xref: /linux/arch/x86/kvm/vmx/vmx.c (revision 06bd48b6cd97ef3889b68c8e09014d81dbc463f1)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Kernel-based Virtual Machine driver for Linux
4  *
5  * This module enables machines with Intel VT-x extensions to run virtual
6  * machines without emulation or binary translation.
7  *
8  * Copyright (C) 2006 Qumranet, Inc.
9  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
10  *
11  * Authors:
12  *   Avi Kivity   <avi@qumranet.com>
13  *   Yaniv Kamay  <yaniv@qumranet.com>
14  */
15 
16 #include <linux/frame.h>
17 #include <linux/highmem.h>
18 #include <linux/hrtimer.h>
19 #include <linux/kernel.h>
20 #include <linux/kvm_host.h>
21 #include <linux/module.h>
22 #include <linux/moduleparam.h>
23 #include <linux/mod_devicetable.h>
24 #include <linux/mm.h>
25 #include <linux/sched.h>
26 #include <linux/sched/smt.h>
27 #include <linux/slab.h>
28 #include <linux/tboot.h>
29 #include <linux/trace_events.h>
30 
31 #include <asm/apic.h>
32 #include <asm/asm.h>
33 #include <asm/cpu.h>
34 #include <asm/cpu_device_id.h>
35 #include <asm/debugreg.h>
36 #include <asm/desc.h>
37 #include <asm/fpu/internal.h>
38 #include <asm/io.h>
39 #include <asm/irq_remapping.h>
40 #include <asm/kexec.h>
41 #include <asm/perf_event.h>
42 #include <asm/mce.h>
43 #include <asm/mmu_context.h>
44 #include <asm/mshyperv.h>
45 #include <asm/mwait.h>
46 #include <asm/spec-ctrl.h>
47 #include <asm/virtext.h>
48 #include <asm/vmx.h>
49 
50 #include "capabilities.h"
51 #include "cpuid.h"
52 #include "evmcs.h"
53 #include "irq.h"
54 #include "kvm_cache_regs.h"
55 #include "lapic.h"
56 #include "mmu.h"
57 #include "nested.h"
58 #include "ops.h"
59 #include "pmu.h"
60 #include "trace.h"
61 #include "vmcs.h"
62 #include "vmcs12.h"
63 #include "vmx.h"
64 #include "x86.h"
65 
66 MODULE_AUTHOR("Qumranet");
67 MODULE_LICENSE("GPL");
68 
69 #ifdef MODULE
70 static const struct x86_cpu_id vmx_cpu_id[] = {
71 	X86_MATCH_FEATURE(X86_FEATURE_VMX, NULL),
72 	{}
73 };
74 MODULE_DEVICE_TABLE(x86cpu, vmx_cpu_id);
75 #endif
76 
77 bool __read_mostly enable_vpid = 1;
78 module_param_named(vpid, enable_vpid, bool, 0444);
79 
80 static bool __read_mostly enable_vnmi = 1;
81 module_param_named(vnmi, enable_vnmi, bool, S_IRUGO);
82 
83 bool __read_mostly flexpriority_enabled = 1;
84 module_param_named(flexpriority, flexpriority_enabled, bool, S_IRUGO);
85 
86 bool __read_mostly enable_ept = 1;
87 module_param_named(ept, enable_ept, bool, S_IRUGO);
88 
89 bool __read_mostly enable_unrestricted_guest = 1;
90 module_param_named(unrestricted_guest,
91 			enable_unrestricted_guest, bool, S_IRUGO);
92 
93 bool __read_mostly enable_ept_ad_bits = 1;
94 module_param_named(eptad, enable_ept_ad_bits, bool, S_IRUGO);
95 
96 static bool __read_mostly emulate_invalid_guest_state = true;
97 module_param(emulate_invalid_guest_state, bool, S_IRUGO);
98 
99 static bool __read_mostly fasteoi = 1;
100 module_param(fasteoi, bool, S_IRUGO);
101 
102 bool __read_mostly enable_apicv = 1;
103 module_param(enable_apicv, bool, S_IRUGO);
104 
105 /*
106  * If nested=1, nested virtualization is supported, i.e., guests may use
107  * VMX and be a hypervisor for its own guests. If nested=0, guests may not
108  * use VMX instructions.
109  */
110 static bool __read_mostly nested = 1;
111 module_param(nested, bool, S_IRUGO);
112 
113 bool __read_mostly enable_pml = 1;
114 module_param_named(pml, enable_pml, bool, S_IRUGO);
115 
116 static bool __read_mostly dump_invalid_vmcs = 0;
117 module_param(dump_invalid_vmcs, bool, 0644);
118 
119 #define MSR_BITMAP_MODE_X2APIC		1
120 #define MSR_BITMAP_MODE_X2APIC_APICV	2
121 
122 #define KVM_VMX_TSC_MULTIPLIER_MAX     0xffffffffffffffffULL
123 
124 /* Guest_tsc -> host_tsc conversion requires 64-bit division.  */
125 static int __read_mostly cpu_preemption_timer_multi;
126 static bool __read_mostly enable_preemption_timer = 1;
127 #ifdef CONFIG_X86_64
128 module_param_named(preemption_timer, enable_preemption_timer, bool, S_IRUGO);
129 #endif
130 
131 #define KVM_VM_CR0_ALWAYS_OFF (X86_CR0_NW | X86_CR0_CD)
132 #define KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST X86_CR0_NE
133 #define KVM_VM_CR0_ALWAYS_ON				\
134 	(KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST | 	\
135 	 X86_CR0_WP | X86_CR0_PG | X86_CR0_PE)
136 #define KVM_CR4_GUEST_OWNED_BITS				      \
137 	(X86_CR4_PVI | X86_CR4_DE | X86_CR4_PCE | X86_CR4_OSFXSR      \
138 	 | X86_CR4_OSXMMEXCPT | X86_CR4_LA57 | X86_CR4_TSD)
139 
140 #define KVM_VM_CR4_ALWAYS_ON_UNRESTRICTED_GUEST X86_CR4_VMXE
141 #define KVM_PMODE_VM_CR4_ALWAYS_ON (X86_CR4_PAE | X86_CR4_VMXE)
142 #define KVM_RMODE_VM_CR4_ALWAYS_ON (X86_CR4_VME | X86_CR4_PAE | X86_CR4_VMXE)
143 
144 #define RMODE_GUEST_OWNED_EFLAGS_BITS (~(X86_EFLAGS_IOPL | X86_EFLAGS_VM))
145 
146 #define MSR_IA32_RTIT_STATUS_MASK (~(RTIT_STATUS_FILTEREN | \
147 	RTIT_STATUS_CONTEXTEN | RTIT_STATUS_TRIGGEREN | \
148 	RTIT_STATUS_ERROR | RTIT_STATUS_STOPPED | \
149 	RTIT_STATUS_BYTECNT))
150 
151 #define MSR_IA32_RTIT_OUTPUT_BASE_MASK \
152 	(~((1UL << cpuid_query_maxphyaddr(vcpu)) - 1) | 0x7f)
153 
154 /*
155  * These 2 parameters are used to config the controls for Pause-Loop Exiting:
156  * ple_gap:    upper bound on the amount of time between two successive
157  *             executions of PAUSE in a loop. Also indicate if ple enabled.
158  *             According to test, this time is usually smaller than 128 cycles.
159  * ple_window: upper bound on the amount of time a guest is allowed to execute
160  *             in a PAUSE loop. Tests indicate that most spinlocks are held for
161  *             less than 2^12 cycles
162  * Time is measured based on a counter that runs at the same rate as the TSC,
163  * refer SDM volume 3b section 21.6.13 & 22.1.3.
164  */
165 static unsigned int ple_gap = KVM_DEFAULT_PLE_GAP;
166 module_param(ple_gap, uint, 0444);
167 
168 static unsigned int ple_window = KVM_VMX_DEFAULT_PLE_WINDOW;
169 module_param(ple_window, uint, 0444);
170 
171 /* Default doubles per-vcpu window every exit. */
172 static unsigned int ple_window_grow = KVM_DEFAULT_PLE_WINDOW_GROW;
173 module_param(ple_window_grow, uint, 0444);
174 
175 /* Default resets per-vcpu window every exit to ple_window. */
176 static unsigned int ple_window_shrink = KVM_DEFAULT_PLE_WINDOW_SHRINK;
177 module_param(ple_window_shrink, uint, 0444);
178 
179 /* Default is to compute the maximum so we can never overflow. */
180 static unsigned int ple_window_max        = KVM_VMX_DEFAULT_PLE_WINDOW_MAX;
181 module_param(ple_window_max, uint, 0444);
182 
183 /* Default is SYSTEM mode, 1 for host-guest mode */
184 int __read_mostly pt_mode = PT_MODE_SYSTEM;
185 module_param(pt_mode, int, S_IRUGO);
186 
187 static DEFINE_STATIC_KEY_FALSE(vmx_l1d_should_flush);
188 static DEFINE_STATIC_KEY_FALSE(vmx_l1d_flush_cond);
189 static DEFINE_MUTEX(vmx_l1d_flush_mutex);
190 
191 /* Storage for pre module init parameter parsing */
192 static enum vmx_l1d_flush_state __read_mostly vmentry_l1d_flush_param = VMENTER_L1D_FLUSH_AUTO;
193 
194 static const struct {
195 	const char *option;
196 	bool for_parse;
197 } vmentry_l1d_param[] = {
198 	[VMENTER_L1D_FLUSH_AUTO]	 = {"auto", true},
199 	[VMENTER_L1D_FLUSH_NEVER]	 = {"never", true},
200 	[VMENTER_L1D_FLUSH_COND]	 = {"cond", true},
201 	[VMENTER_L1D_FLUSH_ALWAYS]	 = {"always", true},
202 	[VMENTER_L1D_FLUSH_EPT_DISABLED] = {"EPT disabled", false},
203 	[VMENTER_L1D_FLUSH_NOT_REQUIRED] = {"not required", false},
204 };
205 
206 #define L1D_CACHE_ORDER 4
207 static void *vmx_l1d_flush_pages;
208 
209 static int vmx_setup_l1d_flush(enum vmx_l1d_flush_state l1tf)
210 {
211 	struct page *page;
212 	unsigned int i;
213 
214 	if (!boot_cpu_has_bug(X86_BUG_L1TF)) {
215 		l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_NOT_REQUIRED;
216 		return 0;
217 	}
218 
219 	if (!enable_ept) {
220 		l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_EPT_DISABLED;
221 		return 0;
222 	}
223 
224 	if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES)) {
225 		u64 msr;
226 
227 		rdmsrl(MSR_IA32_ARCH_CAPABILITIES, msr);
228 		if (msr & ARCH_CAP_SKIP_VMENTRY_L1DFLUSH) {
229 			l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_NOT_REQUIRED;
230 			return 0;
231 		}
232 	}
233 
234 	/* If set to auto use the default l1tf mitigation method */
235 	if (l1tf == VMENTER_L1D_FLUSH_AUTO) {
236 		switch (l1tf_mitigation) {
237 		case L1TF_MITIGATION_OFF:
238 			l1tf = VMENTER_L1D_FLUSH_NEVER;
239 			break;
240 		case L1TF_MITIGATION_FLUSH_NOWARN:
241 		case L1TF_MITIGATION_FLUSH:
242 		case L1TF_MITIGATION_FLUSH_NOSMT:
243 			l1tf = VMENTER_L1D_FLUSH_COND;
244 			break;
245 		case L1TF_MITIGATION_FULL:
246 		case L1TF_MITIGATION_FULL_FORCE:
247 			l1tf = VMENTER_L1D_FLUSH_ALWAYS;
248 			break;
249 		}
250 	} else if (l1tf_mitigation == L1TF_MITIGATION_FULL_FORCE) {
251 		l1tf = VMENTER_L1D_FLUSH_ALWAYS;
252 	}
253 
254 	if (l1tf != VMENTER_L1D_FLUSH_NEVER && !vmx_l1d_flush_pages &&
255 	    !boot_cpu_has(X86_FEATURE_FLUSH_L1D)) {
256 		/*
257 		 * This allocation for vmx_l1d_flush_pages is not tied to a VM
258 		 * lifetime and so should not be charged to a memcg.
259 		 */
260 		page = alloc_pages(GFP_KERNEL, L1D_CACHE_ORDER);
261 		if (!page)
262 			return -ENOMEM;
263 		vmx_l1d_flush_pages = page_address(page);
264 
265 		/*
266 		 * Initialize each page with a different pattern in
267 		 * order to protect against KSM in the nested
268 		 * virtualization case.
269 		 */
270 		for (i = 0; i < 1u << L1D_CACHE_ORDER; ++i) {
271 			memset(vmx_l1d_flush_pages + i * PAGE_SIZE, i + 1,
272 			       PAGE_SIZE);
273 		}
274 	}
275 
276 	l1tf_vmx_mitigation = l1tf;
277 
278 	if (l1tf != VMENTER_L1D_FLUSH_NEVER)
279 		static_branch_enable(&vmx_l1d_should_flush);
280 	else
281 		static_branch_disable(&vmx_l1d_should_flush);
282 
283 	if (l1tf == VMENTER_L1D_FLUSH_COND)
284 		static_branch_enable(&vmx_l1d_flush_cond);
285 	else
286 		static_branch_disable(&vmx_l1d_flush_cond);
287 	return 0;
288 }
289 
290 static int vmentry_l1d_flush_parse(const char *s)
291 {
292 	unsigned int i;
293 
294 	if (s) {
295 		for (i = 0; i < ARRAY_SIZE(vmentry_l1d_param); i++) {
296 			if (vmentry_l1d_param[i].for_parse &&
297 			    sysfs_streq(s, vmentry_l1d_param[i].option))
298 				return i;
299 		}
300 	}
301 	return -EINVAL;
302 }
303 
304 static int vmentry_l1d_flush_set(const char *s, const struct kernel_param *kp)
305 {
306 	int l1tf, ret;
307 
308 	l1tf = vmentry_l1d_flush_parse(s);
309 	if (l1tf < 0)
310 		return l1tf;
311 
312 	if (!boot_cpu_has(X86_BUG_L1TF))
313 		return 0;
314 
315 	/*
316 	 * Has vmx_init() run already? If not then this is the pre init
317 	 * parameter parsing. In that case just store the value and let
318 	 * vmx_init() do the proper setup after enable_ept has been
319 	 * established.
320 	 */
321 	if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_AUTO) {
322 		vmentry_l1d_flush_param = l1tf;
323 		return 0;
324 	}
325 
326 	mutex_lock(&vmx_l1d_flush_mutex);
327 	ret = vmx_setup_l1d_flush(l1tf);
328 	mutex_unlock(&vmx_l1d_flush_mutex);
329 	return ret;
330 }
331 
332 static int vmentry_l1d_flush_get(char *s, const struct kernel_param *kp)
333 {
334 	if (WARN_ON_ONCE(l1tf_vmx_mitigation >= ARRAY_SIZE(vmentry_l1d_param)))
335 		return sprintf(s, "???\n");
336 
337 	return sprintf(s, "%s\n", vmentry_l1d_param[l1tf_vmx_mitigation].option);
338 }
339 
340 static const struct kernel_param_ops vmentry_l1d_flush_ops = {
341 	.set = vmentry_l1d_flush_set,
342 	.get = vmentry_l1d_flush_get,
343 };
344 module_param_cb(vmentry_l1d_flush, &vmentry_l1d_flush_ops, NULL, 0644);
345 
346 static bool guest_state_valid(struct kvm_vcpu *vcpu);
347 static u32 vmx_segment_access_rights(struct kvm_segment *var);
348 static __always_inline void vmx_disable_intercept_for_msr(unsigned long *msr_bitmap,
349 							  u32 msr, int type);
350 
351 void vmx_vmexit(void);
352 
353 #define vmx_insn_failed(fmt...)		\
354 do {					\
355 	WARN_ONCE(1, fmt);		\
356 	pr_warn_ratelimited(fmt);	\
357 } while (0)
358 
359 asmlinkage void vmread_error(unsigned long field, bool fault)
360 {
361 	if (fault)
362 		kvm_spurious_fault();
363 	else
364 		vmx_insn_failed("kvm: vmread failed: field=%lx\n", field);
365 }
366 
367 noinline void vmwrite_error(unsigned long field, unsigned long value)
368 {
369 	vmx_insn_failed("kvm: vmwrite failed: field=%lx val=%lx err=%d\n",
370 			field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
371 }
372 
373 noinline void vmclear_error(struct vmcs *vmcs, u64 phys_addr)
374 {
375 	vmx_insn_failed("kvm: vmclear failed: %p/%llx\n", vmcs, phys_addr);
376 }
377 
378 noinline void vmptrld_error(struct vmcs *vmcs, u64 phys_addr)
379 {
380 	vmx_insn_failed("kvm: vmptrld failed: %p/%llx\n", vmcs, phys_addr);
381 }
382 
383 noinline void invvpid_error(unsigned long ext, u16 vpid, gva_t gva)
384 {
385 	vmx_insn_failed("kvm: invvpid failed: ext=0x%lx vpid=%u gva=0x%lx\n",
386 			ext, vpid, gva);
387 }
388 
389 noinline void invept_error(unsigned long ext, u64 eptp, gpa_t gpa)
390 {
391 	vmx_insn_failed("kvm: invept failed: ext=0x%lx eptp=%llx gpa=0x%llx\n",
392 			ext, eptp, gpa);
393 }
394 
395 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
396 DEFINE_PER_CPU(struct vmcs *, current_vmcs);
397 /*
398  * We maintain a per-CPU linked-list of VMCS loaded on that CPU. This is needed
399  * when a CPU is brought down, and we need to VMCLEAR all VMCSs loaded on it.
