xref: /openbmc/linux/arch/x86/kvm/vmx/vmx.c (revision 2645d8d0)
1 /*
2  * Kernel-based Virtual Machine driver for Linux
3  *
4  * This module enables machines with Intel VT-x extensions to run virtual
5  * machines without emulation or binary translation.
6  *
7  * Copyright (C) 2006 Qumranet, Inc.
8  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
9  *
10  * Authors:
11  *   Avi Kivity   <avi@qumranet.com>
12  *   Yaniv Kamay  <yaniv@qumranet.com>
13  *
14  * This work is licensed under the terms of the GNU GPL, version 2.  See
15  * the COPYING file in the top-level directory.
16  *
17  */
18 
19 #include <linux/frame.h>
20 #include <linux/highmem.h>
21 #include <linux/hrtimer.h>
22 #include <linux/kernel.h>
23 #include <linux/kvm_host.h>
24 #include <linux/module.h>
25 #include <linux/moduleparam.h>
26 #include <linux/mod_devicetable.h>
27 #include <linux/mm.h>
28 #include <linux/sched.h>
29 #include <linux/sched/smt.h>
30 #include <linux/slab.h>
31 #include <linux/tboot.h>
32 #include <linux/trace_events.h>
33 
34 #include <asm/apic.h>
35 #include <asm/asm.h>
36 #include <asm/cpu.h>
37 #include <asm/debugreg.h>
38 #include <asm/desc.h>
39 #include <asm/fpu/internal.h>
40 #include <asm/io.h>
41 #include <asm/irq_remapping.h>
42 #include <asm/kexec.h>
43 #include <asm/perf_event.h>
44 #include <asm/mce.h>
45 #include <asm/mmu_context.h>
46 #include <asm/mshyperv.h>
47 #include <asm/spec-ctrl.h>
48 #include <asm/virtext.h>
49 #include <asm/vmx.h>
50 
51 #include "capabilities.h"
52 #include "cpuid.h"
53 #include "evmcs.h"
54 #include "irq.h"
55 #include "kvm_cache_regs.h"
56 #include "lapic.h"
57 #include "mmu.h"
58 #include "nested.h"
59 #include "ops.h"
60 #include "pmu.h"
61 #include "trace.h"
62 #include "vmcs.h"
63 #include "vmcs12.h"
64 #include "vmx.h"
65 #include "x86.h"
66 
67 MODULE_AUTHOR("Qumranet");
68 MODULE_LICENSE("GPL");
69 
70 static const struct x86_cpu_id vmx_cpu_id[] = {
71 	X86_FEATURE_MATCH(X86_FEATURE_VMX),
72 	{}
73 };
74 MODULE_DEVICE_TABLE(x86cpu, vmx_cpu_id);
75 
76 bool __read_mostly enable_vpid = 1;
77 module_param_named(vpid, enable_vpid, bool, 0444);
78 
79 static bool __read_mostly enable_vnmi = 1;
80 module_param_named(vnmi, enable_vnmi, bool, S_IRUGO);
81 
82 bool __read_mostly flexpriority_enabled = 1;
83 module_param_named(flexpriority, flexpriority_enabled, bool, S_IRUGO);
84 
85 bool __read_mostly enable_ept = 1;
86 module_param_named(ept, enable_ept, bool, S_IRUGO);
87 
88 bool __read_mostly enable_unrestricted_guest = 1;
89 module_param_named(unrestricted_guest,
90 			enable_unrestricted_guest, bool, S_IRUGO);
91 
92 bool __read_mostly enable_ept_ad_bits = 1;
93 module_param_named(eptad, enable_ept_ad_bits, bool, S_IRUGO);
94 
95 static bool __read_mostly emulate_invalid_guest_state = true;
96 module_param(emulate_invalid_guest_state, bool, S_IRUGO);
97 
98 static bool __read_mostly fasteoi = 1;
99 module_param(fasteoi, bool, S_IRUGO);
100 
101 static bool __read_mostly enable_apicv = 1;
102 module_param(enable_apicv, bool, S_IRUGO);
103 
104 /*
105  * If nested=1, nested virtualization is supported, i.e., guests may use
106  * VMX and be a hypervisor for its own guests. If nested=0, guests may not
107  * use VMX instructions.
108  */
109 static bool __read_mostly nested = 1;
110 module_param(nested, bool, S_IRUGO);
111 
112 static u64 __read_mostly host_xss;
113 
114 bool __read_mostly enable_pml = 1;
115 module_param_named(pml, enable_pml, bool, S_IRUGO);
116 
117 #define MSR_BITMAP_MODE_X2APIC		1
118 #define MSR_BITMAP_MODE_X2APIC_APICV	2
119 
120 #define KVM_VMX_TSC_MULTIPLIER_MAX     0xffffffffffffffffULL
121 
122 /* Guest_tsc -> host_tsc conversion requires 64-bit division.  */
123 static int __read_mostly cpu_preemption_timer_multi;
124 static bool __read_mostly enable_preemption_timer = 1;
125 #ifdef CONFIG_X86_64
126 module_param_named(preemption_timer, enable_preemption_timer, bool, S_IRUGO);
127 #endif
128 
129 #define KVM_VM_CR0_ALWAYS_OFF (X86_CR0_NW | X86_CR0_CD)
130 #define KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST X86_CR0_NE
131 #define KVM_VM_CR0_ALWAYS_ON				\
132 	(KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST | 	\
133 	 X86_CR0_WP | X86_CR0_PG | X86_CR0_PE)
134 #define KVM_CR4_GUEST_OWNED_BITS				      \
135 	(X86_CR4_PVI | X86_CR4_DE | X86_CR4_PCE | X86_CR4_OSFXSR      \
136 	 | X86_CR4_OSXMMEXCPT | X86_CR4_LA57 | X86_CR4_TSD)
137 
138 #define KVM_VM_CR4_ALWAYS_ON_UNRESTRICTED_GUEST X86_CR4_VMXE
139 #define KVM_PMODE_VM_CR4_ALWAYS_ON (X86_CR4_PAE | X86_CR4_VMXE)
140 #define KVM_RMODE_VM_CR4_ALWAYS_ON (X86_CR4_VME | X86_CR4_PAE | X86_CR4_VMXE)
141 
142 #define RMODE_GUEST_OWNED_EFLAGS_BITS (~(X86_EFLAGS_IOPL | X86_EFLAGS_VM))
143 
144 #define MSR_IA32_RTIT_STATUS_MASK (~(RTIT_STATUS_FILTEREN | \
145 	RTIT_STATUS_CONTEXTEN | RTIT_STATUS_TRIGGEREN | \
146 	RTIT_STATUS_ERROR | RTIT_STATUS_STOPPED | \
147 	RTIT_STATUS_BYTECNT))
148 
149 #define MSR_IA32_RTIT_OUTPUT_BASE_MASK \
150 	(~((1UL << cpuid_query_maxphyaddr(vcpu)) - 1) | 0x7f)
151 
152 /*
153  * These 2 parameters are used to config the controls for Pause-Loop Exiting:
154  * ple_gap:    upper bound on the amount of time between two successive
155  *             executions of PAUSE in a loop. Also indicate if ple enabled.
156  *             According to test, this time is usually smaller than 128 cycles.
157  * ple_window: upper bound on the amount of time a guest is allowed to execute
158  *             in a PAUSE loop. Tests indicate that most spinlocks are held for
159  *             less than 2^12 cycles
160  * Time is measured based on a counter that runs at the same rate as the TSC,
161  * refer SDM volume 3b section 21.6.13 & 22.1.3.
162  */
163 static unsigned int ple_gap = KVM_DEFAULT_PLE_GAP;
164 module_param(ple_gap, uint, 0444);
165 
166 static unsigned int ple_window = KVM_VMX_DEFAULT_PLE_WINDOW;
167 module_param(ple_window, uint, 0444);
168 
169 /* Default doubles per-vcpu window every exit. */
170 static unsigned int ple_window_grow = KVM_DEFAULT_PLE_WINDOW_GROW;
171 module_param(ple_window_grow, uint, 0444);
172 
173 /* Default resets per-vcpu window every exit to ple_window. */
174 static unsigned int ple_window_shrink = KVM_DEFAULT_PLE_WINDOW_SHRINK;
175 module_param(ple_window_shrink, uint, 0444);
176 
177 /* Default is to compute the maximum so we can never overflow. */
178 static unsigned int ple_window_max        = KVM_VMX_DEFAULT_PLE_WINDOW_MAX;
179 module_param(ple_window_max, uint, 0444);
180 
181 /* Default is SYSTEM mode, 1 for host-guest mode */
182 int __read_mostly pt_mode = PT_MODE_SYSTEM;
183 module_param(pt_mode, int, S_IRUGO);
184 
185 static DEFINE_STATIC_KEY_FALSE(vmx_l1d_should_flush);
186 static DEFINE_STATIC_KEY_FALSE(vmx_l1d_flush_cond);
187 static DEFINE_MUTEX(vmx_l1d_flush_mutex);
188 
189 /* Storage for pre module init parameter parsing */
190 static enum vmx_l1d_flush_state __read_mostly vmentry_l1d_flush_param = VMENTER_L1D_FLUSH_AUTO;
191 
192 static const struct {
193 	const char *option;
194 	bool for_parse;
195 } vmentry_l1d_param[] = {
196 	[VMENTER_L1D_FLUSH_AUTO]	 = {"auto", true},
197 	[VMENTER_L1D_FLUSH_NEVER]	 = {"never", true},
198 	[VMENTER_L1D_FLUSH_COND]	 = {"cond", true},
199 	[VMENTER_L1D_FLUSH_ALWAYS]	 = {"always", true},
200 	[VMENTER_L1D_FLUSH_EPT_DISABLED] = {"EPT disabled", false},
201 	[VMENTER_L1D_FLUSH_NOT_REQUIRED] = {"not required", false},
202 };
203 
204 #define L1D_CACHE_ORDER 4
205 static void *vmx_l1d_flush_pages;
206 
207 static int vmx_setup_l1d_flush(enum vmx_l1d_flush_state l1tf)
208 {
209 	struct page *page;
210 	unsigned int i;
211 
212 	if (!enable_ept) {
213 		l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_EPT_DISABLED;
214 		return 0;
215 	}
216 
217 	if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES)) {
218 		u64 msr;
219 
220 		rdmsrl(MSR_IA32_ARCH_CAPABILITIES, msr);
221 		if (msr & ARCH_CAP_SKIP_VMENTRY_L1DFLUSH) {
222 			l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_NOT_REQUIRED;
223 			return 0;
224 		}
225 	}
226 
227 	/* If set to auto use the default l1tf mitigation method */
228 	if (l1tf == VMENTER_L1D_FLUSH_AUTO) {
229 		switch (l1tf_mitigation) {
230 		case L1TF_MITIGATION_OFF:
231 			l1tf = VMENTER_L1D_FLUSH_NEVER;
232 			break;
233 		case L1TF_MITIGATION_FLUSH_NOWARN:
234 		case L1TF_MITIGATION_FLUSH:
235 		case L1TF_MITIGATION_FLUSH_NOSMT:
236 			l1tf = VMENTER_L1D_FLUSH_COND;
237 			break;
238 		case L1TF_MITIGATION_FULL:
239 		case L1TF_MITIGATION_FULL_FORCE:
240 			l1tf = VMENTER_L1D_FLUSH_ALWAYS;
241 			break;
242 		}
243 	} else if (l1tf_mitigation == L1TF_MITIGATION_FULL_FORCE) {
244 		l1tf = VMENTER_L1D_FLUSH_ALWAYS;
245 	}
246 
247 	if (l1tf != VMENTER_L1D_FLUSH_NEVER && !vmx_l1d_flush_pages &&
248 	    !boot_cpu_has(X86_FEATURE_FLUSH_L1D)) {
249 		/*
250 		 * This allocation for vmx_l1d_flush_pages is not tied to a VM
251 		 * lifetime and so should not be charged to a memcg.
252 		 */
253 		page = alloc_pages(GFP_KERNEL, L1D_CACHE_ORDER);
254 		if (!page)
255 			return -ENOMEM;
256 		vmx_l1d_flush_pages = page_address(page);
257 
258 		/*
259 		 * Initialize each page with a different pattern in
260 		 * order to protect against KSM in the nested
261 		 * virtualization case.
262 		 */
263 		for (i = 0; i < 1u << L1D_CACHE_ORDER; ++i) {
264 			memset(vmx_l1d_flush_pages + i * PAGE_SIZE, i + 1,
265 			       PAGE_SIZE);
266 		}
267 	}
268 
269 	l1tf_vmx_mitigation = l1tf;
270 
271 	if (l1tf != VMENTER_L1D_FLUSH_NEVER)
272 		static_branch_enable(&vmx_l1d_should_flush);
273 	else
274 		static_branch_disable(&vmx_l1d_should_flush);
275 
276 	if (l1tf == VMENTER_L1D_FLUSH_COND)
277 		static_branch_enable(&vmx_l1d_flush_cond);
278 	else
279 		static_branch_disable(&vmx_l1d_flush_cond);
280 	return 0;
281 }
282 
283 static int vmentry_l1d_flush_parse(const char *s)
284 {
285 	unsigned int i;
286 
287 	if (s) {
288 		for (i = 0; i < ARRAY_SIZE(vmentry_l1d_param); i++) {
289 			if (vmentry_l1d_param[i].for_parse &&
290 			    sysfs_streq(s, vmentry_l1d_param[i].option))
291 				return i;
292 		}
293 	}
294 	return -EINVAL;
295 }
296 
297 static int vmentry_l1d_flush_set(const char *s, const struct kernel_param *kp)
298 {
299 	int l1tf, ret;
300 
301 	l1tf = vmentry_l1d_flush_parse(s);
302 	if (l1tf < 0)
303 		return l1tf;
304 
305 	if (!boot_cpu_has(X86_BUG_L1TF))
306 		return 0;
307 
308 	/*
309 	 * Has vmx_init() run already? If not then this is the pre init
310 	 * parameter parsing. In that case just store the value and let
311 	 * vmx_init() do the proper setup after enable_ept has been
312 	 * established.
313 	 */
314 	if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_AUTO) {
315 		vmentry_l1d_flush_param = l1tf;
316 		return 0;
317 	}
318 
319 	mutex_lock(&vmx_l1d_flush_mutex);
320 	ret = vmx_setup_l1d_flush(l1tf);
321 	mutex_unlock(&vmx_l1d_flush_mutex);
322 	return ret;
323 }
324 
325 static int vmentry_l1d_flush_get(char *s, const struct kernel_param *kp)
326 {
327 	if (WARN_ON_ONCE(l1tf_vmx_mitigation >= ARRAY_SIZE(vmentry_l1d_param)))
328 		return sprintf(s, "???\n");
329 
330 	return sprintf(s, "%s\n", vmentry_l1d_param[l1tf_vmx_mitigation].option);
331 }
332 
333 static const struct kernel_param_ops vmentry_l1d_flush_ops = {
334 	.set = vmentry_l1d_flush_set,
335 	.get = vmentry_l1d_flush_get,
336 };
337 module_param_cb(vmentry_l1d_flush, &vmentry_l1d_flush_ops, NULL, 0644);
338 
339 static bool guest_state_valid(struct kvm_vcpu *vcpu);
340 static u32 vmx_segment_access_rights(struct kvm_segment *var);
341 static __always_inline void vmx_disable_intercept_for_msr(unsigned long *msr_bitmap,
342 							  u32 msr, int type);
343 
344 void vmx_vmexit(void);
345 
346 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
347 DEFINE_PER_CPU(struct vmcs *, current_vmcs);
348 /*
349  * We maintain a per-CPU linked-list of VMCS loaded on that CPU. This is needed
350  * when a CPU is brought down, and we need to VMCLEAR all VMCSs loaded on it.
351  */
352 static DEFINE_PER_CPU(struct list_head, loaded_vmcss_on_cpu);
353 
354 /*
355  * We maintian a per-CPU linked-list of vCPU, so in wakeup_handler() we
356  * can find which vCPU should be waken up.
357  */
358 static DEFINE_PER_CPU(struct list_head, blocked_vcpu_on_cpu);
359 static DEFINE_PER_CPU(spinlock_t, blocked_vcpu_on_cpu_lock);
360 
361 static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
362 static DEFINE_SPINLOCK(vmx_vpid_lock);
363 
364 struct vmcs_config vmcs_config;
365 struct vmx_capability vmx_capability;
366 
367 #define VMX_SEGMENT_FIELD(seg)					\
368 	[VCPU_SREG_##seg] = {                                   \
369 		.selector = GUEST_##seg##_SELECTOR,		\
370 		.base = GUEST_##seg##_BASE,		   	\
371 		.limit = GUEST_##seg##_LIMIT,		   	\
372 		.ar_bytes = GUEST_##seg##_AR_BYTES,	   	\
373 	}
374 
375 static const struct kvm_vmx_segment_field {
376 	unsigned selector;
377 	unsigned base;
378 	unsigned limit;
379 	unsigned ar_bytes;
380 } kvm_vmx_segment_fields[] = {
381 	VMX_SEGMENT_FIELD(CS),
382 	VMX_SEGMENT_FIELD(DS),
383 	VMX_SEGMENT_FIELD(ES),
384 	VMX_SEGMENT_FIELD(FS),
385 	VMX_SEGMENT_FIELD(GS),
386 	VMX_SEGMENT_FIELD(SS),
387 	VMX_SEGMENT_FIELD(TR),
388 	VMX_SEGMENT_FIELD(LDTR),
389 };
390 
391 u64 host_efer;
392 
393 /*
394  * Though SYSCALL is only supported in 64-bit mode on Intel CPUs, kvm
395  * will emulate SYSCALL in legacy mode if the vendor string in guest
396  * CPUID.0:{EBX,ECX,EDX} is "AuthenticAMD" or "AMDisbetter!" To
397  * support this emulation, IA32_STAR must always be included in
398  * vmx_msr_index[], even in i386 builds.
399  */
400 const u32 vmx_msr_index[] = {
401 #ifdef CONFIG_X86_64
402 	MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
403 #endif
404 	MSR_EFER, MSR_TSC_AUX, MSR_STAR,
405 };
406 
407 #if IS_ENABLED(CONFIG_HYPERV)
408 static bool __read_mostly enlightened_vmcs = true;
409 module_param(enlightened_vmcs, bool, 0444);
410 
411 /* check_ept_pointer() should be under protection of ept_pointer_lock. */
412 static void check_ept_pointer_match(struct kvm *kvm)
413 {
414 	struct kvm_vcpu *vcpu;
415 	u64 tmp_eptp = INVALID_PAGE;
416 	int i;
417 
418 	kvm_for_each_vcpu(i, vcpu, kvm) {
419 		if (!VALID_PAGE(tmp_eptp)) {
420 			tmp_eptp = to_vmx(vcpu)->ept_pointer;
421 		} else if (tmp_eptp != to_vmx(vcpu)->ept_pointer) {
422 			to_kvm_vmx(kvm)->ept_pointers_match
423 				= EPT_POINTERS_MISMATCH;
424 			return;
425 		}
426 	}
427 
428 	to_kvm_vmx(kvm)->ept_pointers_match = EPT_POINTERS_MATCH;
429 }
430 
431 static int kvm_fill_hv_flush_list_func(struct hv_guest_mapping_flush_list *flush,
432 		void *data)
433 {
434 	struct kvm_tlb_range *range = data;
435 
436 	return hyperv_fill_flush_guest_mapping_list(flush, range->start_gfn,
437 			range->pages);
438 }
439 
440 static inline int __hv_remote_flush_tlb_with_range(struct kvm *kvm,
441 		struct kvm_vcpu *vcpu, struct kvm_tlb_range *range)
442 {
443 	u64 ept_pointer = to_vmx(vcpu)->ept_pointer;
444 
445 	/*
446 	 * FLUSH_GUEST_PHYSICAL_ADDRESS_SPACE hypercall needs address
447 	 * of the base of EPT PML4 table, strip off EPT configuration
448 	 * information.
449 	 */
450 	if (range)
451 		return hyperv_flush_guest_mapping_range(ept_pointer & PAGE_MASK,
452 				kvm_fill_hv_flush_list_func, (void *)range);
453 	else
454 		return hyperv_flush_guest_mapping(ept_pointer & PAGE_MASK);
455 }
456 
457 static int hv_remote_flush_tlb_with_range(struct kvm *kvm,
458 		struct kvm_tlb_range *range)
459 {
460 	struct kvm_vcpu *vcpu;
461 	int ret = 0, i;
462 
463 	spin_lock(&to_kvm_vmx(kvm)->ept_pointer_lock);
464 
465 	if (to_kvm_vmx(kvm)->ept_pointers_match == EPT_POINTERS_CHECK)
466 		check_ept_pointer_match(kvm);
467 
468 	if (to_kvm_vmx(kvm)->ept_pointers_match != EPT_POINTERS_MATCH) {
469 		kvm_for_each_vcpu(i, vcpu, kvm) {
470 			/* If ept_pointer is invalid pointer, bypass flush request. */
471 			if (VALID_PAGE(to_vmx(vcpu)->ept_pointer))
472 				ret |= __hv_remote_flush_tlb_with_range(
473 					kvm, vcpu, range);
474 		}
475 	} else {
476 		ret = __hv_remote_flush_tlb_with_range(kvm,
477 				kvm_get_vcpu(kvm, 0), range);
478 	}
479 
480 	spin_unlock(&to_kvm_vmx(kvm)->ept_pointer_lock);
481 	return ret;
482 }
483 static int hv_remote_flush_tlb(struct kvm *kvm)
484 {
485 	return hv_remote_flush_tlb_with_range(kvm, NULL);
486 }
487 
488 #endif /* IS_ENABLED(CONFIG_HYPERV) */
489 
490 /*
491  * Comment's format: document - errata name - stepping - processor name.
492  * Refer from
493  * https://www.virtualbox.org/svn/vbox/trunk/src/VBox/VMM/VMMR0/HMR0.cpp
494  */
495 static u32 vmx_preemption_cpu_tfms[] = {
496 /* 323344.pdf - BA86   - D0 - Xeon 7500 Series */
497 0x000206E6,
498 /* 323056.pdf - AAX65  - C2 - Xeon L3406 */
499 /* 322814.pdf - AAT59  - C2 - i7-600, i5-500, i5-400 and i3-300 Mobile */
500 /* 322911.pdf - AAU65  - C2 - i5-600, i3-500 Desktop and Pentium G6950 */
501 0x00020652,
502 /* 322911.pdf - AAU65  - K0 - i5-600, i3-500 Desktop and Pentium G6950 */
503 0x00020655,
504 /* 322373.pdf - AAO95  - B1 - Xeon 3400 Series */
505 /* 322166.pdf - AAN92  - B1 - i7-800 and i5-700 Desktop */
506 /*
507  * 320767.pdf - AAP86  - B1 -
508  * i7-900 Mobile Extreme, i7-800 and i7-700 Mobile
509  */
510 0x000106E5,
511 /* 321333.pdf - AAM126 - C0 - Xeon 3500 */
512 0x000106A0,
513 /* 321333.pdf - AAM126 - C1 - Xeon 3500 */
514 0x000106A1,
515 /* 320836.pdf - AAJ124 - C0 - i7-900 Desktop Extreme and i7-900 Desktop */
516 0x000106A4,
517  /* 321333.pdf - AAM126 - D0 - Xeon 3500 */
518  /* 321324.pdf - AAK139 - D0 - Xeon 5500 */
519  /* 320836.pdf - AAJ124 - D0 - i7-900 Extreme and i7-900 Desktop */
520 0x000106A5,
521  /* Xeon E3-1220 V2 */
522 0x000306A8,
523 };
524 
525 static inline bool cpu_has_broken_vmx_preemption_timer(void)
526 {
527 	u32 eax = cpuid_eax(0x00000001), i;
528 
529 	/* Clear the reserved bits */
530 	eax &= ~(0x3U << 14 | 0xfU << 28);
531 	for (i = 0; i < ARRAY_SIZE(vmx_preemption_cpu_tfms); i++)
532 		if (eax == vmx_preemption_cpu_tfms[i])
533 			return true;
534 
535 	return false;
536 }
537 
538 static inline bool cpu_need_virtualize_apic_accesses(struct kvm_vcpu *vcpu)
539 {
540 	return flexpriority_enabled && lapic_in_kernel(vcpu);
541 }
542 
543 static inline bool report_flexpriority(void)
544 {
545 	return flexpriority_enabled;
546 }
547 
548 static inline int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
549 {
550 	int i;
551 
552 	for (i = 0; i < vmx->nmsrs; ++i)
553 		if (vmx_msr_index[vmx->guest_msrs[i].index] == msr)
554 			return i;
555 	return -1;
556 }
557 
558 struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
559 {
560 	int i;
561 
562 	i = __find_msr_index(vmx, msr);
563 	if (i >= 0)
564 		return &vmx->guest_msrs[i];
565 	return NULL;
566 }
567 
568 void loaded_vmcs_init(struct loaded_vmcs *loaded_vmcs)
569 {
570 	vmcs_clear(loaded_vmcs->vmcs);
571 	if (loaded_vmcs->shadow_vmcs && loaded_vmcs->launched)
572 		vmcs_clear(loaded_vmcs->shadow_vmcs);
573 	loaded_vmcs->cpu = -1;
574 	loaded_vmcs->launched = 0;
575 }
576 
577 #ifdef CONFIG_KEXEC_CORE
578 /*
579  * This bitmap is used to indicate whether the vmclear
580  * operation is enabled on all cpus. All disabled by
581  * default.
582  */
583 static cpumask_t crash_vmclear_enabled_bitmap = CPU_MASK_NONE;
584 
585 static inline void crash_enable_local_vmclear(int cpu)
586 {
587 	cpumask_set_cpu(cpu, &crash_vmclear_enabled_bitmap);
588 }
589 
590 static inline void crash_disable_local_vmclear(int cpu)
591 {
592 	cpumask_clear_cpu(cpu, &crash_vmclear_enabled_bitmap);
593 }
594 
595 static inline int crash_local_vmclear_enabled(int cpu)
596 {
597 	return cpumask_test_cpu(cpu, &crash_vmclear_enabled_bitmap);
598 }
599 
600 static void crash_vmclear_local_loaded_vmcss(void)
601 {
602 	int cpu = raw_smp_processor_id();
603 	struct loaded_vmcs *v;
604 
605 	if (!crash_local_vmclear_enabled(cpu))
606 		return;
607 
608 	list_for_each_entry(v, &per_cpu(loaded_vmcss_on_cpu, cpu),
609 			    loaded_vmcss_on_cpu_link)
610 		vmcs_clear(v->vmcs);
611 }
612 #else
613 static inline void crash_enable_local_vmclear(int cpu) { }
614 static inline void crash_disable_local_vmclear(int cpu) { }
615 #endif /* CONFIG_KEXEC_CORE */
616 
617 static void __loaded_vmcs_clear(void *arg)
618 {
619 	struct loaded_vmcs *loaded_vmcs = arg;
620 	int cpu = raw_smp_processor_id();
621 
622 	if (loaded_vmcs->cpu != cpu)
623 		return; /* vcpu migration can race with cpu offline */
624 	if (per_cpu(current_vmcs, cpu) == loaded_vmcs->vmcs)
625 		per_cpu(current_vmcs, cpu) = NULL;
626 	crash_disable_local_vmclear(cpu);
627 	list_del(&loaded_vmcs->loaded_vmcss_on_cpu_link);
628 
629 	/*
630 	 * we should ensure updating loaded_vmcs->loaded_vmcss_on_cpu_link
631 	 * is before setting loaded_vmcs->vcpu to -1 which is done in
632 	 * loaded_vmcs_init. Otherwise, other cpu can see vcpu = -1 fist
633 	 * then adds the vmcs into percpu list before it is deleted.
634 	 */
635 	smp_wmb();
636 
637 	loaded_vmcs_init(loaded_vmcs);
638 	crash_enable_local_vmclear(cpu);
639 }
640 
641 void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs)
642 {
643 	int cpu = loaded_vmcs->cpu;
644 
645 	if (cpu != -1)
646 		smp_call_function_single(cpu,
647 			 __loaded_vmcs_clear, loaded_vmcs, 1);
648 }
649 
650 static bool vmx_segment_cache_test_set(struct vcpu_vmx *vmx, unsigned seg,
651 				       unsigned field)
652 {
653 	bool ret;
654 	u32 mask = 1 << (seg * SEG_FIELD_NR + field);
655 
656 	if (!(vmx->vcpu.arch.regs_avail & (1 << VCPU_EXREG_SEGMENTS))) {
657 		vmx->vcpu.arch.regs_avail |= (1 << VCPU_EXREG_SEGMENTS);
658 		vmx->segment_cache.bitmask = 0;
659 	}
660 	ret = vmx->segment_cache.bitmask & mask;
661 	vmx->segment_cache.bitmask |= mask;
662 	return ret;
663 }
664 
665 static u16 vmx_read_guest_seg_selector(struct vcpu_vmx *vmx, unsigned seg)
666 {
667 	u16 *p = &vmx->segment_cache.seg[seg].selector;
668 
669 	if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_SEL))
670 		*p = vmcs_read16(kvm_vmx_segment_fields[seg].selector);
671 	return *p;
672 }
673 
674 static ulong vmx_read_guest_seg_base(struct vcpu_vmx *vmx, unsigned seg)
675 {
676 	ulong *p = &vmx->segment_cache.seg[seg].base;
677 
678 	if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_BASE))
679 		*p = vmcs_readl(kvm_vmx_segment_fields[seg].base);
680 	return *p;
681 }
682 
683 static u32 vmx_read_guest_seg_limit(struct vcpu_vmx *vmx, unsigned seg)
684 {
685 	u32 *p = &vmx->segment_cache.seg[seg].limit;
686 
687 	if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_LIMIT))
688 		*p = vmcs_read32(kvm_vmx_segment_fields[seg].limit);
689 	return *p;
690 }
691 
692 static u32 vmx_read_guest_seg_ar(struct vcpu_vmx *vmx, unsigned seg)
693 {
694 	u32 *p = &vmx->segment_cache.seg[seg].ar;
695 
696 	if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_AR))
697 		*p = vmcs_read32(kvm_vmx_segment_fields[seg].ar_bytes);
698 	return *p;
699 }
700 
701 void update_exception_bitmap(struct kvm_vcpu *vcpu)
702 {
703 	u32 eb;
704 
705 	eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR) |
706 	     (1u << DB_VECTOR) | (1u << AC_VECTOR);
707 	/*
708 	 * Guest access to VMware backdoor ports could legitimately
709 	 * trigger #GP because of TSS I/O permission bitmap.
710 	 * We intercept those #GP and allow access to them anyway
711 	 * as VMware does.
712 	 */
713 	if (enable_vmware_backdoor)
714 		eb |= (1u << GP_VECTOR);
715 	if ((vcpu->guest_debug &
716 	     (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) ==
717 	    (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP))
718 		eb |= 1u << BP_VECTOR;
719 	if (to_vmx(vcpu)->rmode.vm86_active)
720 		eb = ~0;
721 	if (enable_ept)
722 		eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */
723 
724 	/* When we are running a nested L2 guest and L1 specified for it a
725 	 * certain exception bitmap, we must trap the same exceptions and pass
726 	 * them to L1. When running L2, we will only handle the exceptions
727 	 * specified above if L1 did not want them.
728 	 */
729 	if (is_guest_mode(vcpu))
730 		eb |= get_vmcs12(vcpu)->exception_bitmap;
731 
732 	vmcs_write32(EXCEPTION_BITMAP, eb);
733 }
734 
735 /*
736  * Check if MSR is intercepted for currently loaded MSR bitmap.
737  */
738 static bool msr_write_intercepted(struct kvm_vcpu *vcpu, u32 msr)
739 {
740 	unsigned long *msr_bitmap;
741 	int f = sizeof(unsigned long);
742 
743 	if (!cpu_has_vmx_msr_bitmap())
744 		return true;
745 
746 	msr_bitmap = to_vmx(vcpu)->loaded_vmcs->msr_bitmap;
747 
748 	if (msr <= 0x1fff) {
749 		return !!test_bit(msr, msr_bitmap + 0x800 / f);
750 	} else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
751 		msr &= 0x1fff;
752 		return !!test_bit(msr, msr_bitmap + 0xc00 / f);
753 	}
754 
755 	return true;
756 }
757 
758 static void clear_atomic_switch_msr_special(struct vcpu_vmx *vmx,
759 		unsigned long entry, unsigned long exit)
760 {
761 	vm_entry_controls_clearbit(vmx, entry);
762 	vm_exit_controls_clearbit(vmx, exit);
763 }
764 
765 static int find_msr(struct vmx_msrs *m, unsigned int msr)
766 {
767 	unsigned int i;
768 
769 	for (i = 0; i < m->nr; ++i) {
770 		if (m->val[i].index == msr)
771 			return i;
772 	}
773 	return -ENOENT;
774 }
775 
776 static void clear_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr)
777 {
778 	int i;
779 	struct msr_autoload *m = &vmx->msr_autoload;
780 
781 	switch (msr) {
782 	case MSR_EFER:
783 		if (cpu_has_load_ia32_efer()) {
784 			clear_atomic_switch_msr_special(vmx,
785 					VM_ENTRY_LOAD_IA32_EFER,
786 					VM_EXIT_LOAD_IA32_EFER);
787 			return;
788 		}
789 		break;
790 	case MSR_CORE_PERF_GLOBAL_CTRL:
791 		if (cpu_has_load_perf_global_ctrl()) {
792 			clear_atomic_switch_msr_special(vmx,
793 					VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
794 					VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
795 			return;
796 		}
797 		break;
798 	}
799 	i = find_msr(&m->guest, msr);
800 	if (i < 0)
801 		goto skip_guest;
802 	--m->guest.nr;
803 	m->guest.val[i] = m->guest.val[m->guest.nr];
804 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->guest.nr);
805 
806 skip_guest:
807 	i = find_msr(&m->host, msr);
808 	if (i < 0)
809 		return;
810 
811 	--m->host.nr;
812 	m->host.val[i] = m->host.val[m->host.nr];
813 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->host.nr);
814 }
815 
816 static void add_atomic_switch_msr_special(struct vcpu_vmx *vmx,
817 		unsigned long entry, unsigned long exit,
818 		unsigned long guest_val_vmcs, unsigned long host_val_vmcs,
819 		u64 guest_val, u64 host_val)
820 {
821 	vmcs_write64(guest_val_vmcs, guest_val);
822 	if (host_val_vmcs != HOST_IA32_EFER)
823 		vmcs_write64(host_val_vmcs, host_val);
824 	vm_entry_controls_setbit(vmx, entry);
825 	vm_exit_controls_setbit(vmx, exit);
826 }
827 
828 static void add_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr,
829 				  u64 guest_val, u64 host_val, bool entry_only)
830 {
831 	int i, j = 0;
832 	struct msr_autoload *m = &vmx->msr_autoload;
833 
834 	switch (msr) {
835 	case MSR_EFER:
836 		if (cpu_has_load_ia32_efer()) {
837 			add_atomic_switch_msr_special(vmx,
838 					VM_ENTRY_LOAD_IA32_EFER,
839 					VM_EXIT_LOAD_IA32_EFER,
840 					GUEST_IA32_EFER,
841 					HOST_IA32_EFER,
842 					guest_val, host_val);
843 			return;
844 		}
845 		break;
846 	case MSR_CORE_PERF_GLOBAL_CTRL:
847 		if (cpu_has_load_perf_global_ctrl()) {
848 			add_atomic_switch_msr_special(vmx,
849 					VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
850 					VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL,
851 					GUEST_IA32_PERF_GLOBAL_CTRL,
852 					HOST_IA32_PERF_GLOBAL_CTRL,
853 					guest_val, host_val);
854 			return;
855 		}
856 		break;
857 	case MSR_IA32_PEBS_ENABLE:
858 		/* PEBS needs a quiescent period after being disabled (to write
859 		 * a record).  Disabling PEBS through VMX MSR swapping doesn't
860 		 * provide that period, so a CPU could write host's record into
861 		 * guest's memory.
862 		 */
863 		wrmsrl(MSR_IA32_PEBS_ENABLE, 0);
864 	}
865 
866 	i = find_msr(&m->guest, msr);
867 	if (!entry_only)
868 		j = find_msr(&m->host, msr);
869 
870 	if ((i < 0 && m->guest.nr == NR_AUTOLOAD_MSRS) ||
871 		(j < 0 &&  m->host.nr == NR_AUTOLOAD_MSRS)) {
872 		printk_once(KERN_WARNING "Not enough msr switch entries. "
873 				"Can't add msr %x\n", msr);
874 		return;
875 	}
876 	if (i < 0) {
877 		i = m->guest.nr++;
878 		vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->guest.nr);
879 	}
880 	m->guest.val[i].index = msr;
881 	m->guest.val[i].value = guest_val;
882 
883 	if (entry_only)
884 		return;
885 
886 	if (j < 0) {
887 		j = m->host.nr++;
888 		vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->host.nr);
889 	}
890 	m->host.val[j].index = msr;
891 	m->host.val[j].value = host_val;
892 }
893 
894 static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset)
895 {
896 	u64 guest_efer = vmx->vcpu.arch.efer;
897 	u64 ignore_bits = 0;
898 
899 	if (!enable_ept) {
900 		/*
901 		 * NX is needed to handle CR0.WP=1, CR4.SMEP=1.  Testing
902 		 * host CPUID is more efficient than testing guest CPUID
903 		 * or CR4.  Host SMEP is anyway a requirement for guest SMEP.
904 		 */
905 		if (boot_cpu_has(X86_FEATURE_SMEP))
906 			guest_efer |= EFER_NX;
907 		else if (!(guest_efer & EFER_NX))
908 			ignore_bits |= EFER_NX;
909 	}
910 
911 	/*
912 	 * LMA and LME handled by hardware; SCE meaningless outside long mode.
913 	 */
914 	ignore_bits |= EFER_SCE;
915 #ifdef CONFIG_X86_64
916 	ignore_bits |= EFER_LMA | EFER_LME;
917 	/* SCE is meaningful only in long mode on Intel */
918 	if (guest_efer & EFER_LMA)
919 		ignore_bits &= ~(u64)EFER_SCE;
920 #endif
921 
922 	/*
923 	 * On EPT, we can't emulate NX, so we must switch EFER atomically.
924 	 * On CPUs that support "load IA32_EFER", always switch EFER
925 	 * atomically, since it's faster than switching it manually.
926 	 */
927 	if (cpu_has_load_ia32_efer() ||
928 	    (enable_ept && ((vmx->vcpu.arch.efer ^ host_efer) & EFER_NX))) {
929 		if (!(guest_efer & EFER_LMA))
930 			guest_efer &= ~EFER_LME;
931 		if (guest_efer != host_efer)
932 			add_atomic_switch_msr(vmx, MSR_EFER,
933 					      guest_efer, host_efer, false);
934 		else
935 			clear_atomic_switch_msr(vmx, MSR_EFER);
936 		return false;
937 	} else {
938 		clear_atomic_switch_msr(vmx, MSR_EFER);
939 
940 		guest_efer &= ~ignore_bits;
941 		guest_efer |= host_efer & ignore_bits;
942 
943 		vmx->guest_msrs[efer_offset].data = guest_efer;
944 		vmx->guest_msrs[efer_offset].mask = ~ignore_bits;
945 
946 		return true;
947 	}
948 }
949 
950 #ifdef CONFIG_X86_32
951 /*
952  * On 32-bit kernels, VM exits still load the FS and GS bases from the
953  * VMCS rather than the segment table.  KVM uses this helper to figure
954  * out the current bases to poke them into the VMCS before entry.
