xref: /openbmc/linux/arch/x86/kvm/vmx/vmx.c (revision 5b4cb650)
1 /*
2  * Kernel-based Virtual Machine driver for Linux
3  *
4  * This module enables machines with Intel VT-x extensions to run virtual
5  * machines without emulation or binary translation.
6  *
7  * Copyright (C) 2006 Qumranet, Inc.
8  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
9  *
10  * Authors:
11  *   Avi Kivity   <avi@qumranet.com>
12  *   Yaniv Kamay  <yaniv@qumranet.com>
13  *
14  * This work is licensed under the terms of the GNU GPL, version 2.  See
15  * the COPYING file in the top-level directory.
16  *
17  */
18 
19 #include <linux/frame.h>
20 #include <linux/highmem.h>
21 #include <linux/hrtimer.h>
22 #include <linux/kernel.h>
23 #include <linux/kvm_host.h>
24 #include <linux/module.h>
25 #include <linux/moduleparam.h>
26 #include <linux/mod_devicetable.h>
27 #include <linux/mm.h>
28 #include <linux/sched.h>
29 #include <linux/slab.h>
30 #include <linux/tboot.h>
31 #include <linux/trace_events.h>
32 
33 #include <asm/apic.h>
34 #include <asm/asm.h>
35 #include <asm/cpu.h>
36 #include <asm/debugreg.h>
37 #include <asm/desc.h>
38 #include <asm/fpu/internal.h>
39 #include <asm/io.h>
40 #include <asm/irq_remapping.h>
41 #include <asm/kexec.h>
42 #include <asm/perf_event.h>
43 #include <asm/mce.h>
44 #include <asm/mmu_context.h>
45 #include <asm/mshyperv.h>
46 #include <asm/spec-ctrl.h>
47 #include <asm/virtext.h>
48 #include <asm/vmx.h>
49 
50 #include "capabilities.h"
51 #include "cpuid.h"
52 #include "evmcs.h"
53 #include "irq.h"
54 #include "kvm_cache_regs.h"
55 #include "lapic.h"
56 #include "mmu.h"
57 #include "nested.h"
58 #include "ops.h"
59 #include "pmu.h"
60 #include "trace.h"
61 #include "vmcs.h"
62 #include "vmcs12.h"
63 #include "vmx.h"
64 #include "x86.h"
65 
66 MODULE_AUTHOR("Qumranet");
67 MODULE_LICENSE("GPL");
68 
69 static const struct x86_cpu_id vmx_cpu_id[] = {
70 	X86_FEATURE_MATCH(X86_FEATURE_VMX),
71 	{}
72 };
73 MODULE_DEVICE_TABLE(x86cpu, vmx_cpu_id);
74 
75 bool __read_mostly enable_vpid = 1;
76 module_param_named(vpid, enable_vpid, bool, 0444);
77 
78 static bool __read_mostly enable_vnmi = 1;
79 module_param_named(vnmi, enable_vnmi, bool, S_IRUGO);
80 
81 bool __read_mostly flexpriority_enabled = 1;
82 module_param_named(flexpriority, flexpriority_enabled, bool, S_IRUGO);
83 
84 bool __read_mostly enable_ept = 1;
85 module_param_named(ept, enable_ept, bool, S_IRUGO);
86 
87 bool __read_mostly enable_unrestricted_guest = 1;
88 module_param_named(unrestricted_guest,
89 			enable_unrestricted_guest, bool, S_IRUGO);
90 
91 bool __read_mostly enable_ept_ad_bits = 1;
92 module_param_named(eptad, enable_ept_ad_bits, bool, S_IRUGO);
93 
94 static bool __read_mostly emulate_invalid_guest_state = true;
95 module_param(emulate_invalid_guest_state, bool, S_IRUGO);
96 
97 static bool __read_mostly fasteoi = 1;
98 module_param(fasteoi, bool, S_IRUGO);
99 
100 static bool __read_mostly enable_apicv = 1;
101 module_param(enable_apicv, bool, S_IRUGO);
102 
103 /*
104  * If nested=1, nested virtualization is supported, i.e., guests may use
105  * VMX and be a hypervisor for its own guests. If nested=0, guests may not
106  * use VMX instructions.
107  */
108 static bool __read_mostly nested = 1;
109 module_param(nested, bool, S_IRUGO);
110 
111 static u64 __read_mostly host_xss;
112 
113 bool __read_mostly enable_pml = 1;
114 module_param_named(pml, enable_pml, bool, S_IRUGO);
115 
116 #define MSR_BITMAP_MODE_X2APIC		1
117 #define MSR_BITMAP_MODE_X2APIC_APICV	2
118 
119 #define KVM_VMX_TSC_MULTIPLIER_MAX     0xffffffffffffffffULL
120 
121 /* Guest_tsc -> host_tsc conversion requires 64-bit division.  */
122 static int __read_mostly cpu_preemption_timer_multi;
123 static bool __read_mostly enable_preemption_timer = 1;
124 #ifdef CONFIG_X86_64
125 module_param_named(preemption_timer, enable_preemption_timer, bool, S_IRUGO);
126 #endif
127 
128 #define KVM_VM_CR0_ALWAYS_OFF (X86_CR0_NW | X86_CR0_CD)
129 #define KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST X86_CR0_NE
130 #define KVM_VM_CR0_ALWAYS_ON				\
131 	(KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST | 	\
132 	 X86_CR0_WP | X86_CR0_PG | X86_CR0_PE)
133 #define KVM_CR4_GUEST_OWNED_BITS				      \
134 	(X86_CR4_PVI | X86_CR4_DE | X86_CR4_PCE | X86_CR4_OSFXSR      \
135 	 | X86_CR4_OSXMMEXCPT | X86_CR4_LA57 | X86_CR4_TSD)
136 
137 #define KVM_VM_CR4_ALWAYS_ON_UNRESTRICTED_GUEST X86_CR4_VMXE
138 #define KVM_PMODE_VM_CR4_ALWAYS_ON (X86_CR4_PAE | X86_CR4_VMXE)
139 #define KVM_RMODE_VM_CR4_ALWAYS_ON (X86_CR4_VME | X86_CR4_PAE | X86_CR4_VMXE)
140 
141 #define RMODE_GUEST_OWNED_EFLAGS_BITS (~(X86_EFLAGS_IOPL | X86_EFLAGS_VM))
142 
143 #define MSR_IA32_RTIT_STATUS_MASK (~(RTIT_STATUS_FILTEREN | \
144 	RTIT_STATUS_CONTEXTEN | RTIT_STATUS_TRIGGEREN | \
145 	RTIT_STATUS_ERROR | RTIT_STATUS_STOPPED | \
146 	RTIT_STATUS_BYTECNT))
147 
148 #define MSR_IA32_RTIT_OUTPUT_BASE_MASK \
149 	(~((1UL << cpuid_query_maxphyaddr(vcpu)) - 1) | 0x7f)
150 
151 /*
152  * These 2 parameters are used to config the controls for Pause-Loop Exiting:
153  * ple_gap:    upper bound on the amount of time between two successive
154  *             executions of PAUSE in a loop. Also indicate if ple enabled.
155  *             According to test, this time is usually smaller than 128 cycles.
156  * ple_window: upper bound on the amount of time a guest is allowed to execute
157  *             in a PAUSE loop. Tests indicate that most spinlocks are held for
158  *             less than 2^12 cycles
159  * Time is measured based on a counter that runs at the same rate as the TSC,
160  * refer SDM volume 3b section 21.6.13 & 22.1.3.
161  */
162 static unsigned int ple_gap = KVM_DEFAULT_PLE_GAP;
163 module_param(ple_gap, uint, 0444);
164 
165 static unsigned int ple_window = KVM_VMX_DEFAULT_PLE_WINDOW;
166 module_param(ple_window, uint, 0444);
167 
168 /* Default doubles per-vcpu window every exit. */
169 static unsigned int ple_window_grow = KVM_DEFAULT_PLE_WINDOW_GROW;
170 module_param(ple_window_grow, uint, 0444);
171 
172 /* Default resets per-vcpu window every exit to ple_window. */
173 static unsigned int ple_window_shrink = KVM_DEFAULT_PLE_WINDOW_SHRINK;
174 module_param(ple_window_shrink, uint, 0444);
175 
176 /* Default is to compute the maximum so we can never overflow. */
177 static unsigned int ple_window_max        = KVM_VMX_DEFAULT_PLE_WINDOW_MAX;
178 module_param(ple_window_max, uint, 0444);
179 
180 /* Default is SYSTEM mode, 1 for host-guest mode */
181 int __read_mostly pt_mode = PT_MODE_SYSTEM;
182 module_param(pt_mode, int, S_IRUGO);
183 
184 static DEFINE_STATIC_KEY_FALSE(vmx_l1d_should_flush);
185 static DEFINE_STATIC_KEY_FALSE(vmx_l1d_flush_cond);
186 static DEFINE_MUTEX(vmx_l1d_flush_mutex);
187 
188 /* Storage for pre module init parameter parsing */
189 static enum vmx_l1d_flush_state __read_mostly vmentry_l1d_flush_param = VMENTER_L1D_FLUSH_AUTO;
190 
191 static const struct {
192 	const char *option;
193 	bool for_parse;
194 } vmentry_l1d_param[] = {
195 	[VMENTER_L1D_FLUSH_AUTO]	 = {"auto", true},
196 	[VMENTER_L1D_FLUSH_NEVER]	 = {"never", true},
197 	[VMENTER_L1D_FLUSH_COND]	 = {"cond", true},
198 	[VMENTER_L1D_FLUSH_ALWAYS]	 = {"always", true},
199 	[VMENTER_L1D_FLUSH_EPT_DISABLED] = {"EPT disabled", false},
200 	[VMENTER_L1D_FLUSH_NOT_REQUIRED] = {"not required", false},
201 };
202 
203 #define L1D_CACHE_ORDER 4
204 static void *vmx_l1d_flush_pages;
205 
206 static int vmx_setup_l1d_flush(enum vmx_l1d_flush_state l1tf)
207 {
208 	struct page *page;
209 	unsigned int i;
210 
211 	if (!enable_ept) {
212 		l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_EPT_DISABLED;
213 		return 0;
214 	}
215 
216 	if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES)) {
217 		u64 msr;
218 
219 		rdmsrl(MSR_IA32_ARCH_CAPABILITIES, msr);
220 		if (msr & ARCH_CAP_SKIP_VMENTRY_L1DFLUSH) {
221 			l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_NOT_REQUIRED;
222 			return 0;
223 		}
224 	}
225 
226 	/* If set to auto use the default l1tf mitigation method */
227 	if (l1tf == VMENTER_L1D_FLUSH_AUTO) {
228 		switch (l1tf_mitigation) {
229 		case L1TF_MITIGATION_OFF:
230 			l1tf = VMENTER_L1D_FLUSH_NEVER;
231 			break;
232 		case L1TF_MITIGATION_FLUSH_NOWARN:
233 		case L1TF_MITIGATION_FLUSH:
234 		case L1TF_MITIGATION_FLUSH_NOSMT:
235 			l1tf = VMENTER_L1D_FLUSH_COND;
236 			break;
237 		case L1TF_MITIGATION_FULL:
238 		case L1TF_MITIGATION_FULL_FORCE:
239 			l1tf = VMENTER_L1D_FLUSH_ALWAYS;
240 			break;
241 		}
242 	} else if (l1tf_mitigation == L1TF_MITIGATION_FULL_FORCE) {
243 		l1tf = VMENTER_L1D_FLUSH_ALWAYS;
244 	}
245 
246 	if (l1tf != VMENTER_L1D_FLUSH_NEVER && !vmx_l1d_flush_pages &&
247 	    !boot_cpu_has(X86_FEATURE_FLUSH_L1D)) {
248 		page = alloc_pages(GFP_KERNEL, L1D_CACHE_ORDER);
249 		if (!page)
250 			return -ENOMEM;
251 		vmx_l1d_flush_pages = page_address(page);
252 
253 		/*
254 		 * Initialize each page with a different pattern in
255 		 * order to protect against KSM in the nested
256 		 * virtualization case.
257 		 */
258 		for (i = 0; i < 1u << L1D_CACHE_ORDER; ++i) {
259 			memset(vmx_l1d_flush_pages + i * PAGE_SIZE, i + 1,
260 			       PAGE_SIZE);
261 		}
262 	}
263 
264 	l1tf_vmx_mitigation = l1tf;
265 
266 	if (l1tf != VMENTER_L1D_FLUSH_NEVER)
267 		static_branch_enable(&vmx_l1d_should_flush);
268 	else
269 		static_branch_disable(&vmx_l1d_should_flush);
270 
271 	if (l1tf == VMENTER_L1D_FLUSH_COND)
272 		static_branch_enable(&vmx_l1d_flush_cond);
273 	else
274 		static_branch_disable(&vmx_l1d_flush_cond);
275 	return 0;
276 }
277 
278 static int vmentry_l1d_flush_parse(const char *s)
279 {
280 	unsigned int i;
281 
282 	if (s) {
283 		for (i = 0; i < ARRAY_SIZE(vmentry_l1d_param); i++) {
284 			if (vmentry_l1d_param[i].for_parse &&
285 			    sysfs_streq(s, vmentry_l1d_param[i].option))
286 				return i;
287 		}
288 	}
289 	return -EINVAL;
290 }
291 
292 static int vmentry_l1d_flush_set(const char *s, const struct kernel_param *kp)
293 {
294 	int l1tf, ret;
295 
296 	l1tf = vmentry_l1d_flush_parse(s);
297 	if (l1tf < 0)
298 		return l1tf;
299 
300 	if (!boot_cpu_has(X86_BUG_L1TF))
301 		return 0;
302 
303 	/*
304 	 * Has vmx_init() run already? If not then this is the pre init
305 	 * parameter parsing. In that case just store the value and let
306 	 * vmx_init() do the proper setup after enable_ept has been
307 	 * established.
308 	 */
309 	if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_AUTO) {
310 		vmentry_l1d_flush_param = l1tf;
311 		return 0;
312 	}
313 
314 	mutex_lock(&vmx_l1d_flush_mutex);
315 	ret = vmx_setup_l1d_flush(l1tf);
316 	mutex_unlock(&vmx_l1d_flush_mutex);
317 	return ret;
318 }
319 
320 static int vmentry_l1d_flush_get(char *s, const struct kernel_param *kp)
321 {
322 	if (WARN_ON_ONCE(l1tf_vmx_mitigation >= ARRAY_SIZE(vmentry_l1d_param)))
323 		return sprintf(s, "???\n");
324 
325 	return sprintf(s, "%s\n", vmentry_l1d_param[l1tf_vmx_mitigation].option);
326 }
327 
328 static const struct kernel_param_ops vmentry_l1d_flush_ops = {
329 	.set = vmentry_l1d_flush_set,
330 	.get = vmentry_l1d_flush_get,
331 };
332 module_param_cb(vmentry_l1d_flush, &vmentry_l1d_flush_ops, NULL, 0644);
333 
334 static bool guest_state_valid(struct kvm_vcpu *vcpu);
335 static u32 vmx_segment_access_rights(struct kvm_segment *var);
336 static __always_inline void vmx_disable_intercept_for_msr(unsigned long *msr_bitmap,
337 							  u32 msr, int type);
338 
339 void vmx_vmexit(void);
340 
341 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
342 DEFINE_PER_CPU(struct vmcs *, current_vmcs);
343 /*
344  * We maintain a per-CPU linked-list of VMCS loaded on that CPU. This is needed
345  * when a CPU is brought down, and we need to VMCLEAR all VMCSs loaded on it.
346  */
347 static DEFINE_PER_CPU(struct list_head, loaded_vmcss_on_cpu);
348 
349 /*
350  * We maintian a per-CPU linked-list of vCPU, so in wakeup_handler() we
351  * can find which vCPU should be waken up.
352  */
353 static DEFINE_PER_CPU(struct list_head, blocked_vcpu_on_cpu);
354 static DEFINE_PER_CPU(spinlock_t, blocked_vcpu_on_cpu_lock);
355 
356 static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
357 static DEFINE_SPINLOCK(vmx_vpid_lock);
358 
359 struct vmcs_config vmcs_config;
360 struct vmx_capability vmx_capability;
361 
362 #define VMX_SEGMENT_FIELD(seg)					\
363 	[VCPU_SREG_##seg] = {                                   \
364 		.selector = GUEST_##seg##_SELECTOR,		\
365 		.base = GUEST_##seg##_BASE,		   	\
366 		.limit = GUEST_##seg##_LIMIT,		   	\
367 		.ar_bytes = GUEST_##seg##_AR_BYTES,	   	\
368 	}
369 
370 static const struct kvm_vmx_segment_field {
371 	unsigned selector;
372 	unsigned base;
373 	unsigned limit;
374 	unsigned ar_bytes;
375 } kvm_vmx_segment_fields[] = {
376 	VMX_SEGMENT_FIELD(CS),
377 	VMX_SEGMENT_FIELD(DS),
378 	VMX_SEGMENT_FIELD(ES),
379 	VMX_SEGMENT_FIELD(FS),
380 	VMX_SEGMENT_FIELD(GS),
381 	VMX_SEGMENT_FIELD(SS),
382 	VMX_SEGMENT_FIELD(TR),
383 	VMX_SEGMENT_FIELD(LDTR),
384 };
385 
386 u64 host_efer;
387 
388 /*
389  * Though SYSCALL is only supported in 64-bit mode on Intel CPUs, kvm
390  * will emulate SYSCALL in legacy mode if the vendor string in guest
391  * CPUID.0:{EBX,ECX,EDX} is "AuthenticAMD" or "AMDisbetter!" To
392  * support this emulation, IA32_STAR must always be included in
393  * vmx_msr_index[], even in i386 builds.
394  */
395 const u32 vmx_msr_index[] = {
396 #ifdef CONFIG_X86_64
397 	MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
398 #endif
399 	MSR_EFER, MSR_TSC_AUX, MSR_STAR,
400 };
401 
402 #if IS_ENABLED(CONFIG_HYPERV)
403 static bool __read_mostly enlightened_vmcs = true;
404 module_param(enlightened_vmcs, bool, 0444);
405 
406 /* check_ept_pointer() should be under protection of ept_pointer_lock. */
407 static void check_ept_pointer_match(struct kvm *kvm)
408 {
409 	struct kvm_vcpu *vcpu;
410 	u64 tmp_eptp = INVALID_PAGE;
411 	int i;
412 
413 	kvm_for_each_vcpu(i, vcpu, kvm) {
414 		if (!VALID_PAGE(tmp_eptp)) {
415 			tmp_eptp = to_vmx(vcpu)->ept_pointer;
416 		} else if (tmp_eptp != to_vmx(vcpu)->ept_pointer) {
417 			to_kvm_vmx(kvm)->ept_pointers_match
418 				= EPT_POINTERS_MISMATCH;
419 			return;
420 		}
421 	}
422 
423 	to_kvm_vmx(kvm)->ept_pointers_match = EPT_POINTERS_MATCH;
424 }
425 
426 int kvm_fill_hv_flush_list_func(struct hv_guest_mapping_flush_list *flush,
427 		void *data)
428 {
429 	struct kvm_tlb_range *range = data;
430 
431 	return hyperv_fill_flush_guest_mapping_list(flush, range->start_gfn,
432 			range->pages);
433 }
434 
435 static inline int __hv_remote_flush_tlb_with_range(struct kvm *kvm,
436 		struct kvm_vcpu *vcpu, struct kvm_tlb_range *range)
437 {
438 	u64 ept_pointer = to_vmx(vcpu)->ept_pointer;
439 
440 	/*
441 	 * FLUSH_GUEST_PHYSICAL_ADDRESS_SPACE hypercall needs address
442 	 * of the base of EPT PML4 table, strip off EPT configuration
443 	 * information.
444 	 */
445 	if (range)
446 		return hyperv_flush_guest_mapping_range(ept_pointer & PAGE_MASK,
447 				kvm_fill_hv_flush_list_func, (void *)range);
448 	else
449 		return hyperv_flush_guest_mapping(ept_pointer & PAGE_MASK);
450 }
451 
452 static int hv_remote_flush_tlb_with_range(struct kvm *kvm,
453 		struct kvm_tlb_range *range)
454 {
455 	struct kvm_vcpu *vcpu;
456 	int ret = -ENOTSUPP, i;
457 
458 	spin_lock(&to_kvm_vmx(kvm)->ept_pointer_lock);
459 
460 	if (to_kvm_vmx(kvm)->ept_pointers_match == EPT_POINTERS_CHECK)
461 		check_ept_pointer_match(kvm);
462 
463 	if (to_kvm_vmx(kvm)->ept_pointers_match != EPT_POINTERS_MATCH) {
464 		kvm_for_each_vcpu(i, vcpu, kvm) {
465 			/* If ept_pointer is invalid pointer, bypass flush request. */
466 			if (VALID_PAGE(to_vmx(vcpu)->ept_pointer))
467 				ret |= __hv_remote_flush_tlb_with_range(
468 					kvm, vcpu, range);
469 		}
470 	} else {
471 		ret = __hv_remote_flush_tlb_with_range(kvm,
472 				kvm_get_vcpu(kvm, 0), range);
473 	}
474 
475 	spin_unlock(&to_kvm_vmx(kvm)->ept_pointer_lock);
476 	return ret;
477 }
478 static int hv_remote_flush_tlb(struct kvm *kvm)
479 {
480 	return hv_remote_flush_tlb_with_range(kvm, NULL);
481 }
482 
483 #endif /* IS_ENABLED(CONFIG_HYPERV) */
484 
485 /*
486  * Comment's format: document - errata name - stepping - processor name.
487  * Refer from
488  * https://www.virtualbox.org/svn/vbox/trunk/src/VBox/VMM/VMMR0/HMR0.cpp
489  */
490 static u32 vmx_preemption_cpu_tfms[] = {
491 /* 323344.pdf - BA86   - D0 - Xeon 7500 Series */
492 0x000206E6,
493 /* 323056.pdf - AAX65  - C2 - Xeon L3406 */
494 /* 322814.pdf - AAT59  - C2 - i7-600, i5-500, i5-400 and i3-300 Mobile */
495 /* 322911.pdf - AAU65  - C2 - i5-600, i3-500 Desktop and Pentium G6950 */
496 0x00020652,
497 /* 322911.pdf - AAU65  - K0 - i5-600, i3-500 Desktop and Pentium G6950 */
498 0x00020655,
499 /* 322373.pdf - AAO95  - B1 - Xeon 3400 Series */
500 /* 322166.pdf - AAN92  - B1 - i7-800 and i5-700 Desktop */
501 /*
502  * 320767.pdf - AAP86  - B1 -
503  * i7-900 Mobile Extreme, i7-800 and i7-700 Mobile
504  */
505 0x000106E5,
506 /* 321333.pdf - AAM126 - C0 - Xeon 3500 */
507 0x000106A0,
508 /* 321333.pdf - AAM126 - C1 - Xeon 3500 */
509 0x000106A1,
510 /* 320836.pdf - AAJ124 - C0 - i7-900 Desktop Extreme and i7-900 Desktop */
511 0x000106A4,
512  /* 321333.pdf - AAM126 - D0 - Xeon 3500 */
513  /* 321324.pdf - AAK139 - D0 - Xeon 5500 */
514  /* 320836.pdf - AAJ124 - D0 - i7-900 Extreme and i7-900 Desktop */
515 0x000106A5,
516  /* Xeon E3-1220 V2 */
517 0x000306A8,
518 };
519 
520 static inline bool cpu_has_broken_vmx_preemption_timer(void)
521 {
522 	u32 eax = cpuid_eax(0x00000001), i;
523 
524 	/* Clear the reserved bits */
525 	eax &= ~(0x3U << 14 | 0xfU << 28);
526 	for (i = 0; i < ARRAY_SIZE(vmx_preemption_cpu_tfms); i++)
527 		if (eax == vmx_preemption_cpu_tfms[i])
528 			return true;
529 
530 	return false;
531 }
532 
533 static inline bool cpu_need_virtualize_apic_accesses(struct kvm_vcpu *vcpu)
534 {
535 	return flexpriority_enabled && lapic_in_kernel(vcpu);
536 }
537 
538 static inline bool report_flexpriority(void)
539 {
540 	return flexpriority_enabled;
541 }
542 
543 static inline int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
544 {
545 	int i;
546 
547 	for (i = 0; i < vmx->nmsrs; ++i)
548 		if (vmx_msr_index[vmx->guest_msrs[i].index] == msr)
549 			return i;
550 	return -1;
551 }
552 
553 struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
554 {
555 	int i;
556 
557 	i = __find_msr_index(vmx, msr);
558 	if (i >= 0)
559 		return &vmx->guest_msrs[i];
560 	return NULL;
561 }
562 
563 void loaded_vmcs_init(struct loaded_vmcs *loaded_vmcs)
564 {
565 	vmcs_clear(loaded_vmcs->vmcs);
566 	if (loaded_vmcs->shadow_vmcs && loaded_vmcs->launched)
567 		vmcs_clear(loaded_vmcs->shadow_vmcs);
568 	loaded_vmcs->cpu = -1;
569 	loaded_vmcs->launched = 0;
570 }
571 
572 #ifdef CONFIG_KEXEC_CORE
573 /*
574  * This bitmap is used to indicate whether the vmclear
575  * operation is enabled on all cpus. All disabled by
576  * default.
577  */
578 static cpumask_t crash_vmclear_enabled_bitmap = CPU_MASK_NONE;
579 
580 static inline void crash_enable_local_vmclear(int cpu)
581 {
582 	cpumask_set_cpu(cpu, &crash_vmclear_enabled_bitmap);
583 }
584 
585 static inline void crash_disable_local_vmclear(int cpu)
586 {
587 	cpumask_clear_cpu(cpu, &crash_vmclear_enabled_bitmap);
588 }
589 
590 static inline int crash_local_vmclear_enabled(int cpu)
591 {
592 	return cpumask_test_cpu(cpu, &crash_vmclear_enabled_bitmap);
593 }
594 
595 static void crash_vmclear_local_loaded_vmcss(void)
596 {
597 	int cpu = raw_smp_processor_id();
598 	struct loaded_vmcs *v;
599 
600 	if (!crash_local_vmclear_enabled(cpu))
601 		return;
602 
603 	list_for_each_entry(v, &per_cpu(loaded_vmcss_on_cpu, cpu),
604 			    loaded_vmcss_on_cpu_link)
605 		vmcs_clear(v->vmcs);
606 }
607 #else
608 static inline void crash_enable_local_vmclear(int cpu) { }
609 static inline void crash_disable_local_vmclear(int cpu) { }
610 #endif /* CONFIG_KEXEC_CORE */
611 
612 static void __loaded_vmcs_clear(void *arg)
613 {
614 	struct loaded_vmcs *loaded_vmcs = arg;
615 	int cpu = raw_smp_processor_id();
616 
617 	if (loaded_vmcs->cpu != cpu)
618 		return; /* vcpu migration can race with cpu offline */
619 	if (per_cpu(current_vmcs, cpu) == loaded_vmcs->vmcs)
620 		per_cpu(current_vmcs, cpu) = NULL;
621 	crash_disable_local_vmclear(cpu);
622 	list_del(&loaded_vmcs->loaded_vmcss_on_cpu_link);
623 
624 	/*
625 	 * we should ensure updating loaded_vmcs->loaded_vmcss_on_cpu_link
626 	 * is before setting loaded_vmcs->vcpu to -1 which is done in
627 	 * loaded_vmcs_init. Otherwise, other cpu can see vcpu = -1 fist
628 	 * then adds the vmcs into percpu list before it is deleted.
629 	 */
630 	smp_wmb();
631 
632 	loaded_vmcs_init(loaded_vmcs);
633 	crash_enable_local_vmclear(cpu);
634 }
635 
636 void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs)
637 {
638 	int cpu = loaded_vmcs->cpu;
639 
640 	if (cpu != -1)
641 		smp_call_function_single(cpu,
642 			 __loaded_vmcs_clear, loaded_vmcs, 1);
643 }
644 
645 static bool vmx_segment_cache_test_set(struct vcpu_vmx *vmx, unsigned seg,
646 				       unsigned field)
647 {
648 	bool ret;
649 	u32 mask = 1 << (seg * SEG_FIELD_NR + field);
650 
651 	if (!(vmx->vcpu.arch.regs_avail & (1 << VCPU_EXREG_SEGMENTS))) {
652 		vmx->vcpu.arch.regs_avail |= (1 << VCPU_EXREG_SEGMENTS);
653 		vmx->segment_cache.bitmask = 0;
654 	}
655 	ret = vmx->segment_cache.bitmask & mask;
656 	vmx->segment_cache.bitmask |= mask;
657 	return ret;
658 }
659 
660 static u16 vmx_read_guest_seg_selector(struct vcpu_vmx *vmx, unsigned seg)
661 {
662 	u16 *p = &vmx->segment_cache.seg[seg].selector;
663 
664 	if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_SEL))
665 		*p = vmcs_read16(kvm_vmx_segment_fields[seg].selector);
666 	return *p;
667 }
668 
669 static ulong vmx_read_guest_seg_base(struct vcpu_vmx *vmx, unsigned seg)
670 {
671 	ulong *p = &vmx->segment_cache.seg[seg].base;
672 
673 	if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_BASE))
674 		*p = vmcs_readl(kvm_vmx_segment_fields[seg].base);
675 	return *p;
676 }
677 
678 static u32 vmx_read_guest_seg_limit(struct vcpu_vmx *vmx, unsigned seg)
679 {
680 	u32 *p = &vmx->segment_cache.seg[seg].limit;
681 
682 	if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_LIMIT))
683 		*p = vmcs_read32(kvm_vmx_segment_fields[seg].limit);
684 	return *p;
685 }
686 
687 static u32 vmx_read_guest_seg_ar(struct vcpu_vmx *vmx, unsigned seg)
688 {
689 	u32 *p = &vmx->segment_cache.seg[seg].ar;
690 
691 	if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_AR))
692 		*p = vmcs_read32(kvm_vmx_segment_fields[seg].ar_bytes);
693 	return *p;
694 }
695 
696 void update_exception_bitmap(struct kvm_vcpu *vcpu)
697 {
698 	u32 eb;
699 
700 	eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR) |
701 	     (1u << DB_VECTOR) | (1u << AC_VECTOR);
702 	/*
703 	 * Guest access to VMware backdoor ports could legitimately
704 	 * trigger #GP because of TSS I/O permission bitmap.
705 	 * We intercept those #GP and allow access to them anyway
706 	 * as VMware does.
707 	 */
708 	if (enable_vmware_backdoor)
709 		eb |= (1u << GP_VECTOR);
710 	if ((vcpu->guest_debug &
711 	     (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) ==
712 	    (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP))
713 		eb |= 1u << BP_VECTOR;
714 	if (to_vmx(vcpu)->rmode.vm86_active)
715 		eb = ~0;
716 	if (enable_ept)
717 		eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */
718 
719 	/* When we are running a nested L2 guest and L1 specified for it a
720 	 * certain exception bitmap, we must trap the same exceptions and pass
721 	 * them to L1. When running L2, we will only handle the exceptions
722 	 * specified above if L1 did not want them.
723 	 */
724 	if (is_guest_mode(vcpu))
725 		eb |= get_vmcs12(vcpu)->exception_bitmap;
726 
727 	vmcs_write32(EXCEPTION_BITMAP, eb);
728 }
729 
730 /*
731  * Check if MSR is intercepted for currently loaded MSR bitmap.
732  */
733 static bool msr_write_intercepted(struct kvm_vcpu *vcpu, u32 msr)
734 {
735 	unsigned long *msr_bitmap;
736 	int f = sizeof(unsigned long);
737 
738 	if (!cpu_has_vmx_msr_bitmap())
739 		return true;
740 
741 	msr_bitmap = to_vmx(vcpu)->loaded_vmcs->msr_bitmap;
742 
743 	if (msr <= 0x1fff) {
744 		return !!test_bit(msr, msr_bitmap + 0x800 / f);
745 	} else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
746 		msr &= 0x1fff;
747 		return !!test_bit(msr, msr_bitmap + 0xc00 / f);
748 	}
749 
750 	return true;
751 }
752 
753 static void clear_atomic_switch_msr_special(struct vcpu_vmx *vmx,
754 		unsigned long entry, unsigned long exit)
755 {
756 	vm_entry_controls_clearbit(vmx, entry);
757 	vm_exit_controls_clearbit(vmx, exit);
758 }
759 
760 static int find_msr(struct vmx_msrs *m, unsigned int msr)
761 {
762 	unsigned int i;
763 
764 	for (i = 0; i < m->nr; ++i) {
765 		if (m->val[i].index == msr)
766 			return i;
767 	}
768 	return -ENOENT;
769 }
770 
771 static void clear_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr)
772 {
773 	int i;
774 	struct msr_autoload *m = &vmx->msr_autoload;
775 
776 	switch (msr) {
777 	case MSR_EFER:
778 		if (cpu_has_load_ia32_efer()) {
779 			clear_atomic_switch_msr_special(vmx,
780 					VM_ENTRY_LOAD_IA32_EFER,
781 					VM_EXIT_LOAD_IA32_EFER);
782 			return;
783 		}
784 		break;
785 	case MSR_CORE_PERF_GLOBAL_CTRL:
786 		if (cpu_has_load_perf_global_ctrl()) {
787 			clear_atomic_switch_msr_special(vmx,
788 					VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
789 					VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
790 			return;
791 		}
792 		break;
793 	}
794 	i = find_msr(&m->guest, msr);
795 	if (i < 0)
796 		goto skip_guest;
797 	--m->guest.nr;
798 	m->guest.val[i] = m->guest.val[m->guest.nr];
799 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->guest.nr);
800 
801 skip_guest:
802 	i = find_msr(&m->host, msr);
803 	if (i < 0)
804 		return;
805 
806 	--m->host.nr;
807 	m->host.val[i] = m->host.val[m->host.nr];
808 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->host.nr);
809 }
810 
811 static void add_atomic_switch_msr_special(struct vcpu_vmx *vmx,
812 		unsigned long entry, unsigned long exit,
813 		unsigned long guest_val_vmcs, unsigned long host_val_vmcs,
814 		u64 guest_val, u64 host_val)
815 {
816 	vmcs_write64(guest_val_vmcs, guest_val);
817 	if (host_val_vmcs != HOST_IA32_EFER)
818 		vmcs_write64(host_val_vmcs, host_val);
819 	vm_entry_controls_setbit(vmx, entry);
820 	vm_exit_controls_setbit(vmx, exit);
821 }
822 
823 static void add_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr,
824 				  u64 guest_val, u64 host_val, bool entry_only)
825 {
826 	int i, j = 0;
827 	struct msr_autoload *m = &vmx->msr_autoload;
828 
829 	switch (msr) {
830 	case MSR_EFER:
831 		if (cpu_has_load_ia32_efer()) {
832 			add_atomic_switch_msr_special(vmx,
833 					VM_ENTRY_LOAD_IA32_EFER,
834 					VM_EXIT_LOAD_IA32_EFER,
835 					GUEST_IA32_EFER,
836 					HOST_IA32_EFER,
837 					guest_val, host_val);
838 			return;
839 		}
840 		break;
841 	case MSR_CORE_PERF_GLOBAL_CTRL:
842 		if (cpu_has_load_perf_global_ctrl()) {
843 			add_atomic_switch_msr_special(vmx,
844 					VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
845 					VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL,
846 					GUEST_IA32_PERF_GLOBAL_CTRL,
847 					HOST_IA32_PERF_GLOBAL_CTRL,
848 					guest_val, host_val);
849 			return;
850 		}
851 		break;
852 	case MSR_IA32_PEBS_ENABLE:
853 		/* PEBS needs a quiescent period after being disabled (to write
854 		 * a record).  Disabling PEBS through VMX MSR swapping doesn't
855 		 * provide that period, so a CPU could write host's record into
856 		 * guest's memory.
857 		 */
858 		wrmsrl(MSR_IA32_PEBS_ENABLE, 0);
859 	}
860 
861 	i = find_msr(&m->guest, msr);
862 	if (!entry_only)
863 		j = find_msr(&m->host, msr);
864 
865 	if (i == NR_AUTOLOAD_MSRS || j == NR_AUTOLOAD_MSRS) {
866 		printk_once(KERN_WARNING "Not enough msr switch entries. "
867 				"Can't add msr %x\n", msr);
868 		return;
869 	}
870 	if (i < 0) {
871 		i = m->guest.nr++;
872 		vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->guest.nr);
873 	}
874 	m->guest.val[i].index = msr;
875 	m->guest.val[i].value = guest_val;
876 
877 	if (entry_only)
878 		return;
879 
880 	if (j < 0) {
881 		j = m->host.nr++;
882 		vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->host.nr);
883 	}
884 	m->host.val[j].index = msr;
885 	m->host.val[j].value = host_val;
886 }
887 
888 static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset)
889 {
890 	u64 guest_efer = vmx->vcpu.arch.efer;
891 	u64 ignore_bits = 0;
892 
893 	if (!enable_ept) {
894 		/*
895 		 * NX is needed to handle CR0.WP=1, CR4.SMEP=1.  Testing
896 		 * host CPUID is more efficient than testing guest CPUID
897 		 * or CR4.  Host SMEP is anyway a requirement for guest SMEP.
898 		 */
899 		if (boot_cpu_has(X86_FEATURE_SMEP))
900 			guest_efer |= EFER_NX;
901 		else if (!(guest_efer & EFER_NX))
902 			ignore_bits |= EFER_NX;
903 	}
904 
905 	/*
906 	 * LMA and LME handled by hardware; SCE meaningless outside long mode.
907 	 */
908 	ignore_bits |= EFER_SCE;
909 #ifdef CONFIG_X86_64
910 	ignore_bits |= EFER_LMA | EFER_LME;
911 	/* SCE is meaningful only in long mode on Intel */
912 	if (guest_efer & EFER_LMA)
913 		ignore_bits &= ~(u64)EFER_SCE;
914 #endif
915 
916 	/*
917 	 * On EPT, we can't emulate NX, so we must switch EFER atomically.
918 	 * On CPUs that support "load IA32_EFER", always switch EFER
919 	 * atomically, since it's faster than switching it manually.
920 	 */
921 	if (cpu_has_load_ia32_efer() ||
922 	    (enable_ept && ((vmx->vcpu.arch.efer ^ host_efer) & EFER_NX))) {
923 		if (!(guest_efer & EFER_LMA))
924 			guest_efer &= ~EFER_LME;
925 		if (guest_efer != host_efer)
926 			add_atomic_switch_msr(vmx, MSR_EFER,
927 					      guest_efer, host_efer, false);
928 		else
929 			clear_atomic_switch_msr(vmx, MSR_EFER);
930 		return false;
931 	} else {
932 		clear_atomic_switch_msr(vmx, MSR_EFER);
933 
934 		guest_efer &= ~ignore_bits;
935 		guest_efer |= host_efer & ignore_bits;
936 
937 		vmx->guest_msrs[efer_offset].data = guest_efer;
938 		vmx->guest_msrs[efer_offset].mask = ~ignore_bits;
939 
940 		return true;
941 	}
942 }
943 
944 #ifdef CONFIG_X86_32
945 /*
946  * On 32-bit kernels, VM exits still load the FS and GS bases from the
947  * VMCS rather than the segment table.  KVM uses this helper to figure
948  * out the current bases to poke them into the VMCS before entry.
949  */
950 static unsigned long segment_base(u16 selector)
951 {
952 	struct desc_struct *table;
953 	unsigned long v;
954 
955 	if (!(selector & ~SEGMENT_RPL_MASK))
956 		return 0;
957 
958 	table = get_current_gdt_ro();
959 
960 	if ((selector & SEGMENT_TI_MASK) == SEGMENT_LDT) {
961 		u16 ldt_selector = kvm_read_ldt();
962 
963 		if (!(ldt_selector & ~SEGMENT_RPL_MASK))
964 			return 0;
965 
966 		table = (struct desc_struct *)segment_base(ldt_selector);
967 	}
968 	v = get_desc_base(&table[selector >> 3]);
969 	return v;
970 }
971 #endif
972 
973 static inline void pt_load_msr(struct pt_ctx *ctx, u32 addr_range)
974 {
975 	u32 i;
976 
977 	wrmsrl(MSR_IA32_RTIT_STATUS, ctx->status);
978 	wrmsrl(MSR_IA32_RTIT_OUTPUT_BASE, ctx->output_base);
979 	wrmsrl(MSR_IA32_RTIT_OUTPUT_MASK, ctx->output_mask);
980 	wrmsrl(MSR_IA32_RTIT_CR3_MATCH, ctx->cr3_match);
981 	for (i = 0; i < addr_range; i++) {
982 		wrmsrl(MSR_IA32_RTIT_ADDR0_A + i * 2, ctx->addr_a[i]);
983 		wrmsrl(MSR_IA32_RTIT_ADDR0_B + i * 2, ctx->addr_b[i]);
984 	}
985 }
986 
987 static inline void pt_save_msr(struct pt_ctx *ctx, u32 addr_range)
988 {
989 	u32 i;
990 
991 	rdmsrl(MSR_IA32_RTIT_STATUS, ctx->status);
992 	rdmsrl(MSR_IA32_RTIT_OUTPUT_BASE, ctx->output_base);
993 	rdmsrl(MSR_IA32_RTIT_OUTPUT_MASK, ctx->output_mask);
994 	rdmsrl(MSR_IA32_RTIT_CR3_MATCH, ctx->cr3_match);
995 	for (i = 0; i < addr_range; i++) {
996 		rdmsrl(MSR_IA32_RTIT_ADDR0_A + i * 2, ctx->addr_a[i]);
997 		rdmsrl(MSR_IA32_RTIT_ADDR0_B + i * 2, ctx->addr_b[i]);
998 	}
999 }
1000 
1001 static void pt_guest_enter(struct vcpu_vmx *vmx)
1002 {
1003 	if (pt_mode == PT_MODE_SYSTEM)
1004 		return;
1005 
1006 	/*
1007 	 * GUEST_IA32_RTIT_CTL is already set in the VMCS.
