xref: /linux/arch/x86/kvm/vmx/vmx.c (revision a6d5f9dca42eab3526e2f73aa5b7df2a5fec2c9d)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Kernel-based Virtual Machine driver for Linux
4  *
5  * This module enables machines with Intel VT-x extensions to run virtual
6  * machines without emulation or binary translation.
7  *
8  * Copyright (C) 2006 Qumranet, Inc.
9  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
10  *
11  * Authors:
12  *   Avi Kivity   <avi@qumranet.com>
13  *   Yaniv Kamay  <yaniv@qumranet.com>
14  */
15 
16 #include <linux/frame.h>
17 #include <linux/highmem.h>
18 #include <linux/hrtimer.h>
19 #include <linux/kernel.h>
20 #include <linux/kvm_host.h>
21 #include <linux/module.h>
22 #include <linux/moduleparam.h>
23 #include <linux/mod_devicetable.h>
24 #include <linux/mm.h>
25 #include <linux/sched.h>
26 #include <linux/sched/smt.h>
27 #include <linux/slab.h>
28 #include <linux/tboot.h>
29 #include <linux/trace_events.h>
30 
31 #include <asm/apic.h>
32 #include <asm/asm.h>
33 #include <asm/cpu.h>
34 #include <asm/debugreg.h>
35 #include <asm/desc.h>
36 #include <asm/fpu/internal.h>
37 #include <asm/io.h>
38 #include <asm/irq_remapping.h>
39 #include <asm/kexec.h>
40 #include <asm/perf_event.h>
41 #include <asm/mce.h>
42 #include <asm/mmu_context.h>
43 #include <asm/mshyperv.h>
44 #include <asm/spec-ctrl.h>
45 #include <asm/virtext.h>
46 #include <asm/vmx.h>
47 
48 #include "capabilities.h"
49 #include "cpuid.h"
50 #include "evmcs.h"
51 #include "irq.h"
52 #include "kvm_cache_regs.h"
53 #include "lapic.h"
54 #include "mmu.h"
55 #include "nested.h"
56 #include "ops.h"
57 #include "pmu.h"
58 #include "trace.h"
59 #include "vmcs.h"
60 #include "vmcs12.h"
61 #include "vmx.h"
62 #include "x86.h"
63 
64 MODULE_AUTHOR("Qumranet");
65 MODULE_LICENSE("GPL");
66 
67 static const struct x86_cpu_id vmx_cpu_id[] = {
68 	X86_FEATURE_MATCH(X86_FEATURE_VMX),
69 	{}
70 };
71 MODULE_DEVICE_TABLE(x86cpu, vmx_cpu_id);
72 
73 bool __read_mostly enable_vpid = 1;
74 module_param_named(vpid, enable_vpid, bool, 0444);
75 
76 static bool __read_mostly enable_vnmi = 1;
77 module_param_named(vnmi, enable_vnmi, bool, S_IRUGO);
78 
79 bool __read_mostly flexpriority_enabled = 1;
80 module_param_named(flexpriority, flexpriority_enabled, bool, S_IRUGO);
81 
82 bool __read_mostly enable_ept = 1;
83 module_param_named(ept, enable_ept, bool, S_IRUGO);
84 
85 bool __read_mostly enable_unrestricted_guest = 1;
86 module_param_named(unrestricted_guest,
87 			enable_unrestricted_guest, bool, S_IRUGO);
88 
89 bool __read_mostly enable_ept_ad_bits = 1;
90 module_param_named(eptad, enable_ept_ad_bits, bool, S_IRUGO);
91 
92 static bool __read_mostly emulate_invalid_guest_state = true;
93 module_param(emulate_invalid_guest_state, bool, S_IRUGO);
94 
95 static bool __read_mostly fasteoi = 1;
96 module_param(fasteoi, bool, S_IRUGO);
97 
98 static bool __read_mostly enable_apicv = 1;
99 module_param(enable_apicv, bool, S_IRUGO);
100 
101 /*
102  * If nested=1, nested virtualization is supported, i.e., guests may use
103  * VMX and be a hypervisor for its own guests. If nested=0, guests may not
104  * use VMX instructions.
105  */
106 static bool __read_mostly nested = 1;
107 module_param(nested, bool, S_IRUGO);
108 
109 bool __read_mostly enable_pml = 1;
110 module_param_named(pml, enable_pml, bool, S_IRUGO);
111 
112 static bool __read_mostly dump_invalid_vmcs = 0;
113 module_param(dump_invalid_vmcs, bool, 0644);
114 
115 #define MSR_BITMAP_MODE_X2APIC		1
116 #define MSR_BITMAP_MODE_X2APIC_APICV	2
117 
118 #define KVM_VMX_TSC_MULTIPLIER_MAX     0xffffffffffffffffULL
119 
120 /* Guest_tsc -> host_tsc conversion requires 64-bit division.  */
121 static int __read_mostly cpu_preemption_timer_multi;
122 static bool __read_mostly enable_preemption_timer = 1;
123 #ifdef CONFIG_X86_64
124 module_param_named(preemption_timer, enable_preemption_timer, bool, S_IRUGO);
125 #endif
126 
127 #define KVM_VM_CR0_ALWAYS_OFF (X86_CR0_NW | X86_CR0_CD)
128 #define KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST X86_CR0_NE
129 #define KVM_VM_CR0_ALWAYS_ON				\
130 	(KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST | 	\
131 	 X86_CR0_WP | X86_CR0_PG | X86_CR0_PE)
132 #define KVM_CR4_GUEST_OWNED_BITS				      \
133 	(X86_CR4_PVI | X86_CR4_DE | X86_CR4_PCE | X86_CR4_OSFXSR      \
134 	 | X86_CR4_OSXMMEXCPT | X86_CR4_LA57 | X86_CR4_TSD)
135 
136 #define KVM_VM_CR4_ALWAYS_ON_UNRESTRICTED_GUEST X86_CR4_VMXE
137 #define KVM_PMODE_VM_CR4_ALWAYS_ON (X86_CR4_PAE | X86_CR4_VMXE)
138 #define KVM_RMODE_VM_CR4_ALWAYS_ON (X86_CR4_VME | X86_CR4_PAE | X86_CR4_VMXE)
139 
140 #define RMODE_GUEST_OWNED_EFLAGS_BITS (~(X86_EFLAGS_IOPL | X86_EFLAGS_VM))
141 
142 #define MSR_IA32_RTIT_STATUS_MASK (~(RTIT_STATUS_FILTEREN | \
143 	RTIT_STATUS_CONTEXTEN | RTIT_STATUS_TRIGGEREN | \
144 	RTIT_STATUS_ERROR | RTIT_STATUS_STOPPED | \
145 	RTIT_STATUS_BYTECNT))
146 
147 #define MSR_IA32_RTIT_OUTPUT_BASE_MASK \
148 	(~((1UL << cpuid_query_maxphyaddr(vcpu)) - 1) | 0x7f)
149 
150 /*
151  * These 2 parameters are used to config the controls for Pause-Loop Exiting:
152  * ple_gap:    upper bound on the amount of time between two successive
153  *             executions of PAUSE in a loop. Also indicate if ple enabled.
154  *             According to test, this time is usually smaller than 128 cycles.
155  * ple_window: upper bound on the amount of time a guest is allowed to execute
156  *             in a PAUSE loop. Tests indicate that most spinlocks are held for
157  *             less than 2^12 cycles
158  * Time is measured based on a counter that runs at the same rate as the TSC,
159  * refer SDM volume 3b section 21.6.13 & 22.1.3.
160  */
161 static unsigned int ple_gap = KVM_DEFAULT_PLE_GAP;
162 module_param(ple_gap, uint, 0444);
163 
164 static unsigned int ple_window = KVM_VMX_DEFAULT_PLE_WINDOW;
165 module_param(ple_window, uint, 0444);
166 
167 /* Default doubles per-vcpu window every exit. */
168 static unsigned int ple_window_grow = KVM_DEFAULT_PLE_WINDOW_GROW;
169 module_param(ple_window_grow, uint, 0444);
170 
171 /* Default resets per-vcpu window every exit to ple_window. */
172 static unsigned int ple_window_shrink = KVM_DEFAULT_PLE_WINDOW_SHRINK;
173 module_param(ple_window_shrink, uint, 0444);
174 
175 /* Default is to compute the maximum so we can never overflow. */
176 static unsigned int ple_window_max        = KVM_VMX_DEFAULT_PLE_WINDOW_MAX;
177 module_param(ple_window_max, uint, 0444);
178 
179 /* Default is SYSTEM mode, 1 for host-guest mode */
180 int __read_mostly pt_mode = PT_MODE_SYSTEM;
181 module_param(pt_mode, int, S_IRUGO);
182 
183 static DEFINE_STATIC_KEY_FALSE(vmx_l1d_should_flush);
184 static DEFINE_STATIC_KEY_FALSE(vmx_l1d_flush_cond);
185 static DEFINE_MUTEX(vmx_l1d_flush_mutex);
186 
187 /* Storage for pre module init parameter parsing */
188 static enum vmx_l1d_flush_state __read_mostly vmentry_l1d_flush_param = VMENTER_L1D_FLUSH_AUTO;
189 
190 static const struct {
191 	const char *option;
192 	bool for_parse;
193 } vmentry_l1d_param[] = {
194 	[VMENTER_L1D_FLUSH_AUTO]	 = {"auto", true},
195 	[VMENTER_L1D_FLUSH_NEVER]	 = {"never", true},
196 	[VMENTER_L1D_FLUSH_COND]	 = {"cond", true},
197 	[VMENTER_L1D_FLUSH_ALWAYS]	 = {"always", true},
198 	[VMENTER_L1D_FLUSH_EPT_DISABLED] = {"EPT disabled", false},
199 	[VMENTER_L1D_FLUSH_NOT_REQUIRED] = {"not required", false},
200 };
201 
202 #define L1D_CACHE_ORDER 4
203 static void *vmx_l1d_flush_pages;
204 
205 static int vmx_setup_l1d_flush(enum vmx_l1d_flush_state l1tf)
206 {
207 	struct page *page;
208 	unsigned int i;
209 
210 	if (!boot_cpu_has_bug(X86_BUG_L1TF)) {
211 		l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_NOT_REQUIRED;
212 		return 0;
213 	}
214 
215 	if (!enable_ept) {
216 		l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_EPT_DISABLED;
217 		return 0;
218 	}
219 
220 	if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES)) {
221 		u64 msr;
222 
223 		rdmsrl(MSR_IA32_ARCH_CAPABILITIES, msr);
224 		if (msr & ARCH_CAP_SKIP_VMENTRY_L1DFLUSH) {
225 			l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_NOT_REQUIRED;
226 			return 0;
227 		}
228 	}
229 
230 	/* If set to auto use the default l1tf mitigation method */
231 	if (l1tf == VMENTER_L1D_FLUSH_AUTO) {
232 		switch (l1tf_mitigation) {
233 		case L1TF_MITIGATION_OFF:
234 			l1tf = VMENTER_L1D_FLUSH_NEVER;
235 			break;
236 		case L1TF_MITIGATION_FLUSH_NOWARN:
237 		case L1TF_MITIGATION_FLUSH:
238 		case L1TF_MITIGATION_FLUSH_NOSMT:
239 			l1tf = VMENTER_L1D_FLUSH_COND;
240 			break;
241 		case L1TF_MITIGATION_FULL:
242 		case L1TF_MITIGATION_FULL_FORCE:
243 			l1tf = VMENTER_L1D_FLUSH_ALWAYS;
244 			break;
245 		}
246 	} else if (l1tf_mitigation == L1TF_MITIGATION_FULL_FORCE) {
247 		l1tf = VMENTER_L1D_FLUSH_ALWAYS;
248 	}
249 
250 	if (l1tf != VMENTER_L1D_FLUSH_NEVER && !vmx_l1d_flush_pages &&
251 	    !boot_cpu_has(X86_FEATURE_FLUSH_L1D)) {
252 		/*
253 		 * This allocation for vmx_l1d_flush_pages is not tied to a VM
254 		 * lifetime and so should not be charged to a memcg.
255 		 */
256 		page = alloc_pages(GFP_KERNEL, L1D_CACHE_ORDER);
257 		if (!page)
258 			return -ENOMEM;
259 		vmx_l1d_flush_pages = page_address(page);
260 
261 		/*
262 		 * Initialize each page with a different pattern in
263 		 * order to protect against KSM in the nested
264 		 * virtualization case.
265 		 */
266 		for (i = 0; i < 1u << L1D_CACHE_ORDER; ++i) {
267 			memset(vmx_l1d_flush_pages + i * PAGE_SIZE, i + 1,
268 			       PAGE_SIZE);
269 		}
270 	}
271 
272 	l1tf_vmx_mitigation = l1tf;
273 
274 	if (l1tf != VMENTER_L1D_FLUSH_NEVER)
275 		static_branch_enable(&vmx_l1d_should_flush);
276 	else
277 		static_branch_disable(&vmx_l1d_should_flush);
278 
279 	if (l1tf == VMENTER_L1D_FLUSH_COND)
280 		static_branch_enable(&vmx_l1d_flush_cond);
281 	else
282 		static_branch_disable(&vmx_l1d_flush_cond);
283 	return 0;
284 }
285 
286 static int vmentry_l1d_flush_parse(const char *s)
287 {
288 	unsigned int i;
289 
290 	if (s) {
291 		for (i = 0; i < ARRAY_SIZE(vmentry_l1d_param); i++) {
292 			if (vmentry_l1d_param[i].for_parse &&
293 			    sysfs_streq(s, vmentry_l1d_param[i].option))
294 				return i;
295 		}
296 	}
297 	return -EINVAL;
298 }
299 
300 static int vmentry_l1d_flush_set(const char *s, const struct kernel_param *kp)
301 {
302 	int l1tf, ret;
303 
304 	l1tf = vmentry_l1d_flush_parse(s);
305 	if (l1tf < 0)
306 		return l1tf;
307 
308 	if (!boot_cpu_has(X86_BUG_L1TF))
309 		return 0;
310 
311 	/*
312 	 * Has vmx_init() run already? If not then this is the pre init
313 	 * parameter parsing. In that case just store the value and let
314 	 * vmx_init() do the proper setup after enable_ept has been
315 	 * established.
316 	 */
317 	if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_AUTO) {
318 		vmentry_l1d_flush_param = l1tf;
319 		return 0;
320 	}
321 
322 	mutex_lock(&vmx_l1d_flush_mutex);
323 	ret = vmx_setup_l1d_flush(l1tf);
324 	mutex_unlock(&vmx_l1d_flush_mutex);
325 	return ret;
326 }
327 
328 static int vmentry_l1d_flush_get(char *s, const struct kernel_param *kp)
329 {
330 	if (WARN_ON_ONCE(l1tf_vmx_mitigation >= ARRAY_SIZE(vmentry_l1d_param)))
331 		return sprintf(s, "???\n");
332 
333 	return sprintf(s, "%s\n", vmentry_l1d_param[l1tf_vmx_mitigation].option);
334 }
335 
336 static const struct kernel_param_ops vmentry_l1d_flush_ops = {
337 	.set = vmentry_l1d_flush_set,
338 	.get = vmentry_l1d_flush_get,
339 };
340 module_param_cb(vmentry_l1d_flush, &vmentry_l1d_flush_ops, NULL, 0644);
341 
342 static bool guest_state_valid(struct kvm_vcpu *vcpu);
343 static u32 vmx_segment_access_rights(struct kvm_segment *var);
344 static __always_inline void vmx_disable_intercept_for_msr(unsigned long *msr_bitmap,
345 							  u32 msr, int type);
346 
347 void vmx_vmexit(void);
348 
349 #define vmx_insn_failed(fmt...)		\
350 do {					\
351 	WARN_ONCE(1, fmt);		\
352 	pr_warn_ratelimited(fmt);	\
353 } while (0)
354 
355 asmlinkage void vmread_error(unsigned long field, bool fault)
356 {
357 	if (fault)
358 		kvm_spurious_fault();
359 	else
360 		vmx_insn_failed("kvm: vmread failed: field=%lx\n", field);
361 }
362 
363 noinline void vmwrite_error(unsigned long field, unsigned long value)
364 {
365 	vmx_insn_failed("kvm: vmwrite failed: field=%lx val=%lx err=%d\n",
366 			field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
367 }
368 
369 noinline void vmclear_error(struct vmcs *vmcs, u64 phys_addr)
370 {
371 	vmx_insn_failed("kvm: vmclear failed: %p/%llx\n", vmcs, phys_addr);
372 }
373 
374 noinline void vmptrld_error(struct vmcs *vmcs, u64 phys_addr)
375 {
376 	vmx_insn_failed("kvm: vmptrld failed: %p/%llx\n", vmcs, phys_addr);
377 }
378 
379 noinline void invvpid_error(unsigned long ext, u16 vpid, gva_t gva)
380 {
381 	vmx_insn_failed("kvm: invvpid failed: ext=0x%lx vpid=%u gva=0x%lx\n",
382 			ext, vpid, gva);
383 }
384 
385 noinline void invept_error(unsigned long ext, u64 eptp, gpa_t gpa)
386 {
387 	vmx_insn_failed("kvm: invept failed: ext=0x%lx eptp=%llx gpa=0x%llx\n",
388 			ext, eptp, gpa);
389 }
390 
391 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
392 DEFINE_PER_CPU(struct vmcs *, current_vmcs);
393 /*
394  * We maintain a per-CPU linked-list of VMCS loaded on that CPU. This is needed
395  * when a CPU is brought down, and we need to VMCLEAR all VMCSs loaded on it.
396  */
397 static DEFINE_PER_CPU(struct list_head, loaded_vmcss_on_cpu);
398 
399 /*
400  * We maintian a per-CPU linked-list of vCPU, so in wakeup_handler() we
401  * can find which vCPU should be waken up.
402  */
403 static DEFINE_PER_CPU(struct list_head, blocked_vcpu_on_cpu);
404 static DEFINE_PER_CPU(spinlock_t, blocked_vcpu_on_cpu_lock);
405 
406 static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
407 static DEFINE_SPINLOCK(vmx_vpid_lock);
408 
409 struct vmcs_config vmcs_config;
410 struct vmx_capability vmx_capability;
411 
412 #define VMX_SEGMENT_FIELD(seg)					\
413 	[VCPU_SREG_##seg] = {                                   \
414 		.selector = GUEST_##seg##_SELECTOR,		\
415 		.base = GUEST_##seg##_BASE,		   	\
416 		.limit = GUEST_##seg##_LIMIT,		   	\
417 		.ar_bytes = GUEST_##seg##_AR_BYTES,	   	\
418 	}
419 
420 static const struct kvm_vmx_segment_field {
421 	unsigned selector;
422 	unsigned base;
423 	unsigned limit;
424 	unsigned ar_bytes;
425 } kvm_vmx_segment_fields[] = {
426 	VMX_SEGMENT_FIELD(CS),
427 	VMX_SEGMENT_FIELD(DS),
428 	VMX_SEGMENT_FIELD(ES),
429 	VMX_SEGMENT_FIELD(FS),
430 	VMX_SEGMENT_FIELD(GS),
431 	VMX_SEGMENT_FIELD(SS),
432 	VMX_SEGMENT_FIELD(TR),
433 	VMX_SEGMENT_FIELD(LDTR),
434 };
435 
436 u64 host_efer;
437 static unsigned long host_idt_base;
438 
439 /*
440  * Though SYSCALL is only supported in 64-bit mode on Intel CPUs, kvm
441  * will emulate SYSCALL in legacy mode if the vendor string in guest
442  * CPUID.0:{EBX,ECX,EDX} is "AuthenticAMD" or "AMDisbetter!" To
443  * support this emulation, IA32_STAR must always be included in
444  * vmx_msr_index[], even in i386 builds.
445  */
446 const u32 vmx_msr_index[] = {
447 #ifdef CONFIG_X86_64
448 	MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
449 #endif
450 	MSR_EFER, MSR_TSC_AUX, MSR_STAR,
451 	MSR_IA32_TSX_CTRL,
452 };
453 
454 #if IS_ENABLED(CONFIG_HYPERV)
455 static bool __read_mostly enlightened_vmcs = true;
456 module_param(enlightened_vmcs, bool, 0444);
457 
458 /* check_ept_pointer() should be under protection of ept_pointer_lock. */
459 static void check_ept_pointer_match(struct kvm *kvm)
460 {
461 	struct kvm_vcpu *vcpu;
462 	u64 tmp_eptp = INVALID_PAGE;
463 	int i;
464 
465 	kvm_for_each_vcpu(i, vcpu, kvm) {
466 		if (!VALID_PAGE(tmp_eptp)) {
467 			tmp_eptp = to_vmx(vcpu)->ept_pointer;
468 		} else if (tmp_eptp != to_vmx(vcpu)->ept_pointer) {
469 			to_kvm_vmx(kvm)->ept_pointers_match
470 				= EPT_POINTERS_MISMATCH;
471 			return;
472 		}
473 	}
474 
475 	to_kvm_vmx(kvm)->ept_pointers_match = EPT_POINTERS_MATCH;
476 }
477 
478 static int kvm_fill_hv_flush_list_func(struct hv_guest_mapping_flush_list *flush,
479 		void *data)
480 {
481 	struct kvm_tlb_range *range = data;
482 
483 	return hyperv_fill_flush_guest_mapping_list(flush, range->start_gfn,
484 			range->pages);
485 }
486 
487 static inline int __hv_remote_flush_tlb_with_range(struct kvm *kvm,
488 		struct kvm_vcpu *vcpu, struct kvm_tlb_range *range)
489 {
490 	u64 ept_pointer = to_vmx(vcpu)->ept_pointer;
491 
492 	/*
493 	 * FLUSH_GUEST_PHYSICAL_ADDRESS_SPACE hypercall needs address
494 	 * of the base of EPT PML4 table, strip off EPT configuration
495 	 * information.
496 	 */
497 	if (range)
498 		return hyperv_flush_guest_mapping_range(ept_pointer & PAGE_MASK,
499 				kvm_fill_hv_flush_list_func, (void *)range);
500 	else
501 		return hyperv_flush_guest_mapping(ept_pointer & PAGE_MASK);
502 }
503 
504 static int hv_remote_flush_tlb_with_range(struct kvm *kvm,
505 		struct kvm_tlb_range *range)
506 {
507 	struct kvm_vcpu *vcpu;
508 	int ret = 0, i;
509 
510 	spin_lock(&to_kvm_vmx(kvm)->ept_pointer_lock);
511 
512 	if (to_kvm_vmx(kvm)->ept_pointers_match == EPT_POINTERS_CHECK)
513 		check_ept_pointer_match(kvm);
514 
515 	if (to_kvm_vmx(kvm)->ept_pointers_match != EPT_POINTERS_MATCH) {
516 		kvm_for_each_vcpu(i, vcpu, kvm) {
517 			/* If ept_pointer is invalid pointer, bypass flush request. */
518 			if (VALID_PAGE(to_vmx(vcpu)->ept_pointer))
519 				ret |= __hv_remote_flush_tlb_with_range(
520 					kvm, vcpu, range);
521 		}
522 	} else {
523 		ret = __hv_remote_flush_tlb_with_range(kvm,
524 				kvm_get_vcpu(kvm, 0), range);
525 	}
526 
527 	spin_unlock(&to_kvm_vmx(kvm)->ept_pointer_lock);
528 	return ret;
529 }
530 static int hv_remote_flush_tlb(struct kvm *kvm)
531 {
532 	return hv_remote_flush_tlb_with_range(kvm, NULL);
533 }
534 
535 static int hv_enable_direct_tlbflush(struct kvm_vcpu *vcpu)
536 {
537 	struct hv_enlightened_vmcs *evmcs;
538 	struct hv_partition_assist_pg **p_hv_pa_pg =
539 			&vcpu->kvm->arch.hyperv.hv_pa_pg;
540 	/*
541 	 * Synthetic VM-Exit is not enabled in current code and so All
542 	 * evmcs in singe VM shares same assist page.
543 	 */
544 	if (!*p_hv_pa_pg)
545 		*p_hv_pa_pg = kzalloc(PAGE_SIZE, GFP_KERNEL);
546 
547 	if (!*p_hv_pa_pg)
548 		return -ENOMEM;
549 
550 	evmcs = (struct hv_enlightened_vmcs *)to_vmx(vcpu)->loaded_vmcs->vmcs;
551 
552 	evmcs->partition_assist_page =
553 		__pa(*p_hv_pa_pg);
554 	evmcs->hv_vm_id = (unsigned long)vcpu->kvm;
555 	evmcs->hv_enlightenments_control.nested_flush_hypercall = 1;
556 
557 	return 0;
558 }
559 
560 #endif /* IS_ENABLED(CONFIG_HYPERV) */
561 
562 /*
563  * Comment's format: document - errata name - stepping - processor name.
564  * Refer from
565  * https://www.virtualbox.org/svn/vbox/trunk/src/VBox/VMM/VMMR0/HMR0.cpp
566  */
567 static u32 vmx_preemption_cpu_tfms[] = {
568 /* 323344.pdf - BA86   - D0 - Xeon 7500 Series */
569 0x000206E6,
570 /* 323056.pdf - AAX65  - C2 - Xeon L3406 */
571 /* 322814.pdf - AAT59  - C2 - i7-600, i5-500, i5-400 and i3-300 Mobile */
572 /* 322911.pdf - AAU65  - C2 - i5-600, i3-500 Desktop and Pentium G6950 */
573 0x00020652,
574 /* 322911.pdf - AAU65  - K0 - i5-600, i3-500 Desktop and Pentium G6950 */
575 0x00020655,
576 /* 322373.pdf - AAO95  - B1 - Xeon 3400 Series */
577 /* 322166.pdf - AAN92  - B1 - i7-800 and i5-700 Desktop */
578 /*
579  * 320767.pdf - AAP86  - B1 -
580  * i7-900 Mobile Extreme, i7-800 and i7-700 Mobile
581  */
582 0x000106E5,
583 /* 321333.pdf - AAM126 - C0 - Xeon 3500 */
584 0x000106A0,
585 /* 321333.pdf - AAM126 - C1 - Xeon 3500 */
586 0x000106A1,
587 /* 320836.pdf - AAJ124 - C0 - i7-900 Desktop Extreme and i7-900 Desktop */
588 0x000106A4,
589  /* 321333.pdf - AAM126 - D0 - Xeon 3500 */
590  /* 321324.pdf - AAK139 - D0 - Xeon 5500 */
591  /* 320836.pdf - AAJ124 - D0 - i7-900 Extreme and i7-900 Desktop */
592 0x000106A5,
593  /* Xeon E3-1220 V2 */
594 0x000306A8,
595 };
596 
597 static inline bool cpu_has_broken_vmx_preemption_timer(void)
598 {
599 	u32 eax = cpuid_eax(0x00000001), i;
600 
601 	/* Clear the reserved bits */
602 	eax &= ~(0x3U << 14 | 0xfU << 28);
603 	for (i = 0; i < ARRAY_SIZE(vmx_preemption_cpu_tfms); i++)
604 		if (eax == vmx_preemption_cpu_tfms[i])
605 			return true;
606 
607 	return false;
608 }
609 
610 static inline bool cpu_need_virtualize_apic_accesses(struct kvm_vcpu *vcpu)
611 {
612 	return flexpriority_enabled && lapic_in_kernel(vcpu);
613 }
614 
615 static inline bool report_flexpriority(void)
616 {
617 	return flexpriority_enabled;
618 }
619 
620 static inline int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
621 {
622 	int i;
623 
624 	for (i = 0; i < vmx->nmsrs; ++i)
625 		if (vmx_msr_index[vmx->guest_msrs[i].index] == msr)
626 			return i;
627 	return -1;
628 }
629 
630 struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
631 {
632 	int i;
633 
634 	i = __find_msr_index(vmx, msr);
635 	if (i >= 0)
636 		return &vmx->guest_msrs[i];
637 	return NULL;
638 }
639 
640 static int vmx_set_guest_msr(struct vcpu_vmx *vmx, struct shared_msr_entry *msr, u64 data)
641 {
642 	int ret = 0;
643 
644 	u64 old_msr_data = msr->data;
645 	msr->data = data;
646 	if (msr - vmx->guest_msrs < vmx->save_nmsrs) {
647 		preempt_disable();
648 		ret = kvm_set_shared_msr(msr->index, msr->data,
649 					 msr->mask);
650 		preempt_enable();
651 		if (ret)
652 			msr->data = old_msr_data;
653 	}
654 	return ret;
655 }
656 
657 void loaded_vmcs_init(struct loaded_vmcs *loaded_vmcs)
658 {
659 	vmcs_clear(loaded_vmcs->vmcs);
660 	if (loaded_vmcs->shadow_vmcs && loaded_vmcs->launched)
661 		vmcs_clear(loaded_vmcs->shadow_vmcs);
662 	loaded_vmcs->cpu = -1;
663 	loaded_vmcs->launched = 0;
664 }
665 
666 #ifdef CONFIG_KEXEC_CORE
667 /*
668  * This bitmap is used to indicate whether the vmclear
669  * operation is enabled on all cpus. All disabled by
670  * default.
671  */
672 static cpumask_t crash_vmclear_enabled_bitmap = CPU_MASK_NONE;
673 
674 static inline void crash_enable_local_vmclear(int cpu)
675 {
676 	cpumask_set_cpu(cpu, &crash_vmclear_enabled_bitmap);
677 }
678 
679 static inline void crash_disable_local_vmclear(int cpu)
680 {
681 	cpumask_clear_cpu(cpu, &crash_vmclear_enabled_bitmap);
682 }
683 
684 static inline int crash_local_vmclear_enabled(int cpu)
685 {
686 	return cpumask_test_cpu(cpu, &crash_vmclear_enabled_bitmap);
687 }
688 
689 static void crash_vmclear_local_loaded_vmcss(void)
690 {
691 	int cpu = raw_smp_processor_id();
692 	struct loaded_vmcs *v;
693 
694 	if (!crash_local_vmclear_enabled(cpu))
695 		return;
696 
697 	list_for_each_entry(v, &per_cpu(loaded_vmcss_on_cpu, cpu),
698 			    loaded_vmcss_on_cpu_link)
699 		vmcs_clear(v->vmcs);
700 }
701 #else
702 static inline void crash_enable_local_vmclear(int cpu) { }
703 static inline void crash_disable_local_vmclear(int cpu) { }
704 #endif /* CONFIG_KEXEC_CORE */
705 
706 static void __loaded_vmcs_clear(void *arg)
707 {
708 	struct loaded_vmcs *loaded_vmcs = arg;
709 	int cpu = raw_smp_processor_id();
710 
711 	if (loaded_vmcs->cpu != cpu)
712 		return; /* vcpu migration can race with cpu offline */
713 	if (per_cpu(current_vmcs, cpu) == loaded_vmcs->vmcs)
714 		per_cpu(current_vmcs, cpu) = NULL;
715 	crash_disable_local_vmclear(cpu);
716 	list_del(&loaded_vmcs->loaded_vmcss_on_cpu_link);
717 
718 	/*
719 	 * we should ensure updating loaded_vmcs->loaded_vmcss_on_cpu_link
720 	 * is before setting loaded_vmcs->vcpu to -1 which is done in
721 	 * loaded_vmcs_init. Otherwise, other cpu can see vcpu = -1 fist
722 	 * then adds the vmcs into percpu list before it is deleted.
723 	 */
724 	smp_wmb();
725 
726 	loaded_vmcs_init(loaded_vmcs);
727 	crash_enable_local_vmclear(cpu);
728 }
729 
730 void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs)
731 {
732 	int cpu = loaded_vmcs->cpu;
733 
734 	if (cpu != -1)
735 		smp_call_function_single(cpu,
736 			 __loaded_vmcs_clear, loaded_vmcs, 1);
737 }
738 
739 static bool vmx_segment_cache_test_set(struct vcpu_vmx *vmx, unsigned seg,
740 				       unsigned field)
741 {
742 	bool ret;
743 	u32 mask = 1 << (seg * SEG_FIELD_NR + field);
744 
745 	if (!kvm_register_is_available(&vmx->vcpu, VCPU_EXREG_SEGMENTS)) {
746 		kvm_register_mark_available(&vmx->vcpu, VCPU_EXREG_SEGMENTS);
747 		vmx->segment_cache.bitmask = 0;
748 	}
749 	ret = vmx->segment_cache.bitmask & mask;
750 	vmx->segment_cache.bitmask |= mask;
751 	return ret;
752 }
753 
754 static u16 vmx_read_guest_seg_selector(struct vcpu_vmx *vmx, unsigned seg)
755 {
756 	u16 *p = &vmx->segment_cache.seg[seg].selector;
757 
758 	if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_SEL))
759 		*p = vmcs_read16(kvm_vmx_segment_fields[seg].selector);
760 	return *p;
761 }
762 
763 static ulong vmx_read_guest_seg_base(struct vcpu_vmx *vmx, unsigned seg)
764 {
765 	ulong *p = &vmx->segment_cache.seg[seg].base;
766 
767 	if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_BASE))
768 		*p = vmcs_readl(kvm_vmx_segment_fields[seg].base);
769 	return *p;
770 }
771 
772 static u32 vmx_read_guest_seg_limit(struct vcpu_vmx *vmx, unsigned seg)
773 {
774 	u32 *p = &vmx->segment_cache.seg[seg].limit;
775 
776 	if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_LIMIT))
777 		*p = vmcs_read32(kvm_vmx_segment_fields[seg].limit);
778 	return *p;
779 }
780 
781 static u32 vmx_read_guest_seg_ar(struct vcpu_vmx *vmx, unsigned seg)
782 {
783 	u32 *p = &vmx->segment_cache.seg[seg].ar;
784 
785 	if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_AR))
786 		*p = vmcs_read32(kvm_vmx_segment_fields[seg].ar_bytes);
787 	return *p;
788 }
789 
790 void update_exception_bitmap(struct kvm_vcpu *vcpu)
791 {
792 	u32 eb;
793 
794 	eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR) |
795 	     (1u << DB_VECTOR) | (1u << AC_VECTOR);
796 	/*
797 	 * Guest access to VMware backdoor ports could legitimately
798 	 * trigger #GP because of TSS I/O permission bitmap.
799 	 * We intercept those #GP and allow access to them anyway
800 	 * as VMware does.
801 	 */
802 	if (enable_vmware_backdoor)
803 		eb |= (1u << GP_VECTOR);
804 	if ((vcpu->guest_debug &
805 	     (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) ==
806 	    (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP))
807 		eb |= 1u << BP_VECTOR;
808 	if (to_vmx(vcpu)->rmode.vm86_active)
809 		eb = ~0;
810 	if (enable_ept)
811 		eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */
812 
813 	/* When we are running a nested L2 guest and L1 specified for it a
814 	 * certain exception bitmap, we must trap the same exceptions and pass
815 	 * them to L1. When running L2, we will only handle the exceptions
816 	 * specified above if L1 did not want them.
817 	 */
818 	if (is_guest_mode(vcpu))
819 		eb |= get_vmcs12(vcpu)->exception_bitmap;
820 
821 	vmcs_write32(EXCEPTION_BITMAP, eb);
822 }
823 
824 /*
825  * Check if MSR is intercepted for currently loaded MSR bitmap.
826  */
827 static bool msr_write_intercepted(struct kvm_vcpu *vcpu, u32 msr)
828 {
829 	unsigned long *msr_bitmap;
830 	int f = sizeof(unsigned long);
831 
832 	if (!cpu_has_vmx_msr_bitmap())
833 		return true;
834 
835 	msr_bitmap = to_vmx(vcpu)->loaded_vmcs->msr_bitmap;
836 
837 	if (msr <= 0x1fff) {
838 		return !!test_bit(msr, msr_bitmap + 0x800 / f);
839 	} else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
840 		msr &= 0x1fff;
841 		return !!test_bit(msr, msr_bitmap + 0xc00 / f);
842 	}
843 
844 	return true;
845 }
846 
847 static void clear_atomic_switch_msr_special(struct vcpu_vmx *vmx,
848 		unsigned long entry, unsigned long exit)
849 {
850 	vm_entry_controls_clearbit(vmx, entry);
851 	vm_exit_controls_clearbit(vmx, exit);
852 }
853 
854 int vmx_find_msr_index(struct vmx_msrs *m, u32 msr)
855 {
856 	unsigned int i;
857 
858 	for (i = 0; i < m->nr; ++i) {
859 		if (m->val[i].index == msr)
860 			return i;
861 	}
862 	return -ENOENT;
863 }
864 
865 static void clear_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr)
866 {
867 	int i;
868 	struct msr_autoload *m = &vmx->msr_autoload;
869 
870 	switch (msr) {
871 	case MSR_EFER:
872 		if (cpu_has_load_ia32_efer()) {
873 			clear_atomic_switch_msr_special(vmx,
874 					VM_ENTRY_LOAD_IA32_EFER,
875 					VM_EXIT_LOAD_IA32_EFER);
876 			return;
877 		}
878 		break;
879 	case MSR_CORE_PERF_GLOBAL_CTRL:
880 		if (cpu_has_load_perf_global_ctrl()) {
881 			clear_atomic_switch_msr_special(vmx,
882 					VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
883 					VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
884 			return;
885 		}
886 		break;
887 	}
888 	i = vmx_find_msr_index(&m->guest, msr);
889 	if (i < 0)
890 		goto skip_guest;
891 	--m->guest.nr;
892 	m->guest.val[i] = m->guest.val[m->guest.nr];
893 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->guest.nr);
894 
895 skip_guest:
896 	i = vmx_find_msr_index(&m->host, msr);
897 	if (i < 0)
898 		return;
899 
900 	--m->host.nr;
901 	m->host.val[i] = m->host.val[m->host.nr];
902 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->host.nr);
903 }
904 
905 static void add_atomic_switch_msr_special(struct vcpu_vmx *vmx,
906 		unsigned long entry, unsigned long exit,
907 		unsigned long guest_val_vmcs, unsigned long host_val_vmcs,
908 		u64 guest_val, u64 host_val)
909 {
910 	vmcs_write64(guest_val_vmcs, guest_val);
911 	if (host_val_vmcs != HOST_IA32_EFER)
912 		vmcs_write64(host_val_vmcs, host_val);
913 	vm_entry_controls_setbit(vmx, entry);
914 	vm_exit_controls_setbit(vmx, exit);
915 }
916 
917 static void add_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr,
918 				  u64 guest_val, u64 host_val, bool entry_only)
919 {
920 	int i, j = 0;
921 	struct msr_autoload *m = &vmx->msr_autoload;
922 
923 	switch (msr) {
924 	case MSR_EFER:
925 		if (cpu_has_load_ia32_efer()) {
926 			add_atomic_switch_msr_special(vmx,
927 					VM_ENTRY_LOAD_IA32_EFER,
928 					VM_EXIT_LOAD_IA32_EFER,
929 					GUEST_IA32_EFER,
930 					HOST_IA32_EFER,
931 					guest_val, host_val);
932 			return;
933 		}
934 		break;
935 	case MSR_CORE_PERF_GLOBAL_CTRL:
936 		if (cpu_has_load_perf_global_ctrl()) {
937 			add_atomic_switch_msr_special(vmx,
938 					VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
939 					VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL,
940 					GUEST_IA32_PERF_GLOBAL_CTRL,
941 					HOST_IA32_PERF_GLOBAL_CTRL,
942 					guest_val, host_val);
943 			return;
944 		}
945 		break;
946 	case MSR_IA32_PEBS_ENABLE:
947 		/* PEBS needs a quiescent period after being disabled (to write
948 		 * a record).  Disabling PEBS through VMX MSR swapping doesn't
949 		 * provide that period, so a CPU could write host's record into
950 		 * guest's memory.
951 		 */
952 		wrmsrl(MSR_IA32_PEBS_ENABLE, 0);
953 	}
954 
955 	i = vmx_find_msr_index(&m->guest, msr);
956 	if (!entry_only)
957 		j = vmx_find_msr_index(&m->host, msr);
958 
959 	if ((i < 0 && m->guest.nr == NR_LOADSTORE_MSRS) ||
960 		(j < 0 &&  m->host.nr == NR_LOADSTORE_MSRS)) {
961 		printk_once(KERN_WARNING "Not enough msr switch entries. "
962 				"Can't add msr %x\n", msr);
963 		return;
964 	}
965 	if (i < 0) {
966 		i = m->guest.nr++;
967 		vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->guest.nr);
968 	}
969 	m->guest.val[i].index = msr;
970 	m->guest.val[i].value = guest_val;
971 
972 	if (entry_only)
973 		return;
974 
975 	if (j < 0) {
976 		j = m->host.nr++;
977 		vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->host.nr);
978 	}
979 	m->host.val[j].index = msr;
980 	m->host.val[j].value = host_val;
981 }
982 
983 static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset)
984 {
985 	u64 guest_efer = vmx->vcpu.arch.efer;
986 	u64 ignore_bits = 0;
987 
988 	/* Shadow paging assumes NX to be available.  */
989 	if (!enable_ept)
990 		guest_efer |= EFER_NX;
991 
992 	/*
993 	 * LMA and LME handled by hardware; SCE meaningless outside long mode.
994 	 */
995 	ignore_bits |= EFER_SCE;
996 #ifdef CONFIG_X86_64
997 	ignore_bits |= EFER_LMA | EFER_LME;
998 	/* SCE is meaningful only in long mode on Intel */
999 	if (guest_efer & EFER_LMA)
1000 		ignore_bits &= ~(u64)EFER_SCE;
1001 #endif
1002 
1003 	/*
1004 	 * On EPT, we can't emulate NX, so we must switch EFER atomically.
