xref: /openbmc/linux/arch/x86/kvm/vmx/vmx.c (revision d6e0cbb1)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Kernel-based Virtual Machine driver for Linux
4  *
5  * This module enables machines with Intel VT-x extensions to run virtual
6  * machines without emulation or binary translation.
7  *
8  * Copyright (C) 2006 Qumranet, Inc.
9  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
10  *
11  * Authors:
12  *   Avi Kivity   <avi@qumranet.com>
13  *   Yaniv Kamay  <yaniv@qumranet.com>
14  */
15 
16 #include <linux/frame.h>
17 #include <linux/highmem.h>
18 #include <linux/hrtimer.h>
19 #include <linux/kernel.h>
20 #include <linux/kvm_host.h>
21 #include <linux/module.h>
22 #include <linux/moduleparam.h>
23 #include <linux/mod_devicetable.h>
24 #include <linux/mm.h>
25 #include <linux/sched.h>
26 #include <linux/sched/smt.h>
27 #include <linux/slab.h>
28 #include <linux/tboot.h>
29 #include <linux/trace_events.h>
30 
31 #include <asm/apic.h>
32 #include <asm/asm.h>
33 #include <asm/cpu.h>
34 #include <asm/debugreg.h>
35 #include <asm/desc.h>
36 #include <asm/fpu/internal.h>
37 #include <asm/io.h>
38 #include <asm/irq_remapping.h>
39 #include <asm/kexec.h>
40 #include <asm/perf_event.h>
41 #include <asm/mce.h>
42 #include <asm/mmu_context.h>
43 #include <asm/mshyperv.h>
44 #include <asm/spec-ctrl.h>
45 #include <asm/virtext.h>
46 #include <asm/vmx.h>
47 
48 #include "capabilities.h"
49 #include "cpuid.h"
50 #include "evmcs.h"
51 #include "irq.h"
52 #include "kvm_cache_regs.h"
53 #include "lapic.h"
54 #include "mmu.h"
55 #include "nested.h"
56 #include "ops.h"
57 #include "pmu.h"
58 #include "trace.h"
59 #include "vmcs.h"
60 #include "vmcs12.h"
61 #include "vmx.h"
62 #include "x86.h"
63 
64 MODULE_AUTHOR("Qumranet");
65 MODULE_LICENSE("GPL");
66 
67 static const struct x86_cpu_id vmx_cpu_id[] = {
68 	X86_FEATURE_MATCH(X86_FEATURE_VMX),
69 	{}
70 };
71 MODULE_DEVICE_TABLE(x86cpu, vmx_cpu_id);
72 
73 bool __read_mostly enable_vpid = 1;
74 module_param_named(vpid, enable_vpid, bool, 0444);
75 
76 static bool __read_mostly enable_vnmi = 1;
77 module_param_named(vnmi, enable_vnmi, bool, S_IRUGO);
78 
79 bool __read_mostly flexpriority_enabled = 1;
80 module_param_named(flexpriority, flexpriority_enabled, bool, S_IRUGO);
81 
82 bool __read_mostly enable_ept = 1;
83 module_param_named(ept, enable_ept, bool, S_IRUGO);
84 
85 bool __read_mostly enable_unrestricted_guest = 1;
86 module_param_named(unrestricted_guest,
87 			enable_unrestricted_guest, bool, S_IRUGO);
88 
89 bool __read_mostly enable_ept_ad_bits = 1;
90 module_param_named(eptad, enable_ept_ad_bits, bool, S_IRUGO);
91 
92 static bool __read_mostly emulate_invalid_guest_state = true;
93 module_param(emulate_invalid_guest_state, bool, S_IRUGO);
94 
95 static bool __read_mostly fasteoi = 1;
96 module_param(fasteoi, bool, S_IRUGO);
97 
98 static bool __read_mostly enable_apicv = 1;
99 module_param(enable_apicv, bool, S_IRUGO);
100 
101 /*
102  * If nested=1, nested virtualization is supported, i.e., guests may use
103  * VMX and be a hypervisor for its own guests. If nested=0, guests may not
104  * use VMX instructions.
105  */
106 static bool __read_mostly nested = 1;
107 module_param(nested, bool, S_IRUGO);
108 
109 static u64 __read_mostly host_xss;
110 
111 bool __read_mostly enable_pml = 1;
112 module_param_named(pml, enable_pml, bool, S_IRUGO);
113 
114 static bool __read_mostly dump_invalid_vmcs = 0;
115 module_param(dump_invalid_vmcs, bool, 0644);
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 static unsigned long host_idt_base;
393 
394 /*
395  * Though SYSCALL is only supported in 64-bit mode on Intel CPUs, kvm
396  * will emulate SYSCALL in legacy mode if the vendor string in guest
397  * CPUID.0:{EBX,ECX,EDX} is "AuthenticAMD" or "AMDisbetter!" To
398  * support this emulation, IA32_STAR must always be included in
399  * vmx_msr_index[], even in i386 builds.
400  */
401 const u32 vmx_msr_index[] = {
402 #ifdef CONFIG_X86_64
403 	MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
404 #endif
405 	MSR_EFER, MSR_TSC_AUX, MSR_STAR,
406 };
407 
408 #if IS_ENABLED(CONFIG_HYPERV)
409 static bool __read_mostly enlightened_vmcs = true;
410 module_param(enlightened_vmcs, bool, 0444);
411 
412 /* check_ept_pointer() should be under protection of ept_pointer_lock. */
413 static void check_ept_pointer_match(struct kvm *kvm)
414 {
415 	struct kvm_vcpu *vcpu;
416 	u64 tmp_eptp = INVALID_PAGE;
417 	int i;
418 
419 	kvm_for_each_vcpu(i, vcpu, kvm) {
420 		if (!VALID_PAGE(tmp_eptp)) {
421 			tmp_eptp = to_vmx(vcpu)->ept_pointer;
422 		} else if (tmp_eptp != to_vmx(vcpu)->ept_pointer) {
423 			to_kvm_vmx(kvm)->ept_pointers_match
424 				= EPT_POINTERS_MISMATCH;
425 			return;
426 		}
427 	}
428 
429 	to_kvm_vmx(kvm)->ept_pointers_match = EPT_POINTERS_MATCH;
430 }
431 
432 static int kvm_fill_hv_flush_list_func(struct hv_guest_mapping_flush_list *flush,
433 		void *data)
434 {
435 	struct kvm_tlb_range *range = data;
436 
437 	return hyperv_fill_flush_guest_mapping_list(flush, range->start_gfn,
438 			range->pages);
439 }
440 
441 static inline int __hv_remote_flush_tlb_with_range(struct kvm *kvm,
442 		struct kvm_vcpu *vcpu, struct kvm_tlb_range *range)
443 {
444 	u64 ept_pointer = to_vmx(vcpu)->ept_pointer;
445 
446 	/*
447 	 * FLUSH_GUEST_PHYSICAL_ADDRESS_SPACE hypercall needs address
448 	 * of the base of EPT PML4 table, strip off EPT configuration
449 	 * information.
450 	 */
451 	if (range)
452 		return hyperv_flush_guest_mapping_range(ept_pointer & PAGE_MASK,
453 				kvm_fill_hv_flush_list_func, (void *)range);
454 	else
455 		return hyperv_flush_guest_mapping(ept_pointer & PAGE_MASK);
456 }
457 
458 static int hv_remote_flush_tlb_with_range(struct kvm *kvm,
459 		struct kvm_tlb_range *range)
460 {
461 	struct kvm_vcpu *vcpu;
462 	int ret = 0, i;
463 
464 	spin_lock(&to_kvm_vmx(kvm)->ept_pointer_lock);
465 
466 	if (to_kvm_vmx(kvm)->ept_pointers_match == EPT_POINTERS_CHECK)
467 		check_ept_pointer_match(kvm);
468 
469 	if (to_kvm_vmx(kvm)->ept_pointers_match != EPT_POINTERS_MATCH) {
470 		kvm_for_each_vcpu(i, vcpu, kvm) {
471 			/* If ept_pointer is invalid pointer, bypass flush request. */
472 			if (VALID_PAGE(to_vmx(vcpu)->ept_pointer))
473 				ret |= __hv_remote_flush_tlb_with_range(
474 					kvm, vcpu, range);
475 		}
476 	} else {
477 		ret = __hv_remote_flush_tlb_with_range(kvm,
478 				kvm_get_vcpu(kvm, 0), range);
479 	}
480 
481 	spin_unlock(&to_kvm_vmx(kvm)->ept_pointer_lock);
482 	return ret;
483 }
484 static int hv_remote_flush_tlb(struct kvm *kvm)
485 {
486 	return hv_remote_flush_tlb_with_range(kvm, NULL);
487 }
488 
489 #endif /* IS_ENABLED(CONFIG_HYPERV) */
490 
491 /*
492  * Comment's format: document - errata name - stepping - processor name.
493  * Refer from
494  * https://www.virtualbox.org/svn/vbox/trunk/src/VBox/VMM/VMMR0/HMR0.cpp
495  */
496 static u32 vmx_preemption_cpu_tfms[] = {
497 /* 323344.pdf - BA86   - D0 - Xeon 7500 Series */
498 0x000206E6,
499 /* 323056.pdf - AAX65  - C2 - Xeon L3406 */
500 /* 322814.pdf - AAT59  - C2 - i7-600, i5-500, i5-400 and i3-300 Mobile */
501 /* 322911.pdf - AAU65  - C2 - i5-600, i3-500 Desktop and Pentium G6950 */
502 0x00020652,
503 /* 322911.pdf - AAU65  - K0 - i5-600, i3-500 Desktop and Pentium G6950 */
504 0x00020655,
505 /* 322373.pdf - AAO95  - B1 - Xeon 3400 Series */
506 /* 322166.pdf - AAN92  - B1 - i7-800 and i5-700 Desktop */
507 /*
508  * 320767.pdf - AAP86  - B1 -
509  * i7-900 Mobile Extreme, i7-800 and i7-700 Mobile
510  */
511 0x000106E5,
512 /* 321333.pdf - AAM126 - C0 - Xeon 3500 */
513 0x000106A0,
514 /* 321333.pdf - AAM126 - C1 - Xeon 3500 */
515 0x000106A1,
516 /* 320836.pdf - AAJ124 - C0 - i7-900 Desktop Extreme and i7-900 Desktop */
517 0x000106A4,
518  /* 321333.pdf - AAM126 - D0 - Xeon 3500 */
519  /* 321324.pdf - AAK139 - D0 - Xeon 5500 */
520  /* 320836.pdf - AAJ124 - D0 - i7-900 Extreme and i7-900 Desktop */
521 0x000106A5,
522  /* Xeon E3-1220 V2 */
523 0x000306A8,
524 };
525 
526 static inline bool cpu_has_broken_vmx_preemption_timer(void)
527 {
528 	u32 eax = cpuid_eax(0x00000001), i;
529 
530 	/* Clear the reserved bits */
531 	eax &= ~(0x3U << 14 | 0xfU << 28);
532 	for (i = 0; i < ARRAY_SIZE(vmx_preemption_cpu_tfms); i++)
533 		if (eax == vmx_preemption_cpu_tfms[i])
534 			return true;
535 
536 	return false;
537 }
538 
539 static inline bool cpu_need_virtualize_apic_accesses(struct kvm_vcpu *vcpu)
540 {
541 	return flexpriority_enabled && lapic_in_kernel(vcpu);
542 }
543 
544 static inline bool report_flexpriority(void)
545 {
546 	return flexpriority_enabled;
547 }
548 
549 static inline int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
550 {
551 	int i;
552 
553 	for (i = 0; i < vmx->nmsrs; ++i)
554 		if (vmx_msr_index[vmx->guest_msrs[i].index] == msr)
555 			return i;
556 	return -1;
557 }
558 
559 struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
560 {
561 	int i;
562 
563 	i = __find_msr_index(vmx, msr);
564 	if (i >= 0)
565 		return &vmx->guest_msrs[i];
566 	return NULL;
567 }
568 
569 void loaded_vmcs_init(struct loaded_vmcs *loaded_vmcs)
570 {
571 	vmcs_clear(loaded_vmcs->vmcs);
572 	if (loaded_vmcs->shadow_vmcs && loaded_vmcs->launched)
573 		vmcs_clear(loaded_vmcs->shadow_vmcs);
574 	loaded_vmcs->cpu = -1;
575 	loaded_vmcs->launched = 0;
576 }
577 
578 #ifdef CONFIG_KEXEC_CORE
579 /*
580  * This bitmap is used to indicate whether the vmclear
581  * operation is enabled on all cpus. All disabled by
582  * default.
583  */
584 static cpumask_t crash_vmclear_enabled_bitmap = CPU_MASK_NONE;
585 
586 static inline void crash_enable_local_vmclear(int cpu)
587 {
588 	cpumask_set_cpu(cpu, &crash_vmclear_enabled_bitmap);
589 }
590 
591 static inline void crash_disable_local_vmclear(int cpu)
592 {
593 	cpumask_clear_cpu(cpu, &crash_vmclear_enabled_bitmap);
594 }
595 
596 static inline int crash_local_vmclear_enabled(int cpu)
597 {
598 	return cpumask_test_cpu(cpu, &crash_vmclear_enabled_bitmap);
599 }
600 
601 static void crash_vmclear_local_loaded_vmcss(void)
602 {
603 	int cpu = raw_smp_processor_id();
604 	struct loaded_vmcs *v;
605 
606 	if (!crash_local_vmclear_enabled(cpu))
607 		return;
608 
609 	list_for_each_entry(v, &per_cpu(loaded_vmcss_on_cpu, cpu),
610 			    loaded_vmcss_on_cpu_link)
611 		vmcs_clear(v->vmcs);
612 }
613 #else
614 static inline void crash_enable_local_vmclear(int cpu) { }
615 static inline void crash_disable_local_vmclear(int cpu) { }
616 #endif /* CONFIG_KEXEC_CORE */
617 
618 static void __loaded_vmcs_clear(void *arg)
619 {
620 	struct loaded_vmcs *loaded_vmcs = arg;
621 	int cpu = raw_smp_processor_id();
622 
623 	if (loaded_vmcs->cpu != cpu)
624 		return; /* vcpu migration can race with cpu offline */
625 	if (per_cpu(current_vmcs, cpu) == loaded_vmcs->vmcs)
626 		per_cpu(current_vmcs, cpu) = NULL;
627 	crash_disable_local_vmclear(cpu);
628 	list_del(&loaded_vmcs->loaded_vmcss_on_cpu_link);
629 
630 	/*
631 	 * we should ensure updating loaded_vmcs->loaded_vmcss_on_cpu_link
632 	 * is before setting loaded_vmcs->vcpu to -1 which is done in
633 	 * loaded_vmcs_init. Otherwise, other cpu can see vcpu = -1 fist
634 	 * then adds the vmcs into percpu list before it is deleted.
635 	 */
636 	smp_wmb();
637 
638 	loaded_vmcs_init(loaded_vmcs);
639 	crash_enable_local_vmclear(cpu);
640 }
641 
642 void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs)
643 {
644 	int cpu = loaded_vmcs->cpu;
645 
646 	if (cpu != -1)
647 		smp_call_function_single(cpu,
648 			 __loaded_vmcs_clear, loaded_vmcs, 1);
649 }
650 
651 static bool vmx_segment_cache_test_set(struct vcpu_vmx *vmx, unsigned seg,
652 				       unsigned field)
653 {
654 	bool ret;
655 	u32 mask = 1 << (seg * SEG_FIELD_NR + field);
656 
657 	if (!(vmx->vcpu.arch.regs_avail & (1 << VCPU_EXREG_SEGMENTS))) {
658 		vmx->vcpu.arch.regs_avail |= (1 << VCPU_EXREG_SEGMENTS);
659 		vmx->segment_cache.bitmask = 0;
660 	}
661 	ret = vmx->segment_cache.bitmask & mask;
662 	vmx->segment_cache.bitmask |= mask;
663 	return ret;
664 }
665 
666 static u16 vmx_read_guest_seg_selector(struct vcpu_vmx *vmx, unsigned seg)
667 {
668 	u16 *p = &vmx->segment_cache.seg[seg].selector;
669 
670 	if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_SEL))
671 		*p = vmcs_read16(kvm_vmx_segment_fields[seg].selector);
672 	return *p;
673 }
674 
675 static ulong vmx_read_guest_seg_base(struct vcpu_vmx *vmx, unsigned seg)
676 {
677 	ulong *p = &vmx->segment_cache.seg[seg].base;
678 
679 	if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_BASE))
680 		*p = vmcs_readl(kvm_vmx_segment_fields[seg].base);
681 	return *p;
682 }
683 
684 static u32 vmx_read_guest_seg_limit(struct vcpu_vmx *vmx, unsigned seg)
685 {
686 	u32 *p = &vmx->segment_cache.seg[seg].limit;
687 
688 	if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_LIMIT))
689 		*p = vmcs_read32(kvm_vmx_segment_fields[seg].limit);
690 	return *p;
691 }
692 
693 static u32 vmx_read_guest_seg_ar(struct vcpu_vmx *vmx, unsigned seg)
694 {
695 	u32 *p = &vmx->segment_cache.seg[seg].ar;
696 
697 	if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_AR))
698 		*p = vmcs_read32(kvm_vmx_segment_fields[seg].ar_bytes);
699 	return *p;
700 }
701 
702 void update_exception_bitmap(struct kvm_vcpu *vcpu)
703 {
704 	u32 eb;
705 
706 	eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR) |
707 	     (1u << DB_VECTOR) | (1u << AC_VECTOR);
708 	/*
709 	 * Guest access to VMware backdoor ports could legitimately
710 	 * trigger #GP because of TSS I/O permission bitmap.
711 	 * We intercept those #GP and allow access to them anyway
712 	 * as VMware does.
713 	 */
714 	if (enable_vmware_backdoor)
715 		eb |= (1u << GP_VECTOR);
716 	if ((vcpu->guest_debug &
717 	     (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) ==
718 	    (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP))
719 		eb |= 1u << BP_VECTOR;
720 	if (to_vmx(vcpu)->rmode.vm86_active)
721 		eb = ~0;
722 	if (enable_ept)
723 		eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */
724 
725 	/* When we are running a nested L2 guest and L1 specified for it a
726 	 * certain exception bitmap, we must trap the same exceptions and pass
727 	 * them to L1. When running L2, we will only handle the exceptions
728 	 * specified above if L1 did not want them.
729 	 */
730 	if (is_guest_mode(vcpu))
731 		eb |= get_vmcs12(vcpu)->exception_bitmap;
732 
733 	vmcs_write32(EXCEPTION_BITMAP, eb);
734 }
735 
736 /*
737  * Check if MSR is intercepted for currently loaded MSR bitmap.
738  */
739 static bool msr_write_intercepted(struct kvm_vcpu *vcpu, u32 msr)
740 {
741 	unsigned long *msr_bitmap;
742 	int f = sizeof(unsigned long);
743 
744 	if (!cpu_has_vmx_msr_bitmap())
745 		return true;
746 
747 	msr_bitmap = to_vmx(vcpu)->loaded_vmcs->msr_bitmap;
748 
749 	if (msr <= 0x1fff) {
750 		return !!test_bit(msr, msr_bitmap + 0x800 / f);
751 	} else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
752 		msr &= 0x1fff;
753 		return !!test_bit(msr, msr_bitmap + 0xc00 / f);
754 	}
755 
756 	return true;
757 }
758 
759 static void clear_atomic_switch_msr_special(struct vcpu_vmx *vmx,
760 		unsigned long entry, unsigned long exit)
761 {
762 	vm_entry_controls_clearbit(vmx, entry);
763 	vm_exit_controls_clearbit(vmx, exit);
764 }
765 
766 static int find_msr(struct vmx_msrs *m, unsigned int msr)
767 {
768 	unsigned int i;
769 
770 	for (i = 0; i < m->nr; ++i) {
771 		if (m->val[i].index == msr)
772 			return i;
773 	}
774 	return -ENOENT;
775 }
776 
777 static void clear_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr)
778 {
779 	int i;
780 	struct msr_autoload *m = &vmx->msr_autoload;
781 
782 	switch (msr) {
783 	case MSR_EFER:
784 		if (cpu_has_load_ia32_efer()) {
785 			clear_atomic_switch_msr_special(vmx,
786 					VM_ENTRY_LOAD_IA32_EFER,
787 					VM_EXIT_LOAD_IA32_EFER);
788 			return;
789 		}
790 		break;
791 	case MSR_CORE_PERF_GLOBAL_CTRL:
792 		if (cpu_has_load_perf_global_ctrl()) {
793 			clear_atomic_switch_msr_special(vmx,
794 					VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
795 					VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
796 			return;
797 		}
798 		break;
799 	}
800 	i = find_msr(&m->guest, msr);
801 	if (i < 0)
802 		goto skip_guest;
803 	--m->guest.nr;
804 	m->guest.val[i] = m->guest.val[m->guest.nr];
805 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->guest.nr);
806 
807 skip_guest:
808 	i = find_msr(&m->host, msr);
809 	if (i < 0)
810 		return;
811 
812 	--m->host.nr;
813 	m->host.val[i] = m->host.val[m->host.nr];
814 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->host.nr);
815 }
816 
817 static void add_atomic_switch_msr_special(struct vcpu_vmx *vmx,
818 		unsigned long entry, unsigned long exit,
819 		unsigned long guest_val_vmcs, unsigned long host_val_vmcs,
820 		u64 guest_val, u64 host_val)
821 {
822 	vmcs_write64(guest_val_vmcs, guest_val);
823 	if (host_val_vmcs != HOST_IA32_EFER)
824 		vmcs_write64(host_val_vmcs, host_val);
825 	vm_entry_controls_setbit(vmx, entry);
826 	vm_exit_controls_setbit(vmx, exit);
827 }
828 
829 static void add_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr,
830 				  u64 guest_val, u64 host_val, bool entry_only)
831 {
832 	int i, j = 0;
833 	struct msr_autoload *m = &vmx->msr_autoload;
834 
835 	switch (msr) {
836 	case MSR_EFER:
837 		if (cpu_has_load_ia32_efer()) {
838 			add_atomic_switch_msr_special(vmx,
839 					VM_ENTRY_LOAD_IA32_EFER,
840 					VM_EXIT_LOAD_IA32_EFER,
841 					GUEST_IA32_EFER,
842 					HOST_IA32_EFER,
843 					guest_val, host_val);
844 			return;
845 		}
846 		break;
847 	case MSR_CORE_PERF_GLOBAL_CTRL:
848 		if (cpu_has_load_perf_global_ctrl()) {
849 			add_atomic_switch_msr_special(vmx,
850 					VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
851 					VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL,
852 					GUEST_IA32_PERF_GLOBAL_CTRL,
853 					HOST_IA32_PERF_GLOBAL_CTRL,
854 					guest_val, host_val);
855 			return;
856 		}
857 		break;
858 	case MSR_IA32_PEBS_ENABLE:
859 		/* PEBS needs a quiescent period after being disabled (to write
860 		 * a record).  Disabling PEBS through VMX MSR swapping doesn't
861 		 * provide that period, so a CPU could write host's record into
862 		 * guest's memory.
863 		 */
864 		wrmsrl(MSR_IA32_PEBS_ENABLE, 0);
865 	}
866 
867 	i = find_msr(&m->guest, msr);
868 	if (!entry_only)
869 		j = find_msr(&m->host, msr);
870 
871 	if ((i < 0 && m->guest.nr == NR_AUTOLOAD_MSRS) ||
872 		(j < 0 &&  m->host.nr == NR_AUTOLOAD_MSRS)) {
873 		printk_once(KERN_WARNING "Not enough msr switch entries. "
874 				"Can't add msr %x\n", msr);
875 		return;
876 	}
877 	if (i < 0) {
878 		i = m->guest.nr++;
879 		vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->guest.nr);
880 	}
881 	m->guest.val[i].index = msr;
882 	m->guest.val[i].value = guest_val;
883 
884 	if (entry_only)
885 		return;
886 
887 	if (j < 0) {
888 		j = m->host.nr++;
889 		vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->host.nr);
890 	}
891 	m->host.val[j].index = msr;
892 	m->host.val[j].value = host_val;
893 }
894 
895 static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset)
896 {
897 	u64 guest_efer = vmx->vcpu.arch.efer;
898 	u64 ignore_bits = 0;
899 
900 	if (!enable_ept) {
901 		/*
902 		 * NX is needed to handle CR0.WP=1, CR4.SMEP=1.  Testing
903 		 * host CPUID is more efficient than testing guest CPUID
904 		 * or CR4.  Host SMEP is anyway a requirement for guest SMEP.
905 		 */
906 		if (boot_cpu_has(X86_FEATURE_SMEP))
907 			guest_efer |= EFER_NX;
908 		else if (!(guest_efer & EFER_NX))
909 			ignore_bits |= EFER_NX;
910 	}
911 
912 	/*
913 	 * LMA and LME handled by hardware; SCE meaningless outside long mode.
914 	 */
915 	ignore_bits |= EFER_SCE;
916 #ifdef CONFIG_X86_64
917 	ignore_bits |= EFER_LMA | EFER_LME;
918 	/* SCE is meaningful only in long mode on Intel */
919 	if (guest_efer & EFER_LMA)
920 		ignore_bits &= ~(u64)EFER_SCE;
921 #endif
922 
923 	/*
924 	 * On EPT, we can't emulate NX, so we must switch EFER atomically.
925 	 * On CPUs that support "load IA32_EFER", always switch EFER
926 	 * atomically, since it's faster than switching it manually.
927 	 */
928 	if (cpu_has_load_ia32_efer() ||
929 	    (enable_ept && ((vmx->vcpu.arch.efer ^ host_efer) & EFER_NX))) {
930 		if (!(guest_efer & EFER_LMA))
931 			guest_efer &= ~EFER_LME;
932 		if (guest_efer != host_efer)
933 			add_atomic_switch_msr(vmx, MSR_EFER,
934 					      guest_efer, host_efer, false);
935 		else
936 			clear_atomic_switch_msr(vmx, MSR_EFER);
937 		return false;
938 	} else {
939 		clear_atomic_switch_msr(vmx, MSR_EFER);
940 
941 		guest_efer &= ~ignore_bits;
942 		guest_efer |= host_efer & ignore_bits;
943 
944 		vmx->guest_msrs[efer_offset].data = guest_efer;
945 		vmx->guest_msrs[efer_offset].mask = ~ignore_bits;
946 
947 		return true;
948 	}
949 }
950 
951 #ifdef CONFIG_X86_32
952 /*
953  * On 32-bit kernels, VM exits still load the FS and GS bases from the
954  * VMCS rather than the segment table.  KVM uses this helper to figure
955  * out the current bases to poke them into the VMCS before entry.
956  */
957 static unsigned long segment_base(u16 selector)
958 {
959 	struct desc_struct *table;
960 	unsigned long v;
961 
962 	if (!(selector & ~SEGMENT_RPL_MASK))
963 		return 0;
964 
965 	table = get_current_gdt_ro();
966 
967 	if ((selector & SEGMENT_TI_MASK) == SEGMENT_LDT) {
968 		u16 ldt_selector = kvm_read_ldt();
969 
970 		if (!(ldt_selector & ~SEGMENT_RPL_MASK))
971 			return 0;
972 
973 		table = (struct desc_struct *)segment_base(ldt_selector);
974 	}
975 	v = get_desc_base(&table[selector >> 3]);
976 	return v;
977 }
978 #endif
979 
980 static inline void pt_load_msr(struct pt_ctx *ctx, u32 addr_range)
981 {
982 	u32 i;
983 
984 	wrmsrl(MSR_IA32_RTIT_STATUS, ctx->status);
985 	wrmsrl(MSR_IA32_RTIT_OUTPUT_BASE, ctx->output_base);
986 	wrmsrl(MSR_IA32_RTIT_OUTPUT_MASK, ctx->output_mask);
987 	wrmsrl(MSR_IA32_RTIT_CR3_MATCH, ctx->cr3_match);
988 	for (i = 0; i < addr_range; i++) {
989 		wrmsrl(MSR_IA32_RTIT_ADDR0_A + i * 2, ctx->addr_a[i]);
990 		wrmsrl(MSR_IA32_RTIT_ADDR0_B + i * 2, ctx->addr_b[i]);
991 	}
992 }
993 
994 static inline void pt_save_msr(struct pt_ctx *ctx, u32 addr_range)
995 {
996 	u32 i;
997 
998 	rdmsrl(MSR_IA32_RTIT_STATUS, ctx->status);
999 	rdmsrl(MSR_IA32_RTIT_OUTPUT_BASE, ctx->output_base);
1000 	rdmsrl(MSR_IA32_RTIT_OUTPUT_MASK, ctx->output_mask);
1001 	rdmsrl(MSR_IA32_RTIT_CR3_MATCH, ctx->cr3_match);
1002 	for (i = 0; i < addr_range; i++) {
1003 		rdmsrl(MSR_IA32_RTIT_ADDR0_A + i * 2, ctx->addr_a[i]);
1004 		rdmsrl(MSR_IA32_RTIT_ADDR0_B + i * 2, ctx->addr_b[i]);
1005 	}
1006 }
1007 
1008 static void pt_guest_enter(struct vcpu_vmx *vmx)
1009 {
1010 	if (pt_mode == PT_MODE_SYSTEM)
1011 		return;
1012 
1013 	/*
1014 	 * GUEST_IA32_RTIT_CTL is already set in the VMCS.
1015 	 * Save host state before VM entry.
1016 	 */
1017 	rdmsrl(MSR_IA32_RTIT_CTL, vmx->pt_desc.host.ctl);
1018 	if (vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) {
1019 		wrmsrl(MSR_IA32_RTIT_CTL, 0);
1020 		pt_save_msr(&vmx->pt_desc.host, vmx->pt_desc.addr_range);
1021 		pt_load_msr(&vmx->pt_desc.guest, vmx->pt_desc.addr_range);
1022 	}
1023 }
1024 
1025 static void pt_guest_exit(struct vcpu_vmx *vmx)
1026 {
1027 	if (pt_mode == PT_MODE_SYSTEM)
1028 		return;
1029 
1030 	if (vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) {
1031 		pt_save_msr(&vmx->pt_desc.guest, vmx->pt_desc.addr_range);
1032 		pt_load_msr(&vmx->pt_desc.host, vmx->pt_desc.addr_range);
1033 	}
1034 
1035 	/* Reload host state (IA32_RTIT_CTL will be cleared on VM exit). */
1036 	wrmsrl(MSR_IA32_RTIT_CTL, vmx->pt_desc.host.ctl);
1037 }
1038 
1039 void vmx_set_host_fs_gs(struct vmcs_host_state *host, u16 fs_sel, u16 gs_sel,
1040 			unsigned long fs_base, unsigned long gs_base)
1041 {
1042 	if (unlikely(fs_sel != host->fs_sel)) {
1043 		if (!(fs_sel & 7))
1044 			vmcs_write16(HOST_FS_SELECTOR, fs_sel);
1045 		else
1046 			vmcs_write16(HOST_FS_SELECTOR, 0);
1047 		host->fs_sel = fs_sel;
1048 	}
1049 	if (unlikely(gs_sel != host->gs_sel)) {
1050 		if (!(gs_sel & 7))
1051 			vmcs_write16(HOST_GS_SELECTOR, gs_sel);
1052 		else
1053 			vmcs_write16(HOST_GS_SELECTOR, 0);
1054 		host->gs_sel = gs_sel;
1055 	}
1056 	if (unlikely(fs_base != host->fs_base)) {
1057 		vmcs_writel(HOST_FS_BASE, fs_base);
1058 		host->fs_base = fs_base;
1059 	}
1060 	if (unlikely(gs_base != host->gs_base)) {
1061 		vmcs_writel(HOST_GS_BASE, gs_base);
1062 		host->gs_base = gs_base;
1063 	}
1064 }
1065 
1066 void vmx_prepare_switch_to_guest(struct kvm_vcpu *vcpu)
1067 {
1068 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1069 	struct vmcs_host_state *host_state;
1070 #ifdef CONFIG_X86_64
1071 	int cpu = raw_smp_processor_id();
1072 #endif
1073 	unsigned long fs_base, gs_base;
1074 	u16 fs_sel, gs_sel;
1075 	int i;
1076 
1077 	vmx->req_immediate_exit = false;
1078 
1079 	/*
1080 	 * Note that guest MSRs to be saved/restored can also be changed
1081 	 * when guest state is loaded. This happens when guest transitions
1082 	 * to/from long-mode by setting MSR_EFER.LMA.
1083 	 */
1084 	if (!vmx->guest_msrs_ready) {
1085 		vmx->guest_msrs_ready = true;
1086 		for (i = 0; i < vmx->save_nmsrs; ++i)
1087 			kvm_set_shared_msr(vmx->guest_msrs[i].index,
1088 					   vmx->guest_msrs[i].data,
1089 					   vmx->guest_msrs[i].mask);
1090 
1091 	}
1092 	if (vmx->guest_state_loaded)
1093 		return;
1094 
1095 	host_state = &vmx->loaded_vmcs->host_state;
1096 
1097 	/*
1098 	 * Set host fs and gs selectors.  Unfortunately, 22.2.3 does not
1099 	 * allow segment selectors with cpl > 0 or ti == 1.
1100 	 */
1101 	host_state->ldt_sel = kvm_read_ldt();
1102 
1103 #ifdef CONFIG_X86_64
1104 	savesegment(ds, host_state->ds_sel);
1105 	savesegment(es, host_state->es_sel);
1106 
1107 	gs_base = cpu_kernelmode_gs_base(cpu);
1108 	if (likely(is_64bit_mm(current->mm))) {
1109 		save_fsgs_for_kvm();
1110 		fs_sel = current->thread.fsindex;
1111 		gs_sel = current->thread.gsindex;
1112 		fs_base = current->thread.fsbase;
1113 		vmx->msr_host_kernel_gs_base = current->thread.gsbase;
1114 	} else {
1115 		savesegment(fs, fs_sel);
1116 		savesegment(gs, gs_sel);
1117 		fs_base = read_msr(MSR_FS_BASE);
1118 		vmx->msr_host_kernel_gs_base = read_msr(MSR_KERNEL_GS_BASE);
1119 	}
1120 
1121 	wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1122 #else
1123 	savesegment(fs, fs_sel);
1124 	savesegment(gs, gs_sel);
1125 	fs_base = segment_base(fs_sel);
1126 	gs_base = segment_base(gs_sel);
1127 #endif
1128 
1129 	vmx_set_host_fs_gs(host_state, fs_sel, gs_sel, fs_base, gs_base);
1130 	vmx->guest_state_loaded = true;
1131 }
1132 
1133 static void vmx_prepare_switch_to_host(struct vcpu_vmx *vmx)
1134 {
1135 	struct vmcs_host_state *host_state;
1136 
1137 	if (!vmx->guest_state_loaded)
1138 		return;
1139 
1140 	host_state = &vmx->loaded_vmcs->host_state;
1141 
1142 	++vmx->vcpu.stat.host_state_reload;
1143 
1144 #ifdef CONFIG_X86_64
1145 	rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1146 #endif
1147 	if (host_state->ldt_sel || (host_state->gs_sel & 7)) {
1148 		kvm_load_ldt(host_state->ldt_sel);
1149 #ifdef CONFIG_X86_64
1150 		load_gs_index(host_state->gs_sel);
1151 #else
1152 		loadsegment(gs, host_state->gs_sel);
1153 #endif
1154 	}
1155 	if (host_state->fs_sel & 7)
1156 		loadsegment(fs, host_state->fs_sel);
1157 #ifdef CONFIG_X86_64
1158 	if (unlikely(host_state->ds_sel | host_state->es_sel)) {
1159 		loadsegment(ds, host_state->ds_sel);
1160 		loadsegment(es, host_state->es_sel);
1161 	}
1162 #endif
1163 	invalidate_tss_limit();
1164 #ifdef CONFIG_X86_64
1165 	wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
1166 #endif
1167 	load_fixmap_gdt(raw_smp_processor_id());
1168 	vmx->guest_state_loaded = false;
1169 	vmx->guest_msrs_ready = false;
1170 }
1171 
1172 #ifdef CONFIG_X86_64
1173 static u64 vmx_read_guest_kernel_gs_base(struct vcpu_vmx *vmx)
1174 {
1175 	preempt_disable();
1176 	if (vmx->guest_state_loaded)
1177 		rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1178 	preempt_enable();
1179 	return vmx->msr_guest_kernel_gs_base;
1180 }
1181 
1182 static void vmx_write_guest_kernel_gs_base(struct vcpu_vmx *vmx, u64 data)
1183 {
1184 	preempt_disable();
1185 	if (vmx->guest_state_loaded)
1186 		wrmsrl(MSR_KERNEL_GS_BASE, data);
1187 	preempt_enable();
1188 	vmx->msr_guest_kernel_gs_base = data;
1189 }
1190 #endif
1191 
1192 static void vmx_vcpu_pi_load(struct kvm_vcpu *vcpu, int cpu)
1193 {
1194 	struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
1195 	struct pi_desc old, new;
1196 	unsigned int dest;
1197 
1198 	/*
1199 	 * In case of hot-plug or hot-unplug, we may have to undo
1200 	 * vmx_vcpu_pi_put even if there is no assigned device.  And we
1201 	 * always keep PI.NDST up to date for simplicity: it makes the
1202 	 * code easier, and CPU migration is not a fast path.
1203 	 */
1204 	if (!pi_test_sn(pi_desc) && vcpu->cpu == cpu)
1205 		return;
1206 
1207 	/* The full case.  */
1208 	do {
1209 		old.control = new.control = pi_desc->control;
1210 
1211 		dest = cpu_physical_id(cpu);
1212 
1213 		if (x2apic_enabled())
1214 			new.ndst = dest;
1215 		else
1216 			new.ndst = (dest << 8) & 0xFF00;
1217 
1218 		new.sn = 0;
1219 	} while (cmpxchg64(&pi_desc->control, old.control,
1220 			   new.control) != old.control);
1221 
1222 	/*
1223 	 * Clear SN before reading the bitmap.  The VT-d firmware
1224 	 * writes the bitmap and reads SN atomically (5.2.3 in the
1225 	 * spec), so it doesn't really have a memory barrier that
1226 	 * pairs with this, but we cannot do that and we need one.
1227 	 */
1228 	smp_mb__after_atomic();
1229 
1230 	if (!bitmap_empty((unsigned long *)pi_desc->pir, NR_VECTORS))
1231 		pi_set_on(pi_desc);
1232 }
1233 
1234 void vmx_vcpu_load_vmcs(struct kvm_vcpu *vcpu, int cpu)
1235 {
1236 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1237 	bool already_loaded = vmx->loaded_vmcs->cpu == cpu;
1238 
1239 	if (!already_loaded) {
1240 		loaded_vmcs_clear(vmx->loaded_vmcs);
1241 		local_irq_disable();
1242 		crash_disable_local_vmclear(cpu);
1243 
1244 		/*
1245 		 * Read loaded_vmcs->cpu should be before fetching
1246 		 * loaded_vmcs->loaded_vmcss_on_cpu_link.
1247 		 * See the comments in __loaded_vmcs_clear().