955  */
956 static unsigned long segment_base(u16 selector)
957 {
958 	struct desc_struct *table;
959 	unsigned long v;
960 
961 	if (!(selector & ~SEGMENT_RPL_MASK))
962 		return 0;
963 
964 	table = get_current_gdt_ro();
965 
966 	if ((selector & SEGMENT_TI_MASK) == SEGMENT_LDT) {
967 		u16 ldt_selector = kvm_read_ldt();
968 
969 		if (!(ldt_selector & ~SEGMENT_RPL_MASK))
970 			return 0;
971 
972 		table = (struct desc_struct *)segment_base(ldt_selector);
973 	}
974 	v = get_desc_base(&table[selector >> 3]);
975 	return v;
976 }
977 #endif
978 
979 static inline void pt_load_msr(struct pt_ctx *ctx, u32 addr_range)
980 {
981 	u32 i;
982 
983 	wrmsrl(MSR_IA32_RTIT_STATUS, ctx->status);
984 	wrmsrl(MSR_IA32_RTIT_OUTPUT_BASE, ctx->output_base);
985 	wrmsrl(MSR_IA32_RTIT_OUTPUT_MASK, ctx->output_mask);
986 	wrmsrl(MSR_IA32_RTIT_CR3_MATCH, ctx->cr3_match);
987 	for (i = 0; i < addr_range; i++) {
988 		wrmsrl(MSR_IA32_RTIT_ADDR0_A + i * 2, ctx->addr_a[i]);
989 		wrmsrl(MSR_IA32_RTIT_ADDR0_B + i * 2, ctx->addr_b[i]);
990 	}
991 }
992 
993 static inline void pt_save_msr(struct pt_ctx *ctx, u32 addr_range)
994 {
995 	u32 i;
996 
997 	rdmsrl(MSR_IA32_RTIT_STATUS, ctx->status);
998 	rdmsrl(MSR_IA32_RTIT_OUTPUT_BASE, ctx->output_base);
999 	rdmsrl(MSR_IA32_RTIT_OUTPUT_MASK, ctx->output_mask);
1000 	rdmsrl(MSR_IA32_RTIT_CR3_MATCH, ctx->cr3_match);
1001 	for (i = 0; i < addr_range; i++) {
1002 		rdmsrl(MSR_IA32_RTIT_ADDR0_A + i * 2, ctx->addr_a[i]);
1003 		rdmsrl(MSR_IA32_RTIT_ADDR0_B + i * 2, ctx->addr_b[i]);
1004 	}
1005 }
1006 
1007 static void pt_guest_enter(struct vcpu_vmx *vmx)
1008 {
1009 	if (pt_mode == PT_MODE_SYSTEM)
1010 		return;
1011 
1012 	/*
1013 	 * GUEST_IA32_RTIT_CTL is already set in the VMCS.
1014 	 * Save host state before VM entry.
1015 	 */
1016 	rdmsrl(MSR_IA32_RTIT_CTL, vmx->pt_desc.host.ctl);
1017 	if (vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) {
1018 		wrmsrl(MSR_IA32_RTIT_CTL, 0);
1019 		pt_save_msr(&vmx->pt_desc.host, vmx->pt_desc.addr_range);
1020 		pt_load_msr(&vmx->pt_desc.guest, vmx->pt_desc.addr_range);
1021 	}
1022 }
1023 
1024 static void pt_guest_exit(struct vcpu_vmx *vmx)
1025 {
1026 	if (pt_mode == PT_MODE_SYSTEM)
1027 		return;
1028 
1029 	if (vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) {
1030 		pt_save_msr(&vmx->pt_desc.guest, vmx->pt_desc.addr_range);
1031 		pt_load_msr(&vmx->pt_desc.host, vmx->pt_desc.addr_range);
1032 	}
1033 
1034 	/* Reload host state (IA32_RTIT_CTL will be cleared on VM exit). */
1035 	wrmsrl(MSR_IA32_RTIT_CTL, vmx->pt_desc.host.ctl);
1036 }
1037 
1038 void vmx_prepare_switch_to_guest(struct kvm_vcpu *vcpu)
1039 {
1040 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1041 	struct vmcs_host_state *host_state;
1042 #ifdef CONFIG_X86_64
1043 	int cpu = raw_smp_processor_id();
1044 #endif
1045 	unsigned long fs_base, gs_base;
1046 	u16 fs_sel, gs_sel;
1047 	int i;
1048 
1049 	vmx->req_immediate_exit = false;
1050 
1051 	/*
1052 	 * Note that guest MSRs to be saved/restored can also be changed
1053 	 * when guest state is loaded. This happens when guest transitions
1054 	 * to/from long-mode by setting MSR_EFER.LMA.
1055 	 */
1056 	if (!vmx->loaded_cpu_state || vmx->guest_msrs_dirty) {
1057 		vmx->guest_msrs_dirty = false;
1058 		for (i = 0; i < vmx->save_nmsrs; ++i)
1059 			kvm_set_shared_msr(vmx->guest_msrs[i].index,
1060 					   vmx->guest_msrs[i].data,
1061 					   vmx->guest_msrs[i].mask);
1062 
1063 	}
1064 
1065 	if (vmx->loaded_cpu_state)
1066 		return;
1067 
1068 	vmx->loaded_cpu_state = vmx->loaded_vmcs;
1069 	host_state = &vmx->loaded_cpu_state->host_state;
1070 
1071 	/*
1072 	 * Set host fs and gs selectors.  Unfortunately, 22.2.3 does not
1073 	 * allow segment selectors with cpl > 0 or ti == 1.
1074 	 */
1075 	host_state->ldt_sel = kvm_read_ldt();
1076 
1077 #ifdef CONFIG_X86_64
1078 	savesegment(ds, host_state->ds_sel);
1079 	savesegment(es, host_state->es_sel);
1080 
1081 	gs_base = cpu_kernelmode_gs_base(cpu);
1082 	if (likely(is_64bit_mm(current->mm))) {
1083 		save_fsgs_for_kvm();
1084 		fs_sel = current->thread.fsindex;
1085 		gs_sel = current->thread.gsindex;
1086 		fs_base = current->thread.fsbase;
1087 		vmx->msr_host_kernel_gs_base = current->thread.gsbase;
1088 	} else {
1089 		savesegment(fs, fs_sel);
1090 		savesegment(gs, gs_sel);
1091 		fs_base = read_msr(MSR_FS_BASE);
1092 		vmx->msr_host_kernel_gs_base = read_msr(MSR_KERNEL_GS_BASE);
1093 	}
1094 
1095 	wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1096 #else
1097 	savesegment(fs, fs_sel);
1098 	savesegment(gs, gs_sel);
1099 	fs_base = segment_base(fs_sel);
1100 	gs_base = segment_base(gs_sel);
1101 #endif
1102 
1103 	if (unlikely(fs_sel != host_state->fs_sel)) {
1104 		if (!(fs_sel & 7))
1105 			vmcs_write16(HOST_FS_SELECTOR, fs_sel);
1106 		else
1107 			vmcs_write16(HOST_FS_SELECTOR, 0);
1108 		host_state->fs_sel = fs_sel;
1109 	}
1110 	if (unlikely(gs_sel != host_state->gs_sel)) {
1111 		if (!(gs_sel & 7))
1112 			vmcs_write16(HOST_GS_SELECTOR, gs_sel);
1113 		else
1114 			vmcs_write16(HOST_GS_SELECTOR, 0);
1115 		host_state->gs_sel = gs_sel;
1116 	}
1117 	if (unlikely(fs_base != host_state->fs_base)) {
1118 		vmcs_writel(HOST_FS_BASE, fs_base);
1119 		host_state->fs_base = fs_base;
1120 	}
1121 	if (unlikely(gs_base != host_state->gs_base)) {
1122 		vmcs_writel(HOST_GS_BASE, gs_base);
1123 		host_state->gs_base = gs_base;
1124 	}
1125 }
1126 
1127 static void vmx_prepare_switch_to_host(struct vcpu_vmx *vmx)
1128 {
1129 	struct vmcs_host_state *host_state;
1130 
1131 	if (!vmx->loaded_cpu_state)
1132 		return;
1133 
1134 	WARN_ON_ONCE(vmx->loaded_cpu_state != vmx->loaded_vmcs);
1135 	host_state = &vmx->loaded_cpu_state->host_state;
1136 
1137 	++vmx->vcpu.stat.host_state_reload;
1138 	vmx->loaded_cpu_state = NULL;
1139 
1140 #ifdef CONFIG_X86_64
1141 	rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1142 #endif
1143 	if (host_state->ldt_sel || (host_state->gs_sel & 7)) {
1144 		kvm_load_ldt(host_state->ldt_sel);
1145 #ifdef CONFIG_X86_64
1146 		load_gs_index(host_state->gs_sel);
1147 #else
1148 		loadsegment(gs, host_state->gs_sel);
1149 #endif
1150 	}
1151 	if (host_state->fs_sel & 7)
1152 		loadsegment(fs, host_state->fs_sel);
1153 #ifdef CONFIG_X86_64
1154 	if (unlikely(host_state->ds_sel | host_state->es_sel)) {
1155 		loadsegment(ds, host_state->ds_sel);
1156 		loadsegment(es, host_state->es_sel);
1157 	}
1158 #endif
1159 	invalidate_tss_limit();
1160 #ifdef CONFIG_X86_64
1161 	wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
1162 #endif
1163 	load_fixmap_gdt(raw_smp_processor_id());
1164 }
1165 
1166 #ifdef CONFIG_X86_64
1167 static u64 vmx_read_guest_kernel_gs_base(struct vcpu_vmx *vmx)
1168 {
1169 	preempt_disable();
1170 	if (vmx->loaded_cpu_state)
1171 		rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1172 	preempt_enable();
1173 	return vmx->msr_guest_kernel_gs_base;
1174 }
1175 
1176 static void vmx_write_guest_kernel_gs_base(struct vcpu_vmx *vmx, u64 data)
1177 {
1178 	preempt_disable();
1179 	if (vmx->loaded_cpu_state)
1180 		wrmsrl(MSR_KERNEL_GS_BASE, data);
1181 	preempt_enable();
1182 	vmx->msr_guest_kernel_gs_base = data;
1183 }
1184 #endif
1185 
1186 static void vmx_vcpu_pi_load(struct kvm_vcpu *vcpu, int cpu)
1187 {
1188 	struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
1189 	struct pi_desc old, new;
1190 	unsigned int dest;
1191 
1192 	/*
1193 	 * In case of hot-plug or hot-unplug, we may have to undo
1194 	 * vmx_vcpu_pi_put even if there is no assigned device.  And we
1195 	 * always keep PI.NDST up to date for simplicity: it makes the
1196 	 * code easier, and CPU migration is not a fast path.
1197 	 */
1198 	if (!pi_test_sn(pi_desc) && vcpu->cpu == cpu)
1199 		return;
1200 
1201 	/* The full case.  */
1202 	do {
1203 		old.control = new.control = pi_desc->control;
1204 
1205 		dest = cpu_physical_id(cpu);
1206 
1207 		if (x2apic_enabled())
1208 			new.ndst = dest;
1209 		else
1210 			new.ndst = (dest << 8) & 0xFF00;
1211 
1212 		new.sn = 0;
1213 	} while (cmpxchg64(&pi_desc->control, old.control,
1214 			   new.control) != old.control);
1215 
1216 	/*
1217 	 * Clear SN before reading the bitmap.  The VT-d firmware
1218 	 * writes the bitmap and reads SN atomically (5.2.3 in the
1219 	 * spec), so it doesn't really have a memory barrier that
1220 	 * pairs with this, but we cannot do that and we need one.
1221 	 */
1222 	smp_mb__after_atomic();
1223 
1224 	if (!bitmap_empty((unsigned long *)pi_desc->pir, NR_VECTORS))
1225 		pi_set_on(pi_desc);
1226 }
1227 
1228 /*
1229  * Switches to specified vcpu, until a matching vcpu_put(), but assumes
1230  * vcpu mutex is already taken.
1231  */
1232 void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1233 {
1234 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1235 	bool already_loaded = vmx->loaded_vmcs->cpu == cpu;
1236 
1237 	if (!already_loaded) {
1238 		loaded_vmcs_clear(vmx->loaded_vmcs);
1239 		local_irq_disable();
1240 		crash_disable_local_vmclear(cpu);
1241 
1242 		/*
1243 		 * Read loaded_vmcs->cpu should be before fetching
1244 		 * loaded_vmcs->loaded_vmcss_on_cpu_link.
1245 		 * See the comments in __loaded_vmcs_clear().
1246 		 */
1247 		smp_rmb();
1248 
1249 		list_add(&vmx->loaded_vmcs->loaded_vmcss_on_cpu_link,
1250 			 &per_cpu(loaded_vmcss_on_cpu, cpu));
1251 		crash_enable_local_vmclear(cpu);
1252 		local_irq_enable();
1253 	}
1254 
1255 	if (per_cpu(current_vmcs, cpu) != vmx->loaded_vmcs->vmcs) {
1256 		per_cpu(current_vmcs, cpu) = vmx->loaded_vmcs->vmcs;
1257 		vmcs_load(vmx->loaded_vmcs->vmcs);
1258 		indirect_branch_prediction_barrier();
1259 	}
1260 
1261 	if (!already_loaded) {
1262 		void *gdt = get_current_gdt_ro();
1263 		unsigned long sysenter_esp;
1264 
1265 		kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1266 
1267 		/*
1268 		 * Linux uses per-cpu TSS and GDT, so set these when switching
1269 		 * processors.  See 22.2.4.
1270 		 */
1271 		vmcs_writel(HOST_TR_BASE,
1272 			    (unsigned long)&get_cpu_entry_area(cpu)->tss.x86_tss);
1273 		vmcs_writel(HOST_GDTR_BASE, (unsigned long)gdt);   /* 22.2.4 */
1274 
1275 		/*
1276 		 * VM exits change the host TR limit to 0x67 after a VM
1277 		 * exit.  This is okay, since 0x67 covers everything except
1278 		 * the IO bitmap and have have code to handle the IO bitmap
1279 		 * being lost after a VM exit.
1280 		 */
1281 		BUILD_BUG_ON(IO_BITMAP_OFFSET - 1 != 0x67);
1282 
1283 		rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
1284 		vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
1285 
1286 		vmx->loaded_vmcs->cpu = cpu;
1287 	}
1288 
1289 	/* Setup TSC multiplier */
1290 	if (kvm_has_tsc_control &&
1291 	    vmx->current_tsc_ratio != vcpu->arch.tsc_scaling_ratio)
1292 		decache_tsc_multiplier(vmx);
1293 
1294 	vmx_vcpu_pi_load(vcpu, cpu);
1295 	vmx->host_pkru = read_pkru();
1296 	vmx->host_debugctlmsr = get_debugctlmsr();
1297 }
1298 
1299 static void vmx_vcpu_pi_put(struct kvm_vcpu *vcpu)
1300 {
1301 	struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
1302 
1303 	if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
1304 		!irq_remapping_cap(IRQ_POSTING_CAP)  ||
1305 		!kvm_vcpu_apicv_active(vcpu))
1306 		return;
1307 
1308 	/* Set SN when the vCPU is preempted */
1309 	if (vcpu->preempted)
1310 		pi_set_sn(pi_desc);
1311 }
1312 
1313 void vmx_vcpu_put(struct kvm_vcpu *vcpu)
1314 {
1315 	vmx_vcpu_pi_put(vcpu);
1316 
1317 	vmx_prepare_switch_to_host(to_vmx(vcpu));
1318 }
1319 
1320 static bool emulation_required(struct kvm_vcpu *vcpu)
1321 {
1322 	return emulate_invalid_guest_state && !guest_state_valid(vcpu);
1323 }
1324 
1325 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu);
1326 
1327 unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
1328 {
1329 	unsigned long rflags, save_rflags;
1330 
1331 	if (!test_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail)) {
1332 		__set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
1333 		rflags = vmcs_readl(GUEST_RFLAGS);
1334 		if (to_vmx(vcpu)->rmode.vm86_active) {
1335 			rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
1336 			save_rflags = to_vmx(vcpu)->rmode.save_rflags;
1337 			rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
1338 		}
1339 		to_vmx(vcpu)->rflags = rflags;
1340 	}
1341 	return to_vmx(vcpu)->rflags;
1342 }
1343 
1344 void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
1345 {
1346 	unsigned long old_rflags = vmx_get_rflags(vcpu);
1347 
1348 	__set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
1349 	to_vmx(vcpu)->rflags = rflags;
1350 	if (to_vmx(vcpu)->rmode.vm86_active) {
1351 		to_vmx(vcpu)->rmode.save_rflags = rflags;
1352 		rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
1353 	}
1354 	vmcs_writel(GUEST_RFLAGS, rflags);
1355 
1356 	if ((old_rflags ^ to_vmx(vcpu)->rflags) & X86_EFLAGS_VM)
1357 		to_vmx(vcpu)->emulation_required = emulation_required(vcpu);
1358 }
1359 
1360 u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu)
1361 {
1362 	u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1363 	int ret = 0;
1364 
1365 	if (interruptibility & GUEST_INTR_STATE_STI)
1366 		ret |= KVM_X86_SHADOW_INT_STI;
1367 	if (interruptibility & GUEST_INTR_STATE_MOV_SS)
1368 		ret |= KVM_X86_SHADOW_INT_MOV_SS;
1369 
1370 	return ret;
1371 }
1372 
1373 void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
1374 {
1375 	u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1376 	u32 interruptibility = interruptibility_old;
1377 
1378 	interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
1379 
1380 	if (mask & KVM_X86_SHADOW_INT_MOV_SS)
1381 		interruptibility |= GUEST_INTR_STATE_MOV_SS;
1382 	else if (mask & KVM_X86_SHADOW_INT_STI)
1383 		interruptibility |= GUEST_INTR_STATE_STI;
1384 
1385 	if ((interruptibility != interruptibility_old))
1386 		vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
1387 }
1388 
1389 static int vmx_rtit_ctl_check(struct kvm_vcpu *vcpu, u64 data)
1390 {
1391 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1392 	unsigned long value;
1393 
1394 	/*
1395 	 * Any MSR write that attempts to change bits marked reserved will
1396 	 * case a #GP fault.
1397 	 */
1398 	if (data & vmx->pt_desc.ctl_bitmask)
1399 		return 1;
1400 
1401 	/*
1402 	 * Any attempt to modify IA32_RTIT_CTL while TraceEn is set will
1403 	 * result in a #GP unless the same write also clears TraceEn.
1404 	 */
1405 	if ((vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) &&
1406 		((vmx->pt_desc.guest.ctl ^ data) & ~RTIT_CTL_TRACEEN))
1407 		return 1;
1408 
1409 	/*
1410 	 * WRMSR to IA32_RTIT_CTL that sets TraceEn but clears this bit
1411 	 * and FabricEn would cause #GP, if
1412 	 * CPUID.(EAX=14H, ECX=0):ECX.SNGLRGNOUT[bit 2] = 0
1413 	 */
1414 	if ((data & RTIT_CTL_TRACEEN) && !(data & RTIT_CTL_TOPA) &&
1415 		!(data & RTIT_CTL_FABRIC_EN) &&
1416 		!intel_pt_validate_cap(vmx->pt_desc.caps,
1417 					PT_CAP_single_range_output))
1418 		return 1;
1419 
1420 	/*
1421 	 * MTCFreq, CycThresh and PSBFreq encodings check, any MSR write that
1422 	 * utilize encodings marked reserved will casue a #GP fault.
1423 	 */
1424 	value = intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc_periods);
1425 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc) &&
1426 			!test_bit((data & RTIT_CTL_MTC_RANGE) >>
1427 			RTIT_CTL_MTC_RANGE_OFFSET, &value))
1428 		return 1;
1429 	value = intel_pt_validate_cap(vmx->pt_desc.caps,
1430 						PT_CAP_cycle_thresholds);
1431 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc) &&
1432 			!test_bit((data & RTIT_CTL_CYC_THRESH) >>
1433 			RTIT_CTL_CYC_THRESH_OFFSET, &value))
1434 		return 1;
1435 	value = intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_periods);
1436 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc) &&
1437 			!test_bit((data & RTIT_CTL_PSB_FREQ) >>
1438 			RTIT_CTL_PSB_FREQ_OFFSET, &value))
1439 		return 1;
1440 
1441 	/*
1442 	 * If ADDRx_CFG is reserved or the encodings is >2 will
1443 	 * cause a #GP fault.
1444 	 */
1445 	value = (data & RTIT_CTL_ADDR0) >> RTIT_CTL_ADDR0_OFFSET;
1446 	if ((value && (vmx->pt_desc.addr_range < 1)) || (value > 2))
1447 		return 1;
1448 	value = (data & RTIT_CTL_ADDR1) >> RTIT_CTL_ADDR1_OFFSET;
1449 	if ((value && (vmx->pt_desc.addr_range < 2)) || (value > 2))
1450 		return 1;
1451 	value = (data & RTIT_CTL_ADDR2) >> RTIT_CTL_ADDR2_OFFSET;
1452 	if ((value && (vmx->pt_desc.addr_range < 3)) || (value > 2))
1453 		return 1;
1454 	value = (data & RTIT_CTL_ADDR3) >> RTIT_CTL_ADDR3_OFFSET;
1455 	if ((value && (vmx->pt_desc.addr_range < 4)) || (value > 2))
1456 		return 1;
1457 
1458 	return 0;
1459 }
1460 
1461 
1462 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
1463 {
1464 	unsigned long rip;
1465 
1466 	rip = kvm_rip_read(vcpu);
1467 	rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
1468 	kvm_rip_write(vcpu, rip);
1469 
1470 	/* skipping an emulated instruction also counts */
1471 	vmx_set_interrupt_shadow(vcpu, 0);
1472 }
1473 
1474 static void vmx_clear_hlt(struct kvm_vcpu *vcpu)
1475 {
1476 	/*
1477 	 * Ensure that we clear the HLT state in the VMCS.  We don't need to
1478 	 * explicitly skip the instruction because if the HLT state is set,
1479 	 * then the instruction is already executing and RIP has already been
1480 	 * advanced.
1481 	 */
1482 	if (kvm_hlt_in_guest(vcpu->kvm) &&
1483 			vmcs_read32(GUEST_ACTIVITY_STATE) == GUEST_ACTIVITY_HLT)
1484 		vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
1485 }
1486 
1487 static void vmx_queue_exception(struct kvm_vcpu *vcpu)
1488 {
1489 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1490 	unsigned nr = vcpu->arch.exception.nr;
1491 	bool has_error_code = vcpu->arch.exception.has_error_code;
1492 	u32 error_code = vcpu->arch.exception.error_code;
1493 	u32 intr_info = nr | INTR_INFO_VALID_MASK;
1494 
1495 	kvm_deliver_exception_payload(vcpu);
1496 
1497 	if (has_error_code) {
1498 		vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
1499 		intr_info |= INTR_INFO_DELIVER_CODE_MASK;
1500 	}
1501 
1502 	if (vmx->rmode.vm86_active) {
1503 		int inc_eip = 0;
1504 		if (kvm_exception_is_soft(nr))
1505 			inc_eip = vcpu->arch.event_exit_inst_len;
1506 		if (kvm_inject_realmode_interrupt(vcpu, nr, inc_eip) != EMULATE_DONE)
1507 			kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
1508 		return;
1509 	}
1510 
1511 	WARN_ON_ONCE(vmx->emulation_required);
1512 
1513 	if (kvm_exception_is_soft(nr)) {
1514 		vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
1515 			     vmx->vcpu.arch.event_exit_inst_len);
1516 		intr_info |= INTR_TYPE_SOFT_EXCEPTION;
1517 	} else
1518 		intr_info |= INTR_TYPE_HARD_EXCEPTION;
1519 
1520 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
1521 
1522 	vmx_clear_hlt(vcpu);
1523 }
1524 
1525 static bool vmx_rdtscp_supported(void)
1526 {
1527 	return cpu_has_vmx_rdtscp();
1528 }
1529 
1530 static bool vmx_invpcid_supported(void)
1531 {
1532 	return cpu_has_vmx_invpcid();
1533 }
1534 
1535 /*
1536  * Swap MSR entry in host/guest MSR entry array.
1537  */
1538 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
1539 {
1540 	struct shared_msr_entry tmp;
1541 
1542 	tmp = vmx->guest_msrs[to];
1543 	vmx->guest_msrs[to] = vmx->guest_msrs[from];
1544 	vmx->guest_msrs[from] = tmp;
1545 }
1546 
1547 /*
1548  * Set up the vmcs to automatically save and restore system
1549  * msrs.  Don't touch the 64-bit msrs if the guest is in legacy
1550  * mode, as fiddling with msrs is very expensive.
1551  */
1552 static void setup_msrs(struct vcpu_vmx *vmx)
1553 {
1554 	int save_nmsrs, index;
1555 
1556 	save_nmsrs = 0;
1557 #ifdef CONFIG_X86_64
1558 	/*
1559 	 * The SYSCALL MSRs are only needed on long mode guests, and only
1560 	 * when EFER.SCE is set.
1561 	 */
1562 	if (is_long_mode(&vmx->vcpu) && (vmx->vcpu.arch.efer & EFER_SCE)) {
1563 		index = __find_msr_index(vmx, MSR_STAR);
1564 		if (index >= 0)
1565 			move_msr_up(vmx, index, save_nmsrs++);
1566 		index = __find_msr_index(vmx, MSR_LSTAR);
1567 		if (index >= 0)
1568 			move_msr_up(vmx, index, save_nmsrs++);
1569 		index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
1570 		if (index >= 0)
1571 			move_msr_up(vmx, index, save_nmsrs++);
1572 	}
1573 #endif
1574 	index = __find_msr_index(vmx, MSR_EFER);
1575 	if (index >= 0 && update_transition_efer(vmx, index))
1576 		move_msr_up(vmx, index, save_nmsrs++);
1577 	index = __find_msr_index(vmx, MSR_TSC_AUX);
1578 	if (index >= 0 && guest_cpuid_has(&vmx->vcpu, X86_FEATURE_RDTSCP))
1579 		move_msr_up(vmx, index, save_nmsrs++);
1580 
1581 	vmx->save_nmsrs = save_nmsrs;
1582 	vmx->guest_msrs_dirty = true;
1583 
1584 	if (cpu_has_vmx_msr_bitmap())
1585 		vmx_update_msr_bitmap(&vmx->vcpu);
1586 }
1587 
1588 static u64 vmx_read_l1_tsc_offset(struct kvm_vcpu *vcpu)
1589 {
1590 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1591 
1592 	if (is_guest_mode(vcpu) &&
1593 	    (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING))
1594 		return vcpu->arch.tsc_offset - vmcs12->tsc_offset;
1595 
1596 	return vcpu->arch.tsc_offset;
1597 }
1598 
1599 static u64 vmx_write_l1_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1600 {
1601 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1602 	u64 g_tsc_offset = 0;
1603 
1604 	/*
1605 	 * We're here if L1 chose not to trap WRMSR to TSC. According
1606 	 * to the spec, this should set L1's TSC; The offset that L1
1607 	 * set for L2 remains unchanged, and still needs to be added
1608 	 * to the newly set TSC to get L2's TSC.
1609 	 */
1610 	if (is_guest_mode(vcpu) &&
1611 	    (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING))
1612 		g_tsc_offset = vmcs12->tsc_offset;
1613 
1614 	trace_kvm_write_tsc_offset(vcpu->vcpu_id,
1615 				   vcpu->arch.tsc_offset - g_tsc_offset,
1616 				   offset);
1617 	vmcs_write64(TSC_OFFSET, offset + g_tsc_offset);
1618 	return offset + g_tsc_offset;
1619 }
1620 
1621 /*
1622  * nested_vmx_allowed() checks whether a guest should be allowed to use VMX
1623  * instructions and MSRs (i.e., nested VMX). Nested VMX is disabled for
1624  * all guests if the "nested" module option is off, and can also be disabled
1625  * for a single guest by disabling its VMX cpuid bit.
1626  */
1627 bool nested_vmx_allowed(struct kvm_vcpu *vcpu)
1628 {
1629 	return nested && guest_cpuid_has(vcpu, X86_FEATURE_VMX);
1630 }
1631 
1632 static inline bool vmx_feature_control_msr_valid(struct kvm_vcpu *vcpu,
1633 						 uint64_t val)
1634 {
1635 	uint64_t valid_bits = to_vmx(vcpu)->msr_ia32_feature_control_valid_bits;
1636 
1637 	return !(val & ~valid_bits);
1638 }
1639 
1640 static int vmx_get_msr_feature(struct kvm_msr_entry *msr)
1641 {
1642 	switch (msr->index) {
1643 	case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
1644 		if (!nested)
1645 			return 1;
1646 		return vmx_get_vmx_msr(&vmcs_config.nested, msr->index, &msr->data);
1647 	default:
1648 		return 1;
1649 	}
1650 
1651 	return 0;
1652 }
1653 
1654 /*
1655  * Reads an msr value (of 'msr_index') into 'pdata'.
1656  * Returns 0 on success, non-0 otherwise.
1657  * Assumes vcpu_load() was already called.
1658  */
1659 static int vmx_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1660 {
1661 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1662 	struct shared_msr_entry *msr;
1663 	u32 index;
1664 
1665 	switch (msr_info->index) {
1666 #ifdef CONFIG_X86_64
1667 	case MSR_FS_BASE:
1668 		msr_info->data = vmcs_readl(GUEST_FS_BASE);
1669 		break;
1670 	case MSR_GS_BASE:
1671 		msr_info->data = vmcs_readl(GUEST_GS_BASE);
1672 		break;
1673 	case MSR_KERNEL_GS_BASE:
1674 		msr_info->data = vmx_read_guest_kernel_gs_base(vmx);
1675 		break;
1676 #endif
1677 	case MSR_EFER:
1678 		return kvm_get_msr_common(vcpu, msr_info);
1679 	case MSR_IA32_SPEC_CTRL:
1680 		if (!msr_info->host_initiated &&
1681 		    !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
1682 			return 1;
1683 
1684 		msr_info->data = to_vmx(vcpu)->spec_ctrl;
1685 		break;
1686 	case MSR_IA32_ARCH_CAPABILITIES:
1687 		if (!msr_info->host_initiated &&
1688 		    !guest_cpuid_has(vcpu, X86_FEATURE_ARCH_CAPABILITIES))
1689 			return 1;
1690 		msr_info->data = to_vmx(vcpu)->arch_capabilities;
1691 		break;
1692 	case MSR_IA32_SYSENTER_CS:
1693 		msr_info->data = vmcs_read32(GUEST_SYSENTER_CS);
1694 		break;
1695 	case MSR_IA32_SYSENTER_EIP:
1696 		msr_info->data = vmcs_readl(GUEST_SYSENTER_EIP);
1697 		break;
1698 	case MSR_IA32_SYSENTER_ESP:
1699 		msr_info->data = vmcs_readl(GUEST_SYSENTER_ESP);
1700 		break;
1701 	case MSR_IA32_BNDCFGS:
1702 		if (!kvm_mpx_supported() ||
1703 		    (!msr_info->host_initiated &&
1704 		     !guest_cpuid_has(vcpu, X86_FEATURE_MPX)))
1705 			return 1;
1706 		msr_info->data = vmcs_read64(GUEST_BNDCFGS);
1707 		break;
1708 	case MSR_IA32_MCG_EXT_CTL:
1709 		if (!msr_info->host_initiated &&
1710 		    !(vmx->msr_ia32_feature_control &
1711 		      FEATURE_CONTROL_LMCE))
1712 			return 1;
1713 		msr_info->data = vcpu->arch.mcg_ext_ctl;
1714 		break;
1715 	case MSR_IA32_FEATURE_CONTROL:
1716 		msr_info->data = vmx->msr_ia32_feature_control;
1717 		break;
1718 	case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
1719 		if (!nested_vmx_allowed(vcpu))
1720 			return 1;
1721 		return vmx_get_vmx_msr(&vmx->nested.msrs, msr_info->index,
1722 				       &msr_info->data);
1723 	case MSR_IA32_XSS:
1724 		if (!vmx_xsaves_supported())
1725 			return 1;
1726 		msr_info->data = vcpu->arch.ia32_xss;
1727 		break;
1728 	case MSR_IA32_RTIT_CTL:
1729 		if (pt_mode != PT_MODE_HOST_GUEST)
1730 			return 1;
1731 		msr_info->data = vmx->pt_desc.guest.ctl;
1732 		break;
1733 	case MSR_IA32_RTIT_STATUS:
1734 		if (pt_mode != PT_MODE_HOST_GUEST)
1735 			return 1;
1736 		msr_info->data = vmx->pt_desc.guest.status;
1737 		break;
1738 	case MSR_IA32_RTIT_CR3_MATCH:
1739 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
1740 			!intel_pt_validate_cap(vmx->pt_desc.caps,
1741 						PT_CAP_cr3_filtering))
1742 			return 1;
1743 		msr_info->data = vmx->pt_desc.guest.cr3_match;
1744 		break;
1745 	case MSR_IA32_RTIT_OUTPUT_BASE:
1746 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
1747 			(!intel_pt_validate_cap(vmx->pt_desc.caps,
1748 					PT_CAP_topa_output) &&
1749 			 !intel_pt_validate_cap(vmx->pt_desc.caps,
1750 					PT_CAP_single_range_output)))
1751 			return 1;
1752 		msr_info->data = vmx->pt_desc.guest.output_base;
1753 		break;
1754 	case MSR_IA32_RTIT_OUTPUT_MASK:
1755 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
1756 			(!intel_pt_validate_cap(vmx->pt_desc.caps,
1757 					PT_CAP_topa_output) &&
1758 			 !intel_pt_validate_cap(vmx->pt_desc.caps,
1759 					PT_CAP_single_range_output)))
1760 			return 1;
1761 		msr_info->data = vmx->pt_desc.guest.output_mask;
1762 		break;
1763 	case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
1764 		index = msr_info->index - MSR_IA32_RTIT_ADDR0_A;
1765 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
1766 			(index >= 2 * intel_pt_validate_cap(vmx->pt_desc.caps,
1767 					PT_CAP_num_address_ranges)))
1768 			return 1;
1769 		if (index % 2)
1770 			msr_info->data = vmx->pt_desc.guest.addr_b[index / 2];
1771 		else
1772 			msr_info->data = vmx->pt_desc.guest.addr_a[index / 2];
1773 		break;
1774 	case MSR_TSC_AUX:
1775 		if (!msr_info->host_initiated &&
1776 		    !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP))
1777 			return 1;
1778 		/* Else, falls through */
1779 	default:
1780 		msr = find_msr_entry(vmx, msr_info->index);
1781 		if (msr) {
1782 			msr_info->data = msr->data;
1783 			break;
1784 		}
1785 		return kvm_get_msr_common(vcpu, msr_info);
1786 	}
1787 
1788 	return 0;
1789 }
1790 
1791 /*
1792  * Writes msr value into into the appropriate "register".
1793  * Returns 0 on success, non-0 otherwise.
1794  * Assumes vcpu_load() was already called.
1795  */
1796 static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1797 {
1798 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1799 	struct shared_msr_entry *msr;
1800 	int ret = 0;
1801 	u32 msr_index = msr_info->index;
1802 	u64 data = msr_info->data;
1803 	u32 index;
1804 
1805 	switch (msr_index) {
1806 	case MSR_EFER:
1807 		ret = kvm_set_msr_common(vcpu, msr_info);
1808 		break;
1809 #ifdef CONFIG_X86_64
1810 	case MSR_FS_BASE:
1811 		vmx_segment_cache_clear(vmx);
1812 		vmcs_writel(GUEST_FS_BASE, data);
1813 		break;
1814 	case MSR_GS_BASE:
1815 		vmx_segment_cache_clear(vmx);
1816 		vmcs_writel(GUEST_GS_BASE, data);
1817 		break;
1818 	case MSR_KERNEL_GS_BASE:
1819 		vmx_write_guest_kernel_gs_base(vmx, data);
1820 		break;
1821 #endif
1822 	case MSR_IA32_SYSENTER_CS:
1823 		vmcs_write32(GUEST_SYSENTER_CS, data);
1824 		break;
1825 	case MSR_IA32_SYSENTER_EIP:
1826 		vmcs_writel(GUEST_SYSENTER_EIP, data);
1827 		break;
1828 	case MSR_IA32_SYSENTER_ESP:
1829 		vmcs_writel(GUEST_SYSENTER_ESP, data);
1830 		break;
1831 	case MSR_IA32_BNDCFGS:
1832 		if (!kvm_mpx_supported() ||
1833 		    (!msr_info->host_initiated &&
1834 		     !guest_cpuid_has(vcpu, X86_FEATURE_MPX)))
1835 			return 1;
1836 		if (is_noncanonical_address(data & PAGE_MASK, vcpu) ||
1837 		    (data & MSR_IA32_BNDCFGS_RSVD))
1838 			return 1;
1839 		vmcs_write64(GUEST_BNDCFGS, data);
1840 		break;
1841 	case MSR_IA32_SPEC_CTRL:
1842 		if (!msr_info->host_initiated &&
1843 		    !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
1844 			return 1;
1845 
1846 		/* The STIBP bit doesn't fault even if it's not advertised */
1847 		if (data & ~(SPEC_CTRL_IBRS | SPEC_CTRL_STIBP | SPEC_CTRL_SSBD))
1848 			return 1;
1849 
1850 		vmx->spec_ctrl = data;
1851 
1852 		if (!data)
1853 			break;
1854 
1855 		/*
1856 		 * For non-nested:
1857 		 * When it's written (to non-zero) for the first time, pass
1858 		 * it through.
1859 		 *
1860 		 * For nested:
1861 		 * The handling of the MSR bitmap for L2 guests is done in
1862 		 * nested_vmx_merge_msr_bitmap. We should not touch the
1863 		 * vmcs02.msr_bitmap here since it gets completely overwritten
1864 		 * in the merging. We update the vmcs01 here for L1 as well
1865 		 * since it will end up touching the MSR anyway now.
1866 		 */
1867 		vmx_disable_intercept_for_msr(vmx->vmcs01.msr_bitmap,
1868 					      MSR_IA32_SPEC_CTRL,
1869 					      MSR_TYPE_RW);
1870 		break;
1871 	case MSR_IA32_PRED_CMD:
1872 		if (!msr_info->host_initiated &&
1873 		    !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
1874 			return 1;
1875 
1876 		if (data & ~PRED_CMD_IBPB)
1877 			return 1;
1878 
1879 		if (!data)
1880 			break;
1881 
1882 		wrmsrl(MSR_IA32_PRED_CMD, PRED_CMD_IBPB);
1883 
1884 		/*
1885 		 * For non-nested:
1886 		 * When it's written (to non-zero) for the first time, pass
1887 		 * it through.
1888 		 *
1889 		 * For nested:
1890 		 * The handling of the MSR bitmap for L2 guests is done in
1891 		 * nested_vmx_merge_msr_bitmap. We should not touch the
1892 		 * vmcs02.msr_bitmap here since it gets completely overwritten
1893 		 * in the merging.
1894 		 */
1895 		vmx_disable_intercept_for_msr(vmx->vmcs01.msr_bitmap, MSR_IA32_PRED_CMD,
1896 					      MSR_TYPE_W);
1897 		break;
1898 	case MSR_IA32_ARCH_CAPABILITIES:
1899 		if (!msr_info->host_initiated)
1900 			return 1;
1901 		vmx->arch_capabilities = data;
1902 		break;
1903 	case MSR_IA32_CR_PAT:
1904 		if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
1905 			if (!kvm_mtrr_valid(vcpu, MSR_IA32_CR_PAT, data))
1906 				return 1;
1907 			vmcs_write64(GUEST_IA32_PAT, data);
1908 			vcpu->arch.pat = data;
1909 			break;
1910 		}
1911 		ret = kvm_set_msr_common(vcpu, msr_info);
1912 		break;
1913 	case MSR_IA32_TSC_ADJUST:
1914 		ret = kvm_set_msr_common(vcpu, msr_info);
1915 		break;
1916 	case MSR_IA32_MCG_EXT_CTL:
1917 		if ((!msr_info->host_initiated &&
1918 		     !(to_vmx(vcpu)->msr_ia32_feature_control &
1919 		       FEATURE_CONTROL_LMCE)) ||
1920 		    (data & ~MCG_EXT_CTL_LMCE_EN))
1921 			return 1;
1922 		vcpu->arch.mcg_ext_ctl = data;
1923 		break;
1924 	case MSR_IA32_FEATURE_CONTROL:
1925 		if (!vmx_feature_control_msr_valid(vcpu, data) ||
1926 		    (to_vmx(vcpu)->msr_ia32_feature_control &
1927 		     FEATURE_CONTROL_LOCKED && !msr_info->host_initiated))
1928 			return 1;
1929 		vmx->msr_ia32_feature_control = data;
1930 		if (msr_info->host_initiated && data == 0)
1931 			vmx_leave_nested(vcpu);
1932 		break;
1933 	case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
1934 		if (!msr_info->host_initiated)
1935 			return 1; /* they are read-only */
1936 		if (!nested_vmx_allowed(vcpu))
1937 			return 1;
1938 		return vmx_set_vmx_msr(vcpu, msr_index, data);
1939 	case MSR_IA32_XSS:
1940 		if (!vmx_xsaves_supported())
1941 			return 1;
1942 		/*
1943 		 * The only supported bit as of Skylake is bit 8, but
1944 		 * it is not supported on KVM.