1005 	 * On CPUs that support "load IA32_EFER", always switch EFER
1006 	 * atomically, since it's faster than switching it manually.
1007 	 */
1008 	if (cpu_has_load_ia32_efer() ||
1009 	    (enable_ept && ((vmx->vcpu.arch.efer ^ host_efer) & EFER_NX))) {
1010 		if (!(guest_efer & EFER_LMA))
1011 			guest_efer &= ~EFER_LME;
1012 		if (guest_efer != host_efer)
1013 			add_atomic_switch_msr(vmx, MSR_EFER,
1014 					      guest_efer, host_efer, false);
1015 		else
1016 			clear_atomic_switch_msr(vmx, MSR_EFER);
1017 		return false;
1018 	} else {
1019 		clear_atomic_switch_msr(vmx, MSR_EFER);
1020 
1021 		guest_efer &= ~ignore_bits;
1022 		guest_efer |= host_efer & ignore_bits;
1023 
1024 		vmx->guest_msrs[efer_offset].data = guest_efer;
1025 		vmx->guest_msrs[efer_offset].mask = ~ignore_bits;
1026 
1027 		return true;
1028 	}
1029 }
1030 
1031 #ifdef CONFIG_X86_32
1032 /*
1033  * On 32-bit kernels, VM exits still load the FS and GS bases from the
1034  * VMCS rather than the segment table.  KVM uses this helper to figure
1035  * out the current bases to poke them into the VMCS before entry.
1036  */
1037 static unsigned long segment_base(u16 selector)
1038 {
1039 	struct desc_struct *table;
1040 	unsigned long v;
1041 
1042 	if (!(selector & ~SEGMENT_RPL_MASK))
1043 		return 0;
1044 
1045 	table = get_current_gdt_ro();
1046 
1047 	if ((selector & SEGMENT_TI_MASK) == SEGMENT_LDT) {
1048 		u16 ldt_selector = kvm_read_ldt();
1049 
1050 		if (!(ldt_selector & ~SEGMENT_RPL_MASK))
1051 			return 0;
1052 
1053 		table = (struct desc_struct *)segment_base(ldt_selector);
1054 	}
1055 	v = get_desc_base(&table[selector >> 3]);
1056 	return v;
1057 }
1058 #endif
1059 
1060 static inline bool pt_can_write_msr(struct vcpu_vmx *vmx)
1061 {
1062 	return (pt_mode == PT_MODE_HOST_GUEST) &&
1063 	       !(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN);
1064 }
1065 
1066 static inline void pt_load_msr(struct pt_ctx *ctx, u32 addr_range)
1067 {
1068 	u32 i;
1069 
1070 	wrmsrl(MSR_IA32_RTIT_STATUS, ctx->status);
1071 	wrmsrl(MSR_IA32_RTIT_OUTPUT_BASE, ctx->output_base);
1072 	wrmsrl(MSR_IA32_RTIT_OUTPUT_MASK, ctx->output_mask);
1073 	wrmsrl(MSR_IA32_RTIT_CR3_MATCH, ctx->cr3_match);
1074 	for (i = 0; i < addr_range; i++) {
1075 		wrmsrl(MSR_IA32_RTIT_ADDR0_A + i * 2, ctx->addr_a[i]);
1076 		wrmsrl(MSR_IA32_RTIT_ADDR0_B + i * 2, ctx->addr_b[i]);
1077 	}
1078 }
1079 
1080 static inline void pt_save_msr(struct pt_ctx *ctx, u32 addr_range)
1081 {
1082 	u32 i;
1083 
1084 	rdmsrl(MSR_IA32_RTIT_STATUS, ctx->status);
1085 	rdmsrl(MSR_IA32_RTIT_OUTPUT_BASE, ctx->output_base);
1086 	rdmsrl(MSR_IA32_RTIT_OUTPUT_MASK, ctx->output_mask);
1087 	rdmsrl(MSR_IA32_RTIT_CR3_MATCH, ctx->cr3_match);
1088 	for (i = 0; i < addr_range; i++) {
1089 		rdmsrl(MSR_IA32_RTIT_ADDR0_A + i * 2, ctx->addr_a[i]);
1090 		rdmsrl(MSR_IA32_RTIT_ADDR0_B + i * 2, ctx->addr_b[i]);
1091 	}
1092 }
1093 
1094 static void pt_guest_enter(struct vcpu_vmx *vmx)
1095 {
1096 	if (pt_mode == PT_MODE_SYSTEM)
1097 		return;
1098 
1099 	/*
1100 	 * GUEST_IA32_RTIT_CTL is already set in the VMCS.
1101 	 * Save host state before VM entry.
1102 	 */
1103 	rdmsrl(MSR_IA32_RTIT_CTL, vmx->pt_desc.host.ctl);
1104 	if (vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) {
1105 		wrmsrl(MSR_IA32_RTIT_CTL, 0);
1106 		pt_save_msr(&vmx->pt_desc.host, vmx->pt_desc.addr_range);
1107 		pt_load_msr(&vmx->pt_desc.guest, vmx->pt_desc.addr_range);
1108 	}
1109 }
1110 
1111 static void pt_guest_exit(struct vcpu_vmx *vmx)
1112 {
1113 	if (pt_mode == PT_MODE_SYSTEM)
1114 		return;
1115 
1116 	if (vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) {
1117 		pt_save_msr(&vmx->pt_desc.guest, vmx->pt_desc.addr_range);
1118 		pt_load_msr(&vmx->pt_desc.host, vmx->pt_desc.addr_range);
1119 	}
1120 
1121 	/* Reload host state (IA32_RTIT_CTL will be cleared on VM exit). */
1122 	wrmsrl(MSR_IA32_RTIT_CTL, vmx->pt_desc.host.ctl);
1123 }
1124 
1125 void vmx_set_host_fs_gs(struct vmcs_host_state *host, u16 fs_sel, u16 gs_sel,
1126 			unsigned long fs_base, unsigned long gs_base)
1127 {
1128 	if (unlikely(fs_sel != host->fs_sel)) {
1129 		if (!(fs_sel & 7))
1130 			vmcs_write16(HOST_FS_SELECTOR, fs_sel);
1131 		else
1132 			vmcs_write16(HOST_FS_SELECTOR, 0);
1133 		host->fs_sel = fs_sel;
1134 	}
1135 	if (unlikely(gs_sel != host->gs_sel)) {
1136 		if (!(gs_sel & 7))
1137 			vmcs_write16(HOST_GS_SELECTOR, gs_sel);
1138 		else
1139 			vmcs_write16(HOST_GS_SELECTOR, 0);
1140 		host->gs_sel = gs_sel;
1141 	}
1142 	if (unlikely(fs_base != host->fs_base)) {
1143 		vmcs_writel(HOST_FS_BASE, fs_base);
1144 		host->fs_base = fs_base;
1145 	}
1146 	if (unlikely(gs_base != host->gs_base)) {
1147 		vmcs_writel(HOST_GS_BASE, gs_base);
1148 		host->gs_base = gs_base;
1149 	}
1150 }
1151 
1152 void vmx_prepare_switch_to_guest(struct kvm_vcpu *vcpu)
1153 {
1154 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1155 	struct vmcs_host_state *host_state;
1156 #ifdef CONFIG_X86_64
1157 	int cpu = raw_smp_processor_id();
1158 #endif
1159 	unsigned long fs_base, gs_base;
1160 	u16 fs_sel, gs_sel;
1161 	int i;
1162 
1163 	vmx->req_immediate_exit = false;
1164 
1165 	/*
1166 	 * Note that guest MSRs to be saved/restored can also be changed
1167 	 * when guest state is loaded. This happens when guest transitions
1168 	 * to/from long-mode by setting MSR_EFER.LMA.
1169 	 */
1170 	if (!vmx->guest_msrs_ready) {
1171 		vmx->guest_msrs_ready = true;
1172 		for (i = 0; i < vmx->save_nmsrs; ++i)
1173 			kvm_set_shared_msr(vmx->guest_msrs[i].index,
1174 					   vmx->guest_msrs[i].data,
1175 					   vmx->guest_msrs[i].mask);
1176 
1177 	}
1178 	if (vmx->guest_state_loaded)
1179 		return;
1180 
1181 	host_state = &vmx->loaded_vmcs->host_state;
1182 
1183 	/*
1184 	 * Set host fs and gs selectors.  Unfortunately, 22.2.3 does not
1185 	 * allow segment selectors with cpl > 0 or ti == 1.
1186 	 */
1187 	host_state->ldt_sel = kvm_read_ldt();
1188 
1189 #ifdef CONFIG_X86_64
1190 	savesegment(ds, host_state->ds_sel);
1191 	savesegment(es, host_state->es_sel);
1192 
1193 	gs_base = cpu_kernelmode_gs_base(cpu);
1194 	if (likely(is_64bit_mm(current->mm))) {
1195 		save_fsgs_for_kvm();
1196 		fs_sel = current->thread.fsindex;
1197 		gs_sel = current->thread.gsindex;
1198 		fs_base = current->thread.fsbase;
1199 		vmx->msr_host_kernel_gs_base = current->thread.gsbase;
1200 	} else {
1201 		savesegment(fs, fs_sel);
1202 		savesegment(gs, gs_sel);
1203 		fs_base = read_msr(MSR_FS_BASE);
1204 		vmx->msr_host_kernel_gs_base = read_msr(MSR_KERNEL_GS_BASE);
1205 	}
1206 
1207 	wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1208 #else
1209 	savesegment(fs, fs_sel);
1210 	savesegment(gs, gs_sel);
1211 	fs_base = segment_base(fs_sel);
1212 	gs_base = segment_base(gs_sel);
1213 #endif
1214 
1215 	vmx_set_host_fs_gs(host_state, fs_sel, gs_sel, fs_base, gs_base);
1216 	vmx->guest_state_loaded = true;
1217 }
1218 
1219 static void vmx_prepare_switch_to_host(struct vcpu_vmx *vmx)
1220 {
1221 	struct vmcs_host_state *host_state;
1222 
1223 	if (!vmx->guest_state_loaded)
1224 		return;
1225 
1226 	host_state = &vmx->loaded_vmcs->host_state;
1227 
1228 	++vmx->vcpu.stat.host_state_reload;
1229 
1230 #ifdef CONFIG_X86_64
1231 	rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1232 #endif
1233 	if (host_state->ldt_sel || (host_state->gs_sel & 7)) {
1234 		kvm_load_ldt(host_state->ldt_sel);
1235 #ifdef CONFIG_X86_64
1236 		load_gs_index(host_state->gs_sel);
1237 #else
1238 		loadsegment(gs, host_state->gs_sel);
1239 #endif
1240 	}
1241 	if (host_state->fs_sel & 7)
1242 		loadsegment(fs, host_state->fs_sel);
1243 #ifdef CONFIG_X86_64
1244 	if (unlikely(host_state->ds_sel | host_state->es_sel)) {
1245 		loadsegment(ds, host_state->ds_sel);
1246 		loadsegment(es, host_state->es_sel);
1247 	}
1248 #endif
1249 	invalidate_tss_limit();
1250 #ifdef CONFIG_X86_64
1251 	wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
1252 #endif
1253 	load_fixmap_gdt(raw_smp_processor_id());
1254 	vmx->guest_state_loaded = false;
1255 	vmx->guest_msrs_ready = false;
1256 }
1257 
1258 #ifdef CONFIG_X86_64
1259 static u64 vmx_read_guest_kernel_gs_base(struct vcpu_vmx *vmx)
1260 {
1261 	preempt_disable();
1262 	if (vmx->guest_state_loaded)
1263 		rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1264 	preempt_enable();
1265 	return vmx->msr_guest_kernel_gs_base;
1266 }
1267 
1268 static void vmx_write_guest_kernel_gs_base(struct vcpu_vmx *vmx, u64 data)
1269 {
1270 	preempt_disable();
1271 	if (vmx->guest_state_loaded)
1272 		wrmsrl(MSR_KERNEL_GS_BASE, data);
1273 	preempt_enable();
1274 	vmx->msr_guest_kernel_gs_base = data;
1275 }
1276 #endif
1277 
1278 static void vmx_vcpu_pi_load(struct kvm_vcpu *vcpu, int cpu)
1279 {
1280 	struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
1281 	struct pi_desc old, new;
1282 	unsigned int dest;
1283 
1284 	/*
1285 	 * In case of hot-plug or hot-unplug, we may have to undo
1286 	 * vmx_vcpu_pi_put even if there is no assigned device.  And we
1287 	 * always keep PI.NDST up to date for simplicity: it makes the
1288 	 * code easier, and CPU migration is not a fast path.
1289 	 */
1290 	if (!pi_test_sn(pi_desc) && vcpu->cpu == cpu)
1291 		return;
1292 
1293 	/*
1294 	 * If the 'nv' field is POSTED_INTR_WAKEUP_VECTOR, do not change
1295 	 * PI.NDST: pi_post_block is the one expected to change PID.NDST and the
1296 	 * wakeup handler expects the vCPU to be on the blocked_vcpu_list that
1297 	 * matches PI.NDST. Otherwise, a vcpu may not be able to be woken up
1298 	 * correctly.
1299 	 */
1300 	if (pi_desc->nv == POSTED_INTR_WAKEUP_VECTOR || vcpu->cpu == cpu) {
1301 		pi_clear_sn(pi_desc);
1302 		goto after_clear_sn;
1303 	}
1304 
1305 	/* The full case.  */
1306 	do {
1307 		old.control = new.control = pi_desc->control;
1308 
1309 		dest = cpu_physical_id(cpu);
1310 
1311 		if (x2apic_enabled())
1312 			new.ndst = dest;
1313 		else
1314 			new.ndst = (dest << 8) & 0xFF00;
1315 
1316 		new.sn = 0;
1317 	} while (cmpxchg64(&pi_desc->control, old.control,
1318 			   new.control) != old.control);
1319 
1320 after_clear_sn:
1321 
1322 	/*
1323 	 * Clear SN before reading the bitmap.  The VT-d firmware
1324 	 * writes the bitmap and reads SN atomically (5.2.3 in the
1325 	 * spec), so it doesn't really have a memory barrier that
1326 	 * pairs with this, but we cannot do that and we need one.
1327 	 */
1328 	smp_mb__after_atomic();
1329 
1330 	if (!pi_is_pir_empty(pi_desc))
1331 		pi_set_on(pi_desc);
1332 }
1333 
1334 void vmx_vcpu_load_vmcs(struct kvm_vcpu *vcpu, int cpu)
1335 {
1336 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1337 	bool already_loaded = vmx->loaded_vmcs->cpu == cpu;
1338 
1339 	if (!already_loaded) {
1340 		loaded_vmcs_clear(vmx->loaded_vmcs);
1341 		local_irq_disable();
1342 		crash_disable_local_vmclear(cpu);
1343 
1344 		/*
1345 		 * Read loaded_vmcs->cpu should be before fetching
1346 		 * loaded_vmcs->loaded_vmcss_on_cpu_link.
1347 		 * See the comments in __loaded_vmcs_clear().
1348 		 */
1349 		smp_rmb();
1350 
1351 		list_add(&vmx->loaded_vmcs->loaded_vmcss_on_cpu_link,
1352 			 &per_cpu(loaded_vmcss_on_cpu, cpu));
1353 		crash_enable_local_vmclear(cpu);
1354 		local_irq_enable();
1355 	}
1356 
1357 	if (per_cpu(current_vmcs, cpu) != vmx->loaded_vmcs->vmcs) {
1358 		per_cpu(current_vmcs, cpu) = vmx->loaded_vmcs->vmcs;
1359 		vmcs_load(vmx->loaded_vmcs->vmcs);
1360 		indirect_branch_prediction_barrier();
1361 	}
1362 
1363 	if (!already_loaded) {
1364 		void *gdt = get_current_gdt_ro();
1365 		unsigned long sysenter_esp;
1366 
1367 		kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1368 
1369 		/*
1370 		 * Linux uses per-cpu TSS and GDT, so set these when switching
1371 		 * processors.  See 22.2.4.
1372 		 */
1373 		vmcs_writel(HOST_TR_BASE,
1374 			    (unsigned long)&get_cpu_entry_area(cpu)->tss.x86_tss);
1375 		vmcs_writel(HOST_GDTR_BASE, (unsigned long)gdt);   /* 22.2.4 */
1376 
1377 		rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
1378 		vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
1379 
1380 		vmx->loaded_vmcs->cpu = cpu;
1381 	}
1382 
1383 	/* Setup TSC multiplier */
1384 	if (kvm_has_tsc_control &&
1385 	    vmx->current_tsc_ratio != vcpu->arch.tsc_scaling_ratio)
1386 		decache_tsc_multiplier(vmx);
1387 }
1388 
1389 /*
1390  * Switches to specified vcpu, until a matching vcpu_put(), but assumes
1391  * vcpu mutex is already taken.
1392  */
1393 void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1394 {
1395 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1396 
1397 	vmx_vcpu_load_vmcs(vcpu, cpu);
1398 
1399 	vmx_vcpu_pi_load(vcpu, cpu);
1400 
1401 	vmx->host_pkru = read_pkru();
1402 	vmx->host_debugctlmsr = get_debugctlmsr();
1403 }
1404 
1405 static void vmx_vcpu_pi_put(struct kvm_vcpu *vcpu)
1406 {
1407 	struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
1408 
1409 	if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
1410 		!irq_remapping_cap(IRQ_POSTING_CAP)  ||
1411 		!kvm_vcpu_apicv_active(vcpu))
1412 		return;
1413 
1414 	/* Set SN when the vCPU is preempted */
1415 	if (vcpu->preempted)
1416 		pi_set_sn(pi_desc);
1417 }
1418 
1419 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
1420 {
1421 	vmx_vcpu_pi_put(vcpu);
1422 
1423 	vmx_prepare_switch_to_host(to_vmx(vcpu));
1424 }
1425 
1426 static bool emulation_required(struct kvm_vcpu *vcpu)
1427 {
1428 	return emulate_invalid_guest_state && !guest_state_valid(vcpu);
1429 }
1430 
1431 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu);
1432 
1433 unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
1434 {
1435 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1436 	unsigned long rflags, save_rflags;
1437 
1438 	if (!kvm_register_is_available(vcpu, VCPU_EXREG_RFLAGS)) {
1439 		kvm_register_mark_available(vcpu, VCPU_EXREG_RFLAGS);
1440 		rflags = vmcs_readl(GUEST_RFLAGS);
1441 		if (vmx->rmode.vm86_active) {
1442 			rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
1443 			save_rflags = vmx->rmode.save_rflags;
1444 			rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
1445 		}
1446 		vmx->rflags = rflags;
1447 	}
1448 	return vmx->rflags;
1449 }
1450 
1451 void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
1452 {
1453 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1454 	unsigned long old_rflags;
1455 
1456 	if (enable_unrestricted_guest) {
1457 		kvm_register_mark_available(vcpu, VCPU_EXREG_RFLAGS);
1458 		vmx->rflags = rflags;
1459 		vmcs_writel(GUEST_RFLAGS, rflags);
1460 		return;
1461 	}
1462 
1463 	old_rflags = vmx_get_rflags(vcpu);
1464 	vmx->rflags = rflags;
1465 	if (vmx->rmode.vm86_active) {
1466 		vmx->rmode.save_rflags = rflags;
1467 		rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
1468 	}
1469 	vmcs_writel(GUEST_RFLAGS, rflags);
1470 
1471 	if ((old_rflags ^ vmx->rflags) & X86_EFLAGS_VM)
1472 		vmx->emulation_required = emulation_required(vcpu);
1473 }
1474 
1475 u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu)
1476 {
1477 	u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1478 	int ret = 0;
1479 
1480 	if (interruptibility & GUEST_INTR_STATE_STI)
1481 		ret |= KVM_X86_SHADOW_INT_STI;
1482 	if (interruptibility & GUEST_INTR_STATE_MOV_SS)
1483 		ret |= KVM_X86_SHADOW_INT_MOV_SS;
1484 
1485 	return ret;
1486 }
1487 
1488 void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
1489 {
1490 	u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1491 	u32 interruptibility = interruptibility_old;
1492 
1493 	interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
1494 
1495 	if (mask & KVM_X86_SHADOW_INT_MOV_SS)
1496 		interruptibility |= GUEST_INTR_STATE_MOV_SS;
1497 	else if (mask & KVM_X86_SHADOW_INT_STI)
1498 		interruptibility |= GUEST_INTR_STATE_STI;
1499 
1500 	if ((interruptibility != interruptibility_old))
1501 		vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
1502 }
1503 
1504 static int vmx_rtit_ctl_check(struct kvm_vcpu *vcpu, u64 data)
1505 {
1506 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1507 	unsigned long value;
1508 
1509 	/*
1510 	 * Any MSR write that attempts to change bits marked reserved will
1511 	 * case a #GP fault.
1512 	 */
1513 	if (data & vmx->pt_desc.ctl_bitmask)
1514 		return 1;
1515 
1516 	/*
1517 	 * Any attempt to modify IA32_RTIT_CTL while TraceEn is set will
1518 	 * result in a #GP unless the same write also clears TraceEn.
1519 	 */
1520 	if ((vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) &&
1521 		((vmx->pt_desc.guest.ctl ^ data) & ~RTIT_CTL_TRACEEN))
1522 		return 1;
1523 
1524 	/*
1525 	 * WRMSR to IA32_RTIT_CTL that sets TraceEn but clears this bit
1526 	 * and FabricEn would cause #GP, if
1527 	 * CPUID.(EAX=14H, ECX=0):ECX.SNGLRGNOUT[bit 2] = 0
1528 	 */
1529 	if ((data & RTIT_CTL_TRACEEN) && !(data & RTIT_CTL_TOPA) &&
1530 		!(data & RTIT_CTL_FABRIC_EN) &&
1531 		!intel_pt_validate_cap(vmx->pt_desc.caps,
1532 					PT_CAP_single_range_output))
1533 		return 1;
1534 
1535 	/*
1536 	 * MTCFreq, CycThresh and PSBFreq encodings check, any MSR write that
1537 	 * utilize encodings marked reserved will casue a #GP fault.
1538 	 */
1539 	value = intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc_periods);
1540 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc) &&
1541 			!test_bit((data & RTIT_CTL_MTC_RANGE) >>
1542 			RTIT_CTL_MTC_RANGE_OFFSET, &value))
1543 		return 1;
1544 	value = intel_pt_validate_cap(vmx->pt_desc.caps,
1545 						PT_CAP_cycle_thresholds);
1546 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc) &&
1547 			!test_bit((data & RTIT_CTL_CYC_THRESH) >>
1548 			RTIT_CTL_CYC_THRESH_OFFSET, &value))
1549 		return 1;
1550 	value = intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_periods);
1551 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc) &&
1552 			!test_bit((data & RTIT_CTL_PSB_FREQ) >>
1553 			RTIT_CTL_PSB_FREQ_OFFSET, &value))
1554 		return 1;
1555 
1556 	/*
1557 	 * If ADDRx_CFG is reserved or the encodings is >2 will
1558 	 * cause a #GP fault.
1559 	 */
1560 	value = (data & RTIT_CTL_ADDR0) >> RTIT_CTL_ADDR0_OFFSET;
1561 	if ((value && (vmx->pt_desc.addr_range < 1)) || (value > 2))
1562 		return 1;
1563 	value = (data & RTIT_CTL_ADDR1) >> RTIT_CTL_ADDR1_OFFSET;
1564 	if ((value && (vmx->pt_desc.addr_range < 2)) || (value > 2))
1565 		return 1;
1566 	value = (data & RTIT_CTL_ADDR2) >> RTIT_CTL_ADDR2_OFFSET;
1567 	if ((value && (vmx->pt_desc.addr_range < 3)) || (value > 2))
1568 		return 1;
1569 	value = (data & RTIT_CTL_ADDR3) >> RTIT_CTL_ADDR3_OFFSET;
1570 	if ((value && (vmx->pt_desc.addr_range < 4)) || (value > 2))
1571 		return 1;
1572 
1573 	return 0;
1574 }
1575 
1576 static int skip_emulated_instruction(struct kvm_vcpu *vcpu)
1577 {
1578 	unsigned long rip;
1579 
1580 	/*
1581 	 * Using VMCS.VM_EXIT_INSTRUCTION_LEN on EPT misconfig depends on
1582 	 * undefined behavior: Intel's SDM doesn't mandate the VMCS field be
1583 	 * set when EPT misconfig occurs.  In practice, real hardware updates
1584 	 * VM_EXIT_INSTRUCTION_LEN on EPT misconfig, but other hypervisors
1585 	 * (namely Hyper-V) don't set it due to it being undefined behavior,
1586 	 * i.e. we end up advancing IP with some random value.
1587 	 */
1588 	if (!static_cpu_has(X86_FEATURE_HYPERVISOR) ||
1589 	    to_vmx(vcpu)->exit_reason != EXIT_REASON_EPT_MISCONFIG) {
1590 		rip = kvm_rip_read(vcpu);
1591 		rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
1592 		kvm_rip_write(vcpu, rip);
1593 	} else {
1594 		if (!kvm_emulate_instruction(vcpu, EMULTYPE_SKIP))
1595 			return 0;
1596 	}
1597 
1598 	/* skipping an emulated instruction also counts */
1599 	vmx_set_interrupt_shadow(vcpu, 0);
1600 
1601 	return 1;
1602 }
1603 
1604 static void vmx_clear_hlt(struct kvm_vcpu *vcpu)
1605 {
1606 	/*
1607 	 * Ensure that we clear the HLT state in the VMCS.  We don't need to
1608 	 * explicitly skip the instruction because if the HLT state is set,
1609 	 * then the instruction is already executing and RIP has already been
1610 	 * advanced.
1611 	 */
1612 	if (kvm_hlt_in_guest(vcpu->kvm) &&
1613 			vmcs_read32(GUEST_ACTIVITY_STATE) == GUEST_ACTIVITY_HLT)
1614 		vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
1615 }
1616 
1617 static void vmx_queue_exception(struct kvm_vcpu *vcpu)
1618 {
1619 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1620 	unsigned nr = vcpu->arch.exception.nr;
1621 	bool has_error_code = vcpu->arch.exception.has_error_code;
1622 	u32 error_code = vcpu->arch.exception.error_code;
1623 	u32 intr_info = nr | INTR_INFO_VALID_MASK;
1624 
1625 	kvm_deliver_exception_payload(vcpu);
1626 
1627 	if (has_error_code) {
1628 		vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
1629 		intr_info |= INTR_INFO_DELIVER_CODE_MASK;
1630 	}
1631 
1632 	if (vmx->rmode.vm86_active) {
1633 		int inc_eip = 0;
1634 		if (kvm_exception_is_soft(nr))
1635 			inc_eip = vcpu->arch.event_exit_inst_len;
1636 		kvm_inject_realmode_interrupt(vcpu, nr, inc_eip);
1637 		return;
1638 	}
1639 
1640 	WARN_ON_ONCE(vmx->emulation_required);
1641 
1642 	if (kvm_exception_is_soft(nr)) {
1643 		vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
1644 			     vmx->vcpu.arch.event_exit_inst_len);
1645 		intr_info |= INTR_TYPE_SOFT_EXCEPTION;
1646 	} else
1647 		intr_info |= INTR_TYPE_HARD_EXCEPTION;
1648 
1649 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
1650 
1651 	vmx_clear_hlt(vcpu);
1652 }
1653 
1654 static bool vmx_rdtscp_supported(void)
1655 {
1656 	return cpu_has_vmx_rdtscp();
1657 }
1658 
1659 static bool vmx_invpcid_supported(void)
1660 {
1661 	return cpu_has_vmx_invpcid();
1662 }
1663 
1664 /*
1665  * Swap MSR entry in host/guest MSR entry array.
1666  */
1667 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
1668 {
1669 	struct shared_msr_entry tmp;
1670 
1671 	tmp = vmx->guest_msrs[to];
1672 	vmx->guest_msrs[to] = vmx->guest_msrs[from];
1673 	vmx->guest_msrs[from] = tmp;
1674 }
1675 
1676 /*
1677  * Set up the vmcs to automatically save and restore system
1678  * msrs.  Don't touch the 64-bit msrs if the guest is in legacy
1679  * mode, as fiddling with msrs is very expensive.
1680  */
1681 static void setup_msrs(struct vcpu_vmx *vmx)
1682 {
1683 	int save_nmsrs, index;
1684 
1685 	save_nmsrs = 0;
1686 #ifdef CONFIG_X86_64
1687 	/*
1688 	 * The SYSCALL MSRs are only needed on long mode guests, and only
1689 	 * when EFER.SCE is set.
1690 	 */
1691 	if (is_long_mode(&vmx->vcpu) && (vmx->vcpu.arch.efer & EFER_SCE)) {
1692 		index = __find_msr_index(vmx, MSR_STAR);
1693 		if (index >= 0)
1694 			move_msr_up(vmx, index, save_nmsrs++);
1695 		index = __find_msr_index(vmx, MSR_LSTAR);
1696 		if (index >= 0)
1697 			move_msr_up(vmx, index, save_nmsrs++);
1698 		index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
1699 		if (index >= 0)
1700 			move_msr_up(vmx, index, save_nmsrs++);
1701 	}
1702 #endif
1703 	index = __find_msr_index(vmx, MSR_EFER);
1704 	if (index >= 0 && update_transition_efer(vmx, index))
1705 		move_msr_up(vmx, index, save_nmsrs++);
1706 	index = __find_msr_index(vmx, MSR_TSC_AUX);
1707 	if (index >= 0 && guest_cpuid_has(&vmx->vcpu, X86_FEATURE_RDTSCP))
1708 		move_msr_up(vmx, index, save_nmsrs++);
1709 	index = __find_msr_index(vmx, MSR_IA32_TSX_CTRL);
1710 	if (index >= 0)
1711 		move_msr_up(vmx, index, save_nmsrs++);
1712 
1713 	vmx->save_nmsrs = save_nmsrs;
1714 	vmx->guest_msrs_ready = false;
1715 
1716 	if (cpu_has_vmx_msr_bitmap())
1717 		vmx_update_msr_bitmap(&vmx->vcpu);
1718 }
1719 
1720 static u64 vmx_read_l1_tsc_offset(struct kvm_vcpu *vcpu)
1721 {
1722 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1723 
1724 	if (is_guest_mode(vcpu) &&
1725 	    (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETTING))
1726 		return vcpu->arch.tsc_offset - vmcs12->tsc_offset;
1727 
1728 	return vcpu->arch.tsc_offset;
1729 }
1730 
1731 static u64 vmx_write_l1_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1732 {
1733 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1734 	u64 g_tsc_offset = 0;
1735 
1736 	/*
1737 	 * We're here if L1 chose not to trap WRMSR to TSC. According
1738 	 * to the spec, this should set L1's TSC; The offset that L1
1739 	 * set for L2 remains unchanged, and still needs to be added
1740 	 * to the newly set TSC to get L2's TSC.
1741 	 */
1742 	if (is_guest_mode(vcpu) &&
1743 	    (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETTING))
1744 		g_tsc_offset = vmcs12->tsc_offset;
1745 
1746 	trace_kvm_write_tsc_offset(vcpu->vcpu_id,
1747 				   vcpu->arch.tsc_offset - g_tsc_offset,
1748 				   offset);
1749 	vmcs_write64(TSC_OFFSET, offset + g_tsc_offset);
1750 	return offset + g_tsc_offset;
1751 }
1752 
1753 /*
1754  * nested_vmx_allowed() checks whether a guest should be allowed to use VMX
1755  * instructions and MSRs (i.e., nested VMX). Nested VMX is disabled for
1756  * all guests if the "nested" module option is off, and can also be disabled
1757  * for a single guest by disabling its VMX cpuid bit.
1758  */
1759 bool nested_vmx_allowed(struct kvm_vcpu *vcpu)
1760 {
1761 	return nested && guest_cpuid_has(vcpu, X86_FEATURE_VMX);
1762 }
1763 
1764 static inline bool vmx_feature_control_msr_valid(struct kvm_vcpu *vcpu,
1765 						 uint64_t val)
1766 {
1767 	uint64_t valid_bits = to_vmx(vcpu)->msr_ia32_feature_control_valid_bits;
1768 
1769 	return !(val & ~valid_bits);
1770 }
1771 
1772 static int vmx_get_msr_feature(struct kvm_msr_entry *msr)
1773 {
1774 	switch (msr->index) {
1775 	case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
1776 		if (!nested)
1777 			return 1;
1778 		return vmx_get_vmx_msr(&vmcs_config.nested, msr->index, &msr->data);
1779 	default:
1780 		return 1;
1781 	}
1782 }
1783 
1784 /*
1785  * Reads an msr value (of 'msr_index') into 'pdata'.
1786  * Returns 0 on success, non-0 otherwise.
1787  * Assumes vcpu_load() was already called.
1788  */
1789 static int vmx_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1790 {
1791 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1792 	struct shared_msr_entry *msr;
1793 	u32 index;
1794 
1795 	switch (msr_info->index) {
1796 #ifdef CONFIG_X86_64
1797 	case MSR_FS_BASE:
1798 		msr_info->data = vmcs_readl(GUEST_FS_BASE);
1799 		break;
1800 	case MSR_GS_BASE:
1801 		msr_info->data = vmcs_readl(GUEST_GS_BASE);
1802 		break;
1803 	case MSR_KERNEL_GS_BASE:
1804 		msr_info->data = vmx_read_guest_kernel_gs_base(vmx);
1805 		break;
1806 #endif
1807 	case MSR_EFER:
1808 		return kvm_get_msr_common(vcpu, msr_info);
1809 	case MSR_IA32_TSX_CTRL:
1810 		if (!msr_info->host_initiated &&
1811 		    !(vcpu->arch.arch_capabilities & ARCH_CAP_TSX_CTRL_MSR))
1812 			return 1;
1813 		goto find_shared_msr;
1814 	case MSR_IA32_UMWAIT_CONTROL:
1815 		if (!msr_info->host_initiated && !vmx_has_waitpkg(vmx))
1816 			return 1;
1817 
1818 		msr_info->data = vmx->msr_ia32_umwait_control;
1819 		break;
1820 	case MSR_IA32_SPEC_CTRL:
1821 		if (!msr_info->host_initiated &&
1822 		    !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
1823 			return 1;
1824 
1825 		msr_info->data = to_vmx(vcpu)->spec_ctrl;
1826 		break;
1827 	case MSR_IA32_SYSENTER_CS:
1828 		msr_info->data = vmcs_read32(GUEST_SYSENTER_CS);
1829 		break;
1830 	case MSR_IA32_SYSENTER_EIP:
1831 		msr_info->data = vmcs_readl(GUEST_SYSENTER_EIP);
1832 		break;
1833 	case MSR_IA32_SYSENTER_ESP:
1834 		msr_info->data = vmcs_readl(GUEST_SYSENTER_ESP);
1835 		break;
1836 	case MSR_IA32_BNDCFGS:
1837 		if (!kvm_mpx_supported() ||
1838 		    (!msr_info->host_initiated &&
1839 		     !guest_cpuid_has(vcpu, X86_FEATURE_MPX)))
1840 			return 1;
1841 		msr_info->data = vmcs_read64(GUEST_BNDCFGS);
1842 		break;
1843 	case MSR_IA32_MCG_EXT_CTL:
1844 		if (!msr_info->host_initiated &&
1845 		    !(vmx->msr_ia32_feature_control &
1846 		      FEAT_CTL_LMCE_ENABLED))
1847 			return 1;
1848 		msr_info->data = vcpu->arch.mcg_ext_ctl;
1849 		break;
1850 	case MSR_IA32_FEAT_CTL:
1851 		msr_info->data = vmx->msr_ia32_feature_control;
1852 		break;
1853 	case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
1854 		if (!nested_vmx_allowed(vcpu))
1855 			return 1;
1856 		return vmx_get_vmx_msr(&vmx->nested.msrs, msr_info->index,
1857 				       &msr_info->data);
1858 	case MSR_IA32_RTIT_CTL:
1859 		if (pt_mode != PT_MODE_HOST_GUEST)
1860 			return 1;
1861 		msr_info->data = vmx->pt_desc.guest.ctl;
1862 		break;
1863 	case MSR_IA32_RTIT_STATUS:
1864 		if (pt_mode != PT_MODE_HOST_GUEST)
1865 			return 1;
1866 		msr_info->data = vmx->pt_desc.guest.status;
1867 		break;
1868 	case MSR_IA32_RTIT_CR3_MATCH:
1869 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
1870 			!intel_pt_validate_cap(vmx->pt_desc.caps,
1871 						PT_CAP_cr3_filtering))
1872 			return 1;
1873 		msr_info->data = vmx->pt_desc.guest.cr3_match;
1874 		break;
1875 	case MSR_IA32_RTIT_OUTPUT_BASE:
1876 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
1877 			(!intel_pt_validate_cap(vmx->pt_desc.caps,
1878 					PT_CAP_topa_output) &&
1879 			 !intel_pt_validate_cap(vmx->pt_desc.caps,
1880 					PT_CAP_single_range_output)))
1881 			return 1;
1882 		msr_info->data = vmx->pt_desc.guest.output_base;
1883 		break;
1884 	case MSR_IA32_RTIT_OUTPUT_MASK:
1885 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
1886 			(!intel_pt_validate_cap(vmx->pt_desc.caps,
1887 					PT_CAP_topa_output) &&
1888 			 !intel_pt_validate_cap(vmx->pt_desc.caps,
1889 					PT_CAP_single_range_output)))
1890 			return 1;
1891 		msr_info->data = vmx->pt_desc.guest.output_mask;
1892 		break;
1893 	case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
1894 		index = msr_info->index - MSR_IA32_RTIT_ADDR0_A;
1895 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
1896 			(index >= 2 * intel_pt_validate_cap(vmx->pt_desc.caps,
1897 					PT_CAP_num_address_ranges)))
1898 			return 1;
1899 		if (index % 2)
1900 			msr_info->data = vmx->pt_desc.guest.addr_b[index / 2];
1901 		else
1902 			msr_info->data = vmx->pt_desc.guest.addr_a[index / 2];
1903 		break;
1904 	case MSR_TSC_AUX:
1905 		if (!msr_info->host_initiated &&
1906 		    !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP))
1907 			return 1;
1908 		goto find_shared_msr;
1909 	default:
1910 	find_shared_msr:
1911 		msr = find_msr_entry(vmx, msr_info->index);
1912 		if (msr) {
1913 			msr_info->data = msr->data;
1914 			break;
1915 		}
1916 		return kvm_get_msr_common(vcpu, msr_info);
1917 	}
1918 
1919 	return 0;
1920 }
1921 
1922 /*
1923  * Writes msr value into the appropriate "register".
1924  * Returns 0 on success, non-0 otherwise.
1925  * Assumes vcpu_load() was already called.
1926  */
1927 static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1928 {
1929 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1930 	struct shared_msr_entry *msr;
1931 	int ret = 0;
1932 	u32 msr_index = msr_info->index;
1933 	u64 data = msr_info->data;
1934 	u32 index;
1935 
1936 	switch (msr_index) {
1937 	case MSR_EFER:
1938 		ret = kvm_set_msr_common(vcpu, msr_info);
1939 		break;
1940 #ifdef CONFIG_X86_64
1941 	case MSR_FS_BASE:
1942 		vmx_segment_cache_clear(vmx);
1943 		vmcs_writel(GUEST_FS_BASE, data);
1944 		break;
1945 	case MSR_GS_BASE:
1946 		vmx_segment_cache_clear(vmx);
1947 		vmcs_writel(GUEST_GS_BASE, data);
1948 		break;
1949 	case MSR_KERNEL_GS_BASE:
1950 		vmx_write_guest_kernel_gs_base(vmx, data);
1951 		break;
1952 #endif
1953 	case MSR_IA32_SYSENTER_CS:
1954 		if (is_guest_mode(vcpu))
1955 			get_vmcs12(vcpu)->guest_sysenter_cs = data;
1956 		vmcs_write32(GUEST_SYSENTER_CS, data);
1957 		break;
1958 	case MSR_IA32_SYSENTER_EIP:
1959 		if (is_guest_mode(vcpu))
1960 			get_vmcs12(vcpu)->guest_sysenter_eip = data;
1961 		vmcs_writel(GUEST_SYSENTER_EIP, data);
1962 		break;
1963 	case MSR_IA32_SYSENTER_ESP:
1964 		if (is_guest_mode(vcpu))
1965 			get_vmcs12(vcpu)->guest_sysenter_esp = data;
1966 		vmcs_writel(GUEST_SYSENTER_ESP, data);
1967 		break;
1968 	case MSR_IA32_DEBUGCTLMSR:
1969 		if (is_guest_mode(vcpu) && get_vmcs12(vcpu)->vm_exit_controls &
1970 						VM_EXIT_SAVE_DEBUG_CONTROLS)
1971 			get_vmcs12(vcpu)->guest_ia32_debugctl = data;
1972 
1973 		ret = kvm_set_msr_common(vcpu, msr_info);
1974 		break;
1975 
1976 	case MSR_IA32_BNDCFGS:
1977 		if (!kvm_mpx_supported() ||
1978 		    (!msr_info->host_initiated &&
1979 		     !guest_cpuid_has(vcpu, X86_FEATURE_MPX)))
1980 			return 1;
1981 		if (is_noncanonical_address(data & PAGE_MASK, vcpu) ||
1982 		    (data & MSR_IA32_BNDCFGS_RSVD))
1983 			return 1;
1984 		vmcs_write64(GUEST_BNDCFGS, data);
1985 		break;
1986 	case MSR_IA32_UMWAIT_CONTROL:
1987 		if (!msr_info->host_initiated && !vmx_has_waitpkg(vmx))
1988 			return 1;
1989 
1990 		/* The reserved bit 1 and non-32 bit [63:32] should be zero */
1991 		if (data & (BIT_ULL(1) | GENMASK_ULL(63, 32)))
1992 			return 1;
1993 
1994 		vmx->msr_ia32_umwait_control = data;
1995 		break;
1996 	case MSR_IA32_SPEC_CTRL:
1997 		if (!msr_info->host_initiated &&
1998 		    !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
1999 			return 1;
2000 
2001 		if (data & ~kvm_spec_ctrl_valid_bits(vcpu))
2002 			return 1;
2003 
2004 		vmx->spec_ctrl = data;
2005 		if (!data)
2006 			break;
2007 
2008 		/*
2009 		 * For non-nested:
2010 		 * When it's written (to non-zero) for the first time, pass
2011 		 * it through.
