xref: /openbmc/linux/arch/x86/kvm/vmx/vmx.c (revision f8e17c17)
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 unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
1432 {
1433 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1434 	unsigned long rflags, save_rflags;
1435 
1436 	if (!kvm_register_is_available(vcpu, VCPU_EXREG_RFLAGS)) {
1437 		kvm_register_mark_available(vcpu, VCPU_EXREG_RFLAGS);
1438 		rflags = vmcs_readl(GUEST_RFLAGS);
1439 		if (vmx->rmode.vm86_active) {
1440 			rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
1441 			save_rflags = vmx->rmode.save_rflags;
1442 			rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
1443 		}
1444 		vmx->rflags = rflags;
1445 	}
1446 	return vmx->rflags;
1447 }
1448 
1449 void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
1450 {
1451 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1452 	unsigned long old_rflags;
1453 
1454 	if (enable_unrestricted_guest) {
1455 		kvm_register_mark_available(vcpu, VCPU_EXREG_RFLAGS);
1456 		vmx->rflags = rflags;
1457 		vmcs_writel(GUEST_RFLAGS, rflags);
1458 		return;
1459 	}
1460 
1461 	old_rflags = vmx_get_rflags(vcpu);
1462 	vmx->rflags = rflags;
1463 	if (vmx->rmode.vm86_active) {
1464 		vmx->rmode.save_rflags = rflags;
1465 		rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
1466 	}
1467 	vmcs_writel(GUEST_RFLAGS, rflags);
1468 
1469 	if ((old_rflags ^ vmx->rflags) & X86_EFLAGS_VM)
1470 		vmx->emulation_required = emulation_required(vcpu);
1471 }
1472 
1473 u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu)
1474 {
1475 	u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1476 	int ret = 0;
1477 
1478 	if (interruptibility & GUEST_INTR_STATE_STI)
1479 		ret |= KVM_X86_SHADOW_INT_STI;
1480 	if (interruptibility & GUEST_INTR_STATE_MOV_SS)
1481 		ret |= KVM_X86_SHADOW_INT_MOV_SS;
1482 
1483 	return ret;
1484 }
1485 
1486 void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
1487 {
1488 	u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1489 	u32 interruptibility = interruptibility_old;
1490 
1491 	interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
1492 
1493 	if (mask & KVM_X86_SHADOW_INT_MOV_SS)
1494 		interruptibility |= GUEST_INTR_STATE_MOV_SS;
1495 	else if (mask & KVM_X86_SHADOW_INT_STI)
1496 		interruptibility |= GUEST_INTR_STATE_STI;
1497 
1498 	if ((interruptibility != interruptibility_old))
1499 		vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
1500 }
1501 
1502 static int vmx_rtit_ctl_check(struct kvm_vcpu *vcpu, u64 data)
1503 {
1504 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1505 	unsigned long value;
1506 
1507 	/*
1508 	 * Any MSR write that attempts to change bits marked reserved will
1509 	 * case a #GP fault.
1510 	 */
1511 	if (data & vmx->pt_desc.ctl_bitmask)
1512 		return 1;
1513 
1514 	/*
1515 	 * Any attempt to modify IA32_RTIT_CTL while TraceEn is set will
1516 	 * result in a #GP unless the same write also clears TraceEn.
1517 	 */
1518 	if ((vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) &&
1519 		((vmx->pt_desc.guest.ctl ^ data) & ~RTIT_CTL_TRACEEN))
1520 		return 1;
1521 
1522 	/*
1523 	 * WRMSR to IA32_RTIT_CTL that sets TraceEn but clears this bit
1524 	 * and FabricEn would cause #GP, if
1525 	 * CPUID.(EAX=14H, ECX=0):ECX.SNGLRGNOUT[bit 2] = 0
1526 	 */
1527 	if ((data & RTIT_CTL_TRACEEN) && !(data & RTIT_CTL_TOPA) &&
1528 		!(data & RTIT_CTL_FABRIC_EN) &&
1529 		!intel_pt_validate_cap(vmx->pt_desc.caps,
1530 					PT_CAP_single_range_output))
1531 		return 1;
1532 
1533 	/*
1534 	 * MTCFreq, CycThresh and PSBFreq encodings check, any MSR write that
1535 	 * utilize encodings marked reserved will casue a #GP fault.
1536 	 */
1537 	value = intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc_periods);
1538 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc) &&
1539 			!test_bit((data & RTIT_CTL_MTC_RANGE) >>
1540 			RTIT_CTL_MTC_RANGE_OFFSET, &value))
1541 		return 1;
1542 	value = intel_pt_validate_cap(vmx->pt_desc.caps,
1543 						PT_CAP_cycle_thresholds);
1544 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc) &&
1545 			!test_bit((data & RTIT_CTL_CYC_THRESH) >>
1546 			RTIT_CTL_CYC_THRESH_OFFSET, &value))
1547 		return 1;
1548 	value = intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_periods);
1549 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc) &&
1550 			!test_bit((data & RTIT_CTL_PSB_FREQ) >>
1551 			RTIT_CTL_PSB_FREQ_OFFSET, &value))
1552 		return 1;
1553 
1554 	/*
1555 	 * If ADDRx_CFG is reserved or the encodings is >2 will
1556 	 * cause a #GP fault.
1557 	 */
1558 	value = (data & RTIT_CTL_ADDR0) >> RTIT_CTL_ADDR0_OFFSET;
1559 	if ((value && (vmx->pt_desc.addr_range < 1)) || (value > 2))
1560 		return 1;
1561 	value = (data & RTIT_CTL_ADDR1) >> RTIT_CTL_ADDR1_OFFSET;
1562 	if ((value && (vmx->pt_desc.addr_range < 2)) || (value > 2))
1563 		return 1;
1564 	value = (data & RTIT_CTL_ADDR2) >> RTIT_CTL_ADDR2_OFFSET;
1565 	if ((value && (vmx->pt_desc.addr_range < 3)) || (value > 2))
1566 		return 1;
1567 	value = (data & RTIT_CTL_ADDR3) >> RTIT_CTL_ADDR3_OFFSET;
1568 	if ((value && (vmx->pt_desc.addr_range < 4)) || (value > 2))
1569 		return 1;
1570 
1571 	return 0;
1572 }
1573 
1574 static int skip_emulated_instruction(struct kvm_vcpu *vcpu)
1575 {
1576 	unsigned long rip;
1577 
1578 	/*
1579 	 * Using VMCS.VM_EXIT_INSTRUCTION_LEN on EPT misconfig depends on
1580 	 * undefined behavior: Intel's SDM doesn't mandate the VMCS field be
1581 	 * set when EPT misconfig occurs.  In practice, real hardware updates
1582 	 * VM_EXIT_INSTRUCTION_LEN on EPT misconfig, but other hypervisors
1583 	 * (namely Hyper-V) don't set it due to it being undefined behavior,
1584 	 * i.e. we end up advancing IP with some random value.
1585 	 */
1586 	if (!static_cpu_has(X86_FEATURE_HYPERVISOR) ||
1587 	    to_vmx(vcpu)->exit_reason != EXIT_REASON_EPT_MISCONFIG) {
1588 		rip = kvm_rip_read(vcpu);
1589 		rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
1590 		kvm_rip_write(vcpu, rip);
1591 	} else {
1592 		if (!kvm_emulate_instruction(vcpu, EMULTYPE_SKIP))
1593 			return 0;
1594 	}
1595 
1596 	/* skipping an emulated instruction also counts */
1597 	vmx_set_interrupt_shadow(vcpu, 0);
1598 
1599 	return 1;
1600 }
1601 
1602 static void vmx_clear_hlt(struct kvm_vcpu *vcpu)
1603 {
1604 	/*
1605 	 * Ensure that we clear the HLT state in the VMCS.  We don't need to
1606 	 * explicitly skip the instruction because if the HLT state is set,
1607 	 * then the instruction is already executing and RIP has already been
1608 	 * advanced.
1609 	 */
1610 	if (kvm_hlt_in_guest(vcpu->kvm) &&
1611 			vmcs_read32(GUEST_ACTIVITY_STATE) == GUEST_ACTIVITY_HLT)
1612 		vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
1613 }
1614 
1615 static void vmx_queue_exception(struct kvm_vcpu *vcpu)
1616 {
1617 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1618 	unsigned nr = vcpu->arch.exception.nr;
1619 	bool has_error_code = vcpu->arch.exception.has_error_code;
1620 	u32 error_code = vcpu->arch.exception.error_code;
1621 	u32 intr_info = nr | INTR_INFO_VALID_MASK;
1622 
1623 	kvm_deliver_exception_payload(vcpu);
1624 
1625 	if (has_error_code) {
1626 		vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
1627 		intr_info |= INTR_INFO_DELIVER_CODE_MASK;
1628 	}
1629 
1630 	if (vmx->rmode.vm86_active) {
1631 		int inc_eip = 0;
1632 		if (kvm_exception_is_soft(nr))
1633 			inc_eip = vcpu->arch.event_exit_inst_len;
1634 		kvm_inject_realmode_interrupt(vcpu, nr, inc_eip);
1635 		return;
1636 	}
1637 
1638 	WARN_ON_ONCE(vmx->emulation_required);
1639 
1640 	if (kvm_exception_is_soft(nr)) {
1641 		vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
1642 			     vmx->vcpu.arch.event_exit_inst_len);
1643 		intr_info |= INTR_TYPE_SOFT_EXCEPTION;
1644 	} else
1645 		intr_info |= INTR_TYPE_HARD_EXCEPTION;
1646 
1647 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
1648 
1649 	vmx_clear_hlt(vcpu);
1650 }
1651 
1652 static bool vmx_rdtscp_supported(void)
1653 {
1654 	return cpu_has_vmx_rdtscp();
1655 }
1656 
1657 static bool vmx_invpcid_supported(void)
1658 {
1659 	return cpu_has_vmx_invpcid();
1660 }
1661 
1662 /*
1663  * Swap MSR entry in host/guest MSR entry array.
1664  */
1665 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
1666 {
1667 	struct shared_msr_entry tmp;
1668 
1669 	tmp = vmx->guest_msrs[to];
1670 	vmx->guest_msrs[to] = vmx->guest_msrs[from];
1671 	vmx->guest_msrs[from] = tmp;
1672 }
1673 
1674 /*
1675  * Set up the vmcs to automatically save and restore system
1676  * msrs.  Don't touch the 64-bit msrs if the guest is in legacy
1677  * mode, as fiddling with msrs is very expensive.
1678  */
1679 static void setup_msrs(struct vcpu_vmx *vmx)
1680 {
1681 	int save_nmsrs, index;
1682 
1683 	save_nmsrs = 0;
1684 #ifdef CONFIG_X86_64
1685 	/*
1686 	 * The SYSCALL MSRs are only needed on long mode guests, and only
1687 	 * when EFER.SCE is set.
1688 	 */
1689 	if (is_long_mode(&vmx->vcpu) && (vmx->vcpu.arch.efer & EFER_SCE)) {
1690 		index = __find_msr_index(vmx, MSR_STAR);
1691 		if (index >= 0)
1692 			move_msr_up(vmx, index, save_nmsrs++);
1693 		index = __find_msr_index(vmx, MSR_LSTAR);
1694 		if (index >= 0)
1695 			move_msr_up(vmx, index, save_nmsrs++);
1696 		index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
1697 		if (index >= 0)
1698 			move_msr_up(vmx, index, save_nmsrs++);
1699 	}
1700 #endif
1701 	index = __find_msr_index(vmx, MSR_EFER);
1702 	if (index >= 0 && update_transition_efer(vmx, index))
1703 		move_msr_up(vmx, index, save_nmsrs++);
1704 	index = __find_msr_index(vmx, MSR_TSC_AUX);
1705 	if (index >= 0 && guest_cpuid_has(&vmx->vcpu, X86_FEATURE_RDTSCP))
1706 		move_msr_up(vmx, index, save_nmsrs++);
1707 	index = __find_msr_index(vmx, MSR_IA32_TSX_CTRL);
1708 	if (index >= 0)
1709 		move_msr_up(vmx, index, save_nmsrs++);
1710 
1711 	vmx->save_nmsrs = save_nmsrs;
1712 	vmx->guest_msrs_ready = false;
1713 
1714 	if (cpu_has_vmx_msr_bitmap())
1715 		vmx_update_msr_bitmap(&vmx->vcpu);
1716 }
1717 
1718 static u64 vmx_read_l1_tsc_offset(struct kvm_vcpu *vcpu)
1719 {
1720 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1721 
1722 	if (is_guest_mode(vcpu) &&
1723 	    (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETTING))
1724 		return vcpu->arch.tsc_offset - vmcs12->tsc_offset;
1725 
1726 	return vcpu->arch.tsc_offset;
1727 }
1728 
1729 static u64 vmx_write_l1_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1730 {
1731 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1732 	u64 g_tsc_offset = 0;
1733 
1734 	/*
1735 	 * We're here if L1 chose not to trap WRMSR to TSC. According
1736 	 * to the spec, this should set L1's TSC; The offset that L1
1737 	 * set for L2 remains unchanged, and still needs to be added
1738 	 * to the newly set TSC to get L2's TSC.
1739 	 */
1740 	if (is_guest_mode(vcpu) &&
1741 	    (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETTING))
1742 		g_tsc_offset = vmcs12->tsc_offset;
1743 
1744 	trace_kvm_write_tsc_offset(vcpu->vcpu_id,
1745 				   vcpu->arch.tsc_offset - g_tsc_offset,
1746 				   offset);
1747 	vmcs_write64(TSC_OFFSET, offset + g_tsc_offset);
1748 	return offset + g_tsc_offset;
1749 }
1750 
1751 /*
1752  * nested_vmx_allowed() checks whether a guest should be allowed to use VMX
1753  * instructions and MSRs (i.e., nested VMX). Nested VMX is disabled for
1754  * all guests if the "nested" module option is off, and can also be disabled
1755  * for a single guest by disabling its VMX cpuid bit.
1756  */
1757 bool nested_vmx_allowed(struct kvm_vcpu *vcpu)
1758 {
1759 	return nested && guest_cpuid_has(vcpu, X86_FEATURE_VMX);
1760 }
1761 
1762 static inline bool vmx_feature_control_msr_valid(struct kvm_vcpu *vcpu,
1763 						 uint64_t val)
1764 {
1765 	uint64_t valid_bits = to_vmx(vcpu)->msr_ia32_feature_control_valid_bits;
1766 
1767 	return !(val & ~valid_bits);
1768 }
1769 
1770 static int vmx_get_msr_feature(struct kvm_msr_entry *msr)
1771 {
1772 	switch (msr->index) {
1773 	case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
1774 		if (!nested)
1775 			return 1;
1776 		return vmx_get_vmx_msr(&vmcs_config.nested, msr->index, &msr->data);
1777 	default:
1778 		return 1;
1779 	}
1780 }
1781 
1782 /*
1783  * Reads an msr value (of 'msr_index') into 'pdata'.
1784  * Returns 0 on success, non-0 otherwise.
1785  * Assumes vcpu_load() was already called.
1786  */
1787 static int vmx_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1788 {
1789 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1790 	struct shared_msr_entry *msr;
1791 	u32 index;
1792 
1793 	switch (msr_info->index) {
1794 #ifdef CONFIG_X86_64
1795 	case MSR_FS_BASE:
1796 		msr_info->data = vmcs_readl(GUEST_FS_BASE);
1797 		break;
1798 	case MSR_GS_BASE:
1799 		msr_info->data = vmcs_readl(GUEST_GS_BASE);
1800 		break;
1801 	case MSR_KERNEL_GS_BASE:
1802 		msr_info->data = vmx_read_guest_kernel_gs_base(vmx);
1803 		break;
1804 #endif
1805 	case MSR_EFER:
1806 		return kvm_get_msr_common(vcpu, msr_info);
1807 	case MSR_IA32_TSX_CTRL:
1808 		if (!msr_info->host_initiated &&
1809 		    !(vcpu->arch.arch_capabilities & ARCH_CAP_TSX_CTRL_MSR))
1810 			return 1;
1811 		goto find_shared_msr;
1812 	case MSR_IA32_UMWAIT_CONTROL:
1813 		if (!msr_info->host_initiated && !vmx_has_waitpkg(vmx))
1814 			return 1;
1815 
1816 		msr_info->data = vmx->msr_ia32_umwait_control;
1817 		break;
1818 	case MSR_IA32_SPEC_CTRL:
1819 		if (!msr_info->host_initiated &&
1820 		    !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
1821 			return 1;
1822 
1823 		msr_info->data = to_vmx(vcpu)->spec_ctrl;
1824 		break;
1825 	case MSR_IA32_SYSENTER_CS:
1826 		msr_info->data = vmcs_read32(GUEST_SYSENTER_CS);
1827 		break;
1828 	case MSR_IA32_SYSENTER_EIP:
1829 		msr_info->data = vmcs_readl(GUEST_SYSENTER_EIP);
1830 		break;
1831 	case MSR_IA32_SYSENTER_ESP:
1832 		msr_info->data = vmcs_readl(GUEST_SYSENTER_ESP);
1833 		break;
1834 	case MSR_IA32_BNDCFGS:
1835 		if (!kvm_mpx_supported() ||
1836 		    (!msr_info->host_initiated &&
1837 		     !guest_cpuid_has(vcpu, X86_FEATURE_MPX)))
1838 			return 1;
1839 		msr_info->data = vmcs_read64(GUEST_BNDCFGS);
1840 		break;
1841 	case MSR_IA32_MCG_EXT_CTL:
1842 		if (!msr_info->host_initiated &&
1843 		    !(vmx->msr_ia32_feature_control &
1844 		      FEAT_CTL_LMCE_ENABLED))
1845 			return 1;
1846 		msr_info->data = vcpu->arch.mcg_ext_ctl;
1847 		break;
1848 	case MSR_IA32_FEAT_CTL:
1849 		msr_info->data = vmx->msr_ia32_feature_control;
1850 		break;
1851 	case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
1852 		if (!nested_vmx_allowed(vcpu))
1853 			return 1;
1854 		if (vmx_get_vmx_msr(&vmx->nested.msrs, msr_info->index,
1855 				    &msr_info->data))
1856 			return 1;
1857 		/*
1858 		 * Enlightened VMCS v1 doesn't have certain fields, but buggy
1859 		 * Hyper-V versions are still trying to use corresponding
1860 		 * features when they are exposed. Filter out the essential
1861 		 * minimum.
1862 		 */
1863 		if (!msr_info->host_initiated &&
1864 		    vmx->nested.enlightened_vmcs_enabled)
1865 			nested_evmcs_filter_control_msr(msr_info->index,
1866 							&msr_info->data);
1867 		break;
1868 	case MSR_IA32_RTIT_CTL:
1869 		if (pt_mode != PT_MODE_HOST_GUEST)
1870 			return 1;
1871 		msr_info->data = vmx->pt_desc.guest.ctl;
1872 		break;
1873 	case MSR_IA32_RTIT_STATUS:
1874 		if (pt_mode != PT_MODE_HOST_GUEST)
1875 			return 1;
1876 		msr_info->data = vmx->pt_desc.guest.status;
1877 		break;
1878 	case MSR_IA32_RTIT_CR3_MATCH:
1879 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
1880 			!intel_pt_validate_cap(vmx->pt_desc.caps,
1881 						PT_CAP_cr3_filtering))
1882 			return 1;
1883 		msr_info->data = vmx->pt_desc.guest.cr3_match;
1884 		break;
1885 	case MSR_IA32_RTIT_OUTPUT_BASE:
1886 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
1887 			(!intel_pt_validate_cap(vmx->pt_desc.caps,
1888 					PT_CAP_topa_output) &&
1889 			 !intel_pt_validate_cap(vmx->pt_desc.caps,
1890 					PT_CAP_single_range_output)))
1891 			return 1;
1892 		msr_info->data = vmx->pt_desc.guest.output_base;
1893 		break;
1894 	case MSR_IA32_RTIT_OUTPUT_MASK:
1895 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
1896 			(!intel_pt_validate_cap(vmx->pt_desc.caps,
1897 					PT_CAP_topa_output) &&
1898 			 !intel_pt_validate_cap(vmx->pt_desc.caps,
1899 					PT_CAP_single_range_output)))
1900 			return 1;
1901 		msr_info->data = vmx->pt_desc.guest.output_mask;
1902 		break;
1903 	case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
1904 		index = msr_info->index - MSR_IA32_RTIT_ADDR0_A;
1905 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
1906 			(index >= 2 * intel_pt_validate_cap(vmx->pt_desc.caps,
1907 					PT_CAP_num_address_ranges)))
1908 			return 1;
1909 		if (index % 2)
1910 			msr_info->data = vmx->pt_desc.guest.addr_b[index / 2];
1911 		else
1912 			msr_info->data = vmx->pt_desc.guest.addr_a[index / 2];
1913 		break;
1914 	case MSR_TSC_AUX:
1915 		if (!msr_info->host_initiated &&
1916 		    !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP))
1917 			return 1;
1918 		goto find_shared_msr;
1919 	default:
1920 	find_shared_msr:
1921 		msr = find_msr_entry(vmx, msr_info->index);
1922 		if (msr) {
1923 			msr_info->data = msr->data;
1924 			break;
1925 		}
1926 		return kvm_get_msr_common(vcpu, msr_info);
1927 	}
1928 
1929 	return 0;
1930 }
1931 
1932 /*
1933  * Writes msr value into the appropriate "register".
1934  * Returns 0 on success, non-0 otherwise.
1935  * Assumes vcpu_load() was already called.
1936  */
1937 static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1938 {
1939 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1940 	struct shared_msr_entry *msr;
1941 	int ret = 0;
1942 	u32 msr_index = msr_info->index;
1943 	u64 data = msr_info->data;
1944 	u32 index;
1945 
1946 	switch (msr_index) {
1947 	case MSR_EFER:
1948 		ret = kvm_set_msr_common(vcpu, msr_info);
1949 		break;
1950 #ifdef CONFIG_X86_64
1951 	case MSR_FS_BASE:
1952 		vmx_segment_cache_clear(vmx);
1953 		vmcs_writel(GUEST_FS_BASE, data);
1954 		break;
1955 	case MSR_GS_BASE:
1956 		vmx_segment_cache_clear(vmx);
1957 		vmcs_writel(GUEST_GS_BASE, data);
1958 		break;
1959 	case MSR_KERNEL_GS_BASE:
1960 		vmx_write_guest_kernel_gs_base(vmx, data);
1961 		break;
1962 #endif
1963 	case MSR_IA32_SYSENTER_CS:
1964 		if (is_guest_mode(vcpu))
1965 			get_vmcs12(vcpu)->guest_sysenter_cs = data;
1966 		vmcs_write32(GUEST_SYSENTER_CS, data);
1967 		break;
1968 	case MSR_IA32_SYSENTER_EIP:
1969 		if (is_guest_mode(vcpu))
1970 			get_vmcs12(vcpu)->guest_sysenter_eip = data;
1971 		vmcs_writel(GUEST_SYSENTER_EIP, data);
1972 		break;
1973 	case MSR_IA32_SYSENTER_ESP:
1974 		if (is_guest_mode(vcpu))
1975 			get_vmcs12(vcpu)->guest_sysenter_esp = data;
1976 		vmcs_writel(GUEST_SYSENTER_ESP, data);
1977 		break;
1978 	case MSR_IA32_DEBUGCTLMSR:
1979 		if (is_guest_mode(vcpu) && get_vmcs12(vcpu)->vm_exit_controls &
1980 						VM_EXIT_SAVE_DEBUG_CONTROLS)
1981 			get_vmcs12(vcpu)->guest_ia32_debugctl = data;
1982 
1983 		ret = kvm_set_msr_common(vcpu, msr_info);
1984 		break;
1985 
1986 	case MSR_IA32_BNDCFGS:
1987 		if (!kvm_mpx_supported() ||
1988 		    (!msr_info->host_initiated &&
1989 		     !guest_cpuid_has(vcpu, X86_FEATURE_MPX)))
1990 			return 1;
1991 		if (is_noncanonical_address(data & PAGE_MASK, vcpu) ||
1992 		    (data & MSR_IA32_BNDCFGS_RSVD))
1993 			return 1;
1994 		vmcs_write64(GUEST_BNDCFGS, data);
1995 		break;
1996 	case MSR_IA32_UMWAIT_CONTROL:
1997 		if (!msr_info->host_initiated && !vmx_has_waitpkg(vmx))
1998 			return 1;
1999 
2000 		/* The reserved bit 1 and non-32 bit [63:32] should be zero */
2001 		if (data & (BIT_ULL(1) | GENMASK_ULL(63, 32)))
2002 			return 1;
2003 
2004 		vmx->msr_ia32_umwait_control = data;
2005 		break;
2006 	case MSR_IA32_SPEC_CTRL:
2007 		if (!msr_info->host_initiated &&
2008 		    !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
2009 			return 1;
2010 
2011 		if (data & ~kvm_spec_ctrl_valid_bits(vcpu))
2012 			return 1;
2013 
2014 		vmx->spec_ctrl = data;
2015 		if (!data)
2016 			break;
2017 
2018 		/*
2019 		 * For non-nested:
2020 		 * When it's written (to non-zero) for the first time, pass
2021 		 * it through.
2022 		 *
2023 		 * For nested:
2024 		 * The handling of the MSR bitmap for L2 guests is done in
2025 		 * nested_vmx_prepare_msr_bitmap. We should not touch the
2026 		 * vmcs02.msr_bitmap here since it gets completely overwritten
2027 		 * in the merging. We update the vmcs01 here for L1 as well
2028 		 * since it will end up touching the MSR anyway now.
2029 		 */
2030 		vmx_disable_intercept_for_msr(vmx->vmcs01.msr_bitmap,
2031 					      MSR_IA32_SPEC_CTRL,
2032 					      MSR_TYPE_RW);
2033 		break;
2034 	case MSR_IA32_TSX_CTRL:
2035 		if (!msr_info->host_initiated &&
2036 		    !(vcpu->arch.arch_capabilities & ARCH_CAP_TSX_CTRL_MSR))
2037 			return 1;
2038 		if (data & ~(TSX_CTRL_RTM_DISABLE | TSX_CTRL_CPUID_CLEAR))
2039 			return 1;
2040 		goto find_shared_msr;
2041 	case MSR_IA32_PRED_CMD:
2042 		if (!msr_info->host_initiated &&
2043 		    !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
2044 			return 1;
2045 
2046 		if (data & ~PRED_CMD_IBPB)
2047 			return 1;
2048 		if (!boot_cpu_has(X86_FEATURE_SPEC_CTRL))
2049 			return 1;
2050 		if (!data)
2051 			break;
2052 
2053 		wrmsrl(MSR_IA32_PRED_CMD, PRED_CMD_IBPB);
2054 
2055 		/*
2056 		 * For non-nested:
2057 		 * When it's written (to non-zero) for the first time, pass
2058 		 * it through.
2059 		 *
2060 		 * For nested:
2061 		 * The handling of the MSR bitmap for L2 guests is done in
2062 		 * nested_vmx_prepare_msr_bitmap. We should not touch the
2063 		 * vmcs02.msr_bitmap here since it gets completely overwritten
2064 		 * in the merging.
2065 		 */
2066 		vmx_disable_intercept_for_msr(vmx->vmcs01.msr_bitmap, MSR_IA32_PRED_CMD,
2067 					      MSR_TYPE_W);
2068 		break;
2069 	case MSR_IA32_CR_PAT:
2070 		if (!kvm_pat_valid(data))
2071 			return 1;
2072 
2073 		if (is_guest_mode(vcpu) &&
2074 		    get_vmcs12(vcpu)->vm_exit_controls & VM_EXIT_SAVE_IA32_PAT)
2075 			get_vmcs12(vcpu)->guest_ia32_pat = data;
2076 
2077 		if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2078 			vmcs_write64(GUEST_IA32_PAT, data);
2079 			vcpu->arch.pat = data;
2080 			break;
2081 		}
2082 		ret = kvm_set_msr_common(vcpu, msr_info);
2083 		break;
2084 	case MSR_IA32_TSC_ADJUST:
2085 		ret = kvm_set_msr_common(vcpu, msr_info);
2086 		break;
2087 	case MSR_IA32_MCG_EXT_CTL:
2088 		if ((!msr_info->host_initiated &&
2089 		     !(to_vmx(vcpu)->msr_ia32_feature_control &
2090 		       FEAT_CTL_LMCE_ENABLED)) ||
2091 		    (data & ~MCG_EXT_CTL_LMCE_EN))
2092 			return 1;
2093 		vcpu->arch.mcg_ext_ctl = data;
2094 		break;
2095 	case MSR_IA32_FEAT_CTL:
2096 		if (!vmx_feature_control_msr_valid(vcpu, data) ||
2097 		    (to_vmx(vcpu)->msr_ia32_feature_control &
2098 		     FEAT_CTL_LOCKED && !msr_info->host_initiated))
2099 			return 1;
2100 		vmx->msr_ia32_feature_control = data;
2101 		if (msr_info->host_initiated && data == 0)
2102 			vmx_leave_nested(vcpu);
2103 		break;
2104 	case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
2105 		if (!msr_info->host_initiated)
2106 			return 1; /* they are read-only */
2107 		if (!nested_vmx_allowed(vcpu))
2108 			return 1;
2109 		return vmx_set_vmx_msr(vcpu, msr_index, data);
2110 	case MSR_IA32_RTIT_CTL:
2111 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
2112 			vmx_rtit_ctl_check(vcpu, data) ||
2113 			vmx->nested.vmxon)
2114 			return 1;
2115 		vmcs_write64(GUEST_IA32_RTIT_CTL, data);
2116 		vmx->pt_desc.guest.ctl = data;
2117 		pt_update_intercept_for_msr(vmx);
2118 		break;
2119 	case MSR_IA32_RTIT_STATUS:
2120 		if (!pt_can_write_msr(vmx))
2121 			return 1;
2122 		if (data & MSR_IA32_RTIT_STATUS_MASK)
2123 			return 1;
2124 		vmx->pt_desc.guest.status = data;
2125 		break;
2126 	case MSR_IA32_RTIT_CR3_MATCH:
2127 		if (!pt_can_write_msr(vmx))
2128 			return 1;
2129 		if (!intel_pt_validate_cap(vmx->pt_desc.caps,
2130 					   PT_CAP_cr3_filtering))
2131 			return 1;
2132 		vmx->pt_desc.guest.cr3_match = data;
2133 		break;
2134 	case MSR_IA32_RTIT_OUTPUT_BASE:
2135 		if (!pt_can_write_msr(vmx))
2136 			return 1;
2137 		if (!intel_pt_validate_cap(vmx->pt_desc.caps,
2138 					   PT_CAP_topa_output) &&
2139 		    !intel_pt_validate_cap(vmx->pt_desc.caps,
2140 					   PT_CAP_single_range_output))
2141 			return 1;
2142 		if (data & MSR_IA32_RTIT_OUTPUT_BASE_MASK)
2143 			return 1;
2144 		vmx->pt_desc.guest.output_base = data;
2145 		break;
2146 	case MSR_IA32_RTIT_OUTPUT_MASK:
2147 		if (!pt_can_write_msr(vmx))
2148 			return 1;
2149 		if (!intel_pt_validate_cap(vmx->pt_desc.caps,
2150 					   PT_CAP_topa_output) &&
2151 		    !intel_pt_validate_cap(vmx->pt_desc.caps,
2152 					   PT_CAP_single_range_output))
2153 			return 1;
2154 		vmx->pt_desc.guest.output_mask = data;
2155 		break;
2156 	case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
2157 		if (!pt_can_write_msr(vmx))
2158 			return 1;
2159 		index = msr_info->index - MSR_IA32_RTIT_ADDR0_A;
2160 		if (index >= 2 * intel_pt_validate_cap(vmx->pt_desc.caps,
2161 						       PT_CAP_num_address_ranges))
2162 			return 1;
2163 		if (is_noncanonical_address(data, vcpu))
2164 			return 1;
2165 		if (index % 2)
2166 			vmx->pt_desc.guest.addr_b[index / 2] = data;
2167 		else
2168 			vmx->pt_desc.guest.addr_a[index / 2] = data;
2169 		break;
2170 	case MSR_TSC_AUX:
2171 		if (!msr_info->host_initiated &&
2172 		    !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP))
2173 			return 1;
2174 		/* Check reserved bit, higher 32 bits should be zero */
2175 		if ((data >> 32) != 0)
2176 			return 1;
2177 		goto find_shared_msr;
2178 
2179 	default:
2180 	find_shared_msr:
2181 		msr = find_msr_entry(vmx, msr_index);
2182 		if (msr)
2183 			ret = vmx_set_guest_msr(vmx, msr, data);
2184 		else
2185 			ret = kvm_set_msr_common(vcpu, msr_info);
2186 	}
2187 
2188 	return ret;
2189 }
2190 
2191 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
2192 {
2193 	kvm_register_mark_available(vcpu, reg);
2194 
2195 	switch (reg) {
2196 	case VCPU_REGS_RSP:
2197 		vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
2198 		break;
2199 	case VCPU_REGS_RIP:
2200 		vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
2201 		break;
2202 	case VCPU_EXREG_PDPTR:
2203 		if (enable_ept)
2204 			ept_save_pdptrs(vcpu);
2205 		break;
2206 	case VCPU_EXREG_CR3:
2207 		if (enable_unrestricted_guest || (enable_ept && is_paging(vcpu)))
2208 			vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
2209 		break;
2210 	default:
2211 		WARN_ON_ONCE(1);
2212 		break;
2213 	}
2214 }
2215 
2216 static __init int cpu_has_kvm_support(void)
2217 {
2218 	return cpu_has_vmx();
2219 }
2220 
2221 static __init int vmx_disabled_by_bios(void)
2222 {
2223 	return !boot_cpu_has(X86_FEATURE_MSR_IA32_FEAT_CTL) ||
2224 	       !boot_cpu_has(X86_FEATURE_VMX);
2225 }
2226 
2227 static void kvm_cpu_vmxon(u64 addr)
2228 {
2229 	cr4_set_bits(X86_CR4_VMXE);
2230 	intel_pt_handle_vmx(1);
2231 
2232 	asm volatile ("vmxon %0" : : "m"(addr));
2233 }
2234 
2235 static int hardware_enable(void)
2236 {
2237 	int cpu = raw_smp_processor_id();
2238 	u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
2239 
2240 	if (cr4_read_shadow() & X86_CR4_VMXE)
2241 		return -EBUSY;
2242 
2243 	/*
2244 	 * This can happen if we hot-added a CPU but failed to allocate
2245 	 * VP assist page for it.