1248 		 */
1249 		smp_rmb();
1250 
1251 		list_add(&vmx->loaded_vmcs->loaded_vmcss_on_cpu_link,
1252 			 &per_cpu(loaded_vmcss_on_cpu, cpu));
1253 		crash_enable_local_vmclear(cpu);
1254 		local_irq_enable();
1255 	}
1256 
1257 	if (per_cpu(current_vmcs, cpu) != vmx->loaded_vmcs->vmcs) {
1258 		per_cpu(current_vmcs, cpu) = vmx->loaded_vmcs->vmcs;
1259 		vmcs_load(vmx->loaded_vmcs->vmcs);
1260 		indirect_branch_prediction_barrier();
1261 	}
1262 
1263 	if (!already_loaded) {
1264 		void *gdt = get_current_gdt_ro();
1265 		unsigned long sysenter_esp;
1266 
1267 		kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1268 
1269 		/*
1270 		 * Linux uses per-cpu TSS and GDT, so set these when switching
1271 		 * processors.  See 22.2.4.
1272 		 */
1273 		vmcs_writel(HOST_TR_BASE,
1274 			    (unsigned long)&get_cpu_entry_area(cpu)->tss.x86_tss);
1275 		vmcs_writel(HOST_GDTR_BASE, (unsigned long)gdt);   /* 22.2.4 */
1276 
1277 		/*
1278 		 * VM exits change the host TR limit to 0x67 after a VM
1279 		 * exit.  This is okay, since 0x67 covers everything except
1280 		 * the IO bitmap and have have code to handle the IO bitmap
1281 		 * being lost after a VM exit.
1282 		 */
1283 		BUILD_BUG_ON(IO_BITMAP_OFFSET - 1 != 0x67);
1284 
1285 		rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
1286 		vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
1287 
1288 		vmx->loaded_vmcs->cpu = cpu;
1289 	}
1290 
1291 	/* Setup TSC multiplier */
1292 	if (kvm_has_tsc_control &&
1293 	    vmx->current_tsc_ratio != vcpu->arch.tsc_scaling_ratio)
1294 		decache_tsc_multiplier(vmx);
1295 }
1296 
1297 /*
1298  * Switches to specified vcpu, until a matching vcpu_put(), but assumes
1299  * vcpu mutex is already taken.
1300  */
1301 void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1302 {
1303 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1304 
1305 	vmx_vcpu_load_vmcs(vcpu, cpu);
1306 
1307 	vmx_vcpu_pi_load(vcpu, cpu);
1308 
1309 	vmx->host_pkru = read_pkru();
1310 	vmx->host_debugctlmsr = get_debugctlmsr();
1311 }
1312 
1313 static void vmx_vcpu_pi_put(struct kvm_vcpu *vcpu)
1314 {
1315 	struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
1316 
1317 	if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
1318 		!irq_remapping_cap(IRQ_POSTING_CAP)  ||
1319 		!kvm_vcpu_apicv_active(vcpu))
1320 		return;
1321 
1322 	/* Set SN when the vCPU is preempted */
1323 	if (vcpu->preempted)
1324 		pi_set_sn(pi_desc);
1325 }
1326 
1327 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
1328 {
1329 	vmx_vcpu_pi_put(vcpu);
1330 
1331 	vmx_prepare_switch_to_host(to_vmx(vcpu));
1332 }
1333 
1334 static bool emulation_required(struct kvm_vcpu *vcpu)
1335 {
1336 	return emulate_invalid_guest_state && !guest_state_valid(vcpu);
1337 }
1338 
1339 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu);
1340 
1341 unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
1342 {
1343 	unsigned long rflags, save_rflags;
1344 
1345 	if (!test_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail)) {
1346 		__set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
1347 		rflags = vmcs_readl(GUEST_RFLAGS);
1348 		if (to_vmx(vcpu)->rmode.vm86_active) {
1349 			rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
1350 			save_rflags = to_vmx(vcpu)->rmode.save_rflags;
1351 			rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
1352 		}
1353 		to_vmx(vcpu)->rflags = rflags;
1354 	}
1355 	return to_vmx(vcpu)->rflags;
1356 }
1357 
1358 void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
1359 {
1360 	unsigned long old_rflags = vmx_get_rflags(vcpu);
1361 
1362 	__set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
1363 	to_vmx(vcpu)->rflags = rflags;
1364 	if (to_vmx(vcpu)->rmode.vm86_active) {
1365 		to_vmx(vcpu)->rmode.save_rflags = rflags;
1366 		rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
1367 	}
1368 	vmcs_writel(GUEST_RFLAGS, rflags);
1369 
1370 	if ((old_rflags ^ to_vmx(vcpu)->rflags) & X86_EFLAGS_VM)
1371 		to_vmx(vcpu)->emulation_required = emulation_required(vcpu);
1372 }
1373 
1374 u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu)
1375 {
1376 	u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1377 	int ret = 0;
1378 
1379 	if (interruptibility & GUEST_INTR_STATE_STI)
1380 		ret |= KVM_X86_SHADOW_INT_STI;
1381 	if (interruptibility & GUEST_INTR_STATE_MOV_SS)
1382 		ret |= KVM_X86_SHADOW_INT_MOV_SS;
1383 
1384 	return ret;
1385 }
1386 
1387 void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
1388 {
1389 	u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1390 	u32 interruptibility = interruptibility_old;
1391 
1392 	interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
1393 
1394 	if (mask & KVM_X86_SHADOW_INT_MOV_SS)
1395 		interruptibility |= GUEST_INTR_STATE_MOV_SS;
1396 	else if (mask & KVM_X86_SHADOW_INT_STI)
1397 		interruptibility |= GUEST_INTR_STATE_STI;
1398 
1399 	if ((interruptibility != interruptibility_old))
1400 		vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
1401 }
1402 
1403 static int vmx_rtit_ctl_check(struct kvm_vcpu *vcpu, u64 data)
1404 {
1405 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1406 	unsigned long value;
1407 
1408 	/*
1409 	 * Any MSR write that attempts to change bits marked reserved will
1410 	 * case a #GP fault.
1411 	 */
1412 	if (data & vmx->pt_desc.ctl_bitmask)
1413 		return 1;
1414 
1415 	/*
1416 	 * Any attempt to modify IA32_RTIT_CTL while TraceEn is set will
1417 	 * result in a #GP unless the same write also clears TraceEn.
1418 	 */
1419 	if ((vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) &&
1420 		((vmx->pt_desc.guest.ctl ^ data) & ~RTIT_CTL_TRACEEN))
1421 		return 1;
1422 
1423 	/*
1424 	 * WRMSR to IA32_RTIT_CTL that sets TraceEn but clears this bit
1425 	 * and FabricEn would cause #GP, if
1426 	 * CPUID.(EAX=14H, ECX=0):ECX.SNGLRGNOUT[bit 2] = 0
1427 	 */
1428 	if ((data & RTIT_CTL_TRACEEN) && !(data & RTIT_CTL_TOPA) &&
1429 		!(data & RTIT_CTL_FABRIC_EN) &&
1430 		!intel_pt_validate_cap(vmx->pt_desc.caps,
1431 					PT_CAP_single_range_output))
1432 		return 1;
1433 
1434 	/*
1435 	 * MTCFreq, CycThresh and PSBFreq encodings check, any MSR write that
1436 	 * utilize encodings marked reserved will casue a #GP fault.
1437 	 */
1438 	value = intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc_periods);
1439 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc) &&
1440 			!test_bit((data & RTIT_CTL_MTC_RANGE) >>
1441 			RTIT_CTL_MTC_RANGE_OFFSET, &value))
1442 		return 1;
1443 	value = intel_pt_validate_cap(vmx->pt_desc.caps,
1444 						PT_CAP_cycle_thresholds);
1445 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc) &&
1446 			!test_bit((data & RTIT_CTL_CYC_THRESH) >>
1447 			RTIT_CTL_CYC_THRESH_OFFSET, &value))
1448 		return 1;
1449 	value = intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_periods);
1450 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc) &&
1451 			!test_bit((data & RTIT_CTL_PSB_FREQ) >>
1452 			RTIT_CTL_PSB_FREQ_OFFSET, &value))
1453 		return 1;
1454 
1455 	/*
1456 	 * If ADDRx_CFG is reserved or the encodings is >2 will
1457 	 * cause a #GP fault.
1458 	 */
1459 	value = (data & RTIT_CTL_ADDR0) >> RTIT_CTL_ADDR0_OFFSET;
1460 	if ((value && (vmx->pt_desc.addr_range < 1)) || (value > 2))
1461 		return 1;
1462 	value = (data & RTIT_CTL_ADDR1) >> RTIT_CTL_ADDR1_OFFSET;
1463 	if ((value && (vmx->pt_desc.addr_range < 2)) || (value > 2))
1464 		return 1;
1465 	value = (data & RTIT_CTL_ADDR2) >> RTIT_CTL_ADDR2_OFFSET;
1466 	if ((value && (vmx->pt_desc.addr_range < 3)) || (value > 2))
1467 		return 1;
1468 	value = (data & RTIT_CTL_ADDR3) >> RTIT_CTL_ADDR3_OFFSET;
1469 	if ((value && (vmx->pt_desc.addr_range < 4)) || (value > 2))
1470 		return 1;
1471 
1472 	return 0;
1473 }
1474 
1475 
1476 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
1477 {
1478 	unsigned long rip;
1479 
1480 	rip = kvm_rip_read(vcpu);
1481 	rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
1482 	kvm_rip_write(vcpu, rip);
1483 
1484 	/* skipping an emulated instruction also counts */
1485 	vmx_set_interrupt_shadow(vcpu, 0);
1486 }
1487 
1488 static void vmx_clear_hlt(struct kvm_vcpu *vcpu)
1489 {
1490 	/*
1491 	 * Ensure that we clear the HLT state in the VMCS.  We don't need to
1492 	 * explicitly skip the instruction because if the HLT state is set,
1493 	 * then the instruction is already executing and RIP has already been
1494 	 * advanced.
1495 	 */
1496 	if (kvm_hlt_in_guest(vcpu->kvm) &&
1497 			vmcs_read32(GUEST_ACTIVITY_STATE) == GUEST_ACTIVITY_HLT)
1498 		vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
1499 }
1500 
1501 static void vmx_queue_exception(struct kvm_vcpu *vcpu)
1502 {
1503 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1504 	unsigned nr = vcpu->arch.exception.nr;
1505 	bool has_error_code = vcpu->arch.exception.has_error_code;
1506 	u32 error_code = vcpu->arch.exception.error_code;
1507 	u32 intr_info = nr | INTR_INFO_VALID_MASK;
1508 
1509 	kvm_deliver_exception_payload(vcpu);
1510 
1511 	if (has_error_code) {
1512 		vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
1513 		intr_info |= INTR_INFO_DELIVER_CODE_MASK;
1514 	}
1515 
1516 	if (vmx->rmode.vm86_active) {
1517 		int inc_eip = 0;
1518 		if (kvm_exception_is_soft(nr))
1519 			inc_eip = vcpu->arch.event_exit_inst_len;
1520 		if (kvm_inject_realmode_interrupt(vcpu, nr, inc_eip) != EMULATE_DONE)
1521 			kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
1522 		return;
1523 	}
1524 
1525 	WARN_ON_ONCE(vmx->emulation_required);
1526 
1527 	if (kvm_exception_is_soft(nr)) {
1528 		vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
1529 			     vmx->vcpu.arch.event_exit_inst_len);
1530 		intr_info |= INTR_TYPE_SOFT_EXCEPTION;
1531 	} else
1532 		intr_info |= INTR_TYPE_HARD_EXCEPTION;
1533 
1534 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
1535 
1536 	vmx_clear_hlt(vcpu);
1537 }
1538 
1539 static bool vmx_rdtscp_supported(void)
1540 {
1541 	return cpu_has_vmx_rdtscp();
1542 }
1543 
1544 static bool vmx_invpcid_supported(void)
1545 {
1546 	return cpu_has_vmx_invpcid();
1547 }
1548 
1549 /*
1550  * Swap MSR entry in host/guest MSR entry array.
1551  */
1552 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
1553 {
1554 	struct shared_msr_entry tmp;
1555 
1556 	tmp = vmx->guest_msrs[to];
1557 	vmx->guest_msrs[to] = vmx->guest_msrs[from];
1558 	vmx->guest_msrs[from] = tmp;
1559 }
1560 
1561 /*
1562  * Set up the vmcs to automatically save and restore system
1563  * msrs.  Don't touch the 64-bit msrs if the guest is in legacy
1564  * mode, as fiddling with msrs is very expensive.
1565  */
1566 static void setup_msrs(struct vcpu_vmx *vmx)
1567 {
1568 	int save_nmsrs, index;
1569 
1570 	save_nmsrs = 0;
1571 #ifdef CONFIG_X86_64
1572 	/*
1573 	 * The SYSCALL MSRs are only needed on long mode guests, and only
1574 	 * when EFER.SCE is set.
1575 	 */
1576 	if (is_long_mode(&vmx->vcpu) && (vmx->vcpu.arch.efer & EFER_SCE)) {
1577 		index = __find_msr_index(vmx, MSR_STAR);
1578 		if (index >= 0)
1579 			move_msr_up(vmx, index, save_nmsrs++);
1580 		index = __find_msr_index(vmx, MSR_LSTAR);
1581 		if (index >= 0)
1582 			move_msr_up(vmx, index, save_nmsrs++);
1583 		index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
1584 		if (index >= 0)
1585 			move_msr_up(vmx, index, save_nmsrs++);
1586 	}
1587 #endif
1588 	index = __find_msr_index(vmx, MSR_EFER);
1589 	if (index >= 0 && update_transition_efer(vmx, index))
1590 		move_msr_up(vmx, index, save_nmsrs++);
1591 	index = __find_msr_index(vmx, MSR_TSC_AUX);
1592 	if (index >= 0 && guest_cpuid_has(&vmx->vcpu, X86_FEATURE_RDTSCP))
1593 		move_msr_up(vmx, index, save_nmsrs++);
1594 
1595 	vmx->save_nmsrs = save_nmsrs;
1596 	vmx->guest_msrs_ready = false;
1597 
1598 	if (cpu_has_vmx_msr_bitmap())
1599 		vmx_update_msr_bitmap(&vmx->vcpu);
1600 }
1601 
1602 static u64 vmx_read_l1_tsc_offset(struct kvm_vcpu *vcpu)
1603 {
1604 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1605 
1606 	if (is_guest_mode(vcpu) &&
1607 	    (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING))
1608 		return vcpu->arch.tsc_offset - vmcs12->tsc_offset;
1609 
1610 	return vcpu->arch.tsc_offset;
1611 }
1612 
1613 static u64 vmx_write_l1_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1614 {
1615 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1616 	u64 g_tsc_offset = 0;
1617 
1618 	/*
1619 	 * We're here if L1 chose not to trap WRMSR to TSC. According
1620 	 * to the spec, this should set L1's TSC; The offset that L1
1621 	 * set for L2 remains unchanged, and still needs to be added
1622 	 * to the newly set TSC to get L2's TSC.
1623 	 */
1624 	if (is_guest_mode(vcpu) &&
1625 	    (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING))
1626 		g_tsc_offset = vmcs12->tsc_offset;
1627 
1628 	trace_kvm_write_tsc_offset(vcpu->vcpu_id,
1629 				   vcpu->arch.tsc_offset - g_tsc_offset,
1630 				   offset);
1631 	vmcs_write64(TSC_OFFSET, offset + g_tsc_offset);
1632 	return offset + g_tsc_offset;
1633 }
1634 
1635 /*
1636  * nested_vmx_allowed() checks whether a guest should be allowed to use VMX
1637  * instructions and MSRs (i.e., nested VMX). Nested VMX is disabled for
1638  * all guests if the "nested" module option is off, and can also be disabled
1639  * for a single guest by disabling its VMX cpuid bit.
1640  */
1641 bool nested_vmx_allowed(struct kvm_vcpu *vcpu)
1642 {
1643 	return nested && guest_cpuid_has(vcpu, X86_FEATURE_VMX);
1644 }
1645 
1646 static inline bool vmx_feature_control_msr_valid(struct kvm_vcpu *vcpu,
1647 						 uint64_t val)
1648 {
1649 	uint64_t valid_bits = to_vmx(vcpu)->msr_ia32_feature_control_valid_bits;
1650 
1651 	return !(val & ~valid_bits);
1652 }
1653 
1654 static int vmx_get_msr_feature(struct kvm_msr_entry *msr)
1655 {
1656 	switch (msr->index) {
1657 	case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
1658 		if (!nested)
1659 			return 1;
1660 		return vmx_get_vmx_msr(&vmcs_config.nested, msr->index, &msr->data);
1661 	default:
1662 		return 1;
1663 	}
1664 
1665 	return 0;
1666 }
1667 
1668 /*
1669  * Reads an msr value (of 'msr_index') into 'pdata'.
1670  * Returns 0 on success, non-0 otherwise.
1671  * Assumes vcpu_load() was already called.
1672  */
1673 static int vmx_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1674 {
1675 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1676 	struct shared_msr_entry *msr;
1677 	u32 index;
1678 
1679 	switch (msr_info->index) {
1680 #ifdef CONFIG_X86_64
1681 	case MSR_FS_BASE:
1682 		msr_info->data = vmcs_readl(GUEST_FS_BASE);
1683 		break;
1684 	case MSR_GS_BASE:
1685 		msr_info->data = vmcs_readl(GUEST_GS_BASE);
1686 		break;
1687 	case MSR_KERNEL_GS_BASE:
1688 		msr_info->data = vmx_read_guest_kernel_gs_base(vmx);
1689 		break;
1690 #endif
1691 	case MSR_EFER:
1692 		return kvm_get_msr_common(vcpu, msr_info);
1693 	case MSR_IA32_SPEC_CTRL:
1694 		if (!msr_info->host_initiated &&
1695 		    !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
1696 			return 1;
1697 
1698 		msr_info->data = to_vmx(vcpu)->spec_ctrl;
1699 		break;
1700 	case MSR_IA32_SYSENTER_CS:
1701 		msr_info->data = vmcs_read32(GUEST_SYSENTER_CS);
1702 		break;
1703 	case MSR_IA32_SYSENTER_EIP:
1704 		msr_info->data = vmcs_readl(GUEST_SYSENTER_EIP);
1705 		break;
1706 	case MSR_IA32_SYSENTER_ESP:
1707 		msr_info->data = vmcs_readl(GUEST_SYSENTER_ESP);
1708 		break;
1709 	case MSR_IA32_BNDCFGS:
1710 		if (!kvm_mpx_supported() ||
1711 		    (!msr_info->host_initiated &&
1712 		     !guest_cpuid_has(vcpu, X86_FEATURE_MPX)))
1713 			return 1;
1714 		msr_info->data = vmcs_read64(GUEST_BNDCFGS);
1715 		break;
1716 	case MSR_IA32_MCG_EXT_CTL:
1717 		if (!msr_info->host_initiated &&
1718 		    !(vmx->msr_ia32_feature_control &
1719 		      FEATURE_CONTROL_LMCE))
1720 			return 1;
1721 		msr_info->data = vcpu->arch.mcg_ext_ctl;
1722 		break;
1723 	case MSR_IA32_FEATURE_CONTROL:
1724 		msr_info->data = vmx->msr_ia32_feature_control;
1725 		break;
1726 	case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
1727 		if (!nested_vmx_allowed(vcpu))
1728 			return 1;
1729 		return vmx_get_vmx_msr(&vmx->nested.msrs, msr_info->index,
1730 				       &msr_info->data);
1731 	case MSR_IA32_XSS:
1732 		if (!vmx_xsaves_supported() ||
1733 		    (!msr_info->host_initiated &&
1734 		     !(guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
1735 		       guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))))
1736 			return 1;
1737 		msr_info->data = vcpu->arch.ia32_xss;
1738 		break;
1739 	case MSR_IA32_RTIT_CTL:
1740 		if (pt_mode != PT_MODE_HOST_GUEST)
1741 			return 1;
1742 		msr_info->data = vmx->pt_desc.guest.ctl;
1743 		break;
1744 	case MSR_IA32_RTIT_STATUS:
1745 		if (pt_mode != PT_MODE_HOST_GUEST)
1746 			return 1;
1747 		msr_info->data = vmx->pt_desc.guest.status;
1748 		break;
1749 	case MSR_IA32_RTIT_CR3_MATCH:
1750 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
1751 			!intel_pt_validate_cap(vmx->pt_desc.caps,
1752 						PT_CAP_cr3_filtering))
1753 			return 1;
1754 		msr_info->data = vmx->pt_desc.guest.cr3_match;
1755 		break;
1756 	case MSR_IA32_RTIT_OUTPUT_BASE:
1757 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
1758 			(!intel_pt_validate_cap(vmx->pt_desc.caps,
1759 					PT_CAP_topa_output) &&
1760 			 !intel_pt_validate_cap(vmx->pt_desc.caps,
1761 					PT_CAP_single_range_output)))
1762 			return 1;
1763 		msr_info->data = vmx->pt_desc.guest.output_base;
1764 		break;
1765 	case MSR_IA32_RTIT_OUTPUT_MASK:
1766 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
1767 			(!intel_pt_validate_cap(vmx->pt_desc.caps,
1768 					PT_CAP_topa_output) &&
1769 			 !intel_pt_validate_cap(vmx->pt_desc.caps,
1770 					PT_CAP_single_range_output)))
1771 			return 1;
1772 		msr_info->data = vmx->pt_desc.guest.output_mask;
1773 		break;
1774 	case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
1775 		index = msr_info->index - MSR_IA32_RTIT_ADDR0_A;
1776 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
1777 			(index >= 2 * intel_pt_validate_cap(vmx->pt_desc.caps,
1778 					PT_CAP_num_address_ranges)))
1779 			return 1;
1780 		if (index % 2)
1781 			msr_info->data = vmx->pt_desc.guest.addr_b[index / 2];
1782 		else
1783 			msr_info->data = vmx->pt_desc.guest.addr_a[index / 2];
1784 		break;
1785 	case MSR_TSC_AUX:
1786 		if (!msr_info->host_initiated &&
1787 		    !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP))
1788 			return 1;
1789 		/* Else, falls through */
1790 	default:
1791 		msr = find_msr_entry(vmx, msr_info->index);
1792 		if (msr) {
1793 			msr_info->data = msr->data;
1794 			break;
1795 		}
1796 		return kvm_get_msr_common(vcpu, msr_info);
1797 	}
1798 
1799 	return 0;
1800 }
1801 
1802 /*
1803  * Writes msr value into into the appropriate "register".
1804  * Returns 0 on success, non-0 otherwise.
1805  * Assumes vcpu_load() was already called.
1806  */
1807 static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1808 {
1809 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1810 	struct shared_msr_entry *msr;
1811 	int ret = 0;
1812 	u32 msr_index = msr_info->index;
1813 	u64 data = msr_info->data;
1814 	u32 index;
1815 
1816 	switch (msr_index) {
1817 	case MSR_EFER:
1818 		ret = kvm_set_msr_common(vcpu, msr_info);
1819 		break;
1820 #ifdef CONFIG_X86_64
1821 	case MSR_FS_BASE:
1822 		vmx_segment_cache_clear(vmx);
1823 		vmcs_writel(GUEST_FS_BASE, data);
1824 		break;
1825 	case MSR_GS_BASE:
1826 		vmx_segment_cache_clear(vmx);
1827 		vmcs_writel(GUEST_GS_BASE, data);
1828 		break;
1829 	case MSR_KERNEL_GS_BASE:
1830 		vmx_write_guest_kernel_gs_base(vmx, data);
1831 		break;
1832 #endif
1833 	case MSR_IA32_SYSENTER_CS:
1834 		if (is_guest_mode(vcpu))
1835 			get_vmcs12(vcpu)->guest_sysenter_cs = data;
1836 		vmcs_write32(GUEST_SYSENTER_CS, data);
1837 		break;
1838 	case MSR_IA32_SYSENTER_EIP:
1839 		if (is_guest_mode(vcpu))
1840 			get_vmcs12(vcpu)->guest_sysenter_eip = data;
1841 		vmcs_writel(GUEST_SYSENTER_EIP, data);
1842 		break;
1843 	case MSR_IA32_SYSENTER_ESP:
1844 		if (is_guest_mode(vcpu))
1845 			get_vmcs12(vcpu)->guest_sysenter_esp = data;
1846 		vmcs_writel(GUEST_SYSENTER_ESP, data);
1847 		break;
1848 	case MSR_IA32_DEBUGCTLMSR:
1849 		if (is_guest_mode(vcpu) && get_vmcs12(vcpu)->vm_exit_controls &
1850 						VM_EXIT_SAVE_DEBUG_CONTROLS)
1851 			get_vmcs12(vcpu)->guest_ia32_debugctl = data;
1852 
1853 		ret = kvm_set_msr_common(vcpu, msr_info);
1854 		break;
1855 
1856 	case MSR_IA32_BNDCFGS:
1857 		if (!kvm_mpx_supported() ||
1858 		    (!msr_info->host_initiated &&
1859 		     !guest_cpuid_has(vcpu, X86_FEATURE_MPX)))
1860 			return 1;
1861 		if (is_noncanonical_address(data & PAGE_MASK, vcpu) ||
1862 		    (data & MSR_IA32_BNDCFGS_RSVD))
1863 			return 1;
1864 		vmcs_write64(GUEST_BNDCFGS, data);
1865 		break;
1866 	case MSR_IA32_SPEC_CTRL:
1867 		if (!msr_info->host_initiated &&
1868 		    !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
1869 			return 1;
1870 
1871 		/* The STIBP bit doesn't fault even if it's not advertised */
1872 		if (data & ~(SPEC_CTRL_IBRS | SPEC_CTRL_STIBP | SPEC_CTRL_SSBD))
1873 			return 1;
1874 
1875 		vmx->spec_ctrl = data;
1876 
1877 		if (!data)
1878 			break;
1879 
1880 		/*
1881 		 * For non-nested:
1882 		 * When it's written (to non-zero) for the first time, pass
1883 		 * it through.
1884 		 *
1885 		 * For nested:
1886 		 * The handling of the MSR bitmap for L2 guests is done in
1887 		 * nested_vmx_merge_msr_bitmap. We should not touch the
1888 		 * vmcs02.msr_bitmap here since it gets completely overwritten
1889 		 * in the merging. We update the vmcs01 here for L1 as well
1890 		 * since it will end up touching the MSR anyway now.
1891 		 */
1892 		vmx_disable_intercept_for_msr(vmx->vmcs01.msr_bitmap,
1893 					      MSR_IA32_SPEC_CTRL,
1894 					      MSR_TYPE_RW);
1895 		break;
1896 	case MSR_IA32_PRED_CMD:
1897 		if (!msr_info->host_initiated &&
1898 		    !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
1899 			return 1;
1900 
1901 		if (data & ~PRED_CMD_IBPB)
1902 			return 1;
1903 
1904 		if (!data)
1905 			break;
1906 
1907 		wrmsrl(MSR_IA32_PRED_CMD, PRED_CMD_IBPB);
1908 
1909 		/*
1910 		 * For non-nested:
1911 		 * When it's written (to non-zero) for the first time, pass
1912 		 * it through.
1913 		 *
1914 		 * For nested:
1915 		 * The handling of the MSR bitmap for L2 guests is done in
1916 		 * nested_vmx_merge_msr_bitmap. We should not touch the
1917 		 * vmcs02.msr_bitmap here since it gets completely overwritten
1918 		 * in the merging.
1919 		 */
1920 		vmx_disable_intercept_for_msr(vmx->vmcs01.msr_bitmap, MSR_IA32_PRED_CMD,
1921 					      MSR_TYPE_W);
1922 		break;
1923 	case MSR_IA32_CR_PAT:
1924 		if (!kvm_pat_valid(data))
1925 			return 1;
1926 
1927 		if (is_guest_mode(vcpu) &&
1928 		    get_vmcs12(vcpu)->vm_exit_controls & VM_EXIT_SAVE_IA32_PAT)
1929 			get_vmcs12(vcpu)->guest_ia32_pat = data;
1930 
1931 		if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
1932 			vmcs_write64(GUEST_IA32_PAT, data);
1933 			vcpu->arch.pat = data;
1934 			break;
1935 		}
1936 		ret = kvm_set_msr_common(vcpu, msr_info);
1937 		break;
1938 	case MSR_IA32_TSC_ADJUST:
1939 		ret = kvm_set_msr_common(vcpu, msr_info);
1940 		break;
1941 	case MSR_IA32_MCG_EXT_CTL:
1942 		if ((!msr_info->host_initiated &&
1943 		     !(to_vmx(vcpu)->msr_ia32_feature_control &
1944 		       FEATURE_CONTROL_LMCE)) ||
1945 		    (data & ~MCG_EXT_CTL_LMCE_EN))
1946 			return 1;
1947 		vcpu->arch.mcg_ext_ctl = data;
1948 		break;
1949 	case MSR_IA32_FEATURE_CONTROL:
1950 		if (!vmx_feature_control_msr_valid(vcpu, data) ||
1951 		    (to_vmx(vcpu)->msr_ia32_feature_control &
1952 		     FEATURE_CONTROL_LOCKED && !msr_info->host_initiated))
1953 			return 1;
1954 		vmx->msr_ia32_feature_control = data;
1955 		if (msr_info->host_initiated && data == 0)
1956 			vmx_leave_nested(vcpu);
1957 		break;
1958 	case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
1959 		if (!msr_info->host_initiated)
1960 			return 1; /* they are read-only */
1961 		if (!nested_vmx_allowed(vcpu))
1962 			return 1;
1963 		return vmx_set_vmx_msr(vcpu, msr_index, data);
1964 	case MSR_IA32_XSS:
1965 		if (!vmx_xsaves_supported() ||
1966 		    (!msr_info->host_initiated &&
1967 		     !(guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
1968 		       guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))))
1969 			return 1;
1970 		/*
1971 		 * The only supported bit as of Skylake is bit 8, but
1972 		 * it is not supported on KVM.
1973 		 */
1974 		if (data != 0)
1975 			return 1;
1976 		vcpu->arch.ia32_xss = data;
1977 		if (vcpu->arch.ia32_xss != host_xss)
1978 			add_atomic_switch_msr(vmx, MSR_IA32_XSS,
1979 				vcpu->arch.ia32_xss, host_xss, false);
1980 		else
1981 			clear_atomic_switch_msr(vmx, MSR_IA32_XSS);
1982 		break;
1983 	case MSR_IA32_RTIT_CTL:
1984 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
1985 			vmx_rtit_ctl_check(vcpu, data) ||
1986 			vmx->nested.vmxon)
1987 			return 1;
1988 		vmcs_write64(GUEST_IA32_RTIT_CTL, data);
1989 		vmx->pt_desc.guest.ctl = data;
1990 		pt_update_intercept_for_msr(vmx);
1991 		break;
1992 	case MSR_IA32_RTIT_STATUS:
1993 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
1994 			(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) ||
1995 			(data & MSR_IA32_RTIT_STATUS_MASK))
1996 			return 1;
1997 		vmx->pt_desc.guest.status = data;
1998 		break;
1999 	case MSR_IA32_RTIT_CR3_MATCH:
2000 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
2001 			(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) ||
2002 			!intel_pt_validate_cap(vmx->pt_desc.caps,
2003 						PT_CAP_cr3_filtering))
2004 			return 1;
2005 		vmx->pt_desc.guest.cr3_match = data;
2006 		break;
2007 	case MSR_IA32_RTIT_OUTPUT_BASE:
2008 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
2009 			(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) ||
2010 			(!intel_pt_validate_cap(vmx->pt_desc.caps,
2011 					PT_CAP_topa_output) &&
2012 			 !intel_pt_validate_cap(vmx->pt_desc.caps,
2013 					PT_CAP_single_range_output)) ||
2014 			(data & MSR_IA32_RTIT_OUTPUT_BASE_MASK))
2015 			return 1;
2016 		vmx->pt_desc.guest.output_base = data;
2017 		break;
2018 	case MSR_IA32_RTIT_OUTPUT_MASK:
2019 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
2020 			(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) ||
2021 			(!intel_pt_validate_cap(vmx->pt_desc.caps,
2022 					PT_CAP_topa_output) &&
2023 			 !intel_pt_validate_cap(vmx->pt_desc.caps,
2024 					PT_CAP_single_range_output)))
2025 			return 1;
2026 		vmx->pt_desc.guest.output_mask = data;
2027 		break;
2028 	case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
2029 		index = msr_info->index - MSR_IA32_RTIT_ADDR0_A;
2030 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
2031 			(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) ||
2032 			(index >= 2 * intel_pt_validate_cap(vmx->pt_desc.caps,
2033 					PT_CAP_num_address_ranges)))
2034 			return 1;
2035 		if (index % 2)
2036 			vmx->pt_desc.guest.addr_b[index / 2] = data;
2037 		else
2038 			vmx->pt_desc.guest.addr_a[index / 2] = data;
2039 		break;
2040 	case MSR_TSC_AUX:
2041 		if (!msr_info->host_initiated &&
2042 		    !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP))
2043 			return 1;
2044 		/* Check reserved bit, higher 32 bits should be zero */
2045 		if ((data >> 32) != 0)
2046 			return 1;
2047 		/* Else, falls through */
2048 	default:
2049 		msr = find_msr_entry(vmx, msr_index);
2050 		if (msr) {
2051 			u64 old_msr_data = msr->data;
2052 			msr->data = data;
2053 			if (msr - vmx->guest_msrs < vmx->save_nmsrs) {
2054 				preempt_disable();
2055 				ret = kvm_set_shared_msr(msr->index, msr->data,
2056 							 msr->mask);
2057 				preempt_enable();
2058 				if (ret)
2059 					msr->data = old_msr_data;
2060 			}
2061 			break;
2062 		}
2063 		ret = kvm_set_msr_common(vcpu, msr_info);
2064 	}
2065 
2066 	return ret;
2067 }
2068 
2069 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
2070 {
2071 	__set_bit(reg, (unsigned long *)&vcpu->arch.regs_avail);
2072 	switch (reg) {
2073 	case VCPU_REGS_RSP:
2074 		vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
2075 		break;
2076 	case VCPU_REGS_RIP:
2077 		vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
2078 		break;
2079 	case VCPU_EXREG_PDPTR:
2080 		if (enable_ept)
2081 			ept_save_pdptrs(vcpu);
2082 		break;
2083 	default:
2084 		break;
2085 	}
2086 }
2087 
2088 static __init int cpu_has_kvm_support(void)
2089 {
2090 	return cpu_has_vmx();
2091 }
2092 
2093 static __init int vmx_disabled_by_bios(void)
2094 {
2095 	u64 msr;
2096 
2097 	rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
2098 	if (msr & FEATURE_CONTROL_LOCKED) {
2099 		/* launched w/ TXT and VMX disabled */
2100 		if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
2101 			&& tboot_enabled())
2102 			return 1;
2103 		/* launched w/o TXT and VMX only enabled w/ TXT */
2104 		if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
2105 			&& (msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
2106 			&& !tboot_enabled()) {
2107 			printk(KERN_WARNING "kvm: disable TXT in the BIOS or "
2108 				"activate TXT before enabling KVM\n");
2109 			return 1;
2110 		}
2111 		/* launched w/o TXT and VMX disabled */
2112 		if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
2113 			&& !tboot_enabled())
2114 			return 1;
2115 	}
2116 
2117 	return 0;
2118 }
2119 
2120 static void kvm_cpu_vmxon(u64 addr)
2121 {
2122 	cr4_set_bits(X86_CR4_VMXE);
2123 	intel_pt_handle_vmx(1);
2124 
2125 	asm volatile ("vmxon %0" : : "m"(addr));
2126 }
2127 
2128 static int hardware_enable(void)
2129 {
2130 	int cpu = raw_smp_processor_id();
2131 	u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
2132 	u64 old, test_bits;
2133 
2134 	if (cr4_read_shadow() & X86_CR4_VMXE)
2135 		return -EBUSY;
2136 
2137 	/*
2138 	 * This can happen if we hot-added a CPU but failed to allocate
2139 	 * VP assist page for it.
2140 	 */
2141 	if (static_branch_unlikely(&enable_evmcs) &&
2142 	    !hv_get_vp_assist_page(cpu))
2143 		return -EFAULT;
2144 
2145 	INIT_LIST_HEAD(&per_cpu(loaded_vmcss_on_cpu, cpu));
2146 	INIT_LIST_HEAD(&per_cpu(blocked_vcpu_on_cpu, cpu));
2147 	spin_lock_init(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
2148 
2149 	/*
2150 	 * Now we can enable the vmclear operation in kdump
2151 	 * since the loaded_vmcss_on_cpu list on this cpu
2152 	 * has been initialized.
2153 	 *
2154 	 * Though the cpu is not in VMX operation now, there
2155 	 * is no problem to enable the vmclear operation
2156 	 * for the loaded_vmcss_on_cpu list is empty!
2157 	 */
2158 	crash_enable_local_vmclear(cpu);
2159 
2160 	rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
2161 
2162 	test_bits = FEATURE_CONTROL_LOCKED;
2163 	test_bits |= FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
2164 	if (tboot_enabled())
2165 		test_bits |= FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX;
2166 
2167 	if ((old & test_bits) != test_bits) {
2168 		/* enable and lock */
2169 		wrmsrl(MSR_IA32_FEATURE_CONTROL, old | test_bits);
2170 	}
2171 	kvm_cpu_vmxon(phys_addr);
2172 	if (enable_ept)
2173 		ept_sync_global();
2174 
2175 	return 0;
2176 }
2177 
2178 static void vmclear_local_loaded_vmcss(void)
2179 {
2180 	int cpu = raw_smp_processor_id();
2181 	struct loaded_vmcs *v, *n;
2182 
2183 	list_for_each_entry_safe(v, n, &per_cpu(loaded_vmcss_on_cpu, cpu),
2184 				 loaded_vmcss_on_cpu_link)
2185 		__loaded_vmcs_clear(v);
2186 }
2187 
2188 
2189 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
2190  * tricks.