2012 		 *
2013 		 * For nested:
2014 		 * The handling of the MSR bitmap for L2 guests is done in
2015 		 * nested_vmx_prepare_msr_bitmap. We should not touch the
2016 		 * vmcs02.msr_bitmap here since it gets completely overwritten
2017 		 * in the merging. We update the vmcs01 here for L1 as well
2018 		 * since it will end up touching the MSR anyway now.
2019 		 */
2020 		vmx_disable_intercept_for_msr(vmx->vmcs01.msr_bitmap,
2021 					      MSR_IA32_SPEC_CTRL,
2022 					      MSR_TYPE_RW);
2023 		break;
2024 	case MSR_IA32_TSX_CTRL:
2025 		if (!msr_info->host_initiated &&
2026 		    !(vcpu->arch.arch_capabilities & ARCH_CAP_TSX_CTRL_MSR))
2027 			return 1;
2028 		if (data & ~(TSX_CTRL_RTM_DISABLE | TSX_CTRL_CPUID_CLEAR))
2029 			return 1;
2030 		goto find_shared_msr;
2031 	case MSR_IA32_PRED_CMD:
2032 		if (!msr_info->host_initiated &&
2033 		    !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
2034 			return 1;
2035 
2036 		if (data & ~PRED_CMD_IBPB)
2037 			return 1;
2038 		if (!boot_cpu_has(X86_FEATURE_SPEC_CTRL))
2039 			return 1;
2040 		if (!data)
2041 			break;
2042 
2043 		wrmsrl(MSR_IA32_PRED_CMD, PRED_CMD_IBPB);
2044 
2045 		/*
2046 		 * For non-nested:
2047 		 * When it's written (to non-zero) for the first time, pass
2048 		 * it through.
2049 		 *
2050 		 * For nested:
2051 		 * The handling of the MSR bitmap for L2 guests is done in
2052 		 * nested_vmx_prepare_msr_bitmap. We should not touch the
2053 		 * vmcs02.msr_bitmap here since it gets completely overwritten
2054 		 * in the merging.
2055 		 */
2056 		vmx_disable_intercept_for_msr(vmx->vmcs01.msr_bitmap, MSR_IA32_PRED_CMD,
2057 					      MSR_TYPE_W);
2058 		break;
2059 	case MSR_IA32_CR_PAT:
2060 		if (!kvm_pat_valid(data))
2061 			return 1;
2062 
2063 		if (is_guest_mode(vcpu) &&
2064 		    get_vmcs12(vcpu)->vm_exit_controls & VM_EXIT_SAVE_IA32_PAT)
2065 			get_vmcs12(vcpu)->guest_ia32_pat = data;
2066 
2067 		if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2068 			vmcs_write64(GUEST_IA32_PAT, data);
2069 			vcpu->arch.pat = data;
2070 			break;
2071 		}
2072 		ret = kvm_set_msr_common(vcpu, msr_info);
2073 		break;
2074 	case MSR_IA32_TSC_ADJUST:
2075 		ret = kvm_set_msr_common(vcpu, msr_info);
2076 		break;
2077 	case MSR_IA32_MCG_EXT_CTL:
2078 		if ((!msr_info->host_initiated &&
2079 		     !(to_vmx(vcpu)->msr_ia32_feature_control &
2080 		       FEAT_CTL_LMCE_ENABLED)) ||
2081 		    (data & ~MCG_EXT_CTL_LMCE_EN))
2082 			return 1;
2083 		vcpu->arch.mcg_ext_ctl = data;
2084 		break;
2085 	case MSR_IA32_FEAT_CTL:
2086 		if (!vmx_feature_control_msr_valid(vcpu, data) ||
2087 		    (to_vmx(vcpu)->msr_ia32_feature_control &
2088 		     FEAT_CTL_LOCKED && !msr_info->host_initiated))
2089 			return 1;
2090 		vmx->msr_ia32_feature_control = data;
2091 		if (msr_info->host_initiated && data == 0)
2092 			vmx_leave_nested(vcpu);
2093 		break;
2094 	case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
2095 		if (!msr_info->host_initiated)
2096 			return 1; /* they are read-only */
2097 		if (!nested_vmx_allowed(vcpu))
2098 			return 1;
2099 		return vmx_set_vmx_msr(vcpu, msr_index, data);
2100 	case MSR_IA32_RTIT_CTL:
2101 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
2102 			vmx_rtit_ctl_check(vcpu, data) ||
2103 			vmx->nested.vmxon)
2104 			return 1;
2105 		vmcs_write64(GUEST_IA32_RTIT_CTL, data);
2106 		vmx->pt_desc.guest.ctl = data;
2107 		pt_update_intercept_for_msr(vmx);
2108 		break;
2109 	case MSR_IA32_RTIT_STATUS:
2110 		if (!pt_can_write_msr(vmx))
2111 			return 1;
2112 		if (data & MSR_IA32_RTIT_STATUS_MASK)
2113 			return 1;
2114 		vmx->pt_desc.guest.status = data;
2115 		break;
2116 	case MSR_IA32_RTIT_CR3_MATCH:
2117 		if (!pt_can_write_msr(vmx))
2118 			return 1;
2119 		if (!intel_pt_validate_cap(vmx->pt_desc.caps,
2120 					   PT_CAP_cr3_filtering))
2121 			return 1;
2122 		vmx->pt_desc.guest.cr3_match = data;
2123 		break;
2124 	case MSR_IA32_RTIT_OUTPUT_BASE:
2125 		if (!pt_can_write_msr(vmx))
2126 			return 1;
2127 		if (!intel_pt_validate_cap(vmx->pt_desc.caps,
2128 					   PT_CAP_topa_output) &&
2129 		    !intel_pt_validate_cap(vmx->pt_desc.caps,
2130 					   PT_CAP_single_range_output))
2131 			return 1;
2132 		if (data & MSR_IA32_RTIT_OUTPUT_BASE_MASK)
2133 			return 1;
2134 		vmx->pt_desc.guest.output_base = data;
2135 		break;
2136 	case MSR_IA32_RTIT_OUTPUT_MASK:
2137 		if (!pt_can_write_msr(vmx))
2138 			return 1;
2139 		if (!intel_pt_validate_cap(vmx->pt_desc.caps,
2140 					   PT_CAP_topa_output) &&
2141 		    !intel_pt_validate_cap(vmx->pt_desc.caps,
2142 					   PT_CAP_single_range_output))
2143 			return 1;
2144 		vmx->pt_desc.guest.output_mask = data;
2145 		break;
2146 	case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
2147 		if (!pt_can_write_msr(vmx))
2148 			return 1;
2149 		index = msr_info->index - MSR_IA32_RTIT_ADDR0_A;
2150 		if (index >= 2 * intel_pt_validate_cap(vmx->pt_desc.caps,
2151 						       PT_CAP_num_address_ranges))
2152 			return 1;
2153 		if (is_noncanonical_address(data, vcpu))
2154 			return 1;
2155 		if (index % 2)
2156 			vmx->pt_desc.guest.addr_b[index / 2] = data;
2157 		else
2158 			vmx->pt_desc.guest.addr_a[index / 2] = data;
2159 		break;
2160 	case MSR_TSC_AUX:
2161 		if (!msr_info->host_initiated &&
2162 		    !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP))
2163 			return 1;
2164 		/* Check reserved bit, higher 32 bits should be zero */
2165 		if ((data >> 32) != 0)
2166 			return 1;
2167 		goto find_shared_msr;
2168 
2169 	default:
2170 	find_shared_msr:
2171 		msr = find_msr_entry(vmx, msr_index);
2172 		if (msr)
2173 			ret = vmx_set_guest_msr(vmx, msr, data);
2174 		else
2175 			ret = kvm_set_msr_common(vcpu, msr_info);
2176 	}
2177 
2178 	return ret;
2179 }
2180 
2181 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
2182 {
2183 	kvm_register_mark_available(vcpu, reg);
2184 
2185 	switch (reg) {
2186 	case VCPU_REGS_RSP:
2187 		vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
2188 		break;
2189 	case VCPU_REGS_RIP:
2190 		vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
2191 		break;
2192 	case VCPU_EXREG_PDPTR:
2193 		if (enable_ept)
2194 			ept_save_pdptrs(vcpu);
2195 		break;
2196 	case VCPU_EXREG_CR3:
2197 		if (enable_unrestricted_guest || (enable_ept && is_paging(vcpu)))
2198 			vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
2199 		break;
2200 	default:
2201 		WARN_ON_ONCE(1);
2202 		break;
2203 	}
2204 }
2205 
2206 static __init int cpu_has_kvm_support(void)
2207 {
2208 	return cpu_has_vmx();
2209 }
2210 
2211 static __init int vmx_disabled_by_bios(void)
2212 {
2213 	return !boot_cpu_has(X86_FEATURE_MSR_IA32_FEAT_CTL) ||
2214 	       !boot_cpu_has(X86_FEATURE_VMX);
2215 }
2216 
2217 static void kvm_cpu_vmxon(u64 addr)
2218 {
2219 	cr4_set_bits(X86_CR4_VMXE);
2220 	intel_pt_handle_vmx(1);
2221 
2222 	asm volatile ("vmxon %0" : : "m"(addr));
2223 }
2224 
2225 static int hardware_enable(void)
2226 {
2227 	int cpu = raw_smp_processor_id();
2228 	u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
2229 
2230 	if (cr4_read_shadow() & X86_CR4_VMXE)
2231 		return -EBUSY;
2232 
2233 	/*
2234 	 * This can happen if we hot-added a CPU but failed to allocate
2235 	 * VP assist page for it.
2236 	 */
2237 	if (static_branch_unlikely(&enable_evmcs) &&
2238 	    !hv_get_vp_assist_page(cpu))
2239 		return -EFAULT;
2240 
2241 	INIT_LIST_HEAD(&per_cpu(loaded_vmcss_on_cpu, cpu));
2242 	INIT_LIST_HEAD(&per_cpu(blocked_vcpu_on_cpu, cpu));
2243 	spin_lock_init(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
2244 
2245 	/*
2246 	 * Now we can enable the vmclear operation in kdump
2247 	 * since the loaded_vmcss_on_cpu list on this cpu
2248 	 * has been initialized.
2249 	 *
2250 	 * Though the cpu is not in VMX operation now, there
2251 	 * is no problem to enable the vmclear operation
2252 	 * for the loaded_vmcss_on_cpu list is empty!
2253 	 */
2254 	crash_enable_local_vmclear(cpu);
2255 
2256 	kvm_cpu_vmxon(phys_addr);
2257 	if (enable_ept)
2258 		ept_sync_global();
2259 
2260 	return 0;
2261 }
2262 
2263 static void vmclear_local_loaded_vmcss(void)
2264 {
2265 	int cpu = raw_smp_processor_id();
2266 	struct loaded_vmcs *v, *n;
2267 
2268 	list_for_each_entry_safe(v, n, &per_cpu(loaded_vmcss_on_cpu, cpu),
2269 				 loaded_vmcss_on_cpu_link)
2270 		__loaded_vmcs_clear(v);
2271 }
2272 
2273 
2274 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
2275  * tricks.
2276  */
2277 static void kvm_cpu_vmxoff(void)
2278 {
2279 	asm volatile (__ex("vmxoff"));
2280 
2281 	intel_pt_handle_vmx(0);
2282 	cr4_clear_bits(X86_CR4_VMXE);
2283 }
2284 
2285 static void hardware_disable(void)
2286 {
2287 	vmclear_local_loaded_vmcss();
2288 	kvm_cpu_vmxoff();
2289 }
2290 
2291 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
2292 				      u32 msr, u32 *result)
2293 {
2294 	u32 vmx_msr_low, vmx_msr_high;
2295 	u32 ctl = ctl_min | ctl_opt;
2296 
2297 	rdmsr(msr, vmx_msr_low, vmx_msr_high);
2298 
2299 	ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
2300 	ctl |= vmx_msr_low;  /* bit == 1 in low word  ==> must be one  */
2301 
2302 	/* Ensure minimum (required) set of control bits are supported. */
2303 	if (ctl_min & ~ctl)
2304 		return -EIO;
2305 
2306 	*result = ctl;
2307 	return 0;
2308 }
2309 
2310 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf,
2311 				    struct vmx_capability *vmx_cap)
2312 {
2313 	u32 vmx_msr_low, vmx_msr_high;
2314 	u32 min, opt, min2, opt2;
2315 	u32 _pin_based_exec_control = 0;
2316 	u32 _cpu_based_exec_control = 0;
2317 	u32 _cpu_based_2nd_exec_control = 0;
2318 	u32 _vmexit_control = 0;
2319 	u32 _vmentry_control = 0;
2320 
2321 	memset(vmcs_conf, 0, sizeof(*vmcs_conf));
2322 	min = CPU_BASED_HLT_EXITING |
2323 #ifdef CONFIG_X86_64
2324 	      CPU_BASED_CR8_LOAD_EXITING |
2325 	      CPU_BASED_CR8_STORE_EXITING |
2326 #endif
2327 	      CPU_BASED_CR3_LOAD_EXITING |
2328 	      CPU_BASED_CR3_STORE_EXITING |
2329 	      CPU_BASED_UNCOND_IO_EXITING |
2330 	      CPU_BASED_MOV_DR_EXITING |
2331 	      CPU_BASED_USE_TSC_OFFSETTING |
2332 	      CPU_BASED_MWAIT_EXITING |
2333 	      CPU_BASED_MONITOR_EXITING |
2334 	      CPU_BASED_INVLPG_EXITING |
2335 	      CPU_BASED_RDPMC_EXITING;
2336 
2337 	opt = CPU_BASED_TPR_SHADOW |
2338 	      CPU_BASED_USE_MSR_BITMAPS |
2339 	      CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
2340 	if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
2341 				&_cpu_based_exec_control) < 0)
2342 		return -EIO;
2343 #ifdef CONFIG_X86_64
2344 	if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2345 		_cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
2346 					   ~CPU_BASED_CR8_STORE_EXITING;
2347 #endif
2348 	if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
2349 		min2 = 0;
2350 		opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2351 			SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2352 			SECONDARY_EXEC_WBINVD_EXITING |
2353 			SECONDARY_EXEC_ENABLE_VPID |
2354 			SECONDARY_EXEC_ENABLE_EPT |
2355 			SECONDARY_EXEC_UNRESTRICTED_GUEST |
2356 			SECONDARY_EXEC_PAUSE_LOOP_EXITING |
2357 			SECONDARY_EXEC_DESC |
2358 			SECONDARY_EXEC_RDTSCP |
2359 			SECONDARY_EXEC_ENABLE_INVPCID |
2360 			SECONDARY_EXEC_APIC_REGISTER_VIRT |
2361 			SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
2362 			SECONDARY_EXEC_SHADOW_VMCS |
2363 			SECONDARY_EXEC_XSAVES |
2364 			SECONDARY_EXEC_RDSEED_EXITING |
2365 			SECONDARY_EXEC_RDRAND_EXITING |
2366 			SECONDARY_EXEC_ENABLE_PML |
2367 			SECONDARY_EXEC_TSC_SCALING |
2368 			SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE |
2369 			SECONDARY_EXEC_PT_USE_GPA |
2370 			SECONDARY_EXEC_PT_CONCEAL_VMX |
2371 			SECONDARY_EXEC_ENABLE_VMFUNC |
2372 			SECONDARY_EXEC_ENCLS_EXITING;
2373 		if (adjust_vmx_controls(min2, opt2,
2374 					MSR_IA32_VMX_PROCBASED_CTLS2,
2375 					&_cpu_based_2nd_exec_control) < 0)
2376 			return -EIO;
2377 	}
2378 #ifndef CONFIG_X86_64
2379 	if (!(_cpu_based_2nd_exec_control &
2380 				SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
2381 		_cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
2382 #endif
2383 
2384 	if (!(_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2385 		_cpu_based_2nd_exec_control &= ~(
2386 				SECONDARY_EXEC_APIC_REGISTER_VIRT |
2387 				SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2388 				SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
2389 
2390 	rdmsr_safe(MSR_IA32_VMX_EPT_VPID_CAP,
2391 		&vmx_cap->ept, &vmx_cap->vpid);
2392 
2393 	if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
2394 		/* CR3 accesses and invlpg don't need to cause VM Exits when EPT
2395 		   enabled */
2396 		_cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
2397 					     CPU_BASED_CR3_STORE_EXITING |
2398 					     CPU_BASED_INVLPG_EXITING);
2399 	} else if (vmx_cap->ept) {
2400 		vmx_cap->ept = 0;
2401 		pr_warn_once("EPT CAP should not exist if not support "
2402 				"1-setting enable EPT VM-execution control\n");
2403 	}
2404 	if (!(_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_VPID) &&
2405 		vmx_cap->vpid) {
2406 		vmx_cap->vpid = 0;
2407 		pr_warn_once("VPID CAP should not exist if not support "
2408 				"1-setting enable VPID VM-execution control\n");
2409 	}
2410 
2411 	min = VM_EXIT_SAVE_DEBUG_CONTROLS | VM_EXIT_ACK_INTR_ON_EXIT;
2412 #ifdef CONFIG_X86_64
2413 	min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
2414 #endif
2415 	opt = VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL |
2416 	      VM_EXIT_LOAD_IA32_PAT |
2417 	      VM_EXIT_LOAD_IA32_EFER |
2418 	      VM_EXIT_CLEAR_BNDCFGS |
2419 	      VM_EXIT_PT_CONCEAL_PIP |
2420 	      VM_EXIT_CLEAR_IA32_RTIT_CTL;
2421 	if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
2422 				&_vmexit_control) < 0)
2423 		return -EIO;
2424 
2425 	min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
2426 	opt = PIN_BASED_VIRTUAL_NMIS | PIN_BASED_POSTED_INTR |
2427 		 PIN_BASED_VMX_PREEMPTION_TIMER;
2428 	if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
2429 				&_pin_based_exec_control) < 0)
2430 		return -EIO;
2431 
2432 	if (cpu_has_broken_vmx_preemption_timer())
2433 		_pin_based_exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
2434 	if (!(_cpu_based_2nd_exec_control &
2435 		SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY))
2436 		_pin_based_exec_control &= ~PIN_BASED_POSTED_INTR;
2437 
2438 	min = VM_ENTRY_LOAD_DEBUG_CONTROLS;
2439 	opt = VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL |
2440 	      VM_ENTRY_LOAD_IA32_PAT |
2441 	      VM_ENTRY_LOAD_IA32_EFER |
2442 	      VM_ENTRY_LOAD_BNDCFGS |
2443 	      VM_ENTRY_PT_CONCEAL_PIP |
2444 	      VM_ENTRY_LOAD_IA32_RTIT_CTL;
2445 	if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
2446 				&_vmentry_control) < 0)
2447 		return -EIO;
2448 
2449 	/*
2450 	 * Some cpus support VM_{ENTRY,EXIT}_IA32_PERF_GLOBAL_CTRL but they
2451 	 * can't be used due to an errata where VM Exit may incorrectly clear
2452 	 * IA32_PERF_GLOBAL_CTRL[34:32].  Workaround the errata by using the
2453 	 * MSR load mechanism to switch IA32_PERF_GLOBAL_CTRL.
2454 	 */
2455 	if (boot_cpu_data.x86 == 0x6) {
2456 		switch (boot_cpu_data.x86_model) {
2457 		case 26: /* AAK155 */
2458 		case 30: /* AAP115 */
2459 		case 37: /* AAT100 */
2460 		case 44: /* BC86,AAY89,BD102 */
2461 		case 46: /* BA97 */
2462 			_vmentry_control &= ~VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL;
2463 			_vmexit_control &= ~VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL;
2464 			pr_warn_once("kvm: VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL "
2465 					"does not work properly. Using workaround\n");
2466 			break;
2467 		default:
2468 			break;
2469 		}
2470 	}
2471 
2472 
2473 	rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
2474 
2475 	/* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
2476 	if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
2477 		return -EIO;
2478 
2479 #ifdef CONFIG_X86_64
2480 	/* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
2481 	if (vmx_msr_high & (1u<<16))
2482 		return -EIO;
2483 #endif
2484 
2485 	/* Require Write-Back (WB) memory type for VMCS accesses. */
2486 	if (((vmx_msr_high >> 18) & 15) != 6)
2487 		return -EIO;
2488 
2489 	vmcs_conf->size = vmx_msr_high & 0x1fff;
2490 	vmcs_conf->order = get_order(vmcs_conf->size);
2491 	vmcs_conf->basic_cap = vmx_msr_high & ~0x1fff;
2492 
2493 	vmcs_conf->revision_id = vmx_msr_low;
2494 
2495 	vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
2496 	vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
2497 	vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
2498 	vmcs_conf->vmexit_ctrl         = _vmexit_control;
2499 	vmcs_conf->vmentry_ctrl        = _vmentry_control;
2500 
2501 	if (static_branch_unlikely(&enable_evmcs))
2502 		evmcs_sanitize_exec_ctrls(vmcs_conf);
2503 
2504 	return 0;
2505 }
2506 
2507 struct vmcs *alloc_vmcs_cpu(bool shadow, int cpu, gfp_t flags)
2508 {
2509 	int node = cpu_to_node(cpu);
2510 	struct page *pages;
2511 	struct vmcs *vmcs;
2512 
2513 	pages = __alloc_pages_node(node, flags, vmcs_config.order);
2514 	if (!pages)
2515 		return NULL;
2516 	vmcs = page_address(pages);
2517 	memset(vmcs, 0, vmcs_config.size);
2518 
2519 	/* KVM supports Enlightened VMCS v1 only */
2520 	if (static_branch_unlikely(&enable_evmcs))
2521 		vmcs->hdr.revision_id = KVM_EVMCS_VERSION;
2522 	else
2523 		vmcs->hdr.revision_id = vmcs_config.revision_id;
2524 
2525 	if (shadow)
2526 		vmcs->hdr.shadow_vmcs = 1;
2527 	return vmcs;
2528 }
2529 
2530 void free_vmcs(struct vmcs *vmcs)
2531 {
2532 	free_pages((unsigned long)vmcs, vmcs_config.order);
2533 }
2534 
2535 /*
2536  * Free a VMCS, but before that VMCLEAR it on the CPU where it was last loaded
2537  */
2538 void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
2539 {
2540 	if (!loaded_vmcs->vmcs)
2541 		return;
2542 	loaded_vmcs_clear(loaded_vmcs);
2543 	free_vmcs(loaded_vmcs->vmcs);
2544 	loaded_vmcs->vmcs = NULL;
2545 	if (loaded_vmcs->msr_bitmap)
2546 		free_page((unsigned long)loaded_vmcs->msr_bitmap);
2547 	WARN_ON(loaded_vmcs->shadow_vmcs != NULL);
2548 }
2549 
2550 int alloc_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
2551 {
2552 	loaded_vmcs->vmcs = alloc_vmcs(false);
2553 	if (!loaded_vmcs->vmcs)
2554 		return -ENOMEM;
2555 
2556 	loaded_vmcs->shadow_vmcs = NULL;
2557 	loaded_vmcs->hv_timer_soft_disabled = false;
2558 	loaded_vmcs_init(loaded_vmcs);
2559 
2560 	if (cpu_has_vmx_msr_bitmap()) {
2561 		loaded_vmcs->msr_bitmap = (unsigned long *)
2562 				__get_free_page(GFP_KERNEL_ACCOUNT);
2563 		if (!loaded_vmcs->msr_bitmap)
2564 			goto out_vmcs;
2565 		memset(loaded_vmcs->msr_bitmap, 0xff, PAGE_SIZE);
2566 
2567 		if (IS_ENABLED(CONFIG_HYPERV) &&
2568 		    static_branch_unlikely(&enable_evmcs) &&
2569 		    (ms_hyperv.nested_features & HV_X64_NESTED_MSR_BITMAP)) {
2570 			struct hv_enlightened_vmcs *evmcs =
2571 				(struct hv_enlightened_vmcs *)loaded_vmcs->vmcs;
2572 
2573 			evmcs->hv_enlightenments_control.msr_bitmap = 1;
2574 		}
2575 	}
2576 
2577 	memset(&loaded_vmcs->host_state, 0, sizeof(struct vmcs_host_state));
2578 	memset(&loaded_vmcs->controls_shadow, 0,
2579 		sizeof(struct vmcs_controls_shadow));
2580 
2581 	return 0;
2582 
2583 out_vmcs:
2584 	free_loaded_vmcs(loaded_vmcs);
2585 	return -ENOMEM;
2586 }
2587 
2588 static void free_kvm_area(void)
2589 {
2590 	int cpu;
2591 
2592 	for_each_possible_cpu(cpu) {
2593 		free_vmcs(per_cpu(vmxarea, cpu));
2594 		per_cpu(vmxarea, cpu) = NULL;
2595 	}
2596 }
2597 
2598 static __init int alloc_kvm_area(void)
2599 {
2600 	int cpu;
2601 
2602 	for_each_possible_cpu(cpu) {
2603 		struct vmcs *vmcs;
2604 
2605 		vmcs = alloc_vmcs_cpu(false, cpu, GFP_KERNEL);
2606 		if (!vmcs) {
2607 			free_kvm_area();
2608 			return -ENOMEM;
2609 		}
2610 
2611 		/*
2612 		 * When eVMCS is enabled, alloc_vmcs_cpu() sets
2613 		 * vmcs->revision_id to KVM_EVMCS_VERSION instead of
2614 		 * revision_id reported by MSR_IA32_VMX_BASIC.
2615 		 *
2616 		 * However, even though not explicitly documented by
2617 		 * TLFS, VMXArea passed as VMXON argument should
2618 		 * still be marked with revision_id reported by
2619 		 * physical CPU.
2620 		 */
2621 		if (static_branch_unlikely(&enable_evmcs))
2622 			vmcs->hdr.revision_id = vmcs_config.revision_id;
2623 
2624 		per_cpu(vmxarea, cpu) = vmcs;
2625 	}
2626 	return 0;
2627 }
2628 
2629 static void fix_pmode_seg(struct kvm_vcpu *vcpu, int seg,
2630 		struct kvm_segment *save)
2631 {
2632 	if (!emulate_invalid_guest_state) {
2633 		/*
2634 		 * CS and SS RPL should be equal during guest entry according
2635 		 * to VMX spec, but in reality it is not always so. Since vcpu
2636 		 * is in the middle of the transition from real mode to
2637 		 * protected mode it is safe to assume that RPL 0 is a good
2638 		 * default value.
2639 		 */
2640 		if (seg == VCPU_SREG_CS || seg == VCPU_SREG_SS)
2641 			save->selector &= ~SEGMENT_RPL_MASK;
2642 		save->dpl = save->selector & SEGMENT_RPL_MASK;
2643 		save->s = 1;
2644 	}
2645 	vmx_set_segment(vcpu, save, seg);
2646 }
2647 
2648 static void enter_pmode(struct kvm_vcpu *vcpu)
2649 {
2650 	unsigned long flags;
2651 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2652 
2653 	/*
2654 	 * Update real mode segment cache. It may be not up-to-date if sement
2655 	 * register was written while vcpu was in a guest mode.
2656 	 */
2657 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
2658 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
2659 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
2660 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
2661 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
2662 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
2663 
2664 	vmx->rmode.vm86_active = 0;
2665 
2666 	vmx_set_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
2667 
2668 	flags = vmcs_readl(GUEST_RFLAGS);
2669 	flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
2670 	flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
2671 	vmcs_writel(GUEST_RFLAGS, flags);
2672 
2673 	vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
2674 			(vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
2675 
2676 	update_exception_bitmap(vcpu);
2677 
2678 	fix_pmode_seg(vcpu, VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
2679 	fix_pmode_seg(vcpu, VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
2680 	fix_pmode_seg(vcpu, VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
2681 	fix_pmode_seg(vcpu, VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
2682 	fix_pmode_seg(vcpu, VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
2683 	fix_pmode_seg(vcpu, VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
2684 }
2685 
2686 static void fix_rmode_seg(int seg, struct kvm_segment *save)
2687 {
2688 	const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2689 	struct kvm_segment var = *save;
2690 
2691 	var.dpl = 0x3;
2692 	if (seg == VCPU_SREG_CS)
2693 		var.type = 0x3;
2694 
2695 	if (!emulate_invalid_guest_state) {
2696 		var.selector = var.base >> 4;
2697 		var.base = var.base & 0xffff0;
2698 		var.limit = 0xffff;
2699 		var.g = 0;
2700 		var.db = 0;
2701 		var.present = 1;
2702 		var.s = 1;
2703 		var.l = 0;
2704 		var.unusable = 0;
2705 		var.type = 0x3;
2706 		var.avl = 0;
2707 		if (save->base & 0xf)
2708 			printk_once(KERN_WARNING "kvm: segment base is not "
2709 					"paragraph aligned when entering "
2710 					"protected mode (seg=%d)", seg);
2711 	}
2712 
2713 	vmcs_write16(sf->selector, var.selector);
2714 	vmcs_writel(sf->base, var.base);
2715 	vmcs_write32(sf->limit, var.limit);
2716 	vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(&var));
2717 }
2718 
2719 static void enter_rmode(struct kvm_vcpu *vcpu)
2720 {
2721 	unsigned long flags;
2722 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2723 	struct kvm_vmx *kvm_vmx = to_kvm_vmx(vcpu->kvm);
2724 
2725 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
2726 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
2727 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
2728 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
2729 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
2730 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
2731 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
2732 
2733 	vmx->rmode.vm86_active = 1;
2734 
2735 	/*
2736 	 * Very old userspace does not call KVM_SET_TSS_ADDR before entering
2737 	 * vcpu. Warn the user that an update is overdue.
2738 	 */
2739 	if (!kvm_vmx->tss_addr)
2740 		printk_once(KERN_WARNING "kvm: KVM_SET_TSS_ADDR need to be "
2741 			     "called before entering vcpu\n");
2742 
2743 	vmx_segment_cache_clear(vmx);
2744 
2745 	vmcs_writel(GUEST_TR_BASE, kvm_vmx->tss_addr);
2746 	vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
2747 	vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
2748 
2749 	flags = vmcs_readl(GUEST_RFLAGS);
2750 	vmx->rmode.save_rflags = flags;
2751 
2752 	flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
2753 
2754 	vmcs_writel(GUEST_RFLAGS, flags);
2755 	vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
2756 	update_exception_bitmap(vcpu);
2757 
2758 	fix_rmode_seg(VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
2759 	fix_rmode_seg(VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
2760 	fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
2761 	fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
2762 	fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
2763 	fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
2764 
2765 	kvm_mmu_reset_context(vcpu);
2766 }
2767 
2768 void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
2769 {
2770 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2771 	struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
2772 
2773 	if (!msr)
2774 		return;
2775 
2776 	vcpu->arch.efer = efer;
2777 	if (efer & EFER_LMA) {
2778 		vm_entry_controls_setbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
2779 		msr->data = efer;
2780 	} else {
2781 		vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
2782 
2783 		msr->data = efer & ~EFER_LME;
2784 	}
2785 	setup_msrs(vmx);
2786 }
2787 
2788 #ifdef CONFIG_X86_64
2789 
2790 static void enter_lmode(struct kvm_vcpu *vcpu)
2791 {
2792 	u32 guest_tr_ar;
2793 
2794 	vmx_segment_cache_clear(to_vmx(vcpu));
2795 
2796 	guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
2797 	if ((guest_tr_ar & VMX_AR_TYPE_MASK) != VMX_AR_TYPE_BUSY_64_TSS) {
2798 		pr_debug_ratelimited("%s: tss fixup for long mode. \n",
2799 				     __func__);
2800 		vmcs_write32(GUEST_TR_AR_BYTES,
2801 			     (guest_tr_ar & ~VMX_AR_TYPE_MASK)
2802 			     | VMX_AR_TYPE_BUSY_64_TSS);
2803 	}
2804 	vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
2805 }
2806 
2807 static void exit_lmode(struct kvm_vcpu *vcpu)
2808 {
2809 	vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
2810 	vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
2811 }
2812 
2813 #endif
2814 
2815 static void vmx_flush_tlb_gva(struct kvm_vcpu *vcpu, gva_t addr)
2816 {
2817 	int vpid = to_vmx(vcpu)->vpid;
2818 
2819 	if (!vpid_sync_vcpu_addr(vpid, addr))
2820 		vpid_sync_context(vpid);
2821 
2822 	/*
2823 	 * If VPIDs are not supported or enabled, then the above is a no-op.
2824 	 * But we don't really need a TLB flush in that case anyway, because
2825 	 * each VM entry/exit includes an implicit flush when VPID is 0.
2826 	 */
2827 }
2828 
2829 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
2830 {
2831 	ulong cr0_guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
2832 
2833 	vcpu->arch.cr0 &= ~cr0_guest_owned_bits;
2834 	vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & cr0_guest_owned_bits;
2835 }
2836 
2837 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
2838 {
2839 	ulong cr4_guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
2840 
2841 	vcpu->arch.cr4 &= ~cr4_guest_owned_bits;
2842 	vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & cr4_guest_owned_bits;
2843 }
2844 
2845 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
2846 {
2847 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
2848 
2849 	if (!kvm_register_is_dirty(vcpu, VCPU_EXREG_PDPTR))
2850 		return;
2851 
2852 	if (is_pae_paging(vcpu)) {
2853 		vmcs_write64(GUEST_PDPTR0, mmu->pdptrs[0]);
2854 		vmcs_write64(GUEST_PDPTR1, mmu->pdptrs[1]);
2855 		vmcs_write64(GUEST_PDPTR2, mmu->pdptrs[2]);
2856 		vmcs_write64(GUEST_PDPTR3, mmu->pdptrs[3]);
2857 	}
2858 }
2859 
2860 void ept_save_pdptrs(struct kvm_vcpu *vcpu)
2861 {
2862 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
2863 
2864 	if (is_pae_paging(vcpu)) {
2865 		mmu->pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
2866 		mmu->pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
2867 		mmu->pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
2868 		mmu->pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
2869 	}
2870 
2871 	kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
2872 }
2873 
2874 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
2875 					unsigned long cr0,
2876 					struct kvm_vcpu *vcpu)
2877 {
2878 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2879 
2880 	if (!kvm_register_is_available(vcpu, VCPU_EXREG_CR3))
2881 		vmx_cache_reg(vcpu, VCPU_EXREG_CR3);
2882 	if (!(cr0 & X86_CR0_PG)) {
2883 		/* From paging/starting to nonpaging */
2884 		exec_controls_setbit(vmx, CPU_BASED_CR3_LOAD_EXITING |
2885 					  CPU_BASED_CR3_STORE_EXITING);
2886 		vcpu->arch.cr0 = cr0;
2887 		vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
2888 	} else if (!is_paging(vcpu)) {
2889 		/* From nonpaging to paging */
2890 		exec_controls_clearbit(vmx, CPU_BASED_CR3_LOAD_EXITING |
2891 					    CPU_BASED_CR3_STORE_EXITING);
2892 		vcpu->arch.cr0 = cr0;
2893 		vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
2894 	}
2895 
2896 	if (!(cr0 & X86_CR0_WP))
2897 		*hw_cr0 &= ~X86_CR0_WP;
2898 }
2899 
2900 void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
2901 {
2902 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2903 	unsigned long hw_cr0;
2904 
2905 	hw_cr0 = (cr0 & ~KVM_VM_CR0_ALWAYS_OFF);
2906 	if (enable_unrestricted_guest)
2907 		hw_cr0 |= KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
2908 	else {
2909 		hw_cr0 |= KVM_VM_CR0_ALWAYS_ON;
2910 
2911 		if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
2912 			enter_pmode(vcpu);
2913 
2914 		if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
2915 			enter_rmode(vcpu);
2916 	}
2917 
2918 #ifdef CONFIG_X86_64
2919 	if (vcpu->arch.efer & EFER_LME) {
2920 		if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
2921 			enter_lmode(vcpu);
2922 		if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
2923 			exit_lmode(vcpu);
2924 	}
2925 #endif
2926 
2927 	if (enable_ept && !enable_unrestricted_guest)
2928 		ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
2929 
2930 	vmcs_writel(CR0_READ_SHADOW, cr0);
2931 	vmcs_writel(GUEST_CR0, hw_cr0);
2932 	vcpu->arch.cr0 = cr0;
2933 
2934 	/* depends on vcpu->arch.cr0 to be set to a new value */
2935 	vmx->emulation_required = emulation_required(vcpu);
2936 }
2937 
2938 static int get_ept_level(struct kvm_vcpu *vcpu)
2939 {
2940 	if (cpu_has_vmx_ept_5levels() && (cpuid_maxphyaddr(vcpu) > 48))
2941 		return 5;
2942 	return 4;
2943 }
2944 
2945 u64 construct_eptp(struct kvm_vcpu *vcpu, unsigned long root_hpa)
2946 {
2947 	u64 eptp = VMX_EPTP_MT_WB;
2948 
2949 	eptp |= (get_ept_level(vcpu) == 5) ? VMX_EPTP_PWL_5 : VMX_EPTP_PWL_4;
2950 
2951 	if (enable_ept_ad_bits &&
2952 	    (!is_guest_mode(vcpu) || nested_ept_ad_enabled(vcpu)))
2953 		eptp |= VMX_EPTP_AD_ENABLE_BIT;
2954 	eptp |= (root_hpa & PAGE_MASK);
2955 
2956 	return eptp;
2957 }
2958 
2959 void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
2960 {
2961 	struct kvm *kvm = vcpu->kvm;
2962 	bool update_guest_cr3 = true;
2963 	unsigned long guest_cr3;
2964 	u64 eptp;
2965 
2966 	guest_cr3 = cr3;
2967 	if (enable_ept) {
2968 		eptp = construct_eptp(vcpu, cr3);
2969 		vmcs_write64(EPT_POINTER, eptp);
2970 
2971 		if (kvm_x86_ops->tlb_remote_flush) {
2972 			spin_lock(&to_kvm_vmx(kvm)->ept_pointer_lock);
2973 			to_vmx(vcpu)->ept_pointer = eptp;
2974 			to_kvm_vmx(kvm)->ept_pointers_match
2975 				= EPT_POINTERS_CHECK;
2976 			spin_unlock(&to_kvm_vmx(kvm)->ept_pointer_lock);
2977 		}
2978 
2979 		/* Loading vmcs02.GUEST_CR3 is handled by nested VM-Enter. */
2980 		if (is_guest_mode(vcpu))
2981 			update_guest_cr3 = false;
2982 		else if (!enable_unrestricted_guest && !is_paging(vcpu))
2983 			guest_cr3 = to_kvm_vmx(kvm)->ept_identity_map_addr;
2984 		else if (test_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail))
2985 			guest_cr3 = vcpu->arch.cr3;
2986 		else /* vmcs01.GUEST_CR3 is already up-to-date. */
2987 			update_guest_cr3 = false;
2988 		ept_load_pdptrs(vcpu);
2989 	}
2990 
2991 	if (update_guest_cr3)
2992 		vmcs_writel(GUEST_CR3, guest_cr3);
2993 }
2994 
2995 int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
2996 {
2997 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2998 	/*
2999 	 * Pass through host's Machine Check Enable value to hw_cr4, which
3000 	 * is in force while we are in guest mode.  Do not let guests control
3001 	 * this bit, even if host CR4.MCE == 0.
3002 	 */
3003 	unsigned long hw_cr4;
3004 
3005 	hw_cr4 = (cr4_read_shadow() & X86_CR4_MCE) | (cr4 & ~X86_CR4_MCE);
3006 	if (enable_unrestricted_guest)
3007 		hw_cr4 |= KVM_VM_CR4_ALWAYS_ON_UNRESTRICTED_GUEST;
3008 	else if (vmx->rmode.vm86_active)
3009 		hw_cr4 |= KVM_RMODE_VM_CR4_ALWAYS_ON;
3010 	else
3011 		hw_cr4 |= KVM_PMODE_VM_CR4_ALWAYS_ON;
3012 
3013 	if (!boot_cpu_has(X86_FEATURE_UMIP) && vmx_umip_emulated()) {
3014 		if (cr4 & X86_CR4_UMIP) {
3015 			secondary_exec_controls_setbit(vmx, SECONDARY_EXEC_DESC);
3016 			hw_cr4 &= ~X86_CR4_UMIP;
3017 		} else if (!is_guest_mode(vcpu) ||
3018 			!nested_cpu_has2(get_vmcs12(vcpu), SECONDARY_EXEC_DESC)) {
3019 			secondary_exec_controls_clearbit(vmx, SECONDARY_EXEC_DESC);
3020 		}
3021 	}
3022 
3023 	if (cr4 & X86_CR4_VMXE) {
3024 		/*
3025 		 * To use VMXON (and later other VMX instructions), a guest
3026 		 * must first be able to turn on cr4.VMXE (see handle_vmon()).