2246 	 */
2247 	if (static_branch_unlikely(&enable_evmcs) &&
2248 	    !hv_get_vp_assist_page(cpu))
2249 		return -EFAULT;
2250 
2251 	INIT_LIST_HEAD(&per_cpu(loaded_vmcss_on_cpu, cpu));
2252 	INIT_LIST_HEAD(&per_cpu(blocked_vcpu_on_cpu, cpu));
2253 	spin_lock_init(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
2254 
2255 	/*
2256 	 * Now we can enable the vmclear operation in kdump
2257 	 * since the loaded_vmcss_on_cpu list on this cpu
2258 	 * has been initialized.
2259 	 *
2260 	 * Though the cpu is not in VMX operation now, there
2261 	 * is no problem to enable the vmclear operation
2262 	 * for the loaded_vmcss_on_cpu list is empty!
2263 	 */
2264 	crash_enable_local_vmclear(cpu);
2265 
2266 	kvm_cpu_vmxon(phys_addr);
2267 	if (enable_ept)
2268 		ept_sync_global();
2269 
2270 	return 0;
2271 }
2272 
2273 static void vmclear_local_loaded_vmcss(void)
2274 {
2275 	int cpu = raw_smp_processor_id();
2276 	struct loaded_vmcs *v, *n;
2277 
2278 	list_for_each_entry_safe(v, n, &per_cpu(loaded_vmcss_on_cpu, cpu),
2279 				 loaded_vmcss_on_cpu_link)
2280 		__loaded_vmcs_clear(v);
2281 }
2282 
2283 
2284 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
2285  * tricks.
2286  */
2287 static void kvm_cpu_vmxoff(void)
2288 {
2289 	asm volatile (__ex("vmxoff"));
2290 
2291 	intel_pt_handle_vmx(0);
2292 	cr4_clear_bits(X86_CR4_VMXE);
2293 }
2294 
2295 static void hardware_disable(void)
2296 {
2297 	vmclear_local_loaded_vmcss();
2298 	kvm_cpu_vmxoff();
2299 }
2300 
2301 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
2302 				      u32 msr, u32 *result)
2303 {
2304 	u32 vmx_msr_low, vmx_msr_high;
2305 	u32 ctl = ctl_min | ctl_opt;
2306 
2307 	rdmsr(msr, vmx_msr_low, vmx_msr_high);
2308 
2309 	ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
2310 	ctl |= vmx_msr_low;  /* bit == 1 in low word  ==> must be one  */
2311 
2312 	/* Ensure minimum (required) set of control bits are supported. */
2313 	if (ctl_min & ~ctl)
2314 		return -EIO;
2315 
2316 	*result = ctl;
2317 	return 0;
2318 }
2319 
2320 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf,
2321 				    struct vmx_capability *vmx_cap)
2322 {
2323 	u32 vmx_msr_low, vmx_msr_high;
2324 	u32 min, opt, min2, opt2;
2325 	u32 _pin_based_exec_control = 0;
2326 	u32 _cpu_based_exec_control = 0;
2327 	u32 _cpu_based_2nd_exec_control = 0;
2328 	u32 _vmexit_control = 0;
2329 	u32 _vmentry_control = 0;
2330 
2331 	memset(vmcs_conf, 0, sizeof(*vmcs_conf));
2332 	min = CPU_BASED_HLT_EXITING |
2333 #ifdef CONFIG_X86_64
2334 	      CPU_BASED_CR8_LOAD_EXITING |
2335 	      CPU_BASED_CR8_STORE_EXITING |
2336 #endif
2337 	      CPU_BASED_CR3_LOAD_EXITING |
2338 	      CPU_BASED_CR3_STORE_EXITING |
2339 	      CPU_BASED_UNCOND_IO_EXITING |
2340 	      CPU_BASED_MOV_DR_EXITING |
2341 	      CPU_BASED_USE_TSC_OFFSETTING |
2342 	      CPU_BASED_MWAIT_EXITING |
2343 	      CPU_BASED_MONITOR_EXITING |
2344 	      CPU_BASED_INVLPG_EXITING |
2345 	      CPU_BASED_RDPMC_EXITING;
2346 
2347 	opt = CPU_BASED_TPR_SHADOW |
2348 	      CPU_BASED_USE_MSR_BITMAPS |
2349 	      CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
2350 	if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
2351 				&_cpu_based_exec_control) < 0)
2352 		return -EIO;
2353 #ifdef CONFIG_X86_64
2354 	if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2355 		_cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
2356 					   ~CPU_BASED_CR8_STORE_EXITING;
2357 #endif
2358 	if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
2359 		min2 = 0;
2360 		opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2361 			SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2362 			SECONDARY_EXEC_WBINVD_EXITING |
2363 			SECONDARY_EXEC_ENABLE_VPID |
2364 			SECONDARY_EXEC_ENABLE_EPT |
2365 			SECONDARY_EXEC_UNRESTRICTED_GUEST |
2366 			SECONDARY_EXEC_PAUSE_LOOP_EXITING |
2367 			SECONDARY_EXEC_DESC |
2368 			SECONDARY_EXEC_RDTSCP |
2369 			SECONDARY_EXEC_ENABLE_INVPCID |
2370 			SECONDARY_EXEC_APIC_REGISTER_VIRT |
2371 			SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
2372 			SECONDARY_EXEC_SHADOW_VMCS |
2373 			SECONDARY_EXEC_XSAVES |
2374 			SECONDARY_EXEC_RDSEED_EXITING |
2375 			SECONDARY_EXEC_RDRAND_EXITING |
2376 			SECONDARY_EXEC_ENABLE_PML |
2377 			SECONDARY_EXEC_TSC_SCALING |
2378 			SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE |
2379 			SECONDARY_EXEC_PT_USE_GPA |
2380 			SECONDARY_EXEC_PT_CONCEAL_VMX |
2381 			SECONDARY_EXEC_ENABLE_VMFUNC |
2382 			SECONDARY_EXEC_ENCLS_EXITING;
2383 		if (adjust_vmx_controls(min2, opt2,
2384 					MSR_IA32_VMX_PROCBASED_CTLS2,
2385 					&_cpu_based_2nd_exec_control) < 0)
2386 			return -EIO;
2387 	}
2388 #ifndef CONFIG_X86_64
2389 	if (!(_cpu_based_2nd_exec_control &
2390 				SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
2391 		_cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
2392 #endif
2393 
2394 	if (!(_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2395 		_cpu_based_2nd_exec_control &= ~(
2396 				SECONDARY_EXEC_APIC_REGISTER_VIRT |
2397 				SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2398 				SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
2399 
2400 	rdmsr_safe(MSR_IA32_VMX_EPT_VPID_CAP,
2401 		&vmx_cap->ept, &vmx_cap->vpid);
2402 
2403 	if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
2404 		/* CR3 accesses and invlpg don't need to cause VM Exits when EPT
2405 		   enabled */
2406 		_cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
2407 					     CPU_BASED_CR3_STORE_EXITING |
2408 					     CPU_BASED_INVLPG_EXITING);
2409 	} else if (vmx_cap->ept) {
2410 		vmx_cap->ept = 0;
2411 		pr_warn_once("EPT CAP should not exist if not support "
2412 				"1-setting enable EPT VM-execution control\n");
2413 	}
2414 	if (!(_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_VPID) &&
2415 		vmx_cap->vpid) {
2416 		vmx_cap->vpid = 0;
2417 		pr_warn_once("VPID CAP should not exist if not support "
2418 				"1-setting enable VPID VM-execution control\n");
2419 	}
2420 
2421 	min = VM_EXIT_SAVE_DEBUG_CONTROLS | VM_EXIT_ACK_INTR_ON_EXIT;
2422 #ifdef CONFIG_X86_64
2423 	min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
2424 #endif
2425 	opt = VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL |
2426 	      VM_EXIT_LOAD_IA32_PAT |
2427 	      VM_EXIT_LOAD_IA32_EFER |
2428 	      VM_EXIT_CLEAR_BNDCFGS |
2429 	      VM_EXIT_PT_CONCEAL_PIP |
2430 	      VM_EXIT_CLEAR_IA32_RTIT_CTL;
2431 	if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
2432 				&_vmexit_control) < 0)
2433 		return -EIO;
2434 
2435 	min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
2436 	opt = PIN_BASED_VIRTUAL_NMIS | PIN_BASED_POSTED_INTR |
2437 		 PIN_BASED_VMX_PREEMPTION_TIMER;
2438 	if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
2439 				&_pin_based_exec_control) < 0)
2440 		return -EIO;
2441 
2442 	if (cpu_has_broken_vmx_preemption_timer())
2443 		_pin_based_exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
2444 	if (!(_cpu_based_2nd_exec_control &
2445 		SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY))
2446 		_pin_based_exec_control &= ~PIN_BASED_POSTED_INTR;
2447 
2448 	min = VM_ENTRY_LOAD_DEBUG_CONTROLS;
2449 	opt = VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL |
2450 	      VM_ENTRY_LOAD_IA32_PAT |
2451 	      VM_ENTRY_LOAD_IA32_EFER |
2452 	      VM_ENTRY_LOAD_BNDCFGS |
2453 	      VM_ENTRY_PT_CONCEAL_PIP |
2454 	      VM_ENTRY_LOAD_IA32_RTIT_CTL;
2455 	if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
2456 				&_vmentry_control) < 0)
2457 		return -EIO;
2458 
2459 	/*
2460 	 * Some cpus support VM_{ENTRY,EXIT}_IA32_PERF_GLOBAL_CTRL but they
2461 	 * can't be used due to an errata where VM Exit may incorrectly clear
2462 	 * IA32_PERF_GLOBAL_CTRL[34:32].  Workaround the errata by using the
2463 	 * MSR load mechanism to switch IA32_PERF_GLOBAL_CTRL.
2464 	 */
2465 	if (boot_cpu_data.x86 == 0x6) {
2466 		switch (boot_cpu_data.x86_model) {
2467 		case 26: /* AAK155 */
2468 		case 30: /* AAP115 */
2469 		case 37: /* AAT100 */
2470 		case 44: /* BC86,AAY89,BD102 */
2471 		case 46: /* BA97 */
2472 			_vmentry_control &= ~VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL;
2473 			_vmexit_control &= ~VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL;
2474 			pr_warn_once("kvm: VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL "
2475 					"does not work properly. Using workaround\n");
2476 			break;
2477 		default:
2478 			break;
2479 		}
2480 	}
2481 
2482 
2483 	rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
2484 
2485 	/* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
2486 	if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
2487 		return -EIO;
2488 
2489 #ifdef CONFIG_X86_64
2490 	/* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
2491 	if (vmx_msr_high & (1u<<16))
2492 		return -EIO;
2493 #endif
2494 
2495 	/* Require Write-Back (WB) memory type for VMCS accesses. */
2496 	if (((vmx_msr_high >> 18) & 15) != 6)
2497 		return -EIO;
2498 
2499 	vmcs_conf->size = vmx_msr_high & 0x1fff;
2500 	vmcs_conf->order = get_order(vmcs_conf->size);
2501 	vmcs_conf->basic_cap = vmx_msr_high & ~0x1fff;
2502 
2503 	vmcs_conf->revision_id = vmx_msr_low;
2504 
2505 	vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
2506 	vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
2507 	vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
2508 	vmcs_conf->vmexit_ctrl         = _vmexit_control;
2509 	vmcs_conf->vmentry_ctrl        = _vmentry_control;
2510 
2511 	if (static_branch_unlikely(&enable_evmcs))
2512 		evmcs_sanitize_exec_ctrls(vmcs_conf);
2513 
2514 	return 0;
2515 }
2516 
2517 struct vmcs *alloc_vmcs_cpu(bool shadow, int cpu, gfp_t flags)
2518 {
2519 	int node = cpu_to_node(cpu);
2520 	struct page *pages;
2521 	struct vmcs *vmcs;
2522 
2523 	pages = __alloc_pages_node(node, flags, vmcs_config.order);
2524 	if (!pages)
2525 		return NULL;
2526 	vmcs = page_address(pages);
2527 	memset(vmcs, 0, vmcs_config.size);
2528 
2529 	/* KVM supports Enlightened VMCS v1 only */
2530 	if (static_branch_unlikely(&enable_evmcs))
2531 		vmcs->hdr.revision_id = KVM_EVMCS_VERSION;
2532 	else
2533 		vmcs->hdr.revision_id = vmcs_config.revision_id;
2534 
2535 	if (shadow)
2536 		vmcs->hdr.shadow_vmcs = 1;
2537 	return vmcs;
2538 }
2539 
2540 void free_vmcs(struct vmcs *vmcs)
2541 {
2542 	free_pages((unsigned long)vmcs, vmcs_config.order);
2543 }
2544 
2545 /*
2546  * Free a VMCS, but before that VMCLEAR it on the CPU where it was last loaded
2547  */
2548 void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
2549 {
2550 	if (!loaded_vmcs->vmcs)
2551 		return;
2552 	loaded_vmcs_clear(loaded_vmcs);
2553 	free_vmcs(loaded_vmcs->vmcs);
2554 	loaded_vmcs->vmcs = NULL;
2555 	if (loaded_vmcs->msr_bitmap)
2556 		free_page((unsigned long)loaded_vmcs->msr_bitmap);
2557 	WARN_ON(loaded_vmcs->shadow_vmcs != NULL);
2558 }
2559 
2560 int alloc_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
2561 {
2562 	loaded_vmcs->vmcs = alloc_vmcs(false);
2563 	if (!loaded_vmcs->vmcs)
2564 		return -ENOMEM;
2565 
2566 	loaded_vmcs->shadow_vmcs = NULL;
2567 	loaded_vmcs->hv_timer_soft_disabled = false;
2568 	loaded_vmcs_init(loaded_vmcs);
2569 
2570 	if (cpu_has_vmx_msr_bitmap()) {
2571 		loaded_vmcs->msr_bitmap = (unsigned long *)
2572 				__get_free_page(GFP_KERNEL_ACCOUNT);
2573 		if (!loaded_vmcs->msr_bitmap)
2574 			goto out_vmcs;
2575 		memset(loaded_vmcs->msr_bitmap, 0xff, PAGE_SIZE);
2576 
2577 		if (IS_ENABLED(CONFIG_HYPERV) &&
2578 		    static_branch_unlikely(&enable_evmcs) &&
2579 		    (ms_hyperv.nested_features & HV_X64_NESTED_MSR_BITMAP)) {
2580 			struct hv_enlightened_vmcs *evmcs =
2581 				(struct hv_enlightened_vmcs *)loaded_vmcs->vmcs;
2582 
2583 			evmcs->hv_enlightenments_control.msr_bitmap = 1;
2584 		}
2585 	}
2586 
2587 	memset(&loaded_vmcs->host_state, 0, sizeof(struct vmcs_host_state));
2588 	memset(&loaded_vmcs->controls_shadow, 0,
2589 		sizeof(struct vmcs_controls_shadow));
2590 
2591 	return 0;
2592 
2593 out_vmcs:
2594 	free_loaded_vmcs(loaded_vmcs);
2595 	return -ENOMEM;
2596 }
2597 
2598 static void free_kvm_area(void)
2599 {
2600 	int cpu;
2601 
2602 	for_each_possible_cpu(cpu) {
2603 		free_vmcs(per_cpu(vmxarea, cpu));
2604 		per_cpu(vmxarea, cpu) = NULL;
2605 	}
2606 }
2607 
2608 static __init int alloc_kvm_area(void)
2609 {
2610 	int cpu;
2611 
2612 	for_each_possible_cpu(cpu) {
2613 		struct vmcs *vmcs;
2614 
2615 		vmcs = alloc_vmcs_cpu(false, cpu, GFP_KERNEL);
2616 		if (!vmcs) {
2617 			free_kvm_area();
2618 			return -ENOMEM;
2619 		}
2620 
2621 		/*
2622 		 * When eVMCS is enabled, alloc_vmcs_cpu() sets
2623 		 * vmcs->revision_id to KVM_EVMCS_VERSION instead of
2624 		 * revision_id reported by MSR_IA32_VMX_BASIC.
2625 		 *
2626 		 * However, even though not explicitly documented by
2627 		 * TLFS, VMXArea passed as VMXON argument should
2628 		 * still be marked with revision_id reported by
2629 		 * physical CPU.
2630 		 */
2631 		if (static_branch_unlikely(&enable_evmcs))
2632 			vmcs->hdr.revision_id = vmcs_config.revision_id;
2633 
2634 		per_cpu(vmxarea, cpu) = vmcs;
2635 	}
2636 	return 0;
2637 }
2638 
2639 static void fix_pmode_seg(struct kvm_vcpu *vcpu, int seg,
2640 		struct kvm_segment *save)
2641 {
2642 	if (!emulate_invalid_guest_state) {
2643 		/*
2644 		 * CS and SS RPL should be equal during guest entry according
2645 		 * to VMX spec, but in reality it is not always so. Since vcpu
2646 		 * is in the middle of the transition from real mode to
2647 		 * protected mode it is safe to assume that RPL 0 is a good
2648 		 * default value.
2649 		 */
2650 		if (seg == VCPU_SREG_CS || seg == VCPU_SREG_SS)
2651 			save->selector &= ~SEGMENT_RPL_MASK;
2652 		save->dpl = save->selector & SEGMENT_RPL_MASK;
2653 		save->s = 1;
2654 	}
2655 	vmx_set_segment(vcpu, save, seg);
2656 }
2657 
2658 static void enter_pmode(struct kvm_vcpu *vcpu)
2659 {
2660 	unsigned long flags;
2661 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2662 
2663 	/*
2664 	 * Update real mode segment cache. It may be not up-to-date if sement
2665 	 * register was written while vcpu was in a guest mode.
2666 	 */
2667 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
2668 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
2669 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
2670 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
2671 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
2672 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
2673 
2674 	vmx->rmode.vm86_active = 0;
2675 
2676 	vmx_set_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
2677 
2678 	flags = vmcs_readl(GUEST_RFLAGS);
2679 	flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
2680 	flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
2681 	vmcs_writel(GUEST_RFLAGS, flags);
2682 
2683 	vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
2684 			(vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
2685 
2686 	update_exception_bitmap(vcpu);
2687 
2688 	fix_pmode_seg(vcpu, VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
2689 	fix_pmode_seg(vcpu, VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
2690 	fix_pmode_seg(vcpu, VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
2691 	fix_pmode_seg(vcpu, VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
2692 	fix_pmode_seg(vcpu, VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
2693 	fix_pmode_seg(vcpu, VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
2694 }
2695 
2696 static void fix_rmode_seg(int seg, struct kvm_segment *save)
2697 {
2698 	const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2699 	struct kvm_segment var = *save;
2700 
2701 	var.dpl = 0x3;
2702 	if (seg == VCPU_SREG_CS)
2703 		var.type = 0x3;
2704 
2705 	if (!emulate_invalid_guest_state) {
2706 		var.selector = var.base >> 4;
2707 		var.base = var.base & 0xffff0;
2708 		var.limit = 0xffff;
2709 		var.g = 0;
2710 		var.db = 0;
2711 		var.present = 1;
2712 		var.s = 1;
2713 		var.l = 0;
2714 		var.unusable = 0;
2715 		var.type = 0x3;
2716 		var.avl = 0;
2717 		if (save->base & 0xf)
2718 			printk_once(KERN_WARNING "kvm: segment base is not "
2719 					"paragraph aligned when entering "
2720 					"protected mode (seg=%d)", seg);
2721 	}
2722 
2723 	vmcs_write16(sf->selector, var.selector);
2724 	vmcs_writel(sf->base, var.base);
2725 	vmcs_write32(sf->limit, var.limit);
2726 	vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(&var));
2727 }
2728 
2729 static void enter_rmode(struct kvm_vcpu *vcpu)
2730 {
2731 	unsigned long flags;
2732 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2733 	struct kvm_vmx *kvm_vmx = to_kvm_vmx(vcpu->kvm);
2734 
2735 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
2736 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
2737 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
2738 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
2739 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
2740 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
2741 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
2742 
2743 	vmx->rmode.vm86_active = 1;
2744 
2745 	/*
2746 	 * Very old userspace does not call KVM_SET_TSS_ADDR before entering
2747 	 * vcpu. Warn the user that an update is overdue.
2748 	 */
2749 	if (!kvm_vmx->tss_addr)
2750 		printk_once(KERN_WARNING "kvm: KVM_SET_TSS_ADDR need to be "
2751 			     "called before entering vcpu\n");
2752 
2753 	vmx_segment_cache_clear(vmx);
2754 
2755 	vmcs_writel(GUEST_TR_BASE, kvm_vmx->tss_addr);
2756 	vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
2757 	vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
2758 
2759 	flags = vmcs_readl(GUEST_RFLAGS);
2760 	vmx->rmode.save_rflags = flags;
2761 
2762 	flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
2763 
2764 	vmcs_writel(GUEST_RFLAGS, flags);
2765 	vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
2766 	update_exception_bitmap(vcpu);
2767 
2768 	fix_rmode_seg(VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
2769 	fix_rmode_seg(VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
2770 	fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
2771 	fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
2772 	fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
2773 	fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
2774 
2775 	kvm_mmu_reset_context(vcpu);
2776 }
2777 
2778 void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
2779 {
2780 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2781 	struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
2782 
2783 	if (!msr)
2784 		return;
2785 
2786 	vcpu->arch.efer = efer;
2787 	if (efer & EFER_LMA) {
2788 		vm_entry_controls_setbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
2789 		msr->data = efer;
2790 	} else {
2791 		vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
2792 
2793 		msr->data = efer & ~EFER_LME;
2794 	}
2795 	setup_msrs(vmx);
2796 }
2797 
2798 #ifdef CONFIG_X86_64
2799 
2800 static void enter_lmode(struct kvm_vcpu *vcpu)
2801 {
2802 	u32 guest_tr_ar;
2803 
2804 	vmx_segment_cache_clear(to_vmx(vcpu));
2805 
2806 	guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
2807 	if ((guest_tr_ar & VMX_AR_TYPE_MASK) != VMX_AR_TYPE_BUSY_64_TSS) {
2808 		pr_debug_ratelimited("%s: tss fixup for long mode. \n",
2809 				     __func__);
2810 		vmcs_write32(GUEST_TR_AR_BYTES,
2811 			     (guest_tr_ar & ~VMX_AR_TYPE_MASK)
2812 			     | VMX_AR_TYPE_BUSY_64_TSS);
2813 	}
2814 	vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
2815 }
2816 
2817 static void exit_lmode(struct kvm_vcpu *vcpu)
2818 {
2819 	vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
2820 	vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
2821 }
2822 
2823 #endif
2824 
2825 static void vmx_flush_tlb_gva(struct kvm_vcpu *vcpu, gva_t addr)
2826 {
2827 	int vpid = to_vmx(vcpu)->vpid;
2828 
2829 	if (!vpid_sync_vcpu_addr(vpid, addr))
2830 		vpid_sync_context(vpid);
2831 
2832 	/*
2833 	 * If VPIDs are not supported or enabled, then the above is a no-op.
2834 	 * But we don't really need a TLB flush in that case anyway, because
2835 	 * each VM entry/exit includes an implicit flush when VPID is 0.
2836 	 */
2837 }
2838 
2839 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
2840 {
2841 	ulong cr0_guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
2842 
2843 	vcpu->arch.cr0 &= ~cr0_guest_owned_bits;
2844 	vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & cr0_guest_owned_bits;
2845 }
2846 
2847 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
2848 {
2849 	ulong cr4_guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
2850 
2851 	vcpu->arch.cr4 &= ~cr4_guest_owned_bits;
2852 	vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & cr4_guest_owned_bits;
2853 }
2854 
2855 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
2856 {
2857 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
2858 
2859 	if (!kvm_register_is_dirty(vcpu, VCPU_EXREG_PDPTR))
2860 		return;
2861 
2862 	if (is_pae_paging(vcpu)) {
2863 		vmcs_write64(GUEST_PDPTR0, mmu->pdptrs[0]);
2864 		vmcs_write64(GUEST_PDPTR1, mmu->pdptrs[1]);
2865 		vmcs_write64(GUEST_PDPTR2, mmu->pdptrs[2]);
2866 		vmcs_write64(GUEST_PDPTR3, mmu->pdptrs[3]);
2867 	}
2868 }
2869 
2870 void ept_save_pdptrs(struct kvm_vcpu *vcpu)
2871 {
2872 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
2873 
2874 	if (is_pae_paging(vcpu)) {
2875 		mmu->pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
2876 		mmu->pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
2877 		mmu->pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
2878 		mmu->pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
2879 	}
2880 
2881 	kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
2882 }
2883 
2884 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
2885 					unsigned long cr0,
2886 					struct kvm_vcpu *vcpu)
2887 {
2888 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2889 
2890 	if (!kvm_register_is_available(vcpu, VCPU_EXREG_CR3))
2891 		vmx_cache_reg(vcpu, VCPU_EXREG_CR3);
2892 	if (!(cr0 & X86_CR0_PG)) {
2893 		/* From paging/starting to nonpaging */
2894 		exec_controls_setbit(vmx, CPU_BASED_CR3_LOAD_EXITING |
2895 					  CPU_BASED_CR3_STORE_EXITING);
2896 		vcpu->arch.cr0 = cr0;
2897 		vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
2898 	} else if (!is_paging(vcpu)) {
2899 		/* From nonpaging to paging */
2900 		exec_controls_clearbit(vmx, CPU_BASED_CR3_LOAD_EXITING |
2901 					    CPU_BASED_CR3_STORE_EXITING);
2902 		vcpu->arch.cr0 = cr0;
2903 		vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
2904 	}
2905 
2906 	if (!(cr0 & X86_CR0_WP))
2907 		*hw_cr0 &= ~X86_CR0_WP;
2908 }
2909 
2910 void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
2911 {
2912 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2913 	unsigned long hw_cr0;
2914 
2915 	hw_cr0 = (cr0 & ~KVM_VM_CR0_ALWAYS_OFF);
2916 	if (enable_unrestricted_guest)
2917 		hw_cr0 |= KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
2918 	else {
2919 		hw_cr0 |= KVM_VM_CR0_ALWAYS_ON;
2920 
2921 		if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
2922 			enter_pmode(vcpu);
2923 
2924 		if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
2925 			enter_rmode(vcpu);
2926 	}
2927 
2928 #ifdef CONFIG_X86_64
2929 	if (vcpu->arch.efer & EFER_LME) {
2930 		if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
2931 			enter_lmode(vcpu);
2932 		if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
2933 			exit_lmode(vcpu);
2934 	}
2935 #endif
2936 
2937 	if (enable_ept && !enable_unrestricted_guest)
2938 		ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
2939 
2940 	vmcs_writel(CR0_READ_SHADOW, cr0);
2941 	vmcs_writel(GUEST_CR0, hw_cr0);
2942 	vcpu->arch.cr0 = cr0;
2943 
2944 	/* depends on vcpu->arch.cr0 to be set to a new value */
2945 	vmx->emulation_required = emulation_required(vcpu);
2946 }
2947 
2948 static int get_ept_level(struct kvm_vcpu *vcpu)
2949 {
2950 	/* Nested EPT currently only supports 4-level walks. */
2951 	if (is_guest_mode(vcpu) && nested_cpu_has_ept(get_vmcs12(vcpu)))
2952 		return 4;
2953 	if (cpu_has_vmx_ept_5levels() && (cpuid_maxphyaddr(vcpu) > 48))
2954 		return 5;
2955 	return 4;
2956 }
2957 
2958 u64 construct_eptp(struct kvm_vcpu *vcpu, unsigned long root_hpa)
2959 {
2960 	u64 eptp = VMX_EPTP_MT_WB;
2961 
2962 	eptp |= (get_ept_level(vcpu) == 5) ? VMX_EPTP_PWL_5 : VMX_EPTP_PWL_4;
2963 
2964 	if (enable_ept_ad_bits &&
2965 	    (!is_guest_mode(vcpu) || nested_ept_ad_enabled(vcpu)))
2966 		eptp |= VMX_EPTP_AD_ENABLE_BIT;
2967 	eptp |= (root_hpa & PAGE_MASK);
2968 
2969 	return eptp;
2970 }
2971 
2972 void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
2973 {
2974 	struct kvm *kvm = vcpu->kvm;
2975 	bool update_guest_cr3 = true;
2976 	unsigned long guest_cr3;
2977 	u64 eptp;
2978 
2979 	guest_cr3 = cr3;
2980 	if (enable_ept) {
2981 		eptp = construct_eptp(vcpu, cr3);
2982 		vmcs_write64(EPT_POINTER, eptp);
2983 
2984 		if (kvm_x86_ops->tlb_remote_flush) {
2985 			spin_lock(&to_kvm_vmx(kvm)->ept_pointer_lock);
2986 			to_vmx(vcpu)->ept_pointer = eptp;
2987 			to_kvm_vmx(kvm)->ept_pointers_match
2988 				= EPT_POINTERS_CHECK;
2989 			spin_unlock(&to_kvm_vmx(kvm)->ept_pointer_lock);
2990 		}
2991 
2992 		/* Loading vmcs02.GUEST_CR3 is handled by nested VM-Enter. */
2993 		if (is_guest_mode(vcpu))
2994 			update_guest_cr3 = false;
2995 		else if (!enable_unrestricted_guest && !is_paging(vcpu))
2996 			guest_cr3 = to_kvm_vmx(kvm)->ept_identity_map_addr;
2997 		else if (test_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail))
2998 			guest_cr3 = vcpu->arch.cr3;
2999 		else /* vmcs01.GUEST_CR3 is already up-to-date. */
3000 			update_guest_cr3 = false;
3001 		ept_load_pdptrs(vcpu);
3002 	}
3003 
3004 	if (update_guest_cr3)
3005 		vmcs_writel(GUEST_CR3, guest_cr3);
3006 }
3007 
3008 int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
3009 {
3010 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3011 	/*
3012 	 * Pass through host's Machine Check Enable value to hw_cr4, which
3013 	 * is in force while we are in guest mode.  Do not let guests control
3014 	 * this bit, even if host CR4.MCE == 0.
3015 	 */
3016 	unsigned long hw_cr4;
3017 
3018 	hw_cr4 = (cr4_read_shadow() & X86_CR4_MCE) | (cr4 & ~X86_CR4_MCE);
3019 	if (enable_unrestricted_guest)
3020 		hw_cr4 |= KVM_VM_CR4_ALWAYS_ON_UNRESTRICTED_GUEST;
3021 	else if (vmx->rmode.vm86_active)
3022 		hw_cr4 |= KVM_RMODE_VM_CR4_ALWAYS_ON;
3023 	else
3024 		hw_cr4 |= KVM_PMODE_VM_CR4_ALWAYS_ON;
3025 
3026 	if (!boot_cpu_has(X86_FEATURE_UMIP) && vmx_umip_emulated()) {
3027 		if (cr4 & X86_CR4_UMIP) {
3028 			secondary_exec_controls_setbit(vmx, SECONDARY_EXEC_DESC);
3029 			hw_cr4 &= ~X86_CR4_UMIP;
3030 		} else if (!is_guest_mode(vcpu) ||
3031 			!nested_cpu_has2(get_vmcs12(vcpu), SECONDARY_EXEC_DESC)) {
3032 			secondary_exec_controls_clearbit(vmx, SECONDARY_EXEC_DESC);
3033 		}
3034 	}
3035 
3036 	if (cr4 & X86_CR4_VMXE) {
3037 		/*
3038 		 * To use VMXON (and later other VMX instructions), a guest
3039 		 * must first be able to turn on cr4.VMXE (see handle_vmon()).
3040 		 * So basically the check on whether to allow nested VMX
3041 		 * is here.  We operate under the default treatment of SMM,
3042 		 * so VMX cannot be enabled under SMM.
3043 		 */
3044 		if (!nested_vmx_allowed(vcpu) || is_smm(vcpu))
3045 			return 1;
3046 	}
3047 
3048 	if (vmx->nested.vmxon && !nested_cr4_valid(vcpu, cr4))
3049 		return 1;
3050 
3051 	vcpu->arch.cr4 = cr4;
3052 
3053 	if (!enable_unrestricted_guest) {
3054 		if (enable_ept) {
3055 			if (!is_paging(vcpu)) {
3056 				hw_cr4 &= ~X86_CR4_PAE;
3057 				hw_cr4 |= X86_CR4_PSE;
3058 			} else if (!(cr4 & X86_CR4_PAE)) {
3059 				hw_cr4 &= ~X86_CR4_PAE;
3060 			}
3061 		}
3062 
3063 		/*
3064 		 * SMEP/SMAP/PKU is disabled if CPU is in non-paging mode in
3065 		 * hardware.  To emulate this behavior, SMEP/SMAP/PKU needs
3066 		 * to be manually disabled when guest switches to non-paging
3067 		 * mode.
3068 		 *
3069 		 * If !enable_unrestricted_guest, the CPU is always running
3070 		 * with CR0.PG=1 and CR4 needs to be modified.
3071 		 * If enable_unrestricted_guest, the CPU automatically
3072 		 * disables SMEP/SMAP/PKU when the guest sets CR0.PG=0.
3073 		 */
3074 		if (!is_paging(vcpu))
3075 			hw_cr4 &= ~(X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE);
3076 	}
3077 
3078 	vmcs_writel(CR4_READ_SHADOW, cr4);
3079 	vmcs_writel(GUEST_CR4, hw_cr4);
3080 	return 0;
3081 }
3082 
3083 void vmx_get_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg)
3084 {
3085 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3086 	u32 ar;
3087 
3088 	if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3089 		*var = vmx->rmode.segs[seg];
3090 		if (seg == VCPU_SREG_TR
3091 		    || var->selector == vmx_read_guest_seg_selector(vmx, seg))
3092 			return;
3093 		var->base = vmx_read_guest_seg_base(vmx, seg);
3094 		var->selector = vmx_read_guest_seg_selector(vmx, seg);
3095 		return;
3096 	}
3097 	var->base = vmx_read_guest_seg_base(vmx, seg);
3098 	var->limit = vmx_read_guest_seg_limit(vmx, seg);
3099 	var->selector = vmx_read_guest_seg_selector(vmx, seg);
3100 	ar = vmx_read_guest_seg_ar(vmx, seg);
3101 	var->unusable = (ar >> 16) & 1;
3102 	var->type = ar & 15;
3103 	var->s = (ar >> 4) & 1;
3104 	var->dpl = (ar >> 5) & 3;
3105 	/*
3106 	 * Some userspaces do not preserve unusable property. Since usable
3107 	 * segment has to be present according to VMX spec we can use present
3108 	 * property to amend userspace bug by making unusable segment always
3109 	 * nonpresent. vmx_segment_access_rights() already marks nonpresent
3110 	 * segment as unusable.