1008 	 * Save host state before VM entry.
1009 	 */
1010 	rdmsrl(MSR_IA32_RTIT_CTL, vmx->pt_desc.host.ctl);
1011 	if (vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) {
1012 		wrmsrl(MSR_IA32_RTIT_CTL, 0);
1013 		pt_save_msr(&vmx->pt_desc.host, vmx->pt_desc.addr_range);
1014 		pt_load_msr(&vmx->pt_desc.guest, vmx->pt_desc.addr_range);
1015 	}
1016 }
1017 
1018 static void pt_guest_exit(struct vcpu_vmx *vmx)
1019 {
1020 	if (pt_mode == PT_MODE_SYSTEM)
1021 		return;
1022 
1023 	if (vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) {
1024 		pt_save_msr(&vmx->pt_desc.guest, vmx->pt_desc.addr_range);
1025 		pt_load_msr(&vmx->pt_desc.host, vmx->pt_desc.addr_range);
1026 	}
1027 
1028 	/* Reload host state (IA32_RTIT_CTL will be cleared on VM exit). */
1029 	wrmsrl(MSR_IA32_RTIT_CTL, vmx->pt_desc.host.ctl);
1030 }
1031 
1032 void vmx_prepare_switch_to_guest(struct kvm_vcpu *vcpu)
1033 {
1034 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1035 	struct vmcs_host_state *host_state;
1036 #ifdef CONFIG_X86_64
1037 	int cpu = raw_smp_processor_id();
1038 #endif
1039 	unsigned long fs_base, gs_base;
1040 	u16 fs_sel, gs_sel;
1041 	int i;
1042 
1043 	vmx->req_immediate_exit = false;
1044 
1045 	/*
1046 	 * Note that guest MSRs to be saved/restored can also be changed
1047 	 * when guest state is loaded. This happens when guest transitions
1048 	 * to/from long-mode by setting MSR_EFER.LMA.
1049 	 */
1050 	if (!vmx->loaded_cpu_state || vmx->guest_msrs_dirty) {
1051 		vmx->guest_msrs_dirty = false;
1052 		for (i = 0; i < vmx->save_nmsrs; ++i)
1053 			kvm_set_shared_msr(vmx->guest_msrs[i].index,
1054 					   vmx->guest_msrs[i].data,
1055 					   vmx->guest_msrs[i].mask);
1056 
1057 	}
1058 
1059 	if (vmx->loaded_cpu_state)
1060 		return;
1061 
1062 	vmx->loaded_cpu_state = vmx->loaded_vmcs;
1063 	host_state = &vmx->loaded_cpu_state->host_state;
1064 
1065 	/*
1066 	 * Set host fs and gs selectors.  Unfortunately, 22.2.3 does not
1067 	 * allow segment selectors with cpl > 0 or ti == 1.
1068 	 */
1069 	host_state->ldt_sel = kvm_read_ldt();
1070 
1071 #ifdef CONFIG_X86_64
1072 	savesegment(ds, host_state->ds_sel);
1073 	savesegment(es, host_state->es_sel);
1074 
1075 	gs_base = cpu_kernelmode_gs_base(cpu);
1076 	if (likely(is_64bit_mm(current->mm))) {
1077 		save_fsgs_for_kvm();
1078 		fs_sel = current->thread.fsindex;
1079 		gs_sel = current->thread.gsindex;
1080 		fs_base = current->thread.fsbase;
1081 		vmx->msr_host_kernel_gs_base = current->thread.gsbase;
1082 	} else {
1083 		savesegment(fs, fs_sel);
1084 		savesegment(gs, gs_sel);
1085 		fs_base = read_msr(MSR_FS_BASE);
1086 		vmx->msr_host_kernel_gs_base = read_msr(MSR_KERNEL_GS_BASE);
1087 	}
1088 
1089 	wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1090 #else
1091 	savesegment(fs, fs_sel);
1092 	savesegment(gs, gs_sel);
1093 	fs_base = segment_base(fs_sel);
1094 	gs_base = segment_base(gs_sel);
1095 #endif
1096 
1097 	if (unlikely(fs_sel != host_state->fs_sel)) {
1098 		if (!(fs_sel & 7))
1099 			vmcs_write16(HOST_FS_SELECTOR, fs_sel);
1100 		else
1101 			vmcs_write16(HOST_FS_SELECTOR, 0);
1102 		host_state->fs_sel = fs_sel;
1103 	}
1104 	if (unlikely(gs_sel != host_state->gs_sel)) {
1105 		if (!(gs_sel & 7))
1106 			vmcs_write16(HOST_GS_SELECTOR, gs_sel);
1107 		else
1108 			vmcs_write16(HOST_GS_SELECTOR, 0);
1109 		host_state->gs_sel = gs_sel;
1110 	}
1111 	if (unlikely(fs_base != host_state->fs_base)) {
1112 		vmcs_writel(HOST_FS_BASE, fs_base);
1113 		host_state->fs_base = fs_base;
1114 	}
1115 	if (unlikely(gs_base != host_state->gs_base)) {
1116 		vmcs_writel(HOST_GS_BASE, gs_base);
1117 		host_state->gs_base = gs_base;
1118 	}
1119 }
1120 
1121 static void vmx_prepare_switch_to_host(struct vcpu_vmx *vmx)
1122 {
1123 	struct vmcs_host_state *host_state;
1124 
1125 	if (!vmx->loaded_cpu_state)
1126 		return;
1127 
1128 	WARN_ON_ONCE(vmx->loaded_cpu_state != vmx->loaded_vmcs);
1129 	host_state = &vmx->loaded_cpu_state->host_state;
1130 
1131 	++vmx->vcpu.stat.host_state_reload;
1132 	vmx->loaded_cpu_state = NULL;
1133 
1134 #ifdef CONFIG_X86_64
1135 	rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1136 #endif
1137 	if (host_state->ldt_sel || (host_state->gs_sel & 7)) {
1138 		kvm_load_ldt(host_state->ldt_sel);
1139 #ifdef CONFIG_X86_64
1140 		load_gs_index(host_state->gs_sel);
1141 #else
1142 		loadsegment(gs, host_state->gs_sel);
1143 #endif
1144 	}
1145 	if (host_state->fs_sel & 7)
1146 		loadsegment(fs, host_state->fs_sel);
1147 #ifdef CONFIG_X86_64
1148 	if (unlikely(host_state->ds_sel | host_state->es_sel)) {
1149 		loadsegment(ds, host_state->ds_sel);
1150 		loadsegment(es, host_state->es_sel);
1151 	}
1152 #endif
1153 	invalidate_tss_limit();
1154 #ifdef CONFIG_X86_64
1155 	wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
1156 #endif
1157 	load_fixmap_gdt(raw_smp_processor_id());
1158 }
1159 
1160 #ifdef CONFIG_X86_64
1161 static u64 vmx_read_guest_kernel_gs_base(struct vcpu_vmx *vmx)
1162 {
1163 	preempt_disable();
1164 	if (vmx->loaded_cpu_state)
1165 		rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1166 	preempt_enable();
1167 	return vmx->msr_guest_kernel_gs_base;
1168 }
1169 
1170 static void vmx_write_guest_kernel_gs_base(struct vcpu_vmx *vmx, u64 data)
1171 {
1172 	preempt_disable();
1173 	if (vmx->loaded_cpu_state)
1174 		wrmsrl(MSR_KERNEL_GS_BASE, data);
1175 	preempt_enable();
1176 	vmx->msr_guest_kernel_gs_base = data;
1177 }
1178 #endif
1179 
1180 static void vmx_vcpu_pi_load(struct kvm_vcpu *vcpu, int cpu)
1181 {
1182 	struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
1183 	struct pi_desc old, new;
1184 	unsigned int dest;
1185 
1186 	/*
1187 	 * In case of hot-plug or hot-unplug, we may have to undo
1188 	 * vmx_vcpu_pi_put even if there is no assigned device.  And we
1189 	 * always keep PI.NDST up to date for simplicity: it makes the
1190 	 * code easier, and CPU migration is not a fast path.
1191 	 */
1192 	if (!pi_test_sn(pi_desc) && vcpu->cpu == cpu)
1193 		return;
1194 
1195 	/*
1196 	 * First handle the simple case where no cmpxchg is necessary; just
1197 	 * allow posting non-urgent interrupts.
1198 	 *
1199 	 * If the 'nv' field is POSTED_INTR_WAKEUP_VECTOR, do not change
1200 	 * PI.NDST: pi_post_block will do it for us and the wakeup_handler
1201 	 * expects the VCPU to be on the blocked_vcpu_list that matches
1202 	 * PI.NDST.
1203 	 */
1204 	if (pi_desc->nv == POSTED_INTR_WAKEUP_VECTOR ||
1205 	    vcpu->cpu == cpu) {
1206 		pi_clear_sn(pi_desc);
1207 		return;
1208 	}
1209 
1210 	/* The full case.  */
1211 	do {
1212 		old.control = new.control = pi_desc->control;
1213 
1214 		dest = cpu_physical_id(cpu);
1215 
1216 		if (x2apic_enabled())
1217 			new.ndst = dest;
1218 		else
1219 			new.ndst = (dest << 8) & 0xFF00;
1220 
1221 		new.sn = 0;
1222 	} while (cmpxchg64(&pi_desc->control, old.control,
1223 			   new.control) != old.control);
1224 }
1225 
1226 /*
1227  * Switches to specified vcpu, until a matching vcpu_put(), but assumes
1228  * vcpu mutex is already taken.
1229  */
1230 void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1231 {
1232 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1233 	bool already_loaded = vmx->loaded_vmcs->cpu == cpu;
1234 
1235 	if (!already_loaded) {
1236 		loaded_vmcs_clear(vmx->loaded_vmcs);
1237 		local_irq_disable();
1238 		crash_disable_local_vmclear(cpu);
1239 
1240 		/*
1241 		 * Read loaded_vmcs->cpu should be before fetching
1242 		 * loaded_vmcs->loaded_vmcss_on_cpu_link.
1243 		 * See the comments in __loaded_vmcs_clear().
1244 		 */
1245 		smp_rmb();
1246 
1247 		list_add(&vmx->loaded_vmcs->loaded_vmcss_on_cpu_link,
1248 			 &per_cpu(loaded_vmcss_on_cpu, cpu));
1249 		crash_enable_local_vmclear(cpu);
1250 		local_irq_enable();
1251 	}
1252 
1253 	if (per_cpu(current_vmcs, cpu) != vmx->loaded_vmcs->vmcs) {
1254 		per_cpu(current_vmcs, cpu) = vmx->loaded_vmcs->vmcs;
1255 		vmcs_load(vmx->loaded_vmcs->vmcs);
1256 		indirect_branch_prediction_barrier();
1257 	}
1258 
1259 	if (!already_loaded) {
1260 		void *gdt = get_current_gdt_ro();
1261 		unsigned long sysenter_esp;
1262 
1263 		kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1264 
1265 		/*
1266 		 * Linux uses per-cpu TSS and GDT, so set these when switching
1267 		 * processors.  See 22.2.4.
1268 		 */
1269 		vmcs_writel(HOST_TR_BASE,
1270 			    (unsigned long)&get_cpu_entry_area(cpu)->tss.x86_tss);
1271 		vmcs_writel(HOST_GDTR_BASE, (unsigned long)gdt);   /* 22.2.4 */
1272 
1273 		/*
1274 		 * VM exits change the host TR limit to 0x67 after a VM
1275 		 * exit.  This is okay, since 0x67 covers everything except
1276 		 * the IO bitmap and have have code to handle the IO bitmap
1277 		 * being lost after a VM exit.
1278 		 */
1279 		BUILD_BUG_ON(IO_BITMAP_OFFSET - 1 != 0x67);
1280 
1281 		rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
1282 		vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
1283 
1284 		vmx->loaded_vmcs->cpu = cpu;
1285 	}
1286 
1287 	/* Setup TSC multiplier */
1288 	if (kvm_has_tsc_control &&
1289 	    vmx->current_tsc_ratio != vcpu->arch.tsc_scaling_ratio)
1290 		decache_tsc_multiplier(vmx);
1291 
1292 	vmx_vcpu_pi_load(vcpu, cpu);
1293 	vmx->host_pkru = read_pkru();
1294 	vmx->host_debugctlmsr = get_debugctlmsr();
1295 }
1296 
1297 static void vmx_vcpu_pi_put(struct kvm_vcpu *vcpu)
1298 {
1299 	struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
1300 
1301 	if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
1302 		!irq_remapping_cap(IRQ_POSTING_CAP)  ||
1303 		!kvm_vcpu_apicv_active(vcpu))
1304 		return;
1305 
1306 	/* Set SN when the vCPU is preempted */
1307 	if (vcpu->preempted)
1308 		pi_set_sn(pi_desc);
1309 }
1310 
1311 void vmx_vcpu_put(struct kvm_vcpu *vcpu)
1312 {
1313 	vmx_vcpu_pi_put(vcpu);
1314 
1315 	vmx_prepare_switch_to_host(to_vmx(vcpu));
1316 }
1317 
1318 static bool emulation_required(struct kvm_vcpu *vcpu)
1319 {
1320 	return emulate_invalid_guest_state && !guest_state_valid(vcpu);
1321 }
1322 
1323 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu);
1324 
1325 unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
1326 {
1327 	unsigned long rflags, save_rflags;
1328 
1329 	if (!test_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail)) {
1330 		__set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
1331 		rflags = vmcs_readl(GUEST_RFLAGS);
1332 		if (to_vmx(vcpu)->rmode.vm86_active) {
1333 			rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
1334 			save_rflags = to_vmx(vcpu)->rmode.save_rflags;
1335 			rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
1336 		}
1337 		to_vmx(vcpu)->rflags = rflags;
1338 	}
1339 	return to_vmx(vcpu)->rflags;
1340 }
1341 
1342 void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
1343 {
1344 	unsigned long old_rflags = vmx_get_rflags(vcpu);
1345 
1346 	__set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
1347 	to_vmx(vcpu)->rflags = rflags;
1348 	if (to_vmx(vcpu)->rmode.vm86_active) {
1349 		to_vmx(vcpu)->rmode.save_rflags = rflags;
1350 		rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
1351 	}
1352 	vmcs_writel(GUEST_RFLAGS, rflags);
1353 
1354 	if ((old_rflags ^ to_vmx(vcpu)->rflags) & X86_EFLAGS_VM)
1355 		to_vmx(vcpu)->emulation_required = emulation_required(vcpu);
1356 }
1357 
1358 u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu)
1359 {
1360 	u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1361 	int ret = 0;
1362 
1363 	if (interruptibility & GUEST_INTR_STATE_STI)
1364 		ret |= KVM_X86_SHADOW_INT_STI;
1365 	if (interruptibility & GUEST_INTR_STATE_MOV_SS)
1366 		ret |= KVM_X86_SHADOW_INT_MOV_SS;
1367 
1368 	return ret;
1369 }
1370 
1371 void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
1372 {
1373 	u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1374 	u32 interruptibility = interruptibility_old;
1375 
1376 	interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
1377 
1378 	if (mask & KVM_X86_SHADOW_INT_MOV_SS)
1379 		interruptibility |= GUEST_INTR_STATE_MOV_SS;
1380 	else if (mask & KVM_X86_SHADOW_INT_STI)
1381 		interruptibility |= GUEST_INTR_STATE_STI;
1382 
1383 	if ((interruptibility != interruptibility_old))
1384 		vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
1385 }
1386 
1387 static int vmx_rtit_ctl_check(struct kvm_vcpu *vcpu, u64 data)
1388 {
1389 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1390 	unsigned long value;
1391 
1392 	/*
1393 	 * Any MSR write that attempts to change bits marked reserved will
1394 	 * case a #GP fault.
1395 	 */
1396 	if (data & vmx->pt_desc.ctl_bitmask)
1397 		return 1;
1398 
1399 	/*
1400 	 * Any attempt to modify IA32_RTIT_CTL while TraceEn is set will
1401 	 * result in a #GP unless the same write also clears TraceEn.
1402 	 */
1403 	if ((vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) &&
1404 		((vmx->pt_desc.guest.ctl ^ data) & ~RTIT_CTL_TRACEEN))
1405 		return 1;
1406 
1407 	/*
1408 	 * WRMSR to IA32_RTIT_CTL that sets TraceEn but clears this bit
1409 	 * and FabricEn would cause #GP, if
1410 	 * CPUID.(EAX=14H, ECX=0):ECX.SNGLRGNOUT[bit 2] = 0
1411 	 */
1412 	if ((data & RTIT_CTL_TRACEEN) && !(data & RTIT_CTL_TOPA) &&
1413 		!(data & RTIT_CTL_FABRIC_EN) &&
1414 		!intel_pt_validate_cap(vmx->pt_desc.caps,
1415 					PT_CAP_single_range_output))
1416 		return 1;
1417 
1418 	/*
1419 	 * MTCFreq, CycThresh and PSBFreq encodings check, any MSR write that
1420 	 * utilize encodings marked reserved will casue a #GP fault.
1421 	 */
1422 	value = intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc_periods);
1423 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc) &&
1424 			!test_bit((data & RTIT_CTL_MTC_RANGE) >>
1425 			RTIT_CTL_MTC_RANGE_OFFSET, &value))
1426 		return 1;
1427 	value = intel_pt_validate_cap(vmx->pt_desc.caps,
1428 						PT_CAP_cycle_thresholds);
1429 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc) &&
1430 			!test_bit((data & RTIT_CTL_CYC_THRESH) >>
1431 			RTIT_CTL_CYC_THRESH_OFFSET, &value))
1432 		return 1;
1433 	value = intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_periods);
1434 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc) &&
1435 			!test_bit((data & RTIT_CTL_PSB_FREQ) >>
1436 			RTIT_CTL_PSB_FREQ_OFFSET, &value))
1437 		return 1;
1438 
1439 	/*
1440 	 * If ADDRx_CFG is reserved or the encodings is >2 will
1441 	 * cause a #GP fault.
1442 	 */
1443 	value = (data & RTIT_CTL_ADDR0) >> RTIT_CTL_ADDR0_OFFSET;
1444 	if ((value && (vmx->pt_desc.addr_range < 1)) || (value > 2))
1445 		return 1;
1446 	value = (data & RTIT_CTL_ADDR1) >> RTIT_CTL_ADDR1_OFFSET;
1447 	if ((value && (vmx->pt_desc.addr_range < 2)) || (value > 2))
1448 		return 1;
1449 	value = (data & RTIT_CTL_ADDR2) >> RTIT_CTL_ADDR2_OFFSET;
1450 	if ((value && (vmx->pt_desc.addr_range < 3)) || (value > 2))
1451 		return 1;
1452 	value = (data & RTIT_CTL_ADDR3) >> RTIT_CTL_ADDR3_OFFSET;
1453 	if ((value && (vmx->pt_desc.addr_range < 4)) || (value > 2))
1454 		return 1;
1455 
1456 	return 0;
1457 }
1458 
1459 
1460 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
1461 {
1462 	unsigned long rip;
1463 
1464 	rip = kvm_rip_read(vcpu);
1465 	rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
1466 	kvm_rip_write(vcpu, rip);
1467 
1468 	/* skipping an emulated instruction also counts */
1469 	vmx_set_interrupt_shadow(vcpu, 0);
1470 }
1471 
1472 static void vmx_clear_hlt(struct kvm_vcpu *vcpu)
1473 {
1474 	/*
1475 	 * Ensure that we clear the HLT state in the VMCS.  We don't need to
1476 	 * explicitly skip the instruction because if the HLT state is set,
1477 	 * then the instruction is already executing and RIP has already been
1478 	 * advanced.
1479 	 */
1480 	if (kvm_hlt_in_guest(vcpu->kvm) &&
1481 			vmcs_read32(GUEST_ACTIVITY_STATE) == GUEST_ACTIVITY_HLT)
1482 		vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
1483 }
1484 
1485 static void vmx_queue_exception(struct kvm_vcpu *vcpu)
1486 {
1487 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1488 	unsigned nr = vcpu->arch.exception.nr;
1489 	bool has_error_code = vcpu->arch.exception.has_error_code;
1490 	u32 error_code = vcpu->arch.exception.error_code;
1491 	u32 intr_info = nr | INTR_INFO_VALID_MASK;
1492 
1493 	kvm_deliver_exception_payload(vcpu);
1494 
1495 	if (has_error_code) {
1496 		vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
1497 		intr_info |= INTR_INFO_DELIVER_CODE_MASK;
1498 	}
1499 
1500 	if (vmx->rmode.vm86_active) {
1501 		int inc_eip = 0;
1502 		if (kvm_exception_is_soft(nr))
1503 			inc_eip = vcpu->arch.event_exit_inst_len;
1504 		if (kvm_inject_realmode_interrupt(vcpu, nr, inc_eip) != EMULATE_DONE)
1505 			kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
1506 		return;
1507 	}
1508 
1509 	WARN_ON_ONCE(vmx->emulation_required);
1510 
1511 	if (kvm_exception_is_soft(nr)) {
1512 		vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
1513 			     vmx->vcpu.arch.event_exit_inst_len);
1514 		intr_info |= INTR_TYPE_SOFT_EXCEPTION;
1515 	} else
1516 		intr_info |= INTR_TYPE_HARD_EXCEPTION;
1517 
1518 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
1519 
1520 	vmx_clear_hlt(vcpu);
1521 }
1522 
1523 static bool vmx_rdtscp_supported(void)
1524 {
1525 	return cpu_has_vmx_rdtscp();
1526 }
1527 
1528 static bool vmx_invpcid_supported(void)
1529 {
1530 	return cpu_has_vmx_invpcid();
1531 }
1532 
1533 /*
1534  * Swap MSR entry in host/guest MSR entry array.
1535  */
1536 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
1537 {
1538 	struct shared_msr_entry tmp;
1539 
1540 	tmp = vmx->guest_msrs[to];
1541 	vmx->guest_msrs[to] = vmx->guest_msrs[from];
1542 	vmx->guest_msrs[from] = tmp;
1543 }
1544 
1545 /*
1546  * Set up the vmcs to automatically save and restore system
1547  * msrs.  Don't touch the 64-bit msrs if the guest is in legacy
1548  * mode, as fiddling with msrs is very expensive.
1549  */
1550 static void setup_msrs(struct vcpu_vmx *vmx)
1551 {
1552 	int save_nmsrs, index;
1553 
1554 	save_nmsrs = 0;
1555 #ifdef CONFIG_X86_64
1556 	/*
1557 	 * The SYSCALL MSRs are only needed on long mode guests, and only
1558 	 * when EFER.SCE is set.
1559 	 */
1560 	if (is_long_mode(&vmx->vcpu) && (vmx->vcpu.arch.efer & EFER_SCE)) {
1561 		index = __find_msr_index(vmx, MSR_STAR);
1562 		if (index >= 0)
1563 			move_msr_up(vmx, index, save_nmsrs++);
1564 		index = __find_msr_index(vmx, MSR_LSTAR);
1565 		if (index >= 0)
1566 			move_msr_up(vmx, index, save_nmsrs++);
1567 		index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
1568 		if (index >= 0)
1569 			move_msr_up(vmx, index, save_nmsrs++);
1570 	}
1571 #endif
1572 	index = __find_msr_index(vmx, MSR_EFER);
1573 	if (index >= 0 && update_transition_efer(vmx, index))
1574 		move_msr_up(vmx, index, save_nmsrs++);
1575 	index = __find_msr_index(vmx, MSR_TSC_AUX);
1576 	if (index >= 0 && guest_cpuid_has(&vmx->vcpu, X86_FEATURE_RDTSCP))
1577 		move_msr_up(vmx, index, save_nmsrs++);
1578 
1579 	vmx->save_nmsrs = save_nmsrs;
1580 	vmx->guest_msrs_dirty = true;
1581 
1582 	if (cpu_has_vmx_msr_bitmap())
1583 		vmx_update_msr_bitmap(&vmx->vcpu);
1584 }
1585 
1586 static u64 vmx_read_l1_tsc_offset(struct kvm_vcpu *vcpu)
1587 {
1588 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1589 
1590 	if (is_guest_mode(vcpu) &&
1591 	    (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING))
1592 		return vcpu->arch.tsc_offset - vmcs12->tsc_offset;
1593 
1594 	return vcpu->arch.tsc_offset;
1595 }
1596 
1597 static u64 vmx_write_l1_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1598 {
1599 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1600 	u64 g_tsc_offset = 0;
1601 
1602 	/*
1603 	 * We're here if L1 chose not to trap WRMSR to TSC. According
1604 	 * to the spec, this should set L1's TSC; The offset that L1
1605 	 * set for L2 remains unchanged, and still needs to be added
1606 	 * to the newly set TSC to get L2's TSC.
1607 	 */
1608 	if (is_guest_mode(vcpu) &&
1609 	    (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING))
1610 		g_tsc_offset = vmcs12->tsc_offset;
1611 
1612 	trace_kvm_write_tsc_offset(vcpu->vcpu_id,
1613 				   vcpu->arch.tsc_offset - g_tsc_offset,
1614 				   offset);
1615 	vmcs_write64(TSC_OFFSET, offset + g_tsc_offset);
1616 	return offset + g_tsc_offset;
1617 }
1618 
1619 /*
1620  * nested_vmx_allowed() checks whether a guest should be allowed to use VMX
1621  * instructions and MSRs (i.e., nested VMX). Nested VMX is disabled for
1622  * all guests if the "nested" module option is off, and can also be disabled
1623  * for a single guest by disabling its VMX cpuid bit.
1624  */
1625 bool nested_vmx_allowed(struct kvm_vcpu *vcpu)
1626 {
1627 	return nested && guest_cpuid_has(vcpu, X86_FEATURE_VMX);
1628 }
1629 
1630 static inline bool vmx_feature_control_msr_valid(struct kvm_vcpu *vcpu,
1631 						 uint64_t val)
1632 {
1633 	uint64_t valid_bits = to_vmx(vcpu)->msr_ia32_feature_control_valid_bits;
1634 
1635 	return !(val & ~valid_bits);
1636 }
1637 
1638 static int vmx_get_msr_feature(struct kvm_msr_entry *msr)
1639 {
1640 	switch (msr->index) {
1641 	case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
1642 		if (!nested)
1643 			return 1;
1644 		return vmx_get_vmx_msr(&vmcs_config.nested, msr->index, &msr->data);
1645 	default:
1646 		return 1;
1647 	}
1648 
1649 	return 0;
1650 }
1651 
1652 /*
1653  * Reads an msr value (of 'msr_index') into 'pdata'.
1654  * Returns 0 on success, non-0 otherwise.
1655  * Assumes vcpu_load() was already called.
1656  */
1657 static int vmx_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1658 {
1659 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1660 	struct shared_msr_entry *msr;
1661 	u32 index;
1662 
1663 	switch (msr_info->index) {
1664 #ifdef CONFIG_X86_64
1665 	case MSR_FS_BASE:
1666 		msr_info->data = vmcs_readl(GUEST_FS_BASE);
1667 		break;
1668 	case MSR_GS_BASE:
1669 		msr_info->data = vmcs_readl(GUEST_GS_BASE);
1670 		break;
1671 	case MSR_KERNEL_GS_BASE:
1672 		msr_info->data = vmx_read_guest_kernel_gs_base(vmx);
1673 		break;
1674 #endif
1675 	case MSR_EFER:
1676 		return kvm_get_msr_common(vcpu, msr_info);
1677 	case MSR_IA32_SPEC_CTRL:
1678 		if (!msr_info->host_initiated &&
1679 		    !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
1680 			return 1;
1681 
1682 		msr_info->data = to_vmx(vcpu)->spec_ctrl;
1683 		break;
1684 	case MSR_IA32_ARCH_CAPABILITIES:
1685 		if (!msr_info->host_initiated &&
1686 		    !guest_cpuid_has(vcpu, X86_FEATURE_ARCH_CAPABILITIES))
1687 			return 1;
1688 		msr_info->data = to_vmx(vcpu)->arch_capabilities;
1689 		break;
1690 	case MSR_IA32_SYSENTER_CS:
1691 		msr_info->data = vmcs_read32(GUEST_SYSENTER_CS);
1692 		break;
1693 	case MSR_IA32_SYSENTER_EIP:
1694 		msr_info->data = vmcs_readl(GUEST_SYSENTER_EIP);
1695 		break;
1696 	case MSR_IA32_SYSENTER_ESP:
1697 		msr_info->data = vmcs_readl(GUEST_SYSENTER_ESP);
1698 		break;
1699 	case MSR_IA32_BNDCFGS:
1700 		if (!kvm_mpx_supported() ||
1701 		    (!msr_info->host_initiated &&
1702 		     !guest_cpuid_has(vcpu, X86_FEATURE_MPX)))
1703 			return 1;
1704 		msr_info->data = vmcs_read64(GUEST_BNDCFGS);
1705 		break;
1706 	case MSR_IA32_MCG_EXT_CTL:
1707 		if (!msr_info->host_initiated &&
1708 		    !(vmx->msr_ia32_feature_control &
1709 		      FEATURE_CONTROL_LMCE))
1710 			return 1;
1711 		msr_info->data = vcpu->arch.mcg_ext_ctl;
1712 		break;
1713 	case MSR_IA32_FEATURE_CONTROL:
1714 		msr_info->data = vmx->msr_ia32_feature_control;
1715 		break;
1716 	case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
1717 		if (!nested_vmx_allowed(vcpu))
1718 			return 1;
1719 		return vmx_get_vmx_msr(&vmx->nested.msrs, msr_info->index,
1720 				       &msr_info->data);
1721 	case MSR_IA32_XSS:
1722 		if (!vmx_xsaves_supported())
1723 			return 1;
1724 		msr_info->data = vcpu->arch.ia32_xss;
1725 		break;
1726 	case MSR_IA32_RTIT_CTL:
1727 		if (pt_mode != PT_MODE_HOST_GUEST)
1728 			return 1;
1729 		msr_info->data = vmx->pt_desc.guest.ctl;
1730 		break;
1731 	case MSR_IA32_RTIT_STATUS:
1732 		if (pt_mode != PT_MODE_HOST_GUEST)
1733 			return 1;
1734 		msr_info->data = vmx->pt_desc.guest.status;
1735 		break;
1736 	case MSR_IA32_RTIT_CR3_MATCH:
1737 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
1738 			!intel_pt_validate_cap(vmx->pt_desc.caps,
1739 						PT_CAP_cr3_filtering))
1740 			return 1;
1741 		msr_info->data = vmx->pt_desc.guest.cr3_match;
1742 		break;
1743 	case MSR_IA32_RTIT_OUTPUT_BASE:
1744 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
1745 			(!intel_pt_validate_cap(vmx->pt_desc.caps,
1746 					PT_CAP_topa_output) &&
1747 			 !intel_pt_validate_cap(vmx->pt_desc.caps,
1748 					PT_CAP_single_range_output)))
1749 			return 1;
1750 		msr_info->data = vmx->pt_desc.guest.output_base;
1751 		break;
1752 	case MSR_IA32_RTIT_OUTPUT_MASK:
1753 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
1754 			(!intel_pt_validate_cap(vmx->pt_desc.caps,
1755 					PT_CAP_topa_output) &&
1756 			 !intel_pt_validate_cap(vmx->pt_desc.caps,
1757 					PT_CAP_single_range_output)))
1758 			return 1;
1759 		msr_info->data = vmx->pt_desc.guest.output_mask;
1760 		break;
1761 	case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
1762 		index = msr_info->index - MSR_IA32_RTIT_ADDR0_A;
1763 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
1764 			(index >= 2 * intel_pt_validate_cap(vmx->pt_desc.caps,
1765 					PT_CAP_num_address_ranges)))
1766 			return 1;
1767 		if (index % 2)
1768 			msr_info->data = vmx->pt_desc.guest.addr_b[index / 2];
1769 		else
1770 			msr_info->data = vmx->pt_desc.guest.addr_a[index / 2];
1771 		break;
1772 	case MSR_TSC_AUX:
1773 		if (!msr_info->host_initiated &&
1774 		    !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP))
1775 			return 1;
1776 		/* Otherwise falls through */
1777 	default:
1778 		msr = find_msr_entry(vmx, msr_info->index);
1779 		if (msr) {
1780 			msr_info->data = msr->data;
1781 			break;
1782 		}
1783 		return kvm_get_msr_common(vcpu, msr_info);
1784 	}
1785 
1786 	return 0;
1787 }
1788 
1789 /*
1790  * Writes msr value into into the appropriate "register".
1791  * Returns 0 on success, non-0 otherwise.
1792  * Assumes vcpu_load() was already called.
1793  */
1794 static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1795 {
1796 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1797 	struct shared_msr_entry *msr;
1798 	int ret = 0;
1799 	u32 msr_index = msr_info->index;
1800 	u64 data = msr_info->data;
1801 	u32 index;
1802 
1803 	switch (msr_index) {
1804 	case MSR_EFER:
1805 		ret = kvm_set_msr_common(vcpu, msr_info);
1806 		break;
1807 #ifdef CONFIG_X86_64
1808 	case MSR_FS_BASE:
1809 		vmx_segment_cache_clear(vmx);
1810 		vmcs_writel(GUEST_FS_BASE, data);
1811 		break;
1812 	case MSR_GS_BASE:
1813 		vmx_segment_cache_clear(vmx);
1814 		vmcs_writel(GUEST_GS_BASE, data);
1815 		break;
1816 	case MSR_KERNEL_GS_BASE:
1817 		vmx_write_guest_kernel_gs_base(vmx, data);
1818 		break;
1819 #endif
1820 	case MSR_IA32_SYSENTER_CS:
1821 		vmcs_write32(GUEST_SYSENTER_CS, data);
1822 		break;
1823 	case MSR_IA32_SYSENTER_EIP:
1824 		vmcs_writel(GUEST_SYSENTER_EIP, data);
1825 		break;
1826 	case MSR_IA32_SYSENTER_ESP:
1827 		vmcs_writel(GUEST_SYSENTER_ESP, data);
1828 		break;
1829 	case MSR_IA32_BNDCFGS:
1830 		if (!kvm_mpx_supported() ||
1831 		    (!msr_info->host_initiated &&
1832 		     !guest_cpuid_has(vcpu, X86_FEATURE_MPX)))
1833 			return 1;
1834 		if (is_noncanonical_address(data & PAGE_MASK, vcpu) ||
1835 		    (data & MSR_IA32_BNDCFGS_RSVD))
1836 			return 1;
1837 		vmcs_write64(GUEST_BNDCFGS, data);
1838 		break;
1839 	case MSR_IA32_SPEC_CTRL:
1840 		if (!msr_info->host_initiated &&
1841 		    !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
1842 			return 1;
1843 
1844 		/* The STIBP bit doesn't fault even if it's not advertised */
1845 		if (data & ~(SPEC_CTRL_IBRS | SPEC_CTRL_STIBP | SPEC_CTRL_SSBD))
1846 			return 1;
1847 
1848 		vmx->spec_ctrl = data;
1849 
1850 		if (!data)
1851 			break;
1852 
1853 		/*
1854 		 * For non-nested:
1855 		 * When it's written (to non-zero) for the first time, pass
1856 		 * it through.
1857 		 *
1858 		 * For nested:
1859 		 * The handling of the MSR bitmap for L2 guests is done in
1860 		 * nested_vmx_merge_msr_bitmap. We should not touch the
1861 		 * vmcs02.msr_bitmap here since it gets completely overwritten
1862 		 * in the merging. We update the vmcs01 here for L1 as well
1863 		 * since it will end up touching the MSR anyway now.
1864 		 */
1865 		vmx_disable_intercept_for_msr(vmx->vmcs01.msr_bitmap,
1866 					      MSR_IA32_SPEC_CTRL,
1867 					      MSR_TYPE_RW);
1868 		break;
1869 	case MSR_IA32_PRED_CMD:
1870 		if (!msr_info->host_initiated &&
1871 		    !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
1872 			return 1;
1873 
1874 		if (data & ~PRED_CMD_IBPB)
1875 			return 1;
1876 
1877 		if (!data)
1878 			break;
1879 
1880 		wrmsrl(MSR_IA32_PRED_CMD, PRED_CMD_IBPB);
1881 
1882 		/*
1883 		 * For non-nested:
1884 		 * When it's written (to non-zero) for the first time, pass
1885 		 * it through.
1886 		 *
1887 		 * For nested:
1888 		 * The handling of the MSR bitmap for L2 guests is done in
1889 		 * nested_vmx_merge_msr_bitmap. We should not touch the
1890 		 * vmcs02.msr_bitmap here since it gets completely overwritten
1891 		 * in the merging.
1892 		 */
1893 		vmx_disable_intercept_for_msr(vmx->vmcs01.msr_bitmap, MSR_IA32_PRED_CMD,
1894 					      MSR_TYPE_W);
1895 		break;
1896 	case MSR_IA32_ARCH_CAPABILITIES:
1897 		if (!msr_info->host_initiated)
1898 			return 1;
1899 		vmx->arch_capabilities = data;
1900 		break;
1901 	case MSR_IA32_CR_PAT:
1902 		if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
1903 			if (!kvm_mtrr_valid(vcpu, MSR_IA32_CR_PAT, data))
1904 				return 1;
1905 			vmcs_write64(GUEST_IA32_PAT, data);
1906 			vcpu->arch.pat = data;
1907 			break;
1908 		}
1909 		ret = kvm_set_msr_common(vcpu, msr_info);
1910 		break;
1911 	case MSR_IA32_TSC_ADJUST:
1912 		ret = kvm_set_msr_common(vcpu, msr_info);
1913 		break;
1914 	case MSR_IA32_MCG_EXT_CTL:
1915 		if ((!msr_info->host_initiated &&
1916 		     !(to_vmx(vcpu)->msr_ia32_feature_control &
1917 		       FEATURE_CONTROL_LMCE)) ||
1918 		    (data & ~MCG_EXT_CTL_LMCE_EN))
1919 			return 1;
1920 		vcpu->arch.mcg_ext_ctl = data;
1921 		break;
1922 	case MSR_IA32_FEATURE_CONTROL:
1923 		if (!vmx_feature_control_msr_valid(vcpu, data) ||
1924 		    (to_vmx(vcpu)->msr_ia32_feature_control &
1925 		     FEATURE_CONTROL_LOCKED && !msr_info->host_initiated))
1926 			return 1;
1927 		vmx->msr_ia32_feature_control = data;
1928 		if (msr_info->host_initiated && data == 0)
1929 			vmx_leave_nested(vcpu);
1930 		break;
1931 	case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
1932 		if (!msr_info->host_initiated)
1933 			return 1; /* they are read-only */
1934 		if (!nested_vmx_allowed(vcpu))
1935 			return 1;
1936 		return vmx_set_vmx_msr(vcpu, msr_index, data);
1937 	case MSR_IA32_XSS:
1938 		if (!vmx_xsaves_supported())
1939 			return 1;
1940 		/*
1941 		 * The only supported bit as of Skylake is bit 8, but
1942 		 * it is not supported on KVM.
