xref: /openbmc/linux/arch/x86/kvm/x86.c (revision 275876e2)
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 
31 #include <linux/clocksource.h>
32 #include <linux/interrupt.h>
33 #include <linux/kvm.h>
34 #include <linux/fs.h>
35 #include <linux/vmalloc.h>
36 #include <linux/module.h>
37 #include <linux/mman.h>
38 #include <linux/highmem.h>
39 #include <linux/iommu.h>
40 #include <linux/intel-iommu.h>
41 #include <linux/cpufreq.h>
42 #include <linux/user-return-notifier.h>
43 #include <linux/srcu.h>
44 #include <linux/slab.h>
45 #include <linux/perf_event.h>
46 #include <linux/uaccess.h>
47 #include <linux/hash.h>
48 #include <linux/pci.h>
49 #include <linux/timekeeper_internal.h>
50 #include <linux/pvclock_gtod.h>
51 #include <trace/events/kvm.h>
52 
53 #define CREATE_TRACE_POINTS
54 #include "trace.h"
55 
56 #include <asm/debugreg.h>
57 #include <asm/msr.h>
58 #include <asm/desc.h>
59 #include <asm/mtrr.h>
60 #include <asm/mce.h>
61 #include <asm/i387.h>
62 #include <asm/fpu-internal.h> /* Ugh! */
63 #include <asm/xcr.h>
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 bool kvm_has_tsc_control;
102 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
103 u32  kvm_max_guest_tsc_khz;
104 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
105 
106 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
107 static u32 tsc_tolerance_ppm = 250;
108 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
109 
110 static bool backwards_tsc_observed = false;
111 
112 #define KVM_NR_SHARED_MSRS 16
113 
114 struct kvm_shared_msrs_global {
115 	int nr;
116 	u32 msrs[KVM_NR_SHARED_MSRS];
117 };
118 
119 struct kvm_shared_msrs {
120 	struct user_return_notifier urn;
121 	bool registered;
122 	struct kvm_shared_msr_values {
123 		u64 host;
124 		u64 curr;
125 	} values[KVM_NR_SHARED_MSRS];
126 };
127 
128 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
129 static struct kvm_shared_msrs __percpu *shared_msrs;
130 
131 struct kvm_stats_debugfs_item debugfs_entries[] = {
132 	{ "pf_fixed", VCPU_STAT(pf_fixed) },
133 	{ "pf_guest", VCPU_STAT(pf_guest) },
134 	{ "tlb_flush", VCPU_STAT(tlb_flush) },
135 	{ "invlpg", VCPU_STAT(invlpg) },
136 	{ "exits", VCPU_STAT(exits) },
137 	{ "io_exits", VCPU_STAT(io_exits) },
138 	{ "mmio_exits", VCPU_STAT(mmio_exits) },
139 	{ "signal_exits", VCPU_STAT(signal_exits) },
140 	{ "irq_window", VCPU_STAT(irq_window_exits) },
141 	{ "nmi_window", VCPU_STAT(nmi_window_exits) },
142 	{ "halt_exits", VCPU_STAT(halt_exits) },
143 	{ "halt_wakeup", VCPU_STAT(halt_wakeup) },
144 	{ "hypercalls", VCPU_STAT(hypercalls) },
145 	{ "request_irq", VCPU_STAT(request_irq_exits) },
146 	{ "irq_exits", VCPU_STAT(irq_exits) },
147 	{ "host_state_reload", VCPU_STAT(host_state_reload) },
148 	{ "efer_reload", VCPU_STAT(efer_reload) },
149 	{ "fpu_reload", VCPU_STAT(fpu_reload) },
150 	{ "insn_emulation", VCPU_STAT(insn_emulation) },
151 	{ "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
152 	{ "irq_injections", VCPU_STAT(irq_injections) },
153 	{ "nmi_injections", VCPU_STAT(nmi_injections) },
154 	{ "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
155 	{ "mmu_pte_write", VM_STAT(mmu_pte_write) },
156 	{ "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
157 	{ "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
158 	{ "mmu_flooded", VM_STAT(mmu_flooded) },
159 	{ "mmu_recycled", VM_STAT(mmu_recycled) },
160 	{ "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
161 	{ "mmu_unsync", VM_STAT(mmu_unsync) },
162 	{ "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
163 	{ "largepages", VM_STAT(lpages) },
164 	{ NULL }
165 };
166 
167 u64 __read_mostly host_xcr0;
168 
169 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
170 
171 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
172 {
173 	int i;
174 	for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
175 		vcpu->arch.apf.gfns[i] = ~0;
176 }
177 
178 static void kvm_on_user_return(struct user_return_notifier *urn)
179 {
180 	unsigned slot;
181 	struct kvm_shared_msrs *locals
182 		= container_of(urn, struct kvm_shared_msrs, urn);
183 	struct kvm_shared_msr_values *values;
184 
185 	for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
186 		values = &locals->values[slot];
187 		if (values->host != values->curr) {
188 			wrmsrl(shared_msrs_global.msrs[slot], values->host);
189 			values->curr = values->host;
190 		}
191 	}
192 	locals->registered = false;
193 	user_return_notifier_unregister(urn);
194 }
195 
196 static void shared_msr_update(unsigned slot, u32 msr)
197 {
198 	u64 value;
199 	unsigned int cpu = smp_processor_id();
200 	struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
201 
202 	/* only read, and nobody should modify it at this time,
203 	 * so don't need lock */
204 	if (slot >= shared_msrs_global.nr) {
205 		printk(KERN_ERR "kvm: invalid MSR slot!");
206 		return;
207 	}
208 	rdmsrl_safe(msr, &value);
209 	smsr->values[slot].host = value;
210 	smsr->values[slot].curr = value;
211 }
212 
213 void kvm_define_shared_msr(unsigned slot, u32 msr)
214 {
215 	BUG_ON(slot >= KVM_NR_SHARED_MSRS);
216 	if (slot >= shared_msrs_global.nr)
217 		shared_msrs_global.nr = slot + 1;
218 	shared_msrs_global.msrs[slot] = msr;
219 	/* we need ensured the shared_msr_global have been updated */
220 	smp_wmb();
221 }
222 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
223 
224 static void kvm_shared_msr_cpu_online(void)
225 {
226 	unsigned i;
227 
228 	for (i = 0; i < shared_msrs_global.nr; ++i)
229 		shared_msr_update(i, shared_msrs_global.msrs[i]);
230 }
231 
232 void kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
233 {
234 	unsigned int cpu = smp_processor_id();
235 	struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
236 
237 	if (((value ^ smsr->values[slot].curr) & mask) == 0)
238 		return;
239 	smsr->values[slot].curr = value;
240 	wrmsrl(shared_msrs_global.msrs[slot], value);
241 	if (!smsr->registered) {
242 		smsr->urn.on_user_return = kvm_on_user_return;
243 		user_return_notifier_register(&smsr->urn);
244 		smsr->registered = true;
245 	}
246 }
247 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
248 
249 static void drop_user_return_notifiers(void *ignore)
250 {
251 	unsigned int cpu = smp_processor_id();
252 	struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
253 
254 	if (smsr->registered)
255 		kvm_on_user_return(&smsr->urn);
256 }
257 
258 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
259 {
260 	return vcpu->arch.apic_base;
261 }
262 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
263 
264 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
265 {
266 	u64 old_state = vcpu->arch.apic_base &
267 		(MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
268 	u64 new_state = msr_info->data &
269 		(MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
270 	u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) |
271 		0x2ff | (guest_cpuid_has_x2apic(vcpu) ? 0 : X2APIC_ENABLE);
272 
273 	if (!msr_info->host_initiated &&
274 	    ((msr_info->data & reserved_bits) != 0 ||
275 	     new_state == X2APIC_ENABLE ||
276 	     (new_state == MSR_IA32_APICBASE_ENABLE &&
277 	      old_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE)) ||
278 	     (new_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE) &&
279 	      old_state == 0)))
280 		return 1;
281 
282 	kvm_lapic_set_base(vcpu, msr_info->data);
283 	return 0;
284 }
285 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
286 
287 asmlinkage __visible void kvm_spurious_fault(void)
288 {
289 	/* Fault while not rebooting.  We want the trace. */
290 	BUG();
291 }
292 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
293 
294 #define EXCPT_BENIGN		0
295 #define EXCPT_CONTRIBUTORY	1
296 #define EXCPT_PF		2
297 
298 static int exception_class(int vector)
299 {
300 	switch (vector) {
301 	case PF_VECTOR:
302 		return EXCPT_PF;
303 	case DE_VECTOR:
304 	case TS_VECTOR:
305 	case NP_VECTOR:
306 	case SS_VECTOR:
307 	case GP_VECTOR:
308 		return EXCPT_CONTRIBUTORY;
309 	default:
310 		break;
311 	}
312 	return EXCPT_BENIGN;
313 }
314 
315 #define EXCPT_FAULT		0
316 #define EXCPT_TRAP		1
317 #define EXCPT_ABORT		2
318 #define EXCPT_INTERRUPT		3
319 
320 static int exception_type(int vector)
321 {
322 	unsigned int mask;
323 
324 	if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
325 		return EXCPT_INTERRUPT;
326 
327 	mask = 1 << vector;
328 
329 	/* #DB is trap, as instruction watchpoints are handled elsewhere */
330 	if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
331 		return EXCPT_TRAP;
332 
333 	if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
334 		return EXCPT_ABORT;
335 
336 	/* Reserved exceptions will result in fault */
337 	return EXCPT_FAULT;
338 }
339 
340 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
341 		unsigned nr, bool has_error, u32 error_code,
342 		bool reinject)
343 {
344 	u32 prev_nr;
345 	int class1, class2;
346 
347 	kvm_make_request(KVM_REQ_EVENT, vcpu);
348 
349 	if (!vcpu->arch.exception.pending) {
350 	queue:
351 		vcpu->arch.exception.pending = true;
352 		vcpu->arch.exception.has_error_code = has_error;
353 		vcpu->arch.exception.nr = nr;
354 		vcpu->arch.exception.error_code = error_code;
355 		vcpu->arch.exception.reinject = reinject;
356 		return;
357 	}
358 
359 	/* to check exception */
360 	prev_nr = vcpu->arch.exception.nr;
361 	if (prev_nr == DF_VECTOR) {
362 		/* triple fault -> shutdown */
363 		kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
364 		return;
365 	}
366 	class1 = exception_class(prev_nr);
367 	class2 = exception_class(nr);
368 	if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
369 		|| (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
370 		/* generate double fault per SDM Table 5-5 */
371 		vcpu->arch.exception.pending = true;
372 		vcpu->arch.exception.has_error_code = true;
373 		vcpu->arch.exception.nr = DF_VECTOR;
374 		vcpu->arch.exception.error_code = 0;
375 	} else
376 		/* replace previous exception with a new one in a hope
377 		   that instruction re-execution will regenerate lost
378 		   exception */
379 		goto queue;
380 }
381 
382 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
383 {
384 	kvm_multiple_exception(vcpu, nr, false, 0, false);
385 }
386 EXPORT_SYMBOL_GPL(kvm_queue_exception);
387 
388 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
389 {
390 	kvm_multiple_exception(vcpu, nr, false, 0, true);
391 }
392 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
393 
394 void kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
395 {
396 	if (err)
397 		kvm_inject_gp(vcpu, 0);
398 	else
399 		kvm_x86_ops->skip_emulated_instruction(vcpu);
400 }
401 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
402 
403 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
404 {
405 	++vcpu->stat.pf_guest;
406 	vcpu->arch.cr2 = fault->address;
407 	kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
408 }
409 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
410 
411 void kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
412 {
413 	if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
414 		vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
415 	else
416 		vcpu->arch.mmu.inject_page_fault(vcpu, fault);
417 }
418 
419 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
420 {
421 	atomic_inc(&vcpu->arch.nmi_queued);
422 	kvm_make_request(KVM_REQ_NMI, vcpu);
423 }
424 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
425 
426 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
427 {
428 	kvm_multiple_exception(vcpu, nr, true, error_code, false);
429 }
430 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
431 
432 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
433 {
434 	kvm_multiple_exception(vcpu, nr, true, error_code, true);
435 }
436 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
437 
438 /*
439  * Checks if cpl <= required_cpl; if true, return true.  Otherwise queue
440  * a #GP and return false.
441  */
442 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
443 {
444 	if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
445 		return true;
446 	kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
447 	return false;
448 }
449 EXPORT_SYMBOL_GPL(kvm_require_cpl);
450 
451 /*
452  * This function will be used to read from the physical memory of the currently
453  * running guest. The difference to kvm_read_guest_page is that this function
454  * can read from guest physical or from the guest's guest physical memory.
455  */
456 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
457 			    gfn_t ngfn, void *data, int offset, int len,
458 			    u32 access)
459 {
460 	gfn_t real_gfn;
461 	gpa_t ngpa;
462 
463 	ngpa     = gfn_to_gpa(ngfn);
464 	real_gfn = mmu->translate_gpa(vcpu, ngpa, access);
465 	if (real_gfn == UNMAPPED_GVA)
466 		return -EFAULT;
467 
468 	real_gfn = gpa_to_gfn(real_gfn);
469 
470 	return kvm_read_guest_page(vcpu->kvm, real_gfn, data, offset, len);
471 }
472 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
473 
474 int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
475 			       void *data, int offset, int len, u32 access)
476 {
477 	return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
478 				       data, offset, len, access);
479 }
480 
481 /*
482  * Load the pae pdptrs.  Return true is they are all valid.
483  */
484 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
485 {
486 	gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
487 	unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
488 	int i;
489 	int ret;
490 	u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
491 
492 	ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
493 				      offset * sizeof(u64), sizeof(pdpte),
494 				      PFERR_USER_MASK|PFERR_WRITE_MASK);
495 	if (ret < 0) {
496 		ret = 0;
497 		goto out;
498 	}
499 	for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
500 		if (is_present_gpte(pdpte[i]) &&
501 		    (pdpte[i] & vcpu->arch.mmu.rsvd_bits_mask[0][2])) {
502 			ret = 0;
503 			goto out;
504 		}
505 	}
506 	ret = 1;
507 
508 	memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
509 	__set_bit(VCPU_EXREG_PDPTR,
510 		  (unsigned long *)&vcpu->arch.regs_avail);
511 	__set_bit(VCPU_EXREG_PDPTR,
512 		  (unsigned long *)&vcpu->arch.regs_dirty);
513 out:
514 
515 	return ret;
516 }
517 EXPORT_SYMBOL_GPL(load_pdptrs);
518 
519 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
520 {
521 	u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
522 	bool changed = true;
523 	int offset;
524 	gfn_t gfn;
525 	int r;
526 
527 	if (is_long_mode(vcpu) || !is_pae(vcpu))
528 		return false;
529 
530 	if (!test_bit(VCPU_EXREG_PDPTR,
531 		      (unsigned long *)&vcpu->arch.regs_avail))
532 		return true;
533 
534 	gfn = (kvm_read_cr3(vcpu) & ~31u) >> PAGE_SHIFT;
535 	offset = (kvm_read_cr3(vcpu) & ~31u) & (PAGE_SIZE - 1);
536 	r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
537 				       PFERR_USER_MASK | PFERR_WRITE_MASK);
538 	if (r < 0)
539 		goto out;
540 	changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
541 out:
542 
543 	return changed;
544 }
545 
546 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
547 {
548 	unsigned long old_cr0 = kvm_read_cr0(vcpu);
549 	unsigned long update_bits = X86_CR0_PG | X86_CR0_WP |
550 				    X86_CR0_CD | X86_CR0_NW;
551 
552 	cr0 |= X86_CR0_ET;
553 
554 #ifdef CONFIG_X86_64
555 	if (cr0 & 0xffffffff00000000UL)
556 		return 1;
557 #endif
558 
559 	cr0 &= ~CR0_RESERVED_BITS;
560 
561 	if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
562 		return 1;
563 
564 	if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
565 		return 1;
566 
567 	if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
568 #ifdef CONFIG_X86_64
569 		if ((vcpu->arch.efer & EFER_LME)) {
570 			int cs_db, cs_l;
571 
572 			if (!is_pae(vcpu))
573 				return 1;
574 			kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
575 			if (cs_l)
576 				return 1;
577 		} else
578 #endif
579 		if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
580 						 kvm_read_cr3(vcpu)))
581 			return 1;
582 	}
583 
584 	if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
585 		return 1;
586 
587 	kvm_x86_ops->set_cr0(vcpu, cr0);
588 
589 	if ((cr0 ^ old_cr0) & X86_CR0_PG) {
590 		kvm_clear_async_pf_completion_queue(vcpu);
591 		kvm_async_pf_hash_reset(vcpu);
592 	}
593 
594 	if ((cr0 ^ old_cr0) & update_bits)
595 		kvm_mmu_reset_context(vcpu);
596 	return 0;
597 }
598 EXPORT_SYMBOL_GPL(kvm_set_cr0);
599 
600 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
601 {
602 	(void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
603 }
604 EXPORT_SYMBOL_GPL(kvm_lmsw);
605 
606 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
607 {
608 	if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
609 			!vcpu->guest_xcr0_loaded) {
610 		/* kvm_set_xcr() also depends on this */
611 		xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
612 		vcpu->guest_xcr0_loaded = 1;
613 	}
614 }
615 
616 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
617 {
618 	if (vcpu->guest_xcr0_loaded) {
619 		if (vcpu->arch.xcr0 != host_xcr0)
620 			xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
621 		vcpu->guest_xcr0_loaded = 0;
622 	}
623 }
624 
625 int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
626 {
627 	u64 xcr0 = xcr;
628 	u64 old_xcr0 = vcpu->arch.xcr0;
629 	u64 valid_bits;
630 
631 	/* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now  */
632 	if (index != XCR_XFEATURE_ENABLED_MASK)
633 		return 1;
634 	if (!(xcr0 & XSTATE_FP))
635 		return 1;
636 	if ((xcr0 & XSTATE_YMM) && !(xcr0 & XSTATE_SSE))
637 		return 1;
638 
639 	/*
640 	 * Do not allow the guest to set bits that we do not support
641 	 * saving.  However, xcr0 bit 0 is always set, even if the
642 	 * emulated CPU does not support XSAVE (see fx_init).
643 	 */
644 	valid_bits = vcpu->arch.guest_supported_xcr0 | XSTATE_FP;
645 	if (xcr0 & ~valid_bits)
646 		return 1;
647 
648 	if ((!(xcr0 & XSTATE_BNDREGS)) != (!(xcr0 & XSTATE_BNDCSR)))
649 		return 1;
650 
651 	kvm_put_guest_xcr0(vcpu);
652 	vcpu->arch.xcr0 = xcr0;
653 
654 	if ((xcr0 ^ old_xcr0) & XSTATE_EXTEND_MASK)
655 		kvm_update_cpuid(vcpu);
656 	return 0;
657 }
658 
659 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
660 {
661 	if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
662 	    __kvm_set_xcr(vcpu, index, xcr)) {
663 		kvm_inject_gp(vcpu, 0);
664 		return 1;
665 	}
666 	return 0;
667 }
668 EXPORT_SYMBOL_GPL(kvm_set_xcr);
669 
670 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
671 {
672 	unsigned long old_cr4 = kvm_read_cr4(vcpu);
673 	unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE |
674 				   X86_CR4_PAE | X86_CR4_SMEP;
675 	if (cr4 & CR4_RESERVED_BITS)
676 		return 1;
677 
678 	if (!guest_cpuid_has_xsave(vcpu) && (cr4 & X86_CR4_OSXSAVE))
679 		return 1;
680 
681 	if (!guest_cpuid_has_smep(vcpu) && (cr4 & X86_CR4_SMEP))
682 		return 1;
683 
684 	if (!guest_cpuid_has_smap(vcpu) && (cr4 & X86_CR4_SMAP))
685 		return 1;
686 
687 	if (!guest_cpuid_has_fsgsbase(vcpu) && (cr4 & X86_CR4_FSGSBASE))
688 		return 1;
689 
690 	if (is_long_mode(vcpu)) {
691 		if (!(cr4 & X86_CR4_PAE))
692 			return 1;
693 	} else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
694 		   && ((cr4 ^ old_cr4) & pdptr_bits)
695 		   && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
696 				   kvm_read_cr3(vcpu)))
697 		return 1;
698 
699 	if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
700 		if (!guest_cpuid_has_pcid(vcpu))
701 			return 1;
702 
703 		/* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
704 		if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
705 			return 1;
706 	}
707 
708 	if (kvm_x86_ops->set_cr4(vcpu, cr4))
709 		return 1;
710 
711 	if (((cr4 ^ old_cr4) & pdptr_bits) ||
712 	    (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
713 		kvm_mmu_reset_context(vcpu);
714 
715 	if ((cr4 ^ old_cr4) & X86_CR4_SMAP)
716 		update_permission_bitmask(vcpu, vcpu->arch.walk_mmu, false);
717 
718 	if ((cr4 ^ old_cr4) & X86_CR4_OSXSAVE)
719 		kvm_update_cpuid(vcpu);
720 
721 	return 0;
722 }
723 EXPORT_SYMBOL_GPL(kvm_set_cr4);
724 
725 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
726 {
727 	if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
728 		kvm_mmu_sync_roots(vcpu);
729 		kvm_mmu_flush_tlb(vcpu);
730 		return 0;
731 	}
732 
733 	if (is_long_mode(vcpu)) {
734 		if (cr3 & CR3_L_MODE_RESERVED_BITS)
735 			return 1;
736 	} else if (is_pae(vcpu) && is_paging(vcpu) &&
737 		   !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
738 		return 1;
739 
740 	vcpu->arch.cr3 = cr3;
741 	__set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
742 	kvm_mmu_new_cr3(vcpu);
743 	return 0;
744 }
745 EXPORT_SYMBOL_GPL(kvm_set_cr3);
746 
747 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
748 {
749 	if (cr8 & CR8_RESERVED_BITS)
750 		return 1;
751 	if (irqchip_in_kernel(vcpu->kvm))
752 		kvm_lapic_set_tpr(vcpu, cr8);
753 	else
754 		vcpu->arch.cr8 = cr8;
755 	return 0;
756 }
757 EXPORT_SYMBOL_GPL(kvm_set_cr8);
758 
759 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
760 {
761 	if (irqchip_in_kernel(vcpu->kvm))
762 		return kvm_lapic_get_cr8(vcpu);
763 	else
764 		return vcpu->arch.cr8;
765 }
766 EXPORT_SYMBOL_GPL(kvm_get_cr8);
767 
768 static void kvm_update_dr6(struct kvm_vcpu *vcpu)
769 {
770 	if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
771 		kvm_x86_ops->set_dr6(vcpu, vcpu->arch.dr6);
772 }
773 
774 static void kvm_update_dr7(struct kvm_vcpu *vcpu)
775 {
776 	unsigned long dr7;
777 
778 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
779 		dr7 = vcpu->arch.guest_debug_dr7;
780 	else
781 		dr7 = vcpu->arch.dr7;
782 	kvm_x86_ops->set_dr7(vcpu, dr7);
783 	vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
784 	if (dr7 & DR7_BP_EN_MASK)
785 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
786 }
787 
788 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
789 {
790 	u64 fixed = DR6_FIXED_1;
791 
792 	if (!guest_cpuid_has_rtm(vcpu))
793 		fixed |= DR6_RTM;
794 	return fixed;
795 }
796 
797 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
798 {
799 	switch (dr) {
800 	case 0 ... 3:
801 		vcpu->arch.db[dr] = val;
802 		if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
803 			vcpu->arch.eff_db[dr] = val;
804 		break;
805 	case 4:
806 		if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
807 			return 1; /* #UD */
808 		/* fall through */
809 	case 6:
810 		if (val & 0xffffffff00000000ULL)
811 			return -1; /* #GP */
812 		vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
813 		kvm_update_dr6(vcpu);
814 		break;
815 	case 5:
816 		if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
817 			return 1; /* #UD */
818 		/* fall through */
819 	default: /* 7 */
820 		if (val & 0xffffffff00000000ULL)
821 			return -1; /* #GP */
822 		vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
823 		kvm_update_dr7(vcpu);
824 		break;
825 	}
826 
827 	return 0;
828 }
829 
830 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
831 {
832 	int res;
833 
834 	res = __kvm_set_dr(vcpu, dr, val);
835 	if (res > 0)
836 		kvm_queue_exception(vcpu, UD_VECTOR);
837 	else if (res < 0)
838 		kvm_inject_gp(vcpu, 0);
839 
840 	return res;
841 }
842 EXPORT_SYMBOL_GPL(kvm_set_dr);
843 
844 static int _kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
845 {
846 	switch (dr) {
847 	case 0 ... 3:
848 		*val = vcpu->arch.db[dr];
849 		break;
850 	case 4:
851 		if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
852 			return 1;
853 		/* fall through */
854 	case 6:
855 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
856 			*val = vcpu->arch.dr6;
857 		else
858 			*val = kvm_x86_ops->get_dr6(vcpu);
859 		break;
860 	case 5:
861 		if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
862 			return 1;
863 		/* fall through */
864 	default: /* 7 */
865 		*val = vcpu->arch.dr7;
866 		break;
867 	}
868 
869 	return 0;
870 }
871 
872 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
873 {
874 	if (_kvm_get_dr(vcpu, dr, val)) {
875 		kvm_queue_exception(vcpu, UD_VECTOR);
876 		return 1;
877 	}
878 	return 0;
879 }
880 EXPORT_SYMBOL_GPL(kvm_get_dr);
881 
882 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
883 {
884 	u32 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
885 	u64 data;
886 	int err;
887 
888 	err = kvm_pmu_read_pmc(vcpu, ecx, &data);
889 	if (err)
890 		return err;
891 	kvm_register_write(vcpu, VCPU_REGS_RAX, (u32)data);
892 	kvm_register_write(vcpu, VCPU_REGS_RDX, data >> 32);
893 	return err;
894 }
895 EXPORT_SYMBOL_GPL(kvm_rdpmc);
896 
897 /*
898  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
899  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
900  *
901  * This list is modified at module load time to reflect the
902  * capabilities of the host cpu. This capabilities test skips MSRs that are
903  * kvm-specific. Those are put in the beginning of the list.
904  */
905 
906 #define KVM_SAVE_MSRS_BEGIN	12
907 static u32 msrs_to_save[] = {
908 	MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
909 	MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
910 	HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
911 	HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
912 	HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
913 	MSR_KVM_PV_EOI_EN,
914 	MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
915 	MSR_STAR,
916 #ifdef CONFIG_X86_64
917 	MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
918 #endif
919 	MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
920 	MSR_IA32_FEATURE_CONTROL, MSR_IA32_BNDCFGS
921 };
922 
923 static unsigned num_msrs_to_save;
924 
925 static const u32 emulated_msrs[] = {
926 	MSR_IA32_TSC_ADJUST,
927 	MSR_IA32_TSCDEADLINE,
928 	MSR_IA32_MISC_ENABLE,
929 	MSR_IA32_MCG_STATUS,
930 	MSR_IA32_MCG_CTL,
931 };
932 
933 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
934 {
935 	if (efer & efer_reserved_bits)
936 		return false;
937 
938 	if (efer & EFER_FFXSR) {
939 		struct kvm_cpuid_entry2 *feat;
940 
941 		feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
942 		if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
943 			return false;
944 	}
945 
946 	if (efer & EFER_SVME) {
947 		struct kvm_cpuid_entry2 *feat;
948 
949 		feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
950 		if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
951 			return false;
952 	}
953 
954 	return true;
955 }
956 EXPORT_SYMBOL_GPL(kvm_valid_efer);
957 
958 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
959 {
960 	u64 old_efer = vcpu->arch.efer;
961 
962 	if (!kvm_valid_efer(vcpu, efer))
963 		return 1;
964 
965 	if (is_paging(vcpu)
966 	    && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
967 		return 1;
968 
969 	efer &= ~EFER_LMA;
970 	efer |= vcpu->arch.efer & EFER_LMA;
971 
972 	kvm_x86_ops->set_efer(vcpu, efer);
973 
974 	/* Update reserved bits */
975 	if ((efer ^ old_efer) & EFER_NX)
976 		kvm_mmu_reset_context(vcpu);
977 
978 	return 0;
979 }
980 
981 void kvm_enable_efer_bits(u64 mask)
982 {
983        efer_reserved_bits &= ~mask;
984 }
985 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
986 
987 
988 /*
989  * Writes msr value into into the appropriate "register".
990  * Returns 0 on success, non-0 otherwise.
991  * Assumes vcpu_load() was already called.
992  */
993 int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
994 {
995 	return kvm_x86_ops->set_msr(vcpu, msr);
996 }
997 
998 /*
999  * Adapt set_msr() to msr_io()'s calling convention
1000  */
1001 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1002 {
1003 	struct msr_data msr;
1004 
1005 	msr.data = *data;
1006 	msr.index = index;
1007 	msr.host_initiated = true;
1008 	return kvm_set_msr(vcpu, &msr);
1009 }
1010 
1011 #ifdef CONFIG_X86_64
1012 struct pvclock_gtod_data {
1013 	seqcount_t	seq;
1014 
1015 	struct { /* extract of a clocksource struct */
1016 		int vclock_mode;
1017 		cycle_t	cycle_last;
1018 		cycle_t	mask;
1019 		u32	mult;
1020 		u32	shift;
1021 	} clock;
1022 
1023 	u64		boot_ns;
1024 	u64		nsec_base;
1025 };
1026 
1027 static struct pvclock_gtod_data pvclock_gtod_data;
1028 
1029 static void update_pvclock_gtod(struct timekeeper *tk)
1030 {
1031 	struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1032 	u64 boot_ns;
1033 
1034 	boot_ns = ktime_to_ns(ktime_add(tk->tkr.base_mono, tk->offs_boot));
1035 
1036 	write_seqcount_begin(&vdata->seq);
1037 
1038 	/* copy pvclock gtod data */
1039 	vdata->clock.vclock_mode	= tk->tkr.clock->archdata.vclock_mode;
1040 	vdata->clock.cycle_last		= tk->tkr.cycle_last;
1041 	vdata->clock.mask		= tk->tkr.mask;
1042 	vdata->clock.mult		= tk->tkr.mult;
1043 	vdata->clock.shift		= tk->tkr.shift;
1044 
1045 	vdata->boot_ns			= boot_ns;
1046 	vdata->nsec_base		= tk->tkr.xtime_nsec;
1047 
1048 	write_seqcount_end(&vdata->seq);
1049 }
1050 #endif
1051 
1052 
1053 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1054 {
1055 	int version;
1056 	int r;
1057 	struct pvclock_wall_clock wc;
1058 	struct timespec boot;
1059 
1060 	if (!wall_clock)
1061 		return;
1062 
1063 	r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1064 	if (r)
1065 		return;
1066 
1067 	if (version & 1)
1068 		++version;  /* first time write, random junk */
1069 
1070 	++version;
1071 
1072 	kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1073 
1074 	/*
1075 	 * The guest calculates current wall clock time by adding
1076 	 * system time (updated by kvm_guest_time_update below) to the
1077 	 * wall clock specified here.  guest system time equals host
1078 	 * system time for us, thus we must fill in host boot time here.
