 1;
7983 		vcpu->mmio_read_completed = 1;
7984 		return complete_emulated_io(vcpu);
7985 	}
7986 
7987 	run->exit_reason = KVM_EXIT_MMIO;
7988 	run->mmio.phys_addr = frag->gpa;
7989 	if (vcpu->mmio_is_write)
7990 		memcpy(run->mmio.data, frag->data, min(8u, frag->len));
7991 	run->mmio.len = min(8u, frag->len);
7992 	run->mmio.is_write = vcpu->mmio_is_write;
7993 	vcpu->arch.complete_userspace_io = complete_emulated_mmio;
7994 	return 0;
7995 }
7996 
7997 /* Swap (qemu) user FPU context for the guest FPU context. */
7998 static void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
7999 {
8000 	preempt_disable();
8001 	copy_fpregs_to_fpstate(&vcpu->arch.user_fpu);
8002 	/* PKRU is separately restored in kvm_x86_ops->run.  */
8003 	__copy_kernel_to_fpregs(&vcpu->arch.guest_fpu.state,
8004 				~XFEATURE_MASK_PKRU);
8005 	preempt_enable();
8006 	trace_kvm_fpu(1);
8007 }
8008 
8009 /* When vcpu_run ends, restore user space FPU context. */
8010 static void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
8011 {
8012 	preempt_disable();
8013 	copy_fpregs_to_fpstate(&vcpu->arch.guest_fpu);
8014 	copy_kernel_to_fpregs(&vcpu->arch.user_fpu.state);
8015 	preempt_enable();
8016 	++vcpu->stat.fpu_reload;
8017 	trace_kvm_fpu(0);
8018 }
8019 
8020 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
8021 {
8022 	int r;
8023 
8024 	vcpu_load(vcpu);
8025 	kvm_sigset_activate(vcpu);
8026 	kvm_load_guest_fpu(vcpu);
8027 
8028 	if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
8029 		if (kvm_run->immediate_exit) {
8030 			r = -EINTR;
8031 			goto out;
8032 		}
8033 		kvm_vcpu_block(vcpu);
8034 		kvm_apic_accept_events(vcpu);
8035 		kvm_clear_request(KVM_REQ_UNHALT, vcpu);
8036 		r = -EAGAIN;
8037 		if (signal_pending(current)) {
8038 			r = -EINTR;
8039 			vcpu->run->exit_reason = KVM_EXIT_INTR;
8040 			++vcpu->stat.signal_exits;
8041 		}
8042 		goto out;
8043 	}
8044 
8045 	if (vcpu->run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) {
8046 		r = -EINVAL;
8047 		goto out;
8048 	}
8049 
8050 	if (vcpu->run->kvm_dirty_regs) {
8051 		r = sync_regs(vcpu);
8052 		if (r != 0)
8053 			goto out;
8054 	}
8055 
8056 	/* re-sync apic's tpr */
8057 	if (!lapic_in_kernel(vcpu)) {
8058 		if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
8059 			r = -EINVAL;
8060 			goto out;
8061 		}
8062 	}
8063 
8064 	if (unlikely(vcpu->arch.complete_userspace_io)) {
8065 		int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
8066 		vcpu->arch.complete_userspace_io = NULL;
8067 		r = cui(vcpu);
8068 		if (r <= 0)
8069 			goto out;
8070 	} else
8071 		WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
8072 
8073 	if (kvm_run->immediate_exit)
8074 		r = -EINTR;
8075 	else
8076 		r = vcpu_run(vcpu);
8077 
8078 out:
8079 	kvm_put_guest_fpu(vcpu);
8080 	if (vcpu->run->kvm_valid_regs)
8081 		store_regs(vcpu);
8082 	post_kvm_run_save(vcpu);
8083 	kvm_sigset_deactivate(vcpu);
8084 
8085 	vcpu_put(vcpu);
8086 	return r;
8087 }
8088 
8089 static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8090 {
8091 	if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
8092 		/*
8093 		 * We are here if userspace calls get_regs() in the middle of
8094 		 * instruction emulation. Registers state needs to be copied
8095 		 * back from emulation context to vcpu. Userspace shouldn't do
8096 		 * that usually, but some bad designed PV devices (vmware
8097 		 * backdoor interface) need this to work
8098 		 */
8099 		emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
8100 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8101 	}
8102 	regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
8103 	regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
8104 	regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
8105 	regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
8106 	regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
8107 	regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
8108 	regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
8109 	regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
8110 #ifdef CONFIG_X86_64
8111 	regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
8112 	regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
8113 	regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
8114 	regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
8115 	regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
8116 	regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
8117 	regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
8118 	regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
8119 #endif
8120 
8121 	regs->rip = kvm_rip_read(vcpu);
8122 	regs->rflags = kvm_get_rflags(vcpu);
8123 }
8124 
8125 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8126 {
8127 	vcpu_load(vcpu);
8128 	__get_regs(vcpu, regs);
8129 	vcpu_put(vcpu);
8130 	return 0;
8131 }
8132 
8133 static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8134 {
8135 	vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
8136 	vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8137 
8138 	kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
8139 	kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
8140 	kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
8141 	kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
8142 	kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
8143 	kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
8144 	kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
8145 	kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
8146 #ifdef CONFIG_X86_64
8147 	kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
8148 	kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
8149 	kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
8150 	kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
8151 	kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
8152 	kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
8153 	kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
8154 	kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
8155 #endif
8156 
8157 	kvm_rip_write(vcpu, regs->rip);
8158 	kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
8159 
8160 	vcpu->arch.exception.pending = false;
8161 
8162 	kvm_make_request(KVM_REQ_EVENT, vcpu);
8163 }
8164 
8165 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8166 {
8167 	vcpu_load(vcpu);
8168 	__set_regs(vcpu, regs);
8169 	vcpu_put(vcpu);
8170 	return 0;
8171 }
8172 
8173 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
8174 {
8175 	struct kvm_segment cs;
8176 
8177 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
8178 	*db = cs.db;
8179 	*l = cs.l;
8180 }
8181 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
8182 
8183 static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8184 {
8185 	struct desc_ptr dt;
8186 
8187 	kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
8188 	kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
8189 	kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
8190 	kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
8191 	kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
8192 	kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
8193 
8194 	kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
8195 	kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
8196 
8197 	kvm_x86_ops->get_idt(vcpu, &dt);
8198 	sregs->idt.limit = dt.size;
8199 	sregs->idt.base = dt.address;
8200 	kvm_x86_ops->get_gdt(vcpu, &dt);
8201 	sregs->gdt.limit = dt.size;