2191  */
2192 static void kvm_cpu_vmxoff(void)
2193 {
2194 	asm volatile (__ex("vmxoff"));
2195 
2196 	intel_pt_handle_vmx(0);
2197 	cr4_clear_bits(X86_CR4_VMXE);
2198 }
2199 
2200 static void hardware_disable(void)
2201 {
2202 	vmclear_local_loaded_vmcss();
2203 	kvm_cpu_vmxoff();
2204 }
2205 
2206 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
2207 				      u32 msr, u32 *result)
2208 {
2209 	u32 vmx_msr_low, vmx_msr_high;
2210 	u32 ctl = ctl_min | ctl_opt;
2211 
2212 	rdmsr(msr, vmx_msr_low, vmx_msr_high);
2213 
2214 	ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
2215 	ctl |= vmx_msr_low;  /* bit == 1 in low word  ==> must be one  */
2216 
2217 	/* Ensure minimum (required) set of control bits are supported. */
2218 	if (ctl_min & ~ctl)
2219 		return -EIO;
2220 
2221 	*result = ctl;
2222 	return 0;
2223 }
2224 
2225 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf,
2226 				    struct vmx_capability *vmx_cap)
2227 {
2228 	u32 vmx_msr_low, vmx_msr_high;
2229 	u32 min, opt, min2, opt2;
2230 	u32 _pin_based_exec_control = 0;
2231 	u32 _cpu_based_exec_control = 0;
2232 	u32 _cpu_based_2nd_exec_control = 0;
2233 	u32 _vmexit_control = 0;
2234 	u32 _vmentry_control = 0;
2235 
2236 	memset(vmcs_conf, 0, sizeof(*vmcs_conf));
2237 	min = CPU_BASED_HLT_EXITING |
2238 #ifdef CONFIG_X86_64
2239 	      CPU_BASED_CR8_LOAD_EXITING |
2240 	      CPU_BASED_CR8_STORE_EXITING |
2241 #endif
2242 	      CPU_BASED_CR3_LOAD_EXITING |
2243 	      CPU_BASED_CR3_STORE_EXITING |
2244 	      CPU_BASED_UNCOND_IO_EXITING |
2245 	      CPU_BASED_MOV_DR_EXITING |
2246 	      CPU_BASED_USE_TSC_OFFSETING |
2247 	      CPU_BASED_MWAIT_EXITING |
2248 	      CPU_BASED_MONITOR_EXITING |
2249 	      CPU_BASED_INVLPG_EXITING |
2250 	      CPU_BASED_RDPMC_EXITING;
2251 
2252 	opt = CPU_BASED_TPR_SHADOW |
2253 	      CPU_BASED_USE_MSR_BITMAPS |
2254 	      CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
2255 	if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
2256 				&_cpu_based_exec_control) < 0)
2257 		return -EIO;
2258 #ifdef CONFIG_X86_64
2259 	if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2260 		_cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
2261 					   ~CPU_BASED_CR8_STORE_EXITING;
2262 #endif
2263 	if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
2264 		min2 = 0;
2265 		opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2266 			SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2267 			SECONDARY_EXEC_WBINVD_EXITING |
2268 			SECONDARY_EXEC_ENABLE_VPID |
2269 			SECONDARY_EXEC_ENABLE_EPT |
2270 			SECONDARY_EXEC_UNRESTRICTED_GUEST |
2271 			SECONDARY_EXEC_PAUSE_LOOP_EXITING |
2272 			SECONDARY_EXEC_DESC |
2273 			SECONDARY_EXEC_RDTSCP |
2274 			SECONDARY_EXEC_ENABLE_INVPCID |
2275 			SECONDARY_EXEC_APIC_REGISTER_VIRT |
2276 			SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
2277 			SECONDARY_EXEC_SHADOW_VMCS |
2278 			SECONDARY_EXEC_XSAVES |
2279 			SECONDARY_EXEC_RDSEED_EXITING |
2280 			SECONDARY_EXEC_RDRAND_EXITING |
2281 			SECONDARY_EXEC_ENABLE_PML |
2282 			SECONDARY_EXEC_TSC_SCALING |
2283 			SECONDARY_EXEC_PT_USE_GPA |
2284 			SECONDARY_EXEC_PT_CONCEAL_VMX |
2285 			SECONDARY_EXEC_ENABLE_VMFUNC |
2286 			SECONDARY_EXEC_ENCLS_EXITING;
2287 		if (adjust_vmx_controls(min2, opt2,
2288 					MSR_IA32_VMX_PROCBASED_CTLS2,
2289 					&_cpu_based_2nd_exec_control) < 0)
2290 			return -EIO;
2291 	}
2292 #ifndef CONFIG_X86_64
2293 	if (!(_cpu_based_2nd_exec_control &
2294 				SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
2295 		_cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
2296 #endif
2297 
2298 	if (!(_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2299 		_cpu_based_2nd_exec_control &= ~(
2300 				SECONDARY_EXEC_APIC_REGISTER_VIRT |
2301 				SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2302 				SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
2303 
2304 	rdmsr_safe(MSR_IA32_VMX_EPT_VPID_CAP,
2305 		&vmx_cap->ept, &vmx_cap->vpid);
2306 
2307 	if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
2308 		/* CR3 accesses and invlpg don't need to cause VM Exits when EPT
2309 		   enabled */
2310 		_cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
2311 					     CPU_BASED_CR3_STORE_EXITING |
2312 					     CPU_BASED_INVLPG_EXITING);
2313 	} else if (vmx_cap->ept) {
2314 		vmx_cap->ept = 0;
2315 		pr_warn_once("EPT CAP should not exist if not support "
2316 				"1-setting enable EPT VM-execution control\n");
2317 	}
2318 	if (!(_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_VPID) &&
2319 		vmx_cap->vpid) {
2320 		vmx_cap->vpid = 0;
2321 		pr_warn_once("VPID CAP should not exist if not support "
2322 				"1-setting enable VPID VM-execution control\n");
2323 	}
2324 
2325 	min = VM_EXIT_SAVE_DEBUG_CONTROLS | VM_EXIT_ACK_INTR_ON_EXIT;
2326 #ifdef CONFIG_X86_64
2327 	min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
2328 #endif
2329 	opt = VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL |
2330 	      VM_EXIT_LOAD_IA32_PAT |
2331 	      VM_EXIT_LOAD_IA32_EFER |
2332 	      VM_EXIT_CLEAR_BNDCFGS |
2333 	      VM_EXIT_PT_CONCEAL_PIP |
2334 	      VM_EXIT_CLEAR_IA32_RTIT_CTL;
2335 	if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
2336 				&_vmexit_control) < 0)
2337 		return -EIO;
2338 
2339 	min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
2340 	opt = PIN_BASED_VIRTUAL_NMIS | PIN_BASED_POSTED_INTR |
2341 		 PIN_BASED_VMX_PREEMPTION_TIMER;
2342 	if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
2343 				&_pin_based_exec_control) < 0)
2344 		return -EIO;
2345 
2346 	if (cpu_has_broken_vmx_preemption_timer())
2347 		_pin_based_exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
2348 	if (!(_cpu_based_2nd_exec_control &
2349 		SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY))
2350 		_pin_based_exec_control &= ~PIN_BASED_POSTED_INTR;
2351 
2352 	min = VM_ENTRY_LOAD_DEBUG_CONTROLS;
2353 	opt = VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL |
2354 	      VM_ENTRY_LOAD_IA32_PAT |
2355 	      VM_ENTRY_LOAD_IA32_EFER |
2356 	      VM_ENTRY_LOAD_BNDCFGS |
2357 	      VM_ENTRY_PT_CONCEAL_PIP |
2358 	      VM_ENTRY_LOAD_IA32_RTIT_CTL;
2359 	if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
2360 				&_vmentry_control) < 0)
2361 		return -EIO;
2362 
2363 	/*
2364 	 * Some cpus support VM_{ENTRY,EXIT}_IA32_PERF_GLOBAL_CTRL but they
2365 	 * can't be used due to an errata where VM Exit may incorrectly clear
2366 	 * IA32_PERF_GLOBAL_CTRL[34:32].  Workaround the errata by using the
2367 	 * MSR load mechanism to switch IA32_PERF_GLOBAL_CTRL.
2368 	 */
2369 	if (boot_cpu_data.x86 == 0x6) {
2370 		switch (boot_cpu_data.x86_model) {
2371 		case 26: /* AAK155 */
2372 		case 30: /* AAP115 */
2373 		case 37: /* AAT100 */
2374 		case 44: /* BC86,AAY89,BD102 */
2375 		case 46: /* BA97 */
2376 			_vmentry_control &= ~VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL;
2377 			_vmexit_control &= ~VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL;
2378 			pr_warn_once("kvm: VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL "
2379 					"does not work properly. Using workaround\n");
2380 			break;
2381 		default:
2382 			break;
2383 		}
2384 	}
2385 
2386 
2387 	rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
2388 
2389 	/* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
2390 	if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
2391 		return -EIO;
2392 
2393 #ifdef CONFIG_X86_64
2394 	/* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
2395 	if (vmx_msr_high & (1u<<16))
2396 		return -EIO;
2397 #endif
2398 
2399 	/* Require Write-Back (WB) memory type for VMCS accesses. */
2400 	if (((vmx_msr_high >> 18) & 15) != 6)
2401 		return -EIO;
2402 
2403 	vmcs_conf->size = vmx_msr_high & 0x1fff;
2404 	vmcs_conf->order = get_order(vmcs_conf->size);
2405 	vmcs_conf->basic_cap = vmx_msr_high & ~0x1fff;
2406 
2407 	vmcs_conf->revision_id = vmx_msr_low;
2408 
2409 	vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
2410 	vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
2411 	vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
2412 	vmcs_conf->vmexit_ctrl         = _vmexit_control;
2413 	vmcs_conf->vmentry_ctrl        = _vmentry_control;
2414 
2415 	if (static_branch_unlikely(&enable_evmcs))
2416 		evmcs_sanitize_exec_ctrls(vmcs_conf);
2417 
2418 	return 0;
2419 }
2420 
2421 struct vmcs *alloc_vmcs_cpu(bool shadow, int cpu, gfp_t flags)
2422 {
2423 	int node = cpu_to_node(cpu);
2424 	struct page *pages;
2425 	struct vmcs *vmcs;
2426 
2427 	pages = __alloc_pages_node(node, flags, vmcs_config.order);
2428 	if (!pages)
2429 		return NULL;
2430 	vmcs = page_address(pages);
2431 	memset(vmcs, 0, vmcs_config.size);
2432 
2433 	/* KVM supports Enlightened VMCS v1 only */
2434 	if (static_branch_unlikely(&enable_evmcs))
2435 		vmcs->hdr.revision_id = KVM_EVMCS_VERSION;
2436 	else
2437 		vmcs->hdr.revision_id = vmcs_config.revision_id;
2438 
2439 	if (shadow)
2440 		vmcs->hdr.shadow_vmcs = 1;
2441 	return vmcs;
2442 }
2443 
2444 void free_vmcs(struct vmcs *vmcs)
2445 {
2446 	free_pages((unsigned long)vmcs, vmcs_config.order);
2447 }
2448 
2449 /*
2450  * Free a VMCS, but before that VMCLEAR it on the CPU where it was last loaded
2451  */
2452 void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
2453 {
2454 	if (!loaded_vmcs->vmcs)
2455 		return;
2456 	loaded_vmcs_clear(loaded_vmcs);
2457 	free_vmcs(loaded_vmcs->vmcs);
2458 	loaded_vmcs->vmcs = NULL;
2459 	if (loaded_vmcs->msr_bitmap)
2460 		free_page((unsigned long)loaded_vmcs->msr_bitmap);
2461 	WARN_ON(loaded_vmcs->shadow_vmcs != NULL);
2462 }
2463 
2464 int alloc_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
2465 {
2466 	loaded_vmcs->vmcs = alloc_vmcs(false);
2467 	if (!loaded_vmcs->vmcs)
2468 		return -ENOMEM;
2469 
2470 	loaded_vmcs->shadow_vmcs = NULL;
2471 	loaded_vmcs->hv_timer_soft_disabled = false;
2472 	loaded_vmcs_init(loaded_vmcs);
2473 
2474 	if (cpu_has_vmx_msr_bitmap()) {
2475 		loaded_vmcs->msr_bitmap = (unsigned long *)
2476 				__get_free_page(GFP_KERNEL_ACCOUNT);
2477 		if (!loaded_vmcs->msr_bitmap)
2478 			goto out_vmcs;
2479 		memset(loaded_vmcs->msr_bitmap, 0xff, PAGE_SIZE);
2480 
2481 		if (IS_ENABLED(CONFIG_HYPERV) &&
2482 		    static_branch_unlikely(&enable_evmcs) &&
2483 		    (ms_hyperv.nested_features & HV_X64_NESTED_MSR_BITMAP)) {
2484 			struct hv_enlightened_vmcs *evmcs =
2485 				(struct hv_enlightened_vmcs *)loaded_vmcs->vmcs;
2486 
2487 			evmcs->hv_enlightenments_control.msr_bitmap = 1;
2488 		}
2489 	}
2490 
2491 	memset(&loaded_vmcs->host_state, 0, sizeof(struct vmcs_host_state));
2492 	memset(&loaded_vmcs->controls_shadow, 0,
2493 		sizeof(struct vmcs_controls_shadow));
2494 
2495 	return 0;
2496 
2497 out_vmcs:
2498 	free_loaded_vmcs(loaded_vmcs);
2499 	return -ENOMEM;
2500 }
2501 
2502 static void free_kvm_area(void)
2503 {
2504 	int cpu;
2505 
2506 	for_each_possible_cpu(cpu) {
2507 		free_vmcs(per_cpu(vmxarea, cpu));
2508 		per_cpu(vmxarea, cpu) = NULL;
2509 	}
2510 }
2511 
2512 static __init int alloc_kvm_area(void)
2513 {
2514 	int cpu;
2515 
2516 	for_each_possible_cpu(cpu) {
2517 		struct vmcs *vmcs;
2518 
2519 		vmcs = alloc_vmcs_cpu(false, cpu, GFP_KERNEL);
2520 		if (!vmcs) {
2521 			free_kvm_area();
2522 			return -ENOMEM;
2523 		}
2524 
2525 		/*
2526 		 * When eVMCS is enabled, alloc_vmcs_cpu() sets
2527 		 * vmcs->revision_id to KVM_EVMCS_VERSION instead of
2528 		 * revision_id reported by MSR_IA32_VMX_BASIC.
2529 		 *
2530 		 * However, even though not explicitly documented by
2531 		 * TLFS, VMXArea passed as VMXON argument should
2532 		 * still be marked with revision_id reported by
2533 		 * physical CPU.
2534 		 */
2535 		if (static_branch_unlikely(&enable_evmcs))
2536 			vmcs->hdr.revision_id = vmcs_config.revision_id;
2537 
2538 		per_cpu(vmxarea, cpu) = vmcs;
2539 	}
2540 	return 0;
2541 }
2542 
2543 static void fix_pmode_seg(struct kvm_vcpu *vcpu, int seg,
2544 		struct kvm_segment *save)
2545 {
2546 	if (!emulate_invalid_guest_state) {
2547 		/*
2548 		 * CS and SS RPL should be equal during guest entry according
2549 		 * to VMX spec, but in reality it is not always so. Since vcpu
2550 		 * is in the middle of the transition from real mode to
2551 		 * protected mode it is safe to assume that RPL 0 is a good
2552 		 * default value.
2553 		 */
2554 		if (seg == VCPU_SREG_CS || seg == VCPU_SREG_SS)
2555 			save->selector &= ~SEGMENT_RPL_MASK;
2556 		save->dpl = save->selector & SEGMENT_RPL_MASK;
2557 		save->s = 1;
2558 	}
2559 	vmx_set_segment(vcpu, save, seg);
2560 }
2561 
2562 static void enter_pmode(struct kvm_vcpu *vcpu)
2563 {
2564 	unsigned long flags;
2565 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2566 
2567 	/*
2568 	 * Update real mode segment cache. It may be not up-to-date if sement
2569 	 * register was written while vcpu was in a guest mode.
2570 	 */
2571 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
2572 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
2573 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
2574 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
2575 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
2576 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
2577 
2578 	vmx->rmode.vm86_active = 0;
2579 
2580 	vmx_segment_cache_clear(vmx);
2581 
2582 	vmx_set_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
2583 
2584 	flags = vmcs_readl(GUEST_RFLAGS);
2585 	flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
2586 	flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
2587 	vmcs_writel(GUEST_RFLAGS, flags);
2588 
2589 	vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
2590 			(vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
2591 
2592 	update_exception_bitmap(vcpu);
2593 
2594 	fix_pmode_seg(vcpu, VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
2595 	fix_pmode_seg(vcpu, VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
2596 	fix_pmode_seg(vcpu, VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
2597 	fix_pmode_seg(vcpu, VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
2598 	fix_pmode_seg(vcpu, VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
2599 	fix_pmode_seg(vcpu, VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
2600 }
2601 
2602 static void fix_rmode_seg(int seg, struct kvm_segment *save)
2603 {
2604 	const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2605 	struct kvm_segment var = *save;
2606 
2607 	var.dpl = 0x3;
2608 	if (seg == VCPU_SREG_CS)
2609 		var.type = 0x3;
2610 
2611 	if (!emulate_invalid_guest_state) {
2612 		var.selector = var.base >> 4;
2613 		var.base = var.base & 0xffff0;
2614 		var.limit = 0xffff;
2615 		var.g = 0;
2616 		var.db = 0;
2617 		var.present = 1;
2618 		var.s = 1;
2619 		var.l = 0;
2620 		var.unusable = 0;
2621 		var.type = 0x3;
2622 		var.avl = 0;
2623 		if (save->base & 0xf)
2624 			printk_once(KERN_WARNING "kvm: segment base is not "
2625 					"paragraph aligned when entering "
2626 					"protected mode (seg=%d)", seg);
2627 	}
2628 
2629 	vmcs_write16(sf->selector, var.selector);
2630 	vmcs_writel(sf->base, var.base);
2631 	vmcs_write32(sf->limit, var.limit);
2632 	vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(&var));
2633 }
2634 
2635 static void enter_rmode(struct kvm_vcpu *vcpu)
2636 {
2637 	unsigned long flags;
2638 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2639 	struct kvm_vmx *kvm_vmx = to_kvm_vmx(vcpu->kvm);
2640 
2641 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
2642 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
2643 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
2644 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
2645 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
2646 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
2647 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
2648 
2649 	vmx->rmode.vm86_active = 1;
2650 
2651 	/*
2652 	 * Very old userspace does not call KVM_SET_TSS_ADDR before entering
2653 	 * vcpu. Warn the user that an update is overdue.
2654 	 */
2655 	if (!kvm_vmx->tss_addr)
2656 		printk_once(KERN_WARNING "kvm: KVM_SET_TSS_ADDR need to be "
2657 			     "called before entering vcpu\n");
2658 
2659 	vmx_segment_cache_clear(vmx);
2660 
2661 	vmcs_writel(GUEST_TR_BASE, kvm_vmx->tss_addr);
2662 	vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
2663 	vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
2664 
2665 	flags = vmcs_readl(GUEST_RFLAGS);
2666 	vmx->rmode.save_rflags = flags;
2667 
2668 	flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
2669 
2670 	vmcs_writel(GUEST_RFLAGS, flags);
2671 	vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
2672 	update_exception_bitmap(vcpu);
2673 
2674 	fix_rmode_seg(VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
2675 	fix_rmode_seg(VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
2676 	fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
2677 	fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
2678 	fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
2679 	fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
2680 
2681 	kvm_mmu_reset_context(vcpu);
2682 }
2683 
2684 void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
2685 {
2686 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2687 	struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
2688 
2689 	if (!msr)
2690 		return;
2691 
2692 	vcpu->arch.efer = efer;
2693 	if (efer & EFER_LMA) {
2694 		vm_entry_controls_setbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
2695 		msr->data = efer;
2696 	} else {
2697 		vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
2698 
2699 		msr->data = efer & ~EFER_LME;
2700 	}
2701 	setup_msrs(vmx);
2702 }
2703 
2704 #ifdef CONFIG_X86_64
2705 
2706 static void enter_lmode(struct kvm_vcpu *vcpu)
2707 {
2708 	u32 guest_tr_ar;
2709 
2710 	vmx_segment_cache_clear(to_vmx(vcpu));
2711 
2712 	guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
2713 	if ((guest_tr_ar & VMX_AR_TYPE_MASK) != VMX_AR_TYPE_BUSY_64_TSS) {
2714 		pr_debug_ratelimited("%s: tss fixup for long mode. \n",
2715 				     __func__);
2716 		vmcs_write32(GUEST_TR_AR_BYTES,
2717 			     (guest_tr_ar & ~VMX_AR_TYPE_MASK)
2718 			     | VMX_AR_TYPE_BUSY_64_TSS);
2719 	}
2720 	vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
2721 }
2722 
2723 static void exit_lmode(struct kvm_vcpu *vcpu)
2724 {
2725 	vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
2726 	vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
2727 }
2728 
2729 #endif
2730 
2731 static void vmx_flush_tlb_gva(struct kvm_vcpu *vcpu, gva_t addr)
2732 {
2733 	int vpid = to_vmx(vcpu)->vpid;
2734 
2735 	if (!vpid_sync_vcpu_addr(vpid, addr))
2736 		vpid_sync_context(vpid);
2737 
2738 	/*
2739 	 * If VPIDs are not supported or enabled, then the above is a no-op.
2740 	 * But we don't really need a TLB flush in that case anyway, because
2741 	 * each VM entry/exit includes an implicit flush when VPID is 0.
2742 	 */
2743 }
2744 
2745 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
2746 {
2747 	ulong cr0_guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
2748 
2749 	vcpu->arch.cr0 &= ~cr0_guest_owned_bits;
2750 	vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & cr0_guest_owned_bits;
2751 }
2752 
2753 static void vmx_decache_cr3(struct kvm_vcpu *vcpu)
2754 {
2755 	if (enable_unrestricted_guest || (enable_ept && is_paging(vcpu)))
2756 		vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
2757 	__set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
2758 }
2759 
2760 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
2761 {
2762 	ulong cr4_guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
2763 
2764 	vcpu->arch.cr4 &= ~cr4_guest_owned_bits;
2765 	vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & cr4_guest_owned_bits;
2766 }
2767 
2768 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
2769 {
2770 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
2771 
2772 	if (!test_bit(VCPU_EXREG_PDPTR,
2773 		      (unsigned long *)&vcpu->arch.regs_dirty))
2774 		return;
2775 
2776 	if (is_pae_paging(vcpu)) {
2777 		vmcs_write64(GUEST_PDPTR0, mmu->pdptrs[0]);
2778 		vmcs_write64(GUEST_PDPTR1, mmu->pdptrs[1]);
2779 		vmcs_write64(GUEST_PDPTR2, mmu->pdptrs[2]);
2780 		vmcs_write64(GUEST_PDPTR3, mmu->pdptrs[3]);
2781 	}
2782 }
2783 
2784 void ept_save_pdptrs(struct kvm_vcpu *vcpu)
2785 {
2786 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
2787 
2788 	if (is_pae_paging(vcpu)) {
2789 		mmu->pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
2790 		mmu->pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
2791 		mmu->pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
2792 		mmu->pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
2793 	}
2794 
2795 	__set_bit(VCPU_EXREG_PDPTR,
2796 		  (unsigned long *)&vcpu->arch.regs_avail);
2797 	__set_bit(VCPU_EXREG_PDPTR,
2798 		  (unsigned long *)&vcpu->arch.regs_dirty);
2799 }
2800 
2801 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
2802 					unsigned long cr0,
2803 					struct kvm_vcpu *vcpu)
2804 {
2805 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2806 
2807 	if (!test_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail))
2808 		vmx_decache_cr3(vcpu);
2809 	if (!(cr0 & X86_CR0_PG)) {
2810 		/* From paging/starting to nonpaging */
2811 		exec_controls_setbit(vmx, CPU_BASED_CR3_LOAD_EXITING |
2812 					  CPU_BASED_CR3_STORE_EXITING);
2813 		vcpu->arch.cr0 = cr0;
2814 		vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
2815 	} else if (!is_paging(vcpu)) {
2816 		/* From nonpaging to paging */
2817 		exec_controls_clearbit(vmx, CPU_BASED_CR3_LOAD_EXITING |
2818 					    CPU_BASED_CR3_STORE_EXITING);
2819 		vcpu->arch.cr0 = cr0;
2820 		vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
2821 	}
2822 
2823 	if (!(cr0 & X86_CR0_WP))
2824 		*hw_cr0 &= ~X86_CR0_WP;
2825 }
2826 
2827 void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
2828 {
2829 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2830 	unsigned long hw_cr0;
2831 
2832 	hw_cr0 = (cr0 & ~KVM_VM_CR0_ALWAYS_OFF);
2833 	if (enable_unrestricted_guest)
2834 		hw_cr0 |= KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
2835 	else {
2836 		hw_cr0 |= KVM_VM_CR0_ALWAYS_ON;
2837 
2838 		if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
2839 			enter_pmode(vcpu);
2840 
2841 		if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
2842 			enter_rmode(vcpu);
2843 	}
2844 
2845 #ifdef CONFIG_X86_64
2846 	if (vcpu->arch.efer & EFER_LME) {
2847 		if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
2848 			enter_lmode(vcpu);
2849 		if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
2850 			exit_lmode(vcpu);
2851 	}
2852 #endif
2853 
2854 	if (enable_ept && !enable_unrestricted_guest)
2855 		ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
2856 
2857 	vmcs_writel(CR0_READ_SHADOW, cr0);
2858 	vmcs_writel(GUEST_CR0, hw_cr0);
2859 	vcpu->arch.cr0 = cr0;
2860 
2861 	/* depends on vcpu->arch.cr0 to be set to a new value */
2862 	vmx->emulation_required = emulation_required(vcpu);
2863 }
2864 
2865 static int get_ept_level(struct kvm_vcpu *vcpu)
2866 {
2867 	if (cpu_has_vmx_ept_5levels() && (cpuid_maxphyaddr(vcpu) > 48))
2868 		return 5;
2869 	return 4;
2870 }
2871 
2872 u64 construct_eptp(struct kvm_vcpu *vcpu, unsigned long root_hpa)
2873 {
2874 	u64 eptp = VMX_EPTP_MT_WB;
2875 
2876 	eptp |= (get_ept_level(vcpu) == 5) ? VMX_EPTP_PWL_5 : VMX_EPTP_PWL_4;
2877 
2878 	if (enable_ept_ad_bits &&
2879 	    (!is_guest_mode(vcpu) || nested_ept_ad_enabled(vcpu)))
2880 		eptp |= VMX_EPTP_AD_ENABLE_BIT;
2881 	eptp |= (root_hpa & PAGE_MASK);
2882 
2883 	return eptp;
2884 }
2885 
2886 void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
2887 {
2888 	struct kvm *kvm = vcpu->kvm;
2889 	unsigned long guest_cr3;
2890 	u64 eptp;
2891 
2892 	guest_cr3 = cr3;
2893 	if (enable_ept) {
2894 		eptp = construct_eptp(vcpu, cr3);
2895 		vmcs_write64(EPT_POINTER, eptp);
2896 
2897 		if (kvm_x86_ops->tlb_remote_flush) {
2898 			spin_lock(&to_kvm_vmx(kvm)->ept_pointer_lock);
2899 			to_vmx(vcpu)->ept_pointer = eptp;
2900 			to_kvm_vmx(kvm)->ept_pointers_match
2901 				= EPT_POINTERS_CHECK;
2902 			spin_unlock(&to_kvm_vmx(kvm)->ept_pointer_lock);
2903 		}
2904 
2905 		if (enable_unrestricted_guest || is_paging(vcpu) ||
2906 		    is_guest_mode(vcpu))
2907 			guest_cr3 = kvm_read_cr3(vcpu);
2908 		else
2909 			guest_cr3 = to_kvm_vmx(kvm)->ept_identity_map_addr;
2910 		ept_load_pdptrs(vcpu);
2911 	}
2912 
2913 	vmcs_writel(GUEST_CR3, guest_cr3);
2914 }
2915 
2916 int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
2917 {
2918 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2919 	/*
2920 	 * Pass through host's Machine Check Enable value to hw_cr4, which
2921 	 * is in force while we are in guest mode.  Do not let guests control
2922 	 * this bit, even if host CR4.MCE == 0.
2923 	 */
2924 	unsigned long hw_cr4;
2925 
2926 	hw_cr4 = (cr4_read_shadow() & X86_CR4_MCE) | (cr4 & ~X86_CR4_MCE);
2927 	if (enable_unrestricted_guest)
2928 		hw_cr4 |= KVM_VM_CR4_ALWAYS_ON_UNRESTRICTED_GUEST;
2929 	else if (vmx->rmode.vm86_active)
2930 		hw_cr4 |= KVM_RMODE_VM_CR4_ALWAYS_ON;
2931 	else
2932 		hw_cr4 |= KVM_PMODE_VM_CR4_ALWAYS_ON;
2933 
2934 	if (!boot_cpu_has(X86_FEATURE_UMIP) && vmx_umip_emulated()) {
2935 		if (cr4 & X86_CR4_UMIP) {
2936 			secondary_exec_controls_setbit(vmx, SECONDARY_EXEC_DESC);
2937 			hw_cr4 &= ~X86_CR4_UMIP;
2938 		} else if (!is_guest_mode(vcpu) ||
2939 			!nested_cpu_has2(get_vmcs12(vcpu), SECONDARY_EXEC_DESC)) {
2940 			secondary_exec_controls_clearbit(vmx, SECONDARY_EXEC_DESC);
2941 		}
2942 	}
2943 
2944 	if (cr4 & X86_CR4_VMXE) {
2945 		/*
2946 		 * To use VMXON (and later other VMX instructions), a guest
2947 		 * must first be able to turn on cr4.VMXE (see handle_vmon()).
2948 		 * So basically the check on whether to allow nested VMX
2949 		 * is here.  We operate under the default treatment of SMM,
2950 		 * so VMX cannot be enabled under SMM.
2951 		 */
2952 		if (!nested_vmx_allowed(vcpu) || is_smm(vcpu))
2953 			return 1;
2954 	}
2955 
2956 	if (vmx->nested.vmxon && !nested_cr4_valid(vcpu, cr4))
2957 		return 1;
2958 
2959 	vcpu->arch.cr4 = cr4;
2960 
2961 	if (!enable_unrestricted_guest) {
2962 		if (enable_ept) {
2963 			if (!is_paging(vcpu)) {
2964 				hw_cr4 &= ~X86_CR4_PAE;
2965 				hw_cr4 |= X86_CR4_PSE;
2966 			} else if (!(cr4 & X86_CR4_PAE)) {
2967 				hw_cr4 &= ~X86_CR4_PAE;
2968 			}
2969 		}
2970 
2971 		/*
2972 		 * SMEP/SMAP/PKU is disabled if CPU is in non-paging mode in
2973 		 * hardware.  To emulate this behavior, SMEP/SMAP/PKU needs
2974 		 * to be manually disabled when guest switches to non-paging
2975 		 * mode.
2976 		 *
2977 		 * If !enable_unrestricted_guest, the CPU is always running
2978 		 * with CR0.PG=1 and CR4 needs to be modified.
2979 		 * If enable_unrestricted_guest, the CPU automatically
2980 		 * disables SMEP/SMAP/PKU when the guest sets CR0.PG=0.
2981 		 */
2982 		if (!is_paging(vcpu))
2983 			hw_cr4 &= ~(X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE);
2984 	}
2985 
2986 	vmcs_writel(CR4_READ_SHADOW, cr4);
2987 	vmcs_writel(GUEST_CR4, hw_cr4);
2988 	return 0;
2989 }
2990 
2991 void vmx_get_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg)
2992 {
2993 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2994 	u32 ar;
2995 
2996 	if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
2997 		*var = vmx->rmode.segs[seg];
2998 		if (seg == VCPU_SREG_TR
2999 		    || var->selector == vmx_read_guest_seg_selector(vmx, seg))
3000 			return;
3001 		var->base = vmx_read_guest_seg_base(vmx, seg);
3002 		var->selector = vmx_read_guest_seg_selector(vmx, seg);
3003 		return;
3004 	}
3005 	var->base = vmx_read_guest_seg_base(vmx, seg);
3006 	var->limit = vmx_read_guest_seg_limit(vmx, seg);
3007 	var->selector = vmx_read_guest_seg_selector(vmx, seg);
3008 	ar = vmx_read_guest_seg_ar(vmx, seg);
3009 	var->unusable = (ar >> 16) & 1;
3010 	var->type = ar & 15;
3011 	var->s = (ar >> 4) & 1;
3012 	var->dpl = (ar >> 5) & 3;
3013 	/*
3014 	 * Some userspaces do not preserve unusable property. Since usable
3015 	 * segment has to be present according to VMX spec we can use present
3016 	 * property to amend userspace bug by making unusable segment always
3017 	 * nonpresent. vmx_segment_access_rights() already marks nonpresent
3018 	 * segment as unusable.
3019 	 */
3020 	var->present = !var->unusable;
3021 	var->avl = (ar >> 12) & 1;
3022 	var->l = (ar >> 13) & 1;
3023 	var->db = (ar >> 14) & 1;
3024 	var->g = (ar >> 15) & 1;
3025 }
3026 
3027 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
3028 {
3029 	struct kvm_segment s;
3030 
3031 	if (to_vmx(vcpu)->rmode.vm86_active) {
3032 		vmx_get_segment(vcpu, &s, seg);
3033 		return s.base;
3034 	}
3035 	return vmx_read_guest_seg_base(to_vmx(vcpu), seg);
3036 }
3037 
3038 int vmx_get_cpl(struct kvm_vcpu *vcpu)
3039 {
3040 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3041 
3042 	if (unlikely(vmx->rmode.vm86_active))
3043 		return 0;
3044 	else {
3045 		int ar = vmx_read_guest_seg_ar(vmx, VCPU_SREG_SS);
3046 		return VMX_AR_DPL(ar);
3047 	}
3048 }
3049 
3050 static u32 vmx_segment_access_rights(struct kvm_segment *var)
3051 {
3052 	u32 ar;
3053 
3054 	if (var->unusable || !var->present)
3055 		ar = 1 << 16;
3056 	else {
3057 		ar = var->type & 15;
3058 		ar |= (var->s & 1) << 4;
3059 		ar |= (var->dpl & 3) << 5;
3060 		ar |= (var->present & 1) << 7;
3061 		ar |= (var->avl & 1) << 12;
3062 		ar |= (var->l & 1) << 13;
3063 		ar |= (var->db & 1) << 14;
3064 		ar |= (var->g & 1) << 15;
3065 	}
3066 
3067 	return ar;
3068 }
3069 
3070 void vmx_set_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg)
3071 {
3072 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3073 	const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3074 
3075 	vmx_segment_cache_clear(vmx);
3076 
3077 	if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3078 		vmx->rmode.segs[seg] = *var;
3079 		if (seg == VCPU_SREG_TR)
3080 			vmcs_write16(sf->selector, var->selector);
3081 		else if (var->s)
3082 			fix_rmode_seg(seg, &vmx->rmode.segs[seg]);
3083 		goto out;
3084 	}
3085 
3086 	vmcs_writel(sf->base, var->base);
3087 	vmcs_write32(sf->limit, var->limit);
3088 	vmcs_write16(sf->selector, var->selector);
3089 
3090 	/*
3091 	 *   Fix the "Accessed" bit in AR field of segment registers for older
3092 	 * qemu binaries.
3093 	 *   IA32 arch specifies that at the time of processor reset the
3094 	 * "Accessed" bit in the AR field of segment registers is 1. And qemu
3095 	 * is setting it to 0 in the userland code. This causes invalid guest
3096 	 * state vmexit when "unrestricted guest" mode is turned on.
3097 	 *    Fix for this setup issue in cpu_reset is being pushed in the qemu
3098 	 * tree. Newer qemu binaries with that qemu fix would not need this
3099 	 * kvm hack.
3100 	 */
3101 	if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
3102 		var->type |= 0x1; /* Accessed */
3103 
3104 	vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(var));
3105 
3106 out:
3107 	vmx->emulation_required = emulation_required(vcpu);
3108 }
3109 
3110 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3111 {
3112 	u32 ar = vmx_read_guest_seg_ar(to_vmx(vcpu), VCPU_SREG_CS);
3113 
3114 	*db = (ar >> 14) & 1;
3115 	*l = (ar >> 13) & 1;
3116 }
3117 
3118 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3119 {
3120 	dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
3121 	dt->address = vmcs_readl(GUEST_IDTR_BASE);
3122 }
3123 
3124 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3125 {
3126 	vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
3127 	vmcs_writel(GUEST_IDTR_BASE, dt->address);
3128 }
3129 
3130 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3131 {
3132 	dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
3133 	dt->address = vmcs_readl(GUEST_GDTR_BASE);
3134 }
3135 
3136 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3137 {
3138 	vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
3139 	vmcs_writel(GUEST_GDTR_BASE, dt->address);
3140 }
3141 
3142 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
3143 {
3144 	struct kvm_segment var;
3145 	u32 ar;
3146 
3147 	vmx_get_segment(vcpu, &var, seg);
3148 	var.dpl = 0x3;
3149 	if (seg == VCPU_SREG_CS)
3150 		var.type = 0x3;
3151 	ar = vmx_segment_access_rights(&var);
3152 
3153 	if (var.base != (var.selector << 4))
3154 		return false;
3155 	if (var.limit != 0xffff)
3156 		return false;
3157 	if (ar != 0xf3)
3158 		return false;
3159 
3160 	return true;
3161 }
3162 
3163 static bool code_segment_valid(struct kvm_vcpu *vcpu)
3164 {
3165 	struct kvm_segment cs;
3166 	unsigned int cs_rpl;
3167 
3168 	vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3169 	cs_rpl = cs.selector & SEGMENT_RPL_MASK;
3170 
3171 	if (cs.unusable)
3172 		return false;
3173 	if (~cs.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_ACCESSES_MASK))
3174 		return false;
3175 	if (!cs.s)
3176 		return false;
3177 	if (cs.type & VMX_AR_TYPE_WRITEABLE_MASK) {
3178 		if (cs.dpl > cs_rpl)
3179 			return false;
3180 	} else {
3181 		if (cs.dpl != cs_rpl)
3182 			return false;
3183 	}
3184 	if (!cs.present)
3185 		return false;
3186 
3187 	/* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
3188 	return true;
3189 }
3190 
3191 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
3192 {
3193 	struct kvm_segment ss;
3194 	unsigned int ss_rpl;
3195 
3196 	vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3197 	ss_rpl = ss.selector & SEGMENT_RPL_MASK;
3198 
3199 	if (ss.unusable)
3200 		return true;
3201 	if (ss.type != 3 && ss.type != 7)
3202 		return false;
3203 	if (!ss.s)
3204 		return false;
3205 	if (ss.dpl != ss_rpl) /* DPL != RPL */
3206 		return false;
3207 	if (!ss.present)
3208 		return false;
3209 
3210 	return true;
3211 }
3212 
3213 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
3214 {
3215 	struct kvm_segment var;
3216 	unsigned int rpl;
3217 
3218 	vmx_get_segment(vcpu, &var, seg);
3219 	rpl = var.selector & SEGMENT_RPL_MASK;
3220 
3221 	if (var.unusable)
3222 		return true;
3223 	if (!var.s)
3224 		return false;
3225 	if (!var.present)
3226 		return false;
3227 	if (~var.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_WRITEABLE_MASK)) {
3228 		if (var.dpl < rpl) /* DPL < RPL */
3229 			return false;
3230 	}
3231 
3232 	/* TODO: Add other members to kvm_segment_field to allow checking for other access
3233 	 * rights flags
3234 	 */
3235 	return true;
3236 }
3237 
3238 static bool tr_valid(struct kvm_vcpu *vcpu)
3239 {
3240 	struct kvm_segment tr;
3241 
3242 	vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
3243 
3244 	if (tr.unusable)
3245 		return false;
3246 	if (tr.selector & SEGMENT_TI_MASK)	/* TI = 1 */
3247 		return false;
3248 	if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
3249 		return false;
3250 	if (!tr.present)
3251 		return false;
3252 
3253 	return true;
3254 }
3255 
3256 static bool ldtr_valid(struct kvm_vcpu *vcpu)
3257 {
3258 	struct kvm_segment ldtr;
3259 
3260 	vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
3261 
3262 	if (ldtr.unusable)
3263 		return true;
3264 	if (ldtr.selector & SEGMENT_TI_MASK)	/* TI = 1 */
3265 		return false;
3266 	if (ldtr.type != 2)
3267 		return false;
3268 	if (!ldtr.present)
3269 		return false;
3270 
3271 	return true;
3272 }
3273 
3274 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
3275 {
3276 	struct kvm_segment cs, ss;
3277 
3278 	vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3279 	vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3280 
3281 	return ((cs.selector & SEGMENT_RPL_MASK) ==
3282 		 (ss.selector & SEGMENT_RPL_MASK));
3283 }
3284 
3285 /*
3286  * Check if guest state is valid. Returns true if valid, false if
3287  * not.