400  */
401 static DEFINE_PER_CPU(struct list_head, loaded_vmcss_on_cpu);
402 
403 /*
404  * We maintian a per-CPU linked-list of vCPU, so in wakeup_handler() we
405  * can find which vCPU should be waken up.
406  */
407 static DEFINE_PER_CPU(struct list_head, blocked_vcpu_on_cpu);
408 static DEFINE_PER_CPU(spinlock_t, blocked_vcpu_on_cpu_lock);
409 
410 static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
411 static DEFINE_SPINLOCK(vmx_vpid_lock);
412 
413 struct vmcs_config vmcs_config;
414 struct vmx_capability vmx_capability;
415 
416 #define VMX_SEGMENT_FIELD(seg)					\
417 	[VCPU_SREG_##seg] = {                                   \
418 		.selector = GUEST_##seg##_SELECTOR,		\
419 		.base = GUEST_##seg##_BASE,		   	\
420 		.limit = GUEST_##seg##_LIMIT,		   	\
421 		.ar_bytes = GUEST_##seg##_AR_BYTES,	   	\
422 	}
423 
424 static const struct kvm_vmx_segment_field {
425 	unsigned selector;
426 	unsigned base;
427 	unsigned limit;
428 	unsigned ar_bytes;
429 } kvm_vmx_segment_fields[] = {
430 	VMX_SEGMENT_FIELD(CS),
431 	VMX_SEGMENT_FIELD(DS),
432 	VMX_SEGMENT_FIELD(ES),
433 	VMX_SEGMENT_FIELD(FS),
434 	VMX_SEGMENT_FIELD(GS),
435 	VMX_SEGMENT_FIELD(SS),
436 	VMX_SEGMENT_FIELD(TR),
437 	VMX_SEGMENT_FIELD(LDTR),
438 };
439 
440 static unsigned long host_idt_base;
441 
442 /*
443  * Though SYSCALL is only supported in 64-bit mode on Intel CPUs, kvm
444  * will emulate SYSCALL in legacy mode if the vendor string in guest
445  * CPUID.0:{EBX,ECX,EDX} is "AuthenticAMD" or "AMDisbetter!" To
446  * support this emulation, IA32_STAR must always be included in
447  * vmx_msr_index[], even in i386 builds.
448  */
449 const u32 vmx_msr_index[] = {
450 #ifdef CONFIG_X86_64
451 	MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
452 #endif
453 	MSR_EFER, MSR_TSC_AUX, MSR_STAR,
454 	MSR_IA32_TSX_CTRL,
455 };
456 
457 #if IS_ENABLED(CONFIG_HYPERV)
458 static bool __read_mostly enlightened_vmcs = true;
459 module_param(enlightened_vmcs, bool, 0444);
460 
461 /* check_ept_pointer() should be under protection of ept_pointer_lock. */
462 static void check_ept_pointer_match(struct kvm *kvm)
463 {
464 	struct kvm_vcpu *vcpu;
465 	u64 tmp_eptp = INVALID_PAGE;
466 	int i;
467 
468 	kvm_for_each_vcpu(i, vcpu, kvm) {
469 		if (!VALID_PAGE(tmp_eptp)) {
470 			tmp_eptp = to_vmx(vcpu)->ept_pointer;
471 		} else if (tmp_eptp != to_vmx(vcpu)->ept_pointer) {
472 			to_kvm_vmx(kvm)->ept_pointers_match
473 				= EPT_POINTERS_MISMATCH;
474 			return;
475 		}
476 	}
477 
478 	to_kvm_vmx(kvm)->ept_pointers_match = EPT_POINTERS_MATCH;
479 }
480 
481 static int kvm_fill_hv_flush_list_func(struct hv_guest_mapping_flush_list *flush,
482 		void *data)
483 {
484 	struct kvm_tlb_range *range = data;
485 
486 	return hyperv_fill_flush_guest_mapping_list(flush, range->start_gfn,
487 			range->pages);
488 }
489 
490 static inline int __hv_remote_flush_tlb_with_range(struct kvm *kvm,
491 		struct kvm_vcpu *vcpu, struct kvm_tlb_range *range)
492 {
493 	u64 ept_pointer = to_vmx(vcpu)->ept_pointer;
494 
495 	/*
496 	 * FLUSH_GUEST_PHYSICAL_ADDRESS_SPACE hypercall needs address
497 	 * of the base of EPT PML4 table, strip off EPT configuration
498 	 * information.
499 	 */
500 	if (range)
501 		return hyperv_flush_guest_mapping_range(ept_pointer & PAGE_MASK,
502 				kvm_fill_hv_flush_list_func, (void *)range);
503 	else
504 		return hyperv_flush_guest_mapping(ept_pointer & PAGE_MASK);
505 }
506 
507 static int hv_remote_flush_tlb_with_range(struct kvm *kvm,
508 		struct kvm_tlb_range *range)
509 {
510 	struct kvm_vcpu *vcpu;
511 	int ret = 0, i;
512 
513 	spin_lock(&to_kvm_vmx(kvm)->ept_pointer_lock);
514 
515 	if (to_kvm_vmx(kvm)->ept_pointers_match == EPT_POINTERS_CHECK)
516 		check_ept_pointer_match(kvm);
517 
518 	if (to_kvm_vmx(kvm)->ept_pointers_match != EPT_POINTERS_MATCH) {
519 		kvm_for_each_vcpu(i, vcpu, kvm) {
520 			/* If ept_pointer is invalid pointer, bypass flush request. */
521 			if (VALID_PAGE(to_vmx(vcpu)->ept_pointer))
522 				ret |= __hv_remote_flush_tlb_with_range(
523 					kvm, vcpu, range);
524 		}
525 	} else {
526 		ret = __hv_remote_flush_tlb_with_range(kvm,
527 				kvm_get_vcpu(kvm, 0), range);
528 	}
529 
530 	spin_unlock(&to_kvm_vmx(kvm)->ept_pointer_lock);
531 	return ret;
532 }
533 static int hv_remote_flush_tlb(struct kvm *kvm)
534 {
535 	return hv_remote_flush_tlb_with_range(kvm, NULL);
536 }
537 
538 static int hv_enable_direct_tlbflush(struct kvm_vcpu *vcpu)
539 {
540 	struct hv_enlightened_vmcs *evmcs;
541 	struct hv_partition_assist_pg **p_hv_pa_pg =
542 			&vcpu->kvm->arch.hyperv.hv_pa_pg;
543 	/*
544 	 * Synthetic VM-Exit is not enabled in current code and so All
545 	 * evmcs in singe VM shares same assist page.
546 	 */
547 	if (!*p_hv_pa_pg)
548 		*p_hv_pa_pg = kzalloc(PAGE_SIZE, GFP_KERNEL);
549 
550 	if (!*p_hv_pa_pg)
551 		return -ENOMEM;
552 
553 	evmcs = (struct hv_enlightened_vmcs *)to_vmx(vcpu)->loaded_vmcs->vmcs;
554 
555 	evmcs->partition_assist_page =
556 		__pa(*p_hv_pa_pg);
557 	evmcs->hv_vm_id = (unsigned long)vcpu->kvm;
558 	evmcs->hv_enlightenments_control.nested_flush_hypercall = 1;
559 
560 	return 0;
561 }
562 
563 #endif /* IS_ENABLED(CONFIG_HYPERV) */
564 
565 /*
566  * Comment's format: document - errata name - stepping - processor name.
567  * Refer from
568  * https://www.virtualbox.org/svn/vbox/trunk/src/VBox/VMM/VMMR0/HMR0.cpp
569  */
570 static u32 vmx_preemption_cpu_tfms[] = {
571 /* 323344.pdf - BA86   - D0 - Xeon 7500 Series */
572 0x000206E6,
573 /* 323056.pdf - AAX65  - C2 - Xeon L3406 */
574 /* 322814.pdf - AAT59  - C2 - i7-600, i5-500, i5-400 and i3-300 Mobile */
575 /* 322911.pdf - AAU65  - C2 - i5-600, i3-500 Desktop and Pentium G6950 */
576 0x00020652,
577 /* 322911.pdf - AAU65  - K0 - i5-600, i3-500 Desktop and Pentium G6950 */
578 0x00020655,
579 /* 322373.pdf - AAO95  - B1 - Xeon 3400 Series */
580 /* 322166.pdf - AAN92  - B1 - i7-800 and i5-700 Desktop */
581 /*
582  * 320767.pdf - AAP86  - B1 -
583  * i7-900 Mobile Extreme, i7-800 and i7-700 Mobile
584  */
585 0x000106E5,
586 /* 321333.pdf - AAM126 - C0 - Xeon 3500 */
587 0x000106A0,
588 /* 321333.pdf - AAM126 - C1 - Xeon 3500 */
589 0x000106A1,
590 /* 320836.pdf - AAJ124 - C0 - i7-900 Desktop Extreme and i7-900 Desktop */
591 0x000106A4,
592  /* 321333.pdf - AAM126 - D0 - Xeon 3500 */
593  /* 321324.pdf - AAK139 - D0 - Xeon 5500 */
594  /* 320836.pdf - AAJ124 - D0 - i7-900 Extreme and i7-900 Desktop */
595 0x000106A5,
596  /* Xeon E3-1220 V2 */
597 0x000306A8,
598 };
599 
600 static inline bool cpu_has_broken_vmx_preemption_timer(void)
601 {
602 	u32 eax = cpuid_eax(0x00000001), i;
603 
604 	/* Clear the reserved bits */
605 	eax &= ~(0x3U << 14 | 0xfU << 28);
606 	for (i = 0; i < ARRAY_SIZE(vmx_preemption_cpu_tfms); i++)
607 		if (eax == vmx_preemption_cpu_tfms[i])
608 			return true;
609 
610 	return false;
611 }
612 
613 static inline bool cpu_need_virtualize_apic_accesses(struct kvm_vcpu *vcpu)
614 {
615 	return flexpriority_enabled && lapic_in_kernel(vcpu);
616 }
617 
618 static inline bool report_flexpriority(void)
619 {
620 	return flexpriority_enabled;
621 }
622 
623 static inline int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
624 {
625 	int i;
626 
627 	for (i = 0; i < vmx->nmsrs; ++i)
628 		if (vmx_msr_index[vmx->guest_msrs[i].index] == msr)
629 			return i;
630 	return -1;
631 }
632 
633 struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
634 {
635 	int i;
636 
637 	i = __find_msr_index(vmx, msr);
638 	if (i >= 0)
639 		return &vmx->guest_msrs[i];
640 	return NULL;
641 }
642 
643 static int vmx_set_guest_msr(struct vcpu_vmx *vmx, struct shared_msr_entry *msr, u64 data)
644 {
645 	int ret = 0;
646 
647 	u64 old_msr_data = msr->data;
648 	msr->data = data;
649 	if (msr - vmx->guest_msrs < vmx->save_nmsrs) {
650 		preempt_disable();
651 		ret = kvm_set_shared_msr(msr->index, msr->data,
652 					 msr->mask);
653 		preempt_enable();
654 		if (ret)
655 			msr->data = old_msr_data;
656 	}
657 	return ret;
658 }
659 
660 #ifdef CONFIG_KEXEC_CORE
661 static void crash_vmclear_local_loaded_vmcss(void)
662 {
663 	int cpu = raw_smp_processor_id();
664 	struct loaded_vmcs *v;
665 
666 	list_for_each_entry(v, &per_cpu(loaded_vmcss_on_cpu, cpu),
667 			    loaded_vmcss_on_cpu_link)
668 		vmcs_clear(v->vmcs);
669 }
670 #endif /* CONFIG_KEXEC_CORE */
671 
672 static void __loaded_vmcs_clear(void *arg)
673 {
674 	struct loaded_vmcs *loaded_vmcs = arg;
675 	int cpu = raw_smp_processor_id();
676 
677 	if (loaded_vmcs->cpu != cpu)
678 		return; /* vcpu migration can race with cpu offline */
679 	if (per_cpu(current_vmcs, cpu) == loaded_vmcs->vmcs)
680 		per_cpu(current_vmcs, cpu) = NULL;
681 
682 	vmcs_clear(loaded_vmcs->vmcs);
683 	if (loaded_vmcs->shadow_vmcs && loaded_vmcs->launched)
684 		vmcs_clear(loaded_vmcs->shadow_vmcs);
685 
686 	list_del(&loaded_vmcs->loaded_vmcss_on_cpu_link);
687 
688 	/*
689 	 * Ensure all writes to loaded_vmcs, including deleting it from its
690 	 * current percpu list, complete before setting loaded_vmcs->vcpu to
691 	 * -1, otherwise a different cpu can see vcpu == -1 first and add
692 	 * loaded_vmcs to its percpu list before it's deleted from this cpu's
693 	 * list. Pairs with the smp_rmb() in vmx_vcpu_load_vmcs().
694 	 */
695 	smp_wmb();
696 
697 	loaded_vmcs->cpu = -1;
698 	loaded_vmcs->launched = 0;
699 }
700 
701 void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs)
702 {
703 	int cpu = loaded_vmcs->cpu;
704 
705 	if (cpu != -1)
706 		smp_call_function_single(cpu,
707 			 __loaded_vmcs_clear, loaded_vmcs, 1);
708 }
709 
710 static bool vmx_segment_cache_test_set(struct vcpu_vmx *vmx, unsigned seg,
711 				       unsigned field)
712 {
713 	bool ret;
714 	u32 mask = 1 << (seg * SEG_FIELD_NR + field);
715 
716 	if (!kvm_register_is_available(&vmx->vcpu, VCPU_EXREG_SEGMENTS)) {
717 		kvm_register_mark_available(&vmx->vcpu, VCPU_EXREG_SEGMENTS);
718 		vmx->segment_cache.bitmask = 0;
719 	}
720 	ret = vmx->segment_cache.bitmask & mask;
721 	vmx->segment_cache.bitmask |= mask;
722 	return ret;
723 }
724 
725 static u16 vmx_read_guest_seg_selector(struct vcpu_vmx *vmx, unsigned seg)
726 {
727 	u16 *p = &vmx->segment_cache.seg[seg].selector;
728 
729 	if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_SEL))
730 		*p = vmcs_read16(kvm_vmx_segment_fields[seg].selector);
731 	return *p;
732 }
733 
734 static ulong vmx_read_guest_seg_base(struct vcpu_vmx *vmx, unsigned seg)
735 {
736 	ulong *p = &vmx->segment_cache.seg[seg].base;
737 
738 	if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_BASE))
739 		*p = vmcs_readl(kvm_vmx_segment_fields[seg].base);
740 	return *p;
741 }
742 
743 static u32 vmx_read_guest_seg_limit(struct vcpu_vmx *vmx, unsigned seg)
744 {
745 	u32 *p = &vmx->segment_cache.seg[seg].limit;
746 
747 	if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_LIMIT))
748 		*p = vmcs_read32(kvm_vmx_segment_fields[seg].limit);
749 	return *p;
750 }
751 
752 static u32 vmx_read_guest_seg_ar(struct vcpu_vmx *vmx, unsigned seg)
753 {
754 	u32 *p = &vmx->segment_cache.seg[seg].ar;
755 
756 	if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_AR))
757 		*p = vmcs_read32(kvm_vmx_segment_fields[seg].ar_bytes);
758 	return *p;
759 }
760 
761 void update_exception_bitmap(struct kvm_vcpu *vcpu)
762 {
763 	u32 eb;
764 
765 	eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR) |
766 	     (1u << DB_VECTOR) | (1u << AC_VECTOR);
767 	/*
768 	 * Guest access to VMware backdoor ports could legitimately
769 	 * trigger #GP because of TSS I/O permission bitmap.