1943 		 */
1944 		if (data != 0)
1945 			return 1;
1946 		vcpu->arch.ia32_xss = data;
1947 		if (vcpu->arch.ia32_xss != host_xss)
1948 			add_atomic_switch_msr(vmx, MSR_IA32_XSS,
1949 				vcpu->arch.ia32_xss, host_xss, false);
1950 		else
1951 			clear_atomic_switch_msr(vmx, MSR_IA32_XSS);
1952 		break;
1953 	case MSR_IA32_RTIT_CTL:
1954 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
1955 			vmx_rtit_ctl_check(vcpu, data) ||
1956 			vmx->nested.vmxon)
1957 			return 1;
1958 		vmcs_write64(GUEST_IA32_RTIT_CTL, data);
1959 		vmx->pt_desc.guest.ctl = data;
1960 		pt_update_intercept_for_msr(vmx);
1961 		break;
1962 	case MSR_IA32_RTIT_STATUS:
1963 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
1964 			(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) ||
1965 			(data & MSR_IA32_RTIT_STATUS_MASK))
1966 			return 1;
1967 		vmx->pt_desc.guest.status = data;
1968 		break;
1969 	case MSR_IA32_RTIT_CR3_MATCH:
1970 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
1971 			(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) ||
1972 			!intel_pt_validate_cap(vmx->pt_desc.caps,
1973 						PT_CAP_cr3_filtering))
1974 			return 1;
1975 		vmx->pt_desc.guest.cr3_match = data;
1976 		break;
1977 	case MSR_IA32_RTIT_OUTPUT_BASE:
1978 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
1979 			(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) ||
1980 			(!intel_pt_validate_cap(vmx->pt_desc.caps,
1981 					PT_CAP_topa_output) &&
1982 			 !intel_pt_validate_cap(vmx->pt_desc.caps,
1983 					PT_CAP_single_range_output)) ||
1984 			(data & MSR_IA32_RTIT_OUTPUT_BASE_MASK))
1985 			return 1;
1986 		vmx->pt_desc.guest.output_base = data;
1987 		break;
1988 	case MSR_IA32_RTIT_OUTPUT_MASK:
1989 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
1990 			(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) ||
1991 			(!intel_pt_validate_cap(vmx->pt_desc.caps,
1992 					PT_CAP_topa_output) &&
1993 			 !intel_pt_validate_cap(vmx->pt_desc.caps,
1994 					PT_CAP_single_range_output)))
1995 			return 1;
1996 		vmx->pt_desc.guest.output_mask = data;
1997 		break;
1998 	case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
1999 		index = msr_info->index - MSR_IA32_RTIT_ADDR0_A;
2000 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
2001 			(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) ||
2002 			(index >= 2 * intel_pt_validate_cap(vmx->pt_desc.caps,
2003 					PT_CAP_num_address_ranges)))
2004 			return 1;
2005 		if (index % 2)
2006 			vmx->pt_desc.guest.addr_b[index / 2] = data;
2007 		else
2008 			vmx->pt_desc.guest.addr_a[index / 2] = data;
2009 		break;
2010 	case MSR_TSC_AUX:
2011 		if (!msr_info->host_initiated &&
2012 		    !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP))
2013 			return 1;
2014 		/* Check reserved bit, higher 32 bits should be zero */
2015 		if ((data >> 32) != 0)
2016 			return 1;
2017 		/* Otherwise falls through */
2018 	default:
2019 		msr = find_msr_entry(vmx, msr_index);
2020 		if (msr) {
2021 			u64 old_msr_data = msr->data;
2022 			msr->data = data;
2023 			if (msr - vmx->guest_msrs < vmx->save_nmsrs) {
2024 				preempt_disable();
2025 				ret = kvm_set_shared_msr(msr->index, msr->data,
2026 							 msr->mask);
2027 				preempt_enable();
2028 				if (ret)
2029 					msr->data = old_msr_data;
2030 			}
2031 			break;
2032 		}
2033 		ret = kvm_set_msr_common(vcpu, msr_info);
2034 	}
2035 
2036 	return ret;
2037 }
2038 
2039 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
2040 {
2041 	__set_bit(reg, (unsigned long *)&vcpu->arch.regs_avail);
2042 	switch (reg) {
2043 	case VCPU_REGS_RSP:
2044 		vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
2045 		break;
2046 	case VCPU_REGS_RIP:
2047 		vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
2048 		break;
2049 	case VCPU_EXREG_PDPTR:
2050 		if (enable_ept)
2051 			ept_save_pdptrs(vcpu);
2052 		break;
2053 	default:
2054 		break;
2055 	}
2056 }
2057 
2058 static __init int cpu_has_kvm_support(void)
2059 {
2060 	return cpu_has_vmx();
2061 }
2062 
2063 static __init int vmx_disabled_by_bios(void)
2064 {
2065 	u64 msr;
2066 
2067 	rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
2068 	if (msr & FEATURE_CONTROL_LOCKED) {
2069 		/* launched w/ TXT and VMX disabled */
2070 		if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
2071 			&& tboot_enabled())
2072 			return 1;
2073 		/* launched w/o TXT and VMX only enabled w/ TXT */
2074 		if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
2075 			&& (msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
2076 			&& !tboot_enabled()) {
2077 			printk(KERN_WARNING "kvm: disable TXT in the BIOS or "
2078 				"activate TXT before enabling KVM\n");
2079 			return 1;
2080 		}
2081 		/* launched w/o TXT and VMX disabled */
2082 		if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
2083 			&& !tboot_enabled())
2084 			return 1;
2085 	}
2086 
2087 	return 0;
2088 }
2089 
2090 static void kvm_cpu_vmxon(u64 addr)
2091 {
2092 	cr4_set_bits(X86_CR4_VMXE);
2093 	intel_pt_handle_vmx(1);
2094 
2095 	asm volatile ("vmxon %0" : : "m"(addr));
2096 }
2097 
2098 static int hardware_enable(void)
2099 {
2100 	int cpu = raw_smp_processor_id();
2101 	u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
2102 	u64 old, test_bits;
2103 
2104 	if (cr4_read_shadow() & X86_CR4_VMXE)
2105 		return -EBUSY;
2106 
2107 	/*
2108 	 * This can happen if we hot-added a CPU but failed to allocate
2109 	 * VP assist page for it.
2110 	 */
2111 	if (static_branch_unlikely(&enable_evmcs) &&
2112 	    !hv_get_vp_assist_page(cpu))
2113 		return -EFAULT;
2114 
2115 	INIT_LIST_HEAD(&per_cpu(loaded_vmcss_on_cpu, cpu));
2116 	INIT_LIST_HEAD(&per_cpu(blocked_vcpu_on_cpu, cpu));
2117 	spin_lock_init(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
2118 
2119 	/*
2120 	 * Now we can enable the vmclear operation in kdump
2121 	 * since the loaded_vmcss_on_cpu list on this cpu
2122 	 * has been initialized.
2123 	 *
2124 	 * Though the cpu is not in VMX operation now, there
2125 	 * is no problem to enable the vmclear operation
2126 	 * for the loaded_vmcss_on_cpu list is empty!
2127 	 */
2128 	crash_enable_local_vmclear(cpu);
2129 
2130 	rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
2131 
2132 	test_bits = FEATURE_CONTROL_LOCKED;
2133 	test_bits |= FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
2134 	if (tboot_enabled())
2135 		test_bits |= FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX;
2136 
2137 	if ((old & test_bits) != test_bits) {
2138 		/* enable and lock */
2139 		wrmsrl(MSR_IA32_FEATURE_CONTROL, old | test_bits);
2140 	}
2141 	kvm_cpu_vmxon(phys_addr);
2142 	if (enable_ept)
2143 		ept_sync_global();
2144 
2145 	return 0;
2146 }
2147 
2148 static void vmclear_local_loaded_vmcss(void)
2149 {
2150 	int cpu = raw_smp_processor_id();
2151 	struct loaded_vmcs *v, *n;
2152 
2153 	list_for_each_entry_safe(v, n, &per_cpu(loaded_vmcss_on_cpu, cpu),
2154 				 loaded_vmcss_on_cpu_link)
2155 		__loaded_vmcs_clear(v);
2156 }
2157 
2158 
2159 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
2160  * tricks.
2161  */
2162 static void kvm_cpu_vmxoff(void)
2163 {
2164 	asm volatile (__ex("vmxoff"));
2165 
2166 	intel_pt_handle_vmx(0);
2167 	cr4_clear_bits(X86_CR4_VMXE);
2168 }
2169 
2170 static void hardware_disable(void)
2171 {
2172 	vmclear_local_loaded_vmcss();
2173 	kvm_cpu_vmxoff();
2174 }
2175 
2176 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
2177 				      u32 msr, u32 *result)
2178 {
2179 	u32 vmx_msr_low, vmx_msr_high;
2180 	u32 ctl = ctl_min | ctl_opt;
2181 
2182 	rdmsr(msr, vmx_msr_low, vmx_msr_high);
2183 
2184 	ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
2185 	ctl |= vmx_msr_low;  /* bit == 1 in low word  ==> must be one  */
2186 
2187 	/* Ensure minimum (required) set of control bits are supported. */
2188 	if (ctl_min & ~ctl)
2189 		return -EIO;
2190 
2191 	*result = ctl;
2192 	return 0;
2193 }
2194 
2195 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf,
2196 				    struct vmx_capability *vmx_cap)
2197 {
2198 	u32 vmx_msr_low, vmx_msr_high;
2199 	u32 min, opt, min2, opt2;
2200 	u32 _pin_based_exec_control = 0;
2201 	u32 _cpu_based_exec_control = 0;
2202 	u32 _cpu_based_2nd_exec_control = 0;
2203 	u32 _vmexit_control = 0;
2204 	u32 _vmentry_control = 0;
2205 
2206 	memset(vmcs_conf, 0, sizeof(*vmcs_conf));
2207 	min = CPU_BASED_HLT_EXITING |
2208 #ifdef CONFIG_X86_64
2209 	      CPU_BASED_CR8_LOAD_EXITING |
2210 	      CPU_BASED_CR8_STORE_EXITING |
2211 #endif
2212 	      CPU_BASED_CR3_LOAD_EXITING |
2213 	      CPU_BASED_CR3_STORE_EXITING |
2214 	      CPU_BASED_UNCOND_IO_EXITING |
2215 	      CPU_BASED_MOV_DR_EXITING |
2216 	      CPU_BASED_USE_TSC_OFFSETING |
2217 	      CPU_BASED_MWAIT_EXITING |
2218 	      CPU_BASED_MONITOR_EXITING |
2219 	      CPU_BASED_INVLPG_EXITING |
2220 	      CPU_BASED_RDPMC_EXITING;
2221 
2222 	opt = CPU_BASED_TPR_SHADOW |
2223 	      CPU_BASED_USE_MSR_BITMAPS |
2224 	      CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
2225 	if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
2226 				&_cpu_based_exec_control) < 0)
2227 		return -EIO;
2228 #ifdef CONFIG_X86_64
2229 	if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2230 		_cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
2231 					   ~CPU_BASED_CR8_STORE_EXITING;
2232 #endif
2233 	if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
2234 		min2 = 0;
2235 		opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2236 			SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2237 			SECONDARY_EXEC_WBINVD_EXITING |
2238 			SECONDARY_EXEC_ENABLE_VPID |
2239 			SECONDARY_EXEC_ENABLE_EPT |
2240 			SECONDARY_EXEC_UNRESTRICTED_GUEST |
2241 			SECONDARY_EXEC_PAUSE_LOOP_EXITING |
2242 			SECONDARY_EXEC_DESC |
2243 			SECONDARY_EXEC_RDTSCP |
2244 			SECONDARY_EXEC_ENABLE_INVPCID |
2245 			SECONDARY_EXEC_APIC_REGISTER_VIRT |
2246 			SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
2247 			SECONDARY_EXEC_SHADOW_VMCS |
2248 			SECONDARY_EXEC_XSAVES |
2249 			SECONDARY_EXEC_RDSEED_EXITING |
2250 			SECONDARY_EXEC_RDRAND_EXITING |
2251 			SECONDARY_EXEC_ENABLE_PML |
2252 			SECONDARY_EXEC_TSC_SCALING |
2253 			SECONDARY_EXEC_PT_USE_GPA |
2254 			SECONDARY_EXEC_PT_CONCEAL_VMX |
2255 			SECONDARY_EXEC_ENABLE_VMFUNC |
2256 			SECONDARY_EXEC_ENCLS_EXITING;
2257 		if (adjust_vmx_controls(min2, opt2,
2258 					MSR_IA32_VMX_PROCBASED_CTLS2,
2259 					&_cpu_based_2nd_exec_control) < 0)
2260 			return -EIO;
2261 	}
2262 #ifndef CONFIG_X86_64
2263 	if (!(_cpu_based_2nd_exec_control &
2264 				SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
2265 		_cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
2266 #endif
2267 
2268 	if (!(_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2269 		_cpu_based_2nd_exec_control &= ~(
2270 				SECONDARY_EXEC_APIC_REGISTER_VIRT |
2271 				SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2272 				SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
2273 
2274 	rdmsr_safe(MSR_IA32_VMX_EPT_VPID_CAP,
2275 		&vmx_cap->ept, &vmx_cap->vpid);
2276 
2277 	if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
2278 		/* CR3 accesses and invlpg don't need to cause VM Exits when EPT
2279 		   enabled */
2280 		_cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
2281 					     CPU_BASED_CR3_STORE_EXITING |
2282 					     CPU_BASED_INVLPG_EXITING);
2283 	} else if (vmx_cap->ept) {
2284 		vmx_cap->ept = 0;
2285 		pr_warn_once("EPT CAP should not exist if not support "
2286 				"1-setting enable EPT VM-execution control\n");
2287 	}
2288 	if (!(_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_VPID) &&
2289 		vmx_cap->vpid) {
2290 		vmx_cap->vpid = 0;
2291 		pr_warn_once("VPID CAP should not exist if not support "
2292 				"1-setting enable VPID VM-execution control\n");
2293 	}
2294 
2295 	min = VM_EXIT_SAVE_DEBUG_CONTROLS | VM_EXIT_ACK_INTR_ON_EXIT;
2296 #ifdef CONFIG_X86_64
2297 	min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
2298 #endif
2299 	opt = VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL |
2300 	      VM_EXIT_SAVE_IA32_PAT |
2301 	      VM_EXIT_LOAD_IA32_PAT |
2302 	      VM_EXIT_LOAD_IA32_EFER |
2303 	      VM_EXIT_CLEAR_BNDCFGS |
2304 	      VM_EXIT_PT_CONCEAL_PIP |
2305 	      VM_EXIT_CLEAR_IA32_RTIT_CTL;
2306 	if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
2307 				&_vmexit_control) < 0)
2308 		return -EIO;
2309 
2310 	min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
2311 	opt = PIN_BASED_VIRTUAL_NMIS | PIN_BASED_POSTED_INTR |
2312 		 PIN_BASED_VMX_PREEMPTION_TIMER;
2313 	if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
2314 				&_pin_based_exec_control) < 0)
2315 		return -EIO;
2316 
2317 	if (cpu_has_broken_vmx_preemption_timer())
2318 		_pin_based_exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
2319 	if (!(_cpu_based_2nd_exec_control &
2320 		SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY))
2321 		_pin_based_exec_control &= ~PIN_BASED_POSTED_INTR;
2322 
2323 	min = VM_ENTRY_LOAD_DEBUG_CONTROLS;
2324 	opt = VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL |
2325 	      VM_ENTRY_LOAD_IA32_PAT |
2326 	      VM_ENTRY_LOAD_IA32_EFER |
2327 	      VM_ENTRY_LOAD_BNDCFGS |
2328 	      VM_ENTRY_PT_CONCEAL_PIP |
2329 	      VM_ENTRY_LOAD_IA32_RTIT_CTL;
2330 	if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
2331 				&_vmentry_control) < 0)
2332 		return -EIO;
2333 
2334 	/*
2335 	 * Some cpus support VM_{ENTRY,EXIT}_IA32_PERF_GLOBAL_CTRL but they
2336 	 * can't be used due to an errata where VM Exit may incorrectly clear
2337 	 * IA32_PERF_GLOBAL_CTRL[34:32].  Workaround the errata by using the
2338 	 * MSR load mechanism to switch IA32_PERF_GLOBAL_CTRL.
2339 	 */
2340 	if (boot_cpu_data.x86 == 0x6) {
2341 		switch (boot_cpu_data.x86_model) {
2342 		case 26: /* AAK155 */
2343 		case 30: /* AAP115 */
2344 		case 37: /* AAT100 */
2345 		case 44: /* BC86,AAY89,BD102 */
2346 		case 46: /* BA97 */
2347 			_vmexit_control &= ~VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL;
2348 			_vmexit_control &= ~VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL;
2349 			pr_warn_once("kvm: VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL "
2350 					"does not work properly. Using workaround\n");
2351 			break;
2352 		default:
2353 			break;
2354 		}
2355 	}
2356 
2357 
2358 	rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
2359 
2360 	/* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
2361 	if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
2362 		return -EIO;
2363 
2364 #ifdef CONFIG_X86_64
2365 	/* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
2366 	if (vmx_msr_high & (1u<<16))
2367 		return -EIO;
2368 #endif
2369 
2370 	/* Require Write-Back (WB) memory type for VMCS accesses. */
2371 	if (((vmx_msr_high >> 18) & 15) != 6)
2372 		return -EIO;
2373 
2374 	vmcs_conf->size = vmx_msr_high & 0x1fff;
2375 	vmcs_conf->order = get_order(vmcs_conf->size);
2376 	vmcs_conf->basic_cap = vmx_msr_high & ~0x1fff;
2377 
2378 	vmcs_conf->revision_id = vmx_msr_low;
2379 
2380 	vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
2381 	vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
2382 	vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
2383 	vmcs_conf->vmexit_ctrl         = _vmexit_control;
2384 	vmcs_conf->vmentry_ctrl        = _vmentry_control;
2385 
2386 	if (static_branch_unlikely(&enable_evmcs))
2387 		evmcs_sanitize_exec_ctrls(vmcs_conf);
2388 
2389 	return 0;
2390 }
2391 
2392 struct vmcs *alloc_vmcs_cpu(bool shadow, int cpu)
2393 {
2394 	int node = cpu_to_node(cpu);
2395 	struct page *pages;
2396 	struct vmcs *vmcs;
2397 
2398 	pages = __alloc_pages_node(node, GFP_KERNEL, vmcs_config.order);
2399 	if (!pages)
2400 		return NULL;
2401 	vmcs = page_address(pages);
2402 	memset(vmcs, 0, vmcs_config.size);
2403 
2404 	/* KVM supports Enlightened VMCS v1 only */
2405 	if (static_branch_unlikely(&enable_evmcs))
2406 		vmcs->hdr.revision_id = KVM_EVMCS_VERSION;
2407 	else
2408 		vmcs->hdr.revision_id = vmcs_config.revision_id;
2409 
2410 	if (shadow)
2411 		vmcs->hdr.shadow_vmcs = 1;
2412 	return vmcs;
2413 }
2414 
2415 void free_vmcs(struct vmcs *vmcs)
2416 {
2417 	free_pages((unsigned long)vmcs, vmcs_config.order);
2418 }
2419 
2420 /*
2421  * Free a VMCS, but before that VMCLEAR it on the CPU where it was last loaded
2422  */
2423 void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
2424 {
2425 	if (!loaded_vmcs->vmcs)
2426 		return;
2427 	loaded_vmcs_clear(loaded_vmcs);
2428 	free_vmcs(loaded_vmcs->vmcs);
2429 	loaded_vmcs->vmcs = NULL;
2430 	if (loaded_vmcs->msr_bitmap)
2431 		free_page((unsigned long)loaded_vmcs->msr_bitmap);
2432 	WARN_ON(loaded_vmcs->shadow_vmcs != NULL);
2433 }
2434 
2435 int alloc_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
2436 {
2437 	loaded_vmcs->vmcs = alloc_vmcs(false);
2438 	if (!loaded_vmcs->vmcs)
2439 		return -ENOMEM;
2440 
2441 	loaded_vmcs->shadow_vmcs = NULL;
2442 	loaded_vmcs_init(loaded_vmcs);
2443 
2444 	if (cpu_has_vmx_msr_bitmap()) {
2445 		loaded_vmcs->msr_bitmap = (unsigned long *)__get_free_page(GFP_KERNEL);
2446 		if (!loaded_vmcs->msr_bitmap)
2447 			goto out_vmcs;
2448 		memset(loaded_vmcs->msr_bitmap, 0xff, PAGE_SIZE);
2449 
2450 		if (IS_ENABLED(CONFIG_HYPERV) &&
2451 		    static_branch_unlikely(&enable_evmcs) &&
2452 		    (ms_hyperv.nested_features & HV_X64_NESTED_MSR_BITMAP)) {
2453 			struct hv_enlightened_vmcs *evmcs =
2454 				(struct hv_enlightened_vmcs *)loaded_vmcs->vmcs;
2455 
2456 			evmcs->hv_enlightenments_control.msr_bitmap = 1;
2457 		}
2458 	}
2459 
2460 	memset(&loaded_vmcs->host_state, 0, sizeof(struct vmcs_host_state));
2461 
2462 	return 0;
2463 
2464 out_vmcs:
2465 	free_loaded_vmcs(loaded_vmcs);
2466 	return -ENOMEM;
2467 }
2468 
2469 static void free_kvm_area(void)
2470 {
2471 	int cpu;
2472 
2473 	for_each_possible_cpu(cpu) {
2474 		free_vmcs(per_cpu(vmxarea, cpu));
2475 		per_cpu(vmxarea, cpu) = NULL;
2476 	}
2477 }
2478 
2479 static __init int alloc_kvm_area(void)
2480 {
2481 	int cpu;
2482 
2483 	for_each_possible_cpu(cpu) {
2484 		struct vmcs *vmcs;
2485 
2486 		vmcs = alloc_vmcs_cpu(false, cpu);
2487 		if (!vmcs) {
2488 			free_kvm_area();
2489 			return -ENOMEM;
2490 		}
2491 
2492 		/*
2493 		 * When eVMCS is enabled, alloc_vmcs_cpu() sets
2494 		 * vmcs->revision_id to KVM_EVMCS_VERSION instead of
2495 		 * revision_id reported by MSR_IA32_VMX_BASIC.
2496 		 *
2497 		 * However, even though not explicitly documented by
2498 		 * TLFS, VMXArea passed as VMXON argument should
2499 		 * still be marked with revision_id reported by
2500 		 * physical CPU.
2501 		 */
2502 		if (static_branch_unlikely(&enable_evmcs))
2503 			vmcs->hdr.revision_id = vmcs_config.revision_id;
2504 
2505 		per_cpu(vmxarea, cpu) = vmcs;
2506 	}
2507 	return 0;
2508 }
2509 
2510 static void fix_pmode_seg(struct kvm_vcpu *vcpu, int seg,
2511 		struct kvm_segment *save)
2512 {
2513 	if (!emulate_invalid_guest_state) {
2514 		/*
2515 		 * CS and SS RPL should be equal during guest entry according
2516 		 * to VMX spec, but in reality it is not always so. Since vcpu
2517 		 * is in the middle of the transition from real mode to
2518 		 * protected mode it is safe to assume that RPL 0 is a good
2519 		 * default value.
2520 		 */
2521 		if (seg == VCPU_SREG_CS || seg == VCPU_SREG_SS)
2522 			save->selector &= ~SEGMENT_RPL_MASK;
2523 		save->dpl = save->selector & SEGMENT_RPL_MASK;
2524 		save->s = 1;
2525 	}
2526 	vmx_set_segment(vcpu, save, seg);
2527 }
2528 
2529 static void enter_pmode(struct kvm_vcpu *vcpu)
2530 {
2531 	unsigned long flags;
2532 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2533 
2534 	/*
2535 	 * Update real mode segment cache. It may be not up-to-date if sement
2536 	 * register was written while vcpu was in a guest mode.
2537 	 */
2538 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
2539 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
2540 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
2541 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
2542 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
2543 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
2544 
2545 	vmx->rmode.vm86_active = 0;
2546 
2547 	vmx_segment_cache_clear(vmx);
2548 
2549 	vmx_set_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
2550 
2551 	flags = vmcs_readl(GUEST_RFLAGS);
2552 	flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
2553 	flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
2554 	vmcs_writel(GUEST_RFLAGS, flags);
2555 
2556 	vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
2557 			(vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
2558 
2559 	update_exception_bitmap(vcpu);
2560 
2561 	fix_pmode_seg(vcpu, VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
2562 	fix_pmode_seg(vcpu, VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
2563 	fix_pmode_seg(vcpu, VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
2564 	fix_pmode_seg(vcpu, VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
2565 	fix_pmode_seg(vcpu, VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
2566 	fix_pmode_seg(vcpu, VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
2567 }
2568 
2569 static void fix_rmode_seg(int seg, struct kvm_segment *save)
2570 {
2571 	const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2572 	struct kvm_segment var = *save;
2573 
2574 	var.dpl = 0x3;
2575 	if (seg == VCPU_SREG_CS)
2576 		var.type = 0x3;
2577 
2578 	if (!emulate_invalid_guest_state) {
2579 		var.selector = var.base >> 4;
2580 		var.base = var.base & 0xffff0;
2581 		var.limit = 0xffff;
2582 		var.g = 0;
2583 		var.db = 0;
2584 		var.present = 1;
2585 		var.s = 1;
2586 		var.l = 0;
2587 		var.unusable = 0;
2588 		var.type = 0x3;
2589 		var.avl = 0;
2590 		if (save->base & 0xf)
2591 			printk_once(KERN_WARNING "kvm: segment base is not "
2592 					"paragraph aligned when entering "
2593 					"protected mode (seg=%d)", seg);
2594 	}
2595 
2596 	vmcs_write16(sf->selector, var.selector);
2597 	vmcs_writel(sf->base, var.base);
2598 	vmcs_write32(sf->limit, var.limit);
2599 	vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(&var));
2600 }
2601 
2602 static void enter_rmode(struct kvm_vcpu *vcpu)
2603 {
2604 	unsigned long flags;
2605 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2606 	struct kvm_vmx *kvm_vmx = to_kvm_vmx(vcpu->kvm);
2607 
2608 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
2609 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
2610 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
2611 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
2612 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
2613 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
2614 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
2615 
2616 	vmx->rmode.vm86_active = 1;
2617 
2618 	/*
2619 	 * Very old userspace does not call KVM_SET_TSS_ADDR before entering
2620 	 * vcpu. Warn the user that an update is overdue.
2621 	 */
2622 	if (!kvm_vmx->tss_addr)
2623 		printk_once(KERN_WARNING "kvm: KVM_SET_TSS_ADDR need to be "
2624 			     "called before entering vcpu\n");
2625 
2626 	vmx_segment_cache_clear(vmx);
2627 
2628 	vmcs_writel(GUEST_TR_BASE, kvm_vmx->tss_addr);
2629 	vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
2630 	vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
2631 
2632 	flags = vmcs_readl(GUEST_RFLAGS);
2633 	vmx->rmode.save_rflags = flags;
2634 
2635 	flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
2636 
2637 	vmcs_writel(GUEST_RFLAGS, flags);
2638 	vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
2639 	update_exception_bitmap(vcpu);
2640 
2641 	fix_rmode_seg(VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
2642 	fix_rmode_seg(VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
2643 	fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
2644 	fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
2645 	fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
2646 	fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
2647 
2648 	kvm_mmu_reset_context(vcpu);
2649 }
2650 
2651 void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
2652 {
2653 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2654 	struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
2655 
2656 	if (!msr)
2657 		return;
2658 
2659 	vcpu->arch.efer = efer;
2660 	if (efer & EFER_LMA) {
2661 		vm_entry_controls_setbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
2662 		msr->data = efer;
2663 	} else {
2664 		vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
2665 
2666 		msr->data = efer & ~EFER_LME;
2667 	}
2668 	setup_msrs(vmx);
2669 }
2670 
2671 #ifdef CONFIG_X86_64
2672 
2673 static void enter_lmode(struct kvm_vcpu *vcpu)
2674 {
2675 	u32 guest_tr_ar;
2676 
2677 	vmx_segment_cache_clear(to_vmx(vcpu));
2678 
2679 	guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
2680 	if ((guest_tr_ar & VMX_AR_TYPE_MASK) != VMX_AR_TYPE_BUSY_64_TSS) {
2681 		pr_debug_ratelimited("%s: tss fixup for long mode. \n",
2682 				     __func__);
2683 		vmcs_write32(GUEST_TR_AR_BYTES,
2684 			     (guest_tr_ar & ~VMX_AR_TYPE_MASK)
2685 			     | VMX_AR_TYPE_BUSY_64_TSS);
2686 	}
2687 	vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
2688 }
2689 
2690 static void exit_lmode(struct kvm_vcpu *vcpu)
2691 {
2692 	vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
2693 	vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
2694 }
2695 
2696 #endif
2697 
2698 static void vmx_flush_tlb_gva(struct kvm_vcpu *vcpu, gva_t addr)
2699 {
2700 	int vpid = to_vmx(vcpu)->vpid;
2701 
2702 	if (!vpid_sync_vcpu_addr(vpid, addr))
2703 		vpid_sync_context(vpid);
2704 
2705 	/*
2706 	 * If VPIDs are not supported or enabled, then the above is a no-op.
2707 	 * But we don't really need a TLB flush in that case anyway, because
2708 	 * each VM entry/exit includes an implicit flush when VPID is 0.
2709 	 */
2710 }
2711 
2712 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
2713 {
2714 	ulong cr0_guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
2715 
2716 	vcpu->arch.cr0 &= ~cr0_guest_owned_bits;
2717 	vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & cr0_guest_owned_bits;
2718 }
2719 
2720 static void vmx_decache_cr3(struct kvm_vcpu *vcpu)
2721 {
2722 	if (enable_unrestricted_guest || (enable_ept && is_paging(vcpu)))
2723 		vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
2724 	__set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
2725 }
2726 
2727 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
2728 {
2729 	ulong cr4_guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
2730 
2731 	vcpu->arch.cr4 &= ~cr4_guest_owned_bits;
2732 	vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & cr4_guest_owned_bits;
2733 }
2734 
2735 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
2736 {
2737 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
2738 
2739 	if (!test_bit(VCPU_EXREG_PDPTR,
2740 		      (unsigned long *)&vcpu->arch.regs_dirty))
2741 		return;
2742 
2743 	if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
2744 		vmcs_write64(GUEST_PDPTR0, mmu->pdptrs[0]);
2745 		vmcs_write64(GUEST_PDPTR1, mmu->pdptrs[1]);
2746 		vmcs_write64(GUEST_PDPTR2, mmu->pdptrs[2]);
2747 		vmcs_write64(GUEST_PDPTR3, mmu->pdptrs[3]);
2748 	}
2749 }
2750 
2751 void ept_save_pdptrs(struct kvm_vcpu *vcpu)
2752 {
2753 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
2754 
2755 	if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
2756 		mmu->pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
2757 		mmu->pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
2758 		mmu->pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
2759 		mmu->pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
2760 	}
2761 
2762 	__set_bit(VCPU_EXREG_PDPTR,
2763 		  (unsigned long *)&vcpu->arch.regs_avail);
2764 	__set_bit(VCPU_EXREG_PDPTR,
2765 		  (unsigned long *)&vcpu->arch.regs_dirty);
2766 }
2767 
2768 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
2769 					unsigned long cr0,
2770 					struct kvm_vcpu *vcpu)
2771 {
2772 	if (!test_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail))
2773 		vmx_decache_cr3(vcpu);
2774 	if (!(cr0 & X86_CR0_PG)) {
2775 		/* From paging/starting to nonpaging */
2776 		vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
2777 			     vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) |
2778 			     (CPU_BASED_CR3_LOAD_EXITING |
2779 			      CPU_BASED_CR3_STORE_EXITING));
2780 		vcpu->arch.cr0 = cr0;
2781 		vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
2782 	} else if (!is_paging(vcpu)) {
2783 		/* From nonpaging to paging */
2784 		vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
2785 			     vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) &
2786 			     ~(CPU_BASED_CR3_LOAD_EXITING |
2787 			       CPU_BASED_CR3_STORE_EXITING));
2788 		vcpu->arch.cr0 = cr0;
2789 		vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
2790 	}
2791 
2792 	if (!(cr0 & X86_CR0_WP))
2793 		*hw_cr0 &= ~X86_CR0_WP;
2794 }
2795 
2796 void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
2797 {
2798 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2799 	unsigned long hw_cr0;
2800 
2801 	hw_cr0 = (cr0 & ~KVM_VM_CR0_ALWAYS_OFF);
2802 	if (enable_unrestricted_guest)
2803 		hw_cr0 |= KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
2804 	else {
2805 		hw_cr0 |= KVM_VM_CR0_ALWAYS_ON;
2806 
2807 		if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
2808 			enter_pmode(vcpu);
2809 
2810 		if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
2811 			enter_rmode(vcpu);
2812 	}
2813 
2814 #ifdef CONFIG_X86_64
2815 	if (vcpu->arch.efer & EFER_LME) {
2816 		if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
2817 			enter_lmode(vcpu);
2818 		if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
2819 			exit_lmode(vcpu);
2820 	}
2821 #endif
2822 
2823 	if (enable_ept && !enable_unrestricted_guest)
2824 		ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
2825 
2826 	vmcs_writel(CR0_READ_SHADOW, cr0);
2827 	vmcs_writel(GUEST_CR0, hw_cr0);
2828 	vcpu->arch.cr0 = cr0;
2829 
2830 	/* depends on vcpu->arch.cr0 to be set to a new value */
2831 	vmx->emulation_required = emulation_required(vcpu);
2832 }
2833 
2834 static int get_ept_level(struct kvm_vcpu *vcpu)
2835 {
2836 	if (cpu_has_vmx_ept_5levels() && (cpuid_maxphyaddr(vcpu) > 48))
2837 		return 5;
2838 	return 4;
2839 }
2840 
2841 u64 construct_eptp(struct kvm_vcpu *vcpu, unsigned long root_hpa)
2842 {
2843 	u64 eptp = VMX_EPTP_MT_WB;
2844 
2845 	eptp |= (get_ept_level(vcpu) == 5) ? VMX_EPTP_PWL_5 : VMX_EPTP_PWL_4;
2846 
2847 	if (enable_ept_ad_bits &&
2848 	    (!is_guest_mode(vcpu) || nested_ept_ad_enabled(vcpu)))
2849 		eptp |= VMX_EPTP_AD_ENABLE_BIT;
2850 	eptp |= (root_hpa & PAGE_MASK);
2851 
2852 	return eptp;
2853 }
2854 
2855 void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
2856 {
2857 	struct kvm *kvm = vcpu->kvm;
2858 	unsigned long guest_cr3;
2859 	u64 eptp;
2860 
2861 	guest_cr3 = cr3;
2862 	if (enable_ept) {
2863 		eptp = construct_eptp(vcpu, cr3);
2864 		vmcs_write64(EPT_POINTER, eptp);
2865 
2866 		if (kvm_x86_ops->tlb_remote_flush) {
2867 			spin_lock(&to_kvm_vmx(kvm)->ept_pointer_lock);
2868 			to_vmx(vcpu)->ept_pointer = eptp;
2869 			to_kvm_vmx(kvm)->ept_pointers_match
2870 				= EPT_POINTERS_CHECK;
2871 			spin_unlock(&to_kvm_vmx(kvm)->ept_pointer_lock);
2872 		}
2873 
2874 		if (enable_unrestricted_guest || is_paging(vcpu) ||
2875 		    is_guest_mode(vcpu))
2876 			guest_cr3 = kvm_read_cr3(vcpu);
2877 		else
2878 			guest_cr3 = to_kvm_vmx(kvm)->ept_identity_map_addr;
2879 		ept_load_pdptrs(vcpu);
2880 	}
2881 
2882 	vmcs_writel(GUEST_CR3, guest_cr3);
2883 }
2884 
2885 int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
2886 {
2887 	/*
2888 	 * Pass through host's Machine Check Enable value to hw_cr4, which
2889 	 * is in force while we are in guest mode.  Do not let guests control
2890 	 * this bit, even if host CR4.MCE == 0.
2891 	 */
2892 	unsigned long hw_cr4;
2893 
2894 	hw_cr4 = (cr4_read_shadow() & X86_CR4_MCE) | (cr4 & ~X86_CR4_MCE);
2895 	if (enable_unrestricted_guest)
2896 		hw_cr4 |= KVM_VM_CR4_ALWAYS_ON_UNRESTRICTED_GUEST;
2897 	else if (to_vmx(vcpu)->rmode.vm86_active)
2898 		hw_cr4 |= KVM_RMODE_VM_CR4_ALWAYS_ON;
2899 	else
2900 		hw_cr4 |= KVM_PMODE_VM_CR4_ALWAYS_ON;
2901 
2902 	if (!boot_cpu_has(X86_FEATURE_UMIP) && vmx_umip_emulated()) {
2903 		if (cr4 & X86_CR4_UMIP) {
2904 			vmcs_set_bits(SECONDARY_VM_EXEC_CONTROL,
2905 				SECONDARY_EXEC_DESC);
2906 			hw_cr4 &= ~X86_CR4_UMIP;
2907 		} else if (!is_guest_mode(vcpu) ||
2908 			!nested_cpu_has2(get_vmcs12(vcpu), SECONDARY_EXEC_DESC))
2909 			vmcs_clear_bits(SECONDARY_VM_EXEC_CONTROL,
2910 					SECONDARY_EXEC_DESC);
2911 	}
2912 
2913 	if (cr4 & X86_CR4_VMXE) {
2914 		/*
2915 		 * To use VMXON (and later other VMX instructions), a guest
2916 		 * must first be able to turn on cr4.VMXE (see handle_vmon()).
2917 		 * So basically the check on whether to allow nested VMX
2918 		 * is here.  We operate under the default treatment of SMM,
2919 		 * so VMX cannot be enabled under SMM.
2920 		 */
2921 		if (!nested_vmx_allowed(vcpu) || is_smm(vcpu))
2922 			return 1;
2923 	}
2924 
2925 	if (to_vmx(vcpu)->nested.vmxon && !nested_cr4_valid(vcpu, cr4))
2926 		return 1;
2927 
2928 	vcpu->arch.cr4 = cr4;
2929 
2930 	if (!enable_unrestricted_guest) {
2931 		if (enable_ept) {
2932 			if (!is_paging(vcpu)) {
2933 				hw_cr4 &= ~X86_CR4_PAE;
2934 				hw_cr4 |= X86_CR4_PSE;
2935 			} else if (!(cr4 & X86_CR4_PAE)) {
2936 				hw_cr4 &= ~X86_CR4_PAE;
2937 			}
2938 		}
2939 
2940 		/*
2941 		 * SMEP/SMAP/PKU is disabled if CPU is in non-paging mode in
2942 		 * hardware.  To emulate this behavior, SMEP/SMAP/PKU needs
2943 		 * to be manually disabled when guest switches to non-paging
2944 		 * mode.
2945 		 *
2946 		 * If !enable_unrestricted_guest, the CPU is always running
2947 		 * with CR0.PG=1 and CR4 needs to be modified.
2948 		 * If enable_unrestricted_guest, the CPU automatically
2949 		 * disables SMEP/SMAP/PKU when the guest sets CR0.PG=0.
2950 		 */
2951 		if (!is_paging(vcpu))
2952 			hw_cr4 &= ~(X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE);
2953 	}
2954 
2955 	vmcs_writel(CR4_READ_SHADOW, cr4);
2956 	vmcs_writel(GUEST_CR4, hw_cr4);
2957 	return 0;
2958 }
2959 
2960 void vmx_get_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg)
2961 {
2962 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2963 	u32 ar;
2964 
2965 	if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
2966 		*var = vmx->rmode.segs[seg];
2967 		if (seg == VCPU_SREG_TR
2968 		    || var->selector == vmx_read_guest_seg_selector(vmx, seg))
2969 			return;
2970 		var->base = vmx_read_guest_seg_base(vmx, seg);
2971 		var->selector = vmx_read_guest_seg_selector(vmx, seg);
2972 		return;
2973 	}
2974 	var->base = vmx_read_guest_seg_base(vmx, seg);
2975 	var->limit = vmx_read_guest_seg_limit(vmx, seg);
2976 	var->selector = vmx_read_guest_seg_selector(vmx, seg);
2977 	ar = vmx_read_guest_seg_ar(vmx, seg);
2978 	var->unusable = (ar >> 16) & 1;
2979 	var->type = ar & 15;
2980 	var->s = (ar >> 4) & 1;
2981 	var->dpl = (ar >> 5) & 3;
2982 	/*
2983 	 * Some userspaces do not preserve unusable property. Since usable
2984 	 * segment has to be present according to VMX spec we can use present
2985 	 * property to amend userspace bug by making unusable segment always
2986 	 * nonpresent. vmx_segment_access_rights() already marks nonpresent
2987 	 * segment as unusable.
2988 	 */
2989 	var->present = !var->unusable;
2990 	var->avl = (ar >> 12) & 1;
2991 	var->l = (ar >> 13) & 1;
2992 	var->db = (ar >> 14) & 1;
2993 	var->g = (ar >> 15) & 1;
2994 }
2995 
2996 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
2997 {
2998 	struct kvm_segment s;
2999 
3000 	if (to_vmx(vcpu)->rmode.vm86_active) {
3001 		vmx_get_segment(vcpu, &s, seg);
3002 		return s.base;
3003 	}
3004 	return vmx_read_guest_seg_base(to_vmx(vcpu), seg);
3005 }
3006 
3007 int vmx_get_cpl(struct kvm_vcpu *vcpu)
3008 {
3009 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3010 
3011 	if (unlikely(vmx->rmode.vm86_active))
3012 		return 0;
3013 	else {
3014 		int ar = vmx_read_guest_seg_ar(vmx, VCPU_SREG_SS);
3015 		return VMX_AR_DPL(ar);
3016 	}
3017 }
3018 
3019 static u32 vmx_segment_access_rights(struct kvm_segment *var)
3020 {
3021 	u32 ar;
3022 
3023 	if (var->unusable || !var->present)
3024 		ar = 1 << 16;
3025 	else {
3026 		ar = var->type & 15;
3027 		ar |= (var->s & 1) << 4;
3028 		ar |= (var->dpl & 3) << 5;
3029 		ar |= (var->present & 1) << 7;
3030 		ar |= (var->avl & 1) << 12;
3031 		ar |= (var->l & 1) << 13;
3032 		ar |= (var->db & 1) << 14;
3033 		ar |= (var->g & 1) << 15;
3034 	}
3035 
3036 	return ar;
3037 }
3038 
3039 void vmx_set_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg)
3040 {
3041 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3042 	const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3043 
3044 	vmx_segment_cache_clear(vmx);
3045 
3046 	if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3047 		vmx->rmode.segs[seg] = *var;
3048 		if (seg == VCPU_SREG_TR)
3049 			vmcs_write16(sf->selector, var->selector);
3050 		else if (var->s)
3051 			fix_rmode_seg(seg, &vmx->rmode.segs[seg]);
3052 		goto out;
3053 	}
3054 
3055 	vmcs_writel(sf->base, var->base);
3056 	vmcs_write32(sf->limit, var->limit);
3057 	vmcs_write16(sf->selector, var->selector);
3058 
3059 	/*
3060 	 *   Fix the "Accessed" bit in AR field of segment registers for older
3061 	 * qemu binaries.