1945 		 */
1946 		if (data != 0)
1947 			return 1;
1948 		vcpu->arch.ia32_xss = data;
1949 		if (vcpu->arch.ia32_xss != host_xss)
1950 			add_atomic_switch_msr(vmx, MSR_IA32_XSS,
1951 				vcpu->arch.ia32_xss, host_xss, false);
1952 		else
1953 			clear_atomic_switch_msr(vmx, MSR_IA32_XSS);
1954 		break;
1955 	case MSR_IA32_RTIT_CTL:
1956 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
1957 			vmx_rtit_ctl_check(vcpu, data) ||
1958 			vmx->nested.vmxon)
1959 			return 1;
1960 		vmcs_write64(GUEST_IA32_RTIT_CTL, data);
1961 		vmx->pt_desc.guest.ctl = data;
1962 		pt_update_intercept_for_msr(vmx);
1963 		break;
1964 	case MSR_IA32_RTIT_STATUS:
1965 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
1966 			(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) ||
1967 			(data & MSR_IA32_RTIT_STATUS_MASK))
1968 			return 1;
1969 		vmx->pt_desc.guest.status = data;
1970 		break;
1971 	case MSR_IA32_RTIT_CR3_MATCH:
1972 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
1973 			(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) ||
1974 			!intel_pt_validate_cap(vmx->pt_desc.caps,
1975 						PT_CAP_cr3_filtering))
1976 			return 1;
1977 		vmx->pt_desc.guest.cr3_match = data;
1978 		break;
1979 	case MSR_IA32_RTIT_OUTPUT_BASE:
1980 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
1981 			(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) ||
1982 			(!intel_pt_validate_cap(vmx->pt_desc.caps,
1983 					PT_CAP_topa_output) &&
1984 			 !intel_pt_validate_cap(vmx->pt_desc.caps,
1985 					PT_CAP_single_range_output)) ||
1986 			(data & MSR_IA32_RTIT_OUTPUT_BASE_MASK))
1987 			return 1;
1988 		vmx->pt_desc.guest.output_base = data;
1989 		break;
1990 	case MSR_IA32_RTIT_OUTPUT_MASK:
1991 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
1992 			(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) ||
1993 			(!intel_pt_validate_cap(vmx->pt_desc.caps,
1994 					PT_CAP_topa_output) &&
1995 			 !intel_pt_validate_cap(vmx->pt_desc.caps,
1996 					PT_CAP_single_range_output)))
1997 			return 1;
1998 		vmx->pt_desc.guest.output_mask = data;
1999 		break;
2000 	case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
2001 		index = msr_info->index - MSR_IA32_RTIT_ADDR0_A;
2002 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
2003 			(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) ||
2004 			(index >= 2 * intel_pt_validate_cap(vmx->pt_desc.caps,
2005 					PT_CAP_num_address_ranges)))
2006 			return 1;
2007 		if (index % 2)
2008 			vmx->pt_desc.guest.addr_b[index / 2] = data;
2009 		else
2010 			vmx->pt_desc.guest.addr_a[index / 2] = data;
2011 		break;
2012 	case MSR_TSC_AUX:
2013 		if (!msr_info->host_initiated &&
2014 		    !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP))
2015 			return 1;
2016 		/* Check reserved bit, higher 32 bits should be zero */
2017 		if ((data >> 32) != 0)
2018 			return 1;
2019 		/* Else, falls through */
2020 	default:
2021 		msr = find_msr_entry(vmx, msr_index);
2022 		if (msr) {
2023 			u64 old_msr_data = msr->data;
2024 			msr->data = data;
2025 			if (msr - vmx->guest_msrs < vmx->save_nmsrs) {
2026 				preempt_disable();
2027 				ret = kvm_set_shared_msr(msr->index, msr->data,
2028 							 msr->mask);
2029 				preempt_enable();
2030 				if (ret)
2031 					msr->data = old_msr_data;
2032 			}
2033 			break;
2034 		}
2035 		ret = kvm_set_msr_common(vcpu, msr_info);
2036 	}
2037 
2038 	return ret;
2039 }
2040 
2041 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
2042 {
2043 	__set_bit(reg, (unsigned long *)&vcpu->arch.regs_avail);
2044 	switch (reg) {
2045 	case VCPU_REGS_RSP:
2046 		vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
2047 		break;
2048 	case VCPU_REGS_RIP:
2049 		vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
2050 		break;
2051 	case VCPU_EXREG_PDPTR:
2052 		if (enable_ept)
2053 			ept_save_pdptrs(vcpu);
2054 		break;
2055 	default:
2056 		break;
2057 	}
2058 }
2059 
2060 static __init int cpu_has_kvm_support(void)
2061 {
2062 	return cpu_has_vmx();
2063 }
2064 
2065 static __init int vmx_disabled_by_bios(void)
2066 {
2067 	u64 msr;
2068 
2069 	rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
2070 	if (msr & FEATURE_CONTROL_LOCKED) {
2071 		/* launched w/ TXT and VMX disabled */
2072 		if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
2073 			&& tboot_enabled())
2074 			return 1;
2075 		/* launched w/o TXT and VMX only enabled w/ TXT */
2076 		if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
2077 			&& (msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
2078 			&& !tboot_enabled()) {
2079 			printk(KERN_WARNING "kvm: disable TXT in the BIOS or "
2080 				"activate TXT before enabling KVM\n");
2081 			return 1;
2082 		}
2083 		/* launched w/o TXT and VMX disabled */
2084 		if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
2085 			&& !tboot_enabled())
2086 			return 1;
2087 	}
2088 
2089 	return 0;
2090 }
2091 
2092 static void kvm_cpu_vmxon(u64 addr)
2093 {
2094 	cr4_set_bits(X86_CR4_VMXE);
2095 	intel_pt_handle_vmx(1);
2096 
2097 	asm volatile ("vmxon %0" : : "m"(addr));
2098 }
2099 
2100 static int hardware_enable(void)
2101 {
2102 	int cpu = raw_smp_processor_id();
2103 	u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
2104 	u64 old, test_bits;
2105 
2106 	if (cr4_read_shadow() & X86_CR4_VMXE)
2107 		return -EBUSY;
2108 
2109 	/*
2110 	 * This can happen if we hot-added a CPU but failed to allocate
2111 	 * VP assist page for it.
2112 	 */
2113 	if (static_branch_unlikely(&enable_evmcs) &&
2114 	    !hv_get_vp_assist_page(cpu))
2115 		return -EFAULT;
2116 
2117 	INIT_LIST_HEAD(&per_cpu(loaded_vmcss_on_cpu, cpu));
2118 	INIT_LIST_HEAD(&per_cpu(blocked_vcpu_on_cpu, cpu));
2119 	spin_lock_init(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
2120 
2121 	/*
2122 	 * Now we can enable the vmclear operation in kdump
2123 	 * since the loaded_vmcss_on_cpu list on this cpu
2124 	 * has been initialized.
2125 	 *
2126 	 * Though the cpu is not in VMX operation now, there
2127 	 * is no problem to enable the vmclear operation
2128 	 * for the loaded_vmcss_on_cpu list is empty!
2129 	 */
2130 	crash_enable_local_vmclear(cpu);
2131 
2132 	rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
2133 
2134 	test_bits = FEATURE_CONTROL_LOCKED;
2135 	test_bits |= FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
2136 	if (tboot_enabled())
2137 		test_bits |= FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX;
2138 
2139 	if ((old & test_bits) != test_bits) {
2140 		/* enable and lock */
2141 		wrmsrl(MSR_IA32_FEATURE_CONTROL, old | test_bits);
2142 	}
2143 	kvm_cpu_vmxon(phys_addr);
2144 	if (enable_ept)
2145 		ept_sync_global();
2146 
2147 	return 0;
2148 }
2149 
2150 static void vmclear_local_loaded_vmcss(void)
2151 {
2152 	int cpu = raw_smp_processor_id();
2153 	struct loaded_vmcs *v, *n;
2154 
2155 	list_for_each_entry_safe(v, n, &per_cpu(loaded_vmcss_on_cpu, cpu),
2156 				 loaded_vmcss_on_cpu_link)
2157 		__loaded_vmcs_clear(v);
2158 }
2159 
2160 
2161 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
2162  * tricks.
2163  */
2164 static void kvm_cpu_vmxoff(void)
2165 {
2166 	asm volatile (__ex("vmxoff"));
2167 
2168 	intel_pt_handle_vmx(0);
2169 	cr4_clear_bits(X86_CR4_VMXE);
2170 }
2171 
2172 static void hardware_disable(void)
2173 {
2174 	vmclear_local_loaded_vmcss();
2175 	kvm_cpu_vmxoff();
2176 }
2177 
2178 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
2179 				      u32 msr, u32 *result)
2180 {
2181 	u32 vmx_msr_low, vmx_msr_high;
2182 	u32 ctl = ctl_min | ctl_opt;
2183 
2184 	rdmsr(msr, vmx_msr_low, vmx_msr_high);
2185 
2186 	ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
2187 	ctl |= vmx_msr_low;  /* bit == 1 in low word  ==> must be one  */
2188 
2189 	/* Ensure minimum (required) set of control bits are supported. */
2190 	if (ctl_min & ~ctl)
2191 		return -EIO;
2192 
2193 	*result = ctl;
2194 	return 0;
2195 }
2196 
2197 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf,
2198 				    struct vmx_capability *vmx_cap)
2199 {
2200 	u32 vmx_msr_low, vmx_msr_high;
2201 	u32 min, opt, min2, opt2;
2202 	u32 _pin_based_exec_control = 0;
2203 	u32 _cpu_based_exec_control = 0;
2204 	u32 _cpu_based_2nd_exec_control = 0;
2205 	u32 _vmexit_control = 0;
2206 	u32 _vmentry_control = 0;
2207 
2208 	memset(vmcs_conf, 0, sizeof(*vmcs_conf));
2209 	min = CPU_BASED_HLT_EXITING |
2210 #ifdef CONFIG_X86_64
2211 	      CPU_BASED_CR8_LOAD_EXITING |
2212 	      CPU_BASED_CR8_STORE_EXITING |
2213 #endif
2214 	      CPU_BASED_CR3_LOAD_EXITING |
2215 	      CPU_BASED_CR3_STORE_EXITING |
2216 	      CPU_BASED_UNCOND_IO_EXITING |
2217 	      CPU_BASED_MOV_DR_EXITING |
2218 	      CPU_BASED_USE_TSC_OFFSETING |
2219 	      CPU_BASED_MWAIT_EXITING |
2220 	      CPU_BASED_MONITOR_EXITING |
2221 	      CPU_BASED_INVLPG_EXITING |
2222 	      CPU_BASED_RDPMC_EXITING;
2223 
2224 	opt = CPU_BASED_TPR_SHADOW |
2225 	      CPU_BASED_USE_MSR_BITMAPS |
2226 	      CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
2227 	if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
2228 				&_cpu_based_exec_control) < 0)
2229 		return -EIO;
2230 #ifdef CONFIG_X86_64
2231 	if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2232 		_cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
2233 					   ~CPU_BASED_CR8_STORE_EXITING;
2234 #endif
2235 	if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
2236 		min2 = 0;
2237 		opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2238 			SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2239 			SECONDARY_EXEC_WBINVD_EXITING |
2240 			SECONDARY_EXEC_ENABLE_VPID |
2241 			SECONDARY_EXEC_ENABLE_EPT |
2242 			SECONDARY_EXEC_UNRESTRICTED_GUEST |
2243 			SECONDARY_EXEC_PAUSE_LOOP_EXITING |
2244 			SECONDARY_EXEC_DESC |
2245 			SECONDARY_EXEC_RDTSCP |
2246 			SECONDARY_EXEC_ENABLE_INVPCID |
2247 			SECONDARY_EXEC_APIC_REGISTER_VIRT |
2248 			SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
2249 			SECONDARY_EXEC_SHADOW_VMCS |
2250 			SECONDARY_EXEC_XSAVES |
2251 			SECONDARY_EXEC_RDSEED_EXITING |
2252 			SECONDARY_EXEC_RDRAND_EXITING |
2253 			SECONDARY_EXEC_ENABLE_PML |
2254 			SECONDARY_EXEC_TSC_SCALING |
2255 			SECONDARY_EXEC_PT_USE_GPA |
2256 			SECONDARY_EXEC_PT_CONCEAL_VMX |
2257 			SECONDARY_EXEC_ENABLE_VMFUNC |
2258 			SECONDARY_EXEC_ENCLS_EXITING;
2259 		if (adjust_vmx_controls(min2, opt2,
2260 					MSR_IA32_VMX_PROCBASED_CTLS2,
2261 					&_cpu_based_2nd_exec_control) < 0)
2262 			return -EIO;
2263 	}
2264 #ifndef CONFIG_X86_64
2265 	if (!(_cpu_based_2nd_exec_control &
2266 				SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
2267 		_cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
2268 #endif
2269 
2270 	if (!(_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2271 		_cpu_based_2nd_exec_control &= ~(
2272 				SECONDARY_EXEC_APIC_REGISTER_VIRT |
2273 				SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2274 				SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
2275 
2276 	rdmsr_safe(MSR_IA32_VMX_EPT_VPID_CAP,
2277 		&vmx_cap->ept, &vmx_cap->vpid);
2278 
2279 	if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
2280 		/* CR3 accesses and invlpg don't need to cause VM Exits when EPT
2281 		   enabled */
2282 		_cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
2283 					     CPU_BASED_CR3_STORE_EXITING |
2284 					     CPU_BASED_INVLPG_EXITING);
2285 	} else if (vmx_cap->ept) {
2286 		vmx_cap->ept = 0;
2287 		pr_warn_once("EPT CAP should not exist if not support "
2288 				"1-setting enable EPT VM-execution control\n");
2289 	}
2290 	if (!(_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_VPID) &&
2291 		vmx_cap->vpid) {
2292 		vmx_cap->vpid = 0;
2293 		pr_warn_once("VPID CAP should not exist if not support "
2294 				"1-setting enable VPID VM-execution control\n");
2295 	}
2296 
2297 	min = VM_EXIT_SAVE_DEBUG_CONTROLS | VM_EXIT_ACK_INTR_ON_EXIT;
2298 #ifdef CONFIG_X86_64
2299 	min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
2300 #endif
2301 	opt = VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL |
2302 	      VM_EXIT_SAVE_IA32_PAT |
2303 	      VM_EXIT_LOAD_IA32_PAT |
2304 	      VM_EXIT_LOAD_IA32_EFER |
2305 	      VM_EXIT_CLEAR_BNDCFGS |
2306 	      VM_EXIT_PT_CONCEAL_PIP |
2307 	      VM_EXIT_CLEAR_IA32_RTIT_CTL;
2308 	if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
2309 				&_vmexit_control) < 0)
2310 		return -EIO;
2311 
2312 	min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
2313 	opt = PIN_BASED_VIRTUAL_NMIS | PIN_BASED_POSTED_INTR |
2314 		 PIN_BASED_VMX_PREEMPTION_TIMER;
2315 	if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
2316 				&_pin_based_exec_control) < 0)
2317 		return -EIO;
2318 
2319 	if (cpu_has_broken_vmx_preemption_timer())
2320 		_pin_based_exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
2321 	if (!(_cpu_based_2nd_exec_control &
2322 		SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY))
2323 		_pin_based_exec_control &= ~PIN_BASED_POSTED_INTR;
2324 
2325 	min = VM_ENTRY_LOAD_DEBUG_CONTROLS;
2326 	opt = VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL |
2327 	      VM_ENTRY_LOAD_IA32_PAT |
2328 	      VM_ENTRY_LOAD_IA32_EFER |
2329 	      VM_ENTRY_LOAD_BNDCFGS |
2330 	      VM_ENTRY_PT_CONCEAL_PIP |
2331 	      VM_ENTRY_LOAD_IA32_RTIT_CTL;
2332 	if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
2333 				&_vmentry_control) < 0)
2334 		return -EIO;
2335 
2336 	/*
2337 	 * Some cpus support VM_{ENTRY,EXIT}_IA32_PERF_GLOBAL_CTRL but they
2338 	 * can't be used due to an errata where VM Exit may incorrectly clear
2339 	 * IA32_PERF_GLOBAL_CTRL[34:32].  Workaround the errata by using the
2340 	 * MSR load mechanism to switch IA32_PERF_GLOBAL_CTRL.
2341 	 */
2342 	if (boot_cpu_data.x86 == 0x6) {
2343 		switch (boot_cpu_data.x86_model) {
2344 		case 26: /* AAK155 */
2345 		case 30: /* AAP115 */
2346 		case 37: /* AAT100 */
2347 		case 44: /* BC86,AAY89,BD102 */
2348 		case 46: /* BA97 */
2349 			_vmentry_control &= ~VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL;
2350 			_vmexit_control &= ~VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL;
2351 			pr_warn_once("kvm: VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL "
2352 					"does not work properly. Using workaround\n");
2353 			break;
2354 		default:
2355 			break;
2356 		}
2357 	}
2358 
2359 
2360 	rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
2361 
2362 	/* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
2363 	if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
2364 		return -EIO;
2365 
2366 #ifdef CONFIG_X86_64
2367 	/* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
2368 	if (vmx_msr_high & (1u<<16))
2369 		return -EIO;
2370 #endif
2371 
2372 	/* Require Write-Back (WB) memory type for VMCS accesses. */
2373 	if (((vmx_msr_high >> 18) & 15) != 6)
2374 		return -EIO;
2375 
2376 	vmcs_conf->size = vmx_msr_high & 0x1fff;
2377 	vmcs_conf->order = get_order(vmcs_conf->size);
2378 	vmcs_conf->basic_cap = vmx_msr_high & ~0x1fff;
2379 
2380 	vmcs_conf->revision_id = vmx_msr_low;
2381 
2382 	vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
2383 	vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
2384 	vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
2385 	vmcs_conf->vmexit_ctrl         = _vmexit_control;
2386 	vmcs_conf->vmentry_ctrl        = _vmentry_control;
2387 
2388 	if (static_branch_unlikely(&enable_evmcs))
2389 		evmcs_sanitize_exec_ctrls(vmcs_conf);
2390 
2391 	return 0;
2392 }
2393 
2394 struct vmcs *alloc_vmcs_cpu(bool shadow, int cpu, gfp_t flags)
2395 {
2396 	int node = cpu_to_node(cpu);
2397 	struct page *pages;
2398 	struct vmcs *vmcs;
2399 
2400 	pages = __alloc_pages_node(node, flags, vmcs_config.order);
2401 	if (!pages)
2402 		return NULL;
2403 	vmcs = page_address(pages);
2404 	memset(vmcs, 0, vmcs_config.size);
2405 
2406 	/* KVM supports Enlightened VMCS v1 only */
2407 	if (static_branch_unlikely(&enable_evmcs))
2408 		vmcs->hdr.revision_id = KVM_EVMCS_VERSION;
2409 	else
2410 		vmcs->hdr.revision_id = vmcs_config.revision_id;
2411 
2412 	if (shadow)
2413 		vmcs->hdr.shadow_vmcs = 1;
2414 	return vmcs;
2415 }
2416 
2417 void free_vmcs(struct vmcs *vmcs)
2418 {
2419 	free_pages((unsigned long)vmcs, vmcs_config.order);
2420 }
2421 
2422 /*
2423  * Free a VMCS, but before that VMCLEAR it on the CPU where it was last loaded
2424  */
2425 void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
2426 {
2427 	if (!loaded_vmcs->vmcs)
2428 		return;
2429 	loaded_vmcs_clear(loaded_vmcs);
2430 	free_vmcs(loaded_vmcs->vmcs);
2431 	loaded_vmcs->vmcs = NULL;
2432 	if (loaded_vmcs->msr_bitmap)
2433 		free_page((unsigned long)loaded_vmcs->msr_bitmap);
2434 	WARN_ON(loaded_vmcs->shadow_vmcs != NULL);
2435 }
2436 
2437 int alloc_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
2438 {
2439 	loaded_vmcs->vmcs = alloc_vmcs(false);
2440 	if (!loaded_vmcs->vmcs)
2441 		return -ENOMEM;
2442 
2443 	loaded_vmcs->shadow_vmcs = NULL;
2444 	loaded_vmcs_init(loaded_vmcs);
2445 
2446 	if (cpu_has_vmx_msr_bitmap()) {
2447 		loaded_vmcs->msr_bitmap = (unsigned long *)
2448 				__get_free_page(GFP_KERNEL_ACCOUNT);
2449 		if (!loaded_vmcs->msr_bitmap)
2450 			goto out_vmcs;
2451 		memset(loaded_vmcs->msr_bitmap, 0xff, PAGE_SIZE);
2452 
2453 		if (IS_ENABLED(CONFIG_HYPERV) &&
2454 		    static_branch_unlikely(&enable_evmcs) &&
2455 		    (ms_hyperv.nested_features & HV_X64_NESTED_MSR_BITMAP)) {
2456 			struct hv_enlightened_vmcs *evmcs =
2457 				(struct hv_enlightened_vmcs *)loaded_vmcs->vmcs;
2458 
2459 			evmcs->hv_enlightenments_control.msr_bitmap = 1;
2460 		}
2461 	}
2462 
2463 	memset(&loaded_vmcs->host_state, 0, sizeof(struct vmcs_host_state));
2464 
2465 	return 0;
2466 
2467 out_vmcs:
2468 	free_loaded_vmcs(loaded_vmcs);
2469 	return -ENOMEM;
2470 }
2471 
2472 static void free_kvm_area(void)
2473 {
2474 	int cpu;
2475 
2476 	for_each_possible_cpu(cpu) {
2477 		free_vmcs(per_cpu(vmxarea, cpu));
2478 		per_cpu(vmxarea, cpu) = NULL;
2479 	}
2480 }
2481 
2482 static __init int alloc_kvm_area(void)
2483 {
2484 	int cpu;
2485 
2486 	for_each_possible_cpu(cpu) {
2487 		struct vmcs *vmcs;
2488 
2489 		vmcs = alloc_vmcs_cpu(false, cpu, GFP_KERNEL);
2490 		if (!vmcs) {
2491 			free_kvm_area();
2492 			return -ENOMEM;
2493 		}
2494 
2495 		/*
2496 		 * When eVMCS is enabled, alloc_vmcs_cpu() sets
2497 		 * vmcs->revision_id to KVM_EVMCS_VERSION instead of
2498 		 * revision_id reported by MSR_IA32_VMX_BASIC.
2499 		 *
2500 		 * However, even though not explicitly documented by
2501 		 * TLFS, VMXArea passed as VMXON argument should
2502 		 * still be marked with revision_id reported by
2503 		 * physical CPU.
2504 		 */
2505 		if (static_branch_unlikely(&enable_evmcs))
2506 			vmcs->hdr.revision_id = vmcs_config.revision_id;
2507 
2508 		per_cpu(vmxarea, cpu) = vmcs;
2509 	}
2510 	return 0;
2511 }
2512 
2513 static void fix_pmode_seg(struct kvm_vcpu *vcpu, int seg,
2514 		struct kvm_segment *save)
2515 {
2516 	if (!emulate_invalid_guest_state) {
2517 		/*
2518 		 * CS and SS RPL should be equal during guest entry according
2519 		 * to VMX spec, but in reality it is not always so. Since vcpu
2520 		 * is in the middle of the transition from real mode to
2521 		 * protected mode it is safe to assume that RPL 0 is a good
2522 		 * default value.
2523 		 */
2524 		if (seg == VCPU_SREG_CS || seg == VCPU_SREG_SS)
2525 			save->selector &= ~SEGMENT_RPL_MASK;
2526 		save->dpl = save->selector & SEGMENT_RPL_MASK;
2527 		save->s = 1;
2528 	}
2529 	vmx_set_segment(vcpu, save, seg);
2530 }
2531 
2532 static void enter_pmode(struct kvm_vcpu *vcpu)
2533 {
2534 	unsigned long flags;
2535 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2536 
2537 	/*
2538 	 * Update real mode segment cache. It may be not up-to-date if sement
2539 	 * register was written while vcpu was in a guest mode.
2540 	 */
2541 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
2542 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
2543 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
2544 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
2545 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
2546 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
2547 
2548 	vmx->rmode.vm86_active = 0;
2549 
2550 	vmx_segment_cache_clear(vmx);
2551 
2552 	vmx_set_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
2553 
2554 	flags = vmcs_readl(GUEST_RFLAGS);
2555 	flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
2556 	flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
2557 	vmcs_writel(GUEST_RFLAGS, flags);
2558 
2559 	vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
2560 			(vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
2561 
2562 	update_exception_bitmap(vcpu);
2563 
2564 	fix_pmode_seg(vcpu, VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
2565 	fix_pmode_seg(vcpu, VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
2566 	fix_pmode_seg(vcpu, VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
2567 	fix_pmode_seg(vcpu, VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
2568 	fix_pmode_seg(vcpu, VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
2569 	fix_pmode_seg(vcpu, VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
2570 }
2571 
2572 static void fix_rmode_seg(int seg, struct kvm_segment *save)
2573 {
2574 	const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2575 	struct kvm_segment var = *save;
2576 
2577 	var.dpl = 0x3;
2578 	if (seg == VCPU_SREG_CS)
2579 		var.type = 0x3;
2580 
2581 	if (!emulate_invalid_guest_state) {
2582 		var.selector = var.base >> 4;
2583 		var.base = var.base & 0xffff0;
2584 		var.limit = 0xffff;
2585 		var.g = 0;
2586 		var.db = 0;
2587 		var.present = 1;
2588 		var.s = 1;
2589 		var.l = 0;
2590 		var.unusable = 0;
2591 		var.type = 0x3;
2592 		var.avl = 0;
2593 		if (save->base & 0xf)
2594 			printk_once(KERN_WARNING "kvm: segment base is not "
2595 					"paragraph aligned when entering "
2596 					"protected mode (seg=%d)", seg);
2597 	}
2598 
2599 	vmcs_write16(sf->selector, var.selector);
2600 	vmcs_writel(sf->base, var.base);
2601 	vmcs_write32(sf->limit, var.limit);
2602 	vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(&var));
2603 }
2604 
2605 static void enter_rmode(struct kvm_vcpu *vcpu)
2606 {
2607 	unsigned long flags;
2608 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2609 	struct kvm_vmx *kvm_vmx = to_kvm_vmx(vcpu->kvm);
2610 
2611 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
2612 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
2613 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
2614 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
2615 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
2616 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
2617 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
2618 
2619 	vmx->rmode.vm86_active = 1;
2620 
2621 	/*
2622 	 * Very old userspace does not call KVM_SET_TSS_ADDR before entering
2623 	 * vcpu. Warn the user that an update is overdue.
2624 	 */
2625 	if (!kvm_vmx->tss_addr)
2626 		printk_once(KERN_WARNING "kvm: KVM_SET_TSS_ADDR need to be "
2627 			     "called before entering vcpu\n");
2628 
2629 	vmx_segment_cache_clear(vmx);
2630 
2631 	vmcs_writel(GUEST_TR_BASE, kvm_vmx->tss_addr);
2632 	vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
2633 	vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
2634 
2635 	flags = vmcs_readl(GUEST_RFLAGS);
2636 	vmx->rmode.save_rflags = flags;
2637 
2638 	flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
2639 
2640 	vmcs_writel(GUEST_RFLAGS, flags);
2641 	vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
2642 	update_exception_bitmap(vcpu);
2643 
2644 	fix_rmode_seg(VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
2645 	fix_rmode_seg(VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
2646 	fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
2647 	fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
2648 	fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
2649 	fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
2650 
2651 	kvm_mmu_reset_context(vcpu);
2652 }
2653 
2654 void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
2655 {
2656 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2657 	struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
2658 
2659 	if (!msr)
2660 		return;
2661 
2662 	vcpu->arch.efer = efer;
2663 	if (efer & EFER_LMA) {
2664 		vm_entry_controls_setbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
2665 		msr->data = efer;
2666 	} else {
2667 		vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
2668 
2669 		msr->data = efer & ~EFER_LME;
2670 	}
2671 	setup_msrs(vmx);
2672 }
2673 
2674 #ifdef CONFIG_X86_64
2675 
2676 static void enter_lmode(struct kvm_vcpu *vcpu)
2677 {
2678 	u32 guest_tr_ar;
2679 
2680 	vmx_segment_cache_clear(to_vmx(vcpu));
2681 
2682 	guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
2683 	if ((guest_tr_ar & VMX_AR_TYPE_MASK) != VMX_AR_TYPE_BUSY_64_TSS) {
2684 		pr_debug_ratelimited("%s: tss fixup for long mode. \n",
2685 				     __func__);
2686 		vmcs_write32(GUEST_TR_AR_BYTES,
2687 			     (guest_tr_ar & ~VMX_AR_TYPE_MASK)
2688 			     | VMX_AR_TYPE_BUSY_64_TSS);
2689 	}
2690 	vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
2691 }
2692 
2693 static void exit_lmode(struct kvm_vcpu *vcpu)
2694 {
2695 	vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
2696 	vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
2697 }
2698 
2699 #endif
2700 
2701 static void vmx_flush_tlb_gva(struct kvm_vcpu *vcpu, gva_t addr)
2702 {
2703 	int vpid = to_vmx(vcpu)->vpid;
2704 
2705 	if (!vpid_sync_vcpu_addr(vpid, addr))
2706 		vpid_sync_context(vpid);
2707 
2708 	/*
2709 	 * If VPIDs are not supported or enabled, then the above is a no-op.
2710 	 * But we don't really need a TLB flush in that case anyway, because
2711 	 * each VM entry/exit includes an implicit flush when VPID is 0.
2712 	 */
2713 }
2714 
2715 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
2716 {
2717 	ulong cr0_guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
2718 
2719 	vcpu->arch.cr0 &= ~cr0_guest_owned_bits;
2720 	vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & cr0_guest_owned_bits;
2721 }
2722 
2723 static void vmx_decache_cr3(struct kvm_vcpu *vcpu)
2724 {
2725 	if (enable_unrestricted_guest || (enable_ept && is_paging(vcpu)))
2726 		vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
2727 	__set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
2728 }
2729 
2730 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
2731 {
2732 	ulong cr4_guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
2733 
2734 	vcpu->arch.cr4 &= ~cr4_guest_owned_bits;
2735 	vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & cr4_guest_owned_bits;
2736 }
2737 
2738 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
2739 {
2740 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
2741 
2742 	if (!test_bit(VCPU_EXREG_PDPTR,
2743 		      (unsigned long *)&vcpu->arch.regs_dirty))
2744 		return;
2745 
2746 	if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
2747 		vmcs_write64(GUEST_PDPTR0, mmu->pdptrs[0]);
2748 		vmcs_write64(GUEST_PDPTR1, mmu->pdptrs[1]);
2749 		vmcs_write64(GUEST_PDPTR2, mmu->pdptrs[2]);
2750 		vmcs_write64(GUEST_PDPTR3, mmu->pdptrs[3]);
2751 	}
2752 }
2753 
2754 void ept_save_pdptrs(struct kvm_vcpu *vcpu)
2755 {
2756 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
2757 
2758 	if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
2759 		mmu->pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
2760 		mmu->pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
2761 		mmu->pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
2762 		mmu->pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
2763 	}
2764 
2765 	__set_bit(VCPU_EXREG_PDPTR,
2766 		  (unsigned long *)&vcpu->arch.regs_avail);
2767 	__set_bit(VCPU_EXREG_PDPTR,
2768 		  (unsigned long *)&vcpu->arch.regs_dirty);
2769 }
2770 
2771 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
2772 					unsigned long cr0,
2773 					struct kvm_vcpu *vcpu)
2774 {
2775 	if (!test_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail))
2776 		vmx_decache_cr3(vcpu);
2777 	if (!(cr0 & X86_CR0_PG)) {
2778 		/* From paging/starting to nonpaging */
2779 		vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
2780 			     vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) |
2781 			     (CPU_BASED_CR3_LOAD_EXITING |
2782 			      CPU_BASED_CR3_STORE_EXITING));
2783 		vcpu->arch.cr0 = cr0;
2784 		vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
2785 	} else if (!is_paging(vcpu)) {
2786 		/* From nonpaging to paging */
2787 		vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
2788 			     vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) &
2789 			     ~(CPU_BASED_CR3_LOAD_EXITING |
2790 			       CPU_BASED_CR3_STORE_EXITING));
2791 		vcpu->arch.cr0 = cr0;
2792 		vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
2793 	}
2794 
2795 	if (!(cr0 & X86_CR0_WP))
2796 		*hw_cr0 &= ~X86_CR0_WP;
2797 }
2798 
2799 void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
2800 {
2801 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2802 	unsigned long hw_cr0;
2803 
2804 	hw_cr0 = (cr0 & ~KVM_VM_CR0_ALWAYS_OFF);
2805 	if (enable_unrestricted_guest)
2806 		hw_cr0 |= KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
2807 	else {
2808 		hw_cr0 |= KVM_VM_CR0_ALWAYS_ON;
2809 
2810 		if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
2811 			enter_pmode(vcpu);
2812 
2813 		if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
2814 			enter_rmode(vcpu);
2815 	}
2816 
2817 #ifdef CONFIG_X86_64
2818 	if (vcpu->arch.efer & EFER_LME) {
2819 		if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
2820 			enter_lmode(vcpu);
2821 		if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
2822 			exit_lmode(vcpu);
2823 	}
2824 #endif
2825 
2826 	if (enable_ept && !enable_unrestricted_guest)
2827 		ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
2828 
2829 	vmcs_writel(CR0_READ_SHADOW, cr0);
2830 	vmcs_writel(GUEST_CR0, hw_cr0);
2831 	vcpu->arch.cr0 = cr0;
2832 
2833 	/* depends on vcpu->arch.cr0 to be set to a new value */
2834 	vmx->emulation_required = emulation_required(vcpu);
2835 }
2836 
2837 static int get_ept_level(struct kvm_vcpu *vcpu)
2838 {
2839 	if (cpu_has_vmx_ept_5levels() && (cpuid_maxphyaddr(vcpu) > 48))
2840 		return 5;
2841 	return 4;
2842 }
2843 
2844 u64 construct_eptp(struct kvm_vcpu *vcpu, unsigned long root_hpa)
2845 {
2846 	u64 eptp = VMX_EPTP_MT_WB;
2847 
2848 	eptp |= (get_ept_level(vcpu) == 5) ? VMX_EPTP_PWL_5 : VMX_EPTP_PWL_4;
2849 
2850 	if (enable_ept_ad_bits &&
2851 	    (!is_guest_mode(vcpu) || nested_ept_ad_enabled(vcpu)))
2852 		eptp |= VMX_EPTP_AD_ENABLE_BIT;
2853 	eptp |= (root_hpa & PAGE_MASK);
2854 
2855 	return eptp;
2856 }
2857 
2858 void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
2859 {
2860 	struct kvm *kvm = vcpu->kvm;
2861 	unsigned long guest_cr3;
2862 	u64 eptp;
2863 
2864 	guest_cr3 = cr3;
2865 	if (enable_ept) {
2866 		eptp = construct_eptp(vcpu, cr3);
2867 		vmcs_write64(EPT_POINTER, eptp);
2868 
2869 		if (kvm_x86_ops->tlb_remote_flush) {
2870 			spin_lock(&to_kvm_vmx(kvm)->ept_pointer_lock);
2871 			to_vmx(vcpu)->ept_pointer = eptp;
2872 			to_kvm_vmx(kvm)->ept_pointers_match
2873 				= EPT_POINTERS_CHECK;
2874 			spin_unlock(&to_kvm_vmx(kvm)->ept_pointer_lock);
2875 		}
2876 
2877 		if (enable_unrestricted_guest || is_paging(vcpu) ||
2878 		    is_guest_mode(vcpu))
2879 			guest_cr3 = kvm_read_cr3(vcpu);
2880 		else
2881 			guest_cr3 = to_kvm_vmx(kvm)->ept_identity_map_addr;
2882 		ept_load_pdptrs(vcpu);
2883 	}
2884 
2885 	vmcs_writel(GUEST_CR3, guest_cr3);
2886 }
2887 
2888 int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
2889 {
2890 	/*
2891 	 * Pass through host's Machine Check Enable value to hw_cr4, which
2892 	 * is in force while we are in guest mode.  Do not let guests control
2893 	 * this bit, even if host CR4.MCE == 0.
2894 	 */
2895 	unsigned long hw_cr4;
2896 
2897 	hw_cr4 = (cr4_read_shadow() & X86_CR4_MCE) | (cr4 & ~X86_CR4_MCE);
2898 	if (enable_unrestricted_guest)
2899 		hw_cr4 |= KVM_VM_CR4_ALWAYS_ON_UNRESTRICTED_GUEST;
2900 	else if (to_vmx(vcpu)->rmode.vm86_active)
2901 		hw_cr4 |= KVM_RMODE_VM_CR4_ALWAYS_ON;
2902 	else
2903 		hw_cr4 |= KVM_PMODE_VM_CR4_ALWAYS_ON;
2904 
2905 	if (!boot_cpu_has(X86_FEATURE_UMIP) && vmx_umip_emulated()) {
2906 		if (cr4 & X86_CR4_UMIP) {
2907 			vmcs_set_bits(SECONDARY_VM_EXEC_CONTROL,
2908 				SECONDARY_EXEC_DESC);
2909 			hw_cr4 &= ~X86_CR4_UMIP;
2910 		} else if (!is_guest_mode(vcpu) ||
2911 			!nested_cpu_has2(get_vmcs12(vcpu), SECONDARY_EXEC_DESC))
2912 			vmcs_clear_bits(SECONDARY_VM_EXEC_CONTROL,
2913 					SECONDARY_EXEC_DESC);
2914 	}
2915 
2916 	if (cr4 & X86_CR4_VMXE) {
2917 		/*
2918 		 * To use VMXON (and later other VMX instructions), a guest
2919 		 * must first be able to turn on cr4.VMXE (see handle_vmon()).
2920 		 * So basically the check on whether to allow nested VMX
2921 		 * is here.  We operate under the default treatment of SMM,
2922 		 * so VMX cannot be enabled under SMM.
2923 		 */
2924 		if (!nested_vmx_allowed(vcpu) || is_smm(vcpu))
2925 			return 1;
2926 	}
2927 
2928 	if (to_vmx(vcpu)->nested.vmxon && !nested_cr4_valid(vcpu, cr4))
2929 		return 1;
2930 
2931 	vcpu->arch.cr4 = cr4;
2932 
2933 	if (!enable_unrestricted_guest) {
2934 		if (enable_ept) {
2935 			if (!is_paging(vcpu)) {
2936 				hw_cr4 &= ~X86_CR4_PAE;
2937 				hw_cr4 |= X86_CR4_PSE;
2938 			} else if (!(cr4 & X86_CR4_PAE)) {
2939 				hw_cr4 &= ~X86_CR4_PAE;
2940 			}
2941 		}
2942 
2943 		/*
2944 		 * SMEP/SMAP/PKU is disabled if CPU is in non-paging mode in
2945 		 * hardware.  To emulate this behavior, SMEP/SMAP/PKU needs
2946 		 * to be manually disabled when guest switches to non-paging
2947 		 * mode.
2948 		 *
2949 		 * If !enable_unrestricted_guest, the CPU is always running
2950 		 * with CR0.PG=1 and CR4 needs to be modified.
2951 		 * If enable_unrestricted_guest, the CPU automatically
2952 		 * disables SMEP/SMAP/PKU when the guest sets CR0.PG=0.
