xref: /openbmc/linux/arch/x86/kvm/vmx/vmx.c (revision c6fddb28)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Kernel-based Virtual Machine driver for Linux
4  *
5  * This module enables machines with Intel VT-x extensions to run virtual
6  * machines without emulation or binary translation.
7  *
8  * Copyright (C) 2006 Qumranet, Inc.
9  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
10  *
11  * Authors:
12  *   Avi Kivity   <avi@qumranet.com>
13  *   Yaniv Kamay  <yaniv@qumranet.com>
14  */
15 
16 #include <linux/frame.h>
17 #include <linux/highmem.h>
18 #include <linux/hrtimer.h>
19 #include <linux/kernel.h>
20 #include <linux/kvm_host.h>
21 #include <linux/module.h>
22 #include <linux/moduleparam.h>
23 #include <linux/mod_devicetable.h>
24 #include <linux/mm.h>
25 #include <linux/sched.h>
26 #include <linux/sched/smt.h>
27 #include <linux/slab.h>
28 #include <linux/tboot.h>
29 #include <linux/trace_events.h>
30 
31 #include <asm/apic.h>
32 #include <asm/asm.h>
33 #include <asm/cpu.h>
34 #include <asm/cpu_device_id.h>
35 #include <asm/debugreg.h>
36 #include <asm/desc.h>
37 #include <asm/fpu/internal.h>
38 #include <asm/io.h>
39 #include <asm/irq_remapping.h>
40 #include <asm/kexec.h>
41 #include <asm/perf_event.h>
42 #include <asm/mce.h>
43 #include <asm/mmu_context.h>
44 #include <asm/mshyperv.h>
45 #include <asm/mwait.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 #ifdef MODULE
70 static const struct x86_cpu_id vmx_cpu_id[] = {
71 	X86_MATCH_FEATURE(X86_FEATURE_VMX, NULL),
72 	{}
73 };
74 MODULE_DEVICE_TABLE(x86cpu, vmx_cpu_id);
75 #endif
76 
77 bool __read_mostly enable_vpid = 1;
78 module_param_named(vpid, enable_vpid, bool, 0444);
79 
80 static bool __read_mostly enable_vnmi = 1;
81 module_param_named(vnmi, enable_vnmi, bool, S_IRUGO);
82 
83 bool __read_mostly flexpriority_enabled = 1;
84 module_param_named(flexpriority, flexpriority_enabled, bool, S_IRUGO);
85 
86 bool __read_mostly enable_ept = 1;
87 module_param_named(ept, enable_ept, bool, S_IRUGO);
88 
89 bool __read_mostly enable_unrestricted_guest = 1;
90 module_param_named(unrestricted_guest,
91 			enable_unrestricted_guest, bool, S_IRUGO);
92 
93 bool __read_mostly enable_ept_ad_bits = 1;
94 module_param_named(eptad, enable_ept_ad_bits, bool, S_IRUGO);
95 
96 static bool __read_mostly emulate_invalid_guest_state = true;
97 module_param(emulate_invalid_guest_state, bool, S_IRUGO);
98 
99 static bool __read_mostly fasteoi = 1;
100 module_param(fasteoi, bool, S_IRUGO);
101 
102 bool __read_mostly enable_apicv = 1;
103 module_param(enable_apicv, bool, S_IRUGO);
104 
105 /*
106  * If nested=1, nested virtualization is supported, i.e., guests may use
107  * VMX and be a hypervisor for its own guests. If nested=0, guests may not
108  * use VMX instructions.
109  */
110 static bool __read_mostly nested = 1;
111 module_param(nested, bool, S_IRUGO);
112 
113 bool __read_mostly enable_pml = 1;
114 module_param_named(pml, enable_pml, bool, S_IRUGO);
115 
116 static bool __read_mostly dump_invalid_vmcs = 0;
117 module_param(dump_invalid_vmcs, bool, 0644);
118 
119 #define MSR_BITMAP_MODE_X2APIC		1
120 #define MSR_BITMAP_MODE_X2APIC_APICV	2
121 
122 #define KVM_VMX_TSC_MULTIPLIER_MAX     0xffffffffffffffffULL
123 
124 /* Guest_tsc -> host_tsc conversion requires 64-bit division.  */
125 static int __read_mostly cpu_preemption_timer_multi;
126 static bool __read_mostly enable_preemption_timer = 1;
127 #ifdef CONFIG_X86_64
128 module_param_named(preemption_timer, enable_preemption_timer, bool, S_IRUGO);
129 #endif
130 
131 #define KVM_VM_CR0_ALWAYS_OFF (X86_CR0_NW | X86_CR0_CD)
132 #define KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST X86_CR0_NE
133 #define KVM_VM_CR0_ALWAYS_ON				\
134 	(KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST | 	\
135 	 X86_CR0_WP | X86_CR0_PG | X86_CR0_PE)
136 #define KVM_CR4_GUEST_OWNED_BITS				      \
137 	(X86_CR4_PVI | X86_CR4_DE | X86_CR4_PCE | X86_CR4_OSFXSR      \
138 	 | X86_CR4_OSXMMEXCPT | X86_CR4_LA57 | X86_CR4_TSD)
139 
140 #define KVM_VM_CR4_ALWAYS_ON_UNRESTRICTED_GUEST X86_CR4_VMXE
141 #define KVM_PMODE_VM_CR4_ALWAYS_ON (X86_CR4_PAE | X86_CR4_VMXE)
142 #define KVM_RMODE_VM_CR4_ALWAYS_ON (X86_CR4_VME | X86_CR4_PAE | X86_CR4_VMXE)
143 
144 #define RMODE_GUEST_OWNED_EFLAGS_BITS (~(X86_EFLAGS_IOPL | X86_EFLAGS_VM))
145 
146 #define MSR_IA32_RTIT_STATUS_MASK (~(RTIT_STATUS_FILTEREN | \
147 	RTIT_STATUS_CONTEXTEN | RTIT_STATUS_TRIGGEREN | \
148 	RTIT_STATUS_ERROR | RTIT_STATUS_STOPPED | \
149 	RTIT_STATUS_BYTECNT))
150 
151 #define MSR_IA32_RTIT_OUTPUT_BASE_MASK \
152 	(~((1UL << cpuid_query_maxphyaddr(vcpu)) - 1) | 0x7f)
153 
154 /*
155  * These 2 parameters are used to config the controls for Pause-Loop Exiting:
156  * ple_gap:    upper bound on the amount of time between two successive
157  *             executions of PAUSE in a loop. Also indicate if ple enabled.
158  *             According to test, this time is usually smaller than 128 cycles.
159  * ple_window: upper bound on the amount of time a guest is allowed to execute
160  *             in a PAUSE loop. Tests indicate that most spinlocks are held for
161  *             less than 2^12 cycles
162  * Time is measured based on a counter that runs at the same rate as the TSC,
163  * refer SDM volume 3b section 21.6.13 & 22.1.3.
164  */
165 static unsigned int ple_gap = KVM_DEFAULT_PLE_GAP;
166 module_param(ple_gap, uint, 0444);
167 
168 static unsigned int ple_window = KVM_VMX_DEFAULT_PLE_WINDOW;
169 module_param(ple_window, uint, 0444);
170 
171 /* Default doubles per-vcpu window every exit. */
172 static unsigned int ple_window_grow = KVM_DEFAULT_PLE_WINDOW_GROW;
173 module_param(ple_window_grow, uint, 0444);
174 
175 /* Default resets per-vcpu window every exit to ple_window. */
176 static unsigned int ple_window_shrink = KVM_DEFAULT_PLE_WINDOW_SHRINK;
177 module_param(ple_window_shrink, uint, 0444);
178 
179 /* Default is to compute the maximum so we can never overflow. */
180 static unsigned int ple_window_max        = KVM_VMX_DEFAULT_PLE_WINDOW_MAX;
181 module_param(ple_window_max, uint, 0444);
182 
183 /* Default is SYSTEM mode, 1 for host-guest mode */
184 int __read_mostly pt_mode = PT_MODE_SYSTEM;
185 module_param(pt_mode, int, S_IRUGO);
186 
187 static DEFINE_STATIC_KEY_FALSE(vmx_l1d_should_flush);
188 static DEFINE_STATIC_KEY_FALSE(vmx_l1d_flush_cond);
189 static DEFINE_MUTEX(vmx_l1d_flush_mutex);
190 
191 /* Storage for pre module init parameter parsing */
192 static enum vmx_l1d_flush_state __read_mostly vmentry_l1d_flush_param = VMENTER_L1D_FLUSH_AUTO;
193 
194 static const struct {
195 	const char *option;
196 	bool for_parse;
197 } vmentry_l1d_param[] = {
198 	[VMENTER_L1D_FLUSH_AUTO]	 = {"auto", true},
199 	[VMENTER_L1D_FLUSH_NEVER]	 = {"never", true},
200 	[VMENTER_L1D_FLUSH_COND]	 = {"cond", true},
201 	[VMENTER_L1D_FLUSH_ALWAYS]	 = {"always", true},
202 	[VMENTER_L1D_FLUSH_EPT_DISABLED] = {"EPT disabled", false},
203 	[VMENTER_L1D_FLUSH_NOT_REQUIRED] = {"not required", false},
204 };
205 
206 #define L1D_CACHE_ORDER 4
207 static void *vmx_l1d_flush_pages;
208 
209 static int vmx_setup_l1d_flush(enum vmx_l1d_flush_state l1tf)
210 {
211 	struct page *page;
212 	unsigned int i;
213 
214 	if (!boot_cpu_has_bug(X86_BUG_L1TF)) {
215 		l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_NOT_REQUIRED;
216 		return 0;
217 	}
218 
219 	if (!enable_ept) {
220 		l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_EPT_DISABLED;
221 		return 0;
222 	}
223 
224 	if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES)) {
225 		u64 msr;
226 
227 		rdmsrl(MSR_IA32_ARCH_CAPABILITIES, msr);
228 		if (msr & ARCH_CAP_SKIP_VMENTRY_L1DFLUSH) {
229 			l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_NOT_REQUIRED;
230 			return 0;
231 		}
232 	}
233 
234 	/* If set to auto use the default l1tf mitigation method */
235 	if (l1tf == VMENTER_L1D_FLUSH_AUTO) {
236 		switch (l1tf_mitigation) {
237 		case L1TF_MITIGATION_OFF:
238 			l1tf = VMENTER_L1D_FLUSH_NEVER;
239 			break;
240 		case L1TF_MITIGATION_FLUSH_NOWARN:
241 		case L1TF_MITIGATION_FLUSH:
242 		case L1TF_MITIGATION_FLUSH_NOSMT:
243 			l1tf = VMENTER_L1D_FLUSH_COND;
244 			break;
245 		case L1TF_MITIGATION_FULL:
246 		case L1TF_MITIGATION_FULL_FORCE:
247 			l1tf = VMENTER_L1D_FLUSH_ALWAYS;
248 			break;
249 		}
250 	} else if (l1tf_mitigation == L1TF_MITIGATION_FULL_FORCE) {
251 		l1tf = VMENTER_L1D_FLUSH_ALWAYS;
252 	}
253 
254 	if (l1tf != VMENTER_L1D_FLUSH_NEVER && !vmx_l1d_flush_pages &&
255 	    !boot_cpu_has(X86_FEATURE_FLUSH_L1D)) {
256 		/*
257 		 * This allocation for vmx_l1d_flush_pages is not tied to a VM
258 		 * lifetime and so should not be charged to a memcg.
259 		 */
260 		page = alloc_pages(GFP_KERNEL, L1D_CACHE_ORDER);
261 		if (!page)
262 			return -ENOMEM;
263 		vmx_l1d_flush_pages = page_address(page);
264 
265 		/*
266 		 * Initialize each page with a different pattern in
267 		 * order to protect against KSM in the nested
268 		 * virtualization case.
269 		 */
270 		for (i = 0; i < 1u << L1D_CACHE_ORDER; ++i) {
271 			memset(vmx_l1d_flush_pages + i * PAGE_SIZE, i + 1,
272 			       PAGE_SIZE);
273 		}
274 	}
275 
276 	l1tf_vmx_mitigation = l1tf;
277 
278 	if (l1tf != VMENTER_L1D_FLUSH_NEVER)
279 		static_branch_enable(&vmx_l1d_should_flush);
280 	else
281 		static_branch_disable(&vmx_l1d_should_flush);
282 
283 	if (l1tf == VMENTER_L1D_FLUSH_COND)
284 		static_branch_enable(&vmx_l1d_flush_cond);
285 	else
286 		static_branch_disable(&vmx_l1d_flush_cond);
287 	return 0;
288 }
289 
290 static int vmentry_l1d_flush_parse(const char *s)
291 {
292 	unsigned int i;
293 
294 	if (s) {
295 		for (i = 0; i < ARRAY_SIZE(vmentry_l1d_param); i++) {
296 			if (vmentry_l1d_param[i].for_parse &&
297 			    sysfs_streq(s, vmentry_l1d_param[i].option))
298 				return i;
299 		}
300 	}
301 	return -EINVAL;
302 }
303 
304 static int vmentry_l1d_flush_set(const char *s, const struct kernel_param *kp)
305 {
306 	int l1tf, ret;
307 
308 	l1tf = vmentry_l1d_flush_parse(s);
309 	if (l1tf < 0)
310 		return l1tf;
311 
312 	if (!boot_cpu_has(X86_BUG_L1TF))
313 		return 0;
314 
315 	/*
316 	 * Has vmx_init() run already? If not then this is the pre init
317 	 * parameter parsing. In that case just store the value and let
318 	 * vmx_init() do the proper setup after enable_ept has been
319 	 * established.
320 	 */
321 	if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_AUTO) {
322 		vmentry_l1d_flush_param = l1tf;
323 		return 0;
324 	}
325 
326 	mutex_lock(&vmx_l1d_flush_mutex);
327 	ret = vmx_setup_l1d_flush(l1tf);
328 	mutex_unlock(&vmx_l1d_flush_mutex);
329 	return ret;
330 }
331 
332 static int vmentry_l1d_flush_get(char *s, const struct kernel_param *kp)
333 {
334 	if (WARN_ON_ONCE(l1tf_vmx_mitigation >= ARRAY_SIZE(vmentry_l1d_param)))
335 		return sprintf(s, "???\n");
336 
337 	return sprintf(s, "%s\n", vmentry_l1d_param[l1tf_vmx_mitigation].option);
338 }
339 
340 static const struct kernel_param_ops vmentry_l1d_flush_ops = {
341 	.set = vmentry_l1d_flush_set,
342 	.get = vmentry_l1d_flush_get,
343 };
344 module_param_cb(vmentry_l1d_flush, &vmentry_l1d_flush_ops, NULL, 0644);
345 
346 static bool guest_state_valid(struct kvm_vcpu *vcpu);
347 static u32 vmx_segment_access_rights(struct kvm_segment *var);
348 static __always_inline void vmx_disable_intercept_for_msr(unsigned long *msr_bitmap,
349 							  u32 msr, int type);
350 
351 void vmx_vmexit(void);
352 
353 #define vmx_insn_failed(fmt...)		\
354 do {					\
355 	WARN_ONCE(1, fmt);		\
356 	pr_warn_ratelimited(fmt);	\
357 } while (0)
358 
359 asmlinkage void vmread_error(unsigned long field, bool fault)
360 {
361 	if (fault)
362 		kvm_spurious_fault();
363 	else
364 		vmx_insn_failed("kvm: vmread failed: field=%lx\n", field);
365 }
366 
367 noinline void vmwrite_error(unsigned long field, unsigned long value)
368 {
369 	vmx_insn_failed("kvm: vmwrite failed: field=%lx val=%lx err=%d\n",
370 			field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
371 }
372 
373 noinline void vmclear_error(struct vmcs *vmcs, u64 phys_addr)
374 {
375 	vmx_insn_failed("kvm: vmclear failed: %p/%llx\n", vmcs, phys_addr);
376 }
377 
378 noinline void vmptrld_error(struct vmcs *vmcs, u64 phys_addr)
379 {
380 	vmx_insn_failed("kvm: vmptrld failed: %p/%llx\n", vmcs, phys_addr);
381 }
382 
383 noinline void invvpid_error(unsigned long ext, u16 vpid, gva_t gva)
384 {
385 	vmx_insn_failed("kvm: invvpid failed: ext=0x%lx vpid=%u gva=0x%lx\n",
386 			ext, vpid, gva);
387 }
388 
389 noinline void invept_error(unsigned long ext, u64 eptp, gpa_t gpa)
390 {
391 	vmx_insn_failed("kvm: invept failed: ext=0x%lx eptp=%llx gpa=0x%llx\n",
392 			ext, eptp, gpa);
393 }
394 
395 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
396 DEFINE_PER_CPU(struct vmcs *, current_vmcs);
397 /*
398  * We maintain a per-CPU linked-list of VMCS loaded on that CPU. This is needed
399  * when a CPU is brought down, and we need to VMCLEAR all VMCSs loaded on it.
400  */
401 static DEFINE_PER_CPU(struct list_head, loaded_vmcss_on_cpu);
402 
403 /*
404  * We maintian a per-CPU linked-list of vCPU, so in wakeup_handler() we
405  * can find which vCPU should be waken up.
406  */
407 static DEFINE_PER_CPU(struct list_head, blocked_vcpu_on_cpu);
408 static DEFINE_PER_CPU(spinlock_t, blocked_vcpu_on_cpu_lock);
409 
410 static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
411 static DEFINE_SPINLOCK(vmx_vpid_lock);
412 
413 struct vmcs_config vmcs_config;
414 struct vmx_capability vmx_capability;
415 
416 #define VMX_SEGMENT_FIELD(seg)					\
417 	[VCPU_SREG_##seg] = {                                   \
418 		.selector = GUEST_##seg##_SELECTOR,		\
419 		.base = GUEST_##seg##_BASE,		   	\
420 		.limit = GUEST_##seg##_LIMIT,		   	\
421 		.ar_bytes = GUEST_##seg##_AR_BYTES,	   	\
422 	}
423 
424 static const struct kvm_vmx_segment_field {
425 	unsigned selector;
426 	unsigned base;
427 	unsigned limit;
428 	unsigned ar_bytes;
429 } kvm_vmx_segment_fields[] = {
430 	VMX_SEGMENT_FIELD(CS),
431 	VMX_SEGMENT_FIELD(DS),
432 	VMX_SEGMENT_FIELD(ES),
433 	VMX_SEGMENT_FIELD(FS),
434 	VMX_SEGMENT_FIELD(GS),
435 	VMX_SEGMENT_FIELD(SS),
436 	VMX_SEGMENT_FIELD(TR),
437 	VMX_SEGMENT_FIELD(LDTR),
438 };
439 
440 static unsigned long host_idt_base;
441 
442 /*
443  * Though SYSCALL is only supported in 64-bit mode on Intel CPUs, kvm
444  * will emulate SYSCALL in legacy mode if the vendor string in guest
445  * CPUID.0:{EBX,ECX,EDX} is "AuthenticAMD" or "AMDisbetter!" To
446  * support this emulation, IA32_STAR must always be included in
447  * vmx_msr_index[], even in i386 builds.
448  */
449 const u32 vmx_msr_index[] = {
450 #ifdef CONFIG_X86_64
451 	MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
452 #endif
453 	MSR_EFER, MSR_TSC_AUX, MSR_STAR,
454 	MSR_IA32_TSX_CTRL,
455 };
456 
457 #if IS_ENABLED(CONFIG_HYPERV)
458 static bool __read_mostly enlightened_vmcs = true;
459 module_param(enlightened_vmcs, bool, 0444);
460 
461 /* check_ept_pointer() should be under protection of ept_pointer_lock. */
462 static void check_ept_pointer_match(struct kvm *kvm)
463 {
464 	struct kvm_vcpu *vcpu;
465 	u64 tmp_eptp = INVALID_PAGE;
466 	int i;
467 
468 	kvm_for_each_vcpu(i, vcpu, kvm) {
469 		if (!VALID_PAGE(tmp_eptp)) {
470 			tmp_eptp = to_vmx(vcpu)->ept_pointer;
471 		} else if (tmp_eptp != to_vmx(vcpu)->ept_pointer) {
472 			to_kvm_vmx(kvm)->ept_pointers_match
473 				= EPT_POINTERS_MISMATCH;
474 			return;
475 		}
476 	}
477 
478 	to_kvm_vmx(kvm)->ept_pointers_match = EPT_POINTERS_MATCH;
479 }
480 
481 static int kvm_fill_hv_flush_list_func(struct hv_guest_mapping_flush_list *flush,
482 		void *data)
483 {
484 	struct kvm_tlb_range *range = data;
485 
486 	return hyperv_fill_flush_guest_mapping_list(flush, range->start_gfn,
487 			range->pages);
488 }
489 
490 static inline int __hv_remote_flush_tlb_with_range(struct kvm *kvm,
491 		struct kvm_vcpu *vcpu, struct kvm_tlb_range *range)
492 {
493 	u64 ept_pointer = to_vmx(vcpu)->ept_pointer;
494 
495 	/*
496 	 * FLUSH_GUEST_PHYSICAL_ADDRESS_SPACE hypercall needs address
497 	 * of the base of EPT PML4 table, strip off EPT configuration
498 	 * information.
499 	 */
500 	if (range)
501 		return hyperv_flush_guest_mapping_range(ept_pointer & PAGE_MASK,
502 				kvm_fill_hv_flush_list_func, (void *)range);
503 	else
504 		return hyperv_flush_guest_mapping(ept_pointer & PAGE_MASK);
505 }
506 
507 static int hv_remote_flush_tlb_with_range(struct kvm *kvm,
508 		struct kvm_tlb_range *range)
509 {
510 	struct kvm_vcpu *vcpu;
511 	int ret = 0, i;
512 
513 	spin_lock(&to_kvm_vmx(kvm)->ept_pointer_lock);
514 
515 	if (to_kvm_vmx(kvm)->ept_pointers_match == EPT_POINTERS_CHECK)
516 		check_ept_pointer_match(kvm);
517 
518 	if (to_kvm_vmx(kvm)->ept_pointers_match != EPT_POINTERS_MATCH) {
519 		kvm_for_each_vcpu(i, vcpu, kvm) {
520 			/* If ept_pointer is invalid pointer, bypass flush request. */
521 			if (VALID_PAGE(to_vmx(vcpu)->ept_pointer))
522 				ret |= __hv_remote_flush_tlb_with_range(
523 					kvm, vcpu, range);
524 		}
525 	} else {
526 		ret = __hv_remote_flush_tlb_with_range(kvm,
527 				kvm_get_vcpu(kvm, 0), range);
528 	}
529 
530 	spin_unlock(&to_kvm_vmx(kvm)->ept_pointer_lock);
531 	return ret;
532 }
533 static int hv_remote_flush_tlb(struct kvm *kvm)
534 {
535 	return hv_remote_flush_tlb_with_range(kvm, NULL);
536 }
537 
538 static int hv_enable_direct_tlbflush(struct kvm_vcpu *vcpu)
539 {
540 	struct hv_enlightened_vmcs *evmcs;
541 	struct hv_partition_assist_pg **p_hv_pa_pg =
542 			&vcpu->kvm->arch.hyperv.hv_pa_pg;
543 	/*
544 	 * Synthetic VM-Exit is not enabled in current code and so All
545 	 * evmcs in singe VM shares same assist page.
546 	 */
547 	if (!*p_hv_pa_pg)
548 		*p_hv_pa_pg = kzalloc(PAGE_SIZE, GFP_KERNEL);
549 
550 	if (!*p_hv_pa_pg)
551 		return -ENOMEM;
552 
553 	evmcs = (struct hv_enlightened_vmcs *)to_vmx(vcpu)->loaded_vmcs->vmcs;
554 
555 	evmcs->partition_assist_page =
556 		__pa(*p_hv_pa_pg);
557 	evmcs->hv_vm_id = (unsigned long)vcpu->kvm;
558 	evmcs->hv_enlightenments_control.nested_flush_hypercall = 1;
559 
560 	return 0;
561 }
562 
563 #endif /* IS_ENABLED(CONFIG_HYPERV) */
564 
565 /*
566  * Comment's format: document - errata name - stepping - processor name.
567  * Refer from
568  * https://www.virtualbox.org/svn/vbox/trunk/src/VBox/VMM/VMMR0/HMR0.cpp
569  */
570 static u32 vmx_preemption_cpu_tfms[] = {
571 /* 323344.pdf - BA86   - D0 - Xeon 7500 Series */
572 0x000206E6,
573 /* 323056.pdf - AAX65  - C2 - Xeon L3406 */
574 /* 322814.pdf - AAT59  - C2 - i7-600, i5-500, i5-400 and i3-300 Mobile */
575 /* 322911.pdf - AAU65  - C2 - i5-600, i3-500 Desktop and Pentium G6950 */
576 0x00020652,
577 /* 322911.pdf - AAU65  - K0 - i5-600, i3-500 Desktop and Pentium G6950 */
578 0x00020655,
579 /* 322373.pdf - AAO95  - B1 - Xeon 3400 Series */
580 /* 322166.pdf - AAN92  - B1 - i7-800 and i5-700 Desktop */
581 /*
582  * 320767.pdf - AAP86  - B1 -
583  * i7-900 Mobile Extreme, i7-800 and i7-700 Mobile
584  */
585 0x000106E5,
586 /* 321333.pdf - AAM126 - C0 - Xeon 3500 */
587 0x000106A0,
588 /* 321333.pdf - AAM126 - C1 - Xeon 3500 */
589 0x000106A1,
590 /* 320836.pdf - AAJ124 - C0 - i7-900 Desktop Extreme and i7-900 Desktop */
591 0x000106A4,
592  /* 321333.pdf - AAM126 - D0 - Xeon 3500 */
593  /* 321324.pdf - AAK139 - D0 - Xeon 5500 */
594  /* 320836.pdf - AAJ124 - D0 - i7-900 Extreme and i7-900 Desktop */
595 0x000106A5,
596  /* Xeon E3-1220 V2 */
597 0x000306A8,
598 };
599 
600 static inline bool cpu_has_broken_vmx_preemption_timer(void)
601 {
602 	u32 eax = cpuid_eax(0x00000001), i;
603 
604 	/* Clear the reserved bits */
605 	eax &= ~(0x3U << 14 | 0xfU << 28);
606 	for (i = 0; i < ARRAY_SIZE(vmx_preemption_cpu_tfms); i++)
607 		if (eax == vmx_preemption_cpu_tfms[i])
608 			return true;
609 
610 	return false;
611 }
612 
613 static inline bool cpu_need_virtualize_apic_accesses(struct kvm_vcpu *vcpu)
614 {
615 	return flexpriority_enabled && lapic_in_kernel(vcpu);
616 }
617 
618 static inline bool report_flexpriority(void)
619 {
620 	return flexpriority_enabled;
621 }
622 
623 static inline int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
624 {
625 	int i;
626 
627 	for (i = 0; i < vmx->nmsrs; ++i)
628 		if (vmx_msr_index[vmx->guest_msrs[i].index] == msr)
629 			return i;
630 	return -1;
631 }
632 
633 struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
634 {
635 	int i;
636 
637 	i = __find_msr_index(vmx, msr);
638 	if (i >= 0)
639 		return &vmx->guest_msrs[i];
640 	return NULL;
641 }
642 
643 static int vmx_set_guest_msr(struct vcpu_vmx *vmx, struct shared_msr_entry *msr, u64 data)
644 {
645 	int ret = 0;
646 
647 	u64 old_msr_data = msr->data;
648 	msr->data = data;
649 	if (msr - vmx->guest_msrs < vmx->save_nmsrs) {
650 		preempt_disable();
651 		ret = kvm_set_shared_msr(msr->index, msr->data,
652 					 msr->mask);
653 		preempt_enable();
654 		if (ret)
655 			msr->data = old_msr_data;
656 	}
657 	return ret;
658 }
659 
660 #ifdef CONFIG_KEXEC_CORE
661 static void crash_vmclear_local_loaded_vmcss(void)
662 {
663 	int cpu = raw_smp_processor_id();
664 	struct loaded_vmcs *v;
665 
666 	list_for_each_entry(v, &per_cpu(loaded_vmcss_on_cpu, cpu),
667 			    loaded_vmcss_on_cpu_link)
668 		vmcs_clear(v->vmcs);
669 }
670 #endif /* CONFIG_KEXEC_CORE */
671 
672 static void __loaded_vmcs_clear(void *arg)
673 {
674 	struct loaded_vmcs *loaded_vmcs = arg;
675 	int cpu = raw_smp_processor_id();
676 
677 	if (loaded_vmcs->cpu != cpu)
678 		return; /* vcpu migration can race with cpu offline */
679 	if (per_cpu(current_vmcs, cpu) == loaded_vmcs->vmcs)
680 		per_cpu(current_vmcs, cpu) = NULL;
681 
682 	vmcs_clear(loaded_vmcs->vmcs);
683 	if (loaded_vmcs->shadow_vmcs && loaded_vmcs->launched)
684 		vmcs_clear(loaded_vmcs->shadow_vmcs);
685 
686 	list_del(&loaded_vmcs->loaded_vmcss_on_cpu_link);
687 
688 	/*
689 	 * Ensure all writes to loaded_vmcs, including deleting it from its
690 	 * current percpu list, complete before setting loaded_vmcs->vcpu to
691 	 * -1, otherwise a different cpu can see vcpu == -1 first and add
692 	 * loaded_vmcs to its percpu list before it's deleted from this cpu's
693 	 * list. Pairs with the smp_rmb() in vmx_vcpu_load_vmcs().
694 	 */
695 	smp_wmb();
696 
697 	loaded_vmcs->cpu = -1;
698 	loaded_vmcs->launched = 0;
699 }
700 
701 void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs)
702 {
703 	int cpu = loaded_vmcs->cpu;
704 
705 	if (cpu != -1)
706 		smp_call_function_single(cpu,
707 			 __loaded_vmcs_clear, loaded_vmcs, 1);
708 }
709 
710 static bool vmx_segment_cache_test_set(struct vcpu_vmx *vmx, unsigned seg,
711 				       unsigned field)
712 {
713 	bool ret;
714 	u32 mask = 1 << (seg * SEG_FIELD_NR + field);
715 
716 	if (!kvm_register_is_available(&vmx->vcpu, VCPU_EXREG_SEGMENTS)) {
717 		kvm_register_mark_available(&vmx->vcpu, VCPU_EXREG_SEGMENTS);
718 		vmx->segment_cache.bitmask = 0;
719 	}
720 	ret = vmx->segment_cache.bitmask & mask;
721 	vmx->segment_cache.bitmask |= mask;
722 	return ret;
723 }
724 
725 static u16 vmx_read_guest_seg_selector(struct vcpu_vmx *vmx, unsigned seg)
726 {
727 	u16 *p = &vmx->segment_cache.seg[seg].selector;
728 
729 	if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_SEL))
730 		*p = vmcs_read16(kvm_vmx_segment_fields[seg].selector);
731 	return *p;
732 }
733 
734 static ulong vmx_read_guest_seg_base(struct vcpu_vmx *vmx, unsigned seg)
735 {
736 	ulong *p = &vmx->segment_cache.seg[seg].base;
737 
738 	if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_BASE))
739 		*p = vmcs_readl(kvm_vmx_segment_fields[seg].base);
740 	return *p;
741 }
742 
743 static u32 vmx_read_guest_seg_limit(struct vcpu_vmx *vmx, unsigned seg)
744 {
745 	u32 *p = &vmx->segment_cache.seg[seg].limit;
746 
747 	if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_LIMIT))
748 		*p = vmcs_read32(kvm_vmx_segment_fields[seg].limit);
749 	return *p;
750 }
751 
752 static u32 vmx_read_guest_seg_ar(struct vcpu_vmx *vmx, unsigned seg)
753 {
754 	u32 *p = &vmx->segment_cache.seg[seg].ar;
755 
756 	if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_AR))
757 		*p = vmcs_read32(kvm_vmx_segment_fields[seg].ar_bytes);
758 	return *p;
759 }
760 
761 void update_exception_bitmap(struct kvm_vcpu *vcpu)
762 {
763 	u32 eb;
764 
765 	eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR) |
766 	     (1u << DB_VECTOR) | (1u << AC_VECTOR);
767 	/*
768 	 * Guest access to VMware backdoor ports could legitimately
769 	 * trigger #GP because of TSS I/O permission bitmap.
770 	 * We intercept those #GP and allow access to them anyway
771 	 * as VMware does.
772 	 */
773 	if (enable_vmware_backdoor)
774 		eb |= (1u << GP_VECTOR);
775 	if ((vcpu->guest_debug &
776 	     (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) ==
777 	    (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP))
778 		eb |= 1u << BP_VECTOR;
779 	if (to_vmx(vcpu)->rmode.vm86_active)
780 		eb = ~0;
781 	if (enable_ept)
782 		eb &= ~(1u << PF_VECTOR);
783 
784 	/* When we are running a nested L2 guest and L1 specified for it a
785 	 * certain exception bitmap, we must trap the same exceptions and pass
786 	 * them to L1. When running L2, we will only handle the exceptions
787 	 * specified above if L1 did not want them.
788 	 */
789 	if (is_guest_mode(vcpu))
790 		eb |= get_vmcs12(vcpu)->exception_bitmap;
791 
792 	vmcs_write32(EXCEPTION_BITMAP, eb);
793 }
794 
795 /*
796  * Check if MSR is intercepted for currently loaded MSR bitmap.
797  */
798 static bool msr_write_intercepted(struct kvm_vcpu *vcpu, u32 msr)
799 {
800 	unsigned long *msr_bitmap;
801 	int f = sizeof(unsigned long);
802 
803 	if (!cpu_has_vmx_msr_bitmap())
804 		return true;
805 
806 	msr_bitmap = to_vmx(vcpu)->loaded_vmcs->msr_bitmap;
807 
808 	if (msr <= 0x1fff) {
809 		return !!test_bit(msr, msr_bitmap + 0x800 / f);
810 	} else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
811 		msr &= 0x1fff;
812 		return !!test_bit(msr, msr_bitmap + 0xc00 / f);
813 	}
814 
815 	return true;
816 }
817 
818 static void clear_atomic_switch_msr_special(struct vcpu_vmx *vmx,
819 		unsigned long entry, unsigned long exit)
820 {
821 	vm_entry_controls_clearbit(vmx, entry);
822 	vm_exit_controls_clearbit(vmx, exit);
823 }
824 
825 int vmx_find_msr_index(struct vmx_msrs *m, u32 msr)
826 {
827 	unsigned int i;
828 
829 	for (i = 0; i < m->nr; ++i) {
830 		if (m->val[i].index == msr)
831 			return i;
832 	}
833 	return -ENOENT;
834 }
835 
836 static void clear_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr)
837 {
838 	int i;
839 	struct msr_autoload *m = &vmx->msr_autoload;
840 
841 	switch (msr) {
842 	case MSR_EFER:
843 		if (cpu_has_load_ia32_efer()) {
844 			clear_atomic_switch_msr_special(vmx,
845 					VM_ENTRY_LOAD_IA32_EFER,
846 					VM_EXIT_LOAD_IA32_EFER);
847 			return;
848 		}
849 		break;
850 	case MSR_CORE_PERF_GLOBAL_CTRL:
851 		if (cpu_has_load_perf_global_ctrl()) {
852 			clear_atomic_switch_msr_special(vmx,
853 					VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
854 					VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
855 			return;
856 		}
857 		break;
858 	}
859 	i = vmx_find_msr_index(&m->guest, msr);
860 	if (i < 0)
861 		goto skip_guest;
862 	--m->guest.nr;
863 	m->guest.val[i] = m->guest.val[m->guest.nr];
864 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->guest.nr);
865 
866 skip_guest:
867 	i = vmx_find_msr_index(&m->host, msr);
868 	if (i < 0)
869 		return;
870 
871 	--m->host.nr;
872 	m->host.val[i] = m->host.val[m->host.nr];
873 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->host.nr);
874 }
875 
876 static void add_atomic_switch_msr_special(struct vcpu_vmx *vmx,
877 		unsigned long entry, unsigned long exit,
878 		unsigned long guest_val_vmcs, unsigned long host_val_vmcs,
879 		u64 guest_val, u64 host_val)
880 {
881 	vmcs_write64(guest_val_vmcs, guest_val);
882 	if (host_val_vmcs != HOST_IA32_EFER)
883 		vmcs_write64(host_val_vmcs, host_val);
884 	vm_entry_controls_setbit(vmx, entry);
885 	vm_exit_controls_setbit(vmx, exit);
886 }
887 
888 static void add_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr,
889 				  u64 guest_val, u64 host_val, bool entry_only)
890 {
891 	int i, j = 0;
892 	struct msr_autoload *m = &vmx->msr_autoload;
893 
894 	switch (msr) {
895 	case MSR_EFER:
896 		if (cpu_has_load_ia32_efer()) {
897 			add_atomic_switch_msr_special(vmx,
898 					VM_ENTRY_LOAD_IA32_EFER,
899 					VM_EXIT_LOAD_IA32_EFER,
900 					GUEST_IA32_EFER,
901 					HOST_IA32_EFER,
902 					guest_val, host_val);
903 			return;
904 		}
905 		break;
906 	case MSR_CORE_PERF_GLOBAL_CTRL:
907 		if (cpu_has_load_perf_global_ctrl()) {
908 			add_atomic_switch_msr_special(vmx,
909 					VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
910 					VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL,
911 					GUEST_IA32_PERF_GLOBAL_CTRL,
912 					HOST_IA32_PERF_GLOBAL_CTRL,
913 					guest_val, host_val);
914 			return;
915 		}
916 		break;
917 	case MSR_IA32_PEBS_ENABLE:
918 		/* PEBS needs a quiescent period after being disabled (to write
919 		 * a record).  Disabling PEBS through VMX MSR swapping doesn't
920 		 * provide that period, so a CPU could write host's record into
921 		 * guest's memory.
922 		 */
923 		wrmsrl(MSR_IA32_PEBS_ENABLE, 0);
924 	}
925 
926 	i = vmx_find_msr_index(&m->guest, msr);
927 	if (!entry_only)
928 		j = vmx_find_msr_index(&m->host, msr);
929 
930 	if ((i < 0 && m->guest.nr == NR_LOADSTORE_MSRS) ||
931 		(j < 0 &&  m->host.nr == NR_LOADSTORE_MSRS)) {
932 		printk_once(KERN_WARNING "Not enough msr switch entries. "
933 				"Can't add msr %x\n", msr);
934 		return;
935 	}
936 	if (i < 0) {
937 		i = m->guest.nr++;
938 		vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->guest.nr);
939 	}
940 	m->guest.val[i].index = msr;
941 	m->guest.val[i].value = guest_val;
942 
943 	if (entry_only)
944 		return;
945 
946 	if (j < 0) {
947 		j = m->host.nr++;
948 		vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->host.nr);
949 	}
950 	m->host.val[j].index = msr;
951 	m->host.val[j].value = host_val;
952 }
953 
954 static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset)
955 {
956 	u64 guest_efer = vmx->vcpu.arch.efer;
957 	u64 ignore_bits = 0;
958 
959 	/* Shadow paging assumes NX to be available.  */
960 	if (!enable_ept)
961 		guest_efer |= EFER_NX;
962 
963 	/*
964 	 * LMA and LME handled by hardware; SCE meaningless outside long mode.
965 	 */
966 	ignore_bits |= EFER_SCE;
967 #ifdef CONFIG_X86_64
968 	ignore_bits |= EFER_LMA | EFER_LME;
969 	/* SCE is meaningful only in long mode on Intel */
970 	if (guest_efer & EFER_LMA)
971 		ignore_bits &= ~(u64)EFER_SCE;
972 #endif
973 
974 	/*
975 	 * On EPT, we can't emulate NX, so we must switch EFER atomically.
976 	 * On CPUs that support "load IA32_EFER", always switch EFER
977 	 * atomically, since it's faster than switching it manually.
978 	 */
979 	if (cpu_has_load_ia32_efer() ||
980 	    (enable_ept && ((vmx->vcpu.arch.efer ^ host_efer) & EFER_NX))) {
981 		if (!(guest_efer & EFER_LMA))
982 			guest_efer &= ~EFER_LME;
983 		if (guest_efer != host_efer)
984 			add_atomic_switch_msr(vmx, MSR_EFER,
985 					      guest_efer, host_efer, false);
986 		else
987 			clear_atomic_switch_msr(vmx, MSR_EFER);
988 		return false;
989 	} else {
990 		clear_atomic_switch_msr(vmx, MSR_EFER);
991 
992 		guest_efer &= ~ignore_bits;
993 		guest_efer |= host_efer & ignore_bits;
994 
995 		vmx->guest_msrs[efer_offset].data = guest_efer;
996 		vmx->guest_msrs[efer_offset].mask = ~ignore_bits;
997 
998 		return true;
999 	}
1000 }
1001 
1002 #ifdef CONFIG_X86_32
1003 /*
1004  * On 32-bit kernels, VM exits still load the FS and GS bases from the
1005  * VMCS rather than the segment table.  KVM uses this helper to figure
1006  * out the current bases to poke them into the VMCS before entry.
