xref: /openbmc/linux/arch/x86/kvm/x86.c (revision babbdf5b)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Kernel-based Virtual Machine driver for Linux
4  *
5  * derived from drivers/kvm/kvm_main.c
6  *
7  * Copyright (C) 2006 Qumranet, Inc.
8  * Copyright (C) 2008 Qumranet, Inc.
9  * Copyright IBM Corporation, 2008
10  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
11  *
12  * Authors:
13  *   Avi Kivity   <avi@qumranet.com>
14  *   Yaniv Kamay  <yaniv@qumranet.com>
15  *   Amit Shah    <amit.shah@qumranet.com>
16  *   Ben-Ami Yassour <benami@il.ibm.com>
17  */
18 
19 #include <linux/kvm_host.h>
20 #include "irq.h"
21 #include "ioapic.h"
22 #include "mmu.h"
23 #include "i8254.h"
24 #include "tss.h"
25 #include "kvm_cache_regs.h"
26 #include "kvm_emulate.h"
27 #include "x86.h"
28 #include "cpuid.h"
29 #include "pmu.h"
30 #include "hyperv.h"
31 #include "lapic.h"
32 #include "xen.h"
33 
34 #include <linux/clocksource.h>
35 #include <linux/interrupt.h>
36 #include <linux/kvm.h>
37 #include <linux/fs.h>
38 #include <linux/vmalloc.h>
39 #include <linux/export.h>
40 #include <linux/moduleparam.h>
41 #include <linux/mman.h>
42 #include <linux/highmem.h>
43 #include <linux/iommu.h>
44 #include <linux/intel-iommu.h>
45 #include <linux/cpufreq.h>
46 #include <linux/user-return-notifier.h>
47 #include <linux/srcu.h>
48 #include <linux/slab.h>
49 #include <linux/perf_event.h>
50 #include <linux/uaccess.h>
51 #include <linux/hash.h>
52 #include <linux/pci.h>
53 #include <linux/timekeeper_internal.h>
54 #include <linux/pvclock_gtod.h>
55 #include <linux/kvm_irqfd.h>
56 #include <linux/irqbypass.h>
57 #include <linux/sched/stat.h>
58 #include <linux/sched/isolation.h>
59 #include <linux/mem_encrypt.h>
60 #include <linux/entry-kvm.h>
61 
62 #include <trace/events/kvm.h>
63 
64 #include <asm/debugreg.h>
65 #include <asm/msr.h>
66 #include <asm/desc.h>
67 #include <asm/mce.h>
68 #include <linux/kernel_stat.h>
69 #include <asm/fpu/internal.h> /* Ugh! */
70 #include <asm/pvclock.h>
71 #include <asm/div64.h>
72 #include <asm/irq_remapping.h>
73 #include <asm/mshyperv.h>
74 #include <asm/hypervisor.h>
75 #include <asm/tlbflush.h>
76 #include <asm/intel_pt.h>
77 #include <asm/emulate_prefix.h>
78 #include <asm/sgx.h>
79 #include <clocksource/hyperv_timer.h>
80 
81 #define CREATE_TRACE_POINTS
82 #include "trace.h"
83 
84 #define MAX_IO_MSRS 256
85 #define KVM_MAX_MCE_BANKS 32
86 u64 __read_mostly kvm_mce_cap_supported = MCG_CTL_P | MCG_SER_P;
87 EXPORT_SYMBOL_GPL(kvm_mce_cap_supported);
88 
89 #define emul_to_vcpu(ctxt) \
90 	((struct kvm_vcpu *)(ctxt)->vcpu)
91 
92 /* EFER defaults:
93  * - enable syscall per default because its emulated by KVM
94  * - enable LME and LMA per default on 64 bit KVM
95  */
96 #ifdef CONFIG_X86_64
97 static
98 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
99 #else
100 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
101 #endif
102 
103 static u64 __read_mostly cr4_reserved_bits = CR4_RESERVED_BITS;
104 
105 #define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \
106                                     KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
107 
108 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
109 static void process_nmi(struct kvm_vcpu *vcpu);
110 static void process_smi(struct kvm_vcpu *vcpu);
111 static void enter_smm(struct kvm_vcpu *vcpu);
112 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
113 static void store_regs(struct kvm_vcpu *vcpu);
114 static int sync_regs(struct kvm_vcpu *vcpu);
115 
116 struct kvm_x86_ops kvm_x86_ops __read_mostly;
117 EXPORT_SYMBOL_GPL(kvm_x86_ops);
118 
119 #define KVM_X86_OP(func)					     \
120 	DEFINE_STATIC_CALL_NULL(kvm_x86_##func,			     \
121 				*(((struct kvm_x86_ops *)0)->func));
122 #define KVM_X86_OP_NULL KVM_X86_OP
123 #include <asm/kvm-x86-ops.h>
124 EXPORT_STATIC_CALL_GPL(kvm_x86_get_cs_db_l_bits);
125 EXPORT_STATIC_CALL_GPL(kvm_x86_cache_reg);
126 EXPORT_STATIC_CALL_GPL(kvm_x86_tlb_flush_current);
127 
128 static bool __read_mostly ignore_msrs = 0;
129 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
130 
131 bool __read_mostly report_ignored_msrs = true;
132 module_param(report_ignored_msrs, bool, S_IRUGO | S_IWUSR);
133 EXPORT_SYMBOL_GPL(report_ignored_msrs);
134 
135 unsigned int min_timer_period_us = 200;
136 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
137 
138 static bool __read_mostly kvmclock_periodic_sync = true;
139 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
140 
141 bool __read_mostly kvm_has_tsc_control;
142 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
143 u32  __read_mostly kvm_max_guest_tsc_khz;
144 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
145 u8   __read_mostly kvm_tsc_scaling_ratio_frac_bits;
146 EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
147 u64  __read_mostly kvm_max_tsc_scaling_ratio;
148 EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
149 u64 __read_mostly kvm_default_tsc_scaling_ratio;
150 EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio);
151 bool __read_mostly kvm_has_bus_lock_exit;
152 EXPORT_SYMBOL_GPL(kvm_has_bus_lock_exit);
153 
154 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
155 static u32 __read_mostly tsc_tolerance_ppm = 250;
156 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
157 
158 /*
159  * lapic timer advance (tscdeadline mode only) in nanoseconds.  '-1' enables
160  * adaptive tuning starting from default advancement of 1000ns.  '0' disables
161  * advancement entirely.  Any other value is used as-is and disables adaptive
162  * tuning, i.e. allows privileged userspace to set an exact advancement time.
163  */
164 static int __read_mostly lapic_timer_advance_ns = -1;
165 module_param(lapic_timer_advance_ns, int, S_IRUGO | S_IWUSR);
166 
167 static bool __read_mostly vector_hashing = true;
168 module_param(vector_hashing, bool, S_IRUGO);
169 
170 bool __read_mostly enable_vmware_backdoor = false;
171 module_param(enable_vmware_backdoor, bool, S_IRUGO);
172 EXPORT_SYMBOL_GPL(enable_vmware_backdoor);
173 
174 static bool __read_mostly force_emulation_prefix = false;
175 module_param(force_emulation_prefix, bool, S_IRUGO);
176 
177 int __read_mostly pi_inject_timer = -1;
178 module_param(pi_inject_timer, bint, S_IRUGO | S_IWUSR);
179 
180 /*
181  * Restoring the host value for MSRs that are only consumed when running in
182  * usermode, e.g. SYSCALL MSRs and TSC_AUX, can be deferred until the CPU
183  * returns to userspace, i.e. the kernel can run with the guest's value.
184  */
185 #define KVM_MAX_NR_USER_RETURN_MSRS 16
186 
187 struct kvm_user_return_msrs {
188 	struct user_return_notifier urn;
189 	bool registered;
190 	struct kvm_user_return_msr_values {
191 		u64 host;
192 		u64 curr;
193 	} values[KVM_MAX_NR_USER_RETURN_MSRS];
194 };
195 
196 u32 __read_mostly kvm_nr_uret_msrs;
197 EXPORT_SYMBOL_GPL(kvm_nr_uret_msrs);
198 static u32 __read_mostly kvm_uret_msrs_list[KVM_MAX_NR_USER_RETURN_MSRS];
199 static struct kvm_user_return_msrs __percpu *user_return_msrs;
200 
201 #define KVM_SUPPORTED_XCR0     (XFEATURE_MASK_FP | XFEATURE_MASK_SSE \
202 				| XFEATURE_MASK_YMM | XFEATURE_MASK_BNDREGS \
203 				| XFEATURE_MASK_BNDCSR | XFEATURE_MASK_AVX512 \
204 				| XFEATURE_MASK_PKRU)
205 
206 u64 __read_mostly host_efer;
207 EXPORT_SYMBOL_GPL(host_efer);
208 
209 bool __read_mostly allow_smaller_maxphyaddr = 0;
210 EXPORT_SYMBOL_GPL(allow_smaller_maxphyaddr);
211 
212 u64 __read_mostly host_xss;
213 EXPORT_SYMBOL_GPL(host_xss);
214 u64 __read_mostly supported_xss;
215 EXPORT_SYMBOL_GPL(supported_xss);
216 
217 struct kvm_stats_debugfs_item debugfs_entries[] = {
218 	VCPU_STAT("pf_fixed", pf_fixed),
219 	VCPU_STAT("pf_guest", pf_guest),
220 	VCPU_STAT("tlb_flush", tlb_flush),
221 	VCPU_STAT("invlpg", invlpg),
222 	VCPU_STAT("exits", exits),
223 	VCPU_STAT("io_exits", io_exits),
224 	VCPU_STAT("mmio_exits", mmio_exits),
225 	VCPU_STAT("signal_exits", signal_exits),
226 	VCPU_STAT("irq_window", irq_window_exits),
227 	VCPU_STAT("nmi_window", nmi_window_exits),
228 	VCPU_STAT("halt_exits", halt_exits),
229 	VCPU_STAT("halt_successful_poll", halt_successful_poll),
230 	VCPU_STAT("halt_attempted_poll", halt_attempted_poll),
231 	VCPU_STAT("halt_poll_invalid", halt_poll_invalid),
232 	VCPU_STAT("halt_wakeup", halt_wakeup),
233 	VCPU_STAT("hypercalls", hypercalls),
234 	VCPU_STAT("request_irq", request_irq_exits),
235 	VCPU_STAT("irq_exits", irq_exits),
236 	VCPU_STAT("host_state_reload", host_state_reload),
237 	VCPU_STAT("fpu_reload", fpu_reload),
238 	VCPU_STAT("insn_emulation", insn_emulation),
239 	VCPU_STAT("insn_emulation_fail", insn_emulation_fail),
240 	VCPU_STAT("irq_injections", irq_injections),
241 	VCPU_STAT("nmi_injections", nmi_injections),
242 	VCPU_STAT("req_event", req_event),
243 	VCPU_STAT("l1d_flush", l1d_flush),
244 	VCPU_STAT("halt_poll_success_ns", halt_poll_success_ns),
245 	VCPU_STAT("halt_poll_fail_ns", halt_poll_fail_ns),
246 	VCPU_STAT("nested_run", nested_run),
247 	VCPU_STAT("directed_yield_attempted", directed_yield_attempted),
248 	VCPU_STAT("directed_yield_successful", directed_yield_successful),
249 	VM_STAT("mmu_shadow_zapped", mmu_shadow_zapped),
250 	VM_STAT("mmu_pte_write", mmu_pte_write),
251 	VM_STAT("mmu_pde_zapped", mmu_pde_zapped),
252 	VM_STAT("mmu_flooded", mmu_flooded),
253 	VM_STAT("mmu_recycled", mmu_recycled),
254 	VM_STAT("mmu_cache_miss", mmu_cache_miss),
255 	VM_STAT("mmu_unsync", mmu_unsync),
256 	VM_STAT("remote_tlb_flush", remote_tlb_flush),
257 	VM_STAT("largepages", lpages, .mode = 0444),
258 	VM_STAT("nx_largepages_splitted", nx_lpage_splits, .mode = 0444),
259 	VM_STAT("max_mmu_page_hash_collisions", max_mmu_page_hash_collisions),
260 	{ NULL }
261 };
262 
263 u64 __read_mostly host_xcr0;
264 u64 __read_mostly supported_xcr0;
265 EXPORT_SYMBOL_GPL(supported_xcr0);
266 
267 static struct kmem_cache *x86_fpu_cache;
268 
269 static struct kmem_cache *x86_emulator_cache;
270 
271 /*
272  * When called, it means the previous get/set msr reached an invalid msr.
273  * Return true if we want to ignore/silent this failed msr access.
274  */
275 static bool kvm_msr_ignored_check(u32 msr, u64 data, bool write)
276 {
277 	const char *op = write ? "wrmsr" : "rdmsr";
278 
279 	if (ignore_msrs) {
280 		if (report_ignored_msrs)
281 			kvm_pr_unimpl("ignored %s: 0x%x data 0x%llx\n",
282 				      op, msr, data);
283 		/* Mask the error */
284 		return true;
285 	} else {
286 		kvm_debug_ratelimited("unhandled %s: 0x%x data 0x%llx\n",
287 				      op, msr, data);
288 		return false;
289 	}
290 }
291 
292 static struct kmem_cache *kvm_alloc_emulator_cache(void)
293 {
294 	unsigned int useroffset = offsetof(struct x86_emulate_ctxt, src);
295 	unsigned int size = sizeof(struct x86_emulate_ctxt);
296 
297 	return kmem_cache_create_usercopy("x86_emulator", size,
298 					  __alignof__(struct x86_emulate_ctxt),
299 					  SLAB_ACCOUNT, useroffset,
300 					  size - useroffset, NULL);
301 }
302 
303 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
304 
305 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
306 {
307 	int i;
308 	for (i = 0; i < ASYNC_PF_PER_VCPU; i++)
309 		vcpu->arch.apf.gfns[i] = ~0;
310 }
311 
312 static void kvm_on_user_return(struct user_return_notifier *urn)
313 {
314 	unsigned slot;
315 	struct kvm_user_return_msrs *msrs
316 		= container_of(urn, struct kvm_user_return_msrs, urn);
317 	struct kvm_user_return_msr_values *values;
318 	unsigned long flags;
319 
320 	/*
321 	 * Disabling irqs at this point since the following code could be
322 	 * interrupted and executed through kvm_arch_hardware_disable()
323 	 */
324 	local_irq_save(flags);
325 	if (msrs->registered) {
326 		msrs->registered = false;
327 		user_return_notifier_unregister(urn);
328 	}
329 	local_irq_restore(flags);
330 	for (slot = 0; slot < kvm_nr_uret_msrs; ++slot) {
331 		values = &msrs->values[slot];
332 		if (values->host != values->curr) {
333 			wrmsrl(kvm_uret_msrs_list[slot], values->host);
334 			values->curr = values->host;
335 		}
336 	}
337 }
338 
339 static int kvm_probe_user_return_msr(u32 msr)
340 {
341 	u64 val;
342 	int ret;
343 
344 	preempt_disable();
345 	ret = rdmsrl_safe(msr, &val);
346 	if (ret)
347 		goto out;
348 	ret = wrmsrl_safe(msr, val);
349 out:
350 	preempt_enable();
351 	return ret;
352 }
353 
354 int kvm_add_user_return_msr(u32 msr)
355 {
356 	BUG_ON(kvm_nr_uret_msrs >= KVM_MAX_NR_USER_RETURN_MSRS);
357 
358 	if (kvm_probe_user_return_msr(msr))
359 		return -1;
360 
361 	kvm_uret_msrs_list[kvm_nr_uret_msrs] = msr;
362 	return kvm_nr_uret_msrs++;
363 }
364 EXPORT_SYMBOL_GPL(kvm_add_user_return_msr);
365 
366 int kvm_find_user_return_msr(u32 msr)
367 {
368 	int i;
369 
370 	for (i = 0; i < kvm_nr_uret_msrs; ++i) {
371 		if (kvm_uret_msrs_list[i] == msr)
372 			return i;
373 	}
374 	return -1;
375 }
376 EXPORT_SYMBOL_GPL(kvm_find_user_return_msr);
377 
378 static void kvm_user_return_msr_cpu_online(void)
379 {
380 	unsigned int cpu = smp_processor_id();
381 	struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
382 	u64 value;
383 	int i;
384 
385 	for (i = 0; i < kvm_nr_uret_msrs; ++i) {
386 		rdmsrl_safe(kvm_uret_msrs_list[i], &value);
387 		msrs->values[i].host = value;
388 		msrs->values[i].curr = value;
389 	}
390 }
391 
392 int kvm_set_user_return_msr(unsigned slot, u64 value, u64 mask)
393 {
394 	unsigned int cpu = smp_processor_id();
395 	struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
396 	int err;
397 
398 	value = (value & mask) | (msrs->values[slot].host & ~mask);
399 	if (value == msrs->values[slot].curr)
400 		return 0;
401 	err = wrmsrl_safe(kvm_uret_msrs_list[slot], value);
402 	if (err)
403 		return 1;
404 
405 	msrs->values[slot].curr = value;
406 	if (!msrs->registered) {
407 		msrs->urn.on_user_return = kvm_on_user_return;
408 		user_return_notifier_register(&msrs->urn);
409 		msrs->registered = true;
410 	}
411 	return 0;
412 }
413 EXPORT_SYMBOL_GPL(kvm_set_user_return_msr);
414 
415 static void drop_user_return_notifiers(void)
416 {
417 	unsigned int cpu = smp_processor_id();
418 	struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
419 
420 	if (msrs->registered)
421 		kvm_on_user_return(&msrs->urn);
422 }
423 
424 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
425 {
426 	return vcpu->arch.apic_base;
427 }
428 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
429 
430 enum lapic_mode kvm_get_apic_mode(struct kvm_vcpu *vcpu)
431 {
432 	return kvm_apic_mode(kvm_get_apic_base(vcpu));
433 }
434 EXPORT_SYMBOL_GPL(kvm_get_apic_mode);
435 
436 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
437 {
438 	enum lapic_mode old_mode = kvm_get_apic_mode(vcpu);
439 	enum lapic_mode new_mode = kvm_apic_mode(msr_info->data);
440 	u64 reserved_bits = kvm_vcpu_reserved_gpa_bits_raw(vcpu) | 0x2ff |
441 		(guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
442 
443 	if ((msr_info->data & reserved_bits) != 0 || new_mode == LAPIC_MODE_INVALID)
444 		return 1;
445 	if (!msr_info->host_initiated) {
446 		if (old_mode == LAPIC_MODE_X2APIC && new_mode == LAPIC_MODE_XAPIC)
447 			return 1;
448 		if (old_mode == LAPIC_MODE_DISABLED && new_mode == LAPIC_MODE_X2APIC)
449 			return 1;
450 	}
451 
452 	kvm_lapic_set_base(vcpu, msr_info->data);
453 	kvm_recalculate_apic_map(vcpu->kvm);
454 	return 0;
455 }
456 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
457 
458 asmlinkage __visible noinstr void kvm_spurious_fault(void)
459 {
460 	/* Fault while not rebooting.  We want the trace. */
461 	BUG_ON(!kvm_rebooting);
462 }
463 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
464 
465 #define EXCPT_BENIGN		0
466 #define EXCPT_CONTRIBUTORY	1
467 #define EXCPT_PF		2
468 
469 static int exception_class(int vector)
470 {
471 	switch (vector) {
472 	case PF_VECTOR:
473 		return EXCPT_PF;
474 	case DE_VECTOR:
475 	case TS_VECTOR:
476 	case NP_VECTOR:
477 	case SS_VECTOR:
478 	case GP_VECTOR:
479 		return EXCPT_CONTRIBUTORY;
480 	default:
481 		break;
482 	}
483 	return EXCPT_BENIGN;
484 }
485 
486 #define EXCPT_FAULT		0
487 #define EXCPT_TRAP		1
488 #define EXCPT_ABORT		2
489 #define EXCPT_INTERRUPT		3
490 
491 static int exception_type(int vector)
492 {
493 	unsigned int mask;
494 
495 	if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
496 		return EXCPT_INTERRUPT;
497 
498 	mask = 1 << vector;
499 
500 	/* #DB is trap, as instruction watchpoints are handled elsewhere */
501 	if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
502 		return EXCPT_TRAP;
503 
504 	if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
505 		return EXCPT_ABORT;
506 
507 	/* Reserved exceptions will result in fault */
508 	return EXCPT_FAULT;
509 }
510 
511 void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu)
512 {
513 	unsigned nr = vcpu->arch.exception.nr;
514 	bool has_payload = vcpu->arch.exception.has_payload;
515 	unsigned long payload = vcpu->arch.exception.payload;
516 
517 	if (!has_payload)
518 		return;
519 
520 	switch (nr) {
521 	case DB_VECTOR:
522 		/*
523 		 * "Certain debug exceptions may clear bit 0-3.  The
524 		 * remaining contents of the DR6 register are never
525 		 * cleared by the processor".
526 		 */
527 		vcpu->arch.dr6 &= ~DR_TRAP_BITS;
528 		/*
529 		 * In order to reflect the #DB exception payload in guest
530 		 * dr6, three components need to be considered: active low
531 		 * bit, FIXED_1 bits and active high bits (e.g. DR6_BD,
532 		 * DR6_BS and DR6_BT)
533 		 * DR6_ACTIVE_LOW contains the FIXED_1 and active low bits.
534 		 * In the target guest dr6:
535 		 * FIXED_1 bits should always be set.
536 		 * Active low bits should be cleared if 1-setting in payload.
537 		 * Active high bits should be set if 1-setting in payload.
538 		 *
539 		 * Note, the payload is compatible with the pending debug
540 		 * exceptions/exit qualification under VMX, that active_low bits
541 		 * are active high in payload.
542 		 * So they need to be flipped for DR6.
543 		 */
544 		vcpu->arch.dr6 |= DR6_ACTIVE_LOW;
545 		vcpu->arch.dr6 |= payload;
546 		vcpu->arch.dr6 ^= payload & DR6_ACTIVE_LOW;
547 
548 		/*
549 		 * The #DB payload is defined as compatible with the 'pending
550 		 * debug exceptions' field under VMX, not DR6. While bit 12 is
551 		 * defined in the 'pending debug exceptions' field (enabled
552 		 * breakpoint), it is reserved and must be zero in DR6.
553 		 */
554 		vcpu->arch.dr6 &= ~BIT(12);
555 		break;
556 	case PF_VECTOR:
557 		vcpu->arch.cr2 = payload;
558 		break;
559 	}
560 
561 	vcpu->arch.exception.has_payload = false;
562 	vcpu->arch.exception.payload = 0;
563 }
564 EXPORT_SYMBOL_GPL(kvm_deliver_exception_payload);
565 
566 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
567 		unsigned nr, bool has_error, u32 error_code,
568 	        bool has_payload, unsigned long payload, bool reinject)
569 {
570 	u32 prev_nr;
571 	int class1, class2;
572 
573 	kvm_make_request(KVM_REQ_EVENT, vcpu);
574 
575 	if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) {
576 	queue:
577 		if (reinject) {
578 			/*
579 			 * On vmentry, vcpu->arch.exception.pending is only
580 			 * true if an event injection was blocked by
581 			 * nested_run_pending.  In that case, however,
582 			 * vcpu_enter_guest requests an immediate exit,
583 			 * and the guest shouldn't proceed far enough to
584 			 * need reinjection.
585 			 */
586 			WARN_ON_ONCE(vcpu->arch.exception.pending);
587 			vcpu->arch.exception.injected = true;
588 			if (WARN_ON_ONCE(has_payload)) {
589 				/*
590 				 * A reinjected event has already
591 				 * delivered its payload.
592 				 */
593 				has_payload = false;
594 				payload = 0;
595 			}
596 		} else {
597 			vcpu->arch.exception.pending = true;
598 			vcpu->arch.exception.injected = false;
599 		}
600 		vcpu->arch.exception.has_error_code = has_error;
601 		vcpu->arch.exception.nr = nr;
602 		vcpu->arch.exception.error_code = error_code;
603 		vcpu->arch.exception.has_payload = has_payload;
604 		vcpu->arch.exception.payload = payload;
605 		if (!is_guest_mode(vcpu))
606 			kvm_deliver_exception_payload(vcpu);
607 		return;
608 	}
609 
610 	/* to check exception */
611 	prev_nr = vcpu->arch.exception.nr;
612 	if (prev_nr == DF_VECTOR) {
613 		/* triple fault -> shutdown */
614 		kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
615 		return;
616 	}
617 	class1 = exception_class(prev_nr);
618 	class2 = exception_class(nr);
619 	if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
620 		|| (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
621 		/*
622 		 * Generate double fault per SDM Table 5-5.  Set
623 		 * exception.pending = true so that the double fault
624 		 * can trigger a nested vmexit.
625 		 */
626 		vcpu->arch.exception.pending = true;
627 		vcpu->arch.exception.injected = false;
628 		vcpu->arch.exception.has_error_code = true;
629 		vcpu->arch.exception.nr = DF_VECTOR;
630 		vcpu->arch.exception.error_code = 0;
631 		vcpu->arch.exception.has_payload = false;
632 		vcpu->arch.exception.payload = 0;
633 	} else
634 		/* replace previous exception with a new one in a hope
635 		   that instruction re-execution will regenerate lost
636 		   exception */
637 		goto queue;
638 }
639 
640 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
641 {
642 	kvm_multiple_exception(vcpu, nr, false, 0, false, 0, false);
643 }
644 EXPORT_SYMBOL_GPL(kvm_queue_exception);
645 
646 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
647 {
648 	kvm_multiple_exception(vcpu, nr, false, 0, false, 0, true);
649 }
650 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
651 
652 void kvm_queue_exception_p(struct kvm_vcpu *vcpu, unsigned nr,
653 			   unsigned long payload)
654 {
655 	kvm_multiple_exception(vcpu, nr, false, 0, true, payload, false);
656 }
657 EXPORT_SYMBOL_GPL(kvm_queue_exception_p);
658 
659 static void kvm_queue_exception_e_p(struct kvm_vcpu *vcpu, unsigned nr,
660 				    u32 error_code, unsigned long payload)
661 {
662 	kvm_multiple_exception(vcpu, nr, true, error_code,
663 			       true, payload, false);
664 }
665 
666 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
667 {
668 	if (err)
669 		kvm_inject_gp(vcpu, 0);
670 	else
671 		return kvm_skip_emulated_instruction(vcpu);
672 
673 	return 1;
674 }
675 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
676 
677 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
678 {
679 	++vcpu->stat.pf_guest;
680 	vcpu->arch.exception.nested_apf =
681 		is_guest_mode(vcpu) && fault->async_page_fault;
682 	if (vcpu->arch.exception.nested_apf) {
683 		vcpu->arch.apf.nested_apf_token = fault->address;
684 		kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
685 	} else {
686 		kvm_queue_exception_e_p(vcpu, PF_VECTOR, fault->error_code,
687 					fault->address);
688 	}
689 }
690 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
691 
692 bool kvm_inject_emulated_page_fault(struct kvm_vcpu *vcpu,
693 				    struct x86_exception *fault)
694 {
695 	struct kvm_mmu *fault_mmu;
696 	WARN_ON_ONCE(fault->vector != PF_VECTOR);
697 
698 	fault_mmu = fault->nested_page_fault ? vcpu->arch.mmu :
699 					       vcpu->arch.walk_mmu;
700 
701 	/*
702 	 * Invalidate the TLB entry for the faulting address, if it exists,
703 	 * else the access will fault indefinitely (and to emulate hardware).
704 	 */
705 	if ((fault->error_code & PFERR_PRESENT_MASK) &&
706 	    !(fault->error_code & PFERR_RSVD_MASK))
707 		kvm_mmu_invalidate_gva(vcpu, fault_mmu, fault->address,
708 				       fault_mmu->root_hpa);
709 
710 	fault_mmu->inject_page_fault(vcpu, fault);
711 	return fault->nested_page_fault;
712 }
713 EXPORT_SYMBOL_GPL(kvm_inject_emulated_page_fault);
714 
715 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
716 {
717 	atomic_inc(&vcpu->arch.nmi_queued);
718 	kvm_make_request(KVM_REQ_NMI, vcpu);
719 }
720 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
721 
722 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
723 {
724 	kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, false);
725 }
726 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
727 
728 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
729 {
730 	kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, true);
731 }
732 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
733 
734 /*
735  * Checks if cpl <= required_cpl; if true, return true.  Otherwise queue
736  * a #GP and return false.
737  */
738 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
739 {
740 	if (static_call(kvm_x86_get_cpl)(vcpu) <= required_cpl)
741 		return true;
742 	kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
743 	return false;
744 }
745 EXPORT_SYMBOL_GPL(kvm_require_cpl);
746 
747 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
748 {
749 	if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
750 		return true;
751 
752 	kvm_queue_exception(vcpu, UD_VECTOR);
753 	return false;
754 }
755 EXPORT_SYMBOL_GPL(kvm_require_dr);
756 
757 /*
758  * This function will be used to read from the physical memory of the currently
759  * running guest. The difference to kvm_vcpu_read_guest_page is that this function
760  * can read from guest physical or from the guest's guest physical memory.
761  */
762 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
763 			    gfn_t ngfn, void *data, int offset, int len,
764 			    u32 access)
765 {
766 	struct x86_exception exception;
767 	gfn_t real_gfn;
768 	gpa_t ngpa;
769 
770 	ngpa     = gfn_to_gpa(ngfn);
771 	real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
772 	if (real_gfn == UNMAPPED_GVA)
773 		return -EFAULT;
774 
775 	real_gfn = gpa_to_gfn(real_gfn);
776 
777 	return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
778 }
779 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
780 
781 static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
782 			       void *data, int offset, int len, u32 access)
783 {
784 	return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
785 				       data, offset, len, access);
786 }
787 
788 static inline u64 pdptr_rsvd_bits(struct kvm_vcpu *vcpu)
789 {
790 	return vcpu->arch.reserved_gpa_bits | rsvd_bits(5, 8) | rsvd_bits(1, 2);
791 }
792 
793 /*
794  * Load the pae pdptrs.  Return 1 if they are all valid, 0 otherwise.
795  */
796 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
797 {
798 	gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
799 	unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
800 	int i;
801 	int ret;
802 	u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
803 
804 	ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
805 				      offset * sizeof(u64), sizeof(pdpte),
806 				      PFERR_USER_MASK|PFERR_WRITE_MASK);
807 	if (ret < 0) {
808 		ret = 0;
809 		goto out;
810 	}
811 	for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
812 		if ((pdpte[i] & PT_PRESENT_MASK) &&
813 		    (pdpte[i] & pdptr_rsvd_bits(vcpu))) {
814 			ret = 0;
815 			goto out;
816 		}
817 	}
818 	ret = 1;
819 
820 	memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
821 	kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
822 
823 out:
824 
825 	return ret;
826 }
827 EXPORT_SYMBOL_GPL(load_pdptrs);
828 
829 bool pdptrs_changed(struct kvm_vcpu *vcpu)
830 {
831 	u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
832 	int offset;
833 	gfn_t gfn;
834 	int r;
835 
836 	if (!is_pae_paging(vcpu))
837 		return false;
838 
839 	if (!kvm_register_is_available(vcpu, VCPU_EXREG_PDPTR))
840 		return true;
841 
842 	gfn = (kvm_read_cr3(vcpu) & 0xffffffe0ul) >> PAGE_SHIFT;
843 	offset = (kvm_read_cr3(vcpu) & 0xffffffe0ul) & (PAGE_SIZE - 1);
844 	r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
845 				       PFERR_USER_MASK | PFERR_WRITE_MASK);
846 	if (r < 0)
847 		return true;
848 
849 	return memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
850 }
851 EXPORT_SYMBOL_GPL(pdptrs_changed);
852 
853 void kvm_post_set_cr0(struct kvm_vcpu *vcpu, unsigned long old_cr0, unsigned long cr0)
854 {
855 	unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
856 
857 	if ((cr0 ^ old_cr0) & X86_CR0_PG) {
858 		kvm_clear_async_pf_completion_queue(vcpu);
859 		kvm_async_pf_hash_reset(vcpu);
860 	}
861 
862 	if ((cr0 ^ old_cr0) & update_bits)
863 		kvm_mmu_reset_context(vcpu);
864 
865 	if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
866 	    kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
867 	    !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
868 		kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
869 }
870 EXPORT_SYMBOL_GPL(kvm_post_set_cr0);
871 
872 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
873 {
874 	unsigned long old_cr0 = kvm_read_cr0(vcpu);
875 	unsigned long pdptr_bits = X86_CR0_CD | X86_CR0_NW | X86_CR0_PG;
876 
877 	cr0 |= X86_CR0_ET;
878 
879 #ifdef CONFIG_X86_64
880 	if (cr0 & 0xffffffff00000000UL)
881 		return 1;
882 #endif
883 
884 	cr0 &= ~CR0_RESERVED_BITS;
885 
886 	if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
887 		return 1;
888 
889 	if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
890 		return 1;
891 
892 #ifdef CONFIG_X86_64
893 	if ((vcpu->arch.efer & EFER_LME) && !is_paging(vcpu) &&
894 	    (cr0 & X86_CR0_PG)) {
895 		int cs_db, cs_l;
896 
897 		if (!is_pae(vcpu))
898 			return 1;
899 		static_call(kvm_x86_get_cs_db_l_bits)(vcpu, &cs_db, &cs_l);
900 		if (cs_l)
901 			return 1;
902 	}
903 #endif
904 	if (!(vcpu->arch.efer & EFER_LME) && (cr0 & X86_CR0_PG) &&
905 	    is_pae(vcpu) && ((cr0 ^ old_cr0) & pdptr_bits) &&
906 	    !load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu)))
907 		return 1;
908 
909 	if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
910 		return 1;
911 
912 	static_call(kvm_x86_set_cr0)(vcpu, cr0);
913 
914 	kvm_post_set_cr0(vcpu, old_cr0, cr0);
915 
916 	return 0;
917 }
918 EXPORT_SYMBOL_GPL(kvm_set_cr0);
919 
920 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
921 {
922 	(void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
923 }
924 EXPORT_SYMBOL_GPL(kvm_lmsw);
925 
926 void kvm_load_guest_xsave_state(struct kvm_vcpu *vcpu)
927 {
928 	if (vcpu->arch.guest_state_protected)
929 		return;
930 
931 	if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
932 
933 		if (vcpu->arch.xcr0 != host_xcr0)
934 			xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
935 
936 		if (vcpu->arch.xsaves_enabled &&
937 		    vcpu->arch.ia32_xss != host_xss)
938 			wrmsrl(MSR_IA32_XSS, vcpu->arch.ia32_xss);
939 	}
940 
941 	if (static_cpu_has(X86_FEATURE_PKU) &&
942 	    (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) ||
943 	     (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU)) &&
944 	    vcpu->arch.pkru != vcpu->arch.host_pkru)
945 		__write_pkru(vcpu->arch.pkru);
946 }
947 EXPORT_SYMBOL_GPL(kvm_load_guest_xsave_state);
948 
949 void kvm_load_host_xsave_state(struct kvm_vcpu *vcpu)
950 {
951 	if (vcpu->arch.guest_state_protected)
952 		return;
953 
954 	if (static_cpu_has(X86_FEATURE_PKU) &&
955 	    (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) ||
956 	     (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU))) {
957 		vcpu->arch.pkru = rdpkru();
958 		if (vcpu->arch.pkru != vcpu->arch.host_pkru)
959 			__write_pkru(vcpu->arch.host_pkru);
960 	}
961 
962 	if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
963 
964 		if (vcpu->arch.xcr0 != host_xcr0)
965 			xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
966 
967 		if (vcpu->arch.xsaves_enabled &&
968 		    vcpu->arch.ia32_xss != host_xss)
969 			wrmsrl(MSR_IA32_XSS, host_xss);
970 	}
971 
972 }
973 EXPORT_SYMBOL_GPL(kvm_load_host_xsave_state);
974 
975 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
976 {
977 	u64 xcr0 = xcr;
978 	u64 old_xcr0 = vcpu->arch.xcr0;
979 	u64 valid_bits;
980 
981 	/* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now  */
982 	if (index != XCR_XFEATURE_ENABLED_MASK)
983 		return 1;
984 	if (!(xcr0 & XFEATURE_MASK_FP))
985 		return 1;
986 	if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
987 		return 1;
988 
989 	/*
990 	 * Do not allow the guest to set bits that we do not support
991 	 * saving.  However, xcr0 bit 0 is always set, even if the
992 	 * emulated CPU does not support XSAVE (see fx_init).
993 	 */
994 	valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
995 	if (xcr0 & ~valid_bits)
996 		return 1;
997 
998 	if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
999 	    (!(xcr0 & XFEATURE_MASK_BNDCSR)))
1000 		return 1;
1001 
1002 	if (xcr0 & XFEATURE_MASK_AVX512) {
1003 		if (!(xcr0 & XFEATURE_MASK_YMM))
1004 			return 1;
1005 		if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
1006 			return 1;
1007 	}
1008 	vcpu->arch.xcr0 = xcr0;
1009 
1010 	if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
1011 		kvm_update_cpuid_runtime(vcpu);
1012 	return 0;
1013 }
1014 
1015 int kvm_emulate_xsetbv(struct kvm_vcpu *vcpu)
1016 {
1017 	if (static_call(kvm_x86_get_cpl)(vcpu) != 0 ||
1018 	    __kvm_set_xcr(vcpu, kvm_rcx_read(vcpu), kvm_read_edx_eax(vcpu))) {
1019 		kvm_inject_gp(vcpu, 0);
1020 		return 1;
1021 	}
1022 
1023 	return kvm_skip_emulated_instruction(vcpu);
1024 }
1025 EXPORT_SYMBOL_GPL(kvm_emulate_xsetbv);
1026 
1027 bool kvm_is_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1028 {
1029 	if (cr4 & cr4_reserved_bits)
1030 		return false;
1031 
1032 	if (cr4 & vcpu->arch.cr4_guest_rsvd_bits)
1033 		return false;
1034 
1035 	return static_call(kvm_x86_is_valid_cr4)(vcpu, cr4);
1036 }
1037 EXPORT_SYMBOL_GPL(kvm_is_valid_cr4);
1038 
1039 void kvm_post_set_cr4(struct kvm_vcpu *vcpu, unsigned long old_cr4, unsigned long cr4)
1040 {
1041 	unsigned long mmu_role_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
1042 				      X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE;
1043 
1044 	if (((cr4 ^ old_cr4) & mmu_role_bits) ||
1045 	    (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
1046 		kvm_mmu_reset_context(vcpu);
1047 }
1048 EXPORT_SYMBOL_GPL(kvm_post_set_cr4);
1049 
1050 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1051 {
1052 	unsigned long old_cr4 = kvm_read_cr4(vcpu);
1053 	unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
1054 				   X86_CR4_SMEP;
1055 
1056 	if (!kvm_is_valid_cr4(vcpu, cr4))
1057 		return 1;
1058 
1059 	if (is_long_mode(vcpu)) {
1060 		if (!(cr4 & X86_CR4_PAE))
1061 			return 1;
1062 		if ((cr4 ^ old_cr4) & X86_CR4_LA57)
1063 			return 1;
1064 	} else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
1065 		   && ((cr4 ^ old_cr4) & pdptr_bits)
1066 		   && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
1067 				   kvm_read_cr3(vcpu)))
1068 		return 1;
1069 
1070 	if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
1071 		if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
1072 			return 1;
1073 
1074 		/* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
1075 		if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
1076 			return 1;
1077 	}
1078 
1079 	static_call(kvm_x86_set_cr4)(vcpu, cr4);
1080 
1081 	kvm_post_set_cr4(vcpu, old_cr4, cr4);
1082 
1083 	return 0;
1084 }
1085 EXPORT_SYMBOL_GPL(kvm_set_cr4);
1086 
1087 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
1088 {
1089 	bool skip_tlb_flush = false;
1090 #ifdef CONFIG_X86_64
1091 	bool pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
1092 
1093 	if (pcid_enabled) {
1094 		skip_tlb_flush = cr3 & X86_CR3_PCID_NOFLUSH;
1095 		cr3 &= ~X86_CR3_PCID_NOFLUSH;
1096 	}
1097 #endif
1098 
1099 	if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
1100 		if (!skip_tlb_flush) {
1101 			kvm_mmu_sync_roots(vcpu);
1102 			kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
1103 		}
1104 		return 0;
1105 	}
1106 
1107 	/*
1108 	 * Do not condition the GPA check on long mode, this helper is used to
1109 	 * stuff CR3, e.g. for RSM emulation, and there is no guarantee that
1110 	 * the current vCPU mode is accurate.
1111 	 */
1112 	if (kvm_vcpu_is_illegal_gpa(vcpu, cr3))
1113 		return 1;
1114 
1115 	if (is_pae_paging(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
1116 		return 1;
1117 
1118 	kvm_mmu_new_pgd(vcpu, cr3, skip_tlb_flush, skip_tlb_flush);
1119 	vcpu->arch.cr3 = cr3;
1120 	kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
1121 
1122 	return 0;
1123 }
1124 EXPORT_SYMBOL_GPL(kvm_set_cr3);
1125 
1126 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
1127 {
1128 	if (cr8 & CR8_RESERVED_BITS)
1129 		return 1;
1130 	if (lapic_in_kernel(vcpu))
1131 		kvm_lapic_set_tpr(vcpu, cr8);
1132 	else
1133 		vcpu->arch.cr8 = cr8;
1134 	return 0;
1135 }
1136 EXPORT_SYMBOL_GPL(kvm_set_cr8);
1137 
1138 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
1139 {
1140 	if (lapic_in_kernel(vcpu))
1141 		return kvm_lapic_get_cr8(vcpu);
1142 	else
1143 		return vcpu->arch.cr8;
1144 }
1145 EXPORT_SYMBOL_GPL(kvm_get_cr8);
1146 
1147 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
1148 {
1149 	int i;
1150 
1151 	if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
1152 		for (i = 0; i < KVM_NR_DB_REGS; i++)
1153 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
1154 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
1155 	}
1156 }
1157 
1158 void kvm_update_dr7(struct kvm_vcpu *vcpu)
1159 {
1160 	unsigned long dr7;
1161 
1162 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1163 		dr7 = vcpu->arch.guest_debug_dr7;
1164 	else
1165 		dr7 = vcpu->arch.dr7;
1166 	static_call(kvm_x86_set_dr7)(vcpu, dr7);
1167 	vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
1168 	if (dr7 & DR7_BP_EN_MASK)
1169 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
1170 }
1171 EXPORT_SYMBOL_GPL(kvm_update_dr7);
1172 
1173 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
1174 {
1175 	u64 fixed = DR6_FIXED_1;
1176 
1177 	if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
1178 		fixed |= DR6_RTM;
1179 
1180 	if (!guest_cpuid_has(vcpu, X86_FEATURE_BUS_LOCK_DETECT))
1181 		fixed |= DR6_BUS_LOCK;
1182 	return fixed;
1183 }
1184 
1185 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1186 {
1187 	size_t size = ARRAY_SIZE(vcpu->arch.db);
1188 
1189 	switch (dr) {
1190 	case 0 ... 3:
1191 		vcpu->arch.db[array_index_nospec(dr, size)] = val;
1192 		if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1193 			vcpu->arch.eff_db[dr] = val;
1194 		break;
1195 	case 4:
1196 	case 6:
1197 		if (!kvm_dr6_valid(val))
1198 			return 1; /* #GP */
1199 		vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
1200 		break;
1201 	case 5:
1202 	default: /* 7 */
1203 		if (!kvm_dr7_valid(val))
1204 			return 1; /* #GP */
1205 		vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
1206 		kvm_update_dr7(vcpu);
1207 		break;
1208 	}
1209 
1210 	return 0;
1211 }
1212 EXPORT_SYMBOL_GPL(kvm_set_dr);
1213 
1214 void kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
1215 {
1216 	size_t size = ARRAY_SIZE(vcpu->arch.db);
1217 
1218 	switch (dr) {
1219 	case 0 ... 3:
1220 		*val = vcpu->arch.db[array_index_nospec(dr, size)];
1221 		break;
1222 	case 4:
1223 	case 6:
1224 		*val = vcpu->arch.dr6;
1225 		break;
1226 	case 5:
1227 	default: /* 7 */
1228 		*val = vcpu->arch.dr7;
1229 		break;
1230 	}
1231 }
1232 EXPORT_SYMBOL_GPL(kvm_get_dr);
1233 
1234 int kvm_emulate_rdpmc(struct kvm_vcpu *vcpu)
1235 {
1236 	u32 ecx = kvm_rcx_read(vcpu);
1237 	u64 data;
1238 
1239 	if (kvm_pmu_rdpmc(vcpu, ecx, &data)) {
1240 		kvm_inject_gp(vcpu, 0);
1241 		return 1;
1242 	}
1243 
1244 	kvm_rax_write(vcpu, (u32)data);
1245 	kvm_rdx_write(vcpu, data >> 32);
1246 	return kvm_skip_emulated_instruction(vcpu);
1247 }
1248 EXPORT_SYMBOL_GPL(kvm_emulate_rdpmc);
1249 
1250 /*
1251  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1252  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1253  *
1254  * The three MSR lists(msrs_to_save, emulated_msrs, msr_based_features)
1255  * extract the supported MSRs from the related const lists.
1256  * msrs_to_save is selected from the msrs_to_save_all to reflect the
1257  * capabilities of the host cpu. This capabilities test skips MSRs that are
1258  * kvm-specific. Those are put in emulated_msrs_all; filtering of emulated_msrs
1259  * may depend on host virtualization features rather than host cpu features.
1260  */
1261 
1262 static const u32 msrs_to_save_all[] = {
1263 	MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1264 	MSR_STAR,
1265 #ifdef CONFIG_X86_64
1266 	MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1267 #endif
1268 	MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
1269 	MSR_IA32_FEAT_CTL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
1270 	MSR_IA32_SPEC_CTRL,
1271 	MSR_IA32_RTIT_CTL, MSR_IA32_RTIT_STATUS, MSR_IA32_RTIT_CR3_MATCH,
1272 	MSR_IA32_RTIT_OUTPUT_BASE, MSR_IA32_RTIT_OUTPUT_MASK,
1273 	MSR_IA32_RTIT_ADDR0_A, MSR_IA32_RTIT_ADDR0_B,
1274 	MSR_IA32_RTIT_ADDR1_A, MSR_IA32_RTIT_ADDR1_B,
1275 	MSR_IA32_RTIT_ADDR2_A, MSR_IA32_RTIT_ADDR2_B,
1276 	MSR_IA32_RTIT_ADDR3_A, MSR_IA32_RTIT_ADDR3_B,
1277 	MSR_IA32_UMWAIT_CONTROL,
1278 
1279 	MSR_ARCH_PERFMON_FIXED_CTR0, MSR_ARCH_PERFMON_FIXED_CTR1,
1280 	MSR_ARCH_PERFMON_FIXED_CTR0 + 2, MSR_ARCH_PERFMON_FIXED_CTR0 + 3,
1281 	MSR_CORE_PERF_FIXED_CTR_CTRL, MSR_CORE_PERF_GLOBAL_STATUS,
1282 	MSR_CORE_PERF_GLOBAL_CTRL, MSR_CORE_PERF_GLOBAL_OVF_CTRL,
1283 	MSR_ARCH_PERFMON_PERFCTR0, MSR_ARCH_PERFMON_PERFCTR1,
1284 	MSR_ARCH_PERFMON_PERFCTR0 + 2, MSR_ARCH_PERFMON_PERFCTR0 + 3,
1285 	MSR_ARCH_PERFMON_PERFCTR0 + 4, MSR_ARCH_PERFMON_PERFCTR0 + 5,
1286 	MSR_ARCH_PERFMON_PERFCTR0 + 6, MSR_ARCH_PERFMON_PERFCTR0 + 7,
1287 	MSR_ARCH_PERFMON_PERFCTR0 + 8, MSR_ARCH_PERFMON_PERFCTR0 + 9,
1288 	MSR_ARCH_PERFMON_PERFCTR0 + 10, MSR_ARCH_PERFMON_PERFCTR0 + 11,
1289 	MSR_ARCH_PERFMON_PERFCTR0 + 12, MSR_ARCH_PERFMON_PERFCTR0 + 13,
1290 	MSR_ARCH_PERFMON_PERFCTR0 + 14, MSR_ARCH_PERFMON_PERFCTR0 + 15,
1291 	MSR_ARCH_PERFMON_PERFCTR0 + 16, MSR_ARCH_PERFMON_PERFCTR0 + 17,
1292 	MSR_ARCH_PERFMON_EVENTSEL0, MSR_ARCH_PERFMON_EVENTSEL1,
1293 	MSR_ARCH_PERFMON_EVENTSEL0 + 2, MSR_ARCH_PERFMON_EVENTSEL0 + 3,
1294 	MSR_ARCH_PERFMON_EVENTSEL0 + 4, MSR_ARCH_PERFMON_EVENTSEL0 + 5,
1295 	MSR_ARCH_PERFMON_EVENTSEL0 + 6, MSR_ARCH_PERFMON_EVENTSEL0 + 7,
1296 	MSR_ARCH_PERFMON_EVENTSEL0 + 8, MSR_ARCH_PERFMON_EVENTSEL0 + 9,
1297 	MSR_ARCH_PERFMON_EVENTSEL0 + 10, MSR_ARCH_PERFMON_EVENTSEL0 + 11,
1298 	MSR_ARCH_PERFMON_EVENTSEL0 + 12, MSR_ARCH_PERFMON_EVENTSEL0 + 13,
1299 	MSR_ARCH_PERFMON_EVENTSEL0 + 14, MSR_ARCH_PERFMON_EVENTSEL0 + 15,
1300 	MSR_ARCH_PERFMON_EVENTSEL0 + 16, MSR_ARCH_PERFMON_EVENTSEL0 + 17,
1301 };
1302 
1303 static u32 msrs_to_save[ARRAY_SIZE(msrs_to_save_all)];
1304 static unsigned num_msrs_to_save;
1305 
1306 static const u32 emulated_msrs_all[] = {
1307 	MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1308 	MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1309 	HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1310 	HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
1311 	HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
1312 	HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1313 	HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1314 	HV_X64_MSR_RESET,
1315 	HV_X64_MSR_VP_INDEX,
1316 	HV_X64_MSR_VP_RUNTIME,
1317 	HV_X64_MSR_SCONTROL,
1318 	HV_X64_MSR_STIMER0_CONFIG,
1319 	HV_X64_MSR_VP_ASSIST_PAGE,
1320 	HV_X64_MSR_REENLIGHTENMENT_CONTROL, HV_X64_MSR_TSC_EMULATION_CONTROL,
1321 	HV_X64_MSR_TSC_EMULATION_STATUS,
1322 	HV_X64_MSR_SYNDBG_OPTIONS,
1323 	HV_X64_MSR_SYNDBG_CONTROL, HV_X64_MSR_SYNDBG_STATUS,
1324 	HV_X64_MSR_SYNDBG_SEND_BUFFER, HV_X64_MSR_SYNDBG_RECV_BUFFER,
1325 	HV_X64_MSR_SYNDBG_PENDING_BUFFER,
1326 
1327 	MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1328 	MSR_KVM_PV_EOI_EN, MSR_KVM_ASYNC_PF_INT, MSR_KVM_ASYNC_PF_ACK,
1329 
1330 	MSR_IA32_TSC_ADJUST,
1331 	MSR_IA32_TSC_DEADLINE,
1332 	MSR_IA32_ARCH_CAPABILITIES,
1333 	MSR_IA32_PERF_CAPABILITIES,
1334 	MSR_IA32_MISC_ENABLE,
1335 	MSR_IA32_MCG_STATUS,
1336 	MSR_IA32_MCG_CTL,
1337 	MSR_IA32_MCG_EXT_CTL,
1338 	MSR_IA32_SMBASE,
1339 	MSR_SMI_COUNT,
1340 	MSR_PLATFORM_INFO,
1341 	MSR_MISC_FEATURES_ENABLES,
1342 	MSR_AMD64_VIRT_SPEC_CTRL,
1343 	MSR_IA32_POWER_CTL,
1344 	MSR_IA32_UCODE_REV,
1345 
1346 	/*
1347 	 * The following list leaves out MSRs whose values are determined
1348 	 * by arch/x86/kvm/vmx/nested.c based on CPUID or other MSRs.
1349 	 * We always support the "true" VMX control MSRs, even if the host
1350 	 * processor does not, so I am putting these registers here rather
1351 	 * than in msrs_to_save_all.
1352 	 */
1353 	MSR_IA32_VMX_BASIC,
1354 	MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1355 	MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1356 	MSR_IA32_VMX_TRUE_EXIT_CTLS,
1357 	MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1358 	MSR_IA32_VMX_MISC,
1359 	MSR_IA32_VMX_CR0_FIXED0,
1360 	MSR_IA32_VMX_CR4_FIXED0,
1361 	MSR_IA32_VMX_VMCS_ENUM,
1362 	MSR_IA32_VMX_PROCBASED_CTLS2,
1363 	MSR_IA32_VMX_EPT_VPID_CAP,
1364 	MSR_IA32_VMX_VMFUNC,
1365 
1366 	MSR_K7_HWCR,
1367 	MSR_KVM_POLL_CONTROL,
1368 };
1369 
1370 static u32 emulated_msrs[ARRAY_SIZE(emulated_msrs_all)];
1371 static unsigned num_emulated_msrs;
1372 
1373 /*
1374  * List of msr numbers which are used to expose MSR-based features that
1375  * can be used by a hypervisor to validate requested CPU features.
1376  */
1377 static const u32 msr_based_features_all[] = {
1378 	MSR_IA32_VMX_BASIC,
1379 	MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1380 	MSR_IA32_VMX_PINBASED_CTLS,
1381 	MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1382 	MSR_IA32_VMX_PROCBASED_CTLS,
1383 	MSR_IA32_VMX_TRUE_EXIT_CTLS,
1384 	MSR_IA32_VMX_EXIT_CTLS,
1385 	MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1386 	MSR_IA32_VMX_ENTRY_CTLS,
1387 	MSR_IA32_VMX_MISC,
1388 	MSR_IA32_VMX_CR0_FIXED0,
1389 	MSR_IA32_VMX_CR0_FIXED1,
1390 	MSR_IA32_VMX_CR4_FIXED0,
1391 	MSR_IA32_VMX_CR4_FIXED1,
1392 	MSR_IA32_VMX_VMCS_ENUM,
1393 	MSR_IA32_VMX_PROCBASED_CTLS2,
1394 	MSR_IA32_VMX_EPT_VPID_CAP,
1395 	MSR_IA32_VMX_VMFUNC,
1396 
1397 	MSR_F10H_DECFG,
1398 	MSR_IA32_UCODE_REV,
1399 	MSR_IA32_ARCH_CAPABILITIES,
1400 	MSR_IA32_PERF_CAPABILITIES,
1401 };
1402 
1403 static u32 msr_based_features[ARRAY_SIZE(msr_based_features_all)];
1404 static unsigned int num_msr_based_features;
1405 
1406 static u64 kvm_get_arch_capabilities(void)
1407 {
1408 	u64 data = 0;
1409 
1410 	if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES))
1411 		rdmsrl(MSR_IA32_ARCH_CAPABILITIES, data);
1412 
1413 	/*
1414 	 * If nx_huge_pages is enabled, KVM's shadow paging will ensure that
1415 	 * the nested hypervisor runs with NX huge pages.  If it is not,
1416 	 * L1 is anyway vulnerable to ITLB_MULTIHIT exploits from other
1417 	 * L1 guests, so it need not worry about its own (L2) guests.