3111 	 */
3112 	var->present = !var->unusable;
3113 	var->avl = (ar >> 12) & 1;
3114 	var->l = (ar >> 13) & 1;
3115 	var->db = (ar >> 14) & 1;
3116 	var->g = (ar >> 15) & 1;
3117 }
3118 
3119 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
3120 {
3121 	struct kvm_segment s;
3122 
3123 	if (to_vmx(vcpu)->rmode.vm86_active) {
3124 		vmx_get_segment(vcpu, &s, seg);
3125 		return s.base;
3126 	}
3127 	return vmx_read_guest_seg_base(to_vmx(vcpu), seg);
3128 }
3129 
3130 int vmx_get_cpl(struct kvm_vcpu *vcpu)
3131 {
3132 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3133 
3134 	if (unlikely(vmx->rmode.vm86_active))
3135 		return 0;
3136 	else {
3137 		int ar = vmx_read_guest_seg_ar(vmx, VCPU_SREG_SS);
3138 		return VMX_AR_DPL(ar);
3139 	}
3140 }
3141 
3142 static u32 vmx_segment_access_rights(struct kvm_segment *var)
3143 {
3144 	u32 ar;
3145 
3146 	if (var->unusable || !var->present)
3147 		ar = 1 << 16;
3148 	else {
3149 		ar = var->type & 15;
3150 		ar |= (var->s & 1) << 4;
3151 		ar |= (var->dpl & 3) << 5;
3152 		ar |= (var->present & 1) << 7;
3153 		ar |= (var->avl & 1) << 12;
3154 		ar |= (var->l & 1) << 13;
3155 		ar |= (var->db & 1) << 14;
3156 		ar |= (var->g & 1) << 15;
3157 	}
3158 
3159 	return ar;
3160 }
3161 
3162 void vmx_set_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg)
3163 {
3164 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3165 	const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3166 
3167 	vmx_segment_cache_clear(vmx);
3168 
3169 	if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3170 		vmx->rmode.segs[seg] = *var;
3171 		if (seg == VCPU_SREG_TR)
3172 			vmcs_write16(sf->selector, var->selector);
3173 		else if (var->s)
3174 			fix_rmode_seg(seg, &vmx->rmode.segs[seg]);
3175 		goto out;
3176 	}
3177 
3178 	vmcs_writel(sf->base, var->base);
3179 	vmcs_write32(sf->limit, var->limit);
3180 	vmcs_write16(sf->selector, var->selector);
3181 
3182 	/*
3183 	 *   Fix the "Accessed" bit in AR field of segment registers for older
3184 	 * qemu binaries.
3185 	 *   IA32 arch specifies that at the time of processor reset the
3186 	 * "Accessed" bit in the AR field of segment registers is 1. And qemu
3187 	 * is setting it to 0 in the userland code. This causes invalid guest
3188 	 * state vmexit when "unrestricted guest" mode is turned on.
3189 	 *    Fix for this setup issue in cpu_reset is being pushed in the qemu
3190 	 * tree. Newer qemu binaries with that qemu fix would not need this
3191 	 * kvm hack.
3192 	 */
3193 	if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
3194 		var->type |= 0x1; /* Accessed */
3195 
3196 	vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(var));
3197 
3198 out:
3199 	vmx->emulation_required = emulation_required(vcpu);
3200 }
3201 
3202 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3203 {
3204 	u32 ar = vmx_read_guest_seg_ar(to_vmx(vcpu), VCPU_SREG_CS);
3205 
3206 	*db = (ar >> 14) & 1;
3207 	*l = (ar >> 13) & 1;
3208 }
3209 
3210 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3211 {
3212 	dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
3213 	dt->address = vmcs_readl(GUEST_IDTR_BASE);
3214 }
3215 
3216 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3217 {
3218 	vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
3219 	vmcs_writel(GUEST_IDTR_BASE, dt->address);
3220 }
3221 
3222 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3223 {
3224 	dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
3225 	dt->address = vmcs_readl(GUEST_GDTR_BASE);
3226 }
3227 
3228 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3229 {
3230 	vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
3231 	vmcs_writel(GUEST_GDTR_BASE, dt->address);
3232 }
3233 
3234 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
3235 {
3236 	struct kvm_segment var;
3237 	u32 ar;
3238 
3239 	vmx_get_segment(vcpu, &var, seg);
3240 	var.dpl = 0x3;
3241 	if (seg == VCPU_SREG_CS)
3242 		var.type = 0x3;
3243 	ar = vmx_segment_access_rights(&var);
3244 
3245 	if (var.base != (var.selector << 4))
3246 		return false;
3247 	if (var.limit != 0xffff)
3248 		return false;
3249 	if (ar != 0xf3)
3250 		return false;
3251 
3252 	return true;
3253 }
3254 
3255 static bool code_segment_valid(struct kvm_vcpu *vcpu)
3256 {
3257 	struct kvm_segment cs;
3258 	unsigned int cs_rpl;
3259 
3260 	vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3261 	cs_rpl = cs.selector & SEGMENT_RPL_MASK;
3262 
3263 	if (cs.unusable)
3264 		return false;
3265 	if (~cs.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_ACCESSES_MASK))
3266 		return false;
3267 	if (!cs.s)
3268 		return false;
3269 	if (cs.type & VMX_AR_TYPE_WRITEABLE_MASK) {
3270 		if (cs.dpl > cs_rpl)
3271 			return false;
3272 	} else {
3273 		if (cs.dpl != cs_rpl)
3274 			return false;
3275 	}
3276 	if (!cs.present)
3277 		return false;
3278 
3279 	/* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
3280 	return true;
3281 }
3282 
3283 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
3284 {
3285 	struct kvm_segment ss;
3286 	unsigned int ss_rpl;
3287 
3288 	vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3289 	ss_rpl = ss.selector & SEGMENT_RPL_MASK;
3290 
3291 	if (ss.unusable)
3292 		return true;
3293 	if (ss.type != 3 && ss.type != 7)
3294 		return false;
3295 	if (!ss.s)
3296 		return false;
3297 	if (ss.dpl != ss_rpl) /* DPL != RPL */
3298 		return false;
3299 	if (!ss.present)
3300 		return false;
3301 
3302 	return true;
3303 }
3304 
3305 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
3306 {
3307 	struct kvm_segment var;
3308 	unsigned int rpl;
3309 
3310 	vmx_get_segment(vcpu, &var, seg);
3311 	rpl = var.selector & SEGMENT_RPL_MASK;
3312 
3313 	if (var.unusable)
3314 		return true;
3315 	if (!var.s)
3316 		return false;
3317 	if (!var.present)
3318 		return false;
3319 	if (~var.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_WRITEABLE_MASK)) {
3320 		if (var.dpl < rpl) /* DPL < RPL */
3321 			return false;
3322 	}
3323 
3324 	/* TODO: Add other members to kvm_segment_field to allow checking for other access
3325 	 * rights flags
3326 	 */
3327 	return true;
3328 }
3329 
3330 static bool tr_valid(struct kvm_vcpu *vcpu)
3331 {
3332 	struct kvm_segment tr;
3333 
3334 	vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
3335 
3336 	if (tr.unusable)
3337 		return false;
3338 	if (tr.selector & SEGMENT_TI_MASK)	/* TI = 1 */
3339 		return false;
3340 	if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
3341 		return false;
3342 	if (!tr.present)
3343 		return false;
3344 
3345 	return true;
3346 }
3347 
3348 static bool ldtr_valid(struct kvm_vcpu *vcpu)
3349 {
3350 	struct kvm_segment ldtr;
3351 
3352 	vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
3353 
3354 	if (ldtr.unusable)
3355 		return true;
3356 	if (ldtr.selector & SEGMENT_TI_MASK)	/* TI = 1 */
3357 		return false;
3358 	if (ldtr.type != 2)
3359 		return false;
3360 	if (!ldtr.present)
3361 		return false;
3362 
3363 	return true;
3364 }
3365 
3366 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
3367 {
3368 	struct kvm_segment cs, ss;
3369 
3370 	vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3371 	vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3372 
3373 	return ((cs.selector & SEGMENT_RPL_MASK) ==
3374 		 (ss.selector & SEGMENT_RPL_MASK));
3375 }
3376 
3377 /*
3378  * Check if guest state is valid. Returns true if valid, false if
3379  * not.
3380  * We assume that registers are always usable
3381  */
3382 static bool guest_state_valid(struct kvm_vcpu *vcpu)
3383 {
3384 	if (enable_unrestricted_guest)
3385 		return true;
3386 
3387 	/* real mode guest state checks */
3388 	if (!is_protmode(vcpu) || (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
3389 		if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
3390 			return false;
3391 		if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
3392 			return false;
3393 		if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
3394 			return false;
3395 		if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
3396 			return false;
3397 		if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
3398 			return false;
3399 		if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
3400 			return false;
3401 	} else {
3402 	/* protected mode guest state checks */
3403 		if (!cs_ss_rpl_check(vcpu))
3404 			return false;
3405 		if (!code_segment_valid(vcpu))
3406 			return false;
3407 		if (!stack_segment_valid(vcpu))
3408 			return false;
3409 		if (!data_segment_valid(vcpu, VCPU_SREG_DS))
3410 			return false;
3411 		if (!data_segment_valid(vcpu, VCPU_SREG_ES))
3412 			return false;
3413 		if (!data_segment_valid(vcpu, VCPU_SREG_FS))
3414 			return false;
3415 		if (!data_segment_valid(vcpu, VCPU_SREG_GS))
3416 			return false;
3417 		if (!tr_valid(vcpu))
3418 			return false;
3419 		if (!ldtr_valid(vcpu))
3420 			return false;
3421 	}
3422 	/* TODO:
3423 	 * - Add checks on RIP
3424 	 * - Add checks on RFLAGS
3425 	 */
3426 
3427 	return true;
3428 }
3429 
3430 static int init_rmode_tss(struct kvm *kvm)
3431 {
3432 	gfn_t fn;
3433 	u16 data = 0;
3434 	int idx, r;
3435 
3436 	idx = srcu_read_lock(&kvm->srcu);
3437 	fn = to_kvm_vmx(kvm)->tss_addr >> PAGE_SHIFT;
3438 	r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3439 	if (r < 0)
3440 		goto out;
3441 	data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
3442 	r = kvm_write_guest_page(kvm, fn++, &data,
3443 			TSS_IOPB_BASE_OFFSET, sizeof(u16));
3444 	if (r < 0)
3445 		goto out;
3446 	r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
3447 	if (r < 0)
3448 		goto out;
3449 	r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3450 	if (r < 0)
3451 		goto out;
3452 	data = ~0;
3453 	r = kvm_write_guest_page(kvm, fn, &data,
3454 				 RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
3455 				 sizeof(u8));
3456 out:
3457 	srcu_read_unlock(&kvm->srcu, idx);
3458 	return r;
3459 }
3460 
3461 static int init_rmode_identity_map(struct kvm *kvm)
3462 {
3463 	struct kvm_vmx *kvm_vmx = to_kvm_vmx(kvm);
3464 	int i, r = 0;
3465 	kvm_pfn_t identity_map_pfn;
3466 	u32 tmp;
3467 
3468 	/* Protect kvm_vmx->ept_identity_pagetable_done. */
3469 	mutex_lock(&kvm->slots_lock);
3470 
3471 	if (likely(kvm_vmx->ept_identity_pagetable_done))
3472 		goto out;
3473 
3474 	if (!kvm_vmx->ept_identity_map_addr)
3475 		kvm_vmx->ept_identity_map_addr = VMX_EPT_IDENTITY_PAGETABLE_ADDR;
3476 	identity_map_pfn = kvm_vmx->ept_identity_map_addr >> PAGE_SHIFT;
3477 
3478 	r = __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
3479 				    kvm_vmx->ept_identity_map_addr, PAGE_SIZE);
3480 	if (r < 0)
3481 		goto out;
3482 
3483 	r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
3484 	if (r < 0)
3485 		goto out;
3486 	/* Set up identity-mapping pagetable for EPT in real mode */
3487 	for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
3488 		tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
3489 			_PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
3490 		r = kvm_write_guest_page(kvm, identity_map_pfn,
3491 				&tmp, i * sizeof(tmp), sizeof(tmp));
3492 		if (r < 0)
3493 			goto out;
3494 	}
3495 	kvm_vmx->ept_identity_pagetable_done = true;
3496 
3497 out:
3498 	mutex_unlock(&kvm->slots_lock);
3499 	return r;
3500 }
3501 
3502 static void seg_setup(int seg)
3503 {
3504 	const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3505 	unsigned int ar;
3506 
3507 	vmcs_write16(sf->selector, 0);
3508 	vmcs_writel(sf->base, 0);
3509 	vmcs_write32(sf->limit, 0xffff);
3510 	ar = 0x93;
3511 	if (seg == VCPU_SREG_CS)
3512 		ar |= 0x08; /* code segment */
3513 
3514 	vmcs_write32(sf->ar_bytes, ar);
3515 }
3516 
3517 static int alloc_apic_access_page(struct kvm *kvm)
3518 {
3519 	struct page *page;
3520 	int r = 0;
3521 
3522 	mutex_lock(&kvm->slots_lock);
3523 	if (kvm->arch.apic_access_page_done)
3524 		goto out;
3525 	r = __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
3526 				    APIC_DEFAULT_PHYS_BASE, PAGE_SIZE);
3527 	if (r)
3528 		goto out;
3529 
3530 	page = gfn_to_page(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
3531 	if (is_error_page(page)) {
3532 		r = -EFAULT;
3533 		goto out;
3534 	}
3535 
3536 	/*
3537 	 * Do not pin the page in memory, so that memory hot-unplug
3538 	 * is able to migrate it.
3539 	 */
3540 	put_page(page);
3541 	kvm->arch.apic_access_page_done = true;
3542 out:
3543 	mutex_unlock(&kvm->slots_lock);
3544 	return r;
3545 }
3546 
3547 int allocate_vpid(void)
3548 {
3549 	int vpid;
3550 
3551 	if (!enable_vpid)
3552 		return 0;
3553 	spin_lock(&vmx_vpid_lock);
3554 	vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
3555 	if (vpid < VMX_NR_VPIDS)
3556 		__set_bit(vpid, vmx_vpid_bitmap);
3557 	else
3558 		vpid = 0;
3559 	spin_unlock(&vmx_vpid_lock);
3560 	return vpid;
3561 }
3562 
3563 void free_vpid(int vpid)
3564 {
3565 	if (!enable_vpid || vpid == 0)
3566 		return;
3567 	spin_lock(&vmx_vpid_lock);
3568 	__clear_bit(vpid, vmx_vpid_bitmap);
3569 	spin_unlock(&vmx_vpid_lock);
3570 }
3571 
3572 static __always_inline void vmx_disable_intercept_for_msr(unsigned long *msr_bitmap,
3573 							  u32 msr, int type)
3574 {
3575 	int f = sizeof(unsigned long);
3576 
3577 	if (!cpu_has_vmx_msr_bitmap())
3578 		return;
3579 
3580 	if (static_branch_unlikely(&enable_evmcs))
3581 		evmcs_touch_msr_bitmap();
3582 
3583 	/*
3584 	 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
3585 	 * have the write-low and read-high bitmap offsets the wrong way round.
3586 	 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
3587 	 */
3588 	if (msr <= 0x1fff) {
3589 		if (type & MSR_TYPE_R)
3590 			/* read-low */
3591 			__clear_bit(msr, msr_bitmap + 0x000 / f);
3592 
3593 		if (type & MSR_TYPE_W)
3594 			/* write-low */
3595 			__clear_bit(msr, msr_bitmap + 0x800 / f);
3596 
3597 	} else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
3598 		msr &= 0x1fff;
3599 		if (type & MSR_TYPE_R)
3600 			/* read-high */
3601 			__clear_bit(msr, msr_bitmap + 0x400 / f);
3602 
3603 		if (type & MSR_TYPE_W)
3604 			/* write-high */
3605 			__clear_bit(msr, msr_bitmap + 0xc00 / f);
3606 
3607 	}
3608 }
3609 
3610 static __always_inline void vmx_enable_intercept_for_msr(unsigned long *msr_bitmap,
3611 							 u32 msr, int type)
3612 {
3613 	int f = sizeof(unsigned long);
3614 
3615 	if (!cpu_has_vmx_msr_bitmap())
3616 		return;
3617 
3618 	if (static_branch_unlikely(&enable_evmcs))
3619 		evmcs_touch_msr_bitmap();
3620 
3621 	/*
3622 	 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
3623 	 * have the write-low and read-high bitmap offsets the wrong way round.
3624 	 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
3625 	 */
3626 	if (msr <= 0x1fff) {
3627 		if (type & MSR_TYPE_R)
3628 			/* read-low */
3629 			__set_bit(msr, msr_bitmap + 0x000 / f);
3630 
3631 		if (type & MSR_TYPE_W)
3632 			/* write-low */
3633 			__set_bit(msr, msr_bitmap + 0x800 / f);
3634 
3635 	} else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
3636 		msr &= 0x1fff;
3637 		if (type & MSR_TYPE_R)
3638 			/* read-high */
3639 			__set_bit(msr, msr_bitmap + 0x400 / f);
3640 
3641 		if (type & MSR_TYPE_W)
3642 			/* write-high */
3643 			__set_bit(msr, msr_bitmap + 0xc00 / f);
3644 
3645 	}
3646 }
3647 
3648 static __always_inline void vmx_set_intercept_for_msr(unsigned long *msr_bitmap,
3649 			     			      u32 msr, int type, bool value)
3650 {
3651 	if (value)
3652 		vmx_enable_intercept_for_msr(msr_bitmap, msr, type);
3653 	else
3654 		vmx_disable_intercept_for_msr(msr_bitmap, msr, type);
3655 }
3656 
3657 static u8 vmx_msr_bitmap_mode(struct kvm_vcpu *vcpu)
3658 {
3659 	u8 mode = 0;
3660 
3661 	if (cpu_has_secondary_exec_ctrls() &&
3662 	    (secondary_exec_controls_get(to_vmx(vcpu)) &
3663 	     SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE)) {
3664 		mode |= MSR_BITMAP_MODE_X2APIC;
3665 		if (enable_apicv && kvm_vcpu_apicv_active(vcpu))
3666 			mode |= MSR_BITMAP_MODE_X2APIC_APICV;
3667 	}
3668 
3669 	return mode;
3670 }
3671 
3672 static void vmx_update_msr_bitmap_x2apic(unsigned long *msr_bitmap,
3673 					 u8 mode)
3674 {
3675 	int msr;
3676 
3677 	for (msr = 0x800; msr <= 0x8ff; msr += BITS_PER_LONG) {
3678 		unsigned word = msr / BITS_PER_LONG;
3679 		msr_bitmap[word] = (mode & MSR_BITMAP_MODE_X2APIC_APICV) ? 0 : ~0;
3680 		msr_bitmap[word + (0x800 / sizeof(long))] = ~0;
3681 	}
3682 
3683 	if (mode & MSR_BITMAP_MODE_X2APIC) {
3684 		/*
3685 		 * TPR reads and writes can be virtualized even if virtual interrupt
3686 		 * delivery is not in use.
3687 		 */
3688 		vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_TASKPRI), MSR_TYPE_RW);
3689 		if (mode & MSR_BITMAP_MODE_X2APIC_APICV) {
3690 			vmx_enable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_TMCCT), MSR_TYPE_R);
3691 			vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_EOI), MSR_TYPE_W);
3692 			vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_SELF_IPI), MSR_TYPE_W);
3693 		}
3694 	}
3695 }
3696 
3697 void vmx_update_msr_bitmap(struct kvm_vcpu *vcpu)
3698 {
3699 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3700 	unsigned long *msr_bitmap = vmx->vmcs01.msr_bitmap;
3701 	u8 mode = vmx_msr_bitmap_mode(vcpu);
3702 	u8 changed = mode ^ vmx->msr_bitmap_mode;
3703 
3704 	if (!changed)
3705 		return;
3706 
3707 	if (changed & (MSR_BITMAP_MODE_X2APIC | MSR_BITMAP_MODE_X2APIC_APICV))
3708 		vmx_update_msr_bitmap_x2apic(msr_bitmap, mode);
3709 
3710 	vmx->msr_bitmap_mode = mode;
3711 }
3712 
3713 void pt_update_intercept_for_msr(struct vcpu_vmx *vmx)
3714 {
3715 	unsigned long *msr_bitmap = vmx->vmcs01.msr_bitmap;
3716 	bool flag = !(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN);
3717 	u32 i;
3718 
3719 	vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_STATUS,
3720 							MSR_TYPE_RW, flag);
3721 	vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_OUTPUT_BASE,
3722 							MSR_TYPE_RW, flag);
3723 	vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_OUTPUT_MASK,
3724 							MSR_TYPE_RW, flag);
3725 	vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_CR3_MATCH,
3726 							MSR_TYPE_RW, flag);
3727 	for (i = 0; i < vmx->pt_desc.addr_range; i++) {
3728 		vmx_set_intercept_for_msr(msr_bitmap,
3729 			MSR_IA32_RTIT_ADDR0_A + i * 2, MSR_TYPE_RW, flag);
3730 		vmx_set_intercept_for_msr(msr_bitmap,
3731 			MSR_IA32_RTIT_ADDR0_B + i * 2, MSR_TYPE_RW, flag);
3732 	}
3733 }
3734 
3735 static bool vmx_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
3736 {
3737 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3738 	void *vapic_page;
3739 	u32 vppr;
3740 	int rvi;
3741 
3742 	if (WARN_ON_ONCE(!is_guest_mode(vcpu)) ||
3743 		!nested_cpu_has_vid(get_vmcs12(vcpu)) ||
3744 		WARN_ON_ONCE(!vmx->nested.virtual_apic_map.gfn))
3745 		return false;
3746 
3747 	rvi = vmx_get_rvi();
3748 
3749 	vapic_page = vmx->nested.virtual_apic_map.hva;
3750 	vppr = *((u32 *)(vapic_page + APIC_PROCPRI));
3751 
3752 	return ((rvi & 0xf0) > (vppr & 0xf0));
3753 }
3754 
3755 static inline bool kvm_vcpu_trigger_posted_interrupt(struct kvm_vcpu *vcpu,
3756 						     bool nested)
3757 {
3758 #ifdef CONFIG_SMP
3759 	int pi_vec = nested ? POSTED_INTR_NESTED_VECTOR : POSTED_INTR_VECTOR;
3760 
3761 	if (vcpu->mode == IN_GUEST_MODE) {
3762 		/*
3763 		 * The vector of interrupt to be delivered to vcpu had
3764 		 * been set in PIR before this function.
3765 		 *
3766 		 * Following cases will be reached in this block, and
3767 		 * we always send a notification event in all cases as
3768 		 * explained below.
3769 		 *
3770 		 * Case 1: vcpu keeps in non-root mode. Sending a
3771 		 * notification event posts the interrupt to vcpu.
3772 		 *
3773 		 * Case 2: vcpu exits to root mode and is still
3774 		 * runnable. PIR will be synced to vIRR before the
3775 		 * next vcpu entry. Sending a notification event in
3776 		 * this case has no effect, as vcpu is not in root
3777 		 * mode.
3778 		 *
3779 		 * Case 3: vcpu exits to root mode and is blocked.
3780 		 * vcpu_block() has already synced PIR to vIRR and
3781 		 * never blocks vcpu if vIRR is not cleared. Therefore,
3782 		 * a blocked vcpu here does not wait for any requested
3783 		 * interrupts in PIR, and sending a notification event
3784 		 * which has no effect is safe here.
3785 		 */
3786 
3787 		apic->send_IPI_mask(get_cpu_mask(vcpu->cpu), pi_vec);
3788 		return true;
3789 	}
3790 #endif
3791 	return false;
3792 }
3793 
3794 static int vmx_deliver_nested_posted_interrupt(struct kvm_vcpu *vcpu,
3795 						int vector)
3796 {
3797 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3798 
3799 	if (is_guest_mode(vcpu) &&
3800 	    vector == vmx->nested.posted_intr_nv) {
3801 		/*
3802 		 * If a posted intr is not recognized by hardware,
3803 		 * we will accomplish it in the next vmentry.
3804 		 */
3805 		vmx->nested.pi_pending = true;
3806 		kvm_make_request(KVM_REQ_EVENT, vcpu);
3807 		/* the PIR and ON have been set by L1. */
3808 		if (!kvm_vcpu_trigger_posted_interrupt(vcpu, true))
3809 			kvm_vcpu_kick(vcpu);
3810 		return 0;
3811 	}
3812 	return -1;
3813 }
3814 /*
3815  * Send interrupt to vcpu via posted interrupt way.
3816  * 1. If target vcpu is running(non-root mode), send posted interrupt
3817  * notification to vcpu and hardware will sync PIR to vIRR atomically.
3818  * 2. If target vcpu isn't running(root mode), kick it to pick up the
3819  * interrupt from PIR in next vmentry.
3820  */
3821 static void vmx_deliver_posted_interrupt(struct kvm_vcpu *vcpu, int vector)
3822 {
3823 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3824 	int r;
3825 
3826 	r = vmx_deliver_nested_posted_interrupt(vcpu, vector);
3827 	if (!r)
3828 		return;
3829 
3830 	if (pi_test_and_set_pir(vector, &vmx->pi_desc))
3831 		return;
3832 
3833 	/* If a previous notification has sent the IPI, nothing to do.  */
3834 	if (pi_test_and_set_on(&vmx->pi_desc))
3835 		return;
3836 
3837 	if (!kvm_vcpu_trigger_posted_interrupt(vcpu, false))
3838 		kvm_vcpu_kick(vcpu);
3839 }
3840 
3841 /*
3842  * Set up the vmcs's constant host-state fields, i.e., host-state fields that
3843  * will not change in the lifetime of the guest.
3844  * Note that host-state that does change is set elsewhere. E.g., host-state
3845  * that is set differently for each CPU is set in vmx_vcpu_load(), not here.
3846  */
3847 void vmx_set_constant_host_state(struct vcpu_vmx *vmx)
3848 {
3849 	u32 low32, high32;
3850 	unsigned long tmpl;
3851 	unsigned long cr0, cr3, cr4;
3852 
3853 	cr0 = read_cr0();
3854 	WARN_ON(cr0 & X86_CR0_TS);
3855 	vmcs_writel(HOST_CR0, cr0);  /* 22.2.3 */
3856 
3857 	/*
3858 	 * Save the most likely value for this task's CR3 in the VMCS.
3859 	 * We can't use __get_current_cr3_fast() because we're not atomic.
3860 	 */
3861 	cr3 = __read_cr3();
3862 	vmcs_writel(HOST_CR3, cr3);		/* 22.2.3  FIXME: shadow tables */
3863 	vmx->loaded_vmcs->host_state.cr3 = cr3;
3864 
3865 	/* Save the most likely value for this task's CR4 in the VMCS. */
3866 	cr4 = cr4_read_shadow();
3867 	vmcs_writel(HOST_CR4, cr4);			/* 22.2.3, 22.2.5 */
3868 	vmx->loaded_vmcs->host_state.cr4 = cr4;
3869 
3870 	vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS);  /* 22.2.4 */
3871 #ifdef CONFIG_X86_64
3872 	/*
3873 	 * Load null selectors, so we can avoid reloading them in
3874 	 * vmx_prepare_switch_to_host(), in case userspace uses
3875 	 * the null selectors too (the expected case).
3876 	 */
3877 	vmcs_write16(HOST_DS_SELECTOR, 0);
3878 	vmcs_write16(HOST_ES_SELECTOR, 0);
3879 #else
3880 	vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
3881 	vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
3882 #endif
3883 	vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
3884 	vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8);  /* 22.2.4 */
3885 
3886 	vmcs_writel(HOST_IDTR_BASE, host_idt_base);   /* 22.2.4 */
3887 
3888 	vmcs_writel(HOST_RIP, (unsigned long)vmx_vmexit); /* 22.2.5 */
3889 
3890 	rdmsr(MSR_IA32_SYSENTER_CS, low32, high32);
3891 	vmcs_write32(HOST_IA32_SYSENTER_CS, low32);
3892 	rdmsrl(MSR_IA32_SYSENTER_EIP, tmpl);
3893 	vmcs_writel(HOST_IA32_SYSENTER_EIP, tmpl);   /* 22.2.3 */
3894 
3895 	if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
3896 		rdmsr(MSR_IA32_CR_PAT, low32, high32);
3897 		vmcs_write64(HOST_IA32_PAT, low32 | ((u64) high32 << 32));
3898 	}
3899 
3900 	if (cpu_has_load_ia32_efer())
3901 		vmcs_write64(HOST_IA32_EFER, host_efer);
3902 }
3903 
3904 void set_cr4_guest_host_mask(struct vcpu_vmx *vmx)
3905 {
3906 	vmx->vcpu.arch.cr4_guest_owned_bits = KVM_CR4_GUEST_OWNED_BITS;
3907 	if (enable_ept)
3908 		vmx->vcpu.arch.cr4_guest_owned_bits |= X86_CR4_PGE;
3909 	if (is_guest_mode(&vmx->vcpu))
3910 		vmx->vcpu.arch.cr4_guest_owned_bits &=
3911 			~get_vmcs12(&vmx->vcpu)->cr4_guest_host_mask;
3912 	vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits);
3913 }
3914 
3915 u32 vmx_pin_based_exec_ctrl(struct vcpu_vmx *vmx)
3916 {
3917 	u32 pin_based_exec_ctrl = vmcs_config.pin_based_exec_ctrl;
3918 
3919 	if (!kvm_vcpu_apicv_active(&vmx->vcpu))
3920 		pin_based_exec_ctrl &= ~PIN_BASED_POSTED_INTR;
3921 
3922 	if (!enable_vnmi)
3923 		pin_based_exec_ctrl &= ~PIN_BASED_VIRTUAL_NMIS;
3924 
3925 	if (!enable_preemption_timer)
3926 		pin_based_exec_ctrl &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
3927 
3928 	return pin_based_exec_ctrl;
3929 }
3930 
3931 static void vmx_refresh_apicv_exec_ctrl(struct kvm_vcpu *vcpu)
3932 {
3933 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3934 
3935 	pin_controls_set(vmx, vmx_pin_based_exec_ctrl(vmx));
3936 	if (cpu_has_secondary_exec_ctrls()) {
3937 		if (kvm_vcpu_apicv_active(vcpu))
3938 			secondary_exec_controls_setbit(vmx,
3939 				      SECONDARY_EXEC_APIC_REGISTER_VIRT |
3940 				      SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
3941 		else
3942 			secondary_exec_controls_clearbit(vmx,
3943 					SECONDARY_EXEC_APIC_REGISTER_VIRT |
3944 					SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
3945 	}
3946 
3947 	if (cpu_has_vmx_msr_bitmap())
3948 		vmx_update_msr_bitmap(vcpu);
3949 }
3950 
3951 u32 vmx_exec_control(struct vcpu_vmx *vmx)
3952 {
3953 	u32 exec_control = vmcs_config.cpu_based_exec_ctrl;
3954 
3955 	if (vmx->vcpu.arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)
3956 		exec_control &= ~CPU_BASED_MOV_DR_EXITING;
3957 
3958 	if (!cpu_need_tpr_shadow(&vmx->vcpu)) {
3959 		exec_control &= ~CPU_BASED_TPR_SHADOW;
3960 #ifdef CONFIG_X86_64
3961 		exec_control |= CPU_BASED_CR8_STORE_EXITING |
3962 				CPU_BASED_CR8_LOAD_EXITING;
3963 #endif
3964 	}
3965 	if (!enable_ept)
3966 		exec_control |= CPU_BASED_CR3_STORE_EXITING |
3967 				CPU_BASED_CR3_LOAD_EXITING  |
3968 				CPU_BASED_INVLPG_EXITING;
3969 	if (kvm_mwait_in_guest(vmx->vcpu.kvm))
3970 		exec_control &= ~(CPU_BASED_MWAIT_EXITING |
3971 				CPU_BASED_MONITOR_EXITING);
3972 	if (kvm_hlt_in_guest(vmx->vcpu.kvm))
3973 		exec_control &= ~CPU_BASED_HLT_EXITING;
3974 	return exec_control;
3975 }
3976 
3977 
3978 static void vmx_compute_secondary_exec_control(struct vcpu_vmx *vmx)
3979 {
3980 	struct kvm_vcpu *vcpu = &vmx->vcpu;
3981 
3982 	u32 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
3983 
3984 	if (pt_mode == PT_MODE_SYSTEM)
3985 		exec_control &= ~(SECONDARY_EXEC_PT_USE_GPA | SECONDARY_EXEC_PT_CONCEAL_VMX);
3986 	if (!cpu_need_virtualize_apic_accesses(vcpu))
3987 		exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
3988 	if (vmx->vpid == 0)
3989 		exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
3990 	if (!enable_ept) {
3991 		exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
3992 		enable_unrestricted_guest = 0;
3993 	}
3994 	if (!enable_unrestricted_guest)
3995 		exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
3996 	if (kvm_pause_in_guest(vmx->vcpu.kvm))
3997 		exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
3998 	if (!kvm_vcpu_apicv_active(vcpu))
3999 		exec_control &= ~(SECONDARY_EXEC_APIC_REGISTER_VIRT |
4000 				  SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
4001 	exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
4002 
4003 	/* SECONDARY_EXEC_DESC is enabled/disabled on writes to CR4.UMIP,
4004 	 * in vmx_set_cr4.  */
4005 	exec_control &= ~SECONDARY_EXEC_DESC;
4006 
4007 	/* SECONDARY_EXEC_SHADOW_VMCS is enabled when L1 executes VMPTRLD
4008 	   (handle_vmptrld).
