xref: /linux/arch/x86/kvm/vmx/vmx.c (revision 86287543715ac2a6d92d561cc105d79306511457)
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 u64 host_efer;
441 static unsigned long host_idt_base;
442 
443 /*
444  * Though SYSCALL is only supported in 64-bit mode on Intel CPUs, kvm
445  * will emulate SYSCALL in legacy mode if the vendor string in guest
446  * CPUID.0:{EBX,ECX,EDX} is "AuthenticAMD" or "AMDisbetter!" To
447  * support this emulation, IA32_STAR must always be included in
448  * vmx_msr_index[], even in i386 builds.
449  */
450 const u32 vmx_msr_index[] = {
451 #ifdef CONFIG_X86_64
452 	MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
453 #endif
454 	MSR_EFER, MSR_TSC_AUX, MSR_STAR,
455 	MSR_IA32_TSX_CTRL,
456 };
457 
458 #if IS_ENABLED(CONFIG_HYPERV)
459 static bool __read_mostly enlightened_vmcs = true;
460 module_param(enlightened_vmcs, bool, 0444);
461 
462 /* check_ept_pointer() should be under protection of ept_pointer_lock. */
463 static void check_ept_pointer_match(struct kvm *kvm)
464 {
465 	struct kvm_vcpu *vcpu;
466 	u64 tmp_eptp = INVALID_PAGE;
467 	int i;
468 
469 	kvm_for_each_vcpu(i, vcpu, kvm) {
470 		if (!VALID_PAGE(tmp_eptp)) {
471 			tmp_eptp = to_vmx(vcpu)->ept_pointer;
472 		} else if (tmp_eptp != to_vmx(vcpu)->ept_pointer) {
473 			to_kvm_vmx(kvm)->ept_pointers_match
474 				= EPT_POINTERS_MISMATCH;
475 			return;
476 		}
477 	}
478 
479 	to_kvm_vmx(kvm)->ept_pointers_match = EPT_POINTERS_MATCH;
480 }
481 
482 static int kvm_fill_hv_flush_list_func(struct hv_guest_mapping_flush_list *flush,
483 		void *data)
484 {
485 	struct kvm_tlb_range *range = data;
486 
487 	return hyperv_fill_flush_guest_mapping_list(flush, range->start_gfn,
488 			range->pages);
489 }
490 
491 static inline int __hv_remote_flush_tlb_with_range(struct kvm *kvm,
492 		struct kvm_vcpu *vcpu, struct kvm_tlb_range *range)
493 {
494 	u64 ept_pointer = to_vmx(vcpu)->ept_pointer;
495 
496 	/*
497 	 * FLUSH_GUEST_PHYSICAL_ADDRESS_SPACE hypercall needs address
498 	 * of the base of EPT PML4 table, strip off EPT configuration
499 	 * information.
500 	 */
501 	if (range)
502 		return hyperv_flush_guest_mapping_range(ept_pointer & PAGE_MASK,
503 				kvm_fill_hv_flush_list_func, (void *)range);
504 	else
505 		return hyperv_flush_guest_mapping(ept_pointer & PAGE_MASK);
506 }
507 
508 static int hv_remote_flush_tlb_with_range(struct kvm *kvm,
509 		struct kvm_tlb_range *range)
510 {
511 	struct kvm_vcpu *vcpu;
512 	int ret = 0, i;
513 
514 	spin_lock(&to_kvm_vmx(kvm)->ept_pointer_lock);
515 
516 	if (to_kvm_vmx(kvm)->ept_pointers_match == EPT_POINTERS_CHECK)
517 		check_ept_pointer_match(kvm);
518 
519 	if (to_kvm_vmx(kvm)->ept_pointers_match != EPT_POINTERS_MATCH) {
520 		kvm_for_each_vcpu(i, vcpu, kvm) {
521 			/* If ept_pointer is invalid pointer, bypass flush request. */
522 			if (VALID_PAGE(to_vmx(vcpu)->ept_pointer))
523 				ret |= __hv_remote_flush_tlb_with_range(
524 					kvm, vcpu, range);
525 		}
526 	} else {
527 		ret = __hv_remote_flush_tlb_with_range(kvm,
528 				kvm_get_vcpu(kvm, 0), range);
529 	}
530 
531 	spin_unlock(&to_kvm_vmx(kvm)->ept_pointer_lock);
532 	return ret;
533 }
534 static int hv_remote_flush_tlb(struct kvm *kvm)
535 {
536 	return hv_remote_flush_tlb_with_range(kvm, NULL);
537 }
538 
539 static int hv_enable_direct_tlbflush(struct kvm_vcpu *vcpu)
540 {
541 	struct hv_enlightened_vmcs *evmcs;
542 	struct hv_partition_assist_pg **p_hv_pa_pg =
543 			&vcpu->kvm->arch.hyperv.hv_pa_pg;
544 	/*
545 	 * Synthetic VM-Exit is not enabled in current code and so All
546 	 * evmcs in singe VM shares same assist page.
547 	 */
548 	if (!*p_hv_pa_pg)
549 		*p_hv_pa_pg = kzalloc(PAGE_SIZE, GFP_KERNEL);
550 
551 	if (!*p_hv_pa_pg)
552 		return -ENOMEM;
553 
554 	evmcs = (struct hv_enlightened_vmcs *)to_vmx(vcpu)->loaded_vmcs->vmcs;
555 
556 	evmcs->partition_assist_page =
557 		__pa(*p_hv_pa_pg);
558 	evmcs->hv_vm_id = (unsigned long)vcpu->kvm;
559 	evmcs->hv_enlightenments_control.nested_flush_hypercall = 1;
560 
561 	return 0;
562 }
563 
564 #endif /* IS_ENABLED(CONFIG_HYPERV) */
565 
566 /*
567  * Comment's format: document - errata name - stepping - processor name.
568  * Refer from
569  * https://www.virtualbox.org/svn/vbox/trunk/src/VBox/VMM/VMMR0/HMR0.cpp
570  */
571 static u32 vmx_preemption_cpu_tfms[] = {
572 /* 323344.pdf - BA86   - D0 - Xeon 7500 Series */
573 0x000206E6,
574 /* 323056.pdf - AAX65  - C2 - Xeon L3406 */
575 /* 322814.pdf - AAT59  - C2 - i7-600, i5-500, i5-400 and i3-300 Mobile */
576 /* 322911.pdf - AAU65  - C2 - i5-600, i3-500 Desktop and Pentium G6950 */
577 0x00020652,
578 /* 322911.pdf - AAU65  - K0 - i5-600, i3-500 Desktop and Pentium G6950 */
579 0x00020655,
580 /* 322373.pdf - AAO95  - B1 - Xeon 3400 Series */
581 /* 322166.pdf - AAN92  - B1 - i7-800 and i5-700 Desktop */
582 /*
583  * 320767.pdf - AAP86  - B1 -
584  * i7-900 Mobile Extreme, i7-800 and i7-700 Mobile
585  */
586 0x000106E5,
587 /* 321333.pdf - AAM126 - C0 - Xeon 3500 */
588 0x000106A0,
589 /* 321333.pdf - AAM126 - C1 - Xeon 3500 */
590 0x000106A1,
591 /* 320836.pdf - AAJ124 - C0 - i7-900 Desktop Extreme and i7-900 Desktop */
592 0x000106A4,
593  /* 321333.pdf - AAM126 - D0 - Xeon 3500 */
594  /* 321324.pdf - AAK139 - D0 - Xeon 5500 */
595  /* 320836.pdf - AAJ124 - D0 - i7-900 Extreme and i7-900 Desktop */
596 0x000106A5,
597  /* Xeon E3-1220 V2 */
598 0x000306A8,
599 };
600 
601 static inline bool cpu_has_broken_vmx_preemption_timer(void)
602 {
603 	u32 eax = cpuid_eax(0x00000001), i;
604 
605 	/* Clear the reserved bits */
606 	eax &= ~(0x3U << 14 | 0xfU << 28);
607 	for (i = 0; i < ARRAY_SIZE(vmx_preemption_cpu_tfms); i++)
608 		if (eax == vmx_preemption_cpu_tfms[i])
609 			return true;
610 
611 	return false;
612 }
613 
614 static inline bool cpu_need_virtualize_apic_accesses(struct kvm_vcpu *vcpu)
615 {
616 	return flexpriority_enabled && lapic_in_kernel(vcpu);
617 }
618 
619 static inline bool report_flexpriority(void)
620 {
621 	return flexpriority_enabled;
622 }
623 
624 static inline int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
625 {
626 	int i;
627 
628 	for (i = 0; i < vmx->nmsrs; ++i)
629 		if (vmx_msr_index[vmx->guest_msrs[i].index] == msr)
630 			return i;
631 	return -1;
632 }
633 
634 struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
635 {
636 	int i;
637 
638 	i = __find_msr_index(vmx, msr);
639 	if (i >= 0)
640 		return &vmx->guest_msrs[i];
641 	return NULL;
642 }
643 
644 static int vmx_set_guest_msr(struct vcpu_vmx *vmx, struct shared_msr_entry *msr, u64 data)
645 {
646 	int ret = 0;
647 
648 	u64 old_msr_data = msr->data;
649 	msr->data = data;
650 	if (msr - vmx->guest_msrs < vmx->save_nmsrs) {
651 		preempt_disable();
652 		ret = kvm_set_shared_msr(msr->index, msr->data,
653 					 msr->mask);
654 		preempt_enable();
655 		if (ret)
656 			msr->data = old_msr_data;
657 	}
658 	return ret;
659 }
660 
661 void loaded_vmcs_init(struct loaded_vmcs *loaded_vmcs)
662 {
663 	vmcs_clear(loaded_vmcs->vmcs);
664 	if (loaded_vmcs->shadow_vmcs && loaded_vmcs->launched)
665 		vmcs_clear(loaded_vmcs->shadow_vmcs);
666 	loaded_vmcs->cpu = -1;
667 	loaded_vmcs->launched = 0;
668 }
669 
670 #ifdef CONFIG_KEXEC_CORE
671 /*
672  * This bitmap is used to indicate whether the vmclear
673  * operation is enabled on all cpus. All disabled by
674  * default.
675  */
676 static cpumask_t crash_vmclear_enabled_bitmap = CPU_MASK_NONE;
677 
678 static inline void crash_enable_local_vmclear(int cpu)
679 {
680 	cpumask_set_cpu(cpu, &crash_vmclear_enabled_bitmap);
681 }
682 
683 static inline void crash_disable_local_vmclear(int cpu)
684 {
685 	cpumask_clear_cpu(cpu, &crash_vmclear_enabled_bitmap);
686 }
687 
688 static inline int crash_local_vmclear_enabled(int cpu)
689 {
690 	return cpumask_test_cpu(cpu, &crash_vmclear_enabled_bitmap);
691 }
692 
693 static void crash_vmclear_local_loaded_vmcss(void)
694 {
695 	int cpu = raw_smp_processor_id();
696 	struct loaded_vmcs *v;
697 
698 	if (!crash_local_vmclear_enabled(cpu))
699 		return;
700 
701 	list_for_each_entry(v, &per_cpu(loaded_vmcss_on_cpu, cpu),
702 			    loaded_vmcss_on_cpu_link)
703 		vmcs_clear(v->vmcs);
704 }
705 #else
706 static inline void crash_enable_local_vmclear(int cpu) { }
707 static inline void crash_disable_local_vmclear(int cpu) { }
708 #endif /* CONFIG_KEXEC_CORE */
709 
710 static void __loaded_vmcs_clear(void *arg)
711 {
712 	struct loaded_vmcs *loaded_vmcs = arg;
713 	int cpu = raw_smp_processor_id();
714 
715 	if (loaded_vmcs->cpu != cpu)
716 		return; /* vcpu migration can race with cpu offline */
717 	if (per_cpu(current_vmcs, cpu) == loaded_vmcs->vmcs)
718 		per_cpu(current_vmcs, cpu) = NULL;
719 	crash_disable_local_vmclear(cpu);
720 	list_del(&loaded_vmcs->loaded_vmcss_on_cpu_link);
721 
722 	/*
723 	 * we should ensure updating loaded_vmcs->loaded_vmcss_on_cpu_link
724 	 * is before setting loaded_vmcs->vcpu to -1 which is done in
725 	 * loaded_vmcs_init. Otherwise, other cpu can see vcpu = -1 fist
726 	 * then adds the vmcs into percpu list before it is deleted.
727 	 */
728 	smp_wmb();
729 
730 	loaded_vmcs_init(loaded_vmcs);
731 	crash_enable_local_vmclear(cpu);
732 }
733 
734 void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs)
735 {
736 	int cpu = loaded_vmcs->cpu;
737 
738 	if (cpu != -1)
739 		smp_call_function_single(cpu,
740 			 __loaded_vmcs_clear, loaded_vmcs, 1);
741 }
742 
743 static bool vmx_segment_cache_test_set(struct vcpu_vmx *vmx, unsigned seg,
744 				       unsigned field)
745 {
746 	bool ret;
747 	u32 mask = 1 << (seg * SEG_FIELD_NR + field);
748 
749 	if (!kvm_register_is_available(&vmx->vcpu, VCPU_EXREG_SEGMENTS)) {
750 		kvm_register_mark_available(&vmx->vcpu, VCPU_EXREG_SEGMENTS);
751 		vmx->segment_cache.bitmask = 0;
752 	}
753 	ret = vmx->segment_cache.bitmask & mask;
754 	vmx->segment_cache.bitmask |= mask;
755 	return ret;
756 }
757 
758 static u16 vmx_read_guest_seg_selector(struct vcpu_vmx *vmx, unsigned seg)
759 {
760 	u16 *p = &vmx->segment_cache.seg[seg].selector;
761 
762 	if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_SEL))
763 		*p = vmcs_read16(kvm_vmx_segment_fields[seg].selector);
764 	return *p;
765 }
766 
767 static ulong vmx_read_guest_seg_base(struct vcpu_vmx *vmx, unsigned seg)
768 {
769 	ulong *p = &vmx->segment_cache.seg[seg].base;
770 
771 	if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_BASE))
772 		*p = vmcs_readl(kvm_vmx_segment_fields[seg].base);
773 	return *p;
774 }
775 
776 static u32 vmx_read_guest_seg_limit(struct vcpu_vmx *vmx, unsigned seg)
777 {
778 	u32 *p = &vmx->segment_cache.seg[seg].limit;
779 
780 	if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_LIMIT))
781 		*p = vmcs_read32(kvm_vmx_segment_fields[seg].limit);
782 	return *p;
783 }
784 
785 static u32 vmx_read_guest_seg_ar(struct vcpu_vmx *vmx, unsigned seg)
786 {
787 	u32 *p = &vmx->segment_cache.seg[seg].ar;
788 
789 	if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_AR))
790 		*p = vmcs_read32(kvm_vmx_segment_fields[seg].ar_bytes);
791 	return *p;
792 }
793 
794 void update_exception_bitmap(struct kvm_vcpu *vcpu)
795 {
796 	u32 eb;
797 
798 	eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR) |
799 	     (1u << DB_VECTOR) | (1u << AC_VECTOR);
800 	/*
801 	 * Guest access to VMware backdoor ports could legitimately
802 	 * trigger #GP because of TSS I/O permission bitmap.
803 	 * We intercept those #GP and allow access to them anyway
804 	 * as VMware does.
805 	 */
806 	if (enable_vmware_backdoor)
807 		eb |= (1u << GP_VECTOR);
808 	if ((vcpu->guest_debug &
809 	     (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) ==
810 	    (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP))
811 		eb |= 1u << BP_VECTOR;
812 	if (to_vmx(vcpu)->rmode.vm86_active)
813 		eb = ~0;
814 	if (enable_ept)
815 		eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */
816 
817 	/* When we are running a nested L2 guest and L1 specified for it a
818 	 * certain exception bitmap, we must trap the same exceptions and pass
819 	 * them to L1. When running L2, we will only handle the exceptions
820 	 * specified above if L1 did not want them.
821 	 */
822 	if (is_guest_mode(vcpu))
823 		eb |= get_vmcs12(vcpu)->exception_bitmap;
824 
825 	vmcs_write32(EXCEPTION_BITMAP, eb);
826 }
827 
828 /*
829  * Check if MSR is intercepted for currently loaded MSR bitmap.
830  */
831 static bool msr_write_intercepted(struct kvm_vcpu *vcpu, u32 msr)
832 {
833 	unsigned long *msr_bitmap;
834 	int f = sizeof(unsigned long);
835 
836 	if (!cpu_has_vmx_msr_bitmap())
837 		return true;
838 
839 	msr_bitmap = to_vmx(vcpu)->loaded_vmcs->msr_bitmap;
840 
841 	if (msr <= 0x1fff) {
842 		return !!test_bit(msr, msr_bitmap + 0x800 / f);
843 	} else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
844 		msr &= 0x1fff;
845 		return !!test_bit(msr, msr_bitmap + 0xc00 / f);
846 	}
847 
848 	return true;
849 }
850 
851 static void clear_atomic_switch_msr_special(struct vcpu_vmx *vmx,
852 		unsigned long entry, unsigned long exit)
853 {
854 	vm_entry_controls_clearbit(vmx, entry);
855 	vm_exit_controls_clearbit(vmx, exit);
856 }
857 
858 int vmx_find_msr_index(struct vmx_msrs *m, u32 msr)
859 {
860 	unsigned int i;
861 
862 	for (i = 0; i < m->nr; ++i) {
863 		if (m->val[i].index == msr)
864 			return i;
865 	}
866 	return -ENOENT;
867 }
868 
869 static void clear_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr)
870 {
871 	int i;
872 	struct msr_autoload *m = &vmx->msr_autoload;
873 
874 	switch (msr) {
875 	case MSR_EFER:
876 		if (cpu_has_load_ia32_efer()) {
877 			clear_atomic_switch_msr_special(vmx,
878 					VM_ENTRY_LOAD_IA32_EFER,
879 					VM_EXIT_LOAD_IA32_EFER);
880 			return;
881 		}
882 		break;
883 	case MSR_CORE_PERF_GLOBAL_CTRL:
884 		if (cpu_has_load_perf_global_ctrl()) {
885 			clear_atomic_switch_msr_special(vmx,
886 					VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
887 					VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
888 			return;
889 		}
890 		break;
891 	}
892 	i = vmx_find_msr_index(&m->guest, msr);
893 	if (i < 0)
894 		goto skip_guest;
895 	--m->guest.nr;
896 	m->guest.val[i] = m->guest.val[m->guest.nr];
897 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->guest.nr);
898 
899 skip_guest:
900 	i = vmx_find_msr_index(&m->host, msr);
901 	if (i < 0)
902 		return;
903 
904 	--m->host.nr;
905 	m->host.val[i] = m->host.val[m->host.nr];
906 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->host.nr);
907 }
908 
909 static void add_atomic_switch_msr_special(struct vcpu_vmx *vmx,
910 		unsigned long entry, unsigned long exit,
911 		unsigned long guest_val_vmcs, unsigned long host_val_vmcs,
912 		u64 guest_val, u64 host_val)
913 {
914 	vmcs_write64(guest_val_vmcs, guest_val);
915 	if (host_val_vmcs != HOST_IA32_EFER)
916 		vmcs_write64(host_val_vmcs, host_val);
917 	vm_entry_controls_setbit(vmx, entry);
918 	vm_exit_controls_setbit(vmx, exit);
919 }
920 
921 static void add_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr,
922 				  u64 guest_val, u64 host_val, bool entry_only)
923 {
924 	int i, j = 0;
925 	struct msr_autoload *m = &vmx->msr_autoload;
926 
927 	switch (msr) {
928 	case MSR_EFER:
929 		if (cpu_has_load_ia32_efer()) {
930 			add_atomic_switch_msr_special(vmx,
931 					VM_ENTRY_LOAD_IA32_EFER,
932 					VM_EXIT_LOAD_IA32_EFER,
933 					GUEST_IA32_EFER,
934 					HOST_IA32_EFER,
935 					guest_val, host_val);
936 			return;
937 		}
938 		break;
939 	case MSR_CORE_PERF_GLOBAL_CTRL:
940 		if (cpu_has_load_perf_global_ctrl()) {
941 			add_atomic_switch_msr_special(vmx,
942 					VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
943 					VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL,
944 					GUEST_IA32_PERF_GLOBAL_CTRL,
945 					HOST_IA32_PERF_GLOBAL_CTRL,
946 					guest_val, host_val);
947 			return;
948 		}
949 		break;
950 	case MSR_IA32_PEBS_ENABLE:
951 		/* PEBS needs a quiescent period after being disabled (to write
952 		 * a record).  Disabling PEBS through VMX MSR swapping doesn't
953 		 * provide that period, so a CPU could write host's record into
954 		 * guest's memory.
955 		 */
956 		wrmsrl(MSR_IA32_PEBS_ENABLE, 0);
957 	}
958 
959 	i = vmx_find_msr_index(&m->guest, msr);
960 	if (!entry_only)
961 		j = vmx_find_msr_index(&m->host, msr);
962 
963 	if ((i < 0 && m->guest.nr == NR_LOADSTORE_MSRS) ||
964 		(j < 0 &&  m->host.nr == NR_LOADSTORE_MSRS)) {
965 		printk_once(KERN_WARNING "Not enough msr switch entries. "
966 				"Can't add msr %x\n", msr);
967 		return;
968 	}
969 	if (i < 0) {
970 		i = m->guest.nr++;
971 		vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->guest.nr);
972 	}
973 	m->guest.val[i].index = msr;
974 	m->guest.val[i].value = guest_val;
975 
976 	if (entry_only)
977 		return;
978 
979 	if (j < 0) {
980 		j = m->host.nr++;
981 		vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->host.nr);
982 	}
983 	m->host.val[j].index = msr;
984 	m->host.val[j].value = host_val;
985 }
986 
987 static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset)
988 {
989 	u64 guest_efer = vmx->vcpu.arch.efer;
990 	u64 ignore_bits = 0;
991 
992 	/* Shadow paging assumes NX to be available.  */
993 	if (!enable_ept)
994 		guest_efer |= EFER_NX;
995 
996 	/*
997 	 * LMA and LME handled by hardware; SCE meaningless outside long mode.
998 	 */
999 	ignore_bits |= EFER_SCE;
1000 #ifdef CONFIG_X86_64
1001 	ignore_bits |= EFER_LMA | EFER_LME;
1002 	/* SCE is meaningful only in long mode on Intel */
1003 	if (guest_efer & EFER_LMA)
1004 		ignore_bits &= ~(u64)EFER_SCE;
1005 #endif
1006 
1007 	/*
1008 	 * On EPT, we can't emulate NX, so we must switch EFER atomically.
1009 	 * On CPUs that support "load IA32_EFER", always switch EFER
1010 	 * atomically, since it's faster than switching it manually.
1011 	 */
1012 	if (cpu_has_load_ia32_efer() ||
1013 	    (enable_ept && ((vmx->vcpu.arch.efer ^ host_efer) & EFER_NX))) {
1014 		if (!(guest_efer & EFER_LMA))
1015 			guest_efer &= ~EFER_LME;
1016 		if (guest_efer != host_efer)
1017 			add_atomic_switch_msr(vmx, MSR_EFER,
1018 					      guest_efer, host_efer, false);
1019 		else
1020 			clear_atomic_switch_msr(vmx, MSR_EFER);
1021 		return false;
1022 	} else {
1023 		clear_atomic_switch_msr(vmx, MSR_EFER);
1024 
1025 		guest_efer &= ~ignore_bits;
1026 		guest_efer |= host_efer & ignore_bits;
1027 
1028 		vmx->guest_msrs[efer_offset].data = guest_efer;
1029 		vmx->guest_msrs[efer_offset].mask = ~ignore_bits;
1030 
1031 		return true;
1032 	}
1033 }
1034 
1035 #ifdef CONFIG_X86_32
1036 /*
1037  * On 32-bit kernels, VM exits still load the FS and GS bases from the
1038  * VMCS rather than the segment table.  KVM uses this helper to figure
1039  * out the current bases to poke them into the VMCS before entry.
1040  */
1041 static unsigned long segment_base(u16 selector)
1042 {
1043 	struct desc_struct *table;
1044 	unsigned long v;
1045 
1046 	if (!(selector & ~SEGMENT_RPL_MASK))
1047 		return 0;
1048 
1049 	table = get_current_gdt_ro();
1050 
1051 	if ((selector & SEGMENT_TI_MASK) == SEGMENT_LDT) {
1052 		u16 ldt_selector = kvm_read_ldt();
1053 
1054 		if (!(ldt_selector & ~SEGMENT_RPL_MASK))
1055 			return 0;
1056 
1057 		table = (struct desc_struct *)segment_base(ldt_selector);
1058 	}
1059 	v = get_desc_base(&table[selector >> 3]);
1060 	return v;
1061 }
1062 #endif
1063 
1064 static inline bool pt_can_write_msr(struct vcpu_vmx *vmx)
1065 {
1066 	return (pt_mode == PT_MODE_HOST_GUEST) &&
1067 	       !(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN);
1068 }
1069 
1070 static inline void pt_load_msr(struct pt_ctx *ctx, u32 addr_range)
1071 {
1072 	u32 i;
1073 
1074 	wrmsrl(MSR_IA32_RTIT_STATUS, ctx->status);
1075 	wrmsrl(MSR_IA32_RTIT_OUTPUT_BASE, ctx->output_base);
1076 	wrmsrl(MSR_IA32_RTIT_OUTPUT_MASK, ctx->output_mask);
1077 	wrmsrl(MSR_IA32_RTIT_CR3_MATCH, ctx->cr3_match);
1078 	for (i = 0; i < addr_range; i++) {
1079 		wrmsrl(MSR_IA32_RTIT_ADDR0_A + i * 2, ctx->addr_a[i]);
1080 		wrmsrl(MSR_IA32_RTIT_ADDR0_B + i * 2, ctx->addr_b[i]);
1081 	}
1082 }
1083 
1084 static inline void pt_save_msr(struct pt_ctx *ctx, u32 addr_range)
1085 {
1086 	u32 i;
1087 
1088 	rdmsrl(MSR_IA32_RTIT_STATUS, ctx->status);
1089 	rdmsrl(MSR_IA32_RTIT_OUTPUT_BASE, ctx->output_base);
1090 	rdmsrl(MSR_IA32_RTIT_OUTPUT_MASK, ctx->output_mask);
1091 	rdmsrl(MSR_IA32_RTIT_CR3_MATCH, ctx->cr3_match);
1092 	for (i = 0; i < addr_range; i++) {
1093 		rdmsrl(MSR_IA32_RTIT_ADDR0_A + i * 2, ctx->addr_a[i]);
1094 		rdmsrl(MSR_IA32_RTIT_ADDR0_B + i * 2, ctx->addr_b[i]);
1095 	}
1096 }
1097 
1098 static void pt_guest_enter(struct vcpu_vmx *vmx)
1099 {
1100 	if (pt_mode == PT_MODE_SYSTEM)
1101 		return;
1102 
1103 	/*
1104 	 * GUEST_IA32_RTIT_CTL is already set in the VMCS.
1105 	 * Save host state before VM entry.
1106 	 */
1107 	rdmsrl(MSR_IA32_RTIT_CTL, vmx->pt_desc.host.ctl);
1108 	if (vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) {
1109 		wrmsrl(MSR_IA32_RTIT_CTL, 0);
1110 		pt_save_msr(&vmx->pt_desc.host, vmx->pt_desc.addr_range);
1111 		pt_load_msr(&vmx->pt_desc.guest, vmx->pt_desc.addr_range);
1112 	}
1113 }
1114 
1115 static void pt_guest_exit(struct vcpu_vmx *vmx)
1116 {
1117 	if (pt_mode == PT_MODE_SYSTEM)
1118 		return;
1119 
1120 	if (vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) {
1121 		pt_save_msr(&vmx->pt_desc.guest, vmx->pt_desc.addr_range);
1122 		pt_load_msr(&vmx->pt_desc.host, vmx->pt_desc.addr_range);
1123 	}
1124 
1125 	/* Reload host state (IA32_RTIT_CTL will be cleared on VM exit). */
1126 	wrmsrl(MSR_IA32_RTIT_CTL, vmx->pt_desc.host.ctl);
1127 }
1128 
1129 void vmx_set_host_fs_gs(struct vmcs_host_state *host, u16 fs_sel, u16 gs_sel,
1130 			unsigned long fs_base, unsigned long gs_base)
1131 {
1132 	if (unlikely(fs_sel != host->fs_sel)) {
1133 		if (!(fs_sel & 7))
1134 			vmcs_write16(HOST_FS_SELECTOR, fs_sel);
1135 		else
1136 			vmcs_write16(HOST_FS_SELECTOR, 0);
1137 		host->fs_sel = fs_sel;
1138 	}
1139 	if (unlikely(gs_sel != host->gs_sel)) {
1140 		if (!(gs_sel & 7))
1141 			vmcs_write16(HOST_GS_SELECTOR, gs_sel);
1142 		else
1143 			vmcs_write16(HOST_GS_SELECTOR, 0);
1144 		host->gs_sel = gs_sel;
1145 	}
1146 	if (unlikely(fs_base != host->fs_base)) {
1147 		vmcs_writel(HOST_FS_BASE, fs_base);
1148 		host->fs_base = fs_base;
1149 	}
1150 	if (unlikely(gs_base != host->gs_base)) {
1151 		vmcs_writel(HOST_GS_BASE, gs_base);
1152 		host->gs_base = gs_base;
1153 	}
1154 }
1155 
1156 void vmx_prepare_switch_to_guest(struct kvm_vcpu *vcpu)
1157 {
1158 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1159 	struct vmcs_host_state *host_state;
1160 #ifdef CONFIG_X86_64
1161 	int cpu = raw_smp_processor_id();
1162 #endif
1163 	unsigned long fs_base, gs_base;
1164 	u16 fs_sel, gs_sel;
1165 	int i;
1166 
1167 	vmx->req_immediate_exit = false;
1168 
1169 	/*
1170 	 * Note that guest MSRs to be saved/restored can also be changed
1171 	 * when guest state is loaded. This happens when guest transitions
1172 	 * to/from long-mode by setting MSR_EFER.LMA.
1173 	 */
1174 	if (!vmx->guest_msrs_ready) {
1175 		vmx->guest_msrs_ready = true;
1176 		for (i = 0; i < vmx->save_nmsrs; ++i)
1177 			kvm_set_shared_msr(vmx->guest_msrs[i].index,
1178 					   vmx->guest_msrs[i].data,
1179 					   vmx->guest_msrs[i].mask);
1180 
1181 	}
1182 
1183     	if (vmx->nested.need_vmcs12_to_shadow_sync)
1184 		nested_sync_vmcs12_to_shadow(vcpu);
1185 
1186 	if (vmx->guest_state_loaded)
1187 		return;
1188 
1189 	host_state = &vmx->loaded_vmcs->host_state;
1190 
1191 	/*
1192 	 * Set host fs and gs selectors.  Unfortunately, 22.2.3 does not
1193 	 * allow segment selectors with cpl > 0 or ti == 1.
1194 	 */
1195 	host_state->ldt_sel = kvm_read_ldt();
1196 
1197 #ifdef CONFIG_X86_64
1198 	savesegment(ds, host_state->ds_sel);
1199 	savesegment(es, host_state->es_sel);
1200 
1201 	gs_base = cpu_kernelmode_gs_base(cpu);
1202 	if (likely(is_64bit_mm(current->mm))) {
1203 		save_fsgs_for_kvm();
1204 		fs_sel = current->thread.fsindex;
1205 		gs_sel = current->thread.gsindex;
1206 		fs_base = current->thread.fsbase;
1207 		vmx->msr_host_kernel_gs_base = current->thread.gsbase;
1208 	} else {
1209 		savesegment(fs, fs_sel);
1210 		savesegment(gs, gs_sel);
1211 		fs_base = read_msr(MSR_FS_BASE);
1212 		vmx->msr_host_kernel_gs_base = read_msr(MSR_KERNEL_GS_BASE);
1213 	}
1214 
1215 	wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1216 #else
1217 	savesegment(fs, fs_sel);
1218 	savesegment(gs, gs_sel);
1219 	fs_base = segment_base(fs_sel);
1220 	gs_base = segment_base(gs_sel);
1221 #endif
1222 
1223 	vmx_set_host_fs_gs(host_state, fs_sel, gs_sel, fs_base, gs_base);
1224 	vmx->guest_state_loaded = true;
1225 }
1226 
1227 static void vmx_prepare_switch_to_host(struct vcpu_vmx *vmx)
1228 {
1229 	struct vmcs_host_state *host_state;
1230 
1231 	if (!vmx->guest_state_loaded)
1232 		return;
1233 
1234 	host_state = &vmx->loaded_vmcs->host_state;
1235 
1236 	++vmx->vcpu.stat.host_state_reload;
1237 
1238 #ifdef CONFIG_X86_64
1239 	rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1240 #endif
1241 	if (host_state->ldt_sel || (host_state->gs_sel & 7)) {
1242 		kvm_load_ldt(host_state->ldt_sel);
1243 #ifdef CONFIG_X86_64
1244 		load_gs_index(host_state->gs_sel);
1245 #else
1246 		loadsegment(gs, host_state->gs_sel);
1247 #endif
1248 	}
1249 	if (host_state->fs_sel & 7)
1250 		loadsegment(fs, host_state->fs_sel);
1251 #ifdef CONFIG_X86_64
1252 	if (unlikely(host_state->ds_sel | host_state->es_sel)) {
1253 		loadsegment(ds, host_state->ds_sel);
1254 		loadsegment(es, host_state->es_sel);
1255 	}
1256 #endif
1257 	invalidate_tss_limit();
1258 #ifdef CONFIG_X86_64
1259 	wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
1260 #endif
1261 	load_fixmap_gdt(raw_smp_processor_id());
1262 	vmx->guest_state_loaded = false;
1263 	vmx->guest_msrs_ready = false;
1264 }
1265 
1266 #ifdef CONFIG_X86_64
1267 static u64 vmx_read_guest_kernel_gs_base(struct vcpu_vmx *vmx)
1268 {
1269 	preempt_disable();
1270 	if (vmx->guest_state_loaded)
1271 		rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1272 	preempt_enable();
1273 	return vmx->msr_guest_kernel_gs_base;
1274 }
1275 
1276 static void vmx_write_guest_kernel_gs_base(struct vcpu_vmx *vmx, u64 data)
1277 {
1278 	preempt_disable();
1279 	if (vmx->guest_state_loaded)
1280 		wrmsrl(MSR_KERNEL_GS_BASE, data);
1281 	preempt_enable();
1282 	vmx->msr_guest_kernel_gs_base = data;
1283 }
1284 #endif
1285 
1286 static void vmx_vcpu_pi_load(struct kvm_vcpu *vcpu, int cpu)
1287 {
1288 	struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
1289 	struct pi_desc old, new;
1290 	unsigned int dest;
1291 
1292 	/*
1293 	 * In case of hot-plug or hot-unplug, we may have to undo
1294 	 * vmx_vcpu_pi_put even if there is no assigned device.  And we
1295 	 * always keep PI.NDST up to date for simplicity: it makes the
1296 	 * code easier, and CPU migration is not a fast path.
1297 	 */
1298 	if (!pi_test_sn(pi_desc) && vcpu->cpu == cpu)
1299 		return;
1300 
1301 	/*
1302 	 * If the 'nv' field is POSTED_INTR_WAKEUP_VECTOR, do not change
1303 	 * PI.NDST: pi_post_block is the one expected to change PID.NDST and the
1304 	 * wakeup handler expects the vCPU to be on the blocked_vcpu_list that
1305 	 * matches PI.NDST. Otherwise, a vcpu may not be able to be woken up
1306 	 * correctly.
1307 	 */
1308 	if (pi_desc->nv == POSTED_INTR_WAKEUP_VECTOR || vcpu->cpu == cpu) {
1309 		pi_clear_sn(pi_desc);
1310 		goto after_clear_sn;
1311 	}
1312 
1313 	/* The full case.  */
1314 	do {
1315 		old.control = new.control = pi_desc->control;
1316 
1317 		dest = cpu_physical_id(cpu);
1318 
1319 		if (x2apic_enabled())
1320 			new.ndst = dest;
1321 		else
1322 			new.ndst = (dest << 8) & 0xFF00;
1323 
1324 		new.sn = 0;
1325 	} while (cmpxchg64(&pi_desc->control, old.control,
1326 			   new.control) != old.control);
1327 
1328 after_clear_sn:
1329 
1330 	/*
1331 	 * Clear SN before reading the bitmap.  The VT-d firmware
1332 	 * writes the bitmap and reads SN atomically (5.2.3 in the
1333 	 * spec), so it doesn't really have a memory barrier that
1334 	 * pairs with this, but we cannot do that and we need one.
1335 	 */
1336 	smp_mb__after_atomic();
1337 
1338 	if (!pi_is_pir_empty(pi_desc))
1339 		pi_set_on(pi_desc);
1340 }
1341 
1342 void vmx_vcpu_load_vmcs(struct kvm_vcpu *vcpu, int cpu)
1343 {
1344 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1345 	bool already_loaded = vmx->loaded_vmcs->cpu == cpu;
1346 
1347 	if (!already_loaded) {
1348 		loaded_vmcs_clear(vmx->loaded_vmcs);
1349 		local_irq_disable();
1350 		crash_disable_local_vmclear(cpu);
1351 
1352 		/*
1353 		 * Read loaded_vmcs->cpu should be before fetching
1354 		 * loaded_vmcs->loaded_vmcss_on_cpu_link.
1355 		 * See the comments in __loaded_vmcs_clear().
1356 		 */
1357 		smp_rmb();
1358 
1359 		list_add(&vmx->loaded_vmcs->loaded_vmcss_on_cpu_link,
1360 			 &per_cpu(loaded_vmcss_on_cpu, cpu));
1361 		crash_enable_local_vmclear(cpu);
1362 		local_irq_enable();
1363 	}
1364 
1365 	if (per_cpu(current_vmcs, cpu) != vmx->loaded_vmcs->vmcs) {
1366 		per_cpu(current_vmcs, cpu) = vmx->loaded_vmcs->vmcs;
1367 		vmcs_load(vmx->loaded_vmcs->vmcs);
1368 		indirect_branch_prediction_barrier();
1369 	}
1370 
1371 	if (!already_loaded) {
1372 		void *gdt = get_current_gdt_ro();
1373 		unsigned long sysenter_esp;
1374 
1375 		kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1376 
1377 		/*
1378 		 * Linux uses per-cpu TSS and GDT, so set these when switching
1379 		 * processors.  See 22.2.4.
1380 		 */
1381 		vmcs_writel(HOST_TR_BASE,
1382 			    (unsigned long)&get_cpu_entry_area(cpu)->tss.x86_tss);
1383 		vmcs_writel(HOST_GDTR_BASE, (unsigned long)gdt);   /* 22.2.4 */
1384 
1385 		rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
1386 		vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
1387 
1388 		vmx->loaded_vmcs->cpu = cpu;
1389 	}
1390 
1391 	/* Setup TSC multiplier */
1392 	if (kvm_has_tsc_control &&
1393 	    vmx->current_tsc_ratio != vcpu->arch.tsc_scaling_ratio)
1394 		decache_tsc_multiplier(vmx);
1395 }
1396 
1397 /*
1398  * Switches to specified vcpu, until a matching vcpu_put(), but assumes
1399  * vcpu mutex is already taken.