1418 	 */
1419 	data |= ARCH_CAP_PSCHANGE_MC_NO;
1420 
1421 	/*
1422 	 * If we're doing cache flushes (either "always" or "cond")
1423 	 * we will do one whenever the guest does a vmlaunch/vmresume.
1424 	 * If an outer hypervisor is doing the cache flush for us
1425 	 * (VMENTER_L1D_FLUSH_NESTED_VM), we can safely pass that
1426 	 * capability to the guest too, and if EPT is disabled we're not
1427 	 * vulnerable.  Overall, only VMENTER_L1D_FLUSH_NEVER will
1428 	 * require a nested hypervisor to do a flush of its own.
1429 	 */
1430 	if (l1tf_vmx_mitigation != VMENTER_L1D_FLUSH_NEVER)
1431 		data |= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH;
1432 
1433 	if (!boot_cpu_has_bug(X86_BUG_CPU_MELTDOWN))
1434 		data |= ARCH_CAP_RDCL_NO;
1435 	if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
1436 		data |= ARCH_CAP_SSB_NO;
1437 	if (!boot_cpu_has_bug(X86_BUG_MDS))
1438 		data |= ARCH_CAP_MDS_NO;
1439 
1440 	if (!boot_cpu_has(X86_FEATURE_RTM)) {
1441 		/*
1442 		 * If RTM=0 because the kernel has disabled TSX, the host might
1443 		 * have TAA_NO or TSX_CTRL.  Clear TAA_NO (the guest sees RTM=0
1444 		 * and therefore knows that there cannot be TAA) but keep
1445 		 * TSX_CTRL: some buggy userspaces leave it set on tsx=on hosts,
1446 		 * and we want to allow migrating those guests to tsx=off hosts.
1447 		 */
1448 		data &= ~ARCH_CAP_TAA_NO;
1449 	} else if (!boot_cpu_has_bug(X86_BUG_TAA)) {
1450 		data |= ARCH_CAP_TAA_NO;
1451 	} else {
1452 		/*
1453 		 * Nothing to do here; we emulate TSX_CTRL if present on the
1454 		 * host so the guest can choose between disabling TSX or
1455 		 * using VERW to clear CPU buffers.
1456 		 */
1457 	}
1458 
1459 	return data;
1460 }
1461 
1462 static int kvm_get_msr_feature(struct kvm_msr_entry *msr)
1463 {
1464 	switch (msr->index) {
1465 	case MSR_IA32_ARCH_CAPABILITIES:
1466 		msr->data = kvm_get_arch_capabilities();
1467 		break;
1468 	case MSR_IA32_UCODE_REV:
1469 		rdmsrl_safe(msr->index, &msr->data);
1470 		break;
1471 	default:
1472 		return static_call(kvm_x86_get_msr_feature)(msr);
1473 	}
1474 	return 0;
1475 }
1476 
1477 static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1478 {
1479 	struct kvm_msr_entry msr;
1480 	int r;
1481 
1482 	msr.index = index;
1483 	r = kvm_get_msr_feature(&msr);
1484 
1485 	if (r == KVM_MSR_RET_INVALID) {
1486 		/* Unconditionally clear the output for simplicity */
1487 		*data = 0;
1488 		if (kvm_msr_ignored_check(index, 0, false))
1489 			r = 0;
1490 	}
1491 
1492 	if (r)
1493 		return r;
1494 
1495 	*data = msr.data;
1496 
1497 	return 0;
1498 }
1499 
1500 static bool __kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1501 {
1502 	if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1503 		return false;
1504 
1505 	if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1506 		return false;
1507 
1508 	if (efer & (EFER_LME | EFER_LMA) &&
1509 	    !guest_cpuid_has(vcpu, X86_FEATURE_LM))
1510 		return false;
1511 
1512 	if (efer & EFER_NX && !guest_cpuid_has(vcpu, X86_FEATURE_NX))
1513 		return false;
1514 
1515 	return true;
1516 
1517 }
1518 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1519 {
1520 	if (efer & efer_reserved_bits)
1521 		return false;
1522 
1523 	return __kvm_valid_efer(vcpu, efer);
1524 }
1525 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1526 
1527 static int set_efer(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1528 {
1529 	u64 old_efer = vcpu->arch.efer;
1530 	u64 efer = msr_info->data;
1531 	int r;
1532 
1533 	if (efer & efer_reserved_bits)
1534 		return 1;
1535 
1536 	if (!msr_info->host_initiated) {
1537 		if (!__kvm_valid_efer(vcpu, efer))
1538 			return 1;
1539 
1540 		if (is_paging(vcpu) &&
1541 		    (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1542 			return 1;
1543 	}
1544 
1545 	efer &= ~EFER_LMA;
1546 	efer |= vcpu->arch.efer & EFER_LMA;
1547 
1548 	r = static_call(kvm_x86_set_efer)(vcpu, efer);
1549 	if (r) {
1550 		WARN_ON(r > 0);
1551 		return r;
1552 	}
1553 
1554 	/* Update reserved bits */
1555 	if ((efer ^ old_efer) & EFER_NX)
1556 		kvm_mmu_reset_context(vcpu);
1557 
1558 	return 0;
1559 }
1560 
1561 void kvm_enable_efer_bits(u64 mask)
1562 {
1563        efer_reserved_bits &= ~mask;
1564 }
1565 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1566 
1567 bool kvm_msr_allowed(struct kvm_vcpu *vcpu, u32 index, u32 type)
1568 {
1569 	struct kvm_x86_msr_filter *msr_filter;
1570 	struct msr_bitmap_range *ranges;
1571 	struct kvm *kvm = vcpu->kvm;
1572 	bool allowed;
1573 	int idx;
1574 	u32 i;
1575 
1576 	/* x2APIC MSRs do not support filtering. */
1577 	if (index >= 0x800 && index <= 0x8ff)
1578 		return true;
1579 
1580 	idx = srcu_read_lock(&kvm->srcu);
1581 
1582 	msr_filter = srcu_dereference(kvm->arch.msr_filter, &kvm->srcu);
1583 	if (!msr_filter) {
1584 		allowed = true;
1585 		goto out;
1586 	}
1587 
1588 	allowed = msr_filter->default_allow;
1589 	ranges = msr_filter->ranges;
1590 
1591 	for (i = 0; i < msr_filter->count; i++) {
1592 		u32 start = ranges[i].base;
1593 		u32 end = start + ranges[i].nmsrs;
1594 		u32 flags = ranges[i].flags;
1595 		unsigned long *bitmap = ranges[i].bitmap;
1596 
1597 		if ((index >= start) && (index < end) && (flags & type)) {
1598 			allowed = !!test_bit(index - start, bitmap);
1599 			break;
1600 		}
1601 	}
1602 
1603 out:
1604 	srcu_read_unlock(&kvm->srcu, idx);
1605 
1606 	return allowed;
1607 }
1608 EXPORT_SYMBOL_GPL(kvm_msr_allowed);
1609 
1610 /*
1611  * Write @data into the MSR specified by @index.  Select MSR specific fault
1612  * checks are bypassed if @host_initiated is %true.
1613  * Returns 0 on success, non-0 otherwise.
1614  * Assumes vcpu_load() was already called.
1615  */
1616 static int __kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data,
1617 			 bool host_initiated)
1618 {
1619 	struct msr_data msr;
1620 
1621 	if (!host_initiated && !kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_WRITE))
1622 		return KVM_MSR_RET_FILTERED;
1623 
1624 	switch (index) {
1625 	case MSR_FS_BASE:
1626 	case MSR_GS_BASE:
1627 	case MSR_KERNEL_GS_BASE:
1628 	case MSR_CSTAR:
1629 	case MSR_LSTAR:
1630 		if (is_noncanonical_address(data, vcpu))
1631 			return 1;
1632 		break;
1633 	case MSR_IA32_SYSENTER_EIP:
1634 	case MSR_IA32_SYSENTER_ESP:
1635 		/*
1636 		 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1637 		 * non-canonical address is written on Intel but not on
1638 		 * AMD (which ignores the top 32-bits, because it does
1639 		 * not implement 64-bit SYSENTER).
1640 		 *
1641 		 * 64-bit code should hence be able to write a non-canonical
1642 		 * value on AMD.  Making the address canonical ensures that
1643 		 * vmentry does not fail on Intel after writing a non-canonical
1644 		 * value, and that something deterministic happens if the guest
1645 		 * invokes 64-bit SYSENTER.
1646 		 */
1647 		data = get_canonical(data, vcpu_virt_addr_bits(vcpu));
1648 		break;
1649 	case MSR_TSC_AUX:
1650 		if (!kvm_is_supported_user_return_msr(MSR_TSC_AUX))
1651 			return 1;
1652 
1653 		if (!host_initiated &&
1654 		    !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP) &&
1655 		    !guest_cpuid_has(vcpu, X86_FEATURE_RDPID))
1656 			return 1;
1657 
1658 		/*
1659 		 * Per Intel's SDM, bits 63:32 are reserved, but AMD's APM has
1660 		 * incomplete and conflicting architectural behavior.  Current
1661 		 * AMD CPUs completely ignore bits 63:32, i.e. they aren't
1662 		 * reserved and always read as zeros.  Enforce Intel's reserved
1663 		 * bits check if and only if the guest CPU is Intel, and clear
1664 		 * the bits in all other cases.  This ensures cross-vendor
1665 		 * migration will provide consistent behavior for the guest.
1666 		 */
1667 		if (guest_cpuid_is_intel(vcpu) && (data >> 32) != 0)
1668 			return 1;
1669 
1670 		data = (u32)data;
1671 		break;
1672 	}
1673 
1674 	msr.data = data;
1675 	msr.index = index;
1676 	msr.host_initiated = host_initiated;
1677 
1678 	return static_call(kvm_x86_set_msr)(vcpu, &msr);
1679 }
1680 
1681 static int kvm_set_msr_ignored_check(struct kvm_vcpu *vcpu,
1682 				     u32 index, u64 data, bool host_initiated)
1683 {
1684 	int ret = __kvm_set_msr(vcpu, index, data, host_initiated);
1685 
1686 	if (ret == KVM_MSR_RET_INVALID)
1687 		if (kvm_msr_ignored_check(index, data, true))
1688 			ret = 0;
1689 
1690 	return ret;
1691 }
1692 
1693 /*
1694  * Read the MSR specified by @index into @data.  Select MSR specific fault
1695  * checks are bypassed if @host_initiated is %true.
1696  * Returns 0 on success, non-0 otherwise.
1697  * Assumes vcpu_load() was already called.
1698  */
1699 int __kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data,
1700 		  bool host_initiated)
1701 {
1702 	struct msr_data msr;
1703 	int ret;
1704 
1705 	if (!host_initiated && !kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_READ))
1706 		return KVM_MSR_RET_FILTERED;
1707 
1708 	switch (index) {
1709 	case MSR_TSC_AUX:
1710 		if (!kvm_is_supported_user_return_msr(MSR_TSC_AUX))
1711 			return 1;
1712 
1713 		if (!host_initiated &&
1714 		    !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP) &&
1715 		    !guest_cpuid_has(vcpu, X86_FEATURE_RDPID))
1716 			return 1;
1717 		break;
1718 	}
1719 
1720 	msr.index = index;
1721 	msr.host_initiated = host_initiated;
1722 
1723 	ret = static_call(kvm_x86_get_msr)(vcpu, &msr);
1724 	if (!ret)
1725 		*data = msr.data;
1726 	return ret;
1727 }
1728 
1729 static int kvm_get_msr_ignored_check(struct kvm_vcpu *vcpu,
1730 				     u32 index, u64 *data, bool host_initiated)
1731 {
1732 	int ret = __kvm_get_msr(vcpu, index, data, host_initiated);
1733 
1734 	if (ret == KVM_MSR_RET_INVALID) {
1735 		/* Unconditionally clear *data for simplicity */
1736 		*data = 0;
1737 		if (kvm_msr_ignored_check(index, 0, false))
1738 			ret = 0;
1739 	}
1740 
1741 	return ret;
1742 }
1743 
1744 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data)
1745 {
1746 	return kvm_get_msr_ignored_check(vcpu, index, data, false);
1747 }
1748 EXPORT_SYMBOL_GPL(kvm_get_msr);
1749 
1750 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data)
1751 {
1752 	return kvm_set_msr_ignored_check(vcpu, index, data, false);
1753 }
1754 EXPORT_SYMBOL_GPL(kvm_set_msr);
1755 
1756 static int complete_emulated_rdmsr(struct kvm_vcpu *vcpu)
1757 {
1758 	int err = vcpu->run->msr.error;
1759 	if (!err) {
1760 		kvm_rax_write(vcpu, (u32)vcpu->run->msr.data);
1761 		kvm_rdx_write(vcpu, vcpu->run->msr.data >> 32);
1762 	}
1763 
1764 	return static_call(kvm_x86_complete_emulated_msr)(vcpu, err);
1765 }
1766 
1767 static int complete_emulated_wrmsr(struct kvm_vcpu *vcpu)
1768 {
1769 	return static_call(kvm_x86_complete_emulated_msr)(vcpu, vcpu->run->msr.error);
1770 }
1771 
1772 static u64 kvm_msr_reason(int r)
1773 {
1774 	switch (r) {
1775 	case KVM_MSR_RET_INVALID:
1776 		return KVM_MSR_EXIT_REASON_UNKNOWN;
1777 	case KVM_MSR_RET_FILTERED:
1778 		return KVM_MSR_EXIT_REASON_FILTER;
1779 	default:
1780 		return KVM_MSR_EXIT_REASON_INVAL;
1781 	}
1782 }
1783 
1784 static int kvm_msr_user_space(struct kvm_vcpu *vcpu, u32 index,
1785 			      u32 exit_reason, u64 data,
1786 			      int (*completion)(struct kvm_vcpu *vcpu),
1787 			      int r)
1788 {
1789 	u64 msr_reason = kvm_msr_reason(r);
1790 
1791 	/* Check if the user wanted to know about this MSR fault */
1792 	if (!(vcpu->kvm->arch.user_space_msr_mask & msr_reason))
1793 		return 0;
1794 
1795 	vcpu->run->exit_reason = exit_reason;
1796 	vcpu->run->msr.error = 0;
1797 	memset(vcpu->run->msr.pad, 0, sizeof(vcpu->run->msr.pad));
1798 	vcpu->run->msr.reason = msr_reason;
1799 	vcpu->run->msr.index = index;
1800 	vcpu->run->msr.data = data;
1801 	vcpu->arch.complete_userspace_io = completion;
1802 
1803 	return 1;
1804 }
1805 
1806 static int kvm_get_msr_user_space(struct kvm_vcpu *vcpu, u32 index, int r)
1807 {
1808 	return kvm_msr_user_space(vcpu, index, KVM_EXIT_X86_RDMSR, 0,
1809 				   complete_emulated_rdmsr, r);
1810 }
1811 
1812 static int kvm_set_msr_user_space(struct kvm_vcpu *vcpu, u32 index, u64 data, int r)
1813 {
1814 	return kvm_msr_user_space(vcpu, index, KVM_EXIT_X86_WRMSR, data,
1815 				   complete_emulated_wrmsr, r);
1816 }
1817 
1818 int kvm_emulate_rdmsr(struct kvm_vcpu *vcpu)
1819 {
1820 	u32 ecx = kvm_rcx_read(vcpu);
1821 	u64 data;
1822 	int r;
1823 
1824 	r = kvm_get_msr(vcpu, ecx, &data);
1825 
1826 	/* MSR read failed? See if we should ask user space */
1827 	if (r && kvm_get_msr_user_space(vcpu, ecx, r)) {
1828 		/* Bounce to user space */
1829 		return 0;
1830 	}
1831 
1832 	if (!r) {
1833 		trace_kvm_msr_read(ecx, data);
1834 
1835 		kvm_rax_write(vcpu, data & -1u);
1836 		kvm_rdx_write(vcpu, (data >> 32) & -1u);
1837 	} else {
1838 		trace_kvm_msr_read_ex(ecx);
1839 	}
1840 
1841 	return static_call(kvm_x86_complete_emulated_msr)(vcpu, r);
1842 }
1843 EXPORT_SYMBOL_GPL(kvm_emulate_rdmsr);
1844 
1845 int kvm_emulate_wrmsr(struct kvm_vcpu *vcpu)
1846 {
1847 	u32 ecx = kvm_rcx_read(vcpu);
1848 	u64 data = kvm_read_edx_eax(vcpu);
1849 	int r;
1850 
1851 	r = kvm_set_msr(vcpu, ecx, data);
1852 
1853 	/* MSR write failed? See if we should ask user space */
1854 	if (r && kvm_set_msr_user_space(vcpu, ecx, data, r))
1855 		/* Bounce to user space */
1856 		return 0;
1857 
1858 	/* Signal all other negative errors to userspace */
1859 	if (r < 0)
1860 		return r;
1861 
1862 	if (!r)
1863 		trace_kvm_msr_write(ecx, data);
1864 	else
1865 		trace_kvm_msr_write_ex(ecx, data);
1866 
1867 	return static_call(kvm_x86_complete_emulated_msr)(vcpu, r);
1868 }
1869 EXPORT_SYMBOL_GPL(kvm_emulate_wrmsr);
1870 
1871 int kvm_emulate_as_nop(struct kvm_vcpu *vcpu)
1872 {
1873 	return kvm_skip_emulated_instruction(vcpu);
1874 }
1875 EXPORT_SYMBOL_GPL(kvm_emulate_as_nop);
1876 
1877 int kvm_emulate_invd(struct kvm_vcpu *vcpu)
1878 {
1879 	/* Treat an INVD instruction as a NOP and just skip it. */
1880 	return kvm_emulate_as_nop(vcpu);
1881 }
1882 EXPORT_SYMBOL_GPL(kvm_emulate_invd);
1883 
1884 int kvm_emulate_mwait(struct kvm_vcpu *vcpu)
1885 {
1886 	pr_warn_once("kvm: MWAIT instruction emulated as NOP!\n");
1887 	return kvm_emulate_as_nop(vcpu);
1888 }
1889 EXPORT_SYMBOL_GPL(kvm_emulate_mwait);
1890 
1891 int kvm_handle_invalid_op(struct kvm_vcpu *vcpu)
1892 {
1893 	kvm_queue_exception(vcpu, UD_VECTOR);
1894 	return 1;
1895 }
1896 EXPORT_SYMBOL_GPL(kvm_handle_invalid_op);
1897 
1898 int kvm_emulate_monitor(struct kvm_vcpu *vcpu)
1899 {
1900 	pr_warn_once("kvm: MONITOR instruction emulated as NOP!\n");
1901 	return kvm_emulate_as_nop(vcpu);
1902 }
1903 EXPORT_SYMBOL_GPL(kvm_emulate_monitor);
1904 
1905 static inline bool kvm_vcpu_exit_request(struct kvm_vcpu *vcpu)
1906 {
1907 	xfer_to_guest_mode_prepare();
1908 	return vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu) ||
1909 		xfer_to_guest_mode_work_pending();
1910 }
1911 
1912 /*
1913  * The fast path for frequent and performance sensitive wrmsr emulation,
1914  * i.e. the sending of IPI, sending IPI early in the VM-Exit flow reduces
1915  * the latency of virtual IPI by avoiding the expensive bits of transitioning
1916  * from guest to host, e.g. reacquiring KVM's SRCU lock. In contrast to the
1917  * other cases which must be called after interrupts are enabled on the host.
1918  */
1919 static int handle_fastpath_set_x2apic_icr_irqoff(struct kvm_vcpu *vcpu, u64 data)
1920 {
1921 	if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(vcpu->arch.apic))
1922 		return 1;
1923 
1924 	if (((data & APIC_SHORT_MASK) == APIC_DEST_NOSHORT) &&
1925 		((data & APIC_DEST_MASK) == APIC_DEST_PHYSICAL) &&
1926 		((data & APIC_MODE_MASK) == APIC_DM_FIXED) &&
1927 		((u32)(data >> 32) != X2APIC_BROADCAST)) {
1928 
1929 		data &= ~(1 << 12);
1930 		kvm_apic_send_ipi(vcpu->arch.apic, (u32)data, (u32)(data >> 32));
1931 		kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR2, (u32)(data >> 32));
1932 		kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR, (u32)data);
1933 		trace_kvm_apic_write(APIC_ICR, (u32)data);
1934 		return 0;
1935 	}
1936 
1937 	return 1;
1938 }
1939 
1940 static int handle_fastpath_set_tscdeadline(struct kvm_vcpu *vcpu, u64 data)
1941 {
1942 	if (!kvm_can_use_hv_timer(vcpu))
1943 		return 1;
1944 
1945 	kvm_set_lapic_tscdeadline_msr(vcpu, data);
1946 	return 0;
1947 }
1948 
1949 fastpath_t handle_fastpath_set_msr_irqoff(struct kvm_vcpu *vcpu)
1950 {
1951 	u32 msr = kvm_rcx_read(vcpu);
1952 	u64 data;
1953 	fastpath_t ret = EXIT_FASTPATH_NONE;
1954 
1955 	switch (msr) {
1956 	case APIC_BASE_MSR + (APIC_ICR >> 4):
1957 		data = kvm_read_edx_eax(vcpu);
1958 		if (!handle_fastpath_set_x2apic_icr_irqoff(vcpu, data)) {
1959 			kvm_skip_emulated_instruction(vcpu);
1960 			ret = EXIT_FASTPATH_EXIT_HANDLED;
1961 		}
1962 		break;
1963 	case MSR_IA32_TSC_DEADLINE:
1964 		data = kvm_read_edx_eax(vcpu);
1965 		if (!handle_fastpath_set_tscdeadline(vcpu, data)) {
1966 			kvm_skip_emulated_instruction(vcpu);
1967 			ret = EXIT_FASTPATH_REENTER_GUEST;
1968 		}
1969 		break;
1970 	default:
1971 		break;
1972 	}
1973 
1974 	if (ret != EXIT_FASTPATH_NONE)
1975 		trace_kvm_msr_write(msr, data);
1976 
1977 	return ret;
1978 }
1979 EXPORT_SYMBOL_GPL(handle_fastpath_set_msr_irqoff);
1980 
1981 /*
1982  * Adapt set_msr() to msr_io()'s calling convention
1983  */
1984 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1985 {
1986 	return kvm_get_msr_ignored_check(vcpu, index, data, true);
1987 }
1988 
1989 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1990 {
1991 	return kvm_set_msr_ignored_check(vcpu, index, *data, true);
1992 }
1993 
1994 #ifdef CONFIG_X86_64
1995 struct pvclock_clock {
1996 	int vclock_mode;
1997 	u64 cycle_last;
1998 	u64 mask;
1999 	u32 mult;
2000 	u32 shift;
2001 	u64 base_cycles;
2002 	u64 offset;
2003 };
2004 
2005 struct pvclock_gtod_data {
2006 	seqcount_t	seq;
2007 
2008 	struct pvclock_clock clock; /* extract of a clocksource struct */
2009 	struct pvclock_clock raw_clock; /* extract of a clocksource struct */
2010 
2011 	ktime_t		offs_boot;
2012 	u64		wall_time_sec;
2013 };
2014 
2015 static struct pvclock_gtod_data pvclock_gtod_data;
2016 
2017 static void update_pvclock_gtod(struct timekeeper *tk)
2018 {
2019 	struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
2020 
2021 	write_seqcount_begin(&vdata->seq);
2022 
2023 	/* copy pvclock gtod data */
2024 	vdata->clock.vclock_mode	= tk->tkr_mono.clock->vdso_clock_mode;
2025 	vdata->clock.cycle_last		= tk->tkr_mono.cycle_last;
2026 	vdata->clock.mask		= tk->tkr_mono.mask;
2027 	vdata->clock.mult		= tk->tkr_mono.mult;
2028 	vdata->clock.shift		= tk->tkr_mono.shift;
2029 	vdata->clock.base_cycles	= tk->tkr_mono.xtime_nsec;
2030 	vdata->clock.offset		= tk->tkr_mono.base;
2031 
2032 	vdata->raw_clock.vclock_mode	= tk->tkr_raw.clock->vdso_clock_mode;
2033 	vdata->raw_clock.cycle_last	= tk->tkr_raw.cycle_last;
2034 	vdata->raw_clock.mask		= tk->tkr_raw.mask;
2035 	vdata->raw_clock.mult		= tk->tkr_raw.mult;
2036 	vdata->raw_clock.shift		= tk->tkr_raw.shift;
2037 	vdata->raw_clock.base_cycles	= tk->tkr_raw.xtime_nsec;
2038 	vdata->raw_clock.offset		= tk->tkr_raw.base;
2039 
2040 	vdata->wall_time_sec            = tk->xtime_sec;
2041 
2042 	vdata->offs_boot		= tk->offs_boot;
2043 
2044 	write_seqcount_end(&vdata->seq);
2045 }
2046 
2047 static s64 get_kvmclock_base_ns(void)
2048 {
2049 	/* Count up from boot time, but with the frequency of the raw clock.  */
2050 	return ktime_to_ns(ktime_add(ktime_get_raw(), pvclock_gtod_data.offs_boot));
2051 }
2052 #else
2053 static s64 get_kvmclock_base_ns(void)
2054 {
2055 	/* Master clock not used, so we can just use CLOCK_BOOTTIME.  */
2056 	return ktime_get_boottime_ns();
2057 }
2058 #endif
2059 
2060 void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock, int sec_hi_ofs)
2061 {
2062 	int version;
2063 	int r;
2064 	struct pvclock_wall_clock wc;
2065 	u32 wc_sec_hi;
2066 	u64 wall_nsec;
2067 
2068 	if (!wall_clock)
2069 		return;
2070 
2071 	r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
2072 	if (r)
2073 		return;
2074 
2075 	if (version & 1)
2076 		++version;  /* first time write, random junk */
2077 
2078 	++version;
2079 
2080 	if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
2081 		return;
2082 
2083 	/*
2084 	 * The guest calculates current wall clock time by adding
2085 	 * system time (updated by kvm_guest_time_update below) to the
2086 	 * wall clock specified here.  We do the reverse here.
2087 	 */
2088 	wall_nsec = ktime_get_real_ns() - get_kvmclock_ns(kvm);
2089 
2090 	wc.nsec = do_div(wall_nsec, 1000000000);
2091 	wc.sec = (u32)wall_nsec; /* overflow in 2106 guest time */
2092 	wc.version = version;
2093 
2094 	kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
2095 
2096 	if (sec_hi_ofs) {
2097 		wc_sec_hi = wall_nsec >> 32;
2098 		kvm_write_guest(kvm, wall_clock + sec_hi_ofs,
2099 				&wc_sec_hi, sizeof(wc_sec_hi));
2100 	}
2101 
2102 	version++;
2103 	kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
2104 }
2105 
2106 static void kvm_write_system_time(struct kvm_vcpu *vcpu, gpa_t system_time,
2107 				  bool old_msr, bool host_initiated)
2108 {
2109 	struct kvm_arch *ka = &vcpu->kvm->arch;
2110 
2111 	if (vcpu->vcpu_id == 0 && !host_initiated) {
2112 		if (ka->boot_vcpu_runs_old_kvmclock != old_msr)
2113 			kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2114 
2115 		ka->boot_vcpu_runs_old_kvmclock = old_msr;
2116 	}
2117 
2118 	vcpu->arch.time = system_time;
2119 	kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2120 
2121 	/* we verify if the enable bit is set... */
2122 	vcpu->arch.pv_time_enabled = false;
2123 	if (!(system_time & 1))
2124 		return;
2125 
2126 	if (!kvm_gfn_to_hva_cache_init(vcpu->kvm,
2127 				       &vcpu->arch.pv_time, system_time & ~1ULL,
2128 				       sizeof(struct pvclock_vcpu_time_info)))
2129 		vcpu->arch.pv_time_enabled = true;
2130 
2131 	return;
2132 }
2133 
2134 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
2135 {
2136 	do_shl32_div32(dividend, divisor);
2137 	return dividend;
2138 }
2139 
2140 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
2141 			       s8 *pshift, u32 *pmultiplier)
2142 {
2143 	uint64_t scaled64;
2144 	int32_t  shift = 0;
2145 	uint64_t tps64;
2146 	uint32_t tps32;
2147 
2148 	tps64 = base_hz;
2149 	scaled64 = scaled_hz;
2150 	while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
2151 		tps64 >>= 1;
2152 		shift--;
2153 	}
2154 
2155 	tps32 = (uint32_t)tps64;
2156 	while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
2157 		if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
2158 			scaled64 >>= 1;
2159 		else
2160 			tps32 <<= 1;
2161 		shift++;
2162 	}
2163 
2164 	*pshift = shift;
2165 	*pmultiplier = div_frac(scaled64, tps32);
2166 }
2167 
2168 #ifdef CONFIG_X86_64
2169 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
2170 #endif
2171 
2172 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
2173 static unsigned long max_tsc_khz;
2174 
2175 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
2176 {
2177 	u64 v = (u64)khz * (1000000 + ppm);
2178 	do_div(v, 1000000);
2179 	return v;
2180 }
2181 
2182 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
2183 {
2184 	u64 ratio;
2185 
2186 	/* Guest TSC same frequency as host TSC? */
2187 	if (!scale) {
2188 		vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
2189 		return 0;
2190 	}
2191 
2192 	/* TSC scaling supported? */
2193 	if (!kvm_has_tsc_control) {
2194 		if (user_tsc_khz > tsc_khz) {
2195 			vcpu->arch.tsc_catchup = 1;
2196 			vcpu->arch.tsc_always_catchup = 1;
2197 			return 0;
2198 		} else {
2199 			pr_warn_ratelimited("user requested TSC rate below hardware speed\n");
2200 			return -1;
2201 		}
2202 	}
2203 
2204 	/* TSC scaling required  - calculate ratio */
2205 	ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
2206 				user_tsc_khz, tsc_khz);
2207 
2208 	if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
2209 		pr_warn_ratelimited("Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
2210 			            user_tsc_khz);
2211 		return -1;
2212 	}
2213 
2214 	vcpu->arch.tsc_scaling_ratio = ratio;
2215 	return 0;
2216 }
2217 
2218 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
2219 {
2220 	u32 thresh_lo, thresh_hi;
2221 	int use_scaling = 0;
2222 
2223 	/* tsc_khz can be zero if TSC calibration fails */
2224 	if (user_tsc_khz == 0) {
2225 		/* set tsc_scaling_ratio to a safe value */
2226 		vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
2227 		return -1;
2228 	}
2229 
2230 	/* Compute a scale to convert nanoseconds in TSC cycles */
2231 	kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
2232 			   &vcpu->arch.virtual_tsc_shift,
2233 			   &vcpu->arch.virtual_tsc_mult);
2234 	vcpu->arch.virtual_tsc_khz = user_tsc_khz;
2235 
2236 	/*
2237 	 * Compute the variation in TSC rate which is acceptable
2238 	 * within the range of tolerance and decide if the
2239 	 * rate being applied is within that bounds of the hardware
2240 	 * rate.  If so, no scaling or compensation need be done.
2241 	 */
2242 	thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
2243 	thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
2244 	if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
2245 		pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
2246 		use_scaling = 1;
2247 	}
2248 	return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
2249 }
2250 
2251 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
2252 {
2253 	u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
2254 				      vcpu->arch.virtual_tsc_mult,
2255 				      vcpu->arch.virtual_tsc_shift);
2256 	tsc += vcpu->arch.this_tsc_write;
2257 	return tsc;
2258 }
2259 
2260 static inline int gtod_is_based_on_tsc(int mode)
2261 {
2262 	return mode == VDSO_CLOCKMODE_TSC || mode == VDSO_CLOCKMODE_HVCLOCK;
2263 }
2264 
2265 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
2266 {
2267 #ifdef CONFIG_X86_64
2268 	bool vcpus_matched;
2269 	struct kvm_arch *ka = &vcpu->kvm->arch;
2270 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2271 
2272 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2273 			 atomic_read(&vcpu->kvm->online_vcpus));
2274 
2275 	/*
2276 	 * Once the masterclock is enabled, always perform request in
2277 	 * order to update it.
2278 	 *
2279 	 * In order to enable masterclock, the host clocksource must be TSC
2280 	 * and the vcpus need to have matched TSCs.  When that happens,
2281 	 * perform request to enable masterclock.
2282 	 */
2283 	if (ka->use_master_clock ||
2284 	    (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched))
2285 		kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2286 
2287 	trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
2288 			    atomic_read(&vcpu->kvm->online_vcpus),
2289 		            ka->use_master_clock, gtod->clock.vclock_mode);
2290 #endif
2291 }
2292 
2293 /*
2294  * Multiply tsc by a fixed point number represented by ratio.
2295  *
2296  * The most significant 64-N bits (mult) of ratio represent the
2297  * integral part of the fixed point number; the remaining N bits
2298  * (frac) represent the fractional part, ie. ratio represents a fixed
2299  * point number (mult + frac * 2^(-N)).
2300  *
2301  * N equals to kvm_tsc_scaling_ratio_frac_bits.
2302  */
2303 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
2304 {
2305 	return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
2306 }
2307 
2308 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
2309 {
2310 	u64 _tsc = tsc;
2311 	u64 ratio = vcpu->arch.tsc_scaling_ratio;
2312 
2313 	if (ratio != kvm_default_tsc_scaling_ratio)
2314 		_tsc = __scale_tsc(ratio, tsc);
2315 
2316 	return _tsc;
2317 }
2318 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
2319 
2320 static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
2321 {
2322 	u64 tsc;
2323 
2324 	tsc = kvm_scale_tsc(vcpu, rdtsc());
2325 
2326 	return target_tsc - tsc;
2327 }
2328 
2329 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
2330 {
2331 	return vcpu->arch.l1_tsc_offset + kvm_scale_tsc(vcpu, host_tsc);
2332 }
2333 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
2334 
2335 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
2336 {
2337 	vcpu->arch.l1_tsc_offset = offset;
2338 	vcpu->arch.tsc_offset = static_call(kvm_x86_write_l1_tsc_offset)(vcpu, offset);
2339 }
2340 
2341 static inline bool kvm_check_tsc_unstable(void)
2342 {
2343 #ifdef CONFIG_X86_64
2344 	/*
2345 	 * TSC is marked unstable when we're running on Hyper-V,
2346 	 * 'TSC page' clocksource is good.
2347 	 */
2348 	if (pvclock_gtod_data.clock.vclock_mode == VDSO_CLOCKMODE_HVCLOCK)
2349 		return false;
2350 #endif
2351 	return check_tsc_unstable();
2352 }
2353 
2354 static void kvm_synchronize_tsc(struct kvm_vcpu *vcpu, u64 data)
2355 {
2356 	struct kvm *kvm = vcpu->kvm;
2357 	u64 offset, ns, elapsed;
2358 	unsigned long flags;
2359 	bool matched;
2360 	bool already_matched;
2361 	bool synchronizing = false;
2362 
2363 	raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
2364 	offset = kvm_compute_tsc_offset(vcpu, data);
2365 	ns = get_kvmclock_base_ns();
2366 	elapsed = ns - kvm->arch.last_tsc_nsec;
2367 
2368 	if (vcpu->arch.virtual_tsc_khz) {
2369 		if (data == 0) {
2370 			/*
2371 			 * detection of vcpu initialization -- need to sync
2372 			 * with other vCPUs. This particularly helps to keep
2373 			 * kvm_clock stable after CPU hotplug
2374 			 */
2375 			synchronizing = true;
2376 		} else {
2377 			u64 tsc_exp = kvm->arch.last_tsc_write +
2378 						nsec_to_cycles(vcpu, elapsed);
2379 			u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
2380 			/*
2381 			 * Special case: TSC write with a small delta (1 second)
2382 			 * of virtual cycle time against real time is
2383 			 * interpreted as an attempt to synchronize the CPU.
2384 			 */
2385 			synchronizing = data < tsc_exp + tsc_hz &&
2386 					data + tsc_hz > tsc_exp;
2387 		}
2388 	}
2389 
2390 	/*
2391 	 * For a reliable TSC, we can match TSC offsets, and for an unstable
2392 	 * TSC, we add elapsed time in this computation.  We could let the
2393 	 * compensation code attempt to catch up if we fall behind, but
2394 	 * it's better to try to match offsets from the beginning.
2395          */
2396 	if (synchronizing &&
2397 	    vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
2398 		if (!kvm_check_tsc_unstable()) {
2399 			offset = kvm->arch.cur_tsc_offset;
2400 		} else {
2401 			u64 delta = nsec_to_cycles(vcpu, elapsed);
2402 			data += delta;
2403 			offset = kvm_compute_tsc_offset(vcpu, data);
2404 		}
2405 		matched = true;
2406 		already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
2407 	} else {
2408 		/*
2409 		 * We split periods of matched TSC writes into generations.
2410 		 * For each generation, we track the original measured
2411 		 * nanosecond time, offset, and write, so if TSCs are in
2412 		 * sync, we can match exact offset, and if not, we can match
2413 		 * exact software computation in compute_guest_tsc()
2414 		 *
2415 		 * These values are tracked in kvm->arch.cur_xxx variables.
2416 		 */
2417 		kvm->arch.cur_tsc_generation++;
2418 		kvm->arch.cur_tsc_nsec = ns;
2419 		kvm->arch.cur_tsc_write = data;
2420 		kvm->arch.cur_tsc_offset = offset;
2421 		matched = false;
2422 	}
2423 
2424 	/*
2425 	 * We also track th most recent recorded KHZ, write and time to
2426 	 * allow the matching interval to be extended at each write.
2427 	 */
2428 	kvm->arch.last_tsc_nsec = ns;
2429 	kvm->arch.last_tsc_write = data;
2430 	kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
2431 
2432 	vcpu->arch.last_guest_tsc = data;
2433 
2434 	/* Keep track of which generation this VCPU has synchronized to */
2435 	vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
2436 	vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
2437 	vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
2438 
2439 	kvm_vcpu_write_tsc_offset(vcpu, offset);
2440 	raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
2441 
2442 	spin_lock_irqsave(&kvm->arch.pvclock_gtod_sync_lock, flags);
2443 	if (!matched) {
2444 		kvm->arch.nr_vcpus_matched_tsc = 0;
2445 	} else if (!already_matched) {
2446 		kvm->arch.nr_vcpus_matched_tsc++;
2447 	}
2448 
2449 	kvm_track_tsc_matching(vcpu);
2450 	spin_unlock_irqrestore(&kvm->arch.pvclock_gtod_sync_lock, flags);
2451 }
2452 
2453 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
2454 					   s64 adjustment)
2455 {
2456 	u64 tsc_offset = vcpu->arch.l1_tsc_offset;
2457 	kvm_vcpu_write_tsc_offset(vcpu, tsc_offset + adjustment);
2458 }
2459 
2460 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
2461 {
2462 	if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
2463 		WARN_ON(adjustment < 0);
2464 	adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
2465 	adjust_tsc_offset_guest(vcpu, adjustment);
2466 }
2467 
2468 #ifdef CONFIG_X86_64
2469 
2470 static u64 read_tsc(void)
2471 {
2472 	u64 ret = (u64)rdtsc_ordered();
2473 	u64 last = pvclock_gtod_data.clock.cycle_last;
2474 
2475 	if (likely(ret >= last))
2476 		return ret;
2477 
2478 	/*
2479 	 * GCC likes to generate cmov here, but this branch is extremely
2480 	 * predictable (it's just a function of time and the likely is
2481 	 * very likely) and there's a data dependence, so force GCC
2482 	 * to generate a branch instead.  I don't barrier() because
2483 	 * we don't actually need a barrier, and if this function
2484 	 * ever gets inlined it will generate worse code.
2485 	 */
2486 	asm volatile ("");
2487 	return last;
2488 }
2489 
2490 static inline u64 vgettsc(struct pvclock_clock *clock, u64 *tsc_timestamp,
2491 			  int *mode)
2492 {
2493 	long v;
2494 	u64 tsc_pg_val;
2495 
2496 	switch (clock->vclock_mode) {
2497 	case VDSO_CLOCKMODE_HVCLOCK:
2498 		tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(),
2499 						  tsc_timestamp);
2500 		if (tsc_pg_val != U64_MAX) {
2501 			/* TSC page valid */
2502 			*mode = VDSO_CLOCKMODE_HVCLOCK;
2503 			v = (tsc_pg_val - clock->cycle_last) &
2504 				clock->mask;
2505 		} else {
2506 			/* TSC page invalid */
2507 			*mode = VDSO_CLOCKMODE_NONE;
2508 		}
2509 		break;
2510 	case VDSO_CLOCKMODE_TSC:
2511 		*mode = VDSO_CLOCKMODE_TSC;
2512 		*tsc_timestamp = read_tsc();
2513 		v = (*tsc_timestamp - clock->cycle_last) &
2514 			clock->mask;
2515 		break;
2516 	default:
2517 		*mode = VDSO_CLOCKMODE_NONE;
2518 	}
2519 
2520 	if (*mode == VDSO_CLOCKMODE_NONE)
2521 		*tsc_timestamp = v = 0;
2522 
2523 	return v * clock->mult;
2524 }
2525 
2526 static int do_monotonic_raw(s64 *t, u64 *tsc_timestamp)
2527 {
2528 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2529 	unsigned long seq;
2530 	int mode;
2531 	u64 ns;
2532 
2533 	do {
2534 		seq = read_seqcount_begin(&gtod->seq);
2535 		ns = gtod->raw_clock.base_cycles;
2536 		ns += vgettsc(&gtod->raw_clock, tsc_timestamp, &mode);
2537 		ns >>= gtod->raw_clock.shift;
2538 		ns += ktime_to_ns(ktime_add(gtod->raw_clock.offset, gtod->offs_boot));
2539 	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
2540 	*t = ns;
2541 
2542 	return mode;
2543 }
2544 
2545 static int do_realtime(struct timespec64 *ts, u64 *tsc_timestamp)
2546 {
2547 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2548 	unsigned long seq;
2549 	int mode;
2550 	u64 ns;
2551 
2552 	do {
2553 		seq = read_seqcount_begin(&gtod->seq);
2554 		ts->tv_sec = gtod->wall_time_sec;
2555 		ns = gtod->clock.base_cycles;
2556 		ns += vgettsc(&gtod->clock, tsc_timestamp, &mode);
2557 		ns >>= gtod->clock.shift;
2558 	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
2559 
2560 	ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
2561 	ts->tv_nsec = ns;
2562 
2563 	return mode;
2564 }
2565 
2566 /* returns true if host is using TSC based clocksource */
2567 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp)
2568 {
2569 	/* checked again under seqlock below */
2570 	if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2571 		return false;
2572 
2573 	return gtod_is_based_on_tsc(do_monotonic_raw(kernel_ns,
2574 						      tsc_timestamp));
2575 }
2576 
2577 /* returns true if host is using TSC based clocksource */
2578 static bool kvm_get_walltime_and_clockread(struct timespec64 *ts,
2579 					   u64 *tsc_timestamp)
2580 {
2581 	/* checked again under seqlock below */
2582 	if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2583 		return false;
2584 
2585 	return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp));
2586 }
2587 #endif
2588 
2589 /*
2590  *
2591  * Assuming a stable TSC across physical CPUS, and a stable TSC
2592  * across virtual CPUs, the following condition is possible.
2593  * Each numbered line represents an event visible to both
2594  * CPUs at the next numbered event.
2595  *
2596  * "timespecX" represents host monotonic time. "tscX" represents
2597  * RDTSC value.
2598  *
2599  * 		VCPU0 on CPU0		|	VCPU1 on CPU1
2600  *
2601  * 1.  read timespec0,tsc0
2602  * 2.					| timespec1 = timespec0 + N
2603  * 					| tsc1 = tsc0 + M
2604  * 3. transition to guest		| transition to guest
2605  * 4. ret0 = timespec0 + (rdtsc - tsc0) |
2606  * 5.				        | ret1 = timespec1 + (rdtsc - tsc1)
2607  * 				        | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
2608  *
2609  * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
2610  *
2611  * 	- ret0 < ret1
2612  *	- timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
2613  *		...
2614  *	- 0 < N - M => M < N
2615  *
2616  * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
2617  * always the case (the difference between two distinct xtime instances
2618  * might be smaller then the difference between corresponding TSC reads,
2619  * when updating guest vcpus pvclock areas).
2620  *
2621  * To avoid that problem, do not allow visibility of distinct
2622  * system_timestamp/tsc_timestamp values simultaneously: use a master
2623  * copy of host monotonic time values. Update that master copy
2624  * in lockstep.
2625  *
2626  * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
2627  *
2628  */
2629 
2630 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
2631 {
2632 #ifdef CONFIG_X86_64
2633 	struct kvm_arch *ka = &kvm->arch;
2634 	int vclock_mode;
2635 	bool host_tsc_clocksource, vcpus_matched;
2636 
2637 	vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2638 			atomic_read(&kvm->online_vcpus));
2639 
2640 	/*
2641 	 * If the host uses TSC clock, then passthrough TSC as stable
2642 	 * to the guest.
2643 	 */
2644 	host_tsc_clocksource = kvm_get_time_and_clockread(
2645 					&ka->master_kernel_ns,
2646 					&ka->master_cycle_now);
2647 
2648 	ka->use_master_clock = host_tsc_clocksource && vcpus_matched
2649 				&& !ka->backwards_tsc_observed
2650 				&& !ka->boot_vcpu_runs_old_kvmclock;
2651 
2652 	if (ka->use_master_clock)
2653 		atomic_set(&kvm_guest_has_master_clock, 1);
2654 
2655 	vclock_mode = pvclock_gtod_data.clock.vclock_mode;
2656 	trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
2657 					vcpus_matched);
2658 #endif
2659 }
2660 
2661 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
2662 {
2663 	kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
2664 }
2665 
2666 static void kvm_gen_update_masterclock(struct kvm *kvm)
2667 {
2668 #ifdef CONFIG_X86_64
2669 	int i;
2670 	struct kvm_vcpu *vcpu;
2671 	struct kvm_arch *ka = &kvm->arch;
2672 	unsigned long flags;
2673 
2674 	kvm_hv_invalidate_tsc_page(kvm);
2675 
2676 	kvm_make_mclock_inprogress_request(kvm);
2677 
2678 	/* no guest entries from this point */
2679 	spin_lock_irqsave(&ka->pvclock_gtod_sync_lock, flags);
2680 	pvclock_update_vm_gtod_copy(kvm);
2681 	spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
2682 
2683 	kvm_for_each_vcpu(i, vcpu, kvm)
2684 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2685 
2686 	/* guest entries allowed */
2687 	kvm_for_each_vcpu(i, vcpu, kvm)
2688 		kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
2689 #endif
2690 }
2691 
2692 u64 get_kvmclock_ns(struct kvm *kvm)
2693 {
2694 	struct kvm_arch *ka = &kvm->arch;
2695 	struct pvclock_vcpu_time_info hv_clock;
2696 	unsigned long flags;
2697 	u64 ret;
2698 
2699 	spin_lock_irqsave(&ka->pvclock_gtod_sync_lock, flags);
2700 	if (!ka->use_master_clock) {
2701 		spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
2702 		return get_kvmclock_base_ns() + ka->kvmclock_offset;
2703 	}
2704 
2705 	hv_clock.tsc_timestamp = ka->master_cycle_now;
2706 	hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
2707 	spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
2708 
2709 	/* both __this_cpu_read() and rdtsc() should be on the same cpu */
2710 	get_cpu();
2711 
2712 	if (__this_cpu_read(cpu_tsc_khz)) {
2713 		kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
2714 				   &hv_clock.tsc_shift,
2715 				   &hv_clock.tsc_to_system_mul);
2716 		ret = __pvclock_read_cycles(&hv_clock, rdtsc());
2717 	} else
2718 		ret = get_kvmclock_base_ns() + ka->kvmclock_offset;
2719 
2720 	put_cpu();
2721 
2722 	return ret;
2723 }
2724 
2725 static void kvm_setup_pvclock_page(struct kvm_vcpu *v,
2726 				   struct gfn_to_hva_cache *cache,
2727 				   unsigned int offset)
2728 {
2729 	struct kvm_vcpu_arch *vcpu = &v->arch;
2730 	struct pvclock_vcpu_time_info guest_hv_clock;
2731 
2732 	if (unlikely(kvm_read_guest_offset_cached(v->kvm, cache,
2733 		&guest_hv_clock, offset, sizeof(guest_hv_clock))))
2734 		return;
2735 
2736 	/* This VCPU is paused, but it's legal for a guest to read another
2737 	 * VCPU's kvmclock, so we really have to follow the specification where
2738 	 * it says that version is odd if data is being modified, and even after
2739 	 * it is consistent.
2740 	 *
2741 	 * Version field updates must be kept separate.  This is because
2742 	 * kvm_write_guest_cached might use a "rep movs" instruction, and
2743 	 * writes within a string instruction are weakly ordered.  So there
2744 	 * are three writes overall.
2745 	 *
2746 	 * As a small optimization, only write the version field in the first
2747 	 * and third write.  The vcpu->pv_time cache is still valid, because the
2748 	 * version field is the first in the struct.
2749 	 */
2750 	BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
2751 
2752 	if (guest_hv_clock.version & 1)
2753 		++guest_hv_clock.version;  /* first time write, random junk */
2754 
2755 	vcpu->hv_clock.version = guest_hv_clock.version + 1;
2756 	kvm_write_guest_offset_cached(v->kvm, cache,
2757 				      &vcpu->hv_clock, offset,
2758 				      sizeof(vcpu->hv_clock.version));
2759 
2760 	smp_wmb();
2761 
2762 	/* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
2763 	vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
2764 
2765 	if (vcpu->pvclock_set_guest_stopped_request) {
2766 		vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
2767 		vcpu->pvclock_set_guest_stopped_request = false;
2768 	}
2769 
2770 	trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
2771 
2772 	kvm_write_guest_offset_cached(v->kvm, cache,
2773 				      &vcpu->hv_clock, offset,
2774 				      sizeof(vcpu->hv_clock));
2775 
2776 	smp_wmb();
2777 
2778 	vcpu->hv_clock.version++;
2779 	kvm_write_guest_offset_cached(v->kvm, cache,
2780 				     &vcpu->hv_clock, offset,
2781 				     sizeof(vcpu->hv_clock.version));
2782 }
2783 
2784 static int kvm_guest_time_update(struct kvm_vcpu *v)
2785 {
2786 	unsigned long flags, tgt_tsc_khz;
2787 	struct kvm_vcpu_arch *vcpu = &v->arch;
2788 	struct kvm_arch *ka = &v->kvm->arch;
2789 	s64 kernel_ns;
2790 	u64 tsc_timestamp, host_tsc;
2791 	u8 pvclock_flags;
2792 	bool use_master_clock;
2793 
2794 	kernel_ns = 0;
2795 	host_tsc = 0;
2796 
2797 	/*
2798 	 * If the host uses TSC clock, then passthrough TSC as stable
2799 	 * to the guest.
2800 	 */
2801 	spin_lock_irqsave(&ka->pvclock_gtod_sync_lock, flags);
2802 	use_master_clock = ka->use_master_clock;
2803 	if (use_master_clock) {
2804 		host_tsc = ka->master_cycle_now;
2805 		kernel_ns = ka->master_kernel_ns;
2806 	}
2807 	spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
2808 
2809 	/* Keep irq disabled to prevent changes to the clock */
2810 	local_irq_save(flags);
2811 	tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
2812 	if (unlikely(tgt_tsc_khz == 0)) {
2813 		local_irq_restore(flags);
2814 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2815 		return 1;
2816 	}
2817 	if (!use_master_clock) {
2818 		host_tsc = rdtsc();
2819 		kernel_ns = get_kvmclock_base_ns();
2820 	}
2821 
2822 	tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
2823 
2824 	/*
2825 	 * We may have to catch up the TSC to match elapsed wall clock
2826 	 * time for two reasons, even if kvmclock is used.
2827 	 *   1) CPU could have been running below the maximum TSC rate
2828 	 *   2) Broken TSC compensation resets the base at each VCPU
2829 	 *      entry to avoid unknown leaps of TSC even when running
2830 	 *      again on the same CPU.  This may cause apparent elapsed
2831 	 *      time to disappear, and the guest to stand still or run
2832 	 *	very slowly.
2833 	 */
2834 	if (vcpu->tsc_catchup) {
2835 		u64 tsc = compute_guest_tsc(v, kernel_ns);
2836 		if (tsc > tsc_timestamp) {
2837 			adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
2838 			tsc_timestamp = tsc;
2839 		}
2840 	}
2841 
2842 	local_irq_restore(flags);
2843 
2844 	/* With all the info we got, fill in the values */
2845 
2846 	if (kvm_has_tsc_control)
2847 		tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
2848 
2849 	if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
2850 		kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
2851 				   &vcpu->hv_clock.tsc_shift,
2852 				   &vcpu->hv_clock.tsc_to_system_mul);
2853 		vcpu->hw_tsc_khz = tgt_tsc_khz;
2854 	}
2855 
2856 	vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
2857 	vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
2858 	vcpu->last_guest_tsc = tsc_timestamp;
2859 
2860 	/* If the host uses TSC clocksource, then it is stable */
2861 	pvclock_flags = 0;
2862 	if (use_master_clock)
2863 		pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
2864 
2865 	vcpu->hv_clock.flags = pvclock_flags;
2866 
2867 	if (vcpu->pv_time_enabled)
2868 		kvm_setup_pvclock_page(v, &vcpu->pv_time, 0);
2869 	if (vcpu->xen.vcpu_info_set)
2870 		kvm_setup_pvclock_page(v, &vcpu->xen.vcpu_info_cache,
2871 				       offsetof(struct compat_vcpu_info, time));
2872 	if (vcpu->xen.vcpu_time_info_set)
2873 		kvm_setup_pvclock_page(v, &vcpu->xen.vcpu_time_info_cache, 0);
2874 	if (v == kvm_get_vcpu(v->kvm, 0))
2875 		kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
2876 	return 0;
2877 }
2878 
2879 /*
2880  * kvmclock updates which are isolated to a given vcpu, such as
2881  * vcpu->cpu migration, should not allow system_timestamp from
2882  * the rest of the vcpus to remain static. Otherwise ntp frequency
2883  * correction applies to one vcpu's system_timestamp but not
2884  * the others.
