xref: /linux/arch/x86/kvm/x86.c (revision b889fcf63cb62e7fdb7816565e28f44dbe4a76a5)
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 
91 struct kvm_x86_ops *kvm_x86_ops;
92 EXPORT_SYMBOL_GPL(kvm_x86_ops);
93 
94 static bool ignore_msrs = 0;
95 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
96 
97 bool kvm_has_tsc_control;
98 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
99 u32  kvm_max_guest_tsc_khz;
100 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
101 
102 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
103 static u32 tsc_tolerance_ppm = 250;
104 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
105 
106 #define KVM_NR_SHARED_MSRS 16
107 
108 struct kvm_shared_msrs_global {
109 	int nr;
110 	u32 msrs[KVM_NR_SHARED_MSRS];
111 };
112 
113 struct kvm_shared_msrs {
114 	struct user_return_notifier urn;
115 	bool registered;
116 	struct kvm_shared_msr_values {
117 		u64 host;
118 		u64 curr;
119 	} values[KVM_NR_SHARED_MSRS];
120 };
121 
122 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
123 static DEFINE_PER_CPU(struct kvm_shared_msrs, shared_msrs);
124 
125 struct kvm_stats_debugfs_item debugfs_entries[] = {
126 	{ "pf_fixed", VCPU_STAT(pf_fixed) },
127 	{ "pf_guest", VCPU_STAT(pf_guest) },
128 	{ "tlb_flush", VCPU_STAT(tlb_flush) },
129 	{ "invlpg", VCPU_STAT(invlpg) },
130 	{ "exits", VCPU_STAT(exits) },
131 	{ "io_exits", VCPU_STAT(io_exits) },
132 	{ "mmio_exits", VCPU_STAT(mmio_exits) },
133 	{ "signal_exits", VCPU_STAT(signal_exits) },
134 	{ "irq_window", VCPU_STAT(irq_window_exits) },
135 	{ "nmi_window", VCPU_STAT(nmi_window_exits) },
136 	{ "halt_exits", VCPU_STAT(halt_exits) },
137 	{ "halt_wakeup", VCPU_STAT(halt_wakeup) },
138 	{ "hypercalls", VCPU_STAT(hypercalls) },
139 	{ "request_irq", VCPU_STAT(request_irq_exits) },
140 	{ "irq_exits", VCPU_STAT(irq_exits) },
141 	{ "host_state_reload", VCPU_STAT(host_state_reload) },
142 	{ "efer_reload", VCPU_STAT(efer_reload) },
143 	{ "fpu_reload", VCPU_STAT(fpu_reload) },
144 	{ "insn_emulation", VCPU_STAT(insn_emulation) },
145 	{ "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
146 	{ "irq_injections", VCPU_STAT(irq_injections) },
147 	{ "nmi_injections", VCPU_STAT(nmi_injections) },
148 	{ "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
149 	{ "mmu_pte_write", VM_STAT(mmu_pte_write) },
150 	{ "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
151 	{ "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
152 	{ "mmu_flooded", VM_STAT(mmu_flooded) },
153 	{ "mmu_recycled", VM_STAT(mmu_recycled) },
154 	{ "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
155 	{ "mmu_unsync", VM_STAT(mmu_unsync) },
156 	{ "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
157 	{ "largepages", VM_STAT(lpages) },
158 	{ NULL }
159 };
160 
161 u64 __read_mostly host_xcr0;
162 
163 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
164 
165 static int kvm_vcpu_reset(struct kvm_vcpu *vcpu);
166 
167 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
168 {
169 	int i;
170 	for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
171 		vcpu->arch.apf.gfns[i] = ~0;
172 }
173 
174 static void kvm_on_user_return(struct user_return_notifier *urn)
175 {
176 	unsigned slot;
177 	struct kvm_shared_msrs *locals
178 		= container_of(urn, struct kvm_shared_msrs, urn);
179 	struct kvm_shared_msr_values *values;
180 
181 	for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
182 		values = &locals->values[slot];
183 		if (values->host != values->curr) {
184 			wrmsrl(shared_msrs_global.msrs[slot], values->host);
185 			values->curr = values->host;
186 		}
187 	}
188 	locals->registered = false;
189 	user_return_notifier_unregister(urn);
190 }
191 
192 static void shared_msr_update(unsigned slot, u32 msr)
193 {
194 	struct kvm_shared_msrs *smsr;
195 	u64 value;
196 
197 	smsr = &__get_cpu_var(shared_msrs);
198 	/* only read, and nobody should modify it at this time,
199 	 * so don't need lock */
200 	if (slot >= shared_msrs_global.nr) {
201 		printk(KERN_ERR "kvm: invalid MSR slot!");
202 		return;
203 	}
204 	rdmsrl_safe(msr, &value);
205 	smsr->values[slot].host = value;
206 	smsr->values[slot].curr = value;
207 }
208 
209 void kvm_define_shared_msr(unsigned slot, u32 msr)
210 {
211 	if (slot >= shared_msrs_global.nr)
212 		shared_msrs_global.nr = slot + 1;
213 	shared_msrs_global.msrs[slot] = msr;
214 	/* we need ensured the shared_msr_global have been updated */
215 	smp_wmb();
216 }
217 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
218 
219 static void kvm_shared_msr_cpu_online(void)
220 {
221 	unsigned i;
222 
223 	for (i = 0; i < shared_msrs_global.nr; ++i)
224 		shared_msr_update(i, shared_msrs_global.msrs[i]);
225 }
226 
227 void kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
228 {
229 	struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
230 
231 	if (((value ^ smsr->values[slot].curr) & mask) == 0)
232 		return;
233 	smsr->values[slot].curr = value;
234 	wrmsrl(shared_msrs_global.msrs[slot], value);
235 	if (!smsr->registered) {
236 		smsr->urn.on_user_return = kvm_on_user_return;
237 		user_return_notifier_register(&smsr->urn);
238 		smsr->registered = true;
239 	}
240 }
241 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
242 
243 static void drop_user_return_notifiers(void *ignore)
244 {
245 	struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
246 
247 	if (smsr->registered)
248 		kvm_on_user_return(&smsr->urn);
249 }
250 
251 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
252 {
253 	return vcpu->arch.apic_base;
254 }
255 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
256 
257 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
258 {
259 	/* TODO: reserve bits check */
260 	kvm_lapic_set_base(vcpu, data);
261 }
262 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
263 
264 #define EXCPT_BENIGN		0
265 #define EXCPT_CONTRIBUTORY	1
266 #define EXCPT_PF		2
267 
268 static int exception_class(int vector)
269 {
270 	switch (vector) {
271 	case PF_VECTOR:
272 		return EXCPT_PF;
273 	case DE_VECTOR:
274 	case TS_VECTOR:
275 	case NP_VECTOR:
276 	case SS_VECTOR:
277 	case GP_VECTOR:
278 		return EXCPT_CONTRIBUTORY;
279 	default:
280 		break;
281 	}
282 	return EXCPT_BENIGN;
283 }
284 
285 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
286 		unsigned nr, bool has_error, u32 error_code,
287 		bool reinject)
288 {
289 	u32 prev_nr;
290 	int class1, class2;
291 
292 	kvm_make_request(KVM_REQ_EVENT, vcpu);
293 
294 	if (!vcpu->arch.exception.pending) {
295 	queue:
296 		vcpu->arch.exception.pending = true;
297 		vcpu->arch.exception.has_error_code = has_error;
298 		vcpu->arch.exception.nr = nr;
299 		vcpu->arch.exception.error_code = error_code;
300 		vcpu->arch.exception.reinject = reinject;
301 		return;
302 	}
303 
304 	/* to check exception */
305 	prev_nr = vcpu->arch.exception.nr;
306 	if (prev_nr == DF_VECTOR) {
307 		/* triple fault -> shutdown */
308 		kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
309 		return;
310 	}
311 	class1 = exception_class(prev_nr);
312 	class2 = exception_class(nr);
313 	if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
314 		|| (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
315 		/* generate double fault per SDM Table 5-5 */
316 		vcpu->arch.exception.pending = true;
317 		vcpu->arch.exception.has_error_code = true;
318 		vcpu->arch.exception.nr = DF_VECTOR;
319 		vcpu->arch.exception.error_code = 0;
320 	} else
321 		/* replace previous exception with a new one in a hope
322 		   that instruction re-execution will regenerate lost
323 		   exception */
324 		goto queue;
325 }
326 
327 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
328 {
329 	kvm_multiple_exception(vcpu, nr, false, 0, false);
330 }
331 EXPORT_SYMBOL_GPL(kvm_queue_exception);
332 
333 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
334 {
335 	kvm_multiple_exception(vcpu, nr, false, 0, true);
336 }
337 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
338 
339 void kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
340 {
341 	if (err)
342 		kvm_inject_gp(vcpu, 0);
343 	else
344 		kvm_x86_ops->skip_emulated_instruction(vcpu);
345 }
346 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
347 
348 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
349 {
350 	++vcpu->stat.pf_guest;
351 	vcpu->arch.cr2 = fault->address;
352 	kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
353 }
354 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
355 
356 void kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
357 {
358 	if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
359 		vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
360 	else
361 		vcpu->arch.mmu.inject_page_fault(vcpu, fault);
362 }
363 
364 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
365 {
366 	atomic_inc(&vcpu->arch.nmi_queued);
367 	kvm_make_request(KVM_REQ_NMI, vcpu);
368 }
369 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
370 
371 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
372 {
373 	kvm_multiple_exception(vcpu, nr, true, error_code, false);
374 }
375 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
376 
377 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
378 {
379 	kvm_multiple_exception(vcpu, nr, true, error_code, true);
380 }
381 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
382 
383 /*
384  * Checks if cpl <= required_cpl; if true, return true.  Otherwise queue
385  * a #GP and return false.
386  */
387 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
388 {
389 	if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
390 		return true;
391 	kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
392 	return false;
393 }
394 EXPORT_SYMBOL_GPL(kvm_require_cpl);
395 
396 /*
397  * This function will be used to read from the physical memory of the currently
398  * running guest. The difference to kvm_read_guest_page is that this function
399  * can read from guest physical or from the guest's guest physical memory.
400  */
401 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
402 			    gfn_t ngfn, void *data, int offset, int len,
403 			    u32 access)
404 {
405 	gfn_t real_gfn;
406 	gpa_t ngpa;
407 
408 	ngpa     = gfn_to_gpa(ngfn);
409 	real_gfn = mmu->translate_gpa(vcpu, ngpa, access);
410 	if (real_gfn == UNMAPPED_GVA)
411 		return -EFAULT;
412 
413 	real_gfn = gpa_to_gfn(real_gfn);
414 
415 	return kvm_read_guest_page(vcpu->kvm, real_gfn, data, offset, len);
416 }
417 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
418 
419 int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
420 			       void *data, int offset, int len, u32 access)
421 {
422 	return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
423 				       data, offset, len, access);
424 }
425 
426 /*
427  * Load the pae pdptrs.  Return true is they are all valid.
428  */
429 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
430 {
431 	gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
432 	unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
433 	int i;
434 	int ret;
435 	u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
436 
437 	ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
438 				      offset * sizeof(u64), sizeof(pdpte),
439 				      PFERR_USER_MASK|PFERR_WRITE_MASK);
440 	if (ret < 0) {
441 		ret = 0;
442 		goto out;
443 	}
444 	for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
445 		if (is_present_gpte(pdpte[i]) &&
446 		    (pdpte[i] & vcpu->arch.mmu.rsvd_bits_mask[0][2])) {
447 			ret = 0;
448 			goto out;
449 		}
450 	}
451 	ret = 1;
452 
453 	memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
454 	__set_bit(VCPU_EXREG_PDPTR,
455 		  (unsigned long *)&vcpu->arch.regs_avail);
456 	__set_bit(VCPU_EXREG_PDPTR,
457 		  (unsigned long *)&vcpu->arch.regs_dirty);
458 out:
459 
460 	return ret;
461 }
462 EXPORT_SYMBOL_GPL(load_pdptrs);
463 
464 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
465 {
466 	u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
467 	bool changed = true;
468 	int offset;
469 	gfn_t gfn;
470 	int r;
471 
472 	if (is_long_mode(vcpu) || !is_pae(vcpu))
473 		return false;
474 
475 	if (!test_bit(VCPU_EXREG_PDPTR,
476 		      (unsigned long *)&vcpu->arch.regs_avail))
477 		return true;
478 
479 	gfn = (kvm_read_cr3(vcpu) & ~31u) >> PAGE_SHIFT;
480 	offset = (kvm_read_cr3(vcpu) & ~31u) & (PAGE_SIZE - 1);
481 	r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
482 				       PFERR_USER_MASK | PFERR_WRITE_MASK);
483 	if (r < 0)
484 		goto out;
485 	changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
486 out:
487 
488 	return changed;
489 }
490 
491 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
492 {
493 	unsigned long old_cr0 = kvm_read_cr0(vcpu);
494 	unsigned long update_bits = X86_CR0_PG | X86_CR0_WP |
495 				    X86_CR0_CD | X86_CR0_NW;
496 
497 	cr0 |= X86_CR0_ET;
498 
499 #ifdef CONFIG_X86_64
500 	if (cr0 & 0xffffffff00000000UL)
501 		return 1;
502 #endif
503 
504 	cr0 &= ~CR0_RESERVED_BITS;
505 
506 	if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
507 		return 1;
508 
509 	if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
510 		return 1;
511 
512 	if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
513 #ifdef CONFIG_X86_64
514 		if ((vcpu->arch.efer & EFER_LME)) {
515 			int cs_db, cs_l;
516 
517 			if (!is_pae(vcpu))
518 				return 1;
519 			kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
520 			if (cs_l)
521 				return 1;
522 		} else
523 #endif
524 		if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
525 						 kvm_read_cr3(vcpu)))
526 			return 1;
527 	}
528 
529 	if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
530 		return 1;
531 
532 	kvm_x86_ops->set_cr0(vcpu, cr0);
533 
534 	if ((cr0 ^ old_cr0) & X86_CR0_PG) {
535 		kvm_clear_async_pf_completion_queue(vcpu);
536 		kvm_async_pf_hash_reset(vcpu);
537 	}
538 
539 	if ((cr0 ^ old_cr0) & update_bits)
540 		kvm_mmu_reset_context(vcpu);
541 	return 0;
542 }
543 EXPORT_SYMBOL_GPL(kvm_set_cr0);
544 
545 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
546 {
547 	(void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
548 }
549 EXPORT_SYMBOL_GPL(kvm_lmsw);
550 
551 int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
552 {
553 	u64 xcr0;
554 
555 	/* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now  */
556 	if (index != XCR_XFEATURE_ENABLED_MASK)
557 		return 1;
558 	xcr0 = xcr;
559 	if (kvm_x86_ops->get_cpl(vcpu) != 0)
560 		return 1;
561 	if (!(xcr0 & XSTATE_FP))
562 		return 1;
563 	if ((xcr0 & XSTATE_YMM) && !(xcr0 & XSTATE_SSE))
564 		return 1;
565 	if (xcr0 & ~host_xcr0)
566 		return 1;
567 	vcpu->arch.xcr0 = xcr0;
568 	vcpu->guest_xcr0_loaded = 0;
569 	return 0;
570 }
571 
572 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
573 {
574 	if (__kvm_set_xcr(vcpu, index, xcr)) {
575 		kvm_inject_gp(vcpu, 0);
576 		return 1;
577 	}
578 	return 0;
579 }
580 EXPORT_SYMBOL_GPL(kvm_set_xcr);
581 
582 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
583 {
584 	unsigned long old_cr4 = kvm_read_cr4(vcpu);
585 	unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE |
586 				   X86_CR4_PAE | X86_CR4_SMEP;
587 	if (cr4 & CR4_RESERVED_BITS)
588 		return 1;
589 
590 	if (!guest_cpuid_has_xsave(vcpu) && (cr4 & X86_CR4_OSXSAVE))
591 		return 1;
592 
593 	if (!guest_cpuid_has_smep(vcpu) && (cr4 & X86_CR4_SMEP))
594 		return 1;
595 
596 	if (!guest_cpuid_has_fsgsbase(vcpu) && (cr4 & X86_CR4_RDWRGSFS))
597 		return 1;
598 
599 	if (is_long_mode(vcpu)) {
600 		if (!(cr4 & X86_CR4_PAE))
601 			return 1;
602 	} else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
603 		   && ((cr4 ^ old_cr4) & pdptr_bits)
604 		   && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
605 				   kvm_read_cr3(vcpu)))
606 		return 1;
607 
608 	if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
609 		if (!guest_cpuid_has_pcid(vcpu))
610 			return 1;
611 
612 		/* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
613 		if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
614 			return 1;
615 	}
616 
617 	if (kvm_x86_ops->set_cr4(vcpu, cr4))
618 		return 1;
619 
620 	if (((cr4 ^ old_cr4) & pdptr_bits) ||
621 	    (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
622 		kvm_mmu_reset_context(vcpu);
623 
624 	if ((cr4 ^ old_cr4) & X86_CR4_OSXSAVE)
625 		kvm_update_cpuid(vcpu);
626 
627 	return 0;
628 }
629 EXPORT_SYMBOL_GPL(kvm_set_cr4);
630 
631 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
632 {
633 	if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
634 		kvm_mmu_sync_roots(vcpu);
635 		kvm_mmu_flush_tlb(vcpu);
636 		return 0;
637 	}
638 
639 	if (is_long_mode(vcpu)) {
640 		if (kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE)) {
641 			if (cr3 & CR3_PCID_ENABLED_RESERVED_BITS)
642 				return 1;
643 		} else
644 			if (cr3 & CR3_L_MODE_RESERVED_BITS)
645 				return 1;
646 	} else {
647 		if (is_pae(vcpu)) {
648 			if (cr3 & CR3_PAE_RESERVED_BITS)
649 				return 1;
650 			if (is_paging(vcpu) &&
651 			    !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
652 				return 1;
653 		}
654 		/*
655 		 * We don't check reserved bits in nonpae mode, because
656 		 * this isn't enforced, and VMware depends on this.
657 		 */
658 	}
659 
660 	/*
661 	 * Does the new cr3 value map to physical memory? (Note, we
662 	 * catch an invalid cr3 even in real-mode, because it would
663 	 * cause trouble later on when we turn on paging anyway.)
664 	 *
665 	 * A real CPU would silently accept an invalid cr3 and would
666 	 * attempt to use it - with largely undefined (and often hard
667 	 * to debug) behavior on the guest side.
668 	 */
669 	if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
670 		return 1;
671 	vcpu->arch.cr3 = cr3;
672 	__set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
673 	vcpu->arch.mmu.new_cr3(vcpu);
674 	return 0;
675 }
676 EXPORT_SYMBOL_GPL(kvm_set_cr3);
677 
678 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
679 {
680 	if (cr8 & CR8_RESERVED_BITS)
681 		return 1;
682 	if (irqchip_in_kernel(vcpu->kvm))
683 		kvm_lapic_set_tpr(vcpu, cr8);
684 	else
685 		vcpu->arch.cr8 = cr8;
686 	return 0;
687 }
688 EXPORT_SYMBOL_GPL(kvm_set_cr8);
689 
690 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
691 {
692 	if (irqchip_in_kernel(vcpu->kvm))
693 		return kvm_lapic_get_cr8(vcpu);
694 	else
695 		return vcpu->arch.cr8;
696 }
697 EXPORT_SYMBOL_GPL(kvm_get_cr8);
698 
699 static void kvm_update_dr7(struct kvm_vcpu *vcpu)
700 {
701 	unsigned long dr7;
702 
703 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
704 		dr7 = vcpu->arch.guest_debug_dr7;
705 	else
706 		dr7 = vcpu->arch.dr7;
707 	kvm_x86_ops->set_dr7(vcpu, dr7);
708 	vcpu->arch.switch_db_regs = (dr7 & DR7_BP_EN_MASK);
709 }
710 
711 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
712 {
713 	switch (dr) {
714 	case 0 ... 3:
715 		vcpu->arch.db[dr] = val;
716 		if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
717 			vcpu->arch.eff_db[dr] = val;
718 		break;
719 	case 4:
720 		if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
721 			return 1; /* #UD */
722 		/* fall through */
723 	case 6:
724 		if (val & 0xffffffff00000000ULL)
725 			return -1; /* #GP */
726 		vcpu->arch.dr6 = (val & DR6_VOLATILE) | DR6_FIXED_1;
727 		break;
728 	case 5:
729 		if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
730 			return 1; /* #UD */
731 		/* fall through */
732 	default: /* 7 */
733 		if (val & 0xffffffff00000000ULL)
734 			return -1; /* #GP */
735 		vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
736 		kvm_update_dr7(vcpu);
737 		break;
738 	}
739 
740 	return 0;
741 }
742 
743 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
744 {
745 	int res;
746 
747 	res = __kvm_set_dr(vcpu, dr, val);
748 	if (res > 0)
749 		kvm_queue_exception(vcpu, UD_VECTOR);
750 	else if (res < 0)
751 		kvm_inject_gp(vcpu, 0);
752 
753 	return res;
754 }
755 EXPORT_SYMBOL_GPL(kvm_set_dr);
756 
757 static int _kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
758 {
759 	switch (dr) {
760 	case 0 ... 3:
761 		*val = vcpu->arch.db[dr];
762 		break;
763 	case 4:
764 		if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
765 			return 1;
766 		/* fall through */
767 	case 6:
768 		*val = vcpu->arch.dr6;
769 		break;
770 	case 5:
771 		if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
772 			return 1;
773 		/* fall through */
774 	default: /* 7 */
775 		*val = vcpu->arch.dr7;
776 		break;
777 	}
778 
779 	return 0;
780 }
781 
782 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
783 {
784 	if (_kvm_get_dr(vcpu, dr, val)) {
785 		kvm_queue_exception(vcpu, UD_VECTOR);
786 		return 1;
787 	}
788 	return 0;
789 }
790 EXPORT_SYMBOL_GPL(kvm_get_dr);
791 
792 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
793 {
794 	u32 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
795 	u64 data;
796 	int err;
797 
798 	err = kvm_pmu_read_pmc(vcpu, ecx, &data);
799 	if (err)
800 		return err;
801 	kvm_register_write(vcpu, VCPU_REGS_RAX, (u32)data);
802 	kvm_register_write(vcpu, VCPU_REGS_RDX, data >> 32);
803 	return err;
804 }
805 EXPORT_SYMBOL_GPL(kvm_rdpmc);
806 
807 /*
808  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
809  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
810  *
811  * This list is modified at module load time to reflect the
812  * capabilities of the host cpu. This capabilities test skips MSRs that are
813  * kvm-specific. Those are put in the beginning of the list.
814  */
815 
816 #define KVM_SAVE_MSRS_BEGIN	10
817 static u32 msrs_to_save[] = {
818 	MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
819 	MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
820 	HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
821 	HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
822 	MSR_KVM_PV_EOI_EN,
823 	MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
824 	MSR_STAR,
825 #ifdef CONFIG_X86_64
826 	MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
827 #endif
828 	MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA
829 };
830 
831 static unsigned num_msrs_to_save;
832 
833 static const u32 emulated_msrs[] = {
834 	MSR_IA32_TSC_ADJUST,
835 	MSR_IA32_TSCDEADLINE,
836 	MSR_IA32_MISC_ENABLE,
837 	MSR_IA32_MCG_STATUS,
838 	MSR_IA32_MCG_CTL,
839 };
840 
841 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
842 {
843 	u64 old_efer = vcpu->arch.efer;
844 
845 	if (efer & efer_reserved_bits)
846 		return 1;
847 
848 	if (is_paging(vcpu)
849 	    && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
850 		return 1;
851 
852 	if (efer & EFER_FFXSR) {
853 		struct kvm_cpuid_entry2 *feat;
854 
855 		feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
856 		if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
857 			return 1;
858 	}
859 
860 	if (efer & EFER_SVME) {
861 		struct kvm_cpuid_entry2 *feat;
862 
863 		feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
864 		if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
865 			return 1;
866 	}
867 
868 	efer &= ~EFER_LMA;
869 	efer |= vcpu->arch.efer & EFER_LMA;
870 
871 	kvm_x86_ops->set_efer(vcpu, efer);
872 
873 	vcpu->arch.mmu.base_role.nxe = (efer & EFER_NX) && !tdp_enabled;
874 
875 	/* Update reserved bits */
876 	if ((efer ^ old_efer) & EFER_NX)
877 		kvm_mmu_reset_context(vcpu);
878 
879 	return 0;
880 }
881 
882 void kvm_enable_efer_bits(u64 mask)
883 {
884        efer_reserved_bits &= ~mask;
885 }
886 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
887 
888 
889 /*
890  * Writes msr value into into the appropriate "register".
891  * Returns 0 on success, non-0 otherwise.
892  * Assumes vcpu_load() was already called.
893  */
894 int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
895 {
896 	return kvm_x86_ops->set_msr(vcpu, msr);
897 }
898 
899 /*
900  * Adapt set_msr() to msr_io()'s calling convention
901  */
902 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
903 {
904 	struct msr_data msr;
905 
906 	msr.data = *data;
907 	msr.index = index;
908 	msr.host_initiated = true;
909 	return kvm_set_msr(vcpu, &msr);
910 }
911 
912 #ifdef CONFIG_X86_64
913 struct pvclock_gtod_data {
914 	seqcount_t	seq;
915 
916 	struct { /* extract of a clocksource struct */
917 		int vclock_mode;
918 		cycle_t	cycle_last;
919 		cycle_t	mask;
920 		u32	mult;
921 		u32	shift;
922 	} clock;
923 
924 	/* open coded 'struct timespec' */
925 	u64		monotonic_time_snsec;
926 	time_t		monotonic_time_sec;
927 };
928 
929 static struct pvclock_gtod_data pvclock_gtod_data;
930 
931 static void update_pvclock_gtod(struct timekeeper *tk)
932 {
933 	struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
934 
935 	write_seqcount_begin(&vdata->seq);
936 
937 	/* copy pvclock gtod data */
938 	vdata->clock.vclock_mode	= tk->clock->archdata.vclock_mode;
939 	vdata->clock.cycle_last		= tk->clock->cycle_last;
940 	vdata->clock.mask		= tk->clock->mask;
941 	vdata->clock.mult		= tk->mult;
942 	vdata->clock.shift		= tk->shift;
943 
944 	vdata->monotonic_time_sec	= tk->xtime_sec
945 					+ tk->wall_to_monotonic.tv_sec;
946 	vdata->monotonic_time_snsec	= tk->xtime_nsec
947 					+ (tk->wall_to_monotonic.tv_nsec
948 						<< tk->shift);
949 	while (vdata->monotonic_time_snsec >=
950 					(((u64)NSEC_PER_SEC) << tk->shift)) {
951 		vdata->monotonic_time_snsec -=
952 					((u64)NSEC_PER_SEC) << tk->shift;
953 		vdata->monotonic_time_sec++;
954 	}
955 
956 	write_seqcount_end(&vdata->seq);
957 }
958 #endif
959 
960 
961 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
962 {
963 	int version;
964 	int r;
965 	struct pvclock_wall_clock wc;
966 	struct timespec boot;
967 
968 	if (!wall_clock)
969 		return;
970 
971 	r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
972 	if (r)
973 		return;
974 
975 	if (version & 1)
976 		++version;  /* first time write, random junk */
977 
978 	++version;
979 
980 	kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
981 
982 	/*
983 	 * The guest calculates current wall clock time by adding
984 	 * system time (updated by kvm_guest_time_update below) to the
985 	 * wall clock specified here.  guest system time equals host
986 	 * system time for us, thus we must fill in host boot time here.