2953 		 */
2954 		if (!is_paging(vcpu))
2955 			hw_cr4 &= ~(X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE);
2956 	}
2957 
2958 	vmcs_writel(CR4_READ_SHADOW, cr4);
2959 	vmcs_writel(GUEST_CR4, hw_cr4);
2960 	return 0;
2961 }
2962 
2963 void vmx_get_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg)
2964 {
2965 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2966 	u32 ar;
2967 
2968 	if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
2969 		*var = vmx->rmode.segs[seg];
2970 		if (seg == VCPU_SREG_TR
2971 		    || var->selector == vmx_read_guest_seg_selector(vmx, seg))
2972 			return;
2973 		var->base = vmx_read_guest_seg_base(vmx, seg);
2974 		var->selector = vmx_read_guest_seg_selector(vmx, seg);
2975 		return;
2976 	}
2977 	var->base = vmx_read_guest_seg_base(vmx, seg);
2978 	var->limit = vmx_read_guest_seg_limit(vmx, seg);
2979 	var->selector = vmx_read_guest_seg_selector(vmx, seg);
2980 	ar = vmx_read_guest_seg_ar(vmx, seg);
2981 	var->unusable = (ar >> 16) & 1;
2982 	var->type = ar & 15;
2983 	var->s = (ar >> 4) & 1;
2984 	var->dpl = (ar >> 5) & 3;
2985 	/*
2986 	 * Some userspaces do not preserve unusable property. Since usable
2987 	 * segment has to be present according to VMX spec we can use present
2988 	 * property to amend userspace bug by making unusable segment always
2989 	 * nonpresent. vmx_segment_access_rights() already marks nonpresent
2990 	 * segment as unusable.
2991 	 */
2992 	var->present = !var->unusable;
2993 	var->avl = (ar >> 12) & 1;
2994 	var->l = (ar >> 13) & 1;
2995 	var->db = (ar >> 14) & 1;
2996 	var->g = (ar >> 15) & 1;
2997 }
2998 
2999 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
3000 {
3001 	struct kvm_segment s;
3002 
3003 	if (to_vmx(vcpu)->rmode.vm86_active) {
3004 		vmx_get_segment(vcpu, &s, seg);
3005 		return s.base;
3006 	}
3007 	return vmx_read_guest_seg_base(to_vmx(vcpu), seg);
3008 }
3009 
3010 int vmx_get_cpl(struct kvm_vcpu *vcpu)
3011 {
3012 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3013 
3014 	if (unlikely(vmx->rmode.vm86_active))
3015 		return 0;
3016 	else {
3017 		int ar = vmx_read_guest_seg_ar(vmx, VCPU_SREG_SS);
3018 		return VMX_AR_DPL(ar);
3019 	}
3020 }
3021 
3022 static u32 vmx_segment_access_rights(struct kvm_segment *var)
3023 {
3024 	u32 ar;
3025 
3026 	if (var->unusable || !var->present)
3027 		ar = 1 << 16;
3028 	else {
3029 		ar = var->type & 15;
3030 		ar |= (var->s & 1) << 4;
3031 		ar |= (var->dpl & 3) << 5;
3032 		ar |= (var->present & 1) << 7;
3033 		ar |= (var->avl & 1) << 12;
3034 		ar |= (var->l & 1) << 13;
3035 		ar |= (var->db & 1) << 14;
3036 		ar |= (var->g & 1) << 15;
3037 	}
3038 
3039 	return ar;
3040 }
3041 
3042 void vmx_set_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg)
3043 {
3044 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3045 	const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3046 
3047 	vmx_segment_cache_clear(vmx);
3048 
3049 	if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3050 		vmx->rmode.segs[seg] = *var;
3051 		if (seg == VCPU_SREG_TR)
3052 			vmcs_write16(sf->selector, var->selector);
3053 		else if (var->s)
3054 			fix_rmode_seg(seg, &vmx->rmode.segs[seg]);
3055 		goto out;
3056 	}
3057 
3058 	vmcs_writel(sf->base, var->base);
3059 	vmcs_write32(sf->limit, var->limit);
3060 	vmcs_write16(sf->selector, var->selector);
3061 
3062 	/*
3063 	 *   Fix the "Accessed" bit in AR field of segment registers for older
3064 	 * qemu binaries.
3065 	 *   IA32 arch specifies that at the time of processor reset the
3066 	 * "Accessed" bit in the AR field of segment registers is 1. And qemu
3067 	 * is setting it to 0 in the userland code. This causes invalid guest
3068 	 * state vmexit when "unrestricted guest" mode is turned on.
3069 	 *    Fix for this setup issue in cpu_reset is being pushed in the qemu
3070 	 * tree. Newer qemu binaries with that qemu fix would not need this
3071 	 * kvm hack.
3072 	 */
3073 	if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
3074 		var->type |= 0x1; /* Accessed */
3075 
3076 	vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(var));
3077 
3078 out:
3079 	vmx->emulation_required = emulation_required(vcpu);
3080 }
3081 
3082 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3083 {
3084 	u32 ar = vmx_read_guest_seg_ar(to_vmx(vcpu), VCPU_SREG_CS);
3085 
3086 	*db = (ar >> 14) & 1;
3087 	*l = (ar >> 13) & 1;
3088 }
3089 
3090 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3091 {
3092 	dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
3093 	dt->address = vmcs_readl(GUEST_IDTR_BASE);
3094 }
3095 
3096 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3097 {
3098 	vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
3099 	vmcs_writel(GUEST_IDTR_BASE, dt->address);
3100 }
3101 
3102 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3103 {
3104 	dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
3105 	dt->address = vmcs_readl(GUEST_GDTR_BASE);
3106 }
3107 
3108 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3109 {
3110 	vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
3111 	vmcs_writel(GUEST_GDTR_BASE, dt->address);
3112 }
3113 
3114 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
3115 {
3116 	struct kvm_segment var;
3117 	u32 ar;
3118 
3119 	vmx_get_segment(vcpu, &var, seg);
3120 	var.dpl = 0x3;
3121 	if (seg == VCPU_SREG_CS)
3122 		var.type = 0x3;
3123 	ar = vmx_segment_access_rights(&var);
3124 
3125 	if (var.base != (var.selector << 4))
3126 		return false;
3127 	if (var.limit != 0xffff)
3128 		return false;
3129 	if (ar != 0xf3)
3130 		return false;
3131 
3132 	return true;
3133 }
3134 
3135 static bool code_segment_valid(struct kvm_vcpu *vcpu)
3136 {
3137 	struct kvm_segment cs;
3138 	unsigned int cs_rpl;
3139 
3140 	vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3141 	cs_rpl = cs.selector & SEGMENT_RPL_MASK;
3142 
3143 	if (cs.unusable)
3144 		return false;
3145 	if (~cs.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_ACCESSES_MASK))
3146 		return false;
3147 	if (!cs.s)
3148 		return false;
3149 	if (cs.type & VMX_AR_TYPE_WRITEABLE_MASK) {
3150 		if (cs.dpl > cs_rpl)
3151 			return false;
3152 	} else {
3153 		if (cs.dpl != cs_rpl)
3154 			return false;
3155 	}
3156 	if (!cs.present)
3157 		return false;
3158 
3159 	/* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
3160 	return true;
3161 }
3162 
3163 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
3164 {
3165 	struct kvm_segment ss;
3166 	unsigned int ss_rpl;
3167 
3168 	vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3169 	ss_rpl = ss.selector & SEGMENT_RPL_MASK;
3170 
3171 	if (ss.unusable)
3172 		return true;
3173 	if (ss.type != 3 && ss.type != 7)
3174 		return false;
3175 	if (!ss.s)
3176 		return false;
3177 	if (ss.dpl != ss_rpl) /* DPL != RPL */
3178 		return false;
3179 	if (!ss.present)
3180 		return false;
3181 
3182 	return true;
3183 }
3184 
3185 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
3186 {
3187 	struct kvm_segment var;
3188 	unsigned int rpl;
3189 
3190 	vmx_get_segment(vcpu, &var, seg);
3191 	rpl = var.selector & SEGMENT_RPL_MASK;
3192 
3193 	if (var.unusable)
3194 		return true;
3195 	if (!var.s)
3196 		return false;
3197 	if (!var.present)
3198 		return false;
3199 	if (~var.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_WRITEABLE_MASK)) {
3200 		if (var.dpl < rpl) /* DPL < RPL */
3201 			return false;
3202 	}
3203 
3204 	/* TODO: Add other members to kvm_segment_field to allow checking for other access
3205 	 * rights flags
3206 	 */
3207 	return true;
3208 }
3209 
3210 static bool tr_valid(struct kvm_vcpu *vcpu)
3211 {
3212 	struct kvm_segment tr;
3213 
3214 	vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
3215 
3216 	if (tr.unusable)
3217 		return false;
3218 	if (tr.selector & SEGMENT_TI_MASK)	/* TI = 1 */
3219 		return false;
3220 	if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
3221 		return false;
3222 	if (!tr.present)
3223 		return false;
3224 
3225 	return true;
3226 }
3227 
3228 static bool ldtr_valid(struct kvm_vcpu *vcpu)
3229 {
3230 	struct kvm_segment ldtr;
3231 
3232 	vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
3233 
3234 	if (ldtr.unusable)
3235 		return true;
3236 	if (ldtr.selector & SEGMENT_TI_MASK)	/* TI = 1 */
3237 		return false;
3238 	if (ldtr.type != 2)
3239 		return false;
3240 	if (!ldtr.present)
3241 		return false;
3242 
3243 	return true;
3244 }
3245 
3246 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
3247 {
3248 	struct kvm_segment cs, ss;
3249 
3250 	vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3251 	vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3252 
3253 	return ((cs.selector & SEGMENT_RPL_MASK) ==
3254 		 (ss.selector & SEGMENT_RPL_MASK));
3255 }
3256 
3257 /*
3258  * Check if guest state is valid. Returns true if valid, false if
3259  * not.
3260  * We assume that registers are always usable
3261  */
3262 static bool guest_state_valid(struct kvm_vcpu *vcpu)
3263 {
3264 	if (enable_unrestricted_guest)
3265 		return true;
3266 
3267 	/* real mode guest state checks */
3268 	if (!is_protmode(vcpu) || (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
3269 		if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
3270 			return false;
3271 		if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
3272 			return false;
3273 		if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
3274 			return false;
3275 		if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
3276 			return false;
3277 		if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
3278 			return false;
3279 		if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
3280 			return false;
3281 	} else {
3282 	/* protected mode guest state checks */
3283 		if (!cs_ss_rpl_check(vcpu))
3284 			return false;
3285 		if (!code_segment_valid(vcpu))
3286 			return false;
3287 		if (!stack_segment_valid(vcpu))
3288 			return false;
3289 		if (!data_segment_valid(vcpu, VCPU_SREG_DS))
3290 			return false;
3291 		if (!data_segment_valid(vcpu, VCPU_SREG_ES))
3292 			return false;
3293 		if (!data_segment_valid(vcpu, VCPU_SREG_FS))
3294 			return false;
3295 		if (!data_segment_valid(vcpu, VCPU_SREG_GS))
3296 			return false;
3297 		if (!tr_valid(vcpu))
3298 			return false;
3299 		if (!ldtr_valid(vcpu))
3300 			return false;
3301 	}
3302 	/* TODO:
3303 	 * - Add checks on RIP
3304 	 * - Add checks on RFLAGS
3305 	 */
3306 
3307 	return true;
3308 }
3309 
3310 static int init_rmode_tss(struct kvm *kvm)
3311 {
3312 	gfn_t fn;
3313 	u16 data = 0;
3314 	int idx, r;
3315 
3316 	idx = srcu_read_lock(&kvm->srcu);
3317 	fn = to_kvm_vmx(kvm)->tss_addr >> PAGE_SHIFT;
3318 	r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3319 	if (r < 0)
3320 		goto out;
3321 	data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
3322 	r = kvm_write_guest_page(kvm, fn++, &data,
3323 			TSS_IOPB_BASE_OFFSET, sizeof(u16));
3324 	if (r < 0)
3325 		goto out;
3326 	r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
3327 	if (r < 0)
3328 		goto out;
3329 	r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3330 	if (r < 0)
3331 		goto out;
3332 	data = ~0;
3333 	r = kvm_write_guest_page(kvm, fn, &data,
3334 				 RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
3335 				 sizeof(u8));
3336 out:
3337 	srcu_read_unlock(&kvm->srcu, idx);
3338 	return r;
3339 }
3340 
3341 static int init_rmode_identity_map(struct kvm *kvm)
3342 {
3343 	struct kvm_vmx *kvm_vmx = to_kvm_vmx(kvm);
3344 	int i, idx, r = 0;
3345 	kvm_pfn_t identity_map_pfn;
3346 	u32 tmp;
3347 
3348 	/* Protect kvm_vmx->ept_identity_pagetable_done. */
3349 	mutex_lock(&kvm->slots_lock);
3350 
3351 	if (likely(kvm_vmx->ept_identity_pagetable_done))
3352 		goto out2;
3353 
3354 	if (!kvm_vmx->ept_identity_map_addr)
3355 		kvm_vmx->ept_identity_map_addr = VMX_EPT_IDENTITY_PAGETABLE_ADDR;
3356 	identity_map_pfn = kvm_vmx->ept_identity_map_addr >> PAGE_SHIFT;
3357 
3358 	r = __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
3359 				    kvm_vmx->ept_identity_map_addr, PAGE_SIZE);
3360 	if (r < 0)
3361 		goto out2;
3362 
3363 	idx = srcu_read_lock(&kvm->srcu);
3364 	r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
3365 	if (r < 0)
3366 		goto out;
3367 	/* Set up identity-mapping pagetable for EPT in real mode */
3368 	for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
3369 		tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
3370 			_PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
3371 		r = kvm_write_guest_page(kvm, identity_map_pfn,
3372 				&tmp, i * sizeof(tmp), sizeof(tmp));
3373 		if (r < 0)
3374 			goto out;
3375 	}
3376 	kvm_vmx->ept_identity_pagetable_done = true;
3377 
3378 out:
3379 	srcu_read_unlock(&kvm->srcu, idx);
3380 
3381 out2:
3382 	mutex_unlock(&kvm->slots_lock);
3383 	return r;
3384 }
3385 
3386 static void seg_setup(int seg)
3387 {
3388 	const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3389 	unsigned int ar;
3390 
3391 	vmcs_write16(sf->selector, 0);
3392 	vmcs_writel(sf->base, 0);
3393 	vmcs_write32(sf->limit, 0xffff);
3394 	ar = 0x93;
3395 	if (seg == VCPU_SREG_CS)
3396 		ar |= 0x08; /* code segment */
3397 
3398 	vmcs_write32(sf->ar_bytes, ar);
3399 }
3400 
3401 static int alloc_apic_access_page(struct kvm *kvm)
3402 {
3403 	struct page *page;
3404 	int r = 0;
3405 
3406 	mutex_lock(&kvm->slots_lock);
3407 	if (kvm->arch.apic_access_page_done)
3408 		goto out;
3409 	r = __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
3410 				    APIC_DEFAULT_PHYS_BASE, PAGE_SIZE);
3411 	if (r)
3412 		goto out;
3413 
3414 	page = gfn_to_page(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
3415 	if (is_error_page(page)) {
3416 		r = -EFAULT;
3417 		goto out;
3418 	}
3419 
3420 	/*
3421 	 * Do not pin the page in memory, so that memory hot-unplug
3422 	 * is able to migrate it.
3423 	 */
3424 	put_page(page);
3425 	kvm->arch.apic_access_page_done = true;
3426 out:
3427 	mutex_unlock(&kvm->slots_lock);
3428 	return r;
3429 }
3430 
3431 int allocate_vpid(void)
3432 {
3433 	int vpid;
3434 
3435 	if (!enable_vpid)
3436 		return 0;
3437 	spin_lock(&vmx_vpid_lock);
3438 	vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
3439 	if (vpid < VMX_NR_VPIDS)
3440 		__set_bit(vpid, vmx_vpid_bitmap);
3441 	else
3442 		vpid = 0;
3443 	spin_unlock(&vmx_vpid_lock);
3444 	return vpid;
3445 }
3446 
3447 void free_vpid(int vpid)
3448 {
3449 	if (!enable_vpid || vpid == 0)
3450 		return;
3451 	spin_lock(&vmx_vpid_lock);
3452 	__clear_bit(vpid, vmx_vpid_bitmap);
3453 	spin_unlock(&vmx_vpid_lock);
3454 }
3455 
3456 static __always_inline void vmx_disable_intercept_for_msr(unsigned long *msr_bitmap,
3457 							  u32 msr, int type)
3458 {
3459 	int f = sizeof(unsigned long);
3460 
3461 	if (!cpu_has_vmx_msr_bitmap())
3462 		return;
3463 
3464 	if (static_branch_unlikely(&enable_evmcs))
3465 		evmcs_touch_msr_bitmap();
3466 
3467 	/*
3468 	 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
3469 	 * have the write-low and read-high bitmap offsets the wrong way round.
3470 	 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
3471 	 */
3472 	if (msr <= 0x1fff) {
3473 		if (type & MSR_TYPE_R)
3474 			/* read-low */
3475 			__clear_bit(msr, msr_bitmap + 0x000 / f);
3476 
3477 		if (type & MSR_TYPE_W)
3478 			/* write-low */
3479 			__clear_bit(msr, msr_bitmap + 0x800 / f);
3480 
3481 	} else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
3482 		msr &= 0x1fff;
3483 		if (type & MSR_TYPE_R)
3484 			/* read-high */
3485 			__clear_bit(msr, msr_bitmap + 0x400 / f);
3486 
3487 		if (type & MSR_TYPE_W)
3488 			/* write-high */
3489 			__clear_bit(msr, msr_bitmap + 0xc00 / f);
3490 
3491 	}
3492 }
3493 
3494 static __always_inline void vmx_enable_intercept_for_msr(unsigned long *msr_bitmap,
3495 							 u32 msr, int type)
3496 {
3497 	int f = sizeof(unsigned long);
3498 
3499 	if (!cpu_has_vmx_msr_bitmap())
3500 		return;
3501 
3502 	if (static_branch_unlikely(&enable_evmcs))
3503 		evmcs_touch_msr_bitmap();
3504 
3505 	/*
3506 	 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
3507 	 * have the write-low and read-high bitmap offsets the wrong way round.
3508 	 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
3509 	 */
3510 	if (msr <= 0x1fff) {
3511 		if (type & MSR_TYPE_R)
3512 			/* read-low */
3513 			__set_bit(msr, msr_bitmap + 0x000 / f);
3514 
3515 		if (type & MSR_TYPE_W)
3516 			/* write-low */
3517 			__set_bit(msr, msr_bitmap + 0x800 / f);
3518 
3519 	} else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
3520 		msr &= 0x1fff;
3521 		if (type & MSR_TYPE_R)
3522 			/* read-high */
3523 			__set_bit(msr, msr_bitmap + 0x400 / f);
3524 
3525 		if (type & MSR_TYPE_W)
3526 			/* write-high */
3527 			__set_bit(msr, msr_bitmap + 0xc00 / f);
3528 
3529 	}
3530 }
3531 
3532 static __always_inline void vmx_set_intercept_for_msr(unsigned long *msr_bitmap,
3533 			     			      u32 msr, int type, bool value)
3534 {
3535 	if (value)
3536 		vmx_enable_intercept_for_msr(msr_bitmap, msr, type);
3537 	else
3538 		vmx_disable_intercept_for_msr(msr_bitmap, msr, type);
3539 }
3540 
3541 static u8 vmx_msr_bitmap_mode(struct kvm_vcpu *vcpu)
3542 {
3543 	u8 mode = 0;
3544 
3545 	if (cpu_has_secondary_exec_ctrls() &&
3546 	    (vmcs_read32(SECONDARY_VM_EXEC_CONTROL) &
3547 	     SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE)) {
3548 		mode |= MSR_BITMAP_MODE_X2APIC;
3549 		if (enable_apicv && kvm_vcpu_apicv_active(vcpu))
3550 			mode |= MSR_BITMAP_MODE_X2APIC_APICV;
3551 	}
3552 
3553 	return mode;
3554 }
3555 
3556 static void vmx_update_msr_bitmap_x2apic(unsigned long *msr_bitmap,
3557 					 u8 mode)
3558 {
3559 	int msr;
3560 
3561 	for (msr = 0x800; msr <= 0x8ff; msr += BITS_PER_LONG) {
3562 		unsigned word = msr / BITS_PER_LONG;
3563 		msr_bitmap[word] = (mode & MSR_BITMAP_MODE_X2APIC_APICV) ? 0 : ~0;
3564 		msr_bitmap[word + (0x800 / sizeof(long))] = ~0;
3565 	}
3566 
3567 	if (mode & MSR_BITMAP_MODE_X2APIC) {
3568 		/*
3569 		 * TPR reads and writes can be virtualized even if virtual interrupt
3570 		 * delivery is not in use.
3571 		 */
3572 		vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_TASKPRI), MSR_TYPE_RW);
3573 		if (mode & MSR_BITMAP_MODE_X2APIC_APICV) {
3574 			vmx_enable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_TMCCT), MSR_TYPE_R);
3575 			vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_EOI), MSR_TYPE_W);
3576 			vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_SELF_IPI), MSR_TYPE_W);
3577 		}
3578 	}
3579 }
3580 
3581 void vmx_update_msr_bitmap(struct kvm_vcpu *vcpu)
3582 {
3583 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3584 	unsigned long *msr_bitmap = vmx->vmcs01.msr_bitmap;
3585 	u8 mode = vmx_msr_bitmap_mode(vcpu);
3586 	u8 changed = mode ^ vmx->msr_bitmap_mode;
3587 
3588 	if (!changed)
3589 		return;
3590 
3591 	if (changed & (MSR_BITMAP_MODE_X2APIC | MSR_BITMAP_MODE_X2APIC_APICV))
3592 		vmx_update_msr_bitmap_x2apic(msr_bitmap, mode);
3593 
3594 	vmx->msr_bitmap_mode = mode;
3595 }
3596 
3597 void pt_update_intercept_for_msr(struct vcpu_vmx *vmx)
3598 {
3599 	unsigned long *msr_bitmap = vmx->vmcs01.msr_bitmap;
3600 	bool flag = !(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN);
3601 	u32 i;
3602 
3603 	vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_STATUS,
3604 							MSR_TYPE_RW, flag);
3605 	vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_OUTPUT_BASE,
3606 							MSR_TYPE_RW, flag);
3607 	vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_OUTPUT_MASK,
3608 							MSR_TYPE_RW, flag);
3609 	vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_CR3_MATCH,
3610 							MSR_TYPE_RW, flag);
3611 	for (i = 0; i < vmx->pt_desc.addr_range; i++) {
3612 		vmx_set_intercept_for_msr(msr_bitmap,
3613 			MSR_IA32_RTIT_ADDR0_A + i * 2, MSR_TYPE_RW, flag);
3614 		vmx_set_intercept_for_msr(msr_bitmap,
3615 			MSR_IA32_RTIT_ADDR0_B + i * 2, MSR_TYPE_RW, flag);
3616 	}
3617 }
3618 
3619 static bool vmx_get_enable_apicv(struct kvm_vcpu *vcpu)
3620 {
3621 	return enable_apicv;
3622 }
3623 
3624 static bool vmx_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
3625 {
3626 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3627 	void *vapic_page;
3628 	u32 vppr;
3629 	int rvi;
3630 
3631 	if (WARN_ON_ONCE(!is_guest_mode(vcpu)) ||
3632 		!nested_cpu_has_vid(get_vmcs12(vcpu)) ||
3633 		WARN_ON_ONCE(!vmx->nested.virtual_apic_page))
3634 		return false;
3635 
3636 	rvi = vmx_get_rvi();
3637 
3638 	vapic_page = kmap(vmx->nested.virtual_apic_page);
3639 	vppr = *((u32 *)(vapic_page + APIC_PROCPRI));
3640 	kunmap(vmx->nested.virtual_apic_page);
3641 
3642 	return ((rvi & 0xf0) > (vppr & 0xf0));
3643 }
3644 
3645 static inline bool kvm_vcpu_trigger_posted_interrupt(struct kvm_vcpu *vcpu,
3646 						     bool nested)
3647 {
3648 #ifdef CONFIG_SMP
3649 	int pi_vec = nested ? POSTED_INTR_NESTED_VECTOR : POSTED_INTR_VECTOR;
3650 
3651 	if (vcpu->mode == IN_GUEST_MODE) {
3652 		/*
3653 		 * The vector of interrupt to be delivered to vcpu had
3654 		 * been set in PIR before this function.
3655 		 *
3656 		 * Following cases will be reached in this block, and
3657 		 * we always send a notification event in all cases as
3658 		 * explained below.
3659 		 *
3660 		 * Case 1: vcpu keeps in non-root mode. Sending a
3661 		 * notification event posts the interrupt to vcpu.
3662 		 *
3663 		 * Case 2: vcpu exits to root mode and is still
3664 		 * runnable. PIR will be synced to vIRR before the
3665 		 * next vcpu entry. Sending a notification event in
3666 		 * this case has no effect, as vcpu is not in root
3667 		 * mode.
3668 		 *
3669 		 * Case 3: vcpu exits to root mode and is blocked.
3670 		 * vcpu_block() has already synced PIR to vIRR and
3671 		 * never blocks vcpu if vIRR is not cleared. Therefore,
3672 		 * a blocked vcpu here does not wait for any requested
3673 		 * interrupts in PIR, and sending a notification event
3674 		 * which has no effect is safe here.
3675 		 */
3676 
3677 		apic->send_IPI_mask(get_cpu_mask(vcpu->cpu), pi_vec);
3678 		return true;
3679 	}
3680 #endif
3681 	return false;
3682 }
3683 
3684 static int vmx_deliver_nested_posted_interrupt(struct kvm_vcpu *vcpu,
3685 						int vector)
3686 {
3687 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3688 
3689 	if (is_guest_mode(vcpu) &&
3690 	    vector == vmx->nested.posted_intr_nv) {
3691 		/*
3692 		 * If a posted intr is not recognized by hardware,
3693 		 * we will accomplish it in the next vmentry.
3694 		 */
3695 		vmx->nested.pi_pending = true;
3696 		kvm_make_request(KVM_REQ_EVENT, vcpu);
3697 		/* the PIR and ON have been set by L1. */
3698 		if (!kvm_vcpu_trigger_posted_interrupt(vcpu, true))
3699 			kvm_vcpu_kick(vcpu);
3700 		return 0;
3701 	}
3702 	return -1;
3703 }
3704 /*
3705  * Send interrupt to vcpu via posted interrupt way.
3706  * 1. If target vcpu is running(non-root mode), send posted interrupt
3707  * notification to vcpu and hardware will sync PIR to vIRR atomically.
3708  * 2. If target vcpu isn't running(root mode), kick it to pick up the
3709  * interrupt from PIR in next vmentry.
3710  */
3711 static void vmx_deliver_posted_interrupt(struct kvm_vcpu *vcpu, int vector)
3712 {
3713 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3714 	int r;
3715 
3716 	r = vmx_deliver_nested_posted_interrupt(vcpu, vector);
3717 	if (!r)
3718 		return;
3719 
3720 	if (pi_test_and_set_pir(vector, &vmx->pi_desc))
3721 		return;
3722 
3723 	/* If a previous notification has sent the IPI, nothing to do.  */
3724 	if (pi_test_and_set_on(&vmx->pi_desc))
3725 		return;
3726 
3727 	if (!kvm_vcpu_trigger_posted_interrupt(vcpu, false))
3728 		kvm_vcpu_kick(vcpu);
3729 }
3730 
3731 /*
3732  * Set up the vmcs's constant host-state fields, i.e., host-state fields that
3733  * will not change in the lifetime of the guest.
3734  * Note that host-state that does change is set elsewhere. E.g., host-state
3735  * that is set differently for each CPU is set in vmx_vcpu_load(), not here.
3736  */
3737 void vmx_set_constant_host_state(struct vcpu_vmx *vmx)
3738 {
3739 	u32 low32, high32;
3740 	unsigned long tmpl;
3741 	struct desc_ptr dt;
3742 	unsigned long cr0, cr3, cr4;
3743 
3744 	cr0 = read_cr0();
3745 	WARN_ON(cr0 & X86_CR0_TS);
3746 	vmcs_writel(HOST_CR0, cr0);  /* 22.2.3 */
3747 
3748 	/*
3749 	 * Save the most likely value for this task's CR3 in the VMCS.
3750 	 * We can't use __get_current_cr3_fast() because we're not atomic.
3751 	 */
3752 	cr3 = __read_cr3();
3753 	vmcs_writel(HOST_CR3, cr3);		/* 22.2.3  FIXME: shadow tables */
3754 	vmx->loaded_vmcs->host_state.cr3 = cr3;
3755 
3756 	/* Save the most likely value for this task's CR4 in the VMCS. */
3757 	cr4 = cr4_read_shadow();
3758 	vmcs_writel(HOST_CR4, cr4);			/* 22.2.3, 22.2.5 */
3759 	vmx->loaded_vmcs->host_state.cr4 = cr4;
3760 
3761 	vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS);  /* 22.2.4 */
3762 #ifdef CONFIG_X86_64
3763 	/*
3764 	 * Load null selectors, so we can avoid reloading them in
3765 	 * vmx_prepare_switch_to_host(), in case userspace uses
3766 	 * the null selectors too (the expected case).
3767 	 */
3768 	vmcs_write16(HOST_DS_SELECTOR, 0);
3769 	vmcs_write16(HOST_ES_SELECTOR, 0);
3770 #else
3771 	vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
3772 	vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
3773 #endif
3774 	vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
3775 	vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8);  /* 22.2.4 */
3776 
3777 	store_idt(&dt);
3778 	vmcs_writel(HOST_IDTR_BASE, dt.address);   /* 22.2.4 */
3779 	vmx->host_idt_base = dt.address;
3780 
3781 	vmcs_writel(HOST_RIP, (unsigned long)vmx_vmexit); /* 22.2.5 */
3782 
3783 	rdmsr(MSR_IA32_SYSENTER_CS, low32, high32);
3784 	vmcs_write32(HOST_IA32_SYSENTER_CS, low32);
3785 	rdmsrl(MSR_IA32_SYSENTER_EIP, tmpl);
3786 	vmcs_writel(HOST_IA32_SYSENTER_EIP, tmpl);   /* 22.2.3 */
3787 
3788 	if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
3789 		rdmsr(MSR_IA32_CR_PAT, low32, high32);
3790 		vmcs_write64(HOST_IA32_PAT, low32 | ((u64) high32 << 32));
3791 	}
3792 
3793 	if (cpu_has_load_ia32_efer())
3794 		vmcs_write64(HOST_IA32_EFER, host_efer);
3795 }
3796 
3797 void set_cr4_guest_host_mask(struct vcpu_vmx *vmx)
3798 {
3799 	vmx->vcpu.arch.cr4_guest_owned_bits = KVM_CR4_GUEST_OWNED_BITS;
3800 	if (enable_ept)
3801 		vmx->vcpu.arch.cr4_guest_owned_bits |= X86_CR4_PGE;
3802 	if (is_guest_mode(&vmx->vcpu))
3803 		vmx->vcpu.arch.cr4_guest_owned_bits &=
3804 			~get_vmcs12(&vmx->vcpu)->cr4_guest_host_mask;
3805 	vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits);
3806 }
3807 
3808 static u32 vmx_pin_based_exec_ctrl(struct vcpu_vmx *vmx)
3809 {
3810 	u32 pin_based_exec_ctrl = vmcs_config.pin_based_exec_ctrl;
3811 
3812 	if (!kvm_vcpu_apicv_active(&vmx->vcpu))
3813 		pin_based_exec_ctrl &= ~PIN_BASED_POSTED_INTR;
3814 
3815 	if (!enable_vnmi)
3816 		pin_based_exec_ctrl &= ~PIN_BASED_VIRTUAL_NMIS;
3817 
3818 	/* Enable the preemption timer dynamically */
3819 	pin_based_exec_ctrl &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
3820 	return pin_based_exec_ctrl;
3821 }
3822 
3823 static void vmx_refresh_apicv_exec_ctrl(struct kvm_vcpu *vcpu)
3824 {
3825 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3826 
3827 	vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, vmx_pin_based_exec_ctrl(vmx));
3828 	if (cpu_has_secondary_exec_ctrls()) {
3829 		if (kvm_vcpu_apicv_active(vcpu))
3830 			vmcs_set_bits(SECONDARY_VM_EXEC_CONTROL,
3831 				      SECONDARY_EXEC_APIC_REGISTER_VIRT |
3832 				      SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
3833 		else
3834 			vmcs_clear_bits(SECONDARY_VM_EXEC_CONTROL,
3835 					SECONDARY_EXEC_APIC_REGISTER_VIRT |
3836 					SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
3837 	}
3838 
3839 	if (cpu_has_vmx_msr_bitmap())
3840 		vmx_update_msr_bitmap(vcpu);
3841 }
3842 
3843 u32 vmx_exec_control(struct vcpu_vmx *vmx)
3844 {
3845 	u32 exec_control = vmcs_config.cpu_based_exec_ctrl;
3846 
3847 	if (vmx->vcpu.arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)
3848 		exec_control &= ~CPU_BASED_MOV_DR_EXITING;
3849 
3850 	if (!cpu_need_tpr_shadow(&vmx->vcpu)) {
3851 		exec_control &= ~CPU_BASED_TPR_SHADOW;
3852 #ifdef CONFIG_X86_64
3853 		exec_control |= CPU_BASED_CR8_STORE_EXITING |
3854 				CPU_BASED_CR8_LOAD_EXITING;
3855 #endif
3856 	}
3857 	if (!enable_ept)
3858 		exec_control |= CPU_BASED_CR3_STORE_EXITING |
3859 				CPU_BASED_CR3_LOAD_EXITING  |
3860 				CPU_BASED_INVLPG_EXITING;
3861 	if (kvm_mwait_in_guest(vmx->vcpu.kvm))
3862 		exec_control &= ~(CPU_BASED_MWAIT_EXITING |
3863 				CPU_BASED_MONITOR_EXITING);
3864 	if (kvm_hlt_in_guest(vmx->vcpu.kvm))
3865 		exec_control &= ~CPU_BASED_HLT_EXITING;
3866 	return exec_control;
3867 }
3868 
3869 
3870 static void vmx_compute_secondary_exec_control(struct vcpu_vmx *vmx)
3871 {
3872 	struct kvm_vcpu *vcpu = &vmx->vcpu;
3873 
3874 	u32 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
3875 
3876 	if (pt_mode == PT_MODE_SYSTEM)
3877 		exec_control &= ~(SECONDARY_EXEC_PT_USE_GPA | SECONDARY_EXEC_PT_CONCEAL_VMX);
3878 	if (!cpu_need_virtualize_apic_accesses(vcpu))
3879 		exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
3880 	if (vmx->vpid == 0)
3881 		exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
3882 	if (!enable_ept) {
3883 		exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
3884 		enable_unrestricted_guest = 0;
3885 	}
3886 	if (!enable_unrestricted_guest)
3887 		exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
3888 	if (kvm_pause_in_guest(vmx->vcpu.kvm))
3889 		exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
3890 	if (!kvm_vcpu_apicv_active(vcpu))
3891 		exec_control &= ~(SECONDARY_EXEC_APIC_REGISTER_VIRT |
3892 				  SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
3893 	exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
3894 
3895 	/* SECONDARY_EXEC_DESC is enabled/disabled on writes to CR4.UMIP,
3896 	 * in vmx_set_cr4.  */
3897 	exec_control &= ~SECONDARY_EXEC_DESC;
3898 
3899 	/* SECONDARY_EXEC_SHADOW_VMCS is enabled when L1 executes VMPTRLD
3900 	   (handle_vmptrld).
3901 	   We can NOT enable shadow_vmcs here because we don't have yet
3902 	   a current VMCS12
3903 	*/
3904 	exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
3905 
3906 	if (!enable_pml)
3907 		exec_control &= ~SECONDARY_EXEC_ENABLE_PML;
3908 
3909 	if (vmx_xsaves_supported()) {
3910 		/* Exposing XSAVES only when XSAVE is exposed */
3911 		bool xsaves_enabled =
3912 			guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
3913 			guest_cpuid_has(vcpu, X86_FEATURE_XSAVES);
3914 
3915 		if (!xsaves_enabled)
3916 			exec_control &= ~SECONDARY_EXEC_XSAVES;
3917 
3918 		if (nested) {
3919 			if (xsaves_enabled)
3920 				vmx->nested.msrs.secondary_ctls_high |=
3921 					SECONDARY_EXEC_XSAVES;
3922 			else
3923 				vmx->nested.msrs.secondary_ctls_high &=
3924 					~SECONDARY_EXEC_XSAVES;
3925 		}
3926 	}
3927 
3928 	if (vmx_rdtscp_supported()) {
3929 		bool rdtscp_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP);
3930 		if (!rdtscp_enabled)
3931 			exec_control &= ~SECONDARY_EXEC_RDTSCP;
3932 
3933 		if (nested) {
3934 			if (rdtscp_enabled)
3935 				vmx->nested.msrs.secondary_ctls_high |=
3936 					SECONDARY_EXEC_RDTSCP;
3937 			else
3938 				vmx->nested.msrs.secondary_ctls_high &=
3939 					~SECONDARY_EXEC_RDTSCP;
3940 		}
3941 	}
3942 
3943 	if (vmx_invpcid_supported()) {
3944 		/* Exposing INVPCID only when PCID is exposed */
3945 		bool invpcid_enabled =
3946 			guest_cpuid_has(vcpu, X86_FEATURE_INVPCID) &&
3947 			guest_cpuid_has(vcpu, X86_FEATURE_PCID);
3948 
3949 		if (!invpcid_enabled) {
3950 			exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID;
3951 			guest_cpuid_clear(vcpu, X86_FEATURE_INVPCID);
3952 		}
3953 
3954 		if (nested) {
3955 			if (invpcid_enabled)
3956 				vmx->nested.msrs.secondary_ctls_high |=
3957 					SECONDARY_EXEC_ENABLE_INVPCID;
3958 			else
3959 				vmx->nested.msrs.secondary_ctls_high &=
3960 					~SECONDARY_EXEC_ENABLE_INVPCID;
3961 		}
3962 	}
3963 
3964 	if (vmx_rdrand_supported()) {
3965 		bool rdrand_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDRAND);
3966 		if (rdrand_enabled)
3967 			exec_control &= ~SECONDARY_EXEC_RDRAND_EXITING;
3968 
3969 		if (nested) {
3970 			if (rdrand_enabled)
3971 				vmx->nested.msrs.secondary_ctls_high |=
3972 					SECONDARY_EXEC_RDRAND_EXITING;
3973 			else
3974 				vmx->nested.msrs.secondary_ctls_high &=
3975 					~SECONDARY_EXEC_RDRAND_EXITING;
3976 		}
3977 	}
3978 
3979 	if (vmx_rdseed_supported()) {
3980 		bool rdseed_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDSEED);
3981 		if (rdseed_enabled)
3982 			exec_control &= ~SECONDARY_EXEC_RDSEED_EXITING;
3983 
3984 		if (nested) {
3985 			if (rdseed_enabled)
3986 				vmx->nested.msrs.secondary_ctls_high |=
3987 					SECONDARY_EXEC_RDSEED_EXITING;
3988 			else
3989 				vmx->nested.msrs.secondary_ctls_high &=
3990 					~SECONDARY_EXEC_RDSEED_EXITING;
3991 		}
3992 	}
3993 
3994 	vmx->secondary_exec_control = exec_control;
3995 }
3996 
3997 static void ept_set_mmio_spte_mask(void)
3998 {
3999 	/*
4000 	 * EPT Misconfigurations can be generated if the value of bits 2:0
4001 	 * of an EPT paging-structure entry is 110b (write/execute).
4002 	 */
4003 	kvm_mmu_set_mmio_spte_mask(VMX_EPT_RWX_MASK,
4004 				   VMX_EPT_MISCONFIG_WX_VALUE);
4005 }
4006 
4007 #define VMX_XSS_EXIT_BITMAP 0
4008 
4009 /*
4010  * Sets up the vmcs for emulated real mode.