1007  */
1008 static unsigned long segment_base(u16 selector)
1009 {
1010 	struct desc_struct *table;
1011 	unsigned long v;
1012 
1013 	if (!(selector & ~SEGMENT_RPL_MASK))
1014 		return 0;
1015 
1016 	table = get_current_gdt_ro();
1017 
1018 	if ((selector & SEGMENT_TI_MASK) == SEGMENT_LDT) {
1019 		u16 ldt_selector = kvm_read_ldt();
1020 
1021 		if (!(ldt_selector & ~SEGMENT_RPL_MASK))
1022 			return 0;
1023 
1024 		table = (struct desc_struct *)segment_base(ldt_selector);
1025 	}
1026 	v = get_desc_base(&table[selector >> 3]);
1027 	return v;
1028 }
1029 #endif
1030 
1031 static inline bool pt_can_write_msr(struct vcpu_vmx *vmx)
1032 {
1033 	return vmx_pt_mode_is_host_guest() &&
1034 	       !(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN);
1035 }
1036 
1037 static inline void pt_load_msr(struct pt_ctx *ctx, u32 addr_range)
1038 {
1039 	u32 i;
1040 
1041 	wrmsrl(MSR_IA32_RTIT_STATUS, ctx->status);
1042 	wrmsrl(MSR_IA32_RTIT_OUTPUT_BASE, ctx->output_base);
1043 	wrmsrl(MSR_IA32_RTIT_OUTPUT_MASK, ctx->output_mask);
1044 	wrmsrl(MSR_IA32_RTIT_CR3_MATCH, ctx->cr3_match);
1045 	for (i = 0; i < addr_range; i++) {
1046 		wrmsrl(MSR_IA32_RTIT_ADDR0_A + i * 2, ctx->addr_a[i]);
1047 		wrmsrl(MSR_IA32_RTIT_ADDR0_B + i * 2, ctx->addr_b[i]);
1048 	}
1049 }
1050 
1051 static inline void pt_save_msr(struct pt_ctx *ctx, u32 addr_range)
1052 {
1053 	u32 i;
1054 
1055 	rdmsrl(MSR_IA32_RTIT_STATUS, ctx->status);
1056 	rdmsrl(MSR_IA32_RTIT_OUTPUT_BASE, ctx->output_base);
1057 	rdmsrl(MSR_IA32_RTIT_OUTPUT_MASK, ctx->output_mask);
1058 	rdmsrl(MSR_IA32_RTIT_CR3_MATCH, ctx->cr3_match);
1059 	for (i = 0; i < addr_range; i++) {
1060 		rdmsrl(MSR_IA32_RTIT_ADDR0_A + i * 2, ctx->addr_a[i]);
1061 		rdmsrl(MSR_IA32_RTIT_ADDR0_B + i * 2, ctx->addr_b[i]);
1062 	}
1063 }
1064 
1065 static void pt_guest_enter(struct vcpu_vmx *vmx)
1066 {
1067 	if (vmx_pt_mode_is_system())
1068 		return;
1069 
1070 	/*
1071 	 * GUEST_IA32_RTIT_CTL is already set in the VMCS.
1072 	 * Save host state before VM entry.
1073 	 */
1074 	rdmsrl(MSR_IA32_RTIT_CTL, vmx->pt_desc.host.ctl);
1075 	if (vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) {
1076 		wrmsrl(MSR_IA32_RTIT_CTL, 0);
1077 		pt_save_msr(&vmx->pt_desc.host, vmx->pt_desc.addr_range);
1078 		pt_load_msr(&vmx->pt_desc.guest, vmx->pt_desc.addr_range);
1079 	}
1080 }
1081 
1082 static void pt_guest_exit(struct vcpu_vmx *vmx)
1083 {
1084 	if (vmx_pt_mode_is_system())
1085 		return;
1086 
1087 	if (vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) {
1088 		pt_save_msr(&vmx->pt_desc.guest, vmx->pt_desc.addr_range);
1089 		pt_load_msr(&vmx->pt_desc.host, vmx->pt_desc.addr_range);
1090 	}
1091 
1092 	/* Reload host state (IA32_RTIT_CTL will be cleared on VM exit). */
1093 	wrmsrl(MSR_IA32_RTIT_CTL, vmx->pt_desc.host.ctl);
1094 }
1095 
1096 void vmx_set_host_fs_gs(struct vmcs_host_state *host, u16 fs_sel, u16 gs_sel,
1097 			unsigned long fs_base, unsigned long gs_base)
1098 {
1099 	if (unlikely(fs_sel != host->fs_sel)) {
1100 		if (!(fs_sel & 7))
1101 			vmcs_write16(HOST_FS_SELECTOR, fs_sel);
1102 		else
1103 			vmcs_write16(HOST_FS_SELECTOR, 0);
1104 		host->fs_sel = fs_sel;
1105 	}
1106 	if (unlikely(gs_sel != host->gs_sel)) {
1107 		if (!(gs_sel & 7))
1108 			vmcs_write16(HOST_GS_SELECTOR, gs_sel);
1109 		else
1110 			vmcs_write16(HOST_GS_SELECTOR, 0);
1111 		host->gs_sel = gs_sel;
1112 	}
1113 	if (unlikely(fs_base != host->fs_base)) {
1114 		vmcs_writel(HOST_FS_BASE, fs_base);
1115 		host->fs_base = fs_base;
1116 	}
1117 	if (unlikely(gs_base != host->gs_base)) {
1118 		vmcs_writel(HOST_GS_BASE, gs_base);
1119 		host->gs_base = gs_base;
1120 	}
1121 }
1122 
1123 void vmx_prepare_switch_to_guest(struct kvm_vcpu *vcpu)
1124 {
1125 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1126 	struct vmcs_host_state *host_state;
1127 #ifdef CONFIG_X86_64
1128 	int cpu = raw_smp_processor_id();
1129 #endif
1130 	unsigned long fs_base, gs_base;
1131 	u16 fs_sel, gs_sel;
1132 	int i;
1133 
1134 	vmx->req_immediate_exit = false;
1135 
1136 	/*
1137 	 * Note that guest MSRs to be saved/restored can also be changed
1138 	 * when guest state is loaded. This happens when guest transitions
1139 	 * to/from long-mode by setting MSR_EFER.LMA.
1140 	 */
1141 	if (!vmx->guest_msrs_ready) {
1142 		vmx->guest_msrs_ready = true;
1143 		for (i = 0; i < vmx->save_nmsrs; ++i)
1144 			kvm_set_shared_msr(vmx->guest_msrs[i].index,
1145 					   vmx->guest_msrs[i].data,
1146 					   vmx->guest_msrs[i].mask);
1147 
1148 	}
1149 
1150     	if (vmx->nested.need_vmcs12_to_shadow_sync)
1151 		nested_sync_vmcs12_to_shadow(vcpu);
1152 
1153 	if (vmx->guest_state_loaded)
1154 		return;
1155 
1156 	host_state = &vmx->loaded_vmcs->host_state;
1157 
1158 	/*
1159 	 * Set host fs and gs selectors.  Unfortunately, 22.2.3 does not
1160 	 * allow segment selectors with cpl > 0 or ti == 1.
1161 	 */
1162 	host_state->ldt_sel = kvm_read_ldt();
1163 
1164 #ifdef CONFIG_X86_64
1165 	savesegment(ds, host_state->ds_sel);
1166 	savesegment(es, host_state->es_sel);
1167 
1168 	gs_base = cpu_kernelmode_gs_base(cpu);
1169 	if (likely(is_64bit_mm(current->mm))) {
1170 		save_fsgs_for_kvm();
1171 		fs_sel = current->thread.fsindex;
1172 		gs_sel = current->thread.gsindex;
1173 		fs_base = current->thread.fsbase;
1174 		vmx->msr_host_kernel_gs_base = current->thread.gsbase;
1175 	} else {
1176 		savesegment(fs, fs_sel);
1177 		savesegment(gs, gs_sel);
1178 		fs_base = read_msr(MSR_FS_BASE);
1179 		vmx->msr_host_kernel_gs_base = read_msr(MSR_KERNEL_GS_BASE);
1180 	}
1181 
1182 	wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1183 #else
1184 	savesegment(fs, fs_sel);
1185 	savesegment(gs, gs_sel);
1186 	fs_base = segment_base(fs_sel);
1187 	gs_base = segment_base(gs_sel);
1188 #endif
1189 
1190 	vmx_set_host_fs_gs(host_state, fs_sel, gs_sel, fs_base, gs_base);
1191 	vmx->guest_state_loaded = true;
1192 }
1193 
1194 static void vmx_prepare_switch_to_host(struct vcpu_vmx *vmx)
1195 {
1196 	struct vmcs_host_state *host_state;
1197 
1198 	if (!vmx->guest_state_loaded)
1199 		return;
1200 
1201 	host_state = &vmx->loaded_vmcs->host_state;
1202 
1203 	++vmx->vcpu.stat.host_state_reload;
1204 
1205 #ifdef CONFIG_X86_64
1206 	rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1207 #endif
1208 	if (host_state->ldt_sel || (host_state->gs_sel & 7)) {
1209 		kvm_load_ldt(host_state->ldt_sel);
1210 #ifdef CONFIG_X86_64
1211 		load_gs_index(host_state->gs_sel);
1212 #else
1213 		loadsegment(gs, host_state->gs_sel);
1214 #endif
1215 	}
1216 	if (host_state->fs_sel & 7)
1217 		loadsegment(fs, host_state->fs_sel);
1218 #ifdef CONFIG_X86_64
1219 	if (unlikely(host_state->ds_sel | host_state->es_sel)) {
1220 		loadsegment(ds, host_state->ds_sel);
1221 		loadsegment(es, host_state->es_sel);
1222 	}
1223 #endif
1224 	invalidate_tss_limit();
1225 #ifdef CONFIG_X86_64
1226 	wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
1227 #endif
1228 	load_fixmap_gdt(raw_smp_processor_id());
1229 	vmx->guest_state_loaded = false;
1230 	vmx->guest_msrs_ready = false;
1231 }
1232 
1233 #ifdef CONFIG_X86_64
1234 static u64 vmx_read_guest_kernel_gs_base(struct vcpu_vmx *vmx)
1235 {
1236 	preempt_disable();
1237 	if (vmx->guest_state_loaded)
1238 		rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1239 	preempt_enable();
1240 	return vmx->msr_guest_kernel_gs_base;
1241 }
1242 
1243 static void vmx_write_guest_kernel_gs_base(struct vcpu_vmx *vmx, u64 data)
1244 {
1245 	preempt_disable();
1246 	if (vmx->guest_state_loaded)
1247 		wrmsrl(MSR_KERNEL_GS_BASE, data);
1248 	preempt_enable();
1249 	vmx->msr_guest_kernel_gs_base = data;
1250 }
1251 #endif
1252 
1253 static void vmx_vcpu_pi_load(struct kvm_vcpu *vcpu, int cpu)
1254 {
1255 	struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
1256 	struct pi_desc old, new;
1257 	unsigned int dest;
1258 
1259 	/*
1260 	 * In case of hot-plug or hot-unplug, we may have to undo
1261 	 * vmx_vcpu_pi_put even if there is no assigned device.  And we
1262 	 * always keep PI.NDST up to date for simplicity: it makes the
1263 	 * code easier, and CPU migration is not a fast path.
1264 	 */
1265 	if (!pi_test_sn(pi_desc) && vcpu->cpu == cpu)
1266 		return;
1267 
1268 	/*
1269 	 * If the 'nv' field is POSTED_INTR_WAKEUP_VECTOR, do not change
1270 	 * PI.NDST: pi_post_block is the one expected to change PID.NDST and the
1271 	 * wakeup handler expects the vCPU to be on the blocked_vcpu_list that
1272 	 * matches PI.NDST. Otherwise, a vcpu may not be able to be woken up
1273 	 * correctly.
1274 	 */
1275 	if (pi_desc->nv == POSTED_INTR_WAKEUP_VECTOR || vcpu->cpu == cpu) {
1276 		pi_clear_sn(pi_desc);
1277 		goto after_clear_sn;
1278 	}
1279 
1280 	/* The full case.  */
1281 	do {
1282 		old.control = new.control = pi_desc->control;
1283 
1284 		dest = cpu_physical_id(cpu);
1285 
1286 		if (x2apic_enabled())
1287 			new.ndst = dest;
1288 		else
1289 			new.ndst = (dest << 8) & 0xFF00;
1290 
1291 		new.sn = 0;
1292 	} while (cmpxchg64(&pi_desc->control, old.control,
1293 			   new.control) != old.control);
1294 
1295 after_clear_sn:
1296 
1297 	/*
1298 	 * Clear SN before reading the bitmap.  The VT-d firmware
1299 	 * writes the bitmap and reads SN atomically (5.2.3 in the
1300 	 * spec), so it doesn't really have a memory barrier that
1301 	 * pairs with this, but we cannot do that and we need one.
1302 	 */
1303 	smp_mb__after_atomic();
1304 
1305 	if (!pi_is_pir_empty(pi_desc))
1306 		pi_set_on(pi_desc);
1307 }
1308 
1309 void vmx_vcpu_load_vmcs(struct kvm_vcpu *vcpu, int cpu)
1310 {
1311 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1312 	bool already_loaded = vmx->loaded_vmcs->cpu == cpu;
1313 
1314 	if (!already_loaded) {
1315 		loaded_vmcs_clear(vmx->loaded_vmcs);
1316 		local_irq_disable();
1317 
1318 		/*
1319 		 * Ensure loaded_vmcs->cpu is read before adding loaded_vmcs to
1320 		 * this cpu's percpu list, otherwise it may not yet be deleted
1321 		 * from its previous cpu's percpu list.  Pairs with the
1322 		 * smb_wmb() in __loaded_vmcs_clear().
1323 		 */
1324 		smp_rmb();
1325 
1326 		list_add(&vmx->loaded_vmcs->loaded_vmcss_on_cpu_link,
1327 			 &per_cpu(loaded_vmcss_on_cpu, cpu));
1328 		local_irq_enable();
1329 	}
1330 
1331 	if (per_cpu(current_vmcs, cpu) != vmx->loaded_vmcs->vmcs) {
1332 		per_cpu(current_vmcs, cpu) = vmx->loaded_vmcs->vmcs;
1333 		vmcs_load(vmx->loaded_vmcs->vmcs);
1334 		indirect_branch_prediction_barrier();
1335 	}
1336 
1337 	if (!already_loaded) {
1338 		void *gdt = get_current_gdt_ro();
1339 		unsigned long sysenter_esp;
1340 
1341 		kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1342 
1343 		/*
1344 		 * Linux uses per-cpu TSS and GDT, so set these when switching
1345 		 * processors.  See 22.2.4.
1346 		 */
1347 		vmcs_writel(HOST_TR_BASE,
1348 			    (unsigned long)&get_cpu_entry_area(cpu)->tss.x86_tss);
1349 		vmcs_writel(HOST_GDTR_BASE, (unsigned long)gdt);   /* 22.2.4 */
1350 
1351 		rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
1352 		vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
1353 
1354 		vmx->loaded_vmcs->cpu = cpu;
1355 	}
1356 
1357 	/* Setup TSC multiplier */
1358 	if (kvm_has_tsc_control &&
1359 	    vmx->current_tsc_ratio != vcpu->arch.tsc_scaling_ratio)
1360 		decache_tsc_multiplier(vmx);
1361 }
1362 
1363 /*
1364  * Switches to specified vcpu, until a matching vcpu_put(), but assumes
1365  * vcpu mutex is already taken.
1366  */
1367 void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1368 {
1369 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1370 
1371 	vmx_vcpu_load_vmcs(vcpu, cpu);
1372 
1373 	vmx_vcpu_pi_load(vcpu, cpu);
1374 
1375 	vmx->host_pkru = read_pkru();
1376 	vmx->host_debugctlmsr = get_debugctlmsr();
1377 }
1378 
1379 static void vmx_vcpu_pi_put(struct kvm_vcpu *vcpu)
1380 {
1381 	struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
1382 
1383 	if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
1384 		!irq_remapping_cap(IRQ_POSTING_CAP)  ||
1385 		!kvm_vcpu_apicv_active(vcpu))
1386 		return;
1387 
1388 	/* Set SN when the vCPU is preempted */
1389 	if (vcpu->preempted)
1390 		pi_set_sn(pi_desc);
1391 }
1392 
1393 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
1394 {
1395 	vmx_vcpu_pi_put(vcpu);
1396 
1397 	vmx_prepare_switch_to_host(to_vmx(vcpu));
1398 }
1399 
1400 static bool emulation_required(struct kvm_vcpu *vcpu)
1401 {
1402 	return emulate_invalid_guest_state && !guest_state_valid(vcpu);
1403 }
1404 
1405 unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
1406 {
1407 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1408 	unsigned long rflags, save_rflags;
1409 
1410 	if (!kvm_register_is_available(vcpu, VCPU_EXREG_RFLAGS)) {
1411 		kvm_register_mark_available(vcpu, VCPU_EXREG_RFLAGS);
1412 		rflags = vmcs_readl(GUEST_RFLAGS);
1413 		if (vmx->rmode.vm86_active) {
1414 			rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
1415 			save_rflags = vmx->rmode.save_rflags;
1416 			rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
1417 		}
1418 		vmx->rflags = rflags;
1419 	}
1420 	return vmx->rflags;
1421 }
1422 
1423 void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
1424 {
1425 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1426 	unsigned long old_rflags;
1427 
1428 	if (enable_unrestricted_guest) {
1429 		kvm_register_mark_available(vcpu, VCPU_EXREG_RFLAGS);
1430 		vmx->rflags = rflags;
1431 		vmcs_writel(GUEST_RFLAGS, rflags);
1432 		return;
1433 	}
1434 
1435 	old_rflags = vmx_get_rflags(vcpu);
1436 	vmx->rflags = rflags;
1437 	if (vmx->rmode.vm86_active) {
1438 		vmx->rmode.save_rflags = rflags;
1439 		rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
1440 	}
1441 	vmcs_writel(GUEST_RFLAGS, rflags);
1442 
1443 	if ((old_rflags ^ vmx->rflags) & X86_EFLAGS_VM)
1444 		vmx->emulation_required = emulation_required(vcpu);
1445 }
1446 
1447 u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu)
1448 {
1449 	u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1450 	int ret = 0;
1451 
1452 	if (interruptibility & GUEST_INTR_STATE_STI)
1453 		ret |= KVM_X86_SHADOW_INT_STI;
1454 	if (interruptibility & GUEST_INTR_STATE_MOV_SS)
1455 		ret |= KVM_X86_SHADOW_INT_MOV_SS;
1456 
1457 	return ret;
1458 }
1459 
1460 void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
1461 {
1462 	u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1463 	u32 interruptibility = interruptibility_old;
1464 
1465 	interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
1466 
1467 	if (mask & KVM_X86_SHADOW_INT_MOV_SS)
1468 		interruptibility |= GUEST_INTR_STATE_MOV_SS;
1469 	else if (mask & KVM_X86_SHADOW_INT_STI)
1470 		interruptibility |= GUEST_INTR_STATE_STI;
1471 
1472 	if ((interruptibility != interruptibility_old))
1473 		vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
1474 }
1475 
1476 static int vmx_rtit_ctl_check(struct kvm_vcpu *vcpu, u64 data)
1477 {
1478 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1479 	unsigned long value;
1480 
1481 	/*
1482 	 * Any MSR write that attempts to change bits marked reserved will
1483 	 * case a #GP fault.
1484 	 */
1485 	if (data & vmx->pt_desc.ctl_bitmask)
1486 		return 1;
1487 
1488 	/*
1489 	 * Any attempt to modify IA32_RTIT_CTL while TraceEn is set will
1490 	 * result in a #GP unless the same write also clears TraceEn.
1491 	 */
1492 	if ((vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) &&
1493 		((vmx->pt_desc.guest.ctl ^ data) & ~RTIT_CTL_TRACEEN))
1494 		return 1;
1495 
1496 	/*
1497 	 * WRMSR to IA32_RTIT_CTL that sets TraceEn but clears this bit
1498 	 * and FabricEn would cause #GP, if
1499 	 * CPUID.(EAX=14H, ECX=0):ECX.SNGLRGNOUT[bit 2] = 0
1500 	 */
1501 	if ((data & RTIT_CTL_TRACEEN) && !(data & RTIT_CTL_TOPA) &&
1502 		!(data & RTIT_CTL_FABRIC_EN) &&
1503 		!intel_pt_validate_cap(vmx->pt_desc.caps,
1504 					PT_CAP_single_range_output))
1505 		return 1;
1506 
1507 	/*
1508 	 * MTCFreq, CycThresh and PSBFreq encodings check, any MSR write that
1509 	 * utilize encodings marked reserved will casue a #GP fault.
1510 	 */
1511 	value = intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc_periods);
1512 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc) &&
1513 			!test_bit((data & RTIT_CTL_MTC_RANGE) >>
1514 			RTIT_CTL_MTC_RANGE_OFFSET, &value))
1515 		return 1;
1516 	value = intel_pt_validate_cap(vmx->pt_desc.caps,
1517 						PT_CAP_cycle_thresholds);
1518 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc) &&
1519 			!test_bit((data & RTIT_CTL_CYC_THRESH) >>
1520 			RTIT_CTL_CYC_THRESH_OFFSET, &value))
1521 		return 1;
1522 	value = intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_periods);
1523 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc) &&
1524 			!test_bit((data & RTIT_CTL_PSB_FREQ) >>
1525 			RTIT_CTL_PSB_FREQ_OFFSET, &value))
1526 		return 1;
1527 
1528 	/*
1529 	 * If ADDRx_CFG is reserved or the encodings is >2 will
1530 	 * cause a #GP fault.
1531 	 */
1532 	value = (data & RTIT_CTL_ADDR0) >> RTIT_CTL_ADDR0_OFFSET;
1533 	if ((value && (vmx->pt_desc.addr_range < 1)) || (value > 2))
1534 		return 1;
1535 	value = (data & RTIT_CTL_ADDR1) >> RTIT_CTL_ADDR1_OFFSET;
1536 	if ((value && (vmx->pt_desc.addr_range < 2)) || (value > 2))
1537 		return 1;
1538 	value = (data & RTIT_CTL_ADDR2) >> RTIT_CTL_ADDR2_OFFSET;
1539 	if ((value && (vmx->pt_desc.addr_range < 3)) || (value > 2))
1540 		return 1;
1541 	value = (data & RTIT_CTL_ADDR3) >> RTIT_CTL_ADDR3_OFFSET;
1542 	if ((value && (vmx->pt_desc.addr_range < 4)) || (value > 2))
1543 		return 1;
1544 
1545 	return 0;
1546 }
1547 
1548 static int skip_emulated_instruction(struct kvm_vcpu *vcpu)
1549 {
1550 	unsigned long rip;
1551 
1552 	/*
1553 	 * Using VMCS.VM_EXIT_INSTRUCTION_LEN on EPT misconfig depends on
1554 	 * undefined behavior: Intel's SDM doesn't mandate the VMCS field be
1555 	 * set when EPT misconfig occurs.  In practice, real hardware updates
1556 	 * VM_EXIT_INSTRUCTION_LEN on EPT misconfig, but other hypervisors
1557 	 * (namely Hyper-V) don't set it due to it being undefined behavior,
1558 	 * i.e. we end up advancing IP with some random value.
1559 	 */
1560 	if (!static_cpu_has(X86_FEATURE_HYPERVISOR) ||
1561 	    to_vmx(vcpu)->exit_reason != EXIT_REASON_EPT_MISCONFIG) {
1562 		rip = kvm_rip_read(vcpu);
1563 		rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
1564 		kvm_rip_write(vcpu, rip);
1565 	} else {
1566 		if (!kvm_emulate_instruction(vcpu, EMULTYPE_SKIP))
1567 			return 0;
1568 	}
1569 
1570 	/* skipping an emulated instruction also counts */
1571 	vmx_set_interrupt_shadow(vcpu, 0);
1572 
1573 	return 1;
1574 }
1575 
1576 
1577 /*
1578  * Recognizes a pending MTF VM-exit and records the nested state for later
1579  * delivery.
1580  */
1581 static void vmx_update_emulated_instruction(struct kvm_vcpu *vcpu)
1582 {
1583 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1584 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1585 
1586 	if (!is_guest_mode(vcpu))
1587 		return;
1588 
1589 	/*
1590 	 * Per the SDM, MTF takes priority over debug-trap exceptions besides
1591 	 * T-bit traps. As instruction emulation is completed (i.e. at the
1592 	 * instruction boundary), any #DB exception pending delivery must be a
1593 	 * debug-trap. Record the pending MTF state to be delivered in
1594 	 * vmx_check_nested_events().
1595 	 */
1596 	if (nested_cpu_has_mtf(vmcs12) &&
1597 	    (!vcpu->arch.exception.pending ||
1598 	     vcpu->arch.exception.nr == DB_VECTOR))
1599 		vmx->nested.mtf_pending = true;
1600 	else
1601 		vmx->nested.mtf_pending = false;
1602 }
1603 
1604 static int vmx_skip_emulated_instruction(struct kvm_vcpu *vcpu)
1605 {
1606 	vmx_update_emulated_instruction(vcpu);
1607 	return skip_emulated_instruction(vcpu);
1608 }
1609 
1610 static void vmx_clear_hlt(struct kvm_vcpu *vcpu)
1611 {
1612 	/*
1613 	 * Ensure that we clear the HLT state in the VMCS.  We don't need to
1614 	 * explicitly skip the instruction because if the HLT state is set,
1615 	 * then the instruction is already executing and RIP has already been
1616 	 * advanced.
1617 	 */
1618 	if (kvm_hlt_in_guest(vcpu->kvm) &&
1619 			vmcs_read32(GUEST_ACTIVITY_STATE) == GUEST_ACTIVITY_HLT)
1620 		vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
1621 }
1622 
1623 static void vmx_queue_exception(struct kvm_vcpu *vcpu)
1624 {
1625 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1626 	unsigned nr = vcpu->arch.exception.nr;
1627 	bool has_error_code = vcpu->arch.exception.has_error_code;
1628 	u32 error_code = vcpu->arch.exception.error_code;
1629 	u32 intr_info = nr | INTR_INFO_VALID_MASK;
1630 
1631 	kvm_deliver_exception_payload(vcpu);
1632 
1633 	if (has_error_code) {
1634 		vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
1635 		intr_info |= INTR_INFO_DELIVER_CODE_MASK;
1636 	}
1637 
1638 	if (vmx->rmode.vm86_active) {
1639 		int inc_eip = 0;
1640 		if (kvm_exception_is_soft(nr))
1641 			inc_eip = vcpu->arch.event_exit_inst_len;
1642 		kvm_inject_realmode_interrupt(vcpu, nr, inc_eip);
1643 		return;
1644 	}
1645 
1646 	WARN_ON_ONCE(vmx->emulation_required);
1647 
1648 	if (kvm_exception_is_soft(nr)) {
1649 		vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
1650 			     vmx->vcpu.arch.event_exit_inst_len);
1651 		intr_info |= INTR_TYPE_SOFT_EXCEPTION;
1652 	} else
1653 		intr_info |= INTR_TYPE_HARD_EXCEPTION;
1654 
1655 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
1656 
1657 	vmx_clear_hlt(vcpu);
1658 }
1659 
1660 /*
1661  * Swap MSR entry in host/guest MSR entry array.
1662  */
1663 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
1664 {
1665 	struct shared_msr_entry tmp;
1666 
1667 	tmp = vmx->guest_msrs[to];
1668 	vmx->guest_msrs[to] = vmx->guest_msrs[from];
1669 	vmx->guest_msrs[from] = tmp;
1670 }
1671 
1672 /*
1673  * Set up the vmcs to automatically save and restore system
1674  * msrs.  Don't touch the 64-bit msrs if the guest is in legacy
1675  * mode, as fiddling with msrs is very expensive.
1676  */
1677 static void setup_msrs(struct vcpu_vmx *vmx)
1678 {
1679 	int save_nmsrs, index;
1680 
1681 	save_nmsrs = 0;
1682 #ifdef CONFIG_X86_64
1683 	/*
1684 	 * The SYSCALL MSRs are only needed on long mode guests, and only
1685 	 * when EFER.SCE is set.
1686 	 */
1687 	if (is_long_mode(&vmx->vcpu) && (vmx->vcpu.arch.efer & EFER_SCE)) {
1688 		index = __find_msr_index(vmx, MSR_STAR);
1689 		if (index >= 0)
1690 			move_msr_up(vmx, index, save_nmsrs++);
1691 		index = __find_msr_index(vmx, MSR_LSTAR);
1692 		if (index >= 0)
1693 			move_msr_up(vmx, index, save_nmsrs++);
1694 		index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
1695 		if (index >= 0)
1696 			move_msr_up(vmx, index, save_nmsrs++);
1697 	}
1698 #endif
1699 	index = __find_msr_index(vmx, MSR_EFER);
1700 	if (index >= 0 && update_transition_efer(vmx, index))
1701 		move_msr_up(vmx, index, save_nmsrs++);
1702 	index = __find_msr_index(vmx, MSR_TSC_AUX);
1703 	if (index >= 0 && guest_cpuid_has(&vmx->vcpu, X86_FEATURE_RDTSCP))
1704 		move_msr_up(vmx, index, save_nmsrs++);
1705 	index = __find_msr_index(vmx, MSR_IA32_TSX_CTRL);
1706 	if (index >= 0)
1707 		move_msr_up(vmx, index, save_nmsrs++);
1708 
1709 	vmx->save_nmsrs = save_nmsrs;
1710 	vmx->guest_msrs_ready = false;
1711 
1712 	if (cpu_has_vmx_msr_bitmap())
1713 		vmx_update_msr_bitmap(&vmx->vcpu);
1714 }
1715 
1716 static u64 vmx_read_l1_tsc_offset(struct kvm_vcpu *vcpu)
1717 {
1718 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1719 
1720 	if (is_guest_mode(vcpu) &&
1721 	    (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETTING))
1722 		return vcpu->arch.tsc_offset - vmcs12->tsc_offset;
1723 
1724 	return vcpu->arch.tsc_offset;
1725 }
1726 
1727 static u64 vmx_write_l1_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1728 {
1729 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1730 	u64 g_tsc_offset = 0;
1731 
1732 	/*
1733 	 * We're here if L1 chose not to trap WRMSR to TSC. According
1734 	 * to the spec, this should set L1's TSC; The offset that L1
1735 	 * set for L2 remains unchanged, and still needs to be added
1736 	 * to the newly set TSC to get L2's TSC.
1737 	 */
1738 	if (is_guest_mode(vcpu) &&
1739 	    (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETTING))
1740 		g_tsc_offset = vmcs12->tsc_offset;
1741 
1742 	trace_kvm_write_tsc_offset(vcpu->vcpu_id,
1743 				   vcpu->arch.tsc_offset - g_tsc_offset,
1744 				   offset);
1745 	vmcs_write64(TSC_OFFSET, offset + g_tsc_offset);
1746 	return offset + g_tsc_offset;
1747 }
1748 
1749 /*
1750  * nested_vmx_allowed() checks whether a guest should be allowed to use VMX
1751  * instructions and MSRs (i.e., nested VMX). Nested VMX is disabled for
1752  * all guests if the "nested" module option is off, and can also be disabled
1753  * for a single guest by disabling its VMX cpuid bit.
1754  */
1755 bool nested_vmx_allowed(struct kvm_vcpu *vcpu)
1756 {
1757 	return nested && guest_cpuid_has(vcpu, X86_FEATURE_VMX);
1758 }
1759 
1760 static inline bool vmx_feature_control_msr_valid(struct kvm_vcpu *vcpu,
1761 						 uint64_t val)
1762 {
1763 	uint64_t valid_bits = to_vmx(vcpu)->msr_ia32_feature_control_valid_bits;
1764 
1765 	return !(val & ~valid_bits);
1766 }
1767 
1768 static int vmx_get_msr_feature(struct kvm_msr_entry *msr)
1769 {
1770 	switch (msr->index) {
1771 	case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
1772 		if (!nested)
1773 			return 1;
1774 		return vmx_get_vmx_msr(&vmcs_config.nested, msr->index, &msr->data);
1775 	default:
1776 		return 1;
1777 	}
1778 }
1779 
1780 /*
1781  * Reads an msr value (of 'msr_index') into 'pdata'.
1782  * Returns 0 on success, non-0 otherwise.
1783  * Assumes vcpu_load() was already called.
1784  */
1785 static int vmx_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1786 {
1787 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1788 	struct shared_msr_entry *msr;
1789 	u32 index;
1790 
1791 	switch (msr_info->index) {
1792 #ifdef CONFIG_X86_64
1793 	case MSR_FS_BASE:
1794 		msr_info->data = vmcs_readl(GUEST_FS_BASE);
1795 		break;
1796 	case MSR_GS_BASE:
1797 		msr_info->data = vmcs_readl(GUEST_GS_BASE);
1798 		break;
1799 	case MSR_KERNEL_GS_BASE:
1800 		msr_info->data = vmx_read_guest_kernel_gs_base(vmx);
1801 		break;
1802 #endif
1803 	case MSR_EFER:
1804 		return kvm_get_msr_common(vcpu, msr_info);
1805 	case MSR_IA32_TSX_CTRL:
1806 		if (!msr_info->host_initiated &&
1807 		    !(vcpu->arch.arch_capabilities & ARCH_CAP_TSX_CTRL_MSR))
1808 			return 1;
1809 		goto find_shared_msr;
1810 	case MSR_IA32_UMWAIT_CONTROL:
1811 		if (!msr_info->host_initiated && !vmx_has_waitpkg(vmx))
1812 			return 1;
1813 
1814 		msr_info->data = vmx->msr_ia32_umwait_control;
1815 		break;
1816 	case MSR_IA32_SPEC_CTRL:
1817 		if (!msr_info->host_initiated &&
1818 		    !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
1819 			return 1;
1820 
1821 		msr_info->data = to_vmx(vcpu)->spec_ctrl;
1822 		break;
1823 	case MSR_IA32_SYSENTER_CS:
1824 		msr_info->data = vmcs_read32(GUEST_SYSENTER_CS);
1825 		break;
1826 	case MSR_IA32_SYSENTER_EIP:
1827 		msr_info->data = vmcs_readl(GUEST_SYSENTER_EIP);
1828 		break;
1829 	case MSR_IA32_SYSENTER_ESP:
1830 		msr_info->data = vmcs_readl(GUEST_SYSENTER_ESP);
1831 		break;
1832 	case MSR_IA32_BNDCFGS:
1833 		if (!kvm_mpx_supported() ||
1834 		    (!msr_info->host_initiated &&
1835 		     !guest_cpuid_has(vcpu, X86_FEATURE_MPX)))
1836 			return 1;
1837 		msr_info->data = vmcs_read64(GUEST_BNDCFGS);
1838 		break;
1839 	case MSR_IA32_MCG_EXT_CTL:
1840 		if (!msr_info->host_initiated &&
1841 		    !(vmx->msr_ia32_feature_control &
1842 		      FEAT_CTL_LMCE_ENABLED))
1843 			return 1;
1844 		msr_info->data = vcpu->arch.mcg_ext_ctl;
1845 		break;
1846 	case MSR_IA32_FEAT_CTL:
1847 		msr_info->data = vmx->msr_ia32_feature_control;
1848 		break;
1849 	case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
1850 		if (!nested_vmx_allowed(vcpu))
1851 			return 1;
1852 		if (vmx_get_vmx_msr(&vmx->nested.msrs, msr_info->index,
1853 				    &msr_info->data))
1854 			return 1;
1855 		/*
1856 		 * Enlightened VMCS v1 doesn't have certain fields, but buggy
1857 		 * Hyper-V versions are still trying to use corresponding
1858 		 * features when they are exposed. Filter out the essential
1859 		 * minimum.
1860 		 */
1861 		if (!msr_info->host_initiated &&
1862 		    vmx->nested.enlightened_vmcs_enabled)
1863 			nested_evmcs_filter_control_msr(msr_info->index,
1864 							&msr_info->data);
1865 		break;
1866 	case MSR_IA32_RTIT_CTL:
1867 		if (!vmx_pt_mode_is_host_guest())
1868 			return 1;
1869 		msr_info->data = vmx->pt_desc.guest.ctl;
1870 		break;
1871 	case MSR_IA32_RTIT_STATUS:
1872 		if (!vmx_pt_mode_is_host_guest())
1873 			return 1;
1874 		msr_info->data = vmx->pt_desc.guest.status;
1875 		break;
1876 	case MSR_IA32_RTIT_CR3_MATCH:
1877 		if (!vmx_pt_mode_is_host_guest() ||
1878 			!intel_pt_validate_cap(vmx->pt_desc.caps,
1879 						PT_CAP_cr3_filtering))
1880 			return 1;
1881 		msr_info->data = vmx->pt_desc.guest.cr3_match;
1882 		break;
1883 	case MSR_IA32_RTIT_OUTPUT_BASE:
1884 		if (!vmx_pt_mode_is_host_guest() ||
1885 			(!intel_pt_validate_cap(vmx->pt_desc.caps,
1886 					PT_CAP_topa_output) &&
1887 			 !intel_pt_validate_cap(vmx->pt_desc.caps,
1888 					PT_CAP_single_range_output)))
1889 			return 1;
1890 		msr_info->data = vmx->pt_desc.guest.output_base;
1891 		break;
1892 	case MSR_IA32_RTIT_OUTPUT_MASK:
1893 		if (!vmx_pt_mode_is_host_guest() ||
1894 			(!intel_pt_validate_cap(vmx->pt_desc.caps,
1895 					PT_CAP_topa_output) &&
1896 			 !intel_pt_validate_cap(vmx->pt_desc.caps,
1897 					PT_CAP_single_range_output)))
1898 			return 1;
1899 		msr_info->data = vmx->pt_desc.guest.output_mask;
1900 		break;
1901 	case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
1902 		index = msr_info->index - MSR_IA32_RTIT_ADDR0_A;
1903 		if (!vmx_pt_mode_is_host_guest() ||
1904 			(index >= 2 * intel_pt_validate_cap(vmx->pt_desc.caps,
1905 					PT_CAP_num_address_ranges)))
1906 			return 1;
1907 		if (index % 2)
1908 			msr_info->data = vmx->pt_desc.guest.addr_b[index / 2];
1909 		else
1910 			msr_info->data = vmx->pt_desc.guest.addr_a[index / 2];
1911 		break;
1912 	case MSR_TSC_AUX:
1913 		if (!msr_info->host_initiated &&
1914 		    !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP))
1915 			return 1;
1916 		goto find_shared_msr;
1917 	default:
1918 	find_shared_msr:
1919 		msr = find_msr_entry(vmx, msr_info->index);
1920 		if (msr) {
1921 			msr_info->data = msr->data;
1922 			break;
1923 		}
1924 		return kvm_get_msr_common(vcpu, msr_info);
1925 	}
1926 
1927 	return 0;
1928 }
1929 
1930 /*
1931  * Writes msr value into the appropriate "register".
1932  * Returns 0 on success, non-0 otherwise.
1933  * Assumes vcpu_load() was already called.
1934  */
1935 static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1936 {
1937 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1938 	struct shared_msr_entry *msr;
1939 	int ret = 0;
1940 	u32 msr_index = msr_info->index;
1941 	u64 data = msr_info->data;
1942 	u32 index;
1943 
1944 	switch (msr_index) {
1945 	case MSR_EFER:
1946 		ret = kvm_set_msr_common(vcpu, msr_info);
1947 		break;
1948 #ifdef CONFIG_X86_64
1949 	case MSR_FS_BASE:
1950 		vmx_segment_cache_clear(vmx);
1951 		vmcs_writel(GUEST_FS_BASE, data);
1952 		break;
1953 	case MSR_GS_BASE:
1954 		vmx_segment_cache_clear(vmx);
1955 		vmcs_writel(GUEST_GS_BASE, data);
1956 		break;
1957 	case MSR_KERNEL_GS_BASE:
1958 		vmx_write_guest_kernel_gs_base(vmx, data);
1959 		break;
1960 #endif
1961 	case MSR_IA32_SYSENTER_CS:
1962 		if (is_guest_mode(vcpu))
1963 			get_vmcs12(vcpu)->guest_sysenter_cs = data;
1964 		vmcs_write32(GUEST_SYSENTER_CS, data);
1965 		break;
1966 	case MSR_IA32_SYSENTER_EIP:
1967 		if (is_guest_mode(vcpu))
1968 			get_vmcs12(vcpu)->guest_sysenter_eip = data;
1969 		vmcs_writel(GUEST_SYSENTER_EIP, data);
1970 		break;
1971 	case MSR_IA32_SYSENTER_ESP:
1972 		if (is_guest_mode(vcpu))
1973 			get_vmcs12(vcpu)->guest_sysenter_esp = data;
1974 		vmcs_writel(GUEST_SYSENTER_ESP, data);
1975 		break;
1976 	case MSR_IA32_DEBUGCTLMSR:
1977 		if (is_guest_mode(vcpu) && get_vmcs12(vcpu)->vm_exit_controls &
1978 						VM_EXIT_SAVE_DEBUG_CONTROLS)
1979 			get_vmcs12(vcpu)->guest_ia32_debugctl = data;
1980 
1981 		ret = kvm_set_msr_common(vcpu, msr_info);
1982 		break;
1983 
1984 	case MSR_IA32_BNDCFGS:
1985 		if (!kvm_mpx_supported() ||
1986 		    (!msr_info->host_initiated &&
1987 		     !guest_cpuid_has(vcpu, X86_FEATURE_MPX)))
1988 			return 1;
1989 		if (is_noncanonical_address(data & PAGE_MASK, vcpu) ||
1990 		    (data & MSR_IA32_BNDCFGS_RSVD))
1991 			return 1;
1992 		vmcs_write64(GUEST_BNDCFGS, data);
1993 		break;
1994 	case MSR_IA32_UMWAIT_CONTROL:
1995 		if (!msr_info->host_initiated && !vmx_has_waitpkg(vmx))
1996 			return 1;
1997 
1998 		/* The reserved bit 1 and non-32 bit [63:32] should be zero */
1999 		if (data & (BIT_ULL(1) | GENMASK_ULL(63, 32)))
2000 			return 1;
2001 
2002 		vmx->msr_ia32_umwait_control = data;
2003 		break;
2004 	case MSR_IA32_SPEC_CTRL:
2005 		if (!msr_info->host_initiated &&
2006 		    !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
2007 			return 1;
2008 
2009 		if (data & ~kvm_spec_ctrl_valid_bits(vcpu))
2010 			return 1;
2011 
2012 		vmx->spec_ctrl = data;
2013 		if (!data)
2014 			break;
2015 
2016 		/*
2017 		 * For non-nested:
2018 		 * When it's written (to non-zero) for the first time, pass
2019 		 * it through.
2020 		 *
2021 		 * For nested:
2022 		 * The handling of the MSR bitmap for L2 guests is done in
2023 		 * nested_vmx_prepare_msr_bitmap. We should not touch the
2024 		 * vmcs02.msr_bitmap here since it gets completely overwritten
2025 		 * in the merging. We update the vmcs01 here for L1 as well
2026 		 * since it will end up touching the MSR anyway now.