987 	 */
988 	getboottime(&boot);
989 
990 	if (kvm->arch.kvmclock_offset) {
991 		struct timespec ts = ns_to_timespec(kvm->arch.kvmclock_offset);
992 		boot = timespec_sub(boot, ts);
993 	}
994 	wc.sec = boot.tv_sec;
995 	wc.nsec = boot.tv_nsec;
996 	wc.version = version;
997 
998 	kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
999 
1000 	version++;
1001 	kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1002 }
1003 
1004 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1005 {
1006 	uint32_t quotient, remainder;
1007 
1008 	/* Don't try to replace with do_div(), this one calculates
1009 	 * "(dividend << 32) / divisor" */
1010 	__asm__ ( "divl %4"
1011 		  : "=a" (quotient), "=d" (remainder)
1012 		  : "0" (0), "1" (dividend), "r" (divisor) );
1013 	return quotient;
1014 }
1015 
1016 static void kvm_get_time_scale(uint32_t scaled_khz, uint32_t base_khz,
1017 			       s8 *pshift, u32 *pmultiplier)
1018 {
1019 	uint64_t scaled64;
1020 	int32_t  shift = 0;
1021 	uint64_t tps64;
1022 	uint32_t tps32;
1023 
1024 	tps64 = base_khz * 1000LL;
1025 	scaled64 = scaled_khz * 1000LL;
1026 	while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1027 		tps64 >>= 1;
1028 		shift--;
1029 	}
1030 
1031 	tps32 = (uint32_t)tps64;
1032 	while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1033 		if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1034 			scaled64 >>= 1;
1035 		else
1036 			tps32 <<= 1;
1037 		shift++;
1038 	}
1039 
1040 	*pshift = shift;
1041 	*pmultiplier = div_frac(scaled64, tps32);
1042 
1043 	pr_debug("%s: base_khz %u => %u, shift %d, mul %u\n",
1044 		 __func__, base_khz, scaled_khz, shift, *pmultiplier);
1045 }
1046 
1047 static inline u64 get_kernel_ns(void)
1048 {
1049 	struct timespec ts;
1050 
1051 	WARN_ON(preemptible());
1052 	ktime_get_ts(&ts);
1053 	monotonic_to_bootbased(&ts);
1054 	return timespec_to_ns(&ts);
1055 }
1056 
1057 #ifdef CONFIG_X86_64
1058 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1059 #endif
1060 
1061 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1062 unsigned long max_tsc_khz;
1063 
1064 static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec)
1065 {
1066 	return pvclock_scale_delta(nsec, vcpu->arch.virtual_tsc_mult,
1067 				   vcpu->arch.virtual_tsc_shift);
1068 }
1069 
1070 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1071 {
1072 	u64 v = (u64)khz * (1000000 + ppm);
1073 	do_div(v, 1000000);
1074 	return v;
1075 }
1076 
1077 static void kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 this_tsc_khz)
1078 {
1079 	u32 thresh_lo, thresh_hi;
1080 	int use_scaling = 0;
1081 
1082 	/* Compute a scale to convert nanoseconds in TSC cycles */
1083 	kvm_get_time_scale(this_tsc_khz, NSEC_PER_SEC / 1000,
1084 			   &vcpu->arch.virtual_tsc_shift,
1085 			   &vcpu->arch.virtual_tsc_mult);
1086 	vcpu->arch.virtual_tsc_khz = this_tsc_khz;
1087 
1088 	/*
1089 	 * Compute the variation in TSC rate which is acceptable
1090 	 * within the range of tolerance and decide if the
1091 	 * rate being applied is within that bounds of the hardware
1092 	 * rate.  If so, no scaling or compensation need be done.
1093 	 */
1094 	thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1095 	thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1096 	if (this_tsc_khz < thresh_lo || this_tsc_khz > thresh_hi) {
1097 		pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", this_tsc_khz, thresh_lo, thresh_hi);
1098 		use_scaling = 1;
1099 	}
1100 	kvm_x86_ops->set_tsc_khz(vcpu, this_tsc_khz, use_scaling);
1101 }
1102 
1103 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1104 {
1105 	u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1106 				      vcpu->arch.virtual_tsc_mult,
1107 				      vcpu->arch.virtual_tsc_shift);
1108 	tsc += vcpu->arch.this_tsc_write;
1109 	return tsc;
1110 }
1111 
1112 void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1113 {
1114 #ifdef CONFIG_X86_64
1115 	bool vcpus_matched;
1116 	bool do_request = false;
1117 	struct kvm_arch *ka = &vcpu->kvm->arch;
1118 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1119 
1120 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1121 			 atomic_read(&vcpu->kvm->online_vcpus));
1122 
1123 	if (vcpus_matched && gtod->clock.vclock_mode == VCLOCK_TSC)
1124 		if (!ka->use_master_clock)
1125 			do_request = 1;
1126 
1127 	if (!vcpus_matched && ka->use_master_clock)
1128 			do_request = 1;
1129 
1130 	if (do_request)
1131 		kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1132 
1133 	trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1134 			    atomic_read(&vcpu->kvm->online_vcpus),
1135 		            ka->use_master_clock, gtod->clock.vclock_mode);
1136 #endif
1137 }
1138 
1139 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1140 {
1141 	u64 curr_offset = kvm_x86_ops->read_tsc_offset(vcpu);
1142 	vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1143 }
1144 
1145 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1146 {
1147 	struct kvm *kvm = vcpu->kvm;
1148 	u64 offset, ns, elapsed;
1149 	unsigned long flags;
1150 	s64 usdiff;
1151 	bool matched;
1152 	u64 data = msr->data;
1153 
1154 	raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1155 	offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
1156 	ns = get_kernel_ns();
1157 	elapsed = ns - kvm->arch.last_tsc_nsec;
1158 
1159 	/* n.b - signed multiplication and division required */
1160 	usdiff = data - kvm->arch.last_tsc_write;
1161 #ifdef CONFIG_X86_64
1162 	usdiff = (usdiff * 1000) / vcpu->arch.virtual_tsc_khz;
1163 #else
1164 	/* do_div() only does unsigned */
1165 	asm("idivl %2; xor %%edx, %%edx"
1166 	    : "=A"(usdiff)
1167 	    : "A"(usdiff * 1000), "rm"(vcpu->arch.virtual_tsc_khz));
1168 #endif
1169 	do_div(elapsed, 1000);
1170 	usdiff -= elapsed;
1171 	if (usdiff < 0)
1172 		usdiff = -usdiff;
1173 
1174 	/*
1175 	 * Special case: TSC write with a small delta (1 second) of virtual
1176 	 * cycle time against real time is interpreted as an attempt to
1177 	 * synchronize the CPU.
1178          *
1179 	 * For a reliable TSC, we can match TSC offsets, and for an unstable
1180 	 * TSC, we add elapsed time in this computation.  We could let the
1181 	 * compensation code attempt to catch up if we fall behind, but
1182 	 * it's better to try to match offsets from the beginning.
1183          */
1184 	if (usdiff < USEC_PER_SEC &&
1185 	    vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1186 		if (!check_tsc_unstable()) {
1187 			offset = kvm->arch.cur_tsc_offset;
1188 			pr_debug("kvm: matched tsc offset for %llu\n", data);
1189 		} else {
1190 			u64 delta = nsec_to_cycles(vcpu, elapsed);
1191 			data += delta;
1192 			offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
1193 			pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1194 		}
1195 		matched = true;
1196 	} else {
1197 		/*
1198 		 * We split periods of matched TSC writes into generations.
1199 		 * For each generation, we track the original measured
1200 		 * nanosecond time, offset, and write, so if TSCs are in
1201 		 * sync, we can match exact offset, and if not, we can match
1202 		 * exact software computation in compute_guest_tsc()
1203 		 *
1204 		 * These values are tracked in kvm->arch.cur_xxx variables.
1205 		 */
1206 		kvm->arch.cur_tsc_generation++;
1207 		kvm->arch.cur_tsc_nsec = ns;
1208 		kvm->arch.cur_tsc_write = data;
1209 		kvm->arch.cur_tsc_offset = offset;
1210 		matched = false;
1211 		pr_debug("kvm: new tsc generation %u, clock %llu\n",
1212 			 kvm->arch.cur_tsc_generation, data);
1213 	}
1214 
1215 	/*
1216 	 * We also track th most recent recorded KHZ, write and time to
1217 	 * allow the matching interval to be extended at each write.
1218 	 */
1219 	kvm->arch.last_tsc_nsec = ns;
1220 	kvm->arch.last_tsc_write = data;
1221 	kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1222 
1223 	/* Reset of TSC must disable overshoot protection below */
1224 	vcpu->arch.hv_clock.tsc_timestamp = 0;
1225 	vcpu->arch.last_guest_tsc = data;
1226 
1227 	/* Keep track of which generation this VCPU has synchronized to */
1228 	vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1229 	vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1230 	vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1231 
1232 	if (guest_cpuid_has_tsc_adjust(vcpu) && !msr->host_initiated)
1233 		update_ia32_tsc_adjust_msr(vcpu, offset);
1234 	kvm_x86_ops->write_tsc_offset(vcpu, offset);
1235 	raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1236 
1237 	spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1238 	if (matched)
1239 		kvm->arch.nr_vcpus_matched_tsc++;
1240 	else
1241 		kvm->arch.nr_vcpus_matched_tsc = 0;
1242 
1243 	kvm_track_tsc_matching(vcpu);
1244 	spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
1245 }
1246 
1247 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1248 
1249 #ifdef CONFIG_X86_64
1250 
1251 static cycle_t read_tsc(void)
1252 {
1253 	cycle_t ret;
1254 	u64 last;
1255 
1256 	/*
1257 	 * Empirically, a fence (of type that depends on the CPU)
1258 	 * before rdtsc is enough to ensure that rdtsc is ordered
1259 	 * with respect to loads.  The various CPU manuals are unclear
1260 	 * as to whether rdtsc can be reordered with later loads,
1261 	 * but no one has ever seen it happen.
1262 	 */
1263 	rdtsc_barrier();
1264 	ret = (cycle_t)vget_cycles();
1265 
1266 	last = pvclock_gtod_data.clock.cycle_last;
1267 
1268 	if (likely(ret >= last))
1269 		return ret;
1270 
1271 	/*
1272 	 * GCC likes to generate cmov here, but this branch is extremely
1273 	 * predictable (it's just a funciton of time and the likely is
1274 	 * very likely) and there's a data dependence, so force GCC
1275 	 * to generate a branch instead.  I don't barrier() because
1276 	 * we don't actually need a barrier, and if this function
1277 	 * ever gets inlined it will generate worse code.
1278 	 */
1279 	asm volatile ("");
1280 	return last;
1281 }
1282 
1283 static inline u64 vgettsc(cycle_t *cycle_now)
1284 {
1285 	long v;
1286 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1287 
1288 	*cycle_now = read_tsc();
1289 
1290 	v = (*cycle_now - gtod->clock.cycle_last) & gtod->clock.mask;
1291 	return v * gtod->clock.mult;
1292 }
1293 
1294 static int do_monotonic(struct timespec *ts, cycle_t *cycle_now)
1295 {
1296 	unsigned long seq;
1297 	u64 ns;
1298 	int mode;
1299 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1300 
1301 	ts->tv_nsec = 0;
1302 	do {
1303 		seq = read_seqcount_begin(&gtod->seq);
1304 		mode = gtod->clock.vclock_mode;
1305 		ts->tv_sec = gtod->monotonic_time_sec;
1306 		ns = gtod->monotonic_time_snsec;
1307 		ns += vgettsc(cycle_now);
1308 		ns >>= gtod->clock.shift;
1309 	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
1310 	timespec_add_ns(ts, ns);
1311 
1312 	return mode;
1313 }
1314 
1315 /* returns true if host is using tsc clocksource */
1316 static bool kvm_get_time_and_clockread(s64 *kernel_ns, cycle_t *cycle_now)
1317 {
1318 	struct timespec ts;
1319 
1320 	/* checked again under seqlock below */
1321 	if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC)
1322 		return false;
1323 
1324 	if (do_monotonic(&ts, cycle_now) != VCLOCK_TSC)
1325 		return false;
1326 
1327 	monotonic_to_bootbased(&ts);
1328 	*kernel_ns = timespec_to_ns(&ts);
1329 
1330 	return true;
1331 }
1332 #endif
1333 
1334 /*
1335  *
1336  * Assuming a stable TSC across physical CPUS, and a stable TSC
1337  * across virtual CPUs, the following condition is possible.
1338  * Each numbered line represents an event visible to both
1339  * CPUs at the next numbered event.
1340  *
1341  * "timespecX" represents host monotonic time. "tscX" represents
1342  * RDTSC value.
1343  *
1344  * 		VCPU0 on CPU0		|	VCPU1 on CPU1
1345  *
1346  * 1.  read timespec0,tsc0
1347  * 2.					| timespec1 = timespec0 + N
1348  * 					| tsc1 = tsc0 + M
1349  * 3. transition to guest		| transition to guest
1350  * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1351  * 5.				        | ret1 = timespec1 + (rdtsc - tsc1)
1352  * 				        | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1353  *
1354  * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1355  *
1356  * 	- ret0 < ret1
1357  *	- timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1358  *		...
1359  *	- 0 < N - M => M < N
1360  *
1361  * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1362  * always the case (the difference between two distinct xtime instances
1363  * might be smaller then the difference between corresponding TSC reads,
1364  * when updating guest vcpus pvclock areas).
1365  *
1366  * To avoid that problem, do not allow visibility of distinct
1367  * system_timestamp/tsc_timestamp values simultaneously: use a master
1368  * copy of host monotonic time values. Update that master copy
1369  * in lockstep.
1370  *
1371  * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
1372  *
1373  */
1374 
1375 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
1376 {
1377 #ifdef CONFIG_X86_64
1378 	struct kvm_arch *ka = &kvm->arch;
1379 	int vclock_mode;
1380 	bool host_tsc_clocksource, vcpus_matched;
1381 
1382 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1383 			atomic_read(&kvm->online_vcpus));
1384 
1385 	/*
1386 	 * If the host uses TSC clock, then passthrough TSC as stable
1387 	 * to the guest.
1388 	 */
1389 	host_tsc_clocksource = kvm_get_time_and_clockread(
1390 					&ka->master_kernel_ns,
1391 					&ka->master_cycle_now);
1392 
1393 	ka->use_master_clock = host_tsc_clocksource & vcpus_matched;
1394 
1395 	if (ka->use_master_clock)
1396 		atomic_set(&kvm_guest_has_master_clock, 1);
1397 
1398 	vclock_mode = pvclock_gtod_data.clock.vclock_mode;
1399 	trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
1400 					vcpus_matched);
1401 #endif
1402 }
1403 
1404 static int kvm_guest_time_update(struct kvm_vcpu *v)
1405 {
1406 	unsigned long flags, this_tsc_khz;
1407 	struct kvm_vcpu_arch *vcpu = &v->arch;
1408 	struct kvm_arch *ka = &v->kvm->arch;
1409 	void *shared_kaddr;
1410 	s64 kernel_ns, max_kernel_ns;
1411 	u64 tsc_timestamp, host_tsc;
1412 	struct pvclock_vcpu_time_info *guest_hv_clock;
1413 	u8 pvclock_flags;
1414 	bool use_master_clock;
1415 
1416 	kernel_ns = 0;
1417 	host_tsc = 0;
1418 
1419 	/* Keep irq disabled to prevent changes to the clock */
1420 	local_irq_save(flags);
1421 	this_tsc_khz = __get_cpu_var(cpu_tsc_khz);
1422 	if (unlikely(this_tsc_khz == 0)) {
1423 		local_irq_restore(flags);
1424 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1425 		return 1;
1426 	}
1427 
1428 	/*
1429 	 * If the host uses TSC clock, then passthrough TSC as stable
1430 	 * to the guest.
1431 	 */
1432 	spin_lock(&ka->pvclock_gtod_sync_lock);
1433 	use_master_clock = ka->use_master_clock;
1434 	if (use_master_clock) {
1435 		host_tsc = ka->master_cycle_now;
1436 		kernel_ns = ka->master_kernel_ns;
1437 	}
1438 	spin_unlock(&ka->pvclock_gtod_sync_lock);
1439 	if (!use_master_clock) {
1440 		host_tsc = native_read_tsc();
1441 		kernel_ns = get_kernel_ns();
1442 	}
1443 
1444 	tsc_timestamp = kvm_x86_ops->read_l1_tsc(v, host_tsc);
1445 
1446 	/*
1447 	 * We may have to catch up the TSC to match elapsed wall clock
1448 	 * time for two reasons, even if kvmclock is used.
1449 	 *   1) CPU could have been running below the maximum TSC rate
1450 	 *   2) Broken TSC compensation resets the base at each VCPU
1451 	 *      entry to avoid unknown leaps of TSC even when running
1452 	 *      again on the same CPU.  This may cause apparent elapsed
1453 	 *      time to disappear, and the guest to stand still or run
1454 	 *	very slowly.
1455 	 */
1456 	if (vcpu->tsc_catchup) {
1457 		u64 tsc = compute_guest_tsc(v, kernel_ns);
1458 		if (tsc > tsc_timestamp) {
1459 			adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
1460 			tsc_timestamp = tsc;
1461 		}
1462 	}
1463 
1464 	local_irq_restore(flags);
1465 
1466 	if (!vcpu->time_page)
1467 		return 0;
1468 
1469 	/*
1470 	 * Time as measured by the TSC may go backwards when resetting the base
1471 	 * tsc_timestamp.  The reason for this is that the TSC resolution is
1472 	 * higher than the resolution of the other clock scales.  Thus, many
1473 	 * possible measurments of the TSC correspond to one measurement of any
1474 	 * other clock, and so a spread of values is possible.  This is not a
1475 	 * problem for the computation of the nanosecond clock; with TSC rates
1476 	 * around 1GHZ, there can only be a few cycles which correspond to one
1477 	 * nanosecond value, and any path through this code will inevitably
1478 	 * take longer than that.  However, with the kernel_ns value itself,
1479 	 * the precision may be much lower, down to HZ granularity.  If the
1480 	 * first sampling of TSC against kernel_ns ends in the low part of the
1481 	 * range, and the second in the high end of the range, we can get:
1482 	 *
1483 	 * (TSC - offset_low) * S + kns_old > (TSC - offset_high) * S + kns_new
1484 	 *
1485 	 * As the sampling errors potentially range in the thousands of cycles,
1486 	 * it is possible such a time value has already been observed by the
1487 	 * guest.  To protect against this, we must compute the system time as
1488 	 * observed by the guest and ensure the new system time is greater.
1489 	 */
1490 	max_kernel_ns = 0;
1491 	if (vcpu->hv_clock.tsc_timestamp) {
1492 		max_kernel_ns = vcpu->last_guest_tsc -
1493 				vcpu->hv_clock.tsc_timestamp;
1494 		max_kernel_ns = pvclock_scale_delta(max_kernel_ns,
1495 				    vcpu->hv_clock.tsc_to_system_mul,
1496 				    vcpu->hv_clock.tsc_shift);
1497 		max_kernel_ns += vcpu->last_kernel_ns;
1498 	}
1499 
1500 	if (unlikely(vcpu->hw_tsc_khz != this_tsc_khz)) {
1501 		kvm_get_time_scale(NSEC_PER_SEC / 1000, this_tsc_khz,
1502 				   &vcpu->hv_clock.tsc_shift,
1503 				   &vcpu->hv_clock.tsc_to_system_mul);
1504 		vcpu->hw_tsc_khz = this_tsc_khz;
1505 	}
1506 
1507 	/* with a master <monotonic time, tsc value> tuple,
1508 	 * pvclock clock reads always increase at the (scaled) rate
1509 	 * of guest TSC - no need to deal with sampling errors.
1510 	 */
1511 	if (!use_master_clock) {
1512 		if (max_kernel_ns > kernel_ns)
1513 			kernel_ns = max_kernel_ns;
1514 	}
1515 	/* With all the info we got, fill in the values */
1516 	vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
1517 	vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1518 	vcpu->last_kernel_ns = kernel_ns;
1519 	vcpu->last_guest_tsc = tsc_timestamp;
1520 
1521 	/*
1522 	 * The interface expects us to write an even number signaling that the
1523 	 * update is finished. Since the guest won't see the intermediate
1524 	 * state, we just increase by 2 at the end.
1525 	 */
1526 	vcpu->hv_clock.version += 2;
1527 
1528 	shared_kaddr = kmap_atomic(vcpu->time_page);
1529 
1530 	guest_hv_clock = shared_kaddr + vcpu->time_offset;
1531 
1532 	/* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
1533 	pvclock_flags = (guest_hv_clock->flags & PVCLOCK_GUEST_STOPPED);
1534 
1535 	if (vcpu->pvclock_set_guest_stopped_request) {
1536 		pvclock_flags |= PVCLOCK_GUEST_STOPPED;
1537 		vcpu->pvclock_set_guest_stopped_request = false;
1538 	}
1539 
1540 	/* If the host uses TSC clocksource, then it is stable */
1541 	if (use_master_clock)
1542 		pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
1543 
1544 	vcpu->hv_clock.flags = pvclock_flags;
1545 
1546 	memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
1547 	       sizeof(vcpu->hv_clock));
1548 
1549 	kunmap_atomic(shared_kaddr);
1550 
1551 	mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
1552 	return 0;
1553 }
1554 
1555 static bool msr_mtrr_valid(unsigned msr)
1556 {
1557 	switch (msr) {
1558 	case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
1559 	case MSR_MTRRfix64K_00000:
1560 	case MSR_MTRRfix16K_80000:
1561 	case MSR_MTRRfix16K_A0000:
1562 	case MSR_MTRRfix4K_C0000:
1563 	case MSR_MTRRfix4K_C8000:
1564 	case MSR_MTRRfix4K_D0000:
1565 	case MSR_MTRRfix4K_D8000:
1566 	case MSR_MTRRfix4K_E0000:
1567 	case MSR_MTRRfix4K_E8000:
1568 	case MSR_MTRRfix4K_F0000:
1569 	case MSR_MTRRfix4K_F8000:
1570 	case MSR_MTRRdefType:
1571 	case MSR_IA32_CR_PAT:
1572 		return true;
1573 	case 0x2f8:
1574 		return true;
1575 	}
1576 	return false;
1577 }
1578 
1579 static bool valid_pat_type(unsigned t)
1580 {
1581 	return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
1582 }
1583 
1584 static bool valid_mtrr_type(unsigned t)
1585 {
1586 	return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
1587 }
1588 
1589 static bool mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1590 {
1591 	int i;
1592 
1593 	if (!msr_mtrr_valid(msr))
1594 		return false;
1595 
1596 	if (msr == MSR_IA32_CR_PAT) {
1597 		for (i = 0; i < 8; i++)
1598 			if (!valid_pat_type((data >> (i * 8)) & 0xff))
1599 				return false;
1600 		return true;
1601 	} else if (msr == MSR_MTRRdefType) {
1602 		if (data & ~0xcff)
1603 			return false;
1604 		return valid_mtrr_type(data & 0xff);
1605 	} else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
1606 		for (i = 0; i < 8 ; i++)
1607 			if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
1608 				return false;
1609 		return true;
1610 	}
1611 
1612 	/* variable MTRRs */
1613 	return valid_mtrr_type(data & 0xff);
1614 }
1615 
1616 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1617 {
1618 	u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1619 
1620 	if (!mtrr_valid(vcpu, msr, data))
1621 		return 1;
1622 
1623 	if (msr == MSR_MTRRdefType) {
1624 		vcpu->arch.mtrr_state.def_type = data;
1625 		vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
1626 	} else if (msr == MSR_MTRRfix64K_00000)
1627 		p[0] = data;
1628 	else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1629 		p[1 + msr - MSR_MTRRfix16K_80000] = data;
1630 	else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1631 		p[3 + msr - MSR_MTRRfix4K_C0000] = data;
1632 	else if (msr == MSR_IA32_CR_PAT)
1633 		vcpu->arch.pat = data;
1634 	else {	/* Variable MTRRs */
1635 		int idx, is_mtrr_mask;
1636 		u64 *pt;
1637 
1638 		idx = (msr - 0x200) / 2;
1639 		is_mtrr_mask = msr - 0x200 - 2 * idx;
1640 		if (!is_mtrr_mask)
1641 			pt =
1642 			  (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1643 		else
1644 			pt =
1645 			  (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1646 		*pt = data;
1647 	}
1648 
1649 	kvm_mmu_reset_context(vcpu);
1650 	return 0;
1651 }
1652 
1653 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1654 {
1655 	u64 mcg_cap = vcpu->arch.mcg_cap;
1656 	unsigned bank_num = mcg_cap & 0xff;
1657 
1658 	switch (msr) {
1659 	case MSR_IA32_MCG_STATUS:
1660 		vcpu->arch.mcg_status = data;
1661 		break;
1662 	case MSR_IA32_MCG_CTL:
1663 		if (!(mcg_cap & MCG_CTL_P))
1664 			return 1;
1665 		if (data != 0 && data != ~(u64)0)
1666 			return -1;
1667 		vcpu->arch.mcg_ctl = data;
1668 		break;
1669 	default:
1670 		if (msr >= MSR_IA32_MC0_CTL &&
1671 		    msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1672 			u32 offset = msr - MSR_IA32_MC0_CTL;
1673 			/* only 0 or all 1s can be written to IA32_MCi_CTL
1674 			 * some Linux kernels though clear bit 10 in bank 4 to
1675 			 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1676 			 * this to avoid an uncatched #GP in the guest
1677 			 */
1678 			if ((offset & 0x3) == 0 &&
1679 			    data != 0 && (data | (1 << 10)) != ~(u64)0)
1680 				return -1;
1681 			vcpu->arch.mce_banks[offset] = data;
1682 			break;
1683 		}
1684 		return 1;
1685 	}
1686 	return 0;
1687 }
1688 
1689 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1690 {
1691 	struct kvm *kvm = vcpu->kvm;
1692 	int lm = is_long_mode(vcpu);
1693 	u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1694 		: (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1695 	u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1696 		: kvm->arch.xen_hvm_config.blob_size_32;
1697 	u32 page_num = data & ~PAGE_MASK;
1698 	u64 page_addr = data & PAGE_MASK;
1699 	u8 *page;
1700 	int r;
1701 
1702 	r = -E2BIG;
1703 	if (page_num >= blob_size)
1704 		goto out;
1705 	r = -ENOMEM;
1706 	page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
1707 	if (IS_ERR(page)) {
1708 		r = PTR_ERR(page);
1709 		goto out;
1710 	}
1711 	if (kvm_write_guest(kvm, page_addr, page, PAGE_SIZE))
1712 		goto out_free;
1713 	r = 0;
1714 out_free:
1715 	kfree(page);
1716 out:
1717 	return r;
1718 }
1719 
1720 static bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1721 {
1722 	return kvm->arch.hv_hypercall & HV_X64_MSR_HYPERCALL_ENABLE;
1723 }
1724 
1725 static bool kvm_hv_msr_partition_wide(u32 msr)
1726 {
1727 	bool r = false;
1728 	switch (msr) {
1729 	case HV_X64_MSR_GUEST_OS_ID:
1730 	case HV_X64_MSR_HYPERCALL:
1731 		r = true;
1732 		break;
1733 	}
1734 
1735 	return r;
1736 }
1737 
1738 static int set_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1739 {
1740 	struct kvm *kvm = vcpu->kvm;
1741 
1742 	switch (msr) {
1743 	case HV_X64_MSR_GUEST_OS_ID:
1744 		kvm->arch.hv_guest_os_id = data;
1745 		/* setting guest os id to zero disables hypercall page */
1746 		if (!kvm->arch.hv_guest_os_id)
1747 			kvm->arch.hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1748 		break;
1749 	case HV_X64_MSR_HYPERCALL: {
1750 		u64 gfn;
1751 		unsigned long addr;
1752 		u8 instructions[4];
1753 
1754 		/* if guest os id is not set hypercall should remain disabled */
1755 		if (!kvm->arch.hv_guest_os_id)
1756 			break;
1757 		if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1758 			kvm->arch.hv_hypercall = data;
1759 			break;
1760 		}
1761 		gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
1762 		addr = gfn_to_hva(kvm, gfn);
1763 		if (kvm_is_error_hva(addr))
1764 			return 1;
1765 		kvm_x86_ops->patch_hypercall(vcpu, instructions);
1766 		((unsigned char *)instructions)[3] = 0xc3; /* ret */
1767 		if (__copy_to_user((void __user *)addr, instructions, 4))
1768 			return 1;
1769 		kvm->arch.hv_hypercall = data;
1770 		break;
1771 	}
1772 	default:
1773 		vcpu_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1774 			    "data 0x%llx\n", msr, data);
1775 		return 1;
1776 	}
1777 	return 0;
1778 }
1779 
1780 static int set_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1781 {
1782 	switch (msr) {
1783 	case HV_X64_MSR_APIC_ASSIST_PAGE: {
1784 		unsigned long addr;
1785 
1786 		if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) {
1787 			vcpu->arch.hv_vapic = data;
1788 			break;
1789 		}
1790 		addr = gfn_to_hva(vcpu->kvm, data >>
1791 				  HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT);
1792 		if (kvm_is_error_hva(addr))
1793 			return 1;
1794 		if (__clear_user((void __user *)addr, PAGE_SIZE))
1795 			return 1;
1796 		vcpu->arch.hv_vapic = data;
1797 		break;
1798 	}
1799 	case HV_X64_MSR_EOI:
1800 		return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1801 	case HV_X64_MSR_ICR:
1802 		return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1803 	case HV_X64_MSR_TPR:
1804 		return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1805 	default:
1806 		vcpu_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1807 			    "data 0x%llx\n", msr, data);
1808 		return 1;
1809 	}
1810 
1811 	return 0;
1812 }
1813 
1814 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
1815 {
1816 	gpa_t gpa = data & ~0x3f;
1817 
1818 	/* Bits 2:5 are reserved, Should be zero */
1819 	if (data & 0x3c)
1820 		return 1;
1821 
1822 	vcpu->arch.apf.msr_val = data;
1823 
1824 	if (!(data & KVM_ASYNC_PF_ENABLED)) {
1825 		kvm_clear_async_pf_completion_queue(vcpu);
1826 		kvm_async_pf_hash_reset(vcpu);
1827 		return 0;
1828 	}
1829 
1830 	if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa))
1831 		return 1;
1832 
1833 	vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
1834 	kvm_async_pf_wakeup_all(vcpu);
1835 	return 0;
1836 }
1837 
1838 static void kvmclock_reset(struct kvm_vcpu *vcpu)
1839 {
1840 	if (vcpu->arch.time_page) {
1841 		kvm_release_page_dirty(vcpu->arch.time_page);
1842 		vcpu->arch.time_page = NULL;
1843 	}
1844 }
1845 
1846 static void accumulate_steal_time(struct kvm_vcpu *vcpu)
1847 {
1848 	u64 delta;
1849 
1850 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
1851 		return;
1852 
1853 	delta = current->sched_info.run_delay - vcpu->arch.st.last_steal;
1854 	vcpu->arch.st.last_steal = current->sched_info.run_delay;
1855 	vcpu->arch.st.accum_steal = delta;
1856 }
1857 
1858 static void record_steal_time(struct kvm_vcpu *vcpu)
1859 {
1860 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
1861 		return;
1862 
1863 	if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
1864 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
1865 		return;
1866 
1867 	vcpu->arch.st.steal.steal += vcpu->arch.st.accum_steal;
1868 	vcpu->arch.st.steal.version += 2;
1869 	vcpu->arch.st.accum_steal = 0;
1870 
1871 	kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
1872 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
1873 }
1874 
1875 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1876 {
1877 	bool pr = false;
1878 	u32 msr = msr_info->index;
1879 	u64 data = msr_info->data;
1880 
1881 	switch (msr) {
1882 	case MSR_EFER:
1883 		return set_efer(vcpu, data);
1884 	case MSR_K7_HWCR:
1885 		data &= ~(u64)0x40;	/* ignore flush filter disable */
1886 		data &= ~(u64)0x100;	/* ignore ignne emulation enable */
1887 		data &= ~(u64)0x8;	/* ignore TLB cache disable */
1888 		if (data != 0) {
1889 			vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
1890 				    data);
1891 			return 1;
1892 		}
1893 		break;
1894 	case MSR_FAM10H_MMIO_CONF_BASE:
1895 		if (data != 0) {
1896 			vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
1897 				    "0x%llx\n", data);
1898 			return 1;
1899 		}
1900 		break;
1901 	case MSR_AMD64_NB_CFG:
1902 		break;
1903 	case MSR_IA32_DEBUGCTLMSR:
1904 		if (!data) {
1905 			/* We support the non-activated case already */
1906 			break;
1907 		} else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
1908 			/* Values other than LBR and BTF are vendor-specific,
1909 			   thus reserved and should throw a #GP */
1910 			return 1;
1911 		}
1912 		vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
1913 			    __func__, data);
1914 		break;
1915 	case MSR_IA32_UCODE_REV:
1916 	case MSR_IA32_UCODE_WRITE:
1917 	case MSR_VM_HSAVE_PA:
1918 	case MSR_AMD64_PATCH_LOADER:
1919 		break;
1920 	case 0x200 ... 0x2ff:
1921 		return set_msr_mtrr(vcpu, msr, data);
1922 	case MSR_IA32_APICBASE:
1923 		kvm_set_apic_base(vcpu, data);
1924 		break;
1925 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1926 		return kvm_x2apic_msr_write(vcpu, msr, data);
1927 	case MSR_IA32_TSCDEADLINE:
1928 		kvm_set_lapic_tscdeadline_msr(vcpu, data);
1929 		break;
1930 	case MSR_IA32_TSC_ADJUST:
1931 		if (guest_cpuid_has_tsc_adjust(vcpu)) {
1932 			if (!msr_info->host_initiated) {
1933 				u64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
1934 				kvm_x86_ops->adjust_tsc_offset(vcpu, adj, true);
1935 			}
1936 			vcpu->arch.ia32_tsc_adjust_msr = data;
1937 		}
1938 		break;
1939 	case MSR_IA32_MISC_ENABLE:
1940 		vcpu->arch.ia32_misc_enable_msr = data;
1941 		break;
1942 	case MSR_KVM_WALL_CLOCK_NEW:
1943 	case MSR_KVM_WALL_CLOCK:
1944 		vcpu->kvm->arch.wall_clock = data;
1945 		kvm_write_wall_clock(vcpu->kvm, data);
1946 		break;
1947 	case MSR_KVM_SYSTEM_TIME_NEW:
1948 	case MSR_KVM_SYSTEM_TIME: {
1949 		kvmclock_reset(vcpu);
1950 
1951 		vcpu->arch.time = data;
1952 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1953 
1954 		/* we verify if the enable bit is set... */
1955 		if (!(data & 1))
1956 			break;
1957 
1958 		/* ...but clean it before doing the actual write */
1959 		vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
1960 
1961 		vcpu->arch.time_page =
1962 				gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
1963 
1964 		if (is_error_page(vcpu->arch.time_page))
1965 			vcpu->arch.time_page = NULL;
1966 
1967 		break;
1968 	}
1969 	case MSR_KVM_ASYNC_PF_EN:
1970 		if (kvm_pv_enable_async_pf(vcpu, data))
1971 			return 1;
1972 		break;
1973 	case MSR_KVM_STEAL_TIME:
1974 
1975 		if (unlikely(!sched_info_on()))
1976 			return 1;
1977 
1978 		if (data & KVM_STEAL_RESERVED_MASK)
1979 			return 1;
1980 
1981 		if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
1982 							data & KVM_STEAL_VALID_BITS))
1983 			return 1;
1984 
1985 		vcpu->arch.st.msr_val = data;
1986 
1987 		if (!(data & KVM_MSR_ENABLED))
1988 			break;
1989 
1990 		vcpu->arch.st.last_steal = current->sched_info.run_delay;
1991 
1992 		preempt_disable();
1993 		accumulate_steal_time(vcpu);
1994 		preempt_enable();
1995 
1996 		kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
1997 
1998 		break;
1999 	case MSR_KVM_PV_EOI_EN:
2000 		if (kvm_lapic_enable_pv_eoi(vcpu, data))
2001 			return 1;
2002 		break;
2003 
2004 	case MSR_IA32_MCG_CTL:
2005 	case MSR_IA32_MCG_STATUS:
2006 	case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
2007 		return set_msr_mce(vcpu, msr, data);
2008 
2009 	/* Performance counters are not protected by a CPUID bit,
2010 	 * so we should check all of them in the generic path for the sake of
2011 	 * cross vendor migration.