1400  */
1401 void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1402 {
1403 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1404 
1405 	vmx_vcpu_load_vmcs(vcpu, cpu);
1406 
1407 	vmx_vcpu_pi_load(vcpu, cpu);
1408 
1409 	vmx->host_pkru = read_pkru();
1410 	vmx->host_debugctlmsr = get_debugctlmsr();
1411 }
1412 
1413 static void vmx_vcpu_pi_put(struct kvm_vcpu *vcpu)
1414 {
1415 	struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
1416 
1417 	if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
1418 		!irq_remapping_cap(IRQ_POSTING_CAP)  ||
1419 		!kvm_vcpu_apicv_active(vcpu))
1420 		return;
1421 
1422 	/* Set SN when the vCPU is preempted */
1423 	if (vcpu->preempted)
1424 		pi_set_sn(pi_desc);
1425 }
1426 
1427 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
1428 {
1429 	vmx_vcpu_pi_put(vcpu);
1430 
1431 	vmx_prepare_switch_to_host(to_vmx(vcpu));
1432 }
1433 
1434 static bool emulation_required(struct kvm_vcpu *vcpu)
1435 {
1436 	return emulate_invalid_guest_state && !guest_state_valid(vcpu);
1437 }
1438 
1439 unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
1440 {
1441 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1442 	unsigned long rflags, save_rflags;
1443 
1444 	if (!kvm_register_is_available(vcpu, VCPU_EXREG_RFLAGS)) {
1445 		kvm_register_mark_available(vcpu, VCPU_EXREG_RFLAGS);
1446 		rflags = vmcs_readl(GUEST_RFLAGS);
1447 		if (vmx->rmode.vm86_active) {
1448 			rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
1449 			save_rflags = vmx->rmode.save_rflags;
1450 			rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
1451 		}
1452 		vmx->rflags = rflags;
1453 	}
1454 	return vmx->rflags;
1455 }
1456 
1457 void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
1458 {
1459 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1460 	unsigned long old_rflags;
1461 
1462 	if (enable_unrestricted_guest) {
1463 		kvm_register_mark_available(vcpu, VCPU_EXREG_RFLAGS);
1464 		vmx->rflags = rflags;
1465 		vmcs_writel(GUEST_RFLAGS, rflags);
1466 		return;
1467 	}
1468 
1469 	old_rflags = vmx_get_rflags(vcpu);
1470 	vmx->rflags = rflags;
1471 	if (vmx->rmode.vm86_active) {
1472 		vmx->rmode.save_rflags = rflags;
1473 		rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
1474 	}
1475 	vmcs_writel(GUEST_RFLAGS, rflags);
1476 
1477 	if ((old_rflags ^ vmx->rflags) & X86_EFLAGS_VM)
1478 		vmx->emulation_required = emulation_required(vcpu);
1479 }
1480 
1481 u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu)
1482 {
1483 	u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1484 	int ret = 0;
1485 
1486 	if (interruptibility & GUEST_INTR_STATE_STI)
1487 		ret |= KVM_X86_SHADOW_INT_STI;
1488 	if (interruptibility & GUEST_INTR_STATE_MOV_SS)
1489 		ret |= KVM_X86_SHADOW_INT_MOV_SS;
1490 
1491 	return ret;
1492 }
1493 
1494 void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
1495 {
1496 	u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1497 	u32 interruptibility = interruptibility_old;
1498 
1499 	interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
1500 
1501 	if (mask & KVM_X86_SHADOW_INT_MOV_SS)
1502 		interruptibility |= GUEST_INTR_STATE_MOV_SS;
1503 	else if (mask & KVM_X86_SHADOW_INT_STI)
1504 		interruptibility |= GUEST_INTR_STATE_STI;
1505 
1506 	if ((interruptibility != interruptibility_old))
1507 		vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
1508 }
1509 
1510 static int vmx_rtit_ctl_check(struct kvm_vcpu *vcpu, u64 data)
1511 {
1512 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1513 	unsigned long value;
1514 
1515 	/*
1516 	 * Any MSR write that attempts to change bits marked reserved will
1517 	 * case a #GP fault.
1518 	 */
1519 	if (data & vmx->pt_desc.ctl_bitmask)
1520 		return 1;
1521 
1522 	/*
1523 	 * Any attempt to modify IA32_RTIT_CTL while TraceEn is set will
1524 	 * result in a #GP unless the same write also clears TraceEn.
1525 	 */
1526 	if ((vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) &&
1527 		((vmx->pt_desc.guest.ctl ^ data) & ~RTIT_CTL_TRACEEN))
1528 		return 1;
1529 
1530 	/*
1531 	 * WRMSR to IA32_RTIT_CTL that sets TraceEn but clears this bit
1532 	 * and FabricEn would cause #GP, if
1533 	 * CPUID.(EAX=14H, ECX=0):ECX.SNGLRGNOUT[bit 2] = 0
1534 	 */
1535 	if ((data & RTIT_CTL_TRACEEN) && !(data & RTIT_CTL_TOPA) &&
1536 		!(data & RTIT_CTL_FABRIC_EN) &&
1537 		!intel_pt_validate_cap(vmx->pt_desc.caps,
1538 					PT_CAP_single_range_output))
1539 		return 1;
1540 
1541 	/*
1542 	 * MTCFreq, CycThresh and PSBFreq encodings check, any MSR write that
1543 	 * utilize encodings marked reserved will casue a #GP fault.
1544 	 */
1545 	value = intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc_periods);
1546 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc) &&
1547 			!test_bit((data & RTIT_CTL_MTC_RANGE) >>
1548 			RTIT_CTL_MTC_RANGE_OFFSET, &value))
1549 		return 1;
1550 	value = intel_pt_validate_cap(vmx->pt_desc.caps,
1551 						PT_CAP_cycle_thresholds);
1552 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc) &&
1553 			!test_bit((data & RTIT_CTL_CYC_THRESH) >>
1554 			RTIT_CTL_CYC_THRESH_OFFSET, &value))
1555 		return 1;
1556 	value = intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_periods);
1557 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc) &&
1558 			!test_bit((data & RTIT_CTL_PSB_FREQ) >>
1559 			RTIT_CTL_PSB_FREQ_OFFSET, &value))
1560 		return 1;
1561 
1562 	/*
1563 	 * If ADDRx_CFG is reserved or the encodings is >2 will
1564 	 * cause a #GP fault.
1565 	 */
1566 	value = (data & RTIT_CTL_ADDR0) >> RTIT_CTL_ADDR0_OFFSET;
1567 	if ((value && (vmx->pt_desc.addr_range < 1)) || (value > 2))
1568 		return 1;
1569 	value = (data & RTIT_CTL_ADDR1) >> RTIT_CTL_ADDR1_OFFSET;
1570 	if ((value && (vmx->pt_desc.addr_range < 2)) || (value > 2))
1571 		return 1;
1572 	value = (data & RTIT_CTL_ADDR2) >> RTIT_CTL_ADDR2_OFFSET;
1573 	if ((value && (vmx->pt_desc.addr_range < 3)) || (value > 2))
1574 		return 1;
1575 	value = (data & RTIT_CTL_ADDR3) >> RTIT_CTL_ADDR3_OFFSET;
1576 	if ((value && (vmx->pt_desc.addr_range < 4)) || (value > 2))
1577 		return 1;
1578 
1579 	return 0;
1580 }
1581 
1582 static int skip_emulated_instruction(struct kvm_vcpu *vcpu)
1583 {
1584 	unsigned long rip;
1585 
1586 	/*
1587 	 * Using VMCS.VM_EXIT_INSTRUCTION_LEN on EPT misconfig depends on
1588 	 * undefined behavior: Intel's SDM doesn't mandate the VMCS field be
1589 	 * set when EPT misconfig occurs.  In practice, real hardware updates
1590 	 * VM_EXIT_INSTRUCTION_LEN on EPT misconfig, but other hypervisors
1591 	 * (namely Hyper-V) don't set it due to it being undefined behavior,
1592 	 * i.e. we end up advancing IP with some random value.
1593 	 */
1594 	if (!static_cpu_has(X86_FEATURE_HYPERVISOR) ||
1595 	    to_vmx(vcpu)->exit_reason != EXIT_REASON_EPT_MISCONFIG) {
1596 		rip = kvm_rip_read(vcpu);
1597 		rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
1598 		kvm_rip_write(vcpu, rip);
1599 	} else {
1600 		if (!kvm_emulate_instruction(vcpu, EMULTYPE_SKIP))
1601 			return 0;
1602 	}
1603 
1604 	/* skipping an emulated instruction also counts */
1605 	vmx_set_interrupt_shadow(vcpu, 0);
1606 
1607 	return 1;
1608 }
1609 
1610 
1611 /*
1612  * Recognizes a pending MTF VM-exit and records the nested state for later
1613  * delivery.
1614  */
1615 static void vmx_update_emulated_instruction(struct kvm_vcpu *vcpu)
1616 {
1617 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1618 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1619 
1620 	if (!is_guest_mode(vcpu))
1621 		return;
1622 
1623 	/*
1624 	 * Per the SDM, MTF takes priority over debug-trap exceptions besides
1625 	 * T-bit traps. As instruction emulation is completed (i.e. at the
1626 	 * instruction boundary), any #DB exception pending delivery must be a
1627 	 * debug-trap. Record the pending MTF state to be delivered in
1628 	 * vmx_check_nested_events().
1629 	 */
1630 	if (nested_cpu_has_mtf(vmcs12) &&
1631 	    (!vcpu->arch.exception.pending ||
1632 	     vcpu->arch.exception.nr == DB_VECTOR))
1633 		vmx->nested.mtf_pending = true;
1634 	else
1635 		vmx->nested.mtf_pending = false;
1636 }
1637 
1638 static int vmx_skip_emulated_instruction(struct kvm_vcpu *vcpu)
1639 {
1640 	vmx_update_emulated_instruction(vcpu);
1641 	return skip_emulated_instruction(vcpu);
1642 }
1643 
1644 static void vmx_clear_hlt(struct kvm_vcpu *vcpu)
1645 {
1646 	/*
1647 	 * Ensure that we clear the HLT state in the VMCS.  We don't need to
1648 	 * explicitly skip the instruction because if the HLT state is set,
1649 	 * then the instruction is already executing and RIP has already been
1650 	 * advanced.
1651 	 */
1652 	if (kvm_hlt_in_guest(vcpu->kvm) &&
1653 			vmcs_read32(GUEST_ACTIVITY_STATE) == GUEST_ACTIVITY_HLT)
1654 		vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
1655 }
1656 
1657 static void vmx_queue_exception(struct kvm_vcpu *vcpu)
1658 {
1659 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1660 	unsigned nr = vcpu->arch.exception.nr;
1661 	bool has_error_code = vcpu->arch.exception.has_error_code;
1662 	u32 error_code = vcpu->arch.exception.error_code;
1663 	u32 intr_info = nr | INTR_INFO_VALID_MASK;
1664 
1665 	kvm_deliver_exception_payload(vcpu);
1666 
1667 	if (has_error_code) {
1668 		vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
1669 		intr_info |= INTR_INFO_DELIVER_CODE_MASK;
1670 	}
1671 
1672 	if (vmx->rmode.vm86_active) {
1673 		int inc_eip = 0;
1674 		if (kvm_exception_is_soft(nr))
1675 			inc_eip = vcpu->arch.event_exit_inst_len;
1676 		kvm_inject_realmode_interrupt(vcpu, nr, inc_eip);
1677 		return;
1678 	}
1679 
1680 	WARN_ON_ONCE(vmx->emulation_required);
1681 
1682 	if (kvm_exception_is_soft(nr)) {
1683 		vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
1684 			     vmx->vcpu.arch.event_exit_inst_len);
1685 		intr_info |= INTR_TYPE_SOFT_EXCEPTION;
1686 	} else
1687 		intr_info |= INTR_TYPE_HARD_EXCEPTION;
1688 
1689 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
1690 
1691 	vmx_clear_hlt(vcpu);
1692 }
1693 
1694 static bool vmx_rdtscp_supported(void)
1695 {
1696 	return cpu_has_vmx_rdtscp();
1697 }
1698 
1699 static bool vmx_invpcid_supported(void)
1700 {
1701 	return cpu_has_vmx_invpcid();
1702 }
1703 
1704 /*
1705  * Swap MSR entry in host/guest MSR entry array.
1706  */
1707 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
1708 {
1709 	struct shared_msr_entry tmp;
1710 
1711 	tmp = vmx->guest_msrs[to];
1712 	vmx->guest_msrs[to] = vmx->guest_msrs[from];
1713 	vmx->guest_msrs[from] = tmp;
1714 }
1715 
1716 /*
1717  * Set up the vmcs to automatically save and restore system
1718  * msrs.  Don't touch the 64-bit msrs if the guest is in legacy
1719  * mode, as fiddling with msrs is very expensive.
1720  */
1721 static void setup_msrs(struct vcpu_vmx *vmx)
1722 {
1723 	int save_nmsrs, index;
1724 
1725 	save_nmsrs = 0;
1726 #ifdef CONFIG_X86_64
1727 	/*
1728 	 * The SYSCALL MSRs are only needed on long mode guests, and only
1729 	 * when EFER.SCE is set.
1730 	 */
1731 	if (is_long_mode(&vmx->vcpu) && (vmx->vcpu.arch.efer & EFER_SCE)) {
1732 		index = __find_msr_index(vmx, MSR_STAR);
1733 		if (index >= 0)
1734 			move_msr_up(vmx, index, save_nmsrs++);
1735 		index = __find_msr_index(vmx, MSR_LSTAR);
1736 		if (index >= 0)
1737 			move_msr_up(vmx, index, save_nmsrs++);
1738 		index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
1739 		if (index >= 0)
1740 			move_msr_up(vmx, index, save_nmsrs++);
1741 	}
1742 #endif
1743 	index = __find_msr_index(vmx, MSR_EFER);
1744 	if (index >= 0 && update_transition_efer(vmx, index))
1745 		move_msr_up(vmx, index, save_nmsrs++);
1746 	index = __find_msr_index(vmx, MSR_TSC_AUX);
1747 	if (index >= 0 && guest_cpuid_has(&vmx->vcpu, X86_FEATURE_RDTSCP))
1748 		move_msr_up(vmx, index, save_nmsrs++);
1749 	index = __find_msr_index(vmx, MSR_IA32_TSX_CTRL);
1750 	if (index >= 0)
1751 		move_msr_up(vmx, index, save_nmsrs++);
1752 
1753 	vmx->save_nmsrs = save_nmsrs;
1754 	vmx->guest_msrs_ready = false;
1755 
1756 	if (cpu_has_vmx_msr_bitmap())
1757 		vmx_update_msr_bitmap(&vmx->vcpu);
1758 }
1759 
1760 static u64 vmx_read_l1_tsc_offset(struct kvm_vcpu *vcpu)
1761 {
1762 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1763 
1764 	if (is_guest_mode(vcpu) &&
1765 	    (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETTING))
1766 		return vcpu->arch.tsc_offset - vmcs12->tsc_offset;
1767 
1768 	return vcpu->arch.tsc_offset;
1769 }
1770 
1771 static u64 vmx_write_l1_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1772 {
1773 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1774 	u64 g_tsc_offset = 0;
1775 
1776 	/*
1777 	 * We're here if L1 chose not to trap WRMSR to TSC. According
1778 	 * to the spec, this should set L1's TSC; The offset that L1
1779 	 * set for L2 remains unchanged, and still needs to be added
1780 	 * to the newly set TSC to get L2's TSC.
1781 	 */
1782 	if (is_guest_mode(vcpu) &&
1783 	    (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETTING))
1784 		g_tsc_offset = vmcs12->tsc_offset;
1785 
1786 	trace_kvm_write_tsc_offset(vcpu->vcpu_id,
1787 				   vcpu->arch.tsc_offset - g_tsc_offset,
1788 				   offset);
1789 	vmcs_write64(TSC_OFFSET, offset + g_tsc_offset);
1790 	return offset + g_tsc_offset;
1791 }
1792 
1793 /*
1794  * nested_vmx_allowed() checks whether a guest should be allowed to use VMX
1795  * instructions and MSRs (i.e., nested VMX). Nested VMX is disabled for
1796  * all guests if the "nested" module option is off, and can also be disabled
1797  * for a single guest by disabling its VMX cpuid bit.
1798  */
1799 bool nested_vmx_allowed(struct kvm_vcpu *vcpu)
1800 {
1801 	return nested && guest_cpuid_has(vcpu, X86_FEATURE_VMX);
1802 }
1803 
1804 static inline bool vmx_feature_control_msr_valid(struct kvm_vcpu *vcpu,
1805 						 uint64_t val)
1806 {
1807 	uint64_t valid_bits = to_vmx(vcpu)->msr_ia32_feature_control_valid_bits;
1808 
1809 	return !(val & ~valid_bits);
1810 }
1811 
1812 static int vmx_get_msr_feature(struct kvm_msr_entry *msr)
1813 {
1814 	switch (msr->index) {
1815 	case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
1816 		if (!nested)
1817 			return 1;
1818 		return vmx_get_vmx_msr(&vmcs_config.nested, msr->index, &msr->data);
1819 	default:
1820 		return 1;
1821 	}
1822 }
1823 
1824 /*
1825  * Reads an msr value (of 'msr_index') into 'pdata'.
1826  * Returns 0 on success, non-0 otherwise.
1827  * Assumes vcpu_load() was already called.
1828  */
1829 static int vmx_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1830 {
1831 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1832 	struct shared_msr_entry *msr;
1833 	u32 index;
1834 
1835 	switch (msr_info->index) {
1836 #ifdef CONFIG_X86_64
1837 	case MSR_FS_BASE:
1838 		msr_info->data = vmcs_readl(GUEST_FS_BASE);
1839 		break;
1840 	case MSR_GS_BASE:
1841 		msr_info->data = vmcs_readl(GUEST_GS_BASE);
1842 		break;
1843 	case MSR_KERNEL_GS_BASE:
1844 		msr_info->data = vmx_read_guest_kernel_gs_base(vmx);
1845 		break;
1846 #endif
1847 	case MSR_EFER:
1848 		return kvm_get_msr_common(vcpu, msr_info);
1849 	case MSR_IA32_TSX_CTRL:
1850 		if (!msr_info->host_initiated &&
1851 		    !(vcpu->arch.arch_capabilities & ARCH_CAP_TSX_CTRL_MSR))
1852 			return 1;
1853 		goto find_shared_msr;
1854 	case MSR_IA32_UMWAIT_CONTROL:
1855 		if (!msr_info->host_initiated && !vmx_has_waitpkg(vmx))
1856 			return 1;
1857 
1858 		msr_info->data = vmx->msr_ia32_umwait_control;
1859 		break;
1860 	case MSR_IA32_SPEC_CTRL:
1861 		if (!msr_info->host_initiated &&
1862 		    !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
1863 			return 1;
1864 
1865 		msr_info->data = to_vmx(vcpu)->spec_ctrl;
1866 		break;
1867 	case MSR_IA32_SYSENTER_CS:
1868 		msr_info->data = vmcs_read32(GUEST_SYSENTER_CS);
1869 		break;
1870 	case MSR_IA32_SYSENTER_EIP:
1871 		msr_info->data = vmcs_readl(GUEST_SYSENTER_EIP);
1872 		break;
1873 	case MSR_IA32_SYSENTER_ESP:
1874 		msr_info->data = vmcs_readl(GUEST_SYSENTER_ESP);
1875 		break;
1876 	case MSR_IA32_BNDCFGS:
1877 		if (!kvm_mpx_supported() ||
1878 		    (!msr_info->host_initiated &&
1879 		     !guest_cpuid_has(vcpu, X86_FEATURE_MPX)))
1880 			return 1;
1881 		msr_info->data = vmcs_read64(GUEST_BNDCFGS);
1882 		break;
1883 	case MSR_IA32_MCG_EXT_CTL:
1884 		if (!msr_info->host_initiated &&
1885 		    !(vmx->msr_ia32_feature_control &
1886 		      FEAT_CTL_LMCE_ENABLED))
1887 			return 1;
1888 		msr_info->data = vcpu->arch.mcg_ext_ctl;
1889 		break;
1890 	case MSR_IA32_FEAT_CTL:
1891 		msr_info->data = vmx->msr_ia32_feature_control;
1892 		break;
1893 	case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
1894 		if (!nested_vmx_allowed(vcpu))
1895 			return 1;
1896 		if (vmx_get_vmx_msr(&vmx->nested.msrs, msr_info->index,
1897 				    &msr_info->data))
1898 			return 1;
1899 		/*
1900 		 * Enlightened VMCS v1 doesn't have certain fields, but buggy
1901 		 * Hyper-V versions are still trying to use corresponding
1902 		 * features when they are exposed. Filter out the essential
1903 		 * minimum.
1904 		 */
1905 		if (!msr_info->host_initiated &&
1906 		    vmx->nested.enlightened_vmcs_enabled)
1907 			nested_evmcs_filter_control_msr(msr_info->index,
1908 							&msr_info->data);
1909 		break;
1910 	case MSR_IA32_RTIT_CTL:
1911 		if (pt_mode != PT_MODE_HOST_GUEST)
1912 			return 1;
1913 		msr_info->data = vmx->pt_desc.guest.ctl;
1914 		break;
1915 	case MSR_IA32_RTIT_STATUS:
1916 		if (pt_mode != PT_MODE_HOST_GUEST)
1917 			return 1;
1918 		msr_info->data = vmx->pt_desc.guest.status;
1919 		break;
1920 	case MSR_IA32_RTIT_CR3_MATCH:
1921 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
1922 			!intel_pt_validate_cap(vmx->pt_desc.caps,
1923 						PT_CAP_cr3_filtering))
1924 			return 1;
1925 		msr_info->data = vmx->pt_desc.guest.cr3_match;
1926 		break;
1927 	case MSR_IA32_RTIT_OUTPUT_BASE:
1928 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
1929 			(!intel_pt_validate_cap(vmx->pt_desc.caps,
1930 					PT_CAP_topa_output) &&
1931 			 !intel_pt_validate_cap(vmx->pt_desc.caps,
1932 					PT_CAP_single_range_output)))
1933 			return 1;
1934 		msr_info->data = vmx->pt_desc.guest.output_base;
1935 		break;
1936 	case MSR_IA32_RTIT_OUTPUT_MASK:
1937 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
1938 			(!intel_pt_validate_cap(vmx->pt_desc.caps,
1939 					PT_CAP_topa_output) &&
1940 			 !intel_pt_validate_cap(vmx->pt_desc.caps,
1941 					PT_CAP_single_range_output)))
1942 			return 1;
1943 		msr_info->data = vmx->pt_desc.guest.output_mask;
1944 		break;
1945 	case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
1946 		index = msr_info->index - MSR_IA32_RTIT_ADDR0_A;
1947 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
1948 			(index >= 2 * intel_pt_validate_cap(vmx->pt_desc.caps,
1949 					PT_CAP_num_address_ranges)))
1950 			return 1;
1951 		if (index % 2)
1952 			msr_info->data = vmx->pt_desc.guest.addr_b[index / 2];
1953 		else
1954 			msr_info->data = vmx->pt_desc.guest.addr_a[index / 2];
1955 		break;
1956 	case MSR_TSC_AUX:
1957 		if (!msr_info->host_initiated &&
1958 		    !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP))
1959 			return 1;
1960 		goto find_shared_msr;
1961 	default:
1962 	find_shared_msr:
1963 		msr = find_msr_entry(vmx, msr_info->index);
1964 		if (msr) {
1965 			msr_info->data = msr->data;
1966 			break;
1967 		}
1968 		return kvm_get_msr_common(vcpu, msr_info);
1969 	}
1970 
1971 	return 0;
1972 }
1973 
1974 /*
1975  * Writes msr value into the appropriate "register".
1976  * Returns 0 on success, non-0 otherwise.
1977  * Assumes vcpu_load() was already called.
1978  */
1979 static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1980 {
1981 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1982 	struct shared_msr_entry *msr;
1983 	int ret = 0;
1984 	u32 msr_index = msr_info->index;
1985 	u64 data = msr_info->data;
1986 	u32 index;
1987 
1988 	switch (msr_index) {
1989 	case MSR_EFER:
1990 		ret = kvm_set_msr_common(vcpu, msr_info);
1991 		break;
1992 #ifdef CONFIG_X86_64
1993 	case MSR_FS_BASE:
1994 		vmx_segment_cache_clear(vmx);
1995 		vmcs_writel(GUEST_FS_BASE, data);
1996 		break;
1997 	case MSR_GS_BASE:
1998 		vmx_segment_cache_clear(vmx);
1999 		vmcs_writel(GUEST_GS_BASE, data);
2000 		break;
2001 	case MSR_KERNEL_GS_BASE:
2002 		vmx_write_guest_kernel_gs_base(vmx, data);
2003 		break;
2004 #endif
2005 	case MSR_IA32_SYSENTER_CS:
2006 		if (is_guest_mode(vcpu))
2007 			get_vmcs12(vcpu)->guest_sysenter_cs = data;
2008 		vmcs_write32(GUEST_SYSENTER_CS, data);
2009 		break;
2010 	case MSR_IA32_SYSENTER_EIP:
2011 		if (is_guest_mode(vcpu))
2012 			get_vmcs12(vcpu)->guest_sysenter_eip = data;
2013 		vmcs_writel(GUEST_SYSENTER_EIP, data);
2014 		break;
2015 	case MSR_IA32_SYSENTER_ESP:
2016 		if (is_guest_mode(vcpu))
2017 			get_vmcs12(vcpu)->guest_sysenter_esp = data;
2018 		vmcs_writel(GUEST_SYSENTER_ESP, data);
2019 		break;
2020 	case MSR_IA32_DEBUGCTLMSR:
2021 		if (is_guest_mode(vcpu) && get_vmcs12(vcpu)->vm_exit_controls &
2022 						VM_EXIT_SAVE_DEBUG_CONTROLS)
2023 			get_vmcs12(vcpu)->guest_ia32_debugctl = data;
2024 
2025 		ret = kvm_set_msr_common(vcpu, msr_info);
2026 		break;
2027 
2028 	case MSR_IA32_BNDCFGS:
2029 		if (!kvm_mpx_supported() ||
2030 		    (!msr_info->host_initiated &&
2031 		     !guest_cpuid_has(vcpu, X86_FEATURE_MPX)))
2032 			return 1;
2033 		if (is_noncanonical_address(data & PAGE_MASK, vcpu) ||
2034 		    (data & MSR_IA32_BNDCFGS_RSVD))
2035 			return 1;
2036 		vmcs_write64(GUEST_BNDCFGS, data);
2037 		break;
2038 	case MSR_IA32_UMWAIT_CONTROL:
2039 		if (!msr_info->host_initiated && !vmx_has_waitpkg(vmx))
2040 			return 1;
2041 
2042 		/* The reserved bit 1 and non-32 bit [63:32] should be zero */
2043 		if (data & (BIT_ULL(1) | GENMASK_ULL(63, 32)))
2044 			return 1;
2045 
2046 		vmx->msr_ia32_umwait_control = data;
2047 		break;
2048 	case MSR_IA32_SPEC_CTRL:
2049 		if (!msr_info->host_initiated &&
2050 		    !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
2051 			return 1;
2052 
2053 		if (data & ~kvm_spec_ctrl_valid_bits(vcpu))
2054 			return 1;
2055 
2056 		vmx->spec_ctrl = data;
2057 		if (!data)
2058 			break;
2059 
2060 		/*
2061 		 * For non-nested:
2062 		 * When it's written (to non-zero) for the first time, pass
2063 		 * it through.
2064 		 *
2065 		 * For nested:
2066 		 * The handling of the MSR bitmap for L2 guests is done in
2067 		 * nested_vmx_prepare_msr_bitmap. We should not touch the
2068 		 * vmcs02.msr_bitmap here since it gets completely overwritten
2069 		 * in the merging. We update the vmcs01 here for L1 as well
2070 		 * since it will end up touching the MSR anyway now.
2071 		 */
2072 		vmx_disable_intercept_for_msr(vmx->vmcs01.msr_bitmap,
2073 					      MSR_IA32_SPEC_CTRL,
2074 					      MSR_TYPE_RW);
2075 		break;
2076 	case MSR_IA32_TSX_CTRL:
2077 		if (!msr_info->host_initiated &&
2078 		    !(vcpu->arch.arch_capabilities & ARCH_CAP_TSX_CTRL_MSR))
2079 			return 1;
2080 		if (data & ~(TSX_CTRL_RTM_DISABLE | TSX_CTRL_CPUID_CLEAR))
2081 			return 1;
2082 		goto find_shared_msr;
2083 	case MSR_IA32_PRED_CMD:
2084 		if (!msr_info->host_initiated &&
2085 		    !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
2086 			return 1;
2087 
2088 		if (data & ~PRED_CMD_IBPB)
2089 			return 1;
2090 		if (!boot_cpu_has(X86_FEATURE_SPEC_CTRL))
2091 			return 1;
2092 		if (!data)
2093 			break;
2094 
2095 		wrmsrl(MSR_IA32_PRED_CMD, PRED_CMD_IBPB);
2096 
2097 		/*
2098 		 * For non-nested:
2099 		 * When it's written (to non-zero) for the first time, pass
2100 		 * it through.
2101 		 *
2102 		 * For nested:
2103 		 * The handling of the MSR bitmap for L2 guests is done in
2104 		 * nested_vmx_prepare_msr_bitmap. We should not touch the
2105 		 * vmcs02.msr_bitmap here since it gets completely overwritten
2106 		 * in the merging.
2107 		 */
2108 		vmx_disable_intercept_for_msr(vmx->vmcs01.msr_bitmap, MSR_IA32_PRED_CMD,
2109 					      MSR_TYPE_W);
2110 		break;
2111 	case MSR_IA32_CR_PAT:
2112 		if (!kvm_pat_valid(data))
2113 			return 1;
2114 
2115 		if (is_guest_mode(vcpu) &&
2116 		    get_vmcs12(vcpu)->vm_exit_controls & VM_EXIT_SAVE_IA32_PAT)
2117 			get_vmcs12(vcpu)->guest_ia32_pat = data;
2118 
2119 		if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2120 			vmcs_write64(GUEST_IA32_PAT, data);
2121 			vcpu->arch.pat = data;
2122 			break;
2123 		}
2124 		ret = kvm_set_msr_common(vcpu, msr_info);
2125 		break;
2126 	case MSR_IA32_TSC_ADJUST:
2127 		ret = kvm_set_msr_common(vcpu, msr_info);
2128 		break;
2129 	case MSR_IA32_MCG_EXT_CTL:
2130 		if ((!msr_info->host_initiated &&
2131 		     !(to_vmx(vcpu)->msr_ia32_feature_control &
2132 		       FEAT_CTL_LMCE_ENABLED)) ||
2133 		    (data & ~MCG_EXT_CTL_LMCE_EN))
2134 			return 1;
2135 		vcpu->arch.mcg_ext_ctl = data;
2136 		break;
2137 	case MSR_IA32_FEAT_CTL:
2138 		if (!vmx_feature_control_msr_valid(vcpu, data) ||
2139 		    (to_vmx(vcpu)->msr_ia32_feature_control &
2140 		     FEAT_CTL_LOCKED && !msr_info->host_initiated))
2141 			return 1;
2142 		vmx->msr_ia32_feature_control = data;
2143 		if (msr_info->host_initiated && data == 0)
2144 			vmx_leave_nested(vcpu);
2145 		break;
2146 	case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
2147 		if (!msr_info->host_initiated)
2148 			return 1; /* they are read-only */
2149 		if (!nested_vmx_allowed(vcpu))
2150 			return 1;
2151 		return vmx_set_vmx_msr(vcpu, msr_index, data);
2152 	case MSR_IA32_RTIT_CTL:
2153 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
2154 			vmx_rtit_ctl_check(vcpu, data) ||
2155 			vmx->nested.vmxon)
2156 			return 1;
2157 		vmcs_write64(GUEST_IA32_RTIT_CTL, data);
2158 		vmx->pt_desc.guest.ctl = data;
2159 		pt_update_intercept_for_msr(vmx);
2160 		break;
2161 	case MSR_IA32_RTIT_STATUS:
2162 		if (!pt_can_write_msr(vmx))
2163 			return 1;
2164 		if (data & MSR_IA32_RTIT_STATUS_MASK)
2165 			return 1;
2166 		vmx->pt_desc.guest.status = data;
2167 		break;
2168 	case MSR_IA32_RTIT_CR3_MATCH:
2169 		if (!pt_can_write_msr(vmx))
2170 			return 1;
2171 		if (!intel_pt_validate_cap(vmx->pt_desc.caps,
2172 					   PT_CAP_cr3_filtering))
2173 			return 1;
2174 		vmx->pt_desc.guest.cr3_match = data;
2175 		break;
2176 	case MSR_IA32_RTIT_OUTPUT_BASE:
2177 		if (!pt_can_write_msr(vmx))
2178 			return 1;
2179 		if (!intel_pt_validate_cap(vmx->pt_desc.caps,
2180 					   PT_CAP_topa_output) &&
2181 		    !intel_pt_validate_cap(vmx->pt_desc.caps,
2182 					   PT_CAP_single_range_output))
2183 			return 1;
2184 		if (data & MSR_IA32_RTIT_OUTPUT_BASE_MASK)
2185 			return 1;
2186 		vmx->pt_desc.guest.output_base = data;
2187 		break;
2188 	case MSR_IA32_RTIT_OUTPUT_MASK:
2189 		if (!pt_can_write_msr(vmx))
2190 			return 1;
2191 		if (!intel_pt_validate_cap(vmx->pt_desc.caps,
2192 					   PT_CAP_topa_output) &&
2193 		    !intel_pt_validate_cap(vmx->pt_desc.caps,
2194 					   PT_CAP_single_range_output))
2195 			return 1;
2196 		vmx->pt_desc.guest.output_mask = data;
2197 		break;
2198 	case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
2199 		if (!pt_can_write_msr(vmx))
2200 			return 1;
2201 		index = msr_info->index - MSR_IA32_RTIT_ADDR0_A;
2202 		if (index >= 2 * intel_pt_validate_cap(vmx->pt_desc.caps,
2203 						       PT_CAP_num_address_ranges))
2204 			return 1;
2205 		if (is_noncanonical_address(data, vcpu))
2206 			return 1;
2207 		if (index % 2)
2208 			vmx->pt_desc.guest.addr_b[index / 2] = data;
2209 		else
2210 			vmx->pt_desc.guest.addr_a[index / 2] = data;
2211 		break;
2212 	case MSR_TSC_AUX:
2213 		if (!msr_info->host_initiated &&
2214 		    !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP))
2215 			return 1;
2216 		/* Check reserved bit, higher 32 bits should be zero */
2217 		if ((data >> 32) != 0)
2218 			return 1;
2219 		goto find_shared_msr;
2220 
2221 	default:
2222 	find_shared_msr:
2223 		msr = find_msr_entry(vmx, msr_index);
2224 		if (msr)
2225 			ret = vmx_set_guest_msr(vmx, msr, data);
2226 		else
2227 			ret = kvm_set_msr_common(vcpu, msr_info);
2228 	}
2229 
2230 	return ret;
2231 }
2232 
2233 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
2234 {
2235 	kvm_register_mark_available(vcpu, reg);
2236 
2237 	switch (reg) {
2238 	case VCPU_REGS_RSP:
2239 		vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
2240 		break;
2241 	case VCPU_REGS_RIP:
2242 		vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
2243 		break;
2244 	case VCPU_EXREG_PDPTR:
2245 		if (enable_ept)
2246 			ept_save_pdptrs(vcpu);
2247 		break;
2248 	case VCPU_EXREG_CR3:
2249 		if (enable_unrestricted_guest || (enable_ept && is_paging(vcpu)))
2250 			vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
2251 		break;
2252 	default:
2253 		WARN_ON_ONCE(1);
2254 		break;
2255 	}
2256 }
2257 
2258 static __init int cpu_has_kvm_support(void)
2259 {
2260 	return cpu_has_vmx();
2261 }
2262 
2263 static __init int vmx_disabled_by_bios(void)
2264 {
2265 	return !boot_cpu_has(X86_FEATURE_MSR_IA32_FEAT_CTL) ||
2266 	       !boot_cpu_has(X86_FEATURE_VMX);
2267 }
2268 
2269 static void kvm_cpu_vmxon(u64 addr)
2270 {
2271 	cr4_set_bits(X86_CR4_VMXE);
2272 	intel_pt_handle_vmx(1);
2273 
2274 	asm volatile ("vmxon %0" : : "m"(addr));
2275 }
2276 
2277 static int hardware_enable(void)
2278 {
2279 	int cpu = raw_smp_processor_id();
2280 	u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
2281 
2282 	if (cr4_read_shadow() & X86_CR4_VMXE)
2283 		return -EBUSY;
2284 
2285 	/*
2286 	 * This can happen if we hot-added a CPU but failed to allocate
2287 	 * VP assist page for it.
2288 	 */
2289 	if (static_branch_unlikely(&enable_evmcs) &&
2290 	    !hv_get_vp_assist_page(cpu))
2291 		return -EFAULT;
2292 
2293 	INIT_LIST_HEAD(&per_cpu(loaded_vmcss_on_cpu, cpu));
2294 	INIT_LIST_HEAD(&per_cpu(blocked_vcpu_on_cpu, cpu));
2295 	spin_lock_init(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
2296 
2297 	/*
2298 	 * Now we can enable the vmclear operation in kdump
2299 	 * since the loaded_vmcss_on_cpu list on this cpu
2300 	 * has been initialized.
2301 	 *
2302 	 * Though the cpu is not in VMX operation now, there
2303 	 * is no problem to enable the vmclear operation
2304 	 * for the loaded_vmcss_on_cpu list is empty!
2305 	 */
2306 	crash_enable_local_vmclear(cpu);
2307 
2308 	kvm_cpu_vmxon(phys_addr);
2309 	if (enable_ept)
2310 		ept_sync_global();
2311 
2312 	return 0;
2313 }
2314 
2315 static void vmclear_local_loaded_vmcss(void)
2316 {
2317 	int cpu = raw_smp_processor_id();
2318 	struct loaded_vmcs *v, *n;
2319 
2320 	list_for_each_entry_safe(v, n, &per_cpu(loaded_vmcss_on_cpu, cpu),
2321 				 loaded_vmcss_on_cpu_link)
2322 		__loaded_vmcs_clear(v);
2323 }
2324 
2325 
2326 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
2327  * tricks.
2328  */
2329 static void kvm_cpu_vmxoff(void)
2330 {
2331 	asm volatile (__ex("vmxoff"));
2332 
2333 	intel_pt_handle_vmx(0);
2334 	cr4_clear_bits(X86_CR4_VMXE);
2335 }
2336 
2337 static void hardware_disable(void)
2338 {
2339 	vmclear_local_loaded_vmcss();
2340 	kvm_cpu_vmxoff();
2341 }
2342 
2343 /*
2344  * There is no X86_FEATURE for SGX yet, but anyway we need to query CPUID
2345  * directly instead of going through cpu_has(), to ensure KVM is trapping
2346  * ENCLS whenever it's supported in hardware.  It does not matter whether
2347  * the host OS supports or has enabled SGX.
