xref: /openbmc/linux/arch/x86/kvm/x86.c (revision 45471cd98decae5fced8b38e46c223f54a924814)
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 "assigned-dev.h"
31 #include "pmu.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/module.h>
39 #include <linux/mman.h>
40 #include <linux/highmem.h>
41 #include <linux/iommu.h>
42 #include <linux/intel-iommu.h>
43 #include <linux/cpufreq.h>
44 #include <linux/user-return-notifier.h>
45 #include <linux/srcu.h>
46 #include <linux/slab.h>
47 #include <linux/perf_event.h>
48 #include <linux/uaccess.h>
49 #include <linux/hash.h>
50 #include <linux/pci.h>
51 #include <linux/timekeeper_internal.h>
52 #include <linux/pvclock_gtod.h>
53 #include <trace/events/kvm.h>
54 
55 #define CREATE_TRACE_POINTS
56 #include "trace.h"
57 
58 #include <asm/debugreg.h>
59 #include <asm/msr.h>
60 #include <asm/desc.h>
61 #include <asm/mce.h>
62 #include <linux/kernel_stat.h>
63 #include <asm/fpu/internal.h> /* Ugh! */
64 #include <asm/pvclock.h>
65 #include <asm/div64.h>
66 
67 #define MAX_IO_MSRS 256
68 #define KVM_MAX_MCE_BANKS 32
69 #define KVM_MCE_CAP_SUPPORTED (MCG_CTL_P | MCG_SER_P)
70 
71 #define emul_to_vcpu(ctxt) \
72 	container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
73 
74 /* EFER defaults:
75  * - enable syscall per default because its emulated by KVM
76  * - enable LME and LMA per default on 64 bit KVM
77  */
78 #ifdef CONFIG_X86_64
79 static
80 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
81 #else
82 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
83 #endif
84 
85 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
86 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
87 
88 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
89 static void process_nmi(struct kvm_vcpu *vcpu);
90 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
91 
92 struct kvm_x86_ops *kvm_x86_ops;
93 EXPORT_SYMBOL_GPL(kvm_x86_ops);
94 
95 static bool ignore_msrs = 0;
96 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
97 
98 unsigned int min_timer_period_us = 500;
99 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
100 
101 static bool __read_mostly kvmclock_periodic_sync = true;
102 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
103 
104 bool kvm_has_tsc_control;
105 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
106 u32  kvm_max_guest_tsc_khz;
107 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
108 
109 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
110 static u32 tsc_tolerance_ppm = 250;
111 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
112 
113 /* lapic timer advance (tscdeadline mode only) in nanoseconds */
114 unsigned int lapic_timer_advance_ns = 0;
115 module_param(lapic_timer_advance_ns, uint, S_IRUGO | S_IWUSR);
116 
117 static bool backwards_tsc_observed = false;
118 
119 #define KVM_NR_SHARED_MSRS 16
120 
121 struct kvm_shared_msrs_global {
122 	int nr;
123 	u32 msrs[KVM_NR_SHARED_MSRS];
124 };
125 
126 struct kvm_shared_msrs {
127 	struct user_return_notifier urn;
128 	bool registered;
129 	struct kvm_shared_msr_values {
130 		u64 host;
131 		u64 curr;
132 	} values[KVM_NR_SHARED_MSRS];
133 };
134 
135 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
136 static struct kvm_shared_msrs __percpu *shared_msrs;
137 
138 struct kvm_stats_debugfs_item debugfs_entries[] = {
139 	{ "pf_fixed", VCPU_STAT(pf_fixed) },
140 	{ "pf_guest", VCPU_STAT(pf_guest) },
141 	{ "tlb_flush", VCPU_STAT(tlb_flush) },
142 	{ "invlpg", VCPU_STAT(invlpg) },
143 	{ "exits", VCPU_STAT(exits) },
144 	{ "io_exits", VCPU_STAT(io_exits) },
145 	{ "mmio_exits", VCPU_STAT(mmio_exits) },
146 	{ "signal_exits", VCPU_STAT(signal_exits) },
147 	{ "irq_window", VCPU_STAT(irq_window_exits) },
148 	{ "nmi_window", VCPU_STAT(nmi_window_exits) },
149 	{ "halt_exits", VCPU_STAT(halt_exits) },
150 	{ "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
151 	{ "halt_wakeup", VCPU_STAT(halt_wakeup) },
152 	{ "hypercalls", VCPU_STAT(hypercalls) },
153 	{ "request_irq", VCPU_STAT(request_irq_exits) },
154 	{ "irq_exits", VCPU_STAT(irq_exits) },
155 	{ "host_state_reload", VCPU_STAT(host_state_reload) },
156 	{ "efer_reload", VCPU_STAT(efer_reload) },
157 	{ "fpu_reload", VCPU_STAT(fpu_reload) },
158 	{ "insn_emulation", VCPU_STAT(insn_emulation) },
159 	{ "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
160 	{ "irq_injections", VCPU_STAT(irq_injections) },
161 	{ "nmi_injections", VCPU_STAT(nmi_injections) },
162 	{ "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
163 	{ "mmu_pte_write", VM_STAT(mmu_pte_write) },
164 	{ "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
165 	{ "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
166 	{ "mmu_flooded", VM_STAT(mmu_flooded) },
167 	{ "mmu_recycled", VM_STAT(mmu_recycled) },
168 	{ "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
169 	{ "mmu_unsync", VM_STAT(mmu_unsync) },
170 	{ "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
171 	{ "largepages", VM_STAT(lpages) },
172 	{ NULL }
173 };
174 
175 u64 __read_mostly host_xcr0;
176 
177 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
178 
179 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
180 {
181 	int i;
182 	for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
183 		vcpu->arch.apf.gfns[i] = ~0;
184 }
185 
186 static void kvm_on_user_return(struct user_return_notifier *urn)
187 {
188 	unsigned slot;
189 	struct kvm_shared_msrs *locals
190 		= container_of(urn, struct kvm_shared_msrs, urn);
191 	struct kvm_shared_msr_values *values;
192 
193 	for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
194 		values = &locals->values[slot];
195 		if (values->host != values->curr) {
196 			wrmsrl(shared_msrs_global.msrs[slot], values->host);
197 			values->curr = values->host;
198 		}
199 	}
200 	locals->registered = false;
201 	user_return_notifier_unregister(urn);
202 }
203 
204 static void shared_msr_update(unsigned slot, u32 msr)
205 {
206 	u64 value;
207 	unsigned int cpu = smp_processor_id();
208 	struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
209 
210 	/* only read, and nobody should modify it at this time,
211 	 * so don't need lock */
212 	if (slot >= shared_msrs_global.nr) {
213 		printk(KERN_ERR "kvm: invalid MSR slot!");
214 		return;
215 	}
216 	rdmsrl_safe(msr, &value);
217 	smsr->values[slot].host = value;
218 	smsr->values[slot].curr = value;
219 }
220 
221 void kvm_define_shared_msr(unsigned slot, u32 msr)
222 {
223 	BUG_ON(slot >= KVM_NR_SHARED_MSRS);
224 	if (slot >= shared_msrs_global.nr)
225 		shared_msrs_global.nr = slot + 1;
226 	shared_msrs_global.msrs[slot] = msr;
227 	/* we need ensured the shared_msr_global have been updated */
228 	smp_wmb();
229 }
230 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
231 
232 static void kvm_shared_msr_cpu_online(void)
233 {
234 	unsigned i;
235 
236 	for (i = 0; i < shared_msrs_global.nr; ++i)
237 		shared_msr_update(i, shared_msrs_global.msrs[i]);
238 }
239 
240 int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
241 {
242 	unsigned int cpu = smp_processor_id();
243 	struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
244 	int err;
245 
246 	if (((value ^ smsr->values[slot].curr) & mask) == 0)
247 		return 0;
248 	smsr->values[slot].curr = value;
249 	err = wrmsrl_safe(shared_msrs_global.msrs[slot], value);
250 	if (err)
251 		return 1;
252 
253 	if (!smsr->registered) {
254 		smsr->urn.on_user_return = kvm_on_user_return;
255 		user_return_notifier_register(&smsr->urn);
256 		smsr->registered = true;
257 	}
258 	return 0;
259 }
260 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
261 
262 static void drop_user_return_notifiers(void)
263 {
264 	unsigned int cpu = smp_processor_id();
265 	struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
266 
267 	if (smsr->registered)
268 		kvm_on_user_return(&smsr->urn);
269 }
270 
271 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
272 {
273 	return vcpu->arch.apic_base;
274 }
275 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
276 
277 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
278 {
279 	u64 old_state = vcpu->arch.apic_base &
280 		(MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
281 	u64 new_state = msr_info->data &
282 		(MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
283 	u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) |
284 		0x2ff | (guest_cpuid_has_x2apic(vcpu) ? 0 : X2APIC_ENABLE);
285 
286 	if (!msr_info->host_initiated &&
287 	    ((msr_info->data & reserved_bits) != 0 ||
288 	     new_state == X2APIC_ENABLE ||
289 	     (new_state == MSR_IA32_APICBASE_ENABLE &&
290 	      old_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE)) ||
291 	     (new_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE) &&
292 	      old_state == 0)))
293 		return 1;
294 
295 	kvm_lapic_set_base(vcpu, msr_info->data);
296 	return 0;
297 }
298 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
299 
300 asmlinkage __visible void kvm_spurious_fault(void)
301 {
302 	/* Fault while not rebooting.  We want the trace. */
303 	BUG();
304 }
305 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
306 
307 #define EXCPT_BENIGN		0
308 #define EXCPT_CONTRIBUTORY	1
309 #define EXCPT_PF		2
310 
311 static int exception_class(int vector)
312 {
313 	switch (vector) {
314 	case PF_VECTOR:
315 		return EXCPT_PF;
316 	case DE_VECTOR:
317 	case TS_VECTOR:
318 	case NP_VECTOR:
319 	case SS_VECTOR:
320 	case GP_VECTOR:
321 		return EXCPT_CONTRIBUTORY;
322 	default:
323 		break;
324 	}
325 	return EXCPT_BENIGN;
326 }
327 
328 #define EXCPT_FAULT		0
329 #define EXCPT_TRAP		1
330 #define EXCPT_ABORT		2
331 #define EXCPT_INTERRUPT		3
332 
333 static int exception_type(int vector)
334 {
335 	unsigned int mask;
336 
337 	if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
338 		return EXCPT_INTERRUPT;
339 
340 	mask = 1 << vector;
341 
342 	/* #DB is trap, as instruction watchpoints are handled elsewhere */
343 	if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
344 		return EXCPT_TRAP;
345 
346 	if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
347 		return EXCPT_ABORT;
348 
349 	/* Reserved exceptions will result in fault */
350 	return EXCPT_FAULT;
351 }
352 
353 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
354 		unsigned nr, bool has_error, u32 error_code,
355 		bool reinject)
356 {
357 	u32 prev_nr;
358 	int class1, class2;
359 
360 	kvm_make_request(KVM_REQ_EVENT, vcpu);
361 
362 	if (!vcpu->arch.exception.pending) {
363 	queue:
364 		if (has_error && !is_protmode(vcpu))
365 			has_error = false;
366 		vcpu->arch.exception.pending = true;
367 		vcpu->arch.exception.has_error_code = has_error;
368 		vcpu->arch.exception.nr = nr;
369 		vcpu->arch.exception.error_code = error_code;
370 		vcpu->arch.exception.reinject = reinject;
371 		return;
372 	}
373 
374 	/* to check exception */
375 	prev_nr = vcpu->arch.exception.nr;
376 	if (prev_nr == DF_VECTOR) {
377 		/* triple fault -> shutdown */
378 		kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
379 		return;
380 	}
381 	class1 = exception_class(prev_nr);
382 	class2 = exception_class(nr);
383 	if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
384 		|| (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
385 		/* generate double fault per SDM Table 5-5 */
386 		vcpu->arch.exception.pending = true;
387 		vcpu->arch.exception.has_error_code = true;
388 		vcpu->arch.exception.nr = DF_VECTOR;
389 		vcpu->arch.exception.error_code = 0;
390 	} else
391 		/* replace previous exception with a new one in a hope
392 		   that instruction re-execution will regenerate lost
393 		   exception */
394 		goto queue;
395 }
396 
397 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
398 {
399 	kvm_multiple_exception(vcpu, nr, false, 0, false);
400 }
401 EXPORT_SYMBOL_GPL(kvm_queue_exception);
402 
403 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
404 {
405 	kvm_multiple_exception(vcpu, nr, false, 0, true);
406 }
407 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
408 
409 void kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
410 {
411 	if (err)
412 		kvm_inject_gp(vcpu, 0);
413 	else
414 		kvm_x86_ops->skip_emulated_instruction(vcpu);
415 }
416 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
417 
418 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
419 {
420 	++vcpu->stat.pf_guest;
421 	vcpu->arch.cr2 = fault->address;
422 	kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
423 }
424 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
425 
426 static bool kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
427 {
428 	if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
429 		vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
430 	else
431 		vcpu->arch.mmu.inject_page_fault(vcpu, fault);
432 
433 	return fault->nested_page_fault;
434 }
435 
436 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
437 {
438 	atomic_inc(&vcpu->arch.nmi_queued);
439 	kvm_make_request(KVM_REQ_NMI, vcpu);
440 }
441 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
442 
443 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
444 {
445 	kvm_multiple_exception(vcpu, nr, true, error_code, false);
446 }
447 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
448 
449 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
450 {
451 	kvm_multiple_exception(vcpu, nr, true, error_code, true);
452 }
453 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
454 
455 /*
456  * Checks if cpl <= required_cpl; if true, return true.  Otherwise queue
457  * a #GP and return false.
458  */
459 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
460 {
461 	if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
462 		return true;
463 	kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
464 	return false;
465 }
466 EXPORT_SYMBOL_GPL(kvm_require_cpl);
467 
468 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
469 {
470 	if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
471 		return true;
472 
473 	kvm_queue_exception(vcpu, UD_VECTOR);
474 	return false;
475 }
476 EXPORT_SYMBOL_GPL(kvm_require_dr);
477 
478 /*
479  * This function will be used to read from the physical memory of the currently
480  * running guest. The difference to kvm_vcpu_read_guest_page is that this function
481  * can read from guest physical or from the guest's guest physical memory.
482  */
483 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
484 			    gfn_t ngfn, void *data, int offset, int len,
485 			    u32 access)
486 {
487 	struct x86_exception exception;
488 	gfn_t real_gfn;
489 	gpa_t ngpa;
490 
491 	ngpa     = gfn_to_gpa(ngfn);
492 	real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
493 	if (real_gfn == UNMAPPED_GVA)
494 		return -EFAULT;
495 
496 	real_gfn = gpa_to_gfn(real_gfn);
497 
498 	return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
499 }
500 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
501 
502 static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
503 			       void *data, int offset, int len, u32 access)
504 {
505 	return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
506 				       data, offset, len, access);
507 }
508 
509 /*
510  * Load the pae pdptrs.  Return true is they are all valid.
511  */
512 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
513 {
514 	gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
515 	unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
516 	int i;
517 	int ret;
518 	u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
519 
520 	ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
521 				      offset * sizeof(u64), sizeof(pdpte),
522 				      PFERR_USER_MASK|PFERR_WRITE_MASK);
523 	if (ret < 0) {
524 		ret = 0;
525 		goto out;
526 	}
527 	for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
528 		if (is_present_gpte(pdpte[i]) &&
529 		    (pdpte[i] & vcpu->arch.mmu.rsvd_bits_mask[0][2])) {
530 			ret = 0;
531 			goto out;
532 		}
533 	}
534 	ret = 1;
535 
536 	memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
537 	__set_bit(VCPU_EXREG_PDPTR,
538 		  (unsigned long *)&vcpu->arch.regs_avail);
539 	__set_bit(VCPU_EXREG_PDPTR,
540 		  (unsigned long *)&vcpu->arch.regs_dirty);
541 out:
542 
543 	return ret;
544 }
545 EXPORT_SYMBOL_GPL(load_pdptrs);
546 
547 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
548 {
549 	u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
550 	bool changed = true;
551 	int offset;
552 	gfn_t gfn;
553 	int r;
554 
555 	if (is_long_mode(vcpu) || !is_pae(vcpu))
556 		return false;
557 
558 	if (!test_bit(VCPU_EXREG_PDPTR,
559 		      (unsigned long *)&vcpu->arch.regs_avail))
560 		return true;
561 
562 	gfn = (kvm_read_cr3(vcpu) & ~31u) >> PAGE_SHIFT;
563 	offset = (kvm_read_cr3(vcpu) & ~31u) & (PAGE_SIZE - 1);
564 	r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
565 				       PFERR_USER_MASK | PFERR_WRITE_MASK);
566 	if (r < 0)
567 		goto out;
568 	changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
569 out:
570 
571 	return changed;
572 }
573 
574 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
575 {
576 	unsigned long old_cr0 = kvm_read_cr0(vcpu);
577 	unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
578 
579 	cr0 |= X86_CR0_ET;
580 
581 #ifdef CONFIG_X86_64
582 	if (cr0 & 0xffffffff00000000UL)
583 		return 1;
584 #endif
585 
586 	cr0 &= ~CR0_RESERVED_BITS;
587 
588 	if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
589 		return 1;
590 
591 	if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
592 		return 1;
593 
594 	if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
595 #ifdef CONFIG_X86_64
596 		if ((vcpu->arch.efer & EFER_LME)) {
597 			int cs_db, cs_l;
598 
599 			if (!is_pae(vcpu))
600 				return 1;
601 			kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
602 			if (cs_l)
603 				return 1;
604 		} else
605 #endif
606 		if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
607 						 kvm_read_cr3(vcpu)))
608 			return 1;
609 	}
610 
611 	if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
612 		return 1;
613 
614 	kvm_x86_ops->set_cr0(vcpu, cr0);
615 
616 	if ((cr0 ^ old_cr0) & X86_CR0_PG) {
617 		kvm_clear_async_pf_completion_queue(vcpu);
618 		kvm_async_pf_hash_reset(vcpu);
619 	}
620 
621 	if ((cr0 ^ old_cr0) & update_bits)
622 		kvm_mmu_reset_context(vcpu);
623 
624 	if ((cr0 ^ old_cr0) & X86_CR0_CD)
625 		kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
626 
627 	return 0;
628 }
629 EXPORT_SYMBOL_GPL(kvm_set_cr0);
630 
631 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
632 {
633 	(void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
634 }
635 EXPORT_SYMBOL_GPL(kvm_lmsw);
636 
637 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
638 {
639 	if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
640 			!vcpu->guest_xcr0_loaded) {
641 		/* kvm_set_xcr() also depends on this */
642 		xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
643 		vcpu->guest_xcr0_loaded = 1;
644 	}
645 }
646 
647 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
648 {
649 	if (vcpu->guest_xcr0_loaded) {
650 		if (vcpu->arch.xcr0 != host_xcr0)
651 			xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
652 		vcpu->guest_xcr0_loaded = 0;
653 	}
654 }
655 
656 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
657 {
658 	u64 xcr0 = xcr;
659 	u64 old_xcr0 = vcpu->arch.xcr0;
660 	u64 valid_bits;
661 
662 	/* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now  */
663 	if (index != XCR_XFEATURE_ENABLED_MASK)
664 		return 1;
665 	if (!(xcr0 & XSTATE_FP))
666 		return 1;
667 	if ((xcr0 & XSTATE_YMM) && !(xcr0 & XSTATE_SSE))
668 		return 1;
669 
670 	/*
671 	 * Do not allow the guest to set bits that we do not support
672 	 * saving.  However, xcr0 bit 0 is always set, even if the
673 	 * emulated CPU does not support XSAVE (see fx_init).
674 	 */
675 	valid_bits = vcpu->arch.guest_supported_xcr0 | XSTATE_FP;
676 	if (xcr0 & ~valid_bits)
677 		return 1;
678 
679 	if ((!(xcr0 & XSTATE_BNDREGS)) != (!(xcr0 & XSTATE_BNDCSR)))
680 		return 1;
681 
682 	if (xcr0 & XSTATE_AVX512) {
683 		if (!(xcr0 & XSTATE_YMM))
684 			return 1;
685 		if ((xcr0 & XSTATE_AVX512) != XSTATE_AVX512)
686 			return 1;
687 	}
688 	kvm_put_guest_xcr0(vcpu);
689 	vcpu->arch.xcr0 = xcr0;
690 
691 	if ((xcr0 ^ old_xcr0) & XSTATE_EXTEND_MASK)
692 		kvm_update_cpuid(vcpu);
693 	return 0;
694 }
695 
696 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
697 {
698 	if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
699 	    __kvm_set_xcr(vcpu, index, xcr)) {
700 		kvm_inject_gp(vcpu, 0);
701 		return 1;
702 	}
703 	return 0;
704 }
705 EXPORT_SYMBOL_GPL(kvm_set_xcr);
706 
707 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
708 {
709 	unsigned long old_cr4 = kvm_read_cr4(vcpu);
710 	unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
711 				   X86_CR4_SMEP | X86_CR4_SMAP;
712 
713 	if (cr4 & CR4_RESERVED_BITS)
714 		return 1;
715 
716 	if (!guest_cpuid_has_xsave(vcpu) && (cr4 & X86_CR4_OSXSAVE))
717 		return 1;
718 
719 	if (!guest_cpuid_has_smep(vcpu) && (cr4 & X86_CR4_SMEP))
720 		return 1;
721 
722 	if (!guest_cpuid_has_smap(vcpu) && (cr4 & X86_CR4_SMAP))
723 		return 1;
724 
725 	if (!guest_cpuid_has_fsgsbase(vcpu) && (cr4 & X86_CR4_FSGSBASE))
726 		return 1;
727 
728 	if (is_long_mode(vcpu)) {
729 		if (!(cr4 & X86_CR4_PAE))
730 			return 1;
731 	} else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
732 		   && ((cr4 ^ old_cr4) & pdptr_bits)
733 		   && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
734 				   kvm_read_cr3(vcpu)))
735 		return 1;
736 
737 	if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
738 		if (!guest_cpuid_has_pcid(vcpu))
739 			return 1;
740 
741 		/* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
742 		if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
743 			return 1;
744 	}
745 
746 	if (kvm_x86_ops->set_cr4(vcpu, cr4))
747 		return 1;
748 
749 	if (((cr4 ^ old_cr4) & pdptr_bits) ||
750 	    (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
751 		kvm_mmu_reset_context(vcpu);
752 
753 	if ((cr4 ^ old_cr4) & X86_CR4_OSXSAVE)
754 		kvm_update_cpuid(vcpu);
755 
756 	return 0;
757 }
758 EXPORT_SYMBOL_GPL(kvm_set_cr4);
759 
760 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
761 {
762 #ifdef CONFIG_X86_64
763 	cr3 &= ~CR3_PCID_INVD;
764 #endif
765 
766 	if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
767 		kvm_mmu_sync_roots(vcpu);
768 		kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
769 		return 0;
770 	}
771 
772 	if (is_long_mode(vcpu)) {
773 		if (cr3 & CR3_L_MODE_RESERVED_BITS)
774 			return 1;
775 	} else if (is_pae(vcpu) && is_paging(vcpu) &&
776 		   !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
777 		return 1;
778 
779 	vcpu->arch.cr3 = cr3;
780 	__set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
781 	kvm_mmu_new_cr3(vcpu);
782 	return 0;
783 }
784 EXPORT_SYMBOL_GPL(kvm_set_cr3);
785 
786 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
787 {
788 	if (cr8 & CR8_RESERVED_BITS)
789 		return 1;
790 	if (irqchip_in_kernel(vcpu->kvm))
791 		kvm_lapic_set_tpr(vcpu, cr8);
792 	else
793 		vcpu->arch.cr8 = cr8;
794 	return 0;
795 }
796 EXPORT_SYMBOL_GPL(kvm_set_cr8);
797 
798 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
799 {
800 	if (irqchip_in_kernel(vcpu->kvm))
801 		return kvm_lapic_get_cr8(vcpu);
802 	else
803 		return vcpu->arch.cr8;
804 }
805 EXPORT_SYMBOL_GPL(kvm_get_cr8);
806 
807 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
808 {
809 	int i;
810 
811 	if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
812 		for (i = 0; i < KVM_NR_DB_REGS; i++)
813 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
814 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
815 	}
816 }
817 
818 static void kvm_update_dr6(struct kvm_vcpu *vcpu)
819 {
820 	if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
821 		kvm_x86_ops->set_dr6(vcpu, vcpu->arch.dr6);
822 }
823 
824 static void kvm_update_dr7(struct kvm_vcpu *vcpu)
825 {
826 	unsigned long dr7;
827 
828 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
829 		dr7 = vcpu->arch.guest_debug_dr7;
830 	else
831 		dr7 = vcpu->arch.dr7;
832 	kvm_x86_ops->set_dr7(vcpu, dr7);
833 	vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
834 	if (dr7 & DR7_BP_EN_MASK)
835 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
836 }
837 
838 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
839 {
840 	u64 fixed = DR6_FIXED_1;
841 
842 	if (!guest_cpuid_has_rtm(vcpu))
843 		fixed |= DR6_RTM;
844 	return fixed;
845 }
846 
847 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
848 {
849 	switch (dr) {
850 	case 0 ... 3:
851 		vcpu->arch.db[dr] = val;
852 		if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
853 			vcpu->arch.eff_db[dr] = val;
854 		break;
855 	case 4:
856 		/* fall through */
857 	case 6:
858 		if (val & 0xffffffff00000000ULL)
859 			return -1; /* #GP */
860 		vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
861 		kvm_update_dr6(vcpu);
862 		break;
863 	case 5:
864 		/* fall through */
865 	default: /* 7 */
866 		if (val & 0xffffffff00000000ULL)
867 			return -1; /* #GP */
868 		vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
869 		kvm_update_dr7(vcpu);
870 		break;
871 	}
872 
873 	return 0;
874 }
875 
876 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
877 {
878 	if (__kvm_set_dr(vcpu, dr, val)) {
879 		kvm_inject_gp(vcpu, 0);
880 		return 1;
881 	}
882 	return 0;
883 }
884 EXPORT_SYMBOL_GPL(kvm_set_dr);
885 
886 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
887 {
888 	switch (dr) {
889 	case 0 ... 3:
890 		*val = vcpu->arch.db[dr];
891 		break;
892 	case 4:
893 		/* fall through */
894 	case 6:
895 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
896 			*val = vcpu->arch.dr6;
897 		else
898 			*val = kvm_x86_ops->get_dr6(vcpu);
899 		break;
900 	case 5:
901 		/* fall through */
902 	default: /* 7 */
903 		*val = vcpu->arch.dr7;
904 		break;
905 	}
906 	return 0;
907 }
908 EXPORT_SYMBOL_GPL(kvm_get_dr);
909 
910 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
911 {
912 	u32 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
913 	u64 data;
914 	int err;
915 
916 	err = kvm_pmu_rdpmc(vcpu, ecx, &data);
917 	if (err)
918 		return err;
919 	kvm_register_write(vcpu, VCPU_REGS_RAX, (u32)data);
920 	kvm_register_write(vcpu, VCPU_REGS_RDX, data >> 32);
921 	return err;
922 }
923 EXPORT_SYMBOL_GPL(kvm_rdpmc);
924 
925 /*
926  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
927  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
928  *
929  * This list is modified at module load time to reflect the
930  * capabilities of the host cpu. This capabilities test skips MSRs that are
931  * kvm-specific. Those are put in emulated_msrs; filtering of emulated_msrs
932  * may depend on host virtualization features rather than host cpu features.
933  */
934 
935 static u32 msrs_to_save[] = {
936 	MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
937 	MSR_STAR,
938 #ifdef CONFIG_X86_64
939 	MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
940 #endif
941 	MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
942 	MSR_IA32_FEATURE_CONTROL, MSR_IA32_BNDCFGS
943 };
944 
945 static unsigned num_msrs_to_save;
946 
947 static u32 emulated_msrs[] = {
948 	MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
949 	MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
950 	HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
951 	HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
952 	HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
953 	MSR_KVM_PV_EOI_EN,
954 
955 	MSR_IA32_TSC_ADJUST,
956 	MSR_IA32_TSCDEADLINE,
957 	MSR_IA32_MISC_ENABLE,
958 	MSR_IA32_MCG_STATUS,
959 	MSR_IA32_MCG_CTL,
960 	MSR_IA32_SMBASE,
961 };
962 
963 static unsigned num_emulated_msrs;
964 
965 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
966 {
967 	if (efer & efer_reserved_bits)
968 		return false;
969 
970 	if (efer & EFER_FFXSR) {
971 		struct kvm_cpuid_entry2 *feat;
972 
973 		feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
974 		if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
975 			return false;
976 	}
977 
978 	if (efer & EFER_SVME) {
979 		struct kvm_cpuid_entry2 *feat;
980 
981 		feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
982 		if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
983 			return false;
984 	}
985 
986 	return true;
987 }
988 EXPORT_SYMBOL_GPL(kvm_valid_efer);
989 
990 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
991 {
992 	u64 old_efer = vcpu->arch.efer;
993 
994 	if (!kvm_valid_efer(vcpu, efer))
995 		return 1;
996 
997 	if (is_paging(vcpu)
998 	    && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
999 		return 1;
1000 
1001 	efer &= ~EFER_LMA;
1002 	efer |= vcpu->arch.efer & EFER_LMA;
1003 
1004 	kvm_x86_ops->set_efer(vcpu, efer);
1005 
1006 	/* Update reserved bits */
1007 	if ((efer ^ old_efer) & EFER_NX)
1008 		kvm_mmu_reset_context(vcpu);
1009 
1010 	return 0;
1011 }
1012 
1013 void kvm_enable_efer_bits(u64 mask)
1014 {
1015        efer_reserved_bits &= ~mask;
1016 }
1017 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1018 
1019 /*
1020  * Writes msr value into into the appropriate "register".
1021  * Returns 0 on success, non-0 otherwise.
1022  * Assumes vcpu_load() was already called.
1023  */
1024 int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
1025 {
1026 	switch (msr->index) {
1027 	case MSR_FS_BASE:
1028 	case MSR_GS_BASE:
1029 	case MSR_KERNEL_GS_BASE:
1030 	case MSR_CSTAR:
1031 	case MSR_LSTAR:
1032 		if (is_noncanonical_address(msr->data))
1033 			return 1;
1034 		break;
1035 	case MSR_IA32_SYSENTER_EIP:
1036 	case MSR_IA32_SYSENTER_ESP:
1037 		/*
1038 		 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1039 		 * non-canonical address is written on Intel but not on
1040 		 * AMD (which ignores the top 32-bits, because it does
1041 		 * not implement 64-bit SYSENTER).
1042 		 *
1043 		 * 64-bit code should hence be able to write a non-canonical
1044 		 * value on AMD.  Making the address canonical ensures that
1045 		 * vmentry does not fail on Intel after writing a non-canonical
1046 		 * value, and that something deterministic happens if the guest
1047 		 * invokes 64-bit SYSENTER.
1048 		 */
1049 		msr->data = get_canonical(msr->data);
1050 	}
1051 	return kvm_x86_ops->set_msr(vcpu, msr);
1052 }
1053 EXPORT_SYMBOL_GPL(kvm_set_msr);
1054 
1055 /*
1056  * Adapt set_msr() to msr_io()'s calling convention
1057  */
1058 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1059 {
1060 	struct msr_data msr;
1061 	int r;
1062 
1063 	msr.index = index;
1064 	msr.host_initiated = true;
1065 	r = kvm_get_msr(vcpu, &msr);
1066 	if (r)
1067 		return r;
1068 
1069 	*data = msr.data;
1070 	return 0;
1071 }
1072 
1073 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1074 {
1075 	struct msr_data msr;
1076 
1077 	msr.data = *data;
1078 	msr.index = index;
1079 	msr.host_initiated = true;
1080 	return kvm_set_msr(vcpu, &msr);
1081 }
1082 
1083 #ifdef CONFIG_X86_64
1084 struct pvclock_gtod_data {
1085 	seqcount_t	seq;
1086 
1087 	struct { /* extract of a clocksource struct */
1088 		int vclock_mode;
1089 		cycle_t	cycle_last;
1090 		cycle_t	mask;
1091 		u32	mult;
1092 		u32	shift;
1093 	} clock;
1094 
1095 	u64		boot_ns;
1096 	u64		nsec_base;
1097 };
1098 
1099 static struct pvclock_gtod_data pvclock_gtod_data;
1100 
1101 static void update_pvclock_gtod(struct timekeeper *tk)
1102 {
1103 	struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1104 	u64 boot_ns;
1105 
1106 	boot_ns = ktime_to_ns(ktime_add(tk->tkr_mono.base, tk->offs_boot));
1107 
1108 	write_seqcount_begin(&vdata->seq);
1109 
1110 	/* copy pvclock gtod data */
1111 	vdata->clock.vclock_mode	= tk->tkr_mono.clock->archdata.vclock_mode;
1112 	vdata->clock.cycle_last		= tk->tkr_mono.cycle_last;
1113 	vdata->clock.mask		= tk->tkr_mono.mask;
1114 	vdata->clock.mult		= tk->tkr_mono.mult;
1115 	vdata->clock.shift		= tk->tkr_mono.shift;
1116 
1117 	vdata->boot_ns			= boot_ns;
1118 	vdata->nsec_base		= tk->tkr_mono.xtime_nsec;
1119 
1120 	write_seqcount_end(&vdata->seq);
1121 }
1122 #endif
1123 
1124 void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1125 {
1126 	/*
1127 	 * Note: KVM_REQ_PENDING_TIMER is implicitly checked in
1128 	 * vcpu_enter_guest.  This function is only called from
1129 	 * the physical CPU that is running vcpu.