8202 	sregs->gdt.base = dt.address;
8203 
8204 	sregs->cr0 = kvm_read_cr0(vcpu);
8205 	sregs->cr2 = vcpu->arch.cr2;
8206 	sregs->cr3 = kvm_read_cr3(vcpu);
8207 	sregs->cr4 = kvm_read_cr4(vcpu);
8208 	sregs->cr8 = kvm_get_cr8(vcpu);
8209 	sregs->efer = vcpu->arch.efer;
8210 	sregs->apic_base = kvm_get_apic_base(vcpu);
8211 
8212 	memset(sregs->interrupt_bitmap, 0, sizeof(sregs->interrupt_bitmap));
8213 
8214 	if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft)
8215 		set_bit(vcpu->arch.interrupt.nr,
8216 			(unsigned long *)sregs->interrupt_bitmap);
8217 }
8218 
8219 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
8220 				  struct kvm_sregs *sregs)
8221 {
8222 	vcpu_load(vcpu);
8223 	__get_sregs(vcpu, sregs);
8224 	vcpu_put(vcpu);
8225 	return 0;
8226 }
8227 
8228 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
8229 				    struct kvm_mp_state *mp_state)
8230 {
8231 	vcpu_load(vcpu);
8232 
8233 	kvm_apic_accept_events(vcpu);
8234 	if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
8235 					vcpu->arch.pv.pv_unhalted)
8236 		mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
8237 	else
8238 		mp_state->mp_state = vcpu->arch.mp_state;
8239 
8240 	vcpu_put(vcpu);
8241 	return 0;
8242 }
8243 
8244 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
8245 				    struct kvm_mp_state *mp_state)
8246 {
8247 	int ret = -EINVAL;
8248 
8249 	vcpu_load(vcpu);
8250 
8251 	if (!lapic_in_kernel(vcpu) &&
8252 	    mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
8253 		goto out;
8254 
8255 	/* INITs are latched while in SMM */
8256 	if ((is_smm(vcpu) || vcpu->arch.smi_pending) &&
8257 	    (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
8258 	     mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
8259 		goto out;
8260 
8261 	if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
8262 		vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
8263 		set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
8264 	} else
8265 		vcpu->arch.mp_state = mp_state->mp_state;
8266 	kvm_make_request(KVM_REQ_EVENT, vcpu);
8267 
8268 	ret = 0;
8269 out:
8270 	vcpu_put(vcpu);
8271 	return ret;
8272 }
8273 
8274 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
8275 		    int reason, bool has_error_code, u32 error_code)
8276 {
8277 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
8278 	int ret;
8279 
8280 	init_emulate_ctxt(vcpu);
8281 
8282 	ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
8283 				   has_error_code, error_code);
8284 
8285 	if (ret)
8286 		return EMULATE_FAIL;
8287 
8288 	kvm_rip_write(vcpu, ctxt->eip);
8289 	kvm_set_rflags(vcpu, ctxt->eflags);
8290 	kvm_make_request(KVM_REQ_EVENT, vcpu);
8291 	return EMULATE_DONE;
8292 }
8293 EXPORT_SYMBOL_GPL(kvm_task_switch);
8294 
8295 static int kvm_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8296 {
8297 	if (!guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
8298 			(sregs->cr4 & X86_CR4_OSXSAVE))
8299 		return  -EINVAL;
8300 
8301 	if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
8302 		/*
8303 		 * When EFER.LME and CR0.PG are set, the processor is in
8304 		 * 64-bit mode (though maybe in a 32-bit code segment).
8305 		 * CR4.PAE and EFER.LMA must be set.
8306 		 */
8307 		if (!(sregs->cr4 & X86_CR4_PAE)
8308 		    || !(sregs->efer & EFER_LMA))
8309 			return -EINVAL;
8310 	} else {
8311 		/*
8312 		 * Not in 64-bit mode: EFER.LMA is clear and the code
8313 		 * segment cannot be 64-bit.
8314 		 */
8315 		if (sregs->efer & EFER_LMA || sregs->cs.l)
8316 			return -EINVAL;
8317 	}
8318 
8319 	return 0;
8320 }
8321 
8322 static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8323 {
8324 	struct msr_data apic_base_msr;
8325 	int mmu_reset_needed = 0;
8326 	int cpuid_update_needed = 0;
8327 	int pending_vec, max_bits, idx;
8328 	struct desc_ptr dt;
8329 	int ret = -EINVAL;
8330 
8331 	if (kvm_valid_sregs(vcpu, sregs))
8332 		goto out;
8333 
8334 	apic_base_msr.data = sregs->apic_base;
8335 	apic_base_msr.host_initiated = true;
8336 	if (kvm_set_apic_base(vcpu, &apic_base_msr))
8337 		goto out;
8338 
8339 	dt.size = sregs->idt.limit;
8340 	dt.address = sregs->idt.base;
8341 	kvm_x86_ops->set_idt(vcpu, &dt);
8342 	dt.size = sregs->gdt.limit;
8343 	dt.address = sregs->gdt.base;
8344 	kvm_x86_ops->set_gdt(vcpu, &dt);
8345 
8346 	vcpu->arch.cr2 = sregs->cr2;
8347 	mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
8348 	vcpu->arch.cr3 = sregs->cr3;
8349 	__set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
8350 
8351 	kvm_set_cr8(vcpu, sregs->cr8);
8352 
8353 	mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
8354 	kvm_x86_ops->set_efer(vcpu, sregs->efer);
8355 
8356 	mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
8357 	kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
8358 	vcpu->arch.cr0 = sregs->cr0;
8359 
8360 	mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
8361 	cpuid_update_needed |= ((kvm_read_cr4(vcpu) ^ sregs->cr4) &
8362 				(X86_CR4_OSXSAVE | X86_CR4_PKE));
8363 	kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
8364 	if (cpuid_update_needed)
8365 		kvm_update_cpuid(vcpu);
8366 
8367 	idx = srcu_read_lock(&vcpu->kvm->srcu);
8368 	if (!is_long_mode(vcpu) && is_pae(vcpu) && is_paging(vcpu)) {
8369 		load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
8370 		mmu_reset_needed = 1;
8371 	}
8372 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
8373 
8374 	if (mmu_reset_needed)
8375 		kvm_mmu_reset_context(vcpu);
8376 
8377 	max_bits = KVM_NR_INTERRUPTS;
8378 	pending_vec = find_first_bit(
8379 		(const unsigned long *)sregs->interrupt_bitmap, max_bits);
8380 	if (pending_vec < max_bits) {
8381 		kvm_queue_interrupt(vcpu, pending_vec, false);
8382 		pr_debug("Set back pending irq %d\n", pending_vec);
8383 	}
8384 
8385 	kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
8386 	kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
8387 	kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
8388 	kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
8389 	kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
8390 	kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
8391 
8392 	kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
8393 	kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
8394 
8395 	update_cr8_intercept(vcpu);
8396 
8397 	/* Older userspace won't unhalt the vcpu on reset. */
8398 	if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
8399 	    sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
8400 	    !is_protmode(vcpu))
8401 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
8402 
8403 	kvm_make_request(KVM_REQ_EVENT, vcpu);
8404 
8405 	ret = 0;
8406 out:
8407 	return ret;
8408 }
8409 
8410 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
8411 				  struct kvm_sregs *sregs)
8412 {
8413 	int ret;
8414 
8415 	vcpu_load(vcpu);
8416 	ret = __set_sregs(vcpu, sregs);
8417 	vcpu_put(vcpu);
8418 	return ret;
8419 }
8420 
8421 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
8422 					struct kvm_guest_debug *dbg)
8423 {
8424 	unsigned long rflags;
8425 	int i, r;
8426 
8427 	vcpu_load(vcpu);
8428 
8429 	if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
8430 		r = -EBUSY;
8431 		if (vcpu->arch.exception.pending)
8432 			goto out;
8433 		if (dbg->control & KVM_GUESTDBG_INJECT_DB)
8434 			kvm_queue_exception(vcpu, DB_VECTOR);
8435 		else
8436 			kvm_queue_exception(vcpu, BP_VECTOR);
8437 	}
8438 
8439 	/*
8440 	 * Read rflags as long as potentially injected trace flags are still
8441 	 * filtered out.