2348  */
2349 static bool cpu_has_sgx(void)
2350 {
2351 	return cpuid_eax(0) >= 0x12 && (cpuid_eax(0x12) & BIT(0));
2352 }
2353 
2354 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
2355 				      u32 msr, u32 *result)
2356 {
2357 	u32 vmx_msr_low, vmx_msr_high;
2358 	u32 ctl = ctl_min | ctl_opt;
2359 
2360 	rdmsr(msr, vmx_msr_low, vmx_msr_high);
2361 
2362 	ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
2363 	ctl |= vmx_msr_low;  /* bit == 1 in low word  ==> must be one  */
2364 
2365 	/* Ensure minimum (required) set of control bits are supported. */
2366 	if (ctl_min & ~ctl)
2367 		return -EIO;
2368 
2369 	*result = ctl;
2370 	return 0;
2371 }
2372 
2373 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf,
2374 				    struct vmx_capability *vmx_cap)
2375 {
2376 	u32 vmx_msr_low, vmx_msr_high;
2377 	u32 min, opt, min2, opt2;
2378 	u32 _pin_based_exec_control = 0;
2379 	u32 _cpu_based_exec_control = 0;
2380 	u32 _cpu_based_2nd_exec_control = 0;
2381 	u32 _vmexit_control = 0;
2382 	u32 _vmentry_control = 0;
2383 
2384 	memset(vmcs_conf, 0, sizeof(*vmcs_conf));
2385 	min = CPU_BASED_HLT_EXITING |
2386 #ifdef CONFIG_X86_64
2387 	      CPU_BASED_CR8_LOAD_EXITING |
2388 	      CPU_BASED_CR8_STORE_EXITING |
2389 #endif
2390 	      CPU_BASED_CR3_LOAD_EXITING |
2391 	      CPU_BASED_CR3_STORE_EXITING |
2392 	      CPU_BASED_UNCOND_IO_EXITING |
2393 	      CPU_BASED_MOV_DR_EXITING |
2394 	      CPU_BASED_USE_TSC_OFFSETTING |
2395 	      CPU_BASED_MWAIT_EXITING |
2396 	      CPU_BASED_MONITOR_EXITING |
2397 	      CPU_BASED_INVLPG_EXITING |
2398 	      CPU_BASED_RDPMC_EXITING;
2399 
2400 	opt = CPU_BASED_TPR_SHADOW |
2401 	      CPU_BASED_USE_MSR_BITMAPS |
2402 	      CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
2403 	if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
2404 				&_cpu_based_exec_control) < 0)
2405 		return -EIO;
2406 #ifdef CONFIG_X86_64
2407 	if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2408 		_cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
2409 					   ~CPU_BASED_CR8_STORE_EXITING;
2410 #endif
2411 	if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
2412 		min2 = 0;
2413 		opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2414 			SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2415 			SECONDARY_EXEC_WBINVD_EXITING |
2416 			SECONDARY_EXEC_ENABLE_VPID |
2417 			SECONDARY_EXEC_ENABLE_EPT |
2418 			SECONDARY_EXEC_UNRESTRICTED_GUEST |
2419 			SECONDARY_EXEC_PAUSE_LOOP_EXITING |
2420 			SECONDARY_EXEC_DESC |
2421 			SECONDARY_EXEC_RDTSCP |
2422 			SECONDARY_EXEC_ENABLE_INVPCID |
2423 			SECONDARY_EXEC_APIC_REGISTER_VIRT |
2424 			SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
2425 			SECONDARY_EXEC_SHADOW_VMCS |
2426 			SECONDARY_EXEC_XSAVES |
2427 			SECONDARY_EXEC_RDSEED_EXITING |
2428 			SECONDARY_EXEC_RDRAND_EXITING |
2429 			SECONDARY_EXEC_ENABLE_PML |
2430 			SECONDARY_EXEC_TSC_SCALING |
2431 			SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE |
2432 			SECONDARY_EXEC_PT_USE_GPA |
2433 			SECONDARY_EXEC_PT_CONCEAL_VMX |
2434 			SECONDARY_EXEC_ENABLE_VMFUNC;
2435 		if (cpu_has_sgx())
2436 			opt2 |= SECONDARY_EXEC_ENCLS_EXITING;
2437 		if (adjust_vmx_controls(min2, opt2,
2438 					MSR_IA32_VMX_PROCBASED_CTLS2,
2439 					&_cpu_based_2nd_exec_control) < 0)
2440 			return -EIO;
2441 	}
2442 #ifndef CONFIG_X86_64
2443 	if (!(_cpu_based_2nd_exec_control &
2444 				SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
2445 		_cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
2446 #endif
2447 
2448 	if (!(_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2449 		_cpu_based_2nd_exec_control &= ~(
2450 				SECONDARY_EXEC_APIC_REGISTER_VIRT |
2451 				SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2452 				SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
2453 
2454 	rdmsr_safe(MSR_IA32_VMX_EPT_VPID_CAP,
2455 		&vmx_cap->ept, &vmx_cap->vpid);
2456 
2457 	if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
2458 		/* CR3 accesses and invlpg don't need to cause VM Exits when EPT
2459 		   enabled */
2460 		_cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
2461 					     CPU_BASED_CR3_STORE_EXITING |
2462 					     CPU_BASED_INVLPG_EXITING);
2463 	} else if (vmx_cap->ept) {
2464 		vmx_cap->ept = 0;
2465 		pr_warn_once("EPT CAP should not exist if not support "
2466 				"1-setting enable EPT VM-execution control\n");
2467 	}
2468 	if (!(_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_VPID) &&
2469 		vmx_cap->vpid) {
2470 		vmx_cap->vpid = 0;
2471 		pr_warn_once("VPID CAP should not exist if not support "
2472 				"1-setting enable VPID VM-execution control\n");
2473 	}
2474 
2475 	min = VM_EXIT_SAVE_DEBUG_CONTROLS | VM_EXIT_ACK_INTR_ON_EXIT;
2476 #ifdef CONFIG_X86_64
2477 	min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
2478 #endif
2479 	opt = VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL |
2480 	      VM_EXIT_LOAD_IA32_PAT |
2481 	      VM_EXIT_LOAD_IA32_EFER |
2482 	      VM_EXIT_CLEAR_BNDCFGS |
2483 	      VM_EXIT_PT_CONCEAL_PIP |
2484 	      VM_EXIT_CLEAR_IA32_RTIT_CTL;
2485 	if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
2486 				&_vmexit_control) < 0)
2487 		return -EIO;
2488 
2489 	min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
2490 	opt = PIN_BASED_VIRTUAL_NMIS | PIN_BASED_POSTED_INTR |
2491 		 PIN_BASED_VMX_PREEMPTION_TIMER;
2492 	if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
2493 				&_pin_based_exec_control) < 0)
2494 		return -EIO;
2495 
2496 	if (cpu_has_broken_vmx_preemption_timer())
2497 		_pin_based_exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
2498 	if (!(_cpu_based_2nd_exec_control &
2499 		SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY))
2500 		_pin_based_exec_control &= ~PIN_BASED_POSTED_INTR;
2501 
2502 	min = VM_ENTRY_LOAD_DEBUG_CONTROLS;
2503 	opt = VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL |
2504 	      VM_ENTRY_LOAD_IA32_PAT |
2505 	      VM_ENTRY_LOAD_IA32_EFER |
2506 	      VM_ENTRY_LOAD_BNDCFGS |
2507 	      VM_ENTRY_PT_CONCEAL_PIP |
2508 	      VM_ENTRY_LOAD_IA32_RTIT_CTL;
2509 	if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
2510 				&_vmentry_control) < 0)
2511 		return -EIO;
2512 
2513 	/*
2514 	 * Some cpus support VM_{ENTRY,EXIT}_IA32_PERF_GLOBAL_CTRL but they
2515 	 * can't be used due to an errata where VM Exit may incorrectly clear
2516 	 * IA32_PERF_GLOBAL_CTRL[34:32].  Workaround the errata by using the
2517 	 * MSR load mechanism to switch IA32_PERF_GLOBAL_CTRL.
2518 	 */
2519 	if (boot_cpu_data.x86 == 0x6) {
2520 		switch (boot_cpu_data.x86_model) {
2521 		case 26: /* AAK155 */
2522 		case 30: /* AAP115 */
2523 		case 37: /* AAT100 */
2524 		case 44: /* BC86,AAY89,BD102 */
2525 		case 46: /* BA97 */
2526 			_vmentry_control &= ~VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL;
2527 			_vmexit_control &= ~VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL;
2528 			pr_warn_once("kvm: VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL "
2529 					"does not work properly. Using workaround\n");
2530 			break;
2531 		default:
2532 			break;
2533 		}
2534 	}
2535 
2536 
2537 	rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
2538 
2539 	/* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
2540 	if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
2541 		return -EIO;
2542 
2543 #ifdef CONFIG_X86_64
2544 	/* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
2545 	if (vmx_msr_high & (1u<<16))
2546 		return -EIO;
2547 #endif
2548 
2549 	/* Require Write-Back (WB) memory type for VMCS accesses. */
2550 	if (((vmx_msr_high >> 18) & 15) != 6)
2551 		return -EIO;
2552 
2553 	vmcs_conf->size = vmx_msr_high & 0x1fff;
2554 	vmcs_conf->order = get_order(vmcs_conf->size);
2555 	vmcs_conf->basic_cap = vmx_msr_high & ~0x1fff;
2556 
2557 	vmcs_conf->revision_id = vmx_msr_low;
2558 
2559 	vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
2560 	vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
2561 	vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
2562 	vmcs_conf->vmexit_ctrl         = _vmexit_control;
2563 	vmcs_conf->vmentry_ctrl        = _vmentry_control;
2564 
2565 	if (static_branch_unlikely(&enable_evmcs))
2566 		evmcs_sanitize_exec_ctrls(vmcs_conf);
2567 
2568 	return 0;
2569 }
2570 
2571 struct vmcs *alloc_vmcs_cpu(bool shadow, int cpu, gfp_t flags)
2572 {
2573 	int node = cpu_to_node(cpu);
2574 	struct page *pages;
2575 	struct vmcs *vmcs;
2576 
2577 	pages = __alloc_pages_node(node, flags, vmcs_config.order);
2578 	if (!pages)
2579 		return NULL;
2580 	vmcs = page_address(pages);
2581 	memset(vmcs, 0, vmcs_config.size);
2582 
2583 	/* KVM supports Enlightened VMCS v1 only */
2584 	if (static_branch_unlikely(&enable_evmcs))
2585 		vmcs->hdr.revision_id = KVM_EVMCS_VERSION;
2586 	else
2587 		vmcs->hdr.revision_id = vmcs_config.revision_id;
2588 
2589 	if (shadow)
2590 		vmcs->hdr.shadow_vmcs = 1;
2591 	return vmcs;
2592 }
2593 
2594 void free_vmcs(struct vmcs *vmcs)
2595 {
2596 	free_pages((unsigned long)vmcs, vmcs_config.order);
2597 }
2598 
2599 /*
2600  * Free a VMCS, but before that VMCLEAR it on the CPU where it was last loaded
2601  */
2602 void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
2603 {
2604 	if (!loaded_vmcs->vmcs)
2605 		return;
2606 	loaded_vmcs_clear(loaded_vmcs);
2607 	free_vmcs(loaded_vmcs->vmcs);
2608 	loaded_vmcs->vmcs = NULL;
2609 	if (loaded_vmcs->msr_bitmap)
2610 		free_page((unsigned long)loaded_vmcs->msr_bitmap);
2611 	WARN_ON(loaded_vmcs->shadow_vmcs != NULL);
2612 }
2613 
2614 int alloc_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
2615 {
2616 	loaded_vmcs->vmcs = alloc_vmcs(false);
2617 	if (!loaded_vmcs->vmcs)
2618 		return -ENOMEM;
2619 
2620 	loaded_vmcs->shadow_vmcs = NULL;
2621 	loaded_vmcs->hv_timer_soft_disabled = false;
2622 	loaded_vmcs_init(loaded_vmcs);
2623 
2624 	if (cpu_has_vmx_msr_bitmap()) {
2625 		loaded_vmcs->msr_bitmap = (unsigned long *)
2626 				__get_free_page(GFP_KERNEL_ACCOUNT);
2627 		if (!loaded_vmcs->msr_bitmap)
2628 			goto out_vmcs;
2629 		memset(loaded_vmcs->msr_bitmap, 0xff, PAGE_SIZE);
2630 
2631 		if (IS_ENABLED(CONFIG_HYPERV) &&
2632 		    static_branch_unlikely(&enable_evmcs) &&
2633 		    (ms_hyperv.nested_features & HV_X64_NESTED_MSR_BITMAP)) {
2634 			struct hv_enlightened_vmcs *evmcs =
2635 				(struct hv_enlightened_vmcs *)loaded_vmcs->vmcs;
2636 
2637 			evmcs->hv_enlightenments_control.msr_bitmap = 1;
2638 		}
2639 	}
2640 
2641 	memset(&loaded_vmcs->host_state, 0, sizeof(struct vmcs_host_state));
2642 	memset(&loaded_vmcs->controls_shadow, 0,
2643 		sizeof(struct vmcs_controls_shadow));
2644 
2645 	return 0;
2646 
2647 out_vmcs:
2648 	free_loaded_vmcs(loaded_vmcs);
2649 	return -ENOMEM;
2650 }
2651 
2652 static void free_kvm_area(void)
2653 {
2654 	int cpu;
2655 
2656 	for_each_possible_cpu(cpu) {
2657 		free_vmcs(per_cpu(vmxarea, cpu));
2658 		per_cpu(vmxarea, cpu) = NULL;
2659 	}
2660 }
2661 
2662 static __init int alloc_kvm_area(void)
2663 {
2664 	int cpu;
2665 
2666 	for_each_possible_cpu(cpu) {
2667 		struct vmcs *vmcs;
2668 
2669 		vmcs = alloc_vmcs_cpu(false, cpu, GFP_KERNEL);
2670 		if (!vmcs) {
2671 			free_kvm_area();
2672 			return -ENOMEM;
2673 		}
2674 
2675 		/*
2676 		 * When eVMCS is enabled, alloc_vmcs_cpu() sets
2677 		 * vmcs->revision_id to KVM_EVMCS_VERSION instead of
2678 		 * revision_id reported by MSR_IA32_VMX_BASIC.
2679 		 *
2680 		 * However, even though not explicitly documented by
2681 		 * TLFS, VMXArea passed as VMXON argument should
2682 		 * still be marked with revision_id reported by
2683 		 * physical CPU.
2684 		 */
2685 		if (static_branch_unlikely(&enable_evmcs))
2686 			vmcs->hdr.revision_id = vmcs_config.revision_id;
2687 
2688 		per_cpu(vmxarea, cpu) = vmcs;
2689 	}
2690 	return 0;
2691 }
2692 
2693 static void fix_pmode_seg(struct kvm_vcpu *vcpu, int seg,
2694 		struct kvm_segment *save)
2695 {
2696 	if (!emulate_invalid_guest_state) {
2697 		/*
2698 		 * CS and SS RPL should be equal during guest entry according
2699 		 * to VMX spec, but in reality it is not always so. Since vcpu
2700 		 * is in the middle of the transition from real mode to
2701 		 * protected mode it is safe to assume that RPL 0 is a good
2702 		 * default value.
2703 		 */
2704 		if (seg == VCPU_SREG_CS || seg == VCPU_SREG_SS)
2705 			save->selector &= ~SEGMENT_RPL_MASK;
2706 		save->dpl = save->selector & SEGMENT_RPL_MASK;
2707 		save->s = 1;
2708 	}
2709 	vmx_set_segment(vcpu, save, seg);
2710 }
2711 
2712 static void enter_pmode(struct kvm_vcpu *vcpu)
2713 {
2714 	unsigned long flags;
2715 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2716 
2717 	/*
2718 	 * Update real mode segment cache. It may be not up-to-date if sement
2719 	 * register was written while vcpu was in a guest mode.
2720 	 */
2721 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
2722 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
2723 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
2724 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
2725 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
2726 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
2727 
2728 	vmx->rmode.vm86_active = 0;
2729 
2730 	vmx_set_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
2731 
2732 	flags = vmcs_readl(GUEST_RFLAGS);
2733 	flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
2734 	flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
2735 	vmcs_writel(GUEST_RFLAGS, flags);
2736 
2737 	vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
2738 			(vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
2739 
2740 	update_exception_bitmap(vcpu);
2741 
2742 	fix_pmode_seg(vcpu, VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
2743 	fix_pmode_seg(vcpu, VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
2744 	fix_pmode_seg(vcpu, VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
2745 	fix_pmode_seg(vcpu, VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
2746 	fix_pmode_seg(vcpu, VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
2747 	fix_pmode_seg(vcpu, VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
2748 }
2749 
2750 static void fix_rmode_seg(int seg, struct kvm_segment *save)
2751 {
2752 	const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2753 	struct kvm_segment var = *save;
2754 
2755 	var.dpl = 0x3;
2756 	if (seg == VCPU_SREG_CS)
2757 		var.type = 0x3;
2758 
2759 	if (!emulate_invalid_guest_state) {
2760 		var.selector = var.base >> 4;
2761 		var.base = var.base & 0xffff0;
2762 		var.limit = 0xffff;
2763 		var.g = 0;
2764 		var.db = 0;
2765 		var.present = 1;
2766 		var.s = 1;
2767 		var.l = 0;
2768 		var.unusable = 0;
2769 		var.type = 0x3;
2770 		var.avl = 0;
2771 		if (save->base & 0xf)
2772 			printk_once(KERN_WARNING "kvm: segment base is not "
2773 					"paragraph aligned when entering "
2774 					"protected mode (seg=%d)", seg);
2775 	}
2776 
2777 	vmcs_write16(sf->selector, var.selector);
2778 	vmcs_writel(sf->base, var.base);
2779 	vmcs_write32(sf->limit, var.limit);
2780 	vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(&var));
2781 }
2782 
2783 static void enter_rmode(struct kvm_vcpu *vcpu)
2784 {
2785 	unsigned long flags;
2786 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2787 	struct kvm_vmx *kvm_vmx = to_kvm_vmx(vcpu->kvm);
2788 
2789 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
2790 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
2791 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
2792 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
2793 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
2794 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
2795 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
2796 
2797 	vmx->rmode.vm86_active = 1;
2798 
2799 	/*
2800 	 * Very old userspace does not call KVM_SET_TSS_ADDR before entering
2801 	 * vcpu. Warn the user that an update is overdue.
2802 	 */
2803 	if (!kvm_vmx->tss_addr)
2804 		printk_once(KERN_WARNING "kvm: KVM_SET_TSS_ADDR need to be "
2805 			     "called before entering vcpu\n");
2806 
2807 	vmx_segment_cache_clear(vmx);
2808 
2809 	vmcs_writel(GUEST_TR_BASE, kvm_vmx->tss_addr);
2810 	vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
2811 	vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
2812 
2813 	flags = vmcs_readl(GUEST_RFLAGS);
2814 	vmx->rmode.save_rflags = flags;
2815 
2816 	flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
2817 
2818 	vmcs_writel(GUEST_RFLAGS, flags);
2819 	vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
2820 	update_exception_bitmap(vcpu);
2821 
2822 	fix_rmode_seg(VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
2823 	fix_rmode_seg(VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
2824 	fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
2825 	fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
2826 	fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
2827 	fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
2828 
2829 	kvm_mmu_reset_context(vcpu);
2830 }
2831 
2832 void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
2833 {
2834 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2835 	struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
2836 
2837 	if (!msr)
2838 		return;
2839 
2840 	vcpu->arch.efer = efer;
2841 	if (efer & EFER_LMA) {
2842 		vm_entry_controls_setbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
2843 		msr->data = efer;
2844 	} else {
2845 		vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
2846 
2847 		msr->data = efer & ~EFER_LME;
2848 	}
2849 	setup_msrs(vmx);
2850 }
2851 
2852 #ifdef CONFIG_X86_64
2853 
2854 static void enter_lmode(struct kvm_vcpu *vcpu)
2855 {
2856 	u32 guest_tr_ar;
2857 
2858 	vmx_segment_cache_clear(to_vmx(vcpu));
2859 
2860 	guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
2861 	if ((guest_tr_ar & VMX_AR_TYPE_MASK) != VMX_AR_TYPE_BUSY_64_TSS) {
2862 		pr_debug_ratelimited("%s: tss fixup for long mode. \n",
2863 				     __func__);
2864 		vmcs_write32(GUEST_TR_AR_BYTES,
2865 			     (guest_tr_ar & ~VMX_AR_TYPE_MASK)
2866 			     | VMX_AR_TYPE_BUSY_64_TSS);
2867 	}
2868 	vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
2869 }
2870 
2871 static void exit_lmode(struct kvm_vcpu *vcpu)
2872 {
2873 	vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
2874 	vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
2875 }
2876 
2877 #endif
2878 
2879 static void vmx_flush_tlb_gva(struct kvm_vcpu *vcpu, gva_t addr)
2880 {
2881 	int vpid = to_vmx(vcpu)->vpid;
2882 
2883 	if (!vpid_sync_vcpu_addr(vpid, addr))
2884 		vpid_sync_context(vpid);
2885 
2886 	/*
2887 	 * If VPIDs are not supported or enabled, then the above is a no-op.
2888 	 * But we don't really need a TLB flush in that case anyway, because
2889 	 * each VM entry/exit includes an implicit flush when VPID is 0.
2890 	 */
2891 }
2892 
2893 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
2894 {
2895 	ulong cr0_guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
2896 
2897 	vcpu->arch.cr0 &= ~cr0_guest_owned_bits;
2898 	vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & cr0_guest_owned_bits;
2899 }
2900 
2901 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
2902 {
2903 	ulong cr4_guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
2904 
2905 	vcpu->arch.cr4 &= ~cr4_guest_owned_bits;
2906 	vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & cr4_guest_owned_bits;
2907 }
2908 
2909 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
2910 {
2911 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
2912 
2913 	if (!kvm_register_is_dirty(vcpu, VCPU_EXREG_PDPTR))
2914 		return;
2915 
2916 	if (is_pae_paging(vcpu)) {
2917 		vmcs_write64(GUEST_PDPTR0, mmu->pdptrs[0]);
2918 		vmcs_write64(GUEST_PDPTR1, mmu->pdptrs[1]);
2919 		vmcs_write64(GUEST_PDPTR2, mmu->pdptrs[2]);
2920 		vmcs_write64(GUEST_PDPTR3, mmu->pdptrs[3]);
2921 	}
2922 }
2923 
2924 void ept_save_pdptrs(struct kvm_vcpu *vcpu)
2925 {
2926 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
2927 
2928 	if (is_pae_paging(vcpu)) {
2929 		mmu->pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
2930 		mmu->pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
2931 		mmu->pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
2932 		mmu->pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
2933 	}
2934 
2935 	kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
2936 }
2937 
2938 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
2939 					unsigned long cr0,
2940 					struct kvm_vcpu *vcpu)
2941 {
2942 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2943 
2944 	if (!kvm_register_is_available(vcpu, VCPU_EXREG_CR3))
2945 		vmx_cache_reg(vcpu, VCPU_EXREG_CR3);
2946 	if (!(cr0 & X86_CR0_PG)) {
2947 		/* From paging/starting to nonpaging */
2948 		exec_controls_setbit(vmx, CPU_BASED_CR3_LOAD_EXITING |
2949 					  CPU_BASED_CR3_STORE_EXITING);
2950 		vcpu->arch.cr0 = cr0;
2951 		vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
2952 	} else if (!is_paging(vcpu)) {
2953 		/* From nonpaging to paging */
2954 		exec_controls_clearbit(vmx, CPU_BASED_CR3_LOAD_EXITING |
2955 					    CPU_BASED_CR3_STORE_EXITING);
2956 		vcpu->arch.cr0 = cr0;
2957 		vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
2958 	}
2959 
2960 	if (!(cr0 & X86_CR0_WP))
2961 		*hw_cr0 &= ~X86_CR0_WP;
2962 }
2963 
2964 void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
2965 {
2966 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2967 	unsigned long hw_cr0;
2968 
2969 	hw_cr0 = (cr0 & ~KVM_VM_CR0_ALWAYS_OFF);
2970 	if (enable_unrestricted_guest)
2971 		hw_cr0 |= KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
2972 	else {
2973 		hw_cr0 |= KVM_VM_CR0_ALWAYS_ON;
2974 
2975 		if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
2976 			enter_pmode(vcpu);
2977 
2978 		if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
2979 			enter_rmode(vcpu);
2980 	}
2981 
2982 #ifdef CONFIG_X86_64
2983 	if (vcpu->arch.efer & EFER_LME) {
2984 		if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
2985 			enter_lmode(vcpu);
2986 		if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
2987 			exit_lmode(vcpu);
2988 	}
2989 #endif
2990 
2991 	if (enable_ept && !enable_unrestricted_guest)
2992 		ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
2993 
2994 	vmcs_writel(CR0_READ_SHADOW, cr0);
2995 	vmcs_writel(GUEST_CR0, hw_cr0);
2996 	vcpu->arch.cr0 = cr0;
2997 
2998 	/* depends on vcpu->arch.cr0 to be set to a new value */
2999 	vmx->emulation_required = emulation_required(vcpu);
3000 }
3001 
3002 static int get_ept_level(struct kvm_vcpu *vcpu)
3003 {
3004 	/* Nested EPT currently only supports 4-level walks. */
3005 	if (is_guest_mode(vcpu) && nested_cpu_has_ept(get_vmcs12(vcpu)))
3006 		return 4;
3007 	if (cpu_has_vmx_ept_5levels() && (cpuid_maxphyaddr(vcpu) > 48))
3008 		return 5;
3009 	return 4;
3010 }
3011 
3012 u64 construct_eptp(struct kvm_vcpu *vcpu, unsigned long root_hpa)
3013 {
3014 	u64 eptp = VMX_EPTP_MT_WB;
3015 
3016 	eptp |= (get_ept_level(vcpu) == 5) ? VMX_EPTP_PWL_5 : VMX_EPTP_PWL_4;
3017 
3018 	if (enable_ept_ad_bits &&
3019 	    (!is_guest_mode(vcpu) || nested_ept_ad_enabled(vcpu)))
3020 		eptp |= VMX_EPTP_AD_ENABLE_BIT;
3021 	eptp |= (root_hpa & PAGE_MASK);
3022 
3023 	return eptp;
3024 }
3025 
3026 void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
3027 {
3028 	struct kvm *kvm = vcpu->kvm;
3029 	bool update_guest_cr3 = true;
3030 	unsigned long guest_cr3;
3031 	u64 eptp;
3032 
3033 	guest_cr3 = cr3;
3034 	if (enable_ept) {
3035 		eptp = construct_eptp(vcpu, cr3);
3036 		vmcs_write64(EPT_POINTER, eptp);
3037 
3038 		if (kvm_x86_ops->tlb_remote_flush) {
3039 			spin_lock(&to_kvm_vmx(kvm)->ept_pointer_lock);
3040 			to_vmx(vcpu)->ept_pointer = eptp;
3041 			to_kvm_vmx(kvm)->ept_pointers_match
3042 				= EPT_POINTERS_CHECK;
3043 			spin_unlock(&to_kvm_vmx(kvm)->ept_pointer_lock);
3044 		}
3045 
3046 		/* Loading vmcs02.GUEST_CR3 is handled by nested VM-Enter. */
3047 		if (is_guest_mode(vcpu))
3048 			update_guest_cr3 = false;
3049 		else if (!enable_unrestricted_guest && !is_paging(vcpu))
3050 			guest_cr3 = to_kvm_vmx(kvm)->ept_identity_map_addr;
3051 		else if (test_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail))
3052 			guest_cr3 = vcpu->arch.cr3;
3053 		else /* vmcs01.GUEST_CR3 is already up-to-date. */
3054 			update_guest_cr3 = false;
3055 		ept_load_pdptrs(vcpu);
3056 	}
3057 
3058 	if (update_guest_cr3)
3059 		vmcs_writel(GUEST_CR3, guest_cr3);
3060 }
3061 
3062 int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
3063 {
3064 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3065 	/*
3066 	 * Pass through host's Machine Check Enable value to hw_cr4, which
3067 	 * is in force while we are in guest mode.  Do not let guests control
3068 	 * this bit, even if host CR4.MCE == 0.
3069 	 */
3070 	unsigned long hw_cr4;
3071 
3072 	hw_cr4 = (cr4_read_shadow() & X86_CR4_MCE) | (cr4 & ~X86_CR4_MCE);
3073 	if (enable_unrestricted_guest)
3074 		hw_cr4 |= KVM_VM_CR4_ALWAYS_ON_UNRESTRICTED_GUEST;
3075 	else if (vmx->rmode.vm86_active)
3076 		hw_cr4 |= KVM_RMODE_VM_CR4_ALWAYS_ON;
3077 	else
3078 		hw_cr4 |= KVM_PMODE_VM_CR4_ALWAYS_ON;
3079 
3080 	if (!boot_cpu_has(X86_FEATURE_UMIP) && vmx_umip_emulated()) {
3081 		if (cr4 & X86_CR4_UMIP) {
3082 			secondary_exec_controls_setbit(vmx, SECONDARY_EXEC_DESC);
3083 			hw_cr4 &= ~X86_CR4_UMIP;
3084 		} else if (!is_guest_mode(vcpu) ||
3085 			!nested_cpu_has2(get_vmcs12(vcpu), SECONDARY_EXEC_DESC)) {
3086 			secondary_exec_controls_clearbit(vmx, SECONDARY_EXEC_DESC);
3087 		}
3088 	}
3089 
3090 	if (cr4 & X86_CR4_VMXE) {
3091 		/*
3092 		 * To use VMXON (and later other VMX instructions), a guest
3093 		 * must first be able to turn on cr4.VMXE (see handle_vmon()).
3094 		 * So basically the check on whether to allow nested VMX
3095 		 * is here.  We operate under the default treatment of SMM,
3096 		 * so VMX cannot be enabled under SMM.
3097 		 */
3098 		if (!nested_vmx_allowed(vcpu) || is_smm(vcpu))
3099 			return 1;
3100 	}
3101 
3102 	if (vmx->nested.vmxon && !nested_cr4_valid(vcpu, cr4))
3103 		return 1;
3104 
3105 	vcpu->arch.cr4 = cr4;
3106 
3107 	if (!enable_unrestricted_guest) {
3108 		if (enable_ept) {
3109 			if (!is_paging(vcpu)) {
3110 				hw_cr4 &= ~X86_CR4_PAE;
3111 				hw_cr4 |= X86_CR4_PSE;
3112 			} else if (!(cr4 & X86_CR4_PAE)) {
3113 				hw_cr4 &= ~X86_CR4_PAE;
3114 			}
3115 		}
3116 
3117 		/*
3118 		 * SMEP/SMAP/PKU is disabled if CPU is in non-paging mode in
3119 		 * hardware.  To emulate this behavior, SMEP/SMAP/PKU needs
3120 		 * to be manually disabled when guest switches to non-paging
3121 		 * mode.
3122 		 *
3123 		 * If !enable_unrestricted_guest, the CPU is always running
3124 		 * with CR0.PG=1 and CR4 needs to be modified.
3125 		 * If enable_unrestricted_guest, the CPU automatically
3126 		 * disables SMEP/SMAP/PKU when the guest sets CR0.PG=0.
3127 		 */
3128 		if (!is_paging(vcpu))
3129 			hw_cr4 &= ~(X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE);
3130 	}
3131 
3132 	vmcs_writel(CR4_READ_SHADOW, cr4);
3133 	vmcs_writel(GUEST_CR4, hw_cr4);
3134 	return 0;
3135 }
3136 
3137 void vmx_get_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg)
3138 {
3139 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3140 	u32 ar;
3141 
3142 	if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3143 		*var = vmx->rmode.segs[seg];
3144 		if (seg == VCPU_SREG_TR
3145 		    || var->selector == vmx_read_guest_seg_selector(vmx, seg))
3146 			return;
3147 		var->base = vmx_read_guest_seg_base(vmx, seg);
3148 		var->selector = vmx_read_guest_seg_selector(vmx, seg);
3149 		return;
3150 	}
3151 	var->base = vmx_read_guest_seg_base(vmx, seg);
3152 	var->limit = vmx_read_guest_seg_limit(vmx, seg);
3153 	var->selector = vmx_read_guest_seg_selector(vmx, seg);
3154 	ar = vmx_read_guest_seg_ar(vmx, seg);
3155 	var->unusable = (ar >> 16) & 1;
3156 	var->type = ar & 15;
3157 	var->s = (ar >> 4) & 1;
3158 	var->dpl = (ar >> 5) & 3;
3159 	/*
3160 	 * Some userspaces do not preserve unusable property. Since usable
3161 	 * segment has to be present according to VMX spec we can use present
3162 	 * property to amend userspace bug by making unusable segment always
3163 	 * nonpresent. vmx_segment_access_rights() already marks nonpresent
3164 	 * segment as unusable.
3165 	 */
3166 	var->present = !var->unusable;
3167 	var->avl = (ar >> 12) & 1;
3168 	var->l = (ar >> 13) & 1;
3169 	var->db = (ar >> 14) & 1;
3170 	var->g = (ar >> 15) & 1;
3171 }
3172 
3173 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
3174 {
3175 	struct kvm_segment s;
3176 
3177 	if (to_vmx(vcpu)->rmode.vm86_active) {
3178 		vmx_get_segment(vcpu, &s, seg);
3179 		return s.base;
3180 	}
3181 	return vmx_read_guest_seg_base(to_vmx(vcpu), seg);
3182 }
3183 
3184 int vmx_get_cpl(struct kvm_vcpu *vcpu)
3185 {
3186 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3187 
3188 	if (unlikely(vmx->rmode.vm86_active))
3189 		return 0;
3190 	else {
3191 		int ar = vmx_read_guest_seg_ar(vmx, VCPU_SREG_SS);
3192 		return VMX_AR_DPL(ar);
3193 	}
3194 }
3195 
3196 static u32 vmx_segment_access_rights(struct kvm_segment *var)
3197 {
3198 	u32 ar;
3199 
3200 	if (var->unusable || !var->present)
3201 		ar = 1 << 16;
3202 	else {
3203 		ar = var->type & 15;
3204 		ar |= (var->s & 1) << 4;
3205 		ar |= (var->dpl & 3) << 5;
3206 		ar |= (var->present & 1) << 7;
3207 		ar |= (var->avl & 1) << 12;
3208 		ar |= (var->l & 1) << 13;
3209 		ar |= (var->db & 1) << 14;
3210 		ar |= (var->g & 1) << 15;
3211 	}
3212 
3213 	return ar;
3214 }
3215 
3216 void vmx_set_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg)
3217 {
3218 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3219 	const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3220 
3221 	vmx_segment_cache_clear(vmx);
3222 
3223 	if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3224 		vmx->rmode.segs[seg] = *var;
3225 		if (seg == VCPU_SREG_TR)
3226 			vmcs_write16(sf->selector, var->selector);
3227 		else if (var->s)
3228 			fix_rmode_seg(seg, &vmx->rmode.segs[seg]);
3229 		goto out;
3230 	}
3231 
3232 	vmcs_writel(sf->base, var->base);
3233 	vmcs_write32(sf->limit, var->limit);
3234 	vmcs_write16(sf->selector, var->selector);
3235 
3236 	/*
3237 	 *   Fix the "Accessed" bit in AR field of segment registers for older
3238 	 * qemu binaries.
3239 	 *   IA32 arch specifies that at the time of processor reset the
3240 	 * "Accessed" bit in the AR field of segment registers is 1. And qemu
3241 	 * is setting it to 0 in the userland code. This causes invalid guest
3242 	 * state vmexit when "unrestricted guest" mode is turned on.
3243 	 *    Fix for this setup issue in cpu_reset is being pushed in the qemu
3244 	 * tree. Newer qemu binaries with that qemu fix would not need this
3245 	 * kvm hack.
3246 	 */
3247 	if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
3248 		var->type |= 0x1; /* Accessed */
3249 
3250 	vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(var));
3251 
3252 out:
3253 	vmx->emulation_required = emulation_required(vcpu);
3254 }
3255 
3256 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3257 {
3258 	u32 ar = vmx_read_guest_seg_ar(to_vmx(vcpu), VCPU_SREG_CS);
3259 
3260 	*db = (ar >> 14) & 1;
3261 	*l = (ar >> 13) & 1;
3262 }
3263 
3264 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3265 {
3266 	dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
3267 	dt->address = vmcs_readl(GUEST_IDTR_BASE);
3268 }
3269 
3270 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3271 {
3272 	vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
3273 	vmcs_writel(GUEST_IDTR_BASE, dt->address);
3274 }
3275 
3276 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3277 {
3278 	dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
3279 	dt->address = vmcs_readl(GUEST_GDTR_BASE);
3280 }
3281 
3282 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3283 {
3284 	vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
3285 	vmcs_writel(GUEST_GDTR_BASE, dt->address);
3286 }
3287 
3288 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
3289 {
3290 	struct kvm_segment var;
3291 	u32 ar;
3292 
3293 	vmx_get_segment(vcpu, &var, seg);
3294 	var.dpl = 0x3;
3295 	if (seg == VCPU_SREG_CS)
3296 		var.type = 0x3;
3297 	ar = vmx_segment_access_rights(&var);
3298 
3299 	if (var.base != (var.selector << 4))
3300 		return false;
3301 	if (var.limit != 0xffff)
3302 		return false;
3303 	if (ar != 0xf3)
3304 		return false;
3305 
3306 	return true;
3307 }
3308 
3309 static bool code_segment_valid(struct kvm_vcpu *vcpu)
3310 {
3311 	struct kvm_segment cs;
3312 	unsigned int cs_rpl;
3313 
3314 	vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3315 	cs_rpl = cs.selector & SEGMENT_RPL_MASK;
3316 
3317 	if (cs.unusable)
3318 		return false;
3319 	if (~cs.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_ACCESSES_MASK))
3320 		return false;
3321 	if (!cs.s)
3322 		return false;
3323 	if (cs.type & VMX_AR_TYPE_WRITEABLE_MASK) {
3324 		if (cs.dpl > cs_rpl)
3325 			return false;
3326 	} else {
3327 		if (cs.dpl != cs_rpl)
3328 			return false;
3329 	}
3330 	if (!cs.present)
3331 		return false;
3332 
3333 	/* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
3334 	return true;
3335 }
3336 
3337 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
3338 {
3339 	struct kvm_segment ss;
3340 	unsigned int ss_rpl;
3341 
3342 	vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3343 	ss_rpl = ss.selector & SEGMENT_RPL_MASK;
3344 
3345 	if (ss.unusable)
3346 		return true;
3347 	if (ss.type != 3 && ss.type != 7)
3348 		return false;
3349 	if (!ss.s)
3350 		return false;
3351 	if (ss.dpl != ss_rpl) /* DPL != RPL */
3352 		return false;
3353 	if (!ss.present)
3354 		return false;
3355 
3356 	return true;
3357 }
3358 
3359 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
3360 {
3361 	struct kvm_segment var;
3362 	unsigned int rpl;
3363 
3364 	vmx_get_segment(vcpu, &var, seg);
3365 	rpl = var.selector & SEGMENT_RPL_MASK;
3366 
3367 	if (var.unusable)
3368 		return true;
3369 	if (!var.s)
3370 		return false;
3371 	if (!var.present)
3372 		return false;
3373 	if (~var.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_WRITEABLE_MASK)) {
3374 		if (var.dpl < rpl) /* DPL < RPL */
3375 			return false;
3376 	}
3377 
3378 	/* TODO: Add other members to kvm_segment_field to allow checking for other access
3379 	 * rights flags
3380 	 */
3381 	return true;
3382 }
3383 
3384 static bool tr_valid(struct kvm_vcpu *vcpu)
3385 {
3386 	struct kvm_segment tr;
3387 
3388 	vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
3389 
3390 	if (tr.unusable)
3391 		return false;
3392 	if (tr.selector & SEGMENT_TI_MASK)	/* TI = 1 */
3393 		return false;
3394 	if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
3395 		return false;
3396 	if (!tr.present)
3397 		return false;
3398 
3399 	return true;
3400 }
3401 
3402 static bool ldtr_valid(struct kvm_vcpu *vcpu)
3403 {
3404 	struct kvm_segment ldtr;
3405 
3406 	vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
3407 
3408 	if (ldtr.unusable)
3409 		return true;
3410 	if (ldtr.selector & SEGMENT_TI_MASK)	/* TI = 1 */
3411 		return false;
3412 	if (ldtr.type != 2)
3413 		return false;
3414 	if (!ldtr.present)
3415 		return false;
3416 
3417 	return true;
3418 }
3419 
3420 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
3421 {
3422 	struct kvm_segment cs, ss;
3423 
3424 	vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3425 	vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3426 
3427 	return ((cs.selector & SEGMENT_RPL_MASK) ==
3428 		 (ss.selector & SEGMENT_RPL_MASK));
3429 }
3430 
3431 /*
3432  * Check if guest state is valid. Returns true if valid, false if
3433  * not.