1079 	 */
1080 	getboottime(&boot);
1081 
1082 	if (kvm->arch.kvmclock_offset) {
1083 		struct timespec ts = ns_to_timespec(kvm->arch.kvmclock_offset);
1084 		boot = timespec_sub(boot, ts);
1085 	}
1086 	wc.sec = boot.tv_sec;
1087 	wc.nsec = boot.tv_nsec;
1088 	wc.version = version;
1089 
1090 	kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1091 
1092 	version++;
1093 	kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1094 }
1095 
1096 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1097 {
1098 	uint32_t quotient, remainder;
1099 
1100 	/* Don't try to replace with do_div(), this one calculates
1101 	 * "(dividend << 32) / divisor" */
1102 	__asm__ ( "divl %4"
1103 		  : "=a" (quotient), "=d" (remainder)
1104 		  : "0" (0), "1" (dividend), "r" (divisor) );
1105 	return quotient;
1106 }
1107 
1108 static void kvm_get_time_scale(uint32_t scaled_khz, uint32_t base_khz,
1109 			       s8 *pshift, u32 *pmultiplier)
1110 {
1111 	uint64_t scaled64;
1112 	int32_t  shift = 0;
1113 	uint64_t tps64;
1114 	uint32_t tps32;
1115 
1116 	tps64 = base_khz * 1000LL;
1117 	scaled64 = scaled_khz * 1000LL;
1118 	while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1119 		tps64 >>= 1;
1120 		shift--;
1121 	}
1122 
1123 	tps32 = (uint32_t)tps64;
1124 	while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1125 		if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1126 			scaled64 >>= 1;
1127 		else
1128 			tps32 <<= 1;
1129 		shift++;
1130 	}
1131 
1132 	*pshift = shift;
1133 	*pmultiplier = div_frac(scaled64, tps32);
1134 
1135 	pr_debug("%s: base_khz %u => %u, shift %d, mul %u\n",
1136 		 __func__, base_khz, scaled_khz, shift, *pmultiplier);
1137 }
1138 
1139 static inline u64 get_kernel_ns(void)
1140 {
1141 	return ktime_get_boot_ns();
1142 }
1143 
1144 #ifdef CONFIG_X86_64
1145 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1146 #endif
1147 
1148 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1149 unsigned long max_tsc_khz;
1150 
1151 static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec)
1152 {
1153 	return pvclock_scale_delta(nsec, vcpu->arch.virtual_tsc_mult,
1154 				   vcpu->arch.virtual_tsc_shift);
1155 }
1156 
1157 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1158 {
1159 	u64 v = (u64)khz * (1000000 + ppm);
1160 	do_div(v, 1000000);
1161 	return v;
1162 }
1163 
1164 static void kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 this_tsc_khz)
1165 {
1166 	u32 thresh_lo, thresh_hi;
1167 	int use_scaling = 0;
1168 
1169 	/* tsc_khz can be zero if TSC calibration fails */
1170 	if (this_tsc_khz == 0)
1171 		return;
1172 
1173 	/* Compute a scale to convert nanoseconds in TSC cycles */
1174 	kvm_get_time_scale(this_tsc_khz, NSEC_PER_SEC / 1000,
1175 			   &vcpu->arch.virtual_tsc_shift,
1176 			   &vcpu->arch.virtual_tsc_mult);
1177 	vcpu->arch.virtual_tsc_khz = this_tsc_khz;
1178 
1179 	/*
1180 	 * Compute the variation in TSC rate which is acceptable
1181 	 * within the range of tolerance and decide if the
1182 	 * rate being applied is within that bounds of the hardware
1183 	 * rate.  If so, no scaling or compensation need be done.
1184 	 */
1185 	thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1186 	thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1187 	if (this_tsc_khz < thresh_lo || this_tsc_khz > thresh_hi) {
1188 		pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", this_tsc_khz, thresh_lo, thresh_hi);
1189 		use_scaling = 1;
1190 	}
1191 	kvm_x86_ops->set_tsc_khz(vcpu, this_tsc_khz, use_scaling);
1192 }
1193 
1194 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1195 {
1196 	u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1197 				      vcpu->arch.virtual_tsc_mult,
1198 				      vcpu->arch.virtual_tsc_shift);
1199 	tsc += vcpu->arch.this_tsc_write;
1200 	return tsc;
1201 }
1202 
1203 void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1204 {
1205 #ifdef CONFIG_X86_64
1206 	bool vcpus_matched;
1207 	bool do_request = false;
1208 	struct kvm_arch *ka = &vcpu->kvm->arch;
1209 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1210 
1211 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1212 			 atomic_read(&vcpu->kvm->online_vcpus));
1213 
1214 	if (vcpus_matched && gtod->clock.vclock_mode == VCLOCK_TSC)
1215 		if (!ka->use_master_clock)
1216 			do_request = 1;
1217 
1218 	if (!vcpus_matched && ka->use_master_clock)
1219 			do_request = 1;
1220 
1221 	if (do_request)
1222 		kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1223 
1224 	trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1225 			    atomic_read(&vcpu->kvm->online_vcpus),
1226 		            ka->use_master_clock, gtod->clock.vclock_mode);
1227 #endif
1228 }
1229 
1230 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1231 {
1232 	u64 curr_offset = kvm_x86_ops->read_tsc_offset(vcpu);
1233 	vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1234 }
1235 
1236 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1237 {
1238 	struct kvm *kvm = vcpu->kvm;
1239 	u64 offset, ns, elapsed;
1240 	unsigned long flags;
1241 	s64 usdiff;
1242 	bool matched;
1243 	bool already_matched;
1244 	u64 data = msr->data;
1245 
1246 	raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1247 	offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
1248 	ns = get_kernel_ns();
1249 	elapsed = ns - kvm->arch.last_tsc_nsec;
1250 
1251 	if (vcpu->arch.virtual_tsc_khz) {
1252 		int faulted = 0;
1253 
1254 		/* n.b - signed multiplication and division required */
1255 		usdiff = data - kvm->arch.last_tsc_write;
1256 #ifdef CONFIG_X86_64
1257 		usdiff = (usdiff * 1000) / vcpu->arch.virtual_tsc_khz;
1258 #else
1259 		/* do_div() only does unsigned */
1260 		asm("1: idivl %[divisor]\n"
1261 		    "2: xor %%edx, %%edx\n"
1262 		    "   movl $0, %[faulted]\n"
1263 		    "3:\n"
1264 		    ".section .fixup,\"ax\"\n"
1265 		    "4: movl $1, %[faulted]\n"
1266 		    "   jmp  3b\n"
1267 		    ".previous\n"
1268 
1269 		_ASM_EXTABLE(1b, 4b)
1270 
1271 		: "=A"(usdiff), [faulted] "=r" (faulted)
1272 		: "A"(usdiff * 1000), [divisor] "rm"(vcpu->arch.virtual_tsc_khz));
1273 
1274 #endif
1275 		do_div(elapsed, 1000);
1276 		usdiff -= elapsed;
1277 		if (usdiff < 0)
1278 			usdiff = -usdiff;
1279 
1280 		/* idivl overflow => difference is larger than USEC_PER_SEC */
1281 		if (faulted)
1282 			usdiff = USEC_PER_SEC;
1283 	} else
1284 		usdiff = USEC_PER_SEC; /* disable TSC match window below */
1285 
1286 	/*
1287 	 * Special case: TSC write with a small delta (1 second) of virtual
1288 	 * cycle time against real time is interpreted as an attempt to
1289 	 * synchronize the CPU.
1290          *
1291 	 * For a reliable TSC, we can match TSC offsets, and for an unstable
1292 	 * TSC, we add elapsed time in this computation.  We could let the
1293 	 * compensation code attempt to catch up if we fall behind, but
1294 	 * it's better to try to match offsets from the beginning.
1295          */
1296 	if (usdiff < USEC_PER_SEC &&
1297 	    vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1298 		if (!check_tsc_unstable()) {
1299 			offset = kvm->arch.cur_tsc_offset;
1300 			pr_debug("kvm: matched tsc offset for %llu\n", data);
1301 		} else {
1302 			u64 delta = nsec_to_cycles(vcpu, elapsed);
1303 			data += delta;
1304 			offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
1305 			pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1306 		}
1307 		matched = true;
1308 		already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
1309 	} else {
1310 		/*
1311 		 * We split periods of matched TSC writes into generations.
1312 		 * For each generation, we track the original measured
1313 		 * nanosecond time, offset, and write, so if TSCs are in
1314 		 * sync, we can match exact offset, and if not, we can match
1315 		 * exact software computation in compute_guest_tsc()
1316 		 *
1317 		 * These values are tracked in kvm->arch.cur_xxx variables.
1318 		 */
1319 		kvm->arch.cur_tsc_generation++;
1320 		kvm->arch.cur_tsc_nsec = ns;
1321 		kvm->arch.cur_tsc_write = data;
1322 		kvm->arch.cur_tsc_offset = offset;
1323 		matched = false;
1324 		pr_debug("kvm: new tsc generation %llu, clock %llu\n",
1325 			 kvm->arch.cur_tsc_generation, data);
1326 	}
1327 
1328 	/*
1329 	 * We also track th most recent recorded KHZ, write and time to
1330 	 * allow the matching interval to be extended at each write.
1331 	 */
1332 	kvm->arch.last_tsc_nsec = ns;
1333 	kvm->arch.last_tsc_write = data;
1334 	kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1335 
1336 	vcpu->arch.last_guest_tsc = data;
1337 
1338 	/* Keep track of which generation this VCPU has synchronized to */
1339 	vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1340 	vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1341 	vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1342 
1343 	if (guest_cpuid_has_tsc_adjust(vcpu) && !msr->host_initiated)
1344 		update_ia32_tsc_adjust_msr(vcpu, offset);
1345 	kvm_x86_ops->write_tsc_offset(vcpu, offset);
1346 	raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1347 
1348 	spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1349 	if (!matched) {
1350 		kvm->arch.nr_vcpus_matched_tsc = 0;
1351 	} else if (!already_matched) {
1352 		kvm->arch.nr_vcpus_matched_tsc++;
1353 	}
1354 
1355 	kvm_track_tsc_matching(vcpu);
1356 	spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
1357 }
1358 
1359 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1360 
1361 #ifdef CONFIG_X86_64
1362 
1363 static cycle_t read_tsc(void)
1364 {
1365 	cycle_t ret;
1366 	u64 last;
1367 
1368 	/*
1369 	 * Empirically, a fence (of type that depends on the CPU)
1370 	 * before rdtsc is enough to ensure that rdtsc is ordered
1371 	 * with respect to loads.  The various CPU manuals are unclear
1372 	 * as to whether rdtsc can be reordered with later loads,
1373 	 * but no one has ever seen it happen.
1374 	 */
1375 	rdtsc_barrier();
1376 	ret = (cycle_t)vget_cycles();
1377 
1378 	last = pvclock_gtod_data.clock.cycle_last;
1379 
1380 	if (likely(ret >= last))
1381 		return ret;
1382 
1383 	/*
1384 	 * GCC likes to generate cmov here, but this branch is extremely
1385 	 * predictable (it's just a funciton of time and the likely is
1386 	 * very likely) and there's a data dependence, so force GCC
1387 	 * to generate a branch instead.  I don't barrier() because
1388 	 * we don't actually need a barrier, and if this function
1389 	 * ever gets inlined it will generate worse code.
1390 	 */
1391 	asm volatile ("");
1392 	return last;
1393 }
1394 
1395 static inline u64 vgettsc(cycle_t *cycle_now)
1396 {
1397 	long v;
1398 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1399 
1400 	*cycle_now = read_tsc();
1401 
1402 	v = (*cycle_now - gtod->clock.cycle_last) & gtod->clock.mask;
1403 	return v * gtod->clock.mult;
1404 }
1405 
1406 static int do_monotonic_boot(s64 *t, cycle_t *cycle_now)
1407 {
1408 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1409 	unsigned long seq;
1410 	int mode;
1411 	u64 ns;
1412 
1413 	do {
1414 		seq = read_seqcount_begin(&gtod->seq);
1415 		mode = gtod->clock.vclock_mode;
1416 		ns = gtod->nsec_base;
1417 		ns += vgettsc(cycle_now);
1418 		ns >>= gtod->clock.shift;
1419 		ns += gtod->boot_ns;
1420 	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
1421 	*t = ns;
1422 
1423 	return mode;
1424 }
1425 
1426 /* returns true if host is using tsc clocksource */
1427 static bool kvm_get_time_and_clockread(s64 *kernel_ns, cycle_t *cycle_now)
1428 {
1429 	/* checked again under seqlock below */
1430 	if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC)
1431 		return false;
1432 
1433 	return do_monotonic_boot(kernel_ns, cycle_now) == VCLOCK_TSC;
1434 }
1435 #endif
1436 
1437 /*
1438  *
1439  * Assuming a stable TSC across physical CPUS, and a stable TSC
1440  * across virtual CPUs, the following condition is possible.
1441  * Each numbered line represents an event visible to both
1442  * CPUs at the next numbered event.
1443  *
1444  * "timespecX" represents host monotonic time. "tscX" represents
1445  * RDTSC value.
1446  *
1447  * 		VCPU0 on CPU0		|	VCPU1 on CPU1
1448  *
1449  * 1.  read timespec0,tsc0
1450  * 2.					| timespec1 = timespec0 + N
1451  * 					| tsc1 = tsc0 + M
1452  * 3. transition to guest		| transition to guest
1453  * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1454  * 5.				        | ret1 = timespec1 + (rdtsc - tsc1)
1455  * 				        | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1456  *
1457  * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1458  *
1459  * 	- ret0 < ret1
1460  *	- timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1461  *		...
1462  *	- 0 < N - M => M < N
1463  *
1464  * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1465  * always the case (the difference between two distinct xtime instances
1466  * might be smaller then the difference between corresponding TSC reads,
1467  * when updating guest vcpus pvclock areas).
1468  *
1469  * To avoid that problem, do not allow visibility of distinct
1470  * system_timestamp/tsc_timestamp values simultaneously: use a master
1471  * copy of host monotonic time values. Update that master copy
1472  * in lockstep.
1473  *
1474  * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
1475  *
1476  */
1477 
1478 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
1479 {
1480 #ifdef CONFIG_X86_64
1481 	struct kvm_arch *ka = &kvm->arch;
1482 	int vclock_mode;
1483 	bool host_tsc_clocksource, vcpus_matched;
1484 
1485 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1486 			atomic_read(&kvm->online_vcpus));
1487 
1488 	/*
1489 	 * If the host uses TSC clock, then passthrough TSC as stable
1490 	 * to the guest.
1491 	 */
1492 	host_tsc_clocksource = kvm_get_time_and_clockread(
1493 					&ka->master_kernel_ns,
1494 					&ka->master_cycle_now);
1495 
1496 	ka->use_master_clock = host_tsc_clocksource && vcpus_matched
1497 				&& !backwards_tsc_observed;
1498 
1499 	if (ka->use_master_clock)
1500 		atomic_set(&kvm_guest_has_master_clock, 1);
1501 
1502 	vclock_mode = pvclock_gtod_data.clock.vclock_mode;
1503 	trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
1504 					vcpus_matched);
1505 #endif
1506 }
1507 
1508 static void kvm_gen_update_masterclock(struct kvm *kvm)
1509 {
1510 #ifdef CONFIG_X86_64
1511 	int i;
1512 	struct kvm_vcpu *vcpu;
1513 	struct kvm_arch *ka = &kvm->arch;
1514 
1515 	spin_lock(&ka->pvclock_gtod_sync_lock);
1516 	kvm_make_mclock_inprogress_request(kvm);
1517 	/* no guest entries from this point */
1518 	pvclock_update_vm_gtod_copy(kvm);
1519 
1520 	kvm_for_each_vcpu(i, vcpu, kvm)
1521 		set_bit(KVM_REQ_CLOCK_UPDATE, &vcpu->requests);
1522 
1523 	/* guest entries allowed */
1524 	kvm_for_each_vcpu(i, vcpu, kvm)
1525 		clear_bit(KVM_REQ_MCLOCK_INPROGRESS, &vcpu->requests);
1526 
1527 	spin_unlock(&ka->pvclock_gtod_sync_lock);
1528 #endif
1529 }
1530 
1531 static int kvm_guest_time_update(struct kvm_vcpu *v)
1532 {
1533 	unsigned long flags, this_tsc_khz;
1534 	struct kvm_vcpu_arch *vcpu = &v->arch;
1535 	struct kvm_arch *ka = &v->kvm->arch;
1536 	s64 kernel_ns;
1537 	u64 tsc_timestamp, host_tsc;
1538 	struct pvclock_vcpu_time_info guest_hv_clock;
1539 	u8 pvclock_flags;
1540 	bool use_master_clock;
1541 
1542 	kernel_ns = 0;
1543 	host_tsc = 0;
1544 
1545 	/*
1546 	 * If the host uses TSC clock, then passthrough TSC as stable
1547 	 * to the guest.
1548 	 */
1549 	spin_lock(&ka->pvclock_gtod_sync_lock);
1550 	use_master_clock = ka->use_master_clock;
1551 	if (use_master_clock) {
1552 		host_tsc = ka->master_cycle_now;
1553 		kernel_ns = ka->master_kernel_ns;
1554 	}
1555 	spin_unlock(&ka->pvclock_gtod_sync_lock);
1556 
1557 	/* Keep irq disabled to prevent changes to the clock */
1558 	local_irq_save(flags);
1559 	this_tsc_khz = __get_cpu_var(cpu_tsc_khz);
1560 	if (unlikely(this_tsc_khz == 0)) {
1561 		local_irq_restore(flags);
1562 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1563 		return 1;
1564 	}
1565 	if (!use_master_clock) {
1566 		host_tsc = native_read_tsc();
1567 		kernel_ns = get_kernel_ns();
1568 	}
1569 
1570 	tsc_timestamp = kvm_x86_ops->read_l1_tsc(v, host_tsc);
1571 
1572 	/*
1573 	 * We may have to catch up the TSC to match elapsed wall clock
1574 	 * time for two reasons, even if kvmclock is used.
1575 	 *   1) CPU could have been running below the maximum TSC rate
1576 	 *   2) Broken TSC compensation resets the base at each VCPU
1577 	 *      entry to avoid unknown leaps of TSC even when running
1578 	 *      again on the same CPU.  This may cause apparent elapsed
1579 	 *      time to disappear, and the guest to stand still or run
1580 	 *	very slowly.
1581 	 */
1582 	if (vcpu->tsc_catchup) {
1583 		u64 tsc = compute_guest_tsc(v, kernel_ns);
1584 		if (tsc > tsc_timestamp) {
1585 			adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
1586 			tsc_timestamp = tsc;
1587 		}
1588 	}
1589 
1590 	local_irq_restore(flags);
1591 
1592 	if (!vcpu->pv_time_enabled)
1593 		return 0;
1594 
1595 	if (unlikely(vcpu->hw_tsc_khz != this_tsc_khz)) {
1596 		kvm_get_time_scale(NSEC_PER_SEC / 1000, this_tsc_khz,
1597 				   &vcpu->hv_clock.tsc_shift,
1598 				   &vcpu->hv_clock.tsc_to_system_mul);
1599 		vcpu->hw_tsc_khz = this_tsc_khz;
1600 	}
1601 
1602 	/* With all the info we got, fill in the values */
1603 	vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
1604 	vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1605 	vcpu->last_guest_tsc = tsc_timestamp;
1606 
1607 	/*
1608 	 * The interface expects us to write an even number signaling that the
1609 	 * update is finished. Since the guest won't see the intermediate
1610 	 * state, we just increase by 2 at the end.
1611 	 */
1612 	vcpu->hv_clock.version += 2;
1613 
1614 	if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
1615 		&guest_hv_clock, sizeof(guest_hv_clock))))
1616 		return 0;
1617 
1618 	/* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
1619 	pvclock_flags = (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
1620 
1621 	if (vcpu->pvclock_set_guest_stopped_request) {
1622 		pvclock_flags |= PVCLOCK_GUEST_STOPPED;
1623 		vcpu->pvclock_set_guest_stopped_request = false;
1624 	}
1625 
1626 	/* If the host uses TSC clocksource, then it is stable */
1627 	if (use_master_clock)
1628 		pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
1629 
1630 	vcpu->hv_clock.flags = pvclock_flags;
1631 
1632 	kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1633 				&vcpu->hv_clock,
1634 				sizeof(vcpu->hv_clock));
1635 	return 0;
1636 }
1637 
1638 /*
1639  * kvmclock updates which are isolated to a given vcpu, such as
1640  * vcpu->cpu migration, should not allow system_timestamp from
1641  * the rest of the vcpus to remain static. Otherwise ntp frequency
1642  * correction applies to one vcpu's system_timestamp but not
1643  * the others.
1644  *
1645  * So in those cases, request a kvmclock update for all vcpus.
1646  * We need to rate-limit these requests though, as they can
1647  * considerably slow guests that have a large number of vcpus.
1648  * The time for a remote vcpu to update its kvmclock is bound
1649  * by the delay we use to rate-limit the updates.
1650  */
1651 
1652 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
1653 
1654 static void kvmclock_update_fn(struct work_struct *work)
1655 {
1656 	int i;
1657 	struct delayed_work *dwork = to_delayed_work(work);
1658 	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1659 					   kvmclock_update_work);
1660 	struct kvm *kvm = container_of(ka, struct kvm, arch);
1661 	struct kvm_vcpu *vcpu;
1662 
1663 	kvm_for_each_vcpu(i, vcpu, kvm) {
1664 		set_bit(KVM_REQ_CLOCK_UPDATE, &vcpu->requests);
1665 		kvm_vcpu_kick(vcpu);
1666 	}
1667 }
1668 
1669 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
1670 {
1671 	struct kvm *kvm = v->kvm;
1672 
1673 	set_bit(KVM_REQ_CLOCK_UPDATE, &v->requests);
1674 	schedule_delayed_work(&kvm->arch.kvmclock_update_work,
1675 					KVMCLOCK_UPDATE_DELAY);
1676 }
1677 
1678 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
1679 
1680 static void kvmclock_sync_fn(struct work_struct *work)
1681 {
1682 	struct delayed_work *dwork = to_delayed_work(work);
1683 	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1684 					   kvmclock_sync_work);
1685 	struct kvm *kvm = container_of(ka, struct kvm, arch);
1686 
1687 	schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
1688 	schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
1689 					KVMCLOCK_SYNC_PERIOD);
1690 }
1691 
1692 static bool msr_mtrr_valid(unsigned msr)
1693 {
1694 	switch (msr) {
1695 	case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
1696 	case MSR_MTRRfix64K_00000:
1697 	case MSR_MTRRfix16K_80000:
1698 	case MSR_MTRRfix16K_A0000:
1699 	case MSR_MTRRfix4K_C0000:
1700 	case MSR_MTRRfix4K_C8000:
1701 	case MSR_MTRRfix4K_D0000:
1702 	case MSR_MTRRfix4K_D8000:
1703 	case MSR_MTRRfix4K_E0000:
1704 	case MSR_MTRRfix4K_E8000:
1705 	case MSR_MTRRfix4K_F0000:
1706 	case MSR_MTRRfix4K_F8000:
1707 	case MSR_MTRRdefType:
1708 	case MSR_IA32_CR_PAT:
1709 		return true;
1710 	case 0x2f8:
1711 		return true;
1712 	}
1713 	return false;
1714 }
1715 
1716 static bool valid_pat_type(unsigned t)
1717 {
1718 	return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
1719 }
1720 
1721 static bool valid_mtrr_type(unsigned t)
1722 {
1723 	return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
1724 }
1725 
1726 static bool mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1727 {
1728 	int i;
1729 
1730 	if (!msr_mtrr_valid(msr))
1731 		return false;
1732 
1733 	if (msr == MSR_IA32_CR_PAT) {
1734 		for (i = 0; i < 8; i++)
1735 			if (!valid_pat_type((data >> (i * 8)) & 0xff))
1736 				return false;
1737 		return true;
1738 	} else if (msr == MSR_MTRRdefType) {
1739 		if (data & ~0xcff)
1740 			return false;
1741 		return valid_mtrr_type(data & 0xff);
1742 	} else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
1743 		for (i = 0; i < 8 ; i++)
1744 			if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
1745 				return false;
1746 		return true;
1747 	}
1748 
1749 	/* variable MTRRs */
1750 	return valid_mtrr_type(data & 0xff);
1751 }
1752 
1753 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1754 {
1755 	u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1756 
1757 	if (!mtrr_valid(vcpu, msr, data))
1758 		return 1;
1759 
1760 	if (msr == MSR_MTRRdefType) {
1761 		vcpu->arch.mtrr_state.def_type = data;
1762 		vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
1763 	} else if (msr == MSR_MTRRfix64K_00000)
1764 		p[0] = data;
1765 	else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1766 		p[1 + msr - MSR_MTRRfix16K_80000] = data;
1767 	else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1768 		p[3 + msr - MSR_MTRRfix4K_C0000] = data;
1769 	else if (msr == MSR_IA32_CR_PAT)
1770 		vcpu->arch.pat = data;
1771 	else {	/* Variable MTRRs */
1772 		int idx, is_mtrr_mask;
1773 		u64 *pt;
1774 
1775 		idx = (msr - 0x200) / 2;
1776 		is_mtrr_mask = msr - 0x200 - 2 * idx;
1777 		if (!is_mtrr_mask)
1778 			pt =
1779 			  (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1780 		else
1781 			pt =
1782 			  (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1783 		*pt = data;
1784 	}
1785 
1786 	kvm_mmu_reset_context(vcpu);
1787 	return 0;
1788 }
1789 
1790 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1791 {
1792 	u64 mcg_cap = vcpu->arch.mcg_cap;
1793 	unsigned bank_num = mcg_cap & 0xff;
1794 
1795 	switch (msr) {
1796 	case MSR_IA32_MCG_STATUS:
1797 		vcpu->arch.mcg_status = data;
1798 		break;
1799 	case MSR_IA32_MCG_CTL:
1800 		if (!(mcg_cap & MCG_CTL_P))
1801 			return 1;
1802 		if (data != 0 && data != ~(u64)0)
1803 			return -1;
1804 		vcpu->arch.mcg_ctl = data;
1805 		break;
1806 	default:
1807 		if (msr >= MSR_IA32_MC0_CTL &&
1808 		    msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1809 			u32 offset = msr - MSR_IA32_MC0_CTL;
1810 			/* only 0 or all 1s can be written to IA32_MCi_CTL
1811 			 * some Linux kernels though clear bit 10 in bank 4 to
1812 			 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1813 			 * this to avoid an uncatched #GP in the guest
1814 			 */
1815 			if ((offset & 0x3) == 0 &&
1816 			    data != 0 && (data | (1 << 10)) != ~(u64)0)
1817 				return -1;
1818 			vcpu->arch.mce_banks[offset] = data;
1819 			break;
1820 		}
1821 		return 1;
1822 	}
1823 	return 0;
1824 }
1825 
1826 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1827 {
1828 	struct kvm *kvm = vcpu->kvm;
1829 	int lm = is_long_mode(vcpu);
1830 	u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1831 		: (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1832 	u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1833 		: kvm->arch.xen_hvm_config.blob_size_32;
1834 	u32 page_num = data & ~PAGE_MASK;
1835 	u64 page_addr = data & PAGE_MASK;
1836 	u8 *page;
1837 	int r;
1838 
1839 	r = -E2BIG;
1840 	if (page_num >= blob_size)
1841 		goto out;
1842 	r = -ENOMEM;
1843 	page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
1844 	if (IS_ERR(page)) {
1845 		r = PTR_ERR(page);
1846 		goto out;
1847 	}
1848 	if (kvm_write_guest(kvm, page_addr, page, PAGE_SIZE))
1849 		goto out_free;
1850 	r = 0;
1851 out_free:
1852 	kfree(page);
1853 out:
1854 	return r;
1855 }
1856 
1857 static bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1858 {
1859 	return kvm->arch.hv_hypercall & HV_X64_MSR_HYPERCALL_ENABLE;
1860 }
1861 
1862 static bool kvm_hv_msr_partition_wide(u32 msr)
1863 {
1864 	bool r = false;
1865 	switch (msr) {
1866 	case HV_X64_MSR_GUEST_OS_ID:
1867 	case HV_X64_MSR_HYPERCALL:
1868 	case HV_X64_MSR_REFERENCE_TSC:
1869 	case HV_X64_MSR_TIME_REF_COUNT:
1870 		r = true;
1871 		break;
1872 	}
1873 
1874 	return r;
1875 }
1876 
1877 static int set_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1878 {
1879 	struct kvm *kvm = vcpu->kvm;
1880 
1881 	switch (msr) {
1882 	case HV_X64_MSR_GUEST_OS_ID:
1883 		kvm->arch.hv_guest_os_id = data;
1884 		/* setting guest os id to zero disables hypercall page */
1885 		if (!kvm->arch.hv_guest_os_id)
1886 			kvm->arch.hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1887 		break;
1888 	case HV_X64_MSR_HYPERCALL: {
1889 		u64 gfn;
1890 		unsigned long addr;
1891 		u8 instructions[4];
1892 
1893 		/* if guest os id is not set hypercall should remain disabled */
1894 		if (!kvm->arch.hv_guest_os_id)
1895 			break;
1896 		if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1897 			kvm->arch.hv_hypercall = data;
1898 			break;
1899 		}
1900 		gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
1901 		addr = gfn_to_hva(kvm, gfn);
1902 		if (kvm_is_error_hva(addr))
1903 			return 1;
1904 		kvm_x86_ops->patch_hypercall(vcpu, instructions);
1905 		((unsigned char *)instructions)[3] = 0xc3; /* ret */
1906 		if (__copy_to_user((void __user *)addr, instructions, 4))
1907 			return 1;
1908 		kvm->arch.hv_hypercall = data;
1909 		mark_page_dirty(kvm, gfn);
1910 		break;
1911 	}
1912 	case HV_X64_MSR_REFERENCE_TSC: {
1913 		u64 gfn;
1914 		HV_REFERENCE_TSC_PAGE tsc_ref;
1915 		memset(&tsc_ref, 0, sizeof(tsc_ref));
1916 		kvm->arch.hv_tsc_page = data;
1917 		if (!(data & HV_X64_MSR_TSC_REFERENCE_ENABLE))
1918 			break;
1919 		gfn = data >> HV_X64_MSR_TSC_REFERENCE_ADDRESS_SHIFT;
1920 		if (kvm_write_guest(kvm, gfn << HV_X64_MSR_TSC_REFERENCE_ADDRESS_SHIFT,
1921 			&tsc_ref, sizeof(tsc_ref)))
1922 			return 1;
1923 		mark_page_dirty(kvm, gfn);
1924 		break;
1925 	}
1926 	default:
1927 		vcpu_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1928 			    "data 0x%llx\n", msr, data);
1929 		return 1;
1930 	}
1931 	return 0;
1932 }
1933 
1934 static int set_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1935 {
1936 	switch (msr) {
1937 	case HV_X64_MSR_APIC_ASSIST_PAGE: {
1938 		u64 gfn;
1939 		unsigned long addr;
1940 
1941 		if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) {
1942 			vcpu->arch.hv_vapic = data;
1943 			if (kvm_lapic_enable_pv_eoi(vcpu, 0))
1944 				return 1;
1945 			break;
1946 		}
1947 		gfn = data >> HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT;
1948 		addr = gfn_to_hva(vcpu->kvm, gfn);
1949 		if (kvm_is_error_hva(addr))
1950 			return 1;
1951 		if (__clear_user((void __user *)addr, PAGE_SIZE))
1952 			return 1;
1953 		vcpu->arch.hv_vapic = data;
1954 		mark_page_dirty(vcpu->kvm, gfn);
1955 		if (kvm_lapic_enable_pv_eoi(vcpu, gfn_to_gpa(gfn) | KVM_MSR_ENABLED))
1956 			return 1;
1957 		break;
1958 	}
1959 	case HV_X64_MSR_EOI:
1960 		return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1961 	case HV_X64_MSR_ICR:
1962 		return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1963 	case HV_X64_MSR_TPR:
1964 		return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1965 	default:
1966 		vcpu_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1967 			    "data 0x%llx\n", msr, data);
1968 		return 1;
1969 	}
1970 
1971 	return 0;
1972 }
1973 
1974 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
1975 {
1976 	gpa_t gpa = data & ~0x3f;
1977 
1978 	/* Bits 2:5 are reserved, Should be zero */
1979 	if (data & 0x3c)
1980 		return 1;
1981 
1982 	vcpu->arch.apf.msr_val = data;
1983 
1984 	if (!(data & KVM_ASYNC_PF_ENABLED)) {
1985 		kvm_clear_async_pf_completion_queue(vcpu);
1986 		kvm_async_pf_hash_reset(vcpu);
1987 		return 0;
1988 	}
1989 
1990 	if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
1991 					sizeof(u32)))
1992 		return 1;
1993 
1994 	vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
1995 	kvm_async_pf_wakeup_all(vcpu);
1996 	return 0;
1997 }
1998 
1999 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2000 {
2001 	vcpu->arch.pv_time_enabled = false;
2002 }
2003 
2004 static void accumulate_steal_time(struct kvm_vcpu *vcpu)
2005 {
2006 	u64 delta;
2007 
2008 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2009 		return;
2010 
2011 	delta = current->sched_info.run_delay - vcpu->arch.st.last_steal;
2012 	vcpu->arch.st.last_steal = current->sched_info.run_delay;
2013 	vcpu->arch.st.accum_steal = delta;
2014 }
2015 
2016 static void record_steal_time(struct kvm_vcpu *vcpu)
2017 {
2018 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2019 		return;
2020 
2021 	if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2022 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
2023 		return;
2024 
2025 	vcpu->arch.st.steal.steal += vcpu->arch.st.accum_steal;
2026 	vcpu->arch.st.steal.version += 2;
2027 	vcpu->arch.st.accum_steal = 0;
2028 
2029 	kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2030 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2031 }
2032 
2033 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2034 {
2035 	bool pr = false;
2036 	u32 msr = msr_info->index;
2037 	u64 data = msr_info->data;
2038 
2039 	switch (msr) {
2040 	case MSR_AMD64_NB_CFG:
2041 	case MSR_IA32_UCODE_REV:
2042 	case MSR_IA32_UCODE_WRITE:
2043 	case MSR_VM_HSAVE_PA:
2044 	case MSR_AMD64_PATCH_LOADER:
2045 	case MSR_AMD64_BU_CFG2:
2046 		break;
2047 
2048 	case MSR_EFER:
2049 		return set_efer(vcpu, data);
2050 	case MSR_K7_HWCR:
2051 		data &= ~(u64)0x40;	/* ignore flush filter disable */
2052 		data &= ~(u64)0x100;	/* ignore ignne emulation enable */
2053 		data &= ~(u64)0x8;	/* ignore TLB cache disable */
2054 		data &= ~(u64)0x40000;  /* ignore Mc status write enable */
2055 		if (data != 0) {
2056 			vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2057 				    data);
2058 			return 1;
2059 		}
2060 		break;
2061 	case MSR_FAM10H_MMIO_CONF_BASE:
2062 		if (data != 0) {
2063 			vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2064 				    "0x%llx\n", data);
2065 			return 1;
2066 		}
2067 		break;
2068 	case MSR_IA32_DEBUGCTLMSR:
2069 		if (!data) {
2070 			/* We support the non-activated case already */
2071 			break;
2072 		} else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2073 			/* Values other than LBR and BTF are vendor-specific,
2074 			   thus reserved and should throw a #GP */
2075 			return 1;
2076 		}
2077 		vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2078 			    __func__, data);
2079 		break;
2080 	case 0x200 ... 0x2ff:
2081 		return set_msr_mtrr(vcpu, msr, data);
2082 	case MSR_IA32_APICBASE:
2083 		return kvm_set_apic_base(vcpu, msr_info);
2084 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2085 		return kvm_x2apic_msr_write(vcpu, msr, data);
2086 	case MSR_IA32_TSCDEADLINE:
2087 		kvm_set_lapic_tscdeadline_msr(vcpu, data);
2088 		break;
2089 	case MSR_IA32_TSC_ADJUST:
2090 		if (guest_cpuid_has_tsc_adjust(vcpu)) {
2091 			if (!msr_info->host_initiated) {
2092 				u64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2093 				kvm_x86_ops->adjust_tsc_offset(vcpu, adj, true);
2094 			}
2095 			vcpu->arch.ia32_tsc_adjust_msr = data;
2096 		}
2097 		break;
2098 	case MSR_IA32_MISC_ENABLE:
2099 		vcpu->arch.ia32_misc_enable_msr = data;
2100 		break;
2101 	case MSR_KVM_WALL_CLOCK_NEW:
2102 	case MSR_KVM_WALL_CLOCK:
2103 		vcpu->kvm->arch.wall_clock = data;
2104 		kvm_write_wall_clock(vcpu->kvm, data);
2105 		break;
2106 	case MSR_KVM_SYSTEM_TIME_NEW:
2107 	case MSR_KVM_SYSTEM_TIME: {
2108 		u64 gpa_offset;
2109 		kvmclock_reset(vcpu);
2110 
2111 		vcpu->arch.time = data;
2112 		kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2113 
2114 		/* we verify if the enable bit is set... */
2115 		if (!(data & 1))
2116 			break;
2117 
2118 		gpa_offset = data & ~(PAGE_MASK | 1);
2119 
2120 		if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2121 		     &vcpu->arch.pv_time, data & ~1ULL,
2122 		     sizeof(struct pvclock_vcpu_time_info)))
2123 			vcpu->arch.pv_time_enabled = false;
2124 		else
2125 			vcpu->arch.pv_time_enabled = true;
2126 
2127 		break;
2128 	}
2129 	case MSR_KVM_ASYNC_PF_EN:
2130 		if (kvm_pv_enable_async_pf(vcpu, data))
2131 			return 1;
2132 		break;
2133 	case MSR_KVM_STEAL_TIME:
2134 
2135 		if (unlikely(!sched_info_on()))
2136 			return 1;
2137 
2138 		if (data & KVM_STEAL_RESERVED_MASK)
2139 			return 1;
2140 
2141 		if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
2142 						data & KVM_STEAL_VALID_BITS,
2143 						sizeof(struct kvm_steal_time)))
2144 			return 1;
2145 
2146 		vcpu->arch.st.msr_val = data;
2147 
2148 		if (!(data & KVM_MSR_ENABLED))
2149 			break;
2150 
2151 		vcpu->arch.st.last_steal = current->sched_info.run_delay;
2152 
2153 		preempt_disable();
2154 		accumulate_steal_time(vcpu);
2155 		preempt_enable();
2156 
2157 		kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2158 
2159 		break;
2160 	case MSR_KVM_PV_EOI_EN:
2161 		if (kvm_lapic_enable_pv_eoi(vcpu, data))
2162 			return 1;
2163 		break;
2164 
2165 	case MSR_IA32_MCG_CTL:
2166 	case MSR_IA32_MCG_STATUS:
2167 	case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
2168 		return set_msr_mce(vcpu, msr, data);
2169 
2170 	/* Performance counters are not protected by a CPUID bit,
2171 	 * so we should check all of them in the generic path for the sake of
2172 	 * cross vendor migration.