1130 	 */
1131 	kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1132 }
1133 
1134 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1135 {
1136 	int version;
1137 	int r;
1138 	struct pvclock_wall_clock wc;
1139 	struct timespec boot;
1140 
1141 	if (!wall_clock)
1142 		return;
1143 
1144 	r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1145 	if (r)
1146 		return;
1147 
1148 	if (version & 1)
1149 		++version;  /* first time write, random junk */
1150 
1151 	++version;
1152 
1153 	kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1154 
1155 	/*
1156 	 * The guest calculates current wall clock time by adding
1157 	 * system time (updated by kvm_guest_time_update below) to the
1158 	 * wall clock specified here.  guest system time equals host
1159 	 * system time for us, thus we must fill in host boot time here.
1160 	 */
1161 	getboottime(&boot);
1162 
1163 	if (kvm->arch.kvmclock_offset) {
1164 		struct timespec ts = ns_to_timespec(kvm->arch.kvmclock_offset);
1165 		boot = timespec_sub(boot, ts);
1166 	}
1167 	wc.sec = boot.tv_sec;
1168 	wc.nsec = boot.tv_nsec;
1169 	wc.version = version;
1170 
1171 	kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1172 
1173 	version++;
1174 	kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1175 }
1176 
1177 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1178 {
1179 	uint32_t quotient, remainder;
1180 
1181 	/* Don't try to replace with do_div(), this one calculates
1182 	 * "(dividend << 32) / divisor" */
1183 	__asm__ ( "divl %4"
1184 		  : "=a" (quotient), "=d" (remainder)
1185 		  : "0" (0), "1" (dividend), "r" (divisor) );
1186 	return quotient;
1187 }
1188 
1189 static void kvm_get_time_scale(uint32_t scaled_khz, uint32_t base_khz,
1190 			       s8 *pshift, u32 *pmultiplier)
1191 {
1192 	uint64_t scaled64;
1193 	int32_t  shift = 0;
1194 	uint64_t tps64;
1195 	uint32_t tps32;
1196 
1197 	tps64 = base_khz * 1000LL;
1198 	scaled64 = scaled_khz * 1000LL;
1199 	while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1200 		tps64 >>= 1;
1201 		shift--;
1202 	}
1203 
1204 	tps32 = (uint32_t)tps64;
1205 	while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1206 		if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1207 			scaled64 >>= 1;
1208 		else
1209 			tps32 <<= 1;
1210 		shift++;
1211 	}
1212 
1213 	*pshift = shift;
1214 	*pmultiplier = div_frac(scaled64, tps32);
1215 
1216 	pr_debug("%s: base_khz %u => %u, shift %d, mul %u\n",
1217 		 __func__, base_khz, scaled_khz, shift, *pmultiplier);
1218 }
1219 
1220 static inline u64 get_kernel_ns(void)
1221 {
1222 	return ktime_get_boot_ns();
1223 }
1224 
1225 #ifdef CONFIG_X86_64
1226 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1227 #endif
1228 
1229 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1230 static unsigned long max_tsc_khz;
1231 
1232 static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec)
1233 {
1234 	return pvclock_scale_delta(nsec, vcpu->arch.virtual_tsc_mult,
1235 				   vcpu->arch.virtual_tsc_shift);
1236 }
1237 
1238 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1239 {
1240 	u64 v = (u64)khz * (1000000 + ppm);
1241 	do_div(v, 1000000);
1242 	return v;
1243 }
1244 
1245 static void kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 this_tsc_khz)
1246 {
1247 	u32 thresh_lo, thresh_hi;
1248 	int use_scaling = 0;
1249 
1250 	/* tsc_khz can be zero if TSC calibration fails */
1251 	if (this_tsc_khz == 0)
1252 		return;
1253 
1254 	/* Compute a scale to convert nanoseconds in TSC cycles */
1255 	kvm_get_time_scale(this_tsc_khz, NSEC_PER_SEC / 1000,
1256 			   &vcpu->arch.virtual_tsc_shift,
1257 			   &vcpu->arch.virtual_tsc_mult);
1258 	vcpu->arch.virtual_tsc_khz = this_tsc_khz;
1259 
1260 	/*
1261 	 * Compute the variation in TSC rate which is acceptable
1262 	 * within the range of tolerance and decide if the
1263 	 * rate being applied is within that bounds of the hardware
1264 	 * rate.  If so, no scaling or compensation need be done.
1265 	 */
1266 	thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1267 	thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1268 	if (this_tsc_khz < thresh_lo || this_tsc_khz > thresh_hi) {
1269 		pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", this_tsc_khz, thresh_lo, thresh_hi);
1270 		use_scaling = 1;
1271 	}
1272 	kvm_x86_ops->set_tsc_khz(vcpu, this_tsc_khz, use_scaling);
1273 }
1274 
1275 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1276 {
1277 	u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1278 				      vcpu->arch.virtual_tsc_mult,
1279 				      vcpu->arch.virtual_tsc_shift);
1280 	tsc += vcpu->arch.this_tsc_write;
1281 	return tsc;
1282 }
1283 
1284 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1285 {
1286 #ifdef CONFIG_X86_64
1287 	bool vcpus_matched;
1288 	struct kvm_arch *ka = &vcpu->kvm->arch;
1289 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1290 
1291 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1292 			 atomic_read(&vcpu->kvm->online_vcpus));
1293 
1294 	/*
1295 	 * Once the masterclock is enabled, always perform request in
1296 	 * order to update it.
1297 	 *
1298 	 * In order to enable masterclock, the host clocksource must be TSC
1299 	 * and the vcpus need to have matched TSCs.  When that happens,
1300 	 * perform request to enable masterclock.
1301 	 */
1302 	if (ka->use_master_clock ||
1303 	    (gtod->clock.vclock_mode == VCLOCK_TSC && vcpus_matched))
1304 		kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1305 
1306 	trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1307 			    atomic_read(&vcpu->kvm->online_vcpus),
1308 		            ka->use_master_clock, gtod->clock.vclock_mode);
1309 #endif
1310 }
1311 
1312 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1313 {
1314 	u64 curr_offset = kvm_x86_ops->read_tsc_offset(vcpu);
1315 	vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1316 }
1317 
1318 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1319 {
1320 	struct kvm *kvm = vcpu->kvm;
1321 	u64 offset, ns, elapsed;
1322 	unsigned long flags;
1323 	s64 usdiff;
1324 	bool matched;
1325 	bool already_matched;
1326 	u64 data = msr->data;
1327 
1328 	raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1329 	offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
1330 	ns = get_kernel_ns();
1331 	elapsed = ns - kvm->arch.last_tsc_nsec;
1332 
1333 	if (vcpu->arch.virtual_tsc_khz) {
1334 		int faulted = 0;
1335 
1336 		/* n.b - signed multiplication and division required */
1337 		usdiff = data - kvm->arch.last_tsc_write;
1338 #ifdef CONFIG_X86_64
1339 		usdiff = (usdiff * 1000) / vcpu->arch.virtual_tsc_khz;
1340 #else
1341 		/* do_div() only does unsigned */
1342 		asm("1: idivl %[divisor]\n"
1343 		    "2: xor %%edx, %%edx\n"
1344 		    "   movl $0, %[faulted]\n"
1345 		    "3:\n"
1346 		    ".section .fixup,\"ax\"\n"
1347 		    "4: movl $1, %[faulted]\n"
1348 		    "   jmp  3b\n"
1349 		    ".previous\n"
1350 
1351 		_ASM_EXTABLE(1b, 4b)
1352 
1353 		: "=A"(usdiff), [faulted] "=r" (faulted)
1354 		: "A"(usdiff * 1000), [divisor] "rm"(vcpu->arch.virtual_tsc_khz));
1355 
1356 #endif
1357 		do_div(elapsed, 1000);
1358 		usdiff -= elapsed;
1359 		if (usdiff < 0)
1360 			usdiff = -usdiff;
1361 
1362 		/* idivl overflow => difference is larger than USEC_PER_SEC */
1363 		if (faulted)
1364 			usdiff = USEC_PER_SEC;
1365 	} else
1366 		usdiff = USEC_PER_SEC; /* disable TSC match window below */
1367 
1368 	/*
1369 	 * Special case: TSC write with a small delta (1 second) of virtual
1370 	 * cycle time against real time is interpreted as an attempt to
1371 	 * synchronize the CPU.
1372          *
1373 	 * For a reliable TSC, we can match TSC offsets, and for an unstable
1374 	 * TSC, we add elapsed time in this computation.  We could let the
1375 	 * compensation code attempt to catch up if we fall behind, but
1376 	 * it's better to try to match offsets from the beginning.
1377          */
1378 	if (usdiff < USEC_PER_SEC &&
1379 	    vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1380 		if (!check_tsc_unstable()) {
1381 			offset = kvm->arch.cur_tsc_offset;
1382 			pr_debug("kvm: matched tsc offset for %llu\n", data);
1383 		} else {
1384 			u64 delta = nsec_to_cycles(vcpu, elapsed);
1385 			data += delta;
1386 			offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
1387 			pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1388 		}
1389 		matched = true;
1390 		already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
1391 	} else {
1392 		/*
1393 		 * We split periods of matched TSC writes into generations.
1394 		 * For each generation, we track the original measured
1395 		 * nanosecond time, offset, and write, so if TSCs are in
1396 		 * sync, we can match exact offset, and if not, we can match
1397 		 * exact software computation in compute_guest_tsc()
1398 		 *
1399 		 * These values are tracked in kvm->arch.cur_xxx variables.
1400 		 */
1401 		kvm->arch.cur_tsc_generation++;
1402 		kvm->arch.cur_tsc_nsec = ns;
1403 		kvm->arch.cur_tsc_write = data;
1404 		kvm->arch.cur_tsc_offset = offset;
1405 		matched = false;
1406 		pr_debug("kvm: new tsc generation %llu, clock %llu\n",
1407 			 kvm->arch.cur_tsc_generation, data);
1408 	}
1409 
1410 	/*
1411 	 * We also track th most recent recorded KHZ, write and time to
1412 	 * allow the matching interval to be extended at each write.
1413 	 */
1414 	kvm->arch.last_tsc_nsec = ns;
1415 	kvm->arch.last_tsc_write = data;
1416 	kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1417 
1418 	vcpu->arch.last_guest_tsc = data;
1419 
1420 	/* Keep track of which generation this VCPU has synchronized to */
1421 	vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1422 	vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1423 	vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1424 
1425 	if (guest_cpuid_has_tsc_adjust(vcpu) && !msr->host_initiated)
1426 		update_ia32_tsc_adjust_msr(vcpu, offset);
1427 	kvm_x86_ops->write_tsc_offset(vcpu, offset);
1428 	raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1429 
1430 	spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1431 	if (!matched) {
1432 		kvm->arch.nr_vcpus_matched_tsc = 0;
1433 	} else if (!already_matched) {
1434 		kvm->arch.nr_vcpus_matched_tsc++;
1435 	}
1436 
1437 	kvm_track_tsc_matching(vcpu);
1438 	spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
1439 }
1440 
1441 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1442 
1443 #ifdef CONFIG_X86_64
1444 
1445 static cycle_t read_tsc(void)
1446 {
1447 	cycle_t ret;
1448 	u64 last;
1449 
1450 	/*
1451 	 * Empirically, a fence (of type that depends on the CPU)
1452 	 * before rdtsc is enough to ensure that rdtsc is ordered
1453 	 * with respect to loads.  The various CPU manuals are unclear
1454 	 * as to whether rdtsc can be reordered with later loads,
1455 	 * but no one has ever seen it happen.
1456 	 */
1457 	rdtsc_barrier();
1458 	ret = (cycle_t)vget_cycles();
1459 
1460 	last = pvclock_gtod_data.clock.cycle_last;
1461 
1462 	if (likely(ret >= last))
1463 		return ret;
1464 
1465 	/*
1466 	 * GCC likes to generate cmov here, but this branch is extremely
1467 	 * predictable (it's just a funciton of time and the likely is
1468 	 * very likely) and there's a data dependence, so force GCC
1469 	 * to generate a branch instead.  I don't barrier() because
1470 	 * we don't actually need a barrier, and if this function
1471 	 * ever gets inlined it will generate worse code.
1472 	 */
1473 	asm volatile ("");
1474 	return last;
1475 }
1476 
1477 static inline u64 vgettsc(cycle_t *cycle_now)
1478 {
1479 	long v;
1480 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1481 
1482 	*cycle_now = read_tsc();
1483 
1484 	v = (*cycle_now - gtod->clock.cycle_last) & gtod->clock.mask;
1485 	return v * gtod->clock.mult;
1486 }
1487 
1488 static int do_monotonic_boot(s64 *t, cycle_t *cycle_now)
1489 {
1490 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1491 	unsigned long seq;
1492 	int mode;
1493 	u64 ns;
1494 
1495 	do {
1496 		seq = read_seqcount_begin(&gtod->seq);
1497 		mode = gtod->clock.vclock_mode;
1498 		ns = gtod->nsec_base;
1499 		ns += vgettsc(cycle_now);
1500 		ns >>= gtod->clock.shift;
1501 		ns += gtod->boot_ns;
1502 	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
1503 	*t = ns;
1504 
1505 	return mode;
1506 }
1507 
1508 /* returns true if host is using tsc clocksource */
1509 static bool kvm_get_time_and_clockread(s64 *kernel_ns, cycle_t *cycle_now)
1510 {
1511 	/* checked again under seqlock below */
1512 	if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC)
1513 		return false;
1514 
1515 	return do_monotonic_boot(kernel_ns, cycle_now) == VCLOCK_TSC;
1516 }
1517 #endif
1518 
1519 /*
1520  *
1521  * Assuming a stable TSC across physical CPUS, and a stable TSC
1522  * across virtual CPUs, the following condition is possible.
1523  * Each numbered line represents an event visible to both
1524  * CPUs at the next numbered event.
1525  *
1526  * "timespecX" represents host monotonic time. "tscX" represents
1527  * RDTSC value.
1528  *
1529  * 		VCPU0 on CPU0		|	VCPU1 on CPU1
1530  *
1531  * 1.  read timespec0,tsc0
1532  * 2.					| timespec1 = timespec0 + N
1533  * 					| tsc1 = tsc0 + M
1534  * 3. transition to guest		| transition to guest
1535  * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1536  * 5.				        | ret1 = timespec1 + (rdtsc - tsc1)
1537  * 				        | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1538  *
1539  * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1540  *
1541  * 	- ret0 < ret1
1542  *	- timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1543  *		...
1544  *	- 0 < N - M => M < N
1545  *
1546  * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1547  * always the case (the difference between two distinct xtime instances
1548  * might be smaller then the difference between corresponding TSC reads,
1549  * when updating guest vcpus pvclock areas).
1550  *
1551  * To avoid that problem, do not allow visibility of distinct
1552  * system_timestamp/tsc_timestamp values simultaneously: use a master
1553  * copy of host monotonic time values. Update that master copy
1554  * in lockstep.
1555  *
1556  * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
1557  *
1558  */
1559 
1560 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
1561 {
1562 #ifdef CONFIG_X86_64
1563 	struct kvm_arch *ka = &kvm->arch;
1564 	int vclock_mode;
1565 	bool host_tsc_clocksource, vcpus_matched;
1566 
1567 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1568 			atomic_read(&kvm->online_vcpus));
1569 
1570 	/*
1571 	 * If the host uses TSC clock, then passthrough TSC as stable
1572 	 * to the guest.
1573 	 */
1574 	host_tsc_clocksource = kvm_get_time_and_clockread(
1575 					&ka->master_kernel_ns,
1576 					&ka->master_cycle_now);
1577 
1578 	ka->use_master_clock = host_tsc_clocksource && vcpus_matched
1579 				&& !backwards_tsc_observed
1580 				&& !ka->boot_vcpu_runs_old_kvmclock;
1581 
1582 	if (ka->use_master_clock)
1583 		atomic_set(&kvm_guest_has_master_clock, 1);
1584 
1585 	vclock_mode = pvclock_gtod_data.clock.vclock_mode;
1586 	trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
1587 					vcpus_matched);
1588 #endif
1589 }
1590 
1591 static void kvm_gen_update_masterclock(struct kvm *kvm)
1592 {
1593 #ifdef CONFIG_X86_64
1594 	int i;
1595 	struct kvm_vcpu *vcpu;
1596 	struct kvm_arch *ka = &kvm->arch;
1597 
1598 	spin_lock(&ka->pvclock_gtod_sync_lock);
1599 	kvm_make_mclock_inprogress_request(kvm);
1600 	/* no guest entries from this point */
1601 	pvclock_update_vm_gtod_copy(kvm);
1602 
1603 	kvm_for_each_vcpu(i, vcpu, kvm)
1604 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1605 
1606 	/* guest entries allowed */
1607 	kvm_for_each_vcpu(i, vcpu, kvm)
1608 		clear_bit(KVM_REQ_MCLOCK_INPROGRESS, &vcpu->requests);
1609 
1610 	spin_unlock(&ka->pvclock_gtod_sync_lock);
1611 #endif
1612 }
1613 
1614 static int kvm_guest_time_update(struct kvm_vcpu *v)
1615 {
1616 	unsigned long flags, this_tsc_khz;
1617 	struct kvm_vcpu_arch *vcpu = &v->arch;
1618 	struct kvm_arch *ka = &v->kvm->arch;
1619 	s64 kernel_ns;
1620 	u64 tsc_timestamp, host_tsc;
1621 	struct pvclock_vcpu_time_info guest_hv_clock;
1622 	u8 pvclock_flags;
1623 	bool use_master_clock;
1624 
1625 	kernel_ns = 0;
1626 	host_tsc = 0;
1627 
1628 	/*
1629 	 * If the host uses TSC clock, then passthrough TSC as stable
1630 	 * to the guest.
1631 	 */
1632 	spin_lock(&ka->pvclock_gtod_sync_lock);
1633 	use_master_clock = ka->use_master_clock;
1634 	if (use_master_clock) {
1635 		host_tsc = ka->master_cycle_now;
1636 		kernel_ns = ka->master_kernel_ns;
1637 	}
1638 	spin_unlock(&ka->pvclock_gtod_sync_lock);
1639 
1640 	/* Keep irq disabled to prevent changes to the clock */
1641 	local_irq_save(flags);
1642 	this_tsc_khz = __this_cpu_read(cpu_tsc_khz);
1643 	if (unlikely(this_tsc_khz == 0)) {
1644 		local_irq_restore(flags);
1645 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1646 		return 1;
1647 	}
1648 	if (!use_master_clock) {
1649 		host_tsc = native_read_tsc();
1650 		kernel_ns = get_kernel_ns();
1651 	}
1652 
1653 	tsc_timestamp = kvm_x86_ops->read_l1_tsc(v, host_tsc);
1654 
1655 	/*
1656 	 * We may have to catch up the TSC to match elapsed wall clock
1657 	 * time for two reasons, even if kvmclock is used.
1658 	 *   1) CPU could have been running below the maximum TSC rate
1659 	 *   2) Broken TSC compensation resets the base at each VCPU
1660 	 *      entry to avoid unknown leaps of TSC even when running
1661 	 *      again on the same CPU.  This may cause apparent elapsed
1662 	 *      time to disappear, and the guest to stand still or run
1663 	 *	very slowly.
1664 	 */
1665 	if (vcpu->tsc_catchup) {
1666 		u64 tsc = compute_guest_tsc(v, kernel_ns);
1667 		if (tsc > tsc_timestamp) {
1668 			adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
1669 			tsc_timestamp = tsc;
1670 		}
1671 	}
1672 
1673 	local_irq_restore(flags);
1674 
1675 	if (!vcpu->pv_time_enabled)
1676 		return 0;
1677 
1678 	if (unlikely(vcpu->hw_tsc_khz != this_tsc_khz)) {
1679 		kvm_get_time_scale(NSEC_PER_SEC / 1000, this_tsc_khz,
1680 				   &vcpu->hv_clock.tsc_shift,
1681 				   &vcpu->hv_clock.tsc_to_system_mul);
1682 		vcpu->hw_tsc_khz = this_tsc_khz;
1683 	}
1684 
1685 	/* With all the info we got, fill in the values */
1686 	vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
1687 	vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1688 	vcpu->last_guest_tsc = tsc_timestamp;
1689 
1690 	if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
1691 		&guest_hv_clock, sizeof(guest_hv_clock))))
1692 		return 0;
1693 
1694 	/* This VCPU is paused, but it's legal for a guest to read another
1695 	 * VCPU's kvmclock, so we really have to follow the specification where
1696 	 * it says that version is odd if data is being modified, and even after
1697 	 * it is consistent.
1698 	 *
1699 	 * Version field updates must be kept separate.  This is because
1700 	 * kvm_write_guest_cached might use a "rep movs" instruction, and
1701 	 * writes within a string instruction are weakly ordered.  So there
1702 	 * are three writes overall.
1703 	 *
1704 	 * As a small optimization, only write the version field in the first
1705 	 * and third write.  The vcpu->pv_time cache is still valid, because the
1706 	 * version field is the first in the struct.
1707 	 */
1708 	BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
1709 
1710 	vcpu->hv_clock.version = guest_hv_clock.version + 1;
1711 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1712 				&vcpu->hv_clock,
1713 				sizeof(vcpu->hv_clock.version));
1714 
1715 	smp_wmb();
1716 
1717 	/* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
1718 	pvclock_flags = (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
1719 
1720 	if (vcpu->pvclock_set_guest_stopped_request) {
1721 		pvclock_flags |= PVCLOCK_GUEST_STOPPED;
1722 		vcpu->pvclock_set_guest_stopped_request = false;
1723 	}
1724 
1725 	pvclock_flags |= PVCLOCK_COUNTS_FROM_ZERO;
1726 
1727 	/* If the host uses TSC clocksource, then it is stable */
1728 	if (use_master_clock)
1729 		pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
1730 
1731 	vcpu->hv_clock.flags = pvclock_flags;
1732 
1733 	trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
1734 
1735 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1736 				&vcpu->hv_clock,
1737 				sizeof(vcpu->hv_clock));
1738 
1739 	smp_wmb();
1740 
1741 	vcpu->hv_clock.version++;
1742 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1743 				&vcpu->hv_clock,
1744 				sizeof(vcpu->hv_clock.version));
1745 	return 0;
1746 }
1747 
1748 /*
1749  * kvmclock updates which are isolated to a given vcpu, such as
1750  * vcpu->cpu migration, should not allow system_timestamp from
1751  * the rest of the vcpus to remain static. Otherwise ntp frequency
1752  * correction applies to one vcpu's system_timestamp but not
1753  * the others.
1754  *
1755  * So in those cases, request a kvmclock update for all vcpus.
1756  * We need to rate-limit these requests though, as they can
1757  * considerably slow guests that have a large number of vcpus.
1758  * The time for a remote vcpu to update its kvmclock is bound
1759  * by the delay we use to rate-limit the updates.
1760  */
1761 
1762 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
1763 
1764 static void kvmclock_update_fn(struct work_struct *work)
1765 {
1766 	int i;
1767 	struct delayed_work *dwork = to_delayed_work(work);
1768 	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1769 					   kvmclock_update_work);
1770 	struct kvm *kvm = container_of(ka, struct kvm, arch);
1771 	struct kvm_vcpu *vcpu;
1772 
1773 	kvm_for_each_vcpu(i, vcpu, kvm) {
1774 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1775 		kvm_vcpu_kick(vcpu);
1776 	}
1777 }
1778 
1779 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
1780 {
1781 	struct kvm *kvm = v->kvm;
1782 
1783 	kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1784 	schedule_delayed_work(&kvm->arch.kvmclock_update_work,
1785 					KVMCLOCK_UPDATE_DELAY);
1786 }
1787 
1788 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
1789 
1790 static void kvmclock_sync_fn(struct work_struct *work)
1791 {
1792 	struct delayed_work *dwork = to_delayed_work(work);
1793 	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1794 					   kvmclock_sync_work);
1795 	struct kvm *kvm = container_of(ka, struct kvm, arch);
1796 
1797 	if (!kvmclock_periodic_sync)
1798 		return;
1799 
1800 	schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
1801 	schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
1802 					KVMCLOCK_SYNC_PERIOD);
1803 }
1804 
1805 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1806 {
1807 	u64 mcg_cap = vcpu->arch.mcg_cap;
1808 	unsigned bank_num = mcg_cap & 0xff;
1809 
1810 	switch (msr) {
1811 	case MSR_IA32_MCG_STATUS:
1812 		vcpu->arch.mcg_status = data;
1813 		break;
1814 	case MSR_IA32_MCG_CTL:
1815 		if (!(mcg_cap & MCG_CTL_P))
1816 			return 1;
1817 		if (data != 0 && data != ~(u64)0)
1818 			return -1;
1819 		vcpu->arch.mcg_ctl = data;
1820 		break;
1821 	default:
1822 		if (msr >= MSR_IA32_MC0_CTL &&
1823 		    msr < MSR_IA32_MCx_CTL(bank_num)) {
1824 			u32 offset = msr - MSR_IA32_MC0_CTL;
1825 			/* only 0 or all 1s can be written to IA32_MCi_CTL
1826 			 * some Linux kernels though clear bit 10 in bank 4 to
1827 			 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1828 			 * this to avoid an uncatched #GP in the guest
1829 			 */
1830 			if ((offset & 0x3) == 0 &&
1831 			    data != 0 && (data | (1 << 10)) != ~(u64)0)
1832 				return -1;
1833 			vcpu->arch.mce_banks[offset] = data;
1834 			break;
1835 		}
1836 		return 1;
1837 	}
1838 	return 0;
1839 }
1840 
1841 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1842 {
1843 	struct kvm *kvm = vcpu->kvm;
1844 	int lm = is_long_mode(vcpu);
1845 	u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1846 		: (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1847 	u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1848 		: kvm->arch.xen_hvm_config.blob_size_32;
1849 	u32 page_num = data & ~PAGE_MASK;
1850 	u64 page_addr = data & PAGE_MASK;
1851 	u8 *page;
1852 	int r;
1853 
1854 	r = -E2BIG;
1855 	if (page_num >= blob_size)
1856 		goto out;
1857 	r = -ENOMEM;
1858 	page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
1859 	if (IS_ERR(page)) {
1860 		r = PTR_ERR(page);
1861 		goto out;
1862 	}
1863 	if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
1864 		goto out_free;
1865 	r = 0;
1866 out_free:
1867 	kfree(page);
1868 out:
1869 	return r;
1870 }
1871 
1872 static bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1873 {
1874 	return kvm->arch.hv_hypercall & HV_X64_MSR_HYPERCALL_ENABLE;
1875 }
1876 
1877 static bool kvm_hv_msr_partition_wide(u32 msr)
1878 {
1879 	bool r = false;
1880 	switch (msr) {
1881 	case HV_X64_MSR_GUEST_OS_ID:
1882 	case HV_X64_MSR_HYPERCALL:
1883 	case HV_X64_MSR_REFERENCE_TSC:
1884 	case HV_X64_MSR_TIME_REF_COUNT:
1885 		r = true;
1886 		break;
1887 	}
1888 
1889 	return r;
1890 }
1891 
1892 static int set_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1893 {
1894 	struct kvm *kvm = vcpu->kvm;
1895 
1896 	switch (msr) {
1897 	case HV_X64_MSR_GUEST_OS_ID:
1898 		kvm->arch.hv_guest_os_id = data;
1899 		/* setting guest os id to zero disables hypercall page */
1900 		if (!kvm->arch.hv_guest_os_id)
1901 			kvm->arch.hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1902 		break;
1903 	case HV_X64_MSR_HYPERCALL: {
1904 		u64 gfn;
1905 		unsigned long addr;
1906 		u8 instructions[4];
1907 
1908 		/* if guest os id is not set hypercall should remain disabled */
1909 		if (!kvm->arch.hv_guest_os_id)
1910 			break;
1911 		if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1912 			kvm->arch.hv_hypercall = data;
1913 			break;
1914 		}
1915 		gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
1916 		addr = gfn_to_hva(kvm, gfn);
1917 		if (kvm_is_error_hva(addr))
1918 			return 1;
1919 		kvm_x86_ops->patch_hypercall(vcpu, instructions);
1920 		((unsigned char *)instructions)[3] = 0xc3; /* ret */
1921 		if (__copy_to_user((void __user *)addr, instructions, 4))
1922 			return 1;
1923 		kvm->arch.hv_hypercall = data;
1924 		mark_page_dirty(kvm, gfn);
1925 		break;
1926 	}
1927 	case HV_X64_MSR_REFERENCE_TSC: {
1928 		u64 gfn;
1929 		HV_REFERENCE_TSC_PAGE tsc_ref;
1930 		memset(&tsc_ref, 0, sizeof(tsc_ref));
1931 		kvm->arch.hv_tsc_page = data;
1932 		if (!(data & HV_X64_MSR_TSC_REFERENCE_ENABLE))
1933 			break;
1934 		gfn = data >> HV_X64_MSR_TSC_REFERENCE_ADDRESS_SHIFT;
1935 		if (kvm_write_guest(kvm, gfn << HV_X64_MSR_TSC_REFERENCE_ADDRESS_SHIFT,
1936 			&tsc_ref, sizeof(tsc_ref)))
1937 			return 1;
1938 		mark_page_dirty(kvm, gfn);
1939 		break;
1940 	}
1941 	default:
1942 		vcpu_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1943 			    "data 0x%llx\n", msr, data);
1944 		return 1;
1945 	}
1946 	return 0;
1947 }
1948 
1949 static int set_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1950 {
1951 	switch (msr) {
1952 	case HV_X64_MSR_APIC_ASSIST_PAGE: {
1953 		u64 gfn;
1954 		unsigned long addr;
1955 
1956 		if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) {
1957 			vcpu->arch.hv_vapic = data;
1958 			if (kvm_lapic_enable_pv_eoi(vcpu, 0))
1959 				return 1;
1960 			break;
1961 		}
1962 		gfn = data >> HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT;
1963 		addr = kvm_vcpu_gfn_to_hva(vcpu, gfn);
1964 		if (kvm_is_error_hva(addr))
1965 			return 1;
1966 		if (__clear_user((void __user *)addr, PAGE_SIZE))
1967 			return 1;
1968 		vcpu->arch.hv_vapic = data;
1969 		kvm_vcpu_mark_page_dirty(vcpu, gfn);
1970 		if (kvm_lapic_enable_pv_eoi(vcpu, gfn_to_gpa(gfn) | KVM_MSR_ENABLED))
1971 			return 1;
1972 		break;
1973 	}
1974 	case HV_X64_MSR_EOI:
1975 		return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1976 	case HV_X64_MSR_ICR:
1977 		return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1978 	case HV_X64_MSR_TPR:
1979 		return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1980 	default:
1981 		vcpu_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1982 			    "data 0x%llx\n", msr, data);
1983 		return 1;
1984 	}
1985 
1986 	return 0;
1987 }
1988 
1989 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
1990 {
1991 	gpa_t gpa = data & ~0x3f;
1992 
1993 	/* Bits 2:5 are reserved, Should be zero */
1994 	if (data & 0x3c)
1995 		return 1;
1996 
1997 	vcpu->arch.apf.msr_val = data;
1998 
1999 	if (!(data & KVM_ASYNC_PF_ENABLED)) {
2000 		kvm_clear_async_pf_completion_queue(vcpu);
2001 		kvm_async_pf_hash_reset(vcpu);
2002 		return 0;
2003 	}
2004 
2005 	if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
2006 					sizeof(u32)))
2007 		return 1;
2008 
2009 	vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
2010 	kvm_async_pf_wakeup_all(vcpu);
2011 	return 0;
2012 }
2013 
2014 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2015 {
2016 	vcpu->arch.pv_time_enabled = false;
2017 }
2018 
2019 static void accumulate_steal_time(struct kvm_vcpu *vcpu)
2020 {
2021 	u64 delta;
2022 
2023 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2024 		return;
2025 
2026 	delta = current->sched_info.run_delay - vcpu->arch.st.last_steal;
2027 	vcpu->arch.st.last_steal = current->sched_info.run_delay;
2028 	vcpu->arch.st.accum_steal = delta;
2029 }
2030 
2031 static void record_steal_time(struct kvm_vcpu *vcpu)
2032 {
2033 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2034 		return;
2035 
2036 	if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2037 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
2038 		return;
2039 
2040 	vcpu->arch.st.steal.steal += vcpu->arch.st.accum_steal;
2041 	vcpu->arch.st.steal.version += 2;
2042 	vcpu->arch.st.accum_steal = 0;
2043 
2044 	kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2045 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2046 }
2047 
2048 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2049 {
2050 	bool pr = false;
2051 	u32 msr = msr_info->index;
2052 	u64 data = msr_info->data;
2053 
2054 	switch (msr) {
2055 	case MSR_AMD64_NB_CFG:
2056 	case MSR_IA32_UCODE_REV:
2057 	case MSR_IA32_UCODE_WRITE:
2058 	case MSR_VM_HSAVE_PA:
2059 	case MSR_AMD64_PATCH_LOADER:
2060 	case MSR_AMD64_BU_CFG2:
2061 		break;
2062 
2063 	case MSR_EFER:
2064 		return set_efer(vcpu, data);
2065 	case MSR_K7_HWCR:
2066 		data &= ~(u64)0x40;	/* ignore flush filter disable */
2067 		data &= ~(u64)0x100;	/* ignore ignne emulation enable */
2068 		data &= ~(u64)0x8;	/* ignore TLB cache disable */
2069 		data &= ~(u64)0x40000;  /* ignore Mc status write enable */
2070 		if (data != 0) {
2071 			vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2072 				    data);
2073 			return 1;
2074 		}
2075 		break;
2076 	case MSR_FAM10H_MMIO_CONF_BASE:
2077 		if (data != 0) {
2078 			vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2079 				    "0x%llx\n", data);
2080 			return 1;
2081 		}
2082 		break;
2083 	case MSR_IA32_DEBUGCTLMSR:
2084 		if (!data) {
2085 			/* We support the non-activated case already */
2086 			break;
2087 		} else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2088 			/* Values other than LBR and BTF are vendor-specific,
2089 			   thus reserved and should throw a #GP */
2090 			return 1;
2091 		}
2092 		vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2093 			    __func__, data);
2094 		break;
2095 	case 0x200 ... 0x2ff:
2096 		return kvm_mtrr_set_msr(vcpu, msr, data);
2097 	case MSR_IA32_APICBASE:
2098 		return kvm_set_apic_base(vcpu, msr_info);
2099 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2100 		return kvm_x2apic_msr_write(vcpu, msr, data);
2101 	case MSR_IA32_TSCDEADLINE:
2102 		kvm_set_lapic_tscdeadline_msr(vcpu, data);
2103 		break;
2104 	case MSR_IA32_TSC_ADJUST:
2105 		if (guest_cpuid_has_tsc_adjust(vcpu)) {
2106 			if (!msr_info->host_initiated) {
2107 				s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2108 				kvm_x86_ops->adjust_tsc_offset(vcpu, adj, true);
2109 			}
2110 			vcpu->arch.ia32_tsc_adjust_msr = data;
2111 		}
2112 		break;
2113 	case MSR_IA32_MISC_ENABLE:
2114 		vcpu->arch.ia32_misc_enable_msr = data;
2115 		break;
2116 	case MSR_IA32_SMBASE:
2117 		if (!msr_info->host_initiated)
2118 			return 1;
2119 		vcpu->arch.smbase = data;
2120 		break;
2121 	case MSR_KVM_WALL_CLOCK_NEW:
2122 	case MSR_KVM_WALL_CLOCK:
2123 		vcpu->kvm->arch.wall_clock = data;
2124 		kvm_write_wall_clock(vcpu->kvm, data);
2125 		break;
2126 	case MSR_KVM_SYSTEM_TIME_NEW:
2127 	case MSR_KVM_SYSTEM_TIME: {
2128 		u64 gpa_offset;
2129 		struct kvm_arch *ka = &vcpu->kvm->arch;
2130 
2131 		kvmclock_reset(vcpu);
2132 
2133 		if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2134 			bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2135 
2136 			if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2137 				set_bit(KVM_REQ_MASTERCLOCK_UPDATE,
2138 					&vcpu->requests);
2139 
2140 			ka->boot_vcpu_runs_old_kvmclock = tmp;
2141 
2142 			ka->kvmclock_offset = -get_kernel_ns();
2143 		}
2144 
2145 		vcpu->arch.time = data;
2146 		kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2147 
2148 		/* we verify if the enable bit is set... */
2149 		if (!(data & 1))
2150 			break;
2151 
2152 		gpa_offset = data & ~(PAGE_MASK | 1);
2153 
2154 		if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2155 		     &vcpu->arch.pv_time, data & ~1ULL,
2156 		     sizeof(struct pvclock_vcpu_time_info)))
2157 			vcpu->arch.pv_time_enabled = false;
2158 		else
2159 			vcpu->arch.pv_time_enabled = true;
2160 
2161 		break;
2162 	}
2163 	case MSR_KVM_ASYNC_PF_EN:
2164 		if (kvm_pv_enable_async_pf(vcpu, data))
2165 			return 1;
2166 		break;
2167 	case MSR_KVM_STEAL_TIME:
2168 
2169 		if (unlikely(!sched_info_on()))
2170 			return 1;
2171 
2172 		if (data & KVM_STEAL_RESERVED_MASK)
2173 			return 1;
2174 
2175 		if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
2176 						data & KVM_STEAL_VALID_BITS,
2177 						sizeof(struct kvm_steal_time)))
2178 			return 1;
2179 
2180 		vcpu->arch.st.msr_val = data;
2181 
2182 		if (!(data & KVM_MSR_ENABLED))
2183 			break;
2184 
2185 		vcpu->arch.st.last_steal = current->sched_info.run_delay;
2186 
2187 		preempt_disable();
2188 		accumulate_steal_time(vcpu);
2189 		preempt_enable();
2190 
2191 		kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2192 
2193 		break;
2194 	case MSR_KVM_PV_EOI_EN:
2195 		if (kvm_lapic_enable_pv_eoi(vcpu, data))
2196 			return 1;
2197 		break;
2198 
2199 	case MSR_IA32_MCG_CTL:
2200 	case MSR_IA32_MCG_STATUS:
2201 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2202 		return set_msr_mce(vcpu, msr, data);
2203 
2204 	case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2205 	case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2206 		pr = true; /* fall through */
2207 	case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2208 	case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2209 		if (kvm_pmu_is_valid_msr(vcpu, msr))
2210 			return kvm_pmu_set_msr(vcpu, msr_info);
2211 
2212 		if (pr || data != 0)
2213 			vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2214 				    "0x%x data 0x%llx\n", msr, data);
2215 		break;
2216 	case MSR_K7_CLK_CTL:
2217 		/*
2218 		 * Ignore all writes to this no longer documented MSR.