3062 	 *   IA32 arch specifies that at the time of processor reset the
3063 	 * "Accessed" bit in the AR field of segment registers is 1. And qemu
3064 	 * is setting it to 0 in the userland code. This causes invalid guest
3065 	 * state vmexit when "unrestricted guest" mode is turned on.
3066 	 *    Fix for this setup issue in cpu_reset is being pushed in the qemu
3067 	 * tree. Newer qemu binaries with that qemu fix would not need this
3068 	 * kvm hack.
3069 	 */
3070 	if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
3071 		var->type |= 0x1; /* Accessed */
3072 
3073 	vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(var));
3074 
3075 out:
3076 	vmx->emulation_required = emulation_required(vcpu);
3077 }
3078 
3079 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3080 {
3081 	u32 ar = vmx_read_guest_seg_ar(to_vmx(vcpu), VCPU_SREG_CS);
3082 
3083 	*db = (ar >> 14) & 1;
3084 	*l = (ar >> 13) & 1;
3085 }
3086 
3087 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3088 {
3089 	dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
3090 	dt->address = vmcs_readl(GUEST_IDTR_BASE);
3091 }
3092 
3093 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3094 {
3095 	vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
3096 	vmcs_writel(GUEST_IDTR_BASE, dt->address);
3097 }
3098 
3099 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3100 {
3101 	dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
3102 	dt->address = vmcs_readl(GUEST_GDTR_BASE);
3103 }
3104 
3105 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3106 {
3107 	vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
3108 	vmcs_writel(GUEST_GDTR_BASE, dt->address);
3109 }
3110 
3111 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
3112 {
3113 	struct kvm_segment var;
3114 	u32 ar;
3115 
3116 	vmx_get_segment(vcpu, &var, seg);
3117 	var.dpl = 0x3;
3118 	if (seg == VCPU_SREG_CS)
3119 		var.type = 0x3;
3120 	ar = vmx_segment_access_rights(&var);
3121 
3122 	if (var.base != (var.selector << 4))
3123 		return false;
3124 	if (var.limit != 0xffff)
3125 		return false;
3126 	if (ar != 0xf3)
3127 		return false;
3128 
3129 	return true;
3130 }
3131 
3132 static bool code_segment_valid(struct kvm_vcpu *vcpu)
3133 {
3134 	struct kvm_segment cs;
3135 	unsigned int cs_rpl;
3136 
3137 	vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3138 	cs_rpl = cs.selector & SEGMENT_RPL_MASK;
3139 
3140 	if (cs.unusable)
3141 		return false;
3142 	if (~cs.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_ACCESSES_MASK))
3143 		return false;
3144 	if (!cs.s)
3145 		return false;
3146 	if (cs.type & VMX_AR_TYPE_WRITEABLE_MASK) {
3147 		if (cs.dpl > cs_rpl)
3148 			return false;
3149 	} else {
3150 		if (cs.dpl != cs_rpl)
3151 			return false;
3152 	}
3153 	if (!cs.present)
3154 		return false;
3155 
3156 	/* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
3157 	return true;
3158 }
3159 
3160 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
3161 {
3162 	struct kvm_segment ss;
3163 	unsigned int ss_rpl;
3164 
3165 	vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3166 	ss_rpl = ss.selector & SEGMENT_RPL_MASK;
3167 
3168 	if (ss.unusable)
3169 		return true;
3170 	if (ss.type != 3 && ss.type != 7)
3171 		return false;
3172 	if (!ss.s)
3173 		return false;
3174 	if (ss.dpl != ss_rpl) /* DPL != RPL */
3175 		return false;
3176 	if (!ss.present)
3177 		return false;
3178 
3179 	return true;
3180 }
3181 
3182 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
3183 {
3184 	struct kvm_segment var;
3185 	unsigned int rpl;
3186 
3187 	vmx_get_segment(vcpu, &var, seg);
3188 	rpl = var.selector & SEGMENT_RPL_MASK;
3189 
3190 	if (var.unusable)
3191 		return true;
3192 	if (!var.s)
3193 		return false;
3194 	if (!var.present)
3195 		return false;
3196 	if (~var.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_WRITEABLE_MASK)) {
3197 		if (var.dpl < rpl) /* DPL < RPL */
3198 			return false;
3199 	}
3200 
3201 	/* TODO: Add other members to kvm_segment_field to allow checking for other access
3202 	 * rights flags
3203 	 */
3204 	return true;
3205 }
3206 
3207 static bool tr_valid(struct kvm_vcpu *vcpu)
3208 {
3209 	struct kvm_segment tr;
3210 
3211 	vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
3212 
3213 	if (tr.unusable)
3214 		return false;
3215 	if (tr.selector & SEGMENT_TI_MASK)	/* TI = 1 */
3216 		return false;
3217 	if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
3218 		return false;
3219 	if (!tr.present)
3220 		return false;
3221 
3222 	return true;
3223 }
3224 
3225 static bool ldtr_valid(struct kvm_vcpu *vcpu)
3226 {
3227 	struct kvm_segment ldtr;
3228 
3229 	vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
3230 
3231 	if (ldtr.unusable)
3232 		return true;
3233 	if (ldtr.selector & SEGMENT_TI_MASK)	/* TI = 1 */
3234 		return false;
3235 	if (ldtr.type != 2)
3236 		return false;
3237 	if (!ldtr.present)
3238 		return false;
3239 
3240 	return true;
3241 }
3242 
3243 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
3244 {
3245 	struct kvm_segment cs, ss;
3246 
3247 	vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3248 	vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3249 
3250 	return ((cs.selector & SEGMENT_RPL_MASK) ==
3251 		 (ss.selector & SEGMENT_RPL_MASK));
3252 }
3253 
3254 /*
3255  * Check if guest state is valid. Returns true if valid, false if
3256  * not.
3257  * We assume that registers are always usable
3258  */
3259 static bool guest_state_valid(struct kvm_vcpu *vcpu)
3260 {
3261 	if (enable_unrestricted_guest)
3262 		return true;
3263 
3264 	/* real mode guest state checks */
3265 	if (!is_protmode(vcpu) || (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
3266 		if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
3267 			return false;
3268 		if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
3269 			return false;
3270 		if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
3271 			return false;
3272 		if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
3273 			return false;
3274 		if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
3275 			return false;
3276 		if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
3277 			return false;
3278 	} else {
3279 	/* protected mode guest state checks */
3280 		if (!cs_ss_rpl_check(vcpu))
3281 			return false;
3282 		if (!code_segment_valid(vcpu))
3283 			return false;
3284 		if (!stack_segment_valid(vcpu))
3285 			return false;
3286 		if (!data_segment_valid(vcpu, VCPU_SREG_DS))
3287 			return false;
3288 		if (!data_segment_valid(vcpu, VCPU_SREG_ES))
3289 			return false;
3290 		if (!data_segment_valid(vcpu, VCPU_SREG_FS))
3291 			return false;
3292 		if (!data_segment_valid(vcpu, VCPU_SREG_GS))
3293 			return false;
3294 		if (!tr_valid(vcpu))
3295 			return false;
3296 		if (!ldtr_valid(vcpu))
3297 			return false;
3298 	}
3299 	/* TODO:
3300 	 * - Add checks on RIP
3301 	 * - Add checks on RFLAGS
3302 	 */
3303 
3304 	return true;
3305 }
3306 
3307 static int init_rmode_tss(struct kvm *kvm)
3308 {
3309 	gfn_t fn;
3310 	u16 data = 0;
3311 	int idx, r;
3312 
3313 	idx = srcu_read_lock(&kvm->srcu);
3314 	fn = to_kvm_vmx(kvm)->tss_addr >> PAGE_SHIFT;
3315 	r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3316 	if (r < 0)
3317 		goto out;
3318 	data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
3319 	r = kvm_write_guest_page(kvm, fn++, &data,
3320 			TSS_IOPB_BASE_OFFSET, sizeof(u16));
3321 	if (r < 0)
3322 		goto out;
3323 	r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
3324 	if (r < 0)
3325 		goto out;
3326 	r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3327 	if (r < 0)
3328 		goto out;
3329 	data = ~0;
3330 	r = kvm_write_guest_page(kvm, fn, &data,
3331 				 RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
3332 				 sizeof(u8));
3333 out:
3334 	srcu_read_unlock(&kvm->srcu, idx);
3335 	return r;
3336 }
3337 
3338 static int init_rmode_identity_map(struct kvm *kvm)
3339 {
3340 	struct kvm_vmx *kvm_vmx = to_kvm_vmx(kvm);
3341 	int i, idx, r = 0;
3342 	kvm_pfn_t identity_map_pfn;
3343 	u32 tmp;
3344 
3345 	/* Protect kvm_vmx->ept_identity_pagetable_done. */
3346 	mutex_lock(&kvm->slots_lock);
3347 
3348 	if (likely(kvm_vmx->ept_identity_pagetable_done))
3349 		goto out2;
3350 
3351 	if (!kvm_vmx->ept_identity_map_addr)
3352 		kvm_vmx->ept_identity_map_addr = VMX_EPT_IDENTITY_PAGETABLE_ADDR;
3353 	identity_map_pfn = kvm_vmx->ept_identity_map_addr >> PAGE_SHIFT;
3354 
3355 	r = __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
3356 				    kvm_vmx->ept_identity_map_addr, PAGE_SIZE);
3357 	if (r < 0)
3358 		goto out2;
3359 
3360 	idx = srcu_read_lock(&kvm->srcu);
3361 	r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
3362 	if (r < 0)
3363 		goto out;
3364 	/* Set up identity-mapping pagetable for EPT in real mode */
3365 	for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
3366 		tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
3367 			_PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
3368 		r = kvm_write_guest_page(kvm, identity_map_pfn,
3369 				&tmp, i * sizeof(tmp), sizeof(tmp));
3370 		if (r < 0)
3371 			goto out;
3372 	}
3373 	kvm_vmx->ept_identity_pagetable_done = true;
3374 
3375 out:
3376 	srcu_read_unlock(&kvm->srcu, idx);
3377 
3378 out2:
3379 	mutex_unlock(&kvm->slots_lock);
3380 	return r;
3381 }
3382 
3383 static void seg_setup(int seg)
3384 {
3385 	const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3386 	unsigned int ar;
3387 
3388 	vmcs_write16(sf->selector, 0);
3389 	vmcs_writel(sf->base, 0);
3390 	vmcs_write32(sf->limit, 0xffff);
3391 	ar = 0x93;
3392 	if (seg == VCPU_SREG_CS)
3393 		ar |= 0x08; /* code segment */
3394 
3395 	vmcs_write32(sf->ar_bytes, ar);
3396 }
3397 
3398 static int alloc_apic_access_page(struct kvm *kvm)
3399 {
3400 	struct page *page;
3401 	int r = 0;
3402 
3403 	mutex_lock(&kvm->slots_lock);
3404 	if (kvm->arch.apic_access_page_done)
3405 		goto out;
3406 	r = __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
3407 				    APIC_DEFAULT_PHYS_BASE, PAGE_SIZE);
3408 	if (r)
3409 		goto out;
3410 
3411 	page = gfn_to_page(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
3412 	if (is_error_page(page)) {
3413 		r = -EFAULT;
3414 		goto out;
3415 	}
3416 
3417 	/*
3418 	 * Do not pin the page in memory, so that memory hot-unplug
3419 	 * is able to migrate it.
3420 	 */
3421 	put_page(page);
3422 	kvm->arch.apic_access_page_done = true;
3423 out:
3424 	mutex_unlock(&kvm->slots_lock);
3425 	return r;
3426 }
3427 
3428 int allocate_vpid(void)
3429 {
3430 	int vpid;
3431 
3432 	if (!enable_vpid)
3433 		return 0;
3434 	spin_lock(&vmx_vpid_lock);
3435 	vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
3436 	if (vpid < VMX_NR_VPIDS)
3437 		__set_bit(vpid, vmx_vpid_bitmap);
3438 	else
3439 		vpid = 0;
3440 	spin_unlock(&vmx_vpid_lock);
3441 	return vpid;
3442 }
3443 
3444 void free_vpid(int vpid)
3445 {
3446 	if (!enable_vpid || vpid == 0)
3447 		return;
3448 	spin_lock(&vmx_vpid_lock);
3449 	__clear_bit(vpid, vmx_vpid_bitmap);
3450 	spin_unlock(&vmx_vpid_lock);
3451 }
3452 
3453 static __always_inline void vmx_disable_intercept_for_msr(unsigned long *msr_bitmap,
3454 							  u32 msr, int type)
3455 {
3456 	int f = sizeof(unsigned long);
3457 
3458 	if (!cpu_has_vmx_msr_bitmap())
3459 		return;
3460 
3461 	if (static_branch_unlikely(&enable_evmcs))
3462 		evmcs_touch_msr_bitmap();
3463 
3464 	/*
3465 	 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
3466 	 * have the write-low and read-high bitmap offsets the wrong way round.
3467 	 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
3468 	 */
3469 	if (msr <= 0x1fff) {
3470 		if (type & MSR_TYPE_R)
3471 			/* read-low */
3472 			__clear_bit(msr, msr_bitmap + 0x000 / f);
3473 
3474 		if (type & MSR_TYPE_W)
3475 			/* write-low */
3476 			__clear_bit(msr, msr_bitmap + 0x800 / f);
3477 
3478 	} else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
3479 		msr &= 0x1fff;
3480 		if (type & MSR_TYPE_R)
3481 			/* read-high */
3482 			__clear_bit(msr, msr_bitmap + 0x400 / f);
3483 
3484 		if (type & MSR_TYPE_W)
3485 			/* write-high */
3486 			__clear_bit(msr, msr_bitmap + 0xc00 / f);
3487 
3488 	}
3489 }
3490 
3491 static __always_inline void vmx_enable_intercept_for_msr(unsigned long *msr_bitmap,
3492 							 u32 msr, int type)
3493 {
3494 	int f = sizeof(unsigned long);
3495 
3496 	if (!cpu_has_vmx_msr_bitmap())
3497 		return;
3498 
3499 	if (static_branch_unlikely(&enable_evmcs))
3500 		evmcs_touch_msr_bitmap();
3501 
3502 	/*
3503 	 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
3504 	 * have the write-low and read-high bitmap offsets the wrong way round.
3505 	 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
3506 	 */
3507 	if (msr <= 0x1fff) {
3508 		if (type & MSR_TYPE_R)
3509 			/* read-low */
3510 			__set_bit(msr, msr_bitmap + 0x000 / f);
3511 
3512 		if (type & MSR_TYPE_W)
3513 			/* write-low */
3514 			__set_bit(msr, msr_bitmap + 0x800 / f);
3515 
3516 	} else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
3517 		msr &= 0x1fff;
3518 		if (type & MSR_TYPE_R)
3519 			/* read-high */
3520 			__set_bit(msr, msr_bitmap + 0x400 / f);
3521 
3522 		if (type & MSR_TYPE_W)
3523 			/* write-high */
3524 			__set_bit(msr, msr_bitmap + 0xc00 / f);
3525 
3526 	}
3527 }
3528 
3529 static __always_inline void vmx_set_intercept_for_msr(unsigned long *msr_bitmap,
3530 			     			      u32 msr, int type, bool value)
3531 {
3532 	if (value)
3533 		vmx_enable_intercept_for_msr(msr_bitmap, msr, type);
3534 	else
3535 		vmx_disable_intercept_for_msr(msr_bitmap, msr, type);
3536 }
3537 
3538 static u8 vmx_msr_bitmap_mode(struct kvm_vcpu *vcpu)
3539 {
3540 	u8 mode = 0;
3541 
3542 	if (cpu_has_secondary_exec_ctrls() &&
3543 	    (vmcs_read32(SECONDARY_VM_EXEC_CONTROL) &
3544 	     SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE)) {
3545 		mode |= MSR_BITMAP_MODE_X2APIC;
3546 		if (enable_apicv && kvm_vcpu_apicv_active(vcpu))
3547 			mode |= MSR_BITMAP_MODE_X2APIC_APICV;
3548 	}
3549 
3550 	return mode;
3551 }
3552 
3553 static void vmx_update_msr_bitmap_x2apic(unsigned long *msr_bitmap,
3554 					 u8 mode)
3555 {
3556 	int msr;
3557 
3558 	for (msr = 0x800; msr <= 0x8ff; msr += BITS_PER_LONG) {
3559 		unsigned word = msr / BITS_PER_LONG;
3560 		msr_bitmap[word] = (mode & MSR_BITMAP_MODE_X2APIC_APICV) ? 0 : ~0;
3561 		msr_bitmap[word + (0x800 / sizeof(long))] = ~0;
3562 	}
3563 
3564 	if (mode & MSR_BITMAP_MODE_X2APIC) {
3565 		/*
3566 		 * TPR reads and writes can be virtualized even if virtual interrupt
3567 		 * delivery is not in use.
3568 		 */
3569 		vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_TASKPRI), MSR_TYPE_RW);
3570 		if (mode & MSR_BITMAP_MODE_X2APIC_APICV) {
3571 			vmx_enable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_TMCCT), MSR_TYPE_R);
3572 			vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_EOI), MSR_TYPE_W);
3573 			vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_SELF_IPI), MSR_TYPE_W);
3574 		}
3575 	}
3576 }
3577 
3578 void vmx_update_msr_bitmap(struct kvm_vcpu *vcpu)
3579 {
3580 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3581 	unsigned long *msr_bitmap = vmx->vmcs01.msr_bitmap;
3582 	u8 mode = vmx_msr_bitmap_mode(vcpu);
3583 	u8 changed = mode ^ vmx->msr_bitmap_mode;
3584 
3585 	if (!changed)
3586 		return;
3587 
3588 	if (changed & (MSR_BITMAP_MODE_X2APIC | MSR_BITMAP_MODE_X2APIC_APICV))
3589 		vmx_update_msr_bitmap_x2apic(msr_bitmap, mode);
3590 
3591 	vmx->msr_bitmap_mode = mode;
3592 }
3593 
3594 void pt_update_intercept_for_msr(struct vcpu_vmx *vmx)
3595 {
3596 	unsigned long *msr_bitmap = vmx->vmcs01.msr_bitmap;
3597 	bool flag = !(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN);
3598 	u32 i;
3599 
3600 	vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_STATUS,
3601 							MSR_TYPE_RW, flag);
3602 	vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_OUTPUT_BASE,
3603 							MSR_TYPE_RW, flag);
3604 	vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_OUTPUT_MASK,
3605 							MSR_TYPE_RW, flag);
3606 	vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_CR3_MATCH,
3607 							MSR_TYPE_RW, flag);
3608 	for (i = 0; i < vmx->pt_desc.addr_range; i++) {
3609 		vmx_set_intercept_for_msr(msr_bitmap,
3610 			MSR_IA32_RTIT_ADDR0_A + i * 2, MSR_TYPE_RW, flag);
3611 		vmx_set_intercept_for_msr(msr_bitmap,
3612 			MSR_IA32_RTIT_ADDR0_B + i * 2, MSR_TYPE_RW, flag);
3613 	}
3614 }
3615 
3616 static bool vmx_get_enable_apicv(struct kvm_vcpu *vcpu)
3617 {
3618 	return enable_apicv;
3619 }
3620 
3621 static bool vmx_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
3622 {
3623 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3624 	void *vapic_page;
3625 	u32 vppr;
3626 	int rvi;
3627 
3628 	if (WARN_ON_ONCE(!is_guest_mode(vcpu)) ||
3629 		!nested_cpu_has_vid(get_vmcs12(vcpu)) ||
3630 		WARN_ON_ONCE(!vmx->nested.virtual_apic_page))
3631 		return false;
3632 
3633 	rvi = vmx_get_rvi();
3634 
3635 	vapic_page = kmap(vmx->nested.virtual_apic_page);
3636 	vppr = *((u32 *)(vapic_page + APIC_PROCPRI));
3637 	kunmap(vmx->nested.virtual_apic_page);
3638 
3639 	return ((rvi & 0xf0) > (vppr & 0xf0));
3640 }
3641 
3642 static inline bool kvm_vcpu_trigger_posted_interrupt(struct kvm_vcpu *vcpu,
3643 						     bool nested)
3644 {
3645 #ifdef CONFIG_SMP
3646 	int pi_vec = nested ? POSTED_INTR_NESTED_VECTOR : POSTED_INTR_VECTOR;
3647 
3648 	if (vcpu->mode == IN_GUEST_MODE) {
3649 		/*
3650 		 * The vector of interrupt to be delivered to vcpu had
3651 		 * been set in PIR before this function.
3652 		 *
3653 		 * Following cases will be reached in this block, and
3654 		 * we always send a notification event in all cases as
3655 		 * explained below.
3656 		 *
3657 		 * Case 1: vcpu keeps in non-root mode. Sending a
3658 		 * notification event posts the interrupt to vcpu.
3659 		 *
3660 		 * Case 2: vcpu exits to root mode and is still
3661 		 * runnable. PIR will be synced to vIRR before the
3662 		 * next vcpu entry. Sending a notification event in
3663 		 * this case has no effect, as vcpu is not in root
3664 		 * mode.
3665 		 *
3666 		 * Case 3: vcpu exits to root mode and is blocked.
3667 		 * vcpu_block() has already synced PIR to vIRR and
3668 		 * never blocks vcpu if vIRR is not cleared. Therefore,
3669 		 * a blocked vcpu here does not wait for any requested
3670 		 * interrupts in PIR, and sending a notification event
3671 		 * which has no effect is safe here.
3672 		 */
3673 
3674 		apic->send_IPI_mask(get_cpu_mask(vcpu->cpu), pi_vec);
3675 		return true;
3676 	}
3677 #endif
3678 	return false;
3679 }
3680 
3681 static int vmx_deliver_nested_posted_interrupt(struct kvm_vcpu *vcpu,
3682 						int vector)
3683 {
3684 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3685 
3686 	if (is_guest_mode(vcpu) &&
3687 	    vector == vmx->nested.posted_intr_nv) {
3688 		/*
3689 		 * If a posted intr is not recognized by hardware,
3690 		 * we will accomplish it in the next vmentry.
3691 		 */
3692 		vmx->nested.pi_pending = true;
3693 		kvm_make_request(KVM_REQ_EVENT, vcpu);
3694 		/* the PIR and ON have been set by L1. */
3695 		if (!kvm_vcpu_trigger_posted_interrupt(vcpu, true))
3696 			kvm_vcpu_kick(vcpu);
3697 		return 0;
3698 	}
3699 	return -1;
3700 }
3701 /*
3702  * Send interrupt to vcpu via posted interrupt way.
3703  * 1. If target vcpu is running(non-root mode), send posted interrupt
3704  * notification to vcpu and hardware will sync PIR to vIRR atomically.
3705  * 2. If target vcpu isn't running(root mode), kick it to pick up the
3706  * interrupt from PIR in next vmentry.
3707  */
3708 static void vmx_deliver_posted_interrupt(struct kvm_vcpu *vcpu, int vector)
3709 {
3710 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3711 	int r;
3712 
3713 	r = vmx_deliver_nested_posted_interrupt(vcpu, vector);
3714 	if (!r)
3715 		return;
3716 
3717 	if (pi_test_and_set_pir(vector, &vmx->pi_desc))
3718 		return;
3719 
3720 	/* If a previous notification has sent the IPI, nothing to do.  */
3721 	if (pi_test_and_set_on(&vmx->pi_desc))
3722 		return;
3723 
3724 	if (!kvm_vcpu_trigger_posted_interrupt(vcpu, false))
3725 		kvm_vcpu_kick(vcpu);
3726 }
3727 
3728 /*
3729  * Set up the vmcs's constant host-state fields, i.e., host-state fields that
3730  * will not change in the lifetime of the guest.
3731  * Note that host-state that does change is set elsewhere. E.g., host-state
3732  * that is set differently for each CPU is set in vmx_vcpu_load(), not here.
3733  */
3734 void vmx_set_constant_host_state(struct vcpu_vmx *vmx)
3735 {
3736 	u32 low32, high32;
3737 	unsigned long tmpl;
3738 	struct desc_ptr dt;
3739 	unsigned long cr0, cr3, cr4;
3740 
3741 	cr0 = read_cr0();
3742 	WARN_ON(cr0 & X86_CR0_TS);
3743 	vmcs_writel(HOST_CR0, cr0);  /* 22.2.3 */
3744 
3745 	/*
3746 	 * Save the most likely value for this task's CR3 in the VMCS.
3747 	 * We can't use __get_current_cr3_fast() because we're not atomic.
3748 	 */
3749 	cr3 = __read_cr3();
3750 	vmcs_writel(HOST_CR3, cr3);		/* 22.2.3  FIXME: shadow tables */
3751 	vmx->loaded_vmcs->host_state.cr3 = cr3;
3752 
3753 	/* Save the most likely value for this task's CR4 in the VMCS. */
3754 	cr4 = cr4_read_shadow();
3755 	vmcs_writel(HOST_CR4, cr4);			/* 22.2.3, 22.2.5 */
3756 	vmx->loaded_vmcs->host_state.cr4 = cr4;
3757 
3758 	vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS);  /* 22.2.4 */
3759 #ifdef CONFIG_X86_64
3760 	/*
3761 	 * Load null selectors, so we can avoid reloading them in
3762 	 * vmx_prepare_switch_to_host(), in case userspace uses
3763 	 * the null selectors too (the expected case).
3764 	 */
3765 	vmcs_write16(HOST_DS_SELECTOR, 0);
3766 	vmcs_write16(HOST_ES_SELECTOR, 0);
3767 #else
3768 	vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
3769 	vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
3770 #endif
3771 	vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
3772 	vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8);  /* 22.2.4 */
3773 
3774 	store_idt(&dt);
3775 	vmcs_writel(HOST_IDTR_BASE, dt.address);   /* 22.2.4 */
3776 	vmx->host_idt_base = dt.address;
3777 
3778 	vmcs_writel(HOST_RIP, (unsigned long)vmx_vmexit); /* 22.2.5 */
3779 
3780 	rdmsr(MSR_IA32_SYSENTER_CS, low32, high32);
3781 	vmcs_write32(HOST_IA32_SYSENTER_CS, low32);
3782 	rdmsrl(MSR_IA32_SYSENTER_EIP, tmpl);
3783 	vmcs_writel(HOST_IA32_SYSENTER_EIP, tmpl);   /* 22.2.3 */
3784 
3785 	if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
3786 		rdmsr(MSR_IA32_CR_PAT, low32, high32);
3787 		vmcs_write64(HOST_IA32_PAT, low32 | ((u64) high32 << 32));
3788 	}
3789 
3790 	if (cpu_has_load_ia32_efer())
3791 		vmcs_write64(HOST_IA32_EFER, host_efer);
3792 }
3793 
3794 void set_cr4_guest_host_mask(struct vcpu_vmx *vmx)
3795 {
3796 	vmx->vcpu.arch.cr4_guest_owned_bits = KVM_CR4_GUEST_OWNED_BITS;
3797 	if (enable_ept)
3798 		vmx->vcpu.arch.cr4_guest_owned_bits |= X86_CR4_PGE;
3799 	if (is_guest_mode(&vmx->vcpu))
3800 		vmx->vcpu.arch.cr4_guest_owned_bits &=
3801 			~get_vmcs12(&vmx->vcpu)->cr4_guest_host_mask;
3802 	vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits);
3803 }
3804 
3805 static u32 vmx_pin_based_exec_ctrl(struct vcpu_vmx *vmx)
3806 {
3807 	u32 pin_based_exec_ctrl = vmcs_config.pin_based_exec_ctrl;
3808 
3809 	if (!kvm_vcpu_apicv_active(&vmx->vcpu))
3810 		pin_based_exec_ctrl &= ~PIN_BASED_POSTED_INTR;
3811 
3812 	if (!enable_vnmi)
3813 		pin_based_exec_ctrl &= ~PIN_BASED_VIRTUAL_NMIS;
3814 
3815 	/* Enable the preemption timer dynamically */
3816 	pin_based_exec_ctrl &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
3817 	return pin_based_exec_ctrl;
3818 }
3819 
3820 static void vmx_refresh_apicv_exec_ctrl(struct kvm_vcpu *vcpu)
3821 {
3822 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3823 
3824 	vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, vmx_pin_based_exec_ctrl(vmx));
3825 	if (cpu_has_secondary_exec_ctrls()) {
3826 		if (kvm_vcpu_apicv_active(vcpu))
3827 			vmcs_set_bits(SECONDARY_VM_EXEC_CONTROL,
3828 				      SECONDARY_EXEC_APIC_REGISTER_VIRT |
3829 				      SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
3830 		else
3831 			vmcs_clear_bits(SECONDARY_VM_EXEC_CONTROL,
3832 					SECONDARY_EXEC_APIC_REGISTER_VIRT |
3833 					SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
3834 	}
3835 
3836 	if (cpu_has_vmx_msr_bitmap())
3837 		vmx_update_msr_bitmap(vcpu);
3838 }
3839 
3840 u32 vmx_exec_control(struct vcpu_vmx *vmx)
3841 {
3842 	u32 exec_control = vmcs_config.cpu_based_exec_ctrl;
3843 
3844 	if (vmx->vcpu.arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)
3845 		exec_control &= ~CPU_BASED_MOV_DR_EXITING;
3846 
3847 	if (!cpu_need_tpr_shadow(&vmx->vcpu)) {
3848 		exec_control &= ~CPU_BASED_TPR_SHADOW;
3849 #ifdef CONFIG_X86_64
3850 		exec_control |= CPU_BASED_CR8_STORE_EXITING |
3851 				CPU_BASED_CR8_LOAD_EXITING;
3852 #endif
3853 	}
3854 	if (!enable_ept)
3855 		exec_control |= CPU_BASED_CR3_STORE_EXITING |
3856 				CPU_BASED_CR3_LOAD_EXITING  |
3857 				CPU_BASED_INVLPG_EXITING;
3858 	if (kvm_mwait_in_guest(vmx->vcpu.kvm))
3859 		exec_control &= ~(CPU_BASED_MWAIT_EXITING |
3860 				CPU_BASED_MONITOR_EXITING);
3861 	if (kvm_hlt_in_guest(vmx->vcpu.kvm))
3862 		exec_control &= ~CPU_BASED_HLT_EXITING;
3863 	return exec_control;
3864 }
3865 
3866 
3867 static void vmx_compute_secondary_exec_control(struct vcpu_vmx *vmx)
3868 {
3869 	struct kvm_vcpu *vcpu = &vmx->vcpu;
3870 
3871 	u32 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
3872 
3873 	if (pt_mode == PT_MODE_SYSTEM)
3874 		exec_control &= ~(SECONDARY_EXEC_PT_USE_GPA | SECONDARY_EXEC_PT_CONCEAL_VMX);
3875 	if (!cpu_need_virtualize_apic_accesses(vcpu))
3876 		exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
3877 	if (vmx->vpid == 0)
3878 		exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
3879 	if (!enable_ept) {
3880 		exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
3881 		enable_unrestricted_guest = 0;
3882 	}
3883 	if (!enable_unrestricted_guest)
3884 		exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
3885 	if (kvm_pause_in_guest(vmx->vcpu.kvm))
3886 		exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
3887 	if (!kvm_vcpu_apicv_active(vcpu))
3888 		exec_control &= ~(SECONDARY_EXEC_APIC_REGISTER_VIRT |
3889 				  SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
3890 	exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
3891 
3892 	/* SECONDARY_EXEC_DESC is enabled/disabled on writes to CR4.UMIP,
3893 	 * in vmx_set_cr4.  */
3894 	exec_control &= ~SECONDARY_EXEC_DESC;
3895 
3896 	/* SECONDARY_EXEC_SHADOW_VMCS is enabled when L1 executes VMPTRLD
3897 	   (handle_vmptrld).
3898 	   We can NOT enable shadow_vmcs here because we don't have yet
3899 	   a current VMCS12
3900 	*/
3901 	exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
3902 
3903 	if (!enable_pml)
3904 		exec_control &= ~SECONDARY_EXEC_ENABLE_PML;
3905 
3906 	if (vmx_xsaves_supported()) {
3907 		/* Exposing XSAVES only when XSAVE is exposed */
3908 		bool xsaves_enabled =
3909 			guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
3910 			guest_cpuid_has(vcpu, X86_FEATURE_XSAVES);
3911 
3912 		if (!xsaves_enabled)
3913 			exec_control &= ~SECONDARY_EXEC_XSAVES;
3914 
3915 		if (nested) {
3916 			if (xsaves_enabled)
3917 				vmx->nested.msrs.secondary_ctls_high |=
3918 					SECONDARY_EXEC_XSAVES;
3919 			else
3920 				vmx->nested.msrs.secondary_ctls_high &=
3921 					~SECONDARY_EXEC_XSAVES;
3922 		}
3923 	}
3924 
3925 	if (vmx_rdtscp_supported()) {
3926 		bool rdtscp_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP);
3927 		if (!rdtscp_enabled)
3928 			exec_control &= ~SECONDARY_EXEC_RDTSCP;
3929 
3930 		if (nested) {
3931 			if (rdtscp_enabled)
3932 				vmx->nested.msrs.secondary_ctls_high |=
3933 					SECONDARY_EXEC_RDTSCP;
3934 			else
3935 				vmx->nested.msrs.secondary_ctls_high &=
3936 					~SECONDARY_EXEC_RDTSCP;
3937 		}
3938 	}
3939 
3940 	if (vmx_invpcid_supported()) {
3941 		/* Exposing INVPCID only when PCID is exposed */
3942 		bool invpcid_enabled =
3943 			guest_cpuid_has(vcpu, X86_FEATURE_INVPCID) &&
3944 			guest_cpuid_has(vcpu, X86_FEATURE_PCID);
3945 
3946 		if (!invpcid_enabled) {
3947 			exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID;
3948 			guest_cpuid_clear(vcpu, X86_FEATURE_INVPCID);
3949 		}
3950 
3951 		if (nested) {
3952 			if (invpcid_enabled)
3953 				vmx->nested.msrs.secondary_ctls_high |=
3954 					SECONDARY_EXEC_ENABLE_INVPCID;
3955 			else
3956 				vmx->nested.msrs.secondary_ctls_high &=
3957 					~SECONDARY_EXEC_ENABLE_INVPCID;
3958 		}
3959 	}
3960 
3961 	if (vmx_rdrand_supported()) {
3962 		bool rdrand_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDRAND);
3963 		if (rdrand_enabled)
3964 			exec_control &= ~SECONDARY_EXEC_RDRAND_EXITING;
3965 
3966 		if (nested) {
3967 			if (rdrand_enabled)
3968 				vmx->nested.msrs.secondary_ctls_high |=
3969 					SECONDARY_EXEC_RDRAND_EXITING;
3970 			else
3971 				vmx->nested.msrs.secondary_ctls_high &=
3972 					~SECONDARY_EXEC_RDRAND_EXITING;
3973 		}
3974 	}
3975 
3976 	if (vmx_rdseed_supported()) {
3977 		bool rdseed_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDSEED);
3978 		if (rdseed_enabled)
3979 			exec_control &= ~SECONDARY_EXEC_RDSEED_EXITING;
3980 
3981 		if (nested) {
3982 			if (rdseed_enabled)
3983 				vmx->nested.msrs.secondary_ctls_high |=
3984 					SECONDARY_EXEC_RDSEED_EXITING;
3985 			else
3986 				vmx->nested.msrs.secondary_ctls_high &=
3987 					~SECONDARY_EXEC_RDSEED_EXITING;
3988 		}
3989 	}
3990 
3991 	vmx->secondary_exec_control = exec_control;
3992 }
3993 
3994 static void ept_set_mmio_spte_mask(void)
3995 {
3996 	/*
3997 	 * EPT Misconfigurations can be generated if the value of bits 2:0
3998 	 * of an EPT paging-structure entry is 110b (write/execute).
3999 	 */
4000 	kvm_mmu_set_mmio_spte_mask(VMX_EPT_RWX_MASK,
4001 				   VMX_EPT_MISCONFIG_WX_VALUE);
4002 }
4003 
4004 #define VMX_XSS_EXIT_BITMAP 0
4005 
4006 /*
4007  * Sets up the vmcs for emulated real mode.