2012 	 * Writing a zero into the event select MSRs disables them,
2013 	 * which we perfectly emulate ;-). Any other value should be at least
2014 	 * reported, some guests depend on them.
2015 	 */
2016 	case MSR_K7_EVNTSEL0:
2017 	case MSR_K7_EVNTSEL1:
2018 	case MSR_K7_EVNTSEL2:
2019 	case MSR_K7_EVNTSEL3:
2020 		if (data != 0)
2021 			vcpu_unimpl(vcpu, "unimplemented perfctr wrmsr: "
2022 				    "0x%x data 0x%llx\n", msr, data);
2023 		break;
2024 	/* at least RHEL 4 unconditionally writes to the perfctr registers,
2025 	 * so we ignore writes to make it happy.
2026 	 */
2027 	case MSR_K7_PERFCTR0:
2028 	case MSR_K7_PERFCTR1:
2029 	case MSR_K7_PERFCTR2:
2030 	case MSR_K7_PERFCTR3:
2031 		vcpu_unimpl(vcpu, "unimplemented perfctr wrmsr: "
2032 			    "0x%x data 0x%llx\n", msr, data);
2033 		break;
2034 	case MSR_P6_PERFCTR0:
2035 	case MSR_P6_PERFCTR1:
2036 		pr = true;
2037 	case MSR_P6_EVNTSEL0:
2038 	case MSR_P6_EVNTSEL1:
2039 		if (kvm_pmu_msr(vcpu, msr))
2040 			return kvm_pmu_set_msr(vcpu, msr, data);
2041 
2042 		if (pr || data != 0)
2043 			vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2044 				    "0x%x data 0x%llx\n", msr, data);
2045 		break;
2046 	case MSR_K7_CLK_CTL:
2047 		/*
2048 		 * Ignore all writes to this no longer documented MSR.
2049 		 * Writes are only relevant for old K7 processors,
2050 		 * all pre-dating SVM, but a recommended workaround from
2051 		 * AMD for these chips. It is possible to specify the
2052 		 * affected processor models on the command line, hence
2053 		 * the need to ignore the workaround.
2054 		 */
2055 		break;
2056 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2057 		if (kvm_hv_msr_partition_wide(msr)) {
2058 			int r;
2059 			mutex_lock(&vcpu->kvm->lock);
2060 			r = set_msr_hyperv_pw(vcpu, msr, data);
2061 			mutex_unlock(&vcpu->kvm->lock);
2062 			return r;
2063 		} else
2064 			return set_msr_hyperv(vcpu, msr, data);
2065 		break;
2066 	case MSR_IA32_BBL_CR_CTL3:
2067 		/* Drop writes to this legacy MSR -- see rdmsr
2068 		 * counterpart for further detail.
2069 		 */
2070 		vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n", msr, data);
2071 		break;
2072 	case MSR_AMD64_OSVW_ID_LENGTH:
2073 		if (!guest_cpuid_has_osvw(vcpu))
2074 			return 1;
2075 		vcpu->arch.osvw.length = data;
2076 		break;
2077 	case MSR_AMD64_OSVW_STATUS:
2078 		if (!guest_cpuid_has_osvw(vcpu))
2079 			return 1;
2080 		vcpu->arch.osvw.status = data;
2081 		break;
2082 	default:
2083 		if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2084 			return xen_hvm_config(vcpu, data);
2085 		if (kvm_pmu_msr(vcpu, msr))
2086 			return kvm_pmu_set_msr(vcpu, msr, data);
2087 		if (!ignore_msrs) {
2088 			vcpu_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
2089 				    msr, data);
2090 			return 1;
2091 		} else {
2092 			vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
2093 				    msr, data);
2094 			break;
2095 		}
2096 	}
2097 	return 0;
2098 }
2099 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2100 
2101 
2102 /*
2103  * Reads an msr value (of 'msr_index') into 'pdata'.
2104  * Returns 0 on success, non-0 otherwise.
2105  * Assumes vcpu_load() was already called.
2106  */
2107 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
2108 {
2109 	return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
2110 }
2111 
2112 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2113 {
2114 	u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
2115 
2116 	if (!msr_mtrr_valid(msr))
2117 		return 1;
2118 
2119 	if (msr == MSR_MTRRdefType)
2120 		*pdata = vcpu->arch.mtrr_state.def_type +
2121 			 (vcpu->arch.mtrr_state.enabled << 10);
2122 	else if (msr == MSR_MTRRfix64K_00000)
2123 		*pdata = p[0];
2124 	else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
2125 		*pdata = p[1 + msr - MSR_MTRRfix16K_80000];
2126 	else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
2127 		*pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
2128 	else if (msr == MSR_IA32_CR_PAT)
2129 		*pdata = vcpu->arch.pat;
2130 	else {	/* Variable MTRRs */
2131 		int idx, is_mtrr_mask;
2132 		u64 *pt;
2133 
2134 		idx = (msr - 0x200) / 2;
2135 		is_mtrr_mask = msr - 0x200 - 2 * idx;
2136 		if (!is_mtrr_mask)
2137 			pt =
2138 			  (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
2139 		else
2140 			pt =
2141 			  (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
2142 		*pdata = *pt;
2143 	}
2144 
2145 	return 0;
2146 }
2147 
2148 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2149 {
2150 	u64 data;
2151 	u64 mcg_cap = vcpu->arch.mcg_cap;
2152 	unsigned bank_num = mcg_cap & 0xff;
2153 
2154 	switch (msr) {
2155 	case MSR_IA32_P5_MC_ADDR:
2156 	case MSR_IA32_P5_MC_TYPE:
2157 		data = 0;
2158 		break;
2159 	case MSR_IA32_MCG_CAP:
2160 		data = vcpu->arch.mcg_cap;
2161 		break;
2162 	case MSR_IA32_MCG_CTL:
2163 		if (!(mcg_cap & MCG_CTL_P))
2164 			return 1;
2165 		data = vcpu->arch.mcg_ctl;
2166 		break;
2167 	case MSR_IA32_MCG_STATUS:
2168 		data = vcpu->arch.mcg_status;
2169 		break;
2170 	default:
2171 		if (msr >= MSR_IA32_MC0_CTL &&
2172 		    msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
2173 			u32 offset = msr - MSR_IA32_MC0_CTL;
2174 			data = vcpu->arch.mce_banks[offset];
2175 			break;
2176 		}
2177 		return 1;
2178 	}
2179 	*pdata = data;
2180 	return 0;
2181 }
2182 
2183 static int get_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2184 {
2185 	u64 data = 0;
2186 	struct kvm *kvm = vcpu->kvm;
2187 
2188 	switch (msr) {
2189 	case HV_X64_MSR_GUEST_OS_ID:
2190 		data = kvm->arch.hv_guest_os_id;
2191 		break;
2192 	case HV_X64_MSR_HYPERCALL:
2193 		data = kvm->arch.hv_hypercall;
2194 		break;
2195 	default:
2196 		vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
2197 		return 1;
2198 	}
2199 
2200 	*pdata = data;
2201 	return 0;
2202 }
2203 
2204 static int get_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2205 {
2206 	u64 data = 0;
2207 
2208 	switch (msr) {
2209 	case HV_X64_MSR_VP_INDEX: {
2210 		int r;
2211 		struct kvm_vcpu *v;
2212 		kvm_for_each_vcpu(r, v, vcpu->kvm)
2213 			if (v == vcpu)
2214 				data = r;
2215 		break;
2216 	}
2217 	case HV_X64_MSR_EOI:
2218 		return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
2219 	case HV_X64_MSR_ICR:
2220 		return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
2221 	case HV_X64_MSR_TPR:
2222 		return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
2223 	case HV_X64_MSR_APIC_ASSIST_PAGE:
2224 		data = vcpu->arch.hv_vapic;
2225 		break;
2226 	default:
2227 		vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
2228 		return 1;
2229 	}
2230 	*pdata = data;
2231 	return 0;
2232 }
2233 
2234 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2235 {
2236 	u64 data;
2237 
2238 	switch (msr) {
2239 	case MSR_IA32_PLATFORM_ID:
2240 	case MSR_IA32_EBL_CR_POWERON:
2241 	case MSR_IA32_DEBUGCTLMSR:
2242 	case MSR_IA32_LASTBRANCHFROMIP:
2243 	case MSR_IA32_LASTBRANCHTOIP:
2244 	case MSR_IA32_LASTINTFROMIP:
2245 	case MSR_IA32_LASTINTTOIP:
2246 	case MSR_K8_SYSCFG:
2247 	case MSR_K7_HWCR:
2248 	case MSR_VM_HSAVE_PA:
2249 	case MSR_K7_EVNTSEL0:
2250 	case MSR_K7_PERFCTR0:
2251 	case MSR_K8_INT_PENDING_MSG:
2252 	case MSR_AMD64_NB_CFG:
2253 	case MSR_FAM10H_MMIO_CONF_BASE:
2254 		data = 0;
2255 		break;
2256 	case MSR_P6_PERFCTR0:
2257 	case MSR_P6_PERFCTR1:
2258 	case MSR_P6_EVNTSEL0:
2259 	case MSR_P6_EVNTSEL1:
2260 		if (kvm_pmu_msr(vcpu, msr))
2261 			return kvm_pmu_get_msr(vcpu, msr, pdata);
2262 		data = 0;
2263 		break;
2264 	case MSR_IA32_UCODE_REV:
2265 		data = 0x100000000ULL;
2266 		break;
2267 	case MSR_MTRRcap:
2268 		data = 0x500 | KVM_NR_VAR_MTRR;
2269 		break;
2270 	case 0x200 ... 0x2ff:
2271 		return get_msr_mtrr(vcpu, msr, pdata);
2272 	case 0xcd: /* fsb frequency */
2273 		data = 3;
2274 		break;
2275 		/*
2276 		 * MSR_EBC_FREQUENCY_ID
2277 		 * Conservative value valid for even the basic CPU models.
2278 		 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2279 		 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2280 		 * and 266MHz for model 3, or 4. Set Core Clock
2281 		 * Frequency to System Bus Frequency Ratio to 1 (bits
2282 		 * 31:24) even though these are only valid for CPU
2283 		 * models > 2, however guests may end up dividing or
2284 		 * multiplying by zero otherwise.
2285 		 */
2286 	case MSR_EBC_FREQUENCY_ID:
2287 		data = 1 << 24;
2288 		break;
2289 	case MSR_IA32_APICBASE:
2290 		data = kvm_get_apic_base(vcpu);
2291 		break;
2292 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2293 		return kvm_x2apic_msr_read(vcpu, msr, pdata);
2294 		break;
2295 	case MSR_IA32_TSCDEADLINE:
2296 		data = kvm_get_lapic_tscdeadline_msr(vcpu);
2297 		break;
2298 	case MSR_IA32_TSC_ADJUST:
2299 		data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
2300 		break;
2301 	case MSR_IA32_MISC_ENABLE:
2302 		data = vcpu->arch.ia32_misc_enable_msr;
2303 		break;
2304 	case MSR_IA32_PERF_STATUS:
2305 		/* TSC increment by tick */
2306 		data = 1000ULL;
2307 		/* CPU multiplier */
2308 		data |= (((uint64_t)4ULL) << 40);
2309 		break;
2310 	case MSR_EFER:
2311 		data = vcpu->arch.efer;
2312 		break;
2313 	case MSR_KVM_WALL_CLOCK:
2314 	case MSR_KVM_WALL_CLOCK_NEW:
2315 		data = vcpu->kvm->arch.wall_clock;
2316 		break;
2317 	case MSR_KVM_SYSTEM_TIME:
2318 	case MSR_KVM_SYSTEM_TIME_NEW:
2319 		data = vcpu->arch.time;
2320 		break;
2321 	case MSR_KVM_ASYNC_PF_EN:
2322 		data = vcpu->arch.apf.msr_val;
2323 		break;
2324 	case MSR_KVM_STEAL_TIME:
2325 		data = vcpu->arch.st.msr_val;
2326 		break;
2327 	case MSR_KVM_PV_EOI_EN:
2328 		data = vcpu->arch.pv_eoi.msr_val;
2329 		break;
2330 	case MSR_IA32_P5_MC_ADDR:
2331 	case MSR_IA32_P5_MC_TYPE:
2332 	case MSR_IA32_MCG_CAP:
2333 	case MSR_IA32_MCG_CTL:
2334 	case MSR_IA32_MCG_STATUS:
2335 	case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
2336 		return get_msr_mce(vcpu, msr, pdata);
2337 	case MSR_K7_CLK_CTL:
2338 		/*
2339 		 * Provide expected ramp-up count for K7. All other
2340 		 * are set to zero, indicating minimum divisors for
2341 		 * every field.
2342 		 *
2343 		 * This prevents guest kernels on AMD host with CPU
2344 		 * type 6, model 8 and higher from exploding due to
2345 		 * the rdmsr failing.
2346 		 */
2347 		data = 0x20000000;
2348 		break;
2349 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2350 		if (kvm_hv_msr_partition_wide(msr)) {
2351 			int r;
2352 			mutex_lock(&vcpu->kvm->lock);
2353 			r = get_msr_hyperv_pw(vcpu, msr, pdata);
2354 			mutex_unlock(&vcpu->kvm->lock);
2355 			return r;
2356 		} else
2357 			return get_msr_hyperv(vcpu, msr, pdata);
2358 		break;
2359 	case MSR_IA32_BBL_CR_CTL3:
2360 		/* This legacy MSR exists but isn't fully documented in current
2361 		 * silicon.  It is however accessed by winxp in very narrow
2362 		 * scenarios where it sets bit #19, itself documented as
2363 		 * a "reserved" bit.  Best effort attempt to source coherent
2364 		 * read data here should the balance of the register be
2365 		 * interpreted by the guest:
2366 		 *
2367 		 * L2 cache control register 3: 64GB range, 256KB size,
2368 		 * enabled, latency 0x1, configured
2369 		 */
2370 		data = 0xbe702111;
2371 		break;
2372 	case MSR_AMD64_OSVW_ID_LENGTH:
2373 		if (!guest_cpuid_has_osvw(vcpu))
2374 			return 1;
2375 		data = vcpu->arch.osvw.length;
2376 		break;
2377 	case MSR_AMD64_OSVW_STATUS:
2378 		if (!guest_cpuid_has_osvw(vcpu))
2379 			return 1;
2380 		data = vcpu->arch.osvw.status;
2381 		break;
2382 	default:
2383 		if (kvm_pmu_msr(vcpu, msr))
2384 			return kvm_pmu_get_msr(vcpu, msr, pdata);
2385 		if (!ignore_msrs) {
2386 			vcpu_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
2387 			return 1;
2388 		} else {
2389 			vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr);
2390 			data = 0;
2391 		}
2392 		break;
2393 	}
2394 	*pdata = data;
2395 	return 0;
2396 }
2397 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2398 
2399 /*
2400  * Read or write a bunch of msrs. All parameters are kernel addresses.
2401  *
2402  * @return number of msrs set successfully.
2403  */
2404 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2405 		    struct kvm_msr_entry *entries,
2406 		    int (*do_msr)(struct kvm_vcpu *vcpu,
2407 				  unsigned index, u64 *data))
2408 {
2409 	int i, idx;
2410 
2411 	idx = srcu_read_lock(&vcpu->kvm->srcu);
2412 	for (i = 0; i < msrs->nmsrs; ++i)
2413 		if (do_msr(vcpu, entries[i].index, &entries[i].data))
2414 			break;
2415 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
2416 
2417 	return i;
2418 }
2419 
2420 /*
2421  * Read or write a bunch of msrs. Parameters are user addresses.
2422  *
2423  * @return number of msrs set successfully.