8442 	 */
8443 	rflags = kvm_get_rflags(vcpu);
8444 
8445 	vcpu->guest_debug = dbg->control;
8446 	if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
8447 		vcpu->guest_debug = 0;
8448 
8449 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
8450 		for (i = 0; i < KVM_NR_DB_REGS; ++i)
8451 			vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
8452 		vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
8453 	} else {
8454 		for (i = 0; i < KVM_NR_DB_REGS; i++)
8455 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
8456 	}
8457 	kvm_update_dr7(vcpu);
8458 
8459 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
8460 		vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
8461 			get_segment_base(vcpu, VCPU_SREG_CS);
8462 
8463 	/*
8464 	 * Trigger an rflags update that will inject or remove the trace
8465 	 * flags.
8466 	 */
8467 	kvm_set_rflags(vcpu, rflags);
8468 
8469 	kvm_x86_ops->update_bp_intercept(vcpu);
8470 
8471 	r = 0;
8472 
8473 out:
8474 	vcpu_put(vcpu);
8475 	return r;
8476 }
8477 
8478 /*
8479  * Translate a guest virtual address to a guest physical address.
8480  */
8481 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
8482 				    struct kvm_translation *tr)
8483 {
8484 	unsigned long vaddr = tr->linear_address;
8485 	gpa_t gpa;
8486 	int idx;
8487 
8488 	vcpu_load(vcpu);
8489 
8490 	idx = srcu_read_lock(&vcpu->kvm->srcu);
8491 	gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
8492 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
8493 	tr->physical_address = gpa;
8494 	tr->valid = gpa != UNMAPPED_GVA;
8495 	tr->writeable = 1;
8496 	tr->usermode = 0;
8497 
8498 	vcpu_put(vcpu);
8499 	return 0;
8500 }
8501 
8502 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
8503 {
8504 	struct fxregs_state *fxsave;
8505 
8506 	vcpu_load(vcpu);
8507 
8508 	fxsave = &vcpu->arch.guest_fpu.state.fxsave;
8509 	memcpy(fpu->fpr, fxsave->st_space, 128);
8510 	fpu->fcw = fxsave->cwd;
8511 	fpu->fsw = fxsave->swd;
8512 	fpu->ftwx = fxsave->twd;
8513 	fpu->last_opcode = fxsave->fop;
8514 	fpu->last_ip = fxsave->rip;
8515 	fpu->last_dp = fxsave->rdp;
8516 	memcpy(fpu->xmm, fxsave->xmm_space, sizeof(fxsave->xmm_space));
8517 
8518 	vcpu_put(vcpu);
8519 	return 0;
8520 }
8521 
8522 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
8523 {
8524 	struct fxregs_state *fxsave;
8525 
8526 	vcpu_load(vcpu);
8527 
8528 	fxsave = &vcpu->arch.guest_fpu.state.fxsave;
8529 
8530 	memcpy(fxsave->st_space, fpu->fpr, 128);
8531 	fxsave->cwd = fpu->fcw;
8532 	fxsave->swd = fpu->fsw;
8533 	fxsave->twd = fpu->ftwx;
8534 	fxsave->fop = fpu->last_opcode;
8535 	fxsave->rip = fpu->last_ip;
8536 	fxsave->rdp = fpu->last_dp;
8537 	memcpy(fxsave->xmm_space, fpu->xmm, sizeof(fxsave->xmm_space));
8538 
8539 	vcpu_put(vcpu);
8540 	return 0;
8541 }
8542 
8543 static void store_regs(struct kvm_vcpu *vcpu)
8544 {
8545 	BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
8546 
8547 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
8548 		__get_regs(vcpu, &vcpu->run->s.regs.regs);
8549 
8550 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
8551 		__get_sregs(vcpu, &vcpu->run->s.regs.sregs);
8552 
8553 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
8554 		kvm_vcpu_ioctl_x86_get_vcpu_events(
8555 				vcpu, &vcpu->run->s.regs.events);
8556 }
8557 
8558 static int sync_regs(struct kvm_vcpu *vcpu)
8559 {
8560 	if (vcpu->run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS)
8561 		return -EINVAL;
8562 
8563 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
8564 		__set_regs(vcpu, &vcpu->run->s.regs.regs);
8565 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
8566 	}
8567 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
8568 		if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs))
8569 			return -EINVAL;
8570 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
8571 	}
8572 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
8573 		if (kvm_vcpu_ioctl_x86_set_vcpu_events(
8574 				vcpu, &vcpu->run->s.regs.events))
8575 			return -EINVAL;
8576 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS;
8577 	}
8578 
8579 	return 0;
8580 }
8581 
8582 static void fx_init(struct kvm_vcpu *vcpu)
8583 {
8584 	fpstate_init(&vcpu->arch.guest_fpu.state);
8585 	if (boot_cpu_has(X86_FEATURE_XSAVES))
8586 		vcpu->arch.guest_fpu.state.xsave.header.xcomp_bv =
8587 			host_xcr0 | XSTATE_COMPACTION_ENABLED;
8588 
8589 	/*
8590 	 * Ensure guest xcr0 is valid for loading
8591 	 */
8592 	vcpu->arch.xcr0 = XFEATURE_MASK_FP;
8593 
8594 	vcpu->arch.cr0 |= X86_CR0_ET;
8595 }
8596 
8597 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
8598 {
8599 	void *wbinvd_dirty_mask = vcpu->arch.wbinvd_dirty_mask;
8600 
8601 	kvmclock_reset(vcpu);
8602 
8603 	kvm_x86_ops->vcpu_free(vcpu);
8604 	free_cpumask_var(wbinvd_dirty_mask);
8605 }
8606 
8607 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
8608 						unsigned int id)
8609 {
8610 	struct kvm_vcpu *vcpu;
8611 
8612 	if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
8613 		printk_once(KERN_WARNING
8614 		"kvm: SMP vm created on host with unstable TSC; "
8615 		"guest TSC will not be reliable\n");
8616 
8617 	vcpu = kvm_x86_ops->vcpu_create(kvm, id);
8618 
8619 	return vcpu;
8620 }
8621 
8622 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
8623 {
8624 	kvm_vcpu_mtrr_init(vcpu);
8625 	vcpu_load(vcpu);
8626 	kvm_vcpu_reset(vcpu, false);
8627 	kvm_init_mmu(vcpu, false);
8628 	vcpu_put(vcpu);
8629 	return 0;
8630 }
8631 
8632 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
8633 {
8634 	struct msr_data msr;
8635 	struct kvm *kvm = vcpu->kvm;
8636 
8637 	kvm_hv_vcpu_postcreate(vcpu);
8638 
8639 	if (mutex_lock_killable(&vcpu->mutex))
8640 		return;
8641 	vcpu_load(vcpu);
8642 	msr.data = 0x0;
8643 	msr.index = MSR_IA32_TSC;
8644 	msr.host_initiated = true;
8645 	kvm_write_tsc(vcpu, &msr);
8646 	vcpu_put(vcpu);
8647 	mutex_unlock(&vcpu->mutex);
8648 
8649 	if (!kvmclock_periodic_sync)
8650 		return;
8651 
8652 	schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
8653 					KVMCLOCK_SYNC_PERIOD);
8654 }
8655 
8656 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
8657 {
8658 	vcpu->arch.apf.msr_val = 0;
8659 
8660 	vcpu_load(vcpu);
8661 	kvm_mmu_unload(vcpu);
8662 	vcpu_put(vcpu);
8663 
8664 	kvm_x86_ops->vcpu_free(vcpu);
8665 }
8666 
8667 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
8668 {
8669 	kvm_lapic_reset(vcpu, init_event);
8670 
8671 	vcpu->arch.hflags = 0;
8672 
8673 	vcpu->arch.smi_pending = 0;
8674 	vcpu->arch.smi_count = 0;
8675 	atomic_set(&vcpu->arch.nmi_queued, 0);
8676 	vcpu->arch.nmi_pending = 0;
8677 	vcpu->arch.nmi_injected = false;
8678 	kvm_clear_interrupt_queue(vcpu);
8679 	kvm_clear_exception_queue(vcpu);
8680 	vcpu->arch.exception.pending = false;
8681 
8682 	memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
8683 	kvm_update_dr0123(vcpu);
8684 	vcpu->arch.dr6 = DR6_INIT;
8685 	kvm_update_dr6(vcpu);
8686 	vcpu->arch.dr7 = DR7_FIXED_1;
8687 	kvm_update_dr7(vcpu);
8688 
8689 	vcpu->arch.cr2 = 0;
8690 
8691 	kvm_make_request(KVM_REQ_EVENT, vcpu);
8692 	vcpu->arch.apf.msr_val = 0;
8693 	vcpu->arch.st.msr_val = 0;
8694 
8695 	kvmclock_reset(vcpu);
8696 
8697 	kvm_clear_async_pf_completion_queue(vcpu);
8698 	kvm_async_pf_hash_reset(vcpu);
8699 	vcpu->arch.apf.halted = false;
8700 
8701 	if (kvm_mpx_supported()) {
8702 		void *mpx_state_buffer;
8703 
8704 		/*
8705 		 * To avoid have the INIT path from kvm_apic_has_events() that be
8706 		 * called with loaded FPU and does not let userspace fix the state.