2219 		 * Writes are only relevant for old K7 processors,
2220 		 * all pre-dating SVM, but a recommended workaround from
2221 		 * AMD for these chips. It is possible to specify the
2222 		 * affected processor models on the command line, hence
2223 		 * the need to ignore the workaround.
2224 		 */
2225 		break;
2226 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2227 		if (kvm_hv_msr_partition_wide(msr)) {
2228 			int r;
2229 			mutex_lock(&vcpu->kvm->lock);
2230 			r = set_msr_hyperv_pw(vcpu, msr, data);
2231 			mutex_unlock(&vcpu->kvm->lock);
2232 			return r;
2233 		} else
2234 			return set_msr_hyperv(vcpu, msr, data);
2235 		break;
2236 	case MSR_IA32_BBL_CR_CTL3:
2237 		/* Drop writes to this legacy MSR -- see rdmsr
2238 		 * counterpart for further detail.
2239 		 */
2240 		vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n", msr, data);
2241 		break;
2242 	case MSR_AMD64_OSVW_ID_LENGTH:
2243 		if (!guest_cpuid_has_osvw(vcpu))
2244 			return 1;
2245 		vcpu->arch.osvw.length = data;
2246 		break;
2247 	case MSR_AMD64_OSVW_STATUS:
2248 		if (!guest_cpuid_has_osvw(vcpu))
2249 			return 1;
2250 		vcpu->arch.osvw.status = data;
2251 		break;
2252 	default:
2253 		if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2254 			return xen_hvm_config(vcpu, data);
2255 		if (kvm_pmu_is_valid_msr(vcpu, msr))
2256 			return kvm_pmu_set_msr(vcpu, msr_info);
2257 		if (!ignore_msrs) {
2258 			vcpu_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
2259 				    msr, data);
2260 			return 1;
2261 		} else {
2262 			vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
2263 				    msr, data);
2264 			break;
2265 		}
2266 	}
2267 	return 0;
2268 }
2269 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2270 
2271 
2272 /*
2273  * Reads an msr value (of 'msr_index') into 'pdata'.
2274  * Returns 0 on success, non-0 otherwise.
2275  * Assumes vcpu_load() was already called.
2276  */
2277 int kvm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
2278 {
2279 	return kvm_x86_ops->get_msr(vcpu, msr);
2280 }
2281 EXPORT_SYMBOL_GPL(kvm_get_msr);
2282 
2283 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2284 {
2285 	u64 data;
2286 	u64 mcg_cap = vcpu->arch.mcg_cap;
2287 	unsigned bank_num = mcg_cap & 0xff;
2288 
2289 	switch (msr) {
2290 	case MSR_IA32_P5_MC_ADDR:
2291 	case MSR_IA32_P5_MC_TYPE:
2292 		data = 0;
2293 		break;
2294 	case MSR_IA32_MCG_CAP:
2295 		data = vcpu->arch.mcg_cap;
2296 		break;
2297 	case MSR_IA32_MCG_CTL:
2298 		if (!(mcg_cap & MCG_CTL_P))
2299 			return 1;
2300 		data = vcpu->arch.mcg_ctl;
2301 		break;
2302 	case MSR_IA32_MCG_STATUS:
2303 		data = vcpu->arch.mcg_status;
2304 		break;
2305 	default:
2306 		if (msr >= MSR_IA32_MC0_CTL &&
2307 		    msr < MSR_IA32_MCx_CTL(bank_num)) {
2308 			u32 offset = msr - MSR_IA32_MC0_CTL;
2309 			data = vcpu->arch.mce_banks[offset];
2310 			break;
2311 		}
2312 		return 1;
2313 	}
2314 	*pdata = data;
2315 	return 0;
2316 }
2317 
2318 static int get_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2319 {
2320 	u64 data = 0;
2321 	struct kvm *kvm = vcpu->kvm;
2322 
2323 	switch (msr) {
2324 	case HV_X64_MSR_GUEST_OS_ID:
2325 		data = kvm->arch.hv_guest_os_id;
2326 		break;
2327 	case HV_X64_MSR_HYPERCALL:
2328 		data = kvm->arch.hv_hypercall;
2329 		break;
2330 	case HV_X64_MSR_TIME_REF_COUNT: {
2331 		data =
2332 		     div_u64(get_kernel_ns() + kvm->arch.kvmclock_offset, 100);
2333 		break;
2334 	}
2335 	case HV_X64_MSR_REFERENCE_TSC:
2336 		data = kvm->arch.hv_tsc_page;
2337 		break;
2338 	default:
2339 		vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
2340 		return 1;
2341 	}
2342 
2343 	*pdata = data;
2344 	return 0;
2345 }
2346 
2347 static int get_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2348 {
2349 	u64 data = 0;
2350 
2351 	switch (msr) {
2352 	case HV_X64_MSR_VP_INDEX: {
2353 		int r;
2354 		struct kvm_vcpu *v;
2355 		kvm_for_each_vcpu(r, v, vcpu->kvm) {
2356 			if (v == vcpu) {
2357 				data = r;
2358 				break;
2359 			}
2360 		}
2361 		break;
2362 	}
2363 	case HV_X64_MSR_EOI:
2364 		return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
2365 	case HV_X64_MSR_ICR:
2366 		return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
2367 	case HV_X64_MSR_TPR:
2368 		return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
2369 	case HV_X64_MSR_APIC_ASSIST_PAGE:
2370 		data = vcpu->arch.hv_vapic;
2371 		break;
2372 	default:
2373 		vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
2374 		return 1;
2375 	}
2376 	*pdata = data;
2377 	return 0;
2378 }
2379 
2380 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2381 {
2382 	u64 data;
2383 
2384 	switch (msr_info->index) {
2385 	case MSR_IA32_PLATFORM_ID:
2386 	case MSR_IA32_EBL_CR_POWERON:
2387 	case MSR_IA32_DEBUGCTLMSR:
2388 	case MSR_IA32_LASTBRANCHFROMIP:
2389 	case MSR_IA32_LASTBRANCHTOIP:
2390 	case MSR_IA32_LASTINTFROMIP:
2391 	case MSR_IA32_LASTINTTOIP:
2392 	case MSR_K8_SYSCFG:
2393 	case MSR_K7_HWCR:
2394 	case MSR_VM_HSAVE_PA:
2395 	case MSR_K8_INT_PENDING_MSG:
2396 	case MSR_AMD64_NB_CFG:
2397 	case MSR_FAM10H_MMIO_CONF_BASE:
2398 	case MSR_AMD64_BU_CFG2:
2399 		msr_info->data = 0;
2400 		break;
2401 	case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2402 	case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2403 	case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2404 	case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2405 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2406 			return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2407 		msr_info->data = 0;
2408 		break;
2409 	case MSR_IA32_UCODE_REV:
2410 		msr_info->data = 0x100000000ULL;
2411 		break;
2412 	case MSR_MTRRcap:
2413 	case 0x200 ... 0x2ff:
2414 		return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
2415 	case 0xcd: /* fsb frequency */
2416 		msr_info->data = 3;
2417 		break;
2418 		/*
2419 		 * MSR_EBC_FREQUENCY_ID
2420 		 * Conservative value valid for even the basic CPU models.
2421 		 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2422 		 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2423 		 * and 266MHz for model 3, or 4. Set Core Clock
2424 		 * Frequency to System Bus Frequency Ratio to 1 (bits
2425 		 * 31:24) even though these are only valid for CPU
2426 		 * models > 2, however guests may end up dividing or
2427 		 * multiplying by zero otherwise.
2428 		 */
2429 	case MSR_EBC_FREQUENCY_ID:
2430 		msr_info->data = 1 << 24;
2431 		break;
2432 	case MSR_IA32_APICBASE:
2433 		msr_info->data = kvm_get_apic_base(vcpu);
2434 		break;
2435 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2436 		return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
2437 		break;
2438 	case MSR_IA32_TSCDEADLINE:
2439 		msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
2440 		break;
2441 	case MSR_IA32_TSC_ADJUST:
2442 		msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
2443 		break;
2444 	case MSR_IA32_MISC_ENABLE:
2445 		msr_info->data = vcpu->arch.ia32_misc_enable_msr;
2446 		break;
2447 	case MSR_IA32_SMBASE:
2448 		if (!msr_info->host_initiated)
2449 			return 1;
2450 		msr_info->data = vcpu->arch.smbase;
2451 		break;
2452 	case MSR_IA32_PERF_STATUS:
2453 		/* TSC increment by tick */
2454 		msr_info->data = 1000ULL;
2455 		/* CPU multiplier */
2456 		data |= (((uint64_t)4ULL) << 40);
2457 		break;
2458 	case MSR_EFER:
2459 		msr_info->data = vcpu->arch.efer;
2460 		break;
2461 	case MSR_KVM_WALL_CLOCK:
2462 	case MSR_KVM_WALL_CLOCK_NEW:
2463 		msr_info->data = vcpu->kvm->arch.wall_clock;
2464 		break;
2465 	case MSR_KVM_SYSTEM_TIME:
2466 	case MSR_KVM_SYSTEM_TIME_NEW:
2467 		msr_info->data = vcpu->arch.time;
2468 		break;
2469 	case MSR_KVM_ASYNC_PF_EN:
2470 		msr_info->data = vcpu->arch.apf.msr_val;
2471 		break;
2472 	case MSR_KVM_STEAL_TIME:
2473 		msr_info->data = vcpu->arch.st.msr_val;
2474 		break;
2475 	case MSR_KVM_PV_EOI_EN:
2476 		msr_info->data = vcpu->arch.pv_eoi.msr_val;
2477 		break;
2478 	case MSR_IA32_P5_MC_ADDR:
2479 	case MSR_IA32_P5_MC_TYPE:
2480 	case MSR_IA32_MCG_CAP:
2481 	case MSR_IA32_MCG_CTL:
2482 	case MSR_IA32_MCG_STATUS:
2483 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2484 		return get_msr_mce(vcpu, msr_info->index, &msr_info->data);
2485 	case MSR_K7_CLK_CTL:
2486 		/*
2487 		 * Provide expected ramp-up count for K7. All other
2488 		 * are set to zero, indicating minimum divisors for
2489 		 * every field.
2490 		 *
2491 		 * This prevents guest kernels on AMD host with CPU
2492 		 * type 6, model 8 and higher from exploding due to
2493 		 * the rdmsr failing.
2494 		 */
2495 		msr_info->data = 0x20000000;
2496 		break;
2497 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2498 		if (kvm_hv_msr_partition_wide(msr_info->index)) {
2499 			int r;
2500 			mutex_lock(&vcpu->kvm->lock);
2501 			r = get_msr_hyperv_pw(vcpu, msr_info->index, &msr_info->data);
2502 			mutex_unlock(&vcpu->kvm->lock);
2503 			return r;
2504 		} else
2505 			return get_msr_hyperv(vcpu, msr_info->index, &msr_info->data);
2506 		break;
2507 	case MSR_IA32_BBL_CR_CTL3:
2508 		/* This legacy MSR exists but isn't fully documented in current
2509 		 * silicon.  It is however accessed by winxp in very narrow
2510 		 * scenarios where it sets bit #19, itself documented as
2511 		 * a "reserved" bit.  Best effort attempt to source coherent
2512 		 * read data here should the balance of the register be
2513 		 * interpreted by the guest:
2514 		 *
2515 		 * L2 cache control register 3: 64GB range, 256KB size,
2516 		 * enabled, latency 0x1, configured
2517 		 */
2518 		msr_info->data = 0xbe702111;
2519 		break;
2520 	case MSR_AMD64_OSVW_ID_LENGTH:
2521 		if (!guest_cpuid_has_osvw(vcpu))
2522 			return 1;
2523 		msr_info->data = vcpu->arch.osvw.length;
2524 		break;
2525 	case MSR_AMD64_OSVW_STATUS:
2526 		if (!guest_cpuid_has_osvw(vcpu))
2527 			return 1;
2528 		msr_info->data = vcpu->arch.osvw.status;
2529 		break;
2530 	default:
2531 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2532 			return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2533 		if (!ignore_msrs) {
2534 			vcpu_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr_info->index);
2535 			return 1;
2536 		} else {
2537 			vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr_info->index);
2538 			msr_info->data = 0;
2539 		}
2540 		break;
2541 	}
2542 	return 0;
2543 }
2544 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2545 
2546 /*
2547  * Read or write a bunch of msrs. All parameters are kernel addresses.
2548  *
2549  * @return number of msrs set successfully.
2550  */
2551 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2552 		    struct kvm_msr_entry *entries,
2553 		    int (*do_msr)(struct kvm_vcpu *vcpu,
2554 				  unsigned index, u64 *data))
2555 {
2556 	int i, idx;
2557 
2558 	idx = srcu_read_lock(&vcpu->kvm->srcu);
2559 	for (i = 0; i < msrs->nmsrs; ++i)
2560 		if (do_msr(vcpu, entries[i].index, &entries[i].data))
2561 			break;
2562 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
2563 
2564 	return i;
2565 }
2566 
2567 /*
2568  * Read or write a bunch of msrs. Parameters are user addresses.
2569  *
2570  * @return number of msrs set successfully.
2571  */
2572 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2573 		  int (*do_msr)(struct kvm_vcpu *vcpu,
2574 				unsigned index, u64 *data),
2575 		  int writeback)
2576 {
2577 	struct kvm_msrs msrs;
2578 	struct kvm_msr_entry *entries;
2579 	int r, n;
2580 	unsigned size;
2581 
2582 	r = -EFAULT;
2583 	if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2584 		goto out;
2585 
2586 	r = -E2BIG;
2587 	if (msrs.nmsrs >= MAX_IO_MSRS)
2588 		goto out;
2589 
2590 	size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2591 	entries = memdup_user(user_msrs->entries, size);
2592 	if (IS_ERR(entries)) {
2593 		r = PTR_ERR(entries);
2594 		goto out;
2595 	}
2596 
2597 	r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2598 	if (r < 0)
2599 		goto out_free;
2600 
2601 	r = -EFAULT;
2602 	if (writeback && copy_to_user(user_msrs->entries, entries, size))
2603 		goto out_free;
2604 
2605 	r = n;
2606 
2607 out_free:
2608 	kfree(entries);
2609 out:
2610 	return r;
2611 }
2612 
2613 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
2614 {
2615 	int r;
2616 
2617 	switch (ext) {
2618 	case KVM_CAP_IRQCHIP:
2619 	case KVM_CAP_HLT:
2620 	case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2621 	case KVM_CAP_SET_TSS_ADDR:
2622 	case KVM_CAP_EXT_CPUID:
2623 	case KVM_CAP_EXT_EMUL_CPUID:
2624 	case KVM_CAP_CLOCKSOURCE:
2625 	case KVM_CAP_PIT:
2626 	case KVM_CAP_NOP_IO_DELAY:
2627 	case KVM_CAP_MP_STATE:
2628 	case KVM_CAP_SYNC_MMU:
2629 	case KVM_CAP_USER_NMI:
2630 	case KVM_CAP_REINJECT_CONTROL:
2631 	case KVM_CAP_IRQ_INJECT_STATUS:
2632 	case KVM_CAP_IOEVENTFD:
2633 	case KVM_CAP_IOEVENTFD_NO_LENGTH:
2634 	case KVM_CAP_PIT2:
2635 	case KVM_CAP_PIT_STATE2:
2636 	case KVM_CAP_SET_IDENTITY_MAP_ADDR:
2637 	case KVM_CAP_XEN_HVM:
2638 	case KVM_CAP_ADJUST_CLOCK:
2639 	case KVM_CAP_VCPU_EVENTS:
2640 	case KVM_CAP_HYPERV:
2641 	case KVM_CAP_HYPERV_VAPIC:
2642 	case KVM_CAP_HYPERV_SPIN:
2643 	case KVM_CAP_PCI_SEGMENT:
2644 	case KVM_CAP_DEBUGREGS:
2645 	case KVM_CAP_X86_ROBUST_SINGLESTEP:
2646 	case KVM_CAP_XSAVE:
2647 	case KVM_CAP_ASYNC_PF:
2648 	case KVM_CAP_GET_TSC_KHZ:
2649 	case KVM_CAP_KVMCLOCK_CTRL:
2650 	case KVM_CAP_READONLY_MEM:
2651 	case KVM_CAP_HYPERV_TIME:
2652 	case KVM_CAP_IOAPIC_POLARITY_IGNORED:
2653 	case KVM_CAP_TSC_DEADLINE_TIMER:
2654 	case KVM_CAP_ENABLE_CAP_VM:
2655 	case KVM_CAP_DISABLE_QUIRKS:
2656 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2657 	case KVM_CAP_ASSIGN_DEV_IRQ:
2658 	case KVM_CAP_PCI_2_3:
2659 #endif
2660 		r = 1;
2661 		break;
2662 	case KVM_CAP_X86_SMM:
2663 		/* SMBASE is usually relocated above 1M on modern chipsets,
2664 		 * and SMM handlers might indeed rely on 4G segment limits,
2665 		 * so do not report SMM to be available if real mode is
2666 		 * emulated via vm86 mode.  Still, do not go to great lengths
2667 		 * to avoid userspace's usage of the feature, because it is a
2668 		 * fringe case that is not enabled except via specific settings
2669 		 * of the module parameters.
2670 		 */
2671 		r = kvm_x86_ops->cpu_has_high_real_mode_segbase();
2672 		break;
2673 	case KVM_CAP_COALESCED_MMIO:
2674 		r = KVM_COALESCED_MMIO_PAGE_OFFSET;
2675 		break;
2676 	case KVM_CAP_VAPIC:
2677 		r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2678 		break;
2679 	case KVM_CAP_NR_VCPUS:
2680 		r = KVM_SOFT_MAX_VCPUS;
2681 		break;
2682 	case KVM_CAP_MAX_VCPUS:
2683 		r = KVM_MAX_VCPUS;
2684 		break;
2685 	case KVM_CAP_NR_MEMSLOTS:
2686 		r = KVM_USER_MEM_SLOTS;
2687 		break;
2688 	case KVM_CAP_PV_MMU:	/* obsolete */
2689 		r = 0;
2690 		break;
2691 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2692 	case KVM_CAP_IOMMU:
2693 		r = iommu_present(&pci_bus_type);
2694 		break;
2695 #endif
2696 	case KVM_CAP_MCE:
2697 		r = KVM_MAX_MCE_BANKS;
2698 		break;
2699 	case KVM_CAP_XCRS:
2700 		r = cpu_has_xsave;
2701 		break;
2702 	case KVM_CAP_TSC_CONTROL:
2703 		r = kvm_has_tsc_control;
2704 		break;
2705 	default:
2706 		r = 0;
2707 		break;
2708 	}
2709 	return r;
2710 
2711 }
2712 
2713 long kvm_arch_dev_ioctl(struct file *filp,
2714 			unsigned int ioctl, unsigned long arg)
2715 {
2716 	void __user *argp = (void __user *)arg;
2717 	long r;
2718 
2719 	switch (ioctl) {
2720 	case KVM_GET_MSR_INDEX_LIST: {
2721 		struct kvm_msr_list __user *user_msr_list = argp;
2722 		struct kvm_msr_list msr_list;
2723 		unsigned n;
2724 
2725 		r = -EFAULT;
2726 		if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2727 			goto out;
2728 		n = msr_list.nmsrs;
2729 		msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
2730 		if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2731 			goto out;
2732 		r = -E2BIG;
2733 		if (n < msr_list.nmsrs)
2734 			goto out;
2735 		r = -EFAULT;
2736 		if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2737 				 num_msrs_to_save * sizeof(u32)))
2738 			goto out;
2739 		if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2740 				 &emulated_msrs,
2741 				 num_emulated_msrs * sizeof(u32)))
2742 			goto out;
2743 		r = 0;
2744 		break;
2745 	}
2746 	case KVM_GET_SUPPORTED_CPUID:
2747 	case KVM_GET_EMULATED_CPUID: {
2748 		struct kvm_cpuid2 __user *cpuid_arg = argp;
2749 		struct kvm_cpuid2 cpuid;
2750 
2751 		r = -EFAULT;
2752 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2753 			goto out;
2754 
2755 		r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
2756 					    ioctl);
2757 		if (r)
2758 			goto out;
2759 
2760 		r = -EFAULT;
2761 		if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2762 			goto out;
2763 		r = 0;
2764 		break;
2765 	}
2766 	case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2767 		u64 mce_cap;
2768 
2769 		mce_cap = KVM_MCE_CAP_SUPPORTED;
2770 		r = -EFAULT;
2771 		if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
2772 			goto out;
2773 		r = 0;
2774 		break;
2775 	}
2776 	default:
2777 		r = -EINVAL;
2778 	}
2779 out:
2780 	return r;
2781 }
2782 
2783 static void wbinvd_ipi(void *garbage)
2784 {
2785 	wbinvd();
2786 }
2787 
2788 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2789 {
2790 	return kvm_arch_has_noncoherent_dma(vcpu->kvm);
2791 }
2792 
2793 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2794 {
2795 	/* Address WBINVD may be executed by guest */
2796 	if (need_emulate_wbinvd(vcpu)) {
2797 		if (kvm_x86_ops->has_wbinvd_exit())
2798 			cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2799 		else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2800 			smp_call_function_single(vcpu->cpu,
2801 					wbinvd_ipi, NULL, 1);
2802 	}
2803 
2804 	kvm_x86_ops->vcpu_load(vcpu, cpu);
2805 
2806 	/* Apply any externally detected TSC adjustments (due to suspend) */
2807 	if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
2808 		adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
2809 		vcpu->arch.tsc_offset_adjustment = 0;
2810 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2811 	}
2812 
2813 	if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
2814 		s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
2815 				native_read_tsc() - vcpu->arch.last_host_tsc;
2816 		if (tsc_delta < 0)
2817 			mark_tsc_unstable("KVM discovered backwards TSC");
2818 		if (check_tsc_unstable()) {
2819 			u64 offset = kvm_x86_ops->compute_tsc_offset(vcpu,
2820 						vcpu->arch.last_guest_tsc);
2821 			kvm_x86_ops->write_tsc_offset(vcpu, offset);
2822 			vcpu->arch.tsc_catchup = 1;
2823 		}
2824 		/*
2825 		 * On a host with synchronized TSC, there is no need to update
2826 		 * kvmclock on vcpu->cpu migration
2827 		 */
2828 		if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
2829 			kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2830 		if (vcpu->cpu != cpu)
2831 			kvm_migrate_timers(vcpu);
2832 		vcpu->cpu = cpu;
2833 	}
2834 
2835 	accumulate_steal_time(vcpu);
2836 	kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2837 }
2838 
2839 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2840 {
2841 	kvm_x86_ops->vcpu_put(vcpu);
2842 	kvm_put_guest_fpu(vcpu);
2843 	vcpu->arch.last_host_tsc = native_read_tsc();
2844 }
2845 
2846 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2847 				    struct kvm_lapic_state *s)
2848 {
2849 	kvm_x86_ops->sync_pir_to_irr(vcpu);
2850 	memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
2851 
2852 	return 0;
2853 }
2854 
2855 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2856 				    struct kvm_lapic_state *s)
2857 {
2858 	kvm_apic_post_state_restore(vcpu, s);
2859 	update_cr8_intercept(vcpu);
2860 
2861 	return 0;
2862 }
2863 
2864 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2865 				    struct kvm_interrupt *irq)
2866 {
2867 	if (irq->irq >= KVM_NR_INTERRUPTS)
2868 		return -EINVAL;
2869 	if (irqchip_in_kernel(vcpu->kvm))
2870 		return -ENXIO;
2871 
2872 	kvm_queue_interrupt(vcpu, irq->irq, false);
2873 	kvm_make_request(KVM_REQ_EVENT, vcpu);
2874 
2875 	return 0;
2876 }
2877 
2878 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2879 {
2880 	kvm_inject_nmi(vcpu);
2881 
2882 	return 0;
2883 }
2884 
2885 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
2886 {
2887 	kvm_make_request(KVM_REQ_SMI, vcpu);
2888 
2889 	return 0;
2890 }
2891 
2892 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2893 					   struct kvm_tpr_access_ctl *tac)
2894 {
2895 	if (tac->flags)
2896 		return -EINVAL;
2897 	vcpu->arch.tpr_access_reporting = !!tac->enabled;
2898 	return 0;
2899 }
2900 
2901 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2902 					u64 mcg_cap)
2903 {
2904 	int r;
2905 	unsigned bank_num = mcg_cap & 0xff, bank;
2906 
2907 	r = -EINVAL;
2908 	if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2909 		goto out;
2910 	if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2911 		goto out;
2912 	r = 0;
2913 	vcpu->arch.mcg_cap = mcg_cap;
2914 	/* Init IA32_MCG_CTL to all 1s */
2915 	if (mcg_cap & MCG_CTL_P)
2916 		vcpu->arch.mcg_ctl = ~(u64)0;
2917 	/* Init IA32_MCi_CTL to all 1s */
2918 	for (bank = 0; bank < bank_num; bank++)
2919 		vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2920 out:
2921 	return r;
2922 }
2923 
2924 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2925 				      struct kvm_x86_mce *mce)
2926 {
2927 	u64 mcg_cap = vcpu->arch.mcg_cap;
2928 	unsigned bank_num = mcg_cap & 0xff;
2929 	u64 *banks = vcpu->arch.mce_banks;
2930 
2931 	if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2932 		return -EINVAL;
2933 	/*
2934 	 * if IA32_MCG_CTL is not all 1s, the uncorrected error
2935 	 * reporting is disabled
2936 	 */
2937 	if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2938 	    vcpu->arch.mcg_ctl != ~(u64)0)
2939 		return 0;
2940 	banks += 4 * mce->bank;
2941 	/*
2942 	 * if IA32_MCi_CTL is not all 1s, the uncorrected error
2943 	 * reporting is disabled for the bank
2944 	 */
2945 	if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2946 		return 0;
2947 	if (mce->status & MCI_STATUS_UC) {
2948 		if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
2949 		    !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
2950 			kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2951 			return 0;
2952 		}
2953 		if (banks[1] & MCI_STATUS_VAL)
2954 			mce->status |= MCI_STATUS_OVER;
2955 		banks[2] = mce->addr;
2956 		banks[3] = mce->misc;
2957 		vcpu->arch.mcg_status = mce->mcg_status;
2958 		banks[1] = mce->status;
2959 		kvm_queue_exception(vcpu, MC_VECTOR);
2960 	} else if (!(banks[1] & MCI_STATUS_VAL)
2961 		   || !(banks[1] & MCI_STATUS_UC)) {
2962 		if (banks[1] & MCI_STATUS_VAL)
2963 			mce->status |= MCI_STATUS_OVER;
2964 		banks[2] = mce->addr;
2965 		banks[3] = mce->misc;
2966 		banks[1] = mce->status;
2967 	} else
2968 		banks[1] |= MCI_STATUS_OVER;
2969 	return 0;
2970 }
2971 
2972 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2973 					       struct kvm_vcpu_events *events)
2974 {
2975 	process_nmi(vcpu);
2976 	events->exception.injected =
2977 		vcpu->arch.exception.pending &&
2978 		!kvm_exception_is_soft(vcpu->arch.exception.nr);
2979 	events->exception.nr = vcpu->arch.exception.nr;
2980 	events->exception.has_error_code = vcpu->arch.exception.has_error_code;
2981 	events->exception.pad = 0;
2982 	events->exception.error_code = vcpu->arch.exception.error_code;
2983 
2984 	events->interrupt.injected =
2985 		vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
2986 	events->interrupt.nr = vcpu->arch.interrupt.nr;
2987 	events->interrupt.soft = 0;
2988 	events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
2989 
2990 	events->nmi.injected = vcpu->arch.nmi_injected;
2991 	events->nmi.pending = vcpu->arch.nmi_pending != 0;
2992 	events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
2993 	events->nmi.pad = 0;
2994 
2995 	events->sipi_vector = 0; /* never valid when reporting to user space */
2996 
2997 	events->smi.smm = is_smm(vcpu);
2998 	events->smi.pending = vcpu->arch.smi_pending;
2999 	events->smi.smm_inside_nmi =
3000 		!!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
3001 	events->smi.latched_init = kvm_lapic_latched_init(vcpu);
3002 
3003 	events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
3004 			 | KVM_VCPUEVENT_VALID_SHADOW
3005 			 | KVM_VCPUEVENT_VALID_SMM);
3006 	memset(&events->reserved, 0, sizeof(events->reserved));
3007 }
3008 
3009 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
3010 					      struct kvm_vcpu_events *events)
3011 {
3012 	if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
3013 			      | KVM_VCPUEVENT_VALID_SIPI_VECTOR
3014 			      | KVM_VCPUEVENT_VALID_SHADOW
3015 			      | KVM_VCPUEVENT_VALID_SMM))
3016 		return -EINVAL;
3017 
3018 	process_nmi(vcpu);
3019 	vcpu->arch.exception.pending = events->exception.injected;
3020 	vcpu->arch.exception.nr = events->exception.nr;
3021 	vcpu->arch.exception.has_error_code = events->exception.has_error_code;
3022 	vcpu->arch.exception.error_code = events->exception.error_code;
3023 
3024 	vcpu->arch.interrupt.pending = events->interrupt.injected;
3025 	vcpu->arch.interrupt.nr = events->interrupt.nr;
3026 	vcpu->arch.interrupt.soft = events->interrupt.soft;
3027 	if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
3028 		kvm_x86_ops->set_interrupt_shadow(vcpu,
3029 						  events->interrupt.shadow);
3030 
3031 	vcpu->arch.nmi_injected = events->nmi.injected;
3032 	if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
3033 		vcpu->arch.nmi_pending = events->nmi.pending;
3034 	kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
3035 
3036 	if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
3037 	    kvm_vcpu_has_lapic(vcpu))
3038 		vcpu->arch.apic->sipi_vector = events->sipi_vector;
3039 
3040 	if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
3041 		if (events->smi.smm)
3042 			vcpu->arch.hflags |= HF_SMM_MASK;
3043 		else
3044 			vcpu->arch.hflags &= ~HF_SMM_MASK;
3045 		vcpu->arch.smi_pending = events->smi.pending;
3046 		if (events->smi.smm_inside_nmi)
3047 			vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
3048 		else
3049 			vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
3050 		if (kvm_vcpu_has_lapic(vcpu)) {
3051 			if (events->smi.latched_init)
3052 				set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3053 			else
3054 				clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3055 		}
3056 	}
3057 
3058 	kvm_make_request(KVM_REQ_EVENT, vcpu);
3059 
3060 	return 0;
3061 }
3062 
3063 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3064 					     struct kvm_debugregs *dbgregs)
3065 {
3066 	unsigned long val;
3067 
3068 	memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
3069 	kvm_get_dr(vcpu, 6, &val);
3070 	dbgregs->dr6 = val;
3071 	dbgregs->dr7 = vcpu->arch.dr7;
3072 	dbgregs->flags = 0;
3073 	memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
3074 }
3075 
3076 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
3077 					    struct kvm_debugregs *dbgregs)
3078 {
3079 	if (dbgregs->flags)
3080 		return -EINVAL;
3081 
3082 	memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
3083 	kvm_update_dr0123(vcpu);
3084 	vcpu->arch.dr6 = dbgregs->dr6;
3085 	kvm_update_dr6(vcpu);
3086 	vcpu->arch.dr7 = dbgregs->dr7;
3087 	kvm_update_dr7(vcpu);
3088 
3089 	return 0;
3090 }
3091 
3092 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
3093 
3094 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
3095 {
3096 	struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
3097 	u64 xstate_bv = xsave->header.xfeatures;
3098 	u64 valid;
3099 
3100 	/*
3101 	 * Copy legacy XSAVE area, to avoid complications with CPUID
3102 	 * leaves 0 and 1 in the loop below.