2424  */
2425 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2426 		  int (*do_msr)(struct kvm_vcpu *vcpu,
2427 				unsigned index, u64 *data),
2428 		  int writeback)
2429 {
2430 	struct kvm_msrs msrs;
2431 	struct kvm_msr_entry *entries;
2432 	int r, n;
2433 	unsigned size;
2434 
2435 	r = -EFAULT;
2436 	if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2437 		goto out;
2438 
2439 	r = -E2BIG;
2440 	if (msrs.nmsrs >= MAX_IO_MSRS)
2441 		goto out;
2442 
2443 	size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2444 	entries = memdup_user(user_msrs->entries, size);
2445 	if (IS_ERR(entries)) {
2446 		r = PTR_ERR(entries);
2447 		goto out;
2448 	}
2449 
2450 	r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2451 	if (r < 0)
2452 		goto out_free;
2453 
2454 	r = -EFAULT;
2455 	if (writeback && copy_to_user(user_msrs->entries, entries, size))
2456 		goto out_free;
2457 
2458 	r = n;
2459 
2460 out_free:
2461 	kfree(entries);
2462 out:
2463 	return r;
2464 }
2465 
2466 int kvm_dev_ioctl_check_extension(long ext)
2467 {
2468 	int r;
2469 
2470 	switch (ext) {
2471 	case KVM_CAP_IRQCHIP:
2472 	case KVM_CAP_HLT:
2473 	case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2474 	case KVM_CAP_SET_TSS_ADDR:
2475 	case KVM_CAP_EXT_CPUID:
2476 	case KVM_CAP_CLOCKSOURCE:
2477 	case KVM_CAP_PIT:
2478 	case KVM_CAP_NOP_IO_DELAY:
2479 	case KVM_CAP_MP_STATE:
2480 	case KVM_CAP_SYNC_MMU:
2481 	case KVM_CAP_USER_NMI:
2482 	case KVM_CAP_REINJECT_CONTROL:
2483 	case KVM_CAP_IRQ_INJECT_STATUS:
2484 	case KVM_CAP_ASSIGN_DEV_IRQ:
2485 	case KVM_CAP_IRQFD:
2486 	case KVM_CAP_IOEVENTFD:
2487 	case KVM_CAP_PIT2:
2488 	case KVM_CAP_PIT_STATE2:
2489 	case KVM_CAP_SET_IDENTITY_MAP_ADDR:
2490 	case KVM_CAP_XEN_HVM:
2491 	case KVM_CAP_ADJUST_CLOCK:
2492 	case KVM_CAP_VCPU_EVENTS:
2493 	case KVM_CAP_HYPERV:
2494 	case KVM_CAP_HYPERV_VAPIC:
2495 	case KVM_CAP_HYPERV_SPIN:
2496 	case KVM_CAP_PCI_SEGMENT:
2497 	case KVM_CAP_DEBUGREGS:
2498 	case KVM_CAP_X86_ROBUST_SINGLESTEP:
2499 	case KVM_CAP_XSAVE:
2500 	case KVM_CAP_ASYNC_PF:
2501 	case KVM_CAP_GET_TSC_KHZ:
2502 	case KVM_CAP_PCI_2_3:
2503 	case KVM_CAP_KVMCLOCK_CTRL:
2504 	case KVM_CAP_READONLY_MEM:
2505 	case KVM_CAP_IRQFD_RESAMPLE:
2506 		r = 1;
2507 		break;
2508 	case KVM_CAP_COALESCED_MMIO:
2509 		r = KVM_COALESCED_MMIO_PAGE_OFFSET;
2510 		break;
2511 	case KVM_CAP_VAPIC:
2512 		r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2513 		break;
2514 	case KVM_CAP_NR_VCPUS:
2515 		r = KVM_SOFT_MAX_VCPUS;
2516 		break;
2517 	case KVM_CAP_MAX_VCPUS:
2518 		r = KVM_MAX_VCPUS;
2519 		break;
2520 	case KVM_CAP_NR_MEMSLOTS:
2521 		r = KVM_MEMORY_SLOTS;
2522 		break;
2523 	case KVM_CAP_PV_MMU:	/* obsolete */
2524 		r = 0;
2525 		break;
2526 	case KVM_CAP_IOMMU:
2527 		r = iommu_present(&pci_bus_type);
2528 		break;
2529 	case KVM_CAP_MCE:
2530 		r = KVM_MAX_MCE_BANKS;
2531 		break;
2532 	case KVM_CAP_XCRS:
2533 		r = cpu_has_xsave;
2534 		break;
2535 	case KVM_CAP_TSC_CONTROL:
2536 		r = kvm_has_tsc_control;
2537 		break;
2538 	case KVM_CAP_TSC_DEADLINE_TIMER:
2539 		r = boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER);
2540 		break;
2541 	default:
2542 		r = 0;
2543 		break;
2544 	}
2545 	return r;
2546 
2547 }
2548 
2549 long kvm_arch_dev_ioctl(struct file *filp,
2550 			unsigned int ioctl, unsigned long arg)
2551 {
2552 	void __user *argp = (void __user *)arg;
2553 	long r;
2554 
2555 	switch (ioctl) {
2556 	case KVM_GET_MSR_INDEX_LIST: {
2557 		struct kvm_msr_list __user *user_msr_list = argp;
2558 		struct kvm_msr_list msr_list;
2559 		unsigned n;
2560 
2561 		r = -EFAULT;
2562 		if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2563 			goto out;
2564 		n = msr_list.nmsrs;
2565 		msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2566 		if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2567 			goto out;
2568 		r = -E2BIG;
2569 		if (n < msr_list.nmsrs)
2570 			goto out;
2571 		r = -EFAULT;
2572 		if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2573 				 num_msrs_to_save * sizeof(u32)))
2574 			goto out;
2575 		if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2576 				 &emulated_msrs,
2577 				 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2578 			goto out;
2579 		r = 0;
2580 		break;
2581 	}
2582 	case KVM_GET_SUPPORTED_CPUID: {
2583 		struct kvm_cpuid2 __user *cpuid_arg = argp;
2584 		struct kvm_cpuid2 cpuid;
2585 
2586 		r = -EFAULT;
2587 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2588 			goto out;
2589 		r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
2590 						      cpuid_arg->entries);
2591 		if (r)
2592 			goto out;
2593 
2594 		r = -EFAULT;
2595 		if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2596 			goto out;
2597 		r = 0;
2598 		break;
2599 	}
2600 	case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2601 		u64 mce_cap;
2602 
2603 		mce_cap = KVM_MCE_CAP_SUPPORTED;
2604 		r = -EFAULT;
2605 		if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
2606 			goto out;
2607 		r = 0;
2608 		break;
2609 	}
2610 	default:
2611 		r = -EINVAL;
2612 	}
2613 out:
2614 	return r;
2615 }
2616 
2617 static void wbinvd_ipi(void *garbage)
2618 {
2619 	wbinvd();
2620 }
2621 
2622 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2623 {
2624 	return vcpu->kvm->arch.iommu_domain &&
2625 		!(vcpu->kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY);
2626 }
2627 
2628 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2629 {
2630 	/* Address WBINVD may be executed by guest */
2631 	if (need_emulate_wbinvd(vcpu)) {
2632 		if (kvm_x86_ops->has_wbinvd_exit())
2633 			cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2634 		else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2635 			smp_call_function_single(vcpu->cpu,
2636 					wbinvd_ipi, NULL, 1);
2637 	}
2638 
2639 	kvm_x86_ops->vcpu_load(vcpu, cpu);
2640 
2641 	/* Apply any externally detected TSC adjustments (due to suspend) */
2642 	if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
2643 		adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
2644 		vcpu->arch.tsc_offset_adjustment = 0;
2645 		set_bit(KVM_REQ_CLOCK_UPDATE, &vcpu->requests);
2646 	}
2647 
2648 	if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
2649 		s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
2650 				native_read_tsc() - vcpu->arch.last_host_tsc;
2651 		if (tsc_delta < 0)
2652 			mark_tsc_unstable("KVM discovered backwards TSC");
2653 		if (check_tsc_unstable()) {
2654 			u64 offset = kvm_x86_ops->compute_tsc_offset(vcpu,
2655 						vcpu->arch.last_guest_tsc);
2656 			kvm_x86_ops->write_tsc_offset(vcpu, offset);
2657 			vcpu->arch.tsc_catchup = 1;
2658 		}
2659 		/*
2660 		 * On a host with synchronized TSC, there is no need to update
2661 		 * kvmclock on vcpu->cpu migration
2662 		 */
2663 		if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
2664 			kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2665 		if (vcpu->cpu != cpu)
2666 			kvm_migrate_timers(vcpu);
2667 		vcpu->cpu = cpu;
2668 	}
2669 
2670 	accumulate_steal_time(vcpu);
2671 	kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2672 }
2673 
2674 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2675 {
2676 	kvm_x86_ops->vcpu_put(vcpu);
2677 	kvm_put_guest_fpu(vcpu);
2678 	vcpu->arch.last_host_tsc = native_read_tsc();
2679 }
2680 
2681 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2682 				    struct kvm_lapic_state *s)
2683 {
2684 	memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
2685 
2686 	return 0;
2687 }
2688 
2689 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2690 				    struct kvm_lapic_state *s)
2691 {
2692 	kvm_apic_post_state_restore(vcpu, s);
2693 	update_cr8_intercept(vcpu);
2694 
2695 	return 0;
2696 }
2697 
2698 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2699 				    struct kvm_interrupt *irq)
2700 {
2701 	if (irq->irq < 0 || irq->irq >= KVM_NR_INTERRUPTS)
2702 		return -EINVAL;
2703 	if (irqchip_in_kernel(vcpu->kvm))
2704 		return -ENXIO;
2705 
2706 	kvm_queue_interrupt(vcpu, irq->irq, false);
2707 	kvm_make_request(KVM_REQ_EVENT, vcpu);
2708 
2709 	return 0;
2710 }
2711 
2712 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2713 {
2714 	kvm_inject_nmi(vcpu);
2715 
2716 	return 0;
2717 }
2718 
2719 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2720 					   struct kvm_tpr_access_ctl *tac)
2721 {
2722 	if (tac->flags)
2723 		return -EINVAL;
2724 	vcpu->arch.tpr_access_reporting = !!tac->enabled;
2725 	return 0;
2726 }
2727 
2728 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2729 					u64 mcg_cap)
2730 {
2731 	int r;
2732 	unsigned bank_num = mcg_cap & 0xff, bank;
2733 
2734 	r = -EINVAL;
2735 	if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2736 		goto out;
2737 	if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2738 		goto out;
2739 	r = 0;
2740 	vcpu->arch.mcg_cap = mcg_cap;
2741 	/* Init IA32_MCG_CTL to all 1s */
2742 	if (mcg_cap & MCG_CTL_P)
2743 		vcpu->arch.mcg_ctl = ~(u64)0;
2744 	/* Init IA32_MCi_CTL to all 1s */
2745 	for (bank = 0; bank < bank_num; bank++)
2746 		vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2747 out:
2748 	return r;
2749 }
2750 
2751 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2752 				      struct kvm_x86_mce *mce)
2753 {
2754 	u64 mcg_cap = vcpu->arch.mcg_cap;
2755 	unsigned bank_num = mcg_cap & 0xff;
2756 	u64 *banks = vcpu->arch.mce_banks;
2757 
2758 	if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2759 		return -EINVAL;
2760 	/*
2761 	 * if IA32_MCG_CTL is not all 1s, the uncorrected error
2762 	 * reporting is disabled
2763 	 */
2764 	if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2765 	    vcpu->arch.mcg_ctl != ~(u64)0)
2766 		return 0;
2767 	banks += 4 * mce->bank;
2768 	/*
2769 	 * if IA32_MCi_CTL is not all 1s, the uncorrected error
2770 	 * reporting is disabled for the bank
2771 	 */
2772 	if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2773 		return 0;
2774 	if (mce->status & MCI_STATUS_UC) {
2775 		if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
2776 		    !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
2777 			kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2778 			return 0;
2779 		}
2780 		if (banks[1] & MCI_STATUS_VAL)
2781 			mce->status |= MCI_STATUS_OVER;
2782 		banks[2] = mce->addr;
2783 		banks[3] = mce->misc;
2784 		vcpu->arch.mcg_status = mce->mcg_status;
2785 		banks[1] = mce->status;
2786 		kvm_queue_exception(vcpu, MC_VECTOR);
2787 	} else if (!(banks[1] & MCI_STATUS_VAL)
2788 		   || !(banks[1] & MCI_STATUS_UC)) {
2789 		if (banks[1] & MCI_STATUS_VAL)
2790 			mce->status |= MCI_STATUS_OVER;
2791 		banks[2] = mce->addr;
2792 		banks[3] = mce->misc;
2793 		banks[1] = mce->status;
2794 	} else
2795 		banks[1] |= MCI_STATUS_OVER;
2796 	return 0;
2797 }
2798 
2799 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2800 					       struct kvm_vcpu_events *events)
2801 {
2802 	process_nmi(vcpu);
2803 	events->exception.injected =
2804 		vcpu->arch.exception.pending &&
2805 		!kvm_exception_is_soft(vcpu->arch.exception.nr);
2806 	events->exception.nr = vcpu->arch.exception.nr;
2807 	events->exception.has_error_code = vcpu->arch.exception.has_error_code;
2808 	events->exception.pad = 0;
2809 	events->exception.error_code = vcpu->arch.exception.error_code;
2810 
2811 	events->interrupt.injected =
2812 		vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
2813 	events->interrupt.nr = vcpu->arch.interrupt.nr;
2814 	events->interrupt.soft = 0;
2815 	events->interrupt.shadow =
2816 		kvm_x86_ops->get_interrupt_shadow(vcpu,
2817 			KVM_X86_SHADOW_INT_MOV_SS | KVM_X86_SHADOW_INT_STI);
2818 
2819 	events->nmi.injected = vcpu->arch.nmi_injected;
2820 	events->nmi.pending = vcpu->arch.nmi_pending != 0;
2821 	events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
2822 	events->nmi.pad = 0;
2823 
2824 	events->sipi_vector = vcpu->arch.sipi_vector;
2825 
2826 	events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
2827 			 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2828 			 | KVM_VCPUEVENT_VALID_SHADOW);
2829 	memset(&events->reserved, 0, sizeof(events->reserved));
2830 }
2831 
2832 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2833 					      struct kvm_vcpu_events *events)
2834 {
2835 	if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
2836 			      | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2837 			      | KVM_VCPUEVENT_VALID_SHADOW))
2838 		return -EINVAL;
2839 
2840 	process_nmi(vcpu);
2841 	vcpu->arch.exception.pending = events->exception.injected;
2842 	vcpu->arch.exception.nr = events->exception.nr;
2843 	vcpu->arch.exception.has_error_code = events->exception.has_error_code;
2844 	vcpu->arch.exception.error_code = events->exception.error_code;
2845 
2846 	vcpu->arch.interrupt.pending = events->interrupt.injected;
2847 	vcpu->arch.interrupt.nr = events->interrupt.nr;
2848 	vcpu->arch.interrupt.soft = events->interrupt.soft;
2849 	if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
2850 		kvm_x86_ops->set_interrupt_shadow(vcpu,
2851 						  events->interrupt.shadow);
2852 
2853 	vcpu->arch.nmi_injected = events->nmi.injected;
2854 	if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
2855 		vcpu->arch.nmi_pending = events->nmi.pending;
2856 	kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
2857 
2858 	if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR)
2859 		vcpu->arch.sipi_vector = events->sipi_vector;
2860 
2861 	kvm_make_request(KVM_REQ_EVENT, vcpu);
2862 
2863 	return 0;
2864 }
2865 
2866 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
2867 					     struct kvm_debugregs *dbgregs)
2868 {
2869 	memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
2870 	dbgregs->dr6 = vcpu->arch.dr6;
2871 	dbgregs->dr7 = vcpu->arch.dr7;
2872 	dbgregs->flags = 0;
2873 	memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
2874 }
2875 
2876 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
2877 					    struct kvm_debugregs *dbgregs)
2878 {
2879 	if (dbgregs->flags)
2880 		return -EINVAL;
2881 
2882 	memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
2883 	vcpu->arch.dr6 = dbgregs->dr6;
2884 	vcpu->arch.dr7 = dbgregs->dr7;
2885 
2886 	return 0;
2887 }
2888 
2889 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
2890 					 struct kvm_xsave *guest_xsave)
2891 {
2892 	if (cpu_has_xsave)
2893 		memcpy(guest_xsave->region,
2894 			&vcpu->arch.guest_fpu.state->xsave,
2895 			xstate_size);
2896 	else {
2897 		memcpy(guest_xsave->region,
2898 			&vcpu->arch.guest_fpu.state->fxsave,
2899 			sizeof(struct i387_fxsave_struct));
2900 		*(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
2901 			XSTATE_FPSSE;
2902 	}
2903 }
2904 
2905 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
2906 					struct kvm_xsave *guest_xsave)
2907 {
2908 	u64 xstate_bv =
2909 		*(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
2910 
2911 	if (cpu_has_xsave)
2912 		memcpy(&vcpu->arch.guest_fpu.state->xsave,
2913 			guest_xsave->region, xstate_size);
2914 	else {
2915 		if (xstate_bv & ~XSTATE_FPSSE)
2916 			return -EINVAL;
2917 		memcpy(&vcpu->arch.guest_fpu.state->fxsave,
2918 			guest_xsave->region, sizeof(struct i387_fxsave_struct));
2919 	}
2920 	return 0;
2921 }
2922 
2923 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
2924 					struct kvm_xcrs *guest_xcrs)
2925 {
2926 	if (!cpu_has_xsave) {
2927 		guest_xcrs->nr_xcrs = 0;
2928 		return;
2929 	}
2930 
2931 	guest_xcrs->nr_xcrs = 1;
2932 	guest_xcrs->flags = 0;
2933 	guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
2934 	guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
2935 }
2936 
2937 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
2938 				       struct kvm_xcrs *guest_xcrs)
2939 {
2940 	int i, r = 0;
2941 
2942 	if (!cpu_has_xsave)
2943 		return -EINVAL;
2944 
2945 	if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
2946 		return -EINVAL;
2947 
2948 	for (i = 0; i < guest_xcrs->nr_xcrs; i++)
2949 		/* Only support XCR0 currently */
2950 		if (guest_xcrs->xcrs[0].xcr == XCR_XFEATURE_ENABLED_MASK) {
2951 			r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
2952 				guest_xcrs->xcrs[0].value);
2953 			break;
2954 		}
2955 	if (r)
2956 		r = -EINVAL;
2957 	return r;
2958 }
2959 
2960 /*
2961  * kvm_set_guest_paused() indicates to the guest kernel that it has been
2962  * stopped by the hypervisor.  This function will be called from the host only.
2963  * EINVAL is returned when the host attempts to set the flag for a guest that
2964  * does not support pv clocks.
2965  */
2966 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
2967 {
2968 	if (!vcpu->arch.time_page)
2969 		return -EINVAL;
2970 	vcpu->arch.pvclock_set_guest_stopped_request = true;
2971 	kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2972 	return 0;
2973 }
2974 
2975 long kvm_arch_vcpu_ioctl(struct file *filp,
2976 			 unsigned int ioctl, unsigned long arg)
2977 {
2978 	struct kvm_vcpu *vcpu = filp->private_data;
2979 	void __user *argp = (void __user *)arg;
2980 	int r;
2981 	union {
2982 		struct kvm_lapic_state *lapic;
2983 		struct kvm_xsave *xsave;
2984 		struct kvm_xcrs *xcrs;
2985 		void *buffer;
2986 	} u;
2987 
2988 	u.buffer = NULL;
2989 	switch (ioctl) {
2990 	case KVM_GET_LAPIC: {
2991 		r = -EINVAL;
2992 		if (!vcpu->arch.apic)
2993 			goto out;
2994 		u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
2995 
2996 		r = -ENOMEM;
2997 		if (!u.lapic)
2998 			goto out;
2999 		r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
3000 		if (r)
3001 			goto out;
3002 		r = -EFAULT;
3003 		if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
3004 			goto out;
3005 		r = 0;
3006 		break;
3007 	}
3008 	case KVM_SET_LAPIC: {
3009 		r = -EINVAL;
3010 		if (!vcpu->arch.apic)
3011 			goto out;
3012 		u.lapic = memdup_user(argp, sizeof(*u.lapic));
3013 		if (IS_ERR(u.lapic))
3014 			return PTR_ERR(u.lapic);
3015 
3016 		r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
3017 		break;
3018 	}
3019 	case KVM_INTERRUPT: {
3020 		struct kvm_interrupt irq;
3021 
3022 		r = -EFAULT;
3023 		if (copy_from_user(&irq, argp, sizeof irq))
3024 			goto out;
3025 		r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
3026 		break;
3027 	}
3028 	case KVM_NMI: {
3029 		r = kvm_vcpu_ioctl_nmi(vcpu);
3030 		break;
3031 	}
3032 	case KVM_SET_CPUID: {
3033 		struct kvm_cpuid __user *cpuid_arg = argp;
3034 		struct kvm_cpuid cpuid;
3035 
3036 		r = -EFAULT;
3037 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3038 			goto out;
3039 		r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3040 		break;
3041 	}
3042 	case KVM_SET_CPUID2: {
3043 		struct kvm_cpuid2 __user *cpuid_arg = argp;
3044 		struct kvm_cpuid2 cpuid;
3045 
3046 		r = -EFAULT;
3047 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3048 			goto out;
3049 		r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
3050 					      cpuid_arg->entries);
3051 		break;
3052 	}
3053 	case KVM_GET_CPUID2: {
3054 		struct kvm_cpuid2 __user *cpuid_arg = argp;
3055 		struct kvm_cpuid2 cpuid;
3056 
3057 		r = -EFAULT;
3058 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3059 			goto out;
3060 		r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
3061 					      cpuid_arg->entries);
3062 		if (r)
3063 			goto out;
3064 		r = -EFAULT;
3065 		if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3066 			goto out;
3067 		r = 0;
3068 		break;
3069 	}
3070 	case KVM_GET_MSRS:
3071 		r = msr_io(vcpu, argp, kvm_get_msr, 1);
3072 		break;
3073 	case KVM_SET_MSRS:
3074 		r = msr_io(vcpu, argp, do_set_msr, 0);
3075 		break;
3076 	case KVM_TPR_ACCESS_REPORTING: {
3077 		struct kvm_tpr_access_ctl tac;
3078 
3079 		r = -EFAULT;
3080 		if (copy_from_user(&tac, argp, sizeof tac))
3081 			goto out;
3082 		r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3083 		if (r)
3084 			goto out;
3085 		r = -EFAULT;
3086 		if (copy_to_user(argp, &tac, sizeof tac))
3087 			goto out;
3088 		r = 0;
3089 		break;
3090 	};
3091 	case KVM_SET_VAPIC_ADDR: {
3092 		struct kvm_vapic_addr va;
3093 
3094 		r = -EINVAL;
3095 		if (!irqchip_in_kernel(vcpu->kvm))
3096 			goto out;
3097 		r = -EFAULT;
3098 		if (copy_from_user(&va, argp, sizeof va))
3099 			goto out;
3100 		r = 0;
3101 		kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
3102 		break;
3103 	}
3104 	case KVM_X86_SETUP_MCE: {
3105 		u64 mcg_cap;
3106 
3107 		r = -EFAULT;
3108 		if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
3109 			goto out;
3110 		r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3111 		break;
3112 	}
3113 	case KVM_X86_SET_MCE: {
3114 		struct kvm_x86_mce mce;
3115 
3116 		r = -EFAULT;
3117 		if (copy_from_user(&mce, argp, sizeof mce))
3118 			goto out;
3119 		r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3120 		break;
3121 	}
3122 	case KVM_GET_VCPU_EVENTS: {
3123 		struct kvm_vcpu_events events;
3124 
3125 		kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3126 
3127 		r = -EFAULT;
3128 		if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3129 			break;
3130 		r = 0;
3131 		break;
3132 	}
3133 	case KVM_SET_VCPU_EVENTS: {
3134 		struct kvm_vcpu_events events;
3135 
3136 		r = -EFAULT;
3137 		if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3138 			break;
3139 
3140 		r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3141 		break;
3142 	}
3143 	case KVM_GET_DEBUGREGS: {
3144 		struct kvm_debugregs dbgregs;
3145 
3146 		kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3147 
3148 		r = -EFAULT;
3149 		if (copy_to_user(argp, &dbgregs,
3150 				 sizeof(struct kvm_debugregs)))
3151 			break;
3152 		r = 0;
3153 		break;
3154 	}
3155 	case KVM_SET_DEBUGREGS: {
3156 		struct kvm_debugregs dbgregs;
3157 
3158 		r = -EFAULT;
3159 		if (copy_from_user(&dbgregs, argp,
3160 				   sizeof(struct kvm_debugregs)))
3161 			break;
3162 
3163 		r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3164 		break;
3165 	}
3166 	case KVM_GET_XSAVE: {
3167 		u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3168 		r = -ENOMEM;
3169 		if (!u.xsave)
3170 			break;
3171 
3172 		kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
3173 
3174 		r = -EFAULT;
3175 		if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
3176 			break;
3177 		r = 0;
3178 		break;
3179 	}
3180 	case KVM_SET_XSAVE: {
3181 		u.xsave = memdup_user(argp, sizeof(*u.xsave));
3182 		if (IS_ERR(u.xsave))
3183 			return PTR_ERR(u.xsave);
3184 
3185 		r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
3186 		break;
3187 	}
3188 	case KVM_GET_XCRS: {
3189 		u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3190 		r = -ENOMEM;
3191 		if (!u.xcrs)
3192 			break;
3193 
3194 		kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
3195 
3196 		r = -EFAULT;
3197 		if (copy_to_user(argp, u.xcrs,
3198 				 sizeof(struct kvm_xcrs)))
3199 			break;
3200 		r = 0;
3201 		break;
3202 	}
3203 	case KVM_SET_XCRS: {
3204 		u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
3205 		if (IS_ERR(u.xcrs))
3206 			return PTR_ERR(u.xcrs);
3207 
3208 		r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3209 		break;
3210 	}
3211 	case KVM_SET_TSC_KHZ: {
3212 		u32 user_tsc_khz;
3213 
3214 		r = -EINVAL;
3215 		user_tsc_khz = (u32)arg;
3216 
3217 		if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3218 			goto out;
3219 
3220 		if (user_tsc_khz == 0)
3221 			user_tsc_khz = tsc_khz;
3222 
3223 		kvm_set_tsc_khz(vcpu, user_tsc_khz);
3224 
3225 		r = 0;
3226 		goto out;
3227 	}
3228 	case KVM_GET_TSC_KHZ: {
3229 		r = vcpu->arch.virtual_tsc_khz;
3230 		goto out;
3231 	}
3232 	case KVM_KVMCLOCK_CTRL: {
3233 		r = kvm_set_guest_paused(vcpu);
3234 		goto out;
3235 	}
3236 	default:
3237 		r = -EINVAL;
3238 	}
3239 out:
3240 	kfree(u.buffer);
3241 	return r;
3242 }
3243 
3244 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
3245 {
3246 	return VM_FAULT_SIGBUS;
3247 }
3248 
3249 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3250 {
3251 	int ret;
3252 
3253 	if (addr > (unsigned int)(-3 * PAGE_SIZE))
3254 		return -EINVAL;
3255 	ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3256 	return ret;
3257 }
3258 
3259 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3260 					      u64 ident_addr)
3261 {
3262 	kvm->arch.ept_identity_map_addr = ident_addr;
3263 	return 0;
3264 }
3265 
3266 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3267 					  u32 kvm_nr_mmu_pages)
3268 {
3269 	if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3270 		return -EINVAL;
3271 
3272 	mutex_lock(&kvm->slots_lock);
3273 	spin_lock(&kvm->mmu_lock);
3274 
3275 	kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3276 	kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3277 
3278 	spin_unlock(&kvm->mmu_lock);
3279 	mutex_unlock(&kvm->slots_lock);
3280 	return 0;
3281 }
3282 
3283 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3284 {
3285 	return kvm->arch.n_max_mmu_pages;
3286 }
3287 
3288 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3289 {
3290 	int r;
3291 
3292 	r = 0;
3293 	switch (chip->chip_id) {
3294 	case KVM_IRQCHIP_PIC_MASTER:
3295 		memcpy(&chip->chip.pic,
3296 			&pic_irqchip(kvm)->pics[0],
3297 			sizeof(struct kvm_pic_state));
3298 		break;
3299 	case KVM_IRQCHIP_PIC_SLAVE:
3300 		memcpy(&chip->chip.pic,
3301 			&pic_irqchip(kvm)->pics[1],
3302 			sizeof(struct kvm_pic_state));
3303 		break;
3304 	case KVM_IRQCHIP_IOAPIC:
3305 		r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
3306 		break;
3307 	default:
3308 		r = -EINVAL;
3309 		break;
3310 	}
3311 	return r;
3312 }
3313 
3314 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3315 {
3316 	int r;
3317 
3318 	r = 0;
3319 	switch (chip->chip_id) {
3320 	case KVM_IRQCHIP_PIC_MASTER:
3321 		spin_lock(&pic_irqchip(kvm)->lock);
3322 		memcpy(&pic_irqchip(kvm)->pics[0],
3323 			&chip->chip.pic,
3324 			sizeof(struct kvm_pic_state));
3325 		spin_unlock(&pic_irqchip(kvm)->lock);
3326 		break;
3327 	case KVM_IRQCHIP_PIC_SLAVE:
3328 		spin_lock(&pic_irqchip(kvm)->lock);
3329 		memcpy(&pic_irqchip(kvm)->pics[1],
3330 			&chip->chip.pic,
3331 			sizeof(struct kvm_pic_state));
3332 		spin_unlock(&pic_irqchip(kvm)->lock);
3333 		break;
3334 	case KVM_IRQCHIP_IOAPIC:
3335 		r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
3336 		break;
3337 	default:
3338 		r = -EINVAL;
3339 		break;
3340 	}
3341 	kvm_pic_update_irq(pic_irqchip(kvm));
3342 	return r;
3343 }
3344 
3345 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3346 {
3347 	int r = 0;
3348 
3349 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
3350 	memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
3351 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3352 	return r;
3353 }
3354 
3355 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3356 {
3357 	int r = 0;
3358 
3359 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
3360 	memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
3361 	kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
3362 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3363 	return r;
3364 }
3365 
3366 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3367 {
3368 	int r = 0;
3369 
3370 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
3371 	memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3372 		sizeof(ps->channels));
3373 	ps->flags = kvm->arch.vpit->pit_state.flags;
3374 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3375 	memset(&ps->reserved, 0, sizeof(ps->reserved));
3376 	return r;
3377 }
3378 
3379 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3380 {
3381 	int r = 0, start = 0;
3382 	u32 prev_legacy, cur_legacy;
3383 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
3384 	prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3385 	cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3386 	if (!prev_legacy && cur_legacy)
3387 		start = 1;
3388 	memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
3389 	       sizeof(kvm->arch.vpit->pit_state.channels));
3390 	kvm->arch.vpit->pit_state.flags = ps->flags;
3391 	kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
3392 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3393 	return r;
3394 }
3395 
3396 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3397 				 struct kvm_reinject_control *control)
3398 {
3399 	if (!kvm->arch.vpit)
3400 		return -ENXIO;
3401 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
3402 	kvm->arch.vpit->pit_state.reinject = control->pit_reinject;
3403 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3404 	return 0;
3405 }
3406 
3407 /**
3408  * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
3409  * @kvm: kvm instance
3410  * @log: slot id and address to which we copy the log
3411  *
3412  * We need to keep it in mind that VCPU threads can write to the bitmap
3413  * concurrently.  So, to avoid losing data, we keep the following order for
3414  * each bit:
3415  *
3416  *   1. Take a snapshot of the bit and clear it if needed.
3417  *   2. Write protect the corresponding page.
3418  *   3. Flush TLB's if needed.
3419  *   4. Copy the snapshot to the userspace.
3420  *
3421  * Between 2 and 3, the guest may write to the page using the remaining TLB
3422  * entry.  This is not a problem because the page will be reported dirty at
3423  * step 4 using the snapshot taken before and step 3 ensures that successive
3424  * writes will be logged for the next call.
3425  */
3426 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
3427 {
3428 	int r;
3429 	struct kvm_memory_slot *memslot;
3430 	unsigned long n, i;
3431 	unsigned long *dirty_bitmap;
3432 	unsigned long *dirty_bitmap_buffer;
3433 	bool is_dirty = false;
3434 
3435 	mutex_lock(&kvm->slots_lock);
3436 
3437 	r = -EINVAL;
3438 	if (log->slot >= KVM_MEMORY_SLOTS)
3439 		goto out;
3440 
3441 	memslot = id_to_memslot(kvm->memslots, log->slot);
3442 
3443 	dirty_bitmap = memslot->dirty_bitmap;
3444 	r = -ENOENT;
3445 	if (!dirty_bitmap)
3446 		goto out;
3447 
3448 	n = kvm_dirty_bitmap_bytes(memslot);
3449 
3450 	dirty_bitmap_buffer = dirty_bitmap + n / sizeof(long);
3451 	memset(dirty_bitmap_buffer, 0, n);
3452 
3453 	spin_lock(&kvm->mmu_lock);
3454 
3455 	for (i = 0; i < n / sizeof(long); i++) {
3456 		unsigned long mask;
3457 		gfn_t offset;
3458 
3459 		if (!dirty_bitmap[i])
3460 			continue;
3461 
3462 		is_dirty = true;
3463 
3464 		mask = xchg(&dirty_bitmap[i], 0);
3465 		dirty_bitmap_buffer[i] = mask;
3466 
3467 		offset = i * BITS_PER_LONG;
3468 		kvm_mmu_write_protect_pt_masked(kvm, memslot, offset, mask);
3469 	}
3470 	if (is_dirty)
3471 		kvm_flush_remote_tlbs(kvm);
3472 
3473 	spin_unlock(&kvm->mmu_lock);
3474 
3475 	r = -EFAULT;
3476 	if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
3477 		goto out;
3478 
3479 	r = 0;
3480 out:
3481 	mutex_unlock(&kvm->slots_lock);
3482 	return r;
3483 }
3484 
3485 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event)
3486 {
3487 	if (!irqchip_in_kernel(kvm))
3488 		return -ENXIO;
3489 
3490 	irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3491 					irq_event->irq, irq_event->level);
3492 	return 0;
3493 }
3494 
3495 long kvm_arch_vm_ioctl(struct file *filp,
3496 		       unsigned int ioctl, unsigned long arg)
3497 {
3498 	struct kvm *kvm = filp->private_data;
3499 	void __user *argp = (void __user *)arg;
3500 	int r = -ENOTTY;
3501 	/*
3502 	 * This union makes it completely explicit to gcc-3.x
3503 	 * that these two variables' stack usage should be
3504 	 * combined, not added together.