4009 	   We can NOT enable shadow_vmcs here because we don't have yet
4010 	   a current VMCS12
4011 	*/
4012 	exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
4013 
4014 	if (!enable_pml)
4015 		exec_control &= ~SECONDARY_EXEC_ENABLE_PML;
4016 
4017 	if (vmx_xsaves_supported()) {
4018 		/* Exposing XSAVES only when XSAVE is exposed */
4019 		bool xsaves_enabled =
4020 			boot_cpu_has(X86_FEATURE_XSAVE) &&
4021 			guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
4022 			guest_cpuid_has(vcpu, X86_FEATURE_XSAVES);
4023 
4024 		vcpu->arch.xsaves_enabled = xsaves_enabled;
4025 
4026 		if (!xsaves_enabled)
4027 			exec_control &= ~SECONDARY_EXEC_XSAVES;
4028 
4029 		if (nested) {
4030 			if (xsaves_enabled)
4031 				vmx->nested.msrs.secondary_ctls_high |=
4032 					SECONDARY_EXEC_XSAVES;
4033 			else
4034 				vmx->nested.msrs.secondary_ctls_high &=
4035 					~SECONDARY_EXEC_XSAVES;
4036 		}
4037 	}
4038 
4039 	if (vmx_rdtscp_supported()) {
4040 		bool rdtscp_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP);
4041 		if (!rdtscp_enabled)
4042 			exec_control &= ~SECONDARY_EXEC_RDTSCP;
4043 
4044 		if (nested) {
4045 			if (rdtscp_enabled)
4046 				vmx->nested.msrs.secondary_ctls_high |=
4047 					SECONDARY_EXEC_RDTSCP;
4048 			else
4049 				vmx->nested.msrs.secondary_ctls_high &=
4050 					~SECONDARY_EXEC_RDTSCP;
4051 		}
4052 	}
4053 
4054 	if (vmx_invpcid_supported()) {
4055 		/* Exposing INVPCID only when PCID is exposed */
4056 		bool invpcid_enabled =
4057 			guest_cpuid_has(vcpu, X86_FEATURE_INVPCID) &&
4058 			guest_cpuid_has(vcpu, X86_FEATURE_PCID);
4059 
4060 		if (!invpcid_enabled) {
4061 			exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID;
4062 			guest_cpuid_clear(vcpu, X86_FEATURE_INVPCID);
4063 		}
4064 
4065 		if (nested) {
4066 			if (invpcid_enabled)
4067 				vmx->nested.msrs.secondary_ctls_high |=
4068 					SECONDARY_EXEC_ENABLE_INVPCID;
4069 			else
4070 				vmx->nested.msrs.secondary_ctls_high &=
4071 					~SECONDARY_EXEC_ENABLE_INVPCID;
4072 		}
4073 	}
4074 
4075 	if (vmx_rdrand_supported()) {
4076 		bool rdrand_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDRAND);
4077 		if (rdrand_enabled)
4078 			exec_control &= ~SECONDARY_EXEC_RDRAND_EXITING;
4079 
4080 		if (nested) {
4081 			if (rdrand_enabled)
4082 				vmx->nested.msrs.secondary_ctls_high |=
4083 					SECONDARY_EXEC_RDRAND_EXITING;
4084 			else
4085 				vmx->nested.msrs.secondary_ctls_high &=
4086 					~SECONDARY_EXEC_RDRAND_EXITING;
4087 		}
4088 	}
4089 
4090 	if (vmx_rdseed_supported()) {
4091 		bool rdseed_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDSEED);
4092 		if (rdseed_enabled)
4093 			exec_control &= ~SECONDARY_EXEC_RDSEED_EXITING;
4094 
4095 		if (nested) {
4096 			if (rdseed_enabled)
4097 				vmx->nested.msrs.secondary_ctls_high |=
4098 					SECONDARY_EXEC_RDSEED_EXITING;
4099 			else
4100 				vmx->nested.msrs.secondary_ctls_high &=
4101 					~SECONDARY_EXEC_RDSEED_EXITING;
4102 		}
4103 	}
4104 
4105 	if (vmx_waitpkg_supported()) {
4106 		bool waitpkg_enabled =
4107 			guest_cpuid_has(vcpu, X86_FEATURE_WAITPKG);
4108 
4109 		if (!waitpkg_enabled)
4110 			exec_control &= ~SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE;
4111 
4112 		if (nested) {
4113 			if (waitpkg_enabled)
4114 				vmx->nested.msrs.secondary_ctls_high |=
4115 					SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE;
4116 			else
4117 				vmx->nested.msrs.secondary_ctls_high &=
4118 					~SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE;
4119 		}
4120 	}
4121 
4122 	vmx->secondary_exec_control = exec_control;
4123 }
4124 
4125 static void ept_set_mmio_spte_mask(void)
4126 {
4127 	/*
4128 	 * EPT Misconfigurations can be generated if the value of bits 2:0
4129 	 * of an EPT paging-structure entry is 110b (write/execute).
4130 	 */
4131 	kvm_mmu_set_mmio_spte_mask(VMX_EPT_RWX_MASK,
4132 				   VMX_EPT_MISCONFIG_WX_VALUE, 0);
4133 }
4134 
4135 #define VMX_XSS_EXIT_BITMAP 0
4136 
4137 /*
4138  * Noting that the initialization of Guest-state Area of VMCS is in
4139  * vmx_vcpu_reset().
4140  */
4141 static void init_vmcs(struct vcpu_vmx *vmx)
4142 {
4143 	if (nested)
4144 		nested_vmx_set_vmcs_shadowing_bitmap();
4145 
4146 	if (cpu_has_vmx_msr_bitmap())
4147 		vmcs_write64(MSR_BITMAP, __pa(vmx->vmcs01.msr_bitmap));
4148 
4149 	vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
4150 
4151 	/* Control */
4152 	pin_controls_set(vmx, vmx_pin_based_exec_ctrl(vmx));
4153 
4154 	exec_controls_set(vmx, vmx_exec_control(vmx));
4155 
4156 	if (cpu_has_secondary_exec_ctrls()) {
4157 		vmx_compute_secondary_exec_control(vmx);
4158 		secondary_exec_controls_set(vmx, vmx->secondary_exec_control);
4159 	}
4160 
4161 	if (kvm_vcpu_apicv_active(&vmx->vcpu)) {
4162 		vmcs_write64(EOI_EXIT_BITMAP0, 0);
4163 		vmcs_write64(EOI_EXIT_BITMAP1, 0);
4164 		vmcs_write64(EOI_EXIT_BITMAP2, 0);
4165 		vmcs_write64(EOI_EXIT_BITMAP3, 0);
4166 
4167 		vmcs_write16(GUEST_INTR_STATUS, 0);
4168 
4169 		vmcs_write16(POSTED_INTR_NV, POSTED_INTR_VECTOR);
4170 		vmcs_write64(POSTED_INTR_DESC_ADDR, __pa((&vmx->pi_desc)));
4171 	}
4172 
4173 	if (!kvm_pause_in_guest(vmx->vcpu.kvm)) {
4174 		vmcs_write32(PLE_GAP, ple_gap);
4175 		vmx->ple_window = ple_window;
4176 		vmx->ple_window_dirty = true;
4177 	}
4178 
4179 	vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
4180 	vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
4181 	vmcs_write32(CR3_TARGET_COUNT, 0);           /* 22.2.1 */
4182 
4183 	vmcs_write16(HOST_FS_SELECTOR, 0);            /* 22.2.4 */
4184 	vmcs_write16(HOST_GS_SELECTOR, 0);            /* 22.2.4 */
4185 	vmx_set_constant_host_state(vmx);
4186 	vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
4187 	vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
4188 
4189 	if (cpu_has_vmx_vmfunc())
4190 		vmcs_write64(VM_FUNCTION_CONTROL, 0);
4191 
4192 	vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
4193 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
4194 	vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host.val));
4195 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
4196 	vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest.val));
4197 
4198 	if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
4199 		vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
4200 
4201 	vm_exit_controls_set(vmx, vmx_vmexit_ctrl());
4202 
4203 	/* 22.2.1, 20.8.1 */
4204 	vm_entry_controls_set(vmx, vmx_vmentry_ctrl());
4205 
4206 	vmx->vcpu.arch.cr0_guest_owned_bits = X86_CR0_TS;
4207 	vmcs_writel(CR0_GUEST_HOST_MASK, ~X86_CR0_TS);
4208 
4209 	set_cr4_guest_host_mask(vmx);
4210 
4211 	if (vmx->vpid != 0)
4212 		vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
4213 
4214 	if (vmx_xsaves_supported())
4215 		vmcs_write64(XSS_EXIT_BITMAP, VMX_XSS_EXIT_BITMAP);
4216 
4217 	if (enable_pml) {
4218 		vmcs_write64(PML_ADDRESS, page_to_phys(vmx->pml_pg));
4219 		vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
4220 	}
4221 
4222 	if (cpu_has_vmx_encls_vmexit())
4223 		vmcs_write64(ENCLS_EXITING_BITMAP, -1ull);
4224 
4225 	if (pt_mode == PT_MODE_HOST_GUEST) {
4226 		memset(&vmx->pt_desc, 0, sizeof(vmx->pt_desc));
4227 		/* Bit[6~0] are forced to 1, writes are ignored. */
4228 		vmx->pt_desc.guest.output_mask = 0x7F;
4229 		vmcs_write64(GUEST_IA32_RTIT_CTL, 0);
4230 	}
4231 }
4232 
4233 static void vmx_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
4234 {
4235 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4236 	struct msr_data apic_base_msr;
4237 	u64 cr0;
4238 
4239 	vmx->rmode.vm86_active = 0;
4240 	vmx->spec_ctrl = 0;
4241 
4242 	vmx->msr_ia32_umwait_control = 0;
4243 
4244 	vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
4245 	vmx->hv_deadline_tsc = -1;
4246 	kvm_set_cr8(vcpu, 0);
4247 
4248 	if (!init_event) {
4249 		apic_base_msr.data = APIC_DEFAULT_PHYS_BASE |
4250 				     MSR_IA32_APICBASE_ENABLE;
4251 		if (kvm_vcpu_is_reset_bsp(vcpu))
4252 			apic_base_msr.data |= MSR_IA32_APICBASE_BSP;
4253 		apic_base_msr.host_initiated = true;
4254 		kvm_set_apic_base(vcpu, &apic_base_msr);
4255 	}
4256 
4257 	vmx_segment_cache_clear(vmx);
4258 
4259 	seg_setup(VCPU_SREG_CS);
4260 	vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
4261 	vmcs_writel(GUEST_CS_BASE, 0xffff0000ul);
4262 
4263 	seg_setup(VCPU_SREG_DS);
4264 	seg_setup(VCPU_SREG_ES);
4265 	seg_setup(VCPU_SREG_FS);
4266 	seg_setup(VCPU_SREG_GS);
4267 	seg_setup(VCPU_SREG_SS);
4268 
4269 	vmcs_write16(GUEST_TR_SELECTOR, 0);
4270 	vmcs_writel(GUEST_TR_BASE, 0);
4271 	vmcs_write32(GUEST_TR_LIMIT, 0xffff);
4272 	vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
4273 
4274 	vmcs_write16(GUEST_LDTR_SELECTOR, 0);
4275 	vmcs_writel(GUEST_LDTR_BASE, 0);
4276 	vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
4277 	vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
4278 
4279 	if (!init_event) {
4280 		vmcs_write32(GUEST_SYSENTER_CS, 0);
4281 		vmcs_writel(GUEST_SYSENTER_ESP, 0);
4282 		vmcs_writel(GUEST_SYSENTER_EIP, 0);
4283 		vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
4284 	}
4285 
4286 	kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
4287 	kvm_rip_write(vcpu, 0xfff0);
4288 
4289 	vmcs_writel(GUEST_GDTR_BASE, 0);
4290 	vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
4291 
4292 	vmcs_writel(GUEST_IDTR_BASE, 0);
4293 	vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
4294 
4295 	vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
4296 	vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
4297 	vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS, 0);
4298 	if (kvm_mpx_supported())
4299 		vmcs_write64(GUEST_BNDCFGS, 0);
4300 
4301 	setup_msrs(vmx);
4302 
4303 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);  /* 22.2.1 */
4304 
4305 	if (cpu_has_vmx_tpr_shadow() && !init_event) {
4306 		vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
4307 		if (cpu_need_tpr_shadow(vcpu))
4308 			vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
4309 				     __pa(vcpu->arch.apic->regs));
4310 		vmcs_write32(TPR_THRESHOLD, 0);
4311 	}
4312 
4313 	kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
4314 
4315 	cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
4316 	vmx->vcpu.arch.cr0 = cr0;
4317 	vmx_set_cr0(vcpu, cr0); /* enter rmode */
4318 	vmx_set_cr4(vcpu, 0);
4319 	vmx_set_efer(vcpu, 0);
4320 
4321 	update_exception_bitmap(vcpu);
4322 
4323 	vpid_sync_context(vmx->vpid);
4324 	if (init_event)
4325 		vmx_clear_hlt(vcpu);
4326 }
4327 
4328 static void enable_irq_window(struct kvm_vcpu *vcpu)
4329 {
4330 	exec_controls_setbit(to_vmx(vcpu), CPU_BASED_INTR_WINDOW_EXITING);
4331 }
4332 
4333 static void enable_nmi_window(struct kvm_vcpu *vcpu)
4334 {
4335 	if (!enable_vnmi ||
4336 	    vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_STI) {
4337 		enable_irq_window(vcpu);
4338 		return;
4339 	}
4340 
4341 	exec_controls_setbit(to_vmx(vcpu), CPU_BASED_NMI_WINDOW_EXITING);
4342 }
4343 
4344 static void vmx_inject_irq(struct kvm_vcpu *vcpu)
4345 {
4346 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4347 	uint32_t intr;
4348 	int irq = vcpu->arch.interrupt.nr;
4349 
4350 	trace_kvm_inj_virq(irq);
4351 
4352 	++vcpu->stat.irq_injections;
4353 	if (vmx->rmode.vm86_active) {
4354 		int inc_eip = 0;
4355 		if (vcpu->arch.interrupt.soft)
4356 			inc_eip = vcpu->arch.event_exit_inst_len;
4357 		kvm_inject_realmode_interrupt(vcpu, irq, inc_eip);
4358 		return;
4359 	}
4360 	intr = irq | INTR_INFO_VALID_MASK;
4361 	if (vcpu->arch.interrupt.soft) {
4362 		intr |= INTR_TYPE_SOFT_INTR;
4363 		vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
4364 			     vmx->vcpu.arch.event_exit_inst_len);
4365 	} else
4366 		intr |= INTR_TYPE_EXT_INTR;
4367 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
4368 
4369 	vmx_clear_hlt(vcpu);
4370 }
4371 
4372 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
4373 {
4374 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4375 
4376 	if (!enable_vnmi) {
4377 		/*
4378 		 * Tracking the NMI-blocked state in software is built upon
4379 		 * finding the next open IRQ window. This, in turn, depends on
4380 		 * well-behaving guests: They have to keep IRQs disabled at
4381 		 * least as long as the NMI handler runs. Otherwise we may
4382 		 * cause NMI nesting, maybe breaking the guest. But as this is
4383 		 * highly unlikely, we can live with the residual risk.
4384 		 */
4385 		vmx->loaded_vmcs->soft_vnmi_blocked = 1;
4386 		vmx->loaded_vmcs->vnmi_blocked_time = 0;
4387 	}
4388 
4389 	++vcpu->stat.nmi_injections;
4390 	vmx->loaded_vmcs->nmi_known_unmasked = false;
4391 
4392 	if (vmx->rmode.vm86_active) {
4393 		kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR, 0);
4394 		return;
4395 	}
4396 
4397 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
4398 			INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
4399 
4400 	vmx_clear_hlt(vcpu);
4401 }
4402 
4403 bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
4404 {
4405 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4406 	bool masked;
4407 
4408 	if (!enable_vnmi)
4409 		return vmx->loaded_vmcs->soft_vnmi_blocked;
4410 	if (vmx->loaded_vmcs->nmi_known_unmasked)
4411 		return false;
4412 	masked = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI;
4413 	vmx->loaded_vmcs->nmi_known_unmasked = !masked;
4414 	return masked;
4415 }
4416 
4417 void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
4418 {
4419 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4420 
4421 	if (!enable_vnmi) {
4422 		if (vmx->loaded_vmcs->soft_vnmi_blocked != masked) {
4423 			vmx->loaded_vmcs->soft_vnmi_blocked = masked;
4424 			vmx->loaded_vmcs->vnmi_blocked_time = 0;
4425 		}
4426 	} else {
4427 		vmx->loaded_vmcs->nmi_known_unmasked = !masked;
4428 		if (masked)
4429 			vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
4430 				      GUEST_INTR_STATE_NMI);
4431 		else
4432 			vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
4433 					GUEST_INTR_STATE_NMI);
4434 	}
4435 }
4436 
4437 static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
4438 {
4439 	if (to_vmx(vcpu)->nested.nested_run_pending)
4440 		return 0;
4441 
4442 	if (!enable_vnmi &&
4443 	    to_vmx(vcpu)->loaded_vmcs->soft_vnmi_blocked)
4444 		return 0;
4445 
4446 	return	!(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4447 		  (GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_STI
4448 		   | GUEST_INTR_STATE_NMI));
4449 }
4450 
4451 static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
4452 {
4453 	return (!to_vmx(vcpu)->nested.nested_run_pending &&
4454 		vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
4455 		!(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4456 			(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
4457 }
4458 
4459 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
4460 {
4461 	int ret;
4462 
4463 	if (enable_unrestricted_guest)
4464 		return 0;
4465 
4466 	mutex_lock(&kvm->slots_lock);
4467 	ret = __x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, addr,
4468 				      PAGE_SIZE * 3);
4469 	mutex_unlock(&kvm->slots_lock);
4470 
4471 	if (ret)
4472 		return ret;
4473 	to_kvm_vmx(kvm)->tss_addr = addr;
4474 	return init_rmode_tss(kvm);
4475 }
4476 
4477 static int vmx_set_identity_map_addr(struct kvm *kvm, u64 ident_addr)
4478 {
4479 	to_kvm_vmx(kvm)->ept_identity_map_addr = ident_addr;
4480 	return 0;
4481 }
4482 
4483 static bool rmode_exception(struct kvm_vcpu *vcpu, int vec)
4484 {
4485 	switch (vec) {
4486 	case BP_VECTOR:
4487 		/*
4488 		 * Update instruction length as we may reinject the exception
4489 		 * from user space while in guest debugging mode.
4490 		 */
4491 		to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
4492 			vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4493 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
4494 			return false;
4495 		/* fall through */
4496 	case DB_VECTOR:
4497 		if (vcpu->guest_debug &
4498 			(KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
4499 			return false;
4500 		/* fall through */
4501 	case DE_VECTOR:
4502 	case OF_VECTOR:
4503 	case BR_VECTOR:
4504 	case UD_VECTOR:
4505 	case DF_VECTOR:
4506 	case SS_VECTOR:
4507 	case GP_VECTOR:
4508 	case MF_VECTOR:
4509 		return true;
4510 	break;
4511 	}
4512 	return false;
4513 }
4514 
4515 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
4516 				  int vec, u32 err_code)
4517 {
4518 	/*
4519 	 * Instruction with address size override prefix opcode 0x67
4520 	 * Cause the #SS fault with 0 error code in VM86 mode.
4521 	 */
4522 	if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0) {
4523 		if (kvm_emulate_instruction(vcpu, 0)) {
4524 			if (vcpu->arch.halt_request) {
4525 				vcpu->arch.halt_request = 0;
4526 				return kvm_vcpu_halt(vcpu);
4527 			}
4528 			return 1;
4529 		}
4530 		return 0;
4531 	}
4532 
4533 	/*
4534 	 * Forward all other exceptions that are valid in real mode.
4535 	 * FIXME: Breaks guest debugging in real mode, needs to be fixed with
4536 	 *        the required debugging infrastructure rework.
4537 	 */
4538 	kvm_queue_exception(vcpu, vec);
4539 	return 1;
4540 }
4541 
4542 /*
4543  * Trigger machine check on the host. We assume all the MSRs are already set up
4544  * by the CPU and that we still run on the same CPU as the MCE occurred on.
4545  * We pass a fake environment to the machine check handler because we want
4546  * the guest to be always treated like user space, no matter what context
4547  * it used internally.
4548  */
4549 static void kvm_machine_check(void)
4550 {
4551 #if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
4552 	struct pt_regs regs = {
4553 		.cs = 3, /* Fake ring 3 no matter what the guest ran on */
4554 		.flags = X86_EFLAGS_IF,
4555 	};
4556 
4557 	do_machine_check(&regs, 0);
4558 #endif
4559 }
4560 
4561 static int handle_machine_check(struct kvm_vcpu *vcpu)
4562 {
4563 	/* handled by vmx_vcpu_run() */
4564 	return 1;
4565 }
4566 
4567 static int handle_exception_nmi(struct kvm_vcpu *vcpu)
4568 {
4569 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4570 	struct kvm_run *kvm_run = vcpu->run;
4571 	u32 intr_info, ex_no, error_code;
4572 	unsigned long cr2, rip, dr6;
4573 	u32 vect_info;
4574 
4575 	vect_info = vmx->idt_vectoring_info;
4576 	intr_info = vmx->exit_intr_info;
4577 
4578 	if (is_machine_check(intr_info) || is_nmi(intr_info))
4579 		return 1; /* handled by handle_exception_nmi_irqoff() */
4580 
4581 	if (is_invalid_opcode(intr_info))
4582 		return handle_ud(vcpu);
4583 
4584 	error_code = 0;
4585 	if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
4586 		error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
4587 
4588 	if (!vmx->rmode.vm86_active && is_gp_fault(intr_info)) {
4589 		WARN_ON_ONCE(!enable_vmware_backdoor);
4590 
4591 		/*
4592 		 * VMware backdoor emulation on #GP interception only handles
4593 		 * IN{S}, OUT{S}, and RDPMC, none of which generate a non-zero
4594 		 * error code on #GP.
4595 		 */
4596 		if (error_code) {
4597 			kvm_queue_exception_e(vcpu, GP_VECTOR, error_code);
4598 			return 1;
4599 		}
4600 		return kvm_emulate_instruction(vcpu, EMULTYPE_VMWARE_GP);
4601 	}
4602 
4603 	/*
4604 	 * The #PF with PFEC.RSVD = 1 indicates the guest is accessing
4605 	 * MMIO, it is better to report an internal error.
4606 	 * See the comments in vmx_handle_exit.
4607 	 */
4608 	if ((vect_info & VECTORING_INFO_VALID_MASK) &&
4609 	    !(is_page_fault(intr_info) && !(error_code & PFERR_RSVD_MASK))) {
4610 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4611 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
4612 		vcpu->run->internal.ndata = 3;
4613 		vcpu->run->internal.data[0] = vect_info;
4614 		vcpu->run->internal.data[1] = intr_info;
4615 		vcpu->run->internal.data[2] = error_code;
4616 		return 0;
4617 	}
4618 
4619 	if (is_page_fault(intr_info)) {
4620 		cr2 = vmcs_readl(EXIT_QUALIFICATION);
4621 		/* EPT won't cause page fault directly */
4622 		WARN_ON_ONCE(!vcpu->arch.apf.host_apf_reason && enable_ept);
4623 		return kvm_handle_page_fault(vcpu, error_code, cr2, NULL, 0);
4624 	}
4625 
4626 	ex_no = intr_info & INTR_INFO_VECTOR_MASK;
4627 
4628 	if (vmx->rmode.vm86_active && rmode_exception(vcpu, ex_no))
4629 		return handle_rmode_exception(vcpu, ex_no, error_code);
4630 
4631 	switch (ex_no) {
4632 	case AC_VECTOR:
4633 		kvm_queue_exception_e(vcpu, AC_VECTOR, error_code);
4634 		return 1;
4635 	case DB_VECTOR:
4636 		dr6 = vmcs_readl(EXIT_QUALIFICATION);
4637 		if (!(vcpu->guest_debug &
4638 		      (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
4639 			vcpu->arch.dr6 &= ~DR_TRAP_BITS;
4640 			vcpu->arch.dr6 |= dr6 | DR6_RTM;
4641 			if (is_icebp(intr_info))
4642 				WARN_ON(!skip_emulated_instruction(vcpu));
4643 
4644 			kvm_queue_exception(vcpu, DB_VECTOR);
4645 			return 1;
4646 		}
4647 		kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
4648 		kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
4649 		/* fall through */
4650 	case BP_VECTOR:
4651 		/*
4652 		 * Update instruction length as we may reinject #BP from
4653 		 * user space while in guest debugging mode. Reading it for
4654 		 * #DB as well causes no harm, it is not used in that case.
4655 		 */
4656 		vmx->vcpu.arch.event_exit_inst_len =
4657 			vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4658 		kvm_run->exit_reason = KVM_EXIT_DEBUG;
4659 		rip = kvm_rip_read(vcpu);
4660 		kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
4661 		kvm_run->debug.arch.exception = ex_no;
4662 		break;
4663 	default:
4664 		kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
4665 		kvm_run->ex.exception = ex_no;
4666 		kvm_run->ex.error_code = error_code;
4667 		break;
4668 	}
4669 	return 0;
4670 }
4671 
4672 static __always_inline int handle_external_interrupt(struct kvm_vcpu *vcpu)
4673 {
4674 	++vcpu->stat.irq_exits;
4675 	return 1;
4676 }
4677 
4678 static int handle_triple_fault(struct kvm_vcpu *vcpu)
4679 {
4680 	vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
4681 	vcpu->mmio_needed = 0;
4682 	return 0;
4683 }
4684 
4685 static int handle_io(struct kvm_vcpu *vcpu)
4686 {
4687 	unsigned long exit_qualification;
4688 	int size, in, string;
4689 	unsigned port;
4690 
4691 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4692 	string = (exit_qualification & 16) != 0;
4693 
4694 	++vcpu->stat.io_exits;
4695 
4696 	if (string)
4697 		return kvm_emulate_instruction(vcpu, 0);
4698 
4699 	port = exit_qualification >> 16;
4700 	size = (exit_qualification & 7) + 1;
4701 	in = (exit_qualification & 8) != 0;
4702 
4703 	return kvm_fast_pio(vcpu, size, port, in);
4704 }
4705 
4706 static void
4707 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
4708 {
4709 	/*
4710 	 * Patch in the VMCALL instruction:
4711 	 */
4712 	hypercall[0] = 0x0f;
4713 	hypercall[1] = 0x01;
4714 	hypercall[2] = 0xc1;
4715 }
4716 
4717 /* called to set cr0 as appropriate for a mov-to-cr0 exit. */
4718 static int handle_set_cr0(struct kvm_vcpu *vcpu, unsigned long val)
4719 {
4720 	if (is_guest_mode(vcpu)) {
4721 		struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4722 		unsigned long orig_val = val;
4723 
4724 		/*
4725 		 * We get here when L2 changed cr0 in a way that did not change
4726 		 * any of L1's shadowed bits (see nested_vmx_exit_handled_cr),
4727 		 * but did change L0 shadowed bits. So we first calculate the
4728 		 * effective cr0 value that L1 would like to write into the
4729 		 * hardware. It consists of the L2-owned bits from the new
4730 		 * value combined with the L1-owned bits from L1's guest_cr0.
4731 		 */
4732 		val = (val & ~vmcs12->cr0_guest_host_mask) |
4733 			(vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask);
4734 
4735 		if (!nested_guest_cr0_valid(vcpu, val))
4736 			return 1;
4737 
4738 		if (kvm_set_cr0(vcpu, val))
4739 			return 1;
4740 		vmcs_writel(CR0_READ_SHADOW, orig_val);
4741 		return 0;
4742 	} else {
4743 		if (to_vmx(vcpu)->nested.vmxon &&
4744 		    !nested_host_cr0_valid(vcpu, val))
4745 			return 1;
4746 
4747 		return kvm_set_cr0(vcpu, val);
4748 	}
4749 }
4750 
4751 static int handle_set_cr4(struct kvm_vcpu *vcpu, unsigned long val)
4752 {
4753 	if (is_guest_mode(vcpu)) {
4754 		struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4755 		unsigned long orig_val = val;
4756 
4757 		/* analogously to handle_set_cr0 */
4758 		val = (val & ~vmcs12->cr4_guest_host_mask) |
4759 			(vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask);
4760 		if (kvm_set_cr4(vcpu, val))
4761 			return 1;
4762 		vmcs_writel(CR4_READ_SHADOW, orig_val);
4763 		return 0;
4764 	} else
4765 		return kvm_set_cr4(vcpu, val);
4766 }
4767 
4768 static int handle_desc(struct kvm_vcpu *vcpu)
4769 {
4770 	WARN_ON(!(vcpu->arch.cr4 & X86_CR4_UMIP));
4771 	return kvm_emulate_instruction(vcpu, 0);
4772 }
4773 
4774 static int handle_cr(struct kvm_vcpu *vcpu)
4775 {
4776 	unsigned long exit_qualification, val;
4777 	int cr;
4778 	int reg;
4779 	int err;
4780 	int ret;
4781 
4782 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4783 	cr = exit_qualification & 15;
4784 	reg = (exit_qualification >> 8) & 15;
4785 	switch ((exit_qualification >> 4) & 3) {
4786 	case 0: /* mov to cr */
4787 		val = kvm_register_readl(vcpu, reg);
4788 		trace_kvm_cr_write(cr, val);
4789 		switch (cr) {
4790 		case 0:
4791 			err = handle_set_cr0(vcpu, val);
4792 			return kvm_complete_insn_gp(vcpu, err);
4793 		case 3:
4794 			WARN_ON_ONCE(enable_unrestricted_guest);
4795 			err = kvm_set_cr3(vcpu, val);
4796 			return kvm_complete_insn_gp(vcpu, err);
4797 		case 4:
4798 			err = handle_set_cr4(vcpu, val);
4799 			return kvm_complete_insn_gp(vcpu, err);
4800 		case 8: {
4801 				u8 cr8_prev = kvm_get_cr8(vcpu);
4802 				u8 cr8 = (u8)val;
4803 				err = kvm_set_cr8(vcpu, cr8);
4804 				ret = kvm_complete_insn_gp(vcpu, err);
4805 				if (lapic_in_kernel(vcpu))
4806 					return ret;
4807 				if (cr8_prev <= cr8)
4808 					return ret;
4809 				/*
4810 				 * TODO: we might be squashing a
4811 				 * KVM_GUESTDBG_SINGLESTEP-triggered
4812 				 * KVM_EXIT_DEBUG here.
4813 				 */
4814 				vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
4815 				return 0;
4816 			}
4817 		}
4818 		break;
4819 	case 2: /* clts */
4820 		WARN_ONCE(1, "Guest should always own CR0.TS");
4821 		vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
4822 		trace_kvm_cr_write(0, kvm_read_cr0(vcpu));
4823 		return kvm_skip_emulated_instruction(vcpu);
4824 	case 1: /*mov from cr*/
4825 		switch (cr) {
4826 		case 3:
4827 			WARN_ON_ONCE(enable_unrestricted_guest);
4828 			val = kvm_read_cr3(vcpu);
4829 			kvm_register_write(vcpu, reg, val);
4830 			trace_kvm_cr_read(cr, val);
4831 			return kvm_skip_emulated_instruction(vcpu);
4832 		case 8:
4833 			val = kvm_get_cr8(vcpu);
4834 			kvm_register_write(vcpu, reg, val);
4835 			trace_kvm_cr_read(cr, val);
4836 			return kvm_skip_emulated_instruction(vcpu);
4837 		}
4838 		break;
4839 	case 3: /* lmsw */
4840 		val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
4841 		trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val);
4842 		kvm_lmsw(vcpu, val);
4843 
4844 		return kvm_skip_emulated_instruction(vcpu);
4845 	default:
4846 		break;
4847 	}
4848 	vcpu->run->exit_reason = 0;
4849 	vcpu_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
4850 	       (int)(exit_qualification >> 4) & 3, cr);
4851 	return 0;
4852 }
4853 
4854 static int handle_dr(struct kvm_vcpu *vcpu)
4855 {
4856 	unsigned long exit_qualification;
4857 	int dr, dr7, reg;
4858 
4859 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4860 	dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
4861 
4862 	/* First, if DR does not exist, trigger UD */
4863 	if (!kvm_require_dr(vcpu, dr))
4864 		return 1;
4865 
4866 	/* Do not handle if the CPL > 0, will trigger GP on re-entry */
4867 	if (!kvm_require_cpl(vcpu, 0))
4868 		return 1;
4869 	dr7 = vmcs_readl(GUEST_DR7);
4870 	if (dr7 & DR7_GD) {
4871 		/*
4872 		 * As the vm-exit takes precedence over the debug trap, we
4873 		 * need to emulate the latter, either for the host or the
4874 		 * guest debugging itself.
4875 		 */
4876 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
4877 			vcpu->run->debug.arch.dr6 = vcpu->arch.dr6;
4878 			vcpu->run->debug.arch.dr7 = dr7;
4879 			vcpu->run->debug.arch.pc = kvm_get_linear_rip(vcpu);
4880 			vcpu->run->debug.arch.exception = DB_VECTOR;
4881 			vcpu->run->exit_reason = KVM_EXIT_DEBUG;
4882 			return 0;
4883 		} else {
4884 			vcpu->arch.dr6 &= ~DR_TRAP_BITS;
4885 			vcpu->arch.dr6 |= DR6_BD | DR6_RTM;
4886 			kvm_queue_exception(vcpu, DB_VECTOR);
4887 			return 1;
4888 		}
4889 	}
4890 
4891 	if (vcpu->guest_debug == 0) {
4892 		exec_controls_clearbit(to_vmx(vcpu), CPU_BASED_MOV_DR_EXITING);
4893 
4894 		/*
4895 		 * No more DR vmexits; force a reload of the debug registers
4896 		 * and reenter on this instruction.  The next vmexit will
4897 		 * retrieve the full state of the debug registers.