3027 		 * So basically the check on whether to allow nested VMX
3028 		 * is here.  We operate under the default treatment of SMM,
3029 		 * so VMX cannot be enabled under SMM.
3030 		 */
3031 		if (!nested_vmx_allowed(vcpu) || is_smm(vcpu))
3032 			return 1;
3033 	}
3034 
3035 	if (vmx->nested.vmxon && !nested_cr4_valid(vcpu, cr4))
3036 		return 1;
3037 
3038 	vcpu->arch.cr4 = cr4;
3039 
3040 	if (!enable_unrestricted_guest) {
3041 		if (enable_ept) {
3042 			if (!is_paging(vcpu)) {
3043 				hw_cr4 &= ~X86_CR4_PAE;
3044 				hw_cr4 |= X86_CR4_PSE;
3045 			} else if (!(cr4 & X86_CR4_PAE)) {
3046 				hw_cr4 &= ~X86_CR4_PAE;
3047 			}
3048 		}
3049 
3050 		/*
3051 		 * SMEP/SMAP/PKU is disabled if CPU is in non-paging mode in
3052 		 * hardware.  To emulate this behavior, SMEP/SMAP/PKU needs
3053 		 * to be manually disabled when guest switches to non-paging
3054 		 * mode.
3055 		 *
3056 		 * If !enable_unrestricted_guest, the CPU is always running
3057 		 * with CR0.PG=1 and CR4 needs to be modified.
3058 		 * If enable_unrestricted_guest, the CPU automatically
3059 		 * disables SMEP/SMAP/PKU when the guest sets CR0.PG=0.
3060 		 */
3061 		if (!is_paging(vcpu))
3062 			hw_cr4 &= ~(X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE);
3063 	}
3064 
3065 	vmcs_writel(CR4_READ_SHADOW, cr4);
3066 	vmcs_writel(GUEST_CR4, hw_cr4);
3067 	return 0;
3068 }
3069 
3070 void vmx_get_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg)
3071 {
3072 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3073 	u32 ar;
3074 
3075 	if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3076 		*var = vmx->rmode.segs[seg];
3077 		if (seg == VCPU_SREG_TR
3078 		    || var->selector == vmx_read_guest_seg_selector(vmx, seg))
3079 			return;
3080 		var->base = vmx_read_guest_seg_base(vmx, seg);
3081 		var->selector = vmx_read_guest_seg_selector(vmx, seg);
3082 		return;
3083 	}
3084 	var->base = vmx_read_guest_seg_base(vmx, seg);
3085 	var->limit = vmx_read_guest_seg_limit(vmx, seg);
3086 	var->selector = vmx_read_guest_seg_selector(vmx, seg);
3087 	ar = vmx_read_guest_seg_ar(vmx, seg);
3088 	var->unusable = (ar >> 16) & 1;
3089 	var->type = ar & 15;
3090 	var->s = (ar >> 4) & 1;
3091 	var->dpl = (ar >> 5) & 3;
3092 	/*
3093 	 * Some userspaces do not preserve unusable property. Since usable
3094 	 * segment has to be present according to VMX spec we can use present
3095 	 * property to amend userspace bug by making unusable segment always
3096 	 * nonpresent. vmx_segment_access_rights() already marks nonpresent
3097 	 * segment as unusable.
3098 	 */
3099 	var->present = !var->unusable;
3100 	var->avl = (ar >> 12) & 1;
3101 	var->l = (ar >> 13) & 1;
3102 	var->db = (ar >> 14) & 1;
3103 	var->g = (ar >> 15) & 1;
3104 }
3105 
3106 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
3107 {
3108 	struct kvm_segment s;
3109 
3110 	if (to_vmx(vcpu)->rmode.vm86_active) {
3111 		vmx_get_segment(vcpu, &s, seg);
3112 		return s.base;
3113 	}
3114 	return vmx_read_guest_seg_base(to_vmx(vcpu), seg);
3115 }
3116 
3117 int vmx_get_cpl(struct kvm_vcpu *vcpu)
3118 {
3119 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3120 
3121 	if (unlikely(vmx->rmode.vm86_active))
3122 		return 0;
3123 	else {
3124 		int ar = vmx_read_guest_seg_ar(vmx, VCPU_SREG_SS);
3125 		return VMX_AR_DPL(ar);
3126 	}
3127 }
3128 
3129 static u32 vmx_segment_access_rights(struct kvm_segment *var)
3130 {
3131 	u32 ar;
3132 
3133 	if (var->unusable || !var->present)
3134 		ar = 1 << 16;
3135 	else {
3136 		ar = var->type & 15;
3137 		ar |= (var->s & 1) << 4;
3138 		ar |= (var->dpl & 3) << 5;
3139 		ar |= (var->present & 1) << 7;
3140 		ar |= (var->avl & 1) << 12;
3141 		ar |= (var->l & 1) << 13;
3142 		ar |= (var->db & 1) << 14;
3143 		ar |= (var->g & 1) << 15;
3144 	}
3145 
3146 	return ar;
3147 }
3148 
3149 void vmx_set_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg)
3150 {
3151 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3152 	const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3153 
3154 	vmx_segment_cache_clear(vmx);
3155 
3156 	if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3157 		vmx->rmode.segs[seg] = *var;
3158 		if (seg == VCPU_SREG_TR)
3159 			vmcs_write16(sf->selector, var->selector);
3160 		else if (var->s)
3161 			fix_rmode_seg(seg, &vmx->rmode.segs[seg]);
3162 		goto out;
3163 	}
3164 
3165 	vmcs_writel(sf->base, var->base);
3166 	vmcs_write32(sf->limit, var->limit);
3167 	vmcs_write16(sf->selector, var->selector);
3168 
3169 	/*
3170 	 *   Fix the "Accessed" bit in AR field of segment registers for older
3171 	 * qemu binaries.
3172 	 *   IA32 arch specifies that at the time of processor reset the
3173 	 * "Accessed" bit in the AR field of segment registers is 1. And qemu
3174 	 * is setting it to 0 in the userland code. This causes invalid guest
3175 	 * state vmexit when "unrestricted guest" mode is turned on.
3176 	 *    Fix for this setup issue in cpu_reset is being pushed in the qemu
3177 	 * tree. Newer qemu binaries with that qemu fix would not need this
3178 	 * kvm hack.
3179 	 */
3180 	if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
3181 		var->type |= 0x1; /* Accessed */
3182 
3183 	vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(var));
3184 
3185 out:
3186 	vmx->emulation_required = emulation_required(vcpu);
3187 }
3188 
3189 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3190 {
3191 	u32 ar = vmx_read_guest_seg_ar(to_vmx(vcpu), VCPU_SREG_CS);
3192 
3193 	*db = (ar >> 14) & 1;
3194 	*l = (ar >> 13) & 1;
3195 }
3196 
3197 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3198 {
3199 	dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
3200 	dt->address = vmcs_readl(GUEST_IDTR_BASE);
3201 }
3202 
3203 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3204 {
3205 	vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
3206 	vmcs_writel(GUEST_IDTR_BASE, dt->address);
3207 }
3208 
3209 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3210 {
3211 	dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
3212 	dt->address = vmcs_readl(GUEST_GDTR_BASE);
3213 }
3214 
3215 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3216 {
3217 	vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
3218 	vmcs_writel(GUEST_GDTR_BASE, dt->address);
3219 }
3220 
3221 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
3222 {
3223 	struct kvm_segment var;
3224 	u32 ar;
3225 
3226 	vmx_get_segment(vcpu, &var, seg);
3227 	var.dpl = 0x3;
3228 	if (seg == VCPU_SREG_CS)
3229 		var.type = 0x3;
3230 	ar = vmx_segment_access_rights(&var);
3231 
3232 	if (var.base != (var.selector << 4))
3233 		return false;
3234 	if (var.limit != 0xffff)
3235 		return false;
3236 	if (ar != 0xf3)
3237 		return false;
3238 
3239 	return true;
3240 }
3241 
3242 static bool code_segment_valid(struct kvm_vcpu *vcpu)
3243 {
3244 	struct kvm_segment cs;
3245 	unsigned int cs_rpl;
3246 
3247 	vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3248 	cs_rpl = cs.selector & SEGMENT_RPL_MASK;
3249 
3250 	if (cs.unusable)
3251 		return false;
3252 	if (~cs.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_ACCESSES_MASK))
3253 		return false;
3254 	if (!cs.s)
3255 		return false;
3256 	if (cs.type & VMX_AR_TYPE_WRITEABLE_MASK) {
3257 		if (cs.dpl > cs_rpl)
3258 			return false;
3259 	} else {
3260 		if (cs.dpl != cs_rpl)
3261 			return false;
3262 	}
3263 	if (!cs.present)
3264 		return false;
3265 
3266 	/* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
3267 	return true;
3268 }
3269 
3270 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
3271 {
3272 	struct kvm_segment ss;
3273 	unsigned int ss_rpl;
3274 
3275 	vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3276 	ss_rpl = ss.selector & SEGMENT_RPL_MASK;
3277 
3278 	if (ss.unusable)
3279 		return true;
3280 	if (ss.type != 3 && ss.type != 7)
3281 		return false;
3282 	if (!ss.s)
3283 		return false;
3284 	if (ss.dpl != ss_rpl) /* DPL != RPL */
3285 		return false;
3286 	if (!ss.present)
3287 		return false;
3288 
3289 	return true;
3290 }
3291 
3292 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
3293 {
3294 	struct kvm_segment var;
3295 	unsigned int rpl;
3296 
3297 	vmx_get_segment(vcpu, &var, seg);
3298 	rpl = var.selector & SEGMENT_RPL_MASK;
3299 
3300 	if (var.unusable)
3301 		return true;
3302 	if (!var.s)
3303 		return false;
3304 	if (!var.present)
3305 		return false;
3306 	if (~var.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_WRITEABLE_MASK)) {
3307 		if (var.dpl < rpl) /* DPL < RPL */
3308 			return false;
3309 	}
3310 
3311 	/* TODO: Add other members to kvm_segment_field to allow checking for other access
3312 	 * rights flags
3313 	 */
3314 	return true;
3315 }
3316 
3317 static bool tr_valid(struct kvm_vcpu *vcpu)
3318 {
3319 	struct kvm_segment tr;
3320 
3321 	vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
3322 
3323 	if (tr.unusable)
3324 		return false;
3325 	if (tr.selector & SEGMENT_TI_MASK)	/* TI = 1 */
3326 		return false;
3327 	if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
3328 		return false;
3329 	if (!tr.present)
3330 		return false;
3331 
3332 	return true;
3333 }
3334 
3335 static bool ldtr_valid(struct kvm_vcpu *vcpu)
3336 {
3337 	struct kvm_segment ldtr;
3338 
3339 	vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
3340 
3341 	if (ldtr.unusable)
3342 		return true;
3343 	if (ldtr.selector & SEGMENT_TI_MASK)	/* TI = 1 */
3344 		return false;
3345 	if (ldtr.type != 2)
3346 		return false;
3347 	if (!ldtr.present)
3348 		return false;
3349 
3350 	return true;
3351 }
3352 
3353 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
3354 {
3355 	struct kvm_segment cs, ss;
3356 
3357 	vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3358 	vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3359 
3360 	return ((cs.selector & SEGMENT_RPL_MASK) ==
3361 		 (ss.selector & SEGMENT_RPL_MASK));
3362 }
3363 
3364 /*
3365  * Check if guest state is valid. Returns true if valid, false if
3366  * not.
3367  * We assume that registers are always usable
3368  */
3369 static bool guest_state_valid(struct kvm_vcpu *vcpu)
3370 {
3371 	if (enable_unrestricted_guest)
3372 		return true;
3373 
3374 	/* real mode guest state checks */
3375 	if (!is_protmode(vcpu) || (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
3376 		if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
3377 			return false;
3378 		if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
3379 			return false;
3380 		if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
3381 			return false;
3382 		if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
3383 			return false;
3384 		if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
3385 			return false;
3386 		if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
3387 			return false;
3388 	} else {
3389 	/* protected mode guest state checks */
3390 		if (!cs_ss_rpl_check(vcpu))
3391 			return false;
3392 		if (!code_segment_valid(vcpu))
3393 			return false;
3394 		if (!stack_segment_valid(vcpu))
3395 			return false;
3396 		if (!data_segment_valid(vcpu, VCPU_SREG_DS))
3397 			return false;
3398 		if (!data_segment_valid(vcpu, VCPU_SREG_ES))
3399 			return false;
3400 		if (!data_segment_valid(vcpu, VCPU_SREG_FS))
3401 			return false;
3402 		if (!data_segment_valid(vcpu, VCPU_SREG_GS))
3403 			return false;
3404 		if (!tr_valid(vcpu))
3405 			return false;
3406 		if (!ldtr_valid(vcpu))
3407 			return false;
3408 	}
3409 	/* TODO:
3410 	 * - Add checks on RIP
3411 	 * - Add checks on RFLAGS
3412 	 */
3413 
3414 	return true;
3415 }
3416 
3417 static int init_rmode_tss(struct kvm *kvm)
3418 {
3419 	gfn_t fn;
3420 	u16 data = 0;
3421 	int idx, r;
3422 
3423 	idx = srcu_read_lock(&kvm->srcu);
3424 	fn = to_kvm_vmx(kvm)->tss_addr >> PAGE_SHIFT;
3425 	r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3426 	if (r < 0)
3427 		goto out;
3428 	data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
3429 	r = kvm_write_guest_page(kvm, fn++, &data,
3430 			TSS_IOPB_BASE_OFFSET, sizeof(u16));
3431 	if (r < 0)
3432 		goto out;
3433 	r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
3434 	if (r < 0)
3435 		goto out;
3436 	r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3437 	if (r < 0)
3438 		goto out;
3439 	data = ~0;
3440 	r = kvm_write_guest_page(kvm, fn, &data,
3441 				 RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
3442 				 sizeof(u8));
3443 out:
3444 	srcu_read_unlock(&kvm->srcu, idx);
3445 	return r;
3446 }
3447 
3448 static int init_rmode_identity_map(struct kvm *kvm)
3449 {
3450 	struct kvm_vmx *kvm_vmx = to_kvm_vmx(kvm);
3451 	int i, r = 0;
3452 	kvm_pfn_t identity_map_pfn;
3453 	u32 tmp;
3454 
3455 	/* Protect kvm_vmx->ept_identity_pagetable_done. */
3456 	mutex_lock(&kvm->slots_lock);
3457 
3458 	if (likely(kvm_vmx->ept_identity_pagetable_done))
3459 		goto out;
3460 
3461 	if (!kvm_vmx->ept_identity_map_addr)
3462 		kvm_vmx->ept_identity_map_addr = VMX_EPT_IDENTITY_PAGETABLE_ADDR;
3463 	identity_map_pfn = kvm_vmx->ept_identity_map_addr >> PAGE_SHIFT;
3464 
3465 	r = __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
3466 				    kvm_vmx->ept_identity_map_addr, PAGE_SIZE);
3467 	if (r < 0)
3468 		goto out;
3469 
3470 	r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
3471 	if (r < 0)
3472 		goto out;
3473 	/* Set up identity-mapping pagetable for EPT in real mode */
3474 	for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
3475 		tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
3476 			_PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
3477 		r = kvm_write_guest_page(kvm, identity_map_pfn,
3478 				&tmp, i * sizeof(tmp), sizeof(tmp));
3479 		if (r < 0)
3480 			goto out;
3481 	}
3482 	kvm_vmx->ept_identity_pagetable_done = true;
3483 
3484 out:
3485 	mutex_unlock(&kvm->slots_lock);
3486 	return r;
3487 }
3488 
3489 static void seg_setup(int seg)
3490 {
3491 	const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3492 	unsigned int ar;
3493 
3494 	vmcs_write16(sf->selector, 0);
3495 	vmcs_writel(sf->base, 0);
3496 	vmcs_write32(sf->limit, 0xffff);
3497 	ar = 0x93;
3498 	if (seg == VCPU_SREG_CS)
3499 		ar |= 0x08; /* code segment */
3500 
3501 	vmcs_write32(sf->ar_bytes, ar);
3502 }
3503 
3504 static int alloc_apic_access_page(struct kvm *kvm)
3505 {
3506 	struct page *page;
3507 	int r = 0;
3508 
3509 	mutex_lock(&kvm->slots_lock);
3510 	if (kvm->arch.apic_access_page_done)
3511 		goto out;
3512 	r = __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
3513 				    APIC_DEFAULT_PHYS_BASE, PAGE_SIZE);
3514 	if (r)
3515 		goto out;
3516 
3517 	page = gfn_to_page(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
3518 	if (is_error_page(page)) {
3519 		r = -EFAULT;
3520 		goto out;
3521 	}
3522 
3523 	/*
3524 	 * Do not pin the page in memory, so that memory hot-unplug
3525 	 * is able to migrate it.
3526 	 */
3527 	put_page(page);
3528 	kvm->arch.apic_access_page_done = true;
3529 out:
3530 	mutex_unlock(&kvm->slots_lock);
3531 	return r;
3532 }
3533 
3534 int allocate_vpid(void)
3535 {
3536 	int vpid;
3537 
3538 	if (!enable_vpid)
3539 		return 0;
3540 	spin_lock(&vmx_vpid_lock);
3541 	vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
3542 	if (vpid < VMX_NR_VPIDS)
3543 		__set_bit(vpid, vmx_vpid_bitmap);
3544 	else
3545 		vpid = 0;
3546 	spin_unlock(&vmx_vpid_lock);
3547 	return vpid;
3548 }
3549 
3550 void free_vpid(int vpid)
3551 {
3552 	if (!enable_vpid || vpid == 0)
3553 		return;
3554 	spin_lock(&vmx_vpid_lock);
3555 	__clear_bit(vpid, vmx_vpid_bitmap);
3556 	spin_unlock(&vmx_vpid_lock);
3557 }
3558 
3559 static __always_inline void vmx_disable_intercept_for_msr(unsigned long *msr_bitmap,
3560 							  u32 msr, int type)
3561 {
3562 	int f = sizeof(unsigned long);
3563 
3564 	if (!cpu_has_vmx_msr_bitmap())
3565 		return;
3566 
3567 	if (static_branch_unlikely(&enable_evmcs))
3568 		evmcs_touch_msr_bitmap();
3569 
3570 	/*
3571 	 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
3572 	 * have the write-low and read-high bitmap offsets the wrong way round.
3573 	 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
3574 	 */
3575 	if (msr <= 0x1fff) {
3576 		if (type & MSR_TYPE_R)
3577 			/* read-low */
3578 			__clear_bit(msr, msr_bitmap + 0x000 / f);
3579 
3580 		if (type & MSR_TYPE_W)
3581 			/* write-low */
3582 			__clear_bit(msr, msr_bitmap + 0x800 / f);
3583 
3584 	} else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
3585 		msr &= 0x1fff;
3586 		if (type & MSR_TYPE_R)
3587 			/* read-high */
3588 			__clear_bit(msr, msr_bitmap + 0x400 / f);
3589 
3590 		if (type & MSR_TYPE_W)
3591 			/* write-high */
3592 			__clear_bit(msr, msr_bitmap + 0xc00 / f);
3593 
3594 	}
3595 }
3596 
3597 static __always_inline void vmx_enable_intercept_for_msr(unsigned long *msr_bitmap,
3598 							 u32 msr, int type)
3599 {
3600 	int f = sizeof(unsigned long);
3601 
3602 	if (!cpu_has_vmx_msr_bitmap())
3603 		return;
3604 
3605 	if (static_branch_unlikely(&enable_evmcs))
3606 		evmcs_touch_msr_bitmap();
3607 
3608 	/*
3609 	 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
3610 	 * have the write-low and read-high bitmap offsets the wrong way round.
3611 	 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
3612 	 */
3613 	if (msr <= 0x1fff) {
3614 		if (type & MSR_TYPE_R)
3615 			/* read-low */
3616 			__set_bit(msr, msr_bitmap + 0x000 / f);
3617 
3618 		if (type & MSR_TYPE_W)
3619 			/* write-low */
3620 			__set_bit(msr, msr_bitmap + 0x800 / f);
3621 
3622 	} else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
3623 		msr &= 0x1fff;
3624 		if (type & MSR_TYPE_R)
3625 			/* read-high */
3626 			__set_bit(msr, msr_bitmap + 0x400 / f);
3627 
3628 		if (type & MSR_TYPE_W)
3629 			/* write-high */
3630 			__set_bit(msr, msr_bitmap + 0xc00 / f);
3631 
3632 	}
3633 }
3634 
3635 static __always_inline void vmx_set_intercept_for_msr(unsigned long *msr_bitmap,
3636 			     			      u32 msr, int type, bool value)
3637 {
3638 	if (value)
3639 		vmx_enable_intercept_for_msr(msr_bitmap, msr, type);
3640 	else
3641 		vmx_disable_intercept_for_msr(msr_bitmap, msr, type);
3642 }
3643 
3644 static u8 vmx_msr_bitmap_mode(struct kvm_vcpu *vcpu)
3645 {
3646 	u8 mode = 0;
3647 
3648 	if (cpu_has_secondary_exec_ctrls() &&
3649 	    (secondary_exec_controls_get(to_vmx(vcpu)) &
3650 	     SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE)) {
3651 		mode |= MSR_BITMAP_MODE_X2APIC;
3652 		if (enable_apicv && kvm_vcpu_apicv_active(vcpu))
3653 			mode |= MSR_BITMAP_MODE_X2APIC_APICV;
3654 	}
3655 
3656 	return mode;
3657 }
3658 
3659 static void vmx_update_msr_bitmap_x2apic(unsigned long *msr_bitmap,
3660 					 u8 mode)
3661 {
3662 	int msr;
3663 
3664 	for (msr = 0x800; msr <= 0x8ff; msr += BITS_PER_LONG) {
3665 		unsigned word = msr / BITS_PER_LONG;
3666 		msr_bitmap[word] = (mode & MSR_BITMAP_MODE_X2APIC_APICV) ? 0 : ~0;
3667 		msr_bitmap[word + (0x800 / sizeof(long))] = ~0;
3668 	}
3669 
3670 	if (mode & MSR_BITMAP_MODE_X2APIC) {
3671 		/*
3672 		 * TPR reads and writes can be virtualized even if virtual interrupt
3673 		 * delivery is not in use.
3674 		 */
3675 		vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_TASKPRI), MSR_TYPE_RW);
3676 		if (mode & MSR_BITMAP_MODE_X2APIC_APICV) {
3677 			vmx_enable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_TMCCT), MSR_TYPE_R);
3678 			vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_EOI), MSR_TYPE_W);
3679 			vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_SELF_IPI), MSR_TYPE_W);
3680 		}
3681 	}
3682 }
3683 
3684 void vmx_update_msr_bitmap(struct kvm_vcpu *vcpu)
3685 {
3686 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3687 	unsigned long *msr_bitmap = vmx->vmcs01.msr_bitmap;
3688 	u8 mode = vmx_msr_bitmap_mode(vcpu);
3689 	u8 changed = mode ^ vmx->msr_bitmap_mode;
3690 
3691 	if (!changed)
3692 		return;
3693 
3694 	if (changed & (MSR_BITMAP_MODE_X2APIC | MSR_BITMAP_MODE_X2APIC_APICV))
3695 		vmx_update_msr_bitmap_x2apic(msr_bitmap, mode);
3696 
3697 	vmx->msr_bitmap_mode = mode;
3698 }
3699 
3700 void pt_update_intercept_for_msr(struct vcpu_vmx *vmx)
3701 {
3702 	unsigned long *msr_bitmap = vmx->vmcs01.msr_bitmap;
3703 	bool flag = !(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN);
3704 	u32 i;
3705 
3706 	vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_STATUS,
3707 							MSR_TYPE_RW, flag);
3708 	vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_OUTPUT_BASE,
3709 							MSR_TYPE_RW, flag);
3710 	vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_OUTPUT_MASK,
3711 							MSR_TYPE_RW, flag);
3712 	vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_CR3_MATCH,
3713 							MSR_TYPE_RW, flag);
3714 	for (i = 0; i < vmx->pt_desc.addr_range; i++) {
3715 		vmx_set_intercept_for_msr(msr_bitmap,
3716 			MSR_IA32_RTIT_ADDR0_A + i * 2, MSR_TYPE_RW, flag);
3717 		vmx_set_intercept_for_msr(msr_bitmap,
3718 			MSR_IA32_RTIT_ADDR0_B + i * 2, MSR_TYPE_RW, flag);
3719 	}
3720 }
3721 
3722 static bool vmx_get_enable_apicv(struct kvm *kvm)
3723 {
3724 	return enable_apicv;
3725 }
3726 
3727 static bool vmx_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
3728 {
3729 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3730 	void *vapic_page;
3731 	u32 vppr;
3732 	int rvi;
3733 
3734 	if (WARN_ON_ONCE(!is_guest_mode(vcpu)) ||
3735 		!nested_cpu_has_vid(get_vmcs12(vcpu)) ||
3736 		WARN_ON_ONCE(!vmx->nested.virtual_apic_map.gfn))
3737 		return false;
3738 
3739 	rvi = vmx_get_rvi();
3740 
3741 	vapic_page = vmx->nested.virtual_apic_map.hva;
3742 	vppr = *((u32 *)(vapic_page + APIC_PROCPRI));
3743 
3744 	return ((rvi & 0xf0) > (vppr & 0xf0));
3745 }
3746 
3747 static inline bool kvm_vcpu_trigger_posted_interrupt(struct kvm_vcpu *vcpu,
3748 						     bool nested)
3749 {
3750 #ifdef CONFIG_SMP
3751 	int pi_vec = nested ? POSTED_INTR_NESTED_VECTOR : POSTED_INTR_VECTOR;
3752 
3753 	if (vcpu->mode == IN_GUEST_MODE) {
3754 		/*
3755 		 * The vector of interrupt to be delivered to vcpu had
3756 		 * been set in PIR before this function.
3757 		 *
3758 		 * Following cases will be reached in this block, and
3759 		 * we always send a notification event in all cases as
3760 		 * explained below.
3761 		 *
3762 		 * Case 1: vcpu keeps in non-root mode. Sending a
3763 		 * notification event posts the interrupt to vcpu.
3764 		 *
3765 		 * Case 2: vcpu exits to root mode and is still
3766 		 * runnable. PIR will be synced to vIRR before the
3767 		 * next vcpu entry. Sending a notification event in
3768 		 * this case has no effect, as vcpu is not in root
3769 		 * mode.
3770 		 *
3771 		 * Case 3: vcpu exits to root mode and is blocked.
3772 		 * vcpu_block() has already synced PIR to vIRR and
3773 		 * never blocks vcpu if vIRR is not cleared. Therefore,
3774 		 * a blocked vcpu here does not wait for any requested
3775 		 * interrupts in PIR, and sending a notification event
3776 		 * which has no effect is safe here.
3777 		 */
3778 
3779 		apic->send_IPI_mask(get_cpu_mask(vcpu->cpu), pi_vec);
3780 		return true;
3781 	}
3782 #endif
3783 	return false;
3784 }
3785 
3786 static int vmx_deliver_nested_posted_interrupt(struct kvm_vcpu *vcpu,
3787 						int vector)
3788 {
3789 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3790 
3791 	if (is_guest_mode(vcpu) &&
3792 	    vector == vmx->nested.posted_intr_nv) {
3793 		/*
3794 		 * If a posted intr is not recognized by hardware,
3795 		 * we will accomplish it in the next vmentry.
3796 		 */
3797 		vmx->nested.pi_pending = true;
3798 		kvm_make_request(KVM_REQ_EVENT, vcpu);
3799 		/* the PIR and ON have been set by L1. */
3800 		if (!kvm_vcpu_trigger_posted_interrupt(vcpu, true))
3801 			kvm_vcpu_kick(vcpu);
3802 		return 0;
3803 	}
3804 	return -1;
3805 }
3806 /*
3807  * Send interrupt to vcpu via posted interrupt way.
3808  * 1. If target vcpu is running(non-root mode), send posted interrupt
3809  * notification to vcpu and hardware will sync PIR to vIRR atomically.
3810  * 2. If target vcpu isn't running(root mode), kick it to pick up the
3811  * interrupt from PIR in next vmentry.
3812  */
3813 static void vmx_deliver_posted_interrupt(struct kvm_vcpu *vcpu, int vector)
3814 {
3815 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3816 	int r;
3817 
3818 	r = vmx_deliver_nested_posted_interrupt(vcpu, vector);
3819 	if (!r)
3820 		return;
3821 
3822 	if (pi_test_and_set_pir(vector, &vmx->pi_desc))
3823 		return;
3824 
3825 	/* If a previous notification has sent the IPI, nothing to do.  */
3826 	if (pi_test_and_set_on(&vmx->pi_desc))
3827 		return;
3828 
3829 	if (!kvm_vcpu_trigger_posted_interrupt(vcpu, false))
3830 		kvm_vcpu_kick(vcpu);
3831 }
3832 
3833 /*
3834  * Set up the vmcs's constant host-state fields, i.e., host-state fields that
3835  * will not change in the lifetime of the guest.
3836  * Note that host-state that does change is set elsewhere. E.g., host-state
3837  * that is set differently for each CPU is set in vmx_vcpu_load(), not here.
3838  */
3839 void vmx_set_constant_host_state(struct vcpu_vmx *vmx)
3840 {
3841 	u32 low32, high32;
3842 	unsigned long tmpl;
3843 	unsigned long cr0, cr3, cr4;
3844 
3845 	cr0 = read_cr0();
3846 	WARN_ON(cr0 & X86_CR0_TS);
3847 	vmcs_writel(HOST_CR0, cr0);  /* 22.2.3 */
3848 
3849 	/*
3850 	 * Save the most likely value for this task's CR3 in the VMCS.
3851 	 * We can't use __get_current_cr3_fast() because we're not atomic.
3852 	 */
3853 	cr3 = __read_cr3();
3854 	vmcs_writel(HOST_CR3, cr3);		/* 22.2.3  FIXME: shadow tables */
3855 	vmx->loaded_vmcs->host_state.cr3 = cr3;
3856 
3857 	/* Save the most likely value for this task's CR4 in the VMCS. */
3858 	cr4 = cr4_read_shadow();
3859 	vmcs_writel(HOST_CR4, cr4);			/* 22.2.3, 22.2.5 */
3860 	vmx->loaded_vmcs->host_state.cr4 = cr4;
3861 
3862 	vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS);  /* 22.2.4 */
3863 #ifdef CONFIG_X86_64
3864 	/*
3865 	 * Load null selectors, so we can avoid reloading them in
3866 	 * vmx_prepare_switch_to_host(), in case userspace uses
3867 	 * the null selectors too (the expected case).
3868 	 */
3869 	vmcs_write16(HOST_DS_SELECTOR, 0);
3870 	vmcs_write16(HOST_ES_SELECTOR, 0);
3871 #else
3872 	vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
3873 	vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
3874 #endif
3875 	vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
3876 	vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8);  /* 22.2.4 */
3877 
3878 	vmcs_writel(HOST_IDTR_BASE, host_idt_base);   /* 22.2.4 */
3879 
3880 	vmcs_writel(HOST_RIP, (unsigned long)vmx_vmexit); /* 22.2.5 */
3881 
3882 	rdmsr(MSR_IA32_SYSENTER_CS, low32, high32);
3883 	vmcs_write32(HOST_IA32_SYSENTER_CS, low32);
3884 	rdmsrl(MSR_IA32_SYSENTER_EIP, tmpl);
3885 	vmcs_writel(HOST_IA32_SYSENTER_EIP, tmpl);   /* 22.2.3 */
3886 
3887 	if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
3888 		rdmsr(MSR_IA32_CR_PAT, low32, high32);
3889 		vmcs_write64(HOST_IA32_PAT, low32 | ((u64) high32 << 32));
3890 	}
3891 
3892 	if (cpu_has_load_ia32_efer())
3893 		vmcs_write64(HOST_IA32_EFER, host_efer);
3894 }
3895 
3896 void set_cr4_guest_host_mask(struct vcpu_vmx *vmx)
3897 {
3898 	vmx->vcpu.arch.cr4_guest_owned_bits = KVM_CR4_GUEST_OWNED_BITS;
3899 	if (enable_ept)
3900 		vmx->vcpu.arch.cr4_guest_owned_bits |= X86_CR4_PGE;
3901 	if (is_guest_mode(&vmx->vcpu))
3902 		vmx->vcpu.arch.cr4_guest_owned_bits &=
3903 			~get_vmcs12(&vmx->vcpu)->cr4_guest_host_mask;
3904 	vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits);
3905 }
3906 
3907 u32 vmx_pin_based_exec_ctrl(struct vcpu_vmx *vmx)
3908 {
3909 	u32 pin_based_exec_ctrl = vmcs_config.pin_based_exec_ctrl;
3910 
3911 	if (!kvm_vcpu_apicv_active(&vmx->vcpu))
3912 		pin_based_exec_ctrl &= ~PIN_BASED_POSTED_INTR;
3913 
3914 	if (!enable_vnmi)
3915 		pin_based_exec_ctrl &= ~PIN_BASED_VIRTUAL_NMIS;
3916 
3917 	if (!enable_preemption_timer)
3918 		pin_based_exec_ctrl &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
3919 
3920 	return pin_based_exec_ctrl;
3921 }
3922 
3923 static void vmx_refresh_apicv_exec_ctrl(struct kvm_vcpu *vcpu)
3924 {
3925 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3926 
3927 	pin_controls_set(vmx, vmx_pin_based_exec_ctrl(vmx));
3928 	if (cpu_has_secondary_exec_ctrls()) {
3929 		if (kvm_vcpu_apicv_active(vcpu))
3930 			secondary_exec_controls_setbit(vmx,
3931 				      SECONDARY_EXEC_APIC_REGISTER_VIRT |
3932 				      SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
3933 		else
3934 			secondary_exec_controls_clearbit(vmx,
3935 					SECONDARY_EXEC_APIC_REGISTER_VIRT |
3936 					SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
3937 	}
3938 
3939 	if (cpu_has_vmx_msr_bitmap())
3940 		vmx_update_msr_bitmap(vcpu);
3941 }
3942 
3943 u32 vmx_exec_control(struct vcpu_vmx *vmx)
3944 {
3945 	u32 exec_control = vmcs_config.cpu_based_exec_ctrl;
3946 
3947 	if (vmx->vcpu.arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)
3948 		exec_control &= ~CPU_BASED_MOV_DR_EXITING;
3949 
3950 	if (!cpu_need_tpr_shadow(&vmx->vcpu)) {
3951 		exec_control &= ~CPU_BASED_TPR_SHADOW;
3952 #ifdef CONFIG_X86_64
3953 		exec_control |= CPU_BASED_CR8_STORE_EXITING |
3954 				CPU_BASED_CR8_LOAD_EXITING;
3955 #endif
3956 	}
3957 	if (!enable_ept)
3958 		exec_control |= CPU_BASED_CR3_STORE_EXITING |
3959 				CPU_BASED_CR3_LOAD_EXITING  |
3960 				CPU_BASED_INVLPG_EXITING;
3961 	if (kvm_mwait_in_guest(vmx->vcpu.kvm))
3962 		exec_control &= ~(CPU_BASED_MWAIT_EXITING |
3963 				CPU_BASED_MONITOR_EXITING);
3964 	if (kvm_hlt_in_guest(vmx->vcpu.kvm))
3965 		exec_control &= ~CPU_BASED_HLT_EXITING;
3966 	return exec_control;
3967 }
3968 
3969 
3970 static void vmx_compute_secondary_exec_control(struct vcpu_vmx *vmx)
3971 {
3972 	struct kvm_vcpu *vcpu = &vmx->vcpu;
3973 
3974 	u32 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
3975 
3976 	if (pt_mode == PT_MODE_SYSTEM)
3977 		exec_control &= ~(SECONDARY_EXEC_PT_USE_GPA | SECONDARY_EXEC_PT_CONCEAL_VMX);
3978 	if (!cpu_need_virtualize_apic_accesses(vcpu))
3979 		exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
3980 	if (vmx->vpid == 0)
3981 		exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
3982 	if (!enable_ept) {
3983 		exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
3984 		enable_unrestricted_guest = 0;
3985 	}
3986 	if (!enable_unrestricted_guest)
3987 		exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
3988 	if (kvm_pause_in_guest(vmx->vcpu.kvm))
3989 		exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
3990 	if (!kvm_vcpu_apicv_active(vcpu))
3991 		exec_control &= ~(SECONDARY_EXEC_APIC_REGISTER_VIRT |
3992 				  SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
3993 	exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
3994 
3995 	/* SECONDARY_EXEC_DESC is enabled/disabled on writes to CR4.UMIP,
3996 	 * in vmx_set_cr4.  */
3997 	exec_control &= ~SECONDARY_EXEC_DESC;
3998 
3999 	/* SECONDARY_EXEC_SHADOW_VMCS is enabled when L1 executes VMPTRLD
4000 	   (handle_vmptrld).
4001 	   We can NOT enable shadow_vmcs here because we don't have yet
4002 	   a current VMCS12
4003 	*/
4004 	exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
4005 
4006 	if (!enable_pml)
4007 		exec_control &= ~SECONDARY_EXEC_ENABLE_PML;
4008 
4009 	if (vmx_xsaves_supported()) {
4010 		/* Exposing XSAVES only when XSAVE is exposed */
4011 		bool xsaves_enabled =
4012 			boot_cpu_has(X86_FEATURE_XSAVE) &&
4013 			guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
4014 			guest_cpuid_has(vcpu, X86_FEATURE_XSAVES);
4015 
4016 		vcpu->arch.xsaves_enabled = xsaves_enabled;
4017 
4018 		if (!xsaves_enabled)
4019 			exec_control &= ~SECONDARY_EXEC_XSAVES;
4020 
4021 		if (nested) {
4022 			if (xsaves_enabled)
4023 				vmx->nested.msrs.secondary_ctls_high |=
4024 					SECONDARY_EXEC_XSAVES;
4025 			else
4026 				vmx->nested.msrs.secondary_ctls_high &=
4027 					~SECONDARY_EXEC_XSAVES;
4028 		}
4029 	}
4030 
4031 	if (vmx_rdtscp_supported()) {
4032 		bool rdtscp_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP);
4033 		if (!rdtscp_enabled)
4034 			exec_control &= ~SECONDARY_EXEC_RDTSCP;
4035 
4036 		if (nested) {
4037 			if (rdtscp_enabled)
4038 				vmx->nested.msrs.secondary_ctls_high |=
4039 					SECONDARY_EXEC_RDTSCP;
4040 			else
4041 				vmx->nested.msrs.secondary_ctls_high &=
4042 					~SECONDARY_EXEC_RDTSCP;
4043 		}
4044 	}
4045 
4046 	if (vmx_invpcid_supported()) {
4047 		/* Exposing INVPCID only when PCID is exposed */
4048 		bool invpcid_enabled =
4049 			guest_cpuid_has(vcpu, X86_FEATURE_INVPCID) &&
4050 			guest_cpuid_has(vcpu, X86_FEATURE_PCID);
4051 
4052 		if (!invpcid_enabled) {
4053 			exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID;
4054 			guest_cpuid_clear(vcpu, X86_FEATURE_INVPCID);
4055 		}
4056 
4057 		if (nested) {
4058 			if (invpcid_enabled)
4059 				vmx->nested.msrs.secondary_ctls_high |=
4060 					SECONDARY_EXEC_ENABLE_INVPCID;
4061 			else
4062 				vmx->nested.msrs.secondary_ctls_high &=
4063 					~SECONDARY_EXEC_ENABLE_INVPCID;
4064 		}
4065 	}
4066 
4067 	if (vmx_rdrand_supported()) {
4068 		bool rdrand_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDRAND);
4069 		if (rdrand_enabled)
4070 			exec_control &= ~SECONDARY_EXEC_RDRAND_EXITING;
4071 
4072 		if (nested) {
4073 			if (rdrand_enabled)
4074 				vmx->nested.msrs.secondary_ctls_high |=
4075 					SECONDARY_EXEC_RDRAND_EXITING;
4076 			else
4077 				vmx->nested.msrs.secondary_ctls_high &=
4078 					~SECONDARY_EXEC_RDRAND_EXITING;
4079 		}
4080 	}
4081 
4082 	if (vmx_rdseed_supported()) {
4083 		bool rdseed_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDSEED);
4084 		if (rdseed_enabled)
4085 			exec_control &= ~SECONDARY_EXEC_RDSEED_EXITING;
4086 
4087 		if (nested) {
4088 			if (rdseed_enabled)
4089 				vmx->nested.msrs.secondary_ctls_high |=
4090 					SECONDARY_EXEC_RDSEED_EXITING;
4091 			else
4092 				vmx->nested.msrs.secondary_ctls_high &=
4093 					~SECONDARY_EXEC_RDSEED_EXITING;
4094 		}
4095 	}
4096 
4097 	if (vmx_waitpkg_supported()) {
4098 		bool waitpkg_enabled =
4099 			guest_cpuid_has(vcpu, X86_FEATURE_WAITPKG);
4100 
4101 		if (!waitpkg_enabled)
4102 			exec_control &= ~SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE;
4103 
4104 		if (nested) {
4105 			if (waitpkg_enabled)
4106 				vmx->nested.msrs.secondary_ctls_high |=
4107 					SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE;
4108 			else
4109 				vmx->nested.msrs.secondary_ctls_high &=
4110 					~SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE;
4111 		}
4112 	}
4113 
4114 	vmx->secondary_exec_control = exec_control;
4115 }
4116 
4117 static void ept_set_mmio_spte_mask(void)
4118 {
4119 	/*
4120 	 * EPT Misconfigurations can be generated if the value of bits 2:0
4121 	 * of an EPT paging-structure entry is 110b (write/execute).