770 	 * We intercept those #GP and allow access to them anyway
771 	 * as VMware does.
772 	 */
773 	if (enable_vmware_backdoor)
774 		eb |= (1u << GP_VECTOR);
775 	if ((vcpu->guest_debug &
776 	     (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) ==
777 	    (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP))
778 		eb |= 1u << BP_VECTOR;
779 	if (to_vmx(vcpu)->rmode.vm86_active)
780 		eb = ~0;
781 	if (enable_ept)
782 		eb &= ~(1u << PF_VECTOR);
783 
784 	/* When we are running a nested L2 guest and L1 specified for it a
785 	 * certain exception bitmap, we must trap the same exceptions and pass
786 	 * them to L1. When running L2, we will only handle the exceptions
787 	 * specified above if L1 did not want them.
788 	 */
789 	if (is_guest_mode(vcpu))
790 		eb |= get_vmcs12(vcpu)->exception_bitmap;
791 
792 	vmcs_write32(EXCEPTION_BITMAP, eb);
793 }
794 
795 /*
796  * Check if MSR is intercepted for currently loaded MSR bitmap.
797  */
798 static bool msr_write_intercepted(struct kvm_vcpu *vcpu, u32 msr)
799 {
800 	unsigned long *msr_bitmap;
801 	int f = sizeof(unsigned long);
802 
803 	if (!cpu_has_vmx_msr_bitmap())
804 		return true;
805 
806 	msr_bitmap = to_vmx(vcpu)->loaded_vmcs->msr_bitmap;
807 
808 	if (msr <= 0x1fff) {
809 		return !!test_bit(msr, msr_bitmap + 0x800 / f);
810 	} else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
811 		msr &= 0x1fff;
812 		return !!test_bit(msr, msr_bitmap + 0xc00 / f);
813 	}
814 
815 	return true;
816 }
817 
818 static void clear_atomic_switch_msr_special(struct vcpu_vmx *vmx,
819 		unsigned long entry, unsigned long exit)
820 {
821 	vm_entry_controls_clearbit(vmx, entry);
822 	vm_exit_controls_clearbit(vmx, exit);
823 }
824 
825 int vmx_find_msr_index(struct vmx_msrs *m, u32 msr)
826 {
827 	unsigned int i;
828 
829 	for (i = 0; i < m->nr; ++i) {
830 		if (m->val[i].index == msr)
831 			return i;
832 	}
833 	return -ENOENT;
834 }
835 
836 static void clear_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr)
837 {
838 	int i;
839 	struct msr_autoload *m = &vmx->msr_autoload;
840 
841 	switch (msr) {
842 	case MSR_EFER:
843 		if (cpu_has_load_ia32_efer()) {
844 			clear_atomic_switch_msr_special(vmx,
845 					VM_ENTRY_LOAD_IA32_EFER,
846 					VM_EXIT_LOAD_IA32_EFER);
847 			return;
848 		}
849 		break;
850 	case MSR_CORE_PERF_GLOBAL_CTRL:
851 		if (cpu_has_load_perf_global_ctrl()) {
852 			clear_atomic_switch_msr_special(vmx,
853 					VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
854 					VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
855 			return;
856 		}
857 		break;
858 	}
859 	i = vmx_find_msr_index(&m->guest, msr);
860 	if (i < 0)
861 		goto skip_guest;
862 	--m->guest.nr;
863 	m->guest.val[i] = m->guest.val[m->guest.nr];
864 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->guest.nr);
865 
866 skip_guest:
867 	i = vmx_find_msr_index(&m->host, msr);
868 	if (i < 0)
869 		return;
870 
871 	--m->host.nr;
872 	m->host.val[i] = m->host.val[m->host.nr];
873 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->host.nr);
874 }
875 
876 static void add_atomic_switch_msr_special(struct vcpu_vmx *vmx,
877 		unsigned long entry, unsigned long exit,
878 		unsigned long guest_val_vmcs, unsigned long host_val_vmcs,
879 		u64 guest_val, u64 host_val)
880 {
881 	vmcs_write64(guest_val_vmcs, guest_val);
882 	if (host_val_vmcs != HOST_IA32_EFER)
883 		vmcs_write64(host_val_vmcs, host_val);
884 	vm_entry_controls_setbit(vmx, entry);
885 	vm_exit_controls_setbit(vmx, exit);
886 }
887 
888 static void add_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr,
889 				  u64 guest_val, u64 host_val, bool entry_only)
890 {
891 	int i, j = 0;
892 	struct msr_autoload *m = &vmx->msr_autoload;
893 
894 	switch (msr) {
895 	case MSR_EFER:
896 		if (cpu_has_load_ia32_efer()) {
897 			add_atomic_switch_msr_special(vmx,
898 					VM_ENTRY_LOAD_IA32_EFER,
899 					VM_EXIT_LOAD_IA32_EFER,
900 					GUEST_IA32_EFER,
901 					HOST_IA32_EFER,
902 					guest_val, host_val);
903 			return;
904 		}
905 		break;
906 	case MSR_CORE_PERF_GLOBAL_CTRL:
907 		if (cpu_has_load_perf_global_ctrl()) {
908 			add_atomic_switch_msr_special(vmx,
909 					VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
910 					VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL,
911 					GUEST_IA32_PERF_GLOBAL_CTRL,
912 					HOST_IA32_PERF_GLOBAL_CTRL,
913 					guest_val, host_val);
914 			return;
915 		}
916 		break;
917 	case MSR_IA32_PEBS_ENABLE:
918 		/* PEBS needs a quiescent period after being disabled (to write
919 		 * a record).  Disabling PEBS through VMX MSR swapping doesn't
920 		 * provide that period, so a CPU could write host's record into
921 		 * guest's memory.
922 		 */
923 		wrmsrl(MSR_IA32_PEBS_ENABLE, 0);
924 	}
925 
926 	i = vmx_find_msr_index(&m->guest, msr);
927 	if (!entry_only)
928 		j = vmx_find_msr_index(&m->host, msr);
929 
930 	if ((i < 0 && m->guest.nr == NR_LOADSTORE_MSRS) ||
931 		(j < 0 &&  m->host.nr == NR_LOADSTORE_MSRS)) {
932 		printk_once(KERN_WARNING "Not enough msr switch entries. "
933 				"Can't add msr %x\n", msr);
934 		return;
935 	}
936 	if (i < 0) {
937 		i = m->guest.nr++;
938 		vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->guest.nr);
939 	}
940 	m->guest.val[i].index = msr;
941 	m->guest.val[i].value = guest_val;
942 
943 	if (entry_only)
944 		return;
945 
946 	if (j < 0) {
947 		j = m->host.nr++;
948 		vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->host.nr);
949 	}
950 	m->host.val[j].index = msr;
951 	m->host.val[j].value = host_val;
952 }
953 
954 static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset)
955 {
956 	u64 guest_efer = vmx->vcpu.arch.efer;
957 	u64 ignore_bits = 0;
958 
959 	/* Shadow paging assumes NX to be available.  */
960 	if (!enable_ept)
961 		guest_efer |= EFER_NX;
962 
963 	/*
964 	 * LMA and LME handled by hardware; SCE meaningless outside long mode.
965 	 */
966 	ignore_bits |= EFER_SCE;
967 #ifdef CONFIG_X86_64
968 	ignore_bits |= EFER_LMA | EFER_LME;
969 	/* SCE is meaningful only in long mode on Intel */
970 	if (guest_efer & EFER_LMA)
971 		ignore_bits &= ~(u64)EFER_SCE;
972 #endif
973 
974 	/*
975 	 * On EPT, we can't emulate NX, so we must switch EFER atomically.
976 	 * On CPUs that support "load IA32_EFER", always switch EFER
977 	 * atomically, since it's faster than switching it manually.
978 	 */
979 	if (cpu_has_load_ia32_efer() ||
980 	    (enable_ept && ((vmx->vcpu.arch.efer ^ host_efer) & EFER_NX))) {
981 		if (!(guest_efer & EFER_LMA))
982 			guest_efer &= ~EFER_LME;
983 		if (guest_efer != host_efer)
984 			add_atomic_switch_msr(vmx, MSR_EFER,
985 					      guest_efer, host_efer, false);
986 		else
987 			clear_atomic_switch_msr(vmx, MSR_EFER);
988 		return false;
989 	} else {
990 		clear_atomic_switch_msr(vmx, MSR_EFER);
991 
992 		guest_efer &= ~ignore_bits;
993 		guest_efer |= host_efer & ignore_bits;
994 
995 		vmx->guest_msrs[efer_offset].data = guest_efer;
996 		vmx->guest_msrs[efer_offset].mask = ~ignore_bits;
997 
998 		return true;
999 	}
1000 }
1001 
1002 #ifdef CONFIG_X86_32
1003 /*
1004  * On 32-bit kernels, VM exits still load the FS and GS bases from the
1005  * VMCS rather than the segment table.  KVM uses this helper to figure
1006  * out the current bases to poke them into the VMCS before entry.
1007  */
1008 static unsigned long segment_base(u16 selector)
1009 {
1010 	struct desc_struct *table;
1011 	unsigned long v;
1012 
1013 	if (!(selector & ~SEGMENT_RPL_MASK))
1014 		return 0;
1015 
1016 	table = get_current_gdt_ro();
1017 
1018 	if ((selector & SEGMENT_TI_MASK) == SEGMENT_LDT) {
1019 		u16 ldt_selector = kvm_read_ldt();
1020 
1021 		if (!(ldt_selector & ~SEGMENT_RPL_MASK))
1022 			return 0;
1023 
1024 		table = (struct desc_struct *)segment_base(ldt_selector);
1025 	}
1026 	v = get_desc_base(&table[selector >> 3]);
1027 	return v;
1028 }
1029 #endif
1030 
1031 static inline bool pt_can_write_msr(struct vcpu_vmx *vmx)
1032 {
1033 	return vmx_pt_mode_is_host_guest() &&
1034 	       !(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN);
1035 }
1036 
1037 static inline void pt_load_msr(struct pt_ctx *ctx, u32 addr_range)
1038 {
1039 	u32 i;
1040 
1041 	wrmsrl(MSR_IA32_RTIT_STATUS, ctx->status);
1042 	wrmsrl(MSR_IA32_RTIT_OUTPUT_BASE, ctx->output_base);
1043 	wrmsrl(MSR_IA32_RTIT_OUTPUT_MASK, ctx->output_mask);
1044 	wrmsrl(MSR_IA32_RTIT_CR3_MATCH, ctx->cr3_match);
1045 	for (i = 0; i < addr_range; i++) {
1046 		wrmsrl(MSR_IA32_RTIT_ADDR0_A + i * 2, ctx->addr_a[i]);
1047 		wrmsrl(MSR_IA32_RTIT_ADDR0_B + i * 2, ctx->addr_b[i]);
1048 	}
1049 }
1050 
1051 static inline void pt_save_msr(struct pt_ctx *ctx, u32 addr_range)
1052 {
1053 	u32 i;
1054 
1055 	rdmsrl(MSR_IA32_RTIT_STATUS, ctx->status);
1056 	rdmsrl(MSR_IA32_RTIT_OUTPUT_BASE, ctx->output_base);
1057 	rdmsrl(MSR_IA32_RTIT_OUTPUT_MASK, ctx->output_mask);
1058 	rdmsrl(MSR_IA32_RTIT_CR3_MATCH, ctx->cr3_match);
1059 	for (i = 0; i < addr_range; i++) {
1060 		rdmsrl(MSR_IA32_RTIT_ADDR0_A + i * 2, ctx->addr_a[i]);
1061 		rdmsrl(MSR_IA32_RTIT_ADDR0_B + i * 2, ctx->addr_b[i]);
1062 	}
1063 }
1064 
1065 static void pt_guest_enter(struct vcpu_vmx *vmx)
1066 {
1067 	if (vmx_pt_mode_is_system())
1068 		return;
1069 
1070 	/*
1071 	 * GUEST_IA32_RTIT_CTL is already set in the VMCS.
1072 	 * Save host state before VM entry.
1073 	 */
1074 	rdmsrl(MSR_IA32_RTIT_CTL, vmx->pt_desc.host.ctl);
1075 	if (vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) {
1076 		wrmsrl(MSR_IA32_RTIT_CTL, 0);
1077 		pt_save_msr(&vmx->pt_desc.host, vmx->pt_desc.addr_range);
1078 		pt_load_msr(&vmx->pt_desc.guest, vmx->pt_desc.addr_range);
1079 	}
1080 }
1081 
1082 static void pt_guest_exit(struct vcpu_vmx *vmx)
1083 {
1084 	if (vmx_pt_mode_is_system())
1085 		return;
1086 
1087 	if (vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) {
1088 		pt_save_msr(&vmx->pt_desc.guest, vmx->pt_desc.addr_range);
1089 		pt_load_msr(&vmx->pt_desc.host, vmx->pt_desc.addr_range);
1090 	}
1091 
1092 	/* Reload host state (IA32_RTIT_CTL will be cleared on VM exit). */
1093 	wrmsrl(MSR_IA32_RTIT_CTL, vmx->pt_desc.host.ctl);
1094 }
1095 
1096 void vmx_set_host_fs_gs(struct vmcs_host_state *host, u16 fs_sel, u16 gs_sel,
1097 			unsigned long fs_base, unsigned long gs_base)
1098 {
1099 	if (unlikely(fs_sel != host->fs_sel)) {
1100 		if (!(fs_sel & 7))
1101 			vmcs_write16(HOST_FS_SELECTOR, fs_sel);
1102 		else
1103 			vmcs_write16(HOST_FS_SELECTOR, 0);
1104 		host->fs_sel = fs_sel;
1105 	}
1106 	if (unlikely(gs_sel != host->gs_sel)) {
1107 		if (!(gs_sel & 7))
1108 			vmcs_write16(HOST_GS_SELECTOR, gs_sel);
1109 		else
1110 			vmcs_write16(HOST_GS_SELECTOR, 0);
1111 		host->gs_sel = gs_sel;
1112 	}
1113 	if (unlikely(fs_base != host->fs_base)) {
1114 		vmcs_writel(HOST_FS_BASE, fs_base);
1115 		host->fs_base = fs_base;
1116 	}
1117 	if (unlikely(gs_base != host->gs_base)) {
1118 		vmcs_writel(HOST_GS_BASE, gs_base);
1119 		host->gs_base = gs_base;
1120 	}
1121 }
1122 
1123 void vmx_prepare_switch_to_guest(struct kvm_vcpu *vcpu)
1124 {
1125 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1126 	struct vmcs_host_state *host_state;
1127 #ifdef CONFIG_X86_64
1128 	int cpu = raw_smp_processor_id();
1129 #endif
1130 	unsigned long fs_base, gs_base;
1131 	u16 fs_sel, gs_sel;
1132 	int i;
1133 
1134 	vmx->req_immediate_exit = false;
1135 
1136 	/*
1137 	 * Note that guest MSRs to be saved/restored can also be changed
1138 	 * when guest state is loaded. This happens when guest transitions
1139 	 * to/from long-mode by setting MSR_EFER.LMA.
1140 	 */
1141 	if (!vmx->guest_msrs_ready) {
1142 		vmx->guest_msrs_ready = true;
1143 		for (i = 0; i < vmx->save_nmsrs; ++i)
1144 			kvm_set_shared_msr(vmx->guest_msrs[i].index,
1145 					   vmx->guest_msrs[i].data,
1146 					   vmx->guest_msrs[i].mask);
1147 
1148 	}
1149 
1150     	if (vmx->nested.need_vmcs12_to_shadow_sync)
1151 		nested_sync_vmcs12_to_shadow(vcpu);
1152 
1153 	if (vmx->guest_state_loaded)
1154 		return;
1155 
1156 	host_state = &vmx->loaded_vmcs->host_state;
1157 
1158 	/*
1159 	 * Set host fs and gs selectors.  Unfortunately, 22.2.3 does not
1160 	 * allow segment selectors with cpl > 0 or ti == 1.