2027 		 */
2028 		vmx_disable_intercept_for_msr(vmx->vmcs01.msr_bitmap,
2029 					      MSR_IA32_SPEC_CTRL,
2030 					      MSR_TYPE_RW);
2031 		break;
2032 	case MSR_IA32_TSX_CTRL:
2033 		if (!msr_info->host_initiated &&
2034 		    !(vcpu->arch.arch_capabilities & ARCH_CAP_TSX_CTRL_MSR))
2035 			return 1;
2036 		if (data & ~(TSX_CTRL_RTM_DISABLE | TSX_CTRL_CPUID_CLEAR))
2037 			return 1;
2038 		goto find_shared_msr;
2039 	case MSR_IA32_PRED_CMD:
2040 		if (!msr_info->host_initiated &&
2041 		    !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
2042 			return 1;
2043 
2044 		if (data & ~PRED_CMD_IBPB)
2045 			return 1;
2046 		if (!boot_cpu_has(X86_FEATURE_SPEC_CTRL))
2047 			return 1;
2048 		if (!data)
2049 			break;
2050 
2051 		wrmsrl(MSR_IA32_PRED_CMD, PRED_CMD_IBPB);
2052 
2053 		/*
2054 		 * For non-nested:
2055 		 * When it's written (to non-zero) for the first time, pass
2056 		 * it through.
2057 		 *
2058 		 * For nested:
2059 		 * The handling of the MSR bitmap for L2 guests is done in
2060 		 * nested_vmx_prepare_msr_bitmap. We should not touch the
2061 		 * vmcs02.msr_bitmap here since it gets completely overwritten
2062 		 * in the merging.
2063 		 */
2064 		vmx_disable_intercept_for_msr(vmx->vmcs01.msr_bitmap, MSR_IA32_PRED_CMD,
2065 					      MSR_TYPE_W);
2066 		break;
2067 	case MSR_IA32_CR_PAT:
2068 		if (!kvm_pat_valid(data))
2069 			return 1;
2070 
2071 		if (is_guest_mode(vcpu) &&
2072 		    get_vmcs12(vcpu)->vm_exit_controls & VM_EXIT_SAVE_IA32_PAT)
2073 			get_vmcs12(vcpu)->guest_ia32_pat = data;
2074 
2075 		if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2076 			vmcs_write64(GUEST_IA32_PAT, data);
2077 			vcpu->arch.pat = data;
2078 			break;
2079 		}
2080 		ret = kvm_set_msr_common(vcpu, msr_info);
2081 		break;
2082 	case MSR_IA32_TSC_ADJUST:
2083 		ret = kvm_set_msr_common(vcpu, msr_info);
2084 		break;
2085 	case MSR_IA32_MCG_EXT_CTL:
2086 		if ((!msr_info->host_initiated &&
2087 		     !(to_vmx(vcpu)->msr_ia32_feature_control &
2088 		       FEAT_CTL_LMCE_ENABLED)) ||
2089 		    (data & ~MCG_EXT_CTL_LMCE_EN))
2090 			return 1;
2091 		vcpu->arch.mcg_ext_ctl = data;
2092 		break;
2093 	case MSR_IA32_FEAT_CTL:
2094 		if (!vmx_feature_control_msr_valid(vcpu, data) ||
2095 		    (to_vmx(vcpu)->msr_ia32_feature_control &
2096 		     FEAT_CTL_LOCKED && !msr_info->host_initiated))
2097 			return 1;
2098 		vmx->msr_ia32_feature_control = data;
2099 		if (msr_info->host_initiated && data == 0)
2100 			vmx_leave_nested(vcpu);
2101 		break;
2102 	case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
2103 		if (!msr_info->host_initiated)
2104 			return 1; /* they are read-only */
2105 		if (!nested_vmx_allowed(vcpu))
2106 			return 1;
2107 		return vmx_set_vmx_msr(vcpu, msr_index, data);
2108 	case MSR_IA32_RTIT_CTL:
2109 		if (!vmx_pt_mode_is_host_guest() ||
2110 			vmx_rtit_ctl_check(vcpu, data) ||
2111 			vmx->nested.vmxon)
2112 			return 1;
2113 		vmcs_write64(GUEST_IA32_RTIT_CTL, data);
2114 		vmx->pt_desc.guest.ctl = data;
2115 		pt_update_intercept_for_msr(vmx);
2116 		break;
2117 	case MSR_IA32_RTIT_STATUS:
2118 		if (!pt_can_write_msr(vmx))
2119 			return 1;
2120 		if (data & MSR_IA32_RTIT_STATUS_MASK)
2121 			return 1;
2122 		vmx->pt_desc.guest.status = data;
2123 		break;
2124 	case MSR_IA32_RTIT_CR3_MATCH:
2125 		if (!pt_can_write_msr(vmx))
2126 			return 1;
2127 		if (!intel_pt_validate_cap(vmx->pt_desc.caps,
2128 					   PT_CAP_cr3_filtering))
2129 			return 1;
2130 		vmx->pt_desc.guest.cr3_match = data;
2131 		break;
2132 	case MSR_IA32_RTIT_OUTPUT_BASE:
2133 		if (!pt_can_write_msr(vmx))
2134 			return 1;
2135 		if (!intel_pt_validate_cap(vmx->pt_desc.caps,
2136 					   PT_CAP_topa_output) &&
2137 		    !intel_pt_validate_cap(vmx->pt_desc.caps,
2138 					   PT_CAP_single_range_output))
2139 			return 1;
2140 		if (data & MSR_IA32_RTIT_OUTPUT_BASE_MASK)
2141 			return 1;
2142 		vmx->pt_desc.guest.output_base = data;
2143 		break;
2144 	case MSR_IA32_RTIT_OUTPUT_MASK:
2145 		if (!pt_can_write_msr(vmx))
2146 			return 1;
2147 		if (!intel_pt_validate_cap(vmx->pt_desc.caps,
2148 					   PT_CAP_topa_output) &&
2149 		    !intel_pt_validate_cap(vmx->pt_desc.caps,
2150 					   PT_CAP_single_range_output))
2151 			return 1;
2152 		vmx->pt_desc.guest.output_mask = data;
2153 		break;
2154 	case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
2155 		if (!pt_can_write_msr(vmx))
2156 			return 1;
2157 		index = msr_info->index - MSR_IA32_RTIT_ADDR0_A;
2158 		if (index >= 2 * intel_pt_validate_cap(vmx->pt_desc.caps,
2159 						       PT_CAP_num_address_ranges))
2160 			return 1;
2161 		if (is_noncanonical_address(data, vcpu))
2162 			return 1;
2163 		if (index % 2)
2164 			vmx->pt_desc.guest.addr_b[index / 2] = data;
2165 		else
2166 			vmx->pt_desc.guest.addr_a[index / 2] = data;
2167 		break;
2168 	case MSR_TSC_AUX:
2169 		if (!msr_info->host_initiated &&
2170 		    !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP))
2171 			return 1;
2172 		/* Check reserved bit, higher 32 bits should be zero */
2173 		if ((data >> 32) != 0)
2174 			return 1;
2175 		goto find_shared_msr;
2176 
2177 	default:
2178 	find_shared_msr:
2179 		msr = find_msr_entry(vmx, msr_index);
2180 		if (msr)
2181 			ret = vmx_set_guest_msr(vmx, msr, data);
2182 		else
2183 			ret = kvm_set_msr_common(vcpu, msr_info);
2184 	}
2185 
2186 	return ret;
2187 }
2188 
2189 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
2190 {
2191 	kvm_register_mark_available(vcpu, reg);
2192 
2193 	switch (reg) {
2194 	case VCPU_REGS_RSP:
2195 		vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
2196 		break;
2197 	case VCPU_REGS_RIP:
2198 		vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
2199 		break;
2200 	case VCPU_EXREG_PDPTR:
2201 		if (enable_ept)
2202 			ept_save_pdptrs(vcpu);
2203 		break;
2204 	case VCPU_EXREG_CR3:
2205 		if (enable_unrestricted_guest || (enable_ept && is_paging(vcpu)))
2206 			vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
2207 		break;
2208 	default:
2209 		WARN_ON_ONCE(1);
2210 		break;
2211 	}
2212 }
2213 
2214 static __init int cpu_has_kvm_support(void)
2215 {
2216 	return cpu_has_vmx();
2217 }
2218 
2219 static __init int vmx_disabled_by_bios(void)
2220 {
2221 	return !boot_cpu_has(X86_FEATURE_MSR_IA32_FEAT_CTL) ||
2222 	       !boot_cpu_has(X86_FEATURE_VMX);
2223 }
2224 
2225 static int kvm_cpu_vmxon(u64 vmxon_pointer)
2226 {
2227 	u64 msr;
2228 
2229 	cr4_set_bits(X86_CR4_VMXE);
2230 	intel_pt_handle_vmx(1);
2231 
2232 	asm_volatile_goto("1: vmxon %[vmxon_pointer]\n\t"
2233 			  _ASM_EXTABLE(1b, %l[fault])
2234 			  : : [vmxon_pointer] "m"(vmxon_pointer)
2235 			  : : fault);
2236 	return 0;
2237 
2238 fault:
2239 	WARN_ONCE(1, "VMXON faulted, MSR_IA32_FEAT_CTL (0x3a) = 0x%llx\n",
2240 		  rdmsrl_safe(MSR_IA32_FEAT_CTL, &msr) ? 0xdeadbeef : msr);
2241 	intel_pt_handle_vmx(0);
2242 	cr4_clear_bits(X86_CR4_VMXE);
2243 
2244 	return -EFAULT;
2245 }
2246 
2247 static int hardware_enable(void)
2248 {
2249 	int cpu = raw_smp_processor_id();
2250 	u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
2251 	int r;
2252 
2253 	if (cr4_read_shadow() & X86_CR4_VMXE)
2254 		return -EBUSY;
2255 
2256 	/*
2257 	 * This can happen if we hot-added a CPU but failed to allocate
2258 	 * VP assist page for it.
2259 	 */
2260 	if (static_branch_unlikely(&enable_evmcs) &&
2261 	    !hv_get_vp_assist_page(cpu))
2262 		return -EFAULT;
2263 
2264 	r = kvm_cpu_vmxon(phys_addr);
2265 	if (r)
2266 		return r;
2267 
2268 	if (enable_ept)
2269 		ept_sync_global();
2270 
2271 	return 0;
2272 }
2273 
2274 static void vmclear_local_loaded_vmcss(void)
2275 {
2276 	int cpu = raw_smp_processor_id();
2277 	struct loaded_vmcs *v, *n;
2278 
2279 	list_for_each_entry_safe(v, n, &per_cpu(loaded_vmcss_on_cpu, cpu),
2280 				 loaded_vmcss_on_cpu_link)
2281 		__loaded_vmcs_clear(v);
2282 }
2283 
2284 
2285 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
2286  * tricks.
2287  */
2288 static void kvm_cpu_vmxoff(void)
2289 {
2290 	asm volatile (__ex("vmxoff"));
2291 
2292 	intel_pt_handle_vmx(0);
2293 	cr4_clear_bits(X86_CR4_VMXE);
2294 }
2295 
2296 static void hardware_disable(void)
2297 {
2298 	vmclear_local_loaded_vmcss();
2299 	kvm_cpu_vmxoff();
2300 }
2301 
2302 /*
2303  * There is no X86_FEATURE for SGX yet, but anyway we need to query CPUID
2304  * directly instead of going through cpu_has(), to ensure KVM is trapping
2305  * ENCLS whenever it's supported in hardware.  It does not matter whether
2306  * the host OS supports or has enabled SGX.
2307  */
2308 static bool cpu_has_sgx(void)
2309 {
2310 	return cpuid_eax(0) >= 0x12 && (cpuid_eax(0x12) & BIT(0));
2311 }
2312 
2313 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
2314 				      u32 msr, u32 *result)
2315 {
2316 	u32 vmx_msr_low, vmx_msr_high;
2317 	u32 ctl = ctl_min | ctl_opt;
2318 
2319 	rdmsr(msr, vmx_msr_low, vmx_msr_high);
2320 
2321 	ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
2322 	ctl |= vmx_msr_low;  /* bit == 1 in low word  ==> must be one  */
2323 
2324 	/* Ensure minimum (required) set of control bits are supported. */
2325 	if (ctl_min & ~ctl)
2326 		return -EIO;
2327 
2328 	*result = ctl;
2329 	return 0;
2330 }
2331 
2332 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf,
2333 				    struct vmx_capability *vmx_cap)
2334 {
2335 	u32 vmx_msr_low, vmx_msr_high;
2336 	u32 min, opt, min2, opt2;
2337 	u32 _pin_based_exec_control = 0;
2338 	u32 _cpu_based_exec_control = 0;
2339 	u32 _cpu_based_2nd_exec_control = 0;
2340 	u32 _vmexit_control = 0;
2341 	u32 _vmentry_control = 0;
2342 
2343 	memset(vmcs_conf, 0, sizeof(*vmcs_conf));
2344 	min = CPU_BASED_HLT_EXITING |
2345 #ifdef CONFIG_X86_64
2346 	      CPU_BASED_CR8_LOAD_EXITING |
2347 	      CPU_BASED_CR8_STORE_EXITING |
2348 #endif
2349 	      CPU_BASED_CR3_LOAD_EXITING |
2350 	      CPU_BASED_CR3_STORE_EXITING |
2351 	      CPU_BASED_UNCOND_IO_EXITING |
2352 	      CPU_BASED_MOV_DR_EXITING |
2353 	      CPU_BASED_USE_TSC_OFFSETTING |
2354 	      CPU_BASED_MWAIT_EXITING |
2355 	      CPU_BASED_MONITOR_EXITING |
2356 	      CPU_BASED_INVLPG_EXITING |
2357 	      CPU_BASED_RDPMC_EXITING;
2358 
2359 	opt = CPU_BASED_TPR_SHADOW |
2360 	      CPU_BASED_USE_MSR_BITMAPS |
2361 	      CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
2362 	if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
2363 				&_cpu_based_exec_control) < 0)
2364 		return -EIO;
2365 #ifdef CONFIG_X86_64
2366 	if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2367 		_cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
2368 					   ~CPU_BASED_CR8_STORE_EXITING;
2369 #endif
2370 	if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
2371 		min2 = 0;
2372 		opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2373 			SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2374 			SECONDARY_EXEC_WBINVD_EXITING |
2375 			SECONDARY_EXEC_ENABLE_VPID |
2376 			SECONDARY_EXEC_ENABLE_EPT |
2377 			SECONDARY_EXEC_UNRESTRICTED_GUEST |
2378 			SECONDARY_EXEC_PAUSE_LOOP_EXITING |
2379 			SECONDARY_EXEC_DESC |
2380 			SECONDARY_EXEC_RDTSCP |
2381 			SECONDARY_EXEC_ENABLE_INVPCID |
2382 			SECONDARY_EXEC_APIC_REGISTER_VIRT |
2383 			SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
2384 			SECONDARY_EXEC_SHADOW_VMCS |
2385 			SECONDARY_EXEC_XSAVES |
2386 			SECONDARY_EXEC_RDSEED_EXITING |
2387 			SECONDARY_EXEC_RDRAND_EXITING |
2388 			SECONDARY_EXEC_ENABLE_PML |
2389 			SECONDARY_EXEC_TSC_SCALING |
2390 			SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE |
2391 			SECONDARY_EXEC_PT_USE_GPA |
2392 			SECONDARY_EXEC_PT_CONCEAL_VMX |
2393 			SECONDARY_EXEC_ENABLE_VMFUNC;
2394 		if (cpu_has_sgx())
2395 			opt2 |= SECONDARY_EXEC_ENCLS_EXITING;
2396 		if (adjust_vmx_controls(min2, opt2,
2397 					MSR_IA32_VMX_PROCBASED_CTLS2,
2398 					&_cpu_based_2nd_exec_control) < 0)
2399 			return -EIO;
2400 	}
2401 #ifndef CONFIG_X86_64
2402 	if (!(_cpu_based_2nd_exec_control &
2403 				SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
2404 		_cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
2405 #endif
2406 
2407 	if (!(_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2408 		_cpu_based_2nd_exec_control &= ~(
2409 				SECONDARY_EXEC_APIC_REGISTER_VIRT |
2410 				SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2411 				SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
2412 
2413 	rdmsr_safe(MSR_IA32_VMX_EPT_VPID_CAP,
2414 		&vmx_cap->ept, &vmx_cap->vpid);
2415 
2416 	if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
2417 		/* CR3 accesses and invlpg don't need to cause VM Exits when EPT
2418 		   enabled */
2419 		_cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
2420 					     CPU_BASED_CR3_STORE_EXITING |
2421 					     CPU_BASED_INVLPG_EXITING);
2422 	} else if (vmx_cap->ept) {
2423 		vmx_cap->ept = 0;
2424 		pr_warn_once("EPT CAP should not exist if not support "
2425 				"1-setting enable EPT VM-execution control\n");
2426 	}
2427 	if (!(_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_VPID) &&
2428 		vmx_cap->vpid) {
2429 		vmx_cap->vpid = 0;
2430 		pr_warn_once("VPID CAP should not exist if not support "
2431 				"1-setting enable VPID VM-execution control\n");
2432 	}
2433 
2434 	min = VM_EXIT_SAVE_DEBUG_CONTROLS | VM_EXIT_ACK_INTR_ON_EXIT;
2435 #ifdef CONFIG_X86_64
2436 	min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
2437 #endif
2438 	opt = VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL |
2439 	      VM_EXIT_LOAD_IA32_PAT |
2440 	      VM_EXIT_LOAD_IA32_EFER |
2441 	      VM_EXIT_CLEAR_BNDCFGS |
2442 	      VM_EXIT_PT_CONCEAL_PIP |
2443 	      VM_EXIT_CLEAR_IA32_RTIT_CTL;
2444 	if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
2445 				&_vmexit_control) < 0)
2446 		return -EIO;
2447 
2448 	min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
2449 	opt = PIN_BASED_VIRTUAL_NMIS | PIN_BASED_POSTED_INTR |
2450 		 PIN_BASED_VMX_PREEMPTION_TIMER;
2451 	if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
2452 				&_pin_based_exec_control) < 0)
2453 		return -EIO;
2454 
2455 	if (cpu_has_broken_vmx_preemption_timer())
2456 		_pin_based_exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
2457 	if (!(_cpu_based_2nd_exec_control &
2458 		SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY))
2459 		_pin_based_exec_control &= ~PIN_BASED_POSTED_INTR;
2460 
2461 	min = VM_ENTRY_LOAD_DEBUG_CONTROLS;
2462 	opt = VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL |
2463 	      VM_ENTRY_LOAD_IA32_PAT |
2464 	      VM_ENTRY_LOAD_IA32_EFER |
2465 	      VM_ENTRY_LOAD_BNDCFGS |
2466 	      VM_ENTRY_PT_CONCEAL_PIP |
2467 	      VM_ENTRY_LOAD_IA32_RTIT_CTL;
2468 	if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
2469 				&_vmentry_control) < 0)
2470 		return -EIO;
2471 
2472 	/*
2473 	 * Some cpus support VM_{ENTRY,EXIT}_IA32_PERF_GLOBAL_CTRL but they
2474 	 * can't be used due to an errata where VM Exit may incorrectly clear
2475 	 * IA32_PERF_GLOBAL_CTRL[34:32].  Workaround the errata by using the
2476 	 * MSR load mechanism to switch IA32_PERF_GLOBAL_CTRL.
2477 	 */
2478 	if (boot_cpu_data.x86 == 0x6) {
2479 		switch (boot_cpu_data.x86_model) {
2480 		case 26: /* AAK155 */
2481 		case 30: /* AAP115 */
2482 		case 37: /* AAT100 */
2483 		case 44: /* BC86,AAY89,BD102 */
2484 		case 46: /* BA97 */
2485 			_vmentry_control &= ~VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL;
2486 			_vmexit_control &= ~VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL;
2487 			pr_warn_once("kvm: VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL "
2488 					"does not work properly. Using workaround\n");
2489 			break;
2490 		default:
2491 			break;
2492 		}
2493 	}
2494 
2495 
2496 	rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
2497 
2498 	/* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
2499 	if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
2500 		return -EIO;
2501 
2502 #ifdef CONFIG_X86_64
2503 	/* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
2504 	if (vmx_msr_high & (1u<<16))
2505 		return -EIO;
2506 #endif
2507 
2508 	/* Require Write-Back (WB) memory type for VMCS accesses. */
2509 	if (((vmx_msr_high >> 18) & 15) != 6)
2510 		return -EIO;
2511 
2512 	vmcs_conf->size = vmx_msr_high & 0x1fff;
2513 	vmcs_conf->order = get_order(vmcs_conf->size);
2514 	vmcs_conf->basic_cap = vmx_msr_high & ~0x1fff;
2515 
2516 	vmcs_conf->revision_id = vmx_msr_low;
2517 
2518 	vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
2519 	vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
2520 	vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
2521 	vmcs_conf->vmexit_ctrl         = _vmexit_control;
2522 	vmcs_conf->vmentry_ctrl        = _vmentry_control;
2523 
2524 	if (static_branch_unlikely(&enable_evmcs))
2525 		evmcs_sanitize_exec_ctrls(vmcs_conf);
2526 
2527 	return 0;
2528 }
2529 
2530 struct vmcs *alloc_vmcs_cpu(bool shadow, int cpu, gfp_t flags)
2531 {
2532 	int node = cpu_to_node(cpu);
2533 	struct page *pages;
2534 	struct vmcs *vmcs;
2535 
2536 	pages = __alloc_pages_node(node, flags, vmcs_config.order);
2537 	if (!pages)
2538 		return NULL;
2539 	vmcs = page_address(pages);
2540 	memset(vmcs, 0, vmcs_config.size);
2541 
2542 	/* KVM supports Enlightened VMCS v1 only */
2543 	if (static_branch_unlikely(&enable_evmcs))
2544 		vmcs->hdr.revision_id = KVM_EVMCS_VERSION;
2545 	else
2546 		vmcs->hdr.revision_id = vmcs_config.revision_id;
2547 
2548 	if (shadow)
2549 		vmcs->hdr.shadow_vmcs = 1;
2550 	return vmcs;
2551 }
2552 
2553 void free_vmcs(struct vmcs *vmcs)
2554 {
2555 	free_pages((unsigned long)vmcs, vmcs_config.order);
2556 }
2557 
2558 /*
2559  * Free a VMCS, but before that VMCLEAR it on the CPU where it was last loaded
2560  */
2561 void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
2562 {
2563 	if (!loaded_vmcs->vmcs)
2564 		return;
2565 	loaded_vmcs_clear(loaded_vmcs);
2566 	free_vmcs(loaded_vmcs->vmcs);
2567 	loaded_vmcs->vmcs = NULL;
2568 	if (loaded_vmcs->msr_bitmap)
2569 		free_page((unsigned long)loaded_vmcs->msr_bitmap);
2570 	WARN_ON(loaded_vmcs->shadow_vmcs != NULL);
2571 }
2572 
2573 int alloc_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
2574 {
2575 	loaded_vmcs->vmcs = alloc_vmcs(false);
2576 	if (!loaded_vmcs->vmcs)
2577 		return -ENOMEM;
2578 
2579 	vmcs_clear(loaded_vmcs->vmcs);
2580 
2581 	loaded_vmcs->shadow_vmcs = NULL;
2582 	loaded_vmcs->hv_timer_soft_disabled = false;
2583 	loaded_vmcs->cpu = -1;
2584 	loaded_vmcs->launched = 0;
2585 
2586 	if (cpu_has_vmx_msr_bitmap()) {
2587 		loaded_vmcs->msr_bitmap = (unsigned long *)
2588 				__get_free_page(GFP_KERNEL_ACCOUNT);
2589 		if (!loaded_vmcs->msr_bitmap)
2590 			goto out_vmcs;
2591 		memset(loaded_vmcs->msr_bitmap, 0xff, PAGE_SIZE);
2592 
2593 		if (IS_ENABLED(CONFIG_HYPERV) &&
2594 		    static_branch_unlikely(&enable_evmcs) &&
2595 		    (ms_hyperv.nested_features & HV_X64_NESTED_MSR_BITMAP)) {
2596 			struct hv_enlightened_vmcs *evmcs =
2597 				(struct hv_enlightened_vmcs *)loaded_vmcs->vmcs;
2598 
2599 			evmcs->hv_enlightenments_control.msr_bitmap = 1;
2600 		}
2601 	}
2602 
2603 	memset(&loaded_vmcs->host_state, 0, sizeof(struct vmcs_host_state));
2604 	memset(&loaded_vmcs->controls_shadow, 0,
2605 		sizeof(struct vmcs_controls_shadow));
2606 
2607 	return 0;
2608 
2609 out_vmcs:
2610 	free_loaded_vmcs(loaded_vmcs);
2611 	return -ENOMEM;
2612 }
2613 
2614 static void free_kvm_area(void)
2615 {
2616 	int cpu;
2617 
2618 	for_each_possible_cpu(cpu) {
2619 		free_vmcs(per_cpu(vmxarea, cpu));
2620 		per_cpu(vmxarea, cpu) = NULL;
2621 	}
2622 }
2623 
2624 static __init int alloc_kvm_area(void)
2625 {
2626 	int cpu;
2627 
2628 	for_each_possible_cpu(cpu) {
2629 		struct vmcs *vmcs;
2630 
2631 		vmcs = alloc_vmcs_cpu(false, cpu, GFP_KERNEL);
2632 		if (!vmcs) {
2633 			free_kvm_area();
2634 			return -ENOMEM;
2635 		}
2636 
2637 		/*
2638 		 * When eVMCS is enabled, alloc_vmcs_cpu() sets
2639 		 * vmcs->revision_id to KVM_EVMCS_VERSION instead of
2640 		 * revision_id reported by MSR_IA32_VMX_BASIC.
2641 		 *
2642 		 * However, even though not explicitly documented by
2643 		 * TLFS, VMXArea passed as VMXON argument should
2644 		 * still be marked with revision_id reported by
2645 		 * physical CPU.
2646 		 */
2647 		if (static_branch_unlikely(&enable_evmcs))
2648 			vmcs->hdr.revision_id = vmcs_config.revision_id;
2649 
2650 		per_cpu(vmxarea, cpu) = vmcs;
2651 	}
2652 	return 0;
2653 }
2654 
2655 static void fix_pmode_seg(struct kvm_vcpu *vcpu, int seg,
2656 		struct kvm_segment *save)
2657 {
2658 	if (!emulate_invalid_guest_state) {
2659 		/*
2660 		 * CS and SS RPL should be equal during guest entry according
2661 		 * to VMX spec, but in reality it is not always so. Since vcpu
2662 		 * is in the middle of the transition from real mode to
2663 		 * protected mode it is safe to assume that RPL 0 is a good
2664 		 * default value.
2665 		 */
2666 		if (seg == VCPU_SREG_CS || seg == VCPU_SREG_SS)
2667 			save->selector &= ~SEGMENT_RPL_MASK;
2668 		save->dpl = save->selector & SEGMENT_RPL_MASK;
2669 		save->s = 1;
2670 	}
2671 	vmx_set_segment(vcpu, save, seg);
2672 }
2673 
2674 static void enter_pmode(struct kvm_vcpu *vcpu)
2675 {
2676 	unsigned long flags;
2677 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2678 
2679 	/*
2680 	 * Update real mode segment cache. It may be not up-to-date if sement
2681 	 * register was written while vcpu was in a guest mode.
2682 	 */
2683 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
2684 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
2685 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
2686 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
2687 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
2688 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
2689 
2690 	vmx->rmode.vm86_active = 0;
2691 
2692 	vmx_set_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
2693 
2694 	flags = vmcs_readl(GUEST_RFLAGS);
2695 	flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
2696 	flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
2697 	vmcs_writel(GUEST_RFLAGS, flags);
2698 
2699 	vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
2700 			(vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
2701 
2702 	update_exception_bitmap(vcpu);
2703 
2704 	fix_pmode_seg(vcpu, VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
2705 	fix_pmode_seg(vcpu, VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
2706 	fix_pmode_seg(vcpu, VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
2707 	fix_pmode_seg(vcpu, VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
2708 	fix_pmode_seg(vcpu, VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
2709 	fix_pmode_seg(vcpu, VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
2710 }
2711 
2712 static void fix_rmode_seg(int seg, struct kvm_segment *save)
2713 {
2714 	const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2715 	struct kvm_segment var = *save;
2716 
2717 	var.dpl = 0x3;
2718 	if (seg == VCPU_SREG_CS)
2719 		var.type = 0x3;
2720 
2721 	if (!emulate_invalid_guest_state) {
2722 		var.selector = var.base >> 4;
2723 		var.base = var.base & 0xffff0;
2724 		var.limit = 0xffff;
2725 		var.g = 0;
2726 		var.db = 0;
2727 		var.present = 1;
2728 		var.s = 1;
2729 		var.l = 0;
2730 		var.unusable = 0;
2731 		var.type = 0x3;
2732 		var.avl = 0;
2733 		if (save->base & 0xf)
2734 			printk_once(KERN_WARNING "kvm: segment base is not "
2735 					"paragraph aligned when entering "
2736 					"protected mode (seg=%d)", seg);
2737 	}
2738 
2739 	vmcs_write16(sf->selector, var.selector);
2740 	vmcs_writel(sf->base, var.base);
2741 	vmcs_write32(sf->limit, var.limit);
2742 	vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(&var));
2743 }
2744 
2745 static void enter_rmode(struct kvm_vcpu *vcpu)
2746 {
2747 	unsigned long flags;
2748 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2749 	struct kvm_vmx *kvm_vmx = to_kvm_vmx(vcpu->kvm);
2750 
2751 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
2752 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
2753 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
2754 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
2755 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
2756 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
2757 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
2758 
2759 	vmx->rmode.vm86_active = 1;
2760 
2761 	/*
2762 	 * Very old userspace does not call KVM_SET_TSS_ADDR before entering
2763 	 * vcpu. Warn the user that an update is overdue.
2764 	 */
2765 	if (!kvm_vmx->tss_addr)
2766 		printk_once(KERN_WARNING "kvm: KVM_SET_TSS_ADDR need to be "
2767 			     "called before entering vcpu\n");
2768 
2769 	vmx_segment_cache_clear(vmx);
2770 
2771 	vmcs_writel(GUEST_TR_BASE, kvm_vmx->tss_addr);
2772 	vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
2773 	vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
2774 
2775 	flags = vmcs_readl(GUEST_RFLAGS);
2776 	vmx->rmode.save_rflags = flags;
2777 
2778 	flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
2779 
2780 	vmcs_writel(GUEST_RFLAGS, flags);
2781 	vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
2782 	update_exception_bitmap(vcpu);
2783 
2784 	fix_rmode_seg(VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
2785 	fix_rmode_seg(VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
2786 	fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
2787 	fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
2788 	fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
2789 	fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
2790 
2791 	kvm_mmu_reset_context(vcpu);
2792 }
2793 
2794 void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
2795 {
2796 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2797 	struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
2798 
2799 	if (!msr)
2800 		return;
2801 
2802 	vcpu->arch.efer = efer;
2803 	if (efer & EFER_LMA) {
2804 		vm_entry_controls_setbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
2805 		msr->data = efer;
2806 	} else {
2807 		vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
2808 
2809 		msr->data = efer & ~EFER_LME;
2810 	}
2811 	setup_msrs(vmx);
2812 }
2813 
2814 #ifdef CONFIG_X86_64
2815 
2816 static void enter_lmode(struct kvm_vcpu *vcpu)
2817 {
2818 	u32 guest_tr_ar;
2819 
2820 	vmx_segment_cache_clear(to_vmx(vcpu));
2821 
2822 	guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
2823 	if ((guest_tr_ar & VMX_AR_TYPE_MASK) != VMX_AR_TYPE_BUSY_64_TSS) {
2824 		pr_debug_ratelimited("%s: tss fixup for long mode. \n",
2825 				     __func__);
2826 		vmcs_write32(GUEST_TR_AR_BYTES,
2827 			     (guest_tr_ar & ~VMX_AR_TYPE_MASK)
2828 			     | VMX_AR_TYPE_BUSY_64_TSS);
2829 	}
2830 	vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
2831 }
2832 
2833 static void exit_lmode(struct kvm_vcpu *vcpu)
2834 {
2835 	vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
2836 	vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
2837 }
2838 
2839 #endif
2840 
2841 static void vmx_flush_tlb_gva(struct kvm_vcpu *vcpu, gva_t addr)
2842 {
2843 	int vpid = to_vmx(vcpu)->vpid;
2844 
2845 	if (!vpid_sync_vcpu_addr(vpid, addr))
2846 		vpid_sync_context(vpid);
2847 
2848 	/*
2849 	 * If VPIDs are not supported or enabled, then the above is a no-op.
2850 	 * But we don't really need a TLB flush in that case anyway, because
2851 	 * each VM entry/exit includes an implicit flush when VPID is 0.
2852 	 */
2853 }
2854 
2855 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
2856 {
2857 	ulong cr0_guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
2858 
2859 	vcpu->arch.cr0 &= ~cr0_guest_owned_bits;
2860 	vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & cr0_guest_owned_bits;
2861 }
2862 
2863 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
2864 {
2865 	ulong cr4_guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
2866 
2867 	vcpu->arch.cr4 &= ~cr4_guest_owned_bits;
2868 	vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & cr4_guest_owned_bits;
2869 }
2870 
2871 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
2872 {
2873 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
2874 
2875 	if (!kvm_register_is_dirty(vcpu, VCPU_EXREG_PDPTR))
2876 		return;
2877 
2878 	if (is_pae_paging(vcpu)) {
2879 		vmcs_write64(GUEST_PDPTR0, mmu->pdptrs[0]);
2880 		vmcs_write64(GUEST_PDPTR1, mmu->pdptrs[1]);
2881 		vmcs_write64(GUEST_PDPTR2, mmu->pdptrs[2]);
2882 		vmcs_write64(GUEST_PDPTR3, mmu->pdptrs[3]);
2883 	}
2884 }
2885 
2886 void ept_save_pdptrs(struct kvm_vcpu *vcpu)
2887 {
2888 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
2889 
2890 	if (is_pae_paging(vcpu)) {
2891 		mmu->pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
2892 		mmu->pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
2893 		mmu->pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
2894 		mmu->pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
2895 	}
2896 
2897 	kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
2898 }
2899 
2900 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
2901 					unsigned long cr0,
2902 					struct kvm_vcpu *vcpu)
2903 {
2904 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2905 
2906 	if (!kvm_register_is_available(vcpu, VCPU_EXREG_CR3))
2907 		vmx_cache_reg(vcpu, VCPU_EXREG_CR3);
2908 	if (!(cr0 & X86_CR0_PG)) {
2909 		/* From paging/starting to nonpaging */
2910 		exec_controls_setbit(vmx, CPU_BASED_CR3_LOAD_EXITING |
2911 					  CPU_BASED_CR3_STORE_EXITING);
2912 		vcpu->arch.cr0 = cr0;
2913 		vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
2914 	} else if (!is_paging(vcpu)) {
2915 		/* From nonpaging to paging */
2916 		exec_controls_clearbit(vmx, CPU_BASED_CR3_LOAD_EXITING |
2917 					    CPU_BASED_CR3_STORE_EXITING);
2918 		vcpu->arch.cr0 = cr0;
2919 		vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
2920 	}
2921 
2922 	if (!(cr0 & X86_CR0_WP))
2923 		*hw_cr0 &= ~X86_CR0_WP;
2924 }
2925 
2926 void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
2927 {
2928 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2929 	unsigned long hw_cr0;
2930 
2931 	hw_cr0 = (cr0 & ~KVM_VM_CR0_ALWAYS_OFF);
2932 	if (enable_unrestricted_guest)
2933 		hw_cr0 |= KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
2934 	else {
2935 		hw_cr0 |= KVM_VM_CR0_ALWAYS_ON;
2936 
2937 		if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
2938 			enter_pmode(vcpu);
2939 
2940 		if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
2941 			enter_rmode(vcpu);
2942 	}
2943 
2944 #ifdef CONFIG_X86_64
2945 	if (vcpu->arch.efer & EFER_LME) {
2946 		if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
2947 			enter_lmode(vcpu);
2948 		if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
2949 			exit_lmode(vcpu);
2950 	}
2951 #endif
2952 
2953 	if (enable_ept && !enable_unrestricted_guest)
2954 		ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
2955 
2956 	vmcs_writel(CR0_READ_SHADOW, cr0);
2957 	vmcs_writel(GUEST_CR0, hw_cr0);
2958 	vcpu->arch.cr0 = cr0;
2959 
2960 	/* depends on vcpu->arch.cr0 to be set to a new value */
2961 	vmx->emulation_required = emulation_required(vcpu);
2962 }
2963 
2964 static int get_ept_level(struct kvm_vcpu *vcpu)
2965 {
2966 	if (is_guest_mode(vcpu) && nested_cpu_has_ept(get_vmcs12(vcpu)))
2967 		return vmx_eptp_page_walk_level(nested_ept_get_eptp(vcpu));
2968 	if (cpu_has_vmx_ept_5levels() && (cpuid_maxphyaddr(vcpu) > 48))
2969 		return 5;
2970 	return 4;
2971 }
2972 
2973 u64 construct_eptp(struct kvm_vcpu *vcpu, unsigned long root_hpa)
2974 {
2975 	u64 eptp = VMX_EPTP_MT_WB;
2976 
2977 	eptp |= (get_ept_level(vcpu) == 5) ? VMX_EPTP_PWL_5 : VMX_EPTP_PWL_4;
2978 
2979 	if (enable_ept_ad_bits &&
2980 	    (!is_guest_mode(vcpu) || nested_ept_ad_enabled(vcpu)))
2981 		eptp |= VMX_EPTP_AD_ENABLE_BIT;
2982 	eptp |= (root_hpa & PAGE_MASK);
2983 
2984 	return eptp;
2985 }
2986 
2987 void vmx_load_mmu_pgd(struct kvm_vcpu *vcpu, unsigned long cr3)
2988 {
2989 	struct kvm *kvm = vcpu->kvm;
2990 	bool update_guest_cr3 = true;
2991 	unsigned long guest_cr3;
2992 	u64 eptp;
2993 
2994 	guest_cr3 = cr3;
2995 	if (enable_ept) {
2996 		eptp = construct_eptp(vcpu, cr3);
2997 		vmcs_write64(EPT_POINTER, eptp);
2998 
2999 		if (kvm_x86_ops.tlb_remote_flush) {
3000 			spin_lock(&to_kvm_vmx(kvm)->ept_pointer_lock);
3001 			to_vmx(vcpu)->ept_pointer = eptp;
3002 			to_kvm_vmx(kvm)->ept_pointers_match
3003 				= EPT_POINTERS_CHECK;
3004 			spin_unlock(&to_kvm_vmx(kvm)->ept_pointer_lock);
3005 		}
3006 
3007 		/* Loading vmcs02.GUEST_CR3 is handled by nested VM-Enter. */
3008 		if (is_guest_mode(vcpu))
3009 			update_guest_cr3 = false;
3010 		else if (!enable_unrestricted_guest && !is_paging(vcpu))
3011 			guest_cr3 = to_kvm_vmx(kvm)->ept_identity_map_addr;
3012 		else if (test_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail))
3013 			guest_cr3 = vcpu->arch.cr3;
3014 		else /* vmcs01.GUEST_CR3 is already up-to-date. */
3015 			update_guest_cr3 = false;
3016 		ept_load_pdptrs(vcpu);
3017 	}
3018 
3019 	if (update_guest_cr3)
3020 		vmcs_writel(GUEST_CR3, guest_cr3);
3021 }
3022 
3023 int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
3024 {
3025 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3026 	/*
3027 	 * Pass through host's Machine Check Enable value to hw_cr4, which
3028 	 * is in force while we are in guest mode.  Do not let guests control
3029 	 * this bit, even if host CR4.MCE == 0.
3030 	 */
3031 	unsigned long hw_cr4;
3032 
3033 	hw_cr4 = (cr4_read_shadow() & X86_CR4_MCE) | (cr4 & ~X86_CR4_MCE);
3034 	if (enable_unrestricted_guest)
3035 		hw_cr4 |= KVM_VM_CR4_ALWAYS_ON_UNRESTRICTED_GUEST;
3036 	else if (vmx->rmode.vm86_active)
3037 		hw_cr4 |= KVM_RMODE_VM_CR4_ALWAYS_ON;
3038 	else
3039 		hw_cr4 |= KVM_PMODE_VM_CR4_ALWAYS_ON;
3040 
3041 	if (!boot_cpu_has(X86_FEATURE_UMIP) && vmx_umip_emulated()) {
3042 		if (cr4 & X86_CR4_UMIP) {
3043 			secondary_exec_controls_setbit(vmx, SECONDARY_EXEC_DESC);
3044 			hw_cr4 &= ~X86_CR4_UMIP;
3045 		} else if (!is_guest_mode(vcpu) ||
3046 			!nested_cpu_has2(get_vmcs12(vcpu), SECONDARY_EXEC_DESC)) {
3047 			secondary_exec_controls_clearbit(vmx, SECONDARY_EXEC_DESC);
3048 		}
3049 	}
3050 
3051 	if (cr4 & X86_CR4_VMXE) {
3052 		/*
3053 		 * To use VMXON (and later other VMX instructions), a guest
3054 		 * must first be able to turn on cr4.VMXE (see handle_vmon()).
3055 		 * So basically the check on whether to allow nested VMX
3056 		 * is here.  We operate under the default treatment of SMM,
3057 		 * so VMX cannot be enabled under SMM.
3058 		 */
3059 		if (!nested_vmx_allowed(vcpu) || is_smm(vcpu))
3060 			return 1;
3061 	}
3062 
3063 	if (vmx->nested.vmxon && !nested_cr4_valid(vcpu, cr4))
3064 		return 1;
3065 
3066 	vcpu->arch.cr4 = cr4;
3067 
3068 	if (!enable_unrestricted_guest) {
3069 		if (enable_ept) {
3070 			if (!is_paging(vcpu)) {
3071 				hw_cr4 &= ~X86_CR4_PAE;
3072 				hw_cr4 |= X86_CR4_PSE;
3073 			} else if (!(cr4 & X86_CR4_PAE)) {
3074 				hw_cr4 &= ~X86_CR4_PAE;
3075 			}
3076 		}
3077 
3078 		/*
3079 		 * SMEP/SMAP/PKU is disabled if CPU is in non-paging mode in
3080 		 * hardware.  To emulate this behavior, SMEP/SMAP/PKU needs
3081 		 * to be manually disabled when guest switches to non-paging
3082 		 * mode.
3083 		 *
3084 		 * If !enable_unrestricted_guest, the CPU is always running
3085 		 * with CR0.PG=1 and CR4 needs to be modified.
3086 		 * If enable_unrestricted_guest, the CPU automatically
3087 		 * disables SMEP/SMAP/PKU when the guest sets CR0.PG=0.