4011  */
4012 static void vmx_vcpu_setup(struct vcpu_vmx *vmx)
4013 {
4014 	int i;
4015 
4016 	if (nested)
4017 		nested_vmx_vcpu_setup();
4018 
4019 	if (cpu_has_vmx_msr_bitmap())
4020 		vmcs_write64(MSR_BITMAP, __pa(vmx->vmcs01.msr_bitmap));
4021 
4022 	vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
4023 
4024 	/* Control */
4025 	vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, vmx_pin_based_exec_ctrl(vmx));
4026 	vmx->hv_deadline_tsc = -1;
4027 
4028 	vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, vmx_exec_control(vmx));
4029 
4030 	if (cpu_has_secondary_exec_ctrls()) {
4031 		vmx_compute_secondary_exec_control(vmx);
4032 		vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
4033 			     vmx->secondary_exec_control);
4034 	}
4035 
4036 	if (kvm_vcpu_apicv_active(&vmx->vcpu)) {
4037 		vmcs_write64(EOI_EXIT_BITMAP0, 0);
4038 		vmcs_write64(EOI_EXIT_BITMAP1, 0);
4039 		vmcs_write64(EOI_EXIT_BITMAP2, 0);
4040 		vmcs_write64(EOI_EXIT_BITMAP3, 0);
4041 
4042 		vmcs_write16(GUEST_INTR_STATUS, 0);
4043 
4044 		vmcs_write16(POSTED_INTR_NV, POSTED_INTR_VECTOR);
4045 		vmcs_write64(POSTED_INTR_DESC_ADDR, __pa((&vmx->pi_desc)));
4046 	}
4047 
4048 	if (!kvm_pause_in_guest(vmx->vcpu.kvm)) {
4049 		vmcs_write32(PLE_GAP, ple_gap);
4050 		vmx->ple_window = ple_window;
4051 		vmx->ple_window_dirty = true;
4052 	}
4053 
4054 	vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
4055 	vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
4056 	vmcs_write32(CR3_TARGET_COUNT, 0);           /* 22.2.1 */
4057 
4058 	vmcs_write16(HOST_FS_SELECTOR, 0);            /* 22.2.4 */
4059 	vmcs_write16(HOST_GS_SELECTOR, 0);            /* 22.2.4 */
4060 	vmx_set_constant_host_state(vmx);
4061 	vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
4062 	vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
4063 
4064 	if (cpu_has_vmx_vmfunc())
4065 		vmcs_write64(VM_FUNCTION_CONTROL, 0);
4066 
4067 	vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
4068 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
4069 	vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host.val));
4070 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
4071 	vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest.val));
4072 
4073 	if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
4074 		vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
4075 
4076 	for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i) {
4077 		u32 index = vmx_msr_index[i];
4078 		u32 data_low, data_high;
4079 		int j = vmx->nmsrs;
4080 
4081 		if (rdmsr_safe(index, &data_low, &data_high) < 0)
4082 			continue;
4083 		if (wrmsr_safe(index, data_low, data_high) < 0)
4084 			continue;
4085 		vmx->guest_msrs[j].index = i;
4086 		vmx->guest_msrs[j].data = 0;
4087 		vmx->guest_msrs[j].mask = -1ull;
4088 		++vmx->nmsrs;
4089 	}
4090 
4091 	vmx->arch_capabilities = kvm_get_arch_capabilities();
4092 
4093 	vm_exit_controls_init(vmx, vmx_vmexit_ctrl());
4094 
4095 	/* 22.2.1, 20.8.1 */
4096 	vm_entry_controls_init(vmx, vmx_vmentry_ctrl());
4097 
4098 	vmx->vcpu.arch.cr0_guest_owned_bits = X86_CR0_TS;
4099 	vmcs_writel(CR0_GUEST_HOST_MASK, ~X86_CR0_TS);
4100 
4101 	set_cr4_guest_host_mask(vmx);
4102 
4103 	if (vmx_xsaves_supported())
4104 		vmcs_write64(XSS_EXIT_BITMAP, VMX_XSS_EXIT_BITMAP);
4105 
4106 	if (enable_pml) {
4107 		vmcs_write64(PML_ADDRESS, page_to_phys(vmx->pml_pg));
4108 		vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
4109 	}
4110 
4111 	if (cpu_has_vmx_encls_vmexit())
4112 		vmcs_write64(ENCLS_EXITING_BITMAP, -1ull);
4113 
4114 	if (pt_mode == PT_MODE_HOST_GUEST) {
4115 		memset(&vmx->pt_desc, 0, sizeof(vmx->pt_desc));
4116 		/* Bit[6~0] are forced to 1, writes are ignored. */
4117 		vmx->pt_desc.guest.output_mask = 0x7F;
4118 		vmcs_write64(GUEST_IA32_RTIT_CTL, 0);
4119 	}
4120 }
4121 
4122 static void vmx_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
4123 {
4124 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4125 	struct msr_data apic_base_msr;
4126 	u64 cr0;
4127 
4128 	vmx->rmode.vm86_active = 0;
4129 	vmx->spec_ctrl = 0;
4130 
4131 	vcpu->arch.microcode_version = 0x100000000ULL;
4132 	vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
4133 	kvm_set_cr8(vcpu, 0);
4134 
4135 	if (!init_event) {
4136 		apic_base_msr.data = APIC_DEFAULT_PHYS_BASE |
4137 				     MSR_IA32_APICBASE_ENABLE;
4138 		if (kvm_vcpu_is_reset_bsp(vcpu))
4139 			apic_base_msr.data |= MSR_IA32_APICBASE_BSP;
4140 		apic_base_msr.host_initiated = true;
4141 		kvm_set_apic_base(vcpu, &apic_base_msr);
4142 	}
4143 
4144 	vmx_segment_cache_clear(vmx);
4145 
4146 	seg_setup(VCPU_SREG_CS);
4147 	vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
4148 	vmcs_writel(GUEST_CS_BASE, 0xffff0000ul);
4149 
4150 	seg_setup(VCPU_SREG_DS);
4151 	seg_setup(VCPU_SREG_ES);
4152 	seg_setup(VCPU_SREG_FS);
4153 	seg_setup(VCPU_SREG_GS);
4154 	seg_setup(VCPU_SREG_SS);
4155 
4156 	vmcs_write16(GUEST_TR_SELECTOR, 0);
4157 	vmcs_writel(GUEST_TR_BASE, 0);
4158 	vmcs_write32(GUEST_TR_LIMIT, 0xffff);
4159 	vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
4160 
4161 	vmcs_write16(GUEST_LDTR_SELECTOR, 0);
4162 	vmcs_writel(GUEST_LDTR_BASE, 0);
4163 	vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
4164 	vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
4165 
4166 	if (!init_event) {
4167 		vmcs_write32(GUEST_SYSENTER_CS, 0);
4168 		vmcs_writel(GUEST_SYSENTER_ESP, 0);
4169 		vmcs_writel(GUEST_SYSENTER_EIP, 0);
4170 		vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
4171 	}
4172 
4173 	kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
4174 	kvm_rip_write(vcpu, 0xfff0);
4175 
4176 	vmcs_writel(GUEST_GDTR_BASE, 0);
4177 	vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
4178 
4179 	vmcs_writel(GUEST_IDTR_BASE, 0);
4180 	vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
4181 
4182 	vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
4183 	vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
4184 	vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS, 0);
4185 	if (kvm_mpx_supported())
4186 		vmcs_write64(GUEST_BNDCFGS, 0);
4187 
4188 	setup_msrs(vmx);
4189 
4190 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);  /* 22.2.1 */
4191 
4192 	if (cpu_has_vmx_tpr_shadow() && !init_event) {
4193 		vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
4194 		if (cpu_need_tpr_shadow(vcpu))
4195 			vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
4196 				     __pa(vcpu->arch.apic->regs));
4197 		vmcs_write32(TPR_THRESHOLD, 0);
4198 	}
4199 
4200 	kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
4201 
4202 	if (vmx->vpid != 0)
4203 		vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
4204 
4205 	cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
4206 	vmx->vcpu.arch.cr0 = cr0;
4207 	vmx_set_cr0(vcpu, cr0); /* enter rmode */
4208 	vmx_set_cr4(vcpu, 0);
4209 	vmx_set_efer(vcpu, 0);
4210 
4211 	update_exception_bitmap(vcpu);
4212 
4213 	vpid_sync_context(vmx->vpid);
4214 	if (init_event)
4215 		vmx_clear_hlt(vcpu);
4216 }
4217 
4218 static void enable_irq_window(struct kvm_vcpu *vcpu)
4219 {
4220 	vmcs_set_bits(CPU_BASED_VM_EXEC_CONTROL,
4221 		      CPU_BASED_VIRTUAL_INTR_PENDING);
4222 }
4223 
4224 static void enable_nmi_window(struct kvm_vcpu *vcpu)
4225 {
4226 	if (!enable_vnmi ||
4227 	    vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_STI) {
4228 		enable_irq_window(vcpu);
4229 		return;
4230 	}
4231 
4232 	vmcs_set_bits(CPU_BASED_VM_EXEC_CONTROL,
4233 		      CPU_BASED_VIRTUAL_NMI_PENDING);
4234 }
4235 
4236 static void vmx_inject_irq(struct kvm_vcpu *vcpu)
4237 {
4238 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4239 	uint32_t intr;
4240 	int irq = vcpu->arch.interrupt.nr;
4241 
4242 	trace_kvm_inj_virq(irq);
4243 
4244 	++vcpu->stat.irq_injections;
4245 	if (vmx->rmode.vm86_active) {
4246 		int inc_eip = 0;
4247 		if (vcpu->arch.interrupt.soft)
4248 			inc_eip = vcpu->arch.event_exit_inst_len;
4249 		if (kvm_inject_realmode_interrupt(vcpu, irq, inc_eip) != EMULATE_DONE)
4250 			kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4251 		return;
4252 	}
4253 	intr = irq | INTR_INFO_VALID_MASK;
4254 	if (vcpu->arch.interrupt.soft) {
4255 		intr |= INTR_TYPE_SOFT_INTR;
4256 		vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
4257 			     vmx->vcpu.arch.event_exit_inst_len);
4258 	} else
4259 		intr |= INTR_TYPE_EXT_INTR;
4260 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
4261 
4262 	vmx_clear_hlt(vcpu);
4263 }
4264 
4265 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
4266 {
4267 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4268 
4269 	if (!enable_vnmi) {
4270 		/*
4271 		 * Tracking the NMI-blocked state in software is built upon
4272 		 * finding the next open IRQ window. This, in turn, depends on
4273 		 * well-behaving guests: They have to keep IRQs disabled at
4274 		 * least as long as the NMI handler runs. Otherwise we may
4275 		 * cause NMI nesting, maybe breaking the guest. But as this is
4276 		 * highly unlikely, we can live with the residual risk.
4277 		 */
4278 		vmx->loaded_vmcs->soft_vnmi_blocked = 1;
4279 		vmx->loaded_vmcs->vnmi_blocked_time = 0;
4280 	}
4281 
4282 	++vcpu->stat.nmi_injections;
4283 	vmx->loaded_vmcs->nmi_known_unmasked = false;
4284 
4285 	if (vmx->rmode.vm86_active) {
4286 		if (kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR, 0) != EMULATE_DONE)
4287 			kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4288 		return;
4289 	}
4290 
4291 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
4292 			INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
4293 
4294 	vmx_clear_hlt(vcpu);
4295 }
4296 
4297 bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
4298 {
4299 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4300 	bool masked;
4301 
4302 	if (!enable_vnmi)
4303 		return vmx->loaded_vmcs->soft_vnmi_blocked;
4304 	if (vmx->loaded_vmcs->nmi_known_unmasked)
4305 		return false;
4306 	masked = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI;
4307 	vmx->loaded_vmcs->nmi_known_unmasked = !masked;
4308 	return masked;
4309 }
4310 
4311 void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
4312 {
4313 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4314 
4315 	if (!enable_vnmi) {
4316 		if (vmx->loaded_vmcs->soft_vnmi_blocked != masked) {
4317 			vmx->loaded_vmcs->soft_vnmi_blocked = masked;
4318 			vmx->loaded_vmcs->vnmi_blocked_time = 0;
4319 		}
4320 	} else {
4321 		vmx->loaded_vmcs->nmi_known_unmasked = !masked;
4322 		if (masked)
4323 			vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
4324 				      GUEST_INTR_STATE_NMI);
4325 		else
4326 			vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
4327 					GUEST_INTR_STATE_NMI);
4328 	}
4329 }
4330 
4331 static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
4332 {
4333 	if (to_vmx(vcpu)->nested.nested_run_pending)
4334 		return 0;
4335 
4336 	if (!enable_vnmi &&
4337 	    to_vmx(vcpu)->loaded_vmcs->soft_vnmi_blocked)
4338 		return 0;
4339 
4340 	return	!(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4341 		  (GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_STI
4342 		   | GUEST_INTR_STATE_NMI));
4343 }
4344 
4345 static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
4346 {
4347 	return (!to_vmx(vcpu)->nested.nested_run_pending &&
4348 		vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
4349 		!(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4350 			(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
4351 }
4352 
4353 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
4354 {
4355 	int ret;
4356 
4357 	if (enable_unrestricted_guest)
4358 		return 0;
4359 
4360 	ret = x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, addr,
4361 				    PAGE_SIZE * 3);
4362 	if (ret)
4363 		return ret;
4364 	to_kvm_vmx(kvm)->tss_addr = addr;
4365 	return init_rmode_tss(kvm);
4366 }
4367 
4368 static int vmx_set_identity_map_addr(struct kvm *kvm, u64 ident_addr)
4369 {
4370 	to_kvm_vmx(kvm)->ept_identity_map_addr = ident_addr;
4371 	return 0;
4372 }
4373 
4374 static bool rmode_exception(struct kvm_vcpu *vcpu, int vec)
4375 {
4376 	switch (vec) {
4377 	case BP_VECTOR:
4378 		/*
4379 		 * Update instruction length as we may reinject the exception
4380 		 * from user space while in guest debugging mode.
4381 		 */
4382 		to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
4383 			vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4384 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
4385 			return false;
4386 		/* fall through */
4387 	case DB_VECTOR:
4388 		if (vcpu->guest_debug &
4389 			(KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
4390 			return false;
4391 		/* fall through */
4392 	case DE_VECTOR:
4393 	case OF_VECTOR:
4394 	case BR_VECTOR:
4395 	case UD_VECTOR:
4396 	case DF_VECTOR:
4397 	case SS_VECTOR:
4398 	case GP_VECTOR:
4399 	case MF_VECTOR:
4400 		return true;
4401 	break;
4402 	}
4403 	return false;
4404 }
4405 
4406 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
4407 				  int vec, u32 err_code)
4408 {
4409 	/*
4410 	 * Instruction with address size override prefix opcode 0x67
4411 	 * Cause the #SS fault with 0 error code in VM86 mode.
4412 	 */
4413 	if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0) {
4414 		if (kvm_emulate_instruction(vcpu, 0) == EMULATE_DONE) {
4415 			if (vcpu->arch.halt_request) {
4416 				vcpu->arch.halt_request = 0;
4417 				return kvm_vcpu_halt(vcpu);
4418 			}
4419 			return 1;
4420 		}
4421 		return 0;
4422 	}
4423 
4424 	/*
4425 	 * Forward all other exceptions that are valid in real mode.
4426 	 * FIXME: Breaks guest debugging in real mode, needs to be fixed with
4427 	 *        the required debugging infrastructure rework.
4428 	 */
4429 	kvm_queue_exception(vcpu, vec);
4430 	return 1;
4431 }
4432 
4433 /*
4434  * Trigger machine check on the host. We assume all the MSRs are already set up
4435  * by the CPU and that we still run on the same CPU as the MCE occurred on.
4436  * We pass a fake environment to the machine check handler because we want
4437  * the guest to be always treated like user space, no matter what context
4438  * it used internally.
4439  */
4440 static void kvm_machine_check(void)
4441 {
4442 #if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
4443 	struct pt_regs regs = {
4444 		.cs = 3, /* Fake ring 3 no matter what the guest ran on */
4445 		.flags = X86_EFLAGS_IF,
4446 	};
4447 
4448 	do_machine_check(&regs, 0);
4449 #endif
4450 }
4451 
4452 static int handle_machine_check(struct kvm_vcpu *vcpu)
4453 {
4454 	/* already handled by vcpu_run */
4455 	return 1;
4456 }
4457 
4458 static int handle_exception(struct kvm_vcpu *vcpu)
4459 {
4460 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4461 	struct kvm_run *kvm_run = vcpu->run;
4462 	u32 intr_info, ex_no, error_code;
4463 	unsigned long cr2, rip, dr6;
4464 	u32 vect_info;
4465 	enum emulation_result er;
4466 
4467 	vect_info = vmx->idt_vectoring_info;
4468 	intr_info = vmx->exit_intr_info;
4469 
4470 	if (is_machine_check(intr_info))
4471 		return handle_machine_check(vcpu);
4472 
4473 	if (is_nmi(intr_info))
4474 		return 1;  /* already handled by vmx_vcpu_run() */
4475 
4476 	if (is_invalid_opcode(intr_info))
4477 		return handle_ud(vcpu);
4478 
4479 	error_code = 0;
4480 	if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
4481 		error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
4482 
4483 	if (!vmx->rmode.vm86_active && is_gp_fault(intr_info)) {
4484 		WARN_ON_ONCE(!enable_vmware_backdoor);
4485 		er = kvm_emulate_instruction(vcpu,
4486 			EMULTYPE_VMWARE | EMULTYPE_NO_UD_ON_FAIL);
4487 		if (er == EMULATE_USER_EXIT)
4488 			return 0;
4489 		else if (er != EMULATE_DONE)
4490 			kvm_queue_exception_e(vcpu, GP_VECTOR, error_code);
4491 		return 1;
4492 	}
4493 
4494 	/*
4495 	 * The #PF with PFEC.RSVD = 1 indicates the guest is accessing
4496 	 * MMIO, it is better to report an internal error.
4497 	 * See the comments in vmx_handle_exit.
4498 	 */
4499 	if ((vect_info & VECTORING_INFO_VALID_MASK) &&
4500 	    !(is_page_fault(intr_info) && !(error_code & PFERR_RSVD_MASK))) {
4501 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4502 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
4503 		vcpu->run->internal.ndata = 3;
4504 		vcpu->run->internal.data[0] = vect_info;
4505 		vcpu->run->internal.data[1] = intr_info;
4506 		vcpu->run->internal.data[2] = error_code;
4507 		return 0;
4508 	}
4509 
4510 	if (is_page_fault(intr_info)) {
4511 		cr2 = vmcs_readl(EXIT_QUALIFICATION);
4512 		/* EPT won't cause page fault directly */
4513 		WARN_ON_ONCE(!vcpu->arch.apf.host_apf_reason && enable_ept);
4514 		return kvm_handle_page_fault(vcpu, error_code, cr2, NULL, 0);
4515 	}
4516 
4517 	ex_no = intr_info & INTR_INFO_VECTOR_MASK;
4518 
4519 	if (vmx->rmode.vm86_active && rmode_exception(vcpu, ex_no))
4520 		return handle_rmode_exception(vcpu, ex_no, error_code);
4521 
4522 	switch (ex_no) {
4523 	case AC_VECTOR:
4524 		kvm_queue_exception_e(vcpu, AC_VECTOR, error_code);
4525 		return 1;
4526 	case DB_VECTOR:
4527 		dr6 = vmcs_readl(EXIT_QUALIFICATION);
4528 		if (!(vcpu->guest_debug &
4529 		      (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
4530 			vcpu->arch.dr6 &= ~15;
4531 			vcpu->arch.dr6 |= dr6 | DR6_RTM;
4532 			if (is_icebp(intr_info))
4533 				skip_emulated_instruction(vcpu);
4534 
4535 			kvm_queue_exception(vcpu, DB_VECTOR);
4536 			return 1;
4537 		}
4538 		kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
4539 		kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
4540 		/* fall through */
4541 	case BP_VECTOR:
4542 		/*
4543 		 * Update instruction length as we may reinject #BP from
4544 		 * user space while in guest debugging mode. Reading it for
4545 		 * #DB as well causes no harm, it is not used in that case.
4546 		 */
4547 		vmx->vcpu.arch.event_exit_inst_len =
4548 			vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4549 		kvm_run->exit_reason = KVM_EXIT_DEBUG;
4550 		rip = kvm_rip_read(vcpu);
4551 		kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
4552 		kvm_run->debug.arch.exception = ex_no;
4553 		break;
4554 	default:
4555 		kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
4556 		kvm_run->ex.exception = ex_no;
4557 		kvm_run->ex.error_code = error_code;
4558 		break;
4559 	}
4560 	return 0;
4561 }
4562 
4563 static int handle_external_interrupt(struct kvm_vcpu *vcpu)
4564 {
4565 	++vcpu->stat.irq_exits;
4566 	return 1;
4567 }
4568 
4569 static int handle_triple_fault(struct kvm_vcpu *vcpu)
4570 {
4571 	vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
4572 	vcpu->mmio_needed = 0;
4573 	return 0;
4574 }
4575 
4576 static int handle_io(struct kvm_vcpu *vcpu)
4577 {
4578 	unsigned long exit_qualification;
4579 	int size, in, string;
4580 	unsigned port;
4581 
4582 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4583 	string = (exit_qualification & 16) != 0;
4584 
4585 	++vcpu->stat.io_exits;
4586 
4587 	if (string)
4588 		return kvm_emulate_instruction(vcpu, 0) == EMULATE_DONE;
4589 
4590 	port = exit_qualification >> 16;
4591 	size = (exit_qualification & 7) + 1;
4592 	in = (exit_qualification & 8) != 0;
4593 
4594 	return kvm_fast_pio(vcpu, size, port, in);
4595 }
4596 
4597 static void
4598 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
4599 {
4600 	/*
4601 	 * Patch in the VMCALL instruction:
4602 	 */
4603 	hypercall[0] = 0x0f;
4604 	hypercall[1] = 0x01;
4605 	hypercall[2] = 0xc1;
4606 }
4607 
4608 /* called to set cr0 as appropriate for a mov-to-cr0 exit. */
4609 static int handle_set_cr0(struct kvm_vcpu *vcpu, unsigned long val)
4610 {
4611 	if (is_guest_mode(vcpu)) {
4612 		struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4613 		unsigned long orig_val = val;
4614 
4615 		/*
4616 		 * We get here when L2 changed cr0 in a way that did not change
4617 		 * any of L1's shadowed bits (see nested_vmx_exit_handled_cr),
4618 		 * but did change L0 shadowed bits. So we first calculate the
4619 		 * effective cr0 value that L1 would like to write into the
4620 		 * hardware. It consists of the L2-owned bits from the new
4621 		 * value combined with the L1-owned bits from L1's guest_cr0.
4622 		 */
4623 		val = (val & ~vmcs12->cr0_guest_host_mask) |
4624 			(vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask);
4625 
4626 		if (!nested_guest_cr0_valid(vcpu, val))
4627 			return 1;
4628 
4629 		if (kvm_set_cr0(vcpu, val))
4630 			return 1;
4631 		vmcs_writel(CR0_READ_SHADOW, orig_val);
4632 		return 0;
4633 	} else {
4634 		if (to_vmx(vcpu)->nested.vmxon &&
4635 		    !nested_host_cr0_valid(vcpu, val))
4636 			return 1;
4637 
4638 		return kvm_set_cr0(vcpu, val);
4639 	}
4640 }
4641 
4642 static int handle_set_cr4(struct kvm_vcpu *vcpu, unsigned long val)
4643 {
4644 	if (is_guest_mode(vcpu)) {
4645 		struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4646 		unsigned long orig_val = val;
4647 
4648 		/* analogously to handle_set_cr0 */
4649 		val = (val & ~vmcs12->cr4_guest_host_mask) |
4650 			(vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask);
4651 		if (kvm_set_cr4(vcpu, val))
4652 			return 1;
4653 		vmcs_writel(CR4_READ_SHADOW, orig_val);
4654 		return 0;
4655 	} else
4656 		return kvm_set_cr4(vcpu, val);
4657 }
4658 
4659 static int handle_desc(struct kvm_vcpu *vcpu)
4660 {
4661 	WARN_ON(!(vcpu->arch.cr4 & X86_CR4_UMIP));
4662 	return kvm_emulate_instruction(vcpu, 0) == EMULATE_DONE;
4663 }
4664 
4665 static int handle_cr(struct kvm_vcpu *vcpu)
4666 {
4667 	unsigned long exit_qualification, val;
4668 	int cr;
4669 	int reg;
4670 	int err;
4671 	int ret;
4672 
4673 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4674 	cr = exit_qualification & 15;
4675 	reg = (exit_qualification >> 8) & 15;
4676 	switch ((exit_qualification >> 4) & 3) {
4677 	case 0: /* mov to cr */
4678 		val = kvm_register_readl(vcpu, reg);
4679 		trace_kvm_cr_write(cr, val);
4680 		switch (cr) {
4681 		case 0:
4682 			err = handle_set_cr0(vcpu, val);
4683 			return kvm_complete_insn_gp(vcpu, err);
4684 		case 3:
4685 			WARN_ON_ONCE(enable_unrestricted_guest);
4686 			err = kvm_set_cr3(vcpu, val);
4687 			return kvm_complete_insn_gp(vcpu, err);
4688 		case 4:
4689 			err = handle_set_cr4(vcpu, val);
4690 			return kvm_complete_insn_gp(vcpu, err);
4691 		case 8: {
4692 				u8 cr8_prev = kvm_get_cr8(vcpu);
4693 				u8 cr8 = (u8)val;
4694 				err = kvm_set_cr8(vcpu, cr8);
4695 				ret = kvm_complete_insn_gp(vcpu, err);
4696 				if (lapic_in_kernel(vcpu))
4697 					return ret;
4698 				if (cr8_prev <= cr8)
4699 					return ret;
4700 				/*
4701 				 * TODO: we might be squashing a
4702 				 * KVM_GUESTDBG_SINGLESTEP-triggered
4703 				 * KVM_EXIT_DEBUG here.
4704 				 */
4705 				vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
4706 				return 0;
4707 			}
4708 		}
4709 		break;
4710 	case 2: /* clts */
4711 		WARN_ONCE(1, "Guest should always own CR0.TS");
4712 		vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
4713 		trace_kvm_cr_write(0, kvm_read_cr0(vcpu));
4714 		return kvm_skip_emulated_instruction(vcpu);
4715 	case 1: /*mov from cr*/
4716 		switch (cr) {
4717 		case 3:
4718 			WARN_ON_ONCE(enable_unrestricted_guest);
4719 			val = kvm_read_cr3(vcpu);
4720 			kvm_register_write(vcpu, reg, val);
4721 			trace_kvm_cr_read(cr, val);
4722 			return kvm_skip_emulated_instruction(vcpu);
4723 		case 8:
4724 			val = kvm_get_cr8(vcpu);
4725 			kvm_register_write(vcpu, reg, val);
4726 			trace_kvm_cr_read(cr, val);
4727 			return kvm_skip_emulated_instruction(vcpu);
4728 		}
4729 		break;
4730 	case 3: /* lmsw */
4731 		val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
4732 		trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val);
4733 		kvm_lmsw(vcpu, val);
4734 
4735 		return kvm_skip_emulated_instruction(vcpu);
4736 	default:
4737 		break;
4738 	}
4739 	vcpu->run->exit_reason = 0;
4740 	vcpu_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
4741 	       (int)(exit_qualification >> 4) & 3, cr);
4742 	return 0;
4743 }
4744 
4745 static int handle_dr(struct kvm_vcpu *vcpu)
4746 {
4747 	unsigned long exit_qualification;
4748 	int dr, dr7, reg;
4749 
4750 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4751 	dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
4752 
4753 	/* First, if DR does not exist, trigger UD */
4754 	if (!kvm_require_dr(vcpu, dr))
4755 		return 1;
4756 
4757 	/* Do not handle if the CPL > 0, will trigger GP on re-entry */
4758 	if (!kvm_require_cpl(vcpu, 0))
4759 		return 1;
4760 	dr7 = vmcs_readl(GUEST_DR7);
4761 	if (dr7 & DR7_GD) {
4762 		/*
4763 		 * As the vm-exit takes precedence over the debug trap, we
4764 		 * need to emulate the latter, either for the host or the
4765 		 * guest debugging itself.
4766 		 */
4767 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
4768 			vcpu->run->debug.arch.dr6 = vcpu->arch.dr6;
4769 			vcpu->run->debug.arch.dr7 = dr7;
4770 			vcpu->run->debug.arch.pc = kvm_get_linear_rip(vcpu);
4771 			vcpu->run->debug.arch.exception = DB_VECTOR;
4772 			vcpu->run->exit_reason = KVM_EXIT_DEBUG;
4773 			return 0;
4774 		} else {
4775 			vcpu->arch.dr6 &= ~15;
4776 			vcpu->arch.dr6 |= DR6_BD | DR6_RTM;
4777 			kvm_queue_exception(vcpu, DB_VECTOR);
4778 			return 1;
4779 		}
4780 	}
4781 
4782 	if (vcpu->guest_debug == 0) {
4783 		vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL,
4784 				CPU_BASED_MOV_DR_EXITING);
4785 
4786 		/*
4787 		 * No more DR vmexits; force a reload of the debug registers
4788 		 * and reenter on this instruction.  The next vmexit will
4789 		 * retrieve the full state of the debug registers.
4790 		 */
4791 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
4792 		return 1;
4793 	}
4794 
4795 	reg = DEBUG_REG_ACCESS_REG(exit_qualification);
4796 	if (exit_qualification & TYPE_MOV_FROM_DR) {
4797 		unsigned long val;
4798 
4799 		if (kvm_get_dr(vcpu, dr, &val))
4800 			return 1;
4801 		kvm_register_write(vcpu, reg, val);
4802 	} else
4803 		if (kvm_set_dr(vcpu, dr, kvm_register_readl(vcpu, reg)))
4804 			return 1;
4805 
4806 	return kvm_skip_emulated_instruction(vcpu);
4807 }
4808 
4809 static u64 vmx_get_dr6(struct kvm_vcpu *vcpu)
4810 {
4811 	return vcpu->arch.dr6;
4812 }
4813 
4814 static void vmx_set_dr6(struct kvm_vcpu *vcpu, unsigned long val)
4815 {
4816 }
4817 
4818 static void vmx_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
4819 {
4820 	get_debugreg(vcpu->arch.db[0], 0);
4821 	get_debugreg(vcpu->arch.db[1], 1);
4822 	get_debugreg(vcpu->arch.db[2], 2);
4823 	get_debugreg(vcpu->arch.db[3], 3);
4824 	get_debugreg(vcpu->arch.dr6, 6);
4825 	vcpu->arch.dr7 = vmcs_readl(GUEST_DR7);
4826 
4827 	vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
4828 	vmcs_set_bits(CPU_BASED_VM_EXEC_CONTROL, CPU_BASED_MOV_DR_EXITING);
4829 }
4830 
4831 static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
4832 {
4833 	vmcs_writel(GUEST_DR7, val);
4834 }
4835 
4836 static int handle_cpuid(struct kvm_vcpu *vcpu)
4837 {
4838 	return kvm_emulate_cpuid(vcpu);
4839 }
4840 
4841 static int handle_rdmsr(struct kvm_vcpu *vcpu)
4842 {
4843 	u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
4844 	struct msr_data msr_info;
4845 
4846 	msr_info.index = ecx;
4847 	msr_info.host_initiated = false;
4848 	if (vmx_get_msr(vcpu, &msr_info)) {
4849 		trace_kvm_msr_read_ex(ecx);
4850 		kvm_inject_gp(vcpu, 0);
4851 		return 1;
4852 	}
4853 
4854 	trace_kvm_msr_read(ecx, msr_info.data);
4855 
4856 	/* FIXME: handling of bits 32:63 of rax, rdx */
4857 	vcpu->arch.regs[VCPU_REGS_RAX] = msr_info.data & -1u;
4858 	vcpu->arch.regs[VCPU_REGS_RDX] = (msr_info.data >> 32) & -1u;
4859 	return kvm_skip_emulated_instruction(vcpu);
4860 }
4861 
4862 static int handle_wrmsr(struct kvm_vcpu *vcpu)
4863 {
4864 	struct msr_data msr;
4865 	u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
4866 	u64 data = (vcpu->arch.regs[VCPU_REGS_RAX] & -1u)
4867 		| ((u64)(vcpu->arch.regs[VCPU_REGS_RDX] & -1u) << 32);
4868 
4869 	msr.data = data;
4870 	msr.index = ecx;
4871 	msr.host_initiated = false;
4872 	if (kvm_set_msr(vcpu, &msr) != 0) {
4873 		trace_kvm_msr_write_ex(ecx, data);
4874 		kvm_inject_gp(vcpu, 0);
4875 		return 1;
4876 	}
4877 
4878 	trace_kvm_msr_write(ecx, data);
4879 	return kvm_skip_emulated_instruction(vcpu);
4880 }
4881 
4882 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
4883 {
4884 	kvm_apic_update_ppr(vcpu);
4885 	return 1;
4886 }
4887 
4888 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
4889 {
4890 	vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL,
4891 			CPU_BASED_VIRTUAL_INTR_PENDING);
4892 
4893 	kvm_make_request(KVM_REQ_EVENT, vcpu);
4894 
4895 	++vcpu->stat.irq_window_exits;
4896 	return 1;
4897 }
4898 
4899 static int handle_halt(struct kvm_vcpu *vcpu)
4900 {
4901 	return kvm_emulate_halt(vcpu);
4902 }
4903 
4904 static int handle_vmcall(struct kvm_vcpu *vcpu)
4905 {
4906 	return kvm_emulate_hypercall(vcpu);
4907 }
4908 
4909 static int handle_invd(struct kvm_vcpu *vcpu)
4910 {
4911 	return kvm_emulate_instruction(vcpu, 0) == EMULATE_DONE;
4912 }
4913 
4914 static int handle_invlpg(struct kvm_vcpu *vcpu)
4915 {
4916 	unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4917 
4918 	kvm_mmu_invlpg(vcpu, exit_qualification);
4919 	return kvm_skip_emulated_instruction(vcpu);
4920 }
4921 
4922 static int handle_rdpmc(struct kvm_vcpu *vcpu)
4923 {
4924 	int err;
4925 
4926 	err = kvm_rdpmc(vcpu);
4927 	return kvm_complete_insn_gp(vcpu, err);
4928 }
4929 
4930 static int handle_wbinvd(struct kvm_vcpu *vcpu)
4931 {
4932 	return kvm_emulate_wbinvd(vcpu);
4933 }
4934 
4935 static int handle_xsetbv(struct kvm_vcpu *vcpu)
4936 {
4937 	u64 new_bv = kvm_read_edx_eax(vcpu);
4938 	u32 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
4939 
4940 	if (kvm_set_xcr(vcpu, index, new_bv) == 0)
4941 		return kvm_skip_emulated_instruction(vcpu);
4942 	return 1;
4943 }
4944 
4945 static int handle_xsaves(struct kvm_vcpu *vcpu)
4946 {
4947 	kvm_skip_emulated_instruction(vcpu);
4948 	WARN(1, "this should never happen\n");
4949 	return 1;
4950 }
4951 
4952 static int handle_xrstors(struct kvm_vcpu *vcpu)
4953 {
4954 	kvm_skip_emulated_instruction(vcpu);
4955 	WARN(1, "this should never happen\n");
4956 	return 1;
4957 }
4958 
4959 static int handle_apic_access(struct kvm_vcpu *vcpu)
4960 {
4961 	if (likely(fasteoi)) {
4962 		unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4963 		int access_type, offset;
4964 
4965 		access_type = exit_qualification & APIC_ACCESS_TYPE;
4966 		offset = exit_qualification & APIC_ACCESS_OFFSET;
4967 		/*
4968 		 * Sane guest uses MOV to write EOI, with written value
4969 		 * not cared. So make a short-circuit here by avoiding
4970 		 * heavy instruction emulation.
4971 		 */
4972 		if ((access_type == TYPE_LINEAR_APIC_INST_WRITE) &&
4973 		    (offset == APIC_EOI)) {
4974 			kvm_lapic_set_eoi(vcpu);
4975 			return kvm_skip_emulated_instruction(vcpu);
4976 		}
4977 	}
4978 	return kvm_emulate_instruction(vcpu, 0) == EMULATE_DONE;
4979 }
4980 
4981 static int handle_apic_eoi_induced(struct kvm_vcpu *vcpu)
4982 {
4983 	unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4984 	int vector = exit_qualification & 0xff;
4985 
4986 	/* EOI-induced VM exit is trap-like and thus no need to adjust IP */
4987 	kvm_apic_set_eoi_accelerated(vcpu, vector);
4988 	return 1;
4989 }
4990 
4991 static int handle_apic_write(struct kvm_vcpu *vcpu)
4992 {
4993 	unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4994 	u32 offset = exit_qualification & 0xfff;
4995 
4996 	/* APIC-write VM exit is trap-like and thus no need to adjust IP */
4997 	kvm_apic_write_nodecode(vcpu, offset);
4998 	return 1;
4999 }
5000 
5001 static int handle_task_switch(struct kvm_vcpu *vcpu)
5002 {
5003 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5004 	unsigned long exit_qualification;
5005 	bool has_error_code = false;
5006 	u32 error_code = 0;
5007 	u16 tss_selector;
5008 	int reason, type, idt_v, idt_index;
5009 
5010 	idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
5011 	idt_index = (vmx->idt_vectoring_info & VECTORING_INFO_VECTOR_MASK);
5012 	type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
5013 
5014 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5015 
5016 	reason = (u32)exit_qualification >> 30;
5017 	if (reason == TASK_SWITCH_GATE && idt_v) {
5018 		switch (type) {
5019 		case INTR_TYPE_NMI_INTR:
5020 			vcpu->arch.nmi_injected = false;
5021 			vmx_set_nmi_mask(vcpu, true);
5022 			break;
5023 		case INTR_TYPE_EXT_INTR:
5024 		case INTR_TYPE_SOFT_INTR:
5025 			kvm_clear_interrupt_queue(vcpu);
5026 			break;
5027 		case INTR_TYPE_HARD_EXCEPTION:
5028 			if (vmx->idt_vectoring_info &
5029 			    VECTORING_INFO_DELIVER_CODE_MASK) {
5030 				has_error_code = true;
5031 				error_code =
5032 					vmcs_read32(IDT_VECTORING_ERROR_CODE);
5033 			}
5034 			/* fall through */
5035 		case INTR_TYPE_SOFT_EXCEPTION:
5036 			kvm_clear_exception_queue(vcpu);
5037 			break;
5038 		default:
5039 			break;
5040 		}
5041 	}
5042 	tss_selector = exit_qualification;
5043 
5044 	if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
5045 		       type != INTR_TYPE_EXT_INTR &&
5046 		       type != INTR_TYPE_NMI_INTR))
5047 		skip_emulated_instruction(vcpu);
5048 
5049 	if (kvm_task_switch(vcpu, tss_selector,
5050 			    type == INTR_TYPE_SOFT_INTR ? idt_index : -1, reason,
5051 			    has_error_code, error_code) == EMULATE_FAIL) {
5052 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5053 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5054 		vcpu->run->internal.ndata = 0;
5055 		return 0;
5056 	}
5057 
5058 	/*
5059 	 * TODO: What about debug traps on tss switch?
5060 	 *       Are we supposed to inject them and update dr6?
5061 	 */
5062 
5063 	return 1;
5064 }
5065 
5066 static int handle_ept_violation(struct kvm_vcpu *vcpu)
5067 {
5068 	unsigned long exit_qualification;
5069 	gpa_t gpa;
5070 	u64 error_code;
5071 
5072 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5073 
5074 	/*
5075 	 * EPT violation happened while executing iret from NMI,
5076 	 * "blocked by NMI" bit has to be set before next VM entry.
5077 	 * There are errata that may cause this bit to not be set:
5078 	 * AAK134, BY25.
5079 	 */
5080 	if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
5081 			enable_vnmi &&
5082 			(exit_qualification & INTR_INFO_UNBLOCK_NMI))
5083 		vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO, GUEST_INTR_STATE_NMI);
5084 
5085 	gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5086 	trace_kvm_page_fault(gpa, exit_qualification);
5087 
5088 	/* Is it a read fault? */
5089 	error_code = (exit_qualification & EPT_VIOLATION_ACC_READ)
5090 		     ? PFERR_USER_MASK : 0;
5091 	/* Is it a write fault? */
5092 	error_code |= (exit_qualification & EPT_VIOLATION_ACC_WRITE)
5093 		      ? PFERR_WRITE_MASK : 0;
5094 	/* Is it a fetch fault? */
5095 	error_code |= (exit_qualification & EPT_VIOLATION_ACC_INSTR)
5096 		      ? PFERR_FETCH_MASK : 0;
5097 	/* ept page table entry is present? */
5098 	error_code |= (exit_qualification &
5099 		       (EPT_VIOLATION_READABLE | EPT_VIOLATION_WRITABLE |
5100 			EPT_VIOLATION_EXECUTABLE))
5101 		      ? PFERR_PRESENT_MASK : 0;
5102 
5103 	error_code |= (exit_qualification & 0x100) != 0 ?