8707 		 */
8708 		if (init_event)
8709 			kvm_put_guest_fpu(vcpu);
8710 		mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu.state.xsave,
8711 					XFEATURE_MASK_BNDREGS);
8712 		if (mpx_state_buffer)
8713 			memset(mpx_state_buffer, 0, sizeof(struct mpx_bndreg_state));
8714 		mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu.state.xsave,
8715 					XFEATURE_MASK_BNDCSR);
8716 		if (mpx_state_buffer)
8717 			memset(mpx_state_buffer, 0, sizeof(struct mpx_bndcsr));
8718 		if (init_event)
8719 			kvm_load_guest_fpu(vcpu);
8720 	}
8721 
8722 	if (!init_event) {
8723 		kvm_pmu_reset(vcpu);
8724 		vcpu->arch.smbase = 0x30000;
8725 
8726 		vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
8727 		vcpu->arch.msr_misc_features_enables = 0;
8728 
8729 		vcpu->arch.xcr0 = XFEATURE_MASK_FP;
8730 	}
8731 
8732 	memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
8733 	vcpu->arch.regs_avail = ~0;
8734 	vcpu->arch.regs_dirty = ~0;
8735 
8736 	vcpu->arch.ia32_xss = 0;
8737 
8738 	kvm_x86_ops->vcpu_reset(vcpu, init_event);
8739 }
8740 
8741 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
8742 {
8743 	struct kvm_segment cs;
8744 
8745 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
8746 	cs.selector = vector << 8;
8747 	cs.base = vector << 12;
8748 	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
8749 	kvm_rip_write(vcpu, 0);
8750 }
8751 
8752 int kvm_arch_hardware_enable(void)
8753 {
8754 	struct kvm *kvm;
8755 	struct kvm_vcpu *vcpu;
8756 	int i;
8757 	int ret;
8758 	u64 local_tsc;
8759 	u64 max_tsc = 0;
8760 	bool stable, backwards_tsc = false;
8761 
8762 	kvm_shared_msr_cpu_online();
8763 	ret = kvm_x86_ops->hardware_enable();
8764 	if (ret != 0)
8765 		return ret;
8766 
8767 	local_tsc = rdtsc();
8768 	stable = !kvm_check_tsc_unstable();
8769 	list_for_each_entry(kvm, &vm_list, vm_list) {
8770 		kvm_for_each_vcpu(i, vcpu, kvm) {
8771 			if (!stable && vcpu->cpu == smp_processor_id())
8772 				kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
8773 			if (stable && vcpu->arch.last_host_tsc > local_tsc) {
8774 				backwards_tsc = true;
8775 				if (vcpu->arch.last_host_tsc > max_tsc)
8776 					max_tsc = vcpu->arch.last_host_tsc;
8777 			}
8778 		}
8779 	}
8780 
8781 	/*
8782 	 * Sometimes, even reliable TSCs go backwards.  This happens on
8783 	 * platforms that reset TSC during suspend or hibernate actions, but
8784 	 * maintain synchronization.  We must compensate.  Fortunately, we can
8785 	 * detect that condition here, which happens early in CPU bringup,
8786 	 * before any KVM threads can be running.  Unfortunately, we can't
8787 	 * bring the TSCs fully up to date with real time, as we aren't yet far
8788 	 * enough into CPU bringup that we know how much real time has actually
8789 	 * elapsed; our helper function, ktime_get_boot_ns() will be using boot
8790 	 * variables that haven't been updated yet.
8791 	 *
8792 	 * So we simply find the maximum observed TSC above, then record the
8793 	 * adjustment to TSC in each VCPU.  When the VCPU later gets loaded,
8794 	 * the adjustment will be applied.  Note that we accumulate
8795 	 * adjustments, in case multiple suspend cycles happen before some VCPU
8796 	 * gets a chance to run again.  In the event that no KVM threads get a
8797 	 * chance to run, we will miss the entire elapsed period, as we'll have
8798 	 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
8799 	 * loose cycle time.  This isn't too big a deal, since the loss will be
8800 	 * uniform across all VCPUs (not to mention the scenario is extremely
8801 	 * unlikely). It is possible that a second hibernate recovery happens
8802 	 * much faster than a first, causing the observed TSC here to be
8803 	 * smaller; this would require additional padding adjustment, which is
8804 	 * why we set last_host_tsc to the local tsc observed here.
8805 	 *
8806 	 * N.B. - this code below runs only on platforms with reliable TSC,
8807 	 * as that is the only way backwards_tsc is set above.  Also note
8808 	 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
8809 	 * have the same delta_cyc adjustment applied if backwards_tsc
8810 	 * is detected.  Note further, this adjustment is only done once,
8811 	 * as we reset last_host_tsc on all VCPUs to stop this from being
8812 	 * called multiple times (one for each physical CPU bringup).
8813 	 *
8814 	 * Platforms with unreliable TSCs don't have to deal with this, they
8815 	 * will be compensated by the logic in vcpu_load, which sets the TSC to
8816 	 * catchup mode.  This will catchup all VCPUs to real time, but cannot
8817 	 * guarantee that they stay in perfect synchronization.
8818 	 */
8819 	if (backwards_tsc) {
8820 		u64 delta_cyc = max_tsc - local_tsc;
8821 		list_for_each_entry(kvm, &vm_list, vm_list) {
8822 			kvm->arch.backwards_tsc_observed = true;
8823 			kvm_for_each_vcpu(i, vcpu, kvm) {
8824 				vcpu->arch.tsc_offset_adjustment += delta_cyc;
8825 				vcpu->arch.last_host_tsc = local_tsc;
8826 				kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
8827 			}
8828 
8829 			/*
8830 			 * We have to disable TSC offset matching.. if you were
8831 			 * booting a VM while issuing an S4 host suspend....