2173 	 * Writing a zero into the event select MSRs disables them,
2174 	 * which we perfectly emulate ;-). Any other value should be at least
2175 	 * reported, some guests depend on them.
2176 	 */
2177 	case MSR_K7_EVNTSEL0:
2178 	case MSR_K7_EVNTSEL1:
2179 	case MSR_K7_EVNTSEL2:
2180 	case MSR_K7_EVNTSEL3:
2181 		if (data != 0)
2182 			vcpu_unimpl(vcpu, "unimplemented perfctr wrmsr: "
2183 				    "0x%x data 0x%llx\n", msr, data);
2184 		break;
2185 	/* at least RHEL 4 unconditionally writes to the perfctr registers,
2186 	 * so we ignore writes to make it happy.
2187 	 */
2188 	case MSR_K7_PERFCTR0:
2189 	case MSR_K7_PERFCTR1:
2190 	case MSR_K7_PERFCTR2:
2191 	case MSR_K7_PERFCTR3:
2192 		vcpu_unimpl(vcpu, "unimplemented perfctr wrmsr: "
2193 			    "0x%x data 0x%llx\n", msr, data);
2194 		break;
2195 	case MSR_P6_PERFCTR0:
2196 	case MSR_P6_PERFCTR1:
2197 		pr = true;
2198 	case MSR_P6_EVNTSEL0:
2199 	case MSR_P6_EVNTSEL1:
2200 		if (kvm_pmu_msr(vcpu, msr))
2201 			return kvm_pmu_set_msr(vcpu, msr_info);
2202 
2203 		if (pr || data != 0)
2204 			vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2205 				    "0x%x data 0x%llx\n", msr, data);
2206 		break;
2207 	case MSR_K7_CLK_CTL:
2208 		/*
2209 		 * Ignore all writes to this no longer documented MSR.
2210 		 * Writes are only relevant for old K7 processors,
2211 		 * all pre-dating SVM, but a recommended workaround from
2212 		 * AMD for these chips. It is possible to specify the
2213 		 * affected processor models on the command line, hence
2214 		 * the need to ignore the workaround.
2215 		 */
2216 		break;
2217 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2218 		if (kvm_hv_msr_partition_wide(msr)) {
2219 			int r;
2220 			mutex_lock(&vcpu->kvm->lock);
2221 			r = set_msr_hyperv_pw(vcpu, msr, data);
2222 			mutex_unlock(&vcpu->kvm->lock);
2223 			return r;
2224 		} else
2225 			return set_msr_hyperv(vcpu, msr, data);
2226 		break;
2227 	case MSR_IA32_BBL_CR_CTL3:
2228 		/* Drop writes to this legacy MSR -- see rdmsr
2229 		 * counterpart for further detail.
2230 		 */
2231 		vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n", msr, data);
2232 		break;
2233 	case MSR_AMD64_OSVW_ID_LENGTH:
2234 		if (!guest_cpuid_has_osvw(vcpu))
2235 			return 1;
2236 		vcpu->arch.osvw.length = data;
2237 		break;
2238 	case MSR_AMD64_OSVW_STATUS:
2239 		if (!guest_cpuid_has_osvw(vcpu))
2240 			return 1;
2241 		vcpu->arch.osvw.status = data;
2242 		break;
2243 	default:
2244 		if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2245 			return xen_hvm_config(vcpu, data);
2246 		if (kvm_pmu_msr(vcpu, msr))
2247 			return kvm_pmu_set_msr(vcpu, msr_info);
2248 		if (!ignore_msrs) {
2249 			vcpu_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
2250 				    msr, data);
2251 			return 1;
2252 		} else {
2253 			vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
2254 				    msr, data);
2255 			break;
2256 		}
2257 	}
2258 	return 0;
2259 }
2260 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2261 
2262 
2263 /*
2264  * Reads an msr value (of 'msr_index') into 'pdata'.
2265  * Returns 0 on success, non-0 otherwise.
2266  * Assumes vcpu_load() was already called.
2267  */
2268 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
2269 {
2270 	return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
2271 }
2272 
2273 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2274 {
2275 	u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
2276 
2277 	if (!msr_mtrr_valid(msr))
2278 		return 1;
2279 
2280 	if (msr == MSR_MTRRdefType)
2281 		*pdata = vcpu->arch.mtrr_state.def_type +
2282 			 (vcpu->arch.mtrr_state.enabled << 10);
2283 	else if (msr == MSR_MTRRfix64K_00000)
2284 		*pdata = p[0];
2285 	else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
2286 		*pdata = p[1 + msr - MSR_MTRRfix16K_80000];
2287 	else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
2288 		*pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
2289 	else if (msr == MSR_IA32_CR_PAT)
2290 		*pdata = vcpu->arch.pat;
2291 	else {	/* Variable MTRRs */
2292 		int idx, is_mtrr_mask;
2293 		u64 *pt;
2294 
2295 		idx = (msr - 0x200) / 2;
2296 		is_mtrr_mask = msr - 0x200 - 2 * idx;
2297 		if (!is_mtrr_mask)
2298 			pt =
2299 			  (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
2300 		else
2301 			pt =
2302 			  (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
2303 		*pdata = *pt;
2304 	}
2305 
2306 	return 0;
2307 }
2308 
2309 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2310 {
2311 	u64 data;
2312 	u64 mcg_cap = vcpu->arch.mcg_cap;
2313 	unsigned bank_num = mcg_cap & 0xff;
2314 
2315 	switch (msr) {
2316 	case MSR_IA32_P5_MC_ADDR:
2317 	case MSR_IA32_P5_MC_TYPE:
2318 		data = 0;
2319 		break;
2320 	case MSR_IA32_MCG_CAP:
2321 		data = vcpu->arch.mcg_cap;
2322 		break;
2323 	case MSR_IA32_MCG_CTL:
2324 		if (!(mcg_cap & MCG_CTL_P))
2325 			return 1;
2326 		data = vcpu->arch.mcg_ctl;
2327 		break;
2328 	case MSR_IA32_MCG_STATUS:
2329 		data = vcpu->arch.mcg_status;
2330 		break;
2331 	default:
2332 		if (msr >= MSR_IA32_MC0_CTL &&
2333 		    msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
2334 			u32 offset = msr - MSR_IA32_MC0_CTL;
2335 			data = vcpu->arch.mce_banks[offset];
2336 			break;
2337 		}
2338 		return 1;
2339 	}
2340 	*pdata = data;
2341 	return 0;
2342 }
2343 
2344 static int get_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2345 {
2346 	u64 data = 0;
2347 	struct kvm *kvm = vcpu->kvm;
2348 
2349 	switch (msr) {
2350 	case HV_X64_MSR_GUEST_OS_ID:
2351 		data = kvm->arch.hv_guest_os_id;
2352 		break;
2353 	case HV_X64_MSR_HYPERCALL:
2354 		data = kvm->arch.hv_hypercall;
2355 		break;
2356 	case HV_X64_MSR_TIME_REF_COUNT: {
2357 		data =
2358 		     div_u64(get_kernel_ns() + kvm->arch.kvmclock_offset, 100);
2359 		break;
2360 	}
2361 	case HV_X64_MSR_REFERENCE_TSC:
2362 		data = kvm->arch.hv_tsc_page;
2363 		break;
2364 	default:
2365 		vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
2366 		return 1;
2367 	}
2368 
2369 	*pdata = data;
2370 	return 0;
2371 }
2372 
2373 static int get_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2374 {
2375 	u64 data = 0;
2376 
2377 	switch (msr) {
2378 	case HV_X64_MSR_VP_INDEX: {
2379 		int r;
2380 		struct kvm_vcpu *v;
2381 		kvm_for_each_vcpu(r, v, vcpu->kvm) {
2382 			if (v == vcpu) {
2383 				data = r;
2384 				break;
2385 			}
2386 		}
2387 		break;
2388 	}
2389 	case HV_X64_MSR_EOI:
2390 		return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
2391 	case HV_X64_MSR_ICR:
2392 		return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
2393 	case HV_X64_MSR_TPR:
2394 		return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
2395 	case HV_X64_MSR_APIC_ASSIST_PAGE:
2396 		data = vcpu->arch.hv_vapic;
2397 		break;
2398 	default:
2399 		vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
2400 		return 1;
2401 	}
2402 	*pdata = data;
2403 	return 0;
2404 }
2405 
2406 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2407 {
2408 	u64 data;
2409 
2410 	switch (msr) {
2411 	case MSR_IA32_PLATFORM_ID:
2412 	case MSR_IA32_EBL_CR_POWERON:
2413 	case MSR_IA32_DEBUGCTLMSR:
2414 	case MSR_IA32_LASTBRANCHFROMIP:
2415 	case MSR_IA32_LASTBRANCHTOIP:
2416 	case MSR_IA32_LASTINTFROMIP:
2417 	case MSR_IA32_LASTINTTOIP:
2418 	case MSR_K8_SYSCFG:
2419 	case MSR_K7_HWCR:
2420 	case MSR_VM_HSAVE_PA:
2421 	case MSR_K7_EVNTSEL0:
2422 	case MSR_K7_PERFCTR0:
2423 	case MSR_K8_INT_PENDING_MSG:
2424 	case MSR_AMD64_NB_CFG:
2425 	case MSR_FAM10H_MMIO_CONF_BASE:
2426 	case MSR_AMD64_BU_CFG2:
2427 		data = 0;
2428 		break;
2429 	case MSR_P6_PERFCTR0:
2430 	case MSR_P6_PERFCTR1:
2431 	case MSR_P6_EVNTSEL0:
2432 	case MSR_P6_EVNTSEL1:
2433 		if (kvm_pmu_msr(vcpu, msr))
2434 			return kvm_pmu_get_msr(vcpu, msr, pdata);
2435 		data = 0;
2436 		break;
2437 	case MSR_IA32_UCODE_REV:
2438 		data = 0x100000000ULL;
2439 		break;
2440 	case MSR_MTRRcap:
2441 		data = 0x500 | KVM_NR_VAR_MTRR;
2442 		break;
2443 	case 0x200 ... 0x2ff:
2444 		return get_msr_mtrr(vcpu, msr, pdata);
2445 	case 0xcd: /* fsb frequency */
2446 		data = 3;
2447 		break;
2448 		/*
2449 		 * MSR_EBC_FREQUENCY_ID
2450 		 * Conservative value valid for even the basic CPU models.
2451 		 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2452 		 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2453 		 * and 266MHz for model 3, or 4. Set Core Clock
2454 		 * Frequency to System Bus Frequency Ratio to 1 (bits
2455 		 * 31:24) even though these are only valid for CPU
2456 		 * models > 2, however guests may end up dividing or
2457 		 * multiplying by zero otherwise.
2458 		 */
2459 	case MSR_EBC_FREQUENCY_ID:
2460 		data = 1 << 24;
2461 		break;
2462 	case MSR_IA32_APICBASE:
2463 		data = kvm_get_apic_base(vcpu);
2464 		break;
2465 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2466 		return kvm_x2apic_msr_read(vcpu, msr, pdata);
2467 		break;
2468 	case MSR_IA32_TSCDEADLINE:
2469 		data = kvm_get_lapic_tscdeadline_msr(vcpu);
2470 		break;
2471 	case MSR_IA32_TSC_ADJUST:
2472 		data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
2473 		break;
2474 	case MSR_IA32_MISC_ENABLE:
2475 		data = vcpu->arch.ia32_misc_enable_msr;
2476 		break;
2477 	case MSR_IA32_PERF_STATUS:
2478 		/* TSC increment by tick */
2479 		data = 1000ULL;
2480 		/* CPU multiplier */
2481 		data |= (((uint64_t)4ULL) << 40);
2482 		break;
2483 	case MSR_EFER:
2484 		data = vcpu->arch.efer;
2485 		break;
2486 	case MSR_KVM_WALL_CLOCK:
2487 	case MSR_KVM_WALL_CLOCK_NEW:
2488 		data = vcpu->kvm->arch.wall_clock;
2489 		break;
2490 	case MSR_KVM_SYSTEM_TIME:
2491 	case MSR_KVM_SYSTEM_TIME_NEW:
2492 		data = vcpu->arch.time;
2493 		break;
2494 	case MSR_KVM_ASYNC_PF_EN:
2495 		data = vcpu->arch.apf.msr_val;
2496 		break;
2497 	case MSR_KVM_STEAL_TIME:
2498 		data = vcpu->arch.st.msr_val;
2499 		break;
2500 	case MSR_KVM_PV_EOI_EN:
2501 		data = vcpu->arch.pv_eoi.msr_val;
2502 		break;
2503 	case MSR_IA32_P5_MC_ADDR:
2504 	case MSR_IA32_P5_MC_TYPE:
2505 	case MSR_IA32_MCG_CAP:
2506 	case MSR_IA32_MCG_CTL:
2507 	case MSR_IA32_MCG_STATUS:
2508 	case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
2509 		return get_msr_mce(vcpu, msr, pdata);
2510 	case MSR_K7_CLK_CTL:
2511 		/*
2512 		 * Provide expected ramp-up count for K7. All other
2513 		 * are set to zero, indicating minimum divisors for
2514 		 * every field.
2515 		 *
2516 		 * This prevents guest kernels on AMD host with CPU
2517 		 * type 6, model 8 and higher from exploding due to
2518 		 * the rdmsr failing.
2519 		 */
2520 		data = 0x20000000;
2521 		break;
2522 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2523 		if (kvm_hv_msr_partition_wide(msr)) {
2524 			int r;
2525 			mutex_lock(&vcpu->kvm->lock);
2526 			r = get_msr_hyperv_pw(vcpu, msr, pdata);
2527 			mutex_unlock(&vcpu->kvm->lock);
2528 			return r;
2529 		} else
2530 			return get_msr_hyperv(vcpu, msr, pdata);
2531 		break;
2532 	case MSR_IA32_BBL_CR_CTL3:
2533 		/* This legacy MSR exists but isn't fully documented in current
2534 		 * silicon.  It is however accessed by winxp in very narrow
2535 		 * scenarios where it sets bit #19, itself documented as
2536 		 * a "reserved" bit.  Best effort attempt to source coherent
2537 		 * read data here should the balance of the register be
2538 		 * interpreted by the guest:
2539 		 *
2540 		 * L2 cache control register 3: 64GB range, 256KB size,
2541 		 * enabled, latency 0x1, configured
2542 		 */
2543 		data = 0xbe702111;
2544 		break;
2545 	case MSR_AMD64_OSVW_ID_LENGTH:
2546 		if (!guest_cpuid_has_osvw(vcpu))
2547 			return 1;
2548 		data = vcpu->arch.osvw.length;
2549 		break;
2550 	case MSR_AMD64_OSVW_STATUS:
2551 		if (!guest_cpuid_has_osvw(vcpu))
2552 			return 1;
2553 		data = vcpu->arch.osvw.status;
2554 		break;
2555 	default:
2556 		if (kvm_pmu_msr(vcpu, msr))
2557 			return kvm_pmu_get_msr(vcpu, msr, pdata);
2558 		if (!ignore_msrs) {
2559 			vcpu_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
2560 			return 1;
2561 		} else {
2562 			vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr);
2563 			data = 0;
2564 		}
2565 		break;
2566 	}
2567 	*pdata = data;
2568 	return 0;
2569 }
2570 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2571 
2572 /*
2573  * Read or write a bunch of msrs. All parameters are kernel addresses.
2574  *
2575  * @return number of msrs set successfully.
2576  */
2577 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2578 		    struct kvm_msr_entry *entries,
2579 		    int (*do_msr)(struct kvm_vcpu *vcpu,
2580 				  unsigned index, u64 *data))
2581 {
2582 	int i, idx;
2583 
2584 	idx = srcu_read_lock(&vcpu->kvm->srcu);
2585 	for (i = 0; i < msrs->nmsrs; ++i)
2586 		if (do_msr(vcpu, entries[i].index, &entries[i].data))
2587 			break;
2588 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
2589 
2590 	return i;
2591 }
2592 
2593 /*
2594  * Read or write a bunch of msrs. Parameters are user addresses.
2595  *
2596  * @return number of msrs set successfully.