3288  * We assume that registers are always usable
3289  */
3290 static bool guest_state_valid(struct kvm_vcpu *vcpu)
3291 {
3292 	if (enable_unrestricted_guest)
3293 		return true;
3294 
3295 	/* real mode guest state checks */
3296 	if (!is_protmode(vcpu) || (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
3297 		if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
3298 			return false;
3299 		if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
3300 			return false;
3301 		if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
3302 			return false;
3303 		if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
3304 			return false;
3305 		if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
3306 			return false;
3307 		if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
3308 			return false;
3309 	} else {
3310 	/* protected mode guest state checks */
3311 		if (!cs_ss_rpl_check(vcpu))
3312 			return false;
3313 		if (!code_segment_valid(vcpu))
3314 			return false;
3315 		if (!stack_segment_valid(vcpu))
3316 			return false;
3317 		if (!data_segment_valid(vcpu, VCPU_SREG_DS))
3318 			return false;
3319 		if (!data_segment_valid(vcpu, VCPU_SREG_ES))
3320 			return false;
3321 		if (!data_segment_valid(vcpu, VCPU_SREG_FS))
3322 			return false;
3323 		if (!data_segment_valid(vcpu, VCPU_SREG_GS))
3324 			return false;
3325 		if (!tr_valid(vcpu))
3326 			return false;
3327 		if (!ldtr_valid(vcpu))
3328 			return false;
3329 	}
3330 	/* TODO:
3331 	 * - Add checks on RIP
3332 	 * - Add checks on RFLAGS
3333 	 */
3334 
3335 	return true;
3336 }
3337 
3338 static int init_rmode_tss(struct kvm *kvm)
3339 {
3340 	gfn_t fn;
3341 	u16 data = 0;
3342 	int idx, r;
3343 
3344 	idx = srcu_read_lock(&kvm->srcu);
3345 	fn = to_kvm_vmx(kvm)->tss_addr >> PAGE_SHIFT;
3346 	r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3347 	if (r < 0)
3348 		goto out;
3349 	data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
3350 	r = kvm_write_guest_page(kvm, fn++, &data,
3351 			TSS_IOPB_BASE_OFFSET, sizeof(u16));
3352 	if (r < 0)
3353 		goto out;
3354 	r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
3355 	if (r < 0)
3356 		goto out;
3357 	r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3358 	if (r < 0)
3359 		goto out;
3360 	data = ~0;
3361 	r = kvm_write_guest_page(kvm, fn, &data,
3362 				 RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
3363 				 sizeof(u8));
3364 out:
3365 	srcu_read_unlock(&kvm->srcu, idx);
3366 	return r;
3367 }
3368 
3369 static int init_rmode_identity_map(struct kvm *kvm)
3370 {
3371 	struct kvm_vmx *kvm_vmx = to_kvm_vmx(kvm);
3372 	int i, idx, r = 0;
3373 	kvm_pfn_t identity_map_pfn;
3374 	u32 tmp;
3375 
3376 	/* Protect kvm_vmx->ept_identity_pagetable_done. */
3377 	mutex_lock(&kvm->slots_lock);
3378 
3379 	if (likely(kvm_vmx->ept_identity_pagetable_done))
3380 		goto out2;
3381 
3382 	if (!kvm_vmx->ept_identity_map_addr)
3383 		kvm_vmx->ept_identity_map_addr = VMX_EPT_IDENTITY_PAGETABLE_ADDR;
3384 	identity_map_pfn = kvm_vmx->ept_identity_map_addr >> PAGE_SHIFT;
3385 
3386 	r = __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
3387 				    kvm_vmx->ept_identity_map_addr, PAGE_SIZE);
3388 	if (r < 0)
3389 		goto out2;
3390 
3391 	idx = srcu_read_lock(&kvm->srcu);
3392 	r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
3393 	if (r < 0)
3394 		goto out;
3395 	/* Set up identity-mapping pagetable for EPT in real mode */
3396 	for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
3397 		tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
3398 			_PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
3399 		r = kvm_write_guest_page(kvm, identity_map_pfn,
3400 				&tmp, i * sizeof(tmp), sizeof(tmp));
3401 		if (r < 0)
3402 			goto out;
3403 	}
3404 	kvm_vmx->ept_identity_pagetable_done = true;
3405 
3406 out:
3407 	srcu_read_unlock(&kvm->srcu, idx);
3408 
3409 out2:
3410 	mutex_unlock(&kvm->slots_lock);
3411 	return r;
3412 }
3413 
3414 static void seg_setup(int seg)
3415 {
3416 	const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3417 	unsigned int ar;
3418 
3419 	vmcs_write16(sf->selector, 0);
3420 	vmcs_writel(sf->base, 0);
3421 	vmcs_write32(sf->limit, 0xffff);
3422 	ar = 0x93;
3423 	if (seg == VCPU_SREG_CS)
3424 		ar |= 0x08; /* code segment */
3425 
3426 	vmcs_write32(sf->ar_bytes, ar);
3427 }
3428 
3429 static int alloc_apic_access_page(struct kvm *kvm)
3430 {
3431 	struct page *page;
3432 	int r = 0;
3433 
3434 	mutex_lock(&kvm->slots_lock);
3435 	if (kvm->arch.apic_access_page_done)
3436 		goto out;
3437 	r = __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
3438 				    APIC_DEFAULT_PHYS_BASE, PAGE_SIZE);
3439 	if (r)
3440 		goto out;
3441 
3442 	page = gfn_to_page(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
3443 	if (is_error_page(page)) {
3444 		r = -EFAULT;
3445 		goto out;
3446 	}
3447 
3448 	/*
3449 	 * Do not pin the page in memory, so that memory hot-unplug
3450 	 * is able to migrate it.
3451 	 */
3452 	put_page(page);
3453 	kvm->arch.apic_access_page_done = true;
3454 out:
3455 	mutex_unlock(&kvm->slots_lock);
3456 	return r;
3457 }
3458 
3459 int allocate_vpid(void)
3460 {
3461 	int vpid;
3462 
3463 	if (!enable_vpid)
3464 		return 0;
3465 	spin_lock(&vmx_vpid_lock);
3466 	vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
3467 	if (vpid < VMX_NR_VPIDS)
3468 		__set_bit(vpid, vmx_vpid_bitmap);
3469 	else
3470 		vpid = 0;
3471 	spin_unlock(&vmx_vpid_lock);
3472 	return vpid;
3473 }
3474 
3475 void free_vpid(int vpid)
3476 {
3477 	if (!enable_vpid || vpid == 0)
3478 		return;
3479 	spin_lock(&vmx_vpid_lock);
3480 	__clear_bit(vpid, vmx_vpid_bitmap);
3481 	spin_unlock(&vmx_vpid_lock);
3482 }
3483 
3484 static __always_inline void vmx_disable_intercept_for_msr(unsigned long *msr_bitmap,
3485 							  u32 msr, int type)
3486 {
3487 	int f = sizeof(unsigned long);
3488 
3489 	if (!cpu_has_vmx_msr_bitmap())
3490 		return;
3491 
3492 	if (static_branch_unlikely(&enable_evmcs))
3493 		evmcs_touch_msr_bitmap();
3494 
3495 	/*
3496 	 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
3497 	 * have the write-low and read-high bitmap offsets the wrong way round.
3498 	 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
3499 	 */
3500 	if (msr <= 0x1fff) {
3501 		if (type & MSR_TYPE_R)
3502 			/* read-low */
3503 			__clear_bit(msr, msr_bitmap + 0x000 / f);
3504 
3505 		if (type & MSR_TYPE_W)
3506 			/* write-low */
3507 			__clear_bit(msr, msr_bitmap + 0x800 / f);
3508 
3509 	} else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
3510 		msr &= 0x1fff;
3511 		if (type & MSR_TYPE_R)
3512 			/* read-high */
3513 			__clear_bit(msr, msr_bitmap + 0x400 / f);
3514 
3515 		if (type & MSR_TYPE_W)
3516 			/* write-high */
3517 			__clear_bit(msr, msr_bitmap + 0xc00 / f);
3518 
3519 	}
3520 }
3521 
3522 static __always_inline void vmx_enable_intercept_for_msr(unsigned long *msr_bitmap,
3523 							 u32 msr, int type)
3524 {
3525 	int f = sizeof(unsigned long);
3526 
3527 	if (!cpu_has_vmx_msr_bitmap())
3528 		return;
3529 
3530 	if (static_branch_unlikely(&enable_evmcs))
3531 		evmcs_touch_msr_bitmap();
3532 
3533 	/*
3534 	 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
3535 	 * have the write-low and read-high bitmap offsets the wrong way round.
3536 	 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
3537 	 */
3538 	if (msr <= 0x1fff) {
3539 		if (type & MSR_TYPE_R)
3540 			/* read-low */
3541 			__set_bit(msr, msr_bitmap + 0x000 / f);
3542 
3543 		if (type & MSR_TYPE_W)
3544 			/* write-low */
3545 			__set_bit(msr, msr_bitmap + 0x800 / f);
3546 
3547 	} else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
3548 		msr &= 0x1fff;
3549 		if (type & MSR_TYPE_R)
3550 			/* read-high */
3551 			__set_bit(msr, msr_bitmap + 0x400 / f);
3552 
3553 		if (type & MSR_TYPE_W)
3554 			/* write-high */
3555 			__set_bit(msr, msr_bitmap + 0xc00 / f);
3556 
3557 	}
3558 }
3559 
3560 static __always_inline void vmx_set_intercept_for_msr(unsigned long *msr_bitmap,
3561 			     			      u32 msr, int type, bool value)
3562 {
3563 	if (value)
3564 		vmx_enable_intercept_for_msr(msr_bitmap, msr, type);
3565 	else
3566 		vmx_disable_intercept_for_msr(msr_bitmap, msr, type);
3567 }
3568 
3569 static u8 vmx_msr_bitmap_mode(struct kvm_vcpu *vcpu)
3570 {
3571 	u8 mode = 0;
3572 
3573 	if (cpu_has_secondary_exec_ctrls() &&
3574 	    (secondary_exec_controls_get(to_vmx(vcpu)) &
3575 	     SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE)) {
3576 		mode |= MSR_BITMAP_MODE_X2APIC;
3577 		if (enable_apicv && kvm_vcpu_apicv_active(vcpu))
3578 			mode |= MSR_BITMAP_MODE_X2APIC_APICV;
3579 	}
3580 
3581 	return mode;
3582 }
3583 
3584 static void vmx_update_msr_bitmap_x2apic(unsigned long *msr_bitmap,
3585 					 u8 mode)
3586 {
3587 	int msr;
3588 
3589 	for (msr = 0x800; msr <= 0x8ff; msr += BITS_PER_LONG) {
3590 		unsigned word = msr / BITS_PER_LONG;
3591 		msr_bitmap[word] = (mode & MSR_BITMAP_MODE_X2APIC_APICV) ? 0 : ~0;
3592 		msr_bitmap[word + (0x800 / sizeof(long))] = ~0;
3593 	}
3594 
3595 	if (mode & MSR_BITMAP_MODE_X2APIC) {
3596 		/*
3597 		 * TPR reads and writes can be virtualized even if virtual interrupt
3598 		 * delivery is not in use.
3599 		 */
3600 		vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_TASKPRI), MSR_TYPE_RW);
3601 		if (mode & MSR_BITMAP_MODE_X2APIC_APICV) {
3602 			vmx_enable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_TMCCT), MSR_TYPE_R);
3603 			vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_EOI), MSR_TYPE_W);
3604 			vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_SELF_IPI), MSR_TYPE_W);
3605 		}
3606 	}
3607 }
3608 
3609 void vmx_update_msr_bitmap(struct kvm_vcpu *vcpu)
3610 {
3611 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3612 	unsigned long *msr_bitmap = vmx->vmcs01.msr_bitmap;
3613 	u8 mode = vmx_msr_bitmap_mode(vcpu);
3614 	u8 changed = mode ^ vmx->msr_bitmap_mode;
3615 
3616 	if (!changed)
3617 		return;
3618 
3619 	if (changed & (MSR_BITMAP_MODE_X2APIC | MSR_BITMAP_MODE_X2APIC_APICV))
3620 		vmx_update_msr_bitmap_x2apic(msr_bitmap, mode);
3621 
3622 	vmx->msr_bitmap_mode = mode;
3623 }
3624 
3625 void pt_update_intercept_for_msr(struct vcpu_vmx *vmx)
3626 {
3627 	unsigned long *msr_bitmap = vmx->vmcs01.msr_bitmap;
3628 	bool flag = !(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN);
3629 	u32 i;
3630 
3631 	vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_STATUS,
3632 							MSR_TYPE_RW, flag);
3633 	vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_OUTPUT_BASE,
3634 							MSR_TYPE_RW, flag);
3635 	vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_OUTPUT_MASK,
3636 							MSR_TYPE_RW, flag);
3637 	vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_CR3_MATCH,
3638 							MSR_TYPE_RW, flag);
3639 	for (i = 0; i < vmx->pt_desc.addr_range; i++) {
3640 		vmx_set_intercept_for_msr(msr_bitmap,
3641 			MSR_IA32_RTIT_ADDR0_A + i * 2, MSR_TYPE_RW, flag);
3642 		vmx_set_intercept_for_msr(msr_bitmap,
3643 			MSR_IA32_RTIT_ADDR0_B + i * 2, MSR_TYPE_RW, flag);
3644 	}
3645 }
3646 
3647 static bool vmx_get_enable_apicv(struct kvm_vcpu *vcpu)
3648 {
3649 	return enable_apicv;
3650 }
3651 
3652 static bool vmx_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
3653 {
3654 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3655 	void *vapic_page;
3656 	u32 vppr;
3657 	int rvi;
3658 
3659 	if (WARN_ON_ONCE(!is_guest_mode(vcpu)) ||
3660 		!nested_cpu_has_vid(get_vmcs12(vcpu)) ||
3661 		WARN_ON_ONCE(!vmx->nested.virtual_apic_map.gfn))
3662 		return false;
3663 
3664 	rvi = vmx_get_rvi();
3665 
3666 	vapic_page = vmx->nested.virtual_apic_map.hva;
3667 	vppr = *((u32 *)(vapic_page + APIC_PROCPRI));
3668 
3669 	return ((rvi & 0xf0) > (vppr & 0xf0));
3670 }
3671 
3672 static inline bool kvm_vcpu_trigger_posted_interrupt(struct kvm_vcpu *vcpu,
3673 						     bool nested)
3674 {
3675 #ifdef CONFIG_SMP
3676 	int pi_vec = nested ? POSTED_INTR_NESTED_VECTOR : POSTED_INTR_VECTOR;
3677 
3678 	if (vcpu->mode == IN_GUEST_MODE) {
3679 		/*
3680 		 * The vector of interrupt to be delivered to vcpu had
3681 		 * been set in PIR before this function.
3682 		 *
3683 		 * Following cases will be reached in this block, and
3684 		 * we always send a notification event in all cases as
3685 		 * explained below.
3686 		 *
3687 		 * Case 1: vcpu keeps in non-root mode. Sending a
3688 		 * notification event posts the interrupt to vcpu.
3689 		 *
3690 		 * Case 2: vcpu exits to root mode and is still
3691 		 * runnable. PIR will be synced to vIRR before the
3692 		 * next vcpu entry. Sending a notification event in
3693 		 * this case has no effect, as vcpu is not in root
3694 		 * mode.
3695 		 *
3696 		 * Case 3: vcpu exits to root mode and is blocked.
3697 		 * vcpu_block() has already synced PIR to vIRR and
3698 		 * never blocks vcpu if vIRR is not cleared. Therefore,
3699 		 * a blocked vcpu here does not wait for any requested
3700 		 * interrupts in PIR, and sending a notification event
3701 		 * which has no effect is safe here.
3702 		 */
3703 
3704 		apic->send_IPI_mask(get_cpu_mask(vcpu->cpu), pi_vec);
3705 		return true;
3706 	}
3707 #endif
3708 	return false;
3709 }
3710 
3711 static int vmx_deliver_nested_posted_interrupt(struct kvm_vcpu *vcpu,
3712 						int vector)
3713 {
3714 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3715 
3716 	if (is_guest_mode(vcpu) &&
3717 	    vector == vmx->nested.posted_intr_nv) {
3718 		/*
3719 		 * If a posted intr is not recognized by hardware,
3720 		 * we will accomplish it in the next vmentry.
3721 		 */
3722 		vmx->nested.pi_pending = true;
3723 		kvm_make_request(KVM_REQ_EVENT, vcpu);
3724 		/* the PIR and ON have been set by L1. */
3725 		if (!kvm_vcpu_trigger_posted_interrupt(vcpu, true))
3726 			kvm_vcpu_kick(vcpu);
3727 		return 0;
3728 	}
3729 	return -1;
3730 }
3731 /*
3732  * Send interrupt to vcpu via posted interrupt way.
3733  * 1. If target vcpu is running(non-root mode), send posted interrupt
3734  * notification to vcpu and hardware will sync PIR to vIRR atomically.
3735  * 2. If target vcpu isn't running(root mode), kick it to pick up the
3736  * interrupt from PIR in next vmentry.
3737  */
3738 static void vmx_deliver_posted_interrupt(struct kvm_vcpu *vcpu, int vector)
3739 {
3740 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3741 	int r;
3742 
3743 	r = vmx_deliver_nested_posted_interrupt(vcpu, vector);
3744 	if (!r)
3745 		return;
3746 
3747 	if (pi_test_and_set_pir(vector, &vmx->pi_desc))
3748 		return;
3749 
3750 	/* If a previous notification has sent the IPI, nothing to do.  */
3751 	if (pi_test_and_set_on(&vmx->pi_desc))
3752 		return;
3753 
3754 	if (!kvm_vcpu_trigger_posted_interrupt(vcpu, false))
3755 		kvm_vcpu_kick(vcpu);
3756 }
3757 
3758 /*
3759  * Set up the vmcs's constant host-state fields, i.e., host-state fields that
3760  * will not change in the lifetime of the guest.
3761  * Note that host-state that does change is set elsewhere. E.g., host-state
3762  * that is set differently for each CPU is set in vmx_vcpu_load(), not here.
3763  */
3764 void vmx_set_constant_host_state(struct vcpu_vmx *vmx)
3765 {
3766 	u32 low32, high32;
3767 	unsigned long tmpl;
3768 	unsigned long cr0, cr3, cr4;
3769 
3770 	cr0 = read_cr0();
3771 	WARN_ON(cr0 & X86_CR0_TS);
3772 	vmcs_writel(HOST_CR0, cr0);  /* 22.2.3 */
3773 
3774 	/*
3775 	 * Save the most likely value for this task's CR3 in the VMCS.
3776 	 * We can't use __get_current_cr3_fast() because we're not atomic.
3777 	 */
3778 	cr3 = __read_cr3();
3779 	vmcs_writel(HOST_CR3, cr3);		/* 22.2.3  FIXME: shadow tables */
3780 	vmx->loaded_vmcs->host_state.cr3 = cr3;
3781 
3782 	/* Save the most likely value for this task's CR4 in the VMCS. */
3783 	cr4 = cr4_read_shadow();
3784 	vmcs_writel(HOST_CR4, cr4);			/* 22.2.3, 22.2.5 */
3785 	vmx->loaded_vmcs->host_state.cr4 = cr4;
3786 
3787 	vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS);  /* 22.2.4 */
3788 #ifdef CONFIG_X86_64
3789 	/*
3790 	 * Load null selectors, so we can avoid reloading them in
3791 	 * vmx_prepare_switch_to_host(), in case userspace uses
3792 	 * the null selectors too (the expected case).
3793 	 */
3794 	vmcs_write16(HOST_DS_SELECTOR, 0);
3795 	vmcs_write16(HOST_ES_SELECTOR, 0);
3796 #else
3797 	vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
3798 	vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
3799 #endif
3800 	vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
3801 	vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8);  /* 22.2.4 */
3802 
3803 	vmcs_writel(HOST_IDTR_BASE, host_idt_base);   /* 22.2.4 */
3804 
3805 	vmcs_writel(HOST_RIP, (unsigned long)vmx_vmexit); /* 22.2.5 */
3806 
3807 	rdmsr(MSR_IA32_SYSENTER_CS, low32, high32);
3808 	vmcs_write32(HOST_IA32_SYSENTER_CS, low32);
3809 	rdmsrl(MSR_IA32_SYSENTER_EIP, tmpl);
3810 	vmcs_writel(HOST_IA32_SYSENTER_EIP, tmpl);   /* 22.2.3 */
3811 
3812 	if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
3813 		rdmsr(MSR_IA32_CR_PAT, low32, high32);
3814 		vmcs_write64(HOST_IA32_PAT, low32 | ((u64) high32 << 32));
3815 	}
3816 
3817 	if (cpu_has_load_ia32_efer())
3818 		vmcs_write64(HOST_IA32_EFER, host_efer);
3819 }
3820 
3821 void set_cr4_guest_host_mask(struct vcpu_vmx *vmx)
3822 {
3823 	vmx->vcpu.arch.cr4_guest_owned_bits = KVM_CR4_GUEST_OWNED_BITS;
3824 	if (enable_ept)
3825 		vmx->vcpu.arch.cr4_guest_owned_bits |= X86_CR4_PGE;
3826 	if (is_guest_mode(&vmx->vcpu))
3827 		vmx->vcpu.arch.cr4_guest_owned_bits &=
3828 			~get_vmcs12(&vmx->vcpu)->cr4_guest_host_mask;
3829 	vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits);
3830 }
3831 
3832 u32 vmx_pin_based_exec_ctrl(struct vcpu_vmx *vmx)
3833 {
3834 	u32 pin_based_exec_ctrl = vmcs_config.pin_based_exec_ctrl;
3835 
3836 	if (!kvm_vcpu_apicv_active(&vmx->vcpu))
3837 		pin_based_exec_ctrl &= ~PIN_BASED_POSTED_INTR;
3838 
3839 	if (!enable_vnmi)
3840 		pin_based_exec_ctrl &= ~PIN_BASED_VIRTUAL_NMIS;
3841 
3842 	if (!enable_preemption_timer)
3843 		pin_based_exec_ctrl &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
3844 
3845 	return pin_based_exec_ctrl;
3846 }
3847 
3848 static void vmx_refresh_apicv_exec_ctrl(struct kvm_vcpu *vcpu)
3849 {
3850 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3851 
3852 	pin_controls_set(vmx, vmx_pin_based_exec_ctrl(vmx));
3853 	if (cpu_has_secondary_exec_ctrls()) {
3854 		if (kvm_vcpu_apicv_active(vcpu))
3855 			secondary_exec_controls_setbit(vmx,
3856 				      SECONDARY_EXEC_APIC_REGISTER_VIRT |
3857 				      SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
3858 		else
3859 			secondary_exec_controls_clearbit(vmx,
3860 					SECONDARY_EXEC_APIC_REGISTER_VIRT |
3861 					SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
3862 	}
3863 
3864 	if (cpu_has_vmx_msr_bitmap())
3865 		vmx_update_msr_bitmap(vcpu);
3866 }
3867 
3868 u32 vmx_exec_control(struct vcpu_vmx *vmx)
3869 {
3870 	u32 exec_control = vmcs_config.cpu_based_exec_ctrl;
3871 
3872 	if (vmx->vcpu.arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)
3873 		exec_control &= ~CPU_BASED_MOV_DR_EXITING;
3874 
3875 	if (!cpu_need_tpr_shadow(&vmx->vcpu)) {
3876 		exec_control &= ~CPU_BASED_TPR_SHADOW;
3877 #ifdef CONFIG_X86_64
3878 		exec_control |= CPU_BASED_CR8_STORE_EXITING |
3879 				CPU_BASED_CR8_LOAD_EXITING;
3880 #endif
3881 	}
3882 	if (!enable_ept)
3883 		exec_control |= CPU_BASED_CR3_STORE_EXITING |
3884 				CPU_BASED_CR3_LOAD_EXITING  |
3885 				CPU_BASED_INVLPG_EXITING;
3886 	if (kvm_mwait_in_guest(vmx->vcpu.kvm))
3887 		exec_control &= ~(CPU_BASED_MWAIT_EXITING |
3888 				CPU_BASED_MONITOR_EXITING);
3889 	if (kvm_hlt_in_guest(vmx->vcpu.kvm))
3890 		exec_control &= ~CPU_BASED_HLT_EXITING;
3891 	return exec_control;
3892 }
3893 
3894 
3895 static void vmx_compute_secondary_exec_control(struct vcpu_vmx *vmx)
3896 {
3897 	struct kvm_vcpu *vcpu = &vmx->vcpu;
3898 
3899 	u32 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
3900 
3901 	if (pt_mode == PT_MODE_SYSTEM)
3902 		exec_control &= ~(SECONDARY_EXEC_PT_USE_GPA | SECONDARY_EXEC_PT_CONCEAL_VMX);
3903 	if (!cpu_need_virtualize_apic_accesses(vcpu))
3904 		exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
3905 	if (vmx->vpid == 0)
3906 		exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
3907 	if (!enable_ept) {
3908 		exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
3909 		enable_unrestricted_guest = 0;
3910 	}
3911 	if (!enable_unrestricted_guest)
3912 		exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
3913 	if (kvm_pause_in_guest(vmx->vcpu.kvm))
3914 		exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
3915 	if (!kvm_vcpu_apicv_active(vcpu))
3916 		exec_control &= ~(SECONDARY_EXEC_APIC_REGISTER_VIRT |
3917 				  SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
3918 	exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
3919 
3920 	/* SECONDARY_EXEC_DESC is enabled/disabled on writes to CR4.UMIP,
3921 	 * in vmx_set_cr4.  */
3922 	exec_control &= ~SECONDARY_EXEC_DESC;
3923 
3924 	/* SECONDARY_EXEC_SHADOW_VMCS is enabled when L1 executes VMPTRLD
3925 	   (handle_vmptrld).
3926 	   We can NOT enable shadow_vmcs here because we don't have yet
3927 	   a current VMCS12
3928 	*/
3929 	exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
3930 
3931 	if (!enable_pml)
3932 		exec_control &= ~SECONDARY_EXEC_ENABLE_PML;
3933 
3934 	if (vmx_xsaves_supported()) {
3935 		/* Exposing XSAVES only when XSAVE is exposed */
3936 		bool xsaves_enabled =
3937 			guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
3938 			guest_cpuid_has(vcpu, X86_FEATURE_XSAVES);
3939 
3940 		if (!xsaves_enabled)
3941 			exec_control &= ~SECONDARY_EXEC_XSAVES;
3942 
3943 		if (nested) {
3944 			if (xsaves_enabled)
3945 				vmx->nested.msrs.secondary_ctls_high |=
3946 					SECONDARY_EXEC_XSAVES;
3947 			else
3948 				vmx->nested.msrs.secondary_ctls_high &=
3949 					~SECONDARY_EXEC_XSAVES;
3950 		}
3951 	}
3952 
3953 	if (vmx_rdtscp_supported()) {
3954 		bool rdtscp_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP);
3955 		if (!rdtscp_enabled)
3956 			exec_control &= ~SECONDARY_EXEC_RDTSCP;
3957 
3958 		if (nested) {
3959 			if (rdtscp_enabled)
3960 				vmx->nested.msrs.secondary_ctls_high |=
3961 					SECONDARY_EXEC_RDTSCP;
3962 			else
3963 				vmx->nested.msrs.secondary_ctls_high &=
3964 					~SECONDARY_EXEC_RDTSCP;
3965 		}
3966 	}
3967 
3968 	if (vmx_invpcid_supported()) {
3969 		/* Exposing INVPCID only when PCID is exposed */
3970 		bool invpcid_enabled =
3971 			guest_cpuid_has(vcpu, X86_FEATURE_INVPCID) &&
3972 			guest_cpuid_has(vcpu, X86_FEATURE_PCID);
3973 
3974 		if (!invpcid_enabled) {
3975 			exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID;
3976 			guest_cpuid_clear(vcpu, X86_FEATURE_INVPCID);
3977 		}
3978 
3979 		if (nested) {
3980 			if (invpcid_enabled)
3981 				vmx->nested.msrs.secondary_ctls_high |=
3982 					SECONDARY_EXEC_ENABLE_INVPCID;
3983 			else
3984 				vmx->nested.msrs.secondary_ctls_high &=
3985 					~SECONDARY_EXEC_ENABLE_INVPCID;
3986 		}
3987 	}
3988 
3989 	if (vmx_rdrand_supported()) {
3990 		bool rdrand_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDRAND);
3991 		if (rdrand_enabled)
3992 			exec_control &= ~SECONDARY_EXEC_RDRAND_EXITING;
3993 
3994 		if (nested) {
3995 			if (rdrand_enabled)
3996 				vmx->nested.msrs.secondary_ctls_high |=
3997 					SECONDARY_EXEC_RDRAND_EXITING;
3998 			else
3999 				vmx->nested.msrs.secondary_ctls_high &=
4000 					~SECONDARY_EXEC_RDRAND_EXITING;
4001 		}
4002 	}
4003 
4004 	if (vmx_rdseed_supported()) {
4005 		bool rdseed_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDSEED);
4006 		if (rdseed_enabled)
4007 			exec_control &= ~SECONDARY_EXEC_RDSEED_EXITING;
4008 
4009 		if (nested) {
4010 			if (rdseed_enabled)
4011 				vmx->nested.msrs.secondary_ctls_high |=
4012 					SECONDARY_EXEC_RDSEED_EXITING;
4013 			else
4014 				vmx->nested.msrs.secondary_ctls_high &=
4015 					~SECONDARY_EXEC_RDSEED_EXITING;
4016 		}
4017 	}
4018 
4019 	vmx->secondary_exec_control = exec_control;
4020 }
4021 
4022 static void ept_set_mmio_spte_mask(void)
4023 {
4024 	/*
4025 	 * EPT Misconfigurations can be generated if the value of bits 2:0
4026 	 * of an EPT paging-structure entry is 110b (write/execute).
4027 	 */
4028 	kvm_mmu_set_mmio_spte_mask(VMX_EPT_RWX_MASK,
4029 				   VMX_EPT_MISCONFIG_WX_VALUE);
4030 }
4031 
4032 #define VMX_XSS_EXIT_BITMAP 0
4033 
4034 /*
4035  * Sets up the vmcs for emulated real mode.
4036  */
4037 static void vmx_vcpu_setup(struct vcpu_vmx *vmx)
4038 {
4039 	int i;
4040 
4041 	if (nested)
4042 		nested_vmx_vcpu_setup();
4043 
4044 	if (cpu_has_vmx_msr_bitmap())
4045 		vmcs_write64(MSR_BITMAP, __pa(vmx->vmcs01.msr_bitmap));
4046 
4047 	vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
4048 
4049 	/* Control */
4050 	pin_controls_set(vmx, vmx_pin_based_exec_ctrl(vmx));
4051 	vmx->hv_deadline_tsc = -1;
4052 
4053 	exec_controls_set(vmx, vmx_exec_control(vmx));
4054 
4055 	if (cpu_has_secondary_exec_ctrls()) {
4056 		vmx_compute_secondary_exec_control(vmx);
4057 		secondary_exec_controls_set(vmx, vmx->secondary_exec_control);
4058 	}
4059 
4060 	if (kvm_vcpu_apicv_active(&vmx->vcpu)) {
4061 		vmcs_write64(EOI_EXIT_BITMAP0, 0);
4062 		vmcs_write64(EOI_EXIT_BITMAP1, 0);
4063 		vmcs_write64(EOI_EXIT_BITMAP2, 0);
4064 		vmcs_write64(EOI_EXIT_BITMAP3, 0);
4065 
4066 		vmcs_write16(GUEST_INTR_STATUS, 0);
4067 
4068 		vmcs_write16(POSTED_INTR_NV, POSTED_INTR_VECTOR);
4069 		vmcs_write64(POSTED_INTR_DESC_ADDR, __pa((&vmx->pi_desc)));
4070 	}
4071 
4072 	if (!kvm_pause_in_guest(vmx->vcpu.kvm)) {
4073 		vmcs_write32(PLE_GAP, ple_gap);
4074 		vmx->ple_window = ple_window;
4075 		vmx->ple_window_dirty = true;
4076 	}
4077 
4078 	vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
4079 	vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
4080 	vmcs_write32(CR3_TARGET_COUNT, 0);           /* 22.2.1 */
4081 
4082 	vmcs_write16(HOST_FS_SELECTOR, 0);            /* 22.2.4 */
4083 	vmcs_write16(HOST_GS_SELECTOR, 0);            /* 22.2.4 */
4084 	vmx_set_constant_host_state(vmx);
4085 	vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
4086 	vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
4087 
4088 	if (cpu_has_vmx_vmfunc())
4089 		vmcs_write64(VM_FUNCTION_CONTROL, 0);
4090 
4091 	vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
4092 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
4093 	vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host.val));
4094 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
4095 	vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest.val));
4096 
4097 	if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
4098 		vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
4099 
4100 	for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i) {
4101 		u32 index = vmx_msr_index[i];
4102 		u32 data_low, data_high;
4103 		int j = vmx->nmsrs;
4104 
4105 		if (rdmsr_safe(index, &data_low, &data_high) < 0)
4106 			continue;
4107 		if (wrmsr_safe(index, data_low, data_high) < 0)
4108 			continue;
4109 		vmx->guest_msrs[j].index = i;
4110 		vmx->guest_msrs[j].data = 0;
4111 		vmx->guest_msrs[j].mask = -1ull;
4112 		++vmx->nmsrs;
4113 	}
4114 
4115 	vm_exit_controls_set(vmx, vmx_vmexit_ctrl());
4116 
4117 	/* 22.2.1, 20.8.1 */
4118 	vm_entry_controls_set(vmx, vmx_vmentry_ctrl());
4119 
4120 	vmx->vcpu.arch.cr0_guest_owned_bits = X86_CR0_TS;
4121 	vmcs_writel(CR0_GUEST_HOST_MASK, ~X86_CR0_TS);
4122 
4123 	set_cr4_guest_host_mask(vmx);
4124 
4125 	if (vmx_xsaves_supported())
4126 		vmcs_write64(XSS_EXIT_BITMAP, VMX_XSS_EXIT_BITMAP);
4127 
4128 	if (enable_pml) {
4129 		vmcs_write64(PML_ADDRESS, page_to_phys(vmx->pml_pg));
4130 		vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
4131 	}
4132 
4133 	if (cpu_has_vmx_encls_vmexit())
4134 		vmcs_write64(ENCLS_EXITING_BITMAP, -1ull);
4135 
4136 	if (pt_mode == PT_MODE_HOST_GUEST) {
4137 		memset(&vmx->pt_desc, 0, sizeof(vmx->pt_desc));
4138 		/* Bit[6~0] are forced to 1, writes are ignored. */
4139 		vmx->pt_desc.guest.output_mask = 0x7F;
4140 		vmcs_write64(GUEST_IA32_RTIT_CTL, 0);
4141 	}
4142 }
4143 
4144 static void vmx_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
4145 {
4146 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4147 	struct msr_data apic_base_msr;
4148 	u64 cr0;
4149 
4150 	vmx->rmode.vm86_active = 0;
4151 	vmx->spec_ctrl = 0;
4152 
4153 	vcpu->arch.microcode_version = 0x100000000ULL;
4154 	vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
4155 	kvm_set_cr8(vcpu, 0);
4156 
4157 	if (!init_event) {
4158 		apic_base_msr.data = APIC_DEFAULT_PHYS_BASE |
4159 				     MSR_IA32_APICBASE_ENABLE;
4160 		if (kvm_vcpu_is_reset_bsp(vcpu))
4161 			apic_base_msr.data |= MSR_IA32_APICBASE_BSP;
4162 		apic_base_msr.host_initiated = true;
4163 		kvm_set_apic_base(vcpu, &apic_base_msr);
4164 	}
4165 
4166 	vmx_segment_cache_clear(vmx);
4167 
4168 	seg_setup(VCPU_SREG_CS);
4169 	vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
4170 	vmcs_writel(GUEST_CS_BASE, 0xffff0000ul);
4171 
4172 	seg_setup(VCPU_SREG_DS);
4173 	seg_setup(VCPU_SREG_ES);
4174 	seg_setup(VCPU_SREG_FS);
4175 	seg_setup(VCPU_SREG_GS);
4176 	seg_setup(VCPU_SREG_SS);
4177 
4178 	vmcs_write16(GUEST_TR_SELECTOR, 0);
4179 	vmcs_writel(GUEST_TR_BASE, 0);
4180 	vmcs_write32(GUEST_TR_LIMIT, 0xffff);
4181 	vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
4182 
4183 	vmcs_write16(GUEST_LDTR_SELECTOR, 0);
4184 	vmcs_writel(GUEST_LDTR_BASE, 0);
4185 	vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
4186 	vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
4187 
4188 	if (!init_event) {
4189 		vmcs_write32(GUEST_SYSENTER_CS, 0);
4190 		vmcs_writel(GUEST_SYSENTER_ESP, 0);
4191 		vmcs_writel(GUEST_SYSENTER_EIP, 0);
4192 		vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
4193 	}
4194 
4195 	kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
4196 	kvm_rip_write(vcpu, 0xfff0);
4197 
4198 	vmcs_writel(GUEST_GDTR_BASE, 0);
4199 	vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
4200 
4201 	vmcs_writel(GUEST_IDTR_BASE, 0);
4202 	vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
4203 
4204 	vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
4205 	vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
4206 	vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS, 0);
4207 	if (kvm_mpx_supported())
4208 		vmcs_write64(GUEST_BNDCFGS, 0);
4209 
4210 	setup_msrs(vmx);
4211 
4212 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);  /* 22.2.1 */
4213 
4214 	if (cpu_has_vmx_tpr_shadow() && !init_event) {
4215 		vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
4216 		if (cpu_need_tpr_shadow(vcpu))
4217 			vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
4218 				     __pa(vcpu->arch.apic->regs));
4219 		vmcs_write32(TPR_THRESHOLD, 0);
4220 	}
4221 
4222 	kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
4223 
4224 	if (vmx->vpid != 0)
4225 		vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
4226 
4227 	cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
4228 	vmx->vcpu.arch.cr0 = cr0;
4229 	vmx_set_cr0(vcpu, cr0); /* enter rmode */
4230 	vmx_set_cr4(vcpu, 0);
4231 	vmx_set_efer(vcpu, 0);
4232 
4233 	update_exception_bitmap(vcpu);
4234 
4235 	vpid_sync_context(vmx->vpid);
4236 	if (init_event)
4237 		vmx_clear_hlt(vcpu);
4238 }
4239 
4240 static void enable_irq_window(struct kvm_vcpu *vcpu)
4241 {
4242 	exec_controls_setbit(to_vmx(vcpu), CPU_BASED_VIRTUAL_INTR_PENDING);
4243 }
4244 
4245 static void enable_nmi_window(struct kvm_vcpu *vcpu)
4246 {
4247 	if (!enable_vnmi ||
4248 	    vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_STI) {
4249 		enable_irq_window(vcpu);
4250 		return;
4251 	}
4252 
4253 	exec_controls_setbit(to_vmx(vcpu), CPU_BASED_VIRTUAL_NMI_PENDING);
4254 }
4255 
4256 static void vmx_inject_irq(struct kvm_vcpu *vcpu)
4257 {
4258 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4259 	uint32_t intr;
4260 	int irq = vcpu->arch.interrupt.nr;
4261 
4262 	trace_kvm_inj_virq(irq);
4263 
4264 	++vcpu->stat.irq_injections;
4265 	if (vmx->rmode.vm86_active) {
4266 		int inc_eip = 0;
4267 		if (vcpu->arch.interrupt.soft)
4268 			inc_eip = vcpu->arch.event_exit_inst_len;
4269 		if (kvm_inject_realmode_interrupt(vcpu, irq, inc_eip) != EMULATE_DONE)
4270 			kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4271 		return;
4272 	}
4273 	intr = irq | INTR_INFO_VALID_MASK;
4274 	if (vcpu->arch.interrupt.soft) {
4275 		intr |= INTR_TYPE_SOFT_INTR;
4276 		vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
4277 			     vmx->vcpu.arch.event_exit_inst_len);
4278 	} else
4279 		intr |= INTR_TYPE_EXT_INTR;
4280 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
4281 
4282 	vmx_clear_hlt(vcpu);
4283 }
4284 
4285 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
4286 {
4287 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4288 
4289 	if (!enable_vnmi) {
4290 		/*
4291 		 * Tracking the NMI-blocked state in software is built upon
4292 		 * finding the next open IRQ window. This, in turn, depends on
4293 		 * well-behaving guests: They have to keep IRQs disabled at
4294 		 * least as long as the NMI handler runs. Otherwise we may
4295 		 * cause NMI nesting, maybe breaking the guest. But as this is
4296 		 * highly unlikely, we can live with the residual risk.