3088 		 */
3089 		if (!is_paging(vcpu))
3090 			hw_cr4 &= ~(X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE);
3091 	}
3092 
3093 	vmcs_writel(CR4_READ_SHADOW, cr4);
3094 	vmcs_writel(GUEST_CR4, hw_cr4);
3095 	return 0;
3096 }
3097 
3098 void vmx_get_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg)
3099 {
3100 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3101 	u32 ar;
3102 
3103 	if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3104 		*var = vmx->rmode.segs[seg];
3105 		if (seg == VCPU_SREG_TR
3106 		    || var->selector == vmx_read_guest_seg_selector(vmx, seg))
3107 			return;
3108 		var->base = vmx_read_guest_seg_base(vmx, seg);
3109 		var->selector = vmx_read_guest_seg_selector(vmx, seg);
3110 		return;
3111 	}
3112 	var->base = vmx_read_guest_seg_base(vmx, seg);
3113 	var->limit = vmx_read_guest_seg_limit(vmx, seg);
3114 	var->selector = vmx_read_guest_seg_selector(vmx, seg);
3115 	ar = vmx_read_guest_seg_ar(vmx, seg);
3116 	var->unusable = (ar >> 16) & 1;
3117 	var->type = ar & 15;
3118 	var->s = (ar >> 4) & 1;
3119 	var->dpl = (ar >> 5) & 3;
3120 	/*
3121 	 * Some userspaces do not preserve unusable property. Since usable
3122 	 * segment has to be present according to VMX spec we can use present
3123 	 * property to amend userspace bug by making unusable segment always
3124 	 * nonpresent. vmx_segment_access_rights() already marks nonpresent
3125 	 * segment as unusable.
3126 	 */
3127 	var->present = !var->unusable;
3128 	var->avl = (ar >> 12) & 1;
3129 	var->l = (ar >> 13) & 1;
3130 	var->db = (ar >> 14) & 1;
3131 	var->g = (ar >> 15) & 1;
3132 }
3133 
3134 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
3135 {
3136 	struct kvm_segment s;
3137 
3138 	if (to_vmx(vcpu)->rmode.vm86_active) {
3139 		vmx_get_segment(vcpu, &s, seg);
3140 		return s.base;
3141 	}
3142 	return vmx_read_guest_seg_base(to_vmx(vcpu), seg);
3143 }
3144 
3145 int vmx_get_cpl(struct kvm_vcpu *vcpu)
3146 {
3147 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3148 
3149 	if (unlikely(vmx->rmode.vm86_active))
3150 		return 0;
3151 	else {
3152 		int ar = vmx_read_guest_seg_ar(vmx, VCPU_SREG_SS);
3153 		return VMX_AR_DPL(ar);
3154 	}
3155 }
3156 
3157 static u32 vmx_segment_access_rights(struct kvm_segment *var)
3158 {
3159 	u32 ar;
3160 
3161 	if (var->unusable || !var->present)
3162 		ar = 1 << 16;
3163 	else {
3164 		ar = var->type & 15;
3165 		ar |= (var->s & 1) << 4;
3166 		ar |= (var->dpl & 3) << 5;
3167 		ar |= (var->present & 1) << 7;
3168 		ar |= (var->avl & 1) << 12;
3169 		ar |= (var->l & 1) << 13;
3170 		ar |= (var->db & 1) << 14;
3171 		ar |= (var->g & 1) << 15;
3172 	}
3173 
3174 	return ar;
3175 }
3176 
3177 void vmx_set_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg)
3178 {
3179 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3180 	const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3181 
3182 	vmx_segment_cache_clear(vmx);
3183 
3184 	if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3185 		vmx->rmode.segs[seg] = *var;
3186 		if (seg == VCPU_SREG_TR)
3187 			vmcs_write16(sf->selector, var->selector);
3188 		else if (var->s)
3189 			fix_rmode_seg(seg, &vmx->rmode.segs[seg]);
3190 		goto out;
3191 	}
3192 
3193 	vmcs_writel(sf->base, var->base);
3194 	vmcs_write32(sf->limit, var->limit);
3195 	vmcs_write16(sf->selector, var->selector);
3196 
3197 	/*
3198 	 *   Fix the "Accessed" bit in AR field of segment registers for older
3199 	 * qemu binaries.
3200 	 *   IA32 arch specifies that at the time of processor reset the
3201 	 * "Accessed" bit in the AR field of segment registers is 1. And qemu
3202 	 * is setting it to 0 in the userland code. This causes invalid guest
3203 	 * state vmexit when "unrestricted guest" mode is turned on.
3204 	 *    Fix for this setup issue in cpu_reset is being pushed in the qemu
3205 	 * tree. Newer qemu binaries with that qemu fix would not need this
3206 	 * kvm hack.
3207 	 */
3208 	if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
3209 		var->type |= 0x1; /* Accessed */
3210 
3211 	vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(var));
3212 
3213 out:
3214 	vmx->emulation_required = emulation_required(vcpu);
3215 }
3216 
3217 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3218 {
3219 	u32 ar = vmx_read_guest_seg_ar(to_vmx(vcpu), VCPU_SREG_CS);
3220 
3221 	*db = (ar >> 14) & 1;
3222 	*l = (ar >> 13) & 1;
3223 }
3224 
3225 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3226 {
3227 	dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
3228 	dt->address = vmcs_readl(GUEST_IDTR_BASE);
3229 }
3230 
3231 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3232 {
3233 	vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
3234 	vmcs_writel(GUEST_IDTR_BASE, dt->address);
3235 }
3236 
3237 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3238 {
3239 	dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
3240 	dt->address = vmcs_readl(GUEST_GDTR_BASE);
3241 }
3242 
3243 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3244 {
3245 	vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
3246 	vmcs_writel(GUEST_GDTR_BASE, dt->address);
3247 }
3248 
3249 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
3250 {
3251 	struct kvm_segment var;
3252 	u32 ar;
3253 
3254 	vmx_get_segment(vcpu, &var, seg);
3255 	var.dpl = 0x3;
3256 	if (seg == VCPU_SREG_CS)
3257 		var.type = 0x3;
3258 	ar = vmx_segment_access_rights(&var);
3259 
3260 	if (var.base != (var.selector << 4))
3261 		return false;
3262 	if (var.limit != 0xffff)
3263 		return false;
3264 	if (ar != 0xf3)
3265 		return false;
3266 
3267 	return true;
3268 }
3269 
3270 static bool code_segment_valid(struct kvm_vcpu *vcpu)
3271 {
3272 	struct kvm_segment cs;
3273 	unsigned int cs_rpl;
3274 
3275 	vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3276 	cs_rpl = cs.selector & SEGMENT_RPL_MASK;
3277 
3278 	if (cs.unusable)
3279 		return false;
3280 	if (~cs.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_ACCESSES_MASK))
3281 		return false;
3282 	if (!cs.s)
3283 		return false;
3284 	if (cs.type & VMX_AR_TYPE_WRITEABLE_MASK) {
3285 		if (cs.dpl > cs_rpl)
3286 			return false;
3287 	} else {
3288 		if (cs.dpl != cs_rpl)
3289 			return false;
3290 	}
3291 	if (!cs.present)
3292 		return false;
3293 
3294 	/* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
3295 	return true;
3296 }
3297 
3298 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
3299 {
3300 	struct kvm_segment ss;
3301 	unsigned int ss_rpl;
3302 
3303 	vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3304 	ss_rpl = ss.selector & SEGMENT_RPL_MASK;
3305 
3306 	if (ss.unusable)
3307 		return true;
3308 	if (ss.type != 3 && ss.type != 7)
3309 		return false;
3310 	if (!ss.s)
3311 		return false;
3312 	if (ss.dpl != ss_rpl) /* DPL != RPL */
3313 		return false;
3314 	if (!ss.present)
3315 		return false;
3316 
3317 	return true;
3318 }
3319 
3320 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
3321 {
3322 	struct kvm_segment var;
3323 	unsigned int rpl;
3324 
3325 	vmx_get_segment(vcpu, &var, seg);
3326 	rpl = var.selector & SEGMENT_RPL_MASK;
3327 
3328 	if (var.unusable)
3329 		return true;
3330 	if (!var.s)
3331 		return false;
3332 	if (!var.present)
3333 		return false;
3334 	if (~var.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_WRITEABLE_MASK)) {
3335 		if (var.dpl < rpl) /* DPL < RPL */
3336 			return false;
3337 	}
3338 
3339 	/* TODO: Add other members to kvm_segment_field to allow checking for other access
3340 	 * rights flags
3341 	 */
3342 	return true;
3343 }
3344 
3345 static bool tr_valid(struct kvm_vcpu *vcpu)
3346 {
3347 	struct kvm_segment tr;
3348 
3349 	vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
3350 
3351 	if (tr.unusable)
3352 		return false;
3353 	if (tr.selector & SEGMENT_TI_MASK)	/* TI = 1 */
3354 		return false;
3355 	if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
3356 		return false;
3357 	if (!tr.present)
3358 		return false;
3359 
3360 	return true;
3361 }
3362 
3363 static bool ldtr_valid(struct kvm_vcpu *vcpu)
3364 {
3365 	struct kvm_segment ldtr;
3366 
3367 	vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
3368 
3369 	if (ldtr.unusable)
3370 		return true;
3371 	if (ldtr.selector & SEGMENT_TI_MASK)	/* TI = 1 */
3372 		return false;
3373 	if (ldtr.type != 2)
3374 		return false;
3375 	if (!ldtr.present)
3376 		return false;
3377 
3378 	return true;
3379 }
3380 
3381 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
3382 {
3383 	struct kvm_segment cs, ss;
3384 
3385 	vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3386 	vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3387 
3388 	return ((cs.selector & SEGMENT_RPL_MASK) ==
3389 		 (ss.selector & SEGMENT_RPL_MASK));
3390 }
3391 
3392 /*
3393  * Check if guest state is valid. Returns true if valid, false if
3394  * not.
3395  * We assume that registers are always usable
3396  */
3397 static bool guest_state_valid(struct kvm_vcpu *vcpu)
3398 {
3399 	if (enable_unrestricted_guest)
3400 		return true;
3401 
3402 	/* real mode guest state checks */
3403 	if (!is_protmode(vcpu) || (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
3404 		if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
3405 			return false;
3406 		if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
3407 			return false;
3408 		if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
3409 			return false;
3410 		if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
3411 			return false;
3412 		if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
3413 			return false;
3414 		if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
3415 			return false;
3416 	} else {
3417 	/* protected mode guest state checks */
3418 		if (!cs_ss_rpl_check(vcpu))
3419 			return false;
3420 		if (!code_segment_valid(vcpu))
3421 			return false;
3422 		if (!stack_segment_valid(vcpu))
3423 			return false;
3424 		if (!data_segment_valid(vcpu, VCPU_SREG_DS))
3425 			return false;
3426 		if (!data_segment_valid(vcpu, VCPU_SREG_ES))
3427 			return false;
3428 		if (!data_segment_valid(vcpu, VCPU_SREG_FS))
3429 			return false;
3430 		if (!data_segment_valid(vcpu, VCPU_SREG_GS))
3431 			return false;
3432 		if (!tr_valid(vcpu))
3433 			return false;
3434 		if (!ldtr_valid(vcpu))
3435 			return false;
3436 	}
3437 	/* TODO:
3438 	 * - Add checks on RIP
3439 	 * - Add checks on RFLAGS
3440 	 */
3441 
3442 	return true;
3443 }
3444 
3445 static int init_rmode_tss(struct kvm *kvm)
3446 {
3447 	gfn_t fn;
3448 	u16 data = 0;
3449 	int idx, r;
3450 
3451 	idx = srcu_read_lock(&kvm->srcu);
3452 	fn = to_kvm_vmx(kvm)->tss_addr >> PAGE_SHIFT;
3453 	r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3454 	if (r < 0)
3455 		goto out;
3456 	data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
3457 	r = kvm_write_guest_page(kvm, fn++, &data,
3458 			TSS_IOPB_BASE_OFFSET, sizeof(u16));
3459 	if (r < 0)
3460 		goto out;
3461 	r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
3462 	if (r < 0)
3463 		goto out;
3464 	r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3465 	if (r < 0)
3466 		goto out;
3467 	data = ~0;
3468 	r = kvm_write_guest_page(kvm, fn, &data,
3469 				 RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
3470 				 sizeof(u8));
3471 out:
3472 	srcu_read_unlock(&kvm->srcu, idx);
3473 	return r;
3474 }
3475 
3476 static int init_rmode_identity_map(struct kvm *kvm)
3477 {
3478 	struct kvm_vmx *kvm_vmx = to_kvm_vmx(kvm);
3479 	int i, r = 0;
3480 	kvm_pfn_t identity_map_pfn;
3481 	u32 tmp;
3482 
3483 	/* Protect kvm_vmx->ept_identity_pagetable_done. */
3484 	mutex_lock(&kvm->slots_lock);
3485 
3486 	if (likely(kvm_vmx->ept_identity_pagetable_done))
3487 		goto out;
3488 
3489 	if (!kvm_vmx->ept_identity_map_addr)
3490 		kvm_vmx->ept_identity_map_addr = VMX_EPT_IDENTITY_PAGETABLE_ADDR;
3491 	identity_map_pfn = kvm_vmx->ept_identity_map_addr >> PAGE_SHIFT;
3492 
3493 	r = __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
3494 				    kvm_vmx->ept_identity_map_addr, PAGE_SIZE);
3495 	if (r < 0)
3496 		goto out;
3497 
3498 	r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
3499 	if (r < 0)
3500 		goto out;
3501 	/* Set up identity-mapping pagetable for EPT in real mode */
3502 	for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
3503 		tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
3504 			_PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
3505 		r = kvm_write_guest_page(kvm, identity_map_pfn,
3506 				&tmp, i * sizeof(tmp), sizeof(tmp));
3507 		if (r < 0)
3508 			goto out;
3509 	}
3510 	kvm_vmx->ept_identity_pagetable_done = true;
3511 
3512 out:
3513 	mutex_unlock(&kvm->slots_lock);
3514 	return r;
3515 }
3516 
3517 static void seg_setup(int seg)
3518 {
3519 	const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3520 	unsigned int ar;
3521 
3522 	vmcs_write16(sf->selector, 0);
3523 	vmcs_writel(sf->base, 0);
3524 	vmcs_write32(sf->limit, 0xffff);
3525 	ar = 0x93;
3526 	if (seg == VCPU_SREG_CS)
3527 		ar |= 0x08; /* code segment */
3528 
3529 	vmcs_write32(sf->ar_bytes, ar);
3530 }
3531 
3532 static int alloc_apic_access_page(struct kvm *kvm)
3533 {
3534 	struct page *page;
3535 	int r = 0;
3536 
3537 	mutex_lock(&kvm->slots_lock);
3538 	if (kvm->arch.apic_access_page_done)
3539 		goto out;
3540 	r = __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
3541 				    APIC_DEFAULT_PHYS_BASE, PAGE_SIZE);
3542 	if (r)
3543 		goto out;
3544 
3545 	page = gfn_to_page(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
3546 	if (is_error_page(page)) {
3547 		r = -EFAULT;
3548 		goto out;
3549 	}
3550 
3551 	/*
3552 	 * Do not pin the page in memory, so that memory hot-unplug
3553 	 * is able to migrate it.
3554 	 */
3555 	put_page(page);
3556 	kvm->arch.apic_access_page_done = true;
3557 out:
3558 	mutex_unlock(&kvm->slots_lock);
3559 	return r;
3560 }
3561 
3562 int allocate_vpid(void)
3563 {
3564 	int vpid;
3565 
3566 	if (!enable_vpid)
3567 		return 0;
3568 	spin_lock(&vmx_vpid_lock);
3569 	vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
3570 	if (vpid < VMX_NR_VPIDS)
3571 		__set_bit(vpid, vmx_vpid_bitmap);
3572 	else
3573 		vpid = 0;
3574 	spin_unlock(&vmx_vpid_lock);
3575 	return vpid;
3576 }
3577 
3578 void free_vpid(int vpid)
3579 {
3580 	if (!enable_vpid || vpid == 0)
3581 		return;
3582 	spin_lock(&vmx_vpid_lock);
3583 	__clear_bit(vpid, vmx_vpid_bitmap);
3584 	spin_unlock(&vmx_vpid_lock);
3585 }
3586 
3587 static __always_inline void vmx_disable_intercept_for_msr(unsigned long *msr_bitmap,
3588 							  u32 msr, int type)
3589 {
3590 	int f = sizeof(unsigned long);
3591 
3592 	if (!cpu_has_vmx_msr_bitmap())
3593 		return;
3594 
3595 	if (static_branch_unlikely(&enable_evmcs))
3596 		evmcs_touch_msr_bitmap();
3597 
3598 	/*
3599 	 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
3600 	 * have the write-low and read-high bitmap offsets the wrong way round.
3601 	 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
3602 	 */
3603 	if (msr <= 0x1fff) {
3604 		if (type & MSR_TYPE_R)
3605 			/* read-low */
3606 			__clear_bit(msr, msr_bitmap + 0x000 / f);
3607 
3608 		if (type & MSR_TYPE_W)
3609 			/* write-low */
3610 			__clear_bit(msr, msr_bitmap + 0x800 / f);
3611 
3612 	} else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
3613 		msr &= 0x1fff;
3614 		if (type & MSR_TYPE_R)
3615 			/* read-high */
3616 			__clear_bit(msr, msr_bitmap + 0x400 / f);
3617 
3618 		if (type & MSR_TYPE_W)
3619 			/* write-high */
3620 			__clear_bit(msr, msr_bitmap + 0xc00 / f);
3621 
3622 	}
3623 }
3624 
3625 static __always_inline void vmx_enable_intercept_for_msr(unsigned long *msr_bitmap,
3626 							 u32 msr, int type)
3627 {
3628 	int f = sizeof(unsigned long);
3629 
3630 	if (!cpu_has_vmx_msr_bitmap())
3631 		return;
3632 
3633 	if (static_branch_unlikely(&enable_evmcs))
3634 		evmcs_touch_msr_bitmap();
3635 
3636 	/*
3637 	 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
3638 	 * have the write-low and read-high bitmap offsets the wrong way round.
3639 	 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
3640 	 */
3641 	if (msr <= 0x1fff) {
3642 		if (type & MSR_TYPE_R)
3643 			/* read-low */
3644 			__set_bit(msr, msr_bitmap + 0x000 / f);
3645 
3646 		if (type & MSR_TYPE_W)
3647 			/* write-low */
3648 			__set_bit(msr, msr_bitmap + 0x800 / f);
3649 
3650 	} else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
3651 		msr &= 0x1fff;
3652 		if (type & MSR_TYPE_R)
3653 			/* read-high */
3654 			__set_bit(msr, msr_bitmap + 0x400 / f);
3655 
3656 		if (type & MSR_TYPE_W)
3657 			/* write-high */
3658 			__set_bit(msr, msr_bitmap + 0xc00 / f);
3659 
3660 	}
3661 }
3662 
3663 static __always_inline void vmx_set_intercept_for_msr(unsigned long *msr_bitmap,
3664 			     			      u32 msr, int type, bool value)
3665 {
3666 	if (value)
3667 		vmx_enable_intercept_for_msr(msr_bitmap, msr, type);
3668 	else
3669 		vmx_disable_intercept_for_msr(msr_bitmap, msr, type);
3670 }
3671 
3672 static u8 vmx_msr_bitmap_mode(struct kvm_vcpu *vcpu)
3673 {
3674 	u8 mode = 0;
3675 
3676 	if (cpu_has_secondary_exec_ctrls() &&
3677 	    (secondary_exec_controls_get(to_vmx(vcpu)) &
3678 	     SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE)) {
3679 		mode |= MSR_BITMAP_MODE_X2APIC;
3680 		if (enable_apicv && kvm_vcpu_apicv_active(vcpu))
3681 			mode |= MSR_BITMAP_MODE_X2APIC_APICV;
3682 	}
3683 
3684 	return mode;
3685 }
3686 
3687 static void vmx_update_msr_bitmap_x2apic(unsigned long *msr_bitmap,
3688 					 u8 mode)
3689 {
3690 	int msr;
3691 
3692 	for (msr = 0x800; msr <= 0x8ff; msr += BITS_PER_LONG) {
3693 		unsigned word = msr / BITS_PER_LONG;
3694 		msr_bitmap[word] = (mode & MSR_BITMAP_MODE_X2APIC_APICV) ? 0 : ~0;
3695 		msr_bitmap[word + (0x800 / sizeof(long))] = ~0;
3696 	}
3697 
3698 	if (mode & MSR_BITMAP_MODE_X2APIC) {
3699 		/*
3700 		 * TPR reads and writes can be virtualized even if virtual interrupt
3701 		 * delivery is not in use.
3702 		 */
3703 		vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_TASKPRI), MSR_TYPE_RW);
3704 		if (mode & MSR_BITMAP_MODE_X2APIC_APICV) {
3705 			vmx_enable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_TMCCT), MSR_TYPE_R);
3706 			vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_EOI), MSR_TYPE_W);
3707 			vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_SELF_IPI), MSR_TYPE_W);
3708 		}
3709 	}
3710 }
3711 
3712 void vmx_update_msr_bitmap(struct kvm_vcpu *vcpu)
3713 {
3714 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3715 	unsigned long *msr_bitmap = vmx->vmcs01.msr_bitmap;
3716 	u8 mode = vmx_msr_bitmap_mode(vcpu);
3717 	u8 changed = mode ^ vmx->msr_bitmap_mode;
3718 
3719 	if (!changed)
3720 		return;
3721 
3722 	if (changed & (MSR_BITMAP_MODE_X2APIC | MSR_BITMAP_MODE_X2APIC_APICV))
3723 		vmx_update_msr_bitmap_x2apic(msr_bitmap, mode);
3724 
3725 	vmx->msr_bitmap_mode = mode;
3726 }
3727 
3728 void pt_update_intercept_for_msr(struct vcpu_vmx *vmx)
3729 {
3730 	unsigned long *msr_bitmap = vmx->vmcs01.msr_bitmap;
3731 	bool flag = !(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN);
3732 	u32 i;
3733 
3734 	vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_STATUS,
3735 							MSR_TYPE_RW, flag);
3736 	vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_OUTPUT_BASE,
3737 							MSR_TYPE_RW, flag);
3738 	vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_OUTPUT_MASK,
3739 							MSR_TYPE_RW, flag);
3740 	vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_CR3_MATCH,
3741 							MSR_TYPE_RW, flag);
3742 	for (i = 0; i < vmx->pt_desc.addr_range; i++) {
3743 		vmx_set_intercept_for_msr(msr_bitmap,
3744 			MSR_IA32_RTIT_ADDR0_A + i * 2, MSR_TYPE_RW, flag);
3745 		vmx_set_intercept_for_msr(msr_bitmap,
3746 			MSR_IA32_RTIT_ADDR0_B + i * 2, MSR_TYPE_RW, flag);
3747 	}
3748 }
3749 
3750 static bool vmx_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
3751 {
3752 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3753 	void *vapic_page;
3754 	u32 vppr;
3755 	int rvi;
3756 
3757 	if (WARN_ON_ONCE(!is_guest_mode(vcpu)) ||
3758 		!nested_cpu_has_vid(get_vmcs12(vcpu)) ||
3759 		WARN_ON_ONCE(!vmx->nested.virtual_apic_map.gfn))
3760 		return false;
3761 
3762 	rvi = vmx_get_rvi();
3763 
3764 	vapic_page = vmx->nested.virtual_apic_map.hva;
3765 	vppr = *((u32 *)(vapic_page + APIC_PROCPRI));
3766 
3767 	return ((rvi & 0xf0) > (vppr & 0xf0));
3768 }
3769 
3770 static inline bool kvm_vcpu_trigger_posted_interrupt(struct kvm_vcpu *vcpu,
3771 						     bool nested)
3772 {
3773 #ifdef CONFIG_SMP
3774 	int pi_vec = nested ? POSTED_INTR_NESTED_VECTOR : POSTED_INTR_VECTOR;
3775 
3776 	if (vcpu->mode == IN_GUEST_MODE) {
3777 		/*
3778 		 * The vector of interrupt to be delivered to vcpu had
3779 		 * been set in PIR before this function.
3780 		 *
3781 		 * Following cases will be reached in this block, and
3782 		 * we always send a notification event in all cases as
3783 		 * explained below.
3784 		 *
3785 		 * Case 1: vcpu keeps in non-root mode. Sending a
3786 		 * notification event posts the interrupt to vcpu.
3787 		 *
3788 		 * Case 2: vcpu exits to root mode and is still
3789 		 * runnable. PIR will be synced to vIRR before the
3790 		 * next vcpu entry. Sending a notification event in
3791 		 * this case has no effect, as vcpu is not in root
3792 		 * mode.
3793 		 *
3794 		 * Case 3: vcpu exits to root mode and is blocked.
3795 		 * vcpu_block() has already synced PIR to vIRR and
3796 		 * never blocks vcpu if vIRR is not cleared. Therefore,
3797 		 * a blocked vcpu here does not wait for any requested
3798 		 * interrupts in PIR, and sending a notification event
3799 		 * which has no effect is safe here.
3800 		 */
3801 
3802 		apic->send_IPI_mask(get_cpu_mask(vcpu->cpu), pi_vec);
3803 		return true;
3804 	}
3805 #endif
3806 	return false;
3807 }
3808 
3809 static int vmx_deliver_nested_posted_interrupt(struct kvm_vcpu *vcpu,
3810 						int vector)
3811 {
3812 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3813 
3814 	if (is_guest_mode(vcpu) &&
3815 	    vector == vmx->nested.posted_intr_nv) {
3816 		/*
3817 		 * If a posted intr is not recognized by hardware,
3818 		 * we will accomplish it in the next vmentry.
3819 		 */
3820 		vmx->nested.pi_pending = true;
3821 		kvm_make_request(KVM_REQ_EVENT, vcpu);
3822 		/* the PIR and ON have been set by L1. */
3823 		if (!kvm_vcpu_trigger_posted_interrupt(vcpu, true))
3824 			kvm_vcpu_kick(vcpu);
3825 		return 0;
3826 	}
3827 	return -1;
3828 }
3829 /*
3830  * Send interrupt to vcpu via posted interrupt way.
3831  * 1. If target vcpu is running(non-root mode), send posted interrupt
3832  * notification to vcpu and hardware will sync PIR to vIRR atomically.
3833  * 2. If target vcpu isn't running(root mode), kick it to pick up the
3834  * interrupt from PIR in next vmentry.
3835  */
3836 static int vmx_deliver_posted_interrupt(struct kvm_vcpu *vcpu, int vector)
3837 {
3838 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3839 	int r;
3840 
3841 	r = vmx_deliver_nested_posted_interrupt(vcpu, vector);
3842 	if (!r)
3843 		return 0;
3844 
3845 	if (!vcpu->arch.apicv_active)
3846 		return -1;
3847 
3848 	if (pi_test_and_set_pir(vector, &vmx->pi_desc))
3849 		return 0;
3850 
3851 	/* If a previous notification has sent the IPI, nothing to do.  */
3852 	if (pi_test_and_set_on(&vmx->pi_desc))
3853 		return 0;
3854 
3855 	if (!kvm_vcpu_trigger_posted_interrupt(vcpu, false))
3856 		kvm_vcpu_kick(vcpu);
3857 
3858 	return 0;
3859 }
3860 
3861 /*
3862  * Set up the vmcs's constant host-state fields, i.e., host-state fields that
3863  * will not change in the lifetime of the guest.
3864  * Note that host-state that does change is set elsewhere. E.g., host-state
3865  * that is set differently for each CPU is set in vmx_vcpu_load(), not here.
3866  */
3867 void vmx_set_constant_host_state(struct vcpu_vmx *vmx)
3868 {
3869 	u32 low32, high32;
3870 	unsigned long tmpl;
3871 	unsigned long cr0, cr3, cr4;
3872 
3873 	cr0 = read_cr0();
3874 	WARN_ON(cr0 & X86_CR0_TS);
3875 	vmcs_writel(HOST_CR0, cr0);  /* 22.2.3 */
3876 
3877 	/*
3878 	 * Save the most likely value for this task's CR3 in the VMCS.
3879 	 * We can't use __get_current_cr3_fast() because we're not atomic.
3880 	 */
3881 	cr3 = __read_cr3();
3882 	vmcs_writel(HOST_CR3, cr3);		/* 22.2.3  FIXME: shadow tables */
3883 	vmx->loaded_vmcs->host_state.cr3 = cr3;
3884 
3885 	/* Save the most likely value for this task's CR4 in the VMCS. */
3886 	cr4 = cr4_read_shadow();
3887 	vmcs_writel(HOST_CR4, cr4);			/* 22.2.3, 22.2.5 */
3888 	vmx->loaded_vmcs->host_state.cr4 = cr4;
3889 
3890 	vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS);  /* 22.2.4 */
3891 #ifdef CONFIG_X86_64
3892 	/*
3893 	 * Load null selectors, so we can avoid reloading them in
3894 	 * vmx_prepare_switch_to_host(), in case userspace uses
3895 	 * the null selectors too (the expected case).
3896 	 */
3897 	vmcs_write16(HOST_DS_SELECTOR, 0);
3898 	vmcs_write16(HOST_ES_SELECTOR, 0);
3899 #else
3900 	vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
3901 	vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
3902 #endif
3903 	vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
3904 	vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8);  /* 22.2.4 */
3905 
3906 	vmcs_writel(HOST_IDTR_BASE, host_idt_base);   /* 22.2.4 */
3907 
3908 	vmcs_writel(HOST_RIP, (unsigned long)vmx_vmexit); /* 22.2.5 */
3909 
3910 	rdmsr(MSR_IA32_SYSENTER_CS, low32, high32);
3911 	vmcs_write32(HOST_IA32_SYSENTER_CS, low32);
3912 	rdmsrl(MSR_IA32_SYSENTER_EIP, tmpl);
3913 	vmcs_writel(HOST_IA32_SYSENTER_EIP, tmpl);   /* 22.2.3 */
3914 
3915 	if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
3916 		rdmsr(MSR_IA32_CR_PAT, low32, high32);
3917 		vmcs_write64(HOST_IA32_PAT, low32 | ((u64) high32 << 32));
3918 	}
3919 
3920 	if (cpu_has_load_ia32_efer())
3921 		vmcs_write64(HOST_IA32_EFER, host_efer);
3922 }
3923 
3924 void set_cr4_guest_host_mask(struct vcpu_vmx *vmx)
3925 {
3926 	vmx->vcpu.arch.cr4_guest_owned_bits = KVM_CR4_GUEST_OWNED_BITS;
3927 	if (enable_ept)
3928 		vmx->vcpu.arch.cr4_guest_owned_bits |= X86_CR4_PGE;
3929 	if (is_guest_mode(&vmx->vcpu))
3930 		vmx->vcpu.arch.cr4_guest_owned_bits &=
3931 			~get_vmcs12(&vmx->vcpu)->cr4_guest_host_mask;
3932 	vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits);
3933 }
3934 
3935 u32 vmx_pin_based_exec_ctrl(struct vcpu_vmx *vmx)
3936 {
3937 	u32 pin_based_exec_ctrl = vmcs_config.pin_based_exec_ctrl;
3938 
3939 	if (!kvm_vcpu_apicv_active(&vmx->vcpu))
3940 		pin_based_exec_ctrl &= ~PIN_BASED_POSTED_INTR;
3941 
3942 	if (!enable_vnmi)
3943 		pin_based_exec_ctrl &= ~PIN_BASED_VIRTUAL_NMIS;
3944 
3945 	if (!enable_preemption_timer)
3946 		pin_based_exec_ctrl &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
3947 
3948 	return pin_based_exec_ctrl;
3949 }
3950 
3951 static void vmx_refresh_apicv_exec_ctrl(struct kvm_vcpu *vcpu)
3952 {
3953 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3954 
3955 	pin_controls_set(vmx, vmx_pin_based_exec_ctrl(vmx));
3956 	if (cpu_has_secondary_exec_ctrls()) {
3957 		if (kvm_vcpu_apicv_active(vcpu))
3958 			secondary_exec_controls_setbit(vmx,
3959 				      SECONDARY_EXEC_APIC_REGISTER_VIRT |
3960 				      SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
3961 		else
3962 			secondary_exec_controls_clearbit(vmx,
3963 					SECONDARY_EXEC_APIC_REGISTER_VIRT |
3964 					SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
3965 	}
3966 
3967 	if (cpu_has_vmx_msr_bitmap())
3968 		vmx_update_msr_bitmap(vcpu);
3969 }
3970 
3971 u32 vmx_exec_control(struct vcpu_vmx *vmx)
3972 {
3973 	u32 exec_control = vmcs_config.cpu_based_exec_ctrl;
3974 
3975 	if (vmx->vcpu.arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)
3976 		exec_control &= ~CPU_BASED_MOV_DR_EXITING;
3977 
3978 	if (!cpu_need_tpr_shadow(&vmx->vcpu)) {
3979 		exec_control &= ~CPU_BASED_TPR_SHADOW;
3980 #ifdef CONFIG_X86_64
3981 		exec_control |= CPU_BASED_CR8_STORE_EXITING |
3982 				CPU_BASED_CR8_LOAD_EXITING;
3983 #endif
3984 	}
3985 	if (!enable_ept)
3986 		exec_control |= CPU_BASED_CR3_STORE_EXITING |
3987 				CPU_BASED_CR3_LOAD_EXITING  |
3988 				CPU_BASED_INVLPG_EXITING;
3989 	if (kvm_mwait_in_guest(vmx->vcpu.kvm))
3990 		exec_control &= ~(CPU_BASED_MWAIT_EXITING |
3991 				CPU_BASED_MONITOR_EXITING);
3992 	if (kvm_hlt_in_guest(vmx->vcpu.kvm))
3993 		exec_control &= ~CPU_BASED_HLT_EXITING;
3994 	return exec_control;
3995 }
3996 
3997 
3998 static void vmx_compute_secondary_exec_control(struct vcpu_vmx *vmx)
3999 {
4000 	struct kvm_vcpu *vcpu = &vmx->vcpu;
4001 
4002 	u32 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
4003 
4004 	if (vmx_pt_mode_is_system())
4005 		exec_control &= ~(SECONDARY_EXEC_PT_USE_GPA | SECONDARY_EXEC_PT_CONCEAL_VMX);
4006 	if (!cpu_need_virtualize_apic_accesses(vcpu))
4007 		exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
4008 	if (vmx->vpid == 0)
4009 		exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
4010 	if (!enable_ept) {
4011 		exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
4012 		enable_unrestricted_guest = 0;
4013 	}
4014 	if (!enable_unrestricted_guest)
4015 		exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
4016 	if (kvm_pause_in_guest(vmx->vcpu.kvm))
4017 		exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
4018 	if (!kvm_vcpu_apicv_active(vcpu))
4019 		exec_control &= ~(SECONDARY_EXEC_APIC_REGISTER_VIRT |
4020 				  SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
4021 	exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
4022 
4023 	/* SECONDARY_EXEC_DESC is enabled/disabled on writes to CR4.UMIP,
4024 	 * in vmx_set_cr4.  */
4025 	exec_control &= ~SECONDARY_EXEC_DESC;
4026 
4027 	/* SECONDARY_EXEC_SHADOW_VMCS is enabled when L1 executes VMPTRLD
4028 	   (handle_vmptrld).
4029 	   We can NOT enable shadow_vmcs here because we don't have yet
4030 	   a current VMCS12
4031 	*/
4032 	exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
4033 
4034 	if (!enable_pml)
4035 		exec_control &= ~SECONDARY_EXEC_ENABLE_PML;
4036 
4037 	if (vmx_xsaves_supported()) {
4038 		/* Exposing XSAVES only when XSAVE is exposed */
4039 		bool xsaves_enabled =
4040 			boot_cpu_has(X86_FEATURE_XSAVE) &&
4041 			guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
4042 			guest_cpuid_has(vcpu, X86_FEATURE_XSAVES);
4043 
4044 		vcpu->arch.xsaves_enabled = xsaves_enabled;
4045 
4046 		if (!xsaves_enabled)
4047 			exec_control &= ~SECONDARY_EXEC_XSAVES;
4048 
4049 		if (nested) {
4050 			if (xsaves_enabled)
4051 				vmx->nested.msrs.secondary_ctls_high |=
4052 					SECONDARY_EXEC_XSAVES;
4053 			else
4054 				vmx->nested.msrs.secondary_ctls_high &=
4055 					~SECONDARY_EXEC_XSAVES;
4056 		}
4057 	}
4058 
4059 	if (cpu_has_vmx_rdtscp()) {
4060 		bool rdtscp_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP);
4061 		if (!rdtscp_enabled)
4062 			exec_control &= ~SECONDARY_EXEC_RDTSCP;
4063 
4064 		if (nested) {
4065 			if (rdtscp_enabled)
4066 				vmx->nested.msrs.secondary_ctls_high |=
4067 					SECONDARY_EXEC_RDTSCP;
4068 			else
4069 				vmx->nested.msrs.secondary_ctls_high &=
4070 					~SECONDARY_EXEC_RDTSCP;
4071 		}
4072 	}
4073 
4074 	if (cpu_has_vmx_invpcid()) {
4075 		/* Exposing INVPCID only when PCID is exposed */
4076 		bool invpcid_enabled =
4077 			guest_cpuid_has(vcpu, X86_FEATURE_INVPCID) &&
4078 			guest_cpuid_has(vcpu, X86_FEATURE_PCID);
4079 
4080 		if (!invpcid_enabled) {
4081 			exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID;
4082 			guest_cpuid_clear(vcpu, X86_FEATURE_INVPCID);
4083 		}
4084 
4085 		if (nested) {
4086 			if (invpcid_enabled)
4087 				vmx->nested.msrs.secondary_ctls_high |=
4088 					SECONDARY_EXEC_ENABLE_INVPCID;
4089 			else
4090 				vmx->nested.msrs.secondary_ctls_high &=
4091 					~SECONDARY_EXEC_ENABLE_INVPCID;
4092 		}
4093 	}
4094 
4095 	if (vmx_rdrand_supported()) {
4096 		bool rdrand_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDRAND);
4097 		if (rdrand_enabled)
4098 			exec_control &= ~SECONDARY_EXEC_RDRAND_EXITING;
4099 
4100 		if (nested) {
4101 			if (rdrand_enabled)
4102 				vmx->nested.msrs.secondary_ctls_high |=
4103 					SECONDARY_EXEC_RDRAND_EXITING;
4104 			else
4105 				vmx->nested.msrs.secondary_ctls_high &=
4106 					~SECONDARY_EXEC_RDRAND_EXITING;
4107 		}
4108 	}
4109 
4110 	if (vmx_rdseed_supported()) {
4111 		bool rdseed_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDSEED);
4112 		if (rdseed_enabled)
4113 			exec_control &= ~SECONDARY_EXEC_RDSEED_EXITING;
4114 
4115 		if (nested) {
4116 			if (rdseed_enabled)
4117 				vmx->nested.msrs.secondary_ctls_high |=
4118 					SECONDARY_EXEC_RDSEED_EXITING;
4119 			else
4120 				vmx->nested.msrs.secondary_ctls_high &=
4121 					~SECONDARY_EXEC_RDSEED_EXITING;
4122 		}
4123 	}
4124 
4125 	if (vmx_waitpkg_supported()) {
4126 		bool waitpkg_enabled =
4127 			guest_cpuid_has(vcpu, X86_FEATURE_WAITPKG);
4128 
4129 		if (!waitpkg_enabled)
4130 			exec_control &= ~SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE;
4131 
4132 		if (nested) {
4133 			if (waitpkg_enabled)
4134 				vmx->nested.msrs.secondary_ctls_high |=
4135 					SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE;
4136 			else
4137 				vmx->nested.msrs.secondary_ctls_high &=
4138 					~SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE;
4139 		}
4140 	}
4141 
4142 	vmx->secondary_exec_control = exec_control;
4143 }
4144 
4145 static void ept_set_mmio_spte_mask(void)
4146 {
4147 	/*
4148 	 * EPT Misconfigurations can be generated if the value of bits 2:0
4149 	 * of an EPT paging-structure entry is 110b (write/execute).
4150 	 */
4151 	kvm_mmu_set_mmio_spte_mask(VMX_EPT_RWX_MASK,
4152 				   VMX_EPT_MISCONFIG_WX_VALUE, 0);
4153 }
4154 
4155 #define VMX_XSS_EXIT_BITMAP 0
4156 
4157 /*
4158  * Noting that the initialization of Guest-state Area of VMCS is in
4159  * vmx_vcpu_reset().