2597  */
2598 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2599 		  int (*do_msr)(struct kvm_vcpu *vcpu,
2600 				unsigned index, u64 *data),
2601 		  int writeback)
2602 {
2603 	struct kvm_msrs msrs;
2604 	struct kvm_msr_entry *entries;
2605 	int r, n;
2606 	unsigned size;
2607 
2608 	r = -EFAULT;
2609 	if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2610 		goto out;
2611 
2612 	r = -E2BIG;
2613 	if (msrs.nmsrs >= MAX_IO_MSRS)
2614 		goto out;
2615 
2616 	size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2617 	entries = memdup_user(user_msrs->entries, size);
2618 	if (IS_ERR(entries)) {
2619 		r = PTR_ERR(entries);
2620 		goto out;
2621 	}
2622 
2623 	r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2624 	if (r < 0)
2625 		goto out_free;
2626 
2627 	r = -EFAULT;
2628 	if (writeback && copy_to_user(user_msrs->entries, entries, size))
2629 		goto out_free;
2630 
2631 	r = n;
2632 
2633 out_free:
2634 	kfree(entries);
2635 out:
2636 	return r;
2637 }
2638 
2639 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
2640 {
2641 	int r;
2642 
2643 	switch (ext) {
2644 	case KVM_CAP_IRQCHIP:
2645 	case KVM_CAP_HLT:
2646 	case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2647 	case KVM_CAP_SET_TSS_ADDR:
2648 	case KVM_CAP_EXT_CPUID:
2649 	case KVM_CAP_EXT_EMUL_CPUID:
2650 	case KVM_CAP_CLOCKSOURCE:
2651 	case KVM_CAP_PIT:
2652 	case KVM_CAP_NOP_IO_DELAY:
2653 	case KVM_CAP_MP_STATE:
2654 	case KVM_CAP_SYNC_MMU:
2655 	case KVM_CAP_USER_NMI:
2656 	case KVM_CAP_REINJECT_CONTROL:
2657 	case KVM_CAP_IRQ_INJECT_STATUS:
2658 	case KVM_CAP_IRQFD:
2659 	case KVM_CAP_IOEVENTFD:
2660 	case KVM_CAP_IOEVENTFD_NO_LENGTH:
2661 	case KVM_CAP_PIT2:
2662 	case KVM_CAP_PIT_STATE2:
2663 	case KVM_CAP_SET_IDENTITY_MAP_ADDR:
2664 	case KVM_CAP_XEN_HVM:
2665 	case KVM_CAP_ADJUST_CLOCK:
2666 	case KVM_CAP_VCPU_EVENTS:
2667 	case KVM_CAP_HYPERV:
2668 	case KVM_CAP_HYPERV_VAPIC:
2669 	case KVM_CAP_HYPERV_SPIN:
2670 	case KVM_CAP_PCI_SEGMENT:
2671 	case KVM_CAP_DEBUGREGS:
2672 	case KVM_CAP_X86_ROBUST_SINGLESTEP:
2673 	case KVM_CAP_XSAVE:
2674 	case KVM_CAP_ASYNC_PF:
2675 	case KVM_CAP_GET_TSC_KHZ:
2676 	case KVM_CAP_KVMCLOCK_CTRL:
2677 	case KVM_CAP_READONLY_MEM:
2678 	case KVM_CAP_HYPERV_TIME:
2679 	case KVM_CAP_IOAPIC_POLARITY_IGNORED:
2680 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2681 	case KVM_CAP_ASSIGN_DEV_IRQ:
2682 	case KVM_CAP_PCI_2_3:
2683 #endif
2684 		r = 1;
2685 		break;
2686 	case KVM_CAP_COALESCED_MMIO:
2687 		r = KVM_COALESCED_MMIO_PAGE_OFFSET;
2688 		break;
2689 	case KVM_CAP_VAPIC:
2690 		r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2691 		break;
2692 	case KVM_CAP_NR_VCPUS:
2693 		r = KVM_SOFT_MAX_VCPUS;
2694 		break;
2695 	case KVM_CAP_MAX_VCPUS:
2696 		r = KVM_MAX_VCPUS;
2697 		break;
2698 	case KVM_CAP_NR_MEMSLOTS:
2699 		r = KVM_USER_MEM_SLOTS;
2700 		break;
2701 	case KVM_CAP_PV_MMU:	/* obsolete */
2702 		r = 0;
2703 		break;
2704 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2705 	case KVM_CAP_IOMMU:
2706 		r = iommu_present(&pci_bus_type);
2707 		break;
2708 #endif
2709 	case KVM_CAP_MCE:
2710 		r = KVM_MAX_MCE_BANKS;
2711 		break;
2712 	case KVM_CAP_XCRS:
2713 		r = cpu_has_xsave;
2714 		break;
2715 	case KVM_CAP_TSC_CONTROL:
2716 		r = kvm_has_tsc_control;
2717 		break;
2718 	case KVM_CAP_TSC_DEADLINE_TIMER:
2719 		r = boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER);
2720 		break;
2721 	default:
2722 		r = 0;
2723 		break;
2724 	}
2725 	return r;
2726 
2727 }
2728 
2729 long kvm_arch_dev_ioctl(struct file *filp,
2730 			unsigned int ioctl, unsigned long arg)
2731 {
2732 	void __user *argp = (void __user *)arg;
2733 	long r;
2734 
2735 	switch (ioctl) {
2736 	case KVM_GET_MSR_INDEX_LIST: {
2737 		struct kvm_msr_list __user *user_msr_list = argp;
2738 		struct kvm_msr_list msr_list;
2739 		unsigned n;
2740 
2741 		r = -EFAULT;
2742 		if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2743 			goto out;
2744 		n = msr_list.nmsrs;
2745 		msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2746 		if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2747 			goto out;
2748 		r = -E2BIG;
2749 		if (n < msr_list.nmsrs)
2750 			goto out;
2751 		r = -EFAULT;
2752 		if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2753 				 num_msrs_to_save * sizeof(u32)))
2754 			goto out;
2755 		if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2756 				 &emulated_msrs,
2757 				 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2758 			goto out;
2759 		r = 0;
2760 		break;
2761 	}
2762 	case KVM_GET_SUPPORTED_CPUID:
2763 	case KVM_GET_EMULATED_CPUID: {
2764 		struct kvm_cpuid2 __user *cpuid_arg = argp;
2765 		struct kvm_cpuid2 cpuid;
2766 
2767 		r = -EFAULT;
2768 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2769 			goto out;
2770 
2771 		r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
2772 					    ioctl);
2773 		if (r)
2774 			goto out;
2775 
2776 		r = -EFAULT;
2777 		if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2778 			goto out;
2779 		r = 0;
2780 		break;
2781 	}
2782 	case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2783 		u64 mce_cap;
2784 
2785 		mce_cap = KVM_MCE_CAP_SUPPORTED;
2786 		r = -EFAULT;
2787 		if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
2788 			goto out;
2789 		r = 0;
2790 		break;
2791 	}
2792 	default:
2793 		r = -EINVAL;
2794 	}
2795 out:
2796 	return r;
2797 }
2798 
2799 static void wbinvd_ipi(void *garbage)
2800 {
2801 	wbinvd();
2802 }
2803 
2804 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2805 {
2806 	return kvm_arch_has_noncoherent_dma(vcpu->kvm);
2807 }
2808 
2809 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2810 {
2811 	/* Address WBINVD may be executed by guest */
2812 	if (need_emulate_wbinvd(vcpu)) {
2813 		if (kvm_x86_ops->has_wbinvd_exit())
2814 			cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2815 		else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2816 			smp_call_function_single(vcpu->cpu,
2817 					wbinvd_ipi, NULL, 1);
2818 	}
2819 
2820 	kvm_x86_ops->vcpu_load(vcpu, cpu);
2821 
2822 	/* Apply any externally detected TSC adjustments (due to suspend) */
2823 	if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
2824 		adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
2825 		vcpu->arch.tsc_offset_adjustment = 0;
2826 		set_bit(KVM_REQ_CLOCK_UPDATE, &vcpu->requests);
2827 	}
2828 
2829 	if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
2830 		s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
2831 				native_read_tsc() - vcpu->arch.last_host_tsc;
2832 		if (tsc_delta < 0)
2833 			mark_tsc_unstable("KVM discovered backwards TSC");
2834 		if (check_tsc_unstable()) {
2835 			u64 offset = kvm_x86_ops->compute_tsc_offset(vcpu,
2836 						vcpu->arch.last_guest_tsc);
2837 			kvm_x86_ops->write_tsc_offset(vcpu, offset);
2838 			vcpu->arch.tsc_catchup = 1;
2839 		}
2840 		/*
2841 		 * On a host with synchronized TSC, there is no need to update
2842 		 * kvmclock on vcpu->cpu migration
2843 		 */
2844 		if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
2845 			kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2846 		if (vcpu->cpu != cpu)
2847 			kvm_migrate_timers(vcpu);
2848 		vcpu->cpu = cpu;
2849 	}
2850 
2851 	accumulate_steal_time(vcpu);
2852 	kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2853 }
2854 
2855 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2856 {
2857 	kvm_x86_ops->vcpu_put(vcpu);
2858 	kvm_put_guest_fpu(vcpu);
2859 	vcpu->arch.last_host_tsc = native_read_tsc();
2860 }
2861 
2862 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2863 				    struct kvm_lapic_state *s)
2864 {
2865 	kvm_x86_ops->sync_pir_to_irr(vcpu);
2866 	memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
2867 
2868 	return 0;
2869 }
2870 
2871 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2872 				    struct kvm_lapic_state *s)
2873 {
2874 	kvm_apic_post_state_restore(vcpu, s);
2875 	update_cr8_intercept(vcpu);
2876 
2877 	return 0;
2878 }
2879 
2880 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2881 				    struct kvm_interrupt *irq)
2882 {
2883 	if (irq->irq >= KVM_NR_INTERRUPTS)
2884 		return -EINVAL;
2885 	if (irqchip_in_kernel(vcpu->kvm))
2886 		return -ENXIO;
2887 
2888 	kvm_queue_interrupt(vcpu, irq->irq, false);
2889 	kvm_make_request(KVM_REQ_EVENT, vcpu);
2890 
2891 	return 0;
2892 }
2893 
2894 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2895 {
2896 	kvm_inject_nmi(vcpu);
2897 
2898 	return 0;
2899 }
2900 
2901 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2902 					   struct kvm_tpr_access_ctl *tac)
2903 {
2904 	if (tac->flags)
2905 		return -EINVAL;
2906 	vcpu->arch.tpr_access_reporting = !!tac->enabled;
2907 	return 0;
2908 }
2909 
2910 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2911 					u64 mcg_cap)
2912 {
2913 	int r;
2914 	unsigned bank_num = mcg_cap & 0xff, bank;
2915 
2916 	r = -EINVAL;
2917 	if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2918 		goto out;
2919 	if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2920 		goto out;
2921 	r = 0;
2922 	vcpu->arch.mcg_cap = mcg_cap;
2923 	/* Init IA32_MCG_CTL to all 1s */
2924 	if (mcg_cap & MCG_CTL_P)
2925 		vcpu->arch.mcg_ctl = ~(u64)0;
2926 	/* Init IA32_MCi_CTL to all 1s */
2927 	for (bank = 0; bank < bank_num; bank++)
2928 		vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2929 out:
2930 	return r;
2931 }
2932 
2933 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2934 				      struct kvm_x86_mce *mce)
2935 {
2936 	u64 mcg_cap = vcpu->arch.mcg_cap;
2937 	unsigned bank_num = mcg_cap & 0xff;
2938 	u64 *banks = vcpu->arch.mce_banks;
2939 
2940 	if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2941 		return -EINVAL;
2942 	/*
2943 	 * if IA32_MCG_CTL is not all 1s, the uncorrected error
2944 	 * reporting is disabled
2945 	 */
2946 	if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2947 	    vcpu->arch.mcg_ctl != ~(u64)0)
2948 		return 0;
2949 	banks += 4 * mce->bank;
2950 	/*
2951 	 * if IA32_MCi_CTL is not all 1s, the uncorrected error
2952 	 * reporting is disabled for the bank
2953 	 */
2954 	if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2955 		return 0;
2956 	if (mce->status & MCI_STATUS_UC) {
2957 		if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
2958 		    !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
2959 			kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2960 			return 0;
2961 		}
2962 		if (banks[1] & MCI_STATUS_VAL)
2963 			mce->status |= MCI_STATUS_OVER;
2964 		banks[2] = mce->addr;
2965 		banks[3] = mce->misc;
2966 		vcpu->arch.mcg_status = mce->mcg_status;
2967 		banks[1] = mce->status;
2968 		kvm_queue_exception(vcpu, MC_VECTOR);
2969 	} else if (!(banks[1] & MCI_STATUS_VAL)
2970 		   || !(banks[1] & MCI_STATUS_UC)) {
2971 		if (banks[1] & MCI_STATUS_VAL)
2972 			mce->status |= MCI_STATUS_OVER;
2973 		banks[2] = mce->addr;
2974 		banks[3] = mce->misc;
2975 		banks[1] = mce->status;
2976 	} else
2977 		banks[1] |= MCI_STATUS_OVER;
2978 	return 0;
2979 }
2980 
2981 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2982 					       struct kvm_vcpu_events *events)
2983 {
2984 	process_nmi(vcpu);
2985 	events->exception.injected =
2986 		vcpu->arch.exception.pending &&
2987 		!kvm_exception_is_soft(vcpu->arch.exception.nr);
2988 	events->exception.nr = vcpu->arch.exception.nr;
2989 	events->exception.has_error_code = vcpu->arch.exception.has_error_code;
2990 	events->exception.pad = 0;
2991 	events->exception.error_code = vcpu->arch.exception.error_code;
2992 
2993 	events->interrupt.injected =
2994 		vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
2995 	events->interrupt.nr = vcpu->arch.interrupt.nr;
2996 	events->interrupt.soft = 0;
2997 	events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
2998 
2999 	events->nmi.injected = vcpu->arch.nmi_injected;
3000 	events->nmi.pending = vcpu->arch.nmi_pending != 0;
3001 	events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
3002 	events->nmi.pad = 0;
3003 
3004 	events->sipi_vector = 0; /* never valid when reporting to user space */
3005 
3006 	events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
3007 			 | KVM_VCPUEVENT_VALID_SHADOW);
3008 	memset(&events->reserved, 0, sizeof(events->reserved));
3009 }
3010 
3011 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
3012 					      struct kvm_vcpu_events *events)
3013 {
3014 	if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
3015 			      | KVM_VCPUEVENT_VALID_SIPI_VECTOR
3016 			      | KVM_VCPUEVENT_VALID_SHADOW))
3017 		return -EINVAL;
3018 
3019 	process_nmi(vcpu);
3020 	vcpu->arch.exception.pending = events->exception.injected;
3021 	vcpu->arch.exception.nr = events->exception.nr;
3022 	vcpu->arch.exception.has_error_code = events->exception.has_error_code;
3023 	vcpu->arch.exception.error_code = events->exception.error_code;
3024 
3025 	vcpu->arch.interrupt.pending = events->interrupt.injected;
3026 	vcpu->arch.interrupt.nr = events->interrupt.nr;
3027 	vcpu->arch.interrupt.soft = events->interrupt.soft;
3028 	if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
3029 		kvm_x86_ops->set_interrupt_shadow(vcpu,
3030 						  events->interrupt.shadow);
3031 
3032 	vcpu->arch.nmi_injected = events->nmi.injected;
3033 	if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
3034 		vcpu->arch.nmi_pending = events->nmi.pending;
3035 	kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
3036 
3037 	if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
3038 	    kvm_vcpu_has_lapic(vcpu))
3039 		vcpu->arch.apic->sipi_vector = events->sipi_vector;
3040 
3041 	kvm_make_request(KVM_REQ_EVENT, vcpu);
3042 
3043 	return 0;
3044 }
3045 
3046 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3047 					     struct kvm_debugregs *dbgregs)
3048 {
3049 	unsigned long val;
3050 
3051 	memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
3052 	_kvm_get_dr(vcpu, 6, &val);
3053 	dbgregs->dr6 = val;
3054 	dbgregs->dr7 = vcpu->arch.dr7;
3055 	dbgregs->flags = 0;
3056 	memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
3057 }
3058 
3059 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
3060 					    struct kvm_debugregs *dbgregs)
3061 {
3062 	if (dbgregs->flags)
3063 		return -EINVAL;
3064 
3065 	memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
3066 	vcpu->arch.dr6 = dbgregs->dr6;
3067 	kvm_update_dr6(vcpu);
3068 	vcpu->arch.dr7 = dbgregs->dr7;
3069 	kvm_update_dr7(vcpu);
3070 
3071 	return 0;
3072 }
3073 
3074 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
3075 					 struct kvm_xsave *guest_xsave)
3076 {
3077 	if (cpu_has_xsave) {
3078 		memcpy(guest_xsave->region,
3079 			&vcpu->arch.guest_fpu.state->xsave,
3080 			vcpu->arch.guest_xstate_size);
3081 		*(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] &=
3082 			vcpu->arch.guest_supported_xcr0 | XSTATE_FPSSE;
3083 	} else {
3084 		memcpy(guest_xsave->region,
3085 			&vcpu->arch.guest_fpu.state->fxsave,
3086 			sizeof(struct i387_fxsave_struct));
3087 		*(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
3088 			XSTATE_FPSSE;
3089 	}
3090 }
3091 
3092 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
3093 					struct kvm_xsave *guest_xsave)
3094 {
3095 	u64 xstate_bv =
3096 		*(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
3097 
3098 	if (cpu_has_xsave) {
3099 		/*
3100 		 * Here we allow setting states that are not present in
3101 		 * CPUID leaf 0xD, index 0, EDX:EAX.  This is for compatibility
3102 		 * with old userspace.
3103 		 */
3104 		if (xstate_bv & ~kvm_supported_xcr0())
3105 			return -EINVAL;
3106 		memcpy(&vcpu->arch.guest_fpu.state->xsave,
3107 			guest_xsave->region, vcpu->arch.guest_xstate_size);
3108 	} else {
3109 		if (xstate_bv & ~XSTATE_FPSSE)
3110 			return -EINVAL;
3111 		memcpy(&vcpu->arch.guest_fpu.state->fxsave,
3112 			guest_xsave->region, sizeof(struct i387_fxsave_struct));
3113 	}
3114 	return 0;
3115 }
3116 
3117 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3118 					struct kvm_xcrs *guest_xcrs)
3119 {
3120 	if (!cpu_has_xsave) {
3121 		guest_xcrs->nr_xcrs = 0;
3122 		return;
3123 	}
3124 
3125 	guest_xcrs->nr_xcrs = 1;
3126 	guest_xcrs->flags = 0;
3127 	guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3128 	guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3129 }
3130 
3131 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3132 				       struct kvm_xcrs *guest_xcrs)
3133 {
3134 	int i, r = 0;
3135 
3136 	if (!cpu_has_xsave)
3137 		return -EINVAL;
3138 
3139 	if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
3140 		return -EINVAL;
3141 
3142 	for (i = 0; i < guest_xcrs->nr_xcrs; i++)
3143 		/* Only support XCR0 currently */
3144 		if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
3145 			r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
3146 				guest_xcrs->xcrs[i].value);
3147 			break;
3148 		}
3149 	if (r)
3150 		r = -EINVAL;
3151 	return r;
3152 }
3153 
3154 /*
3155  * kvm_set_guest_paused() indicates to the guest kernel that it has been
3156  * stopped by the hypervisor.  This function will be called from the host only.
3157  * EINVAL is returned when the host attempts to set the flag for a guest that
3158  * does not support pv clocks.
3159  */
3160 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
3161 {
3162 	if (!vcpu->arch.pv_time_enabled)
3163 		return -EINVAL;
3164 	vcpu->arch.pvclock_set_guest_stopped_request = true;
3165 	kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3166 	return 0;
3167 }
3168 
3169 long kvm_arch_vcpu_ioctl(struct file *filp,
3170 			 unsigned int ioctl, unsigned long arg)
3171 {
3172 	struct kvm_vcpu *vcpu = filp->private_data;
3173 	void __user *argp = (void __user *)arg;
3174 	int r;
3175 	union {
3176 		struct kvm_lapic_state *lapic;
3177 		struct kvm_xsave *xsave;
3178 		struct kvm_xcrs *xcrs;
3179 		void *buffer;
3180 	} u;
3181 
3182 	u.buffer = NULL;
3183 	switch (ioctl) {
3184 	case KVM_GET_LAPIC: {
3185 		r = -EINVAL;
3186 		if (!vcpu->arch.apic)
3187 			goto out;
3188 		u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
3189 
3190 		r = -ENOMEM;
3191 		if (!u.lapic)
3192 			goto out;
3193 		r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
3194 		if (r)
3195 			goto out;
3196 		r = -EFAULT;
3197 		if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
3198 			goto out;
3199 		r = 0;
3200 		break;
3201 	}
3202 	case KVM_SET_LAPIC: {
3203 		r = -EINVAL;
3204 		if (!vcpu->arch.apic)
3205 			goto out;
3206 		u.lapic = memdup_user(argp, sizeof(*u.lapic));
3207 		if (IS_ERR(u.lapic))
3208 			return PTR_ERR(u.lapic);
3209 
3210 		r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
3211 		break;
3212 	}
3213 	case KVM_INTERRUPT: {
3214 		struct kvm_interrupt irq;
3215 
3216 		r = -EFAULT;
3217 		if (copy_from_user(&irq, argp, sizeof irq))
3218 			goto out;
3219 		r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
3220 		break;
3221 	}
3222 	case KVM_NMI: {
3223 		r = kvm_vcpu_ioctl_nmi(vcpu);
3224 		break;
3225 	}
3226 	case KVM_SET_CPUID: {
3227 		struct kvm_cpuid __user *cpuid_arg = argp;
3228 		struct kvm_cpuid cpuid;
3229 
3230 		r = -EFAULT;
3231 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3232 			goto out;
3233 		r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3234 		break;
3235 	}
3236 	case KVM_SET_CPUID2: {
3237 		struct kvm_cpuid2 __user *cpuid_arg = argp;
3238 		struct kvm_cpuid2 cpuid;
3239 
3240 		r = -EFAULT;
3241 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3242 			goto out;
3243 		r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
3244 					      cpuid_arg->entries);
3245 		break;
3246 	}
3247 	case KVM_GET_CPUID2: {
3248 		struct kvm_cpuid2 __user *cpuid_arg = argp;
3249 		struct kvm_cpuid2 cpuid;
3250 
3251 		r = -EFAULT;
3252 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3253 			goto out;
3254 		r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
3255 					      cpuid_arg->entries);
3256 		if (r)
3257 			goto out;
3258 		r = -EFAULT;
3259 		if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3260 			goto out;
3261 		r = 0;
3262 		break;
3263 	}
3264 	case KVM_GET_MSRS:
3265 		r = msr_io(vcpu, argp, kvm_get_msr, 1);
3266 		break;
3267 	case KVM_SET_MSRS:
3268 		r = msr_io(vcpu, argp, do_set_msr, 0);
3269 		break;
3270 	case KVM_TPR_ACCESS_REPORTING: {
3271 		struct kvm_tpr_access_ctl tac;
3272 
3273 		r = -EFAULT;
3274 		if (copy_from_user(&tac, argp, sizeof tac))
3275 			goto out;
3276 		r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3277 		if (r)
3278 			goto out;
3279 		r = -EFAULT;
3280 		if (copy_to_user(argp, &tac, sizeof tac))
3281 			goto out;
3282 		r = 0;
3283 		break;
3284 	};
3285 	case KVM_SET_VAPIC_ADDR: {
3286 		struct kvm_vapic_addr va;
3287 
3288 		r = -EINVAL;
3289 		if (!irqchip_in_kernel(vcpu->kvm))
3290 			goto out;
3291 		r = -EFAULT;
3292 		if (copy_from_user(&va, argp, sizeof va))
3293 			goto out;
3294 		r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
3295 		break;
3296 	}
3297 	case KVM_X86_SETUP_MCE: {
3298 		u64 mcg_cap;
3299 
3300 		r = -EFAULT;
3301 		if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
3302 			goto out;
3303 		r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3304 		break;
3305 	}
3306 	case KVM_X86_SET_MCE: {
3307 		struct kvm_x86_mce mce;
3308 
3309 		r = -EFAULT;
3310 		if (copy_from_user(&mce, argp, sizeof mce))
3311 			goto out;
3312 		r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3313 		break;
3314 	}
3315 	case KVM_GET_VCPU_EVENTS: {
3316 		struct kvm_vcpu_events events;
3317 
3318 		kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3319 
3320 		r = -EFAULT;
3321 		if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3322 			break;
3323 		r = 0;
3324 		break;
3325 	}
3326 	case KVM_SET_VCPU_EVENTS: {
3327 		struct kvm_vcpu_events events;
3328 
3329 		r = -EFAULT;
3330 		if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3331 			break;
3332 
3333 		r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3334 		break;
3335 	}
3336 	case KVM_GET_DEBUGREGS: {
3337 		struct kvm_debugregs dbgregs;
3338 
3339 		kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3340 
3341 		r = -EFAULT;
3342 		if (copy_to_user(argp, &dbgregs,
3343 				 sizeof(struct kvm_debugregs)))
3344 			break;
3345 		r = 0;
3346 		break;
3347 	}
3348 	case KVM_SET_DEBUGREGS: {
3349 		struct kvm_debugregs dbgregs;
3350 
3351 		r = -EFAULT;
3352 		if (copy_from_user(&dbgregs, argp,
3353 				   sizeof(struct kvm_debugregs)))
3354 			break;
3355 
3356 		r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3357 		break;
3358 	}
3359 	case KVM_GET_XSAVE: {
3360 		u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3361 		r = -ENOMEM;
3362 		if (!u.xsave)
3363 			break;
3364 
3365 		kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
3366 
3367 		r = -EFAULT;
3368 		if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
3369 			break;
3370 		r = 0;
3371 		break;
3372 	}
3373 	case KVM_SET_XSAVE: {
3374 		u.xsave = memdup_user(argp, sizeof(*u.xsave));
3375 		if (IS_ERR(u.xsave))
3376 			return PTR_ERR(u.xsave);
3377 
3378 		r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
3379 		break;
3380 	}
3381 	case KVM_GET_XCRS: {
3382 		u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3383 		r = -ENOMEM;
3384 		if (!u.xcrs)
3385 			break;
3386 
3387 		kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
3388 
3389 		r = -EFAULT;
3390 		if (copy_to_user(argp, u.xcrs,
3391 				 sizeof(struct kvm_xcrs)))
3392 			break;
3393 		r = 0;
3394 		break;
3395 	}
3396 	case KVM_SET_XCRS: {
3397 		u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
3398 		if (IS_ERR(u.xcrs))
3399 			return PTR_ERR(u.xcrs);
3400 
3401 		r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3402 		break;
3403 	}
3404 	case KVM_SET_TSC_KHZ: {
3405 		u32 user_tsc_khz;
3406 
3407 		r = -EINVAL;
3408 		user_tsc_khz = (u32)arg;
3409 
3410 		if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3411 			goto out;
3412 
3413 		if (user_tsc_khz == 0)
3414 			user_tsc_khz = tsc_khz;
3415 
3416 		kvm_set_tsc_khz(vcpu, user_tsc_khz);
3417 
3418 		r = 0;
3419 		goto out;
3420 	}
3421 	case KVM_GET_TSC_KHZ: {
3422 		r = vcpu->arch.virtual_tsc_khz;
3423 		goto out;
3424 	}
3425 	case KVM_KVMCLOCK_CTRL: {
3426 		r = kvm_set_guest_paused(vcpu);
3427 		goto out;
3428 	}
3429 	default:
3430 		r = -EINVAL;
3431 	}
3432 out:
3433 	kfree(u.buffer);
3434 	return r;
3435 }
3436 
3437 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
3438 {
3439 	return VM_FAULT_SIGBUS;
3440 }
3441 
3442 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3443 {
3444 	int ret;
3445 
3446 	if (addr > (unsigned int)(-3 * PAGE_SIZE))
3447 		return -EINVAL;
3448 	ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3449 	return ret;
3450 }
3451 
3452 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3453 					      u64 ident_addr)
3454 {
3455 	kvm->arch.ept_identity_map_addr = ident_addr;
3456 	return 0;
3457 }
3458 
3459 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3460 					  u32 kvm_nr_mmu_pages)
3461 {
3462 	if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3463 		return -EINVAL;
3464 
3465 	mutex_lock(&kvm->slots_lock);
3466 
3467 	kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3468 	kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3469 
3470 	mutex_unlock(&kvm->slots_lock);
3471 	return 0;
3472 }
3473 
3474 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3475 {
3476 	return kvm->arch.n_max_mmu_pages;
3477 }
3478 
3479 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3480 {
3481 	int r;
3482 
3483 	r = 0;
3484 	switch (chip->chip_id) {
3485 	case KVM_IRQCHIP_PIC_MASTER:
3486 		memcpy(&chip->chip.pic,
3487 			&pic_irqchip(kvm)->pics[0],
3488 			sizeof(struct kvm_pic_state));
3489 		break;
3490 	case KVM_IRQCHIP_PIC_SLAVE:
3491 		memcpy(&chip->chip.pic,
3492 			&pic_irqchip(kvm)->pics[1],
3493 			sizeof(struct kvm_pic_state));
3494 		break;
3495 	case KVM_IRQCHIP_IOAPIC:
3496 		r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
3497 		break;
3498 	default:
3499 		r = -EINVAL;
3500 		break;
3501 	}
3502 	return r;
3503 }
3504 
3505 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3506 {
3507 	int r;
3508 
3509 	r = 0;
3510 	switch (chip->chip_id) {
3511 	case KVM_IRQCHIP_PIC_MASTER:
3512 		spin_lock(&pic_irqchip(kvm)->lock);
3513 		memcpy(&pic_irqchip(kvm)->pics[0],
3514 			&chip->chip.pic,
3515 			sizeof(struct kvm_pic_state));
3516 		spin_unlock(&pic_irqchip(kvm)->lock);
3517 		break;
3518 	case KVM_IRQCHIP_PIC_SLAVE:
3519 		spin_lock(&pic_irqchip(kvm)->lock);
3520 		memcpy(&pic_irqchip(kvm)->pics[1],
3521 			&chip->chip.pic,
3522 			sizeof(struct kvm_pic_state));
3523 		spin_unlock(&pic_irqchip(kvm)->lock);
3524 		break;
3525 	case KVM_IRQCHIP_IOAPIC:
3526 		r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
3527 		break;
3528 	default:
3529 		r = -EINVAL;
3530 		break;
3531 	}
3532 	kvm_pic_update_irq(pic_irqchip(kvm));
3533 	return r;
3534 }
3535 
3536 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3537 {
3538 	int r = 0;
3539 
3540 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
3541 	memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
3542 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3543 	return r;
3544 }
3545 
3546 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3547 {
3548 	int r = 0;
3549 
3550 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
3551 	memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
3552 	kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
3553 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3554 	return r;
3555 }
3556 
3557 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3558 {
3559 	int r = 0;
3560 
3561 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
3562 	memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3563 		sizeof(ps->channels));
3564 	ps->flags = kvm->arch.vpit->pit_state.flags;
3565 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3566 	memset(&ps->reserved, 0, sizeof(ps->reserved));
3567 	return r;
3568 }
3569 
3570 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3571 {
3572 	int r = 0, start = 0;
3573 	u32 prev_legacy, cur_legacy;
3574 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
3575 	prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3576 	cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3577 	if (!prev_legacy && cur_legacy)
3578 		start = 1;
3579 	memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
3580 	       sizeof(kvm->arch.vpit->pit_state.channels));
3581 	kvm->arch.vpit->pit_state.flags = ps->flags;
3582 	kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
3583 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3584 	return r;
3585 }
3586 
3587 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3588 				 struct kvm_reinject_control *control)
3589 {
3590 	if (!kvm->arch.vpit)
3591 		return -ENXIO;
3592 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
3593 	kvm->arch.vpit->pit_state.reinject = control->pit_reinject;
3594 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3595 	return 0;
3596 }
3597 
3598 /**
3599  * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
3600  * @kvm: kvm instance
3601  * @log: slot id and address to which we copy the log
3602  *
3603  * We need to keep it in mind that VCPU threads can write to the bitmap
3604  * concurrently.  So, to avoid losing data, we keep the following order for
3605  * each bit:
3606  *
3607  *   1. Take a snapshot of the bit and clear it if needed.
3608  *   2. Write protect the corresponding page.
3609  *   3. Flush TLB's if needed.
3610  *   4. Copy the snapshot to the userspace.
3611  *
3612  * Between 2 and 3, the guest may write to the page using the remaining TLB
3613  * entry.  This is not a problem because the page will be reported dirty at
3614  * step 4 using the snapshot taken before and step 3 ensures that successive
3615  * writes will be logged for the next call.
3616  */
3617 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
3618 {
3619 	int r;
3620 	struct kvm_memory_slot *memslot;
3621 	unsigned long n, i;
3622 	unsigned long *dirty_bitmap;
3623 	unsigned long *dirty_bitmap_buffer;
3624 	bool is_dirty = false;
3625 
3626 	mutex_lock(&kvm->slots_lock);
3627 
3628 	r = -EINVAL;
3629 	if (log->slot >= KVM_USER_MEM_SLOTS)
3630 		goto out;
3631 
3632 	memslot = id_to_memslot(kvm->memslots, log->slot);
3633 
3634 	dirty_bitmap = memslot->dirty_bitmap;
3635 	r = -ENOENT;
3636 	if (!dirty_bitmap)
3637 		goto out;
3638 
3639 	n = kvm_dirty_bitmap_bytes(memslot);
3640 
3641 	dirty_bitmap_buffer = dirty_bitmap + n / sizeof(long);
3642 	memset(dirty_bitmap_buffer, 0, n);
3643 
3644 	spin_lock(&kvm->mmu_lock);
3645 
3646 	for (i = 0; i < n / sizeof(long); i++) {
3647 		unsigned long mask;
3648 		gfn_t offset;
3649 
3650 		if (!dirty_bitmap[i])
3651 			continue;
3652 
3653 		is_dirty = true;
3654 
3655 		mask = xchg(&dirty_bitmap[i], 0);
3656 		dirty_bitmap_buffer[i] = mask;
3657 
3658 		offset = i * BITS_PER_LONG;
3659 		kvm_mmu_write_protect_pt_masked(kvm, memslot, offset, mask);
3660 	}
3661 
3662 	spin_unlock(&kvm->mmu_lock);
3663 
3664 	/* See the comments in kvm_mmu_slot_remove_write_access(). */
3665 	lockdep_assert_held(&kvm->slots_lock);
3666 
3667 	/*
3668 	 * All the TLBs can be flushed out of mmu lock, see the comments in
3669 	 * kvm_mmu_slot_remove_write_access().
3670 	 */
3671 	if (is_dirty)
3672 		kvm_flush_remote_tlbs(kvm);
3673 
3674 	r = -EFAULT;
3675 	if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
3676 		goto out;
3677 
3678 	r = 0;
3679 out:
3680 	mutex_unlock(&kvm->slots_lock);
3681 	return r;
3682 }
3683 
3684 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
3685 			bool line_status)
3686 {
3687 	if (!irqchip_in_kernel(kvm))
3688 		return -ENXIO;
3689 
3690 	irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3691 					irq_event->irq, irq_event->level,
3692 					line_status);
3693 	return 0;
3694 }
3695 
3696 long kvm_arch_vm_ioctl(struct file *filp,
3697 		       unsigned int ioctl, unsigned long arg)
3698 {
3699 	struct kvm *kvm = filp->private_data;
3700 	void __user *argp = (void __user *)arg;
3701 	int r = -ENOTTY;
3702 	/*
3703 	 * This union makes it completely explicit to gcc-3.x
3704 	 * that these two variables' stack usage should be
3705 	 * combined, not added together.