3103 	 */
3104 	memcpy(dest, xsave, XSAVE_HDR_OFFSET);
3105 
3106 	/* Set XSTATE_BV */
3107 	*(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
3108 
3109 	/*
3110 	 * Copy each region from the possibly compacted offset to the
3111 	 * non-compacted offset.
3112 	 */
3113 	valid = xstate_bv & ~XSTATE_FPSSE;
3114 	while (valid) {
3115 		u64 feature = valid & -valid;
3116 		int index = fls64(feature) - 1;
3117 		void *src = get_xsave_addr(xsave, feature);
3118 
3119 		if (src) {
3120 			u32 size, offset, ecx, edx;
3121 			cpuid_count(XSTATE_CPUID, index,
3122 				    &size, &offset, &ecx, &edx);
3123 			memcpy(dest + offset, src, size);
3124 		}
3125 
3126 		valid -= feature;
3127 	}
3128 }
3129 
3130 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
3131 {
3132 	struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
3133 	u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
3134 	u64 valid;
3135 
3136 	/*
3137 	 * Copy legacy XSAVE area, to avoid complications with CPUID
3138 	 * leaves 0 and 1 in the loop below.
3139 	 */
3140 	memcpy(xsave, src, XSAVE_HDR_OFFSET);
3141 
3142 	/* Set XSTATE_BV and possibly XCOMP_BV.  */
3143 	xsave->header.xfeatures = xstate_bv;
3144 	if (cpu_has_xsaves)
3145 		xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
3146 
3147 	/*
3148 	 * Copy each region from the non-compacted offset to the
3149 	 * possibly compacted offset.
3150 	 */
3151 	valid = xstate_bv & ~XSTATE_FPSSE;
3152 	while (valid) {
3153 		u64 feature = valid & -valid;
3154 		int index = fls64(feature) - 1;
3155 		void *dest = get_xsave_addr(xsave, feature);
3156 
3157 		if (dest) {
3158 			u32 size, offset, ecx, edx;
3159 			cpuid_count(XSTATE_CPUID, index,
3160 				    &size, &offset, &ecx, &edx);
3161 			memcpy(dest, src + offset, size);
3162 		} else
3163 			WARN_ON_ONCE(1);
3164 
3165 		valid -= feature;
3166 	}
3167 }
3168 
3169 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
3170 					 struct kvm_xsave *guest_xsave)
3171 {
3172 	if (cpu_has_xsave) {
3173 		memset(guest_xsave, 0, sizeof(struct kvm_xsave));
3174 		fill_xsave((u8 *) guest_xsave->region, vcpu);
3175 	} else {
3176 		memcpy(guest_xsave->region,
3177 			&vcpu->arch.guest_fpu.state.fxsave,
3178 			sizeof(struct fxregs_state));
3179 		*(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
3180 			XSTATE_FPSSE;
3181 	}
3182 }
3183 
3184 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
3185 					struct kvm_xsave *guest_xsave)
3186 {
3187 	u64 xstate_bv =
3188 		*(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
3189 
3190 	if (cpu_has_xsave) {
3191 		/*
3192 		 * Here we allow setting states that are not present in
3193 		 * CPUID leaf 0xD, index 0, EDX:EAX.  This is for compatibility
3194 		 * with old userspace.
3195 		 */
3196 		if (xstate_bv & ~kvm_supported_xcr0())
3197 			return -EINVAL;
3198 		load_xsave(vcpu, (u8 *)guest_xsave->region);
3199 	} else {
3200 		if (xstate_bv & ~XSTATE_FPSSE)
3201 			return -EINVAL;
3202 		memcpy(&vcpu->arch.guest_fpu.state.fxsave,
3203 			guest_xsave->region, sizeof(struct fxregs_state));
3204 	}
3205 	return 0;
3206 }
3207 
3208 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3209 					struct kvm_xcrs *guest_xcrs)
3210 {
3211 	if (!cpu_has_xsave) {
3212 		guest_xcrs->nr_xcrs = 0;
3213 		return;
3214 	}
3215 
3216 	guest_xcrs->nr_xcrs = 1;
3217 	guest_xcrs->flags = 0;
3218 	guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3219 	guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3220 }
3221 
3222 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3223 				       struct kvm_xcrs *guest_xcrs)
3224 {
3225 	int i, r = 0;
3226 
3227 	if (!cpu_has_xsave)
3228 		return -EINVAL;
3229 
3230 	if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
3231 		return -EINVAL;
3232 
3233 	for (i = 0; i < guest_xcrs->nr_xcrs; i++)
3234 		/* Only support XCR0 currently */
3235 		if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
3236 			r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
3237 				guest_xcrs->xcrs[i].value);
3238 			break;
3239 		}
3240 	if (r)
3241 		r = -EINVAL;
3242 	return r;
3243 }
3244 
3245 /*
3246  * kvm_set_guest_paused() indicates to the guest kernel that it has been
3247  * stopped by the hypervisor.  This function will be called from the host only.
3248  * EINVAL is returned when the host attempts to set the flag for a guest that
3249  * does not support pv clocks.
3250  */
3251 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
3252 {
3253 	if (!vcpu->arch.pv_time_enabled)
3254 		return -EINVAL;
3255 	vcpu->arch.pvclock_set_guest_stopped_request = true;
3256 	kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3257 	return 0;
3258 }
3259 
3260 long kvm_arch_vcpu_ioctl(struct file *filp,
3261 			 unsigned int ioctl, unsigned long arg)
3262 {
3263 	struct kvm_vcpu *vcpu = filp->private_data;
3264 	void __user *argp = (void __user *)arg;
3265 	int r;
3266 	union {
3267 		struct kvm_lapic_state *lapic;
3268 		struct kvm_xsave *xsave;
3269 		struct kvm_xcrs *xcrs;
3270 		void *buffer;
3271 	} u;
3272 
3273 	u.buffer = NULL;
3274 	switch (ioctl) {
3275 	case KVM_GET_LAPIC: {
3276 		r = -EINVAL;
3277 		if (!vcpu->arch.apic)
3278 			goto out;
3279 		u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
3280 
3281 		r = -ENOMEM;
3282 		if (!u.lapic)
3283 			goto out;
3284 		r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
3285 		if (r)
3286 			goto out;
3287 		r = -EFAULT;
3288 		if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
3289 			goto out;
3290 		r = 0;
3291 		break;
3292 	}
3293 	case KVM_SET_LAPIC: {
3294 		r = -EINVAL;
3295 		if (!vcpu->arch.apic)
3296 			goto out;
3297 		u.lapic = memdup_user(argp, sizeof(*u.lapic));
3298 		if (IS_ERR(u.lapic))
3299 			return PTR_ERR(u.lapic);
3300 
3301 		r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
3302 		break;
3303 	}
3304 	case KVM_INTERRUPT: {
3305 		struct kvm_interrupt irq;
3306 
3307 		r = -EFAULT;
3308 		if (copy_from_user(&irq, argp, sizeof irq))
3309 			goto out;
3310 		r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
3311 		break;
3312 	}
3313 	case KVM_NMI: {
3314 		r = kvm_vcpu_ioctl_nmi(vcpu);
3315 		break;
3316 	}
3317 	case KVM_SMI: {
3318 		r = kvm_vcpu_ioctl_smi(vcpu);
3319 		break;
3320 	}
3321 	case KVM_SET_CPUID: {
3322 		struct kvm_cpuid __user *cpuid_arg = argp;
3323 		struct kvm_cpuid cpuid;
3324 
3325 		r = -EFAULT;
3326 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3327 			goto out;
3328 		r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3329 		break;
3330 	}
3331 	case KVM_SET_CPUID2: {
3332 		struct kvm_cpuid2 __user *cpuid_arg = argp;
3333 		struct kvm_cpuid2 cpuid;
3334 
3335 		r = -EFAULT;
3336 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3337 			goto out;
3338 		r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
3339 					      cpuid_arg->entries);
3340 		break;
3341 	}
3342 	case KVM_GET_CPUID2: {
3343 		struct kvm_cpuid2 __user *cpuid_arg = argp;
3344 		struct kvm_cpuid2 cpuid;
3345 
3346 		r = -EFAULT;
3347 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3348 			goto out;
3349 		r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
3350 					      cpuid_arg->entries);
3351 		if (r)
3352 			goto out;
3353 		r = -EFAULT;
3354 		if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3355 			goto out;
3356 		r = 0;
3357 		break;
3358 	}
3359 	case KVM_GET_MSRS:
3360 		r = msr_io(vcpu, argp, do_get_msr, 1);
3361 		break;
3362 	case KVM_SET_MSRS:
3363 		r = msr_io(vcpu, argp, do_set_msr, 0);
3364 		break;
3365 	case KVM_TPR_ACCESS_REPORTING: {
3366 		struct kvm_tpr_access_ctl tac;
3367 
3368 		r = -EFAULT;
3369 		if (copy_from_user(&tac, argp, sizeof tac))
3370 			goto out;
3371 		r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3372 		if (r)
3373 			goto out;
3374 		r = -EFAULT;
3375 		if (copy_to_user(argp, &tac, sizeof tac))
3376 			goto out;
3377 		r = 0;
3378 		break;
3379 	};
3380 	case KVM_SET_VAPIC_ADDR: {
3381 		struct kvm_vapic_addr va;
3382 
3383 		r = -EINVAL;
3384 		if (!irqchip_in_kernel(vcpu->kvm))
3385 			goto out;
3386 		r = -EFAULT;
3387 		if (copy_from_user(&va, argp, sizeof va))
3388 			goto out;
3389 		r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
3390 		break;
3391 	}
3392 	case KVM_X86_SETUP_MCE: {
3393 		u64 mcg_cap;
3394 
3395 		r = -EFAULT;
3396 		if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
3397 			goto out;
3398 		r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3399 		break;
3400 	}
3401 	case KVM_X86_SET_MCE: {
3402 		struct kvm_x86_mce mce;
3403 
3404 		r = -EFAULT;
3405 		if (copy_from_user(&mce, argp, sizeof mce))
3406 			goto out;
3407 		r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3408 		break;
3409 	}
3410 	case KVM_GET_VCPU_EVENTS: {
3411 		struct kvm_vcpu_events events;
3412 
3413 		kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3414 
3415 		r = -EFAULT;
3416 		if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3417 			break;
3418 		r = 0;
3419 		break;
3420 	}
3421 	case KVM_SET_VCPU_EVENTS: {
3422 		struct kvm_vcpu_events events;
3423 
3424 		r = -EFAULT;
3425 		if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3426 			break;
3427 
3428 		r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3429 		break;
3430 	}
3431 	case KVM_GET_DEBUGREGS: {
3432 		struct kvm_debugregs dbgregs;
3433 
3434 		kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3435 
3436 		r = -EFAULT;
3437 		if (copy_to_user(argp, &dbgregs,
3438 				 sizeof(struct kvm_debugregs)))
3439 			break;
3440 		r = 0;
3441 		break;
3442 	}
3443 	case KVM_SET_DEBUGREGS: {
3444 		struct kvm_debugregs dbgregs;
3445 
3446 		r = -EFAULT;
3447 		if (copy_from_user(&dbgregs, argp,
3448 				   sizeof(struct kvm_debugregs)))
3449 			break;
3450 
3451 		r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3452 		break;
3453 	}
3454 	case KVM_GET_XSAVE: {
3455 		u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3456 		r = -ENOMEM;
3457 		if (!u.xsave)
3458 			break;
3459 
3460 		kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
3461 
3462 		r = -EFAULT;
3463 		if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
3464 			break;
3465 		r = 0;
3466 		break;
3467 	}
3468 	case KVM_SET_XSAVE: {
3469 		u.xsave = memdup_user(argp, sizeof(*u.xsave));
3470 		if (IS_ERR(u.xsave))
3471 			return PTR_ERR(u.xsave);
3472 
3473 		r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
3474 		break;
3475 	}
3476 	case KVM_GET_XCRS: {
3477 		u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3478 		r = -ENOMEM;
3479 		if (!u.xcrs)
3480 			break;
3481 
3482 		kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
3483 
3484 		r = -EFAULT;
3485 		if (copy_to_user(argp, u.xcrs,
3486 				 sizeof(struct kvm_xcrs)))
3487 			break;
3488 		r = 0;
3489 		break;
3490 	}
3491 	case KVM_SET_XCRS: {
3492 		u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
3493 		if (IS_ERR(u.xcrs))
3494 			return PTR_ERR(u.xcrs);
3495 
3496 		r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3497 		break;
3498 	}
3499 	case KVM_SET_TSC_KHZ: {
3500 		u32 user_tsc_khz;
3501 
3502 		r = -EINVAL;
3503 		user_tsc_khz = (u32)arg;
3504 
3505 		if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3506 			goto out;
3507 
3508 		if (user_tsc_khz == 0)
3509 			user_tsc_khz = tsc_khz;
3510 
3511 		kvm_set_tsc_khz(vcpu, user_tsc_khz);
3512 
3513 		r = 0;
3514 		goto out;
3515 	}
3516 	case KVM_GET_TSC_KHZ: {
3517 		r = vcpu->arch.virtual_tsc_khz;
3518 		goto out;
3519 	}
3520 	case KVM_KVMCLOCK_CTRL: {
3521 		r = kvm_set_guest_paused(vcpu);
3522 		goto out;
3523 	}
3524 	default:
3525 		r = -EINVAL;
3526 	}
3527 out:
3528 	kfree(u.buffer);
3529 	return r;
3530 }
3531 
3532 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
3533 {
3534 	return VM_FAULT_SIGBUS;
3535 }
3536 
3537 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3538 {
3539 	int ret;
3540 
3541 	if (addr > (unsigned int)(-3 * PAGE_SIZE))
3542 		return -EINVAL;
3543 	ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3544 	return ret;
3545 }
3546 
3547 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3548 					      u64 ident_addr)
3549 {
3550 	kvm->arch.ept_identity_map_addr = ident_addr;
3551 	return 0;
3552 }
3553 
3554 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3555 					  u32 kvm_nr_mmu_pages)
3556 {
3557 	if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3558 		return -EINVAL;
3559 
3560 	mutex_lock(&kvm->slots_lock);
3561 
3562 	kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3563 	kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3564 
3565 	mutex_unlock(&kvm->slots_lock);
3566 	return 0;
3567 }
3568 
3569 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3570 {
3571 	return kvm->arch.n_max_mmu_pages;
3572 }
3573 
3574 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3575 {
3576 	int r;
3577 
3578 	r = 0;
3579 	switch (chip->chip_id) {
3580 	case KVM_IRQCHIP_PIC_MASTER:
3581 		memcpy(&chip->chip.pic,
3582 			&pic_irqchip(kvm)->pics[0],
3583 			sizeof(struct kvm_pic_state));
3584 		break;
3585 	case KVM_IRQCHIP_PIC_SLAVE:
3586 		memcpy(&chip->chip.pic,
3587 			&pic_irqchip(kvm)->pics[1],
3588 			sizeof(struct kvm_pic_state));
3589 		break;
3590 	case KVM_IRQCHIP_IOAPIC:
3591 		r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
3592 		break;
3593 	default:
3594 		r = -EINVAL;
3595 		break;
3596 	}
3597 	return r;
3598 }
3599 
3600 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3601 {
3602 	int r;
3603 
3604 	r = 0;
3605 	switch (chip->chip_id) {
3606 	case KVM_IRQCHIP_PIC_MASTER:
3607 		spin_lock(&pic_irqchip(kvm)->lock);
3608 		memcpy(&pic_irqchip(kvm)->pics[0],
3609 			&chip->chip.pic,
3610 			sizeof(struct kvm_pic_state));
3611 		spin_unlock(&pic_irqchip(kvm)->lock);
3612 		break;
3613 	case KVM_IRQCHIP_PIC_SLAVE:
3614 		spin_lock(&pic_irqchip(kvm)->lock);
3615 		memcpy(&pic_irqchip(kvm)->pics[1],
3616 			&chip->chip.pic,
3617 			sizeof(struct kvm_pic_state));
3618 		spin_unlock(&pic_irqchip(kvm)->lock);
3619 		break;
3620 	case KVM_IRQCHIP_IOAPIC:
3621 		r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
3622 		break;
3623 	default:
3624 		r = -EINVAL;
3625 		break;
3626 	}
3627 	kvm_pic_update_irq(pic_irqchip(kvm));
3628 	return r;
3629 }
3630 
3631 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3632 {
3633 	int r = 0;
3634 
3635 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
3636 	memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
3637 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3638 	return r;
3639 }
3640 
3641 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3642 {
3643 	int r = 0;
3644 
3645 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
3646 	memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
3647 	kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
3648 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3649 	return r;
3650 }
3651 
3652 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3653 {
3654 	int r = 0;
3655 
3656 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
3657 	memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3658 		sizeof(ps->channels));
3659 	ps->flags = kvm->arch.vpit->pit_state.flags;
3660 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3661 	memset(&ps->reserved, 0, sizeof(ps->reserved));
3662 	return r;
3663 }
3664 
3665 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3666 {
3667 	int r = 0, start = 0;
3668 	u32 prev_legacy, cur_legacy;
3669 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
3670 	prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3671 	cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3672 	if (!prev_legacy && cur_legacy)
3673 		start = 1;
3674 	memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
3675 	       sizeof(kvm->arch.vpit->pit_state.channels));
3676 	kvm->arch.vpit->pit_state.flags = ps->flags;
3677 	kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
3678 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3679 	return r;
3680 }
3681 
3682 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3683 				 struct kvm_reinject_control *control)
3684 {
3685 	if (!kvm->arch.vpit)
3686 		return -ENXIO;
3687 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
3688 	kvm->arch.vpit->pit_state.reinject = control->pit_reinject;
3689 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3690 	return 0;
3691 }
3692 
3693 /**
3694  * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
3695  * @kvm: kvm instance
3696  * @log: slot id and address to which we copy the log
3697  *
3698  * Steps 1-4 below provide general overview of dirty page logging. See
3699  * kvm_get_dirty_log_protect() function description for additional details.
3700  *
3701  * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
3702  * always flush the TLB (step 4) even if previous step failed  and the dirty
3703  * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
3704  * does not preclude user space subsequent dirty log read. Flushing TLB ensures
3705  * writes will be marked dirty for next log read.
3706  *
3707  *   1. Take a snapshot of the bit and clear it if needed.
3708  *   2. Write protect the corresponding page.
3709  *   3. Copy the snapshot to the userspace.
3710  *   4. Flush TLB's if needed.
3711  */
3712 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
3713 {
3714 	bool is_dirty = false;
3715 	int r;
3716 
3717 	mutex_lock(&kvm->slots_lock);
3718 
3719 	/*
3720 	 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
3721 	 */
3722 	if (kvm_x86_ops->flush_log_dirty)
3723 		kvm_x86_ops->flush_log_dirty(kvm);
3724 
3725 	r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
3726 
3727 	/*
3728 	 * All the TLBs can be flushed out of mmu lock, see the comments in
3729 	 * kvm_mmu_slot_remove_write_access().
3730 	 */
3731 	lockdep_assert_held(&kvm->slots_lock);
3732 	if (is_dirty)
3733 		kvm_flush_remote_tlbs(kvm);
3734 
3735 	mutex_unlock(&kvm->slots_lock);
3736 	return r;
3737 }
3738 
3739 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
3740 			bool line_status)
3741 {
3742 	if (!irqchip_in_kernel(kvm))
3743 		return -ENXIO;
3744 
3745 	irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3746 					irq_event->irq, irq_event->level,
3747 					line_status);
3748 	return 0;
3749 }
3750 
3751 static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
3752 				   struct kvm_enable_cap *cap)
3753 {
3754 	int r;
3755 
3756 	if (cap->flags)
3757 		return -EINVAL;
3758 
3759 	switch (cap->cap) {
3760 	case KVM_CAP_DISABLE_QUIRKS:
3761 		kvm->arch.disabled_quirks = cap->args[0];
3762 		r = 0;
3763 		break;
3764 	default:
3765 		r = -EINVAL;
3766 		break;
3767 	}
3768 	return r;
3769 }
3770 
3771 long kvm_arch_vm_ioctl(struct file *filp,
3772 		       unsigned int ioctl, unsigned long arg)
3773 {
3774 	struct kvm *kvm = filp->private_data;
3775 	void __user *argp = (void __user *)arg;
3776 	int r = -ENOTTY;
3777 	/*
3778 	 * This union makes it completely explicit to gcc-3.x
3779 	 * that these two variables' stack usage should be
3780 	 * combined, not added together.
3781 	 */
3782 	union {
3783 		struct kvm_pit_state ps;
3784 		struct kvm_pit_state2 ps2;
3785 		struct kvm_pit_config pit_config;
3786 	} u;
3787 
3788 	switch (ioctl) {
3789 	case KVM_SET_TSS_ADDR:
3790 		r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
3791 		break;
3792 	case KVM_SET_IDENTITY_MAP_ADDR: {
3793 		u64 ident_addr;
3794 
3795 		r = -EFAULT;
3796 		if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3797 			goto out;
3798 		r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
3799 		break;
3800 	}
3801 	case KVM_SET_NR_MMU_PAGES:
3802 		r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
3803 		break;
3804 	case KVM_GET_NR_MMU_PAGES:
3805 		r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3806 		break;
3807 	case KVM_CREATE_IRQCHIP: {
3808 		struct kvm_pic *vpic;
3809 
3810 		mutex_lock(&kvm->lock);
3811 		r = -EEXIST;
3812 		if (kvm->arch.vpic)
3813 			goto create_irqchip_unlock;
3814 		r = -EINVAL;
3815 		if (atomic_read(&kvm->online_vcpus))
3816 			goto create_irqchip_unlock;
3817 		r = -ENOMEM;
3818 		vpic = kvm_create_pic(kvm);
3819 		if (vpic) {
3820 			r = kvm_ioapic_init(kvm);
3821 			if (r) {
3822 				mutex_lock(&kvm->slots_lock);
3823 				kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3824 							  &vpic->dev_master);
3825 				kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3826 							  &vpic->dev_slave);
3827 				kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3828 							  &vpic->dev_eclr);
3829 				mutex_unlock(&kvm->slots_lock);
3830 				kfree(vpic);
3831 				goto create_irqchip_unlock;
3832 			}
3833 		} else
3834 			goto create_irqchip_unlock;
3835 		smp_wmb();
3836 		kvm->arch.vpic = vpic;
3837 		smp_wmb();
3838 		r = kvm_setup_default_irq_routing(kvm);
3839 		if (r) {
3840 			mutex_lock(&kvm->slots_lock);
3841 			mutex_lock(&kvm->irq_lock);
3842 			kvm_ioapic_destroy(kvm);
3843 			kvm_destroy_pic(kvm);
3844 			mutex_unlock(&kvm->irq_lock);
3845 			mutex_unlock(&kvm->slots_lock);
3846 		}
3847 	create_irqchip_unlock:
3848 		mutex_unlock(&kvm->lock);
3849 		break;
3850 	}
3851 	case KVM_CREATE_PIT:
3852 		u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
3853 		goto create_pit;
3854 	case KVM_CREATE_PIT2:
3855 		r = -EFAULT;
3856 		if (copy_from_user(&u.pit_config, argp,
3857 				   sizeof(struct kvm_pit_config)))
3858 			goto out;
3859 	create_pit:
3860 		mutex_lock(&kvm->slots_lock);
3861 		r = -EEXIST;
3862 		if (kvm->arch.vpit)
3863 			goto create_pit_unlock;
3864 		r = -ENOMEM;
3865 		kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
3866 		if (kvm->arch.vpit)
3867 			r = 0;
3868 	create_pit_unlock:
3869 		mutex_unlock(&kvm->slots_lock);
3870 		break;
3871 	case KVM_GET_IRQCHIP: {
3872 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3873 		struct kvm_irqchip *chip;
3874 
3875 		chip = memdup_user(argp, sizeof(*chip));
3876 		if (IS_ERR(chip)) {
3877 			r = PTR_ERR(chip);
3878 			goto out;
3879 		}
3880 
3881 		r = -ENXIO;
3882 		if (!irqchip_in_kernel(kvm))
3883 			goto get_irqchip_out;
3884 		r = kvm_vm_ioctl_get_irqchip(kvm, chip);
3885 		if (r)
3886 			goto get_irqchip_out;
3887 		r = -EFAULT;
3888 		if (copy_to_user(argp, chip, sizeof *chip))
3889 			goto get_irqchip_out;
3890 		r = 0;
3891 	get_irqchip_out:
3892 		kfree(chip);
3893 		break;
3894 	}
3895 	case KVM_SET_IRQCHIP: {
3896 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3897 		struct kvm_irqchip *chip;
3898 
3899 		chip = memdup_user(argp, sizeof(*chip));
3900 		if (IS_ERR(chip)) {
3901 			r = PTR_ERR(chip);
3902 			goto out;
3903 		}
3904 
3905 		r = -ENXIO;
3906 		if (!irqchip_in_kernel(kvm))
3907 			goto set_irqchip_out;
3908 		r = kvm_vm_ioctl_set_irqchip(kvm, chip);
3909 		if (r)
3910 			goto set_irqchip_out;
3911 		r = 0;
3912 	set_irqchip_out:
3913 		kfree(chip);
3914 		break;
3915 	}
3916 	case KVM_GET_PIT: {
3917 		r = -EFAULT;
3918 		if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
3919 			goto out;
3920 		r = -ENXIO;
3921 		if (!kvm->arch.vpit)
3922 			goto out;
3923 		r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
3924 		if (r)
3925 			goto out;
3926 		r = -EFAULT;
3927 		if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
3928 			goto out;
3929 		r = 0;
3930 		break;
3931 	}
3932 	case KVM_SET_PIT: {
3933 		r = -EFAULT;
3934 		if (copy_from_user(&u.ps, argp, sizeof u.ps))
3935 			goto out;
3936 		r = -ENXIO;
3937 		if (!kvm->arch.vpit)
3938 			goto out;
3939 		r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
3940 		break;
3941 	}
3942 	case KVM_GET_PIT2: {
3943 		r = -ENXIO;
3944 		if (!kvm->arch.vpit)
3945 			goto out;
3946 		r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
3947 		if (r)
3948 			goto out;
3949 		r = -EFAULT;
3950 		if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
3951 			goto out;
3952 		r = 0;
3953 		break;
3954 	}
3955 	case KVM_SET_PIT2: {
3956 		r = -EFAULT;
3957 		if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
3958 			goto out;
3959 		r = -ENXIO;
3960 		if (!kvm->arch.vpit)
3961 			goto out;
3962 		r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
3963 		break;
3964 	}
3965 	case KVM_REINJECT_CONTROL: {
3966 		struct kvm_reinject_control control;
3967 		r =  -EFAULT;
3968 		if (copy_from_user(&control, argp, sizeof(control)))
3969 			goto out;
3970 		r = kvm_vm_ioctl_reinject(kvm, &control);
3971 		break;
3972 	}
3973 	case KVM_XEN_HVM_CONFIG: {
3974 		r = -EFAULT;
3975 		if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
3976 				   sizeof(struct kvm_xen_hvm_config)))
3977 			goto out;
3978 		r = -EINVAL;
3979 		if (kvm->arch.xen_hvm_config.flags)
3980 			goto out;
3981 		r = 0;
3982 		break;
3983 	}
3984 	case KVM_SET_CLOCK: {
3985 		struct kvm_clock_data user_ns;
3986 		u64 now_ns;
3987 		s64 delta;
3988 
3989 		r = -EFAULT;
3990 		if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
3991 			goto out;
3992 
3993 		r = -EINVAL;
3994 		if (user_ns.flags)
3995 			goto out;
3996 
3997 		r = 0;
3998 		local_irq_disable();
3999 		now_ns = get_kernel_ns();
4000 		delta = user_ns.clock - now_ns;
4001 		local_irq_enable();
4002 		kvm->arch.kvmclock_offset = delta;
4003 		kvm_gen_update_masterclock(kvm);
4004 		break;
4005 	}
4006 	case KVM_GET_CLOCK: {
4007 		struct kvm_clock_data user_ns;
4008 		u64 now_ns;
4009 
4010 		local_irq_disable();
4011 		now_ns = get_kernel_ns();
4012 		user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
4013 		local_irq_enable();
4014 		user_ns.flags = 0;
4015 		memset(&user_ns.pad, 0, sizeof(user_ns.pad));
4016 
4017 		r = -EFAULT;
4018 		if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
4019 			goto out;
4020 		r = 0;
4021 		break;
4022 	}
4023 	case KVM_ENABLE_CAP: {
4024 		struct kvm_enable_cap cap;
4025 
4026 		r = -EFAULT;
4027 		if (copy_from_user(&cap, argp, sizeof(cap)))
4028 			goto out;
4029 		r = kvm_vm_ioctl_enable_cap(kvm, &cap);
4030 		break;
4031 	}
4032 	default:
4033 		r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
4034 	}
4035 out:
4036 	return r;
4037 }
4038 
4039 static void kvm_init_msr_list(void)
4040 {
4041 	u32 dummy[2];
4042 	unsigned i, j;
4043 
4044 	for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
4045 		if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
4046 			continue;
4047 
4048 		/*
4049 		 * Even MSRs that are valid in the host may not be exposed
4050 		 * to the guests in some cases.  We could work around this
4051 		 * in VMX with the generic MSR save/load machinery, but it
4052 		 * is not really worthwhile since it will really only
4053 		 * happen with nested virtualization.