4160  */
4161 static void init_vmcs(struct vcpu_vmx *vmx)
4162 {
4163 	if (nested)
4164 		nested_vmx_set_vmcs_shadowing_bitmap();
4165 
4166 	if (cpu_has_vmx_msr_bitmap())
4167 		vmcs_write64(MSR_BITMAP, __pa(vmx->vmcs01.msr_bitmap));
4168 
4169 	vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
4170 
4171 	/* Control */
4172 	pin_controls_set(vmx, vmx_pin_based_exec_ctrl(vmx));
4173 
4174 	exec_controls_set(vmx, vmx_exec_control(vmx));
4175 
4176 	if (cpu_has_secondary_exec_ctrls()) {
4177 		vmx_compute_secondary_exec_control(vmx);
4178 		secondary_exec_controls_set(vmx, vmx->secondary_exec_control);
4179 	}
4180 
4181 	if (kvm_vcpu_apicv_active(&vmx->vcpu)) {
4182 		vmcs_write64(EOI_EXIT_BITMAP0, 0);
4183 		vmcs_write64(EOI_EXIT_BITMAP1, 0);
4184 		vmcs_write64(EOI_EXIT_BITMAP2, 0);
4185 		vmcs_write64(EOI_EXIT_BITMAP3, 0);
4186 
4187 		vmcs_write16(GUEST_INTR_STATUS, 0);
4188 
4189 		vmcs_write16(POSTED_INTR_NV, POSTED_INTR_VECTOR);
4190 		vmcs_write64(POSTED_INTR_DESC_ADDR, __pa((&vmx->pi_desc)));
4191 	}
4192 
4193 	if (!kvm_pause_in_guest(vmx->vcpu.kvm)) {
4194 		vmcs_write32(PLE_GAP, ple_gap);
4195 		vmx->ple_window = ple_window;
4196 		vmx->ple_window_dirty = true;
4197 	}
4198 
4199 	vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
4200 	vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
4201 	vmcs_write32(CR3_TARGET_COUNT, 0);           /* 22.2.1 */
4202 
4203 	vmcs_write16(HOST_FS_SELECTOR, 0);            /* 22.2.4 */
4204 	vmcs_write16(HOST_GS_SELECTOR, 0);            /* 22.2.4 */
4205 	vmx_set_constant_host_state(vmx);
4206 	vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
4207 	vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
4208 
4209 	if (cpu_has_vmx_vmfunc())
4210 		vmcs_write64(VM_FUNCTION_CONTROL, 0);
4211 
4212 	vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
4213 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
4214 	vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host.val));
4215 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
4216 	vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest.val));
4217 
4218 	if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
4219 		vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
4220 
4221 	vm_exit_controls_set(vmx, vmx_vmexit_ctrl());
4222 
4223 	/* 22.2.1, 20.8.1 */
4224 	vm_entry_controls_set(vmx, vmx_vmentry_ctrl());
4225 
4226 	vmx->vcpu.arch.cr0_guest_owned_bits = X86_CR0_TS;
4227 	vmcs_writel(CR0_GUEST_HOST_MASK, ~X86_CR0_TS);
4228 
4229 	set_cr4_guest_host_mask(vmx);
4230 
4231 	if (vmx->vpid != 0)
4232 		vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
4233 
4234 	if (vmx_xsaves_supported())
4235 		vmcs_write64(XSS_EXIT_BITMAP, VMX_XSS_EXIT_BITMAP);
4236 
4237 	if (enable_pml) {
4238 		vmcs_write64(PML_ADDRESS, page_to_phys(vmx->pml_pg));
4239 		vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
4240 	}
4241 
4242 	if (cpu_has_vmx_encls_vmexit())
4243 		vmcs_write64(ENCLS_EXITING_BITMAP, -1ull);
4244 
4245 	if (vmx_pt_mode_is_host_guest()) {
4246 		memset(&vmx->pt_desc, 0, sizeof(vmx->pt_desc));
4247 		/* Bit[6~0] are forced to 1, writes are ignored. */
4248 		vmx->pt_desc.guest.output_mask = 0x7F;
4249 		vmcs_write64(GUEST_IA32_RTIT_CTL, 0);
4250 	}
4251 }
4252 
4253 static void vmx_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
4254 {
4255 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4256 	struct msr_data apic_base_msr;
4257 	u64 cr0;
4258 
4259 	vmx->rmode.vm86_active = 0;
4260 	vmx->spec_ctrl = 0;
4261 
4262 	vmx->msr_ia32_umwait_control = 0;
4263 
4264 	vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
4265 	vmx->hv_deadline_tsc = -1;
4266 	kvm_set_cr8(vcpu, 0);
4267 
4268 	if (!init_event) {
4269 		apic_base_msr.data = APIC_DEFAULT_PHYS_BASE |
4270 				     MSR_IA32_APICBASE_ENABLE;
4271 		if (kvm_vcpu_is_reset_bsp(vcpu))
4272 			apic_base_msr.data |= MSR_IA32_APICBASE_BSP;
4273 		apic_base_msr.host_initiated = true;
4274 		kvm_set_apic_base(vcpu, &apic_base_msr);
4275 	}
4276 
4277 	vmx_segment_cache_clear(vmx);
4278 
4279 	seg_setup(VCPU_SREG_CS);
4280 	vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
4281 	vmcs_writel(GUEST_CS_BASE, 0xffff0000ul);
4282 
4283 	seg_setup(VCPU_SREG_DS);
4284 	seg_setup(VCPU_SREG_ES);
4285 	seg_setup(VCPU_SREG_FS);
4286 	seg_setup(VCPU_SREG_GS);
4287 	seg_setup(VCPU_SREG_SS);
4288 
4289 	vmcs_write16(GUEST_TR_SELECTOR, 0);
4290 	vmcs_writel(GUEST_TR_BASE, 0);
4291 	vmcs_write32(GUEST_TR_LIMIT, 0xffff);
4292 	vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
4293 
4294 	vmcs_write16(GUEST_LDTR_SELECTOR, 0);
4295 	vmcs_writel(GUEST_LDTR_BASE, 0);
4296 	vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
4297 	vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
4298 
4299 	if (!init_event) {
4300 		vmcs_write32(GUEST_SYSENTER_CS, 0);
4301 		vmcs_writel(GUEST_SYSENTER_ESP, 0);
4302 		vmcs_writel(GUEST_SYSENTER_EIP, 0);
4303 		vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
4304 	}
4305 
4306 	kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
4307 	kvm_rip_write(vcpu, 0xfff0);
4308 
4309 	vmcs_writel(GUEST_GDTR_BASE, 0);
4310 	vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
4311 
4312 	vmcs_writel(GUEST_IDTR_BASE, 0);
4313 	vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
4314 
4315 	vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
4316 	vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
4317 	vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS, 0);
4318 	if (kvm_mpx_supported())
4319 		vmcs_write64(GUEST_BNDCFGS, 0);
4320 
4321 	setup_msrs(vmx);
4322 
4323 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);  /* 22.2.1 */
4324 
4325 	if (cpu_has_vmx_tpr_shadow() && !init_event) {
4326 		vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
4327 		if (cpu_need_tpr_shadow(vcpu))
4328 			vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
4329 				     __pa(vcpu->arch.apic->regs));
4330 		vmcs_write32(TPR_THRESHOLD, 0);
4331 	}
4332 
4333 	kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
4334 
4335 	cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
4336 	vmx->vcpu.arch.cr0 = cr0;
4337 	vmx_set_cr0(vcpu, cr0); /* enter rmode */
4338 	vmx_set_cr4(vcpu, 0);
4339 	vmx_set_efer(vcpu, 0);
4340 
4341 	update_exception_bitmap(vcpu);
4342 
4343 	vpid_sync_context(vmx->vpid);
4344 	if (init_event)
4345 		vmx_clear_hlt(vcpu);
4346 }
4347 
4348 static void enable_irq_window(struct kvm_vcpu *vcpu)
4349 {
4350 	exec_controls_setbit(to_vmx(vcpu), CPU_BASED_INTR_WINDOW_EXITING);
4351 }
4352 
4353 static void enable_nmi_window(struct kvm_vcpu *vcpu)
4354 {
4355 	if (!enable_vnmi ||
4356 	    vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_STI) {
4357 		enable_irq_window(vcpu);
4358 		return;
4359 	}
4360 
4361 	exec_controls_setbit(to_vmx(vcpu), CPU_BASED_NMI_WINDOW_EXITING);
4362 }
4363 
4364 static void vmx_inject_irq(struct kvm_vcpu *vcpu)
4365 {
4366 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4367 	uint32_t intr;
4368 	int irq = vcpu->arch.interrupt.nr;
4369 
4370 	trace_kvm_inj_virq(irq);
4371 
4372 	++vcpu->stat.irq_injections;
4373 	if (vmx->rmode.vm86_active) {
4374 		int inc_eip = 0;
4375 		if (vcpu->arch.interrupt.soft)
4376 			inc_eip = vcpu->arch.event_exit_inst_len;
4377 		kvm_inject_realmode_interrupt(vcpu, irq, inc_eip);
4378 		return;
4379 	}
4380 	intr = irq | INTR_INFO_VALID_MASK;
4381 	if (vcpu->arch.interrupt.soft) {
4382 		intr |= INTR_TYPE_SOFT_INTR;
4383 		vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
4384 			     vmx->vcpu.arch.event_exit_inst_len);
4385 	} else
4386 		intr |= INTR_TYPE_EXT_INTR;
4387 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
4388 
4389 	vmx_clear_hlt(vcpu);
4390 }
4391 
4392 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
4393 {
4394 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4395 
4396 	if (!enable_vnmi) {
4397 		/*
4398 		 * Tracking the NMI-blocked state in software is built upon
4399 		 * finding the next open IRQ window. This, in turn, depends on
4400 		 * well-behaving guests: They have to keep IRQs disabled at
4401 		 * least as long as the NMI handler runs. Otherwise we may
4402 		 * cause NMI nesting, maybe breaking the guest. But as this is
4403 		 * highly unlikely, we can live with the residual risk.
4404 		 */
4405 		vmx->loaded_vmcs->soft_vnmi_blocked = 1;
4406 		vmx->loaded_vmcs->vnmi_blocked_time = 0;
4407 	}
4408 
4409 	++vcpu->stat.nmi_injections;
4410 	vmx->loaded_vmcs->nmi_known_unmasked = false;
4411 
4412 	if (vmx->rmode.vm86_active) {
4413 		kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR, 0);
4414 		return;
4415 	}
4416 
4417 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
4418 			INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
4419 
4420 	vmx_clear_hlt(vcpu);
4421 }
4422 
4423 bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
4424 {
4425 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4426 	bool masked;
4427 
4428 	if (!enable_vnmi)
4429 		return vmx->loaded_vmcs->soft_vnmi_blocked;
4430 	if (vmx->loaded_vmcs->nmi_known_unmasked)
4431 		return false;
4432 	masked = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI;
4433 	vmx->loaded_vmcs->nmi_known_unmasked = !masked;
4434 	return masked;
4435 }
4436 
4437 void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
4438 {
4439 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4440 
4441 	if (!enable_vnmi) {
4442 		if (vmx->loaded_vmcs->soft_vnmi_blocked != masked) {
4443 			vmx->loaded_vmcs->soft_vnmi_blocked = masked;
4444 			vmx->loaded_vmcs->vnmi_blocked_time = 0;
4445 		}
4446 	} else {
4447 		vmx->loaded_vmcs->nmi_known_unmasked = !masked;
4448 		if (masked)
4449 			vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
4450 				      GUEST_INTR_STATE_NMI);
4451 		else
4452 			vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
4453 					GUEST_INTR_STATE_NMI);
4454 	}
4455 }
4456 
4457 static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
4458 {
4459 	if (to_vmx(vcpu)->nested.nested_run_pending)
4460 		return 0;
4461 
4462 	if (!enable_vnmi &&
4463 	    to_vmx(vcpu)->loaded_vmcs->soft_vnmi_blocked)
4464 		return 0;
4465 
4466 	return	!(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4467 		  (GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_STI
4468 		   | GUEST_INTR_STATE_NMI));
4469 }
4470 
4471 static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
4472 {
4473 	if (to_vmx(vcpu)->nested.nested_run_pending)
4474 		return false;
4475 
4476 	if (is_guest_mode(vcpu) && nested_exit_on_intr(vcpu))
4477 		return true;
4478 
4479 	return (vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
4480 		!(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4481 			(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
4482 }
4483 
4484 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
4485 {
4486 	int ret;
4487 
4488 	if (enable_unrestricted_guest)
4489 		return 0;
4490 
4491 	mutex_lock(&kvm->slots_lock);
4492 	ret = __x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, addr,
4493 				      PAGE_SIZE * 3);
4494 	mutex_unlock(&kvm->slots_lock);
4495 
4496 	if (ret)
4497 		return ret;
4498 	to_kvm_vmx(kvm)->tss_addr = addr;
4499 	return init_rmode_tss(kvm);
4500 }
4501 
4502 static int vmx_set_identity_map_addr(struct kvm *kvm, u64 ident_addr)
4503 {
4504 	to_kvm_vmx(kvm)->ept_identity_map_addr = ident_addr;
4505 	return 0;
4506 }
4507 
4508 static bool rmode_exception(struct kvm_vcpu *vcpu, int vec)
4509 {
4510 	switch (vec) {
4511 	case BP_VECTOR:
4512 		/*
4513 		 * Update instruction length as we may reinject the exception
4514 		 * from user space while in guest debugging mode.
4515 		 */
4516 		to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
4517 			vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4518 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
4519 			return false;
4520 		/* fall through */
4521 	case DB_VECTOR:
4522 		if (vcpu->guest_debug &
4523 			(KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
4524 			return false;
4525 		/* fall through */
4526 	case DE_VECTOR:
4527 	case OF_VECTOR:
4528 	case BR_VECTOR:
4529 	case UD_VECTOR:
4530 	case DF_VECTOR:
4531 	case SS_VECTOR:
4532 	case GP_VECTOR:
4533 	case MF_VECTOR:
4534 		return true;
4535 	}
4536 	return false;
4537 }
4538 
4539 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
4540 				  int vec, u32 err_code)
4541 {
4542 	/*
4543 	 * Instruction with address size override prefix opcode 0x67
4544 	 * Cause the #SS fault with 0 error code in VM86 mode.
4545 	 */
4546 	if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0) {
4547 		if (kvm_emulate_instruction(vcpu, 0)) {
4548 			if (vcpu->arch.halt_request) {
4549 				vcpu->arch.halt_request = 0;
4550 				return kvm_vcpu_halt(vcpu);
4551 			}
4552 			return 1;
4553 		}
4554 		return 0;
4555 	}
4556 
4557 	/*
4558 	 * Forward all other exceptions that are valid in real mode.
4559 	 * FIXME: Breaks guest debugging in real mode, needs to be fixed with
4560 	 *        the required debugging infrastructure rework.
4561 	 */
4562 	kvm_queue_exception(vcpu, vec);
4563 	return 1;
4564 }
4565 
4566 /*
4567  * Trigger machine check on the host. We assume all the MSRs are already set up
4568  * by the CPU and that we still run on the same CPU as the MCE occurred on.
4569  * We pass a fake environment to the machine check handler because we want
4570  * the guest to be always treated like user space, no matter what context
4571  * it used internally.
4572  */
4573 static void kvm_machine_check(void)
4574 {
4575 #if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
4576 	struct pt_regs regs = {
4577 		.cs = 3, /* Fake ring 3 no matter what the guest ran on */
4578 		.flags = X86_EFLAGS_IF,
4579 	};
4580 
4581 	do_machine_check(&regs, 0);
4582 #endif
4583 }
4584 
4585 static int handle_machine_check(struct kvm_vcpu *vcpu)
4586 {
4587 	/* handled by vmx_vcpu_run() */
4588 	return 1;
4589 }
4590 
4591 /*
4592  * If the host has split lock detection disabled, then #AC is
4593  * unconditionally injected into the guest, which is the pre split lock
4594  * detection behaviour.
4595  *
4596  * If the host has split lock detection enabled then #AC is
4597  * only injected into the guest when:
4598  *  - Guest CPL == 3 (user mode)
4599  *  - Guest has #AC detection enabled in CR0
4600  *  - Guest EFLAGS has AC bit set
4601  */
4602 static inline bool guest_inject_ac(struct kvm_vcpu *vcpu)
4603 {
4604 	if (!boot_cpu_has(X86_FEATURE_SPLIT_LOCK_DETECT))
4605 		return true;
4606 
4607 	return vmx_get_cpl(vcpu) == 3 && kvm_read_cr0_bits(vcpu, X86_CR0_AM) &&
4608 	       (kvm_get_rflags(vcpu) & X86_EFLAGS_AC);
4609 }
4610 
4611 static int handle_exception_nmi(struct kvm_vcpu *vcpu)
4612 {
4613 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4614 	struct kvm_run *kvm_run = vcpu->run;
4615 	u32 intr_info, ex_no, error_code;
4616 	unsigned long cr2, rip, dr6;
4617 	u32 vect_info;
4618 
4619 	vect_info = vmx->idt_vectoring_info;
4620 	intr_info = vmx->exit_intr_info;
4621 
4622 	if (is_machine_check(intr_info) || is_nmi(intr_info))
4623 		return 1; /* handled by handle_exception_nmi_irqoff() */
4624 
4625 	if (is_invalid_opcode(intr_info))
4626 		return handle_ud(vcpu);
4627 
4628 	error_code = 0;
4629 	if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
4630 		error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
4631 
4632 	if (!vmx->rmode.vm86_active && is_gp_fault(intr_info)) {
4633 		WARN_ON_ONCE(!enable_vmware_backdoor);
4634 
4635 		/*
4636 		 * VMware backdoor emulation on #GP interception only handles
4637 		 * IN{S}, OUT{S}, and RDPMC, none of which generate a non-zero
4638 		 * error code on #GP.
4639 		 */
4640 		if (error_code) {
4641 			kvm_queue_exception_e(vcpu, GP_VECTOR, error_code);
4642 			return 1;
4643 		}
4644 		return kvm_emulate_instruction(vcpu, EMULTYPE_VMWARE_GP);
4645 	}
4646 
4647 	/*
4648 	 * The #PF with PFEC.RSVD = 1 indicates the guest is accessing
4649 	 * MMIO, it is better to report an internal error.
4650 	 * See the comments in vmx_handle_exit.
4651 	 */
4652 	if ((vect_info & VECTORING_INFO_VALID_MASK) &&
4653 	    !(is_page_fault(intr_info) && !(error_code & PFERR_RSVD_MASK))) {
4654 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4655 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
4656 		vcpu->run->internal.ndata = 3;
4657 		vcpu->run->internal.data[0] = vect_info;
4658 		vcpu->run->internal.data[1] = intr_info;
4659 		vcpu->run->internal.data[2] = error_code;
4660 		return 0;
4661 	}
4662 
4663 	if (is_page_fault(intr_info)) {
4664 		cr2 = vmcs_readl(EXIT_QUALIFICATION);
4665 		/* EPT won't cause page fault directly */
4666 		WARN_ON_ONCE(!vcpu->arch.apf.host_apf_reason && enable_ept);
4667 		return kvm_handle_page_fault(vcpu, error_code, cr2, NULL, 0);
4668 	}
4669 
4670 	ex_no = intr_info & INTR_INFO_VECTOR_MASK;
4671 
4672 	if (vmx->rmode.vm86_active && rmode_exception(vcpu, ex_no))
4673 		return handle_rmode_exception(vcpu, ex_no, error_code);
4674 
4675 	switch (ex_no) {
4676 	case DB_VECTOR:
4677 		dr6 = vmcs_readl(EXIT_QUALIFICATION);
4678 		if (!(vcpu->guest_debug &
4679 		      (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
4680 			vcpu->arch.dr6 &= ~DR_TRAP_BITS;
4681 			vcpu->arch.dr6 |= dr6 | DR6_RTM;
4682 			if (is_icebp(intr_info))
4683 				WARN_ON(!skip_emulated_instruction(vcpu));
4684 
4685 			kvm_queue_exception(vcpu, DB_VECTOR);
4686 			return 1;
4687 		}
4688 		kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
4689 		kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
4690 		/* fall through */
4691 	case BP_VECTOR:
4692 		/*
4693 		 * Update instruction length as we may reinject #BP from
4694 		 * user space while in guest debugging mode. Reading it for
4695 		 * #DB as well causes no harm, it is not used in that case.
4696 		 */
4697 		vmx->vcpu.arch.event_exit_inst_len =
4698 			vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4699 		kvm_run->exit_reason = KVM_EXIT_DEBUG;
4700 		rip = kvm_rip_read(vcpu);
4701 		kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
4702 		kvm_run->debug.arch.exception = ex_no;
4703 		break;
4704 	case AC_VECTOR:
4705 		if (guest_inject_ac(vcpu)) {
4706 			kvm_queue_exception_e(vcpu, AC_VECTOR, error_code);
4707 			return 1;
4708 		}
4709 
4710 		/*
4711 		 * Handle split lock. Depending on detection mode this will
4712 		 * either warn and disable split lock detection for this
4713 		 * task or force SIGBUS on it.
4714 		 */
4715 		if (handle_guest_split_lock(kvm_rip_read(vcpu)))
4716 			return 1;
4717 		fallthrough;
4718 	default:
4719 		kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
4720 		kvm_run->ex.exception = ex_no;
4721 		kvm_run->ex.error_code = error_code;
4722 		break;
4723 	}
4724 	return 0;
4725 }
4726 
4727 static __always_inline int handle_external_interrupt(struct kvm_vcpu *vcpu)
4728 {
4729 	++vcpu->stat.irq_exits;
4730 	return 1;
4731 }
4732 
4733 static int handle_triple_fault(struct kvm_vcpu *vcpu)
4734 {
4735 	vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
4736 	vcpu->mmio_needed = 0;
4737 	return 0;
4738 }
4739 
4740 static int handle_io(struct kvm_vcpu *vcpu)
4741 {
4742 	unsigned long exit_qualification;
4743 	int size, in, string;
4744 	unsigned port;
4745 
4746 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4747 	string = (exit_qualification & 16) != 0;
4748 
4749 	++vcpu->stat.io_exits;
4750 
4751 	if (string)
4752 		return kvm_emulate_instruction(vcpu, 0);
4753 
4754 	port = exit_qualification >> 16;
4755 	size = (exit_qualification & 7) + 1;
4756 	in = (exit_qualification & 8) != 0;
4757 
4758 	return kvm_fast_pio(vcpu, size, port, in);
4759 }
4760 
4761 static void
4762 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
4763 {
4764 	/*
4765 	 * Patch in the VMCALL instruction:
4766 	 */
4767 	hypercall[0] = 0x0f;
4768 	hypercall[1] = 0x01;
4769 	hypercall[2] = 0xc1;
4770 }
4771 
4772 /* called to set cr0 as appropriate for a mov-to-cr0 exit. */
4773 static int handle_set_cr0(struct kvm_vcpu *vcpu, unsigned long val)
4774 {
4775 	if (is_guest_mode(vcpu)) {
4776 		struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4777 		unsigned long orig_val = val;
4778 
4779 		/*
4780 		 * We get here when L2 changed cr0 in a way that did not change
4781 		 * any of L1's shadowed bits (see nested_vmx_exit_handled_cr),
4782 		 * but did change L0 shadowed bits. So we first calculate the
4783 		 * effective cr0 value that L1 would like to write into the
4784 		 * hardware. It consists of the L2-owned bits from the new
4785 		 * value combined with the L1-owned bits from L1's guest_cr0.
4786 		 */
4787 		val = (val & ~vmcs12->cr0_guest_host_mask) |
4788 			(vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask);
4789 
4790 		if (!nested_guest_cr0_valid(vcpu, val))
4791 			return 1;
4792 
4793 		if (kvm_set_cr0(vcpu, val))
4794 			return 1;
4795 		vmcs_writel(CR0_READ_SHADOW, orig_val);
4796 		return 0;
4797 	} else {
4798 		if (to_vmx(vcpu)->nested.vmxon &&
4799 		    !nested_host_cr0_valid(vcpu, val))
4800 			return 1;
4801 
4802 		return kvm_set_cr0(vcpu, val);
4803 	}
4804 }
4805 
4806 static int handle_set_cr4(struct kvm_vcpu *vcpu, unsigned long val)
4807 {
4808 	if (is_guest_mode(vcpu)) {
4809 		struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4810 		unsigned long orig_val = val;
4811 
4812 		/* analogously to handle_set_cr0 */
4813 		val = (val & ~vmcs12->cr4_guest_host_mask) |
4814 			(vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask);
4815 		if (kvm_set_cr4(vcpu, val))
4816 			return 1;
4817 		vmcs_writel(CR4_READ_SHADOW, orig_val);
4818 		return 0;
4819 	} else
4820 		return kvm_set_cr4(vcpu, val);
4821 }
4822 
4823 static int handle_desc(struct kvm_vcpu *vcpu)
4824 {
4825 	WARN_ON(!(vcpu->arch.cr4 & X86_CR4_UMIP));
4826 	return kvm_emulate_instruction(vcpu, 0);
4827 }
4828 
4829 static int handle_cr(struct kvm_vcpu *vcpu)
4830 {
4831 	unsigned long exit_qualification, val;
4832 	int cr;
4833 	int reg;
4834 	int err;
4835 	int ret;
4836 
4837 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4838 	cr = exit_qualification & 15;
4839 	reg = (exit_qualification >> 8) & 15;
4840 	switch ((exit_qualification >> 4) & 3) {
4841 	case 0: /* mov to cr */
4842 		val = kvm_register_readl(vcpu, reg);
4843 		trace_kvm_cr_write(cr, val);
4844 		switch (cr) {
4845 		case 0:
4846 			err = handle_set_cr0(vcpu, val);
4847 			return kvm_complete_insn_gp(vcpu, err);
4848 		case 3:
4849 			WARN_ON_ONCE(enable_unrestricted_guest);
4850 			err = kvm_set_cr3(vcpu, val);
4851 			return kvm_complete_insn_gp(vcpu, err);
4852 		case 4:
4853 			err = handle_set_cr4(vcpu, val);
4854 			return kvm_complete_insn_gp(vcpu, err);
4855 		case 8: {
4856 				u8 cr8_prev = kvm_get_cr8(vcpu);
4857 				u8 cr8 = (u8)val;
4858 				err = kvm_set_cr8(vcpu, cr8);
4859 				ret = kvm_complete_insn_gp(vcpu, err);
4860 				if (lapic_in_kernel(vcpu))
4861 					return ret;
4862 				if (cr8_prev <= cr8)
4863 					return ret;
4864 				/*
4865 				 * TODO: we might be squashing a
4866 				 * KVM_GUESTDBG_SINGLESTEP-triggered
4867 				 * KVM_EXIT_DEBUG here.
4868 				 */
4869 				vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
4870 				return 0;
4871 			}
4872 		}
4873 		break;
4874 	case 2: /* clts */
4875 		WARN_ONCE(1, "Guest should always own CR0.TS");
4876 		vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
4877 		trace_kvm_cr_write(0, kvm_read_cr0(vcpu));
4878 		return kvm_skip_emulated_instruction(vcpu);
4879 	case 1: /*mov from cr*/
4880 		switch (cr) {
4881 		case 3:
4882 			WARN_ON_ONCE(enable_unrestricted_guest);
4883 			val = kvm_read_cr3(vcpu);
4884 			kvm_register_write(vcpu, reg, val);
4885 			trace_kvm_cr_read(cr, val);
4886 			return kvm_skip_emulated_instruction(vcpu);
4887 		case 8:
4888 			val = kvm_get_cr8(vcpu);
4889 			kvm_register_write(vcpu, reg, val);
4890 			trace_kvm_cr_read(cr, val);
4891 			return kvm_skip_emulated_instruction(vcpu);
4892 		}
4893 		break;
4894 	case 3: /* lmsw */
4895 		val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
4896 		trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val);
4897 		kvm_lmsw(vcpu, val);
4898 
4899 		return kvm_skip_emulated_instruction(vcpu);
4900 	default:
4901 		break;
4902 	}
4903 	vcpu->run->exit_reason = 0;
4904 	vcpu_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
4905 	       (int)(exit_qualification >> 4) & 3, cr);
4906 	return 0;
4907 }
4908 
4909 static int handle_dr(struct kvm_vcpu *vcpu)
4910 {
4911 	unsigned long exit_qualification;
4912 	int dr, dr7, reg;
4913 
4914 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4915 	dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
4916 
4917 	/* First, if DR does not exist, trigger UD */
4918 	if (!kvm_require_dr(vcpu, dr))
4919 		return 1;
4920 
4921 	/* Do not handle if the CPL > 0, will trigger GP on re-entry */
4922 	if (!kvm_require_cpl(vcpu, 0))
4923 		return 1;
4924 	dr7 = vmcs_readl(GUEST_DR7);
4925 	if (dr7 & DR7_GD) {
4926 		/*
4927 		 * As the vm-exit takes precedence over the debug trap, we
4928 		 * need to emulate the latter, either for the host or the
4929 		 * guest debugging itself.
4930 		 */
4931 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
4932 			vcpu->run->debug.arch.dr6 = vcpu->arch.dr6;
4933 			vcpu->run->debug.arch.dr7 = dr7;
4934 			vcpu->run->debug.arch.pc = kvm_get_linear_rip(vcpu);
4935 			vcpu->run->debug.arch.exception = DB_VECTOR;
4936 			vcpu->run->exit_reason = KVM_EXIT_DEBUG;
4937 			return 0;
4938 		} else {
4939 			vcpu->arch.dr6 &= ~DR_TRAP_BITS;
4940 			vcpu->arch.dr6 |= DR6_BD | DR6_RTM;
4941 			kvm_queue_exception(vcpu, DB_VECTOR);
4942 			return 1;
4943 		}
4944 	}
4945 
4946 	if (vcpu->guest_debug == 0) {
4947 		exec_controls_clearbit(to_vmx(vcpu), CPU_BASED_MOV_DR_EXITING);
4948 
4949 		/*
4950 		 * No more DR vmexits; force a reload of the debug registers
4951 		 * and reenter on this instruction.  The next vmexit will
4952 		 * retrieve the full state of the debug registers.
4953 		 */
4954 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
4955 		return 1;
4956 	}
4957 
4958 	reg = DEBUG_REG_ACCESS_REG(exit_qualification);
4959 	if (exit_qualification & TYPE_MOV_FROM_DR) {
4960 		unsigned long val;
4961 
4962 		if (kvm_get_dr(vcpu, dr, &val))
4963 			return 1;
4964 		kvm_register_write(vcpu, reg, val);
4965 	} else
4966 		if (kvm_set_dr(vcpu, dr, kvm_register_readl(vcpu, reg)))
4967 			return 1;
4968 
4969 	return kvm_skip_emulated_instruction(vcpu);
4970 }
4971 
4972 static u64 vmx_get_dr6(struct kvm_vcpu *vcpu)
4973 {
4974 	return vcpu->arch.dr6;
4975 }
4976 
4977 static void vmx_set_dr6(struct kvm_vcpu *vcpu, unsigned long val)
4978 {
4979 }
4980 
4981 static void vmx_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
4982 {
4983 	get_debugreg(vcpu->arch.db[0], 0);
4984 	get_debugreg(vcpu->arch.db[1], 1);
4985 	get_debugreg(vcpu->arch.db[2], 2);
4986 	get_debugreg(vcpu->arch.db[3], 3);
4987 	get_debugreg(vcpu->arch.dr6, 6);
4988 	vcpu->arch.dr7 = vmcs_readl(GUEST_DR7);
4989 
4990 	vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
4991 	exec_controls_setbit(to_vmx(vcpu), CPU_BASED_MOV_DR_EXITING);
4992 }
4993 
4994 static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
4995 {
4996 	vmcs_writel(GUEST_DR7, val);
4997 }
4998 
4999 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
5000 {
5001 	kvm_apic_update_ppr(vcpu);
5002 	return 1;
5003 }
5004 
5005 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
5006 {
5007 	exec_controls_clearbit(to_vmx(vcpu), CPU_BASED_INTR_WINDOW_EXITING);
5008 
5009 	kvm_make_request(KVM_REQ_EVENT, vcpu);
5010 
5011 	++vcpu->stat.irq_window_exits;
5012 	return 1;
5013 }
5014 
5015 static int handle_vmcall(struct kvm_vcpu *vcpu)
5016 {
5017 	return kvm_emulate_hypercall(vcpu);
5018 }
5019 
5020 static int handle_invd(struct kvm_vcpu *vcpu)
5021 {
5022 	return kvm_emulate_instruction(vcpu, 0);
5023 }
5024 
5025 static int handle_invlpg(struct kvm_vcpu *vcpu)
5026 {
5027 	unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5028 
5029 	kvm_mmu_invlpg(vcpu, exit_qualification);
5030 	return kvm_skip_emulated_instruction(vcpu);
5031 }
5032 
5033 static int handle_rdpmc(struct kvm_vcpu *vcpu)
5034 {
5035 	int err;
5036 
5037 	err = kvm_rdpmc(vcpu);
5038 	return kvm_complete_insn_gp(vcpu, err);
5039 }
5040 
5041 static int handle_wbinvd(struct kvm_vcpu *vcpu)
5042 {
5043 	return kvm_emulate_wbinvd(vcpu);
5044 }
5045 
5046 static int handle_xsetbv(struct kvm_vcpu *vcpu)
5047 {
5048 	u64 new_bv = kvm_read_edx_eax(vcpu);
5049 	u32 index = kvm_rcx_read(vcpu);
5050 
5051 	if (kvm_set_xcr(vcpu, index, new_bv) == 0)
5052 		return kvm_skip_emulated_instruction(vcpu);
5053 	return 1;
5054 }
5055 
5056 static int handle_apic_access(struct kvm_vcpu *vcpu)
5057 {
5058 	if (likely(fasteoi)) {
5059 		unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5060 		int access_type, offset;
5061 
5062 		access_type = exit_qualification & APIC_ACCESS_TYPE;
5063 		offset = exit_qualification & APIC_ACCESS_OFFSET;
5064 		/*
5065 		 * Sane guest uses MOV to write EOI, with written value
5066 		 * not cared. So make a short-circuit here by avoiding
5067 		 * heavy instruction emulation.
5068 		 */
5069 		if ((access_type == TYPE_LINEAR_APIC_INST_WRITE) &&
5070 		    (offset == APIC_EOI)) {
5071 			kvm_lapic_set_eoi(vcpu);
5072 			return kvm_skip_emulated_instruction(vcpu);
5073 		}
5074 	}
5075 	return kvm_emulate_instruction(vcpu, 0);
5076 }
5077 
5078 static int handle_apic_eoi_induced(struct kvm_vcpu *vcpu)
5079 {
5080 	unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5081 	int vector = exit_qualification & 0xff;
5082 
5083 	/* EOI-induced VM exit is trap-like and thus no need to adjust IP */
5084 	kvm_apic_set_eoi_accelerated(vcpu, vector);
5085 	return 1;
5086 }
5087 
5088 static int handle_apic_write(struct kvm_vcpu *vcpu)
5089 {
5090 	unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5091 	u32 offset = exit_qualification & 0xfff;
5092 
5093 	/* APIC-write VM exit is trap-like and thus no need to adjust IP */
5094 	kvm_apic_write_nodecode(vcpu, offset);
5095 	return 1;
5096 }
5097 
5098 static int handle_task_switch(struct kvm_vcpu *vcpu)
5099 {
5100 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5101 	unsigned long exit_qualification;
5102 	bool has_error_code = false;
5103 	u32 error_code = 0;
5104 	u16 tss_selector;
5105 	int reason, type, idt_v, idt_index;
5106 
5107 	idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
5108 	idt_index = (vmx->idt_vectoring_info & VECTORING_INFO_VECTOR_MASK);
5109 	type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
5110 
5111 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5112 
5113 	reason = (u32)exit_qualification >> 30;
5114 	if (reason == TASK_SWITCH_GATE && idt_v) {
5115 		switch (type) {
5116 		case INTR_TYPE_NMI_INTR:
5117 			vcpu->arch.nmi_injected = false;
5118 			vmx_set_nmi_mask(vcpu, true);
5119 			break;
5120 		case INTR_TYPE_EXT_INTR:
5121 		case INTR_TYPE_SOFT_INTR:
5122 			kvm_clear_interrupt_queue(vcpu);
5123 			break;
5124 		case INTR_TYPE_HARD_EXCEPTION:
5125 			if (vmx->idt_vectoring_info &
5126 			    VECTORING_INFO_DELIVER_CODE_MASK) {
5127 				has_error_code = true;
5128 				error_code =
5129 					vmcs_read32(IDT_VECTORING_ERROR_CODE);
5130 			}
5131 			/* fall through */
5132 		case INTR_TYPE_SOFT_EXCEPTION:
5133 			kvm_clear_exception_queue(vcpu);
5134 			break;
5135 		default:
5136 			break;
5137 		}
5138 	}
5139 	tss_selector = exit_qualification;
5140 
5141 	if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
5142 		       type != INTR_TYPE_EXT_INTR &&
5143 		       type != INTR_TYPE_NMI_INTR))
5144 		WARN_ON(!skip_emulated_instruction(vcpu));
5145 
5146 	/*
5147 	 * TODO: What about debug traps on tss switch?
5148 	 *       Are we supposed to inject them and update dr6?
5149 	 */
5150 	return kvm_task_switch(vcpu, tss_selector,
5151 			       type == INTR_TYPE_SOFT_INTR ? idt_index : -1,
5152 			       reason, has_error_code, error_code);
5153 }
5154 
5155 static int handle_ept_violation(struct kvm_vcpu *vcpu)
5156 {
5157 	unsigned long exit_qualification;
5158 	gpa_t gpa;
5159 	u64 error_code;
5160 
5161 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5162 
5163 	/*
5164 	 * EPT violation happened while executing iret from NMI,
5165 	 * "blocked by NMI" bit has to be set before next VM entry.
5166 	 * There are errata that may cause this bit to not be set:
5167 	 * AAK134, BY25.
5168 	 */
5169 	if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
5170 			enable_vnmi &&
5171 			(exit_qualification & INTR_INFO_UNBLOCK_NMI))
5172 		vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO, GUEST_INTR_STATE_NMI);
5173 
5174 	gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5175 	trace_kvm_page_fault(gpa, exit_qualification);
5176 
5177 	/* Is it a read fault? */
5178 	error_code = (exit_qualification & EPT_VIOLATION_ACC_READ)
5179 		     ? PFERR_USER_MASK : 0;
5180 	/* Is it a write fault? */
5181 	error_code |= (exit_qualification & EPT_VIOLATION_ACC_WRITE)
5182 		      ? PFERR_WRITE_MASK : 0;
5183 	/* Is it a fetch fault? */
5184 	error_code |= (exit_qualification & EPT_VIOLATION_ACC_INSTR)
5185 		      ? PFERR_FETCH_MASK : 0;
5186 	/* ept page table entry is present? */
5187 	error_code |= (exit_qualification &
5188 		       (EPT_VIOLATION_READABLE | EPT_VIOLATION_WRITABLE |
5189 			EPT_VIOLATION_EXECUTABLE))
5190 		      ? PFERR_PRESENT_MASK : 0;
5191 
5192 	error_code |= (exit_qualification & 0x100) != 0 ?
5193 	       PFERR_GUEST_FINAL_MASK : PFERR_GUEST_PAGE_MASK;
5194 
5195 	vcpu->arch.exit_qualification = exit_qualification;
5196 	return kvm_mmu_page_fault(vcpu, gpa, error_code, NULL, 0);
5197 }
5198 
5199 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
5200 {
5201 	gpa_t gpa;
5202 
5203 	/*
5204 	 * A nested guest cannot optimize MMIO vmexits, because we have an
5205 	 * nGPA here instead of the required GPA.
5206 	 */
5207 	gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5208 	if (!is_guest_mode(vcpu) &&
5209 	    !kvm_io_bus_write(vcpu, KVM_FAST_MMIO_BUS, gpa, 0, NULL)) {
5210 		trace_kvm_fast_mmio(gpa);
5211 		return kvm_skip_emulated_instruction(vcpu);
5212 	}
5213 
5214 	return kvm_mmu_page_fault(vcpu, gpa, PFERR_RSVD_MASK, NULL, 0);
5215 }
5216 
5217 static int handle_nmi_window(struct kvm_vcpu *vcpu)
5218 {
5219 	WARN_ON_ONCE(!enable_vnmi);
5220 	exec_controls_clearbit(to_vmx(vcpu), CPU_BASED_NMI_WINDOW_EXITING);
5221 	++vcpu->stat.nmi_window_exits;
5222 	kvm_make_request(KVM_REQ_EVENT, vcpu);
5223 
5224 	return 1;
5225 }
5226 
5227 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
5228 {
5229 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5230 	bool intr_window_requested;
5231 	unsigned count = 130;
5232 
5233 	/*
5234 	 * We should never reach the point where we are emulating L2
5235 	 * due to invalid guest state as that means we incorrectly
5236 	 * allowed a nested VMEntry with an invalid vmcs12.
5237 	 */
5238 	WARN_ON_ONCE(vmx->emulation_required && vmx->nested.nested_run_pending);
5239 
5240 	intr_window_requested = exec_controls_get(vmx) &
5241 				CPU_BASED_INTR_WINDOW_EXITING;
5242 
5243 	while (vmx->emulation_required && count-- != 0) {
5244 		if (intr_window_requested && vmx_interrupt_allowed(vcpu))
5245 			return handle_interrupt_window(&vmx->vcpu);
5246 
5247 		if (kvm_test_request(KVM_REQ_EVENT, vcpu))
5248 			return 1;
5249 
5250 		if (!kvm_emulate_instruction(vcpu, 0))
5251 			return 0;
5252 
5253 		if (vmx->emulation_required && !vmx->rmode.vm86_active &&
5254 		    vcpu->arch.exception.pending) {
5255 			vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5256 			vcpu->run->internal.suberror =
5257 						KVM_INTERNAL_ERROR_EMULATION;
5258 			vcpu->run->internal.ndata = 0;
5259 			return 0;
5260 		}
5261 
5262 		if (vcpu->arch.halt_request) {
5263 			vcpu->arch.halt_request = 0;
5264 			return kvm_vcpu_halt(vcpu);
5265 		}
5266 
5267 		/*
5268 		 * Note, return 1 and not 0, vcpu_run() is responsible for
5269 		 * morphing the pending signal into the proper return code.
5270 		 */
5271 		if (signal_pending(current))
5272 			return 1;
5273 
5274 		if (need_resched())
5275 			schedule();
5276 	}
5277 
5278 	return 1;
5279 }
5280 
5281 static void grow_ple_window(struct kvm_vcpu *vcpu)
5282 {
5283 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5284 	unsigned int old = vmx->ple_window;
5285 
5286 	vmx->ple_window = __grow_ple_window(old, ple_window,
5287 					    ple_window_grow,
5288 					    ple_window_max);
5289 
5290 	if (vmx->ple_window != old) {
5291 		vmx->ple_window_dirty = true;
5292 		trace_kvm_ple_window_update(vcpu->vcpu_id,
5293 					    vmx->ple_window, old);
5294 	}
5295 }
5296 
5297 static void shrink_ple_window(struct kvm_vcpu *vcpu)
5298 {
5299 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5300 	unsigned int old = vmx->ple_window;
5301 
5302 	vmx->ple_window = __shrink_ple_window(old, ple_window,
5303 					      ple_window_shrink,
5304 					      ple_window);
5305 
5306 	if (vmx->ple_window != old) {
5307 		vmx->ple_window_dirty = true;
5308 		trace_kvm_ple_window_update(vcpu->vcpu_id,
5309 					    vmx->ple_window, old);
5310 	}
5311 }
5312 
5313 /*
5314  * Handler for POSTED_INTERRUPT_WAKEUP_VECTOR.