4898 		 */
4899 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
4900 		return 1;
4901 	}
4902 
4903 	reg = DEBUG_REG_ACCESS_REG(exit_qualification);
4904 	if (exit_qualification & TYPE_MOV_FROM_DR) {
4905 		unsigned long val;
4906 
4907 		if (kvm_get_dr(vcpu, dr, &val))
4908 			return 1;
4909 		kvm_register_write(vcpu, reg, val);
4910 	} else
4911 		if (kvm_set_dr(vcpu, dr, kvm_register_readl(vcpu, reg)))
4912 			return 1;
4913 
4914 	return kvm_skip_emulated_instruction(vcpu);
4915 }
4916 
4917 static u64 vmx_get_dr6(struct kvm_vcpu *vcpu)
4918 {
4919 	return vcpu->arch.dr6;
4920 }
4921 
4922 static void vmx_set_dr6(struct kvm_vcpu *vcpu, unsigned long val)
4923 {
4924 }
4925 
4926 static void vmx_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
4927 {
4928 	get_debugreg(vcpu->arch.db[0], 0);
4929 	get_debugreg(vcpu->arch.db[1], 1);
4930 	get_debugreg(vcpu->arch.db[2], 2);
4931 	get_debugreg(vcpu->arch.db[3], 3);
4932 	get_debugreg(vcpu->arch.dr6, 6);
4933 	vcpu->arch.dr7 = vmcs_readl(GUEST_DR7);
4934 
4935 	vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
4936 	exec_controls_setbit(to_vmx(vcpu), CPU_BASED_MOV_DR_EXITING);
4937 }
4938 
4939 static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
4940 {
4941 	vmcs_writel(GUEST_DR7, val);
4942 }
4943 
4944 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
4945 {
4946 	kvm_apic_update_ppr(vcpu);
4947 	return 1;
4948 }
4949 
4950 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
4951 {
4952 	exec_controls_clearbit(to_vmx(vcpu), CPU_BASED_INTR_WINDOW_EXITING);
4953 
4954 	kvm_make_request(KVM_REQ_EVENT, vcpu);
4955 
4956 	++vcpu->stat.irq_window_exits;
4957 	return 1;
4958 }
4959 
4960 static int handle_vmcall(struct kvm_vcpu *vcpu)
4961 {
4962 	return kvm_emulate_hypercall(vcpu);
4963 }
4964 
4965 static int handle_invd(struct kvm_vcpu *vcpu)
4966 {
4967 	return kvm_emulate_instruction(vcpu, 0);
4968 }
4969 
4970 static int handle_invlpg(struct kvm_vcpu *vcpu)
4971 {
4972 	unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4973 
4974 	kvm_mmu_invlpg(vcpu, exit_qualification);
4975 	return kvm_skip_emulated_instruction(vcpu);
4976 }
4977 
4978 static int handle_rdpmc(struct kvm_vcpu *vcpu)
4979 {
4980 	int err;
4981 
4982 	err = kvm_rdpmc(vcpu);
4983 	return kvm_complete_insn_gp(vcpu, err);
4984 }
4985 
4986 static int handle_wbinvd(struct kvm_vcpu *vcpu)
4987 {
4988 	return kvm_emulate_wbinvd(vcpu);
4989 }
4990 
4991 static int handle_xsetbv(struct kvm_vcpu *vcpu)
4992 {
4993 	u64 new_bv = kvm_read_edx_eax(vcpu);
4994 	u32 index = kvm_rcx_read(vcpu);
4995 
4996 	if (kvm_set_xcr(vcpu, index, new_bv) == 0)
4997 		return kvm_skip_emulated_instruction(vcpu);
4998 	return 1;
4999 }
5000 
5001 static int handle_apic_access(struct kvm_vcpu *vcpu)
5002 {
5003 	if (likely(fasteoi)) {
5004 		unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5005 		int access_type, offset;
5006 
5007 		access_type = exit_qualification & APIC_ACCESS_TYPE;
5008 		offset = exit_qualification & APIC_ACCESS_OFFSET;
5009 		/*
5010 		 * Sane guest uses MOV to write EOI, with written value
5011 		 * not cared. So make a short-circuit here by avoiding
5012 		 * heavy instruction emulation.
5013 		 */
5014 		if ((access_type == TYPE_LINEAR_APIC_INST_WRITE) &&
5015 		    (offset == APIC_EOI)) {
5016 			kvm_lapic_set_eoi(vcpu);
5017 			return kvm_skip_emulated_instruction(vcpu);
5018 		}
5019 	}
5020 	return kvm_emulate_instruction(vcpu, 0);
5021 }
5022 
5023 static int handle_apic_eoi_induced(struct kvm_vcpu *vcpu)
5024 {
5025 	unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5026 	int vector = exit_qualification & 0xff;
5027 
5028 	/* EOI-induced VM exit is trap-like and thus no need to adjust IP */
5029 	kvm_apic_set_eoi_accelerated(vcpu, vector);
5030 	return 1;
5031 }
5032 
5033 static int handle_apic_write(struct kvm_vcpu *vcpu)
5034 {
5035 	unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5036 	u32 offset = exit_qualification & 0xfff;
5037 
5038 	/* APIC-write VM exit is trap-like and thus no need to adjust IP */
5039 	kvm_apic_write_nodecode(vcpu, offset);
5040 	return 1;
5041 }
5042 
5043 static int handle_task_switch(struct kvm_vcpu *vcpu)
5044 {
5045 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5046 	unsigned long exit_qualification;
5047 	bool has_error_code = false;
5048 	u32 error_code = 0;
5049 	u16 tss_selector;
5050 	int reason, type, idt_v, idt_index;
5051 
5052 	idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
5053 	idt_index = (vmx->idt_vectoring_info & VECTORING_INFO_VECTOR_MASK);
5054 	type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
5055 
5056 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5057 
5058 	reason = (u32)exit_qualification >> 30;
5059 	if (reason == TASK_SWITCH_GATE && idt_v) {
5060 		switch (type) {
5061 		case INTR_TYPE_NMI_INTR:
5062 			vcpu->arch.nmi_injected = false;
5063 			vmx_set_nmi_mask(vcpu, true);
5064 			break;
5065 		case INTR_TYPE_EXT_INTR:
5066 		case INTR_TYPE_SOFT_INTR:
5067 			kvm_clear_interrupt_queue(vcpu);
5068 			break;
5069 		case INTR_TYPE_HARD_EXCEPTION:
5070 			if (vmx->idt_vectoring_info &
5071 			    VECTORING_INFO_DELIVER_CODE_MASK) {
5072 				has_error_code = true;
5073 				error_code =
5074 					vmcs_read32(IDT_VECTORING_ERROR_CODE);
5075 			}
5076 			/* fall through */
5077 		case INTR_TYPE_SOFT_EXCEPTION:
5078 			kvm_clear_exception_queue(vcpu);
5079 			break;
5080 		default:
5081 			break;
5082 		}
5083 	}
5084 	tss_selector = exit_qualification;
5085 
5086 	if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
5087 		       type != INTR_TYPE_EXT_INTR &&
5088 		       type != INTR_TYPE_NMI_INTR))
5089 		WARN_ON(!skip_emulated_instruction(vcpu));
5090 
5091 	/*
5092 	 * TODO: What about debug traps on tss switch?
5093 	 *       Are we supposed to inject them and update dr6?
5094 	 */
5095 	return kvm_task_switch(vcpu, tss_selector,
5096 			       type == INTR_TYPE_SOFT_INTR ? idt_index : -1,
5097 			       reason, has_error_code, error_code);
5098 }
5099 
5100 static int handle_ept_violation(struct kvm_vcpu *vcpu)
5101 {
5102 	unsigned long exit_qualification;
5103 	gpa_t gpa;
5104 	u64 error_code;
5105 
5106 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5107 
5108 	/*
5109 	 * EPT violation happened while executing iret from NMI,
5110 	 * "blocked by NMI" bit has to be set before next VM entry.
5111 	 * There are errata that may cause this bit to not be set:
5112 	 * AAK134, BY25.
5113 	 */
5114 	if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
5115 			enable_vnmi &&
5116 			(exit_qualification & INTR_INFO_UNBLOCK_NMI))
5117 		vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO, GUEST_INTR_STATE_NMI);
5118 
5119 	gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5120 	trace_kvm_page_fault(gpa, exit_qualification);
5121 
5122 	/* Is it a read fault? */
5123 	error_code = (exit_qualification & EPT_VIOLATION_ACC_READ)
5124 		     ? PFERR_USER_MASK : 0;
5125 	/* Is it a write fault? */
5126 	error_code |= (exit_qualification & EPT_VIOLATION_ACC_WRITE)
5127 		      ? PFERR_WRITE_MASK : 0;
5128 	/* Is it a fetch fault? */
5129 	error_code |= (exit_qualification & EPT_VIOLATION_ACC_INSTR)
5130 		      ? PFERR_FETCH_MASK : 0;
5131 	/* ept page table entry is present? */
5132 	error_code |= (exit_qualification &
5133 		       (EPT_VIOLATION_READABLE | EPT_VIOLATION_WRITABLE |
5134 			EPT_VIOLATION_EXECUTABLE))
5135 		      ? PFERR_PRESENT_MASK : 0;
5136 
5137 	error_code |= (exit_qualification & 0x100) != 0 ?
5138 	       PFERR_GUEST_FINAL_MASK : PFERR_GUEST_PAGE_MASK;
5139 
5140 	vcpu->arch.exit_qualification = exit_qualification;
5141 	return kvm_mmu_page_fault(vcpu, gpa, error_code, NULL, 0);
5142 }
5143 
5144 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
5145 {
5146 	gpa_t gpa;
5147 
5148 	/*
5149 	 * A nested guest cannot optimize MMIO vmexits, because we have an
5150 	 * nGPA here instead of the required GPA.
5151 	 */
5152 	gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5153 	if (!is_guest_mode(vcpu) &&
5154 	    !kvm_io_bus_write(vcpu, KVM_FAST_MMIO_BUS, gpa, 0, NULL)) {
5155 		trace_kvm_fast_mmio(gpa);
5156 		return kvm_skip_emulated_instruction(vcpu);
5157 	}
5158 
5159 	return kvm_mmu_page_fault(vcpu, gpa, PFERR_RSVD_MASK, NULL, 0);
5160 }
5161 
5162 static int handle_nmi_window(struct kvm_vcpu *vcpu)
5163 {
5164 	WARN_ON_ONCE(!enable_vnmi);
5165 	exec_controls_clearbit(to_vmx(vcpu), CPU_BASED_NMI_WINDOW_EXITING);
5166 	++vcpu->stat.nmi_window_exits;
5167 	kvm_make_request(KVM_REQ_EVENT, vcpu);
5168 
5169 	return 1;
5170 }
5171 
5172 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
5173 {
5174 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5175 	bool intr_window_requested;
5176 	unsigned count = 130;
5177 
5178 	/*
5179 	 * We should never reach the point where we are emulating L2
5180 	 * due to invalid guest state as that means we incorrectly
5181 	 * allowed a nested VMEntry with an invalid vmcs12.
5182 	 */
5183 	WARN_ON_ONCE(vmx->emulation_required && vmx->nested.nested_run_pending);
5184 
5185 	intr_window_requested = exec_controls_get(vmx) &
5186 				CPU_BASED_INTR_WINDOW_EXITING;
5187 
5188 	while (vmx->emulation_required && count-- != 0) {
5189 		if (intr_window_requested && vmx_interrupt_allowed(vcpu))
5190 			return handle_interrupt_window(&vmx->vcpu);
5191 
5192 		if (kvm_test_request(KVM_REQ_EVENT, vcpu))
5193 			return 1;
5194 
5195 		if (!kvm_emulate_instruction(vcpu, 0))
5196 			return 0;
5197 
5198 		if (vmx->emulation_required && !vmx->rmode.vm86_active &&
5199 		    vcpu->arch.exception.pending) {
5200 			vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5201 			vcpu->run->internal.suberror =
5202 						KVM_INTERNAL_ERROR_EMULATION;
5203 			vcpu->run->internal.ndata = 0;
5204 			return 0;
5205 		}
5206 
5207 		if (vcpu->arch.halt_request) {
5208 			vcpu->arch.halt_request = 0;
5209 			return kvm_vcpu_halt(vcpu);
5210 		}
5211 
5212 		/*
5213 		 * Note, return 1 and not 0, vcpu_run() is responsible for
5214 		 * morphing the pending signal into the proper return code.
5215 		 */
5216 		if (signal_pending(current))
5217 			return 1;
5218 
5219 		if (need_resched())
5220 			schedule();
5221 	}
5222 
5223 	return 1;
5224 }
5225 
5226 static void grow_ple_window(struct kvm_vcpu *vcpu)
5227 {
5228 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5229 	unsigned int old = vmx->ple_window;
5230 
5231 	vmx->ple_window = __grow_ple_window(old, ple_window,
5232 					    ple_window_grow,
5233 					    ple_window_max);
5234 
5235 	if (vmx->ple_window != old) {
5236 		vmx->ple_window_dirty = true;
5237 		trace_kvm_ple_window_update(vcpu->vcpu_id,
5238 					    vmx->ple_window, old);
5239 	}
5240 }
5241 
5242 static void shrink_ple_window(struct kvm_vcpu *vcpu)
5243 {
5244 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5245 	unsigned int old = vmx->ple_window;
5246 
5247 	vmx->ple_window = __shrink_ple_window(old, ple_window,
5248 					      ple_window_shrink,
5249 					      ple_window);
5250 
5251 	if (vmx->ple_window != old) {
5252 		vmx->ple_window_dirty = true;
5253 		trace_kvm_ple_window_update(vcpu->vcpu_id,
5254 					    vmx->ple_window, old);
5255 	}
5256 }
5257 
5258 /*
5259  * Handler for POSTED_INTERRUPT_WAKEUP_VECTOR.
5260  */
5261 static void wakeup_handler(void)
5262 {
5263 	struct kvm_vcpu *vcpu;
5264 	int cpu = smp_processor_id();
5265 
5266 	spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
5267 	list_for_each_entry(vcpu, &per_cpu(blocked_vcpu_on_cpu, cpu),
5268 			blocked_vcpu_list) {
5269 		struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
5270 
5271 		if (pi_test_on(pi_desc) == 1)
5272 			kvm_vcpu_kick(vcpu);
5273 	}
5274 	spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
5275 }
5276 
5277 static void vmx_enable_tdp(void)
5278 {
5279 	kvm_mmu_set_mask_ptes(VMX_EPT_READABLE_MASK,
5280 		enable_ept_ad_bits ? VMX_EPT_ACCESS_BIT : 0ull,
5281 		enable_ept_ad_bits ? VMX_EPT_DIRTY_BIT : 0ull,
5282 		0ull, VMX_EPT_EXECUTABLE_MASK,
5283 		cpu_has_vmx_ept_execute_only() ? 0ull : VMX_EPT_READABLE_MASK,
5284 		VMX_EPT_RWX_MASK, 0ull);
5285 
5286 	ept_set_mmio_spte_mask();
5287 	kvm_enable_tdp();
5288 }
5289 
5290 /*
5291  * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
5292  * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
5293  */
5294 static int handle_pause(struct kvm_vcpu *vcpu)
5295 {
5296 	if (!kvm_pause_in_guest(vcpu->kvm))
5297 		grow_ple_window(vcpu);
5298 
5299 	/*
5300 	 * Intel sdm vol3 ch-25.1.3 says: The "PAUSE-loop exiting"
5301 	 * VM-execution control is ignored if CPL > 0. OTOH, KVM
5302 	 * never set PAUSE_EXITING and just set PLE if supported,
5303 	 * so the vcpu must be CPL=0 if it gets a PAUSE exit.
5304 	 */
5305 	kvm_vcpu_on_spin(vcpu, true);
5306 	return kvm_skip_emulated_instruction(vcpu);
5307 }
5308 
5309 static int handle_nop(struct kvm_vcpu *vcpu)
5310 {
5311 	return kvm_skip_emulated_instruction(vcpu);
5312 }
5313 
5314 static int handle_mwait(struct kvm_vcpu *vcpu)
5315 {
5316 	printk_once(KERN_WARNING "kvm: MWAIT instruction emulated as NOP!\n");
5317 	return handle_nop(vcpu);
5318 }
5319 
5320 static int handle_invalid_op(struct kvm_vcpu *vcpu)
5321 {
5322 	kvm_queue_exception(vcpu, UD_VECTOR);
5323 	return 1;
5324 }
5325 
5326 static int handle_monitor_trap(struct kvm_vcpu *vcpu)
5327 {
5328 	return 1;
5329 }
5330 
5331 static int handle_monitor(struct kvm_vcpu *vcpu)
5332 {
5333 	printk_once(KERN_WARNING "kvm: MONITOR instruction emulated as NOP!\n");
5334 	return handle_nop(vcpu);
5335 }
5336 
5337 static int handle_invpcid(struct kvm_vcpu *vcpu)
5338 {
5339 	u32 vmx_instruction_info;
5340 	unsigned long type;
5341 	bool pcid_enabled;
5342 	gva_t gva;
5343 	struct x86_exception e;
5344 	unsigned i;
5345 	unsigned long roots_to_free = 0;
5346 	struct {
5347 		u64 pcid;
5348 		u64 gla;
5349 	} operand;
5350 
5351 	if (!guest_cpuid_has(vcpu, X86_FEATURE_INVPCID)) {
5352 		kvm_queue_exception(vcpu, UD_VECTOR);
5353 		return 1;
5354 	}
5355 
5356 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5357 	type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
5358 
5359 	if (type > 3) {
5360 		kvm_inject_gp(vcpu, 0);
5361 		return 1;
5362 	}
5363 
5364 	/* According to the Intel instruction reference, the memory operand
5365 	 * is read even if it isn't needed (e.g., for type==all)
5366 	 */
5367 	if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
5368 				vmx_instruction_info, false,
5369 				sizeof(operand), &gva))
5370 		return 1;
5371 
5372 	if (kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e)) {
5373 		kvm_inject_page_fault(vcpu, &e);
5374 		return 1;
5375 	}
5376 
5377 	if (operand.pcid >> 12 != 0) {
5378 		kvm_inject_gp(vcpu, 0);
5379 		return 1;
5380 	}
5381 
5382 	pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
5383 
5384 	switch (type) {
5385 	case INVPCID_TYPE_INDIV_ADDR:
5386 		if ((!pcid_enabled && (operand.pcid != 0)) ||
5387 		    is_noncanonical_address(operand.gla, vcpu)) {
5388 			kvm_inject_gp(vcpu, 0);
5389 			return 1;
5390 		}
5391 		kvm_mmu_invpcid_gva(vcpu, operand.gla, operand.pcid);
5392 		return kvm_skip_emulated_instruction(vcpu);
5393 
5394 	case INVPCID_TYPE_SINGLE_CTXT:
5395 		if (!pcid_enabled && (operand.pcid != 0)) {
5396 			kvm_inject_gp(vcpu, 0);
5397 			return 1;
5398 		}
5399 
5400 		if (kvm_get_active_pcid(vcpu) == operand.pcid) {
5401 			kvm_mmu_sync_roots(vcpu);
5402 			kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
5403 		}
5404 
5405 		for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++)
5406 			if (kvm_get_pcid(vcpu, vcpu->arch.mmu->prev_roots[i].cr3)
5407 			    == operand.pcid)
5408 				roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i);
5409 
5410 		kvm_mmu_free_roots(vcpu, vcpu->arch.mmu, roots_to_free);
5411 		/*
5412 		 * If neither the current cr3 nor any of the prev_roots use the
5413 		 * given PCID, then nothing needs to be done here because a
5414 		 * resync will happen anyway before switching to any other CR3.
5415 		 */
5416 
5417 		return kvm_skip_emulated_instruction(vcpu);
5418 
5419 	case INVPCID_TYPE_ALL_NON_GLOBAL:
5420 		/*
5421 		 * Currently, KVM doesn't mark global entries in the shadow
5422 		 * page tables, so a non-global flush just degenerates to a
5423 		 * global flush. If needed, we could optimize this later by
5424 		 * keeping track of global entries in shadow page tables.
5425 		 */
5426 
5427 		/* fall-through */
5428 	case INVPCID_TYPE_ALL_INCL_GLOBAL:
5429 		kvm_mmu_unload(vcpu);
5430 		return kvm_skip_emulated_instruction(vcpu);
5431 
5432 	default:
5433 		BUG(); /* We have already checked above that type <= 3 */
5434 	}
5435 }
5436 
5437 static int handle_pml_full(struct kvm_vcpu *vcpu)
5438 {
5439 	unsigned long exit_qualification;
5440 
5441 	trace_kvm_pml_full(vcpu->vcpu_id);
5442 
5443 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5444 
5445 	/*
5446 	 * PML buffer FULL happened while executing iret from NMI,
5447 	 * "blocked by NMI" bit has to be set before next VM entry.
5448 	 */
5449 	if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
5450 			enable_vnmi &&
5451 			(exit_qualification & INTR_INFO_UNBLOCK_NMI))
5452 		vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
5453 				GUEST_INTR_STATE_NMI);
5454 
5455 	/*
5456 	 * PML buffer already flushed at beginning of VMEXIT. Nothing to do
5457 	 * here.., and there's no userspace involvement needed for PML.
5458 	 */
5459 	return 1;
5460 }
5461 
5462 static int handle_preemption_timer(struct kvm_vcpu *vcpu)
5463 {
5464 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5465 
5466 	if (!vmx->req_immediate_exit &&
5467 	    !unlikely(vmx->loaded_vmcs->hv_timer_soft_disabled))
5468 		kvm_lapic_expired_hv_timer(vcpu);
5469 
5470 	return 1;
5471 }
5472 
5473 /*
5474  * When nested=0, all VMX instruction VM Exits filter here.  The handlers
5475  * are overwritten by nested_vmx_setup() when nested=1.
5476  */
5477 static int handle_vmx_instruction(struct kvm_vcpu *vcpu)
5478 {
5479 	kvm_queue_exception(vcpu, UD_VECTOR);
5480 	return 1;
5481 }
5482 
5483 static int handle_encls(struct kvm_vcpu *vcpu)
5484 {
5485 	/*
5486 	 * SGX virtualization is not yet supported.  There is no software
5487 	 * enable bit for SGX, so we have to trap ENCLS and inject a #UD
5488 	 * to prevent the guest from executing ENCLS.
5489 	 */
5490 	kvm_queue_exception(vcpu, UD_VECTOR);
5491 	return 1;
5492 }
5493 
5494 /*
5495  * The exit handlers return 1 if the exit was handled fully and guest execution
5496  * may resume.  Otherwise they set the kvm_run parameter to indicate what needs
5497  * to be done to userspace and return 0.
5498  */
5499 static int (*kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
5500 	[EXIT_REASON_EXCEPTION_NMI]           = handle_exception_nmi,
5501 	[EXIT_REASON_EXTERNAL_INTERRUPT]      = handle_external_interrupt,
5502 	[EXIT_REASON_TRIPLE_FAULT]            = handle_triple_fault,
5503 	[EXIT_REASON_NMI_WINDOW]	      = handle_nmi_window,
5504 	[EXIT_REASON_IO_INSTRUCTION]          = handle_io,
5505 	[EXIT_REASON_CR_ACCESS]               = handle_cr,
5506 	[EXIT_REASON_DR_ACCESS]               = handle_dr,
5507 	[EXIT_REASON_CPUID]                   = kvm_emulate_cpuid,
5508 	[EXIT_REASON_MSR_READ]                = kvm_emulate_rdmsr,
5509 	[EXIT_REASON_MSR_WRITE]               = kvm_emulate_wrmsr,
5510 	[EXIT_REASON_INTERRUPT_WINDOW]        = handle_interrupt_window,
5511 	[EXIT_REASON_HLT]                     = kvm_emulate_halt,
5512 	[EXIT_REASON_INVD]		      = handle_invd,
5513 	[EXIT_REASON_INVLPG]		      = handle_invlpg,
5514 	[EXIT_REASON_RDPMC]                   = handle_rdpmc,
5515 	[EXIT_REASON_VMCALL]                  = handle_vmcall,
5516 	[EXIT_REASON_VMCLEAR]		      = handle_vmx_instruction,
5517 	[EXIT_REASON_VMLAUNCH]		      = handle_vmx_instruction,
5518 	[EXIT_REASON_VMPTRLD]		      = handle_vmx_instruction,
5519 	[EXIT_REASON_VMPTRST]		      = handle_vmx_instruction,
5520 	[EXIT_REASON_VMREAD]		      = handle_vmx_instruction,
5521 	[EXIT_REASON_VMRESUME]		      = handle_vmx_instruction,
5522 	[EXIT_REASON_VMWRITE]		      = handle_vmx_instruction,
5523 	[EXIT_REASON_VMOFF]		      = handle_vmx_instruction,
5524 	[EXIT_REASON_VMON]		      = handle_vmx_instruction,
5525 	[EXIT_REASON_TPR_BELOW_THRESHOLD]     = handle_tpr_below_threshold,
5526 	[EXIT_REASON_APIC_ACCESS]             = handle_apic_access,
5527 	[EXIT_REASON_APIC_WRITE]              = handle_apic_write,
5528 	[EXIT_REASON_EOI_INDUCED]             = handle_apic_eoi_induced,
5529 	[EXIT_REASON_WBINVD]                  = handle_wbinvd,
5530 	[EXIT_REASON_XSETBV]                  = handle_xsetbv,
5531 	[EXIT_REASON_TASK_SWITCH]             = handle_task_switch,
5532 	[EXIT_REASON_MCE_DURING_VMENTRY]      = handle_machine_check,
5533 	[EXIT_REASON_GDTR_IDTR]		      = handle_desc,
5534 	[EXIT_REASON_LDTR_TR]		      = handle_desc,
5535 	[EXIT_REASON_EPT_VIOLATION]	      = handle_ept_violation,
5536 	[EXIT_REASON_EPT_MISCONFIG]           = handle_ept_misconfig,
5537 	[EXIT_REASON_PAUSE_INSTRUCTION]       = handle_pause,
5538 	[EXIT_REASON_MWAIT_INSTRUCTION]	      = handle_mwait,
5539 	[EXIT_REASON_MONITOR_TRAP_FLAG]       = handle_monitor_trap,
5540 	[EXIT_REASON_MONITOR_INSTRUCTION]     = handle_monitor,
5541 	[EXIT_REASON_INVEPT]                  = handle_vmx_instruction,
5542 	[EXIT_REASON_INVVPID]                 = handle_vmx_instruction,
5543 	[EXIT_REASON_RDRAND]                  = handle_invalid_op,
5544 	[EXIT_REASON_RDSEED]                  = handle_invalid_op,
5545 	[EXIT_REASON_PML_FULL]		      = handle_pml_full,
5546 	[EXIT_REASON_INVPCID]                 = handle_invpcid,
5547 	[EXIT_REASON_VMFUNC]		      = handle_vmx_instruction,
5548 	[EXIT_REASON_PREEMPTION_TIMER]	      = handle_preemption_timer,
5549 	[EXIT_REASON_ENCLS]		      = handle_encls,
5550 };
5551 
5552 static const int kvm_vmx_max_exit_handlers =
5553 	ARRAY_SIZE(kvm_vmx_exit_handlers);
5554 
5555 static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
5556 {
5557 	*info1 = vmcs_readl(EXIT_QUALIFICATION);
5558 	*info2 = vmcs_read32(VM_EXIT_INTR_INFO);
5559 }
5560 
5561 static void vmx_destroy_pml_buffer(struct vcpu_vmx *vmx)
5562 {
5563 	if (vmx->pml_pg) {
5564 		__free_page(vmx->pml_pg);
5565 		vmx->pml_pg = NULL;
5566 	}
5567 }
5568 
5569 static void vmx_flush_pml_buffer(struct kvm_vcpu *vcpu)
5570 {
5571 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5572 	u64 *pml_buf;
5573 	u16 pml_idx;
5574 
5575 	pml_idx = vmcs_read16(GUEST_PML_INDEX);
5576 
5577 	/* Do nothing if PML buffer is empty */
5578 	if (pml_idx == (PML_ENTITY_NUM - 1))
5579 		return;
5580 
5581 	/* PML index always points to next available PML buffer entity */
5582 	if (pml_idx >= PML_ENTITY_NUM)
5583 		pml_idx = 0;
5584 	else
5585 		pml_idx++;
5586 
5587 	pml_buf = page_address(vmx->pml_pg);
5588 	for (; pml_idx < PML_ENTITY_NUM; pml_idx++) {
5589 		u64 gpa;
5590 
5591 		gpa = pml_buf[pml_idx];
5592 		WARN_ON(gpa & (PAGE_SIZE - 1));
5593 		kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
5594 	}
5595 
5596 	/* reset PML index */
5597 	vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
5598 }
5599 
5600 /*
5601  * Flush all vcpus' PML buffer and update logged GPAs to dirty_bitmap.
5602  * Called before reporting dirty_bitmap to userspace.
5603  */
5604 static void kvm_flush_pml_buffers(struct kvm *kvm)
5605 {
5606 	int i;
5607 	struct kvm_vcpu *vcpu;
5608 	/*
5609 	 * We only need to kick vcpu out of guest mode here, as PML buffer
5610 	 * is flushed at beginning of all VMEXITs, and it's obvious that only
5611 	 * vcpus running in guest are possible to have unflushed GPAs in PML
5612 	 * buffer.