2885  *
2886  * So in those cases, request a kvmclock update for all vcpus.
2887  * We need to rate-limit these requests though, as they can
2888  * considerably slow guests that have a large number of vcpus.
2889  * The time for a remote vcpu to update its kvmclock is bound
2890  * by the delay we use to rate-limit the updates.
2891  */
2892 
2893 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
2894 
2895 static void kvmclock_update_fn(struct work_struct *work)
2896 {
2897 	int i;
2898 	struct delayed_work *dwork = to_delayed_work(work);
2899 	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2900 					   kvmclock_update_work);
2901 	struct kvm *kvm = container_of(ka, struct kvm, arch);
2902 	struct kvm_vcpu *vcpu;
2903 
2904 	kvm_for_each_vcpu(i, vcpu, kvm) {
2905 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2906 		kvm_vcpu_kick(vcpu);
2907 	}
2908 }
2909 
2910 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
2911 {
2912 	struct kvm *kvm = v->kvm;
2913 
2914 	kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2915 	schedule_delayed_work(&kvm->arch.kvmclock_update_work,
2916 					KVMCLOCK_UPDATE_DELAY);
2917 }
2918 
2919 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
2920 
2921 static void kvmclock_sync_fn(struct work_struct *work)
2922 {
2923 	struct delayed_work *dwork = to_delayed_work(work);
2924 	struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2925 					   kvmclock_sync_work);
2926 	struct kvm *kvm = container_of(ka, struct kvm, arch);
2927 
2928 	if (!kvmclock_periodic_sync)
2929 		return;
2930 
2931 	schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
2932 	schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
2933 					KVMCLOCK_SYNC_PERIOD);
2934 }
2935 
2936 /*
2937  * On AMD, HWCR[McStatusWrEn] controls whether setting MCi_STATUS results in #GP.
2938  */
2939 static bool can_set_mci_status(struct kvm_vcpu *vcpu)
2940 {
2941 	/* McStatusWrEn enabled? */
2942 	if (guest_cpuid_is_amd_or_hygon(vcpu))
2943 		return !!(vcpu->arch.msr_hwcr & BIT_ULL(18));
2944 
2945 	return false;
2946 }
2947 
2948 static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2949 {
2950 	u64 mcg_cap = vcpu->arch.mcg_cap;
2951 	unsigned bank_num = mcg_cap & 0xff;
2952 	u32 msr = msr_info->index;
2953 	u64 data = msr_info->data;
2954 
2955 	switch (msr) {
2956 	case MSR_IA32_MCG_STATUS:
2957 		vcpu->arch.mcg_status = data;
2958 		break;
2959 	case MSR_IA32_MCG_CTL:
2960 		if (!(mcg_cap & MCG_CTL_P) &&
2961 		    (data || !msr_info->host_initiated))
2962 			return 1;
2963 		if (data != 0 && data != ~(u64)0)
2964 			return 1;
2965 		vcpu->arch.mcg_ctl = data;
2966 		break;
2967 	default:
2968 		if (msr >= MSR_IA32_MC0_CTL &&
2969 		    msr < MSR_IA32_MCx_CTL(bank_num)) {
2970 			u32 offset = array_index_nospec(
2971 				msr - MSR_IA32_MC0_CTL,
2972 				MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
2973 
2974 			/* only 0 or all 1s can be written to IA32_MCi_CTL
2975 			 * some Linux kernels though clear bit 10 in bank 4 to
2976 			 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
2977 			 * this to avoid an uncatched #GP in the guest
2978 			 */
2979 			if ((offset & 0x3) == 0 &&
2980 			    data != 0 && (data | (1 << 10)) != ~(u64)0)
2981 				return -1;
2982 
2983 			/* MCi_STATUS */
2984 			if (!msr_info->host_initiated &&
2985 			    (offset & 0x3) == 1 && data != 0) {
2986 				if (!can_set_mci_status(vcpu))
2987 					return -1;
2988 			}
2989 
2990 			vcpu->arch.mce_banks[offset] = data;
2991 			break;
2992 		}
2993 		return 1;
2994 	}
2995 	return 0;
2996 }
2997 
2998 static inline bool kvm_pv_async_pf_enabled(struct kvm_vcpu *vcpu)
2999 {
3000 	u64 mask = KVM_ASYNC_PF_ENABLED | KVM_ASYNC_PF_DELIVERY_AS_INT;
3001 
3002 	return (vcpu->arch.apf.msr_en_val & mask) == mask;
3003 }
3004 
3005 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
3006 {
3007 	gpa_t gpa = data & ~0x3f;
3008 
3009 	/* Bits 4:5 are reserved, Should be zero */
3010 	if (data & 0x30)
3011 		return 1;
3012 
3013 	if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_VMEXIT) &&
3014 	    (data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT))
3015 		return 1;
3016 
3017 	if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT) &&
3018 	    (data & KVM_ASYNC_PF_DELIVERY_AS_INT))
3019 		return 1;
3020 
3021 	if (!lapic_in_kernel(vcpu))
3022 		return data ? 1 : 0;
3023 
3024 	vcpu->arch.apf.msr_en_val = data;
3025 
3026 	if (!kvm_pv_async_pf_enabled(vcpu)) {
3027 		kvm_clear_async_pf_completion_queue(vcpu);
3028 		kvm_async_pf_hash_reset(vcpu);
3029 		return 0;
3030 	}
3031 
3032 	if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
3033 					sizeof(u64)))
3034 		return 1;
3035 
3036 	vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
3037 	vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
3038 
3039 	kvm_async_pf_wakeup_all(vcpu);
3040 
3041 	return 0;
3042 }
3043 
3044 static int kvm_pv_enable_async_pf_int(struct kvm_vcpu *vcpu, u64 data)
3045 {
3046 	/* Bits 8-63 are reserved */
3047 	if (data >> 8)
3048 		return 1;
3049 
3050 	if (!lapic_in_kernel(vcpu))
3051 		return 1;
3052 
3053 	vcpu->arch.apf.msr_int_val = data;
3054 
3055 	vcpu->arch.apf.vec = data & KVM_ASYNC_PF_VEC_MASK;
3056 
3057 	return 0;
3058 }
3059 
3060 static void kvmclock_reset(struct kvm_vcpu *vcpu)
3061 {
3062 	vcpu->arch.pv_time_enabled = false;
3063 	vcpu->arch.time = 0;
3064 }
3065 
3066 static void kvm_vcpu_flush_tlb_all(struct kvm_vcpu *vcpu)
3067 {
3068 	++vcpu->stat.tlb_flush;
3069 	static_call(kvm_x86_tlb_flush_all)(vcpu);
3070 }
3071 
3072 static void kvm_vcpu_flush_tlb_guest(struct kvm_vcpu *vcpu)
3073 {
3074 	++vcpu->stat.tlb_flush;
3075 
3076 	if (!tdp_enabled) {
3077                /*
3078 		 * A TLB flush on behalf of the guest is equivalent to
3079 		 * INVPCID(all), toggling CR4.PGE, etc., which requires
3080 		 * a forced sync of the shadow page tables.  Unload the
3081 		 * entire MMU here and the subsequent load will sync the
3082 		 * shadow page tables, and also flush the TLB.
3083 		 */
3084 		kvm_mmu_unload(vcpu);
3085 		return;
3086 	}
3087 
3088 	static_call(kvm_x86_tlb_flush_guest)(vcpu);
3089 }
3090 
3091 static void record_steal_time(struct kvm_vcpu *vcpu)
3092 {
3093 	struct kvm_host_map map;
3094 	struct kvm_steal_time *st;
3095 
3096 	if (kvm_xen_msr_enabled(vcpu->kvm)) {
3097 		kvm_xen_runstate_set_running(vcpu);
3098 		return;
3099 	}
3100 
3101 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
3102 		return;
3103 
3104 	/* -EAGAIN is returned in atomic context so we can just return. */
3105 	if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT,
3106 			&map, &vcpu->arch.st.cache, false))
3107 		return;
3108 
3109 	st = map.hva +
3110 		offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS);
3111 
3112 	/*
3113 	 * Doing a TLB flush here, on the guest's behalf, can avoid
3114 	 * expensive IPIs.
3115 	 */
3116 	if (guest_pv_has(vcpu, KVM_FEATURE_PV_TLB_FLUSH)) {
3117 		u8 st_preempted = xchg(&st->preempted, 0);
3118 
3119 		trace_kvm_pv_tlb_flush(vcpu->vcpu_id,
3120 				       st_preempted & KVM_VCPU_FLUSH_TLB);
3121 		if (st_preempted & KVM_VCPU_FLUSH_TLB)
3122 			kvm_vcpu_flush_tlb_guest(vcpu);
3123 	} else {
3124 		st->preempted = 0;
3125 	}
3126 
3127 	vcpu->arch.st.preempted = 0;
3128 
3129 	if (st->version & 1)
3130 		st->version += 1;  /* first time write, random junk */
3131 
3132 	st->version += 1;
3133 
3134 	smp_wmb();
3135 
3136 	st->steal += current->sched_info.run_delay -
3137 		vcpu->arch.st.last_steal;
3138 	vcpu->arch.st.last_steal = current->sched_info.run_delay;
3139 
3140 	smp_wmb();
3141 
3142 	st->version += 1;
3143 
3144 	kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, false);
3145 }
3146 
3147 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3148 {
3149 	bool pr = false;
3150 	u32 msr = msr_info->index;
3151 	u64 data = msr_info->data;
3152 
3153 	if (msr && msr == vcpu->kvm->arch.xen_hvm_config.msr)
3154 		return kvm_xen_write_hypercall_page(vcpu, data);
3155 
3156 	switch (msr) {
3157 	case MSR_AMD64_NB_CFG:
3158 	case MSR_IA32_UCODE_WRITE:
3159 	case MSR_VM_HSAVE_PA:
3160 	case MSR_AMD64_PATCH_LOADER:
3161 	case MSR_AMD64_BU_CFG2:
3162 	case MSR_AMD64_DC_CFG:
3163 	case MSR_F15H_EX_CFG:
3164 		break;
3165 
3166 	case MSR_IA32_UCODE_REV:
3167 		if (msr_info->host_initiated)
3168 			vcpu->arch.microcode_version = data;
3169 		break;
3170 	case MSR_IA32_ARCH_CAPABILITIES:
3171 		if (!msr_info->host_initiated)
3172 			return 1;
3173 		vcpu->arch.arch_capabilities = data;
3174 		break;
3175 	case MSR_IA32_PERF_CAPABILITIES: {
3176 		struct kvm_msr_entry msr_ent = {.index = msr, .data = 0};
3177 
3178 		if (!msr_info->host_initiated)
3179 			return 1;
3180 		if (guest_cpuid_has(vcpu, X86_FEATURE_PDCM) && kvm_get_msr_feature(&msr_ent))
3181 			return 1;
3182 		if (data & ~msr_ent.data)
3183 			return 1;
3184 
3185 		vcpu->arch.perf_capabilities = data;
3186 
3187 		return 0;
3188 		}
3189 	case MSR_EFER:
3190 		return set_efer(vcpu, msr_info);
3191 	case MSR_K7_HWCR:
3192 		data &= ~(u64)0x40;	/* ignore flush filter disable */
3193 		data &= ~(u64)0x100;	/* ignore ignne emulation enable */
3194 		data &= ~(u64)0x8;	/* ignore TLB cache disable */
3195 
3196 		/* Handle McStatusWrEn */
3197 		if (data == BIT_ULL(18)) {
3198 			vcpu->arch.msr_hwcr = data;
3199 		} else if (data != 0) {
3200 			vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
3201 				    data);
3202 			return 1;
3203 		}
3204 		break;
3205 	case MSR_FAM10H_MMIO_CONF_BASE:
3206 		if (data != 0) {
3207 			vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
3208 				    "0x%llx\n", data);
3209 			return 1;
3210 		}
3211 		break;
3212 	case 0x200 ... 0x2ff:
3213 		return kvm_mtrr_set_msr(vcpu, msr, data);
3214 	case MSR_IA32_APICBASE:
3215 		return kvm_set_apic_base(vcpu, msr_info);
3216 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
3217 		return kvm_x2apic_msr_write(vcpu, msr, data);
3218 	case MSR_IA32_TSC_DEADLINE:
3219 		kvm_set_lapic_tscdeadline_msr(vcpu, data);
3220 		break;
3221 	case MSR_IA32_TSC_ADJUST:
3222 		if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
3223 			if (!msr_info->host_initiated) {
3224 				s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
3225 				adjust_tsc_offset_guest(vcpu, adj);
3226 			}
3227 			vcpu->arch.ia32_tsc_adjust_msr = data;
3228 		}
3229 		break;
3230 	case MSR_IA32_MISC_ENABLE:
3231 		if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT) &&
3232 		    ((vcpu->arch.ia32_misc_enable_msr ^ data) & MSR_IA32_MISC_ENABLE_MWAIT)) {
3233 			if (!guest_cpuid_has(vcpu, X86_FEATURE_XMM3))
3234 				return 1;
3235 			vcpu->arch.ia32_misc_enable_msr = data;
3236 			kvm_update_cpuid_runtime(vcpu);
3237 		} else {
3238 			vcpu->arch.ia32_misc_enable_msr = data;
3239 		}
3240 		break;
3241 	case MSR_IA32_SMBASE:
3242 		if (!msr_info->host_initiated)
3243 			return 1;
3244 		vcpu->arch.smbase = data;
3245 		break;
3246 	case MSR_IA32_POWER_CTL:
3247 		vcpu->arch.msr_ia32_power_ctl = data;
3248 		break;
3249 	case MSR_IA32_TSC:
3250 		if (msr_info->host_initiated) {
3251 			kvm_synchronize_tsc(vcpu, data);
3252 		} else {
3253 			u64 adj = kvm_compute_tsc_offset(vcpu, data) - vcpu->arch.l1_tsc_offset;
3254 			adjust_tsc_offset_guest(vcpu, adj);
3255 			vcpu->arch.ia32_tsc_adjust_msr += adj;
3256 		}
3257 		break;
3258 	case MSR_IA32_XSS:
3259 		if (!msr_info->host_initiated &&
3260 		    !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3261 			return 1;
3262 		/*
3263 		 * KVM supports exposing PT to the guest, but does not support
3264 		 * IA32_XSS[bit 8]. Guests have to use RDMSR/WRMSR rather than
3265 		 * XSAVES/XRSTORS to save/restore PT MSRs.
3266 		 */
3267 		if (data & ~supported_xss)
3268 			return 1;
3269 		vcpu->arch.ia32_xss = data;
3270 		break;
3271 	case MSR_SMI_COUNT:
3272 		if (!msr_info->host_initiated)
3273 			return 1;
3274 		vcpu->arch.smi_count = data;
3275 		break;
3276 	case MSR_KVM_WALL_CLOCK_NEW:
3277 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3278 			return 1;
3279 
3280 		vcpu->kvm->arch.wall_clock = data;
3281 		kvm_write_wall_clock(vcpu->kvm, data, 0);
3282 		break;
3283 	case MSR_KVM_WALL_CLOCK:
3284 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3285 			return 1;
3286 
3287 		vcpu->kvm->arch.wall_clock = data;
3288 		kvm_write_wall_clock(vcpu->kvm, data, 0);
3289 		break;
3290 	case MSR_KVM_SYSTEM_TIME_NEW:
3291 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3292 			return 1;
3293 
3294 		kvm_write_system_time(vcpu, data, false, msr_info->host_initiated);
3295 		break;
3296 	case MSR_KVM_SYSTEM_TIME:
3297 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3298 			return 1;
3299 
3300 		kvm_write_system_time(vcpu, data, true,  msr_info->host_initiated);
3301 		break;
3302 	case MSR_KVM_ASYNC_PF_EN:
3303 		if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3304 			return 1;
3305 
3306 		if (kvm_pv_enable_async_pf(vcpu, data))
3307 			return 1;
3308 		break;
3309 	case MSR_KVM_ASYNC_PF_INT:
3310 		if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3311 			return 1;
3312 
3313 		if (kvm_pv_enable_async_pf_int(vcpu, data))
3314 			return 1;
3315 		break;
3316 	case MSR_KVM_ASYNC_PF_ACK:
3317 		if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3318 			return 1;
3319 		if (data & 0x1) {
3320 			vcpu->arch.apf.pageready_pending = false;
3321 			kvm_check_async_pf_completion(vcpu);
3322 		}
3323 		break;
3324 	case MSR_KVM_STEAL_TIME:
3325 		if (!guest_pv_has(vcpu, KVM_FEATURE_STEAL_TIME))
3326 			return 1;
3327 
3328 		if (unlikely(!sched_info_on()))
3329 			return 1;
3330 
3331 		if (data & KVM_STEAL_RESERVED_MASK)
3332 			return 1;
3333 
3334 		vcpu->arch.st.msr_val = data;
3335 
3336 		if (!(data & KVM_MSR_ENABLED))
3337 			break;
3338 
3339 		kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3340 
3341 		break;
3342 	case MSR_KVM_PV_EOI_EN:
3343 		if (!guest_pv_has(vcpu, KVM_FEATURE_PV_EOI))
3344 			return 1;
3345 
3346 		if (kvm_lapic_enable_pv_eoi(vcpu, data, sizeof(u8)))
3347 			return 1;
3348 		break;
3349 
3350 	case MSR_KVM_POLL_CONTROL:
3351 		if (!guest_pv_has(vcpu, KVM_FEATURE_POLL_CONTROL))
3352 			return 1;
3353 
3354 		/* only enable bit supported */
3355 		if (data & (-1ULL << 1))
3356 			return 1;
3357 
3358 		vcpu->arch.msr_kvm_poll_control = data;
3359 		break;
3360 
3361 	case MSR_IA32_MCG_CTL:
3362 	case MSR_IA32_MCG_STATUS:
3363 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3364 		return set_msr_mce(vcpu, msr_info);
3365 
3366 	case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3367 	case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3368 		pr = true;
3369 		fallthrough;
3370 	case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3371 	case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3372 		if (kvm_pmu_is_valid_msr(vcpu, msr))
3373 			return kvm_pmu_set_msr(vcpu, msr_info);
3374 
3375 		if (pr || data != 0)
3376 			vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
3377 				    "0x%x data 0x%llx\n", msr, data);
3378 		break;
3379 	case MSR_K7_CLK_CTL:
3380 		/*
3381 		 * Ignore all writes to this no longer documented MSR.
3382 		 * Writes are only relevant for old K7 processors,
3383 		 * all pre-dating SVM, but a recommended workaround from
3384 		 * AMD for these chips. It is possible to specify the
3385 		 * affected processor models on the command line, hence
3386 		 * the need to ignore the workaround.
3387 		 */
3388 		break;
3389 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3390 	case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
3391 	case HV_X64_MSR_SYNDBG_OPTIONS:
3392 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3393 	case HV_X64_MSR_CRASH_CTL:
3394 	case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3395 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3396 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
3397 	case HV_X64_MSR_TSC_EMULATION_STATUS:
3398 		return kvm_hv_set_msr_common(vcpu, msr, data,
3399 					     msr_info->host_initiated);
3400 	case MSR_IA32_BBL_CR_CTL3:
3401 		/* Drop writes to this legacy MSR -- see rdmsr
3402 		 * counterpart for further detail.
3403 		 */
3404 		if (report_ignored_msrs)
3405 			vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
3406 				msr, data);
3407 		break;
3408 	case MSR_AMD64_OSVW_ID_LENGTH:
3409 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3410 			return 1;
3411 		vcpu->arch.osvw.length = data;
3412 		break;
3413 	case MSR_AMD64_OSVW_STATUS:
3414 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3415 			return 1;
3416 		vcpu->arch.osvw.status = data;
3417 		break;
3418 	case MSR_PLATFORM_INFO:
3419 		if (!msr_info->host_initiated ||
3420 		    (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
3421 		     cpuid_fault_enabled(vcpu)))
3422 			return 1;
3423 		vcpu->arch.msr_platform_info = data;
3424 		break;
3425 	case MSR_MISC_FEATURES_ENABLES:
3426 		if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
3427 		    (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
3428 		     !supports_cpuid_fault(vcpu)))
3429 			return 1;
3430 		vcpu->arch.msr_misc_features_enables = data;
3431 		break;
3432 	default:
3433 		if (kvm_pmu_is_valid_msr(vcpu, msr))
3434 			return kvm_pmu_set_msr(vcpu, msr_info);
3435 		return KVM_MSR_RET_INVALID;
3436 	}
3437 	return 0;
3438 }
3439 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
3440 
3441 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
3442 {
3443 	u64 data;
3444 	u64 mcg_cap = vcpu->arch.mcg_cap;
3445 	unsigned bank_num = mcg_cap & 0xff;
3446 
3447 	switch (msr) {
3448 	case MSR_IA32_P5_MC_ADDR:
3449 	case MSR_IA32_P5_MC_TYPE:
3450 		data = 0;
3451 		break;
3452 	case MSR_IA32_MCG_CAP:
3453 		data = vcpu->arch.mcg_cap;
3454 		break;
3455 	case MSR_IA32_MCG_CTL:
3456 		if (!(mcg_cap & MCG_CTL_P) && !host)
3457 			return 1;
3458 		data = vcpu->arch.mcg_ctl;
3459 		break;
3460 	case MSR_IA32_MCG_STATUS:
3461 		data = vcpu->arch.mcg_status;
3462 		break;
3463 	default:
3464 		if (msr >= MSR_IA32_MC0_CTL &&
3465 		    msr < MSR_IA32_MCx_CTL(bank_num)) {
3466 			u32 offset = array_index_nospec(
3467 				msr - MSR_IA32_MC0_CTL,
3468 				MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
3469 
3470 			data = vcpu->arch.mce_banks[offset];
3471 			break;
3472 		}
3473 		return 1;
3474 	}
3475 	*pdata = data;
3476 	return 0;
3477 }
3478 
3479 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3480 {
3481 	switch (msr_info->index) {
3482 	case MSR_IA32_PLATFORM_ID:
3483 	case MSR_IA32_EBL_CR_POWERON:
3484 	case MSR_IA32_LASTBRANCHFROMIP:
3485 	case MSR_IA32_LASTBRANCHTOIP:
3486 	case MSR_IA32_LASTINTFROMIP:
3487 	case MSR_IA32_LASTINTTOIP:
3488 	case MSR_AMD64_SYSCFG:
3489 	case MSR_K8_TSEG_ADDR:
3490 	case MSR_K8_TSEG_MASK:
3491 	case MSR_VM_HSAVE_PA:
3492 	case MSR_K8_INT_PENDING_MSG:
3493 	case MSR_AMD64_NB_CFG:
3494 	case MSR_FAM10H_MMIO_CONF_BASE:
3495 	case MSR_AMD64_BU_CFG2:
3496 	case MSR_IA32_PERF_CTL:
3497 	case MSR_AMD64_DC_CFG:
3498 	case MSR_F15H_EX_CFG:
3499 	/*
3500 	 * Intel Sandy Bridge CPUs must support the RAPL (running average power
3501 	 * limit) MSRs. Just return 0, as we do not want to expose the host
3502 	 * data here. Do not conditionalize this on CPUID, as KVM does not do
3503 	 * so for existing CPU-specific MSRs.
3504 	 */
3505 	case MSR_RAPL_POWER_UNIT:
3506 	case MSR_PP0_ENERGY_STATUS:	/* Power plane 0 (core) */
3507 	case MSR_PP1_ENERGY_STATUS:	/* Power plane 1 (graphics uncore) */
3508 	case MSR_PKG_ENERGY_STATUS:	/* Total package */
3509 	case MSR_DRAM_ENERGY_STATUS:	/* DRAM controller */
3510 		msr_info->data = 0;
3511 		break;
3512 	case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
3513 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3514 			return kvm_pmu_get_msr(vcpu, msr_info);
3515 		if (!msr_info->host_initiated)
3516 			return 1;
3517 		msr_info->data = 0;
3518 		break;
3519 	case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3520 	case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3521 	case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3522 	case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3523 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3524 			return kvm_pmu_get_msr(vcpu, msr_info);
3525 		msr_info->data = 0;
3526 		break;
3527 	case MSR_IA32_UCODE_REV:
3528 		msr_info->data = vcpu->arch.microcode_version;
3529 		break;
3530 	case MSR_IA32_ARCH_CAPABILITIES:
3531 		if (!msr_info->host_initiated &&
3532 		    !guest_cpuid_has(vcpu, X86_FEATURE_ARCH_CAPABILITIES))
3533 			return 1;
3534 		msr_info->data = vcpu->arch.arch_capabilities;
3535 		break;
3536 	case MSR_IA32_PERF_CAPABILITIES:
3537 		if (!msr_info->host_initiated &&
3538 		    !guest_cpuid_has(vcpu, X86_FEATURE_PDCM))
3539 			return 1;
3540 		msr_info->data = vcpu->arch.perf_capabilities;
3541 		break;
3542 	case MSR_IA32_POWER_CTL:
3543 		msr_info->data = vcpu->arch.msr_ia32_power_ctl;
3544 		break;
3545 	case MSR_IA32_TSC: {
3546 		/*
3547 		 * Intel SDM states that MSR_IA32_TSC read adds the TSC offset
3548 		 * even when not intercepted. AMD manual doesn't explicitly
3549 		 * state this but appears to behave the same.
3550 		 *
3551 		 * On userspace reads and writes, however, we unconditionally
3552 		 * return L1's TSC value to ensure backwards-compatible
3553 		 * behavior for migration.
3554 		 */
3555 		u64 tsc_offset = msr_info->host_initiated ? vcpu->arch.l1_tsc_offset :
3556 							    vcpu->arch.tsc_offset;
3557 
3558 		msr_info->data = kvm_scale_tsc(vcpu, rdtsc()) + tsc_offset;
3559 		break;
3560 	}
3561 	case MSR_MTRRcap:
3562 	case 0x200 ... 0x2ff:
3563 		return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
3564 	case 0xcd: /* fsb frequency */
3565 		msr_info->data = 3;
3566 		break;
3567 		/*
3568 		 * MSR_EBC_FREQUENCY_ID
3569 		 * Conservative value valid for even the basic CPU models.
3570 		 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
3571 		 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
3572 		 * and 266MHz for model 3, or 4. Set Core Clock
3573 		 * Frequency to System Bus Frequency Ratio to 1 (bits
3574 		 * 31:24) even though these are only valid for CPU
3575 		 * models > 2, however guests may end up dividing or
3576 		 * multiplying by zero otherwise.
3577 		 */
3578 	case MSR_EBC_FREQUENCY_ID:
3579 		msr_info->data = 1 << 24;
3580 		break;
3581 	case MSR_IA32_APICBASE:
3582 		msr_info->data = kvm_get_apic_base(vcpu);
3583 		break;
3584 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
3585 		return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
3586 	case MSR_IA32_TSC_DEADLINE:
3587 		msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
3588 		break;
3589 	case MSR_IA32_TSC_ADJUST:
3590 		msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
3591 		break;
3592 	case MSR_IA32_MISC_ENABLE:
3593 		msr_info->data = vcpu->arch.ia32_misc_enable_msr;
3594 		break;
3595 	case MSR_IA32_SMBASE:
3596 		if (!msr_info->host_initiated)
3597 			return 1;
3598 		msr_info->data = vcpu->arch.smbase;
3599 		break;
3600 	case MSR_SMI_COUNT:
3601 		msr_info->data = vcpu->arch.smi_count;
3602 		break;
3603 	case MSR_IA32_PERF_STATUS:
3604 		/* TSC increment by tick */
3605 		msr_info->data = 1000ULL;
3606 		/* CPU multiplier */
3607 		msr_info->data |= (((uint64_t)4ULL) << 40);
3608 		break;
3609 	case MSR_EFER:
3610 		msr_info->data = vcpu->arch.efer;
3611 		break;
3612 	case MSR_KVM_WALL_CLOCK:
3613 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3614 			return 1;
3615 
3616 		msr_info->data = vcpu->kvm->arch.wall_clock;
3617 		break;
3618 	case MSR_KVM_WALL_CLOCK_NEW:
3619 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3620 			return 1;
3621 
3622 		msr_info->data = vcpu->kvm->arch.wall_clock;
3623 		break;
3624 	case MSR_KVM_SYSTEM_TIME:
3625 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3626 			return 1;
3627 
3628 		msr_info->data = vcpu->arch.time;
3629 		break;
3630 	case MSR_KVM_SYSTEM_TIME_NEW:
3631 		if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3632 			return 1;
3633 
3634 		msr_info->data = vcpu->arch.time;
3635 		break;
3636 	case MSR_KVM_ASYNC_PF_EN:
3637 		if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3638 			return 1;
3639 
3640 		msr_info->data = vcpu->arch.apf.msr_en_val;
3641 		break;
3642 	case MSR_KVM_ASYNC_PF_INT:
3643 		if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3644 			return 1;
3645 
3646 		msr_info->data = vcpu->arch.apf.msr_int_val;
3647 		break;
3648 	case MSR_KVM_ASYNC_PF_ACK:
3649 		if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3650 			return 1;
3651 
3652 		msr_info->data = 0;
3653 		break;
3654 	case MSR_KVM_STEAL_TIME:
3655 		if (!guest_pv_has(vcpu, KVM_FEATURE_STEAL_TIME))
3656 			return 1;
3657 
3658 		msr_info->data = vcpu->arch.st.msr_val;
3659 		break;
3660 	case MSR_KVM_PV_EOI_EN:
3661 		if (!guest_pv_has(vcpu, KVM_FEATURE_PV_EOI))
3662 			return 1;
3663 
3664 		msr_info->data = vcpu->arch.pv_eoi.msr_val;
3665 		break;
3666 	case MSR_KVM_POLL_CONTROL:
3667 		if (!guest_pv_has(vcpu, KVM_FEATURE_POLL_CONTROL))
3668 			return 1;
3669 
3670 		msr_info->data = vcpu->arch.msr_kvm_poll_control;
3671 		break;
3672 	case MSR_IA32_P5_MC_ADDR:
3673 	case MSR_IA32_P5_MC_TYPE:
3674 	case MSR_IA32_MCG_CAP:
3675 	case MSR_IA32_MCG_CTL:
3676 	case MSR_IA32_MCG_STATUS:
3677 	case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3678 		return get_msr_mce(vcpu, msr_info->index, &msr_info->data,
3679 				   msr_info->host_initiated);
3680 	case MSR_IA32_XSS:
3681 		if (!msr_info->host_initiated &&
3682 		    !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3683 			return 1;
3684 		msr_info->data = vcpu->arch.ia32_xss;
3685 		break;
3686 	case MSR_K7_CLK_CTL:
3687 		/*
3688 		 * Provide expected ramp-up count for K7. All other
3689 		 * are set to zero, indicating minimum divisors for
3690 		 * every field.
3691 		 *
3692 		 * This prevents guest kernels on AMD host with CPU
3693 		 * type 6, model 8 and higher from exploding due to
3694 		 * the rdmsr failing.
3695 		 */
3696 		msr_info->data = 0x20000000;
3697 		break;
3698 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3699 	case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
3700 	case HV_X64_MSR_SYNDBG_OPTIONS:
3701 	case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3702 	case HV_X64_MSR_CRASH_CTL:
3703 	case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3704 	case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3705 	case HV_X64_MSR_TSC_EMULATION_CONTROL:
3706 	case HV_X64_MSR_TSC_EMULATION_STATUS:
3707 		return kvm_hv_get_msr_common(vcpu,
3708 					     msr_info->index, &msr_info->data,
3709 					     msr_info->host_initiated);
3710 	case MSR_IA32_BBL_CR_CTL3:
3711 		/* This legacy MSR exists but isn't fully documented in current
3712 		 * silicon.  It is however accessed by winxp in very narrow
3713 		 * scenarios where it sets bit #19, itself documented as
3714 		 * a "reserved" bit.  Best effort attempt to source coherent
3715 		 * read data here should the balance of the register be
3716 		 * interpreted by the guest:
3717 		 *
3718 		 * L2 cache control register 3: 64GB range, 256KB size,
3719 		 * enabled, latency 0x1, configured
3720 		 */
3721 		msr_info->data = 0xbe702111;
3722 		break;
3723 	case MSR_AMD64_OSVW_ID_LENGTH:
3724 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3725 			return 1;
3726 		msr_info->data = vcpu->arch.osvw.length;
3727 		break;
3728 	case MSR_AMD64_OSVW_STATUS:
3729 		if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3730 			return 1;
3731 		msr_info->data = vcpu->arch.osvw.status;
3732 		break;
3733 	case MSR_PLATFORM_INFO:
3734 		if (!msr_info->host_initiated &&
3735 		    !vcpu->kvm->arch.guest_can_read_msr_platform_info)
3736 			return 1;
3737 		msr_info->data = vcpu->arch.msr_platform_info;
3738 		break;
3739 	case MSR_MISC_FEATURES_ENABLES:
3740 		msr_info->data = vcpu->arch.msr_misc_features_enables;
3741 		break;
3742 	case MSR_K7_HWCR:
3743 		msr_info->data = vcpu->arch.msr_hwcr;
3744 		break;
3745 	default:
3746 		if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3747 			return kvm_pmu_get_msr(vcpu, msr_info);
3748 		return KVM_MSR_RET_INVALID;
3749 	}
3750 	return 0;
3751 }
3752 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
3753 
3754 /*
3755  * Read or write a bunch of msrs. All parameters are kernel addresses.
3756  *
3757  * @return number of msrs set successfully.
3758  */
3759 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
3760 		    struct kvm_msr_entry *entries,
3761 		    int (*do_msr)(struct kvm_vcpu *vcpu,
3762 				  unsigned index, u64 *data))
3763 {
3764 	int i;
3765 
3766 	for (i = 0; i < msrs->nmsrs; ++i)
3767 		if (do_msr(vcpu, entries[i].index, &entries[i].data))
3768 			break;
3769 
3770 	return i;
3771 }
3772 
3773 /*
3774  * Read or write a bunch of msrs. Parameters are user addresses.
3775  *
3776  * @return number of msrs set successfully.
3777  */
3778 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
3779 		  int (*do_msr)(struct kvm_vcpu *vcpu,
3780 				unsigned index, u64 *data),
3781 		  int writeback)
3782 {
3783 	struct kvm_msrs msrs;
3784 	struct kvm_msr_entry *entries;
3785 	int r, n;
3786 	unsigned size;
3787 
3788 	r = -EFAULT;
3789 	if (copy_from_user(&msrs, user_msrs, sizeof(msrs)))
3790 		goto out;
3791 
3792 	r = -E2BIG;
3793 	if (msrs.nmsrs >= MAX_IO_MSRS)
3794 		goto out;
3795 
3796 	size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
3797 	entries = memdup_user(user_msrs->entries, size);
3798 	if (IS_ERR(entries)) {
3799 		r = PTR_ERR(entries);
3800 		goto out;
3801 	}
3802 
3803 	r = n = __msr_io(vcpu, &msrs, entries, do_msr);
3804 	if (r < 0)
3805 		goto out_free;
3806 
3807 	r = -EFAULT;
3808 	if (writeback && copy_to_user(user_msrs->entries, entries, size))
3809 		goto out_free;
3810 
3811 	r = n;
3812 
3813 out_free:
3814 	kfree(entries);
3815 out:
3816 	return r;
3817 }
3818 
3819 static inline bool kvm_can_mwait_in_guest(void)
3820 {
3821 	return boot_cpu_has(X86_FEATURE_MWAIT) &&
3822 		!boot_cpu_has_bug(X86_BUG_MONITOR) &&
3823 		boot_cpu_has(X86_FEATURE_ARAT);
3824 }
3825 
3826 static int kvm_ioctl_get_supported_hv_cpuid(struct kvm_vcpu *vcpu,
3827 					    struct kvm_cpuid2 __user *cpuid_arg)
3828 {
3829 	struct kvm_cpuid2 cpuid;
3830 	int r;
3831 
3832 	r = -EFAULT;
3833 	if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3834 		return r;
3835 
3836 	r = kvm_get_hv_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3837 	if (r)
3838 		return r;
3839 
3840 	r = -EFAULT;
3841 	if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
3842 		return r;
3843 
3844 	return 0;
3845 }
3846 
3847 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
3848 {
3849 	int r = 0;
3850 
3851 	switch (ext) {
3852 	case KVM_CAP_IRQCHIP:
3853 	case KVM_CAP_HLT:
3854 	case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
3855 	case KVM_CAP_SET_TSS_ADDR:
3856 	case KVM_CAP_EXT_CPUID:
3857 	case KVM_CAP_EXT_EMUL_CPUID:
3858 	case KVM_CAP_CLOCKSOURCE:
3859 	case KVM_CAP_PIT:
3860 	case KVM_CAP_NOP_IO_DELAY:
3861 	case KVM_CAP_MP_STATE:
3862 	case KVM_CAP_SYNC_MMU:
3863 	case KVM_CAP_USER_NMI:
3864 	case KVM_CAP_REINJECT_CONTROL:
3865 	case KVM_CAP_IRQ_INJECT_STATUS:
3866 	case KVM_CAP_IOEVENTFD:
3867 	case KVM_CAP_IOEVENTFD_NO_LENGTH:
3868 	case KVM_CAP_PIT2:
3869 	case KVM_CAP_PIT_STATE2:
3870 	case KVM_CAP_SET_IDENTITY_MAP_ADDR:
3871 	case KVM_CAP_VCPU_EVENTS:
3872 	case KVM_CAP_HYPERV:
3873 	case KVM_CAP_HYPERV_VAPIC:
3874 	case KVM_CAP_HYPERV_SPIN:
3875 	case KVM_CAP_HYPERV_SYNIC:
3876 	case KVM_CAP_HYPERV_SYNIC2:
3877 	case KVM_CAP_HYPERV_VP_INDEX:
3878 	case KVM_CAP_HYPERV_EVENTFD:
3879 	case KVM_CAP_HYPERV_TLBFLUSH:
3880 	case KVM_CAP_HYPERV_SEND_IPI:
3881 	case KVM_CAP_HYPERV_CPUID:
3882 	case KVM_CAP_SYS_HYPERV_CPUID:
3883 	case KVM_CAP_PCI_SEGMENT:
3884 	case KVM_CAP_DEBUGREGS:
3885 	case KVM_CAP_X86_ROBUST_SINGLESTEP:
3886 	case KVM_CAP_XSAVE:
3887 	case KVM_CAP_ASYNC_PF:
3888 	case KVM_CAP_ASYNC_PF_INT:
3889 	case KVM_CAP_GET_TSC_KHZ:
3890 	case KVM_CAP_KVMCLOCK_CTRL:
3891 	case KVM_CAP_READONLY_MEM:
3892 	case KVM_CAP_HYPERV_TIME:
3893 	case KVM_CAP_IOAPIC_POLARITY_IGNORED:
3894 	case KVM_CAP_TSC_DEADLINE_TIMER:
3895 	case KVM_CAP_DISABLE_QUIRKS:
3896 	case KVM_CAP_SET_BOOT_CPU_ID:
3897  	case KVM_CAP_SPLIT_IRQCHIP:
3898 	case KVM_CAP_IMMEDIATE_EXIT:
3899 	case KVM_CAP_PMU_EVENT_FILTER:
3900 	case KVM_CAP_GET_MSR_FEATURES:
3901 	case KVM_CAP_MSR_PLATFORM_INFO:
3902 	case KVM_CAP_EXCEPTION_PAYLOAD:
3903 	case KVM_CAP_SET_GUEST_DEBUG:
3904 	case KVM_CAP_LAST_CPU:
3905 	case KVM_CAP_X86_USER_SPACE_MSR:
3906 	case KVM_CAP_X86_MSR_FILTER:
3907 	case KVM_CAP_ENFORCE_PV_FEATURE_CPUID:
3908 #ifdef CONFIG_X86_SGX_KVM
3909 	case KVM_CAP_SGX_ATTRIBUTE:
3910 #endif
3911 	case KVM_CAP_VM_COPY_ENC_CONTEXT_FROM:
3912 		r = 1;
3913 		break;
3914 	case KVM_CAP_SET_GUEST_DEBUG2:
3915 		return KVM_GUESTDBG_VALID_MASK;
3916 #ifdef CONFIG_KVM_XEN
3917 	case KVM_CAP_XEN_HVM:
3918 		r = KVM_XEN_HVM_CONFIG_HYPERCALL_MSR |
3919 		    KVM_XEN_HVM_CONFIG_INTERCEPT_HCALL |
3920 		    KVM_XEN_HVM_CONFIG_SHARED_INFO;
3921 		if (sched_info_on())
3922 			r |= KVM_XEN_HVM_CONFIG_RUNSTATE;
3923 		break;
3924 #endif
3925 	case KVM_CAP_SYNC_REGS:
3926 		r = KVM_SYNC_X86_VALID_FIELDS;
3927 		break;
3928 	case KVM_CAP_ADJUST_CLOCK:
3929 		r = KVM_CLOCK_TSC_STABLE;
3930 		break;
3931 	case KVM_CAP_X86_DISABLE_EXITS:
3932 		r |=  KVM_X86_DISABLE_EXITS_HLT | KVM_X86_DISABLE_EXITS_PAUSE |
3933 		      KVM_X86_DISABLE_EXITS_CSTATE;
3934 		if(kvm_can_mwait_in_guest())
3935 			r |= KVM_X86_DISABLE_EXITS_MWAIT;
3936 		break;
3937 	case KVM_CAP_X86_SMM:
3938 		/* SMBASE is usually relocated above 1M on modern chipsets,
3939 		 * and SMM handlers might indeed rely on 4G segment limits,
3940 		 * so do not report SMM to be available if real mode is
3941 		 * emulated via vm86 mode.  Still, do not go to great lengths
3942 		 * to avoid userspace's usage of the feature, because it is a
3943 		 * fringe case that is not enabled except via specific settings
3944 		 * of the module parameters.
3945 		 */
3946 		r = static_call(kvm_x86_has_emulated_msr)(kvm, MSR_IA32_SMBASE);
3947 		break;
3948 	case KVM_CAP_VAPIC:
3949 		r = !static_call(kvm_x86_cpu_has_accelerated_tpr)();
3950 		break;
3951 	case KVM_CAP_NR_VCPUS:
3952 		r = KVM_SOFT_MAX_VCPUS;
3953 		break;
3954 	case KVM_CAP_MAX_VCPUS:
3955 		r = KVM_MAX_VCPUS;
3956 		break;
3957 	case KVM_CAP_MAX_VCPU_ID:
3958 		r = KVM_MAX_VCPU_ID;
3959 		break;
3960 	case KVM_CAP_PV_MMU:	/* obsolete */
3961 		r = 0;
3962 		break;
3963 	case KVM_CAP_MCE:
3964 		r = KVM_MAX_MCE_BANKS;
3965 		break;
3966 	case KVM_CAP_XCRS:
3967 		r = boot_cpu_has(X86_FEATURE_XSAVE);
3968 		break;
3969 	case KVM_CAP_TSC_CONTROL:
3970 		r = kvm_has_tsc_control;
3971 		break;
3972 	case KVM_CAP_X2APIC_API:
3973 		r = KVM_X2APIC_API_VALID_FLAGS;
3974 		break;
3975 	case KVM_CAP_NESTED_STATE:
3976 		r = kvm_x86_ops.nested_ops->get_state ?
3977 			kvm_x86_ops.nested_ops->get_state(NULL, NULL, 0) : 0;
3978 		break;
3979 	case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
3980 		r = kvm_x86_ops.enable_direct_tlbflush != NULL;
3981 		break;
3982 	case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
3983 		r = kvm_x86_ops.nested_ops->enable_evmcs != NULL;
3984 		break;
3985 	case KVM_CAP_SMALLER_MAXPHYADDR:
3986 		r = (int) allow_smaller_maxphyaddr;
3987 		break;
3988 	case KVM_CAP_STEAL_TIME:
3989 		r = sched_info_on();
3990 		break;
3991 	case KVM_CAP_X86_BUS_LOCK_EXIT:
3992 		if (kvm_has_bus_lock_exit)
3993 			r = KVM_BUS_LOCK_DETECTION_OFF |
3994 			    KVM_BUS_LOCK_DETECTION_EXIT;
3995 		else
3996 			r = 0;
3997 		break;
3998 	default:
3999 		break;
4000 	}
4001 	return r;
4002 
4003 }
4004 
4005 long kvm_arch_dev_ioctl(struct file *filp,
4006 			unsigned int ioctl, unsigned long arg)
4007 {
4008 	void __user *argp = (void __user *)arg;
4009 	long r;
4010 
4011 	switch (ioctl) {
4012 	case KVM_GET_MSR_INDEX_LIST: {
4013 		struct kvm_msr_list __user *user_msr_list = argp;
4014 		struct kvm_msr_list msr_list;
4015 		unsigned n;
4016 
4017 		r = -EFAULT;
4018 		if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
4019 			goto out;
4020 		n = msr_list.nmsrs;
4021 		msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
4022 		if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
4023 			goto out;
4024 		r = -E2BIG;
4025 		if (n < msr_list.nmsrs)
4026 			goto out;
4027 		r = -EFAULT;
4028 		if (copy_to_user(user_msr_list->indices, &msrs_to_save,
4029 				 num_msrs_to_save * sizeof(u32)))
4030 			goto out;
4031 		if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
4032 				 &emulated_msrs,
4033 				 num_emulated_msrs * sizeof(u32)))
4034 			goto out;
4035 		r = 0;
4036 		break;
4037 	}
4038 	case KVM_GET_SUPPORTED_CPUID:
4039 	case KVM_GET_EMULATED_CPUID: {
4040 		struct kvm_cpuid2 __user *cpuid_arg = argp;
4041 		struct kvm_cpuid2 cpuid;
4042 
4043 		r = -EFAULT;
4044 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4045 			goto out;
4046 
4047 		r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
4048 					    ioctl);
4049 		if (r)
4050 			goto out;
4051 
4052 		r = -EFAULT;
4053 		if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4054 			goto out;
4055 		r = 0;
4056 		break;
4057 	}
4058 	case KVM_X86_GET_MCE_CAP_SUPPORTED:
4059 		r = -EFAULT;
4060 		if (copy_to_user(argp, &kvm_mce_cap_supported,
4061 				 sizeof(kvm_mce_cap_supported)))
4062 			goto out;
4063 		r = 0;
4064 		break;
4065 	case KVM_GET_MSR_FEATURE_INDEX_LIST: {
4066 		struct kvm_msr_list __user *user_msr_list = argp;
4067 		struct kvm_msr_list msr_list;
4068 		unsigned int n;
4069 
4070 		r = -EFAULT;
4071 		if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
4072 			goto out;
4073 		n = msr_list.nmsrs;
4074 		msr_list.nmsrs = num_msr_based_features;
4075 		if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
4076 			goto out;
4077 		r = -E2BIG;
4078 		if (n < msr_list.nmsrs)
4079 			goto out;
4080 		r = -EFAULT;
4081 		if (copy_to_user(user_msr_list->indices, &msr_based_features,
4082 				 num_msr_based_features * sizeof(u32)))
4083 			goto out;
4084 		r = 0;
4085 		break;
4086 	}
4087 	case KVM_GET_MSRS:
4088 		r = msr_io(NULL, argp, do_get_msr_feature, 1);
4089 		break;
4090 	case KVM_GET_SUPPORTED_HV_CPUID:
4091 		r = kvm_ioctl_get_supported_hv_cpuid(NULL, argp);
4092 		break;
4093 	default:
4094 		r = -EINVAL;
4095 		break;
4096 	}
4097 out:
4098 	return r;
4099 }
4100 
4101 static void wbinvd_ipi(void *garbage)
4102 {
4103 	wbinvd();
4104 }
4105 
4106 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
4107 {
4108 	return kvm_arch_has_noncoherent_dma(vcpu->kvm);
4109 }
4110 
4111 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
4112 {
4113 	/* Address WBINVD may be executed by guest */
4114 	if (need_emulate_wbinvd(vcpu)) {
4115 		if (static_call(kvm_x86_has_wbinvd_exit)())
4116 			cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4117 		else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
4118 			smp_call_function_single(vcpu->cpu,
4119 					wbinvd_ipi, NULL, 1);
4120 	}
4121 
4122 	static_call(kvm_x86_vcpu_load)(vcpu, cpu);
4123 
4124 	/* Save host pkru register if supported */
4125 	vcpu->arch.host_pkru = read_pkru();
4126 
4127 	/* Apply any externally detected TSC adjustments (due to suspend) */
4128 	if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
4129 		adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
4130 		vcpu->arch.tsc_offset_adjustment = 0;
4131 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4132 	}
4133 
4134 	if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) {
4135 		s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
4136 				rdtsc() - vcpu->arch.last_host_tsc;
4137 		if (tsc_delta < 0)
4138 			mark_tsc_unstable("KVM discovered backwards TSC");
4139 
4140 		if (kvm_check_tsc_unstable()) {
4141 			u64 offset = kvm_compute_tsc_offset(vcpu,
4142 						vcpu->arch.last_guest_tsc);
4143 			kvm_vcpu_write_tsc_offset(vcpu, offset);
4144 			vcpu->arch.tsc_catchup = 1;
4145 		}
4146 
4147 		if (kvm_lapic_hv_timer_in_use(vcpu))
4148 			kvm_lapic_restart_hv_timer(vcpu);
4149 
4150 		/*
4151 		 * On a host with synchronized TSC, there is no need to update
4152 		 * kvmclock on vcpu->cpu migration
4153 		 */
4154 		if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
4155 			kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
4156 		if (vcpu->cpu != cpu)
4157 			kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
4158 		vcpu->cpu = cpu;
4159 	}
4160 
4161 	kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
4162 }
4163 
4164 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
4165 {
4166 	struct kvm_host_map map;
4167 	struct kvm_steal_time *st;
4168 
4169 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
4170 		return;
4171 
4172 	if (vcpu->arch.st.preempted)
4173 		return;
4174 
4175 	if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT, &map,
4176 			&vcpu->arch.st.cache, true))
4177 		return;
4178 
4179 	st = map.hva +
4180 		offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS);
4181 
4182 	st->preempted = vcpu->arch.st.preempted = KVM_VCPU_PREEMPTED;
4183 
4184 	kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, true);
4185 }
4186 
4187 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
4188 {
4189 	int idx;
4190 
4191 	if (vcpu->preempted && !vcpu->arch.guest_state_protected)
4192 		vcpu->arch.preempted_in_kernel = !static_call(kvm_x86_get_cpl)(vcpu);
4193 
4194 	/*
4195 	 * Take the srcu lock as memslots will be accessed to check the gfn
4196 	 * cache generation against the memslots generation.
4197 	 */
4198 	idx = srcu_read_lock(&vcpu->kvm->srcu);
4199 	if (kvm_xen_msr_enabled(vcpu->kvm))
4200 		kvm_xen_runstate_set_preempted(vcpu);
4201 	else
4202 		kvm_steal_time_set_preempted(vcpu);
4203 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
4204 
4205 	static_call(kvm_x86_vcpu_put)(vcpu);
4206 	vcpu->arch.last_host_tsc = rdtsc();
4207 	/*
4208 	 * If userspace has set any breakpoints or watchpoints, dr6 is restored
4209 	 * on every vmexit, but if not, we might have a stale dr6 from the
4210 	 * guest. do_debug expects dr6 to be cleared after it runs, do the same.
4211 	 */
4212 	set_debugreg(0, 6);
4213 }
4214 
4215 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
4216 				    struct kvm_lapic_state *s)
4217 {
4218 	if (vcpu->arch.apicv_active)
4219 		static_call(kvm_x86_sync_pir_to_irr)(vcpu);
4220 
4221 	return kvm_apic_get_state(vcpu, s);
4222 }
4223 
4224 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
4225 				    struct kvm_lapic_state *s)
4226 {
4227 	int r;
4228 
4229 	r = kvm_apic_set_state(vcpu, s);
4230 	if (r)
4231 		return r;
4232 	update_cr8_intercept(vcpu);
4233 
4234 	return 0;
4235 }
4236 
4237 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
4238 {
4239 	/*
4240 	 * We can accept userspace's request for interrupt injection
4241 	 * as long as we have a place to store the interrupt number.
4242 	 * The actual injection will happen when the CPU is able to
4243 	 * deliver the interrupt.
4244 	 */
4245 	if (kvm_cpu_has_extint(vcpu))
4246 		return false;
4247 
4248 	/* Acknowledging ExtINT does not happen if LINT0 is masked.  */
4249 	return (!lapic_in_kernel(vcpu) ||
4250 		kvm_apic_accept_pic_intr(vcpu));
4251 }
4252 
4253 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
4254 {
4255 	return kvm_arch_interrupt_allowed(vcpu) &&
4256 		kvm_cpu_accept_dm_intr(vcpu);
4257 }
4258 
4259 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
4260 				    struct kvm_interrupt *irq)
4261 {
4262 	if (irq->irq >= KVM_NR_INTERRUPTS)
4263 		return -EINVAL;
4264 
4265 	if (!irqchip_in_kernel(vcpu->kvm)) {
4266 		kvm_queue_interrupt(vcpu, irq->irq, false);
4267 		kvm_make_request(KVM_REQ_EVENT, vcpu);
4268 		return 0;
4269 	}
4270 
4271 	/*
4272 	 * With in-kernel LAPIC, we only use this to inject EXTINT, so
4273 	 * fail for in-kernel 8259.