3706 	 */
3707 	union {
3708 		struct kvm_pit_state ps;
3709 		struct kvm_pit_state2 ps2;
3710 		struct kvm_pit_config pit_config;
3711 	} u;
3712 
3713 	switch (ioctl) {
3714 	case KVM_SET_TSS_ADDR:
3715 		r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
3716 		break;
3717 	case KVM_SET_IDENTITY_MAP_ADDR: {
3718 		u64 ident_addr;
3719 
3720 		r = -EFAULT;
3721 		if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3722 			goto out;
3723 		r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
3724 		break;
3725 	}
3726 	case KVM_SET_NR_MMU_PAGES:
3727 		r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
3728 		break;
3729 	case KVM_GET_NR_MMU_PAGES:
3730 		r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3731 		break;
3732 	case KVM_CREATE_IRQCHIP: {
3733 		struct kvm_pic *vpic;
3734 
3735 		mutex_lock(&kvm->lock);
3736 		r = -EEXIST;
3737 		if (kvm->arch.vpic)
3738 			goto create_irqchip_unlock;
3739 		r = -EINVAL;
3740 		if (atomic_read(&kvm->online_vcpus))
3741 			goto create_irqchip_unlock;
3742 		r = -ENOMEM;
3743 		vpic = kvm_create_pic(kvm);
3744 		if (vpic) {
3745 			r = kvm_ioapic_init(kvm);
3746 			if (r) {
3747 				mutex_lock(&kvm->slots_lock);
3748 				kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3749 							  &vpic->dev_master);
3750 				kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3751 							  &vpic->dev_slave);
3752 				kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3753 							  &vpic->dev_eclr);
3754 				mutex_unlock(&kvm->slots_lock);
3755 				kfree(vpic);
3756 				goto create_irqchip_unlock;
3757 			}
3758 		} else
3759 			goto create_irqchip_unlock;
3760 		smp_wmb();
3761 		kvm->arch.vpic = vpic;
3762 		smp_wmb();
3763 		r = kvm_setup_default_irq_routing(kvm);
3764 		if (r) {
3765 			mutex_lock(&kvm->slots_lock);
3766 			mutex_lock(&kvm->irq_lock);
3767 			kvm_ioapic_destroy(kvm);
3768 			kvm_destroy_pic(kvm);
3769 			mutex_unlock(&kvm->irq_lock);
3770 			mutex_unlock(&kvm->slots_lock);
3771 		}
3772 	create_irqchip_unlock:
3773 		mutex_unlock(&kvm->lock);
3774 		break;
3775 	}
3776 	case KVM_CREATE_PIT:
3777 		u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
3778 		goto create_pit;
3779 	case KVM_CREATE_PIT2:
3780 		r = -EFAULT;
3781 		if (copy_from_user(&u.pit_config, argp,
3782 				   sizeof(struct kvm_pit_config)))
3783 			goto out;
3784 	create_pit:
3785 		mutex_lock(&kvm->slots_lock);
3786 		r = -EEXIST;
3787 		if (kvm->arch.vpit)
3788 			goto create_pit_unlock;
3789 		r = -ENOMEM;
3790 		kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
3791 		if (kvm->arch.vpit)
3792 			r = 0;
3793 	create_pit_unlock:
3794 		mutex_unlock(&kvm->slots_lock);
3795 		break;
3796 	case KVM_GET_IRQCHIP: {
3797 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3798 		struct kvm_irqchip *chip;
3799 
3800 		chip = memdup_user(argp, sizeof(*chip));
3801 		if (IS_ERR(chip)) {
3802 			r = PTR_ERR(chip);
3803 			goto out;
3804 		}
3805 
3806 		r = -ENXIO;
3807 		if (!irqchip_in_kernel(kvm))
3808 			goto get_irqchip_out;
3809 		r = kvm_vm_ioctl_get_irqchip(kvm, chip);
3810 		if (r)
3811 			goto get_irqchip_out;
3812 		r = -EFAULT;
3813 		if (copy_to_user(argp, chip, sizeof *chip))
3814 			goto get_irqchip_out;
3815 		r = 0;
3816 	get_irqchip_out:
3817 		kfree(chip);
3818 		break;
3819 	}
3820 	case KVM_SET_IRQCHIP: {
3821 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3822 		struct kvm_irqchip *chip;
3823 
3824 		chip = memdup_user(argp, sizeof(*chip));
3825 		if (IS_ERR(chip)) {
3826 			r = PTR_ERR(chip);
3827 			goto out;
3828 		}
3829 
3830 		r = -ENXIO;
3831 		if (!irqchip_in_kernel(kvm))
3832 			goto set_irqchip_out;
3833 		r = kvm_vm_ioctl_set_irqchip(kvm, chip);
3834 		if (r)
3835 			goto set_irqchip_out;
3836 		r = 0;
3837 	set_irqchip_out:
3838 		kfree(chip);
3839 		break;
3840 	}
3841 	case KVM_GET_PIT: {
3842 		r = -EFAULT;
3843 		if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
3844 			goto out;
3845 		r = -ENXIO;
3846 		if (!kvm->arch.vpit)
3847 			goto out;
3848 		r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
3849 		if (r)
3850 			goto out;
3851 		r = -EFAULT;
3852 		if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
3853 			goto out;
3854 		r = 0;
3855 		break;
3856 	}
3857 	case KVM_SET_PIT: {
3858 		r = -EFAULT;
3859 		if (copy_from_user(&u.ps, argp, sizeof u.ps))
3860 			goto out;
3861 		r = -ENXIO;
3862 		if (!kvm->arch.vpit)
3863 			goto out;
3864 		r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
3865 		break;
3866 	}
3867 	case KVM_GET_PIT2: {
3868 		r = -ENXIO;
3869 		if (!kvm->arch.vpit)
3870 			goto out;
3871 		r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
3872 		if (r)
3873 			goto out;
3874 		r = -EFAULT;
3875 		if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
3876 			goto out;
3877 		r = 0;
3878 		break;
3879 	}
3880 	case KVM_SET_PIT2: {
3881 		r = -EFAULT;
3882 		if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
3883 			goto out;
3884 		r = -ENXIO;
3885 		if (!kvm->arch.vpit)
3886 			goto out;
3887 		r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
3888 		break;
3889 	}
3890 	case KVM_REINJECT_CONTROL: {
3891 		struct kvm_reinject_control control;
3892 		r =  -EFAULT;
3893 		if (copy_from_user(&control, argp, sizeof(control)))
3894 			goto out;
3895 		r = kvm_vm_ioctl_reinject(kvm, &control);
3896 		break;
3897 	}
3898 	case KVM_XEN_HVM_CONFIG: {
3899 		r = -EFAULT;
3900 		if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
3901 				   sizeof(struct kvm_xen_hvm_config)))
3902 			goto out;
3903 		r = -EINVAL;
3904 		if (kvm->arch.xen_hvm_config.flags)
3905 			goto out;
3906 		r = 0;
3907 		break;
3908 	}
3909 	case KVM_SET_CLOCK: {
3910 		struct kvm_clock_data user_ns;
3911 		u64 now_ns;
3912 		s64 delta;
3913 
3914 		r = -EFAULT;
3915 		if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
3916 			goto out;
3917 
3918 		r = -EINVAL;
3919 		if (user_ns.flags)
3920 			goto out;
3921 
3922 		r = 0;
3923 		local_irq_disable();
3924 		now_ns = get_kernel_ns();
3925 		delta = user_ns.clock - now_ns;
3926 		local_irq_enable();
3927 		kvm->arch.kvmclock_offset = delta;
3928 		kvm_gen_update_masterclock(kvm);
3929 		break;
3930 	}
3931 	case KVM_GET_CLOCK: {
3932 		struct kvm_clock_data user_ns;
3933 		u64 now_ns;
3934 
3935 		local_irq_disable();
3936 		now_ns = get_kernel_ns();
3937 		user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
3938 		local_irq_enable();
3939 		user_ns.flags = 0;
3940 		memset(&user_ns.pad, 0, sizeof(user_ns.pad));
3941 
3942 		r = -EFAULT;
3943 		if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
3944 			goto out;
3945 		r = 0;
3946 		break;
3947 	}
3948 
3949 	default:
3950 		;
3951 	}
3952 out:
3953 	return r;
3954 }
3955 
3956 static void kvm_init_msr_list(void)
3957 {
3958 	u32 dummy[2];
3959 	unsigned i, j;
3960 
3961 	/* skip the first msrs in the list. KVM-specific */
3962 	for (i = j = KVM_SAVE_MSRS_BEGIN; i < ARRAY_SIZE(msrs_to_save); i++) {
3963 		if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
3964 			continue;
3965 
3966 		/*
3967 		 * Even MSRs that are valid in the host may not be exposed
3968 		 * to the guests in some cases.  We could work around this
3969 		 * in VMX with the generic MSR save/load machinery, but it
3970 		 * is not really worthwhile since it will really only
3971 		 * happen with nested virtualization.
3972 		 */
3973 		switch (msrs_to_save[i]) {
3974 		case MSR_IA32_BNDCFGS:
3975 			if (!kvm_x86_ops->mpx_supported())
3976 				continue;
3977 			break;
3978 		default:
3979 			break;
3980 		}
3981 
3982 		if (j < i)
3983 			msrs_to_save[j] = msrs_to_save[i];
3984 		j++;
3985 	}
3986 	num_msrs_to_save = j;
3987 }
3988 
3989 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
3990 			   const void *v)
3991 {
3992 	int handled = 0;
3993 	int n;
3994 
3995 	do {
3996 		n = min(len, 8);
3997 		if (!(vcpu->arch.apic &&
3998 		      !kvm_iodevice_write(&vcpu->arch.apic->dev, addr, n, v))
3999 		    && kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
4000 			break;
4001 		handled += n;
4002 		addr += n;
4003 		len -= n;
4004 		v += n;
4005 	} while (len);
4006 
4007 	return handled;
4008 }
4009 
4010 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
4011 {
4012 	int handled = 0;
4013 	int n;
4014 
4015 	do {
4016 		n = min(len, 8);
4017 		if (!(vcpu->arch.apic &&
4018 		      !kvm_iodevice_read(&vcpu->arch.apic->dev, addr, n, v))
4019 		    && kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
4020 			break;
4021 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
4022 		handled += n;
4023 		addr += n;
4024 		len -= n;
4025 		v += n;
4026 	} while (len);
4027 
4028 	return handled;
4029 }
4030 
4031 static void kvm_set_segment(struct kvm_vcpu *vcpu,
4032 			struct kvm_segment *var, int seg)
4033 {
4034 	kvm_x86_ops->set_segment(vcpu, var, seg);
4035 }
4036 
4037 void kvm_get_segment(struct kvm_vcpu *vcpu,
4038 		     struct kvm_segment *var, int seg)
4039 {
4040 	kvm_x86_ops->get_segment(vcpu, var, seg);
4041 }
4042 
4043 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access)
4044 {
4045 	gpa_t t_gpa;
4046 	struct x86_exception exception;
4047 
4048 	BUG_ON(!mmu_is_nested(vcpu));
4049 
4050 	/* NPT walks are always user-walks */
4051 	access |= PFERR_USER_MASK;
4052 	t_gpa  = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, &exception);
4053 
4054 	return t_gpa;
4055 }
4056 
4057 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
4058 			      struct x86_exception *exception)
4059 {
4060 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4061 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4062 }
4063 
4064  gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
4065 				struct x86_exception *exception)
4066 {
4067 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4068 	access |= PFERR_FETCH_MASK;
4069 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4070 }
4071 
4072 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
4073 			       struct x86_exception *exception)
4074 {
4075 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4076 	access |= PFERR_WRITE_MASK;
4077 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4078 }
4079 
4080 /* uses this to access any guest's mapped memory without checking CPL */
4081 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
4082 				struct x86_exception *exception)
4083 {
4084 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
4085 }
4086 
4087 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
4088 				      struct kvm_vcpu *vcpu, u32 access,
4089 				      struct x86_exception *exception)
4090 {
4091 	void *data = val;
4092 	int r = X86EMUL_CONTINUE;
4093 
4094 	while (bytes) {
4095 		gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
4096 							    exception);
4097 		unsigned offset = addr & (PAGE_SIZE-1);
4098 		unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
4099 		int ret;
4100 
4101 		if (gpa == UNMAPPED_GVA)
4102 			return X86EMUL_PROPAGATE_FAULT;
4103 		ret = kvm_read_guest_page(vcpu->kvm, gpa >> PAGE_SHIFT, data,
4104 					  offset, toread);
4105 		if (ret < 0) {
4106 			r = X86EMUL_IO_NEEDED;
4107 			goto out;
4108 		}
4109 
4110 		bytes -= toread;
4111 		data += toread;
4112 		addr += toread;
4113 	}
4114 out:
4115 	return r;
4116 }
4117 
4118 /* used for instruction fetching */
4119 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
4120 				gva_t addr, void *val, unsigned int bytes,
4121 				struct x86_exception *exception)
4122 {
4123 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4124 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4125 	unsigned offset;
4126 	int ret;
4127 
4128 	/* Inline kvm_read_guest_virt_helper for speed.  */
4129 	gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
4130 						    exception);
4131 	if (unlikely(gpa == UNMAPPED_GVA))
4132 		return X86EMUL_PROPAGATE_FAULT;
4133 
4134 	offset = addr & (PAGE_SIZE-1);
4135 	if (WARN_ON(offset + bytes > PAGE_SIZE))
4136 		bytes = (unsigned)PAGE_SIZE - offset;
4137 	ret = kvm_read_guest_page(vcpu->kvm, gpa >> PAGE_SHIFT, val,
4138 				  offset, bytes);
4139 	if (unlikely(ret < 0))
4140 		return X86EMUL_IO_NEEDED;
4141 
4142 	return X86EMUL_CONTINUE;
4143 }
4144 
4145 int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
4146 			       gva_t addr, void *val, unsigned int bytes,
4147 			       struct x86_exception *exception)
4148 {
4149 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4150 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4151 
4152 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
4153 					  exception);
4154 }
4155 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
4156 
4157 static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4158 				      gva_t addr, void *val, unsigned int bytes,
4159 				      struct x86_exception *exception)
4160 {
4161 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4162 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
4163 }
4164 
4165 int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4166 				       gva_t addr, void *val,
4167 				       unsigned int bytes,
4168 				       struct x86_exception *exception)
4169 {
4170 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4171 	void *data = val;
4172 	int r = X86EMUL_CONTINUE;
4173 
4174 	while (bytes) {
4175 		gpa_t gpa =  vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
4176 							     PFERR_WRITE_MASK,
4177 							     exception);
4178 		unsigned offset = addr & (PAGE_SIZE-1);
4179 		unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
4180 		int ret;
4181 
4182 		if (gpa == UNMAPPED_GVA)
4183 			return X86EMUL_PROPAGATE_FAULT;
4184 		ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
4185 		if (ret < 0) {
4186 			r = X86EMUL_IO_NEEDED;
4187 			goto out;
4188 		}
4189 
4190 		bytes -= towrite;
4191 		data += towrite;
4192 		addr += towrite;
4193 	}
4194 out:
4195 	return r;
4196 }
4197 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
4198 
4199 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4200 				gpa_t *gpa, struct x86_exception *exception,
4201 				bool write)
4202 {
4203 	u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
4204 		| (write ? PFERR_WRITE_MASK : 0);
4205 
4206 	if (vcpu_match_mmio_gva(vcpu, gva)
4207 	    && !permission_fault(vcpu, vcpu->arch.walk_mmu,
4208 				 vcpu->arch.access, access)) {
4209 		*gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
4210 					(gva & (PAGE_SIZE - 1));
4211 		trace_vcpu_match_mmio(gva, *gpa, write, false);
4212 		return 1;
4213 	}
4214 
4215 	*gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4216 
4217 	if (*gpa == UNMAPPED_GVA)
4218 		return -1;
4219 
4220 	/* For APIC access vmexit */
4221 	if ((*gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4222 		return 1;
4223 
4224 	if (vcpu_match_mmio_gpa(vcpu, *gpa)) {
4225 		trace_vcpu_match_mmio(gva, *gpa, write, true);
4226 		return 1;
4227 	}
4228 
4229 	return 0;
4230 }
4231 
4232 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
4233 			const void *val, int bytes)
4234 {
4235 	int ret;
4236 
4237 	ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
4238 	if (ret < 0)
4239 		return 0;
4240 	kvm_mmu_pte_write(vcpu, gpa, val, bytes);
4241 	return 1;
4242 }
4243 
4244 struct read_write_emulator_ops {
4245 	int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
4246 				  int bytes);
4247 	int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
4248 				  void *val, int bytes);
4249 	int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4250 			       int bytes, void *val);
4251 	int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4252 				    void *val, int bytes);
4253 	bool write;
4254 };
4255 
4256 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
4257 {
4258 	if (vcpu->mmio_read_completed) {
4259 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
4260 			       vcpu->mmio_fragments[0].gpa, *(u64 *)val);
4261 		vcpu->mmio_read_completed = 0;
4262 		return 1;
4263 	}
4264 
4265 	return 0;
4266 }
4267 
4268 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4269 			void *val, int bytes)
4270 {
4271 	return !kvm_read_guest(vcpu->kvm, gpa, val, bytes);
4272 }
4273 
4274 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4275 			 void *val, int bytes)
4276 {
4277 	return emulator_write_phys(vcpu, gpa, val, bytes);
4278 }
4279 
4280 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
4281 {
4282 	trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
4283 	return vcpu_mmio_write(vcpu, gpa, bytes, val);
4284 }
4285 
4286 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4287 			  void *val, int bytes)
4288 {
4289 	trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
4290 	return X86EMUL_IO_NEEDED;
4291 }
4292 
4293 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4294 			   void *val, int bytes)
4295 {
4296 	struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
4297 
4298 	memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
4299 	return X86EMUL_CONTINUE;
4300 }
4301 
4302 static const struct read_write_emulator_ops read_emultor = {
4303 	.read_write_prepare = read_prepare,
4304 	.read_write_emulate = read_emulate,
4305 	.read_write_mmio = vcpu_mmio_read,
4306 	.read_write_exit_mmio = read_exit_mmio,
4307 };
4308 
4309 static const struct read_write_emulator_ops write_emultor = {
4310 	.read_write_emulate = write_emulate,
4311 	.read_write_mmio = write_mmio,
4312 	.read_write_exit_mmio = write_exit_mmio,
4313 	.write = true,
4314 };
4315 
4316 static int emulator_read_write_onepage(unsigned long addr, void *val,
4317 				       unsigned int bytes,
4318 				       struct x86_exception *exception,
4319 				       struct kvm_vcpu *vcpu,
4320 				       const struct read_write_emulator_ops *ops)
4321 {
4322 	gpa_t gpa;
4323 	int handled, ret;
4324 	bool write = ops->write;
4325 	struct kvm_mmio_fragment *frag;
4326 
4327 	ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
4328 
4329 	if (ret < 0)
4330 		return X86EMUL_PROPAGATE_FAULT;
4331 
4332 	/* For APIC access vmexit */
4333 	if (ret)
4334 		goto mmio;
4335 
4336 	if (ops->read_write_emulate(vcpu, gpa, val, bytes))
4337 		return X86EMUL_CONTINUE;
4338 
4339 mmio:
4340 	/*
4341 	 * Is this MMIO handled locally?
4342 	 */
4343 	handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
4344 	if (handled == bytes)
4345 		return X86EMUL_CONTINUE;
4346 
4347 	gpa += handled;
4348 	bytes -= handled;
4349 	val += handled;
4350 
4351 	WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
4352 	frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
4353 	frag->gpa = gpa;
4354 	frag->data = val;
4355 	frag->len = bytes;
4356 	return X86EMUL_CONTINUE;
4357 }
4358 
4359 int emulator_read_write(struct x86_emulate_ctxt *ctxt, unsigned long addr,
4360 			void *val, unsigned int bytes,
4361 			struct x86_exception *exception,
4362 			const struct read_write_emulator_ops *ops)
4363 {
4364 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4365 	gpa_t gpa;
4366 	int rc;
4367 
4368 	if (ops->read_write_prepare &&
4369 		  ops->read_write_prepare(vcpu, val, bytes))
4370 		return X86EMUL_CONTINUE;
4371 
4372 	vcpu->mmio_nr_fragments = 0;
4373 
4374 	/* Crossing a page boundary? */
4375 	if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
4376 		int now;
4377 
4378 		now = -addr & ~PAGE_MASK;
4379 		rc = emulator_read_write_onepage(addr, val, now, exception,
4380 						 vcpu, ops);
4381 
4382 		if (rc != X86EMUL_CONTINUE)
4383 			return rc;
4384 		addr += now;
4385 		val += now;
4386 		bytes -= now;
4387 	}
4388 
4389 	rc = emulator_read_write_onepage(addr, val, bytes, exception,
4390 					 vcpu, ops);
4391 	if (rc != X86EMUL_CONTINUE)
4392 		return rc;
4393 
4394 	if (!vcpu->mmio_nr_fragments)
4395 		return rc;
4396 
4397 	gpa = vcpu->mmio_fragments[0].gpa;
4398 
4399 	vcpu->mmio_needed = 1;
4400 	vcpu->mmio_cur_fragment = 0;
4401 
4402 	vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
4403 	vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
4404 	vcpu->run->exit_reason = KVM_EXIT_MMIO;
4405 	vcpu->run->mmio.phys_addr = gpa;
4406 
4407 	return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
4408 }
4409 
4410 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
4411 				  unsigned long addr,
4412 				  void *val,
4413 				  unsigned int bytes,
4414 				  struct x86_exception *exception)
4415 {
4416 	return emulator_read_write(ctxt, addr, val, bytes,
4417 				   exception, &read_emultor);
4418 }
4419 
4420 int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
4421 			    unsigned long addr,
4422 			    const void *val,
4423 			    unsigned int bytes,
4424 			    struct x86_exception *exception)
4425 {
4426 	return emulator_read_write(ctxt, addr, (void *)val, bytes,
4427 				   exception, &write_emultor);
4428 }
4429 
4430 #define CMPXCHG_TYPE(t, ptr, old, new) \
4431 	(cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4432 
4433 #ifdef CONFIG_X86_64
4434 #  define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4435 #else
4436 #  define CMPXCHG64(ptr, old, new) \
4437 	(cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
4438 #endif
4439 
4440 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
4441 				     unsigned long addr,
4442 				     const void *old,
4443 				     const void *new,
4444 				     unsigned int bytes,
4445 				     struct x86_exception *exception)
4446 {
4447 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4448 	gpa_t gpa;
4449 	struct page *page;
4450 	char *kaddr;
4451 	bool exchanged;
4452 
4453 	/* guests cmpxchg8b have to be emulated atomically */
4454 	if (bytes > 8 || (bytes & (bytes - 1)))
4455 		goto emul_write;
4456 
4457 	gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
4458 
4459 	if (gpa == UNMAPPED_GVA ||
4460 	    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4461 		goto emul_write;
4462 
4463 	if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
4464 		goto emul_write;
4465 
4466 	page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
4467 	if (is_error_page(page))
4468 		goto emul_write;
4469 
4470 	kaddr = kmap_atomic(page);
4471 	kaddr += offset_in_page(gpa);
4472 	switch (bytes) {
4473 	case 1:
4474 		exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
4475 		break;
4476 	case 2:
4477 		exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
4478 		break;
4479 	case 4:
4480 		exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
4481 		break;
4482 	case 8:
4483 		exchanged = CMPXCHG64(kaddr, old, new);
4484 		break;
4485 	default:
4486 		BUG();
4487 	}
4488 	kunmap_atomic(kaddr);
4489 	kvm_release_page_dirty(page);
4490 
4491 	if (!exchanged)
4492 		return X86EMUL_CMPXCHG_FAILED;
4493 
4494 	mark_page_dirty(vcpu->kvm, gpa >> PAGE_SHIFT);
4495 	kvm_mmu_pte_write(vcpu, gpa, new, bytes);
4496 
4497 	return X86EMUL_CONTINUE;
4498 
4499 emul_write:
4500 	printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
4501 
4502 	return emulator_write_emulated(ctxt, addr, new, bytes, exception);
4503 }
4504 
4505 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4506 {
4507 	/* TODO: String I/O for in kernel device */
4508 	int r;
4509 
4510 	if (vcpu->arch.pio.in)
4511 		r = kvm_io_bus_read(vcpu->kvm, KVM_PIO_BUS, vcpu->arch.pio.port,
4512 				    vcpu->arch.pio.size, pd);
4513 	else
4514 		r = kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS,
4515 				     vcpu->arch.pio.port, vcpu->arch.pio.size,
4516 				     pd);
4517 	return r;
4518 }
4519 
4520 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
4521 			       unsigned short port, void *val,
4522 			       unsigned int count, bool in)
4523 {
4524 	vcpu->arch.pio.port = port;
4525 	vcpu->arch.pio.in = in;
4526 	vcpu->arch.pio.count  = count;
4527 	vcpu->arch.pio.size = size;
4528 
4529 	if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4530 		vcpu->arch.pio.count = 0;
4531 		return 1;
4532 	}
4533 
4534 	vcpu->run->exit_reason = KVM_EXIT_IO;
4535 	vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
4536 	vcpu->run->io.size = size;
4537 	vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4538 	vcpu->run->io.count = count;
4539 	vcpu->run->io.port = port;
4540 
4541 	return 0;
4542 }
4543 
4544 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
4545 				    int size, unsigned short port, void *val,
4546 				    unsigned int count)
4547 {
4548 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4549 	int ret;
4550 
4551 	if (vcpu->arch.pio.count)
4552 		goto data_avail;
4553 
4554 	ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
4555 	if (ret) {
4556 data_avail:
4557 		memcpy(val, vcpu->arch.pio_data, size * count);
4558 		trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
4559 		vcpu->arch.pio.count = 0;
4560 		return 1;
4561 	}
4562 
4563 	return 0;
4564 }
4565 
4566 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
4567 				     int size, unsigned short port,
4568 				     const void *val, unsigned int count)
4569 {
4570 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4571 
4572 	memcpy(vcpu->arch.pio_data, val, size * count);
4573 	trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
4574 	return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
4575 }
4576 
4577 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4578 {
4579 	return kvm_x86_ops->get_segment_base(vcpu, seg);
4580 }
4581 
4582 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
4583 {
4584 	kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
4585 }
4586 
4587 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4588 {
4589 	if (!need_emulate_wbinvd(vcpu))
4590 		return X86EMUL_CONTINUE;
4591 
4592 	if (kvm_x86_ops->has_wbinvd_exit()) {
4593 		int cpu = get_cpu();
4594 
4595 		cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4596 		smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4597 				wbinvd_ipi, NULL, 1);
4598 		put_cpu();
4599 		cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
4600 	} else
4601 		wbinvd();
4602 	return X86EMUL_CONTINUE;
4603 }
4604 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4605 
4606 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
4607 {
4608 	kvm_emulate_wbinvd(emul_to_vcpu(ctxt));
4609 }
4610 
4611 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
4612 {
4613 	return _kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
4614 }
4615 
4616 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
4617 {
4618 
4619 	return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
4620 }
4621 
4622 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4623 {
4624 	return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4625 }
4626 
4627 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
4628 {
4629 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4630 	unsigned long value;
4631 
4632 	switch (cr) {
4633 	case 0:
4634 		value = kvm_read_cr0(vcpu);
4635 		break;
4636 	case 2:
4637 		value = vcpu->arch.cr2;
4638 		break;
4639 	case 3:
4640 		value = kvm_read_cr3(vcpu);
4641 		break;
4642 	case 4:
4643 		value = kvm_read_cr4(vcpu);
4644 		break;
4645 	case 8:
4646 		value = kvm_get_cr8(vcpu);
4647 		break;
4648 	default:
4649 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
4650 		return 0;
4651 	}
4652 
4653 	return value;
4654 }
4655 
4656 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
4657 {
4658 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4659 	int res = 0;
4660 
4661 	switch (cr) {
4662 	case 0:
4663 		res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
4664 		break;
4665 	case 2:
4666 		vcpu->arch.cr2 = val;
4667 		break;
4668 	case 3:
4669 		res = kvm_set_cr3(vcpu, val);
4670 		break;
4671 	case 4:
4672 		res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
4673 		break;
4674 	case 8:
4675 		res = kvm_set_cr8(vcpu, val);
4676 		break;
4677 	default:
4678 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
4679 		res = -1;
4680 	}
4681 
4682 	return res;
4683 }
4684 
4685 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
4686 {
4687 	return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
4688 }
4689 
4690 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4691 {
4692 	kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
4693 }
4694 
4695 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4696 {
4697 	kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
4698 }
4699 
4700 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4701 {
4702 	kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
4703 }
4704 
4705 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4706 {
4707 	kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
4708 }
4709 
4710 static unsigned long emulator_get_cached_segment_base(
4711 	struct x86_emulate_ctxt *ctxt, int seg)
4712 {
4713 	return get_segment_base(emul_to_vcpu(ctxt), seg);
4714 }
4715 
4716 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
4717 				 struct desc_struct *desc, u32 *base3,
4718 				 int seg)
4719 {
4720 	struct kvm_segment var;
4721 
4722 	kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
4723 	*selector = var.selector;
4724 
4725 	if (var.unusable) {
4726 		memset(desc, 0, sizeof(*desc));
4727 		return false;
4728 	}
4729 
4730 	if (var.g)
4731 		var.limit >>= 12;
4732 	set_desc_limit(desc, var.limit);
4733 	set_desc_base(desc, (unsigned long)var.base);
4734 #ifdef CONFIG_X86_64
4735 	if (base3)
4736 		*base3 = var.base >> 32;
4737 #endif
4738 	desc->type = var.type;
4739 	desc->s = var.s;
4740 	desc->dpl = var.dpl;
4741 	desc->p = var.present;
4742 	desc->avl = var.avl;
4743 	desc->l = var.l;
4744 	desc->d = var.db;
4745 	desc->g = var.g;
4746 
4747 	return true;
4748 }
4749 
4750 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
4751 				 struct desc_struct *desc, u32 base3,
4752 				 int seg)
4753 {
4754 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4755 	struct kvm_segment var;
4756 
4757 	var.selector = selector;
4758 	var.base = get_desc_base(desc);
4759 #ifdef CONFIG_X86_64
4760 	var.base |= ((u64)base3) << 32;
4761 #endif
4762 	var.limit = get_desc_limit(desc);
4763 	if (desc->g)
4764 		var.limit = (var.limit << 12) | 0xfff;
4765 	var.type = desc->type;
4766 	var.dpl = desc->dpl;
4767 	var.db = desc->d;
4768 	var.s = desc->s;
4769 	var.l = desc->l;
4770 	var.g = desc->g;
4771 	var.avl = desc->avl;
4772 	var.present = desc->p;
4773 	var.unusable = !var.present;
4774 	var.padding = 0;
4775 
4776 	kvm_set_segment(vcpu, &var, seg);
4777 	return;
4778 }
4779 
4780 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
4781 			    u32 msr_index, u64 *pdata)
4782 {
4783 	return kvm_get_msr(emul_to_vcpu(ctxt), msr_index, pdata);
4784 }
4785 
4786 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
4787 			    u32 msr_index, u64 data)
4788 {
4789 	struct msr_data msr;
4790 
4791 	msr.data = data;
4792 	msr.index = msr_index;
4793 	msr.host_initiated = false;
4794 	return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
4795 }
4796 
4797 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
4798 			      u32 pmc)
4799 {
4800 	return kvm_pmu_check_pmc(emul_to_vcpu(ctxt), pmc);
4801 }
4802 
4803 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
4804 			     u32 pmc, u64 *pdata)
4805 {
4806 	return kvm_pmu_read_pmc(emul_to_vcpu(ctxt), pmc, pdata);
4807 }
4808 
4809 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
4810 {
4811 	emul_to_vcpu(ctxt)->arch.halt_request = 1;
4812 }
4813 
4814 static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
4815 {
4816 	preempt_disable();
4817 	kvm_load_guest_fpu(emul_to_vcpu(ctxt));
4818 	/*
4819 	 * CR0.TS may reference the host fpu state, not the guest fpu state,
4820 	 * so it may be clear at this point.