5613 	 */
5614 	kvm_for_each_vcpu(i, vcpu, kvm)
5615 		kvm_vcpu_kick(vcpu);
5616 }
5617 
5618 static void vmx_dump_sel(char *name, uint32_t sel)
5619 {
5620 	pr_err("%s sel=0x%04x, attr=0x%05x, limit=0x%08x, base=0x%016lx\n",
5621 	       name, vmcs_read16(sel),
5622 	       vmcs_read32(sel + GUEST_ES_AR_BYTES - GUEST_ES_SELECTOR),
5623 	       vmcs_read32(sel + GUEST_ES_LIMIT - GUEST_ES_SELECTOR),
5624 	       vmcs_readl(sel + GUEST_ES_BASE - GUEST_ES_SELECTOR));
5625 }
5626 
5627 static void vmx_dump_dtsel(char *name, uint32_t limit)
5628 {
5629 	pr_err("%s                           limit=0x%08x, base=0x%016lx\n",
5630 	       name, vmcs_read32(limit),
5631 	       vmcs_readl(limit + GUEST_GDTR_BASE - GUEST_GDTR_LIMIT));
5632 }
5633 
5634 void dump_vmcs(void)
5635 {
5636 	u32 vmentry_ctl, vmexit_ctl;
5637 	u32 cpu_based_exec_ctrl, pin_based_exec_ctrl, secondary_exec_control;
5638 	unsigned long cr4;
5639 	u64 efer;
5640 	int i, n;
5641 
5642 	if (!dump_invalid_vmcs) {
5643 		pr_warn_ratelimited("set kvm_intel.dump_invalid_vmcs=1 to dump internal KVM state.\n");
5644 		return;
5645 	}
5646 
5647 	vmentry_ctl = vmcs_read32(VM_ENTRY_CONTROLS);
5648 	vmexit_ctl = vmcs_read32(VM_EXIT_CONTROLS);
5649 	cpu_based_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5650 	pin_based_exec_ctrl = vmcs_read32(PIN_BASED_VM_EXEC_CONTROL);
5651 	cr4 = vmcs_readl(GUEST_CR4);
5652 	efer = vmcs_read64(GUEST_IA32_EFER);
5653 	secondary_exec_control = 0;
5654 	if (cpu_has_secondary_exec_ctrls())
5655 		secondary_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
5656 
5657 	pr_err("*** Guest State ***\n");
5658 	pr_err("CR0: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
5659 	       vmcs_readl(GUEST_CR0), vmcs_readl(CR0_READ_SHADOW),
5660 	       vmcs_readl(CR0_GUEST_HOST_MASK));
5661 	pr_err("CR4: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
5662 	       cr4, vmcs_readl(CR4_READ_SHADOW), vmcs_readl(CR4_GUEST_HOST_MASK));
5663 	pr_err("CR3 = 0x%016lx\n", vmcs_readl(GUEST_CR3));
5664 	if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT) &&
5665 	    (cr4 & X86_CR4_PAE) && !(efer & EFER_LMA))
5666 	{
5667 		pr_err("PDPTR0 = 0x%016llx  PDPTR1 = 0x%016llx\n",
5668 		       vmcs_read64(GUEST_PDPTR0), vmcs_read64(GUEST_PDPTR1));
5669 		pr_err("PDPTR2 = 0x%016llx  PDPTR3 = 0x%016llx\n",
5670 		       vmcs_read64(GUEST_PDPTR2), vmcs_read64(GUEST_PDPTR3));
5671 	}
5672 	pr_err("RSP = 0x%016lx  RIP = 0x%016lx\n",
5673 	       vmcs_readl(GUEST_RSP), vmcs_readl(GUEST_RIP));
5674 	pr_err("RFLAGS=0x%08lx         DR7 = 0x%016lx\n",
5675 	       vmcs_readl(GUEST_RFLAGS), vmcs_readl(GUEST_DR7));
5676 	pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
5677 	       vmcs_readl(GUEST_SYSENTER_ESP),
5678 	       vmcs_read32(GUEST_SYSENTER_CS), vmcs_readl(GUEST_SYSENTER_EIP));
5679 	vmx_dump_sel("CS:  ", GUEST_CS_SELECTOR);
5680 	vmx_dump_sel("DS:  ", GUEST_DS_SELECTOR);
5681 	vmx_dump_sel("SS:  ", GUEST_SS_SELECTOR);
5682 	vmx_dump_sel("ES:  ", GUEST_ES_SELECTOR);
5683 	vmx_dump_sel("FS:  ", GUEST_FS_SELECTOR);
5684 	vmx_dump_sel("GS:  ", GUEST_GS_SELECTOR);
5685 	vmx_dump_dtsel("GDTR:", GUEST_GDTR_LIMIT);
5686 	vmx_dump_sel("LDTR:", GUEST_LDTR_SELECTOR);
5687 	vmx_dump_dtsel("IDTR:", GUEST_IDTR_LIMIT);
5688 	vmx_dump_sel("TR:  ", GUEST_TR_SELECTOR);
5689 	if ((vmexit_ctl & (VM_EXIT_SAVE_IA32_PAT | VM_EXIT_SAVE_IA32_EFER)) ||
5690 	    (vmentry_ctl & (VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_IA32_EFER)))
5691 		pr_err("EFER =     0x%016llx  PAT = 0x%016llx\n",
5692 		       efer, vmcs_read64(GUEST_IA32_PAT));
5693 	pr_err("DebugCtl = 0x%016llx  DebugExceptions = 0x%016lx\n",
5694 	       vmcs_read64(GUEST_IA32_DEBUGCTL),
5695 	       vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS));
5696 	if (cpu_has_load_perf_global_ctrl() &&
5697 	    vmentry_ctl & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
5698 		pr_err("PerfGlobCtl = 0x%016llx\n",
5699 		       vmcs_read64(GUEST_IA32_PERF_GLOBAL_CTRL));
5700 	if (vmentry_ctl & VM_ENTRY_LOAD_BNDCFGS)
5701 		pr_err("BndCfgS = 0x%016llx\n", vmcs_read64(GUEST_BNDCFGS));
5702 	pr_err("Interruptibility = %08x  ActivityState = %08x\n",
5703 	       vmcs_read32(GUEST_INTERRUPTIBILITY_INFO),
5704 	       vmcs_read32(GUEST_ACTIVITY_STATE));
5705 	if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)
5706 		pr_err("InterruptStatus = %04x\n",
5707 		       vmcs_read16(GUEST_INTR_STATUS));
5708 
5709 	pr_err("*** Host State ***\n");
5710 	pr_err("RIP = 0x%016lx  RSP = 0x%016lx\n",
5711 	       vmcs_readl(HOST_RIP), vmcs_readl(HOST_RSP));
5712 	pr_err("CS=%04x SS=%04x DS=%04x ES=%04x FS=%04x GS=%04x TR=%04x\n",
5713 	       vmcs_read16(HOST_CS_SELECTOR), vmcs_read16(HOST_SS_SELECTOR),
5714 	       vmcs_read16(HOST_DS_SELECTOR), vmcs_read16(HOST_ES_SELECTOR),
5715 	       vmcs_read16(HOST_FS_SELECTOR), vmcs_read16(HOST_GS_SELECTOR),
5716 	       vmcs_read16(HOST_TR_SELECTOR));
5717 	pr_err("FSBase=%016lx GSBase=%016lx TRBase=%016lx\n",
5718 	       vmcs_readl(HOST_FS_BASE), vmcs_readl(HOST_GS_BASE),
5719 	       vmcs_readl(HOST_TR_BASE));
5720 	pr_err("GDTBase=%016lx IDTBase=%016lx\n",
5721 	       vmcs_readl(HOST_GDTR_BASE), vmcs_readl(HOST_IDTR_BASE));
5722 	pr_err("CR0=%016lx CR3=%016lx CR4=%016lx\n",
5723 	       vmcs_readl(HOST_CR0), vmcs_readl(HOST_CR3),
5724 	       vmcs_readl(HOST_CR4));
5725 	pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
5726 	       vmcs_readl(HOST_IA32_SYSENTER_ESP),
5727 	       vmcs_read32(HOST_IA32_SYSENTER_CS),
5728 	       vmcs_readl(HOST_IA32_SYSENTER_EIP));
5729 	if (vmexit_ctl & (VM_EXIT_LOAD_IA32_PAT | VM_EXIT_LOAD_IA32_EFER))
5730 		pr_err("EFER = 0x%016llx  PAT = 0x%016llx\n",
5731 		       vmcs_read64(HOST_IA32_EFER),
5732 		       vmcs_read64(HOST_IA32_PAT));
5733 	if (cpu_has_load_perf_global_ctrl() &&
5734 	    vmexit_ctl & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
5735 		pr_err("PerfGlobCtl = 0x%016llx\n",
5736 		       vmcs_read64(HOST_IA32_PERF_GLOBAL_CTRL));
5737 
5738 	pr_err("*** Control State ***\n");
5739 	pr_err("PinBased=%08x CPUBased=%08x SecondaryExec=%08x\n",
5740 	       pin_based_exec_ctrl, cpu_based_exec_ctrl, secondary_exec_control);
5741 	pr_err("EntryControls=%08x ExitControls=%08x\n", vmentry_ctl, vmexit_ctl);
5742 	pr_err("ExceptionBitmap=%08x PFECmask=%08x PFECmatch=%08x\n",
5743 	       vmcs_read32(EXCEPTION_BITMAP),
5744 	       vmcs_read32(PAGE_FAULT_ERROR_CODE_MASK),
5745 	       vmcs_read32(PAGE_FAULT_ERROR_CODE_MATCH));
5746 	pr_err("VMEntry: intr_info=%08x errcode=%08x ilen=%08x\n",
5747 	       vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
5748 	       vmcs_read32(VM_ENTRY_EXCEPTION_ERROR_CODE),
5749 	       vmcs_read32(VM_ENTRY_INSTRUCTION_LEN));
5750 	pr_err("VMExit: intr_info=%08x errcode=%08x ilen=%08x\n",
5751 	       vmcs_read32(VM_EXIT_INTR_INFO),
5752 	       vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
5753 	       vmcs_read32(VM_EXIT_INSTRUCTION_LEN));
5754 	pr_err("        reason=%08x qualification=%016lx\n",
5755 	       vmcs_read32(VM_EXIT_REASON), vmcs_readl(EXIT_QUALIFICATION));
5756 	pr_err("IDTVectoring: info=%08x errcode=%08x\n",
5757 	       vmcs_read32(IDT_VECTORING_INFO_FIELD),
5758 	       vmcs_read32(IDT_VECTORING_ERROR_CODE));
5759 	pr_err("TSC Offset = 0x%016llx\n", vmcs_read64(TSC_OFFSET));
5760 	if (secondary_exec_control & SECONDARY_EXEC_TSC_SCALING)
5761 		pr_err("TSC Multiplier = 0x%016llx\n",
5762 		       vmcs_read64(TSC_MULTIPLIER));
5763 	if (cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW) {
5764 		if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY) {
5765 			u16 status = vmcs_read16(GUEST_INTR_STATUS);
5766 			pr_err("SVI|RVI = %02x|%02x ", status >> 8, status & 0xff);
5767 		}
5768 		pr_cont("TPR Threshold = 0x%02x\n", vmcs_read32(TPR_THRESHOLD));
5769 		if (secondary_exec_control & SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)
5770 			pr_err("APIC-access addr = 0x%016llx ", vmcs_read64(APIC_ACCESS_ADDR));
5771 		pr_cont("virt-APIC addr = 0x%016llx\n", vmcs_read64(VIRTUAL_APIC_PAGE_ADDR));
5772 	}
5773 	if (pin_based_exec_ctrl & PIN_BASED_POSTED_INTR)
5774 		pr_err("PostedIntrVec = 0x%02x\n", vmcs_read16(POSTED_INTR_NV));
5775 	if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT))
5776 		pr_err("EPT pointer = 0x%016llx\n", vmcs_read64(EPT_POINTER));
5777 	n = vmcs_read32(CR3_TARGET_COUNT);
5778 	for (i = 0; i + 1 < n; i += 4)
5779 		pr_err("CR3 target%u=%016lx target%u=%016lx\n",
5780 		       i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2),
5781 		       i + 1, vmcs_readl(CR3_TARGET_VALUE0 + i * 2 + 2));
5782 	if (i < n)
5783 		pr_err("CR3 target%u=%016lx\n",
5784 		       i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2));
5785 	if (secondary_exec_control & SECONDARY_EXEC_PAUSE_LOOP_EXITING)
5786 		pr_err("PLE Gap=%08x Window=%08x\n",
5787 		       vmcs_read32(PLE_GAP), vmcs_read32(PLE_WINDOW));
5788 	if (secondary_exec_control & SECONDARY_EXEC_ENABLE_VPID)
5789 		pr_err("Virtual processor ID = 0x%04x\n",
5790 		       vmcs_read16(VIRTUAL_PROCESSOR_ID));
5791 }
5792 
5793 /*
5794  * The guest has exited.  See if we can fix it or if we need userspace
5795  * assistance.
5796  */
5797 static int vmx_handle_exit(struct kvm_vcpu *vcpu,
5798 	enum exit_fastpath_completion exit_fastpath)
5799 {
5800 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5801 	u32 exit_reason = vmx->exit_reason;
5802 	u32 vectoring_info = vmx->idt_vectoring_info;
5803 
5804 	trace_kvm_exit(exit_reason, vcpu, KVM_ISA_VMX);
5805 
5806 	/*
5807 	 * Flush logged GPAs PML buffer, this will make dirty_bitmap more
5808 	 * updated. Another good is, in kvm_vm_ioctl_get_dirty_log, before
5809 	 * querying dirty_bitmap, we only need to kick all vcpus out of guest
5810 	 * mode as if vcpus is in root mode, the PML buffer must has been
5811 	 * flushed already.
5812 	 */
5813 	if (enable_pml)
5814 		vmx_flush_pml_buffer(vcpu);
5815 
5816 	/* If guest state is invalid, start emulating */
5817 	if (vmx->emulation_required)
5818 		return handle_invalid_guest_state(vcpu);
5819 
5820 	if (is_guest_mode(vcpu) && nested_vmx_exit_reflected(vcpu, exit_reason))
5821 		return nested_vmx_reflect_vmexit(vcpu, exit_reason);
5822 
5823 	if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) {
5824 		dump_vmcs();
5825 		vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
5826 		vcpu->run->fail_entry.hardware_entry_failure_reason
5827 			= exit_reason;
5828 		return 0;
5829 	}
5830 
5831 	if (unlikely(vmx->fail)) {
5832 		dump_vmcs();
5833 		vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
5834 		vcpu->run->fail_entry.hardware_entry_failure_reason
5835 			= vmcs_read32(VM_INSTRUCTION_ERROR);
5836 		return 0;
5837 	}
5838 
5839 	/*
5840 	 * Note:
5841 	 * Do not try to fix EXIT_REASON_EPT_MISCONFIG if it caused by
5842 	 * delivery event since it indicates guest is accessing MMIO.
5843 	 * The vm-exit can be triggered again after return to guest that
5844 	 * will cause infinite loop.
5845 	 */
5846 	if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
5847 			(exit_reason != EXIT_REASON_EXCEPTION_NMI &&
5848 			exit_reason != EXIT_REASON_EPT_VIOLATION &&
5849 			exit_reason != EXIT_REASON_PML_FULL &&
5850 			exit_reason != EXIT_REASON_TASK_SWITCH)) {
5851 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5852 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_DELIVERY_EV;
5853 		vcpu->run->internal.ndata = 3;
5854 		vcpu->run->internal.data[0] = vectoring_info;
5855 		vcpu->run->internal.data[1] = exit_reason;
5856 		vcpu->run->internal.data[2] = vcpu->arch.exit_qualification;
5857 		if (exit_reason == EXIT_REASON_EPT_MISCONFIG) {
5858 			vcpu->run->internal.ndata++;
5859 			vcpu->run->internal.data[3] =
5860 				vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5861 		}
5862 		return 0;
5863 	}
5864 
5865 	if (unlikely(!enable_vnmi &&
5866 		     vmx->loaded_vmcs->soft_vnmi_blocked)) {
5867 		if (vmx_interrupt_allowed(vcpu)) {
5868 			vmx->loaded_vmcs->soft_vnmi_blocked = 0;
5869 		} else if (vmx->loaded_vmcs->vnmi_blocked_time > 1000000000LL &&
5870 			   vcpu->arch.nmi_pending) {
5871 			/*
5872 			 * This CPU don't support us in finding the end of an
5873 			 * NMI-blocked window if the guest runs with IRQs
5874 			 * disabled. So we pull the trigger after 1 s of
5875 			 * futile waiting, but inform the user about this.
5876 			 */
5877 			printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
5878 			       "state on VCPU %d after 1 s timeout\n",
5879 			       __func__, vcpu->vcpu_id);
5880 			vmx->loaded_vmcs->soft_vnmi_blocked = 0;
5881 		}
5882 	}
5883 
5884 	if (exit_fastpath == EXIT_FASTPATH_SKIP_EMUL_INS) {
5885 		kvm_skip_emulated_instruction(vcpu);
5886 		return 1;
5887 	}
5888 
5889 	if (exit_reason >= kvm_vmx_max_exit_handlers)
5890 		goto unexpected_vmexit;
5891 #ifdef CONFIG_RETPOLINE
5892 	if (exit_reason == EXIT_REASON_MSR_WRITE)
5893 		return kvm_emulate_wrmsr(vcpu);
5894 	else if (exit_reason == EXIT_REASON_PREEMPTION_TIMER)
5895 		return handle_preemption_timer(vcpu);
5896 	else if (exit_reason == EXIT_REASON_INTERRUPT_WINDOW)
5897 		return handle_interrupt_window(vcpu);
5898 	else if (exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT)
5899 		return handle_external_interrupt(vcpu);
5900 	else if (exit_reason == EXIT_REASON_HLT)
5901 		return kvm_emulate_halt(vcpu);
5902 	else if (exit_reason == EXIT_REASON_EPT_MISCONFIG)
5903 		return handle_ept_misconfig(vcpu);
5904 #endif
5905 
5906 	exit_reason = array_index_nospec(exit_reason,
5907 					 kvm_vmx_max_exit_handlers);
5908 	if (!kvm_vmx_exit_handlers[exit_reason])
5909 		goto unexpected_vmexit;
5910 
5911 	return kvm_vmx_exit_handlers[exit_reason](vcpu);
5912 
5913 unexpected_vmexit:
5914 	vcpu_unimpl(vcpu, "vmx: unexpected exit reason 0x%x\n", exit_reason);
5915 	dump_vmcs();
5916 	vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5917 	vcpu->run->internal.suberror =
5918 			KVM_INTERNAL_ERROR_UNEXPECTED_EXIT_REASON;
5919 	vcpu->run->internal.ndata = 1;
5920 	vcpu->run->internal.data[0] = exit_reason;
5921 	return 0;
5922 }
5923 
5924 /*
5925  * Software based L1D cache flush which is used when microcode providing
5926  * the cache control MSR is not loaded.
5927  *
5928  * The L1D cache is 32 KiB on Nehalem and later microarchitectures, but to
5929  * flush it is required to read in 64 KiB because the replacement algorithm
5930  * is not exactly LRU. This could be sized at runtime via topology
5931  * information but as all relevant affected CPUs have 32KiB L1D cache size
5932  * there is no point in doing so.
5933  */
5934 static void vmx_l1d_flush(struct kvm_vcpu *vcpu)
5935 {
5936 	int size = PAGE_SIZE << L1D_CACHE_ORDER;
5937 
5938 	/*
5939 	 * This code is only executed when the the flush mode is 'cond' or
5940 	 * 'always'
5941 	 */
5942 	if (static_branch_likely(&vmx_l1d_flush_cond)) {
5943 		bool flush_l1d;
5944 
5945 		/*
5946 		 * Clear the per-vcpu flush bit, it gets set again
5947 		 * either from vcpu_run() or from one of the unsafe
5948 		 * VMEXIT handlers.
5949 		 */
5950 		flush_l1d = vcpu->arch.l1tf_flush_l1d;
5951 		vcpu->arch.l1tf_flush_l1d = false;
5952 
5953 		/*
5954 		 * Clear the per-cpu flush bit, it gets set again from
5955 		 * the interrupt handlers.
5956 		 */
5957 		flush_l1d |= kvm_get_cpu_l1tf_flush_l1d();
5958 		kvm_clear_cpu_l1tf_flush_l1d();
5959 
5960 		if (!flush_l1d)
5961 			return;
5962 	}
5963 
5964 	vcpu->stat.l1d_flush++;
5965 
5966 	if (static_cpu_has(X86_FEATURE_FLUSH_L1D)) {
5967 		wrmsrl(MSR_IA32_FLUSH_CMD, L1D_FLUSH);
5968 		return;
5969 	}
5970 
5971 	asm volatile(
5972 		/* First ensure the pages are in the TLB */
5973 		"xorl	%%eax, %%eax\n"
5974 		".Lpopulate_tlb:\n\t"
5975 		"movzbl	(%[flush_pages], %%" _ASM_AX "), %%ecx\n\t"
5976 		"addl	$4096, %%eax\n\t"
5977 		"cmpl	%%eax, %[size]\n\t"
5978 		"jne	.Lpopulate_tlb\n\t"
5979 		"xorl	%%eax, %%eax\n\t"
5980 		"cpuid\n\t"
5981 		/* Now fill the cache */
5982 		"xorl	%%eax, %%eax\n"
5983 		".Lfill_cache:\n"
5984 		"movzbl	(%[flush_pages], %%" _ASM_AX "), %%ecx\n\t"
5985 		"addl	$64, %%eax\n\t"
5986 		"cmpl	%%eax, %[size]\n\t"
5987 		"jne	.Lfill_cache\n\t"
5988 		"lfence\n"
5989 		:: [flush_pages] "r" (vmx_l1d_flush_pages),
5990 		    [size] "r" (size)
5991 		: "eax", "ebx", "ecx", "edx");
5992 }
5993 
5994 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
5995 {
5996 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5997 	int tpr_threshold;
5998 
5999 	if (is_guest_mode(vcpu) &&
6000 		nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
6001 		return;
6002 
6003 	tpr_threshold = (irr == -1 || tpr < irr) ? 0 : irr;
6004 	if (is_guest_mode(vcpu))
6005 		to_vmx(vcpu)->nested.l1_tpr_threshold = tpr_threshold;
6006 	else
6007 		vmcs_write32(TPR_THRESHOLD, tpr_threshold);
6008 }
6009 
6010 void vmx_set_virtual_apic_mode(struct kvm_vcpu *vcpu)
6011 {
6012 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6013 	u32 sec_exec_control;
6014 
6015 	if (!lapic_in_kernel(vcpu))
6016 		return;
6017 
6018 	if (!flexpriority_enabled &&
6019 	    !cpu_has_vmx_virtualize_x2apic_mode())
6020 		return;
6021 
6022 	/* Postpone execution until vmcs01 is the current VMCS. */
6023 	if (is_guest_mode(vcpu)) {
6024 		vmx->nested.change_vmcs01_virtual_apic_mode = true;
6025 		return;
6026 	}
6027 
6028 	sec_exec_control = secondary_exec_controls_get(vmx);
6029 	sec_exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
6030 			      SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE);
6031 
6032 	switch (kvm_get_apic_mode(vcpu)) {
6033 	case LAPIC_MODE_INVALID:
6034 		WARN_ONCE(true, "Invalid local APIC state");
6035 	case LAPIC_MODE_DISABLED:
6036 		break;
6037 	case LAPIC_MODE_XAPIC:
6038 		if (flexpriority_enabled) {
6039 			sec_exec_control |=
6040 				SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
6041 			vmx_flush_tlb(vcpu, true);
6042 		}
6043 		break;
6044 	case LAPIC_MODE_X2APIC:
6045 		if (cpu_has_vmx_virtualize_x2apic_mode())
6046 			sec_exec_control |=
6047 				SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
6048 		break;
6049 	}
6050 	secondary_exec_controls_set(vmx, sec_exec_control);
6051 
6052 	vmx_update_msr_bitmap(vcpu);
6053 }
6054 
6055 static void vmx_set_apic_access_page_addr(struct kvm_vcpu *vcpu, hpa_t hpa)
6056 {
6057 	if (!is_guest_mode(vcpu)) {
6058 		vmcs_write64(APIC_ACCESS_ADDR, hpa);
6059 		vmx_flush_tlb(vcpu, true);
6060 	}
6061 }
6062 
6063 static void vmx_hwapic_isr_update(struct kvm_vcpu *vcpu, int max_isr)
6064 {
6065 	u16 status;
6066 	u8 old;
6067 
6068 	if (max_isr == -1)
6069 		max_isr = 0;
6070 
6071 	status = vmcs_read16(GUEST_INTR_STATUS);
6072 	old = status >> 8;
6073 	if (max_isr != old) {
6074 		status &= 0xff;
6075 		status |= max_isr << 8;
6076 		vmcs_write16(GUEST_INTR_STATUS, status);
6077 	}
6078 }
6079 
6080 static void vmx_set_rvi(int vector)
6081 {
6082 	u16 status;
6083 	u8 old;
6084 
6085 	if (vector == -1)
6086 		vector = 0;
6087 
6088 	status = vmcs_read16(GUEST_INTR_STATUS);
6089 	old = (u8)status & 0xff;
6090 	if ((u8)vector != old) {
6091 		status &= ~0xff;
6092 		status |= (u8)vector;
6093 		vmcs_write16(GUEST_INTR_STATUS, status);
6094 	}
6095 }
6096 
6097 static void vmx_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr)
6098 {
6099 	/*
6100 	 * When running L2, updating RVI is only relevant when
6101 	 * vmcs12 virtual-interrupt-delivery enabled.
6102 	 * However, it can be enabled only when L1 also
6103 	 * intercepts external-interrupts and in that case
6104 	 * we should not update vmcs02 RVI but instead intercept
6105 	 * interrupt. Therefore, do nothing when running L2.
6106 	 */
6107 	if (!is_guest_mode(vcpu))
6108 		vmx_set_rvi(max_irr);
6109 }
6110 
6111 static int vmx_sync_pir_to_irr(struct kvm_vcpu *vcpu)
6112 {
6113 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6114 	int max_irr;
6115 	bool max_irr_updated;
6116 
6117 	WARN_ON(!vcpu->arch.apicv_active);
6118 	if (pi_test_on(&vmx->pi_desc)) {
6119 		pi_clear_on(&vmx->pi_desc);
6120 		/*
6121 		 * IOMMU can write to PID.ON, so the barrier matters even on UP.
6122 		 * But on x86 this is just a compiler barrier anyway.
6123 		 */
6124 		smp_mb__after_atomic();
6125 		max_irr_updated =
6126 			kvm_apic_update_irr(vcpu, vmx->pi_desc.pir, &max_irr);
6127 
6128 		/*
6129 		 * If we are running L2 and L1 has a new pending interrupt
6130 		 * which can be injected, we should re-evaluate
6131 		 * what should be done with this new L1 interrupt.
6132 		 * If L1 intercepts external-interrupts, we should
6133 		 * exit from L2 to L1. Otherwise, interrupt should be
6134 		 * delivered directly to L2.
6135 		 */
6136 		if (is_guest_mode(vcpu) && max_irr_updated) {
6137 			if (nested_exit_on_intr(vcpu))
6138 				kvm_vcpu_exiting_guest_mode(vcpu);
6139 			else
6140 				kvm_make_request(KVM_REQ_EVENT, vcpu);
6141 		}
6142 	} else {
6143 		max_irr = kvm_lapic_find_highest_irr(vcpu);
6144 	}
6145 	vmx_hwapic_irr_update(vcpu, max_irr);
6146 	return max_irr;
6147 }
6148 
6149 static bool vmx_dy_apicv_has_pending_interrupt(struct kvm_vcpu *vcpu)
6150 {
6151 	struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
6152 
6153 	return pi_test_on(pi_desc) ||
6154 		(pi_test_sn(pi_desc) && !pi_is_pir_empty(pi_desc));
6155 }
6156 
6157 static void vmx_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap)
6158 {
6159 	if (!kvm_vcpu_apicv_active(vcpu))
6160 		return;
6161 
6162 	vmcs_write64(EOI_EXIT_BITMAP0, eoi_exit_bitmap[0]);
6163 	vmcs_write64(EOI_EXIT_BITMAP1, eoi_exit_bitmap[1]);
6164 	vmcs_write64(EOI_EXIT_BITMAP2, eoi_exit_bitmap[2]);
6165 	vmcs_write64(EOI_EXIT_BITMAP3, eoi_exit_bitmap[3]);
6166 }
6167 
6168 static void vmx_apicv_post_state_restore(struct kvm_vcpu *vcpu)
6169 {
6170 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6171 
6172 	pi_clear_on(&vmx->pi_desc);
6173 	memset(vmx->pi_desc.pir, 0, sizeof(vmx->pi_desc.pir));
6174 }
6175 
6176 static void handle_exception_nmi_irqoff(struct vcpu_vmx *vmx)
6177 {
6178 	vmx->exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6179 
6180 	/* if exit due to PF check for async PF */
6181 	if (is_page_fault(vmx->exit_intr_info))
6182 		vmx->vcpu.arch.apf.host_apf_reason = kvm_read_and_reset_pf_reason();
6183 
6184 	/* Handle machine checks before interrupts are enabled */
6185 	if (is_machine_check(vmx->exit_intr_info))
6186 		kvm_machine_check();
6187 
6188 	/* We need to handle NMIs before interrupts are enabled */
6189 	if (is_nmi(vmx->exit_intr_info)) {
6190 		kvm_before_interrupt(&vmx->vcpu);
6191 		asm("int $2");
6192 		kvm_after_interrupt(&vmx->vcpu);
6193 	}
6194 }
6195 
6196 static void handle_external_interrupt_irqoff(struct kvm_vcpu *vcpu)
6197 {
6198 	unsigned int vector;
6199 	unsigned long entry;
6200 #ifdef CONFIG_X86_64
6201 	unsigned long tmp;
6202 #endif
6203 	gate_desc *desc;
6204 	u32 intr_info;
6205 
6206 	intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6207 	if (WARN_ONCE(!is_external_intr(intr_info),
6208 	    "KVM: unexpected VM-Exit interrupt info: 0x%x", intr_info))
6209 		return;
6210 
6211 	vector = intr_info & INTR_INFO_VECTOR_MASK;
6212 	desc = (gate_desc *)host_idt_base + vector;
6213 	entry = gate_offset(desc);
6214 
6215 	kvm_before_interrupt(vcpu);
6216 
6217 	asm volatile(
6218 #ifdef CONFIG_X86_64
6219 		"mov %%" _ASM_SP ", %[sp]\n\t"
6220 		"and $0xfffffffffffffff0, %%" _ASM_SP "\n\t"
6221 		"push $%c[ss]\n\t"
6222 		"push %[sp]\n\t"
6223 #endif
6224 		"pushf\n\t"
6225 		__ASM_SIZE(push) " $%c[cs]\n\t"
6226 		CALL_NOSPEC
6227 		:
6228 #ifdef CONFIG_X86_64
6229 		[sp]"=&r"(tmp),
6230 #endif
6231 		ASM_CALL_CONSTRAINT
6232 		:
6233 		THUNK_TARGET(entry),
6234 		[ss]"i"(__KERNEL_DS),
6235 		[cs]"i"(__KERNEL_CS)
6236 	);
6237 
6238 	kvm_after_interrupt(vcpu);
6239 }
6240 STACK_FRAME_NON_STANDARD(handle_external_interrupt_irqoff);
6241 
6242 static void vmx_handle_exit_irqoff(struct kvm_vcpu *vcpu,
6243 	enum exit_fastpath_completion *exit_fastpath)
6244 {
6245 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6246 
6247 	if (vmx->exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT)
6248 		handle_external_interrupt_irqoff(vcpu);
6249 	else if (vmx->exit_reason == EXIT_REASON_EXCEPTION_NMI)
6250 		handle_exception_nmi_irqoff(vmx);
6251 	else if (!is_guest_mode(vcpu) &&
6252 		vmx->exit_reason == EXIT_REASON_MSR_WRITE)
6253 		*exit_fastpath = handle_fastpath_set_msr_irqoff(vcpu);
6254 }
6255 
6256 static bool vmx_has_emulated_msr(int index)
6257 {
6258 	switch (index) {
6259 	case MSR_IA32_SMBASE:
6260 		/*
6261 		 * We cannot do SMM unless we can run the guest in big
6262 		 * real mode.
6263 		 */
6264 		return enable_unrestricted_guest || emulate_invalid_guest_state;
6265 	case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
6266 		return nested;
6267 	case MSR_AMD64_VIRT_SPEC_CTRL:
6268 		/* This is AMD only.  */
6269 		return false;
6270 	default:
6271 		return true;
6272 	}
6273 }
6274 
6275 static bool vmx_pt_supported(void)
6276 {
6277 	return pt_mode == PT_MODE_HOST_GUEST;
6278 }
6279 
6280 static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx)
6281 {
6282 	u32 exit_intr_info;
6283 	bool unblock_nmi;
6284 	u8 vector;
6285 	bool idtv_info_valid;
6286 
6287 	idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK;
6288 
6289 	if (enable_vnmi) {
6290 		if (vmx->loaded_vmcs->nmi_known_unmasked)
6291 			return;
6292 		/*
6293 		 * Can't use vmx->exit_intr_info since we're not sure what
6294 		 * the exit reason is.
6295 		 */
6296 		exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6297 		unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
6298 		vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
6299 		/*
6300 		 * SDM 3: 27.7.1.2 (September 2008)
6301 		 * Re-set bit "block by NMI" before VM entry if vmexit caused by
6302 		 * a guest IRET fault.
6303 		 * SDM 3: 23.2.2 (September 2008)
6304 		 * Bit 12 is undefined in any of the following cases:
6305 		 *  If the VM exit sets the valid bit in the IDT-vectoring
6306 		 *   information field.
6307 		 *  If the VM exit is due to a double fault.
6308 		 */
6309 		if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
6310 		    vector != DF_VECTOR && !idtv_info_valid)
6311 			vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
6312 				      GUEST_INTR_STATE_NMI);
6313 		else
6314 			vmx->loaded_vmcs->nmi_known_unmasked =
6315 				!(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO)
6316 				  & GUEST_INTR_STATE_NMI);
6317 	} else if (unlikely(vmx->loaded_vmcs->soft_vnmi_blocked))
6318 		vmx->loaded_vmcs->vnmi_blocked_time +=
6319 			ktime_to_ns(ktime_sub(ktime_get(),
6320 					      vmx->loaded_vmcs->entry_time));
6321 }
6322 
6323 static void __vmx_complete_interrupts(struct kvm_vcpu *vcpu,
6324 				      u32 idt_vectoring_info,
6325 				      int instr_len_field,
6326 				      int error_code_field)
6327 {
6328 	u8 vector;
6329 	int type;
6330 	bool idtv_info_valid;
6331 
6332 	idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
6333 
6334 	vcpu->arch.nmi_injected = false;
6335 	kvm_clear_exception_queue(vcpu);
6336 	kvm_clear_interrupt_queue(vcpu);
6337 
6338 	if (!idtv_info_valid)
6339 		return;
6340 
6341 	kvm_make_request(KVM_REQ_EVENT, vcpu);
6342 
6343 	vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
6344 	type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
6345 
6346 	switch (type) {
6347 	case INTR_TYPE_NMI_INTR:
6348 		vcpu->arch.nmi_injected = true;
6349 		/*
6350 		 * SDM 3: 27.7.1.2 (September 2008)
6351 		 * Clear bit "block by NMI" before VM entry if a NMI
6352 		 * delivery faulted.