4054 		 */
4055 		switch (msrs_to_save[i]) {
4056 		case MSR_IA32_BNDCFGS:
4057 			if (!kvm_x86_ops->mpx_supported())
4058 				continue;
4059 			break;
4060 		default:
4061 			break;
4062 		}
4063 
4064 		if (j < i)
4065 			msrs_to_save[j] = msrs_to_save[i];
4066 		j++;
4067 	}
4068 	num_msrs_to_save = j;
4069 
4070 	for (i = j = 0; i < ARRAY_SIZE(emulated_msrs); i++) {
4071 		switch (emulated_msrs[i]) {
4072 		case MSR_IA32_SMBASE:
4073 			if (!kvm_x86_ops->cpu_has_high_real_mode_segbase())
4074 				continue;
4075 			break;
4076 		default:
4077 			break;
4078 		}
4079 
4080 		if (j < i)
4081 			emulated_msrs[j] = emulated_msrs[i];
4082 		j++;
4083 	}
4084 	num_emulated_msrs = j;
4085 }
4086 
4087 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
4088 			   const void *v)
4089 {
4090 	int handled = 0;
4091 	int n;
4092 
4093 	do {
4094 		n = min(len, 8);
4095 		if (!(vcpu->arch.apic &&
4096 		      !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
4097 		    && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
4098 			break;
4099 		handled += n;
4100 		addr += n;
4101 		len -= n;
4102 		v += n;
4103 	} while (len);
4104 
4105 	return handled;
4106 }
4107 
4108 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
4109 {
4110 	int handled = 0;
4111 	int n;
4112 
4113 	do {
4114 		n = min(len, 8);
4115 		if (!(vcpu->arch.apic &&
4116 		      !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
4117 					 addr, n, v))
4118 		    && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
4119 			break;
4120 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
4121 		handled += n;
4122 		addr += n;
4123 		len -= n;
4124 		v += n;
4125 	} while (len);
4126 
4127 	return handled;
4128 }
4129 
4130 static void kvm_set_segment(struct kvm_vcpu *vcpu,
4131 			struct kvm_segment *var, int seg)
4132 {
4133 	kvm_x86_ops->set_segment(vcpu, var, seg);
4134 }
4135 
4136 void kvm_get_segment(struct kvm_vcpu *vcpu,
4137 		     struct kvm_segment *var, int seg)
4138 {
4139 	kvm_x86_ops->get_segment(vcpu, var, seg);
4140 }
4141 
4142 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
4143 			   struct x86_exception *exception)
4144 {
4145 	gpa_t t_gpa;
4146 
4147 	BUG_ON(!mmu_is_nested(vcpu));
4148 
4149 	/* NPT walks are always user-walks */
4150 	access |= PFERR_USER_MASK;
4151 	t_gpa  = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, exception);
4152 
4153 	return t_gpa;
4154 }
4155 
4156 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
4157 			      struct x86_exception *exception)
4158 {
4159 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4160 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4161 }
4162 
4163  gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
4164 				struct x86_exception *exception)
4165 {
4166 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4167 	access |= PFERR_FETCH_MASK;
4168 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4169 }
4170 
4171 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
4172 			       struct x86_exception *exception)
4173 {
4174 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4175 	access |= PFERR_WRITE_MASK;
4176 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4177 }
4178 
4179 /* uses this to access any guest's mapped memory without checking CPL */
4180 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
4181 				struct x86_exception *exception)
4182 {
4183 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
4184 }
4185 
4186 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
4187 				      struct kvm_vcpu *vcpu, u32 access,
4188 				      struct x86_exception *exception)
4189 {
4190 	void *data = val;
4191 	int r = X86EMUL_CONTINUE;
4192 
4193 	while (bytes) {
4194 		gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
4195 							    exception);
4196 		unsigned offset = addr & (PAGE_SIZE-1);
4197 		unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
4198 		int ret;
4199 
4200 		if (gpa == UNMAPPED_GVA)
4201 			return X86EMUL_PROPAGATE_FAULT;
4202 		ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
4203 					       offset, toread);
4204 		if (ret < 0) {
4205 			r = X86EMUL_IO_NEEDED;
4206 			goto out;
4207 		}
4208 
4209 		bytes -= toread;
4210 		data += toread;
4211 		addr += toread;
4212 	}
4213 out:
4214 	return r;
4215 }
4216 
4217 /* used for instruction fetching */
4218 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
4219 				gva_t addr, void *val, unsigned int bytes,
4220 				struct x86_exception *exception)
4221 {
4222 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4223 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4224 	unsigned offset;
4225 	int ret;
4226 
4227 	/* Inline kvm_read_guest_virt_helper for speed.  */
4228 	gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
4229 						    exception);
4230 	if (unlikely(gpa == UNMAPPED_GVA))
4231 		return X86EMUL_PROPAGATE_FAULT;
4232 
4233 	offset = addr & (PAGE_SIZE-1);
4234 	if (WARN_ON(offset + bytes > PAGE_SIZE))
4235 		bytes = (unsigned)PAGE_SIZE - offset;
4236 	ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
4237 				       offset, bytes);
4238 	if (unlikely(ret < 0))
4239 		return X86EMUL_IO_NEEDED;
4240 
4241 	return X86EMUL_CONTINUE;
4242 }
4243 
4244 int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
4245 			       gva_t addr, void *val, unsigned int bytes,
4246 			       struct x86_exception *exception)
4247 {
4248 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4249 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4250 
4251 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
4252 					  exception);
4253 }
4254 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
4255 
4256 static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4257 				      gva_t addr, void *val, unsigned int bytes,
4258 				      struct x86_exception *exception)
4259 {
4260 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4261 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
4262 }
4263 
4264 int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4265 				       gva_t addr, void *val,
4266 				       unsigned int bytes,
4267 				       struct x86_exception *exception)
4268 {
4269 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4270 	void *data = val;
4271 	int r = X86EMUL_CONTINUE;
4272 
4273 	while (bytes) {
4274 		gpa_t gpa =  vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
4275 							     PFERR_WRITE_MASK,
4276 							     exception);
4277 		unsigned offset = addr & (PAGE_SIZE-1);
4278 		unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
4279 		int ret;
4280 
4281 		if (gpa == UNMAPPED_GVA)
4282 			return X86EMUL_PROPAGATE_FAULT;
4283 		ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
4284 		if (ret < 0) {
4285 			r = X86EMUL_IO_NEEDED;
4286 			goto out;
4287 		}
4288 
4289 		bytes -= towrite;
4290 		data += towrite;
4291 		addr += towrite;
4292 	}
4293 out:
4294 	return r;
4295 }
4296 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
4297 
4298 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4299 				gpa_t *gpa, struct x86_exception *exception,
4300 				bool write)
4301 {
4302 	u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
4303 		| (write ? PFERR_WRITE_MASK : 0);
4304 
4305 	if (vcpu_match_mmio_gva(vcpu, gva)
4306 	    && !permission_fault(vcpu, vcpu->arch.walk_mmu,
4307 				 vcpu->arch.access, access)) {
4308 		*gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
4309 					(gva & (PAGE_SIZE - 1));
4310 		trace_vcpu_match_mmio(gva, *gpa, write, false);
4311 		return 1;
4312 	}
4313 
4314 	*gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4315 
4316 	if (*gpa == UNMAPPED_GVA)
4317 		return -1;
4318 
4319 	/* For APIC access vmexit */
4320 	if ((*gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4321 		return 1;
4322 
4323 	if (vcpu_match_mmio_gpa(vcpu, *gpa)) {
4324 		trace_vcpu_match_mmio(gva, *gpa, write, true);
4325 		return 1;
4326 	}
4327 
4328 	return 0;
4329 }
4330 
4331 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
4332 			const void *val, int bytes)
4333 {
4334 	int ret;
4335 
4336 	ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
4337 	if (ret < 0)
4338 		return 0;
4339 	kvm_mmu_pte_write(vcpu, gpa, val, bytes);
4340 	return 1;
4341 }
4342 
4343 struct read_write_emulator_ops {
4344 	int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
4345 				  int bytes);
4346 	int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
4347 				  void *val, int bytes);
4348 	int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4349 			       int bytes, void *val);
4350 	int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4351 				    void *val, int bytes);
4352 	bool write;
4353 };
4354 
4355 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
4356 {
4357 	if (vcpu->mmio_read_completed) {
4358 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
4359 			       vcpu->mmio_fragments[0].gpa, *(u64 *)val);
4360 		vcpu->mmio_read_completed = 0;
4361 		return 1;
4362 	}
4363 
4364 	return 0;
4365 }
4366 
4367 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4368 			void *val, int bytes)
4369 {
4370 	return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
4371 }
4372 
4373 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4374 			 void *val, int bytes)
4375 {
4376 	return emulator_write_phys(vcpu, gpa, val, bytes);
4377 }
4378 
4379 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
4380 {
4381 	trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
4382 	return vcpu_mmio_write(vcpu, gpa, bytes, val);
4383 }
4384 
4385 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4386 			  void *val, int bytes)
4387 {
4388 	trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
4389 	return X86EMUL_IO_NEEDED;
4390 }
4391 
4392 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4393 			   void *val, int bytes)
4394 {
4395 	struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
4396 
4397 	memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
4398 	return X86EMUL_CONTINUE;
4399 }
4400 
4401 static const struct read_write_emulator_ops read_emultor = {
4402 	.read_write_prepare = read_prepare,
4403 	.read_write_emulate = read_emulate,
4404 	.read_write_mmio = vcpu_mmio_read,
4405 	.read_write_exit_mmio = read_exit_mmio,
4406 };
4407 
4408 static const struct read_write_emulator_ops write_emultor = {
4409 	.read_write_emulate = write_emulate,
4410 	.read_write_mmio = write_mmio,
4411 	.read_write_exit_mmio = write_exit_mmio,
4412 	.write = true,
4413 };
4414 
4415 static int emulator_read_write_onepage(unsigned long addr, void *val,
4416 				       unsigned int bytes,
4417 				       struct x86_exception *exception,
4418 				       struct kvm_vcpu *vcpu,
4419 				       const struct read_write_emulator_ops *ops)
4420 {
4421 	gpa_t gpa;
4422 	int handled, ret;
4423 	bool write = ops->write;
4424 	struct kvm_mmio_fragment *frag;
4425 
4426 	ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
4427 
4428 	if (ret < 0)
4429 		return X86EMUL_PROPAGATE_FAULT;
4430 
4431 	/* For APIC access vmexit */
4432 	if (ret)
4433 		goto mmio;
4434 
4435 	if (ops->read_write_emulate(vcpu, gpa, val, bytes))
4436 		return X86EMUL_CONTINUE;
4437 
4438 mmio:
4439 	/*
4440 	 * Is this MMIO handled locally?
4441 	 */
4442 	handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
4443 	if (handled == bytes)
4444 		return X86EMUL_CONTINUE;
4445 
4446 	gpa += handled;
4447 	bytes -= handled;
4448 	val += handled;
4449 
4450 	WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
4451 	frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
4452 	frag->gpa = gpa;
4453 	frag->data = val;
4454 	frag->len = bytes;
4455 	return X86EMUL_CONTINUE;
4456 }
4457 
4458 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
4459 			unsigned long addr,
4460 			void *val, unsigned int bytes,
4461 			struct x86_exception *exception,
4462 			const struct read_write_emulator_ops *ops)
4463 {
4464 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4465 	gpa_t gpa;
4466 	int rc;
4467 
4468 	if (ops->read_write_prepare &&
4469 		  ops->read_write_prepare(vcpu, val, bytes))
4470 		return X86EMUL_CONTINUE;
4471 
4472 	vcpu->mmio_nr_fragments = 0;
4473 
4474 	/* Crossing a page boundary? */
4475 	if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
4476 		int now;
4477 
4478 		now = -addr & ~PAGE_MASK;
4479 		rc = emulator_read_write_onepage(addr, val, now, exception,
4480 						 vcpu, ops);
4481 
4482 		if (rc != X86EMUL_CONTINUE)
4483 			return rc;
4484 		addr += now;
4485 		if (ctxt->mode != X86EMUL_MODE_PROT64)
4486 			addr = (u32)addr;
4487 		val += now;
4488 		bytes -= now;
4489 	}
4490 
4491 	rc = emulator_read_write_onepage(addr, val, bytes, exception,
4492 					 vcpu, ops);
4493 	if (rc != X86EMUL_CONTINUE)
4494 		return rc;
4495 
4496 	if (!vcpu->mmio_nr_fragments)
4497 		return rc;
4498 
4499 	gpa = vcpu->mmio_fragments[0].gpa;
4500 
4501 	vcpu->mmio_needed = 1;
4502 	vcpu->mmio_cur_fragment = 0;
4503 
4504 	vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
4505 	vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
4506 	vcpu->run->exit_reason = KVM_EXIT_MMIO;
4507 	vcpu->run->mmio.phys_addr = gpa;
4508 
4509 	return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
4510 }
4511 
4512 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
4513 				  unsigned long addr,
4514 				  void *val,
4515 				  unsigned int bytes,
4516 				  struct x86_exception *exception)
4517 {
4518 	return emulator_read_write(ctxt, addr, val, bytes,
4519 				   exception, &read_emultor);
4520 }
4521 
4522 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
4523 			    unsigned long addr,
4524 			    const void *val,
4525 			    unsigned int bytes,
4526 			    struct x86_exception *exception)
4527 {
4528 	return emulator_read_write(ctxt, addr, (void *)val, bytes,
4529 				   exception, &write_emultor);
4530 }
4531 
4532 #define CMPXCHG_TYPE(t, ptr, old, new) \
4533 	(cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4534 
4535 #ifdef CONFIG_X86_64
4536 #  define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4537 #else
4538 #  define CMPXCHG64(ptr, old, new) \
4539 	(cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
4540 #endif
4541 
4542 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
4543 				     unsigned long addr,
4544 				     const void *old,
4545 				     const void *new,
4546 				     unsigned int bytes,
4547 				     struct x86_exception *exception)
4548 {
4549 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4550 	gpa_t gpa;
4551 	struct page *page;
4552 	char *kaddr;
4553 	bool exchanged;
4554 
4555 	/* guests cmpxchg8b have to be emulated atomically */
4556 	if (bytes > 8 || (bytes & (bytes - 1)))
4557 		goto emul_write;
4558 
4559 	gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
4560 
4561 	if (gpa == UNMAPPED_GVA ||
4562 	    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4563 		goto emul_write;
4564 
4565 	if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
4566 		goto emul_write;
4567 
4568 	page = kvm_vcpu_gfn_to_page(vcpu, gpa >> PAGE_SHIFT);
4569 	if (is_error_page(page))
4570 		goto emul_write;
4571 
4572 	kaddr = kmap_atomic(page);
4573 	kaddr += offset_in_page(gpa);
4574 	switch (bytes) {
4575 	case 1:
4576 		exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
4577 		break;
4578 	case 2:
4579 		exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
4580 		break;
4581 	case 4:
4582 		exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
4583 		break;
4584 	case 8:
4585 		exchanged = CMPXCHG64(kaddr, old, new);
4586 		break;
4587 	default:
4588 		BUG();
4589 	}
4590 	kunmap_atomic(kaddr);
4591 	kvm_release_page_dirty(page);
4592 
4593 	if (!exchanged)
4594 		return X86EMUL_CMPXCHG_FAILED;
4595 
4596 	kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
4597 	kvm_mmu_pte_write(vcpu, gpa, new, bytes);
4598 
4599 	return X86EMUL_CONTINUE;
4600 
4601 emul_write:
4602 	printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
4603 
4604 	return emulator_write_emulated(ctxt, addr, new, bytes, exception);
4605 }
4606 
4607 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4608 {
4609 	/* TODO: String I/O for in kernel device */
4610 	int r;
4611 
4612 	if (vcpu->arch.pio.in)
4613 		r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
4614 				    vcpu->arch.pio.size, pd);
4615 	else
4616 		r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
4617 				     vcpu->arch.pio.port, vcpu->arch.pio.size,
4618 				     pd);
4619 	return r;
4620 }
4621 
4622 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
4623 			       unsigned short port, void *val,
4624 			       unsigned int count, bool in)
4625 {
4626 	vcpu->arch.pio.port = port;
4627 	vcpu->arch.pio.in = in;
4628 	vcpu->arch.pio.count  = count;
4629 	vcpu->arch.pio.size = size;
4630 
4631 	if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4632 		vcpu->arch.pio.count = 0;
4633 		return 1;
4634 	}
4635 
4636 	vcpu->run->exit_reason = KVM_EXIT_IO;
4637 	vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
4638 	vcpu->run->io.size = size;
4639 	vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4640 	vcpu->run->io.count = count;
4641 	vcpu->run->io.port = port;
4642 
4643 	return 0;
4644 }
4645 
4646 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
4647 				    int size, unsigned short port, void *val,
4648 				    unsigned int count)
4649 {
4650 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4651 	int ret;
4652 
4653 	if (vcpu->arch.pio.count)
4654 		goto data_avail;
4655 
4656 	ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
4657 	if (ret) {
4658 data_avail:
4659 		memcpy(val, vcpu->arch.pio_data, size * count);
4660 		trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
4661 		vcpu->arch.pio.count = 0;
4662 		return 1;
4663 	}
4664 
4665 	return 0;
4666 }
4667 
4668 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
4669 				     int size, unsigned short port,
4670 				     const void *val, unsigned int count)
4671 {
4672 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4673 
4674 	memcpy(vcpu->arch.pio_data, val, size * count);
4675 	trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
4676 	return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
4677 }
4678 
4679 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4680 {
4681 	return kvm_x86_ops->get_segment_base(vcpu, seg);
4682 }
4683 
4684 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
4685 {
4686 	kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
4687 }
4688 
4689 int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
4690 {
4691 	if (!need_emulate_wbinvd(vcpu))
4692 		return X86EMUL_CONTINUE;
4693 
4694 	if (kvm_x86_ops->has_wbinvd_exit()) {
4695 		int cpu = get_cpu();
4696 
4697 		cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4698 		smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4699 				wbinvd_ipi, NULL, 1);
4700 		put_cpu();
4701 		cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
4702 	} else
4703 		wbinvd();
4704 	return X86EMUL_CONTINUE;
4705 }
4706 
4707 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4708 {
4709 	kvm_x86_ops->skip_emulated_instruction(vcpu);
4710 	return kvm_emulate_wbinvd_noskip(vcpu);
4711 }
4712 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4713 
4714 
4715 
4716 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
4717 {
4718 	kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
4719 }
4720 
4721 static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
4722 			   unsigned long *dest)
4723 {
4724 	return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
4725 }
4726 
4727 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
4728 			   unsigned long value)
4729 {
4730 
4731 	return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
4732 }
4733 
4734 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4735 {
4736 	return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4737 }
4738 
4739 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
4740 {
4741 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4742 	unsigned long value;
4743 
4744 	switch (cr) {
4745 	case 0:
4746 		value = kvm_read_cr0(vcpu);
4747 		break;
4748 	case 2:
4749 		value = vcpu->arch.cr2;
4750 		break;
4751 	case 3:
4752 		value = kvm_read_cr3(vcpu);
4753 		break;
4754 	case 4:
4755 		value = kvm_read_cr4(vcpu);
4756 		break;
4757 	case 8:
4758 		value = kvm_get_cr8(vcpu);
4759 		break;
4760 	default:
4761 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
4762 		return 0;
4763 	}
4764 
4765 	return value;
4766 }
4767 
4768 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
4769 {
4770 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4771 	int res = 0;
4772 
4773 	switch (cr) {
4774 	case 0:
4775 		res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
4776 		break;
4777 	case 2:
4778 		vcpu->arch.cr2 = val;
4779 		break;
4780 	case 3:
4781 		res = kvm_set_cr3(vcpu, val);
4782 		break;
4783 	case 4:
4784 		res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
4785 		break;
4786 	case 8:
4787 		res = kvm_set_cr8(vcpu, val);
4788 		break;
4789 	default:
4790 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
4791 		res = -1;
4792 	}
4793 
4794 	return res;
4795 }
4796 
4797 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
4798 {
4799 	return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
4800 }
4801 
4802 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4803 {
4804 	kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
4805 }
4806 
4807 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4808 {
4809 	kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
4810 }
4811 
4812 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4813 {
4814 	kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
4815 }
4816 
4817 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4818 {
4819 	kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
4820 }
4821 
4822 static unsigned long emulator_get_cached_segment_base(
4823 	struct x86_emulate_ctxt *ctxt, int seg)
4824 {
4825 	return get_segment_base(emul_to_vcpu(ctxt), seg);
4826 }
4827 
4828 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
4829 				 struct desc_struct *desc, u32 *base3,
4830 				 int seg)
4831 {
4832 	struct kvm_segment var;
4833 
4834 	kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
4835 	*selector = var.selector;
4836 
4837 	if (var.unusable) {
4838 		memset(desc, 0, sizeof(*desc));
4839 		return false;
4840 	}
4841 
4842 	if (var.g)
4843 		var.limit >>= 12;
4844 	set_desc_limit(desc, var.limit);
4845 	set_desc_base(desc, (unsigned long)var.base);
4846 #ifdef CONFIG_X86_64
4847 	if (base3)
4848 		*base3 = var.base >> 32;
4849 #endif
4850 	desc->type = var.type;
4851 	desc->s = var.s;
4852 	desc->dpl = var.dpl;
4853 	desc->p = var.present;
4854 	desc->avl = var.avl;
4855 	desc->l = var.l;
4856 	desc->d = var.db;
4857 	desc->g = var.g;
4858 
4859 	return true;
4860 }
4861 
4862 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
4863 				 struct desc_struct *desc, u32 base3,
4864 				 int seg)
4865 {
4866 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4867 	struct kvm_segment var;
4868 
4869 	var.selector = selector;
4870 	var.base = get_desc_base(desc);
4871 #ifdef CONFIG_X86_64
4872 	var.base |= ((u64)base3) << 32;
4873 #endif
4874 	var.limit = get_desc_limit(desc);
4875 	if (desc->g)
4876 		var.limit = (var.limit << 12) | 0xfff;
4877 	var.type = desc->type;
4878 	var.dpl = desc->dpl;
4879 	var.db = desc->d;
4880 	var.s = desc->s;
4881 	var.l = desc->l;
4882 	var.g = desc->g;
4883 	var.avl = desc->avl;
4884 	var.present = desc->p;
4885 	var.unusable = !var.present;
4886 	var.padding = 0;
4887 
4888 	kvm_set_segment(vcpu, &var, seg);
4889 	return;
4890 }
4891 
4892 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
4893 			    u32 msr_index, u64 *pdata)
4894 {
4895 	struct msr_data msr;
4896 	int r;
4897 
4898 	msr.index = msr_index;
4899 	msr.host_initiated = false;
4900 	r = kvm_get_msr(emul_to_vcpu(ctxt), &msr);
4901 	if (r)
4902 		return r;
4903 
4904 	*pdata = msr.data;
4905 	return 0;
4906 }
4907 
4908 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
4909 			    u32 msr_index, u64 data)
4910 {
4911 	struct msr_data msr;
4912 
4913 	msr.data = data;
4914 	msr.index = msr_index;
4915 	msr.host_initiated = false;
4916 	return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
4917 }
4918 
4919 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
4920 {
4921 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4922 
4923 	return vcpu->arch.smbase;
4924 }
4925 
4926 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
4927 {
4928 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4929 
4930 	vcpu->arch.smbase = smbase;
4931 }
4932 
4933 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
4934 			      u32 pmc)
4935 {
4936 	return kvm_pmu_is_valid_msr_idx(emul_to_vcpu(ctxt), pmc);
4937 }
4938 
4939 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
4940 			     u32 pmc, u64 *pdata)
4941 {
4942 	return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
4943 }
4944 
4945 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
4946 {
4947 	emul_to_vcpu(ctxt)->arch.halt_request = 1;
4948 }
4949 
4950 static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
4951 {
4952 	preempt_disable();
4953 	kvm_load_guest_fpu(emul_to_vcpu(ctxt));
4954 	/*
4955 	 * CR0.TS may reference the host fpu state, not the guest fpu state,
4956 	 * so it may be clear at this point.
4957 	 */
4958 	clts();
4959 }
4960 
4961 static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
4962 {
4963 	preempt_enable();
4964 }
4965 
4966 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
4967 			      struct x86_instruction_info *info,
4968 			      enum x86_intercept_stage stage)
4969 {
4970 	return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
4971 }
4972 
4973 static void emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
4974 			       u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
4975 {
4976 	kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx);
4977 }
4978 
4979 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
4980 {
4981 	return kvm_register_read(emul_to_vcpu(ctxt), reg);
4982 }
4983 
4984 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
4985 {
4986 	kvm_register_write(emul_to_vcpu(ctxt), reg, val);
4987 }
4988 
4989 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
4990 {
4991 	kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked);
4992 }
4993 
4994 static const struct x86_emulate_ops emulate_ops = {
4995 	.read_gpr            = emulator_read_gpr,
4996 	.write_gpr           = emulator_write_gpr,
4997 	.read_std            = kvm_read_guest_virt_system,
4998 	.write_std           = kvm_write_guest_virt_system,
4999 	.fetch               = kvm_fetch_guest_virt,
5000 	.read_emulated       = emulator_read_emulated,
5001 	.write_emulated      = emulator_write_emulated,
5002 	.cmpxchg_emulated    = emulator_cmpxchg_emulated,
5003 	.invlpg              = emulator_invlpg,
5004 	.pio_in_emulated     = emulator_pio_in_emulated,
5005 	.pio_out_emulated    = emulator_pio_out_emulated,
5006 	.get_segment         = emulator_get_segment,
5007 	.set_segment         = emulator_set_segment,
5008 	.get_cached_segment_base = emulator_get_cached_segment_base,
5009 	.get_gdt             = emulator_get_gdt,
5010 	.get_idt	     = emulator_get_idt,
5011 	.set_gdt             = emulator_set_gdt,
5012 	.set_idt	     = emulator_set_idt,
5013 	.get_cr              = emulator_get_cr,
5014 	.set_cr              = emulator_set_cr,
5015 	.cpl                 = emulator_get_cpl,
5016 	.get_dr              = emulator_get_dr,
5017 	.set_dr              = emulator_set_dr,
5018 	.get_smbase          = emulator_get_smbase,
5019 	.set_smbase          = emulator_set_smbase,
5020 	.set_msr             = emulator_set_msr,
5021 	.get_msr             = emulator_get_msr,
5022 	.check_pmc	     = emulator_check_pmc,
5023 	.read_pmc            = emulator_read_pmc,
5024 	.halt                = emulator_halt,
5025 	.wbinvd              = emulator_wbinvd,
5026 	.fix_hypercall       = emulator_fix_hypercall,
5027 	.get_fpu             = emulator_get_fpu,
5028 	.put_fpu             = emulator_put_fpu,
5029 	.intercept           = emulator_intercept,
5030 	.get_cpuid           = emulator_get_cpuid,
5031 	.set_nmi_mask        = emulator_set_nmi_mask,
5032 };
5033 
5034 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
5035 {
5036 	u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
5037 	/*
5038 	 * an sti; sti; sequence only disable interrupts for the first
5039 	 * instruction. So, if the last instruction, be it emulated or
5040 	 * not, left the system with the INT_STI flag enabled, it
5041 	 * means that the last instruction is an sti. We should not
5042 	 * leave the flag on in this case. The same goes for mov ss
5043 	 */
5044 	if (int_shadow & mask)
5045 		mask = 0;
5046 	if (unlikely(int_shadow || mask)) {
5047 		kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
5048 		if (!mask)
5049 			kvm_make_request(KVM_REQ_EVENT, vcpu);
5050 	}
5051 }
5052 
5053 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
5054 {
5055 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5056 	if (ctxt->exception.vector == PF_VECTOR)
5057 		return kvm_propagate_fault(vcpu, &ctxt->exception);
5058 
5059 	if (ctxt->exception.error_code_valid)
5060 		kvm_queue_exception_e(vcpu, ctxt->exception.vector,
5061 				      ctxt->exception.error_code);
5062 	else
5063 		kvm_queue_exception(vcpu, ctxt->exception.vector);
5064 	return false;
5065 }
5066 
5067 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
5068 {
5069 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5070 	int cs_db, cs_l;
5071 
5072 	kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
5073 
5074 	ctxt->eflags = kvm_get_rflags(vcpu);
5075 	ctxt->eip = kvm_rip_read(vcpu);
5076 	ctxt->mode = (!is_protmode(vcpu))		? X86EMUL_MODE_REAL :
5077 		     (ctxt->eflags & X86_EFLAGS_VM)	? X86EMUL_MODE_VM86 :
5078 		     (cs_l && is_long_mode(vcpu))	? X86EMUL_MODE_PROT64 :
5079 		     cs_db				? X86EMUL_MODE_PROT32 :
5080 							  X86EMUL_MODE_PROT16;
5081 	BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
5082 	BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
5083 	BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
5084 	ctxt->emul_flags = vcpu->arch.hflags;
5085 
5086 	init_decode_cache(ctxt);
5087 	vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5088 }
5089 
5090 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
5091 {
5092 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5093 	int ret;
5094 
5095 	init_emulate_ctxt(vcpu);
5096 
5097 	ctxt->op_bytes = 2;
5098 	ctxt->ad_bytes = 2;
5099 	ctxt->_eip = ctxt->eip + inc_eip;
5100 	ret = emulate_int_real(ctxt, irq);
5101 
5102 	if (ret != X86EMUL_CONTINUE)
5103 		return EMULATE_FAIL;
5104 
5105 	ctxt->eip = ctxt->_eip;
5106 	kvm_rip_write(vcpu, ctxt->eip);
5107 	kvm_set_rflags(vcpu, ctxt->eflags);
5108 
5109 	if (irq == NMI_VECTOR)
5110 		vcpu->arch.nmi_pending = 0;
5111 	else
5112 		vcpu->arch.interrupt.pending = false;
5113 
5114 	return EMULATE_DONE;
5115 }
5116 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
5117 
5118 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
5119 {
5120 	int r = EMULATE_DONE;
5121 
5122 	++vcpu->stat.insn_emulation_fail;
5123 	trace_kvm_emulate_insn_failed(vcpu);
5124 	if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
5125 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5126 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5127 		vcpu->run->internal.ndata = 0;
5128 		r = EMULATE_FAIL;
5129 	}
5130 	kvm_queue_exception(vcpu, UD_VECTOR);
5131 
5132 	return r;
5133 }
5134 
5135 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
5136 				  bool write_fault_to_shadow_pgtable,
5137 				  int emulation_type)
5138 {
5139 	gpa_t gpa = cr2;
5140 	pfn_t pfn;
5141 
5142 	if (emulation_type & EMULTYPE_NO_REEXECUTE)
5143 		return false;
5144 
5145 	if (!vcpu->arch.mmu.direct_map) {
5146 		/*
5147 		 * Write permission should be allowed since only
5148 		 * write access need to be emulated.
5149 		 */
5150 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5151 
5152 		/*
5153 		 * If the mapping is invalid in guest, let cpu retry
5154 		 * it to generate fault.
5155 		 */
5156 		if (gpa == UNMAPPED_GVA)
5157 			return true;
5158 	}
5159 
5160 	/*
5161 	 * Do not retry the unhandleable instruction if it faults on the
5162 	 * readonly host memory, otherwise it will goto a infinite loop:
5163 	 * retry instruction -> write #PF -> emulation fail -> retry
5164 	 * instruction -> ...
5165 	 */
5166 	pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
5167 
5168 	/*
5169 	 * If the instruction failed on the error pfn, it can not be fixed,
5170 	 * report the error to userspace.
5171 	 */
5172 	if (is_error_noslot_pfn(pfn))
5173 		return false;
5174 
5175 	kvm_release_pfn_clean(pfn);
5176 
5177 	/* The instructions are well-emulated on direct mmu. */
5178 	if (vcpu->arch.mmu.direct_map) {
5179 		unsigned int indirect_shadow_pages;
5180 
5181 		spin_lock(&vcpu->kvm->mmu_lock);
5182 		indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
5183 		spin_unlock(&vcpu->kvm->mmu_lock);
5184 
5185 		if (indirect_shadow_pages)
5186 			kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5187 
5188 		return true;
5189 	}
5190 
5191 	/*
5192 	 * if emulation was due to access to shadowed page table
5193 	 * and it failed try to unshadow page and re-enter the
5194 	 * guest to let CPU execute the instruction.
5195 	 */
5196 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5197 
5198 	/*
5199 	 * If the access faults on its page table, it can not
5200 	 * be fixed by unprotecting shadow page and it should
5201 	 * be reported to userspace.
5202 	 */
5203 	return !write_fault_to_shadow_pgtable;
5204 }
5205 
5206 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
5207 			      unsigned long cr2,  int emulation_type)
5208 {
5209 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5210 	unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
5211 
5212 	last_retry_eip = vcpu->arch.last_retry_eip;
5213 	last_retry_addr = vcpu->arch.last_retry_addr;
5214 
5215 	/*
5216 	 * If the emulation is caused by #PF and it is non-page_table
5217 	 * writing instruction, it means the VM-EXIT is caused by shadow
5218 	 * page protected, we can zap the shadow page and retry this
5219 	 * instruction directly.
5220 	 *
5221 	 * Note: if the guest uses a non-page-table modifying instruction
5222 	 * on the PDE that points to the instruction, then we will unmap
5223 	 * the instruction and go to an infinite loop. So, we cache the
5224 	 * last retried eip and the last fault address, if we meet the eip
5225 	 * and the address again, we can break out of the potential infinite
5226 	 * loop.