1161 	 */
1162 	host_state->ldt_sel = kvm_read_ldt();
1163 
1164 #ifdef CONFIG_X86_64
1165 	savesegment(ds, host_state->ds_sel);
1166 	savesegment(es, host_state->es_sel);
1167 
1168 	gs_base = cpu_kernelmode_gs_base(cpu);
1169 	if (likely(is_64bit_mm(current->mm))) {
1170 		save_fsgs_for_kvm();
1171 		fs_sel = current->thread.fsindex;
1172 		gs_sel = current->thread.gsindex;
1173 		fs_base = current->thread.fsbase;
1174 		vmx->msr_host_kernel_gs_base = current->thread.gsbase;
1175 	} else {
1176 		savesegment(fs, fs_sel);
1177 		savesegment(gs, gs_sel);
1178 		fs_base = read_msr(MSR_FS_BASE);
1179 		vmx->msr_host_kernel_gs_base = read_msr(MSR_KERNEL_GS_BASE);
1180 	}
1181 
1182 	wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1183 #else
1184 	savesegment(fs, fs_sel);
1185 	savesegment(gs, gs_sel);
1186 	fs_base = segment_base(fs_sel);
1187 	gs_base = segment_base(gs_sel);
1188 #endif
1189 
1190 	vmx_set_host_fs_gs(host_state, fs_sel, gs_sel, fs_base, gs_base);
1191 	vmx->guest_state_loaded = true;
1192 }
1193 
1194 static void vmx_prepare_switch_to_host(struct vcpu_vmx *vmx)
1195 {
1196 	struct vmcs_host_state *host_state;
1197 
1198 	if (!vmx->guest_state_loaded)
1199 		return;
1200 
1201 	host_state = &vmx->loaded_vmcs->host_state;
1202 
1203 	++vmx->vcpu.stat.host_state_reload;
1204 
1205 #ifdef CONFIG_X86_64
1206 	rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1207 #endif
1208 	if (host_state->ldt_sel || (host_state->gs_sel & 7)) {
1209 		kvm_load_ldt(host_state->ldt_sel);
1210 #ifdef CONFIG_X86_64
1211 		load_gs_index(host_state->gs_sel);
1212 #else
1213 		loadsegment(gs, host_state->gs_sel);
1214 #endif
1215 	}
1216 	if (host_state->fs_sel & 7)
1217 		loadsegment(fs, host_state->fs_sel);
1218 #ifdef CONFIG_X86_64
1219 	if (unlikely(host_state->ds_sel | host_state->es_sel)) {
1220 		loadsegment(ds, host_state->ds_sel);
1221 		loadsegment(es, host_state->es_sel);
1222 	}
1223 #endif
1224 	invalidate_tss_limit();
1225 #ifdef CONFIG_X86_64
1226 	wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
1227 #endif
1228 	load_fixmap_gdt(raw_smp_processor_id());
1229 	vmx->guest_state_loaded = false;
1230 	vmx->guest_msrs_ready = false;
1231 }
1232 
1233 #ifdef CONFIG_X86_64
1234 static u64 vmx_read_guest_kernel_gs_base(struct vcpu_vmx *vmx)
1235 {
1236 	preempt_disable();
1237 	if (vmx->guest_state_loaded)
1238 		rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1239 	preempt_enable();
1240 	return vmx->msr_guest_kernel_gs_base;
1241 }
1242 
1243 static void vmx_write_guest_kernel_gs_base(struct vcpu_vmx *vmx, u64 data)
1244 {
1245 	preempt_disable();
1246 	if (vmx->guest_state_loaded)
1247 		wrmsrl(MSR_KERNEL_GS_BASE, data);
1248 	preempt_enable();
1249 	vmx->msr_guest_kernel_gs_base = data;
1250 }
1251 #endif
1252 
1253 static void vmx_vcpu_pi_load(struct kvm_vcpu *vcpu, int cpu)
1254 {
1255 	struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
1256 	struct pi_desc old, new;
1257 	unsigned int dest;
1258 
1259 	/*
1260 	 * In case of hot-plug or hot-unplug, we may have to undo
1261 	 * vmx_vcpu_pi_put even if there is no assigned device.  And we
1262 	 * always keep PI.NDST up to date for simplicity: it makes the
1263 	 * code easier, and CPU migration is not a fast path.
1264 	 */
1265 	if (!pi_test_sn(pi_desc) && vcpu->cpu == cpu)
1266 		return;
1267 
1268 	/*
1269 	 * If the 'nv' field is POSTED_INTR_WAKEUP_VECTOR, do not change
1270 	 * PI.NDST: pi_post_block is the one expected to change PID.NDST and the
1271 	 * wakeup handler expects the vCPU to be on the blocked_vcpu_list that
1272 	 * matches PI.NDST. Otherwise, a vcpu may not be able to be woken up
1273 	 * correctly.
1274 	 */
1275 	if (pi_desc->nv == POSTED_INTR_WAKEUP_VECTOR || vcpu->cpu == cpu) {
1276 		pi_clear_sn(pi_desc);
1277 		goto after_clear_sn;
1278 	}
1279 
1280 	/* The full case.  */
1281 	do {
1282 		old.control = new.control = pi_desc->control;
1283 
1284 		dest = cpu_physical_id(cpu);
1285 
1286 		if (x2apic_enabled())
1287 			new.ndst = dest;
1288 		else
1289 			new.ndst = (dest << 8) & 0xFF00;
1290 
1291 		new.sn = 0;
1292 	} while (cmpxchg64(&pi_desc->control, old.control,
1293 			   new.control) != old.control);
1294 
1295 after_clear_sn:
1296 
1297 	/*
1298 	 * Clear SN before reading the bitmap.  The VT-d firmware
1299 	 * writes the bitmap and reads SN atomically (5.2.3 in the
1300 	 * spec), so it doesn't really have a memory barrier that
1301 	 * pairs with this, but we cannot do that and we need one.
1302 	 */
1303 	smp_mb__after_atomic();
1304 
1305 	if (!pi_is_pir_empty(pi_desc))
1306 		pi_set_on(pi_desc);
1307 }
1308 
1309 void vmx_vcpu_load_vmcs(struct kvm_vcpu *vcpu, int cpu)
1310 {
1311 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1312 	bool already_loaded = vmx->loaded_vmcs->cpu == cpu;
1313 
1314 	if (!already_loaded) {
1315 		loaded_vmcs_clear(vmx->loaded_vmcs);
1316 		local_irq_disable();
1317 
1318 		/*
1319 		 * Ensure loaded_vmcs->cpu is read before adding loaded_vmcs to
1320 		 * this cpu's percpu list, otherwise it may not yet be deleted
1321 		 * from its previous cpu's percpu list.  Pairs with the
1322 		 * smb_wmb() in __loaded_vmcs_clear().
1323 		 */
1324 		smp_rmb();
1325 
1326 		list_add(&vmx->loaded_vmcs->loaded_vmcss_on_cpu_link,
1327 			 &per_cpu(loaded_vmcss_on_cpu, cpu));
1328 		local_irq_enable();
1329 	}
1330 
1331 	if (per_cpu(current_vmcs, cpu) != vmx->loaded_vmcs->vmcs) {
1332 		per_cpu(current_vmcs, cpu) = vmx->loaded_vmcs->vmcs;
1333 		vmcs_load(vmx->loaded_vmcs->vmcs);
1334 		indirect_branch_prediction_barrier();
1335 	}
1336 
1337 	if (!already_loaded) {
1338 		void *gdt = get_current_gdt_ro();
1339 		unsigned long sysenter_esp;
1340 
1341 		kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1342 
1343 		/*
1344 		 * Linux uses per-cpu TSS and GDT, so set these when switching
1345 		 * processors.  See 22.2.4.
1346 		 */
1347 		vmcs_writel(HOST_TR_BASE,
1348 			    (unsigned long)&get_cpu_entry_area(cpu)->tss.x86_tss);
1349 		vmcs_writel(HOST_GDTR_BASE, (unsigned long)gdt);   /* 22.2.4 */
1350 
1351 		rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
1352 		vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
1353 
1354 		vmx->loaded_vmcs->cpu = cpu;
1355 	}
1356 
1357 	/* Setup TSC multiplier */
1358 	if (kvm_has_tsc_control &&
1359 	    vmx->current_tsc_ratio != vcpu->arch.tsc_scaling_ratio)
1360 		decache_tsc_multiplier(vmx);
1361 }
1362 
1363 /*
1364  * Switches to specified vcpu, until a matching vcpu_put(), but assumes
1365  * vcpu mutex is already taken.
1366  */
1367 void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1368 {
1369 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1370 
1371 	vmx_vcpu_load_vmcs(vcpu, cpu);
1372 
1373 	vmx_vcpu_pi_load(vcpu, cpu);
1374 
1375 	vmx->host_pkru = read_pkru();
1376 	vmx->host_debugctlmsr = get_debugctlmsr();
1377 }
1378 
1379 static void vmx_vcpu_pi_put(struct kvm_vcpu *vcpu)
1380 {
1381 	struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
1382 
1383 	if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
1384 		!irq_remapping_cap(IRQ_POSTING_CAP)  ||
1385 		!kvm_vcpu_apicv_active(vcpu))
1386 		return;
1387 
1388 	/* Set SN when the vCPU is preempted */
1389 	if (vcpu->preempted)
1390 		pi_set_sn(pi_desc);
1391 }
1392 
1393 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
1394 {
1395 	vmx_vcpu_pi_put(vcpu);
1396 
1397 	vmx_prepare_switch_to_host(to_vmx(vcpu));
1398 }
1399 
1400 static bool emulation_required(struct kvm_vcpu *vcpu)
1401 {
1402 	return emulate_invalid_guest_state && !guest_state_valid(vcpu);
1403 }
1404 
1405 unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
1406 {
1407 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1408 	unsigned long rflags, save_rflags;
1409 
1410 	if (!kvm_register_is_available(vcpu, VCPU_EXREG_RFLAGS)) {
1411 		kvm_register_mark_available(vcpu, VCPU_EXREG_RFLAGS);
1412 		rflags = vmcs_readl(GUEST_RFLAGS);
1413 		if (vmx->rmode.vm86_active) {
1414 			rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
1415 			save_rflags = vmx->rmode.save_rflags;
1416 			rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
1417 		}
1418 		vmx->rflags = rflags;
1419 	}
1420 	return vmx->rflags;
1421 }
1422 
1423 void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
1424 {
1425 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1426 	unsigned long old_rflags;
1427 
1428 	if (enable_unrestricted_guest) {
1429 		kvm_register_mark_available(vcpu, VCPU_EXREG_RFLAGS);
1430 		vmx->rflags = rflags;
1431 		vmcs_writel(GUEST_RFLAGS, rflags);
1432 		return;
1433 	}
1434 
1435 	old_rflags = vmx_get_rflags(vcpu);
1436 	vmx->rflags = rflags;
1437 	if (vmx->rmode.vm86_active) {
1438 		vmx->rmode.save_rflags = rflags;
1439 		rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
1440 	}
1441 	vmcs_writel(GUEST_RFLAGS, rflags);
1442 
1443 	if ((old_rflags ^ vmx->rflags) & X86_EFLAGS_VM)
1444 		vmx->emulation_required = emulation_required(vcpu);
1445 }
1446 
1447 u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu)
1448 {
1449 	u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1450 	int ret = 0;
1451 
1452 	if (interruptibility & GUEST_INTR_STATE_STI)
1453 		ret |= KVM_X86_SHADOW_INT_STI;
1454 	if (interruptibility & GUEST_INTR_STATE_MOV_SS)
1455 		ret |= KVM_X86_SHADOW_INT_MOV_SS;
1456 
1457 	return ret;
1458 }
1459 
1460 void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
1461 {
1462 	u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1463 	u32 interruptibility = interruptibility_old;
1464 
1465 	interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
1466 
1467 	if (mask & KVM_X86_SHADOW_INT_MOV_SS)
1468 		interruptibility |= GUEST_INTR_STATE_MOV_SS;
1469 	else if (mask & KVM_X86_SHADOW_INT_STI)
1470 		interruptibility |= GUEST_INTR_STATE_STI;
1471 
1472 	if ((interruptibility != interruptibility_old))
1473 		vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
1474 }
1475 
1476 static int vmx_rtit_ctl_check(struct kvm_vcpu *vcpu, u64 data)
1477 {
1478 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1479 	unsigned long value;
1480 
1481 	/*
1482 	 * Any MSR write that attempts to change bits marked reserved will
1483 	 * case a #GP fault.
1484 	 */
1485 	if (data & vmx->pt_desc.ctl_bitmask)
1486 		return 1;
1487 
1488 	/*
1489 	 * Any attempt to modify IA32_RTIT_CTL while TraceEn is set will
1490 	 * result in a #GP unless the same write also clears TraceEn.
1491 	 */
1492 	if ((vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) &&
1493 		((vmx->pt_desc.guest.ctl ^ data) & ~RTIT_CTL_TRACEEN))
1494 		return 1;
1495 
1496 	/*
1497 	 * WRMSR to IA32_RTIT_CTL that sets TraceEn but clears this bit
1498 	 * and FabricEn would cause #GP, if
1499 	 * CPUID.(EAX=14H, ECX=0):ECX.SNGLRGNOUT[bit 2] = 0
1500 	 */
1501 	if ((data & RTIT_CTL_TRACEEN) && !(data & RTIT_CTL_TOPA) &&
1502 		!(data & RTIT_CTL_FABRIC_EN) &&
1503 		!intel_pt_validate_cap(vmx->pt_desc.caps,
1504 					PT_CAP_single_range_output))
1505 		return 1;
1506 
1507 	/*
1508 	 * MTCFreq, CycThresh and PSBFreq encodings check, any MSR write that
1509 	 * utilize encodings marked reserved will casue a #GP fault.
1510 	 */
1511 	value = intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc_periods);
1512 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc) &&
1513 			!test_bit((data & RTIT_CTL_MTC_RANGE) >>
1514 			RTIT_CTL_MTC_RANGE_OFFSET, &value))
1515 		return 1;
1516 	value = intel_pt_validate_cap(vmx->pt_desc.caps,
1517 						PT_CAP_cycle_thresholds);
1518 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc) &&
1519 			!test_bit((data & RTIT_CTL_CYC_THRESH) >>
1520 			RTIT_CTL_CYC_THRESH_OFFSET, &value))
1521 		return 1;
1522 	value = intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_periods);
1523 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc) &&
1524 			!test_bit((data & RTIT_CTL_PSB_FREQ) >>
1525 			RTIT_CTL_PSB_FREQ_OFFSET, &value))
1526 		return 1;
1527 
1528 	/*
1529 	 * If ADDRx_CFG is reserved or the encodings is >2 will
1530 	 * cause a #GP fault.
1531 	 */
1532 	value = (data & RTIT_CTL_ADDR0) >> RTIT_CTL_ADDR0_OFFSET;
1533 	if ((value && (vmx->pt_desc.addr_range < 1)) || (value > 2))
1534 		return 1;
1535 	value = (data & RTIT_CTL_ADDR1) >> RTIT_CTL_ADDR1_OFFSET;
1536 	if ((value && (vmx->pt_desc.addr_range < 2)) || (value > 2))
1537 		return 1;
1538 	value = (data & RTIT_CTL_ADDR2) >> RTIT_CTL_ADDR2_OFFSET;
1539 	if ((value && (vmx->pt_desc.addr_range < 3)) || (value > 2))
1540 		return 1;
1541 	value = (data & RTIT_CTL_ADDR3) >> RTIT_CTL_ADDR3_OFFSET;
1542 	if ((value && (vmx->pt_desc.addr_range < 4)) || (value > 2))
1543 		return 1;
1544 
1545 	return 0;
1546 }
1547 
1548 static int skip_emulated_instruction(struct kvm_vcpu *vcpu)
1549 {
1550 	unsigned long rip;
1551 
1552 	/*
1553 	 * Using VMCS.VM_EXIT_INSTRUCTION_LEN on EPT misconfig depends on
1554 	 * undefined behavior: Intel's SDM doesn't mandate the VMCS field be
1555 	 * set when EPT misconfig occurs.  In practice, real hardware updates
1556 	 * VM_EXIT_INSTRUCTION_LEN on EPT misconfig, but other hypervisors
1557 	 * (namely Hyper-V) don't set it due to it being undefined behavior,
1558 	 * i.e. we end up advancing IP with some random value.