3505 	 */
3506 	union {
3507 		struct kvm_pit_state ps;
3508 		struct kvm_pit_state2 ps2;
3509 		struct kvm_pit_config pit_config;
3510 	} u;
3511 
3512 	switch (ioctl) {
3513 	case KVM_SET_TSS_ADDR:
3514 		r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
3515 		break;
3516 	case KVM_SET_IDENTITY_MAP_ADDR: {
3517 		u64 ident_addr;
3518 
3519 		r = -EFAULT;
3520 		if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3521 			goto out;
3522 		r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
3523 		break;
3524 	}
3525 	case KVM_SET_NR_MMU_PAGES:
3526 		r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
3527 		break;
3528 	case KVM_GET_NR_MMU_PAGES:
3529 		r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3530 		break;
3531 	case KVM_CREATE_IRQCHIP: {
3532 		struct kvm_pic *vpic;
3533 
3534 		mutex_lock(&kvm->lock);
3535 		r = -EEXIST;
3536 		if (kvm->arch.vpic)
3537 			goto create_irqchip_unlock;
3538 		r = -EINVAL;
3539 		if (atomic_read(&kvm->online_vcpus))
3540 			goto create_irqchip_unlock;
3541 		r = -ENOMEM;
3542 		vpic = kvm_create_pic(kvm);
3543 		if (vpic) {
3544 			r = kvm_ioapic_init(kvm);
3545 			if (r) {
3546 				mutex_lock(&kvm->slots_lock);
3547 				kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3548 							  &vpic->dev_master);
3549 				kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3550 							  &vpic->dev_slave);
3551 				kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3552 							  &vpic->dev_eclr);
3553 				mutex_unlock(&kvm->slots_lock);
3554 				kfree(vpic);
3555 				goto create_irqchip_unlock;
3556 			}
3557 		} else
3558 			goto create_irqchip_unlock;
3559 		smp_wmb();
3560 		kvm->arch.vpic = vpic;
3561 		smp_wmb();
3562 		r = kvm_setup_default_irq_routing(kvm);
3563 		if (r) {
3564 			mutex_lock(&kvm->slots_lock);
3565 			mutex_lock(&kvm->irq_lock);
3566 			kvm_ioapic_destroy(kvm);
3567 			kvm_destroy_pic(kvm);
3568 			mutex_unlock(&kvm->irq_lock);
3569 			mutex_unlock(&kvm->slots_lock);
3570 		}
3571 	create_irqchip_unlock:
3572 		mutex_unlock(&kvm->lock);
3573 		break;
3574 	}
3575 	case KVM_CREATE_PIT:
3576 		u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
3577 		goto create_pit;
3578 	case KVM_CREATE_PIT2:
3579 		r = -EFAULT;
3580 		if (copy_from_user(&u.pit_config, argp,
3581 				   sizeof(struct kvm_pit_config)))
3582 			goto out;
3583 	create_pit:
3584 		mutex_lock(&kvm->slots_lock);
3585 		r = -EEXIST;
3586 		if (kvm->arch.vpit)
3587 			goto create_pit_unlock;
3588 		r = -ENOMEM;
3589 		kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
3590 		if (kvm->arch.vpit)
3591 			r = 0;
3592 	create_pit_unlock:
3593 		mutex_unlock(&kvm->slots_lock);
3594 		break;
3595 	case KVM_GET_IRQCHIP: {
3596 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3597 		struct kvm_irqchip *chip;
3598 
3599 		chip = memdup_user(argp, sizeof(*chip));
3600 		if (IS_ERR(chip)) {
3601 			r = PTR_ERR(chip);
3602 			goto out;
3603 		}
3604 
3605 		r = -ENXIO;
3606 		if (!irqchip_in_kernel(kvm))
3607 			goto get_irqchip_out;
3608 		r = kvm_vm_ioctl_get_irqchip(kvm, chip);
3609 		if (r)
3610 			goto get_irqchip_out;
3611 		r = -EFAULT;
3612 		if (copy_to_user(argp, chip, sizeof *chip))
3613 			goto get_irqchip_out;
3614 		r = 0;
3615 	get_irqchip_out:
3616 		kfree(chip);
3617 		break;
3618 	}
3619 	case KVM_SET_IRQCHIP: {
3620 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3621 		struct kvm_irqchip *chip;
3622 
3623 		chip = memdup_user(argp, sizeof(*chip));
3624 		if (IS_ERR(chip)) {
3625 			r = PTR_ERR(chip);
3626 			goto out;
3627 		}
3628 
3629 		r = -ENXIO;
3630 		if (!irqchip_in_kernel(kvm))
3631 			goto set_irqchip_out;
3632 		r = kvm_vm_ioctl_set_irqchip(kvm, chip);
3633 		if (r)
3634 			goto set_irqchip_out;
3635 		r = 0;
3636 	set_irqchip_out:
3637 		kfree(chip);
3638 		break;
3639 	}
3640 	case KVM_GET_PIT: {
3641 		r = -EFAULT;
3642 		if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
3643 			goto out;
3644 		r = -ENXIO;
3645 		if (!kvm->arch.vpit)
3646 			goto out;
3647 		r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
3648 		if (r)
3649 			goto out;
3650 		r = -EFAULT;
3651 		if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
3652 			goto out;
3653 		r = 0;
3654 		break;
3655 	}
3656 	case KVM_SET_PIT: {
3657 		r = -EFAULT;
3658 		if (copy_from_user(&u.ps, argp, sizeof u.ps))
3659 			goto out;
3660 		r = -ENXIO;
3661 		if (!kvm->arch.vpit)
3662 			goto out;
3663 		r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
3664 		break;
3665 	}
3666 	case KVM_GET_PIT2: {
3667 		r = -ENXIO;
3668 		if (!kvm->arch.vpit)
3669 			goto out;
3670 		r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
3671 		if (r)
3672 			goto out;
3673 		r = -EFAULT;
3674 		if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
3675 			goto out;
3676 		r = 0;
3677 		break;
3678 	}
3679 	case KVM_SET_PIT2: {
3680 		r = -EFAULT;
3681 		if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
3682 			goto out;
3683 		r = -ENXIO;
3684 		if (!kvm->arch.vpit)
3685 			goto out;
3686 		r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
3687 		break;
3688 	}
3689 	case KVM_REINJECT_CONTROL: {
3690 		struct kvm_reinject_control control;
3691 		r =  -EFAULT;
3692 		if (copy_from_user(&control, argp, sizeof(control)))
3693 			goto out;
3694 		r = kvm_vm_ioctl_reinject(kvm, &control);
3695 		break;
3696 	}
3697 	case KVM_XEN_HVM_CONFIG: {
3698 		r = -EFAULT;
3699 		if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
3700 				   sizeof(struct kvm_xen_hvm_config)))
3701 			goto out;
3702 		r = -EINVAL;
3703 		if (kvm->arch.xen_hvm_config.flags)
3704 			goto out;
3705 		r = 0;
3706 		break;
3707 	}
3708 	case KVM_SET_CLOCK: {
3709 		struct kvm_clock_data user_ns;
3710 		u64 now_ns;
3711 		s64 delta;
3712 
3713 		r = -EFAULT;
3714 		if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
3715 			goto out;
3716 
3717 		r = -EINVAL;
3718 		if (user_ns.flags)
3719 			goto out;
3720 
3721 		r = 0;
3722 		local_irq_disable();
3723 		now_ns = get_kernel_ns();
3724 		delta = user_ns.clock - now_ns;
3725 		local_irq_enable();
3726 		kvm->arch.kvmclock_offset = delta;
3727 		break;
3728 	}
3729 	case KVM_GET_CLOCK: {
3730 		struct kvm_clock_data user_ns;
3731 		u64 now_ns;
3732 
3733 		local_irq_disable();
3734 		now_ns = get_kernel_ns();
3735 		user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
3736 		local_irq_enable();
3737 		user_ns.flags = 0;
3738 		memset(&user_ns.pad, 0, sizeof(user_ns.pad));
3739 
3740 		r = -EFAULT;
3741 		if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
3742 			goto out;
3743 		r = 0;
3744 		break;
3745 	}
3746 
3747 	default:
3748 		;
3749 	}
3750 out:
3751 	return r;
3752 }
3753 
3754 static void kvm_init_msr_list(void)
3755 {
3756 	u32 dummy[2];
3757 	unsigned i, j;
3758 
3759 	/* skip the first msrs in the list. KVM-specific */
3760 	for (i = j = KVM_SAVE_MSRS_BEGIN; i < ARRAY_SIZE(msrs_to_save); i++) {
3761 		if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
3762 			continue;
3763 		if (j < i)
3764 			msrs_to_save[j] = msrs_to_save[i];
3765 		j++;
3766 	}
3767 	num_msrs_to_save = j;
3768 }
3769 
3770 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
3771 			   const void *v)
3772 {
3773 	int handled = 0;
3774 	int n;
3775 
3776 	do {
3777 		n = min(len, 8);
3778 		if (!(vcpu->arch.apic &&
3779 		      !kvm_iodevice_write(&vcpu->arch.apic->dev, addr, n, v))
3780 		    && kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
3781 			break;
3782 		handled += n;
3783 		addr += n;
3784 		len -= n;
3785 		v += n;
3786 	} while (len);
3787 
3788 	return handled;
3789 }
3790 
3791 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
3792 {
3793 	int handled = 0;
3794 	int n;
3795 
3796 	do {
3797 		n = min(len, 8);
3798 		if (!(vcpu->arch.apic &&
3799 		      !kvm_iodevice_read(&vcpu->arch.apic->dev, addr, n, v))
3800 		    && kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
3801 			break;
3802 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
3803 		handled += n;
3804 		addr += n;
3805 		len -= n;
3806 		v += n;
3807 	} while (len);
3808 
3809 	return handled;
3810 }
3811 
3812 static void kvm_set_segment(struct kvm_vcpu *vcpu,
3813 			struct kvm_segment *var, int seg)
3814 {
3815 	kvm_x86_ops->set_segment(vcpu, var, seg);
3816 }
3817 
3818 void kvm_get_segment(struct kvm_vcpu *vcpu,
3819 		     struct kvm_segment *var, int seg)
3820 {
3821 	kvm_x86_ops->get_segment(vcpu, var, seg);
3822 }
3823 
3824 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access)
3825 {
3826 	gpa_t t_gpa;
3827 	struct x86_exception exception;
3828 
3829 	BUG_ON(!mmu_is_nested(vcpu));
3830 
3831 	/* NPT walks are always user-walks */
3832 	access |= PFERR_USER_MASK;
3833 	t_gpa  = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, &exception);
3834 
3835 	return t_gpa;
3836 }
3837 
3838 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
3839 			      struct x86_exception *exception)
3840 {
3841 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3842 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3843 }
3844 
3845  gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
3846 				struct x86_exception *exception)
3847 {
3848 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3849 	access |= PFERR_FETCH_MASK;
3850 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3851 }
3852 
3853 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
3854 			       struct x86_exception *exception)
3855 {
3856 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3857 	access |= PFERR_WRITE_MASK;
3858 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3859 }
3860 
3861 /* uses this to access any guest's mapped memory without checking CPL */
3862 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
3863 				struct x86_exception *exception)
3864 {
3865 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
3866 }
3867 
3868 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
3869 				      struct kvm_vcpu *vcpu, u32 access,
3870 				      struct x86_exception *exception)
3871 {
3872 	void *data = val;
3873 	int r = X86EMUL_CONTINUE;
3874 
3875 	while (bytes) {
3876 		gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
3877 							    exception);
3878 		unsigned offset = addr & (PAGE_SIZE-1);
3879 		unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
3880 		int ret;
3881 
3882 		if (gpa == UNMAPPED_GVA)
3883 			return X86EMUL_PROPAGATE_FAULT;
3884 		ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
3885 		if (ret < 0) {
3886 			r = X86EMUL_IO_NEEDED;
3887 			goto out;
3888 		}
3889 
3890 		bytes -= toread;
3891 		data += toread;
3892 		addr += toread;
3893 	}
3894 out:
3895 	return r;
3896 }
3897 
3898 /* used for instruction fetching */
3899 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
3900 				gva_t addr, void *val, unsigned int bytes,
3901 				struct x86_exception *exception)
3902 {
3903 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3904 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3905 
3906 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu,
3907 					  access | PFERR_FETCH_MASK,
3908 					  exception);
3909 }
3910 
3911 int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
3912 			       gva_t addr, void *val, unsigned int bytes,
3913 			       struct x86_exception *exception)
3914 {
3915 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3916 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3917 
3918 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
3919 					  exception);
3920 }
3921 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
3922 
3923 static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
3924 				      gva_t addr, void *val, unsigned int bytes,
3925 				      struct x86_exception *exception)
3926 {
3927 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3928 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
3929 }
3930 
3931 int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
3932 				       gva_t addr, void *val,
3933 				       unsigned int bytes,
3934 				       struct x86_exception *exception)
3935 {
3936 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3937 	void *data = val;
3938 	int r = X86EMUL_CONTINUE;
3939 
3940 	while (bytes) {
3941 		gpa_t gpa =  vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
3942 							     PFERR_WRITE_MASK,
3943 							     exception);
3944 		unsigned offset = addr & (PAGE_SIZE-1);
3945 		unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
3946 		int ret;
3947 
3948 		if (gpa == UNMAPPED_GVA)
3949 			return X86EMUL_PROPAGATE_FAULT;
3950 		ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
3951 		if (ret < 0) {
3952 			r = X86EMUL_IO_NEEDED;
3953 			goto out;
3954 		}
3955 
3956 		bytes -= towrite;
3957 		data += towrite;
3958 		addr += towrite;
3959 	}
3960 out:
3961 	return r;
3962 }
3963 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
3964 
3965 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
3966 				gpa_t *gpa, struct x86_exception *exception,
3967 				bool write)
3968 {
3969 	u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
3970 		| (write ? PFERR_WRITE_MASK : 0);
3971 
3972 	if (vcpu_match_mmio_gva(vcpu, gva)
3973 	    && !permission_fault(vcpu->arch.walk_mmu, vcpu->arch.access, access)) {
3974 		*gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
3975 					(gva & (PAGE_SIZE - 1));
3976 		trace_vcpu_match_mmio(gva, *gpa, write, false);
3977 		return 1;
3978 	}
3979 
3980 	*gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3981 
3982 	if (*gpa == UNMAPPED_GVA)
3983 		return -1;
3984 
3985 	/* For APIC access vmexit */
3986 	if ((*gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3987 		return 1;
3988 
3989 	if (vcpu_match_mmio_gpa(vcpu, *gpa)) {
3990 		trace_vcpu_match_mmio(gva, *gpa, write, true);
3991 		return 1;
3992 	}
3993 
3994 	return 0;
3995 }
3996 
3997 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
3998 			const void *val, int bytes)
3999 {
4000 	int ret;
4001 
4002 	ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
4003 	if (ret < 0)
4004 		return 0;
4005 	kvm_mmu_pte_write(vcpu, gpa, val, bytes);
4006 	return 1;
4007 }
4008 
4009 struct read_write_emulator_ops {
4010 	int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
4011 				  int bytes);
4012 	int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
4013 				  void *val, int bytes);
4014 	int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4015 			       int bytes, void *val);
4016 	int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4017 				    void *val, int bytes);
4018 	bool write;
4019 };
4020 
4021 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
4022 {
4023 	if (vcpu->mmio_read_completed) {
4024 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
4025 			       vcpu->mmio_fragments[0].gpa, *(u64 *)val);
4026 		vcpu->mmio_read_completed = 0;
4027 		return 1;
4028 	}
4029 
4030 	return 0;
4031 }
4032 
4033 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4034 			void *val, int bytes)
4035 {
4036 	return !kvm_read_guest(vcpu->kvm, gpa, val, bytes);
4037 }
4038 
4039 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4040 			 void *val, int bytes)
4041 {
4042 	return emulator_write_phys(vcpu, gpa, val, bytes);
4043 }
4044 
4045 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
4046 {
4047 	trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
4048 	return vcpu_mmio_write(vcpu, gpa, bytes, val);
4049 }
4050 
4051 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4052 			  void *val, int bytes)
4053 {
4054 	trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
4055 	return X86EMUL_IO_NEEDED;
4056 }
4057 
4058 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4059 			   void *val, int bytes)
4060 {
4061 	struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
4062 
4063 	memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
4064 	return X86EMUL_CONTINUE;
4065 }
4066 
4067 static const struct read_write_emulator_ops read_emultor = {
4068 	.read_write_prepare = read_prepare,
4069 	.read_write_emulate = read_emulate,
4070 	.read_write_mmio = vcpu_mmio_read,
4071 	.read_write_exit_mmio = read_exit_mmio,
4072 };
4073 
4074 static const struct read_write_emulator_ops write_emultor = {
4075 	.read_write_emulate = write_emulate,
4076 	.read_write_mmio = write_mmio,
4077 	.read_write_exit_mmio = write_exit_mmio,
4078 	.write = true,
4079 };
4080 
4081 static int emulator_read_write_onepage(unsigned long addr, void *val,
4082 				       unsigned int bytes,
4083 				       struct x86_exception *exception,
4084 				       struct kvm_vcpu *vcpu,
4085 				       const struct read_write_emulator_ops *ops)
4086 {
4087 	gpa_t gpa;
4088 	int handled, ret;
4089 	bool write = ops->write;
4090 	struct kvm_mmio_fragment *frag;
4091 
4092 	ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
4093 
4094 	if (ret < 0)
4095 		return X86EMUL_PROPAGATE_FAULT;
4096 
4097 	/* For APIC access vmexit */
4098 	if (ret)
4099 		goto mmio;
4100 
4101 	if (ops->read_write_emulate(vcpu, gpa, val, bytes))
4102 		return X86EMUL_CONTINUE;
4103 
4104 mmio:
4105 	/*
4106 	 * Is this MMIO handled locally?
4107 	 */
4108 	handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
4109 	if (handled == bytes)
4110 		return X86EMUL_CONTINUE;
4111 
4112 	gpa += handled;
4113 	bytes -= handled;
4114 	val += handled;
4115 
4116 	WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
4117 	frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
4118 	frag->gpa = gpa;
4119 	frag->data = val;
4120 	frag->len = bytes;
4121 	return X86EMUL_CONTINUE;
4122 }
4123 
4124 int emulator_read_write(struct x86_emulate_ctxt *ctxt, unsigned long addr,
4125 			void *val, unsigned int bytes,
4126 			struct x86_exception *exception,
4127 			const struct read_write_emulator_ops *ops)
4128 {
4129 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4130 	gpa_t gpa;
4131 	int rc;
4132 
4133 	if (ops->read_write_prepare &&
4134 		  ops->read_write_prepare(vcpu, val, bytes))
4135 		return X86EMUL_CONTINUE;
4136 
4137 	vcpu->mmio_nr_fragments = 0;
4138 
4139 	/* Crossing a page boundary? */
4140 	if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
4141 		int now;
4142 
4143 		now = -addr & ~PAGE_MASK;
4144 		rc = emulator_read_write_onepage(addr, val, now, exception,
4145 						 vcpu, ops);
4146 
4147 		if (rc != X86EMUL_CONTINUE)
4148 			return rc;
4149 		addr += now;
4150 		val += now;
4151 		bytes -= now;
4152 	}
4153 
4154 	rc = emulator_read_write_onepage(addr, val, bytes, exception,
4155 					 vcpu, ops);
4156 	if (rc != X86EMUL_CONTINUE)
4157 		return rc;
4158 
4159 	if (!vcpu->mmio_nr_fragments)
4160 		return rc;
4161 
4162 	gpa = vcpu->mmio_fragments[0].gpa;
4163 
4164 	vcpu->mmio_needed = 1;
4165 	vcpu->mmio_cur_fragment = 0;
4166 
4167 	vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
4168 	vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
4169 	vcpu->run->exit_reason = KVM_EXIT_MMIO;
4170 	vcpu->run->mmio.phys_addr = gpa;
4171 
4172 	return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
4173 }
4174 
4175 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
4176 				  unsigned long addr,
4177 				  void *val,
4178 				  unsigned int bytes,
4179 				  struct x86_exception *exception)
4180 {
4181 	return emulator_read_write(ctxt, addr, val, bytes,
4182 				   exception, &read_emultor);
4183 }
4184 
4185 int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
4186 			    unsigned long addr,
4187 			    const void *val,
4188 			    unsigned int bytes,
4189 			    struct x86_exception *exception)
4190 {
4191 	return emulator_read_write(ctxt, addr, (void *)val, bytes,
4192 				   exception, &write_emultor);
4193 }
4194 
4195 #define CMPXCHG_TYPE(t, ptr, old, new) \
4196 	(cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4197 
4198 #ifdef CONFIG_X86_64
4199 #  define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4200 #else
4201 #  define CMPXCHG64(ptr, old, new) \
4202 	(cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
4203 #endif
4204 
4205 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
4206 				     unsigned long addr,
4207 				     const void *old,
4208 				     const void *new,
4209 				     unsigned int bytes,
4210 				     struct x86_exception *exception)
4211 {
4212 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4213 	gpa_t gpa;
4214 	struct page *page;
4215 	char *kaddr;
4216 	bool exchanged;
4217 
4218 	/* guests cmpxchg8b have to be emulated atomically */
4219 	if (bytes > 8 || (bytes & (bytes - 1)))
4220 		goto emul_write;
4221 
4222 	gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
4223 
4224 	if (gpa == UNMAPPED_GVA ||
4225 	    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4226 		goto emul_write;
4227 
4228 	if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
4229 		goto emul_write;
4230 
4231 	page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
4232 	if (is_error_page(page))
4233 		goto emul_write;
4234 
4235 	kaddr = kmap_atomic(page);
4236 	kaddr += offset_in_page(gpa);
4237 	switch (bytes) {
4238 	case 1:
4239 		exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
4240 		break;
4241 	case 2:
4242 		exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
4243 		break;
4244 	case 4:
4245 		exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
4246 		break;
4247 	case 8:
4248 		exchanged = CMPXCHG64(kaddr, old, new);
4249 		break;
4250 	default:
4251 		BUG();
4252 	}
4253 	kunmap_atomic(kaddr);
4254 	kvm_release_page_dirty(page);
4255 
4256 	if (!exchanged)
4257 		return X86EMUL_CMPXCHG_FAILED;
4258 
4259 	kvm_mmu_pte_write(vcpu, gpa, new, bytes);
4260 
4261 	return X86EMUL_CONTINUE;
4262 
4263 emul_write:
4264 	printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
4265 
4266 	return emulator_write_emulated(ctxt, addr, new, bytes, exception);
4267 }
4268 
4269 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4270 {
4271 	/* TODO: String I/O for in kernel device */
4272 	int r;
4273 
4274 	if (vcpu->arch.pio.in)
4275 		r = kvm_io_bus_read(vcpu->kvm, KVM_PIO_BUS, vcpu->arch.pio.port,
4276 				    vcpu->arch.pio.size, pd);
4277 	else
4278 		r = kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS,
4279 				     vcpu->arch.pio.port, vcpu->arch.pio.size,
4280 				     pd);
4281 	return r;
4282 }
4283 
4284 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
4285 			       unsigned short port, void *val,
4286 			       unsigned int count, bool in)
4287 {
4288 	trace_kvm_pio(!in, port, size, count);
4289 
4290 	vcpu->arch.pio.port = port;
4291 	vcpu->arch.pio.in = in;
4292 	vcpu->arch.pio.count  = count;
4293 	vcpu->arch.pio.size = size;
4294 
4295 	if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4296 		vcpu->arch.pio.count = 0;
4297 		return 1;
4298 	}
4299 
4300 	vcpu->run->exit_reason = KVM_EXIT_IO;
4301 	vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
4302 	vcpu->run->io.size = size;
4303 	vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4304 	vcpu->run->io.count = count;
4305 	vcpu->run->io.port = port;
4306 
4307 	return 0;
4308 }
4309 
4310 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
4311 				    int size, unsigned short port, void *val,
4312 				    unsigned int count)
4313 {
4314 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4315 	int ret;
4316 
4317 	if (vcpu->arch.pio.count)
4318 		goto data_avail;
4319 
4320 	ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
4321 	if (ret) {
4322 data_avail:
4323 		memcpy(val, vcpu->arch.pio_data, size * count);
4324 		vcpu->arch.pio.count = 0;
4325 		return 1;
4326 	}
4327 
4328 	return 0;
4329 }
4330 
4331 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
4332 				     int size, unsigned short port,
4333 				     const void *val, unsigned int count)
4334 {
4335 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4336 
4337 	memcpy(vcpu->arch.pio_data, val, size * count);
4338 	return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
4339 }
4340 
4341 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4342 {
4343 	return kvm_x86_ops->get_segment_base(vcpu, seg);
4344 }
4345 
4346 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
4347 {
4348 	kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
4349 }
4350 
4351 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4352 {
4353 	if (!need_emulate_wbinvd(vcpu))
4354 		return X86EMUL_CONTINUE;
4355 
4356 	if (kvm_x86_ops->has_wbinvd_exit()) {
4357 		int cpu = get_cpu();
4358 
4359 		cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4360 		smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4361 				wbinvd_ipi, NULL, 1);
4362 		put_cpu();
4363 		cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
4364 	} else
4365 		wbinvd();
4366 	return X86EMUL_CONTINUE;
4367 }
4368 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4369 
4370 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
4371 {
4372 	kvm_emulate_wbinvd(emul_to_vcpu(ctxt));
4373 }
4374 
4375 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
4376 {
4377 	return _kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
4378 }
4379 
4380 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
4381 {
4382 
4383 	return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
4384 }
4385 
4386 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4387 {
4388 	return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4389 }
4390 
4391 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
4392 {
4393 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4394 	unsigned long value;
4395 
4396 	switch (cr) {
4397 	case 0:
4398 		value = kvm_read_cr0(vcpu);
4399 		break;
4400 	case 2:
4401 		value = vcpu->arch.cr2;
4402 		break;
4403 	case 3:
4404 		value = kvm_read_cr3(vcpu);
4405 		break;
4406 	case 4:
4407 		value = kvm_read_cr4(vcpu);
4408 		break;
4409 	case 8:
4410 		value = kvm_get_cr8(vcpu);
4411 		break;
4412 	default:
4413 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
4414 		return 0;
4415 	}
4416 
4417 	return value;
4418 }
4419 
4420 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
4421 {
4422 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4423 	int res = 0;
4424 
4425 	switch (cr) {
4426 	case 0:
4427 		res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
4428 		break;
4429 	case 2:
4430 		vcpu->arch.cr2 = val;
4431 		break;
4432 	case 3:
4433 		res = kvm_set_cr3(vcpu, val);
4434 		break;
4435 	case 4:
4436 		res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
4437 		break;
4438 	case 8:
4439 		res = kvm_set_cr8(vcpu, val);
4440 		break;
4441 	default:
4442 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
4443 		res = -1;
4444 	}
4445 
4446 	return res;
4447 }
4448 
4449 static void emulator_set_rflags(struct x86_emulate_ctxt *ctxt, ulong val)
4450 {
4451 	kvm_set_rflags(emul_to_vcpu(ctxt), val);
4452 }
4453 
4454 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
4455 {
4456 	return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
4457 }
4458 
4459 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4460 {
4461 	kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
4462 }
4463 
4464 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4465 {
4466 	kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
4467 }
4468 
4469 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4470 {
4471 	kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
4472 }
4473 
4474 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4475 {
4476 	kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
4477 }
4478 
4479 static unsigned long emulator_get_cached_segment_base(
4480 	struct x86_emulate_ctxt *ctxt, int seg)
4481 {
4482 	return get_segment_base(emul_to_vcpu(ctxt), seg);
4483 }
4484 
4485 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
4486 				 struct desc_struct *desc, u32 *base3,
4487 				 int seg)
4488 {
4489 	struct kvm_segment var;
4490 
4491 	kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
4492 	*selector = var.selector;
4493 
4494 	if (var.unusable)
4495 		return false;
4496 
4497 	if (var.g)
4498 		var.limit >>= 12;
4499 	set_desc_limit(desc, var.limit);
4500 	set_desc_base(desc, (unsigned long)var.base);
4501 #ifdef CONFIG_X86_64
4502 	if (base3)
4503 		*base3 = var.base >> 32;
4504 #endif
4505 	desc->type = var.type;
4506 	desc->s = var.s;
4507 	desc->dpl = var.dpl;
4508 	desc->p = var.present;
4509 	desc->avl = var.avl;
4510 	desc->l = var.l;
4511 	desc->d = var.db;
4512 	desc->g = var.g;
4513 
4514 	return true;
4515 }
4516 
4517 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
4518 				 struct desc_struct *desc, u32 base3,
4519 				 int seg)
4520 {
4521 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4522 	struct kvm_segment var;
4523 
4524 	var.selector = selector;
4525 	var.base = get_desc_base(desc);
4526 #ifdef CONFIG_X86_64
4527 	var.base |= ((u64)base3) << 32;
4528 #endif
4529 	var.limit = get_desc_limit(desc);
4530 	if (desc->g)
4531 		var.limit = (var.limit << 12) | 0xfff;
4532 	var.type = desc->type;
4533 	var.present = desc->p;
4534 	var.dpl = desc->dpl;
4535 	var.db = desc->d;
4536 	var.s = desc->s;
4537 	var.l = desc->l;
4538 	var.g = desc->g;
4539 	var.avl = desc->avl;
4540 	var.present = desc->p;
4541 	var.unusable = !var.present;
4542 	var.padding = 0;
4543 
4544 	kvm_set_segment(vcpu, &var, seg);
4545 	return;
4546 }
4547 
4548 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
4549 			    u32 msr_index, u64 *pdata)
4550 {
4551 	return kvm_get_msr(emul_to_vcpu(ctxt), msr_index, pdata);
4552 }
4553 
4554 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
4555 			    u32 msr_index, u64 data)
4556 {
4557 	struct msr_data msr;
4558 
4559 	msr.data = data;
4560 	msr.index = msr_index;
4561 	msr.host_initiated = false;
4562 	return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
4563 }
4564 
4565 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
4566 			     u32 pmc, u64 *pdata)
4567 {
4568 	return kvm_pmu_read_pmc(emul_to_vcpu(ctxt), pmc, pdata);
4569 }
4570 
4571 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
4572 {
4573 	emul_to_vcpu(ctxt)->arch.halt_request = 1;
4574 }
4575 
4576 static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
4577 {
4578 	preempt_disable();
4579 	kvm_load_guest_fpu(emul_to_vcpu(ctxt));
4580 	/*
4581 	 * CR0.TS may reference the host fpu state, not the guest fpu state,
4582 	 * so it may be clear at this point.