6353 		 */
6354 		vmx_set_nmi_mask(vcpu, false);
6355 		break;
6356 	case INTR_TYPE_SOFT_EXCEPTION:
6357 		vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
6358 		/* fall through */
6359 	case INTR_TYPE_HARD_EXCEPTION:
6360 		if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
6361 			u32 err = vmcs_read32(error_code_field);
6362 			kvm_requeue_exception_e(vcpu, vector, err);
6363 		} else
6364 			kvm_requeue_exception(vcpu, vector);
6365 		break;
6366 	case INTR_TYPE_SOFT_INTR:
6367 		vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
6368 		/* fall through */
6369 	case INTR_TYPE_EXT_INTR:
6370 		kvm_queue_interrupt(vcpu, vector, type == INTR_TYPE_SOFT_INTR);
6371 		break;
6372 	default:
6373 		break;
6374 	}
6375 }
6376 
6377 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
6378 {
6379 	__vmx_complete_interrupts(&vmx->vcpu, vmx->idt_vectoring_info,
6380 				  VM_EXIT_INSTRUCTION_LEN,
6381 				  IDT_VECTORING_ERROR_CODE);
6382 }
6383 
6384 static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
6385 {
6386 	__vmx_complete_interrupts(vcpu,
6387 				  vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
6388 				  VM_ENTRY_INSTRUCTION_LEN,
6389 				  VM_ENTRY_EXCEPTION_ERROR_CODE);
6390 
6391 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
6392 }
6393 
6394 static void atomic_switch_perf_msrs(struct vcpu_vmx *vmx)
6395 {
6396 	int i, nr_msrs;
6397 	struct perf_guest_switch_msr *msrs;
6398 
6399 	msrs = perf_guest_get_msrs(&nr_msrs);
6400 
6401 	if (!msrs)
6402 		return;
6403 
6404 	for (i = 0; i < nr_msrs; i++)
6405 		if (msrs[i].host == msrs[i].guest)
6406 			clear_atomic_switch_msr(vmx, msrs[i].msr);
6407 		else
6408 			add_atomic_switch_msr(vmx, msrs[i].msr, msrs[i].guest,
6409 					msrs[i].host, false);
6410 }
6411 
6412 static void atomic_switch_umwait_control_msr(struct vcpu_vmx *vmx)
6413 {
6414 	u32 host_umwait_control;
6415 
6416 	if (!vmx_has_waitpkg(vmx))
6417 		return;
6418 
6419 	host_umwait_control = get_umwait_control_msr();
6420 
6421 	if (vmx->msr_ia32_umwait_control != host_umwait_control)
6422 		add_atomic_switch_msr(vmx, MSR_IA32_UMWAIT_CONTROL,
6423 			vmx->msr_ia32_umwait_control,
6424 			host_umwait_control, false);
6425 	else
6426 		clear_atomic_switch_msr(vmx, MSR_IA32_UMWAIT_CONTROL);
6427 }
6428 
6429 static void vmx_update_hv_timer(struct kvm_vcpu *vcpu)
6430 {
6431 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6432 	u64 tscl;
6433 	u32 delta_tsc;
6434 
6435 	if (vmx->req_immediate_exit) {
6436 		vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, 0);
6437 		vmx->loaded_vmcs->hv_timer_soft_disabled = false;
6438 	} else if (vmx->hv_deadline_tsc != -1) {
6439 		tscl = rdtsc();
6440 		if (vmx->hv_deadline_tsc > tscl)
6441 			/* set_hv_timer ensures the delta fits in 32-bits */
6442 			delta_tsc = (u32)((vmx->hv_deadline_tsc - tscl) >>
6443 				cpu_preemption_timer_multi);
6444 		else
6445 			delta_tsc = 0;
6446 
6447 		vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, delta_tsc);
6448 		vmx->loaded_vmcs->hv_timer_soft_disabled = false;
6449 	} else if (!vmx->loaded_vmcs->hv_timer_soft_disabled) {
6450 		vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, -1);
6451 		vmx->loaded_vmcs->hv_timer_soft_disabled = true;
6452 	}
6453 }
6454 
6455 void vmx_update_host_rsp(struct vcpu_vmx *vmx, unsigned long host_rsp)
6456 {
6457 	if (unlikely(host_rsp != vmx->loaded_vmcs->host_state.rsp)) {
6458 		vmx->loaded_vmcs->host_state.rsp = host_rsp;
6459 		vmcs_writel(HOST_RSP, host_rsp);
6460 	}
6461 }
6462 
6463 bool __vmx_vcpu_run(struct vcpu_vmx *vmx, unsigned long *regs, bool launched);
6464 
6465 static void vmx_vcpu_run(struct kvm_vcpu *vcpu)
6466 {
6467 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6468 	unsigned long cr3, cr4;
6469 
6470 	/* Record the guest's net vcpu time for enforced NMI injections. */
6471 	if (unlikely(!enable_vnmi &&
6472 		     vmx->loaded_vmcs->soft_vnmi_blocked))
6473 		vmx->loaded_vmcs->entry_time = ktime_get();
6474 
6475 	/* Don't enter VMX if guest state is invalid, let the exit handler
6476 	   start emulation until we arrive back to a valid state */
6477 	if (vmx->emulation_required)
6478 		return;
6479 
6480 	if (vmx->ple_window_dirty) {
6481 		vmx->ple_window_dirty = false;
6482 		vmcs_write32(PLE_WINDOW, vmx->ple_window);
6483 	}
6484 
6485 	if (vmx->nested.need_vmcs12_to_shadow_sync)
6486 		nested_sync_vmcs12_to_shadow(vcpu);
6487 
6488 	if (kvm_register_is_dirty(vcpu, VCPU_REGS_RSP))
6489 		vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
6490 	if (kvm_register_is_dirty(vcpu, VCPU_REGS_RIP))
6491 		vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
6492 
6493 	cr3 = __get_current_cr3_fast();
6494 	if (unlikely(cr3 != vmx->loaded_vmcs->host_state.cr3)) {
6495 		vmcs_writel(HOST_CR3, cr3);
6496 		vmx->loaded_vmcs->host_state.cr3 = cr3;
6497 	}
6498 
6499 	cr4 = cr4_read_shadow();
6500 	if (unlikely(cr4 != vmx->loaded_vmcs->host_state.cr4)) {
6501 		vmcs_writel(HOST_CR4, cr4);
6502 		vmx->loaded_vmcs->host_state.cr4 = cr4;
6503 	}
6504 
6505 	/* When single-stepping over STI and MOV SS, we must clear the
6506 	 * corresponding interruptibility bits in the guest state. Otherwise
6507 	 * vmentry fails as it then expects bit 14 (BS) in pending debug
6508 	 * exceptions being set, but that's not correct for the guest debugging
6509 	 * case. */
6510 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6511 		vmx_set_interrupt_shadow(vcpu, 0);
6512 
6513 	kvm_load_guest_xsave_state(vcpu);
6514 
6515 	if (static_cpu_has(X86_FEATURE_PKU) &&
6516 	    kvm_read_cr4_bits(vcpu, X86_CR4_PKE) &&
6517 	    vcpu->arch.pkru != vmx->host_pkru)
6518 		__write_pkru(vcpu->arch.pkru);
6519 
6520 	pt_guest_enter(vmx);
6521 
6522 	atomic_switch_perf_msrs(vmx);
6523 	atomic_switch_umwait_control_msr(vmx);
6524 
6525 	if (enable_preemption_timer)
6526 		vmx_update_hv_timer(vcpu);
6527 
6528 	if (lapic_in_kernel(vcpu) &&
6529 		vcpu->arch.apic->lapic_timer.timer_advance_ns)
6530 		kvm_wait_lapic_expire(vcpu);
6531 
6532 	/*
6533 	 * If this vCPU has touched SPEC_CTRL, restore the guest's value if
6534 	 * it's non-zero. Since vmentry is serialising on affected CPUs, there
6535 	 * is no need to worry about the conditional branch over the wrmsr
6536 	 * being speculatively taken.
6537 	 */
6538 	x86_spec_ctrl_set_guest(vmx->spec_ctrl, 0);
6539 
6540 	/* L1D Flush includes CPU buffer clear to mitigate MDS */
6541 	if (static_branch_unlikely(&vmx_l1d_should_flush))
6542 		vmx_l1d_flush(vcpu);
6543 	else if (static_branch_unlikely(&mds_user_clear))
6544 		mds_clear_cpu_buffers();
6545 
6546 	if (vcpu->arch.cr2 != read_cr2())
6547 		write_cr2(vcpu->arch.cr2);
6548 
6549 	vmx->fail = __vmx_vcpu_run(vmx, (unsigned long *)&vcpu->arch.regs,
6550 				   vmx->loaded_vmcs->launched);
6551 
6552 	vcpu->arch.cr2 = read_cr2();
6553 
6554 	/*
6555 	 * We do not use IBRS in the kernel. If this vCPU has used the
6556 	 * SPEC_CTRL MSR it may have left it on; save the value and
6557 	 * turn it off. This is much more efficient than blindly adding
6558 	 * it to the atomic save/restore list. Especially as the former
6559 	 * (Saving guest MSRs on vmexit) doesn't even exist in KVM.
6560 	 *
6561 	 * For non-nested case:
6562 	 * If the L01 MSR bitmap does not intercept the MSR, then we need to
6563 	 * save it.
6564 	 *
6565 	 * For nested case:
6566 	 * If the L02 MSR bitmap does not intercept the MSR, then we need to
6567 	 * save it.
6568 	 */
6569 	if (unlikely(!msr_write_intercepted(vcpu, MSR_IA32_SPEC_CTRL)))
6570 		vmx->spec_ctrl = native_read_msr(MSR_IA32_SPEC_CTRL);
6571 
6572 	x86_spec_ctrl_restore_host(vmx->spec_ctrl, 0);
6573 
6574 	/* All fields are clean at this point */
6575 	if (static_branch_unlikely(&enable_evmcs))
6576 		current_evmcs->hv_clean_fields |=
6577 			HV_VMX_ENLIGHTENED_CLEAN_FIELD_ALL;
6578 
6579 	if (static_branch_unlikely(&enable_evmcs))
6580 		current_evmcs->hv_vp_id = vcpu->arch.hyperv.vp_index;
6581 
6582 	/* MSR_IA32_DEBUGCTLMSR is zeroed on vmexit. Restore it if needed */
6583 	if (vmx->host_debugctlmsr)
6584 		update_debugctlmsr(vmx->host_debugctlmsr);
6585 
6586 #ifndef CONFIG_X86_64
6587 	/*
6588 	 * The sysexit path does not restore ds/es, so we must set them to
6589 	 * a reasonable value ourselves.
6590 	 *
6591 	 * We can't defer this to vmx_prepare_switch_to_host() since that
6592 	 * function may be executed in interrupt context, which saves and
6593 	 * restore segments around it, nullifying its effect.
6594 	 */
6595 	loadsegment(ds, __USER_DS);
6596 	loadsegment(es, __USER_DS);
6597 #endif
6598 
6599 	vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
6600 				  | (1 << VCPU_EXREG_RFLAGS)
6601 				  | (1 << VCPU_EXREG_PDPTR)
6602 				  | (1 << VCPU_EXREG_SEGMENTS)
6603 				  | (1 << VCPU_EXREG_CR3));
6604 	vcpu->arch.regs_dirty = 0;
6605 
6606 	pt_guest_exit(vmx);
6607 
6608 	/*
6609 	 * eager fpu is enabled if PKEY is supported and CR4 is switched
6610 	 * back on host, so it is safe to read guest PKRU from current
6611 	 * XSAVE.
6612 	 */
6613 	if (static_cpu_has(X86_FEATURE_PKU) &&
6614 	    kvm_read_cr4_bits(vcpu, X86_CR4_PKE)) {
6615 		vcpu->arch.pkru = rdpkru();
6616 		if (vcpu->arch.pkru != vmx->host_pkru)
6617 			__write_pkru(vmx->host_pkru);
6618 	}
6619 
6620 	kvm_load_host_xsave_state(vcpu);
6621 
6622 	vmx->nested.nested_run_pending = 0;
6623 	vmx->idt_vectoring_info = 0;
6624 
6625 	vmx->exit_reason = vmx->fail ? 0xdead : vmcs_read32(VM_EXIT_REASON);
6626 	if ((u16)vmx->exit_reason == EXIT_REASON_MCE_DURING_VMENTRY)
6627 		kvm_machine_check();
6628 
6629 	if (vmx->fail || (vmx->exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY))
6630 		return;
6631 
6632 	vmx->loaded_vmcs->launched = 1;
6633 	vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
6634 
6635 	vmx_recover_nmi_blocking(vmx);
6636 	vmx_complete_interrupts(vmx);
6637 }
6638 
6639 static struct kvm *vmx_vm_alloc(void)
6640 {
6641 	struct kvm_vmx *kvm_vmx = __vmalloc(sizeof(struct kvm_vmx),
6642 					    GFP_KERNEL_ACCOUNT | __GFP_ZERO,
6643 					    PAGE_KERNEL);
6644 	return &kvm_vmx->kvm;
6645 }
6646 
6647 static void vmx_vm_free(struct kvm *kvm)
6648 {
6649 	kfree(kvm->arch.hyperv.hv_pa_pg);
6650 	vfree(to_kvm_vmx(kvm));
6651 }
6652 
6653 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
6654 {
6655 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6656 
6657 	if (enable_pml)
6658 		vmx_destroy_pml_buffer(vmx);
6659 	free_vpid(vmx->vpid);
6660 	nested_vmx_free_vcpu(vcpu);
6661 	free_loaded_vmcs(vmx->loaded_vmcs);
6662 }
6663 
6664 static int vmx_create_vcpu(struct kvm_vcpu *vcpu)
6665 {
6666 	struct vcpu_vmx *vmx;
6667 	unsigned long *msr_bitmap;
6668 	int i, cpu, err;
6669 
6670 	BUILD_BUG_ON(offsetof(struct vcpu_vmx, vcpu) != 0);
6671 	vmx = to_vmx(vcpu);
6672 
6673 	err = -ENOMEM;
6674 
6675 	vmx->vpid = allocate_vpid();
6676 
6677 	/*
6678 	 * If PML is turned on, failure on enabling PML just results in failure
6679 	 * of creating the vcpu, therefore we can simplify PML logic (by
6680 	 * avoiding dealing with cases, such as enabling PML partially on vcpus
6681 	 * for the guest), etc.
6682 	 */
6683 	if (enable_pml) {
6684 		vmx->pml_pg = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
6685 		if (!vmx->pml_pg)
6686 			goto free_vpid;
6687 	}
6688 
6689 	BUILD_BUG_ON(ARRAY_SIZE(vmx_msr_index) != NR_SHARED_MSRS);
6690 
6691 	for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i) {
6692 		u32 index = vmx_msr_index[i];
6693 		u32 data_low, data_high;
6694 		int j = vmx->nmsrs;
6695 
6696 		if (rdmsr_safe(index, &data_low, &data_high) < 0)
6697 			continue;
6698 		if (wrmsr_safe(index, data_low, data_high) < 0)
6699 			continue;
6700 
6701 		vmx->guest_msrs[j].index = i;
6702 		vmx->guest_msrs[j].data = 0;
6703 		switch (index) {
6704 		case MSR_IA32_TSX_CTRL:
6705 			/*
6706 			 * No need to pass TSX_CTRL_CPUID_CLEAR through, so
6707 			 * let's avoid changing CPUID bits under the host
6708 			 * kernel's feet.
6709 			 */
6710 			vmx->guest_msrs[j].mask = ~(u64)TSX_CTRL_CPUID_CLEAR;
6711 			break;
6712 		default:
6713 			vmx->guest_msrs[j].mask = -1ull;
6714 			break;
6715 		}
6716 		++vmx->nmsrs;
6717 	}
6718 
6719 	err = alloc_loaded_vmcs(&vmx->vmcs01);
6720 	if (err < 0)
6721 		goto free_pml;
6722 
6723 	msr_bitmap = vmx->vmcs01.msr_bitmap;
6724 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_TSC, MSR_TYPE_R);
6725 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_FS_BASE, MSR_TYPE_RW);
6726 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_GS_BASE, MSR_TYPE_RW);
6727 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_KERNEL_GS_BASE, MSR_TYPE_RW);
6728 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_CS, MSR_TYPE_RW);
6729 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_ESP, MSR_TYPE_RW);
6730 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_EIP, MSR_TYPE_RW);
6731 	if (kvm_cstate_in_guest(vcpu->kvm)) {
6732 		vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C1_RES, MSR_TYPE_R);
6733 		vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C3_RESIDENCY, MSR_TYPE_R);
6734 		vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C6_RESIDENCY, MSR_TYPE_R);
6735 		vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C7_RESIDENCY, MSR_TYPE_R);
6736 	}
6737 	vmx->msr_bitmap_mode = 0;
6738 
6739 	vmx->loaded_vmcs = &vmx->vmcs01;
6740 	cpu = get_cpu();
6741 	vmx_vcpu_load(vcpu, cpu);
6742 	vcpu->cpu = cpu;
6743 	init_vmcs(vmx);
6744 	vmx_vcpu_put(vcpu);
6745 	put_cpu();
6746 	if (cpu_need_virtualize_apic_accesses(vcpu)) {
6747 		err = alloc_apic_access_page(vcpu->kvm);
6748 		if (err)
6749 			goto free_vmcs;
6750 	}
6751 
6752 	if (enable_ept && !enable_unrestricted_guest) {
6753 		err = init_rmode_identity_map(vcpu->kvm);
6754 		if (err)
6755 			goto free_vmcs;
6756 	}
6757 
6758 	if (nested)
6759 		nested_vmx_setup_ctls_msrs(&vmx->nested.msrs,
6760 					   vmx_capability.ept,
6761 					   kvm_vcpu_apicv_active(vcpu));
6762 	else
6763 		memset(&vmx->nested.msrs, 0, sizeof(vmx->nested.msrs));
6764 
6765 	vmx->nested.posted_intr_nv = -1;
6766 	vmx->nested.current_vmptr = -1ull;
6767 
6768 	vcpu->arch.microcode_version = 0x100000000ULL;
6769 	vmx->msr_ia32_feature_control_valid_bits = FEAT_CTL_LOCKED;
6770 
6771 	/*
6772 	 * Enforce invariant: pi_desc.nv is always either POSTED_INTR_VECTOR
6773 	 * or POSTED_INTR_WAKEUP_VECTOR.
6774 	 */
6775 	vmx->pi_desc.nv = POSTED_INTR_VECTOR;
6776 	vmx->pi_desc.sn = 1;
6777 
6778 	vmx->ept_pointer = INVALID_PAGE;
6779 
6780 	return 0;
6781 
6782 free_vmcs:
6783 	free_loaded_vmcs(vmx->loaded_vmcs);
6784 free_pml:
6785 	vmx_destroy_pml_buffer(vmx);
6786 free_vpid:
6787 	free_vpid(vmx->vpid);
6788 	return err;
6789 }
6790 
6791 #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"
6792 #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"
6793 
6794 static int vmx_vm_init(struct kvm *kvm)
6795 {
6796 	spin_lock_init(&to_kvm_vmx(kvm)->ept_pointer_lock);
6797 
6798 	if (!ple_gap)
6799 		kvm->arch.pause_in_guest = true;
6800 
6801 	if (boot_cpu_has(X86_BUG_L1TF) && enable_ept) {
6802 		switch (l1tf_mitigation) {
6803 		case L1TF_MITIGATION_OFF:
6804 		case L1TF_MITIGATION_FLUSH_NOWARN:
6805 			/* 'I explicitly don't care' is set */
6806 			break;
6807 		case L1TF_MITIGATION_FLUSH:
6808 		case L1TF_MITIGATION_FLUSH_NOSMT:
6809 		case L1TF_MITIGATION_FULL:
6810 			/*
6811 			 * Warn upon starting the first VM in a potentially
6812 			 * insecure environment.
6813 			 */
6814 			if (sched_smt_active())
6815 				pr_warn_once(L1TF_MSG_SMT);
6816 			if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_NEVER)
6817 				pr_warn_once(L1TF_MSG_L1D);
6818 			break;
6819 		case L1TF_MITIGATION_FULL_FORCE:
6820 			/* Flush is enforced */
6821 			break;
6822 		}
6823 	}
6824 	kvm_apicv_init(kvm, enable_apicv);
6825 	return 0;
6826 }
6827 
6828 static int __init vmx_check_processor_compat(void)
6829 {
6830 	struct vmcs_config vmcs_conf;
6831 	struct vmx_capability vmx_cap;
6832 
6833 	if (!this_cpu_has(X86_FEATURE_MSR_IA32_FEAT_CTL) ||
6834 	    !this_cpu_has(X86_FEATURE_VMX)) {
6835 		pr_err("kvm: VMX is disabled on CPU %d\n", smp_processor_id());
6836 		return -EIO;
6837 	}
6838 
6839 	if (setup_vmcs_config(&vmcs_conf, &vmx_cap) < 0)
6840 		return -EIO;
6841 	if (nested)
6842 		nested_vmx_setup_ctls_msrs(&vmcs_conf.nested, vmx_cap.ept,
6843 					   enable_apicv);
6844 	if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
6845 		printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
6846 				smp_processor_id());
6847 		return -EIO;
6848 	}
6849 	return 0;
6850 }
6851 
6852 static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
6853 {
6854 	u8 cache;
6855 	u64 ipat = 0;
6856 
6857 	/* For VT-d and EPT combination
6858 	 * 1. MMIO: always map as UC
6859 	 * 2. EPT with VT-d:
6860 	 *   a. VT-d without snooping control feature: can't guarantee the
6861 	 *	result, try to trust guest.
6862 	 *   b. VT-d with snooping control feature: snooping control feature of
6863 	 *	VT-d engine can guarantee the cache correctness. Just set it
6864 	 *	to WB to keep consistent with host. So the same as item 3.
6865 	 * 3. EPT without VT-d: always map as WB and set IPAT=1 to keep
6866 	 *    consistent with host MTRR
6867 	 */
6868 	if (is_mmio) {
6869 		cache = MTRR_TYPE_UNCACHABLE;
6870 		goto exit;
6871 	}
6872 
6873 	if (!kvm_arch_has_noncoherent_dma(vcpu->kvm)) {
6874 		ipat = VMX_EPT_IPAT_BIT;
6875 		cache = MTRR_TYPE_WRBACK;
6876 		goto exit;
6877 	}
6878 
6879 	if (kvm_read_cr0(vcpu) & X86_CR0_CD) {
6880 		ipat = VMX_EPT_IPAT_BIT;
6881 		if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
6882 			cache = MTRR_TYPE_WRBACK;
6883 		else
6884 			cache = MTRR_TYPE_UNCACHABLE;
6885 		goto exit;
6886 	}
6887 
6888 	cache = kvm_mtrr_get_guest_memory_type(vcpu, gfn);
6889 
6890 exit:
6891 	return (cache << VMX_EPT_MT_EPTE_SHIFT) | ipat;
6892 }
6893 
6894 static int vmx_get_lpage_level(void)
6895 {
6896 	if (enable_ept && !cpu_has_vmx_ept_1g_page())
6897 		return PT_DIRECTORY_LEVEL;
6898 	else
6899 		/* For shadow and EPT supported 1GB page */
6900 		return PT_PDPE_LEVEL;
6901 }
6902 
6903 static void vmcs_set_secondary_exec_control(struct vcpu_vmx *vmx)
6904 {
6905 	/*
6906 	 * These bits in the secondary execution controls field
6907 	 * are dynamic, the others are mostly based on the hypervisor
6908 	 * architecture and the guest's CPUID.  Do not touch the
6909 	 * dynamic bits.
6910 	 */
6911 	u32 mask =
6912 		SECONDARY_EXEC_SHADOW_VMCS |
6913 		SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
6914 		SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
6915 		SECONDARY_EXEC_DESC;
6916 
6917 	u32 new_ctl = vmx->secondary_exec_control;
6918 	u32 cur_ctl = secondary_exec_controls_get(vmx);
6919 
6920 	secondary_exec_controls_set(vmx, (new_ctl & ~mask) | (cur_ctl & mask));
6921 }
6922 
6923 /*
6924  * Generate MSR_IA32_VMX_CR{0,4}_FIXED1 according to CPUID. Only set bits
6925  * (indicating "allowed-1") if they are supported in the guest's CPUID.
6926  */
6927 static void nested_vmx_cr_fixed1_bits_update(struct kvm_vcpu *vcpu)
6928 {
6929 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6930 	struct kvm_cpuid_entry2 *entry;
6931 
6932 	vmx->nested.msrs.cr0_fixed1 = 0xffffffff;
6933 	vmx->nested.msrs.cr4_fixed1 = X86_CR4_PCE;
6934 
6935 #define cr4_fixed1_update(_cr4_mask, _reg, _cpuid_mask) do {		\
6936 	if (entry && (entry->_reg & (_cpuid_mask)))			\
6937 		vmx->nested.msrs.cr4_fixed1 |= (_cr4_mask);	\
6938 } while (0)
6939 
6940 	entry = kvm_find_cpuid_entry(vcpu, 0x1, 0);
6941 	cr4_fixed1_update(X86_CR4_VME,        edx, feature_bit(VME));
6942 	cr4_fixed1_update(X86_CR4_PVI,        edx, feature_bit(VME));
6943 	cr4_fixed1_update(X86_CR4_TSD,        edx, feature_bit(TSC));
6944 	cr4_fixed1_update(X86_CR4_DE,         edx, feature_bit(DE));
6945 	cr4_fixed1_update(X86_CR4_PSE,        edx, feature_bit(PSE));
6946 	cr4_fixed1_update(X86_CR4_PAE,        edx, feature_bit(PAE));
6947 	cr4_fixed1_update(X86_CR4_MCE,        edx, feature_bit(MCE));
6948 	cr4_fixed1_update(X86_CR4_PGE,        edx, feature_bit(PGE));
6949 	cr4_fixed1_update(X86_CR4_OSFXSR,     edx, feature_bit(FXSR));
6950 	cr4_fixed1_update(X86_CR4_OSXMMEXCPT, edx, feature_bit(XMM));
6951 	cr4_fixed1_update(X86_CR4_VMXE,       ecx, feature_bit(VMX));
6952 	cr4_fixed1_update(X86_CR4_SMXE,       ecx, feature_bit(SMX));
6953 	cr4_fixed1_update(X86_CR4_PCIDE,      ecx, feature_bit(PCID));
6954 	cr4_fixed1_update(X86_CR4_OSXSAVE,    ecx, feature_bit(XSAVE));
6955 
6956 	entry = kvm_find_cpuid_entry(vcpu, 0x7, 0);
6957 	cr4_fixed1_update(X86_CR4_FSGSBASE,   ebx, feature_bit(FSGSBASE));
6958 	cr4_fixed1_update(X86_CR4_SMEP,       ebx, feature_bit(SMEP));
6959 	cr4_fixed1_update(X86_CR4_SMAP,       ebx, feature_bit(SMAP));
6960 	cr4_fixed1_update(X86_CR4_PKE,        ecx, feature_bit(PKU));
6961 	cr4_fixed1_update(X86_CR4_UMIP,       ecx, feature_bit(UMIP));
6962 	cr4_fixed1_update(X86_CR4_LA57,       ecx, feature_bit(LA57));
6963 
6964 #undef cr4_fixed1_update
6965 }
6966 
6967 static void nested_vmx_entry_exit_ctls_update(struct kvm_vcpu *vcpu)
6968 {
6969 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6970 
6971 	if (kvm_mpx_supported()) {
6972 		bool mpx_enabled = guest_cpuid_has(vcpu, X86_FEATURE_MPX);
6973 
6974 		if (mpx_enabled) {
6975 			vmx->nested.msrs.entry_ctls_high |= VM_ENTRY_LOAD_BNDCFGS;
6976 			vmx->nested.msrs.exit_ctls_high |= VM_EXIT_CLEAR_BNDCFGS;
6977 		} else {
6978 			vmx->nested.msrs.entry_ctls_high &= ~VM_ENTRY_LOAD_BNDCFGS;
6979 			vmx->nested.msrs.exit_ctls_high &= ~VM_EXIT_CLEAR_BNDCFGS;
6980 		}
6981 	}
6982 }
6983 
6984 static void update_intel_pt_cfg(struct kvm_vcpu *vcpu)
6985 {
6986 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6987 	struct kvm_cpuid_entry2 *best = NULL;
6988 	int i;
6989 
6990 	for (i = 0; i < PT_CPUID_LEAVES; i++) {
6991 		best = kvm_find_cpuid_entry(vcpu, 0x14, i);
6992 		if (!best)
6993 			return;
6994 		vmx->pt_desc.caps[CPUID_EAX + i*PT_CPUID_REGS_NUM] = best->eax;
6995 		vmx->pt_desc.caps[CPUID_EBX + i*PT_CPUID_REGS_NUM] = best->ebx;
6996 		vmx->pt_desc.caps[CPUID_ECX + i*PT_CPUID_REGS_NUM] = best->ecx;
6997 		vmx->pt_desc.caps[CPUID_EDX + i*PT_CPUID_REGS_NUM] = best->edx;
6998 	}
6999 
7000 	/* Get the number of configurable Address Ranges for filtering */
7001 	vmx->pt_desc.addr_range = intel_pt_validate_cap(vmx->pt_desc.caps,
7002 						PT_CAP_num_address_ranges);
7003 
7004 	/* Initialize and clear the no dependency bits */
7005 	vmx->pt_desc.ctl_bitmask = ~(RTIT_CTL_TRACEEN | RTIT_CTL_OS |
7006 			RTIT_CTL_USR | RTIT_CTL_TSC_EN | RTIT_CTL_DISRETC);
7007 
7008 	/*
7009 	 * If CPUID.(EAX=14H,ECX=0):EBX[0]=1 CR3Filter can be set otherwise
7010 	 * will inject an #GP
7011 	 */
7012 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_cr3_filtering))
7013 		vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_CR3EN;
7014 
7015 	/*
7016 	 * If CPUID.(EAX=14H,ECX=0):EBX[1]=1 CYCEn, CycThresh and
7017 	 * PSBFreq can be set
7018 	 */
7019 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc))
7020 		vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_CYCLEACC |
7021 				RTIT_CTL_CYC_THRESH | RTIT_CTL_PSB_FREQ);
7022 
7023 	/*
7024 	 * If CPUID.(EAX=14H,ECX=0):EBX[3]=1 MTCEn BranchEn and
7025 	 * MTCFreq can be set
7026 	 */
7027 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc))
7028 		vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_MTC_EN |
7029 				RTIT_CTL_BRANCH_EN | RTIT_CTL_MTC_RANGE);
7030 
7031 	/* If CPUID.(EAX=14H,ECX=0):EBX[4]=1 FUPonPTW and PTWEn can be set */
7032 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_ptwrite))
7033 		vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_FUP_ON_PTW |
7034 							RTIT_CTL_PTW_EN);
7035 
7036 	/* If CPUID.(EAX=14H,ECX=0):EBX[5]=1 PwrEvEn can be set */
7037 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_power_event_trace))
7038 		vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_PWR_EVT_EN;
7039 
7040 	/* If CPUID.(EAX=14H,ECX=0):ECX[0]=1 ToPA can be set */
7041 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_topa_output))
7042 		vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_TOPA;
7043 
7044 	/* If CPUID.(EAX=14H,ECX=0):ECX[3]=1 FabircEn can be set */
7045 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_output_subsys))
7046 		vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_FABRIC_EN;
7047 
7048 	/* unmask address range configure area */
7049 	for (i = 0; i < vmx->pt_desc.addr_range; i++)
7050 		vmx->pt_desc.ctl_bitmask &= ~(0xfULL << (32 + i * 4));
7051 }
7052 
7053 static void vmx_cpuid_update(struct kvm_vcpu *vcpu)
7054 {
7055 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7056 
7057 	/* xsaves_enabled is recomputed in vmx_compute_secondary_exec_control(). */
7058 	vcpu->arch.xsaves_enabled = false;
7059 
7060 	if (cpu_has_secondary_exec_ctrls()) {
7061 		vmx_compute_secondary_exec_control(vmx);
7062 		vmcs_set_secondary_exec_control(vmx);
7063 	}
7064 
7065 	if (nested_vmx_allowed(vcpu))
7066 		to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
7067 			FEAT_CTL_VMX_ENABLED_INSIDE_SMX |
7068 			FEAT_CTL_VMX_ENABLED_OUTSIDE_SMX;
7069 	else
7070 		to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
7071 			~(FEAT_CTL_VMX_ENABLED_INSIDE_SMX |
7072 			  FEAT_CTL_VMX_ENABLED_OUTSIDE_SMX);
7073 
7074 	if (nested_vmx_allowed(vcpu)) {
7075 		nested_vmx_cr_fixed1_bits_update(vcpu);
7076 		nested_vmx_entry_exit_ctls_update(vcpu);
7077 	}
7078 
7079 	if (boot_cpu_has(X86_FEATURE_INTEL_PT) &&
7080 			guest_cpuid_has(vcpu, X86_FEATURE_INTEL_PT))
7081 		update_intel_pt_cfg(vcpu);
7082 
7083 	if (boot_cpu_has(X86_FEATURE_RTM)) {
7084 		struct shared_msr_entry *msr;
7085 		msr = find_msr_entry(vmx, MSR_IA32_TSX_CTRL);
7086 		if (msr) {
7087 			bool enabled = guest_cpuid_has(vcpu, X86_FEATURE_RTM);
7088 			vmx_set_guest_msr(vmx, msr, enabled ? 0 : TSX_CTRL_RTM_DISABLE);
7089 		}
7090 	}
7091 }
7092 
7093 static void vmx_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
7094 {
7095 	if (func == 1 && nested)
7096 		entry->ecx |= feature_bit(VMX);
7097 }
7098 
7099 static void vmx_request_immediate_exit(struct kvm_vcpu *vcpu)
7100 {
7101 	to_vmx(vcpu)->req_immediate_exit = true;
7102 }
7103 
7104 static int vmx_check_intercept(struct kvm_vcpu *vcpu,
7105 			       struct x86_instruction_info *info,
7106 			       enum x86_intercept_stage stage)
7107 {
7108 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
7109 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
7110 
7111 	/*
7112 	 * RDPID causes #UD if disabled through secondary execution controls.
7113 	 * Because it is marked as EmulateOnUD, we need to intercept it here.
7114 	 */
7115 	if (info->intercept == x86_intercept_rdtscp &&
7116 	    !nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDTSCP)) {
7117 		ctxt->exception.vector = UD_VECTOR;
7118 		ctxt->exception.error_code_valid = false;
7119 		return X86EMUL_PROPAGATE_FAULT;
7120 	}
7121 
7122 	/* TODO: check more intercepts... */
7123 	return X86EMUL_CONTINUE;
7124 }
7125 
7126 #ifdef CONFIG_X86_64
7127 /* (a << shift) / divisor, return 1 if overflow otherwise 0 */
7128 static inline int u64_shl_div_u64(u64 a, unsigned int shift,
7129 				  u64 divisor, u64 *result)
7130 {
7131 	u64 low = a << shift, high = a >> (64 - shift);
7132 
7133 	/* To avoid the overflow on divq */
7134 	if (high >= divisor)
7135 		return 1;
7136 
7137 	/* Low hold the result, high hold rem which is discarded */
7138 	asm("divq %2\n\t" : "=a" (low), "=d" (high) :
7139 	    "rm" (divisor), "0" (low), "1" (high));
7140 	*result = low;
7141 
7142 	return 0;
7143 }
7144 
7145 static int vmx_set_hv_timer(struct kvm_vcpu *vcpu, u64 guest_deadline_tsc,
7146 			    bool *expired)
7147 {
7148 	struct vcpu_vmx *vmx;
7149 	u64 tscl, guest_tscl, delta_tsc, lapic_timer_advance_cycles;
7150 	struct kvm_timer *ktimer = &vcpu->arch.apic->lapic_timer;
7151 
7152 	if (kvm_mwait_in_guest(vcpu->kvm) ||
7153 		kvm_can_post_timer_interrupt(vcpu))
7154 		return -EOPNOTSUPP;
7155 
7156 	vmx = to_vmx(vcpu);
7157 	tscl = rdtsc();
7158 	guest_tscl = kvm_read_l1_tsc(vcpu, tscl);
7159 	delta_tsc = max(guest_deadline_tsc, guest_tscl) - guest_tscl;
7160 	lapic_timer_advance_cycles = nsec_to_cycles(vcpu,
7161 						    ktimer->timer_advance_ns);
7162 
7163 	if (delta_tsc > lapic_timer_advance_cycles)
7164 		delta_tsc -= lapic_timer_advance_cycles;
7165 	else
7166 		delta_tsc = 0;
7167 
7168 	/* Convert to host delta tsc if tsc scaling is enabled */
7169 	if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio &&
7170 	    delta_tsc && u64_shl_div_u64(delta_tsc,
7171 				kvm_tsc_scaling_ratio_frac_bits,
7172 				vcpu->arch.tsc_scaling_ratio, &delta_tsc))
7173 		return -ERANGE;
7174 
7175 	/*
7176 	 * If the delta tsc can't fit in the 32 bit after the multi shift,
7177 	 * we can't use the preemption timer.
7178 	 * It's possible that it fits on later vmentries, but checking
7179 	 * on every vmentry is costly so we just use an hrtimer.