4122 	 */
4123 	kvm_mmu_set_mmio_spte_mask(VMX_EPT_RWX_MASK,
4124 				   VMX_EPT_MISCONFIG_WX_VALUE, 0);
4125 }
4126 
4127 #define VMX_XSS_EXIT_BITMAP 0
4128 
4129 /*
4130  * Noting that the initialization of Guest-state Area of VMCS is in
4131  * vmx_vcpu_reset().
4132  */
4133 static void init_vmcs(struct vcpu_vmx *vmx)
4134 {
4135 	if (nested)
4136 		nested_vmx_set_vmcs_shadowing_bitmap();
4137 
4138 	if (cpu_has_vmx_msr_bitmap())
4139 		vmcs_write64(MSR_BITMAP, __pa(vmx->vmcs01.msr_bitmap));
4140 
4141 	vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
4142 
4143 	/* Control */
4144 	pin_controls_set(vmx, vmx_pin_based_exec_ctrl(vmx));
4145 
4146 	exec_controls_set(vmx, vmx_exec_control(vmx));
4147 
4148 	if (cpu_has_secondary_exec_ctrls()) {
4149 		vmx_compute_secondary_exec_control(vmx);
4150 		secondary_exec_controls_set(vmx, vmx->secondary_exec_control);
4151 	}
4152 
4153 	if (kvm_vcpu_apicv_active(&vmx->vcpu)) {
4154 		vmcs_write64(EOI_EXIT_BITMAP0, 0);
4155 		vmcs_write64(EOI_EXIT_BITMAP1, 0);
4156 		vmcs_write64(EOI_EXIT_BITMAP2, 0);
4157 		vmcs_write64(EOI_EXIT_BITMAP3, 0);
4158 
4159 		vmcs_write16(GUEST_INTR_STATUS, 0);
4160 
4161 		vmcs_write16(POSTED_INTR_NV, POSTED_INTR_VECTOR);
4162 		vmcs_write64(POSTED_INTR_DESC_ADDR, __pa((&vmx->pi_desc)));
4163 	}
4164 
4165 	if (!kvm_pause_in_guest(vmx->vcpu.kvm)) {
4166 		vmcs_write32(PLE_GAP, ple_gap);
4167 		vmx->ple_window = ple_window;
4168 		vmx->ple_window_dirty = true;
4169 	}
4170 
4171 	vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
4172 	vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
4173 	vmcs_write32(CR3_TARGET_COUNT, 0);           /* 22.2.1 */
4174 
4175 	vmcs_write16(HOST_FS_SELECTOR, 0);            /* 22.2.4 */
4176 	vmcs_write16(HOST_GS_SELECTOR, 0);            /* 22.2.4 */
4177 	vmx_set_constant_host_state(vmx);
4178 	vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
4179 	vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
4180 
4181 	if (cpu_has_vmx_vmfunc())
4182 		vmcs_write64(VM_FUNCTION_CONTROL, 0);
4183 
4184 	vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
4185 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
4186 	vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host.val));
4187 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
4188 	vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest.val));
4189 
4190 	if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
4191 		vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
4192 
4193 	vm_exit_controls_set(vmx, vmx_vmexit_ctrl());
4194 
4195 	/* 22.2.1, 20.8.1 */
4196 	vm_entry_controls_set(vmx, vmx_vmentry_ctrl());
4197 
4198 	vmx->vcpu.arch.cr0_guest_owned_bits = X86_CR0_TS;
4199 	vmcs_writel(CR0_GUEST_HOST_MASK, ~X86_CR0_TS);
4200 
4201 	set_cr4_guest_host_mask(vmx);
4202 
4203 	if (vmx->vpid != 0)
4204 		vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
4205 
4206 	if (vmx_xsaves_supported())
4207 		vmcs_write64(XSS_EXIT_BITMAP, VMX_XSS_EXIT_BITMAP);
4208 
4209 	if (enable_pml) {
4210 		vmcs_write64(PML_ADDRESS, page_to_phys(vmx->pml_pg));
4211 		vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
4212 	}
4213 
4214 	if (cpu_has_vmx_encls_vmexit())
4215 		vmcs_write64(ENCLS_EXITING_BITMAP, -1ull);
4216 
4217 	if (pt_mode == PT_MODE_HOST_GUEST) {
4218 		memset(&vmx->pt_desc, 0, sizeof(vmx->pt_desc));
4219 		/* Bit[6~0] are forced to 1, writes are ignored. */
4220 		vmx->pt_desc.guest.output_mask = 0x7F;
4221 		vmcs_write64(GUEST_IA32_RTIT_CTL, 0);
4222 	}
4223 }
4224 
4225 static void vmx_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
4226 {
4227 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4228 	struct msr_data apic_base_msr;
4229 	u64 cr0;
4230 
4231 	vmx->rmode.vm86_active = 0;
4232 	vmx->spec_ctrl = 0;
4233 
4234 	vmx->msr_ia32_umwait_control = 0;
4235 
4236 	vcpu->arch.microcode_version = 0x100000000ULL;
4237 	vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
4238 	vmx->hv_deadline_tsc = -1;
4239 	kvm_set_cr8(vcpu, 0);
4240 
4241 	if (!init_event) {
4242 		apic_base_msr.data = APIC_DEFAULT_PHYS_BASE |
4243 				     MSR_IA32_APICBASE_ENABLE;
4244 		if (kvm_vcpu_is_reset_bsp(vcpu))
4245 			apic_base_msr.data |= MSR_IA32_APICBASE_BSP;
4246 		apic_base_msr.host_initiated = true;
4247 		kvm_set_apic_base(vcpu, &apic_base_msr);
4248 	}
4249 
4250 	vmx_segment_cache_clear(vmx);
4251 
4252 	seg_setup(VCPU_SREG_CS);
4253 	vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
4254 	vmcs_writel(GUEST_CS_BASE, 0xffff0000ul);
4255 
4256 	seg_setup(VCPU_SREG_DS);
4257 	seg_setup(VCPU_SREG_ES);
4258 	seg_setup(VCPU_SREG_FS);
4259 	seg_setup(VCPU_SREG_GS);
4260 	seg_setup(VCPU_SREG_SS);
4261 
4262 	vmcs_write16(GUEST_TR_SELECTOR, 0);
4263 	vmcs_writel(GUEST_TR_BASE, 0);
4264 	vmcs_write32(GUEST_TR_LIMIT, 0xffff);
4265 	vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
4266 
4267 	vmcs_write16(GUEST_LDTR_SELECTOR, 0);
4268 	vmcs_writel(GUEST_LDTR_BASE, 0);
4269 	vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
4270 	vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
4271 
4272 	if (!init_event) {
4273 		vmcs_write32(GUEST_SYSENTER_CS, 0);
4274 		vmcs_writel(GUEST_SYSENTER_ESP, 0);
4275 		vmcs_writel(GUEST_SYSENTER_EIP, 0);
4276 		vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
4277 	}
4278 
4279 	kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
4280 	kvm_rip_write(vcpu, 0xfff0);
4281 
4282 	vmcs_writel(GUEST_GDTR_BASE, 0);
4283 	vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
4284 
4285 	vmcs_writel(GUEST_IDTR_BASE, 0);
4286 	vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
4287 
4288 	vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
4289 	vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
4290 	vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS, 0);
4291 	if (kvm_mpx_supported())
4292 		vmcs_write64(GUEST_BNDCFGS, 0);
4293 
4294 	setup_msrs(vmx);
4295 
4296 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);  /* 22.2.1 */
4297 
4298 	if (cpu_has_vmx_tpr_shadow() && !init_event) {
4299 		vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
4300 		if (cpu_need_tpr_shadow(vcpu))
4301 			vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
4302 				     __pa(vcpu->arch.apic->regs));
4303 		vmcs_write32(TPR_THRESHOLD, 0);
4304 	}
4305 
4306 	kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
4307 
4308 	cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
4309 	vmx->vcpu.arch.cr0 = cr0;
4310 	vmx_set_cr0(vcpu, cr0); /* enter rmode */
4311 	vmx_set_cr4(vcpu, 0);
4312 	vmx_set_efer(vcpu, 0);
4313 
4314 	update_exception_bitmap(vcpu);
4315 
4316 	vpid_sync_context(vmx->vpid);
4317 	if (init_event)
4318 		vmx_clear_hlt(vcpu);
4319 }
4320 
4321 static void enable_irq_window(struct kvm_vcpu *vcpu)
4322 {
4323 	exec_controls_setbit(to_vmx(vcpu), CPU_BASED_INTR_WINDOW_EXITING);
4324 }
4325 
4326 static void enable_nmi_window(struct kvm_vcpu *vcpu)
4327 {
4328 	if (!enable_vnmi ||
4329 	    vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_STI) {
4330 		enable_irq_window(vcpu);
4331 		return;
4332 	}
4333 
4334 	exec_controls_setbit(to_vmx(vcpu), CPU_BASED_NMI_WINDOW_EXITING);
4335 }
4336 
4337 static void vmx_inject_irq(struct kvm_vcpu *vcpu)
4338 {
4339 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4340 	uint32_t intr;
4341 	int irq = vcpu->arch.interrupt.nr;
4342 
4343 	trace_kvm_inj_virq(irq);
4344 
4345 	++vcpu->stat.irq_injections;
4346 	if (vmx->rmode.vm86_active) {
4347 		int inc_eip = 0;
4348 		if (vcpu->arch.interrupt.soft)
4349 			inc_eip = vcpu->arch.event_exit_inst_len;
4350 		kvm_inject_realmode_interrupt(vcpu, irq, inc_eip);
4351 		return;
4352 	}
4353 	intr = irq | INTR_INFO_VALID_MASK;
4354 	if (vcpu->arch.interrupt.soft) {
4355 		intr |= INTR_TYPE_SOFT_INTR;
4356 		vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
4357 			     vmx->vcpu.arch.event_exit_inst_len);
4358 	} else
4359 		intr |= INTR_TYPE_EXT_INTR;
4360 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
4361 
4362 	vmx_clear_hlt(vcpu);
4363 }
4364 
4365 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
4366 {
4367 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4368 
4369 	if (!enable_vnmi) {
4370 		/*
4371 		 * Tracking the NMI-blocked state in software is built upon
4372 		 * finding the next open IRQ window. This, in turn, depends on
4373 		 * well-behaving guests: They have to keep IRQs disabled at
4374 		 * least as long as the NMI handler runs. Otherwise we may
4375 		 * cause NMI nesting, maybe breaking the guest. But as this is
4376 		 * highly unlikely, we can live with the residual risk.
4377 		 */
4378 		vmx->loaded_vmcs->soft_vnmi_blocked = 1;
4379 		vmx->loaded_vmcs->vnmi_blocked_time = 0;
4380 	}
4381 
4382 	++vcpu->stat.nmi_injections;
4383 	vmx->loaded_vmcs->nmi_known_unmasked = false;
4384 
4385 	if (vmx->rmode.vm86_active) {
4386 		kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR, 0);
4387 		return;
4388 	}
4389 
4390 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
4391 			INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
4392 
4393 	vmx_clear_hlt(vcpu);
4394 }
4395 
4396 bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
4397 {
4398 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4399 	bool masked;
4400 
4401 	if (!enable_vnmi)
4402 		return vmx->loaded_vmcs->soft_vnmi_blocked;
4403 	if (vmx->loaded_vmcs->nmi_known_unmasked)
4404 		return false;
4405 	masked = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI;
4406 	vmx->loaded_vmcs->nmi_known_unmasked = !masked;
4407 	return masked;
4408 }
4409 
4410 void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
4411 {
4412 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4413 
4414 	if (!enable_vnmi) {
4415 		if (vmx->loaded_vmcs->soft_vnmi_blocked != masked) {
4416 			vmx->loaded_vmcs->soft_vnmi_blocked = masked;
4417 			vmx->loaded_vmcs->vnmi_blocked_time = 0;
4418 		}
4419 	} else {
4420 		vmx->loaded_vmcs->nmi_known_unmasked = !masked;
4421 		if (masked)
4422 			vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
4423 				      GUEST_INTR_STATE_NMI);
4424 		else
4425 			vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
4426 					GUEST_INTR_STATE_NMI);
4427 	}
4428 }
4429 
4430 static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
4431 {
4432 	if (to_vmx(vcpu)->nested.nested_run_pending)
4433 		return 0;
4434 
4435 	if (!enable_vnmi &&
4436 	    to_vmx(vcpu)->loaded_vmcs->soft_vnmi_blocked)
4437 		return 0;
4438 
4439 	return	!(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4440 		  (GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_STI
4441 		   | GUEST_INTR_STATE_NMI));
4442 }
4443 
4444 static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
4445 {
4446 	return (!to_vmx(vcpu)->nested.nested_run_pending &&
4447 		vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
4448 		!(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4449 			(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
4450 }
4451 
4452 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
4453 {
4454 	int ret;
4455 
4456 	if (enable_unrestricted_guest)
4457 		return 0;
4458 
4459 	mutex_lock(&kvm->slots_lock);
4460 	ret = __x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, addr,
4461 				      PAGE_SIZE * 3);
4462 	mutex_unlock(&kvm->slots_lock);
4463 
4464 	if (ret)
4465 		return ret;
4466 	to_kvm_vmx(kvm)->tss_addr = addr;
4467 	return init_rmode_tss(kvm);
4468 }
4469 
4470 static int vmx_set_identity_map_addr(struct kvm *kvm, u64 ident_addr)
4471 {
4472 	to_kvm_vmx(kvm)->ept_identity_map_addr = ident_addr;
4473 	return 0;
4474 }
4475 
4476 static bool rmode_exception(struct kvm_vcpu *vcpu, int vec)
4477 {
4478 	switch (vec) {
4479 	case BP_VECTOR:
4480 		/*
4481 		 * Update instruction length as we may reinject the exception
4482 		 * from user space while in guest debugging mode.
4483 		 */
4484 		to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
4485 			vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4486 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
4487 			return false;
4488 		/* fall through */
4489 	case DB_VECTOR:
4490 		if (vcpu->guest_debug &
4491 			(KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
4492 			return false;
4493 		/* fall through */
4494 	case DE_VECTOR:
4495 	case OF_VECTOR:
4496 	case BR_VECTOR:
4497 	case UD_VECTOR:
4498 	case DF_VECTOR:
4499 	case SS_VECTOR:
4500 	case GP_VECTOR:
4501 	case MF_VECTOR:
4502 		return true;
4503 	break;
4504 	}
4505 	return false;
4506 }
4507 
4508 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
4509 				  int vec, u32 err_code)
4510 {
4511 	/*
4512 	 * Instruction with address size override prefix opcode 0x67
4513 	 * Cause the #SS fault with 0 error code in VM86 mode.
4514 	 */
4515 	if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0) {
4516 		if (kvm_emulate_instruction(vcpu, 0)) {
4517 			if (vcpu->arch.halt_request) {
4518 				vcpu->arch.halt_request = 0;
4519 				return kvm_vcpu_halt(vcpu);
4520 			}
4521 			return 1;
4522 		}
4523 		return 0;
4524 	}
4525 
4526 	/*
4527 	 * Forward all other exceptions that are valid in real mode.
4528 	 * FIXME: Breaks guest debugging in real mode, needs to be fixed with
4529 	 *        the required debugging infrastructure rework.
4530 	 */
4531 	kvm_queue_exception(vcpu, vec);
4532 	return 1;
4533 }
4534 
4535 /*
4536  * Trigger machine check on the host. We assume all the MSRs are already set up
4537  * by the CPU and that we still run on the same CPU as the MCE occurred on.
4538  * We pass a fake environment to the machine check handler because we want
4539  * the guest to be always treated like user space, no matter what context
4540  * it used internally.
4541  */
4542 static void kvm_machine_check(void)
4543 {
4544 #if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
4545 	struct pt_regs regs = {
4546 		.cs = 3, /* Fake ring 3 no matter what the guest ran on */
4547 		.flags = X86_EFLAGS_IF,
4548 	};
4549 
4550 	do_machine_check(&regs, 0);
4551 #endif
4552 }
4553 
4554 static int handle_machine_check(struct kvm_vcpu *vcpu)
4555 {
4556 	/* handled by vmx_vcpu_run() */
4557 	return 1;
4558 }
4559 
4560 static int handle_exception_nmi(struct kvm_vcpu *vcpu)
4561 {
4562 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4563 	struct kvm_run *kvm_run = vcpu->run;
4564 	u32 intr_info, ex_no, error_code;
4565 	unsigned long cr2, rip, dr6;
4566 	u32 vect_info;
4567 
4568 	vect_info = vmx->idt_vectoring_info;
4569 	intr_info = vmx->exit_intr_info;
4570 
4571 	if (is_machine_check(intr_info) || is_nmi(intr_info))
4572 		return 1; /* handled by handle_exception_nmi_irqoff() */
4573 
4574 	if (is_invalid_opcode(intr_info))
4575 		return handle_ud(vcpu);
4576 
4577 	error_code = 0;
4578 	if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
4579 		error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
4580 
4581 	if (!vmx->rmode.vm86_active && is_gp_fault(intr_info)) {
4582 		WARN_ON_ONCE(!enable_vmware_backdoor);
4583 
4584 		/*
4585 		 * VMware backdoor emulation on #GP interception only handles
4586 		 * IN{S}, OUT{S}, and RDPMC, none of which generate a non-zero
4587 		 * error code on #GP.
4588 		 */
4589 		if (error_code) {
4590 			kvm_queue_exception_e(vcpu, GP_VECTOR, error_code);
4591 			return 1;
4592 		}
4593 		return kvm_emulate_instruction(vcpu, EMULTYPE_VMWARE_GP);
4594 	}
4595 
4596 	/*
4597 	 * The #PF with PFEC.RSVD = 1 indicates the guest is accessing
4598 	 * MMIO, it is better to report an internal error.
4599 	 * See the comments in vmx_handle_exit.
4600 	 */
4601 	if ((vect_info & VECTORING_INFO_VALID_MASK) &&
4602 	    !(is_page_fault(intr_info) && !(error_code & PFERR_RSVD_MASK))) {
4603 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4604 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
4605 		vcpu->run->internal.ndata = 3;
4606 		vcpu->run->internal.data[0] = vect_info;
4607 		vcpu->run->internal.data[1] = intr_info;
4608 		vcpu->run->internal.data[2] = error_code;
4609 		return 0;
4610 	}
4611 
4612 	if (is_page_fault(intr_info)) {
4613 		cr2 = vmcs_readl(EXIT_QUALIFICATION);
4614 		/* EPT won't cause page fault directly */
4615 		WARN_ON_ONCE(!vcpu->arch.apf.host_apf_reason && enable_ept);
4616 		return kvm_handle_page_fault(vcpu, error_code, cr2, NULL, 0);
4617 	}
4618 
4619 	ex_no = intr_info & INTR_INFO_VECTOR_MASK;
4620 
4621 	if (vmx->rmode.vm86_active && rmode_exception(vcpu, ex_no))
4622 		return handle_rmode_exception(vcpu, ex_no, error_code);
4623 
4624 	switch (ex_no) {
4625 	case AC_VECTOR:
4626 		kvm_queue_exception_e(vcpu, AC_VECTOR, error_code);
4627 		return 1;
4628 	case DB_VECTOR:
4629 		dr6 = vmcs_readl(EXIT_QUALIFICATION);
4630 		if (!(vcpu->guest_debug &
4631 		      (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
4632 			vcpu->arch.dr6 &= ~DR_TRAP_BITS;
4633 			vcpu->arch.dr6 |= dr6 | DR6_RTM;
4634 			if (is_icebp(intr_info))
4635 				WARN_ON(!skip_emulated_instruction(vcpu));
4636 
4637 			kvm_queue_exception(vcpu, DB_VECTOR);
4638 			return 1;
4639 		}
4640 		kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
4641 		kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
4642 		/* fall through */
4643 	case BP_VECTOR:
4644 		/*
4645 		 * Update instruction length as we may reinject #BP from
4646 		 * user space while in guest debugging mode. Reading it for
4647 		 * #DB as well causes no harm, it is not used in that case.
4648 		 */
4649 		vmx->vcpu.arch.event_exit_inst_len =
4650 			vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4651 		kvm_run->exit_reason = KVM_EXIT_DEBUG;
4652 		rip = kvm_rip_read(vcpu);
4653 		kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
4654 		kvm_run->debug.arch.exception = ex_no;
4655 		break;
4656 	default:
4657 		kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
4658 		kvm_run->ex.exception = ex_no;
4659 		kvm_run->ex.error_code = error_code;
4660 		break;
4661 	}
4662 	return 0;
4663 }
4664 
4665 static __always_inline int handle_external_interrupt(struct kvm_vcpu *vcpu)
4666 {
4667 	++vcpu->stat.irq_exits;
4668 	return 1;
4669 }
4670 
4671 static int handle_triple_fault(struct kvm_vcpu *vcpu)
4672 {
4673 	vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
4674 	vcpu->mmio_needed = 0;
4675 	return 0;
4676 }
4677 
4678 static int handle_io(struct kvm_vcpu *vcpu)
4679 {
4680 	unsigned long exit_qualification;
4681 	int size, in, string;
4682 	unsigned port;
4683 
4684 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4685 	string = (exit_qualification & 16) != 0;
4686 
4687 	++vcpu->stat.io_exits;
4688 
4689 	if (string)
4690 		return kvm_emulate_instruction(vcpu, 0);
4691 
4692 	port = exit_qualification >> 16;
4693 	size = (exit_qualification & 7) + 1;
4694 	in = (exit_qualification & 8) != 0;
4695 
4696 	return kvm_fast_pio(vcpu, size, port, in);
4697 }
4698 
4699 static void
4700 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
4701 {
4702 	/*
4703 	 * Patch in the VMCALL instruction:
4704 	 */
4705 	hypercall[0] = 0x0f;
4706 	hypercall[1] = 0x01;
4707 	hypercall[2] = 0xc1;
4708 }
4709 
4710 /* called to set cr0 as appropriate for a mov-to-cr0 exit. */
4711 static int handle_set_cr0(struct kvm_vcpu *vcpu, unsigned long val)
4712 {
4713 	if (is_guest_mode(vcpu)) {
4714 		struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4715 		unsigned long orig_val = val;
4716 
4717 		/*
4718 		 * We get here when L2 changed cr0 in a way that did not change
4719 		 * any of L1's shadowed bits (see nested_vmx_exit_handled_cr),
4720 		 * but did change L0 shadowed bits. So we first calculate the
4721 		 * effective cr0 value that L1 would like to write into the
4722 		 * hardware. It consists of the L2-owned bits from the new
4723 		 * value combined with the L1-owned bits from L1's guest_cr0.
4724 		 */
4725 		val = (val & ~vmcs12->cr0_guest_host_mask) |
4726 			(vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask);
4727 
4728 		if (!nested_guest_cr0_valid(vcpu, val))
4729 			return 1;
4730 
4731 		if (kvm_set_cr0(vcpu, val))
4732 			return 1;
4733 		vmcs_writel(CR0_READ_SHADOW, orig_val);
4734 		return 0;
4735 	} else {
4736 		if (to_vmx(vcpu)->nested.vmxon &&
4737 		    !nested_host_cr0_valid(vcpu, val))
4738 			return 1;
4739 
4740 		return kvm_set_cr0(vcpu, val);
4741 	}
4742 }
4743 
4744 static int handle_set_cr4(struct kvm_vcpu *vcpu, unsigned long val)
4745 {
4746 	if (is_guest_mode(vcpu)) {
4747 		struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4748 		unsigned long orig_val = val;
4749 
4750 		/* analogously to handle_set_cr0 */
4751 		val = (val & ~vmcs12->cr4_guest_host_mask) |
4752 			(vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask);
4753 		if (kvm_set_cr4(vcpu, val))
4754 			return 1;
4755 		vmcs_writel(CR4_READ_SHADOW, orig_val);
4756 		return 0;
4757 	} else
4758 		return kvm_set_cr4(vcpu, val);
4759 }
4760 
4761 static int handle_desc(struct kvm_vcpu *vcpu)
4762 {
4763 	WARN_ON(!(vcpu->arch.cr4 & X86_CR4_UMIP));
4764 	return kvm_emulate_instruction(vcpu, 0);
4765 }
4766 
4767 static int handle_cr(struct kvm_vcpu *vcpu)
4768 {
4769 	unsigned long exit_qualification, val;
4770 	int cr;
4771 	int reg;
4772 	int err;
4773 	int ret;
4774 
4775 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4776 	cr = exit_qualification & 15;
4777 	reg = (exit_qualification >> 8) & 15;
4778 	switch ((exit_qualification >> 4) & 3) {
4779 	case 0: /* mov to cr */
4780 		val = kvm_register_readl(vcpu, reg);
4781 		trace_kvm_cr_write(cr, val);
4782 		switch (cr) {
4783 		case 0:
4784 			err = handle_set_cr0(vcpu, val);
4785 			return kvm_complete_insn_gp(vcpu, err);
4786 		case 3:
4787 			WARN_ON_ONCE(enable_unrestricted_guest);
4788 			err = kvm_set_cr3(vcpu, val);
4789 			return kvm_complete_insn_gp(vcpu, err);
4790 		case 4:
4791 			err = handle_set_cr4(vcpu, val);
4792 			return kvm_complete_insn_gp(vcpu, err);
4793 		case 8: {
4794 				u8 cr8_prev = kvm_get_cr8(vcpu);
4795 				u8 cr8 = (u8)val;
4796 				err = kvm_set_cr8(vcpu, cr8);
4797 				ret = kvm_complete_insn_gp(vcpu, err);
4798 				if (lapic_in_kernel(vcpu))
4799 					return ret;
4800 				if (cr8_prev <= cr8)
4801 					return ret;
4802 				/*
4803 				 * TODO: we might be squashing a
4804 				 * KVM_GUESTDBG_SINGLESTEP-triggered
4805 				 * KVM_EXIT_DEBUG here.
4806 				 */
4807 				vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
4808 				return 0;
4809 			}
4810 		}
4811 		break;
4812 	case 2: /* clts */
4813 		WARN_ONCE(1, "Guest should always own CR0.TS");
4814 		vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
4815 		trace_kvm_cr_write(0, kvm_read_cr0(vcpu));
4816 		return kvm_skip_emulated_instruction(vcpu);
4817 	case 1: /*mov from cr*/
4818 		switch (cr) {
4819 		case 3:
4820 			WARN_ON_ONCE(enable_unrestricted_guest);
4821 			val = kvm_read_cr3(vcpu);
4822 			kvm_register_write(vcpu, reg, val);
4823 			trace_kvm_cr_read(cr, val);
4824 			return kvm_skip_emulated_instruction(vcpu);
4825 		case 8:
4826 			val = kvm_get_cr8(vcpu);
4827 			kvm_register_write(vcpu, reg, val);
4828 			trace_kvm_cr_read(cr, val);
4829 			return kvm_skip_emulated_instruction(vcpu);
4830 		}
4831 		break;
4832 	case 3: /* lmsw */
4833 		val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
4834 		trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val);
4835 		kvm_lmsw(vcpu, val);
4836 
4837 		return kvm_skip_emulated_instruction(vcpu);
4838 	default:
4839 		break;
4840 	}
4841 	vcpu->run->exit_reason = 0;
4842 	vcpu_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
4843 	       (int)(exit_qualification >> 4) & 3, cr);
4844 	return 0;
4845 }
4846 
4847 static int handle_dr(struct kvm_vcpu *vcpu)
4848 {
4849 	unsigned long exit_qualification;
4850 	int dr, dr7, reg;
4851 
4852 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4853 	dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
4854 
4855 	/* First, if DR does not exist, trigger UD */
4856 	if (!kvm_require_dr(vcpu, dr))
4857 		return 1;
4858 
4859 	/* Do not handle if the CPL > 0, will trigger GP on re-entry */
4860 	if (!kvm_require_cpl(vcpu, 0))
4861 		return 1;
4862 	dr7 = vmcs_readl(GUEST_DR7);
4863 	if (dr7 & DR7_GD) {
4864 		/*
4865 		 * As the vm-exit takes precedence over the debug trap, we
4866 		 * need to emulate the latter, either for the host or the
4867 		 * guest debugging itself.
4868 		 */
4869 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
4870 			vcpu->run->debug.arch.dr6 = vcpu->arch.dr6;
4871 			vcpu->run->debug.arch.dr7 = dr7;
4872 			vcpu->run->debug.arch.pc = kvm_get_linear_rip(vcpu);
4873 			vcpu->run->debug.arch.exception = DB_VECTOR;
4874 			vcpu->run->exit_reason = KVM_EXIT_DEBUG;
4875 			return 0;
4876 		} else {
4877 			vcpu->arch.dr6 &= ~DR_TRAP_BITS;
4878 			vcpu->arch.dr6 |= DR6_BD | DR6_RTM;
4879 			kvm_queue_exception(vcpu, DB_VECTOR);
4880 			return 1;
4881 		}
4882 	}
4883 
4884 	if (vcpu->guest_debug == 0) {
4885 		exec_controls_clearbit(to_vmx(vcpu), CPU_BASED_MOV_DR_EXITING);
4886 
4887 		/*
4888 		 * No more DR vmexits; force a reload of the debug registers
4889 		 * and reenter on this instruction.  The next vmexit will
4890 		 * retrieve the full state of the debug registers.
4891 		 */
4892 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
4893 		return 1;
4894 	}
4895 
4896 	reg = DEBUG_REG_ACCESS_REG(exit_qualification);
4897 	if (exit_qualification & TYPE_MOV_FROM_DR) {
4898 		unsigned long val;
4899 
4900 		if (kvm_get_dr(vcpu, dr, &val))
4901 			return 1;
4902 		kvm_register_write(vcpu, reg, val);
4903 	} else
4904 		if (kvm_set_dr(vcpu, dr, kvm_register_readl(vcpu, reg)))
4905 			return 1;
4906 
4907 	return kvm_skip_emulated_instruction(vcpu);
4908 }
4909 
4910 static u64 vmx_get_dr6(struct kvm_vcpu *vcpu)
4911 {
4912 	return vcpu->arch.dr6;
4913 }
4914 
4915 static void vmx_set_dr6(struct kvm_vcpu *vcpu, unsigned long val)
4916 {
4917 }
4918 
4919 static void vmx_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
4920 {
4921 	get_debugreg(vcpu->arch.db[0], 0);
4922 	get_debugreg(vcpu->arch.db[1], 1);
4923 	get_debugreg(vcpu->arch.db[2], 2);
4924 	get_debugreg(vcpu->arch.db[3], 3);
4925 	get_debugreg(vcpu->arch.dr6, 6);
4926 	vcpu->arch.dr7 = vmcs_readl(GUEST_DR7);
4927 
4928 	vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
4929 	exec_controls_setbit(to_vmx(vcpu), CPU_BASED_MOV_DR_EXITING);
4930 }
4931 
4932 static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
4933 {
4934 	vmcs_writel(GUEST_DR7, val);
4935 }
4936 
4937 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
4938 {
4939 	kvm_apic_update_ppr(vcpu);
4940 	return 1;
4941 }
4942 
4943 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
4944 {
4945 	exec_controls_clearbit(to_vmx(vcpu), CPU_BASED_INTR_WINDOW_EXITING);
4946 
4947 	kvm_make_request(KVM_REQ_EVENT, vcpu);
4948 
4949 	++vcpu->stat.irq_window_exits;
4950 	return 1;
4951 }
4952 
4953 static int handle_vmcall(struct kvm_vcpu *vcpu)
4954 {
4955 	return kvm_emulate_hypercall(vcpu);
4956 }
4957 
4958 static int handle_invd(struct kvm_vcpu *vcpu)
4959 {
4960 	return kvm_emulate_instruction(vcpu, 0);
4961 }
4962 
4963 static int handle_invlpg(struct kvm_vcpu *vcpu)
4964 {
4965 	unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4966 
4967 	kvm_mmu_invlpg(vcpu, exit_qualification);
4968 	return kvm_skip_emulated_instruction(vcpu);
4969 }
4970 
4971 static int handle_rdpmc(struct kvm_vcpu *vcpu)
4972 {
4973 	int err;
4974 
4975 	err = kvm_rdpmc(vcpu);
4976 	return kvm_complete_insn_gp(vcpu, err);
4977 }
4978 
4979 static int handle_wbinvd(struct kvm_vcpu *vcpu)
4980 {
4981 	return kvm_emulate_wbinvd(vcpu);
4982 }
4983 
4984 static int handle_xsetbv(struct kvm_vcpu *vcpu)
4985 {
4986 	u64 new_bv = kvm_read_edx_eax(vcpu);
4987 	u32 index = kvm_rcx_read(vcpu);
4988 
4989 	if (kvm_set_xcr(vcpu, index, new_bv) == 0)
4990 		return kvm_skip_emulated_instruction(vcpu);
4991 	return 1;
4992 }
4993 
4994 static int handle_apic_access(struct kvm_vcpu *vcpu)
4995 {
4996 	if (likely(fasteoi)) {
4997 		unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4998 		int access_type, offset;
4999 
5000 		access_type = exit_qualification & APIC_ACCESS_TYPE;
5001 		offset = exit_qualification & APIC_ACCESS_OFFSET;
5002 		/*
5003 		 * Sane guest uses MOV to write EOI, with written value
5004 		 * not cared. So make a short-circuit here by avoiding
5005 		 * heavy instruction emulation.
5006 		 */
5007 		if ((access_type == TYPE_LINEAR_APIC_INST_WRITE) &&
5008 		    (offset == APIC_EOI)) {
5009 			kvm_lapic_set_eoi(vcpu);
5010 			return kvm_skip_emulated_instruction(vcpu);
5011 		}
5012 	}
5013 	return kvm_emulate_instruction(vcpu, 0);
5014 }
5015 
5016 static int handle_apic_eoi_induced(struct kvm_vcpu *vcpu)
5017 {
5018 	unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5019 	int vector = exit_qualification & 0xff;
5020 
5021 	/* EOI-induced VM exit is trap-like and thus no need to adjust IP */
5022 	kvm_apic_set_eoi_accelerated(vcpu, vector);
5023 	return 1;
5024 }
5025 
5026 static int handle_apic_write(struct kvm_vcpu *vcpu)
5027 {
5028 	unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5029 	u32 offset = exit_qualification & 0xfff;
5030 
5031 	/* APIC-write VM exit is trap-like and thus no need to adjust IP */
5032 	kvm_apic_write_nodecode(vcpu, offset);
5033 	return 1;
5034 }
5035 
5036 static int handle_task_switch(struct kvm_vcpu *vcpu)
5037 {
5038 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5039 	unsigned long exit_qualification;
5040 	bool has_error_code = false;
5041 	u32 error_code = 0;
5042 	u16 tss_selector;
5043 	int reason, type, idt_v, idt_index;
5044 
5045 	idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
5046 	idt_index = (vmx->idt_vectoring_info & VECTORING_INFO_VECTOR_MASK);
5047 	type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
5048 
5049 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5050 
5051 	reason = (u32)exit_qualification >> 30;
5052 	if (reason == TASK_SWITCH_GATE && idt_v) {
5053 		switch (type) {
5054 		case INTR_TYPE_NMI_INTR:
5055 			vcpu->arch.nmi_injected = false;
5056 			vmx_set_nmi_mask(vcpu, true);
5057 			break;
5058 		case INTR_TYPE_EXT_INTR:
5059 		case INTR_TYPE_SOFT_INTR:
5060 			kvm_clear_interrupt_queue(vcpu);
5061 			break;
5062 		case INTR_TYPE_HARD_EXCEPTION:
5063 			if (vmx->idt_vectoring_info &
5064 			    VECTORING_INFO_DELIVER_CODE_MASK) {
5065 				has_error_code = true;
5066 				error_code =
5067 					vmcs_read32(IDT_VECTORING_ERROR_CODE);
5068 			}
5069 			/* fall through */
5070 		case INTR_TYPE_SOFT_EXCEPTION:
5071 			kvm_clear_exception_queue(vcpu);
5072 			break;
5073 		default:
5074 			break;
5075 		}
5076 	}
5077 	tss_selector = exit_qualification;
5078 
5079 	if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
5080 		       type != INTR_TYPE_EXT_INTR &&
5081 		       type != INTR_TYPE_NMI_INTR))
5082 		WARN_ON(!skip_emulated_instruction(vcpu));
5083 
5084 	/*
5085 	 * TODO: What about debug traps on tss switch?
5086 	 *       Are we supposed to inject them and update dr6?
5087 	 */
5088 	return kvm_task_switch(vcpu, tss_selector,
5089 			       type == INTR_TYPE_SOFT_INTR ? idt_index : -1,
5090 			       reason, has_error_code, error_code);
5091 }
5092 
5093 static int handle_ept_violation(struct kvm_vcpu *vcpu)
5094 {
5095 	unsigned long exit_qualification;
5096 	gpa_t gpa;
5097 	u64 error_code;
5098 
5099 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5100 
5101 	/*
5102 	 * EPT violation happened while executing iret from NMI,
5103 	 * "blocked by NMI" bit has to be set before next VM entry.
5104 	 * There are errata that may cause this bit to not be set:
5105 	 * AAK134, BY25.
5106 	 */
5107 	if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
5108 			enable_vnmi &&
5109 			(exit_qualification & INTR_INFO_UNBLOCK_NMI))
5110 		vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO, GUEST_INTR_STATE_NMI);
5111 
5112 	gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5113 	trace_kvm_page_fault(gpa, exit_qualification);
5114 
5115 	/* Is it a read fault? */
5116 	error_code = (exit_qualification & EPT_VIOLATION_ACC_READ)
5117 		     ? PFERR_USER_MASK : 0;
5118 	/* Is it a write fault? */
5119 	error_code |= (exit_qualification & EPT_VIOLATION_ACC_WRITE)
5120 		      ? PFERR_WRITE_MASK : 0;
5121 	/* Is it a fetch fault? */
5122 	error_code |= (exit_qualification & EPT_VIOLATION_ACC_INSTR)
5123 		      ? PFERR_FETCH_MASK : 0;
5124 	/* ept page table entry is present? */
5125 	error_code |= (exit_qualification &
5126 		       (EPT_VIOLATION_READABLE | EPT_VIOLATION_WRITABLE |
5127 			EPT_VIOLATION_EXECUTABLE))
5128 		      ? PFERR_PRESENT_MASK : 0;
5129 
5130 	error_code |= (exit_qualification & 0x100) != 0 ?
5131 	       PFERR_GUEST_FINAL_MASK : PFERR_GUEST_PAGE_MASK;
5132 
5133 	vcpu->arch.exit_qualification = exit_qualification;
5134 	return kvm_mmu_page_fault(vcpu, gpa, error_code, NULL, 0);
5135 }
5136 
5137 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
5138 {
5139 	gpa_t gpa;
5140 
5141 	/*
5142 	 * A nested guest cannot optimize MMIO vmexits, because we have an
5143 	 * nGPA here instead of the required GPA.
5144 	 */
5145 	gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5146 	if (!is_guest_mode(vcpu) &&
5147 	    !kvm_io_bus_write(vcpu, KVM_FAST_MMIO_BUS, gpa, 0, NULL)) {
5148 		trace_kvm_fast_mmio(gpa);
5149 		return kvm_skip_emulated_instruction(vcpu);
5150 	}
5151 
5152 	return kvm_mmu_page_fault(vcpu, gpa, PFERR_RSVD_MASK, NULL, 0);
5153 }
5154 
5155 static int handle_nmi_window(struct kvm_vcpu *vcpu)
5156 {
5157 	WARN_ON_ONCE(!enable_vnmi);
5158 	exec_controls_clearbit(to_vmx(vcpu), CPU_BASED_NMI_WINDOW_EXITING);
5159 	++vcpu->stat.nmi_window_exits;
5160 	kvm_make_request(KVM_REQ_EVENT, vcpu);
5161 
5162 	return 1;
5163 }
5164 
5165 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
5166 {
5167 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5168 	bool intr_window_requested;
5169 	unsigned count = 130;
5170 
5171 	/*
5172 	 * We should never reach the point where we are emulating L2
5173 	 * due to invalid guest state as that means we incorrectly
5174 	 * allowed a nested VMEntry with an invalid vmcs12.
5175 	 */
5176 	WARN_ON_ONCE(vmx->emulation_required && vmx->nested.nested_run_pending);
5177 
5178 	intr_window_requested = exec_controls_get(vmx) &
5179 				CPU_BASED_INTR_WINDOW_EXITING;
5180 
5181 	while (vmx->emulation_required && count-- != 0) {
5182 		if (intr_window_requested && vmx_interrupt_allowed(vcpu))
5183 			return handle_interrupt_window(&vmx->vcpu);
5184 
5185 		if (kvm_test_request(KVM_REQ_EVENT, vcpu))
5186 			return 1;
5187 
5188 		if (!kvm_emulate_instruction(vcpu, 0))
5189 			return 0;
5190 
5191 		if (vmx->emulation_required && !vmx->rmode.vm86_active &&
5192 		    vcpu->arch.exception.pending) {
5193 			vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5194 			vcpu->run->internal.suberror =
5195 						KVM_INTERNAL_ERROR_EMULATION;
5196 			vcpu->run->internal.ndata = 0;
5197 			return 0;
5198 		}
5199 
5200 		if (vcpu->arch.halt_request) {
5201 			vcpu->arch.halt_request = 0;
5202 			return kvm_vcpu_halt(vcpu);
5203 		}
5204 
5205 		/*
5206 		 * Note, return 1 and not 0, vcpu_run() is responsible for
5207 		 * morphing the pending signal into the proper return code.