3434  * We assume that registers are always usable
3435  */
3436 static bool guest_state_valid(struct kvm_vcpu *vcpu)
3437 {
3438 	if (enable_unrestricted_guest)
3439 		return true;
3440 
3441 	/* real mode guest state checks */
3442 	if (!is_protmode(vcpu) || (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
3443 		if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
3444 			return false;
3445 		if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
3446 			return false;
3447 		if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
3448 			return false;
3449 		if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
3450 			return false;
3451 		if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
3452 			return false;
3453 		if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
3454 			return false;
3455 	} else {
3456 	/* protected mode guest state checks */
3457 		if (!cs_ss_rpl_check(vcpu))
3458 			return false;
3459 		if (!code_segment_valid(vcpu))
3460 			return false;
3461 		if (!stack_segment_valid(vcpu))
3462 			return false;
3463 		if (!data_segment_valid(vcpu, VCPU_SREG_DS))
3464 			return false;
3465 		if (!data_segment_valid(vcpu, VCPU_SREG_ES))
3466 			return false;
3467 		if (!data_segment_valid(vcpu, VCPU_SREG_FS))
3468 			return false;
3469 		if (!data_segment_valid(vcpu, VCPU_SREG_GS))
3470 			return false;
3471 		if (!tr_valid(vcpu))
3472 			return false;
3473 		if (!ldtr_valid(vcpu))
3474 			return false;
3475 	}
3476 	/* TODO:
3477 	 * - Add checks on RIP
3478 	 * - Add checks on RFLAGS
3479 	 */
3480 
3481 	return true;
3482 }
3483 
3484 static int init_rmode_tss(struct kvm *kvm)
3485 {
3486 	gfn_t fn;
3487 	u16 data = 0;
3488 	int idx, r;
3489 
3490 	idx = srcu_read_lock(&kvm->srcu);
3491 	fn = to_kvm_vmx(kvm)->tss_addr >> PAGE_SHIFT;
3492 	r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3493 	if (r < 0)
3494 		goto out;
3495 	data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
3496 	r = kvm_write_guest_page(kvm, fn++, &data,
3497 			TSS_IOPB_BASE_OFFSET, sizeof(u16));
3498 	if (r < 0)
3499 		goto out;
3500 	r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
3501 	if (r < 0)
3502 		goto out;
3503 	r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3504 	if (r < 0)
3505 		goto out;
3506 	data = ~0;
3507 	r = kvm_write_guest_page(kvm, fn, &data,
3508 				 RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
3509 				 sizeof(u8));
3510 out:
3511 	srcu_read_unlock(&kvm->srcu, idx);
3512 	return r;
3513 }
3514 
3515 static int init_rmode_identity_map(struct kvm *kvm)
3516 {
3517 	struct kvm_vmx *kvm_vmx = to_kvm_vmx(kvm);
3518 	int i, r = 0;
3519 	kvm_pfn_t identity_map_pfn;
3520 	u32 tmp;
3521 
3522 	/* Protect kvm_vmx->ept_identity_pagetable_done. */
3523 	mutex_lock(&kvm->slots_lock);
3524 
3525 	if (likely(kvm_vmx->ept_identity_pagetable_done))
3526 		goto out;
3527 
3528 	if (!kvm_vmx->ept_identity_map_addr)
3529 		kvm_vmx->ept_identity_map_addr = VMX_EPT_IDENTITY_PAGETABLE_ADDR;
3530 	identity_map_pfn = kvm_vmx->ept_identity_map_addr >> PAGE_SHIFT;
3531 
3532 	r = __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
3533 				    kvm_vmx->ept_identity_map_addr, PAGE_SIZE);
3534 	if (r < 0)
3535 		goto out;
3536 
3537 	r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
3538 	if (r < 0)
3539 		goto out;
3540 	/* Set up identity-mapping pagetable for EPT in real mode */
3541 	for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
3542 		tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
3543 			_PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
3544 		r = kvm_write_guest_page(kvm, identity_map_pfn,
3545 				&tmp, i * sizeof(tmp), sizeof(tmp));
3546 		if (r < 0)
3547 			goto out;
3548 	}
3549 	kvm_vmx->ept_identity_pagetable_done = true;
3550 
3551 out:
3552 	mutex_unlock(&kvm->slots_lock);
3553 	return r;
3554 }
3555 
3556 static void seg_setup(int seg)
3557 {
3558 	const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3559 	unsigned int ar;
3560 
3561 	vmcs_write16(sf->selector, 0);
3562 	vmcs_writel(sf->base, 0);
3563 	vmcs_write32(sf->limit, 0xffff);
3564 	ar = 0x93;
3565 	if (seg == VCPU_SREG_CS)
3566 		ar |= 0x08; /* code segment */
3567 
3568 	vmcs_write32(sf->ar_bytes, ar);
3569 }
3570 
3571 static int alloc_apic_access_page(struct kvm *kvm)
3572 {
3573 	struct page *page;
3574 	int r = 0;
3575 
3576 	mutex_lock(&kvm->slots_lock);
3577 	if (kvm->arch.apic_access_page_done)
3578 		goto out;
3579 	r = __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
3580 				    APIC_DEFAULT_PHYS_BASE, PAGE_SIZE);
3581 	if (r)
3582 		goto out;
3583 
3584 	page = gfn_to_page(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
3585 	if (is_error_page(page)) {
3586 		r = -EFAULT;
3587 		goto out;
3588 	}
3589 
3590 	/*
3591 	 * Do not pin the page in memory, so that memory hot-unplug
3592 	 * is able to migrate it.
3593 	 */
3594 	put_page(page);
3595 	kvm->arch.apic_access_page_done = true;
3596 out:
3597 	mutex_unlock(&kvm->slots_lock);
3598 	return r;
3599 }
3600 
3601 int allocate_vpid(void)
3602 {
3603 	int vpid;
3604 
3605 	if (!enable_vpid)
3606 		return 0;
3607 	spin_lock(&vmx_vpid_lock);
3608 	vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
3609 	if (vpid < VMX_NR_VPIDS)
3610 		__set_bit(vpid, vmx_vpid_bitmap);
3611 	else
3612 		vpid = 0;
3613 	spin_unlock(&vmx_vpid_lock);
3614 	return vpid;
3615 }
3616 
3617 void free_vpid(int vpid)
3618 {
3619 	if (!enable_vpid || vpid == 0)
3620 		return;
3621 	spin_lock(&vmx_vpid_lock);
3622 	__clear_bit(vpid, vmx_vpid_bitmap);
3623 	spin_unlock(&vmx_vpid_lock);
3624 }
3625 
3626 static __always_inline void vmx_disable_intercept_for_msr(unsigned long *msr_bitmap,
3627 							  u32 msr, int type)
3628 {
3629 	int f = sizeof(unsigned long);
3630 
3631 	if (!cpu_has_vmx_msr_bitmap())
3632 		return;
3633 
3634 	if (static_branch_unlikely(&enable_evmcs))
3635 		evmcs_touch_msr_bitmap();
3636 
3637 	/*
3638 	 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
3639 	 * have the write-low and read-high bitmap offsets the wrong way round.
3640 	 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
3641 	 */
3642 	if (msr <= 0x1fff) {
3643 		if (type & MSR_TYPE_R)
3644 			/* read-low */
3645 			__clear_bit(msr, msr_bitmap + 0x000 / f);
3646 
3647 		if (type & MSR_TYPE_W)
3648 			/* write-low */
3649 			__clear_bit(msr, msr_bitmap + 0x800 / f);
3650 
3651 	} else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
3652 		msr &= 0x1fff;
3653 		if (type & MSR_TYPE_R)
3654 			/* read-high */
3655 			__clear_bit(msr, msr_bitmap + 0x400 / f);
3656 
3657 		if (type & MSR_TYPE_W)
3658 			/* write-high */
3659 			__clear_bit(msr, msr_bitmap + 0xc00 / f);
3660 
3661 	}
3662 }
3663 
3664 static __always_inline void vmx_enable_intercept_for_msr(unsigned long *msr_bitmap,
3665 							 u32 msr, int type)
3666 {
3667 	int f = sizeof(unsigned long);
3668 
3669 	if (!cpu_has_vmx_msr_bitmap())
3670 		return;
3671 
3672 	if (static_branch_unlikely(&enable_evmcs))
3673 		evmcs_touch_msr_bitmap();
3674 
3675 	/*
3676 	 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
3677 	 * have the write-low and read-high bitmap offsets the wrong way round.
3678 	 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
3679 	 */
3680 	if (msr <= 0x1fff) {
3681 		if (type & MSR_TYPE_R)
3682 			/* read-low */
3683 			__set_bit(msr, msr_bitmap + 0x000 / f);
3684 
3685 		if (type & MSR_TYPE_W)
3686 			/* write-low */
3687 			__set_bit(msr, msr_bitmap + 0x800 / f);
3688 
3689 	} else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
3690 		msr &= 0x1fff;
3691 		if (type & MSR_TYPE_R)
3692 			/* read-high */
3693 			__set_bit(msr, msr_bitmap + 0x400 / f);
3694 
3695 		if (type & MSR_TYPE_W)
3696 			/* write-high */
3697 			__set_bit(msr, msr_bitmap + 0xc00 / f);
3698 
3699 	}
3700 }
3701 
3702 static __always_inline void vmx_set_intercept_for_msr(unsigned long *msr_bitmap,
3703 			     			      u32 msr, int type, bool value)
3704 {
3705 	if (value)
3706 		vmx_enable_intercept_for_msr(msr_bitmap, msr, type);
3707 	else
3708 		vmx_disable_intercept_for_msr(msr_bitmap, msr, type);
3709 }
3710 
3711 static u8 vmx_msr_bitmap_mode(struct kvm_vcpu *vcpu)
3712 {
3713 	u8 mode = 0;
3714 
3715 	if (cpu_has_secondary_exec_ctrls() &&
3716 	    (secondary_exec_controls_get(to_vmx(vcpu)) &
3717 	     SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE)) {
3718 		mode |= MSR_BITMAP_MODE_X2APIC;
3719 		if (enable_apicv && kvm_vcpu_apicv_active(vcpu))
3720 			mode |= MSR_BITMAP_MODE_X2APIC_APICV;
3721 	}
3722 
3723 	return mode;
3724 }
3725 
3726 static void vmx_update_msr_bitmap_x2apic(unsigned long *msr_bitmap,
3727 					 u8 mode)
3728 {
3729 	int msr;
3730 
3731 	for (msr = 0x800; msr <= 0x8ff; msr += BITS_PER_LONG) {
3732 		unsigned word = msr / BITS_PER_LONG;
3733 		msr_bitmap[word] = (mode & MSR_BITMAP_MODE_X2APIC_APICV) ? 0 : ~0;
3734 		msr_bitmap[word + (0x800 / sizeof(long))] = ~0;
3735 	}
3736 
3737 	if (mode & MSR_BITMAP_MODE_X2APIC) {
3738 		/*
3739 		 * TPR reads and writes can be virtualized even if virtual interrupt
3740 		 * delivery is not in use.
3741 		 */
3742 		vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_TASKPRI), MSR_TYPE_RW);
3743 		if (mode & MSR_BITMAP_MODE_X2APIC_APICV) {
3744 			vmx_enable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_TMCCT), MSR_TYPE_R);
3745 			vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_EOI), MSR_TYPE_W);
3746 			vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_SELF_IPI), MSR_TYPE_W);
3747 		}
3748 	}
3749 }
3750 
3751 void vmx_update_msr_bitmap(struct kvm_vcpu *vcpu)
3752 {
3753 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3754 	unsigned long *msr_bitmap = vmx->vmcs01.msr_bitmap;
3755 	u8 mode = vmx_msr_bitmap_mode(vcpu);
3756 	u8 changed = mode ^ vmx->msr_bitmap_mode;
3757 
3758 	if (!changed)
3759 		return;
3760 
3761 	if (changed & (MSR_BITMAP_MODE_X2APIC | MSR_BITMAP_MODE_X2APIC_APICV))
3762 		vmx_update_msr_bitmap_x2apic(msr_bitmap, mode);
3763 
3764 	vmx->msr_bitmap_mode = mode;
3765 }
3766 
3767 void pt_update_intercept_for_msr(struct vcpu_vmx *vmx)
3768 {
3769 	unsigned long *msr_bitmap = vmx->vmcs01.msr_bitmap;
3770 	bool flag = !(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN);
3771 	u32 i;
3772 
3773 	vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_STATUS,
3774 							MSR_TYPE_RW, flag);
3775 	vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_OUTPUT_BASE,
3776 							MSR_TYPE_RW, flag);
3777 	vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_OUTPUT_MASK,
3778 							MSR_TYPE_RW, flag);
3779 	vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_CR3_MATCH,
3780 							MSR_TYPE_RW, flag);
3781 	for (i = 0; i < vmx->pt_desc.addr_range; i++) {
3782 		vmx_set_intercept_for_msr(msr_bitmap,
3783 			MSR_IA32_RTIT_ADDR0_A + i * 2, MSR_TYPE_RW, flag);
3784 		vmx_set_intercept_for_msr(msr_bitmap,
3785 			MSR_IA32_RTIT_ADDR0_B + i * 2, MSR_TYPE_RW, flag);
3786 	}
3787 }
3788 
3789 static bool vmx_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
3790 {
3791 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3792 	void *vapic_page;
3793 	u32 vppr;
3794 	int rvi;
3795 
3796 	if (WARN_ON_ONCE(!is_guest_mode(vcpu)) ||
3797 		!nested_cpu_has_vid(get_vmcs12(vcpu)) ||
3798 		WARN_ON_ONCE(!vmx->nested.virtual_apic_map.gfn))
3799 		return false;
3800 
3801 	rvi = vmx_get_rvi();
3802 
3803 	vapic_page = vmx->nested.virtual_apic_map.hva;
3804 	vppr = *((u32 *)(vapic_page + APIC_PROCPRI));
3805 
3806 	return ((rvi & 0xf0) > (vppr & 0xf0));
3807 }
3808 
3809 static inline bool kvm_vcpu_trigger_posted_interrupt(struct kvm_vcpu *vcpu,
3810 						     bool nested)
3811 {
3812 #ifdef CONFIG_SMP
3813 	int pi_vec = nested ? POSTED_INTR_NESTED_VECTOR : POSTED_INTR_VECTOR;
3814 
3815 	if (vcpu->mode == IN_GUEST_MODE) {
3816 		/*
3817 		 * The vector of interrupt to be delivered to vcpu had
3818 		 * been set in PIR before this function.
3819 		 *
3820 		 * Following cases will be reached in this block, and
3821 		 * we always send a notification event in all cases as
3822 		 * explained below.
3823 		 *
3824 		 * Case 1: vcpu keeps in non-root mode. Sending a
3825 		 * notification event posts the interrupt to vcpu.
3826 		 *
3827 		 * Case 2: vcpu exits to root mode and is still
3828 		 * runnable. PIR will be synced to vIRR before the
3829 		 * next vcpu entry. Sending a notification event in
3830 		 * this case has no effect, as vcpu is not in root
3831 		 * mode.
3832 		 *
3833 		 * Case 3: vcpu exits to root mode and is blocked.
3834 		 * vcpu_block() has already synced PIR to vIRR and
3835 		 * never blocks vcpu if vIRR is not cleared. Therefore,
3836 		 * a blocked vcpu here does not wait for any requested
3837 		 * interrupts in PIR, and sending a notification event
3838 		 * which has no effect is safe here.
3839 		 */
3840 
3841 		apic->send_IPI_mask(get_cpu_mask(vcpu->cpu), pi_vec);
3842 		return true;
3843 	}
3844 #endif
3845 	return false;
3846 }
3847 
3848 static int vmx_deliver_nested_posted_interrupt(struct kvm_vcpu *vcpu,
3849 						int vector)
3850 {
3851 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3852 
3853 	if (is_guest_mode(vcpu) &&
3854 	    vector == vmx->nested.posted_intr_nv) {
3855 		/*
3856 		 * If a posted intr is not recognized by hardware,
3857 		 * we will accomplish it in the next vmentry.
3858 		 */
3859 		vmx->nested.pi_pending = true;
3860 		kvm_make_request(KVM_REQ_EVENT, vcpu);
3861 		/* the PIR and ON have been set by L1. */
3862 		if (!kvm_vcpu_trigger_posted_interrupt(vcpu, true))
3863 			kvm_vcpu_kick(vcpu);
3864 		return 0;
3865 	}
3866 	return -1;
3867 }
3868 /*
3869  * Send interrupt to vcpu via posted interrupt way.
3870  * 1. If target vcpu is running(non-root mode), send posted interrupt
3871  * notification to vcpu and hardware will sync PIR to vIRR atomically.
3872  * 2. If target vcpu isn't running(root mode), kick it to pick up the
3873  * interrupt from PIR in next vmentry.
3874  */
3875 static int vmx_deliver_posted_interrupt(struct kvm_vcpu *vcpu, int vector)
3876 {
3877 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3878 	int r;
3879 
3880 	r = vmx_deliver_nested_posted_interrupt(vcpu, vector);
3881 	if (!r)
3882 		return 0;
3883 
3884 	if (!vcpu->arch.apicv_active)
3885 		return -1;
3886 
3887 	if (pi_test_and_set_pir(vector, &vmx->pi_desc))
3888 		return 0;
3889 
3890 	/* If a previous notification has sent the IPI, nothing to do.  */
3891 	if (pi_test_and_set_on(&vmx->pi_desc))
3892 		return 0;
3893 
3894 	if (!kvm_vcpu_trigger_posted_interrupt(vcpu, false))
3895 		kvm_vcpu_kick(vcpu);
3896 
3897 	return 0;
3898 }
3899 
3900 /*
3901  * Set up the vmcs's constant host-state fields, i.e., host-state fields that
3902  * will not change in the lifetime of the guest.
3903  * Note that host-state that does change is set elsewhere. E.g., host-state
3904  * that is set differently for each CPU is set in vmx_vcpu_load(), not here.
3905  */
3906 void vmx_set_constant_host_state(struct vcpu_vmx *vmx)
3907 {
3908 	u32 low32, high32;
3909 	unsigned long tmpl;
3910 	unsigned long cr0, cr3, cr4;
3911 
3912 	cr0 = read_cr0();
3913 	WARN_ON(cr0 & X86_CR0_TS);
3914 	vmcs_writel(HOST_CR0, cr0);  /* 22.2.3 */
3915 
3916 	/*
3917 	 * Save the most likely value for this task's CR3 in the VMCS.
3918 	 * We can't use __get_current_cr3_fast() because we're not atomic.
3919 	 */
3920 	cr3 = __read_cr3();
3921 	vmcs_writel(HOST_CR3, cr3);		/* 22.2.3  FIXME: shadow tables */
3922 	vmx->loaded_vmcs->host_state.cr3 = cr3;
3923 
3924 	/* Save the most likely value for this task's CR4 in the VMCS. */
3925 	cr4 = cr4_read_shadow();
3926 	vmcs_writel(HOST_CR4, cr4);			/* 22.2.3, 22.2.5 */
3927 	vmx->loaded_vmcs->host_state.cr4 = cr4;
3928 
3929 	vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS);  /* 22.2.4 */
3930 #ifdef CONFIG_X86_64
3931 	/*
3932 	 * Load null selectors, so we can avoid reloading them in
3933 	 * vmx_prepare_switch_to_host(), in case userspace uses
3934 	 * the null selectors too (the expected case).
3935 	 */
3936 	vmcs_write16(HOST_DS_SELECTOR, 0);
3937 	vmcs_write16(HOST_ES_SELECTOR, 0);
3938 #else
3939 	vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
3940 	vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
3941 #endif
3942 	vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
3943 	vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8);  /* 22.2.4 */
3944 
3945 	vmcs_writel(HOST_IDTR_BASE, host_idt_base);   /* 22.2.4 */
3946 
3947 	vmcs_writel(HOST_RIP, (unsigned long)vmx_vmexit); /* 22.2.5 */
3948 
3949 	rdmsr(MSR_IA32_SYSENTER_CS, low32, high32);
3950 	vmcs_write32(HOST_IA32_SYSENTER_CS, low32);
3951 	rdmsrl(MSR_IA32_SYSENTER_EIP, tmpl);
3952 	vmcs_writel(HOST_IA32_SYSENTER_EIP, tmpl);   /* 22.2.3 */
3953 
3954 	if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
3955 		rdmsr(MSR_IA32_CR_PAT, low32, high32);
3956 		vmcs_write64(HOST_IA32_PAT, low32 | ((u64) high32 << 32));
3957 	}
3958 
3959 	if (cpu_has_load_ia32_efer())
3960 		vmcs_write64(HOST_IA32_EFER, host_efer);
3961 }
3962 
3963 void set_cr4_guest_host_mask(struct vcpu_vmx *vmx)
3964 {
3965 	vmx->vcpu.arch.cr4_guest_owned_bits = KVM_CR4_GUEST_OWNED_BITS;
3966 	if (enable_ept)
3967 		vmx->vcpu.arch.cr4_guest_owned_bits |= X86_CR4_PGE;
3968 	if (is_guest_mode(&vmx->vcpu))
3969 		vmx->vcpu.arch.cr4_guest_owned_bits &=
3970 			~get_vmcs12(&vmx->vcpu)->cr4_guest_host_mask;
3971 	vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits);
3972 }
3973 
3974 u32 vmx_pin_based_exec_ctrl(struct vcpu_vmx *vmx)
3975 {
3976 	u32 pin_based_exec_ctrl = vmcs_config.pin_based_exec_ctrl;
3977 
3978 	if (!kvm_vcpu_apicv_active(&vmx->vcpu))
3979 		pin_based_exec_ctrl &= ~PIN_BASED_POSTED_INTR;
3980 
3981 	if (!enable_vnmi)
3982 		pin_based_exec_ctrl &= ~PIN_BASED_VIRTUAL_NMIS;
3983 
3984 	if (!enable_preemption_timer)
3985 		pin_based_exec_ctrl &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
3986 
3987 	return pin_based_exec_ctrl;
3988 }
3989 
3990 static void vmx_refresh_apicv_exec_ctrl(struct kvm_vcpu *vcpu)
3991 {
3992 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3993 
3994 	pin_controls_set(vmx, vmx_pin_based_exec_ctrl(vmx));
3995 	if (cpu_has_secondary_exec_ctrls()) {
3996 		if (kvm_vcpu_apicv_active(vcpu))
3997 			secondary_exec_controls_setbit(vmx,
3998 				      SECONDARY_EXEC_APIC_REGISTER_VIRT |
3999 				      SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
4000 		else
4001 			secondary_exec_controls_clearbit(vmx,
4002 					SECONDARY_EXEC_APIC_REGISTER_VIRT |
4003 					SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
4004 	}
4005 
4006 	if (cpu_has_vmx_msr_bitmap())
4007 		vmx_update_msr_bitmap(vcpu);
4008 }
4009 
4010 u32 vmx_exec_control(struct vcpu_vmx *vmx)
4011 {
4012 	u32 exec_control = vmcs_config.cpu_based_exec_ctrl;
4013 
4014 	if (vmx->vcpu.arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)
4015 		exec_control &= ~CPU_BASED_MOV_DR_EXITING;
4016 
4017 	if (!cpu_need_tpr_shadow(&vmx->vcpu)) {
4018 		exec_control &= ~CPU_BASED_TPR_SHADOW;
4019 #ifdef CONFIG_X86_64
4020 		exec_control |= CPU_BASED_CR8_STORE_EXITING |
4021 				CPU_BASED_CR8_LOAD_EXITING;
4022 #endif
4023 	}
4024 	if (!enable_ept)
4025 		exec_control |= CPU_BASED_CR3_STORE_EXITING |
4026 				CPU_BASED_CR3_LOAD_EXITING  |
4027 				CPU_BASED_INVLPG_EXITING;
4028 	if (kvm_mwait_in_guest(vmx->vcpu.kvm))
4029 		exec_control &= ~(CPU_BASED_MWAIT_EXITING |
4030 				CPU_BASED_MONITOR_EXITING);
4031 	if (kvm_hlt_in_guest(vmx->vcpu.kvm))
4032 		exec_control &= ~CPU_BASED_HLT_EXITING;
4033 	return exec_control;
4034 }
4035 
4036 
4037 static void vmx_compute_secondary_exec_control(struct vcpu_vmx *vmx)
4038 {
4039 	struct kvm_vcpu *vcpu = &vmx->vcpu;
4040 
4041 	u32 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
4042 
4043 	if (pt_mode == PT_MODE_SYSTEM)
4044 		exec_control &= ~(SECONDARY_EXEC_PT_USE_GPA | SECONDARY_EXEC_PT_CONCEAL_VMX);
4045 	if (!cpu_need_virtualize_apic_accesses(vcpu))
4046 		exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
4047 	if (vmx->vpid == 0)
4048 		exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
4049 	if (!enable_ept) {
4050 		exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
4051 		enable_unrestricted_guest = 0;
4052 	}
4053 	if (!enable_unrestricted_guest)
4054 		exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
4055 	if (kvm_pause_in_guest(vmx->vcpu.kvm))
4056 		exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
4057 	if (!kvm_vcpu_apicv_active(vcpu))
4058 		exec_control &= ~(SECONDARY_EXEC_APIC_REGISTER_VIRT |
4059 				  SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
4060 	exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
4061 
4062 	/* SECONDARY_EXEC_DESC is enabled/disabled on writes to CR4.UMIP,
4063 	 * in vmx_set_cr4.  */
4064 	exec_control &= ~SECONDARY_EXEC_DESC;
4065 
4066 	/* SECONDARY_EXEC_SHADOW_VMCS is enabled when L1 executes VMPTRLD
4067 	   (handle_vmptrld).
4068 	   We can NOT enable shadow_vmcs here because we don't have yet
4069 	   a current VMCS12
4070 	*/
4071 	exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
4072 
4073 	if (!enable_pml)
4074 		exec_control &= ~SECONDARY_EXEC_ENABLE_PML;
4075 
4076 	if (vmx_xsaves_supported()) {
4077 		/* Exposing XSAVES only when XSAVE is exposed */
4078 		bool xsaves_enabled =
4079 			boot_cpu_has(X86_FEATURE_XSAVE) &&
4080 			guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
4081 			guest_cpuid_has(vcpu, X86_FEATURE_XSAVES);
4082 
4083 		vcpu->arch.xsaves_enabled = xsaves_enabled;
4084 
4085 		if (!xsaves_enabled)
4086 			exec_control &= ~SECONDARY_EXEC_XSAVES;
4087 
4088 		if (nested) {
4089 			if (xsaves_enabled)
4090 				vmx->nested.msrs.secondary_ctls_high |=
4091 					SECONDARY_EXEC_XSAVES;
4092 			else
4093 				vmx->nested.msrs.secondary_ctls_high &=
4094 					~SECONDARY_EXEC_XSAVES;
4095 		}
4096 	}
4097 
4098 	if (vmx_rdtscp_supported()) {
4099 		bool rdtscp_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP);
4100 		if (!rdtscp_enabled)
4101 			exec_control &= ~SECONDARY_EXEC_RDTSCP;
4102 
4103 		if (nested) {
4104 			if (rdtscp_enabled)
4105 				vmx->nested.msrs.secondary_ctls_high |=
4106 					SECONDARY_EXEC_RDTSCP;
4107 			else
4108 				vmx->nested.msrs.secondary_ctls_high &=
4109 					~SECONDARY_EXEC_RDTSCP;
4110 		}
4111 	}
4112 
4113 	if (vmx_invpcid_supported()) {
4114 		/* Exposing INVPCID only when PCID is exposed */
4115 		bool invpcid_enabled =
4116 			guest_cpuid_has(vcpu, X86_FEATURE_INVPCID) &&
4117 			guest_cpuid_has(vcpu, X86_FEATURE_PCID);
4118 
4119 		if (!invpcid_enabled) {
4120 			exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID;
4121 			guest_cpuid_clear(vcpu, X86_FEATURE_INVPCID);
4122 		}
4123 
4124 		if (nested) {
4125 			if (invpcid_enabled)
4126 				vmx->nested.msrs.secondary_ctls_high |=
4127 					SECONDARY_EXEC_ENABLE_INVPCID;
4128 			else
4129 				vmx->nested.msrs.secondary_ctls_high &=
4130 					~SECONDARY_EXEC_ENABLE_INVPCID;
4131 		}
4132 	}
4133 
4134 	if (vmx_rdrand_supported()) {
4135 		bool rdrand_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDRAND);
4136 		if (rdrand_enabled)
4137 			exec_control &= ~SECONDARY_EXEC_RDRAND_EXITING;
4138 
4139 		if (nested) {
4140 			if (rdrand_enabled)
4141 				vmx->nested.msrs.secondary_ctls_high |=
4142 					SECONDARY_EXEC_RDRAND_EXITING;
4143 			else
4144 				vmx->nested.msrs.secondary_ctls_high &=
4145 					~SECONDARY_EXEC_RDRAND_EXITING;
4146 		}
4147 	}
4148 
4149 	if (vmx_rdseed_supported()) {
4150 		bool rdseed_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDSEED);
4151 		if (rdseed_enabled)
4152 			exec_control &= ~SECONDARY_EXEC_RDSEED_EXITING;
4153 
4154 		if (nested) {
4155 			if (rdseed_enabled)
4156 				vmx->nested.msrs.secondary_ctls_high |=
4157 					SECONDARY_EXEC_RDSEED_EXITING;
4158 			else
4159 				vmx->nested.msrs.secondary_ctls_high &=
4160 					~SECONDARY_EXEC_RDSEED_EXITING;
4161 		}
4162 	}
4163 
4164 	if (vmx_waitpkg_supported()) {
4165 		bool waitpkg_enabled =
4166 			guest_cpuid_has(vcpu, X86_FEATURE_WAITPKG);
4167 
4168 		if (!waitpkg_enabled)
4169 			exec_control &= ~SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE;
4170 
4171 		if (nested) {
4172 			if (waitpkg_enabled)
4173 				vmx->nested.msrs.secondary_ctls_high |=
4174 					SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE;
4175 			else
4176 				vmx->nested.msrs.secondary_ctls_high &=
4177 					~SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE;
4178 		}
4179 	}
4180 
4181 	vmx->secondary_exec_control = exec_control;
4182 }
4183 
4184 static void ept_set_mmio_spte_mask(void)
4185 {
4186 	/*
4187 	 * EPT Misconfigurations can be generated if the value of bits 2:0
4188 	 * of an EPT paging-structure entry is 110b (write/execute).
4189 	 */
4190 	kvm_mmu_set_mmio_spte_mask(VMX_EPT_RWX_MASK,
4191 				   VMX_EPT_MISCONFIG_WX_VALUE, 0);
4192 }
4193 
4194 #define VMX_XSS_EXIT_BITMAP 0
4195 
4196 /*
4197  * Noting that the initialization of Guest-state Area of VMCS is in
4198  * vmx_vcpu_reset().
4199  */
4200 static void init_vmcs(struct vcpu_vmx *vmx)
4201 {
4202 	if (nested)
4203 		nested_vmx_set_vmcs_shadowing_bitmap();
4204 
4205 	if (cpu_has_vmx_msr_bitmap())
4206 		vmcs_write64(MSR_BITMAP, __pa(vmx->vmcs01.msr_bitmap));
4207 
4208 	vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
4209 
4210 	/* Control */
4211 	pin_controls_set(vmx, vmx_pin_based_exec_ctrl(vmx));
4212 
4213 	exec_controls_set(vmx, vmx_exec_control(vmx));
4214 
4215 	if (cpu_has_secondary_exec_ctrls()) {
4216 		vmx_compute_secondary_exec_control(vmx);
4217 		secondary_exec_controls_set(vmx, vmx->secondary_exec_control);
4218 	}
4219 
4220 	if (kvm_vcpu_apicv_active(&vmx->vcpu)) {
4221 		vmcs_write64(EOI_EXIT_BITMAP0, 0);
4222 		vmcs_write64(EOI_EXIT_BITMAP1, 0);
4223 		vmcs_write64(EOI_EXIT_BITMAP2, 0);
4224 		vmcs_write64(EOI_EXIT_BITMAP3, 0);
4225 
4226 		vmcs_write16(GUEST_INTR_STATUS, 0);
4227 
4228 		vmcs_write16(POSTED_INTR_NV, POSTED_INTR_VECTOR);
4229 		vmcs_write64(POSTED_INTR_DESC_ADDR, __pa((&vmx->pi_desc)));
4230 	}
4231 
4232 	if (!kvm_pause_in_guest(vmx->vcpu.kvm)) {
4233 		vmcs_write32(PLE_GAP, ple_gap);
4234 		vmx->ple_window = ple_window;
4235 		vmx->ple_window_dirty = true;
4236 	}
4237 
4238 	vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
4239 	vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
4240 	vmcs_write32(CR3_TARGET_COUNT, 0);           /* 22.2.1 */
4241 
4242 	vmcs_write16(HOST_FS_SELECTOR, 0);            /* 22.2.4 */
4243 	vmcs_write16(HOST_GS_SELECTOR, 0);            /* 22.2.4 */
4244 	vmx_set_constant_host_state(vmx);
4245 	vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
4246 	vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
4247 
4248 	if (cpu_has_vmx_vmfunc())
4249 		vmcs_write64(VM_FUNCTION_CONTROL, 0);
4250 
4251 	vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
4252 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
4253 	vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host.val));
4254 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
4255 	vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest.val));
4256 
4257 	if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
4258 		vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
4259 
4260 	vm_exit_controls_set(vmx, vmx_vmexit_ctrl());
4261 
4262 	/* 22.2.1, 20.8.1 */
4263 	vm_entry_controls_set(vmx, vmx_vmentry_ctrl());
4264 
4265 	vmx->vcpu.arch.cr0_guest_owned_bits = X86_CR0_TS;
4266 	vmcs_writel(CR0_GUEST_HOST_MASK, ~X86_CR0_TS);
4267 
4268 	set_cr4_guest_host_mask(vmx);
4269 
4270 	if (vmx->vpid != 0)
4271 		vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
4272 
4273 	if (vmx_xsaves_supported())
4274 		vmcs_write64(XSS_EXIT_BITMAP, VMX_XSS_EXIT_BITMAP);
4275 
4276 	if (enable_pml) {
4277 		vmcs_write64(PML_ADDRESS, page_to_phys(vmx->pml_pg));
4278 		vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
4279 	}
4280 
4281 	if (cpu_has_vmx_encls_vmexit())
4282 		vmcs_write64(ENCLS_EXITING_BITMAP, -1ull);
4283 
4284 	if (pt_mode == PT_MODE_HOST_GUEST) {
4285 		memset(&vmx->pt_desc, 0, sizeof(vmx->pt_desc));
4286 		/* Bit[6~0] are forced to 1, writes are ignored. */
4287 		vmx->pt_desc.guest.output_mask = 0x7F;
4288 		vmcs_write64(GUEST_IA32_RTIT_CTL, 0);
4289 	}
4290 }
4291 
4292 static void vmx_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
4293 {
4294 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4295 	struct msr_data apic_base_msr;
4296 	u64 cr0;
4297 
4298 	vmx->rmode.vm86_active = 0;
4299 	vmx->spec_ctrl = 0;
4300 
4301 	vmx->msr_ia32_umwait_control = 0;
4302 
4303 	vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
4304 	vmx->hv_deadline_tsc = -1;
4305 	kvm_set_cr8(vcpu, 0);
4306 
4307 	if (!init_event) {
4308 		apic_base_msr.data = APIC_DEFAULT_PHYS_BASE |
4309 				     MSR_IA32_APICBASE_ENABLE;
4310 		if (kvm_vcpu_is_reset_bsp(vcpu))
4311 			apic_base_msr.data |= MSR_IA32_APICBASE_BSP;
4312 		apic_base_msr.host_initiated = true;
4313 		kvm_set_apic_base(vcpu, &apic_base_msr);
4314 	}
4315 
4316 	vmx_segment_cache_clear(vmx);
4317 
4318 	seg_setup(VCPU_SREG_CS);
4319 	vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
4320 	vmcs_writel(GUEST_CS_BASE, 0xffff0000ul);
4321 
4322 	seg_setup(VCPU_SREG_DS);
4323 	seg_setup(VCPU_SREG_ES);
4324 	seg_setup(VCPU_SREG_FS);
4325 	seg_setup(VCPU_SREG_GS);
4326 	seg_setup(VCPU_SREG_SS);
4327 
4328 	vmcs_write16(GUEST_TR_SELECTOR, 0);
4329 	vmcs_writel(GUEST_TR_BASE, 0);
4330 	vmcs_write32(GUEST_TR_LIMIT, 0xffff);
4331 	vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
4332 
4333 	vmcs_write16(GUEST_LDTR_SELECTOR, 0);
4334 	vmcs_writel(GUEST_LDTR_BASE, 0);
4335 	vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
4336 	vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
4337 
4338 	if (!init_event) {
4339 		vmcs_write32(GUEST_SYSENTER_CS, 0);
4340 		vmcs_writel(GUEST_SYSENTER_ESP, 0);
4341 		vmcs_writel(GUEST_SYSENTER_EIP, 0);
4342 		vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
4343 	}
4344 
4345 	kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
4346 	kvm_rip_write(vcpu, 0xfff0);
4347 
4348 	vmcs_writel(GUEST_GDTR_BASE, 0);
4349 	vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
4350 
4351 	vmcs_writel(GUEST_IDTR_BASE, 0);
4352 	vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
4353 
4354 	vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
4355 	vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
4356 	vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS, 0);
4357 	if (kvm_mpx_supported())
4358 		vmcs_write64(GUEST_BNDCFGS, 0);
4359 
4360 	setup_msrs(vmx);
4361 
4362 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);  /* 22.2.1 */
4363 
4364 	if (cpu_has_vmx_tpr_shadow() && !init_event) {
4365 		vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
4366 		if (cpu_need_tpr_shadow(vcpu))
4367 			vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
4368 				     __pa(vcpu->arch.apic->regs));
4369 		vmcs_write32(TPR_THRESHOLD, 0);
4370 	}
4371 
4372 	kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
4373 
4374 	cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
4375 	vmx->vcpu.arch.cr0 = cr0;
4376 	vmx_set_cr0(vcpu, cr0); /* enter rmode */
4377 	vmx_set_cr4(vcpu, 0);
4378 	vmx_set_efer(vcpu, 0);
4379 
4380 	update_exception_bitmap(vcpu);
4381 
4382 	vpid_sync_context(vmx->vpid);
4383 	if (init_event)
4384 		vmx_clear_hlt(vcpu);
4385 }
4386 
4387 static void enable_irq_window(struct kvm_vcpu *vcpu)
4388 {
4389 	exec_controls_setbit(to_vmx(vcpu), CPU_BASED_INTR_WINDOW_EXITING);
4390 }
4391 
4392 static void enable_nmi_window(struct kvm_vcpu *vcpu)
4393 {
4394 	if (!enable_vnmi ||
4395 	    vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_STI) {
4396 		enable_irq_window(vcpu);
4397 		return;
4398 	}
4399 
4400 	exec_controls_setbit(to_vmx(vcpu), CPU_BASED_NMI_WINDOW_EXITING);
4401 }
4402 
4403 static void vmx_inject_irq(struct kvm_vcpu *vcpu)
4404 {
4405 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4406 	uint32_t intr;
4407 	int irq = vcpu->arch.interrupt.nr;
4408 
4409 	trace_kvm_inj_virq(irq);
4410 
4411 	++vcpu->stat.irq_injections;
4412 	if (vmx->rmode.vm86_active) {
4413 		int inc_eip = 0;
4414 		if (vcpu->arch.interrupt.soft)
4415 			inc_eip = vcpu->arch.event_exit_inst_len;
4416 		kvm_inject_realmode_interrupt(vcpu, irq, inc_eip);
4417 		return;
4418 	}
4419 	intr = irq | INTR_INFO_VALID_MASK;
4420 	if (vcpu->arch.interrupt.soft) {
4421 		intr |= INTR_TYPE_SOFT_INTR;
4422 		vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
4423 			     vmx->vcpu.arch.event_exit_inst_len);
4424 	} else
4425 		intr |= INTR_TYPE_EXT_INTR;
4426 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
4427 
4428 	vmx_clear_hlt(vcpu);
4429 }
4430 
4431 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
4432 {
4433 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4434 
4435 	if (!enable_vnmi) {
4436 		/*
4437 		 * Tracking the NMI-blocked state in software is built upon
4438 		 * finding the next open IRQ window. This, in turn, depends on
4439 		 * well-behaving guests: They have to keep IRQs disabled at
4440 		 * least as long as the NMI handler runs. Otherwise we may
4441 		 * cause NMI nesting, maybe breaking the guest. But as this is
4442 		 * highly unlikely, we can live with the residual risk.