1559 	 */
1560 	if (!static_cpu_has(X86_FEATURE_HYPERVISOR) ||
1561 	    to_vmx(vcpu)->exit_reason != EXIT_REASON_EPT_MISCONFIG) {
1562 		rip = kvm_rip_read(vcpu);
1563 		rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
1564 		kvm_rip_write(vcpu, rip);
1565 	} else {
1566 		if (!kvm_emulate_instruction(vcpu, EMULTYPE_SKIP))
1567 			return 0;
1568 	}
1569 
1570 	/* skipping an emulated instruction also counts */
1571 	vmx_set_interrupt_shadow(vcpu, 0);
1572 
1573 	return 1;
1574 }
1575 
1576 
1577 /*
1578  * Recognizes a pending MTF VM-exit and records the nested state for later
1579  * delivery.
1580  */
1581 static void vmx_update_emulated_instruction(struct kvm_vcpu *vcpu)
1582 {
1583 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1584 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1585 
1586 	if (!is_guest_mode(vcpu))
1587 		return;
1588 
1589 	/*
1590 	 * Per the SDM, MTF takes priority over debug-trap exceptions besides
1591 	 * T-bit traps. As instruction emulation is completed (i.e. at the
1592 	 * instruction boundary), any #DB exception pending delivery must be a
1593 	 * debug-trap. Record the pending MTF state to be delivered in
1594 	 * vmx_check_nested_events().
1595 	 */
1596 	if (nested_cpu_has_mtf(vmcs12) &&
1597 	    (!vcpu->arch.exception.pending ||
1598 	     vcpu->arch.exception.nr == DB_VECTOR))
1599 		vmx->nested.mtf_pending = true;
1600 	else
1601 		vmx->nested.mtf_pending = false;
1602 }
1603 
1604 static int vmx_skip_emulated_instruction(struct kvm_vcpu *vcpu)
1605 {
1606 	vmx_update_emulated_instruction(vcpu);
1607 	return skip_emulated_instruction(vcpu);
1608 }
1609 
1610 static void vmx_clear_hlt(struct kvm_vcpu *vcpu)
1611 {
1612 	/*
1613 	 * Ensure that we clear the HLT state in the VMCS.  We don't need to
1614 	 * explicitly skip the instruction because if the HLT state is set,
1615 	 * then the instruction is already executing and RIP has already been
1616 	 * advanced.
1617 	 */
1618 	if (kvm_hlt_in_guest(vcpu->kvm) &&
1619 			vmcs_read32(GUEST_ACTIVITY_STATE) == GUEST_ACTIVITY_HLT)
1620 		vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
1621 }
1622 
1623 static void vmx_queue_exception(struct kvm_vcpu *vcpu)
1624 {
1625 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1626 	unsigned nr = vcpu->arch.exception.nr;
1627 	bool has_error_code = vcpu->arch.exception.has_error_code;
1628 	u32 error_code = vcpu->arch.exception.error_code;
1629 	u32 intr_info = nr | INTR_INFO_VALID_MASK;
1630 
1631 	kvm_deliver_exception_payload(vcpu);
1632 
1633 	if (has_error_code) {
1634 		vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
1635 		intr_info |= INTR_INFO_DELIVER_CODE_MASK;
1636 	}
1637 
1638 	if (vmx->rmode.vm86_active) {
1639 		int inc_eip = 0;
1640 		if (kvm_exception_is_soft(nr))
1641 			inc_eip = vcpu->arch.event_exit_inst_len;
1642 		kvm_inject_realmode_interrupt(vcpu, nr, inc_eip);
1643 		return;
1644 	}
1645 
1646 	WARN_ON_ONCE(vmx->emulation_required);
1647 
1648 	if (kvm_exception_is_soft(nr)) {
1649 		vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
1650 			     vmx->vcpu.arch.event_exit_inst_len);
1651 		intr_info |= INTR_TYPE_SOFT_EXCEPTION;
1652 	} else
1653 		intr_info |= INTR_TYPE_HARD_EXCEPTION;
1654 
1655 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
1656 
1657 	vmx_clear_hlt(vcpu);
1658 }
1659 
1660 /*
1661  * Swap MSR entry in host/guest MSR entry array.
1662  */
1663 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
1664 {
1665 	struct shared_msr_entry tmp;
1666 
1667 	tmp = vmx->guest_msrs[to];
1668 	vmx->guest_msrs[to] = vmx->guest_msrs[from];
1669 	vmx->guest_msrs[from] = tmp;
1670 }
1671 
1672 /*
1673  * Set up the vmcs to automatically save and restore system
1674  * msrs.  Don't touch the 64-bit msrs if the guest is in legacy
1675  * mode, as fiddling with msrs is very expensive.
1676  */
1677 static void setup_msrs(struct vcpu_vmx *vmx)
1678 {
1679 	int save_nmsrs, index;
1680 
1681 	save_nmsrs = 0;
1682 #ifdef CONFIG_X86_64
1683 	/*
1684 	 * The SYSCALL MSRs are only needed on long mode guests, and only
1685 	 * when EFER.SCE is set.
1686 	 */
1687 	if (is_long_mode(&vmx->vcpu) && (vmx->vcpu.arch.efer & EFER_SCE)) {
1688 		index = __find_msr_index(vmx, MSR_STAR);
1689 		if (index >= 0)
1690 			move_msr_up(vmx, index, save_nmsrs++);
1691 		index = __find_msr_index(vmx, MSR_LSTAR);
1692 		if (index >= 0)
1693 			move_msr_up(vmx, index, save_nmsrs++);
1694 		index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
1695 		if (index >= 0)
1696 			move_msr_up(vmx, index, save_nmsrs++);
1697 	}
1698 #endif
1699 	index = __find_msr_index(vmx, MSR_EFER);
1700 	if (index >= 0 && update_transition_efer(vmx, index))
1701 		move_msr_up(vmx, index, save_nmsrs++);
1702 	index = __find_msr_index(vmx, MSR_TSC_AUX);
1703 	if (index >= 0 && guest_cpuid_has(&vmx->vcpu, X86_FEATURE_RDTSCP))
1704 		move_msr_up(vmx, index, save_nmsrs++);
1705 	index = __find_msr_index(vmx, MSR_IA32_TSX_CTRL);
1706 	if (index >= 0)
1707 		move_msr_up(vmx, index, save_nmsrs++);
1708 
1709 	vmx->save_nmsrs = save_nmsrs;
1710 	vmx->guest_msrs_ready = false;
1711 
1712 	if (cpu_has_vmx_msr_bitmap())
1713 		vmx_update_msr_bitmap(&vmx->vcpu);
1714 }
1715 
1716 static u64 vmx_read_l1_tsc_offset(struct kvm_vcpu *vcpu)
1717 {
1718 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1719 
1720 	if (is_guest_mode(vcpu) &&
1721 	    (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETTING))
1722 		return vcpu->arch.tsc_offset - vmcs12->tsc_offset;
1723 
1724 	return vcpu->arch.tsc_offset;
1725 }
1726 
1727 static u64 vmx_write_l1_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1728 {
1729 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1730 	u64 g_tsc_offset = 0;
1731 
1732 	/*
1733 	 * We're here if L1 chose not to trap WRMSR to TSC. According
1734 	 * to the spec, this should set L1's TSC; The offset that L1
1735 	 * set for L2 remains unchanged, and still needs to be added
1736 	 * to the newly set TSC to get L2's TSC.
1737 	 */
1738 	if (is_guest_mode(vcpu) &&
1739 	    (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETTING))
1740 		g_tsc_offset = vmcs12->tsc_offset;
1741 
1742 	trace_kvm_write_tsc_offset(vcpu->vcpu_id,
1743 				   vcpu->arch.tsc_offset - g_tsc_offset,
1744 				   offset);
1745 	vmcs_write64(TSC_OFFSET, offset + g_tsc_offset);
1746 	return offset + g_tsc_offset;
1747 }
1748 
1749 /*
1750  * nested_vmx_allowed() checks whether a guest should be allowed to use VMX
1751  * instructions and MSRs (i.e., nested VMX). Nested VMX is disabled for
1752  * all guests if the "nested" module option is off, and can also be disabled
1753  * for a single guest by disabling its VMX cpuid bit.
1754  */
1755 bool nested_vmx_allowed(struct kvm_vcpu *vcpu)
1756 {
1757 	return nested && guest_cpuid_has(vcpu, X86_FEATURE_VMX);
1758 }
1759 
1760 static inline bool vmx_feature_control_msr_valid(struct kvm_vcpu *vcpu,
1761 						 uint64_t val)
1762 {
1763 	uint64_t valid_bits = to_vmx(vcpu)->msr_ia32_feature_control_valid_bits;
1764 
1765 	return !(val & ~valid_bits);
1766 }
1767 
1768 static int vmx_get_msr_feature(struct kvm_msr_entry *msr)
1769 {
1770 	switch (msr->index) {
1771 	case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
1772 		if (!nested)
1773 			return 1;
1774 		return vmx_get_vmx_msr(&vmcs_config.nested, msr->index, &msr->data);
1775 	default:
1776 		return 1;
1777 	}
1778 }
1779 
1780 /*
1781  * Reads an msr value (of 'msr_index') into 'pdata'.
1782  * Returns 0 on success, non-0 otherwise.
1783  * Assumes vcpu_load() was already called.
1784  */
1785 static int vmx_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1786 {
1787 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1788 	struct shared_msr_entry *msr;
1789 	u32 index;
1790 
1791 	switch (msr_info->index) {
1792 #ifdef CONFIG_X86_64
1793 	case MSR_FS_BASE:
1794 		msr_info->data = vmcs_readl(GUEST_FS_BASE);
1795 		break;
1796 	case MSR_GS_BASE:
1797 		msr_info->data = vmcs_readl(GUEST_GS_BASE);
1798 		break;
1799 	case MSR_KERNEL_GS_BASE:
1800 		msr_info->data = vmx_read_guest_kernel_gs_base(vmx);
1801 		break;
1802 #endif
1803 	case MSR_EFER:
1804 		return kvm_get_msr_common(vcpu, msr_info);
1805 	case MSR_IA32_TSX_CTRL:
1806 		if (!msr_info->host_initiated &&
1807 		    !(vcpu->arch.arch_capabilities & ARCH_CAP_TSX_CTRL_MSR))
1808 			return 1;
1809 		goto find_shared_msr;
1810 	case MSR_IA32_UMWAIT_CONTROL:
1811 		if (!msr_info->host_initiated && !vmx_has_waitpkg(vmx))
1812 			return 1;
1813 
1814 		msr_info->data = vmx->msr_ia32_umwait_control;
1815 		break;
1816 	case MSR_IA32_SPEC_CTRL:
1817 		if (!msr_info->host_initiated &&
1818 		    !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
1819 			return 1;
1820 
1821 		msr_info->data = to_vmx(vcpu)->spec_ctrl;
1822 		break;
1823 	case MSR_IA32_SYSENTER_CS:
1824 		msr_info->data = vmcs_read32(GUEST_SYSENTER_CS);
1825 		break;
1826 	case MSR_IA32_SYSENTER_EIP:
1827 		msr_info->data = vmcs_readl(GUEST_SYSENTER_EIP);
1828 		break;
1829 	case MSR_IA32_SYSENTER_ESP:
1830 		msr_info->data = vmcs_readl(GUEST_SYSENTER_ESP);
1831 		break;
1832 	case MSR_IA32_BNDCFGS:
1833 		if (!kvm_mpx_supported() ||
1834 		    (!msr_info->host_initiated &&
1835 		     !guest_cpuid_has(vcpu, X86_FEATURE_MPX)))
1836 			return 1;
1837 		msr_info->data = vmcs_read64(GUEST_BNDCFGS);
1838 		break;
1839 	case MSR_IA32_MCG_EXT_CTL:
1840 		if (!msr_info->host_initiated &&
1841 		    !(vmx->msr_ia32_feature_control &
1842 		      FEAT_CTL_LMCE_ENABLED))
1843 			return 1;
1844 		msr_info->data = vcpu->arch.mcg_ext_ctl;
1845 		break;
1846 	case MSR_IA32_FEAT_CTL:
1847 		msr_info->data = vmx->msr_ia32_feature_control;
1848 		break;
1849 	case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
1850 		if (!nested_vmx_allowed(vcpu))
1851 			return 1;
1852 		if (vmx_get_vmx_msr(&vmx->nested.msrs, msr_info->index,
1853 				    &msr_info->data))
1854 			return 1;
1855 		/*
1856 		 * Enlightened VMCS v1 doesn't have certain fields, but buggy
1857 		 * Hyper-V versions are still trying to use corresponding
1858 		 * features when they are exposed. Filter out the essential
1859 		 * minimum.
1860 		 */
1861 		if (!msr_info->host_initiated &&
1862 		    vmx->nested.enlightened_vmcs_enabled)
1863 			nested_evmcs_filter_control_msr(msr_info->index,
1864 							&msr_info->data);
1865 		break;
1866 	case MSR_IA32_RTIT_CTL:
1867 		if (!vmx_pt_mode_is_host_guest())
1868 			return 1;
1869 		msr_info->data = vmx->pt_desc.guest.ctl;
1870 		break;
1871 	case MSR_IA32_RTIT_STATUS:
1872 		if (!vmx_pt_mode_is_host_guest())
1873 			return 1;
1874 		msr_info->data = vmx->pt_desc.guest.status;
1875 		break;
1876 	case MSR_IA32_RTIT_CR3_MATCH:
1877 		if (!vmx_pt_mode_is_host_guest() ||
1878 			!intel_pt_validate_cap(vmx->pt_desc.caps,
1879 						PT_CAP_cr3_filtering))
1880 			return 1;
1881 		msr_info->data = vmx->pt_desc.guest.cr3_match;
1882 		break;
1883 	case MSR_IA32_RTIT_OUTPUT_BASE:
1884 		if (!vmx_pt_mode_is_host_guest() ||
1885 			(!intel_pt_validate_cap(vmx->pt_desc.caps,
1886 					PT_CAP_topa_output) &&
1887 			 !intel_pt_validate_cap(vmx->pt_desc.caps,
1888 					PT_CAP_single_range_output)))
1889 			return 1;
1890 		msr_info->data = vmx->pt_desc.guest.output_base;
1891 		break;
1892 	case MSR_IA32_RTIT_OUTPUT_MASK:
1893 		if (!vmx_pt_mode_is_host_guest() ||
1894 			(!intel_pt_validate_cap(vmx->pt_desc.caps,
1895 					PT_CAP_topa_output) &&
1896 			 !intel_pt_validate_cap(vmx->pt_desc.caps,
1897 					PT_CAP_single_range_output)))
1898 			return 1;
1899 		msr_info->data = vmx->pt_desc.guest.output_mask;
1900 		break;
1901 	case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
1902 		index = msr_info->index - MSR_IA32_RTIT_ADDR0_A;
1903 		if (!vmx_pt_mode_is_host_guest() ||
1904 			(index >= 2 * intel_pt_validate_cap(vmx->pt_desc.caps,
1905 					PT_CAP_num_address_ranges)))
1906 			return 1;
1907 		if (index % 2)
1908 			msr_info->data = vmx->pt_desc.guest.addr_b[index / 2];
1909 		else
1910 			msr_info->data = vmx->pt_desc.guest.addr_a[index / 2];
1911 		break;
1912 	case MSR_TSC_AUX:
1913 		if (!msr_info->host_initiated &&
1914 		    !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP))
1915 			return 1;
1916 		goto find_shared_msr;
1917 	default:
1918 	find_shared_msr:
1919 		msr = find_msr_entry(vmx, msr_info->index);
1920 		if (msr) {
1921 			msr_info->data = msr->data;
1922 			break;
1923 		}
1924 		return kvm_get_msr_common(vcpu, msr_info);
1925 	}
1926 
1927 	return 0;
1928 }
1929 
1930 /*
1931  * Writes msr value into the appropriate "register".
1932  * Returns 0 on success, non-0 otherwise.
1933  * Assumes vcpu_load() was already called.