4297 		 */
4298 		vmx->loaded_vmcs->soft_vnmi_blocked = 1;
4299 		vmx->loaded_vmcs->vnmi_blocked_time = 0;
4300 	}
4301 
4302 	++vcpu->stat.nmi_injections;
4303 	vmx->loaded_vmcs->nmi_known_unmasked = false;
4304 
4305 	if (vmx->rmode.vm86_active) {
4306 		if (kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR, 0) != EMULATE_DONE)
4307 			kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4308 		return;
4309 	}
4310 
4311 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
4312 			INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
4313 
4314 	vmx_clear_hlt(vcpu);
4315 }
4316 
4317 bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
4318 {
4319 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4320 	bool masked;
4321 
4322 	if (!enable_vnmi)
4323 		return vmx->loaded_vmcs->soft_vnmi_blocked;
4324 	if (vmx->loaded_vmcs->nmi_known_unmasked)
4325 		return false;
4326 	masked = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI;
4327 	vmx->loaded_vmcs->nmi_known_unmasked = !masked;
4328 	return masked;
4329 }
4330 
4331 void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
4332 {
4333 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4334 
4335 	if (!enable_vnmi) {
4336 		if (vmx->loaded_vmcs->soft_vnmi_blocked != masked) {
4337 			vmx->loaded_vmcs->soft_vnmi_blocked = masked;
4338 			vmx->loaded_vmcs->vnmi_blocked_time = 0;
4339 		}
4340 	} else {
4341 		vmx->loaded_vmcs->nmi_known_unmasked = !masked;
4342 		if (masked)
4343 			vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
4344 				      GUEST_INTR_STATE_NMI);
4345 		else
4346 			vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
4347 					GUEST_INTR_STATE_NMI);
4348 	}
4349 }
4350 
4351 static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
4352 {
4353 	if (to_vmx(vcpu)->nested.nested_run_pending)
4354 		return 0;
4355 
4356 	if (!enable_vnmi &&
4357 	    to_vmx(vcpu)->loaded_vmcs->soft_vnmi_blocked)
4358 		return 0;
4359 
4360 	return	!(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4361 		  (GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_STI
4362 		   | GUEST_INTR_STATE_NMI));
4363 }
4364 
4365 static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
4366 {
4367 	return (!to_vmx(vcpu)->nested.nested_run_pending &&
4368 		vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
4369 		!(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4370 			(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
4371 }
4372 
4373 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
4374 {
4375 	int ret;
4376 
4377 	if (enable_unrestricted_guest)
4378 		return 0;
4379 
4380 	ret = x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, addr,
4381 				    PAGE_SIZE * 3);
4382 	if (ret)
4383 		return ret;
4384 	to_kvm_vmx(kvm)->tss_addr = addr;
4385 	return init_rmode_tss(kvm);
4386 }
4387 
4388 static int vmx_set_identity_map_addr(struct kvm *kvm, u64 ident_addr)
4389 {
4390 	to_kvm_vmx(kvm)->ept_identity_map_addr = ident_addr;
4391 	return 0;
4392 }
4393 
4394 static bool rmode_exception(struct kvm_vcpu *vcpu, int vec)
4395 {
4396 	switch (vec) {
4397 	case BP_VECTOR:
4398 		/*
4399 		 * Update instruction length as we may reinject the exception
4400 		 * from user space while in guest debugging mode.
4401 		 */
4402 		to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
4403 			vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4404 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
4405 			return false;
4406 		/* fall through */
4407 	case DB_VECTOR:
4408 		if (vcpu->guest_debug &
4409 			(KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
4410 			return false;
4411 		/* fall through */
4412 	case DE_VECTOR:
4413 	case OF_VECTOR:
4414 	case BR_VECTOR:
4415 	case UD_VECTOR:
4416 	case DF_VECTOR:
4417 	case SS_VECTOR:
4418 	case GP_VECTOR:
4419 	case MF_VECTOR:
4420 		return true;
4421 	break;
4422 	}
4423 	return false;
4424 }
4425 
4426 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
4427 				  int vec, u32 err_code)
4428 {
4429 	/*
4430 	 * Instruction with address size override prefix opcode 0x67
4431 	 * Cause the #SS fault with 0 error code in VM86 mode.
4432 	 */
4433 	if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0) {
4434 		if (kvm_emulate_instruction(vcpu, 0) == EMULATE_DONE) {
4435 			if (vcpu->arch.halt_request) {
4436 				vcpu->arch.halt_request = 0;
4437 				return kvm_vcpu_halt(vcpu);
4438 			}
4439 			return 1;
4440 		}
4441 		return 0;
4442 	}
4443 
4444 	/*
4445 	 * Forward all other exceptions that are valid in real mode.
4446 	 * FIXME: Breaks guest debugging in real mode, needs to be fixed with
4447 	 *        the required debugging infrastructure rework.
4448 	 */
4449 	kvm_queue_exception(vcpu, vec);
4450 	return 1;
4451 }
4452 
4453 /*
4454  * Trigger machine check on the host. We assume all the MSRs are already set up
4455  * by the CPU and that we still run on the same CPU as the MCE occurred on.
4456  * We pass a fake environment to the machine check handler because we want
4457  * the guest to be always treated like user space, no matter what context
4458  * it used internally.
4459  */
4460 static void kvm_machine_check(void)
4461 {
4462 #if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
4463 	struct pt_regs regs = {
4464 		.cs = 3, /* Fake ring 3 no matter what the guest ran on */
4465 		.flags = X86_EFLAGS_IF,
4466 	};
4467 
4468 	do_machine_check(&regs, 0);
4469 #endif
4470 }
4471 
4472 static int handle_machine_check(struct kvm_vcpu *vcpu)
4473 {
4474 	/* handled by vmx_vcpu_run() */
4475 	return 1;
4476 }
4477 
4478 static int handle_exception_nmi(struct kvm_vcpu *vcpu)
4479 {
4480 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4481 	struct kvm_run *kvm_run = vcpu->run;
4482 	u32 intr_info, ex_no, error_code;
4483 	unsigned long cr2, rip, dr6;
4484 	u32 vect_info;
4485 	enum emulation_result er;
4486 
4487 	vect_info = vmx->idt_vectoring_info;
4488 	intr_info = vmx->exit_intr_info;
4489 
4490 	if (is_machine_check(intr_info) || is_nmi(intr_info))
4491 		return 1; /* handled by handle_exception_nmi_irqoff() */
4492 
4493 	if (is_invalid_opcode(intr_info))
4494 		return handle_ud(vcpu);
4495 
4496 	error_code = 0;
4497 	if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
4498 		error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
4499 
4500 	if (!vmx->rmode.vm86_active && is_gp_fault(intr_info)) {
4501 		WARN_ON_ONCE(!enable_vmware_backdoor);
4502 		er = kvm_emulate_instruction(vcpu,
4503 			EMULTYPE_VMWARE | EMULTYPE_NO_UD_ON_FAIL);
4504 		if (er == EMULATE_USER_EXIT)
4505 			return 0;
4506 		else if (er != EMULATE_DONE)
4507 			kvm_queue_exception_e(vcpu, GP_VECTOR, error_code);
4508 		return 1;
4509 	}
4510 
4511 	/*
4512 	 * The #PF with PFEC.RSVD = 1 indicates the guest is accessing
4513 	 * MMIO, it is better to report an internal error.
4514 	 * See the comments in vmx_handle_exit.
4515 	 */
4516 	if ((vect_info & VECTORING_INFO_VALID_MASK) &&
4517 	    !(is_page_fault(intr_info) && !(error_code & PFERR_RSVD_MASK))) {
4518 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4519 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
4520 		vcpu->run->internal.ndata = 3;
4521 		vcpu->run->internal.data[0] = vect_info;
4522 		vcpu->run->internal.data[1] = intr_info;
4523 		vcpu->run->internal.data[2] = error_code;
4524 		return 0;
4525 	}
4526 
4527 	if (is_page_fault(intr_info)) {
4528 		cr2 = vmcs_readl(EXIT_QUALIFICATION);
4529 		/* EPT won't cause page fault directly */
4530 		WARN_ON_ONCE(!vcpu->arch.apf.host_apf_reason && enable_ept);
4531 		return kvm_handle_page_fault(vcpu, error_code, cr2, NULL, 0);
4532 	}
4533 
4534 	ex_no = intr_info & INTR_INFO_VECTOR_MASK;
4535 
4536 	if (vmx->rmode.vm86_active && rmode_exception(vcpu, ex_no))
4537 		return handle_rmode_exception(vcpu, ex_no, error_code);
4538 
4539 	switch (ex_no) {
4540 	case AC_VECTOR:
4541 		kvm_queue_exception_e(vcpu, AC_VECTOR, error_code);
4542 		return 1;
4543 	case DB_VECTOR:
4544 		dr6 = vmcs_readl(EXIT_QUALIFICATION);
4545 		if (!(vcpu->guest_debug &
4546 		      (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
4547 			vcpu->arch.dr6 &= ~DR_TRAP_BITS;
4548 			vcpu->arch.dr6 |= dr6 | DR6_RTM;
4549 			if (is_icebp(intr_info))
4550 				skip_emulated_instruction(vcpu);
4551 
4552 			kvm_queue_exception(vcpu, DB_VECTOR);
4553 			return 1;
4554 		}
4555 		kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
4556 		kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
4557 		/* fall through */
4558 	case BP_VECTOR:
4559 		/*
4560 		 * Update instruction length as we may reinject #BP from
4561 		 * user space while in guest debugging mode. Reading it for
4562 		 * #DB as well causes no harm, it is not used in that case.
4563 		 */
4564 		vmx->vcpu.arch.event_exit_inst_len =
4565 			vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4566 		kvm_run->exit_reason = KVM_EXIT_DEBUG;
4567 		rip = kvm_rip_read(vcpu);
4568 		kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
4569 		kvm_run->debug.arch.exception = ex_no;
4570 		break;
4571 	default:
4572 		kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
4573 		kvm_run->ex.exception = ex_no;
4574 		kvm_run->ex.error_code = error_code;
4575 		break;
4576 	}
4577 	return 0;
4578 }
4579 
4580 static int handle_external_interrupt(struct kvm_vcpu *vcpu)
4581 {
4582 	++vcpu->stat.irq_exits;
4583 	return 1;
4584 }
4585 
4586 static int handle_triple_fault(struct kvm_vcpu *vcpu)
4587 {
4588 	vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
4589 	vcpu->mmio_needed = 0;
4590 	return 0;
4591 }
4592 
4593 static int handle_io(struct kvm_vcpu *vcpu)
4594 {
4595 	unsigned long exit_qualification;
4596 	int size, in, string;
4597 	unsigned port;
4598 
4599 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4600 	string = (exit_qualification & 16) != 0;
4601 
4602 	++vcpu->stat.io_exits;
4603 
4604 	if (string)
4605 		return kvm_emulate_instruction(vcpu, 0) == EMULATE_DONE;
4606 
4607 	port = exit_qualification >> 16;
4608 	size = (exit_qualification & 7) + 1;
4609 	in = (exit_qualification & 8) != 0;
4610 
4611 	return kvm_fast_pio(vcpu, size, port, in);
4612 }
4613 
4614 static void
4615 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
4616 {
4617 	/*
4618 	 * Patch in the VMCALL instruction:
4619 	 */
4620 	hypercall[0] = 0x0f;
4621 	hypercall[1] = 0x01;
4622 	hypercall[2] = 0xc1;
4623 }
4624 
4625 /* called to set cr0 as appropriate for a mov-to-cr0 exit. */
4626 static int handle_set_cr0(struct kvm_vcpu *vcpu, unsigned long val)
4627 {
4628 	if (is_guest_mode(vcpu)) {
4629 		struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4630 		unsigned long orig_val = val;
4631 
4632 		/*
4633 		 * We get here when L2 changed cr0 in a way that did not change
4634 		 * any of L1's shadowed bits (see nested_vmx_exit_handled_cr),
4635 		 * but did change L0 shadowed bits. So we first calculate the
4636 		 * effective cr0 value that L1 would like to write into the
4637 		 * hardware. It consists of the L2-owned bits from the new
4638 		 * value combined with the L1-owned bits from L1's guest_cr0.
4639 		 */
4640 		val = (val & ~vmcs12->cr0_guest_host_mask) |
4641 			(vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask);
4642 
4643 		if (!nested_guest_cr0_valid(vcpu, val))
4644 			return 1;
4645 
4646 		if (kvm_set_cr0(vcpu, val))
4647 			return 1;
4648 		vmcs_writel(CR0_READ_SHADOW, orig_val);
4649 		return 0;
4650 	} else {
4651 		if (to_vmx(vcpu)->nested.vmxon &&
4652 		    !nested_host_cr0_valid(vcpu, val))
4653 			return 1;
4654 
4655 		return kvm_set_cr0(vcpu, val);
4656 	}
4657 }
4658 
4659 static int handle_set_cr4(struct kvm_vcpu *vcpu, unsigned long val)
4660 {
4661 	if (is_guest_mode(vcpu)) {
4662 		struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4663 		unsigned long orig_val = val;
4664 
4665 		/* analogously to handle_set_cr0 */
4666 		val = (val & ~vmcs12->cr4_guest_host_mask) |
4667 			(vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask);
4668 		if (kvm_set_cr4(vcpu, val))
4669 			return 1;
4670 		vmcs_writel(CR4_READ_SHADOW, orig_val);
4671 		return 0;
4672 	} else
4673 		return kvm_set_cr4(vcpu, val);
4674 }
4675 
4676 static int handle_desc(struct kvm_vcpu *vcpu)
4677 {
4678 	WARN_ON(!(vcpu->arch.cr4 & X86_CR4_UMIP));
4679 	return kvm_emulate_instruction(vcpu, 0) == EMULATE_DONE;
4680 }
4681 
4682 static int handle_cr(struct kvm_vcpu *vcpu)
4683 {
4684 	unsigned long exit_qualification, val;
4685 	int cr;
4686 	int reg;
4687 	int err;
4688 	int ret;
4689 
4690 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4691 	cr = exit_qualification & 15;
4692 	reg = (exit_qualification >> 8) & 15;
4693 	switch ((exit_qualification >> 4) & 3) {
4694 	case 0: /* mov to cr */
4695 		val = kvm_register_readl(vcpu, reg);
4696 		trace_kvm_cr_write(cr, val);
4697 		switch (cr) {
4698 		case 0:
4699 			err = handle_set_cr0(vcpu, val);
4700 			return kvm_complete_insn_gp(vcpu, err);
4701 		case 3:
4702 			WARN_ON_ONCE(enable_unrestricted_guest);
4703 			err = kvm_set_cr3(vcpu, val);
4704 			return kvm_complete_insn_gp(vcpu, err);
4705 		case 4:
4706 			err = handle_set_cr4(vcpu, val);
4707 			return kvm_complete_insn_gp(vcpu, err);
4708 		case 8: {
4709 				u8 cr8_prev = kvm_get_cr8(vcpu);
4710 				u8 cr8 = (u8)val;
4711 				err = kvm_set_cr8(vcpu, cr8);
4712 				ret = kvm_complete_insn_gp(vcpu, err);
4713 				if (lapic_in_kernel(vcpu))
4714 					return ret;
4715 				if (cr8_prev <= cr8)
4716 					return ret;
4717 				/*
4718 				 * TODO: we might be squashing a
4719 				 * KVM_GUESTDBG_SINGLESTEP-triggered
4720 				 * KVM_EXIT_DEBUG here.
4721 				 */
4722 				vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
4723 				return 0;
4724 			}
4725 		}
4726 		break;
4727 	case 2: /* clts */
4728 		WARN_ONCE(1, "Guest should always own CR0.TS");
4729 		vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
4730 		trace_kvm_cr_write(0, kvm_read_cr0(vcpu));
4731 		return kvm_skip_emulated_instruction(vcpu);
4732 	case 1: /*mov from cr*/
4733 		switch (cr) {
4734 		case 3:
4735 			WARN_ON_ONCE(enable_unrestricted_guest);
4736 			val = kvm_read_cr3(vcpu);
4737 			kvm_register_write(vcpu, reg, val);
4738 			trace_kvm_cr_read(cr, val);
4739 			return kvm_skip_emulated_instruction(vcpu);
4740 		case 8:
4741 			val = kvm_get_cr8(vcpu);
4742 			kvm_register_write(vcpu, reg, val);
4743 			trace_kvm_cr_read(cr, val);
4744 			return kvm_skip_emulated_instruction(vcpu);
4745 		}
4746 		break;
4747 	case 3: /* lmsw */
4748 		val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
4749 		trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val);
4750 		kvm_lmsw(vcpu, val);
4751 
4752 		return kvm_skip_emulated_instruction(vcpu);
4753 	default:
4754 		break;
4755 	}
4756 	vcpu->run->exit_reason = 0;
4757 	vcpu_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
4758 	       (int)(exit_qualification >> 4) & 3, cr);
4759 	return 0;
4760 }
4761 
4762 static int handle_dr(struct kvm_vcpu *vcpu)
4763 {
4764 	unsigned long exit_qualification;
4765 	int dr, dr7, reg;
4766 
4767 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4768 	dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
4769 
4770 	/* First, if DR does not exist, trigger UD */
4771 	if (!kvm_require_dr(vcpu, dr))
4772 		return 1;
4773 
4774 	/* Do not handle if the CPL > 0, will trigger GP on re-entry */
4775 	if (!kvm_require_cpl(vcpu, 0))
4776 		return 1;
4777 	dr7 = vmcs_readl(GUEST_DR7);
4778 	if (dr7 & DR7_GD) {
4779 		/*
4780 		 * As the vm-exit takes precedence over the debug trap, we
4781 		 * need to emulate the latter, either for the host or the
4782 		 * guest debugging itself.
4783 		 */
4784 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
4785 			vcpu->run->debug.arch.dr6 = vcpu->arch.dr6;
4786 			vcpu->run->debug.arch.dr7 = dr7;
4787 			vcpu->run->debug.arch.pc = kvm_get_linear_rip(vcpu);
4788 			vcpu->run->debug.arch.exception = DB_VECTOR;
4789 			vcpu->run->exit_reason = KVM_EXIT_DEBUG;
4790 			return 0;
4791 		} else {
4792 			vcpu->arch.dr6 &= ~DR_TRAP_BITS;
4793 			vcpu->arch.dr6 |= DR6_BD | DR6_RTM;
4794 			kvm_queue_exception(vcpu, DB_VECTOR);
4795 			return 1;
4796 		}
4797 	}
4798 
4799 	if (vcpu->guest_debug == 0) {
4800 		exec_controls_clearbit(to_vmx(vcpu), CPU_BASED_MOV_DR_EXITING);
4801 
4802 		/*
4803 		 * No more DR vmexits; force a reload of the debug registers
4804 		 * and reenter on this instruction.  The next vmexit will
4805 		 * retrieve the full state of the debug registers.
4806 		 */
4807 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
4808 		return 1;
4809 	}
4810 
4811 	reg = DEBUG_REG_ACCESS_REG(exit_qualification);
4812 	if (exit_qualification & TYPE_MOV_FROM_DR) {
4813 		unsigned long val;
4814 
4815 		if (kvm_get_dr(vcpu, dr, &val))
4816 			return 1;
4817 		kvm_register_write(vcpu, reg, val);
4818 	} else
4819 		if (kvm_set_dr(vcpu, dr, kvm_register_readl(vcpu, reg)))
4820 			return 1;
4821 
4822 	return kvm_skip_emulated_instruction(vcpu);
4823 }
4824 
4825 static u64 vmx_get_dr6(struct kvm_vcpu *vcpu)
4826 {
4827 	return vcpu->arch.dr6;
4828 }
4829 
4830 static void vmx_set_dr6(struct kvm_vcpu *vcpu, unsigned long val)
4831 {
4832 }
4833 
4834 static void vmx_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
4835 {
4836 	get_debugreg(vcpu->arch.db[0], 0);
4837 	get_debugreg(vcpu->arch.db[1], 1);
4838 	get_debugreg(vcpu->arch.db[2], 2);
4839 	get_debugreg(vcpu->arch.db[3], 3);
4840 	get_debugreg(vcpu->arch.dr6, 6);
4841 	vcpu->arch.dr7 = vmcs_readl(GUEST_DR7);
4842 
4843 	vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
4844 	exec_controls_setbit(to_vmx(vcpu), CPU_BASED_MOV_DR_EXITING);
4845 }
4846 
4847 static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
4848 {
4849 	vmcs_writel(GUEST_DR7, val);
4850 }
4851 
4852 static int handle_cpuid(struct kvm_vcpu *vcpu)
4853 {
4854 	return kvm_emulate_cpuid(vcpu);
4855 }
4856 
4857 static int handle_rdmsr(struct kvm_vcpu *vcpu)
4858 {
4859 	u32 ecx = kvm_rcx_read(vcpu);
4860 	struct msr_data msr_info;
4861 
4862 	msr_info.index = ecx;
4863 	msr_info.host_initiated = false;
4864 	if (vmx_get_msr(vcpu, &msr_info)) {
4865 		trace_kvm_msr_read_ex(ecx);
4866 		kvm_inject_gp(vcpu, 0);
4867 		return 1;
4868 	}
4869 
4870 	trace_kvm_msr_read(ecx, msr_info.data);
4871 
4872 	kvm_rax_write(vcpu, msr_info.data & -1u);
4873 	kvm_rdx_write(vcpu, (msr_info.data >> 32) & -1u);
4874 	return kvm_skip_emulated_instruction(vcpu);
4875 }
4876 
4877 static int handle_wrmsr(struct kvm_vcpu *vcpu)
4878 {
4879 	struct msr_data msr;
4880 	u32 ecx = kvm_rcx_read(vcpu);
4881 	u64 data = kvm_read_edx_eax(vcpu);
4882 
4883 	msr.data = data;
4884 	msr.index = ecx;
4885 	msr.host_initiated = false;
4886 	if (kvm_set_msr(vcpu, &msr) != 0) {
4887 		trace_kvm_msr_write_ex(ecx, data);
4888 		kvm_inject_gp(vcpu, 0);
4889 		return 1;
4890 	}
4891 
4892 	trace_kvm_msr_write(ecx, data);
4893 	return kvm_skip_emulated_instruction(vcpu);
4894 }
4895 
4896 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
4897 {
4898 	kvm_apic_update_ppr(vcpu);
4899 	return 1;
4900 }
4901 
4902 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
4903 {
4904 	exec_controls_clearbit(to_vmx(vcpu), CPU_BASED_VIRTUAL_INTR_PENDING);
4905 
4906 	kvm_make_request(KVM_REQ_EVENT, vcpu);
4907 
4908 	++vcpu->stat.irq_window_exits;
4909 	return 1;
4910 }
4911 
4912 static int handle_halt(struct kvm_vcpu *vcpu)
4913 {
4914 	return kvm_emulate_halt(vcpu);
4915 }
4916 
4917 static int handle_vmcall(struct kvm_vcpu *vcpu)
4918 {
4919 	return kvm_emulate_hypercall(vcpu);
4920 }
4921 
4922 static int handle_invd(struct kvm_vcpu *vcpu)
4923 {
4924 	return kvm_emulate_instruction(vcpu, 0) == EMULATE_DONE;
4925 }
4926 
4927 static int handle_invlpg(struct kvm_vcpu *vcpu)
4928 {
4929 	unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4930 
4931 	kvm_mmu_invlpg(vcpu, exit_qualification);
4932 	return kvm_skip_emulated_instruction(vcpu);
4933 }
4934 
4935 static int handle_rdpmc(struct kvm_vcpu *vcpu)
4936 {
4937 	int err;
4938 
4939 	err = kvm_rdpmc(vcpu);
4940 	return kvm_complete_insn_gp(vcpu, err);
4941 }
4942 
4943 static int handle_wbinvd(struct kvm_vcpu *vcpu)
4944 {
4945 	return kvm_emulate_wbinvd(vcpu);
4946 }
4947 
4948 static int handle_xsetbv(struct kvm_vcpu *vcpu)
4949 {
4950 	u64 new_bv = kvm_read_edx_eax(vcpu);
4951 	u32 index = kvm_rcx_read(vcpu);
4952 
4953 	if (kvm_set_xcr(vcpu, index, new_bv) == 0)
4954 		return kvm_skip_emulated_instruction(vcpu);
4955 	return 1;
4956 }
4957 
4958 static int handle_xsaves(struct kvm_vcpu *vcpu)
4959 {
4960 	kvm_skip_emulated_instruction(vcpu);
4961 	WARN(1, "this should never happen\n");
4962 	return 1;
4963 }
4964 
4965 static int handle_xrstors(struct kvm_vcpu *vcpu)
4966 {
4967 	kvm_skip_emulated_instruction(vcpu);
4968 	WARN(1, "this should never happen\n");
4969 	return 1;
4970 }
4971 
4972 static int handle_apic_access(struct kvm_vcpu *vcpu)
4973 {
4974 	if (likely(fasteoi)) {
4975 		unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4976 		int access_type, offset;
4977 
4978 		access_type = exit_qualification & APIC_ACCESS_TYPE;
4979 		offset = exit_qualification & APIC_ACCESS_OFFSET;
4980 		/*
4981 		 * Sane guest uses MOV to write EOI, with written value
4982 		 * not cared. So make a short-circuit here by avoiding
4983 		 * heavy instruction emulation.
4984 		 */
4985 		if ((access_type == TYPE_LINEAR_APIC_INST_WRITE) &&
4986 		    (offset == APIC_EOI)) {
4987 			kvm_lapic_set_eoi(vcpu);
4988 			return kvm_skip_emulated_instruction(vcpu);
4989 		}
4990 	}
4991 	return kvm_emulate_instruction(vcpu, 0) == EMULATE_DONE;
4992 }
4993 
4994 static int handle_apic_eoi_induced(struct kvm_vcpu *vcpu)
4995 {
4996 	unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4997 	int vector = exit_qualification & 0xff;
4998 
4999 	/* EOI-induced VM exit is trap-like and thus no need to adjust IP */
5000 	kvm_apic_set_eoi_accelerated(vcpu, vector);
5001 	return 1;
5002 }
5003 
5004 static int handle_apic_write(struct kvm_vcpu *vcpu)
5005 {
5006 	unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5007 	u32 offset = exit_qualification & 0xfff;
5008 
5009 	/* APIC-write VM exit is trap-like and thus no need to adjust IP */
5010 	kvm_apic_write_nodecode(vcpu, offset);
5011 	return 1;
5012 }
5013 
5014 static int handle_task_switch(struct kvm_vcpu *vcpu)
5015 {
5016 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5017 	unsigned long exit_qualification;
5018 	bool has_error_code = false;
5019 	u32 error_code = 0;
5020 	u16 tss_selector;
5021 	int reason, type, idt_v, idt_index;
5022 
5023 	idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
5024 	idt_index = (vmx->idt_vectoring_info & VECTORING_INFO_VECTOR_MASK);
5025 	type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
5026 
5027 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5028 
5029 	reason = (u32)exit_qualification >> 30;
5030 	if (reason == TASK_SWITCH_GATE && idt_v) {
5031 		switch (type) {
5032 		case INTR_TYPE_NMI_INTR:
5033 			vcpu->arch.nmi_injected = false;
5034 			vmx_set_nmi_mask(vcpu, true);
5035 			break;
5036 		case INTR_TYPE_EXT_INTR:
5037 		case INTR_TYPE_SOFT_INTR:
5038 			kvm_clear_interrupt_queue(vcpu);
5039 			break;
5040 		case INTR_TYPE_HARD_EXCEPTION:
5041 			if (vmx->idt_vectoring_info &
5042 			    VECTORING_INFO_DELIVER_CODE_MASK) {
5043 				has_error_code = true;
5044 				error_code =
5045 					vmcs_read32(IDT_VECTORING_ERROR_CODE);
5046 			}
5047 			/* fall through */
5048 		case INTR_TYPE_SOFT_EXCEPTION:
5049 			kvm_clear_exception_queue(vcpu);
5050 			break;
5051 		default:
5052 			break;
5053 		}
5054 	}
5055 	tss_selector = exit_qualification;
5056 
5057 	if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
5058 		       type != INTR_TYPE_EXT_INTR &&
5059 		       type != INTR_TYPE_NMI_INTR))
5060 		skip_emulated_instruction(vcpu);
5061 
5062 	if (kvm_task_switch(vcpu, tss_selector,
5063 			    type == INTR_TYPE_SOFT_INTR ? idt_index : -1, reason,
5064 			    has_error_code, error_code) == EMULATE_FAIL) {
5065 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5066 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5067 		vcpu->run->internal.ndata = 0;
5068 		return 0;
5069 	}
5070 
5071 	/*
5072 	 * TODO: What about debug traps on tss switch?
5073 	 *       Are we supposed to inject them and update dr6?
5074 	 */
5075 
5076 	return 1;
5077 }
5078 
5079 static int handle_ept_violation(struct kvm_vcpu *vcpu)
5080 {
5081 	unsigned long exit_qualification;
5082 	gpa_t gpa;
5083 	u64 error_code;
5084 
5085 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5086 
5087 	/*
5088 	 * EPT violation happened while executing iret from NMI,
5089 	 * "blocked by NMI" bit has to be set before next VM entry.
5090 	 * There are errata that may cause this bit to not be set:
5091 	 * AAK134, BY25.
5092 	 */
5093 	if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
5094 			enable_vnmi &&
5095 			(exit_qualification & INTR_INFO_UNBLOCK_NMI))
5096 		vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO, GUEST_INTR_STATE_NMI);
5097 
5098 	gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5099 	trace_kvm_page_fault(gpa, exit_qualification);
5100 
5101 	/* Is it a read fault? */
5102 	error_code = (exit_qualification & EPT_VIOLATION_ACC_READ)
5103 		     ? PFERR_USER_MASK : 0;
5104 	/* Is it a write fault? */
5105 	error_code |= (exit_qualification & EPT_VIOLATION_ACC_WRITE)
5106 		      ? PFERR_WRITE_MASK : 0;
5107 	/* Is it a fetch fault? */
5108 	error_code |= (exit_qualification & EPT_VIOLATION_ACC_INSTR)
5109 		      ? PFERR_FETCH_MASK : 0;
5110 	/* ept page table entry is present? */
5111 	error_code |= (exit_qualification &
5112 		       (EPT_VIOLATION_READABLE | EPT_VIOLATION_WRITABLE |
5113 			EPT_VIOLATION_EXECUTABLE))
5114 		      ? PFERR_PRESENT_MASK : 0;
5115 
5116 	error_code |= (exit_qualification & 0x100) != 0 ?
5117 	       PFERR_GUEST_FINAL_MASK : PFERR_GUEST_PAGE_MASK;
5118 
5119 	vcpu->arch.exit_qualification = exit_qualification;
5120 	return kvm_mmu_page_fault(vcpu, gpa, error_code, NULL, 0);
5121 }
5122 
5123 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
5124 {
5125 	gpa_t gpa;
5126 
5127 	/*
5128 	 * A nested guest cannot optimize MMIO vmexits, because we have an
5129 	 * nGPA here instead of the required GPA.
5130 	 */
5131 	gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5132 	if (!is_guest_mode(vcpu) &&
5133 	    !kvm_io_bus_write(vcpu, KVM_FAST_MMIO_BUS, gpa, 0, NULL)) {
5134 		trace_kvm_fast_mmio(gpa);
5135 		/*
5136 		 * Doing kvm_skip_emulated_instruction() depends on undefined
5137 		 * behavior: Intel's manual doesn't mandate
5138 		 * VM_EXIT_INSTRUCTION_LEN to be set in VMCS when EPT MISCONFIG
5139 		 * occurs and while on real hardware it was observed to be set,
5140 		 * other hypervisors (namely Hyper-V) don't set it, we end up
5141 		 * advancing IP with some random value. Disable fast mmio when
5142 		 * running nested and keep it for real hardware in hope that
5143 		 * VM_EXIT_INSTRUCTION_LEN will always be set correctly.
5144 		 */
5145 		if (!static_cpu_has(X86_FEATURE_HYPERVISOR))
5146 			return kvm_skip_emulated_instruction(vcpu);
5147 		else
5148 			return kvm_emulate_instruction(vcpu, EMULTYPE_SKIP) ==
5149 								EMULATE_DONE;
5150 	}
5151 
5152 	return kvm_mmu_page_fault(vcpu, gpa, PFERR_RSVD_MASK, NULL, 0);
5153 }
5154 
5155 static int handle_nmi_window(struct kvm_vcpu *vcpu)
5156 {
5157 	WARN_ON_ONCE(!enable_vnmi);
5158 	exec_controls_clearbit(to_vmx(vcpu), CPU_BASED_VIRTUAL_NMI_PENDING);
5159 	++vcpu->stat.nmi_window_exits;
5160 	kvm_make_request(KVM_REQ_EVENT, vcpu);
5161 
5162 	return 1;
5163 }
5164 
5165 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
5166 {
5167 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5168 	enum emulation_result err = EMULATE_DONE;
5169 	int ret = 1;
5170 	bool intr_window_requested;
5171 	unsigned count = 130;
5172 
5173 	/*
5174 	 * We should never reach the point where we are emulating L2
5175 	 * due to invalid guest state as that means we incorrectly
5176 	 * allowed a nested VMEntry with an invalid vmcs12.
5177 	 */
5178 	WARN_ON_ONCE(vmx->emulation_required && vmx->nested.nested_run_pending);
5179 
5180 	intr_window_requested = exec_controls_get(vmx) &
5181 				CPU_BASED_VIRTUAL_INTR_PENDING;
5182 
5183 	while (vmx->emulation_required && count-- != 0) {
5184 		if (intr_window_requested && vmx_interrupt_allowed(vcpu))
5185 			return handle_interrupt_window(&vmx->vcpu);
5186 
5187 		if (kvm_test_request(KVM_REQ_EVENT, vcpu))
5188 			return 1;
5189 
5190 		err = kvm_emulate_instruction(vcpu, 0);
5191 
5192 		if (err == EMULATE_USER_EXIT) {
5193 			++vcpu->stat.mmio_exits;
5194 			ret = 0;
5195 			goto out;
5196 		}
5197 
5198 		if (err != EMULATE_DONE)
5199 			goto emulation_error;
5200 
5201 		if (vmx->emulation_required && !vmx->rmode.vm86_active &&
5202 		    vcpu->arch.exception.pending)
5203 			goto emulation_error;
5204 
5205 		if (vcpu->arch.halt_request) {
5206 			vcpu->arch.halt_request = 0;
5207 			ret = kvm_vcpu_halt(vcpu);
5208 			goto out;
5209 		}
5210 
5211 		if (signal_pending(current))
5212 			goto out;
5213 		if (need_resched())
5214 			schedule();
5215 	}
5216 
5217 out:
5218 	return ret;
5219 
5220 emulation_error:
5221 	vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5222 	vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5223 	vcpu->run->internal.ndata = 0;
5224 	return 0;
5225 }
5226 
5227 static void grow_ple_window(struct kvm_vcpu *vcpu)
5228 {
5229 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5230 	int old = vmx->ple_window;
5231 
5232 	vmx->ple_window = __grow_ple_window(old, ple_window,
5233 					    ple_window_grow,
5234 					    ple_window_max);
5235 
5236 	if (vmx->ple_window != old)
5237 		vmx->ple_window_dirty = true;
5238 
5239 	trace_kvm_ple_window_grow(vcpu->vcpu_id, vmx->ple_window, old);
5240 }
5241 
5242 static void shrink_ple_window(struct kvm_vcpu *vcpu)
5243 {
5244 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5245 	int old = vmx->ple_window;
5246 
5247 	vmx->ple_window = __shrink_ple_window(old, ple_window,
5248 					      ple_window_shrink,
5249 					      ple_window);
5250 
5251 	if (vmx->ple_window != old)
5252 		vmx->ple_window_dirty = true;
5253 
5254 	trace_kvm_ple_window_shrink(vcpu->vcpu_id, vmx->ple_window, old);
5255 }
5256 
5257 /*
5258  * Handler for POSTED_INTERRUPT_WAKEUP_VECTOR.
5259  */
5260 static void wakeup_handler(void)
5261 {
5262 	struct kvm_vcpu *vcpu;
5263 	int cpu = smp_processor_id();
5264 
5265 	spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
5266 	list_for_each_entry(vcpu, &per_cpu(blocked_vcpu_on_cpu, cpu),
5267 			blocked_vcpu_list) {
5268 		struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
5269 
5270 		if (pi_test_on(pi_desc) == 1)
5271 			kvm_vcpu_kick(vcpu);
5272 	}
5273 	spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
5274 }
5275 
5276 static void vmx_enable_tdp(void)
5277 {
5278 	kvm_mmu_set_mask_ptes(VMX_EPT_READABLE_MASK,
5279 		enable_ept_ad_bits ? VMX_EPT_ACCESS_BIT : 0ull,
5280 		enable_ept_ad_bits ? VMX_EPT_DIRTY_BIT : 0ull,
5281 		0ull, VMX_EPT_EXECUTABLE_MASK,
5282 		cpu_has_vmx_ept_execute_only() ? 0ull : VMX_EPT_READABLE_MASK,
5283 		VMX_EPT_RWX_MASK, 0ull);
5284 
5285 	ept_set_mmio_spte_mask();
5286 	kvm_enable_tdp();
5287 }
5288 
5289 /*
5290  * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
5291  * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
5292  */
5293 static int handle_pause(struct kvm_vcpu *vcpu)
5294 {
5295 	if (!kvm_pause_in_guest(vcpu->kvm))
5296 		grow_ple_window(vcpu);
5297 
5298 	/*
5299 	 * Intel sdm vol3 ch-25.1.3 says: The "PAUSE-loop exiting"
5300 	 * VM-execution control is ignored if CPL > 0. OTOH, KVM
5301 	 * never set PAUSE_EXITING and just set PLE if supported,
5302 	 * so the vcpu must be CPL=0 if it gets a PAUSE exit.