4008  */
4009 static void vmx_vcpu_setup(struct vcpu_vmx *vmx)
4010 {
4011 	int i;
4012 
4013 	if (nested)
4014 		nested_vmx_vcpu_setup();
4015 
4016 	if (cpu_has_vmx_msr_bitmap())
4017 		vmcs_write64(MSR_BITMAP, __pa(vmx->vmcs01.msr_bitmap));
4018 
4019 	vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
4020 
4021 	/* Control */
4022 	vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, vmx_pin_based_exec_ctrl(vmx));
4023 	vmx->hv_deadline_tsc = -1;
4024 
4025 	vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, vmx_exec_control(vmx));
4026 
4027 	if (cpu_has_secondary_exec_ctrls()) {
4028 		vmx_compute_secondary_exec_control(vmx);
4029 		vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
4030 			     vmx->secondary_exec_control);
4031 	}
4032 
4033 	if (kvm_vcpu_apicv_active(&vmx->vcpu)) {
4034 		vmcs_write64(EOI_EXIT_BITMAP0, 0);
4035 		vmcs_write64(EOI_EXIT_BITMAP1, 0);
4036 		vmcs_write64(EOI_EXIT_BITMAP2, 0);
4037 		vmcs_write64(EOI_EXIT_BITMAP3, 0);
4038 
4039 		vmcs_write16(GUEST_INTR_STATUS, 0);
4040 
4041 		vmcs_write16(POSTED_INTR_NV, POSTED_INTR_VECTOR);
4042 		vmcs_write64(POSTED_INTR_DESC_ADDR, __pa((&vmx->pi_desc)));
4043 	}
4044 
4045 	if (!kvm_pause_in_guest(vmx->vcpu.kvm)) {
4046 		vmcs_write32(PLE_GAP, ple_gap);
4047 		vmx->ple_window = ple_window;
4048 		vmx->ple_window_dirty = true;
4049 	}
4050 
4051 	vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
4052 	vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
4053 	vmcs_write32(CR3_TARGET_COUNT, 0);           /* 22.2.1 */
4054 
4055 	vmcs_write16(HOST_FS_SELECTOR, 0);            /* 22.2.4 */
4056 	vmcs_write16(HOST_GS_SELECTOR, 0);            /* 22.2.4 */
4057 	vmx_set_constant_host_state(vmx);
4058 	vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
4059 	vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
4060 
4061 	if (cpu_has_vmx_vmfunc())
4062 		vmcs_write64(VM_FUNCTION_CONTROL, 0);
4063 
4064 	vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
4065 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
4066 	vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host.val));
4067 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
4068 	vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest.val));
4069 
4070 	if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
4071 		vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
4072 
4073 	for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i) {
4074 		u32 index = vmx_msr_index[i];
4075 		u32 data_low, data_high;
4076 		int j = vmx->nmsrs;
4077 
4078 		if (rdmsr_safe(index, &data_low, &data_high) < 0)
4079 			continue;
4080 		if (wrmsr_safe(index, data_low, data_high) < 0)
4081 			continue;
4082 		vmx->guest_msrs[j].index = i;
4083 		vmx->guest_msrs[j].data = 0;
4084 		vmx->guest_msrs[j].mask = -1ull;
4085 		++vmx->nmsrs;
4086 	}
4087 
4088 	vmx->arch_capabilities = kvm_get_arch_capabilities();
4089 
4090 	vm_exit_controls_init(vmx, vmx_vmexit_ctrl());
4091 
4092 	/* 22.2.1, 20.8.1 */
4093 	vm_entry_controls_init(vmx, vmx_vmentry_ctrl());
4094 
4095 	vmx->vcpu.arch.cr0_guest_owned_bits = X86_CR0_TS;
4096 	vmcs_writel(CR0_GUEST_HOST_MASK, ~X86_CR0_TS);
4097 
4098 	set_cr4_guest_host_mask(vmx);
4099 
4100 	if (vmx_xsaves_supported())
4101 		vmcs_write64(XSS_EXIT_BITMAP, VMX_XSS_EXIT_BITMAP);
4102 
4103 	if (enable_pml) {
4104 		vmcs_write64(PML_ADDRESS, page_to_phys(vmx->pml_pg));
4105 		vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
4106 	}
4107 
4108 	if (cpu_has_vmx_encls_vmexit())
4109 		vmcs_write64(ENCLS_EXITING_BITMAP, -1ull);
4110 
4111 	if (pt_mode == PT_MODE_HOST_GUEST) {
4112 		memset(&vmx->pt_desc, 0, sizeof(vmx->pt_desc));
4113 		/* Bit[6~0] are forced to 1, writes are ignored. */
4114 		vmx->pt_desc.guest.output_mask = 0x7F;
4115 		vmcs_write64(GUEST_IA32_RTIT_CTL, 0);
4116 	}
4117 }
4118 
4119 static void vmx_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
4120 {
4121 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4122 	struct msr_data apic_base_msr;
4123 	u64 cr0;
4124 
4125 	vmx->rmode.vm86_active = 0;
4126 	vmx->spec_ctrl = 0;
4127 
4128 	vcpu->arch.microcode_version = 0x100000000ULL;
4129 	vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
4130 	kvm_set_cr8(vcpu, 0);
4131 
4132 	if (!init_event) {
4133 		apic_base_msr.data = APIC_DEFAULT_PHYS_BASE |
4134 				     MSR_IA32_APICBASE_ENABLE;
4135 		if (kvm_vcpu_is_reset_bsp(vcpu))
4136 			apic_base_msr.data |= MSR_IA32_APICBASE_BSP;
4137 		apic_base_msr.host_initiated = true;
4138 		kvm_set_apic_base(vcpu, &apic_base_msr);
4139 	}
4140 
4141 	vmx_segment_cache_clear(vmx);
4142 
4143 	seg_setup(VCPU_SREG_CS);
4144 	vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
4145 	vmcs_writel(GUEST_CS_BASE, 0xffff0000ul);
4146 
4147 	seg_setup(VCPU_SREG_DS);
4148 	seg_setup(VCPU_SREG_ES);
4149 	seg_setup(VCPU_SREG_FS);
4150 	seg_setup(VCPU_SREG_GS);
4151 	seg_setup(VCPU_SREG_SS);
4152 
4153 	vmcs_write16(GUEST_TR_SELECTOR, 0);
4154 	vmcs_writel(GUEST_TR_BASE, 0);
4155 	vmcs_write32(GUEST_TR_LIMIT, 0xffff);
4156 	vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
4157 
4158 	vmcs_write16(GUEST_LDTR_SELECTOR, 0);
4159 	vmcs_writel(GUEST_LDTR_BASE, 0);
4160 	vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
4161 	vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
4162 
4163 	if (!init_event) {
4164 		vmcs_write32(GUEST_SYSENTER_CS, 0);
4165 		vmcs_writel(GUEST_SYSENTER_ESP, 0);
4166 		vmcs_writel(GUEST_SYSENTER_EIP, 0);
4167 		vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
4168 	}
4169 
4170 	kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
4171 	kvm_rip_write(vcpu, 0xfff0);
4172 
4173 	vmcs_writel(GUEST_GDTR_BASE, 0);
4174 	vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
4175 
4176 	vmcs_writel(GUEST_IDTR_BASE, 0);
4177 	vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
4178 
4179 	vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
4180 	vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
4181 	vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS, 0);
4182 	if (kvm_mpx_supported())
4183 		vmcs_write64(GUEST_BNDCFGS, 0);
4184 
4185 	setup_msrs(vmx);
4186 
4187 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);  /* 22.2.1 */
4188 
4189 	if (cpu_has_vmx_tpr_shadow() && !init_event) {
4190 		vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
4191 		if (cpu_need_tpr_shadow(vcpu))
4192 			vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
4193 				     __pa(vcpu->arch.apic->regs));
4194 		vmcs_write32(TPR_THRESHOLD, 0);
4195 	}
4196 
4197 	kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
4198 
4199 	if (vmx->vpid != 0)
4200 		vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
4201 
4202 	cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
4203 	vmx->vcpu.arch.cr0 = cr0;
4204 	vmx_set_cr0(vcpu, cr0); /* enter rmode */
4205 	vmx_set_cr4(vcpu, 0);
4206 	vmx_set_efer(vcpu, 0);
4207 
4208 	update_exception_bitmap(vcpu);
4209 
4210 	vpid_sync_context(vmx->vpid);
4211 	if (init_event)
4212 		vmx_clear_hlt(vcpu);
4213 }
4214 
4215 static void enable_irq_window(struct kvm_vcpu *vcpu)
4216 {
4217 	vmcs_set_bits(CPU_BASED_VM_EXEC_CONTROL,
4218 		      CPU_BASED_VIRTUAL_INTR_PENDING);
4219 }
4220 
4221 static void enable_nmi_window(struct kvm_vcpu *vcpu)
4222 {
4223 	if (!enable_vnmi ||
4224 	    vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_STI) {
4225 		enable_irq_window(vcpu);
4226 		return;
4227 	}
4228 
4229 	vmcs_set_bits(CPU_BASED_VM_EXEC_CONTROL,
4230 		      CPU_BASED_VIRTUAL_NMI_PENDING);
4231 }
4232 
4233 static void vmx_inject_irq(struct kvm_vcpu *vcpu)
4234 {
4235 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4236 	uint32_t intr;
4237 	int irq = vcpu->arch.interrupt.nr;
4238 
4239 	trace_kvm_inj_virq(irq);
4240 
4241 	++vcpu->stat.irq_injections;
4242 	if (vmx->rmode.vm86_active) {
4243 		int inc_eip = 0;
4244 		if (vcpu->arch.interrupt.soft)
4245 			inc_eip = vcpu->arch.event_exit_inst_len;
4246 		if (kvm_inject_realmode_interrupt(vcpu, irq, inc_eip) != EMULATE_DONE)
4247 			kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4248 		return;
4249 	}
4250 	intr = irq | INTR_INFO_VALID_MASK;
4251 	if (vcpu->arch.interrupt.soft) {
4252 		intr |= INTR_TYPE_SOFT_INTR;
4253 		vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
4254 			     vmx->vcpu.arch.event_exit_inst_len);
4255 	} else
4256 		intr |= INTR_TYPE_EXT_INTR;
4257 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
4258 
4259 	vmx_clear_hlt(vcpu);
4260 }
4261 
4262 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
4263 {
4264 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4265 
4266 	if (!enable_vnmi) {
4267 		/*
4268 		 * Tracking the NMI-blocked state in software is built upon
4269 		 * finding the next open IRQ window. This, in turn, depends on
4270 		 * well-behaving guests: They have to keep IRQs disabled at
4271 		 * least as long as the NMI handler runs. Otherwise we may
4272 		 * cause NMI nesting, maybe breaking the guest. But as this is
4273 		 * highly unlikely, we can live with the residual risk.
4274 		 */
4275 		vmx->loaded_vmcs->soft_vnmi_blocked = 1;
4276 		vmx->loaded_vmcs->vnmi_blocked_time = 0;
4277 	}
4278 
4279 	++vcpu->stat.nmi_injections;
4280 	vmx->loaded_vmcs->nmi_known_unmasked = false;
4281 
4282 	if (vmx->rmode.vm86_active) {
4283 		if (kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR, 0) != EMULATE_DONE)
4284 			kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4285 		return;
4286 	}
4287 
4288 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
4289 			INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
4290 
4291 	vmx_clear_hlt(vcpu);
4292 }
4293 
4294 bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
4295 {
4296 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4297 	bool masked;
4298 
4299 	if (!enable_vnmi)
4300 		return vmx->loaded_vmcs->soft_vnmi_blocked;
4301 	if (vmx->loaded_vmcs->nmi_known_unmasked)
4302 		return false;
4303 	masked = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI;
4304 	vmx->loaded_vmcs->nmi_known_unmasked = !masked;
4305 	return masked;
4306 }
4307 
4308 void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
4309 {
4310 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4311 
4312 	if (!enable_vnmi) {
4313 		if (vmx->loaded_vmcs->soft_vnmi_blocked != masked) {
4314 			vmx->loaded_vmcs->soft_vnmi_blocked = masked;
4315 			vmx->loaded_vmcs->vnmi_blocked_time = 0;
4316 		}
4317 	} else {
4318 		vmx->loaded_vmcs->nmi_known_unmasked = !masked;
4319 		if (masked)
4320 			vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
4321 				      GUEST_INTR_STATE_NMI);
4322 		else
4323 			vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
4324 					GUEST_INTR_STATE_NMI);
4325 	}
4326 }
4327 
4328 static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
4329 {
4330 	if (to_vmx(vcpu)->nested.nested_run_pending)
4331 		return 0;
4332 
4333 	if (!enable_vnmi &&
4334 	    to_vmx(vcpu)->loaded_vmcs->soft_vnmi_blocked)
4335 		return 0;
4336 
4337 	return	!(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4338 		  (GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_STI
4339 		   | GUEST_INTR_STATE_NMI));
4340 }
4341 
4342 static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
4343 {
4344 	return (!to_vmx(vcpu)->nested.nested_run_pending &&
4345 		vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
4346 		!(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4347 			(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
4348 }
4349 
4350 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
4351 {
4352 	int ret;
4353 
4354 	if (enable_unrestricted_guest)
4355 		return 0;
4356 
4357 	ret = x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, addr,
4358 				    PAGE_SIZE * 3);
4359 	if (ret)
4360 		return ret;
4361 	to_kvm_vmx(kvm)->tss_addr = addr;
4362 	return init_rmode_tss(kvm);
4363 }
4364 
4365 static int vmx_set_identity_map_addr(struct kvm *kvm, u64 ident_addr)
4366 {
4367 	to_kvm_vmx(kvm)->ept_identity_map_addr = ident_addr;
4368 	return 0;
4369 }
4370 
4371 static bool rmode_exception(struct kvm_vcpu *vcpu, int vec)
4372 {
4373 	switch (vec) {
4374 	case BP_VECTOR:
4375 		/*
4376 		 * Update instruction length as we may reinject the exception
4377 		 * from user space while in guest debugging mode.
4378 		 */
4379 		to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
4380 			vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4381 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
4382 			return false;
4383 		/* fall through */
4384 	case DB_VECTOR:
4385 		if (vcpu->guest_debug &
4386 			(KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
4387 			return false;
4388 		/* fall through */
4389 	case DE_VECTOR:
4390 	case OF_VECTOR:
4391 	case BR_VECTOR:
4392 	case UD_VECTOR:
4393 	case DF_VECTOR:
4394 	case SS_VECTOR:
4395 	case GP_VECTOR:
4396 	case MF_VECTOR:
4397 		return true;
4398 	break;
4399 	}
4400 	return false;
4401 }
4402 
4403 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
4404 				  int vec, u32 err_code)
4405 {
4406 	/*
4407 	 * Instruction with address size override prefix opcode 0x67
4408 	 * Cause the #SS fault with 0 error code in VM86 mode.
4409 	 */
4410 	if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0) {
4411 		if (kvm_emulate_instruction(vcpu, 0) == EMULATE_DONE) {
4412 			if (vcpu->arch.halt_request) {
4413 				vcpu->arch.halt_request = 0;
4414 				return kvm_vcpu_halt(vcpu);
4415 			}
4416 			return 1;
4417 		}
4418 		return 0;
4419 	}
4420 
4421 	/*
4422 	 * Forward all other exceptions that are valid in real mode.
4423 	 * FIXME: Breaks guest debugging in real mode, needs to be fixed with
4424 	 *        the required debugging infrastructure rework.
4425 	 */
4426 	kvm_queue_exception(vcpu, vec);
4427 	return 1;
4428 }
4429 
4430 /*
4431  * Trigger machine check on the host. We assume all the MSRs are already set up
4432  * by the CPU and that we still run on the same CPU as the MCE occurred on.
4433  * We pass a fake environment to the machine check handler because we want
4434  * the guest to be always treated like user space, no matter what context
4435  * it used internally.
4436  */
4437 static void kvm_machine_check(void)
4438 {
4439 #if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
4440 	struct pt_regs regs = {
4441 		.cs = 3, /* Fake ring 3 no matter what the guest ran on */
4442 		.flags = X86_EFLAGS_IF,
4443 	};
4444 
4445 	do_machine_check(&regs, 0);
4446 #endif
4447 }
4448 
4449 static int handle_machine_check(struct kvm_vcpu *vcpu)
4450 {
4451 	/* already handled by vcpu_run */
4452 	return 1;
4453 }
4454 
4455 static int handle_exception(struct kvm_vcpu *vcpu)
4456 {
4457 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4458 	struct kvm_run *kvm_run = vcpu->run;
4459 	u32 intr_info, ex_no, error_code;
4460 	unsigned long cr2, rip, dr6;
4461 	u32 vect_info;
4462 	enum emulation_result er;
4463 
4464 	vect_info = vmx->idt_vectoring_info;
4465 	intr_info = vmx->exit_intr_info;
4466 
4467 	if (is_machine_check(intr_info))
4468 		return handle_machine_check(vcpu);
4469 
4470 	if (is_nmi(intr_info))
4471 		return 1;  /* already handled by vmx_vcpu_run() */
4472 
4473 	if (is_invalid_opcode(intr_info))
4474 		return handle_ud(vcpu);
4475 
4476 	error_code = 0;
4477 	if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
4478 		error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
4479 
4480 	if (!vmx->rmode.vm86_active && is_gp_fault(intr_info)) {
4481 		WARN_ON_ONCE(!enable_vmware_backdoor);
4482 		er = kvm_emulate_instruction(vcpu,
4483 			EMULTYPE_VMWARE | EMULTYPE_NO_UD_ON_FAIL);
4484 		if (er == EMULATE_USER_EXIT)
4485 			return 0;
4486 		else if (er != EMULATE_DONE)
4487 			kvm_queue_exception_e(vcpu, GP_VECTOR, error_code);
4488 		return 1;
4489 	}
4490 
4491 	/*
4492 	 * The #PF with PFEC.RSVD = 1 indicates the guest is accessing
4493 	 * MMIO, it is better to report an internal error.
4494 	 * See the comments in vmx_handle_exit.
4495 	 */
4496 	if ((vect_info & VECTORING_INFO_VALID_MASK) &&
4497 	    !(is_page_fault(intr_info) && !(error_code & PFERR_RSVD_MASK))) {
4498 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4499 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
4500 		vcpu->run->internal.ndata = 3;
4501 		vcpu->run->internal.data[0] = vect_info;
4502 		vcpu->run->internal.data[1] = intr_info;
4503 		vcpu->run->internal.data[2] = error_code;
4504 		return 0;
4505 	}
4506 
4507 	if (is_page_fault(intr_info)) {
4508 		cr2 = vmcs_readl(EXIT_QUALIFICATION);
4509 		/* EPT won't cause page fault directly */
4510 		WARN_ON_ONCE(!vcpu->arch.apf.host_apf_reason && enable_ept);
4511 		return kvm_handle_page_fault(vcpu, error_code, cr2, NULL, 0);
4512 	}
4513 
4514 	ex_no = intr_info & INTR_INFO_VECTOR_MASK;
4515 
4516 	if (vmx->rmode.vm86_active && rmode_exception(vcpu, ex_no))
4517 		return handle_rmode_exception(vcpu, ex_no, error_code);
4518 
4519 	switch (ex_no) {
4520 	case AC_VECTOR:
4521 		kvm_queue_exception_e(vcpu, AC_VECTOR, error_code);
4522 		return 1;
4523 	case DB_VECTOR:
4524 		dr6 = vmcs_readl(EXIT_QUALIFICATION);
4525 		if (!(vcpu->guest_debug &
4526 		      (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
4527 			vcpu->arch.dr6 &= ~15;
4528 			vcpu->arch.dr6 |= dr6 | DR6_RTM;
4529 			if (is_icebp(intr_info))
4530 				skip_emulated_instruction(vcpu);
4531 
4532 			kvm_queue_exception(vcpu, DB_VECTOR);
4533 			return 1;
4534 		}
4535 		kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
4536 		kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
4537 		/* fall through */
4538 	case BP_VECTOR:
4539 		/*
4540 		 * Update instruction length as we may reinject #BP from
4541 		 * user space while in guest debugging mode. Reading it for
4542 		 * #DB as well causes no harm, it is not used in that case.
4543 		 */
4544 		vmx->vcpu.arch.event_exit_inst_len =
4545 			vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4546 		kvm_run->exit_reason = KVM_EXIT_DEBUG;
4547 		rip = kvm_rip_read(vcpu);
4548 		kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
4549 		kvm_run->debug.arch.exception = ex_no;
4550 		break;
4551 	default:
4552 		kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
4553 		kvm_run->ex.exception = ex_no;
4554 		kvm_run->ex.error_code = error_code;
4555 		break;
4556 	}
4557 	return 0;
4558 }
4559 
4560 static int handle_external_interrupt(struct kvm_vcpu *vcpu)
4561 {
4562 	++vcpu->stat.irq_exits;
4563 	return 1;
4564 }
4565 
4566 static int handle_triple_fault(struct kvm_vcpu *vcpu)
4567 {
4568 	vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
4569 	vcpu->mmio_needed = 0;
4570 	return 0;
4571 }
4572 
4573 static int handle_io(struct kvm_vcpu *vcpu)
4574 {
4575 	unsigned long exit_qualification;
4576 	int size, in, string;
4577 	unsigned port;
4578 
4579 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4580 	string = (exit_qualification & 16) != 0;
4581 
4582 	++vcpu->stat.io_exits;
4583 
4584 	if (string)
4585 		return kvm_emulate_instruction(vcpu, 0) == EMULATE_DONE;
4586 
4587 	port = exit_qualification >> 16;
4588 	size = (exit_qualification & 7) + 1;
4589 	in = (exit_qualification & 8) != 0;
4590 
4591 	return kvm_fast_pio(vcpu, size, port, in);
4592 }
4593 
4594 static void
4595 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
4596 {
4597 	/*
4598 	 * Patch in the VMCALL instruction:
4599 	 */
4600 	hypercall[0] = 0x0f;
4601 	hypercall[1] = 0x01;
4602 	hypercall[2] = 0xc1;
4603 }
4604 
4605 /* called to set cr0 as appropriate for a mov-to-cr0 exit. */
4606 static int handle_set_cr0(struct kvm_vcpu *vcpu, unsigned long val)
4607 {
4608 	if (is_guest_mode(vcpu)) {
4609 		struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4610 		unsigned long orig_val = val;
4611 
4612 		/*
4613 		 * We get here when L2 changed cr0 in a way that did not change
4614 		 * any of L1's shadowed bits (see nested_vmx_exit_handled_cr),
4615 		 * but did change L0 shadowed bits. So we first calculate the
4616 		 * effective cr0 value that L1 would like to write into the
4617 		 * hardware. It consists of the L2-owned bits from the new
4618 		 * value combined with the L1-owned bits from L1's guest_cr0.
4619 		 */
4620 		val = (val & ~vmcs12->cr0_guest_host_mask) |
4621 			(vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask);
4622 
4623 		if (!nested_guest_cr0_valid(vcpu, val))
4624 			return 1;
4625 
4626 		if (kvm_set_cr0(vcpu, val))
4627 			return 1;
4628 		vmcs_writel(CR0_READ_SHADOW, orig_val);
4629 		return 0;
4630 	} else {
4631 		if (to_vmx(vcpu)->nested.vmxon &&
4632 		    !nested_host_cr0_valid(vcpu, val))
4633 			return 1;
4634 
4635 		return kvm_set_cr0(vcpu, val);
4636 	}
4637 }
4638 
4639 static int handle_set_cr4(struct kvm_vcpu *vcpu, unsigned long val)
4640 {
4641 	if (is_guest_mode(vcpu)) {
4642 		struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4643 		unsigned long orig_val = val;
4644 
4645 		/* analogously to handle_set_cr0 */
4646 		val = (val & ~vmcs12->cr4_guest_host_mask) |
4647 			(vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask);
4648 		if (kvm_set_cr4(vcpu, val))
4649 			return 1;
4650 		vmcs_writel(CR4_READ_SHADOW, orig_val);
4651 		return 0;
4652 	} else
4653 		return kvm_set_cr4(vcpu, val);
4654 }
4655 
4656 static int handle_desc(struct kvm_vcpu *vcpu)
4657 {
4658 	WARN_ON(!(vcpu->arch.cr4 & X86_CR4_UMIP));
4659 	return kvm_emulate_instruction(vcpu, 0) == EMULATE_DONE;
4660 }
4661 
4662 static int handle_cr(struct kvm_vcpu *vcpu)
4663 {
4664 	unsigned long exit_qualification, val;
4665 	int cr;
4666 	int reg;
4667 	int err;
4668 	int ret;
4669 
4670 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4671 	cr = exit_qualification & 15;
4672 	reg = (exit_qualification >> 8) & 15;
4673 	switch ((exit_qualification >> 4) & 3) {
4674 	case 0: /* mov to cr */
4675 		val = kvm_register_readl(vcpu, reg);
4676 		trace_kvm_cr_write(cr, val);
4677 		switch (cr) {
4678 		case 0:
4679 			err = handle_set_cr0(vcpu, val);
4680 			return kvm_complete_insn_gp(vcpu, err);
4681 		case 3:
4682 			WARN_ON_ONCE(enable_unrestricted_guest);
4683 			err = kvm_set_cr3(vcpu, val);
4684 			return kvm_complete_insn_gp(vcpu, err);
4685 		case 4:
4686 			err = handle_set_cr4(vcpu, val);
4687 			return kvm_complete_insn_gp(vcpu, err);
4688 		case 8: {
4689 				u8 cr8_prev = kvm_get_cr8(vcpu);
4690 				u8 cr8 = (u8)val;
4691 				err = kvm_set_cr8(vcpu, cr8);
4692 				ret = kvm_complete_insn_gp(vcpu, err);
4693 				if (lapic_in_kernel(vcpu))
4694 					return ret;
4695 				if (cr8_prev <= cr8)
4696 					return ret;
4697 				/*
4698 				 * TODO: we might be squashing a
4699 				 * KVM_GUESTDBG_SINGLESTEP-triggered
4700 				 * KVM_EXIT_DEBUG here.
4701 				 */
4702 				vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
4703 				return 0;
4704 			}
4705 		}
4706 		break;
4707 	case 2: /* clts */
4708 		WARN_ONCE(1, "Guest should always own CR0.TS");
4709 		vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
4710 		trace_kvm_cr_write(0, kvm_read_cr0(vcpu));
4711 		return kvm_skip_emulated_instruction(vcpu);
4712 	case 1: /*mov from cr*/
4713 		switch (cr) {
4714 		case 3:
4715 			WARN_ON_ONCE(enable_unrestricted_guest);
4716 			val = kvm_read_cr3(vcpu);
4717 			kvm_register_write(vcpu, reg, val);
4718 			trace_kvm_cr_read(cr, val);
4719 			return kvm_skip_emulated_instruction(vcpu);
4720 		case 8:
4721 			val = kvm_get_cr8(vcpu);
4722 			kvm_register_write(vcpu, reg, val);
4723 			trace_kvm_cr_read(cr, val);
4724 			return kvm_skip_emulated_instruction(vcpu);
4725 		}
4726 		break;
4727 	case 3: /* lmsw */
4728 		val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
4729 		trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val);
4730 		kvm_lmsw(vcpu, val);
4731 
4732 		return kvm_skip_emulated_instruction(vcpu);
4733 	default:
4734 		break;
4735 	}
4736 	vcpu->run->exit_reason = 0;
4737 	vcpu_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
4738 	       (int)(exit_qualification >> 4) & 3, cr);
4739 	return 0;
4740 }
4741 
4742 static int handle_dr(struct kvm_vcpu *vcpu)
4743 {
4744 	unsigned long exit_qualification;
4745 	int dr, dr7, reg;
4746 
4747 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4748 	dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
4749 
4750 	/* First, if DR does not exist, trigger UD */
4751 	if (!kvm_require_dr(vcpu, dr))
4752 		return 1;
4753 
4754 	/* Do not handle if the CPL > 0, will trigger GP on re-entry */
4755 	if (!kvm_require_cpl(vcpu, 0))
4756 		return 1;
4757 	dr7 = vmcs_readl(GUEST_DR7);
4758 	if (dr7 & DR7_GD) {
4759 		/*
4760 		 * As the vm-exit takes precedence over the debug trap, we
4761 		 * need to emulate the latter, either for the host or the
4762 		 * guest debugging itself.
4763 		 */
4764 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
4765 			vcpu->run->debug.arch.dr6 = vcpu->arch.dr6;
4766 			vcpu->run->debug.arch.dr7 = dr7;
4767 			vcpu->run->debug.arch.pc = kvm_get_linear_rip(vcpu);
4768 			vcpu->run->debug.arch.exception = DB_VECTOR;
4769 			vcpu->run->exit_reason = KVM_EXIT_DEBUG;
4770 			return 0;
4771 		} else {
4772 			vcpu->arch.dr6 &= ~15;
4773 			vcpu->arch.dr6 |= DR6_BD | DR6_RTM;
4774 			kvm_queue_exception(vcpu, DB_VECTOR);
4775 			return 1;
4776 		}
4777 	}
4778 
4779 	if (vcpu->guest_debug == 0) {
4780 		vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL,
4781 				CPU_BASED_MOV_DR_EXITING);
4782 
4783 		/*
4784 		 * No more DR vmexits; force a reload of the debug registers
4785 		 * and reenter on this instruction.  The next vmexit will
4786 		 * retrieve the full state of the debug registers.
4787 		 */
4788 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
4789 		return 1;
4790 	}
4791 
4792 	reg = DEBUG_REG_ACCESS_REG(exit_qualification);
4793 	if (exit_qualification & TYPE_MOV_FROM_DR) {
4794 		unsigned long val;
4795 
4796 		if (kvm_get_dr(vcpu, dr, &val))
4797 			return 1;
4798 		kvm_register_write(vcpu, reg, val);
4799 	} else
4800 		if (kvm_set_dr(vcpu, dr, kvm_register_readl(vcpu, reg)))
4801 			return 1;
4802 
4803 	return kvm_skip_emulated_instruction(vcpu);
4804 }
4805 
4806 static u64 vmx_get_dr6(struct kvm_vcpu *vcpu)
4807 {
4808 	return vcpu->arch.dr6;
4809 }
4810 
4811 static void vmx_set_dr6(struct kvm_vcpu *vcpu, unsigned long val)
4812 {
4813 }
4814 
4815 static void vmx_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
4816 {
4817 	get_debugreg(vcpu->arch.db[0], 0);
4818 	get_debugreg(vcpu->arch.db[1], 1);
4819 	get_debugreg(vcpu->arch.db[2], 2);
4820 	get_debugreg(vcpu->arch.db[3], 3);
4821 	get_debugreg(vcpu->arch.dr6, 6);
4822 	vcpu->arch.dr7 = vmcs_readl(GUEST_DR7);
4823 
4824 	vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
4825 	vmcs_set_bits(CPU_BASED_VM_EXEC_CONTROL, CPU_BASED_MOV_DR_EXITING);
4826 }
4827 
4828 static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
4829 {
4830 	vmcs_writel(GUEST_DR7, val);
4831 }
4832 
4833 static int handle_cpuid(struct kvm_vcpu *vcpu)
4834 {
4835 	return kvm_emulate_cpuid(vcpu);
4836 }
4837 
4838 static int handle_rdmsr(struct kvm_vcpu *vcpu)
4839 {
4840 	u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
4841 	struct msr_data msr_info;
4842 
4843 	msr_info.index = ecx;
4844 	msr_info.host_initiated = false;
4845 	if (vmx_get_msr(vcpu, &msr_info)) {
4846 		trace_kvm_msr_read_ex(ecx);
4847 		kvm_inject_gp(vcpu, 0);
4848 		return 1;
4849 	}
4850 
4851 	trace_kvm_msr_read(ecx, msr_info.data);
4852 
4853 	/* FIXME: handling of bits 32:63 of rax, rdx */
4854 	vcpu->arch.regs[VCPU_REGS_RAX] = msr_info.data & -1u;
4855 	vcpu->arch.regs[VCPU_REGS_RDX] = (msr_info.data >> 32) & -1u;
4856 	return kvm_skip_emulated_instruction(vcpu);
4857 }
4858 
4859 static int handle_wrmsr(struct kvm_vcpu *vcpu)
4860 {
4861 	struct msr_data msr;
4862 	u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
4863 	u64 data = (vcpu->arch.regs[VCPU_REGS_RAX] & -1u)
4864 		| ((u64)(vcpu->arch.regs[VCPU_REGS_RDX] & -1u) << 32);
4865 
4866 	msr.data = data;
4867 	msr.index = ecx;
4868 	msr.host_initiated = false;
4869 	if (kvm_set_msr(vcpu, &msr) != 0) {
4870 		trace_kvm_msr_write_ex(ecx, data);
4871 		kvm_inject_gp(vcpu, 0);
4872 		return 1;
4873 	}
4874 
4875 	trace_kvm_msr_write(ecx, data);
4876 	return kvm_skip_emulated_instruction(vcpu);
4877 }
4878 
4879 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
4880 {
4881 	kvm_apic_update_ppr(vcpu);
4882 	return 1;
4883 }
4884 
4885 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
4886 {
4887 	vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL,
4888 			CPU_BASED_VIRTUAL_INTR_PENDING);
4889 
4890 	kvm_make_request(KVM_REQ_EVENT, vcpu);
4891 
4892 	++vcpu->stat.irq_window_exits;
4893 	return 1;
4894 }
4895 
4896 static int handle_halt(struct kvm_vcpu *vcpu)
4897 {
4898 	return kvm_emulate_halt(vcpu);
4899 }
4900 
4901 static int handle_vmcall(struct kvm_vcpu *vcpu)
4902 {
4903 	return kvm_emulate_hypercall(vcpu);
4904 }
4905 
4906 static int handle_invd(struct kvm_vcpu *vcpu)
4907 {
4908 	return kvm_emulate_instruction(vcpu, 0) == EMULATE_DONE;
4909 }
4910 
4911 static int handle_invlpg(struct kvm_vcpu *vcpu)
4912 {
4913 	unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4914 
4915 	kvm_mmu_invlpg(vcpu, exit_qualification);
4916 	return kvm_skip_emulated_instruction(vcpu);
4917 }
4918 
4919 static int handle_rdpmc(struct kvm_vcpu *vcpu)
4920 {
4921 	int err;
4922 
4923 	err = kvm_rdpmc(vcpu);
4924 	return kvm_complete_insn_gp(vcpu, err);
4925 }
4926 
4927 static int handle_wbinvd(struct kvm_vcpu *vcpu)
4928 {
4929 	return kvm_emulate_wbinvd(vcpu);
4930 }
4931 
4932 static int handle_xsetbv(struct kvm_vcpu *vcpu)
4933 {
4934 	u64 new_bv = kvm_read_edx_eax(vcpu);
4935 	u32 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
4936 
4937 	if (kvm_set_xcr(vcpu, index, new_bv) == 0)
4938 		return kvm_skip_emulated_instruction(vcpu);
4939 	return 1;
4940 }
4941 
4942 static int handle_xsaves(struct kvm_vcpu *vcpu)
4943 {
4944 	kvm_skip_emulated_instruction(vcpu);
4945 	WARN(1, "this should never happen\n");
4946 	return 1;
4947 }
4948 
4949 static int handle_xrstors(struct kvm_vcpu *vcpu)
4950 {
4951 	kvm_skip_emulated_instruction(vcpu);
4952 	WARN(1, "this should never happen\n");
4953 	return 1;
4954 }
4955 
4956 static int handle_apic_access(struct kvm_vcpu *vcpu)
4957 {
4958 	if (likely(fasteoi)) {
4959 		unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4960 		int access_type, offset;
4961 
4962 		access_type = exit_qualification & APIC_ACCESS_TYPE;
4963 		offset = exit_qualification & APIC_ACCESS_OFFSET;
4964 		/*
4965 		 * Sane guest uses MOV to write EOI, with written value
4966 		 * not cared. So make a short-circuit here by avoiding
4967 		 * heavy instruction emulation.
4968 		 */
4969 		if ((access_type == TYPE_LINEAR_APIC_INST_WRITE) &&
4970 		    (offset == APIC_EOI)) {
4971 			kvm_lapic_set_eoi(vcpu);
4972 			return kvm_skip_emulated_instruction(vcpu);
4973 		}
4974 	}
4975 	return kvm_emulate_instruction(vcpu, 0) == EMULATE_DONE;
4976 }
4977 
4978 static int handle_apic_eoi_induced(struct kvm_vcpu *vcpu)
4979 {
4980 	unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4981 	int vector = exit_qualification & 0xff;
4982 
4983 	/* EOI-induced VM exit is trap-like and thus no need to adjust IP */
4984 	kvm_apic_set_eoi_accelerated(vcpu, vector);
4985 	return 1;
4986 }
4987 
4988 static int handle_apic_write(struct kvm_vcpu *vcpu)
4989 {
4990 	unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4991 	u32 offset = exit_qualification & 0xfff;
4992 
4993 	/* APIC-write VM exit is trap-like and thus no need to adjust IP */
4994 	kvm_apic_write_nodecode(vcpu, offset);
4995 	return 1;
4996 }
4997 
4998 static int handle_task_switch(struct kvm_vcpu *vcpu)
4999 {
5000 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5001 	unsigned long exit_qualification;
5002 	bool has_error_code = false;
5003 	u32 error_code = 0;
5004 	u16 tss_selector;
5005 	int reason, type, idt_v, idt_index;
5006 
5007 	idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
5008 	idt_index = (vmx->idt_vectoring_info & VECTORING_INFO_VECTOR_MASK);
5009 	type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
5010 
5011 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5012 
5013 	reason = (u32)exit_qualification >> 30;
5014 	if (reason == TASK_SWITCH_GATE && idt_v) {
5015 		switch (type) {
5016 		case INTR_TYPE_NMI_INTR:
5017 			vcpu->arch.nmi_injected = false;
5018 			vmx_set_nmi_mask(vcpu, true);
5019 			break;
5020 		case INTR_TYPE_EXT_INTR:
5021 		case INTR_TYPE_SOFT_INTR:
5022 			kvm_clear_interrupt_queue(vcpu);
5023 			break;
5024 		case INTR_TYPE_HARD_EXCEPTION:
5025 			if (vmx->idt_vectoring_info &
5026 			    VECTORING_INFO_DELIVER_CODE_MASK) {
5027 				has_error_code = true;
5028 				error_code =
5029 					vmcs_read32(IDT_VECTORING_ERROR_CODE);
5030 			}
5031 			/* fall through */
5032 		case INTR_TYPE_SOFT_EXCEPTION:
5033 			kvm_clear_exception_queue(vcpu);
5034 			break;
5035 		default:
5036 			break;
5037 		}
5038 	}
5039 	tss_selector = exit_qualification;
5040 
5041 	if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
5042 		       type != INTR_TYPE_EXT_INTR &&
5043 		       type != INTR_TYPE_NMI_INTR))
5044 		skip_emulated_instruction(vcpu);
5045 
5046 	if (kvm_task_switch(vcpu, tss_selector,
5047 			    type == INTR_TYPE_SOFT_INTR ? idt_index : -1, reason,
5048 			    has_error_code, error_code) == EMULATE_FAIL) {
5049 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5050 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5051 		vcpu->run->internal.ndata = 0;
5052 		return 0;
5053 	}
5054 
5055 	/*
5056 	 * TODO: What about debug traps on tss switch?
5057 	 *       Are we supposed to inject them and update dr6?
5058 	 */
5059 
5060 	return 1;
5061 }
5062 
5063 static int handle_ept_violation(struct kvm_vcpu *vcpu)
5064 {
5065 	unsigned long exit_qualification;
5066 	gpa_t gpa;
5067 	u64 error_code;
5068 
5069 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5070 
5071 	/*
5072 	 * EPT violation happened while executing iret from NMI,
5073 	 * "blocked by NMI" bit has to be set before next VM entry.
5074 	 * There are errata that may cause this bit to not be set:
5075 	 * AAK134, BY25.
5076 	 */
5077 	if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
5078 			enable_vnmi &&
5079 			(exit_qualification & INTR_INFO_UNBLOCK_NMI))
5080 		vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO, GUEST_INTR_STATE_NMI);
5081 
5082 	gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5083 	trace_kvm_page_fault(gpa, exit_qualification);
5084 
5085 	/* Is it a read fault? */
5086 	error_code = (exit_qualification & EPT_VIOLATION_ACC_READ)
5087 		     ? PFERR_USER_MASK : 0;
5088 	/* Is it a write fault? */
5089 	error_code |= (exit_qualification & EPT_VIOLATION_ACC_WRITE)
5090 		      ? PFERR_WRITE_MASK : 0;
5091 	/* Is it a fetch fault? */
5092 	error_code |= (exit_qualification & EPT_VIOLATION_ACC_INSTR)
5093 		      ? PFERR_FETCH_MASK : 0;
5094 	/* ept page table entry is present? */
5095 	error_code |= (exit_qualification &
5096 		       (EPT_VIOLATION_READABLE | EPT_VIOLATION_WRITABLE |
5097 			EPT_VIOLATION_EXECUTABLE))
5098 		      ? PFERR_PRESENT_MASK : 0;
5099 
5100 	error_code |= (exit_qualification & 0x100) != 0 ?
5101 	       PFERR_GUEST_FINAL_MASK : PFERR_GUEST_PAGE_MASK;
5102 
5103 	vcpu->arch.exit_qualification = exit_qualification;
5104 	return kvm_mmu_page_fault(vcpu, gpa, error_code, NULL, 0);
5105 }
5106 
5107 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
5108 {
5109 	gpa_t gpa;
5110 
5111 	/*
5112 	 * A nested guest cannot optimize MMIO vmexits, because we have an
5113 	 * nGPA here instead of the required GPA.
5114 	 */
5115 	gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5116 	if (!is_guest_mode(vcpu) &&
5117 	    !kvm_io_bus_write(vcpu, KVM_FAST_MMIO_BUS, gpa, 0, NULL)) {
5118 		trace_kvm_fast_mmio(gpa);
5119 		/*
5120 		 * Doing kvm_skip_emulated_instruction() depends on undefined
5121 		 * behavior: Intel's manual doesn't mandate
5122 		 * VM_EXIT_INSTRUCTION_LEN to be set in VMCS when EPT MISCONFIG
5123 		 * occurs and while on real hardware it was observed to be set,
5124 		 * other hypervisors (namely Hyper-V) don't set it, we end up
5125 		 * advancing IP with some random value. Disable fast mmio when
5126 		 * running nested and keep it for real hardware in hope that
5127 		 * VM_EXIT_INSTRUCTION_LEN will always be set correctly.
5128 		 */
5129 		if (!static_cpu_has(X86_FEATURE_HYPERVISOR))
5130 			return kvm_skip_emulated_instruction(vcpu);
5131 		else
5132 			return kvm_emulate_instruction(vcpu, EMULTYPE_SKIP) ==
5133 								EMULATE_DONE;
5134 	}
5135 
5136 	return kvm_mmu_page_fault(vcpu, gpa, PFERR_RSVD_MASK, NULL, 0);
5137 }
5138 
5139 static int handle_nmi_window(struct kvm_vcpu *vcpu)
5140 {
5141 	WARN_ON_ONCE(!enable_vnmi);
5142 	vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL,
5143 			CPU_BASED_VIRTUAL_NMI_PENDING);
5144 	++vcpu->stat.nmi_window_exits;
5145 	kvm_make_request(KVM_REQ_EVENT, vcpu);
5146 
5147 	return 1;
5148 }
5149 
5150 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
5151 {
5152 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5153 	enum emulation_result err = EMULATE_DONE;
5154 	int ret = 1;
5155 	u32 cpu_exec_ctrl;
5156 	bool intr_window_requested;
5157 	unsigned count = 130;
5158 
5159 	/*
5160 	 * We should never reach the point where we are emulating L2
5161 	 * due to invalid guest state as that means we incorrectly
5162 	 * allowed a nested VMEntry with an invalid vmcs12.
5163 	 */
5164 	WARN_ON_ONCE(vmx->emulation_required && vmx->nested.nested_run_pending);
5165 
5166 	cpu_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5167 	intr_window_requested = cpu_exec_ctrl & CPU_BASED_VIRTUAL_INTR_PENDING;
5168 
5169 	while (vmx->emulation_required && count-- != 0) {
5170 		if (intr_window_requested && vmx_interrupt_allowed(vcpu))
5171 			return handle_interrupt_window(&vmx->vcpu);
5172 
5173 		if (kvm_test_request(KVM_REQ_EVENT, vcpu))
5174 			return 1;
5175 
5176 		err = kvm_emulate_instruction(vcpu, 0);
5177 
5178 		if (err == EMULATE_USER_EXIT) {
5179 			++vcpu->stat.mmio_exits;
5180 			ret = 0;
5181 			goto out;
5182 		}
5183 
5184 		if (err != EMULATE_DONE)
5185 			goto emulation_error;
5186 
5187 		if (vmx->emulation_required && !vmx->rmode.vm86_active &&
5188 		    vcpu->arch.exception.pending)
5189 			goto emulation_error;
5190 
5191 		if (vcpu->arch.halt_request) {
5192 			vcpu->arch.halt_request = 0;
5193 			ret = kvm_vcpu_halt(vcpu);
5194 			goto out;
5195 		}
5196 
5197 		if (signal_pending(current))
5198 			goto out;
5199 		if (need_resched())
5200 			schedule();
5201 	}
5202 
5203 out:
5204 	return ret;
5205 
5206 emulation_error:
5207 	vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5208 	vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5209 	vcpu->run->internal.ndata = 0;
5210 	return 0;
5211 }
5212 
5213 static void grow_ple_window(struct kvm_vcpu *vcpu)
5214 {
5215 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5216 	int old = vmx->ple_window;
5217 
5218 	vmx->ple_window = __grow_ple_window(old, ple_window,
5219 					    ple_window_grow,
5220 					    ple_window_max);
5221 
5222 	if (vmx->ple_window != old)
5223 		vmx->ple_window_dirty = true;
5224 
5225 	trace_kvm_ple_window_grow(vcpu->vcpu_id, vmx->ple_window, old);
5226 }
5227 
5228 static void shrink_ple_window(struct kvm_vcpu *vcpu)
5229 {
5230 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5231 	int old = vmx->ple_window;
5232 
5233 	vmx->ple_window = __shrink_ple_window(old, ple_window,
5234 					      ple_window_shrink,
5235 					      ple_window);
5236 
5237 	if (vmx->ple_window != old)
5238 		vmx->ple_window_dirty = true;
5239 
5240 	trace_kvm_ple_window_shrink(vcpu->vcpu_id, vmx->ple_window, old);
5241 }
5242 
5243 /*
5244  * Handler for POSTED_INTERRUPT_WAKEUP_VECTOR.