5227 	 */
5228 	vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
5229 
5230 	if (!(emulation_type & EMULTYPE_RETRY))
5231 		return false;
5232 
5233 	if (x86_page_table_writing_insn(ctxt))
5234 		return false;
5235 
5236 	if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
5237 		return false;
5238 
5239 	vcpu->arch.last_retry_eip = ctxt->eip;
5240 	vcpu->arch.last_retry_addr = cr2;
5241 
5242 	if (!vcpu->arch.mmu.direct_map)
5243 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5244 
5245 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5246 
5247 	return true;
5248 }
5249 
5250 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
5251 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
5252 
5253 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
5254 {
5255 	if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
5256 		/* This is a good place to trace that we are exiting SMM.  */
5257 		trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
5258 
5259 		if (unlikely(vcpu->arch.smi_pending)) {
5260 			kvm_make_request(KVM_REQ_SMI, vcpu);
5261 			vcpu->arch.smi_pending = 0;
5262 		} else {
5263 			/* Process a latched INIT, if any.  */
5264 			kvm_make_request(KVM_REQ_EVENT, vcpu);
5265 		}
5266 	}
5267 
5268 	kvm_mmu_reset_context(vcpu);
5269 }
5270 
5271 static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags)
5272 {
5273 	unsigned changed = vcpu->arch.hflags ^ emul_flags;
5274 
5275 	vcpu->arch.hflags = emul_flags;
5276 
5277 	if (changed & HF_SMM_MASK)
5278 		kvm_smm_changed(vcpu);
5279 }
5280 
5281 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
5282 				unsigned long *db)
5283 {
5284 	u32 dr6 = 0;
5285 	int i;
5286 	u32 enable, rwlen;
5287 
5288 	enable = dr7;
5289 	rwlen = dr7 >> 16;
5290 	for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
5291 		if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
5292 			dr6 |= (1 << i);
5293 	return dr6;
5294 }
5295 
5296 static void kvm_vcpu_check_singlestep(struct kvm_vcpu *vcpu, unsigned long rflags, int *r)
5297 {
5298 	struct kvm_run *kvm_run = vcpu->run;
5299 
5300 	/*
5301 	 * rflags is the old, "raw" value of the flags.  The new value has
5302 	 * not been saved yet.
5303 	 *
5304 	 * This is correct even for TF set by the guest, because "the
5305 	 * processor will not generate this exception after the instruction
5306 	 * that sets the TF flag".
5307 	 */
5308 	if (unlikely(rflags & X86_EFLAGS_TF)) {
5309 		if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
5310 			kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 |
5311 						  DR6_RTM;
5312 			kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
5313 			kvm_run->debug.arch.exception = DB_VECTOR;
5314 			kvm_run->exit_reason = KVM_EXIT_DEBUG;
5315 			*r = EMULATE_USER_EXIT;
5316 		} else {
5317 			vcpu->arch.emulate_ctxt.eflags &= ~X86_EFLAGS_TF;
5318 			/*
5319 			 * "Certain debug exceptions may clear bit 0-3.  The
5320 			 * remaining contents of the DR6 register are never
5321 			 * cleared by the processor".
5322 			 */
5323 			vcpu->arch.dr6 &= ~15;
5324 			vcpu->arch.dr6 |= DR6_BS | DR6_RTM;
5325 			kvm_queue_exception(vcpu, DB_VECTOR);
5326 		}
5327 	}
5328 }
5329 
5330 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
5331 {
5332 	if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
5333 	    (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
5334 		struct kvm_run *kvm_run = vcpu->run;
5335 		unsigned long eip = kvm_get_linear_rip(vcpu);
5336 		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5337 					   vcpu->arch.guest_debug_dr7,
5338 					   vcpu->arch.eff_db);
5339 
5340 		if (dr6 != 0) {
5341 			kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
5342 			kvm_run->debug.arch.pc = eip;
5343 			kvm_run->debug.arch.exception = DB_VECTOR;
5344 			kvm_run->exit_reason = KVM_EXIT_DEBUG;
5345 			*r = EMULATE_USER_EXIT;
5346 			return true;
5347 		}
5348 	}
5349 
5350 	if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
5351 	    !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
5352 		unsigned long eip = kvm_get_linear_rip(vcpu);
5353 		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5354 					   vcpu->arch.dr7,
5355 					   vcpu->arch.db);
5356 
5357 		if (dr6 != 0) {
5358 			vcpu->arch.dr6 &= ~15;
5359 			vcpu->arch.dr6 |= dr6 | DR6_RTM;
5360 			kvm_queue_exception(vcpu, DB_VECTOR);
5361 			*r = EMULATE_DONE;
5362 			return true;
5363 		}
5364 	}
5365 
5366 	return false;
5367 }
5368 
5369 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
5370 			    unsigned long cr2,
5371 			    int emulation_type,
5372 			    void *insn,
5373 			    int insn_len)
5374 {
5375 	int r;
5376 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5377 	bool writeback = true;
5378 	bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
5379 
5380 	/*
5381 	 * Clear write_fault_to_shadow_pgtable here to ensure it is
5382 	 * never reused.
5383 	 */
5384 	vcpu->arch.write_fault_to_shadow_pgtable = false;
5385 	kvm_clear_exception_queue(vcpu);
5386 
5387 	if (!(emulation_type & EMULTYPE_NO_DECODE)) {
5388 		init_emulate_ctxt(vcpu);
5389 
5390 		/*
5391 		 * We will reenter on the same instruction since
5392 		 * we do not set complete_userspace_io.  This does not
5393 		 * handle watchpoints yet, those would be handled in
5394 		 * the emulate_ops.
5395 		 */
5396 		if (kvm_vcpu_check_breakpoint(vcpu, &r))
5397 			return r;
5398 
5399 		ctxt->interruptibility = 0;
5400 		ctxt->have_exception = false;
5401 		ctxt->exception.vector = -1;
5402 		ctxt->perm_ok = false;
5403 
5404 		ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
5405 
5406 		r = x86_decode_insn(ctxt, insn, insn_len);
5407 
5408 		trace_kvm_emulate_insn_start(vcpu);
5409 		++vcpu->stat.insn_emulation;
5410 		if (r != EMULATION_OK)  {
5411 			if (emulation_type & EMULTYPE_TRAP_UD)
5412 				return EMULATE_FAIL;
5413 			if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5414 						emulation_type))
5415 				return EMULATE_DONE;
5416 			if (emulation_type & EMULTYPE_SKIP)
5417 				return EMULATE_FAIL;
5418 			return handle_emulation_failure(vcpu);
5419 		}
5420 	}
5421 
5422 	if (emulation_type & EMULTYPE_SKIP) {
5423 		kvm_rip_write(vcpu, ctxt->_eip);
5424 		if (ctxt->eflags & X86_EFLAGS_RF)
5425 			kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
5426 		return EMULATE_DONE;
5427 	}
5428 
5429 	if (retry_instruction(ctxt, cr2, emulation_type))
5430 		return EMULATE_DONE;
5431 
5432 	/* this is needed for vmware backdoor interface to work since it
5433 	   changes registers values  during IO operation */
5434 	if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
5435 		vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5436 		emulator_invalidate_register_cache(ctxt);
5437 	}
5438 
5439 restart:
5440 	r = x86_emulate_insn(ctxt);
5441 
5442 	if (r == EMULATION_INTERCEPTED)
5443 		return EMULATE_DONE;
5444 
5445 	if (r == EMULATION_FAILED) {
5446 		if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5447 					emulation_type))
5448 			return EMULATE_DONE;
5449 
5450 		return handle_emulation_failure(vcpu);
5451 	}
5452 
5453 	if (ctxt->have_exception) {
5454 		r = EMULATE_DONE;
5455 		if (inject_emulated_exception(vcpu))
5456 			return r;
5457 	} else if (vcpu->arch.pio.count) {
5458 		if (!vcpu->arch.pio.in) {
5459 			/* FIXME: return into emulator if single-stepping.  */
5460 			vcpu->arch.pio.count = 0;
5461 		} else {
5462 			writeback = false;
5463 			vcpu->arch.complete_userspace_io = complete_emulated_pio;
5464 		}
5465 		r = EMULATE_USER_EXIT;
5466 	} else if (vcpu->mmio_needed) {
5467 		if (!vcpu->mmio_is_write)
5468 			writeback = false;
5469 		r = EMULATE_USER_EXIT;
5470 		vcpu->arch.complete_userspace_io = complete_emulated_mmio;
5471 	} else if (r == EMULATION_RESTART)
5472 		goto restart;
5473 	else
5474 		r = EMULATE_DONE;
5475 
5476 	if (writeback) {
5477 		unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
5478 		toggle_interruptibility(vcpu, ctxt->interruptibility);
5479 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5480 		if (vcpu->arch.hflags != ctxt->emul_flags)
5481 			kvm_set_hflags(vcpu, ctxt->emul_flags);
5482 		kvm_rip_write(vcpu, ctxt->eip);
5483 		if (r == EMULATE_DONE)
5484 			kvm_vcpu_check_singlestep(vcpu, rflags, &r);
5485 		if (!ctxt->have_exception ||
5486 		    exception_type(ctxt->exception.vector) == EXCPT_TRAP)
5487 			__kvm_set_rflags(vcpu, ctxt->eflags);
5488 
5489 		/*
5490 		 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
5491 		 * do nothing, and it will be requested again as soon as
5492 		 * the shadow expires.  But we still need to check here,
5493 		 * because POPF has no interrupt shadow.
5494 		 */
5495 		if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
5496 			kvm_make_request(KVM_REQ_EVENT, vcpu);
5497 	} else
5498 		vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
5499 
5500 	return r;
5501 }
5502 EXPORT_SYMBOL_GPL(x86_emulate_instruction);
5503 
5504 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
5505 {
5506 	unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
5507 	int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
5508 					    size, port, &val, 1);
5509 	/* do not return to emulator after return from userspace */
5510 	vcpu->arch.pio.count = 0;
5511 	return ret;
5512 }
5513 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
5514 
5515 static void tsc_bad(void *info)
5516 {
5517 	__this_cpu_write(cpu_tsc_khz, 0);
5518 }
5519 
5520 static void tsc_khz_changed(void *data)
5521 {
5522 	struct cpufreq_freqs *freq = data;
5523 	unsigned long khz = 0;
5524 
5525 	if (data)
5526 		khz = freq->new;
5527 	else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5528 		khz = cpufreq_quick_get(raw_smp_processor_id());
5529 	if (!khz)
5530 		khz = tsc_khz;
5531 	__this_cpu_write(cpu_tsc_khz, khz);
5532 }
5533 
5534 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
5535 				     void *data)
5536 {
5537 	struct cpufreq_freqs *freq = data;
5538 	struct kvm *kvm;
5539 	struct kvm_vcpu *vcpu;
5540 	int i, send_ipi = 0;
5541 
5542 	/*
5543 	 * We allow guests to temporarily run on slowing clocks,
5544 	 * provided we notify them after, or to run on accelerating
5545 	 * clocks, provided we notify them before.  Thus time never
5546 	 * goes backwards.
5547 	 *
5548 	 * However, we have a problem.  We can't atomically update
5549 	 * the frequency of a given CPU from this function; it is
5550 	 * merely a notifier, which can be called from any CPU.
5551 	 * Changing the TSC frequency at arbitrary points in time
5552 	 * requires a recomputation of local variables related to
5553 	 * the TSC for each VCPU.  We must flag these local variables
5554 	 * to be updated and be sure the update takes place with the
5555 	 * new frequency before any guests proceed.
5556 	 *
5557 	 * Unfortunately, the combination of hotplug CPU and frequency
5558 	 * change creates an intractable locking scenario; the order
5559 	 * of when these callouts happen is undefined with respect to
5560 	 * CPU hotplug, and they can race with each other.  As such,
5561 	 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
5562 	 * undefined; you can actually have a CPU frequency change take
5563 	 * place in between the computation of X and the setting of the
5564 	 * variable.  To protect against this problem, all updates of
5565 	 * the per_cpu tsc_khz variable are done in an interrupt
5566 	 * protected IPI, and all callers wishing to update the value
5567 	 * must wait for a synchronous IPI to complete (which is trivial
5568 	 * if the caller is on the CPU already).  This establishes the
5569 	 * necessary total order on variable updates.
5570 	 *
5571 	 * Note that because a guest time update may take place
5572 	 * anytime after the setting of the VCPU's request bit, the
5573 	 * correct TSC value must be set before the request.  However,
5574 	 * to ensure the update actually makes it to any guest which
5575 	 * starts running in hardware virtualization between the set
5576 	 * and the acquisition of the spinlock, we must also ping the
5577 	 * CPU after setting the request bit.
5578 	 *
5579 	 */
5580 
5581 	if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
5582 		return 0;
5583 	if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
5584 		return 0;
5585 
5586 	smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5587 
5588 	spin_lock(&kvm_lock);
5589 	list_for_each_entry(kvm, &vm_list, vm_list) {
5590 		kvm_for_each_vcpu(i, vcpu, kvm) {
5591 			if (vcpu->cpu != freq->cpu)
5592 				continue;
5593 			kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5594 			if (vcpu->cpu != smp_processor_id())
5595 				send_ipi = 1;
5596 		}
5597 	}
5598 	spin_unlock(&kvm_lock);
5599 
5600 	if (freq->old < freq->new && send_ipi) {
5601 		/*
5602 		 * We upscale the frequency.  Must make the guest
5603 		 * doesn't see old kvmclock values while running with
5604 		 * the new frequency, otherwise we risk the guest sees
5605 		 * time go backwards.
5606 		 *
5607 		 * In case we update the frequency for another cpu
5608 		 * (which might be in guest context) send an interrupt
5609 		 * to kick the cpu out of guest context.  Next time
5610 		 * guest context is entered kvmclock will be updated,
5611 		 * so the guest will not see stale values.
5612 		 */
5613 		smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5614 	}
5615 	return 0;
5616 }
5617 
5618 static struct notifier_block kvmclock_cpufreq_notifier_block = {
5619 	.notifier_call  = kvmclock_cpufreq_notifier
5620 };
5621 
5622 static int kvmclock_cpu_notifier(struct notifier_block *nfb,
5623 					unsigned long action, void *hcpu)
5624 {
5625 	unsigned int cpu = (unsigned long)hcpu;
5626 
5627 	switch (action) {
5628 		case CPU_ONLINE:
5629 		case CPU_DOWN_FAILED:
5630 			smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5631 			break;
5632 		case CPU_DOWN_PREPARE:
5633 			smp_call_function_single(cpu, tsc_bad, NULL, 1);
5634 			break;
5635 	}
5636 	return NOTIFY_OK;
5637 }
5638 
5639 static struct notifier_block kvmclock_cpu_notifier_block = {
5640 	.notifier_call  = kvmclock_cpu_notifier,
5641 	.priority = -INT_MAX
5642 };
5643 
5644 static void kvm_timer_init(void)
5645 {
5646 	int cpu;
5647 
5648 	max_tsc_khz = tsc_khz;
5649 
5650 	cpu_notifier_register_begin();
5651 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5652 #ifdef CONFIG_CPU_FREQ
5653 		struct cpufreq_policy policy;
5654 		memset(&policy, 0, sizeof(policy));
5655 		cpu = get_cpu();
5656 		cpufreq_get_policy(&policy, cpu);
5657 		if (policy.cpuinfo.max_freq)
5658 			max_tsc_khz = policy.cpuinfo.max_freq;
5659 		put_cpu();
5660 #endif
5661 		cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
5662 					  CPUFREQ_TRANSITION_NOTIFIER);
5663 	}
5664 	pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
5665 	for_each_online_cpu(cpu)
5666 		smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5667 
5668 	__register_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5669 	cpu_notifier_register_done();
5670 
5671 }
5672 
5673 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
5674 
5675 int kvm_is_in_guest(void)
5676 {
5677 	return __this_cpu_read(current_vcpu) != NULL;
5678 }
5679 
5680 static int kvm_is_user_mode(void)
5681 {
5682 	int user_mode = 3;
5683 
5684 	if (__this_cpu_read(current_vcpu))
5685 		user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
5686 
5687 	return user_mode != 0;
5688 }
5689 
5690 static unsigned long kvm_get_guest_ip(void)
5691 {
5692 	unsigned long ip = 0;
5693 
5694 	if (__this_cpu_read(current_vcpu))
5695 		ip = kvm_rip_read(__this_cpu_read(current_vcpu));
5696 
5697 	return ip;
5698 }
5699 
5700 static struct perf_guest_info_callbacks kvm_guest_cbs = {
5701 	.is_in_guest		= kvm_is_in_guest,
5702 	.is_user_mode		= kvm_is_user_mode,
5703 	.get_guest_ip		= kvm_get_guest_ip,
5704 };
5705 
5706 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
5707 {
5708 	__this_cpu_write(current_vcpu, vcpu);
5709 }
5710 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
5711 
5712 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
5713 {
5714 	__this_cpu_write(current_vcpu, NULL);
5715 }
5716 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
5717 
5718 static void kvm_set_mmio_spte_mask(void)
5719 {
5720 	u64 mask;
5721 	int maxphyaddr = boot_cpu_data.x86_phys_bits;
5722 
5723 	/*
5724 	 * Set the reserved bits and the present bit of an paging-structure
5725 	 * entry to generate page fault with PFER.RSV = 1.
5726 	 */
5727 	 /* Mask the reserved physical address bits. */
5728 	mask = rsvd_bits(maxphyaddr, 51);
5729 
5730 	/* Bit 62 is always reserved for 32bit host. */
5731 	mask |= 0x3ull << 62;
5732 
5733 	/* Set the present bit. */
5734 	mask |= 1ull;
5735 
5736 #ifdef CONFIG_X86_64
5737 	/*
5738 	 * If reserved bit is not supported, clear the present bit to disable
5739 	 * mmio page fault.
5740 	 */
5741 	if (maxphyaddr == 52)
5742 		mask &= ~1ull;
5743 #endif
5744 
5745 	kvm_mmu_set_mmio_spte_mask(mask);
5746 }
5747 
5748 #ifdef CONFIG_X86_64
5749 static void pvclock_gtod_update_fn(struct work_struct *work)
5750 {
5751 	struct kvm *kvm;
5752 
5753 	struct kvm_vcpu *vcpu;
5754 	int i;
5755 
5756 	spin_lock(&kvm_lock);
5757 	list_for_each_entry(kvm, &vm_list, vm_list)
5758 		kvm_for_each_vcpu(i, vcpu, kvm)
5759 			kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
5760 	atomic_set(&kvm_guest_has_master_clock, 0);
5761 	spin_unlock(&kvm_lock);
5762 }
5763 
5764 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
5765 
5766 /*
5767  * Notification about pvclock gtod data update.
5768  */
5769 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
5770 			       void *priv)
5771 {
5772 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
5773 	struct timekeeper *tk = priv;
5774 
5775 	update_pvclock_gtod(tk);
5776 
5777 	/* disable master clock if host does not trust, or does not
5778 	 * use, TSC clocksource
5779 	 */
5780 	if (gtod->clock.vclock_mode != VCLOCK_TSC &&
5781 	    atomic_read(&kvm_guest_has_master_clock) != 0)
5782 		queue_work(system_long_wq, &pvclock_gtod_work);
5783 
5784 	return 0;
5785 }
5786 
5787 static struct notifier_block pvclock_gtod_notifier = {
5788 	.notifier_call = pvclock_gtod_notify,
5789 };
5790 #endif
5791 
5792 int kvm_arch_init(void *opaque)
5793 {
5794 	int r;
5795 	struct kvm_x86_ops *ops = opaque;
5796 
5797 	if (kvm_x86_ops) {
5798 		printk(KERN_ERR "kvm: already loaded the other module\n");
5799 		r = -EEXIST;
5800 		goto out;
5801 	}
5802 
5803 	if (!ops->cpu_has_kvm_support()) {
5804 		printk(KERN_ERR "kvm: no hardware support\n");
5805 		r = -EOPNOTSUPP;
5806 		goto out;
5807 	}
5808 	if (ops->disabled_by_bios()) {
5809 		printk(KERN_ERR "kvm: disabled by bios\n");
5810 		r = -EOPNOTSUPP;
5811 		goto out;
5812 	}
5813 
5814 	r = -ENOMEM;
5815 	shared_msrs = alloc_percpu(struct kvm_shared_msrs);
5816 	if (!shared_msrs) {
5817 		printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
5818 		goto out;
5819 	}
5820 
5821 	r = kvm_mmu_module_init();
5822 	if (r)
5823 		goto out_free_percpu;
5824 
5825 	kvm_set_mmio_spte_mask();
5826 
5827 	kvm_x86_ops = ops;
5828 
5829 	kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
5830 			PT_DIRTY_MASK, PT64_NX_MASK, 0);
5831 
5832 	kvm_timer_init();
5833 
5834 	perf_register_guest_info_callbacks(&kvm_guest_cbs);
5835 
5836 	if (cpu_has_xsave)
5837 		host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
5838 
5839 	kvm_lapic_init();
5840 #ifdef CONFIG_X86_64
5841 	pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
5842 #endif
5843 
5844 	return 0;
5845 
5846 out_free_percpu:
5847 	free_percpu(shared_msrs);
5848 out:
5849 	return r;
5850 }
5851 
5852 void kvm_arch_exit(void)
5853 {
5854 	perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
5855 
5856 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5857 		cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
5858 					    CPUFREQ_TRANSITION_NOTIFIER);
5859 	unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5860 #ifdef CONFIG_X86_64
5861 	pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
5862 #endif
5863 	kvm_x86_ops = NULL;
5864 	kvm_mmu_module_exit();
5865 	free_percpu(shared_msrs);
5866 }
5867 
5868 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
5869 {
5870 	++vcpu->stat.halt_exits;
5871 	if (irqchip_in_kernel(vcpu->kvm)) {
5872 		vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
5873 		return 1;
5874 	} else {
5875 		vcpu->run->exit_reason = KVM_EXIT_HLT;
5876 		return 0;
5877 	}
5878 }
5879 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
5880 
5881 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
5882 {
5883 	kvm_x86_ops->skip_emulated_instruction(vcpu);
5884 	return kvm_vcpu_halt(vcpu);
5885 }
5886 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
5887 
5888 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
5889 {
5890 	u64 param, ingpa, outgpa, ret;
5891 	uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
5892 	bool fast, longmode;
5893 
5894 	/*
5895 	 * hypercall generates UD from non zero cpl and real mode
5896 	 * per HYPER-V spec
5897 	 */
5898 	if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
5899 		kvm_queue_exception(vcpu, UD_VECTOR);
5900 		return 0;
5901 	}
5902 
5903 	longmode = is_64_bit_mode(vcpu);
5904 
5905 	if (!longmode) {
5906 		param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
5907 			(kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
5908 		ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
5909 			(kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
5910 		outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
5911 			(kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
5912 	}
5913 #ifdef CONFIG_X86_64
5914 	else {
5915 		param = kvm_register_read(vcpu, VCPU_REGS_RCX);
5916 		ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
5917 		outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
5918 	}
5919 #endif
5920 
5921 	code = param & 0xffff;
5922 	fast = (param >> 16) & 0x1;
5923 	rep_cnt = (param >> 32) & 0xfff;
5924 	rep_idx = (param >> 48) & 0xfff;
5925 
5926 	trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
5927 
5928 	switch (code) {
5929 	case HV_X64_HV_NOTIFY_LONG_SPIN_WAIT:
5930 		kvm_vcpu_on_spin(vcpu);
5931 		break;
5932 	default:
5933 		res = HV_STATUS_INVALID_HYPERCALL_CODE;
5934 		break;
5935 	}
5936 
5937 	ret = res | (((u64)rep_done & 0xfff) << 32);
5938 	if (longmode) {
5939 		kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5940 	} else {
5941 		kvm_register_write(vcpu, VCPU_REGS_RDX, ret >> 32);
5942 		kvm_register_write(vcpu, VCPU_REGS_RAX, ret & 0xffffffff);
5943 	}
5944 
5945 	return 1;
5946 }
5947 
5948 /*
5949  * kvm_pv_kick_cpu_op:  Kick a vcpu.
5950  *
5951  * @apicid - apicid of vcpu to be kicked.
5952  */
5953 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
5954 {
5955 	struct kvm_lapic_irq lapic_irq;
5956 
5957 	lapic_irq.shorthand = 0;
5958 	lapic_irq.dest_mode = 0;
5959 	lapic_irq.dest_id = apicid;
5960 	lapic_irq.msi_redir_hint = false;
5961 
5962 	lapic_irq.delivery_mode = APIC_DM_REMRD;
5963 	kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
5964 }
5965 
5966 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
5967 {
5968 	unsigned long nr, a0, a1, a2, a3, ret;
5969 	int op_64_bit, r = 1;
5970 
5971 	kvm_x86_ops->skip_emulated_instruction(vcpu);
5972 
5973 	if (kvm_hv_hypercall_enabled(vcpu->kvm))
5974 		return kvm_hv_hypercall(vcpu);
5975 
5976 	nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
5977 	a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
5978 	a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
5979 	a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
5980 	a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
5981 
5982 	trace_kvm_hypercall(nr, a0, a1, a2, a3);
5983 
5984 	op_64_bit = is_64_bit_mode(vcpu);
5985 	if (!op_64_bit) {
5986 		nr &= 0xFFFFFFFF;
5987 		a0 &= 0xFFFFFFFF;
5988 		a1 &= 0xFFFFFFFF;
5989 		a2 &= 0xFFFFFFFF;
5990 		a3 &= 0xFFFFFFFF;
5991 	}
5992 
5993 	if (kvm_x86_ops->get_cpl(vcpu) != 0) {
5994 		ret = -KVM_EPERM;
5995 		goto out;
5996 	}
5997 
5998 	switch (nr) {
5999 	case KVM_HC_VAPIC_POLL_IRQ:
6000 		ret = 0;
6001 		break;
6002 	case KVM_HC_KICK_CPU:
6003 		kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
6004 		ret = 0;
6005 		break;
6006 	default:
6007 		ret = -KVM_ENOSYS;
6008 		break;
6009 	}
6010 out:
6011 	if (!op_64_bit)
6012 		ret = (u32)ret;
6013 	kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
6014 	++vcpu->stat.hypercalls;
6015 	return r;
6016 }
6017 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
6018 
6019 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
6020 {
6021 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6022 	char instruction[3];
6023 	unsigned long rip = kvm_rip_read(vcpu);
6024 
6025 	kvm_x86_ops->patch_hypercall(vcpu, instruction);
6026 
6027 	return emulator_write_emulated(ctxt, rip, instruction, 3, NULL);
6028 }
6029 
6030 /*
6031  * Check if userspace requested an interrupt window, and that the
6032  * interrupt window is open.
6033  *
6034  * No need to exit to userspace if we already have an interrupt queued.
6035  */
6036 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
6037 {
6038 	return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
6039 		vcpu->run->request_interrupt_window &&
6040 		kvm_arch_interrupt_allowed(vcpu));
6041 }
6042 
6043 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
6044 {
6045 	struct kvm_run *kvm_run = vcpu->run;
6046 
6047 	kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
6048 	kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
6049 	kvm_run->cr8 = kvm_get_cr8(vcpu);
6050 	kvm_run->apic_base = kvm_get_apic_base(vcpu);
6051 	if (irqchip_in_kernel(vcpu->kvm))
6052 		kvm_run->ready_for_interrupt_injection = 1;
6053 	else
6054 		kvm_run->ready_for_interrupt_injection =
6055 			kvm_arch_interrupt_allowed(vcpu) &&
6056 			!kvm_cpu_has_interrupt(vcpu) &&
6057 			!kvm_event_needs_reinjection(vcpu);
6058 }
6059 
6060 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
6061 {
6062 	int max_irr, tpr;
6063 
6064 	if (!kvm_x86_ops->update_cr8_intercept)
6065 		return;
6066 
6067 	if (!vcpu->arch.apic)
6068 		return;
6069 
6070 	if (!vcpu->arch.apic->vapic_addr)
6071 		max_irr = kvm_lapic_find_highest_irr(vcpu);
6072 	else
6073 		max_irr = -1;
6074 
6075 	if (max_irr != -1)
6076 		max_irr >>= 4;
6077 
6078 	tpr = kvm_lapic_get_cr8(vcpu);
6079 
6080 	kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
6081 }
6082 
6083 static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
6084 {
6085 	int r;
6086 
6087 	/* try to reinject previous events if any */
6088 	if (vcpu->arch.exception.pending) {
6089 		trace_kvm_inj_exception(vcpu->arch.exception.nr,
6090 					vcpu->arch.exception.has_error_code,
6091 					vcpu->arch.exception.error_code);
6092 
6093 		if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
6094 			__kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
6095 					     X86_EFLAGS_RF);
6096 
6097 		if (vcpu->arch.exception.nr == DB_VECTOR &&
6098 		    (vcpu->arch.dr7 & DR7_GD)) {
6099 			vcpu->arch.dr7 &= ~DR7_GD;
6100 			kvm_update_dr7(vcpu);
6101 		}
6102 
6103 		kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
6104 					  vcpu->arch.exception.has_error_code,
6105 					  vcpu->arch.exception.error_code,
6106 					  vcpu->arch.exception.reinject);
6107 		return 0;
6108 	}
6109 
6110 	if (vcpu->arch.nmi_injected) {
6111 		kvm_x86_ops->set_nmi(vcpu);
6112 		return 0;
6113 	}
6114 
6115 	if (vcpu->arch.interrupt.pending) {
6116 		kvm_x86_ops->set_irq(vcpu);
6117 		return 0;
6118 	}
6119 
6120 	if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6121 		r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6122 		if (r != 0)
6123 			return r;
6124 	}
6125 
6126 	/* try to inject new event if pending */
6127 	if (vcpu->arch.nmi_pending) {
6128 		if (kvm_x86_ops->nmi_allowed(vcpu)) {
6129 			--vcpu->arch.nmi_pending;
6130 			vcpu->arch.nmi_injected = true;
6131 			kvm_x86_ops->set_nmi(vcpu);
6132 		}
6133 	} else if (kvm_cpu_has_injectable_intr(vcpu)) {
6134 		/*
6135 		 * Because interrupts can be injected asynchronously, we are
6136 		 * calling check_nested_events again here to avoid a race condition.
6137 		 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
6138 		 * proposal and current concerns.  Perhaps we should be setting
6139 		 * KVM_REQ_EVENT only on certain events and not unconditionally?
6140 		 */
6141 		if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6142 			r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6143 			if (r != 0)
6144 				return r;
6145 		}
6146 		if (kvm_x86_ops->interrupt_allowed(vcpu)) {
6147 			kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
6148 					    false);
6149 			kvm_x86_ops->set_irq(vcpu);
6150 		}
6151 	}
6152 	return 0;
6153 }
6154 
6155 static void process_nmi(struct kvm_vcpu *vcpu)
6156 {
6157 	unsigned limit = 2;
6158 
6159 	/*
6160 	 * x86 is limited to one NMI running, and one NMI pending after it.
6161 	 * If an NMI is already in progress, limit further NMIs to just one.
6162 	 * Otherwise, allow two (and we'll inject the first one immediately).