5208 		 */
5209 		if (signal_pending(current))
5210 			return 1;
5211 
5212 		if (need_resched())
5213 			schedule();
5214 	}
5215 
5216 	return 1;
5217 }
5218 
5219 static void grow_ple_window(struct kvm_vcpu *vcpu)
5220 {
5221 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5222 	unsigned int old = vmx->ple_window;
5223 
5224 	vmx->ple_window = __grow_ple_window(old, ple_window,
5225 					    ple_window_grow,
5226 					    ple_window_max);
5227 
5228 	if (vmx->ple_window != old) {
5229 		vmx->ple_window_dirty = true;
5230 		trace_kvm_ple_window_update(vcpu->vcpu_id,
5231 					    vmx->ple_window, old);
5232 	}
5233 }
5234 
5235 static void shrink_ple_window(struct kvm_vcpu *vcpu)
5236 {
5237 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5238 	unsigned int old = vmx->ple_window;
5239 
5240 	vmx->ple_window = __shrink_ple_window(old, ple_window,
5241 					      ple_window_shrink,
5242 					      ple_window);
5243 
5244 	if (vmx->ple_window != old) {
5245 		vmx->ple_window_dirty = true;
5246 		trace_kvm_ple_window_update(vcpu->vcpu_id,
5247 					    vmx->ple_window, old);
5248 	}
5249 }
5250 
5251 /*
5252  * Handler for POSTED_INTERRUPT_WAKEUP_VECTOR.
5253  */
5254 static void wakeup_handler(void)
5255 {
5256 	struct kvm_vcpu *vcpu;
5257 	int cpu = smp_processor_id();
5258 
5259 	spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
5260 	list_for_each_entry(vcpu, &per_cpu(blocked_vcpu_on_cpu, cpu),
5261 			blocked_vcpu_list) {
5262 		struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
5263 
5264 		if (pi_test_on(pi_desc) == 1)
5265 			kvm_vcpu_kick(vcpu);
5266 	}
5267 	spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
5268 }
5269 
5270 static void vmx_enable_tdp(void)
5271 {
5272 	kvm_mmu_set_mask_ptes(VMX_EPT_READABLE_MASK,
5273 		enable_ept_ad_bits ? VMX_EPT_ACCESS_BIT : 0ull,
5274 		enable_ept_ad_bits ? VMX_EPT_DIRTY_BIT : 0ull,
5275 		0ull, VMX_EPT_EXECUTABLE_MASK,
5276 		cpu_has_vmx_ept_execute_only() ? 0ull : VMX_EPT_READABLE_MASK,
5277 		VMX_EPT_RWX_MASK, 0ull);
5278 
5279 	ept_set_mmio_spte_mask();
5280 	kvm_enable_tdp();
5281 }
5282 
5283 /*
5284  * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
5285  * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
5286  */
5287 static int handle_pause(struct kvm_vcpu *vcpu)
5288 {
5289 	if (!kvm_pause_in_guest(vcpu->kvm))
5290 		grow_ple_window(vcpu);
5291 
5292 	/*
5293 	 * Intel sdm vol3 ch-25.1.3 says: The "PAUSE-loop exiting"
5294 	 * VM-execution control is ignored if CPL > 0. OTOH, KVM
5295 	 * never set PAUSE_EXITING and just set PLE if supported,
5296 	 * so the vcpu must be CPL=0 if it gets a PAUSE exit.
5297 	 */
5298 	kvm_vcpu_on_spin(vcpu, true);
5299 	return kvm_skip_emulated_instruction(vcpu);
5300 }
5301 
5302 static int handle_nop(struct kvm_vcpu *vcpu)
5303 {
5304 	return kvm_skip_emulated_instruction(vcpu);
5305 }
5306 
5307 static int handle_mwait(struct kvm_vcpu *vcpu)
5308 {
5309 	printk_once(KERN_WARNING "kvm: MWAIT instruction emulated as NOP!\n");
5310 	return handle_nop(vcpu);
5311 }
5312 
5313 static int handle_invalid_op(struct kvm_vcpu *vcpu)
5314 {
5315 	kvm_queue_exception(vcpu, UD_VECTOR);
5316 	return 1;
5317 }
5318 
5319 static int handle_monitor_trap(struct kvm_vcpu *vcpu)
5320 {
5321 	return 1;
5322 }
5323 
5324 static int handle_monitor(struct kvm_vcpu *vcpu)
5325 {
5326 	printk_once(KERN_WARNING "kvm: MONITOR instruction emulated as NOP!\n");
5327 	return handle_nop(vcpu);
5328 }
5329 
5330 static int handle_invpcid(struct kvm_vcpu *vcpu)
5331 {
5332 	u32 vmx_instruction_info;
5333 	unsigned long type;
5334 	bool pcid_enabled;
5335 	gva_t gva;
5336 	struct x86_exception e;
5337 	unsigned i;
5338 	unsigned long roots_to_free = 0;
5339 	struct {
5340 		u64 pcid;
5341 		u64 gla;
5342 	} operand;
5343 
5344 	if (!guest_cpuid_has(vcpu, X86_FEATURE_INVPCID)) {
5345 		kvm_queue_exception(vcpu, UD_VECTOR);
5346 		return 1;
5347 	}
5348 
5349 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5350 	type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
5351 
5352 	if (type > 3) {
5353 		kvm_inject_gp(vcpu, 0);
5354 		return 1;
5355 	}
5356 
5357 	/* According to the Intel instruction reference, the memory operand
5358 	 * is read even if it isn't needed (e.g., for type==all)
5359 	 */
5360 	if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
5361 				vmx_instruction_info, false,
5362 				sizeof(operand), &gva))
5363 		return 1;
5364 
5365 	if (kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e)) {
5366 		kvm_inject_page_fault(vcpu, &e);
5367 		return 1;
5368 	}
5369 
5370 	if (operand.pcid >> 12 != 0) {
5371 		kvm_inject_gp(vcpu, 0);
5372 		return 1;
5373 	}
5374 
5375 	pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
5376 
5377 	switch (type) {
5378 	case INVPCID_TYPE_INDIV_ADDR:
5379 		if ((!pcid_enabled && (operand.pcid != 0)) ||
5380 		    is_noncanonical_address(operand.gla, vcpu)) {
5381 			kvm_inject_gp(vcpu, 0);
5382 			return 1;
5383 		}
5384 		kvm_mmu_invpcid_gva(vcpu, operand.gla, operand.pcid);
5385 		return kvm_skip_emulated_instruction(vcpu);
5386 
5387 	case INVPCID_TYPE_SINGLE_CTXT:
5388 		if (!pcid_enabled && (operand.pcid != 0)) {
5389 			kvm_inject_gp(vcpu, 0);
5390 			return 1;
5391 		}
5392 
5393 		if (kvm_get_active_pcid(vcpu) == operand.pcid) {
5394 			kvm_mmu_sync_roots(vcpu);
5395 			kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
5396 		}
5397 
5398 		for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++)
5399 			if (kvm_get_pcid(vcpu, vcpu->arch.mmu->prev_roots[i].cr3)
5400 			    == operand.pcid)
5401 				roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i);
5402 
5403 		kvm_mmu_free_roots(vcpu, vcpu->arch.mmu, roots_to_free);
5404 		/*
5405 		 * If neither the current cr3 nor any of the prev_roots use the
5406 		 * given PCID, then nothing needs to be done here because a
5407 		 * resync will happen anyway before switching to any other CR3.
5408 		 */
5409 
5410 		return kvm_skip_emulated_instruction(vcpu);
5411 
5412 	case INVPCID_TYPE_ALL_NON_GLOBAL:
5413 		/*
5414 		 * Currently, KVM doesn't mark global entries in the shadow
5415 		 * page tables, so a non-global flush just degenerates to a
5416 		 * global flush. If needed, we could optimize this later by
5417 		 * keeping track of global entries in shadow page tables.
5418 		 */
5419 
5420 		/* fall-through */
5421 	case INVPCID_TYPE_ALL_INCL_GLOBAL:
5422 		kvm_mmu_unload(vcpu);
5423 		return kvm_skip_emulated_instruction(vcpu);
5424 
5425 	default:
5426 		BUG(); /* We have already checked above that type <= 3 */
5427 	}
5428 }
5429 
5430 static int handle_pml_full(struct kvm_vcpu *vcpu)
5431 {
5432 	unsigned long exit_qualification;
5433 
5434 	trace_kvm_pml_full(vcpu->vcpu_id);
5435 
5436 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5437 
5438 	/*
5439 	 * PML buffer FULL happened while executing iret from NMI,
5440 	 * "blocked by NMI" bit has to be set before next VM entry.
5441 	 */
5442 	if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
5443 			enable_vnmi &&
5444 			(exit_qualification & INTR_INFO_UNBLOCK_NMI))
5445 		vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
5446 				GUEST_INTR_STATE_NMI);
5447 
5448 	/*
5449 	 * PML buffer already flushed at beginning of VMEXIT. Nothing to do
5450 	 * here.., and there's no userspace involvement needed for PML.
5451 	 */
5452 	return 1;
5453 }
5454 
5455 static int handle_preemption_timer(struct kvm_vcpu *vcpu)
5456 {
5457 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5458 
5459 	if (!vmx->req_immediate_exit &&
5460 	    !unlikely(vmx->loaded_vmcs->hv_timer_soft_disabled))
5461 		kvm_lapic_expired_hv_timer(vcpu);
5462 
5463 	return 1;
5464 }
5465 
5466 /*
5467  * When nested=0, all VMX instruction VM Exits filter here.  The handlers
5468  * are overwritten by nested_vmx_setup() when nested=1.
5469  */
5470 static int handle_vmx_instruction(struct kvm_vcpu *vcpu)
5471 {
5472 	kvm_queue_exception(vcpu, UD_VECTOR);
5473 	return 1;
5474 }
5475 
5476 static int handle_encls(struct kvm_vcpu *vcpu)
5477 {
5478 	/*
5479 	 * SGX virtualization is not yet supported.  There is no software
5480 	 * enable bit for SGX, so we have to trap ENCLS and inject a #UD
5481 	 * to prevent the guest from executing ENCLS.
5482 	 */
5483 	kvm_queue_exception(vcpu, UD_VECTOR);
5484 	return 1;
5485 }
5486 
5487 /*
5488  * The exit handlers return 1 if the exit was handled fully and guest execution
5489  * may resume.  Otherwise they set the kvm_run parameter to indicate what needs
5490  * to be done to userspace and return 0.
5491  */
5492 static int (*kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
5493 	[EXIT_REASON_EXCEPTION_NMI]           = handle_exception_nmi,
5494 	[EXIT_REASON_EXTERNAL_INTERRUPT]      = handle_external_interrupt,
5495 	[EXIT_REASON_TRIPLE_FAULT]            = handle_triple_fault,
5496 	[EXIT_REASON_NMI_WINDOW]	      = handle_nmi_window,
5497 	[EXIT_REASON_IO_INSTRUCTION]          = handle_io,
5498 	[EXIT_REASON_CR_ACCESS]               = handle_cr,
5499 	[EXIT_REASON_DR_ACCESS]               = handle_dr,
5500 	[EXIT_REASON_CPUID]                   = kvm_emulate_cpuid,
5501 	[EXIT_REASON_MSR_READ]                = kvm_emulate_rdmsr,
5502 	[EXIT_REASON_MSR_WRITE]               = kvm_emulate_wrmsr,
5503 	[EXIT_REASON_INTERRUPT_WINDOW]        = handle_interrupt_window,
5504 	[EXIT_REASON_HLT]                     = kvm_emulate_halt,
5505 	[EXIT_REASON_INVD]		      = handle_invd,
5506 	[EXIT_REASON_INVLPG]		      = handle_invlpg,
5507 	[EXIT_REASON_RDPMC]                   = handle_rdpmc,
5508 	[EXIT_REASON_VMCALL]                  = handle_vmcall,
5509 	[EXIT_REASON_VMCLEAR]		      = handle_vmx_instruction,
5510 	[EXIT_REASON_VMLAUNCH]		      = handle_vmx_instruction,
5511 	[EXIT_REASON_VMPTRLD]		      = handle_vmx_instruction,
5512 	[EXIT_REASON_VMPTRST]		      = handle_vmx_instruction,
5513 	[EXIT_REASON_VMREAD]		      = handle_vmx_instruction,
5514 	[EXIT_REASON_VMRESUME]		      = handle_vmx_instruction,
5515 	[EXIT_REASON_VMWRITE]		      = handle_vmx_instruction,
5516 	[EXIT_REASON_VMOFF]		      = handle_vmx_instruction,
5517 	[EXIT_REASON_VMON]		      = handle_vmx_instruction,
5518 	[EXIT_REASON_TPR_BELOW_THRESHOLD]     = handle_tpr_below_threshold,
5519 	[EXIT_REASON_APIC_ACCESS]             = handle_apic_access,
5520 	[EXIT_REASON_APIC_WRITE]              = handle_apic_write,
5521 	[EXIT_REASON_EOI_INDUCED]             = handle_apic_eoi_induced,
5522 	[EXIT_REASON_WBINVD]                  = handle_wbinvd,
5523 	[EXIT_REASON_XSETBV]                  = handle_xsetbv,
5524 	[EXIT_REASON_TASK_SWITCH]             = handle_task_switch,
5525 	[EXIT_REASON_MCE_DURING_VMENTRY]      = handle_machine_check,
5526 	[EXIT_REASON_GDTR_IDTR]		      = handle_desc,
5527 	[EXIT_REASON_LDTR_TR]		      = handle_desc,
5528 	[EXIT_REASON_EPT_VIOLATION]	      = handle_ept_violation,
5529 	[EXIT_REASON_EPT_MISCONFIG]           = handle_ept_misconfig,
5530 	[EXIT_REASON_PAUSE_INSTRUCTION]       = handle_pause,
5531 	[EXIT_REASON_MWAIT_INSTRUCTION]	      = handle_mwait,
5532 	[EXIT_REASON_MONITOR_TRAP_FLAG]       = handle_monitor_trap,
5533 	[EXIT_REASON_MONITOR_INSTRUCTION]     = handle_monitor,
5534 	[EXIT_REASON_INVEPT]                  = handle_vmx_instruction,
5535 	[EXIT_REASON_INVVPID]                 = handle_vmx_instruction,
5536 	[EXIT_REASON_RDRAND]                  = handle_invalid_op,
5537 	[EXIT_REASON_RDSEED]                  = handle_invalid_op,
5538 	[EXIT_REASON_PML_FULL]		      = handle_pml_full,
5539 	[EXIT_REASON_INVPCID]                 = handle_invpcid,
5540 	[EXIT_REASON_VMFUNC]		      = handle_vmx_instruction,
5541 	[EXIT_REASON_PREEMPTION_TIMER]	      = handle_preemption_timer,
5542 	[EXIT_REASON_ENCLS]		      = handle_encls,
5543 };
5544 
5545 static const int kvm_vmx_max_exit_handlers =
5546 	ARRAY_SIZE(kvm_vmx_exit_handlers);
5547 
5548 static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
5549 {
5550 	*info1 = vmcs_readl(EXIT_QUALIFICATION);
5551 	*info2 = vmcs_read32(VM_EXIT_INTR_INFO);
5552 }
5553 
5554 static void vmx_destroy_pml_buffer(struct vcpu_vmx *vmx)
5555 {
5556 	if (vmx->pml_pg) {
5557 		__free_page(vmx->pml_pg);
5558 		vmx->pml_pg = NULL;
5559 	}
5560 }
5561 
5562 static void vmx_flush_pml_buffer(struct kvm_vcpu *vcpu)
5563 {
5564 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5565 	u64 *pml_buf;
5566 	u16 pml_idx;
5567 
5568 	pml_idx = vmcs_read16(GUEST_PML_INDEX);
5569 
5570 	/* Do nothing if PML buffer is empty */
5571 	if (pml_idx == (PML_ENTITY_NUM - 1))
5572 		return;
5573 
5574 	/* PML index always points to next available PML buffer entity */
5575 	if (pml_idx >= PML_ENTITY_NUM)
5576 		pml_idx = 0;
5577 	else
5578 		pml_idx++;
5579 
5580 	pml_buf = page_address(vmx->pml_pg);
5581 	for (; pml_idx < PML_ENTITY_NUM; pml_idx++) {
5582 		u64 gpa;
5583 
5584 		gpa = pml_buf[pml_idx];
5585 		WARN_ON(gpa & (PAGE_SIZE - 1));
5586 		kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
5587 	}
5588 
5589 	/* reset PML index */
5590 	vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
5591 }
5592 
5593 /*
5594  * Flush all vcpus' PML buffer and update logged GPAs to dirty_bitmap.
5595  * Called before reporting dirty_bitmap to userspace.
5596  */
5597 static void kvm_flush_pml_buffers(struct kvm *kvm)
5598 {
5599 	int i;
5600 	struct kvm_vcpu *vcpu;
5601 	/*
5602 	 * We only need to kick vcpu out of guest mode here, as PML buffer
5603 	 * is flushed at beginning of all VMEXITs, and it's obvious that only
5604 	 * vcpus running in guest are possible to have unflushed GPAs in PML
5605 	 * buffer.
5606 	 */
5607 	kvm_for_each_vcpu(i, vcpu, kvm)
5608 		kvm_vcpu_kick(vcpu);
5609 }
5610 
5611 static void vmx_dump_sel(char *name, uint32_t sel)
5612 {
5613 	pr_err("%s sel=0x%04x, attr=0x%05x, limit=0x%08x, base=0x%016lx\n",
5614 	       name, vmcs_read16(sel),
5615 	       vmcs_read32(sel + GUEST_ES_AR_BYTES - GUEST_ES_SELECTOR),
5616 	       vmcs_read32(sel + GUEST_ES_LIMIT - GUEST_ES_SELECTOR),
5617 	       vmcs_readl(sel + GUEST_ES_BASE - GUEST_ES_SELECTOR));
5618 }
5619 
5620 static void vmx_dump_dtsel(char *name, uint32_t limit)
5621 {
5622 	pr_err("%s                           limit=0x%08x, base=0x%016lx\n",
5623 	       name, vmcs_read32(limit),
5624 	       vmcs_readl(limit + GUEST_GDTR_BASE - GUEST_GDTR_LIMIT));
5625 }
5626 
5627 void dump_vmcs(void)
5628 {
5629 	u32 vmentry_ctl, vmexit_ctl;
5630 	u32 cpu_based_exec_ctrl, pin_based_exec_ctrl, secondary_exec_control;
5631 	unsigned long cr4;
5632 	u64 efer;
5633 	int i, n;
5634 
5635 	if (!dump_invalid_vmcs) {
5636 		pr_warn_ratelimited("set kvm_intel.dump_invalid_vmcs=1 to dump internal KVM state.\n");
5637 		return;
5638 	}
5639 
5640 	vmentry_ctl = vmcs_read32(VM_ENTRY_CONTROLS);
5641 	vmexit_ctl = vmcs_read32(VM_EXIT_CONTROLS);
5642 	cpu_based_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5643 	pin_based_exec_ctrl = vmcs_read32(PIN_BASED_VM_EXEC_CONTROL);
5644 	cr4 = vmcs_readl(GUEST_CR4);
5645 	efer = vmcs_read64(GUEST_IA32_EFER);
5646 	secondary_exec_control = 0;
5647 	if (cpu_has_secondary_exec_ctrls())
5648 		secondary_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
5649 
5650 	pr_err("*** Guest State ***\n");
5651 	pr_err("CR0: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
5652 	       vmcs_readl(GUEST_CR0), vmcs_readl(CR0_READ_SHADOW),
5653 	       vmcs_readl(CR0_GUEST_HOST_MASK));
5654 	pr_err("CR4: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
5655 	       cr4, vmcs_readl(CR4_READ_SHADOW), vmcs_readl(CR4_GUEST_HOST_MASK));
5656 	pr_err("CR3 = 0x%016lx\n", vmcs_readl(GUEST_CR3));
5657 	if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT) &&
5658 	    (cr4 & X86_CR4_PAE) && !(efer & EFER_LMA))
5659 	{
5660 		pr_err("PDPTR0 = 0x%016llx  PDPTR1 = 0x%016llx\n",
5661 		       vmcs_read64(GUEST_PDPTR0), vmcs_read64(GUEST_PDPTR1));
5662 		pr_err("PDPTR2 = 0x%016llx  PDPTR3 = 0x%016llx\n",
5663 		       vmcs_read64(GUEST_PDPTR2), vmcs_read64(GUEST_PDPTR3));
5664 	}
5665 	pr_err("RSP = 0x%016lx  RIP = 0x%016lx\n",
5666 	       vmcs_readl(GUEST_RSP), vmcs_readl(GUEST_RIP));
5667 	pr_err("RFLAGS=0x%08lx         DR7 = 0x%016lx\n",
5668 	       vmcs_readl(GUEST_RFLAGS), vmcs_readl(GUEST_DR7));
5669 	pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
5670 	       vmcs_readl(GUEST_SYSENTER_ESP),
5671 	       vmcs_read32(GUEST_SYSENTER_CS), vmcs_readl(GUEST_SYSENTER_EIP));
5672 	vmx_dump_sel("CS:  ", GUEST_CS_SELECTOR);
5673 	vmx_dump_sel("DS:  ", GUEST_DS_SELECTOR);
5674 	vmx_dump_sel("SS:  ", GUEST_SS_SELECTOR);
5675 	vmx_dump_sel("ES:  ", GUEST_ES_SELECTOR);
5676 	vmx_dump_sel("FS:  ", GUEST_FS_SELECTOR);
5677 	vmx_dump_sel("GS:  ", GUEST_GS_SELECTOR);
5678 	vmx_dump_dtsel("GDTR:", GUEST_GDTR_LIMIT);
5679 	vmx_dump_sel("LDTR:", GUEST_LDTR_SELECTOR);
5680 	vmx_dump_dtsel("IDTR:", GUEST_IDTR_LIMIT);
5681 	vmx_dump_sel("TR:  ", GUEST_TR_SELECTOR);
5682 	if ((vmexit_ctl & (VM_EXIT_SAVE_IA32_PAT | VM_EXIT_SAVE_IA32_EFER)) ||
5683 	    (vmentry_ctl & (VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_IA32_EFER)))
5684 		pr_err("EFER =     0x%016llx  PAT = 0x%016llx\n",
5685 		       efer, vmcs_read64(GUEST_IA32_PAT));
5686 	pr_err("DebugCtl = 0x%016llx  DebugExceptions = 0x%016lx\n",
5687 	       vmcs_read64(GUEST_IA32_DEBUGCTL),
5688 	       vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS));
5689 	if (cpu_has_load_perf_global_ctrl() &&
5690 	    vmentry_ctl & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
5691 		pr_err("PerfGlobCtl = 0x%016llx\n",
5692 		       vmcs_read64(GUEST_IA32_PERF_GLOBAL_CTRL));
5693 	if (vmentry_ctl & VM_ENTRY_LOAD_BNDCFGS)
5694 		pr_err("BndCfgS = 0x%016llx\n", vmcs_read64(GUEST_BNDCFGS));
5695 	pr_err("Interruptibility = %08x  ActivityState = %08x\n",
5696 	       vmcs_read32(GUEST_INTERRUPTIBILITY_INFO),
5697 	       vmcs_read32(GUEST_ACTIVITY_STATE));
5698 	if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)
5699 		pr_err("InterruptStatus = %04x\n",
5700 		       vmcs_read16(GUEST_INTR_STATUS));
5701 
5702 	pr_err("*** Host State ***\n");
5703 	pr_err("RIP = 0x%016lx  RSP = 0x%016lx\n",
5704 	       vmcs_readl(HOST_RIP), vmcs_readl(HOST_RSP));
5705 	pr_err("CS=%04x SS=%04x DS=%04x ES=%04x FS=%04x GS=%04x TR=%04x\n",
5706 	       vmcs_read16(HOST_CS_SELECTOR), vmcs_read16(HOST_SS_SELECTOR),
5707 	       vmcs_read16(HOST_DS_SELECTOR), vmcs_read16(HOST_ES_SELECTOR),
5708 	       vmcs_read16(HOST_FS_SELECTOR), vmcs_read16(HOST_GS_SELECTOR),
5709 	       vmcs_read16(HOST_TR_SELECTOR));
5710 	pr_err("FSBase=%016lx GSBase=%016lx TRBase=%016lx\n",
5711 	       vmcs_readl(HOST_FS_BASE), vmcs_readl(HOST_GS_BASE),
5712 	       vmcs_readl(HOST_TR_BASE));
5713 	pr_err("GDTBase=%016lx IDTBase=%016lx\n",
5714 	       vmcs_readl(HOST_GDTR_BASE), vmcs_readl(HOST_IDTR_BASE));
5715 	pr_err("CR0=%016lx CR3=%016lx CR4=%016lx\n",
5716 	       vmcs_readl(HOST_CR0), vmcs_readl(HOST_CR3),
5717 	       vmcs_readl(HOST_CR4));
5718 	pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
5719 	       vmcs_readl(HOST_IA32_SYSENTER_ESP),
5720 	       vmcs_read32(HOST_IA32_SYSENTER_CS),
5721 	       vmcs_readl(HOST_IA32_SYSENTER_EIP));
5722 	if (vmexit_ctl & (VM_EXIT_LOAD_IA32_PAT | VM_EXIT_LOAD_IA32_EFER))
5723 		pr_err("EFER = 0x%016llx  PAT = 0x%016llx\n",
5724 		       vmcs_read64(HOST_IA32_EFER),
5725 		       vmcs_read64(HOST_IA32_PAT));
5726 	if (cpu_has_load_perf_global_ctrl() &&
5727 	    vmexit_ctl & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
5728 		pr_err("PerfGlobCtl = 0x%016llx\n",
5729 		       vmcs_read64(HOST_IA32_PERF_GLOBAL_CTRL));
5730 
5731 	pr_err("*** Control State ***\n");
5732 	pr_err("PinBased=%08x CPUBased=%08x SecondaryExec=%08x\n",
5733 	       pin_based_exec_ctrl, cpu_based_exec_ctrl, secondary_exec_control);
5734 	pr_err("EntryControls=%08x ExitControls=%08x\n", vmentry_ctl, vmexit_ctl);
5735 	pr_err("ExceptionBitmap=%08x PFECmask=%08x PFECmatch=%08x\n",
5736 	       vmcs_read32(EXCEPTION_BITMAP),
5737 	       vmcs_read32(PAGE_FAULT_ERROR_CODE_MASK),
5738 	       vmcs_read32(PAGE_FAULT_ERROR_CODE_MATCH));
5739 	pr_err("VMEntry: intr_info=%08x errcode=%08x ilen=%08x\n",
5740 	       vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
5741 	       vmcs_read32(VM_ENTRY_EXCEPTION_ERROR_CODE),
5742 	       vmcs_read32(VM_ENTRY_INSTRUCTION_LEN));
5743 	pr_err("VMExit: intr_info=%08x errcode=%08x ilen=%08x\n",
5744 	       vmcs_read32(VM_EXIT_INTR_INFO),
5745 	       vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
5746 	       vmcs_read32(VM_EXIT_INSTRUCTION_LEN));
5747 	pr_err("        reason=%08x qualification=%016lx\n",
5748 	       vmcs_read32(VM_EXIT_REASON), vmcs_readl(EXIT_QUALIFICATION));
5749 	pr_err("IDTVectoring: info=%08x errcode=%08x\n",
5750 	       vmcs_read32(IDT_VECTORING_INFO_FIELD),
5751 	       vmcs_read32(IDT_VECTORING_ERROR_CODE));
5752 	pr_err("TSC Offset = 0x%016llx\n", vmcs_read64(TSC_OFFSET));
5753 	if (secondary_exec_control & SECONDARY_EXEC_TSC_SCALING)
5754 		pr_err("TSC Multiplier = 0x%016llx\n",
5755 		       vmcs_read64(TSC_MULTIPLIER));
5756 	if (cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW) {
5757 		if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY) {
5758 			u16 status = vmcs_read16(GUEST_INTR_STATUS);
5759 			pr_err("SVI|RVI = %02x|%02x ", status >> 8, status & 0xff);
5760 		}
5761 		pr_cont("TPR Threshold = 0x%02x\n", vmcs_read32(TPR_THRESHOLD));
5762 		if (secondary_exec_control & SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)
5763 			pr_err("APIC-access addr = 0x%016llx ", vmcs_read64(APIC_ACCESS_ADDR));
5764 		pr_cont("virt-APIC addr = 0x%016llx\n", vmcs_read64(VIRTUAL_APIC_PAGE_ADDR));
5765 	}
5766 	if (pin_based_exec_ctrl & PIN_BASED_POSTED_INTR)
5767 		pr_err("PostedIntrVec = 0x%02x\n", vmcs_read16(POSTED_INTR_NV));
5768 	if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT))
5769 		pr_err("EPT pointer = 0x%016llx\n", vmcs_read64(EPT_POINTER));
5770 	n = vmcs_read32(CR3_TARGET_COUNT);
5771 	for (i = 0; i + 1 < n; i += 4)
5772 		pr_err("CR3 target%u=%016lx target%u=%016lx\n",
5773 		       i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2),
5774 		       i + 1, vmcs_readl(CR3_TARGET_VALUE0 + i * 2 + 2));
5775 	if (i < n)
5776 		pr_err("CR3 target%u=%016lx\n",
5777 		       i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2));
5778 	if (secondary_exec_control & SECONDARY_EXEC_PAUSE_LOOP_EXITING)
5779 		pr_err("PLE Gap=%08x Window=%08x\n",
5780 		       vmcs_read32(PLE_GAP), vmcs_read32(PLE_WINDOW));
5781 	if (secondary_exec_control & SECONDARY_EXEC_ENABLE_VPID)
5782 		pr_err("Virtual processor ID = 0x%04x\n",
5783 		       vmcs_read16(VIRTUAL_PROCESSOR_ID));
5784 }
5785 
5786 /*
5787  * The guest has exited.  See if we can fix it or if we need userspace
5788  * assistance.
5789  */
5790 static int vmx_handle_exit(struct kvm_vcpu *vcpu,
5791 	enum exit_fastpath_completion exit_fastpath)
5792 {
5793 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5794 	u32 exit_reason = vmx->exit_reason;
5795 	u32 vectoring_info = vmx->idt_vectoring_info;
5796 
5797 	trace_kvm_exit(exit_reason, vcpu, KVM_ISA_VMX);
5798 
5799 	/*
5800 	 * Flush logged GPAs PML buffer, this will make dirty_bitmap more
5801 	 * updated. Another good is, in kvm_vm_ioctl_get_dirty_log, before
5802 	 * querying dirty_bitmap, we only need to kick all vcpus out of guest
5803 	 * mode as if vcpus is in root mode, the PML buffer must has been
5804 	 * flushed already.
5805 	 */
5806 	if (enable_pml)
5807 		vmx_flush_pml_buffer(vcpu);
5808 
5809 	/* If guest state is invalid, start emulating */
5810 	if (vmx->emulation_required)
5811 		return handle_invalid_guest_state(vcpu);
5812 
5813 	if (is_guest_mode(vcpu) && nested_vmx_exit_reflected(vcpu, exit_reason))
5814 		return nested_vmx_reflect_vmexit(vcpu, exit_reason);
5815 
5816 	if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) {
5817 		dump_vmcs();
5818 		vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
5819 		vcpu->run->fail_entry.hardware_entry_failure_reason
5820 			= exit_reason;
5821 		return 0;
5822 	}
5823 
5824 	if (unlikely(vmx->fail)) {
5825 		dump_vmcs();
5826 		vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
5827 		vcpu->run->fail_entry.hardware_entry_failure_reason
5828 			= vmcs_read32(VM_INSTRUCTION_ERROR);
5829 		return 0;
5830 	}
5831 
5832 	/*
5833 	 * Note:
5834 	 * Do not try to fix EXIT_REASON_EPT_MISCONFIG if it caused by
5835 	 * delivery event since it indicates guest is accessing MMIO.
5836 	 * The vm-exit can be triggered again after return to guest that
5837 	 * will cause infinite loop.
5838 	 */
5839 	if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
5840 			(exit_reason != EXIT_REASON_EXCEPTION_NMI &&
5841 			exit_reason != EXIT_REASON_EPT_VIOLATION &&
5842 			exit_reason != EXIT_REASON_PML_FULL &&
5843 			exit_reason != EXIT_REASON_TASK_SWITCH)) {
5844 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5845 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_DELIVERY_EV;
5846 		vcpu->run->internal.ndata = 3;
5847 		vcpu->run->internal.data[0] = vectoring_info;
5848 		vcpu->run->internal.data[1] = exit_reason;
5849 		vcpu->run->internal.data[2] = vcpu->arch.exit_qualification;
5850 		if (exit_reason == EXIT_REASON_EPT_MISCONFIG) {
5851 			vcpu->run->internal.ndata++;
5852 			vcpu->run->internal.data[3] =
5853 				vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5854 		}
5855 		return 0;
5856 	}
5857 
5858 	if (unlikely(!enable_vnmi &&
5859 		     vmx->loaded_vmcs->soft_vnmi_blocked)) {
5860 		if (vmx_interrupt_allowed(vcpu)) {
5861 			vmx->loaded_vmcs->soft_vnmi_blocked = 0;
5862 		} else if (vmx->loaded_vmcs->vnmi_blocked_time > 1000000000LL &&
5863 			   vcpu->arch.nmi_pending) {
5864 			/*
5865 			 * This CPU don't support us in finding the end of an
5866 			 * NMI-blocked window if the guest runs with IRQs
5867 			 * disabled. So we pull the trigger after 1 s of
5868 			 * futile waiting, but inform the user about this.
5869 			 */
5870 			printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
5871 			       "state on VCPU %d after 1 s timeout\n",
5872 			       __func__, vcpu->vcpu_id);
5873 			vmx->loaded_vmcs->soft_vnmi_blocked = 0;
5874 		}
5875 	}
5876 
5877 	if (exit_fastpath == EXIT_FASTPATH_SKIP_EMUL_INS) {
5878 		kvm_skip_emulated_instruction(vcpu);
5879 		return 1;
5880 	}
5881 
5882 	if (exit_reason >= kvm_vmx_max_exit_handlers)
5883 		goto unexpected_vmexit;
5884 #ifdef CONFIG_RETPOLINE
5885 	if (exit_reason == EXIT_REASON_MSR_WRITE)
5886 		return kvm_emulate_wrmsr(vcpu);
5887 	else if (exit_reason == EXIT_REASON_PREEMPTION_TIMER)
5888 		return handle_preemption_timer(vcpu);
5889 	else if (exit_reason == EXIT_REASON_INTERRUPT_WINDOW)
5890 		return handle_interrupt_window(vcpu);
5891 	else if (exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT)
5892 		return handle_external_interrupt(vcpu);
5893 	else if (exit_reason == EXIT_REASON_HLT)
5894 		return kvm_emulate_halt(vcpu);
5895 	else if (exit_reason == EXIT_REASON_EPT_MISCONFIG)
5896 		return handle_ept_misconfig(vcpu);
5897 #endif
5898 
5899 	exit_reason = array_index_nospec(exit_reason,
5900 					 kvm_vmx_max_exit_handlers);
5901 	if (!kvm_vmx_exit_handlers[exit_reason])
5902 		goto unexpected_vmexit;
5903 
5904 	return kvm_vmx_exit_handlers[exit_reason](vcpu);
5905 
5906 unexpected_vmexit:
5907 	vcpu_unimpl(vcpu, "vmx: unexpected exit reason 0x%x\n", exit_reason);
5908 	dump_vmcs();
5909 	vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5910 	vcpu->run->internal.suberror =
5911 			KVM_INTERNAL_ERROR_UNEXPECTED_EXIT_REASON;
5912 	vcpu->run->internal.ndata = 1;
5913 	vcpu->run->internal.data[0] = exit_reason;
5914 	return 0;
5915 }
5916 
5917 /*
5918  * Software based L1D cache flush which is used when microcode providing
5919  * the cache control MSR is not loaded.
5920  *
5921  * The L1D cache is 32 KiB on Nehalem and later microarchitectures, but to
5922  * flush it is required to read in 64 KiB because the replacement algorithm
5923  * is not exactly LRU. This could be sized at runtime via topology
5924  * information but as all relevant affected CPUs have 32KiB L1D cache size
5925  * there is no point in doing so.
5926  */
5927 static void vmx_l1d_flush(struct kvm_vcpu *vcpu)
5928 {
5929 	int size = PAGE_SIZE << L1D_CACHE_ORDER;
5930 
5931 	/*
5932 	 * This code is only executed when the the flush mode is 'cond' or
5933 	 * 'always'
5934 	 */
5935 	if (static_branch_likely(&vmx_l1d_flush_cond)) {
5936 		bool flush_l1d;
5937 
5938 		/*
5939 		 * Clear the per-vcpu flush bit, it gets set again
5940 		 * either from vcpu_run() or from one of the unsafe
5941 		 * VMEXIT handlers.
5942 		 */
5943 		flush_l1d = vcpu->arch.l1tf_flush_l1d;
5944 		vcpu->arch.l1tf_flush_l1d = false;
5945 
5946 		/*
5947 		 * Clear the per-cpu flush bit, it gets set again from
5948 		 * the interrupt handlers.
5949 		 */
5950 		flush_l1d |= kvm_get_cpu_l1tf_flush_l1d();
5951 		kvm_clear_cpu_l1tf_flush_l1d();
5952 
5953 		if (!flush_l1d)
5954 			return;
5955 	}
5956 
5957 	vcpu->stat.l1d_flush++;
5958 
5959 	if (static_cpu_has(X86_FEATURE_FLUSH_L1D)) {
5960 		wrmsrl(MSR_IA32_FLUSH_CMD, L1D_FLUSH);
5961 		return;
5962 	}
5963 
5964 	asm volatile(
5965 		/* First ensure the pages are in the TLB */
5966 		"xorl	%%eax, %%eax\n"
5967 		".Lpopulate_tlb:\n\t"
5968 		"movzbl	(%[flush_pages], %%" _ASM_AX "), %%ecx\n\t"
5969 		"addl	$4096, %%eax\n\t"
5970 		"cmpl	%%eax, %[size]\n\t"
5971 		"jne	.Lpopulate_tlb\n\t"
5972 		"xorl	%%eax, %%eax\n\t"
5973 		"cpuid\n\t"
5974 		/* Now fill the cache */
5975 		"xorl	%%eax, %%eax\n"
5976 		".Lfill_cache:\n"
5977 		"movzbl	(%[flush_pages], %%" _ASM_AX "), %%ecx\n\t"
5978 		"addl	$64, %%eax\n\t"
5979 		"cmpl	%%eax, %[size]\n\t"
5980 		"jne	.Lfill_cache\n\t"
5981 		"lfence\n"
5982 		:: [flush_pages] "r" (vmx_l1d_flush_pages),
5983 		    [size] "r" (size)
5984 		: "eax", "ebx", "ecx", "edx");
5985 }
5986 
5987 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
5988 {
5989 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5990 	int tpr_threshold;
5991 
5992 	if (is_guest_mode(vcpu) &&
5993 		nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
5994 		return;
5995 
5996 	tpr_threshold = (irr == -1 || tpr < irr) ? 0 : irr;
5997 	if (is_guest_mode(vcpu))
5998 		to_vmx(vcpu)->nested.l1_tpr_threshold = tpr_threshold;
5999 	else
6000 		vmcs_write32(TPR_THRESHOLD, tpr_threshold);
6001 }
6002 
6003 void vmx_set_virtual_apic_mode(struct kvm_vcpu *vcpu)
6004 {
6005 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6006 	u32 sec_exec_control;
6007 
6008 	if (!lapic_in_kernel(vcpu))
6009 		return;
6010 
6011 	if (!flexpriority_enabled &&
6012 	    !cpu_has_vmx_virtualize_x2apic_mode())
6013 		return;
6014 
6015 	/* Postpone execution until vmcs01 is the current VMCS. */
6016 	if (is_guest_mode(vcpu)) {
6017 		vmx->nested.change_vmcs01_virtual_apic_mode = true;
6018 		return;
6019 	}
6020 
6021 	sec_exec_control = secondary_exec_controls_get(vmx);
6022 	sec_exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
6023 			      SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE);
6024 
6025 	switch (kvm_get_apic_mode(vcpu)) {
6026 	case LAPIC_MODE_INVALID:
6027 		WARN_ONCE(true, "Invalid local APIC state");
6028 	case LAPIC_MODE_DISABLED:
6029 		break;
6030 	case LAPIC_MODE_XAPIC:
6031 		if (flexpriority_enabled) {
6032 			sec_exec_control |=
6033 				SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
6034 			vmx_flush_tlb(vcpu, true);
6035 		}
6036 		break;
6037 	case LAPIC_MODE_X2APIC:
6038 		if (cpu_has_vmx_virtualize_x2apic_mode())
6039 			sec_exec_control |=
6040 				SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
6041 		break;
6042 	}
6043 	secondary_exec_controls_set(vmx, sec_exec_control);
6044 
6045 	vmx_update_msr_bitmap(vcpu);
6046 }
6047 
6048 static void vmx_set_apic_access_page_addr(struct kvm_vcpu *vcpu, hpa_t hpa)
6049 {
6050 	if (!is_guest_mode(vcpu)) {
6051 		vmcs_write64(APIC_ACCESS_ADDR, hpa);
6052 		vmx_flush_tlb(vcpu, true);
6053 	}
6054 }
6055 
6056 static void vmx_hwapic_isr_update(struct kvm_vcpu *vcpu, int max_isr)
6057 {
6058 	u16 status;
6059 	u8 old;
6060 
6061 	if (max_isr == -1)
6062 		max_isr = 0;
6063 
6064 	status = vmcs_read16(GUEST_INTR_STATUS);
6065 	old = status >> 8;
6066 	if (max_isr != old) {
6067 		status &= 0xff;
6068 		status |= max_isr << 8;
6069 		vmcs_write16(GUEST_INTR_STATUS, status);
6070 	}
6071 }
6072 
6073 static void vmx_set_rvi(int vector)
6074 {
6075 	u16 status;
6076 	u8 old;
6077 
6078 	if (vector == -1)
6079 		vector = 0;
6080 
6081 	status = vmcs_read16(GUEST_INTR_STATUS);
6082 	old = (u8)status & 0xff;
6083 	if ((u8)vector != old) {
6084 		status &= ~0xff;
6085 		status |= (u8)vector;
6086 		vmcs_write16(GUEST_INTR_STATUS, status);
6087 	}
6088 }
6089 
6090 static void vmx_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr)
6091 {
6092 	/*
6093 	 * When running L2, updating RVI is only relevant when
6094 	 * vmcs12 virtual-interrupt-delivery enabled.