4274 	 */
4275 	if (pic_in_kernel(vcpu->kvm))
4276 		return -ENXIO;
4277 
4278 	if (vcpu->arch.pending_external_vector != -1)
4279 		return -EEXIST;
4280 
4281 	vcpu->arch.pending_external_vector = irq->irq;
4282 	kvm_make_request(KVM_REQ_EVENT, vcpu);
4283 	return 0;
4284 }
4285 
4286 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
4287 {
4288 	kvm_inject_nmi(vcpu);
4289 
4290 	return 0;
4291 }
4292 
4293 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
4294 {
4295 	kvm_make_request(KVM_REQ_SMI, vcpu);
4296 
4297 	return 0;
4298 }
4299 
4300 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
4301 					   struct kvm_tpr_access_ctl *tac)
4302 {
4303 	if (tac->flags)
4304 		return -EINVAL;
4305 	vcpu->arch.tpr_access_reporting = !!tac->enabled;
4306 	return 0;
4307 }
4308 
4309 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
4310 					u64 mcg_cap)
4311 {
4312 	int r;
4313 	unsigned bank_num = mcg_cap & 0xff, bank;
4314 
4315 	r = -EINVAL;
4316 	if (!bank_num || bank_num > KVM_MAX_MCE_BANKS)
4317 		goto out;
4318 	if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
4319 		goto out;
4320 	r = 0;
4321 	vcpu->arch.mcg_cap = mcg_cap;
4322 	/* Init IA32_MCG_CTL to all 1s */
4323 	if (mcg_cap & MCG_CTL_P)
4324 		vcpu->arch.mcg_ctl = ~(u64)0;
4325 	/* Init IA32_MCi_CTL to all 1s */
4326 	for (bank = 0; bank < bank_num; bank++)
4327 		vcpu->arch.mce_banks[bank*4] = ~(u64)0;
4328 
4329 	static_call(kvm_x86_setup_mce)(vcpu);
4330 out:
4331 	return r;
4332 }
4333 
4334 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
4335 				      struct kvm_x86_mce *mce)
4336 {
4337 	u64 mcg_cap = vcpu->arch.mcg_cap;
4338 	unsigned bank_num = mcg_cap & 0xff;
4339 	u64 *banks = vcpu->arch.mce_banks;
4340 
4341 	if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
4342 		return -EINVAL;
4343 	/*
4344 	 * if IA32_MCG_CTL is not all 1s, the uncorrected error
4345 	 * reporting is disabled
4346 	 */
4347 	if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
4348 	    vcpu->arch.mcg_ctl != ~(u64)0)
4349 		return 0;
4350 	banks += 4 * mce->bank;
4351 	/*
4352 	 * if IA32_MCi_CTL is not all 1s, the uncorrected error
4353 	 * reporting is disabled for the bank
4354 	 */
4355 	if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
4356 		return 0;
4357 	if (mce->status & MCI_STATUS_UC) {
4358 		if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
4359 		    !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
4360 			kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4361 			return 0;
4362 		}
4363 		if (banks[1] & MCI_STATUS_VAL)
4364 			mce->status |= MCI_STATUS_OVER;
4365 		banks[2] = mce->addr;
4366 		banks[3] = mce->misc;
4367 		vcpu->arch.mcg_status = mce->mcg_status;
4368 		banks[1] = mce->status;
4369 		kvm_queue_exception(vcpu, MC_VECTOR);
4370 	} else if (!(banks[1] & MCI_STATUS_VAL)
4371 		   || !(banks[1] & MCI_STATUS_UC)) {
4372 		if (banks[1] & MCI_STATUS_VAL)
4373 			mce->status |= MCI_STATUS_OVER;
4374 		banks[2] = mce->addr;
4375 		banks[3] = mce->misc;
4376 		banks[1] = mce->status;
4377 	} else
4378 		banks[1] |= MCI_STATUS_OVER;
4379 	return 0;
4380 }
4381 
4382 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
4383 					       struct kvm_vcpu_events *events)
4384 {
4385 	process_nmi(vcpu);
4386 
4387 	if (kvm_check_request(KVM_REQ_SMI, vcpu))
4388 		process_smi(vcpu);
4389 
4390 	/*
4391 	 * In guest mode, payload delivery should be deferred,
4392 	 * so that the L1 hypervisor can intercept #PF before
4393 	 * CR2 is modified (or intercept #DB before DR6 is
4394 	 * modified under nVMX). Unless the per-VM capability,
4395 	 * KVM_CAP_EXCEPTION_PAYLOAD, is set, we may not defer the delivery of
4396 	 * an exception payload and handle after a KVM_GET_VCPU_EVENTS. Since we
4397 	 * opportunistically defer the exception payload, deliver it if the
4398 	 * capability hasn't been requested before processing a
4399 	 * KVM_GET_VCPU_EVENTS.
4400 	 */
4401 	if (!vcpu->kvm->arch.exception_payload_enabled &&
4402 	    vcpu->arch.exception.pending && vcpu->arch.exception.has_payload)
4403 		kvm_deliver_exception_payload(vcpu);
4404 
4405 	/*
4406 	 * The API doesn't provide the instruction length for software
4407 	 * exceptions, so don't report them. As long as the guest RIP
4408 	 * isn't advanced, we should expect to encounter the exception
4409 	 * again.
4410 	 */
4411 	if (kvm_exception_is_soft(vcpu->arch.exception.nr)) {
4412 		events->exception.injected = 0;
4413 		events->exception.pending = 0;
4414 	} else {
4415 		events->exception.injected = vcpu->arch.exception.injected;
4416 		events->exception.pending = vcpu->arch.exception.pending;
4417 		/*
4418 		 * For ABI compatibility, deliberately conflate
4419 		 * pending and injected exceptions when
4420 		 * KVM_CAP_EXCEPTION_PAYLOAD isn't enabled.
4421 		 */
4422 		if (!vcpu->kvm->arch.exception_payload_enabled)
4423 			events->exception.injected |=
4424 				vcpu->arch.exception.pending;
4425 	}
4426 	events->exception.nr = vcpu->arch.exception.nr;
4427 	events->exception.has_error_code = vcpu->arch.exception.has_error_code;
4428 	events->exception.error_code = vcpu->arch.exception.error_code;
4429 	events->exception_has_payload = vcpu->arch.exception.has_payload;
4430 	events->exception_payload = vcpu->arch.exception.payload;
4431 
4432 	events->interrupt.injected =
4433 		vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft;
4434 	events->interrupt.nr = vcpu->arch.interrupt.nr;
4435 	events->interrupt.soft = 0;
4436 	events->interrupt.shadow = static_call(kvm_x86_get_interrupt_shadow)(vcpu);
4437 
4438 	events->nmi.injected = vcpu->arch.nmi_injected;
4439 	events->nmi.pending = vcpu->arch.nmi_pending != 0;
4440 	events->nmi.masked = static_call(kvm_x86_get_nmi_mask)(vcpu);
4441 	events->nmi.pad = 0;
4442 
4443 	events->sipi_vector = 0; /* never valid when reporting to user space */
4444 
4445 	events->smi.smm = is_smm(vcpu);
4446 	events->smi.pending = vcpu->arch.smi_pending;
4447 	events->smi.smm_inside_nmi =
4448 		!!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
4449 	events->smi.latched_init = kvm_lapic_latched_init(vcpu);
4450 
4451 	events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
4452 			 | KVM_VCPUEVENT_VALID_SHADOW
4453 			 | KVM_VCPUEVENT_VALID_SMM);
4454 	if (vcpu->kvm->arch.exception_payload_enabled)
4455 		events->flags |= KVM_VCPUEVENT_VALID_PAYLOAD;
4456 
4457 	memset(&events->reserved, 0, sizeof(events->reserved));
4458 }
4459 
4460 static void kvm_smm_changed(struct kvm_vcpu *vcpu);
4461 
4462 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
4463 					      struct kvm_vcpu_events *events)
4464 {
4465 	if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
4466 			      | KVM_VCPUEVENT_VALID_SIPI_VECTOR
4467 			      | KVM_VCPUEVENT_VALID_SHADOW
4468 			      | KVM_VCPUEVENT_VALID_SMM
4469 			      | KVM_VCPUEVENT_VALID_PAYLOAD))
4470 		return -EINVAL;
4471 
4472 	if (events->flags & KVM_VCPUEVENT_VALID_PAYLOAD) {
4473 		if (!vcpu->kvm->arch.exception_payload_enabled)
4474 			return -EINVAL;
4475 		if (events->exception.pending)
4476 			events->exception.injected = 0;
4477 		else
4478 			events->exception_has_payload = 0;
4479 	} else {
4480 		events->exception.pending = 0;
4481 		events->exception_has_payload = 0;
4482 	}
4483 
4484 	if ((events->exception.injected || events->exception.pending) &&
4485 	    (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
4486 		return -EINVAL;
4487 
4488 	/* INITs are latched while in SMM */
4489 	if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
4490 	    (events->smi.smm || events->smi.pending) &&
4491 	    vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
4492 		return -EINVAL;
4493 
4494 	process_nmi(vcpu);
4495 	vcpu->arch.exception.injected = events->exception.injected;
4496 	vcpu->arch.exception.pending = events->exception.pending;
4497 	vcpu->arch.exception.nr = events->exception.nr;
4498 	vcpu->arch.exception.has_error_code = events->exception.has_error_code;
4499 	vcpu->arch.exception.error_code = events->exception.error_code;
4500 	vcpu->arch.exception.has_payload = events->exception_has_payload;
4501 	vcpu->arch.exception.payload = events->exception_payload;
4502 
4503 	vcpu->arch.interrupt.injected = events->interrupt.injected;
4504 	vcpu->arch.interrupt.nr = events->interrupt.nr;
4505 	vcpu->arch.interrupt.soft = events->interrupt.soft;
4506 	if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
4507 		static_call(kvm_x86_set_interrupt_shadow)(vcpu,
4508 						events->interrupt.shadow);
4509 
4510 	vcpu->arch.nmi_injected = events->nmi.injected;
4511 	if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
4512 		vcpu->arch.nmi_pending = events->nmi.pending;
4513 	static_call(kvm_x86_set_nmi_mask)(vcpu, events->nmi.masked);
4514 
4515 	if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
4516 	    lapic_in_kernel(vcpu))
4517 		vcpu->arch.apic->sipi_vector = events->sipi_vector;
4518 
4519 	if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
4520 		if (!!(vcpu->arch.hflags & HF_SMM_MASK) != events->smi.smm) {
4521 			if (events->smi.smm)
4522 				vcpu->arch.hflags |= HF_SMM_MASK;
4523 			else
4524 				vcpu->arch.hflags &= ~HF_SMM_MASK;
4525 			kvm_smm_changed(vcpu);
4526 		}
4527 
4528 		vcpu->arch.smi_pending = events->smi.pending;
4529 
4530 		if (events->smi.smm) {
4531 			if (events->smi.smm_inside_nmi)
4532 				vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
4533 			else
4534 				vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
4535 		}
4536 
4537 		if (lapic_in_kernel(vcpu)) {
4538 			if (events->smi.latched_init)
4539 				set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
4540 			else
4541 				clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
4542 		}
4543 	}
4544 
4545 	kvm_make_request(KVM_REQ_EVENT, vcpu);
4546 
4547 	return 0;
4548 }
4549 
4550 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
4551 					     struct kvm_debugregs *dbgregs)
4552 {
4553 	unsigned long val;
4554 
4555 	memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
4556 	kvm_get_dr(vcpu, 6, &val);
4557 	dbgregs->dr6 = val;
4558 	dbgregs->dr7 = vcpu->arch.dr7;
4559 	dbgregs->flags = 0;
4560 	memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
4561 }
4562 
4563 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
4564 					    struct kvm_debugregs *dbgregs)
4565 {
4566 	if (dbgregs->flags)
4567 		return -EINVAL;
4568 
4569 	if (!kvm_dr6_valid(dbgregs->dr6))
4570 		return -EINVAL;
4571 	if (!kvm_dr7_valid(dbgregs->dr7))
4572 		return -EINVAL;
4573 
4574 	memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
4575 	kvm_update_dr0123(vcpu);
4576 	vcpu->arch.dr6 = dbgregs->dr6;
4577 	vcpu->arch.dr7 = dbgregs->dr7;
4578 	kvm_update_dr7(vcpu);
4579 
4580 	return 0;
4581 }
4582 
4583 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
4584 
4585 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
4586 {
4587 	struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
4588 	u64 xstate_bv = xsave->header.xfeatures;
4589 	u64 valid;
4590 
4591 	/*
4592 	 * Copy legacy XSAVE area, to avoid complications with CPUID
4593 	 * leaves 0 and 1 in the loop below.
4594 	 */
4595 	memcpy(dest, xsave, XSAVE_HDR_OFFSET);
4596 
4597 	/* Set XSTATE_BV */
4598 	xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
4599 	*(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
4600 
4601 	/*
4602 	 * Copy each region from the possibly compacted offset to the
4603 	 * non-compacted offset.
4604 	 */
4605 	valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
4606 	while (valid) {
4607 		u64 xfeature_mask = valid & -valid;
4608 		int xfeature_nr = fls64(xfeature_mask) - 1;
4609 		void *src = get_xsave_addr(xsave, xfeature_nr);
4610 
4611 		if (src) {
4612 			u32 size, offset, ecx, edx;
4613 			cpuid_count(XSTATE_CPUID, xfeature_nr,
4614 				    &size, &offset, &ecx, &edx);
4615 			if (xfeature_nr == XFEATURE_PKRU)
4616 				memcpy(dest + offset, &vcpu->arch.pkru,
4617 				       sizeof(vcpu->arch.pkru));
4618 			else
4619 				memcpy(dest + offset, src, size);
4620 
4621 		}
4622 
4623 		valid -= xfeature_mask;
4624 	}
4625 }
4626 
4627 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
4628 {
4629 	struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
4630 	u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
4631 	u64 valid;
4632 
4633 	/*
4634 	 * Copy legacy XSAVE area, to avoid complications with CPUID
4635 	 * leaves 0 and 1 in the loop below.
4636 	 */
4637 	memcpy(xsave, src, XSAVE_HDR_OFFSET);
4638 
4639 	/* Set XSTATE_BV and possibly XCOMP_BV.  */
4640 	xsave->header.xfeatures = xstate_bv;
4641 	if (boot_cpu_has(X86_FEATURE_XSAVES))
4642 		xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
4643 
4644 	/*
4645 	 * Copy each region from the non-compacted offset to the
4646 	 * possibly compacted offset.
4647 	 */
4648 	valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
4649 	while (valid) {
4650 		u64 xfeature_mask = valid & -valid;
4651 		int xfeature_nr = fls64(xfeature_mask) - 1;
4652 		void *dest = get_xsave_addr(xsave, xfeature_nr);
4653 
4654 		if (dest) {
4655 			u32 size, offset, ecx, edx;
4656 			cpuid_count(XSTATE_CPUID, xfeature_nr,
4657 				    &size, &offset, &ecx, &edx);
4658 			if (xfeature_nr == XFEATURE_PKRU)
4659 				memcpy(&vcpu->arch.pkru, src + offset,
4660 				       sizeof(vcpu->arch.pkru));
4661 			else
4662 				memcpy(dest, src + offset, size);
4663 		}
4664 
4665 		valid -= xfeature_mask;
4666 	}
4667 }
4668 
4669 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
4670 					 struct kvm_xsave *guest_xsave)
4671 {
4672 	if (!vcpu->arch.guest_fpu)
4673 		return;
4674 
4675 	if (boot_cpu_has(X86_FEATURE_XSAVE)) {
4676 		memset(guest_xsave, 0, sizeof(struct kvm_xsave));
4677 		fill_xsave((u8 *) guest_xsave->region, vcpu);
4678 	} else {
4679 		memcpy(guest_xsave->region,
4680 			&vcpu->arch.guest_fpu->state.fxsave,
4681 			sizeof(struct fxregs_state));
4682 		*(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
4683 			XFEATURE_MASK_FPSSE;
4684 	}
4685 }
4686 
4687 #define XSAVE_MXCSR_OFFSET 24
4688 
4689 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
4690 					struct kvm_xsave *guest_xsave)
4691 {
4692 	u64 xstate_bv;
4693 	u32 mxcsr;
4694 
4695 	if (!vcpu->arch.guest_fpu)
4696 		return 0;
4697 
4698 	xstate_bv = *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
4699 	mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)];
4700 
4701 	if (boot_cpu_has(X86_FEATURE_XSAVE)) {
4702 		/*
4703 		 * Here we allow setting states that are not present in
4704 		 * CPUID leaf 0xD, index 0, EDX:EAX.  This is for compatibility
4705 		 * with old userspace.
4706 		 */
4707 		if (xstate_bv & ~supported_xcr0 || mxcsr & ~mxcsr_feature_mask)
4708 			return -EINVAL;
4709 		load_xsave(vcpu, (u8 *)guest_xsave->region);
4710 	} else {
4711 		if (xstate_bv & ~XFEATURE_MASK_FPSSE ||
4712 			mxcsr & ~mxcsr_feature_mask)
4713 			return -EINVAL;
4714 		memcpy(&vcpu->arch.guest_fpu->state.fxsave,
4715 			guest_xsave->region, sizeof(struct fxregs_state));
4716 	}
4717 	return 0;
4718 }
4719 
4720 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
4721 					struct kvm_xcrs *guest_xcrs)
4722 {
4723 	if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
4724 		guest_xcrs->nr_xcrs = 0;
4725 		return;
4726 	}
4727 
4728 	guest_xcrs->nr_xcrs = 1;
4729 	guest_xcrs->flags = 0;
4730 	guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
4731 	guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
4732 }
4733 
4734 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
4735 				       struct kvm_xcrs *guest_xcrs)
4736 {
4737 	int i, r = 0;
4738 
4739 	if (!boot_cpu_has(X86_FEATURE_XSAVE))
4740 		return -EINVAL;
4741 
4742 	if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
4743 		return -EINVAL;
4744 
4745 	for (i = 0; i < guest_xcrs->nr_xcrs; i++)
4746 		/* Only support XCR0 currently */
4747 		if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
4748 			r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
4749 				guest_xcrs->xcrs[i].value);
4750 			break;
4751 		}
4752 	if (r)
4753 		r = -EINVAL;
4754 	return r;
4755 }
4756 
4757 /*
4758  * kvm_set_guest_paused() indicates to the guest kernel that it has been
4759  * stopped by the hypervisor.  This function will be called from the host only.
4760  * EINVAL is returned when the host attempts to set the flag for a guest that
4761  * does not support pv clocks.
4762  */
4763 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
4764 {
4765 	if (!vcpu->arch.pv_time_enabled)
4766 		return -EINVAL;
4767 	vcpu->arch.pvclock_set_guest_stopped_request = true;
4768 	kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4769 	return 0;
4770 }
4771 
4772 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
4773 				     struct kvm_enable_cap *cap)
4774 {
4775 	int r;
4776 	uint16_t vmcs_version;
4777 	void __user *user_ptr;
4778 
4779 	if (cap->flags)
4780 		return -EINVAL;
4781 
4782 	switch (cap->cap) {
4783 	case KVM_CAP_HYPERV_SYNIC2:
4784 		if (cap->args[0])
4785 			return -EINVAL;
4786 		fallthrough;
4787 
4788 	case KVM_CAP_HYPERV_SYNIC:
4789 		if (!irqchip_in_kernel(vcpu->kvm))
4790 			return -EINVAL;
4791 		return kvm_hv_activate_synic(vcpu, cap->cap ==
4792 					     KVM_CAP_HYPERV_SYNIC2);
4793 	case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
4794 		if (!kvm_x86_ops.nested_ops->enable_evmcs)
4795 			return -ENOTTY;
4796 		r = kvm_x86_ops.nested_ops->enable_evmcs(vcpu, &vmcs_version);
4797 		if (!r) {
4798 			user_ptr = (void __user *)(uintptr_t)cap->args[0];
4799 			if (copy_to_user(user_ptr, &vmcs_version,
4800 					 sizeof(vmcs_version)))
4801 				r = -EFAULT;
4802 		}
4803 		return r;
4804 	case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
4805 		if (!kvm_x86_ops.enable_direct_tlbflush)
4806 			return -ENOTTY;
4807 
4808 		return static_call(kvm_x86_enable_direct_tlbflush)(vcpu);
4809 
4810 	case KVM_CAP_ENFORCE_PV_FEATURE_CPUID:
4811 		vcpu->arch.pv_cpuid.enforce = cap->args[0];
4812 		if (vcpu->arch.pv_cpuid.enforce)
4813 			kvm_update_pv_runtime(vcpu);
4814 
4815 		return 0;
4816 	default:
4817 		return -EINVAL;
4818 	}
4819 }
4820 
4821 long kvm_arch_vcpu_ioctl(struct file *filp,
4822 			 unsigned int ioctl, unsigned long arg)
4823 {
4824 	struct kvm_vcpu *vcpu = filp->private_data;
4825 	void __user *argp = (void __user *)arg;
4826 	int r;
4827 	union {
4828 		struct kvm_lapic_state *lapic;
4829 		struct kvm_xsave *xsave;
4830 		struct kvm_xcrs *xcrs;
4831 		void *buffer;
4832 	} u;
4833 
4834 	vcpu_load(vcpu);
4835 
4836 	u.buffer = NULL;
4837 	switch (ioctl) {
4838 	case KVM_GET_LAPIC: {
4839 		r = -EINVAL;
4840 		if (!lapic_in_kernel(vcpu))
4841 			goto out;
4842 		u.lapic = kzalloc(sizeof(struct kvm_lapic_state),
4843 				GFP_KERNEL_ACCOUNT);
4844 
4845 		r = -ENOMEM;
4846 		if (!u.lapic)
4847 			goto out;
4848 		r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
4849 		if (r)
4850 			goto out;
4851 		r = -EFAULT;
4852 		if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
4853 			goto out;
4854 		r = 0;
4855 		break;
4856 	}
4857 	case KVM_SET_LAPIC: {
4858 		r = -EINVAL;
4859 		if (!lapic_in_kernel(vcpu))
4860 			goto out;
4861 		u.lapic = memdup_user(argp, sizeof(*u.lapic));
4862 		if (IS_ERR(u.lapic)) {
4863 			r = PTR_ERR(u.lapic);
4864 			goto out_nofree;
4865 		}
4866 
4867 		r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
4868 		break;
4869 	}
4870 	case KVM_INTERRUPT: {
4871 		struct kvm_interrupt irq;
4872 
4873 		r = -EFAULT;
4874 		if (copy_from_user(&irq, argp, sizeof(irq)))
4875 			goto out;
4876 		r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
4877 		break;
4878 	}
4879 	case KVM_NMI: {
4880 		r = kvm_vcpu_ioctl_nmi(vcpu);
4881 		break;
4882 	}
4883 	case KVM_SMI: {
4884 		r = kvm_vcpu_ioctl_smi(vcpu);
4885 		break;
4886 	}
4887 	case KVM_SET_CPUID: {
4888 		struct kvm_cpuid __user *cpuid_arg = argp;
4889 		struct kvm_cpuid cpuid;
4890 
4891 		r = -EFAULT;
4892 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4893 			goto out;
4894 		r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
4895 		break;
4896 	}
4897 	case KVM_SET_CPUID2: {
4898 		struct kvm_cpuid2 __user *cpuid_arg = argp;
4899 		struct kvm_cpuid2 cpuid;
4900 
4901 		r = -EFAULT;
4902 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4903 			goto out;
4904 		r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
4905 					      cpuid_arg->entries);
4906 		break;
4907 	}
4908 	case KVM_GET_CPUID2: {
4909 		struct kvm_cpuid2 __user *cpuid_arg = argp;
4910 		struct kvm_cpuid2 cpuid;
4911 
4912 		r = -EFAULT;
4913 		if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4914 			goto out;
4915 		r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
4916 					      cpuid_arg->entries);
4917 		if (r)
4918 			goto out;
4919 		r = -EFAULT;
4920 		if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4921 			goto out;
4922 		r = 0;
4923 		break;
4924 	}
4925 	case KVM_GET_MSRS: {
4926 		int idx = srcu_read_lock(&vcpu->kvm->srcu);
4927 		r = msr_io(vcpu, argp, do_get_msr, 1);
4928 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
4929 		break;
4930 	}
4931 	case KVM_SET_MSRS: {
4932 		int idx = srcu_read_lock(&vcpu->kvm->srcu);
4933 		r = msr_io(vcpu, argp, do_set_msr, 0);
4934 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
4935 		break;
4936 	}
4937 	case KVM_TPR_ACCESS_REPORTING: {
4938 		struct kvm_tpr_access_ctl tac;
4939 
4940 		r = -EFAULT;
4941 		if (copy_from_user(&tac, argp, sizeof(tac)))
4942 			goto out;
4943 		r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
4944 		if (r)
4945 			goto out;
4946 		r = -EFAULT;
4947 		if (copy_to_user(argp, &tac, sizeof(tac)))
4948 			goto out;
4949 		r = 0;
4950 		break;
4951 	};
4952 	case KVM_SET_VAPIC_ADDR: {
4953 		struct kvm_vapic_addr va;
4954 		int idx;
4955 
4956 		r = -EINVAL;
4957 		if (!lapic_in_kernel(vcpu))
4958 			goto out;
4959 		r = -EFAULT;
4960 		if (copy_from_user(&va, argp, sizeof(va)))
4961 			goto out;
4962 		idx = srcu_read_lock(&vcpu->kvm->srcu);
4963 		r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
4964 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
4965 		break;
4966 	}
4967 	case KVM_X86_SETUP_MCE: {
4968 		u64 mcg_cap;
4969 
4970 		r = -EFAULT;
4971 		if (copy_from_user(&mcg_cap, argp, sizeof(mcg_cap)))
4972 			goto out;
4973 		r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
4974 		break;
4975 	}
4976 	case KVM_X86_SET_MCE: {
4977 		struct kvm_x86_mce mce;
4978 
4979 		r = -EFAULT;
4980 		if (copy_from_user(&mce, argp, sizeof(mce)))
4981 			goto out;
4982 		r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
4983 		break;
4984 	}
4985 	case KVM_GET_VCPU_EVENTS: {
4986 		struct kvm_vcpu_events events;
4987 
4988 		kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
4989 
4990 		r = -EFAULT;
4991 		if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
4992 			break;
4993 		r = 0;
4994 		break;
4995 	}
4996 	case KVM_SET_VCPU_EVENTS: {
4997 		struct kvm_vcpu_events events;
4998 
4999 		r = -EFAULT;
5000 		if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
5001 			break;
5002 
5003 		r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
5004 		break;
5005 	}
5006 	case KVM_GET_DEBUGREGS: {
5007 		struct kvm_debugregs dbgregs;
5008 
5009 		kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
5010 
5011 		r = -EFAULT;
5012 		if (copy_to_user(argp, &dbgregs,
5013 				 sizeof(struct kvm_debugregs)))
5014 			break;
5015 		r = 0;
5016 		break;
5017 	}
5018 	case KVM_SET_DEBUGREGS: {
5019 		struct kvm_debugregs dbgregs;
5020 
5021 		r = -EFAULT;
5022 		if (copy_from_user(&dbgregs, argp,
5023 				   sizeof(struct kvm_debugregs)))
5024 			break;
5025 
5026 		r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
5027 		break;
5028 	}
5029 	case KVM_GET_XSAVE: {
5030 		u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL_ACCOUNT);
5031 		r = -ENOMEM;
5032 		if (!u.xsave)
5033 			break;
5034 
5035 		kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
5036 
5037 		r = -EFAULT;
5038 		if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
5039 			break;
5040 		r = 0;
5041 		break;
5042 	}
5043 	case KVM_SET_XSAVE: {
5044 		u.xsave = memdup_user(argp, sizeof(*u.xsave));
5045 		if (IS_ERR(u.xsave)) {
5046 			r = PTR_ERR(u.xsave);
5047 			goto out_nofree;
5048 		}
5049 
5050 		r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
5051 		break;
5052 	}
5053 	case KVM_GET_XCRS: {
5054 		u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL_ACCOUNT);
5055 		r = -ENOMEM;
5056 		if (!u.xcrs)
5057 			break;
5058 
5059 		kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
5060 
5061 		r = -EFAULT;
5062 		if (copy_to_user(argp, u.xcrs,
5063 				 sizeof(struct kvm_xcrs)))
5064 			break;
5065 		r = 0;
5066 		break;
5067 	}
5068 	case KVM_SET_XCRS: {
5069 		u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
5070 		if (IS_ERR(u.xcrs)) {
5071 			r = PTR_ERR(u.xcrs);
5072 			goto out_nofree;
5073 		}
5074 
5075 		r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
5076 		break;
5077 	}
5078 	case KVM_SET_TSC_KHZ: {
5079 		u32 user_tsc_khz;
5080 
5081 		r = -EINVAL;
5082 		user_tsc_khz = (u32)arg;
5083 
5084 		if (kvm_has_tsc_control &&
5085 		    user_tsc_khz >= kvm_max_guest_tsc_khz)
5086 			goto out;
5087 
5088 		if (user_tsc_khz == 0)
5089 			user_tsc_khz = tsc_khz;
5090 
5091 		if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
5092 			r = 0;
5093 
5094 		goto out;
5095 	}
5096 	case KVM_GET_TSC_KHZ: {
5097 		r = vcpu->arch.virtual_tsc_khz;
5098 		goto out;
5099 	}
5100 	case KVM_KVMCLOCK_CTRL: {
5101 		r = kvm_set_guest_paused(vcpu);
5102 		goto out;
5103 	}
5104 	case KVM_ENABLE_CAP: {
5105 		struct kvm_enable_cap cap;
5106 
5107 		r = -EFAULT;
5108 		if (copy_from_user(&cap, argp, sizeof(cap)))
5109 			goto out;
5110 		r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
5111 		break;
5112 	}
5113 	case KVM_GET_NESTED_STATE: {
5114 		struct kvm_nested_state __user *user_kvm_nested_state = argp;
5115 		u32 user_data_size;
5116 
5117 		r = -EINVAL;
5118 		if (!kvm_x86_ops.nested_ops->get_state)
5119 			break;
5120 
5121 		BUILD_BUG_ON(sizeof(user_data_size) != sizeof(user_kvm_nested_state->size));
5122 		r = -EFAULT;
5123 		if (get_user(user_data_size, &user_kvm_nested_state->size))
5124 			break;
5125 
5126 		r = kvm_x86_ops.nested_ops->get_state(vcpu, user_kvm_nested_state,
5127 						     user_data_size);
5128 		if (r < 0)
5129 			break;
5130 
5131 		if (r > user_data_size) {
5132 			if (put_user(r, &user_kvm_nested_state->size))
5133 				r = -EFAULT;
5134 			else
5135 				r = -E2BIG;
5136 			break;
5137 		}
5138 
5139 		r = 0;
5140 		break;
5141 	}
5142 	case KVM_SET_NESTED_STATE: {
5143 		struct kvm_nested_state __user *user_kvm_nested_state = argp;
5144 		struct kvm_nested_state kvm_state;
5145 		int idx;
5146 
5147 		r = -EINVAL;
5148 		if (!kvm_x86_ops.nested_ops->set_state)
5149 			break;
5150 
5151 		r = -EFAULT;
5152 		if (copy_from_user(&kvm_state, user_kvm_nested_state, sizeof(kvm_state)))
5153 			break;
5154 
5155 		r = -EINVAL;
5156 		if (kvm_state.size < sizeof(kvm_state))
5157 			break;
5158 
5159 		if (kvm_state.flags &
5160 		    ~(KVM_STATE_NESTED_RUN_PENDING | KVM_STATE_NESTED_GUEST_MODE
5161 		      | KVM_STATE_NESTED_EVMCS | KVM_STATE_NESTED_MTF_PENDING
5162 		      | KVM_STATE_NESTED_GIF_SET))
5163 			break;
5164 
5165 		/* nested_run_pending implies guest_mode.  */
5166 		if ((kvm_state.flags & KVM_STATE_NESTED_RUN_PENDING)
5167 		    && !(kvm_state.flags & KVM_STATE_NESTED_GUEST_MODE))
5168 			break;
5169 
5170 		idx = srcu_read_lock(&vcpu->kvm->srcu);
5171 		r = kvm_x86_ops.nested_ops->set_state(vcpu, user_kvm_nested_state, &kvm_state);
5172 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
5173 		break;
5174 	}
5175 	case KVM_GET_SUPPORTED_HV_CPUID:
5176 		r = kvm_ioctl_get_supported_hv_cpuid(vcpu, argp);
5177 		break;
5178 #ifdef CONFIG_KVM_XEN
5179 	case KVM_XEN_VCPU_GET_ATTR: {
5180 		struct kvm_xen_vcpu_attr xva;
5181 
5182 		r = -EFAULT;
5183 		if (copy_from_user(&xva, argp, sizeof(xva)))
5184 			goto out;
5185 		r = kvm_xen_vcpu_get_attr(vcpu, &xva);
5186 		if (!r && copy_to_user(argp, &xva, sizeof(xva)))
5187 			r = -EFAULT;
5188 		break;
5189 	}
5190 	case KVM_XEN_VCPU_SET_ATTR: {
5191 		struct kvm_xen_vcpu_attr xva;
5192 
5193 		r = -EFAULT;
5194 		if (copy_from_user(&xva, argp, sizeof(xva)))
5195 			goto out;
5196 		r = kvm_xen_vcpu_set_attr(vcpu, &xva);
5197 		break;
5198 	}
5199 #endif
5200 	default:
5201 		r = -EINVAL;
5202 	}
5203 out:
5204 	kfree(u.buffer);
5205 out_nofree:
5206 	vcpu_put(vcpu);
5207 	return r;
5208 }
5209 
5210 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
5211 {
5212 	return VM_FAULT_SIGBUS;
5213 }
5214 
5215 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
5216 {
5217 	int ret;
5218 
5219 	if (addr > (unsigned int)(-3 * PAGE_SIZE))
5220 		return -EINVAL;
5221 	ret = static_call(kvm_x86_set_tss_addr)(kvm, addr);
5222 	return ret;
5223 }
5224 
5225 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
5226 					      u64 ident_addr)
5227 {
5228 	return static_call(kvm_x86_set_identity_map_addr)(kvm, ident_addr);
5229 }
5230 
5231 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
5232 					 unsigned long kvm_nr_mmu_pages)
5233 {
5234 	if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
5235 		return -EINVAL;
5236 
5237 	mutex_lock(&kvm->slots_lock);
5238 
5239 	kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
5240 	kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
5241 
5242 	mutex_unlock(&kvm->slots_lock);
5243 	return 0;
5244 }
5245 
5246 static unsigned long kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
5247 {
5248 	return kvm->arch.n_max_mmu_pages;
5249 }
5250 
5251 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
5252 {
5253 	struct kvm_pic *pic = kvm->arch.vpic;
5254 	int r;
5255 
5256 	r = 0;
5257 	switch (chip->chip_id) {
5258 	case KVM_IRQCHIP_PIC_MASTER:
5259 		memcpy(&chip->chip.pic, &pic->pics[0],
5260 			sizeof(struct kvm_pic_state));
5261 		break;
5262 	case KVM_IRQCHIP_PIC_SLAVE:
5263 		memcpy(&chip->chip.pic, &pic->pics[1],
5264 			sizeof(struct kvm_pic_state));
5265 		break;
5266 	case KVM_IRQCHIP_IOAPIC:
5267 		kvm_get_ioapic(kvm, &chip->chip.ioapic);
5268 		break;
5269 	default:
5270 		r = -EINVAL;
5271 		break;
5272 	}
5273 	return r;
5274 }
5275 
5276 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
5277 {
5278 	struct kvm_pic *pic = kvm->arch.vpic;
5279 	int r;
5280 
5281 	r = 0;
5282 	switch (chip->chip_id) {
5283 	case KVM_IRQCHIP_PIC_MASTER:
5284 		spin_lock(&pic->lock);
5285 		memcpy(&pic->pics[0], &chip->chip.pic,
5286 			sizeof(struct kvm_pic_state));
5287 		spin_unlock(&pic->lock);
5288 		break;
5289 	case KVM_IRQCHIP_PIC_SLAVE:
5290 		spin_lock(&pic->lock);
5291 		memcpy(&pic->pics[1], &chip->chip.pic,
5292 			sizeof(struct kvm_pic_state));
5293 		spin_unlock(&pic->lock);
5294 		break;
5295 	case KVM_IRQCHIP_IOAPIC:
5296 		kvm_set_ioapic(kvm, &chip->chip.ioapic);
5297 		break;
5298 	default:
5299 		r = -EINVAL;
5300 		break;
5301 	}
5302 	kvm_pic_update_irq(pic);
5303 	return r;
5304 }
5305 
5306 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
5307 {
5308 	struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
5309 
5310 	BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
5311 
5312 	mutex_lock(&kps->lock);
5313 	memcpy(ps, &kps->channels, sizeof(*ps));
5314 	mutex_unlock(&kps->lock);
5315 	return 0;
5316 }
5317 
5318 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
5319 {
5320 	int i;
5321 	struct kvm_pit *pit = kvm->arch.vpit;
5322 
5323 	mutex_lock(&pit->pit_state.lock);
5324 	memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
5325 	for (i = 0; i < 3; i++)
5326 		kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
5327 	mutex_unlock(&pit->pit_state.lock);
5328 	return 0;
5329 }
5330 
5331 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
5332 {
5333 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
5334 	memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
5335 		sizeof(ps->channels));
5336 	ps->flags = kvm->arch.vpit->pit_state.flags;
5337 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
5338 	memset(&ps->reserved, 0, sizeof(ps->reserved));
5339 	return 0;
5340 }
5341 
5342 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
5343 {
5344 	int start = 0;
5345 	int i;
5346 	u32 prev_legacy, cur_legacy;
5347 	struct kvm_pit *pit = kvm->arch.vpit;
5348 
5349 	mutex_lock(&pit->pit_state.lock);
5350 	prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
5351 	cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
5352 	if (!prev_legacy && cur_legacy)
5353 		start = 1;
5354 	memcpy(&pit->pit_state.channels, &ps->channels,
5355 	       sizeof(pit->pit_state.channels));
5356 	pit->pit_state.flags = ps->flags;
5357 	for (i = 0; i < 3; i++)
5358 		kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
5359 				   start && i == 0);
5360 	mutex_unlock(&pit->pit_state.lock);
5361 	return 0;
5362 }
5363 
5364 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
5365 				 struct kvm_reinject_control *control)
5366 {
5367 	struct kvm_pit *pit = kvm->arch.vpit;
5368 
5369 	/* pit->pit_state.lock was overloaded to prevent userspace from getting
5370 	 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
5371 	 * ioctls in parallel.  Use a separate lock if that ioctl isn't rare.
5372 	 */
5373 	mutex_lock(&pit->pit_state.lock);
5374 	kvm_pit_set_reinject(pit, control->pit_reinject);
5375 	mutex_unlock(&pit->pit_state.lock);
5376 
5377 	return 0;
5378 }
5379 
5380 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
5381 {
5382 
5383 	/*
5384 	 * Flush all CPUs' dirty log buffers to the  dirty_bitmap.  Called
5385 	 * before reporting dirty_bitmap to userspace.  KVM flushes the buffers
5386 	 * on all VM-Exits, thus we only need to kick running vCPUs to force a
5387 	 * VM-Exit.
5388 	 */
5389 	struct kvm_vcpu *vcpu;
5390 	int i;
5391 
5392 	kvm_for_each_vcpu(i, vcpu, kvm)
5393 		kvm_vcpu_kick(vcpu);
5394 }
5395 
5396 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
5397 			bool line_status)
5398 {
5399 	if (!irqchip_in_kernel(kvm))
5400 		return -ENXIO;
5401 
5402 	irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
5403 					irq_event->irq, irq_event->level,
5404 					line_status);
5405 	return 0;
5406 }
5407 
5408 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
5409 			    struct kvm_enable_cap *cap)
5410 {
5411 	int r;
5412 
5413 	if (cap->flags)
5414 		return -EINVAL;
5415 
5416 	switch (cap->cap) {
5417 	case KVM_CAP_DISABLE_QUIRKS:
5418 		kvm->arch.disabled_quirks = cap->args[0];
5419 		r = 0;
5420 		break;
5421 	case KVM_CAP_SPLIT_IRQCHIP: {
5422 		mutex_lock(&kvm->lock);
5423 		r = -EINVAL;
5424 		if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
5425 			goto split_irqchip_unlock;
5426 		r = -EEXIST;
5427 		if (irqchip_in_kernel(kvm))
5428 			goto split_irqchip_unlock;
5429 		if (kvm->created_vcpus)
5430 			goto split_irqchip_unlock;
5431 		r = kvm_setup_empty_irq_routing(kvm);
5432 		if (r)
5433 			goto split_irqchip_unlock;
5434 		/* Pairs with irqchip_in_kernel. */
5435 		smp_wmb();
5436 		kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
5437 		kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
5438 		r = 0;
5439 split_irqchip_unlock:
5440 		mutex_unlock(&kvm->lock);
5441 		break;
5442 	}
5443 	case KVM_CAP_X2APIC_API:
5444 		r = -EINVAL;
5445 		if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
5446 			break;
5447 
5448 		if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
5449 			kvm->arch.x2apic_format = true;
5450 		if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
5451 			kvm->arch.x2apic_broadcast_quirk_disabled = true;
5452 
5453 		r = 0;
5454 		break;
5455 	case KVM_CAP_X86_DISABLE_EXITS:
5456 		r = -EINVAL;
5457 		if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
5458 			break;
5459 
5460 		if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
5461 			kvm_can_mwait_in_guest())
5462 			kvm->arch.mwait_in_guest = true;
5463 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_HLT)
5464 			kvm->arch.hlt_in_guest = true;
5465 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
5466 			kvm->arch.pause_in_guest = true;
5467 		if (cap->args[0] & KVM_X86_DISABLE_EXITS_CSTATE)
5468 			kvm->arch.cstate_in_guest = true;
5469 		r = 0;
5470 		break;
5471 	case KVM_CAP_MSR_PLATFORM_INFO:
5472 		kvm->arch.guest_can_read_msr_platform_info = cap->args[0];
5473 		r = 0;
5474 		break;
5475 	case KVM_CAP_EXCEPTION_PAYLOAD:
5476 		kvm->arch.exception_payload_enabled = cap->args[0];
5477 		r = 0;
5478 		break;
5479 	case KVM_CAP_X86_USER_SPACE_MSR:
5480 		kvm->arch.user_space_msr_mask = cap->args[0];
5481 		r = 0;
5482 		break;
5483 	case KVM_CAP_X86_BUS_LOCK_EXIT:
5484 		r = -EINVAL;
5485 		if (cap->args[0] & ~KVM_BUS_LOCK_DETECTION_VALID_MODE)
5486 			break;
5487 
5488 		if ((cap->args[0] & KVM_BUS_LOCK_DETECTION_OFF) &&
5489 		    (cap->args[0] & KVM_BUS_LOCK_DETECTION_EXIT))
5490 			break;
5491 
5492 		if (kvm_has_bus_lock_exit &&
5493 		    cap->args[0] & KVM_BUS_LOCK_DETECTION_EXIT)
5494 			kvm->arch.bus_lock_detection_enabled = true;
5495 		r = 0;
5496 		break;
5497 #ifdef CONFIG_X86_SGX_KVM
5498 	case KVM_CAP_SGX_ATTRIBUTE: {
5499 		unsigned long allowed_attributes = 0;
5500 
5501 		r = sgx_set_attribute(&allowed_attributes, cap->args[0]);
5502 		if (r)
5503 			break;
5504 
5505 		/* KVM only supports the PROVISIONKEY privileged attribute. */
5506 		if ((allowed_attributes & SGX_ATTR_PROVISIONKEY) &&
5507 		    !(allowed_attributes & ~SGX_ATTR_PROVISIONKEY))
5508 			kvm->arch.sgx_provisioning_allowed = true;
5509 		else
5510 			r = -EINVAL;
5511 		break;
5512 	}
5513 #endif
5514 	case KVM_CAP_VM_COPY_ENC_CONTEXT_FROM:
5515 		r = -EINVAL;
5516 		if (kvm_x86_ops.vm_copy_enc_context_from)
5517 			r = kvm_x86_ops.vm_copy_enc_context_from(kvm, cap->args[0]);
5518 		return r;
5519 	default:
5520 		r = -EINVAL;
5521 		break;
5522 	}
5523 	return r;
5524 }
5525 
5526 static struct kvm_x86_msr_filter *kvm_alloc_msr_filter(bool default_allow)
5527 {
5528 	struct kvm_x86_msr_filter *msr_filter;
5529 
5530 	msr_filter = kzalloc(sizeof(*msr_filter), GFP_KERNEL_ACCOUNT);
5531 	if (!msr_filter)
5532 		return NULL;
5533 
5534 	msr_filter->default_allow = default_allow;
5535 	return msr_filter;
5536 }
5537 
5538 static void kvm_free_msr_filter(struct kvm_x86_msr_filter *msr_filter)
5539 {
5540 	u32 i;
5541 
5542 	if (!msr_filter)
5543 		return;
5544 
5545 	for (i = 0; i < msr_filter->count; i++)
5546 		kfree(msr_filter->ranges[i].bitmap);
5547 
5548 	kfree(msr_filter);
5549 }
5550 
5551 static int kvm_add_msr_filter(struct kvm_x86_msr_filter *msr_filter,
5552 			      struct kvm_msr_filter_range *user_range)
5553 {
5554 	unsigned long *bitmap = NULL;
5555 	size_t bitmap_size;
5556 
5557 	if (!user_range->nmsrs)
5558 		return 0;
5559 
5560 	if (user_range->flags & ~(KVM_MSR_FILTER_READ | KVM_MSR_FILTER_WRITE))
5561 		return -EINVAL;
5562 
5563 	if (!user_range->flags)
5564 		return -EINVAL;
5565 
5566 	bitmap_size = BITS_TO_LONGS(user_range->nmsrs) * sizeof(long);
5567 	if (!bitmap_size || bitmap_size > KVM_MSR_FILTER_MAX_BITMAP_SIZE)
5568 		return -EINVAL;
5569 
5570 	bitmap = memdup_user((__user u8*)user_range->bitmap, bitmap_size);
5571 	if (IS_ERR(bitmap))
5572 		return PTR_ERR(bitmap);
5573 
5574 	msr_filter->ranges[msr_filter->count] = (struct msr_bitmap_range) {
5575 		.flags = user_range->flags,
5576 		.base = user_range->base,
5577 		.nmsrs = user_range->nmsrs,
5578 		.bitmap = bitmap,
5579 	};
5580 
5581 	msr_filter->count++;
5582 	return 0;
5583 }
5584 
5585 static int kvm_vm_ioctl_set_msr_filter(struct kvm *kvm, void __user *argp)
5586 {
5587 	struct kvm_msr_filter __user *user_msr_filter = argp;
5588 	struct kvm_x86_msr_filter *new_filter, *old_filter;
5589 	struct kvm_msr_filter filter;
5590 	bool default_allow;
5591 	bool empty = true;
5592 	int r = 0;
5593 	u32 i;
5594 
5595 	if (copy_from_user(&filter, user_msr_filter, sizeof(filter)))
5596 		return -EFAULT;
5597 
5598 	for (i = 0; i < ARRAY_SIZE(filter.ranges); i++)
5599 		empty &= !filter.ranges[i].nmsrs;
5600 
5601 	default_allow = !(filter.flags & KVM_MSR_FILTER_DEFAULT_DENY);
5602 	if (empty && !default_allow)
5603 		return -EINVAL;
5604 
5605 	new_filter = kvm_alloc_msr_filter(default_allow);
5606 	if (!new_filter)
5607 		return -ENOMEM;
5608 
5609 	for (i = 0; i < ARRAY_SIZE(filter.ranges); i++) {
5610 		r = kvm_add_msr_filter(new_filter, &filter.ranges[i]);
5611 		if (r) {
5612 			kvm_free_msr_filter(new_filter);
5613 			return r;
5614 		}
5615 	}
5616 
5617 	mutex_lock(&kvm->lock);
5618 
5619 	/* The per-VM filter is protected by kvm->lock... */
5620 	old_filter = srcu_dereference_check(kvm->arch.msr_filter, &kvm->srcu, 1);
5621 
5622 	rcu_assign_pointer(kvm->arch.msr_filter, new_filter);
5623 	synchronize_srcu(&kvm->srcu);
5624 
5625 	kvm_free_msr_filter(old_filter);
5626 
5627 	kvm_make_all_cpus_request(kvm, KVM_REQ_MSR_FILTER_CHANGED);
5628 	mutex_unlock(&kvm->lock);
5629 
5630 	return 0;
5631 }
5632 
5633 long kvm_arch_vm_ioctl(struct file *filp,
5634 		       unsigned int ioctl, unsigned long arg)
5635 {
5636 	struct kvm *kvm = filp->private_data;
5637 	void __user *argp = (void __user *)arg;
5638 	int r = -ENOTTY;
5639 	/*
5640 	 * This union makes it completely explicit to gcc-3.x
5641 	 * that these two variables' stack usage should be
5642 	 * combined, not added together.
5643 	 */
5644 	union {
5645 		struct kvm_pit_state ps;
5646 		struct kvm_pit_state2 ps2;
5647 		struct kvm_pit_config pit_config;
5648 	} u;
5649 
5650 	switch (ioctl) {
5651 	case KVM_SET_TSS_ADDR:
5652 		r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
5653 		break;
5654 	case KVM_SET_IDENTITY_MAP_ADDR: {
5655 		u64 ident_addr;
5656 
5657 		mutex_lock(&kvm->lock);
5658 		r = -EINVAL;
5659 		if (kvm->created_vcpus)
5660 			goto set_identity_unlock;
5661 		r = -EFAULT;
5662 		if (copy_from_user(&ident_addr, argp, sizeof(ident_addr)))
5663 			goto set_identity_unlock;
5664 		r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
5665 set_identity_unlock:
5666 		mutex_unlock(&kvm->lock);
5667 		break;
5668 	}
5669 	case KVM_SET_NR_MMU_PAGES:
5670 		r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
5671 		break;
5672 	case KVM_GET_NR_MMU_PAGES:
5673 		r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
5674 		break;
5675 	case KVM_CREATE_IRQCHIP: {
5676 		mutex_lock(&kvm->lock);
5677 
5678 		r = -EEXIST;
5679 		if (irqchip_in_kernel(kvm))
5680 			goto create_irqchip_unlock;
5681 
5682 		r = -EINVAL;
5683 		if (kvm->created_vcpus)
5684 			goto create_irqchip_unlock;
5685 
5686 		r = kvm_pic_init(kvm);
5687 		if (r)
5688 			goto create_irqchip_unlock;
5689 
5690 		r = kvm_ioapic_init(kvm);
5691 		if (r) {
5692 			kvm_pic_destroy(kvm);
5693 			goto create_irqchip_unlock;
5694 		}
5695 
5696 		r = kvm_setup_default_irq_routing(kvm);
5697 		if (r) {
5698 			kvm_ioapic_destroy(kvm);
5699 			kvm_pic_destroy(kvm);
5700 			goto create_irqchip_unlock;
5701 		}
5702 		/* Write kvm->irq_routing before enabling irqchip_in_kernel. */
5703 		smp_wmb();
5704 		kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
5705 	create_irqchip_unlock:
5706 		mutex_unlock(&kvm->lock);
5707 		break;
5708 	}
5709 	case KVM_CREATE_PIT:
5710 		u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
5711 		goto create_pit;
5712 	case KVM_CREATE_PIT2:
5713 		r = -EFAULT;
5714 		if (copy_from_user(&u.pit_config, argp,
5715 				   sizeof(struct kvm_pit_config)))
5716 			goto out;
5717 	create_pit:
5718 		mutex_lock(&kvm->lock);
5719 		r = -EEXIST;
5720 		if (kvm->arch.vpit)
5721 			goto create_pit_unlock;
5722 		r = -ENOMEM;
5723 		kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
5724 		if (kvm->arch.vpit)
5725 			r = 0;
5726 	create_pit_unlock:
5727 		mutex_unlock(&kvm->lock);
5728 		break;
5729 	case KVM_GET_IRQCHIP: {
5730 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
5731 		struct kvm_irqchip *chip;
5732 
5733 		chip = memdup_user(argp, sizeof(*chip));
5734 		if (IS_ERR(chip)) {
5735 			r = PTR_ERR(chip);
5736 			goto out;
5737 		}
5738 
5739 		r = -ENXIO;
5740 		if (!irqchip_kernel(kvm))
5741 			goto get_irqchip_out;
5742 		r = kvm_vm_ioctl_get_irqchip(kvm, chip);
5743 		if (r)
5744 			goto get_irqchip_out;
5745 		r = -EFAULT;
5746 		if (copy_to_user(argp, chip, sizeof(*chip)))
5747 			goto get_irqchip_out;
5748 		r = 0;
5749 	get_irqchip_out:
5750 		kfree(chip);
5751 		break;
5752 	}
5753 	case KVM_SET_IRQCHIP: {
5754 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
5755 		struct kvm_irqchip *chip;
5756 
5757 		chip = memdup_user(argp, sizeof(*chip));
5758 		if (IS_ERR(chip)) {
5759 			r = PTR_ERR(chip);
5760 			goto out;
5761 		}
5762 
5763 		r = -ENXIO;
5764 		if (!irqchip_kernel(kvm))
5765 			goto set_irqchip_out;
5766 		r = kvm_vm_ioctl_set_irqchip(kvm, chip);
5767 	set_irqchip_out:
5768 		kfree(chip);
5769 		break;
5770 	}
5771 	case KVM_GET_PIT: {
5772 		r = -EFAULT;
5773 		if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
5774 			goto out;
5775 		r = -ENXIO;
5776 		if (!kvm->arch.vpit)
5777 			goto out;
5778 		r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
5779 		if (r)
5780 			goto out;
5781 		r = -EFAULT;
5782 		if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
5783 			goto out;
5784 		r = 0;
5785 		break;
5786 	}
5787 	case KVM_SET_PIT: {
5788 		r = -EFAULT;
5789 		if (copy_from_user(&u.ps, argp, sizeof(u.ps)))
5790 			goto out;
5791 		mutex_lock(&kvm->lock);
5792 		r = -ENXIO;
5793 		if (!kvm->arch.vpit)
5794 			goto set_pit_out;
5795 		r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
5796 set_pit_out:
5797 		mutex_unlock(&kvm->lock);
5798 		break;
5799 	}
5800 	case KVM_GET_PIT2: {
5801 		r = -ENXIO;
5802 		if (!kvm->arch.vpit)
5803 			goto out;
5804 		r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
5805 		if (r)
5806 			goto out;
5807 		r = -EFAULT;
5808 		if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
5809 			goto out;
5810 		r = 0;
5811 		break;
5812 	}
5813 	case KVM_SET_PIT2: {
5814 		r = -EFAULT;
5815 		if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
5816 			goto out;
5817 		mutex_lock(&kvm->lock);
5818 		r = -ENXIO;
5819 		if (!kvm->arch.vpit)
5820 			goto set_pit2_out;
5821 		r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
5822 set_pit2_out:
5823 		mutex_unlock(&kvm->lock);
5824 		break;
5825 	}
5826 	case KVM_REINJECT_CONTROL: {
5827 		struct kvm_reinject_control control;
5828 		r =  -EFAULT;
5829 		if (copy_from_user(&control, argp, sizeof(control)))
5830 			goto out;
5831 		r = -ENXIO;
5832 		if (!kvm->arch.vpit)
5833 			goto out;
5834 		r = kvm_vm_ioctl_reinject(kvm, &control);
5835 		break;
5836 	}
5837 	case KVM_SET_BOOT_CPU_ID:
5838 		r = 0;
5839 		mutex_lock(&kvm->lock);
5840 		if (kvm->created_vcpus)
5841 			r = -EBUSY;
5842 		else
5843 			kvm->arch.bsp_vcpu_id = arg;
5844 		mutex_unlock(&kvm->lock);
5845 		break;
5846 #ifdef CONFIG_KVM_XEN
5847 	case KVM_XEN_HVM_CONFIG: {
5848 		struct kvm_xen_hvm_config xhc;
5849 		r = -EFAULT;
5850 		if (copy_from_user(&xhc, argp, sizeof(xhc)))
5851 			goto out;
5852 		r = kvm_xen_hvm_config(kvm, &xhc);
5853 		break;
5854 	}
5855 	case KVM_XEN_HVM_GET_ATTR: {
5856 		struct kvm_xen_hvm_attr xha;
5857 
5858 		r = -EFAULT;
5859 		if (copy_from_user(&xha, argp, sizeof(xha)))
5860 			goto out;
5861 		r = kvm_xen_hvm_get_attr(kvm, &xha);
5862 		if (!r && copy_to_user(argp, &xha, sizeof(xha)))
5863 			r = -EFAULT;
5864 		break;
5865 	}
5866 	case KVM_XEN_HVM_SET_ATTR: {
5867 		struct kvm_xen_hvm_attr xha;
5868 
5869 		r = -EFAULT;
5870 		if (copy_from_user(&xha, argp, sizeof(xha)))
5871 			goto out;
5872 		r = kvm_xen_hvm_set_attr(kvm, &xha);
5873 		break;
5874 	}
5875 #endif
5876 	case KVM_SET_CLOCK: {
5877 		struct kvm_arch *ka = &kvm->arch;
5878 		struct kvm_clock_data user_ns;
5879 		u64 now_ns;
5880 
5881 		r = -EFAULT;
5882 		if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
5883 			goto out;
5884 
5885 		r = -EINVAL;
5886 		if (user_ns.flags)
5887 			goto out;
5888 
5889 		r = 0;
5890 		/*
5891 		 * TODO: userspace has to take care of races with VCPU_RUN, so
5892 		 * kvm_gen_update_masterclock() can be cut down to locked
5893 		 * pvclock_update_vm_gtod_copy().