5245  */
5246 static void wakeup_handler(void)
5247 {
5248 	struct kvm_vcpu *vcpu;
5249 	int cpu = smp_processor_id();
5250 
5251 	spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
5252 	list_for_each_entry(vcpu, &per_cpu(blocked_vcpu_on_cpu, cpu),
5253 			blocked_vcpu_list) {
5254 		struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
5255 
5256 		if (pi_test_on(pi_desc) == 1)
5257 			kvm_vcpu_kick(vcpu);
5258 	}
5259 	spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
5260 }
5261 
5262 static void vmx_enable_tdp(void)
5263 {
5264 	kvm_mmu_set_mask_ptes(VMX_EPT_READABLE_MASK,
5265 		enable_ept_ad_bits ? VMX_EPT_ACCESS_BIT : 0ull,
5266 		enable_ept_ad_bits ? VMX_EPT_DIRTY_BIT : 0ull,
5267 		0ull, VMX_EPT_EXECUTABLE_MASK,
5268 		cpu_has_vmx_ept_execute_only() ? 0ull : VMX_EPT_READABLE_MASK,
5269 		VMX_EPT_RWX_MASK, 0ull);
5270 
5271 	ept_set_mmio_spte_mask();
5272 	kvm_enable_tdp();
5273 }
5274 
5275 /*
5276  * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
5277  * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
5278  */
5279 static int handle_pause(struct kvm_vcpu *vcpu)
5280 {
5281 	if (!kvm_pause_in_guest(vcpu->kvm))
5282 		grow_ple_window(vcpu);
5283 
5284 	/*
5285 	 * Intel sdm vol3 ch-25.1.3 says: The "PAUSE-loop exiting"
5286 	 * VM-execution control is ignored if CPL > 0. OTOH, KVM
5287 	 * never set PAUSE_EXITING and just set PLE if supported,
5288 	 * so the vcpu must be CPL=0 if it gets a PAUSE exit.
5289 	 */
5290 	kvm_vcpu_on_spin(vcpu, true);
5291 	return kvm_skip_emulated_instruction(vcpu);
5292 }
5293 
5294 static int handle_nop(struct kvm_vcpu *vcpu)
5295 {
5296 	return kvm_skip_emulated_instruction(vcpu);
5297 }
5298 
5299 static int handle_mwait(struct kvm_vcpu *vcpu)
5300 {
5301 	printk_once(KERN_WARNING "kvm: MWAIT instruction emulated as NOP!\n");
5302 	return handle_nop(vcpu);
5303 }
5304 
5305 static int handle_invalid_op(struct kvm_vcpu *vcpu)
5306 {
5307 	kvm_queue_exception(vcpu, UD_VECTOR);
5308 	return 1;
5309 }
5310 
5311 static int handle_monitor_trap(struct kvm_vcpu *vcpu)
5312 {
5313 	return 1;
5314 }
5315 
5316 static int handle_monitor(struct kvm_vcpu *vcpu)
5317 {
5318 	printk_once(KERN_WARNING "kvm: MONITOR instruction emulated as NOP!\n");
5319 	return handle_nop(vcpu);
5320 }
5321 
5322 static int handle_invpcid(struct kvm_vcpu *vcpu)
5323 {
5324 	u32 vmx_instruction_info;
5325 	unsigned long type;
5326 	bool pcid_enabled;
5327 	gva_t gva;
5328 	struct x86_exception e;
5329 	unsigned i;
5330 	unsigned long roots_to_free = 0;
5331 	struct {
5332 		u64 pcid;
5333 		u64 gla;
5334 	} operand;
5335 
5336 	if (!guest_cpuid_has(vcpu, X86_FEATURE_INVPCID)) {
5337 		kvm_queue_exception(vcpu, UD_VECTOR);
5338 		return 1;
5339 	}
5340 
5341 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5342 	type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
5343 
5344 	if (type > 3) {
5345 		kvm_inject_gp(vcpu, 0);
5346 		return 1;
5347 	}
5348 
5349 	/* According to the Intel instruction reference, the memory operand
5350 	 * is read even if it isn't needed (e.g., for type==all)
5351 	 */
5352 	if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
5353 				vmx_instruction_info, false, &gva))
5354 		return 1;
5355 
5356 	if (kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e)) {
5357 		kvm_inject_page_fault(vcpu, &e);
5358 		return 1;
5359 	}
5360 
5361 	if (operand.pcid >> 12 != 0) {
5362 		kvm_inject_gp(vcpu, 0);
5363 		return 1;
5364 	}
5365 
5366 	pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
5367 
5368 	switch (type) {
5369 	case INVPCID_TYPE_INDIV_ADDR:
5370 		if ((!pcid_enabled && (operand.pcid != 0)) ||
5371 		    is_noncanonical_address(operand.gla, vcpu)) {
5372 			kvm_inject_gp(vcpu, 0);
5373 			return 1;
5374 		}
5375 		kvm_mmu_invpcid_gva(vcpu, operand.gla, operand.pcid);
5376 		return kvm_skip_emulated_instruction(vcpu);
5377 
5378 	case INVPCID_TYPE_SINGLE_CTXT:
5379 		if (!pcid_enabled && (operand.pcid != 0)) {
5380 			kvm_inject_gp(vcpu, 0);
5381 			return 1;
5382 		}
5383 
5384 		if (kvm_get_active_pcid(vcpu) == operand.pcid) {
5385 			kvm_mmu_sync_roots(vcpu);
5386 			kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
5387 		}
5388 
5389 		for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++)
5390 			if (kvm_get_pcid(vcpu, vcpu->arch.mmu->prev_roots[i].cr3)
5391 			    == operand.pcid)
5392 				roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i);
5393 
5394 		kvm_mmu_free_roots(vcpu, vcpu->arch.mmu, roots_to_free);
5395 		/*
5396 		 * If neither the current cr3 nor any of the prev_roots use the
5397 		 * given PCID, then nothing needs to be done here because a
5398 		 * resync will happen anyway before switching to any other CR3.
5399 		 */
5400 
5401 		return kvm_skip_emulated_instruction(vcpu);
5402 
5403 	case INVPCID_TYPE_ALL_NON_GLOBAL:
5404 		/*
5405 		 * Currently, KVM doesn't mark global entries in the shadow
5406 		 * page tables, so a non-global flush just degenerates to a
5407 		 * global flush. If needed, we could optimize this later by
5408 		 * keeping track of global entries in shadow page tables.
5409 		 */
5410 
5411 		/* fall-through */
5412 	case INVPCID_TYPE_ALL_INCL_GLOBAL:
5413 		kvm_mmu_unload(vcpu);
5414 		return kvm_skip_emulated_instruction(vcpu);
5415 
5416 	default:
5417 		BUG(); /* We have already checked above that type <= 3 */
5418 	}
5419 }
5420 
5421 static int handle_pml_full(struct kvm_vcpu *vcpu)
5422 {
5423 	unsigned long exit_qualification;
5424 
5425 	trace_kvm_pml_full(vcpu->vcpu_id);
5426 
5427 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5428 
5429 	/*
5430 	 * PML buffer FULL happened while executing iret from NMI,
5431 	 * "blocked by NMI" bit has to be set before next VM entry.
5432 	 */
5433 	if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
5434 			enable_vnmi &&
5435 			(exit_qualification & INTR_INFO_UNBLOCK_NMI))
5436 		vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
5437 				GUEST_INTR_STATE_NMI);
5438 
5439 	/*
5440 	 * PML buffer already flushed at beginning of VMEXIT. Nothing to do
5441 	 * here.., and there's no userspace involvement needed for PML.
5442 	 */
5443 	return 1;
5444 }
5445 
5446 static int handle_preemption_timer(struct kvm_vcpu *vcpu)
5447 {
5448 	if (!to_vmx(vcpu)->req_immediate_exit)
5449 		kvm_lapic_expired_hv_timer(vcpu);
5450 	return 1;
5451 }
5452 
5453 /*
5454  * When nested=0, all VMX instruction VM Exits filter here.  The handlers
5455  * are overwritten by nested_vmx_setup() when nested=1.
5456  */
5457 static int handle_vmx_instruction(struct kvm_vcpu *vcpu)
5458 {
5459 	kvm_queue_exception(vcpu, UD_VECTOR);
5460 	return 1;
5461 }
5462 
5463 static int handle_encls(struct kvm_vcpu *vcpu)
5464 {
5465 	/*
5466 	 * SGX virtualization is not yet supported.  There is no software
5467 	 * enable bit for SGX, so we have to trap ENCLS and inject a #UD
5468 	 * to prevent the guest from executing ENCLS.
5469 	 */
5470 	kvm_queue_exception(vcpu, UD_VECTOR);
5471 	return 1;
5472 }
5473 
5474 /*
5475  * The exit handlers return 1 if the exit was handled fully and guest execution
5476  * may resume.  Otherwise they set the kvm_run parameter to indicate what needs
5477  * to be done to userspace and return 0.
5478  */
5479 static int (*kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
5480 	[EXIT_REASON_EXCEPTION_NMI]           = handle_exception,
5481 	[EXIT_REASON_EXTERNAL_INTERRUPT]      = handle_external_interrupt,
5482 	[EXIT_REASON_TRIPLE_FAULT]            = handle_triple_fault,
5483 	[EXIT_REASON_NMI_WINDOW]	      = handle_nmi_window,
5484 	[EXIT_REASON_IO_INSTRUCTION]          = handle_io,
5485 	[EXIT_REASON_CR_ACCESS]               = handle_cr,
5486 	[EXIT_REASON_DR_ACCESS]               = handle_dr,
5487 	[EXIT_REASON_CPUID]                   = handle_cpuid,
5488 	[EXIT_REASON_MSR_READ]                = handle_rdmsr,
5489 	[EXIT_REASON_MSR_WRITE]               = handle_wrmsr,
5490 	[EXIT_REASON_PENDING_INTERRUPT]       = handle_interrupt_window,
5491 	[EXIT_REASON_HLT]                     = handle_halt,
5492 	[EXIT_REASON_INVD]		      = handle_invd,
5493 	[EXIT_REASON_INVLPG]		      = handle_invlpg,
5494 	[EXIT_REASON_RDPMC]                   = handle_rdpmc,
5495 	[EXIT_REASON_VMCALL]                  = handle_vmcall,
5496 	[EXIT_REASON_VMCLEAR]		      = handle_vmx_instruction,
5497 	[EXIT_REASON_VMLAUNCH]		      = handle_vmx_instruction,
5498 	[EXIT_REASON_VMPTRLD]		      = handle_vmx_instruction,
5499 	[EXIT_REASON_VMPTRST]		      = handle_vmx_instruction,
5500 	[EXIT_REASON_VMREAD]		      = handle_vmx_instruction,
5501 	[EXIT_REASON_VMRESUME]		      = handle_vmx_instruction,
5502 	[EXIT_REASON_VMWRITE]		      = handle_vmx_instruction,
5503 	[EXIT_REASON_VMOFF]		      = handle_vmx_instruction,
5504 	[EXIT_REASON_VMON]		      = handle_vmx_instruction,
5505 	[EXIT_REASON_TPR_BELOW_THRESHOLD]     = handle_tpr_below_threshold,
5506 	[EXIT_REASON_APIC_ACCESS]             = handle_apic_access,
5507 	[EXIT_REASON_APIC_WRITE]              = handle_apic_write,
5508 	[EXIT_REASON_EOI_INDUCED]             = handle_apic_eoi_induced,
5509 	[EXIT_REASON_WBINVD]                  = handle_wbinvd,
5510 	[EXIT_REASON_XSETBV]                  = handle_xsetbv,
5511 	[EXIT_REASON_TASK_SWITCH]             = handle_task_switch,
5512 	[EXIT_REASON_MCE_DURING_VMENTRY]      = handle_machine_check,
5513 	[EXIT_REASON_GDTR_IDTR]		      = handle_desc,
5514 	[EXIT_REASON_LDTR_TR]		      = handle_desc,
5515 	[EXIT_REASON_EPT_VIOLATION]	      = handle_ept_violation,
5516 	[EXIT_REASON_EPT_MISCONFIG]           = handle_ept_misconfig,
5517 	[EXIT_REASON_PAUSE_INSTRUCTION]       = handle_pause,
5518 	[EXIT_REASON_MWAIT_INSTRUCTION]	      = handle_mwait,
5519 	[EXIT_REASON_MONITOR_TRAP_FLAG]       = handle_monitor_trap,
5520 	[EXIT_REASON_MONITOR_INSTRUCTION]     = handle_monitor,
5521 	[EXIT_REASON_INVEPT]                  = handle_vmx_instruction,
5522 	[EXIT_REASON_INVVPID]                 = handle_vmx_instruction,
5523 	[EXIT_REASON_RDRAND]                  = handle_invalid_op,
5524 	[EXIT_REASON_RDSEED]                  = handle_invalid_op,
5525 	[EXIT_REASON_XSAVES]                  = handle_xsaves,
5526 	[EXIT_REASON_XRSTORS]                 = handle_xrstors,
5527 	[EXIT_REASON_PML_FULL]		      = handle_pml_full,
5528 	[EXIT_REASON_INVPCID]                 = handle_invpcid,
5529 	[EXIT_REASON_VMFUNC]		      = handle_vmx_instruction,
5530 	[EXIT_REASON_PREEMPTION_TIMER]	      = handle_preemption_timer,
5531 	[EXIT_REASON_ENCLS]		      = handle_encls,
5532 };
5533 
5534 static const int kvm_vmx_max_exit_handlers =
5535 	ARRAY_SIZE(kvm_vmx_exit_handlers);
5536 
5537 static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
5538 {
5539 	*info1 = vmcs_readl(EXIT_QUALIFICATION);
5540 	*info2 = vmcs_read32(VM_EXIT_INTR_INFO);
5541 }
5542 
5543 static void vmx_destroy_pml_buffer(struct vcpu_vmx *vmx)
5544 {
5545 	if (vmx->pml_pg) {
5546 		__free_page(vmx->pml_pg);
5547 		vmx->pml_pg = NULL;
5548 	}
5549 }
5550 
5551 static void vmx_flush_pml_buffer(struct kvm_vcpu *vcpu)
5552 {
5553 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5554 	u64 *pml_buf;
5555 	u16 pml_idx;
5556 
5557 	pml_idx = vmcs_read16(GUEST_PML_INDEX);
5558 
5559 	/* Do nothing if PML buffer is empty */
5560 	if (pml_idx == (PML_ENTITY_NUM - 1))
5561 		return;
5562 
5563 	/* PML index always points to next available PML buffer entity */
5564 	if (pml_idx >= PML_ENTITY_NUM)
5565 		pml_idx = 0;
5566 	else
5567 		pml_idx++;
5568 
5569 	pml_buf = page_address(vmx->pml_pg);
5570 	for (; pml_idx < PML_ENTITY_NUM; pml_idx++) {
5571 		u64 gpa;
5572 
5573 		gpa = pml_buf[pml_idx];
5574 		WARN_ON(gpa & (PAGE_SIZE - 1));
5575 		kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
5576 	}
5577 
5578 	/* reset PML index */
5579 	vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
5580 }
5581 
5582 /*
5583  * Flush all vcpus' PML buffer and update logged GPAs to dirty_bitmap.
5584  * Called before reporting dirty_bitmap to userspace.
5585  */
5586 static void kvm_flush_pml_buffers(struct kvm *kvm)
5587 {
5588 	int i;
5589 	struct kvm_vcpu *vcpu;
5590 	/*
5591 	 * We only need to kick vcpu out of guest mode here, as PML buffer
5592 	 * is flushed at beginning of all VMEXITs, and it's obvious that only
5593 	 * vcpus running in guest are possible to have unflushed GPAs in PML
5594 	 * buffer.
5595 	 */
5596 	kvm_for_each_vcpu(i, vcpu, kvm)
5597 		kvm_vcpu_kick(vcpu);
5598 }
5599 
5600 static void vmx_dump_sel(char *name, uint32_t sel)
5601 {
5602 	pr_err("%s sel=0x%04x, attr=0x%05x, limit=0x%08x, base=0x%016lx\n",
5603 	       name, vmcs_read16(sel),
5604 	       vmcs_read32(sel + GUEST_ES_AR_BYTES - GUEST_ES_SELECTOR),
5605 	       vmcs_read32(sel + GUEST_ES_LIMIT - GUEST_ES_SELECTOR),
5606 	       vmcs_readl(sel + GUEST_ES_BASE - GUEST_ES_SELECTOR));
5607 }
5608 
5609 static void vmx_dump_dtsel(char *name, uint32_t limit)
5610 {
5611 	pr_err("%s                           limit=0x%08x, base=0x%016lx\n",
5612 	       name, vmcs_read32(limit),
5613 	       vmcs_readl(limit + GUEST_GDTR_BASE - GUEST_GDTR_LIMIT));
5614 }
5615 
5616 static void dump_vmcs(void)
5617 {
5618 	u32 vmentry_ctl = vmcs_read32(VM_ENTRY_CONTROLS);
5619 	u32 vmexit_ctl = vmcs_read32(VM_EXIT_CONTROLS);
5620 	u32 cpu_based_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5621 	u32 pin_based_exec_ctrl = vmcs_read32(PIN_BASED_VM_EXEC_CONTROL);
5622 	u32 secondary_exec_control = 0;
5623 	unsigned long cr4 = vmcs_readl(GUEST_CR4);
5624 	u64 efer = vmcs_read64(GUEST_IA32_EFER);
5625 	int i, n;
5626 
5627 	if (cpu_has_secondary_exec_ctrls())
5628 		secondary_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
5629 
5630 	pr_err("*** Guest State ***\n");
5631 	pr_err("CR0: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
5632 	       vmcs_readl(GUEST_CR0), vmcs_readl(CR0_READ_SHADOW),
5633 	       vmcs_readl(CR0_GUEST_HOST_MASK));
5634 	pr_err("CR4: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
5635 	       cr4, vmcs_readl(CR4_READ_SHADOW), vmcs_readl(CR4_GUEST_HOST_MASK));
5636 	pr_err("CR3 = 0x%016lx\n", vmcs_readl(GUEST_CR3));
5637 	if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT) &&
5638 	    (cr4 & X86_CR4_PAE) && !(efer & EFER_LMA))
5639 	{
5640 		pr_err("PDPTR0 = 0x%016llx  PDPTR1 = 0x%016llx\n",
5641 		       vmcs_read64(GUEST_PDPTR0), vmcs_read64(GUEST_PDPTR1));
5642 		pr_err("PDPTR2 = 0x%016llx  PDPTR3 = 0x%016llx\n",
5643 		       vmcs_read64(GUEST_PDPTR2), vmcs_read64(GUEST_PDPTR3));
5644 	}
5645 	pr_err("RSP = 0x%016lx  RIP = 0x%016lx\n",
5646 	       vmcs_readl(GUEST_RSP), vmcs_readl(GUEST_RIP));
5647 	pr_err("RFLAGS=0x%08lx         DR7 = 0x%016lx\n",
5648 	       vmcs_readl(GUEST_RFLAGS), vmcs_readl(GUEST_DR7));
5649 	pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
5650 	       vmcs_readl(GUEST_SYSENTER_ESP),
5651 	       vmcs_read32(GUEST_SYSENTER_CS), vmcs_readl(GUEST_SYSENTER_EIP));
5652 	vmx_dump_sel("CS:  ", GUEST_CS_SELECTOR);
5653 	vmx_dump_sel("DS:  ", GUEST_DS_SELECTOR);
5654 	vmx_dump_sel("SS:  ", GUEST_SS_SELECTOR);
5655 	vmx_dump_sel("ES:  ", GUEST_ES_SELECTOR);
5656 	vmx_dump_sel("FS:  ", GUEST_FS_SELECTOR);
5657 	vmx_dump_sel("GS:  ", GUEST_GS_SELECTOR);
5658 	vmx_dump_dtsel("GDTR:", GUEST_GDTR_LIMIT);
5659 	vmx_dump_sel("LDTR:", GUEST_LDTR_SELECTOR);
5660 	vmx_dump_dtsel("IDTR:", GUEST_IDTR_LIMIT);
5661 	vmx_dump_sel("TR:  ", GUEST_TR_SELECTOR);
5662 	if ((vmexit_ctl & (VM_EXIT_SAVE_IA32_PAT | VM_EXIT_SAVE_IA32_EFER)) ||
5663 	    (vmentry_ctl & (VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_IA32_EFER)))
5664 		pr_err("EFER =     0x%016llx  PAT = 0x%016llx\n",
5665 		       efer, vmcs_read64(GUEST_IA32_PAT));
5666 	pr_err("DebugCtl = 0x%016llx  DebugExceptions = 0x%016lx\n",
5667 	       vmcs_read64(GUEST_IA32_DEBUGCTL),
5668 	       vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS));
5669 	if (cpu_has_load_perf_global_ctrl() &&
5670 	    vmentry_ctl & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
5671 		pr_err("PerfGlobCtl = 0x%016llx\n",
5672 		       vmcs_read64(GUEST_IA32_PERF_GLOBAL_CTRL));
5673 	if (vmentry_ctl & VM_ENTRY_LOAD_BNDCFGS)
5674 		pr_err("BndCfgS = 0x%016llx\n", vmcs_read64(GUEST_BNDCFGS));
5675 	pr_err("Interruptibility = %08x  ActivityState = %08x\n",
5676 	       vmcs_read32(GUEST_INTERRUPTIBILITY_INFO),
5677 	       vmcs_read32(GUEST_ACTIVITY_STATE));
5678 	if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)
5679 		pr_err("InterruptStatus = %04x\n",
5680 		       vmcs_read16(GUEST_INTR_STATUS));
5681 
5682 	pr_err("*** Host State ***\n");
5683 	pr_err("RIP = 0x%016lx  RSP = 0x%016lx\n",
5684 	       vmcs_readl(HOST_RIP), vmcs_readl(HOST_RSP));
5685 	pr_err("CS=%04x SS=%04x DS=%04x ES=%04x FS=%04x GS=%04x TR=%04x\n",
5686 	       vmcs_read16(HOST_CS_SELECTOR), vmcs_read16(HOST_SS_SELECTOR),
5687 	       vmcs_read16(HOST_DS_SELECTOR), vmcs_read16(HOST_ES_SELECTOR),
5688 	       vmcs_read16(HOST_FS_SELECTOR), vmcs_read16(HOST_GS_SELECTOR),
5689 	       vmcs_read16(HOST_TR_SELECTOR));
5690 	pr_err("FSBase=%016lx GSBase=%016lx TRBase=%016lx\n",
5691 	       vmcs_readl(HOST_FS_BASE), vmcs_readl(HOST_GS_BASE),
5692 	       vmcs_readl(HOST_TR_BASE));
5693 	pr_err("GDTBase=%016lx IDTBase=%016lx\n",
5694 	       vmcs_readl(HOST_GDTR_BASE), vmcs_readl(HOST_IDTR_BASE));
5695 	pr_err("CR0=%016lx CR3=%016lx CR4=%016lx\n",
5696 	       vmcs_readl(HOST_CR0), vmcs_readl(HOST_CR3),
5697 	       vmcs_readl(HOST_CR4));
5698 	pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
5699 	       vmcs_readl(HOST_IA32_SYSENTER_ESP),
5700 	       vmcs_read32(HOST_IA32_SYSENTER_CS),
5701 	       vmcs_readl(HOST_IA32_SYSENTER_EIP));
5702 	if (vmexit_ctl & (VM_EXIT_LOAD_IA32_PAT | VM_EXIT_LOAD_IA32_EFER))
5703 		pr_err("EFER = 0x%016llx  PAT = 0x%016llx\n",
5704 		       vmcs_read64(HOST_IA32_EFER),
5705 		       vmcs_read64(HOST_IA32_PAT));
5706 	if (cpu_has_load_perf_global_ctrl() &&
5707 	    vmexit_ctl & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
5708 		pr_err("PerfGlobCtl = 0x%016llx\n",
5709 		       vmcs_read64(HOST_IA32_PERF_GLOBAL_CTRL));
5710 
5711 	pr_err("*** Control State ***\n");
5712 	pr_err("PinBased=%08x CPUBased=%08x SecondaryExec=%08x\n",
5713 	       pin_based_exec_ctrl, cpu_based_exec_ctrl, secondary_exec_control);
5714 	pr_err("EntryControls=%08x ExitControls=%08x\n", vmentry_ctl, vmexit_ctl);
5715 	pr_err("ExceptionBitmap=%08x PFECmask=%08x PFECmatch=%08x\n",
5716 	       vmcs_read32(EXCEPTION_BITMAP),
5717 	       vmcs_read32(PAGE_FAULT_ERROR_CODE_MASK),
5718 	       vmcs_read32(PAGE_FAULT_ERROR_CODE_MATCH));
5719 	pr_err("VMEntry: intr_info=%08x errcode=%08x ilen=%08x\n",
5720 	       vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
5721 	       vmcs_read32(VM_ENTRY_EXCEPTION_ERROR_CODE),
5722 	       vmcs_read32(VM_ENTRY_INSTRUCTION_LEN));
5723 	pr_err("VMExit: intr_info=%08x errcode=%08x ilen=%08x\n",
5724 	       vmcs_read32(VM_EXIT_INTR_INFO),
5725 	       vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
5726 	       vmcs_read32(VM_EXIT_INSTRUCTION_LEN));
5727 	pr_err("        reason=%08x qualification=%016lx\n",
5728 	       vmcs_read32(VM_EXIT_REASON), vmcs_readl(EXIT_QUALIFICATION));
5729 	pr_err("IDTVectoring: info=%08x errcode=%08x\n",
5730 	       vmcs_read32(IDT_VECTORING_INFO_FIELD),
5731 	       vmcs_read32(IDT_VECTORING_ERROR_CODE));
5732 	pr_err("TSC Offset = 0x%016llx\n", vmcs_read64(TSC_OFFSET));
5733 	if (secondary_exec_control & SECONDARY_EXEC_TSC_SCALING)
5734 		pr_err("TSC Multiplier = 0x%016llx\n",
5735 		       vmcs_read64(TSC_MULTIPLIER));
5736 	if (cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW)
5737 		pr_err("TPR Threshold = 0x%02x\n", vmcs_read32(TPR_THRESHOLD));
5738 	if (pin_based_exec_ctrl & PIN_BASED_POSTED_INTR)
5739 		pr_err("PostedIntrVec = 0x%02x\n", vmcs_read16(POSTED_INTR_NV));
5740 	if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT))
5741 		pr_err("EPT pointer = 0x%016llx\n", vmcs_read64(EPT_POINTER));
5742 	n = vmcs_read32(CR3_TARGET_COUNT);
5743 	for (i = 0; i + 1 < n; i += 4)
5744 		pr_err("CR3 target%u=%016lx target%u=%016lx\n",
5745 		       i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2),
5746 		       i + 1, vmcs_readl(CR3_TARGET_VALUE0 + i * 2 + 2));
5747 	if (i < n)
5748 		pr_err("CR3 target%u=%016lx\n",
5749 		       i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2));
5750 	if (secondary_exec_control & SECONDARY_EXEC_PAUSE_LOOP_EXITING)
5751 		pr_err("PLE Gap=%08x Window=%08x\n",
5752 		       vmcs_read32(PLE_GAP), vmcs_read32(PLE_WINDOW));
5753 	if (secondary_exec_control & SECONDARY_EXEC_ENABLE_VPID)
5754 		pr_err("Virtual processor ID = 0x%04x\n",
5755 		       vmcs_read16(VIRTUAL_PROCESSOR_ID));
5756 }
5757 
5758 /*
5759  * The guest has exited.  See if we can fix it or if we need userspace
5760  * assistance.
5761  */
5762 static int vmx_handle_exit(struct kvm_vcpu *vcpu)
5763 {
5764 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5765 	u32 exit_reason = vmx->exit_reason;
5766 	u32 vectoring_info = vmx->idt_vectoring_info;
5767 
5768 	trace_kvm_exit(exit_reason, vcpu, KVM_ISA_VMX);
5769 
5770 	/*
5771 	 * Flush logged GPAs PML buffer, this will make dirty_bitmap more
5772 	 * updated. Another good is, in kvm_vm_ioctl_get_dirty_log, before
5773 	 * querying dirty_bitmap, we only need to kick all vcpus out of guest
5774 	 * mode as if vcpus is in root mode, the PML buffer must has been
5775 	 * flushed already.
5776 	 */
5777 	if (enable_pml)
5778 		vmx_flush_pml_buffer(vcpu);
5779 
5780 	/* If guest state is invalid, start emulating */
5781 	if (vmx->emulation_required)
5782 		return handle_invalid_guest_state(vcpu);
5783 
5784 	if (is_guest_mode(vcpu) && nested_vmx_exit_reflected(vcpu, exit_reason))
5785 		return nested_vmx_reflect_vmexit(vcpu, exit_reason);
5786 
5787 	if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) {
5788 		dump_vmcs();
5789 		vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
5790 		vcpu->run->fail_entry.hardware_entry_failure_reason
5791 			= exit_reason;
5792 		return 0;
5793 	}
5794 
5795 	if (unlikely(vmx->fail)) {
5796 		vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
5797 		vcpu->run->fail_entry.hardware_entry_failure_reason
5798 			= vmcs_read32(VM_INSTRUCTION_ERROR);
5799 		return 0;
5800 	}
5801 
5802 	/*
5803 	 * Note:
5804 	 * Do not try to fix EXIT_REASON_EPT_MISCONFIG if it caused by
5805 	 * delivery event since it indicates guest is accessing MMIO.
5806 	 * The vm-exit can be triggered again after return to guest that
5807 	 * will cause infinite loop.
5808 	 */
5809 	if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
5810 			(exit_reason != EXIT_REASON_EXCEPTION_NMI &&
5811 			exit_reason != EXIT_REASON_EPT_VIOLATION &&
5812 			exit_reason != EXIT_REASON_PML_FULL &&
5813 			exit_reason != EXIT_REASON_TASK_SWITCH)) {
5814 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5815 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_DELIVERY_EV;
5816 		vcpu->run->internal.ndata = 3;
5817 		vcpu->run->internal.data[0] = vectoring_info;
5818 		vcpu->run->internal.data[1] = exit_reason;
5819 		vcpu->run->internal.data[2] = vcpu->arch.exit_qualification;
5820 		if (exit_reason == EXIT_REASON_EPT_MISCONFIG) {
5821 			vcpu->run->internal.ndata++;
5822 			vcpu->run->internal.data[3] =
5823 				vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5824 		}
5825 		return 0;
5826 	}
5827 
5828 	if (unlikely(!enable_vnmi &&
5829 		     vmx->loaded_vmcs->soft_vnmi_blocked)) {
5830 		if (vmx_interrupt_allowed(vcpu)) {
5831 			vmx->loaded_vmcs->soft_vnmi_blocked = 0;
5832 		} else if (vmx->loaded_vmcs->vnmi_blocked_time > 1000000000LL &&
5833 			   vcpu->arch.nmi_pending) {
5834 			/*
5835 			 * This CPU don't support us in finding the end of an
5836 			 * NMI-blocked window if the guest runs with IRQs
5837 			 * disabled. So we pull the trigger after 1 s of
5838 			 * futile waiting, but inform the user about this.
5839 			 */
5840 			printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
5841 			       "state on VCPU %d after 1 s timeout\n",
5842 			       __func__, vcpu->vcpu_id);
5843 			vmx->loaded_vmcs->soft_vnmi_blocked = 0;
5844 		}
5845 	}
5846 
5847 	if (exit_reason < kvm_vmx_max_exit_handlers
5848 	    && kvm_vmx_exit_handlers[exit_reason])
5849 		return kvm_vmx_exit_handlers[exit_reason](vcpu);
5850 	else {
5851 		vcpu_unimpl(vcpu, "vmx: unexpected exit reason 0x%x\n",
5852 				exit_reason);
5853 		kvm_queue_exception(vcpu, UD_VECTOR);
5854 		return 1;
5855 	}
5856 }
5857 
5858 /*
5859  * Software based L1D cache flush which is used when microcode providing
5860  * the cache control MSR is not loaded.
5861  *
5862  * The L1D cache is 32 KiB on Nehalem and later microarchitectures, but to
5863  * flush it is required to read in 64 KiB because the replacement algorithm
5864  * is not exactly LRU. This could be sized at runtime via topology
5865  * information but as all relevant affected CPUs have 32KiB L1D cache size
5866  * there is no point in doing so.
5867  */
5868 static void vmx_l1d_flush(struct kvm_vcpu *vcpu)
5869 {
5870 	int size = PAGE_SIZE << L1D_CACHE_ORDER;
5871 
5872 	/*
5873 	 * This code is only executed when the the flush mode is 'cond' or
5874 	 * 'always'
5875 	 */
5876 	if (static_branch_likely(&vmx_l1d_flush_cond)) {
5877 		bool flush_l1d;
5878 
5879 		/*
5880 		 * Clear the per-vcpu flush bit, it gets set again
5881 		 * either from vcpu_run() or from one of the unsafe
5882 		 * VMEXIT handlers.
5883 		 */
5884 		flush_l1d = vcpu->arch.l1tf_flush_l1d;
5885 		vcpu->arch.l1tf_flush_l1d = false;
5886 
5887 		/*
5888 		 * Clear the per-cpu flush bit, it gets set again from
5889 		 * the interrupt handlers.
5890 		 */
5891 		flush_l1d |= kvm_get_cpu_l1tf_flush_l1d();
5892 		kvm_clear_cpu_l1tf_flush_l1d();
5893 
5894 		if (!flush_l1d)
5895 			return;
5896 	}
5897 
5898 	vcpu->stat.l1d_flush++;
5899 
5900 	if (static_cpu_has(X86_FEATURE_FLUSH_L1D)) {
5901 		wrmsrl(MSR_IA32_FLUSH_CMD, L1D_FLUSH);
5902 		return;
5903 	}
5904 
5905 	asm volatile(
5906 		/* First ensure the pages are in the TLB */
5907 		"xorl	%%eax, %%eax\n"
5908 		".Lpopulate_tlb:\n\t"
5909 		"movzbl	(%[flush_pages], %%" _ASM_AX "), %%ecx\n\t"
5910 		"addl	$4096, %%eax\n\t"
5911 		"cmpl	%%eax, %[size]\n\t"
5912 		"jne	.Lpopulate_tlb\n\t"
5913 		"xorl	%%eax, %%eax\n\t"
5914 		"cpuid\n\t"
5915 		/* Now fill the cache */
5916 		"xorl	%%eax, %%eax\n"
5917 		".Lfill_cache:\n"
5918 		"movzbl	(%[flush_pages], %%" _ASM_AX "), %%ecx\n\t"
5919 		"addl	$64, %%eax\n\t"
5920 		"cmpl	%%eax, %[size]\n\t"
5921 		"jne	.Lfill_cache\n\t"
5922 		"lfence\n"
5923 		:: [flush_pages] "r" (vmx_l1d_flush_pages),
5924 		    [size] "r" (size)
5925 		: "eax", "ebx", "ecx", "edx");
5926 }
5927 
5928 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
5929 {
5930 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5931 
5932 	if (is_guest_mode(vcpu) &&
5933 		nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
5934 		return;
5935 
5936 	if (irr == -1 || tpr < irr) {
5937 		vmcs_write32(TPR_THRESHOLD, 0);
5938 		return;
5939 	}
5940 
5941 	vmcs_write32(TPR_THRESHOLD, irr);
5942 }
5943 
5944 void vmx_set_virtual_apic_mode(struct kvm_vcpu *vcpu)
5945 {
5946 	u32 sec_exec_control;
5947 
5948 	if (!lapic_in_kernel(vcpu))
5949 		return;
5950 
5951 	if (!flexpriority_enabled &&
5952 	    !cpu_has_vmx_virtualize_x2apic_mode())
5953 		return;
5954 
5955 	/* Postpone execution until vmcs01 is the current VMCS. */
5956 	if (is_guest_mode(vcpu)) {
5957 		to_vmx(vcpu)->nested.change_vmcs01_virtual_apic_mode = true;
5958 		return;
5959 	}
5960 
5961 	sec_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
5962 	sec_exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
5963 			      SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE);
5964 
5965 	switch (kvm_get_apic_mode(vcpu)) {
5966 	case LAPIC_MODE_INVALID:
5967 		WARN_ONCE(true, "Invalid local APIC state");
5968 	case LAPIC_MODE_DISABLED:
5969 		break;
5970 	case LAPIC_MODE_XAPIC:
5971 		if (flexpriority_enabled) {
5972 			sec_exec_control |=
5973 				SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
5974 			vmx_flush_tlb(vcpu, true);
5975 		}
5976 		break;
5977 	case LAPIC_MODE_X2APIC:
5978 		if (cpu_has_vmx_virtualize_x2apic_mode())
5979 			sec_exec_control |=
5980 				SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
5981 		break;
5982 	}
5983 	vmcs_write32(SECONDARY_VM_EXEC_CONTROL, sec_exec_control);
5984 
5985 	vmx_update_msr_bitmap(vcpu);
5986 }
5987 
5988 static void vmx_set_apic_access_page_addr(struct kvm_vcpu *vcpu, hpa_t hpa)
5989 {
5990 	if (!is_guest_mode(vcpu)) {
5991 		vmcs_write64(APIC_ACCESS_ADDR, hpa);
5992 		vmx_flush_tlb(vcpu, true);
5993 	}
5994 }
5995 
5996 static void vmx_hwapic_isr_update(struct kvm_vcpu *vcpu, int max_isr)
5997 {
5998 	u16 status;
5999 	u8 old;
6000 
6001 	if (max_isr == -1)
6002 		max_isr = 0;
6003 
6004 	status = vmcs_read16(GUEST_INTR_STATUS);
6005 	old = status >> 8;
6006 	if (max_isr != old) {
6007 		status &= 0xff;
6008 		status |= max_isr << 8;
6009 		vmcs_write16(GUEST_INTR_STATUS, status);
6010 	}
6011 }
6012 
6013 static void vmx_set_rvi(int vector)
6014 {
6015 	u16 status;
6016 	u8 old;
6017 
6018 	if (vector == -1)
6019 		vector = 0;
6020 
6021 	status = vmcs_read16(GUEST_INTR_STATUS);
6022 	old = (u8)status & 0xff;
6023 	if ((u8)vector != old) {
6024 		status &= ~0xff;
6025 		status |= (u8)vector;
6026 		vmcs_write16(GUEST_INTR_STATUS, status);
6027 	}
6028 }
6029 
6030 static void vmx_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr)
6031 {
6032 	/*
6033 	 * When running L2, updating RVI is only relevant when
6034 	 * vmcs12 virtual-interrupt-delivery enabled.
6035 	 * However, it can be enabled only when L1 also
6036 	 * intercepts external-interrupts and in that case
6037 	 * we should not update vmcs02 RVI but instead intercept
6038 	 * interrupt. Therefore, do nothing when running L2.
6039 	 */
6040 	if (!is_guest_mode(vcpu))
6041 		vmx_set_rvi(max_irr);
6042 }
6043 
6044 static int vmx_sync_pir_to_irr(struct kvm_vcpu *vcpu)
6045 {
6046 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6047 	int max_irr;
6048 	bool max_irr_updated;
6049 
6050 	WARN_ON(!vcpu->arch.apicv_active);
6051 	if (pi_test_on(&vmx->pi_desc)) {
6052 		pi_clear_on(&vmx->pi_desc);
6053 		/*
6054 		 * IOMMU can write to PIR.ON, so the barrier matters even on UP.
6055 		 * But on x86 this is just a compiler barrier anyway.
6056 		 */
6057 		smp_mb__after_atomic();
6058 		max_irr_updated =
6059 			kvm_apic_update_irr(vcpu, vmx->pi_desc.pir, &max_irr);
6060 
6061 		/*
6062 		 * If we are running L2 and L1 has a new pending interrupt
6063 		 * which can be injected, we should re-evaluate
6064 		 * what should be done with this new L1 interrupt.
6065 		 * If L1 intercepts external-interrupts, we should
6066 		 * exit from L2 to L1. Otherwise, interrupt should be
6067 		 * delivered directly to L2.