5303 	 */
5304 	kvm_vcpu_on_spin(vcpu, true);
5305 	return kvm_skip_emulated_instruction(vcpu);
5306 }
5307 
5308 static int handle_nop(struct kvm_vcpu *vcpu)
5309 {
5310 	return kvm_skip_emulated_instruction(vcpu);
5311 }
5312 
5313 static int handle_mwait(struct kvm_vcpu *vcpu)
5314 {
5315 	printk_once(KERN_WARNING "kvm: MWAIT instruction emulated as NOP!\n");
5316 	return handle_nop(vcpu);
5317 }
5318 
5319 static int handle_invalid_op(struct kvm_vcpu *vcpu)
5320 {
5321 	kvm_queue_exception(vcpu, UD_VECTOR);
5322 	return 1;
5323 }
5324 
5325 static int handle_monitor_trap(struct kvm_vcpu *vcpu)
5326 {
5327 	return 1;
5328 }
5329 
5330 static int handle_monitor(struct kvm_vcpu *vcpu)
5331 {
5332 	printk_once(KERN_WARNING "kvm: MONITOR instruction emulated as NOP!\n");
5333 	return handle_nop(vcpu);
5334 }
5335 
5336 static int handle_invpcid(struct kvm_vcpu *vcpu)
5337 {
5338 	u32 vmx_instruction_info;
5339 	unsigned long type;
5340 	bool pcid_enabled;
5341 	gva_t gva;
5342 	struct x86_exception e;
5343 	unsigned i;
5344 	unsigned long roots_to_free = 0;
5345 	struct {
5346 		u64 pcid;
5347 		u64 gla;
5348 	} operand;
5349 
5350 	if (!guest_cpuid_has(vcpu, X86_FEATURE_INVPCID)) {
5351 		kvm_queue_exception(vcpu, UD_VECTOR);
5352 		return 1;
5353 	}
5354 
5355 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5356 	type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
5357 
5358 	if (type > 3) {
5359 		kvm_inject_gp(vcpu, 0);
5360 		return 1;
5361 	}
5362 
5363 	/* According to the Intel instruction reference, the memory operand
5364 	 * is read even if it isn't needed (e.g., for type==all)
5365 	 */
5366 	if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
5367 				vmx_instruction_info, false,
5368 				sizeof(operand), &gva))
5369 		return 1;
5370 
5371 	if (kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e)) {
5372 		kvm_inject_page_fault(vcpu, &e);
5373 		return 1;
5374 	}
5375 
5376 	if (operand.pcid >> 12 != 0) {
5377 		kvm_inject_gp(vcpu, 0);
5378 		return 1;
5379 	}
5380 
5381 	pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
5382 
5383 	switch (type) {
5384 	case INVPCID_TYPE_INDIV_ADDR:
5385 		if ((!pcid_enabled && (operand.pcid != 0)) ||
5386 		    is_noncanonical_address(operand.gla, vcpu)) {
5387 			kvm_inject_gp(vcpu, 0);
5388 			return 1;
5389 		}
5390 		kvm_mmu_invpcid_gva(vcpu, operand.gla, operand.pcid);
5391 		return kvm_skip_emulated_instruction(vcpu);
5392 
5393 	case INVPCID_TYPE_SINGLE_CTXT:
5394 		if (!pcid_enabled && (operand.pcid != 0)) {
5395 			kvm_inject_gp(vcpu, 0);
5396 			return 1;
5397 		}
5398 
5399 		if (kvm_get_active_pcid(vcpu) == operand.pcid) {
5400 			kvm_mmu_sync_roots(vcpu);
5401 			kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
5402 		}
5403 
5404 		for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++)
5405 			if (kvm_get_pcid(vcpu, vcpu->arch.mmu->prev_roots[i].cr3)
5406 			    == operand.pcid)
5407 				roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i);
5408 
5409 		kvm_mmu_free_roots(vcpu, vcpu->arch.mmu, roots_to_free);
5410 		/*
5411 		 * If neither the current cr3 nor any of the prev_roots use the
5412 		 * given PCID, then nothing needs to be done here because a
5413 		 * resync will happen anyway before switching to any other CR3.
5414 		 */
5415 
5416 		return kvm_skip_emulated_instruction(vcpu);
5417 
5418 	case INVPCID_TYPE_ALL_NON_GLOBAL:
5419 		/*
5420 		 * Currently, KVM doesn't mark global entries in the shadow
5421 		 * page tables, so a non-global flush just degenerates to a
5422 		 * global flush. If needed, we could optimize this later by
5423 		 * keeping track of global entries in shadow page tables.
5424 		 */
5425 
5426 		/* fall-through */
5427 	case INVPCID_TYPE_ALL_INCL_GLOBAL:
5428 		kvm_mmu_unload(vcpu);
5429 		return kvm_skip_emulated_instruction(vcpu);
5430 
5431 	default:
5432 		BUG(); /* We have already checked above that type <= 3 */
5433 	}
5434 }
5435 
5436 static int handle_pml_full(struct kvm_vcpu *vcpu)
5437 {
5438 	unsigned long exit_qualification;
5439 
5440 	trace_kvm_pml_full(vcpu->vcpu_id);
5441 
5442 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5443 
5444 	/*
5445 	 * PML buffer FULL happened while executing iret from NMI,
5446 	 * "blocked by NMI" bit has to be set before next VM entry.
5447 	 */
5448 	if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
5449 			enable_vnmi &&
5450 			(exit_qualification & INTR_INFO_UNBLOCK_NMI))
5451 		vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
5452 				GUEST_INTR_STATE_NMI);
5453 
5454 	/*
5455 	 * PML buffer already flushed at beginning of VMEXIT. Nothing to do
5456 	 * here.., and there's no userspace involvement needed for PML.
5457 	 */
5458 	return 1;
5459 }
5460 
5461 static int handle_preemption_timer(struct kvm_vcpu *vcpu)
5462 {
5463 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5464 
5465 	if (!vmx->req_immediate_exit &&
5466 	    !unlikely(vmx->loaded_vmcs->hv_timer_soft_disabled))
5467 		kvm_lapic_expired_hv_timer(vcpu);
5468 
5469 	return 1;
5470 }
5471 
5472 /*
5473  * When nested=0, all VMX instruction VM Exits filter here.  The handlers
5474  * are overwritten by nested_vmx_setup() when nested=1.
5475  */
5476 static int handle_vmx_instruction(struct kvm_vcpu *vcpu)
5477 {
5478 	kvm_queue_exception(vcpu, UD_VECTOR);
5479 	return 1;
5480 }
5481 
5482 static int handle_encls(struct kvm_vcpu *vcpu)
5483 {
5484 	/*
5485 	 * SGX virtualization is not yet supported.  There is no software
5486 	 * enable bit for SGX, so we have to trap ENCLS and inject a #UD
5487 	 * to prevent the guest from executing ENCLS.
5488 	 */
5489 	kvm_queue_exception(vcpu, UD_VECTOR);
5490 	return 1;
5491 }
5492 
5493 /*
5494  * The exit handlers return 1 if the exit was handled fully and guest execution
5495  * may resume.  Otherwise they set the kvm_run parameter to indicate what needs
5496  * to be done to userspace and return 0.
5497  */
5498 static int (*kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
5499 	[EXIT_REASON_EXCEPTION_NMI]           = handle_exception_nmi,
5500 	[EXIT_REASON_EXTERNAL_INTERRUPT]      = handle_external_interrupt,
5501 	[EXIT_REASON_TRIPLE_FAULT]            = handle_triple_fault,
5502 	[EXIT_REASON_NMI_WINDOW]	      = handle_nmi_window,
5503 	[EXIT_REASON_IO_INSTRUCTION]          = handle_io,
5504 	[EXIT_REASON_CR_ACCESS]               = handle_cr,
5505 	[EXIT_REASON_DR_ACCESS]               = handle_dr,
5506 	[EXIT_REASON_CPUID]                   = handle_cpuid,
5507 	[EXIT_REASON_MSR_READ]                = handle_rdmsr,
5508 	[EXIT_REASON_MSR_WRITE]               = handle_wrmsr,
5509 	[EXIT_REASON_PENDING_INTERRUPT]       = handle_interrupt_window,
5510 	[EXIT_REASON_HLT]                     = handle_halt,
5511 	[EXIT_REASON_INVD]		      = handle_invd,
5512 	[EXIT_REASON_INVLPG]		      = handle_invlpg,
5513 	[EXIT_REASON_RDPMC]                   = handle_rdpmc,
5514 	[EXIT_REASON_VMCALL]                  = handle_vmcall,
5515 	[EXIT_REASON_VMCLEAR]		      = handle_vmx_instruction,
5516 	[EXIT_REASON_VMLAUNCH]		      = handle_vmx_instruction,
5517 	[EXIT_REASON_VMPTRLD]		      = handle_vmx_instruction,
5518 	[EXIT_REASON_VMPTRST]		      = handle_vmx_instruction,
5519 	[EXIT_REASON_VMREAD]		      = handle_vmx_instruction,
5520 	[EXIT_REASON_VMRESUME]		      = handle_vmx_instruction,
5521 	[EXIT_REASON_VMWRITE]		      = handle_vmx_instruction,
5522 	[EXIT_REASON_VMOFF]		      = handle_vmx_instruction,
5523 	[EXIT_REASON_VMON]		      = handle_vmx_instruction,
5524 	[EXIT_REASON_TPR_BELOW_THRESHOLD]     = handle_tpr_below_threshold,
5525 	[EXIT_REASON_APIC_ACCESS]             = handle_apic_access,
5526 	[EXIT_REASON_APIC_WRITE]              = handle_apic_write,
5527 	[EXIT_REASON_EOI_INDUCED]             = handle_apic_eoi_induced,
5528 	[EXIT_REASON_WBINVD]                  = handle_wbinvd,
5529 	[EXIT_REASON_XSETBV]                  = handle_xsetbv,
5530 	[EXIT_REASON_TASK_SWITCH]             = handle_task_switch,
5531 	[EXIT_REASON_MCE_DURING_VMENTRY]      = handle_machine_check,
5532 	[EXIT_REASON_GDTR_IDTR]		      = handle_desc,
5533 	[EXIT_REASON_LDTR_TR]		      = handle_desc,
5534 	[EXIT_REASON_EPT_VIOLATION]	      = handle_ept_violation,
5535 	[EXIT_REASON_EPT_MISCONFIG]           = handle_ept_misconfig,
5536 	[EXIT_REASON_PAUSE_INSTRUCTION]       = handle_pause,
5537 	[EXIT_REASON_MWAIT_INSTRUCTION]	      = handle_mwait,
5538 	[EXIT_REASON_MONITOR_TRAP_FLAG]       = handle_monitor_trap,
5539 	[EXIT_REASON_MONITOR_INSTRUCTION]     = handle_monitor,
5540 	[EXIT_REASON_INVEPT]                  = handle_vmx_instruction,
5541 	[EXIT_REASON_INVVPID]                 = handle_vmx_instruction,
5542 	[EXIT_REASON_RDRAND]                  = handle_invalid_op,
5543 	[EXIT_REASON_RDSEED]                  = handle_invalid_op,
5544 	[EXIT_REASON_XSAVES]                  = handle_xsaves,
5545 	[EXIT_REASON_XRSTORS]                 = handle_xrstors,
5546 	[EXIT_REASON_PML_FULL]		      = handle_pml_full,
5547 	[EXIT_REASON_INVPCID]                 = handle_invpcid,
5548 	[EXIT_REASON_VMFUNC]		      = handle_vmx_instruction,
5549 	[EXIT_REASON_PREEMPTION_TIMER]	      = handle_preemption_timer,
5550 	[EXIT_REASON_ENCLS]		      = handle_encls,
5551 };
5552 
5553 static const int kvm_vmx_max_exit_handlers =
5554 	ARRAY_SIZE(kvm_vmx_exit_handlers);
5555 
5556 static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
5557 {
5558 	*info1 = vmcs_readl(EXIT_QUALIFICATION);
5559 	*info2 = vmcs_read32(VM_EXIT_INTR_INFO);
5560 }
5561 
5562 static void vmx_destroy_pml_buffer(struct vcpu_vmx *vmx)
5563 {
5564 	if (vmx->pml_pg) {
5565 		__free_page(vmx->pml_pg);
5566 		vmx->pml_pg = NULL;
5567 	}
5568 }
5569 
5570 static void vmx_flush_pml_buffer(struct kvm_vcpu *vcpu)
5571 {
5572 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5573 	u64 *pml_buf;
5574 	u16 pml_idx;
5575 
5576 	pml_idx = vmcs_read16(GUEST_PML_INDEX);
5577 
5578 	/* Do nothing if PML buffer is empty */
5579 	if (pml_idx == (PML_ENTITY_NUM - 1))
5580 		return;
5581 
5582 	/* PML index always points to next available PML buffer entity */
5583 	if (pml_idx >= PML_ENTITY_NUM)
5584 		pml_idx = 0;
5585 	else
5586 		pml_idx++;
5587 
5588 	pml_buf = page_address(vmx->pml_pg);
5589 	for (; pml_idx < PML_ENTITY_NUM; pml_idx++) {
5590 		u64 gpa;
5591 
5592 		gpa = pml_buf[pml_idx];
5593 		WARN_ON(gpa & (PAGE_SIZE - 1));
5594 		kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
5595 	}
5596 
5597 	/* reset PML index */
5598 	vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
5599 }
5600 
5601 /*
5602  * Flush all vcpus' PML buffer and update logged GPAs to dirty_bitmap.
5603  * Called before reporting dirty_bitmap to userspace.
5604  */
5605 static void kvm_flush_pml_buffers(struct kvm *kvm)
5606 {
5607 	int i;
5608 	struct kvm_vcpu *vcpu;
5609 	/*
5610 	 * We only need to kick vcpu out of guest mode here, as PML buffer
5611 	 * is flushed at beginning of all VMEXITs, and it's obvious that only
5612 	 * vcpus running in guest are possible to have unflushed GPAs in PML
5613 	 * buffer.
5614 	 */
5615 	kvm_for_each_vcpu(i, vcpu, kvm)
5616 		kvm_vcpu_kick(vcpu);
5617 }
5618 
5619 static void vmx_dump_sel(char *name, uint32_t sel)
5620 {
5621 	pr_err("%s sel=0x%04x, attr=0x%05x, limit=0x%08x, base=0x%016lx\n",
5622 	       name, vmcs_read16(sel),
5623 	       vmcs_read32(sel + GUEST_ES_AR_BYTES - GUEST_ES_SELECTOR),
5624 	       vmcs_read32(sel + GUEST_ES_LIMIT - GUEST_ES_SELECTOR),
5625 	       vmcs_readl(sel + GUEST_ES_BASE - GUEST_ES_SELECTOR));
5626 }
5627 
5628 static void vmx_dump_dtsel(char *name, uint32_t limit)
5629 {
5630 	pr_err("%s                           limit=0x%08x, base=0x%016lx\n",
5631 	       name, vmcs_read32(limit),
5632 	       vmcs_readl(limit + GUEST_GDTR_BASE - GUEST_GDTR_LIMIT));
5633 }
5634 
5635 void dump_vmcs(void)
5636 {
5637 	u32 vmentry_ctl, vmexit_ctl;
5638 	u32 cpu_based_exec_ctrl, pin_based_exec_ctrl, secondary_exec_control;
5639 	unsigned long cr4;
5640 	u64 efer;
5641 	int i, n;
5642 
5643 	if (!dump_invalid_vmcs) {
5644 		pr_warn_ratelimited("set kvm_intel.dump_invalid_vmcs=1 to dump internal KVM state.\n");
5645 		return;
5646 	}
5647 
5648 	vmentry_ctl = vmcs_read32(VM_ENTRY_CONTROLS);
5649 	vmexit_ctl = vmcs_read32(VM_EXIT_CONTROLS);
5650 	cpu_based_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5651 	pin_based_exec_ctrl = vmcs_read32(PIN_BASED_VM_EXEC_CONTROL);
5652 	cr4 = vmcs_readl(GUEST_CR4);
5653 	efer = vmcs_read64(GUEST_IA32_EFER);
5654 	secondary_exec_control = 0;
5655 	if (cpu_has_secondary_exec_ctrls())
5656 		secondary_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
5657 
5658 	pr_err("*** Guest State ***\n");
5659 	pr_err("CR0: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
5660 	       vmcs_readl(GUEST_CR0), vmcs_readl(CR0_READ_SHADOW),
5661 	       vmcs_readl(CR0_GUEST_HOST_MASK));
5662 	pr_err("CR4: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
5663 	       cr4, vmcs_readl(CR4_READ_SHADOW), vmcs_readl(CR4_GUEST_HOST_MASK));
5664 	pr_err("CR3 = 0x%016lx\n", vmcs_readl(GUEST_CR3));
5665 	if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT) &&
5666 	    (cr4 & X86_CR4_PAE) && !(efer & EFER_LMA))
5667 	{
5668 		pr_err("PDPTR0 = 0x%016llx  PDPTR1 = 0x%016llx\n",
5669 		       vmcs_read64(GUEST_PDPTR0), vmcs_read64(GUEST_PDPTR1));
5670 		pr_err("PDPTR2 = 0x%016llx  PDPTR3 = 0x%016llx\n",
5671 		       vmcs_read64(GUEST_PDPTR2), vmcs_read64(GUEST_PDPTR3));
5672 	}
5673 	pr_err("RSP = 0x%016lx  RIP = 0x%016lx\n",
5674 	       vmcs_readl(GUEST_RSP), vmcs_readl(GUEST_RIP));
5675 	pr_err("RFLAGS=0x%08lx         DR7 = 0x%016lx\n",
5676 	       vmcs_readl(GUEST_RFLAGS), vmcs_readl(GUEST_DR7));
5677 	pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
5678 	       vmcs_readl(GUEST_SYSENTER_ESP),
5679 	       vmcs_read32(GUEST_SYSENTER_CS), vmcs_readl(GUEST_SYSENTER_EIP));
5680 	vmx_dump_sel("CS:  ", GUEST_CS_SELECTOR);
5681 	vmx_dump_sel("DS:  ", GUEST_DS_SELECTOR);
5682 	vmx_dump_sel("SS:  ", GUEST_SS_SELECTOR);
5683 	vmx_dump_sel("ES:  ", GUEST_ES_SELECTOR);
5684 	vmx_dump_sel("FS:  ", GUEST_FS_SELECTOR);
5685 	vmx_dump_sel("GS:  ", GUEST_GS_SELECTOR);
5686 	vmx_dump_dtsel("GDTR:", GUEST_GDTR_LIMIT);
5687 	vmx_dump_sel("LDTR:", GUEST_LDTR_SELECTOR);
5688 	vmx_dump_dtsel("IDTR:", GUEST_IDTR_LIMIT);
5689 	vmx_dump_sel("TR:  ", GUEST_TR_SELECTOR);
5690 	if ((vmexit_ctl & (VM_EXIT_SAVE_IA32_PAT | VM_EXIT_SAVE_IA32_EFER)) ||
5691 	    (vmentry_ctl & (VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_IA32_EFER)))
5692 		pr_err("EFER =     0x%016llx  PAT = 0x%016llx\n",
5693 		       efer, vmcs_read64(GUEST_IA32_PAT));
5694 	pr_err("DebugCtl = 0x%016llx  DebugExceptions = 0x%016lx\n",
5695 	       vmcs_read64(GUEST_IA32_DEBUGCTL),
5696 	       vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS));
5697 	if (cpu_has_load_perf_global_ctrl() &&
5698 	    vmentry_ctl & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
5699 		pr_err("PerfGlobCtl = 0x%016llx\n",
5700 		       vmcs_read64(GUEST_IA32_PERF_GLOBAL_CTRL));
5701 	if (vmentry_ctl & VM_ENTRY_LOAD_BNDCFGS)
5702 		pr_err("BndCfgS = 0x%016llx\n", vmcs_read64(GUEST_BNDCFGS));
5703 	pr_err("Interruptibility = %08x  ActivityState = %08x\n",
5704 	       vmcs_read32(GUEST_INTERRUPTIBILITY_INFO),
5705 	       vmcs_read32(GUEST_ACTIVITY_STATE));
5706 	if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)
5707 		pr_err("InterruptStatus = %04x\n",
5708 		       vmcs_read16(GUEST_INTR_STATUS));
5709 
5710 	pr_err("*** Host State ***\n");
5711 	pr_err("RIP = 0x%016lx  RSP = 0x%016lx\n",
5712 	       vmcs_readl(HOST_RIP), vmcs_readl(HOST_RSP));
5713 	pr_err("CS=%04x SS=%04x DS=%04x ES=%04x FS=%04x GS=%04x TR=%04x\n",
5714 	       vmcs_read16(HOST_CS_SELECTOR), vmcs_read16(HOST_SS_SELECTOR),
5715 	       vmcs_read16(HOST_DS_SELECTOR), vmcs_read16(HOST_ES_SELECTOR),
5716 	       vmcs_read16(HOST_FS_SELECTOR), vmcs_read16(HOST_GS_SELECTOR),
5717 	       vmcs_read16(HOST_TR_SELECTOR));
5718 	pr_err("FSBase=%016lx GSBase=%016lx TRBase=%016lx\n",
5719 	       vmcs_readl(HOST_FS_BASE), vmcs_readl(HOST_GS_BASE),
5720 	       vmcs_readl(HOST_TR_BASE));
5721 	pr_err("GDTBase=%016lx IDTBase=%016lx\n",
5722 	       vmcs_readl(HOST_GDTR_BASE), vmcs_readl(HOST_IDTR_BASE));
5723 	pr_err("CR0=%016lx CR3=%016lx CR4=%016lx\n",
5724 	       vmcs_readl(HOST_CR0), vmcs_readl(HOST_CR3),
5725 	       vmcs_readl(HOST_CR4));
5726 	pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
5727 	       vmcs_readl(HOST_IA32_SYSENTER_ESP),
5728 	       vmcs_read32(HOST_IA32_SYSENTER_CS),
5729 	       vmcs_readl(HOST_IA32_SYSENTER_EIP));
5730 	if (vmexit_ctl & (VM_EXIT_LOAD_IA32_PAT | VM_EXIT_LOAD_IA32_EFER))
5731 		pr_err("EFER = 0x%016llx  PAT = 0x%016llx\n",
5732 		       vmcs_read64(HOST_IA32_EFER),
5733 		       vmcs_read64(HOST_IA32_PAT));
5734 	if (cpu_has_load_perf_global_ctrl() &&
5735 	    vmexit_ctl & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
5736 		pr_err("PerfGlobCtl = 0x%016llx\n",
5737 		       vmcs_read64(HOST_IA32_PERF_GLOBAL_CTRL));
5738 
5739 	pr_err("*** Control State ***\n");
5740 	pr_err("PinBased=%08x CPUBased=%08x SecondaryExec=%08x\n",
5741 	       pin_based_exec_ctrl, cpu_based_exec_ctrl, secondary_exec_control);
5742 	pr_err("EntryControls=%08x ExitControls=%08x\n", vmentry_ctl, vmexit_ctl);
5743 	pr_err("ExceptionBitmap=%08x PFECmask=%08x PFECmatch=%08x\n",
5744 	       vmcs_read32(EXCEPTION_BITMAP),
5745 	       vmcs_read32(PAGE_FAULT_ERROR_CODE_MASK),
5746 	       vmcs_read32(PAGE_FAULT_ERROR_CODE_MATCH));
5747 	pr_err("VMEntry: intr_info=%08x errcode=%08x ilen=%08x\n",
5748 	       vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
5749 	       vmcs_read32(VM_ENTRY_EXCEPTION_ERROR_CODE),
5750 	       vmcs_read32(VM_ENTRY_INSTRUCTION_LEN));
5751 	pr_err("VMExit: intr_info=%08x errcode=%08x ilen=%08x\n",
5752 	       vmcs_read32(VM_EXIT_INTR_INFO),
5753 	       vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
5754 	       vmcs_read32(VM_EXIT_INSTRUCTION_LEN));
5755 	pr_err("        reason=%08x qualification=%016lx\n",
5756 	       vmcs_read32(VM_EXIT_REASON), vmcs_readl(EXIT_QUALIFICATION));
5757 	pr_err("IDTVectoring: info=%08x errcode=%08x\n",
5758 	       vmcs_read32(IDT_VECTORING_INFO_FIELD),
5759 	       vmcs_read32(IDT_VECTORING_ERROR_CODE));
5760 	pr_err("TSC Offset = 0x%016llx\n", vmcs_read64(TSC_OFFSET));
5761 	if (secondary_exec_control & SECONDARY_EXEC_TSC_SCALING)
5762 		pr_err("TSC Multiplier = 0x%016llx\n",
5763 		       vmcs_read64(TSC_MULTIPLIER));
5764 	if (cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW) {
5765 		if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY) {
5766 			u16 status = vmcs_read16(GUEST_INTR_STATUS);
5767 			pr_err("SVI|RVI = %02x|%02x ", status >> 8, status & 0xff);
5768 		}
5769 		pr_cont("TPR Threshold = 0x%02x\n", vmcs_read32(TPR_THRESHOLD));
5770 		if (secondary_exec_control & SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)
5771 			pr_err("APIC-access addr = 0x%016llx ", vmcs_read64(APIC_ACCESS_ADDR));
5772 		pr_cont("virt-APIC addr = 0x%016llx\n", vmcs_read64(VIRTUAL_APIC_PAGE_ADDR));
5773 	}
5774 	if (pin_based_exec_ctrl & PIN_BASED_POSTED_INTR)
5775 		pr_err("PostedIntrVec = 0x%02x\n", vmcs_read16(POSTED_INTR_NV));
5776 	if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT))
5777 		pr_err("EPT pointer = 0x%016llx\n", vmcs_read64(EPT_POINTER));
5778 	n = vmcs_read32(CR3_TARGET_COUNT);
5779 	for (i = 0; i + 1 < n; i += 4)
5780 		pr_err("CR3 target%u=%016lx target%u=%016lx\n",
5781 		       i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2),
5782 		       i + 1, vmcs_readl(CR3_TARGET_VALUE0 + i * 2 + 2));
5783 	if (i < n)
5784 		pr_err("CR3 target%u=%016lx\n",
5785 		       i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2));
5786 	if (secondary_exec_control & SECONDARY_EXEC_PAUSE_LOOP_EXITING)
5787 		pr_err("PLE Gap=%08x Window=%08x\n",
5788 		       vmcs_read32(PLE_GAP), vmcs_read32(PLE_WINDOW));
5789 	if (secondary_exec_control & SECONDARY_EXEC_ENABLE_VPID)
5790 		pr_err("Virtual processor ID = 0x%04x\n",
5791 		       vmcs_read16(VIRTUAL_PROCESSOR_ID));
5792 }
5793 
5794 /*
5795  * The guest has exited.  See if we can fix it or if we need userspace
5796  * assistance.
5797  */
5798 static int vmx_handle_exit(struct kvm_vcpu *vcpu)
5799 {
5800 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5801 	u32 exit_reason = vmx->exit_reason;
5802 	u32 vectoring_info = vmx->idt_vectoring_info;
5803 
5804 	trace_kvm_exit(exit_reason, vcpu, KVM_ISA_VMX);
5805 
5806 	/*
5807 	 * Flush logged GPAs PML buffer, this will make dirty_bitmap more
5808 	 * updated. Another good is, in kvm_vm_ioctl_get_dirty_log, before
5809 	 * querying dirty_bitmap, we only need to kick all vcpus out of guest
5810 	 * mode as if vcpus is in root mode, the PML buffer must has been
5811 	 * flushed already.
5812 	 */
5813 	if (enable_pml)
5814 		vmx_flush_pml_buffer(vcpu);
5815 
5816 	/* If guest state is invalid, start emulating */
5817 	if (vmx->emulation_required)
5818 		return handle_invalid_guest_state(vcpu);
5819 
5820 	if (is_guest_mode(vcpu) && nested_vmx_exit_reflected(vcpu, exit_reason))
5821 		return nested_vmx_reflect_vmexit(vcpu, exit_reason);
5822 
5823 	if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) {
5824 		dump_vmcs();
5825 		vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
5826 		vcpu->run->fail_entry.hardware_entry_failure_reason
5827 			= exit_reason;
5828 		return 0;
5829 	}
5830 
5831 	if (unlikely(vmx->fail)) {
5832 		dump_vmcs();
5833 		vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
5834 		vcpu->run->fail_entry.hardware_entry_failure_reason
5835 			= vmcs_read32(VM_INSTRUCTION_ERROR);
5836 		return 0;
5837 	}
5838 
5839 	/*
5840 	 * Note:
5841 	 * Do not try to fix EXIT_REASON_EPT_MISCONFIG if it caused by
5842 	 * delivery event since it indicates guest is accessing MMIO.
5843 	 * The vm-exit can be triggered again after return to guest that
5844 	 * will cause infinite loop.
5845 	 */
5846 	if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
5847 			(exit_reason != EXIT_REASON_EXCEPTION_NMI &&
5848 			exit_reason != EXIT_REASON_EPT_VIOLATION &&
5849 			exit_reason != EXIT_REASON_PML_FULL &&
5850 			exit_reason != EXIT_REASON_TASK_SWITCH)) {
5851 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5852 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_DELIVERY_EV;
5853 		vcpu->run->internal.ndata = 3;
5854 		vcpu->run->internal.data[0] = vectoring_info;
5855 		vcpu->run->internal.data[1] = exit_reason;
5856 		vcpu->run->internal.data[2] = vcpu->arch.exit_qualification;
5857 		if (exit_reason == EXIT_REASON_EPT_MISCONFIG) {
5858 			vcpu->run->internal.ndata++;
5859 			vcpu->run->internal.data[3] =
5860 				vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5861 		}
5862 		return 0;
5863 	}
5864 
5865 	if (unlikely(!enable_vnmi &&
5866 		     vmx->loaded_vmcs->soft_vnmi_blocked)) {
5867 		if (vmx_interrupt_allowed(vcpu)) {
5868 			vmx->loaded_vmcs->soft_vnmi_blocked = 0;
5869 		} else if (vmx->loaded_vmcs->vnmi_blocked_time > 1000000000LL &&
5870 			   vcpu->arch.nmi_pending) {
5871 			/*
5872 			 * This CPU don't support us in finding the end of an
5873 			 * NMI-blocked window if the guest runs with IRQs
5874 			 * disabled. So we pull the trigger after 1 s of
5875 			 * futile waiting, but inform the user about this.
5876 			 */
5877 			printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
5878 			       "state on VCPU %d after 1 s timeout\n",
5879 			       __func__, vcpu->vcpu_id);
5880 			vmx->loaded_vmcs->soft_vnmi_blocked = 0;
5881 		}
5882 	}
5883 
5884 	if (exit_reason < kvm_vmx_max_exit_handlers
5885 	    && kvm_vmx_exit_handlers[exit_reason])
5886 		return kvm_vmx_exit_handlers[exit_reason](vcpu);
5887 	else {
5888 		vcpu_unimpl(vcpu, "vmx: unexpected exit reason 0x%x\n",
5889 				exit_reason);
5890 		kvm_queue_exception(vcpu, UD_VECTOR);
5891 		return 1;
5892 	}
5893 }
5894 
5895 /*
5896  * Software based L1D cache flush which is used when microcode providing
5897  * the cache control MSR is not loaded.
5898  *
5899  * The L1D cache is 32 KiB on Nehalem and later microarchitectures, but to
5900  * flush it is required to read in 64 KiB because the replacement algorithm
5901  * is not exactly LRU. This could be sized at runtime via topology
5902  * information but as all relevant affected CPUs have 32KiB L1D cache size
5903  * there is no point in doing so.
5904  */
5905 static void vmx_l1d_flush(struct kvm_vcpu *vcpu)
5906 {
5907 	int size = PAGE_SIZE << L1D_CACHE_ORDER;
5908 
5909 	/*
5910 	 * This code is only executed when the the flush mode is 'cond' or
5911 	 * 'always'
5912 	 */
5913 	if (static_branch_likely(&vmx_l1d_flush_cond)) {
5914 		bool flush_l1d;
5915 
5916 		/*
5917 		 * Clear the per-vcpu flush bit, it gets set again
5918 		 * either from vcpu_run() or from one of the unsafe
5919 		 * VMEXIT handlers.
5920 		 */
5921 		flush_l1d = vcpu->arch.l1tf_flush_l1d;
5922 		vcpu->arch.l1tf_flush_l1d = false;
5923 
5924 		/*
5925 		 * Clear the per-cpu flush bit, it gets set again from
5926 		 * the interrupt handlers.
5927 		 */
5928 		flush_l1d |= kvm_get_cpu_l1tf_flush_l1d();
5929 		kvm_clear_cpu_l1tf_flush_l1d();
5930 
5931 		if (!flush_l1d)
5932 			return;
5933 	}
5934 
5935 	vcpu->stat.l1d_flush++;
5936 
5937 	if (static_cpu_has(X86_FEATURE_FLUSH_L1D)) {
5938 		wrmsrl(MSR_IA32_FLUSH_CMD, L1D_FLUSH);
5939 		return;
5940 	}
5941 
5942 	asm volatile(
5943 		/* First ensure the pages are in the TLB */
5944 		"xorl	%%eax, %%eax\n"
5945 		".Lpopulate_tlb:\n\t"
5946 		"movzbl	(%[flush_pages], %%" _ASM_AX "), %%ecx\n\t"
5947 		"addl	$4096, %%eax\n\t"
5948 		"cmpl	%%eax, %[size]\n\t"
5949 		"jne	.Lpopulate_tlb\n\t"
5950 		"xorl	%%eax, %%eax\n\t"
5951 		"cpuid\n\t"
5952 		/* Now fill the cache */
5953 		"xorl	%%eax, %%eax\n"
5954 		".Lfill_cache:\n"
5955 		"movzbl	(%[flush_pages], %%" _ASM_AX "), %%ecx\n\t"
5956 		"addl	$64, %%eax\n\t"
5957 		"cmpl	%%eax, %[size]\n\t"
5958 		"jne	.Lfill_cache\n\t"
5959 		"lfence\n"
5960 		:: [flush_pages] "r" (vmx_l1d_flush_pages),
5961 		    [size] "r" (size)
5962 		: "eax", "ebx", "ecx", "edx");
5963 }
5964 
5965 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
5966 {
5967 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5968 
5969 	if (is_guest_mode(vcpu) &&
5970 		nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
5971 		return;
5972 
5973 	if (irr == -1 || tpr < irr) {
5974 		vmcs_write32(TPR_THRESHOLD, 0);
5975 		return;
5976 	}
5977 
5978 	vmcs_write32(TPR_THRESHOLD, irr);
5979 }
5980 
5981 void vmx_set_virtual_apic_mode(struct kvm_vcpu *vcpu)
5982 {
5983 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5984 	u32 sec_exec_control;
5985 
5986 	if (!lapic_in_kernel(vcpu))
5987 		return;
5988 
5989 	if (!flexpriority_enabled &&
5990 	    !cpu_has_vmx_virtualize_x2apic_mode())
5991 		return;
5992 
5993 	/* Postpone execution until vmcs01 is the current VMCS. */
5994 	if (is_guest_mode(vcpu)) {
5995 		vmx->nested.change_vmcs01_virtual_apic_mode = true;
5996 		return;
5997 	}
5998 
5999 	sec_exec_control = secondary_exec_controls_get(vmx);
6000 	sec_exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
6001 			      SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE);
6002 
6003 	switch (kvm_get_apic_mode(vcpu)) {
6004 	case LAPIC_MODE_INVALID:
6005 		WARN_ONCE(true, "Invalid local APIC state");
6006 	case LAPIC_MODE_DISABLED:
6007 		break;
6008 	case LAPIC_MODE_XAPIC:
6009 		if (flexpriority_enabled) {
6010 			sec_exec_control |=
6011 				SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
6012 			vmx_flush_tlb(vcpu, true);
6013 		}
6014 		break;
6015 	case LAPIC_MODE_X2APIC:
6016 		if (cpu_has_vmx_virtualize_x2apic_mode())
6017 			sec_exec_control |=
6018 				SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
6019 		break;
6020 	}
6021 	secondary_exec_controls_set(vmx, sec_exec_control);
6022 
6023 	vmx_update_msr_bitmap(vcpu);
6024 }
6025 
6026 static void vmx_set_apic_access_page_addr(struct kvm_vcpu *vcpu, hpa_t hpa)
6027 {
6028 	if (!is_guest_mode(vcpu)) {
6029 		vmcs_write64(APIC_ACCESS_ADDR, hpa);
6030 		vmx_flush_tlb(vcpu, true);
6031 	}
6032 }
6033 
6034 static void vmx_hwapic_isr_update(struct kvm_vcpu *vcpu, int max_isr)
6035 {
6036 	u16 status;
6037 	u8 old;
6038 
6039 	if (max_isr == -1)
6040 		max_isr = 0;
6041 
6042 	status = vmcs_read16(GUEST_INTR_STATUS);
6043 	old = status >> 8;
6044 	if (max_isr != old) {
6045 		status &= 0xff;
6046 		status |= max_isr << 8;
6047 		vmcs_write16(GUEST_INTR_STATUS, status);
6048 	}
6049 }
6050 
6051 static void vmx_set_rvi(int vector)
6052 {
6053 	u16 status;
6054 	u8 old;
6055 
6056 	if (vector == -1)
6057 		vector = 0;
6058 
6059 	status = vmcs_read16(GUEST_INTR_STATUS);
6060 	old = (u8)status & 0xff;
6061 	if ((u8)vector != old) {
6062 		status &= ~0xff;
6063 		status |= (u8)vector;
6064 		vmcs_write16(GUEST_INTR_STATUS, status);
6065 	}
6066 }
6067 
6068 static void vmx_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr)
6069 {
6070 	/*
6071 	 * When running L2, updating RVI is only relevant when
6072 	 * vmcs12 virtual-interrupt-delivery enabled.
6073 	 * However, it can be enabled only when L1 also
6074 	 * intercepts external-interrupts and in that case
6075 	 * we should not update vmcs02 RVI but instead intercept
6076 	 * interrupt. Therefore, do nothing when running L2.
6077 	 */
6078 	if (!is_guest_mode(vcpu))
6079 		vmx_set_rvi(max_irr);
6080 }
6081 
6082 static int vmx_sync_pir_to_irr(struct kvm_vcpu *vcpu)
6083 {
6084 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6085 	int max_irr;
6086 	bool max_irr_updated;
6087 
6088 	WARN_ON(!vcpu->arch.apicv_active);
6089 	if (pi_test_on(&vmx->pi_desc)) {
6090 		pi_clear_on(&vmx->pi_desc);
6091 		/*
6092 		 * IOMMU can write to PIR.ON, so the barrier matters even on UP.