4583 	 */
4584 	clts();
4585 }
4586 
4587 static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
4588 {
4589 	preempt_enable();
4590 }
4591 
4592 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
4593 			      struct x86_instruction_info *info,
4594 			      enum x86_intercept_stage stage)
4595 {
4596 	return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
4597 }
4598 
4599 static void emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
4600 			       u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
4601 {
4602 	kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx);
4603 }
4604 
4605 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
4606 {
4607 	return kvm_register_read(emul_to_vcpu(ctxt), reg);
4608 }
4609 
4610 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
4611 {
4612 	kvm_register_write(emul_to_vcpu(ctxt), reg, val);
4613 }
4614 
4615 static const struct x86_emulate_ops emulate_ops = {
4616 	.read_gpr            = emulator_read_gpr,
4617 	.write_gpr           = emulator_write_gpr,
4618 	.read_std            = kvm_read_guest_virt_system,
4619 	.write_std           = kvm_write_guest_virt_system,
4620 	.fetch               = kvm_fetch_guest_virt,
4621 	.read_emulated       = emulator_read_emulated,
4622 	.write_emulated      = emulator_write_emulated,
4623 	.cmpxchg_emulated    = emulator_cmpxchg_emulated,
4624 	.invlpg              = emulator_invlpg,
4625 	.pio_in_emulated     = emulator_pio_in_emulated,
4626 	.pio_out_emulated    = emulator_pio_out_emulated,
4627 	.get_segment         = emulator_get_segment,
4628 	.set_segment         = emulator_set_segment,
4629 	.get_cached_segment_base = emulator_get_cached_segment_base,
4630 	.get_gdt             = emulator_get_gdt,
4631 	.get_idt	     = emulator_get_idt,
4632 	.set_gdt             = emulator_set_gdt,
4633 	.set_idt	     = emulator_set_idt,
4634 	.get_cr              = emulator_get_cr,
4635 	.set_cr              = emulator_set_cr,
4636 	.set_rflags          = emulator_set_rflags,
4637 	.cpl                 = emulator_get_cpl,
4638 	.get_dr              = emulator_get_dr,
4639 	.set_dr              = emulator_set_dr,
4640 	.set_msr             = emulator_set_msr,
4641 	.get_msr             = emulator_get_msr,
4642 	.read_pmc            = emulator_read_pmc,
4643 	.halt                = emulator_halt,
4644 	.wbinvd              = emulator_wbinvd,
4645 	.fix_hypercall       = emulator_fix_hypercall,
4646 	.get_fpu             = emulator_get_fpu,
4647 	.put_fpu             = emulator_put_fpu,
4648 	.intercept           = emulator_intercept,
4649 	.get_cpuid           = emulator_get_cpuid,
4650 };
4651 
4652 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
4653 {
4654 	u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu, mask);
4655 	/*
4656 	 * an sti; sti; sequence only disable interrupts for the first
4657 	 * instruction. So, if the last instruction, be it emulated or
4658 	 * not, left the system with the INT_STI flag enabled, it
4659 	 * means that the last instruction is an sti. We should not
4660 	 * leave the flag on in this case. The same goes for mov ss
4661 	 */
4662 	if (!(int_shadow & mask))
4663 		kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
4664 }
4665 
4666 static void inject_emulated_exception(struct kvm_vcpu *vcpu)
4667 {
4668 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4669 	if (ctxt->exception.vector == PF_VECTOR)
4670 		kvm_propagate_fault(vcpu, &ctxt->exception);
4671 	else if (ctxt->exception.error_code_valid)
4672 		kvm_queue_exception_e(vcpu, ctxt->exception.vector,
4673 				      ctxt->exception.error_code);
4674 	else
4675 		kvm_queue_exception(vcpu, ctxt->exception.vector);
4676 }
4677 
4678 static void init_decode_cache(struct x86_emulate_ctxt *ctxt)
4679 {
4680 	memset(&ctxt->twobyte, 0,
4681 	       (void *)&ctxt->_regs - (void *)&ctxt->twobyte);
4682 
4683 	ctxt->fetch.start = 0;
4684 	ctxt->fetch.end = 0;
4685 	ctxt->io_read.pos = 0;
4686 	ctxt->io_read.end = 0;
4687 	ctxt->mem_read.pos = 0;
4688 	ctxt->mem_read.end = 0;
4689 }
4690 
4691 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
4692 {
4693 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4694 	int cs_db, cs_l;
4695 
4696 	kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4697 
4698 	ctxt->eflags = kvm_get_rflags(vcpu);
4699 	ctxt->eip = kvm_rip_read(vcpu);
4700 	ctxt->mode = (!is_protmode(vcpu))		? X86EMUL_MODE_REAL :
4701 		     (ctxt->eflags & X86_EFLAGS_VM)	? X86EMUL_MODE_VM86 :
4702 		     cs_l				? X86EMUL_MODE_PROT64 :
4703 		     cs_db				? X86EMUL_MODE_PROT32 :
4704 							  X86EMUL_MODE_PROT16;
4705 	ctxt->guest_mode = is_guest_mode(vcpu);
4706 
4707 	init_decode_cache(ctxt);
4708 	vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
4709 }
4710 
4711 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
4712 {
4713 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4714 	int ret;
4715 
4716 	init_emulate_ctxt(vcpu);
4717 
4718 	ctxt->op_bytes = 2;
4719 	ctxt->ad_bytes = 2;
4720 	ctxt->_eip = ctxt->eip + inc_eip;
4721 	ret = emulate_int_real(ctxt, irq);
4722 
4723 	if (ret != X86EMUL_CONTINUE)
4724 		return EMULATE_FAIL;
4725 
4726 	ctxt->eip = ctxt->_eip;
4727 	kvm_rip_write(vcpu, ctxt->eip);
4728 	kvm_set_rflags(vcpu, ctxt->eflags);
4729 
4730 	if (irq == NMI_VECTOR)
4731 		vcpu->arch.nmi_pending = 0;
4732 	else
4733 		vcpu->arch.interrupt.pending = false;
4734 
4735 	return EMULATE_DONE;
4736 }
4737 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
4738 
4739 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
4740 {
4741 	int r = EMULATE_DONE;
4742 
4743 	++vcpu->stat.insn_emulation_fail;
4744 	trace_kvm_emulate_insn_failed(vcpu);
4745 	if (!is_guest_mode(vcpu)) {
4746 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4747 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
4748 		vcpu->run->internal.ndata = 0;
4749 		r = EMULATE_FAIL;
4750 	}
4751 	kvm_queue_exception(vcpu, UD_VECTOR);
4752 
4753 	return r;
4754 }
4755 
4756 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t gva)
4757 {
4758 	gpa_t gpa;
4759 	pfn_t pfn;
4760 
4761 	if (tdp_enabled)
4762 		return false;
4763 
4764 	/*
4765 	 * if emulation was due to access to shadowed page table
4766 	 * and it failed try to unshadow page and re-enter the
4767 	 * guest to let CPU execute the instruction.
4768 	 */
4769 	if (kvm_mmu_unprotect_page_virt(vcpu, gva))
4770 		return true;
4771 
4772 	gpa = kvm_mmu_gva_to_gpa_system(vcpu, gva, NULL);
4773 
4774 	if (gpa == UNMAPPED_GVA)
4775 		return true; /* let cpu generate fault */
4776 
4777 	/*
4778 	 * Do not retry the unhandleable instruction if it faults on the
4779 	 * readonly host memory, otherwise it will goto a infinite loop:
4780 	 * retry instruction -> write #PF -> emulation fail -> retry
4781 	 * instruction -> ...
4782 	 */
4783 	pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
4784 	if (!is_error_noslot_pfn(pfn)) {
4785 		kvm_release_pfn_clean(pfn);
4786 		return true;
4787 	}
4788 
4789 	return false;
4790 }
4791 
4792 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
4793 			      unsigned long cr2,  int emulation_type)
4794 {
4795 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4796 	unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
4797 
4798 	last_retry_eip = vcpu->arch.last_retry_eip;
4799 	last_retry_addr = vcpu->arch.last_retry_addr;
4800 
4801 	/*
4802 	 * If the emulation is caused by #PF and it is non-page_table
4803 	 * writing instruction, it means the VM-EXIT is caused by shadow
4804 	 * page protected, we can zap the shadow page and retry this
4805 	 * instruction directly.
4806 	 *
4807 	 * Note: if the guest uses a non-page-table modifying instruction
4808 	 * on the PDE that points to the instruction, then we will unmap
4809 	 * the instruction and go to an infinite loop. So, we cache the
4810 	 * last retried eip and the last fault address, if we meet the eip
4811 	 * and the address again, we can break out of the potential infinite
4812 	 * loop.
4813 	 */
4814 	vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
4815 
4816 	if (!(emulation_type & EMULTYPE_RETRY))
4817 		return false;
4818 
4819 	if (x86_page_table_writing_insn(ctxt))
4820 		return false;
4821 
4822 	if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
4823 		return false;
4824 
4825 	vcpu->arch.last_retry_eip = ctxt->eip;
4826 	vcpu->arch.last_retry_addr = cr2;
4827 
4828 	if (!vcpu->arch.mmu.direct_map)
4829 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
4830 
4831 	kvm_mmu_unprotect_page(vcpu->kvm, gpa >> PAGE_SHIFT);
4832 
4833 	return true;
4834 }
4835 
4836 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
4837 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
4838 
4839 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
4840 			    unsigned long cr2,
4841 			    int emulation_type,
4842 			    void *insn,
4843 			    int insn_len)
4844 {
4845 	int r;
4846 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4847 	bool writeback = true;
4848 
4849 	kvm_clear_exception_queue(vcpu);
4850 
4851 	if (!(emulation_type & EMULTYPE_NO_DECODE)) {
4852 		init_emulate_ctxt(vcpu);
4853 		ctxt->interruptibility = 0;
4854 		ctxt->have_exception = false;
4855 		ctxt->perm_ok = false;
4856 
4857 		ctxt->only_vendor_specific_insn
4858 			= emulation_type & EMULTYPE_TRAP_UD;
4859 
4860 		r = x86_decode_insn(ctxt, insn, insn_len);
4861 
4862 		trace_kvm_emulate_insn_start(vcpu);
4863 		++vcpu->stat.insn_emulation;
4864 		if (r != EMULATION_OK)  {
4865 			if (emulation_type & EMULTYPE_TRAP_UD)
4866 				return EMULATE_FAIL;
4867 			if (reexecute_instruction(vcpu, cr2))
4868 				return EMULATE_DONE;
4869 			if (emulation_type & EMULTYPE_SKIP)
4870 				return EMULATE_FAIL;
4871 			return handle_emulation_failure(vcpu);
4872 		}
4873 	}
4874 
4875 	if (emulation_type & EMULTYPE_SKIP) {
4876 		kvm_rip_write(vcpu, ctxt->_eip);
4877 		return EMULATE_DONE;
4878 	}
4879 
4880 	if (retry_instruction(ctxt, cr2, emulation_type))
4881 		return EMULATE_DONE;
4882 
4883 	/* this is needed for vmware backdoor interface to work since it
4884 	   changes registers values  during IO operation */
4885 	if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
4886 		vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
4887 		emulator_invalidate_register_cache(ctxt);
4888 	}
4889 
4890 restart:
4891 	r = x86_emulate_insn(ctxt);
4892 
4893 	if (r == EMULATION_INTERCEPTED)
4894 		return EMULATE_DONE;
4895 
4896 	if (r == EMULATION_FAILED) {
4897 		if (reexecute_instruction(vcpu, cr2))
4898 			return EMULATE_DONE;
4899 
4900 		return handle_emulation_failure(vcpu);
4901 	}
4902 
4903 	if (ctxt->have_exception) {
4904 		inject_emulated_exception(vcpu);
4905 		r = EMULATE_DONE;
4906 	} else if (vcpu->arch.pio.count) {
4907 		if (!vcpu->arch.pio.in)
4908 			vcpu->arch.pio.count = 0;
4909 		else {
4910 			writeback = false;
4911 			vcpu->arch.complete_userspace_io = complete_emulated_pio;
4912 		}
4913 		r = EMULATE_DO_MMIO;
4914 	} else if (vcpu->mmio_needed) {
4915 		if (!vcpu->mmio_is_write)
4916 			writeback = false;
4917 		r = EMULATE_DO_MMIO;
4918 		vcpu->arch.complete_userspace_io = complete_emulated_mmio;
4919 	} else if (r == EMULATION_RESTART)
4920 		goto restart;
4921 	else
4922 		r = EMULATE_DONE;
4923 
4924 	if (writeback) {
4925 		toggle_interruptibility(vcpu, ctxt->interruptibility);
4926 		kvm_set_rflags(vcpu, ctxt->eflags);
4927 		kvm_make_request(KVM_REQ_EVENT, vcpu);
4928 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
4929 		kvm_rip_write(vcpu, ctxt->eip);
4930 	} else
4931 		vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
4932 
4933 	return r;
4934 }
4935 EXPORT_SYMBOL_GPL(x86_emulate_instruction);
4936 
4937 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
4938 {
4939 	unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
4940 	int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
4941 					    size, port, &val, 1);
4942 	/* do not return to emulator after return from userspace */
4943 	vcpu->arch.pio.count = 0;
4944 	return ret;
4945 }
4946 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
4947 
4948 static void tsc_bad(void *info)
4949 {
4950 	__this_cpu_write(cpu_tsc_khz, 0);
4951 }
4952 
4953 static void tsc_khz_changed(void *data)
4954 {
4955 	struct cpufreq_freqs *freq = data;
4956 	unsigned long khz = 0;
4957 
4958 	if (data)
4959 		khz = freq->new;
4960 	else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
4961 		khz = cpufreq_quick_get(raw_smp_processor_id());
4962 	if (!khz)
4963 		khz = tsc_khz;
4964 	__this_cpu_write(cpu_tsc_khz, khz);
4965 }
4966 
4967 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
4968 				     void *data)
4969 {
4970 	struct cpufreq_freqs *freq = data;
4971 	struct kvm *kvm;
4972 	struct kvm_vcpu *vcpu;
4973 	int i, send_ipi = 0;
4974 
4975 	/*
4976 	 * We allow guests to temporarily run on slowing clocks,
4977 	 * provided we notify them after, or to run on accelerating
4978 	 * clocks, provided we notify them before.  Thus time never
4979 	 * goes backwards.
4980 	 *
4981 	 * However, we have a problem.  We can't atomically update
4982 	 * the frequency of a given CPU from this function; it is
4983 	 * merely a notifier, which can be called from any CPU.
4984 	 * Changing the TSC frequency at arbitrary points in time
4985 	 * requires a recomputation of local variables related to
4986 	 * the TSC for each VCPU.  We must flag these local variables
4987 	 * to be updated and be sure the update takes place with the
4988 	 * new frequency before any guests proceed.
4989 	 *
4990 	 * Unfortunately, the combination of hotplug CPU and frequency
4991 	 * change creates an intractable locking scenario; the order
4992 	 * of when these callouts happen is undefined with respect to
4993 	 * CPU hotplug, and they can race with each other.  As such,
4994 	 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
4995 	 * undefined; you can actually have a CPU frequency change take
4996 	 * place in between the computation of X and the setting of the
4997 	 * variable.  To protect against this problem, all updates of
4998 	 * the per_cpu tsc_khz variable are done in an interrupt
4999 	 * protected IPI, and all callers wishing to update the value
5000 	 * must wait for a synchronous IPI to complete (which is trivial
5001 	 * if the caller is on the CPU already).  This establishes the
5002 	 * necessary total order on variable updates.
5003 	 *
5004 	 * Note that because a guest time update may take place
5005 	 * anytime after the setting of the VCPU's request bit, the
5006 	 * correct TSC value must be set before the request.  However,
5007 	 * to ensure the update actually makes it to any guest which
5008 	 * starts running in hardware virtualization between the set
5009 	 * and the acquisition of the spinlock, we must also ping the
5010 	 * CPU after setting the request bit.
5011 	 *
5012 	 */
5013 
5014 	if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
5015 		return 0;
5016 	if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
5017 		return 0;
5018 
5019 	smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5020 
5021 	raw_spin_lock(&kvm_lock);
5022 	list_for_each_entry(kvm, &vm_list, vm_list) {
5023 		kvm_for_each_vcpu(i, vcpu, kvm) {
5024 			if (vcpu->cpu != freq->cpu)
5025 				continue;
5026 			kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5027 			if (vcpu->cpu != smp_processor_id())
5028 				send_ipi = 1;
5029 		}
5030 	}
5031 	raw_spin_unlock(&kvm_lock);
5032 
5033 	if (freq->old < freq->new && send_ipi) {
5034 		/*
5035 		 * We upscale the frequency.  Must make the guest
5036 		 * doesn't see old kvmclock values while running with
5037 		 * the new frequency, otherwise we risk the guest sees
5038 		 * time go backwards.
5039 		 *
5040 		 * In case we update the frequency for another cpu
5041 		 * (which might be in guest context) send an interrupt
5042 		 * to kick the cpu out of guest context.  Next time
5043 		 * guest context is entered kvmclock will be updated,
5044 		 * so the guest will not see stale values.
5045 		 */
5046 		smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5047 	}
5048 	return 0;
5049 }
5050 
5051 static struct notifier_block kvmclock_cpufreq_notifier_block = {
5052 	.notifier_call  = kvmclock_cpufreq_notifier
5053 };
5054 
5055 static int kvmclock_cpu_notifier(struct notifier_block *nfb,
5056 					unsigned long action, void *hcpu)
5057 {
5058 	unsigned int cpu = (unsigned long)hcpu;
5059 
5060 	switch (action) {
5061 		case CPU_ONLINE:
5062 		case CPU_DOWN_FAILED:
5063 			smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5064 			break;
5065 		case CPU_DOWN_PREPARE:
5066 			smp_call_function_single(cpu, tsc_bad, NULL, 1);
5067 			break;
5068 	}
5069 	return NOTIFY_OK;
5070 }
5071 
5072 static struct notifier_block kvmclock_cpu_notifier_block = {
5073 	.notifier_call  = kvmclock_cpu_notifier,
5074 	.priority = -INT_MAX
5075 };
5076 
5077 static void kvm_timer_init(void)
5078 {
5079 	int cpu;
5080 
5081 	max_tsc_khz = tsc_khz;
5082 	register_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5083 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5084 #ifdef CONFIG_CPU_FREQ
5085 		struct cpufreq_policy policy;
5086 		memset(&policy, 0, sizeof(policy));
5087 		cpu = get_cpu();
5088 		cpufreq_get_policy(&policy, cpu);
5089 		if (policy.cpuinfo.max_freq)
5090 			max_tsc_khz = policy.cpuinfo.max_freq;
5091 		put_cpu();
5092 #endif
5093 		cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
5094 					  CPUFREQ_TRANSITION_NOTIFIER);
5095 	}
5096 	pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
5097 	for_each_online_cpu(cpu)
5098 		smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5099 }
5100 
5101 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
5102 
5103 int kvm_is_in_guest(void)
5104 {
5105 	return __this_cpu_read(current_vcpu) != NULL;
5106 }
5107 
5108 static int kvm_is_user_mode(void)
5109 {
5110 	int user_mode = 3;
5111 
5112 	if (__this_cpu_read(current_vcpu))
5113 		user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
5114 
5115 	return user_mode != 0;
5116 }
5117 
5118 static unsigned long kvm_get_guest_ip(void)
5119 {
5120 	unsigned long ip = 0;
5121 
5122 	if (__this_cpu_read(current_vcpu))
5123 		ip = kvm_rip_read(__this_cpu_read(current_vcpu));
5124 
5125 	return ip;
5126 }
5127 
5128 static struct perf_guest_info_callbacks kvm_guest_cbs = {
5129 	.is_in_guest		= kvm_is_in_guest,
5130 	.is_user_mode		= kvm_is_user_mode,
5131 	.get_guest_ip		= kvm_get_guest_ip,
5132 };
5133 
5134 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
5135 {
5136 	__this_cpu_write(current_vcpu, vcpu);
5137 }
5138 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
5139 
5140 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
5141 {
5142 	__this_cpu_write(current_vcpu, NULL);
5143 }
5144 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
5145 
5146 static void kvm_set_mmio_spte_mask(void)
5147 {
5148 	u64 mask;
5149 	int maxphyaddr = boot_cpu_data.x86_phys_bits;
5150 
5151 	/*
5152 	 * Set the reserved bits and the present bit of an paging-structure
5153 	 * entry to generate page fault with PFER.RSV = 1.
5154 	 */
5155 	mask = ((1ull << (62 - maxphyaddr + 1)) - 1) << maxphyaddr;
5156 	mask |= 1ull;
5157 
5158 #ifdef CONFIG_X86_64
5159 	/*
5160 	 * If reserved bit is not supported, clear the present bit to disable
5161 	 * mmio page fault.
5162 	 */
5163 	if (maxphyaddr == 52)
5164 		mask &= ~1ull;
5165 #endif
5166 
5167 	kvm_mmu_set_mmio_spte_mask(mask);
5168 }
5169 
5170 #ifdef CONFIG_X86_64
5171 static void pvclock_gtod_update_fn(struct work_struct *work)
5172 {
5173 	struct kvm *kvm;
5174 
5175 	struct kvm_vcpu *vcpu;
5176 	int i;
5177 
5178 	raw_spin_lock(&kvm_lock);
5179 	list_for_each_entry(kvm, &vm_list, vm_list)
5180 		kvm_for_each_vcpu(i, vcpu, kvm)
5181 			set_bit(KVM_REQ_MASTERCLOCK_UPDATE, &vcpu->requests);
5182 	atomic_set(&kvm_guest_has_master_clock, 0);
5183 	raw_spin_unlock(&kvm_lock);
5184 }
5185 
5186 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
5187 
5188 /*
5189  * Notification about pvclock gtod data update.
5190  */
5191 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
5192 			       void *priv)
5193 {
5194 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
5195 	struct timekeeper *tk = priv;
5196 
5197 	update_pvclock_gtod(tk);
5198 
5199 	/* disable master clock if host does not trust, or does not
5200 	 * use, TSC clocksource
5201 	 */
5202 	if (gtod->clock.vclock_mode != VCLOCK_TSC &&
5203 	    atomic_read(&kvm_guest_has_master_clock) != 0)
5204 		queue_work(system_long_wq, &pvclock_gtod_work);
5205 
5206 	return 0;
5207 }
5208 
5209 static struct notifier_block pvclock_gtod_notifier = {
5210 	.notifier_call = pvclock_gtod_notify,
5211 };
5212 #endif
5213 
5214 int kvm_arch_init(void *opaque)
5215 {
5216 	int r;
5217 	struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
5218 
5219 	if (kvm_x86_ops) {
5220 		printk(KERN_ERR "kvm: already loaded the other module\n");
5221 		r = -EEXIST;
5222 		goto out;
5223 	}
5224 
5225 	if (!ops->cpu_has_kvm_support()) {
5226 		printk(KERN_ERR "kvm: no hardware support\n");
5227 		r = -EOPNOTSUPP;
5228 		goto out;
5229 	}
5230 	if (ops->disabled_by_bios()) {
5231 		printk(KERN_ERR "kvm: disabled by bios\n");
5232 		r = -EOPNOTSUPP;
5233 		goto out;
5234 	}
5235 
5236 	r = kvm_mmu_module_init();
5237 	if (r)
5238 		goto out;
5239 
5240 	kvm_set_mmio_spte_mask();
5241 	kvm_init_msr_list();
5242 
5243 	kvm_x86_ops = ops;
5244 	kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
5245 			PT_DIRTY_MASK, PT64_NX_MASK, 0);
5246 
5247 	kvm_timer_init();
5248 
5249 	perf_register_guest_info_callbacks(&kvm_guest_cbs);
5250 
5251 	if (cpu_has_xsave)
5252 		host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
5253 
5254 	kvm_lapic_init();
5255 #ifdef CONFIG_X86_64
5256 	pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
5257 #endif
5258 
5259 	return 0;
5260 
5261 out:
5262 	return r;
5263 }
5264 
5265 void kvm_arch_exit(void)
5266 {
5267 	perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
5268 
5269 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5270 		cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
5271 					    CPUFREQ_TRANSITION_NOTIFIER);
5272 	unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5273 #ifdef CONFIG_X86_64
5274 	pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
5275 #endif
5276 	kvm_x86_ops = NULL;
5277 	kvm_mmu_module_exit();
5278 }
5279 
5280 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
5281 {
5282 	++vcpu->stat.halt_exits;
5283 	if (irqchip_in_kernel(vcpu->kvm)) {
5284 		vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
5285 		return 1;
5286 	} else {
5287 		vcpu->run->exit_reason = KVM_EXIT_HLT;
5288 		return 0;
5289 	}
5290 }
5291 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
5292 
5293 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
5294 {
5295 	u64 param, ingpa, outgpa, ret;
5296 	uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
5297 	bool fast, longmode;
5298 	int cs_db, cs_l;
5299 
5300 	/*
5301 	 * hypercall generates UD from non zero cpl and real mode
5302 	 * per HYPER-V spec
5303 	 */
5304 	if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
5305 		kvm_queue_exception(vcpu, UD_VECTOR);
5306 		return 0;
5307 	}
5308 
5309 	kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
5310 	longmode = is_long_mode(vcpu) && cs_l == 1;
5311 
5312 	if (!longmode) {
5313 		param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
5314 			(kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
5315 		ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
5316 			(kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
5317 		outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
5318 			(kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
5319 	}
5320 #ifdef CONFIG_X86_64
5321 	else {
5322 		param = kvm_register_read(vcpu, VCPU_REGS_RCX);
5323 		ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
5324 		outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
5325 	}
5326 #endif
5327 
5328 	code = param & 0xffff;
5329 	fast = (param >> 16) & 0x1;
5330 	rep_cnt = (param >> 32) & 0xfff;
5331 	rep_idx = (param >> 48) & 0xfff;
5332 
5333 	trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
5334 
5335 	switch (code) {
5336 	case HV_X64_HV_NOTIFY_LONG_SPIN_WAIT:
5337 		kvm_vcpu_on_spin(vcpu);
5338 		break;
5339 	default:
5340 		res = HV_STATUS_INVALID_HYPERCALL_CODE;
5341 		break;
5342 	}
5343 
5344 	ret = res | (((u64)rep_done & 0xfff) << 32);
5345 	if (longmode) {
5346 		kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5347 	} else {
5348 		kvm_register_write(vcpu, VCPU_REGS_RDX, ret >> 32);
5349 		kvm_register_write(vcpu, VCPU_REGS_RAX, ret & 0xffffffff);
5350 	}
5351 
5352 	return 1;
5353 }
5354 
5355 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
5356 {
5357 	unsigned long nr, a0, a1, a2, a3, ret;
5358 	int r = 1;
5359 
5360 	if (kvm_hv_hypercall_enabled(vcpu->kvm))
5361 		return kvm_hv_hypercall(vcpu);
5362 
5363 	nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
5364 	a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
5365 	a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
5366 	a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
5367 	a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
5368 
5369 	trace_kvm_hypercall(nr, a0, a1, a2, a3);
5370 
5371 	if (!is_long_mode(vcpu)) {
5372 		nr &= 0xFFFFFFFF;
5373 		a0 &= 0xFFFFFFFF;
5374 		a1 &= 0xFFFFFFFF;
5375 		a2 &= 0xFFFFFFFF;
5376 		a3 &= 0xFFFFFFFF;
5377 	}
5378 
5379 	if (kvm_x86_ops->get_cpl(vcpu) != 0) {
5380 		ret = -KVM_EPERM;
5381 		goto out;
5382 	}
5383 
5384 	switch (nr) {
5385 	case KVM_HC_VAPIC_POLL_IRQ:
5386 		ret = 0;
5387 		break;
5388 	default:
5389 		ret = -KVM_ENOSYS;
5390 		break;
5391 	}
5392 out:
5393 	kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5394 	++vcpu->stat.hypercalls;
5395 	return r;
5396 }
5397 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
5398 
5399 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
5400 {
5401 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5402 	char instruction[3];
5403 	unsigned long rip = kvm_rip_read(vcpu);
5404 
5405 	/*
5406 	 * Blow out the MMU to ensure that no other VCPU has an active mapping
5407 	 * to ensure that the updated hypercall appears atomically across all
5408 	 * VCPUs.