6068 		 */
6069 		if (is_guest_mode(vcpu) && max_irr_updated) {
6070 			if (nested_exit_on_intr(vcpu))
6071 				kvm_vcpu_exiting_guest_mode(vcpu);
6072 			else
6073 				kvm_make_request(KVM_REQ_EVENT, vcpu);
6074 		}
6075 	} else {
6076 		max_irr = kvm_lapic_find_highest_irr(vcpu);
6077 	}
6078 	vmx_hwapic_irr_update(vcpu, max_irr);
6079 	return max_irr;
6080 }
6081 
6082 static void vmx_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap)
6083 {
6084 	if (!kvm_vcpu_apicv_active(vcpu))
6085 		return;
6086 
6087 	vmcs_write64(EOI_EXIT_BITMAP0, eoi_exit_bitmap[0]);
6088 	vmcs_write64(EOI_EXIT_BITMAP1, eoi_exit_bitmap[1]);
6089 	vmcs_write64(EOI_EXIT_BITMAP2, eoi_exit_bitmap[2]);
6090 	vmcs_write64(EOI_EXIT_BITMAP3, eoi_exit_bitmap[3]);
6091 }
6092 
6093 static void vmx_apicv_post_state_restore(struct kvm_vcpu *vcpu)
6094 {
6095 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6096 
6097 	pi_clear_on(&vmx->pi_desc);
6098 	memset(vmx->pi_desc.pir, 0, sizeof(vmx->pi_desc.pir));
6099 }
6100 
6101 static void vmx_complete_atomic_exit(struct vcpu_vmx *vmx)
6102 {
6103 	u32 exit_intr_info = 0;
6104 	u16 basic_exit_reason = (u16)vmx->exit_reason;
6105 
6106 	if (!(basic_exit_reason == EXIT_REASON_MCE_DURING_VMENTRY
6107 	      || basic_exit_reason == EXIT_REASON_EXCEPTION_NMI))
6108 		return;
6109 
6110 	if (!(vmx->exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY))
6111 		exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6112 	vmx->exit_intr_info = exit_intr_info;
6113 
6114 	/* if exit due to PF check for async PF */
6115 	if (is_page_fault(exit_intr_info))
6116 		vmx->vcpu.arch.apf.host_apf_reason = kvm_read_and_reset_pf_reason();
6117 
6118 	/* Handle machine checks before interrupts are enabled */
6119 	if (basic_exit_reason == EXIT_REASON_MCE_DURING_VMENTRY ||
6120 	    is_machine_check(exit_intr_info))
6121 		kvm_machine_check();
6122 
6123 	/* We need to handle NMIs before interrupts are enabled */
6124 	if (is_nmi(exit_intr_info)) {
6125 		kvm_before_interrupt(&vmx->vcpu);
6126 		asm("int $2");
6127 		kvm_after_interrupt(&vmx->vcpu);
6128 	}
6129 }
6130 
6131 static void vmx_handle_external_intr(struct kvm_vcpu *vcpu)
6132 {
6133 	u32 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6134 
6135 	if ((exit_intr_info & (INTR_INFO_VALID_MASK | INTR_INFO_INTR_TYPE_MASK))
6136 			== (INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR)) {
6137 		unsigned int vector;
6138 		unsigned long entry;
6139 		gate_desc *desc;
6140 		struct vcpu_vmx *vmx = to_vmx(vcpu);
6141 #ifdef CONFIG_X86_64
6142 		unsigned long tmp;
6143 #endif
6144 
6145 		vector =  exit_intr_info & INTR_INFO_VECTOR_MASK;
6146 		desc = (gate_desc *)vmx->host_idt_base + vector;
6147 		entry = gate_offset(desc);
6148 		asm volatile(
6149 #ifdef CONFIG_X86_64
6150 			"mov %%" _ASM_SP ", %[sp]\n\t"
6151 			"and $0xfffffffffffffff0, %%" _ASM_SP "\n\t"
6152 			"push $%c[ss]\n\t"
6153 			"push %[sp]\n\t"
6154 #endif
6155 			"pushf\n\t"
6156 			__ASM_SIZE(push) " $%c[cs]\n\t"
6157 			CALL_NOSPEC
6158 			:
6159 #ifdef CONFIG_X86_64
6160 			[sp]"=&r"(tmp),
6161 #endif
6162 			ASM_CALL_CONSTRAINT
6163 			:
6164 			THUNK_TARGET(entry),
6165 			[ss]"i"(__KERNEL_DS),
6166 			[cs]"i"(__KERNEL_CS)
6167 			);
6168 	}
6169 }
6170 STACK_FRAME_NON_STANDARD(vmx_handle_external_intr);
6171 
6172 static bool vmx_has_emulated_msr(int index)
6173 {
6174 	switch (index) {
6175 	case MSR_IA32_SMBASE:
6176 		/*
6177 		 * We cannot do SMM unless we can run the guest in big
6178 		 * real mode.
6179 		 */
6180 		return enable_unrestricted_guest || emulate_invalid_guest_state;
6181 	case MSR_AMD64_VIRT_SPEC_CTRL:
6182 		/* This is AMD only.  */
6183 		return false;
6184 	default:
6185 		return true;
6186 	}
6187 }
6188 
6189 static bool vmx_pt_supported(void)
6190 {
6191 	return pt_mode == PT_MODE_HOST_GUEST;
6192 }
6193 
6194 static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx)
6195 {
6196 	u32 exit_intr_info;
6197 	bool unblock_nmi;
6198 	u8 vector;
6199 	bool idtv_info_valid;
6200 
6201 	idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK;
6202 
6203 	if (enable_vnmi) {
6204 		if (vmx->loaded_vmcs->nmi_known_unmasked)
6205 			return;
6206 		/*
6207 		 * Can't use vmx->exit_intr_info since we're not sure what
6208 		 * the exit reason is.
6209 		 */
6210 		exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6211 		unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
6212 		vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
6213 		/*
6214 		 * SDM 3: 27.7.1.2 (September 2008)
6215 		 * Re-set bit "block by NMI" before VM entry if vmexit caused by
6216 		 * a guest IRET fault.
6217 		 * SDM 3: 23.2.2 (September 2008)
6218 		 * Bit 12 is undefined in any of the following cases:
6219 		 *  If the VM exit sets the valid bit in the IDT-vectoring
6220 		 *   information field.
6221 		 *  If the VM exit is due to a double fault.
6222 		 */
6223 		if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
6224 		    vector != DF_VECTOR && !idtv_info_valid)
6225 			vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
6226 				      GUEST_INTR_STATE_NMI);
6227 		else
6228 			vmx->loaded_vmcs->nmi_known_unmasked =
6229 				!(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO)
6230 				  & GUEST_INTR_STATE_NMI);
6231 	} else if (unlikely(vmx->loaded_vmcs->soft_vnmi_blocked))
6232 		vmx->loaded_vmcs->vnmi_blocked_time +=
6233 			ktime_to_ns(ktime_sub(ktime_get(),
6234 					      vmx->loaded_vmcs->entry_time));
6235 }
6236 
6237 static void __vmx_complete_interrupts(struct kvm_vcpu *vcpu,
6238 				      u32 idt_vectoring_info,
6239 				      int instr_len_field,
6240 				      int error_code_field)
6241 {
6242 	u8 vector;
6243 	int type;
6244 	bool idtv_info_valid;
6245 
6246 	idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
6247 
6248 	vcpu->arch.nmi_injected = false;
6249 	kvm_clear_exception_queue(vcpu);
6250 	kvm_clear_interrupt_queue(vcpu);
6251 
6252 	if (!idtv_info_valid)
6253 		return;
6254 
6255 	kvm_make_request(KVM_REQ_EVENT, vcpu);
6256 
6257 	vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
6258 	type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
6259 
6260 	switch (type) {
6261 	case INTR_TYPE_NMI_INTR:
6262 		vcpu->arch.nmi_injected = true;
6263 		/*
6264 		 * SDM 3: 27.7.1.2 (September 2008)
6265 		 * Clear bit "block by NMI" before VM entry if a NMI
6266 		 * delivery faulted.
6267 		 */
6268 		vmx_set_nmi_mask(vcpu, false);
6269 		break;
6270 	case INTR_TYPE_SOFT_EXCEPTION:
6271 		vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
6272 		/* fall through */
6273 	case INTR_TYPE_HARD_EXCEPTION:
6274 		if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
6275 			u32 err = vmcs_read32(error_code_field);
6276 			kvm_requeue_exception_e(vcpu, vector, err);
6277 		} else
6278 			kvm_requeue_exception(vcpu, vector);
6279 		break;
6280 	case INTR_TYPE_SOFT_INTR:
6281 		vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
6282 		/* fall through */
6283 	case INTR_TYPE_EXT_INTR:
6284 		kvm_queue_interrupt(vcpu, vector, type == INTR_TYPE_SOFT_INTR);
6285 		break;
6286 	default:
6287 		break;
6288 	}
6289 }
6290 
6291 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
6292 {
6293 	__vmx_complete_interrupts(&vmx->vcpu, vmx->idt_vectoring_info,
6294 				  VM_EXIT_INSTRUCTION_LEN,
6295 				  IDT_VECTORING_ERROR_CODE);
6296 }
6297 
6298 static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
6299 {
6300 	__vmx_complete_interrupts(vcpu,
6301 				  vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
6302 				  VM_ENTRY_INSTRUCTION_LEN,
6303 				  VM_ENTRY_EXCEPTION_ERROR_CODE);
6304 
6305 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
6306 }
6307 
6308 static void atomic_switch_perf_msrs(struct vcpu_vmx *vmx)
6309 {
6310 	int i, nr_msrs;
6311 	struct perf_guest_switch_msr *msrs;
6312 
6313 	msrs = perf_guest_get_msrs(&nr_msrs);
6314 
6315 	if (!msrs)
6316 		return;
6317 
6318 	for (i = 0; i < nr_msrs; i++)
6319 		if (msrs[i].host == msrs[i].guest)
6320 			clear_atomic_switch_msr(vmx, msrs[i].msr);
6321 		else
6322 			add_atomic_switch_msr(vmx, msrs[i].msr, msrs[i].guest,
6323 					msrs[i].host, false);
6324 }
6325 
6326 static void vmx_arm_hv_timer(struct vcpu_vmx *vmx, u32 val)
6327 {
6328 	vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, val);
6329 	if (!vmx->loaded_vmcs->hv_timer_armed)
6330 		vmcs_set_bits(PIN_BASED_VM_EXEC_CONTROL,
6331 			      PIN_BASED_VMX_PREEMPTION_TIMER);
6332 	vmx->loaded_vmcs->hv_timer_armed = true;
6333 }
6334 
6335 static void vmx_update_hv_timer(struct kvm_vcpu *vcpu)
6336 {
6337 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6338 	u64 tscl;
6339 	u32 delta_tsc;
6340 
6341 	if (vmx->req_immediate_exit) {
6342 		vmx_arm_hv_timer(vmx, 0);
6343 		return;
6344 	}
6345 
6346 	if (vmx->hv_deadline_tsc != -1) {
6347 		tscl = rdtsc();
6348 		if (vmx->hv_deadline_tsc > tscl)
6349 			/* set_hv_timer ensures the delta fits in 32-bits */
6350 			delta_tsc = (u32)((vmx->hv_deadline_tsc - tscl) >>
6351 				cpu_preemption_timer_multi);
6352 		else
6353 			delta_tsc = 0;
6354 
6355 		vmx_arm_hv_timer(vmx, delta_tsc);
6356 		return;
6357 	}
6358 
6359 	if (vmx->loaded_vmcs->hv_timer_armed)
6360 		vmcs_clear_bits(PIN_BASED_VM_EXEC_CONTROL,
6361 				PIN_BASED_VMX_PREEMPTION_TIMER);
6362 	vmx->loaded_vmcs->hv_timer_armed = false;
6363 }
6364 
6365 static void vmx_vcpu_run(struct kvm_vcpu *vcpu)
6366 {
6367 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6368 	unsigned long cr3, cr4, evmcs_rsp;
6369 
6370 	/* Record the guest's net vcpu time for enforced NMI injections. */
6371 	if (unlikely(!enable_vnmi &&
6372 		     vmx->loaded_vmcs->soft_vnmi_blocked))
6373 		vmx->loaded_vmcs->entry_time = ktime_get();
6374 
6375 	/* Don't enter VMX if guest state is invalid, let the exit handler
6376 	   start emulation until we arrive back to a valid state */
6377 	if (vmx->emulation_required)
6378 		return;
6379 
6380 	if (vmx->ple_window_dirty) {
6381 		vmx->ple_window_dirty = false;
6382 		vmcs_write32(PLE_WINDOW, vmx->ple_window);
6383 	}
6384 
6385 	if (vmx->nested.need_vmcs12_sync)
6386 		nested_sync_from_vmcs12(vcpu);
6387 
6388 	if (test_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_dirty))
6389 		vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
6390 	if (test_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_dirty))
6391 		vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
6392 
6393 	cr3 = __get_current_cr3_fast();
6394 	if (unlikely(cr3 != vmx->loaded_vmcs->host_state.cr3)) {
6395 		vmcs_writel(HOST_CR3, cr3);
6396 		vmx->loaded_vmcs->host_state.cr3 = cr3;
6397 	}
6398 
6399 	cr4 = cr4_read_shadow();
6400 	if (unlikely(cr4 != vmx->loaded_vmcs->host_state.cr4)) {
6401 		vmcs_writel(HOST_CR4, cr4);
6402 		vmx->loaded_vmcs->host_state.cr4 = cr4;
6403 	}
6404 
6405 	/* When single-stepping over STI and MOV SS, we must clear the
6406 	 * corresponding interruptibility bits in the guest state. Otherwise
6407 	 * vmentry fails as it then expects bit 14 (BS) in pending debug
6408 	 * exceptions being set, but that's not correct for the guest debugging
6409 	 * case. */
6410 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6411 		vmx_set_interrupt_shadow(vcpu, 0);
6412 
6413 	if (static_cpu_has(X86_FEATURE_PKU) &&
6414 	    kvm_read_cr4_bits(vcpu, X86_CR4_PKE) &&
6415 	    vcpu->arch.pkru != vmx->host_pkru)
6416 		__write_pkru(vcpu->arch.pkru);
6417 
6418 	pt_guest_enter(vmx);
6419 
6420 	atomic_switch_perf_msrs(vmx);
6421 
6422 	vmx_update_hv_timer(vcpu);
6423 
6424 	/*
6425 	 * If this vCPU has touched SPEC_CTRL, restore the guest's value if
6426 	 * it's non-zero. Since vmentry is serialising on affected CPUs, there
6427 	 * is no need to worry about the conditional branch over the wrmsr
6428 	 * being speculatively taken.
6429 	 */
6430 	x86_spec_ctrl_set_guest(vmx->spec_ctrl, 0);
6431 
6432 	vmx->__launched = vmx->loaded_vmcs->launched;
6433 
6434 	evmcs_rsp = static_branch_unlikely(&enable_evmcs) ?
6435 		(unsigned long)&current_evmcs->host_rsp : 0;
6436 
6437 	if (static_branch_unlikely(&vmx_l1d_should_flush))
6438 		vmx_l1d_flush(vcpu);
6439 
6440 	asm(
6441 		/* Store host registers */
6442 		"push %%" _ASM_DX "; push %%" _ASM_BP ";"
6443 		"push %%" _ASM_CX " \n\t" /* placeholder for guest rcx */
6444 		"push %%" _ASM_CX " \n\t"
6445 		"sub $%c[wordsize], %%" _ASM_SP "\n\t" /* temporarily adjust RSP for CALL */
6446 		"cmp %%" _ASM_SP ", %c[host_rsp](%%" _ASM_CX ") \n\t"
6447 		"je 1f \n\t"
6448 		"mov %%" _ASM_SP ", %c[host_rsp](%%" _ASM_CX ") \n\t"
6449 		/* Avoid VMWRITE when Enlightened VMCS is in use */
6450 		"test %%" _ASM_SI ", %%" _ASM_SI " \n\t"
6451 		"jz 2f \n\t"
6452 		"mov %%" _ASM_SP ", (%%" _ASM_SI ") \n\t"
6453 		"jmp 1f \n\t"
6454 		"2: \n\t"
6455 		__ex("vmwrite %%" _ASM_SP ", %%" _ASM_DX) "\n\t"
6456 		"1: \n\t"
6457 		"add $%c[wordsize], %%" _ASM_SP "\n\t" /* un-adjust RSP */
6458 
6459 		/* Reload cr2 if changed */
6460 		"mov %c[cr2](%%" _ASM_CX "), %%" _ASM_AX " \n\t"
6461 		"mov %%cr2, %%" _ASM_DX " \n\t"
6462 		"cmp %%" _ASM_AX ", %%" _ASM_DX " \n\t"
6463 		"je 3f \n\t"
6464 		"mov %%" _ASM_AX", %%cr2 \n\t"
6465 		"3: \n\t"
6466 		/* Check if vmlaunch or vmresume is needed */
6467 		"cmpl $0, %c[launched](%%" _ASM_CX ") \n\t"
6468 		/* Load guest registers.  Don't clobber flags. */
6469 		"mov %c[rax](%%" _ASM_CX "), %%" _ASM_AX " \n\t"
6470 		"mov %c[rbx](%%" _ASM_CX "), %%" _ASM_BX " \n\t"
6471 		"mov %c[rdx](%%" _ASM_CX "), %%" _ASM_DX " \n\t"
6472 		"mov %c[rsi](%%" _ASM_CX "), %%" _ASM_SI " \n\t"
6473 		"mov %c[rdi](%%" _ASM_CX "), %%" _ASM_DI " \n\t"
6474 		"mov %c[rbp](%%" _ASM_CX "), %%" _ASM_BP " \n\t"
6475 #ifdef CONFIG_X86_64
6476 		"mov %c[r8](%%" _ASM_CX "),  %%r8  \n\t"
6477 		"mov %c[r9](%%" _ASM_CX "),  %%r9  \n\t"
6478 		"mov %c[r10](%%" _ASM_CX "), %%r10 \n\t"
6479 		"mov %c[r11](%%" _ASM_CX "), %%r11 \n\t"
6480 		"mov %c[r12](%%" _ASM_CX "), %%r12 \n\t"
6481 		"mov %c[r13](%%" _ASM_CX "), %%r13 \n\t"
6482 		"mov %c[r14](%%" _ASM_CX "), %%r14 \n\t"
6483 		"mov %c[r15](%%" _ASM_CX "), %%r15 \n\t"
6484 #endif
6485 		/* Load guest RCX.  This kills the vmx_vcpu pointer! */
6486 		"mov %c[rcx](%%" _ASM_CX "), %%" _ASM_CX " \n\t"
6487 
6488 		/* Enter guest mode */
6489 		"call vmx_vmenter\n\t"
6490 
6491 		/* Save guest's RCX to the stack placeholder (see above) */
6492 		"mov %%" _ASM_CX ", %c[wordsize](%%" _ASM_SP ") \n\t"
6493 
6494 		/* Load host's RCX, i.e. the vmx_vcpu pointer */
6495 		"pop %%" _ASM_CX " \n\t"
6496 
6497 		/* Set vmx->fail based on EFLAGS.{CF,ZF} */
6498 		"setbe %c[fail](%%" _ASM_CX ")\n\t"
6499 
6500 		/* Save all guest registers, including RCX from the stack */
6501 		"mov %%" _ASM_AX ", %c[rax](%%" _ASM_CX ") \n\t"
6502 		"mov %%" _ASM_BX ", %c[rbx](%%" _ASM_CX ") \n\t"
6503 		__ASM_SIZE(pop) " %c[rcx](%%" _ASM_CX ") \n\t"
6504 		"mov %%" _ASM_DX ", %c[rdx](%%" _ASM_CX ") \n\t"
6505 		"mov %%" _ASM_SI ", %c[rsi](%%" _ASM_CX ") \n\t"
6506 		"mov %%" _ASM_DI ", %c[rdi](%%" _ASM_CX ") \n\t"
6507 		"mov %%" _ASM_BP ", %c[rbp](%%" _ASM_CX ") \n\t"
6508 #ifdef CONFIG_X86_64
6509 		"mov %%r8,  %c[r8](%%" _ASM_CX ") \n\t"
6510 		"mov %%r9,  %c[r9](%%" _ASM_CX ") \n\t"
6511 		"mov %%r10, %c[r10](%%" _ASM_CX ") \n\t"
6512 		"mov %%r11, %c[r11](%%" _ASM_CX ") \n\t"
6513 		"mov %%r12, %c[r12](%%" _ASM_CX ") \n\t"
6514 		"mov %%r13, %c[r13](%%" _ASM_CX ") \n\t"
6515 		"mov %%r14, %c[r14](%%" _ASM_CX ") \n\t"
6516 		"mov %%r15, %c[r15](%%" _ASM_CX ") \n\t"
6517 		/*
6518 		* Clear host registers marked as clobbered to prevent
6519 		* speculative use.
6520 		*/
6521 		"xor %%r8d,  %%r8d \n\t"
6522 		"xor %%r9d,  %%r9d \n\t"
6523 		"xor %%r10d, %%r10d \n\t"
6524 		"xor %%r11d, %%r11d \n\t"
6525 		"xor %%r12d, %%r12d \n\t"
6526 		"xor %%r13d, %%r13d \n\t"
6527 		"xor %%r14d, %%r14d \n\t"
6528 		"xor %%r15d, %%r15d \n\t"
6529 #endif
6530 		"mov %%cr2, %%" _ASM_AX "   \n\t"
6531 		"mov %%" _ASM_AX ", %c[cr2](%%" _ASM_CX ") \n\t"
6532 
6533 		"xor %%eax, %%eax \n\t"
6534 		"xor %%ebx, %%ebx \n\t"
6535 		"xor %%esi, %%esi \n\t"
6536 		"xor %%edi, %%edi \n\t"
6537 		"pop  %%" _ASM_BP "; pop  %%" _ASM_DX " \n\t"
6538 	      : ASM_CALL_CONSTRAINT
6539 	      : "c"(vmx), "d"((unsigned long)HOST_RSP), "S"(evmcs_rsp),
6540 		[launched]"i"(offsetof(struct vcpu_vmx, __launched)),
6541 		[fail]"i"(offsetof(struct vcpu_vmx, fail)),
6542 		[host_rsp]"i"(offsetof(struct vcpu_vmx, host_rsp)),
6543 		[rax]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RAX])),
6544 		[rbx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBX])),
6545 		[rcx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RCX])),
6546 		[rdx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDX])),
6547 		[rsi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RSI])),
6548 		[rdi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDI])),
6549 		[rbp]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBP])),
6550 #ifdef CONFIG_X86_64
6551 		[r8]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R8])),
6552 		[r9]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R9])),
6553 		[r10]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R10])),
6554 		[r11]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R11])),
6555 		[r12]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R12])),
6556 		[r13]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R13])),
6557 		[r14]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R14])),
6558 		[r15]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R15])),
6559 #endif
6560 		[cr2]"i"(offsetof(struct vcpu_vmx, vcpu.arch.cr2)),
6561 		[wordsize]"i"(sizeof(ulong))
6562 	      : "cc", "memory"
6563 #ifdef CONFIG_X86_64
6564 		, "rax", "rbx", "rdi"
6565 		, "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15"
6566 #else
6567 		, "eax", "ebx", "edi"
6568 #endif
6569 	      );
6570 
6571 	/*
6572 	 * We do not use IBRS in the kernel. If this vCPU has used the
6573 	 * SPEC_CTRL MSR it may have left it on; save the value and
6574 	 * turn it off. This is much more efficient than blindly adding
6575 	 * it to the atomic save/restore list. Especially as the former
6576 	 * (Saving guest MSRs on vmexit) doesn't even exist in KVM.
6577 	 *
6578 	 * For non-nested case:
6579 	 * If the L01 MSR bitmap does not intercept the MSR, then we need to
6580 	 * save it.
6581 	 *
6582 	 * For nested case:
6583 	 * If the L02 MSR bitmap does not intercept the MSR, then we need to
6584 	 * save it.
6585 	 */
6586 	if (unlikely(!msr_write_intercepted(vcpu, MSR_IA32_SPEC_CTRL)))
6587 		vmx->spec_ctrl = native_read_msr(MSR_IA32_SPEC_CTRL);
6588 
6589 	x86_spec_ctrl_restore_host(vmx->spec_ctrl, 0);
6590 
6591 	/* Eliminate branch target predictions from guest mode */
6592 	vmexit_fill_RSB();
6593 
6594 	/* All fields are clean at this point */
6595 	if (static_branch_unlikely(&enable_evmcs))
6596 		current_evmcs->hv_clean_fields |=
6597 			HV_VMX_ENLIGHTENED_CLEAN_FIELD_ALL;
6598 
6599 	/* MSR_IA32_DEBUGCTLMSR is zeroed on vmexit. Restore it if needed */
6600 	if (vmx->host_debugctlmsr)
6601 		update_debugctlmsr(vmx->host_debugctlmsr);
6602 
6603 #ifndef CONFIG_X86_64
6604 	/*
6605 	 * The sysexit path does not restore ds/es, so we must set them to
6606 	 * a reasonable value ourselves.
6607 	 *
6608 	 * We can't defer this to vmx_prepare_switch_to_host() since that
6609 	 * function may be executed in interrupt context, which saves and
6610 	 * restore segments around it, nullifying its effect.
6611 	 */
6612 	loadsegment(ds, __USER_DS);
6613 	loadsegment(es, __USER_DS);
6614 #endif
6615 
6616 	vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
6617 				  | (1 << VCPU_EXREG_RFLAGS)
6618 				  | (1 << VCPU_EXREG_PDPTR)
6619 				  | (1 << VCPU_EXREG_SEGMENTS)
6620 				  | (1 << VCPU_EXREG_CR3));
6621 	vcpu->arch.regs_dirty = 0;
6622 
6623 	pt_guest_exit(vmx);
6624 
6625 	/*
6626 	 * eager fpu is enabled if PKEY is supported and CR4 is switched
6627 	 * back on host, so it is safe to read guest PKRU from current
6628 	 * XSAVE.
6629 	 */
6630 	if (static_cpu_has(X86_FEATURE_PKU) &&
6631 	    kvm_read_cr4_bits(vcpu, X86_CR4_PKE)) {
6632 		vcpu->arch.pkru = __read_pkru();
6633 		if (vcpu->arch.pkru != vmx->host_pkru)
6634 			__write_pkru(vmx->host_pkru);
6635 	}
6636 
6637 	vmx->nested.nested_run_pending = 0;
6638 	vmx->idt_vectoring_info = 0;
6639 
6640 	vmx->exit_reason = vmx->fail ? 0xdead : vmcs_read32(VM_EXIT_REASON);
6641 	if (vmx->fail || (vmx->exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY))
6642 		return;
6643 
6644 	vmx->loaded_vmcs->launched = 1;
6645 	vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
6646 
6647 	vmx_complete_atomic_exit(vmx);
6648 	vmx_recover_nmi_blocking(vmx);
6649 	vmx_complete_interrupts(vmx);
6650 }
6651 STACK_FRAME_NON_STANDARD(vmx_vcpu_run);
6652 
6653 static struct kvm *vmx_vm_alloc(void)
6654 {
6655 	struct kvm_vmx *kvm_vmx = vzalloc(sizeof(struct kvm_vmx));
6656 	return &kvm_vmx->kvm;
6657 }
6658 
6659 static void vmx_vm_free(struct kvm *kvm)
6660 {
6661 	vfree(to_kvm_vmx(kvm));
6662 }
6663 
6664 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
6665 {
6666 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6667 
6668 	if (enable_pml)
6669 		vmx_destroy_pml_buffer(vmx);
6670 	free_vpid(vmx->vpid);
6671 	leave_guest_mode(vcpu);
6672 	nested_vmx_free_vcpu(vcpu);
6673 	free_loaded_vmcs(vmx->loaded_vmcs);
6674 	kfree(vmx->guest_msrs);
6675 	kvm_vcpu_uninit(vcpu);
6676 	kmem_cache_free(x86_fpu_cache, vmx->vcpu.arch.guest_fpu);
6677 	kmem_cache_free(kvm_vcpu_cache, vmx);
6678 }
6679 
6680 static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
6681 {
6682 	int err;
6683 	struct vcpu_vmx *vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
6684 	unsigned long *msr_bitmap;
6685 	int cpu;
6686 
6687 	if (!vmx)
6688 		return ERR_PTR(-ENOMEM);
6689 
6690 	vmx->vcpu.arch.guest_fpu = kmem_cache_zalloc(x86_fpu_cache, GFP_KERNEL);
6691 	if (!vmx->vcpu.arch.guest_fpu) {
6692 		printk(KERN_ERR "kvm: failed to allocate vcpu's fpu\n");
6693 		err = -ENOMEM;
6694 		goto free_partial_vcpu;
6695 	}
6696 
6697 	vmx->vpid = allocate_vpid();
6698 
6699 	err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
6700 	if (err)
6701 		goto free_vcpu;
6702 
6703 	err = -ENOMEM;
6704 
6705 	/*
6706 	 * If PML is turned on, failure on enabling PML just results in failure
6707 	 * of creating the vcpu, therefore we can simplify PML logic (by
6708 	 * avoiding dealing with cases, such as enabling PML partially on vcpus
6709 	 * for the guest, etc.
6710 	 */
6711 	if (enable_pml) {
6712 		vmx->pml_pg = alloc_page(GFP_KERNEL | __GFP_ZERO);
6713 		if (!vmx->pml_pg)
6714 			goto uninit_vcpu;
6715 	}
6716 
6717 	vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
6718 	BUILD_BUG_ON(ARRAY_SIZE(vmx_msr_index) * sizeof(vmx->guest_msrs[0])
6719 		     > PAGE_SIZE);
6720 
6721 	if (!vmx->guest_msrs)
6722 		goto free_pml;
6723 
6724 	err = alloc_loaded_vmcs(&vmx->vmcs01);
6725 	if (err < 0)
6726 		goto free_msrs;
6727 
6728 	msr_bitmap = vmx->vmcs01.msr_bitmap;
6729 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_TSC, MSR_TYPE_R);
6730 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_FS_BASE, MSR_TYPE_RW);
6731 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_GS_BASE, MSR_TYPE_RW);
6732 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_KERNEL_GS_BASE, MSR_TYPE_RW);
6733 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_CS, MSR_TYPE_RW);
6734 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_ESP, MSR_TYPE_RW);
6735 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_EIP, MSR_TYPE_RW);
6736 	vmx->msr_bitmap_mode = 0;
6737 
6738 	vmx->loaded_vmcs = &vmx->vmcs01;
6739 	cpu = get_cpu();
6740 	vmx_vcpu_load(&vmx->vcpu, cpu);
6741 	vmx->vcpu.cpu = cpu;
6742 	vmx_vcpu_setup(vmx);
6743 	vmx_vcpu_put(&vmx->vcpu);
6744 	put_cpu();
6745 	if (cpu_need_virtualize_apic_accesses(&vmx->vcpu)) {
6746 		err = alloc_apic_access_page(kvm);
6747 		if (err)
6748 			goto free_vmcs;
6749 	}
6750 
6751 	if (enable_ept && !enable_unrestricted_guest) {
6752 		err = init_rmode_identity_map(kvm);
6753 		if (err)
6754 			goto free_vmcs;
6755 	}
6756 
6757 	if (nested)
6758 		nested_vmx_setup_ctls_msrs(&vmx->nested.msrs,
6759 					   vmx_capability.ept,
6760 					   kvm_vcpu_apicv_active(&vmx->vcpu));
6761 	else
6762 		memset(&vmx->nested.msrs, 0, sizeof(vmx->nested.msrs));
6763 
6764 	vmx->nested.posted_intr_nv = -1;
6765 	vmx->nested.current_vmptr = -1ull;
6766 
6767 	vmx->msr_ia32_feature_control_valid_bits = FEATURE_CONTROL_LOCKED;
6768 
6769 	/*
6770 	 * Enforce invariant: pi_desc.nv is always either POSTED_INTR_VECTOR
6771 	 * or POSTED_INTR_WAKEUP_VECTOR.
6772 	 */
6773 	vmx->pi_desc.nv = POSTED_INTR_VECTOR;
6774 	vmx->pi_desc.sn = 1;
6775 
6776 	vmx->ept_pointer = INVALID_PAGE;
6777 
6778 	return &vmx->vcpu;
6779 
6780 free_vmcs:
6781 	free_loaded_vmcs(vmx->loaded_vmcs);
6782 free_msrs:
6783 	kfree(vmx->guest_msrs);
6784 free_pml:
6785 	vmx_destroy_pml_buffer(vmx);
6786 uninit_vcpu:
6787 	kvm_vcpu_uninit(&vmx->vcpu);
6788 free_vcpu:
6789 	free_vpid(vmx->vpid);
6790 	kmem_cache_free(x86_fpu_cache, vmx->vcpu.arch.guest_fpu);
6791 free_partial_vcpu:
6792 	kmem_cache_free(kvm_vcpu_cache, vmx);
6793 	return ERR_PTR(err);
6794 }
6795 
6796 #define L1TF_MSG_SMT "L1TF CPU bug present and SMT on, data leak possible. See CVE-2018-3646 and https://www.kernel.org/doc/html/latest/admin-guide/l1tf.html for details.\n"
6797 #define L1TF_MSG_L1D "L1TF CPU bug present and virtualization mitigation disabled, data leak possible. See CVE-2018-3646 and https://www.kernel.org/doc/html/latest/admin-guide/l1tf.html for details.\n"
6798 
6799 static int vmx_vm_init(struct kvm *kvm)
6800 {
6801 	spin_lock_init(&to_kvm_vmx(kvm)->ept_pointer_lock);
6802 
6803 	if (!ple_gap)
6804 		kvm->arch.pause_in_guest = true;
6805 
6806 	if (boot_cpu_has(X86_BUG_L1TF) && enable_ept) {
6807 		switch (l1tf_mitigation) {
6808 		case L1TF_MITIGATION_OFF:
6809 		case L1TF_MITIGATION_FLUSH_NOWARN:
6810 			/* 'I explicitly don't care' is set */
6811 			break;
6812 		case L1TF_MITIGATION_FLUSH:
6813 		case L1TF_MITIGATION_FLUSH_NOSMT:
6814 		case L1TF_MITIGATION_FULL:
6815 			/*
6816 			 * Warn upon starting the first VM in a potentially
6817 			 * insecure environment.
6818 			 */
6819 			if (cpu_smt_control == CPU_SMT_ENABLED)
6820 				pr_warn_once(L1TF_MSG_SMT);
6821 			if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_NEVER)
6822 				pr_warn_once(L1TF_MSG_L1D);
6823 			break;
6824 		case L1TF_MITIGATION_FULL_FORCE:
6825 			/* Flush is enforced */
6826 			break;
6827 		}
6828 	}
6829 	return 0;
6830 }
6831 
6832 static void __init vmx_check_processor_compat(void *rtn)
6833 {
6834 	struct vmcs_config vmcs_conf;
6835 	struct vmx_capability vmx_cap;
6836 
6837 	*(int *)rtn = 0;
6838 	if (setup_vmcs_config(&vmcs_conf, &vmx_cap) < 0)
6839 		*(int *)rtn = -EIO;
6840 	if (nested)
6841 		nested_vmx_setup_ctls_msrs(&vmcs_conf.nested, vmx_cap.ept,
6842 					   enable_apicv);
6843 	if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
6844 		printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
6845 				smp_processor_id());
6846 		*(int *)rtn = -EIO;
6847 	}
6848 }
6849 
6850 static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
6851 {
6852 	u8 cache;
6853 	u64 ipat = 0;
6854 
6855 	/* For VT-d and EPT combination
6856 	 * 1. MMIO: always map as UC
6857 	 * 2. EPT with VT-d:
6858 	 *   a. VT-d without snooping control feature: can't guarantee the
6859 	 *	result, try to trust guest.
6860 	 *   b. VT-d with snooping control feature: snooping control feature of
6861 	 *	VT-d engine can guarantee the cache correctness. Just set it
6862 	 *	to WB to keep consistent with host. So the same as item 3.
6863 	 * 3. EPT without VT-d: always map as WB and set IPAT=1 to keep
6864 	 *    consistent with host MTRR
6865 	 */
6866 	if (is_mmio) {
6867 		cache = MTRR_TYPE_UNCACHABLE;
6868 		goto exit;
6869 	}
6870 
6871 	if (!kvm_arch_has_noncoherent_dma(vcpu->kvm)) {
6872 		ipat = VMX_EPT_IPAT_BIT;
6873 		cache = MTRR_TYPE_WRBACK;
6874 		goto exit;
6875 	}
6876 
6877 	if (kvm_read_cr0(vcpu) & X86_CR0_CD) {
6878 		ipat = VMX_EPT_IPAT_BIT;
6879 		if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
6880 			cache = MTRR_TYPE_WRBACK;
6881 		else
6882 			cache = MTRR_TYPE_UNCACHABLE;
6883 		goto exit;
6884 	}
6885 
6886 	cache = kvm_mtrr_get_guest_memory_type(vcpu, gfn);
6887 
6888 exit:
6889 	return (cache << VMX_EPT_MT_EPTE_SHIFT) | ipat;
6890 }
6891 
6892 static int vmx_get_lpage_level(void)
6893 {
6894 	if (enable_ept && !cpu_has_vmx_ept_1g_page())
6895 		return PT_DIRECTORY_LEVEL;
6896 	else
6897 		/* For shadow and EPT supported 1GB page */
6898 		return PT_PDPE_LEVEL;
6899 }
6900 
6901 static void vmcs_set_secondary_exec_control(u32 new_ctl)
6902 {
6903 	/*
6904 	 * These bits in the secondary execution controls field
6905 	 * are dynamic, the others are mostly based on the hypervisor
6906 	 * architecture and the guest's CPUID.  Do not touch the
6907 	 * dynamic bits.
6908 	 */
6909 	u32 mask =
6910 		SECONDARY_EXEC_SHADOW_VMCS |
6911 		SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
6912 		SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
6913 		SECONDARY_EXEC_DESC;
6914 
6915 	u32 cur_ctl = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
6916 
6917 	vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
6918 		     (new_ctl & ~mask) | (cur_ctl & mask));
6919 }
6920 
6921 /*
6922  * Generate MSR_IA32_VMX_CR{0,4}_FIXED1 according to CPUID. Only set bits
6923  * (indicating "allowed-1") if they are supported in the guest's CPUID.