5104 	       PFERR_GUEST_FINAL_MASK : PFERR_GUEST_PAGE_MASK;
5105 
5106 	vcpu->arch.exit_qualification = exit_qualification;
5107 	return kvm_mmu_page_fault(vcpu, gpa, error_code, NULL, 0);
5108 }
5109 
5110 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
5111 {
5112 	gpa_t gpa;
5113 
5114 	/*
5115 	 * A nested guest cannot optimize MMIO vmexits, because we have an
5116 	 * nGPA here instead of the required GPA.
5117 	 */
5118 	gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5119 	if (!is_guest_mode(vcpu) &&
5120 	    !kvm_io_bus_write(vcpu, KVM_FAST_MMIO_BUS, gpa, 0, NULL)) {
5121 		trace_kvm_fast_mmio(gpa);
5122 		/*
5123 		 * Doing kvm_skip_emulated_instruction() depends on undefined
5124 		 * behavior: Intel's manual doesn't mandate
5125 		 * VM_EXIT_INSTRUCTION_LEN to be set in VMCS when EPT MISCONFIG
5126 		 * occurs and while on real hardware it was observed to be set,
5127 		 * other hypervisors (namely Hyper-V) don't set it, we end up
5128 		 * advancing IP with some random value. Disable fast mmio when
5129 		 * running nested and keep it for real hardware in hope that
5130 		 * VM_EXIT_INSTRUCTION_LEN will always be set correctly.
5131 		 */
5132 		if (!static_cpu_has(X86_FEATURE_HYPERVISOR))
5133 			return kvm_skip_emulated_instruction(vcpu);
5134 		else
5135 			return kvm_emulate_instruction(vcpu, EMULTYPE_SKIP) ==
5136 								EMULATE_DONE;
5137 	}
5138 
5139 	return kvm_mmu_page_fault(vcpu, gpa, PFERR_RSVD_MASK, NULL, 0);
5140 }
5141 
5142 static int handle_nmi_window(struct kvm_vcpu *vcpu)
5143 {
5144 	WARN_ON_ONCE(!enable_vnmi);
5145 	vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL,
5146 			CPU_BASED_VIRTUAL_NMI_PENDING);
5147 	++vcpu->stat.nmi_window_exits;
5148 	kvm_make_request(KVM_REQ_EVENT, vcpu);
5149 
5150 	return 1;
5151 }
5152 
5153 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
5154 {
5155 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5156 	enum emulation_result err = EMULATE_DONE;
5157 	int ret = 1;
5158 	u32 cpu_exec_ctrl;
5159 	bool intr_window_requested;
5160 	unsigned count = 130;
5161 
5162 	/*
5163 	 * We should never reach the point where we are emulating L2
5164 	 * due to invalid guest state as that means we incorrectly
5165 	 * allowed a nested VMEntry with an invalid vmcs12.
5166 	 */
5167 	WARN_ON_ONCE(vmx->emulation_required && vmx->nested.nested_run_pending);
5168 
5169 	cpu_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5170 	intr_window_requested = cpu_exec_ctrl & CPU_BASED_VIRTUAL_INTR_PENDING;
5171 
5172 	while (vmx->emulation_required && count-- != 0) {
5173 		if (intr_window_requested && vmx_interrupt_allowed(vcpu))
5174 			return handle_interrupt_window(&vmx->vcpu);
5175 
5176 		if (kvm_test_request(KVM_REQ_EVENT, vcpu))
5177 			return 1;
5178 
5179 		err = kvm_emulate_instruction(vcpu, 0);
5180 
5181 		if (err == EMULATE_USER_EXIT) {
5182 			++vcpu->stat.mmio_exits;
5183 			ret = 0;
5184 			goto out;
5185 		}
5186 
5187 		if (err != EMULATE_DONE)
5188 			goto emulation_error;
5189 
5190 		if (vmx->emulation_required && !vmx->rmode.vm86_active &&
5191 		    vcpu->arch.exception.pending)
5192 			goto emulation_error;
5193 
5194 		if (vcpu->arch.halt_request) {
5195 			vcpu->arch.halt_request = 0;
5196 			ret = kvm_vcpu_halt(vcpu);
5197 			goto out;
5198 		}
5199 
5200 		if (signal_pending(current))
5201 			goto out;
5202 		if (need_resched())
5203 			schedule();
5204 	}
5205 
5206 out:
5207 	return ret;
5208 
5209 emulation_error:
5210 	vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5211 	vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5212 	vcpu->run->internal.ndata = 0;
5213 	return 0;
5214 }
5215 
5216 static void grow_ple_window(struct kvm_vcpu *vcpu)
5217 {
5218 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5219 	int old = vmx->ple_window;
5220 
5221 	vmx->ple_window = __grow_ple_window(old, ple_window,
5222 					    ple_window_grow,
5223 					    ple_window_max);
5224 
5225 	if (vmx->ple_window != old)
5226 		vmx->ple_window_dirty = true;
5227 
5228 	trace_kvm_ple_window_grow(vcpu->vcpu_id, vmx->ple_window, old);
5229 }
5230 
5231 static void shrink_ple_window(struct kvm_vcpu *vcpu)
5232 {
5233 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5234 	int old = vmx->ple_window;
5235 
5236 	vmx->ple_window = __shrink_ple_window(old, ple_window,
5237 					      ple_window_shrink,
5238 					      ple_window);
5239 
5240 	if (vmx->ple_window != old)
5241 		vmx->ple_window_dirty = true;
5242 
5243 	trace_kvm_ple_window_shrink(vcpu->vcpu_id, vmx->ple_window, old);
5244 }
5245 
5246 /*
5247  * Handler for POSTED_INTERRUPT_WAKEUP_VECTOR.
5248  */
5249 static void wakeup_handler(void)
5250 {
5251 	struct kvm_vcpu *vcpu;
5252 	int cpu = smp_processor_id();
5253 
5254 	spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
5255 	list_for_each_entry(vcpu, &per_cpu(blocked_vcpu_on_cpu, cpu),
5256 			blocked_vcpu_list) {
5257 		struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
5258 
5259 		if (pi_test_on(pi_desc) == 1)
5260 			kvm_vcpu_kick(vcpu);
5261 	}
5262 	spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
5263 }
5264 
5265 static void vmx_enable_tdp(void)
5266 {
5267 	kvm_mmu_set_mask_ptes(VMX_EPT_READABLE_MASK,
5268 		enable_ept_ad_bits ? VMX_EPT_ACCESS_BIT : 0ull,
5269 		enable_ept_ad_bits ? VMX_EPT_DIRTY_BIT : 0ull,
5270 		0ull, VMX_EPT_EXECUTABLE_MASK,
5271 		cpu_has_vmx_ept_execute_only() ? 0ull : VMX_EPT_READABLE_MASK,
5272 		VMX_EPT_RWX_MASK, 0ull);
5273 
5274 	ept_set_mmio_spte_mask();
5275 	kvm_enable_tdp();
5276 }
5277 
5278 /*
5279  * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
5280  * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
5281  */
5282 static int handle_pause(struct kvm_vcpu *vcpu)
5283 {
5284 	if (!kvm_pause_in_guest(vcpu->kvm))
5285 		grow_ple_window(vcpu);
5286 
5287 	/*
5288 	 * Intel sdm vol3 ch-25.1.3 says: The "PAUSE-loop exiting"
5289 	 * VM-execution control is ignored if CPL > 0. OTOH, KVM
5290 	 * never set PAUSE_EXITING and just set PLE if supported,
5291 	 * so the vcpu must be CPL=0 if it gets a PAUSE exit.
5292 	 */
5293 	kvm_vcpu_on_spin(vcpu, true);
5294 	return kvm_skip_emulated_instruction(vcpu);
5295 }
5296 
5297 static int handle_nop(struct kvm_vcpu *vcpu)
5298 {
5299 	return kvm_skip_emulated_instruction(vcpu);
5300 }
5301 
5302 static int handle_mwait(struct kvm_vcpu *vcpu)
5303 {
5304 	printk_once(KERN_WARNING "kvm: MWAIT instruction emulated as NOP!\n");
5305 	return handle_nop(vcpu);
5306 }
5307 
5308 static int handle_invalid_op(struct kvm_vcpu *vcpu)
5309 {
5310 	kvm_queue_exception(vcpu, UD_VECTOR);
5311 	return 1;
5312 }
5313 
5314 static int handle_monitor_trap(struct kvm_vcpu *vcpu)
5315 {
5316 	return 1;
5317 }
5318 
5319 static int handle_monitor(struct kvm_vcpu *vcpu)
5320 {
5321 	printk_once(KERN_WARNING "kvm: MONITOR instruction emulated as NOP!\n");
5322 	return handle_nop(vcpu);
5323 }
5324 
5325 static int handle_invpcid(struct kvm_vcpu *vcpu)
5326 {
5327 	u32 vmx_instruction_info;
5328 	unsigned long type;
5329 	bool pcid_enabled;
5330 	gva_t gva;
5331 	struct x86_exception e;
5332 	unsigned i;
5333 	unsigned long roots_to_free = 0;
5334 	struct {
5335 		u64 pcid;
5336 		u64 gla;
5337 	} operand;
5338 
5339 	if (!guest_cpuid_has(vcpu, X86_FEATURE_INVPCID)) {
5340 		kvm_queue_exception(vcpu, UD_VECTOR);
5341 		return 1;
5342 	}
5343 
5344 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5345 	type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
5346 
5347 	if (type > 3) {
5348 		kvm_inject_gp(vcpu, 0);
5349 		return 1;
5350 	}
5351 
5352 	/* According to the Intel instruction reference, the memory operand
5353 	 * is read even if it isn't needed (e.g., for type==all)
5354 	 */
5355 	if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
5356 				vmx_instruction_info, false, &gva))
5357 		return 1;
5358 
5359 	if (kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e)) {
5360 		kvm_inject_page_fault(vcpu, &e);
5361 		return 1;
5362 	}
5363 
5364 	if (operand.pcid >> 12 != 0) {
5365 		kvm_inject_gp(vcpu, 0);
5366 		return 1;
5367 	}
5368 
5369 	pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
5370 
5371 	switch (type) {
5372 	case INVPCID_TYPE_INDIV_ADDR:
5373 		if ((!pcid_enabled && (operand.pcid != 0)) ||
5374 		    is_noncanonical_address(operand.gla, vcpu)) {
5375 			kvm_inject_gp(vcpu, 0);
5376 			return 1;
5377 		}
5378 		kvm_mmu_invpcid_gva(vcpu, operand.gla, operand.pcid);
5379 		return kvm_skip_emulated_instruction(vcpu);
5380 
5381 	case INVPCID_TYPE_SINGLE_CTXT:
5382 		if (!pcid_enabled && (operand.pcid != 0)) {
5383 			kvm_inject_gp(vcpu, 0);
5384 			return 1;
5385 		}
5386 
5387 		if (kvm_get_active_pcid(vcpu) == operand.pcid) {
5388 			kvm_mmu_sync_roots(vcpu);
5389 			kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
5390 		}
5391 
5392 		for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++)
5393 			if (kvm_get_pcid(vcpu, vcpu->arch.mmu->prev_roots[i].cr3)
5394 			    == operand.pcid)
5395 				roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i);
5396 
5397 		kvm_mmu_free_roots(vcpu, vcpu->arch.mmu, roots_to_free);
5398 		/*
5399 		 * If neither the current cr3 nor any of the prev_roots use the
5400 		 * given PCID, then nothing needs to be done here because a
5401 		 * resync will happen anyway before switching to any other CR3.
5402 		 */
5403 
5404 		return kvm_skip_emulated_instruction(vcpu);
5405 
5406 	case INVPCID_TYPE_ALL_NON_GLOBAL:
5407 		/*
5408 		 * Currently, KVM doesn't mark global entries in the shadow
5409 		 * page tables, so a non-global flush just degenerates to a
5410 		 * global flush. If needed, we could optimize this later by
5411 		 * keeping track of global entries in shadow page tables.
5412 		 */
5413 
5414 		/* fall-through */
5415 	case INVPCID_TYPE_ALL_INCL_GLOBAL:
5416 		kvm_mmu_unload(vcpu);
5417 		return kvm_skip_emulated_instruction(vcpu);
5418 
5419 	default:
5420 		BUG(); /* We have already checked above that type <= 3 */
5421 	}
5422 }
5423 
5424 static int handle_pml_full(struct kvm_vcpu *vcpu)
5425 {
5426 	unsigned long exit_qualification;
5427 
5428 	trace_kvm_pml_full(vcpu->vcpu_id);
5429 
5430 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5431 
5432 	/*
5433 	 * PML buffer FULL happened while executing iret from NMI,
5434 	 * "blocked by NMI" bit has to be set before next VM entry.
5435 	 */
5436 	if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
5437 			enable_vnmi &&
5438 			(exit_qualification & INTR_INFO_UNBLOCK_NMI))
5439 		vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
5440 				GUEST_INTR_STATE_NMI);
5441 
5442 	/*
5443 	 * PML buffer already flushed at beginning of VMEXIT. Nothing to do
5444 	 * here.., and there's no userspace involvement needed for PML.
5445 	 */
5446 	return 1;
5447 }
5448 
5449 static int handle_preemption_timer(struct kvm_vcpu *vcpu)
5450 {
5451 	if (!to_vmx(vcpu)->req_immediate_exit)
5452 		kvm_lapic_expired_hv_timer(vcpu);
5453 	return 1;
5454 }
5455 
5456 /*
5457  * When nested=0, all VMX instruction VM Exits filter here.  The handlers
5458  * are overwritten by nested_vmx_setup() when nested=1.
5459  */
5460 static int handle_vmx_instruction(struct kvm_vcpu *vcpu)
5461 {
5462 	kvm_queue_exception(vcpu, UD_VECTOR);
5463 	return 1;
5464 }
5465 
5466 static int handle_encls(struct kvm_vcpu *vcpu)
5467 {
5468 	/*
5469 	 * SGX virtualization is not yet supported.  There is no software
5470 	 * enable bit for SGX, so we have to trap ENCLS and inject a #UD
5471 	 * to prevent the guest from executing ENCLS.
5472 	 */
5473 	kvm_queue_exception(vcpu, UD_VECTOR);
5474 	return 1;
5475 }
5476 
5477 /*
5478  * The exit handlers return 1 if the exit was handled fully and guest execution
5479  * may resume.  Otherwise they set the kvm_run parameter to indicate what needs
5480  * to be done to userspace and return 0.
5481  */
5482 static int (*kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
5483 	[EXIT_REASON_EXCEPTION_NMI]           = handle_exception,
5484 	[EXIT_REASON_EXTERNAL_INTERRUPT]      = handle_external_interrupt,
5485 	[EXIT_REASON_TRIPLE_FAULT]            = handle_triple_fault,
5486 	[EXIT_REASON_NMI_WINDOW]	      = handle_nmi_window,
5487 	[EXIT_REASON_IO_INSTRUCTION]          = handle_io,
5488 	[EXIT_REASON_CR_ACCESS]               = handle_cr,
5489 	[EXIT_REASON_DR_ACCESS]               = handle_dr,
5490 	[EXIT_REASON_CPUID]                   = handle_cpuid,
5491 	[EXIT_REASON_MSR_READ]                = handle_rdmsr,
5492 	[EXIT_REASON_MSR_WRITE]               = handle_wrmsr,
5493 	[EXIT_REASON_PENDING_INTERRUPT]       = handle_interrupt_window,
5494 	[EXIT_REASON_HLT]                     = handle_halt,
5495 	[EXIT_REASON_INVD]		      = handle_invd,
5496 	[EXIT_REASON_INVLPG]		      = handle_invlpg,
5497 	[EXIT_REASON_RDPMC]                   = handle_rdpmc,
5498 	[EXIT_REASON_VMCALL]                  = handle_vmcall,
5499 	[EXIT_REASON_VMCLEAR]		      = handle_vmx_instruction,
5500 	[EXIT_REASON_VMLAUNCH]		      = handle_vmx_instruction,
5501 	[EXIT_REASON_VMPTRLD]		      = handle_vmx_instruction,
5502 	[EXIT_REASON_VMPTRST]		      = handle_vmx_instruction,
5503 	[EXIT_REASON_VMREAD]		      = handle_vmx_instruction,
5504 	[EXIT_REASON_VMRESUME]		      = handle_vmx_instruction,
5505 	[EXIT_REASON_VMWRITE]		      = handle_vmx_instruction,
5506 	[EXIT_REASON_VMOFF]		      = handle_vmx_instruction,
5507 	[EXIT_REASON_VMON]		      = handle_vmx_instruction,
5508 	[EXIT_REASON_TPR_BELOW_THRESHOLD]     = handle_tpr_below_threshold,
5509 	[EXIT_REASON_APIC_ACCESS]             = handle_apic_access,
5510 	[EXIT_REASON_APIC_WRITE]              = handle_apic_write,
5511 	[EXIT_REASON_EOI_INDUCED]             = handle_apic_eoi_induced,
5512 	[EXIT_REASON_WBINVD]                  = handle_wbinvd,
5513 	[EXIT_REASON_XSETBV]                  = handle_xsetbv,
5514 	[EXIT_REASON_TASK_SWITCH]             = handle_task_switch,
5515 	[EXIT_REASON_MCE_DURING_VMENTRY]      = handle_machine_check,
5516 	[EXIT_REASON_GDTR_IDTR]		      = handle_desc,
5517 	[EXIT_REASON_LDTR_TR]		      = handle_desc,
5518 	[EXIT_REASON_EPT_VIOLATION]	      = handle_ept_violation,
5519 	[EXIT_REASON_EPT_MISCONFIG]           = handle_ept_misconfig,
5520 	[EXIT_REASON_PAUSE_INSTRUCTION]       = handle_pause,
5521 	[EXIT_REASON_MWAIT_INSTRUCTION]	      = handle_mwait,
5522 	[EXIT_REASON_MONITOR_TRAP_FLAG]       = handle_monitor_trap,
5523 	[EXIT_REASON_MONITOR_INSTRUCTION]     = handle_monitor,
5524 	[EXIT_REASON_INVEPT]                  = handle_vmx_instruction,
5525 	[EXIT_REASON_INVVPID]                 = handle_vmx_instruction,
5526 	[EXIT_REASON_RDRAND]                  = handle_invalid_op,
5527 	[EXIT_REASON_RDSEED]                  = handle_invalid_op,
5528 	[EXIT_REASON_XSAVES]                  = handle_xsaves,
5529 	[EXIT_REASON_XRSTORS]                 = handle_xrstors,
5530 	[EXIT_REASON_PML_FULL]		      = handle_pml_full,
5531 	[EXIT_REASON_INVPCID]                 = handle_invpcid,
5532 	[EXIT_REASON_VMFUNC]		      = handle_vmx_instruction,
5533 	[EXIT_REASON_PREEMPTION_TIMER]	      = handle_preemption_timer,
5534 	[EXIT_REASON_ENCLS]		      = handle_encls,
5535 };
5536 
5537 static const int kvm_vmx_max_exit_handlers =
5538 	ARRAY_SIZE(kvm_vmx_exit_handlers);
5539 
5540 static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
5541 {
5542 	*info1 = vmcs_readl(EXIT_QUALIFICATION);
5543 	*info2 = vmcs_read32(VM_EXIT_INTR_INFO);
5544 }
5545 
5546 static void vmx_destroy_pml_buffer(struct vcpu_vmx *vmx)
5547 {
5548 	if (vmx->pml_pg) {
5549 		__free_page(vmx->pml_pg);
5550 		vmx->pml_pg = NULL;
5551 	}
5552 }
5553 
5554 static void vmx_flush_pml_buffer(struct kvm_vcpu *vcpu)
5555 {
5556 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5557 	u64 *pml_buf;
5558 	u16 pml_idx;
5559 
5560 	pml_idx = vmcs_read16(GUEST_PML_INDEX);
5561 
5562 	/* Do nothing if PML buffer is empty */
5563 	if (pml_idx == (PML_ENTITY_NUM - 1))
5564 		return;
5565 
5566 	/* PML index always points to next available PML buffer entity */
5567 	if (pml_idx >= PML_ENTITY_NUM)
5568 		pml_idx = 0;
5569 	else
5570 		pml_idx++;
5571 
5572 	pml_buf = page_address(vmx->pml_pg);
5573 	for (; pml_idx < PML_ENTITY_NUM; pml_idx++) {
5574 		u64 gpa;
5575 
5576 		gpa = pml_buf[pml_idx];
5577 		WARN_ON(gpa & (PAGE_SIZE - 1));
5578 		kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
5579 	}
5580 
5581 	/* reset PML index */
5582 	vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
5583 }
5584 
5585 /*
5586  * Flush all vcpus' PML buffer and update logged GPAs to dirty_bitmap.
5587  * Called before reporting dirty_bitmap to userspace.
5588  */
5589 static void kvm_flush_pml_buffers(struct kvm *kvm)
5590 {
5591 	int i;
5592 	struct kvm_vcpu *vcpu;
5593 	/*
5594 	 * We only need to kick vcpu out of guest mode here, as PML buffer
5595 	 * is flushed at beginning of all VMEXITs, and it's obvious that only
5596 	 * vcpus running in guest are possible to have unflushed GPAs in PML
5597 	 * buffer.
5598 	 */
5599 	kvm_for_each_vcpu(i, vcpu, kvm)
5600 		kvm_vcpu_kick(vcpu);
5601 }
5602 
5603 static void vmx_dump_sel(char *name, uint32_t sel)
5604 {
5605 	pr_err("%s sel=0x%04x, attr=0x%05x, limit=0x%08x, base=0x%016lx\n",
5606 	       name, vmcs_read16(sel),
5607 	       vmcs_read32(sel + GUEST_ES_AR_BYTES - GUEST_ES_SELECTOR),
5608 	       vmcs_read32(sel + GUEST_ES_LIMIT - GUEST_ES_SELECTOR),
5609 	       vmcs_readl(sel + GUEST_ES_BASE - GUEST_ES_SELECTOR));
5610 }
5611 
5612 static void vmx_dump_dtsel(char *name, uint32_t limit)
5613 {
5614 	pr_err("%s                           limit=0x%08x, base=0x%016lx\n",
5615 	       name, vmcs_read32(limit),
5616 	       vmcs_readl(limit + GUEST_GDTR_BASE - GUEST_GDTR_LIMIT));
5617 }
5618 
5619 static void dump_vmcs(void)
5620 {
5621 	u32 vmentry_ctl = vmcs_read32(VM_ENTRY_CONTROLS);
5622 	u32 vmexit_ctl = vmcs_read32(VM_EXIT_CONTROLS);
5623 	u32 cpu_based_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5624 	u32 pin_based_exec_ctrl = vmcs_read32(PIN_BASED_VM_EXEC_CONTROL);
5625 	u32 secondary_exec_control = 0;
5626 	unsigned long cr4 = vmcs_readl(GUEST_CR4);
5627 	u64 efer = vmcs_read64(GUEST_IA32_EFER);
5628 	int i, n;
5629 
5630 	if (cpu_has_secondary_exec_ctrls())
5631 		secondary_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
5632 
5633 	pr_err("*** Guest State ***\n");
5634 	pr_err("CR0: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
5635 	       vmcs_readl(GUEST_CR0), vmcs_readl(CR0_READ_SHADOW),
5636 	       vmcs_readl(CR0_GUEST_HOST_MASK));
5637 	pr_err("CR4: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
5638 	       cr4, vmcs_readl(CR4_READ_SHADOW), vmcs_readl(CR4_GUEST_HOST_MASK));
5639 	pr_err("CR3 = 0x%016lx\n", vmcs_readl(GUEST_CR3));
5640 	if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT) &&
5641 	    (cr4 & X86_CR4_PAE) && !(efer & EFER_LMA))
5642 	{
5643 		pr_err("PDPTR0 = 0x%016llx  PDPTR1 = 0x%016llx\n",
5644 		       vmcs_read64(GUEST_PDPTR0), vmcs_read64(GUEST_PDPTR1));
5645 		pr_err("PDPTR2 = 0x%016llx  PDPTR3 = 0x%016llx\n",
5646 		       vmcs_read64(GUEST_PDPTR2), vmcs_read64(GUEST_PDPTR3));
5647 	}
5648 	pr_err("RSP = 0x%016lx  RIP = 0x%016lx\n",
5649 	       vmcs_readl(GUEST_RSP), vmcs_readl(GUEST_RIP));
5650 	pr_err("RFLAGS=0x%08lx         DR7 = 0x%016lx\n",
5651 	       vmcs_readl(GUEST_RFLAGS), vmcs_readl(GUEST_DR7));
5652 	pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
5653 	       vmcs_readl(GUEST_SYSENTER_ESP),
5654 	       vmcs_read32(GUEST_SYSENTER_CS), vmcs_readl(GUEST_SYSENTER_EIP));
5655 	vmx_dump_sel("CS:  ", GUEST_CS_SELECTOR);
5656 	vmx_dump_sel("DS:  ", GUEST_DS_SELECTOR);
5657 	vmx_dump_sel("SS:  ", GUEST_SS_SELECTOR);
5658 	vmx_dump_sel("ES:  ", GUEST_ES_SELECTOR);
5659 	vmx_dump_sel("FS:  ", GUEST_FS_SELECTOR);
5660 	vmx_dump_sel("GS:  ", GUEST_GS_SELECTOR);
5661 	vmx_dump_dtsel("GDTR:", GUEST_GDTR_LIMIT);
5662 	vmx_dump_sel("LDTR:", GUEST_LDTR_SELECTOR);
5663 	vmx_dump_dtsel("IDTR:", GUEST_IDTR_LIMIT);
5664 	vmx_dump_sel("TR:  ", GUEST_TR_SELECTOR);
5665 	if ((vmexit_ctl & (VM_EXIT_SAVE_IA32_PAT | VM_EXIT_SAVE_IA32_EFER)) ||
5666 	    (vmentry_ctl & (VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_IA32_EFER)))
5667 		pr_err("EFER =     0x%016llx  PAT = 0x%016llx\n",
5668 		       efer, vmcs_read64(GUEST_IA32_PAT));
5669 	pr_err("DebugCtl = 0x%016llx  DebugExceptions = 0x%016lx\n",
5670 	       vmcs_read64(GUEST_IA32_DEBUGCTL),
5671 	       vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS));
5672 	if (cpu_has_load_perf_global_ctrl() &&
5673 	    vmentry_ctl & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
5674 		pr_err("PerfGlobCtl = 0x%016llx\n",
5675 		       vmcs_read64(GUEST_IA32_PERF_GLOBAL_CTRL));
5676 	if (vmentry_ctl & VM_ENTRY_LOAD_BNDCFGS)
5677 		pr_err("BndCfgS = 0x%016llx\n", vmcs_read64(GUEST_BNDCFGS));
5678 	pr_err("Interruptibility = %08x  ActivityState = %08x\n",
5679 	       vmcs_read32(GUEST_INTERRUPTIBILITY_INFO),
5680 	       vmcs_read32(GUEST_ACTIVITY_STATE));
5681 	if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)
5682 		pr_err("InterruptStatus = %04x\n",
5683 		       vmcs_read16(GUEST_INTR_STATUS));
5684 
5685 	pr_err("*** Host State ***\n");
5686 	pr_err("RIP = 0x%016lx  RSP = 0x%016lx\n",
5687 	       vmcs_readl(HOST_RIP), vmcs_readl(HOST_RSP));
5688 	pr_err("CS=%04x SS=%04x DS=%04x ES=%04x FS=%04x GS=%04x TR=%04x\n",
5689 	       vmcs_read16(HOST_CS_SELECTOR), vmcs_read16(HOST_SS_SELECTOR),
5690 	       vmcs_read16(HOST_DS_SELECTOR), vmcs_read16(HOST_ES_SELECTOR),
5691 	       vmcs_read16(HOST_FS_SELECTOR), vmcs_read16(HOST_GS_SELECTOR),
5692 	       vmcs_read16(HOST_TR_SELECTOR));
5693 	pr_err("FSBase=%016lx GSBase=%016lx TRBase=%016lx\n",
5694 	       vmcs_readl(HOST_FS_BASE), vmcs_readl(HOST_GS_BASE),
5695 	       vmcs_readl(HOST_TR_BASE));
5696 	pr_err("GDTBase=%016lx IDTBase=%016lx\n",
5697 	       vmcs_readl(HOST_GDTR_BASE), vmcs_readl(HOST_IDTR_BASE));
5698 	pr_err("CR0=%016lx CR3=%016lx CR4=%016lx\n",
5699 	       vmcs_readl(HOST_CR0), vmcs_readl(HOST_CR3),
5700 	       vmcs_readl(HOST_CR4));
5701 	pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
5702 	       vmcs_readl(HOST_IA32_SYSENTER_ESP),
5703 	       vmcs_read32(HOST_IA32_SYSENTER_CS),
5704 	       vmcs_readl(HOST_IA32_SYSENTER_EIP));
5705 	if (vmexit_ctl & (VM_EXIT_LOAD_IA32_PAT | VM_EXIT_LOAD_IA32_EFER))
5706 		pr_err("EFER = 0x%016llx  PAT = 0x%016llx\n",
5707 		       vmcs_read64(HOST_IA32_EFER),
5708 		       vmcs_read64(HOST_IA32_PAT));
5709 	if (cpu_has_load_perf_global_ctrl() &&
5710 	    vmexit_ctl & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
5711 		pr_err("PerfGlobCtl = 0x%016llx\n",
5712 		       vmcs_read64(HOST_IA32_PERF_GLOBAL_CTRL));
5713 
5714 	pr_err("*** Control State ***\n");
5715 	pr_err("PinBased=%08x CPUBased=%08x SecondaryExec=%08x\n",
5716 	       pin_based_exec_ctrl, cpu_based_exec_ctrl, secondary_exec_control);
5717 	pr_err("EntryControls=%08x ExitControls=%08x\n", vmentry_ctl, vmexit_ctl);
5718 	pr_err("ExceptionBitmap=%08x PFECmask=%08x PFECmatch=%08x\n",
5719 	       vmcs_read32(EXCEPTION_BITMAP),
5720 	       vmcs_read32(PAGE_FAULT_ERROR_CODE_MASK),
5721 	       vmcs_read32(PAGE_FAULT_ERROR_CODE_MATCH));
5722 	pr_err("VMEntry: intr_info=%08x errcode=%08x ilen=%08x\n",
5723 	       vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
5724 	       vmcs_read32(VM_ENTRY_EXCEPTION_ERROR_CODE),
5725 	       vmcs_read32(VM_ENTRY_INSTRUCTION_LEN));
5726 	pr_err("VMExit: intr_info=%08x errcode=%08x ilen=%08x\n",
5727 	       vmcs_read32(VM_EXIT_INTR_INFO),
5728 	       vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
5729 	       vmcs_read32(VM_EXIT_INSTRUCTION_LEN));
5730 	pr_err("        reason=%08x qualification=%016lx\n",
5731 	       vmcs_read32(VM_EXIT_REASON), vmcs_readl(EXIT_QUALIFICATION));
5732 	pr_err("IDTVectoring: info=%08x errcode=%08x\n",
5733 	       vmcs_read32(IDT_VECTORING_INFO_FIELD),
5734 	       vmcs_read32(IDT_VECTORING_ERROR_CODE));
5735 	pr_err("TSC Offset = 0x%016llx\n", vmcs_read64(TSC_OFFSET));
5736 	if (secondary_exec_control & SECONDARY_EXEC_TSC_SCALING)
5737 		pr_err("TSC Multiplier = 0x%016llx\n",
5738 		       vmcs_read64(TSC_MULTIPLIER));
5739 	if (cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW)
5740 		pr_err("TPR Threshold = 0x%02x\n", vmcs_read32(TPR_THRESHOLD));
5741 	if (pin_based_exec_ctrl & PIN_BASED_POSTED_INTR)
5742 		pr_err("PostedIntrVec = 0x%02x\n", vmcs_read16(POSTED_INTR_NV));
5743 	if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT))
5744 		pr_err("EPT pointer = 0x%016llx\n", vmcs_read64(EPT_POINTER));
5745 	n = vmcs_read32(CR3_TARGET_COUNT);
5746 	for (i = 0; i + 1 < n; i += 4)
5747 		pr_err("CR3 target%u=%016lx target%u=%016lx\n",
5748 		       i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2),
5749 		       i + 1, vmcs_readl(CR3_TARGET_VALUE0 + i * 2 + 2));
5750 	if (i < n)
5751 		pr_err("CR3 target%u=%016lx\n",
5752 		       i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2));
5753 	if (secondary_exec_control & SECONDARY_EXEC_PAUSE_LOOP_EXITING)
5754 		pr_err("PLE Gap=%08x Window=%08x\n",
5755 		       vmcs_read32(PLE_GAP), vmcs_read32(PLE_WINDOW));
5756 	if (secondary_exec_control & SECONDARY_EXEC_ENABLE_VPID)
5757 		pr_err("Virtual processor ID = 0x%04x\n",
5758 		       vmcs_read16(VIRTUAL_PROCESSOR_ID));
5759 }
5760 
5761 /*
5762  * The guest has exited.  See if we can fix it or if we need userspace
5763  * assistance.
5764  */
5765 static int vmx_handle_exit(struct kvm_vcpu *vcpu)
5766 {
5767 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5768 	u32 exit_reason = vmx->exit_reason;
5769 	u32 vectoring_info = vmx->idt_vectoring_info;
5770 
5771 	trace_kvm_exit(exit_reason, vcpu, KVM_ISA_VMX);
5772 
5773 	/*
5774 	 * Flush logged GPAs PML buffer, this will make dirty_bitmap more
5775 	 * updated. Another good is, in kvm_vm_ioctl_get_dirty_log, before
5776 	 * querying dirty_bitmap, we only need to kick all vcpus out of guest
5777 	 * mode as if vcpus is in root mode, the PML buffer must has been
5778 	 * flushed already.
5779 	 */
5780 	if (enable_pml)
5781 		vmx_flush_pml_buffer(vcpu);
5782 
5783 	/* If guest state is invalid, start emulating */
5784 	if (vmx->emulation_required)
5785 		return handle_invalid_guest_state(vcpu);
5786 
5787 	if (is_guest_mode(vcpu) && nested_vmx_exit_reflected(vcpu, exit_reason))
5788 		return nested_vmx_reflect_vmexit(vcpu, exit_reason);
5789 
5790 	if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) {
5791 		dump_vmcs();
5792 		vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
5793 		vcpu->run->fail_entry.hardware_entry_failure_reason
5794 			= exit_reason;
5795 		return 0;
5796 	}
5797 
5798 	if (unlikely(vmx->fail)) {
5799 		vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
5800 		vcpu->run->fail_entry.hardware_entry_failure_reason
5801 			= vmcs_read32(VM_INSTRUCTION_ERROR);
5802 		return 0;
5803 	}
5804 
5805 	/*
5806 	 * Note:
5807 	 * Do not try to fix EXIT_REASON_EPT_MISCONFIG if it caused by
5808 	 * delivery event since it indicates guest is accessing MMIO.
5809 	 * The vm-exit can be triggered again after return to guest that
5810 	 * will cause infinite loop.
5811 	 */
5812 	if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
5813 			(exit_reason != EXIT_REASON_EXCEPTION_NMI &&
5814 			exit_reason != EXIT_REASON_EPT_VIOLATION &&
5815 			exit_reason != EXIT_REASON_PML_FULL &&
5816 			exit_reason != EXIT_REASON_TASK_SWITCH)) {
5817 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5818 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_DELIVERY_EV;
5819 		vcpu->run->internal.ndata = 3;
5820 		vcpu->run->internal.data[0] = vectoring_info;
5821 		vcpu->run->internal.data[1] = exit_reason;
5822 		vcpu->run->internal.data[2] = vcpu->arch.exit_qualification;
5823 		if (exit_reason == EXIT_REASON_EPT_MISCONFIG) {
5824 			vcpu->run->internal.ndata++;
5825 			vcpu->run->internal.data[3] =
5826 				vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5827 		}
5828 		return 0;
5829 	}
5830 
5831 	if (unlikely(!enable_vnmi &&
5832 		     vmx->loaded_vmcs->soft_vnmi_blocked)) {
5833 		if (vmx_interrupt_allowed(vcpu)) {
5834 			vmx->loaded_vmcs->soft_vnmi_blocked = 0;
5835 		} else if (vmx->loaded_vmcs->vnmi_blocked_time > 1000000000LL &&
5836 			   vcpu->arch.nmi_pending) {
5837 			/*
5838 			 * This CPU don't support us in finding the end of an
5839 			 * NMI-blocked window if the guest runs with IRQs
5840 			 * disabled. So we pull the trigger after 1 s of
5841 			 * futile waiting, but inform the user about this.
5842 			 */
5843 			printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
5844 			       "state on VCPU %d after 1 s timeout\n",
5845 			       __func__, vcpu->vcpu_id);
5846 			vmx->loaded_vmcs->soft_vnmi_blocked = 0;
5847 		}
5848 	}
5849 
5850 	if (exit_reason < kvm_vmx_max_exit_handlers
5851 	    && kvm_vmx_exit_handlers[exit_reason])
5852 		return kvm_vmx_exit_handlers[exit_reason](vcpu);
5853 	else {
5854 		vcpu_unimpl(vcpu, "vmx: unexpected exit reason 0x%x\n",
5855 				exit_reason);
5856 		kvm_queue_exception(vcpu, UD_VECTOR);
5857 		return 1;
5858 	}
5859 }
5860 
5861 /*
5862  * Software based L1D cache flush which is used when microcode providing
5863  * the cache control MSR is not loaded.
5864  *
5865  * The L1D cache is 32 KiB on Nehalem and later microarchitectures, but to
5866  * flush it is required to read in 64 KiB because the replacement algorithm
5867  * is not exactly LRU. This could be sized at runtime via topology
5868  * information but as all relevant affected CPUs have 32KiB L1D cache size
5869  * there is no point in doing so.
5870  */
5871 static void vmx_l1d_flush(struct kvm_vcpu *vcpu)
5872 {
5873 	int size = PAGE_SIZE << L1D_CACHE_ORDER;
5874 
5875 	/*
5876 	 * This code is only executed when the the flush mode is 'cond' or
5877 	 * 'always'
5878 	 */
5879 	if (static_branch_likely(&vmx_l1d_flush_cond)) {
5880 		bool flush_l1d;
5881 
5882 		/*
5883 		 * Clear the per-vcpu flush bit, it gets set again
5884 		 * either from vcpu_run() or from one of the unsafe
5885 		 * VMEXIT handlers.
5886 		 */
5887 		flush_l1d = vcpu->arch.l1tf_flush_l1d;
5888 		vcpu->arch.l1tf_flush_l1d = false;
5889 
5890 		/*
5891 		 * Clear the per-cpu flush bit, it gets set again from
5892 		 * the interrupt handlers.