5409 	 */
5410 	kvm_mmu_zap_all(vcpu->kvm);
5411 
5412 	kvm_x86_ops->patch_hypercall(vcpu, instruction);
5413 
5414 	return emulator_write_emulated(ctxt, rip, instruction, 3, NULL);
5415 }
5416 
5417 /*
5418  * Check if userspace requested an interrupt window, and that the
5419  * interrupt window is open.
5420  *
5421  * No need to exit to userspace if we already have an interrupt queued.
5422  */
5423 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
5424 {
5425 	return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
5426 		vcpu->run->request_interrupt_window &&
5427 		kvm_arch_interrupt_allowed(vcpu));
5428 }
5429 
5430 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
5431 {
5432 	struct kvm_run *kvm_run = vcpu->run;
5433 
5434 	kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
5435 	kvm_run->cr8 = kvm_get_cr8(vcpu);
5436 	kvm_run->apic_base = kvm_get_apic_base(vcpu);
5437 	if (irqchip_in_kernel(vcpu->kvm))
5438 		kvm_run->ready_for_interrupt_injection = 1;
5439 	else
5440 		kvm_run->ready_for_interrupt_injection =
5441 			kvm_arch_interrupt_allowed(vcpu) &&
5442 			!kvm_cpu_has_interrupt(vcpu) &&
5443 			!kvm_event_needs_reinjection(vcpu);
5444 }
5445 
5446 static int vapic_enter(struct kvm_vcpu *vcpu)
5447 {
5448 	struct kvm_lapic *apic = vcpu->arch.apic;
5449 	struct page *page;
5450 
5451 	if (!apic || !apic->vapic_addr)
5452 		return 0;
5453 
5454 	page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
5455 	if (is_error_page(page))
5456 		return -EFAULT;
5457 
5458 	vcpu->arch.apic->vapic_page = page;
5459 	return 0;
5460 }
5461 
5462 static void vapic_exit(struct kvm_vcpu *vcpu)
5463 {
5464 	struct kvm_lapic *apic = vcpu->arch.apic;
5465 	int idx;
5466 
5467 	if (!apic || !apic->vapic_addr)
5468 		return;
5469 
5470 	idx = srcu_read_lock(&vcpu->kvm->srcu);
5471 	kvm_release_page_dirty(apic->vapic_page);
5472 	mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
5473 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
5474 }
5475 
5476 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
5477 {
5478 	int max_irr, tpr;
5479 
5480 	if (!kvm_x86_ops->update_cr8_intercept)
5481 		return;
5482 
5483 	if (!vcpu->arch.apic)
5484 		return;
5485 
5486 	if (!vcpu->arch.apic->vapic_addr)
5487 		max_irr = kvm_lapic_find_highest_irr(vcpu);
5488 	else
5489 		max_irr = -1;
5490 
5491 	if (max_irr != -1)
5492 		max_irr >>= 4;
5493 
5494 	tpr = kvm_lapic_get_cr8(vcpu);
5495 
5496 	kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
5497 }
5498 
5499 static void inject_pending_event(struct kvm_vcpu *vcpu)
5500 {
5501 	/* try to reinject previous events if any */
5502 	if (vcpu->arch.exception.pending) {
5503 		trace_kvm_inj_exception(vcpu->arch.exception.nr,
5504 					vcpu->arch.exception.has_error_code,
5505 					vcpu->arch.exception.error_code);
5506 		kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
5507 					  vcpu->arch.exception.has_error_code,
5508 					  vcpu->arch.exception.error_code,
5509 					  vcpu->arch.exception.reinject);
5510 		return;
5511 	}
5512 
5513 	if (vcpu->arch.nmi_injected) {
5514 		kvm_x86_ops->set_nmi(vcpu);
5515 		return;
5516 	}
5517 
5518 	if (vcpu->arch.interrupt.pending) {
5519 		kvm_x86_ops->set_irq(vcpu);
5520 		return;
5521 	}
5522 
5523 	/* try to inject new event if pending */
5524 	if (vcpu->arch.nmi_pending) {
5525 		if (kvm_x86_ops->nmi_allowed(vcpu)) {
5526 			--vcpu->arch.nmi_pending;
5527 			vcpu->arch.nmi_injected = true;
5528 			kvm_x86_ops->set_nmi(vcpu);
5529 		}
5530 	} else if (kvm_cpu_has_interrupt(vcpu)) {
5531 		if (kvm_x86_ops->interrupt_allowed(vcpu)) {
5532 			kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
5533 					    false);
5534 			kvm_x86_ops->set_irq(vcpu);
5535 		}
5536 	}
5537 }
5538 
5539 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
5540 {
5541 	if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
5542 			!vcpu->guest_xcr0_loaded) {
5543 		/* kvm_set_xcr() also depends on this */
5544 		xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
5545 		vcpu->guest_xcr0_loaded = 1;
5546 	}
5547 }
5548 
5549 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
5550 {
5551 	if (vcpu->guest_xcr0_loaded) {
5552 		if (vcpu->arch.xcr0 != host_xcr0)
5553 			xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
5554 		vcpu->guest_xcr0_loaded = 0;
5555 	}
5556 }
5557 
5558 static void process_nmi(struct kvm_vcpu *vcpu)
5559 {
5560 	unsigned limit = 2;
5561 
5562 	/*
5563 	 * x86 is limited to one NMI running, and one NMI pending after it.
5564 	 * If an NMI is already in progress, limit further NMIs to just one.
5565 	 * Otherwise, allow two (and we'll inject the first one immediately).
5566 	 */
5567 	if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
5568 		limit = 1;
5569 
5570 	vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
5571 	vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
5572 	kvm_make_request(KVM_REQ_EVENT, vcpu);
5573 }
5574 
5575 static void kvm_gen_update_masterclock(struct kvm *kvm)
5576 {
5577 #ifdef CONFIG_X86_64
5578 	int i;
5579 	struct kvm_vcpu *vcpu;
5580 	struct kvm_arch *ka = &kvm->arch;
5581 
5582 	spin_lock(&ka->pvclock_gtod_sync_lock);
5583 	kvm_make_mclock_inprogress_request(kvm);
5584 	/* no guest entries from this point */
5585 	pvclock_update_vm_gtod_copy(kvm);
5586 
5587 	kvm_for_each_vcpu(i, vcpu, kvm)
5588 		set_bit(KVM_REQ_CLOCK_UPDATE, &vcpu->requests);
5589 
5590 	/* guest entries allowed */
5591 	kvm_for_each_vcpu(i, vcpu, kvm)
5592 		clear_bit(KVM_REQ_MCLOCK_INPROGRESS, &vcpu->requests);
5593 
5594 	spin_unlock(&ka->pvclock_gtod_sync_lock);
5595 #endif
5596 }
5597 
5598 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
5599 {
5600 	int r;
5601 	bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
5602 		vcpu->run->request_interrupt_window;
5603 	bool req_immediate_exit = 0;
5604 
5605 	if (vcpu->requests) {
5606 		if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
5607 			kvm_mmu_unload(vcpu);
5608 		if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
5609 			__kvm_migrate_timers(vcpu);
5610 		if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
5611 			kvm_gen_update_masterclock(vcpu->kvm);
5612 		if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
5613 			r = kvm_guest_time_update(vcpu);
5614 			if (unlikely(r))
5615 				goto out;
5616 		}
5617 		if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
5618 			kvm_mmu_sync_roots(vcpu);
5619 		if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
5620 			kvm_x86_ops->tlb_flush(vcpu);
5621 		if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
5622 			vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
5623 			r = 0;
5624 			goto out;
5625 		}
5626 		if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
5627 			vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
5628 			r = 0;
5629 			goto out;
5630 		}
5631 		if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
5632 			vcpu->fpu_active = 0;
5633 			kvm_x86_ops->fpu_deactivate(vcpu);
5634 		}
5635 		if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
5636 			/* Page is swapped out. Do synthetic halt */
5637 			vcpu->arch.apf.halted = true;
5638 			r = 1;
5639 			goto out;
5640 		}
5641 		if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
5642 			record_steal_time(vcpu);
5643 		if (kvm_check_request(KVM_REQ_NMI, vcpu))
5644 			process_nmi(vcpu);
5645 		req_immediate_exit =
5646 			kvm_check_request(KVM_REQ_IMMEDIATE_EXIT, vcpu);
5647 		if (kvm_check_request(KVM_REQ_PMU, vcpu))
5648 			kvm_handle_pmu_event(vcpu);
5649 		if (kvm_check_request(KVM_REQ_PMI, vcpu))
5650 			kvm_deliver_pmi(vcpu);
5651 	}
5652 
5653 	if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
5654 		inject_pending_event(vcpu);
5655 
5656 		/* enable NMI/IRQ window open exits if needed */
5657 		if (vcpu->arch.nmi_pending)
5658 			kvm_x86_ops->enable_nmi_window(vcpu);
5659 		else if (kvm_cpu_has_interrupt(vcpu) || req_int_win)
5660 			kvm_x86_ops->enable_irq_window(vcpu);
5661 
5662 		if (kvm_lapic_enabled(vcpu)) {
5663 			update_cr8_intercept(vcpu);
5664 			kvm_lapic_sync_to_vapic(vcpu);
5665 		}
5666 	}
5667 
5668 	r = kvm_mmu_reload(vcpu);
5669 	if (unlikely(r)) {
5670 		goto cancel_injection;
5671 	}
5672 
5673 	preempt_disable();
5674 
5675 	kvm_x86_ops->prepare_guest_switch(vcpu);
5676 	if (vcpu->fpu_active)
5677 		kvm_load_guest_fpu(vcpu);
5678 	kvm_load_guest_xcr0(vcpu);
5679 
5680 	vcpu->mode = IN_GUEST_MODE;
5681 
5682 	/* We should set ->mode before check ->requests,
5683 	 * see the comment in make_all_cpus_request.
5684 	 */
5685 	smp_mb();
5686 
5687 	local_irq_disable();
5688 
5689 	if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests
5690 	    || need_resched() || signal_pending(current)) {
5691 		vcpu->mode = OUTSIDE_GUEST_MODE;
5692 		smp_wmb();
5693 		local_irq_enable();
5694 		preempt_enable();
5695 		r = 1;
5696 		goto cancel_injection;
5697 	}
5698 
5699 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5700 
5701 	if (req_immediate_exit)
5702 		smp_send_reschedule(vcpu->cpu);
5703 
5704 	kvm_guest_enter();
5705 
5706 	if (unlikely(vcpu->arch.switch_db_regs)) {
5707 		set_debugreg(0, 7);
5708 		set_debugreg(vcpu->arch.eff_db[0], 0);
5709 		set_debugreg(vcpu->arch.eff_db[1], 1);
5710 		set_debugreg(vcpu->arch.eff_db[2], 2);
5711 		set_debugreg(vcpu->arch.eff_db[3], 3);
5712 	}
5713 
5714 	trace_kvm_entry(vcpu->vcpu_id);
5715 	kvm_x86_ops->run(vcpu);
5716 
5717 	/*
5718 	 * If the guest has used debug registers, at least dr7
5719 	 * will be disabled while returning to the host.
5720 	 * If we don't have active breakpoints in the host, we don't
5721 	 * care about the messed up debug address registers. But if
5722 	 * we have some of them active, restore the old state.
5723 	 */
5724 	if (hw_breakpoint_active())
5725 		hw_breakpoint_restore();
5726 
5727 	vcpu->arch.last_guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu,
5728 							   native_read_tsc());
5729 
5730 	vcpu->mode = OUTSIDE_GUEST_MODE;
5731 	smp_wmb();
5732 	local_irq_enable();
5733 
5734 	++vcpu->stat.exits;
5735 
5736 	/*
5737 	 * We must have an instruction between local_irq_enable() and
5738 	 * kvm_guest_exit(), so the timer interrupt isn't delayed by
5739 	 * the interrupt shadow.  The stat.exits increment will do nicely.
5740 	 * But we need to prevent reordering, hence this barrier():
5741 	 */
5742 	barrier();
5743 
5744 	kvm_guest_exit();
5745 
5746 	preempt_enable();
5747 
5748 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5749 
5750 	/*
5751 	 * Profile KVM exit RIPs:
5752 	 */
5753 	if (unlikely(prof_on == KVM_PROFILING)) {
5754 		unsigned long rip = kvm_rip_read(vcpu);
5755 		profile_hit(KVM_PROFILING, (void *)rip);
5756 	}
5757 
5758 	if (unlikely(vcpu->arch.tsc_always_catchup))
5759 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5760 
5761 	if (vcpu->arch.apic_attention)
5762 		kvm_lapic_sync_from_vapic(vcpu);
5763 
5764 	r = kvm_x86_ops->handle_exit(vcpu);
5765 	return r;
5766 
5767 cancel_injection:
5768 	kvm_x86_ops->cancel_injection(vcpu);
5769 	if (unlikely(vcpu->arch.apic_attention))
5770 		kvm_lapic_sync_from_vapic(vcpu);
5771 out:
5772 	return r;
5773 }
5774 
5775 
5776 static int __vcpu_run(struct kvm_vcpu *vcpu)
5777 {
5778 	int r;
5779 	struct kvm *kvm = vcpu->kvm;
5780 
5781 	if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
5782 		pr_debug("vcpu %d received sipi with vector # %x\n",
5783 			 vcpu->vcpu_id, vcpu->arch.sipi_vector);
5784 		kvm_lapic_reset(vcpu);
5785 		r = kvm_vcpu_reset(vcpu);
5786 		if (r)
5787 			return r;
5788 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5789 	}
5790 
5791 	vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5792 	r = vapic_enter(vcpu);
5793 	if (r) {
5794 		srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5795 		return r;
5796 	}
5797 
5798 	r = 1;
5799 	while (r > 0) {
5800 		if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
5801 		    !vcpu->arch.apf.halted)
5802 			r = vcpu_enter_guest(vcpu);
5803 		else {
5804 			srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5805 			kvm_vcpu_block(vcpu);
5806 			vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5807 			if (kvm_check_request(KVM_REQ_UNHALT, vcpu))
5808 			{
5809 				switch(vcpu->arch.mp_state) {
5810 				case KVM_MP_STATE_HALTED:
5811 					vcpu->arch.mp_state =
5812 						KVM_MP_STATE_RUNNABLE;
5813 				case KVM_MP_STATE_RUNNABLE:
5814 					vcpu->arch.apf.halted = false;
5815 					break;
5816 				case KVM_MP_STATE_SIPI_RECEIVED:
5817 				default:
5818 					r = -EINTR;
5819 					break;
5820 				}
5821 			}
5822 		}
5823 
5824 		if (r <= 0)
5825 			break;
5826 
5827 		clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
5828 		if (kvm_cpu_has_pending_timer(vcpu))
5829 			kvm_inject_pending_timer_irqs(vcpu);
5830 
5831 		if (dm_request_for_irq_injection(vcpu)) {
5832 			r = -EINTR;
5833 			vcpu->run->exit_reason = KVM_EXIT_INTR;
5834 			++vcpu->stat.request_irq_exits;
5835 		}
5836 
5837 		kvm_check_async_pf_completion(vcpu);
5838 
5839 		if (signal_pending(current)) {
5840 			r = -EINTR;
5841 			vcpu->run->exit_reason = KVM_EXIT_INTR;
5842 			++vcpu->stat.signal_exits;
5843 		}
5844 		if (need_resched()) {
5845 			srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5846 			kvm_resched(vcpu);
5847 			vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5848 		}
5849 	}
5850 
5851 	srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5852 
5853 	vapic_exit(vcpu);
5854 
5855 	return r;
5856 }
5857 
5858 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
5859 {
5860 	int r;
5861 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5862 	r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
5863 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5864 	if (r != EMULATE_DONE)
5865 		return 0;
5866 	return 1;
5867 }
5868 
5869 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
5870 {
5871 	BUG_ON(!vcpu->arch.pio.count);
5872 
5873 	return complete_emulated_io(vcpu);
5874 }
5875 
5876 /*
5877  * Implements the following, as a state machine:
5878  *
5879  * read:
5880  *   for each fragment
5881  *     for each mmio piece in the fragment
5882  *       write gpa, len
5883  *       exit
5884  *       copy data
5885  *   execute insn
5886  *
5887  * write:
5888  *   for each fragment
5889  *     for each mmio piece in the fragment
5890  *       write gpa, len
5891  *       copy data
5892  *       exit
5893  */
5894 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
5895 {
5896 	struct kvm_run *run = vcpu->run;
5897 	struct kvm_mmio_fragment *frag;
5898 	unsigned len;
5899 
5900 	BUG_ON(!vcpu->mmio_needed);
5901 
5902 	/* Complete previous fragment */
5903 	frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
5904 	len = min(8u, frag->len);
5905 	if (!vcpu->mmio_is_write)
5906 		memcpy(frag->data, run->mmio.data, len);
5907 
5908 	if (frag->len <= 8) {
5909 		/* Switch to the next fragment. */
5910 		frag++;
5911 		vcpu->mmio_cur_fragment++;
5912 	} else {
5913 		/* Go forward to the next mmio piece. */
5914 		frag->data += len;
5915 		frag->gpa += len;
5916 		frag->len -= len;
5917 	}
5918 
5919 	if (vcpu->mmio_cur_fragment == vcpu->mmio_nr_fragments) {
5920 		vcpu->mmio_needed = 0;
5921 		if (vcpu->mmio_is_write)
5922 			return 1;
5923 		vcpu->mmio_read_completed = 1;
5924 		return complete_emulated_io(vcpu);
5925 	}
5926 
5927 	run->exit_reason = KVM_EXIT_MMIO;
5928 	run->mmio.phys_addr = frag->gpa;
5929 	if (vcpu->mmio_is_write)
5930 		memcpy(run->mmio.data, frag->data, min(8u, frag->len));
5931 	run->mmio.len = min(8u, frag->len);
5932 	run->mmio.is_write = vcpu->mmio_is_write;
5933 	vcpu->arch.complete_userspace_io = complete_emulated_mmio;
5934 	return 0;
5935 }
5936 
5937 
5938 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
5939 {
5940 	int r;
5941 	sigset_t sigsaved;
5942 
5943 	if (!tsk_used_math(current) && init_fpu(current))
5944 		return -ENOMEM;
5945 
5946 	if (vcpu->sigset_active)
5947 		sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
5948 
5949 	if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
5950 		kvm_vcpu_block(vcpu);
5951 		clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
5952 		r = -EAGAIN;
5953 		goto out;
5954 	}
5955 
5956 	/* re-sync apic's tpr */
5957 	if (!irqchip_in_kernel(vcpu->kvm)) {
5958 		if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
5959 			r = -EINVAL;
5960 			goto out;
5961 		}
5962 	}
5963 
5964 	if (unlikely(vcpu->arch.complete_userspace_io)) {
5965 		int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
5966 		vcpu->arch.complete_userspace_io = NULL;
5967 		r = cui(vcpu);
5968 		if (r <= 0)
5969 			goto out;
5970 	} else
5971 		WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
5972 
5973 	r = __vcpu_run(vcpu);
5974 
5975 out:
5976 	post_kvm_run_save(vcpu);
5977 	if (vcpu->sigset_active)
5978 		sigprocmask(SIG_SETMASK, &sigsaved, NULL);
5979 
5980 	return r;
5981 }
5982 
5983 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
5984 {
5985 	if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
5986 		/*
5987 		 * We are here if userspace calls get_regs() in the middle of
5988 		 * instruction emulation. Registers state needs to be copied
5989 		 * back from emulation context to vcpu. Userspace shouldn't do
5990 		 * that usually, but some bad designed PV devices (vmware
5991 		 * backdoor interface) need this to work
5992 		 */
5993 		emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
5994 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5995 	}
5996 	regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
5997 	regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
5998 	regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
5999 	regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
6000 	regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
6001 	regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
6002 	regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
6003 	regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
6004 #ifdef CONFIG_X86_64
6005 	regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
6006 	regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
6007 	regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
6008 	regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
6009 	regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
6010 	regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
6011 	regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
6012 	regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
6013 #endif
6014 
6015 	regs->rip = kvm_rip_read(vcpu);
6016 	regs->rflags = kvm_get_rflags(vcpu);
6017 
6018 	return 0;
6019 }
6020 
6021 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6022 {
6023 	vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
6024 	vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6025 
6026 	kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
6027 	kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
6028 	kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
6029 	kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
6030 	kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
6031 	kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
6032 	kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
6033 	kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
6034 #ifdef CONFIG_X86_64
6035 	kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
6036 	kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
6037 	kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
6038 	kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
6039 	kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
6040 	kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
6041 	kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
6042 	kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
6043 #endif
6044 
6045 	kvm_rip_write(vcpu, regs->rip);
6046 	kvm_set_rflags(vcpu, regs->rflags);
6047 
6048 	vcpu->arch.exception.pending = false;
6049 
6050 	kvm_make_request(KVM_REQ_EVENT, vcpu);
6051 
6052 	return 0;
6053 }
6054 
6055 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
6056 {
6057 	struct kvm_segment cs;
6058 
6059 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
6060 	*db = cs.db;
6061 	*l = cs.l;
6062 }
6063 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
6064 
6065 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
6066 				  struct kvm_sregs *sregs)
6067 {
6068 	struct desc_ptr dt;
6069 
6070 	kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6071 	kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6072 	kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6073 	kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6074 	kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6075 	kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
6076 
6077 	kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6078 	kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
6079 
6080 	kvm_x86_ops->get_idt(vcpu, &dt);
6081 	sregs->idt.limit = dt.size;
6082 	sregs->idt.base = dt.address;
6083 	kvm_x86_ops->get_gdt(vcpu, &dt);
6084 	sregs->gdt.limit = dt.size;
6085 	sregs->gdt.base = dt.address;
6086 
6087 	sregs->cr0 = kvm_read_cr0(vcpu);
6088 	sregs->cr2 = vcpu->arch.cr2;
6089 	sregs->cr3 = kvm_read_cr3(vcpu);
6090 	sregs->cr4 = kvm_read_cr4(vcpu);
6091 	sregs->cr8 = kvm_get_cr8(vcpu);
6092 	sregs->efer = vcpu->arch.efer;
6093 	sregs->apic_base = kvm_get_apic_base(vcpu);
6094 
6095 	memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
6096 
6097 	if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
6098 		set_bit(vcpu->arch.interrupt.nr,
6099 			(unsigned long *)sregs->interrupt_bitmap);
6100 
6101 	return 0;
6102 }
6103 
6104 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
6105 				    struct kvm_mp_state *mp_state)
6106 {
6107 	mp_state->mp_state = vcpu->arch.mp_state;
6108 	return 0;
6109 }
6110 
6111 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
6112 				    struct kvm_mp_state *mp_state)
6113 {
6114 	vcpu->arch.mp_state = mp_state->mp_state;
6115 	kvm_make_request(KVM_REQ_EVENT, vcpu);
6116 	return 0;
6117 }
6118 
6119 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
6120 		    int reason, bool has_error_code, u32 error_code)
6121 {
6122 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6123 	int ret;
6124 
6125 	init_emulate_ctxt(vcpu);
6126 
6127 	ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
6128 				   has_error_code, error_code);
6129 
6130 	if (ret)
6131 		return EMULATE_FAIL;
6132 
6133 	kvm_rip_write(vcpu, ctxt->eip);
6134 	kvm_set_rflags(vcpu, ctxt->eflags);
6135 	kvm_make_request(KVM_REQ_EVENT, vcpu);
6136 	return EMULATE_DONE;
6137 }
6138 EXPORT_SYMBOL_GPL(kvm_task_switch);
6139 
6140 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
6141 				  struct kvm_sregs *sregs)
6142 {
6143 	int mmu_reset_needed = 0;
6144 	int pending_vec, max_bits, idx;
6145 	struct desc_ptr dt;
6146 
6147 	if (!guest_cpuid_has_xsave(vcpu) && (sregs->cr4 & X86_CR4_OSXSAVE))
6148 		return -EINVAL;
6149 
6150 	dt.size = sregs->idt.limit;
6151 	dt.address = sregs->idt.base;
6152 	kvm_x86_ops->set_idt(vcpu, &dt);
6153 	dt.size = sregs->gdt.limit;
6154 	dt.address = sregs->gdt.base;
6155 	kvm_x86_ops->set_gdt(vcpu, &dt);
6156 
6157 	vcpu->arch.cr2 = sregs->cr2;
6158 	mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
6159 	vcpu->arch.cr3 = sregs->cr3;
6160 	__set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
6161 
6162 	kvm_set_cr8(vcpu, sregs->cr8);
6163 
6164 	mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
6165 	kvm_x86_ops->set_efer(vcpu, sregs->efer);
6166 	kvm_set_apic_base(vcpu, sregs->apic_base);
6167 
6168 	mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
6169 	kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
6170 	vcpu->arch.cr0 = sregs->cr0;
6171 
6172 	mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
6173 	kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
6174 	if (sregs->cr4 & X86_CR4_OSXSAVE)
6175 		kvm_update_cpuid(vcpu);
6176 
6177 	idx = srcu_read_lock(&vcpu->kvm->srcu);
6178 	if (!is_long_mode(vcpu) && is_pae(vcpu)) {
6179 		load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
6180 		mmu_reset_needed = 1;
6181 	}
6182 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
6183 
6184 	if (mmu_reset_needed)
6185 		kvm_mmu_reset_context(vcpu);
6186 
6187 	max_bits = KVM_NR_INTERRUPTS;
6188 	pending_vec = find_first_bit(
6189 		(const unsigned long *)sregs->interrupt_bitmap, max_bits);
6190 	if (pending_vec < max_bits) {
6191 		kvm_queue_interrupt(vcpu, pending_vec, false);
6192 		pr_debug("Set back pending irq %d\n", pending_vec);
6193 	}
6194 
6195 	kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6196 	kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6197 	kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6198 	kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6199 	kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6200 	kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
6201 
6202 	kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6203 	kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
6204 
6205 	update_cr8_intercept(vcpu);
6206 
6207 	/* Older userspace won't unhalt the vcpu on reset. */
6208 	if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
6209 	    sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
6210 	    !is_protmode(vcpu))
6211 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
6212 
6213 	kvm_make_request(KVM_REQ_EVENT, vcpu);
6214 
6215 	return 0;
6216 }
6217 
6218 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
6219 					struct kvm_guest_debug *dbg)
6220 {
6221 	unsigned long rflags;
6222 	int i, r;
6223 
6224 	if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
6225 		r = -EBUSY;
6226 		if (vcpu->arch.exception.pending)
6227 			goto out;
6228 		if (dbg->control & KVM_GUESTDBG_INJECT_DB)
6229 			kvm_queue_exception(vcpu, DB_VECTOR);
6230 		else
6231 			kvm_queue_exception(vcpu, BP_VECTOR);
6232 	}
6233 
6234 	/*
6235 	 * Read rflags as long as potentially injected trace flags are still
6236 	 * filtered out.
6237 	 */
6238 	rflags = kvm_get_rflags(vcpu);
6239 
6240 	vcpu->guest_debug = dbg->control;
6241 	if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
6242 		vcpu->guest_debug = 0;
6243 
6244 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
6245 		for (i = 0; i < KVM_NR_DB_REGS; ++i)
6246 			vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
6247 		vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
6248 	} else {
6249 		for (i = 0; i < KVM_NR_DB_REGS; i++)
6250 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
6251 	}
6252 	kvm_update_dr7(vcpu);
6253 
6254 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6255 		vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
6256 			get_segment_base(vcpu, VCPU_SREG_CS);
6257 
6258 	/*
6259 	 * Trigger an rflags update that will inject or remove the trace
6260 	 * flags.
6261 	 */
6262 	kvm_set_rflags(vcpu, rflags);
6263 
6264 	kvm_x86_ops->update_db_bp_intercept(vcpu);
6265 
6266 	r = 0;
6267 
6268 out:
6269 
6270 	return r;
6271 }
6272 
6273 /*
6274  * Translate a guest virtual address to a guest physical address.