4443 		 */
4444 		vmx->loaded_vmcs->soft_vnmi_blocked = 1;
4445 		vmx->loaded_vmcs->vnmi_blocked_time = 0;
4446 	}
4447 
4448 	++vcpu->stat.nmi_injections;
4449 	vmx->loaded_vmcs->nmi_known_unmasked = false;
4450 
4451 	if (vmx->rmode.vm86_active) {
4452 		kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR, 0);
4453 		return;
4454 	}
4455 
4456 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
4457 			INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
4458 
4459 	vmx_clear_hlt(vcpu);
4460 }
4461 
4462 bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
4463 {
4464 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4465 	bool masked;
4466 
4467 	if (!enable_vnmi)
4468 		return vmx->loaded_vmcs->soft_vnmi_blocked;
4469 	if (vmx->loaded_vmcs->nmi_known_unmasked)
4470 		return false;
4471 	masked = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI;
4472 	vmx->loaded_vmcs->nmi_known_unmasked = !masked;
4473 	return masked;
4474 }
4475 
4476 void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
4477 {
4478 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4479 
4480 	if (!enable_vnmi) {
4481 		if (vmx->loaded_vmcs->soft_vnmi_blocked != masked) {
4482 			vmx->loaded_vmcs->soft_vnmi_blocked = masked;
4483 			vmx->loaded_vmcs->vnmi_blocked_time = 0;
4484 		}
4485 	} else {
4486 		vmx->loaded_vmcs->nmi_known_unmasked = !masked;
4487 		if (masked)
4488 			vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
4489 				      GUEST_INTR_STATE_NMI);
4490 		else
4491 			vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
4492 					GUEST_INTR_STATE_NMI);
4493 	}
4494 }
4495 
4496 static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
4497 {
4498 	if (to_vmx(vcpu)->nested.nested_run_pending)
4499 		return 0;
4500 
4501 	if (!enable_vnmi &&
4502 	    to_vmx(vcpu)->loaded_vmcs->soft_vnmi_blocked)
4503 		return 0;
4504 
4505 	return	!(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4506 		  (GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_STI
4507 		   | GUEST_INTR_STATE_NMI));
4508 }
4509 
4510 static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
4511 {
4512 	return (!to_vmx(vcpu)->nested.nested_run_pending &&
4513 		vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
4514 		!(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4515 			(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
4516 }
4517 
4518 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
4519 {
4520 	int ret;
4521 
4522 	if (enable_unrestricted_guest)
4523 		return 0;
4524 
4525 	mutex_lock(&kvm->slots_lock);
4526 	ret = __x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, addr,
4527 				      PAGE_SIZE * 3);
4528 	mutex_unlock(&kvm->slots_lock);
4529 
4530 	if (ret)
4531 		return ret;
4532 	to_kvm_vmx(kvm)->tss_addr = addr;
4533 	return init_rmode_tss(kvm);
4534 }
4535 
4536 static int vmx_set_identity_map_addr(struct kvm *kvm, u64 ident_addr)
4537 {
4538 	to_kvm_vmx(kvm)->ept_identity_map_addr = ident_addr;
4539 	return 0;
4540 }
4541 
4542 static bool rmode_exception(struct kvm_vcpu *vcpu, int vec)
4543 {
4544 	switch (vec) {
4545 	case BP_VECTOR:
4546 		/*
4547 		 * Update instruction length as we may reinject the exception
4548 		 * from user space while in guest debugging mode.
4549 		 */
4550 		to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
4551 			vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4552 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
4553 			return false;
4554 		/* fall through */
4555 	case DB_VECTOR:
4556 		if (vcpu->guest_debug &
4557 			(KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
4558 			return false;
4559 		/* fall through */
4560 	case DE_VECTOR:
4561 	case OF_VECTOR:
4562 	case BR_VECTOR:
4563 	case UD_VECTOR:
4564 	case DF_VECTOR:
4565 	case SS_VECTOR:
4566 	case GP_VECTOR:
4567 	case MF_VECTOR:
4568 		return true;
4569 	break;
4570 	}
4571 	return false;
4572 }
4573 
4574 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
4575 				  int vec, u32 err_code)
4576 {
4577 	/*
4578 	 * Instruction with address size override prefix opcode 0x67
4579 	 * Cause the #SS fault with 0 error code in VM86 mode.
4580 	 */
4581 	if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0) {
4582 		if (kvm_emulate_instruction(vcpu, 0)) {
4583 			if (vcpu->arch.halt_request) {
4584 				vcpu->arch.halt_request = 0;
4585 				return kvm_vcpu_halt(vcpu);
4586 			}
4587 			return 1;
4588 		}
4589 		return 0;
4590 	}
4591 
4592 	/*
4593 	 * Forward all other exceptions that are valid in real mode.
4594 	 * FIXME: Breaks guest debugging in real mode, needs to be fixed with
4595 	 *        the required debugging infrastructure rework.
4596 	 */
4597 	kvm_queue_exception(vcpu, vec);
4598 	return 1;
4599 }
4600 
4601 /*
4602  * Trigger machine check on the host. We assume all the MSRs are already set up
4603  * by the CPU and that we still run on the same CPU as the MCE occurred on.
4604  * We pass a fake environment to the machine check handler because we want
4605  * the guest to be always treated like user space, no matter what context
4606  * it used internally.
4607  */
4608 static void kvm_machine_check(void)
4609 {
4610 #if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
4611 	struct pt_regs regs = {
4612 		.cs = 3, /* Fake ring 3 no matter what the guest ran on */
4613 		.flags = X86_EFLAGS_IF,
4614 	};
4615 
4616 	do_machine_check(&regs, 0);
4617 #endif
4618 }
4619 
4620 static int handle_machine_check(struct kvm_vcpu *vcpu)
4621 {
4622 	/* handled by vmx_vcpu_run() */
4623 	return 1;
4624 }
4625 
4626 static int handle_exception_nmi(struct kvm_vcpu *vcpu)
4627 {
4628 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4629 	struct kvm_run *kvm_run = vcpu->run;
4630 	u32 intr_info, ex_no, error_code;
4631 	unsigned long cr2, rip, dr6;
4632 	u32 vect_info;
4633 
4634 	vect_info = vmx->idt_vectoring_info;
4635 	intr_info = vmx->exit_intr_info;
4636 
4637 	if (is_machine_check(intr_info) || is_nmi(intr_info))
4638 		return 1; /* handled by handle_exception_nmi_irqoff() */
4639 
4640 	if (is_invalid_opcode(intr_info))
4641 		return handle_ud(vcpu);
4642 
4643 	error_code = 0;
4644 	if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
4645 		error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
4646 
4647 	if (!vmx->rmode.vm86_active && is_gp_fault(intr_info)) {
4648 		WARN_ON_ONCE(!enable_vmware_backdoor);
4649 
4650 		/*
4651 		 * VMware backdoor emulation on #GP interception only handles
4652 		 * IN{S}, OUT{S}, and RDPMC, none of which generate a non-zero
4653 		 * error code on #GP.
4654 		 */
4655 		if (error_code) {
4656 			kvm_queue_exception_e(vcpu, GP_VECTOR, error_code);
4657 			return 1;
4658 		}
4659 		return kvm_emulate_instruction(vcpu, EMULTYPE_VMWARE_GP);
4660 	}
4661 
4662 	/*
4663 	 * The #PF with PFEC.RSVD = 1 indicates the guest is accessing
4664 	 * MMIO, it is better to report an internal error.
4665 	 * See the comments in vmx_handle_exit.
4666 	 */
4667 	if ((vect_info & VECTORING_INFO_VALID_MASK) &&
4668 	    !(is_page_fault(intr_info) && !(error_code & PFERR_RSVD_MASK))) {
4669 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4670 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
4671 		vcpu->run->internal.ndata = 3;
4672 		vcpu->run->internal.data[0] = vect_info;
4673 		vcpu->run->internal.data[1] = intr_info;
4674 		vcpu->run->internal.data[2] = error_code;
4675 		return 0;
4676 	}
4677 
4678 	if (is_page_fault(intr_info)) {
4679 		cr2 = vmcs_readl(EXIT_QUALIFICATION);
4680 		/* EPT won't cause page fault directly */
4681 		WARN_ON_ONCE(!vcpu->arch.apf.host_apf_reason && enable_ept);
4682 		return kvm_handle_page_fault(vcpu, error_code, cr2, NULL, 0);
4683 	}
4684 
4685 	ex_no = intr_info & INTR_INFO_VECTOR_MASK;
4686 
4687 	if (vmx->rmode.vm86_active && rmode_exception(vcpu, ex_no))
4688 		return handle_rmode_exception(vcpu, ex_no, error_code);
4689 
4690 	switch (ex_no) {
4691 	case AC_VECTOR:
4692 		kvm_queue_exception_e(vcpu, AC_VECTOR, error_code);
4693 		return 1;
4694 	case DB_VECTOR:
4695 		dr6 = vmcs_readl(EXIT_QUALIFICATION);
4696 		if (!(vcpu->guest_debug &
4697 		      (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
4698 			vcpu->arch.dr6 &= ~DR_TRAP_BITS;
4699 			vcpu->arch.dr6 |= dr6 | DR6_RTM;
4700 			if (is_icebp(intr_info))
4701 				WARN_ON(!skip_emulated_instruction(vcpu));
4702 
4703 			kvm_queue_exception(vcpu, DB_VECTOR);
4704 			return 1;
4705 		}
4706 		kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
4707 		kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
4708 		/* fall through */
4709 	case BP_VECTOR:
4710 		/*
4711 		 * Update instruction length as we may reinject #BP from
4712 		 * user space while in guest debugging mode. Reading it for
4713 		 * #DB as well causes no harm, it is not used in that case.
4714 		 */
4715 		vmx->vcpu.arch.event_exit_inst_len =
4716 			vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4717 		kvm_run->exit_reason = KVM_EXIT_DEBUG;
4718 		rip = kvm_rip_read(vcpu);
4719 		kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
4720 		kvm_run->debug.arch.exception = ex_no;
4721 		break;
4722 	default:
4723 		kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
4724 		kvm_run->ex.exception = ex_no;
4725 		kvm_run->ex.error_code = error_code;
4726 		break;
4727 	}
4728 	return 0;
4729 }
4730 
4731 static __always_inline int handle_external_interrupt(struct kvm_vcpu *vcpu)
4732 {
4733 	++vcpu->stat.irq_exits;
4734 	return 1;
4735 }
4736 
4737 static int handle_triple_fault(struct kvm_vcpu *vcpu)
4738 {
4739 	vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
4740 	vcpu->mmio_needed = 0;
4741 	return 0;
4742 }
4743 
4744 static int handle_io(struct kvm_vcpu *vcpu)
4745 {
4746 	unsigned long exit_qualification;
4747 	int size, in, string;
4748 	unsigned port;
4749 
4750 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4751 	string = (exit_qualification & 16) != 0;
4752 
4753 	++vcpu->stat.io_exits;
4754 
4755 	if (string)
4756 		return kvm_emulate_instruction(vcpu, 0);
4757 
4758 	port = exit_qualification >> 16;
4759 	size = (exit_qualification & 7) + 1;
4760 	in = (exit_qualification & 8) != 0;
4761 
4762 	return kvm_fast_pio(vcpu, size, port, in);
4763 }
4764 
4765 static void
4766 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
4767 {
4768 	/*
4769 	 * Patch in the VMCALL instruction:
4770 	 */
4771 	hypercall[0] = 0x0f;
4772 	hypercall[1] = 0x01;
4773 	hypercall[2] = 0xc1;
4774 }
4775 
4776 /* called to set cr0 as appropriate for a mov-to-cr0 exit. */
4777 static int handle_set_cr0(struct kvm_vcpu *vcpu, unsigned long val)
4778 {
4779 	if (is_guest_mode(vcpu)) {
4780 		struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4781 		unsigned long orig_val = val;
4782 
4783 		/*
4784 		 * We get here when L2 changed cr0 in a way that did not change
4785 		 * any of L1's shadowed bits (see nested_vmx_exit_handled_cr),
4786 		 * but did change L0 shadowed bits. So we first calculate the
4787 		 * effective cr0 value that L1 would like to write into the
4788 		 * hardware. It consists of the L2-owned bits from the new
4789 		 * value combined with the L1-owned bits from L1's guest_cr0.
4790 		 */
4791 		val = (val & ~vmcs12->cr0_guest_host_mask) |
4792 			(vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask);
4793 
4794 		if (!nested_guest_cr0_valid(vcpu, val))
4795 			return 1;
4796 
4797 		if (kvm_set_cr0(vcpu, val))
4798 			return 1;
4799 		vmcs_writel(CR0_READ_SHADOW, orig_val);
4800 		return 0;
4801 	} else {
4802 		if (to_vmx(vcpu)->nested.vmxon &&
4803 		    !nested_host_cr0_valid(vcpu, val))
4804 			return 1;
4805 
4806 		return kvm_set_cr0(vcpu, val);
4807 	}
4808 }
4809 
4810 static int handle_set_cr4(struct kvm_vcpu *vcpu, unsigned long val)
4811 {
4812 	if (is_guest_mode(vcpu)) {
4813 		struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4814 		unsigned long orig_val = val;
4815 
4816 		/* analogously to handle_set_cr0 */
4817 		val = (val & ~vmcs12->cr4_guest_host_mask) |
4818 			(vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask);
4819 		if (kvm_set_cr4(vcpu, val))
4820 			return 1;
4821 		vmcs_writel(CR4_READ_SHADOW, orig_val);
4822 		return 0;
4823 	} else
4824 		return kvm_set_cr4(vcpu, val);
4825 }
4826 
4827 static int handle_desc(struct kvm_vcpu *vcpu)
4828 {
4829 	WARN_ON(!(vcpu->arch.cr4 & X86_CR4_UMIP));
4830 	return kvm_emulate_instruction(vcpu, 0);
4831 }
4832 
4833 static int handle_cr(struct kvm_vcpu *vcpu)
4834 {
4835 	unsigned long exit_qualification, val;
4836 	int cr;
4837 	int reg;
4838 	int err;
4839 	int ret;
4840 
4841 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4842 	cr = exit_qualification & 15;
4843 	reg = (exit_qualification >> 8) & 15;
4844 	switch ((exit_qualification >> 4) & 3) {
4845 	case 0: /* mov to cr */
4846 		val = kvm_register_readl(vcpu, reg);
4847 		trace_kvm_cr_write(cr, val);
4848 		switch (cr) {
4849 		case 0:
4850 			err = handle_set_cr0(vcpu, val);
4851 			return kvm_complete_insn_gp(vcpu, err);
4852 		case 3:
4853 			WARN_ON_ONCE(enable_unrestricted_guest);
4854 			err = kvm_set_cr3(vcpu, val);
4855 			return kvm_complete_insn_gp(vcpu, err);
4856 		case 4:
4857 			err = handle_set_cr4(vcpu, val);
4858 			return kvm_complete_insn_gp(vcpu, err);
4859 		case 8: {
4860 				u8 cr8_prev = kvm_get_cr8(vcpu);
4861 				u8 cr8 = (u8)val;
4862 				err = kvm_set_cr8(vcpu, cr8);
4863 				ret = kvm_complete_insn_gp(vcpu, err);
4864 				if (lapic_in_kernel(vcpu))
4865 					return ret;
4866 				if (cr8_prev <= cr8)
4867 					return ret;
4868 				/*
4869 				 * TODO: we might be squashing a
4870 				 * KVM_GUESTDBG_SINGLESTEP-triggered
4871 				 * KVM_EXIT_DEBUG here.
4872 				 */
4873 				vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
4874 				return 0;
4875 			}
4876 		}
4877 		break;
4878 	case 2: /* clts */
4879 		WARN_ONCE(1, "Guest should always own CR0.TS");
4880 		vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
4881 		trace_kvm_cr_write(0, kvm_read_cr0(vcpu));
4882 		return kvm_skip_emulated_instruction(vcpu);
4883 	case 1: /*mov from cr*/
4884 		switch (cr) {
4885 		case 3:
4886 			WARN_ON_ONCE(enable_unrestricted_guest);
4887 			val = kvm_read_cr3(vcpu);
4888 			kvm_register_write(vcpu, reg, val);
4889 			trace_kvm_cr_read(cr, val);
4890 			return kvm_skip_emulated_instruction(vcpu);
4891 		case 8:
4892 			val = kvm_get_cr8(vcpu);
4893 			kvm_register_write(vcpu, reg, val);
4894 			trace_kvm_cr_read(cr, val);
4895 			return kvm_skip_emulated_instruction(vcpu);
4896 		}
4897 		break;
4898 	case 3: /* lmsw */
4899 		val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
4900 		trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val);
4901 		kvm_lmsw(vcpu, val);
4902 
4903 		return kvm_skip_emulated_instruction(vcpu);
4904 	default:
4905 		break;
4906 	}
4907 	vcpu->run->exit_reason = 0;
4908 	vcpu_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
4909 	       (int)(exit_qualification >> 4) & 3, cr);
4910 	return 0;
4911 }
4912 
4913 static int handle_dr(struct kvm_vcpu *vcpu)
4914 {
4915 	unsigned long exit_qualification;
4916 	int dr, dr7, reg;
4917 
4918 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4919 	dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
4920 
4921 	/* First, if DR does not exist, trigger UD */
4922 	if (!kvm_require_dr(vcpu, dr))
4923 		return 1;
4924 
4925 	/* Do not handle if the CPL > 0, will trigger GP on re-entry */
4926 	if (!kvm_require_cpl(vcpu, 0))
4927 		return 1;
4928 	dr7 = vmcs_readl(GUEST_DR7);
4929 	if (dr7 & DR7_GD) {
4930 		/*
4931 		 * As the vm-exit takes precedence over the debug trap, we
4932 		 * need to emulate the latter, either for the host or the
4933 		 * guest debugging itself.
4934 		 */
4935 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
4936 			vcpu->run->debug.arch.dr6 = vcpu->arch.dr6;
4937 			vcpu->run->debug.arch.dr7 = dr7;
4938 			vcpu->run->debug.arch.pc = kvm_get_linear_rip(vcpu);
4939 			vcpu->run->debug.arch.exception = DB_VECTOR;
4940 			vcpu->run->exit_reason = KVM_EXIT_DEBUG;
4941 			return 0;
4942 		} else {
4943 			vcpu->arch.dr6 &= ~DR_TRAP_BITS;
4944 			vcpu->arch.dr6 |= DR6_BD | DR6_RTM;
4945 			kvm_queue_exception(vcpu, DB_VECTOR);
4946 			return 1;
4947 		}
4948 	}
4949 
4950 	if (vcpu->guest_debug == 0) {
4951 		exec_controls_clearbit(to_vmx(vcpu), CPU_BASED_MOV_DR_EXITING);
4952 
4953 		/*
4954 		 * No more DR vmexits; force a reload of the debug registers
4955 		 * and reenter on this instruction.  The next vmexit will
4956 		 * retrieve the full state of the debug registers.
4957 		 */
4958 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
4959 		return 1;
4960 	}
4961 
4962 	reg = DEBUG_REG_ACCESS_REG(exit_qualification);
4963 	if (exit_qualification & TYPE_MOV_FROM_DR) {
4964 		unsigned long val;
4965 
4966 		if (kvm_get_dr(vcpu, dr, &val))
4967 			return 1;
4968 		kvm_register_write(vcpu, reg, val);
4969 	} else
4970 		if (kvm_set_dr(vcpu, dr, kvm_register_readl(vcpu, reg)))
4971 			return 1;
4972 
4973 	return kvm_skip_emulated_instruction(vcpu);
4974 }
4975 
4976 static u64 vmx_get_dr6(struct kvm_vcpu *vcpu)
4977 {
4978 	return vcpu->arch.dr6;
4979 }
4980 
4981 static void vmx_set_dr6(struct kvm_vcpu *vcpu, unsigned long val)
4982 {
4983 }
4984 
4985 static void vmx_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
4986 {
4987 	get_debugreg(vcpu->arch.db[0], 0);
4988 	get_debugreg(vcpu->arch.db[1], 1);
4989 	get_debugreg(vcpu->arch.db[2], 2);
4990 	get_debugreg(vcpu->arch.db[3], 3);
4991 	get_debugreg(vcpu->arch.dr6, 6);
4992 	vcpu->arch.dr7 = vmcs_readl(GUEST_DR7);
4993 
4994 	vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
4995 	exec_controls_setbit(to_vmx(vcpu), CPU_BASED_MOV_DR_EXITING);
4996 }
4997 
4998 static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
4999 {
5000 	vmcs_writel(GUEST_DR7, val);
5001 }
5002 
5003 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
5004 {
5005 	kvm_apic_update_ppr(vcpu);
5006 	return 1;
5007 }
5008 
5009 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
5010 {
5011 	exec_controls_clearbit(to_vmx(vcpu), CPU_BASED_INTR_WINDOW_EXITING);
5012 
5013 	kvm_make_request(KVM_REQ_EVENT, vcpu);
5014 
5015 	++vcpu->stat.irq_window_exits;
5016 	return 1;
5017 }
5018 
5019 static int handle_vmcall(struct kvm_vcpu *vcpu)
5020 {
5021 	return kvm_emulate_hypercall(vcpu);
5022 }
5023 
5024 static int handle_invd(struct kvm_vcpu *vcpu)
5025 {
5026 	return kvm_emulate_instruction(vcpu, 0);
5027 }
5028 
5029 static int handle_invlpg(struct kvm_vcpu *vcpu)
5030 {
5031 	unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5032 
5033 	kvm_mmu_invlpg(vcpu, exit_qualification);
5034 	return kvm_skip_emulated_instruction(vcpu);
5035 }
5036 
5037 static int handle_rdpmc(struct kvm_vcpu *vcpu)
5038 {
5039 	int err;
5040 
5041 	err = kvm_rdpmc(vcpu);
5042 	return kvm_complete_insn_gp(vcpu, err);
5043 }
5044 
5045 static int handle_wbinvd(struct kvm_vcpu *vcpu)
5046 {
5047 	return kvm_emulate_wbinvd(vcpu);
5048 }
5049 
5050 static int handle_xsetbv(struct kvm_vcpu *vcpu)
5051 {
5052 	u64 new_bv = kvm_read_edx_eax(vcpu);
5053 	u32 index = kvm_rcx_read(vcpu);
5054 
5055 	if (kvm_set_xcr(vcpu, index, new_bv) == 0)
5056 		return kvm_skip_emulated_instruction(vcpu);
5057 	return 1;
5058 }
5059 
5060 static int handle_apic_access(struct kvm_vcpu *vcpu)
5061 {
5062 	if (likely(fasteoi)) {
5063 		unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5064 		int access_type, offset;
5065 
5066 		access_type = exit_qualification & APIC_ACCESS_TYPE;
5067 		offset = exit_qualification & APIC_ACCESS_OFFSET;
5068 		/*
5069 		 * Sane guest uses MOV to write EOI, with written value
5070 		 * not cared. So make a short-circuit here by avoiding
5071 		 * heavy instruction emulation.
5072 		 */
5073 		if ((access_type == TYPE_LINEAR_APIC_INST_WRITE) &&
5074 		    (offset == APIC_EOI)) {
5075 			kvm_lapic_set_eoi(vcpu);
5076 			return kvm_skip_emulated_instruction(vcpu);
5077 		}
5078 	}
5079 	return kvm_emulate_instruction(vcpu, 0);
5080 }
5081 
5082 static int handle_apic_eoi_induced(struct kvm_vcpu *vcpu)
5083 {
5084 	unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5085 	int vector = exit_qualification & 0xff;
5086 
5087 	/* EOI-induced VM exit is trap-like and thus no need to adjust IP */
5088 	kvm_apic_set_eoi_accelerated(vcpu, vector);
5089 	return 1;
5090 }
5091 
5092 static int handle_apic_write(struct kvm_vcpu *vcpu)
5093 {
5094 	unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5095 	u32 offset = exit_qualification & 0xfff;
5096 
5097 	/* APIC-write VM exit is trap-like and thus no need to adjust IP */
5098 	kvm_apic_write_nodecode(vcpu, offset);
5099 	return 1;
5100 }
5101 
5102 static int handle_task_switch(struct kvm_vcpu *vcpu)
5103 {
5104 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5105 	unsigned long exit_qualification;
5106 	bool has_error_code = false;
5107 	u32 error_code = 0;
5108 	u16 tss_selector;
5109 	int reason, type, idt_v, idt_index;
5110 
5111 	idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
5112 	idt_index = (vmx->idt_vectoring_info & VECTORING_INFO_VECTOR_MASK);
5113 	type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
5114 
5115 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5116 
5117 	reason = (u32)exit_qualification >> 30;
5118 	if (reason == TASK_SWITCH_GATE && idt_v) {
5119 		switch (type) {
5120 		case INTR_TYPE_NMI_INTR:
5121 			vcpu->arch.nmi_injected = false;
5122 			vmx_set_nmi_mask(vcpu, true);
5123 			break;
5124 		case INTR_TYPE_EXT_INTR:
5125 		case INTR_TYPE_SOFT_INTR:
5126 			kvm_clear_interrupt_queue(vcpu);
5127 			break;
5128 		case INTR_TYPE_HARD_EXCEPTION:
5129 			if (vmx->idt_vectoring_info &
5130 			    VECTORING_INFO_DELIVER_CODE_MASK) {
5131 				has_error_code = true;
5132 				error_code =
5133 					vmcs_read32(IDT_VECTORING_ERROR_CODE);
5134 			}
5135 			/* fall through */
5136 		case INTR_TYPE_SOFT_EXCEPTION:
5137 			kvm_clear_exception_queue(vcpu);
5138 			break;
5139 		default:
5140 			break;
5141 		}
5142 	}
5143 	tss_selector = exit_qualification;
5144 
5145 	if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
5146 		       type != INTR_TYPE_EXT_INTR &&
5147 		       type != INTR_TYPE_NMI_INTR))
5148 		WARN_ON(!skip_emulated_instruction(vcpu));
5149 
5150 	/*
5151 	 * TODO: What about debug traps on tss switch?
5152 	 *       Are we supposed to inject them and update dr6?
5153 	 */
5154 	return kvm_task_switch(vcpu, tss_selector,
5155 			       type == INTR_TYPE_SOFT_INTR ? idt_index : -1,
5156 			       reason, has_error_code, error_code);
5157 }
5158 
5159 static int handle_ept_violation(struct kvm_vcpu *vcpu)
5160 {
5161 	unsigned long exit_qualification;
5162 	gpa_t gpa;
5163 	u64 error_code;
5164 
5165 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5166 
5167 	/*
5168 	 * EPT violation happened while executing iret from NMI,
5169 	 * "blocked by NMI" bit has to be set before next VM entry.
5170 	 * There are errata that may cause this bit to not be set:
5171 	 * AAK134, BY25.
5172 	 */
5173 	if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
5174 			enable_vnmi &&
5175 			(exit_qualification & INTR_INFO_UNBLOCK_NMI))
5176 		vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO, GUEST_INTR_STATE_NMI);
5177 
5178 	gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5179 	trace_kvm_page_fault(gpa, exit_qualification);
5180 
5181 	/* Is it a read fault? */
5182 	error_code = (exit_qualification & EPT_VIOLATION_ACC_READ)
5183 		     ? PFERR_USER_MASK : 0;
5184 	/* Is it a write fault? */
5185 	error_code |= (exit_qualification & EPT_VIOLATION_ACC_WRITE)
5186 		      ? PFERR_WRITE_MASK : 0;
5187 	/* Is it a fetch fault? */
5188 	error_code |= (exit_qualification & EPT_VIOLATION_ACC_INSTR)
5189 		      ? PFERR_FETCH_MASK : 0;
5190 	/* ept page table entry is present? */
5191 	error_code |= (exit_qualification &
5192 		       (EPT_VIOLATION_READABLE | EPT_VIOLATION_WRITABLE |
5193 			EPT_VIOLATION_EXECUTABLE))
5194 		      ? PFERR_PRESENT_MASK : 0;
5195 
5196 	error_code |= (exit_qualification & 0x100) != 0 ?
5197 	       PFERR_GUEST_FINAL_MASK : PFERR_GUEST_PAGE_MASK;
5198 
5199 	vcpu->arch.exit_qualification = exit_qualification;
5200 	return kvm_mmu_page_fault(vcpu, gpa, error_code, NULL, 0);
5201 }
5202 
5203 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
5204 {
5205 	gpa_t gpa;
5206 
5207 	/*
5208 	 * A nested guest cannot optimize MMIO vmexits, because we have an
5209 	 * nGPA here instead of the required GPA.
5210 	 */
5211 	gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5212 	if (!is_guest_mode(vcpu) &&
5213 	    !kvm_io_bus_write(vcpu, KVM_FAST_MMIO_BUS, gpa, 0, NULL)) {
5214 		trace_kvm_fast_mmio(gpa);
5215 		return kvm_skip_emulated_instruction(vcpu);
5216 	}
5217 
5218 	return kvm_mmu_page_fault(vcpu, gpa, PFERR_RSVD_MASK, NULL, 0);
5219 }
5220 
5221 static int handle_nmi_window(struct kvm_vcpu *vcpu)
5222 {
5223 	WARN_ON_ONCE(!enable_vnmi);
5224 	exec_controls_clearbit(to_vmx(vcpu), CPU_BASED_NMI_WINDOW_EXITING);
5225 	++vcpu->stat.nmi_window_exits;
5226 	kvm_make_request(KVM_REQ_EVENT, vcpu);
5227 
5228 	return 1;
5229 }
5230 
5231 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
5232 {
5233 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5234 	bool intr_window_requested;
5235 	unsigned count = 130;
5236 
5237 	/*
5238 	 * We should never reach the point where we are emulating L2
5239 	 * due to invalid guest state as that means we incorrectly
5240 	 * allowed a nested VMEntry with an invalid vmcs12.
5241 	 */
5242 	WARN_ON_ONCE(vmx->emulation_required && vmx->nested.nested_run_pending);
5243 
5244 	intr_window_requested = exec_controls_get(vmx) &
5245 				CPU_BASED_INTR_WINDOW_EXITING;
5246 
5247 	while (vmx->emulation_required && count-- != 0) {
5248 		if (intr_window_requested && vmx_interrupt_allowed(vcpu))
5249 			return handle_interrupt_window(&vmx->vcpu);
5250 
5251 		if (kvm_test_request(KVM_REQ_EVENT, vcpu))
5252 			return 1;
5253 
5254 		if (!kvm_emulate_instruction(vcpu, 0))
5255 			return 0;
5256 
5257 		if (vmx->emulation_required && !vmx->rmode.vm86_active &&
5258 		    vcpu->arch.exception.pending) {
5259 			vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5260 			vcpu->run->internal.suberror =
5261 						KVM_INTERNAL_ERROR_EMULATION;
5262 			vcpu->run->internal.ndata = 0;
5263 			return 0;
5264 		}
5265 
5266 		if (vcpu->arch.halt_request) {
5267 			vcpu->arch.halt_request = 0;
5268 			return kvm_vcpu_halt(vcpu);
5269 		}
5270 
5271 		/*
5272 		 * Note, return 1 and not 0, vcpu_run() is responsible for
5273 		 * morphing the pending signal into the proper return code.
5274 		 */
5275 		if (signal_pending(current))
5276 			return 1;
5277 
5278 		if (need_resched())
5279 			schedule();
5280 	}
5281 
5282 	return 1;
5283 }
5284 
5285 static void grow_ple_window(struct kvm_vcpu *vcpu)
5286 {
5287 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5288 	unsigned int old = vmx->ple_window;
5289 
5290 	vmx->ple_window = __grow_ple_window(old, ple_window,
5291 					    ple_window_grow,
5292 					    ple_window_max);
5293 
5294 	if (vmx->ple_window != old) {
5295 		vmx->ple_window_dirty = true;
5296 		trace_kvm_ple_window_update(vcpu->vcpu_id,
5297 					    vmx->ple_window, old);
5298 	}
5299 }
5300 
5301 static void shrink_ple_window(struct kvm_vcpu *vcpu)
5302 {
5303 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5304 	unsigned int old = vmx->ple_window;
5305 
5306 	vmx->ple_window = __shrink_ple_window(old, ple_window,
5307 					      ple_window_shrink,
5308 					      ple_window);
5309 
5310 	if (vmx->ple_window != old) {
5311 		vmx->ple_window_dirty = true;
5312 		trace_kvm_ple_window_update(vcpu->vcpu_id,
5313 					    vmx->ple_window, old);
5314 	}
5315 }
5316 
5317 /*
5318  * Handler for POSTED_INTERRUPT_WAKEUP_VECTOR.
5319  */
5320 static void wakeup_handler(void)
5321 {
5322 	struct kvm_vcpu *vcpu;
5323 	int cpu = smp_processor_id();
5324 
5325 	spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
5326 	list_for_each_entry(vcpu, &per_cpu(blocked_vcpu_on_cpu, cpu),
5327 			blocked_vcpu_list) {
5328 		struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
5329 
5330 		if (pi_test_on(pi_desc) == 1)
5331 			kvm_vcpu_kick(vcpu);
5332 	}
5333 	spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
5334 }
5335 
5336 static void vmx_enable_tdp(void)
5337 {
5338 	kvm_mmu_set_mask_ptes(VMX_EPT_READABLE_MASK,
5339 		enable_ept_ad_bits ? VMX_EPT_ACCESS_BIT : 0ull,
5340 		enable_ept_ad_bits ? VMX_EPT_DIRTY_BIT : 0ull,
5341 		0ull, VMX_EPT_EXECUTABLE_MASK,
5342 		cpu_has_vmx_ept_execute_only() ? 0ull : VMX_EPT_READABLE_MASK,
5343 		VMX_EPT_RWX_MASK, 0ull);
5344 
5345 	ept_set_mmio_spte_mask();
5346 	kvm_enable_tdp();
5347 }
5348 
5349 /*
5350  * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
5351  * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
5352  */
5353 static int handle_pause(struct kvm_vcpu *vcpu)
5354 {
5355 	if (!kvm_pause_in_guest(vcpu->kvm))
5356 		grow_ple_window(vcpu);
5357 
5358 	/*
5359 	 * Intel sdm vol3 ch-25.1.3 says: The "PAUSE-loop exiting"
5360 	 * VM-execution control is ignored if CPL > 0. OTOH, KVM
5361 	 * never set PAUSE_EXITING and just set PLE if supported,
5362 	 * so the vcpu must be CPL=0 if it gets a PAUSE exit.
5363 	 */
5364 	kvm_vcpu_on_spin(vcpu, true);
5365 	return kvm_skip_emulated_instruction(vcpu);
5366 }
5367 
5368 static int handle_nop(struct kvm_vcpu *vcpu)
5369 {
5370 	return kvm_skip_emulated_instruction(vcpu);
5371 }
5372 
5373 static int handle_mwait(struct kvm_vcpu *vcpu)
5374 {
5375 	printk_once(KERN_WARNING "kvm: MWAIT instruction emulated as NOP!\n");
5376 	return handle_nop(vcpu);
5377 }
5378 
5379 static int handle_invalid_op(struct kvm_vcpu *vcpu)
5380 {
5381 	kvm_queue_exception(vcpu, UD_VECTOR);
5382 	return 1;
5383 }
5384 
5385 static int handle_monitor_trap(struct kvm_vcpu *vcpu)
5386 {
5387 	return 1;
5388 }
5389 
5390 static int handle_monitor(struct kvm_vcpu *vcpu)
5391 {
5392 	printk_once(KERN_WARNING "kvm: MONITOR instruction emulated as NOP!\n");
5393 	return handle_nop(vcpu);
5394 }
5395 
5396 static int handle_invpcid(struct kvm_vcpu *vcpu)
5397 {
5398 	u32 vmx_instruction_info;
5399 	unsigned long type;
5400 	bool pcid_enabled;
5401 	gva_t gva;
5402 	struct x86_exception e;
5403 	unsigned i;
5404 	unsigned long roots_to_free = 0;
5405 	struct {
5406 		u64 pcid;
5407 		u64 gla;
5408 	} operand;
5409 
5410 	if (!guest_cpuid_has(vcpu, X86_FEATURE_INVPCID)) {
5411 		kvm_queue_exception(vcpu, UD_VECTOR);
5412 		return 1;
5413 	}
5414 
5415 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5416 	type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
5417 
5418 	if (type > 3) {
5419 		kvm_inject_gp(vcpu, 0);
5420 		return 1;
5421 	}
5422 
5423 	/* According to the Intel instruction reference, the memory operand
5424 	 * is read even if it isn't needed (e.g., for type==all)
5425 	 */
5426 	if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
5427 				vmx_instruction_info, false,
5428 				sizeof(operand), &gva))
5429 		return 1;
5430 
5431 	if (kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e)) {
5432 		kvm_inject_page_fault(vcpu, &e);
5433 		return 1;
5434 	}
5435 
5436 	if (operand.pcid >> 12 != 0) {
5437 		kvm_inject_gp(vcpu, 0);
5438 		return 1;
5439 	}
5440 
5441 	pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
5442 
5443 	switch (type) {
5444 	case INVPCID_TYPE_INDIV_ADDR:
5445 		if ((!pcid_enabled && (operand.pcid != 0)) ||
5446 		    is_noncanonical_address(operand.gla, vcpu)) {
5447 			kvm_inject_gp(vcpu, 0);
5448 			return 1;
5449 		}
5450 		kvm_mmu_invpcid_gva(vcpu, operand.gla, operand.pcid);
5451 		return kvm_skip_emulated_instruction(vcpu);
5452 
5453 	case INVPCID_TYPE_SINGLE_CTXT:
5454 		if (!pcid_enabled && (operand.pcid != 0)) {
5455 			kvm_inject_gp(vcpu, 0);
5456 			return 1;
5457 		}
5458 
5459 		if (kvm_get_active_pcid(vcpu) == operand.pcid) {
5460 			kvm_mmu_sync_roots(vcpu);
5461 			kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
5462 		}
5463 
5464 		for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++)
5465 			if (kvm_get_pcid(vcpu, vcpu->arch.mmu->prev_roots[i].cr3)
5466 			    == operand.pcid)
5467 				roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i);
5468 
5469 		kvm_mmu_free_roots(vcpu, vcpu->arch.mmu, roots_to_free);
5470 		/*
5471 		 * If neither the current cr3 nor any of the prev_roots use the
5472 		 * given PCID, then nothing needs to be done here because a
5473 		 * resync will happen anyway before switching to any other CR3.