5893 		 */
5894 		flush_l1d |= kvm_get_cpu_l1tf_flush_l1d();
5895 		kvm_clear_cpu_l1tf_flush_l1d();
5896 
5897 		if (!flush_l1d)
5898 			return;
5899 	}
5900 
5901 	vcpu->stat.l1d_flush++;
5902 
5903 	if (static_cpu_has(X86_FEATURE_FLUSH_L1D)) {
5904 		wrmsrl(MSR_IA32_FLUSH_CMD, L1D_FLUSH);
5905 		return;
5906 	}
5907 
5908 	asm volatile(
5909 		/* First ensure the pages are in the TLB */
5910 		"xorl	%%eax, %%eax\n"
5911 		".Lpopulate_tlb:\n\t"
5912 		"movzbl	(%[flush_pages], %%" _ASM_AX "), %%ecx\n\t"
5913 		"addl	$4096, %%eax\n\t"
5914 		"cmpl	%%eax, %[size]\n\t"
5915 		"jne	.Lpopulate_tlb\n\t"
5916 		"xorl	%%eax, %%eax\n\t"
5917 		"cpuid\n\t"
5918 		/* Now fill the cache */
5919 		"xorl	%%eax, %%eax\n"
5920 		".Lfill_cache:\n"
5921 		"movzbl	(%[flush_pages], %%" _ASM_AX "), %%ecx\n\t"
5922 		"addl	$64, %%eax\n\t"
5923 		"cmpl	%%eax, %[size]\n\t"
5924 		"jne	.Lfill_cache\n\t"
5925 		"lfence\n"
5926 		:: [flush_pages] "r" (vmx_l1d_flush_pages),
5927 		    [size] "r" (size)
5928 		: "eax", "ebx", "ecx", "edx");
5929 }
5930 
5931 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
5932 {
5933 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5934 
5935 	if (is_guest_mode(vcpu) &&
5936 		nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
5937 		return;
5938 
5939 	if (irr == -1 || tpr < irr) {
5940 		vmcs_write32(TPR_THRESHOLD, 0);
5941 		return;
5942 	}
5943 
5944 	vmcs_write32(TPR_THRESHOLD, irr);
5945 }
5946 
5947 void vmx_set_virtual_apic_mode(struct kvm_vcpu *vcpu)
5948 {
5949 	u32 sec_exec_control;
5950 
5951 	if (!lapic_in_kernel(vcpu))
5952 		return;
5953 
5954 	if (!flexpriority_enabled &&
5955 	    !cpu_has_vmx_virtualize_x2apic_mode())
5956 		return;
5957 
5958 	/* Postpone execution until vmcs01 is the current VMCS. */
5959 	if (is_guest_mode(vcpu)) {
5960 		to_vmx(vcpu)->nested.change_vmcs01_virtual_apic_mode = true;
5961 		return;
5962 	}
5963 
5964 	sec_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
5965 	sec_exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
5966 			      SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE);
5967 
5968 	switch (kvm_get_apic_mode(vcpu)) {
5969 	case LAPIC_MODE_INVALID:
5970 		WARN_ONCE(true, "Invalid local APIC state");
5971 	case LAPIC_MODE_DISABLED:
5972 		break;
5973 	case LAPIC_MODE_XAPIC:
5974 		if (flexpriority_enabled) {
5975 			sec_exec_control |=
5976 				SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
5977 			vmx_flush_tlb(vcpu, true);
5978 		}
5979 		break;
5980 	case LAPIC_MODE_X2APIC:
5981 		if (cpu_has_vmx_virtualize_x2apic_mode())
5982 			sec_exec_control |=
5983 				SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
5984 		break;
5985 	}
5986 	vmcs_write32(SECONDARY_VM_EXEC_CONTROL, sec_exec_control);
5987 
5988 	vmx_update_msr_bitmap(vcpu);
5989 }
5990 
5991 static void vmx_set_apic_access_page_addr(struct kvm_vcpu *vcpu, hpa_t hpa)
5992 {
5993 	if (!is_guest_mode(vcpu)) {
5994 		vmcs_write64(APIC_ACCESS_ADDR, hpa);
5995 		vmx_flush_tlb(vcpu, true);
5996 	}
5997 }
5998 
5999 static void vmx_hwapic_isr_update(struct kvm_vcpu *vcpu, int max_isr)
6000 {
6001 	u16 status;
6002 	u8 old;
6003 
6004 	if (max_isr == -1)
6005 		max_isr = 0;
6006 
6007 	status = vmcs_read16(GUEST_INTR_STATUS);
6008 	old = status >> 8;
6009 	if (max_isr != old) {
6010 		status &= 0xff;
6011 		status |= max_isr << 8;
6012 		vmcs_write16(GUEST_INTR_STATUS, status);
6013 	}
6014 }
6015 
6016 static void vmx_set_rvi(int vector)
6017 {
6018 	u16 status;
6019 	u8 old;
6020 
6021 	if (vector == -1)
6022 		vector = 0;
6023 
6024 	status = vmcs_read16(GUEST_INTR_STATUS);
6025 	old = (u8)status & 0xff;
6026 	if ((u8)vector != old) {
6027 		status &= ~0xff;
6028 		status |= (u8)vector;
6029 		vmcs_write16(GUEST_INTR_STATUS, status);
6030 	}
6031 }
6032 
6033 static void vmx_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr)
6034 {
6035 	/*
6036 	 * When running L2, updating RVI is only relevant when
6037 	 * vmcs12 virtual-interrupt-delivery enabled.
6038 	 * However, it can be enabled only when L1 also
6039 	 * intercepts external-interrupts and in that case
6040 	 * we should not update vmcs02 RVI but instead intercept
6041 	 * interrupt. Therefore, do nothing when running L2.
6042 	 */
6043 	if (!is_guest_mode(vcpu))
6044 		vmx_set_rvi(max_irr);
6045 }
6046 
6047 static int vmx_sync_pir_to_irr(struct kvm_vcpu *vcpu)
6048 {
6049 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6050 	int max_irr;
6051 	bool max_irr_updated;
6052 
6053 	WARN_ON(!vcpu->arch.apicv_active);
6054 	if (pi_test_on(&vmx->pi_desc)) {
6055 		pi_clear_on(&vmx->pi_desc);
6056 		/*
6057 		 * IOMMU can write to PIR.ON, so the barrier matters even on UP.
6058 		 * But on x86 this is just a compiler barrier anyway.
6059 		 */
6060 		smp_mb__after_atomic();
6061 		max_irr_updated =
6062 			kvm_apic_update_irr(vcpu, vmx->pi_desc.pir, &max_irr);
6063 
6064 		/*
6065 		 * If we are running L2 and L1 has a new pending interrupt
6066 		 * which can be injected, we should re-evaluate
6067 		 * what should be done with this new L1 interrupt.
6068 		 * If L1 intercepts external-interrupts, we should
6069 		 * exit from L2 to L1. Otherwise, interrupt should be
6070 		 * delivered directly to L2.
6071 		 */
6072 		if (is_guest_mode(vcpu) && max_irr_updated) {
6073 			if (nested_exit_on_intr(vcpu))
6074 				kvm_vcpu_exiting_guest_mode(vcpu);
6075 			else
6076 				kvm_make_request(KVM_REQ_EVENT, vcpu);
6077 		}
6078 	} else {
6079 		max_irr = kvm_lapic_find_highest_irr(vcpu);
6080 	}
6081 	vmx_hwapic_irr_update(vcpu, max_irr);
6082 	return max_irr;
6083 }
6084 
6085 static void vmx_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap)
6086 {
6087 	if (!kvm_vcpu_apicv_active(vcpu))
6088 		return;
6089 
6090 	vmcs_write64(EOI_EXIT_BITMAP0, eoi_exit_bitmap[0]);
6091 	vmcs_write64(EOI_EXIT_BITMAP1, eoi_exit_bitmap[1]);
6092 	vmcs_write64(EOI_EXIT_BITMAP2, eoi_exit_bitmap[2]);
6093 	vmcs_write64(EOI_EXIT_BITMAP3, eoi_exit_bitmap[3]);
6094 }
6095 
6096 static void vmx_apicv_post_state_restore(struct kvm_vcpu *vcpu)
6097 {
6098 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6099 
6100 	pi_clear_on(&vmx->pi_desc);
6101 	memset(vmx->pi_desc.pir, 0, sizeof(vmx->pi_desc.pir));
6102 }
6103 
6104 static void vmx_complete_atomic_exit(struct vcpu_vmx *vmx)
6105 {
6106 	u32 exit_intr_info = 0;
6107 	u16 basic_exit_reason = (u16)vmx->exit_reason;
6108 
6109 	if (!(basic_exit_reason == EXIT_REASON_MCE_DURING_VMENTRY
6110 	      || basic_exit_reason == EXIT_REASON_EXCEPTION_NMI))
6111 		return;
6112 
6113 	if (!(vmx->exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY))
6114 		exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6115 	vmx->exit_intr_info = exit_intr_info;
6116 
6117 	/* if exit due to PF check for async PF */
6118 	if (is_page_fault(exit_intr_info))
6119 		vmx->vcpu.arch.apf.host_apf_reason = kvm_read_and_reset_pf_reason();
6120 
6121 	/* Handle machine checks before interrupts are enabled */
6122 	if (basic_exit_reason == EXIT_REASON_MCE_DURING_VMENTRY ||
6123 	    is_machine_check(exit_intr_info))
6124 		kvm_machine_check();
6125 
6126 	/* We need to handle NMIs before interrupts are enabled */
6127 	if (is_nmi(exit_intr_info)) {
6128 		kvm_before_interrupt(&vmx->vcpu);
6129 		asm("int $2");
6130 		kvm_after_interrupt(&vmx->vcpu);
6131 	}
6132 }
6133 
6134 static void vmx_handle_external_intr(struct kvm_vcpu *vcpu)
6135 {
6136 	u32 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6137 
6138 	if ((exit_intr_info & (INTR_INFO_VALID_MASK | INTR_INFO_INTR_TYPE_MASK))
6139 			== (INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR)) {
6140 		unsigned int vector;
6141 		unsigned long entry;
6142 		gate_desc *desc;
6143 		struct vcpu_vmx *vmx = to_vmx(vcpu);
6144 #ifdef CONFIG_X86_64
6145 		unsigned long tmp;
6146 #endif
6147 
6148 		vector =  exit_intr_info & INTR_INFO_VECTOR_MASK;
6149 		desc = (gate_desc *)vmx->host_idt_base + vector;
6150 		entry = gate_offset(desc);
6151 		asm volatile(
6152 #ifdef CONFIG_X86_64
6153 			"mov %%" _ASM_SP ", %[sp]\n\t"
6154 			"and $0xfffffffffffffff0, %%" _ASM_SP "\n\t"
6155 			"push $%c[ss]\n\t"
6156 			"push %[sp]\n\t"
6157 #endif
6158 			"pushf\n\t"
6159 			__ASM_SIZE(push) " $%c[cs]\n\t"
6160 			CALL_NOSPEC
6161 			:
6162 #ifdef CONFIG_X86_64
6163 			[sp]"=&r"(tmp),
6164 #endif
6165 			ASM_CALL_CONSTRAINT
6166 			:
6167 			THUNK_TARGET(entry),
6168 			[ss]"i"(__KERNEL_DS),
6169 			[cs]"i"(__KERNEL_CS)
6170 			);
6171 	}
6172 }
6173 STACK_FRAME_NON_STANDARD(vmx_handle_external_intr);
6174 
6175 static bool vmx_has_emulated_msr(int index)
6176 {
6177 	switch (index) {
6178 	case MSR_IA32_SMBASE:
6179 		/*
6180 		 * We cannot do SMM unless we can run the guest in big
6181 		 * real mode.
6182 		 */
6183 		return enable_unrestricted_guest || emulate_invalid_guest_state;
6184 	case MSR_AMD64_VIRT_SPEC_CTRL:
6185 		/* This is AMD only.  */
6186 		return false;
6187 	default:
6188 		return true;
6189 	}
6190 }
6191 
6192 static bool vmx_pt_supported(void)
6193 {
6194 	return pt_mode == PT_MODE_HOST_GUEST;
6195 }
6196 
6197 static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx)
6198 {
6199 	u32 exit_intr_info;
6200 	bool unblock_nmi;
6201 	u8 vector;
6202 	bool idtv_info_valid;
6203 
6204 	idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK;
6205 
6206 	if (enable_vnmi) {
6207 		if (vmx->loaded_vmcs->nmi_known_unmasked)
6208 			return;
6209 		/*
6210 		 * Can't use vmx->exit_intr_info since we're not sure what
6211 		 * the exit reason is.
6212 		 */
6213 		exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6214 		unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
6215 		vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
6216 		/*
6217 		 * SDM 3: 27.7.1.2 (September 2008)
6218 		 * Re-set bit "block by NMI" before VM entry if vmexit caused by
6219 		 * a guest IRET fault.
6220 		 * SDM 3: 23.2.2 (September 2008)
6221 		 * Bit 12 is undefined in any of the following cases:
6222 		 *  If the VM exit sets the valid bit in the IDT-vectoring
6223 		 *   information field.
6224 		 *  If the VM exit is due to a double fault.
6225 		 */
6226 		if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
6227 		    vector != DF_VECTOR && !idtv_info_valid)
6228 			vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
6229 				      GUEST_INTR_STATE_NMI);
6230 		else
6231 			vmx->loaded_vmcs->nmi_known_unmasked =
6232 				!(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO)
6233 				  & GUEST_INTR_STATE_NMI);
6234 	} else if (unlikely(vmx->loaded_vmcs->soft_vnmi_blocked))
6235 		vmx->loaded_vmcs->vnmi_blocked_time +=
6236 			ktime_to_ns(ktime_sub(ktime_get(),
6237 					      vmx->loaded_vmcs->entry_time));
6238 }
6239 
6240 static void __vmx_complete_interrupts(struct kvm_vcpu *vcpu,
6241 				      u32 idt_vectoring_info,
6242 				      int instr_len_field,
6243 				      int error_code_field)
6244 {
6245 	u8 vector;
6246 	int type;
6247 	bool idtv_info_valid;
6248 
6249 	idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
6250 
6251 	vcpu->arch.nmi_injected = false;
6252 	kvm_clear_exception_queue(vcpu);
6253 	kvm_clear_interrupt_queue(vcpu);
6254 
6255 	if (!idtv_info_valid)
6256 		return;
6257 
6258 	kvm_make_request(KVM_REQ_EVENT, vcpu);
6259 
6260 	vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
6261 	type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
6262 
6263 	switch (type) {
6264 	case INTR_TYPE_NMI_INTR:
6265 		vcpu->arch.nmi_injected = true;
6266 		/*
6267 		 * SDM 3: 27.7.1.2 (September 2008)
6268 		 * Clear bit "block by NMI" before VM entry if a NMI
6269 		 * delivery faulted.
6270 		 */
6271 		vmx_set_nmi_mask(vcpu, false);
6272 		break;
6273 	case INTR_TYPE_SOFT_EXCEPTION:
6274 		vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
6275 		/* fall through */
6276 	case INTR_TYPE_HARD_EXCEPTION:
6277 		if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
6278 			u32 err = vmcs_read32(error_code_field);
6279 			kvm_requeue_exception_e(vcpu, vector, err);
6280 		} else
6281 			kvm_requeue_exception(vcpu, vector);
6282 		break;
6283 	case INTR_TYPE_SOFT_INTR:
6284 		vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
6285 		/* fall through */
6286 	case INTR_TYPE_EXT_INTR:
6287 		kvm_queue_interrupt(vcpu, vector, type == INTR_TYPE_SOFT_INTR);
6288 		break;
6289 	default:
6290 		break;
6291 	}
6292 }
6293 
6294 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
6295 {
6296 	__vmx_complete_interrupts(&vmx->vcpu, vmx->idt_vectoring_info,
6297 				  VM_EXIT_INSTRUCTION_LEN,
6298 				  IDT_VECTORING_ERROR_CODE);
6299 }
6300 
6301 static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
6302 {
6303 	__vmx_complete_interrupts(vcpu,
6304 				  vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
6305 				  VM_ENTRY_INSTRUCTION_LEN,
6306 				  VM_ENTRY_EXCEPTION_ERROR_CODE);
6307 
6308 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
6309 }
6310 
6311 static void atomic_switch_perf_msrs(struct vcpu_vmx *vmx)
6312 {
6313 	int i, nr_msrs;
6314 	struct perf_guest_switch_msr *msrs;
6315 
6316 	msrs = perf_guest_get_msrs(&nr_msrs);
6317 
6318 	if (!msrs)
6319 		return;
6320 
6321 	for (i = 0; i < nr_msrs; i++)
6322 		if (msrs[i].host == msrs[i].guest)
6323 			clear_atomic_switch_msr(vmx, msrs[i].msr);
6324 		else
6325 			add_atomic_switch_msr(vmx, msrs[i].msr, msrs[i].guest,
6326 					msrs[i].host, false);
6327 }
6328 
6329 static void vmx_arm_hv_timer(struct vcpu_vmx *vmx, u32 val)
6330 {
6331 	vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, val);
6332 	if (!vmx->loaded_vmcs->hv_timer_armed)
6333 		vmcs_set_bits(PIN_BASED_VM_EXEC_CONTROL,
6334 			      PIN_BASED_VMX_PREEMPTION_TIMER);
6335 	vmx->loaded_vmcs->hv_timer_armed = true;
6336 }
6337 
6338 static void vmx_update_hv_timer(struct kvm_vcpu *vcpu)
6339 {
6340 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6341 	u64 tscl;
6342 	u32 delta_tsc;
6343 
6344 	if (vmx->req_immediate_exit) {
6345 		vmx_arm_hv_timer(vmx, 0);
6346 		return;
6347 	}
6348 
6349 	if (vmx->hv_deadline_tsc != -1) {
6350 		tscl = rdtsc();
6351 		if (vmx->hv_deadline_tsc > tscl)
6352 			/* set_hv_timer ensures the delta fits in 32-bits */
6353 			delta_tsc = (u32)((vmx->hv_deadline_tsc - tscl) >>
6354 				cpu_preemption_timer_multi);
6355 		else
6356 			delta_tsc = 0;
6357 
6358 		vmx_arm_hv_timer(vmx, delta_tsc);
6359 		return;
6360 	}
6361 
6362 	if (vmx->loaded_vmcs->hv_timer_armed)
6363 		vmcs_clear_bits(PIN_BASED_VM_EXEC_CONTROL,
6364 				PIN_BASED_VMX_PREEMPTION_TIMER);
6365 	vmx->loaded_vmcs->hv_timer_armed = false;
6366 }
6367 
6368 void vmx_update_host_rsp(struct vcpu_vmx *vmx, unsigned long host_rsp)
6369 {
6370 	if (unlikely(host_rsp != vmx->loaded_vmcs->host_state.rsp)) {
6371 		vmx->loaded_vmcs->host_state.rsp = host_rsp;
6372 		vmcs_writel(HOST_RSP, host_rsp);
6373 	}
6374 }
6375 
6376 bool __vmx_vcpu_run(struct vcpu_vmx *vmx, unsigned long *regs, bool launched);
6377 
6378 static void vmx_vcpu_run(struct kvm_vcpu *vcpu)
6379 {
6380 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6381 	unsigned long cr3, cr4;
6382 
6383 	/* Record the guest's net vcpu time for enforced NMI injections. */
6384 	if (unlikely(!enable_vnmi &&
6385 		     vmx->loaded_vmcs->soft_vnmi_blocked))
6386 		vmx->loaded_vmcs->entry_time = ktime_get();
6387 
6388 	/* Don't enter VMX if guest state is invalid, let the exit handler
6389 	   start emulation until we arrive back to a valid state */
6390 	if (vmx->emulation_required)
6391 		return;
6392 
6393 	if (vmx->ple_window_dirty) {
6394 		vmx->ple_window_dirty = false;
6395 		vmcs_write32(PLE_WINDOW, vmx->ple_window);
6396 	}
6397 
6398 	if (vmx->nested.need_vmcs12_sync)
6399 		nested_sync_from_vmcs12(vcpu);
6400 
6401 	if (test_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_dirty))
6402 		vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
6403 	if (test_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_dirty))
6404 		vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
6405 
6406 	cr3 = __get_current_cr3_fast();
6407 	if (unlikely(cr3 != vmx->loaded_vmcs->host_state.cr3)) {
6408 		vmcs_writel(HOST_CR3, cr3);
6409 		vmx->loaded_vmcs->host_state.cr3 = cr3;
6410 	}
6411 
6412 	cr4 = cr4_read_shadow();
6413 	if (unlikely(cr4 != vmx->loaded_vmcs->host_state.cr4)) {
6414 		vmcs_writel(HOST_CR4, cr4);
6415 		vmx->loaded_vmcs->host_state.cr4 = cr4;
6416 	}
6417 
6418 	/* When single-stepping over STI and MOV SS, we must clear the
6419 	 * corresponding interruptibility bits in the guest state. Otherwise
6420 	 * vmentry fails as it then expects bit 14 (BS) in pending debug
6421 	 * exceptions being set, but that's not correct for the guest debugging
6422 	 * case. */
6423 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6424 		vmx_set_interrupt_shadow(vcpu, 0);
6425 
6426 	if (static_cpu_has(X86_FEATURE_PKU) &&
6427 	    kvm_read_cr4_bits(vcpu, X86_CR4_PKE) &&
6428 	    vcpu->arch.pkru != vmx->host_pkru)
6429 		__write_pkru(vcpu->arch.pkru);
6430 
6431 	pt_guest_enter(vmx);
6432 
6433 	atomic_switch_perf_msrs(vmx);
6434 
6435 	vmx_update_hv_timer(vcpu);
6436 
6437 	/*
6438 	 * If this vCPU has touched SPEC_CTRL, restore the guest's value if
6439 	 * it's non-zero. Since vmentry is serialising on affected CPUs, there
6440 	 * is no need to worry about the conditional branch over the wrmsr
6441 	 * being speculatively taken.
6442 	 */
6443 	x86_spec_ctrl_set_guest(vmx->spec_ctrl, 0);
6444 
6445 	if (static_branch_unlikely(&vmx_l1d_should_flush))
6446 		vmx_l1d_flush(vcpu);
6447 
6448 	if (vcpu->arch.cr2 != read_cr2())
6449 		write_cr2(vcpu->arch.cr2);
6450 
6451 	vmx->fail = __vmx_vcpu_run(vmx, (unsigned long *)&vcpu->arch.regs,
6452 				   vmx->loaded_vmcs->launched);
6453 
6454 	vcpu->arch.cr2 = read_cr2();
6455 
6456 	/*
6457 	 * We do not use IBRS in the kernel. If this vCPU has used the
6458 	 * SPEC_CTRL MSR it may have left it on; save the value and
6459 	 * turn it off. This is much more efficient than blindly adding
6460 	 * it to the atomic save/restore list. Especially as the former
6461 	 * (Saving guest MSRs on vmexit) doesn't even exist in KVM.
6462 	 *
6463 	 * For non-nested case:
6464 	 * If the L01 MSR bitmap does not intercept the MSR, then we need to
6465 	 * save it.
6466 	 *
6467 	 * For nested case:
6468 	 * If the L02 MSR bitmap does not intercept the MSR, then we need to
6469 	 * save it.
6470 	 */
6471 	if (unlikely(!msr_write_intercepted(vcpu, MSR_IA32_SPEC_CTRL)))
6472 		vmx->spec_ctrl = native_read_msr(MSR_IA32_SPEC_CTRL);
6473 
6474 	x86_spec_ctrl_restore_host(vmx->spec_ctrl, 0);
6475 
6476 	/* Eliminate branch target predictions from guest mode */
6477 	vmexit_fill_RSB();
6478 
6479 	/* All fields are clean at this point */
6480 	if (static_branch_unlikely(&enable_evmcs))
6481 		current_evmcs->hv_clean_fields |=
6482 			HV_VMX_ENLIGHTENED_CLEAN_FIELD_ALL;
6483 
6484 	/* MSR_IA32_DEBUGCTLMSR is zeroed on vmexit. Restore it if needed */
6485 	if (vmx->host_debugctlmsr)
6486 		update_debugctlmsr(vmx->host_debugctlmsr);
6487 
6488 #ifndef CONFIG_X86_64
6489 	/*
6490 	 * The sysexit path does not restore ds/es, so we must set them to
6491 	 * a reasonable value ourselves.
6492 	 *
6493 	 * We can't defer this to vmx_prepare_switch_to_host() since that
6494 	 * function may be executed in interrupt context, which saves and
6495 	 * restore segments around it, nullifying its effect.
6496 	 */
6497 	loadsegment(ds, __USER_DS);
6498 	loadsegment(es, __USER_DS);
6499 #endif
6500 
6501 	vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
6502 				  | (1 << VCPU_EXREG_RFLAGS)
6503 				  | (1 << VCPU_EXREG_PDPTR)
6504 				  | (1 << VCPU_EXREG_SEGMENTS)
6505 				  | (1 << VCPU_EXREG_CR3));
6506 	vcpu->arch.regs_dirty = 0;
6507 
6508 	pt_guest_exit(vmx);
6509 
6510 	/*
6511 	 * eager fpu is enabled if PKEY is supported and CR4 is switched
6512 	 * back on host, so it is safe to read guest PKRU from current
6513 	 * XSAVE.
6514 	 */
6515 	if (static_cpu_has(X86_FEATURE_PKU) &&
6516 	    kvm_read_cr4_bits(vcpu, X86_CR4_PKE)) {
6517 		vcpu->arch.pkru = __read_pkru();
6518 		if (vcpu->arch.pkru != vmx->host_pkru)
6519 			__write_pkru(vmx->host_pkru);
6520 	}
6521 
6522 	vmx->nested.nested_run_pending = 0;
6523 	vmx->idt_vectoring_info = 0;
6524 
6525 	vmx->exit_reason = vmx->fail ? 0xdead : vmcs_read32(VM_EXIT_REASON);
6526 	if (vmx->fail || (vmx->exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY))
6527 		return;
6528 
6529 	vmx->loaded_vmcs->launched = 1;
6530 	vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
6531 
6532 	vmx_complete_atomic_exit(vmx);
6533 	vmx_recover_nmi_blocking(vmx);
6534 	vmx_complete_interrupts(vmx);
6535 }
6536 
6537 static struct kvm *vmx_vm_alloc(void)
6538 {
6539 	struct kvm_vmx *kvm_vmx = __vmalloc(sizeof(struct kvm_vmx),
6540 					    GFP_KERNEL_ACCOUNT | __GFP_ZERO,
6541 					    PAGE_KERNEL);
6542 	return &kvm_vmx->kvm;
6543 }
6544 
6545 static void vmx_vm_free(struct kvm *kvm)
6546 {
6547 	vfree(to_kvm_vmx(kvm));
6548 }
6549 
6550 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
6551 {
6552 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6553 
6554 	if (enable_pml)
6555 		vmx_destroy_pml_buffer(vmx);
6556 	free_vpid(vmx->vpid);
6557 	nested_vmx_free_vcpu(vcpu);
6558 	free_loaded_vmcs(vmx->loaded_vmcs);
6559 	kfree(vmx->guest_msrs);
6560 	kvm_vcpu_uninit(vcpu);
6561 	kmem_cache_free(x86_fpu_cache, vmx->vcpu.arch.guest_fpu);
6562 	kmem_cache_free(kvm_vcpu_cache, vmx);
6563 }
6564 
6565 static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
6566 {
6567 	int err;
6568 	struct vcpu_vmx *vmx;
6569 	unsigned long *msr_bitmap;
6570 	int cpu;
6571 
6572 	vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL_ACCOUNT);
6573 	if (!vmx)
6574 		return ERR_PTR(-ENOMEM);
6575 
6576 	vmx->vcpu.arch.guest_fpu = kmem_cache_zalloc(x86_fpu_cache,
6577 			GFP_KERNEL_ACCOUNT);
6578 	if (!vmx->vcpu.arch.guest_fpu) {
6579 		printk(KERN_ERR "kvm: failed to allocate vcpu's fpu\n");
6580 		err = -ENOMEM;
6581 		goto free_partial_vcpu;
6582 	}
6583 
6584 	vmx->vpid = allocate_vpid();
6585 
6586 	err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
6587 	if (err)
6588 		goto free_vcpu;
6589 
6590 	err = -ENOMEM;
6591 
6592 	/*
6593 	 * If PML is turned on, failure on enabling PML just results in failure
6594 	 * of creating the vcpu, therefore we can simplify PML logic (by
6595 	 * avoiding dealing with cases, such as enabling PML partially on vcpus
6596 	 * for the guest, etc.
6597 	 */
6598 	if (enable_pml) {
6599 		vmx->pml_pg = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
6600 		if (!vmx->pml_pg)
6601 			goto uninit_vcpu;
6602 	}
6603 
6604 	vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL_ACCOUNT);
6605 	BUILD_BUG_ON(ARRAY_SIZE(vmx_msr_index) * sizeof(vmx->guest_msrs[0])
6606 		     > PAGE_SIZE);
6607 
6608 	if (!vmx->guest_msrs)
6609 		goto free_pml;
6610 
6611 	err = alloc_loaded_vmcs(&vmx->vmcs01);
6612 	if (err < 0)
6613 		goto free_msrs;
6614 
6615 	msr_bitmap = vmx->vmcs01.msr_bitmap;
6616 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_TSC, MSR_TYPE_R);
6617 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_FS_BASE, MSR_TYPE_RW);
6618 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_GS_BASE, MSR_TYPE_RW);
6619 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_KERNEL_GS_BASE, MSR_TYPE_RW);
6620 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_CS, MSR_TYPE_RW);
6621 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_ESP, MSR_TYPE_RW);
6622 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_EIP, MSR_TYPE_RW);
6623 	vmx->msr_bitmap_mode = 0;
6624 
6625 	vmx->loaded_vmcs = &vmx->vmcs01;
6626 	cpu = get_cpu();
6627 	vmx_vcpu_load(&vmx->vcpu, cpu);
6628 	vmx->vcpu.cpu = cpu;
6629 	vmx_vcpu_setup(vmx);
6630 	vmx_vcpu_put(&vmx->vcpu);
6631 	put_cpu();
6632 	if (cpu_need_virtualize_apic_accesses(&vmx->vcpu)) {
6633 		err = alloc_apic_access_page(kvm);
6634 		if (err)
6635 			goto free_vmcs;
6636 	}
6637 
6638 	if (enable_ept && !enable_unrestricted_guest) {
6639 		err = init_rmode_identity_map(kvm);
6640 		if (err)
6641 			goto free_vmcs;
6642 	}
6643 
6644 	if (nested)
6645 		nested_vmx_setup_ctls_msrs(&vmx->nested.msrs,
6646 					   vmx_capability.ept,
6647 					   kvm_vcpu_apicv_active(&vmx->vcpu));
6648 	else
6649 		memset(&vmx->nested.msrs, 0, sizeof(vmx->nested.msrs));
6650 
6651 	vmx->nested.posted_intr_nv = -1;
6652 	vmx->nested.current_vmptr = -1ull;
6653 
6654 	vmx->msr_ia32_feature_control_valid_bits = FEATURE_CONTROL_LOCKED;
6655 
6656 	/*
6657 	 * Enforce invariant: pi_desc.nv is always either POSTED_INTR_VECTOR
6658 	 * or POSTED_INTR_WAKEUP_VECTOR.
6659 	 */
6660 	vmx->pi_desc.nv = POSTED_INTR_VECTOR;
6661 	vmx->pi_desc.sn = 1;
6662 
6663 	vmx->ept_pointer = INVALID_PAGE;
6664 
6665 	return &vmx->vcpu;
6666 
6667 free_vmcs:
6668 	free_loaded_vmcs(vmx->loaded_vmcs);
6669 free_msrs:
6670 	kfree(vmx->guest_msrs);
6671 free_pml:
6672 	vmx_destroy_pml_buffer(vmx);
6673 uninit_vcpu:
6674 	kvm_vcpu_uninit(&vmx->vcpu);
6675 free_vcpu:
6676 	free_vpid(vmx->vpid);
6677 	kmem_cache_free(x86_fpu_cache, vmx->vcpu.arch.guest_fpu);
6678 free_partial_vcpu:
6679 	kmem_cache_free(kvm_vcpu_cache, vmx);
6680 	return ERR_PTR(err);
6681 }
6682 
6683 #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/l1tf.html for details.\n"
6684 #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/l1tf.html for details.\n"
6685 
6686 static int vmx_vm_init(struct kvm *kvm)
6687 {
6688 	spin_lock_init(&to_kvm_vmx(kvm)->ept_pointer_lock);
6689 
6690 	if (!ple_gap)
6691 		kvm->arch.pause_in_guest = true;
6692 
6693 	if (boot_cpu_has(X86_BUG_L1TF) && enable_ept) {
6694 		switch (l1tf_mitigation) {
6695 		case L1TF_MITIGATION_OFF:
6696 		case L1TF_MITIGATION_FLUSH_NOWARN:
6697 			/* 'I explicitly don't care' is set */
6698 			break;
6699 		case L1TF_MITIGATION_FLUSH:
6700 		case L1TF_MITIGATION_FLUSH_NOSMT:
6701 		case L1TF_MITIGATION_FULL:
6702 			/*
6703 			 * Warn upon starting the first VM in a potentially
6704 			 * insecure environment.
6705 			 */
6706 			if (sched_smt_active())
6707 				pr_warn_once(L1TF_MSG_SMT);
6708 			if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_NEVER)
6709 				pr_warn_once(L1TF_MSG_L1D);
6710 			break;
6711 		case L1TF_MITIGATION_FULL_FORCE:
6712 			/* Flush is enforced */
6713 			break;
6714 		}
6715 	}
6716 	return 0;
6717 }
6718 
6719 static void __init vmx_check_processor_compat(void *rtn)
6720 {
6721 	struct vmcs_config vmcs_conf;
6722 	struct vmx_capability vmx_cap;
6723 
6724 	*(int *)rtn = 0;
6725 	if (setup_vmcs_config(&vmcs_conf, &vmx_cap) < 0)
6726 		*(int *)rtn = -EIO;
6727 	if (nested)
6728 		nested_vmx_setup_ctls_msrs(&vmcs_conf.nested, vmx_cap.ept,
6729 					   enable_apicv);
6730 	if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
6731 		printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
6732 				smp_processor_id());
6733 		*(int *)rtn = -EIO;
6734 	}
6735 }
6736 
6737 static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
6738 {
6739 	u8 cache;
6740 	u64 ipat = 0;
6741 
6742 	/* For VT-d and EPT combination
6743 	 * 1. MMIO: always map as UC
6744 	 * 2. EPT with VT-d:
6745 	 *   a. VT-d without snooping control feature: can't guarantee the
6746 	 *	result, try to trust guest.
6747 	 *   b. VT-d with snooping control feature: snooping control feature of
6748 	 *	VT-d engine can guarantee the cache correctness. Just set it
6749 	 *	to WB to keep consistent with host. So the same as item 3.
6750 	 * 3. EPT without VT-d: always map as WB and set IPAT=1 to keep
6751 	 *    consistent with host MTRR
6752 	 */
6753 	if (is_mmio) {
6754 		cache = MTRR_TYPE_UNCACHABLE;
6755 		goto exit;
6756 	}
6757 
6758 	if (!kvm_arch_has_noncoherent_dma(vcpu->kvm)) {
6759 		ipat = VMX_EPT_IPAT_BIT;
6760 		cache = MTRR_TYPE_WRBACK;
6761 		goto exit;
6762 	}
6763 
6764 	if (kvm_read_cr0(vcpu) & X86_CR0_CD) {
6765 		ipat = VMX_EPT_IPAT_BIT;
6766 		if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
6767 			cache = MTRR_TYPE_WRBACK;
6768 		else
6769 			cache = MTRR_TYPE_UNCACHABLE;
6770 		goto exit;
6771 	}
6772 
6773 	cache = kvm_mtrr_get_guest_memory_type(vcpu, gfn);
6774 
6775 exit:
6776 	return (cache << VMX_EPT_MT_EPTE_SHIFT) | ipat;
6777 }
6778 
6779 static int vmx_get_lpage_level(void)
6780 {
6781 	if (enable_ept && !cpu_has_vmx_ept_1g_page())
6782 		return PT_DIRECTORY_LEVEL;
6783 	else
6784 		/* For shadow and EPT supported 1GB page */
6785 		return PT_PDPE_LEVEL;
6786 }
6787 
6788 static void vmcs_set_secondary_exec_control(u32 new_ctl)
6789 {
6790 	/*
6791 	 * These bits in the secondary execution controls field
6792 	 * are dynamic, the others are mostly based on the hypervisor
6793 	 * architecture and the guest's CPUID.  Do not touch the
6794 	 * dynamic bits.
6795 	 */
6796 	u32 mask =
6797 		SECONDARY_EXEC_SHADOW_VMCS |
6798 		SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
6799 		SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
6800 		SECONDARY_EXEC_DESC;
6801 
6802 	u32 cur_ctl = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
6803 
6804 	vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
6805 		     (new_ctl & ~mask) | (cur_ctl & mask));
6806 }
6807 
6808 /*
6809  * Generate MSR_IA32_VMX_CR{0,4}_FIXED1 according to CPUID. Only set bits
6810  * (indicating "allowed-1") if they are supported in the guest's CPUID.