6275  */
6276 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
6277 				    struct kvm_translation *tr)
6278 {
6279 	unsigned long vaddr = tr->linear_address;
6280 	gpa_t gpa;
6281 	int idx;
6282 
6283 	idx = srcu_read_lock(&vcpu->kvm->srcu);
6284 	gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
6285 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
6286 	tr->physical_address = gpa;
6287 	tr->valid = gpa != UNMAPPED_GVA;
6288 	tr->writeable = 1;
6289 	tr->usermode = 0;
6290 
6291 	return 0;
6292 }
6293 
6294 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
6295 {
6296 	struct i387_fxsave_struct *fxsave =
6297 			&vcpu->arch.guest_fpu.state->fxsave;
6298 
6299 	memcpy(fpu->fpr, fxsave->st_space, 128);
6300 	fpu->fcw = fxsave->cwd;
6301 	fpu->fsw = fxsave->swd;
6302 	fpu->ftwx = fxsave->twd;
6303 	fpu->last_opcode = fxsave->fop;
6304 	fpu->last_ip = fxsave->rip;
6305 	fpu->last_dp = fxsave->rdp;
6306 	memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
6307 
6308 	return 0;
6309 }
6310 
6311 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
6312 {
6313 	struct i387_fxsave_struct *fxsave =
6314 			&vcpu->arch.guest_fpu.state->fxsave;
6315 
6316 	memcpy(fxsave->st_space, fpu->fpr, 128);
6317 	fxsave->cwd = fpu->fcw;
6318 	fxsave->swd = fpu->fsw;
6319 	fxsave->twd = fpu->ftwx;
6320 	fxsave->fop = fpu->last_opcode;
6321 	fxsave->rip = fpu->last_ip;
6322 	fxsave->rdp = fpu->last_dp;
6323 	memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
6324 
6325 	return 0;
6326 }
6327 
6328 int fx_init(struct kvm_vcpu *vcpu)
6329 {
6330 	int err;
6331 
6332 	err = fpu_alloc(&vcpu->arch.guest_fpu);
6333 	if (err)
6334 		return err;
6335 
6336 	fpu_finit(&vcpu->arch.guest_fpu);
6337 
6338 	/*
6339 	 * Ensure guest xcr0 is valid for loading
6340 	 */
6341 	vcpu->arch.xcr0 = XSTATE_FP;
6342 
6343 	vcpu->arch.cr0 |= X86_CR0_ET;
6344 
6345 	return 0;
6346 }
6347 EXPORT_SYMBOL_GPL(fx_init);
6348 
6349 static void fx_free(struct kvm_vcpu *vcpu)
6350 {
6351 	fpu_free(&vcpu->arch.guest_fpu);
6352 }
6353 
6354 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
6355 {
6356 	if (vcpu->guest_fpu_loaded)
6357 		return;
6358 
6359 	/*
6360 	 * Restore all possible states in the guest,
6361 	 * and assume host would use all available bits.
6362 	 * Guest xcr0 would be loaded later.
6363 	 */
6364 	kvm_put_guest_xcr0(vcpu);
6365 	vcpu->guest_fpu_loaded = 1;
6366 	__kernel_fpu_begin();
6367 	fpu_restore_checking(&vcpu->arch.guest_fpu);
6368 	trace_kvm_fpu(1);
6369 }
6370 
6371 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
6372 {
6373 	kvm_put_guest_xcr0(vcpu);
6374 
6375 	if (!vcpu->guest_fpu_loaded)
6376 		return;
6377 
6378 	vcpu->guest_fpu_loaded = 0;
6379 	fpu_save_init(&vcpu->arch.guest_fpu);
6380 	__kernel_fpu_end();
6381 	++vcpu->stat.fpu_reload;
6382 	kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu);
6383 	trace_kvm_fpu(0);
6384 }
6385 
6386 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
6387 {
6388 	kvmclock_reset(vcpu);
6389 
6390 	free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
6391 	fx_free(vcpu);
6392 	kvm_x86_ops->vcpu_free(vcpu);
6393 }
6394 
6395 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
6396 						unsigned int id)
6397 {
6398 	if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
6399 		printk_once(KERN_WARNING
6400 		"kvm: SMP vm created on host with unstable TSC; "
6401 		"guest TSC will not be reliable\n");
6402 	return kvm_x86_ops->vcpu_create(kvm, id);
6403 }
6404 
6405 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
6406 {
6407 	int r;
6408 
6409 	vcpu->arch.mtrr_state.have_fixed = 1;
6410 	r = vcpu_load(vcpu);
6411 	if (r)
6412 		return r;
6413 	r = kvm_vcpu_reset(vcpu);
6414 	if (r == 0)
6415 		r = kvm_mmu_setup(vcpu);
6416 	vcpu_put(vcpu);
6417 
6418 	return r;
6419 }
6420 
6421 int kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
6422 {
6423 	int r;
6424 	struct msr_data msr;
6425 
6426 	r = vcpu_load(vcpu);
6427 	if (r)
6428 		return r;
6429 	msr.data = 0x0;
6430 	msr.index = MSR_IA32_TSC;
6431 	msr.host_initiated = true;
6432 	kvm_write_tsc(vcpu, &msr);
6433 	vcpu_put(vcpu);
6434 
6435 	return r;
6436 }
6437 
6438 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
6439 {
6440 	int r;
6441 	vcpu->arch.apf.msr_val = 0;
6442 
6443 	r = vcpu_load(vcpu);
6444 	BUG_ON(r);
6445 	kvm_mmu_unload(vcpu);
6446 	vcpu_put(vcpu);
6447 
6448 	fx_free(vcpu);
6449 	kvm_x86_ops->vcpu_free(vcpu);
6450 }
6451 
6452 static int kvm_vcpu_reset(struct kvm_vcpu *vcpu)
6453 {
6454 	atomic_set(&vcpu->arch.nmi_queued, 0);
6455 	vcpu->arch.nmi_pending = 0;
6456 	vcpu->arch.nmi_injected = false;
6457 
6458 	memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
6459 	vcpu->arch.dr6 = DR6_FIXED_1;
6460 	vcpu->arch.dr7 = DR7_FIXED_1;
6461 	kvm_update_dr7(vcpu);
6462 
6463 	kvm_make_request(KVM_REQ_EVENT, vcpu);
6464 	vcpu->arch.apf.msr_val = 0;
6465 	vcpu->arch.st.msr_val = 0;
6466 
6467 	kvmclock_reset(vcpu);
6468 
6469 	kvm_clear_async_pf_completion_queue(vcpu);
6470 	kvm_async_pf_hash_reset(vcpu);
6471 	vcpu->arch.apf.halted = false;
6472 
6473 	kvm_pmu_reset(vcpu);
6474 
6475 	memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
6476 	vcpu->arch.regs_avail = ~0;
6477 	vcpu->arch.regs_dirty = ~0;
6478 
6479 	return kvm_x86_ops->vcpu_reset(vcpu);
6480 }
6481 
6482 int kvm_arch_hardware_enable(void *garbage)
6483 {
6484 	struct kvm *kvm;
6485 	struct kvm_vcpu *vcpu;
6486 	int i;
6487 	int ret;
6488 	u64 local_tsc;
6489 	u64 max_tsc = 0;
6490 	bool stable, backwards_tsc = false;
6491 
6492 	kvm_shared_msr_cpu_online();
6493 	ret = kvm_x86_ops->hardware_enable(garbage);
6494 	if (ret != 0)
6495 		return ret;
6496 
6497 	local_tsc = native_read_tsc();
6498 	stable = !check_tsc_unstable();
6499 	list_for_each_entry(kvm, &vm_list, vm_list) {
6500 		kvm_for_each_vcpu(i, vcpu, kvm) {
6501 			if (!stable && vcpu->cpu == smp_processor_id())
6502 				set_bit(KVM_REQ_CLOCK_UPDATE, &vcpu->requests);
6503 			if (stable && vcpu->arch.last_host_tsc > local_tsc) {
6504 				backwards_tsc = true;
6505 				if (vcpu->arch.last_host_tsc > max_tsc)
6506 					max_tsc = vcpu->arch.last_host_tsc;
6507 			}
6508 		}
6509 	}
6510 
6511 	/*
6512 	 * Sometimes, even reliable TSCs go backwards.  This happens on
6513 	 * platforms that reset TSC during suspend or hibernate actions, but
6514 	 * maintain synchronization.  We must compensate.  Fortunately, we can
6515 	 * detect that condition here, which happens early in CPU bringup,
6516 	 * before any KVM threads can be running.  Unfortunately, we can't
6517 	 * bring the TSCs fully up to date with real time, as we aren't yet far
6518 	 * enough into CPU bringup that we know how much real time has actually
6519 	 * elapsed; our helper function, get_kernel_ns() will be using boot
6520 	 * variables that haven't been updated yet.
6521 	 *
6522 	 * So we simply find the maximum observed TSC above, then record the
6523 	 * adjustment to TSC in each VCPU.  When the VCPU later gets loaded,
6524 	 * the adjustment will be applied.  Note that we accumulate
6525 	 * adjustments, in case multiple suspend cycles happen before some VCPU
6526 	 * gets a chance to run again.  In the event that no KVM threads get a
6527 	 * chance to run, we will miss the entire elapsed period, as we'll have
6528 	 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
6529 	 * loose cycle time.  This isn't too big a deal, since the loss will be
6530 	 * uniform across all VCPUs (not to mention the scenario is extremely
6531 	 * unlikely). It is possible that a second hibernate recovery happens
6532 	 * much faster than a first, causing the observed TSC here to be
6533 	 * smaller; this would require additional padding adjustment, which is
6534 	 * why we set last_host_tsc to the local tsc observed here.
6535 	 *
6536 	 * N.B. - this code below runs only on platforms with reliable TSC,
6537 	 * as that is the only way backwards_tsc is set above.  Also note
6538 	 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
6539 	 * have the same delta_cyc adjustment applied if backwards_tsc
6540 	 * is detected.  Note further, this adjustment is only done once,
6541 	 * as we reset last_host_tsc on all VCPUs to stop this from being
6542 	 * called multiple times (one for each physical CPU bringup).
6543 	 *
6544 	 * Platforms with unreliable TSCs don't have to deal with this, they
6545 	 * will be compensated by the logic in vcpu_load, which sets the TSC to
6546 	 * catchup mode.  This will catchup all VCPUs to real time, but cannot
6547 	 * guarantee that they stay in perfect synchronization.
6548 	 */
6549 	if (backwards_tsc) {
6550 		u64 delta_cyc = max_tsc - local_tsc;
6551 		list_for_each_entry(kvm, &vm_list, vm_list) {
6552 			kvm_for_each_vcpu(i, vcpu, kvm) {
6553 				vcpu->arch.tsc_offset_adjustment += delta_cyc;
6554 				vcpu->arch.last_host_tsc = local_tsc;
6555 				set_bit(KVM_REQ_MASTERCLOCK_UPDATE,
6556 					&vcpu->requests);
6557 			}
6558 
6559 			/*
6560 			 * We have to disable TSC offset matching.. if you were
6561 			 * booting a VM while issuing an S4 host suspend....
6562 			 * you may have some problem.  Solving this issue is
6563 			 * left as an exercise to the reader.
6564 			 */
6565 			kvm->arch.last_tsc_nsec = 0;
6566 			kvm->arch.last_tsc_write = 0;
6567 		}
6568 
6569 	}
6570 	return 0;
6571 }
6572 
6573 void kvm_arch_hardware_disable(void *garbage)
6574 {
6575 	kvm_x86_ops->hardware_disable(garbage);
6576 	drop_user_return_notifiers(garbage);
6577 }
6578 
6579 int kvm_arch_hardware_setup(void)
6580 {
6581 	return kvm_x86_ops->hardware_setup();
6582 }
6583 
6584 void kvm_arch_hardware_unsetup(void)
6585 {
6586 	kvm_x86_ops->hardware_unsetup();
6587 }
6588 
6589 void kvm_arch_check_processor_compat(void *rtn)
6590 {
6591 	kvm_x86_ops->check_processor_compatibility(rtn);
6592 }
6593 
6594 bool kvm_vcpu_compatible(struct kvm_vcpu *vcpu)
6595 {
6596 	return irqchip_in_kernel(vcpu->kvm) == (vcpu->arch.apic != NULL);
6597 }
6598 
6599 struct static_key kvm_no_apic_vcpu __read_mostly;
6600 
6601 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
6602 {
6603 	struct page *page;
6604 	struct kvm *kvm;
6605 	int r;
6606 
6607 	BUG_ON(vcpu->kvm == NULL);
6608 	kvm = vcpu->kvm;
6609 
6610 	vcpu->arch.emulate_ctxt.ops = &emulate_ops;
6611 	if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
6612 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
6613 	else
6614 		vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
6615 
6616 	page = alloc_page(GFP_KERNEL | __GFP_ZERO);
6617 	if (!page) {
6618 		r = -ENOMEM;
6619 		goto fail;
6620 	}
6621 	vcpu->arch.pio_data = page_address(page);
6622 
6623 	kvm_set_tsc_khz(vcpu, max_tsc_khz);
6624 
6625 	r = kvm_mmu_create(vcpu);
6626 	if (r < 0)
6627 		goto fail_free_pio_data;
6628 
6629 	if (irqchip_in_kernel(kvm)) {
6630 		r = kvm_create_lapic(vcpu);
6631 		if (r < 0)
6632 			goto fail_mmu_destroy;
6633 	} else
6634 		static_key_slow_inc(&kvm_no_apic_vcpu);
6635 
6636 	vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
6637 				       GFP_KERNEL);
6638 	if (!vcpu->arch.mce_banks) {
6639 		r = -ENOMEM;
6640 		goto fail_free_lapic;
6641 	}
6642 	vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
6643 
6644 	if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL))
6645 		goto fail_free_mce_banks;
6646 
6647 	r = fx_init(vcpu);
6648 	if (r)
6649 		goto fail_free_wbinvd_dirty_mask;
6650 
6651 	vcpu->arch.ia32_tsc_adjust_msr = 0x0;
6652 	kvm_async_pf_hash_reset(vcpu);
6653 	kvm_pmu_init(vcpu);
6654 
6655 	return 0;
6656 fail_free_wbinvd_dirty_mask:
6657 	free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
6658 fail_free_mce_banks:
6659 	kfree(vcpu->arch.mce_banks);
6660 fail_free_lapic:
6661 	kvm_free_lapic(vcpu);
6662 fail_mmu_destroy:
6663 	kvm_mmu_destroy(vcpu);
6664 fail_free_pio_data:
6665 	free_page((unsigned long)vcpu->arch.pio_data);
6666 fail:
6667 	return r;
6668 }
6669 
6670 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
6671 {
6672 	int idx;
6673 
6674 	kvm_pmu_destroy(vcpu);
6675 	kfree(vcpu->arch.mce_banks);
6676 	kvm_free_lapic(vcpu);
6677 	idx = srcu_read_lock(&vcpu->kvm->srcu);
6678 	kvm_mmu_destroy(vcpu);
6679 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
6680 	free_page((unsigned long)vcpu->arch.pio_data);
6681 	if (!irqchip_in_kernel(vcpu->kvm))
6682 		static_key_slow_dec(&kvm_no_apic_vcpu);
6683 }
6684 
6685 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
6686 {
6687 	if (type)
6688 		return -EINVAL;
6689 
6690 	INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
6691 	INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
6692 
6693 	/* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
6694 	set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
6695 	/* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
6696 	set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
6697 		&kvm->arch.irq_sources_bitmap);
6698 
6699 	raw_spin_lock_init(&kvm->arch.tsc_write_lock);
6700 	mutex_init(&kvm->arch.apic_map_lock);
6701 	spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
6702 
6703 	pvclock_update_vm_gtod_copy(kvm);
6704 
6705 	return 0;
6706 }
6707 
6708 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
6709 {
6710 	int r;
6711 	r = vcpu_load(vcpu);
6712 	BUG_ON(r);
6713 	kvm_mmu_unload(vcpu);
6714 	vcpu_put(vcpu);
6715 }
6716 
6717 static void kvm_free_vcpus(struct kvm *kvm)
6718 {
6719 	unsigned int i;
6720 	struct kvm_vcpu *vcpu;
6721 
6722 	/*
6723 	 * Unpin any mmu pages first.
6724 	 */
6725 	kvm_for_each_vcpu(i, vcpu, kvm) {
6726 		kvm_clear_async_pf_completion_queue(vcpu);
6727 		kvm_unload_vcpu_mmu(vcpu);
6728 	}
6729 	kvm_for_each_vcpu(i, vcpu, kvm)
6730 		kvm_arch_vcpu_free(vcpu);
6731 
6732 	mutex_lock(&kvm->lock);
6733 	for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
6734 		kvm->vcpus[i] = NULL;
6735 
6736 	atomic_set(&kvm->online_vcpus, 0);
6737 	mutex_unlock(&kvm->lock);
6738 }
6739 
6740 void kvm_arch_sync_events(struct kvm *kvm)
6741 {
6742 	kvm_free_all_assigned_devices(kvm);
6743 	kvm_free_pit(kvm);
6744 }
6745 
6746 void kvm_arch_destroy_vm(struct kvm *kvm)
6747 {
6748 	kvm_iommu_unmap_guest(kvm);
6749 	kfree(kvm->arch.vpic);
6750 	kfree(kvm->arch.vioapic);
6751 	kvm_free_vcpus(kvm);
6752 	if (kvm->arch.apic_access_page)
6753 		put_page(kvm->arch.apic_access_page);
6754 	if (kvm->arch.ept_identity_pagetable)
6755 		put_page(kvm->arch.ept_identity_pagetable);
6756 	kfree(rcu_dereference_check(kvm->arch.apic_map, 1));
6757 }
6758 
6759 void kvm_arch_free_memslot(struct kvm_memory_slot *free,
6760 			   struct kvm_memory_slot *dont)
6761 {
6762 	int i;
6763 
6764 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
6765 		if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
6766 			kvm_kvfree(free->arch.rmap[i]);
6767 			free->arch.rmap[i] = NULL;
6768 		}
6769 		if (i == 0)
6770 			continue;
6771 
6772 		if (!dont || free->arch.lpage_info[i - 1] !=
6773 			     dont->arch.lpage_info[i - 1]) {
6774 			kvm_kvfree(free->arch.lpage_info[i - 1]);
6775 			free->arch.lpage_info[i - 1] = NULL;
6776 		}
6777 	}
6778 }
6779 
6780 int kvm_arch_create_memslot(struct kvm_memory_slot *slot, unsigned long npages)
6781 {
6782 	int i;
6783 
6784 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
6785 		unsigned long ugfn;
6786 		int lpages;
6787 		int level = i + 1;
6788 
6789 		lpages = gfn_to_index(slot->base_gfn + npages - 1,
6790 				      slot->base_gfn, level) + 1;
6791 
6792 		slot->arch.rmap[i] =
6793 			kvm_kvzalloc(lpages * sizeof(*slot->arch.rmap[i]));
6794 		if (!slot->arch.rmap[i])
6795 			goto out_free;
6796 		if (i == 0)
6797 			continue;
6798 
6799 		slot->arch.lpage_info[i - 1] = kvm_kvzalloc(lpages *
6800 					sizeof(*slot->arch.lpage_info[i - 1]));
6801 		if (!slot->arch.lpage_info[i - 1])
6802 			goto out_free;
6803 
6804 		if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
6805 			slot->arch.lpage_info[i - 1][0].write_count = 1;
6806 		if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
6807 			slot->arch.lpage_info[i - 1][lpages - 1].write_count = 1;
6808 		ugfn = slot->userspace_addr >> PAGE_SHIFT;
6809 		/*
6810 		 * If the gfn and userspace address are not aligned wrt each
6811 		 * other, or if explicitly asked to, disable large page
6812 		 * support for this slot
6813 		 */
6814 		if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
6815 		    !kvm_largepages_enabled()) {
6816 			unsigned long j;
6817 
6818 			for (j = 0; j < lpages; ++j)
6819 				slot->arch.lpage_info[i - 1][j].write_count = 1;
6820 		}
6821 	}
6822 
6823 	return 0;
6824 
6825 out_free:
6826 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
6827 		kvm_kvfree(slot->arch.rmap[i]);
6828 		slot->arch.rmap[i] = NULL;
6829 		if (i == 0)
6830 			continue;
6831 
6832 		kvm_kvfree(slot->arch.lpage_info[i - 1]);
6833 		slot->arch.lpage_info[i - 1] = NULL;
6834 	}
6835 	return -ENOMEM;
6836 }
6837 
6838 int kvm_arch_prepare_memory_region(struct kvm *kvm,
6839 				struct kvm_memory_slot *memslot,
6840 				struct kvm_memory_slot old,
6841 				struct kvm_userspace_memory_region *mem,
6842 				int user_alloc)
6843 {
6844 	int npages = memslot->npages;
6845 	int map_flags = MAP_PRIVATE | MAP_ANONYMOUS;
6846 
6847 	/* Prevent internal slot pages from being moved by fork()/COW. */
6848 	if (memslot->id >= KVM_MEMORY_SLOTS)
6849 		map_flags = MAP_SHARED | MAP_ANONYMOUS;
6850 
6851 	/*To keep backward compatibility with older userspace,
6852 	 *x86 needs to handle !user_alloc case.
6853 	 */
6854 	if (!user_alloc) {
6855 		if (npages && !old.npages) {
6856 			unsigned long userspace_addr;
6857 
6858 			userspace_addr = vm_mmap(NULL, 0,
6859 						 npages * PAGE_SIZE,
6860 						 PROT_READ | PROT_WRITE,
6861 						 map_flags,
6862 						 0);
6863 
6864 			if (IS_ERR((void *)userspace_addr))
6865 				return PTR_ERR((void *)userspace_addr);
6866 
6867 			memslot->userspace_addr = userspace_addr;
6868 		}
6869 	}
6870 
6871 
6872 	return 0;
6873 }
6874 
6875 void kvm_arch_commit_memory_region(struct kvm *kvm,
6876 				struct kvm_userspace_memory_region *mem,
6877 				struct kvm_memory_slot old,
6878 				int user_alloc)
6879 {
6880 
6881 	int nr_mmu_pages = 0, npages = mem->memory_size >> PAGE_SHIFT;
6882 
6883 	if (!user_alloc && !old.user_alloc && old.npages && !npages) {
6884 		int ret;
6885 
6886 		ret = vm_munmap(old.userspace_addr,
6887 				old.npages * PAGE_SIZE);
6888 		if (ret < 0)
6889 			printk(KERN_WARNING
6890 			       "kvm_vm_ioctl_set_memory_region: "
6891 			       "failed to munmap memory\n");
6892 	}
6893 
6894 	if (!kvm->arch.n_requested_mmu_pages)
6895 		nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
6896 
6897 	spin_lock(&kvm->mmu_lock);
6898 	if (nr_mmu_pages)
6899 		kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
6900 	kvm_mmu_slot_remove_write_access(kvm, mem->slot);
6901 	spin_unlock(&kvm->mmu_lock);
6902 	/*
6903 	 * If memory slot is created, or moved, we need to clear all
6904 	 * mmio sptes.
6905 	 */
6906 	if (npages && old.base_gfn != mem->guest_phys_addr >> PAGE_SHIFT) {
6907 		kvm_mmu_zap_all(kvm);
6908 		kvm_reload_remote_mmus(kvm);
6909 	}
6910 }
6911 
6912 void kvm_arch_flush_shadow_all(struct kvm *kvm)
6913 {
6914 	kvm_mmu_zap_all(kvm);
6915 	kvm_reload_remote_mmus(kvm);
6916 }
6917 
6918 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
6919 				   struct kvm_memory_slot *slot)
6920 {
6921 	kvm_arch_flush_shadow_all(kvm);
6922 }
6923 
6924 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
6925 {
6926 	return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
6927 		!vcpu->arch.apf.halted)
6928 		|| !list_empty_careful(&vcpu->async_pf.done)
6929 		|| vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
6930 		|| atomic_read(&vcpu->arch.nmi_queued) ||
6931 		(kvm_arch_interrupt_allowed(vcpu) &&
6932 		 kvm_cpu_has_interrupt(vcpu));
6933 }
6934 
6935 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
6936 {
6937 	return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
6938 }
6939 
6940 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
6941 {
6942 	return kvm_x86_ops->interrupt_allowed(vcpu);
6943 }
6944 
6945 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
6946 {
6947 	unsigned long current_rip = kvm_rip_read(vcpu) +
6948 		get_segment_base(vcpu, VCPU_SREG_CS);
6949 
6950 	return current_rip == linear_rip;
6951 }
6952 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
6953 
6954 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
6955 {
6956 	unsigned long rflags;
6957 
6958 	rflags = kvm_x86_ops->get_rflags(vcpu);
6959 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6960 		rflags &= ~X86_EFLAGS_TF;
6961 	return rflags;
6962 }
6963 EXPORT_SYMBOL_GPL(kvm_get_rflags);
6964 
6965 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
6966 {
6967 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
6968 	    kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
6969 		rflags |= X86_EFLAGS_TF;
6970 	kvm_x86_ops->set_rflags(vcpu, rflags);
6971 	kvm_make_request(KVM_REQ_EVENT, vcpu);
6972 }
6973 EXPORT_SYMBOL_GPL(kvm_set_rflags);
6974 
6975 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
6976 {
6977 	int r;
6978 
6979 	if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
6980 	      is_error_page(work->page))
6981 		return;
6982 
6983 	r = kvm_mmu_reload(vcpu);
6984 	if (unlikely(r))
6985 		return;
6986 
6987 	if (!vcpu->arch.mmu.direct_map &&
6988 	      work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
6989 		return;
6990 
6991 	vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
6992 }
6993 
6994 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
6995 {
6996 	return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
6997 }
6998 
6999 static inline u32 kvm_async_pf_next_probe(u32 key)
7000 {
7001 	return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
7002 }
7003 
7004 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7005 {
7006 	u32 key = kvm_async_pf_hash_fn(gfn);
7007 
7008 	while (vcpu->arch.apf.gfns[key] != ~0)
7009 		key = kvm_async_pf_next_probe(key);
7010 
7011 	vcpu->arch.apf.gfns[key] = gfn;
7012 }
7013 
7014 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
7015 {
7016 	int i;
7017 	u32 key = kvm_async_pf_hash_fn(gfn);
7018 
7019 	for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
7020 		     (vcpu->arch.apf.gfns[key] != gfn &&
7021 		      vcpu->arch.apf.gfns[key] != ~0); i++)
7022 		key = kvm_async_pf_next_probe(key);
7023 
7024 	return key;
7025 }
7026 
7027 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7028 {
7029 	return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
7030 }
7031 
7032 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7033 {
7034 	u32 i, j, k;
7035 
7036 	i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
7037 	while (true) {
7038 		vcpu->arch.apf.gfns[i] = ~0;
7039 		do {
7040 			j = kvm_async_pf_next_probe(j);
7041 			if (vcpu->arch.apf.gfns[j] == ~0)
7042 				return;
7043 			k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
7044 			/*
7045 			 * k lies cyclically in ]i,j]
7046 			 * |    i.k.j |
7047 			 * |....j i.k.| or  |.k..j i...|
7048 			 */
7049 		} while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
7050 		vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
7051 		i = j;
7052 	}
7053 }
7054 
7055 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
7056 {
7057 
7058 	return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
7059 				      sizeof(val));
7060 }
7061 
7062 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
7063 				     struct kvm_async_pf *work)
7064 {
7065 	struct x86_exception fault;
7066 
7067 	trace_kvm_async_pf_not_present(work->arch.token, work->gva);
7068 	kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
7069 
7070 	if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
7071 	    (vcpu->arch.apf.send_user_only &&
7072 	     kvm_x86_ops->get_cpl(vcpu) == 0))
7073 		kvm_make_request(KVM_REQ_APF_HALT, vcpu);
7074 	else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
7075 		fault.vector = PF_VECTOR;
7076 		fault.error_code_valid = true;
7077 		fault.error_code = 0;
7078 		fault.nested_page_fault = false;
7079 		fault.address = work->arch.token;
7080 		kvm_inject_page_fault(vcpu, &fault);
7081 	}
7082 }
7083 
7084 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
7085 				 struct kvm_async_pf *work)
7086 {
7087 	struct x86_exception fault;
7088 
7089 	trace_kvm_async_pf_ready(work->arch.token, work->gva);
7090 	if (is_error_page(work->page))
7091 		work->arch.token = ~0; /* broadcast wakeup */
7092 	else
7093 		kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
7094 
7095 	if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
7096 	    !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
7097 		fault.vector = PF_VECTOR;
7098 		fault.error_code_valid = true;
7099 		fault.error_code = 0;
7100 		fault.nested_page_fault = false;
7101 		fault.address = work->arch.token;
7102 		kvm_inject_page_fault(vcpu, &fault);
7103 	}
7104 	vcpu->arch.apf.halted = false;
7105 	vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7106 }
7107 
7108 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
7109 {
7110 	if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
7111 		return true;
7112 	else
7113 		return !kvm_event_needs_reinjection(vcpu) &&
7114 			kvm_x86_ops->interrupt_allowed(vcpu);
7115 }
7116 
7117 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
7118 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
7119 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
7120 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
7121 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
7122 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
7123 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
7124 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
7125 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
7126 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
7127 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
7128 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
7129