8832 			 * you may have some problem.  Solving this issue is
8833 			 * left as an exercise to the reader.
8834 			 */
8835 			kvm->arch.last_tsc_nsec = 0;
8836 			kvm->arch.last_tsc_write = 0;
8837 		}
8838 
8839 	}
8840 	return 0;
8841 }
8842 
8843 void kvm_arch_hardware_disable(void)
8844 {
8845 	kvm_x86_ops->hardware_disable();
8846 	drop_user_return_notifiers();
8847 }
8848 
8849 int kvm_arch_hardware_setup(void)
8850 {
8851 	int r;
8852 
8853 	r = kvm_x86_ops->hardware_setup();
8854 	if (r != 0)
8855 		return r;
8856 
8857 	if (kvm_has_tsc_control) {
8858 		/*
8859 		 * Make sure the user can only configure tsc_khz values that
8860 		 * fit into a signed integer.
8861 		 * A min value is not calculated because it will always
8862 		 * be 1 on all machines.
8863 		 */
8864 		u64 max = min(0x7fffffffULL,
8865 			      __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
8866 		kvm_max_guest_tsc_khz = max;
8867 
8868 		kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
8869 	}
8870 
8871 	kvm_init_msr_list();
8872 	return 0;
8873 }
8874 
8875 void kvm_arch_hardware_unsetup(void)
8876 {
8877 	kvm_x86_ops->hardware_unsetup();
8878 }
8879 
8880 void kvm_arch_check_processor_compat(void *rtn)
8881 {
8882 	kvm_x86_ops->check_processor_compatibility(rtn);
8883 }
8884 
8885 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
8886 {
8887 	return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
8888 }
8889 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
8890 
8891 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
8892 {
8893 	return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
8894 }
8895 
8896 struct static_key kvm_no_apic_vcpu __read_mostly;
8897 EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu);
8898 
8899 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
8900 {
8901 	struct page *page;
8902 	int r;
8903 
8904 	vcpu->arch.apicv_active = kvm_x86_ops->get_enable_apicv(vcpu);
8905 	vcpu->arch.emulate_ctxt.ops = &emulate_ops;
8906 	if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
8907 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
8908 	else
8909 		vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
8910 
8911 	page = alloc_page(GFP_KERNEL | __GFP_ZERO);
8912 	if (!page) {
8913 		r = -ENOMEM;
8914 		goto fail;
8915 	}
8916 	vcpu->arch.pio_data = page_address(page);
8917 
8918 	kvm_set_tsc_khz(vcpu, max_tsc_khz);
8919 
8920 	r = kvm_mmu_create(vcpu);
8921 	if (r < 0)
8922 		goto fail_free_pio_data;
8923 
8924 	if (irqchip_in_kernel(vcpu->kvm)) {
8925 		r = kvm_create_lapic(vcpu);
8926 		if (r < 0)
8927 			goto fail_mmu_destroy;
8928 	} else
8929 		static_key_slow_inc(&kvm_no_apic_vcpu);
8930 
8931 	vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
8932 				       GFP_KERNEL);
8933 	if (!vcpu->arch.mce_banks) {
8934 		r = -ENOMEM;
8935 		goto fail_free_lapic;
8936 	}
8937 	vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
8938 
8939 	if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL)) {
8940 		r = -ENOMEM;
8941 		goto fail_free_mce_banks;
8942 	}
8943 
8944 	fx_init(vcpu);
8945 
8946 	vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
8947 
8948 	vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
8949 
8950 	vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
8951 
8952 	kvm_async_pf_hash_reset(vcpu);
8953 	kvm_pmu_init(vcpu);
8954 
8955 	vcpu->arch.pending_external_vector = -1;
8956 	vcpu->arch.preempted_in_kernel = false;
8957 
8958 	kvm_hv_vcpu_init(vcpu);
8959 
8960 	return 0;
8961 
8962 fail_free_mce_banks:
8963 	kfree(vcpu->arch.mce_banks);
8964 fail_free_lapic:
8965 	kvm_free_lapic(vcpu);
8966 fail_mmu_destroy:
8967 	kvm_mmu_destroy(vcpu);
8968 fail_free_pio_data:
8969 	free_page((unsigned long)vcpu->arch.pio_data);
8970 fail:
8971 	return r;
8972 }
8973 
8974 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
8975 {
8976 	int idx;
8977 
8978 	kvm_hv_vcpu_uninit(vcpu);
8979 	kvm_pmu_destroy(vcpu);
8980 	kfree(vcpu->arch.mce_banks);
8981 	kvm_free_lapic(vcpu);
8982 	idx = srcu_read_lock(&vcpu->kvm->srcu);
8983 	kvm_mmu_destroy(vcpu);
8984 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
8985 	free_page((unsigned long)vcpu->arch.pio_data);
8986 	if (!lapic_in_kernel(vcpu))
8987 		static_key_slow_dec(&kvm_no_apic_vcpu);
8988 }
8989 
8990 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
8991 {
8992 	vcpu->arch.l1tf_flush_l1d = true;
8993 	kvm_x86_ops->sched_in(vcpu, cpu);
8994 }
8995 
8996 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
8997 {
8998 	if (type)
8999 		return -EINVAL;
9000 
9001 	INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
9002 	INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
9003 	INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
9004 	INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
9005 	atomic_set(&kvm->arch.noncoherent_dma_count, 0);
9006 
9007 	/* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
9008 	set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
9009 	/* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
9010 	set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
9011 		&kvm->arch.irq_sources_bitmap);
9012 
9013 	raw_spin_lock_init(&kvm->arch.tsc_write_lock);
9014 	mutex_init(&kvm->arch.apic_map_lock);
9015 	spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
9016 
9017 	kvm->arch.kvmclock_offset = -ktime_get_boot_ns();
9018 	pvclock_update_vm_gtod_copy(kvm);
9019 
9020 	kvm->arch.guest_can_read_msr_platform_info = true;
9021 
9022 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
9023 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
9024 
9025 	kvm_hv_init_vm(kvm);
9026 	kvm_page_track_init(kvm);
9027 	kvm_mmu_init_vm(kvm);
9028 
9029 	if (kvm_x86_ops->vm_init)
9030 		return kvm_x86_ops->vm_init(kvm);
9031 
9032 	return 0;
9033 }
9034 
9035 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
9036 {
9037 	vcpu_load(vcpu);
9038 	kvm_mmu_unload(vcpu);
9039 	vcpu_put(vcpu);
9040 }
9041 
9042 static void kvm_free_vcpus(struct kvm *kvm)
9043 {
9044 	unsigned int i;
9045 	struct kvm_vcpu *vcpu;
9046 
9047 	/*
9048 	 * Unpin any mmu pages first.