5315  */
5316 static void wakeup_handler(void)
5317 {
5318 	struct kvm_vcpu *vcpu;
5319 	int cpu = smp_processor_id();
5320 
5321 	spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
5322 	list_for_each_entry(vcpu, &per_cpu(blocked_vcpu_on_cpu, cpu),
5323 			blocked_vcpu_list) {
5324 		struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
5325 
5326 		if (pi_test_on(pi_desc) == 1)
5327 			kvm_vcpu_kick(vcpu);
5328 	}
5329 	spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
5330 }
5331 
5332 static void vmx_enable_tdp(void)
5333 {
5334 	kvm_mmu_set_mask_ptes(VMX_EPT_READABLE_MASK,
5335 		enable_ept_ad_bits ? VMX_EPT_ACCESS_BIT : 0ull,
5336 		enable_ept_ad_bits ? VMX_EPT_DIRTY_BIT : 0ull,
5337 		0ull, VMX_EPT_EXECUTABLE_MASK,
5338 		cpu_has_vmx_ept_execute_only() ? 0ull : VMX_EPT_READABLE_MASK,
5339 		VMX_EPT_RWX_MASK, 0ull);
5340 
5341 	ept_set_mmio_spte_mask();
5342 }
5343 
5344 /*
5345  * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
5346  * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
5347  */
5348 static int handle_pause(struct kvm_vcpu *vcpu)
5349 {
5350 	if (!kvm_pause_in_guest(vcpu->kvm))
5351 		grow_ple_window(vcpu);
5352 
5353 	/*
5354 	 * Intel sdm vol3 ch-25.1.3 says: The "PAUSE-loop exiting"
5355 	 * VM-execution control is ignored if CPL > 0. OTOH, KVM
5356 	 * never set PAUSE_EXITING and just set PLE if supported,
5357 	 * so the vcpu must be CPL=0 if it gets a PAUSE exit.
5358 	 */
5359 	kvm_vcpu_on_spin(vcpu, true);
5360 	return kvm_skip_emulated_instruction(vcpu);
5361 }
5362 
5363 static int handle_nop(struct kvm_vcpu *vcpu)
5364 {
5365 	return kvm_skip_emulated_instruction(vcpu);
5366 }
5367 
5368 static int handle_mwait(struct kvm_vcpu *vcpu)
5369 {
5370 	printk_once(KERN_WARNING "kvm: MWAIT instruction emulated as NOP!\n");
5371 	return handle_nop(vcpu);
5372 }
5373 
5374 static int handle_invalid_op(struct kvm_vcpu *vcpu)
5375 {
5376 	kvm_queue_exception(vcpu, UD_VECTOR);
5377 	return 1;
5378 }
5379 
5380 static int handle_monitor_trap(struct kvm_vcpu *vcpu)
5381 {
5382 	return 1;
5383 }
5384 
5385 static int handle_monitor(struct kvm_vcpu *vcpu)
5386 {
5387 	printk_once(KERN_WARNING "kvm: MONITOR instruction emulated as NOP!\n");
5388 	return handle_nop(vcpu);
5389 }
5390 
5391 static int handle_invpcid(struct kvm_vcpu *vcpu)
5392 {
5393 	u32 vmx_instruction_info;
5394 	unsigned long type;
5395 	bool pcid_enabled;
5396 	gva_t gva;
5397 	struct x86_exception e;
5398 	unsigned i;
5399 	unsigned long roots_to_free = 0;
5400 	struct {
5401 		u64 pcid;
5402 		u64 gla;
5403 	} operand;
5404 
5405 	if (!guest_cpuid_has(vcpu, X86_FEATURE_INVPCID)) {
5406 		kvm_queue_exception(vcpu, UD_VECTOR);
5407 		return 1;
5408 	}
5409 
5410 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5411 	type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
5412 
5413 	if (type > 3) {
5414 		kvm_inject_gp(vcpu, 0);
5415 		return 1;
5416 	}
5417 
5418 	/* According to the Intel instruction reference, the memory operand
5419 	 * is read even if it isn't needed (e.g., for type==all)
5420 	 */
5421 	if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
5422 				vmx_instruction_info, false,
5423 				sizeof(operand), &gva))
5424 		return 1;
5425 
5426 	if (kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e)) {
5427 		kvm_inject_page_fault(vcpu, &e);
5428 		return 1;
5429 	}
5430 
5431 	if (operand.pcid >> 12 != 0) {
5432 		kvm_inject_gp(vcpu, 0);
5433 		return 1;
5434 	}
5435 
5436 	pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
5437 
5438 	switch (type) {
5439 	case INVPCID_TYPE_INDIV_ADDR:
5440 		if ((!pcid_enabled && (operand.pcid != 0)) ||
5441 		    is_noncanonical_address(operand.gla, vcpu)) {
5442 			kvm_inject_gp(vcpu, 0);
5443 			return 1;
5444 		}
5445 		kvm_mmu_invpcid_gva(vcpu, operand.gla, operand.pcid);
5446 		return kvm_skip_emulated_instruction(vcpu);
5447 
5448 	case INVPCID_TYPE_SINGLE_CTXT:
5449 		if (!pcid_enabled && (operand.pcid != 0)) {
5450 			kvm_inject_gp(vcpu, 0);
5451 			return 1;
5452 		}
5453 
5454 		if (kvm_get_active_pcid(vcpu) == operand.pcid) {
5455 			kvm_mmu_sync_roots(vcpu);
5456 			kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
5457 		}
5458 
5459 		for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++)
5460 			if (kvm_get_pcid(vcpu, vcpu->arch.mmu->prev_roots[i].cr3)
5461 			    == operand.pcid)
5462 				roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i);
5463 
5464 		kvm_mmu_free_roots(vcpu, vcpu->arch.mmu, roots_to_free);
5465 		/*
5466 		 * If neither the current cr3 nor any of the prev_roots use the
5467 		 * given PCID, then nothing needs to be done here because a
5468 		 * resync will happen anyway before switching to any other CR3.
5469 		 */
5470 
5471 		return kvm_skip_emulated_instruction(vcpu);
5472 
5473 	case INVPCID_TYPE_ALL_NON_GLOBAL:
5474 		/*
5475 		 * Currently, KVM doesn't mark global entries in the shadow
5476 		 * page tables, so a non-global flush just degenerates to a
5477 		 * global flush. If needed, we could optimize this later by
5478 		 * keeping track of global entries in shadow page tables.
5479 		 */
5480 
5481 		/* fall-through */
5482 	case INVPCID_TYPE_ALL_INCL_GLOBAL:
5483 		kvm_mmu_unload(vcpu);
5484 		return kvm_skip_emulated_instruction(vcpu);
5485 
5486 	default:
5487 		BUG(); /* We have already checked above that type <= 3 */
5488 	}
5489 }
5490 
5491 static int handle_pml_full(struct kvm_vcpu *vcpu)
5492 {
5493 	unsigned long exit_qualification;
5494 
5495 	trace_kvm_pml_full(vcpu->vcpu_id);
5496 
5497 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5498 
5499 	/*
5500 	 * PML buffer FULL happened while executing iret from NMI,
5501 	 * "blocked by NMI" bit has to be set before next VM entry.
5502 	 */
5503 	if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
5504 			enable_vnmi &&
5505 			(exit_qualification & INTR_INFO_UNBLOCK_NMI))
5506 		vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
5507 				GUEST_INTR_STATE_NMI);
5508 
5509 	/*
5510 	 * PML buffer already flushed at beginning of VMEXIT. Nothing to do
5511 	 * here.., and there's no userspace involvement needed for PML.
5512 	 */
5513 	return 1;
5514 }
5515 
5516 static int handle_preemption_timer(struct kvm_vcpu *vcpu)
5517 {
5518 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5519 
5520 	if (!vmx->req_immediate_exit &&
5521 	    !unlikely(vmx->loaded_vmcs->hv_timer_soft_disabled))
5522 		kvm_lapic_expired_hv_timer(vcpu);
5523 
5524 	return 1;
5525 }
5526 
5527 /*
5528  * When nested=0, all VMX instruction VM Exits filter here.  The handlers
5529  * are overwritten by nested_vmx_setup() when nested=1.
5530  */
5531 static int handle_vmx_instruction(struct kvm_vcpu *vcpu)
5532 {
5533 	kvm_queue_exception(vcpu, UD_VECTOR);
5534 	return 1;
5535 }
5536 
5537 static int handle_encls(struct kvm_vcpu *vcpu)
5538 {
5539 	/*
5540 	 * SGX virtualization is not yet supported.  There is no software
5541 	 * enable bit for SGX, so we have to trap ENCLS and inject a #UD
5542 	 * to prevent the guest from executing ENCLS.
5543 	 */
5544 	kvm_queue_exception(vcpu, UD_VECTOR);
5545 	return 1;
5546 }
5547 
5548 /*
5549  * The exit handlers return 1 if the exit was handled fully and guest execution
5550  * may resume.  Otherwise they set the kvm_run parameter to indicate what needs
5551  * to be done to userspace and return 0.
5552  */
5553 static int (*kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
5554 	[EXIT_REASON_EXCEPTION_NMI]           = handle_exception_nmi,
5555 	[EXIT_REASON_EXTERNAL_INTERRUPT]      = handle_external_interrupt,
5556 	[EXIT_REASON_TRIPLE_FAULT]            = handle_triple_fault,
5557 	[EXIT_REASON_NMI_WINDOW]	      = handle_nmi_window,
5558 	[EXIT_REASON_IO_INSTRUCTION]          = handle_io,
5559 	[EXIT_REASON_CR_ACCESS]               = handle_cr,
5560 	[EXIT_REASON_DR_ACCESS]               = handle_dr,
5561 	[EXIT_REASON_CPUID]                   = kvm_emulate_cpuid,
5562 	[EXIT_REASON_MSR_READ]                = kvm_emulate_rdmsr,
5563 	[EXIT_REASON_MSR_WRITE]               = kvm_emulate_wrmsr,
5564 	[EXIT_REASON_INTERRUPT_WINDOW]        = handle_interrupt_window,
5565 	[EXIT_REASON_HLT]                     = kvm_emulate_halt,
5566 	[EXIT_REASON_INVD]		      = handle_invd,
5567 	[EXIT_REASON_INVLPG]		      = handle_invlpg,
5568 	[EXIT_REASON_RDPMC]                   = handle_rdpmc,
5569 	[EXIT_REASON_VMCALL]                  = handle_vmcall,
5570 	[EXIT_REASON_VMCLEAR]		      = handle_vmx_instruction,
5571 	[EXIT_REASON_VMLAUNCH]		      = handle_vmx_instruction,
5572 	[EXIT_REASON_VMPTRLD]		      = handle_vmx_instruction,
5573 	[EXIT_REASON_VMPTRST]		      = handle_vmx_instruction,
5574 	[EXIT_REASON_VMREAD]		      = handle_vmx_instruction,
5575 	[EXIT_REASON_VMRESUME]		      = handle_vmx_instruction,
5576 	[EXIT_REASON_VMWRITE]		      = handle_vmx_instruction,
5577 	[EXIT_REASON_VMOFF]		      = handle_vmx_instruction,
5578 	[EXIT_REASON_VMON]		      = handle_vmx_instruction,
5579 	[EXIT_REASON_TPR_BELOW_THRESHOLD]     = handle_tpr_below_threshold,
5580 	[EXIT_REASON_APIC_ACCESS]             = handle_apic_access,
5581 	[EXIT_REASON_APIC_WRITE]              = handle_apic_write,
5582 	[EXIT_REASON_EOI_INDUCED]             = handle_apic_eoi_induced,
5583 	[EXIT_REASON_WBINVD]                  = handle_wbinvd,
5584 	[EXIT_REASON_XSETBV]                  = handle_xsetbv,
5585 	[EXIT_REASON_TASK_SWITCH]             = handle_task_switch,
5586 	[EXIT_REASON_MCE_DURING_VMENTRY]      = handle_machine_check,
5587 	[EXIT_REASON_GDTR_IDTR]		      = handle_desc,
5588 	[EXIT_REASON_LDTR_TR]		      = handle_desc,
5589 	[EXIT_REASON_EPT_VIOLATION]	      = handle_ept_violation,
5590 	[EXIT_REASON_EPT_MISCONFIG]           = handle_ept_misconfig,
5591 	[EXIT_REASON_PAUSE_INSTRUCTION]       = handle_pause,
5592 	[EXIT_REASON_MWAIT_INSTRUCTION]	      = handle_mwait,
5593 	[EXIT_REASON_MONITOR_TRAP_FLAG]       = handle_monitor_trap,
5594 	[EXIT_REASON_MONITOR_INSTRUCTION]     = handle_monitor,
5595 	[EXIT_REASON_INVEPT]                  = handle_vmx_instruction,
5596 	[EXIT_REASON_INVVPID]                 = handle_vmx_instruction,
5597 	[EXIT_REASON_RDRAND]                  = handle_invalid_op,
5598 	[EXIT_REASON_RDSEED]                  = handle_invalid_op,
5599 	[EXIT_REASON_PML_FULL]		      = handle_pml_full,
5600 	[EXIT_REASON_INVPCID]                 = handle_invpcid,
5601 	[EXIT_REASON_VMFUNC]		      = handle_vmx_instruction,
5602 	[EXIT_REASON_PREEMPTION_TIMER]	      = handle_preemption_timer,
5603 	[EXIT_REASON_ENCLS]		      = handle_encls,
5604 };
5605 
5606 static const int kvm_vmx_max_exit_handlers =
5607 	ARRAY_SIZE(kvm_vmx_exit_handlers);
5608 
5609 static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
5610 {
5611 	*info1 = vmcs_readl(EXIT_QUALIFICATION);
5612 	*info2 = vmcs_read32(VM_EXIT_INTR_INFO);
5613 }
5614 
5615 static void vmx_destroy_pml_buffer(struct vcpu_vmx *vmx)
5616 {
5617 	if (vmx->pml_pg) {
5618 		__free_page(vmx->pml_pg);
5619 		vmx->pml_pg = NULL;
5620 	}
5621 }
5622 
5623 static void vmx_flush_pml_buffer(struct kvm_vcpu *vcpu)
5624 {
5625 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5626 	u64 *pml_buf;
5627 	u16 pml_idx;
5628 
5629 	pml_idx = vmcs_read16(GUEST_PML_INDEX);
5630 
5631 	/* Do nothing if PML buffer is empty */
5632 	if (pml_idx == (PML_ENTITY_NUM - 1))
5633 		return;
5634 
5635 	/* PML index always points to next available PML buffer entity */
5636 	if (pml_idx >= PML_ENTITY_NUM)
5637 		pml_idx = 0;
5638 	else
5639 		pml_idx++;
5640 
5641 	pml_buf = page_address(vmx->pml_pg);
5642 	for (; pml_idx < PML_ENTITY_NUM; pml_idx++) {
5643 		u64 gpa;
5644 
5645 		gpa = pml_buf[pml_idx];
5646 		WARN_ON(gpa & (PAGE_SIZE - 1));
5647 		kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
5648 	}
5649 
5650 	/* reset PML index */
5651 	vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
5652 }
5653 
5654 /*
5655  * Flush all vcpus' PML buffer and update logged GPAs to dirty_bitmap.
5656  * Called before reporting dirty_bitmap to userspace.
5657  */
5658 static void kvm_flush_pml_buffers(struct kvm *kvm)
5659 {
5660 	int i;
5661 	struct kvm_vcpu *vcpu;
5662 	/*
5663 	 * We only need to kick vcpu out of guest mode here, as PML buffer
5664 	 * is flushed at beginning of all VMEXITs, and it's obvious that only
5665 	 * vcpus running in guest are possible to have unflushed GPAs in PML
5666 	 * buffer.
5667 	 */
5668 	kvm_for_each_vcpu(i, vcpu, kvm)
5669 		kvm_vcpu_kick(vcpu);
5670 }
5671 
5672 static void vmx_dump_sel(char *name, uint32_t sel)
5673 {
5674 	pr_err("%s sel=0x%04x, attr=0x%05x, limit=0x%08x, base=0x%016lx\n",
5675 	       name, vmcs_read16(sel),
5676 	       vmcs_read32(sel + GUEST_ES_AR_BYTES - GUEST_ES_SELECTOR),
5677 	       vmcs_read32(sel + GUEST_ES_LIMIT - GUEST_ES_SELECTOR),
5678 	       vmcs_readl(sel + GUEST_ES_BASE - GUEST_ES_SELECTOR));
5679 }
5680 
5681 static void vmx_dump_dtsel(char *name, uint32_t limit)
5682 {
5683 	pr_err("%s                           limit=0x%08x, base=0x%016lx\n",
5684 	       name, vmcs_read32(limit),
5685 	       vmcs_readl(limit + GUEST_GDTR_BASE - GUEST_GDTR_LIMIT));
5686 }
5687 
5688 void dump_vmcs(void)
5689 {
5690 	u32 vmentry_ctl, vmexit_ctl;
5691 	u32 cpu_based_exec_ctrl, pin_based_exec_ctrl, secondary_exec_control;
5692 	unsigned long cr4;
5693 	u64 efer;
5694 	int i, n;
5695 
5696 	if (!dump_invalid_vmcs) {
5697 		pr_warn_ratelimited("set kvm_intel.dump_invalid_vmcs=1 to dump internal KVM state.\n");
5698 		return;
5699 	}
5700 
5701 	vmentry_ctl = vmcs_read32(VM_ENTRY_CONTROLS);
5702 	vmexit_ctl = vmcs_read32(VM_EXIT_CONTROLS);
5703 	cpu_based_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5704 	pin_based_exec_ctrl = vmcs_read32(PIN_BASED_VM_EXEC_CONTROL);
5705 	cr4 = vmcs_readl(GUEST_CR4);
5706 	efer = vmcs_read64(GUEST_IA32_EFER);
5707 	secondary_exec_control = 0;
5708 	if (cpu_has_secondary_exec_ctrls())
5709 		secondary_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
5710 
5711 	pr_err("*** Guest State ***\n");
5712 	pr_err("CR0: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
5713 	       vmcs_readl(GUEST_CR0), vmcs_readl(CR0_READ_SHADOW),
5714 	       vmcs_readl(CR0_GUEST_HOST_MASK));
5715 	pr_err("CR4: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
5716 	       cr4, vmcs_readl(CR4_READ_SHADOW), vmcs_readl(CR4_GUEST_HOST_MASK));
5717 	pr_err("CR3 = 0x%016lx\n", vmcs_readl(GUEST_CR3));
5718 	if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT) &&
5719 	    (cr4 & X86_CR4_PAE) && !(efer & EFER_LMA))
5720 	{
5721 		pr_err("PDPTR0 = 0x%016llx  PDPTR1 = 0x%016llx\n",
5722 		       vmcs_read64(GUEST_PDPTR0), vmcs_read64(GUEST_PDPTR1));
5723 		pr_err("PDPTR2 = 0x%016llx  PDPTR3 = 0x%016llx\n",
5724 		       vmcs_read64(GUEST_PDPTR2), vmcs_read64(GUEST_PDPTR3));
5725 	}
5726 	pr_err("RSP = 0x%016lx  RIP = 0x%016lx\n",
5727 	       vmcs_readl(GUEST_RSP), vmcs_readl(GUEST_RIP));
5728 	pr_err("RFLAGS=0x%08lx         DR7 = 0x%016lx\n",
5729 	       vmcs_readl(GUEST_RFLAGS), vmcs_readl(GUEST_DR7));
5730 	pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
5731 	       vmcs_readl(GUEST_SYSENTER_ESP),
5732 	       vmcs_read32(GUEST_SYSENTER_CS), vmcs_readl(GUEST_SYSENTER_EIP));
5733 	vmx_dump_sel("CS:  ", GUEST_CS_SELECTOR);
5734 	vmx_dump_sel("DS:  ", GUEST_DS_SELECTOR);
5735 	vmx_dump_sel("SS:  ", GUEST_SS_SELECTOR);
5736 	vmx_dump_sel("ES:  ", GUEST_ES_SELECTOR);
5737 	vmx_dump_sel("FS:  ", GUEST_FS_SELECTOR);
5738 	vmx_dump_sel("GS:  ", GUEST_GS_SELECTOR);
5739 	vmx_dump_dtsel("GDTR:", GUEST_GDTR_LIMIT);
5740 	vmx_dump_sel("LDTR:", GUEST_LDTR_SELECTOR);
5741 	vmx_dump_dtsel("IDTR:", GUEST_IDTR_LIMIT);
5742 	vmx_dump_sel("TR:  ", GUEST_TR_SELECTOR);
5743 	if ((vmexit_ctl & (VM_EXIT_SAVE_IA32_PAT | VM_EXIT_SAVE_IA32_EFER)) ||
5744 	    (vmentry_ctl & (VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_IA32_EFER)))
5745 		pr_err("EFER =     0x%016llx  PAT = 0x%016llx\n",
5746 		       efer, vmcs_read64(GUEST_IA32_PAT));
5747 	pr_err("DebugCtl = 0x%016llx  DebugExceptions = 0x%016lx\n",
5748 	       vmcs_read64(GUEST_IA32_DEBUGCTL),
5749 	       vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS));
5750 	if (cpu_has_load_perf_global_ctrl() &&
5751 	    vmentry_ctl & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
5752 		pr_err("PerfGlobCtl = 0x%016llx\n",
5753 		       vmcs_read64(GUEST_IA32_PERF_GLOBAL_CTRL));
5754 	if (vmentry_ctl & VM_ENTRY_LOAD_BNDCFGS)
5755 		pr_err("BndCfgS = 0x%016llx\n", vmcs_read64(GUEST_BNDCFGS));
5756 	pr_err("Interruptibility = %08x  ActivityState = %08x\n",
5757 	       vmcs_read32(GUEST_INTERRUPTIBILITY_INFO),
5758 	       vmcs_read32(GUEST_ACTIVITY_STATE));
5759 	if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)
5760 		pr_err("InterruptStatus = %04x\n",
5761 		       vmcs_read16(GUEST_INTR_STATUS));
5762 
5763 	pr_err("*** Host State ***\n");
5764 	pr_err("RIP = 0x%016lx  RSP = 0x%016lx\n",
5765 	       vmcs_readl(HOST_RIP), vmcs_readl(HOST_RSP));
5766 	pr_err("CS=%04x SS=%04x DS=%04x ES=%04x FS=%04x GS=%04x TR=%04x\n",
5767 	       vmcs_read16(HOST_CS_SELECTOR), vmcs_read16(HOST_SS_SELECTOR),
5768 	       vmcs_read16(HOST_DS_SELECTOR), vmcs_read16(HOST_ES_SELECTOR),
5769 	       vmcs_read16(HOST_FS_SELECTOR), vmcs_read16(HOST_GS_SELECTOR),
5770 	       vmcs_read16(HOST_TR_SELECTOR));
5771 	pr_err("FSBase=%016lx GSBase=%016lx TRBase=%016lx\n",
5772 	       vmcs_readl(HOST_FS_BASE), vmcs_readl(HOST_GS_BASE),
5773 	       vmcs_readl(HOST_TR_BASE));
5774 	pr_err("GDTBase=%016lx IDTBase=%016lx\n",
5775 	       vmcs_readl(HOST_GDTR_BASE), vmcs_readl(HOST_IDTR_BASE));
5776 	pr_err("CR0=%016lx CR3=%016lx CR4=%016lx\n",
5777 	       vmcs_readl(HOST_CR0), vmcs_readl(HOST_CR3),
5778 	       vmcs_readl(HOST_CR4));
5779 	pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
5780 	       vmcs_readl(HOST_IA32_SYSENTER_ESP),
5781 	       vmcs_read32(HOST_IA32_SYSENTER_CS),
5782 	       vmcs_readl(HOST_IA32_SYSENTER_EIP));
5783 	if (vmexit_ctl & (VM_EXIT_LOAD_IA32_PAT | VM_EXIT_LOAD_IA32_EFER))
5784 		pr_err("EFER = 0x%016llx  PAT = 0x%016llx\n",
5785 		       vmcs_read64(HOST_IA32_EFER),
5786 		       vmcs_read64(HOST_IA32_PAT));
5787 	if (cpu_has_load_perf_global_ctrl() &&
5788 	    vmexit_ctl & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
5789 		pr_err("PerfGlobCtl = 0x%016llx\n",
5790 		       vmcs_read64(HOST_IA32_PERF_GLOBAL_CTRL));
5791 
5792 	pr_err("*** Control State ***\n");
5793 	pr_err("PinBased=%08x CPUBased=%08x SecondaryExec=%08x\n",
5794 	       pin_based_exec_ctrl, cpu_based_exec_ctrl, secondary_exec_control);
5795 	pr_err("EntryControls=%08x ExitControls=%08x\n", vmentry_ctl, vmexit_ctl);
5796 	pr_err("ExceptionBitmap=%08x PFECmask=%08x PFECmatch=%08x\n",
5797 	       vmcs_read32(EXCEPTION_BITMAP),
5798 	       vmcs_read32(PAGE_FAULT_ERROR_CODE_MASK),
5799 	       vmcs_read32(PAGE_FAULT_ERROR_CODE_MATCH));
5800 	pr_err("VMEntry: intr_info=%08x errcode=%08x ilen=%08x\n",
5801 	       vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
5802 	       vmcs_read32(VM_ENTRY_EXCEPTION_ERROR_CODE),
5803 	       vmcs_read32(VM_ENTRY_INSTRUCTION_LEN));
5804 	pr_err("VMExit: intr_info=%08x errcode=%08x ilen=%08x\n",
5805 	       vmcs_read32(VM_EXIT_INTR_INFO),
5806 	       vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
5807 	       vmcs_read32(VM_EXIT_INSTRUCTION_LEN));
5808 	pr_err("        reason=%08x qualification=%016lx\n",
5809 	       vmcs_read32(VM_EXIT_REASON), vmcs_readl(EXIT_QUALIFICATION));
5810 	pr_err("IDTVectoring: info=%08x errcode=%08x\n",
5811 	       vmcs_read32(IDT_VECTORING_INFO_FIELD),
5812 	       vmcs_read32(IDT_VECTORING_ERROR_CODE));
5813 	pr_err("TSC Offset = 0x%016llx\n", vmcs_read64(TSC_OFFSET));
5814 	if (secondary_exec_control & SECONDARY_EXEC_TSC_SCALING)
5815 		pr_err("TSC Multiplier = 0x%016llx\n",
5816 		       vmcs_read64(TSC_MULTIPLIER));
5817 	if (cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW) {
5818 		if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY) {
5819 			u16 status = vmcs_read16(GUEST_INTR_STATUS);
5820 			pr_err("SVI|RVI = %02x|%02x ", status >> 8, status & 0xff);
5821 		}
5822 		pr_cont("TPR Threshold = 0x%02x\n", vmcs_read32(TPR_THRESHOLD));
5823 		if (secondary_exec_control & SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)
5824 			pr_err("APIC-access addr = 0x%016llx ", vmcs_read64(APIC_ACCESS_ADDR));
5825 		pr_cont("virt-APIC addr = 0x%016llx\n", vmcs_read64(VIRTUAL_APIC_PAGE_ADDR));
5826 	}
5827 	if (pin_based_exec_ctrl & PIN_BASED_POSTED_INTR)
5828 		pr_err("PostedIntrVec = 0x%02x\n", vmcs_read16(POSTED_INTR_NV));
5829 	if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT))
5830 		pr_err("EPT pointer = 0x%016llx\n", vmcs_read64(EPT_POINTER));
5831 	n = vmcs_read32(CR3_TARGET_COUNT);
5832 	for (i = 0; i + 1 < n; i += 4)
5833 		pr_err("CR3 target%u=%016lx target%u=%016lx\n",
5834 		       i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2),
5835 		       i + 1, vmcs_readl(CR3_TARGET_VALUE0 + i * 2 + 2));
5836 	if (i < n)
5837 		pr_err("CR3 target%u=%016lx\n",
5838 		       i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2));
5839 	if (secondary_exec_control & SECONDARY_EXEC_PAUSE_LOOP_EXITING)
5840 		pr_err("PLE Gap=%08x Window=%08x\n",
5841 		       vmcs_read32(PLE_GAP), vmcs_read32(PLE_WINDOW));
5842 	if (secondary_exec_control & SECONDARY_EXEC_ENABLE_VPID)
5843 		pr_err("Virtual processor ID = 0x%04x\n",
5844 		       vmcs_read16(VIRTUAL_PROCESSOR_ID));
5845 }
5846 
5847 /*
5848  * The guest has exited.  See if we can fix it or if we need userspace
5849  * assistance.
5850  */
5851 static int vmx_handle_exit(struct kvm_vcpu *vcpu,
5852 	enum exit_fastpath_completion exit_fastpath)
5853 {
5854 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5855 	u32 exit_reason = vmx->exit_reason;
5856 	u32 vectoring_info = vmx->idt_vectoring_info;
5857 
5858 	trace_kvm_exit(exit_reason, vcpu, KVM_ISA_VMX);
5859 
5860 	/*
5861 	 * Flush logged GPAs PML buffer, this will make dirty_bitmap more
5862 	 * updated. Another good is, in kvm_vm_ioctl_get_dirty_log, before
5863 	 * querying dirty_bitmap, we only need to kick all vcpus out of guest
5864 	 * mode as if vcpus is in root mode, the PML buffer must has been
5865 	 * flushed already.
5866 	 */
5867 	if (enable_pml)
5868 		vmx_flush_pml_buffer(vcpu);
5869 
5870 	/* If guest state is invalid, start emulating */
5871 	if (vmx->emulation_required)
5872 		return handle_invalid_guest_state(vcpu);
5873 
5874 	if (is_guest_mode(vcpu)) {
5875 		/*
5876 		 * The host physical addresses of some pages of guest memory
5877 		 * are loaded into the vmcs02 (e.g. vmcs12's Virtual APIC
5878 		 * Page). The CPU may write to these pages via their host
5879 		 * physical address while L2 is running, bypassing any
5880 		 * address-translation-based dirty tracking (e.g. EPT write
5881 		 * protection).
5882 		 *
5883 		 * Mark them dirty on every exit from L2 to prevent them from
5884 		 * getting out of sync with dirty tracking.
5885 		 */
5886 		nested_mark_vmcs12_pages_dirty(vcpu);
5887 
5888 		if (nested_vmx_exit_reflected(vcpu, exit_reason))
5889 			return nested_vmx_reflect_vmexit(vcpu, exit_reason);
5890 	}
5891 
5892 	if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) {
5893 		dump_vmcs();
5894 		vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
5895 		vcpu->run->fail_entry.hardware_entry_failure_reason
5896 			= exit_reason;
5897 		return 0;
5898 	}
5899 
5900 	if (unlikely(vmx->fail)) {
5901 		dump_vmcs();
5902 		vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
5903 		vcpu->run->fail_entry.hardware_entry_failure_reason
5904 			= vmcs_read32(VM_INSTRUCTION_ERROR);
5905 		return 0;
5906 	}
5907 
5908 	/*
5909 	 * Note:
5910 	 * Do not try to fix EXIT_REASON_EPT_MISCONFIG if it caused by
5911 	 * delivery event since it indicates guest is accessing MMIO.
5912 	 * The vm-exit can be triggered again after return to guest that
5913 	 * will cause infinite loop.
5914 	 */
5915 	if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
5916 			(exit_reason != EXIT_REASON_EXCEPTION_NMI &&
5917 			exit_reason != EXIT_REASON_EPT_VIOLATION &&
5918 			exit_reason != EXIT_REASON_PML_FULL &&
5919 			exit_reason != EXIT_REASON_TASK_SWITCH)) {
5920 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5921 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_DELIVERY_EV;
5922 		vcpu->run->internal.ndata = 3;
5923 		vcpu->run->internal.data[0] = vectoring_info;
5924 		vcpu->run->internal.data[1] = exit_reason;
5925 		vcpu->run->internal.data[2] = vcpu->arch.exit_qualification;
5926 		if (exit_reason == EXIT_REASON_EPT_MISCONFIG) {
5927 			vcpu->run->internal.ndata++;
5928 			vcpu->run->internal.data[3] =
5929 				vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5930 		}
5931 		return 0;
5932 	}
5933 
5934 	if (unlikely(!enable_vnmi &&
5935 		     vmx->loaded_vmcs->soft_vnmi_blocked)) {
5936 		if (vmx_interrupt_allowed(vcpu)) {
5937 			vmx->loaded_vmcs->soft_vnmi_blocked = 0;
5938 		} else if (vmx->loaded_vmcs->vnmi_blocked_time > 1000000000LL &&
5939 			   vcpu->arch.nmi_pending) {
5940 			/*
5941 			 * This CPU don't support us in finding the end of an
5942 			 * NMI-blocked window if the guest runs with IRQs
5943 			 * disabled. So we pull the trigger after 1 s of
5944 			 * futile waiting, but inform the user about this.
5945 			 */
5946 			printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
5947 			       "state on VCPU %d after 1 s timeout\n",
5948 			       __func__, vcpu->vcpu_id);
5949 			vmx->loaded_vmcs->soft_vnmi_blocked = 0;
5950 		}
5951 	}
5952 
5953 	if (exit_fastpath == EXIT_FASTPATH_SKIP_EMUL_INS) {
5954 		kvm_skip_emulated_instruction(vcpu);
5955 		return 1;
5956 	}
5957 
5958 	if (exit_reason >= kvm_vmx_max_exit_handlers)
5959 		goto unexpected_vmexit;
5960 #ifdef CONFIG_RETPOLINE
5961 	if (exit_reason == EXIT_REASON_MSR_WRITE)
5962 		return kvm_emulate_wrmsr(vcpu);
5963 	else if (exit_reason == EXIT_REASON_PREEMPTION_TIMER)
5964 		return handle_preemption_timer(vcpu);
5965 	else if (exit_reason == EXIT_REASON_INTERRUPT_WINDOW)
5966 		return handle_interrupt_window(vcpu);
5967 	else if (exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT)
5968 		return handle_external_interrupt(vcpu);
5969 	else if (exit_reason == EXIT_REASON_HLT)
5970 		return kvm_emulate_halt(vcpu);
5971 	else if (exit_reason == EXIT_REASON_EPT_MISCONFIG)
5972 		return handle_ept_misconfig(vcpu);
5973 #endif
5974 
5975 	exit_reason = array_index_nospec(exit_reason,
5976 					 kvm_vmx_max_exit_handlers);
5977 	if (!kvm_vmx_exit_handlers[exit_reason])
5978 		goto unexpected_vmexit;
5979 
5980 	return kvm_vmx_exit_handlers[exit_reason](vcpu);
5981 
5982 unexpected_vmexit:
5983 	vcpu_unimpl(vcpu, "vmx: unexpected exit reason 0x%x\n", exit_reason);
5984 	dump_vmcs();
5985 	vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5986 	vcpu->run->internal.suberror =
5987 			KVM_INTERNAL_ERROR_UNEXPECTED_EXIT_REASON;
5988 	vcpu->run->internal.ndata = 1;
5989 	vcpu->run->internal.data[0] = exit_reason;
5990 	return 0;
5991 }
5992 
5993 /*
5994  * Software based L1D cache flush which is used when microcode providing
5995  * the cache control MSR is not loaded.
5996  *
5997  * The L1D cache is 32 KiB on Nehalem and later microarchitectures, but to
5998  * flush it is required to read in 64 KiB because the replacement algorithm
5999  * is not exactly LRU. This could be sized at runtime via topology
6000  * information but as all relevant affected CPUs have 32KiB L1D cache size
6001  * there is no point in doing so.
6002  */
6003 static void vmx_l1d_flush(struct kvm_vcpu *vcpu)
6004 {
6005 	int size = PAGE_SIZE << L1D_CACHE_ORDER;
6006 
6007 	/*
6008 	 * This code is only executed when the the flush mode is 'cond' or
6009 	 * 'always'
6010 	 */
6011 	if (static_branch_likely(&vmx_l1d_flush_cond)) {
6012 		bool flush_l1d;
6013 
6014 		/*
6015 		 * Clear the per-vcpu flush bit, it gets set again
6016 		 * either from vcpu_run() or from one of the unsafe
6017 		 * VMEXIT handlers.
6018 		 */
6019 		flush_l1d = vcpu->arch.l1tf_flush_l1d;
6020 		vcpu->arch.l1tf_flush_l1d = false;
6021 
6022 		/*
6023 		 * Clear the per-cpu flush bit, it gets set again from
6024 		 * the interrupt handlers.
6025 		 */
6026 		flush_l1d |= kvm_get_cpu_l1tf_flush_l1d();
6027 		kvm_clear_cpu_l1tf_flush_l1d();
6028 
6029 		if (!flush_l1d)
6030 			return;
6031 	}
6032 
6033 	vcpu->stat.l1d_flush++;
6034 
6035 	if (static_cpu_has(X86_FEATURE_FLUSH_L1D)) {
6036 		wrmsrl(MSR_IA32_FLUSH_CMD, L1D_FLUSH);
6037 		return;
6038 	}
6039 
6040 	asm volatile(
6041 		/* First ensure the pages are in the TLB */
6042 		"xorl	%%eax, %%eax\n"
6043 		".Lpopulate_tlb:\n\t"
6044 		"movzbl	(%[flush_pages], %%" _ASM_AX "), %%ecx\n\t"
6045 		"addl	$4096, %%eax\n\t"
6046 		"cmpl	%%eax, %[size]\n\t"
6047 		"jne	.Lpopulate_tlb\n\t"
6048 		"xorl	%%eax, %%eax\n\t"
6049 		"cpuid\n\t"
6050 		/* Now fill the cache */
6051 		"xorl	%%eax, %%eax\n"
6052 		".Lfill_cache:\n"
6053 		"movzbl	(%[flush_pages], %%" _ASM_AX "), %%ecx\n\t"
6054 		"addl	$64, %%eax\n\t"
6055 		"cmpl	%%eax, %[size]\n\t"
6056 		"jne	.Lfill_cache\n\t"
6057 		"lfence\n"
6058 		:: [flush_pages] "r" (vmx_l1d_flush_pages),
6059 		    [size] "r" (size)
6060 		: "eax", "ebx", "ecx", "edx");
6061 }
6062 
6063 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
6064 {
6065 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6066 	int tpr_threshold;
6067 
6068 	if (is_guest_mode(vcpu) &&
6069 		nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
6070 		return;
6071 
6072 	tpr_threshold = (irr == -1 || tpr < irr) ? 0 : irr;
6073 	if (is_guest_mode(vcpu))
6074 		to_vmx(vcpu)->nested.l1_tpr_threshold = tpr_threshold;
6075 	else
6076 		vmcs_write32(TPR_THRESHOLD, tpr_threshold);
6077 }
6078 
6079 void vmx_set_virtual_apic_mode(struct kvm_vcpu *vcpu)
6080 {
6081 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6082 	u32 sec_exec_control;
6083 
6084 	if (!lapic_in_kernel(vcpu))
6085 		return;
6086 
6087 	if (!flexpriority_enabled &&
6088 	    !cpu_has_vmx_virtualize_x2apic_mode())
6089 		return;
6090 
6091 	/* Postpone execution until vmcs01 is the current VMCS. */
6092 	if (is_guest_mode(vcpu)) {
6093 		vmx->nested.change_vmcs01_virtual_apic_mode = true;
6094 		return;
6095 	}
6096 
6097 	sec_exec_control = secondary_exec_controls_get(vmx);
6098 	sec_exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
6099 			      SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE);
6100 
6101 	switch (kvm_get_apic_mode(vcpu)) {
6102 	case LAPIC_MODE_INVALID:
6103 		WARN_ONCE(true, "Invalid local APIC state");
6104 	case LAPIC_MODE_DISABLED:
6105 		break;
6106 	case LAPIC_MODE_XAPIC:
6107 		if (flexpriority_enabled) {
6108 			sec_exec_control |=
6109 				SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
6110 			vmx_flush_tlb(vcpu, true);
6111 		}
6112 		break;
6113 	case LAPIC_MODE_X2APIC:
6114 		if (cpu_has_vmx_virtualize_x2apic_mode())
6115 			sec_exec_control |=
6116 				SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
6117 		break;
6118 	}
6119 	secondary_exec_controls_set(vmx, sec_exec_control);
6120 
6121 	vmx_update_msr_bitmap(vcpu);
6122 }
6123 
6124 static void vmx_set_apic_access_page_addr(struct kvm_vcpu *vcpu, hpa_t hpa)
6125 {
6126 	if (!is_guest_mode(vcpu)) {
6127 		vmcs_write64(APIC_ACCESS_ADDR, hpa);
6128 		vmx_flush_tlb(vcpu, true);
6129 	}
6130 }
6131 
6132 static void vmx_hwapic_isr_update(struct kvm_vcpu *vcpu, int max_isr)
6133 {
6134 	u16 status;
6135 	u8 old;
6136 
6137 	if (max_isr == -1)
6138 		max_isr = 0;
6139 
6140 	status = vmcs_read16(GUEST_INTR_STATUS);
6141 	old = status >> 8;
6142 	if (max_isr != old) {
6143 		status &= 0xff;
6144 		status |= max_isr << 8;
6145 		vmcs_write16(GUEST_INTR_STATUS, status);
6146 	}
6147 }
6148 
6149 static void vmx_set_rvi(int vector)
6150 {
6151 	u16 status;
6152 	u8 old;
6153 
6154 	if (vector == -1)
6155 		vector = 0;
6156 
6157 	status = vmcs_read16(GUEST_INTR_STATUS);
6158 	old = (u8)status & 0xff;
6159 	if ((u8)vector != old) {
6160 		status &= ~0xff;
6161 		status |= (u8)vector;
6162 		vmcs_write16(GUEST_INTR_STATUS, status);
6163 	}
6164 }
6165 
6166 static void vmx_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr)
6167 {
6168 	/*
6169 	 * When running L2, updating RVI is only relevant when
6170 	 * vmcs12 virtual-interrupt-delivery enabled.
6171 	 * However, it can be enabled only when L1 also
6172 	 * intercepts external-interrupts and in that case
6173 	 * we should not update vmcs02 RVI but instead intercept
6174 	 * interrupt. Therefore, do nothing when running L2.
6175 	 */
6176 	if (!is_guest_mode(vcpu))
6177 		vmx_set_rvi(max_irr);
6178 }
6179 
6180 static int vmx_sync_pir_to_irr(struct kvm_vcpu *vcpu)
6181 {
6182 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6183 	int max_irr;
6184 	bool max_irr_updated;
6185 
6186 	WARN_ON(!vcpu->arch.apicv_active);
6187 	if (pi_test_on(&vmx->pi_desc)) {
6188 		pi_clear_on(&vmx->pi_desc);
6189 		/*
6190 		 * IOMMU can write to PID.ON, so the barrier matters even on UP.
6191 		 * But on x86 this is just a compiler barrier anyway.