1934  */
1935 static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1936 {
1937 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1938 	struct shared_msr_entry *msr;
1939 	int ret = 0;
1940 	u32 msr_index = msr_info->index;
1941 	u64 data = msr_info->data;
1942 	u32 index;
1943 
1944 	switch (msr_index) {
1945 	case MSR_EFER:
1946 		ret = kvm_set_msr_common(vcpu, msr_info);
1947 		break;
1948 #ifdef CONFIG_X86_64
1949 	case MSR_FS_BASE:
1950 		vmx_segment_cache_clear(vmx);
1951 		vmcs_writel(GUEST_FS_BASE, data);
1952 		break;
1953 	case MSR_GS_BASE:
1954 		vmx_segment_cache_clear(vmx);
1955 		vmcs_writel(GUEST_GS_BASE, data);
1956 		break;
1957 	case MSR_KERNEL_GS_BASE:
1958 		vmx_write_guest_kernel_gs_base(vmx, data);
1959 		break;
1960 #endif
1961 	case MSR_IA32_SYSENTER_CS:
1962 		if (is_guest_mode(vcpu))
1963 			get_vmcs12(vcpu)->guest_sysenter_cs = data;
1964 		vmcs_write32(GUEST_SYSENTER_CS, data);
1965 		break;
1966 	case MSR_IA32_SYSENTER_EIP:
1967 		if (is_guest_mode(vcpu))
1968 			get_vmcs12(vcpu)->guest_sysenter_eip = data;
1969 		vmcs_writel(GUEST_SYSENTER_EIP, data);
1970 		break;
1971 	case MSR_IA32_SYSENTER_ESP:
1972 		if (is_guest_mode(vcpu))
1973 			get_vmcs12(vcpu)->guest_sysenter_esp = data;
1974 		vmcs_writel(GUEST_SYSENTER_ESP, data);
1975 		break;
1976 	case MSR_IA32_DEBUGCTLMSR:
1977 		if (is_guest_mode(vcpu) && get_vmcs12(vcpu)->vm_exit_controls &
1978 						VM_EXIT_SAVE_DEBUG_CONTROLS)
1979 			get_vmcs12(vcpu)->guest_ia32_debugctl = data;
1980 
1981 		ret = kvm_set_msr_common(vcpu, msr_info);
1982 		break;
1983 
1984 	case MSR_IA32_BNDCFGS:
1985 		if (!kvm_mpx_supported() ||
1986 		    (!msr_info->host_initiated &&
1987 		     !guest_cpuid_has(vcpu, X86_FEATURE_MPX)))
1988 			return 1;
1989 		if (is_noncanonical_address(data & PAGE_MASK, vcpu) ||
1990 		    (data & MSR_IA32_BNDCFGS_RSVD))
1991 			return 1;
1992 		vmcs_write64(GUEST_BNDCFGS, data);
1993 		break;
1994 	case MSR_IA32_UMWAIT_CONTROL:
1995 		if (!msr_info->host_initiated && !vmx_has_waitpkg(vmx))
1996 			return 1;
1997 
1998 		/* The reserved bit 1 and non-32 bit [63:32] should be zero */
1999 		if (data & (BIT_ULL(1) | GENMASK_ULL(63, 32)))
2000 			return 1;
2001 
2002 		vmx->msr_ia32_umwait_control = data;
2003 		break;
2004 	case MSR_IA32_SPEC_CTRL:
2005 		if (!msr_info->host_initiated &&
2006 		    !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
2007 			return 1;
2008 
2009 		if (data & ~kvm_spec_ctrl_valid_bits(vcpu))
2010 			return 1;
2011 
2012 		vmx->spec_ctrl = data;
2013 		if (!data)
2014 			break;
2015 
2016 		/*
2017 		 * For non-nested:
2018 		 * When it's written (to non-zero) for the first time, pass
2019 		 * it through.
2020 		 *
2021 		 * For nested:
2022 		 * The handling of the MSR bitmap for L2 guests is done in
2023 		 * nested_vmx_prepare_msr_bitmap. We should not touch the
2024 		 * vmcs02.msr_bitmap here since it gets completely overwritten
2025 		 * in the merging. We update the vmcs01 here for L1 as well
2026 		 * since it will end up touching the MSR anyway now.
2027 		 */
2028 		vmx_disable_intercept_for_msr(vmx->vmcs01.msr_bitmap,
2029 					      MSR_IA32_SPEC_CTRL,
2030 					      MSR_TYPE_RW);
2031 		break;
2032 	case MSR_IA32_TSX_CTRL:
2033 		if (!msr_info->host_initiated &&
2034 		    !(vcpu->arch.arch_capabilities & ARCH_CAP_TSX_CTRL_MSR))
2035 			return 1;
2036 		if (data & ~(TSX_CTRL_RTM_DISABLE | TSX_CTRL_CPUID_CLEAR))
2037 			return 1;
2038 		goto find_shared_msr;
2039 	case MSR_IA32_PRED_CMD:
2040 		if (!msr_info->host_initiated &&
2041 		    !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
2042 			return 1;
2043 
2044 		if (data & ~PRED_CMD_IBPB)
2045 			return 1;
2046 		if (!boot_cpu_has(X86_FEATURE_SPEC_CTRL))
2047 			return 1;
2048 		if (!data)
2049 			break;
2050 
2051 		wrmsrl(MSR_IA32_PRED_CMD, PRED_CMD_IBPB);
2052 
2053 		/*
2054 		 * For non-nested:
2055 		 * When it's written (to non-zero) for the first time, pass
2056 		 * it through.
2057 		 *
2058 		 * For nested:
2059 		 * The handling of the MSR bitmap for L2 guests is done in
2060 		 * nested_vmx_prepare_msr_bitmap. We should not touch the
2061 		 * vmcs02.msr_bitmap here since it gets completely overwritten
2062 		 * in the merging.
2063 		 */
2064 		vmx_disable_intercept_for_msr(vmx->vmcs01.msr_bitmap, MSR_IA32_PRED_CMD,
2065 					      MSR_TYPE_W);
2066 		break;
2067 	case MSR_IA32_CR_PAT:
2068 		if (!kvm_pat_valid(data))
2069 			return 1;
2070 
2071 		if (is_guest_mode(vcpu) &&
2072 		    get_vmcs12(vcpu)->vm_exit_controls & VM_EXIT_SAVE_IA32_PAT)
2073 			get_vmcs12(vcpu)->guest_ia32_pat = data;
2074 
2075 		if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2076 			vmcs_write64(GUEST_IA32_PAT, data);
2077 			vcpu->arch.pat = data;
2078 			break;
2079 		}
2080 		ret = kvm_set_msr_common(vcpu, msr_info);
2081 		break;
2082 	case MSR_IA32_TSC_ADJUST:
2083 		ret = kvm_set_msr_common(vcpu, msr_info);
2084 		break;
2085 	case MSR_IA32_MCG_EXT_CTL:
2086 		if ((!msr_info->host_initiated &&
2087 		     !(to_vmx(vcpu)->msr_ia32_feature_control &
2088 		       FEAT_CTL_LMCE_ENABLED)) ||
2089 		    (data & ~MCG_EXT_CTL_LMCE_EN))
2090 			return 1;
2091 		vcpu->arch.mcg_ext_ctl = data;
2092 		break;
2093 	case MSR_IA32_FEAT_CTL:
2094 		if (!vmx_feature_control_msr_valid(vcpu, data) ||
2095 		    (to_vmx(vcpu)->msr_ia32_feature_control &
2096 		     FEAT_CTL_LOCKED && !msr_info->host_initiated))
2097 			return 1;
2098 		vmx->msr_ia32_feature_control = data;
2099 		if (msr_info->host_initiated && data == 0)
2100 			vmx_leave_nested(vcpu);
2101 		break;
2102 	case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
2103 		if (!msr_info->host_initiated)
2104 			return 1; /* they are read-only */
2105 		if (!nested_vmx_allowed(vcpu))
2106 			return 1;
2107 		return vmx_set_vmx_msr(vcpu, msr_index, data);
2108 	case MSR_IA32_RTIT_CTL:
2109 		if (!vmx_pt_mode_is_host_guest() ||
2110 			vmx_rtit_ctl_check(vcpu, data) ||
2111 			vmx->nested.vmxon)
2112 			return 1;
2113 		vmcs_write64(GUEST_IA32_RTIT_CTL, data);
2114 		vmx->pt_desc.guest.ctl = data;
2115 		pt_update_intercept_for_msr(vmx);
2116 		break;
2117 	case MSR_IA32_RTIT_STATUS:
2118 		if (!pt_can_write_msr(vmx))
2119 			return 1;
2120 		if (data & MSR_IA32_RTIT_STATUS_MASK)
2121 			return 1;
2122 		vmx->pt_desc.guest.status = data;
2123 		break;
2124 	case MSR_IA32_RTIT_CR3_MATCH:
2125 		if (!pt_can_write_msr(vmx))
2126 			return 1;
2127 		if (!intel_pt_validate_cap(vmx->pt_desc.caps,
2128 					   PT_CAP_cr3_filtering))
2129 			return 1;
2130 		vmx->pt_desc.guest.cr3_match = data;
2131 		break;
2132 	case MSR_IA32_RTIT_OUTPUT_BASE:
2133 		if (!pt_can_write_msr(vmx))
2134 			return 1;
2135 		if (!intel_pt_validate_cap(vmx->pt_desc.caps,
2136 					   PT_CAP_topa_output) &&
2137 		    !intel_pt_validate_cap(vmx->pt_desc.caps,
2138 					   PT_CAP_single_range_output))
2139 			return 1;
2140 		if (data & MSR_IA32_RTIT_OUTPUT_BASE_MASK)
2141 			return 1;
2142 		vmx->pt_desc.guest.output_base = data;
2143 		break;
2144 	case MSR_IA32_RTIT_OUTPUT_MASK:
2145 		if (!pt_can_write_msr(vmx))
2146 			return 1;
2147 		if (!intel_pt_validate_cap(vmx->pt_desc.caps,
2148 					   PT_CAP_topa_output) &&
2149 		    !intel_pt_validate_cap(vmx->pt_desc.caps,
2150 					   PT_CAP_single_range_output))
2151 			return 1;
2152 		vmx->pt_desc.guest.output_mask = data;
2153 		break;
2154 	case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
2155 		if (!pt_can_write_msr(vmx))
2156 			return 1;
2157 		index = msr_info->index - MSR_IA32_RTIT_ADDR0_A;
2158 		if (index >= 2 * intel_pt_validate_cap(vmx->pt_desc.caps,
2159 						       PT_CAP_num_address_ranges))
2160 			return 1;
2161 		if (is_noncanonical_address(data, vcpu))
2162 			return 1;
2163 		if (index % 2)
2164 			vmx->pt_desc.guest.addr_b[index / 2] = data;
2165 		else
2166 			vmx->pt_desc.guest.addr_a[index / 2] = data;
2167 		break;
2168 	case MSR_TSC_AUX:
2169 		if (!msr_info->host_initiated &&
2170 		    !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP))
2171 			return 1;
2172 		/* Check reserved bit, higher 32 bits should be zero */
2173 		if ((data >> 32) != 0)
2174 			return 1;
2175 		goto find_shared_msr;
2176 
2177 	default:
2178 	find_shared_msr:
2179 		msr = find_msr_entry(vmx, msr_index);
2180 		if (msr)
2181 			ret = vmx_set_guest_msr(vmx, msr, data);
2182 		else
2183 			ret = kvm_set_msr_common(vcpu, msr_info);
2184 	}
2185 
2186 	return ret;
2187 }
2188 
2189 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
2190 {
2191 	kvm_register_mark_available(vcpu, reg);
2192 
2193 	switch (reg) {
2194 	case VCPU_REGS_RSP:
2195 		vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
2196 		break;
2197 	case VCPU_REGS_RIP:
2198 		vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
2199 		break;
2200 	case VCPU_EXREG_PDPTR:
2201 		if (enable_ept)
2202 			ept_save_pdptrs(vcpu);
2203 		break;
2204 	case VCPU_EXREG_CR3:
2205 		if (enable_unrestricted_guest || (enable_ept && is_paging(vcpu)))
2206 			vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
2207 		break;
2208 	default:
2209 		WARN_ON_ONCE(1);
2210 		break;
2211 	}
2212 }
2213 
2214 static __init int cpu_has_kvm_support(void)
2215 {
2216 	return cpu_has_vmx();
2217 }
2218 
2219 static __init int vmx_disabled_by_bios(void)
2220 {
2221 	return !boot_cpu_has(X86_FEATURE_MSR_IA32_FEAT_CTL) ||
2222 	       !boot_cpu_has(X86_FEATURE_VMX);
2223 }
2224 
2225 static int kvm_cpu_vmxon(u64 vmxon_pointer)
2226 {
2227 	u64 msr;
2228 
2229 	cr4_set_bits(X86_CR4_VMXE);
2230 	intel_pt_handle_vmx(1);
2231 
2232 	asm_volatile_goto("1: vmxon %[vmxon_pointer]\n\t"
2233 			  _ASM_EXTABLE(1b, %l[fault])
2234 			  : : [vmxon_pointer] "m"(vmxon_pointer)
2235 			  : : fault);
2236 	return 0;
2237 
2238 fault:
2239 	WARN_ONCE(1, "VMXON faulted, MSR_IA32_FEAT_CTL (0x3a) = 0x%llx\n",
2240 		  rdmsrl_safe(MSR_IA32_FEAT_CTL, &msr) ? 0xdeadbeef : msr);
2241 	intel_pt_handle_vmx(0);
2242 	cr4_clear_bits(X86_CR4_VMXE);
2243 
2244 	return -EFAULT;
2245 }
2246 
2247 static int hardware_enable(void)
2248 {
2249 	int cpu = raw_smp_processor_id();
2250 	u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
2251 	int r;
2252 
2253 	if (cr4_read_shadow() & X86_CR4_VMXE)
2254 		return -EBUSY;
2255 
2256 	/*
2257 	 * This can happen if we hot-added a CPU but failed to allocate
2258 	 * VP assist page for it.
2259 	 */
2260 	if (static_branch_unlikely(&enable_evmcs) &&
2261 	    !hv_get_vp_assist_page(cpu))
2262 		return -EFAULT;
2263 
2264 	INIT_LIST_HEAD(&per_cpu(loaded_vmcss_on_cpu, cpu));
2265 	INIT_LIST_HEAD(&per_cpu(blocked_vcpu_on_cpu, cpu));
2266 	spin_lock_init(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
2267 
2268 	r = kvm_cpu_vmxon(phys_addr);
2269 	if (r)
2270 		return r;
2271 
2272 	if (enable_ept)
2273 		ept_sync_global();
2274 
2275 	return 0;
2276 }
2277 
2278 static void vmclear_local_loaded_vmcss(void)
2279 {
2280 	int cpu = raw_smp_processor_id();
2281 	struct loaded_vmcs *v, *n;
2282 
2283 	list_for_each_entry_safe(v, n, &per_cpu(loaded_vmcss_on_cpu, cpu),
2284 				 loaded_vmcss_on_cpu_link)
2285 		__loaded_vmcs_clear(v);
2286 }
2287 
2288 
2289 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
2290  * tricks.
2291  */
2292 static void kvm_cpu_vmxoff(void)
2293 {
2294 	asm volatile (__ex("vmxoff"));
2295 
2296 	intel_pt_handle_vmx(0);
2297 	cr4_clear_bits(X86_CR4_VMXE);
2298 }
2299 
2300 static void hardware_disable(void)
2301 {
2302 	vmclear_local_loaded_vmcss();
2303 	kvm_cpu_vmxoff();
2304 }
2305 
2306 /*
2307  * There is no X86_FEATURE for SGX yet, but anyway we need to query CPUID
2308  * directly instead of going through cpu_has(), to ensure KVM is trapping
2309  * ENCLS whenever it's supported in hardware.  It does not matter whether
2310  * the host OS supports or has enabled SGX.