5894 		 */
5895 		kvm_gen_update_masterclock(kvm);
5896 
5897 		/*
5898 		 * This pairs with kvm_guest_time_update(): when masterclock is
5899 		 * in use, we use master_kernel_ns + kvmclock_offset to set
5900 		 * unsigned 'system_time' so if we use get_kvmclock_ns() (which
5901 		 * is slightly ahead) here we risk going negative on unsigned
5902 		 * 'system_time' when 'user_ns.clock' is very small.
5903 		 */
5904 		spin_lock_irq(&ka->pvclock_gtod_sync_lock);
5905 		if (kvm->arch.use_master_clock)
5906 			now_ns = ka->master_kernel_ns;
5907 		else
5908 			now_ns = get_kvmclock_base_ns();
5909 		ka->kvmclock_offset = user_ns.clock - now_ns;
5910 		spin_unlock_irq(&ka->pvclock_gtod_sync_lock);
5911 
5912 		kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE);
5913 		break;
5914 	}
5915 	case KVM_GET_CLOCK: {
5916 		struct kvm_clock_data user_ns;
5917 		u64 now_ns;
5918 
5919 		now_ns = get_kvmclock_ns(kvm);
5920 		user_ns.clock = now_ns;
5921 		user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
5922 		memset(&user_ns.pad, 0, sizeof(user_ns.pad));
5923 
5924 		r = -EFAULT;
5925 		if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
5926 			goto out;
5927 		r = 0;
5928 		break;
5929 	}
5930 	case KVM_MEMORY_ENCRYPT_OP: {
5931 		r = -ENOTTY;
5932 		if (kvm_x86_ops.mem_enc_op)
5933 			r = static_call(kvm_x86_mem_enc_op)(kvm, argp);
5934 		break;
5935 	}
5936 	case KVM_MEMORY_ENCRYPT_REG_REGION: {
5937 		struct kvm_enc_region region;
5938 
5939 		r = -EFAULT;
5940 		if (copy_from_user(&region, argp, sizeof(region)))
5941 			goto out;
5942 
5943 		r = -ENOTTY;
5944 		if (kvm_x86_ops.mem_enc_reg_region)
5945 			r = static_call(kvm_x86_mem_enc_reg_region)(kvm, &region);
5946 		break;
5947 	}
5948 	case KVM_MEMORY_ENCRYPT_UNREG_REGION: {
5949 		struct kvm_enc_region region;
5950 
5951 		r = -EFAULT;
5952 		if (copy_from_user(&region, argp, sizeof(region)))
5953 			goto out;
5954 
5955 		r = -ENOTTY;
5956 		if (kvm_x86_ops.mem_enc_unreg_region)
5957 			r = static_call(kvm_x86_mem_enc_unreg_region)(kvm, &region);
5958 		break;
5959 	}
5960 	case KVM_HYPERV_EVENTFD: {
5961 		struct kvm_hyperv_eventfd hvevfd;
5962 
5963 		r = -EFAULT;
5964 		if (copy_from_user(&hvevfd, argp, sizeof(hvevfd)))
5965 			goto out;
5966 		r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd);
5967 		break;
5968 	}
5969 	case KVM_SET_PMU_EVENT_FILTER:
5970 		r = kvm_vm_ioctl_set_pmu_event_filter(kvm, argp);
5971 		break;
5972 	case KVM_X86_SET_MSR_FILTER:
5973 		r = kvm_vm_ioctl_set_msr_filter(kvm, argp);
5974 		break;
5975 	default:
5976 		r = -ENOTTY;
5977 	}
5978 out:
5979 	return r;
5980 }
5981 
5982 static void kvm_init_msr_list(void)
5983 {
5984 	struct x86_pmu_capability x86_pmu;
5985 	u32 dummy[2];
5986 	unsigned i;
5987 
5988 	BUILD_BUG_ON_MSG(INTEL_PMC_MAX_FIXED != 4,
5989 			 "Please update the fixed PMCs in msrs_to_saved_all[]");
5990 
5991 	perf_get_x86_pmu_capability(&x86_pmu);
5992 
5993 	num_msrs_to_save = 0;
5994 	num_emulated_msrs = 0;
5995 	num_msr_based_features = 0;
5996 
5997 	for (i = 0; i < ARRAY_SIZE(msrs_to_save_all); i++) {
5998 		if (rdmsr_safe(msrs_to_save_all[i], &dummy[0], &dummy[1]) < 0)
5999 			continue;
6000 
6001 		/*
6002 		 * Even MSRs that are valid in the host may not be exposed
6003 		 * to the guests in some cases.
6004 		 */
6005 		switch (msrs_to_save_all[i]) {
6006 		case MSR_IA32_BNDCFGS:
6007 			if (!kvm_mpx_supported())
6008 				continue;
6009 			break;
6010 		case MSR_TSC_AUX:
6011 			if (!kvm_cpu_cap_has(X86_FEATURE_RDTSCP) &&
6012 			    !kvm_cpu_cap_has(X86_FEATURE_RDPID))
6013 				continue;
6014 			break;
6015 		case MSR_IA32_UMWAIT_CONTROL:
6016 			if (!kvm_cpu_cap_has(X86_FEATURE_WAITPKG))
6017 				continue;
6018 			break;
6019 		case MSR_IA32_RTIT_CTL:
6020 		case MSR_IA32_RTIT_STATUS:
6021 			if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT))
6022 				continue;
6023 			break;
6024 		case MSR_IA32_RTIT_CR3_MATCH:
6025 			if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
6026 			    !intel_pt_validate_hw_cap(PT_CAP_cr3_filtering))
6027 				continue;
6028 			break;
6029 		case MSR_IA32_RTIT_OUTPUT_BASE:
6030 		case MSR_IA32_RTIT_OUTPUT_MASK:
6031 			if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
6032 				(!intel_pt_validate_hw_cap(PT_CAP_topa_output) &&
6033 				 !intel_pt_validate_hw_cap(PT_CAP_single_range_output)))
6034 				continue;
6035 			break;
6036 		case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
6037 			if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
6038 				msrs_to_save_all[i] - MSR_IA32_RTIT_ADDR0_A >=
6039 				intel_pt_validate_hw_cap(PT_CAP_num_address_ranges) * 2)
6040 				continue;
6041 			break;
6042 		case MSR_ARCH_PERFMON_PERFCTR0 ... MSR_ARCH_PERFMON_PERFCTR0 + 17:
6043 			if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_PERFCTR0 >=
6044 			    min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
6045 				continue;
6046 			break;
6047 		case MSR_ARCH_PERFMON_EVENTSEL0 ... MSR_ARCH_PERFMON_EVENTSEL0 + 17:
6048 			if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_EVENTSEL0 >=
6049 			    min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
6050 				continue;
6051 			break;
6052 		default:
6053 			break;
6054 		}
6055 
6056 		msrs_to_save[num_msrs_to_save++] = msrs_to_save_all[i];
6057 	}
6058 
6059 	for (i = 0; i < ARRAY_SIZE(emulated_msrs_all); i++) {
6060 		if (!static_call(kvm_x86_has_emulated_msr)(NULL, emulated_msrs_all[i]))
6061 			continue;
6062 
6063 		emulated_msrs[num_emulated_msrs++] = emulated_msrs_all[i];
6064 	}
6065 
6066 	for (i = 0; i < ARRAY_SIZE(msr_based_features_all); i++) {
6067 		struct kvm_msr_entry msr;
6068 
6069 		msr.index = msr_based_features_all[i];
6070 		if (kvm_get_msr_feature(&msr))
6071 			continue;
6072 
6073 		msr_based_features[num_msr_based_features++] = msr_based_features_all[i];
6074 	}
6075 }
6076 
6077 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
6078 			   const void *v)
6079 {
6080 	int handled = 0;
6081 	int n;
6082 
6083 	do {
6084 		n = min(len, 8);
6085 		if (!(lapic_in_kernel(vcpu) &&
6086 		      !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
6087 		    && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
6088 			break;
6089 		handled += n;
6090 		addr += n;
6091 		len -= n;
6092 		v += n;
6093 	} while (len);
6094 
6095 	return handled;
6096 }
6097 
6098 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
6099 {
6100 	int handled = 0;
6101 	int n;
6102 
6103 	do {
6104 		n = min(len, 8);
6105 		if (!(lapic_in_kernel(vcpu) &&
6106 		      !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
6107 					 addr, n, v))
6108 		    && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
6109 			break;
6110 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
6111 		handled += n;
6112 		addr += n;
6113 		len -= n;
6114 		v += n;
6115 	} while (len);
6116 
6117 	return handled;
6118 }
6119 
6120 static void kvm_set_segment(struct kvm_vcpu *vcpu,
6121 			struct kvm_segment *var, int seg)
6122 {
6123 	static_call(kvm_x86_set_segment)(vcpu, var, seg);
6124 }
6125 
6126 void kvm_get_segment(struct kvm_vcpu *vcpu,
6127 		     struct kvm_segment *var, int seg)
6128 {
6129 	static_call(kvm_x86_get_segment)(vcpu, var, seg);
6130 }
6131 
6132 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
6133 			   struct x86_exception *exception)
6134 {
6135 	gpa_t t_gpa;
6136 
6137 	BUG_ON(!mmu_is_nested(vcpu));
6138 
6139 	/* NPT walks are always user-walks */
6140 	access |= PFERR_USER_MASK;
6141 	t_gpa  = vcpu->arch.mmu->gva_to_gpa(vcpu, gpa, access, exception);
6142 
6143 	return t_gpa;
6144 }
6145 
6146 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
6147 			      struct x86_exception *exception)
6148 {
6149 	u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6150 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6151 }
6152 EXPORT_SYMBOL_GPL(kvm_mmu_gva_to_gpa_read);
6153 
6154  gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
6155 				struct x86_exception *exception)
6156 {
6157 	u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6158 	access |= PFERR_FETCH_MASK;
6159 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6160 }
6161 
6162 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
6163 			       struct x86_exception *exception)
6164 {
6165 	u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6166 	access |= PFERR_WRITE_MASK;
6167 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6168 }
6169 EXPORT_SYMBOL_GPL(kvm_mmu_gva_to_gpa_write);
6170 
6171 /* uses this to access any guest's mapped memory without checking CPL */
6172 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
6173 				struct x86_exception *exception)
6174 {
6175 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
6176 }
6177 
6178 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
6179 				      struct kvm_vcpu *vcpu, u32 access,
6180 				      struct x86_exception *exception)
6181 {
6182 	void *data = val;
6183 	int r = X86EMUL_CONTINUE;
6184 
6185 	while (bytes) {
6186 		gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
6187 							    exception);
6188 		unsigned offset = addr & (PAGE_SIZE-1);
6189 		unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
6190 		int ret;
6191 
6192 		if (gpa == UNMAPPED_GVA)
6193 			return X86EMUL_PROPAGATE_FAULT;
6194 		ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
6195 					       offset, toread);
6196 		if (ret < 0) {
6197 			r = X86EMUL_IO_NEEDED;
6198 			goto out;
6199 		}
6200 
6201 		bytes -= toread;
6202 		data += toread;
6203 		addr += toread;
6204 	}
6205 out:
6206 	return r;
6207 }
6208 
6209 /* used for instruction fetching */
6210 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
6211 				gva_t addr, void *val, unsigned int bytes,
6212 				struct x86_exception *exception)
6213 {
6214 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6215 	u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6216 	unsigned offset;
6217 	int ret;
6218 
6219 	/* Inline kvm_read_guest_virt_helper for speed.  */
6220 	gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
6221 						    exception);
6222 	if (unlikely(gpa == UNMAPPED_GVA))
6223 		return X86EMUL_PROPAGATE_FAULT;
6224 
6225 	offset = addr & (PAGE_SIZE-1);
6226 	if (WARN_ON(offset + bytes > PAGE_SIZE))
6227 		bytes = (unsigned)PAGE_SIZE - offset;
6228 	ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
6229 				       offset, bytes);
6230 	if (unlikely(ret < 0))
6231 		return X86EMUL_IO_NEEDED;
6232 
6233 	return X86EMUL_CONTINUE;
6234 }
6235 
6236 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
6237 			       gva_t addr, void *val, unsigned int bytes,
6238 			       struct x86_exception *exception)
6239 {
6240 	u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6241 
6242 	/*
6243 	 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
6244 	 * is returned, but our callers are not ready for that and they blindly
6245 	 * call kvm_inject_page_fault.  Ensure that they at least do not leak
6246 	 * uninitialized kernel stack memory into cr2 and error code.
6247 	 */
6248 	memset(exception, 0, sizeof(*exception));
6249 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
6250 					  exception);
6251 }
6252 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
6253 
6254 static int emulator_read_std(struct x86_emulate_ctxt *ctxt,
6255 			     gva_t addr, void *val, unsigned int bytes,
6256 			     struct x86_exception *exception, bool system)
6257 {
6258 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6259 	u32 access = 0;
6260 
6261 	if (!system && static_call(kvm_x86_get_cpl)(vcpu) == 3)
6262 		access |= PFERR_USER_MASK;
6263 
6264 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, exception);
6265 }
6266 
6267 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
6268 		unsigned long addr, void *val, unsigned int bytes)
6269 {
6270 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6271 	int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
6272 
6273 	return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
6274 }
6275 
6276 static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
6277 				      struct kvm_vcpu *vcpu, u32 access,
6278 				      struct x86_exception *exception)
6279 {
6280 	void *data = val;
6281 	int r = X86EMUL_CONTINUE;
6282 
6283 	while (bytes) {
6284 		gpa_t gpa =  vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
6285 							     access,
6286 							     exception);
6287 		unsigned offset = addr & (PAGE_SIZE-1);
6288 		unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
6289 		int ret;
6290 
6291 		if (gpa == UNMAPPED_GVA)
6292 			return X86EMUL_PROPAGATE_FAULT;
6293 		ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
6294 		if (ret < 0) {
6295 			r = X86EMUL_IO_NEEDED;
6296 			goto out;
6297 		}
6298 
6299 		bytes -= towrite;
6300 		data += towrite;
6301 		addr += towrite;
6302 	}
6303 out:
6304 	return r;
6305 }
6306 
6307 static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val,
6308 			      unsigned int bytes, struct x86_exception *exception,
6309 			      bool system)
6310 {
6311 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6312 	u32 access = PFERR_WRITE_MASK;
6313 
6314 	if (!system && static_call(kvm_x86_get_cpl)(vcpu) == 3)
6315 		access |= PFERR_USER_MASK;
6316 
6317 	return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
6318 					   access, exception);
6319 }
6320 
6321 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val,
6322 				unsigned int bytes, struct x86_exception *exception)
6323 {
6324 	/* kvm_write_guest_virt_system can pull in tons of pages. */
6325 	vcpu->arch.l1tf_flush_l1d = true;
6326 
6327 	return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
6328 					   PFERR_WRITE_MASK, exception);
6329 }
6330 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
6331 
6332 int handle_ud(struct kvm_vcpu *vcpu)
6333 {
6334 	static const char kvm_emulate_prefix[] = { __KVM_EMULATE_PREFIX };
6335 	int emul_type = EMULTYPE_TRAP_UD;
6336 	char sig[5]; /* ud2; .ascii "kvm" */
6337 	struct x86_exception e;
6338 
6339 	if (unlikely(!static_call(kvm_x86_can_emulate_instruction)(vcpu, NULL, 0)))
6340 		return 1;
6341 
6342 	if (force_emulation_prefix &&
6343 	    kvm_read_guest_virt(vcpu, kvm_get_linear_rip(vcpu),
6344 				sig, sizeof(sig), &e) == 0 &&
6345 	    memcmp(sig, kvm_emulate_prefix, sizeof(sig)) == 0) {
6346 		kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig));
6347 		emul_type = EMULTYPE_TRAP_UD_FORCED;
6348 	}
6349 
6350 	return kvm_emulate_instruction(vcpu, emul_type);
6351 }
6352 EXPORT_SYMBOL_GPL(handle_ud);
6353 
6354 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
6355 			    gpa_t gpa, bool write)
6356 {
6357 	/* For APIC access vmexit */
6358 	if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
6359 		return 1;
6360 
6361 	if (vcpu_match_mmio_gpa(vcpu, gpa)) {
6362 		trace_vcpu_match_mmio(gva, gpa, write, true);
6363 		return 1;
6364 	}
6365 
6366 	return 0;
6367 }
6368 
6369 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
6370 				gpa_t *gpa, struct x86_exception *exception,
6371 				bool write)
6372 {
6373 	u32 access = ((static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0)
6374 		| (write ? PFERR_WRITE_MASK : 0);
6375 
6376 	/*
6377 	 * currently PKRU is only applied to ept enabled guest so
6378 	 * there is no pkey in EPT page table for L1 guest or EPT
6379 	 * shadow page table for L2 guest.
6380 	 */
6381 	if (vcpu_match_mmio_gva(vcpu, gva)
6382 	    && !permission_fault(vcpu, vcpu->arch.walk_mmu,
6383 				 vcpu->arch.mmio_access, 0, access)) {
6384 		*gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
6385 					(gva & (PAGE_SIZE - 1));
6386 		trace_vcpu_match_mmio(gva, *gpa, write, false);
6387 		return 1;
6388 	}
6389 
6390 	*gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6391 
6392 	if (*gpa == UNMAPPED_GVA)
6393 		return -1;
6394 
6395 	return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
6396 }
6397 
6398 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
6399 			const void *val, int bytes)
6400 {
6401 	int ret;
6402 
6403 	ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
6404 	if (ret < 0)
6405 		return 0;
6406 	kvm_page_track_write(vcpu, gpa, val, bytes);
6407 	return 1;
6408 }
6409 
6410 struct read_write_emulator_ops {
6411 	int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
6412 				  int bytes);
6413 	int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
6414 				  void *val, int bytes);
6415 	int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
6416 			       int bytes, void *val);
6417 	int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
6418 				    void *val, int bytes);
6419 	bool write;
6420 };
6421 
6422 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
6423 {
6424 	if (vcpu->mmio_read_completed) {
6425 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
6426 			       vcpu->mmio_fragments[0].gpa, val);
6427 		vcpu->mmio_read_completed = 0;
6428 		return 1;
6429 	}
6430 
6431 	return 0;
6432 }
6433 
6434 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
6435 			void *val, int bytes)
6436 {
6437 	return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
6438 }
6439 
6440 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
6441 			 void *val, int bytes)
6442 {
6443 	return emulator_write_phys(vcpu, gpa, val, bytes);
6444 }
6445 
6446 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
6447 {
6448 	trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
6449 	return vcpu_mmio_write(vcpu, gpa, bytes, val);
6450 }
6451 
6452 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
6453 			  void *val, int bytes)
6454 {
6455 	trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
6456 	return X86EMUL_IO_NEEDED;
6457 }
6458 
6459 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
6460 			   void *val, int bytes)
6461 {
6462 	struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
6463 
6464 	memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
6465 	return X86EMUL_CONTINUE;
6466 }
6467 
6468 static const struct read_write_emulator_ops read_emultor = {
6469 	.read_write_prepare = read_prepare,
6470 	.read_write_emulate = read_emulate,
6471 	.read_write_mmio = vcpu_mmio_read,
6472 	.read_write_exit_mmio = read_exit_mmio,
6473 };
6474 
6475 static const struct read_write_emulator_ops write_emultor = {
6476 	.read_write_emulate = write_emulate,
6477 	.read_write_mmio = write_mmio,
6478 	.read_write_exit_mmio = write_exit_mmio,
6479 	.write = true,
6480 };
6481 
6482 static int emulator_read_write_onepage(unsigned long addr, void *val,
6483 				       unsigned int bytes,
6484 				       struct x86_exception *exception,
6485 				       struct kvm_vcpu *vcpu,
6486 				       const struct read_write_emulator_ops *ops)
6487 {
6488 	gpa_t gpa;
6489 	int handled, ret;
6490 	bool write = ops->write;
6491 	struct kvm_mmio_fragment *frag;
6492 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6493 
6494 	/*
6495 	 * If the exit was due to a NPF we may already have a GPA.
6496 	 * If the GPA is present, use it to avoid the GVA to GPA table walk.
6497 	 * Note, this cannot be used on string operations since string
6498 	 * operation using rep will only have the initial GPA from the NPF
6499 	 * occurred.
6500 	 */
6501 	if (ctxt->gpa_available && emulator_can_use_gpa(ctxt) &&
6502 	    (addr & ~PAGE_MASK) == (ctxt->gpa_val & ~PAGE_MASK)) {
6503 		gpa = ctxt->gpa_val;
6504 		ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
6505 	} else {
6506 		ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
6507 		if (ret < 0)
6508 			return X86EMUL_PROPAGATE_FAULT;
6509 	}
6510 
6511 	if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
6512 		return X86EMUL_CONTINUE;
6513 
6514 	/*
6515 	 * Is this MMIO handled locally?
6516 	 */
6517 	handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
6518 	if (handled == bytes)
6519 		return X86EMUL_CONTINUE;
6520 
6521 	gpa += handled;
6522 	bytes -= handled;
6523 	val += handled;
6524 
6525 	WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
6526 	frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
6527 	frag->gpa = gpa;
6528 	frag->data = val;
6529 	frag->len = bytes;
6530 	return X86EMUL_CONTINUE;
6531 }
6532 
6533 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
6534 			unsigned long addr,
6535 			void *val, unsigned int bytes,
6536 			struct x86_exception *exception,
6537 			const struct read_write_emulator_ops *ops)
6538 {
6539 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6540 	gpa_t gpa;
6541 	int rc;
6542 
6543 	if (ops->read_write_prepare &&
6544 		  ops->read_write_prepare(vcpu, val, bytes))
6545 		return X86EMUL_CONTINUE;
6546 
6547 	vcpu->mmio_nr_fragments = 0;
6548 
6549 	/* Crossing a page boundary? */
6550 	if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
6551 		int now;
6552 
6553 		now = -addr & ~PAGE_MASK;
6554 		rc = emulator_read_write_onepage(addr, val, now, exception,
6555 						 vcpu, ops);
6556 
6557 		if (rc != X86EMUL_CONTINUE)
6558 			return rc;
6559 		addr += now;
6560 		if (ctxt->mode != X86EMUL_MODE_PROT64)
6561 			addr = (u32)addr;
6562 		val += now;
6563 		bytes -= now;
6564 	}
6565 
6566 	rc = emulator_read_write_onepage(addr, val, bytes, exception,
6567 					 vcpu, ops);
6568 	if (rc != X86EMUL_CONTINUE)
6569 		return rc;
6570 
6571 	if (!vcpu->mmio_nr_fragments)
6572 		return rc;
6573 
6574 	gpa = vcpu->mmio_fragments[0].gpa;
6575 
6576 	vcpu->mmio_needed = 1;
6577 	vcpu->mmio_cur_fragment = 0;
6578 
6579 	vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
6580 	vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
6581 	vcpu->run->exit_reason = KVM_EXIT_MMIO;
6582 	vcpu->run->mmio.phys_addr = gpa;
6583 
6584 	return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
6585 }
6586 
6587 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
6588 				  unsigned long addr,
6589 				  void *val,
6590 				  unsigned int bytes,
6591 				  struct x86_exception *exception)
6592 {
6593 	return emulator_read_write(ctxt, addr, val, bytes,
6594 				   exception, &read_emultor);
6595 }
6596 
6597 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
6598 			    unsigned long addr,
6599 			    const void *val,
6600 			    unsigned int bytes,
6601 			    struct x86_exception *exception)
6602 {
6603 	return emulator_read_write(ctxt, addr, (void *)val, bytes,
6604 				   exception, &write_emultor);
6605 }
6606 
6607 #define CMPXCHG_TYPE(t, ptr, old, new) \
6608 	(cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
6609 
6610 #ifdef CONFIG_X86_64
6611 #  define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
6612 #else
6613 #  define CMPXCHG64(ptr, old, new) \
6614 	(cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
6615 #endif
6616 
6617 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
6618 				     unsigned long addr,
6619 				     const void *old,
6620 				     const void *new,
6621 				     unsigned int bytes,
6622 				     struct x86_exception *exception)
6623 {
6624 	struct kvm_host_map map;
6625 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6626 	u64 page_line_mask;
6627 	gpa_t gpa;
6628 	char *kaddr;
6629 	bool exchanged;
6630 
6631 	/* guests cmpxchg8b have to be emulated atomically */
6632 	if (bytes > 8 || (bytes & (bytes - 1)))
6633 		goto emul_write;
6634 
6635 	gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
6636 
6637 	if (gpa == UNMAPPED_GVA ||
6638 	    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
6639 		goto emul_write;
6640 
6641 	/*
6642 	 * Emulate the atomic as a straight write to avoid #AC if SLD is
6643 	 * enabled in the host and the access splits a cache line.
6644 	 */
6645 	if (boot_cpu_has(X86_FEATURE_SPLIT_LOCK_DETECT))
6646 		page_line_mask = ~(cache_line_size() - 1);
6647 	else
6648 		page_line_mask = PAGE_MASK;
6649 
6650 	if (((gpa + bytes - 1) & page_line_mask) != (gpa & page_line_mask))
6651 		goto emul_write;
6652 
6653 	if (kvm_vcpu_map(vcpu, gpa_to_gfn(gpa), &map))
6654 		goto emul_write;
6655 
6656 	kaddr = map.hva + offset_in_page(gpa);
6657 
6658 	switch (bytes) {
6659 	case 1:
6660 		exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
6661 		break;
6662 	case 2:
6663 		exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
6664 		break;
6665 	case 4:
6666 		exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
6667 		break;
6668 	case 8:
6669 		exchanged = CMPXCHG64(kaddr, old, new);
6670 		break;
6671 	default:
6672 		BUG();
6673 	}
6674 
6675 	kvm_vcpu_unmap(vcpu, &map, true);
6676 
6677 	if (!exchanged)
6678 		return X86EMUL_CMPXCHG_FAILED;
6679 
6680 	kvm_page_track_write(vcpu, gpa, new, bytes);
6681 
6682 	return X86EMUL_CONTINUE;
6683 
6684 emul_write:
6685 	printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
6686 
6687 	return emulator_write_emulated(ctxt, addr, new, bytes, exception);
6688 }
6689 
6690 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
6691 {
6692 	int r = 0, i;
6693 
6694 	for (i = 0; i < vcpu->arch.pio.count; i++) {
6695 		if (vcpu->arch.pio.in)
6696 			r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
6697 					    vcpu->arch.pio.size, pd);
6698 		else
6699 			r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
6700 					     vcpu->arch.pio.port, vcpu->arch.pio.size,
6701 					     pd);
6702 		if (r)
6703 			break;
6704 		pd += vcpu->arch.pio.size;
6705 	}
6706 	return r;
6707 }
6708 
6709 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
6710 			       unsigned short port, void *val,
6711 			       unsigned int count, bool in)
6712 {
6713 	vcpu->arch.pio.port = port;
6714 	vcpu->arch.pio.in = in;
6715 	vcpu->arch.pio.count  = count;
6716 	vcpu->arch.pio.size = size;
6717 
6718 	if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
6719 		vcpu->arch.pio.count = 0;
6720 		return 1;
6721 	}
6722 
6723 	vcpu->run->exit_reason = KVM_EXIT_IO;
6724 	vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
6725 	vcpu->run->io.size = size;
6726 	vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
6727 	vcpu->run->io.count = count;
6728 	vcpu->run->io.port = port;
6729 
6730 	return 0;
6731 }
6732 
6733 static int emulator_pio_in(struct kvm_vcpu *vcpu, int size,
6734 			   unsigned short port, void *val, unsigned int count)
6735 {
6736 	int ret;
6737 
6738 	if (vcpu->arch.pio.count)
6739 		goto data_avail;
6740 
6741 	memset(vcpu->arch.pio_data, 0, size * count);
6742 
6743 	ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
6744 	if (ret) {
6745 data_avail:
6746 		memcpy(val, vcpu->arch.pio_data, size * count);
6747 		trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
6748 		vcpu->arch.pio.count = 0;
6749 		return 1;
6750 	}
6751 
6752 	return 0;
6753 }
6754 
6755 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
6756 				    int size, unsigned short port, void *val,
6757 				    unsigned int count)
6758 {
6759 	return emulator_pio_in(emul_to_vcpu(ctxt), size, port, val, count);
6760 
6761 }
6762 
6763 static int emulator_pio_out(struct kvm_vcpu *vcpu, int size,
6764 			    unsigned short port, const void *val,
6765 			    unsigned int count)
6766 {
6767 	memcpy(vcpu->arch.pio_data, val, size * count);
6768 	trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
6769 	return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
6770 }
6771 
6772 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
6773 				     int size, unsigned short port,
6774 				     const void *val, unsigned int count)
6775 {
6776 	return emulator_pio_out(emul_to_vcpu(ctxt), size, port, val, count);
6777 }
6778 
6779 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
6780 {
6781 	return static_call(kvm_x86_get_segment_base)(vcpu, seg);
6782 }
6783 
6784 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
6785 {
6786 	kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
6787 }
6788 
6789 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
6790 {
6791 	if (!need_emulate_wbinvd(vcpu))
6792 		return X86EMUL_CONTINUE;
6793 
6794 	if (static_call(kvm_x86_has_wbinvd_exit)()) {
6795 		int cpu = get_cpu();
6796 
6797 		cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
6798 		on_each_cpu_mask(vcpu->arch.wbinvd_dirty_mask,
6799 				wbinvd_ipi, NULL, 1);
6800 		put_cpu();
6801 		cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
6802 	} else
6803 		wbinvd();
6804 	return X86EMUL_CONTINUE;
6805 }
6806 
6807 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
6808 {
6809 	kvm_emulate_wbinvd_noskip(vcpu);
6810 	return kvm_skip_emulated_instruction(vcpu);
6811 }
6812 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
6813 
6814 
6815 
6816 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
6817 {
6818 	kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
6819 }
6820 
6821 static void emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
6822 			    unsigned long *dest)
6823 {
6824 	kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
6825 }
6826 
6827 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
6828 			   unsigned long value)
6829 {
6830 
6831 	return kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
6832 }
6833 
6834 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
6835 {
6836 	return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
6837 }
6838 
6839 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
6840 {
6841 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6842 	unsigned long value;
6843 
6844 	switch (cr) {
6845 	case 0:
6846 		value = kvm_read_cr0(vcpu);
6847 		break;
6848 	case 2:
6849 		value = vcpu->arch.cr2;
6850 		break;
6851 	case 3:
6852 		value = kvm_read_cr3(vcpu);
6853 		break;
6854 	case 4:
6855 		value = kvm_read_cr4(vcpu);
6856 		break;
6857 	case 8:
6858 		value = kvm_get_cr8(vcpu);
6859 		break;
6860 	default:
6861 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
6862 		return 0;
6863 	}
6864 
6865 	return value;
6866 }
6867 
6868 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
6869 {
6870 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6871 	int res = 0;
6872 
6873 	switch (cr) {
6874 	case 0:
6875 		res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
6876 		break;
6877 	case 2:
6878 		vcpu->arch.cr2 = val;
6879 		break;
6880 	case 3:
6881 		res = kvm_set_cr3(vcpu, val);
6882 		break;
6883 	case 4:
6884 		res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
6885 		break;
6886 	case 8:
6887 		res = kvm_set_cr8(vcpu, val);
6888 		break;
6889 	default:
6890 		kvm_err("%s: unexpected cr %u\n", __func__, cr);
6891 		res = -1;
6892 	}
6893 
6894 	return res;
6895 }
6896 
6897 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
6898 {
6899 	return static_call(kvm_x86_get_cpl)(emul_to_vcpu(ctxt));
6900 }
6901 
6902 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6903 {
6904 	static_call(kvm_x86_get_gdt)(emul_to_vcpu(ctxt), dt);
6905 }
6906 
6907 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6908 {
6909 	static_call(kvm_x86_get_idt)(emul_to_vcpu(ctxt), dt);
6910 }
6911 
6912 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6913 {
6914 	static_call(kvm_x86_set_gdt)(emul_to_vcpu(ctxt), dt);
6915 }
6916 
6917 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6918 {
6919 	static_call(kvm_x86_set_idt)(emul_to_vcpu(ctxt), dt);
6920 }
6921 
6922 static unsigned long emulator_get_cached_segment_base(
6923 	struct x86_emulate_ctxt *ctxt, int seg)
6924 {
6925 	return get_segment_base(emul_to_vcpu(ctxt), seg);
6926 }
6927 
6928 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
6929 				 struct desc_struct *desc, u32 *base3,
6930 				 int seg)
6931 {
6932 	struct kvm_segment var;
6933 
6934 	kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
6935 	*selector = var.selector;
6936 
6937 	if (var.unusable) {
6938 		memset(desc, 0, sizeof(*desc));
6939 		if (base3)
6940 			*base3 = 0;
6941 		return false;
6942 	}
6943 
6944 	if (var.g)
6945 		var.limit >>= 12;
6946 	set_desc_limit(desc, var.limit);
6947 	set_desc_base(desc, (unsigned long)var.base);
6948 #ifdef CONFIG_X86_64
6949 	if (base3)
6950 		*base3 = var.base >> 32;
6951 #endif
6952 	desc->type = var.type;
6953 	desc->s = var.s;
6954 	desc->dpl = var.dpl;
6955 	desc->p = var.present;
6956 	desc->avl = var.avl;
6957 	desc->l = var.l;
6958 	desc->d = var.db;
6959 	desc->g = var.g;
6960 
6961 	return true;
6962 }
6963 
6964 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
6965 				 struct desc_struct *desc, u32 base3,
6966 				 int seg)
6967 {
6968 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6969 	struct kvm_segment var;
6970 
6971 	var.selector = selector;
6972 	var.base = get_desc_base(desc);
6973 #ifdef CONFIG_X86_64
6974 	var.base |= ((u64)base3) << 32;
6975 #endif
6976 	var.limit = get_desc_limit(desc);
6977 	if (desc->g)
6978 		var.limit = (var.limit << 12) | 0xfff;
6979 	var.type = desc->type;
6980 	var.dpl = desc->dpl;
6981 	var.db = desc->d;
6982 	var.s = desc->s;
6983 	var.l = desc->l;
6984 	var.g = desc->g;
6985 	var.avl = desc->avl;
6986 	var.present = desc->p;
6987 	var.unusable = !var.present;
6988 	var.padding = 0;
6989 
6990 	kvm_set_segment(vcpu, &var, seg);
6991 	return;
6992 }
6993 
6994 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
6995 			    u32 msr_index, u64 *pdata)
6996 {
6997 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6998 	int r;
6999 
7000 	r = kvm_get_msr(vcpu, msr_index, pdata);
7001 
7002 	if (r && kvm_get_msr_user_space(vcpu, msr_index, r)) {
7003 		/* Bounce to user space */
7004 		return X86EMUL_IO_NEEDED;
7005 	}
7006 
7007 	return r;
7008 }
7009 
7010 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
7011 			    u32 msr_index, u64 data)
7012 {
7013 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7014 	int r;
7015 
7016 	r = kvm_set_msr(vcpu, msr_index, data);
7017 
7018 	if (r && kvm_set_msr_user_space(vcpu, msr_index, data, r)) {
7019 		/* Bounce to user space */
7020 		return X86EMUL_IO_NEEDED;
7021 	}
7022 
7023 	return r;
7024 }
7025 
7026 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
7027 {
7028 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7029 
7030 	return vcpu->arch.smbase;
7031 }
7032 
7033 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
7034 {
7035 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7036 
7037 	vcpu->arch.smbase = smbase;
7038 }
7039 
7040 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
7041 			      u32 pmc)
7042 {
7043 	return kvm_pmu_is_valid_rdpmc_ecx(emul_to_vcpu(ctxt), pmc);
7044 }
7045 
7046 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
7047 			     u32 pmc, u64 *pdata)
7048 {
7049 	return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
7050 }
7051 
7052 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
7053 {
7054 	emul_to_vcpu(ctxt)->arch.halt_request = 1;
7055 }
7056 
7057 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
7058 			      struct x86_instruction_info *info,
7059 			      enum x86_intercept_stage stage)
7060 {
7061 	return static_call(kvm_x86_check_intercept)(emul_to_vcpu(ctxt), info, stage,
7062 					    &ctxt->exception);
7063 }
7064 
7065 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
7066 			      u32 *eax, u32 *ebx, u32 *ecx, u32 *edx,
7067 			      bool exact_only)
7068 {
7069 	return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, exact_only);
7070 }
7071 
7072 static bool emulator_guest_has_long_mode(struct x86_emulate_ctxt *ctxt)
7073 {
7074 	return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_LM);
7075 }
7076 
7077 static bool emulator_guest_has_movbe(struct x86_emulate_ctxt *ctxt)
7078 {
7079 	return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_MOVBE);
7080 }
7081 
7082 static bool emulator_guest_has_fxsr(struct x86_emulate_ctxt *ctxt)
7083 {
7084 	return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_FXSR);
7085 }
7086 
7087 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
7088 {
7089 	return kvm_register_read_raw(emul_to_vcpu(ctxt), reg);
7090 }
7091 
7092 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
7093 {
7094 	kvm_register_write_raw(emul_to_vcpu(ctxt), reg, val);
7095 }
7096 
7097 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
7098 {
7099 	static_call(kvm_x86_set_nmi_mask)(emul_to_vcpu(ctxt), masked);
7100 }
7101 
7102 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
7103 {
7104 	return emul_to_vcpu(ctxt)->arch.hflags;
7105 }
7106 
7107 static void emulator_set_hflags(struct x86_emulate_ctxt *ctxt, unsigned emul_flags)
7108 {
7109 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7110 
7111 	vcpu->arch.hflags = emul_flags;
7112 	kvm_mmu_reset_context(vcpu);
7113 }
7114 
7115 static int emulator_pre_leave_smm(struct x86_emulate_ctxt *ctxt,
7116 				  const char *smstate)
7117 {
7118 	return static_call(kvm_x86_pre_leave_smm)(emul_to_vcpu(ctxt), smstate);
7119 }
7120 
7121 static void emulator_post_leave_smm(struct x86_emulate_ctxt *ctxt)
7122 {
7123 	kvm_smm_changed(emul_to_vcpu(ctxt));
7124 }
7125 
7126 static int emulator_set_xcr(struct x86_emulate_ctxt *ctxt, u32 index, u64 xcr)
7127 {
7128 	return __kvm_set_xcr(emul_to_vcpu(ctxt), index, xcr);
7129 }
7130 
7131 static const struct x86_emulate_ops emulate_ops = {
7132 	.read_gpr            = emulator_read_gpr,
7133 	.write_gpr           = emulator_write_gpr,
7134 	.read_std            = emulator_read_std,
7135 	.write_std           = emulator_write_std,
7136 	.read_phys           = kvm_read_guest_phys_system,
7137 	.fetch               = kvm_fetch_guest_virt,
7138 	.read_emulated       = emulator_read_emulated,
7139 	.write_emulated      = emulator_write_emulated,
7140 	.cmpxchg_emulated    = emulator_cmpxchg_emulated,
7141 	.invlpg              = emulator_invlpg,
7142 	.pio_in_emulated     = emulator_pio_in_emulated,
7143 	.pio_out_emulated    = emulator_pio_out_emulated,
7144 	.get_segment         = emulator_get_segment,
7145 	.set_segment         = emulator_set_segment,
7146 	.get_cached_segment_base = emulator_get_cached_segment_base,
7147 	.get_gdt             = emulator_get_gdt,
7148 	.get_idt	     = emulator_get_idt,
7149 	.set_gdt             = emulator_set_gdt,
7150 	.set_idt	     = emulator_set_idt,
7151 	.get_cr              = emulator_get_cr,
7152 	.set_cr              = emulator_set_cr,
7153 	.cpl                 = emulator_get_cpl,
7154 	.get_dr              = emulator_get_dr,
7155 	.set_dr              = emulator_set_dr,
7156 	.get_smbase          = emulator_get_smbase,
7157 	.set_smbase          = emulator_set_smbase,
7158 	.set_msr             = emulator_set_msr,
7159 	.get_msr             = emulator_get_msr,
7160 	.check_pmc	     = emulator_check_pmc,
7161 	.read_pmc            = emulator_read_pmc,
7162 	.halt                = emulator_halt,
7163 	.wbinvd              = emulator_wbinvd,
7164 	.fix_hypercall       = emulator_fix_hypercall,
7165 	.intercept           = emulator_intercept,
7166 	.get_cpuid           = emulator_get_cpuid,
7167 	.guest_has_long_mode = emulator_guest_has_long_mode,
7168 	.guest_has_movbe     = emulator_guest_has_movbe,
7169 	.guest_has_fxsr      = emulator_guest_has_fxsr,
7170 	.set_nmi_mask        = emulator_set_nmi_mask,
7171 	.get_hflags          = emulator_get_hflags,
7172 	.set_hflags          = emulator_set_hflags,
7173 	.pre_leave_smm       = emulator_pre_leave_smm,
7174 	.post_leave_smm      = emulator_post_leave_smm,
7175 	.set_xcr             = emulator_set_xcr,
7176 };
7177 
7178 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
7179 {
7180 	u32 int_shadow = static_call(kvm_x86_get_interrupt_shadow)(vcpu);
7181 	/*
7182 	 * an sti; sti; sequence only disable interrupts for the first
7183 	 * instruction. So, if the last instruction, be it emulated or
7184 	 * not, left the system with the INT_STI flag enabled, it
7185 	 * means that the last instruction is an sti. We should not
7186 	 * leave the flag on in this case. The same goes for mov ss
7187 	 */
7188 	if (int_shadow & mask)
7189 		mask = 0;
7190 	if (unlikely(int_shadow || mask)) {
7191 		static_call(kvm_x86_set_interrupt_shadow)(vcpu, mask);
7192 		if (!mask)
7193 			kvm_make_request(KVM_REQ_EVENT, vcpu);
7194 	}
7195 }
7196 
7197 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
7198 {
7199 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7200 	if (ctxt->exception.vector == PF_VECTOR)
7201 		return kvm_inject_emulated_page_fault(vcpu, &ctxt->exception);
7202 
7203 	if (ctxt->exception.error_code_valid)
7204 		kvm_queue_exception_e(vcpu, ctxt->exception.vector,
7205 				      ctxt->exception.error_code);
7206 	else
7207 		kvm_queue_exception(vcpu, ctxt->exception.vector);
7208 	return false;
7209 }
7210 
7211 static struct x86_emulate_ctxt *alloc_emulate_ctxt(struct kvm_vcpu *vcpu)
7212 {
7213 	struct x86_emulate_ctxt *ctxt;
7214 
7215 	ctxt = kmem_cache_zalloc(x86_emulator_cache, GFP_KERNEL_ACCOUNT);
7216 	if (!ctxt) {
7217 		pr_err("kvm: failed to allocate vcpu's emulator\n");
7218 		return NULL;
7219 	}
7220 
7221 	ctxt->vcpu = vcpu;
7222 	ctxt->ops = &emulate_ops;
7223 	vcpu->arch.emulate_ctxt = ctxt;
7224 
7225 	return ctxt;
7226 }
7227 
7228 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
7229 {
7230 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7231 	int cs_db, cs_l;
7232 
7233 	static_call(kvm_x86_get_cs_db_l_bits)(vcpu, &cs_db, &cs_l);
7234 
7235 	ctxt->gpa_available = false;
7236 	ctxt->eflags = kvm_get_rflags(vcpu);
7237 	ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
7238 
7239 	ctxt->eip = kvm_rip_read(vcpu);
7240 	ctxt->mode = (!is_protmode(vcpu))		? X86EMUL_MODE_REAL :
7241 		     (ctxt->eflags & X86_EFLAGS_VM)	? X86EMUL_MODE_VM86 :
7242 		     (cs_l && is_long_mode(vcpu))	? X86EMUL_MODE_PROT64 :
7243 		     cs_db				? X86EMUL_MODE_PROT32 :
7244 							  X86EMUL_MODE_PROT16;
7245 	BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
7246 	BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
7247 	BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
7248 
7249 	ctxt->interruptibility = 0;
7250 	ctxt->have_exception = false;
7251 	ctxt->exception.vector = -1;
7252 	ctxt->perm_ok = false;
7253 
7254 	init_decode_cache(ctxt);
7255 	vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
7256 }
7257 
7258 void kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
7259 {
7260 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7261 	int ret;
7262 
7263 	init_emulate_ctxt(vcpu);
7264 
7265 	ctxt->op_bytes = 2;
7266 	ctxt->ad_bytes = 2;
7267 	ctxt->_eip = ctxt->eip + inc_eip;
7268 	ret = emulate_int_real(ctxt, irq);
7269 
7270 	if (ret != X86EMUL_CONTINUE) {
7271 		kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
7272 	} else {
7273 		ctxt->eip = ctxt->_eip;
7274 		kvm_rip_write(vcpu, ctxt->eip);
7275 		kvm_set_rflags(vcpu, ctxt->eflags);
7276 	}
7277 }
7278 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
7279 
7280 static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type)
7281 {
7282 	++vcpu->stat.insn_emulation_fail;
7283 	trace_kvm_emulate_insn_failed(vcpu);
7284 
7285 	if (emulation_type & EMULTYPE_VMWARE_GP) {
7286 		kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
7287 		return 1;
7288 	}
7289 
7290 	if (emulation_type & EMULTYPE_SKIP) {
7291 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
7292 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
7293 		vcpu->run->internal.ndata = 0;
7294 		return 0;
7295 	}
7296 
7297 	kvm_queue_exception(vcpu, UD_VECTOR);
7298 
7299 	if (!is_guest_mode(vcpu) && static_call(kvm_x86_get_cpl)(vcpu) == 0) {
7300 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
7301 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
7302 		vcpu->run->internal.ndata = 0;
7303 		return 0;
7304 	}
7305 
7306 	return 1;
7307 }
7308 
7309 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
7310 				  bool write_fault_to_shadow_pgtable,
7311 				  int emulation_type)
7312 {
7313 	gpa_t gpa = cr2_or_gpa;
7314 	kvm_pfn_t pfn;
7315 
7316 	if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
7317 		return false;
7318 
7319 	if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
7320 	    WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
7321 		return false;
7322 
7323 	if (!vcpu->arch.mmu->direct_map) {
7324 		/*
7325 		 * Write permission should be allowed since only
7326 		 * write access need to be emulated.
7327 		 */
7328 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
7329 
7330 		/*
7331 		 * If the mapping is invalid in guest, let cpu retry
7332 		 * it to generate fault.
7333 		 */
7334 		if (gpa == UNMAPPED_GVA)
7335 			return true;
7336 	}
7337 
7338 	/*
7339 	 * Do not retry the unhandleable instruction if it faults on the
7340 	 * readonly host memory, otherwise it will goto a infinite loop:
7341 	 * retry instruction -> write #PF -> emulation fail -> retry
7342 	 * instruction -> ...
7343 	 */
7344 	pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
7345 
7346 	/*
7347 	 * If the instruction failed on the error pfn, it can not be fixed,
7348 	 * report the error to userspace.
7349 	 */
7350 	if (is_error_noslot_pfn(pfn))
7351 		return false;
7352 
7353 	kvm_release_pfn_clean(pfn);
7354 
7355 	/* The instructions are well-emulated on direct mmu. */
7356 	if (vcpu->arch.mmu->direct_map) {
7357 		unsigned int indirect_shadow_pages;
7358 
7359 		write_lock(&vcpu->kvm->mmu_lock);
7360 		indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
7361 		write_unlock(&vcpu->kvm->mmu_lock);
7362 
7363 		if (indirect_shadow_pages)
7364 			kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7365 
7366 		return true;
7367 	}
7368 
7369 	/*
7370 	 * if emulation was due to access to shadowed page table
7371 	 * and it failed try to unshadow page and re-enter the
7372 	 * guest to let CPU execute the instruction.
7373 	 */
7374 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7375 
7376 	/*
7377 	 * If the access faults on its page table, it can not
7378 	 * be fixed by unprotecting shadow page and it should
7379 	 * be reported to userspace.
7380 	 */
7381 	return !write_fault_to_shadow_pgtable;
7382 }
7383 
7384 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
7385 			      gpa_t cr2_or_gpa,  int emulation_type)
7386 {
7387 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7388 	unsigned long last_retry_eip, last_retry_addr, gpa = cr2_or_gpa;
7389 
7390 	last_retry_eip = vcpu->arch.last_retry_eip;
7391 	last_retry_addr = vcpu->arch.last_retry_addr;
7392 
7393 	/*
7394 	 * If the emulation is caused by #PF and it is non-page_table
7395 	 * writing instruction, it means the VM-EXIT is caused by shadow
7396 	 * page protected, we can zap the shadow page and retry this
7397 	 * instruction directly.
7398 	 *
7399 	 * Note: if the guest uses a non-page-table modifying instruction
7400 	 * on the PDE that points to the instruction, then we will unmap
7401 	 * the instruction and go to an infinite loop. So, we cache the
7402 	 * last retried eip and the last fault address, if we meet the eip
7403 	 * and the address again, we can break out of the potential infinite
7404 	 * loop.
7405 	 */
7406 	vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
7407 
7408 	if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
7409 		return false;
7410 
7411 	if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
7412 	    WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
7413 		return false;
7414 
7415 	if (x86_page_table_writing_insn(ctxt))
7416 		return false;
7417 
7418 	if (ctxt->eip == last_retry_eip && last_retry_addr == cr2_or_gpa)
7419 		return false;
7420 
7421 	vcpu->arch.last_retry_eip = ctxt->eip;
7422 	vcpu->arch.last_retry_addr = cr2_or_gpa;
7423 
7424 	if (!vcpu->arch.mmu->direct_map)
7425 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
7426 
7427 	kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7428 
7429 	return true;
7430 }
7431 
7432 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
7433 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
7434 
7435 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
7436 {
7437 	if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
7438 		/* This is a good place to trace that we are exiting SMM.  */
7439 		trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
7440 
7441 		/* Process a latched INIT or SMI, if any.  */
7442 		kvm_make_request(KVM_REQ_EVENT, vcpu);
7443 	}
7444 
7445 	kvm_mmu_reset_context(vcpu);
7446 }
7447 
7448 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
7449 				unsigned long *db)
7450 {
7451 	u32 dr6 = 0;
7452 	int i;
7453 	u32 enable, rwlen;
7454 
7455 	enable = dr7;
7456 	rwlen = dr7 >> 16;
7457 	for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
7458 		if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
7459 			dr6 |= (1 << i);
7460 	return dr6;
7461 }
7462 
7463 static int kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu)
7464 {
7465 	struct kvm_run *kvm_run = vcpu->run;
7466 
7467 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
7468 		kvm_run->debug.arch.dr6 = DR6_BS | DR6_ACTIVE_LOW;
7469 		kvm_run->debug.arch.pc = kvm_get_linear_rip(vcpu);
7470 		kvm_run->debug.arch.exception = DB_VECTOR;
7471 		kvm_run->exit_reason = KVM_EXIT_DEBUG;
7472 		return 0;
7473 	}
7474 	kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BS);
7475 	return 1;
7476 }
7477 
7478 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
7479 {
7480 	unsigned long rflags = static_call(kvm_x86_get_rflags)(vcpu);
7481 	int r;
7482 
7483 	r = static_call(kvm_x86_skip_emulated_instruction)(vcpu);
7484 	if (unlikely(!r))
7485 		return 0;
7486 
7487 	/*
7488 	 * rflags is the old, "raw" value of the flags.  The new value has
7489 	 * not been saved yet.
7490 	 *
7491 	 * This is correct even for TF set by the guest, because "the
7492 	 * processor will not generate this exception after the instruction
7493 	 * that sets the TF flag".