6924  */
6925 static void nested_vmx_cr_fixed1_bits_update(struct kvm_vcpu *vcpu)
6926 {
6927 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6928 	struct kvm_cpuid_entry2 *entry;
6929 
6930 	vmx->nested.msrs.cr0_fixed1 = 0xffffffff;
6931 	vmx->nested.msrs.cr4_fixed1 = X86_CR4_PCE;
6932 
6933 #define cr4_fixed1_update(_cr4_mask, _reg, _cpuid_mask) do {		\
6934 	if (entry && (entry->_reg & (_cpuid_mask)))			\
6935 		vmx->nested.msrs.cr4_fixed1 |= (_cr4_mask);	\
6936 } while (0)
6937 
6938 	entry = kvm_find_cpuid_entry(vcpu, 0x1, 0);
6939 	cr4_fixed1_update(X86_CR4_VME,        edx, bit(X86_FEATURE_VME));
6940 	cr4_fixed1_update(X86_CR4_PVI,        edx, bit(X86_FEATURE_VME));
6941 	cr4_fixed1_update(X86_CR4_TSD,        edx, bit(X86_FEATURE_TSC));
6942 	cr4_fixed1_update(X86_CR4_DE,         edx, bit(X86_FEATURE_DE));
6943 	cr4_fixed1_update(X86_CR4_PSE,        edx, bit(X86_FEATURE_PSE));
6944 	cr4_fixed1_update(X86_CR4_PAE,        edx, bit(X86_FEATURE_PAE));
6945 	cr4_fixed1_update(X86_CR4_MCE,        edx, bit(X86_FEATURE_MCE));
6946 	cr4_fixed1_update(X86_CR4_PGE,        edx, bit(X86_FEATURE_PGE));
6947 	cr4_fixed1_update(X86_CR4_OSFXSR,     edx, bit(X86_FEATURE_FXSR));
6948 	cr4_fixed1_update(X86_CR4_OSXMMEXCPT, edx, bit(X86_FEATURE_XMM));
6949 	cr4_fixed1_update(X86_CR4_VMXE,       ecx, bit(X86_FEATURE_VMX));
6950 	cr4_fixed1_update(X86_CR4_SMXE,       ecx, bit(X86_FEATURE_SMX));
6951 	cr4_fixed1_update(X86_CR4_PCIDE,      ecx, bit(X86_FEATURE_PCID));
6952 	cr4_fixed1_update(X86_CR4_OSXSAVE,    ecx, bit(X86_FEATURE_XSAVE));
6953 
6954 	entry = kvm_find_cpuid_entry(vcpu, 0x7, 0);
6955 	cr4_fixed1_update(X86_CR4_FSGSBASE,   ebx, bit(X86_FEATURE_FSGSBASE));
6956 	cr4_fixed1_update(X86_CR4_SMEP,       ebx, bit(X86_FEATURE_SMEP));
6957 	cr4_fixed1_update(X86_CR4_SMAP,       ebx, bit(X86_FEATURE_SMAP));
6958 	cr4_fixed1_update(X86_CR4_PKE,        ecx, bit(X86_FEATURE_PKU));
6959 	cr4_fixed1_update(X86_CR4_UMIP,       ecx, bit(X86_FEATURE_UMIP));
6960 
6961 #undef cr4_fixed1_update
6962 }
6963 
6964 static void nested_vmx_entry_exit_ctls_update(struct kvm_vcpu *vcpu)
6965 {
6966 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6967 
6968 	if (kvm_mpx_supported()) {
6969 		bool mpx_enabled = guest_cpuid_has(vcpu, X86_FEATURE_MPX);
6970 
6971 		if (mpx_enabled) {
6972 			vmx->nested.msrs.entry_ctls_high |= VM_ENTRY_LOAD_BNDCFGS;
6973 			vmx->nested.msrs.exit_ctls_high |= VM_EXIT_CLEAR_BNDCFGS;
6974 		} else {
6975 			vmx->nested.msrs.entry_ctls_high &= ~VM_ENTRY_LOAD_BNDCFGS;
6976 			vmx->nested.msrs.exit_ctls_high &= ~VM_EXIT_CLEAR_BNDCFGS;
6977 		}
6978 	}
6979 }
6980 
6981 static void update_intel_pt_cfg(struct kvm_vcpu *vcpu)
6982 {
6983 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6984 	struct kvm_cpuid_entry2 *best = NULL;
6985 	int i;
6986 
6987 	for (i = 0; i < PT_CPUID_LEAVES; i++) {
6988 		best = kvm_find_cpuid_entry(vcpu, 0x14, i);
6989 		if (!best)
6990 			return;
6991 		vmx->pt_desc.caps[CPUID_EAX + i*PT_CPUID_REGS_NUM] = best->eax;
6992 		vmx->pt_desc.caps[CPUID_EBX + i*PT_CPUID_REGS_NUM] = best->ebx;
6993 		vmx->pt_desc.caps[CPUID_ECX + i*PT_CPUID_REGS_NUM] = best->ecx;
6994 		vmx->pt_desc.caps[CPUID_EDX + i*PT_CPUID_REGS_NUM] = best->edx;
6995 	}
6996 
6997 	/* Get the number of configurable Address Ranges for filtering */
6998 	vmx->pt_desc.addr_range = intel_pt_validate_cap(vmx->pt_desc.caps,
6999 						PT_CAP_num_address_ranges);
7000 
7001 	/* Initialize and clear the no dependency bits */
7002 	vmx->pt_desc.ctl_bitmask = ~(RTIT_CTL_TRACEEN | RTIT_CTL_OS |
7003 			RTIT_CTL_USR | RTIT_CTL_TSC_EN | RTIT_CTL_DISRETC);
7004 
7005 	/*
7006 	 * If CPUID.(EAX=14H,ECX=0):EBX[0]=1 CR3Filter can be set otherwise
7007 	 * will inject an #GP
7008 	 */
7009 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_cr3_filtering))
7010 		vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_CR3EN;
7011 
7012 	/*
7013 	 * If CPUID.(EAX=14H,ECX=0):EBX[1]=1 CYCEn, CycThresh and
7014 	 * PSBFreq can be set
7015 	 */
7016 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc))
7017 		vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_CYCLEACC |
7018 				RTIT_CTL_CYC_THRESH | RTIT_CTL_PSB_FREQ);
7019 
7020 	/*
7021 	 * If CPUID.(EAX=14H,ECX=0):EBX[3]=1 MTCEn BranchEn and
7022 	 * MTCFreq can be set
7023 	 */
7024 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc))
7025 		vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_MTC_EN |
7026 				RTIT_CTL_BRANCH_EN | RTIT_CTL_MTC_RANGE);
7027 
7028 	/* If CPUID.(EAX=14H,ECX=0):EBX[4]=1 FUPonPTW and PTWEn can be set */
7029 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_ptwrite))
7030 		vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_FUP_ON_PTW |
7031 							RTIT_CTL_PTW_EN);
7032 
7033 	/* If CPUID.(EAX=14H,ECX=0):EBX[5]=1 PwrEvEn can be set */
7034 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_power_event_trace))
7035 		vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_PWR_EVT_EN;
7036 
7037 	/* If CPUID.(EAX=14H,ECX=0):ECX[0]=1 ToPA can be set */
7038 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_topa_output))
7039 		vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_TOPA;
7040 
7041 	/* If CPUID.(EAX=14H,ECX=0):ECX[3]=1 FabircEn can be set */
7042 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_output_subsys))
7043 		vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_FABRIC_EN;
7044 
7045 	/* unmask address range configure area */
7046 	for (i = 0; i < vmx->pt_desc.addr_range; i++)
7047 		vmx->pt_desc.ctl_bitmask &= ~(0xf << (32 + i * 4));
7048 }
7049 
7050 static void vmx_cpuid_update(struct kvm_vcpu *vcpu)
7051 {
7052 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7053 
7054 	if (cpu_has_secondary_exec_ctrls()) {
7055 		vmx_compute_secondary_exec_control(vmx);
7056 		vmcs_set_secondary_exec_control(vmx->secondary_exec_control);
7057 	}
7058 
7059 	if (nested_vmx_allowed(vcpu))
7060 		to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
7061 			FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
7062 	else
7063 		to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
7064 			~FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
7065 
7066 	if (nested_vmx_allowed(vcpu)) {
7067 		nested_vmx_cr_fixed1_bits_update(vcpu);
7068 		nested_vmx_entry_exit_ctls_update(vcpu);
7069 	}
7070 
7071 	if (boot_cpu_has(X86_FEATURE_INTEL_PT) &&
7072 			guest_cpuid_has(vcpu, X86_FEATURE_INTEL_PT))
7073 		update_intel_pt_cfg(vcpu);
7074 }
7075 
7076 static void vmx_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
7077 {
7078 	if (func == 1 && nested)
7079 		entry->ecx |= bit(X86_FEATURE_VMX);
7080 }
7081 
7082 static void vmx_request_immediate_exit(struct kvm_vcpu *vcpu)
7083 {
7084 	to_vmx(vcpu)->req_immediate_exit = true;
7085 }
7086 
7087 static int vmx_check_intercept(struct kvm_vcpu *vcpu,
7088 			       struct x86_instruction_info *info,
7089 			       enum x86_intercept_stage stage)
7090 {
7091 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
7092 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
7093 
7094 	/*
7095 	 * RDPID causes #UD if disabled through secondary execution controls.
7096 	 * Because it is marked as EmulateOnUD, we need to intercept it here.
7097 	 */
7098 	if (info->intercept == x86_intercept_rdtscp &&
7099 	    !nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDTSCP)) {
7100 		ctxt->exception.vector = UD_VECTOR;
7101 		ctxt->exception.error_code_valid = false;
7102 		return X86EMUL_PROPAGATE_FAULT;
7103 	}
7104 
7105 	/* TODO: check more intercepts... */
7106 	return X86EMUL_CONTINUE;
7107 }
7108 
7109 #ifdef CONFIG_X86_64
7110 /* (a << shift) / divisor, return 1 if overflow otherwise 0 */
7111 static inline int u64_shl_div_u64(u64 a, unsigned int shift,
7112 				  u64 divisor, u64 *result)
7113 {
7114 	u64 low = a << shift, high = a >> (64 - shift);
7115 
7116 	/* To avoid the overflow on divq */
7117 	if (high >= divisor)
7118 		return 1;
7119 
7120 	/* Low hold the result, high hold rem which is discarded */
7121 	asm("divq %2\n\t" : "=a" (low), "=d" (high) :
7122 	    "rm" (divisor), "0" (low), "1" (high));
7123 	*result = low;
7124 
7125 	return 0;
7126 }
7127 
7128 static int vmx_set_hv_timer(struct kvm_vcpu *vcpu, u64 guest_deadline_tsc)
7129 {
7130 	struct vcpu_vmx *vmx;
7131 	u64 tscl, guest_tscl, delta_tsc, lapic_timer_advance_cycles;
7132 
7133 	if (kvm_mwait_in_guest(vcpu->kvm))
7134 		return -EOPNOTSUPP;
7135 
7136 	vmx = to_vmx(vcpu);
7137 	tscl = rdtsc();
7138 	guest_tscl = kvm_read_l1_tsc(vcpu, tscl);
7139 	delta_tsc = max(guest_deadline_tsc, guest_tscl) - guest_tscl;
7140 	lapic_timer_advance_cycles = nsec_to_cycles(vcpu, lapic_timer_advance_ns);
7141 
7142 	if (delta_tsc > lapic_timer_advance_cycles)
7143 		delta_tsc -= lapic_timer_advance_cycles;
7144 	else
7145 		delta_tsc = 0;
7146 
7147 	/* Convert to host delta tsc if tsc scaling is enabled */
7148 	if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio &&
7149 			u64_shl_div_u64(delta_tsc,
7150 				kvm_tsc_scaling_ratio_frac_bits,
7151 				vcpu->arch.tsc_scaling_ratio,
7152 				&delta_tsc))
7153 		return -ERANGE;
7154 
7155 	/*
7156 	 * If the delta tsc can't fit in the 32 bit after the multi shift,
7157 	 * we can't use the preemption timer.
7158 	 * It's possible that it fits on later vmentries, but checking
7159 	 * on every vmentry is costly so we just use an hrtimer.
7160 	 */
7161 	if (delta_tsc >> (cpu_preemption_timer_multi + 32))
7162 		return -ERANGE;
7163 
7164 	vmx->hv_deadline_tsc = tscl + delta_tsc;
7165 	return delta_tsc == 0;
7166 }
7167 
7168 static void vmx_cancel_hv_timer(struct kvm_vcpu *vcpu)
7169 {
7170 	to_vmx(vcpu)->hv_deadline_tsc = -1;
7171 }
7172 #endif
7173 
7174 static void vmx_sched_in(struct kvm_vcpu *vcpu, int cpu)
7175 {
7176 	if (!kvm_pause_in_guest(vcpu->kvm))
7177 		shrink_ple_window(vcpu);
7178 }
7179 
7180 static void vmx_slot_enable_log_dirty(struct kvm *kvm,
7181 				     struct kvm_memory_slot *slot)
7182 {
7183 	kvm_mmu_slot_leaf_clear_dirty(kvm, slot);
7184 	kvm_mmu_slot_largepage_remove_write_access(kvm, slot);
7185 }
7186 
7187 static void vmx_slot_disable_log_dirty(struct kvm *kvm,
7188 				       struct kvm_memory_slot *slot)
7189 {
7190 	kvm_mmu_slot_set_dirty(kvm, slot);
7191 }
7192 
7193 static void vmx_flush_log_dirty(struct kvm *kvm)
7194 {
7195 	kvm_flush_pml_buffers(kvm);
7196 }
7197 
7198 static int vmx_write_pml_buffer(struct kvm_vcpu *vcpu)
7199 {
7200 	struct vmcs12 *vmcs12;
7201 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7202 	gpa_t gpa;
7203 	struct page *page = NULL;
7204 	u64 *pml_address;
7205 
7206 	if (is_guest_mode(vcpu)) {
7207 		WARN_ON_ONCE(vmx->nested.pml_full);
7208 
7209 		/*
7210 		 * Check if PML is enabled for the nested guest.
7211 		 * Whether eptp bit 6 is set is already checked
7212 		 * as part of A/D emulation.
7213 		 */
7214 		vmcs12 = get_vmcs12(vcpu);
7215 		if (!nested_cpu_has_pml(vmcs12))
7216 			return 0;
7217 
7218 		if (vmcs12->guest_pml_index >= PML_ENTITY_NUM) {
7219 			vmx->nested.pml_full = true;
7220 			return 1;
7221 		}
7222 
7223 		gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS) & ~0xFFFull;
7224 
7225 		page = kvm_vcpu_gpa_to_page(vcpu, vmcs12->pml_address);
7226 		if (is_error_page(page))
7227 			return 0;
7228 
7229 		pml_address = kmap(page);
7230 		pml_address[vmcs12->guest_pml_index--] = gpa;
7231 		kunmap(page);
7232 		kvm_release_page_clean(page);
7233 	}
7234 
7235 	return 0;
7236 }
7237 
7238 static void vmx_enable_log_dirty_pt_masked(struct kvm *kvm,
7239 					   struct kvm_memory_slot *memslot,
7240 					   gfn_t offset, unsigned long mask)
7241 {
7242 	kvm_mmu_clear_dirty_pt_masked(kvm, memslot, offset, mask);
7243 }
7244 
7245 static void __pi_post_block(struct kvm_vcpu *vcpu)
7246 {
7247 	struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
7248 	struct pi_desc old, new;
7249 	unsigned int dest;
7250 
7251 	do {
7252 		old.control = new.control = pi_desc->control;
7253 		WARN(old.nv != POSTED_INTR_WAKEUP_VECTOR,
7254 		     "Wakeup handler not enabled while the VCPU is blocked\n");
7255 
7256 		dest = cpu_physical_id(vcpu->cpu);
7257 
7258 		if (x2apic_enabled())
7259 			new.ndst = dest;
7260 		else
7261 			new.ndst = (dest << 8) & 0xFF00;
7262 
7263 		/* set 'NV' to 'notification vector' */
7264 		new.nv = POSTED_INTR_VECTOR;
7265 	} while (cmpxchg64(&pi_desc->control, old.control,
7266 			   new.control) != old.control);
7267 
7268 	if (!WARN_ON_ONCE(vcpu->pre_pcpu == -1)) {
7269 		spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7270 		list_del(&vcpu->blocked_vcpu_list);
7271 		spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7272 		vcpu->pre_pcpu = -1;
7273 	}
7274 }
7275 
7276 /*
7277  * This routine does the following things for vCPU which is going
7278  * to be blocked if VT-d PI is enabled.
7279  * - Store the vCPU to the wakeup list, so when interrupts happen
7280  *   we can find the right vCPU to wake up.
7281  * - Change the Posted-interrupt descriptor as below:
7282  *      'NDST' <-- vcpu->pre_pcpu
7283  *      'NV' <-- POSTED_INTR_WAKEUP_VECTOR
7284  * - If 'ON' is set during this process, which means at least one
7285  *   interrupt is posted for this vCPU, we cannot block it, in
7286  *   this case, return 1, otherwise, return 0.
7287  *
7288  */
7289 static int pi_pre_block(struct kvm_vcpu *vcpu)
7290 {
7291 	unsigned int dest;
7292 	struct pi_desc old, new;
7293 	struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
7294 
7295 	if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
7296 		!irq_remapping_cap(IRQ_POSTING_CAP)  ||
7297 		!kvm_vcpu_apicv_active(vcpu))
7298 		return 0;
7299 
7300 	WARN_ON(irqs_disabled());
7301 	local_irq_disable();
7302 	if (!WARN_ON_ONCE(vcpu->pre_pcpu != -1)) {
7303 		vcpu->pre_pcpu = vcpu->cpu;
7304 		spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7305 		list_add_tail(&vcpu->blocked_vcpu_list,
7306 			      &per_cpu(blocked_vcpu_on_cpu,
7307 				       vcpu->pre_pcpu));
7308 		spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7309 	}
7310 
7311 	do {
7312 		old.control = new.control = pi_desc->control;
7313 
7314 		WARN((pi_desc->sn == 1),
7315 		     "Warning: SN field of posted-interrupts "
7316 		     "is set before blocking\n");
7317 
7318 		/*
7319 		 * Since vCPU can be preempted during this process,
7320 		 * vcpu->cpu could be different with pre_pcpu, we
7321 		 * need to set pre_pcpu as the destination of wakeup
7322 		 * notification event, then we can find the right vCPU
7323 		 * to wakeup in wakeup handler if interrupts happen
7324 		 * when the vCPU is in blocked state.
7325 		 */
7326 		dest = cpu_physical_id(vcpu->pre_pcpu);
7327 
7328 		if (x2apic_enabled())
7329 			new.ndst = dest;
7330 		else
7331 			new.ndst = (dest << 8) & 0xFF00;
7332 
7333 		/* set 'NV' to 'wakeup vector' */
7334 		new.nv = POSTED_INTR_WAKEUP_VECTOR;
7335 	} while (cmpxchg64(&pi_desc->control, old.control,
7336 			   new.control) != old.control);
7337 
7338 	/* We should not block the vCPU if an interrupt is posted for it.  */
7339 	if (pi_test_on(pi_desc) == 1)
7340 		__pi_post_block(vcpu);
7341 
7342 	local_irq_enable();
7343 	return (vcpu->pre_pcpu == -1);
7344 }
7345 
7346 static int vmx_pre_block(struct kvm_vcpu *vcpu)
7347 {
7348 	if (pi_pre_block(vcpu))
7349 		return 1;
7350 
7351 	if (kvm_lapic_hv_timer_in_use(vcpu))
7352 		kvm_lapic_switch_to_sw_timer(vcpu);
7353 
7354 	return 0;
7355 }
7356 
7357 static void pi_post_block(struct kvm_vcpu *vcpu)
7358 {
7359 	if (vcpu->pre_pcpu == -1)
7360 		return;
7361 
7362 	WARN_ON(irqs_disabled());
7363 	local_irq_disable();
7364 	__pi_post_block(vcpu);
7365 	local_irq_enable();
7366 }
7367 
7368 static void vmx_post_block(struct kvm_vcpu *vcpu)
7369 {
7370 	if (kvm_x86_ops->set_hv_timer)
7371 		kvm_lapic_switch_to_hv_timer(vcpu);
7372 
7373 	pi_post_block(vcpu);
7374 }
7375 
7376 /*
7377  * vmx_update_pi_irte - set IRTE for Posted-Interrupts
7378  *
7379  * @kvm: kvm
7380  * @host_irq: host irq of the interrupt
7381  * @guest_irq: gsi of the interrupt
7382  * @set: set or unset PI
7383  * returns 0 on success, < 0 on failure
7384  */
7385 static int vmx_update_pi_irte(struct kvm *kvm, unsigned int host_irq,
7386 			      uint32_t guest_irq, bool set)
7387 {
7388 	struct kvm_kernel_irq_routing_entry *e;
7389 	struct kvm_irq_routing_table *irq_rt;
7390 	struct kvm_lapic_irq irq;
7391 	struct kvm_vcpu *vcpu;
7392 	struct vcpu_data vcpu_info;
7393 	int idx, ret = 0;
7394 
7395 	if (!kvm_arch_has_assigned_device(kvm) ||
7396 		!irq_remapping_cap(IRQ_POSTING_CAP) ||
7397 		!kvm_vcpu_apicv_active(kvm->vcpus[0]))
7398 		return 0;
7399 
7400 	idx = srcu_read_lock(&kvm->irq_srcu);
7401 	irq_rt = srcu_dereference(kvm->irq_routing, &kvm->irq_srcu);
7402 	if (guest_irq >= irq_rt->nr_rt_entries ||
7403 	    hlist_empty(&irq_rt->map[guest_irq])) {
7404 		pr_warn_once("no route for guest_irq %u/%u (broken user space?)\n",
7405 			     guest_irq, irq_rt->nr_rt_entries);
7406 		goto out;
7407 	}
7408 
7409 	hlist_for_each_entry(e, &irq_rt->map[guest_irq], link) {
7410 		if (e->type != KVM_IRQ_ROUTING_MSI)
7411 			continue;
7412 		/*
7413 		 * VT-d PI cannot support posting multicast/broadcast
7414 		 * interrupts to a vCPU, we still use interrupt remapping
7415 		 * for these kind of interrupts.
7416 		 *
7417 		 * For lowest-priority interrupts, we only support
7418 		 * those with single CPU as the destination, e.g. user
7419 		 * configures the interrupts via /proc/irq or uses
7420 		 * irqbalance to make the interrupts single-CPU.
7421 		 *
7422 		 * We will support full lowest-priority interrupt later.
7423 		 */
7424 
7425 		kvm_set_msi_irq(kvm, e, &irq);
7426 		if (!kvm_intr_is_single_vcpu(kvm, &irq, &vcpu)) {
7427 			/*
7428 			 * Make sure the IRTE is in remapped mode if
7429 			 * we don't handle it in posted mode.
7430 			 */
7431 			ret = irq_set_vcpu_affinity(host_irq, NULL);
7432 			if (ret < 0) {
7433 				printk(KERN_INFO
7434 				   "failed to back to remapped mode, irq: %u\n",
7435 				   host_irq);
7436 				goto out;
7437 			}
7438 
7439 			continue;
7440 		}
7441 
7442 		vcpu_info.pi_desc_addr = __pa(vcpu_to_pi_desc(vcpu));
7443 		vcpu_info.vector = irq.vector;
7444 
7445 		trace_kvm_pi_irte_update(host_irq, vcpu->vcpu_id, e->gsi,
7446 				vcpu_info.vector, vcpu_info.pi_desc_addr, set);
7447 
7448 		if (set)
7449 			ret = irq_set_vcpu_affinity(host_irq, &vcpu_info);
7450 		else
7451 			ret = irq_set_vcpu_affinity(host_irq, NULL);
7452 
7453 		if (ret < 0) {
7454 			printk(KERN_INFO "%s: failed to update PI IRTE\n",
7455 					__func__);
7456 			goto out;
7457 		}
7458 	}
7459 
7460 	ret = 0;
7461 out:
7462 	srcu_read_unlock(&kvm->irq_srcu, idx);
7463 	return ret;
7464 }
7465 
7466 static void vmx_setup_mce(struct kvm_vcpu *vcpu)
7467 {
7468 	if (vcpu->arch.mcg_cap & MCG_LMCE_P)
7469 		to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
7470 			FEATURE_CONTROL_LMCE;
7471 	else
7472 		to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
7473 			~FEATURE_CONTROL_LMCE;
7474 }
7475 
7476 static int vmx_smi_allowed(struct kvm_vcpu *vcpu)
7477 {
7478 	/* we need a nested vmexit to enter SMM, postpone if run is pending */
7479 	if (to_vmx(vcpu)->nested.nested_run_pending)
7480 		return 0;
7481 	return 1;
7482 }
7483 
7484 static int vmx_pre_enter_smm(struct kvm_vcpu *vcpu, char *smstate)
7485 {
7486 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7487 
7488 	vmx->nested.smm.guest_mode = is_guest_mode(vcpu);
7489 	if (vmx->nested.smm.guest_mode)
7490 		nested_vmx_vmexit(vcpu, -1, 0, 0);
7491 
7492 	vmx->nested.smm.vmxon = vmx->nested.vmxon;
7493 	vmx->nested.vmxon = false;
7494 	vmx_clear_hlt(vcpu);
7495 	return 0;
7496 }
7497 
7498 static int vmx_pre_leave_smm(struct kvm_vcpu *vcpu, u64 smbase)
7499 {
7500 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7501 	int ret;
7502 
7503 	if (vmx->nested.smm.vmxon) {
7504 		vmx->nested.vmxon = true;
7505 		vmx->nested.smm.vmxon = false;
7506 	}
7507 
7508 	if (vmx->nested.smm.guest_mode) {
7509 		vcpu->arch.hflags &= ~HF_SMM_MASK;
7510 		ret = nested_vmx_enter_non_root_mode(vcpu, false);
7511 		vcpu->arch.hflags |= HF_SMM_MASK;
7512 		if (ret)
7513 			return ret;
7514 
7515 		vmx->nested.smm.guest_mode = false;
7516 	}
7517 	return 0;
7518 }
7519 
7520 static int enable_smi_window(struct kvm_vcpu *vcpu)
7521 {
7522 	return 0;
7523 }
7524 
7525 static __init int hardware_setup(void)
7526 {
7527 	unsigned long host_bndcfgs;
7528 	int r, i;
7529 
7530 	rdmsrl_safe(MSR_EFER, &host_efer);
7531 
7532 	for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i)
7533 		kvm_define_shared_msr(i, vmx_msr_index[i]);
7534 
7535 	if (setup_vmcs_config(&vmcs_config, &vmx_capability) < 0)
7536 		return -EIO;
7537 
7538 	if (boot_cpu_has(X86_FEATURE_NX))
7539 		kvm_enable_efer_bits(EFER_NX);
7540 
7541 	if (boot_cpu_has(X86_FEATURE_MPX)) {
7542 		rdmsrl(MSR_IA32_BNDCFGS, host_bndcfgs);
7543 		WARN_ONCE(host_bndcfgs, "KVM: BNDCFGS in host will be lost");
7544 	}
7545 
7546 	if (boot_cpu_has(X86_FEATURE_XSAVES))
7547 		rdmsrl(MSR_IA32_XSS, host_xss);
7548 
7549 	if (!cpu_has_vmx_vpid() || !cpu_has_vmx_invvpid() ||
7550 	    !(cpu_has_vmx_invvpid_single() || cpu_has_vmx_invvpid_global()))
7551 		enable_vpid = 0;
7552 
7553 	if (!cpu_has_vmx_ept() ||
7554 	    !cpu_has_vmx_ept_4levels() ||
7555 	    !cpu_has_vmx_ept_mt_wb() ||
7556 	    !cpu_has_vmx_invept_global())
7557 		enable_ept = 0;
7558 
7559 	if (!cpu_has_vmx_ept_ad_bits() || !enable_ept)
7560 		enable_ept_ad_bits = 0;
7561 
7562 	if (!cpu_has_vmx_unrestricted_guest() || !enable_ept)
7563 		enable_unrestricted_guest = 0;
7564 
7565 	if (!cpu_has_vmx_flexpriority())
7566 		flexpriority_enabled = 0;
7567 
7568 	if (!cpu_has_virtual_nmis())
7569 		enable_vnmi = 0;
7570 
7571 	/*
7572 	 * set_apic_access_page_addr() is used to reload apic access
7573 	 * page upon invalidation.  No need to do anything if not
7574 	 * using the APIC_ACCESS_ADDR VMCS field.
7575 	 */
7576 	if (!flexpriority_enabled)
7577 		kvm_x86_ops->set_apic_access_page_addr = NULL;
7578 
7579 	if (!cpu_has_vmx_tpr_shadow())
7580 		kvm_x86_ops->update_cr8_intercept = NULL;
7581 
7582 	if (enable_ept && !cpu_has_vmx_ept_2m_page())
7583 		kvm_disable_largepages();
7584 
7585 #if IS_ENABLED(CONFIG_HYPERV)
7586 	if (ms_hyperv.nested_features & HV_X64_NESTED_GUEST_MAPPING_FLUSH
7587 	    && enable_ept) {
7588 		kvm_x86_ops->tlb_remote_flush = hv_remote_flush_tlb;
7589 		kvm_x86_ops->tlb_remote_flush_with_range =
7590 				hv_remote_flush_tlb_with_range;
7591 	}
7592 #endif
7593 
7594 	if (!cpu_has_vmx_ple()) {
7595 		ple_gap = 0;
7596 		ple_window = 0;
7597 		ple_window_grow = 0;
7598 		ple_window_max = 0;
7599 		ple_window_shrink = 0;
7600 	}
7601 
7602 	if (!cpu_has_vmx_apicv()) {
7603 		enable_apicv = 0;
7604 		kvm_x86_ops->sync_pir_to_irr = NULL;
7605 	}
7606 
7607 	if (cpu_has_vmx_tsc_scaling()) {
7608 		kvm_has_tsc_control = true;
7609 		kvm_max_tsc_scaling_ratio = KVM_VMX_TSC_MULTIPLIER_MAX;
7610 		kvm_tsc_scaling_ratio_frac_bits = 48;
7611 	}
7612 
7613 	set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
7614 
7615 	if (enable_ept)
7616 		vmx_enable_tdp();
7617 	else
7618 		kvm_disable_tdp();
7619 
7620 	/*
7621 	 * Only enable PML when hardware supports PML feature, and both EPT
7622 	 * and EPT A/D bit features are enabled -- PML depends on them to work.
7623 	 */
7624 	if (!enable_ept || !enable_ept_ad_bits || !cpu_has_vmx_pml())
7625 		enable_pml = 0;
7626 
7627 	if (!enable_pml) {
7628 		kvm_x86_ops->slot_enable_log_dirty = NULL;
7629 		kvm_x86_ops->slot_disable_log_dirty = NULL;
7630 		kvm_x86_ops->flush_log_dirty = NULL;
7631 		kvm_x86_ops->enable_log_dirty_pt_masked = NULL;
7632 	}
7633 
7634 	if (!cpu_has_vmx_preemption_timer())
7635 		kvm_x86_ops->request_immediate_exit = __kvm_request_immediate_exit;
7636 
7637 	if (cpu_has_vmx_preemption_timer() && enable_preemption_timer) {
7638 		u64 vmx_msr;
7639 
7640 		rdmsrl(MSR_IA32_VMX_MISC, vmx_msr);
7641 		cpu_preemption_timer_multi =
7642 			vmx_msr & VMX_MISC_PREEMPTION_TIMER_RATE_MASK;
7643 	} else {
7644 		kvm_x86_ops->set_hv_timer = NULL;
7645 		kvm_x86_ops->cancel_hv_timer = NULL;
7646 	}
7647 
7648 	kvm_set_posted_intr_wakeup_handler(wakeup_handler);
7649 
7650 	kvm_mce_cap_supported |= MCG_LMCE_P;
7651 
7652 	if (pt_mode != PT_MODE_SYSTEM && pt_mode != PT_MODE_HOST_GUEST)
7653 		return -EINVAL;
7654 	if (!enable_ept || !cpu_has_vmx_intel_pt())
7655 		pt_mode = PT_MODE_SYSTEM;
7656 
7657 	if (nested) {
7658 		nested_vmx_setup_ctls_msrs(&vmcs_config.nested,
7659 					   vmx_capability.ept, enable_apicv);
7660 
7661 		r = nested_vmx_hardware_setup(kvm_vmx_exit_handlers);
7662 		if (r)
7663 			return r;
7664 	}
7665 
7666 	r = alloc_kvm_area();
7667 	if (r)
7668 		nested_vmx_hardware_unsetup();
7669 	return r;
7670 }
7671 
7672 static __exit void hardware_unsetup(void)
7673 {
7674 	if (nested)
7675 		nested_vmx_hardware_unsetup();
7676 
7677 	free_kvm_area();
7678 }
7679 
7680 static struct kvm_x86_ops vmx_x86_ops __ro_after_init = {
7681 	.cpu_has_kvm_support = cpu_has_kvm_support,
7682 	.disabled_by_bios = vmx_disabled_by_bios,
7683 	.hardware_setup = hardware_setup,
7684 	.hardware_unsetup = hardware_unsetup,
7685 	.check_processor_compatibility = vmx_check_processor_compat,
7686 	.hardware_enable = hardware_enable,
7687 	.hardware_disable = hardware_disable,
7688 	.cpu_has_accelerated_tpr = report_flexpriority,
7689 	.has_emulated_msr = vmx_has_emulated_msr,
7690 
7691 	.vm_init = vmx_vm_init,
7692 	.vm_alloc = vmx_vm_alloc,
7693 	.vm_free = vmx_vm_free,
7694 
7695 	.vcpu_create = vmx_create_vcpu,
7696 	.vcpu_free = vmx_free_vcpu,
7697 	.vcpu_reset = vmx_vcpu_reset,
7698 
7699 	.prepare_guest_switch = vmx_prepare_switch_to_guest,
7700 	.vcpu_load = vmx_vcpu_load,
7701 	.vcpu_put = vmx_vcpu_put,
7702 
7703 	.update_bp_intercept = update_exception_bitmap,
7704 	.get_msr_feature = vmx_get_msr_feature,
7705 	.get_msr = vmx_get_msr,
7706 	.set_msr = vmx_set_msr,
7707 	.get_segment_base = vmx_get_segment_base,
7708 	.get_segment = vmx_get_segment,
7709 	.set_segment = vmx_set_segment,
7710 	.get_cpl = vmx_get_cpl,
7711 	.get_cs_db_l_bits = vmx_get_cs_db_l_bits,
7712 	.decache_cr0_guest_bits = vmx_decache_cr0_guest_bits,
7713 	.decache_cr3 = vmx_decache_cr3,
7714 	.decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
7715 	.set_cr0 = vmx_set_cr0,
7716 	.set_cr3 = vmx_set_cr3,
7717 	.set_cr4 = vmx_set_cr4,
7718 	.set_efer = vmx_set_efer,
7719 	.get_idt = vmx_get_idt,
7720 	.set_idt = vmx_set_idt,
7721 	.get_gdt = vmx_get_gdt,
7722 	.set_gdt = vmx_set_gdt,
7723 	.get_dr6 = vmx_get_dr6,
7724 	.set_dr6 = vmx_set_dr6,
7725 	.set_dr7 = vmx_set_dr7,
7726 	.sync_dirty_debug_regs = vmx_sync_dirty_debug_regs,
7727 	.cache_reg = vmx_cache_reg,
7728 	.get_rflags = vmx_get_rflags,
7729 	.set_rflags = vmx_set_rflags,
7730 
7731 	.tlb_flush = vmx_flush_tlb,
7732 	.tlb_flush_gva = vmx_flush_tlb_gva,
7733 
7734 	.run = vmx_vcpu_run,
7735 	.handle_exit = vmx_handle_exit,
7736 	.skip_emulated_instruction = skip_emulated_instruction,
7737 	.set_interrupt_shadow = vmx_set_interrupt_shadow,
7738 	.get_interrupt_shadow = vmx_get_interrupt_shadow,
7739 	.patch_hypercall = vmx_patch_hypercall,
7740 	.set_irq = vmx_inject_irq,
7741 	.set_nmi = vmx_inject_nmi,
7742 	.queue_exception = vmx_queue_exception,
7743 	.cancel_injection = vmx_cancel_injection,
7744 	.interrupt_allowed = vmx_interrupt_allowed,
7745 	.nmi_allowed = vmx_nmi_allowed,
7746 	.get_nmi_mask = vmx_get_nmi_mask,
7747 	.set_nmi_mask = vmx_set_nmi_mask,
7748 	.enable_nmi_window = enable_nmi_window,
7749 	.enable_irq_window = enable_irq_window,
7750 	.update_cr8_intercept = update_cr8_intercept,
7751 	.set_virtual_apic_mode = vmx_set_virtual_apic_mode,
7752 	.set_apic_access_page_addr = vmx_set_apic_access_page_addr,
7753 	.get_enable_apicv = vmx_get_enable_apicv,
7754 	.refresh_apicv_exec_ctrl = vmx_refresh_apicv_exec_ctrl,
7755 	.load_eoi_exitmap = vmx_load_eoi_exitmap,
7756 	.apicv_post_state_restore = vmx_apicv_post_state_restore,
7757 	.hwapic_irr_update = vmx_hwapic_irr_update,
7758 	.hwapic_isr_update = vmx_hwapic_isr_update,
7759 	.guest_apic_has_interrupt = vmx_guest_apic_has_interrupt,
7760 	.sync_pir_to_irr = vmx_sync_pir_to_irr,
7761 	.deliver_posted_interrupt = vmx_deliver_posted_interrupt,
7762 
7763 	.set_tss_addr = vmx_set_tss_addr,
7764 	.set_identity_map_addr = vmx_set_identity_map_addr,
7765 	.get_tdp_level = get_ept_level,
7766 	.get_mt_mask = vmx_get_mt_mask,
7767 
7768 	.get_exit_info = vmx_get_exit_info,
7769 
7770 	.get_lpage_level = vmx_get_lpage_level,
7771 
7772 	.cpuid_update = vmx_cpuid_update,
7773 
7774 	.rdtscp_supported = vmx_rdtscp_supported,
7775 	.invpcid_supported = vmx_invpcid_supported,
7776 
7777 	.set_supported_cpuid = vmx_set_supported_cpuid,
7778 
7779 	.has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
7780 
7781 	.read_l1_tsc_offset = vmx_read_l1_tsc_offset,
7782 	.write_l1_tsc_offset = vmx_write_l1_tsc_offset,
7783 
7784 	.set_tdp_cr3 = vmx_set_cr3,
7785 
7786 	.check_intercept = vmx_check_intercept,
7787 	.handle_external_intr = vmx_handle_external_intr,
7788 	.mpx_supported = vmx_mpx_supported,
7789 	.xsaves_supported = vmx_xsaves_supported,
7790 	.umip_emulated = vmx_umip_emulated,
7791 	.pt_supported = vmx_pt_supported,
7792 
7793 	.request_immediate_exit = vmx_request_immediate_exit,
7794 
7795 	.sched_in = vmx_sched_in,
7796 
7797 	.slot_enable_log_dirty = vmx_slot_enable_log_dirty,
7798 	.slot_disable_log_dirty = vmx_slot_disable_log_dirty,
7799 	.flush_log_dirty = vmx_flush_log_dirty,
7800 	.enable_log_dirty_pt_masked = vmx_enable_log_dirty_pt_masked,
7801 	.write_log_dirty = vmx_write_pml_buffer,
7802 
7803 	.pre_block = vmx_pre_block,
7804 	.post_block = vmx_post_block,
7805 
7806 	.pmu_ops = &intel_pmu_ops,
7807 
7808 	.update_pi_irte = vmx_update_pi_irte,
7809 
7810 #ifdef CONFIG_X86_64
7811 	.set_hv_timer = vmx_set_hv_timer,
7812 	.cancel_hv_timer = vmx_cancel_hv_timer,
7813 #endif
7814 
7815 	.setup_mce = vmx_setup_mce,
7816 
7817 	.smi_allowed = vmx_smi_allowed,
7818 	.pre_enter_smm = vmx_pre_enter_smm,
7819 	.pre_leave_smm = vmx_pre_leave_smm,
7820 	.enable_smi_window = enable_smi_window,
7821 
7822 	.check_nested_events = NULL,
7823 	.get_nested_state = NULL,
7824 	.set_nested_state = NULL,
7825 	.get_vmcs12_pages = NULL,
7826 	.nested_enable_evmcs = NULL,
7827 };
7828 
7829 static void vmx_cleanup_l1d_flush(void)
7830 {
7831 	if (vmx_l1d_flush_pages) {
7832 		free_pages((unsigned long)vmx_l1d_flush_pages, L1D_CACHE_ORDER);
7833 		vmx_l1d_flush_pages = NULL;
7834 	}
7835 	/* Restore state so sysfs ignores VMX */
7836 	l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_AUTO;
7837 }
7838 
7839 static void vmx_exit(void)
7840 {
7841 #ifdef CONFIG_KEXEC_CORE
7842 	RCU_INIT_POINTER(crash_vmclear_loaded_vmcss, NULL);
7843 	synchronize_rcu();
7844 #endif
7845 
7846 	kvm_exit();
7847 
7848 #if IS_ENABLED(CONFIG_HYPERV)
7849 	if (static_branch_unlikely(&enable_evmcs)) {
7850 		int cpu;
7851 		struct hv_vp_assist_page *vp_ap;
7852 		/*
7853 		 * Reset everything to support using non-enlightened VMCS
7854 		 * access later (e.g. when we reload the module with
7855 		 * enlightened_vmcs=0)
7856 		 */
7857 		for_each_online_cpu(cpu) {
7858 			vp_ap =	hv_get_vp_assist_page(cpu);
7859 
7860 			if (!vp_ap)
7861 				continue;
7862 
7863 			vp_ap->current_nested_vmcs = 0;
7864 			vp_ap->enlighten_vmentry = 0;
7865 		}
7866 
7867 		static_branch_disable(&enable_evmcs);
7868 	}
7869 #endif
7870 	vmx_cleanup_l1d_flush();
7871 }
7872 module_exit(vmx_exit);
7873 
7874 static int __init vmx_init(void)
7875 {
7876 	int r;
7877 
7878 #if IS_ENABLED(CONFIG_HYPERV)
7879 	/*
7880 	 * Enlightened VMCS usage should be recommended and the host needs
7881 	 * to support eVMCS v1 or above. We can also disable eVMCS support
7882 	 * with module parameter.
7883 	 */
7884 	if (enlightened_vmcs &&
7885 	    ms_hyperv.hints & HV_X64_ENLIGHTENED_VMCS_RECOMMENDED &&
7886 	    (ms_hyperv.nested_features & HV_X64_ENLIGHTENED_VMCS_VERSION) >=
7887 	    KVM_EVMCS_VERSION) {
7888 		int cpu;
7889 
7890 		/* Check that we have assist pages on all online CPUs */
7891 		for_each_online_cpu(cpu) {
7892 			if (!hv_get_vp_assist_page(cpu)) {
7893 				enlightened_vmcs = false;
7894 				break;
7895 			}
7896 		}
7897 
7898 		if (enlightened_vmcs) {
7899 			pr_info("KVM: vmx: using Hyper-V Enlightened VMCS\n");
7900 			static_branch_enable(&enable_evmcs);
7901 		}
7902 	} else {
7903 		enlightened_vmcs = false;
7904 	}
7905 #endif
7906 
7907 	r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx),
7908 		     __alignof__(struct vcpu_vmx), THIS_MODULE);
7909 	if (r)
7910 		return r;
7911 
7912 	/*
7913 	 * Must be called after kvm_init() so enable_ept is properly set
7914 	 * up. Hand the parameter mitigation value in which was stored in
7915 	 * the pre module init parser. If no parameter was given, it will
7916 	 * contain 'auto' which will be turned into the default 'cond'
7917 	 * mitigation mode.
7918 	 */
7919 	if (boot_cpu_has(X86_BUG_L1TF)) {
7920 		r = vmx_setup_l1d_flush(vmentry_l1d_flush_param);
7921 		if (r) {
7922 			vmx_exit();
7923 			return r;
7924 		}
7925 	}
7926 
7927 #ifdef CONFIG_KEXEC_CORE
7928 	rcu_assign_pointer(crash_vmclear_loaded_vmcss,
7929 			   crash_vmclear_local_loaded_vmcss);
7930 #endif
7931 	vmx_check_vmcs12_offsets();
7932 
7933 	return 0;
7934 }
7935 module_init(vmx_init);
7936