2311  */
2312 static bool cpu_has_sgx(void)
2313 {
2314 	return cpuid_eax(0) >= 0x12 && (cpuid_eax(0x12) & BIT(0));
2315 }
2316 
2317 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
2318 				      u32 msr, u32 *result)
2319 {
2320 	u32 vmx_msr_low, vmx_msr_high;
2321 	u32 ctl = ctl_min | ctl_opt;
2322 
2323 	rdmsr(msr, vmx_msr_low, vmx_msr_high);
2324 
2325 	ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
2326 	ctl |= vmx_msr_low;  /* bit == 1 in low word  ==> must be one  */
2327 
2328 	/* Ensure minimum (required) set of control bits are supported. */
2329 	if (ctl_min & ~ctl)
2330 		return -EIO;
2331 
2332 	*result = ctl;
2333 	return 0;
2334 }
2335 
2336 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf,
2337 				    struct vmx_capability *vmx_cap)
2338 {
2339 	u32 vmx_msr_low, vmx_msr_high;
2340 	u32 min, opt, min2, opt2;
2341 	u32 _pin_based_exec_control = 0;
2342 	u32 _cpu_based_exec_control = 0;
2343 	u32 _cpu_based_2nd_exec_control = 0;
2344 	u32 _vmexit_control = 0;
2345 	u32 _vmentry_control = 0;
2346 
2347 	memset(vmcs_conf, 0, sizeof(*vmcs_conf));
2348 	min = CPU_BASED_HLT_EXITING |
2349 #ifdef CONFIG_X86_64
2350 	      CPU_BASED_CR8_LOAD_EXITING |
2351 	      CPU_BASED_CR8_STORE_EXITING |
2352 #endif
2353 	      CPU_BASED_CR3_LOAD_EXITING |
2354 	      CPU_BASED_CR3_STORE_EXITING |
2355 	      CPU_BASED_UNCOND_IO_EXITING |
2356 	      CPU_BASED_MOV_DR_EXITING |
2357 	      CPU_BASED_USE_TSC_OFFSETTING |
2358 	      CPU_BASED_MWAIT_EXITING |
2359 	      CPU_BASED_MONITOR_EXITING |
2360 	      CPU_BASED_INVLPG_EXITING |
2361 	      CPU_BASED_RDPMC_EXITING;
2362 
2363 	opt = CPU_BASED_TPR_SHADOW |
2364 	      CPU_BASED_USE_MSR_BITMAPS |
2365 	      CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
2366 	if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
2367 				&_cpu_based_exec_control) < 0)
2368 		return -EIO;
2369 #ifdef CONFIG_X86_64
2370 	if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2371 		_cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
2372 					   ~CPU_BASED_CR8_STORE_EXITING;
2373 #endif
2374 	if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
2375 		min2 = 0;
2376 		opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2377 			SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2378 			SECONDARY_EXEC_WBINVD_EXITING |
2379 			SECONDARY_EXEC_ENABLE_VPID |
2380 			SECONDARY_EXEC_ENABLE_EPT |
2381 			SECONDARY_EXEC_UNRESTRICTED_GUEST |
2382 			SECONDARY_EXEC_PAUSE_LOOP_EXITING |
2383 			SECONDARY_EXEC_DESC |
2384 			SECONDARY_EXEC_RDTSCP |
2385 			SECONDARY_EXEC_ENABLE_INVPCID |
2386 			SECONDARY_EXEC_APIC_REGISTER_VIRT |
2387 			SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
2388 			SECONDARY_EXEC_SHADOW_VMCS |
2389 			SECONDARY_EXEC_XSAVES |
2390 			SECONDARY_EXEC_RDSEED_EXITING |
2391 			SECONDARY_EXEC_RDRAND_EXITING |
2392 			SECONDARY_EXEC_ENABLE_PML |
2393 			SECONDARY_EXEC_TSC_SCALING |
2394 			SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE |
2395 			SECONDARY_EXEC_PT_USE_GPA |
2396 			SECONDARY_EXEC_PT_CONCEAL_VMX |
2397 			SECONDARY_EXEC_ENABLE_VMFUNC;
2398 		if (cpu_has_sgx())
2399 			opt2 |= SECONDARY_EXEC_ENCLS_EXITING;
2400 		if (adjust_vmx_controls(min2, opt2,
2401 					MSR_IA32_VMX_PROCBASED_CTLS2,
2402 					&_cpu_based_2nd_exec_control) < 0)
2403 			return -EIO;
2404 	}
2405 #ifndef CONFIG_X86_64
2406 	if (!(_cpu_based_2nd_exec_control &
2407 				SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
2408 		_cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
2409 #endif
2410 
2411 	if (!(_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2412 		_cpu_based_2nd_exec_control &= ~(
2413 				SECONDARY_EXEC_APIC_REGISTER_VIRT |
2414 				SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2415 				SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
2416 
2417 	rdmsr_safe(MSR_IA32_VMX_EPT_VPID_CAP,
2418 		&vmx_cap->ept, &vmx_cap->vpid);
2419 
2420 	if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
2421 		/* CR3 accesses and invlpg don't need to cause VM Exits when EPT
2422 		   enabled */
2423 		_cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
2424 					     CPU_BASED_CR3_STORE_EXITING |
2425 					     CPU_BASED_INVLPG_EXITING);
2426 	} else if (vmx_cap->ept) {
2427 		vmx_cap->ept = 0;
2428 		pr_warn_once("EPT CAP should not exist if not support "
2429 				"1-setting enable EPT VM-execution control\n");
2430 	}
2431 	if (!(_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_VPID) &&
2432 		vmx_cap->vpid) {
2433 		vmx_cap->vpid = 0;
2434 		pr_warn_once("VPID CAP should not exist if not support "
2435 				"1-setting enable VPID VM-execution control\n");
2436 	}
2437 
2438 	min = VM_EXIT_SAVE_DEBUG_CONTROLS | VM_EXIT_ACK_INTR_ON_EXIT;
2439 #ifdef CONFIG_X86_64
2440 	min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
2441 #endif
2442 	opt = VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL |
2443 	      VM_EXIT_LOAD_IA32_PAT |
2444 	      VM_EXIT_LOAD_IA32_EFER |
2445 	      VM_EXIT_CLEAR_BNDCFGS |
2446 	      VM_EXIT_PT_CONCEAL_PIP |
2447 	      VM_EXIT_CLEAR_IA32_RTIT_CTL;
2448 	if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
2449 				&_vmexit_control) < 0)
2450 		return -EIO;
2451 
2452 	min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
2453 	opt = PIN_BASED_VIRTUAL_NMIS | PIN_BASED_POSTED_INTR |
2454 		 PIN_BASED_VMX_PREEMPTION_TIMER;
2455 	if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
2456 				&_pin_based_exec_control) < 0)
2457 		return -EIO;
2458 
2459 	if (cpu_has_broken_vmx_preemption_timer())
2460 		_pin_based_exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
2461 	if (!(_cpu_based_2nd_exec_control &
2462 		SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY))
2463 		_pin_based_exec_control &= ~PIN_BASED_POSTED_INTR;
2464 
2465 	min = VM_ENTRY_LOAD_DEBUG_CONTROLS;
2466 	opt = VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL |
2467 	      VM_ENTRY_LOAD_IA32_PAT |
2468 	      VM_ENTRY_LOAD_IA32_EFER |
2469 	      VM_ENTRY_LOAD_BNDCFGS |
2470 	      VM_ENTRY_PT_CONCEAL_PIP |
2471 	      VM_ENTRY_LOAD_IA32_RTIT_CTL;
2472 	if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
2473 				&_vmentry_control) < 0)
2474 		return -EIO;
2475 
2476 	/*
2477 	 * Some cpus support VM_{ENTRY,EXIT}_IA32_PERF_GLOBAL_CTRL but they
2478 	 * can't be used due to an errata where VM Exit may incorrectly clear
2479 	 * IA32_PERF_GLOBAL_CTRL[34:32].  Workaround the errata by using the
2480 	 * MSR load mechanism to switch IA32_PERF_GLOBAL_CTRL.
2481 	 */
2482 	if (boot_cpu_data.x86 == 0x6) {
2483 		switch (boot_cpu_data.x86_model) {
2484 		case 26: /* AAK155 */
2485 		case 30: /* AAP115 */
2486 		case 37: /* AAT100 */
2487 		case 44: /* BC86,AAY89,BD102 */
2488 		case 46: /* BA97 */
2489 			_vmentry_control &= ~VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL;
2490 			_vmexit_control &= ~VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL;
2491 			pr_warn_once("kvm: VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL "
2492 					"does not work properly. Using workaround\n");
2493 			break;
2494 		default:
2495 			break;
2496 		}
2497 	}
2498 
2499 
2500 	rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
2501 
2502 	/* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
2503 	if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
2504 		return -EIO;
2505 
2506 #ifdef CONFIG_X86_64
2507 	/* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
2508 	if (vmx_msr_high & (1u<<16))
2509 		return -EIO;
2510 #endif
2511 
2512 	/* Require Write-Back (WB) memory type for VMCS accesses. */
2513 	if (((vmx_msr_high >> 18) & 15) != 6)
2514 		return -EIO;
2515 
2516 	vmcs_conf->size = vmx_msr_high & 0x1fff;
2517 	vmcs_conf->order = get_order(vmcs_conf->size);
2518 	vmcs_conf->basic_cap = vmx_msr_high & ~0x1fff;
2519 
2520 	vmcs_conf->revision_id = vmx_msr_low;
2521 
2522 	vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
2523 	vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
2524 	vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
2525 	vmcs_conf->vmexit_ctrl         = _vmexit_control;
2526 	vmcs_conf->vmentry_ctrl        = _vmentry_control;
2527 
2528 	if (static_branch_unlikely(&enable_evmcs))
2529 		evmcs_sanitize_exec_ctrls(vmcs_conf);
2530 
2531 	return 0;
2532 }
2533 
2534 struct vmcs *alloc_vmcs_cpu(bool shadow, int cpu, gfp_t flags)
2535 {
2536 	int node = cpu_to_node(cpu);
2537 	struct page *pages;
2538 	struct vmcs *vmcs;
2539 
2540 	pages = __alloc_pages_node(node, flags, vmcs_config.order);
2541 	if (!pages)
2542 		return NULL;
2543 	vmcs = page_address(pages);
2544 	memset(vmcs, 0, vmcs_config.size);
2545 
2546 	/* KVM supports Enlightened VMCS v1 only */
2547 	if (static_branch_unlikely(&enable_evmcs))
2548 		vmcs->hdr.revision_id = KVM_EVMCS_VERSION;
2549 	else
2550 		vmcs->hdr.revision_id = vmcs_config.revision_id;
2551 
2552 	if (shadow)
2553 		vmcs->hdr.shadow_vmcs = 1;
2554 	return vmcs;
2555 }
2556 
2557 void free_vmcs(struct vmcs *vmcs)
2558 {
2559 	free_pages((unsigned long)vmcs, vmcs_config.order);
2560 }
2561 
2562 /*
2563  * Free a VMCS, but before that VMCLEAR it on the CPU where it was last loaded
2564  */
2565 void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
2566 {
2567 	if (!loaded_vmcs->vmcs)
2568 		return;
2569 	loaded_vmcs_clear(loaded_vmcs);
2570 	free_vmcs(loaded_vmcs->vmcs);
2571 	loaded_vmcs->vmcs = NULL;
2572 	if (loaded_vmcs->msr_bitmap)
2573 		free_page((unsigned long)loaded_vmcs->msr_bitmap);
2574 	WARN_ON(loaded_vmcs->shadow_vmcs != NULL);
2575 }
2576 
2577 int alloc_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
2578 {
2579 	loaded_vmcs->vmcs = alloc_vmcs(false);
2580 	if (!loaded_vmcs->vmcs)
2581 		return -ENOMEM;
2582 
2583 	vmcs_clear(loaded_vmcs->vmcs);
2584 
2585 	loaded_vmcs->shadow_vmcs = NULL;
2586 	loaded_vmcs->hv_timer_soft_disabled = false;
2587 	loaded_vmcs->cpu = -1;
2588 	loaded_vmcs->launched = 0;
2589 
2590 	if (cpu_has_vmx_msr_bitmap()) {
2591 		loaded_vmcs->msr_bitmap = (unsigned long *)
2592 				__get_free_page(GFP_KERNEL_ACCOUNT);
2593 		if (!loaded_vmcs->msr_bitmap)
2594 			goto out_vmcs;
2595 		memset(loaded_vmcs->msr_bitmap, 0xff, PAGE_SIZE);
2596 
2597 		if (IS_ENABLED(CONFIG_HYPERV) &&
2598 		    static_branch_unlikely(&enable_evmcs) &&
2599 		    (ms_hyperv.nested_features & HV_X64_NESTED_MSR_BITMAP)) {
2600 			struct hv_enlightened_vmcs *evmcs =
2601 				(struct hv_enlightened_vmcs *)loaded_vmcs->vmcs;
2602 
2603 			evmcs->hv_enlightenments_control.msr_bitmap = 1;
2604 		}
2605 	}
2606 
2607 	memset(&loaded_vmcs->host_state, 0, sizeof(struct vmcs_host_state));
2608 	memset(&loaded_vmcs->controls_shadow, 0,
2609 		sizeof(struct vmcs_controls_shadow));
2610 
2611 	return 0;
2612 
2613 out_vmcs:
2614 	free_loaded_vmcs(loaded_vmcs);
2615 	return -ENOMEM;
2616 }
2617 
2618 static void free_kvm_area(void)
2619 {
2620 	int cpu;
2621 
2622 	for_each_possible_cpu(cpu) {
2623 		free_vmcs(per_cpu(vmxarea, cpu));
2624 		per_cpu(vmxarea, cpu) = NULL;
2625 	}
2626 }
2627 
2628 static __init int alloc_kvm_area(void)
2629 {
2630 	int cpu;
2631 
2632 	for_each_possible_cpu(cpu) {
2633 		struct vmcs *vmcs;
2634 
2635 		vmcs = alloc_vmcs_cpu(false, cpu, GFP_KERNEL);
2636 		if (!vmcs) {
2637 			free_kvm_area();
2638 			return -ENOMEM;
2639 		}
2640 
2641 		/*
2642 		 * When eVMCS is enabled, alloc_vmcs_cpu() sets
2643 		 * vmcs->revision_id to KVM_EVMCS_VERSION instead of
2644 		 * revision_id reported by MSR_IA32_VMX_BASIC.
2645 		 *
2646 		 * However, even though not explicitly documented by
2647 		 * TLFS, VMXArea passed as VMXON argument should
2648 		 * still be marked with revision_id reported by
2649 		 * physical CPU.
2650 		 */
2651 		if (static_branch_unlikely(&enable_evmcs))
2652 			vmcs->hdr.revision_id = vmcs_config.revision_id;
2653 
2654 		per_cpu(vmxarea, cpu) = vmcs;
2655 	}
2656 	return 0;
2657 }
2658 
2659 static void fix_pmode_seg(struct kvm_vcpu *vcpu, int seg,
2660 		struct kvm_segment *save)
2661 {
2662 	if (!emulate_invalid_guest_state) {
2663 		/*
2664 		 * CS and SS RPL should be equal during guest entry according
2665 		 * to VMX spec, but in reality it is not always so. Since vcpu
2666 		 * is in the middle of the transition from real mode to
2667 		 * protected mode it is safe to assume that RPL 0 is a good
2668 		 * default value.
2669 		 */
2670 		if (seg == VCPU_SREG_CS || seg == VCPU_SREG_SS)
2671 			save->selector &= ~SEGMENT_RPL_MASK;
2672 		save->dpl = save->selector & SEGMENT_RPL_MASK;
2673 		save->s = 1;
2674 	}
2675 	vmx_set_segment(vcpu, save, seg);
2676 }
2677 
2678 static void enter_pmode(struct kvm_vcpu *vcpu)
2679 {
2680 	unsigned long flags;
2681 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2682 
2683 	/*
2684 	 * Update real mode segment cache. It may be not up-to-date if sement
2685 	 * register was written while vcpu was in a guest mode.
2686 	 */
2687 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
2688 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
2689 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
2690 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
2691 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
2692 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
2693 
2694 	vmx->rmode.vm86_active = 0;
2695 
2696 	vmx_set_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
2697 
2698 	flags = vmcs_readl(GUEST_RFLAGS);
2699 	flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
2700 	flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
2701 	vmcs_writel(GUEST_RFLAGS, flags);
2702 
2703 	vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
2704 			(vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
2705 
2706 	update_exception_bitmap(vcpu);
2707 
2708 	fix_pmode_seg(vcpu, VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
2709 	fix_pmode_seg(vcpu, VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
2710 	fix_pmode_seg(vcpu, VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
2711 	fix_pmode_seg(vcpu, VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
2712 	fix_pmode_seg(vcpu, VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
2713 	fix_pmode_seg(vcpu, VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
2714 }
2715 
2716 static void fix_rmode_seg(int seg, struct kvm_segment *save)
2717 {
2718 	const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2719 	struct kvm_segment var = *save;
2720 
2721 	var.dpl = 0x3;
2722 	if (seg == VCPU_SREG_CS)
2723 		var.type = 0x3;
2724 
2725 	if (!emulate_invalid_guest_state) {
2726 		var.selector = var.base >> 4;
2727 		var.base = var.base & 0xffff0;
2728 		var.limit = 0xffff;
2729 		var.g = 0;
2730 		var.db = 0;
2731 		var.present = 1;
2732 		var.s = 1;
2733 		var.l = 0;
2734 		var.unusable = 0;
2735 		var.type = 0x3;
2736 		var.avl = 0;
2737 		if (save->base & 0xf)
2738 			printk_once(KERN_WARNING "kvm: segment base is not "
2739 					"paragraph aligned when entering "
2740 					"protected mode (seg=%d)", seg);
2741 	}
2742 
2743 	vmcs_write16(sf->selector, var.selector);
2744 	vmcs_writel(sf->base, var.base);
2745 	vmcs_write32(sf->limit, var.limit);
2746 	vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(&var));
2747 }
2748 
2749 static void enter_rmode(struct kvm_vcpu *vcpu)
2750 {
2751 	unsigned long flags;
2752 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2753 	struct kvm_vmx *kvm_vmx = to_kvm_vmx(vcpu->kvm);
2754 
2755 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
2756 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
2757 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
2758 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
2759 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
2760 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
2761 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
2762 
2763 	vmx->rmode.vm86_active = 1;
2764 
2765 	/*
2766 	 * Very old userspace does not call KVM_SET_TSS_ADDR before entering
2767 	 * vcpu. Warn the user that an update is overdue.