7494 	 */
7495 	if (unlikely(rflags & X86_EFLAGS_TF))
7496 		r = kvm_vcpu_do_singlestep(vcpu);
7497 	return r;
7498 }
7499 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
7500 
7501 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
7502 {
7503 	if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
7504 	    (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
7505 		struct kvm_run *kvm_run = vcpu->run;
7506 		unsigned long eip = kvm_get_linear_rip(vcpu);
7507 		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
7508 					   vcpu->arch.guest_debug_dr7,
7509 					   vcpu->arch.eff_db);
7510 
7511 		if (dr6 != 0) {
7512 			kvm_run->debug.arch.dr6 = dr6 | DR6_ACTIVE_LOW;
7513 			kvm_run->debug.arch.pc = eip;
7514 			kvm_run->debug.arch.exception = DB_VECTOR;
7515 			kvm_run->exit_reason = KVM_EXIT_DEBUG;
7516 			*r = 0;
7517 			return true;
7518 		}
7519 	}
7520 
7521 	if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
7522 	    !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
7523 		unsigned long eip = kvm_get_linear_rip(vcpu);
7524 		u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
7525 					   vcpu->arch.dr7,
7526 					   vcpu->arch.db);
7527 
7528 		if (dr6 != 0) {
7529 			kvm_queue_exception_p(vcpu, DB_VECTOR, dr6);
7530 			*r = 1;
7531 			return true;
7532 		}
7533 	}
7534 
7535 	return false;
7536 }
7537 
7538 static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
7539 {
7540 	switch (ctxt->opcode_len) {
7541 	case 1:
7542 		switch (ctxt->b) {
7543 		case 0xe4:	/* IN */
7544 		case 0xe5:
7545 		case 0xec:
7546 		case 0xed:
7547 		case 0xe6:	/* OUT */
7548 		case 0xe7:
7549 		case 0xee:
7550 		case 0xef:
7551 		case 0x6c:	/* INS */
7552 		case 0x6d:
7553 		case 0x6e:	/* OUTS */
7554 		case 0x6f:
7555 			return true;
7556 		}
7557 		break;
7558 	case 2:
7559 		switch (ctxt->b) {
7560 		case 0x33:	/* RDPMC */
7561 			return true;
7562 		}
7563 		break;
7564 	}
7565 
7566 	return false;
7567 }
7568 
7569 /*
7570  * Decode to be emulated instruction. Return EMULATION_OK if success.
7571  */
7572 int x86_decode_emulated_instruction(struct kvm_vcpu *vcpu, int emulation_type,
7573 				    void *insn, int insn_len)
7574 {
7575 	int r = EMULATION_OK;
7576 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7577 
7578 	init_emulate_ctxt(vcpu);
7579 
7580 	/*
7581 	 * We will reenter on the same instruction since we do not set
7582 	 * complete_userspace_io. This does not handle watchpoints yet,
7583 	 * those would be handled in the emulate_ops.
7584 	 */
7585 	if (!(emulation_type & EMULTYPE_SKIP) &&
7586 	    kvm_vcpu_check_breakpoint(vcpu, &r))
7587 		return r;
7588 
7589 	r = x86_decode_insn(ctxt, insn, insn_len, emulation_type);
7590 
7591 	trace_kvm_emulate_insn_start(vcpu);
7592 	++vcpu->stat.insn_emulation;
7593 
7594 	return r;
7595 }
7596 EXPORT_SYMBOL_GPL(x86_decode_emulated_instruction);
7597 
7598 int x86_emulate_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
7599 			    int emulation_type, void *insn, int insn_len)
7600 {
7601 	int r;
7602 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7603 	bool writeback = true;
7604 	bool write_fault_to_spt;
7605 
7606 	if (unlikely(!static_call(kvm_x86_can_emulate_instruction)(vcpu, insn, insn_len)))
7607 		return 1;
7608 
7609 	vcpu->arch.l1tf_flush_l1d = true;
7610 
7611 	/*
7612 	 * Clear write_fault_to_shadow_pgtable here to ensure it is
7613 	 * never reused.
7614 	 */
7615 	write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
7616 	vcpu->arch.write_fault_to_shadow_pgtable = false;
7617 
7618 	if (!(emulation_type & EMULTYPE_NO_DECODE)) {
7619 		kvm_clear_exception_queue(vcpu);
7620 
7621 		r = x86_decode_emulated_instruction(vcpu, emulation_type,
7622 						    insn, insn_len);
7623 		if (r != EMULATION_OK)  {
7624 			if ((emulation_type & EMULTYPE_TRAP_UD) ||
7625 			    (emulation_type & EMULTYPE_TRAP_UD_FORCED)) {
7626 				kvm_queue_exception(vcpu, UD_VECTOR);
7627 				return 1;
7628 			}
7629 			if (reexecute_instruction(vcpu, cr2_or_gpa,
7630 						  write_fault_to_spt,
7631 						  emulation_type))
7632 				return 1;
7633 			if (ctxt->have_exception) {
7634 				/*
7635 				 * #UD should result in just EMULATION_FAILED, and trap-like
7636 				 * exception should not be encountered during decode.
7637 				 */
7638 				WARN_ON_ONCE(ctxt->exception.vector == UD_VECTOR ||
7639 					     exception_type(ctxt->exception.vector) == EXCPT_TRAP);
7640 				inject_emulated_exception(vcpu);
7641 				return 1;
7642 			}
7643 			return handle_emulation_failure(vcpu, emulation_type);
7644 		}
7645 	}
7646 
7647 	if ((emulation_type & EMULTYPE_VMWARE_GP) &&
7648 	    !is_vmware_backdoor_opcode(ctxt)) {
7649 		kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
7650 		return 1;
7651 	}
7652 
7653 	/*
7654 	 * Note, EMULTYPE_SKIP is intended for use *only* by vendor callbacks
7655 	 * for kvm_skip_emulated_instruction().  The caller is responsible for
7656 	 * updating interruptibility state and injecting single-step #DBs.
7657 	 */
7658 	if (emulation_type & EMULTYPE_SKIP) {
7659 		kvm_rip_write(vcpu, ctxt->_eip);
7660 		if (ctxt->eflags & X86_EFLAGS_RF)
7661 			kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
7662 		return 1;
7663 	}
7664 
7665 	if (retry_instruction(ctxt, cr2_or_gpa, emulation_type))
7666 		return 1;
7667 
7668 	/* this is needed for vmware backdoor interface to work since it
7669 	   changes registers values  during IO operation */
7670 	if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
7671 		vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
7672 		emulator_invalidate_register_cache(ctxt);
7673 	}
7674 
7675 restart:
7676 	if (emulation_type & EMULTYPE_PF) {
7677 		/* Save the faulting GPA (cr2) in the address field */
7678 		ctxt->exception.address = cr2_or_gpa;
7679 
7680 		/* With shadow page tables, cr2 contains a GVA or nGPA. */
7681 		if (vcpu->arch.mmu->direct_map) {
7682 			ctxt->gpa_available = true;
7683 			ctxt->gpa_val = cr2_or_gpa;
7684 		}
7685 	} else {
7686 		/* Sanitize the address out of an abundance of paranoia. */
7687 		ctxt->exception.address = 0;
7688 	}
7689 
7690 	r = x86_emulate_insn(ctxt);
7691 
7692 	if (r == EMULATION_INTERCEPTED)
7693 		return 1;
7694 
7695 	if (r == EMULATION_FAILED) {
7696 		if (reexecute_instruction(vcpu, cr2_or_gpa, write_fault_to_spt,
7697 					emulation_type))
7698 			return 1;
7699 
7700 		return handle_emulation_failure(vcpu, emulation_type);
7701 	}
7702 
7703 	if (ctxt->have_exception) {
7704 		r = 1;
7705 		if (inject_emulated_exception(vcpu))
7706 			return r;
7707 	} else if (vcpu->arch.pio.count) {
7708 		if (!vcpu->arch.pio.in) {
7709 			/* FIXME: return into emulator if single-stepping.  */
7710 			vcpu->arch.pio.count = 0;
7711 		} else {
7712 			writeback = false;
7713 			vcpu->arch.complete_userspace_io = complete_emulated_pio;
7714 		}
7715 		r = 0;
7716 	} else if (vcpu->mmio_needed) {
7717 		++vcpu->stat.mmio_exits;
7718 
7719 		if (!vcpu->mmio_is_write)
7720 			writeback = false;
7721 		r = 0;
7722 		vcpu->arch.complete_userspace_io = complete_emulated_mmio;
7723 	} else if (r == EMULATION_RESTART)
7724 		goto restart;
7725 	else
7726 		r = 1;
7727 
7728 	if (writeback) {
7729 		unsigned long rflags = static_call(kvm_x86_get_rflags)(vcpu);
7730 		toggle_interruptibility(vcpu, ctxt->interruptibility);
7731 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
7732 		if (!ctxt->have_exception ||
7733 		    exception_type(ctxt->exception.vector) == EXCPT_TRAP) {
7734 			kvm_rip_write(vcpu, ctxt->eip);
7735 			if (r && (ctxt->tf || (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)))
7736 				r = kvm_vcpu_do_singlestep(vcpu);
7737 			if (kvm_x86_ops.update_emulated_instruction)
7738 				static_call(kvm_x86_update_emulated_instruction)(vcpu);
7739 			__kvm_set_rflags(vcpu, ctxt->eflags);
7740 		}
7741 
7742 		/*
7743 		 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
7744 		 * do nothing, and it will be requested again as soon as
7745 		 * the shadow expires.  But we still need to check here,
7746 		 * because POPF has no interrupt shadow.
7747 		 */
7748 		if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
7749 			kvm_make_request(KVM_REQ_EVENT, vcpu);
7750 	} else
7751 		vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
7752 
7753 	return r;
7754 }
7755 
7756 int kvm_emulate_instruction(struct kvm_vcpu *vcpu, int emulation_type)
7757 {
7758 	return x86_emulate_instruction(vcpu, 0, emulation_type, NULL, 0);
7759 }
7760 EXPORT_SYMBOL_GPL(kvm_emulate_instruction);
7761 
7762 int kvm_emulate_instruction_from_buffer(struct kvm_vcpu *vcpu,
7763 					void *insn, int insn_len)
7764 {
7765 	return x86_emulate_instruction(vcpu, 0, 0, insn, insn_len);
7766 }
7767 EXPORT_SYMBOL_GPL(kvm_emulate_instruction_from_buffer);
7768 
7769 static int complete_fast_pio_out_port_0x7e(struct kvm_vcpu *vcpu)
7770 {
7771 	vcpu->arch.pio.count = 0;
7772 	return 1;
7773 }
7774 
7775 static int complete_fast_pio_out(struct kvm_vcpu *vcpu)
7776 {
7777 	vcpu->arch.pio.count = 0;
7778 
7779 	if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip)))
7780 		return 1;
7781 
7782 	return kvm_skip_emulated_instruction(vcpu);
7783 }
7784 
7785 static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size,
7786 			    unsigned short port)
7787 {
7788 	unsigned long val = kvm_rax_read(vcpu);
7789 	int ret = emulator_pio_out(vcpu, size, port, &val, 1);
7790 
7791 	if (ret)
7792 		return ret;
7793 
7794 	/*
7795 	 * Workaround userspace that relies on old KVM behavior of %rip being
7796 	 * incremented prior to exiting to userspace to handle "OUT 0x7e".
7797 	 */
7798 	if (port == 0x7e &&
7799 	    kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_OUT_7E_INC_RIP)) {
7800 		vcpu->arch.complete_userspace_io =
7801 			complete_fast_pio_out_port_0x7e;
7802 		kvm_skip_emulated_instruction(vcpu);
7803 	} else {
7804 		vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
7805 		vcpu->arch.complete_userspace_io = complete_fast_pio_out;
7806 	}
7807 	return 0;
7808 }
7809 
7810 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
7811 {
7812 	unsigned long val;
7813 
7814 	/* We should only ever be called with arch.pio.count equal to 1 */
7815 	BUG_ON(vcpu->arch.pio.count != 1);
7816 
7817 	if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip))) {
7818 		vcpu->arch.pio.count = 0;
7819 		return 1;
7820 	}
7821 
7822 	/* For size less than 4 we merge, else we zero extend */
7823 	val = (vcpu->arch.pio.size < 4) ? kvm_rax_read(vcpu) : 0;
7824 
7825 	/*
7826 	 * Since vcpu->arch.pio.count == 1 let emulator_pio_in perform
7827 	 * the copy and tracing
7828 	 */
7829 	emulator_pio_in(vcpu, vcpu->arch.pio.size, vcpu->arch.pio.port, &val, 1);
7830 	kvm_rax_write(vcpu, val);
7831 
7832 	return kvm_skip_emulated_instruction(vcpu);
7833 }
7834 
7835 static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size,
7836 			   unsigned short port)
7837 {
7838 	unsigned long val;
7839 	int ret;
7840 
7841 	/* For size less than 4 we merge, else we zero extend */
7842 	val = (size < 4) ? kvm_rax_read(vcpu) : 0;
7843 
7844 	ret = emulator_pio_in(vcpu, size, port, &val, 1);
7845 	if (ret) {
7846 		kvm_rax_write(vcpu, val);
7847 		return ret;
7848 	}
7849 
7850 	vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
7851 	vcpu->arch.complete_userspace_io = complete_fast_pio_in;
7852 
7853 	return 0;
7854 }
7855 
7856 int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in)
7857 {
7858 	int ret;
7859 
7860 	if (in)
7861 		ret = kvm_fast_pio_in(vcpu, size, port);
7862 	else
7863 		ret = kvm_fast_pio_out(vcpu, size, port);
7864 	return ret && kvm_skip_emulated_instruction(vcpu);
7865 }
7866 EXPORT_SYMBOL_GPL(kvm_fast_pio);
7867 
7868 static int kvmclock_cpu_down_prep(unsigned int cpu)
7869 {
7870 	__this_cpu_write(cpu_tsc_khz, 0);
7871 	return 0;
7872 }
7873 
7874 static void tsc_khz_changed(void *data)
7875 {
7876 	struct cpufreq_freqs *freq = data;
7877 	unsigned long khz = 0;
7878 
7879 	if (data)
7880 		khz = freq->new;
7881 	else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
7882 		khz = cpufreq_quick_get(raw_smp_processor_id());
7883 	if (!khz)
7884 		khz = tsc_khz;
7885 	__this_cpu_write(cpu_tsc_khz, khz);
7886 }
7887 
7888 #ifdef CONFIG_X86_64
7889 static void kvm_hyperv_tsc_notifier(void)
7890 {
7891 	struct kvm *kvm;
7892 	struct kvm_vcpu *vcpu;
7893 	int cpu;
7894 	unsigned long flags;
7895 
7896 	mutex_lock(&kvm_lock);
7897 	list_for_each_entry(kvm, &vm_list, vm_list)
7898 		kvm_make_mclock_inprogress_request(kvm);
7899 
7900 	hyperv_stop_tsc_emulation();
7901 
7902 	/* TSC frequency always matches when on Hyper-V */
7903 	for_each_present_cpu(cpu)
7904 		per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
7905 	kvm_max_guest_tsc_khz = tsc_khz;
7906 
7907 	list_for_each_entry(kvm, &vm_list, vm_list) {
7908 		struct kvm_arch *ka = &kvm->arch;
7909 
7910 		spin_lock_irqsave(&ka->pvclock_gtod_sync_lock, flags);
7911 		pvclock_update_vm_gtod_copy(kvm);
7912 		spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
7913 
7914 		kvm_for_each_vcpu(cpu, vcpu, kvm)
7915 			kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7916 
7917 		kvm_for_each_vcpu(cpu, vcpu, kvm)
7918 			kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
7919 	}
7920 	mutex_unlock(&kvm_lock);
7921 }
7922 #endif
7923 
7924 static void __kvmclock_cpufreq_notifier(struct cpufreq_freqs *freq, int cpu)
7925 {
7926 	struct kvm *kvm;
7927 	struct kvm_vcpu *vcpu;
7928 	int i, send_ipi = 0;
7929 
7930 	/*
7931 	 * We allow guests to temporarily run on slowing clocks,
7932 	 * provided we notify them after, or to run on accelerating
7933 	 * clocks, provided we notify them before.  Thus time never
7934 	 * goes backwards.
7935 	 *
7936 	 * However, we have a problem.  We can't atomically update
7937 	 * the frequency of a given CPU from this function; it is
7938 	 * merely a notifier, which can be called from any CPU.
7939 	 * Changing the TSC frequency at arbitrary points in time
7940 	 * requires a recomputation of local variables related to
7941 	 * the TSC for each VCPU.  We must flag these local variables
7942 	 * to be updated and be sure the update takes place with the
7943 	 * new frequency before any guests proceed.
7944 	 *
7945 	 * Unfortunately, the combination of hotplug CPU and frequency
7946 	 * change creates an intractable locking scenario; the order
7947 	 * of when these callouts happen is undefined with respect to
7948 	 * CPU hotplug, and they can race with each other.  As such,
7949 	 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
7950 	 * undefined; you can actually have a CPU frequency change take
7951 	 * place in between the computation of X and the setting of the
7952 	 * variable.  To protect against this problem, all updates of
7953 	 * the per_cpu tsc_khz variable are done in an interrupt
7954 	 * protected IPI, and all callers wishing to update the value
7955 	 * must wait for a synchronous IPI to complete (which is trivial
7956 	 * if the caller is on the CPU already).  This establishes the
7957 	 * necessary total order on variable updates.
7958 	 *
7959 	 * Note that because a guest time update may take place
7960 	 * anytime after the setting of the VCPU's request bit, the
7961 	 * correct TSC value must be set before the request.  However,
7962 	 * to ensure the update actually makes it to any guest which
7963 	 * starts running in hardware virtualization between the set
7964 	 * and the acquisition of the spinlock, we must also ping the
7965 	 * CPU after setting the request bit.
7966 	 *
7967 	 */
7968 
7969 	smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
7970 
7971 	mutex_lock(&kvm_lock);
7972 	list_for_each_entry(kvm, &vm_list, vm_list) {
7973 		kvm_for_each_vcpu(i, vcpu, kvm) {
7974 			if (vcpu->cpu != cpu)
7975 				continue;
7976 			kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7977 			if (vcpu->cpu != raw_smp_processor_id())
7978 				send_ipi = 1;
7979 		}
7980 	}
7981 	mutex_unlock(&kvm_lock);
7982 
7983 	if (freq->old < freq->new && send_ipi) {
7984 		/*
7985 		 * We upscale the frequency.  Must make the guest
7986 		 * doesn't see old kvmclock values while running with
7987 		 * the new frequency, otherwise we risk the guest sees
7988 		 * time go backwards.
7989 		 *
7990 		 * In case we update the frequency for another cpu
7991 		 * (which might be in guest context) send an interrupt
7992 		 * to kick the cpu out of guest context.  Next time
7993 		 * guest context is entered kvmclock will be updated,
7994 		 * so the guest will not see stale values.
7995 		 */
7996 		smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
7997 	}
7998 }
7999 
8000 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
8001 				     void *data)
8002 {
8003 	struct cpufreq_freqs *freq = data;
8004 	int cpu;
8005 
8006 	if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
8007 		return 0;
8008 	if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
8009 		return 0;
8010 
8011 	for_each_cpu(cpu, freq->policy->cpus)
8012 		__kvmclock_cpufreq_notifier(freq, cpu);
8013 
8014 	return 0;
8015 }
8016 
8017 static struct notifier_block kvmclock_cpufreq_notifier_block = {
8018 	.notifier_call  = kvmclock_cpufreq_notifier
8019 };
8020 
8021 static int kvmclock_cpu_online(unsigned int cpu)
8022 {
8023 	tsc_khz_changed(NULL);
8024 	return 0;
8025 }
8026 
8027 static void kvm_timer_init(void)
8028 {
8029 	max_tsc_khz = tsc_khz;
8030 
8031 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
8032 #ifdef CONFIG_CPU_FREQ
8033 		struct cpufreq_policy *policy;
8034 		int cpu;
8035 
8036 		cpu = get_cpu();
8037 		policy = cpufreq_cpu_get(cpu);
8038 		if (policy) {
8039 			if (policy->cpuinfo.max_freq)
8040 				max_tsc_khz = policy->cpuinfo.max_freq;
8041 			cpufreq_cpu_put(policy);
8042 		}
8043 		put_cpu();
8044 #endif
8045 		cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
8046 					  CPUFREQ_TRANSITION_NOTIFIER);
8047 	}
8048 
8049 	cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
8050 			  kvmclock_cpu_online, kvmclock_cpu_down_prep);
8051 }
8052 
8053 DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
8054 EXPORT_PER_CPU_SYMBOL_GPL(current_vcpu);
8055 
8056 int kvm_is_in_guest(void)
8057 {
8058 	return __this_cpu_read(current_vcpu) != NULL;
8059 }
8060 
8061 static int kvm_is_user_mode(void)
8062 {
8063 	int user_mode = 3;
8064 
8065 	if (__this_cpu_read(current_vcpu))
8066 		user_mode = static_call(kvm_x86_get_cpl)(__this_cpu_read(current_vcpu));
8067 
8068 	return user_mode != 0;
8069 }
8070 
8071 static unsigned long kvm_get_guest_ip(void)
8072 {
8073 	unsigned long ip = 0;
8074 
8075 	if (__this_cpu_read(current_vcpu))
8076 		ip = kvm_rip_read(__this_cpu_read(current_vcpu));
8077 
8078 	return ip;
8079 }
8080 
8081 static void kvm_handle_intel_pt_intr(void)
8082 {
8083 	struct kvm_vcpu *vcpu = __this_cpu_read(current_vcpu);
8084 
8085 	kvm_make_request(KVM_REQ_PMI, vcpu);
8086 	__set_bit(MSR_CORE_PERF_GLOBAL_OVF_CTRL_TRACE_TOPA_PMI_BIT,
8087 			(unsigned long *)&vcpu->arch.pmu.global_status);
8088 }
8089 
8090 static struct perf_guest_info_callbacks kvm_guest_cbs = {
8091 	.is_in_guest		= kvm_is_in_guest,
8092 	.is_user_mode		= kvm_is_user_mode,
8093 	.get_guest_ip		= kvm_get_guest_ip,
8094 	.handle_intel_pt_intr	= kvm_handle_intel_pt_intr,
8095 };
8096 
8097 #ifdef CONFIG_X86_64
8098 static void pvclock_gtod_update_fn(struct work_struct *work)
8099 {
8100 	struct kvm *kvm;
8101 
8102 	struct kvm_vcpu *vcpu;
8103 	int i;
8104 
8105 	mutex_lock(&kvm_lock);
8106 	list_for_each_entry(kvm, &vm_list, vm_list)
8107 		kvm_for_each_vcpu(i, vcpu, kvm)
8108 			kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
8109 	atomic_set(&kvm_guest_has_master_clock, 0);
8110 	mutex_unlock(&kvm_lock);
8111 }
8112 
8113 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
8114 
8115 /*
8116  * Indirection to move queue_work() out of the tk_core.seq write held
8117  * region to prevent possible deadlocks against time accessors which
8118  * are invoked with work related locks held.
8119  */
8120 static void pvclock_irq_work_fn(struct irq_work *w)
8121 {
8122 	queue_work(system_long_wq, &pvclock_gtod_work);
8123 }
8124 
8125 static DEFINE_IRQ_WORK(pvclock_irq_work, pvclock_irq_work_fn);
8126 
8127 /*
8128  * Notification about pvclock gtod data update.
8129  */
8130 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
8131 			       void *priv)
8132 {
8133 	struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
8134 	struct timekeeper *tk = priv;
8135 
8136 	update_pvclock_gtod(tk);
8137 
8138 	/*
8139 	 * Disable master clock if host does not trust, or does not use,
8140 	 * TSC based clocksource. Delegate queue_work() to irq_work as
8141 	 * this is invoked with tk_core.seq write held.
8142 	 */
8143 	if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) &&
8144 	    atomic_read(&kvm_guest_has_master_clock) != 0)
8145 		irq_work_queue(&pvclock_irq_work);
8146 	return 0;
8147 }
8148 
8149 static struct notifier_block pvclock_gtod_notifier = {
8150 	.notifier_call = pvclock_gtod_notify,
8151 };
8152 #endif
8153 
8154 int kvm_arch_init(void *opaque)
8155 {
8156 	struct kvm_x86_init_ops *ops = opaque;
8157 	int r;
8158 
8159 	if (kvm_x86_ops.hardware_enable) {
8160 		printk(KERN_ERR "kvm: already loaded the other module\n");
8161 		r = -EEXIST;
8162 		goto out;
8163 	}
8164 
8165 	if (!ops->cpu_has_kvm_support()) {
8166 		pr_err_ratelimited("kvm: no hardware support\n");
8167 		r = -EOPNOTSUPP;
8168 		goto out;
8169 	}
8170 	if (ops->disabled_by_bios()) {
8171 		pr_err_ratelimited("kvm: disabled by bios\n");
8172 		r = -EOPNOTSUPP;
8173 		goto out;
8174 	}
8175 
8176 	/*
8177 	 * KVM explicitly assumes that the guest has an FPU and
8178 	 * FXSAVE/FXRSTOR. For example, the KVM_GET_FPU explicitly casts the
8179 	 * vCPU's FPU state as a fxregs_state struct.
8180 	 */
8181 	if (!boot_cpu_has(X86_FEATURE_FPU) || !boot_cpu_has(X86_FEATURE_FXSR)) {
8182 		printk(KERN_ERR "kvm: inadequate fpu\n");
8183 		r = -EOPNOTSUPP;
8184 		goto out;
8185 	}
8186 
8187 	r = -ENOMEM;
8188 	x86_fpu_cache = kmem_cache_create("x86_fpu", sizeof(struct fpu),
8189 					  __alignof__(struct fpu), SLAB_ACCOUNT,
8190 					  NULL);
8191 	if (!x86_fpu_cache) {
8192 		printk(KERN_ERR "kvm: failed to allocate cache for x86 fpu\n");
8193 		goto out;
8194 	}
8195 
8196 	x86_emulator_cache = kvm_alloc_emulator_cache();
8197 	if (!x86_emulator_cache) {
8198 		pr_err("kvm: failed to allocate cache for x86 emulator\n");
8199 		goto out_free_x86_fpu_cache;
8200 	}
8201 
8202 	user_return_msrs = alloc_percpu(struct kvm_user_return_msrs);
8203 	if (!user_return_msrs) {
8204 		printk(KERN_ERR "kvm: failed to allocate percpu kvm_user_return_msrs\n");
8205 		goto out_free_x86_emulator_cache;
8206 	}
8207 	kvm_nr_uret_msrs = 0;
8208 
8209 	r = kvm_mmu_module_init();
8210 	if (r)
8211 		goto out_free_percpu;
8212 
8213 	kvm_timer_init();
8214 
8215 	perf_register_guest_info_callbacks(&kvm_guest_cbs);
8216 
8217 	if (boot_cpu_has(X86_FEATURE_XSAVE)) {
8218 		host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
8219 		supported_xcr0 = host_xcr0 & KVM_SUPPORTED_XCR0;
8220 	}
8221 
8222 	if (pi_inject_timer == -1)
8223 		pi_inject_timer = housekeeping_enabled(HK_FLAG_TIMER);
8224 #ifdef CONFIG_X86_64
8225 	pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
8226 
8227 	if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
8228 		set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
8229 #endif
8230 
8231 	return 0;
8232 
8233 out_free_percpu:
8234 	free_percpu(user_return_msrs);
8235 out_free_x86_emulator_cache:
8236 	kmem_cache_destroy(x86_emulator_cache);
8237 out_free_x86_fpu_cache:
8238 	kmem_cache_destroy(x86_fpu_cache);
8239 out:
8240 	return r;
8241 }
8242 
8243 void kvm_arch_exit(void)
8244 {
8245 #ifdef CONFIG_X86_64
8246 	if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
8247 		clear_hv_tscchange_cb();
8248 #endif
8249 	kvm_lapic_exit();
8250 	perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
8251 
8252 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
8253 		cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
8254 					    CPUFREQ_TRANSITION_NOTIFIER);
8255 	cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
8256 #ifdef CONFIG_X86_64
8257 	pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
8258 	irq_work_sync(&pvclock_irq_work);
8259 	cancel_work_sync(&pvclock_gtod_work);
8260 #endif
8261 	kvm_x86_ops.hardware_enable = NULL;
8262 	kvm_mmu_module_exit();
8263 	free_percpu(user_return_msrs);
8264 	kmem_cache_destroy(x86_emulator_cache);
8265 	kmem_cache_destroy(x86_fpu_cache);
8266 #ifdef CONFIG_KVM_XEN
8267 	static_key_deferred_flush(&kvm_xen_enabled);
8268 	WARN_ON(static_branch_unlikely(&kvm_xen_enabled.key));
8269 #endif
8270 }
8271 
8272 static int __kvm_vcpu_halt(struct kvm_vcpu *vcpu, int state, int reason)
8273 {
8274 	++vcpu->stat.halt_exits;
8275 	if (lapic_in_kernel(vcpu)) {
8276 		vcpu->arch.mp_state = state;
8277 		return 1;
8278 	} else {
8279 		vcpu->run->exit_reason = reason;
8280 		return 0;
8281 	}
8282 }
8283 
8284 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
8285 {
8286 	return __kvm_vcpu_halt(vcpu, KVM_MP_STATE_HALTED, KVM_EXIT_HLT);
8287 }
8288 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
8289 
8290 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
8291 {
8292 	int ret = kvm_skip_emulated_instruction(vcpu);
8293 	/*
8294 	 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
8295 	 * KVM_EXIT_DEBUG here.
8296 	 */
8297 	return kvm_vcpu_halt(vcpu) && ret;
8298 }
8299 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
8300 
8301 int kvm_emulate_ap_reset_hold(struct kvm_vcpu *vcpu)
8302 {
8303 	int ret = kvm_skip_emulated_instruction(vcpu);
8304 
8305 	return __kvm_vcpu_halt(vcpu, KVM_MP_STATE_AP_RESET_HOLD, KVM_EXIT_AP_RESET_HOLD) && ret;
8306 }
8307 EXPORT_SYMBOL_GPL(kvm_emulate_ap_reset_hold);
8308 
8309 #ifdef CONFIG_X86_64
8310 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
8311 			        unsigned long clock_type)
8312 {
8313 	struct kvm_clock_pairing clock_pairing;
8314 	struct timespec64 ts;
8315 	u64 cycle;
8316 	int ret;
8317 
8318 	if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
8319 		return -KVM_EOPNOTSUPP;
8320 
8321 	if (!kvm_get_walltime_and_clockread(&ts, &cycle))
8322 		return -KVM_EOPNOTSUPP;
8323 
8324 	clock_pairing.sec = ts.tv_sec;
8325 	clock_pairing.nsec = ts.tv_nsec;
8326 	clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
8327 	clock_pairing.flags = 0;
8328 	memset(&clock_pairing.pad, 0, sizeof(clock_pairing.pad));
8329 
8330 	ret = 0;
8331 	if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
8332 			    sizeof(struct kvm_clock_pairing)))
8333 		ret = -KVM_EFAULT;
8334 
8335 	return ret;
8336 }
8337 #endif
8338 
8339 /*
8340  * kvm_pv_kick_cpu_op:  Kick a vcpu.
8341  *
8342  * @apicid - apicid of vcpu to be kicked.
8343  */
8344 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
8345 {
8346 	struct kvm_lapic_irq lapic_irq;
8347 
8348 	lapic_irq.shorthand = APIC_DEST_NOSHORT;
8349 	lapic_irq.dest_mode = APIC_DEST_PHYSICAL;
8350 	lapic_irq.level = 0;
8351 	lapic_irq.dest_id = apicid;
8352 	lapic_irq.msi_redir_hint = false;
8353 
8354 	lapic_irq.delivery_mode = APIC_DM_REMRD;
8355 	kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
8356 }
8357 
8358 bool kvm_apicv_activated(struct kvm *kvm)
8359 {
8360 	return (READ_ONCE(kvm->arch.apicv_inhibit_reasons) == 0);
8361 }
8362 EXPORT_SYMBOL_GPL(kvm_apicv_activated);
8363 
8364 void kvm_apicv_init(struct kvm *kvm, bool enable)
8365 {
8366 	if (enable)
8367 		clear_bit(APICV_INHIBIT_REASON_DISABLE,
8368 			  &kvm->arch.apicv_inhibit_reasons);
8369 	else
8370 		set_bit(APICV_INHIBIT_REASON_DISABLE,
8371 			&kvm->arch.apicv_inhibit_reasons);
8372 }
8373 EXPORT_SYMBOL_GPL(kvm_apicv_init);
8374 
8375 static void kvm_sched_yield(struct kvm_vcpu *vcpu, unsigned long dest_id)
8376 {
8377 	struct kvm_vcpu *target = NULL;
8378 	struct kvm_apic_map *map;
8379 
8380 	vcpu->stat.directed_yield_attempted++;
8381 
8382 	if (single_task_running())
8383 		goto no_yield;
8384 
8385 	rcu_read_lock();
8386 	map = rcu_dereference(vcpu->kvm->arch.apic_map);
8387 
8388 	if (likely(map) && dest_id <= map->max_apic_id && map->phys_map[dest_id])
8389 		target = map->phys_map[dest_id]->vcpu;
8390 
8391 	rcu_read_unlock();
8392 
8393 	if (!target || !READ_ONCE(target->ready))
8394 		goto no_yield;
8395 
8396 	/* Ignore requests to yield to self */
8397 	if (vcpu == target)
8398 		goto no_yield;
8399 
8400 	if (kvm_vcpu_yield_to(target) <= 0)
8401 		goto no_yield;
8402 
8403 	vcpu->stat.directed_yield_successful++;
8404 
8405 no_yield:
8406 	return;
8407 }
8408 
8409 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
8410 {
8411 	unsigned long nr, a0, a1, a2, a3, ret;
8412 	int op_64_bit;
8413 
8414 	if (kvm_xen_hypercall_enabled(vcpu->kvm))
8415 		return kvm_xen_hypercall(vcpu);
8416 
8417 	if (kvm_hv_hypercall_enabled(vcpu))
8418 		return kvm_hv_hypercall(vcpu);
8419 
8420 	nr = kvm_rax_read(vcpu);
8421 	a0 = kvm_rbx_read(vcpu);
8422 	a1 = kvm_rcx_read(vcpu);
8423 	a2 = kvm_rdx_read(vcpu);
8424 	a3 = kvm_rsi_read(vcpu);
8425 
8426 	trace_kvm_hypercall(nr, a0, a1, a2, a3);
8427 
8428 	op_64_bit = is_64_bit_mode(vcpu);
8429 	if (!op_64_bit) {
8430 		nr &= 0xFFFFFFFF;
8431 		a0 &= 0xFFFFFFFF;
8432 		a1 &= 0xFFFFFFFF;
8433 		a2 &= 0xFFFFFFFF;
8434 		a3 &= 0xFFFFFFFF;
8435 	}
8436 
8437 	if (static_call(kvm_x86_get_cpl)(vcpu) != 0) {
8438 		ret = -KVM_EPERM;
8439 		goto out;
8440 	}
8441 
8442 	ret = -KVM_ENOSYS;
8443 
8444 	switch (nr) {
8445 	case KVM_HC_VAPIC_POLL_IRQ:
8446 		ret = 0;
8447 		break;
8448 	case KVM_HC_KICK_CPU:
8449 		if (!guest_pv_has(vcpu, KVM_FEATURE_PV_UNHALT))
8450 			break;
8451 
8452 		kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
8453 		kvm_sched_yield(vcpu, a1);
8454 		ret = 0;
8455 		break;
8456 #ifdef CONFIG_X86_64
8457 	case KVM_HC_CLOCK_PAIRING:
8458 		ret = kvm_pv_clock_pairing(vcpu, a0, a1);
8459 		break;
8460 #endif
8461 	case KVM_HC_SEND_IPI:
8462 		if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SEND_IPI))
8463 			break;
8464 
8465 		ret = kvm_pv_send_ipi(vcpu->kvm, a0, a1, a2, a3, op_64_bit);
8466 		break;
8467 	case KVM_HC_SCHED_YIELD:
8468 		if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SCHED_YIELD))
8469 			break;
8470 
8471 		kvm_sched_yield(vcpu, a0);
8472 		ret = 0;
8473 		break;
8474 	default:
8475 		ret = -KVM_ENOSYS;
8476 		break;
8477 	}
8478 out:
8479 	if (!op_64_bit)
8480 		ret = (u32)ret;
8481 	kvm_rax_write(vcpu, ret);
8482 
8483 	++vcpu->stat.hypercalls;
8484 	return kvm_skip_emulated_instruction(vcpu);
8485 }
8486 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
8487 
8488 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
8489 {
8490 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
8491 	char instruction[3];
8492 	unsigned long rip = kvm_rip_read(vcpu);
8493 
8494 	static_call(kvm_x86_patch_hypercall)(vcpu, instruction);
8495 
8496 	return emulator_write_emulated(ctxt, rip, instruction, 3,
8497 		&ctxt->exception);
8498 }
8499 
8500 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
8501 {
8502 	return vcpu->run->request_interrupt_window &&
8503 		likely(!pic_in_kernel(vcpu->kvm));
8504 }
8505 
8506 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
8507 {
8508 	struct kvm_run *kvm_run = vcpu->run;
8509 
8510 	/*
8511 	 * if_flag is obsolete and useless, so do not bother
8512 	 * setting it for SEV-ES guests.  Userspace can just
8513 	 * use kvm_run->ready_for_interrupt_injection.
8514 	 */
8515 	kvm_run->if_flag = !vcpu->arch.guest_state_protected
8516 		&& (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
8517 
8518 	kvm_run->cr8 = kvm_get_cr8(vcpu);
8519 	kvm_run->apic_base = kvm_get_apic_base(vcpu);
8520 	kvm_run->ready_for_interrupt_injection =
8521 		pic_in_kernel(vcpu->kvm) ||
8522 		kvm_vcpu_ready_for_interrupt_injection(vcpu);
8523 
8524 	if (is_smm(vcpu))
8525 		kvm_run->flags |= KVM_RUN_X86_SMM;
8526 }
8527 
8528 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
8529 {
8530 	int max_irr, tpr;
8531 
8532 	if (!kvm_x86_ops.update_cr8_intercept)
8533 		return;
8534 
8535 	if (!lapic_in_kernel(vcpu))
8536 		return;
8537 
8538 	if (vcpu->arch.apicv_active)
8539 		return;
8540 
8541 	if (!vcpu->arch.apic->vapic_addr)
8542 		max_irr = kvm_lapic_find_highest_irr(vcpu);
8543 	else
8544 		max_irr = -1;
8545 
8546 	if (max_irr != -1)
8547 		max_irr >>= 4;
8548 
8549 	tpr = kvm_lapic_get_cr8(vcpu);
8550 
8551 	static_call(kvm_x86_update_cr8_intercept)(vcpu, tpr, max_irr);
8552 }
8553 
8554 
8555 int kvm_check_nested_events(struct kvm_vcpu *vcpu)
8556 {
8557 	if (WARN_ON_ONCE(!is_guest_mode(vcpu)))
8558 		return -EIO;
8559 
8560 	if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
8561 		kvm_x86_ops.nested_ops->triple_fault(vcpu);
8562 		return 1;
8563 	}
8564 
8565 	return kvm_x86_ops.nested_ops->check_events(vcpu);
8566 }
8567 
8568 static void kvm_inject_exception(struct kvm_vcpu *vcpu)
8569 {
8570 	if (vcpu->arch.exception.error_code && !is_protmode(vcpu))
8571 		vcpu->arch.exception.error_code = false;
8572 	static_call(kvm_x86_queue_exception)(vcpu);
8573 }
8574 
8575 static void inject_pending_event(struct kvm_vcpu *vcpu, bool *req_immediate_exit)
8576 {
8577 	int r;
8578 	bool can_inject = true;
8579 
8580 	/* try to reinject previous events if any */
8581 
8582 	if (vcpu->arch.exception.injected) {
8583 		kvm_inject_exception(vcpu);
8584 		can_inject = false;
8585 	}
8586 	/*
8587 	 * Do not inject an NMI or interrupt if there is a pending
8588 	 * exception.  Exceptions and interrupts are recognized at
8589 	 * instruction boundaries, i.e. the start of an instruction.
8590 	 * Trap-like exceptions, e.g. #DB, have higher priority than
8591 	 * NMIs and interrupts, i.e. traps are recognized before an
8592 	 * NMI/interrupt that's pending on the same instruction.
8593 	 * Fault-like exceptions, e.g. #GP and #PF, are the lowest
8594 	 * priority, but are only generated (pended) during instruction
8595 	 * execution, i.e. a pending fault-like exception means the
8596 	 * fault occurred on the *previous* instruction and must be
8597 	 * serviced prior to recognizing any new events in order to
8598 	 * fully complete the previous instruction.
8599 	 */
8600 	else if (!vcpu->arch.exception.pending) {
8601 		if (vcpu->arch.nmi_injected) {
8602 			static_call(kvm_x86_set_nmi)(vcpu);
8603 			can_inject = false;
8604 		} else if (vcpu->arch.interrupt.injected) {
8605 			static_call(kvm_x86_set_irq)(vcpu);
8606 			can_inject = false;
8607 		}
8608 	}
8609 
8610 	WARN_ON_ONCE(vcpu->arch.exception.injected &&
8611 		     vcpu->arch.exception.pending);
8612 
8613 	/*
8614 	 * Call check_nested_events() even if we reinjected a previous event
8615 	 * in order for caller to determine if it should require immediate-exit
8616 	 * from L2 to L1 due to pending L1 events which require exit
8617 	 * from L2 to L1.
8618 	 */
8619 	if (is_guest_mode(vcpu)) {
8620 		r = kvm_check_nested_events(vcpu);
8621 		if (r < 0)
8622 			goto busy;
8623 	}
8624 
8625 	/* try to inject new event if pending */
8626 	if (vcpu->arch.exception.pending) {
8627 		trace_kvm_inj_exception(vcpu->arch.exception.nr,
8628 					vcpu->arch.exception.has_error_code,
8629 					vcpu->arch.exception.error_code);
8630 
8631 		vcpu->arch.exception.pending = false;
8632 		vcpu->arch.exception.injected = true;
8633 
8634 		if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
8635 			__kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
8636 					     X86_EFLAGS_RF);
8637 
8638 		if (vcpu->arch.exception.nr == DB_VECTOR) {
8639 			kvm_deliver_exception_payload(vcpu);
8640 			if (vcpu->arch.dr7 & DR7_GD) {
8641 				vcpu->arch.dr7 &= ~DR7_GD;
8642 				kvm_update_dr7(vcpu);
8643 			}
8644 		}
8645 
8646 		kvm_inject_exception(vcpu);
8647 		can_inject = false;
8648 	}
8649 
8650 	/*
8651 	 * Finally, inject interrupt events.  If an event cannot be injected
8652 	 * due to architectural conditions (e.g. IF=0) a window-open exit
8653 	 * will re-request KVM_REQ_EVENT.  Sometimes however an event is pending
8654 	 * and can architecturally be injected, but we cannot do it right now:
8655 	 * an interrupt could have arrived just now and we have to inject it
8656 	 * as a vmexit, or there could already an event in the queue, which is
8657 	 * indicated by can_inject.  In that case we request an immediate exit
8658 	 * in order to make progress and get back here for another iteration.
8659 	 * The kvm_x86_ops hooks communicate this by returning -EBUSY.
8660 	 */
8661 	if (vcpu->arch.smi_pending) {
8662 		r = can_inject ? static_call(kvm_x86_smi_allowed)(vcpu, true) : -EBUSY;
8663 		if (r < 0)
8664 			goto busy;
8665 		if (r) {
8666 			vcpu->arch.smi_pending = false;
8667 			++vcpu->arch.smi_count;
8668 			enter_smm(vcpu);
8669 			can_inject = false;
8670 		} else
8671 			static_call(kvm_x86_enable_smi_window)(vcpu);
8672 	}
8673 
8674 	if (vcpu->arch.nmi_pending) {
8675 		r = can_inject ? static_call(kvm_x86_nmi_allowed)(vcpu, true) : -EBUSY;
8676 		if (r < 0)
8677 			goto busy;
8678 		if (r) {
8679 			--vcpu->arch.nmi_pending;
8680 			vcpu->arch.nmi_injected = true;
8681 			static_call(kvm_x86_set_nmi)(vcpu);
8682 			can_inject = false;
8683 			WARN_ON(static_call(kvm_x86_nmi_allowed)(vcpu, true) < 0);
8684 		}
8685 		if (vcpu->arch.nmi_pending)
8686 			static_call(kvm_x86_enable_nmi_window)(vcpu);
8687 	}
8688 
8689 	if (kvm_cpu_has_injectable_intr(vcpu)) {
8690 		r = can_inject ? static_call(kvm_x86_interrupt_allowed)(vcpu, true) : -EBUSY;
8691 		if (r < 0)
8692 			goto busy;
8693 		if (r) {
8694 			kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu), false);
8695 			static_call(kvm_x86_set_irq)(vcpu);
8696 			WARN_ON(static_call(kvm_x86_interrupt_allowed)(vcpu, true) < 0);
8697 		}
8698 		if (kvm_cpu_has_injectable_intr(vcpu))
8699 			static_call(kvm_x86_enable_irq_window)(vcpu);
8700 	}
8701 
8702 	if (is_guest_mode(vcpu) &&
8703 	    kvm_x86_ops.nested_ops->hv_timer_pending &&
8704 	    kvm_x86_ops.nested_ops->hv_timer_pending(vcpu))
8705 		*req_immediate_exit = true;
8706 
8707 	WARN_ON(vcpu->arch.exception.pending);
8708 	return;
8709 
8710 busy:
8711 	*req_immediate_exit = true;
8712 	return;
8713 }
8714 
8715 static void process_nmi(struct kvm_vcpu *vcpu)
8716 {
8717 	unsigned limit = 2;
8718 
8719 	/*
8720 	 * x86 is limited to one NMI running, and one NMI pending after it.
8721 	 * If an NMI is already in progress, limit further NMIs to just one.
8722 	 * Otherwise, allow two (and we'll inject the first one immediately).
8723 	 */
8724 	if (static_call(kvm_x86_get_nmi_mask)(vcpu) || vcpu->arch.nmi_injected)
8725 		limit = 1;
8726 
8727 	vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
8728 	vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
8729 	kvm_make_request(KVM_REQ_EVENT, vcpu);
8730 }
8731 
8732 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
8733 {
8734 	u32 flags = 0;
8735 	flags |= seg->g       << 23;
8736 	flags |= seg->db      << 22;
8737 	flags |= seg->l       << 21;
8738 	flags |= seg->avl     << 20;
8739 	flags |= seg->present << 15;
8740 	flags |= seg->dpl     << 13;
8741 	flags |= seg->s       << 12;
8742 	flags |= seg->type    << 8;
8743 	return flags;
8744 }
8745 
8746 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
8747 {
8748 	struct kvm_segment seg;
8749 	int offset;
8750 
8751 	kvm_get_segment(vcpu, &seg, n);
8752 	put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
8753 
8754 	if (n < 3)
8755 		offset = 0x7f84 + n * 12;
8756 	else
8757 		offset = 0x7f2c + (n - 3) * 12;
8758 
8759 	put_smstate(u32, buf, offset + 8, seg.base);
8760 	put_smstate(u32, buf, offset + 4, seg.limit);
8761 	put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
8762 }
8763 
8764 #ifdef CONFIG_X86_64
8765 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
8766 {
8767 	struct kvm_segment seg;
8768 	int offset;
8769 	u16 flags;
8770 
8771 	kvm_get_segment(vcpu, &seg, n);
8772 	offset = 0x7e00 + n * 16;
8773 
8774 	flags = enter_smm_get_segment_flags(&seg) >> 8;
8775 	put_smstate(u16, buf, offset, seg.selector);
8776 	put_smstate(u16, buf, offset + 2, flags);
8777 	put_smstate(u32, buf, offset + 4, seg.limit);
8778 	put_smstate(u64, buf, offset + 8, seg.base);
8779 }
8780 #endif
8781 
8782 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
8783 {
8784 	struct desc_ptr dt;
8785 	struct kvm_segment seg;
8786 	unsigned long val;
8787 	int i;
8788 
8789 	put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
8790 	put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
8791 	put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
8792 	put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
8793 
8794 	for (i = 0; i < 8; i++)
8795 		put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read_raw(vcpu, i));
8796 
8797 	kvm_get_dr(vcpu, 6, &val);
8798 	put_smstate(u32, buf, 0x7fcc, (u32)val);
8799 	kvm_get_dr(vcpu, 7, &val);
8800 	put_smstate(u32, buf, 0x7fc8, (u32)val);
8801 
8802 	kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
8803 	put_smstate(u32, buf, 0x7fc4, seg.selector);
8804 	put_smstate(u32, buf, 0x7f64, seg.base);
8805 	put_smstate(u32, buf, 0x7f60, seg.limit);
8806 	put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
8807 
8808 	kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
8809 	put_smstate(u32, buf, 0x7fc0, seg.selector);
8810 	put_smstate(u32, buf, 0x7f80, seg.base);
8811 	put_smstate(u32, buf, 0x7f7c, seg.limit);
8812 	put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
8813 
8814 	static_call(kvm_x86_get_gdt)(vcpu, &dt);
8815 	put_smstate(u32, buf, 0x7f74, dt.address);
8816 	put_smstate(u32, buf, 0x7f70, dt.size);
8817 
8818 	static_call(kvm_x86_get_idt)(vcpu, &dt);
8819 	put_smstate(u32, buf, 0x7f58, dt.address);
8820 	put_smstate(u32, buf, 0x7f54, dt.size);
8821 
8822 	for (i = 0; i < 6; i++)
8823 		enter_smm_save_seg_32(vcpu, buf, i);
8824 
8825 	put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
8826 
8827 	/* revision id */
8828 	put_smstate(u32, buf, 0x7efc, 0x00020000);
8829 	put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
8830 }
8831 
8832 #ifdef CONFIG_X86_64
8833 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
8834 {
8835 	struct desc_ptr dt;
8836 	struct kvm_segment seg;
8837 	unsigned long val;
8838 	int i;
8839 
8840 	for (i = 0; i < 16; i++)
8841 		put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read_raw(vcpu, i));
8842 
8843 	put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
8844 	put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
8845 
8846 	kvm_get_dr(vcpu, 6, &val);
8847 	put_smstate(u64, buf, 0x7f68, val);
8848 	kvm_get_dr(vcpu, 7, &val);
8849 	put_smstate(u64, buf, 0x7f60, val);
8850 
8851 	put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
8852 	put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
8853 	put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
8854 
8855 	put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
8856 
8857 	/* revision id */
8858 	put_smstate(u32, buf, 0x7efc, 0x00020064);
8859 
8860 	put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
8861 
8862 	kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
8863 	put_smstate(u16, buf, 0x7e90, seg.selector);
8864 	put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
8865 	put_smstate(u32, buf, 0x7e94, seg.limit);
8866 	put_smstate(u64, buf, 0x7e98, seg.base);
8867 
8868 	static_call(kvm_x86_get_idt)(vcpu, &dt);
8869 	put_smstate(u32, buf, 0x7e84, dt.size);
8870 	put_smstate(u64, buf, 0x7e88, dt.address);
8871 
8872 	kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
8873 	put_smstate(u16, buf, 0x7e70, seg.selector);
8874 	put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
8875 	put_smstate(u32, buf, 0x7e74, seg.limit);
8876 	put_smstate(u64, buf, 0x7e78, seg.base);
8877 
8878 	static_call(kvm_x86_get_gdt)(vcpu, &dt);
8879 	put_smstate(u32, buf, 0x7e64, dt.size);
8880 	put_smstate(u64, buf, 0x7e68, dt.address);
8881 
8882 	for (i = 0; i < 6; i++)
8883 		enter_smm_save_seg_64(vcpu, buf, i);
8884 }
8885 #endif
8886 
8887 static void enter_smm(struct kvm_vcpu *vcpu)
8888 {
8889 	struct kvm_segment cs, ds;
8890 	struct desc_ptr dt;
8891 	char buf[512];
8892 	u32 cr0;
8893 
8894 	trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
8895 	memset(buf, 0, 512);
8896 #ifdef CONFIG_X86_64
8897 	if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
8898 		enter_smm_save_state_64(vcpu, buf);
8899 	else
8900 #endif
8901 		enter_smm_save_state_32(vcpu, buf);
8902 
8903 	/*
8904 	 * Give pre_enter_smm() a chance to make ISA-specific changes to the
8905 	 * vCPU state (e.g. leave guest mode) after we've saved the state into
8906 	 * the SMM state-save area.
8907 	 */
8908 	static_call(kvm_x86_pre_enter_smm)(vcpu, buf);
8909 
8910 	vcpu->arch.hflags |= HF_SMM_MASK;
8911 	kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
8912 
8913 	if (static_call(kvm_x86_get_nmi_mask)(vcpu))
8914 		vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
8915 	else
8916 		static_call(kvm_x86_set_nmi_mask)(vcpu, true);
8917 
8918 	kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
8919 	kvm_rip_write(vcpu, 0x8000);
8920 
8921 	cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
8922 	static_call(kvm_x86_set_cr0)(vcpu, cr0);
8923 	vcpu->arch.cr0 = cr0;
8924 
8925 	static_call(kvm_x86_set_cr4)(vcpu, 0);
8926 
8927 	/* Undocumented: IDT limit is set to zero on entry to SMM.  */
8928 	dt.address = dt.size = 0;
8929 	static_call(kvm_x86_set_idt)(vcpu, &dt);
8930 
8931 	kvm_set_dr(vcpu, 7, DR7_FIXED_1);
8932 
8933 	cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
8934 	cs.base = vcpu->arch.smbase;
8935 
8936 	ds.selector = 0;
8937 	ds.base = 0;
8938 
8939 	cs.limit    = ds.limit = 0xffffffff;
8940 	cs.type     = ds.type = 0x3;
8941 	cs.dpl      = ds.dpl = 0;
8942 	cs.db       = ds.db = 0;
8943 	cs.s        = ds.s = 1;
8944 	cs.l        = ds.l = 0;
8945 	cs.g        = ds.g = 1;
8946 	cs.avl      = ds.avl = 0;
8947 	cs.present  = ds.present = 1;
8948 	cs.unusable = ds.unusable = 0;
8949 	cs.padding  = ds.padding = 0;
8950 
8951 	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
8952 	kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
8953 	kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
8954 	kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
8955 	kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
8956 	kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
8957 
8958 #ifdef CONFIG_X86_64
8959 	if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
8960 		static_call(kvm_x86_set_efer)(vcpu, 0);
8961 #endif
8962 
8963 	kvm_update_cpuid_runtime(vcpu);
8964 	kvm_mmu_reset_context(vcpu);
8965 }
8966 
8967 static void process_smi(struct kvm_vcpu *vcpu)
8968 {
8969 	vcpu->arch.smi_pending = true;
8970 	kvm_make_request(KVM_REQ_EVENT, vcpu);
8971 }
8972 
8973 void kvm_make_scan_ioapic_request_mask(struct kvm *kvm,
8974 				       unsigned long *vcpu_bitmap)
8975 {
8976 	cpumask_var_t cpus;
8977 
8978 	zalloc_cpumask_var(&cpus, GFP_ATOMIC);
8979 
8980 	kvm_make_vcpus_request_mask(kvm, KVM_REQ_SCAN_IOAPIC,
8981 				    NULL, vcpu_bitmap, cpus);
8982 
8983 	free_cpumask_var(cpus);
8984 }
8985 
8986 void kvm_make_scan_ioapic_request(struct kvm *kvm)
8987 {
8988 	kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
8989 }
8990 
8991 void kvm_vcpu_update_apicv(struct kvm_vcpu *vcpu)
8992 {
8993 	if (!lapic_in_kernel(vcpu))
8994 		return;
8995 
8996 	vcpu->arch.apicv_active = kvm_apicv_activated(vcpu->kvm);
8997 	kvm_apic_update_apicv(vcpu);
8998 	static_call(kvm_x86_refresh_apicv_exec_ctrl)(vcpu);
8999 }
9000 EXPORT_SYMBOL_GPL(kvm_vcpu_update_apicv);
9001 
9002 /*
9003  * NOTE: Do not hold any lock prior to calling this.