5474 		 */
5475 
5476 		return kvm_skip_emulated_instruction(vcpu);
5477 
5478 	case INVPCID_TYPE_ALL_NON_GLOBAL:
5479 		/*
5480 		 * Currently, KVM doesn't mark global entries in the shadow
5481 		 * page tables, so a non-global flush just degenerates to a
5482 		 * global flush. If needed, we could optimize this later by
5483 		 * keeping track of global entries in shadow page tables.
5484 		 */
5485 
5486 		/* fall-through */
5487 	case INVPCID_TYPE_ALL_INCL_GLOBAL:
5488 		kvm_mmu_unload(vcpu);
5489 		return kvm_skip_emulated_instruction(vcpu);
5490 
5491 	default:
5492 		BUG(); /* We have already checked above that type <= 3 */
5493 	}
5494 }
5495 
5496 static int handle_pml_full(struct kvm_vcpu *vcpu)
5497 {
5498 	unsigned long exit_qualification;
5499 
5500 	trace_kvm_pml_full(vcpu->vcpu_id);
5501 
5502 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5503 
5504 	/*
5505 	 * PML buffer FULL happened while executing iret from NMI,
5506 	 * "blocked by NMI" bit has to be set before next VM entry.
5507 	 */
5508 	if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
5509 			enable_vnmi &&
5510 			(exit_qualification & INTR_INFO_UNBLOCK_NMI))
5511 		vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
5512 				GUEST_INTR_STATE_NMI);
5513 
5514 	/*
5515 	 * PML buffer already flushed at beginning of VMEXIT. Nothing to do
5516 	 * here.., and there's no userspace involvement needed for PML.
5517 	 */
5518 	return 1;
5519 }
5520 
5521 static int handle_preemption_timer(struct kvm_vcpu *vcpu)
5522 {
5523 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5524 
5525 	if (!vmx->req_immediate_exit &&
5526 	    !unlikely(vmx->loaded_vmcs->hv_timer_soft_disabled))
5527 		kvm_lapic_expired_hv_timer(vcpu);
5528 
5529 	return 1;
5530 }
5531 
5532 /*
5533  * When nested=0, all VMX instruction VM Exits filter here.  The handlers
5534  * are overwritten by nested_vmx_setup() when nested=1.
5535  */
5536 static int handle_vmx_instruction(struct kvm_vcpu *vcpu)
5537 {
5538 	kvm_queue_exception(vcpu, UD_VECTOR);
5539 	return 1;
5540 }
5541 
5542 static int handle_encls(struct kvm_vcpu *vcpu)
5543 {
5544 	/*
5545 	 * SGX virtualization is not yet supported.  There is no software
5546 	 * enable bit for SGX, so we have to trap ENCLS and inject a #UD
5547 	 * to prevent the guest from executing ENCLS.
5548 	 */
5549 	kvm_queue_exception(vcpu, UD_VECTOR);
5550 	return 1;
5551 }
5552 
5553 /*
5554  * The exit handlers return 1 if the exit was handled fully and guest execution
5555  * may resume.  Otherwise they set the kvm_run parameter to indicate what needs
5556  * to be done to userspace and return 0.
5557  */
5558 static int (*kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
5559 	[EXIT_REASON_EXCEPTION_NMI]           = handle_exception_nmi,
5560 	[EXIT_REASON_EXTERNAL_INTERRUPT]      = handle_external_interrupt,
5561 	[EXIT_REASON_TRIPLE_FAULT]            = handle_triple_fault,
5562 	[EXIT_REASON_NMI_WINDOW]	      = handle_nmi_window,
5563 	[EXIT_REASON_IO_INSTRUCTION]          = handle_io,
5564 	[EXIT_REASON_CR_ACCESS]               = handle_cr,
5565 	[EXIT_REASON_DR_ACCESS]               = handle_dr,
5566 	[EXIT_REASON_CPUID]                   = kvm_emulate_cpuid,
5567 	[EXIT_REASON_MSR_READ]                = kvm_emulate_rdmsr,
5568 	[EXIT_REASON_MSR_WRITE]               = kvm_emulate_wrmsr,
5569 	[EXIT_REASON_INTERRUPT_WINDOW]        = handle_interrupt_window,
5570 	[EXIT_REASON_HLT]                     = kvm_emulate_halt,
5571 	[EXIT_REASON_INVD]		      = handle_invd,
5572 	[EXIT_REASON_INVLPG]		      = handle_invlpg,
5573 	[EXIT_REASON_RDPMC]                   = handle_rdpmc,
5574 	[EXIT_REASON_VMCALL]                  = handle_vmcall,
5575 	[EXIT_REASON_VMCLEAR]		      = handle_vmx_instruction,
5576 	[EXIT_REASON_VMLAUNCH]		      = handle_vmx_instruction,
5577 	[EXIT_REASON_VMPTRLD]		      = handle_vmx_instruction,
5578 	[EXIT_REASON_VMPTRST]		      = handle_vmx_instruction,
5579 	[EXIT_REASON_VMREAD]		      = handle_vmx_instruction,
5580 	[EXIT_REASON_VMRESUME]		      = handle_vmx_instruction,
5581 	[EXIT_REASON_VMWRITE]		      = handle_vmx_instruction,
5582 	[EXIT_REASON_VMOFF]		      = handle_vmx_instruction,
5583 	[EXIT_REASON_VMON]		      = handle_vmx_instruction,
5584 	[EXIT_REASON_TPR_BELOW_THRESHOLD]     = handle_tpr_below_threshold,
5585 	[EXIT_REASON_APIC_ACCESS]             = handle_apic_access,
5586 	[EXIT_REASON_APIC_WRITE]              = handle_apic_write,
5587 	[EXIT_REASON_EOI_INDUCED]             = handle_apic_eoi_induced,
5588 	[EXIT_REASON_WBINVD]                  = handle_wbinvd,
5589 	[EXIT_REASON_XSETBV]                  = handle_xsetbv,
5590 	[EXIT_REASON_TASK_SWITCH]             = handle_task_switch,
5591 	[EXIT_REASON_MCE_DURING_VMENTRY]      = handle_machine_check,
5592 	[EXIT_REASON_GDTR_IDTR]		      = handle_desc,
5593 	[EXIT_REASON_LDTR_TR]		      = handle_desc,
5594 	[EXIT_REASON_EPT_VIOLATION]	      = handle_ept_violation,
5595 	[EXIT_REASON_EPT_MISCONFIG]           = handle_ept_misconfig,
5596 	[EXIT_REASON_PAUSE_INSTRUCTION]       = handle_pause,
5597 	[EXIT_REASON_MWAIT_INSTRUCTION]	      = handle_mwait,
5598 	[EXIT_REASON_MONITOR_TRAP_FLAG]       = handle_monitor_trap,
5599 	[EXIT_REASON_MONITOR_INSTRUCTION]     = handle_monitor,
5600 	[EXIT_REASON_INVEPT]                  = handle_vmx_instruction,
5601 	[EXIT_REASON_INVVPID]                 = handle_vmx_instruction,
5602 	[EXIT_REASON_RDRAND]                  = handle_invalid_op,
5603 	[EXIT_REASON_RDSEED]                  = handle_invalid_op,
5604 	[EXIT_REASON_PML_FULL]		      = handle_pml_full,
5605 	[EXIT_REASON_INVPCID]                 = handle_invpcid,
5606 	[EXIT_REASON_VMFUNC]		      = handle_vmx_instruction,
5607 	[EXIT_REASON_PREEMPTION_TIMER]	      = handle_preemption_timer,
5608 	[EXIT_REASON_ENCLS]		      = handle_encls,
5609 };
5610 
5611 static const int kvm_vmx_max_exit_handlers =
5612 	ARRAY_SIZE(kvm_vmx_exit_handlers);
5613 
5614 static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
5615 {
5616 	*info1 = vmcs_readl(EXIT_QUALIFICATION);
5617 	*info2 = vmcs_read32(VM_EXIT_INTR_INFO);
5618 }
5619 
5620 static void vmx_destroy_pml_buffer(struct vcpu_vmx *vmx)
5621 {
5622 	if (vmx->pml_pg) {
5623 		__free_page(vmx->pml_pg);
5624 		vmx->pml_pg = NULL;
5625 	}
5626 }
5627 
5628 static void vmx_flush_pml_buffer(struct kvm_vcpu *vcpu)
5629 {
5630 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5631 	u64 *pml_buf;
5632 	u16 pml_idx;
5633 
5634 	pml_idx = vmcs_read16(GUEST_PML_INDEX);
5635 
5636 	/* Do nothing if PML buffer is empty */
5637 	if (pml_idx == (PML_ENTITY_NUM - 1))
5638 		return;
5639 
5640 	/* PML index always points to next available PML buffer entity */
5641 	if (pml_idx >= PML_ENTITY_NUM)
5642 		pml_idx = 0;
5643 	else
5644 		pml_idx++;
5645 
5646 	pml_buf = page_address(vmx->pml_pg);
5647 	for (; pml_idx < PML_ENTITY_NUM; pml_idx++) {
5648 		u64 gpa;
5649 
5650 		gpa = pml_buf[pml_idx];
5651 		WARN_ON(gpa & (PAGE_SIZE - 1));
5652 		kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
5653 	}
5654 
5655 	/* reset PML index */
5656 	vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
5657 }
5658 
5659 /*
5660  * Flush all vcpus' PML buffer and update logged GPAs to dirty_bitmap.
5661  * Called before reporting dirty_bitmap to userspace.
5662  */
5663 static void kvm_flush_pml_buffers(struct kvm *kvm)
5664 {
5665 	int i;
5666 	struct kvm_vcpu *vcpu;
5667 	/*
5668 	 * We only need to kick vcpu out of guest mode here, as PML buffer
5669 	 * is flushed at beginning of all VMEXITs, and it's obvious that only
5670 	 * vcpus running in guest are possible to have unflushed GPAs in PML
5671 	 * buffer.
5672 	 */
5673 	kvm_for_each_vcpu(i, vcpu, kvm)
5674 		kvm_vcpu_kick(vcpu);
5675 }
5676 
5677 static void vmx_dump_sel(char *name, uint32_t sel)
5678 {
5679 	pr_err("%s sel=0x%04x, attr=0x%05x, limit=0x%08x, base=0x%016lx\n",
5680 	       name, vmcs_read16(sel),
5681 	       vmcs_read32(sel + GUEST_ES_AR_BYTES - GUEST_ES_SELECTOR),
5682 	       vmcs_read32(sel + GUEST_ES_LIMIT - GUEST_ES_SELECTOR),
5683 	       vmcs_readl(sel + GUEST_ES_BASE - GUEST_ES_SELECTOR));
5684 }
5685 
5686 static void vmx_dump_dtsel(char *name, uint32_t limit)
5687 {
5688 	pr_err("%s                           limit=0x%08x, base=0x%016lx\n",
5689 	       name, vmcs_read32(limit),
5690 	       vmcs_readl(limit + GUEST_GDTR_BASE - GUEST_GDTR_LIMIT));
5691 }
5692 
5693 void dump_vmcs(void)
5694 {
5695 	u32 vmentry_ctl, vmexit_ctl;
5696 	u32 cpu_based_exec_ctrl, pin_based_exec_ctrl, secondary_exec_control;
5697 	unsigned long cr4;
5698 	u64 efer;
5699 	int i, n;
5700 
5701 	if (!dump_invalid_vmcs) {
5702 		pr_warn_ratelimited("set kvm_intel.dump_invalid_vmcs=1 to dump internal KVM state.\n");
5703 		return;
5704 	}
5705 
5706 	vmentry_ctl = vmcs_read32(VM_ENTRY_CONTROLS);
5707 	vmexit_ctl = vmcs_read32(VM_EXIT_CONTROLS);
5708 	cpu_based_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5709 	pin_based_exec_ctrl = vmcs_read32(PIN_BASED_VM_EXEC_CONTROL);
5710 	cr4 = vmcs_readl(GUEST_CR4);
5711 	efer = vmcs_read64(GUEST_IA32_EFER);
5712 	secondary_exec_control = 0;
5713 	if (cpu_has_secondary_exec_ctrls())
5714 		secondary_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
5715 
5716 	pr_err("*** Guest State ***\n");
5717 	pr_err("CR0: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
5718 	       vmcs_readl(GUEST_CR0), vmcs_readl(CR0_READ_SHADOW),
5719 	       vmcs_readl(CR0_GUEST_HOST_MASK));
5720 	pr_err("CR4: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
5721 	       cr4, vmcs_readl(CR4_READ_SHADOW), vmcs_readl(CR4_GUEST_HOST_MASK));
5722 	pr_err("CR3 = 0x%016lx\n", vmcs_readl(GUEST_CR3));
5723 	if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT) &&
5724 	    (cr4 & X86_CR4_PAE) && !(efer & EFER_LMA))
5725 	{
5726 		pr_err("PDPTR0 = 0x%016llx  PDPTR1 = 0x%016llx\n",
5727 		       vmcs_read64(GUEST_PDPTR0), vmcs_read64(GUEST_PDPTR1));
5728 		pr_err("PDPTR2 = 0x%016llx  PDPTR3 = 0x%016llx\n",
5729 		       vmcs_read64(GUEST_PDPTR2), vmcs_read64(GUEST_PDPTR3));
5730 	}
5731 	pr_err("RSP = 0x%016lx  RIP = 0x%016lx\n",
5732 	       vmcs_readl(GUEST_RSP), vmcs_readl(GUEST_RIP));
5733 	pr_err("RFLAGS=0x%08lx         DR7 = 0x%016lx\n",
5734 	       vmcs_readl(GUEST_RFLAGS), vmcs_readl(GUEST_DR7));
5735 	pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
5736 	       vmcs_readl(GUEST_SYSENTER_ESP),
5737 	       vmcs_read32(GUEST_SYSENTER_CS), vmcs_readl(GUEST_SYSENTER_EIP));
5738 	vmx_dump_sel("CS:  ", GUEST_CS_SELECTOR);
5739 	vmx_dump_sel("DS:  ", GUEST_DS_SELECTOR);
5740 	vmx_dump_sel("SS:  ", GUEST_SS_SELECTOR);
5741 	vmx_dump_sel("ES:  ", GUEST_ES_SELECTOR);
5742 	vmx_dump_sel("FS:  ", GUEST_FS_SELECTOR);
5743 	vmx_dump_sel("GS:  ", GUEST_GS_SELECTOR);
5744 	vmx_dump_dtsel("GDTR:", GUEST_GDTR_LIMIT);
5745 	vmx_dump_sel("LDTR:", GUEST_LDTR_SELECTOR);
5746 	vmx_dump_dtsel("IDTR:", GUEST_IDTR_LIMIT);
5747 	vmx_dump_sel("TR:  ", GUEST_TR_SELECTOR);
5748 	if ((vmexit_ctl & (VM_EXIT_SAVE_IA32_PAT | VM_EXIT_SAVE_IA32_EFER)) ||
5749 	    (vmentry_ctl & (VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_IA32_EFER)))
5750 		pr_err("EFER =     0x%016llx  PAT = 0x%016llx\n",
5751 		       efer, vmcs_read64(GUEST_IA32_PAT));
5752 	pr_err("DebugCtl = 0x%016llx  DebugExceptions = 0x%016lx\n",
5753 	       vmcs_read64(GUEST_IA32_DEBUGCTL),
5754 	       vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS));
5755 	if (cpu_has_load_perf_global_ctrl() &&
5756 	    vmentry_ctl & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
5757 		pr_err("PerfGlobCtl = 0x%016llx\n",
5758 		       vmcs_read64(GUEST_IA32_PERF_GLOBAL_CTRL));
5759 	if (vmentry_ctl & VM_ENTRY_LOAD_BNDCFGS)
5760 		pr_err("BndCfgS = 0x%016llx\n", vmcs_read64(GUEST_BNDCFGS));
5761 	pr_err("Interruptibility = %08x  ActivityState = %08x\n",
5762 	       vmcs_read32(GUEST_INTERRUPTIBILITY_INFO),
5763 	       vmcs_read32(GUEST_ACTIVITY_STATE));
5764 	if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)
5765 		pr_err("InterruptStatus = %04x\n",
5766 		       vmcs_read16(GUEST_INTR_STATUS));
5767 
5768 	pr_err("*** Host State ***\n");
5769 	pr_err("RIP = 0x%016lx  RSP = 0x%016lx\n",
5770 	       vmcs_readl(HOST_RIP), vmcs_readl(HOST_RSP));
5771 	pr_err("CS=%04x SS=%04x DS=%04x ES=%04x FS=%04x GS=%04x TR=%04x\n",
5772 	       vmcs_read16(HOST_CS_SELECTOR), vmcs_read16(HOST_SS_SELECTOR),
5773 	       vmcs_read16(HOST_DS_SELECTOR), vmcs_read16(HOST_ES_SELECTOR),
5774 	       vmcs_read16(HOST_FS_SELECTOR), vmcs_read16(HOST_GS_SELECTOR),
5775 	       vmcs_read16(HOST_TR_SELECTOR));
5776 	pr_err("FSBase=%016lx GSBase=%016lx TRBase=%016lx\n",
5777 	       vmcs_readl(HOST_FS_BASE), vmcs_readl(HOST_GS_BASE),
5778 	       vmcs_readl(HOST_TR_BASE));
5779 	pr_err("GDTBase=%016lx IDTBase=%016lx\n",
5780 	       vmcs_readl(HOST_GDTR_BASE), vmcs_readl(HOST_IDTR_BASE));
5781 	pr_err("CR0=%016lx CR3=%016lx CR4=%016lx\n",
5782 	       vmcs_readl(HOST_CR0), vmcs_readl(HOST_CR3),
5783 	       vmcs_readl(HOST_CR4));
5784 	pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
5785 	       vmcs_readl(HOST_IA32_SYSENTER_ESP),
5786 	       vmcs_read32(HOST_IA32_SYSENTER_CS),
5787 	       vmcs_readl(HOST_IA32_SYSENTER_EIP));
5788 	if (vmexit_ctl & (VM_EXIT_LOAD_IA32_PAT | VM_EXIT_LOAD_IA32_EFER))
5789 		pr_err("EFER = 0x%016llx  PAT = 0x%016llx\n",
5790 		       vmcs_read64(HOST_IA32_EFER),
5791 		       vmcs_read64(HOST_IA32_PAT));
5792 	if (cpu_has_load_perf_global_ctrl() &&
5793 	    vmexit_ctl & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
5794 		pr_err("PerfGlobCtl = 0x%016llx\n",
5795 		       vmcs_read64(HOST_IA32_PERF_GLOBAL_CTRL));
5796 
5797 	pr_err("*** Control State ***\n");
5798 	pr_err("PinBased=%08x CPUBased=%08x SecondaryExec=%08x\n",
5799 	       pin_based_exec_ctrl, cpu_based_exec_ctrl, secondary_exec_control);
5800 	pr_err("EntryControls=%08x ExitControls=%08x\n", vmentry_ctl, vmexit_ctl);
5801 	pr_err("ExceptionBitmap=%08x PFECmask=%08x PFECmatch=%08x\n",
5802 	       vmcs_read32(EXCEPTION_BITMAP),
5803 	       vmcs_read32(PAGE_FAULT_ERROR_CODE_MASK),
5804 	       vmcs_read32(PAGE_FAULT_ERROR_CODE_MATCH));
5805 	pr_err("VMEntry: intr_info=%08x errcode=%08x ilen=%08x\n",
5806 	       vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
5807 	       vmcs_read32(VM_ENTRY_EXCEPTION_ERROR_CODE),
5808 	       vmcs_read32(VM_ENTRY_INSTRUCTION_LEN));
5809 	pr_err("VMExit: intr_info=%08x errcode=%08x ilen=%08x\n",
5810 	       vmcs_read32(VM_EXIT_INTR_INFO),
5811 	       vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
5812 	       vmcs_read32(VM_EXIT_INSTRUCTION_LEN));
5813 	pr_err("        reason=%08x qualification=%016lx\n",
5814 	       vmcs_read32(VM_EXIT_REASON), vmcs_readl(EXIT_QUALIFICATION));
5815 	pr_err("IDTVectoring: info=%08x errcode=%08x\n",
5816 	       vmcs_read32(IDT_VECTORING_INFO_FIELD),
5817 	       vmcs_read32(IDT_VECTORING_ERROR_CODE));
5818 	pr_err("TSC Offset = 0x%016llx\n", vmcs_read64(TSC_OFFSET));
5819 	if (secondary_exec_control & SECONDARY_EXEC_TSC_SCALING)
5820 		pr_err("TSC Multiplier = 0x%016llx\n",
5821 		       vmcs_read64(TSC_MULTIPLIER));
5822 	if (cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW) {
5823 		if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY) {
5824 			u16 status = vmcs_read16(GUEST_INTR_STATUS);
5825 			pr_err("SVI|RVI = %02x|%02x ", status >> 8, status & 0xff);
5826 		}
5827 		pr_cont("TPR Threshold = 0x%02x\n", vmcs_read32(TPR_THRESHOLD));
5828 		if (secondary_exec_control & SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)
5829 			pr_err("APIC-access addr = 0x%016llx ", vmcs_read64(APIC_ACCESS_ADDR));
5830 		pr_cont("virt-APIC addr = 0x%016llx\n", vmcs_read64(VIRTUAL_APIC_PAGE_ADDR));
5831 	}
5832 	if (pin_based_exec_ctrl & PIN_BASED_POSTED_INTR)
5833 		pr_err("PostedIntrVec = 0x%02x\n", vmcs_read16(POSTED_INTR_NV));
5834 	if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT))
5835 		pr_err("EPT pointer = 0x%016llx\n", vmcs_read64(EPT_POINTER));
5836 	n = vmcs_read32(CR3_TARGET_COUNT);
5837 	for (i = 0; i + 1 < n; i += 4)
5838 		pr_err("CR3 target%u=%016lx target%u=%016lx\n",
5839 		       i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2),
5840 		       i + 1, vmcs_readl(CR3_TARGET_VALUE0 + i * 2 + 2));
5841 	if (i < n)
5842 		pr_err("CR3 target%u=%016lx\n",
5843 		       i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2));
5844 	if (secondary_exec_control & SECONDARY_EXEC_PAUSE_LOOP_EXITING)
5845 		pr_err("PLE Gap=%08x Window=%08x\n",
5846 		       vmcs_read32(PLE_GAP), vmcs_read32(PLE_WINDOW));
5847 	if (secondary_exec_control & SECONDARY_EXEC_ENABLE_VPID)
5848 		pr_err("Virtual processor ID = 0x%04x\n",
5849 		       vmcs_read16(VIRTUAL_PROCESSOR_ID));
5850 }
5851 
5852 /*
5853  * The guest has exited.  See if we can fix it or if we need userspace
5854  * assistance.
5855  */
5856 static int vmx_handle_exit(struct kvm_vcpu *vcpu,
5857 	enum exit_fastpath_completion exit_fastpath)
5858 {
5859 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5860 	u32 exit_reason = vmx->exit_reason;
5861 	u32 vectoring_info = vmx->idt_vectoring_info;
5862 
5863 	trace_kvm_exit(exit_reason, vcpu, KVM_ISA_VMX);
5864 
5865 	/*
5866 	 * Flush logged GPAs PML buffer, this will make dirty_bitmap more
5867 	 * updated. Another good is, in kvm_vm_ioctl_get_dirty_log, before
5868 	 * querying dirty_bitmap, we only need to kick all vcpus out of guest
5869 	 * mode as if vcpus is in root mode, the PML buffer must has been
5870 	 * flushed already.
5871 	 */
5872 	if (enable_pml)
5873 		vmx_flush_pml_buffer(vcpu);
5874 
5875 	/* If guest state is invalid, start emulating */
5876 	if (vmx->emulation_required)
5877 		return handle_invalid_guest_state(vcpu);
5878 
5879 	if (is_guest_mode(vcpu) && nested_vmx_exit_reflected(vcpu, exit_reason))
5880 		return nested_vmx_reflect_vmexit(vcpu, exit_reason);
5881 
5882 	if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) {
5883 		dump_vmcs();
5884 		vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
5885 		vcpu->run->fail_entry.hardware_entry_failure_reason
5886 			= exit_reason;
5887 		return 0;
5888 	}
5889 
5890 	if (unlikely(vmx->fail)) {
5891 		dump_vmcs();
5892 		vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
5893 		vcpu->run->fail_entry.hardware_entry_failure_reason
5894 			= vmcs_read32(VM_INSTRUCTION_ERROR);
5895 		return 0;
5896 	}
5897 
5898 	/*
5899 	 * Note:
5900 	 * Do not try to fix EXIT_REASON_EPT_MISCONFIG if it caused by
5901 	 * delivery event since it indicates guest is accessing MMIO.
5902 	 * The vm-exit can be triggered again after return to guest that
5903 	 * will cause infinite loop.
5904 	 */
5905 	if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
5906 			(exit_reason != EXIT_REASON_EXCEPTION_NMI &&
5907 			exit_reason != EXIT_REASON_EPT_VIOLATION &&
5908 			exit_reason != EXIT_REASON_PML_FULL &&
5909 			exit_reason != EXIT_REASON_TASK_SWITCH)) {
5910 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5911 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_DELIVERY_EV;
5912 		vcpu->run->internal.ndata = 3;
5913 		vcpu->run->internal.data[0] = vectoring_info;
5914 		vcpu->run->internal.data[1] = exit_reason;
5915 		vcpu->run->internal.data[2] = vcpu->arch.exit_qualification;
5916 		if (exit_reason == EXIT_REASON_EPT_MISCONFIG) {
5917 			vcpu->run->internal.ndata++;
5918 			vcpu->run->internal.data[3] =
5919 				vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5920 		}
5921 		return 0;
5922 	}
5923 
5924 	if (unlikely(!enable_vnmi &&
5925 		     vmx->loaded_vmcs->soft_vnmi_blocked)) {
5926 		if (vmx_interrupt_allowed(vcpu)) {
5927 			vmx->loaded_vmcs->soft_vnmi_blocked = 0;
5928 		} else if (vmx->loaded_vmcs->vnmi_blocked_time > 1000000000LL &&
5929 			   vcpu->arch.nmi_pending) {
5930 			/*
5931 			 * This CPU don't support us in finding the end of an
5932 			 * NMI-blocked window if the guest runs with IRQs
5933 			 * disabled. So we pull the trigger after 1 s of
5934 			 * futile waiting, but inform the user about this.
5935 			 */
5936 			printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
5937 			       "state on VCPU %d after 1 s timeout\n",
5938 			       __func__, vcpu->vcpu_id);
5939 			vmx->loaded_vmcs->soft_vnmi_blocked = 0;
5940 		}
5941 	}
5942 
5943 	if (exit_fastpath == EXIT_FASTPATH_SKIP_EMUL_INS) {
5944 		kvm_skip_emulated_instruction(vcpu);
5945 		return 1;
5946 	}
5947 
5948 	if (exit_reason >= kvm_vmx_max_exit_handlers)
5949 		goto unexpected_vmexit;
5950 #ifdef CONFIG_RETPOLINE
5951 	if (exit_reason == EXIT_REASON_MSR_WRITE)
5952 		return kvm_emulate_wrmsr(vcpu);
5953 	else if (exit_reason == EXIT_REASON_PREEMPTION_TIMER)
5954 		return handle_preemption_timer(vcpu);
5955 	else if (exit_reason == EXIT_REASON_INTERRUPT_WINDOW)
5956 		return handle_interrupt_window(vcpu);
5957 	else if (exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT)
5958 		return handle_external_interrupt(vcpu);
5959 	else if (exit_reason == EXIT_REASON_HLT)
5960 		return kvm_emulate_halt(vcpu);
5961 	else if (exit_reason == EXIT_REASON_EPT_MISCONFIG)
5962 		return handle_ept_misconfig(vcpu);
5963 #endif
5964 
5965 	exit_reason = array_index_nospec(exit_reason,
5966 					 kvm_vmx_max_exit_handlers);
5967 	if (!kvm_vmx_exit_handlers[exit_reason])
5968 		goto unexpected_vmexit;
5969 
5970 	return kvm_vmx_exit_handlers[exit_reason](vcpu);
5971 
5972 unexpected_vmexit:
5973 	vcpu_unimpl(vcpu, "vmx: unexpected exit reason 0x%x\n", exit_reason);
5974 	dump_vmcs();
5975 	vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5976 	vcpu->run->internal.suberror =
5977 			KVM_INTERNAL_ERROR_UNEXPECTED_EXIT_REASON;
5978 	vcpu->run->internal.ndata = 1;
5979 	vcpu->run->internal.data[0] = exit_reason;
5980 	return 0;
5981 }
5982 
5983 /*
5984  * Software based L1D cache flush which is used when microcode providing
5985  * the cache control MSR is not loaded.
5986  *
5987  * The L1D cache is 32 KiB on Nehalem and later microarchitectures, but to
5988  * flush it is required to read in 64 KiB because the replacement algorithm
5989  * is not exactly LRU. This could be sized at runtime via topology
5990  * information but as all relevant affected CPUs have 32KiB L1D cache size
5991  * there is no point in doing so.
5992  */
5993 static void vmx_l1d_flush(struct kvm_vcpu *vcpu)
5994 {
5995 	int size = PAGE_SIZE << L1D_CACHE_ORDER;
5996 
5997 	/*
5998 	 * This code is only executed when the the flush mode is 'cond' or
5999 	 * 'always'
6000 	 */
6001 	if (static_branch_likely(&vmx_l1d_flush_cond)) {
6002 		bool flush_l1d;
6003 
6004 		/*
6005 		 * Clear the per-vcpu flush bit, it gets set again
6006 		 * either from vcpu_run() or from one of the unsafe
6007 		 * VMEXIT handlers.
6008 		 */
6009 		flush_l1d = vcpu->arch.l1tf_flush_l1d;
6010 		vcpu->arch.l1tf_flush_l1d = false;
6011 
6012 		/*
6013 		 * Clear the per-cpu flush bit, it gets set again from
6014 		 * the interrupt handlers.
6015 		 */
6016 		flush_l1d |= kvm_get_cpu_l1tf_flush_l1d();
6017 		kvm_clear_cpu_l1tf_flush_l1d();
6018 
6019 		if (!flush_l1d)
6020 			return;
6021 	}
6022 
6023 	vcpu->stat.l1d_flush++;
6024 
6025 	if (static_cpu_has(X86_FEATURE_FLUSH_L1D)) {
6026 		wrmsrl(MSR_IA32_FLUSH_CMD, L1D_FLUSH);
6027 		return;
6028 	}
6029 
6030 	asm volatile(
6031 		/* First ensure the pages are in the TLB */
6032 		"xorl	%%eax, %%eax\n"
6033 		".Lpopulate_tlb:\n\t"
6034 		"movzbl	(%[flush_pages], %%" _ASM_AX "), %%ecx\n\t"
6035 		"addl	$4096, %%eax\n\t"
6036 		"cmpl	%%eax, %[size]\n\t"
6037 		"jne	.Lpopulate_tlb\n\t"
6038 		"xorl	%%eax, %%eax\n\t"
6039 		"cpuid\n\t"
6040 		/* Now fill the cache */
6041 		"xorl	%%eax, %%eax\n"
6042 		".Lfill_cache:\n"
6043 		"movzbl	(%[flush_pages], %%" _ASM_AX "), %%ecx\n\t"
6044 		"addl	$64, %%eax\n\t"
6045 		"cmpl	%%eax, %[size]\n\t"
6046 		"jne	.Lfill_cache\n\t"
6047 		"lfence\n"
6048 		:: [flush_pages] "r" (vmx_l1d_flush_pages),
6049 		    [size] "r" (size)
6050 		: "eax", "ebx", "ecx", "edx");
6051 }
6052 
6053 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
6054 {
6055 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6056 	int tpr_threshold;
6057 
6058 	if (is_guest_mode(vcpu) &&
6059 		nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
6060 		return;
6061 
6062 	tpr_threshold = (irr == -1 || tpr < irr) ? 0 : irr;
6063 	if (is_guest_mode(vcpu))
6064 		to_vmx(vcpu)->nested.l1_tpr_threshold = tpr_threshold;
6065 	else
6066 		vmcs_write32(TPR_THRESHOLD, tpr_threshold);
6067 }
6068 
6069 void vmx_set_virtual_apic_mode(struct kvm_vcpu *vcpu)
6070 {
6071 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6072 	u32 sec_exec_control;
6073 
6074 	if (!lapic_in_kernel(vcpu))
6075 		return;
6076 
6077 	if (!flexpriority_enabled &&
6078 	    !cpu_has_vmx_virtualize_x2apic_mode())
6079 		return;
6080 
6081 	/* Postpone execution until vmcs01 is the current VMCS. */
6082 	if (is_guest_mode(vcpu)) {
6083 		vmx->nested.change_vmcs01_virtual_apic_mode = true;
6084 		return;
6085 	}
6086 
6087 	sec_exec_control = secondary_exec_controls_get(vmx);
6088 	sec_exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
6089 			      SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE);
6090 
6091 	switch (kvm_get_apic_mode(vcpu)) {
6092 	case LAPIC_MODE_INVALID:
6093 		WARN_ONCE(true, "Invalid local APIC state");
6094 	case LAPIC_MODE_DISABLED:
6095 		break;
6096 	case LAPIC_MODE_XAPIC:
6097 		if (flexpriority_enabled) {
6098 			sec_exec_control |=
6099 				SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
6100 			vmx_flush_tlb(vcpu, true);
6101 		}
6102 		break;
6103 	case LAPIC_MODE_X2APIC:
6104 		if (cpu_has_vmx_virtualize_x2apic_mode())
6105 			sec_exec_control |=
6106 				SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
6107 		break;
6108 	}
6109 	secondary_exec_controls_set(vmx, sec_exec_control);
6110 
6111 	vmx_update_msr_bitmap(vcpu);
6112 }
6113 
6114 static void vmx_set_apic_access_page_addr(struct kvm_vcpu *vcpu, hpa_t hpa)
6115 {
6116 	if (!is_guest_mode(vcpu)) {
6117 		vmcs_write64(APIC_ACCESS_ADDR, hpa);
6118 		vmx_flush_tlb(vcpu, true);
6119 	}
6120 }
6121 
6122 static void vmx_hwapic_isr_update(struct kvm_vcpu *vcpu, int max_isr)
6123 {
6124 	u16 status;
6125 	u8 old;
6126 
6127 	if (max_isr == -1)
6128 		max_isr = 0;
6129 
6130 	status = vmcs_read16(GUEST_INTR_STATUS);
6131 	old = status >> 8;
6132 	if (max_isr != old) {
6133 		status &= 0xff;
6134 		status |= max_isr << 8;
6135 		vmcs_write16(GUEST_INTR_STATUS, status);
6136 	}
6137 }
6138 
6139 static void vmx_set_rvi(int vector)
6140 {
6141 	u16 status;
6142 	u8 old;
6143 
6144 	if (vector == -1)
6145 		vector = 0;
6146 
6147 	status = vmcs_read16(GUEST_INTR_STATUS);
6148 	old = (u8)status & 0xff;
6149 	if ((u8)vector != old) {
6150 		status &= ~0xff;
6151 		status |= (u8)vector;
6152 		vmcs_write16(GUEST_INTR_STATUS, status);
6153 	}
6154 }
6155 
6156 static void vmx_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr)
6157 {
6158 	/*
6159 	 * When running L2, updating RVI is only relevant when
6160 	 * vmcs12 virtual-interrupt-delivery enabled.
6161 	 * However, it can be enabled only when L1 also
6162 	 * intercepts external-interrupts and in that case
6163 	 * we should not update vmcs02 RVI but instead intercept
6164 	 * interrupt. Therefore, do nothing when running L2.
6165 	 */
6166 	if (!is_guest_mode(vcpu))
6167 		vmx_set_rvi(max_irr);
6168 }
6169 
6170 static int vmx_sync_pir_to_irr(struct kvm_vcpu *vcpu)
6171 {
6172 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6173 	int max_irr;
6174 	bool max_irr_updated;
6175 
6176 	WARN_ON(!vcpu->arch.apicv_active);
6177 	if (pi_test_on(&vmx->pi_desc)) {
6178 		pi_clear_on(&vmx->pi_desc);
6179 		/*
6180 		 * IOMMU can write to PID.ON, so the barrier matters even on UP.
6181 		 * But on x86 this is just a compiler barrier anyway.
6182 		 */
6183 		smp_mb__after_atomic();
6184 		max_irr_updated =
6185 			kvm_apic_update_irr(vcpu, vmx->pi_desc.pir, &max_irr);
6186 
6187 		/*
6188 		 * If we are running L2 and L1 has a new pending interrupt
6189 		 * which can be injected, we should re-evaluate
6190 		 * what should be done with this new L1 interrupt.
6191 		 * If L1 intercepts external-interrupts, we should
6192 		 * exit from L2 to L1. Otherwise, interrupt should be
6193 		 * delivered directly to L2.
6194 		 */
6195 		if (is_guest_mode(vcpu) && max_irr_updated) {
6196 			if (nested_exit_on_intr(vcpu))
6197 				kvm_vcpu_exiting_guest_mode(vcpu);
6198 			else
6199 				kvm_make_request(KVM_REQ_EVENT, vcpu);
6200 		}
6201 	} else {
6202 		max_irr = kvm_lapic_find_highest_irr(vcpu);
6203 	}
6204 	vmx_hwapic_irr_update(vcpu, max_irr);
6205 	return max_irr;
6206 }
6207 
6208 static bool vmx_dy_apicv_has_pending_interrupt(struct kvm_vcpu *vcpu)
6209 {
6210 	struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
6211 
6212 	return pi_test_on(pi_desc) ||
6213 		(pi_test_sn(pi_desc) && !pi_is_pir_empty(pi_desc));
6214 }
6215 
6216 static void vmx_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap)
6217 {
6218 	if (!kvm_vcpu_apicv_active(vcpu))
6219 		return;
6220 
6221 	vmcs_write64(EOI_EXIT_BITMAP0, eoi_exit_bitmap[0]);
6222 	vmcs_write64(EOI_EXIT_BITMAP1, eoi_exit_bitmap[1]);
6223 	vmcs_write64(EOI_EXIT_BITMAP2, eoi_exit_bitmap[2]);
6224 	vmcs_write64(EOI_EXIT_BITMAP3, eoi_exit_bitmap[3]);
6225 }
6226 
6227 static void vmx_apicv_post_state_restore(struct kvm_vcpu *vcpu)
6228 {
6229 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6230 
6231 	pi_clear_on(&vmx->pi_desc);
6232 	memset(vmx->pi_desc.pir, 0, sizeof(vmx->pi_desc.pir));
6233 }
6234 
6235 static void handle_exception_nmi_irqoff(struct vcpu_vmx *vmx)
6236 {
6237 	vmx->exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6238 
6239 	/* if exit due to PF check for async PF */
6240 	if (is_page_fault(vmx->exit_intr_info))
6241 		vmx->vcpu.arch.apf.host_apf_reason = kvm_read_and_reset_pf_reason();
6242 
6243 	/* Handle machine checks before interrupts are enabled */
6244 	if (is_machine_check(vmx->exit_intr_info))
6245 		kvm_machine_check();
6246 
6247 	/* We need to handle NMIs before interrupts are enabled */
6248 	if (is_nmi(vmx->exit_intr_info)) {
6249 		kvm_before_interrupt(&vmx->vcpu);
6250 		asm("int $2");
6251 		kvm_after_interrupt(&vmx->vcpu);
6252 	}
6253 }
6254 
6255 static void handle_external_interrupt_irqoff(struct kvm_vcpu *vcpu)
6256 {
6257 	unsigned int vector;
6258 	unsigned long entry;
6259 #ifdef CONFIG_X86_64
6260 	unsigned long tmp;
6261 #endif
6262 	gate_desc *desc;
6263 	u32 intr_info;
6264 
6265 	intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6266 	if (WARN_ONCE(!is_external_intr(intr_info),
6267 	    "KVM: unexpected VM-Exit interrupt info: 0x%x", intr_info))
6268 		return;
6269 
6270 	vector = intr_info & INTR_INFO_VECTOR_MASK;
6271 	desc = (gate_desc *)host_idt_base + vector;
6272 	entry = gate_offset(desc);
6273 
6274 	kvm_before_interrupt(vcpu);
6275 
6276 	asm volatile(
6277 #ifdef CONFIG_X86_64
6278 		"mov %%" _ASM_SP ", %[sp]\n\t"
6279 		"and $0xfffffffffffffff0, %%" _ASM_SP "\n\t"
6280 		"push $%c[ss]\n\t"
6281 		"push %[sp]\n\t"
6282 #endif
6283 		"pushf\n\t"
6284 		__ASM_SIZE(push) " $%c[cs]\n\t"
6285 		CALL_NOSPEC
6286 		:
6287 #ifdef CONFIG_X86_64
6288 		[sp]"=&r"(tmp),
6289 #endif
6290 		ASM_CALL_CONSTRAINT
6291 		:
6292 		[thunk_target]"r"(entry),
6293 		[ss]"i"(__KERNEL_DS),
6294 		[cs]"i"(__KERNEL_CS)
6295 	);
6296 
6297 	kvm_after_interrupt(vcpu);
6298 }
6299 STACK_FRAME_NON_STANDARD(handle_external_interrupt_irqoff);
6300 
6301 static void vmx_handle_exit_irqoff(struct kvm_vcpu *vcpu,
6302 	enum exit_fastpath_completion *exit_fastpath)
6303 {
6304 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6305 
6306 	if (vmx->exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT)
6307 		handle_external_interrupt_irqoff(vcpu);
6308 	else if (vmx->exit_reason == EXIT_REASON_EXCEPTION_NMI)
6309 		handle_exception_nmi_irqoff(vmx);
6310 	else if (!is_guest_mode(vcpu) &&
6311 		vmx->exit_reason == EXIT_REASON_MSR_WRITE)
6312 		*exit_fastpath = handle_fastpath_set_msr_irqoff(vcpu);
6313 }
6314 
6315 static bool vmx_has_emulated_msr(int index)
6316 {
6317 	switch (index) {
6318 	case MSR_IA32_SMBASE:
6319 		/*
6320 		 * We cannot do SMM unless we can run the guest in big
6321 		 * real mode.