6192 		 */
6193 		smp_mb__after_atomic();
6194 		max_irr_updated =
6195 			kvm_apic_update_irr(vcpu, vmx->pi_desc.pir, &max_irr);
6196 
6197 		/*
6198 		 * If we are running L2 and L1 has a new pending interrupt
6199 		 * which can be injected, we should re-evaluate
6200 		 * what should be done with this new L1 interrupt.
6201 		 * If L1 intercepts external-interrupts, we should
6202 		 * exit from L2 to L1. Otherwise, interrupt should be
6203 		 * delivered directly to L2.
6204 		 */
6205 		if (is_guest_mode(vcpu) && max_irr_updated) {
6206 			if (nested_exit_on_intr(vcpu))
6207 				kvm_vcpu_exiting_guest_mode(vcpu);
6208 			else
6209 				kvm_make_request(KVM_REQ_EVENT, vcpu);
6210 		}
6211 	} else {
6212 		max_irr = kvm_lapic_find_highest_irr(vcpu);
6213 	}
6214 	vmx_hwapic_irr_update(vcpu, max_irr);
6215 	return max_irr;
6216 }
6217 
6218 static bool vmx_dy_apicv_has_pending_interrupt(struct kvm_vcpu *vcpu)
6219 {
6220 	struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
6221 
6222 	return pi_test_on(pi_desc) ||
6223 		(pi_test_sn(pi_desc) && !pi_is_pir_empty(pi_desc));
6224 }
6225 
6226 static void vmx_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap)
6227 {
6228 	if (!kvm_vcpu_apicv_active(vcpu))
6229 		return;
6230 
6231 	vmcs_write64(EOI_EXIT_BITMAP0, eoi_exit_bitmap[0]);
6232 	vmcs_write64(EOI_EXIT_BITMAP1, eoi_exit_bitmap[1]);
6233 	vmcs_write64(EOI_EXIT_BITMAP2, eoi_exit_bitmap[2]);
6234 	vmcs_write64(EOI_EXIT_BITMAP3, eoi_exit_bitmap[3]);
6235 }
6236 
6237 static void vmx_apicv_post_state_restore(struct kvm_vcpu *vcpu)
6238 {
6239 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6240 
6241 	pi_clear_on(&vmx->pi_desc);
6242 	memset(vmx->pi_desc.pir, 0, sizeof(vmx->pi_desc.pir));
6243 }
6244 
6245 static void handle_exception_nmi_irqoff(struct vcpu_vmx *vmx)
6246 {
6247 	vmx->exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6248 
6249 	/* if exit due to PF check for async PF */
6250 	if (is_page_fault(vmx->exit_intr_info)) {
6251 		vmx->vcpu.arch.apf.host_apf_reason = kvm_read_and_reset_pf_reason();
6252 	/* Handle machine checks before interrupts are enabled */
6253 	} else if (is_machine_check(vmx->exit_intr_info)) {
6254 		kvm_machine_check();
6255 	/* We need to handle NMIs before interrupts are enabled */
6256 	} else if (is_nmi(vmx->exit_intr_info)) {
6257 		kvm_before_interrupt(&vmx->vcpu);
6258 		asm("int $2");
6259 		kvm_after_interrupt(&vmx->vcpu);
6260 	}
6261 }
6262 
6263 static void handle_external_interrupt_irqoff(struct kvm_vcpu *vcpu)
6264 {
6265 	unsigned int vector;
6266 	unsigned long entry;
6267 #ifdef CONFIG_X86_64
6268 	unsigned long tmp;
6269 #endif
6270 	gate_desc *desc;
6271 	u32 intr_info;
6272 
6273 	intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6274 	if (WARN_ONCE(!is_external_intr(intr_info),
6275 	    "KVM: unexpected VM-Exit interrupt info: 0x%x", intr_info))
6276 		return;
6277 
6278 	vector = intr_info & INTR_INFO_VECTOR_MASK;
6279 	desc = (gate_desc *)host_idt_base + vector;
6280 	entry = gate_offset(desc);
6281 
6282 	kvm_before_interrupt(vcpu);
6283 
6284 	asm volatile(
6285 #ifdef CONFIG_X86_64
6286 		"mov %%" _ASM_SP ", %[sp]\n\t"
6287 		"and $0xfffffffffffffff0, %%" _ASM_SP "\n\t"
6288 		"push $%c[ss]\n\t"
6289 		"push %[sp]\n\t"
6290 #endif
6291 		"pushf\n\t"
6292 		__ASM_SIZE(push) " $%c[cs]\n\t"
6293 		CALL_NOSPEC
6294 		:
6295 #ifdef CONFIG_X86_64
6296 		[sp]"=&r"(tmp),
6297 #endif
6298 		ASM_CALL_CONSTRAINT
6299 		:
6300 		[thunk_target]"r"(entry),
6301 		[ss]"i"(__KERNEL_DS),
6302 		[cs]"i"(__KERNEL_CS)
6303 	);
6304 
6305 	kvm_after_interrupt(vcpu);
6306 }
6307 STACK_FRAME_NON_STANDARD(handle_external_interrupt_irqoff);
6308 
6309 static void vmx_handle_exit_irqoff(struct kvm_vcpu *vcpu,
6310 	enum exit_fastpath_completion *exit_fastpath)
6311 {
6312 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6313 
6314 	if (vmx->exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT)
6315 		handle_external_interrupt_irqoff(vcpu);
6316 	else if (vmx->exit_reason == EXIT_REASON_EXCEPTION_NMI)
6317 		handle_exception_nmi_irqoff(vmx);
6318 	else if (!is_guest_mode(vcpu) &&
6319 		vmx->exit_reason == EXIT_REASON_MSR_WRITE)
6320 		*exit_fastpath = handle_fastpath_set_msr_irqoff(vcpu);
6321 }
6322 
6323 static bool vmx_has_emulated_msr(int index)
6324 {
6325 	switch (index) {
6326 	case MSR_IA32_SMBASE:
6327 		/*
6328 		 * We cannot do SMM unless we can run the guest in big
6329 		 * real mode.
6330 		 */
6331 		return enable_unrestricted_guest || emulate_invalid_guest_state;
6332 	case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
6333 		return nested;
6334 	case MSR_AMD64_VIRT_SPEC_CTRL:
6335 		/* This is AMD only.  */
6336 		return false;
6337 	default:
6338 		return true;
6339 	}
6340 }
6341 
6342 static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx)
6343 {
6344 	u32 exit_intr_info;
6345 	bool unblock_nmi;
6346 	u8 vector;
6347 	bool idtv_info_valid;
6348 
6349 	idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK;
6350 
6351 	if (enable_vnmi) {
6352 		if (vmx->loaded_vmcs->nmi_known_unmasked)
6353 			return;
6354 		/*
6355 		 * Can't use vmx->exit_intr_info since we're not sure what
6356 		 * the exit reason is.
6357 		 */
6358 		exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6359 		unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
6360 		vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
6361 		/*
6362 		 * SDM 3: 27.7.1.2 (September 2008)
6363 		 * Re-set bit "block by NMI" before VM entry if vmexit caused by
6364 		 * a guest IRET fault.
6365 		 * SDM 3: 23.2.2 (September 2008)
6366 		 * Bit 12 is undefined in any of the following cases:
6367 		 *  If the VM exit sets the valid bit in the IDT-vectoring
6368 		 *   information field.
6369 		 *  If the VM exit is due to a double fault.
6370 		 */
6371 		if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
6372 		    vector != DF_VECTOR && !idtv_info_valid)
6373 			vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
6374 				      GUEST_INTR_STATE_NMI);
6375 		else
6376 			vmx->loaded_vmcs->nmi_known_unmasked =
6377 				!(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO)
6378 				  & GUEST_INTR_STATE_NMI);
6379 	} else if (unlikely(vmx->loaded_vmcs->soft_vnmi_blocked))
6380 		vmx->loaded_vmcs->vnmi_blocked_time +=
6381 			ktime_to_ns(ktime_sub(ktime_get(),
6382 					      vmx->loaded_vmcs->entry_time));
6383 }
6384 
6385 static void __vmx_complete_interrupts(struct kvm_vcpu *vcpu,
6386 				      u32 idt_vectoring_info,
6387 				      int instr_len_field,
6388 				      int error_code_field)
6389 {
6390 	u8 vector;
6391 	int type;
6392 	bool idtv_info_valid;
6393 
6394 	idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
6395 
6396 	vcpu->arch.nmi_injected = false;
6397 	kvm_clear_exception_queue(vcpu);
6398 	kvm_clear_interrupt_queue(vcpu);
6399 
6400 	if (!idtv_info_valid)
6401 		return;
6402 
6403 	kvm_make_request(KVM_REQ_EVENT, vcpu);
6404 
6405 	vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
6406 	type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
6407 
6408 	switch (type) {
6409 	case INTR_TYPE_NMI_INTR:
6410 		vcpu->arch.nmi_injected = true;
6411 		/*
6412 		 * SDM 3: 27.7.1.2 (September 2008)
6413 		 * Clear bit "block by NMI" before VM entry if a NMI
6414 		 * delivery faulted.
6415 		 */
6416 		vmx_set_nmi_mask(vcpu, false);
6417 		break;
6418 	case INTR_TYPE_SOFT_EXCEPTION:
6419 		vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
6420 		/* fall through */
6421 	case INTR_TYPE_HARD_EXCEPTION:
6422 		if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
6423 			u32 err = vmcs_read32(error_code_field);
6424 			kvm_requeue_exception_e(vcpu, vector, err);
6425 		} else
6426 			kvm_requeue_exception(vcpu, vector);
6427 		break;
6428 	case INTR_TYPE_SOFT_INTR:
6429 		vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
6430 		/* fall through */
6431 	case INTR_TYPE_EXT_INTR:
6432 		kvm_queue_interrupt(vcpu, vector, type == INTR_TYPE_SOFT_INTR);
6433 		break;
6434 	default:
6435 		break;
6436 	}
6437 }
6438 
6439 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
6440 {
6441 	__vmx_complete_interrupts(&vmx->vcpu, vmx->idt_vectoring_info,
6442 				  VM_EXIT_INSTRUCTION_LEN,
6443 				  IDT_VECTORING_ERROR_CODE);
6444 }
6445 
6446 static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
6447 {
6448 	__vmx_complete_interrupts(vcpu,
6449 				  vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
6450 				  VM_ENTRY_INSTRUCTION_LEN,
6451 				  VM_ENTRY_EXCEPTION_ERROR_CODE);
6452 
6453 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
6454 }
6455 
6456 static void atomic_switch_perf_msrs(struct vcpu_vmx *vmx)
6457 {
6458 	int i, nr_msrs;
6459 	struct perf_guest_switch_msr *msrs;
6460 
6461 	msrs = perf_guest_get_msrs(&nr_msrs);
6462 
6463 	if (!msrs)
6464 		return;
6465 
6466 	for (i = 0; i < nr_msrs; i++)
6467 		if (msrs[i].host == msrs[i].guest)
6468 			clear_atomic_switch_msr(vmx, msrs[i].msr);
6469 		else
6470 			add_atomic_switch_msr(vmx, msrs[i].msr, msrs[i].guest,
6471 					msrs[i].host, false);
6472 }
6473 
6474 static void atomic_switch_umwait_control_msr(struct vcpu_vmx *vmx)
6475 {
6476 	u32 host_umwait_control;
6477 
6478 	if (!vmx_has_waitpkg(vmx))
6479 		return;
6480 
6481 	host_umwait_control = get_umwait_control_msr();
6482 
6483 	if (vmx->msr_ia32_umwait_control != host_umwait_control)
6484 		add_atomic_switch_msr(vmx, MSR_IA32_UMWAIT_CONTROL,
6485 			vmx->msr_ia32_umwait_control,
6486 			host_umwait_control, false);
6487 	else
6488 		clear_atomic_switch_msr(vmx, MSR_IA32_UMWAIT_CONTROL);
6489 }
6490 
6491 static void vmx_update_hv_timer(struct kvm_vcpu *vcpu)
6492 {
6493 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6494 	u64 tscl;
6495 	u32 delta_tsc;
6496 
6497 	if (vmx->req_immediate_exit) {
6498 		vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, 0);
6499 		vmx->loaded_vmcs->hv_timer_soft_disabled = false;
6500 	} else if (vmx->hv_deadline_tsc != -1) {
6501 		tscl = rdtsc();
6502 		if (vmx->hv_deadline_tsc > tscl)
6503 			/* set_hv_timer ensures the delta fits in 32-bits */
6504 			delta_tsc = (u32)((vmx->hv_deadline_tsc - tscl) >>
6505 				cpu_preemption_timer_multi);
6506 		else
6507 			delta_tsc = 0;
6508 
6509 		vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, delta_tsc);
6510 		vmx->loaded_vmcs->hv_timer_soft_disabled = false;
6511 	} else if (!vmx->loaded_vmcs->hv_timer_soft_disabled) {
6512 		vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, -1);
6513 		vmx->loaded_vmcs->hv_timer_soft_disabled = true;
6514 	}
6515 }
6516 
6517 void vmx_update_host_rsp(struct vcpu_vmx *vmx, unsigned long host_rsp)
6518 {
6519 	if (unlikely(host_rsp != vmx->loaded_vmcs->host_state.rsp)) {
6520 		vmx->loaded_vmcs->host_state.rsp = host_rsp;
6521 		vmcs_writel(HOST_RSP, host_rsp);
6522 	}
6523 }
6524 
6525 bool __vmx_vcpu_run(struct vcpu_vmx *vmx, unsigned long *regs, bool launched);
6526 
6527 static void vmx_vcpu_run(struct kvm_vcpu *vcpu)
6528 {
6529 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6530 	unsigned long cr3, cr4;
6531 
6532 	/* Record the guest's net vcpu time for enforced NMI injections. */
6533 	if (unlikely(!enable_vnmi &&
6534 		     vmx->loaded_vmcs->soft_vnmi_blocked))
6535 		vmx->loaded_vmcs->entry_time = ktime_get();
6536 
6537 	/* Don't enter VMX if guest state is invalid, let the exit handler
6538 	   start emulation until we arrive back to a valid state */
6539 	if (vmx->emulation_required)
6540 		return;
6541 
6542 	if (vmx->ple_window_dirty) {
6543 		vmx->ple_window_dirty = false;
6544 		vmcs_write32(PLE_WINDOW, vmx->ple_window);
6545 	}
6546 
6547 	/*
6548 	 * We did this in prepare_switch_to_guest, because it needs to
6549 	 * be within srcu_read_lock.
6550 	 */
6551 	WARN_ON_ONCE(vmx->nested.need_vmcs12_to_shadow_sync);
6552 
6553 	if (kvm_register_is_dirty(vcpu, VCPU_REGS_RSP))
6554 		vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
6555 	if (kvm_register_is_dirty(vcpu, VCPU_REGS_RIP))
6556 		vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
6557 
6558 	cr3 = __get_current_cr3_fast();
6559 	if (unlikely(cr3 != vmx->loaded_vmcs->host_state.cr3)) {
6560 		vmcs_writel(HOST_CR3, cr3);
6561 		vmx->loaded_vmcs->host_state.cr3 = cr3;
6562 	}
6563 
6564 	cr4 = cr4_read_shadow();
6565 	if (unlikely(cr4 != vmx->loaded_vmcs->host_state.cr4)) {
6566 		vmcs_writel(HOST_CR4, cr4);
6567 		vmx->loaded_vmcs->host_state.cr4 = cr4;
6568 	}
6569 
6570 	/* When single-stepping over STI and MOV SS, we must clear the
6571 	 * corresponding interruptibility bits in the guest state. Otherwise
6572 	 * vmentry fails as it then expects bit 14 (BS) in pending debug
6573 	 * exceptions being set, but that's not correct for the guest debugging
6574 	 * case. */
6575 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6576 		vmx_set_interrupt_shadow(vcpu, 0);
6577 
6578 	kvm_load_guest_xsave_state(vcpu);
6579 
6580 	if (static_cpu_has(X86_FEATURE_PKU) &&
6581 	    kvm_read_cr4_bits(vcpu, X86_CR4_PKE) &&
6582 	    vcpu->arch.pkru != vmx->host_pkru)
6583 		__write_pkru(vcpu->arch.pkru);
6584 
6585 	pt_guest_enter(vmx);
6586 
6587 	if (vcpu_to_pmu(vcpu)->version)
6588 		atomic_switch_perf_msrs(vmx);
6589 	atomic_switch_umwait_control_msr(vmx);
6590 
6591 	if (enable_preemption_timer)
6592 		vmx_update_hv_timer(vcpu);
6593 
6594 	if (lapic_in_kernel(vcpu) &&
6595 		vcpu->arch.apic->lapic_timer.timer_advance_ns)
6596 		kvm_wait_lapic_expire(vcpu);
6597 
6598 	/*
6599 	 * If this vCPU has touched SPEC_CTRL, restore the guest's value if
6600 	 * it's non-zero. Since vmentry is serialising on affected CPUs, there
6601 	 * is no need to worry about the conditional branch over the wrmsr
6602 	 * being speculatively taken.
6603 	 */
6604 	x86_spec_ctrl_set_guest(vmx->spec_ctrl, 0);
6605 
6606 	/* L1D Flush includes CPU buffer clear to mitigate MDS */
6607 	if (static_branch_unlikely(&vmx_l1d_should_flush))
6608 		vmx_l1d_flush(vcpu);
6609 	else if (static_branch_unlikely(&mds_user_clear))
6610 		mds_clear_cpu_buffers();
6611 
6612 	if (vcpu->arch.cr2 != read_cr2())
6613 		write_cr2(vcpu->arch.cr2);
6614 
6615 	vmx->fail = __vmx_vcpu_run(vmx, (unsigned long *)&vcpu->arch.regs,
6616 				   vmx->loaded_vmcs->launched);
6617 
6618 	vcpu->arch.cr2 = read_cr2();
6619 
6620 	/*
6621 	 * We do not use IBRS in the kernel. If this vCPU has used the
6622 	 * SPEC_CTRL MSR it may have left it on; save the value and
6623 	 * turn it off. This is much more efficient than blindly adding
6624 	 * it to the atomic save/restore list. Especially as the former
6625 	 * (Saving guest MSRs on vmexit) doesn't even exist in KVM.
6626 	 *
6627 	 * For non-nested case:
6628 	 * If the L01 MSR bitmap does not intercept the MSR, then we need to
6629 	 * save it.
6630 	 *
6631 	 * For nested case:
6632 	 * If the L02 MSR bitmap does not intercept the MSR, then we need to
6633 	 * save it.
6634 	 */
6635 	if (unlikely(!msr_write_intercepted(vcpu, MSR_IA32_SPEC_CTRL)))
6636 		vmx->spec_ctrl = native_read_msr(MSR_IA32_SPEC_CTRL);
6637 
6638 	x86_spec_ctrl_restore_host(vmx->spec_ctrl, 0);
6639 
6640 	/* All fields are clean at this point */
6641 	if (static_branch_unlikely(&enable_evmcs))
6642 		current_evmcs->hv_clean_fields |=
6643 			HV_VMX_ENLIGHTENED_CLEAN_FIELD_ALL;
6644 
6645 	if (static_branch_unlikely(&enable_evmcs))
6646 		current_evmcs->hv_vp_id = vcpu->arch.hyperv.vp_index;
6647 
6648 	/* MSR_IA32_DEBUGCTLMSR is zeroed on vmexit. Restore it if needed */
6649 	if (vmx->host_debugctlmsr)
6650 		update_debugctlmsr(vmx->host_debugctlmsr);
6651 
6652 #ifndef CONFIG_X86_64
6653 	/*
6654 	 * The sysexit path does not restore ds/es, so we must set them to
6655 	 * a reasonable value ourselves.
6656 	 *
6657 	 * We can't defer this to vmx_prepare_switch_to_host() since that
6658 	 * function may be executed in interrupt context, which saves and
6659 	 * restore segments around it, nullifying its effect.
6660 	 */
6661 	loadsegment(ds, __USER_DS);
6662 	loadsegment(es, __USER_DS);
6663 #endif
6664 
6665 	vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
6666 				  | (1 << VCPU_EXREG_RFLAGS)
6667 				  | (1 << VCPU_EXREG_PDPTR)
6668 				  | (1 << VCPU_EXREG_SEGMENTS)
6669 				  | (1 << VCPU_EXREG_CR3));
6670 	vcpu->arch.regs_dirty = 0;
6671 
6672 	pt_guest_exit(vmx);
6673 
6674 	/*
6675 	 * eager fpu is enabled if PKEY is supported and CR4 is switched
6676 	 * back on host, so it is safe to read guest PKRU from current
6677 	 * XSAVE.
6678 	 */
6679 	if (static_cpu_has(X86_FEATURE_PKU) &&
6680 	    kvm_read_cr4_bits(vcpu, X86_CR4_PKE)) {
6681 		vcpu->arch.pkru = rdpkru();
6682 		if (vcpu->arch.pkru != vmx->host_pkru)
6683 			__write_pkru(vmx->host_pkru);
6684 	}
6685 
6686 	kvm_load_host_xsave_state(vcpu);
6687 
6688 	vmx->nested.nested_run_pending = 0;
6689 	vmx->idt_vectoring_info = 0;
6690 
6691 	vmx->exit_reason = vmx->fail ? 0xdead : vmcs_read32(VM_EXIT_REASON);
6692 	if ((u16)vmx->exit_reason == EXIT_REASON_MCE_DURING_VMENTRY)
6693 		kvm_machine_check();
6694 
6695 	if (vmx->fail || (vmx->exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY))
6696 		return;
6697 
6698 	vmx->loaded_vmcs->launched = 1;
6699 	vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
6700 
6701 	vmx_recover_nmi_blocking(vmx);
6702 	vmx_complete_interrupts(vmx);
6703 }
6704 
6705 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
6706 {
6707 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6708 
6709 	if (enable_pml)
6710 		vmx_destroy_pml_buffer(vmx);
6711 	free_vpid(vmx->vpid);
6712 	nested_vmx_free_vcpu(vcpu);
6713 	free_loaded_vmcs(vmx->loaded_vmcs);
6714 }
6715 
6716 static int vmx_create_vcpu(struct kvm_vcpu *vcpu)
6717 {
6718 	struct vcpu_vmx *vmx;
6719 	unsigned long *msr_bitmap;
6720 	int i, cpu, err;
6721 
6722 	BUILD_BUG_ON(offsetof(struct vcpu_vmx, vcpu) != 0);
6723 	vmx = to_vmx(vcpu);
6724 
6725 	err = -ENOMEM;
6726 
6727 	vmx->vpid = allocate_vpid();
6728 
6729 	/*
6730 	 * If PML is turned on, failure on enabling PML just results in failure
6731 	 * of creating the vcpu, therefore we can simplify PML logic (by
6732 	 * avoiding dealing with cases, such as enabling PML partially on vcpus
6733 	 * for the guest), etc.
6734 	 */
6735 	if (enable_pml) {
6736 		vmx->pml_pg = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
6737 		if (!vmx->pml_pg)
6738 			goto free_vpid;
6739 	}
6740 
6741 	BUILD_BUG_ON(ARRAY_SIZE(vmx_msr_index) != NR_SHARED_MSRS);
6742 
6743 	for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i) {
6744 		u32 index = vmx_msr_index[i];
6745 		u32 data_low, data_high;
6746 		int j = vmx->nmsrs;
6747 
6748 		if (rdmsr_safe(index, &data_low, &data_high) < 0)
6749 			continue;
6750 		if (wrmsr_safe(index, data_low, data_high) < 0)
6751 			continue;
6752 
6753 		vmx->guest_msrs[j].index = i;
6754 		vmx->guest_msrs[j].data = 0;
6755 		switch (index) {
6756 		case MSR_IA32_TSX_CTRL:
6757 			/*
6758 			 * No need to pass TSX_CTRL_CPUID_CLEAR through, so
6759 			 * let's avoid changing CPUID bits under the host
6760 			 * kernel's feet.
6761 			 */
6762 			vmx->guest_msrs[j].mask = ~(u64)TSX_CTRL_CPUID_CLEAR;
6763 			break;
6764 		default:
6765 			vmx->guest_msrs[j].mask = -1ull;
6766 			break;
6767 		}
6768 		++vmx->nmsrs;
6769 	}
6770 
6771 	err = alloc_loaded_vmcs(&vmx->vmcs01);
6772 	if (err < 0)
6773 		goto free_pml;
6774 
6775 	msr_bitmap = vmx->vmcs01.msr_bitmap;
6776 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_TSC, MSR_TYPE_R);
6777 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_FS_BASE, MSR_TYPE_RW);
6778 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_GS_BASE, MSR_TYPE_RW);
6779 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_KERNEL_GS_BASE, MSR_TYPE_RW);
6780 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_CS, MSR_TYPE_RW);
6781 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_ESP, MSR_TYPE_RW);
6782 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_EIP, MSR_TYPE_RW);
6783 	if (kvm_cstate_in_guest(vcpu->kvm)) {
6784 		vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C1_RES, MSR_TYPE_R);
6785 		vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C3_RESIDENCY, MSR_TYPE_R);
6786 		vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C6_RESIDENCY, MSR_TYPE_R);
6787 		vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C7_RESIDENCY, MSR_TYPE_R);
6788 	}
6789 	vmx->msr_bitmap_mode = 0;
6790 
6791 	vmx->loaded_vmcs = &vmx->vmcs01;
6792 	cpu = get_cpu();
6793 	vmx_vcpu_load(vcpu, cpu);
6794 	vcpu->cpu = cpu;
6795 	init_vmcs(vmx);
6796 	vmx_vcpu_put(vcpu);
6797 	put_cpu();
6798 	if (cpu_need_virtualize_apic_accesses(vcpu)) {
6799 		err = alloc_apic_access_page(vcpu->kvm);
6800 		if (err)
6801 			goto free_vmcs;
6802 	}
6803 
6804 	if (enable_ept && !enable_unrestricted_guest) {
6805 		err = init_rmode_identity_map(vcpu->kvm);
6806 		if (err)
6807 			goto free_vmcs;
6808 	}
6809 
6810 	if (nested)
6811 		nested_vmx_setup_ctls_msrs(&vmx->nested.msrs,
6812 					   vmx_capability.ept);
6813 	else
6814 		memset(&vmx->nested.msrs, 0, sizeof(vmx->nested.msrs));
6815 
6816 	vmx->nested.posted_intr_nv = -1;
6817 	vmx->nested.current_vmptr = -1ull;
6818 
6819 	vcpu->arch.microcode_version = 0x100000000ULL;
6820 	vmx->msr_ia32_feature_control_valid_bits = FEAT_CTL_LOCKED;
6821 
6822 	/*
6823 	 * Enforce invariant: pi_desc.nv is always either POSTED_INTR_VECTOR
6824 	 * or POSTED_INTR_WAKEUP_VECTOR.
6825 	 */
6826 	vmx->pi_desc.nv = POSTED_INTR_VECTOR;
6827 	vmx->pi_desc.sn = 1;
6828 
6829 	vmx->ept_pointer = INVALID_PAGE;
6830 
6831 	return 0;
6832 
6833 free_vmcs:
6834 	free_loaded_vmcs(vmx->loaded_vmcs);
6835 free_pml:
6836 	vmx_destroy_pml_buffer(vmx);
6837 free_vpid:
6838 	free_vpid(vmx->vpid);
6839 	return err;
6840 }
6841 
6842 #define L1TF_MSG_SMT "L1TF CPU bug present and SMT on, data leak possible. See CVE-2018-3646 and https://www.kernel.org/doc/html/latest/admin-guide/hw-vuln/l1tf.html for details.\n"
6843 #define L1TF_MSG_L1D "L1TF CPU bug present and virtualization mitigation disabled, data leak possible. See CVE-2018-3646 and https://www.kernel.org/doc/html/latest/admin-guide/hw-vuln/l1tf.html for details.\n"
6844 
6845 static int vmx_vm_init(struct kvm *kvm)
6846 {
6847 	spin_lock_init(&to_kvm_vmx(kvm)->ept_pointer_lock);
6848 
6849 	if (!ple_gap)
6850 		kvm->arch.pause_in_guest = true;
6851 
6852 	if (boot_cpu_has(X86_BUG_L1TF) && enable_ept) {
6853 		switch (l1tf_mitigation) {
6854 		case L1TF_MITIGATION_OFF:
6855 		case L1TF_MITIGATION_FLUSH_NOWARN:
6856 			/* 'I explicitly don't care' is set */
6857 			break;
6858 		case L1TF_MITIGATION_FLUSH:
6859 		case L1TF_MITIGATION_FLUSH_NOSMT:
6860 		case L1TF_MITIGATION_FULL:
6861 			/*
6862 			 * Warn upon starting the first VM in a potentially
6863 			 * insecure environment.
6864 			 */
6865 			if (sched_smt_active())
6866 				pr_warn_once(L1TF_MSG_SMT);
6867 			if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_NEVER)
6868 				pr_warn_once(L1TF_MSG_L1D);
6869 			break;
6870 		case L1TF_MITIGATION_FULL_FORCE:
6871 			/* Flush is enforced */
6872 			break;
6873 		}
6874 	}
6875 	kvm_apicv_init(kvm, enable_apicv);
6876 	return 0;
6877 }
6878 
6879 static int __init vmx_check_processor_compat(void)
6880 {
6881 	struct vmcs_config vmcs_conf;
6882 	struct vmx_capability vmx_cap;
6883 
6884 	if (!this_cpu_has(X86_FEATURE_MSR_IA32_FEAT_CTL) ||
6885 	    !this_cpu_has(X86_FEATURE_VMX)) {
6886 		pr_err("kvm: VMX is disabled on CPU %d\n", smp_processor_id());
6887 		return -EIO;
6888 	}
6889 
6890 	if (setup_vmcs_config(&vmcs_conf, &vmx_cap) < 0)
6891 		return -EIO;
6892 	if (nested)
6893 		nested_vmx_setup_ctls_msrs(&vmcs_conf.nested, vmx_cap.ept);
6894 	if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
6895 		printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
6896 				smp_processor_id());
6897 		return -EIO;
6898 	}
6899 	return 0;
6900 }
6901 
6902 static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
6903 {
6904 	u8 cache;
6905 	u64 ipat = 0;
6906 
6907 	/* We wanted to honor guest CD/MTRR/PAT, but doing so could result in
6908 	 * memory aliases with conflicting memory types and sometimes MCEs.
6909 	 * We have to be careful as to what are honored and when.
6910 	 *
6911 	 * For MMIO, guest CD/MTRR are ignored.  The EPT memory type is set to
6912 	 * UC.  The effective memory type is UC or WC depending on guest PAT.
6913 	 * This was historically the source of MCEs and we want to be
6914 	 * conservative.
6915 	 *
6916 	 * When there is no need to deal with noncoherent DMA (e.g., no VT-d
6917 	 * or VT-d has snoop control), guest CD/MTRR/PAT are all ignored.  The
6918 	 * EPT memory type is set to WB.  The effective memory type is forced
6919 	 * WB.
6920 	 *
6921 	 * Otherwise, we trust guest.  Guest CD/MTRR/PAT are all honored.  The
6922 	 * EPT memory type is used to emulate guest CD/MTRR.
6923 	 */
6924 
6925 	if (is_mmio) {
6926 		cache = MTRR_TYPE_UNCACHABLE;
6927 		goto exit;
6928 	}
6929 
6930 	if (!kvm_arch_has_noncoherent_dma(vcpu->kvm)) {
6931 		ipat = VMX_EPT_IPAT_BIT;
6932 		cache = MTRR_TYPE_WRBACK;
6933 		goto exit;
6934 	}
6935 
6936 	if (kvm_read_cr0(vcpu) & X86_CR0_CD) {
6937 		ipat = VMX_EPT_IPAT_BIT;
6938 		if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
6939 			cache = MTRR_TYPE_WRBACK;
6940 		else
6941 			cache = MTRR_TYPE_UNCACHABLE;
6942 		goto exit;
6943 	}
6944 
6945 	cache = kvm_mtrr_get_guest_memory_type(vcpu, gfn);
6946 
6947 exit:
6948 	return (cache << VMX_EPT_MT_EPTE_SHIFT) | ipat;
6949 }
6950 
6951 static void vmcs_set_secondary_exec_control(struct vcpu_vmx *vmx)
6952 {
6953 	/*
6954 	 * These bits in the secondary execution controls field
6955 	 * are dynamic, the others are mostly based on the hypervisor
6956 	 * architecture and the guest's CPUID.  Do not touch the
6957 	 * dynamic bits.
6958 	 */
6959 	u32 mask =
6960 		SECONDARY_EXEC_SHADOW_VMCS |
6961 		SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
6962 		SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
6963 		SECONDARY_EXEC_DESC;
6964 
6965 	u32 new_ctl = vmx->secondary_exec_control;
6966 	u32 cur_ctl = secondary_exec_controls_get(vmx);
6967 
6968 	secondary_exec_controls_set(vmx, (new_ctl & ~mask) | (cur_ctl & mask));
6969 }
6970 
6971 /*
6972  * Generate MSR_IA32_VMX_CR{0,4}_FIXED1 according to CPUID. Only set bits
6973  * (indicating "allowed-1") if they are supported in the guest's CPUID.
6974  */
6975 static void nested_vmx_cr_fixed1_bits_update(struct kvm_vcpu *vcpu)
6976 {
6977 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6978 	struct kvm_cpuid_entry2 *entry;
6979 
6980 	vmx->nested.msrs.cr0_fixed1 = 0xffffffff;
6981 	vmx->nested.msrs.cr4_fixed1 = X86_CR4_PCE;
6982 
6983 #define cr4_fixed1_update(_cr4_mask, _reg, _cpuid_mask) do {		\
6984 	if (entry && (entry->_reg & (_cpuid_mask)))			\
6985 		vmx->nested.msrs.cr4_fixed1 |= (_cr4_mask);	\
6986 } while (0)
6987 
6988 	entry = kvm_find_cpuid_entry(vcpu, 0x1, 0);
6989 	cr4_fixed1_update(X86_CR4_VME,        edx, feature_bit(VME));
6990 	cr4_fixed1_update(X86_CR4_PVI,        edx, feature_bit(VME));
6991 	cr4_fixed1_update(X86_CR4_TSD,        edx, feature_bit(TSC));
6992 	cr4_fixed1_update(X86_CR4_DE,         edx, feature_bit(DE));
6993 	cr4_fixed1_update(X86_CR4_PSE,        edx, feature_bit(PSE));
6994 	cr4_fixed1_update(X86_CR4_PAE,        edx, feature_bit(PAE));
6995 	cr4_fixed1_update(X86_CR4_MCE,        edx, feature_bit(MCE));
6996 	cr4_fixed1_update(X86_CR4_PGE,        edx, feature_bit(PGE));
6997 	cr4_fixed1_update(X86_CR4_OSFXSR,     edx, feature_bit(FXSR));
6998 	cr4_fixed1_update(X86_CR4_OSXMMEXCPT, edx, feature_bit(XMM));
6999 	cr4_fixed1_update(X86_CR4_VMXE,       ecx, feature_bit(VMX));
7000 	cr4_fixed1_update(X86_CR4_SMXE,       ecx, feature_bit(SMX));
7001 	cr4_fixed1_update(X86_CR4_PCIDE,      ecx, feature_bit(PCID));
7002 	cr4_fixed1_update(X86_CR4_OSXSAVE,    ecx, feature_bit(XSAVE));
7003 
7004 	entry = kvm_find_cpuid_entry(vcpu, 0x7, 0);
7005 	cr4_fixed1_update(X86_CR4_FSGSBASE,   ebx, feature_bit(FSGSBASE));
7006 	cr4_fixed1_update(X86_CR4_SMEP,       ebx, feature_bit(SMEP));
7007 	cr4_fixed1_update(X86_CR4_SMAP,       ebx, feature_bit(SMAP));
7008 	cr4_fixed1_update(X86_CR4_PKE,        ecx, feature_bit(PKU));
7009 	cr4_fixed1_update(X86_CR4_UMIP,       ecx, feature_bit(UMIP));
7010 	cr4_fixed1_update(X86_CR4_LA57,       ecx, feature_bit(LA57));
7011 
7012 #undef cr4_fixed1_update
7013 }
7014 
7015 static void nested_vmx_entry_exit_ctls_update(struct kvm_vcpu *vcpu)
7016 {
7017 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7018 
7019 	if (kvm_mpx_supported()) {
7020 		bool mpx_enabled = guest_cpuid_has(vcpu, X86_FEATURE_MPX);
7021 
7022 		if (mpx_enabled) {
7023 			vmx->nested.msrs.entry_ctls_high |= VM_ENTRY_LOAD_BNDCFGS;
7024 			vmx->nested.msrs.exit_ctls_high |= VM_EXIT_CLEAR_BNDCFGS;
7025 		} else {
7026 			vmx->nested.msrs.entry_ctls_high &= ~VM_ENTRY_LOAD_BNDCFGS;
7027 			vmx->nested.msrs.exit_ctls_high &= ~VM_EXIT_CLEAR_BNDCFGS;
7028 		}
7029 	}
7030 }
7031 
7032 static void update_intel_pt_cfg(struct kvm_vcpu *vcpu)
7033 {
7034 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7035 	struct kvm_cpuid_entry2 *best = NULL;
7036 	int i;
7037 
7038 	for (i = 0; i < PT_CPUID_LEAVES; i++) {
7039 		best = kvm_find_cpuid_entry(vcpu, 0x14, i);
7040 		if (!best)
7041 			return;
7042 		vmx->pt_desc.caps[CPUID_EAX + i*PT_CPUID_REGS_NUM] = best->eax;
7043 		vmx->pt_desc.caps[CPUID_EBX + i*PT_CPUID_REGS_NUM] = best->ebx;
7044 		vmx->pt_desc.caps[CPUID_ECX + i*PT_CPUID_REGS_NUM] = best->ecx;
7045 		vmx->pt_desc.caps[CPUID_EDX + i*PT_CPUID_REGS_NUM] = best->edx;
7046 	}
7047 
7048 	/* Get the number of configurable Address Ranges for filtering */
7049 	vmx->pt_desc.addr_range = intel_pt_validate_cap(vmx->pt_desc.caps,
7050 						PT_CAP_num_address_ranges);
7051 
7052 	/* Initialize and clear the no dependency bits */
7053 	vmx->pt_desc.ctl_bitmask = ~(RTIT_CTL_TRACEEN | RTIT_CTL_OS |
7054 			RTIT_CTL_USR | RTIT_CTL_TSC_EN | RTIT_CTL_DISRETC);
7055 
7056 	/*
7057 	 * If CPUID.(EAX=14H,ECX=0):EBX[0]=1 CR3Filter can be set otherwise
7058 	 * will inject an #GP
7059 	 */
7060 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_cr3_filtering))
7061 		vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_CR3EN;
7062 
7063 	/*
7064 	 * If CPUID.(EAX=14H,ECX=0):EBX[1]=1 CYCEn, CycThresh and
7065 	 * PSBFreq can be set
7066 	 */
7067 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc))
7068 		vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_CYCLEACC |
7069 				RTIT_CTL_CYC_THRESH | RTIT_CTL_PSB_FREQ);
7070 
7071 	/*
7072 	 * If CPUID.(EAX=14H,ECX=0):EBX[3]=1 MTCEn BranchEn and
7073 	 * MTCFreq can be set
7074 	 */
7075 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc))
7076 		vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_MTC_EN |
7077 				RTIT_CTL_BRANCH_EN | RTIT_CTL_MTC_RANGE);
7078 
7079 	/* If CPUID.(EAX=14H,ECX=0):EBX[4]=1 FUPonPTW and PTWEn can be set */
7080 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_ptwrite))
7081 		vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_FUP_ON_PTW |
7082 							RTIT_CTL_PTW_EN);
7083 
7084 	/* If CPUID.(EAX=14H,ECX=0):EBX[5]=1 PwrEvEn can be set */
7085 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_power_event_trace))
7086 		vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_PWR_EVT_EN;
7087 
7088 	/* If CPUID.(EAX=14H,ECX=0):ECX[0]=1 ToPA can be set */
7089 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_topa_output))
7090 		vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_TOPA;
7091 
7092 	/* If CPUID.(EAX=14H,ECX=0):ECX[3]=1 FabircEn can be set */
7093 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_output_subsys))
7094 		vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_FABRIC_EN;
7095 
7096 	/* unmask address range configure area */
7097 	for (i = 0; i < vmx->pt_desc.addr_range; i++)
7098 		vmx->pt_desc.ctl_bitmask &= ~(0xfULL << (32 + i * 4));
7099 }
7100 
7101 static void vmx_cpuid_update(struct kvm_vcpu *vcpu)
7102 {
7103 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7104 
7105 	/* xsaves_enabled is recomputed in vmx_compute_secondary_exec_control(). */
7106 	vcpu->arch.xsaves_enabled = false;
7107 
7108 	if (cpu_has_secondary_exec_ctrls()) {
7109 		vmx_compute_secondary_exec_control(vmx);
7110 		vmcs_set_secondary_exec_control(vmx);
7111 	}
7112 
7113 	if (nested_vmx_allowed(vcpu))
7114 		to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
7115 			FEAT_CTL_VMX_ENABLED_INSIDE_SMX |
7116 			FEAT_CTL_VMX_ENABLED_OUTSIDE_SMX;
7117 	else
7118 		to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
7119 			~(FEAT_CTL_VMX_ENABLED_INSIDE_SMX |
7120 			  FEAT_CTL_VMX_ENABLED_OUTSIDE_SMX);
7121 
7122 	if (nested_vmx_allowed(vcpu)) {
7123 		nested_vmx_cr_fixed1_bits_update(vcpu);
7124 		nested_vmx_entry_exit_ctls_update(vcpu);
7125 	}
7126 
7127 	if (boot_cpu_has(X86_FEATURE_INTEL_PT) &&
7128 			guest_cpuid_has(vcpu, X86_FEATURE_INTEL_PT))
7129 		update_intel_pt_cfg(vcpu);
7130 
7131 	if (boot_cpu_has(X86_FEATURE_RTM)) {
7132 		struct shared_msr_entry *msr;
7133 		msr = find_msr_entry(vmx, MSR_IA32_TSX_CTRL);
7134 		if (msr) {
7135 			bool enabled = guest_cpuid_has(vcpu, X86_FEATURE_RTM);
7136 			vmx_set_guest_msr(vmx, msr, enabled ? 0 : TSX_CTRL_RTM_DISABLE);
7137 		}
7138 	}
7139 }
7140 
7141 static __init void vmx_set_cpu_caps(void)
7142 {
7143 	kvm_set_cpu_caps();
7144 
7145 	/* CPUID 0x1 */
7146 	if (nested)
7147 		kvm_cpu_cap_set(X86_FEATURE_VMX);
7148 
7149 	/* CPUID 0x7 */
7150 	if (kvm_mpx_supported())
7151 		kvm_cpu_cap_check_and_set(X86_FEATURE_MPX);
7152 	if (cpu_has_vmx_invpcid())
7153 		kvm_cpu_cap_check_and_set(X86_FEATURE_INVPCID);
7154 	if (vmx_pt_mode_is_host_guest())
7155 		kvm_cpu_cap_check_and_set(X86_FEATURE_INTEL_PT);
7156 
7157 	/* PKU is not yet implemented for shadow paging. */
7158 	if (enable_ept && boot_cpu_has(X86_FEATURE_OSPKE))
7159 		kvm_cpu_cap_check_and_set(X86_FEATURE_PKU);
7160 
7161 	if (vmx_umip_emulated())
7162 		kvm_cpu_cap_set(X86_FEATURE_UMIP);
7163 
7164 	/* CPUID 0xD.1 */
7165 	supported_xss = 0;
7166 	if (!vmx_xsaves_supported())
7167 		kvm_cpu_cap_clear(X86_FEATURE_XSAVES);
7168 
7169 	/* CPUID 0x80000001 */
7170 	if (!cpu_has_vmx_rdtscp())
7171 		kvm_cpu_cap_clear(X86_FEATURE_RDTSCP);
7172 }
7173 
7174 static void vmx_request_immediate_exit(struct kvm_vcpu *vcpu)
7175 {
7176 	to_vmx(vcpu)->req_immediate_exit = true;
7177 }
7178 
7179 static int vmx_check_intercept_io(struct kvm_vcpu *vcpu,
7180 				  struct x86_instruction_info *info)
7181 {
7182 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
7183 	unsigned short port;
7184 	bool intercept;
7185 	int size;
7186 
7187 	if (info->intercept == x86_intercept_in ||
7188 	    info->intercept == x86_intercept_ins) {
7189 		port = info->src_val;
7190 		size = info->dst_bytes;
7191 	} else {
7192 		port = info->dst_val;
7193 		size = info->src_bytes;
7194 	}
7195 
7196 	/*
7197 	 * If the 'use IO bitmaps' VM-execution control is 0, IO instruction
7198 	 * VM-exits depend on the 'unconditional IO exiting' VM-execution
7199 	 * control.