9049 	 */
9050 	kvm_for_each_vcpu(i, vcpu, kvm) {
9051 		kvm_clear_async_pf_completion_queue(vcpu);
9052 		kvm_unload_vcpu_mmu(vcpu);
9053 	}
9054 	kvm_for_each_vcpu(i, vcpu, kvm)
9055 		kvm_arch_vcpu_free(vcpu);
9056 
9057 	mutex_lock(&kvm->lock);
9058 	for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
9059 		kvm->vcpus[i] = NULL;
9060 
9061 	atomic_set(&kvm->online_vcpus, 0);
9062 	mutex_unlock(&kvm->lock);
9063 }
9064 
9065 void kvm_arch_sync_events(struct kvm *kvm)
9066 {
9067 	cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
9068 	cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
9069 	kvm_free_pit(kvm);
9070 }
9071 
9072 int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
9073 {
9074 	int i, r;
9075 	unsigned long hva;
9076 	struct kvm_memslots *slots = kvm_memslots(kvm);
9077 	struct kvm_memory_slot *slot, old;
9078 
9079 	/* Called with kvm->slots_lock held.  */
9080 	if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
9081 		return -EINVAL;
9082 
9083 	slot = id_to_memslot(slots, id);
9084 	if (size) {
9085 		if (slot->npages)
9086 			return -EEXIST;
9087 
9088 		/*
9089 		 * MAP_SHARED to prevent internal slot pages from being moved
9090 		 * by fork()/COW.
9091 		 */
9092 		hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
9093 			      MAP_SHARED | MAP_ANONYMOUS, 0);
9094 		if (IS_ERR((void *)hva))
9095 			return PTR_ERR((void *)hva);
9096 	} else {
9097 		if (!slot->npages)
9098 			return 0;
9099 
9100 		hva = 0;
9101 	}
9102 
9103 	old = *slot;
9104 	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
9105 		struct kvm_userspace_memory_region m;
9106 
9107 		m.slot = id | (i << 16);
9108 		m.flags = 0;
9109 		m.guest_phys_addr = gpa;
9110 		m.userspace_addr = hva;
9111 		m.memory_size = size;
9112 		r = __kvm_set_memory_region(kvm, &m);
9113 		if (r < 0)
9114 			return r;
9115 	}
9116 
9117 	if (!size)
9118 		vm_munmap(old.userspace_addr, old.npages * PAGE_SIZE);
9119 
9120 	return 0;
9121 }
9122 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
9123 
9124 int x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
9125 {
9126 	int r;
9127 
9128 	mutex_lock(&kvm->slots_lock);
9129 	r = __x86_set_memory_region(kvm, id, gpa, size);
9130 	mutex_unlock(&kvm->slots_lock);
9131 
9132 	return r;
9133 }
9134 EXPORT_SYMBOL_GPL(x86_set_memory_region);
9135 
9136 void kvm_arch_destroy_vm(struct kvm *kvm)
9137 {
9138 	if (current->mm == kvm->mm) {
9139 		/*
9140 		 * Free memory regions allocated on behalf of userspace,
9141 		 * unless the the memory map has changed due to process exit
9142 		 * or fd copying.
9143 		 */
9144 		x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT, 0, 0);
9145 		x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT, 0, 0);
9146 		x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
9147 	}
9148 	if (kvm_x86_ops->vm_destroy)
9149 		kvm_x86_ops->vm_destroy(kvm);
9150 	kvm_pic_destroy(kvm);
9151 	kvm_ioapic_destroy(kvm);
9152 	kvm_free_vcpus(kvm);
9153 	kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
9154 	kvm_mmu_uninit_vm(kvm);
9155 	kvm_page_track_cleanup(kvm);
9156 	kvm_hv_destroy_vm(kvm);
9157 }
9158 
9159 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
9160 			   struct kvm_memory_slot *dont)
9161 {
9162 	int i;
9163 
9164 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9165 		if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
9166 			kvfree(free->arch.rmap[i]);
9167 			free->arch.rmap[i] = NULL;
9168 		}
9169 		if (i == 0)
9170 			continue;
9171 
9172 		if (!dont || free->arch.lpage_info[i - 1] !=
9173 			     dont->arch.lpage_info[i - 1]) {
9174 			kvfree(free->arch.lpage_info[i - 1]);
9175 			free->arch.lpage_info[i - 1] = NULL;
9176 		}
9177 	}
9178 
9179 	kvm_page_track_free_memslot(free, dont);
9180 }
9181 
9182 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
9183 			    unsigned long npages)
9184 {
9185 	int i;
9186 
9187 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9188 		struct kvm_lpage_info *linfo;
9189 		unsigned long ugfn;
9190 		int lpages;
9191 		int level = i + 1;
9192 
9193 		lpages = gfn_to_index(slot->base_gfn + npages - 1,
9194 				      slot->base_gfn, level) + 1;
9195 
9196 		slot->arch.rmap[i] =
9197 			kvcalloc(lpages, sizeof(*slot->arch.rmap[i]),
9198 				 GFP_KERNEL);
9199 		if (!slot->arch.rmap[i])
9200 			goto out_free;
9201 		if (i == 0)
9202 			continue;
9203 
9204 		linfo = kvcalloc(lpages, sizeof(*linfo), GFP_KERNEL);
9205 		if (!linfo)
9206 			goto out_free;
9207 
9208 		slot->arch.lpage_info[i - 1] = linfo;
9209 
9210 		if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
9211 			linfo[0].disallow_lpage = 1;
9212 		if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
9213 			linfo[lpages - 1].disallow_lpage = 1;
9214 		ugfn = slot->userspace_addr >> PAGE_SHIFT;
9215 		/*
9216 		 * If the gfn and userspace address are not aligned wrt each
9217 		 * other, or if explicitly asked to, disable large page
9218 		 * support for this slot
9219 		 */
9220 		if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
9221 		    !kvm_largepages_enabled()) {
9222 			unsigned long j;
9223 
9224 			for (j = 0; j < lpages; ++j)
9225 				linfo[j].disallow_lpage = 1;
9226 		}
9227 	}
9228 
9229 	if (kvm_page_track_create_memslot(slot, npages))
9230 		goto out_free;
9231 
9232 	return 0;
9233 
9234 out_free:
9235 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9236 		kvfree(slot->arch.rmap[i]);
9237 		slot->arch.rmap[i] = NULL;
9238 		if (i == 0)
9239 			continue;
9240 
9241 		kvfree(slot->arch.lpage_info[i - 1]);
9242 		slot->arch.lpage_info[i - 1] = NULL;
9243 	}
9244 	return -ENOMEM;
9245 }
9246 
9247 void kvm_arch_memslots_updated(struct kvm *kvm, struct kvm_memslots *slots)
9248 {
9249 	/*
9250 	 * memslots->generation has been incremented.
9251 	 * mmio generation may have reached its maximum value.
9252 	 */
9253 	kvm_mmu_invalidate_mmio_sptes(kvm, slots);
9254 }
9255 
9256 int kvm_arch_prepare_memory_region(struct kvm *kvm,
9257 				struct kvm_memory_slot *memslot,
9258 				const struct kvm_userspace_memory_region *mem,
9259 				enum kvm_mr_change change)
9260 {
9261 	return 0;
9262 }
9263 
9264 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
9265 				     struct kvm_memory_slot *new)
9266 {
9267 	/* Still write protect RO slot */
9268 	if (new->flags & KVM_MEM_READONLY) {
9269 		kvm_mmu_slot_remove_write_access(kvm, new);
9270 		return;
9271 	}
9272 
9273 	/*
9274 	 * Call kvm_x86_ops dirty logging hooks when they are valid.