6322 		 */
6323 		return enable_unrestricted_guest || emulate_invalid_guest_state;
6324 	case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
6325 		return nested;
6326 	case MSR_AMD64_VIRT_SPEC_CTRL:
6327 		/* This is AMD only.  */
6328 		return false;
6329 	default:
6330 		return true;
6331 	}
6332 }
6333 
6334 static bool vmx_pt_supported(void)
6335 {
6336 	return pt_mode == PT_MODE_HOST_GUEST;
6337 }
6338 
6339 static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx)
6340 {
6341 	u32 exit_intr_info;
6342 	bool unblock_nmi;
6343 	u8 vector;
6344 	bool idtv_info_valid;
6345 
6346 	idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK;
6347 
6348 	if (enable_vnmi) {
6349 		if (vmx->loaded_vmcs->nmi_known_unmasked)
6350 			return;
6351 		/*
6352 		 * Can't use vmx->exit_intr_info since we're not sure what
6353 		 * the exit reason is.
6354 		 */
6355 		exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6356 		unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
6357 		vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
6358 		/*
6359 		 * SDM 3: 27.7.1.2 (September 2008)
6360 		 * Re-set bit "block by NMI" before VM entry if vmexit caused by
6361 		 * a guest IRET fault.
6362 		 * SDM 3: 23.2.2 (September 2008)
6363 		 * Bit 12 is undefined in any of the following cases:
6364 		 *  If the VM exit sets the valid bit in the IDT-vectoring
6365 		 *   information field.
6366 		 *  If the VM exit is due to a double fault.
6367 		 */
6368 		if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
6369 		    vector != DF_VECTOR && !idtv_info_valid)
6370 			vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
6371 				      GUEST_INTR_STATE_NMI);
6372 		else
6373 			vmx->loaded_vmcs->nmi_known_unmasked =
6374 				!(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO)
6375 				  & GUEST_INTR_STATE_NMI);
6376 	} else if (unlikely(vmx->loaded_vmcs->soft_vnmi_blocked))
6377 		vmx->loaded_vmcs->vnmi_blocked_time +=
6378 			ktime_to_ns(ktime_sub(ktime_get(),
6379 					      vmx->loaded_vmcs->entry_time));
6380 }
6381 
6382 static void __vmx_complete_interrupts(struct kvm_vcpu *vcpu,
6383 				      u32 idt_vectoring_info,
6384 				      int instr_len_field,
6385 				      int error_code_field)
6386 {
6387 	u8 vector;
6388 	int type;
6389 	bool idtv_info_valid;
6390 
6391 	idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
6392 
6393 	vcpu->arch.nmi_injected = false;
6394 	kvm_clear_exception_queue(vcpu);
6395 	kvm_clear_interrupt_queue(vcpu);
6396 
6397 	if (!idtv_info_valid)
6398 		return;
6399 
6400 	kvm_make_request(KVM_REQ_EVENT, vcpu);
6401 
6402 	vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
6403 	type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
6404 
6405 	switch (type) {
6406 	case INTR_TYPE_NMI_INTR:
6407 		vcpu->arch.nmi_injected = true;
6408 		/*
6409 		 * SDM 3: 27.7.1.2 (September 2008)
6410 		 * Clear bit "block by NMI" before VM entry if a NMI
6411 		 * delivery faulted.
6412 		 */
6413 		vmx_set_nmi_mask(vcpu, false);
6414 		break;
6415 	case INTR_TYPE_SOFT_EXCEPTION:
6416 		vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
6417 		/* fall through */
6418 	case INTR_TYPE_HARD_EXCEPTION:
6419 		if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
6420 			u32 err = vmcs_read32(error_code_field);
6421 			kvm_requeue_exception_e(vcpu, vector, err);
6422 		} else
6423 			kvm_requeue_exception(vcpu, vector);
6424 		break;
6425 	case INTR_TYPE_SOFT_INTR:
6426 		vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
6427 		/* fall through */
6428 	case INTR_TYPE_EXT_INTR:
6429 		kvm_queue_interrupt(vcpu, vector, type == INTR_TYPE_SOFT_INTR);
6430 		break;
6431 	default:
6432 		break;
6433 	}
6434 }
6435 
6436 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
6437 {
6438 	__vmx_complete_interrupts(&vmx->vcpu, vmx->idt_vectoring_info,
6439 				  VM_EXIT_INSTRUCTION_LEN,
6440 				  IDT_VECTORING_ERROR_CODE);
6441 }
6442 
6443 static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
6444 {
6445 	__vmx_complete_interrupts(vcpu,
6446 				  vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
6447 				  VM_ENTRY_INSTRUCTION_LEN,
6448 				  VM_ENTRY_EXCEPTION_ERROR_CODE);
6449 
6450 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
6451 }
6452 
6453 static void atomic_switch_perf_msrs(struct vcpu_vmx *vmx)
6454 {
6455 	int i, nr_msrs;
6456 	struct perf_guest_switch_msr *msrs;
6457 
6458 	msrs = perf_guest_get_msrs(&nr_msrs);
6459 
6460 	if (!msrs)
6461 		return;
6462 
6463 	for (i = 0; i < nr_msrs; i++)
6464 		if (msrs[i].host == msrs[i].guest)
6465 			clear_atomic_switch_msr(vmx, msrs[i].msr);
6466 		else
6467 			add_atomic_switch_msr(vmx, msrs[i].msr, msrs[i].guest,
6468 					msrs[i].host, false);
6469 }
6470 
6471 static void atomic_switch_umwait_control_msr(struct vcpu_vmx *vmx)
6472 {
6473 	u32 host_umwait_control;
6474 
6475 	if (!vmx_has_waitpkg(vmx))
6476 		return;
6477 
6478 	host_umwait_control = get_umwait_control_msr();
6479 
6480 	if (vmx->msr_ia32_umwait_control != host_umwait_control)
6481 		add_atomic_switch_msr(vmx, MSR_IA32_UMWAIT_CONTROL,
6482 			vmx->msr_ia32_umwait_control,
6483 			host_umwait_control, false);
6484 	else
6485 		clear_atomic_switch_msr(vmx, MSR_IA32_UMWAIT_CONTROL);
6486 }
6487 
6488 static void vmx_update_hv_timer(struct kvm_vcpu *vcpu)
6489 {
6490 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6491 	u64 tscl;
6492 	u32 delta_tsc;
6493 
6494 	if (vmx->req_immediate_exit) {
6495 		vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, 0);
6496 		vmx->loaded_vmcs->hv_timer_soft_disabled = false;
6497 	} else if (vmx->hv_deadline_tsc != -1) {
6498 		tscl = rdtsc();
6499 		if (vmx->hv_deadline_tsc > tscl)
6500 			/* set_hv_timer ensures the delta fits in 32-bits */
6501 			delta_tsc = (u32)((vmx->hv_deadline_tsc - tscl) >>
6502 				cpu_preemption_timer_multi);
6503 		else
6504 			delta_tsc = 0;
6505 
6506 		vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, delta_tsc);
6507 		vmx->loaded_vmcs->hv_timer_soft_disabled = false;
6508 	} else if (!vmx->loaded_vmcs->hv_timer_soft_disabled) {
6509 		vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, -1);
6510 		vmx->loaded_vmcs->hv_timer_soft_disabled = true;
6511 	}
6512 }
6513 
6514 void vmx_update_host_rsp(struct vcpu_vmx *vmx, unsigned long host_rsp)
6515 {
6516 	if (unlikely(host_rsp != vmx->loaded_vmcs->host_state.rsp)) {
6517 		vmx->loaded_vmcs->host_state.rsp = host_rsp;
6518 		vmcs_writel(HOST_RSP, host_rsp);
6519 	}
6520 }
6521 
6522 bool __vmx_vcpu_run(struct vcpu_vmx *vmx, unsigned long *regs, bool launched);
6523 
6524 static void vmx_vcpu_run(struct kvm_vcpu *vcpu)
6525 {
6526 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6527 	unsigned long cr3, cr4;
6528 
6529 	/* Record the guest's net vcpu time for enforced NMI injections. */
6530 	if (unlikely(!enable_vnmi &&
6531 		     vmx->loaded_vmcs->soft_vnmi_blocked))
6532 		vmx->loaded_vmcs->entry_time = ktime_get();
6533 
6534 	/* Don't enter VMX if guest state is invalid, let the exit handler
6535 	   start emulation until we arrive back to a valid state */
6536 	if (vmx->emulation_required)
6537 		return;
6538 
6539 	if (vmx->ple_window_dirty) {
6540 		vmx->ple_window_dirty = false;
6541 		vmcs_write32(PLE_WINDOW, vmx->ple_window);
6542 	}
6543 
6544 	/*
6545 	 * We did this in prepare_switch_to_guest, because it needs to
6546 	 * be within srcu_read_lock.
6547 	 */
6548 	WARN_ON_ONCE(vmx->nested.need_vmcs12_to_shadow_sync);
6549 
6550 	if (kvm_register_is_dirty(vcpu, VCPU_REGS_RSP))
6551 		vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
6552 	if (kvm_register_is_dirty(vcpu, VCPU_REGS_RIP))
6553 		vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
6554 
6555 	cr3 = __get_current_cr3_fast();
6556 	if (unlikely(cr3 != vmx->loaded_vmcs->host_state.cr3)) {
6557 		vmcs_writel(HOST_CR3, cr3);
6558 		vmx->loaded_vmcs->host_state.cr3 = cr3;
6559 	}
6560 
6561 	cr4 = cr4_read_shadow();
6562 	if (unlikely(cr4 != vmx->loaded_vmcs->host_state.cr4)) {
6563 		vmcs_writel(HOST_CR4, cr4);
6564 		vmx->loaded_vmcs->host_state.cr4 = cr4;
6565 	}
6566 
6567 	/* When single-stepping over STI and MOV SS, we must clear the
6568 	 * corresponding interruptibility bits in the guest state. Otherwise
6569 	 * vmentry fails as it then expects bit 14 (BS) in pending debug
6570 	 * exceptions being set, but that's not correct for the guest debugging
6571 	 * case. */
6572 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6573 		vmx_set_interrupt_shadow(vcpu, 0);
6574 
6575 	kvm_load_guest_xsave_state(vcpu);
6576 
6577 	if (static_cpu_has(X86_FEATURE_PKU) &&
6578 	    kvm_read_cr4_bits(vcpu, X86_CR4_PKE) &&
6579 	    vcpu->arch.pkru != vmx->host_pkru)
6580 		__write_pkru(vcpu->arch.pkru);
6581 
6582 	pt_guest_enter(vmx);
6583 
6584 	atomic_switch_perf_msrs(vmx);
6585 	atomic_switch_umwait_control_msr(vmx);
6586 
6587 	if (enable_preemption_timer)
6588 		vmx_update_hv_timer(vcpu);
6589 
6590 	if (lapic_in_kernel(vcpu) &&
6591 		vcpu->arch.apic->lapic_timer.timer_advance_ns)
6592 		kvm_wait_lapic_expire(vcpu);
6593 
6594 	/*
6595 	 * If this vCPU has touched SPEC_CTRL, restore the guest's value if
6596 	 * it's non-zero. Since vmentry is serialising on affected CPUs, there
6597 	 * is no need to worry about the conditional branch over the wrmsr
6598 	 * being speculatively taken.
6599 	 */
6600 	x86_spec_ctrl_set_guest(vmx->spec_ctrl, 0);
6601 
6602 	/* L1D Flush includes CPU buffer clear to mitigate MDS */
6603 	if (static_branch_unlikely(&vmx_l1d_should_flush))
6604 		vmx_l1d_flush(vcpu);
6605 	else if (static_branch_unlikely(&mds_user_clear))
6606 		mds_clear_cpu_buffers();
6607 
6608 	if (vcpu->arch.cr2 != read_cr2())
6609 		write_cr2(vcpu->arch.cr2);
6610 
6611 	vmx->fail = __vmx_vcpu_run(vmx, (unsigned long *)&vcpu->arch.regs,
6612 				   vmx->loaded_vmcs->launched);
6613 
6614 	vcpu->arch.cr2 = read_cr2();
6615 
6616 	/*
6617 	 * We do not use IBRS in the kernel. If this vCPU has used the
6618 	 * SPEC_CTRL MSR it may have left it on; save the value and
6619 	 * turn it off. This is much more efficient than blindly adding
6620 	 * it to the atomic save/restore list. Especially as the former
6621 	 * (Saving guest MSRs on vmexit) doesn't even exist in KVM.
6622 	 *
6623 	 * For non-nested case:
6624 	 * If the L01 MSR bitmap does not intercept the MSR, then we need to
6625 	 * save it.
6626 	 *
6627 	 * For nested case:
6628 	 * If the L02 MSR bitmap does not intercept the MSR, then we need to
6629 	 * save it.
6630 	 */
6631 	if (unlikely(!msr_write_intercepted(vcpu, MSR_IA32_SPEC_CTRL)))
6632 		vmx->spec_ctrl = native_read_msr(MSR_IA32_SPEC_CTRL);
6633 
6634 	x86_spec_ctrl_restore_host(vmx->spec_ctrl, 0);
6635 
6636 	/* All fields are clean at this point */
6637 	if (static_branch_unlikely(&enable_evmcs))
6638 		current_evmcs->hv_clean_fields |=
6639 			HV_VMX_ENLIGHTENED_CLEAN_FIELD_ALL;
6640 
6641 	if (static_branch_unlikely(&enable_evmcs))
6642 		current_evmcs->hv_vp_id = vcpu->arch.hyperv.vp_index;
6643 
6644 	/* MSR_IA32_DEBUGCTLMSR is zeroed on vmexit. Restore it if needed */
6645 	if (vmx->host_debugctlmsr)
6646 		update_debugctlmsr(vmx->host_debugctlmsr);
6647 
6648 #ifndef CONFIG_X86_64
6649 	/*
6650 	 * The sysexit path does not restore ds/es, so we must set them to
6651 	 * a reasonable value ourselves.
6652 	 *
6653 	 * We can't defer this to vmx_prepare_switch_to_host() since that
6654 	 * function may be executed in interrupt context, which saves and
6655 	 * restore segments around it, nullifying its effect.
6656 	 */
6657 	loadsegment(ds, __USER_DS);
6658 	loadsegment(es, __USER_DS);
6659 #endif
6660 
6661 	vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
6662 				  | (1 << VCPU_EXREG_RFLAGS)
6663 				  | (1 << VCPU_EXREG_PDPTR)
6664 				  | (1 << VCPU_EXREG_SEGMENTS)
6665 				  | (1 << VCPU_EXREG_CR3));
6666 	vcpu->arch.regs_dirty = 0;
6667 
6668 	pt_guest_exit(vmx);
6669 
6670 	/*
6671 	 * eager fpu is enabled if PKEY is supported and CR4 is switched
6672 	 * back on host, so it is safe to read guest PKRU from current
6673 	 * XSAVE.
6674 	 */
6675 	if (static_cpu_has(X86_FEATURE_PKU) &&
6676 	    kvm_read_cr4_bits(vcpu, X86_CR4_PKE)) {
6677 		vcpu->arch.pkru = rdpkru();
6678 		if (vcpu->arch.pkru != vmx->host_pkru)
6679 			__write_pkru(vmx->host_pkru);
6680 	}
6681 
6682 	kvm_load_host_xsave_state(vcpu);
6683 
6684 	vmx->nested.nested_run_pending = 0;
6685 	vmx->idt_vectoring_info = 0;
6686 
6687 	vmx->exit_reason = vmx->fail ? 0xdead : vmcs_read32(VM_EXIT_REASON);
6688 	if ((u16)vmx->exit_reason == EXIT_REASON_MCE_DURING_VMENTRY)
6689 		kvm_machine_check();
6690 
6691 	if (vmx->fail || (vmx->exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY))
6692 		return;
6693 
6694 	vmx->loaded_vmcs->launched = 1;
6695 	vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
6696 
6697 	vmx_recover_nmi_blocking(vmx);
6698 	vmx_complete_interrupts(vmx);
6699 }
6700 
6701 static struct kvm *vmx_vm_alloc(void)
6702 {
6703 	struct kvm_vmx *kvm_vmx = __vmalloc(sizeof(struct kvm_vmx),
6704 					    GFP_KERNEL_ACCOUNT | __GFP_ZERO,
6705 					    PAGE_KERNEL);
6706 	return &kvm_vmx->kvm;
6707 }
6708 
6709 static void vmx_vm_free(struct kvm *kvm)
6710 {
6711 	kfree(kvm->arch.hyperv.hv_pa_pg);
6712 	vfree(to_kvm_vmx(kvm));
6713 }
6714 
6715 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
6716 {
6717 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6718 
6719 	if (enable_pml)
6720 		vmx_destroy_pml_buffer(vmx);
6721 	free_vpid(vmx->vpid);
6722 	nested_vmx_free_vcpu(vcpu);
6723 	free_loaded_vmcs(vmx->loaded_vmcs);
6724 }
6725 
6726 static int vmx_create_vcpu(struct kvm_vcpu *vcpu)
6727 {
6728 	struct vcpu_vmx *vmx;
6729 	unsigned long *msr_bitmap;
6730 	int i, cpu, err;
6731 
6732 	BUILD_BUG_ON(offsetof(struct vcpu_vmx, vcpu) != 0);
6733 	vmx = to_vmx(vcpu);
6734 
6735 	err = -ENOMEM;
6736 
6737 	vmx->vpid = allocate_vpid();
6738 
6739 	/*
6740 	 * If PML is turned on, failure on enabling PML just results in failure
6741 	 * of creating the vcpu, therefore we can simplify PML logic (by
6742 	 * avoiding dealing with cases, such as enabling PML partially on vcpus
6743 	 * for the guest), etc.
6744 	 */
6745 	if (enable_pml) {
6746 		vmx->pml_pg = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
6747 		if (!vmx->pml_pg)
6748 			goto free_vpid;
6749 	}
6750 
6751 	BUILD_BUG_ON(ARRAY_SIZE(vmx_msr_index) != NR_SHARED_MSRS);
6752 
6753 	for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i) {
6754 		u32 index = vmx_msr_index[i];
6755 		u32 data_low, data_high;
6756 		int j = vmx->nmsrs;
6757 
6758 		if (rdmsr_safe(index, &data_low, &data_high) < 0)
6759 			continue;
6760 		if (wrmsr_safe(index, data_low, data_high) < 0)
6761 			continue;
6762 
6763 		vmx->guest_msrs[j].index = i;
6764 		vmx->guest_msrs[j].data = 0;
6765 		switch (index) {
6766 		case MSR_IA32_TSX_CTRL:
6767 			/*
6768 			 * No need to pass TSX_CTRL_CPUID_CLEAR through, so
6769 			 * let's avoid changing CPUID bits under the host
6770 			 * kernel's feet.
6771 			 */
6772 			vmx->guest_msrs[j].mask = ~(u64)TSX_CTRL_CPUID_CLEAR;
6773 			break;
6774 		default:
6775 			vmx->guest_msrs[j].mask = -1ull;
6776 			break;
6777 		}
6778 		++vmx->nmsrs;
6779 	}
6780 
6781 	err = alloc_loaded_vmcs(&vmx->vmcs01);
6782 	if (err < 0)
6783 		goto free_pml;
6784 
6785 	msr_bitmap = vmx->vmcs01.msr_bitmap;
6786 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_TSC, MSR_TYPE_R);
6787 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_FS_BASE, MSR_TYPE_RW);
6788 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_GS_BASE, MSR_TYPE_RW);
6789 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_KERNEL_GS_BASE, MSR_TYPE_RW);
6790 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_CS, MSR_TYPE_RW);
6791 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_ESP, MSR_TYPE_RW);
6792 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_EIP, MSR_TYPE_RW);
6793 	if (kvm_cstate_in_guest(vcpu->kvm)) {
6794 		vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C1_RES, MSR_TYPE_R);
6795 		vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C3_RESIDENCY, MSR_TYPE_R);
6796 		vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C6_RESIDENCY, MSR_TYPE_R);
6797 		vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C7_RESIDENCY, MSR_TYPE_R);
6798 	}
6799 	vmx->msr_bitmap_mode = 0;
6800 
6801 	vmx->loaded_vmcs = &vmx->vmcs01;
6802 	cpu = get_cpu();
6803 	vmx_vcpu_load(vcpu, cpu);
6804 	vcpu->cpu = cpu;
6805 	init_vmcs(vmx);
6806 	vmx_vcpu_put(vcpu);
6807 	put_cpu();
6808 	if (cpu_need_virtualize_apic_accesses(vcpu)) {
6809 		err = alloc_apic_access_page(vcpu->kvm);
6810 		if (err)
6811 			goto free_vmcs;
6812 	}
6813 
6814 	if (enable_ept && !enable_unrestricted_guest) {
6815 		err = init_rmode_identity_map(vcpu->kvm);
6816 		if (err)
6817 			goto free_vmcs;
6818 	}
6819 
6820 	if (nested)
6821 		nested_vmx_setup_ctls_msrs(&vmx->nested.msrs,
6822 					   vmx_capability.ept);
6823 	else
6824 		memset(&vmx->nested.msrs, 0, sizeof(vmx->nested.msrs));
6825 
6826 	vmx->nested.posted_intr_nv = -1;
6827 	vmx->nested.current_vmptr = -1ull;
6828 
6829 	vcpu->arch.microcode_version = 0x100000000ULL;
6830 	vmx->msr_ia32_feature_control_valid_bits = FEAT_CTL_LOCKED;
6831 
6832 	/*
6833 	 * Enforce invariant: pi_desc.nv is always either POSTED_INTR_VECTOR
6834 	 * or POSTED_INTR_WAKEUP_VECTOR.
6835 	 */
6836 	vmx->pi_desc.nv = POSTED_INTR_VECTOR;
6837 	vmx->pi_desc.sn = 1;
6838 
6839 	vmx->ept_pointer = INVALID_PAGE;
6840 
6841 	return 0;
6842 
6843 free_vmcs:
6844 	free_loaded_vmcs(vmx->loaded_vmcs);
6845 free_pml:
6846 	vmx_destroy_pml_buffer(vmx);
6847 free_vpid:
6848 	free_vpid(vmx->vpid);
6849 	return err;
6850 }
6851 
6852 #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"
6853 #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"
6854 
6855 static int vmx_vm_init(struct kvm *kvm)
6856 {
6857 	spin_lock_init(&to_kvm_vmx(kvm)->ept_pointer_lock);
6858 
6859 	if (!ple_gap)
6860 		kvm->arch.pause_in_guest = true;
6861 
6862 	if (boot_cpu_has(X86_BUG_L1TF) && enable_ept) {
6863 		switch (l1tf_mitigation) {
6864 		case L1TF_MITIGATION_OFF:
6865 		case L1TF_MITIGATION_FLUSH_NOWARN:
6866 			/* 'I explicitly don't care' is set */
6867 			break;
6868 		case L1TF_MITIGATION_FLUSH:
6869 		case L1TF_MITIGATION_FLUSH_NOSMT:
6870 		case L1TF_MITIGATION_FULL:
6871 			/*
6872 			 * Warn upon starting the first VM in a potentially
6873 			 * insecure environment.
6874 			 */
6875 			if (sched_smt_active())
6876 				pr_warn_once(L1TF_MSG_SMT);
6877 			if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_NEVER)
6878 				pr_warn_once(L1TF_MSG_L1D);
6879 			break;
6880 		case L1TF_MITIGATION_FULL_FORCE:
6881 			/* Flush is enforced */
6882 			break;
6883 		}
6884 	}
6885 	kvm_apicv_init(kvm, enable_apicv);
6886 	return 0;
6887 }
6888 
6889 static int __init vmx_check_processor_compat(void)
6890 {
6891 	struct vmcs_config vmcs_conf;
6892 	struct vmx_capability vmx_cap;
6893 
6894 	if (!this_cpu_has(X86_FEATURE_MSR_IA32_FEAT_CTL) ||
6895 	    !this_cpu_has(X86_FEATURE_VMX)) {
6896 		pr_err("kvm: VMX is disabled on CPU %d\n", smp_processor_id());
6897 		return -EIO;
6898 	}
6899 
6900 	if (setup_vmcs_config(&vmcs_conf, &vmx_cap) < 0)
6901 		return -EIO;
6902 	if (nested)
6903 		nested_vmx_setup_ctls_msrs(&vmcs_conf.nested, vmx_cap.ept);
6904 	if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
6905 		printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
6906 				smp_processor_id());
6907 		return -EIO;
6908 	}
6909 	return 0;
6910 }
6911 
6912 static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
6913 {
6914 	u8 cache;
6915 	u64 ipat = 0;
6916 
6917 	/* For VT-d and EPT combination
6918 	 * 1. MMIO: always map as UC
6919 	 * 2. EPT with VT-d:
6920 	 *   a. VT-d without snooping control feature: can't guarantee the
6921 	 *	result, try to trust guest.
6922 	 *   b. VT-d with snooping control feature: snooping control feature of
6923 	 *	VT-d engine can guarantee the cache correctness. Just set it
6924 	 *	to WB to keep consistent with host. So the same as item 3.
6925 	 * 3. EPT without VT-d: always map as WB and set IPAT=1 to keep
6926 	 *    consistent with host MTRR
6927 	 */
6928 	if (is_mmio) {
6929 		cache = MTRR_TYPE_UNCACHABLE;
6930 		goto exit;
6931 	}
6932 
6933 	if (!kvm_arch_has_noncoherent_dma(vcpu->kvm)) {
6934 		ipat = VMX_EPT_IPAT_BIT;
6935 		cache = MTRR_TYPE_WRBACK;
6936 		goto exit;
6937 	}
6938 
6939 	if (kvm_read_cr0(vcpu) & X86_CR0_CD) {
6940 		ipat = VMX_EPT_IPAT_BIT;
6941 		if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
6942 			cache = MTRR_TYPE_WRBACK;
6943 		else
6944 			cache = MTRR_TYPE_UNCACHABLE;
6945 		goto exit;
6946 	}
6947 
6948 	cache = kvm_mtrr_get_guest_memory_type(vcpu, gfn);
6949 
6950 exit:
6951 	return (cache << VMX_EPT_MT_EPTE_SHIFT) | ipat;
6952 }
6953 
6954 static int vmx_get_lpage_level(void)
6955 {
6956 	if (enable_ept && !cpu_has_vmx_ept_1g_page())
6957 		return PT_DIRECTORY_LEVEL;
6958 	else
6959 		/* For shadow and EPT supported 1GB page */
6960 		return PT_PDPE_LEVEL;
6961 }
6962 
6963 static void vmcs_set_secondary_exec_control(struct vcpu_vmx *vmx)
6964 {
6965 	/*
6966 	 * These bits in the secondary execution controls field
6967 	 * are dynamic, the others are mostly based on the hypervisor
6968 	 * architecture and the guest's CPUID.  Do not touch the
6969 	 * dynamic bits.
6970 	 */
6971 	u32 mask =
6972 		SECONDARY_EXEC_SHADOW_VMCS |
6973 		SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
6974 		SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
6975 		SECONDARY_EXEC_DESC;
6976 
6977 	u32 new_ctl = vmx->secondary_exec_control;
6978 	u32 cur_ctl = secondary_exec_controls_get(vmx);
6979 
6980 	secondary_exec_controls_set(vmx, (new_ctl & ~mask) | (cur_ctl & mask));
6981 }
6982 
6983 /*
6984  * Generate MSR_IA32_VMX_CR{0,4}_FIXED1 according to CPUID. Only set bits
6985  * (indicating "allowed-1") if they are supported in the guest's CPUID.
6986  */
6987 static void nested_vmx_cr_fixed1_bits_update(struct kvm_vcpu *vcpu)
6988 {
6989 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6990 	struct kvm_cpuid_entry2 *entry;
6991 
6992 	vmx->nested.msrs.cr0_fixed1 = 0xffffffff;
6993 	vmx->nested.msrs.cr4_fixed1 = X86_CR4_PCE;
6994 
6995 #define cr4_fixed1_update(_cr4_mask, _reg, _cpuid_mask) do {		\
6996 	if (entry && (entry->_reg & (_cpuid_mask)))			\
6997 		vmx->nested.msrs.cr4_fixed1 |= (_cr4_mask);	\
6998 } while (0)
6999 
7000 	entry = kvm_find_cpuid_entry(vcpu, 0x1, 0);
7001 	cr4_fixed1_update(X86_CR4_VME,        edx, feature_bit(VME));
7002 	cr4_fixed1_update(X86_CR4_PVI,        edx, feature_bit(VME));
7003 	cr4_fixed1_update(X86_CR4_TSD,        edx, feature_bit(TSC));
7004 	cr4_fixed1_update(X86_CR4_DE,         edx, feature_bit(DE));
7005 	cr4_fixed1_update(X86_CR4_PSE,        edx, feature_bit(PSE));
7006 	cr4_fixed1_update(X86_CR4_PAE,        edx, feature_bit(PAE));
7007 	cr4_fixed1_update(X86_CR4_MCE,        edx, feature_bit(MCE));
7008 	cr4_fixed1_update(X86_CR4_PGE,        edx, feature_bit(PGE));
7009 	cr4_fixed1_update(X86_CR4_OSFXSR,     edx, feature_bit(FXSR));
7010 	cr4_fixed1_update(X86_CR4_OSXMMEXCPT, edx, feature_bit(XMM));
7011 	cr4_fixed1_update(X86_CR4_VMXE,       ecx, feature_bit(VMX));
7012 	cr4_fixed1_update(X86_CR4_SMXE,       ecx, feature_bit(SMX));
7013 	cr4_fixed1_update(X86_CR4_PCIDE,      ecx, feature_bit(PCID));
7014 	cr4_fixed1_update(X86_CR4_OSXSAVE,    ecx, feature_bit(XSAVE));
7015 
7016 	entry = kvm_find_cpuid_entry(vcpu, 0x7, 0);
7017 	cr4_fixed1_update(X86_CR4_FSGSBASE,   ebx, feature_bit(FSGSBASE));
7018 	cr4_fixed1_update(X86_CR4_SMEP,       ebx, feature_bit(SMEP));
7019 	cr4_fixed1_update(X86_CR4_SMAP,       ebx, feature_bit(SMAP));
7020 	cr4_fixed1_update(X86_CR4_PKE,        ecx, feature_bit(PKU));
7021 	cr4_fixed1_update(X86_CR4_UMIP,       ecx, feature_bit(UMIP));
7022 	cr4_fixed1_update(X86_CR4_LA57,       ecx, feature_bit(LA57));
7023 
7024 #undef cr4_fixed1_update
7025 }
7026 
7027 static void nested_vmx_entry_exit_ctls_update(struct kvm_vcpu *vcpu)
7028 {
7029 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7030 
7031 	if (kvm_mpx_supported()) {
7032 		bool mpx_enabled = guest_cpuid_has(vcpu, X86_FEATURE_MPX);
7033 
7034 		if (mpx_enabled) {
7035 			vmx->nested.msrs.entry_ctls_high |= VM_ENTRY_LOAD_BNDCFGS;
7036 			vmx->nested.msrs.exit_ctls_high |= VM_EXIT_CLEAR_BNDCFGS;
7037 		} else {
7038 			vmx->nested.msrs.entry_ctls_high &= ~VM_ENTRY_LOAD_BNDCFGS;
7039 			vmx->nested.msrs.exit_ctls_high &= ~VM_EXIT_CLEAR_BNDCFGS;
7040 		}
7041 	}
7042 }
7043 
7044 static void update_intel_pt_cfg(struct kvm_vcpu *vcpu)
7045 {
7046 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7047 	struct kvm_cpuid_entry2 *best = NULL;
7048 	int i;
7049 
7050 	for (i = 0; i < PT_CPUID_LEAVES; i++) {
7051 		best = kvm_find_cpuid_entry(vcpu, 0x14, i);
7052 		if (!best)
7053 			return;
7054 		vmx->pt_desc.caps[CPUID_EAX + i*PT_CPUID_REGS_NUM] = best->eax;
7055 		vmx->pt_desc.caps[CPUID_EBX + i*PT_CPUID_REGS_NUM] = best->ebx;
7056 		vmx->pt_desc.caps[CPUID_ECX + i*PT_CPUID_REGS_NUM] = best->ecx;
7057 		vmx->pt_desc.caps[CPUID_EDX + i*PT_CPUID_REGS_NUM] = best->edx;
7058 	}
7059 
7060 	/* Get the number of configurable Address Ranges for filtering */
7061 	vmx->pt_desc.addr_range = intel_pt_validate_cap(vmx->pt_desc.caps,
7062 						PT_CAP_num_address_ranges);
7063 
7064 	/* Initialize and clear the no dependency bits */
7065 	vmx->pt_desc.ctl_bitmask = ~(RTIT_CTL_TRACEEN | RTIT_CTL_OS |
7066 			RTIT_CTL_USR | RTIT_CTL_TSC_EN | RTIT_CTL_DISRETC);
7067 
7068 	/*
7069 	 * If CPUID.(EAX=14H,ECX=0):EBX[0]=1 CR3Filter can be set otherwise
7070 	 * will inject an #GP
7071 	 */
7072 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_cr3_filtering))
7073 		vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_CR3EN;
7074 
7075 	/*
7076 	 * If CPUID.(EAX=14H,ECX=0):EBX[1]=1 CYCEn, CycThresh and
7077 	 * PSBFreq can be set
7078 	 */
7079 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc))
7080 		vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_CYCLEACC |
7081 				RTIT_CTL_CYC_THRESH | RTIT_CTL_PSB_FREQ);
7082 
7083 	/*
7084 	 * If CPUID.(EAX=14H,ECX=0):EBX[3]=1 MTCEn BranchEn and
7085 	 * MTCFreq can be set
7086 	 */
7087 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc))
7088 		vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_MTC_EN |
7089 				RTIT_CTL_BRANCH_EN | RTIT_CTL_MTC_RANGE);
7090 
7091 	/* If CPUID.(EAX=14H,ECX=0):EBX[4]=1 FUPonPTW and PTWEn can be set */
7092 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_ptwrite))
7093 		vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_FUP_ON_PTW |
7094 							RTIT_CTL_PTW_EN);
7095 
7096 	/* If CPUID.(EAX=14H,ECX=0):EBX[5]=1 PwrEvEn can be set */
7097 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_power_event_trace))
7098 		vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_PWR_EVT_EN;
7099 
7100 	/* If CPUID.(EAX=14H,ECX=0):ECX[0]=1 ToPA can be set */
7101 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_topa_output))
7102 		vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_TOPA;
7103 
7104 	/* If CPUID.(EAX=14H,ECX=0):ECX[3]=1 FabircEn can be set */
7105 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_output_subsys))
7106 		vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_FABRIC_EN;
7107 
7108 	/* unmask address range configure area */
7109 	for (i = 0; i < vmx->pt_desc.addr_range; i++)
7110 		vmx->pt_desc.ctl_bitmask &= ~(0xfULL << (32 + i * 4));
7111 }
7112 
7113 static void vmx_cpuid_update(struct kvm_vcpu *vcpu)
7114 {
7115 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7116 
7117 	/* xsaves_enabled is recomputed in vmx_compute_secondary_exec_control(). */
7118 	vcpu->arch.xsaves_enabled = false;
7119 
7120 	if (cpu_has_secondary_exec_ctrls()) {
7121 		vmx_compute_secondary_exec_control(vmx);
7122 		vmcs_set_secondary_exec_control(vmx);
7123 	}
7124 
7125 	if (nested_vmx_allowed(vcpu))
7126 		to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
7127 			FEAT_CTL_VMX_ENABLED_INSIDE_SMX |
7128 			FEAT_CTL_VMX_ENABLED_OUTSIDE_SMX;
7129 	else
7130 		to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
7131 			~(FEAT_CTL_VMX_ENABLED_INSIDE_SMX |
7132 			  FEAT_CTL_VMX_ENABLED_OUTSIDE_SMX);
7133 
7134 	if (nested_vmx_allowed(vcpu)) {
7135 		nested_vmx_cr_fixed1_bits_update(vcpu);
7136 		nested_vmx_entry_exit_ctls_update(vcpu);
7137 	}
7138 
7139 	if (boot_cpu_has(X86_FEATURE_INTEL_PT) &&
7140 			guest_cpuid_has(vcpu, X86_FEATURE_INTEL_PT))
7141 		update_intel_pt_cfg(vcpu);
7142 
7143 	if (boot_cpu_has(X86_FEATURE_RTM)) {
7144 		struct shared_msr_entry *msr;
7145 		msr = find_msr_entry(vmx, MSR_IA32_TSX_CTRL);
7146 		if (msr) {
7147 			bool enabled = guest_cpuid_has(vcpu, X86_FEATURE_RTM);
7148 			vmx_set_guest_msr(vmx, msr, enabled ? 0 : TSX_CTRL_RTM_DISABLE);
7149 		}
7150 	}
7151 }
7152 
7153 static void vmx_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
7154 {
7155 	if (func == 1 && nested)
7156 		entry->ecx |= feature_bit(VMX);
7157 }
7158 
7159 static void vmx_request_immediate_exit(struct kvm_vcpu *vcpu)
7160 {
7161 	to_vmx(vcpu)->req_immediate_exit = true;
7162 }
7163 
7164 static int vmx_check_intercept_io(struct kvm_vcpu *vcpu,
7165 				  struct x86_instruction_info *info)
7166 {
7167 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
7168 	unsigned short port;
7169 	bool intercept;
7170 	int size;
7171 
7172 	if (info->intercept == x86_intercept_in ||
7173 	    info->intercept == x86_intercept_ins) {
7174 		port = info->src_val;
7175 		size = info->dst_bytes;
7176 	} else {
7177 		port = info->dst_val;
7178 		size = info->src_bytes;
7179 	}
7180 
7181 	/*
7182 	 * If the 'use IO bitmaps' VM-execution control is 0, IO instruction
7183 	 * VM-exits depend on the 'unconditional IO exiting' VM-execution
7184 	 * control.