9004  *
9005  * In particular, kvm_request_apicv_update() expects kvm->srcu not to be
9006  * locked, because it calls __x86_set_memory_region() which does
9007  * synchronize_srcu(&kvm->srcu).
9008  */
9009 void kvm_request_apicv_update(struct kvm *kvm, bool activate, ulong bit)
9010 {
9011 	struct kvm_vcpu *except;
9012 	unsigned long old, new, expected;
9013 
9014 	if (!kvm_x86_ops.check_apicv_inhibit_reasons ||
9015 	    !static_call(kvm_x86_check_apicv_inhibit_reasons)(bit))
9016 		return;
9017 
9018 	old = READ_ONCE(kvm->arch.apicv_inhibit_reasons);
9019 	do {
9020 		expected = new = old;
9021 		if (activate)
9022 			__clear_bit(bit, &new);
9023 		else
9024 			__set_bit(bit, &new);
9025 		if (new == old)
9026 			break;
9027 		old = cmpxchg(&kvm->arch.apicv_inhibit_reasons, expected, new);
9028 	} while (old != expected);
9029 
9030 	if (!!old == !!new)
9031 		return;
9032 
9033 	trace_kvm_apicv_update_request(activate, bit);
9034 	if (kvm_x86_ops.pre_update_apicv_exec_ctrl)
9035 		static_call(kvm_x86_pre_update_apicv_exec_ctrl)(kvm, activate);
9036 
9037 	/*
9038 	 * Sending request to update APICV for all other vcpus,
9039 	 * while update the calling vcpu immediately instead of
9040 	 * waiting for another #VMEXIT to handle the request.
9041 	 */
9042 	except = kvm_get_running_vcpu();
9043 	kvm_make_all_cpus_request_except(kvm, KVM_REQ_APICV_UPDATE,
9044 					 except);
9045 	if (except)
9046 		kvm_vcpu_update_apicv(except);
9047 }
9048 EXPORT_SYMBOL_GPL(kvm_request_apicv_update);
9049 
9050 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
9051 {
9052 	if (!kvm_apic_present(vcpu))
9053 		return;
9054 
9055 	bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
9056 
9057 	if (irqchip_split(vcpu->kvm))
9058 		kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
9059 	else {
9060 		if (vcpu->arch.apicv_active)
9061 			static_call(kvm_x86_sync_pir_to_irr)(vcpu);
9062 		if (ioapic_in_kernel(vcpu->kvm))
9063 			kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
9064 	}
9065 
9066 	if (is_guest_mode(vcpu))
9067 		vcpu->arch.load_eoi_exitmap_pending = true;
9068 	else
9069 		kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
9070 }
9071 
9072 static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
9073 {
9074 	u64 eoi_exit_bitmap[4];
9075 
9076 	if (!kvm_apic_hw_enabled(vcpu->arch.apic))
9077 		return;
9078 
9079 	if (to_hv_vcpu(vcpu))
9080 		bitmap_or((ulong *)eoi_exit_bitmap,
9081 			  vcpu->arch.ioapic_handled_vectors,
9082 			  to_hv_synic(vcpu)->vec_bitmap, 256);
9083 
9084 	static_call(kvm_x86_load_eoi_exitmap)(vcpu, eoi_exit_bitmap);
9085 }
9086 
9087 void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
9088 					    unsigned long start, unsigned long end)
9089 {
9090 	unsigned long apic_address;
9091 
9092 	/*
9093 	 * The physical address of apic access page is stored in the VMCS.
9094 	 * Update it when it becomes invalid.
9095 	 */
9096 	apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
9097 	if (start <= apic_address && apic_address < end)
9098 		kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
9099 }
9100 
9101 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
9102 {
9103 	if (!lapic_in_kernel(vcpu))
9104 		return;
9105 
9106 	if (!kvm_x86_ops.set_apic_access_page_addr)
9107 		return;
9108 
9109 	static_call(kvm_x86_set_apic_access_page_addr)(vcpu);
9110 }
9111 
9112 void __kvm_request_immediate_exit(struct kvm_vcpu *vcpu)
9113 {
9114 	smp_send_reschedule(vcpu->cpu);
9115 }
9116 EXPORT_SYMBOL_GPL(__kvm_request_immediate_exit);
9117 
9118 /*
9119  * Returns 1 to let vcpu_run() continue the guest execution loop without
9120  * exiting to the userspace.  Otherwise, the value will be returned to the
9121  * userspace.
9122  */
9123 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
9124 {
9125 	int r;
9126 	bool req_int_win =
9127 		dm_request_for_irq_injection(vcpu) &&
9128 		kvm_cpu_accept_dm_intr(vcpu);
9129 	fastpath_t exit_fastpath;
9130 
9131 	bool req_immediate_exit = false;
9132 
9133 	/* Forbid vmenter if vcpu dirty ring is soft-full */
9134 	if (unlikely(vcpu->kvm->dirty_ring_size &&
9135 		     kvm_dirty_ring_soft_full(&vcpu->dirty_ring))) {
9136 		vcpu->run->exit_reason = KVM_EXIT_DIRTY_RING_FULL;
9137 		trace_kvm_dirty_ring_exit(vcpu);
9138 		r = 0;
9139 		goto out;
9140 	}
9141 
9142 	if (kvm_request_pending(vcpu)) {
9143 		if (kvm_check_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu)) {
9144 			if (unlikely(!kvm_x86_ops.nested_ops->get_nested_state_pages(vcpu))) {
9145 				r = 0;
9146 				goto out;
9147 			}
9148 		}
9149 		if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
9150 			kvm_mmu_unload(vcpu);
9151 		if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
9152 			__kvm_migrate_timers(vcpu);
9153 		if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
9154 			kvm_gen_update_masterclock(vcpu->kvm);
9155 		if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
9156 			kvm_gen_kvmclock_update(vcpu);
9157 		if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
9158 			r = kvm_guest_time_update(vcpu);
9159 			if (unlikely(r))
9160 				goto out;
9161 		}
9162 		if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
9163 			kvm_mmu_sync_roots(vcpu);
9164 		if (kvm_check_request(KVM_REQ_LOAD_MMU_PGD, vcpu))
9165 			kvm_mmu_load_pgd(vcpu);
9166 		if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) {
9167 			kvm_vcpu_flush_tlb_all(vcpu);
9168 
9169 			/* Flushing all ASIDs flushes the current ASID... */
9170 			kvm_clear_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
9171 		}
9172 		if (kvm_check_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu))
9173 			kvm_vcpu_flush_tlb_current(vcpu);
9174 		if (kvm_check_request(KVM_REQ_HV_TLB_FLUSH, vcpu))
9175 			kvm_vcpu_flush_tlb_guest(vcpu);
9176 
9177 		if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
9178 			vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
9179 			r = 0;
9180 			goto out;
9181 		}
9182 		if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
9183 			if (is_guest_mode(vcpu)) {
9184 				kvm_x86_ops.nested_ops->triple_fault(vcpu);
9185 			} else {
9186 				vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
9187 				vcpu->mmio_needed = 0;
9188 				r = 0;
9189 				goto out;
9190 			}
9191 		}
9192 		if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
9193 			/* Page is swapped out. Do synthetic halt */
9194 			vcpu->arch.apf.halted = true;
9195 			r = 1;
9196 			goto out;
9197 		}
9198 		if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
9199 			record_steal_time(vcpu);
9200 		if (kvm_check_request(KVM_REQ_SMI, vcpu))
9201 			process_smi(vcpu);
9202 		if (kvm_check_request(KVM_REQ_NMI, vcpu))
9203 			process_nmi(vcpu);
9204 		if (kvm_check_request(KVM_REQ_PMU, vcpu))
9205 			kvm_pmu_handle_event(vcpu);
9206 		if (kvm_check_request(KVM_REQ_PMI, vcpu))
9207 			kvm_pmu_deliver_pmi(vcpu);
9208 		if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
9209 			BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
9210 			if (test_bit(vcpu->arch.pending_ioapic_eoi,
9211 				     vcpu->arch.ioapic_handled_vectors)) {
9212 				vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
9213 				vcpu->run->eoi.vector =
9214 						vcpu->arch.pending_ioapic_eoi;
9215 				r = 0;
9216 				goto out;
9217 			}
9218 		}
9219 		if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
9220 			vcpu_scan_ioapic(vcpu);
9221 		if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
9222 			vcpu_load_eoi_exitmap(vcpu);
9223 		if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
9224 			kvm_vcpu_reload_apic_access_page(vcpu);
9225 		if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
9226 			vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
9227 			vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
9228 			r = 0;
9229 			goto out;
9230 		}
9231 		if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
9232 			vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
9233 			vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
9234 			r = 0;
9235 			goto out;
9236 		}
9237 		if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
9238 			struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
9239 
9240 			vcpu->run->exit_reason = KVM_EXIT_HYPERV;
9241 			vcpu->run->hyperv = hv_vcpu->exit;
9242 			r = 0;
9243 			goto out;
9244 		}
9245 
9246 		/*
9247 		 * KVM_REQ_HV_STIMER has to be processed after
9248 		 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
9249 		 * depend on the guest clock being up-to-date
9250 		 */
9251 		if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
9252 			kvm_hv_process_stimers(vcpu);
9253 		if (kvm_check_request(KVM_REQ_APICV_UPDATE, vcpu))
9254 			kvm_vcpu_update_apicv(vcpu);
9255 		if (kvm_check_request(KVM_REQ_APF_READY, vcpu))
9256 			kvm_check_async_pf_completion(vcpu);
9257 		if (kvm_check_request(KVM_REQ_MSR_FILTER_CHANGED, vcpu))
9258 			static_call(kvm_x86_msr_filter_changed)(vcpu);
9259 
9260 		if (kvm_check_request(KVM_REQ_UPDATE_CPU_DIRTY_LOGGING, vcpu))
9261 			static_call(kvm_x86_update_cpu_dirty_logging)(vcpu);
9262 	}
9263 
9264 	if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win ||
9265 	    kvm_xen_has_interrupt(vcpu)) {
9266 		++vcpu->stat.req_event;
9267 		kvm_apic_accept_events(vcpu);
9268 		if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
9269 			r = 1;
9270 			goto out;
9271 		}
9272 
9273 		inject_pending_event(vcpu, &req_immediate_exit);
9274 		if (req_int_win)
9275 			static_call(kvm_x86_enable_irq_window)(vcpu);
9276 
9277 		if (kvm_lapic_enabled(vcpu)) {
9278 			update_cr8_intercept(vcpu);
9279 			kvm_lapic_sync_to_vapic(vcpu);
9280 		}
9281 	}
9282 
9283 	r = kvm_mmu_reload(vcpu);
9284 	if (unlikely(r)) {
9285 		goto cancel_injection;
9286 	}
9287 
9288 	preempt_disable();
9289 
9290 	static_call(kvm_x86_prepare_guest_switch)(vcpu);
9291 
9292 	/*
9293 	 * Disable IRQs before setting IN_GUEST_MODE.  Posted interrupt
9294 	 * IPI are then delayed after guest entry, which ensures that they
9295 	 * result in virtual interrupt delivery.
9296 	 */
9297 	local_irq_disable();
9298 	vcpu->mode = IN_GUEST_MODE;
9299 
9300 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
9301 
9302 	/*
9303 	 * 1) We should set ->mode before checking ->requests.  Please see
9304 	 * the comment in kvm_vcpu_exiting_guest_mode().
9305 	 *
9306 	 * 2) For APICv, we should set ->mode before checking PID.ON. This
9307 	 * pairs with the memory barrier implicit in pi_test_and_set_on
9308 	 * (see vmx_deliver_posted_interrupt).
9309 	 *
9310 	 * 3) This also orders the write to mode from any reads to the page
9311 	 * tables done while the VCPU is running.  Please see the comment
9312 	 * in kvm_flush_remote_tlbs.
9313 	 */
9314 	smp_mb__after_srcu_read_unlock();
9315 
9316 	/*
9317 	 * This handles the case where a posted interrupt was
9318 	 * notified with kvm_vcpu_kick.
9319 	 */
9320 	if (kvm_lapic_enabled(vcpu) && vcpu->arch.apicv_active)
9321 		static_call(kvm_x86_sync_pir_to_irr)(vcpu);
9322 
9323 	if (kvm_vcpu_exit_request(vcpu)) {
9324 		vcpu->mode = OUTSIDE_GUEST_MODE;
9325 		smp_wmb();
9326 		local_irq_enable();
9327 		preempt_enable();
9328 		vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9329 		r = 1;
9330 		goto cancel_injection;
9331 	}
9332 
9333 	if (req_immediate_exit) {
9334 		kvm_make_request(KVM_REQ_EVENT, vcpu);
9335 		static_call(kvm_x86_request_immediate_exit)(vcpu);
9336 	}
9337 
9338 	fpregs_assert_state_consistent();
9339 	if (test_thread_flag(TIF_NEED_FPU_LOAD))
9340 		switch_fpu_return();
9341 
9342 	if (unlikely(vcpu->arch.switch_db_regs)) {
9343 		set_debugreg(0, 7);
9344 		set_debugreg(vcpu->arch.eff_db[0], 0);
9345 		set_debugreg(vcpu->arch.eff_db[1], 1);
9346 		set_debugreg(vcpu->arch.eff_db[2], 2);
9347 		set_debugreg(vcpu->arch.eff_db[3], 3);
9348 		set_debugreg(vcpu->arch.dr6, 6);
9349 		vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
9350 	}
9351 
9352 	for (;;) {
9353 		exit_fastpath = static_call(kvm_x86_run)(vcpu);
9354 		if (likely(exit_fastpath != EXIT_FASTPATH_REENTER_GUEST))
9355 			break;
9356 
9357                 if (unlikely(kvm_vcpu_exit_request(vcpu))) {
9358 			exit_fastpath = EXIT_FASTPATH_EXIT_HANDLED;
9359 			break;
9360 		}
9361 
9362 		if (vcpu->arch.apicv_active)
9363 			static_call(kvm_x86_sync_pir_to_irr)(vcpu);
9364         }
9365 
9366 	/*
9367 	 * Do this here before restoring debug registers on the host.  And
9368 	 * since we do this before handling the vmexit, a DR access vmexit
9369 	 * can (a) read the correct value of the debug registers, (b) set
9370 	 * KVM_DEBUGREG_WONT_EXIT again.
9371 	 */
9372 	if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
9373 		WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
9374 		static_call(kvm_x86_sync_dirty_debug_regs)(vcpu);
9375 		kvm_update_dr0123(vcpu);
9376 		kvm_update_dr7(vcpu);
9377 		vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
9378 	}
9379 
9380 	/*
9381 	 * If the guest has used debug registers, at least dr7
9382 	 * will be disabled while returning to the host.
9383 	 * If we don't have active breakpoints in the host, we don't
9384 	 * care about the messed up debug address registers. But if
9385 	 * we have some of them active, restore the old state.
9386 	 */
9387 	if (hw_breakpoint_active())
9388 		hw_breakpoint_restore();
9389 
9390 	vcpu->arch.last_vmentry_cpu = vcpu->cpu;
9391 	vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
9392 
9393 	vcpu->mode = OUTSIDE_GUEST_MODE;
9394 	smp_wmb();
9395 
9396 	static_call(kvm_x86_handle_exit_irqoff)(vcpu);
9397 
9398 	/*
9399 	 * Consume any pending interrupts, including the possible source of
9400 	 * VM-Exit on SVM and any ticks that occur between VM-Exit and now.
9401 	 * An instruction is required after local_irq_enable() to fully unblock
9402 	 * interrupts on processors that implement an interrupt shadow, the
9403 	 * stat.exits increment will do nicely.
9404 	 */
9405 	kvm_before_interrupt(vcpu);
9406 	local_irq_enable();
9407 	++vcpu->stat.exits;
9408 	local_irq_disable();
9409 	kvm_after_interrupt(vcpu);
9410 
9411 	/*
9412 	 * Wait until after servicing IRQs to account guest time so that any
9413 	 * ticks that occurred while running the guest are properly accounted
9414 	 * to the guest.  Waiting until IRQs are enabled degrades the accuracy
9415 	 * of accounting via context tracking, but the loss of accuracy is
9416 	 * acceptable for all known use cases.
9417 	 */
9418 	vtime_account_guest_exit();
9419 
9420 	if (lapic_in_kernel(vcpu)) {
9421 		s64 delta = vcpu->arch.apic->lapic_timer.advance_expire_delta;
9422 		if (delta != S64_MIN) {
9423 			trace_kvm_wait_lapic_expire(vcpu->vcpu_id, delta);
9424 			vcpu->arch.apic->lapic_timer.advance_expire_delta = S64_MIN;
9425 		}
9426 	}
9427 
9428 	local_irq_enable();
9429 	preempt_enable();
9430 
9431 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9432 
9433 	/*
9434 	 * Profile KVM exit RIPs:
9435 	 */
9436 	if (unlikely(prof_on == KVM_PROFILING)) {
9437 		unsigned long rip = kvm_rip_read(vcpu);
9438 		profile_hit(KVM_PROFILING, (void *)rip);
9439 	}
9440 
9441 	if (unlikely(vcpu->arch.tsc_always_catchup))
9442 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
9443 
9444 	if (vcpu->arch.apic_attention)
9445 		kvm_lapic_sync_from_vapic(vcpu);
9446 
9447 	r = static_call(kvm_x86_handle_exit)(vcpu, exit_fastpath);
9448 	return r;
9449 
9450 cancel_injection:
9451 	if (req_immediate_exit)
9452 		kvm_make_request(KVM_REQ_EVENT, vcpu);
9453 	static_call(kvm_x86_cancel_injection)(vcpu);
9454 	if (unlikely(vcpu->arch.apic_attention))
9455 		kvm_lapic_sync_from_vapic(vcpu);
9456 out:
9457 	return r;
9458 }
9459 
9460 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
9461 {
9462 	if (!kvm_arch_vcpu_runnable(vcpu) &&
9463 	    (!kvm_x86_ops.pre_block || static_call(kvm_x86_pre_block)(vcpu) == 0)) {
9464 		srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9465 		kvm_vcpu_block(vcpu);
9466 		vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9467 
9468 		if (kvm_x86_ops.post_block)
9469 			static_call(kvm_x86_post_block)(vcpu);
9470 
9471 		if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
9472 			return 1;
9473 	}
9474 
9475 	kvm_apic_accept_events(vcpu);
9476 	switch(vcpu->arch.mp_state) {
9477 	case KVM_MP_STATE_HALTED:
9478 	case KVM_MP_STATE_AP_RESET_HOLD:
9479 		vcpu->arch.pv.pv_unhalted = false;
9480 		vcpu->arch.mp_state =
9481 			KVM_MP_STATE_RUNNABLE;
9482 		fallthrough;
9483 	case KVM_MP_STATE_RUNNABLE:
9484 		vcpu->arch.apf.halted = false;
9485 		break;
9486 	case KVM_MP_STATE_INIT_RECEIVED:
9487 		break;
9488 	default:
9489 		return -EINTR;
9490 	}
9491 	return 1;
9492 }
9493 
9494 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
9495 {
9496 	if (is_guest_mode(vcpu))
9497 		kvm_check_nested_events(vcpu);
9498 
9499 	return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
9500 		!vcpu->arch.apf.halted);
9501 }
9502 
9503 static int vcpu_run(struct kvm_vcpu *vcpu)
9504 {
9505 	int r;
9506 	struct kvm *kvm = vcpu->kvm;
9507 
9508 	vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9509 	vcpu->arch.l1tf_flush_l1d = true;
9510 
9511 	for (;;) {
9512 		if (kvm_vcpu_running(vcpu)) {
9513 			r = vcpu_enter_guest(vcpu);
9514 		} else {
9515 			r = vcpu_block(kvm, vcpu);
9516 		}
9517 
9518 		if (r <= 0)
9519 			break;
9520 
9521 		kvm_clear_request(KVM_REQ_UNBLOCK, vcpu);
9522 		if (kvm_cpu_has_pending_timer(vcpu))
9523 			kvm_inject_pending_timer_irqs(vcpu);
9524 
9525 		if (dm_request_for_irq_injection(vcpu) &&
9526 			kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
9527 			r = 0;
9528 			vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
9529 			++vcpu->stat.request_irq_exits;
9530 			break;
9531 		}
9532 
9533 		if (__xfer_to_guest_mode_work_pending()) {
9534 			srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9535 			r = xfer_to_guest_mode_handle_work(vcpu);
9536 			if (r)
9537 				return r;
9538 			vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9539 		}
9540 	}
9541 
9542 	srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9543 
9544 	return r;
9545 }
9546 
9547 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
9548 {
9549 	int r;
9550 
9551 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9552 	r = kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
9553 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
9554 	return r;
9555 }
9556 
9557 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
9558 {
9559 	BUG_ON(!vcpu->arch.pio.count);
9560 
9561 	return complete_emulated_io(vcpu);
9562 }
9563 
9564 /*
9565  * Implements the following, as a state machine:
9566  *
9567  * read:
9568  *   for each fragment
9569  *     for each mmio piece in the fragment
9570  *       write gpa, len
9571  *       exit
9572  *       copy data
9573  *   execute insn
9574  *
9575  * write:
9576  *   for each fragment
9577  *     for each mmio piece in the fragment
9578  *       write gpa, len
9579  *       copy data
9580  *       exit
9581  */
9582 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
9583 {
9584 	struct kvm_run *run = vcpu->run;
9585 	struct kvm_mmio_fragment *frag;
9586 	unsigned len;
9587 
9588 	BUG_ON(!vcpu->mmio_needed);
9589 
9590 	/* Complete previous fragment */
9591 	frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
9592 	len = min(8u, frag->len);
9593 	if (!vcpu->mmio_is_write)
9594 		memcpy(frag->data, run->mmio.data, len);
9595 
9596 	if (frag->len <= 8) {
9597 		/* Switch to the next fragment. */
9598 		frag++;
9599 		vcpu->mmio_cur_fragment++;
9600 	} else {
9601 		/* Go forward to the next mmio piece. */
9602 		frag->data += len;
9603 		frag->gpa += len;
9604 		frag->len -= len;
9605 	}
9606 
9607 	if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
9608 		vcpu->mmio_needed = 0;
9609 
9610 		/* FIXME: return into emulator if single-stepping.  */
9611 		if (vcpu->mmio_is_write)
9612 			return 1;
9613 		vcpu->mmio_read_completed = 1;
9614 		return complete_emulated_io(vcpu);
9615 	}
9616 
9617 	run->exit_reason = KVM_EXIT_MMIO;
9618 	run->mmio.phys_addr = frag->gpa;
9619 	if (vcpu->mmio_is_write)
9620 		memcpy(run->mmio.data, frag->data, min(8u, frag->len));
9621 	run->mmio.len = min(8u, frag->len);
9622 	run->mmio.is_write = vcpu->mmio_is_write;
9623 	vcpu->arch.complete_userspace_io = complete_emulated_mmio;
9624 	return 0;
9625 }
9626 
9627 static void kvm_save_current_fpu(struct fpu *fpu)
9628 {
9629 	/*
9630 	 * If the target FPU state is not resident in the CPU registers, just
9631 	 * memcpy() from current, else save CPU state directly to the target.
9632 	 */
9633 	if (test_thread_flag(TIF_NEED_FPU_LOAD))
9634 		memcpy(&fpu->state, &current->thread.fpu.state,
9635 		       fpu_kernel_xstate_size);
9636 	else
9637 		copy_fpregs_to_fpstate(fpu);
9638 }
9639 
9640 /* Swap (qemu) user FPU context for the guest FPU context. */
9641 static void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
9642 {
9643 	fpregs_lock();
9644 
9645 	kvm_save_current_fpu(vcpu->arch.user_fpu);
9646 
9647 	/*
9648 	 * Guests with protected state can't have it set by the hypervisor,
9649 	 * so skip trying to set it.
9650 	 */
9651 	if (vcpu->arch.guest_fpu)
9652 		/* PKRU is separately restored in kvm_x86_ops.run. */
9653 		__copy_kernel_to_fpregs(&vcpu->arch.guest_fpu->state,
9654 					~XFEATURE_MASK_PKRU);
9655 
9656 	fpregs_mark_activate();
9657 	fpregs_unlock();
9658 
9659 	trace_kvm_fpu(1);
9660 }
9661 
9662 /* When vcpu_run ends, restore user space FPU context. */
9663 static void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
9664 {
9665 	fpregs_lock();
9666 
9667 	/*
9668 	 * Guests with protected state can't have it read by the hypervisor,
9669 	 * so skip trying to save it.
9670 	 */
9671 	if (vcpu->arch.guest_fpu)
9672 		kvm_save_current_fpu(vcpu->arch.guest_fpu);
9673 
9674 	copy_kernel_to_fpregs(&vcpu->arch.user_fpu->state);
9675 
9676 	fpregs_mark_activate();
9677 	fpregs_unlock();
9678 
9679 	++vcpu->stat.fpu_reload;
9680 	trace_kvm_fpu(0);
9681 }
9682 
9683 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
9684 {
9685 	struct kvm_run *kvm_run = vcpu->run;
9686 	int r;
9687 
9688 	vcpu_load(vcpu);
9689 	kvm_sigset_activate(vcpu);
9690 	kvm_run->flags = 0;
9691 	kvm_load_guest_fpu(vcpu);
9692 
9693 	if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
9694 		if (kvm_run->immediate_exit) {
9695 			r = -EINTR;
9696 			goto out;
9697 		}
9698 		kvm_vcpu_block(vcpu);
9699 		kvm_apic_accept_events(vcpu);
9700 		kvm_clear_request(KVM_REQ_UNHALT, vcpu);
9701 		r = -EAGAIN;
9702 		if (signal_pending(current)) {
9703 			r = -EINTR;
9704 			kvm_run->exit_reason = KVM_EXIT_INTR;
9705 			++vcpu->stat.signal_exits;
9706 		}
9707 		goto out;
9708 	}
9709 
9710 	if (kvm_run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) {
9711 		r = -EINVAL;
9712 		goto out;
9713 	}
9714 
9715 	if (kvm_run->kvm_dirty_regs) {
9716 		r = sync_regs(vcpu);
9717 		if (r != 0)
9718 			goto out;
9719 	}
9720 
9721 	/* re-sync apic's tpr */
9722 	if (!lapic_in_kernel(vcpu)) {
9723 		if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
9724 			r = -EINVAL;
9725 			goto out;
9726 		}
9727 	}
9728 
9729 	if (unlikely(vcpu->arch.complete_userspace_io)) {
9730 		int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
9731 		vcpu->arch.complete_userspace_io = NULL;
9732 		r = cui(vcpu);
9733 		if (r <= 0)
9734 			goto out;
9735 	} else
9736 		WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
9737 
9738 	if (kvm_run->immediate_exit)
9739 		r = -EINTR;
9740 	else
9741 		r = vcpu_run(vcpu);
9742 
9743 out:
9744 	kvm_put_guest_fpu(vcpu);
9745 	if (kvm_run->kvm_valid_regs)
9746 		store_regs(vcpu);
9747 	post_kvm_run_save(vcpu);
9748 	kvm_sigset_deactivate(vcpu);
9749 
9750 	vcpu_put(vcpu);
9751 	return r;
9752 }
9753 
9754 static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9755 {
9756 	if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
9757 		/*
9758 		 * We are here if userspace calls get_regs() in the middle of
9759 		 * instruction emulation. Registers state needs to be copied
9760 		 * back from emulation context to vcpu. Userspace shouldn't do
9761 		 * that usually, but some bad designed PV devices (vmware
9762 		 * backdoor interface) need this to work
9763 		 */
9764 		emulator_writeback_register_cache(vcpu->arch.emulate_ctxt);
9765 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
9766 	}
9767 	regs->rax = kvm_rax_read(vcpu);
9768 	regs->rbx = kvm_rbx_read(vcpu);
9769 	regs->rcx = kvm_rcx_read(vcpu);
9770 	regs->rdx = kvm_rdx_read(vcpu);
9771 	regs->rsi = kvm_rsi_read(vcpu);
9772 	regs->rdi = kvm_rdi_read(vcpu);
9773 	regs->rsp = kvm_rsp_read(vcpu);
9774 	regs->rbp = kvm_rbp_read(vcpu);
9775 #ifdef CONFIG_X86_64
9776 	regs->r8 = kvm_r8_read(vcpu);
9777 	regs->r9 = kvm_r9_read(vcpu);
9778 	regs->r10 = kvm_r10_read(vcpu);
9779 	regs->r11 = kvm_r11_read(vcpu);
9780 	regs->r12 = kvm_r12_read(vcpu);
9781 	regs->r13 = kvm_r13_read(vcpu);
9782 	regs->r14 = kvm_r14_read(vcpu);
9783 	regs->r15 = kvm_r15_read(vcpu);
9784 #endif
9785 
9786 	regs->rip = kvm_rip_read(vcpu);
9787 	regs->rflags = kvm_get_rflags(vcpu);
9788 }
9789 
9790 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9791 {
9792 	vcpu_load(vcpu);
9793 	__get_regs(vcpu, regs);
9794 	vcpu_put(vcpu);
9795 	return 0;
9796 }
9797 
9798 static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9799 {
9800 	vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
9801 	vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
9802 
9803 	kvm_rax_write(vcpu, regs->rax);
9804 	kvm_rbx_write(vcpu, regs->rbx);
9805 	kvm_rcx_write(vcpu, regs->rcx);
9806 	kvm_rdx_write(vcpu, regs->rdx);
9807 	kvm_rsi_write(vcpu, regs->rsi);
9808 	kvm_rdi_write(vcpu, regs->rdi);
9809 	kvm_rsp_write(vcpu, regs->rsp);
9810 	kvm_rbp_write(vcpu, regs->rbp);
9811 #ifdef CONFIG_X86_64
9812 	kvm_r8_write(vcpu, regs->r8);
9813 	kvm_r9_write(vcpu, regs->r9);
9814 	kvm_r10_write(vcpu, regs->r10);
9815 	kvm_r11_write(vcpu, regs->r11);
9816 	kvm_r12_write(vcpu, regs->r12);
9817 	kvm_r13_write(vcpu, regs->r13);
9818 	kvm_r14_write(vcpu, regs->r14);
9819 	kvm_r15_write(vcpu, regs->r15);
9820 #endif
9821 
9822 	kvm_rip_write(vcpu, regs->rip);
9823 	kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
9824 
9825 	vcpu->arch.exception.pending = false;
9826 
9827 	kvm_make_request(KVM_REQ_EVENT, vcpu);
9828 }
9829 
9830 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9831 {
9832 	vcpu_load(vcpu);
9833 	__set_regs(vcpu, regs);
9834 	vcpu_put(vcpu);
9835 	return 0;
9836 }
9837 
9838 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
9839 {
9840 	struct kvm_segment cs;
9841 
9842 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
9843 	*db = cs.db;
9844 	*l = cs.l;
9845 }
9846 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
9847 
9848 static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
9849 {
9850 	struct desc_ptr dt;
9851 
9852 	if (vcpu->arch.guest_state_protected)
9853 		goto skip_protected_regs;
9854 
9855 	kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
9856 	kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
9857 	kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
9858 	kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
9859 	kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
9860 	kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
9861 
9862 	kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
9863 	kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
9864 
9865 	static_call(kvm_x86_get_idt)(vcpu, &dt);
9866 	sregs->idt.limit = dt.size;
9867 	sregs->idt.base = dt.address;
9868 	static_call(kvm_x86_get_gdt)(vcpu, &dt);
9869 	sregs->gdt.limit = dt.size;
9870 	sregs->gdt.base = dt.address;
9871 
9872 	sregs->cr2 = vcpu->arch.cr2;
9873 	sregs->cr3 = kvm_read_cr3(vcpu);
9874 
9875 skip_protected_regs:
9876 	sregs->cr0 = kvm_read_cr0(vcpu);
9877 	sregs->cr4 = kvm_read_cr4(vcpu);
9878 	sregs->cr8 = kvm_get_cr8(vcpu);
9879 	sregs->efer = vcpu->arch.efer;
9880 	sregs->apic_base = kvm_get_apic_base(vcpu);
9881 
9882 	memset(sregs->interrupt_bitmap, 0, sizeof(sregs->interrupt_bitmap));
9883 
9884 	if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft)
9885 		set_bit(vcpu->arch.interrupt.nr,
9886 			(unsigned long *)sregs->interrupt_bitmap);
9887 }
9888 
9889 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
9890 				  struct kvm_sregs *sregs)
9891 {
9892 	vcpu_load(vcpu);
9893 	__get_sregs(vcpu, sregs);
9894 	vcpu_put(vcpu);
9895 	return 0;
9896 }
9897 
9898 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
9899 				    struct kvm_mp_state *mp_state)
9900 {
9901 	vcpu_load(vcpu);
9902 	if (kvm_mpx_supported())
9903 		kvm_load_guest_fpu(vcpu);
9904 
9905 	kvm_apic_accept_events(vcpu);
9906 	if ((vcpu->arch.mp_state == KVM_MP_STATE_HALTED ||
9907 	     vcpu->arch.mp_state == KVM_MP_STATE_AP_RESET_HOLD) &&
9908 	    vcpu->arch.pv.pv_unhalted)
9909 		mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
9910 	else
9911 		mp_state->mp_state = vcpu->arch.mp_state;
9912 
9913 	if (kvm_mpx_supported())
9914 		kvm_put_guest_fpu(vcpu);
9915 	vcpu_put(vcpu);
9916 	return 0;
9917 }
9918 
9919 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
9920 				    struct kvm_mp_state *mp_state)
9921 {
9922 	int ret = -EINVAL;
9923 
9924 	vcpu_load(vcpu);
9925 
9926 	if (!lapic_in_kernel(vcpu) &&
9927 	    mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
9928 		goto out;
9929 
9930 	/*
9931 	 * KVM_MP_STATE_INIT_RECEIVED means the processor is in
9932 	 * INIT state; latched init should be reported using
9933 	 * KVM_SET_VCPU_EVENTS, so reject it here.
9934 	 */
9935 	if ((kvm_vcpu_latch_init(vcpu) || vcpu->arch.smi_pending) &&
9936 	    (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
9937 	     mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
9938 		goto out;
9939 
9940 	if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
9941 		vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
9942 		set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
9943 	} else
9944 		vcpu->arch.mp_state = mp_state->mp_state;
9945 	kvm_make_request(KVM_REQ_EVENT, vcpu);
9946 
9947 	ret = 0;
9948 out:
9949 	vcpu_put(vcpu);
9950 	return ret;
9951 }
9952 
9953 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
9954 		    int reason, bool has_error_code, u32 error_code)
9955 {
9956 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
9957 	int ret;
9958 
9959 	init_emulate_ctxt(vcpu);
9960 
9961 	ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
9962 				   has_error_code, error_code);
9963 	if (ret) {
9964 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
9965 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
9966 		vcpu->run->internal.ndata = 0;
9967 		return 0;
9968 	}
9969 
9970 	kvm_rip_write(vcpu, ctxt->eip);
9971 	kvm_set_rflags(vcpu, ctxt->eflags);
9972 	return 1;
9973 }
9974 EXPORT_SYMBOL_GPL(kvm_task_switch);
9975 
9976 static bool kvm_is_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
9977 {
9978 	if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
9979 		/*
9980 		 * When EFER.LME and CR0.PG are set, the processor is in
9981 		 * 64-bit mode (though maybe in a 32-bit code segment).
9982 		 * CR4.PAE and EFER.LMA must be set.
9983 		 */
9984 		if (!(sregs->cr4 & X86_CR4_PAE) || !(sregs->efer & EFER_LMA))
9985 			return false;
9986 		if (kvm_vcpu_is_illegal_gpa(vcpu, sregs->cr3))
9987 			return false;
9988 	} else {
9989 		/*
9990 		 * Not in 64-bit mode: EFER.LMA is clear and the code
9991 		 * segment cannot be 64-bit.
9992 		 */
9993 		if (sregs->efer & EFER_LMA || sregs->cs.l)
9994 			return false;
9995 	}
9996 
9997 	return kvm_is_valid_cr4(vcpu, sregs->cr4);
9998 }
9999 
10000 static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
10001 {
10002 	struct msr_data apic_base_msr;
10003 	int mmu_reset_needed = 0;
10004 	int pending_vec, max_bits, idx;
10005 	struct desc_ptr dt;
10006 	int ret = -EINVAL;
10007 
10008 	if (!kvm_is_valid_sregs(vcpu, sregs))
10009 		goto out;
10010 
10011 	apic_base_msr.data = sregs->apic_base;
10012 	apic_base_msr.host_initiated = true;
10013 	if (kvm_set_apic_base(vcpu, &apic_base_msr))
10014 		goto out;
10015 
10016 	if (vcpu->arch.guest_state_protected)
10017 		goto skip_protected_regs;
10018 
10019 	dt.size = sregs->idt.limit;
10020 	dt.address = sregs->idt.base;
10021 	static_call(kvm_x86_set_idt)(vcpu, &dt);
10022 	dt.size = sregs->gdt.limit;
10023 	dt.address = sregs->gdt.base;
10024 	static_call(kvm_x86_set_gdt)(vcpu, &dt);
10025 
10026 	vcpu->arch.cr2 = sregs->cr2;
10027 	mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
10028 	vcpu->arch.cr3 = sregs->cr3;
10029 	kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
10030 
10031 	kvm_set_cr8(vcpu, sregs->cr8);
10032 
10033 	mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
10034 	static_call(kvm_x86_set_efer)(vcpu, sregs->efer);
10035 
10036 	mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
10037 	static_call(kvm_x86_set_cr0)(vcpu, sregs->cr0);
10038 	vcpu->arch.cr0 = sregs->cr0;
10039 
10040 	mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
10041 	static_call(kvm_x86_set_cr4)(vcpu, sregs->cr4);
10042 
10043 	idx = srcu_read_lock(&vcpu->kvm->srcu);
10044 	if (is_pae_paging(vcpu)) {
10045 		load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
10046 		mmu_reset_needed = 1;
10047 	}
10048 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
10049 
10050 	if (mmu_reset_needed)
10051 		kvm_mmu_reset_context(vcpu);
10052 
10053 	kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
10054 	kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
10055 	kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
10056 	kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
10057 	kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
10058 	kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
10059 
10060 	kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
10061 	kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
10062 
10063 	update_cr8_intercept(vcpu);
10064 
10065 	/* Older userspace won't unhalt the vcpu on reset. */
10066 	if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
10067 	    sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
10068 	    !is_protmode(vcpu))
10069 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10070 
10071 skip_protected_regs:
10072 	max_bits = KVM_NR_INTERRUPTS;
10073 	pending_vec = find_first_bit(
10074 		(const unsigned long *)sregs->interrupt_bitmap, max_bits);
10075 	if (pending_vec < max_bits) {
10076 		kvm_queue_interrupt(vcpu, pending_vec, false);
10077 		pr_debug("Set back pending irq %d\n", pending_vec);
10078 	}
10079 
10080 	kvm_make_request(KVM_REQ_EVENT, vcpu);
10081 
10082 	ret = 0;
10083 out:
10084 	return ret;
10085 }
10086 
10087 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
10088 				  struct kvm_sregs *sregs)
10089 {
10090 	int ret;
10091 
10092 	vcpu_load(vcpu);
10093 	ret = __set_sregs(vcpu, sregs);
10094 	vcpu_put(vcpu);
10095 	return ret;
10096 }
10097 
10098 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
10099 					struct kvm_guest_debug *dbg)
10100 {
10101 	unsigned long rflags;
10102 	int i, r;
10103 
10104 	if (vcpu->arch.guest_state_protected)
10105 		return -EINVAL;
10106 
10107 	vcpu_load(vcpu);
10108 
10109 	if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
10110 		r = -EBUSY;
10111 		if (vcpu->arch.exception.pending)
10112 			goto out;
10113 		if (dbg->control & KVM_GUESTDBG_INJECT_DB)
10114 			kvm_queue_exception(vcpu, DB_VECTOR);
10115 		else
10116 			kvm_queue_exception(vcpu, BP_VECTOR);
10117 	}
10118 
10119 	/*
10120 	 * Read rflags as long as potentially injected trace flags are still
10121 	 * filtered out.
10122 	 */
10123 	rflags = kvm_get_rflags(vcpu);
10124 
10125 	vcpu->guest_debug = dbg->control;
10126 	if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
10127 		vcpu->guest_debug = 0;
10128 
10129 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
10130 		for (i = 0; i < KVM_NR_DB_REGS; ++i)
10131 			vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
10132 		vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
10133 	} else {
10134 		for (i = 0; i < KVM_NR_DB_REGS; i++)
10135 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
10136 	}
10137 	kvm_update_dr7(vcpu);
10138 
10139 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
10140 		vcpu->arch.singlestep_rip = kvm_get_linear_rip(vcpu);
10141 
10142 	/*
10143 	 * Trigger an rflags update that will inject or remove the trace
10144 	 * flags.
10145 	 */
10146 	kvm_set_rflags(vcpu, rflags);
10147 
10148 	static_call(kvm_x86_update_exception_bitmap)(vcpu);
10149 
10150 	r = 0;
10151 
10152 out:
10153 	vcpu_put(vcpu);
10154 	return r;
10155 }
10156 
10157 /*
10158  * Translate a guest virtual address to a guest physical address.
10159  */
10160 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
10161 				    struct kvm_translation *tr)
10162 {
10163 	unsigned long vaddr = tr->linear_address;
10164 	gpa_t gpa;
10165 	int idx;
10166 
10167 	vcpu_load(vcpu);
10168 
10169 	idx = srcu_read_lock(&vcpu->kvm->srcu);
10170 	gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
10171 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
10172 	tr->physical_address = gpa;
10173 	tr->valid = gpa != UNMAPPED_GVA;
10174 	tr->writeable = 1;
10175 	tr->usermode = 0;
10176 
10177 	vcpu_put(vcpu);
10178 	return 0;
10179 }
10180 
10181 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
10182 {
10183 	struct fxregs_state *fxsave;
10184 
10185 	if (!vcpu->arch.guest_fpu)
10186 		return 0;
10187 
10188 	vcpu_load(vcpu);
10189 
10190 	fxsave = &vcpu->arch.guest_fpu->state.fxsave;
10191 	memcpy(fpu->fpr, fxsave->st_space, 128);
10192 	fpu->fcw = fxsave->cwd;
10193 	fpu->fsw = fxsave->swd;
10194 	fpu->ftwx = fxsave->twd;
10195 	fpu->last_opcode = fxsave->fop;
10196 	fpu->last_ip = fxsave->rip;
10197 	fpu->last_dp = fxsave->rdp;
10198 	memcpy(fpu->xmm, fxsave->xmm_space, sizeof(fxsave->xmm_space));
10199 
10200 	vcpu_put(vcpu);
10201 	return 0;
10202 }
10203 
10204 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
10205 {
10206 	struct fxregs_state *fxsave;
10207 
10208 	if (!vcpu->arch.guest_fpu)
10209 		return 0;
10210 
10211 	vcpu_load(vcpu);
10212 
10213 	fxsave = &vcpu->arch.guest_fpu->state.fxsave;
10214 
10215 	memcpy(fxsave->st_space, fpu->fpr, 128);
10216 	fxsave->cwd = fpu->fcw;
10217 	fxsave->swd = fpu->fsw;
10218 	fxsave->twd = fpu->ftwx;
10219 	fxsave->fop = fpu->last_opcode;
10220 	fxsave->rip = fpu->last_ip;
10221 	fxsave->rdp = fpu->last_dp;
10222 	memcpy(fxsave->xmm_space, fpu->xmm, sizeof(fxsave->xmm_space));
10223 
10224 	vcpu_put(vcpu);
10225 	return 0;
10226 }
10227 
10228 static void store_regs(struct kvm_vcpu *vcpu)
10229 {
10230 	BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
10231 
10232 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
10233 		__get_regs(vcpu, &vcpu->run->s.regs.regs);
10234 
10235 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
10236 		__get_sregs(vcpu, &vcpu->run->s.regs.sregs);
10237 
10238 	if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
10239 		kvm_vcpu_ioctl_x86_get_vcpu_events(
10240 				vcpu, &vcpu->run->s.regs.events);
10241 }
10242 
10243 static int sync_regs(struct kvm_vcpu *vcpu)
10244 {
10245 	if (vcpu->run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS)
10246 		return -EINVAL;
10247 
10248 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
10249 		__set_regs(vcpu, &vcpu->run->s.regs.regs);
10250 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
10251 	}
10252 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
10253 		if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs))
10254 			return -EINVAL;
10255 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
10256 	}
10257 	if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
10258 		if (kvm_vcpu_ioctl_x86_set_vcpu_events(
10259 				vcpu, &vcpu->run->s.regs.events))
10260 			return -EINVAL;
10261 		vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS;
10262 	}
10263 
10264 	return 0;
10265 }
10266 
10267 static void fx_init(struct kvm_vcpu *vcpu)
10268 {
10269 	if (!vcpu->arch.guest_fpu)
10270 		return;
10271 
10272 	fpstate_init(&vcpu->arch.guest_fpu->state);
10273 	if (boot_cpu_has(X86_FEATURE_XSAVES))
10274 		vcpu->arch.guest_fpu->state.xsave.header.xcomp_bv =
10275 			host_xcr0 | XSTATE_COMPACTION_ENABLED;
10276 
10277 	/*
10278 	 * Ensure guest xcr0 is valid for loading
10279 	 */
10280 	vcpu->arch.xcr0 = XFEATURE_MASK_FP;
10281 
10282 	vcpu->arch.cr0 |= X86_CR0_ET;
10283 }
10284 
10285 void kvm_free_guest_fpu(struct kvm_vcpu *vcpu)
10286 {
10287 	if (vcpu->arch.guest_fpu) {
10288 		kmem_cache_free(x86_fpu_cache, vcpu->arch.guest_fpu);
10289 		vcpu->arch.guest_fpu = NULL;
10290 	}
10291 }
10292 EXPORT_SYMBOL_GPL(kvm_free_guest_fpu);
10293 
10294 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
10295 {
10296 	if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
10297 		pr_warn_once("kvm: SMP vm created on host with unstable TSC; "
10298 			     "guest TSC will not be reliable\n");
10299 
10300 	return 0;
10301 }
10302 
10303 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
10304 {
10305 	struct page *page;
10306 	int r;
10307 
10308 	if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
10309 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10310 	else
10311 		vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
10312 
10313 	kvm_set_tsc_khz(vcpu, max_tsc_khz);
10314 
10315 	r = kvm_mmu_create(vcpu);
10316 	if (r < 0)
10317 		return r;
10318 
10319 	if (irqchip_in_kernel(vcpu->kvm)) {
10320 		r = kvm_create_lapic(vcpu, lapic_timer_advance_ns);
10321 		if (r < 0)
10322 			goto fail_mmu_destroy;
10323 		if (kvm_apicv_activated(vcpu->kvm))
10324 			vcpu->arch.apicv_active = true;
10325 	} else
10326 		static_branch_inc(&kvm_has_noapic_vcpu);
10327 
10328 	r = -ENOMEM;
10329 
10330 	page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
10331 	if (!page)
10332 		goto fail_free_lapic;
10333 	vcpu->arch.pio_data = page_address(page);
10334 
10335 	vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
10336 				       GFP_KERNEL_ACCOUNT);
10337 	if (!vcpu->arch.mce_banks)
10338 		goto fail_free_pio_data;
10339 	vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
10340 
10341 	if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask,
10342 				GFP_KERNEL_ACCOUNT))
10343 		goto fail_free_mce_banks;
10344 
10345 	if (!alloc_emulate_ctxt(vcpu))
10346 		goto free_wbinvd_dirty_mask;
10347 
10348 	vcpu->arch.user_fpu = kmem_cache_zalloc(x86_fpu_cache,
10349 						GFP_KERNEL_ACCOUNT);
10350 	if (!vcpu->arch.user_fpu) {
10351 		pr_err("kvm: failed to allocate userspace's fpu\n");
10352 		goto free_emulate_ctxt;
10353 	}
10354 
10355 	vcpu->arch.guest_fpu = kmem_cache_zalloc(x86_fpu_cache,
10356 						 GFP_KERNEL_ACCOUNT);
10357 	if (!vcpu->arch.guest_fpu) {
10358 		pr_err("kvm: failed to allocate vcpu's fpu\n");
10359 		goto free_user_fpu;
10360 	}
10361 	fx_init(vcpu);
10362 
10363 	vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
10364 	vcpu->arch.reserved_gpa_bits = kvm_vcpu_reserved_gpa_bits_raw(vcpu);
10365 
10366 	vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
10367 
10368 	kvm_async_pf_hash_reset(vcpu);
10369 	kvm_pmu_init(vcpu);
10370 
10371 	vcpu->arch.pending_external_vector = -1;
10372 	vcpu->arch.preempted_in_kernel = false;
10373 
10374 	r = static_call(kvm_x86_vcpu_create)(vcpu);
10375 	if (r)
10376 		goto free_guest_fpu;
10377 
10378 	vcpu->arch.arch_capabilities = kvm_get_arch_capabilities();
10379 	vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
10380 	kvm_vcpu_mtrr_init(vcpu);
10381 	vcpu_load(vcpu);
10382 	kvm_vcpu_reset(vcpu, false);
10383 	kvm_init_mmu(vcpu, false);
10384 	vcpu_put(vcpu);
10385 	return 0;
10386 
10387 free_guest_fpu:
10388 	kvm_free_guest_fpu(vcpu);
10389 free_user_fpu:
10390 	kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu);
10391 free_emulate_ctxt:
10392 	kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
10393 free_wbinvd_dirty_mask:
10394 	free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
10395 fail_free_mce_banks:
10396 	kfree(vcpu->arch.mce_banks);
10397 fail_free_pio_data:
10398 	free_page((unsigned long)vcpu->arch.pio_data);
10399 fail_free_lapic:
10400 	kvm_free_lapic(vcpu);
10401 fail_mmu_destroy:
10402 	kvm_mmu_destroy(vcpu);
10403 	return r;
10404 }
10405 
10406 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
10407 {
10408 	struct kvm *kvm = vcpu->kvm;
10409 
10410 	if (mutex_lock_killable(&vcpu->mutex))
10411 		return;
10412 	vcpu_load(vcpu);
10413 	kvm_synchronize_tsc(vcpu, 0);
10414 	vcpu_put(vcpu);
10415 
10416 	/* poll control enabled by default */
10417 	vcpu->arch.msr_kvm_poll_control = 1;
10418 
10419 	mutex_unlock(&vcpu->mutex);
10420 
10421 	if (kvmclock_periodic_sync && vcpu->vcpu_idx == 0)
10422 		schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
10423 						KVMCLOCK_SYNC_PERIOD);
10424 }
10425 
10426 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
10427 {
10428 	struct gfn_to_pfn_cache *cache = &vcpu->arch.st.cache;
10429 	int idx;
10430 
10431 	kvm_release_pfn(cache->pfn, cache->dirty, cache);
10432 
10433 	kvmclock_reset(vcpu);
10434 
10435 	static_call(kvm_x86_vcpu_free)(vcpu);
10436 
10437 	kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
10438 	free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
10439 	kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu);
10440 	kvm_free_guest_fpu(vcpu);
10441 
10442 	kvm_hv_vcpu_uninit(vcpu);
10443 	kvm_pmu_destroy(vcpu);
10444 	kfree(vcpu->arch.mce_banks);
10445 	kvm_free_lapic(vcpu);
10446 	idx = srcu_read_lock(&vcpu->kvm->srcu);
10447 	kvm_mmu_destroy(vcpu);
10448 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
10449 	free_page((unsigned long)vcpu->arch.pio_data);
10450 	kvfree(vcpu->arch.cpuid_entries);
10451 	if (!lapic_in_kernel(vcpu))
10452 		static_branch_dec(&kvm_has_noapic_vcpu);
10453 }
10454 
10455 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
10456 {
10457 	kvm_lapic_reset(vcpu, init_event);
10458 
10459 	vcpu->arch.hflags = 0;
10460 
10461 	vcpu->arch.smi_pending = 0;
10462 	vcpu->arch.smi_count = 0;
10463 	atomic_set(&vcpu->arch.nmi_queued, 0);
10464 	vcpu->arch.nmi_pending = 0;
10465 	vcpu->arch.nmi_injected = false;
10466 	kvm_clear_interrupt_queue(vcpu);
10467 	kvm_clear_exception_queue(vcpu);
10468 
10469 	memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
10470 	kvm_update_dr0123(vcpu);
10471 	vcpu->arch.dr6 = DR6_ACTIVE_LOW;
10472 	vcpu->arch.dr7 = DR7_FIXED_1;
10473 	kvm_update_dr7(vcpu);
10474 
10475 	vcpu->arch.cr2 = 0;
10476 
10477 	kvm_make_request(KVM_REQ_EVENT, vcpu);
10478 	vcpu->arch.apf.msr_en_val = 0;
10479 	vcpu->arch.apf.msr_int_val = 0;
10480 	vcpu->arch.st.msr_val = 0;
10481 
10482 	kvmclock_reset(vcpu);
10483 
10484 	kvm_clear_async_pf_completion_queue(vcpu);
10485 	kvm_async_pf_hash_reset(vcpu);
10486 	vcpu->arch.apf.halted = false;
10487 
10488 	if (vcpu->arch.guest_fpu && kvm_mpx_supported()) {
10489 		void *mpx_state_buffer;
10490 
10491 		/*
10492 		 * To avoid have the INIT path from kvm_apic_has_events() that be
10493 		 * called with loaded FPU and does not let userspace fix the state.