6093 		 * But on x86 this is just a compiler barrier anyway.
6094 		 */
6095 		smp_mb__after_atomic();
6096 		max_irr_updated =
6097 			kvm_apic_update_irr(vcpu, vmx->pi_desc.pir, &max_irr);
6098 
6099 		/*
6100 		 * If we are running L2 and L1 has a new pending interrupt
6101 		 * which can be injected, we should re-evaluate
6102 		 * what should be done with this new L1 interrupt.
6103 		 * If L1 intercepts external-interrupts, we should
6104 		 * exit from L2 to L1. Otherwise, interrupt should be
6105 		 * delivered directly to L2.
6106 		 */
6107 		if (is_guest_mode(vcpu) && max_irr_updated) {
6108 			if (nested_exit_on_intr(vcpu))
6109 				kvm_vcpu_exiting_guest_mode(vcpu);
6110 			else
6111 				kvm_make_request(KVM_REQ_EVENT, vcpu);
6112 		}
6113 	} else {
6114 		max_irr = kvm_lapic_find_highest_irr(vcpu);
6115 	}
6116 	vmx_hwapic_irr_update(vcpu, max_irr);
6117 	return max_irr;
6118 }
6119 
6120 static void vmx_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap)
6121 {
6122 	if (!kvm_vcpu_apicv_active(vcpu))
6123 		return;
6124 
6125 	vmcs_write64(EOI_EXIT_BITMAP0, eoi_exit_bitmap[0]);
6126 	vmcs_write64(EOI_EXIT_BITMAP1, eoi_exit_bitmap[1]);
6127 	vmcs_write64(EOI_EXIT_BITMAP2, eoi_exit_bitmap[2]);
6128 	vmcs_write64(EOI_EXIT_BITMAP3, eoi_exit_bitmap[3]);
6129 }
6130 
6131 static void vmx_apicv_post_state_restore(struct kvm_vcpu *vcpu)
6132 {
6133 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6134 
6135 	pi_clear_on(&vmx->pi_desc);
6136 	memset(vmx->pi_desc.pir, 0, sizeof(vmx->pi_desc.pir));
6137 }
6138 
6139 static void handle_exception_nmi_irqoff(struct vcpu_vmx *vmx)
6140 {
6141 	vmx->exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6142 
6143 	/* if exit due to PF check for async PF */
6144 	if (is_page_fault(vmx->exit_intr_info))
6145 		vmx->vcpu.arch.apf.host_apf_reason = kvm_read_and_reset_pf_reason();
6146 
6147 	/* Handle machine checks before interrupts are enabled */
6148 	if (is_machine_check(vmx->exit_intr_info))
6149 		kvm_machine_check();
6150 
6151 	/* We need to handle NMIs before interrupts are enabled */
6152 	if (is_nmi(vmx->exit_intr_info)) {
6153 		kvm_before_interrupt(&vmx->vcpu);
6154 		asm("int $2");
6155 		kvm_after_interrupt(&vmx->vcpu);
6156 	}
6157 }
6158 
6159 static void handle_external_interrupt_irqoff(struct kvm_vcpu *vcpu)
6160 {
6161 	unsigned int vector;
6162 	unsigned long entry;
6163 #ifdef CONFIG_X86_64
6164 	unsigned long tmp;
6165 #endif
6166 	gate_desc *desc;
6167 	u32 intr_info;
6168 
6169 	intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6170 	if (WARN_ONCE(!is_external_intr(intr_info),
6171 	    "KVM: unexpected VM-Exit interrupt info: 0x%x", intr_info))
6172 		return;
6173 
6174 	vector = intr_info & INTR_INFO_VECTOR_MASK;
6175 	desc = (gate_desc *)host_idt_base + vector;
6176 	entry = gate_offset(desc);
6177 
6178 	kvm_before_interrupt(vcpu);
6179 
6180 	asm volatile(
6181 #ifdef CONFIG_X86_64
6182 		"mov %%" _ASM_SP ", %[sp]\n\t"
6183 		"and $0xfffffffffffffff0, %%" _ASM_SP "\n\t"
6184 		"push $%c[ss]\n\t"
6185 		"push %[sp]\n\t"
6186 #endif
6187 		"pushf\n\t"
6188 		__ASM_SIZE(push) " $%c[cs]\n\t"
6189 		CALL_NOSPEC
6190 		:
6191 #ifdef CONFIG_X86_64
6192 		[sp]"=&r"(tmp),
6193 #endif
6194 		ASM_CALL_CONSTRAINT
6195 		:
6196 		THUNK_TARGET(entry),
6197 		[ss]"i"(__KERNEL_DS),
6198 		[cs]"i"(__KERNEL_CS)
6199 	);
6200 
6201 	kvm_after_interrupt(vcpu);
6202 }
6203 STACK_FRAME_NON_STANDARD(handle_external_interrupt_irqoff);
6204 
6205 static void vmx_handle_exit_irqoff(struct kvm_vcpu *vcpu)
6206 {
6207 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6208 
6209 	if (vmx->exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT)
6210 		handle_external_interrupt_irqoff(vcpu);
6211 	else if (vmx->exit_reason == EXIT_REASON_EXCEPTION_NMI)
6212 		handle_exception_nmi_irqoff(vmx);
6213 }
6214 
6215 static bool vmx_has_emulated_msr(int index)
6216 {
6217 	switch (index) {
6218 	case MSR_IA32_SMBASE:
6219 		/*
6220 		 * We cannot do SMM unless we can run the guest in big
6221 		 * real mode.
6222 		 */
6223 		return enable_unrestricted_guest || emulate_invalid_guest_state;
6224 	case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
6225 		return nested;
6226 	case MSR_AMD64_VIRT_SPEC_CTRL:
6227 		/* This is AMD only.  */
6228 		return false;
6229 	default:
6230 		return true;
6231 	}
6232 }
6233 
6234 static bool vmx_pt_supported(void)
6235 {
6236 	return pt_mode == PT_MODE_HOST_GUEST;
6237 }
6238 
6239 static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx)
6240 {
6241 	u32 exit_intr_info;
6242 	bool unblock_nmi;
6243 	u8 vector;
6244 	bool idtv_info_valid;
6245 
6246 	idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK;
6247 
6248 	if (enable_vnmi) {
6249 		if (vmx->loaded_vmcs->nmi_known_unmasked)
6250 			return;
6251 		/*
6252 		 * Can't use vmx->exit_intr_info since we're not sure what
6253 		 * the exit reason is.
6254 		 */
6255 		exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6256 		unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
6257 		vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
6258 		/*
6259 		 * SDM 3: 27.7.1.2 (September 2008)
6260 		 * Re-set bit "block by NMI" before VM entry if vmexit caused by
6261 		 * a guest IRET fault.
6262 		 * SDM 3: 23.2.2 (September 2008)
6263 		 * Bit 12 is undefined in any of the following cases:
6264 		 *  If the VM exit sets the valid bit in the IDT-vectoring
6265 		 *   information field.
6266 		 *  If the VM exit is due to a double fault.
6267 		 */
6268 		if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
6269 		    vector != DF_VECTOR && !idtv_info_valid)
6270 			vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
6271 				      GUEST_INTR_STATE_NMI);
6272 		else
6273 			vmx->loaded_vmcs->nmi_known_unmasked =
6274 				!(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO)
6275 				  & GUEST_INTR_STATE_NMI);
6276 	} else if (unlikely(vmx->loaded_vmcs->soft_vnmi_blocked))
6277 		vmx->loaded_vmcs->vnmi_blocked_time +=
6278 			ktime_to_ns(ktime_sub(ktime_get(),
6279 					      vmx->loaded_vmcs->entry_time));
6280 }
6281 
6282 static void __vmx_complete_interrupts(struct kvm_vcpu *vcpu,
6283 				      u32 idt_vectoring_info,
6284 				      int instr_len_field,
6285 				      int error_code_field)
6286 {
6287 	u8 vector;
6288 	int type;
6289 	bool idtv_info_valid;
6290 
6291 	idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
6292 
6293 	vcpu->arch.nmi_injected = false;
6294 	kvm_clear_exception_queue(vcpu);
6295 	kvm_clear_interrupt_queue(vcpu);
6296 
6297 	if (!idtv_info_valid)
6298 		return;
6299 
6300 	kvm_make_request(KVM_REQ_EVENT, vcpu);
6301 
6302 	vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
6303 	type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
6304 
6305 	switch (type) {
6306 	case INTR_TYPE_NMI_INTR:
6307 		vcpu->arch.nmi_injected = true;
6308 		/*
6309 		 * SDM 3: 27.7.1.2 (September 2008)
6310 		 * Clear bit "block by NMI" before VM entry if a NMI
6311 		 * delivery faulted.
6312 		 */
6313 		vmx_set_nmi_mask(vcpu, false);
6314 		break;
6315 	case INTR_TYPE_SOFT_EXCEPTION:
6316 		vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
6317 		/* fall through */
6318 	case INTR_TYPE_HARD_EXCEPTION:
6319 		if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
6320 			u32 err = vmcs_read32(error_code_field);
6321 			kvm_requeue_exception_e(vcpu, vector, err);
6322 		} else
6323 			kvm_requeue_exception(vcpu, vector);
6324 		break;
6325 	case INTR_TYPE_SOFT_INTR:
6326 		vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
6327 		/* fall through */
6328 	case INTR_TYPE_EXT_INTR:
6329 		kvm_queue_interrupt(vcpu, vector, type == INTR_TYPE_SOFT_INTR);
6330 		break;
6331 	default:
6332 		break;
6333 	}
6334 }
6335 
6336 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
6337 {
6338 	__vmx_complete_interrupts(&vmx->vcpu, vmx->idt_vectoring_info,
6339 				  VM_EXIT_INSTRUCTION_LEN,
6340 				  IDT_VECTORING_ERROR_CODE);
6341 }
6342 
6343 static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
6344 {
6345 	__vmx_complete_interrupts(vcpu,
6346 				  vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
6347 				  VM_ENTRY_INSTRUCTION_LEN,
6348 				  VM_ENTRY_EXCEPTION_ERROR_CODE);
6349 
6350 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
6351 }
6352 
6353 static void atomic_switch_perf_msrs(struct vcpu_vmx *vmx)
6354 {
6355 	int i, nr_msrs;
6356 	struct perf_guest_switch_msr *msrs;
6357 
6358 	msrs = perf_guest_get_msrs(&nr_msrs);
6359 
6360 	if (!msrs)
6361 		return;
6362 
6363 	for (i = 0; i < nr_msrs; i++)
6364 		if (msrs[i].host == msrs[i].guest)
6365 			clear_atomic_switch_msr(vmx, msrs[i].msr);
6366 		else
6367 			add_atomic_switch_msr(vmx, msrs[i].msr, msrs[i].guest,
6368 					msrs[i].host, false);
6369 }
6370 
6371 static void vmx_update_hv_timer(struct kvm_vcpu *vcpu)
6372 {
6373 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6374 	u64 tscl;
6375 	u32 delta_tsc;
6376 
6377 	if (vmx->req_immediate_exit) {
6378 		vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, 0);
6379 		vmx->loaded_vmcs->hv_timer_soft_disabled = false;
6380 	} else if (vmx->hv_deadline_tsc != -1) {
6381 		tscl = rdtsc();
6382 		if (vmx->hv_deadline_tsc > tscl)
6383 			/* set_hv_timer ensures the delta fits in 32-bits */
6384 			delta_tsc = (u32)((vmx->hv_deadline_tsc - tscl) >>
6385 				cpu_preemption_timer_multi);
6386 		else
6387 			delta_tsc = 0;
6388 
6389 		vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, delta_tsc);
6390 		vmx->loaded_vmcs->hv_timer_soft_disabled = false;
6391 	} else if (!vmx->loaded_vmcs->hv_timer_soft_disabled) {
6392 		vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, -1);
6393 		vmx->loaded_vmcs->hv_timer_soft_disabled = true;
6394 	}
6395 }
6396 
6397 void vmx_update_host_rsp(struct vcpu_vmx *vmx, unsigned long host_rsp)
6398 {
6399 	if (unlikely(host_rsp != vmx->loaded_vmcs->host_state.rsp)) {
6400 		vmx->loaded_vmcs->host_state.rsp = host_rsp;
6401 		vmcs_writel(HOST_RSP, host_rsp);
6402 	}
6403 }
6404 
6405 bool __vmx_vcpu_run(struct vcpu_vmx *vmx, unsigned long *regs, bool launched);
6406 
6407 static void vmx_vcpu_run(struct kvm_vcpu *vcpu)
6408 {
6409 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6410 	unsigned long cr3, cr4;
6411 
6412 	/* Record the guest's net vcpu time for enforced NMI injections. */
6413 	if (unlikely(!enable_vnmi &&
6414 		     vmx->loaded_vmcs->soft_vnmi_blocked))
6415 		vmx->loaded_vmcs->entry_time = ktime_get();
6416 
6417 	/* Don't enter VMX if guest state is invalid, let the exit handler
6418 	   start emulation until we arrive back to a valid state */
6419 	if (vmx->emulation_required)
6420 		return;
6421 
6422 	if (vmx->ple_window_dirty) {
6423 		vmx->ple_window_dirty = false;
6424 		vmcs_write32(PLE_WINDOW, vmx->ple_window);
6425 	}
6426 
6427 	if (vmx->nested.need_vmcs12_to_shadow_sync)
6428 		nested_sync_vmcs12_to_shadow(vcpu);
6429 
6430 	if (test_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_dirty))
6431 		vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
6432 	if (test_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_dirty))
6433 		vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
6434 
6435 	cr3 = __get_current_cr3_fast();
6436 	if (unlikely(cr3 != vmx->loaded_vmcs->host_state.cr3)) {
6437 		vmcs_writel(HOST_CR3, cr3);
6438 		vmx->loaded_vmcs->host_state.cr3 = cr3;
6439 	}
6440 
6441 	cr4 = cr4_read_shadow();
6442 	if (unlikely(cr4 != vmx->loaded_vmcs->host_state.cr4)) {
6443 		vmcs_writel(HOST_CR4, cr4);
6444 		vmx->loaded_vmcs->host_state.cr4 = cr4;
6445 	}
6446 
6447 	/* When single-stepping over STI and MOV SS, we must clear the
6448 	 * corresponding interruptibility bits in the guest state. Otherwise
6449 	 * vmentry fails as it then expects bit 14 (BS) in pending debug
6450 	 * exceptions being set, but that's not correct for the guest debugging
6451 	 * case. */
6452 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6453 		vmx_set_interrupt_shadow(vcpu, 0);
6454 
6455 	kvm_load_guest_xcr0(vcpu);
6456 
6457 	if (static_cpu_has(X86_FEATURE_PKU) &&
6458 	    kvm_read_cr4_bits(vcpu, X86_CR4_PKE) &&
6459 	    vcpu->arch.pkru != vmx->host_pkru)
6460 		__write_pkru(vcpu->arch.pkru);
6461 
6462 	pt_guest_enter(vmx);
6463 
6464 	atomic_switch_perf_msrs(vmx);
6465 
6466 	if (enable_preemption_timer)
6467 		vmx_update_hv_timer(vcpu);
6468 
6469 	if (lapic_in_kernel(vcpu) &&
6470 		vcpu->arch.apic->lapic_timer.timer_advance_ns)
6471 		kvm_wait_lapic_expire(vcpu);
6472 
6473 	/*
6474 	 * If this vCPU has touched SPEC_CTRL, restore the guest's value if
6475 	 * it's non-zero. Since vmentry is serialising on affected CPUs, there
6476 	 * is no need to worry about the conditional branch over the wrmsr
6477 	 * being speculatively taken.
6478 	 */
6479 	x86_spec_ctrl_set_guest(vmx->spec_ctrl, 0);
6480 
6481 	/* L1D Flush includes CPU buffer clear to mitigate MDS */
6482 	if (static_branch_unlikely(&vmx_l1d_should_flush))
6483 		vmx_l1d_flush(vcpu);
6484 	else if (static_branch_unlikely(&mds_user_clear))
6485 		mds_clear_cpu_buffers();
6486 
6487 	if (vcpu->arch.cr2 != read_cr2())
6488 		write_cr2(vcpu->arch.cr2);
6489 
6490 	vmx->fail = __vmx_vcpu_run(vmx, (unsigned long *)&vcpu->arch.regs,
6491 				   vmx->loaded_vmcs->launched);
6492 
6493 	vcpu->arch.cr2 = read_cr2();
6494 
6495 	/*
6496 	 * We do not use IBRS in the kernel. If this vCPU has used the
6497 	 * SPEC_CTRL MSR it may have left it on; save the value and
6498 	 * turn it off. This is much more efficient than blindly adding
6499 	 * it to the atomic save/restore list. Especially as the former
6500 	 * (Saving guest MSRs on vmexit) doesn't even exist in KVM.
6501 	 *
6502 	 * For non-nested case:
6503 	 * If the L01 MSR bitmap does not intercept the MSR, then we need to
6504 	 * save it.
6505 	 *
6506 	 * For nested case:
6507 	 * If the L02 MSR bitmap does not intercept the MSR, then we need to
6508 	 * save it.
6509 	 */
6510 	if (unlikely(!msr_write_intercepted(vcpu, MSR_IA32_SPEC_CTRL)))
6511 		vmx->spec_ctrl = native_read_msr(MSR_IA32_SPEC_CTRL);
6512 
6513 	x86_spec_ctrl_restore_host(vmx->spec_ctrl, 0);
6514 
6515 	/* All fields are clean at this point */
6516 	if (static_branch_unlikely(&enable_evmcs))
6517 		current_evmcs->hv_clean_fields |=
6518 			HV_VMX_ENLIGHTENED_CLEAN_FIELD_ALL;
6519 
6520 	/* MSR_IA32_DEBUGCTLMSR is zeroed on vmexit. Restore it if needed */
6521 	if (vmx->host_debugctlmsr)
6522 		update_debugctlmsr(vmx->host_debugctlmsr);
6523 
6524 #ifndef CONFIG_X86_64
6525 	/*
6526 	 * The sysexit path does not restore ds/es, so we must set them to
6527 	 * a reasonable value ourselves.
6528 	 *
6529 	 * We can't defer this to vmx_prepare_switch_to_host() since that
6530 	 * function may be executed in interrupt context, which saves and
6531 	 * restore segments around it, nullifying its effect.
6532 	 */
6533 	loadsegment(ds, __USER_DS);
6534 	loadsegment(es, __USER_DS);
6535 #endif
6536 
6537 	vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
6538 				  | (1 << VCPU_EXREG_RFLAGS)
6539 				  | (1 << VCPU_EXREG_PDPTR)
6540 				  | (1 << VCPU_EXREG_SEGMENTS)
6541 				  | (1 << VCPU_EXREG_CR3));
6542 	vcpu->arch.regs_dirty = 0;
6543 
6544 	pt_guest_exit(vmx);
6545 
6546 	/*
6547 	 * eager fpu is enabled if PKEY is supported and CR4 is switched
6548 	 * back on host, so it is safe to read guest PKRU from current
6549 	 * XSAVE.
6550 	 */
6551 	if (static_cpu_has(X86_FEATURE_PKU) &&
6552 	    kvm_read_cr4_bits(vcpu, X86_CR4_PKE)) {
6553 		vcpu->arch.pkru = rdpkru();
6554 		if (vcpu->arch.pkru != vmx->host_pkru)
6555 			__write_pkru(vmx->host_pkru);
6556 	}
6557 
6558 	kvm_put_guest_xcr0(vcpu);
6559 
6560 	vmx->nested.nested_run_pending = 0;
6561 	vmx->idt_vectoring_info = 0;
6562 
6563 	vmx->exit_reason = vmx->fail ? 0xdead : vmcs_read32(VM_EXIT_REASON);
6564 	if ((u16)vmx->exit_reason == EXIT_REASON_MCE_DURING_VMENTRY)
6565 		kvm_machine_check();
6566 
6567 	if (vmx->fail || (vmx->exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY))
6568 		return;
6569 
6570 	vmx->loaded_vmcs->launched = 1;
6571 	vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
6572 
6573 	vmx_recover_nmi_blocking(vmx);
6574 	vmx_complete_interrupts(vmx);
6575 }
6576 
6577 static struct kvm *vmx_vm_alloc(void)
6578 {
6579 	struct kvm_vmx *kvm_vmx = __vmalloc(sizeof(struct kvm_vmx),
6580 					    GFP_KERNEL_ACCOUNT | __GFP_ZERO,
6581 					    PAGE_KERNEL);
6582 	return &kvm_vmx->kvm;
6583 }
6584 
6585 static void vmx_vm_free(struct kvm *kvm)
6586 {
6587 	vfree(to_kvm_vmx(kvm));
6588 }
6589 
6590 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
6591 {
6592 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6593 
6594 	if (enable_pml)
6595 		vmx_destroy_pml_buffer(vmx);
6596 	free_vpid(vmx->vpid);
6597 	nested_vmx_free_vcpu(vcpu);
6598 	free_loaded_vmcs(vmx->loaded_vmcs);
6599 	kfree(vmx->guest_msrs);
6600 	kvm_vcpu_uninit(vcpu);
6601 	kmem_cache_free(x86_fpu_cache, vmx->vcpu.arch.user_fpu);
6602 	kmem_cache_free(x86_fpu_cache, vmx->vcpu.arch.guest_fpu);
6603 	kmem_cache_free(kvm_vcpu_cache, vmx);
6604 }
6605 
6606 static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
6607 {
6608 	int err;
6609 	struct vcpu_vmx *vmx;
6610 	unsigned long *msr_bitmap;
6611 	int cpu;
6612 
6613 	vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL_ACCOUNT);
6614 	if (!vmx)
6615 		return ERR_PTR(-ENOMEM);
6616 
6617 	vmx->vcpu.arch.user_fpu = kmem_cache_zalloc(x86_fpu_cache,
6618 			GFP_KERNEL_ACCOUNT);
6619 	if (!vmx->vcpu.arch.user_fpu) {
6620 		printk(KERN_ERR "kvm: failed to allocate kvm userspace's fpu\n");
6621 		err = -ENOMEM;
6622 		goto free_partial_vcpu;
6623 	}
6624 
6625 	vmx->vcpu.arch.guest_fpu = kmem_cache_zalloc(x86_fpu_cache,
6626 			GFP_KERNEL_ACCOUNT);
6627 	if (!vmx->vcpu.arch.guest_fpu) {
6628 		printk(KERN_ERR "kvm: failed to allocate vcpu's fpu\n");
6629 		err = -ENOMEM;
6630 		goto free_user_fpu;
6631 	}
6632 
6633 	vmx->vpid = allocate_vpid();
6634 
6635 	err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
6636 	if (err)
6637 		goto free_vcpu;
6638 
6639 	err = -ENOMEM;
6640 
6641 	/*
6642 	 * If PML is turned on, failure on enabling PML just results in failure
6643 	 * of creating the vcpu, therefore we can simplify PML logic (by
6644 	 * avoiding dealing with cases, such as enabling PML partially on vcpus
6645 	 * for the guest, etc.
6646 	 */
6647 	if (enable_pml) {
6648 		vmx->pml_pg = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
6649 		if (!vmx->pml_pg)
6650 			goto uninit_vcpu;
6651 	}
6652 
6653 	vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL_ACCOUNT);
6654 	BUILD_BUG_ON(ARRAY_SIZE(vmx_msr_index) * sizeof(vmx->guest_msrs[0])
6655 		     > PAGE_SIZE);
6656 
6657 	if (!vmx->guest_msrs)
6658 		goto free_pml;
6659 
6660 	err = alloc_loaded_vmcs(&vmx->vmcs01);
6661 	if (err < 0)
6662 		goto free_msrs;
6663 
6664 	msr_bitmap = vmx->vmcs01.msr_bitmap;
6665 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_TSC, MSR_TYPE_R);
6666 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_FS_BASE, MSR_TYPE_RW);
6667 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_GS_BASE, MSR_TYPE_RW);
6668 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_KERNEL_GS_BASE, MSR_TYPE_RW);
6669 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_CS, MSR_TYPE_RW);
6670 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_ESP, MSR_TYPE_RW);
6671 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_EIP, MSR_TYPE_RW);
6672 	if (kvm_cstate_in_guest(kvm)) {
6673 		vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C1_RES, MSR_TYPE_R);
6674 		vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C3_RESIDENCY, MSR_TYPE_R);
6675 		vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C6_RESIDENCY, MSR_TYPE_R);
6676 		vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C7_RESIDENCY, MSR_TYPE_R);
6677 	}
6678 	vmx->msr_bitmap_mode = 0;
6679 
6680 	vmx->loaded_vmcs = &vmx->vmcs01;
6681 	cpu = get_cpu();
6682 	vmx_vcpu_load(&vmx->vcpu, cpu);
6683 	vmx->vcpu.cpu = cpu;
6684 	vmx_vcpu_setup(vmx);
6685 	vmx_vcpu_put(&vmx->vcpu);
6686 	put_cpu();
6687 	if (cpu_need_virtualize_apic_accesses(&vmx->vcpu)) {
6688 		err = alloc_apic_access_page(kvm);
6689 		if (err)
6690 			goto free_vmcs;
6691 	}
6692 
6693 	if (enable_ept && !enable_unrestricted_guest) {
6694 		err = init_rmode_identity_map(kvm);
6695 		if (err)
6696 			goto free_vmcs;
6697 	}
6698 
6699 	if (nested)
6700 		nested_vmx_setup_ctls_msrs(&vmx->nested.msrs,
6701 					   vmx_capability.ept,
6702 					   kvm_vcpu_apicv_active(&vmx->vcpu));
6703 	else
6704 		memset(&vmx->nested.msrs, 0, sizeof(vmx->nested.msrs));
6705 
6706 	vmx->nested.posted_intr_nv = -1;
6707 	vmx->nested.current_vmptr = -1ull;
6708 
6709 	vmx->msr_ia32_feature_control_valid_bits = FEATURE_CONTROL_LOCKED;
6710 
6711 	/*
6712 	 * Enforce invariant: pi_desc.nv is always either POSTED_INTR_VECTOR
6713 	 * or POSTED_INTR_WAKEUP_VECTOR.
6714 	 */
6715 	vmx->pi_desc.nv = POSTED_INTR_VECTOR;
6716 	vmx->pi_desc.sn = 1;
6717 
6718 	vmx->ept_pointer = INVALID_PAGE;
6719 
6720 	return &vmx->vcpu;
6721 
6722 free_vmcs:
6723 	free_loaded_vmcs(vmx->loaded_vmcs);
6724 free_msrs:
6725 	kfree(vmx->guest_msrs);
6726 free_pml:
6727 	vmx_destroy_pml_buffer(vmx);
6728 uninit_vcpu:
6729 	kvm_vcpu_uninit(&vmx->vcpu);
6730 free_vcpu:
6731 	free_vpid(vmx->vpid);
6732 	kmem_cache_free(x86_fpu_cache, vmx->vcpu.arch.guest_fpu);
6733 free_user_fpu:
6734 	kmem_cache_free(x86_fpu_cache, vmx->vcpu.arch.user_fpu);
6735 free_partial_vcpu:
6736 	kmem_cache_free(kvm_vcpu_cache, vmx);
6737 	return ERR_PTR(err);
6738 }
6739 
6740 #define L1TF_MSG_SMT "L1TF CPU bug present and SMT on, data leak possible. See CVE-2018-3646 and https://www.kernel.org/doc/html/latest/admin-guide/hw-vuln/l1tf.html for details.\n"
6741 #define L1TF_MSG_L1D "L1TF CPU bug present and virtualization mitigation disabled, data leak possible. See CVE-2018-3646 and https://www.kernel.org/doc/html/latest/admin-guide/hw-vuln/l1tf.html for details.\n"
6742 
6743 static int vmx_vm_init(struct kvm *kvm)
6744 {
6745 	spin_lock_init(&to_kvm_vmx(kvm)->ept_pointer_lock);
6746 
6747 	if (!ple_gap)
6748 		kvm->arch.pause_in_guest = true;
6749 
6750 	if (boot_cpu_has(X86_BUG_L1TF) && enable_ept) {
6751 		switch (l1tf_mitigation) {
6752 		case L1TF_MITIGATION_OFF:
6753 		case L1TF_MITIGATION_FLUSH_NOWARN:
6754 			/* 'I explicitly don't care' is set */
6755 			break;
6756 		case L1TF_MITIGATION_FLUSH:
6757 		case L1TF_MITIGATION_FLUSH_NOSMT:
6758 		case L1TF_MITIGATION_FULL:
6759 			/*
6760 			 * Warn upon starting the first VM in a potentially
6761 			 * insecure environment.
6762 			 */
6763 			if (sched_smt_active())
6764 				pr_warn_once(L1TF_MSG_SMT);
6765 			if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_NEVER)
6766 				pr_warn_once(L1TF_MSG_L1D);
6767 			break;
6768 		case L1TF_MITIGATION_FULL_FORCE:
6769 			/* Flush is enforced */
6770 			break;
6771 		}
6772 	}
6773 	return 0;
6774 }
6775 
6776 static int __init vmx_check_processor_compat(void)
6777 {
6778 	struct vmcs_config vmcs_conf;
6779 	struct vmx_capability vmx_cap;
6780 
6781 	if (setup_vmcs_config(&vmcs_conf, &vmx_cap) < 0)
6782 		return -EIO;
6783 	if (nested)
6784 		nested_vmx_setup_ctls_msrs(&vmcs_conf.nested, vmx_cap.ept,
6785 					   enable_apicv);
6786 	if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
6787 		printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
6788 				smp_processor_id());
6789 		return -EIO;
6790 	}
6791 	return 0;
6792 }
6793 
6794 static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
6795 {
6796 	u8 cache;
6797 	u64 ipat = 0;
6798 
6799 	/* For VT-d and EPT combination
6800 	 * 1. MMIO: always map as UC
6801 	 * 2. EPT with VT-d:
6802 	 *   a. VT-d without snooping control feature: can't guarantee the
6803 	 *	result, try to trust guest.
6804 	 *   b. VT-d with snooping control feature: snooping control feature of
6805 	 *	VT-d engine can guarantee the cache correctness. Just set it
6806 	 *	to WB to keep consistent with host. So the same as item 3.
6807 	 * 3. EPT without VT-d: always map as WB and set IPAT=1 to keep
6808 	 *    consistent with host MTRR
6809 	 */
6810 	if (is_mmio) {
6811 		cache = MTRR_TYPE_UNCACHABLE;
6812 		goto exit;
6813 	}
6814 
6815 	if (!kvm_arch_has_noncoherent_dma(vcpu->kvm)) {
6816 		ipat = VMX_EPT_IPAT_BIT;
6817 		cache = MTRR_TYPE_WRBACK;
6818 		goto exit;
6819 	}
6820 
6821 	if (kvm_read_cr0(vcpu) & X86_CR0_CD) {
6822 		ipat = VMX_EPT_IPAT_BIT;
6823 		if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
6824 			cache = MTRR_TYPE_WRBACK;
6825 		else
6826 			cache = MTRR_TYPE_UNCACHABLE;
6827 		goto exit;
6828 	}
6829 
6830 	cache = kvm_mtrr_get_guest_memory_type(vcpu, gfn);
6831 
6832 exit:
6833 	return (cache << VMX_EPT_MT_EPTE_SHIFT) | ipat;
6834 }
6835 
6836 static int vmx_get_lpage_level(void)
6837 {
6838 	if (enable_ept && !cpu_has_vmx_ept_1g_page())
6839 		return PT_DIRECTORY_LEVEL;
6840 	else
6841 		/* For shadow and EPT supported 1GB page */
6842 		return PT_PDPE_LEVEL;
6843 }
6844 
6845 static void vmcs_set_secondary_exec_control(struct vcpu_vmx *vmx)
6846 {
6847 	/*
6848 	 * These bits in the secondary execution controls field
6849 	 * are dynamic, the others are mostly based on the hypervisor
6850 	 * architecture and the guest's CPUID.  Do not touch the
6851 	 * dynamic bits.
6852 	 */
6853 	u32 mask =
6854 		SECONDARY_EXEC_SHADOW_VMCS |
6855 		SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
6856 		SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
6857 		SECONDARY_EXEC_DESC;
6858 
6859 	u32 new_ctl = vmx->secondary_exec_control;
6860 	u32 cur_ctl = secondary_exec_controls_get(vmx);
6861 
6862 	secondary_exec_controls_set(vmx, (new_ctl & ~mask) | (cur_ctl & mask));
6863 }
6864 
6865 /*
6866  * Generate MSR_IA32_VMX_CR{0,4}_FIXED1 according to CPUID. Only set bits
6867  * (indicating "allowed-1") if they are supported in the guest's CPUID.
6868  */
6869 static void nested_vmx_cr_fixed1_bits_update(struct kvm_vcpu *vcpu)
6870 {
6871 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6872 	struct kvm_cpuid_entry2 *entry;
6873 
6874 	vmx->nested.msrs.cr0_fixed1 = 0xffffffff;
6875 	vmx->nested.msrs.cr4_fixed1 = X86_CR4_PCE;
6876 
6877 #define cr4_fixed1_update(_cr4_mask, _reg, _cpuid_mask) do {		\
6878 	if (entry && (entry->_reg & (_cpuid_mask)))			\
6879 		vmx->nested.msrs.cr4_fixed1 |= (_cr4_mask);	\
6880 } while (0)
6881 
6882 	entry = kvm_find_cpuid_entry(vcpu, 0x1, 0);
6883 	cr4_fixed1_update(X86_CR4_VME,        edx, bit(X86_FEATURE_VME));
6884 	cr4_fixed1_update(X86_CR4_PVI,        edx, bit(X86_FEATURE_VME));
6885 	cr4_fixed1_update(X86_CR4_TSD,        edx, bit(X86_FEATURE_TSC));
6886 	cr4_fixed1_update(X86_CR4_DE,         edx, bit(X86_FEATURE_DE));
6887 	cr4_fixed1_update(X86_CR4_PSE,        edx, bit(X86_FEATURE_PSE));
6888 	cr4_fixed1_update(X86_CR4_PAE,        edx, bit(X86_FEATURE_PAE));
6889 	cr4_fixed1_update(X86_CR4_MCE,        edx, bit(X86_FEATURE_MCE));
6890 	cr4_fixed1_update(X86_CR4_PGE,        edx, bit(X86_FEATURE_PGE));
6891 	cr4_fixed1_update(X86_CR4_OSFXSR,     edx, bit(X86_FEATURE_FXSR));
6892 	cr4_fixed1_update(X86_CR4_OSXMMEXCPT, edx, bit(X86_FEATURE_XMM));
6893 	cr4_fixed1_update(X86_CR4_VMXE,       ecx, bit(X86_FEATURE_VMX));
6894 	cr4_fixed1_update(X86_CR4_SMXE,       ecx, bit(X86_FEATURE_SMX));
6895 	cr4_fixed1_update(X86_CR4_PCIDE,      ecx, bit(X86_FEATURE_PCID));
6896 	cr4_fixed1_update(X86_CR4_OSXSAVE,    ecx, bit(X86_FEATURE_XSAVE));
6897 
6898 	entry = kvm_find_cpuid_entry(vcpu, 0x7, 0);
6899 	cr4_fixed1_update(X86_CR4_FSGSBASE,   ebx, bit(X86_FEATURE_FSGSBASE));
6900 	cr4_fixed1_update(X86_CR4_SMEP,       ebx, bit(X86_FEATURE_SMEP));
6901 	cr4_fixed1_update(X86_CR4_SMAP,       ebx, bit(X86_FEATURE_SMAP));
6902 	cr4_fixed1_update(X86_CR4_PKE,        ecx, bit(X86_FEATURE_PKU));
6903 	cr4_fixed1_update(X86_CR4_UMIP,       ecx, bit(X86_FEATURE_UMIP));
6904 
6905 #undef cr4_fixed1_update
6906 }
6907 
6908 static void nested_vmx_entry_exit_ctls_update(struct kvm_vcpu *vcpu)
6909 {
6910 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6911 
6912 	if (kvm_mpx_supported()) {
6913 		bool mpx_enabled = guest_cpuid_has(vcpu, X86_FEATURE_MPX);
6914 
6915 		if (mpx_enabled) {
6916 			vmx->nested.msrs.entry_ctls_high |= VM_ENTRY_LOAD_BNDCFGS;
6917 			vmx->nested.msrs.exit_ctls_high |= VM_EXIT_CLEAR_BNDCFGS;
6918 		} else {
6919 			vmx->nested.msrs.entry_ctls_high &= ~VM_ENTRY_LOAD_BNDCFGS;
6920 			vmx->nested.msrs.exit_ctls_high &= ~VM_EXIT_CLEAR_BNDCFGS;
6921 		}
6922 	}
6923 }
6924 
6925 static void update_intel_pt_cfg(struct kvm_vcpu *vcpu)
6926 {
6927 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6928 	struct kvm_cpuid_entry2 *best = NULL;
6929 	int i;
6930 
6931 	for (i = 0; i < PT_CPUID_LEAVES; i++) {
6932 		best = kvm_find_cpuid_entry(vcpu, 0x14, i);
6933 		if (!best)
6934 			return;
6935 		vmx->pt_desc.caps[CPUID_EAX + i*PT_CPUID_REGS_NUM] = best->eax;
6936 		vmx->pt_desc.caps[CPUID_EBX + i*PT_CPUID_REGS_NUM] = best->ebx;
6937 		vmx->pt_desc.caps[CPUID_ECX + i*PT_CPUID_REGS_NUM] = best->ecx;
6938 		vmx->pt_desc.caps[CPUID_EDX + i*PT_CPUID_REGS_NUM] = best->edx;
6939 	}
6940 
6941 	/* Get the number of configurable Address Ranges for filtering */
6942 	vmx->pt_desc.addr_range = intel_pt_validate_cap(vmx->pt_desc.caps,
6943 						PT_CAP_num_address_ranges);
6944 
6945 	/* Initialize and clear the no dependency bits */
6946 	vmx->pt_desc.ctl_bitmask = ~(RTIT_CTL_TRACEEN | RTIT_CTL_OS |
6947 			RTIT_CTL_USR | RTIT_CTL_TSC_EN | RTIT_CTL_DISRETC);
6948 
6949 	/*
6950 	 * If CPUID.(EAX=14H,ECX=0):EBX[0]=1 CR3Filter can be set otherwise
6951 	 * will inject an #GP
6952 	 */
6953 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_cr3_filtering))
6954 		vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_CR3EN;
6955 
6956 	/*
6957 	 * If CPUID.(EAX=14H,ECX=0):EBX[1]=1 CYCEn, CycThresh and
6958 	 * PSBFreq can be set
6959 	 */
6960 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc))
6961 		vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_CYCLEACC |
6962 				RTIT_CTL_CYC_THRESH | RTIT_CTL_PSB_FREQ);
6963 
6964 	/*
6965 	 * If CPUID.(EAX=14H,ECX=0):EBX[3]=1 MTCEn BranchEn and
6966 	 * MTCFreq can be set
6967 	 */
6968 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc))
6969 		vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_MTC_EN |
6970 				RTIT_CTL_BRANCH_EN | RTIT_CTL_MTC_RANGE);
6971 
6972 	/* If CPUID.(EAX=14H,ECX=0):EBX[4]=1 FUPonPTW and PTWEn can be set */
6973 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_ptwrite))
6974 		vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_FUP_ON_PTW |
6975 							RTIT_CTL_PTW_EN);
6976 
6977 	/* If CPUID.(EAX=14H,ECX=0):EBX[5]=1 PwrEvEn can be set */
6978 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_power_event_trace))
6979 		vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_PWR_EVT_EN;
6980 
6981 	/* If CPUID.(EAX=14H,ECX=0):ECX[0]=1 ToPA can be set */
6982 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_topa_output))
6983 		vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_TOPA;
6984 
6985 	/* If CPUID.(EAX=14H,ECX=0):ECX[3]=1 FabircEn can be set */
6986 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_output_subsys))
6987 		vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_FABRIC_EN;
6988 
6989 	/* unmask address range configure area */
6990 	for (i = 0; i < vmx->pt_desc.addr_range; i++)
6991 		vmx->pt_desc.ctl_bitmask &= ~(0xfULL << (32 + i * 4));
6992 }
6993 
6994 static void vmx_cpuid_update(struct kvm_vcpu *vcpu)
6995 {
6996 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6997 
6998 	if (cpu_has_secondary_exec_ctrls()) {
6999 		vmx_compute_secondary_exec_control(vmx);
7000 		vmcs_set_secondary_exec_control(vmx);
7001 	}
7002 
7003 	if (nested_vmx_allowed(vcpu))
7004 		to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
7005 			FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
7006 	else
7007 		to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
7008 			~FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
7009 
7010 	if (nested_vmx_allowed(vcpu)) {
7011 		nested_vmx_cr_fixed1_bits_update(vcpu);
7012 		nested_vmx_entry_exit_ctls_update(vcpu);
7013 	}
7014 
7015 	if (boot_cpu_has(X86_FEATURE_INTEL_PT) &&
7016 			guest_cpuid_has(vcpu, X86_FEATURE_INTEL_PT))
7017 		update_intel_pt_cfg(vcpu);
7018 }
7019 
7020 static void vmx_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
7021 {
7022 	if (func == 1 && nested)
7023 		entry->ecx |= bit(X86_FEATURE_VMX);
7024 }
7025 
7026 static void vmx_request_immediate_exit(struct kvm_vcpu *vcpu)
7027 {
7028 	to_vmx(vcpu)->req_immediate_exit = true;
7029 }
7030 
7031 static int vmx_check_intercept(struct kvm_vcpu *vcpu,
7032 			       struct x86_instruction_info *info,
7033 			       enum x86_intercept_stage stage)
7034 {
7035 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
7036 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
7037 
7038 	/*
7039 	 * RDPID causes #UD if disabled through secondary execution controls.