9275 	 *
9276 	 * kvm_x86_ops->slot_disable_log_dirty is called when:
9277 	 *
9278 	 *  - KVM_MR_CREATE with dirty logging is disabled
9279 	 *  - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
9280 	 *
9281 	 * The reason is, in case of PML, we need to set D-bit for any slots
9282 	 * with dirty logging disabled in order to eliminate unnecessary GPA
9283 	 * logging in PML buffer (and potential PML buffer full VMEXT). This
9284 	 * guarantees leaving PML enabled during guest's lifetime won't have
9285 	 * any additonal overhead from PML when guest is running with dirty
9286 	 * logging disabled for memory slots.
9287 	 *
9288 	 * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
9289 	 * to dirty logging mode.
9290 	 *
9291 	 * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
9292 	 *
9293 	 * In case of write protect:
9294 	 *
9295 	 * Write protect all pages for dirty logging.
9296 	 *
9297 	 * All the sptes including the large sptes which point to this
9298 	 * slot are set to readonly. We can not create any new large
9299 	 * spte on this slot until the end of the logging.
9300 	 *
9301 	 * See the comments in fast_page_fault().
9302 	 */
9303 	if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
9304 		if (kvm_x86_ops->slot_enable_log_dirty)
9305 			kvm_x86_ops->slot_enable_log_dirty(kvm, new);
9306 		else
9307 			kvm_mmu_slot_remove_write_access(kvm, new);
9308 	} else {
9309 		if (kvm_x86_ops->slot_disable_log_dirty)
9310 			kvm_x86_ops->slot_disable_log_dirty(kvm, new);
9311 	}
9312 }
9313 
9314 void kvm_arch_commit_memory_region(struct kvm *kvm,
9315 				const struct kvm_userspace_memory_region *mem,
9316 				const struct kvm_memory_slot *old,
9317 				const struct kvm_memory_slot *new,
9318 				enum kvm_mr_change change)
9319 {
9320 	int nr_mmu_pages = 0;
9321 
9322 	if (!kvm->arch.n_requested_mmu_pages)
9323 		nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
9324 
9325 	if (nr_mmu_pages)
9326 		kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
9327 
9328 	/*
9329 	 * Dirty logging tracks sptes in 4k granularity, meaning that large
9330 	 * sptes have to be split.  If live migration is successful, the guest
9331 	 * in the source machine will be destroyed and large sptes will be
9332 	 * created in the destination. However, if the guest continues to run
9333 	 * in the source machine (for example if live migration fails), small
9334 	 * sptes will remain around and cause bad performance.
9335 	 *
9336 	 * Scan sptes if dirty logging has been stopped, dropping those
9337 	 * which can be collapsed into a single large-page spte.  Later
9338 	 * page faults will create the large-page sptes.
9339 	 */
9340 	if ((change != KVM_MR_DELETE) &&
9341 		(old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
9342 		!(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
9343 		kvm_mmu_zap_collapsible_sptes(kvm, new);
9344 
9345 	/*
9346 	 * Set up write protection and/or dirty logging for the new slot.
9347 	 *
9348 	 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
9349 	 * been zapped so no dirty logging staff is needed for old slot. For
9350 	 * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
9351 	 * new and it's also covered when dealing with the new slot.
9352 	 *
9353 	 * FIXME: const-ify all uses of struct kvm_memory_slot.
9354 	 */
9355 	if (change != KVM_MR_DELETE)
9356 		kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new);
9357 }
9358 
9359 void kvm_arch_flush_shadow_all(struct kvm *kvm)
9360 {
9361 	kvm_mmu_invalidate_zap_all_pages(kvm);
9362 }
9363 
9364 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
9365 				   struct kvm_memory_slot *slot)
9366 {
9367 	kvm_page_track_flush_slot(kvm, slot);
9368 }
9369 
9370 static inline bool kvm_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
9371 {
9372 	return (is_guest_mode(vcpu) &&
9373 			kvm_x86_ops->guest_apic_has_interrupt &&
9374 			kvm_x86_ops->guest_apic_has_interrupt(vcpu));
9375 }
9376 
9377 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
9378 {
9379 	if (!list_empty_careful(&vcpu->async_pf.done))
9380 		return true;
9381 
9382 	if (kvm_apic_has_events(vcpu))
9383 		return true;
9384 
9385 	if (vcpu->arch.pv.pv_unhalted)
9386 		return true;
9387 
9388 	if (vcpu->arch.exception.pending)
9389 		return true;
9390 
9391 	if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
9392 	    (vcpu->arch.nmi_pending &&
9393 	     kvm_x86_ops->nmi_allowed(vcpu)))
9394 		return true;
9395 
9396 	if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
9397 	    (vcpu->arch.smi_pending && !is_smm(vcpu)))
9398 		return true;
9399 
9400 	if (kvm_arch_interrupt_allowed(vcpu) &&
9401 	    (kvm_cpu_has_interrupt(vcpu) ||
9402 	    kvm_guest_apic_has_interrupt(vcpu)))
9403 		return true;
9404 
9405 	if (kvm_hv_has_stimer_pending(vcpu))
9406 		return true;
9407 
9408 	return false;
9409 }
9410 
9411 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
9412 {
9413 	return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
9414 }
9415 
9416 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
9417 {
9418 	return vcpu->arch.preempted_in_kernel;
9419 }
9420 
9421 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
9422 {
9423 	return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
9424 }
9425 
9426 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
9427 {
9428 	return kvm_x86_ops->interrupt_allowed(vcpu);
9429 }
9430 
9431 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
9432 {
9433 	if (is_64_bit_mode(vcpu))
9434 		return kvm_rip_read(vcpu);
9435 	return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
9436 		     kvm_rip_read(vcpu));
9437 }
9438 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
9439 
9440 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
9441 {
9442 	return kvm_get_linear_rip(vcpu) == linear_rip;
9443 }
9444 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
9445 
9446 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
9447 {
9448 	unsigned long rflags;
9449 
9450 	rflags = kvm_x86_ops->get_rflags(vcpu);
9451 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
9452 		rflags &= ~X86_EFLAGS_TF;
9453 	return rflags;
9454 }
9455 EXPORT_SYMBOL_GPL(kvm_get_rflags);
9456 
9457 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
9458 {
9459 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
9460 	    kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
9461 		rflags |= X86_EFLAGS_TF;
9462 	kvm_x86_ops->set_rflags(vcpu, rflags);
9463 }
9464 
9465 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
9466 {
9467 	__kvm_set_rflags(vcpu, rflags);
9468 	kvm_make_request(KVM_REQ_EVENT, vcpu);
9469 }
9470 EXPORT_SYMBOL_GPL(kvm_set_rflags);
9471 
9472 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
9473 {
9474 	int r;
9475 
9476 	if ((vcpu->arch.mmu->direct_map != work->arch.direct_map) ||
9477 	      work->wakeup_all)
9478 		return;
9479 
9480 	r = kvm_mmu_reload(vcpu);
9481 	if (unlikely(r))
9482 		return;
9483 
9484 	if (!vcpu->arch.mmu->direct_map &&
9485 	      work->arch.cr3 != vcpu->arch.mmu->get_cr3(vcpu))
9486 		return;
9487 
9488 	vcpu->arch.mmu->page_fault(vcpu, work->gva, 0, true);
9489 }
9490 
9491 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
9492 {
9493 	return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
9494 }
9495 
9496 static inline u32 kvm_async_pf_next_probe(u32 key)
9497 {
9498 	return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
9499 }
9500 
9501 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
9502 {
9503 	u32 key = kvm_async_pf_hash_fn(gfn);
9504 
9505 	while (vcpu->arch.apf.gfns[key] != ~0)
9506 		key = kvm_async_pf_next_probe(key);
9507 
9508 	vcpu->arch.apf.gfns[key] = gfn;
9509 }
9510 
9511 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
9512 {
9513 	int i;
9514 	u32 key = kvm_async_pf_hash_fn(gfn);
9515 