4821 	 */
4822 	clts();
4823 }
4824 
4825 static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
4826 {
4827 	preempt_enable();
4828 }
4829 
4830 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
4831 			      struct x86_instruction_info *info,
4832 			      enum x86_intercept_stage stage)
4833 {
4834 	return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
4835 }
4836 
4837 static void emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
4838 			       u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
4839 {
4840 	kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx);
4841 }
4842 
4843 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
4844 {
4845 	return kvm_register_read(emul_to_vcpu(ctxt), reg);
4846 }
4847 
4848 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
4849 {
4850 	kvm_register_write(emul_to_vcpu(ctxt), reg, val);
4851 }
4852 
4853 static const struct x86_emulate_ops emulate_ops = {
4854 	.read_gpr            = emulator_read_gpr,
4855 	.write_gpr           = emulator_write_gpr,
4856 	.read_std            = kvm_read_guest_virt_system,
4857 	.write_std           = kvm_write_guest_virt_system,
4858 	.fetch               = kvm_fetch_guest_virt,
4859 	.read_emulated       = emulator_read_emulated,
4860 	.write_emulated      = emulator_write_emulated,
4861 	.cmpxchg_emulated    = emulator_cmpxchg_emulated,
4862 	.invlpg              = emulator_invlpg,
4863 	.pio_in_emulated     = emulator_pio_in_emulated,
4864 	.pio_out_emulated    = emulator_pio_out_emulated,
4865 	.get_segment         = emulator_get_segment,
4866 	.set_segment         = emulator_set_segment,
4867 	.get_cached_segment_base = emulator_get_cached_segment_base,
4868 	.get_gdt             = emulator_get_gdt,
4869 	.get_idt	     = emulator_get_idt,
4870 	.set_gdt             = emulator_set_gdt,
4871 	.set_idt	     = emulator_set_idt,
4872 	.get_cr              = emulator_get_cr,
4873 	.set_cr              = emulator_set_cr,
4874 	.cpl                 = emulator_get_cpl,
4875 	.get_dr              = emulator_get_dr,
4876 	.set_dr              = emulator_set_dr,
4877 	.set_msr             = emulator_set_msr,
4878 	.get_msr             = emulator_get_msr,
4879 	.check_pmc	     = emulator_check_pmc,
4880 	.read_pmc            = emulator_read_pmc,
4881 	.halt                = emulator_halt,
4882 	.wbinvd              = emulator_wbinvd,
4883 	.fix_hypercall       = emulator_fix_hypercall,
4884 	.get_fpu             = emulator_get_fpu,
4885 	.put_fpu             = emulator_put_fpu,
4886 	.intercept           = emulator_intercept,
4887 	.get_cpuid           = emulator_get_cpuid,
4888 };
4889 
4890 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
4891 {
4892 	u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
4893 	/*
4894 	 * an sti; sti; sequence only disable interrupts for the first
4895 	 * instruction. So, if the last instruction, be it emulated or
4896 	 * not, left the system with the INT_STI flag enabled, it
4897 	 * means that the last instruction is an sti. We should not
4898 	 * leave the flag on in this case. The same goes for mov ss
4899 	 */
4900 	if (int_shadow & mask)
4901 		mask = 0;
4902 	if (unlikely(int_shadow || mask)) {
4903 		kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
4904 		if (!mask)
4905 			kvm_make_request(KVM_REQ_EVENT, vcpu);
4906 	}
4907 }
4908 
4909 static void inject_emulated_exception(struct kvm_vcpu *vcpu)
4910 {
4911 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4912 	if (ctxt->exception.vector == PF_VECTOR)
4913 		kvm_propagate_fault(vcpu, &ctxt->exception);
4914 	else if (ctxt->exception.error_code_valid)
4915 		kvm_queue_exception_e(vcpu, ctxt->exception.vector,
4916 				      ctxt->exception.error_code);
4917 	else
4918 		kvm_queue_exception(vcpu, ctxt->exception.vector);
4919 }
4920 
4921 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
4922 {
4923 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4924 	int cs_db, cs_l;
4925 
4926 	kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4927 
4928 	ctxt->eflags = kvm_get_rflags(vcpu);
4929 	ctxt->eip = kvm_rip_read(vcpu);
4930 	ctxt->mode = (!is_protmode(vcpu))		? X86EMUL_MODE_REAL :
4931 		     (ctxt->eflags & X86_EFLAGS_VM)	? X86EMUL_MODE_VM86 :
4932 		     (cs_l && is_long_mode(vcpu))	? X86EMUL_MODE_PROT64 :
4933 		     cs_db				? X86EMUL_MODE_PROT32 :
4934 							  X86EMUL_MODE_PROT16;
4935 	ctxt->guest_mode = is_guest_mode(vcpu);
4936 
4937 	init_decode_cache(ctxt);
4938 	vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
4939 }
4940 
4941 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
4942 {
4943 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4944 	int ret;
4945 
4946 	init_emulate_ctxt(vcpu);
4947 
4948 	ctxt->op_bytes = 2;
4949 	ctxt->ad_bytes = 2;
4950 	ctxt->_eip = ctxt->eip + inc_eip;
4951 	ret = emulate_int_real(ctxt, irq);
4952 
4953 	if (ret != X86EMUL_CONTINUE)
4954 		return EMULATE_FAIL;
4955 
4956 	ctxt->eip = ctxt->_eip;
4957 	kvm_rip_write(vcpu, ctxt->eip);
4958 	kvm_set_rflags(vcpu, ctxt->eflags);
4959 
4960 	if (irq == NMI_VECTOR)
4961 		vcpu->arch.nmi_pending = 0;
4962 	else
4963 		vcpu->arch.interrupt.pending = false;
4964 
4965 	return EMULATE_DONE;
4966 }
4967 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
4968 
4969 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
4970 {
4971 	int r = EMULATE_DONE;
4972 
4973 	++vcpu->stat.insn_emulation_fail;
4974 	trace_kvm_emulate_insn_failed(vcpu);
4975 	if (!is_guest_mode(vcpu)) {
4976 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4977 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
4978 		vcpu->run->internal.ndata = 0;
4979 		r = EMULATE_FAIL;
4980 	}
4981 	kvm_queue_exception(vcpu, UD_VECTOR);
4982 
4983 	return r;
4984 }
4985 
4986 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
4987 				  bool write_fault_to_shadow_pgtable,
4988 				  int emulation_type)
4989 {
4990 	gpa_t gpa = cr2;
4991 	pfn_t pfn;
4992 
4993 	if (emulation_type & EMULTYPE_NO_REEXECUTE)
4994 		return false;
4995 
4996 	if (!vcpu->arch.mmu.direct_map) {
4997 		/*
4998 		 * Write permission should be allowed since only
4999 		 * write access need to be emulated.
5000 		 */
5001 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5002 
5003 		/*
5004 		 * If the mapping is invalid in guest, let cpu retry
5005 		 * it to generate fault.
5006 		 */
5007 		if (gpa == UNMAPPED_GVA)
5008 			return true;
5009 	}
5010 
5011 	/*
5012 	 * Do not retry the unhandleable instruction if it faults on the
5013 	 * readonly host memory, otherwise it will goto a infinite loop:
5014 	 * retry instruction -> write #PF -> emulation fail -> retry
5015 	 * instruction -> ...
5016 	 */
5017 	pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
5018 
5019 	/*
5020 	 * If the instruction failed on the error pfn, it can not be fixed,
5021 	 * report the error to userspace.
5022 	 */
5023 	if (is_error_noslot_pfn(pfn))
5024 		return false;
5025 
5026 	kvm_release_pfn_clean(pfn);
5027 
5028 	/* The instructions are well-emulated on direct mmu. */
5029 	if (vcpu->arch.mmu.direct_map) {
5030 		unsigned int indirect_shadow_pages;
5031 
5032 		spin_lock(&vcpu->kvm->mmu_lock);
5033 		indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
5034 		spin_unlock(&vcpu->kvm->mmu_lock);
5035 
5036 		if (indirect_shadow_pages)
5037 			kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5038 
5039 		return true;
5040 	}
5041 
5042 	/*
5043 	 * if emulation was due to access to shadowed page table
5044 	 * and it failed try to unshadow page and re-enter the
5045 	 * guest to let CPU execute the instruction.
5046 	 */
5047 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5048 
5049 	/*
5050 	 * If the access faults on its page table, it can not
5051 	 * be fixed by unprotecting shadow page and it should
5052 	 * be reported to userspace.
5053 	 */
5054 	return !write_fault_to_shadow_pgtable;
5055 }
5056 
5057 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
5058 			      unsigned long cr2,  int emulation_type)
5059 {
5060 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5061 	unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
5062 
5063 	last_retry_eip = vcpu->arch.last_retry_eip;
5064 	last_retry_addr = vcpu->arch.last_retry_addr;
5065 
5066 	/*
5067 	 * If the emulation is caused by #PF and it is non-page_table
5068 	 * writing instruction, it means the VM-EXIT is caused by shadow
5069 	 * page protected, we can zap the shadow page and retry this
5070 	 * instruction directly.
5071 	 *
5072 	 * Note: if the guest uses a non-page-table modifying instruction
5073 	 * on the PDE that points to the instruction, then we will unmap
5074 	 * the instruction and go to an infinite loop. So, we cache the
5075 	 * last retried eip and the last fault address, if we meet the eip
5076 	 * and the address again, we can break out of the potential infinite
5077 	 * loop.
5078 	 */
5079 	vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
5080 
5081 	if (!(emulation_type & EMULTYPE_RETRY))
5082 		return false;
5083 
5084 	if (x86_page_table_writing_insn(ctxt))
5085 		return false;
5086 
5087 	if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
5088 		return false;
5089 
5090 	vcpu->arch.last_retry_eip = ctxt->eip;
5091 	vcpu->arch.last_retry_addr = cr2;
5092 
5093 	if (!vcpu->arch.mmu.direct_map)
5094 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5095 
5096 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5097 
5098 	return true;
5099 }
5100 
5101 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
5102 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
5103 
5104 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
5105 				unsigned long *db)
5106 {
5107 	u32 dr6 = 0;
5108 	int i;
5109 	u32 enable, rwlen;
5110 
5111 	enable = dr7;
5112 	rwlen = dr7 >> 16;
5113 	for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
5114 		if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
5115 			dr6 |= (1 << i);
5116 	return dr6;
5117 }
5118 
5119 static void kvm_vcpu_check_singlestep(struct kvm_vcpu *vcpu, unsigned long rflags, int *r)
5120 {
5121 	struct kvm_run *kvm_run = vcpu->run;
5122 
5123 	/*
5124 	 * rflags is the old, "raw" value of the flags.  The new value has
5125 	 * not been saved yet.
5126 	 *
5127 	 * This is correct even for TF set by the guest, because "the
5128 	 * processor will not generate this exception after the instruction
5129 	 * that sets the TF flag".
5130 	 */
5131 	if (unlikely(rflags & X86_EFLAGS_TF)) {
5132 		if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
5133 			kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 |
5134 						  DR6_RTM;
5135 			kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
5136 			kvm_run->debug.arch.exception = DB_VECTOR;
5137 			kvm_run->exit_reason = KVM_EXIT_DEBUG;
5138 			*r = EMULATE_USER_EXIT;
5139 		} else {
5140 			vcpu->arch.emulate_ctxt.eflags &= ~X86_EFLAGS_TF;
5141 			/*
5142 			 * "Certain debug exceptions may clear bit 0-3.  The
5143 			 * remaining contents of the DR6 register are never
5144 			 * cleared by the processor".
5145 			 */
5146 			vcpu->arch.dr6 &= ~15;
5147 			vcpu->arch.dr6 |= DR6_BS | DR6_RTM;
5148 			kvm_queue_exception(vcpu, DB_VECTOR);
5149 		}
5150 	}
5151 }
5152 
5153 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
5154 {
5155 	struct kvm_run *kvm_run = vcpu->run;
5156 	unsigned long eip = vcpu->arch.emulate_ctxt.eip;
5157 	u32 dr6 = 0;
5158 
5159 	if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
5160 	    (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
5161 		dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5162 					   vcpu->arch.guest_debug_dr7,
5163 					   vcpu->arch.eff_db);
5164 
5165 		if (dr6 != 0) {
5166 			kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
5167 			kvm_run->debug.arch.pc = kvm_rip_read(vcpu) +
5168 				get_segment_base(vcpu, VCPU_SREG_CS);
5169 
5170 			kvm_run->debug.arch.exception = DB_VECTOR;
5171 			kvm_run->exit_reason = KVM_EXIT_DEBUG;
5172 			*r = EMULATE_USER_EXIT;
5173 			return true;
5174 		}
5175 	}
5176 
5177 	if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
5178 	    !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
5179 		dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5180 					   vcpu->arch.dr7,
5181 					   vcpu->arch.db);
5182 
5183 		if (dr6 != 0) {
5184 			vcpu->arch.dr6 &= ~15;
5185 			vcpu->arch.dr6 |= dr6 | DR6_RTM;
5186 			kvm_queue_exception(vcpu, DB_VECTOR);
5187 			*r = EMULATE_DONE;
5188 			return true;
5189 		}
5190 	}
5191 
5192 	return false;
5193 }
5194 
5195 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
5196 			    unsigned long cr2,
5197 			    int emulation_type,
5198 			    void *insn,
5199 			    int insn_len)
5200 {
5201 	int r;
5202 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5203 	bool writeback = true;
5204 	bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
5205 
5206 	/*
5207 	 * Clear write_fault_to_shadow_pgtable here to ensure it is
5208 	 * never reused.
5209 	 */
5210 	vcpu->arch.write_fault_to_shadow_pgtable = false;
5211 	kvm_clear_exception_queue(vcpu);
5212 
5213 	if (!(emulation_type & EMULTYPE_NO_DECODE)) {
5214 		init_emulate_ctxt(vcpu);
5215 
5216 		/*
5217 		 * We will reenter on the same instruction since
5218 		 * we do not set complete_userspace_io.  This does not
5219 		 * handle watchpoints yet, those would be handled in
5220 		 * the emulate_ops.
5221 		 */
5222 		if (kvm_vcpu_check_breakpoint(vcpu, &r))
5223 			return r;
5224 
5225 		ctxt->interruptibility = 0;
5226 		ctxt->have_exception = false;
5227 		ctxt->perm_ok = false;
5228 
5229 		ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
5230 
5231 		r = x86_decode_insn(ctxt, insn, insn_len);
5232 
5233 		trace_kvm_emulate_insn_start(vcpu);
5234 		++vcpu->stat.insn_emulation;
5235 		if (r != EMULATION_OK)  {
5236 			if (emulation_type & EMULTYPE_TRAP_UD)
5237 				return EMULATE_FAIL;
5238 			if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5239 						emulation_type))
5240 				return EMULATE_DONE;
5241 			if (emulation_type & EMULTYPE_SKIP)
5242 				return EMULATE_FAIL;
5243 			return handle_emulation_failure(vcpu);
5244 		}
5245 	}
5246 
5247 	if (emulation_type & EMULTYPE_SKIP) {
5248 		kvm_rip_write(vcpu, ctxt->_eip);
5249 		if (ctxt->eflags & X86_EFLAGS_RF)
5250 			kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
5251 		return EMULATE_DONE;
5252 	}
5253 
5254 	if (retry_instruction(ctxt, cr2, emulation_type))
5255 		return EMULATE_DONE;
5256 
5257 	/* this is needed for vmware backdoor interface to work since it
5258 	   changes registers values  during IO operation */
5259 	if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
5260 		vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5261 		emulator_invalidate_register_cache(ctxt);
5262 	}
5263 
5264 restart:
5265 	r = x86_emulate_insn(ctxt);
5266 
5267 	if (r == EMULATION_INTERCEPTED)
5268 		return EMULATE_DONE;
5269 
5270 	if (r == EMULATION_FAILED) {
5271 		if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5272 					emulation_type))
5273 			return EMULATE_DONE;
5274 
5275 		return handle_emulation_failure(vcpu);
5276 	}
5277 
5278 	if (ctxt->have_exception) {
5279 		inject_emulated_exception(vcpu);
5280 		r = EMULATE_DONE;
5281 	} else if (vcpu->arch.pio.count) {
5282 		if (!vcpu->arch.pio.in) {
5283 			/* FIXME: return into emulator if single-stepping.  */
5284 			vcpu->arch.pio.count = 0;
5285 		} else {
5286 			writeback = false;
5287 			vcpu->arch.complete_userspace_io = complete_emulated_pio;
5288 		}
5289 		r = EMULATE_USER_EXIT;
5290 	} else if (vcpu->mmio_needed) {
5291 		if (!vcpu->mmio_is_write)
5292 			writeback = false;
5293 		r = EMULATE_USER_EXIT;
5294 		vcpu->arch.complete_userspace_io = complete_emulated_mmio;
5295 	} else if (r == EMULATION_RESTART)
5296 		goto restart;
5297 	else
5298 		r = EMULATE_DONE;
5299 
5300 	if (writeback) {
5301 		unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
5302 		toggle_interruptibility(vcpu, ctxt->interruptibility);
5303 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5304 		kvm_rip_write(vcpu, ctxt->eip);
5305 		if (r == EMULATE_DONE)
5306 			kvm_vcpu_check_singlestep(vcpu, rflags, &r);
5307 		__kvm_set_rflags(vcpu, ctxt->eflags);
5308 
5309 		/*
5310 		 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
5311 		 * do nothing, and it will be requested again as soon as
5312 		 * the shadow expires.  But we still need to check here,
5313 		 * because POPF has no interrupt shadow.
5314 		 */
5315 		if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
5316 			kvm_make_request(KVM_REQ_EVENT, vcpu);
5317 	} else
5318 		vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
5319 
5320 	return r;
5321 }
5322 EXPORT_SYMBOL_GPL(x86_emulate_instruction);
5323 
5324 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
5325 {
5326 	unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
5327 	int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
5328 					    size, port, &val, 1);
5329 	/* do not return to emulator after return from userspace */
5330 	vcpu->arch.pio.count = 0;
5331 	return ret;
5332 }
5333 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
5334 
5335 static void tsc_bad(void *info)
5336 {
5337 	__this_cpu_write(cpu_tsc_khz, 0);
5338 }
5339 
5340 static void tsc_khz_changed(void *data)
5341 {
5342 	struct cpufreq_freqs *freq = data;
5343 	unsigned long khz = 0;
5344 
5345 	if (data)
5346 		khz = freq->new;
5347 	else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5348 		khz = cpufreq_quick_get(raw_smp_processor_id());
5349 	if (!khz)
5350 		khz = tsc_khz;
5351 	__this_cpu_write(cpu_tsc_khz, khz);
5352 }
5353 
5354 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
5355 				     void *data)
5356 {
5357 	struct cpufreq_freqs *freq = data;
5358 	struct kvm *kvm;
5359 	struct kvm_vcpu *vcpu;
5360 	int i, send_ipi = 0;
5361 
5362 	/*
5363 	 * We allow guests to temporarily run on slowing clocks,
5364 	 * provided we notify them after, or to run on accelerating
5365 	 * clocks, provided we notify them before.  Thus time never
5366 	 * goes backwards.
5367 	 *
5368 	 * However, we have a problem.  We can't atomically update
5369 	 * the frequency of a given CPU from this function; it is
5370 	 * merely a notifier, which can be called from any CPU.
5371 	 * Changing the TSC frequency at arbitrary points in time
5372 	 * requires a recomputation of local variables related to
5373 	 * the TSC for each VCPU.  We must flag these local variables
5374 	 * to be updated and be sure the update takes place with the
5375 	 * new frequency before any guests proceed.
5376 	 *
5377 	 * Unfortunately, the combination of hotplug CPU and frequency
5378 	 * change creates an intractable locking scenario; the order
5379 	 * of when these callouts happen is undefined with respect to
5380 	 * CPU hotplug, and they can race with each other.  As such,
5381 	 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
5382 	 * undefined; you can actually have a CPU frequency change take
5383 	 * place in between the computation of X and the setting of the
5384 	 * variable.  To protect against this problem, all updates of
5385 	 * the per_cpu tsc_khz variable are done in an interrupt
5386 	 * protected IPI, and all callers wishing to update the value
5387 	 * must wait for a synchronous IPI to complete (which is trivial
5388 	 * if the caller is on the CPU already).  This establishes the
5389 	 * necessary total order on variable updates.
5390 	 *
5391 	 * Note that because a guest time update may take place
5392 	 * anytime after the setting of the VCPU's request bit, the
5393 	 * correct TSC value must be set before the request.  However,
5394 	 * to ensure the update actually makes it to any guest which
5395 	 * starts running in hardware virtualization between the set
5396 	 * and the acquisition of the spinlock, we must also ping the
5397 	 * CPU after setting the request bit.
5398 	 *
5399 	 */
5400 
5401 	if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
5402 		return 0;
5403 	if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
5404 		return 0;
5405 
5406 	smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5407 
5408 	spin_lock(&kvm_lock);
5409 	list_for_each_entry(kvm, &vm_list, vm_list) {
5410 		kvm_for_each_vcpu(i, vcpu, kvm) {
5411 			if (vcpu->cpu != freq->cpu)
5412 				continue;
5413 			kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5414 			if (vcpu->cpu != smp_processor_id())
5415 				send_ipi = 1;
5416 		}
5417 	}
5418 	spin_unlock(&kvm_lock);
5419 
5420 	if (freq->old < freq->new && send_ipi) {
5421 		/*
5422 		 * We upscale the frequency.  Must make the guest
5423 		 * doesn't see old kvmclock values while running with
5424 		 * the new frequency, otherwise we risk the guest sees
5425 		 * time go backwards.
5426 		 *
5427 		 * In case we update the frequency for another cpu
5428 		 * (which might be in guest context) send an interrupt
5429 		 * to kick the cpu out of guest context.  Next time
5430 		 * guest context is entered kvmclock will be updated,
5431 		 * so the guest will not see stale values.
5432 		 */
5433 		smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5434 	}
5435 	return 0;
5436 }
5437 
5438 static struct notifier_block kvmclock_cpufreq_notifier_block = {
5439 	.notifier_call  = kvmclock_cpufreq_notifier
5440 };
5441 
5442 static int kvmclock_cpu_notifier(struct notifier_block *nfb,
5443 					unsigned long action, void *hcpu)
5444 {
5445 	unsigned int cpu = (unsigned long)hcpu;
5446 
5447 	switch (action) {
5448 		case CPU_ONLINE:
5449 		case CPU_DOWN_FAILED:
5450 			smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5451 			break;
5452 		case CPU_DOWN_PREPARE:
5453 			smp_call_function_single(cpu, tsc_bad, NULL, 1);
5454 			break;
5455 	}
5456 	return NOTIFY_OK;
5457 }
5458 
5459 static struct notifier_block kvmclock_cpu_notifier_block = {
5460 	.notifier_call  = kvmclock_cpu_notifier,
5461 	.priority = -INT_MAX
5462 };
5463 
5464 static void kvm_timer_init(void)
5465 {
5466 	int cpu;
5467 
5468 	max_tsc_khz = tsc_khz;
5469 
5470 	cpu_notifier_register_begin();
5471 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5472 #ifdef CONFIG_CPU_FREQ
5473 		struct cpufreq_policy policy;
5474 		memset(&policy, 0, sizeof(policy));
5475 		cpu = get_cpu();
5476 		cpufreq_get_policy(&policy, cpu);
5477 		if (policy.cpuinfo.max_freq)
5478 			max_tsc_khz = policy.cpuinfo.max_freq;
5479 		put_cpu();
5480 #endif
5481 		cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
5482 					  CPUFREQ_TRANSITION_NOTIFIER);
5483 	}
5484 	pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
5485 	for_each_online_cpu(cpu)
5486 		smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5487 
5488 	__register_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5489 	cpu_notifier_register_done();
5490 
5491 }
5492 
5493 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
5494 
5495 int kvm_is_in_guest(void)
5496 {
5497 	return __this_cpu_read(current_vcpu) != NULL;
5498 }
5499 
5500 static int kvm_is_user_mode(void)
5501 {
5502 	int user_mode = 3;
5503 
5504 	if (__this_cpu_read(current_vcpu))
5505 		user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
5506 
5507 	return user_mode != 0;
5508 }
5509 
5510 static unsigned long kvm_get_guest_ip(void)
5511 {
5512 	unsigned long ip = 0;
5513 
5514 	if (__this_cpu_read(current_vcpu))
5515 		ip = kvm_rip_read(__this_cpu_read(current_vcpu));
5516 
5517 	return ip;
5518 }
5519 
5520 static struct perf_guest_info_callbacks kvm_guest_cbs = {
5521 	.is_in_guest		= kvm_is_in_guest,
5522 	.is_user_mode		= kvm_is_user_mode,
5523 	.get_guest_ip		= kvm_get_guest_ip,
5524 };
5525 
5526 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
5527 {
5528 	__this_cpu_write(current_vcpu, vcpu);
5529 }
5530 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
5531 
5532 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
5533 {
5534 	__this_cpu_write(current_vcpu, NULL);
5535 }
5536 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
5537 
5538 static void kvm_set_mmio_spte_mask(void)
5539 {
5540 	u64 mask;
5541 	int maxphyaddr = boot_cpu_data.x86_phys_bits;
5542 
5543 	/*
5544 	 * Set the reserved bits and the present bit of an paging-structure
5545 	 * entry to generate page fault with PFER.RSV = 1.
5546 	 */
5547 	 /* Mask the reserved physical address bits. */
5548 	mask = ((1ull << (51 - maxphyaddr + 1)) - 1) << maxphyaddr;
5549 
5550 	/* Bit 62 is always reserved for 32bit host. */
5551 	mask |= 0x3ull << 62;
5552 
5553 	/* Set the present bit. */
5554 	mask |= 1ull;
5555 
5556 #ifdef CONFIG_X86_64
5557 	/*
5558 	 * If reserved bit is not supported, clear the present bit to disable
5559 	 * mmio page fault.
5560 	 */
5561 	if (maxphyaddr == 52)
5562 		mask &= ~1ull;
5563 #endif
5564 
5565 	kvm_mmu_set_mmio_spte_mask(mask);
5566 }
5567 
5568 #ifdef CONFIG_X86_64
5569 static void pvclock_gtod_update_fn(struct work_struct *work)
5570 {
5571 	struct kvm *kvm;
5572 
5573 	struct kvm_vcpu *vcpu;
5574 	int i;
5575 
5576 	spin_lock(&kvm_lock);
5577 	list_for_each_entry(kvm, &vm_list, vm_list)
5578 		kvm_for_each_vcpu(i, vcpu, kvm)
5579 			set_bit(KVM_REQ_MASTERCLOCK_UPDATE, &vcpu->requests);
5580 	atomic_set(&kvm_guest_has_master_clock, 0);
5581 	spin_unlock(&kvm_lock);
5582 }
5583 
5584 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
5585 
5586 /*
5587  * Notification about pvclock gtod data update.
5588  */
5589 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
5590 			       void *priv)
5591 {
5592 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
5593 	struct timekeeper *tk = priv;
5594 
5595 	update_pvclock_gtod(tk);
5596 
5597 	/* disable master clock if host does not trust, or does not
5598 	 * use, TSC clocksource
5599 	 */
5600 	if (gtod->clock.vclock_mode != VCLOCK_TSC &&
5601 	    atomic_read(&kvm_guest_has_master_clock) != 0)
5602 		queue_work(system_long_wq, &pvclock_gtod_work);
5603 
5604 	return 0;
5605 }
5606 
5607 static struct notifier_block pvclock_gtod_notifier = {
5608 	.notifier_call = pvclock_gtod_notify,
5609 };
5610 #endif
5611 
5612 int kvm_arch_init(void *opaque)
5613 {
5614 	int r;
5615 	struct kvm_x86_ops *ops = opaque;
5616 
5617 	if (kvm_x86_ops) {
5618 		printk(KERN_ERR "kvm: already loaded the other module\n");
5619 		r = -EEXIST;
5620 		goto out;
5621 	}
5622 
5623 	if (!ops->cpu_has_kvm_support()) {
5624 		printk(KERN_ERR "kvm: no hardware support\n");
5625 		r = -EOPNOTSUPP;
5626 		goto out;
5627 	}
5628 	if (ops->disabled_by_bios()) {
5629 		printk(KERN_ERR "kvm: disabled by bios\n");
5630 		r = -EOPNOTSUPP;
5631 		goto out;
5632 	}
5633 
5634 	r = -ENOMEM;
5635 	shared_msrs = alloc_percpu(struct kvm_shared_msrs);
5636 	if (!shared_msrs) {
5637 		printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
5638 		goto out;
5639 	}
5640 
5641 	r = kvm_mmu_module_init();
5642 	if (r)
5643 		goto out_free_percpu;
5644 
5645 	kvm_set_mmio_spte_mask();
5646 
5647 	kvm_x86_ops = ops;
5648 	kvm_init_msr_list();
5649 
5650 	kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
5651 			PT_DIRTY_MASK, PT64_NX_MASK, 0);
5652 
5653 	kvm_timer_init();
5654 
5655 	perf_register_guest_info_callbacks(&kvm_guest_cbs);
5656 
5657 	if (cpu_has_xsave)
5658 		host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
5659 
5660 	kvm_lapic_init();
5661 #ifdef CONFIG_X86_64
5662 	pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
5663 #endif
5664 
5665 	return 0;
5666 
5667 out_free_percpu:
5668 	free_percpu(shared_msrs);
5669 out:
5670 	return r;
5671 }
5672 
5673 void kvm_arch_exit(void)
5674 {
5675 	perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
5676 
5677 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5678 		cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
5679 					    CPUFREQ_TRANSITION_NOTIFIER);
5680 	unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5681 #ifdef CONFIG_X86_64
5682 	pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
5683 #endif
5684 	kvm_x86_ops = NULL;
5685 	kvm_mmu_module_exit();
5686 	free_percpu(shared_msrs);
5687 }
5688 
5689 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
5690 {
5691 	++vcpu->stat.halt_exits;
5692 	if (irqchip_in_kernel(vcpu->kvm)) {
5693 		vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
5694 		return 1;
5695 	} else {
5696 		vcpu->run->exit_reason = KVM_EXIT_HLT;
5697 		return 0;
5698 	}
5699 }
5700 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
5701 
5702 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
5703 {
5704 	u64 param, ingpa, outgpa, ret;
5705 	uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
5706 	bool fast, longmode;
5707 
5708 	/*
5709 	 * hypercall generates UD from non zero cpl and real mode
5710 	 * per HYPER-V spec
5711 	 */
5712 	if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
5713 		kvm_queue_exception(vcpu, UD_VECTOR);
5714 		return 0;
5715 	}
5716 
5717 	longmode = is_64_bit_mode(vcpu);
5718 
5719 	if (!longmode) {
5720 		param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
5721 			(kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
5722 		ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
5723 			(kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
5724 		outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
5725 			(kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
5726 	}
5727 #ifdef CONFIG_X86_64
5728 	else {
5729 		param = kvm_register_read(vcpu, VCPU_REGS_RCX);
5730 		ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
5731 		outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
5732 	}
5733 #endif
5734 
5735 	code = param & 0xffff;
5736 	fast = (param >> 16) & 0x1;
5737 	rep_cnt = (param >> 32) & 0xfff;
5738 	rep_idx = (param >> 48) & 0xfff;
5739 
5740 	trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
5741 
5742 	switch (code) {
5743 	case HV_X64_HV_NOTIFY_LONG_SPIN_WAIT:
5744 		kvm_vcpu_on_spin(vcpu);
5745 		break;
5746 	default:
5747 		res = HV_STATUS_INVALID_HYPERCALL_CODE;
5748 		break;
5749 	}
5750 
5751 	ret = res | (((u64)rep_done & 0xfff) << 32);
5752 	if (longmode) {
5753 		kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5754 	} else {
5755 		kvm_register_write(vcpu, VCPU_REGS_RDX, ret >> 32);
5756 		kvm_register_write(vcpu, VCPU_REGS_RAX, ret & 0xffffffff);
5757 	}
5758 
5759 	return 1;
5760 }
5761 
5762 /*
5763  * kvm_pv_kick_cpu_op:  Kick a vcpu.