7040 	 * Because it is marked as EmulateOnUD, we need to intercept it here.
7041 	 */
7042 	if (info->intercept == x86_intercept_rdtscp &&
7043 	    !nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDTSCP)) {
7044 		ctxt->exception.vector = UD_VECTOR;
7045 		ctxt->exception.error_code_valid = false;
7046 		return X86EMUL_PROPAGATE_FAULT;
7047 	}
7048 
7049 	/* TODO: check more intercepts... */
7050 	return X86EMUL_CONTINUE;
7051 }
7052 
7053 #ifdef CONFIG_X86_64
7054 /* (a << shift) / divisor, return 1 if overflow otherwise 0 */
7055 static inline int u64_shl_div_u64(u64 a, unsigned int shift,
7056 				  u64 divisor, u64 *result)
7057 {
7058 	u64 low = a << shift, high = a >> (64 - shift);
7059 
7060 	/* To avoid the overflow on divq */
7061 	if (high >= divisor)
7062 		return 1;
7063 
7064 	/* Low hold the result, high hold rem which is discarded */
7065 	asm("divq %2\n\t" : "=a" (low), "=d" (high) :
7066 	    "rm" (divisor), "0" (low), "1" (high));
7067 	*result = low;
7068 
7069 	return 0;
7070 }
7071 
7072 static int vmx_set_hv_timer(struct kvm_vcpu *vcpu, u64 guest_deadline_tsc,
7073 			    bool *expired)
7074 {
7075 	struct vcpu_vmx *vmx;
7076 	u64 tscl, guest_tscl, delta_tsc, lapic_timer_advance_cycles;
7077 	struct kvm_timer *ktimer = &vcpu->arch.apic->lapic_timer;
7078 
7079 	if (kvm_mwait_in_guest(vcpu->kvm) ||
7080 		kvm_can_post_timer_interrupt(vcpu))
7081 		return -EOPNOTSUPP;
7082 
7083 	vmx = to_vmx(vcpu);
7084 	tscl = rdtsc();
7085 	guest_tscl = kvm_read_l1_tsc(vcpu, tscl);
7086 	delta_tsc = max(guest_deadline_tsc, guest_tscl) - guest_tscl;
7087 	lapic_timer_advance_cycles = nsec_to_cycles(vcpu,
7088 						    ktimer->timer_advance_ns);
7089 
7090 	if (delta_tsc > lapic_timer_advance_cycles)
7091 		delta_tsc -= lapic_timer_advance_cycles;
7092 	else
7093 		delta_tsc = 0;
7094 
7095 	/* Convert to host delta tsc if tsc scaling is enabled */
7096 	if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio &&
7097 	    delta_tsc && u64_shl_div_u64(delta_tsc,
7098 				kvm_tsc_scaling_ratio_frac_bits,
7099 				vcpu->arch.tsc_scaling_ratio, &delta_tsc))
7100 		return -ERANGE;
7101 
7102 	/*
7103 	 * If the delta tsc can't fit in the 32 bit after the multi shift,
7104 	 * we can't use the preemption timer.
7105 	 * It's possible that it fits on later vmentries, but checking
7106 	 * on every vmentry is costly so we just use an hrtimer.
7107 	 */
7108 	if (delta_tsc >> (cpu_preemption_timer_multi + 32))
7109 		return -ERANGE;
7110 
7111 	vmx->hv_deadline_tsc = tscl + delta_tsc;
7112 	*expired = !delta_tsc;
7113 	return 0;
7114 }
7115 
7116 static void vmx_cancel_hv_timer(struct kvm_vcpu *vcpu)
7117 {
7118 	to_vmx(vcpu)->hv_deadline_tsc = -1;
7119 }
7120 #endif
7121 
7122 static void vmx_sched_in(struct kvm_vcpu *vcpu, int cpu)
7123 {
7124 	if (!kvm_pause_in_guest(vcpu->kvm))
7125 		shrink_ple_window(vcpu);
7126 }
7127 
7128 static void vmx_slot_enable_log_dirty(struct kvm *kvm,
7129 				     struct kvm_memory_slot *slot)
7130 {
7131 	kvm_mmu_slot_leaf_clear_dirty(kvm, slot);
7132 	kvm_mmu_slot_largepage_remove_write_access(kvm, slot);
7133 }
7134 
7135 static void vmx_slot_disable_log_dirty(struct kvm *kvm,
7136 				       struct kvm_memory_slot *slot)
7137 {
7138 	kvm_mmu_slot_set_dirty(kvm, slot);
7139 }
7140 
7141 static void vmx_flush_log_dirty(struct kvm *kvm)
7142 {
7143 	kvm_flush_pml_buffers(kvm);
7144 }
7145 
7146 static int vmx_write_pml_buffer(struct kvm_vcpu *vcpu)
7147 {
7148 	struct vmcs12 *vmcs12;
7149 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7150 	gpa_t gpa, dst;
7151 
7152 	if (is_guest_mode(vcpu)) {
7153 		WARN_ON_ONCE(vmx->nested.pml_full);
7154 
7155 		/*
7156 		 * Check if PML is enabled for the nested guest.
7157 		 * Whether eptp bit 6 is set is already checked
7158 		 * as part of A/D emulation.
7159 		 */
7160 		vmcs12 = get_vmcs12(vcpu);
7161 		if (!nested_cpu_has_pml(vmcs12))
7162 			return 0;
7163 
7164 		if (vmcs12->guest_pml_index >= PML_ENTITY_NUM) {
7165 			vmx->nested.pml_full = true;
7166 			return 1;
7167 		}
7168 
7169 		gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS) & ~0xFFFull;
7170 		dst = vmcs12->pml_address + sizeof(u64) * vmcs12->guest_pml_index;
7171 
7172 		if (kvm_write_guest_page(vcpu->kvm, gpa_to_gfn(dst), &gpa,
7173 					 offset_in_page(dst), sizeof(gpa)))
7174 			return 0;
7175 
7176 		vmcs12->guest_pml_index--;
7177 	}
7178 
7179 	return 0;
7180 }
7181 
7182 static void vmx_enable_log_dirty_pt_masked(struct kvm *kvm,
7183 					   struct kvm_memory_slot *memslot,
7184 					   gfn_t offset, unsigned long mask)
7185 {
7186 	kvm_mmu_clear_dirty_pt_masked(kvm, memslot, offset, mask);
7187 }
7188 
7189 static void __pi_post_block(struct kvm_vcpu *vcpu)
7190 {
7191 	struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
7192 	struct pi_desc old, new;
7193 	unsigned int dest;
7194 
7195 	do {
7196 		old.control = new.control = pi_desc->control;
7197 		WARN(old.nv != POSTED_INTR_WAKEUP_VECTOR,
7198 		     "Wakeup handler not enabled while the VCPU is blocked\n");
7199 
7200 		dest = cpu_physical_id(vcpu->cpu);
7201 
7202 		if (x2apic_enabled())
7203 			new.ndst = dest;
7204 		else
7205 			new.ndst = (dest << 8) & 0xFF00;
7206 
7207 		/* set 'NV' to 'notification vector' */
7208 		new.nv = POSTED_INTR_VECTOR;
7209 	} while (cmpxchg64(&pi_desc->control, old.control,
7210 			   new.control) != old.control);
7211 
7212 	if (!WARN_ON_ONCE(vcpu->pre_pcpu == -1)) {
7213 		spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7214 		list_del(&vcpu->blocked_vcpu_list);
7215 		spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7216 		vcpu->pre_pcpu = -1;
7217 	}
7218 }
7219 
7220 /*
7221  * This routine does the following things for vCPU which is going
7222  * to be blocked if VT-d PI is enabled.
7223  * - Store the vCPU to the wakeup list, so when interrupts happen
7224  *   we can find the right vCPU to wake up.
7225  * - Change the Posted-interrupt descriptor as below:
7226  *      'NDST' <-- vcpu->pre_pcpu
7227  *      'NV' <-- POSTED_INTR_WAKEUP_VECTOR
7228  * - If 'ON' is set during this process, which means at least one
7229  *   interrupt is posted for this vCPU, we cannot block it, in
7230  *   this case, return 1, otherwise, return 0.
7231  *
7232  */
7233 static int pi_pre_block(struct kvm_vcpu *vcpu)
7234 {
7235 	unsigned int dest;
7236 	struct pi_desc old, new;
7237 	struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
7238 
7239 	if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
7240 		!irq_remapping_cap(IRQ_POSTING_CAP)  ||
7241 		!kvm_vcpu_apicv_active(vcpu))
7242 		return 0;
7243 
7244 	WARN_ON(irqs_disabled());
7245 	local_irq_disable();
7246 	if (!WARN_ON_ONCE(vcpu->pre_pcpu != -1)) {
7247 		vcpu->pre_pcpu = vcpu->cpu;
7248 		spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7249 		list_add_tail(&vcpu->blocked_vcpu_list,
7250 			      &per_cpu(blocked_vcpu_on_cpu,
7251 				       vcpu->pre_pcpu));
7252 		spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7253 	}
7254 
7255 	do {
7256 		old.control = new.control = pi_desc->control;
7257 
7258 		WARN((pi_desc->sn == 1),
7259 		     "Warning: SN field of posted-interrupts "
7260 		     "is set before blocking\n");
7261 
7262 		/*
7263 		 * Since vCPU can be preempted during this process,
7264 		 * vcpu->cpu could be different with pre_pcpu, we
7265 		 * need to set pre_pcpu as the destination of wakeup
7266 		 * notification event, then we can find the right vCPU
7267 		 * to wakeup in wakeup handler if interrupts happen
7268 		 * when the vCPU is in blocked state.
7269 		 */
7270 		dest = cpu_physical_id(vcpu->pre_pcpu);
7271 
7272 		if (x2apic_enabled())
7273 			new.ndst = dest;
7274 		else
7275 			new.ndst = (dest << 8) & 0xFF00;
7276 
7277 		/* set 'NV' to 'wakeup vector' */
7278 		new.nv = POSTED_INTR_WAKEUP_VECTOR;
7279 	} while (cmpxchg64(&pi_desc->control, old.control,
7280 			   new.control) != old.control);
7281 
7282 	/* We should not block the vCPU if an interrupt is posted for it.  */
7283 	if (pi_test_on(pi_desc) == 1)
7284 		__pi_post_block(vcpu);
7285 
7286 	local_irq_enable();
7287 	return (vcpu->pre_pcpu == -1);
7288 }
7289 
7290 static int vmx_pre_block(struct kvm_vcpu *vcpu)
7291 {
7292 	if (pi_pre_block(vcpu))
7293 		return 1;
7294 
7295 	if (kvm_lapic_hv_timer_in_use(vcpu))
7296 		kvm_lapic_switch_to_sw_timer(vcpu);
7297 
7298 	return 0;
7299 }
7300 
7301 static void pi_post_block(struct kvm_vcpu *vcpu)
7302 {
7303 	if (vcpu->pre_pcpu == -1)
7304 		return;
7305 
7306 	WARN_ON(irqs_disabled());
7307 	local_irq_disable();
7308 	__pi_post_block(vcpu);
7309 	local_irq_enable();
7310 }
7311 
7312 static void vmx_post_block(struct kvm_vcpu *vcpu)
7313 {
7314 	if (kvm_x86_ops->set_hv_timer)
7315 		kvm_lapic_switch_to_hv_timer(vcpu);
7316 
7317 	pi_post_block(vcpu);
7318 }
7319 
7320 /*
7321  * vmx_update_pi_irte - set IRTE for Posted-Interrupts
7322  *
7323  * @kvm: kvm
7324  * @host_irq: host irq of the interrupt
7325  * @guest_irq: gsi of the interrupt
7326  * @set: set or unset PI
7327  * returns 0 on success, < 0 on failure
7328  */
7329 static int vmx_update_pi_irte(struct kvm *kvm, unsigned int host_irq,
7330 			      uint32_t guest_irq, bool set)
7331 {
7332 	struct kvm_kernel_irq_routing_entry *e;
7333 	struct kvm_irq_routing_table *irq_rt;
7334 	struct kvm_lapic_irq irq;
7335 	struct kvm_vcpu *vcpu;
7336 	struct vcpu_data vcpu_info;
7337 	int idx, ret = 0;
7338 
7339 	if (!kvm_arch_has_assigned_device(kvm) ||
7340 		!irq_remapping_cap(IRQ_POSTING_CAP) ||
7341 		!kvm_vcpu_apicv_active(kvm->vcpus[0]))
7342 		return 0;
7343 
7344 	idx = srcu_read_lock(&kvm->irq_srcu);
7345 	irq_rt = srcu_dereference(kvm->irq_routing, &kvm->irq_srcu);
7346 	if (guest_irq >= irq_rt->nr_rt_entries ||
7347 	    hlist_empty(&irq_rt->map[guest_irq])) {
7348 		pr_warn_once("no route for guest_irq %u/%u (broken user space?)\n",
7349 			     guest_irq, irq_rt->nr_rt_entries);
7350 		goto out;
7351 	}
7352 
7353 	hlist_for_each_entry(e, &irq_rt->map[guest_irq], link) {
7354 		if (e->type != KVM_IRQ_ROUTING_MSI)
7355 			continue;
7356 		/*
7357 		 * VT-d PI cannot support posting multicast/broadcast
7358 		 * interrupts to a vCPU, we still use interrupt remapping
7359 		 * for these kind of interrupts.
7360 		 *
7361 		 * For lowest-priority interrupts, we only support
7362 		 * those with single CPU as the destination, e.g. user
7363 		 * configures the interrupts via /proc/irq or uses
7364 		 * irqbalance to make the interrupts single-CPU.
7365 		 *
7366 		 * We will support full lowest-priority interrupt later.
7367 		 */
7368 
7369 		kvm_set_msi_irq(kvm, e, &irq);
7370 		if (!kvm_intr_is_single_vcpu(kvm, &irq, &vcpu)) {
7371 			/*
7372 			 * Make sure the IRTE is in remapped mode if
7373 			 * we don't handle it in posted mode.
7374 			 */
7375 			ret = irq_set_vcpu_affinity(host_irq, NULL);
7376 			if (ret < 0) {
7377 				printk(KERN_INFO
7378 				   "failed to back to remapped mode, irq: %u\n",
7379 				   host_irq);
7380 				goto out;
7381 			}
7382 
7383 			continue;
7384 		}
7385 
7386 		vcpu_info.pi_desc_addr = __pa(vcpu_to_pi_desc(vcpu));
7387 		vcpu_info.vector = irq.vector;
7388 
7389 		trace_kvm_pi_irte_update(host_irq, vcpu->vcpu_id, e->gsi,
7390 				vcpu_info.vector, vcpu_info.pi_desc_addr, set);
7391 
7392 		if (set)
7393 			ret = irq_set_vcpu_affinity(host_irq, &vcpu_info);
7394 		else
7395 			ret = irq_set_vcpu_affinity(host_irq, NULL);
7396 
7397 		if (ret < 0) {
7398 			printk(KERN_INFO "%s: failed to update PI IRTE\n",
7399 					__func__);
7400 			goto out;
7401 		}
7402 	}
7403 
7404 	ret = 0;
7405 out:
7406 	srcu_read_unlock(&kvm->irq_srcu, idx);
7407 	return ret;
7408 }
7409 
7410 static void vmx_setup_mce(struct kvm_vcpu *vcpu)
7411 {
7412 	if (vcpu->arch.mcg_cap & MCG_LMCE_P)
7413 		to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
7414 			FEATURE_CONTROL_LMCE;
7415 	else
7416 		to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
7417 			~FEATURE_CONTROL_LMCE;
7418 }
7419 
7420 static int vmx_smi_allowed(struct kvm_vcpu *vcpu)
7421 {
7422 	/* we need a nested vmexit to enter SMM, postpone if run is pending */
7423 	if (to_vmx(vcpu)->nested.nested_run_pending)
7424 		return 0;
7425 	return 1;
7426 }
7427 
7428 static int vmx_pre_enter_smm(struct kvm_vcpu *vcpu, char *smstate)
7429 {
7430 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7431 
7432 	vmx->nested.smm.guest_mode = is_guest_mode(vcpu);
7433 	if (vmx->nested.smm.guest_mode)
7434 		nested_vmx_vmexit(vcpu, -1, 0, 0);
7435 
7436 	vmx->nested.smm.vmxon = vmx->nested.vmxon;
7437 	vmx->nested.vmxon = false;
7438 	vmx_clear_hlt(vcpu);
7439 	return 0;
7440 }
7441 
7442 static int vmx_pre_leave_smm(struct kvm_vcpu *vcpu, const char *smstate)
7443 {
7444 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7445 	int ret;
7446 
7447 	if (vmx->nested.smm.vmxon) {
7448 		vmx->nested.vmxon = true;
7449 		vmx->nested.smm.vmxon = false;
7450 	}
7451 
7452 	if (vmx->nested.smm.guest_mode) {
7453 		ret = nested_vmx_enter_non_root_mode(vcpu, false);
7454 		if (ret)
7455 			return ret;
7456 
7457 		vmx->nested.smm.guest_mode = false;
7458 	}
7459 	return 0;
7460 }
7461 
7462 static int enable_smi_window(struct kvm_vcpu *vcpu)
7463 {
7464 	return 0;
7465 }
7466 
7467 static bool vmx_need_emulation_on_page_fault(struct kvm_vcpu *vcpu)
7468 {
7469 	return false;
7470 }
7471 
7472 static __init int hardware_setup(void)
7473 {
7474 	unsigned long host_bndcfgs;
7475 	struct desc_ptr dt;
7476 	int r, i;
7477 
7478 	rdmsrl_safe(MSR_EFER, &host_efer);
7479 
7480 	store_idt(&dt);
7481 	host_idt_base = dt.address;
7482 
7483 	for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i)
7484 		kvm_define_shared_msr(i, vmx_msr_index[i]);
7485 
7486 	if (setup_vmcs_config(&vmcs_config, &vmx_capability) < 0)
7487 		return -EIO;
7488 
7489 	if (boot_cpu_has(X86_FEATURE_NX))
7490 		kvm_enable_efer_bits(EFER_NX);
7491 
7492 	if (boot_cpu_has(X86_FEATURE_MPX)) {
7493 		rdmsrl(MSR_IA32_BNDCFGS, host_bndcfgs);
7494 		WARN_ONCE(host_bndcfgs, "KVM: BNDCFGS in host will be lost");
7495 	}
7496 
7497 	if (boot_cpu_has(X86_FEATURE_XSAVES))
7498 		rdmsrl(MSR_IA32_XSS, host_xss);
7499 
7500 	if (!cpu_has_vmx_vpid() || !cpu_has_vmx_invvpid() ||
7501 	    !(cpu_has_vmx_invvpid_single() || cpu_has_vmx_invvpid_global()))
7502 		enable_vpid = 0;
7503 
7504 	if (!cpu_has_vmx_ept() ||
7505 	    !cpu_has_vmx_ept_4levels() ||
7506 	    !cpu_has_vmx_ept_mt_wb() ||
7507 	    !cpu_has_vmx_invept_global())
7508 		enable_ept = 0;
7509 
7510 	if (!cpu_has_vmx_ept_ad_bits() || !enable_ept)
7511 		enable_ept_ad_bits = 0;
7512 
7513 	if (!cpu_has_vmx_unrestricted_guest() || !enable_ept)
7514 		enable_unrestricted_guest = 0;
7515 
7516 	if (!cpu_has_vmx_flexpriority())
7517 		flexpriority_enabled = 0;
7518 
7519 	if (!cpu_has_virtual_nmis())
7520 		enable_vnmi = 0;
7521 
7522 	/*
7523 	 * set_apic_access_page_addr() is used to reload apic access
7524 	 * page upon invalidation.  No need to do anything if not
7525 	 * using the APIC_ACCESS_ADDR VMCS field.
7526 	 */
7527 	if (!flexpriority_enabled)
7528 		kvm_x86_ops->set_apic_access_page_addr = NULL;
7529 
7530 	if (!cpu_has_vmx_tpr_shadow())
7531 		kvm_x86_ops->update_cr8_intercept = NULL;
7532 
7533 	if (enable_ept && !cpu_has_vmx_ept_2m_page())
7534 		kvm_disable_largepages();
7535 
7536 #if IS_ENABLED(CONFIG_HYPERV)
7537 	if (ms_hyperv.nested_features & HV_X64_NESTED_GUEST_MAPPING_FLUSH
7538 	    && enable_ept) {
7539 		kvm_x86_ops->tlb_remote_flush = hv_remote_flush_tlb;
7540 		kvm_x86_ops->tlb_remote_flush_with_range =
7541 				hv_remote_flush_tlb_with_range;
7542 	}
7543 #endif
7544 
7545 	if (!cpu_has_vmx_ple()) {
7546 		ple_gap = 0;
7547 		ple_window = 0;
7548 		ple_window_grow = 0;
7549 		ple_window_max = 0;
7550 		ple_window_shrink = 0;
7551 	}
7552 
7553 	if (!cpu_has_vmx_apicv()) {
7554 		enable_apicv = 0;
7555 		kvm_x86_ops->sync_pir_to_irr = NULL;
7556 	}
7557 
7558 	if (cpu_has_vmx_tsc_scaling()) {
7559 		kvm_has_tsc_control = true;
7560 		kvm_max_tsc_scaling_ratio = KVM_VMX_TSC_MULTIPLIER_MAX;
7561 		kvm_tsc_scaling_ratio_frac_bits = 48;
7562 	}
7563 
7564 	set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
7565 
7566 	if (enable_ept)
7567 		vmx_enable_tdp();
7568 	else
7569 		kvm_disable_tdp();
7570 
7571 	/*
7572 	 * Only enable PML when hardware supports PML feature, and both EPT
7573 	 * and EPT A/D bit features are enabled -- PML depends on them to work.
7574 	 */
7575 	if (!enable_ept || !enable_ept_ad_bits || !cpu_has_vmx_pml())
7576 		enable_pml = 0;
7577 
7578 	if (!enable_pml) {
7579 		kvm_x86_ops->slot_enable_log_dirty = NULL;
7580 		kvm_x86_ops->slot_disable_log_dirty = NULL;
7581 		kvm_x86_ops->flush_log_dirty = NULL;
7582 		kvm_x86_ops->enable_log_dirty_pt_masked = NULL;
7583 	}
7584 
7585 	if (!cpu_has_vmx_preemption_timer())
7586 		enable_preemption_timer = false;
7587 
7588 	if (enable_preemption_timer) {
7589 		u64 use_timer_freq = 5000ULL * 1000 * 1000;
7590 		u64 vmx_msr;
7591 
7592 		rdmsrl(MSR_IA32_VMX_MISC, vmx_msr);
7593 		cpu_preemption_timer_multi =
7594 			vmx_msr & VMX_MISC_PREEMPTION_TIMER_RATE_MASK;
7595 
7596 		if (tsc_khz)
7597 			use_timer_freq = (u64)tsc_khz * 1000;
7598 		use_timer_freq >>= cpu_preemption_timer_multi;
7599 
7600 		/*
7601 		 * KVM "disables" the preemption timer by setting it to its max
7602 		 * value.  Don't use the timer if it might cause spurious exits
7603 		 * at a rate faster than 0.1 Hz (of uninterrupted guest time).
7604 		 */
7605 		if (use_timer_freq > 0xffffffffu / 10)
7606 			enable_preemption_timer = false;
7607 	}
7608 
7609 	if (!enable_preemption_timer) {
7610 		kvm_x86_ops->set_hv_timer = NULL;
7611 		kvm_x86_ops->cancel_hv_timer = NULL;
7612 		kvm_x86_ops->request_immediate_exit = __kvm_request_immediate_exit;
7613 	}
7614 
7615 	kvm_set_posted_intr_wakeup_handler(wakeup_handler);
7616 
7617 	kvm_mce_cap_supported |= MCG_LMCE_P;
7618 
7619 	if (pt_mode != PT_MODE_SYSTEM && pt_mode != PT_MODE_HOST_GUEST)
7620 		return -EINVAL;
7621 	if (!enable_ept || !cpu_has_vmx_intel_pt())
7622 		pt_mode = PT_MODE_SYSTEM;
7623 
7624 	if (nested) {
7625 		nested_vmx_setup_ctls_msrs(&vmcs_config.nested,
7626 					   vmx_capability.ept, enable_apicv);
7627 
7628 		r = nested_vmx_hardware_setup(kvm_vmx_exit_handlers);
7629 		if (r)
7630 			return r;
7631 	}
7632 
7633 	r = alloc_kvm_area();
7634 	if (r)
7635 		nested_vmx_hardware_unsetup();
7636 	return r;
7637 }
7638 
7639 static __exit void hardware_unsetup(void)
7640 {
7641 	if (nested)
7642 		nested_vmx_hardware_unsetup();
7643 
7644 	free_kvm_area();
7645 }
7646 
7647 static struct kvm_x86_ops vmx_x86_ops __ro_after_init = {
7648 	.cpu_has_kvm_support = cpu_has_kvm_support,
7649 	.disabled_by_bios = vmx_disabled_by_bios,
7650 	.hardware_setup = hardware_setup,
7651 	.hardware_unsetup = hardware_unsetup,
7652 	.check_processor_compatibility = vmx_check_processor_compat,
7653 	.hardware_enable = hardware_enable,
7654 	.hardware_disable = hardware_disable,
7655 	.cpu_has_accelerated_tpr = report_flexpriority,
7656 	.has_emulated_msr = vmx_has_emulated_msr,
7657 
7658 	.vm_init = vmx_vm_init,
7659 	.vm_alloc = vmx_vm_alloc,
7660 	.vm_free = vmx_vm_free,
7661 
7662 	.vcpu_create = vmx_create_vcpu,
7663 	.vcpu_free = vmx_free_vcpu,
7664 	.vcpu_reset = vmx_vcpu_reset,
7665 
7666 	.prepare_guest_switch = vmx_prepare_switch_to_guest,
7667 	.vcpu_load = vmx_vcpu_load,
7668 	.vcpu_put = vmx_vcpu_put,
7669 
7670 	.update_bp_intercept = update_exception_bitmap,
7671 	.get_msr_feature = vmx_get_msr_feature,
7672 	.get_msr = vmx_get_msr,
7673 	.set_msr = vmx_set_msr,
7674 	.get_segment_base = vmx_get_segment_base,
7675 	.get_segment = vmx_get_segment,
7676 	.set_segment = vmx_set_segment,
7677 	.get_cpl = vmx_get_cpl,
7678 	.get_cs_db_l_bits = vmx_get_cs_db_l_bits,
7679 	.decache_cr0_guest_bits = vmx_decache_cr0_guest_bits,
7680 	.decache_cr3 = vmx_decache_cr3,
7681 	.decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
7682 	.set_cr0 = vmx_set_cr0,
7683 	.set_cr3 = vmx_set_cr3,
7684 	.set_cr4 = vmx_set_cr4,
7685 	.set_efer = vmx_set_efer,
7686 	.get_idt = vmx_get_idt,
7687 	.set_idt = vmx_set_idt,
7688 	.get_gdt = vmx_get_gdt,
7689 	.set_gdt = vmx_set_gdt,
7690 	.get_dr6 = vmx_get_dr6,
7691 	.set_dr6 = vmx_set_dr6,
7692 	.set_dr7 = vmx_set_dr7,
7693 	.sync_dirty_debug_regs = vmx_sync_dirty_debug_regs,
7694 	.cache_reg = vmx_cache_reg,
7695 	.get_rflags = vmx_get_rflags,
7696 	.set_rflags = vmx_set_rflags,
7697 
7698 	.tlb_flush = vmx_flush_tlb,
7699 	.tlb_flush_gva = vmx_flush_tlb_gva,
7700 
7701 	.run = vmx_vcpu_run,
7702 	.handle_exit = vmx_handle_exit,
7703 	.skip_emulated_instruction = skip_emulated_instruction,
7704 	.set_interrupt_shadow = vmx_set_interrupt_shadow,
7705 	.get_interrupt_shadow = vmx_get_interrupt_shadow,
7706 	.patch_hypercall = vmx_patch_hypercall,
7707 	.set_irq = vmx_inject_irq,
7708 	.set_nmi = vmx_inject_nmi,
7709 	.queue_exception = vmx_queue_exception,
7710 	.cancel_injection = vmx_cancel_injection,
7711 	.interrupt_allowed = vmx_interrupt_allowed,
7712 	.nmi_allowed = vmx_nmi_allowed,
7713 	.get_nmi_mask = vmx_get_nmi_mask,
7714 	.set_nmi_mask = vmx_set_nmi_mask,
7715 	.enable_nmi_window = enable_nmi_window,
7716 	.enable_irq_window = enable_irq_window,
7717 	.update_cr8_intercept = update_cr8_intercept,
7718 	.set_virtual_apic_mode = vmx_set_virtual_apic_mode,
7719 	.set_apic_access_page_addr = vmx_set_apic_access_page_addr,
7720 	.get_enable_apicv = vmx_get_enable_apicv,
7721 	.refresh_apicv_exec_ctrl = vmx_refresh_apicv_exec_ctrl,
7722 	.load_eoi_exitmap = vmx_load_eoi_exitmap,
7723 	.apicv_post_state_restore = vmx_apicv_post_state_restore,
7724 	.hwapic_irr_update = vmx_hwapic_irr_update,
7725 	.hwapic_isr_update = vmx_hwapic_isr_update,
7726 	.guest_apic_has_interrupt = vmx_guest_apic_has_interrupt,
7727 	.sync_pir_to_irr = vmx_sync_pir_to_irr,
7728 	.deliver_posted_interrupt = vmx_deliver_posted_interrupt,
7729 
7730 	.set_tss_addr = vmx_set_tss_addr,
7731 	.set_identity_map_addr = vmx_set_identity_map_addr,
7732 	.get_tdp_level = get_ept_level,
7733 	.get_mt_mask = vmx_get_mt_mask,
7734 
7735 	.get_exit_info = vmx_get_exit_info,
7736 
7737 	.get_lpage_level = vmx_get_lpage_level,
7738 
7739 	.cpuid_update = vmx_cpuid_update,
7740 
7741 	.rdtscp_supported = vmx_rdtscp_supported,
7742 	.invpcid_supported = vmx_invpcid_supported,
7743 
7744 	.set_supported_cpuid = vmx_set_supported_cpuid,
7745 
7746 	.has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
7747 
7748 	.read_l1_tsc_offset = vmx_read_l1_tsc_offset,
7749 	.write_l1_tsc_offset = vmx_write_l1_tsc_offset,
7750 
7751 	.set_tdp_cr3 = vmx_set_cr3,
7752 
7753 	.check_intercept = vmx_check_intercept,
7754 	.handle_exit_irqoff = vmx_handle_exit_irqoff,
7755 	.mpx_supported = vmx_mpx_supported,
7756 	.xsaves_supported = vmx_xsaves_supported,
7757 	.umip_emulated = vmx_umip_emulated,
7758 	.pt_supported = vmx_pt_supported,
7759 
7760 	.request_immediate_exit = vmx_request_immediate_exit,
7761 
7762 	.sched_in = vmx_sched_in,
7763 
7764 	.slot_enable_log_dirty = vmx_slot_enable_log_dirty,
7765 	.slot_disable_log_dirty = vmx_slot_disable_log_dirty,
7766 	.flush_log_dirty = vmx_flush_log_dirty,
7767 	.enable_log_dirty_pt_masked = vmx_enable_log_dirty_pt_masked,
7768 	.write_log_dirty = vmx_write_pml_buffer,
7769 
7770 	.pre_block = vmx_pre_block,
7771 	.post_block = vmx_post_block,
7772 
7773 	.pmu_ops = &intel_pmu_ops,
7774 
7775 	.update_pi_irte = vmx_update_pi_irte,
7776 
7777 #ifdef CONFIG_X86_64
7778 	.set_hv_timer = vmx_set_hv_timer,
7779 	.cancel_hv_timer = vmx_cancel_hv_timer,
7780 #endif
7781 
7782 	.setup_mce = vmx_setup_mce,
7783 
7784 	.smi_allowed = vmx_smi_allowed,
7785 	.pre_enter_smm = vmx_pre_enter_smm,
7786 	.pre_leave_smm = vmx_pre_leave_smm,
7787 	.enable_smi_window = enable_smi_window,
7788 
7789 	.check_nested_events = NULL,
7790 	.get_nested_state = NULL,
7791 	.set_nested_state = NULL,
7792 	.get_vmcs12_pages = NULL,
7793 	.nested_enable_evmcs = NULL,
7794 	.need_emulation_on_page_fault = vmx_need_emulation_on_page_fault,
7795 };
7796 
7797 static void vmx_cleanup_l1d_flush(void)
7798 {
7799 	if (vmx_l1d_flush_pages) {
7800 		free_pages((unsigned long)vmx_l1d_flush_pages, L1D_CACHE_ORDER);
7801 		vmx_l1d_flush_pages = NULL;
7802 	}
7803 	/* Restore state so sysfs ignores VMX */
7804 	l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_AUTO;
7805 }
7806 
7807 static void vmx_exit(void)
7808 {
7809 #ifdef CONFIG_KEXEC_CORE
7810 	RCU_INIT_POINTER(crash_vmclear_loaded_vmcss, NULL);
7811 	synchronize_rcu();
7812 #endif
7813 
7814 	kvm_exit();
7815 
7816 #if IS_ENABLED(CONFIG_HYPERV)
7817 	if (static_branch_unlikely(&enable_evmcs)) {
7818 		int cpu;
7819 		struct hv_vp_assist_page *vp_ap;
7820 		/*
7821 		 * Reset everything to support using non-enlightened VMCS
7822 		 * access later (e.g. when we reload the module with
7823 		 * enlightened_vmcs=0)
7824 		 */
7825 		for_each_online_cpu(cpu) {
7826 			vp_ap =	hv_get_vp_assist_page(cpu);
7827 
7828 			if (!vp_ap)
7829 				continue;
7830 
7831 			vp_ap->current_nested_vmcs = 0;
7832 			vp_ap->enlighten_vmentry = 0;
7833 		}
7834 
7835 		static_branch_disable(&enable_evmcs);
7836 	}
7837 #endif
7838 	vmx_cleanup_l1d_flush();
7839 }
7840 module_exit(vmx_exit);
7841 
7842 static int __init vmx_init(void)
7843 {
7844 	int r;
7845 
7846 #if IS_ENABLED(CONFIG_HYPERV)
7847 	/*
7848 	 * Enlightened VMCS usage should be recommended and the host needs
7849 	 * to support eVMCS v1 or above. We can also disable eVMCS support
7850 	 * with module parameter.
7851 	 */
7852 	if (enlightened_vmcs &&
7853 	    ms_hyperv.hints & HV_X64_ENLIGHTENED_VMCS_RECOMMENDED &&
7854 	    (ms_hyperv.nested_features & HV_X64_ENLIGHTENED_VMCS_VERSION) >=
7855 	    KVM_EVMCS_VERSION) {
7856 		int cpu;
7857 
7858 		/* Check that we have assist pages on all online CPUs */
7859 		for_each_online_cpu(cpu) {
7860 			if (!hv_get_vp_assist_page(cpu)) {
7861 				enlightened_vmcs = false;
7862 				break;
7863 			}
7864 		}
7865 
7866 		if (enlightened_vmcs) {
7867 			pr_info("KVM: vmx: using Hyper-V Enlightened VMCS\n");
7868 			static_branch_enable(&enable_evmcs);
7869 		}
7870 	} else {
7871 		enlightened_vmcs = false;
7872 	}
7873 #endif
7874 
7875 	r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx),
7876 		     __alignof__(struct vcpu_vmx), THIS_MODULE);
7877 	if (r)
7878 		return r;
7879 
7880 	/*
7881 	 * Must be called after kvm_init() so enable_ept is properly set
7882 	 * up. Hand the parameter mitigation value in which was stored in
7883 	 * the pre module init parser. If no parameter was given, it will
7884 	 * contain 'auto' which will be turned into the default 'cond'
7885 	 * mitigation mode.
7886 	 */
7887 	if (boot_cpu_has(X86_BUG_L1TF)) {
7888 		r = vmx_setup_l1d_flush(vmentry_l1d_flush_param);
7889 		if (r) {
7890 			vmx_exit();
7891 			return r;
7892 		}
7893 	}
7894 
7895 #ifdef CONFIG_KEXEC_CORE
7896 	rcu_assign_pointer(crash_vmclear_loaded_vmcss,
7897 			   crash_vmclear_local_loaded_vmcss);
7898 #endif
7899 	vmx_check_vmcs12_offsets();
7900 
7901 	return 0;
7902 }
7903 module_init(vmx_init);
7904