7180 	 */
7181 	if (delta_tsc >> (cpu_preemption_timer_multi + 32))
7182 		return -ERANGE;
7183 
7184 	vmx->hv_deadline_tsc = tscl + delta_tsc;
7185 	*expired = !delta_tsc;
7186 	return 0;
7187 }
7188 
7189 static void vmx_cancel_hv_timer(struct kvm_vcpu *vcpu)
7190 {
7191 	to_vmx(vcpu)->hv_deadline_tsc = -1;
7192 }
7193 #endif
7194 
7195 static void vmx_sched_in(struct kvm_vcpu *vcpu, int cpu)
7196 {
7197 	if (!kvm_pause_in_guest(vcpu->kvm))
7198 		shrink_ple_window(vcpu);
7199 }
7200 
7201 static void vmx_slot_enable_log_dirty(struct kvm *kvm,
7202 				     struct kvm_memory_slot *slot)
7203 {
7204 	kvm_mmu_slot_leaf_clear_dirty(kvm, slot);
7205 	kvm_mmu_slot_largepage_remove_write_access(kvm, slot);
7206 }
7207 
7208 static void vmx_slot_disable_log_dirty(struct kvm *kvm,
7209 				       struct kvm_memory_slot *slot)
7210 {
7211 	kvm_mmu_slot_set_dirty(kvm, slot);
7212 }
7213 
7214 static void vmx_flush_log_dirty(struct kvm *kvm)
7215 {
7216 	kvm_flush_pml_buffers(kvm);
7217 }
7218 
7219 static int vmx_write_pml_buffer(struct kvm_vcpu *vcpu)
7220 {
7221 	struct vmcs12 *vmcs12;
7222 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7223 	gpa_t gpa, dst;
7224 
7225 	if (is_guest_mode(vcpu)) {
7226 		WARN_ON_ONCE(vmx->nested.pml_full);
7227 
7228 		/*
7229 		 * Check if PML is enabled for the nested guest.
7230 		 * Whether eptp bit 6 is set is already checked
7231 		 * as part of A/D emulation.
7232 		 */
7233 		vmcs12 = get_vmcs12(vcpu);
7234 		if (!nested_cpu_has_pml(vmcs12))
7235 			return 0;
7236 
7237 		if (vmcs12->guest_pml_index >= PML_ENTITY_NUM) {
7238 			vmx->nested.pml_full = true;
7239 			return 1;
7240 		}
7241 
7242 		gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS) & ~0xFFFull;
7243 		dst = vmcs12->pml_address + sizeof(u64) * vmcs12->guest_pml_index;
7244 
7245 		if (kvm_write_guest_page(vcpu->kvm, gpa_to_gfn(dst), &gpa,
7246 					 offset_in_page(dst), sizeof(gpa)))
7247 			return 0;
7248 
7249 		vmcs12->guest_pml_index--;
7250 	}
7251 
7252 	return 0;
7253 }
7254 
7255 static void vmx_enable_log_dirty_pt_masked(struct kvm *kvm,
7256 					   struct kvm_memory_slot *memslot,
7257 					   gfn_t offset, unsigned long mask)
7258 {
7259 	kvm_mmu_clear_dirty_pt_masked(kvm, memslot, offset, mask);
7260 }
7261 
7262 static void __pi_post_block(struct kvm_vcpu *vcpu)
7263 {
7264 	struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
7265 	struct pi_desc old, new;
7266 	unsigned int dest;
7267 
7268 	do {
7269 		old.control = new.control = pi_desc->control;
7270 		WARN(old.nv != POSTED_INTR_WAKEUP_VECTOR,
7271 		     "Wakeup handler not enabled while the VCPU is blocked\n");
7272 
7273 		dest = cpu_physical_id(vcpu->cpu);
7274 
7275 		if (x2apic_enabled())
7276 			new.ndst = dest;
7277 		else
7278 			new.ndst = (dest << 8) & 0xFF00;
7279 
7280 		/* set 'NV' to 'notification vector' */
7281 		new.nv = POSTED_INTR_VECTOR;
7282 	} while (cmpxchg64(&pi_desc->control, old.control,
7283 			   new.control) != old.control);
7284 
7285 	if (!WARN_ON_ONCE(vcpu->pre_pcpu == -1)) {
7286 		spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7287 		list_del(&vcpu->blocked_vcpu_list);
7288 		spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7289 		vcpu->pre_pcpu = -1;
7290 	}
7291 }
7292 
7293 /*
7294  * This routine does the following things for vCPU which is going
7295  * to be blocked if VT-d PI is enabled.
7296  * - Store the vCPU to the wakeup list, so when interrupts happen
7297  *   we can find the right vCPU to wake up.
7298  * - Change the Posted-interrupt descriptor as below:
7299  *      'NDST' <-- vcpu->pre_pcpu
7300  *      'NV' <-- POSTED_INTR_WAKEUP_VECTOR
7301  * - If 'ON' is set during this process, which means at least one
7302  *   interrupt is posted for this vCPU, we cannot block it, in
7303  *   this case, return 1, otherwise, return 0.
7304  *
7305  */
7306 static int pi_pre_block(struct kvm_vcpu *vcpu)
7307 {
7308 	unsigned int dest;
7309 	struct pi_desc old, new;
7310 	struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
7311 
7312 	if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
7313 		!irq_remapping_cap(IRQ_POSTING_CAP)  ||
7314 		!kvm_vcpu_apicv_active(vcpu))
7315 		return 0;
7316 
7317 	WARN_ON(irqs_disabled());
7318 	local_irq_disable();
7319 	if (!WARN_ON_ONCE(vcpu->pre_pcpu != -1)) {
7320 		vcpu->pre_pcpu = vcpu->cpu;
7321 		spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7322 		list_add_tail(&vcpu->blocked_vcpu_list,
7323 			      &per_cpu(blocked_vcpu_on_cpu,
7324 				       vcpu->pre_pcpu));
7325 		spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7326 	}
7327 
7328 	do {
7329 		old.control = new.control = pi_desc->control;
7330 
7331 		WARN((pi_desc->sn == 1),
7332 		     "Warning: SN field of posted-interrupts "
7333 		     "is set before blocking\n");
7334 
7335 		/*
7336 		 * Since vCPU can be preempted during this process,
7337 		 * vcpu->cpu could be different with pre_pcpu, we
7338 		 * need to set pre_pcpu as the destination of wakeup
7339 		 * notification event, then we can find the right vCPU
7340 		 * to wakeup in wakeup handler if interrupts happen
7341 		 * when the vCPU is in blocked state.
7342 		 */
7343 		dest = cpu_physical_id(vcpu->pre_pcpu);
7344 
7345 		if (x2apic_enabled())
7346 			new.ndst = dest;
7347 		else
7348 			new.ndst = (dest << 8) & 0xFF00;
7349 
7350 		/* set 'NV' to 'wakeup vector' */
7351 		new.nv = POSTED_INTR_WAKEUP_VECTOR;
7352 	} while (cmpxchg64(&pi_desc->control, old.control,
7353 			   new.control) != old.control);
7354 
7355 	/* We should not block the vCPU if an interrupt is posted for it.  */
7356 	if (pi_test_on(pi_desc) == 1)
7357 		__pi_post_block(vcpu);
7358 
7359 	local_irq_enable();
7360 	return (vcpu->pre_pcpu == -1);
7361 }
7362 
7363 static int vmx_pre_block(struct kvm_vcpu *vcpu)
7364 {
7365 	if (pi_pre_block(vcpu))
7366 		return 1;
7367 
7368 	if (kvm_lapic_hv_timer_in_use(vcpu))
7369 		kvm_lapic_switch_to_sw_timer(vcpu);
7370 
7371 	return 0;
7372 }
7373 
7374 static void pi_post_block(struct kvm_vcpu *vcpu)
7375 {
7376 	if (vcpu->pre_pcpu == -1)
7377 		return;
7378 
7379 	WARN_ON(irqs_disabled());
7380 	local_irq_disable();
7381 	__pi_post_block(vcpu);
7382 	local_irq_enable();
7383 }
7384 
7385 static void vmx_post_block(struct kvm_vcpu *vcpu)
7386 {
7387 	if (kvm_x86_ops->set_hv_timer)
7388 		kvm_lapic_switch_to_hv_timer(vcpu);
7389 
7390 	pi_post_block(vcpu);
7391 }
7392 
7393 /*
7394  * vmx_update_pi_irte - set IRTE for Posted-Interrupts
7395  *
7396  * @kvm: kvm
7397  * @host_irq: host irq of the interrupt
7398  * @guest_irq: gsi of the interrupt
7399  * @set: set or unset PI
7400  * returns 0 on success, < 0 on failure
7401  */
7402 static int vmx_update_pi_irte(struct kvm *kvm, unsigned int host_irq,
7403 			      uint32_t guest_irq, bool set)
7404 {
7405 	struct kvm_kernel_irq_routing_entry *e;
7406 	struct kvm_irq_routing_table *irq_rt;
7407 	struct kvm_lapic_irq irq;
7408 	struct kvm_vcpu *vcpu;
7409 	struct vcpu_data vcpu_info;
7410 	int idx, ret = 0;
7411 
7412 	if (!kvm_arch_has_assigned_device(kvm) ||
7413 		!irq_remapping_cap(IRQ_POSTING_CAP) ||
7414 		!kvm_vcpu_apicv_active(kvm->vcpus[0]))
7415 		return 0;
7416 
7417 	idx = srcu_read_lock(&kvm->irq_srcu);
7418 	irq_rt = srcu_dereference(kvm->irq_routing, &kvm->irq_srcu);
7419 	if (guest_irq >= irq_rt->nr_rt_entries ||
7420 	    hlist_empty(&irq_rt->map[guest_irq])) {
7421 		pr_warn_once("no route for guest_irq %u/%u (broken user space?)\n",
7422 			     guest_irq, irq_rt->nr_rt_entries);
7423 		goto out;
7424 	}
7425 
7426 	hlist_for_each_entry(e, &irq_rt->map[guest_irq], link) {
7427 		if (e->type != KVM_IRQ_ROUTING_MSI)
7428 			continue;
7429 		/*
7430 		 * VT-d PI cannot support posting multicast/broadcast
7431 		 * interrupts to a vCPU, we still use interrupt remapping
7432 		 * for these kind of interrupts.
7433 		 *
7434 		 * For lowest-priority interrupts, we only support
7435 		 * those with single CPU as the destination, e.g. user
7436 		 * configures the interrupts via /proc/irq or uses
7437 		 * irqbalance to make the interrupts single-CPU.
7438 		 *
7439 		 * We will support full lowest-priority interrupt later.
7440 		 *
7441 		 * In addition, we can only inject generic interrupts using
7442 		 * the PI mechanism, refuse to route others through it.
7443 		 */
7444 
7445 		kvm_set_msi_irq(kvm, e, &irq);
7446 		if (!kvm_intr_is_single_vcpu(kvm, &irq, &vcpu) ||
7447 		    !kvm_irq_is_postable(&irq)) {
7448 			/*
7449 			 * Make sure the IRTE is in remapped mode if
7450 			 * we don't handle it in posted mode.
7451 			 */
7452 			ret = irq_set_vcpu_affinity(host_irq, NULL);
7453 			if (ret < 0) {
7454 				printk(KERN_INFO
7455 				   "failed to back to remapped mode, irq: %u\n",
7456 				   host_irq);
7457 				goto out;
7458 			}
7459 
7460 			continue;
7461 		}
7462 
7463 		vcpu_info.pi_desc_addr = __pa(vcpu_to_pi_desc(vcpu));
7464 		vcpu_info.vector = irq.vector;
7465 
7466 		trace_kvm_pi_irte_update(host_irq, vcpu->vcpu_id, e->gsi,
7467 				vcpu_info.vector, vcpu_info.pi_desc_addr, set);
7468 
7469 		if (set)
7470 			ret = irq_set_vcpu_affinity(host_irq, &vcpu_info);
7471 		else
7472 			ret = irq_set_vcpu_affinity(host_irq, NULL);
7473 
7474 		if (ret < 0) {
7475 			printk(KERN_INFO "%s: failed to update PI IRTE\n",
7476 					__func__);
7477 			goto out;
7478 		}
7479 	}
7480 
7481 	ret = 0;
7482 out:
7483 	srcu_read_unlock(&kvm->irq_srcu, idx);
7484 	return ret;
7485 }
7486 
7487 static void vmx_setup_mce(struct kvm_vcpu *vcpu)
7488 {
7489 	if (vcpu->arch.mcg_cap & MCG_LMCE_P)
7490 		to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
7491 			FEAT_CTL_LMCE_ENABLED;
7492 	else
7493 		to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
7494 			~FEAT_CTL_LMCE_ENABLED;
7495 }
7496 
7497 static int vmx_smi_allowed(struct kvm_vcpu *vcpu)
7498 {
7499 	/* we need a nested vmexit to enter SMM, postpone if run is pending */
7500 	if (to_vmx(vcpu)->nested.nested_run_pending)
7501 		return 0;
7502 	return 1;
7503 }
7504 
7505 static int vmx_pre_enter_smm(struct kvm_vcpu *vcpu, char *smstate)
7506 {
7507 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7508 
7509 	vmx->nested.smm.guest_mode = is_guest_mode(vcpu);
7510 	if (vmx->nested.smm.guest_mode)
7511 		nested_vmx_vmexit(vcpu, -1, 0, 0);
7512 
7513 	vmx->nested.smm.vmxon = vmx->nested.vmxon;
7514 	vmx->nested.vmxon = false;
7515 	vmx_clear_hlt(vcpu);
7516 	return 0;
7517 }
7518 
7519 static int vmx_pre_leave_smm(struct kvm_vcpu *vcpu, const char *smstate)
7520 {
7521 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7522 	int ret;
7523 
7524 	if (vmx->nested.smm.vmxon) {
7525 		vmx->nested.vmxon = true;
7526 		vmx->nested.smm.vmxon = false;
7527 	}
7528 
7529 	if (vmx->nested.smm.guest_mode) {
7530 		ret = nested_vmx_enter_non_root_mode(vcpu, false);
7531 		if (ret)
7532 			return ret;
7533 
7534 		vmx->nested.smm.guest_mode = false;
7535 	}
7536 	return 0;
7537 }
7538 
7539 static int enable_smi_window(struct kvm_vcpu *vcpu)
7540 {
7541 	return 0;
7542 }
7543 
7544 static bool vmx_need_emulation_on_page_fault(struct kvm_vcpu *vcpu)
7545 {
7546 	return false;
7547 }
7548 
7549 static bool vmx_apic_init_signal_blocked(struct kvm_vcpu *vcpu)
7550 {
7551 	return to_vmx(vcpu)->nested.vmxon;
7552 }
7553 
7554 static __init int hardware_setup(void)
7555 {
7556 	unsigned long host_bndcfgs;
7557 	struct desc_ptr dt;
7558 	int r, i;
7559 
7560 	rdmsrl_safe(MSR_EFER, &host_efer);
7561 
7562 	store_idt(&dt);
7563 	host_idt_base = dt.address;
7564 
7565 	for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i)
7566 		kvm_define_shared_msr(i, vmx_msr_index[i]);
7567 
7568 	if (setup_vmcs_config(&vmcs_config, &vmx_capability) < 0)
7569 		return -EIO;
7570 
7571 	if (boot_cpu_has(X86_FEATURE_NX))
7572 		kvm_enable_efer_bits(EFER_NX);
7573 
7574 	if (boot_cpu_has(X86_FEATURE_MPX)) {
7575 		rdmsrl(MSR_IA32_BNDCFGS, host_bndcfgs);
7576 		WARN_ONCE(host_bndcfgs, "KVM: BNDCFGS in host will be lost");
7577 	}
7578 
7579 	if (!cpu_has_vmx_vpid() || !cpu_has_vmx_invvpid() ||
7580 	    !(cpu_has_vmx_invvpid_single() || cpu_has_vmx_invvpid_global()))
7581 		enable_vpid = 0;
7582 
7583 	if (!cpu_has_vmx_ept() ||
7584 	    !cpu_has_vmx_ept_4levels() ||
7585 	    !cpu_has_vmx_ept_mt_wb() ||
7586 	    !cpu_has_vmx_invept_global())
7587 		enable_ept = 0;
7588 
7589 	if (!cpu_has_vmx_ept_ad_bits() || !enable_ept)
7590 		enable_ept_ad_bits = 0;
7591 
7592 	if (!cpu_has_vmx_unrestricted_guest() || !enable_ept)
7593 		enable_unrestricted_guest = 0;
7594 
7595 	if (!cpu_has_vmx_flexpriority())
7596 		flexpriority_enabled = 0;
7597 
7598 	if (!cpu_has_virtual_nmis())
7599 		enable_vnmi = 0;
7600 
7601 	/*
7602 	 * set_apic_access_page_addr() is used to reload apic access
7603 	 * page upon invalidation.  No need to do anything if not
7604 	 * using the APIC_ACCESS_ADDR VMCS field.
7605 	 */
7606 	if (!flexpriority_enabled)
7607 		kvm_x86_ops->set_apic_access_page_addr = NULL;
7608 
7609 	if (!cpu_has_vmx_tpr_shadow())
7610 		kvm_x86_ops->update_cr8_intercept = NULL;
7611 
7612 	if (enable_ept && !cpu_has_vmx_ept_2m_page())
7613 		kvm_disable_largepages();
7614 
7615 #if IS_ENABLED(CONFIG_HYPERV)
7616 	if (ms_hyperv.nested_features & HV_X64_NESTED_GUEST_MAPPING_FLUSH
7617 	    && enable_ept) {
7618 		kvm_x86_ops->tlb_remote_flush = hv_remote_flush_tlb;
7619 		kvm_x86_ops->tlb_remote_flush_with_range =
7620 				hv_remote_flush_tlb_with_range;
7621 	}
7622 #endif
7623 
7624 	if (!cpu_has_vmx_ple()) {
7625 		ple_gap = 0;
7626 		ple_window = 0;
7627 		ple_window_grow = 0;
7628 		ple_window_max = 0;
7629 		ple_window_shrink = 0;
7630 	}
7631 
7632 	if (!cpu_has_vmx_apicv()) {
7633 		enable_apicv = 0;
7634 		kvm_x86_ops->sync_pir_to_irr = NULL;
7635 	}
7636 
7637 	if (cpu_has_vmx_tsc_scaling()) {
7638 		kvm_has_tsc_control = true;
7639 		kvm_max_tsc_scaling_ratio = KVM_VMX_TSC_MULTIPLIER_MAX;
7640 		kvm_tsc_scaling_ratio_frac_bits = 48;
7641 	}
7642 
7643 	set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
7644 
7645 	if (enable_ept)
7646 		vmx_enable_tdp();
7647 	else
7648 		kvm_disable_tdp();
7649 
7650 	/*
7651 	 * Only enable PML when hardware supports PML feature, and both EPT
7652 	 * and EPT A/D bit features are enabled -- PML depends on them to work.
7653 	 */
7654 	if (!enable_ept || !enable_ept_ad_bits || !cpu_has_vmx_pml())
7655 		enable_pml = 0;
7656 
7657 	if (!enable_pml) {
7658 		kvm_x86_ops->slot_enable_log_dirty = NULL;
7659 		kvm_x86_ops->slot_disable_log_dirty = NULL;
7660 		kvm_x86_ops->flush_log_dirty = NULL;
7661 		kvm_x86_ops->enable_log_dirty_pt_masked = NULL;
7662 	}
7663 
7664 	if (!cpu_has_vmx_preemption_timer())
7665 		enable_preemption_timer = false;
7666 
7667 	if (enable_preemption_timer) {
7668 		u64 use_timer_freq = 5000ULL * 1000 * 1000;
7669 		u64 vmx_msr;
7670 
7671 		rdmsrl(MSR_IA32_VMX_MISC, vmx_msr);
7672 		cpu_preemption_timer_multi =
7673 			vmx_msr & VMX_MISC_PREEMPTION_TIMER_RATE_MASK;
7674 
7675 		if (tsc_khz)
7676 			use_timer_freq = (u64)tsc_khz * 1000;
7677 		use_timer_freq >>= cpu_preemption_timer_multi;
7678 
7679 		/*
7680 		 * KVM "disables" the preemption timer by setting it to its max
7681 		 * value.  Don't use the timer if it might cause spurious exits
7682 		 * at a rate faster than 0.1 Hz (of uninterrupted guest time).
7683 		 */
7684 		if (use_timer_freq > 0xffffffffu / 10)
7685 			enable_preemption_timer = false;
7686 	}
7687 
7688 	if (!enable_preemption_timer) {
7689 		kvm_x86_ops->set_hv_timer = NULL;
7690 		kvm_x86_ops->cancel_hv_timer = NULL;
7691 		kvm_x86_ops->request_immediate_exit = __kvm_request_immediate_exit;
7692 	}
7693 
7694 	kvm_set_posted_intr_wakeup_handler(wakeup_handler);
7695 
7696 	kvm_mce_cap_supported |= MCG_LMCE_P;
7697 
7698 	if (pt_mode != PT_MODE_SYSTEM && pt_mode != PT_MODE_HOST_GUEST)
7699 		return -EINVAL;
7700 	if (!enable_ept || !cpu_has_vmx_intel_pt())
7701 		pt_mode = PT_MODE_SYSTEM;
7702 
7703 	if (nested) {
7704 		nested_vmx_setup_ctls_msrs(&vmcs_config.nested,
7705 					   vmx_capability.ept, enable_apicv);
7706 
7707 		r = nested_vmx_hardware_setup(kvm_vmx_exit_handlers);
7708 		if (r)
7709 			return r;
7710 	}
7711 
7712 	r = alloc_kvm_area();
7713 	if (r)
7714 		nested_vmx_hardware_unsetup();
7715 	return r;
7716 }
7717 
7718 static __exit void hardware_unsetup(void)
7719 {
7720 	if (nested)
7721 		nested_vmx_hardware_unsetup();
7722 
7723 	free_kvm_area();
7724 }
7725 
7726 static bool vmx_check_apicv_inhibit_reasons(ulong bit)
7727 {
7728 	ulong supported = BIT(APICV_INHIBIT_REASON_DISABLE) |
7729 			  BIT(APICV_INHIBIT_REASON_HYPERV);
7730 
7731 	return supported & BIT(bit);
7732 }
7733 
7734 static struct kvm_x86_ops vmx_x86_ops __ro_after_init = {
7735 	.cpu_has_kvm_support = cpu_has_kvm_support,
7736 	.disabled_by_bios = vmx_disabled_by_bios,
7737 	.hardware_setup = hardware_setup,
7738 	.hardware_unsetup = hardware_unsetup,
7739 	.check_processor_compatibility = vmx_check_processor_compat,
7740 	.hardware_enable = hardware_enable,
7741 	.hardware_disable = hardware_disable,
7742 	.cpu_has_accelerated_tpr = report_flexpriority,
7743 	.has_emulated_msr = vmx_has_emulated_msr,
7744 
7745 	.vm_init = vmx_vm_init,
7746 	.vm_alloc = vmx_vm_alloc,
7747 	.vm_free = vmx_vm_free,
7748 
7749 	.vcpu_create = vmx_create_vcpu,
7750 	.vcpu_free = vmx_free_vcpu,
7751 	.vcpu_reset = vmx_vcpu_reset,
7752 
7753 	.prepare_guest_switch = vmx_prepare_switch_to_guest,
7754 	.vcpu_load = vmx_vcpu_load,
7755 	.vcpu_put = vmx_vcpu_put,
7756 
7757 	.update_bp_intercept = update_exception_bitmap,
7758 	.get_msr_feature = vmx_get_msr_feature,
7759 	.get_msr = vmx_get_msr,
7760 	.set_msr = vmx_set_msr,
7761 	.get_segment_base = vmx_get_segment_base,
7762 	.get_segment = vmx_get_segment,
7763 	.set_segment = vmx_set_segment,
7764 	.get_cpl = vmx_get_cpl,
7765 	.get_cs_db_l_bits = vmx_get_cs_db_l_bits,
7766 	.decache_cr0_guest_bits = vmx_decache_cr0_guest_bits,
7767 	.decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
7768 	.set_cr0 = vmx_set_cr0,
7769 	.set_cr3 = vmx_set_cr3,
7770 	.set_cr4 = vmx_set_cr4,
7771 	.set_efer = vmx_set_efer,
7772 	.get_idt = vmx_get_idt,
7773 	.set_idt = vmx_set_idt,
7774 	.get_gdt = vmx_get_gdt,
7775 	.set_gdt = vmx_set_gdt,
7776 	.get_dr6 = vmx_get_dr6,
7777 	.set_dr6 = vmx_set_dr6,
7778 	.set_dr7 = vmx_set_dr7,
7779 	.sync_dirty_debug_regs = vmx_sync_dirty_debug_regs,
7780 	.cache_reg = vmx_cache_reg,
7781 	.get_rflags = vmx_get_rflags,
7782 	.set_rflags = vmx_set_rflags,
7783 
7784 	.tlb_flush = vmx_flush_tlb,
7785 	.tlb_flush_gva = vmx_flush_tlb_gva,
7786 
7787 	.run = vmx_vcpu_run,
7788 	.handle_exit = vmx_handle_exit,
7789 	.skip_emulated_instruction = skip_emulated_instruction,
7790 	.set_interrupt_shadow = vmx_set_interrupt_shadow,
7791 	.get_interrupt_shadow = vmx_get_interrupt_shadow,
7792 	.patch_hypercall = vmx_patch_hypercall,
7793 	.set_irq = vmx_inject_irq,
7794 	.set_nmi = vmx_inject_nmi,
7795 	.queue_exception = vmx_queue_exception,
7796 	.cancel_injection = vmx_cancel_injection,
7797 	.interrupt_allowed = vmx_interrupt_allowed,
7798 	.nmi_allowed = vmx_nmi_allowed,
7799 	.get_nmi_mask = vmx_get_nmi_mask,
7800 	.set_nmi_mask = vmx_set_nmi_mask,
7801 	.enable_nmi_window = enable_nmi_window,
7802 	.enable_irq_window = enable_irq_window,
7803 	.update_cr8_intercept = update_cr8_intercept,
7804 	.set_virtual_apic_mode = vmx_set_virtual_apic_mode,
7805 	.set_apic_access_page_addr = vmx_set_apic_access_page_addr,
7806 	.refresh_apicv_exec_ctrl = vmx_refresh_apicv_exec_ctrl,
7807 	.load_eoi_exitmap = vmx_load_eoi_exitmap,
7808 	.apicv_post_state_restore = vmx_apicv_post_state_restore,
7809 	.check_apicv_inhibit_reasons = vmx_check_apicv_inhibit_reasons,
7810 	.hwapic_irr_update = vmx_hwapic_irr_update,
7811 	.hwapic_isr_update = vmx_hwapic_isr_update,
7812 	.guest_apic_has_interrupt = vmx_guest_apic_has_interrupt,
7813 	.sync_pir_to_irr = vmx_sync_pir_to_irr,
7814 	.deliver_posted_interrupt = vmx_deliver_posted_interrupt,
7815 	.dy_apicv_has_pending_interrupt = vmx_dy_apicv_has_pending_interrupt,
7816 
7817 	.set_tss_addr = vmx_set_tss_addr,
7818 	.set_identity_map_addr = vmx_set_identity_map_addr,
7819 	.get_tdp_level = get_ept_level,
7820 	.get_mt_mask = vmx_get_mt_mask,
7821 
7822 	.get_exit_info = vmx_get_exit_info,
7823 
7824 	.get_lpage_level = vmx_get_lpage_level,
7825 
7826 	.cpuid_update = vmx_cpuid_update,
7827 
7828 	.rdtscp_supported = vmx_rdtscp_supported,
7829 	.invpcid_supported = vmx_invpcid_supported,
7830 
7831 	.set_supported_cpuid = vmx_set_supported_cpuid,
7832 
7833 	.has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
7834 
7835 	.read_l1_tsc_offset = vmx_read_l1_tsc_offset,
7836 	.write_l1_tsc_offset = vmx_write_l1_tsc_offset,
7837 
7838 	.set_tdp_cr3 = vmx_set_cr3,
7839 
7840 	.check_intercept = vmx_check_intercept,
7841 	.handle_exit_irqoff = vmx_handle_exit_irqoff,
7842 	.mpx_supported = vmx_mpx_supported,
7843 	.xsaves_supported = vmx_xsaves_supported,
7844 	.umip_emulated = vmx_umip_emulated,
7845 	.pt_supported = vmx_pt_supported,
7846 	.pku_supported = vmx_pku_supported,
7847 
7848 	.request_immediate_exit = vmx_request_immediate_exit,
7849 
7850 	.sched_in = vmx_sched_in,
7851 
7852 	.slot_enable_log_dirty = vmx_slot_enable_log_dirty,
7853 	.slot_disable_log_dirty = vmx_slot_disable_log_dirty,
7854 	.flush_log_dirty = vmx_flush_log_dirty,
7855 	.enable_log_dirty_pt_masked = vmx_enable_log_dirty_pt_masked,
7856 	.write_log_dirty = vmx_write_pml_buffer,
7857 
7858 	.pre_block = vmx_pre_block,
7859 	.post_block = vmx_post_block,
7860 
7861 	.pmu_ops = &intel_pmu_ops,
7862 
7863 	.update_pi_irte = vmx_update_pi_irte,
7864 
7865 #ifdef CONFIG_X86_64
7866 	.set_hv_timer = vmx_set_hv_timer,
7867 	.cancel_hv_timer = vmx_cancel_hv_timer,
7868 #endif
7869 
7870 	.setup_mce = vmx_setup_mce,
7871 
7872 	.smi_allowed = vmx_smi_allowed,
7873 	.pre_enter_smm = vmx_pre_enter_smm,
7874 	.pre_leave_smm = vmx_pre_leave_smm,
7875 	.enable_smi_window = enable_smi_window,
7876 
7877 	.check_nested_events = NULL,
7878 	.get_nested_state = NULL,
7879 	.set_nested_state = NULL,
7880 	.get_vmcs12_pages = NULL,
7881 	.nested_enable_evmcs = NULL,
7882 	.nested_get_evmcs_version = NULL,
7883 	.need_emulation_on_page_fault = vmx_need_emulation_on_page_fault,
7884 	.apic_init_signal_blocked = vmx_apic_init_signal_blocked,
7885 };
7886 
7887 static void vmx_cleanup_l1d_flush(void)
7888 {
7889 	if (vmx_l1d_flush_pages) {
7890 		free_pages((unsigned long)vmx_l1d_flush_pages, L1D_CACHE_ORDER);
7891 		vmx_l1d_flush_pages = NULL;
7892 	}
7893 	/* Restore state so sysfs ignores VMX */
7894 	l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_AUTO;
7895 }
7896 
7897 static void vmx_exit(void)
7898 {
7899 #ifdef CONFIG_KEXEC_CORE
7900 	RCU_INIT_POINTER(crash_vmclear_loaded_vmcss, NULL);
7901 	synchronize_rcu();
7902 #endif
7903 
7904 	kvm_exit();
7905 
7906 #if IS_ENABLED(CONFIG_HYPERV)
7907 	if (static_branch_unlikely(&enable_evmcs)) {
7908 		int cpu;
7909 		struct hv_vp_assist_page *vp_ap;
7910 		/*
7911 		 * Reset everything to support using non-enlightened VMCS
7912 		 * access later (e.g. when we reload the module with
7913 		 * enlightened_vmcs=0)
7914 		 */
7915 		for_each_online_cpu(cpu) {
7916 			vp_ap =	hv_get_vp_assist_page(cpu);
7917 
7918 			if (!vp_ap)
7919 				continue;
7920 
7921 			vp_ap->nested_control.features.directhypercall = 0;
7922 			vp_ap->current_nested_vmcs = 0;
7923 			vp_ap->enlighten_vmentry = 0;
7924 		}
7925 
7926 		static_branch_disable(&enable_evmcs);
7927 	}
7928 #endif
7929 	vmx_cleanup_l1d_flush();
7930 }
7931 module_exit(vmx_exit);
7932 
7933 static int __init vmx_init(void)
7934 {
7935 	int r;
7936 
7937 #if IS_ENABLED(CONFIG_HYPERV)
7938 	/*
7939 	 * Enlightened VMCS usage should be recommended and the host needs
7940 	 * to support eVMCS v1 or above. We can also disable eVMCS support
7941 	 * with module parameter.
7942 	 */
7943 	if (enlightened_vmcs &&
7944 	    ms_hyperv.hints & HV_X64_ENLIGHTENED_VMCS_RECOMMENDED &&
7945 	    (ms_hyperv.nested_features & HV_X64_ENLIGHTENED_VMCS_VERSION) >=
7946 	    KVM_EVMCS_VERSION) {
7947 		int cpu;
7948 
7949 		/* Check that we have assist pages on all online CPUs */
7950 		for_each_online_cpu(cpu) {
7951 			if (!hv_get_vp_assist_page(cpu)) {
7952 				enlightened_vmcs = false;
7953 				break;
7954 			}
7955 		}
7956 
7957 		if (enlightened_vmcs) {
7958 			pr_info("KVM: vmx: using Hyper-V Enlightened VMCS\n");
7959 			static_branch_enable(&enable_evmcs);
7960 		}
7961 
7962 		if (ms_hyperv.nested_features & HV_X64_NESTED_DIRECT_FLUSH)
7963 			vmx_x86_ops.enable_direct_tlbflush
7964 				= hv_enable_direct_tlbflush;
7965 
7966 	} else {
7967 		enlightened_vmcs = false;
7968 	}
7969 #endif
7970 
7971 	r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx),
7972 		     __alignof__(struct vcpu_vmx), THIS_MODULE);
7973 	if (r)
7974 		return r;
7975 
7976 	/*
7977 	 * Must be called after kvm_init() so enable_ept is properly set
7978 	 * up. Hand the parameter mitigation value in which was stored in
7979 	 * the pre module init parser. If no parameter was given, it will
7980 	 * contain 'auto' which will be turned into the default 'cond'
7981 	 * mitigation mode.
7982 	 */
7983 	r = vmx_setup_l1d_flush(vmentry_l1d_flush_param);
7984 	if (r) {
7985 		vmx_exit();
7986 		return r;
7987 	}
7988 
7989 #ifdef CONFIG_KEXEC_CORE
7990 	rcu_assign_pointer(crash_vmclear_loaded_vmcss,
7991 			   crash_vmclear_local_loaded_vmcss);
7992 #endif
7993 	vmx_check_vmcs12_offsets();
7994 
7995 	return 0;
7996 }
7997 module_init(vmx_init);
7998