6095 	 * However, it can be enabled only when L1 also
6096 	 * intercepts external-interrupts and in that case
6097 	 * we should not update vmcs02 RVI but instead intercept
6098 	 * interrupt. Therefore, do nothing when running L2.
6099 	 */
6100 	if (!is_guest_mode(vcpu))
6101 		vmx_set_rvi(max_irr);
6102 }
6103 
6104 static int vmx_sync_pir_to_irr(struct kvm_vcpu *vcpu)
6105 {
6106 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6107 	int max_irr;
6108 	bool max_irr_updated;
6109 
6110 	WARN_ON(!vcpu->arch.apicv_active);
6111 	if (pi_test_on(&vmx->pi_desc)) {
6112 		pi_clear_on(&vmx->pi_desc);
6113 		/*
6114 		 * IOMMU can write to PID.ON, so the barrier matters even on UP.
6115 		 * But on x86 this is just a compiler barrier anyway.
6116 		 */
6117 		smp_mb__after_atomic();
6118 		max_irr_updated =
6119 			kvm_apic_update_irr(vcpu, vmx->pi_desc.pir, &max_irr);
6120 
6121 		/*
6122 		 * If we are running L2 and L1 has a new pending interrupt
6123 		 * which can be injected, we should re-evaluate
6124 		 * what should be done with this new L1 interrupt.
6125 		 * If L1 intercepts external-interrupts, we should
6126 		 * exit from L2 to L1. Otherwise, interrupt should be
6127 		 * delivered directly to L2.
6128 		 */
6129 		if (is_guest_mode(vcpu) && max_irr_updated) {
6130 			if (nested_exit_on_intr(vcpu))
6131 				kvm_vcpu_exiting_guest_mode(vcpu);
6132 			else
6133 				kvm_make_request(KVM_REQ_EVENT, vcpu);
6134 		}
6135 	} else {
6136 		max_irr = kvm_lapic_find_highest_irr(vcpu);
6137 	}
6138 	vmx_hwapic_irr_update(vcpu, max_irr);
6139 	return max_irr;
6140 }
6141 
6142 static bool vmx_dy_apicv_has_pending_interrupt(struct kvm_vcpu *vcpu)
6143 {
6144 	struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
6145 
6146 	return pi_test_on(pi_desc) ||
6147 		(pi_test_sn(pi_desc) && !pi_is_pir_empty(pi_desc));
6148 }
6149 
6150 static void vmx_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap)
6151 {
6152 	if (!kvm_vcpu_apicv_active(vcpu))
6153 		return;
6154 
6155 	vmcs_write64(EOI_EXIT_BITMAP0, eoi_exit_bitmap[0]);
6156 	vmcs_write64(EOI_EXIT_BITMAP1, eoi_exit_bitmap[1]);
6157 	vmcs_write64(EOI_EXIT_BITMAP2, eoi_exit_bitmap[2]);
6158 	vmcs_write64(EOI_EXIT_BITMAP3, eoi_exit_bitmap[3]);
6159 }
6160 
6161 static void vmx_apicv_post_state_restore(struct kvm_vcpu *vcpu)
6162 {
6163 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6164 
6165 	pi_clear_on(&vmx->pi_desc);
6166 	memset(vmx->pi_desc.pir, 0, sizeof(vmx->pi_desc.pir));
6167 }
6168 
6169 static void handle_exception_nmi_irqoff(struct vcpu_vmx *vmx)
6170 {
6171 	vmx->exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6172 
6173 	/* if exit due to PF check for async PF */
6174 	if (is_page_fault(vmx->exit_intr_info))
6175 		vmx->vcpu.arch.apf.host_apf_reason = kvm_read_and_reset_pf_reason();
6176 
6177 	/* Handle machine checks before interrupts are enabled */
6178 	if (is_machine_check(vmx->exit_intr_info))
6179 		kvm_machine_check();
6180 
6181 	/* We need to handle NMIs before interrupts are enabled */
6182 	if (is_nmi(vmx->exit_intr_info)) {
6183 		kvm_before_interrupt(&vmx->vcpu);
6184 		asm("int $2");
6185 		kvm_after_interrupt(&vmx->vcpu);
6186 	}
6187 }
6188 
6189 static void handle_external_interrupt_irqoff(struct kvm_vcpu *vcpu)
6190 {
6191 	unsigned int vector;
6192 	unsigned long entry;
6193 #ifdef CONFIG_X86_64
6194 	unsigned long tmp;
6195 #endif
6196 	gate_desc *desc;
6197 	u32 intr_info;
6198 
6199 	intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6200 	if (WARN_ONCE(!is_external_intr(intr_info),
6201 	    "KVM: unexpected VM-Exit interrupt info: 0x%x", intr_info))
6202 		return;
6203 
6204 	vector = intr_info & INTR_INFO_VECTOR_MASK;
6205 	desc = (gate_desc *)host_idt_base + vector;
6206 	entry = gate_offset(desc);
6207 
6208 	kvm_before_interrupt(vcpu);
6209 
6210 	asm volatile(
6211 #ifdef CONFIG_X86_64
6212 		"mov %%" _ASM_SP ", %[sp]\n\t"
6213 		"and $0xfffffffffffffff0, %%" _ASM_SP "\n\t"
6214 		"push $%c[ss]\n\t"
6215 		"push %[sp]\n\t"
6216 #endif
6217 		"pushf\n\t"
6218 		__ASM_SIZE(push) " $%c[cs]\n\t"
6219 		CALL_NOSPEC
6220 		:
6221 #ifdef CONFIG_X86_64
6222 		[sp]"=&r"(tmp),
6223 #endif
6224 		ASM_CALL_CONSTRAINT
6225 		:
6226 		THUNK_TARGET(entry),
6227 		[ss]"i"(__KERNEL_DS),
6228 		[cs]"i"(__KERNEL_CS)
6229 	);
6230 
6231 	kvm_after_interrupt(vcpu);
6232 }
6233 STACK_FRAME_NON_STANDARD(handle_external_interrupt_irqoff);
6234 
6235 static void vmx_handle_exit_irqoff(struct kvm_vcpu *vcpu,
6236 	enum exit_fastpath_completion *exit_fastpath)
6237 {
6238 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6239 
6240 	if (vmx->exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT)
6241 		handle_external_interrupt_irqoff(vcpu);
6242 	else if (vmx->exit_reason == EXIT_REASON_EXCEPTION_NMI)
6243 		handle_exception_nmi_irqoff(vmx);
6244 	else if (!is_guest_mode(vcpu) &&
6245 		vmx->exit_reason == EXIT_REASON_MSR_WRITE)
6246 		*exit_fastpath = handle_fastpath_set_msr_irqoff(vcpu);
6247 }
6248 
6249 static bool vmx_has_emulated_msr(int index)
6250 {
6251 	switch (index) {
6252 	case MSR_IA32_SMBASE:
6253 		/*
6254 		 * We cannot do SMM unless we can run the guest in big
6255 		 * real mode.
6256 		 */
6257 		return enable_unrestricted_guest || emulate_invalid_guest_state;
6258 	case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
6259 		return nested;
6260 	case MSR_AMD64_VIRT_SPEC_CTRL:
6261 		/* This is AMD only.  */
6262 		return false;
6263 	default:
6264 		return true;
6265 	}
6266 }
6267 
6268 static bool vmx_pt_supported(void)
6269 {
6270 	return pt_mode == PT_MODE_HOST_GUEST;
6271 }
6272 
6273 static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx)
6274 {
6275 	u32 exit_intr_info;
6276 	bool unblock_nmi;
6277 	u8 vector;
6278 	bool idtv_info_valid;
6279 
6280 	idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK;
6281 
6282 	if (enable_vnmi) {
6283 		if (vmx->loaded_vmcs->nmi_known_unmasked)
6284 			return;
6285 		/*
6286 		 * Can't use vmx->exit_intr_info since we're not sure what
6287 		 * the exit reason is.
6288 		 */
6289 		exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6290 		unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
6291 		vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
6292 		/*
6293 		 * SDM 3: 27.7.1.2 (September 2008)
6294 		 * Re-set bit "block by NMI" before VM entry if vmexit caused by
6295 		 * a guest IRET fault.
6296 		 * SDM 3: 23.2.2 (September 2008)
6297 		 * Bit 12 is undefined in any of the following cases:
6298 		 *  If the VM exit sets the valid bit in the IDT-vectoring
6299 		 *   information field.
6300 		 *  If the VM exit is due to a double fault.
6301 		 */
6302 		if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
6303 		    vector != DF_VECTOR && !idtv_info_valid)
6304 			vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
6305 				      GUEST_INTR_STATE_NMI);
6306 		else
6307 			vmx->loaded_vmcs->nmi_known_unmasked =
6308 				!(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO)
6309 				  & GUEST_INTR_STATE_NMI);
6310 	} else if (unlikely(vmx->loaded_vmcs->soft_vnmi_blocked))
6311 		vmx->loaded_vmcs->vnmi_blocked_time +=
6312 			ktime_to_ns(ktime_sub(ktime_get(),
6313 					      vmx->loaded_vmcs->entry_time));
6314 }
6315 
6316 static void __vmx_complete_interrupts(struct kvm_vcpu *vcpu,
6317 				      u32 idt_vectoring_info,
6318 				      int instr_len_field,
6319 				      int error_code_field)
6320 {
6321 	u8 vector;
6322 	int type;
6323 	bool idtv_info_valid;
6324 
6325 	idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
6326 
6327 	vcpu->arch.nmi_injected = false;
6328 	kvm_clear_exception_queue(vcpu);
6329 	kvm_clear_interrupt_queue(vcpu);
6330 
6331 	if (!idtv_info_valid)
6332 		return;
6333 
6334 	kvm_make_request(KVM_REQ_EVENT, vcpu);
6335 
6336 	vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
6337 	type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
6338 
6339 	switch (type) {
6340 	case INTR_TYPE_NMI_INTR:
6341 		vcpu->arch.nmi_injected = true;
6342 		/*
6343 		 * SDM 3: 27.7.1.2 (September 2008)
6344 		 * Clear bit "block by NMI" before VM entry if a NMI
6345 		 * delivery faulted.
6346 		 */
6347 		vmx_set_nmi_mask(vcpu, false);
6348 		break;
6349 	case INTR_TYPE_SOFT_EXCEPTION:
6350 		vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
6351 		/* fall through */
6352 	case INTR_TYPE_HARD_EXCEPTION:
6353 		if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
6354 			u32 err = vmcs_read32(error_code_field);
6355 			kvm_requeue_exception_e(vcpu, vector, err);
6356 		} else
6357 			kvm_requeue_exception(vcpu, vector);
6358 		break;
6359 	case INTR_TYPE_SOFT_INTR:
6360 		vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
6361 		/* fall through */
6362 	case INTR_TYPE_EXT_INTR:
6363 		kvm_queue_interrupt(vcpu, vector, type == INTR_TYPE_SOFT_INTR);
6364 		break;
6365 	default:
6366 		break;
6367 	}
6368 }
6369 
6370 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
6371 {
6372 	__vmx_complete_interrupts(&vmx->vcpu, vmx->idt_vectoring_info,
6373 				  VM_EXIT_INSTRUCTION_LEN,
6374 				  IDT_VECTORING_ERROR_CODE);
6375 }
6376 
6377 static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
6378 {
6379 	__vmx_complete_interrupts(vcpu,
6380 				  vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
6381 				  VM_ENTRY_INSTRUCTION_LEN,
6382 				  VM_ENTRY_EXCEPTION_ERROR_CODE);
6383 
6384 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
6385 }
6386 
6387 static void atomic_switch_perf_msrs(struct vcpu_vmx *vmx)
6388 {
6389 	int i, nr_msrs;
6390 	struct perf_guest_switch_msr *msrs;
6391 
6392 	msrs = perf_guest_get_msrs(&nr_msrs);
6393 
6394 	if (!msrs)
6395 		return;
6396 
6397 	for (i = 0; i < nr_msrs; i++)
6398 		if (msrs[i].host == msrs[i].guest)
6399 			clear_atomic_switch_msr(vmx, msrs[i].msr);
6400 		else
6401 			add_atomic_switch_msr(vmx, msrs[i].msr, msrs[i].guest,
6402 					msrs[i].host, false);
6403 }
6404 
6405 static void atomic_switch_umwait_control_msr(struct vcpu_vmx *vmx)
6406 {
6407 	u32 host_umwait_control;
6408 
6409 	if (!vmx_has_waitpkg(vmx))
6410 		return;
6411 
6412 	host_umwait_control = get_umwait_control_msr();
6413 
6414 	if (vmx->msr_ia32_umwait_control != host_umwait_control)
6415 		add_atomic_switch_msr(vmx, MSR_IA32_UMWAIT_CONTROL,
6416 			vmx->msr_ia32_umwait_control,
6417 			host_umwait_control, false);
6418 	else
6419 		clear_atomic_switch_msr(vmx, MSR_IA32_UMWAIT_CONTROL);
6420 }
6421 
6422 static void vmx_update_hv_timer(struct kvm_vcpu *vcpu)
6423 {
6424 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6425 	u64 tscl;
6426 	u32 delta_tsc;
6427 
6428 	if (vmx->req_immediate_exit) {
6429 		vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, 0);
6430 		vmx->loaded_vmcs->hv_timer_soft_disabled = false;
6431 	} else if (vmx->hv_deadline_tsc != -1) {
6432 		tscl = rdtsc();
6433 		if (vmx->hv_deadline_tsc > tscl)
6434 			/* set_hv_timer ensures the delta fits in 32-bits */
6435 			delta_tsc = (u32)((vmx->hv_deadline_tsc - tscl) >>
6436 				cpu_preemption_timer_multi);
6437 		else
6438 			delta_tsc = 0;
6439 
6440 		vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, delta_tsc);
6441 		vmx->loaded_vmcs->hv_timer_soft_disabled = false;
6442 	} else if (!vmx->loaded_vmcs->hv_timer_soft_disabled) {
6443 		vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, -1);
6444 		vmx->loaded_vmcs->hv_timer_soft_disabled = true;
6445 	}
6446 }
6447 
6448 void vmx_update_host_rsp(struct vcpu_vmx *vmx, unsigned long host_rsp)
6449 {
6450 	if (unlikely(host_rsp != vmx->loaded_vmcs->host_state.rsp)) {
6451 		vmx->loaded_vmcs->host_state.rsp = host_rsp;
6452 		vmcs_writel(HOST_RSP, host_rsp);
6453 	}
6454 }
6455 
6456 bool __vmx_vcpu_run(struct vcpu_vmx *vmx, unsigned long *regs, bool launched);
6457 
6458 static void vmx_vcpu_run(struct kvm_vcpu *vcpu)
6459 {
6460 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6461 	unsigned long cr3, cr4;
6462 
6463 	/* Record the guest's net vcpu time for enforced NMI injections. */
6464 	if (unlikely(!enable_vnmi &&
6465 		     vmx->loaded_vmcs->soft_vnmi_blocked))
6466 		vmx->loaded_vmcs->entry_time = ktime_get();
6467 
6468 	/* Don't enter VMX if guest state is invalid, let the exit handler
6469 	   start emulation until we arrive back to a valid state */
6470 	if (vmx->emulation_required)
6471 		return;
6472 
6473 	if (vmx->ple_window_dirty) {
6474 		vmx->ple_window_dirty = false;
6475 		vmcs_write32(PLE_WINDOW, vmx->ple_window);
6476 	}
6477 
6478 	if (vmx->nested.need_vmcs12_to_shadow_sync)
6479 		nested_sync_vmcs12_to_shadow(vcpu);
6480 
6481 	if (kvm_register_is_dirty(vcpu, VCPU_REGS_RSP))
6482 		vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
6483 	if (kvm_register_is_dirty(vcpu, VCPU_REGS_RIP))
6484 		vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
6485 
6486 	cr3 = __get_current_cr3_fast();
6487 	if (unlikely(cr3 != vmx->loaded_vmcs->host_state.cr3)) {
6488 		vmcs_writel(HOST_CR3, cr3);
6489 		vmx->loaded_vmcs->host_state.cr3 = cr3;
6490 	}
6491 
6492 	cr4 = cr4_read_shadow();
6493 	if (unlikely(cr4 != vmx->loaded_vmcs->host_state.cr4)) {
6494 		vmcs_writel(HOST_CR4, cr4);
6495 		vmx->loaded_vmcs->host_state.cr4 = cr4;
6496 	}
6497 
6498 	/* When single-stepping over STI and MOV SS, we must clear the
6499 	 * corresponding interruptibility bits in the guest state. Otherwise
6500 	 * vmentry fails as it then expects bit 14 (BS) in pending debug
6501 	 * exceptions being set, but that's not correct for the guest debugging
6502 	 * case. */
6503 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6504 		vmx_set_interrupt_shadow(vcpu, 0);
6505 
6506 	kvm_load_guest_xsave_state(vcpu);
6507 
6508 	if (static_cpu_has(X86_FEATURE_PKU) &&
6509 	    kvm_read_cr4_bits(vcpu, X86_CR4_PKE) &&
6510 	    vcpu->arch.pkru != vmx->host_pkru)
6511 		__write_pkru(vcpu->arch.pkru);
6512 
6513 	pt_guest_enter(vmx);
6514 
6515 	atomic_switch_perf_msrs(vmx);
6516 	atomic_switch_umwait_control_msr(vmx);
6517 
6518 	if (enable_preemption_timer)
6519 		vmx_update_hv_timer(vcpu);
6520 
6521 	if (lapic_in_kernel(vcpu) &&
6522 		vcpu->arch.apic->lapic_timer.timer_advance_ns)
6523 		kvm_wait_lapic_expire(vcpu);
6524 
6525 	/*
6526 	 * If this vCPU has touched SPEC_CTRL, restore the guest's value if
6527 	 * it's non-zero. Since vmentry is serialising on affected CPUs, there
6528 	 * is no need to worry about the conditional branch over the wrmsr
6529 	 * being speculatively taken.
6530 	 */
6531 	x86_spec_ctrl_set_guest(vmx->spec_ctrl, 0);
6532 
6533 	/* L1D Flush includes CPU buffer clear to mitigate MDS */
6534 	if (static_branch_unlikely(&vmx_l1d_should_flush))
6535 		vmx_l1d_flush(vcpu);
6536 	else if (static_branch_unlikely(&mds_user_clear))
6537 		mds_clear_cpu_buffers();
6538 
6539 	if (vcpu->arch.cr2 != read_cr2())
6540 		write_cr2(vcpu->arch.cr2);
6541 
6542 	vmx->fail = __vmx_vcpu_run(vmx, (unsigned long *)&vcpu->arch.regs,
6543 				   vmx->loaded_vmcs->launched);
6544 
6545 	vcpu->arch.cr2 = read_cr2();
6546 
6547 	/*
6548 	 * We do not use IBRS in the kernel. If this vCPU has used the
6549 	 * SPEC_CTRL MSR it may have left it on; save the value and
6550 	 * turn it off. This is much more efficient than blindly adding
6551 	 * it to the atomic save/restore list. Especially as the former
6552 	 * (Saving guest MSRs on vmexit) doesn't even exist in KVM.
6553 	 *
6554 	 * For non-nested case:
6555 	 * If the L01 MSR bitmap does not intercept the MSR, then we need to
6556 	 * save it.
6557 	 *
6558 	 * For nested case:
6559 	 * If the L02 MSR bitmap does not intercept the MSR, then we need to
6560 	 * save it.
6561 	 */
6562 	if (unlikely(!msr_write_intercepted(vcpu, MSR_IA32_SPEC_CTRL)))
6563 		vmx->spec_ctrl = native_read_msr(MSR_IA32_SPEC_CTRL);
6564 
6565 	x86_spec_ctrl_restore_host(vmx->spec_ctrl, 0);
6566 
6567 	/* All fields are clean at this point */
6568 	if (static_branch_unlikely(&enable_evmcs))
6569 		current_evmcs->hv_clean_fields |=
6570 			HV_VMX_ENLIGHTENED_CLEAN_FIELD_ALL;
6571 
6572 	if (static_branch_unlikely(&enable_evmcs))
6573 		current_evmcs->hv_vp_id = vcpu->arch.hyperv.vp_index;
6574 
6575 	/* MSR_IA32_DEBUGCTLMSR is zeroed on vmexit. Restore it if needed */
6576 	if (vmx->host_debugctlmsr)
6577 		update_debugctlmsr(vmx->host_debugctlmsr);
6578 
6579 #ifndef CONFIG_X86_64
6580 	/*
6581 	 * The sysexit path does not restore ds/es, so we must set them to
6582 	 * a reasonable value ourselves.
6583 	 *
6584 	 * We can't defer this to vmx_prepare_switch_to_host() since that
6585 	 * function may be executed in interrupt context, which saves and
6586 	 * restore segments around it, nullifying its effect.
6587 	 */
6588 	loadsegment(ds, __USER_DS);
6589 	loadsegment(es, __USER_DS);
6590 #endif
6591 
6592 	vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
6593 				  | (1 << VCPU_EXREG_RFLAGS)
6594 				  | (1 << VCPU_EXREG_PDPTR)
6595 				  | (1 << VCPU_EXREG_SEGMENTS)
6596 				  | (1 << VCPU_EXREG_CR3));
6597 	vcpu->arch.regs_dirty = 0;
6598 
6599 	pt_guest_exit(vmx);
6600 
6601 	/*
6602 	 * eager fpu is enabled if PKEY is supported and CR4 is switched
6603 	 * back on host, so it is safe to read guest PKRU from current
6604 	 * XSAVE.
6605 	 */
6606 	if (static_cpu_has(X86_FEATURE_PKU) &&
6607 	    kvm_read_cr4_bits(vcpu, X86_CR4_PKE)) {
6608 		vcpu->arch.pkru = rdpkru();
6609 		if (vcpu->arch.pkru != vmx->host_pkru)
6610 			__write_pkru(vmx->host_pkru);
6611 	}
6612 
6613 	kvm_load_host_xsave_state(vcpu);
6614 
6615 	vmx->nested.nested_run_pending = 0;
6616 	vmx->idt_vectoring_info = 0;
6617 
6618 	vmx->exit_reason = vmx->fail ? 0xdead : vmcs_read32(VM_EXIT_REASON);
6619 	if ((u16)vmx->exit_reason == EXIT_REASON_MCE_DURING_VMENTRY)
6620 		kvm_machine_check();
6621 
6622 	if (vmx->fail || (vmx->exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY))
6623 		return;
6624 
6625 	vmx->loaded_vmcs->launched = 1;
6626 	vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
6627 
6628 	vmx_recover_nmi_blocking(vmx);
6629 	vmx_complete_interrupts(vmx);
6630 }
6631 
6632 static struct kvm *vmx_vm_alloc(void)
6633 {
6634 	struct kvm_vmx *kvm_vmx = __vmalloc(sizeof(struct kvm_vmx),
6635 					    GFP_KERNEL_ACCOUNT | __GFP_ZERO,
6636 					    PAGE_KERNEL);
6637 	return &kvm_vmx->kvm;
6638 }
6639 
6640 static void vmx_vm_free(struct kvm *kvm)
6641 {
6642 	kfree(kvm->arch.hyperv.hv_pa_pg);
6643 	vfree(to_kvm_vmx(kvm));
6644 }
6645 
6646 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
6647 {
6648 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6649 
6650 	if (enable_pml)
6651 		vmx_destroy_pml_buffer(vmx);
6652 	free_vpid(vmx->vpid);
6653 	nested_vmx_free_vcpu(vcpu);
6654 	free_loaded_vmcs(vmx->loaded_vmcs);
6655 }
6656 
6657 static int vmx_create_vcpu(struct kvm_vcpu *vcpu)
6658 {
6659 	struct vcpu_vmx *vmx;
6660 	unsigned long *msr_bitmap;
6661 	int i, cpu, err;
6662 
6663 	BUILD_BUG_ON(offsetof(struct vcpu_vmx, vcpu) != 0);
6664 	vmx = to_vmx(vcpu);
6665 
6666 	err = -ENOMEM;
6667 
6668 	vmx->vpid = allocate_vpid();
6669 
6670 	/*
6671 	 * If PML is turned on, failure on enabling PML just results in failure
6672 	 * of creating the vcpu, therefore we can simplify PML logic (by
6673 	 * avoiding dealing with cases, such as enabling PML partially on vcpus
6674 	 * for the guest), etc.
6675 	 */
6676 	if (enable_pml) {
6677 		vmx->pml_pg = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
6678 		if (!vmx->pml_pg)
6679 			goto free_vpid;
6680 	}
6681 
6682 	BUILD_BUG_ON(ARRAY_SIZE(vmx_msr_index) != NR_SHARED_MSRS);
6683 
6684 	for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i) {
6685 		u32 index = vmx_msr_index[i];
6686 		u32 data_low, data_high;
6687 		int j = vmx->nmsrs;
6688 
6689 		if (rdmsr_safe(index, &data_low, &data_high) < 0)
6690 			continue;
6691 		if (wrmsr_safe(index, data_low, data_high) < 0)
6692 			continue;
6693 
6694 		vmx->guest_msrs[j].index = i;
6695 		vmx->guest_msrs[j].data = 0;
6696 		switch (index) {
6697 		case MSR_IA32_TSX_CTRL:
6698 			/*
6699 			 * No need to pass TSX_CTRL_CPUID_CLEAR through, so
6700 			 * let's avoid changing CPUID bits under the host
6701 			 * kernel's feet.
6702 			 */
6703 			vmx->guest_msrs[j].mask = ~(u64)TSX_CTRL_CPUID_CLEAR;
6704 			break;
6705 		default:
6706 			vmx->guest_msrs[j].mask = -1ull;
6707 			break;
6708 		}
6709 		++vmx->nmsrs;
6710 	}
6711 
6712 	err = alloc_loaded_vmcs(&vmx->vmcs01);
6713 	if (err < 0)
6714 		goto free_pml;
6715 
6716 	msr_bitmap = vmx->vmcs01.msr_bitmap;
6717 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_TSC, MSR_TYPE_R);
6718 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_FS_BASE, MSR_TYPE_RW);
6719 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_GS_BASE, MSR_TYPE_RW);
6720 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_KERNEL_GS_BASE, MSR_TYPE_RW);
6721 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_CS, MSR_TYPE_RW);
6722 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_ESP, MSR_TYPE_RW);
6723 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_EIP, MSR_TYPE_RW);
6724 	if (kvm_cstate_in_guest(vcpu->kvm)) {
6725 		vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C1_RES, MSR_TYPE_R);
6726 		vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C3_RESIDENCY, MSR_TYPE_R);
6727 		vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C6_RESIDENCY, MSR_TYPE_R);
6728 		vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C7_RESIDENCY, MSR_TYPE_R);
6729 	}
6730 	vmx->msr_bitmap_mode = 0;
6731 
6732 	vmx->loaded_vmcs = &vmx->vmcs01;
6733 	cpu = get_cpu();
6734 	vmx_vcpu_load(vcpu, cpu);
6735 	vcpu->cpu = cpu;
6736 	init_vmcs(vmx);
6737 	vmx_vcpu_put(vcpu);
6738 	put_cpu();
6739 	if (cpu_need_virtualize_apic_accesses(vcpu)) {
6740 		err = alloc_apic_access_page(vcpu->kvm);
6741 		if (err)
6742 			goto free_vmcs;
6743 	}
6744 
6745 	if (enable_ept && !enable_unrestricted_guest) {
6746 		err = init_rmode_identity_map(vcpu->kvm);
6747 		if (err)
6748 			goto free_vmcs;
6749 	}
6750 
6751 	if (nested)
6752 		nested_vmx_setup_ctls_msrs(&vmx->nested.msrs,
6753 					   vmx_capability.ept,
6754 					   kvm_vcpu_apicv_active(vcpu));
6755 	else
6756 		memset(&vmx->nested.msrs, 0, sizeof(vmx->nested.msrs));
6757 
6758 	vmx->nested.posted_intr_nv = -1;
6759 	vmx->nested.current_vmptr = -1ull;
6760 
6761 	vmx->msr_ia32_feature_control_valid_bits = FEAT_CTL_LOCKED;
6762 
6763 	/*
6764 	 * Enforce invariant: pi_desc.nv is always either POSTED_INTR_VECTOR
6765 	 * or POSTED_INTR_WAKEUP_VECTOR.
6766 	 */
6767 	vmx->pi_desc.nv = POSTED_INTR_VECTOR;
6768 	vmx->pi_desc.sn = 1;
6769 
6770 	vmx->ept_pointer = INVALID_PAGE;
6771 
6772 	return 0;
6773 
6774 free_vmcs:
6775 	free_loaded_vmcs(vmx->loaded_vmcs);
6776 free_pml:
6777 	vmx_destroy_pml_buffer(vmx);
6778 free_vpid:
6779 	free_vpid(vmx->vpid);
6780 	return err;
6781 }
6782 
6783 #define L1TF_MSG_SMT "L1TF CPU bug present and SMT on, data leak possible. See CVE-2018-3646 and https://www.kernel.org/doc/html/latest/admin-guide/hw-vuln/l1tf.html for details.\n"
6784 #define L1TF_MSG_L1D "L1TF CPU bug present and virtualization mitigation disabled, data leak possible. See CVE-2018-3646 and https://www.kernel.org/doc/html/latest/admin-guide/hw-vuln/l1tf.html for details.\n"
6785 
6786 static int vmx_vm_init(struct kvm *kvm)
6787 {
6788 	spin_lock_init(&to_kvm_vmx(kvm)->ept_pointer_lock);
6789 
6790 	if (!ple_gap)
6791 		kvm->arch.pause_in_guest = true;
6792 
6793 	if (boot_cpu_has(X86_BUG_L1TF) && enable_ept) {
6794 		switch (l1tf_mitigation) {
6795 		case L1TF_MITIGATION_OFF:
6796 		case L1TF_MITIGATION_FLUSH_NOWARN:
6797 			/* 'I explicitly don't care' is set */
6798 			break;
6799 		case L1TF_MITIGATION_FLUSH:
6800 		case L1TF_MITIGATION_FLUSH_NOSMT:
6801 		case L1TF_MITIGATION_FULL:
6802 			/*
6803 			 * Warn upon starting the first VM in a potentially
6804 			 * insecure environment.
6805 			 */
6806 			if (sched_smt_active())
6807 				pr_warn_once(L1TF_MSG_SMT);
6808 			if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_NEVER)
6809 				pr_warn_once(L1TF_MSG_L1D);
6810 			break;
6811 		case L1TF_MITIGATION_FULL_FORCE:
6812 			/* Flush is enforced */
6813 			break;
6814 		}
6815 	}
6816 	return 0;
6817 }
6818 
6819 static int __init vmx_check_processor_compat(void)
6820 {
6821 	struct vmcs_config vmcs_conf;
6822 	struct vmx_capability vmx_cap;
6823 
6824 	if (!this_cpu_has(X86_FEATURE_MSR_IA32_FEAT_CTL) ||
6825 	    !this_cpu_has(X86_FEATURE_VMX)) {
6826 		pr_err("kvm: VMX is disabled on CPU %d\n", smp_processor_id());
6827 		return -EIO;
6828 	}
6829 
6830 	if (setup_vmcs_config(&vmcs_conf, &vmx_cap) < 0)
6831 		return -EIO;
6832 	if (nested)
6833 		nested_vmx_setup_ctls_msrs(&vmcs_conf.nested, vmx_cap.ept,
6834 					   enable_apicv);
6835 	if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
6836 		printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
6837 				smp_processor_id());
6838 		return -EIO;
6839 	}
6840 	return 0;
6841 }
6842 
6843 static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
6844 {
6845 	u8 cache;
6846 	u64 ipat = 0;
6847 
6848 	/* For VT-d and EPT combination
6849 	 * 1. MMIO: always map as UC
6850 	 * 2. EPT with VT-d:
6851 	 *   a. VT-d without snooping control feature: can't guarantee the
6852 	 *	result, try to trust guest.
6853 	 *   b. VT-d with snooping control feature: snooping control feature of
6854 	 *	VT-d engine can guarantee the cache correctness. Just set it
6855 	 *	to WB to keep consistent with host. So the same as item 3.
6856 	 * 3. EPT without VT-d: always map as WB and set IPAT=1 to keep
6857 	 *    consistent with host MTRR
6858 	 */
6859 	if (is_mmio) {
6860 		cache = MTRR_TYPE_UNCACHABLE;
6861 		goto exit;
6862 	}
6863 
6864 	if (!kvm_arch_has_noncoherent_dma(vcpu->kvm)) {
6865 		ipat = VMX_EPT_IPAT_BIT;
6866 		cache = MTRR_TYPE_WRBACK;
6867 		goto exit;
6868 	}
6869 
6870 	if (kvm_read_cr0(vcpu) & X86_CR0_CD) {
6871 		ipat = VMX_EPT_IPAT_BIT;
6872 		if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
6873 			cache = MTRR_TYPE_WRBACK;
6874 		else
6875 			cache = MTRR_TYPE_UNCACHABLE;
6876 		goto exit;
6877 	}
6878 
6879 	cache = kvm_mtrr_get_guest_memory_type(vcpu, gfn);
6880 
6881 exit:
6882 	return (cache << VMX_EPT_MT_EPTE_SHIFT) | ipat;
6883 }
6884 
6885 static int vmx_get_lpage_level(void)
6886 {
6887 	if (enable_ept && !cpu_has_vmx_ept_1g_page())
6888 		return PT_DIRECTORY_LEVEL;
6889 	else
6890 		/* For shadow and EPT supported 1GB page */
6891 		return PT_PDPE_LEVEL;
6892 }
6893 
6894 static void vmcs_set_secondary_exec_control(struct vcpu_vmx *vmx)
6895 {
6896 	/*
6897 	 * These bits in the secondary execution controls field
6898 	 * are dynamic, the others are mostly based on the hypervisor
6899 	 * architecture and the guest's CPUID.  Do not touch the
6900 	 * dynamic bits.
6901 	 */
6902 	u32 mask =
6903 		SECONDARY_EXEC_SHADOW_VMCS |
6904 		SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
6905 		SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
6906 		SECONDARY_EXEC_DESC;
6907 
6908 	u32 new_ctl = vmx->secondary_exec_control;
6909 	u32 cur_ctl = secondary_exec_controls_get(vmx);
6910 
6911 	secondary_exec_controls_set(vmx, (new_ctl & ~mask) | (cur_ctl & mask));
6912 }
6913 
6914 /*
6915  * Generate MSR_IA32_VMX_CR{0,4}_FIXED1 according to CPUID. Only set bits
6916  * (indicating "allowed-1") if they are supported in the guest's CPUID.
6917  */
6918 static void nested_vmx_cr_fixed1_bits_update(struct kvm_vcpu *vcpu)
6919 {
6920 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6921 	struct kvm_cpuid_entry2 *entry;
6922 
6923 	vmx->nested.msrs.cr0_fixed1 = 0xffffffff;
6924 	vmx->nested.msrs.cr4_fixed1 = X86_CR4_PCE;
6925 
6926 #define cr4_fixed1_update(_cr4_mask, _reg, _cpuid_mask) do {		\
6927 	if (entry && (entry->_reg & (_cpuid_mask)))			\
6928 		vmx->nested.msrs.cr4_fixed1 |= (_cr4_mask);	\
6929 } while (0)
6930 
6931 	entry = kvm_find_cpuid_entry(vcpu, 0x1, 0);
6932 	cr4_fixed1_update(X86_CR4_VME,        edx, feature_bit(VME));
6933 	cr4_fixed1_update(X86_CR4_PVI,        edx, feature_bit(VME));
6934 	cr4_fixed1_update(X86_CR4_TSD,        edx, feature_bit(TSC));
6935 	cr4_fixed1_update(X86_CR4_DE,         edx, feature_bit(DE));
6936 	cr4_fixed1_update(X86_CR4_PSE,        edx, feature_bit(PSE));
6937 	cr4_fixed1_update(X86_CR4_PAE,        edx, feature_bit(PAE));
6938 	cr4_fixed1_update(X86_CR4_MCE,        edx, feature_bit(MCE));
6939 	cr4_fixed1_update(X86_CR4_PGE,        edx, feature_bit(PGE));
6940 	cr4_fixed1_update(X86_CR4_OSFXSR,     edx, feature_bit(FXSR));
6941 	cr4_fixed1_update(X86_CR4_OSXMMEXCPT, edx, feature_bit(XMM));
6942 	cr4_fixed1_update(X86_CR4_VMXE,       ecx, feature_bit(VMX));
6943 	cr4_fixed1_update(X86_CR4_SMXE,       ecx, feature_bit(SMX));
6944 	cr4_fixed1_update(X86_CR4_PCIDE,      ecx, feature_bit(PCID));
6945 	cr4_fixed1_update(X86_CR4_OSXSAVE,    ecx, feature_bit(XSAVE));
6946 
6947 	entry = kvm_find_cpuid_entry(vcpu, 0x7, 0);
6948 	cr4_fixed1_update(X86_CR4_FSGSBASE,   ebx, feature_bit(FSGSBASE));
6949 	cr4_fixed1_update(X86_CR4_SMEP,       ebx, feature_bit(SMEP));
6950 	cr4_fixed1_update(X86_CR4_SMAP,       ebx, feature_bit(SMAP));
6951 	cr4_fixed1_update(X86_CR4_PKE,        ecx, feature_bit(PKU));
6952 	cr4_fixed1_update(X86_CR4_UMIP,       ecx, feature_bit(UMIP));
6953 	cr4_fixed1_update(X86_CR4_LA57,       ecx, feature_bit(LA57));
6954 
6955 #undef cr4_fixed1_update
6956 }
6957 
6958 static void nested_vmx_entry_exit_ctls_update(struct kvm_vcpu *vcpu)
6959 {
6960 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6961 
6962 	if (kvm_mpx_supported()) {
6963 		bool mpx_enabled = guest_cpuid_has(vcpu, X86_FEATURE_MPX);
6964 
6965 		if (mpx_enabled) {
6966 			vmx->nested.msrs.entry_ctls_high |= VM_ENTRY_LOAD_BNDCFGS;
6967 			vmx->nested.msrs.exit_ctls_high |= VM_EXIT_CLEAR_BNDCFGS;
6968 		} else {
6969 			vmx->nested.msrs.entry_ctls_high &= ~VM_ENTRY_LOAD_BNDCFGS;
6970 			vmx->nested.msrs.exit_ctls_high &= ~VM_EXIT_CLEAR_BNDCFGS;
6971 		}
6972 	}
6973 }
6974 
6975 static void update_intel_pt_cfg(struct kvm_vcpu *vcpu)
6976 {
6977 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6978 	struct kvm_cpuid_entry2 *best = NULL;
6979 	int i;
6980 
6981 	for (i = 0; i < PT_CPUID_LEAVES; i++) {
6982 		best = kvm_find_cpuid_entry(vcpu, 0x14, i);
6983 		if (!best)
6984 			return;
6985 		vmx->pt_desc.caps[CPUID_EAX + i*PT_CPUID_REGS_NUM] = best->eax;
6986 		vmx->pt_desc.caps[CPUID_EBX + i*PT_CPUID_REGS_NUM] = best->ebx;
6987 		vmx->pt_desc.caps[CPUID_ECX + i*PT_CPUID_REGS_NUM] = best->ecx;
6988 		vmx->pt_desc.caps[CPUID_EDX + i*PT_CPUID_REGS_NUM] = best->edx;
6989 	}
6990 
6991 	/* Get the number of configurable Address Ranges for filtering */
6992 	vmx->pt_desc.addr_range = intel_pt_validate_cap(vmx->pt_desc.caps,
6993 						PT_CAP_num_address_ranges);
6994 
6995 	/* Initialize and clear the no dependency bits */
6996 	vmx->pt_desc.ctl_bitmask = ~(RTIT_CTL_TRACEEN | RTIT_CTL_OS |
6997 			RTIT_CTL_USR | RTIT_CTL_TSC_EN | RTIT_CTL_DISRETC);
6998 
6999 	/*
7000 	 * If CPUID.(EAX=14H,ECX=0):EBX[0]=1 CR3Filter can be set otherwise
7001 	 * will inject an #GP
7002 	 */
7003 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_cr3_filtering))
7004 		vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_CR3EN;
7005 
7006 	/*
7007 	 * If CPUID.(EAX=14H,ECX=0):EBX[1]=1 CYCEn, CycThresh and
7008 	 * PSBFreq can be set
7009 	 */
7010 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc))
7011 		vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_CYCLEACC |
7012 				RTIT_CTL_CYC_THRESH | RTIT_CTL_PSB_FREQ);
7013 
7014 	/*
7015 	 * If CPUID.(EAX=14H,ECX=0):EBX[3]=1 MTCEn BranchEn and
7016 	 * MTCFreq can be set
7017 	 */
7018 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc))
7019 		vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_MTC_EN |
7020 				RTIT_CTL_BRANCH_EN | RTIT_CTL_MTC_RANGE);
7021 
7022 	/* If CPUID.(EAX=14H,ECX=0):EBX[4]=1 FUPonPTW and PTWEn can be set */
7023 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_ptwrite))
7024 		vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_FUP_ON_PTW |
7025 							RTIT_CTL_PTW_EN);
7026 
7027 	/* If CPUID.(EAX=14H,ECX=0):EBX[5]=1 PwrEvEn can be set */
7028 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_power_event_trace))
7029 		vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_PWR_EVT_EN;
7030 
7031 	/* If CPUID.(EAX=14H,ECX=0):ECX[0]=1 ToPA can be set */
7032 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_topa_output))
7033 		vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_TOPA;
7034 
7035 	/* If CPUID.(EAX=14H,ECX=0):ECX[3]=1 FabircEn can be set */
7036 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_output_subsys))
7037 		vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_FABRIC_EN;
7038 
7039 	/* unmask address range configure area */
7040 	for (i = 0; i < vmx->pt_desc.addr_range; i++)
7041 		vmx->pt_desc.ctl_bitmask &= ~(0xfULL << (32 + i * 4));
7042 }
7043 
7044 static void vmx_cpuid_update(struct kvm_vcpu *vcpu)
7045 {
7046 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7047 
7048 	/* xsaves_enabled is recomputed in vmx_compute_secondary_exec_control(). */
7049 	vcpu->arch.xsaves_enabled = false;
7050 
7051 	if (cpu_has_secondary_exec_ctrls()) {
7052 		vmx_compute_secondary_exec_control(vmx);
7053 		vmcs_set_secondary_exec_control(vmx);
7054 	}
7055 
7056 	if (nested_vmx_allowed(vcpu))
7057 		to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
7058 			FEAT_CTL_VMX_ENABLED_INSIDE_SMX |
7059 			FEAT_CTL_VMX_ENABLED_OUTSIDE_SMX;
7060 	else
7061 		to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
7062 			~(FEAT_CTL_VMX_ENABLED_INSIDE_SMX |
7063 			  FEAT_CTL_VMX_ENABLED_OUTSIDE_SMX);
7064 
7065 	if (nested_vmx_allowed(vcpu)) {
7066 		nested_vmx_cr_fixed1_bits_update(vcpu);
7067 		nested_vmx_entry_exit_ctls_update(vcpu);
7068 	}
7069 
7070 	if (boot_cpu_has(X86_FEATURE_INTEL_PT) &&
7071 			guest_cpuid_has(vcpu, X86_FEATURE_INTEL_PT))
7072 		update_intel_pt_cfg(vcpu);
7073 
7074 	if (boot_cpu_has(X86_FEATURE_RTM)) {
7075 		struct shared_msr_entry *msr;
7076 		msr = find_msr_entry(vmx, MSR_IA32_TSX_CTRL);
7077 		if (msr) {
7078 			bool enabled = guest_cpuid_has(vcpu, X86_FEATURE_RTM);
7079 			vmx_set_guest_msr(vmx, msr, enabled ? 0 : TSX_CTRL_RTM_DISABLE);
7080 		}
7081 	}
7082 }
7083 
7084 static void vmx_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
7085 {
7086 	if (func == 1 && nested)
7087 		entry->ecx |= feature_bit(VMX);
7088 }
7089 
7090 static void vmx_request_immediate_exit(struct kvm_vcpu *vcpu)
7091 {
7092 	to_vmx(vcpu)->req_immediate_exit = true;
7093 }
7094 
7095 static int vmx_check_intercept(struct kvm_vcpu *vcpu,
7096 			       struct x86_instruction_info *info,
7097 			       enum x86_intercept_stage stage)
7098 {
7099 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
7100 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
7101 
7102 	/*
7103 	 * RDPID causes #UD if disabled through secondary execution controls.