9516 	for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
9517 		     (vcpu->arch.apf.gfns[key] != gfn &&
9518 		      vcpu->arch.apf.gfns[key] != ~0); i++)
9519 		key = kvm_async_pf_next_probe(key);
9520 
9521 	return key;
9522 }
9523 
9524 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
9525 {
9526 	return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
9527 }
9528 
9529 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
9530 {
9531 	u32 i, j, k;
9532 
9533 	i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
9534 	while (true) {
9535 		vcpu->arch.apf.gfns[i] = ~0;
9536 		do {
9537 			j = kvm_async_pf_next_probe(j);
9538 			if (vcpu->arch.apf.gfns[j] == ~0)
9539 				return;
9540 			k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
9541 			/*
9542 			 * k lies cyclically in ]i,j]
9543 			 * |    i.k.j |
9544 			 * |....j i.k.| or  |.k..j i...|
9545 			 */
9546 		} while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
9547 		vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
9548 		i = j;
9549 	}
9550 }
9551 
9552 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
9553 {
9554 
9555 	return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
9556 				      sizeof(val));
9557 }
9558 
9559 static int apf_get_user(struct kvm_vcpu *vcpu, u32 *val)
9560 {
9561 
9562 	return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, val,
9563 				      sizeof(u32));
9564 }
9565 
9566 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
9567 				     struct kvm_async_pf *work)
9568 {
9569 	struct x86_exception fault;
9570 
9571 	trace_kvm_async_pf_not_present(work->arch.token, work->gva);
9572 	kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
9573 
9574 	if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
9575 	    (vcpu->arch.apf.send_user_only &&
9576 	     kvm_x86_ops->get_cpl(vcpu) == 0))
9577 		kvm_make_request(KVM_REQ_APF_HALT, vcpu);
9578 	else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
9579 		fault.vector = PF_VECTOR;
9580 		fault.error_code_valid = true;
9581 		fault.error_code = 0;
9582 		fault.nested_page_fault = false;
9583 		fault.address = work->arch.token;
9584 		fault.async_page_fault = true;
9585 		kvm_inject_page_fault(vcpu, &fault);
9586 	}
9587 }
9588 
9589 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
9590 				 struct kvm_async_pf *work)
9591 {
9592 	struct x86_exception fault;
9593 	u32 val;
9594 
9595 	if (work->wakeup_all)
9596 		work->arch.token = ~0; /* broadcast wakeup */
9597 	else
9598 		kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
9599 	trace_kvm_async_pf_ready(work->arch.token, work->gva);
9600 
9601 	if (vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED &&
9602 	    !apf_get_user(vcpu, &val)) {
9603 		if (val == KVM_PV_REASON_PAGE_NOT_PRESENT &&
9604 		    vcpu->arch.exception.pending &&
9605 		    vcpu->arch.exception.nr == PF_VECTOR &&
9606 		    !apf_put_user(vcpu, 0)) {
9607 			vcpu->arch.exception.injected = false;
9608 			vcpu->arch.exception.pending = false;
9609 			vcpu->arch.exception.nr = 0;
9610 			vcpu->arch.exception.has_error_code = false;
9611 			vcpu->arch.exception.error_code = 0;
9612 			vcpu->arch.exception.has_payload = false;
9613 			vcpu->arch.exception.payload = 0;
9614 		} else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
9615 			fault.vector = PF_VECTOR;
9616 			fault.error_code_valid = true;
9617 			fault.error_code = 0;
9618 			fault.nested_page_fault = false;
9619 			fault.address = work->arch.token;
9620 			fault.async_page_fault = true;
9621 			kvm_inject_page_fault(vcpu, &fault);
9622 		}
9623 	}
9624 	vcpu->arch.apf.halted = false;
9625 	vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
9626 }
9627 
9628 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
9629 {
9630 	if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
9631 		return true;
9632 	else
9633 		return kvm_can_do_async_pf(vcpu);
9634 }
9635 
9636 void kvm_arch_start_assignment(struct kvm *kvm)
9637 {
9638 	atomic_inc(&kvm->arch.assigned_device_count);
9639 }
9640 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
9641 
9642 void kvm_arch_end_assignment(struct kvm *kvm)
9643 {
9644 	atomic_dec(&kvm->arch.assigned_device_count);
9645 }
9646 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
9647 
9648 bool kvm_arch_has_assigned_device(struct kvm *kvm)
9649 {
9650 	return atomic_read(&kvm->arch.assigned_device_count);
9651 }
9652 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
9653 
9654 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
9655 {
9656 	atomic_inc(&kvm->arch.noncoherent_dma_count);
9657 }
9658 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
9659 
9660 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
9661 {
9662 	atomic_dec(&kvm->arch.noncoherent_dma_count);
9663 }
9664 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
9665 
9666 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
9667 {
9668 	return atomic_read(&kvm->arch.noncoherent_dma_count);
9669 }
9670 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
9671 
9672 bool kvm_arch_has_irq_bypass(void)
9673 {
9674 	return kvm_x86_ops->update_pi_irte != NULL;
9675 }
9676 
9677 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
9678 				      struct irq_bypass_producer *prod)
9679 {
9680 	struct kvm_kernel_irqfd *irqfd =
9681 		container_of(cons, struct kvm_kernel_irqfd, consumer);
9682 
9683 	irqfd->producer = prod;
9684 
9685 	return kvm_x86_ops->update_pi_irte(irqfd->kvm,
9686 					   prod->irq, irqfd->gsi, 1);
9687 }
9688 
9689 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
9690 				      struct irq_bypass_producer *prod)
9691 {
9692 	int ret;
9693 	struct kvm_kernel_irqfd *irqfd =
9694 		container_of(cons, struct kvm_kernel_irqfd, consumer);
9695 
9696 	WARN_ON(irqfd->producer != prod);
9697 	irqfd->producer = NULL;
9698 
9699 	/*
9700 	 * When producer of consumer is unregistered, we change back to
9701 	 * remapped mode, so we can re-use the current implementation
9702 	 * when the irq is masked/disabled or the consumer side (KVM
9703 	 * int this case doesn't want to receive the interrupts.
9704 	*/
9705 	ret = kvm_x86_ops->update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0);
9706 	if (ret)
9707 		printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
9708 		       " fails: %d\n", irqfd->consumer.token, ret);
9709 }
9710 
9711 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
9712 				   uint32_t guest_irq, bool set)
9713 {
9714 	if (!kvm_x86_ops->update_pi_irte)
9715 		return -EINVAL;
9716 
9717 	return kvm_x86_ops->update_pi_irte(kvm, host_irq, guest_irq, set);
9718 }
9719 
9720 bool kvm_vector_hashing_enabled(void)
9721 {
9722 	return vector_hashing;
9723 }
9724 EXPORT_SYMBOL_GPL(kvm_vector_hashing_enabled);
9725 
9726 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
9727 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
9728 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
9729 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
9730 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
9731 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
9732 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
9733 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
9734 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
9735 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
9736 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
9737 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
9738 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
9739 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
9740 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window);
9741 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
9742 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
9743 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
9744 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);
9745