6811  */
6812 static void nested_vmx_cr_fixed1_bits_update(struct kvm_vcpu *vcpu)
6813 {
6814 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6815 	struct kvm_cpuid_entry2 *entry;
6816 
6817 	vmx->nested.msrs.cr0_fixed1 = 0xffffffff;
6818 	vmx->nested.msrs.cr4_fixed1 = X86_CR4_PCE;
6819 
6820 #define cr4_fixed1_update(_cr4_mask, _reg, _cpuid_mask) do {		\
6821 	if (entry && (entry->_reg & (_cpuid_mask)))			\
6822 		vmx->nested.msrs.cr4_fixed1 |= (_cr4_mask);	\
6823 } while (0)
6824 
6825 	entry = kvm_find_cpuid_entry(vcpu, 0x1, 0);
6826 	cr4_fixed1_update(X86_CR4_VME,        edx, bit(X86_FEATURE_VME));
6827 	cr4_fixed1_update(X86_CR4_PVI,        edx, bit(X86_FEATURE_VME));
6828 	cr4_fixed1_update(X86_CR4_TSD,        edx, bit(X86_FEATURE_TSC));
6829 	cr4_fixed1_update(X86_CR4_DE,         edx, bit(X86_FEATURE_DE));
6830 	cr4_fixed1_update(X86_CR4_PSE,        edx, bit(X86_FEATURE_PSE));
6831 	cr4_fixed1_update(X86_CR4_PAE,        edx, bit(X86_FEATURE_PAE));
6832 	cr4_fixed1_update(X86_CR4_MCE,        edx, bit(X86_FEATURE_MCE));
6833 	cr4_fixed1_update(X86_CR4_PGE,        edx, bit(X86_FEATURE_PGE));
6834 	cr4_fixed1_update(X86_CR4_OSFXSR,     edx, bit(X86_FEATURE_FXSR));
6835 	cr4_fixed1_update(X86_CR4_OSXMMEXCPT, edx, bit(X86_FEATURE_XMM));
6836 	cr4_fixed1_update(X86_CR4_VMXE,       ecx, bit(X86_FEATURE_VMX));
6837 	cr4_fixed1_update(X86_CR4_SMXE,       ecx, bit(X86_FEATURE_SMX));
6838 	cr4_fixed1_update(X86_CR4_PCIDE,      ecx, bit(X86_FEATURE_PCID));
6839 	cr4_fixed1_update(X86_CR4_OSXSAVE,    ecx, bit(X86_FEATURE_XSAVE));
6840 
6841 	entry = kvm_find_cpuid_entry(vcpu, 0x7, 0);
6842 	cr4_fixed1_update(X86_CR4_FSGSBASE,   ebx, bit(X86_FEATURE_FSGSBASE));
6843 	cr4_fixed1_update(X86_CR4_SMEP,       ebx, bit(X86_FEATURE_SMEP));
6844 	cr4_fixed1_update(X86_CR4_SMAP,       ebx, bit(X86_FEATURE_SMAP));
6845 	cr4_fixed1_update(X86_CR4_PKE,        ecx, bit(X86_FEATURE_PKU));
6846 	cr4_fixed1_update(X86_CR4_UMIP,       ecx, bit(X86_FEATURE_UMIP));
6847 
6848 #undef cr4_fixed1_update
6849 }
6850 
6851 static void nested_vmx_entry_exit_ctls_update(struct kvm_vcpu *vcpu)
6852 {
6853 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6854 
6855 	if (kvm_mpx_supported()) {
6856 		bool mpx_enabled = guest_cpuid_has(vcpu, X86_FEATURE_MPX);
6857 
6858 		if (mpx_enabled) {
6859 			vmx->nested.msrs.entry_ctls_high |= VM_ENTRY_LOAD_BNDCFGS;
6860 			vmx->nested.msrs.exit_ctls_high |= VM_EXIT_CLEAR_BNDCFGS;
6861 		} else {
6862 			vmx->nested.msrs.entry_ctls_high &= ~VM_ENTRY_LOAD_BNDCFGS;
6863 			vmx->nested.msrs.exit_ctls_high &= ~VM_EXIT_CLEAR_BNDCFGS;
6864 		}
6865 	}
6866 }
6867 
6868 static void update_intel_pt_cfg(struct kvm_vcpu *vcpu)
6869 {
6870 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6871 	struct kvm_cpuid_entry2 *best = NULL;
6872 	int i;
6873 
6874 	for (i = 0; i < PT_CPUID_LEAVES; i++) {
6875 		best = kvm_find_cpuid_entry(vcpu, 0x14, i);
6876 		if (!best)
6877 			return;
6878 		vmx->pt_desc.caps[CPUID_EAX + i*PT_CPUID_REGS_NUM] = best->eax;
6879 		vmx->pt_desc.caps[CPUID_EBX + i*PT_CPUID_REGS_NUM] = best->ebx;
6880 		vmx->pt_desc.caps[CPUID_ECX + i*PT_CPUID_REGS_NUM] = best->ecx;
6881 		vmx->pt_desc.caps[CPUID_EDX + i*PT_CPUID_REGS_NUM] = best->edx;
6882 	}
6883 
6884 	/* Get the number of configurable Address Ranges for filtering */
6885 	vmx->pt_desc.addr_range = intel_pt_validate_cap(vmx->pt_desc.caps,
6886 						PT_CAP_num_address_ranges);
6887 
6888 	/* Initialize and clear the no dependency bits */
6889 	vmx->pt_desc.ctl_bitmask = ~(RTIT_CTL_TRACEEN | RTIT_CTL_OS |
6890 			RTIT_CTL_USR | RTIT_CTL_TSC_EN | RTIT_CTL_DISRETC);
6891 
6892 	/*
6893 	 * If CPUID.(EAX=14H,ECX=0):EBX[0]=1 CR3Filter can be set otherwise
6894 	 * will inject an #GP
6895 	 */
6896 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_cr3_filtering))
6897 		vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_CR3EN;
6898 
6899 	/*
6900 	 * If CPUID.(EAX=14H,ECX=0):EBX[1]=1 CYCEn, CycThresh and
6901 	 * PSBFreq can be set
6902 	 */
6903 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc))
6904 		vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_CYCLEACC |
6905 				RTIT_CTL_CYC_THRESH | RTIT_CTL_PSB_FREQ);
6906 
6907 	/*
6908 	 * If CPUID.(EAX=14H,ECX=0):EBX[3]=1 MTCEn BranchEn and
6909 	 * MTCFreq can be set
6910 	 */
6911 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc))
6912 		vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_MTC_EN |
6913 				RTIT_CTL_BRANCH_EN | RTIT_CTL_MTC_RANGE);
6914 
6915 	/* If CPUID.(EAX=14H,ECX=0):EBX[4]=1 FUPonPTW and PTWEn can be set */
6916 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_ptwrite))
6917 		vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_FUP_ON_PTW |
6918 							RTIT_CTL_PTW_EN);
6919 
6920 	/* If CPUID.(EAX=14H,ECX=0):EBX[5]=1 PwrEvEn can be set */
6921 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_power_event_trace))
6922 		vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_PWR_EVT_EN;
6923 
6924 	/* If CPUID.(EAX=14H,ECX=0):ECX[0]=1 ToPA can be set */
6925 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_topa_output))
6926 		vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_TOPA;
6927 
6928 	/* If CPUID.(EAX=14H,ECX=0):ECX[3]=1 FabircEn can be set */
6929 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_output_subsys))
6930 		vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_FABRIC_EN;
6931 
6932 	/* unmask address range configure area */
6933 	for (i = 0; i < vmx->pt_desc.addr_range; i++)
6934 		vmx->pt_desc.ctl_bitmask &= ~(0xfULL << (32 + i * 4));
6935 }
6936 
6937 static void vmx_cpuid_update(struct kvm_vcpu *vcpu)
6938 {
6939 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6940 
6941 	if (cpu_has_secondary_exec_ctrls()) {
6942 		vmx_compute_secondary_exec_control(vmx);
6943 		vmcs_set_secondary_exec_control(vmx->secondary_exec_control);
6944 	}
6945 
6946 	if (nested_vmx_allowed(vcpu))
6947 		to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
6948 			FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
6949 	else
6950 		to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
6951 			~FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
6952 
6953 	if (nested_vmx_allowed(vcpu)) {
6954 		nested_vmx_cr_fixed1_bits_update(vcpu);
6955 		nested_vmx_entry_exit_ctls_update(vcpu);
6956 	}
6957 
6958 	if (boot_cpu_has(X86_FEATURE_INTEL_PT) &&
6959 			guest_cpuid_has(vcpu, X86_FEATURE_INTEL_PT))
6960 		update_intel_pt_cfg(vcpu);
6961 }
6962 
6963 static void vmx_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
6964 {
6965 	if (func == 1 && nested)
6966 		entry->ecx |= bit(X86_FEATURE_VMX);
6967 }
6968 
6969 static void vmx_request_immediate_exit(struct kvm_vcpu *vcpu)
6970 {
6971 	to_vmx(vcpu)->req_immediate_exit = true;
6972 }
6973 
6974 static int vmx_check_intercept(struct kvm_vcpu *vcpu,
6975 			       struct x86_instruction_info *info,
6976 			       enum x86_intercept_stage stage)
6977 {
6978 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6979 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6980 
6981 	/*
6982 	 * RDPID causes #UD if disabled through secondary execution controls.
6983 	 * Because it is marked as EmulateOnUD, we need to intercept it here.
6984 	 */
6985 	if (info->intercept == x86_intercept_rdtscp &&
6986 	    !nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDTSCP)) {
6987 		ctxt->exception.vector = UD_VECTOR;
6988 		ctxt->exception.error_code_valid = false;
6989 		return X86EMUL_PROPAGATE_FAULT;
6990 	}
6991 
6992 	/* TODO: check more intercepts... */
6993 	return X86EMUL_CONTINUE;
6994 }
6995 
6996 #ifdef CONFIG_X86_64
6997 /* (a << shift) / divisor, return 1 if overflow otherwise 0 */
6998 static inline int u64_shl_div_u64(u64 a, unsigned int shift,
6999 				  u64 divisor, u64 *result)
7000 {
7001 	u64 low = a << shift, high = a >> (64 - shift);
7002 
7003 	/* To avoid the overflow on divq */
7004 	if (high >= divisor)
7005 		return 1;
7006 
7007 	/* Low hold the result, high hold rem which is discarded */
7008 	asm("divq %2\n\t" : "=a" (low), "=d" (high) :
7009 	    "rm" (divisor), "0" (low), "1" (high));
7010 	*result = low;
7011 
7012 	return 0;
7013 }
7014 
7015 static int vmx_set_hv_timer(struct kvm_vcpu *vcpu, u64 guest_deadline_tsc)
7016 {
7017 	struct vcpu_vmx *vmx;
7018 	u64 tscl, guest_tscl, delta_tsc, lapic_timer_advance_cycles;
7019 
7020 	if (kvm_mwait_in_guest(vcpu->kvm))
7021 		return -EOPNOTSUPP;
7022 
7023 	vmx = to_vmx(vcpu);
7024 	tscl = rdtsc();
7025 	guest_tscl = kvm_read_l1_tsc(vcpu, tscl);
7026 	delta_tsc = max(guest_deadline_tsc, guest_tscl) - guest_tscl;
7027 	lapic_timer_advance_cycles = nsec_to_cycles(vcpu, lapic_timer_advance_ns);
7028 
7029 	if (delta_tsc > lapic_timer_advance_cycles)
7030 		delta_tsc -= lapic_timer_advance_cycles;
7031 	else
7032 		delta_tsc = 0;
7033 
7034 	/* Convert to host delta tsc if tsc scaling is enabled */
7035 	if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio &&
7036 			u64_shl_div_u64(delta_tsc,
7037 				kvm_tsc_scaling_ratio_frac_bits,
7038 				vcpu->arch.tsc_scaling_ratio,
7039 				&delta_tsc))
7040 		return -ERANGE;
7041 
7042 	/*
7043 	 * If the delta tsc can't fit in the 32 bit after the multi shift,
7044 	 * we can't use the preemption timer.
7045 	 * It's possible that it fits on later vmentries, but checking
7046 	 * on every vmentry is costly so we just use an hrtimer.
7047 	 */
7048 	if (delta_tsc >> (cpu_preemption_timer_multi + 32))
7049 		return -ERANGE;
7050 
7051 	vmx->hv_deadline_tsc = tscl + delta_tsc;
7052 	return delta_tsc == 0;
7053 }
7054 
7055 static void vmx_cancel_hv_timer(struct kvm_vcpu *vcpu)
7056 {
7057 	to_vmx(vcpu)->hv_deadline_tsc = -1;
7058 }
7059 #endif
7060 
7061 static void vmx_sched_in(struct kvm_vcpu *vcpu, int cpu)
7062 {
7063 	if (!kvm_pause_in_guest(vcpu->kvm))
7064 		shrink_ple_window(vcpu);
7065 }
7066 
7067 static void vmx_slot_enable_log_dirty(struct kvm *kvm,
7068 				     struct kvm_memory_slot *slot)
7069 {
7070 	kvm_mmu_slot_leaf_clear_dirty(kvm, slot);
7071 	kvm_mmu_slot_largepage_remove_write_access(kvm, slot);
7072 }
7073 
7074 static void vmx_slot_disable_log_dirty(struct kvm *kvm,
7075 				       struct kvm_memory_slot *slot)
7076 {
7077 	kvm_mmu_slot_set_dirty(kvm, slot);
7078 }
7079 
7080 static void vmx_flush_log_dirty(struct kvm *kvm)
7081 {
7082 	kvm_flush_pml_buffers(kvm);
7083 }
7084 
7085 static int vmx_write_pml_buffer(struct kvm_vcpu *vcpu)
7086 {
7087 	struct vmcs12 *vmcs12;
7088 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7089 	gpa_t gpa;
7090 	struct page *page = NULL;
7091 	u64 *pml_address;
7092 
7093 	if (is_guest_mode(vcpu)) {
7094 		WARN_ON_ONCE(vmx->nested.pml_full);
7095 
7096 		/*
7097 		 * Check if PML is enabled for the nested guest.
7098 		 * Whether eptp bit 6 is set is already checked
7099 		 * as part of A/D emulation.
7100 		 */
7101 		vmcs12 = get_vmcs12(vcpu);
7102 		if (!nested_cpu_has_pml(vmcs12))
7103 			return 0;
7104 
7105 		if (vmcs12->guest_pml_index >= PML_ENTITY_NUM) {
7106 			vmx->nested.pml_full = true;
7107 			return 1;
7108 		}
7109 
7110 		gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS) & ~0xFFFull;
7111 
7112 		page = kvm_vcpu_gpa_to_page(vcpu, vmcs12->pml_address);
7113 		if (is_error_page(page))
7114 			return 0;
7115 
7116 		pml_address = kmap(page);
7117 		pml_address[vmcs12->guest_pml_index--] = gpa;
7118 		kunmap(page);
7119 		kvm_release_page_clean(page);
7120 	}
7121 
7122 	return 0;
7123 }
7124 
7125 static void vmx_enable_log_dirty_pt_masked(struct kvm *kvm,
7126 					   struct kvm_memory_slot *memslot,
7127 					   gfn_t offset, unsigned long mask)
7128 {
7129 	kvm_mmu_clear_dirty_pt_masked(kvm, memslot, offset, mask);
7130 }
7131 
7132 static void __pi_post_block(struct kvm_vcpu *vcpu)
7133 {
7134 	struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
7135 	struct pi_desc old, new;
7136 	unsigned int dest;
7137 
7138 	do {
7139 		old.control = new.control = pi_desc->control;
7140 		WARN(old.nv != POSTED_INTR_WAKEUP_VECTOR,
7141 		     "Wakeup handler not enabled while the VCPU is blocked\n");
7142 
7143 		dest = cpu_physical_id(vcpu->cpu);
7144 
7145 		if (x2apic_enabled())
7146 			new.ndst = dest;
7147 		else
7148 			new.ndst = (dest << 8) & 0xFF00;
7149 
7150 		/* set 'NV' to 'notification vector' */
7151 		new.nv = POSTED_INTR_VECTOR;
7152 	} while (cmpxchg64(&pi_desc->control, old.control,
7153 			   new.control) != old.control);
7154 
7155 	if (!WARN_ON_ONCE(vcpu->pre_pcpu == -1)) {
7156 		spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7157 		list_del(&vcpu->blocked_vcpu_list);
7158 		spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7159 		vcpu->pre_pcpu = -1;
7160 	}
7161 }
7162 
7163 /*
7164  * This routine does the following things for vCPU which is going
7165  * to be blocked if VT-d PI is enabled.
7166  * - Store the vCPU to the wakeup list, so when interrupts happen
7167  *   we can find the right vCPU to wake up.
7168  * - Change the Posted-interrupt descriptor as below:
7169  *      'NDST' <-- vcpu->pre_pcpu
7170  *      'NV' <-- POSTED_INTR_WAKEUP_VECTOR
7171  * - If 'ON' is set during this process, which means at least one
7172  *   interrupt is posted for this vCPU, we cannot block it, in
7173  *   this case, return 1, otherwise, return 0.
7174  *
7175  */
7176 static int pi_pre_block(struct kvm_vcpu *vcpu)
7177 {
7178 	unsigned int dest;
7179 	struct pi_desc old, new;
7180 	struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
7181 
7182 	if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
7183 		!irq_remapping_cap(IRQ_POSTING_CAP)  ||
7184 		!kvm_vcpu_apicv_active(vcpu))
7185 		return 0;
7186 
7187 	WARN_ON(irqs_disabled());
7188 	local_irq_disable();
7189 	if (!WARN_ON_ONCE(vcpu->pre_pcpu != -1)) {
7190 		vcpu->pre_pcpu = vcpu->cpu;
7191 		spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7192 		list_add_tail(&vcpu->blocked_vcpu_list,
7193 			      &per_cpu(blocked_vcpu_on_cpu,
7194 				       vcpu->pre_pcpu));
7195 		spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7196 	}
7197 
7198 	do {
7199 		old.control = new.control = pi_desc->control;
7200 
7201 		WARN((pi_desc->sn == 1),
7202 		     "Warning: SN field of posted-interrupts "
7203 		     "is set before blocking\n");
7204 
7205 		/*
7206 		 * Since vCPU can be preempted during this process,
7207 		 * vcpu->cpu could be different with pre_pcpu, we
7208 		 * need to set pre_pcpu as the destination of wakeup
7209 		 * notification event, then we can find the right vCPU
7210 		 * to wakeup in wakeup handler if interrupts happen
7211 		 * when the vCPU is in blocked state.
7212 		 */
7213 		dest = cpu_physical_id(vcpu->pre_pcpu);
7214 
7215 		if (x2apic_enabled())
7216 			new.ndst = dest;
7217 		else
7218 			new.ndst = (dest << 8) & 0xFF00;
7219 
7220 		/* set 'NV' to 'wakeup vector' */
7221 		new.nv = POSTED_INTR_WAKEUP_VECTOR;
7222 	} while (cmpxchg64(&pi_desc->control, old.control,
7223 			   new.control) != old.control);
7224 
7225 	/* We should not block the vCPU if an interrupt is posted for it.  */
7226 	if (pi_test_on(pi_desc) == 1)
7227 		__pi_post_block(vcpu);
7228 
7229 	local_irq_enable();
7230 	return (vcpu->pre_pcpu == -1);
7231 }
7232 
7233 static int vmx_pre_block(struct kvm_vcpu *vcpu)
7234 {
7235 	if (pi_pre_block(vcpu))
7236 		return 1;
7237 
7238 	if (kvm_lapic_hv_timer_in_use(vcpu))
7239 		kvm_lapic_switch_to_sw_timer(vcpu);
7240 
7241 	return 0;
7242 }
7243 
7244 static void pi_post_block(struct kvm_vcpu *vcpu)
7245 {
7246 	if (vcpu->pre_pcpu == -1)
7247 		return;
7248 
7249 	WARN_ON(irqs_disabled());
7250 	local_irq_disable();
7251 	__pi_post_block(vcpu);
7252 	local_irq_enable();
7253 }
7254 
7255 static void vmx_post_block(struct kvm_vcpu *vcpu)
7256 {
7257 	if (kvm_x86_ops->set_hv_timer)
7258 		kvm_lapic_switch_to_hv_timer(vcpu);
7259 
7260 	pi_post_block(vcpu);
7261 }
7262 
7263 /*
7264  * vmx_update_pi_irte - set IRTE for Posted-Interrupts
7265  *
7266  * @kvm: kvm
7267  * @host_irq: host irq of the interrupt
7268  * @guest_irq: gsi of the interrupt
7269  * @set: set or unset PI
7270  * returns 0 on success, < 0 on failure
7271  */
7272 static int vmx_update_pi_irte(struct kvm *kvm, unsigned int host_irq,
7273 			      uint32_t guest_irq, bool set)
7274 {
7275 	struct kvm_kernel_irq_routing_entry *e;
7276 	struct kvm_irq_routing_table *irq_rt;
7277 	struct kvm_lapic_irq irq;
7278 	struct kvm_vcpu *vcpu;
7279 	struct vcpu_data vcpu_info;
7280 	int idx, ret = 0;
7281 
7282 	if (!kvm_arch_has_assigned_device(kvm) ||
7283 		!irq_remapping_cap(IRQ_POSTING_CAP) ||
7284 		!kvm_vcpu_apicv_active(kvm->vcpus[0]))
7285 		return 0;
7286 
7287 	idx = srcu_read_lock(&kvm->irq_srcu);
7288 	irq_rt = srcu_dereference(kvm->irq_routing, &kvm->irq_srcu);
7289 	if (guest_irq >= irq_rt->nr_rt_entries ||
7290 	    hlist_empty(&irq_rt->map[guest_irq])) {
7291 		pr_warn_once("no route for guest_irq %u/%u (broken user space?)\n",
7292 			     guest_irq, irq_rt->nr_rt_entries);
7293 		goto out;
7294 	}
7295 
7296 	hlist_for_each_entry(e, &irq_rt->map[guest_irq], link) {
7297 		if (e->type != KVM_IRQ_ROUTING_MSI)
7298 			continue;
7299 		/*
7300 		 * VT-d PI cannot support posting multicast/broadcast
7301 		 * interrupts to a vCPU, we still use interrupt remapping
7302 		 * for these kind of interrupts.
7303 		 *
7304 		 * For lowest-priority interrupts, we only support
7305 		 * those with single CPU as the destination, e.g. user
7306 		 * configures the interrupts via /proc/irq or uses
7307 		 * irqbalance to make the interrupts single-CPU.
7308 		 *
7309 		 * We will support full lowest-priority interrupt later.
7310 		 */
7311 
7312 		kvm_set_msi_irq(kvm, e, &irq);
7313 		if (!kvm_intr_is_single_vcpu(kvm, &irq, &vcpu)) {
7314 			/*
7315 			 * Make sure the IRTE is in remapped mode if
7316 			 * we don't handle it in posted mode.
7317 			 */
7318 			ret = irq_set_vcpu_affinity(host_irq, NULL);
7319 			if (ret < 0) {
7320 				printk(KERN_INFO
7321 				   "failed to back to remapped mode, irq: %u\n",
7322 				   host_irq);
7323 				goto out;
7324 			}
7325 
7326 			continue;
7327 		}
7328 
7329 		vcpu_info.pi_desc_addr = __pa(vcpu_to_pi_desc(vcpu));
7330 		vcpu_info.vector = irq.vector;
7331 
7332 		trace_kvm_pi_irte_update(host_irq, vcpu->vcpu_id, e->gsi,
7333 				vcpu_info.vector, vcpu_info.pi_desc_addr, set);
7334 
7335 		if (set)
7336 			ret = irq_set_vcpu_affinity(host_irq, &vcpu_info);
7337 		else
7338 			ret = irq_set_vcpu_affinity(host_irq, NULL);
7339 
7340 		if (ret < 0) {
7341 			printk(KERN_INFO "%s: failed to update PI IRTE\n",
7342 					__func__);
7343 			goto out;
7344 		}
7345 	}
7346 
7347 	ret = 0;
7348 out:
7349 	srcu_read_unlock(&kvm->irq_srcu, idx);
7350 	return ret;
7351 }
7352 
7353 static void vmx_setup_mce(struct kvm_vcpu *vcpu)
7354 {
7355 	if (vcpu->arch.mcg_cap & MCG_LMCE_P)
7356 		to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
7357 			FEATURE_CONTROL_LMCE;
7358 	else
7359 		to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
7360 			~FEATURE_CONTROL_LMCE;
7361 }
7362 
7363 static int vmx_smi_allowed(struct kvm_vcpu *vcpu)
7364 {
7365 	/* we need a nested vmexit to enter SMM, postpone if run is pending */
7366 	if (to_vmx(vcpu)->nested.nested_run_pending)
7367 		return 0;
7368 	return 1;
7369 }
7370 
7371 static int vmx_pre_enter_smm(struct kvm_vcpu *vcpu, char *smstate)
7372 {
7373 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7374 
7375 	vmx->nested.smm.guest_mode = is_guest_mode(vcpu);
7376 	if (vmx->nested.smm.guest_mode)
7377 		nested_vmx_vmexit(vcpu, -1, 0, 0);
7378 
7379 	vmx->nested.smm.vmxon = vmx->nested.vmxon;
7380 	vmx->nested.vmxon = false;
7381 	vmx_clear_hlt(vcpu);
7382 	return 0;
7383 }
7384 
7385 static int vmx_pre_leave_smm(struct kvm_vcpu *vcpu, u64 smbase)
7386 {
7387 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7388 	int ret;
7389 
7390 	if (vmx->nested.smm.vmxon) {
7391 		vmx->nested.vmxon = true;
7392 		vmx->nested.smm.vmxon = false;
7393 	}
7394 
7395 	if (vmx->nested.smm.guest_mode) {
7396 		vcpu->arch.hflags &= ~HF_SMM_MASK;
7397 		ret = nested_vmx_enter_non_root_mode(vcpu, false);
7398 		vcpu->arch.hflags |= HF_SMM_MASK;
7399 		if (ret)
7400 			return ret;
7401 
7402 		vmx->nested.smm.guest_mode = false;
7403 	}
7404 	return 0;
7405 }
7406 
7407 static int enable_smi_window(struct kvm_vcpu *vcpu)
7408 {
7409 	return 0;
7410 }
7411 
7412 static __init int hardware_setup(void)
7413 {
7414 	unsigned long host_bndcfgs;
7415 	int r, i;
7416 
7417 	rdmsrl_safe(MSR_EFER, &host_efer);
7418 
7419 	for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i)
7420 		kvm_define_shared_msr(i, vmx_msr_index[i]);
7421 
7422 	if (setup_vmcs_config(&vmcs_config, &vmx_capability) < 0)
7423 		return -EIO;
7424 
7425 	if (boot_cpu_has(X86_FEATURE_NX))
7426 		kvm_enable_efer_bits(EFER_NX);
7427 
7428 	if (boot_cpu_has(X86_FEATURE_MPX)) {
7429 		rdmsrl(MSR_IA32_BNDCFGS, host_bndcfgs);
7430 		WARN_ONCE(host_bndcfgs, "KVM: BNDCFGS in host will be lost");
7431 	}
7432 
7433 	if (boot_cpu_has(X86_FEATURE_XSAVES))
7434 		rdmsrl(MSR_IA32_XSS, host_xss);
7435 
7436 	if (!cpu_has_vmx_vpid() || !cpu_has_vmx_invvpid() ||
7437 	    !(cpu_has_vmx_invvpid_single() || cpu_has_vmx_invvpid_global()))
7438 		enable_vpid = 0;
7439 
7440 	if (!cpu_has_vmx_ept() ||
7441 	    !cpu_has_vmx_ept_4levels() ||
7442 	    !cpu_has_vmx_ept_mt_wb() ||
7443 	    !cpu_has_vmx_invept_global())
7444 		enable_ept = 0;
7445 
7446 	if (!cpu_has_vmx_ept_ad_bits() || !enable_ept)
7447 		enable_ept_ad_bits = 0;
7448 
7449 	if (!cpu_has_vmx_unrestricted_guest() || !enable_ept)
7450 		enable_unrestricted_guest = 0;
7451 
7452 	if (!cpu_has_vmx_flexpriority())
7453 		flexpriority_enabled = 0;
7454 
7455 	if (!cpu_has_virtual_nmis())
7456 		enable_vnmi = 0;
7457 
7458 	/*
7459 	 * set_apic_access_page_addr() is used to reload apic access
7460 	 * page upon invalidation.  No need to do anything if not
7461 	 * using the APIC_ACCESS_ADDR VMCS field.
7462 	 */
7463 	if (!flexpriority_enabled)
7464 		kvm_x86_ops->set_apic_access_page_addr = NULL;
7465 
7466 	if (!cpu_has_vmx_tpr_shadow())
7467 		kvm_x86_ops->update_cr8_intercept = NULL;
7468 
7469 	if (enable_ept && !cpu_has_vmx_ept_2m_page())
7470 		kvm_disable_largepages();
7471 
7472 #if IS_ENABLED(CONFIG_HYPERV)
7473 	if (ms_hyperv.nested_features & HV_X64_NESTED_GUEST_MAPPING_FLUSH
7474 	    && enable_ept) {
7475 		kvm_x86_ops->tlb_remote_flush = hv_remote_flush_tlb;
7476 		kvm_x86_ops->tlb_remote_flush_with_range =
7477 				hv_remote_flush_tlb_with_range;
7478 	}
7479 #endif
7480 
7481 	if (!cpu_has_vmx_ple()) {
7482 		ple_gap = 0;
7483 		ple_window = 0;
7484 		ple_window_grow = 0;
7485 		ple_window_max = 0;
7486 		ple_window_shrink = 0;
7487 	}
7488 
7489 	if (!cpu_has_vmx_apicv()) {
7490 		enable_apicv = 0;
7491 		kvm_x86_ops->sync_pir_to_irr = NULL;
7492 	}
7493 
7494 	if (cpu_has_vmx_tsc_scaling()) {
7495 		kvm_has_tsc_control = true;
7496 		kvm_max_tsc_scaling_ratio = KVM_VMX_TSC_MULTIPLIER_MAX;
7497 		kvm_tsc_scaling_ratio_frac_bits = 48;
7498 	}
7499 
7500 	set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
7501 
7502 	if (enable_ept)
7503 		vmx_enable_tdp();
7504 	else
7505 		kvm_disable_tdp();
7506 
7507 	/*
7508 	 * Only enable PML when hardware supports PML feature, and both EPT
7509 	 * and EPT A/D bit features are enabled -- PML depends on them to work.
7510 	 */
7511 	if (!enable_ept || !enable_ept_ad_bits || !cpu_has_vmx_pml())
7512 		enable_pml = 0;
7513 
7514 	if (!enable_pml) {
7515 		kvm_x86_ops->slot_enable_log_dirty = NULL;
7516 		kvm_x86_ops->slot_disable_log_dirty = NULL;
7517 		kvm_x86_ops->flush_log_dirty = NULL;
7518 		kvm_x86_ops->enable_log_dirty_pt_masked = NULL;
7519 	}
7520 
7521 	if (!cpu_has_vmx_preemption_timer())
7522 		kvm_x86_ops->request_immediate_exit = __kvm_request_immediate_exit;
7523 
7524 	if (cpu_has_vmx_preemption_timer() && enable_preemption_timer) {
7525 		u64 vmx_msr;
7526 
7527 		rdmsrl(MSR_IA32_VMX_MISC, vmx_msr);
7528 		cpu_preemption_timer_multi =
7529 			vmx_msr & VMX_MISC_PREEMPTION_TIMER_RATE_MASK;
7530 	} else {
7531 		kvm_x86_ops->set_hv_timer = NULL;
7532 		kvm_x86_ops->cancel_hv_timer = NULL;
7533 	}
7534 
7535 	kvm_set_posted_intr_wakeup_handler(wakeup_handler);
7536 
7537 	kvm_mce_cap_supported |= MCG_LMCE_P;
7538 
7539 	if (pt_mode != PT_MODE_SYSTEM && pt_mode != PT_MODE_HOST_GUEST)
7540 		return -EINVAL;
7541 	if (!enable_ept || !cpu_has_vmx_intel_pt())
7542 		pt_mode = PT_MODE_SYSTEM;
7543 
7544 	if (nested) {
7545 		nested_vmx_setup_ctls_msrs(&vmcs_config.nested,
7546 					   vmx_capability.ept, enable_apicv);
7547 
7548 		r = nested_vmx_hardware_setup(kvm_vmx_exit_handlers);
7549 		if (r)
7550 			return r;
7551 	}
7552 
7553 	r = alloc_kvm_area();
7554 	if (r)
7555 		nested_vmx_hardware_unsetup();
7556 	return r;
7557 }
7558 
7559 static __exit void hardware_unsetup(void)
7560 {
7561 	if (nested)
7562 		nested_vmx_hardware_unsetup();
7563 
7564 	free_kvm_area();
7565 }
7566 
7567 static struct kvm_x86_ops vmx_x86_ops __ro_after_init = {
7568 	.cpu_has_kvm_support = cpu_has_kvm_support,
7569 	.disabled_by_bios = vmx_disabled_by_bios,
7570 	.hardware_setup = hardware_setup,
7571 	.hardware_unsetup = hardware_unsetup,
7572 	.check_processor_compatibility = vmx_check_processor_compat,
7573 	.hardware_enable = hardware_enable,
7574 	.hardware_disable = hardware_disable,
7575 	.cpu_has_accelerated_tpr = report_flexpriority,
7576 	.has_emulated_msr = vmx_has_emulated_msr,
7577 
7578 	.vm_init = vmx_vm_init,
7579 	.vm_alloc = vmx_vm_alloc,
7580 	.vm_free = vmx_vm_free,
7581 
7582 	.vcpu_create = vmx_create_vcpu,
7583 	.vcpu_free = vmx_free_vcpu,
7584 	.vcpu_reset = vmx_vcpu_reset,
7585 
7586 	.prepare_guest_switch = vmx_prepare_switch_to_guest,
7587 	.vcpu_load = vmx_vcpu_load,
7588 	.vcpu_put = vmx_vcpu_put,
7589 
7590 	.update_bp_intercept = update_exception_bitmap,
7591 	.get_msr_feature = vmx_get_msr_feature,
7592 	.get_msr = vmx_get_msr,
7593 	.set_msr = vmx_set_msr,
7594 	.get_segment_base = vmx_get_segment_base,
7595 	.get_segment = vmx_get_segment,
7596 	.set_segment = vmx_set_segment,
7597 	.get_cpl = vmx_get_cpl,
7598 	.get_cs_db_l_bits = vmx_get_cs_db_l_bits,
7599 	.decache_cr0_guest_bits = vmx_decache_cr0_guest_bits,
7600 	.decache_cr3 = vmx_decache_cr3,
7601 	.decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
7602 	.set_cr0 = vmx_set_cr0,
7603 	.set_cr3 = vmx_set_cr3,
7604 	.set_cr4 = vmx_set_cr4,
7605 	.set_efer = vmx_set_efer,
7606 	.get_idt = vmx_get_idt,
7607 	.set_idt = vmx_set_idt,
7608 	.get_gdt = vmx_get_gdt,
7609 	.set_gdt = vmx_set_gdt,
7610 	.get_dr6 = vmx_get_dr6,
7611 	.set_dr6 = vmx_set_dr6,
7612 	.set_dr7 = vmx_set_dr7,
7613 	.sync_dirty_debug_regs = vmx_sync_dirty_debug_regs,
7614 	.cache_reg = vmx_cache_reg,
7615 	.get_rflags = vmx_get_rflags,
7616 	.set_rflags = vmx_set_rflags,
7617 
7618 	.tlb_flush = vmx_flush_tlb,
7619 	.tlb_flush_gva = vmx_flush_tlb_gva,
7620 
7621 	.run = vmx_vcpu_run,
7622 	.handle_exit = vmx_handle_exit,
7623 	.skip_emulated_instruction = skip_emulated_instruction,
7624 	.set_interrupt_shadow = vmx_set_interrupt_shadow,
7625 	.get_interrupt_shadow = vmx_get_interrupt_shadow,
7626 	.patch_hypercall = vmx_patch_hypercall,
7627 	.set_irq = vmx_inject_irq,
7628 	.set_nmi = vmx_inject_nmi,
7629 	.queue_exception = vmx_queue_exception,
7630 	.cancel_injection = vmx_cancel_injection,
7631 	.interrupt_allowed = vmx_interrupt_allowed,
7632 	.nmi_allowed = vmx_nmi_allowed,
7633 	.get_nmi_mask = vmx_get_nmi_mask,
7634 	.set_nmi_mask = vmx_set_nmi_mask,
7635 	.enable_nmi_window = enable_nmi_window,
7636 	.enable_irq_window = enable_irq_window,
7637 	.update_cr8_intercept = update_cr8_intercept,
7638 	.set_virtual_apic_mode = vmx_set_virtual_apic_mode,
7639 	.set_apic_access_page_addr = vmx_set_apic_access_page_addr,
7640 	.get_enable_apicv = vmx_get_enable_apicv,
7641 	.refresh_apicv_exec_ctrl = vmx_refresh_apicv_exec_ctrl,
7642 	.load_eoi_exitmap = vmx_load_eoi_exitmap,
7643 	.apicv_post_state_restore = vmx_apicv_post_state_restore,
7644 	.hwapic_irr_update = vmx_hwapic_irr_update,
7645 	.hwapic_isr_update = vmx_hwapic_isr_update,
7646 	.guest_apic_has_interrupt = vmx_guest_apic_has_interrupt,
7647 	.sync_pir_to_irr = vmx_sync_pir_to_irr,
7648 	.deliver_posted_interrupt = vmx_deliver_posted_interrupt,
7649 
7650 	.set_tss_addr = vmx_set_tss_addr,
7651 	.set_identity_map_addr = vmx_set_identity_map_addr,
7652 	.get_tdp_level = get_ept_level,
7653 	.get_mt_mask = vmx_get_mt_mask,
7654 
7655 	.get_exit_info = vmx_get_exit_info,
7656 
7657 	.get_lpage_level = vmx_get_lpage_level,
7658 
7659 	.cpuid_update = vmx_cpuid_update,
7660 
7661 	.rdtscp_supported = vmx_rdtscp_supported,
7662 	.invpcid_supported = vmx_invpcid_supported,
7663 
7664 	.set_supported_cpuid = vmx_set_supported_cpuid,
7665 
7666 	.has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
7667 
7668 	.read_l1_tsc_offset = vmx_read_l1_tsc_offset,
7669 	.write_l1_tsc_offset = vmx_write_l1_tsc_offset,
7670 
7671 	.set_tdp_cr3 = vmx_set_cr3,
7672 
7673 	.check_intercept = vmx_check_intercept,
7674 	.handle_external_intr = vmx_handle_external_intr,
7675 	.mpx_supported = vmx_mpx_supported,
7676 	.xsaves_supported = vmx_xsaves_supported,
7677 	.umip_emulated = vmx_umip_emulated,
7678 	.pt_supported = vmx_pt_supported,
7679 
7680 	.request_immediate_exit = vmx_request_immediate_exit,
7681 
7682 	.sched_in = vmx_sched_in,
7683 
7684 	.slot_enable_log_dirty = vmx_slot_enable_log_dirty,
7685 	.slot_disable_log_dirty = vmx_slot_disable_log_dirty,
7686 	.flush_log_dirty = vmx_flush_log_dirty,
7687 	.enable_log_dirty_pt_masked = vmx_enable_log_dirty_pt_masked,
7688 	.write_log_dirty = vmx_write_pml_buffer,
7689 
7690 	.pre_block = vmx_pre_block,
7691 	.post_block = vmx_post_block,
7692 
7693 	.pmu_ops = &intel_pmu_ops,
7694 
7695 	.update_pi_irte = vmx_update_pi_irte,
7696 
7697 #ifdef CONFIG_X86_64
7698 	.set_hv_timer = vmx_set_hv_timer,
7699 	.cancel_hv_timer = vmx_cancel_hv_timer,
7700 #endif
7701 
7702 	.setup_mce = vmx_setup_mce,
7703 
7704 	.smi_allowed = vmx_smi_allowed,
7705 	.pre_enter_smm = vmx_pre_enter_smm,
7706 	.pre_leave_smm = vmx_pre_leave_smm,
7707 	.enable_smi_window = enable_smi_window,
7708 
7709 	.check_nested_events = NULL,
7710 	.get_nested_state = NULL,
7711 	.set_nested_state = NULL,
7712 	.get_vmcs12_pages = NULL,
7713 	.nested_enable_evmcs = NULL,
7714 };
7715 
7716 static void vmx_cleanup_l1d_flush(void)
7717 {
7718 	if (vmx_l1d_flush_pages) {
7719 		free_pages((unsigned long)vmx_l1d_flush_pages, L1D_CACHE_ORDER);
7720 		vmx_l1d_flush_pages = NULL;
7721 	}
7722 	/* Restore state so sysfs ignores VMX */
7723 	l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_AUTO;
7724 }
7725 
7726 static void vmx_exit(void)
7727 {
7728 #ifdef CONFIG_KEXEC_CORE
7729 	RCU_INIT_POINTER(crash_vmclear_loaded_vmcss, NULL);
7730 	synchronize_rcu();
7731 #endif
7732 
7733 	kvm_exit();
7734 
7735 #if IS_ENABLED(CONFIG_HYPERV)
7736 	if (static_branch_unlikely(&enable_evmcs)) {
7737 		int cpu;
7738 		struct hv_vp_assist_page *vp_ap;
7739 		/*
7740 		 * Reset everything to support using non-enlightened VMCS
7741 		 * access later (e.g. when we reload the module with
7742 		 * enlightened_vmcs=0)
7743 		 */
7744 		for_each_online_cpu(cpu) {
7745 			vp_ap =	hv_get_vp_assist_page(cpu);
7746 
7747 			if (!vp_ap)
7748 				continue;
7749 
7750 			vp_ap->current_nested_vmcs = 0;
7751 			vp_ap->enlighten_vmentry = 0;
7752 		}
7753 
7754 		static_branch_disable(&enable_evmcs);
7755 	}
7756 #endif
7757 	vmx_cleanup_l1d_flush();
7758 }
7759 module_exit(vmx_exit);
7760 
7761 static int __init vmx_init(void)
7762 {
7763 	int r;
7764 
7765 #if IS_ENABLED(CONFIG_HYPERV)
7766 	/*
7767 	 * Enlightened VMCS usage should be recommended and the host needs
7768 	 * to support eVMCS v1 or above. We can also disable eVMCS support
7769 	 * with module parameter.
7770 	 */
7771 	if (enlightened_vmcs &&
7772 	    ms_hyperv.hints & HV_X64_ENLIGHTENED_VMCS_RECOMMENDED &&
7773 	    (ms_hyperv.nested_features & HV_X64_ENLIGHTENED_VMCS_VERSION) >=
7774 	    KVM_EVMCS_VERSION) {
7775 		int cpu;
7776 
7777 		/* Check that we have assist pages on all online CPUs */
7778 		for_each_online_cpu(cpu) {
7779 			if (!hv_get_vp_assist_page(cpu)) {
7780 				enlightened_vmcs = false;
7781 				break;
7782 			}
7783 		}
7784 
7785 		if (enlightened_vmcs) {
7786 			pr_info("KVM: vmx: using Hyper-V Enlightened VMCS\n");
7787 			static_branch_enable(&enable_evmcs);
7788 		}
7789 	} else {
7790 		enlightened_vmcs = false;
7791 	}
7792 #endif
7793 
7794 	r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx),
7795 		     __alignof__(struct vcpu_vmx), THIS_MODULE);
7796 	if (r)
7797 		return r;
7798 
7799 	/*
7800 	 * Must be called after kvm_init() so enable_ept is properly set
7801 	 * up. Hand the parameter mitigation value in which was stored in
7802 	 * the pre module init parser. If no parameter was given, it will
7803 	 * contain 'auto' which will be turned into the default 'cond'
7804 	 * mitigation mode.
7805 	 */
7806 	if (boot_cpu_has(X86_BUG_L1TF)) {
7807 		r = vmx_setup_l1d_flush(vmentry_l1d_flush_param);
7808 		if (r) {
7809 			vmx_exit();
7810 			return r;
7811 		}
7812 	}
7813 
7814 #ifdef CONFIG_KEXEC_CORE
7815 	rcu_assign_pointer(crash_vmclear_loaded_vmcss,
7816 			   crash_vmclear_local_loaded_vmcss);
7817 #endif
7818 	vmx_check_vmcs12_offsets();
7819 
7820 	return 0;
7821 }
7822 module_init(vmx_init);
7823