10494 		 */
10495 		if (init_event)
10496 			kvm_put_guest_fpu(vcpu);
10497 		mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
10498 					XFEATURE_BNDREGS);
10499 		if (mpx_state_buffer)
10500 			memset(mpx_state_buffer, 0, sizeof(struct mpx_bndreg_state));
10501 		mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
10502 					XFEATURE_BNDCSR);
10503 		if (mpx_state_buffer)
10504 			memset(mpx_state_buffer, 0, sizeof(struct mpx_bndcsr));
10505 		if (init_event)
10506 			kvm_load_guest_fpu(vcpu);
10507 	}
10508 
10509 	if (!init_event) {
10510 		kvm_pmu_reset(vcpu);
10511 		vcpu->arch.smbase = 0x30000;
10512 
10513 		vcpu->arch.msr_misc_features_enables = 0;
10514 
10515 		vcpu->arch.xcr0 = XFEATURE_MASK_FP;
10516 	}
10517 
10518 	memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
10519 	vcpu->arch.regs_avail = ~0;
10520 	vcpu->arch.regs_dirty = ~0;
10521 
10522 	vcpu->arch.ia32_xss = 0;
10523 
10524 	static_call(kvm_x86_vcpu_reset)(vcpu, init_event);
10525 }
10526 
10527 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
10528 {
10529 	struct kvm_segment cs;
10530 
10531 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
10532 	cs.selector = vector << 8;
10533 	cs.base = vector << 12;
10534 	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
10535 	kvm_rip_write(vcpu, 0);
10536 }
10537 EXPORT_SYMBOL_GPL(kvm_vcpu_deliver_sipi_vector);
10538 
10539 int kvm_arch_hardware_enable(void)
10540 {
10541 	struct kvm *kvm;
10542 	struct kvm_vcpu *vcpu;
10543 	int i;
10544 	int ret;
10545 	u64 local_tsc;
10546 	u64 max_tsc = 0;
10547 	bool stable, backwards_tsc = false;
10548 
10549 	kvm_user_return_msr_cpu_online();
10550 	ret = static_call(kvm_x86_hardware_enable)();
10551 	if (ret != 0)
10552 		return ret;
10553 
10554 	local_tsc = rdtsc();
10555 	stable = !kvm_check_tsc_unstable();
10556 	list_for_each_entry(kvm, &vm_list, vm_list) {
10557 		kvm_for_each_vcpu(i, vcpu, kvm) {
10558 			if (!stable && vcpu->cpu == smp_processor_id())
10559 				kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
10560 			if (stable && vcpu->arch.last_host_tsc > local_tsc) {
10561 				backwards_tsc = true;
10562 				if (vcpu->arch.last_host_tsc > max_tsc)
10563 					max_tsc = vcpu->arch.last_host_tsc;
10564 			}
10565 		}
10566 	}
10567 
10568 	/*
10569 	 * Sometimes, even reliable TSCs go backwards.  This happens on
10570 	 * platforms that reset TSC during suspend or hibernate actions, but
10571 	 * maintain synchronization.  We must compensate.  Fortunately, we can
10572 	 * detect that condition here, which happens early in CPU bringup,
10573 	 * before any KVM threads can be running.  Unfortunately, we can't
10574 	 * bring the TSCs fully up to date with real time, as we aren't yet far
10575 	 * enough into CPU bringup that we know how much real time has actually
10576 	 * elapsed; our helper function, ktime_get_boottime_ns() will be using boot
10577 	 * variables that haven't been updated yet.
10578 	 *
10579 	 * So we simply find the maximum observed TSC above, then record the
10580 	 * adjustment to TSC in each VCPU.  When the VCPU later gets loaded,
10581 	 * the adjustment will be applied.  Note that we accumulate
10582 	 * adjustments, in case multiple suspend cycles happen before some VCPU
10583 	 * gets a chance to run again.  In the event that no KVM threads get a
10584 	 * chance to run, we will miss the entire elapsed period, as we'll have
10585 	 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
10586 	 * loose cycle time.  This isn't too big a deal, since the loss will be
10587 	 * uniform across all VCPUs (not to mention the scenario is extremely
10588 	 * unlikely). It is possible that a second hibernate recovery happens
10589 	 * much faster than a first, causing the observed TSC here to be
10590 	 * smaller; this would require additional padding adjustment, which is
10591 	 * why we set last_host_tsc to the local tsc observed here.
10592 	 *
10593 	 * N.B. - this code below runs only on platforms with reliable TSC,
10594 	 * as that is the only way backwards_tsc is set above.  Also note
10595 	 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
10596 	 * have the same delta_cyc adjustment applied if backwards_tsc
10597 	 * is detected.  Note further, this adjustment is only done once,
10598 	 * as we reset last_host_tsc on all VCPUs to stop this from being
10599 	 * called multiple times (one for each physical CPU bringup).
10600 	 *
10601 	 * Platforms with unreliable TSCs don't have to deal with this, they
10602 	 * will be compensated by the logic in vcpu_load, which sets the TSC to
10603 	 * catchup mode.  This will catchup all VCPUs to real time, but cannot
10604 	 * guarantee that they stay in perfect synchronization.
10605 	 */
10606 	if (backwards_tsc) {
10607 		u64 delta_cyc = max_tsc - local_tsc;
10608 		list_for_each_entry(kvm, &vm_list, vm_list) {
10609 			kvm->arch.backwards_tsc_observed = true;
10610 			kvm_for_each_vcpu(i, vcpu, kvm) {
10611 				vcpu->arch.tsc_offset_adjustment += delta_cyc;
10612 				vcpu->arch.last_host_tsc = local_tsc;
10613 				kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
10614 			}
10615 
10616 			/*
10617 			 * We have to disable TSC offset matching.. if you were
10618 			 * booting a VM while issuing an S4 host suspend....
10619 			 * you may have some problem.  Solving this issue is
10620 			 * left as an exercise to the reader.
10621 			 */
10622 			kvm->arch.last_tsc_nsec = 0;
10623 			kvm->arch.last_tsc_write = 0;
10624 		}
10625 
10626 	}
10627 	return 0;
10628 }
10629 
10630 void kvm_arch_hardware_disable(void)
10631 {
10632 	static_call(kvm_x86_hardware_disable)();
10633 	drop_user_return_notifiers();
10634 }
10635 
10636 int kvm_arch_hardware_setup(void *opaque)
10637 {
10638 	struct kvm_x86_init_ops *ops = opaque;
10639 	int r;
10640 
10641 	rdmsrl_safe(MSR_EFER, &host_efer);
10642 
10643 	if (boot_cpu_has(X86_FEATURE_XSAVES))
10644 		rdmsrl(MSR_IA32_XSS, host_xss);
10645 
10646 	r = ops->hardware_setup();
10647 	if (r != 0)
10648 		return r;
10649 
10650 	memcpy(&kvm_x86_ops, ops->runtime_ops, sizeof(kvm_x86_ops));
10651 	kvm_ops_static_call_update();
10652 
10653 	if (!kvm_cpu_cap_has(X86_FEATURE_XSAVES))
10654 		supported_xss = 0;
10655 
10656 #define __kvm_cpu_cap_has(UNUSED_, f) kvm_cpu_cap_has(f)
10657 	cr4_reserved_bits = __cr4_reserved_bits(__kvm_cpu_cap_has, UNUSED_);
10658 #undef __kvm_cpu_cap_has
10659 
10660 	if (kvm_has_tsc_control) {
10661 		/*
10662 		 * Make sure the user can only configure tsc_khz values that
10663 		 * fit into a signed integer.
10664 		 * A min value is not calculated because it will always
10665 		 * be 1 on all machines.
10666 		 */
10667 		u64 max = min(0x7fffffffULL,
10668 			      __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
10669 		kvm_max_guest_tsc_khz = max;
10670 
10671 		kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
10672 	}
10673 
10674 	kvm_init_msr_list();
10675 	return 0;
10676 }
10677 
10678 void kvm_arch_hardware_unsetup(void)
10679 {
10680 	static_call(kvm_x86_hardware_unsetup)();
10681 }
10682 
10683 int kvm_arch_check_processor_compat(void *opaque)
10684 {
10685 	struct cpuinfo_x86 *c = &cpu_data(smp_processor_id());
10686 	struct kvm_x86_init_ops *ops = opaque;
10687 
10688 	WARN_ON(!irqs_disabled());
10689 
10690 	if (__cr4_reserved_bits(cpu_has, c) !=
10691 	    __cr4_reserved_bits(cpu_has, &boot_cpu_data))
10692 		return -EIO;
10693 
10694 	return ops->check_processor_compatibility();
10695 }
10696 
10697 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
10698 {
10699 	return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
10700 }
10701 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
10702 
10703 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
10704 {
10705 	return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
10706 }
10707 
10708 __read_mostly DEFINE_STATIC_KEY_FALSE(kvm_has_noapic_vcpu);
10709 EXPORT_SYMBOL_GPL(kvm_has_noapic_vcpu);
10710 
10711 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
10712 {
10713 	struct kvm_pmu *pmu = vcpu_to_pmu(vcpu);
10714 
10715 	vcpu->arch.l1tf_flush_l1d = true;
10716 	if (pmu->version && unlikely(pmu->event_count)) {
10717 		pmu->need_cleanup = true;
10718 		kvm_make_request(KVM_REQ_PMU, vcpu);
10719 	}
10720 	static_call(kvm_x86_sched_in)(vcpu, cpu);
10721 }
10722 
10723 void kvm_arch_free_vm(struct kvm *kvm)
10724 {
10725 	kfree(to_kvm_hv(kvm)->hv_pa_pg);
10726 	vfree(kvm);
10727 }
10728 
10729 
10730 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
10731 {
10732 	if (type)
10733 		return -EINVAL;
10734 
10735 	INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
10736 	INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
10737 	INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
10738 	INIT_LIST_HEAD(&kvm->arch.lpage_disallowed_mmu_pages);
10739 	INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
10740 	atomic_set(&kvm->arch.noncoherent_dma_count, 0);
10741 
10742 	/* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
10743 	set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
10744 	/* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
10745 	set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
10746 		&kvm->arch.irq_sources_bitmap);
10747 
10748 	raw_spin_lock_init(&kvm->arch.tsc_write_lock);
10749 	mutex_init(&kvm->arch.apic_map_lock);
10750 	spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
10751 
10752 	kvm->arch.kvmclock_offset = -get_kvmclock_base_ns();
10753 	pvclock_update_vm_gtod_copy(kvm);
10754 
10755 	kvm->arch.guest_can_read_msr_platform_info = true;
10756 
10757 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
10758 	INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
10759 
10760 	kvm_hv_init_vm(kvm);
10761 	kvm_page_track_init(kvm);
10762 	kvm_mmu_init_vm(kvm);
10763 
10764 	return static_call(kvm_x86_vm_init)(kvm);
10765 }
10766 
10767 int kvm_arch_post_init_vm(struct kvm *kvm)
10768 {
10769 	return kvm_mmu_post_init_vm(kvm);
10770 }
10771 
10772 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
10773 {
10774 	vcpu_load(vcpu);
10775 	kvm_mmu_unload(vcpu);
10776 	vcpu_put(vcpu);
10777 }
10778 
10779 static void kvm_free_vcpus(struct kvm *kvm)
10780 {
10781 	unsigned int i;
10782 	struct kvm_vcpu *vcpu;
10783 
10784 	/*
10785 	 * Unpin any mmu pages first.
10786 	 */
10787 	kvm_for_each_vcpu(i, vcpu, kvm) {
10788 		kvm_clear_async_pf_completion_queue(vcpu);
10789 		kvm_unload_vcpu_mmu(vcpu);
10790 	}
10791 	kvm_for_each_vcpu(i, vcpu, kvm)
10792 		kvm_vcpu_destroy(vcpu);
10793 
10794 	mutex_lock(&kvm->lock);
10795 	for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
10796 		kvm->vcpus[i] = NULL;
10797 
10798 	atomic_set(&kvm->online_vcpus, 0);
10799 	mutex_unlock(&kvm->lock);
10800 }
10801 
10802 void kvm_arch_sync_events(struct kvm *kvm)
10803 {
10804 	cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
10805 	cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
10806 	kvm_free_pit(kvm);
10807 }
10808 
10809 #define  ERR_PTR_USR(e)  ((void __user *)ERR_PTR(e))
10810 
10811 /**
10812  * __x86_set_memory_region: Setup KVM internal memory slot
10813  *
10814  * @kvm: the kvm pointer to the VM.
10815  * @id: the slot ID to setup.
10816  * @gpa: the GPA to install the slot (unused when @size == 0).
10817  * @size: the size of the slot. Set to zero to uninstall a slot.
10818  *
10819  * This function helps to setup a KVM internal memory slot.  Specify
10820  * @size > 0 to install a new slot, while @size == 0 to uninstall a
10821  * slot.  The return code can be one of the following:
10822  *
10823  *   HVA:           on success (uninstall will return a bogus HVA)
10824  *   -errno:        on error
10825  *
10826  * The caller should always use IS_ERR() to check the return value
10827  * before use.  Note, the KVM internal memory slots are guaranteed to
10828  * remain valid and unchanged until the VM is destroyed, i.e., the
10829  * GPA->HVA translation will not change.  However, the HVA is a user
10830  * address, i.e. its accessibility is not guaranteed, and must be
10831  * accessed via __copy_{to,from}_user().
10832  */
10833 void __user * __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa,
10834 				      u32 size)
10835 {
10836 	int i, r;
10837 	unsigned long hva, old_npages;
10838 	struct kvm_memslots *slots = kvm_memslots(kvm);
10839 	struct kvm_memory_slot *slot;
10840 
10841 	/* Called with kvm->slots_lock held.  */
10842 	if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
10843 		return ERR_PTR_USR(-EINVAL);
10844 
10845 	slot = id_to_memslot(slots, id);
10846 	if (size) {
10847 		if (slot && slot->npages)
10848 			return ERR_PTR_USR(-EEXIST);
10849 
10850 		/*
10851 		 * MAP_SHARED to prevent internal slot pages from being moved
10852 		 * by fork()/COW.
10853 		 */
10854 		hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
10855 			      MAP_SHARED | MAP_ANONYMOUS, 0);
10856 		if (IS_ERR((void *)hva))
10857 			return (void __user *)hva;
10858 	} else {
10859 		if (!slot || !slot->npages)
10860 			return NULL;
10861 
10862 		old_npages = slot->npages;
10863 		hva = slot->userspace_addr;
10864 	}
10865 
10866 	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
10867 		struct kvm_userspace_memory_region m;
10868 
10869 		m.slot = id | (i << 16);
10870 		m.flags = 0;
10871 		m.guest_phys_addr = gpa;
10872 		m.userspace_addr = hva;
10873 		m.memory_size = size;
10874 		r = __kvm_set_memory_region(kvm, &m);
10875 		if (r < 0)
10876 			return ERR_PTR_USR(r);
10877 	}
10878 
10879 	if (!size)
10880 		vm_munmap(hva, old_npages * PAGE_SIZE);
10881 
10882 	return (void __user *)hva;
10883 }
10884 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
10885 
10886 void kvm_arch_pre_destroy_vm(struct kvm *kvm)
10887 {
10888 	kvm_mmu_pre_destroy_vm(kvm);
10889 }
10890 
10891 void kvm_arch_destroy_vm(struct kvm *kvm)
10892 {
10893 	if (current->mm == kvm->mm) {
10894 		/*
10895 		 * Free memory regions allocated on behalf of userspace,
10896 		 * unless the the memory map has changed due to process exit
10897 		 * or fd copying.
10898 		 */
10899 		mutex_lock(&kvm->slots_lock);
10900 		__x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
10901 					0, 0);
10902 		__x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
10903 					0, 0);
10904 		__x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
10905 		mutex_unlock(&kvm->slots_lock);
10906 	}
10907 	static_call_cond(kvm_x86_vm_destroy)(kvm);
10908 	kvm_free_msr_filter(srcu_dereference_check(kvm->arch.msr_filter, &kvm->srcu, 1));
10909 	kvm_pic_destroy(kvm);
10910 	kvm_ioapic_destroy(kvm);
10911 	kvm_free_vcpus(kvm);
10912 	kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
10913 	kfree(srcu_dereference_check(kvm->arch.pmu_event_filter, &kvm->srcu, 1));
10914 	kvm_mmu_uninit_vm(kvm);
10915 	kvm_page_track_cleanup(kvm);
10916 	kvm_xen_destroy_vm(kvm);
10917 	kvm_hv_destroy_vm(kvm);
10918 }
10919 
10920 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
10921 {
10922 	int i;
10923 
10924 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
10925 		kvfree(slot->arch.rmap[i]);
10926 		slot->arch.rmap[i] = NULL;
10927 
10928 		if (i == 0)
10929 			continue;
10930 
10931 		kvfree(slot->arch.lpage_info[i - 1]);
10932 		slot->arch.lpage_info[i - 1] = NULL;
10933 	}
10934 
10935 	kvm_page_track_free_memslot(slot);
10936 }
10937 
10938 static int kvm_alloc_memslot_metadata(struct kvm_memory_slot *slot,
10939 				      unsigned long npages)
10940 {
10941 	int i;
10942 
10943 	/*
10944 	 * Clear out the previous array pointers for the KVM_MR_MOVE case.  The
10945 	 * old arrays will be freed by __kvm_set_memory_region() if installing
10946 	 * the new memslot is successful.
10947 	 */
10948 	memset(&slot->arch, 0, sizeof(slot->arch));
10949 
10950 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
10951 		struct kvm_lpage_info *linfo;
10952 		unsigned long ugfn;
10953 		int lpages;
10954 		int level = i + 1;
10955 
10956 		lpages = gfn_to_index(slot->base_gfn + npages - 1,
10957 				      slot->base_gfn, level) + 1;
10958 
10959 		slot->arch.rmap[i] =
10960 			kvcalloc(lpages, sizeof(*slot->arch.rmap[i]),
10961 				 GFP_KERNEL_ACCOUNT);
10962 		if (!slot->arch.rmap[i])
10963 			goto out_free;
10964 		if (i == 0)
10965 			continue;
10966 
10967 		linfo = kvcalloc(lpages, sizeof(*linfo), GFP_KERNEL_ACCOUNT);
10968 		if (!linfo)
10969 			goto out_free;
10970 
10971 		slot->arch.lpage_info[i - 1] = linfo;
10972 
10973 		if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
10974 			linfo[0].disallow_lpage = 1;
10975 		if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
10976 			linfo[lpages - 1].disallow_lpage = 1;
10977 		ugfn = slot->userspace_addr >> PAGE_SHIFT;
10978 		/*
10979 		 * If the gfn and userspace address are not aligned wrt each
10980 		 * other, disable large page support for this slot.
10981 		 */
10982 		if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1)) {
10983 			unsigned long j;
10984 
10985 			for (j = 0; j < lpages; ++j)
10986 				linfo[j].disallow_lpage = 1;
10987 		}
10988 	}
10989 
10990 	if (kvm_page_track_create_memslot(slot, npages))
10991 		goto out_free;
10992 
10993 	return 0;
10994 
10995 out_free:
10996 	for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
10997 		kvfree(slot->arch.rmap[i]);
10998 		slot->arch.rmap[i] = NULL;
10999 		if (i == 0)
11000 			continue;
11001 
11002 		kvfree(slot->arch.lpage_info[i - 1]);
11003 		slot->arch.lpage_info[i - 1] = NULL;
11004 	}
11005 	return -ENOMEM;
11006 }
11007 
11008 void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen)
11009 {
11010 	struct kvm_vcpu *vcpu;
11011 	int i;
11012 
11013 	/*
11014 	 * memslots->generation has been incremented.
11015 	 * mmio generation may have reached its maximum value.
11016 	 */
11017 	kvm_mmu_invalidate_mmio_sptes(kvm, gen);
11018 
11019 	/* Force re-initialization of steal_time cache */
11020 	kvm_for_each_vcpu(i, vcpu, kvm)
11021 		kvm_vcpu_kick(vcpu);
11022 }
11023 
11024 int kvm_arch_prepare_memory_region(struct kvm *kvm,
11025 				struct kvm_memory_slot *memslot,
11026 				const struct kvm_userspace_memory_region *mem,
11027 				enum kvm_mr_change change)
11028 {
11029 	if (change == KVM_MR_CREATE || change == KVM_MR_MOVE)
11030 		return kvm_alloc_memslot_metadata(memslot,
11031 						  mem->memory_size >> PAGE_SHIFT);
11032 	return 0;
11033 }
11034 
11035 
11036 static void kvm_mmu_update_cpu_dirty_logging(struct kvm *kvm, bool enable)
11037 {
11038 	struct kvm_arch *ka = &kvm->arch;
11039 
11040 	if (!kvm_x86_ops.cpu_dirty_log_size)
11041 		return;
11042 
11043 	if ((enable && ++ka->cpu_dirty_logging_count == 1) ||
11044 	    (!enable && --ka->cpu_dirty_logging_count == 0))
11045 		kvm_make_all_cpus_request(kvm, KVM_REQ_UPDATE_CPU_DIRTY_LOGGING);
11046 
11047 	WARN_ON_ONCE(ka->cpu_dirty_logging_count < 0);
11048 }
11049 
11050 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
11051 				     struct kvm_memory_slot *old,
11052 				     struct kvm_memory_slot *new,
11053 				     enum kvm_mr_change change)
11054 {
11055 	bool log_dirty_pages = new->flags & KVM_MEM_LOG_DIRTY_PAGES;
11056 
11057 	/*
11058 	 * Update CPU dirty logging if dirty logging is being toggled.  This
11059 	 * applies to all operations.
11060 	 */
11061 	if ((old->flags ^ new->flags) & KVM_MEM_LOG_DIRTY_PAGES)
11062 		kvm_mmu_update_cpu_dirty_logging(kvm, log_dirty_pages);
11063 
11064 	/*
11065 	 * Nothing more to do for RO slots (which can't be dirtied and can't be
11066 	 * made writable) or CREATE/MOVE/DELETE of a slot.
11067 	 *
11068 	 * For a memslot with dirty logging disabled:
11069 	 * CREATE:      No dirty mappings will already exist.
11070 	 * MOVE/DELETE: The old mappings will already have been cleaned up by
11071 	 *		kvm_arch_flush_shadow_memslot()
11072 	 *
11073 	 * For a memslot with dirty logging enabled:
11074 	 * CREATE:      No shadow pages exist, thus nothing to write-protect
11075 	 *		and no dirty bits to clear.
11076 	 * MOVE/DELETE: The old mappings will already have been cleaned up by
11077 	 *		kvm_arch_flush_shadow_memslot().
11078 	 */
11079 	if ((change != KVM_MR_FLAGS_ONLY) || (new->flags & KVM_MEM_READONLY))
11080 		return;
11081 
11082 	/*
11083 	 * READONLY and non-flags changes were filtered out above, and the only
11084 	 * other flag is LOG_DIRTY_PAGES, i.e. something is wrong if dirty
11085 	 * logging isn't being toggled on or off.
11086 	 */
11087 	if (WARN_ON_ONCE(!((old->flags ^ new->flags) & KVM_MEM_LOG_DIRTY_PAGES)))
11088 		return;
11089 
11090 	if (!log_dirty_pages) {
11091 		/*
11092 		 * Dirty logging tracks sptes in 4k granularity, meaning that
11093 		 * large sptes have to be split.  If live migration succeeds,
11094 		 * the guest in the source machine will be destroyed and large
11095 		 * sptes will be created in the destination.  However, if the
11096 		 * guest continues to run in the source machine (for example if
11097 		 * live migration fails), small sptes will remain around and
11098 		 * cause bad performance.
11099 		 *
11100 		 * Scan sptes if dirty logging has been stopped, dropping those
11101 		 * which can be collapsed into a single large-page spte.  Later
11102 		 * page faults will create the large-page sptes.
11103 		 */
11104 		kvm_mmu_zap_collapsible_sptes(kvm, new);
11105 	} else {
11106 		/* By default, write-protect everything to log writes. */
11107 		int level = PG_LEVEL_4K;
11108 
11109 		if (kvm_x86_ops.cpu_dirty_log_size) {
11110 			/*
11111 			 * Clear all dirty bits, unless pages are treated as
11112 			 * dirty from the get-go.
11113 			 */
11114 			if (!kvm_dirty_log_manual_protect_and_init_set(kvm))
11115 				kvm_mmu_slot_leaf_clear_dirty(kvm, new);
11116 
11117 			/*
11118 			 * Write-protect large pages on write so that dirty
11119 			 * logging happens at 4k granularity.  No need to
11120 			 * write-protect small SPTEs since write accesses are
11121 			 * logged by the CPU via dirty bits.
11122 			 */
11123 			level = PG_LEVEL_2M;
11124 		} else if (kvm_dirty_log_manual_protect_and_init_set(kvm)) {
11125 			/*
11126 			 * If we're with initial-all-set, we don't need
11127 			 * to write protect any small page because
11128 			 * they're reported as dirty already.  However
11129 			 * we still need to write-protect huge pages
11130 			 * so that the page split can happen lazily on
11131 			 * the first write to the huge page.
11132 			 */
11133 			level = PG_LEVEL_2M;
11134 		}
11135 		kvm_mmu_slot_remove_write_access(kvm, new, level);
11136 	}
11137 }
11138 
11139 void kvm_arch_commit_memory_region(struct kvm *kvm,
11140 				const struct kvm_userspace_memory_region *mem,
11141 				struct kvm_memory_slot *old,
11142 				const struct kvm_memory_slot *new,
11143 				enum kvm_mr_change change)
11144 {
11145 	if (!kvm->arch.n_requested_mmu_pages)
11146 		kvm_mmu_change_mmu_pages(kvm,
11147 				kvm_mmu_calculate_default_mmu_pages(kvm));
11148 
11149 	/*
11150 	 * FIXME: const-ify all uses of struct kvm_memory_slot.
11151 	 */
11152 	kvm_mmu_slot_apply_flags(kvm, old, (struct kvm_memory_slot *) new, change);
11153 
11154 	/* Free the arrays associated with the old memslot. */
11155 	if (change == KVM_MR_MOVE)
11156 		kvm_arch_free_memslot(kvm, old);
11157 }
11158 
11159 void kvm_arch_flush_shadow_all(struct kvm *kvm)
11160 {
11161 	kvm_mmu_zap_all(kvm);
11162 }
11163 
11164 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
11165 				   struct kvm_memory_slot *slot)
11166 {
11167 	kvm_page_track_flush_slot(kvm, slot);
11168 }
11169 
11170 static inline bool kvm_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
11171 {
11172 	return (is_guest_mode(vcpu) &&
11173 			kvm_x86_ops.guest_apic_has_interrupt &&
11174 			static_call(kvm_x86_guest_apic_has_interrupt)(vcpu));
11175 }
11176 
11177 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
11178 {
11179 	if (!list_empty_careful(&vcpu->async_pf.done))
11180 		return true;
11181 
11182 	if (kvm_apic_has_events(vcpu))
11183 		return true;
11184 
11185 	if (vcpu->arch.pv.pv_unhalted)
11186 		return true;
11187 
11188 	if (vcpu->arch.exception.pending)
11189 		return true;
11190 
11191 	if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
11192 	    (vcpu->arch.nmi_pending &&
11193 	     static_call(kvm_x86_nmi_allowed)(vcpu, false)))
11194 		return true;
11195 
11196 	if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
11197 	    (vcpu->arch.smi_pending &&
11198 	     static_call(kvm_x86_smi_allowed)(vcpu, false)))
11199 		return true;
11200 
11201 	if (kvm_arch_interrupt_allowed(vcpu) &&
11202 	    (kvm_cpu_has_interrupt(vcpu) ||
11203 	    kvm_guest_apic_has_interrupt(vcpu)))
11204 		return true;
11205 
11206 	if (kvm_hv_has_stimer_pending(vcpu))
11207 		return true;
11208 
11209 	if (is_guest_mode(vcpu) &&
11210 	    kvm_x86_ops.nested_ops->hv_timer_pending &&
11211 	    kvm_x86_ops.nested_ops->hv_timer_pending(vcpu))
11212 		return true;
11213 
11214 	return false;
11215 }
11216 
11217 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
11218 {
11219 	return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
11220 }
11221 
11222 bool kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu)
11223 {
11224 	if (vcpu->arch.apicv_active && static_call(kvm_x86_dy_apicv_has_pending_interrupt)(vcpu))
11225 		return true;
11226 
11227 	return false;
11228 }
11229 
11230 bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
11231 {
11232 	if (READ_ONCE(vcpu->arch.pv.pv_unhalted))
11233 		return true;
11234 
11235 	if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
11236 		kvm_test_request(KVM_REQ_SMI, vcpu) ||
11237 		 kvm_test_request(KVM_REQ_EVENT, vcpu))
11238 		return true;
11239 
11240 	return kvm_arch_dy_has_pending_interrupt(vcpu);
11241 }
11242 
11243 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
11244 {
11245 	if (vcpu->arch.guest_state_protected)
11246 		return true;
11247 
11248 	return vcpu->arch.preempted_in_kernel;
11249 }
11250 
11251 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
11252 {
11253 	return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
11254 }
11255 
11256 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
11257 {
11258 	return static_call(kvm_x86_interrupt_allowed)(vcpu, false);
11259 }
11260 
11261 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
11262 {
11263 	/* Can't read the RIP when guest state is protected, just return 0 */
11264 	if (vcpu->arch.guest_state_protected)
11265 		return 0;
11266 
11267 	if (is_64_bit_mode(vcpu))
11268 		return kvm_rip_read(vcpu);
11269 	return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
11270 		     kvm_rip_read(vcpu));
11271 }
11272 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
11273 
11274 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
11275 {
11276 	return kvm_get_linear_rip(vcpu) == linear_rip;
11277 }
11278 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
11279 
11280 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
11281 {
11282 	unsigned long rflags;
11283 
11284 	rflags = static_call(kvm_x86_get_rflags)(vcpu);
11285 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
11286 		rflags &= ~X86_EFLAGS_TF;
11287 	return rflags;
11288 }
11289 EXPORT_SYMBOL_GPL(kvm_get_rflags);
11290 
11291 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
11292 {
11293 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
11294 	    kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
11295 		rflags |= X86_EFLAGS_TF;
11296 	static_call(kvm_x86_set_rflags)(vcpu, rflags);
11297 }
11298 
11299 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
11300 {
11301 	__kvm_set_rflags(vcpu, rflags);
11302 	kvm_make_request(KVM_REQ_EVENT, vcpu);
11303 }
11304 EXPORT_SYMBOL_GPL(kvm_set_rflags);
11305 
11306 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
11307 {
11308 	int r;
11309 
11310 	if ((vcpu->arch.mmu->direct_map != work->arch.direct_map) ||
11311 	      work->wakeup_all)
11312 		return;
11313 
11314 	r = kvm_mmu_reload(vcpu);
11315 	if (unlikely(r))
11316 		return;
11317 
11318 	if (!vcpu->arch.mmu->direct_map &&
11319 	      work->arch.cr3 != vcpu->arch.mmu->get_guest_pgd(vcpu))
11320 		return;
11321 
11322 	kvm_mmu_do_page_fault(vcpu, work->cr2_or_gpa, 0, true);
11323 }
11324 
11325 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
11326 {
11327 	BUILD_BUG_ON(!is_power_of_2(ASYNC_PF_PER_VCPU));
11328 
11329 	return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
11330 }
11331 
11332 static inline u32 kvm_async_pf_next_probe(u32 key)
11333 {
11334 	return (key + 1) & (ASYNC_PF_PER_VCPU - 1);
11335 }
11336 
11337 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
11338 {
11339 	u32 key = kvm_async_pf_hash_fn(gfn);
11340 
11341 	while (vcpu->arch.apf.gfns[key] != ~0)
11342 		key = kvm_async_pf_next_probe(key);
11343 
11344 	vcpu->arch.apf.gfns[key] = gfn;
11345 }
11346 
11347 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
11348 {
11349 	int i;
11350 	u32 key = kvm_async_pf_hash_fn(gfn);
11351 
11352 	for (i = 0; i < ASYNC_PF_PER_VCPU &&
11353 		     (vcpu->arch.apf.gfns[key] != gfn &&
11354 		      vcpu->arch.apf.gfns[key] != ~0); i++)
11355 		key = kvm_async_pf_next_probe(key);
11356 
11357 	return key;
11358 }
11359 
11360 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
11361 {
11362 	return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
11363 }
11364 
11365 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
11366 {
11367 	u32 i, j, k;
11368 
11369 	i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
11370 
11371 	if (WARN_ON_ONCE(vcpu->arch.apf.gfns[i] != gfn))
11372 		return;
11373 
11374 	while (true) {
11375 		vcpu->arch.apf.gfns[i] = ~0;
11376 		do {
11377 			j = kvm_async_pf_next_probe(j);
11378 			if (vcpu->arch.apf.gfns[j] == ~0)
11379 				return;
11380 			k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
11381 			/*
11382 			 * k lies cyclically in ]i,j]
11383 			 * |    i.k.j |
11384 			 * |....j i.k.| or  |.k..j i...|
11385 			 */
11386 		} while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
11387 		vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
11388 		i = j;
11389 	}
11390 }
11391 
11392 static inline int apf_put_user_notpresent(struct kvm_vcpu *vcpu)
11393 {
11394 	u32 reason = KVM_PV_REASON_PAGE_NOT_PRESENT;
11395 
11396 	return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &reason,
11397 				      sizeof(reason));
11398 }
11399 
11400 static inline int apf_put_user_ready(struct kvm_vcpu *vcpu, u32 token)
11401 {
11402 	unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
11403 
11404 	return kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
11405 					     &token, offset, sizeof(token));
11406 }
11407 
11408 static inline bool apf_pageready_slot_free(struct kvm_vcpu *vcpu)
11409 {
11410 	unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
11411 	u32 val;
11412 
11413 	if (kvm_read_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
11414 					 &val, offset, sizeof(val)))
11415 		return false;
11416 
11417 	return !val;
11418 }
11419 
11420 static bool kvm_can_deliver_async_pf(struct kvm_vcpu *vcpu)
11421 {
11422 	if (!vcpu->arch.apf.delivery_as_pf_vmexit && is_guest_mode(vcpu))
11423 		return false;
11424 
11425 	if (!kvm_pv_async_pf_enabled(vcpu) ||
11426 	    (vcpu->arch.apf.send_user_only && static_call(kvm_x86_get_cpl)(vcpu) == 0))
11427 		return false;
11428 
11429 	return true;
11430 }
11431 
11432 bool kvm_can_do_async_pf(struct kvm_vcpu *vcpu)
11433 {
11434 	if (unlikely(!lapic_in_kernel(vcpu) ||
11435 		     kvm_event_needs_reinjection(vcpu) ||
11436 		     vcpu->arch.exception.pending))
11437 		return false;
11438 
11439 	if (kvm_hlt_in_guest(vcpu->kvm) && !kvm_can_deliver_async_pf(vcpu))
11440 		return false;
11441 
11442 	/*
11443 	 * If interrupts are off we cannot even use an artificial
11444 	 * halt state.
11445 	 */
11446 	return kvm_arch_interrupt_allowed(vcpu);
11447 }
11448 
11449 bool kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
11450 				     struct kvm_async_pf *work)
11451 {
11452 	struct x86_exception fault;
11453 
11454 	trace_kvm_async_pf_not_present(work->arch.token, work->cr2_or_gpa);
11455 	kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
11456 
11457 	if (kvm_can_deliver_async_pf(vcpu) &&
11458 	    !apf_put_user_notpresent(vcpu)) {
11459 		fault.vector = PF_VECTOR;
11460 		fault.error_code_valid = true;
11461 		fault.error_code = 0;
11462 		fault.nested_page_fault = false;
11463 		fault.address = work->arch.token;
11464 		fault.async_page_fault = true;
11465 		kvm_inject_page_fault(vcpu, &fault);
11466 		return true;
11467 	} else {
11468 		/*
11469 		 * It is not possible to deliver a paravirtualized asynchronous
11470 		 * page fault, but putting the guest in an artificial halt state
11471 		 * can be beneficial nevertheless: if an interrupt arrives, we
11472 		 * can deliver it timely and perhaps the guest will schedule
11473 		 * another process.  When the instruction that triggered a page
11474 		 * fault is retried, hopefully the page will be ready in the host.
11475 		 */
11476 		kvm_make_request(KVM_REQ_APF_HALT, vcpu);
11477 		return false;
11478 	}
11479 }
11480 
11481 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
11482 				 struct kvm_async_pf *work)
11483 {
11484 	struct kvm_lapic_irq irq = {
11485 		.delivery_mode = APIC_DM_FIXED,
11486 		.vector = vcpu->arch.apf.vec
11487 	};
11488 
11489 	if (work->wakeup_all)
11490 		work->arch.token = ~0; /* broadcast wakeup */
11491 	else
11492 		kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
11493 	trace_kvm_async_pf_ready(work->arch.token, work->cr2_or_gpa);
11494 
11495 	if ((work->wakeup_all || work->notpresent_injected) &&
11496 	    kvm_pv_async_pf_enabled(vcpu) &&
11497 	    !apf_put_user_ready(vcpu, work->arch.token)) {
11498 		vcpu->arch.apf.pageready_pending = true;
11499 		kvm_apic_set_irq(vcpu, &irq, NULL);
11500 	}
11501 
11502 	vcpu->arch.apf.halted = false;
11503 	vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
11504 }
11505 
11506 void kvm_arch_async_page_present_queued(struct kvm_vcpu *vcpu)
11507 {
11508 	kvm_make_request(KVM_REQ_APF_READY, vcpu);
11509 	if (!vcpu->arch.apf.pageready_pending)
11510 		kvm_vcpu_kick(vcpu);
11511 }
11512 
11513 bool kvm_arch_can_dequeue_async_page_present(struct kvm_vcpu *vcpu)
11514 {
11515 	if (!kvm_pv_async_pf_enabled(vcpu))
11516 		return true;
11517 	else
11518 		return kvm_lapic_enabled(vcpu) && apf_pageready_slot_free(vcpu);
11519 }
11520 
11521 void kvm_arch_start_assignment(struct kvm *kvm)
11522 {
11523 	if (atomic_inc_return(&kvm->arch.assigned_device_count) == 1)
11524 		static_call_cond(kvm_x86_start_assignment)(kvm);
11525 }
11526 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
11527 
11528 void kvm_arch_end_assignment(struct kvm *kvm)
11529 {
11530 	atomic_dec(&kvm->arch.assigned_device_count);
11531 }
11532 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
11533 
11534 bool kvm_arch_has_assigned_device(struct kvm *kvm)
11535 {
11536 	return atomic_read(&kvm->arch.assigned_device_count);
11537 }
11538 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
11539 
11540 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
11541 {
11542 	atomic_inc(&kvm->arch.noncoherent_dma_count);
11543 }
11544 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
11545 
11546 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
11547 {
11548 	atomic_dec(&kvm->arch.noncoherent_dma_count);
11549 }
11550 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
11551 
11552 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
11553 {
11554 	return atomic_read(&kvm->arch.noncoherent_dma_count);
11555 }
11556 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
11557 
11558 bool kvm_arch_has_irq_bypass(void)
11559 {
11560 	return true;
11561 }
11562 
11563 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
11564 				      struct irq_bypass_producer *prod)
11565 {
11566 	struct kvm_kernel_irqfd *irqfd =
11567 		container_of(cons, struct kvm_kernel_irqfd, consumer);
11568 	int ret;
11569 
11570 	irqfd->producer = prod;
11571 	kvm_arch_start_assignment(irqfd->kvm);
11572 	ret = static_call(kvm_x86_update_pi_irte)(irqfd->kvm,
11573 					 prod->irq, irqfd->gsi, 1);
11574 
11575 	if (ret)
11576 		kvm_arch_end_assignment(irqfd->kvm);
11577 
11578 	return ret;
11579 }
11580 
11581 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
11582 				      struct irq_bypass_producer *prod)
11583 {
11584 	int ret;
11585 	struct kvm_kernel_irqfd *irqfd =
11586 		container_of(cons, struct kvm_kernel_irqfd, consumer);
11587 
11588 	WARN_ON(irqfd->producer != prod);
11589 	irqfd->producer = NULL;
11590 
11591 	/*
11592 	 * When producer of consumer is unregistered, we change back to
11593 	 * remapped mode, so we can re-use the current implementation
11594 	 * when the irq is masked/disabled or the consumer side (KVM
11595 	 * int this case doesn't want to receive the interrupts.
11596 	*/
11597 	ret = static_call(kvm_x86_update_pi_irte)(irqfd->kvm, prod->irq, irqfd->gsi, 0);
11598 	if (ret)
11599 		printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
11600 		       " fails: %d\n", irqfd->consumer.token, ret);
11601 
11602 	kvm_arch_end_assignment(irqfd->kvm);
11603 }
11604 
11605 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
11606 				   uint32_t guest_irq, bool set)
11607 {
11608 	return static_call(kvm_x86_update_pi_irte)(kvm, host_irq, guest_irq, set);
11609 }
11610 
11611 bool kvm_vector_hashing_enabled(void)
11612 {
11613 	return vector_hashing;
11614 }
11615 
11616 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
11617 {
11618 	return (vcpu->arch.msr_kvm_poll_control & 1) == 0;
11619 }
11620 EXPORT_SYMBOL_GPL(kvm_arch_no_poll);
11621 
11622 
11623 int kvm_spec_ctrl_test_value(u64 value)
11624 {
11625 	/*
11626 	 * test that setting IA32_SPEC_CTRL to given value
11627 	 * is allowed by the host processor
11628 	 */
11629 
11630 	u64 saved_value;
11631 	unsigned long flags;
11632 	int ret = 0;
11633 
11634 	local_irq_save(flags);
11635 
11636 	if (rdmsrl_safe(MSR_IA32_SPEC_CTRL, &saved_value))
11637 		ret = 1;
11638 	else if (wrmsrl_safe(MSR_IA32_SPEC_CTRL, value))
11639 		ret = 1;
11640 	else
11641 		wrmsrl(MSR_IA32_SPEC_CTRL, saved_value);
11642 
11643 	local_irq_restore(flags);
11644 
11645 	return ret;
11646 }
11647 EXPORT_SYMBOL_GPL(kvm_spec_ctrl_test_value);
11648 
11649 void kvm_fixup_and_inject_pf_error(struct kvm_vcpu *vcpu, gva_t gva, u16 error_code)
11650 {
11651 	struct x86_exception fault;
11652 	u32 access = error_code &
11653 		(PFERR_WRITE_MASK | PFERR_FETCH_MASK | PFERR_USER_MASK);
11654 
11655 	if (!(error_code & PFERR_PRESENT_MASK) ||
11656 	    vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, &fault) != UNMAPPED_GVA) {
11657 		/*
11658 		 * If vcpu->arch.walk_mmu->gva_to_gpa succeeded, the page
11659 		 * tables probably do not match the TLB.  Just proceed
11660 		 * with the error code that the processor gave.
11661 		 */
11662 		fault.vector = PF_VECTOR;
11663 		fault.error_code_valid = true;
11664 		fault.error_code = error_code;
11665 		fault.nested_page_fault = false;
11666 		fault.address = gva;
11667 	}
11668 	vcpu->arch.walk_mmu->inject_page_fault(vcpu, &fault);
11669 }
11670 EXPORT_SYMBOL_GPL(kvm_fixup_and_inject_pf_error);
11671 
11672 /*
11673  * Handles kvm_read/write_guest_virt*() result and either injects #PF or returns
11674  * KVM_EXIT_INTERNAL_ERROR for cases not currently handled by KVM. Return value
11675  * indicates whether exit to userspace is needed.
11676  */
11677 int kvm_handle_memory_failure(struct kvm_vcpu *vcpu, int r,
11678 			      struct x86_exception *e)
11679 {
11680 	if (r == X86EMUL_PROPAGATE_FAULT) {
11681 		kvm_inject_emulated_page_fault(vcpu, e);
11682 		return 1;
11683 	}
11684 
11685 	/*
11686 	 * In case kvm_read/write_guest_virt*() failed with X86EMUL_IO_NEEDED
11687 	 * while handling a VMX instruction KVM could've handled the request
11688 	 * correctly by exiting to userspace and performing I/O but there
11689 	 * doesn't seem to be a real use-case behind such requests, just return
11690 	 * KVM_EXIT_INTERNAL_ERROR for now.
11691 	 */
11692 	vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
11693 	vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
11694 	vcpu->run->internal.ndata = 0;
11695 
11696 	return 0;
11697 }
11698 EXPORT_SYMBOL_GPL(kvm_handle_memory_failure);
11699 
11700 int kvm_handle_invpcid(struct kvm_vcpu *vcpu, unsigned long type, gva_t gva)
11701 {
11702 	bool pcid_enabled;
11703 	struct x86_exception e;
11704 	unsigned i;
11705 	unsigned long roots_to_free = 0;
11706 	struct {
11707 		u64 pcid;
11708 		u64 gla;
11709 	} operand;
11710 	int r;
11711 
11712 	r = kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e);
11713 	if (r != X86EMUL_CONTINUE)
11714 		return kvm_handle_memory_failure(vcpu, r, &e);
11715 
11716 	if (operand.pcid >> 12 != 0) {
11717 		kvm_inject_gp(vcpu, 0);
11718 		return 1;
11719 	}
11720 
11721 	pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
11722 
11723 	switch (type) {
11724 	case INVPCID_TYPE_INDIV_ADDR:
11725 		if ((!pcid_enabled && (operand.pcid != 0)) ||
11726 		    is_noncanonical_address(operand.gla, vcpu)) {
11727 			kvm_inject_gp(vcpu, 0);
11728 			return 1;
11729 		}
11730 		kvm_mmu_invpcid_gva(vcpu, operand.gla, operand.pcid);
11731 		return kvm_skip_emulated_instruction(vcpu);
11732 
11733 	case INVPCID_TYPE_SINGLE_CTXT:
11734 		if (!pcid_enabled && (operand.pcid != 0)) {
11735 			kvm_inject_gp(vcpu, 0);
11736 			return 1;
11737 		}
11738 
11739 		if (kvm_get_active_pcid(vcpu) == operand.pcid) {
11740 			kvm_mmu_sync_roots(vcpu);
11741 			kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
11742 		}
11743 
11744 		for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++)
11745 			if (kvm_get_pcid(vcpu, vcpu->arch.mmu->prev_roots[i].pgd)
11746 			    == operand.pcid)
11747 				roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i);
11748 
11749 		kvm_mmu_free_roots(vcpu, vcpu->arch.mmu, roots_to_free);
11750 		/*
11751 		 * If neither the current cr3 nor any of the prev_roots use the
11752 		 * given PCID, then nothing needs to be done here because a
11753 		 * resync will happen anyway before switching to any other CR3.
11754 		 */
11755 
11756 		return kvm_skip_emulated_instruction(vcpu);
11757 
11758 	case INVPCID_TYPE_ALL_NON_GLOBAL:
11759 		/*
11760 		 * Currently, KVM doesn't mark global entries in the shadow
11761 		 * page tables, so a non-global flush just degenerates to a
11762 		 * global flush. If needed, we could optimize this later by
11763 		 * keeping track of global entries in shadow page tables.
11764 		 */
11765 
11766 		fallthrough;
11767 	case INVPCID_TYPE_ALL_INCL_GLOBAL:
11768 		kvm_make_request(KVM_REQ_MMU_RELOAD, vcpu);
11769 		return kvm_skip_emulated_instruction(vcpu);
11770 
11771 	default:
11772 		BUG(); /* We have already checked above that type <= 3 */
11773 	}
11774 }
11775 EXPORT_SYMBOL_GPL(kvm_handle_invpcid);
11776 
11777 static int complete_sev_es_emulated_mmio(struct kvm_vcpu *vcpu)
11778 {
11779 	struct kvm_run *run = vcpu->run;
11780 	struct kvm_mmio_fragment *frag;
11781 	unsigned int len;
11782 
11783 	BUG_ON(!vcpu->mmio_needed);
11784 
11785 	/* Complete previous fragment */
11786 	frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
11787 	len = min(8u, frag->len);
11788 	if (!vcpu->mmio_is_write)
11789 		memcpy(frag->data, run->mmio.data, len);
11790 
11791 	if (frag->len <= 8) {
11792 		/* Switch to the next fragment. */
11793 		frag++;
11794 		vcpu->mmio_cur_fragment++;
11795 	} else {
11796 		/* Go forward to the next mmio piece. */
11797 		frag->data += len;
11798 		frag->gpa += len;
11799 		frag->len -= len;
11800 	}
11801 
11802 	if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
11803 		vcpu->mmio_needed = 0;
11804 
11805 		// VMG change, at this point, we're always done
11806 		// RIP has already been advanced
11807 		return 1;
11808 	}
11809 
11810 	// More MMIO is needed
11811 	run->mmio.phys_addr = frag->gpa;
11812 	run->mmio.len = min(8u, frag->len);
11813 	run->mmio.is_write = vcpu->mmio_is_write;
11814 	if (run->mmio.is_write)
11815 		memcpy(run->mmio.data, frag->data, min(8u, frag->len));
11816 	run->exit_reason = KVM_EXIT_MMIO;
11817 
11818 	vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
11819 
11820 	return 0;
11821 }
11822 
11823 int kvm_sev_es_mmio_write(struct kvm_vcpu *vcpu, gpa_t gpa, unsigned int bytes,
11824 			  void *data)
11825 {
11826 	int handled;
11827 	struct kvm_mmio_fragment *frag;
11828 
11829 	if (!data)
11830 		return -EINVAL;
11831 
11832 	handled = write_emultor.read_write_mmio(vcpu, gpa, bytes, data);
11833 	if (handled == bytes)
11834 		return 1;
11835 
11836 	bytes -= handled;
11837 	gpa += handled;
11838 	data += handled;
11839 
11840 	/*TODO: Check if need to increment number of frags */
11841 	frag = vcpu->mmio_fragments;
11842 	vcpu->mmio_nr_fragments = 1;
11843 	frag->len = bytes;
11844 	frag->gpa = gpa;
11845 	frag->data = data;
11846 
11847 	vcpu->mmio_needed = 1;
11848 	vcpu->mmio_cur_fragment = 0;
11849 
11850 	vcpu->run->mmio.phys_addr = gpa;
11851 	vcpu->run->mmio.len = min(8u, frag->len);
11852 	vcpu->run->mmio.is_write = 1;
11853 	memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
11854 	vcpu->run->exit_reason = KVM_EXIT_MMIO;
11855 
11856 	vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
11857 
11858 	return 0;
11859 }
11860 EXPORT_SYMBOL_GPL(kvm_sev_es_mmio_write);
11861 
11862 int kvm_sev_es_mmio_read(struct kvm_vcpu *vcpu, gpa_t gpa, unsigned int bytes,
11863 			 void *data)
11864 {
11865 	int handled;
11866 	struct kvm_mmio_fragment *frag;
11867 
11868 	if (!data)
11869 		return -EINVAL;
11870 
11871 	handled = read_emultor.read_write_mmio(vcpu, gpa, bytes, data);
11872 	if (handled == bytes)
11873 		return 1;
11874 
11875 	bytes -= handled;
11876 	gpa += handled;
11877 	data += handled;
11878 
11879 	/*TODO: Check if need to increment number of frags */
11880 	frag = vcpu->mmio_fragments;
11881 	vcpu->mmio_nr_fragments = 1;
11882 	frag->len = bytes;
11883 	frag->gpa = gpa;
11884 	frag->data = data;
11885 
11886 	vcpu->mmio_needed = 1;
11887 	vcpu->mmio_cur_fragment = 0;
11888 
11889 	vcpu->run->mmio.phys_addr = gpa;
11890 	vcpu->run->mmio.len = min(8u, frag->len);
11891 	vcpu->run->mmio.is_write = 0;
11892 	vcpu->run->exit_reason = KVM_EXIT_MMIO;
11893 
11894 	vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
11895 
11896 	return 0;
11897 }
11898 EXPORT_SYMBOL_GPL(kvm_sev_es_mmio_read);
11899 
11900 static int complete_sev_es_emulated_ins(struct kvm_vcpu *vcpu)
11901 {
11902 	memcpy(vcpu->arch.guest_ins_data, vcpu->arch.pio_data,
11903 	       vcpu->arch.pio.count * vcpu->arch.pio.size);
11904 	vcpu->arch.pio.count = 0;
11905 
11906 	return 1;
11907 }
11908 
11909 static int kvm_sev_es_outs(struct kvm_vcpu *vcpu, unsigned int size,
11910 			   unsigned int port, void *data,  unsigned int count)
11911 {
11912 	int ret;
11913 
11914 	ret = emulator_pio_out_emulated(vcpu->arch.emulate_ctxt, size, port,
11915 					data, count);
11916 	if (ret)
11917 		return ret;
11918 
11919 	vcpu->arch.pio.count = 0;
11920 
11921 	return 0;
11922 }
11923 
11924 static int kvm_sev_es_ins(struct kvm_vcpu *vcpu, unsigned int size,
11925 			  unsigned int port, void *data, unsigned int count)
11926 {
11927 	int ret;
11928 
11929 	ret = emulator_pio_in_emulated(vcpu->arch.emulate_ctxt, size, port,
11930 				       data, count);
11931 	if (ret) {
11932 		vcpu->arch.pio.count = 0;
11933 	} else {
11934 		vcpu->arch.guest_ins_data = data;
11935 		vcpu->arch.complete_userspace_io = complete_sev_es_emulated_ins;
11936 	}
11937 
11938 	return 0;
11939 }
11940 
11941 int kvm_sev_es_string_io(struct kvm_vcpu *vcpu, unsigned int size,
11942 			 unsigned int port, void *data,  unsigned int count,
11943 			 int in)
11944 {
11945 	return in ? kvm_sev_es_ins(vcpu, size, port, data, count)
11946 		  : kvm_sev_es_outs(vcpu, size, port, data, count);
11947 }
11948 EXPORT_SYMBOL_GPL(kvm_sev_es_string_io);
11949 
11950 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_entry);
11951 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
11952 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
11953 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
11954 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
11955 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
11956 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
11957 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
11958 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
11959 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
11960 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
11961 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmenter_failed);
11962 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
11963 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
11964 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
11965 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
11966 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window_update);
11967 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
11968 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
11969 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
11970 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);
11971 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_ga_log);
11972 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_apicv_update_request);
11973 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_enter);
11974 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_exit);
11975 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_msr_protocol_enter);
11976 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_msr_protocol_exit);
11977