5764  *
5765  * @apicid - apicid of vcpu to be kicked.
5766  */
5767 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
5768 {
5769 	struct kvm_lapic_irq lapic_irq;
5770 
5771 	lapic_irq.shorthand = 0;
5772 	lapic_irq.dest_mode = 0;
5773 	lapic_irq.dest_id = apicid;
5774 
5775 	lapic_irq.delivery_mode = APIC_DM_REMRD;
5776 	kvm_irq_delivery_to_apic(kvm, 0, &lapic_irq, NULL);
5777 }
5778 
5779 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
5780 {
5781 	unsigned long nr, a0, a1, a2, a3, ret;
5782 	int op_64_bit, r = 1;
5783 
5784 	if (kvm_hv_hypercall_enabled(vcpu->kvm))
5785 		return kvm_hv_hypercall(vcpu);
5786 
5787 	nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
5788 	a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
5789 	a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
5790 	a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
5791 	a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
5792 
5793 	trace_kvm_hypercall(nr, a0, a1, a2, a3);
5794 
5795 	op_64_bit = is_64_bit_mode(vcpu);
5796 	if (!op_64_bit) {
5797 		nr &= 0xFFFFFFFF;
5798 		a0 &= 0xFFFFFFFF;
5799 		a1 &= 0xFFFFFFFF;
5800 		a2 &= 0xFFFFFFFF;
5801 		a3 &= 0xFFFFFFFF;
5802 	}
5803 
5804 	if (kvm_x86_ops->get_cpl(vcpu) != 0) {
5805 		ret = -KVM_EPERM;
5806 		goto out;
5807 	}
5808 
5809 	switch (nr) {
5810 	case KVM_HC_VAPIC_POLL_IRQ:
5811 		ret = 0;
5812 		break;
5813 	case KVM_HC_KICK_CPU:
5814 		kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
5815 		ret = 0;
5816 		break;
5817 	default:
5818 		ret = -KVM_ENOSYS;
5819 		break;
5820 	}
5821 out:
5822 	if (!op_64_bit)
5823 		ret = (u32)ret;
5824 	kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5825 	++vcpu->stat.hypercalls;
5826 	return r;
5827 }
5828 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
5829 
5830 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
5831 {
5832 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5833 	char instruction[3];
5834 	unsigned long rip = kvm_rip_read(vcpu);
5835 
5836 	kvm_x86_ops->patch_hypercall(vcpu, instruction);
5837 
5838 	return emulator_write_emulated(ctxt, rip, instruction, 3, NULL);
5839 }
5840 
5841 /*
5842  * Check if userspace requested an interrupt window, and that the
5843  * interrupt window is open.
5844  *
5845  * No need to exit to userspace if we already have an interrupt queued.
5846  */
5847 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
5848 {
5849 	return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
5850 		vcpu->run->request_interrupt_window &&
5851 		kvm_arch_interrupt_allowed(vcpu));
5852 }
5853 
5854 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
5855 {
5856 	struct kvm_run *kvm_run = vcpu->run;
5857 
5858 	kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
5859 	kvm_run->cr8 = kvm_get_cr8(vcpu);
5860 	kvm_run->apic_base = kvm_get_apic_base(vcpu);
5861 	if (irqchip_in_kernel(vcpu->kvm))
5862 		kvm_run->ready_for_interrupt_injection = 1;
5863 	else
5864 		kvm_run->ready_for_interrupt_injection =
5865 			kvm_arch_interrupt_allowed(vcpu) &&
5866 			!kvm_cpu_has_interrupt(vcpu) &&
5867 			!kvm_event_needs_reinjection(vcpu);
5868 }
5869 
5870 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
5871 {
5872 	int max_irr, tpr;
5873 
5874 	if (!kvm_x86_ops->update_cr8_intercept)
5875 		return;
5876 
5877 	if (!vcpu->arch.apic)
5878 		return;
5879 
5880 	if (!vcpu->arch.apic->vapic_addr)
5881 		max_irr = kvm_lapic_find_highest_irr(vcpu);
5882 	else
5883 		max_irr = -1;
5884 
5885 	if (max_irr != -1)
5886 		max_irr >>= 4;
5887 
5888 	tpr = kvm_lapic_get_cr8(vcpu);
5889 
5890 	kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
5891 }
5892 
5893 static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
5894 {
5895 	int r;
5896 
5897 	/* try to reinject previous events if any */
5898 	if (vcpu->arch.exception.pending) {
5899 		trace_kvm_inj_exception(vcpu->arch.exception.nr,
5900 					vcpu->arch.exception.has_error_code,
5901 					vcpu->arch.exception.error_code);
5902 
5903 		if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
5904 			__kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
5905 					     X86_EFLAGS_RF);
5906 
5907 		kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
5908 					  vcpu->arch.exception.has_error_code,
5909 					  vcpu->arch.exception.error_code,
5910 					  vcpu->arch.exception.reinject);
5911 		return 0;
5912 	}
5913 
5914 	if (vcpu->arch.nmi_injected) {
5915 		kvm_x86_ops->set_nmi(vcpu);
5916 		return 0;
5917 	}
5918 
5919 	if (vcpu->arch.interrupt.pending) {
5920 		kvm_x86_ops->set_irq(vcpu);
5921 		return 0;
5922 	}
5923 
5924 	if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
5925 		r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
5926 		if (r != 0)
5927 			return r;
5928 	}
5929 
5930 	/* try to inject new event if pending */
5931 	if (vcpu->arch.nmi_pending) {
5932 		if (kvm_x86_ops->nmi_allowed(vcpu)) {
5933 			--vcpu->arch.nmi_pending;
5934 			vcpu->arch.nmi_injected = true;
5935 			kvm_x86_ops->set_nmi(vcpu);
5936 		}
5937 	} else if (kvm_cpu_has_injectable_intr(vcpu)) {
5938 		/*
5939 		 * Because interrupts can be injected asynchronously, we are
5940 		 * calling check_nested_events again here to avoid a race condition.
5941 		 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
5942 		 * proposal and current concerns.  Perhaps we should be setting
5943 		 * KVM_REQ_EVENT only on certain events and not unconditionally?
5944 		 */
5945 		if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
5946 			r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
5947 			if (r != 0)
5948 				return r;
5949 		}
5950 		if (kvm_x86_ops->interrupt_allowed(vcpu)) {
5951 			kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
5952 					    false);
5953 			kvm_x86_ops->set_irq(vcpu);
5954 		}
5955 	}
5956 	return 0;
5957 }
5958 
5959 static void process_nmi(struct kvm_vcpu *vcpu)
5960 {
5961 	unsigned limit = 2;
5962 
5963 	/*
5964 	 * x86 is limited to one NMI running, and one NMI pending after it.
5965 	 * If an NMI is already in progress, limit further NMIs to just one.
5966 	 * Otherwise, allow two (and we'll inject the first one immediately).
5967 	 */
5968 	if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
5969 		limit = 1;
5970 
5971 	vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
5972 	vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
5973 	kvm_make_request(KVM_REQ_EVENT, vcpu);
5974 }
5975 
5976 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
5977 {
5978 	u64 eoi_exit_bitmap[4];
5979 	u32 tmr[8];
5980 
5981 	if (!kvm_apic_hw_enabled(vcpu->arch.apic))
5982 		return;
5983 
5984 	memset(eoi_exit_bitmap, 0, 32);
5985 	memset(tmr, 0, 32);
5986 
5987 	kvm_ioapic_scan_entry(vcpu, eoi_exit_bitmap, tmr);
5988 	kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
5989 	kvm_apic_update_tmr(vcpu, tmr);
5990 }
5991 
5992 /*
5993  * Returns 1 to let __vcpu_run() continue the guest execution loop without
5994  * exiting to the userspace.  Otherwise, the value will be returned to the
5995  * userspace.
5996  */
5997 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
5998 {
5999 	int r;
6000 	bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
6001 		vcpu->run->request_interrupt_window;
6002 	bool req_immediate_exit = false;
6003 
6004 	if (vcpu->requests) {
6005 		if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
6006 			kvm_mmu_unload(vcpu);
6007 		if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
6008 			__kvm_migrate_timers(vcpu);
6009 		if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
6010 			kvm_gen_update_masterclock(vcpu->kvm);
6011 		if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
6012 			kvm_gen_kvmclock_update(vcpu);
6013 		if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
6014 			r = kvm_guest_time_update(vcpu);
6015 			if (unlikely(r))
6016 				goto out;
6017 		}
6018 		if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
6019 			kvm_mmu_sync_roots(vcpu);
6020 		if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
6021 			kvm_x86_ops->tlb_flush(vcpu);
6022 		if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
6023 			vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
6024 			r = 0;
6025 			goto out;
6026 		}
6027 		if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
6028 			vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
6029 			r = 0;
6030 			goto out;
6031 		}
6032 		if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
6033 			vcpu->fpu_active = 0;
6034 			kvm_x86_ops->fpu_deactivate(vcpu);
6035 		}
6036 		if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
6037 			/* Page is swapped out. Do synthetic halt */
6038 			vcpu->arch.apf.halted = true;
6039 			r = 1;
6040 			goto out;
6041 		}
6042 		if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
6043 			record_steal_time(vcpu);
6044 		if (kvm_check_request(KVM_REQ_NMI, vcpu))
6045 			process_nmi(vcpu);
6046 		if (kvm_check_request(KVM_REQ_PMU, vcpu))
6047 			kvm_handle_pmu_event(vcpu);
6048 		if (kvm_check_request(KVM_REQ_PMI, vcpu))
6049 			kvm_deliver_pmi(vcpu);
6050 		if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
6051 			vcpu_scan_ioapic(vcpu);
6052 	}
6053 
6054 	if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
6055 		kvm_apic_accept_events(vcpu);
6056 		if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
6057 			r = 1;
6058 			goto out;
6059 		}
6060 
6061 		if (inject_pending_event(vcpu, req_int_win) != 0)
6062 			req_immediate_exit = true;
6063 		/* enable NMI/IRQ window open exits if needed */
6064 		else if (vcpu->arch.nmi_pending)
6065 			kvm_x86_ops->enable_nmi_window(vcpu);
6066 		else if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
6067 			kvm_x86_ops->enable_irq_window(vcpu);
6068 
6069 		if (kvm_lapic_enabled(vcpu)) {
6070 			/*
6071 			 * Update architecture specific hints for APIC
6072 			 * virtual interrupt delivery.
6073 			 */
6074 			if (kvm_x86_ops->hwapic_irr_update)
6075 				kvm_x86_ops->hwapic_irr_update(vcpu,
6076 					kvm_lapic_find_highest_irr(vcpu));
6077 			update_cr8_intercept(vcpu);
6078 			kvm_lapic_sync_to_vapic(vcpu);
6079 		}
6080 	}
6081 
6082 	r = kvm_mmu_reload(vcpu);
6083 	if (unlikely(r)) {
6084 		goto cancel_injection;
6085 	}
6086 
6087 	preempt_disable();
6088 
6089 	kvm_x86_ops->prepare_guest_switch(vcpu);
6090 	if (vcpu->fpu_active)
6091 		kvm_load_guest_fpu(vcpu);
6092 	kvm_load_guest_xcr0(vcpu);
6093 
6094 	vcpu->mode = IN_GUEST_MODE;
6095 
6096 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6097 
6098 	/* We should set ->mode before check ->requests,
6099 	 * see the comment in make_all_cpus_request.
6100 	 */
6101 	smp_mb__after_srcu_read_unlock();
6102 
6103 	local_irq_disable();
6104 
6105 	if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests
6106 	    || need_resched() || signal_pending(current)) {
6107 		vcpu->mode = OUTSIDE_GUEST_MODE;
6108 		smp_wmb();
6109 		local_irq_enable();
6110 		preempt_enable();
6111 		vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6112 		r = 1;
6113 		goto cancel_injection;
6114 	}
6115 
6116 	if (req_immediate_exit)
6117 		smp_send_reschedule(vcpu->cpu);
6118 
6119 	kvm_guest_enter();
6120 
6121 	if (unlikely(vcpu->arch.switch_db_regs)) {
6122 		set_debugreg(0, 7);
6123 		set_debugreg(vcpu->arch.eff_db[0], 0);
6124 		set_debugreg(vcpu->arch.eff_db[1], 1);
6125 		set_debugreg(vcpu->arch.eff_db[2], 2);
6126 		set_debugreg(vcpu->arch.eff_db[3], 3);
6127 		set_debugreg(vcpu->arch.dr6, 6);
6128 	}
6129 
6130 	trace_kvm_entry(vcpu->vcpu_id);
6131 	kvm_x86_ops->run(vcpu);
6132 
6133 	/*
6134 	 * Do this here before restoring debug registers on the host.  And
6135 	 * since we do this before handling the vmexit, a DR access vmexit
6136 	 * can (a) read the correct value of the debug registers, (b) set
6137 	 * KVM_DEBUGREG_WONT_EXIT again.
6138 	 */
6139 	if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
6140 		int i;
6141 
6142 		WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
6143 		kvm_x86_ops->sync_dirty_debug_regs(vcpu);
6144 		for (i = 0; i < KVM_NR_DB_REGS; i++)
6145 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
6146 	}
6147 
6148 	/*
6149 	 * If the guest has used debug registers, at least dr7
6150 	 * will be disabled while returning to the host.
6151 	 * If we don't have active breakpoints in the host, we don't
6152 	 * care about the messed up debug address registers. But if
6153 	 * we have some of them active, restore the old state.
6154 	 */
6155 	if (hw_breakpoint_active())
6156 		hw_breakpoint_restore();
6157 
6158 	vcpu->arch.last_guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu,
6159 							   native_read_tsc());
6160 
6161 	vcpu->mode = OUTSIDE_GUEST_MODE;
6162 	smp_wmb();
6163 
6164 	/* Interrupt is enabled by handle_external_intr() */
6165 	kvm_x86_ops->handle_external_intr(vcpu);
6166 
6167 	++vcpu->stat.exits;
6168 
6169 	/*
6170 	 * We must have an instruction between local_irq_enable() and
6171 	 * kvm_guest_exit(), so the timer interrupt isn't delayed by
6172 	 * the interrupt shadow.  The stat.exits increment will do nicely.
6173 	 * But we need to prevent reordering, hence this barrier():
6174 	 */
6175 	barrier();
6176 
6177 	kvm_guest_exit();
6178 
6179 	preempt_enable();
6180 
6181 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6182 
6183 	/*
6184 	 * Profile KVM exit RIPs:
6185 	 */
6186 	if (unlikely(prof_on == KVM_PROFILING)) {
6187 		unsigned long rip = kvm_rip_read(vcpu);
6188 		profile_hit(KVM_PROFILING, (void *)rip);
6189 	}
6190 
6191 	if (unlikely(vcpu->arch.tsc_always_catchup))
6192 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6193 
6194 	if (vcpu->arch.apic_attention)
6195 		kvm_lapic_sync_from_vapic(vcpu);
6196 
6197 	r = kvm_x86_ops->handle_exit(vcpu);
6198 	return r;
6199 
6200 cancel_injection:
6201 	kvm_x86_ops->cancel_injection(vcpu);
6202 	if (unlikely(vcpu->arch.apic_attention))
6203 		kvm_lapic_sync_from_vapic(vcpu);
6204 out:
6205 	return r;
6206 }
6207 
6208 
6209 static int __vcpu_run(struct kvm_vcpu *vcpu)
6210 {
6211 	int r;
6212 	struct kvm *kvm = vcpu->kvm;
6213 
6214 	vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6215 
6216 	r = 1;
6217 	while (r > 0) {
6218 		if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
6219 		    !vcpu->arch.apf.halted)
6220 			r = vcpu_enter_guest(vcpu);
6221 		else {
6222 			srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6223 			kvm_vcpu_block(vcpu);
6224 			vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6225 			if (kvm_check_request(KVM_REQ_UNHALT, vcpu)) {
6226 				kvm_apic_accept_events(vcpu);
6227 				switch(vcpu->arch.mp_state) {
6228 				case KVM_MP_STATE_HALTED:
6229 					vcpu->arch.pv.pv_unhalted = false;
6230 					vcpu->arch.mp_state =
6231 						KVM_MP_STATE_RUNNABLE;
6232 				case KVM_MP_STATE_RUNNABLE:
6233 					vcpu->arch.apf.halted = false;
6234 					break;
6235 				case KVM_MP_STATE_INIT_RECEIVED:
6236 					break;
6237 				default:
6238 					r = -EINTR;
6239 					break;
6240 				}
6241 			}
6242 		}
6243 
6244 		if (r <= 0)
6245 			break;
6246 
6247 		clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
6248 		if (kvm_cpu_has_pending_timer(vcpu))
6249 			kvm_inject_pending_timer_irqs(vcpu);
6250 
6251 		if (dm_request_for_irq_injection(vcpu)) {
6252 			r = -EINTR;
6253 			vcpu->run->exit_reason = KVM_EXIT_INTR;
6254 			++vcpu->stat.request_irq_exits;
6255 		}
6256 
6257 		kvm_check_async_pf_completion(vcpu);
6258 
6259 		if (signal_pending(current)) {
6260 			r = -EINTR;
6261 			vcpu->run->exit_reason = KVM_EXIT_INTR;
6262 			++vcpu->stat.signal_exits;
6263 		}
6264 		if (need_resched()) {
6265 			srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6266 			cond_resched();
6267 			vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6268 		}
6269 	}
6270 
6271 	srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6272 
6273 	return r;
6274 }
6275 
6276 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
6277 {
6278 	int r;
6279 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6280 	r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
6281 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6282 	if (r != EMULATE_DONE)
6283 		return 0;
6284 	return 1;
6285 }
6286 
6287 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
6288 {
6289 	BUG_ON(!vcpu->arch.pio.count);
6290 
6291 	return complete_emulated_io(vcpu);
6292 }
6293 
6294 /*
6295  * Implements the following, as a state machine:
6296  *
6297  * read:
6298  *   for each fragment
6299  *     for each mmio piece in the fragment
6300  *       write gpa, len
6301  *       exit
6302  *       copy data
6303  *   execute insn
6304  *
6305  * write:
6306  *   for each fragment
6307  *     for each mmio piece in the fragment
6308  *       write gpa, len
6309  *       copy data
6310  *       exit
6311  */
6312 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
6313 {
6314 	struct kvm_run *run = vcpu->run;
6315 	struct kvm_mmio_fragment *frag;
6316 	unsigned len;
6317 
6318 	BUG_ON(!vcpu->mmio_needed);
6319 
6320 	/* Complete previous fragment */
6321 	frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
6322 	len = min(8u, frag->len);
6323 	if (!vcpu->mmio_is_write)
6324 		memcpy(frag->data, run->mmio.data, len);
6325 
6326 	if (frag->len <= 8) {
6327 		/* Switch to the next fragment. */
6328 		frag++;
6329 		vcpu->mmio_cur_fragment++;
6330 	} else {
6331 		/* Go forward to the next mmio piece. */
6332 		frag->data += len;
6333 		frag->gpa += len;
6334 		frag->len -= len;
6335 	}
6336 
6337 	if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
6338 		vcpu->mmio_needed = 0;
6339 
6340 		/* FIXME: return into emulator if single-stepping.  */
6341 		if (vcpu->mmio_is_write)
6342 			return 1;
6343 		vcpu->mmio_read_completed = 1;
6344 		return complete_emulated_io(vcpu);
6345 	}
6346 
6347 	run->exit_reason = KVM_EXIT_MMIO;
6348 	run->mmio.phys_addr = frag->gpa;
6349 	if (vcpu->mmio_is_write)
6350 		memcpy(run->mmio.data, frag->data, min(8u, frag->len));
6351 	run->mmio.len = min(8u, frag->len);
6352 	run->mmio.is_write = vcpu->mmio_is_write;
6353 	vcpu->arch.complete_userspace_io = complete_emulated_mmio;
6354 	return 0;
6355 }
6356 
6357 
6358 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
6359 {
6360 	int r;
6361 	sigset_t sigsaved;
6362 
6363 	if (!tsk_used_math(current) && init_fpu(current))
6364 		return -ENOMEM;
6365 
6366 	if (vcpu->sigset_active)
6367 		sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
6368 
6369 	if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
6370 		kvm_vcpu_block(vcpu);
6371 		kvm_apic_accept_events(vcpu);
6372 		clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
6373 		r = -EAGAIN;
6374 		goto out;
6375 	}
6376 
6377 	/* re-sync apic's tpr */
6378 	if (!irqchip_in_kernel(vcpu->kvm)) {
6379 		if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
6380 			r = -EINVAL;
6381 			goto out;
6382 		}
6383 	}
6384 
6385 	if (unlikely(vcpu->arch.complete_userspace_io)) {
6386 		int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
6387 		vcpu->arch.complete_userspace_io = NULL;
6388 		r = cui(vcpu);
6389 		if (r <= 0)
6390 			goto out;
6391 	} else
6392 		WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
6393 
6394 	r = __vcpu_run(vcpu);
6395 
6396 out:
6397 	post_kvm_run_save(vcpu);
6398 	if (vcpu->sigset_active)
6399 		sigprocmask(SIG_SETMASK, &sigsaved, NULL);
6400 
6401 	return r;
6402 }
6403 
6404 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6405 {
6406 	if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
6407 		/*
6408 		 * We are here if userspace calls get_regs() in the middle of
6409 		 * instruction emulation. Registers state needs to be copied
6410 		 * back from emulation context to vcpu. Userspace shouldn't do
6411 		 * that usually, but some bad designed PV devices (vmware
6412 		 * backdoor interface) need this to work
6413 		 */
6414 		emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
6415 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6416 	}
6417 	regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
6418 	regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
6419 	regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
6420 	regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
6421 	regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
6422 	regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
6423 	regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
6424 	regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
6425 #ifdef CONFIG_X86_64
6426 	regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
6427 	regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
6428 	regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
6429 	regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
6430 	regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
6431 	regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
6432 	regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
6433 	regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
6434 #endif
6435 
6436 	regs->rip = kvm_rip_read(vcpu);
6437 	regs->rflags = kvm_get_rflags(vcpu);
6438 
6439 	return 0;
6440 }
6441 
6442 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6443 {
6444 	vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
6445 	vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6446 
6447 	kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
6448 	kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
6449 	kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
6450 	kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
6451 	kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
6452 	kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
6453 	kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
6454 	kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
6455 #ifdef CONFIG_X86_64
6456 	kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
6457 	kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
6458 	kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
6459 	kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
6460 	kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
6461 	kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
6462 	kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
6463 	kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
6464 #endif
6465 
6466 	kvm_rip_write(vcpu, regs->rip);
6467 	kvm_set_rflags(vcpu, regs->rflags);
6468 
6469 	vcpu->arch.exception.pending = false;
6470 
6471 	kvm_make_request(KVM_REQ_EVENT, vcpu);
6472 
6473 	return 0;
6474 }
6475 
6476 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
6477 {
6478 	struct kvm_segment cs;
6479 
6480 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
6481 	*db = cs.db;
6482 	*l = cs.l;
6483 }
6484 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
6485 
6486 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
6487 				  struct kvm_sregs *sregs)
6488 {
6489 	struct desc_ptr dt;
6490 
6491 	kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6492 	kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6493 	kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6494 	kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6495 	kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6496 	kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
6497 
6498 	kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6499 	kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
6500 
6501 	kvm_x86_ops->get_idt(vcpu, &dt);
6502 	sregs->idt.limit = dt.size;
6503 	sregs->idt.base = dt.address;
6504 	kvm_x86_ops->get_gdt(vcpu, &dt);
6505 	sregs->gdt.limit = dt.size;
6506 	sregs->gdt.base = dt.address;
6507 
6508 	sregs->cr0 = kvm_read_cr0(vcpu);
6509 	sregs->cr2 = vcpu->arch.cr2;
6510 	sregs->cr3 = kvm_read_cr3(vcpu);
6511 	sregs->cr4 = kvm_read_cr4(vcpu);
6512 	sregs->cr8 = kvm_get_cr8(vcpu);
6513 	sregs->efer = vcpu->arch.efer;
6514 	sregs->apic_base = kvm_get_apic_base(vcpu);
6515 
6516 	memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
6517 
6518 	if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
6519 		set_bit(vcpu->arch.interrupt.nr,
6520 			(unsigned long *)sregs->interrupt_bitmap);
6521 
6522 	return 0;
6523 }
6524 
6525 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
6526 				    struct kvm_mp_state *mp_state)
6527 {
6528 	kvm_apic_accept_events(vcpu);
6529 	if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
6530 					vcpu->arch.pv.pv_unhalted)
6531 		mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
6532 	else
6533 		mp_state->mp_state = vcpu->arch.mp_state;
6534 
6535 	return 0;
6536 }
6537 
6538 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
6539 				    struct kvm_mp_state *mp_state)
6540 {
6541 	if (!kvm_vcpu_has_lapic(vcpu) &&
6542 	    mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
6543 		return -EINVAL;
6544 
6545 	if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
6546 		vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
6547 		set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
6548 	} else
6549 		vcpu->arch.mp_state = mp_state->mp_state;
6550 	kvm_make_request(KVM_REQ_EVENT, vcpu);
6551 	return 0;
6552 }
6553 
6554 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
6555 		    int reason, bool has_error_code, u32 error_code)
6556 {
6557 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6558 	int ret;
6559 
6560 	init_emulate_ctxt(vcpu);
6561 
6562 	ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
6563 				   has_error_code, error_code);
6564 
6565 	if (ret)
6566 		return EMULATE_FAIL;
6567 
6568 	kvm_rip_write(vcpu, ctxt->eip);
6569 	kvm_set_rflags(vcpu, ctxt->eflags);
6570 	kvm_make_request(KVM_REQ_EVENT, vcpu);
6571 	return EMULATE_DONE;
6572 }
6573 EXPORT_SYMBOL_GPL(kvm_task_switch);
6574 
6575 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
6576 				  struct kvm_sregs *sregs)
6577 {
6578 	struct msr_data apic_base_msr;
6579 	int mmu_reset_needed = 0;
6580 	int pending_vec, max_bits, idx;
6581 	struct desc_ptr dt;
6582 
6583 	if (!guest_cpuid_has_xsave(vcpu) && (sregs->cr4 & X86_CR4_OSXSAVE))
6584 		return -EINVAL;
6585 
6586 	dt.size = sregs->idt.limit;
6587 	dt.address = sregs->idt.base;
6588 	kvm_x86_ops->set_idt(vcpu, &dt);
6589 	dt.size = sregs->gdt.limit;
6590 	dt.address = sregs->gdt.base;
6591 	kvm_x86_ops->set_gdt(vcpu, &dt);
6592 
6593 	vcpu->arch.cr2 = sregs->cr2;
6594 	mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
6595 	vcpu->arch.cr3 = sregs->cr3;
6596 	__set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
6597 
6598 	kvm_set_cr8(vcpu, sregs->cr8);
6599 
6600 	mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
6601 	kvm_x86_ops->set_efer(vcpu, sregs->efer);
6602 	apic_base_msr.data = sregs->apic_base;
6603 	apic_base_msr.host_initiated = true;
6604 	kvm_set_apic_base(vcpu, &apic_base_msr);
6605 
6606 	mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
6607 	kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
6608 	vcpu->arch.cr0 = sregs->cr0;
6609 
6610 	mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
6611 	kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
6612 	if (sregs->cr4 & X86_CR4_OSXSAVE)
6613 		kvm_update_cpuid(vcpu);
6614 
6615 	idx = srcu_read_lock(&vcpu->kvm->srcu);
6616 	if (!is_long_mode(vcpu) && is_pae(vcpu)) {
6617 		load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
6618 		mmu_reset_needed = 1;
6619 	}
6620 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
6621 
6622 	if (mmu_reset_needed)
6623 		kvm_mmu_reset_context(vcpu);
6624 
6625 	max_bits = KVM_NR_INTERRUPTS;
6626 	pending_vec = find_first_bit(
6627 		(const unsigned long *)sregs->interrupt_bitmap, max_bits);
6628 	if (pending_vec < max_bits) {
6629 		kvm_queue_interrupt(vcpu, pending_vec, false);
6630 		pr_debug("Set back pending irq %d\n", pending_vec);
6631 	}
6632 
6633 	kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6634 	kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6635 	kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6636 	kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6637 	kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6638 	kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
6639 
6640 	kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6641 	kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
6642 
6643 	update_cr8_intercept(vcpu);
6644 
6645 	/* Older userspace won't unhalt the vcpu on reset. */
6646 	if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
6647 	    sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
6648 	    !is_protmode(vcpu))
6649 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
6650 
6651 	kvm_make_request(KVM_REQ_EVENT, vcpu);
6652 
6653 	return 0;
6654 }
6655 
6656 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
6657 					struct kvm_guest_debug *dbg)
6658 {
6659 	unsigned long rflags;
6660 	int i, r;
6661 
6662 	if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
6663 		r = -EBUSY;
6664 		if (vcpu->arch.exception.pending)
6665 			goto out;
6666 		if (dbg->control & KVM_GUESTDBG_INJECT_DB)
6667 			kvm_queue_exception(vcpu, DB_VECTOR);
6668 		else
6669 			kvm_queue_exception(vcpu, BP_VECTOR);
6670 	}
6671 
6672 	/*
6673 	 * Read rflags as long as potentially injected trace flags are still
6674 	 * filtered out.
6675 	 */
6676 	rflags = kvm_get_rflags(vcpu);
6677 
6678 	vcpu->guest_debug = dbg->control;
6679 	if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
6680 		vcpu->guest_debug = 0;
6681 
6682 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
6683 		for (i = 0; i < KVM_NR_DB_REGS; ++i)
6684 			vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
6685 		vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
6686 	} else {
6687 		for (i = 0; i < KVM_NR_DB_REGS; i++)
6688 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
6689 	}
6690 	kvm_update_dr7(vcpu);
6691 
6692 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6693 		vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
6694 			get_segment_base(vcpu, VCPU_SREG_CS);
6695 
6696 	/*
6697 	 * Trigger an rflags update that will inject or remove the trace
6698 	 * flags.
6699 	 */
6700 	kvm_set_rflags(vcpu, rflags);
6701 
6702 	kvm_x86_ops->update_db_bp_intercept(vcpu);
6703 
6704 	r = 0;
6705 
6706 out:
6707 
6708 	return r;
6709 }
6710 
6711 /*
6712  * Translate a guest virtual address to a guest physical address.