6163 	 */
6164 	if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
6165 		limit = 1;
6166 
6167 	vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
6168 	vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
6169 	kvm_make_request(KVM_REQ_EVENT, vcpu);
6170 }
6171 
6172 #define put_smstate(type, buf, offset, val)			  \
6173 	*(type *)((buf) + (offset) - 0x7e00) = val
6174 
6175 static u32 process_smi_get_segment_flags(struct kvm_segment *seg)
6176 {
6177 	u32 flags = 0;
6178 	flags |= seg->g       << 23;
6179 	flags |= seg->db      << 22;
6180 	flags |= seg->l       << 21;
6181 	flags |= seg->avl     << 20;
6182 	flags |= seg->present << 15;
6183 	flags |= seg->dpl     << 13;
6184 	flags |= seg->s       << 12;
6185 	flags |= seg->type    << 8;
6186 	return flags;
6187 }
6188 
6189 static void process_smi_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
6190 {
6191 	struct kvm_segment seg;
6192 	int offset;
6193 
6194 	kvm_get_segment(vcpu, &seg, n);
6195 	put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
6196 
6197 	if (n < 3)
6198 		offset = 0x7f84 + n * 12;
6199 	else
6200 		offset = 0x7f2c + (n - 3) * 12;
6201 
6202 	put_smstate(u32, buf, offset + 8, seg.base);
6203 	put_smstate(u32, buf, offset + 4, seg.limit);
6204 	put_smstate(u32, buf, offset, process_smi_get_segment_flags(&seg));
6205 }
6206 
6207 static void process_smi_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
6208 {
6209 	struct kvm_segment seg;
6210 	int offset;
6211 	u16 flags;
6212 
6213 	kvm_get_segment(vcpu, &seg, n);
6214 	offset = 0x7e00 + n * 16;
6215 
6216 	flags = process_smi_get_segment_flags(&seg) >> 8;
6217 	put_smstate(u16, buf, offset, seg.selector);
6218 	put_smstate(u16, buf, offset + 2, flags);
6219 	put_smstate(u32, buf, offset + 4, seg.limit);
6220 	put_smstate(u64, buf, offset + 8, seg.base);
6221 }
6222 
6223 static void process_smi_save_state_32(struct kvm_vcpu *vcpu, char *buf)
6224 {
6225 	struct desc_ptr dt;
6226 	struct kvm_segment seg;
6227 	unsigned long val;
6228 	int i;
6229 
6230 	put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
6231 	put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
6232 	put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
6233 	put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
6234 
6235 	for (i = 0; i < 8; i++)
6236 		put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
6237 
6238 	kvm_get_dr(vcpu, 6, &val);
6239 	put_smstate(u32, buf, 0x7fcc, (u32)val);
6240 	kvm_get_dr(vcpu, 7, &val);
6241 	put_smstate(u32, buf, 0x7fc8, (u32)val);
6242 
6243 	kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6244 	put_smstate(u32, buf, 0x7fc4, seg.selector);
6245 	put_smstate(u32, buf, 0x7f64, seg.base);
6246 	put_smstate(u32, buf, 0x7f60, seg.limit);
6247 	put_smstate(u32, buf, 0x7f5c, process_smi_get_segment_flags(&seg));
6248 
6249 	kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6250 	put_smstate(u32, buf, 0x7fc0, seg.selector);
6251 	put_smstate(u32, buf, 0x7f80, seg.base);
6252 	put_smstate(u32, buf, 0x7f7c, seg.limit);
6253 	put_smstate(u32, buf, 0x7f78, process_smi_get_segment_flags(&seg));
6254 
6255 	kvm_x86_ops->get_gdt(vcpu, &dt);
6256 	put_smstate(u32, buf, 0x7f74, dt.address);
6257 	put_smstate(u32, buf, 0x7f70, dt.size);
6258 
6259 	kvm_x86_ops->get_idt(vcpu, &dt);
6260 	put_smstate(u32, buf, 0x7f58, dt.address);
6261 	put_smstate(u32, buf, 0x7f54, dt.size);
6262 
6263 	for (i = 0; i < 6; i++)
6264 		process_smi_save_seg_32(vcpu, buf, i);
6265 
6266 	put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
6267 
6268 	/* revision id */
6269 	put_smstate(u32, buf, 0x7efc, 0x00020000);
6270 	put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
6271 }
6272 
6273 static void process_smi_save_state_64(struct kvm_vcpu *vcpu, char *buf)
6274 {
6275 #ifdef CONFIG_X86_64
6276 	struct desc_ptr dt;
6277 	struct kvm_segment seg;
6278 	unsigned long val;
6279 	int i;
6280 
6281 	for (i = 0; i < 16; i++)
6282 		put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
6283 
6284 	put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
6285 	put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
6286 
6287 	kvm_get_dr(vcpu, 6, &val);
6288 	put_smstate(u64, buf, 0x7f68, val);
6289 	kvm_get_dr(vcpu, 7, &val);
6290 	put_smstate(u64, buf, 0x7f60, val);
6291 
6292 	put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
6293 	put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
6294 	put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
6295 
6296 	put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
6297 
6298 	/* revision id */
6299 	put_smstate(u32, buf, 0x7efc, 0x00020064);
6300 
6301 	put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
6302 
6303 	kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6304 	put_smstate(u16, buf, 0x7e90, seg.selector);
6305 	put_smstate(u16, buf, 0x7e92, process_smi_get_segment_flags(&seg) >> 8);
6306 	put_smstate(u32, buf, 0x7e94, seg.limit);
6307 	put_smstate(u64, buf, 0x7e98, seg.base);
6308 
6309 	kvm_x86_ops->get_idt(vcpu, &dt);
6310 	put_smstate(u32, buf, 0x7e84, dt.size);
6311 	put_smstate(u64, buf, 0x7e88, dt.address);
6312 
6313 	kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6314 	put_smstate(u16, buf, 0x7e70, seg.selector);
6315 	put_smstate(u16, buf, 0x7e72, process_smi_get_segment_flags(&seg) >> 8);
6316 	put_smstate(u32, buf, 0x7e74, seg.limit);
6317 	put_smstate(u64, buf, 0x7e78, seg.base);
6318 
6319 	kvm_x86_ops->get_gdt(vcpu, &dt);
6320 	put_smstate(u32, buf, 0x7e64, dt.size);
6321 	put_smstate(u64, buf, 0x7e68, dt.address);
6322 
6323 	for (i = 0; i < 6; i++)
6324 		process_smi_save_seg_64(vcpu, buf, i);
6325 #else
6326 	WARN_ON_ONCE(1);
6327 #endif
6328 }
6329 
6330 static void process_smi(struct kvm_vcpu *vcpu)
6331 {
6332 	struct kvm_segment cs, ds;
6333 	char buf[512];
6334 	u32 cr0;
6335 
6336 	if (is_smm(vcpu)) {
6337 		vcpu->arch.smi_pending = true;
6338 		return;
6339 	}
6340 
6341 	trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
6342 	vcpu->arch.hflags |= HF_SMM_MASK;
6343 	memset(buf, 0, 512);
6344 	if (guest_cpuid_has_longmode(vcpu))
6345 		process_smi_save_state_64(vcpu, buf);
6346 	else
6347 		process_smi_save_state_32(vcpu, buf);
6348 
6349 	kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
6350 
6351 	if (kvm_x86_ops->get_nmi_mask(vcpu))
6352 		vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
6353 	else
6354 		kvm_x86_ops->set_nmi_mask(vcpu, true);
6355 
6356 	kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
6357 	kvm_rip_write(vcpu, 0x8000);
6358 
6359 	cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
6360 	kvm_x86_ops->set_cr0(vcpu, cr0);
6361 	vcpu->arch.cr0 = cr0;
6362 
6363 	kvm_x86_ops->set_cr4(vcpu, 0);
6364 
6365 	__kvm_set_dr(vcpu, 7, DR7_FIXED_1);
6366 
6367 	cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
6368 	cs.base = vcpu->arch.smbase;
6369 
6370 	ds.selector = 0;
6371 	ds.base = 0;
6372 
6373 	cs.limit    = ds.limit = 0xffffffff;
6374 	cs.type     = ds.type = 0x3;
6375 	cs.dpl      = ds.dpl = 0;
6376 	cs.db       = ds.db = 0;
6377 	cs.s        = ds.s = 1;
6378 	cs.l        = ds.l = 0;
6379 	cs.g        = ds.g = 1;
6380 	cs.avl      = ds.avl = 0;
6381 	cs.present  = ds.present = 1;
6382 	cs.unusable = ds.unusable = 0;
6383 	cs.padding  = ds.padding = 0;
6384 
6385 	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
6386 	kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
6387 	kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
6388 	kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
6389 	kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
6390 	kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
6391 
6392 	if (guest_cpuid_has_longmode(vcpu))
6393 		kvm_x86_ops->set_efer(vcpu, 0);
6394 
6395 	kvm_update_cpuid(vcpu);
6396 	kvm_mmu_reset_context(vcpu);
6397 }
6398 
6399 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
6400 {
6401 	u64 eoi_exit_bitmap[4];
6402 	u32 tmr[8];
6403 
6404 	if (!kvm_apic_hw_enabled(vcpu->arch.apic))
6405 		return;
6406 
6407 	memset(eoi_exit_bitmap, 0, 32);
6408 	memset(tmr, 0, 32);
6409 
6410 	kvm_ioapic_scan_entry(vcpu, eoi_exit_bitmap, tmr);
6411 	kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
6412 	kvm_apic_update_tmr(vcpu, tmr);
6413 }
6414 
6415 static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu)
6416 {
6417 	++vcpu->stat.tlb_flush;
6418 	kvm_x86_ops->tlb_flush(vcpu);
6419 }
6420 
6421 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
6422 {
6423 	struct page *page = NULL;
6424 
6425 	if (!irqchip_in_kernel(vcpu->kvm))
6426 		return;
6427 
6428 	if (!kvm_x86_ops->set_apic_access_page_addr)
6429 		return;
6430 
6431 	page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
6432 	if (is_error_page(page))
6433 		return;
6434 	kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page));
6435 
6436 	/*
6437 	 * Do not pin apic access page in memory, the MMU notifier
6438 	 * will call us again if it is migrated or swapped out.
6439 	 */
6440 	put_page(page);
6441 }
6442 EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page);
6443 
6444 void kvm_arch_mmu_notifier_invalidate_page(struct kvm *kvm,
6445 					   unsigned long address)
6446 {
6447 	/*
6448 	 * The physical address of apic access page is stored in the VMCS.
6449 	 * Update it when it becomes invalid.
6450 	 */
6451 	if (address == gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT))
6452 		kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
6453 }
6454 
6455 /*
6456  * Returns 1 to let vcpu_run() continue the guest execution loop without
6457  * exiting to the userspace.  Otherwise, the value will be returned to the
6458  * userspace.
6459  */
6460 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
6461 {
6462 	int r;
6463 	bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
6464 		vcpu->run->request_interrupt_window;
6465 	bool req_immediate_exit = false;
6466 
6467 	if (vcpu->requests) {
6468 		if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
6469 			kvm_mmu_unload(vcpu);
6470 		if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
6471 			__kvm_migrate_timers(vcpu);
6472 		if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
6473 			kvm_gen_update_masterclock(vcpu->kvm);
6474 		if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
6475 			kvm_gen_kvmclock_update(vcpu);
6476 		if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
6477 			r = kvm_guest_time_update(vcpu);
6478 			if (unlikely(r))
6479 				goto out;
6480 		}
6481 		if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
6482 			kvm_mmu_sync_roots(vcpu);
6483 		if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
6484 			kvm_vcpu_flush_tlb(vcpu);
6485 		if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
6486 			vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
6487 			r = 0;
6488 			goto out;
6489 		}
6490 		if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
6491 			vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
6492 			r = 0;
6493 			goto out;
6494 		}
6495 		if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
6496 			vcpu->fpu_active = 0;
6497 			kvm_x86_ops->fpu_deactivate(vcpu);
6498 		}
6499 		if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
6500 			/* Page is swapped out. Do synthetic halt */
6501 			vcpu->arch.apf.halted = true;
6502 			r = 1;
6503 			goto out;
6504 		}
6505 		if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
6506 			record_steal_time(vcpu);
6507 		if (kvm_check_request(KVM_REQ_SMI, vcpu))
6508 			process_smi(vcpu);
6509 		if (kvm_check_request(KVM_REQ_NMI, vcpu))
6510 			process_nmi(vcpu);
6511 		if (kvm_check_request(KVM_REQ_PMU, vcpu))
6512 			kvm_pmu_handle_event(vcpu);
6513 		if (kvm_check_request(KVM_REQ_PMI, vcpu))
6514 			kvm_pmu_deliver_pmi(vcpu);
6515 		if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
6516 			vcpu_scan_ioapic(vcpu);
6517 		if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
6518 			kvm_vcpu_reload_apic_access_page(vcpu);
6519 	}
6520 
6521 	if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
6522 		kvm_apic_accept_events(vcpu);
6523 		if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
6524 			r = 1;
6525 			goto out;
6526 		}
6527 
6528 		if (inject_pending_event(vcpu, req_int_win) != 0)
6529 			req_immediate_exit = true;
6530 		/* enable NMI/IRQ window open exits if needed */
6531 		else if (vcpu->arch.nmi_pending)
6532 			kvm_x86_ops->enable_nmi_window(vcpu);
6533 		else if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
6534 			kvm_x86_ops->enable_irq_window(vcpu);
6535 
6536 		if (kvm_lapic_enabled(vcpu)) {
6537 			/*
6538 			 * Update architecture specific hints for APIC
6539 			 * virtual interrupt delivery.
6540 			 */
6541 			if (kvm_x86_ops->hwapic_irr_update)
6542 				kvm_x86_ops->hwapic_irr_update(vcpu,
6543 					kvm_lapic_find_highest_irr(vcpu));
6544 			update_cr8_intercept(vcpu);
6545 			kvm_lapic_sync_to_vapic(vcpu);
6546 		}
6547 	}
6548 
6549 	r = kvm_mmu_reload(vcpu);
6550 	if (unlikely(r)) {
6551 		goto cancel_injection;
6552 	}
6553 
6554 	preempt_disable();
6555 
6556 	kvm_x86_ops->prepare_guest_switch(vcpu);
6557 	if (vcpu->fpu_active)
6558 		kvm_load_guest_fpu(vcpu);
6559 	kvm_load_guest_xcr0(vcpu);
6560 
6561 	vcpu->mode = IN_GUEST_MODE;
6562 
6563 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6564 
6565 	/* We should set ->mode before check ->requests,
6566 	 * see the comment in make_all_cpus_request.
6567 	 */
6568 	smp_mb__after_srcu_read_unlock();
6569 
6570 	local_irq_disable();
6571 
6572 	if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests
6573 	    || need_resched() || signal_pending(current)) {
6574 		vcpu->mode = OUTSIDE_GUEST_MODE;
6575 		smp_wmb();
6576 		local_irq_enable();
6577 		preempt_enable();
6578 		vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6579 		r = 1;
6580 		goto cancel_injection;
6581 	}
6582 
6583 	if (req_immediate_exit)
6584 		smp_send_reschedule(vcpu->cpu);
6585 
6586 	__kvm_guest_enter();
6587 
6588 	if (unlikely(vcpu->arch.switch_db_regs)) {
6589 		set_debugreg(0, 7);
6590 		set_debugreg(vcpu->arch.eff_db[0], 0);
6591 		set_debugreg(vcpu->arch.eff_db[1], 1);
6592 		set_debugreg(vcpu->arch.eff_db[2], 2);
6593 		set_debugreg(vcpu->arch.eff_db[3], 3);
6594 		set_debugreg(vcpu->arch.dr6, 6);
6595 		vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
6596 	}
6597 
6598 	trace_kvm_entry(vcpu->vcpu_id);
6599 	wait_lapic_expire(vcpu);
6600 	kvm_x86_ops->run(vcpu);
6601 
6602 	/*
6603 	 * Do this here before restoring debug registers on the host.  And
6604 	 * since we do this before handling the vmexit, a DR access vmexit
6605 	 * can (a) read the correct value of the debug registers, (b) set
6606 	 * KVM_DEBUGREG_WONT_EXIT again.
6607 	 */
6608 	if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
6609 		int i;
6610 
6611 		WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
6612 		kvm_x86_ops->sync_dirty_debug_regs(vcpu);
6613 		for (i = 0; i < KVM_NR_DB_REGS; i++)
6614 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
6615 	}
6616 
6617 	/*
6618 	 * If the guest has used debug registers, at least dr7
6619 	 * will be disabled while returning to the host.
6620 	 * If we don't have active breakpoints in the host, we don't
6621 	 * care about the messed up debug address registers. But if
6622 	 * we have some of them active, restore the old state.
6623 	 */
6624 	if (hw_breakpoint_active())
6625 		hw_breakpoint_restore();
6626 
6627 	vcpu->arch.last_guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu,
6628 							   native_read_tsc());
6629 
6630 	vcpu->mode = OUTSIDE_GUEST_MODE;
6631 	smp_wmb();
6632 
6633 	/* Interrupt is enabled by handle_external_intr() */
6634 	kvm_x86_ops->handle_external_intr(vcpu);
6635 
6636 	++vcpu->stat.exits;
6637 
6638 	/*
6639 	 * We must have an instruction between local_irq_enable() and
6640 	 * kvm_guest_exit(), so the timer interrupt isn't delayed by
6641 	 * the interrupt shadow.  The stat.exits increment will do nicely.
6642 	 * But we need to prevent reordering, hence this barrier():
6643 	 */
6644 	barrier();
6645 
6646 	kvm_guest_exit();
6647 
6648 	preempt_enable();
6649 
6650 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6651 
6652 	/*
6653 	 * Profile KVM exit RIPs:
6654 	 */
6655 	if (unlikely(prof_on == KVM_PROFILING)) {
6656 		unsigned long rip = kvm_rip_read(vcpu);
6657 		profile_hit(KVM_PROFILING, (void *)rip);
6658 	}
6659 
6660 	if (unlikely(vcpu->arch.tsc_always_catchup))
6661 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6662 
6663 	if (vcpu->arch.apic_attention)
6664 		kvm_lapic_sync_from_vapic(vcpu);
6665 
6666 	r = kvm_x86_ops->handle_exit(vcpu);
6667 	return r;
6668 
6669 cancel_injection:
6670 	kvm_x86_ops->cancel_injection(vcpu);
6671 	if (unlikely(vcpu->arch.apic_attention))
6672 		kvm_lapic_sync_from_vapic(vcpu);
6673 out:
6674 	return r;
6675 }
6676 
6677 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
6678 {
6679 	if (!kvm_arch_vcpu_runnable(vcpu)) {
6680 		srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6681 		kvm_vcpu_block(vcpu);
6682 		vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6683 		if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
6684 			return 1;
6685 	}
6686 
6687 	kvm_apic_accept_events(vcpu);
6688 	switch(vcpu->arch.mp_state) {
6689 	case KVM_MP_STATE_HALTED:
6690 		vcpu->arch.pv.pv_unhalted = false;
6691 		vcpu->arch.mp_state =
6692 			KVM_MP_STATE_RUNNABLE;
6693 	case KVM_MP_STATE_RUNNABLE:
6694 		vcpu->arch.apf.halted = false;
6695 		break;
6696 	case KVM_MP_STATE_INIT_RECEIVED:
6697 		break;
6698 	default:
6699 		return -EINTR;
6700 		break;
6701 	}
6702 	return 1;
6703 }
6704 
6705 static int vcpu_run(struct kvm_vcpu *vcpu)
6706 {
6707 	int r;
6708 	struct kvm *kvm = vcpu->kvm;
6709 
6710 	vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6711 
6712 	for (;;) {
6713 		if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
6714 		    !vcpu->arch.apf.halted)
6715 			r = vcpu_enter_guest(vcpu);
6716 		else
6717 			r = vcpu_block(kvm, vcpu);
6718 		if (r <= 0)
6719 			break;
6720 
6721 		clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
6722 		if (kvm_cpu_has_pending_timer(vcpu))
6723 			kvm_inject_pending_timer_irqs(vcpu);
6724 
6725 		if (dm_request_for_irq_injection(vcpu)) {
6726 			r = -EINTR;
6727 			vcpu->run->exit_reason = KVM_EXIT_INTR;
6728 			++vcpu->stat.request_irq_exits;
6729 			break;
6730 		}
6731 
6732 		kvm_check_async_pf_completion(vcpu);
6733 
6734 		if (signal_pending(current)) {
6735 			r = -EINTR;
6736 			vcpu->run->exit_reason = KVM_EXIT_INTR;
6737 			++vcpu->stat.signal_exits;
6738 			break;
6739 		}
6740 		if (need_resched()) {
6741 			srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6742 			cond_resched();
6743 			vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6744 		}
6745 	}
6746 
6747 	srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6748 
6749 	return r;
6750 }
6751 
6752 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
6753 {
6754 	int r;
6755 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6756 	r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
6757 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6758 	if (r != EMULATE_DONE)
6759 		return 0;
6760 	return 1;
6761 }
6762 
6763 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
6764 {
6765 	BUG_ON(!vcpu->arch.pio.count);
6766 
6767 	return complete_emulated_io(vcpu);
6768 }
6769 
6770 /*
6771  * Implements the following, as a state machine:
6772  *
6773  * read:
6774  *   for each fragment
6775  *     for each mmio piece in the fragment
6776  *       write gpa, len
6777  *       exit
6778  *       copy data
6779  *   execute insn
6780  *
6781  * write:
6782  *   for each fragment
6783  *     for each mmio piece in the fragment
6784  *       write gpa, len
6785  *       copy data
6786  *       exit
6787  */
6788 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
6789 {
6790 	struct kvm_run *run = vcpu->run;
6791 	struct kvm_mmio_fragment *frag;
6792 	unsigned len;
6793 
6794 	BUG_ON(!vcpu->mmio_needed);
6795 
6796 	/* Complete previous fragment */
6797 	frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
6798 	len = min(8u, frag->len);
6799 	if (!vcpu->mmio_is_write)
6800 		memcpy(frag->data, run->mmio.data, len);
6801 
6802 	if (frag->len <= 8) {
6803 		/* Switch to the next fragment. */
6804 		frag++;
6805 		vcpu->mmio_cur_fragment++;
6806 	} else {
6807 		/* Go forward to the next mmio piece. */
6808 		frag->data += len;
6809 		frag->gpa += len;
6810 		frag->len -= len;
6811 	}
6812 
6813 	if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
6814 		vcpu->mmio_needed = 0;
6815 
6816 		/* FIXME: return into emulator if single-stepping.  */
6817 		if (vcpu->mmio_is_write)
6818 			return 1;
6819 		vcpu->mmio_read_completed = 1;
6820 		return complete_emulated_io(vcpu);
6821 	}
6822 
6823 	run->exit_reason = KVM_EXIT_MMIO;
6824 	run->mmio.phys_addr = frag->gpa;
6825 	if (vcpu->mmio_is_write)
6826 		memcpy(run->mmio.data, frag->data, min(8u, frag->len));
6827 	run->mmio.len = min(8u, frag->len);
6828 	run->mmio.is_write = vcpu->mmio_is_write;
6829 	vcpu->arch.complete_userspace_io = complete_emulated_mmio;
6830 	return 0;
6831 }
6832 
6833 
6834 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
6835 {
6836 	struct fpu *fpu = &current->thread.fpu;
6837 	int r;
6838 	sigset_t sigsaved;
6839 
6840 	fpu__activate_curr(fpu);
6841 
6842 	if (vcpu->sigset_active)
6843 		sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
6844 
6845 	if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
6846 		kvm_vcpu_block(vcpu);
6847 		kvm_apic_accept_events(vcpu);
6848 		clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
6849 		r = -EAGAIN;
6850 		goto out;
6851 	}
6852 
6853 	/* re-sync apic's tpr */
6854 	if (!irqchip_in_kernel(vcpu->kvm)) {
6855 		if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
6856 			r = -EINVAL;
6857 			goto out;
6858 		}
6859 	}
6860 
6861 	if (unlikely(vcpu->arch.complete_userspace_io)) {
6862 		int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
6863 		vcpu->arch.complete_userspace_io = NULL;
6864 		r = cui(vcpu);
6865 		if (r <= 0)
6866 			goto out;
6867 	} else
6868 		WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
6869 
6870 	r = vcpu_run(vcpu);
6871 
6872 out:
6873 	post_kvm_run_save(vcpu);
6874 	if (vcpu->sigset_active)
6875 		sigprocmask(SIG_SETMASK, &sigsaved, NULL);
6876 
6877 	return r;
6878 }
6879 
6880 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6881 {
6882 	if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
6883 		/*
6884 		 * We are here if userspace calls get_regs() in the middle of
6885 		 * instruction emulation. Registers state needs to be copied
6886 		 * back from emulation context to vcpu. Userspace shouldn't do
6887 		 * that usually, but some bad designed PV devices (vmware
6888 		 * backdoor interface) need this to work
6889 		 */
6890 		emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
6891 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6892 	}
6893 	regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
6894 	regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
6895 	regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
6896 	regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
6897 	regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
6898 	regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
6899 	regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
6900 	regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
6901 #ifdef CONFIG_X86_64
6902 	regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
6903 	regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
6904 	regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
6905 	regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
6906 	regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
6907 	regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
6908 	regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
6909 	regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
6910 #endif
6911 
6912 	regs->rip = kvm_rip_read(vcpu);
6913 	regs->rflags = kvm_get_rflags(vcpu);
6914 
6915 	return 0;
6916 }
6917 
6918 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6919 {
6920 	vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
6921 	vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6922 
6923 	kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
6924 	kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
6925 	kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
6926 	kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
6927 	kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
6928 	kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
6929 	kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
6930 	kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
6931 #ifdef CONFIG_X86_64
6932 	kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
6933 	kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
6934 	kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
6935 	kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
6936 	kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
6937 	kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
6938 	kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
6939 	kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
6940 #endif
6941 
6942 	kvm_rip_write(vcpu, regs->rip);
6943 	kvm_set_rflags(vcpu, regs->rflags);
6944 
6945 	vcpu->arch.exception.pending = false;
6946 
6947 	kvm_make_request(KVM_REQ_EVENT, vcpu);
6948 
6949 	return 0;
6950 }
6951 
6952 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
6953 {
6954 	struct kvm_segment cs;
6955 
6956 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
6957 	*db = cs.db;
6958 	*l = cs.l;
6959 }
6960 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
6961 
6962 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
6963 				  struct kvm_sregs *sregs)
6964 {
6965 	struct desc_ptr dt;
6966 
6967 	kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6968 	kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6969 	kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6970 	kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6971 	kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6972 	kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
6973 
6974 	kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6975 	kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
6976 
6977 	kvm_x86_ops->get_idt(vcpu, &dt);
6978 	sregs->idt.limit = dt.size;
6979 	sregs->idt.base = dt.address;
6980 	kvm_x86_ops->get_gdt(vcpu, &dt);
6981 	sregs->gdt.limit = dt.size;
6982 	sregs->gdt.base = dt.address;
6983 
6984 	sregs->cr0 = kvm_read_cr0(vcpu);
6985 	sregs->cr2 = vcpu->arch.cr2;
6986 	sregs->cr3 = kvm_read_cr3(vcpu);
6987 	sregs->cr4 = kvm_read_cr4(vcpu);
6988 	sregs->cr8 = kvm_get_cr8(vcpu);
6989 	sregs->efer = vcpu->arch.efer;
6990 	sregs->apic_base = kvm_get_apic_base(vcpu);
6991 
6992 	memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
6993 
6994 	if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
6995 		set_bit(vcpu->arch.interrupt.nr,
6996 			(unsigned long *)sregs->interrupt_bitmap);
6997 
6998 	return 0;
6999 }
7000 
7001 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
7002 				    struct kvm_mp_state *mp_state)
7003 {
7004 	kvm_apic_accept_events(vcpu);
7005 	if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
7006 					vcpu->arch.pv.pv_unhalted)
7007 		mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
7008 	else
7009 		mp_state->mp_state = vcpu->arch.mp_state;
7010 
7011 	return 0;
7012 }
7013 
7014 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
7015 				    struct kvm_mp_state *mp_state)
7016 {
7017 	if (!kvm_vcpu_has_lapic(vcpu) &&
7018 	    mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
7019 		return -EINVAL;
7020 
7021 	if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
7022 		vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
7023 		set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
7024 	} else
7025 		vcpu->arch.mp_state = mp_state->mp_state;
7026 	kvm_make_request(KVM_REQ_EVENT, vcpu);
7027 	return 0;
7028 }
7029 
7030 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
7031 		    int reason, bool has_error_code, u32 error_code)
7032 {
7033 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
7034 	int ret;
7035 
7036 	init_emulate_ctxt(vcpu);
7037 
7038 	ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
7039 				   has_error_code, error_code);
7040 
7041 	if (ret)
7042 		return EMULATE_FAIL;
7043 
7044 	kvm_rip_write(vcpu, ctxt->eip);
7045 	kvm_set_rflags(vcpu, ctxt->eflags);
7046 	kvm_make_request(KVM_REQ_EVENT, vcpu);
7047 	return EMULATE_DONE;
7048 }
7049 EXPORT_SYMBOL_GPL(kvm_task_switch);
7050 
7051 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
7052 				  struct kvm_sregs *sregs)
7053 {
7054 	struct msr_data apic_base_msr;
7055 	int mmu_reset_needed = 0;
7056 	int pending_vec, max_bits, idx;
7057 	struct desc_ptr dt;
7058 
7059 	if (!guest_cpuid_has_xsave(vcpu) && (sregs->cr4 & X86_CR4_OSXSAVE))
7060 		return -EINVAL;
7061 
7062 	dt.size = sregs->idt.limit;
7063 	dt.address = sregs->idt.base;
7064 	kvm_x86_ops->set_idt(vcpu, &dt);
7065 	dt.size = sregs->gdt.limit;
7066 	dt.address = sregs->gdt.base;
7067 	kvm_x86_ops->set_gdt(vcpu, &dt);
7068 
7069 	vcpu->arch.cr2 = sregs->cr2;
7070 	mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
7071 	vcpu->arch.cr3 = sregs->cr3;
7072 	__set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
7073 
7074 	kvm_set_cr8(vcpu, sregs->cr8);
7075 
7076 	mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
7077 	kvm_x86_ops->set_efer(vcpu, sregs->efer);
7078 	apic_base_msr.data = sregs->apic_base;
7079 	apic_base_msr.host_initiated = true;
7080 	kvm_set_apic_base(vcpu, &apic_base_msr);
7081 
7082 	mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
7083 	kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
7084 	vcpu->arch.cr0 = sregs->cr0;
7085 
7086 	mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
7087 	kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
7088 	if (sregs->cr4 & X86_CR4_OSXSAVE)
7089 		kvm_update_cpuid(vcpu);
7090 
7091 	idx = srcu_read_lock(&vcpu->kvm->srcu);
7092 	if (!is_long_mode(vcpu) && is_pae(vcpu)) {
7093 		load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
7094 		mmu_reset_needed = 1;
7095 	}
7096 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
7097 
7098 	if (mmu_reset_needed)
7099 		kvm_mmu_reset_context(vcpu);
7100 
7101 	max_bits = KVM_NR_INTERRUPTS;
7102 	pending_vec = find_first_bit(
7103 		(const unsigned long *)sregs->interrupt_bitmap, max_bits);
7104 	if (pending_vec < max_bits) {
7105 		kvm_queue_interrupt(vcpu, pending_vec, false);
7106 		pr_debug("Set back pending irq %d\n", pending_vec);
7107 	}
7108 
7109 	kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
7110 	kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
7111 	kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
7112 	kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
7113 	kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
7114 	kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
7115 
7116 	kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
7117 	kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
7118 
7119 	update_cr8_intercept(vcpu);
7120 
7121 	/* Older userspace won't unhalt the vcpu on reset. */
7122 	if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
7123 	    sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
7124 	    !is_protmode(vcpu))
7125 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7126 
7127 	kvm_make_request(KVM_REQ_EVENT, vcpu);
7128 
7129 	return 0;
7130 }
7131 
7132 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
7133 					struct kvm_guest_debug *dbg)
7134 {
7135 	unsigned long rflags;
7136 	int i, r;
7137 
7138 	if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
7139 		r = -EBUSY;
7140 		if (vcpu->arch.exception.pending)
7141 			goto out;
7142 		if (dbg->control & KVM_GUESTDBG_INJECT_DB)
7143 			kvm_queue_exception(vcpu, DB_VECTOR);
7144 		else
7145 			kvm_queue_exception(vcpu, BP_VECTOR);
7146 	}
7147 
7148 	/*
7149 	 * Read rflags as long as potentially injected trace flags are still
7150 	 * filtered out.
7151 	 */
7152 	rflags = kvm_get_rflags(vcpu);
7153 
7154 	vcpu->guest_debug = dbg->control;
7155 	if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
7156 		vcpu->guest_debug = 0;
7157 
7158 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
7159 		for (i = 0; i < KVM_NR_DB_REGS; ++i)
7160 			vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
7161 		vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
7162 	} else {
7163 		for (i = 0; i < KVM_NR_DB_REGS; i++)
7164 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
7165 	}
7166 	kvm_update_dr7(vcpu);
7167 
7168 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
7169 		vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
7170 			get_segment_base(vcpu, VCPU_SREG_CS);
7171 
7172 	/*
7173 	 * Trigger an rflags update that will inject or remove the trace
7174 	 * flags.
7175 	 */
7176 	kvm_set_rflags(vcpu, rflags);
7177 
7178 	kvm_x86_ops->update_db_bp_intercept(vcpu);
7179 
7180 	r = 0;
7181 
7182 out:
7183 
7184 	return r;
7185 }
7186 
7187 /*
7188  * Translate a guest virtual address to a guest physical address.
7189  */
7190 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
7191 				    struct kvm_translation *tr)
7192 {
7193 	unsigned long vaddr = tr->linear_address;
7194 	gpa_t gpa;
7195 	int idx;
7196 
7197 	idx = srcu_read_lock(&vcpu->kvm->srcu);
7198 	gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
7199 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
7200 	tr->physical_address = gpa;
7201 	tr->valid = gpa != UNMAPPED_GVA;
7202 	tr->writeable = 1;
7203 	tr->usermode = 0;
7204 
7205 	return 0;
7206 }
7207 
7208 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
7209 {
7210 	struct fxregs_state *fxsave =
7211 			&vcpu->arch.guest_fpu.state.fxsave;
7212 
7213 	memcpy(fpu->fpr, fxsave->st_space, 128);
7214 	fpu->fcw = fxsave->cwd;
7215 	fpu->fsw = fxsave->swd;
7216 	fpu->ftwx = fxsave->twd;
7217 	fpu->last_opcode = fxsave->fop;
7218 	fpu->last_ip = fxsave->rip;
7219 	fpu->last_dp = fxsave->rdp;
7220 	memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
7221 
7222 	return 0;
7223 }
7224 
7225 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
7226 {
7227 	struct fxregs_state *fxsave =
7228 			&vcpu->arch.guest_fpu.state.fxsave;
7229 
7230 	memcpy(fxsave->st_space, fpu->fpr, 128);
7231 	fxsave->cwd = fpu->fcw;
7232 	fxsave->swd = fpu->fsw;
7233 	fxsave->twd = fpu->ftwx;
7234 	fxsave->fop = fpu->last_opcode;
7235 	fxsave->rip = fpu->last_ip;
7236 	fxsave->rdp = fpu->last_dp;
7237 	memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
7238 
7239 	return 0;
7240 }
7241 
7242 static void fx_init(struct kvm_vcpu *vcpu)
7243 {
7244 	fpstate_init(&vcpu->arch.guest_fpu.state);
7245 	if (cpu_has_xsaves)
7246 		vcpu->arch.guest_fpu.state.xsave.header.xcomp_bv =
7247 			host_xcr0 | XSTATE_COMPACTION_ENABLED;
7248 
7249 	/*
7250 	 * Ensure guest xcr0 is valid for loading
7251 	 */
7252 	vcpu->arch.xcr0 = XSTATE_FP;
7253 
7254 	vcpu->arch.cr0 |= X86_CR0_ET;
7255 }
7256 
7257 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
7258 {
7259 	if (vcpu->guest_fpu_loaded)
7260 		return;
7261 
7262 	/*
7263 	 * Restore all possible states in the guest,
7264 	 * and assume host would use all available bits.
7265 	 * Guest xcr0 would be loaded later.