7185 	 *
7186 	 * Otherwise, IO instruction VM-exits are controlled by the IO bitmaps.
7187 	 */
7188 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
7189 		intercept = nested_cpu_has(vmcs12,
7190 					   CPU_BASED_UNCOND_IO_EXITING);
7191 	else
7192 		intercept = nested_vmx_check_io_bitmaps(vcpu, port, size);
7193 
7194 	/* FIXME: produce nested vmexit and return X86EMUL_INTERCEPTED.  */
7195 	return intercept ? X86EMUL_UNHANDLEABLE : X86EMUL_CONTINUE;
7196 }
7197 
7198 static int vmx_check_intercept(struct kvm_vcpu *vcpu,
7199 			       struct x86_instruction_info *info,
7200 			       enum x86_intercept_stage stage)
7201 {
7202 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
7203 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
7204 
7205 	switch (info->intercept) {
7206 	/*
7207 	 * RDPID causes #UD if disabled through secondary execution controls.
7208 	 * Because it is marked as EmulateOnUD, we need to intercept it here.
7209 	 */
7210 	case x86_intercept_rdtscp:
7211 		if (!nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDTSCP)) {
7212 			ctxt->exception.vector = UD_VECTOR;
7213 			ctxt->exception.error_code_valid = false;
7214 			return X86EMUL_PROPAGATE_FAULT;
7215 		}
7216 		break;
7217 
7218 	case x86_intercept_in:
7219 	case x86_intercept_ins:
7220 	case x86_intercept_out:
7221 	case x86_intercept_outs:
7222 		return vmx_check_intercept_io(vcpu, info);
7223 
7224 	case x86_intercept_lgdt:
7225 	case x86_intercept_lidt:
7226 	case x86_intercept_lldt:
7227 	case x86_intercept_ltr:
7228 	case x86_intercept_sgdt:
7229 	case x86_intercept_sidt:
7230 	case x86_intercept_sldt:
7231 	case x86_intercept_str:
7232 		if (!nested_cpu_has2(vmcs12, SECONDARY_EXEC_DESC))
7233 			return X86EMUL_CONTINUE;
7234 
7235 		/* FIXME: produce nested vmexit and return X86EMUL_INTERCEPTED.  */
7236 		break;
7237 
7238 	/* TODO: check more intercepts... */
7239 	default:
7240 		break;
7241 	}
7242 
7243 	return X86EMUL_UNHANDLEABLE;
7244 }
7245 
7246 #ifdef CONFIG_X86_64
7247 /* (a << shift) / divisor, return 1 if overflow otherwise 0 */
7248 static inline int u64_shl_div_u64(u64 a, unsigned int shift,
7249 				  u64 divisor, u64 *result)
7250 {
7251 	u64 low = a << shift, high = a >> (64 - shift);
7252 
7253 	/* To avoid the overflow on divq */
7254 	if (high >= divisor)
7255 		return 1;
7256 
7257 	/* Low hold the result, high hold rem which is discarded */
7258 	asm("divq %2\n\t" : "=a" (low), "=d" (high) :
7259 	    "rm" (divisor), "0" (low), "1" (high));
7260 	*result = low;
7261 
7262 	return 0;
7263 }
7264 
7265 static int vmx_set_hv_timer(struct kvm_vcpu *vcpu, u64 guest_deadline_tsc,
7266 			    bool *expired)
7267 {
7268 	struct vcpu_vmx *vmx;
7269 	u64 tscl, guest_tscl, delta_tsc, lapic_timer_advance_cycles;
7270 	struct kvm_timer *ktimer = &vcpu->arch.apic->lapic_timer;
7271 
7272 	if (kvm_mwait_in_guest(vcpu->kvm) ||
7273 		kvm_can_post_timer_interrupt(vcpu))
7274 		return -EOPNOTSUPP;
7275 
7276 	vmx = to_vmx(vcpu);
7277 	tscl = rdtsc();
7278 	guest_tscl = kvm_read_l1_tsc(vcpu, tscl);
7279 	delta_tsc = max(guest_deadline_tsc, guest_tscl) - guest_tscl;
7280 	lapic_timer_advance_cycles = nsec_to_cycles(vcpu,
7281 						    ktimer->timer_advance_ns);
7282 
7283 	if (delta_tsc > lapic_timer_advance_cycles)
7284 		delta_tsc -= lapic_timer_advance_cycles;
7285 	else
7286 		delta_tsc = 0;
7287 
7288 	/* Convert to host delta tsc if tsc scaling is enabled */
7289 	if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio &&
7290 	    delta_tsc && u64_shl_div_u64(delta_tsc,
7291 				kvm_tsc_scaling_ratio_frac_bits,
7292 				vcpu->arch.tsc_scaling_ratio, &delta_tsc))
7293 		return -ERANGE;
7294 
7295 	/*
7296 	 * If the delta tsc can't fit in the 32 bit after the multi shift,
7297 	 * we can't use the preemption timer.
7298 	 * It's possible that it fits on later vmentries, but checking
7299 	 * on every vmentry is costly so we just use an hrtimer.
7300 	 */
7301 	if (delta_tsc >> (cpu_preemption_timer_multi + 32))
7302 		return -ERANGE;
7303 
7304 	vmx->hv_deadline_tsc = tscl + delta_tsc;
7305 	*expired = !delta_tsc;
7306 	return 0;
7307 }
7308 
7309 static void vmx_cancel_hv_timer(struct kvm_vcpu *vcpu)
7310 {
7311 	to_vmx(vcpu)->hv_deadline_tsc = -1;
7312 }
7313 #endif
7314 
7315 static void vmx_sched_in(struct kvm_vcpu *vcpu, int cpu)
7316 {
7317 	if (!kvm_pause_in_guest(vcpu->kvm))
7318 		shrink_ple_window(vcpu);
7319 }
7320 
7321 static void vmx_slot_enable_log_dirty(struct kvm *kvm,
7322 				     struct kvm_memory_slot *slot)
7323 {
7324 	kvm_mmu_slot_leaf_clear_dirty(kvm, slot);
7325 	kvm_mmu_slot_largepage_remove_write_access(kvm, slot);
7326 }
7327 
7328 static void vmx_slot_disable_log_dirty(struct kvm *kvm,
7329 				       struct kvm_memory_slot *slot)
7330 {
7331 	kvm_mmu_slot_set_dirty(kvm, slot);
7332 }
7333 
7334 static void vmx_flush_log_dirty(struct kvm *kvm)
7335 {
7336 	kvm_flush_pml_buffers(kvm);
7337 }
7338 
7339 static int vmx_write_pml_buffer(struct kvm_vcpu *vcpu)
7340 {
7341 	struct vmcs12 *vmcs12;
7342 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7343 	gpa_t gpa, dst;
7344 
7345 	if (is_guest_mode(vcpu)) {
7346 		WARN_ON_ONCE(vmx->nested.pml_full);
7347 
7348 		/*
7349 		 * Check if PML is enabled for the nested guest.
7350 		 * Whether eptp bit 6 is set is already checked
7351 		 * as part of A/D emulation.
7352 		 */
7353 		vmcs12 = get_vmcs12(vcpu);
7354 		if (!nested_cpu_has_pml(vmcs12))
7355 			return 0;
7356 
7357 		if (vmcs12->guest_pml_index >= PML_ENTITY_NUM) {
7358 			vmx->nested.pml_full = true;
7359 			return 1;
7360 		}
7361 
7362 		gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS) & ~0xFFFull;
7363 		dst = vmcs12->pml_address + sizeof(u64) * vmcs12->guest_pml_index;
7364 
7365 		if (kvm_write_guest_page(vcpu->kvm, gpa_to_gfn(dst), &gpa,
7366 					 offset_in_page(dst), sizeof(gpa)))
7367 			return 0;
7368 
7369 		vmcs12->guest_pml_index--;
7370 	}
7371 
7372 	return 0;
7373 }
7374 
7375 static void vmx_enable_log_dirty_pt_masked(struct kvm *kvm,
7376 					   struct kvm_memory_slot *memslot,
7377 					   gfn_t offset, unsigned long mask)
7378 {
7379 	kvm_mmu_clear_dirty_pt_masked(kvm, memslot, offset, mask);
7380 }
7381 
7382 static void __pi_post_block(struct kvm_vcpu *vcpu)
7383 {
7384 	struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
7385 	struct pi_desc old, new;
7386 	unsigned int dest;
7387 
7388 	do {
7389 		old.control = new.control = pi_desc->control;
7390 		WARN(old.nv != POSTED_INTR_WAKEUP_VECTOR,
7391 		     "Wakeup handler not enabled while the VCPU is blocked\n");
7392 
7393 		dest = cpu_physical_id(vcpu->cpu);
7394 
7395 		if (x2apic_enabled())
7396 			new.ndst = dest;
7397 		else
7398 			new.ndst = (dest << 8) & 0xFF00;
7399 
7400 		/* set 'NV' to 'notification vector' */
7401 		new.nv = POSTED_INTR_VECTOR;
7402 	} while (cmpxchg64(&pi_desc->control, old.control,
7403 			   new.control) != old.control);
7404 
7405 	if (!WARN_ON_ONCE(vcpu->pre_pcpu == -1)) {
7406 		spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7407 		list_del(&vcpu->blocked_vcpu_list);
7408 		spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7409 		vcpu->pre_pcpu = -1;
7410 	}
7411 }
7412 
7413 /*
7414  * This routine does the following things for vCPU which is going
7415  * to be blocked if VT-d PI is enabled.
7416  * - Store the vCPU to the wakeup list, so when interrupts happen
7417  *   we can find the right vCPU to wake up.
7418  * - Change the Posted-interrupt descriptor as below:
7419  *      'NDST' <-- vcpu->pre_pcpu
7420  *      'NV' <-- POSTED_INTR_WAKEUP_VECTOR
7421  * - If 'ON' is set during this process, which means at least one
7422  *   interrupt is posted for this vCPU, we cannot block it, in
7423  *   this case, return 1, otherwise, return 0.
7424  *
7425  */
7426 static int pi_pre_block(struct kvm_vcpu *vcpu)
7427 {
7428 	unsigned int dest;
7429 	struct pi_desc old, new;
7430 	struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
7431 
7432 	if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
7433 		!irq_remapping_cap(IRQ_POSTING_CAP)  ||
7434 		!kvm_vcpu_apicv_active(vcpu))
7435 		return 0;
7436 
7437 	WARN_ON(irqs_disabled());
7438 	local_irq_disable();
7439 	if (!WARN_ON_ONCE(vcpu->pre_pcpu != -1)) {
7440 		vcpu->pre_pcpu = vcpu->cpu;
7441 		spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7442 		list_add_tail(&vcpu->blocked_vcpu_list,
7443 			      &per_cpu(blocked_vcpu_on_cpu,
7444 				       vcpu->pre_pcpu));
7445 		spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7446 	}
7447 
7448 	do {
7449 		old.control = new.control = pi_desc->control;
7450 
7451 		WARN((pi_desc->sn == 1),
7452 		     "Warning: SN field of posted-interrupts "
7453 		     "is set before blocking\n");
7454 
7455 		/*
7456 		 * Since vCPU can be preempted during this process,
7457 		 * vcpu->cpu could be different with pre_pcpu, we
7458 		 * need to set pre_pcpu as the destination of wakeup
7459 		 * notification event, then we can find the right vCPU
7460 		 * to wakeup in wakeup handler if interrupts happen
7461 		 * when the vCPU is in blocked state.
7462 		 */
7463 		dest = cpu_physical_id(vcpu->pre_pcpu);
7464 
7465 		if (x2apic_enabled())
7466 			new.ndst = dest;
7467 		else
7468 			new.ndst = (dest << 8) & 0xFF00;
7469 
7470 		/* set 'NV' to 'wakeup vector' */
7471 		new.nv = POSTED_INTR_WAKEUP_VECTOR;
7472 	} while (cmpxchg64(&pi_desc->control, old.control,
7473 			   new.control) != old.control);
7474 
7475 	/* We should not block the vCPU if an interrupt is posted for it.  */
7476 	if (pi_test_on(pi_desc) == 1)
7477 		__pi_post_block(vcpu);
7478 
7479 	local_irq_enable();
7480 	return (vcpu->pre_pcpu == -1);
7481 }
7482 
7483 static int vmx_pre_block(struct kvm_vcpu *vcpu)
7484 {
7485 	if (pi_pre_block(vcpu))
7486 		return 1;
7487 
7488 	if (kvm_lapic_hv_timer_in_use(vcpu))
7489 		kvm_lapic_switch_to_sw_timer(vcpu);
7490 
7491 	return 0;
7492 }
7493 
7494 static void pi_post_block(struct kvm_vcpu *vcpu)
7495 {
7496 	if (vcpu->pre_pcpu == -1)
7497 		return;
7498 
7499 	WARN_ON(irqs_disabled());
7500 	local_irq_disable();
7501 	__pi_post_block(vcpu);
7502 	local_irq_enable();
7503 }
7504 
7505 static void vmx_post_block(struct kvm_vcpu *vcpu)
7506 {
7507 	if (kvm_x86_ops->set_hv_timer)
7508 		kvm_lapic_switch_to_hv_timer(vcpu);
7509 
7510 	pi_post_block(vcpu);
7511 }
7512 
7513 /*
7514  * vmx_update_pi_irte - set IRTE for Posted-Interrupts
7515  *
7516  * @kvm: kvm
7517  * @host_irq: host irq of the interrupt
7518  * @guest_irq: gsi of the interrupt
7519  * @set: set or unset PI
7520  * returns 0 on success, < 0 on failure
7521  */
7522 static int vmx_update_pi_irte(struct kvm *kvm, unsigned int host_irq,
7523 			      uint32_t guest_irq, bool set)
7524 {
7525 	struct kvm_kernel_irq_routing_entry *e;
7526 	struct kvm_irq_routing_table *irq_rt;
7527 	struct kvm_lapic_irq irq;
7528 	struct kvm_vcpu *vcpu;
7529 	struct vcpu_data vcpu_info;
7530 	int idx, ret = 0;
7531 
7532 	if (!kvm_arch_has_assigned_device(kvm) ||
7533 		!irq_remapping_cap(IRQ_POSTING_CAP) ||
7534 		!kvm_vcpu_apicv_active(kvm->vcpus[0]))
7535 		return 0;
7536 
7537 	idx = srcu_read_lock(&kvm->irq_srcu);
7538 	irq_rt = srcu_dereference(kvm->irq_routing, &kvm->irq_srcu);
7539 	if (guest_irq >= irq_rt->nr_rt_entries ||
7540 	    hlist_empty(&irq_rt->map[guest_irq])) {
7541 		pr_warn_once("no route for guest_irq %u/%u (broken user space?)\n",
7542 			     guest_irq, irq_rt->nr_rt_entries);
7543 		goto out;
7544 	}
7545 
7546 	hlist_for_each_entry(e, &irq_rt->map[guest_irq], link) {
7547 		if (e->type != KVM_IRQ_ROUTING_MSI)
7548 			continue;
7549 		/*
7550 		 * VT-d PI cannot support posting multicast/broadcast
7551 		 * interrupts to a vCPU, we still use interrupt remapping
7552 		 * for these kind of interrupts.
7553 		 *
7554 		 * For lowest-priority interrupts, we only support
7555 		 * those with single CPU as the destination, e.g. user
7556 		 * configures the interrupts via /proc/irq or uses
7557 		 * irqbalance to make the interrupts single-CPU.
7558 		 *
7559 		 * We will support full lowest-priority interrupt later.
7560 		 *
7561 		 * In addition, we can only inject generic interrupts using
7562 		 * the PI mechanism, refuse to route others through it.
7563 		 */
7564 
7565 		kvm_set_msi_irq(kvm, e, &irq);
7566 		if (!kvm_intr_is_single_vcpu(kvm, &irq, &vcpu) ||
7567 		    !kvm_irq_is_postable(&irq)) {
7568 			/*
7569 			 * Make sure the IRTE is in remapped mode if
7570 			 * we don't handle it in posted mode.
7571 			 */
7572 			ret = irq_set_vcpu_affinity(host_irq, NULL);
7573 			if (ret < 0) {
7574 				printk(KERN_INFO
7575 				   "failed to back to remapped mode, irq: %u\n",
7576 				   host_irq);
7577 				goto out;
7578 			}
7579 
7580 			continue;
7581 		}
7582 
7583 		vcpu_info.pi_desc_addr = __pa(vcpu_to_pi_desc(vcpu));
7584 		vcpu_info.vector = irq.vector;
7585 
7586 		trace_kvm_pi_irte_update(host_irq, vcpu->vcpu_id, e->gsi,
7587 				vcpu_info.vector, vcpu_info.pi_desc_addr, set);
7588 
7589 		if (set)
7590 			ret = irq_set_vcpu_affinity(host_irq, &vcpu_info);
7591 		else
7592 			ret = irq_set_vcpu_affinity(host_irq, NULL);
7593 
7594 		if (ret < 0) {
7595 			printk(KERN_INFO "%s: failed to update PI IRTE\n",
7596 					__func__);
7597 			goto out;
7598 		}
7599 	}
7600 
7601 	ret = 0;
7602 out:
7603 	srcu_read_unlock(&kvm->irq_srcu, idx);
7604 	return ret;
7605 }
7606 
7607 static void vmx_setup_mce(struct kvm_vcpu *vcpu)
7608 {
7609 	if (vcpu->arch.mcg_cap & MCG_LMCE_P)
7610 		to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
7611 			FEAT_CTL_LMCE_ENABLED;
7612 	else
7613 		to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
7614 			~FEAT_CTL_LMCE_ENABLED;
7615 }
7616 
7617 static int vmx_smi_allowed(struct kvm_vcpu *vcpu)
7618 {
7619 	/* we need a nested vmexit to enter SMM, postpone if run is pending */
7620 	if (to_vmx(vcpu)->nested.nested_run_pending)
7621 		return 0;
7622 	return 1;
7623 }
7624 
7625 static int vmx_pre_enter_smm(struct kvm_vcpu *vcpu, char *smstate)
7626 {
7627 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7628 
7629 	vmx->nested.smm.guest_mode = is_guest_mode(vcpu);
7630 	if (vmx->nested.smm.guest_mode)
7631 		nested_vmx_vmexit(vcpu, -1, 0, 0);
7632 
7633 	vmx->nested.smm.vmxon = vmx->nested.vmxon;
7634 	vmx->nested.vmxon = false;
7635 	vmx_clear_hlt(vcpu);
7636 	return 0;
7637 }
7638 
7639 static int vmx_pre_leave_smm(struct kvm_vcpu *vcpu, const char *smstate)
7640 {
7641 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7642 	int ret;
7643 
7644 	if (vmx->nested.smm.vmxon) {
7645 		vmx->nested.vmxon = true;
7646 		vmx->nested.smm.vmxon = false;
7647 	}
7648 
7649 	if (vmx->nested.smm.guest_mode) {
7650 		ret = nested_vmx_enter_non_root_mode(vcpu, false);
7651 		if (ret)
7652 			return ret;
7653 
7654 		vmx->nested.smm.guest_mode = false;
7655 	}
7656 	return 0;
7657 }
7658 
7659 static int enable_smi_window(struct kvm_vcpu *vcpu)
7660 {
7661 	return 0;
7662 }
7663 
7664 static bool vmx_need_emulation_on_page_fault(struct kvm_vcpu *vcpu)
7665 {
7666 	return false;
7667 }
7668 
7669 static bool vmx_apic_init_signal_blocked(struct kvm_vcpu *vcpu)
7670 {
7671 	return to_vmx(vcpu)->nested.vmxon;
7672 }
7673 
7674 static __init int hardware_setup(void)
7675 {
7676 	unsigned long host_bndcfgs;
7677 	struct desc_ptr dt;
7678 	int r, i;
7679 
7680 	rdmsrl_safe(MSR_EFER, &host_efer);
7681 
7682 	store_idt(&dt);
7683 	host_idt_base = dt.address;
7684 
7685 	for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i)
7686 		kvm_define_shared_msr(i, vmx_msr_index[i]);
7687 
7688 	if (setup_vmcs_config(&vmcs_config, &vmx_capability) < 0)
7689 		return -EIO;
7690 
7691 	if (boot_cpu_has(X86_FEATURE_NX))
7692 		kvm_enable_efer_bits(EFER_NX);
7693 
7694 	if (boot_cpu_has(X86_FEATURE_MPX)) {
7695 		rdmsrl(MSR_IA32_BNDCFGS, host_bndcfgs);
7696 		WARN_ONCE(host_bndcfgs, "KVM: BNDCFGS in host will be lost");
7697 	}
7698 
7699 	if (!cpu_has_vmx_vpid() || !cpu_has_vmx_invvpid() ||
7700 	    !(cpu_has_vmx_invvpid_single() || cpu_has_vmx_invvpid_global()))
7701 		enable_vpid = 0;
7702 
7703 	if (!cpu_has_vmx_ept() ||
7704 	    !cpu_has_vmx_ept_4levels() ||
7705 	    !cpu_has_vmx_ept_mt_wb() ||
7706 	    !cpu_has_vmx_invept_global())
7707 		enable_ept = 0;
7708 
7709 	if (!cpu_has_vmx_ept_ad_bits() || !enable_ept)
7710 		enable_ept_ad_bits = 0;
7711 
7712 	if (!cpu_has_vmx_unrestricted_guest() || !enable_ept)
7713 		enable_unrestricted_guest = 0;
7714 
7715 	if (!cpu_has_vmx_flexpriority())
7716 		flexpriority_enabled = 0;
7717 
7718 	if (!cpu_has_virtual_nmis())
7719 		enable_vnmi = 0;
7720 
7721 	/*
7722 	 * set_apic_access_page_addr() is used to reload apic access
7723 	 * page upon invalidation.  No need to do anything if not
7724 	 * using the APIC_ACCESS_ADDR VMCS field.
7725 	 */
7726 	if (!flexpriority_enabled)
7727 		kvm_x86_ops->set_apic_access_page_addr = NULL;
7728 
7729 	if (!cpu_has_vmx_tpr_shadow())
7730 		kvm_x86_ops->update_cr8_intercept = NULL;
7731 
7732 	if (enable_ept && !cpu_has_vmx_ept_2m_page())
7733 		kvm_disable_largepages();
7734 
7735 #if IS_ENABLED(CONFIG_HYPERV)
7736 	if (ms_hyperv.nested_features & HV_X64_NESTED_GUEST_MAPPING_FLUSH
7737 	    && enable_ept) {
7738 		kvm_x86_ops->tlb_remote_flush = hv_remote_flush_tlb;
7739 		kvm_x86_ops->tlb_remote_flush_with_range =
7740 				hv_remote_flush_tlb_with_range;
7741 	}
7742 #endif
7743 
7744 	if (!cpu_has_vmx_ple()) {
7745 		ple_gap = 0;
7746 		ple_window = 0;
7747 		ple_window_grow = 0;
7748 		ple_window_max = 0;
7749 		ple_window_shrink = 0;
7750 	}
7751 
7752 	if (!cpu_has_vmx_apicv()) {
7753 		enable_apicv = 0;
7754 		kvm_x86_ops->sync_pir_to_irr = NULL;
7755 	}
7756 
7757 	if (cpu_has_vmx_tsc_scaling()) {
7758 		kvm_has_tsc_control = true;
7759 		kvm_max_tsc_scaling_ratio = KVM_VMX_TSC_MULTIPLIER_MAX;
7760 		kvm_tsc_scaling_ratio_frac_bits = 48;
7761 	}
7762 
7763 	set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
7764 
7765 	if (enable_ept)
7766 		vmx_enable_tdp();
7767 	else
7768 		kvm_disable_tdp();
7769 
7770 	/*
7771 	 * Only enable PML when hardware supports PML feature, and both EPT
7772 	 * and EPT A/D bit features are enabled -- PML depends on them to work.
7773 	 */
7774 	if (!enable_ept || !enable_ept_ad_bits || !cpu_has_vmx_pml())
7775 		enable_pml = 0;
7776 
7777 	if (!enable_pml) {
7778 		kvm_x86_ops->slot_enable_log_dirty = NULL;
7779 		kvm_x86_ops->slot_disable_log_dirty = NULL;
7780 		kvm_x86_ops->flush_log_dirty = NULL;
7781 		kvm_x86_ops->enable_log_dirty_pt_masked = NULL;
7782 	}
7783 
7784 	if (!cpu_has_vmx_preemption_timer())
7785 		enable_preemption_timer = false;
7786 
7787 	if (enable_preemption_timer) {
7788 		u64 use_timer_freq = 5000ULL * 1000 * 1000;
7789 		u64 vmx_msr;
7790 
7791 		rdmsrl(MSR_IA32_VMX_MISC, vmx_msr);
7792 		cpu_preemption_timer_multi =
7793 			vmx_msr & VMX_MISC_PREEMPTION_TIMER_RATE_MASK;
7794 
7795 		if (tsc_khz)
7796 			use_timer_freq = (u64)tsc_khz * 1000;
7797 		use_timer_freq >>= cpu_preemption_timer_multi;
7798 
7799 		/*
7800 		 * KVM "disables" the preemption timer by setting it to its max
7801 		 * value.  Don't use the timer if it might cause spurious exits
7802 		 * at a rate faster than 0.1 Hz (of uninterrupted guest time).
7803 		 */
7804 		if (use_timer_freq > 0xffffffffu / 10)
7805 			enable_preemption_timer = false;
7806 	}
7807 
7808 	if (!enable_preemption_timer) {
7809 		kvm_x86_ops->set_hv_timer = NULL;
7810 		kvm_x86_ops->cancel_hv_timer = NULL;
7811 		kvm_x86_ops->request_immediate_exit = __kvm_request_immediate_exit;
7812 	}
7813 
7814 	kvm_set_posted_intr_wakeup_handler(wakeup_handler);
7815 
7816 	kvm_mce_cap_supported |= MCG_LMCE_P;
7817 
7818 	if (pt_mode != PT_MODE_SYSTEM && pt_mode != PT_MODE_HOST_GUEST)
7819 		return -EINVAL;
7820 	if (!enable_ept || !cpu_has_vmx_intel_pt())
7821 		pt_mode = PT_MODE_SYSTEM;
7822 
7823 	if (nested) {
7824 		nested_vmx_setup_ctls_msrs(&vmcs_config.nested,
7825 					   vmx_capability.ept);
7826 
7827 		r = nested_vmx_hardware_setup(kvm_vmx_exit_handlers);
7828 		if (r)
7829 			return r;
7830 	}
7831 
7832 	r = alloc_kvm_area();
7833 	if (r)
7834 		nested_vmx_hardware_unsetup();
7835 	return r;
7836 }
7837 
7838 static __exit void hardware_unsetup(void)
7839 {
7840 	if (nested)
7841 		nested_vmx_hardware_unsetup();
7842 
7843 	free_kvm_area();
7844 }
7845 
7846 static bool vmx_check_apicv_inhibit_reasons(ulong bit)
7847 {
7848 	ulong supported = BIT(APICV_INHIBIT_REASON_DISABLE) |
7849 			  BIT(APICV_INHIBIT_REASON_HYPERV);
7850 
7851 	return supported & BIT(bit);
7852 }
7853 
7854 static struct kvm_x86_ops vmx_x86_ops __ro_after_init = {
7855 	.cpu_has_kvm_support = cpu_has_kvm_support,
7856 	.disabled_by_bios = vmx_disabled_by_bios,
7857 	.hardware_setup = hardware_setup,
7858 	.hardware_unsetup = hardware_unsetup,
7859 	.check_processor_compatibility = vmx_check_processor_compat,
7860 	.hardware_enable = hardware_enable,
7861 	.hardware_disable = hardware_disable,
7862 	.cpu_has_accelerated_tpr = report_flexpriority,
7863 	.has_emulated_msr = vmx_has_emulated_msr,
7864 
7865 	.vm_init = vmx_vm_init,
7866 	.vm_alloc = vmx_vm_alloc,
7867 	.vm_free = vmx_vm_free,
7868 
7869 	.vcpu_create = vmx_create_vcpu,
7870 	.vcpu_free = vmx_free_vcpu,
7871 	.vcpu_reset = vmx_vcpu_reset,
7872 
7873 	.prepare_guest_switch = vmx_prepare_switch_to_guest,
7874 	.vcpu_load = vmx_vcpu_load,
7875 	.vcpu_put = vmx_vcpu_put,
7876 
7877 	.update_bp_intercept = update_exception_bitmap,
7878 	.get_msr_feature = vmx_get_msr_feature,
7879 	.get_msr = vmx_get_msr,
7880 	.set_msr = vmx_set_msr,
7881 	.get_segment_base = vmx_get_segment_base,
7882 	.get_segment = vmx_get_segment,
7883 	.set_segment = vmx_set_segment,
7884 	.get_cpl = vmx_get_cpl,
7885 	.get_cs_db_l_bits = vmx_get_cs_db_l_bits,
7886 	.decache_cr0_guest_bits = vmx_decache_cr0_guest_bits,
7887 	.decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
7888 	.set_cr0 = vmx_set_cr0,
7889 	.set_cr3 = vmx_set_cr3,
7890 	.set_cr4 = vmx_set_cr4,
7891 	.set_efer = vmx_set_efer,
7892 	.get_idt = vmx_get_idt,
7893 	.set_idt = vmx_set_idt,
7894 	.get_gdt = vmx_get_gdt,
7895 	.set_gdt = vmx_set_gdt,
7896 	.get_dr6 = vmx_get_dr6,
7897 	.set_dr6 = vmx_set_dr6,
7898 	.set_dr7 = vmx_set_dr7,
7899 	.sync_dirty_debug_regs = vmx_sync_dirty_debug_regs,
7900 	.cache_reg = vmx_cache_reg,
7901 	.get_rflags = vmx_get_rflags,
7902 	.set_rflags = vmx_set_rflags,
7903 
7904 	.tlb_flush = vmx_flush_tlb,
7905 	.tlb_flush_gva = vmx_flush_tlb_gva,
7906 
7907 	.run = vmx_vcpu_run,
7908 	.handle_exit = vmx_handle_exit,
7909 	.skip_emulated_instruction = vmx_skip_emulated_instruction,
7910 	.update_emulated_instruction = vmx_update_emulated_instruction,
7911 	.set_interrupt_shadow = vmx_set_interrupt_shadow,
7912 	.get_interrupt_shadow = vmx_get_interrupt_shadow,
7913 	.patch_hypercall = vmx_patch_hypercall,
7914 	.set_irq = vmx_inject_irq,
7915 	.set_nmi = vmx_inject_nmi,
7916 	.queue_exception = vmx_queue_exception,
7917 	.cancel_injection = vmx_cancel_injection,
7918 	.interrupt_allowed = vmx_interrupt_allowed,
7919 	.nmi_allowed = vmx_nmi_allowed,
7920 	.get_nmi_mask = vmx_get_nmi_mask,
7921 	.set_nmi_mask = vmx_set_nmi_mask,
7922 	.enable_nmi_window = enable_nmi_window,
7923 	.enable_irq_window = enable_irq_window,
7924 	.update_cr8_intercept = update_cr8_intercept,
7925 	.set_virtual_apic_mode = vmx_set_virtual_apic_mode,
7926 	.set_apic_access_page_addr = vmx_set_apic_access_page_addr,
7927 	.refresh_apicv_exec_ctrl = vmx_refresh_apicv_exec_ctrl,
7928 	.load_eoi_exitmap = vmx_load_eoi_exitmap,
7929 	.apicv_post_state_restore = vmx_apicv_post_state_restore,
7930 	.check_apicv_inhibit_reasons = vmx_check_apicv_inhibit_reasons,
7931 	.hwapic_irr_update = vmx_hwapic_irr_update,
7932 	.hwapic_isr_update = vmx_hwapic_isr_update,
7933 	.guest_apic_has_interrupt = vmx_guest_apic_has_interrupt,
7934 	.sync_pir_to_irr = vmx_sync_pir_to_irr,
7935 	.deliver_posted_interrupt = vmx_deliver_posted_interrupt,
7936 	.dy_apicv_has_pending_interrupt = vmx_dy_apicv_has_pending_interrupt,
7937 
7938 	.set_tss_addr = vmx_set_tss_addr,
7939 	.set_identity_map_addr = vmx_set_identity_map_addr,
7940 	.get_tdp_level = get_ept_level,
7941 	.get_mt_mask = vmx_get_mt_mask,
7942 
7943 	.get_exit_info = vmx_get_exit_info,
7944 
7945 	.get_lpage_level = vmx_get_lpage_level,
7946 
7947 	.cpuid_update = vmx_cpuid_update,
7948 
7949 	.rdtscp_supported = vmx_rdtscp_supported,
7950 	.invpcid_supported = vmx_invpcid_supported,
7951 
7952 	.set_supported_cpuid = vmx_set_supported_cpuid,
7953 
7954 	.has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
7955 
7956 	.read_l1_tsc_offset = vmx_read_l1_tsc_offset,
7957 	.write_l1_tsc_offset = vmx_write_l1_tsc_offset,
7958 
7959 	.set_tdp_cr3 = vmx_set_cr3,
7960 
7961 	.check_intercept = vmx_check_intercept,
7962 	.handle_exit_irqoff = vmx_handle_exit_irqoff,
7963 	.mpx_supported = vmx_mpx_supported,
7964 	.xsaves_supported = vmx_xsaves_supported,
7965 	.umip_emulated = vmx_umip_emulated,
7966 	.pt_supported = vmx_pt_supported,
7967 	.pku_supported = vmx_pku_supported,
7968 
7969 	.request_immediate_exit = vmx_request_immediate_exit,
7970 
7971 	.sched_in = vmx_sched_in,
7972 
7973 	.slot_enable_log_dirty = vmx_slot_enable_log_dirty,
7974 	.slot_disable_log_dirty = vmx_slot_disable_log_dirty,
7975 	.flush_log_dirty = vmx_flush_log_dirty,
7976 	.enable_log_dirty_pt_masked = vmx_enable_log_dirty_pt_masked,
7977 	.write_log_dirty = vmx_write_pml_buffer,
7978 
7979 	.pre_block = vmx_pre_block,
7980 	.post_block = vmx_post_block,
7981 
7982 	.pmu_ops = &intel_pmu_ops,
7983 
7984 	.update_pi_irte = vmx_update_pi_irte,
7985 
7986 #ifdef CONFIG_X86_64
7987 	.set_hv_timer = vmx_set_hv_timer,
7988 	.cancel_hv_timer = vmx_cancel_hv_timer,
7989 #endif
7990 
7991 	.setup_mce = vmx_setup_mce,
7992 
7993 	.smi_allowed = vmx_smi_allowed,
7994 	.pre_enter_smm = vmx_pre_enter_smm,
7995 	.pre_leave_smm = vmx_pre_leave_smm,
7996 	.enable_smi_window = enable_smi_window,
7997 
7998 	.check_nested_events = NULL,
7999 	.get_nested_state = NULL,
8000 	.set_nested_state = NULL,
8001 	.get_vmcs12_pages = NULL,
8002 	.nested_enable_evmcs = NULL,
8003 	.nested_get_evmcs_version = NULL,
8004 	.need_emulation_on_page_fault = vmx_need_emulation_on_page_fault,
8005 	.apic_init_signal_blocked = vmx_apic_init_signal_blocked,
8006 };
8007 
8008 static void vmx_cleanup_l1d_flush(void)
8009 {
8010 	if (vmx_l1d_flush_pages) {
8011 		free_pages((unsigned long)vmx_l1d_flush_pages, L1D_CACHE_ORDER);
8012 		vmx_l1d_flush_pages = NULL;
8013 	}
8014 	/* Restore state so sysfs ignores VMX */
8015 	l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_AUTO;
8016 }
8017 
8018 static void vmx_exit(void)
8019 {
8020 #ifdef CONFIG_KEXEC_CORE
8021 	RCU_INIT_POINTER(crash_vmclear_loaded_vmcss, NULL);
8022 	synchronize_rcu();
8023 #endif
8024 
8025 	kvm_exit();
8026 
8027 #if IS_ENABLED(CONFIG_HYPERV)
8028 	if (static_branch_unlikely(&enable_evmcs)) {
8029 		int cpu;
8030 		struct hv_vp_assist_page *vp_ap;
8031 		/*
8032 		 * Reset everything to support using non-enlightened VMCS
8033 		 * access later (e.g. when we reload the module with
8034 		 * enlightened_vmcs=0)
8035 		 */
8036 		for_each_online_cpu(cpu) {
8037 			vp_ap =	hv_get_vp_assist_page(cpu);
8038 
8039 			if (!vp_ap)
8040 				continue;
8041 
8042 			vp_ap->nested_control.features.directhypercall = 0;
8043 			vp_ap->current_nested_vmcs = 0;
8044 			vp_ap->enlighten_vmentry = 0;
8045 		}
8046 
8047 		static_branch_disable(&enable_evmcs);
8048 	}
8049 #endif
8050 	vmx_cleanup_l1d_flush();
8051 }
8052 module_exit(vmx_exit);
8053 
8054 static int __init vmx_init(void)
8055 {
8056 	int r;
8057 
8058 #if IS_ENABLED(CONFIG_HYPERV)
8059 	/*
8060 	 * Enlightened VMCS usage should be recommended and the host needs
8061 	 * to support eVMCS v1 or above. We can also disable eVMCS support
8062 	 * with module parameter.
8063 	 */
8064 	if (enlightened_vmcs &&
8065 	    ms_hyperv.hints & HV_X64_ENLIGHTENED_VMCS_RECOMMENDED &&
8066 	    (ms_hyperv.nested_features & HV_X64_ENLIGHTENED_VMCS_VERSION) >=
8067 	    KVM_EVMCS_VERSION) {
8068 		int cpu;
8069 
8070 		/* Check that we have assist pages on all online CPUs */
8071 		for_each_online_cpu(cpu) {
8072 			if (!hv_get_vp_assist_page(cpu)) {
8073 				enlightened_vmcs = false;
8074 				break;
8075 			}
8076 		}
8077 
8078 		if (enlightened_vmcs) {
8079 			pr_info("KVM: vmx: using Hyper-V Enlightened VMCS\n");
8080 			static_branch_enable(&enable_evmcs);
8081 		}
8082 
8083 		if (ms_hyperv.nested_features & HV_X64_NESTED_DIRECT_FLUSH)
8084 			vmx_x86_ops.enable_direct_tlbflush
8085 				= hv_enable_direct_tlbflush;
8086 
8087 	} else {
8088 		enlightened_vmcs = false;
8089 	}
8090 #endif
8091 
8092 	r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx),
8093 		     __alignof__(struct vcpu_vmx), THIS_MODULE);
8094 	if (r)
8095 		return r;
8096 
8097 	/*
8098 	 * Must be called after kvm_init() so enable_ept is properly set
8099 	 * up. Hand the parameter mitigation value in which was stored in
8100 	 * the pre module init parser. If no parameter was given, it will
8101 	 * contain 'auto' which will be turned into the default 'cond'
8102 	 * mitigation mode.
8103 	 */
8104 	r = vmx_setup_l1d_flush(vmentry_l1d_flush_param);
8105 	if (r) {
8106 		vmx_exit();
8107 		return r;
8108 	}
8109 
8110 #ifdef CONFIG_KEXEC_CORE
8111 	rcu_assign_pointer(crash_vmclear_loaded_vmcss,
8112 			   crash_vmclear_local_loaded_vmcss);
8113 #endif
8114 	vmx_check_vmcs12_offsets();
8115 
8116 	return 0;
8117 }
8118 module_init(vmx_init);
8119