6713  */
6714 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
6715 				    struct kvm_translation *tr)
6716 {
6717 	unsigned long vaddr = tr->linear_address;
6718 	gpa_t gpa;
6719 	int idx;
6720 
6721 	idx = srcu_read_lock(&vcpu->kvm->srcu);
6722 	gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
6723 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
6724 	tr->physical_address = gpa;
6725 	tr->valid = gpa != UNMAPPED_GVA;
6726 	tr->writeable = 1;
6727 	tr->usermode = 0;
6728 
6729 	return 0;
6730 }
6731 
6732 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
6733 {
6734 	struct i387_fxsave_struct *fxsave =
6735 			&vcpu->arch.guest_fpu.state->fxsave;
6736 
6737 	memcpy(fpu->fpr, fxsave->st_space, 128);
6738 	fpu->fcw = fxsave->cwd;
6739 	fpu->fsw = fxsave->swd;
6740 	fpu->ftwx = fxsave->twd;
6741 	fpu->last_opcode = fxsave->fop;
6742 	fpu->last_ip = fxsave->rip;
6743 	fpu->last_dp = fxsave->rdp;
6744 	memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
6745 
6746 	return 0;
6747 }
6748 
6749 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
6750 {
6751 	struct i387_fxsave_struct *fxsave =
6752 			&vcpu->arch.guest_fpu.state->fxsave;
6753 
6754 	memcpy(fxsave->st_space, fpu->fpr, 128);
6755 	fxsave->cwd = fpu->fcw;
6756 	fxsave->swd = fpu->fsw;
6757 	fxsave->twd = fpu->ftwx;
6758 	fxsave->fop = fpu->last_opcode;
6759 	fxsave->rip = fpu->last_ip;
6760 	fxsave->rdp = fpu->last_dp;
6761 	memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
6762 
6763 	return 0;
6764 }
6765 
6766 int fx_init(struct kvm_vcpu *vcpu)
6767 {
6768 	int err;
6769 
6770 	err = fpu_alloc(&vcpu->arch.guest_fpu);
6771 	if (err)
6772 		return err;
6773 
6774 	fpu_finit(&vcpu->arch.guest_fpu);
6775 
6776 	/*
6777 	 * Ensure guest xcr0 is valid for loading
6778 	 */
6779 	vcpu->arch.xcr0 = XSTATE_FP;
6780 
6781 	vcpu->arch.cr0 |= X86_CR0_ET;
6782 
6783 	return 0;
6784 }
6785 EXPORT_SYMBOL_GPL(fx_init);
6786 
6787 static void fx_free(struct kvm_vcpu *vcpu)
6788 {
6789 	fpu_free(&vcpu->arch.guest_fpu);
6790 }
6791 
6792 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
6793 {
6794 	if (vcpu->guest_fpu_loaded)
6795 		return;
6796 
6797 	/*
6798 	 * Restore all possible states in the guest,
6799 	 * and assume host would use all available bits.
6800 	 * Guest xcr0 would be loaded later.
6801 	 */
6802 	kvm_put_guest_xcr0(vcpu);
6803 	vcpu->guest_fpu_loaded = 1;
6804 	__kernel_fpu_begin();
6805 	fpu_restore_checking(&vcpu->arch.guest_fpu);
6806 	trace_kvm_fpu(1);
6807 }
6808 
6809 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
6810 {
6811 	kvm_put_guest_xcr0(vcpu);
6812 
6813 	if (!vcpu->guest_fpu_loaded)
6814 		return;
6815 
6816 	vcpu->guest_fpu_loaded = 0;
6817 	fpu_save_init(&vcpu->arch.guest_fpu);
6818 	__kernel_fpu_end();
6819 	++vcpu->stat.fpu_reload;
6820 	kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu);
6821 	trace_kvm_fpu(0);
6822 }
6823 
6824 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
6825 {
6826 	kvmclock_reset(vcpu);
6827 
6828 	free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
6829 	fx_free(vcpu);
6830 	kvm_x86_ops->vcpu_free(vcpu);
6831 }
6832 
6833 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
6834 						unsigned int id)
6835 {
6836 	if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
6837 		printk_once(KERN_WARNING
6838 		"kvm: SMP vm created on host with unstable TSC; "
6839 		"guest TSC will not be reliable\n");
6840 	return kvm_x86_ops->vcpu_create(kvm, id);
6841 }
6842 
6843 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
6844 {
6845 	int r;
6846 
6847 	vcpu->arch.mtrr_state.have_fixed = 1;
6848 	r = vcpu_load(vcpu);
6849 	if (r)
6850 		return r;
6851 	kvm_vcpu_reset(vcpu);
6852 	kvm_mmu_setup(vcpu);
6853 	vcpu_put(vcpu);
6854 
6855 	return r;
6856 }
6857 
6858 int kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
6859 {
6860 	int r;
6861 	struct msr_data msr;
6862 	struct kvm *kvm = vcpu->kvm;
6863 
6864 	r = vcpu_load(vcpu);
6865 	if (r)
6866 		return r;
6867 	msr.data = 0x0;
6868 	msr.index = MSR_IA32_TSC;
6869 	msr.host_initiated = true;
6870 	kvm_write_tsc(vcpu, &msr);
6871 	vcpu_put(vcpu);
6872 
6873 	schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
6874 					KVMCLOCK_SYNC_PERIOD);
6875 
6876 	return r;
6877 }
6878 
6879 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
6880 {
6881 	int r;
6882 	vcpu->arch.apf.msr_val = 0;
6883 
6884 	r = vcpu_load(vcpu);
6885 	BUG_ON(r);
6886 	kvm_mmu_unload(vcpu);
6887 	vcpu_put(vcpu);
6888 
6889 	fx_free(vcpu);
6890 	kvm_x86_ops->vcpu_free(vcpu);
6891 }
6892 
6893 void kvm_vcpu_reset(struct kvm_vcpu *vcpu)
6894 {
6895 	atomic_set(&vcpu->arch.nmi_queued, 0);
6896 	vcpu->arch.nmi_pending = 0;
6897 	vcpu->arch.nmi_injected = false;
6898 	kvm_clear_interrupt_queue(vcpu);
6899 	kvm_clear_exception_queue(vcpu);
6900 
6901 	memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
6902 	vcpu->arch.dr6 = DR6_INIT;
6903 	kvm_update_dr6(vcpu);
6904 	vcpu->arch.dr7 = DR7_FIXED_1;
6905 	kvm_update_dr7(vcpu);
6906 
6907 	kvm_make_request(KVM_REQ_EVENT, vcpu);
6908 	vcpu->arch.apf.msr_val = 0;
6909 	vcpu->arch.st.msr_val = 0;
6910 
6911 	kvmclock_reset(vcpu);
6912 
6913 	kvm_clear_async_pf_completion_queue(vcpu);
6914 	kvm_async_pf_hash_reset(vcpu);
6915 	vcpu->arch.apf.halted = false;
6916 
6917 	kvm_pmu_reset(vcpu);
6918 
6919 	memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
6920 	vcpu->arch.regs_avail = ~0;
6921 	vcpu->arch.regs_dirty = ~0;
6922 
6923 	kvm_x86_ops->vcpu_reset(vcpu);
6924 }
6925 
6926 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, unsigned int vector)
6927 {
6928 	struct kvm_segment cs;
6929 
6930 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
6931 	cs.selector = vector << 8;
6932 	cs.base = vector << 12;
6933 	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
6934 	kvm_rip_write(vcpu, 0);
6935 }
6936 
6937 int kvm_arch_hardware_enable(void *garbage)
6938 {
6939 	struct kvm *kvm;
6940 	struct kvm_vcpu *vcpu;
6941 	int i;
6942 	int ret;
6943 	u64 local_tsc;
6944 	u64 max_tsc = 0;
6945 	bool stable, backwards_tsc = false;
6946 
6947 	kvm_shared_msr_cpu_online();
6948 	ret = kvm_x86_ops->hardware_enable(garbage);
6949 	if (ret != 0)
6950 		return ret;
6951 
6952 	local_tsc = native_read_tsc();
6953 	stable = !check_tsc_unstable();
6954 	list_for_each_entry(kvm, &vm_list, vm_list) {
6955 		kvm_for_each_vcpu(i, vcpu, kvm) {
6956 			if (!stable && vcpu->cpu == smp_processor_id())
6957 				set_bit(KVM_REQ_CLOCK_UPDATE, &vcpu->requests);
6958 			if (stable && vcpu->arch.last_host_tsc > local_tsc) {
6959 				backwards_tsc = true;
6960 				if (vcpu->arch.last_host_tsc > max_tsc)
6961 					max_tsc = vcpu->arch.last_host_tsc;
6962 			}
6963 		}
6964 	}
6965 
6966 	/*
6967 	 * Sometimes, even reliable TSCs go backwards.  This happens on
6968 	 * platforms that reset TSC during suspend or hibernate actions, but
6969 	 * maintain synchronization.  We must compensate.  Fortunately, we can
6970 	 * detect that condition here, which happens early in CPU bringup,
6971 	 * before any KVM threads can be running.  Unfortunately, we can't
6972 	 * bring the TSCs fully up to date with real time, as we aren't yet far
6973 	 * enough into CPU bringup that we know how much real time has actually
6974 	 * elapsed; our helper function, get_kernel_ns() will be using boot
6975 	 * variables that haven't been updated yet.
6976 	 *
6977 	 * So we simply find the maximum observed TSC above, then record the
6978 	 * adjustment to TSC in each VCPU.  When the VCPU later gets loaded,
6979 	 * the adjustment will be applied.  Note that we accumulate
6980 	 * adjustments, in case multiple suspend cycles happen before some VCPU
6981 	 * gets a chance to run again.  In the event that no KVM threads get a
6982 	 * chance to run, we will miss the entire elapsed period, as we'll have
6983 	 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
6984 	 * loose cycle time.  This isn't too big a deal, since the loss will be
6985 	 * uniform across all VCPUs (not to mention the scenario is extremely
6986 	 * unlikely). It is possible that a second hibernate recovery happens
6987 	 * much faster than a first, causing the observed TSC here to be
6988 	 * smaller; this would require additional padding adjustment, which is
6989 	 * why we set last_host_tsc to the local tsc observed here.
6990 	 *
6991 	 * N.B. - this code below runs only on platforms with reliable TSC,
6992 	 * as that is the only way backwards_tsc is set above.  Also note
6993 	 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
6994 	 * have the same delta_cyc adjustment applied if backwards_tsc
6995 	 * is detected.  Note further, this adjustment is only done once,
6996 	 * as we reset last_host_tsc on all VCPUs to stop this from being
6997 	 * called multiple times (one for each physical CPU bringup).
6998 	 *
6999 	 * Platforms with unreliable TSCs don't have to deal with this, they
7000 	 * will be compensated by the logic in vcpu_load, which sets the TSC to
7001 	 * catchup mode.  This will catchup all VCPUs to real time, but cannot
7002 	 * guarantee that they stay in perfect synchronization.
7003 	 */
7004 	if (backwards_tsc) {
7005 		u64 delta_cyc = max_tsc - local_tsc;
7006 		backwards_tsc_observed = true;
7007 		list_for_each_entry(kvm, &vm_list, vm_list) {
7008 			kvm_for_each_vcpu(i, vcpu, kvm) {
7009 				vcpu->arch.tsc_offset_adjustment += delta_cyc;
7010 				vcpu->arch.last_host_tsc = local_tsc;
7011 				set_bit(KVM_REQ_MASTERCLOCK_UPDATE,
7012 					&vcpu->requests);
7013 			}
7014 
7015 			/*
7016 			 * We have to disable TSC offset matching.. if you were
7017 			 * booting a VM while issuing an S4 host suspend....
7018 			 * you may have some problem.  Solving this issue is
7019 			 * left as an exercise to the reader.
7020 			 */
7021 			kvm->arch.last_tsc_nsec = 0;
7022 			kvm->arch.last_tsc_write = 0;
7023 		}
7024 
7025 	}
7026 	return 0;
7027 }
7028 
7029 void kvm_arch_hardware_disable(void *garbage)
7030 {
7031 	kvm_x86_ops->hardware_disable(garbage);
7032 	drop_user_return_notifiers(garbage);
7033 }
7034 
7035 int kvm_arch_hardware_setup(void)
7036 {
7037 	return kvm_x86_ops->hardware_setup();
7038 }
7039 
7040 void kvm_arch_hardware_unsetup(void)
7041 {
7042 	kvm_x86_ops->hardware_unsetup();
7043 }
7044 
7045 void kvm_arch_check_processor_compat(void *rtn)
7046 {
7047 	kvm_x86_ops->check_processor_compatibility(rtn);
7048 }
7049 
7050 bool kvm_vcpu_compatible(struct kvm_vcpu *vcpu)
7051 {
7052 	return irqchip_in_kernel(vcpu->kvm) == (vcpu->arch.apic != NULL);
7053 }
7054 
7055 struct static_key kvm_no_apic_vcpu __read_mostly;
7056 
7057 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
7058 {
7059 	struct page *page;
7060 	struct kvm *kvm;
7061 	int r;
7062 
7063 	BUG_ON(vcpu->kvm == NULL);
7064 	kvm = vcpu->kvm;
7065 
7066 	vcpu->arch.pv.pv_unhalted = false;
7067 	vcpu->arch.emulate_ctxt.ops = &emulate_ops;
7068 	if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
7069 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7070 	else
7071 		vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
7072 
7073 	page = alloc_page(GFP_KERNEL | __GFP_ZERO);
7074 	if (!page) {
7075 		r = -ENOMEM;
7076 		goto fail;
7077 	}
7078 	vcpu->arch.pio_data = page_address(page);
7079 
7080 	kvm_set_tsc_khz(vcpu, max_tsc_khz);
7081 
7082 	r = kvm_mmu_create(vcpu);
7083 	if (r < 0)
7084 		goto fail_free_pio_data;
7085 
7086 	if (irqchip_in_kernel(kvm)) {
7087 		r = kvm_create_lapic(vcpu);
7088 		if (r < 0)
7089 			goto fail_mmu_destroy;
7090 	} else
7091 		static_key_slow_inc(&kvm_no_apic_vcpu);
7092 
7093 	vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
7094 				       GFP_KERNEL);
7095 	if (!vcpu->arch.mce_banks) {
7096 		r = -ENOMEM;
7097 		goto fail_free_lapic;
7098 	}
7099 	vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
7100 
7101 	if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL)) {
7102 		r = -ENOMEM;
7103 		goto fail_free_mce_banks;
7104 	}
7105 
7106 	r = fx_init(vcpu);
7107 	if (r)
7108 		goto fail_free_wbinvd_dirty_mask;
7109 
7110 	vcpu->arch.ia32_tsc_adjust_msr = 0x0;
7111 	vcpu->arch.pv_time_enabled = false;
7112 
7113 	vcpu->arch.guest_supported_xcr0 = 0;
7114 	vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
7115 
7116 	kvm_async_pf_hash_reset(vcpu);
7117 	kvm_pmu_init(vcpu);
7118 
7119 	return 0;
7120 fail_free_wbinvd_dirty_mask:
7121 	free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
7122 fail_free_mce_banks:
7123 	kfree(vcpu->arch.mce_banks);
7124 fail_free_lapic:
7125 	kvm_free_lapic(vcpu);
7126 fail_mmu_destroy:
7127 	kvm_mmu_destroy(vcpu);
7128 fail_free_pio_data:
7129 	free_page((unsigned long)vcpu->arch.pio_data);
7130 fail:
7131 	return r;
7132 }
7133 
7134 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
7135 {
7136 	int idx;
7137 
7138 	kvm_pmu_destroy(vcpu);
7139 	kfree(vcpu->arch.mce_banks);
7140 	kvm_free_lapic(vcpu);
7141 	idx = srcu_read_lock(&vcpu->kvm->srcu);
7142 	kvm_mmu_destroy(vcpu);
7143 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
7144 	free_page((unsigned long)vcpu->arch.pio_data);
7145 	if (!irqchip_in_kernel(vcpu->kvm))
7146 		static_key_slow_dec(&kvm_no_apic_vcpu);
7147 }
7148 
7149 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
7150 {
7151 	if (type)
7152 		return -EINVAL;
7153 
7154 	INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
7155 	INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
7156 	INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
7157 	atomic_set(&kvm->arch.noncoherent_dma_count, 0);
7158 
7159 	/* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
7160 	set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
7161 	/* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
7162 	set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
7163 		&kvm->arch.irq_sources_bitmap);
7164 
7165 	raw_spin_lock_init(&kvm->arch.tsc_write_lock);
7166 	mutex_init(&kvm->arch.apic_map_lock);
7167 	spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
7168 
7169 	pvclock_update_vm_gtod_copy(kvm);
7170 
7171 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
7172 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
7173 
7174 	return 0;
7175 }
7176 
7177 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
7178 {
7179 	int r;
7180 	r = vcpu_load(vcpu);
7181 	BUG_ON(r);
7182 	kvm_mmu_unload(vcpu);
7183 	vcpu_put(vcpu);
7184 }
7185 
7186 static void kvm_free_vcpus(struct kvm *kvm)
7187 {
7188 	unsigned int i;
7189 	struct kvm_vcpu *vcpu;
7190 
7191 	/*
7192 	 * Unpin any mmu pages first.
7193 	 */
7194 	kvm_for_each_vcpu(i, vcpu, kvm) {
7195 		kvm_clear_async_pf_completion_queue(vcpu);
7196 		kvm_unload_vcpu_mmu(vcpu);
7197 	}
7198 	kvm_for_each_vcpu(i, vcpu, kvm)
7199 		kvm_arch_vcpu_free(vcpu);
7200 
7201 	mutex_lock(&kvm->lock);
7202 	for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
7203 		kvm->vcpus[i] = NULL;
7204 
7205 	atomic_set(&kvm->online_vcpus, 0);
7206 	mutex_unlock(&kvm->lock);
7207 }
7208 
7209 void kvm_arch_sync_events(struct kvm *kvm)
7210 {
7211 	cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
7212 	cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
7213 	kvm_free_all_assigned_devices(kvm);
7214 	kvm_free_pit(kvm);
7215 }
7216 
7217 void kvm_arch_destroy_vm(struct kvm *kvm)
7218 {
7219 	if (current->mm == kvm->mm) {
7220 		/*
7221 		 * Free memory regions allocated on behalf of userspace,
7222 		 * unless the the memory map has changed due to process exit
7223 		 * or fd copying.
7224 		 */
7225 		struct kvm_userspace_memory_region mem;
7226 		memset(&mem, 0, sizeof(mem));
7227 		mem.slot = APIC_ACCESS_PAGE_PRIVATE_MEMSLOT;
7228 		kvm_set_memory_region(kvm, &mem);
7229 
7230 		mem.slot = IDENTITY_PAGETABLE_PRIVATE_MEMSLOT;
7231 		kvm_set_memory_region(kvm, &mem);
7232 
7233 		mem.slot = TSS_PRIVATE_MEMSLOT;
7234 		kvm_set_memory_region(kvm, &mem);
7235 	}
7236 	kvm_iommu_unmap_guest(kvm);
7237 	kfree(kvm->arch.vpic);
7238 	kfree(kvm->arch.vioapic);
7239 	kvm_free_vcpus(kvm);
7240 	if (kvm->arch.apic_access_page)
7241 		put_page(kvm->arch.apic_access_page);
7242 	if (kvm->arch.ept_identity_pagetable)
7243 		put_page(kvm->arch.ept_identity_pagetable);
7244 	kfree(rcu_dereference_check(kvm->arch.apic_map, 1));
7245 }
7246 
7247 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
7248 			   struct kvm_memory_slot *dont)
7249 {
7250 	int i;
7251 
7252 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7253 		if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
7254 			kvm_kvfree(free->arch.rmap[i]);
7255 			free->arch.rmap[i] = NULL;
7256 		}
7257 		if (i == 0)
7258 			continue;
7259 
7260 		if (!dont || free->arch.lpage_info[i - 1] !=
7261 			     dont->arch.lpage_info[i - 1]) {
7262 			kvm_kvfree(free->arch.lpage_info[i - 1]);
7263 			free->arch.lpage_info[i - 1] = NULL;
7264 		}
7265 	}
7266 }
7267 
7268 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
7269 			    unsigned long npages)
7270 {
7271 	int i;
7272 
7273 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7274 		unsigned long ugfn;
7275 		int lpages;
7276 		int level = i + 1;
7277 
7278 		lpages = gfn_to_index(slot->base_gfn + npages - 1,
7279 				      slot->base_gfn, level) + 1;
7280 
7281 		slot->arch.rmap[i] =
7282 			kvm_kvzalloc(lpages * sizeof(*slot->arch.rmap[i]));
7283 		if (!slot->arch.rmap[i])
7284 			goto out_free;
7285 		if (i == 0)
7286 			continue;
7287 
7288 		slot->arch.lpage_info[i - 1] = kvm_kvzalloc(lpages *
7289 					sizeof(*slot->arch.lpage_info[i - 1]));
7290 		if (!slot->arch.lpage_info[i - 1])
7291 			goto out_free;
7292 
7293 		if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
7294 			slot->arch.lpage_info[i - 1][0].write_count = 1;
7295 		if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
7296 			slot->arch.lpage_info[i - 1][lpages - 1].write_count = 1;
7297 		ugfn = slot->userspace_addr >> PAGE_SHIFT;
7298 		/*
7299 		 * If the gfn and userspace address are not aligned wrt each
7300 		 * other, or if explicitly asked to, disable large page
7301 		 * support for this slot
7302 		 */
7303 		if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
7304 		    !kvm_largepages_enabled()) {
7305 			unsigned long j;
7306 
7307 			for (j = 0; j < lpages; ++j)
7308 				slot->arch.lpage_info[i - 1][j].write_count = 1;
7309 		}
7310 	}
7311 
7312 	return 0;
7313 
7314 out_free:
7315 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7316 		kvm_kvfree(slot->arch.rmap[i]);
7317 		slot->arch.rmap[i] = NULL;
7318 		if (i == 0)
7319 			continue;
7320 
7321 		kvm_kvfree(slot->arch.lpage_info[i - 1]);
7322 		slot->arch.lpage_info[i - 1] = NULL;
7323 	}
7324 	return -ENOMEM;
7325 }
7326 
7327 void kvm_arch_memslots_updated(struct kvm *kvm)
7328 {
7329 	/*
7330 	 * memslots->generation has been incremented.
7331 	 * mmio generation may have reached its maximum value.
7332 	 */
7333 	kvm_mmu_invalidate_mmio_sptes(kvm);
7334 }
7335 
7336 int kvm_arch_prepare_memory_region(struct kvm *kvm,
7337 				struct kvm_memory_slot *memslot,
7338 				struct kvm_userspace_memory_region *mem,
7339 				enum kvm_mr_change change)
7340 {
7341 	/*
7342 	 * Only private memory slots need to be mapped here since
7343 	 * KVM_SET_MEMORY_REGION ioctl is no longer supported.
7344 	 */
7345 	if ((memslot->id >= KVM_USER_MEM_SLOTS) && (change == KVM_MR_CREATE)) {
7346 		unsigned long userspace_addr;
7347 
7348 		/*
7349 		 * MAP_SHARED to prevent internal slot pages from being moved
7350 		 * by fork()/COW.
7351 		 */
7352 		userspace_addr = vm_mmap(NULL, 0, memslot->npages * PAGE_SIZE,
7353 					 PROT_READ | PROT_WRITE,
7354 					 MAP_SHARED | MAP_ANONYMOUS, 0);
7355 
7356 		if (IS_ERR((void *)userspace_addr))
7357 			return PTR_ERR((void *)userspace_addr);
7358 
7359 		memslot->userspace_addr = userspace_addr;
7360 	}
7361 
7362 	return 0;
7363 }
7364 
7365 void kvm_arch_commit_memory_region(struct kvm *kvm,
7366 				struct kvm_userspace_memory_region *mem,
7367 				const struct kvm_memory_slot *old,
7368 				enum kvm_mr_change change)
7369 {
7370 
7371 	int nr_mmu_pages = 0;
7372 
7373 	if ((mem->slot >= KVM_USER_MEM_SLOTS) && (change == KVM_MR_DELETE)) {
7374 		int ret;
7375 
7376 		ret = vm_munmap(old->userspace_addr,
7377 				old->npages * PAGE_SIZE);
7378 		if (ret < 0)
7379 			printk(KERN_WARNING
7380 			       "kvm_vm_ioctl_set_memory_region: "
7381 			       "failed to munmap memory\n");
7382 	}
7383 
7384 	if (!kvm->arch.n_requested_mmu_pages)
7385 		nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
7386 
7387 	if (nr_mmu_pages)
7388 		kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
7389 	/*
7390 	 * Write protect all pages for dirty logging.
7391 	 *
7392 	 * All the sptes including the large sptes which point to this
7393 	 * slot are set to readonly. We can not create any new large
7394 	 * spte on this slot until the end of the logging.
7395 	 *
7396 	 * See the comments in fast_page_fault().
7397 	 */
7398 	if ((change != KVM_MR_DELETE) && (mem->flags & KVM_MEM_LOG_DIRTY_PAGES))
7399 		kvm_mmu_slot_remove_write_access(kvm, mem->slot);
7400 }
7401 
7402 void kvm_arch_flush_shadow_all(struct kvm *kvm)
7403 {
7404 	kvm_mmu_invalidate_zap_all_pages(kvm);
7405 }
7406 
7407 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
7408 				   struct kvm_memory_slot *slot)
7409 {
7410 	kvm_mmu_invalidate_zap_all_pages(kvm);
7411 }
7412 
7413 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
7414 {
7415 	if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
7416 		kvm_x86_ops->check_nested_events(vcpu, false);
7417 
7418 	return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
7419 		!vcpu->arch.apf.halted)
7420 		|| !list_empty_careful(&vcpu->async_pf.done)
7421 		|| kvm_apic_has_events(vcpu)
7422 		|| vcpu->arch.pv.pv_unhalted
7423 		|| atomic_read(&vcpu->arch.nmi_queued) ||
7424 		(kvm_arch_interrupt_allowed(vcpu) &&
7425 		 kvm_cpu_has_interrupt(vcpu));
7426 }
7427 
7428 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
7429 {
7430 	return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
7431 }
7432 
7433 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
7434 {
7435 	return kvm_x86_ops->interrupt_allowed(vcpu);
7436 }
7437 
7438 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
7439 {
7440 	unsigned long current_rip = kvm_rip_read(vcpu) +
7441 		get_segment_base(vcpu, VCPU_SREG_CS);
7442 
7443 	return current_rip == linear_rip;
7444 }
7445 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
7446 
7447 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
7448 {
7449 	unsigned long rflags;
7450 
7451 	rflags = kvm_x86_ops->get_rflags(vcpu);
7452 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
7453 		rflags &= ~X86_EFLAGS_TF;
7454 	return rflags;
7455 }
7456 EXPORT_SYMBOL_GPL(kvm_get_rflags);
7457 
7458 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
7459 {
7460 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
7461 	    kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
7462 		rflags |= X86_EFLAGS_TF;
7463 	kvm_x86_ops->set_rflags(vcpu, rflags);
7464 }
7465 
7466 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
7467 {
7468 	__kvm_set_rflags(vcpu, rflags);
7469 	kvm_make_request(KVM_REQ_EVENT, vcpu);
7470 }
7471 EXPORT_SYMBOL_GPL(kvm_set_rflags);
7472 
7473 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
7474 {
7475 	int r;
7476 
7477 	if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
7478 	      work->wakeup_all)
7479 		return;
7480 
7481 	r = kvm_mmu_reload(vcpu);
7482 	if (unlikely(r))
7483 		return;
7484 
7485 	if (!vcpu->arch.mmu.direct_map &&
7486 	      work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
7487 		return;
7488 
7489 	vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
7490 }
7491 
7492 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
7493 {
7494 	return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
7495 }
7496 
7497 static inline u32 kvm_async_pf_next_probe(u32 key)
7498 {
7499 	return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
7500 }
7501 
7502 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7503 {
7504 	u32 key = kvm_async_pf_hash_fn(gfn);
7505 
7506 	while (vcpu->arch.apf.gfns[key] != ~0)
7507 		key = kvm_async_pf_next_probe(key);
7508 
7509 	vcpu->arch.apf.gfns[key] = gfn;
7510 }
7511 
7512 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
7513 {
7514 	int i;
7515 	u32 key = kvm_async_pf_hash_fn(gfn);
7516 
7517 	for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
7518 		     (vcpu->arch.apf.gfns[key] != gfn &&
7519 		      vcpu->arch.apf.gfns[key] != ~0); i++)
7520 		key = kvm_async_pf_next_probe(key);
7521 
7522 	return key;
7523 }
7524 
7525 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7526 {
7527 	return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
7528 }
7529 
7530 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7531 {
7532 	u32 i, j, k;
7533 
7534 	i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
7535 	while (true) {
7536 		vcpu->arch.apf.gfns[i] = ~0;
7537 		do {
7538 			j = kvm_async_pf_next_probe(j);
7539 			if (vcpu->arch.apf.gfns[j] == ~0)
7540 				return;
7541 			k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
7542 			/*
7543 			 * k lies cyclically in ]i,j]
7544 			 * |    i.k.j |
7545 			 * |....j i.k.| or  |.k..j i...|
7546 			 */
7547 		} while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
7548 		vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
7549 		i = j;
7550 	}
7551 }
7552 
7553 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
7554 {
7555 
7556 	return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
7557 				      sizeof(val));
7558 }
7559 
7560 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
7561 				     struct kvm_async_pf *work)
7562 {
7563 	struct x86_exception fault;
7564 
7565 	trace_kvm_async_pf_not_present(work->arch.token, work->gva);
7566 	kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
7567 
7568 	if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
7569 	    (vcpu->arch.apf.send_user_only &&
7570 	     kvm_x86_ops->get_cpl(vcpu) == 0))
7571 		kvm_make_request(KVM_REQ_APF_HALT, vcpu);
7572 	else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
7573 		fault.vector = PF_VECTOR;
7574 		fault.error_code_valid = true;
7575 		fault.error_code = 0;
7576 		fault.nested_page_fault = false;
7577 		fault.address = work->arch.token;
7578 		kvm_inject_page_fault(vcpu, &fault);
7579 	}
7580 }
7581 
7582 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
7583 				 struct kvm_async_pf *work)
7584 {
7585 	struct x86_exception fault;
7586 
7587 	trace_kvm_async_pf_ready(work->arch.token, work->gva);
7588 	if (work->wakeup_all)
7589 		work->arch.token = ~0; /* broadcast wakeup */
7590 	else
7591 		kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
7592 
7593 	if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
7594 	    !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
7595 		fault.vector = PF_VECTOR;
7596 		fault.error_code_valid = true;
7597 		fault.error_code = 0;
7598 		fault.nested_page_fault = false;
7599 		fault.address = work->arch.token;
7600 		kvm_inject_page_fault(vcpu, &fault);
7601 	}
7602 	vcpu->arch.apf.halted = false;
7603 	vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7604 }
7605 
7606 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
7607 {
7608 	if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
7609 		return true;
7610 	else
7611 		return !kvm_event_needs_reinjection(vcpu) &&
7612 			kvm_x86_ops->interrupt_allowed(vcpu);
7613 }
7614 
7615 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
7616 {
7617 	atomic_inc(&kvm->arch.noncoherent_dma_count);
7618 }
7619 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
7620 
7621 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
7622 {
7623 	atomic_dec(&kvm->arch.noncoherent_dma_count);
7624 }
7625 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
7626 
7627 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
7628 {
7629 	return atomic_read(&kvm->arch.noncoherent_dma_count);
7630 }
7631 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
7632 
7633 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
7634 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
7635 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
7636 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
7637 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
7638 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
7639 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
7640 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
7641 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
7642 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
7643 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
7644 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
7645 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
7646