7104 	 * Because it is marked as EmulateOnUD, we need to intercept it here.
7105 	 */
7106 	if (info->intercept == x86_intercept_rdtscp &&
7107 	    !nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDTSCP)) {
7108 		ctxt->exception.vector = UD_VECTOR;
7109 		ctxt->exception.error_code_valid = false;
7110 		return X86EMUL_PROPAGATE_FAULT;
7111 	}
7112 
7113 	/* TODO: check more intercepts... */
7114 	return X86EMUL_CONTINUE;
7115 }
7116 
7117 #ifdef CONFIG_X86_64
7118 /* (a << shift) / divisor, return 1 if overflow otherwise 0 */
7119 static inline int u64_shl_div_u64(u64 a, unsigned int shift,
7120 				  u64 divisor, u64 *result)
7121 {
7122 	u64 low = a << shift, high = a >> (64 - shift);
7123 
7124 	/* To avoid the overflow on divq */
7125 	if (high >= divisor)
7126 		return 1;
7127 
7128 	/* Low hold the result, high hold rem which is discarded */
7129 	asm("divq %2\n\t" : "=a" (low), "=d" (high) :
7130 	    "rm" (divisor), "0" (low), "1" (high));
7131 	*result = low;
7132 
7133 	return 0;
7134 }
7135 
7136 static int vmx_set_hv_timer(struct kvm_vcpu *vcpu, u64 guest_deadline_tsc,
7137 			    bool *expired)
7138 {
7139 	struct vcpu_vmx *vmx;
7140 	u64 tscl, guest_tscl, delta_tsc, lapic_timer_advance_cycles;
7141 	struct kvm_timer *ktimer = &vcpu->arch.apic->lapic_timer;
7142 
7143 	if (kvm_mwait_in_guest(vcpu->kvm) ||
7144 		kvm_can_post_timer_interrupt(vcpu))
7145 		return -EOPNOTSUPP;
7146 
7147 	vmx = to_vmx(vcpu);
7148 	tscl = rdtsc();
7149 	guest_tscl = kvm_read_l1_tsc(vcpu, tscl);
7150 	delta_tsc = max(guest_deadline_tsc, guest_tscl) - guest_tscl;
7151 	lapic_timer_advance_cycles = nsec_to_cycles(vcpu,
7152 						    ktimer->timer_advance_ns);
7153 
7154 	if (delta_tsc > lapic_timer_advance_cycles)
7155 		delta_tsc -= lapic_timer_advance_cycles;
7156 	else
7157 		delta_tsc = 0;
7158 
7159 	/* Convert to host delta tsc if tsc scaling is enabled */
7160 	if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio &&
7161 	    delta_tsc && u64_shl_div_u64(delta_tsc,
7162 				kvm_tsc_scaling_ratio_frac_bits,
7163 				vcpu->arch.tsc_scaling_ratio, &delta_tsc))
7164 		return -ERANGE;
7165 
7166 	/*
7167 	 * If the delta tsc can't fit in the 32 bit after the multi shift,
7168 	 * we can't use the preemption timer.
7169 	 * It's possible that it fits on later vmentries, but checking
7170 	 * on every vmentry is costly so we just use an hrtimer.
7171 	 */
7172 	if (delta_tsc >> (cpu_preemption_timer_multi + 32))
7173 		return -ERANGE;
7174 
7175 	vmx->hv_deadline_tsc = tscl + delta_tsc;
7176 	*expired = !delta_tsc;
7177 	return 0;
7178 }
7179 
7180 static void vmx_cancel_hv_timer(struct kvm_vcpu *vcpu)
7181 {
7182 	to_vmx(vcpu)->hv_deadline_tsc = -1;
7183 }
7184 #endif
7185 
7186 static void vmx_sched_in(struct kvm_vcpu *vcpu, int cpu)
7187 {
7188 	if (!kvm_pause_in_guest(vcpu->kvm))
7189 		shrink_ple_window(vcpu);
7190 }
7191 
7192 static void vmx_slot_enable_log_dirty(struct kvm *kvm,
7193 				     struct kvm_memory_slot *slot)
7194 {
7195 	kvm_mmu_slot_leaf_clear_dirty(kvm, slot);
7196 	kvm_mmu_slot_largepage_remove_write_access(kvm, slot);
7197 }
7198 
7199 static void vmx_slot_disable_log_dirty(struct kvm *kvm,
7200 				       struct kvm_memory_slot *slot)
7201 {
7202 	kvm_mmu_slot_set_dirty(kvm, slot);
7203 }
7204 
7205 static void vmx_flush_log_dirty(struct kvm *kvm)
7206 {
7207 	kvm_flush_pml_buffers(kvm);
7208 }
7209 
7210 static int vmx_write_pml_buffer(struct kvm_vcpu *vcpu)
7211 {
7212 	struct vmcs12 *vmcs12;
7213 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7214 	gpa_t gpa, dst;
7215 
7216 	if (is_guest_mode(vcpu)) {
7217 		WARN_ON_ONCE(vmx->nested.pml_full);
7218 
7219 		/*
7220 		 * Check if PML is enabled for the nested guest.
7221 		 * Whether eptp bit 6 is set is already checked
7222 		 * as part of A/D emulation.
7223 		 */
7224 		vmcs12 = get_vmcs12(vcpu);
7225 		if (!nested_cpu_has_pml(vmcs12))
7226 			return 0;
7227 
7228 		if (vmcs12->guest_pml_index >= PML_ENTITY_NUM) {
7229 			vmx->nested.pml_full = true;
7230 			return 1;
7231 		}
7232 
7233 		gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS) & ~0xFFFull;
7234 		dst = vmcs12->pml_address + sizeof(u64) * vmcs12->guest_pml_index;
7235 
7236 		if (kvm_write_guest_page(vcpu->kvm, gpa_to_gfn(dst), &gpa,
7237 					 offset_in_page(dst), sizeof(gpa)))
7238 			return 0;
7239 
7240 		vmcs12->guest_pml_index--;
7241 	}
7242 
7243 	return 0;
7244 }
7245 
7246 static void vmx_enable_log_dirty_pt_masked(struct kvm *kvm,
7247 					   struct kvm_memory_slot *memslot,
7248 					   gfn_t offset, unsigned long mask)
7249 {
7250 	kvm_mmu_clear_dirty_pt_masked(kvm, memslot, offset, mask);
7251 }
7252 
7253 static void __pi_post_block(struct kvm_vcpu *vcpu)
7254 {
7255 	struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
7256 	struct pi_desc old, new;
7257 	unsigned int dest;
7258 
7259 	do {
7260 		old.control = new.control = pi_desc->control;
7261 		WARN(old.nv != POSTED_INTR_WAKEUP_VECTOR,
7262 		     "Wakeup handler not enabled while the VCPU is blocked\n");
7263 
7264 		dest = cpu_physical_id(vcpu->cpu);
7265 
7266 		if (x2apic_enabled())
7267 			new.ndst = dest;
7268 		else
7269 			new.ndst = (dest << 8) & 0xFF00;
7270 
7271 		/* set 'NV' to 'notification vector' */
7272 		new.nv = POSTED_INTR_VECTOR;
7273 	} while (cmpxchg64(&pi_desc->control, old.control,
7274 			   new.control) != old.control);
7275 
7276 	if (!WARN_ON_ONCE(vcpu->pre_pcpu == -1)) {
7277 		spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7278 		list_del(&vcpu->blocked_vcpu_list);
7279 		spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7280 		vcpu->pre_pcpu = -1;
7281 	}
7282 }
7283 
7284 /*
7285  * This routine does the following things for vCPU which is going
7286  * to be blocked if VT-d PI is enabled.
7287  * - Store the vCPU to the wakeup list, so when interrupts happen
7288  *   we can find the right vCPU to wake up.
7289  * - Change the Posted-interrupt descriptor as below:
7290  *      'NDST' <-- vcpu->pre_pcpu
7291  *      'NV' <-- POSTED_INTR_WAKEUP_VECTOR
7292  * - If 'ON' is set during this process, which means at least one
7293  *   interrupt is posted for this vCPU, we cannot block it, in
7294  *   this case, return 1, otherwise, return 0.
7295  *
7296  */
7297 static int pi_pre_block(struct kvm_vcpu *vcpu)
7298 {
7299 	unsigned int dest;
7300 	struct pi_desc old, new;
7301 	struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
7302 
7303 	if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
7304 		!irq_remapping_cap(IRQ_POSTING_CAP)  ||
7305 		!kvm_vcpu_apicv_active(vcpu))
7306 		return 0;
7307 
7308 	WARN_ON(irqs_disabled());
7309 	local_irq_disable();
7310 	if (!WARN_ON_ONCE(vcpu->pre_pcpu != -1)) {
7311 		vcpu->pre_pcpu = vcpu->cpu;
7312 		spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7313 		list_add_tail(&vcpu->blocked_vcpu_list,
7314 			      &per_cpu(blocked_vcpu_on_cpu,
7315 				       vcpu->pre_pcpu));
7316 		spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7317 	}
7318 
7319 	do {
7320 		old.control = new.control = pi_desc->control;
7321 
7322 		WARN((pi_desc->sn == 1),
7323 		     "Warning: SN field of posted-interrupts "
7324 		     "is set before blocking\n");
7325 
7326 		/*
7327 		 * Since vCPU can be preempted during this process,
7328 		 * vcpu->cpu could be different with pre_pcpu, we
7329 		 * need to set pre_pcpu as the destination of wakeup
7330 		 * notification event, then we can find the right vCPU
7331 		 * to wakeup in wakeup handler if interrupts happen
7332 		 * when the vCPU is in blocked state.
7333 		 */
7334 		dest = cpu_physical_id(vcpu->pre_pcpu);
7335 
7336 		if (x2apic_enabled())
7337 			new.ndst = dest;
7338 		else
7339 			new.ndst = (dest << 8) & 0xFF00;
7340 
7341 		/* set 'NV' to 'wakeup vector' */
7342 		new.nv = POSTED_INTR_WAKEUP_VECTOR;
7343 	} while (cmpxchg64(&pi_desc->control, old.control,
7344 			   new.control) != old.control);
7345 
7346 	/* We should not block the vCPU if an interrupt is posted for it.  */
7347 	if (pi_test_on(pi_desc) == 1)
7348 		__pi_post_block(vcpu);
7349 
7350 	local_irq_enable();
7351 	return (vcpu->pre_pcpu == -1);
7352 }
7353 
7354 static int vmx_pre_block(struct kvm_vcpu *vcpu)
7355 {
7356 	if (pi_pre_block(vcpu))
7357 		return 1;
7358 
7359 	if (kvm_lapic_hv_timer_in_use(vcpu))
7360 		kvm_lapic_switch_to_sw_timer(vcpu);
7361 
7362 	return 0;
7363 }
7364 
7365 static void pi_post_block(struct kvm_vcpu *vcpu)
7366 {
7367 	if (vcpu->pre_pcpu == -1)
7368 		return;
7369 
7370 	WARN_ON(irqs_disabled());
7371 	local_irq_disable();
7372 	__pi_post_block(vcpu);
7373 	local_irq_enable();
7374 }
7375 
7376 static void vmx_post_block(struct kvm_vcpu *vcpu)
7377 {
7378 	if (kvm_x86_ops->set_hv_timer)
7379 		kvm_lapic_switch_to_hv_timer(vcpu);
7380 
7381 	pi_post_block(vcpu);
7382 }
7383 
7384 /*
7385  * vmx_update_pi_irte - set IRTE for Posted-Interrupts
7386  *
7387  * @kvm: kvm
7388  * @host_irq: host irq of the interrupt
7389  * @guest_irq: gsi of the interrupt
7390  * @set: set or unset PI
7391  * returns 0 on success, < 0 on failure
7392  */
7393 static int vmx_update_pi_irte(struct kvm *kvm, unsigned int host_irq,
7394 			      uint32_t guest_irq, bool set)
7395 {
7396 	struct kvm_kernel_irq_routing_entry *e;
7397 	struct kvm_irq_routing_table *irq_rt;
7398 	struct kvm_lapic_irq irq;
7399 	struct kvm_vcpu *vcpu;
7400 	struct vcpu_data vcpu_info;
7401 	int idx, ret = 0;
7402 
7403 	if (!kvm_arch_has_assigned_device(kvm) ||
7404 		!irq_remapping_cap(IRQ_POSTING_CAP) ||
7405 		!kvm_vcpu_apicv_active(kvm->vcpus[0]))
7406 		return 0;
7407 
7408 	idx = srcu_read_lock(&kvm->irq_srcu);
7409 	irq_rt = srcu_dereference(kvm->irq_routing, &kvm->irq_srcu);
7410 	if (guest_irq >= irq_rt->nr_rt_entries ||
7411 	    hlist_empty(&irq_rt->map[guest_irq])) {
7412 		pr_warn_once("no route for guest_irq %u/%u (broken user space?)\n",
7413 			     guest_irq, irq_rt->nr_rt_entries);
7414 		goto out;
7415 	}
7416 
7417 	hlist_for_each_entry(e, &irq_rt->map[guest_irq], link) {
7418 		if (e->type != KVM_IRQ_ROUTING_MSI)
7419 			continue;
7420 		/*
7421 		 * VT-d PI cannot support posting multicast/broadcast
7422 		 * interrupts to a vCPU, we still use interrupt remapping
7423 		 * for these kind of interrupts.
7424 		 *
7425 		 * For lowest-priority interrupts, we only support
7426 		 * those with single CPU as the destination, e.g. user
7427 		 * configures the interrupts via /proc/irq or uses
7428 		 * irqbalance to make the interrupts single-CPU.
7429 		 *
7430 		 * We will support full lowest-priority interrupt later.
7431 		 *
7432 		 * In addition, we can only inject generic interrupts using
7433 		 * the PI mechanism, refuse to route others through it.
7434 		 */
7435 
7436 		kvm_set_msi_irq(kvm, e, &irq);
7437 		if (!kvm_intr_is_single_vcpu(kvm, &irq, &vcpu) ||
7438 		    !kvm_irq_is_postable(&irq)) {
7439 			/*
7440 			 * Make sure the IRTE is in remapped mode if
7441 			 * we don't handle it in posted mode.
7442 			 */
7443 			ret = irq_set_vcpu_affinity(host_irq, NULL);
7444 			if (ret < 0) {
7445 				printk(KERN_INFO
7446 				   "failed to back to remapped mode, irq: %u\n",
7447 				   host_irq);
7448 				goto out;
7449 			}
7450 
7451 			continue;
7452 		}
7453 
7454 		vcpu_info.pi_desc_addr = __pa(vcpu_to_pi_desc(vcpu));
7455 		vcpu_info.vector = irq.vector;
7456 
7457 		trace_kvm_pi_irte_update(host_irq, vcpu->vcpu_id, e->gsi,
7458 				vcpu_info.vector, vcpu_info.pi_desc_addr, set);
7459 
7460 		if (set)
7461 			ret = irq_set_vcpu_affinity(host_irq, &vcpu_info);
7462 		else
7463 			ret = irq_set_vcpu_affinity(host_irq, NULL);
7464 
7465 		if (ret < 0) {
7466 			printk(KERN_INFO "%s: failed to update PI IRTE\n",
7467 					__func__);
7468 			goto out;
7469 		}
7470 	}
7471 
7472 	ret = 0;
7473 out:
7474 	srcu_read_unlock(&kvm->irq_srcu, idx);
7475 	return ret;
7476 }
7477 
7478 static void vmx_setup_mce(struct kvm_vcpu *vcpu)
7479 {
7480 	if (vcpu->arch.mcg_cap & MCG_LMCE_P)
7481 		to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
7482 			FEAT_CTL_LMCE_ENABLED;
7483 	else
7484 		to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
7485 			~FEAT_CTL_LMCE_ENABLED;
7486 }
7487 
7488 static int vmx_smi_allowed(struct kvm_vcpu *vcpu)
7489 {
7490 	/* we need a nested vmexit to enter SMM, postpone if run is pending */
7491 	if (to_vmx(vcpu)->nested.nested_run_pending)
7492 		return 0;
7493 	return 1;
7494 }
7495 
7496 static int vmx_pre_enter_smm(struct kvm_vcpu *vcpu, char *smstate)
7497 {
7498 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7499 
7500 	vmx->nested.smm.guest_mode = is_guest_mode(vcpu);
7501 	if (vmx->nested.smm.guest_mode)
7502 		nested_vmx_vmexit(vcpu, -1, 0, 0);
7503 
7504 	vmx->nested.smm.vmxon = vmx->nested.vmxon;
7505 	vmx->nested.vmxon = false;
7506 	vmx_clear_hlt(vcpu);
7507 	return 0;
7508 }
7509 
7510 static int vmx_pre_leave_smm(struct kvm_vcpu *vcpu, const char *smstate)
7511 {
7512 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7513 	int ret;
7514 
7515 	if (vmx->nested.smm.vmxon) {
7516 		vmx->nested.vmxon = true;
7517 		vmx->nested.smm.vmxon = false;
7518 	}
7519 
7520 	if (vmx->nested.smm.guest_mode) {
7521 		ret = nested_vmx_enter_non_root_mode(vcpu, false);
7522 		if (ret)
7523 			return ret;
7524 
7525 		vmx->nested.smm.guest_mode = false;
7526 	}
7527 	return 0;
7528 }
7529 
7530 static int enable_smi_window(struct kvm_vcpu *vcpu)
7531 {
7532 	return 0;
7533 }
7534 
7535 static bool vmx_need_emulation_on_page_fault(struct kvm_vcpu *vcpu)
7536 {
7537 	return false;
7538 }
7539 
7540 static bool vmx_apic_init_signal_blocked(struct kvm_vcpu *vcpu)
7541 {
7542 	return to_vmx(vcpu)->nested.vmxon;
7543 }
7544 
7545 static __init int hardware_setup(void)
7546 {
7547 	unsigned long host_bndcfgs;
7548 	struct desc_ptr dt;
7549 	int r, i;
7550 
7551 	rdmsrl_safe(MSR_EFER, &host_efer);
7552 
7553 	store_idt(&dt);
7554 	host_idt_base = dt.address;
7555 
7556 	for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i)
7557 		kvm_define_shared_msr(i, vmx_msr_index[i]);
7558 
7559 	if (setup_vmcs_config(&vmcs_config, &vmx_capability) < 0)
7560 		return -EIO;
7561 
7562 	if (boot_cpu_has(X86_FEATURE_NX))
7563 		kvm_enable_efer_bits(EFER_NX);
7564 
7565 	if (boot_cpu_has(X86_FEATURE_MPX)) {
7566 		rdmsrl(MSR_IA32_BNDCFGS, host_bndcfgs);
7567 		WARN_ONCE(host_bndcfgs, "KVM: BNDCFGS in host will be lost");
7568 	}
7569 
7570 	if (!cpu_has_vmx_vpid() || !cpu_has_vmx_invvpid() ||
7571 	    !(cpu_has_vmx_invvpid_single() || cpu_has_vmx_invvpid_global()))
7572 		enable_vpid = 0;
7573 
7574 	if (!cpu_has_vmx_ept() ||
7575 	    !cpu_has_vmx_ept_4levels() ||
7576 	    !cpu_has_vmx_ept_mt_wb() ||
7577 	    !cpu_has_vmx_invept_global())
7578 		enable_ept = 0;
7579 
7580 	if (!cpu_has_vmx_ept_ad_bits() || !enable_ept)
7581 		enable_ept_ad_bits = 0;
7582 
7583 	if (!cpu_has_vmx_unrestricted_guest() || !enable_ept)
7584 		enable_unrestricted_guest = 0;
7585 
7586 	if (!cpu_has_vmx_flexpriority())
7587 		flexpriority_enabled = 0;
7588 
7589 	if (!cpu_has_virtual_nmis())
7590 		enable_vnmi = 0;
7591 
7592 	/*
7593 	 * set_apic_access_page_addr() is used to reload apic access
7594 	 * page upon invalidation.  No need to do anything if not
7595 	 * using the APIC_ACCESS_ADDR VMCS field.
7596 	 */
7597 	if (!flexpriority_enabled)
7598 		kvm_x86_ops->set_apic_access_page_addr = NULL;
7599 
7600 	if (!cpu_has_vmx_tpr_shadow())
7601 		kvm_x86_ops->update_cr8_intercept = NULL;
7602 
7603 	if (enable_ept && !cpu_has_vmx_ept_2m_page())
7604 		kvm_disable_largepages();
7605 
7606 #if IS_ENABLED(CONFIG_HYPERV)
7607 	if (ms_hyperv.nested_features & HV_X64_NESTED_GUEST_MAPPING_FLUSH
7608 	    && enable_ept) {
7609 		kvm_x86_ops->tlb_remote_flush = hv_remote_flush_tlb;
7610 		kvm_x86_ops->tlb_remote_flush_with_range =
7611 				hv_remote_flush_tlb_with_range;
7612 	}
7613 #endif
7614 
7615 	if (!cpu_has_vmx_ple()) {
7616 		ple_gap = 0;
7617 		ple_window = 0;
7618 		ple_window_grow = 0;
7619 		ple_window_max = 0;
7620 		ple_window_shrink = 0;
7621 	}
7622 
7623 	if (!cpu_has_vmx_apicv()) {
7624 		enable_apicv = 0;
7625 		kvm_x86_ops->sync_pir_to_irr = NULL;
7626 	}
7627 
7628 	if (cpu_has_vmx_tsc_scaling()) {
7629 		kvm_has_tsc_control = true;
7630 		kvm_max_tsc_scaling_ratio = KVM_VMX_TSC_MULTIPLIER_MAX;
7631 		kvm_tsc_scaling_ratio_frac_bits = 48;
7632 	}
7633 
7634 	set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
7635 
7636 	if (enable_ept)
7637 		vmx_enable_tdp();
7638 	else
7639 		kvm_disable_tdp();
7640 
7641 	/*
7642 	 * Only enable PML when hardware supports PML feature, and both EPT
7643 	 * and EPT A/D bit features are enabled -- PML depends on them to work.
7644 	 */
7645 	if (!enable_ept || !enable_ept_ad_bits || !cpu_has_vmx_pml())
7646 		enable_pml = 0;
7647 
7648 	if (!enable_pml) {
7649 		kvm_x86_ops->slot_enable_log_dirty = NULL;
7650 		kvm_x86_ops->slot_disable_log_dirty = NULL;
7651 		kvm_x86_ops->flush_log_dirty = NULL;
7652 		kvm_x86_ops->enable_log_dirty_pt_masked = NULL;
7653 	}
7654 
7655 	if (!cpu_has_vmx_preemption_timer())
7656 		enable_preemption_timer = false;
7657 
7658 	if (enable_preemption_timer) {
7659 		u64 use_timer_freq = 5000ULL * 1000 * 1000;
7660 		u64 vmx_msr;
7661 
7662 		rdmsrl(MSR_IA32_VMX_MISC, vmx_msr);
7663 		cpu_preemption_timer_multi =
7664 			vmx_msr & VMX_MISC_PREEMPTION_TIMER_RATE_MASK;
7665 
7666 		if (tsc_khz)
7667 			use_timer_freq = (u64)tsc_khz * 1000;
7668 		use_timer_freq >>= cpu_preemption_timer_multi;
7669 
7670 		/*
7671 		 * KVM "disables" the preemption timer by setting it to its max
7672 		 * value.  Don't use the timer if it might cause spurious exits
7673 		 * at a rate faster than 0.1 Hz (of uninterrupted guest time).
7674 		 */
7675 		if (use_timer_freq > 0xffffffffu / 10)
7676 			enable_preemption_timer = false;
7677 	}
7678 
7679 	if (!enable_preemption_timer) {
7680 		kvm_x86_ops->set_hv_timer = NULL;
7681 		kvm_x86_ops->cancel_hv_timer = NULL;
7682 		kvm_x86_ops->request_immediate_exit = __kvm_request_immediate_exit;
7683 	}
7684 
7685 	kvm_set_posted_intr_wakeup_handler(wakeup_handler);
7686 
7687 	kvm_mce_cap_supported |= MCG_LMCE_P;
7688 
7689 	if (pt_mode != PT_MODE_SYSTEM && pt_mode != PT_MODE_HOST_GUEST)
7690 		return -EINVAL;
7691 	if (!enable_ept || !cpu_has_vmx_intel_pt())
7692 		pt_mode = PT_MODE_SYSTEM;
7693 
7694 	if (nested) {
7695 		nested_vmx_setup_ctls_msrs(&vmcs_config.nested,
7696 					   vmx_capability.ept, enable_apicv);
7697 
7698 		r = nested_vmx_hardware_setup(kvm_vmx_exit_handlers);
7699 		if (r)
7700 			return r;
7701 	}
7702 
7703 	r = alloc_kvm_area();
7704 	if (r)
7705 		nested_vmx_hardware_unsetup();
7706 	return r;
7707 }
7708 
7709 static __exit void hardware_unsetup(void)
7710 {
7711 	if (nested)
7712 		nested_vmx_hardware_unsetup();
7713 
7714 	free_kvm_area();
7715 }
7716 
7717 static struct kvm_x86_ops vmx_x86_ops __ro_after_init = {
7718 	.cpu_has_kvm_support = cpu_has_kvm_support,
7719 	.disabled_by_bios = vmx_disabled_by_bios,
7720 	.hardware_setup = hardware_setup,
7721 	.hardware_unsetup = hardware_unsetup,
7722 	.check_processor_compatibility = vmx_check_processor_compat,
7723 	.hardware_enable = hardware_enable,
7724 	.hardware_disable = hardware_disable,
7725 	.cpu_has_accelerated_tpr = report_flexpriority,
7726 	.has_emulated_msr = vmx_has_emulated_msr,
7727 
7728 	.vm_init = vmx_vm_init,
7729 	.vm_alloc = vmx_vm_alloc,
7730 	.vm_free = vmx_vm_free,
7731 
7732 	.vcpu_create = vmx_create_vcpu,
7733 	.vcpu_free = vmx_free_vcpu,
7734 	.vcpu_reset = vmx_vcpu_reset,
7735 
7736 	.prepare_guest_switch = vmx_prepare_switch_to_guest,
7737 	.vcpu_load = vmx_vcpu_load,
7738 	.vcpu_put = vmx_vcpu_put,
7739 
7740 	.update_bp_intercept = update_exception_bitmap,
7741 	.get_msr_feature = vmx_get_msr_feature,
7742 	.get_msr = vmx_get_msr,
7743 	.set_msr = vmx_set_msr,
7744 	.get_segment_base = vmx_get_segment_base,
7745 	.get_segment = vmx_get_segment,
7746 	.set_segment = vmx_set_segment,
7747 	.get_cpl = vmx_get_cpl,
7748 	.get_cs_db_l_bits = vmx_get_cs_db_l_bits,
7749 	.decache_cr0_guest_bits = vmx_decache_cr0_guest_bits,
7750 	.decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
7751 	.set_cr0 = vmx_set_cr0,
7752 	.set_cr3 = vmx_set_cr3,
7753 	.set_cr4 = vmx_set_cr4,
7754 	.set_efer = vmx_set_efer,
7755 	.get_idt = vmx_get_idt,
7756 	.set_idt = vmx_set_idt,
7757 	.get_gdt = vmx_get_gdt,
7758 	.set_gdt = vmx_set_gdt,
7759 	.get_dr6 = vmx_get_dr6,
7760 	.set_dr6 = vmx_set_dr6,
7761 	.set_dr7 = vmx_set_dr7,
7762 	.sync_dirty_debug_regs = vmx_sync_dirty_debug_regs,
7763 	.cache_reg = vmx_cache_reg,
7764 	.get_rflags = vmx_get_rflags,
7765 	.set_rflags = vmx_set_rflags,
7766 
7767 	.tlb_flush = vmx_flush_tlb,
7768 	.tlb_flush_gva = vmx_flush_tlb_gva,
7769 
7770 	.run = vmx_vcpu_run,
7771 	.handle_exit = vmx_handle_exit,
7772 	.skip_emulated_instruction = skip_emulated_instruction,
7773 	.set_interrupt_shadow = vmx_set_interrupt_shadow,
7774 	.get_interrupt_shadow = vmx_get_interrupt_shadow,
7775 	.patch_hypercall = vmx_patch_hypercall,
7776 	.set_irq = vmx_inject_irq,
7777 	.set_nmi = vmx_inject_nmi,
7778 	.queue_exception = vmx_queue_exception,
7779 	.cancel_injection = vmx_cancel_injection,
7780 	.interrupt_allowed = vmx_interrupt_allowed,
7781 	.nmi_allowed = vmx_nmi_allowed,
7782 	.get_nmi_mask = vmx_get_nmi_mask,
7783 	.set_nmi_mask = vmx_set_nmi_mask,
7784 	.enable_nmi_window = enable_nmi_window,
7785 	.enable_irq_window = enable_irq_window,
7786 	.update_cr8_intercept = update_cr8_intercept,
7787 	.set_virtual_apic_mode = vmx_set_virtual_apic_mode,
7788 	.set_apic_access_page_addr = vmx_set_apic_access_page_addr,
7789 	.get_enable_apicv = vmx_get_enable_apicv,
7790 	.refresh_apicv_exec_ctrl = vmx_refresh_apicv_exec_ctrl,
7791 	.load_eoi_exitmap = vmx_load_eoi_exitmap,
7792 	.apicv_post_state_restore = vmx_apicv_post_state_restore,
7793 	.hwapic_irr_update = vmx_hwapic_irr_update,
7794 	.hwapic_isr_update = vmx_hwapic_isr_update,
7795 	.guest_apic_has_interrupt = vmx_guest_apic_has_interrupt,
7796 	.sync_pir_to_irr = vmx_sync_pir_to_irr,
7797 	.deliver_posted_interrupt = vmx_deliver_posted_interrupt,
7798 	.dy_apicv_has_pending_interrupt = vmx_dy_apicv_has_pending_interrupt,
7799 
7800 	.set_tss_addr = vmx_set_tss_addr,
7801 	.set_identity_map_addr = vmx_set_identity_map_addr,
7802 	.get_tdp_level = get_ept_level,
7803 	.get_mt_mask = vmx_get_mt_mask,
7804 
7805 	.get_exit_info = vmx_get_exit_info,
7806 
7807 	.get_lpage_level = vmx_get_lpage_level,
7808 
7809 	.cpuid_update = vmx_cpuid_update,
7810 
7811 	.rdtscp_supported = vmx_rdtscp_supported,
7812 	.invpcid_supported = vmx_invpcid_supported,
7813 
7814 	.set_supported_cpuid = vmx_set_supported_cpuid,
7815 
7816 	.has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
7817 
7818 	.read_l1_tsc_offset = vmx_read_l1_tsc_offset,
7819 	.write_l1_tsc_offset = vmx_write_l1_tsc_offset,
7820 
7821 	.set_tdp_cr3 = vmx_set_cr3,
7822 
7823 	.check_intercept = vmx_check_intercept,
7824 	.handle_exit_irqoff = vmx_handle_exit_irqoff,
7825 	.mpx_supported = vmx_mpx_supported,
7826 	.xsaves_supported = vmx_xsaves_supported,
7827 	.umip_emulated = vmx_umip_emulated,
7828 	.pt_supported = vmx_pt_supported,
7829 	.pku_supported = vmx_pku_supported,
7830 
7831 	.request_immediate_exit = vmx_request_immediate_exit,
7832 
7833 	.sched_in = vmx_sched_in,
7834 
7835 	.slot_enable_log_dirty = vmx_slot_enable_log_dirty,
7836 	.slot_disable_log_dirty = vmx_slot_disable_log_dirty,
7837 	.flush_log_dirty = vmx_flush_log_dirty,
7838 	.enable_log_dirty_pt_masked = vmx_enable_log_dirty_pt_masked,
7839 	.write_log_dirty = vmx_write_pml_buffer,
7840 
7841 	.pre_block = vmx_pre_block,
7842 	.post_block = vmx_post_block,
7843 
7844 	.pmu_ops = &intel_pmu_ops,
7845 
7846 	.update_pi_irte = vmx_update_pi_irte,
7847 
7848 #ifdef CONFIG_X86_64
7849 	.set_hv_timer = vmx_set_hv_timer,
7850 	.cancel_hv_timer = vmx_cancel_hv_timer,
7851 #endif
7852 
7853 	.setup_mce = vmx_setup_mce,
7854 
7855 	.smi_allowed = vmx_smi_allowed,
7856 	.pre_enter_smm = vmx_pre_enter_smm,
7857 	.pre_leave_smm = vmx_pre_leave_smm,
7858 	.enable_smi_window = enable_smi_window,
7859 
7860 	.check_nested_events = NULL,
7861 	.get_nested_state = NULL,
7862 	.set_nested_state = NULL,
7863 	.get_vmcs12_pages = NULL,
7864 	.nested_enable_evmcs = NULL,
7865 	.nested_get_evmcs_version = NULL,
7866 	.need_emulation_on_page_fault = vmx_need_emulation_on_page_fault,
7867 	.apic_init_signal_blocked = vmx_apic_init_signal_blocked,
7868 };
7869 
7870 static void vmx_cleanup_l1d_flush(void)
7871 {
7872 	if (vmx_l1d_flush_pages) {
7873 		free_pages((unsigned long)vmx_l1d_flush_pages, L1D_CACHE_ORDER);
7874 		vmx_l1d_flush_pages = NULL;
7875 	}
7876 	/* Restore state so sysfs ignores VMX */
7877 	l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_AUTO;
7878 }
7879 
7880 static void vmx_exit(void)
7881 {
7882 #ifdef CONFIG_KEXEC_CORE
7883 	RCU_INIT_POINTER(crash_vmclear_loaded_vmcss, NULL);
7884 	synchronize_rcu();
7885 #endif
7886 
7887 	kvm_exit();
7888 
7889 #if IS_ENABLED(CONFIG_HYPERV)
7890 	if (static_branch_unlikely(&enable_evmcs)) {
7891 		int cpu;
7892 		struct hv_vp_assist_page *vp_ap;
7893 		/*
7894 		 * Reset everything to support using non-enlightened VMCS
7895 		 * access later (e.g. when we reload the module with
7896 		 * enlightened_vmcs=0)
7897 		 */
7898 		for_each_online_cpu(cpu) {
7899 			vp_ap =	hv_get_vp_assist_page(cpu);
7900 
7901 			if (!vp_ap)
7902 				continue;
7903 
7904 			vp_ap->nested_control.features.directhypercall = 0;
7905 			vp_ap->current_nested_vmcs = 0;
7906 			vp_ap->enlighten_vmentry = 0;
7907 		}
7908 
7909 		static_branch_disable(&enable_evmcs);
7910 	}
7911 #endif
7912 	vmx_cleanup_l1d_flush();
7913 }
7914 module_exit(vmx_exit);
7915 
7916 static int __init vmx_init(void)
7917 {
7918 	int r;
7919 
7920 #if IS_ENABLED(CONFIG_HYPERV)
7921 	/*
7922 	 * Enlightened VMCS usage should be recommended and the host needs
7923 	 * to support eVMCS v1 or above. We can also disable eVMCS support
7924 	 * with module parameter.
7925 	 */
7926 	if (enlightened_vmcs &&
7927 	    ms_hyperv.hints & HV_X64_ENLIGHTENED_VMCS_RECOMMENDED &&
7928 	    (ms_hyperv.nested_features & HV_X64_ENLIGHTENED_VMCS_VERSION) >=
7929 	    KVM_EVMCS_VERSION) {
7930 		int cpu;
7931 
7932 		/* Check that we have assist pages on all online CPUs */
7933 		for_each_online_cpu(cpu) {
7934 			if (!hv_get_vp_assist_page(cpu)) {
7935 				enlightened_vmcs = false;
7936 				break;
7937 			}
7938 		}
7939 
7940 		if (enlightened_vmcs) {
7941 			pr_info("KVM: vmx: using Hyper-V Enlightened VMCS\n");
7942 			static_branch_enable(&enable_evmcs);
7943 		}
7944 
7945 		if (ms_hyperv.nested_features & HV_X64_NESTED_DIRECT_FLUSH)
7946 			vmx_x86_ops.enable_direct_tlbflush
7947 				= hv_enable_direct_tlbflush;
7948 
7949 	} else {
7950 		enlightened_vmcs = false;
7951 	}
7952 #endif
7953 
7954 	r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx),
7955 		     __alignof__(struct vcpu_vmx), THIS_MODULE);
7956 	if (r)
7957 		return r;
7958 
7959 	/*
7960 	 * Must be called after kvm_init() so enable_ept is properly set
7961 	 * up. Hand the parameter mitigation value in which was stored in
7962 	 * the pre module init parser. If no parameter was given, it will
7963 	 * contain 'auto' which will be turned into the default 'cond'
7964 	 * mitigation mode.
7965 	 */
7966 	r = vmx_setup_l1d_flush(vmentry_l1d_flush_param);
7967 	if (r) {
7968 		vmx_exit();
7969 		return r;
7970 	}
7971 
7972 #ifdef CONFIG_KEXEC_CORE
7973 	rcu_assign_pointer(crash_vmclear_loaded_vmcss,
7974 			   crash_vmclear_local_loaded_vmcss);
7975 #endif
7976 	vmx_check_vmcs12_offsets();
7977 
7978 	return 0;
7979 }
7980 module_init(vmx_init);
7981