7200 	 *
7201 	 * Otherwise, IO instruction VM-exits are controlled by the IO bitmaps.
7202 	 */
7203 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
7204 		intercept = nested_cpu_has(vmcs12,
7205 					   CPU_BASED_UNCOND_IO_EXITING);
7206 	else
7207 		intercept = nested_vmx_check_io_bitmaps(vcpu, port, size);
7208 
7209 	/* FIXME: produce nested vmexit and return X86EMUL_INTERCEPTED.  */
7210 	return intercept ? X86EMUL_UNHANDLEABLE : X86EMUL_CONTINUE;
7211 }
7212 
7213 static int vmx_check_intercept(struct kvm_vcpu *vcpu,
7214 			       struct x86_instruction_info *info,
7215 			       enum x86_intercept_stage stage,
7216 			       struct x86_exception *exception)
7217 {
7218 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
7219 
7220 	switch (info->intercept) {
7221 	/*
7222 	 * RDPID causes #UD if disabled through secondary execution controls.
7223 	 * Because it is marked as EmulateOnUD, we need to intercept it here.
7224 	 */
7225 	case x86_intercept_rdtscp:
7226 		if (!nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDTSCP)) {
7227 			exception->vector = UD_VECTOR;
7228 			exception->error_code_valid = false;
7229 			return X86EMUL_PROPAGATE_FAULT;
7230 		}
7231 		break;
7232 
7233 	case x86_intercept_in:
7234 	case x86_intercept_ins:
7235 	case x86_intercept_out:
7236 	case x86_intercept_outs:
7237 		return vmx_check_intercept_io(vcpu, info);
7238 
7239 	case x86_intercept_lgdt:
7240 	case x86_intercept_lidt:
7241 	case x86_intercept_lldt:
7242 	case x86_intercept_ltr:
7243 	case x86_intercept_sgdt:
7244 	case x86_intercept_sidt:
7245 	case x86_intercept_sldt:
7246 	case x86_intercept_str:
7247 		if (!nested_cpu_has2(vmcs12, SECONDARY_EXEC_DESC))
7248 			return X86EMUL_CONTINUE;
7249 
7250 		/* FIXME: produce nested vmexit and return X86EMUL_INTERCEPTED.  */
7251 		break;
7252 
7253 	/* TODO: check more intercepts... */
7254 	default:
7255 		break;
7256 	}
7257 
7258 	return X86EMUL_UNHANDLEABLE;
7259 }
7260 
7261 #ifdef CONFIG_X86_64
7262 /* (a << shift) / divisor, return 1 if overflow otherwise 0 */
7263 static inline int u64_shl_div_u64(u64 a, unsigned int shift,
7264 				  u64 divisor, u64 *result)
7265 {
7266 	u64 low = a << shift, high = a >> (64 - shift);
7267 
7268 	/* To avoid the overflow on divq */
7269 	if (high >= divisor)
7270 		return 1;
7271 
7272 	/* Low hold the result, high hold rem which is discarded */
7273 	asm("divq %2\n\t" : "=a" (low), "=d" (high) :
7274 	    "rm" (divisor), "0" (low), "1" (high));
7275 	*result = low;
7276 
7277 	return 0;
7278 }
7279 
7280 static int vmx_set_hv_timer(struct kvm_vcpu *vcpu, u64 guest_deadline_tsc,
7281 			    bool *expired)
7282 {
7283 	struct vcpu_vmx *vmx;
7284 	u64 tscl, guest_tscl, delta_tsc, lapic_timer_advance_cycles;
7285 	struct kvm_timer *ktimer = &vcpu->arch.apic->lapic_timer;
7286 
7287 	if (kvm_mwait_in_guest(vcpu->kvm) ||
7288 		kvm_can_post_timer_interrupt(vcpu))
7289 		return -EOPNOTSUPP;
7290 
7291 	vmx = to_vmx(vcpu);
7292 	tscl = rdtsc();
7293 	guest_tscl = kvm_read_l1_tsc(vcpu, tscl);
7294 	delta_tsc = max(guest_deadline_tsc, guest_tscl) - guest_tscl;
7295 	lapic_timer_advance_cycles = nsec_to_cycles(vcpu,
7296 						    ktimer->timer_advance_ns);
7297 
7298 	if (delta_tsc > lapic_timer_advance_cycles)
7299 		delta_tsc -= lapic_timer_advance_cycles;
7300 	else
7301 		delta_tsc = 0;
7302 
7303 	/* Convert to host delta tsc if tsc scaling is enabled */
7304 	if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio &&
7305 	    delta_tsc && u64_shl_div_u64(delta_tsc,
7306 				kvm_tsc_scaling_ratio_frac_bits,
7307 				vcpu->arch.tsc_scaling_ratio, &delta_tsc))
7308 		return -ERANGE;
7309 
7310 	/*
7311 	 * If the delta tsc can't fit in the 32 bit after the multi shift,
7312 	 * we can't use the preemption timer.
7313 	 * It's possible that it fits on later vmentries, but checking
7314 	 * on every vmentry is costly so we just use an hrtimer.
7315 	 */
7316 	if (delta_tsc >> (cpu_preemption_timer_multi + 32))
7317 		return -ERANGE;
7318 
7319 	vmx->hv_deadline_tsc = tscl + delta_tsc;
7320 	*expired = !delta_tsc;
7321 	return 0;
7322 }
7323 
7324 static void vmx_cancel_hv_timer(struct kvm_vcpu *vcpu)
7325 {
7326 	to_vmx(vcpu)->hv_deadline_tsc = -1;
7327 }
7328 #endif
7329 
7330 static void vmx_sched_in(struct kvm_vcpu *vcpu, int cpu)
7331 {
7332 	if (!kvm_pause_in_guest(vcpu->kvm))
7333 		shrink_ple_window(vcpu);
7334 }
7335 
7336 static void vmx_slot_enable_log_dirty(struct kvm *kvm,
7337 				     struct kvm_memory_slot *slot)
7338 {
7339 	if (!kvm_dirty_log_manual_protect_and_init_set(kvm))
7340 		kvm_mmu_slot_leaf_clear_dirty(kvm, slot);
7341 	kvm_mmu_slot_largepage_remove_write_access(kvm, slot);
7342 }
7343 
7344 static void vmx_slot_disable_log_dirty(struct kvm *kvm,
7345 				       struct kvm_memory_slot *slot)
7346 {
7347 	kvm_mmu_slot_set_dirty(kvm, slot);
7348 }
7349 
7350 static void vmx_flush_log_dirty(struct kvm *kvm)
7351 {
7352 	kvm_flush_pml_buffers(kvm);
7353 }
7354 
7355 static int vmx_write_pml_buffer(struct kvm_vcpu *vcpu)
7356 {
7357 	struct vmcs12 *vmcs12;
7358 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7359 	gpa_t gpa, dst;
7360 
7361 	if (is_guest_mode(vcpu)) {
7362 		WARN_ON_ONCE(vmx->nested.pml_full);
7363 
7364 		/*
7365 		 * Check if PML is enabled for the nested guest.
7366 		 * Whether eptp bit 6 is set is already checked
7367 		 * as part of A/D emulation.
7368 		 */
7369 		vmcs12 = get_vmcs12(vcpu);
7370 		if (!nested_cpu_has_pml(vmcs12))
7371 			return 0;
7372 
7373 		if (vmcs12->guest_pml_index >= PML_ENTITY_NUM) {
7374 			vmx->nested.pml_full = true;
7375 			return 1;
7376 		}
7377 
7378 		gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS) & ~0xFFFull;
7379 		dst = vmcs12->pml_address + sizeof(u64) * vmcs12->guest_pml_index;
7380 
7381 		if (kvm_write_guest_page(vcpu->kvm, gpa_to_gfn(dst), &gpa,
7382 					 offset_in_page(dst), sizeof(gpa)))
7383 			return 0;
7384 
7385 		vmcs12->guest_pml_index--;
7386 	}
7387 
7388 	return 0;
7389 }
7390 
7391 static void vmx_enable_log_dirty_pt_masked(struct kvm *kvm,
7392 					   struct kvm_memory_slot *memslot,
7393 					   gfn_t offset, unsigned long mask)
7394 {
7395 	kvm_mmu_clear_dirty_pt_masked(kvm, memslot, offset, mask);
7396 }
7397 
7398 static void __pi_post_block(struct kvm_vcpu *vcpu)
7399 {
7400 	struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
7401 	struct pi_desc old, new;
7402 	unsigned int dest;
7403 
7404 	do {
7405 		old.control = new.control = pi_desc->control;
7406 		WARN(old.nv != POSTED_INTR_WAKEUP_VECTOR,
7407 		     "Wakeup handler not enabled while the VCPU is blocked\n");
7408 
7409 		dest = cpu_physical_id(vcpu->cpu);
7410 
7411 		if (x2apic_enabled())
7412 			new.ndst = dest;
7413 		else
7414 			new.ndst = (dest << 8) & 0xFF00;
7415 
7416 		/* set 'NV' to 'notification vector' */
7417 		new.nv = POSTED_INTR_VECTOR;
7418 	} while (cmpxchg64(&pi_desc->control, old.control,
7419 			   new.control) != old.control);
7420 
7421 	if (!WARN_ON_ONCE(vcpu->pre_pcpu == -1)) {
7422 		spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7423 		list_del(&vcpu->blocked_vcpu_list);
7424 		spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7425 		vcpu->pre_pcpu = -1;
7426 	}
7427 }
7428 
7429 /*
7430  * This routine does the following things for vCPU which is going
7431  * to be blocked if VT-d PI is enabled.
7432  * - Store the vCPU to the wakeup list, so when interrupts happen
7433  *   we can find the right vCPU to wake up.
7434  * - Change the Posted-interrupt descriptor as below:
7435  *      'NDST' <-- vcpu->pre_pcpu
7436  *      'NV' <-- POSTED_INTR_WAKEUP_VECTOR
7437  * - If 'ON' is set during this process, which means at least one
7438  *   interrupt is posted for this vCPU, we cannot block it, in
7439  *   this case, return 1, otherwise, return 0.
7440  *
7441  */
7442 static int pi_pre_block(struct kvm_vcpu *vcpu)
7443 {
7444 	unsigned int dest;
7445 	struct pi_desc old, new;
7446 	struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
7447 
7448 	if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
7449 		!irq_remapping_cap(IRQ_POSTING_CAP)  ||
7450 		!kvm_vcpu_apicv_active(vcpu))
7451 		return 0;
7452 
7453 	WARN_ON(irqs_disabled());
7454 	local_irq_disable();
7455 	if (!WARN_ON_ONCE(vcpu->pre_pcpu != -1)) {
7456 		vcpu->pre_pcpu = vcpu->cpu;
7457 		spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7458 		list_add_tail(&vcpu->blocked_vcpu_list,
7459 			      &per_cpu(blocked_vcpu_on_cpu,
7460 				       vcpu->pre_pcpu));
7461 		spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7462 	}
7463 
7464 	do {
7465 		old.control = new.control = pi_desc->control;
7466 
7467 		WARN((pi_desc->sn == 1),
7468 		     "Warning: SN field of posted-interrupts "
7469 		     "is set before blocking\n");
7470 
7471 		/*
7472 		 * Since vCPU can be preempted during this process,
7473 		 * vcpu->cpu could be different with pre_pcpu, we
7474 		 * need to set pre_pcpu as the destination of wakeup
7475 		 * notification event, then we can find the right vCPU
7476 		 * to wakeup in wakeup handler if interrupts happen
7477 		 * when the vCPU is in blocked state.
7478 		 */
7479 		dest = cpu_physical_id(vcpu->pre_pcpu);
7480 
7481 		if (x2apic_enabled())
7482 			new.ndst = dest;
7483 		else
7484 			new.ndst = (dest << 8) & 0xFF00;
7485 
7486 		/* set 'NV' to 'wakeup vector' */
7487 		new.nv = POSTED_INTR_WAKEUP_VECTOR;
7488 	} while (cmpxchg64(&pi_desc->control, old.control,
7489 			   new.control) != old.control);
7490 
7491 	/* We should not block the vCPU if an interrupt is posted for it.  */
7492 	if (pi_test_on(pi_desc) == 1)
7493 		__pi_post_block(vcpu);
7494 
7495 	local_irq_enable();
7496 	return (vcpu->pre_pcpu == -1);
7497 }
7498 
7499 static int vmx_pre_block(struct kvm_vcpu *vcpu)
7500 {
7501 	if (pi_pre_block(vcpu))
7502 		return 1;
7503 
7504 	if (kvm_lapic_hv_timer_in_use(vcpu))
7505 		kvm_lapic_switch_to_sw_timer(vcpu);
7506 
7507 	return 0;
7508 }
7509 
7510 static void pi_post_block(struct kvm_vcpu *vcpu)
7511 {
7512 	if (vcpu->pre_pcpu == -1)
7513 		return;
7514 
7515 	WARN_ON(irqs_disabled());
7516 	local_irq_disable();
7517 	__pi_post_block(vcpu);
7518 	local_irq_enable();
7519 }
7520 
7521 static void vmx_post_block(struct kvm_vcpu *vcpu)
7522 {
7523 	if (kvm_x86_ops.set_hv_timer)
7524 		kvm_lapic_switch_to_hv_timer(vcpu);
7525 
7526 	pi_post_block(vcpu);
7527 }
7528 
7529 /*
7530  * vmx_update_pi_irte - set IRTE for Posted-Interrupts
7531  *
7532  * @kvm: kvm
7533  * @host_irq: host irq of the interrupt
7534  * @guest_irq: gsi of the interrupt
7535  * @set: set or unset PI
7536  * returns 0 on success, < 0 on failure
7537  */
7538 static int vmx_update_pi_irte(struct kvm *kvm, unsigned int host_irq,
7539 			      uint32_t guest_irq, bool set)
7540 {
7541 	struct kvm_kernel_irq_routing_entry *e;
7542 	struct kvm_irq_routing_table *irq_rt;
7543 	struct kvm_lapic_irq irq;
7544 	struct kvm_vcpu *vcpu;
7545 	struct vcpu_data vcpu_info;
7546 	int idx, ret = 0;
7547 
7548 	if (!kvm_arch_has_assigned_device(kvm) ||
7549 		!irq_remapping_cap(IRQ_POSTING_CAP) ||
7550 		!kvm_vcpu_apicv_active(kvm->vcpus[0]))
7551 		return 0;
7552 
7553 	idx = srcu_read_lock(&kvm->irq_srcu);
7554 	irq_rt = srcu_dereference(kvm->irq_routing, &kvm->irq_srcu);
7555 	if (guest_irq >= irq_rt->nr_rt_entries ||
7556 	    hlist_empty(&irq_rt->map[guest_irq])) {
7557 		pr_warn_once("no route for guest_irq %u/%u (broken user space?)\n",
7558 			     guest_irq, irq_rt->nr_rt_entries);
7559 		goto out;
7560 	}
7561 
7562 	hlist_for_each_entry(e, &irq_rt->map[guest_irq], link) {
7563 		if (e->type != KVM_IRQ_ROUTING_MSI)
7564 			continue;
7565 		/*
7566 		 * VT-d PI cannot support posting multicast/broadcast
7567 		 * interrupts to a vCPU, we still use interrupt remapping
7568 		 * for these kind of interrupts.
7569 		 *
7570 		 * For lowest-priority interrupts, we only support
7571 		 * those with single CPU as the destination, e.g. user
7572 		 * configures the interrupts via /proc/irq or uses
7573 		 * irqbalance to make the interrupts single-CPU.
7574 		 *
7575 		 * We will support full lowest-priority interrupt later.
7576 		 *
7577 		 * In addition, we can only inject generic interrupts using
7578 		 * the PI mechanism, refuse to route others through it.
7579 		 */
7580 
7581 		kvm_set_msi_irq(kvm, e, &irq);
7582 		if (!kvm_intr_is_single_vcpu(kvm, &irq, &vcpu) ||
7583 		    !kvm_irq_is_postable(&irq)) {
7584 			/*
7585 			 * Make sure the IRTE is in remapped mode if
7586 			 * we don't handle it in posted mode.
7587 			 */
7588 			ret = irq_set_vcpu_affinity(host_irq, NULL);
7589 			if (ret < 0) {
7590 				printk(KERN_INFO
7591 				   "failed to back to remapped mode, irq: %u\n",
7592 				   host_irq);
7593 				goto out;
7594 			}
7595 
7596 			continue;
7597 		}
7598 
7599 		vcpu_info.pi_desc_addr = __pa(vcpu_to_pi_desc(vcpu));
7600 		vcpu_info.vector = irq.vector;
7601 
7602 		trace_kvm_pi_irte_update(host_irq, vcpu->vcpu_id, e->gsi,
7603 				vcpu_info.vector, vcpu_info.pi_desc_addr, set);
7604 
7605 		if (set)
7606 			ret = irq_set_vcpu_affinity(host_irq, &vcpu_info);
7607 		else
7608 			ret = irq_set_vcpu_affinity(host_irq, NULL);
7609 
7610 		if (ret < 0) {
7611 			printk(KERN_INFO "%s: failed to update PI IRTE\n",
7612 					__func__);
7613 			goto out;
7614 		}
7615 	}
7616 
7617 	ret = 0;
7618 out:
7619 	srcu_read_unlock(&kvm->irq_srcu, idx);
7620 	return ret;
7621 }
7622 
7623 static void vmx_setup_mce(struct kvm_vcpu *vcpu)
7624 {
7625 	if (vcpu->arch.mcg_cap & MCG_LMCE_P)
7626 		to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
7627 			FEAT_CTL_LMCE_ENABLED;
7628 	else
7629 		to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
7630 			~FEAT_CTL_LMCE_ENABLED;
7631 }
7632 
7633 static int vmx_smi_allowed(struct kvm_vcpu *vcpu)
7634 {
7635 	/* we need a nested vmexit to enter SMM, postpone if run is pending */
7636 	if (to_vmx(vcpu)->nested.nested_run_pending)
7637 		return 0;
7638 	return 1;
7639 }
7640 
7641 static int vmx_pre_enter_smm(struct kvm_vcpu *vcpu, char *smstate)
7642 {
7643 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7644 
7645 	vmx->nested.smm.guest_mode = is_guest_mode(vcpu);
7646 	if (vmx->nested.smm.guest_mode)
7647 		nested_vmx_vmexit(vcpu, -1, 0, 0);
7648 
7649 	vmx->nested.smm.vmxon = vmx->nested.vmxon;
7650 	vmx->nested.vmxon = false;
7651 	vmx_clear_hlt(vcpu);
7652 	return 0;
7653 }
7654 
7655 static int vmx_pre_leave_smm(struct kvm_vcpu *vcpu, const char *smstate)
7656 {
7657 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7658 	int ret;
7659 
7660 	if (vmx->nested.smm.vmxon) {
7661 		vmx->nested.vmxon = true;
7662 		vmx->nested.smm.vmxon = false;
7663 	}
7664 
7665 	if (vmx->nested.smm.guest_mode) {
7666 		ret = nested_vmx_enter_non_root_mode(vcpu, false);
7667 		if (ret)
7668 			return ret;
7669 
7670 		vmx->nested.smm.guest_mode = false;
7671 	}
7672 	return 0;
7673 }
7674 
7675 static int enable_smi_window(struct kvm_vcpu *vcpu)
7676 {
7677 	return 0;
7678 }
7679 
7680 static bool vmx_need_emulation_on_page_fault(struct kvm_vcpu *vcpu)
7681 {
7682 	return false;
7683 }
7684 
7685 static bool vmx_apic_init_signal_blocked(struct kvm_vcpu *vcpu)
7686 {
7687 	return to_vmx(vcpu)->nested.vmxon;
7688 }
7689 
7690 static void hardware_unsetup(void)
7691 {
7692 	if (nested)
7693 		nested_vmx_hardware_unsetup();
7694 
7695 	free_kvm_area();
7696 }
7697 
7698 static bool vmx_check_apicv_inhibit_reasons(ulong bit)
7699 {
7700 	ulong supported = BIT(APICV_INHIBIT_REASON_DISABLE) |
7701 			  BIT(APICV_INHIBIT_REASON_HYPERV);
7702 
7703 	return supported & BIT(bit);
7704 }
7705 
7706 static struct kvm_x86_ops vmx_x86_ops __initdata = {
7707 	.hardware_unsetup = hardware_unsetup,
7708 
7709 	.hardware_enable = hardware_enable,
7710 	.hardware_disable = hardware_disable,
7711 	.cpu_has_accelerated_tpr = report_flexpriority,
7712 	.has_emulated_msr = vmx_has_emulated_msr,
7713 
7714 	.vm_size = sizeof(struct kvm_vmx),
7715 	.vm_init = vmx_vm_init,
7716 
7717 	.vcpu_create = vmx_create_vcpu,
7718 	.vcpu_free = vmx_free_vcpu,
7719 	.vcpu_reset = vmx_vcpu_reset,
7720 
7721 	.prepare_guest_switch = vmx_prepare_switch_to_guest,
7722 	.vcpu_load = vmx_vcpu_load,
7723 	.vcpu_put = vmx_vcpu_put,
7724 
7725 	.update_bp_intercept = update_exception_bitmap,
7726 	.get_msr_feature = vmx_get_msr_feature,
7727 	.get_msr = vmx_get_msr,
7728 	.set_msr = vmx_set_msr,
7729 	.get_segment_base = vmx_get_segment_base,
7730 	.get_segment = vmx_get_segment,
7731 	.set_segment = vmx_set_segment,
7732 	.get_cpl = vmx_get_cpl,
7733 	.get_cs_db_l_bits = vmx_get_cs_db_l_bits,
7734 	.decache_cr0_guest_bits = vmx_decache_cr0_guest_bits,
7735 	.decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
7736 	.set_cr0 = vmx_set_cr0,
7737 	.set_cr4 = vmx_set_cr4,
7738 	.set_efer = vmx_set_efer,
7739 	.get_idt = vmx_get_idt,
7740 	.set_idt = vmx_set_idt,
7741 	.get_gdt = vmx_get_gdt,
7742 	.set_gdt = vmx_set_gdt,
7743 	.get_dr6 = vmx_get_dr6,
7744 	.set_dr6 = vmx_set_dr6,
7745 	.set_dr7 = vmx_set_dr7,
7746 	.sync_dirty_debug_regs = vmx_sync_dirty_debug_regs,
7747 	.cache_reg = vmx_cache_reg,
7748 	.get_rflags = vmx_get_rflags,
7749 	.set_rflags = vmx_set_rflags,
7750 
7751 	.tlb_flush = vmx_flush_tlb,
7752 	.tlb_flush_gva = vmx_flush_tlb_gva,
7753 
7754 	.run = vmx_vcpu_run,
7755 	.handle_exit = vmx_handle_exit,
7756 	.skip_emulated_instruction = vmx_skip_emulated_instruction,
7757 	.update_emulated_instruction = vmx_update_emulated_instruction,
7758 	.set_interrupt_shadow = vmx_set_interrupt_shadow,
7759 	.get_interrupt_shadow = vmx_get_interrupt_shadow,
7760 	.patch_hypercall = vmx_patch_hypercall,
7761 	.set_irq = vmx_inject_irq,
7762 	.set_nmi = vmx_inject_nmi,
7763 	.queue_exception = vmx_queue_exception,
7764 	.cancel_injection = vmx_cancel_injection,
7765 	.interrupt_allowed = vmx_interrupt_allowed,
7766 	.nmi_allowed = vmx_nmi_allowed,
7767 	.get_nmi_mask = vmx_get_nmi_mask,
7768 	.set_nmi_mask = vmx_set_nmi_mask,
7769 	.enable_nmi_window = enable_nmi_window,
7770 	.enable_irq_window = enable_irq_window,
7771 	.update_cr8_intercept = update_cr8_intercept,
7772 	.set_virtual_apic_mode = vmx_set_virtual_apic_mode,
7773 	.set_apic_access_page_addr = vmx_set_apic_access_page_addr,
7774 	.refresh_apicv_exec_ctrl = vmx_refresh_apicv_exec_ctrl,
7775 	.load_eoi_exitmap = vmx_load_eoi_exitmap,
7776 	.apicv_post_state_restore = vmx_apicv_post_state_restore,
7777 	.check_apicv_inhibit_reasons = vmx_check_apicv_inhibit_reasons,
7778 	.hwapic_irr_update = vmx_hwapic_irr_update,
7779 	.hwapic_isr_update = vmx_hwapic_isr_update,
7780 	.guest_apic_has_interrupt = vmx_guest_apic_has_interrupt,
7781 	.sync_pir_to_irr = vmx_sync_pir_to_irr,
7782 	.deliver_posted_interrupt = vmx_deliver_posted_interrupt,
7783 	.dy_apicv_has_pending_interrupt = vmx_dy_apicv_has_pending_interrupt,
7784 
7785 	.set_tss_addr = vmx_set_tss_addr,
7786 	.set_identity_map_addr = vmx_set_identity_map_addr,
7787 	.get_tdp_level = get_ept_level,
7788 	.get_mt_mask = vmx_get_mt_mask,
7789 
7790 	.get_exit_info = vmx_get_exit_info,
7791 
7792 	.cpuid_update = vmx_cpuid_update,
7793 
7794 	.has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
7795 
7796 	.read_l1_tsc_offset = vmx_read_l1_tsc_offset,
7797 	.write_l1_tsc_offset = vmx_write_l1_tsc_offset,
7798 
7799 	.load_mmu_pgd = vmx_load_mmu_pgd,
7800 
7801 	.check_intercept = vmx_check_intercept,
7802 	.handle_exit_irqoff = vmx_handle_exit_irqoff,
7803 
7804 	.request_immediate_exit = vmx_request_immediate_exit,
7805 
7806 	.sched_in = vmx_sched_in,
7807 
7808 	.slot_enable_log_dirty = vmx_slot_enable_log_dirty,
7809 	.slot_disable_log_dirty = vmx_slot_disable_log_dirty,
7810 	.flush_log_dirty = vmx_flush_log_dirty,
7811 	.enable_log_dirty_pt_masked = vmx_enable_log_dirty_pt_masked,
7812 	.write_log_dirty = vmx_write_pml_buffer,
7813 
7814 	.pre_block = vmx_pre_block,
7815 	.post_block = vmx_post_block,
7816 
7817 	.pmu_ops = &intel_pmu_ops,
7818 
7819 	.update_pi_irte = vmx_update_pi_irte,
7820 
7821 #ifdef CONFIG_X86_64
7822 	.set_hv_timer = vmx_set_hv_timer,
7823 	.cancel_hv_timer = vmx_cancel_hv_timer,
7824 #endif
7825 
7826 	.setup_mce = vmx_setup_mce,
7827 
7828 	.smi_allowed = vmx_smi_allowed,
7829 	.pre_enter_smm = vmx_pre_enter_smm,
7830 	.pre_leave_smm = vmx_pre_leave_smm,
7831 	.enable_smi_window = enable_smi_window,
7832 
7833 	.check_nested_events = NULL,
7834 	.get_nested_state = NULL,
7835 	.set_nested_state = NULL,
7836 	.get_vmcs12_pages = NULL,
7837 	.nested_enable_evmcs = NULL,
7838 	.nested_get_evmcs_version = NULL,
7839 	.need_emulation_on_page_fault = vmx_need_emulation_on_page_fault,
7840 	.apic_init_signal_blocked = vmx_apic_init_signal_blocked,
7841 };
7842 
7843 static __init int hardware_setup(void)
7844 {
7845 	unsigned long host_bndcfgs;
7846 	struct desc_ptr dt;
7847 	int r, i, ept_lpage_level;
7848 
7849 	store_idt(&dt);
7850 	host_idt_base = dt.address;
7851 
7852 	for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i)
7853 		kvm_define_shared_msr(i, vmx_msr_index[i]);
7854 
7855 	if (setup_vmcs_config(&vmcs_config, &vmx_capability) < 0)
7856 		return -EIO;
7857 
7858 	if (boot_cpu_has(X86_FEATURE_NX))
7859 		kvm_enable_efer_bits(EFER_NX);
7860 
7861 	if (boot_cpu_has(X86_FEATURE_MPX)) {
7862 		rdmsrl(MSR_IA32_BNDCFGS, host_bndcfgs);
7863 		WARN_ONCE(host_bndcfgs, "KVM: BNDCFGS in host will be lost");
7864 	}
7865 
7866 	if (!cpu_has_vmx_mpx())
7867 		supported_xcr0 &= ~(XFEATURE_MASK_BNDREGS |
7868 				    XFEATURE_MASK_BNDCSR);
7869 
7870 	if (!cpu_has_vmx_vpid() || !cpu_has_vmx_invvpid() ||
7871 	    !(cpu_has_vmx_invvpid_single() || cpu_has_vmx_invvpid_global()))
7872 		enable_vpid = 0;
7873 
7874 	if (!cpu_has_vmx_ept() ||
7875 	    !cpu_has_vmx_ept_4levels() ||
7876 	    !cpu_has_vmx_ept_mt_wb() ||
7877 	    !cpu_has_vmx_invept_global())
7878 		enable_ept = 0;
7879 
7880 	if (!cpu_has_vmx_ept_ad_bits() || !enable_ept)
7881 		enable_ept_ad_bits = 0;
7882 
7883 	if (!cpu_has_vmx_unrestricted_guest() || !enable_ept)
7884 		enable_unrestricted_guest = 0;
7885 
7886 	if (!cpu_has_vmx_flexpriority())
7887 		flexpriority_enabled = 0;
7888 
7889 	if (!cpu_has_virtual_nmis())
7890 		enable_vnmi = 0;
7891 
7892 	/*
7893 	 * set_apic_access_page_addr() is used to reload apic access
7894 	 * page upon invalidation.  No need to do anything if not
7895 	 * using the APIC_ACCESS_ADDR VMCS field.
7896 	 */
7897 	if (!flexpriority_enabled)
7898 		vmx_x86_ops.set_apic_access_page_addr = NULL;
7899 
7900 	if (!cpu_has_vmx_tpr_shadow())
7901 		vmx_x86_ops.update_cr8_intercept = NULL;
7902 
7903 #if IS_ENABLED(CONFIG_HYPERV)
7904 	if (ms_hyperv.nested_features & HV_X64_NESTED_GUEST_MAPPING_FLUSH
7905 	    && enable_ept) {
7906 		vmx_x86_ops.tlb_remote_flush = hv_remote_flush_tlb;
7907 		vmx_x86_ops.tlb_remote_flush_with_range =
7908 				hv_remote_flush_tlb_with_range;
7909 	}
7910 #endif
7911 
7912 	if (!cpu_has_vmx_ple()) {
7913 		ple_gap = 0;
7914 		ple_window = 0;
7915 		ple_window_grow = 0;
7916 		ple_window_max = 0;
7917 		ple_window_shrink = 0;
7918 	}
7919 
7920 	if (!cpu_has_vmx_apicv()) {
7921 		enable_apicv = 0;
7922 		vmx_x86_ops.sync_pir_to_irr = NULL;
7923 	}
7924 
7925 	if (cpu_has_vmx_tsc_scaling()) {
7926 		kvm_has_tsc_control = true;
7927 		kvm_max_tsc_scaling_ratio = KVM_VMX_TSC_MULTIPLIER_MAX;
7928 		kvm_tsc_scaling_ratio_frac_bits = 48;
7929 	}
7930 
7931 	set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
7932 
7933 	if (enable_ept)
7934 		vmx_enable_tdp();
7935 
7936 	if (!enable_ept)
7937 		ept_lpage_level = 0;
7938 	else if (cpu_has_vmx_ept_1g_page())
7939 		ept_lpage_level = PT_PDPE_LEVEL;
7940 	else if (cpu_has_vmx_ept_2m_page())
7941 		ept_lpage_level = PT_DIRECTORY_LEVEL;
7942 	else
7943 		ept_lpage_level = PT_PAGE_TABLE_LEVEL;
7944 	kvm_configure_mmu(enable_ept, ept_lpage_level);
7945 
7946 	/*
7947 	 * Only enable PML when hardware supports PML feature, and both EPT
7948 	 * and EPT A/D bit features are enabled -- PML depends on them to work.
7949 	 */
7950 	if (!enable_ept || !enable_ept_ad_bits || !cpu_has_vmx_pml())
7951 		enable_pml = 0;
7952 
7953 	if (!enable_pml) {
7954 		vmx_x86_ops.slot_enable_log_dirty = NULL;
7955 		vmx_x86_ops.slot_disable_log_dirty = NULL;
7956 		vmx_x86_ops.flush_log_dirty = NULL;
7957 		vmx_x86_ops.enable_log_dirty_pt_masked = NULL;
7958 	}
7959 
7960 	if (!cpu_has_vmx_preemption_timer())
7961 		enable_preemption_timer = false;
7962 
7963 	if (enable_preemption_timer) {
7964 		u64 use_timer_freq = 5000ULL * 1000 * 1000;
7965 		u64 vmx_msr;
7966 
7967 		rdmsrl(MSR_IA32_VMX_MISC, vmx_msr);
7968 		cpu_preemption_timer_multi =
7969 			vmx_msr & VMX_MISC_PREEMPTION_TIMER_RATE_MASK;
7970 
7971 		if (tsc_khz)
7972 			use_timer_freq = (u64)tsc_khz * 1000;
7973 		use_timer_freq >>= cpu_preemption_timer_multi;
7974 
7975 		/*
7976 		 * KVM "disables" the preemption timer by setting it to its max
7977 		 * value.  Don't use the timer if it might cause spurious exits
7978 		 * at a rate faster than 0.1 Hz (of uninterrupted guest time).
7979 		 */
7980 		if (use_timer_freq > 0xffffffffu / 10)
7981 			enable_preemption_timer = false;
7982 	}
7983 
7984 	if (!enable_preemption_timer) {
7985 		vmx_x86_ops.set_hv_timer = NULL;
7986 		vmx_x86_ops.cancel_hv_timer = NULL;
7987 		vmx_x86_ops.request_immediate_exit = __kvm_request_immediate_exit;
7988 	}
7989 
7990 	kvm_set_posted_intr_wakeup_handler(wakeup_handler);
7991 
7992 	kvm_mce_cap_supported |= MCG_LMCE_P;
7993 
7994 	if (pt_mode != PT_MODE_SYSTEM && pt_mode != PT_MODE_HOST_GUEST)
7995 		return -EINVAL;
7996 	if (!enable_ept || !cpu_has_vmx_intel_pt())
7997 		pt_mode = PT_MODE_SYSTEM;
7998 
7999 	if (nested) {
8000 		nested_vmx_setup_ctls_msrs(&vmcs_config.nested,
8001 					   vmx_capability.ept);
8002 
8003 		r = nested_vmx_hardware_setup(&vmx_x86_ops,
8004 					      kvm_vmx_exit_handlers);
8005 		if (r)
8006 			return r;
8007 	}
8008 
8009 	vmx_set_cpu_caps();
8010 
8011 	r = alloc_kvm_area();
8012 	if (r)
8013 		nested_vmx_hardware_unsetup();
8014 	return r;
8015 }
8016 
8017 static struct kvm_x86_init_ops vmx_init_ops __initdata = {
8018 	.cpu_has_kvm_support = cpu_has_kvm_support,
8019 	.disabled_by_bios = vmx_disabled_by_bios,
8020 	.check_processor_compatibility = vmx_check_processor_compat,
8021 	.hardware_setup = hardware_setup,
8022 
8023 	.runtime_ops = &vmx_x86_ops,
8024 };
8025 
8026 static void vmx_cleanup_l1d_flush(void)
8027 {
8028 	if (vmx_l1d_flush_pages) {
8029 		free_pages((unsigned long)vmx_l1d_flush_pages, L1D_CACHE_ORDER);
8030 		vmx_l1d_flush_pages = NULL;
8031 	}
8032 	/* Restore state so sysfs ignores VMX */
8033 	l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_AUTO;
8034 }
8035 
8036 static void vmx_exit(void)
8037 {
8038 #ifdef CONFIG_KEXEC_CORE
8039 	RCU_INIT_POINTER(crash_vmclear_loaded_vmcss, NULL);
8040 	synchronize_rcu();
8041 #endif
8042 
8043 	kvm_exit();
8044 
8045 #if IS_ENABLED(CONFIG_HYPERV)
8046 	if (static_branch_unlikely(&enable_evmcs)) {
8047 		int cpu;
8048 		struct hv_vp_assist_page *vp_ap;
8049 		/*
8050 		 * Reset everything to support using non-enlightened VMCS
8051 		 * access later (e.g. when we reload the module with
8052 		 * enlightened_vmcs=0)
8053 		 */
8054 		for_each_online_cpu(cpu) {
8055 			vp_ap =	hv_get_vp_assist_page(cpu);
8056 
8057 			if (!vp_ap)
8058 				continue;
8059 
8060 			vp_ap->nested_control.features.directhypercall = 0;
8061 			vp_ap->current_nested_vmcs = 0;
8062 			vp_ap->enlighten_vmentry = 0;
8063 		}
8064 
8065 		static_branch_disable(&enable_evmcs);
8066 	}
8067 #endif
8068 	vmx_cleanup_l1d_flush();
8069 }
8070 module_exit(vmx_exit);
8071 
8072 static int __init vmx_init(void)
8073 {
8074 	int r, cpu;
8075 
8076 #if IS_ENABLED(CONFIG_HYPERV)
8077 	/*
8078 	 * Enlightened VMCS usage should be recommended and the host needs
8079 	 * to support eVMCS v1 or above. We can also disable eVMCS support
8080 	 * with module parameter.
8081 	 */
8082 	if (enlightened_vmcs &&
8083 	    ms_hyperv.hints & HV_X64_ENLIGHTENED_VMCS_RECOMMENDED &&
8084 	    (ms_hyperv.nested_features & HV_X64_ENLIGHTENED_VMCS_VERSION) >=
8085 	    KVM_EVMCS_VERSION) {
8086 		int cpu;
8087 
8088 		/* Check that we have assist pages on all online CPUs */
8089 		for_each_online_cpu(cpu) {
8090 			if (!hv_get_vp_assist_page(cpu)) {
8091 				enlightened_vmcs = false;
8092 				break;
8093 			}
8094 		}
8095 
8096 		if (enlightened_vmcs) {
8097 			pr_info("KVM: vmx: using Hyper-V Enlightened VMCS\n");
8098 			static_branch_enable(&enable_evmcs);
8099 		}
8100 
8101 		if (ms_hyperv.nested_features & HV_X64_NESTED_DIRECT_FLUSH)
8102 			vmx_x86_ops.enable_direct_tlbflush
8103 				= hv_enable_direct_tlbflush;
8104 
8105 	} else {
8106 		enlightened_vmcs = false;
8107 	}
8108 #endif
8109 
8110 	r = kvm_init(&vmx_init_ops, sizeof(struct vcpu_vmx),
8111 		     __alignof__(struct vcpu_vmx), THIS_MODULE);
8112 	if (r)
8113 		return r;
8114 
8115 	/*
8116 	 * Must be called after kvm_init() so enable_ept is properly set
8117 	 * up. Hand the parameter mitigation value in which was stored in
8118 	 * the pre module init parser. If no parameter was given, it will
8119 	 * contain 'auto' which will be turned into the default 'cond'
8120 	 * mitigation mode.
8121 	 */
8122 	r = vmx_setup_l1d_flush(vmentry_l1d_flush_param);
8123 	if (r) {
8124 		vmx_exit();
8125 		return r;
8126 	}
8127 
8128 	for_each_possible_cpu(cpu) {
8129 		INIT_LIST_HEAD(&per_cpu(loaded_vmcss_on_cpu, cpu));
8130 		INIT_LIST_HEAD(&per_cpu(blocked_vcpu_on_cpu, cpu));
8131 		spin_lock_init(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
8132 	}
8133 
8134 #ifdef CONFIG_KEXEC_CORE
8135 	rcu_assign_pointer(crash_vmclear_loaded_vmcss,
8136 			   crash_vmclear_local_loaded_vmcss);
8137 #endif
8138 	vmx_check_vmcs12_offsets();
8139 
8140 	return 0;
8141 }
8142 module_init(vmx_init);
8143