7266 	 */
7267 	kvm_put_guest_xcr0(vcpu);
7268 	vcpu->guest_fpu_loaded = 1;
7269 	__kernel_fpu_begin();
7270 	__copy_kernel_to_fpregs(&vcpu->arch.guest_fpu.state);
7271 	trace_kvm_fpu(1);
7272 }
7273 
7274 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
7275 {
7276 	kvm_put_guest_xcr0(vcpu);
7277 
7278 	if (!vcpu->guest_fpu_loaded) {
7279 		vcpu->fpu_counter = 0;
7280 		return;
7281 	}
7282 
7283 	vcpu->guest_fpu_loaded = 0;
7284 	copy_fpregs_to_fpstate(&vcpu->arch.guest_fpu);
7285 	__kernel_fpu_end();
7286 	++vcpu->stat.fpu_reload;
7287 	/*
7288 	 * If using eager FPU mode, or if the guest is a frequent user
7289 	 * of the FPU, just leave the FPU active for next time.
7290 	 * Every 255 times fpu_counter rolls over to 0; a guest that uses
7291 	 * the FPU in bursts will revert to loading it on demand.
7292 	 */
7293 	if (!vcpu->arch.eager_fpu) {
7294 		if (++vcpu->fpu_counter < 5)
7295 			kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu);
7296 	}
7297 	trace_kvm_fpu(0);
7298 }
7299 
7300 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
7301 {
7302 	kvmclock_reset(vcpu);
7303 
7304 	free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
7305 	kvm_x86_ops->vcpu_free(vcpu);
7306 }
7307 
7308 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
7309 						unsigned int id)
7310 {
7311 	struct kvm_vcpu *vcpu;
7312 
7313 	if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
7314 		printk_once(KERN_WARNING
7315 		"kvm: SMP vm created on host with unstable TSC; "
7316 		"guest TSC will not be reliable\n");
7317 
7318 	vcpu = kvm_x86_ops->vcpu_create(kvm, id);
7319 
7320 	/*
7321 	 * Activate fpu unconditionally in case the guest needs eager FPU.  It will be
7322 	 * deactivated soon if it doesn't.
7323 	 */
7324 	kvm_x86_ops->fpu_activate(vcpu);
7325 	return vcpu;
7326 }
7327 
7328 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
7329 {
7330 	int r;
7331 
7332 	kvm_vcpu_mtrr_init(vcpu);
7333 	r = vcpu_load(vcpu);
7334 	if (r)
7335 		return r;
7336 	kvm_vcpu_reset(vcpu, false);
7337 	kvm_mmu_setup(vcpu);
7338 	vcpu_put(vcpu);
7339 	return r;
7340 }
7341 
7342 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
7343 {
7344 	struct msr_data msr;
7345 	struct kvm *kvm = vcpu->kvm;
7346 
7347 	if (vcpu_load(vcpu))
7348 		return;
7349 	msr.data = 0x0;
7350 	msr.index = MSR_IA32_TSC;
7351 	msr.host_initiated = true;
7352 	kvm_write_tsc(vcpu, &msr);
7353 	vcpu_put(vcpu);
7354 
7355 	if (!kvmclock_periodic_sync)
7356 		return;
7357 
7358 	schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
7359 					KVMCLOCK_SYNC_PERIOD);
7360 }
7361 
7362 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
7363 {
7364 	int r;
7365 	vcpu->arch.apf.msr_val = 0;
7366 
7367 	r = vcpu_load(vcpu);
7368 	BUG_ON(r);
7369 	kvm_mmu_unload(vcpu);
7370 	vcpu_put(vcpu);
7371 
7372 	kvm_x86_ops->vcpu_free(vcpu);
7373 }
7374 
7375 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
7376 {
7377 	vcpu->arch.hflags = 0;
7378 
7379 	atomic_set(&vcpu->arch.nmi_queued, 0);
7380 	vcpu->arch.nmi_pending = 0;
7381 	vcpu->arch.nmi_injected = false;
7382 	kvm_clear_interrupt_queue(vcpu);
7383 	kvm_clear_exception_queue(vcpu);
7384 
7385 	memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
7386 	kvm_update_dr0123(vcpu);
7387 	vcpu->arch.dr6 = DR6_INIT;
7388 	kvm_update_dr6(vcpu);
7389 	vcpu->arch.dr7 = DR7_FIXED_1;
7390 	kvm_update_dr7(vcpu);
7391 
7392 	vcpu->arch.cr2 = 0;
7393 
7394 	kvm_make_request(KVM_REQ_EVENT, vcpu);
7395 	vcpu->arch.apf.msr_val = 0;
7396 	vcpu->arch.st.msr_val = 0;
7397 
7398 	kvmclock_reset(vcpu);
7399 
7400 	kvm_clear_async_pf_completion_queue(vcpu);
7401 	kvm_async_pf_hash_reset(vcpu);
7402 	vcpu->arch.apf.halted = false;
7403 
7404 	if (!init_event) {
7405 		kvm_pmu_reset(vcpu);
7406 		vcpu->arch.smbase = 0x30000;
7407 	}
7408 
7409 	memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
7410 	vcpu->arch.regs_avail = ~0;
7411 	vcpu->arch.regs_dirty = ~0;
7412 
7413 	kvm_x86_ops->vcpu_reset(vcpu, init_event);
7414 }
7415 
7416 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
7417 {
7418 	struct kvm_segment cs;
7419 
7420 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
7421 	cs.selector = vector << 8;
7422 	cs.base = vector << 12;
7423 	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
7424 	kvm_rip_write(vcpu, 0);
7425 }
7426 
7427 int kvm_arch_hardware_enable(void)
7428 {
7429 	struct kvm *kvm;
7430 	struct kvm_vcpu *vcpu;
7431 	int i;
7432 	int ret;
7433 	u64 local_tsc;
7434 	u64 max_tsc = 0;
7435 	bool stable, backwards_tsc = false;
7436 
7437 	kvm_shared_msr_cpu_online();
7438 	ret = kvm_x86_ops->hardware_enable();
7439 	if (ret != 0)
7440 		return ret;
7441 
7442 	local_tsc = native_read_tsc();
7443 	stable = !check_tsc_unstable();
7444 	list_for_each_entry(kvm, &vm_list, vm_list) {
7445 		kvm_for_each_vcpu(i, vcpu, kvm) {
7446 			if (!stable && vcpu->cpu == smp_processor_id())
7447 				kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7448 			if (stable && vcpu->arch.last_host_tsc > local_tsc) {
7449 				backwards_tsc = true;
7450 				if (vcpu->arch.last_host_tsc > max_tsc)
7451 					max_tsc = vcpu->arch.last_host_tsc;
7452 			}
7453 		}
7454 	}
7455 
7456 	/*
7457 	 * Sometimes, even reliable TSCs go backwards.  This happens on
7458 	 * platforms that reset TSC during suspend or hibernate actions, but
7459 	 * maintain synchronization.  We must compensate.  Fortunately, we can
7460 	 * detect that condition here, which happens early in CPU bringup,
7461 	 * before any KVM threads can be running.  Unfortunately, we can't
7462 	 * bring the TSCs fully up to date with real time, as we aren't yet far
7463 	 * enough into CPU bringup that we know how much real time has actually
7464 	 * elapsed; our helper function, get_kernel_ns() will be using boot
7465 	 * variables that haven't been updated yet.
7466 	 *
7467 	 * So we simply find the maximum observed TSC above, then record the
7468 	 * adjustment to TSC in each VCPU.  When the VCPU later gets loaded,
7469 	 * the adjustment will be applied.  Note that we accumulate
7470 	 * adjustments, in case multiple suspend cycles happen before some VCPU
7471 	 * gets a chance to run again.  In the event that no KVM threads get a
7472 	 * chance to run, we will miss the entire elapsed period, as we'll have
7473 	 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
7474 	 * loose cycle time.  This isn't too big a deal, since the loss will be
7475 	 * uniform across all VCPUs (not to mention the scenario is extremely
7476 	 * unlikely). It is possible that a second hibernate recovery happens
7477 	 * much faster than a first, causing the observed TSC here to be
7478 	 * smaller; this would require additional padding adjustment, which is
7479 	 * why we set last_host_tsc to the local tsc observed here.
7480 	 *
7481 	 * N.B. - this code below runs only on platforms with reliable TSC,
7482 	 * as that is the only way backwards_tsc is set above.  Also note
7483 	 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
7484 	 * have the same delta_cyc adjustment applied if backwards_tsc
7485 	 * is detected.  Note further, this adjustment is only done once,
7486 	 * as we reset last_host_tsc on all VCPUs to stop this from being
7487 	 * called multiple times (one for each physical CPU bringup).
7488 	 *
7489 	 * Platforms with unreliable TSCs don't have to deal with this, they
7490 	 * will be compensated by the logic in vcpu_load, which sets the TSC to
7491 	 * catchup mode.  This will catchup all VCPUs to real time, but cannot
7492 	 * guarantee that they stay in perfect synchronization.
7493 	 */
7494 	if (backwards_tsc) {
7495 		u64 delta_cyc = max_tsc - local_tsc;
7496 		backwards_tsc_observed = true;
7497 		list_for_each_entry(kvm, &vm_list, vm_list) {
7498 			kvm_for_each_vcpu(i, vcpu, kvm) {
7499 				vcpu->arch.tsc_offset_adjustment += delta_cyc;
7500 				vcpu->arch.last_host_tsc = local_tsc;
7501 				kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
7502 			}
7503 
7504 			/*
7505 			 * We have to disable TSC offset matching.. if you were
7506 			 * booting a VM while issuing an S4 host suspend....
7507 			 * you may have some problem.  Solving this issue is
7508 			 * left as an exercise to the reader.
7509 			 */
7510 			kvm->arch.last_tsc_nsec = 0;
7511 			kvm->arch.last_tsc_write = 0;
7512 		}
7513 
7514 	}
7515 	return 0;
7516 }
7517 
7518 void kvm_arch_hardware_disable(void)
7519 {
7520 	kvm_x86_ops->hardware_disable();
7521 	drop_user_return_notifiers();
7522 }
7523 
7524 int kvm_arch_hardware_setup(void)
7525 {
7526 	int r;
7527 
7528 	r = kvm_x86_ops->hardware_setup();
7529 	if (r != 0)
7530 		return r;
7531 
7532 	kvm_init_msr_list();
7533 	return 0;
7534 }
7535 
7536 void kvm_arch_hardware_unsetup(void)
7537 {
7538 	kvm_x86_ops->hardware_unsetup();
7539 }
7540 
7541 void kvm_arch_check_processor_compat(void *rtn)
7542 {
7543 	kvm_x86_ops->check_processor_compatibility(rtn);
7544 }
7545 
7546 bool kvm_vcpu_compatible(struct kvm_vcpu *vcpu)
7547 {
7548 	return irqchip_in_kernel(vcpu->kvm) == (vcpu->arch.apic != NULL);
7549 }
7550 
7551 struct static_key kvm_no_apic_vcpu __read_mostly;
7552 
7553 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
7554 {
7555 	struct page *page;
7556 	struct kvm *kvm;
7557 	int r;
7558 
7559 	BUG_ON(vcpu->kvm == NULL);
7560 	kvm = vcpu->kvm;
7561 
7562 	vcpu->arch.pv.pv_unhalted = false;
7563 	vcpu->arch.emulate_ctxt.ops = &emulate_ops;
7564 	if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_reset_bsp(vcpu))
7565 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7566 	else
7567 		vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
7568 
7569 	page = alloc_page(GFP_KERNEL | __GFP_ZERO);
7570 	if (!page) {
7571 		r = -ENOMEM;
7572 		goto fail;
7573 	}
7574 	vcpu->arch.pio_data = page_address(page);
7575 
7576 	kvm_set_tsc_khz(vcpu, max_tsc_khz);
7577 
7578 	r = kvm_mmu_create(vcpu);
7579 	if (r < 0)
7580 		goto fail_free_pio_data;
7581 
7582 	if (irqchip_in_kernel(kvm)) {
7583 		r = kvm_create_lapic(vcpu);
7584 		if (r < 0)
7585 			goto fail_mmu_destroy;
7586 	} else
7587 		static_key_slow_inc(&kvm_no_apic_vcpu);
7588 
7589 	vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
7590 				       GFP_KERNEL);
7591 	if (!vcpu->arch.mce_banks) {
7592 		r = -ENOMEM;
7593 		goto fail_free_lapic;
7594 	}
7595 	vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
7596 
7597 	if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL)) {
7598 		r = -ENOMEM;
7599 		goto fail_free_mce_banks;
7600 	}
7601 
7602 	fx_init(vcpu);
7603 
7604 	vcpu->arch.ia32_tsc_adjust_msr = 0x0;
7605 	vcpu->arch.pv_time_enabled = false;
7606 
7607 	vcpu->arch.guest_supported_xcr0 = 0;
7608 	vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
7609 
7610 	vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
7611 
7612 	vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
7613 
7614 	kvm_async_pf_hash_reset(vcpu);
7615 	kvm_pmu_init(vcpu);
7616 
7617 	return 0;
7618 
7619 fail_free_mce_banks:
7620 	kfree(vcpu->arch.mce_banks);
7621 fail_free_lapic:
7622 	kvm_free_lapic(vcpu);
7623 fail_mmu_destroy:
7624 	kvm_mmu_destroy(vcpu);
7625 fail_free_pio_data:
7626 	free_page((unsigned long)vcpu->arch.pio_data);
7627 fail:
7628 	return r;
7629 }
7630 
7631 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
7632 {
7633 	int idx;
7634 
7635 	kvm_pmu_destroy(vcpu);
7636 	kfree(vcpu->arch.mce_banks);
7637 	kvm_free_lapic(vcpu);
7638 	idx = srcu_read_lock(&vcpu->kvm->srcu);
7639 	kvm_mmu_destroy(vcpu);
7640 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
7641 	free_page((unsigned long)vcpu->arch.pio_data);
7642 	if (!irqchip_in_kernel(vcpu->kvm))
7643 		static_key_slow_dec(&kvm_no_apic_vcpu);
7644 }
7645 
7646 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
7647 {
7648 	kvm_x86_ops->sched_in(vcpu, cpu);
7649 }
7650 
7651 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
7652 {
7653 	if (type)
7654 		return -EINVAL;
7655 
7656 	INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
7657 	INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
7658 	INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
7659 	INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
7660 	atomic_set(&kvm->arch.noncoherent_dma_count, 0);
7661 
7662 	/* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
7663 	set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
7664 	/* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
7665 	set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
7666 		&kvm->arch.irq_sources_bitmap);
7667 
7668 	raw_spin_lock_init(&kvm->arch.tsc_write_lock);
7669 	mutex_init(&kvm->arch.apic_map_lock);
7670 	spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
7671 
7672 	pvclock_update_vm_gtod_copy(kvm);
7673 
7674 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
7675 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
7676 
7677 	return 0;
7678 }
7679 
7680 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
7681 {
7682 	int r;
7683 	r = vcpu_load(vcpu);
7684 	BUG_ON(r);
7685 	kvm_mmu_unload(vcpu);
7686 	vcpu_put(vcpu);
7687 }
7688 
7689 static void kvm_free_vcpus(struct kvm *kvm)
7690 {
7691 	unsigned int i;
7692 	struct kvm_vcpu *vcpu;
7693 
7694 	/*
7695 	 * Unpin any mmu pages first.
7696 	 */
7697 	kvm_for_each_vcpu(i, vcpu, kvm) {
7698 		kvm_clear_async_pf_completion_queue(vcpu);
7699 		kvm_unload_vcpu_mmu(vcpu);
7700 	}
7701 	kvm_for_each_vcpu(i, vcpu, kvm)
7702 		kvm_arch_vcpu_free(vcpu);
7703 
7704 	mutex_lock(&kvm->lock);
7705 	for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
7706 		kvm->vcpus[i] = NULL;
7707 
7708 	atomic_set(&kvm->online_vcpus, 0);
7709 	mutex_unlock(&kvm->lock);
7710 }
7711 
7712 void kvm_arch_sync_events(struct kvm *kvm)
7713 {
7714 	cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
7715 	cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
7716 	kvm_free_all_assigned_devices(kvm);
7717 	kvm_free_pit(kvm);
7718 }
7719 
7720 int __x86_set_memory_region(struct kvm *kvm,
7721 			    const struct kvm_userspace_memory_region *mem)
7722 {
7723 	int i, r;
7724 
7725 	/* Called with kvm->slots_lock held.  */
7726 	BUG_ON(mem->slot >= KVM_MEM_SLOTS_NUM);
7727 
7728 	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
7729 		struct kvm_userspace_memory_region m = *mem;
7730 
7731 		m.slot |= i << 16;
7732 		r = __kvm_set_memory_region(kvm, &m);
7733 		if (r < 0)
7734 			return r;
7735 	}
7736 
7737 	return 0;
7738 }
7739 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
7740 
7741 int x86_set_memory_region(struct kvm *kvm,
7742 			  const struct kvm_userspace_memory_region *mem)
7743 {
7744 	int r;
7745 
7746 	mutex_lock(&kvm->slots_lock);
7747 	r = __x86_set_memory_region(kvm, mem);
7748 	mutex_unlock(&kvm->slots_lock);
7749 
7750 	return r;
7751 }
7752 EXPORT_SYMBOL_GPL(x86_set_memory_region);
7753 
7754 void kvm_arch_destroy_vm(struct kvm *kvm)
7755 {
7756 	if (current->mm == kvm->mm) {
7757 		/*
7758 		 * Free memory regions allocated on behalf of userspace,
7759 		 * unless the the memory map has changed due to process exit
7760 		 * or fd copying.
7761 		 */
7762 		struct kvm_userspace_memory_region mem;
7763 		memset(&mem, 0, sizeof(mem));
7764 		mem.slot = APIC_ACCESS_PAGE_PRIVATE_MEMSLOT;
7765 		x86_set_memory_region(kvm, &mem);
7766 
7767 		mem.slot = IDENTITY_PAGETABLE_PRIVATE_MEMSLOT;
7768 		x86_set_memory_region(kvm, &mem);
7769 
7770 		mem.slot = TSS_PRIVATE_MEMSLOT;
7771 		x86_set_memory_region(kvm, &mem);
7772 	}
7773 	kvm_iommu_unmap_guest(kvm);
7774 	kfree(kvm->arch.vpic);
7775 	kfree(kvm->arch.vioapic);
7776 	kvm_free_vcpus(kvm);
7777 	kfree(rcu_dereference_check(kvm->arch.apic_map, 1));
7778 }
7779 
7780 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
7781 			   struct kvm_memory_slot *dont)
7782 {
7783 	int i;
7784 
7785 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7786 		if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
7787 			kvfree(free->arch.rmap[i]);
7788 			free->arch.rmap[i] = NULL;
7789 		}
7790 		if (i == 0)
7791 			continue;
7792 
7793 		if (!dont || free->arch.lpage_info[i - 1] !=
7794 			     dont->arch.lpage_info[i - 1]) {
7795 			kvfree(free->arch.lpage_info[i - 1]);
7796 			free->arch.lpage_info[i - 1] = NULL;
7797 		}
7798 	}
7799 }
7800 
7801 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
7802 			    unsigned long npages)
7803 {
7804 	int i;
7805 
7806 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7807 		unsigned long ugfn;
7808 		int lpages;
7809 		int level = i + 1;
7810 
7811 		lpages = gfn_to_index(slot->base_gfn + npages - 1,
7812 				      slot->base_gfn, level) + 1;
7813 
7814 		slot->arch.rmap[i] =
7815 			kvm_kvzalloc(lpages * sizeof(*slot->arch.rmap[i]));
7816 		if (!slot->arch.rmap[i])
7817 			goto out_free;
7818 		if (i == 0)
7819 			continue;
7820 
7821 		slot->arch.lpage_info[i - 1] = kvm_kvzalloc(lpages *
7822 					sizeof(*slot->arch.lpage_info[i - 1]));
7823 		if (!slot->arch.lpage_info[i - 1])
7824 			goto out_free;
7825 
7826 		if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
7827 			slot->arch.lpage_info[i - 1][0].write_count = 1;
7828 		if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
7829 			slot->arch.lpage_info[i - 1][lpages - 1].write_count = 1;
7830 		ugfn = slot->userspace_addr >> PAGE_SHIFT;
7831 		/*
7832 		 * If the gfn and userspace address are not aligned wrt each
7833 		 * other, or if explicitly asked to, disable large page
7834 		 * support for this slot
7835 		 */
7836 		if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
7837 		    !kvm_largepages_enabled()) {
7838 			unsigned long j;
7839 
7840 			for (j = 0; j < lpages; ++j)
7841 				slot->arch.lpage_info[i - 1][j].write_count = 1;
7842 		}
7843 	}
7844 
7845 	return 0;
7846 
7847 out_free:
7848 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7849 		kvfree(slot->arch.rmap[i]);
7850 		slot->arch.rmap[i] = NULL;
7851 		if (i == 0)
7852 			continue;
7853 
7854 		kvfree(slot->arch.lpage_info[i - 1]);
7855 		slot->arch.lpage_info[i - 1] = NULL;
7856 	}
7857 	return -ENOMEM;
7858 }
7859 
7860 void kvm_arch_memslots_updated(struct kvm *kvm, struct kvm_memslots *slots)
7861 {
7862 	/*
7863 	 * memslots->generation has been incremented.
7864 	 * mmio generation may have reached its maximum value.
7865 	 */
7866 	kvm_mmu_invalidate_mmio_sptes(kvm, slots);
7867 }
7868 
7869 int kvm_arch_prepare_memory_region(struct kvm *kvm,
7870 				struct kvm_memory_slot *memslot,
7871 				const struct kvm_userspace_memory_region *mem,
7872 				enum kvm_mr_change change)
7873 {
7874 	/*
7875 	 * Only private memory slots need to be mapped here since
7876 	 * KVM_SET_MEMORY_REGION ioctl is no longer supported.
7877 	 */
7878 	if ((memslot->id >= KVM_USER_MEM_SLOTS) && (change == KVM_MR_CREATE)) {
7879 		unsigned long userspace_addr;
7880 
7881 		/*
7882 		 * MAP_SHARED to prevent internal slot pages from being moved
7883 		 * by fork()/COW.
7884 		 */
7885 		userspace_addr = vm_mmap(NULL, 0, memslot->npages * PAGE_SIZE,
7886 					 PROT_READ | PROT_WRITE,
7887 					 MAP_SHARED | MAP_ANONYMOUS, 0);
7888 
7889 		if (IS_ERR((void *)userspace_addr))
7890 			return PTR_ERR((void *)userspace_addr);
7891 
7892 		memslot->userspace_addr = userspace_addr;
7893 	}
7894 
7895 	return 0;
7896 }
7897 
7898 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
7899 				     struct kvm_memory_slot *new)
7900 {
7901 	/* Still write protect RO slot */
7902 	if (new->flags & KVM_MEM_READONLY) {
7903 		kvm_mmu_slot_remove_write_access(kvm, new);
7904 		return;
7905 	}
7906 
7907 	/*
7908 	 * Call kvm_x86_ops dirty logging hooks when they are valid.
7909 	 *
7910 	 * kvm_x86_ops->slot_disable_log_dirty is called when:
7911 	 *
7912 	 *  - KVM_MR_CREATE with dirty logging is disabled
7913 	 *  - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
7914 	 *
7915 	 * The reason is, in case of PML, we need to set D-bit for any slots
7916 	 * with dirty logging disabled in order to eliminate unnecessary GPA
7917 	 * logging in PML buffer (and potential PML buffer full VMEXT). This
7918 	 * guarantees leaving PML enabled during guest's lifetime won't have
7919 	 * any additonal overhead from PML when guest is running with dirty
7920 	 * logging disabled for memory slots.
7921 	 *
7922 	 * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
7923 	 * to dirty logging mode.
7924 	 *
7925 	 * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
7926 	 *
7927 	 * In case of write protect:
7928 	 *
7929 	 * Write protect all pages for dirty logging.
7930 	 *
7931 	 * All the sptes including the large sptes which point to this
7932 	 * slot are set to readonly. We can not create any new large
7933 	 * spte on this slot until the end of the logging.
7934 	 *
7935 	 * See the comments in fast_page_fault().
7936 	 */
7937 	if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
7938 		if (kvm_x86_ops->slot_enable_log_dirty)
7939 			kvm_x86_ops->slot_enable_log_dirty(kvm, new);
7940 		else
7941 			kvm_mmu_slot_remove_write_access(kvm, new);
7942 	} else {
7943 		if (kvm_x86_ops->slot_disable_log_dirty)
7944 			kvm_x86_ops->slot_disable_log_dirty(kvm, new);
7945 	}
7946 }
7947 
7948 void kvm_arch_commit_memory_region(struct kvm *kvm,
7949 				const struct kvm_userspace_memory_region *mem,
7950 				const struct kvm_memory_slot *old,
7951 				const struct kvm_memory_slot *new,
7952 				enum kvm_mr_change change)
7953 {
7954 	int nr_mmu_pages = 0;
7955 
7956 	if (change == KVM_MR_DELETE && old->id >= KVM_USER_MEM_SLOTS) {
7957 		int ret;
7958 
7959 		ret = vm_munmap(old->userspace_addr,
7960 				old->npages * PAGE_SIZE);
7961 		if (ret < 0)
7962 			printk(KERN_WARNING
7963 			       "kvm_vm_ioctl_set_memory_region: "
7964 			       "failed to munmap memory\n");
7965 	}
7966 
7967 	if (!kvm->arch.n_requested_mmu_pages)
7968 		nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
7969 
7970 	if (nr_mmu_pages)
7971 		kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
7972 
7973 	/*
7974 	 * Dirty logging tracks sptes in 4k granularity, meaning that large
7975 	 * sptes have to be split.  If live migration is successful, the guest
7976 	 * in the source machine will be destroyed and large sptes will be
7977 	 * created in the destination. However, if the guest continues to run
7978 	 * in the source machine (for example if live migration fails), small
7979 	 * sptes will remain around and cause bad performance.
7980 	 *
7981 	 * Scan sptes if dirty logging has been stopped, dropping those
7982 	 * which can be collapsed into a single large-page spte.  Later
7983 	 * page faults will create the large-page sptes.
7984 	 */
7985 	if ((change != KVM_MR_DELETE) &&
7986 		(old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
7987 		!(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
7988 		kvm_mmu_zap_collapsible_sptes(kvm, new);
7989 
7990 	/*
7991 	 * Set up write protection and/or dirty logging for the new slot.
7992 	 *
7993 	 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
7994 	 * been zapped so no dirty logging staff is needed for old slot. For
7995 	 * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
7996 	 * new and it's also covered when dealing with the new slot.
7997 	 *
7998 	 * FIXME: const-ify all uses of struct kvm_memory_slot.
7999 	 */
8000 	if (change != KVM_MR_DELETE)
8001 		kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new);
8002 }
8003 
8004 void kvm_arch_flush_shadow_all(struct kvm *kvm)
8005 {
8006 	kvm_mmu_invalidate_zap_all_pages(kvm);
8007 }
8008 
8009 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
8010 				   struct kvm_memory_slot *slot)
8011 {
8012 	kvm_mmu_invalidate_zap_all_pages(kvm);
8013 }
8014 
8015 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
8016 {
8017 	if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
8018 		kvm_x86_ops->check_nested_events(vcpu, false);
8019 
8020 	return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
8021 		!vcpu->arch.apf.halted)
8022 		|| !list_empty_careful(&vcpu->async_pf.done)
8023 		|| kvm_apic_has_events(vcpu)
8024 		|| vcpu->arch.pv.pv_unhalted
8025 		|| atomic_read(&vcpu->arch.nmi_queued) ||
8026 		(kvm_arch_interrupt_allowed(vcpu) &&
8027 		 kvm_cpu_has_interrupt(vcpu));
8028 }
8029 
8030 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
8031 {
8032 	return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
8033 }
8034 
8035 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
8036 {
8037 	return kvm_x86_ops->interrupt_allowed(vcpu);
8038 }
8039 
8040 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
8041 {
8042 	if (is_64_bit_mode(vcpu))
8043 		return kvm_rip_read(vcpu);
8044 	return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
8045 		     kvm_rip_read(vcpu));
8046 }
8047 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
8048 
8049 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
8050 {
8051 	return kvm_get_linear_rip(vcpu) == linear_rip;
8052 }
8053 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
8054 
8055 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
8056 {
8057 	unsigned long rflags;
8058 
8059 	rflags = kvm_x86_ops->get_rflags(vcpu);
8060 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
8061 		rflags &= ~X86_EFLAGS_TF;
8062 	return rflags;
8063 }
8064 EXPORT_SYMBOL_GPL(kvm_get_rflags);
8065 
8066 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
8067 {
8068 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
8069 	    kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
8070 		rflags |= X86_EFLAGS_TF;
8071 	kvm_x86_ops->set_rflags(vcpu, rflags);
8072 }
8073 
8074 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
8075 {
8076 	__kvm_set_rflags(vcpu, rflags);
8077 	kvm_make_request(KVM_REQ_EVENT, vcpu);
8078 }
8079 EXPORT_SYMBOL_GPL(kvm_set_rflags);
8080 
8081 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
8082 {
8083 	int r;
8084 
8085 	if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
8086 	      work->wakeup_all)
8087 		return;
8088 
8089 	r = kvm_mmu_reload(vcpu);
8090 	if (unlikely(r))
8091 		return;
8092 
8093 	if (!vcpu->arch.mmu.direct_map &&
8094 	      work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
8095 		return;
8096 
8097 	vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
8098 }
8099 
8100 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
8101 {
8102 	return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
8103 }
8104 
8105 static inline u32 kvm_async_pf_next_probe(u32 key)
8106 {
8107 	return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
8108 }
8109 
8110 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8111 {
8112 	u32 key = kvm_async_pf_hash_fn(gfn);
8113 
8114 	while (vcpu->arch.apf.gfns[key] != ~0)
8115 		key = kvm_async_pf_next_probe(key);
8116 
8117 	vcpu->arch.apf.gfns[key] = gfn;
8118 }
8119 
8120 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
8121 {
8122 	int i;
8123 	u32 key = kvm_async_pf_hash_fn(gfn);
8124 
8125 	for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
8126 		     (vcpu->arch.apf.gfns[key] != gfn &&
8127 		      vcpu->arch.apf.gfns[key] != ~0); i++)
8128 		key = kvm_async_pf_next_probe(key);
8129 
8130 	return key;
8131 }
8132 
8133 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8134 {
8135 	return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
8136 }
8137 
8138 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8139 {
8140 	u32 i, j, k;
8141 
8142 	i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
8143 	while (true) {
8144 		vcpu->arch.apf.gfns[i] = ~0;
8145 		do {
8146 			j = kvm_async_pf_next_probe(j);
8147 			if (vcpu->arch.apf.gfns[j] == ~0)
8148 				return;
8149 			k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
8150 			/*
8151 			 * k lies cyclically in ]i,j]
8152 			 * |    i.k.j |
8153 			 * |....j i.k.| or  |.k..j i...|
8154 			 */
8155 		} while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
8156 		vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
8157 		i = j;
8158 	}
8159 }
8160 
8161 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
8162 {
8163 
8164 	return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
8165 				      sizeof(val));
8166 }
8167 
8168 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
8169 				     struct kvm_async_pf *work)
8170 {
8171 	struct x86_exception fault;
8172 
8173 	trace_kvm_async_pf_not_present(work->arch.token, work->gva);
8174 	kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
8175 
8176 	if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
8177 	    (vcpu->arch.apf.send_user_only &&
8178 	     kvm_x86_ops->get_cpl(vcpu) == 0))
8179 		kvm_make_request(KVM_REQ_APF_HALT, vcpu);
8180 	else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
8181 		fault.vector = PF_VECTOR;
8182 		fault.error_code_valid = true;
8183 		fault.error_code = 0;
8184 		fault.nested_page_fault = false;
8185 		fault.address = work->arch.token;
8186 		kvm_inject_page_fault(vcpu, &fault);
8187 	}
8188 }
8189 
8190 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
8191 				 struct kvm_async_pf *work)
8192 {
8193 	struct x86_exception fault;
8194 
8195 	trace_kvm_async_pf_ready(work->arch.token, work->gva);
8196 	if (work->wakeup_all)
8197 		work->arch.token = ~0; /* broadcast wakeup */
8198 	else
8199 		kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
8200 
8201 	if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
8202 	    !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
8203 		fault.vector = PF_VECTOR;
8204 		fault.error_code_valid = true;
8205 		fault.error_code = 0;
8206 		fault.nested_page_fault = false;
8207 		fault.address = work->arch.token;
8208 		kvm_inject_page_fault(vcpu, &fault);
8209 	}
8210 	vcpu->arch.apf.halted = false;
8211 	vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
8212 }
8213 
8214 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
8215 {
8216 	if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
8217 		return true;
8218 	else
8219 		return !kvm_event_needs_reinjection(vcpu) &&
8220 			kvm_x86_ops->interrupt_allowed(vcpu);
8221 }
8222 
8223 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
8224 {
8225 	atomic_inc(&kvm->arch.noncoherent_dma_count);
8226 }
8227 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
8228 
8229 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
8230 {
8231 	atomic_dec(&kvm->arch.noncoherent_dma_count);
8232 }
8233 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
8234 
8235 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
8236 {
8237 	return atomic_read(&kvm->arch.noncoherent_dma_count);
8238 }
8239 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
8240 
8241 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
8242 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
8243 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
8244 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
8245 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
8246 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
8247 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
8248 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
8249 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
8250 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
8251 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
8252 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
8253 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
8254 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window);
8255 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
8256