2768 	 */
2769 	if (!kvm_vmx->tss_addr)
2770 		printk_once(KERN_WARNING "kvm: KVM_SET_TSS_ADDR need to be "
2771 			     "called before entering vcpu\n");
2772 
2773 	vmx_segment_cache_clear(vmx);
2774 
2775 	vmcs_writel(GUEST_TR_BASE, kvm_vmx->tss_addr);
2776 	vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
2777 	vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
2778 
2779 	flags = vmcs_readl(GUEST_RFLAGS);
2780 	vmx->rmode.save_rflags = flags;
2781 
2782 	flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
2783 
2784 	vmcs_writel(GUEST_RFLAGS, flags);
2785 	vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
2786 	update_exception_bitmap(vcpu);
2787 
2788 	fix_rmode_seg(VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
2789 	fix_rmode_seg(VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
2790 	fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
2791 	fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
2792 	fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
2793 	fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
2794 
2795 	kvm_mmu_reset_context(vcpu);
2796 }
2797 
2798 void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
2799 {
2800 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2801 	struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
2802 
2803 	if (!msr)
2804 		return;
2805 
2806 	vcpu->arch.efer = efer;
2807 	if (efer & EFER_LMA) {
2808 		vm_entry_controls_setbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
2809 		msr->data = efer;
2810 	} else {
2811 		vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
2812 
2813 		msr->data = efer & ~EFER_LME;
2814 	}
2815 	setup_msrs(vmx);
2816 }
2817 
2818 #ifdef CONFIG_X86_64
2819 
2820 static void enter_lmode(struct kvm_vcpu *vcpu)
2821 {
2822 	u32 guest_tr_ar;
2823 
2824 	vmx_segment_cache_clear(to_vmx(vcpu));
2825 
2826 	guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
2827 	if ((guest_tr_ar & VMX_AR_TYPE_MASK) != VMX_AR_TYPE_BUSY_64_TSS) {
2828 		pr_debug_ratelimited("%s: tss fixup for long mode. \n",
2829 				     __func__);
2830 		vmcs_write32(GUEST_TR_AR_BYTES,
2831 			     (guest_tr_ar & ~VMX_AR_TYPE_MASK)
2832 			     | VMX_AR_TYPE_BUSY_64_TSS);
2833 	}
2834 	vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
2835 }
2836 
2837 static void exit_lmode(struct kvm_vcpu *vcpu)
2838 {
2839 	vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
2840 	vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
2841 }
2842 
2843 #endif
2844 
2845 static void vmx_flush_tlb_gva(struct kvm_vcpu *vcpu, gva_t addr)
2846 {
2847 	int vpid = to_vmx(vcpu)->vpid;
2848 
2849 	if (!vpid_sync_vcpu_addr(vpid, addr))
2850 		vpid_sync_context(vpid);
2851 
2852 	/*
2853 	 * If VPIDs are not supported or enabled, then the above is a no-op.
2854 	 * But we don't really need a TLB flush in that case anyway, because
2855 	 * each VM entry/exit includes an implicit flush when VPID is 0.
2856 	 */
2857 }
2858 
2859 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
2860 {
2861 	ulong cr0_guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
2862 
2863 	vcpu->arch.cr0 &= ~cr0_guest_owned_bits;
2864 	vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & cr0_guest_owned_bits;
2865 }
2866 
2867 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
2868 {
2869 	ulong cr4_guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
2870 
2871 	vcpu->arch.cr4 &= ~cr4_guest_owned_bits;
2872 	vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & cr4_guest_owned_bits;
2873 }
2874 
2875 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
2876 {
2877 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
2878 
2879 	if (!kvm_register_is_dirty(vcpu, VCPU_EXREG_PDPTR))
2880 		return;
2881 
2882 	if (is_pae_paging(vcpu)) {
2883 		vmcs_write64(GUEST_PDPTR0, mmu->pdptrs[0]);
2884 		vmcs_write64(GUEST_PDPTR1, mmu->pdptrs[1]);
2885 		vmcs_write64(GUEST_PDPTR2, mmu->pdptrs[2]);
2886 		vmcs_write64(GUEST_PDPTR3, mmu->pdptrs[3]);
2887 	}
2888 }
2889 
2890 void ept_save_pdptrs(struct kvm_vcpu *vcpu)
2891 {
2892 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
2893 
2894 	if (is_pae_paging(vcpu)) {
2895 		mmu->pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
2896 		mmu->pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
2897 		mmu->pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
2898 		mmu->pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
2899 	}
2900 
2901 	kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
2902 }
2903 
2904 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
2905 					unsigned long cr0,
2906 					struct kvm_vcpu *vcpu)
2907 {
2908 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2909 
2910 	if (!kvm_register_is_available(vcpu, VCPU_EXREG_CR3))
2911 		vmx_cache_reg(vcpu, VCPU_EXREG_CR3);
2912 	if (!(cr0 & X86_CR0_PG)) {
2913 		/* From paging/starting to nonpaging */
2914 		exec_controls_setbit(vmx, CPU_BASED_CR3_LOAD_EXITING |
2915 					  CPU_BASED_CR3_STORE_EXITING);
2916 		vcpu->arch.cr0 = cr0;
2917 		vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
2918 	} else if (!is_paging(vcpu)) {
2919 		/* From nonpaging to paging */
2920 		exec_controls_clearbit(vmx, CPU_BASED_CR3_LOAD_EXITING |
2921 					    CPU_BASED_CR3_STORE_EXITING);
2922 		vcpu->arch.cr0 = cr0;
2923 		vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
2924 	}
2925 
2926 	if (!(cr0 & X86_CR0_WP))
2927 		*hw_cr0 &= ~X86_CR0_WP;
2928 }
2929 
2930 void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
2931 {
2932 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2933 	unsigned long hw_cr0;
2934 
2935 	hw_cr0 = (cr0 & ~KVM_VM_CR0_ALWAYS_OFF);
2936 	if (enable_unrestricted_guest)
2937 		hw_cr0 |= KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
2938 	else {
2939 		hw_cr0 |= KVM_VM_CR0_ALWAYS_ON;
2940 
2941 		if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
2942 			enter_pmode(vcpu);
2943 
2944 		if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
2945 			enter_rmode(vcpu);
2946 	}
2947 
2948 #ifdef CONFIG_X86_64
2949 	if (vcpu->arch.efer & EFER_LME) {
2950 		if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
2951 			enter_lmode(vcpu);
2952 		if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
2953 			exit_lmode(vcpu);
2954 	}
2955 #endif
2956 
2957 	if (enable_ept && !enable_unrestricted_guest)
2958 		ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
2959 
2960 	vmcs_writel(CR0_READ_SHADOW, cr0);
2961 	vmcs_writel(GUEST_CR0, hw_cr0);
2962 	vcpu->arch.cr0 = cr0;
2963 
2964 	/* depends on vcpu->arch.cr0 to be set to a new value */
2965 	vmx->emulation_required = emulation_required(vcpu);
2966 }
2967 
2968 static int get_ept_level(struct kvm_vcpu *vcpu)
2969 {
2970 	if (is_guest_mode(vcpu) && nested_cpu_has_ept(get_vmcs12(vcpu)))
2971 		return vmx_eptp_page_walk_level(nested_ept_get_eptp(vcpu));
2972 	if (cpu_has_vmx_ept_5levels() && (cpuid_maxphyaddr(vcpu) > 48))
2973 		return 5;
2974 	return 4;
2975 }
2976 
2977 u64 construct_eptp(struct kvm_vcpu *vcpu, unsigned long root_hpa)
2978 {
2979 	u64 eptp = VMX_EPTP_MT_WB;
2980 
2981 	eptp |= (get_ept_level(vcpu) == 5) ? VMX_EPTP_PWL_5 : VMX_EPTP_PWL_4;
2982 
2983 	if (enable_ept_ad_bits &&
2984 	    (!is_guest_mode(vcpu) || nested_ept_ad_enabled(vcpu)))
2985 		eptp |= VMX_EPTP_AD_ENABLE_BIT;
2986 	eptp |= (root_hpa & PAGE_MASK);
2987 
2988 	return eptp;
2989 }
2990 
2991 void vmx_load_mmu_pgd(struct kvm_vcpu *vcpu, unsigned long cr3)
2992 {
2993 	struct kvm *kvm = vcpu->kvm;
2994 	bool update_guest_cr3 = true;
2995 	unsigned long guest_cr3;
2996 	u64 eptp;
2997 
2998 	guest_cr3 = cr3;
2999 	if (enable_ept) {
3000 		eptp = construct_eptp(vcpu, cr3);
3001 		vmcs_write64(EPT_POINTER, eptp);
3002 
3003 		if (kvm_x86_ops.tlb_remote_flush) {
3004 			spin_lock(&to_kvm_vmx(kvm)->ept_pointer_lock);
3005 			to_vmx(vcpu)->ept_pointer = eptp;
3006 			to_kvm_vmx(kvm)->ept_pointers_match
3007 				= EPT_POINTERS_CHECK;
3008 			spin_unlock(&to_kvm_vmx(kvm)->ept_pointer_lock);
3009 		}
3010 
3011 		/* Loading vmcs02.GUEST_CR3 is handled by nested VM-Enter. */
3012 		if (is_guest_mode(vcpu))
3013 			update_guest_cr3 = false;
3014 		else if (!enable_unrestricted_guest && !is_paging(vcpu))
3015 			guest_cr3 = to_kvm_vmx(kvm)->ept_identity_map_addr;
3016 		else if (test_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail))
3017 			guest_cr3 = vcpu->arch.cr3;
3018 		else /* vmcs01.GUEST_CR3 is already up-to-date. */
3019 			update_guest_cr3 = false;
3020 		ept_load_pdptrs(vcpu);
3021 	}
3022 
3023 	if (update_guest_cr3)
3024 		vmcs_writel(GUEST_CR3, guest_cr3);
3025 }
3026 
3027 int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
3028 {
3029 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3030 	/*
3031 	 * Pass through host's Machine Check Enable value to hw_cr4, which
3032 	 * is in force while we are in guest mode.  Do not let guests control
3033 	 * this bit, even if host CR4.MCE == 0.
3034 	 */
3035 	unsigned long hw_cr4;
3036 
3037 	hw_cr4 = (cr4_read_shadow() & X86_CR4_MCE) | (cr4 & ~X86_CR4_MCE);
3038 	if (enable_unrestricted_guest)
3039 		hw_cr4 |= KVM_VM_CR4_ALWAYS_ON_UNRESTRICTED_GUEST;
3040 	else if (vmx->rmode.vm86_active)
3041 		hw_cr4 |= KVM_RMODE_VM_CR4_ALWAYS_ON;
3042 	else
3043 		hw_cr4 |= KVM_PMODE_VM_CR4_ALWAYS_ON;
3044 
3045 	if (!boot_cpu_has(X86_FEATURE_UMIP) && vmx_umip_emulated()) {
3046 		if (cr4 & X86_CR4_UMIP) {
3047 			secondary_exec_controls_setbit(vmx, SECONDARY_EXEC_DESC);
3048 			hw_cr4 &= ~X86_CR4_UMIP;
3049 		} else if (!is_guest_mode(vcpu) ||
3050 			!nested_cpu_has2(get_vmcs12(vcpu), SECONDARY_EXEC_DESC)) {
3051 			secondary_exec_controls_clearbit(vmx, SECONDARY_EXEC_DESC);
3052 		}
3053 	}
3054 
3055 	if (cr4 & X86_CR4_VMXE) {
3056 		/*
3057 		 * To use VMXON (and later other VMX instructions), a guest
3058 		 * must first be able to turn on cr4.VMXE (see handle_vmon()).
3059 		 * So basically the check on whether to allow nested VMX
3060 		 * is here.  We operate under the default treatment of SMM,
3061 		 * so VMX cannot be enabled under SMM.
3062 		 */
3063 		if (!nested_vmx_allowed(vcpu) || is_smm(vcpu))
3064 			return 1;
3065 	}
3066 
3067 	if (vmx->nested.vmxon && !nested_cr4_valid(vcpu, cr4))
3068 		return 1;
3069 
3070 	vcpu->arch.cr4 = cr4;
3071 
3072 	if (!enable_unrestricted_guest) {
3073 		if (enable_ept) {
3074 			if (!is_paging(vcpu)) {
3075 				hw_cr4 &= ~X86_CR4_PAE;
3076 				hw_cr4 |= X86_CR4_PSE;
3077 			} else if (!(cr4 & X86_CR4_PAE)) {
3078 				hw_cr4 &= ~X86_CR4_PAE;
3079 			}
3080 		}
3081 
3082 		/*
3083 		 * SMEP/SMAP/PKU is disabled if CPU is in non-paging mode in
3084 		 * hardware.  To emulate this behavior, SMEP/SMAP/PKU needs
3085 		 * to be manually disabled when guest switches to non-paging
3086 		 * mode.
3087 		 *
3088 		 * If !enable_unrestricted_guest, the CPU is always running
3089 		 * with CR0.PG=1 and CR4 needs to be modified.
3090 		 * If enable_unrestricted_guest, the CPU automatically
3091 		 * disables SMEP/SMAP/PKU when the guest sets CR0.PG=0.
3092 		 */
3093 		if (!is_paging(vcpu))
3094 			hw_cr4 &= ~(X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE);
3095 	}
3096 
3097 	vmcs_writel(CR4_READ_SHADOW, cr4);
3098 	vmcs_writel(GUEST_CR4, hw_cr4);
3099 	return 0;
3100 }
3101 
3102 void vmx_get_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg)
3103 {
3104 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3105 	u32 ar;
3106 
3107 	if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3108 		*var = vmx->rmode.segs[seg];
3109 		if (seg == VCPU_SREG_TR
3110 		    || var->selector == vmx_read_guest_seg_selector(vmx, seg))
3111 			return;
3112 		var->base = vmx_read_guest_seg_base(vmx, seg);
3113 		var->selector = vmx_read_guest_seg_selector(vmx, seg);
3114 		return;
3115 	}
3116 	var->base = vmx_read_guest_seg_base(vmx, seg);
3117 	var->limit = vmx_read_guest_seg_limit(vmx, seg);
3118 	var->selector = vmx_read_guest_seg_selector(vmx, seg);
3119 	ar = vmx_read_guest_seg_ar(vmx, seg);
3120 	var->unusable = (ar >> 16) & 1;
3121 	var->type = ar & 15;
3122 	var->s = (ar >> 4) & 1;
3123 	var->dpl = (ar >> 5) & 3;
3124 	/*
3125 	 * Some userspaces do not preserve unusable property. Since usable
3126 	 * segment has to be present according to VMX spec we can use present
3127 	 * property to amend userspace bug by making unusable segment always
3128 	 * nonpresent. vmx_segment_access_rights() already marks nonpresent
3129 	 * segment as unusable.
3130 	 */
3131 	var->present = !var->unusable;
3132 	var->avl = (ar >> 12) & 1;
3133 	var->l = (ar >> 13) & 1;
3134 	var->db = (ar >> 14) & 1;
3135 	var->g = (ar >> 15) & 1;
3136 }
3137 
3138 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
3139 {
3140 	struct kvm_segment s;
3141 
3142 	if (to_vmx(vcpu)->rmode.vm86_active) {
3143 		vmx_get_segment(vcpu, &s, seg);
3144 		return s.base;
3145 	}
3146 	return vmx_read_guest_seg_base(to_vmx(vcpu), seg);
3147 }
3148 
3149 int vmx_get_cpl(struct kvm_vcpu *vcpu)
3150 {
3151 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3152 
3153 	if (unlikely(vmx->rmode.vm86_active))
3154 		return 0;
3155 	else {
3156 		int ar = vmx_read_guest_seg_ar(vmx, VCPU_SREG_SS);
3157 		return VMX_AR_DPL(ar);
3158 	}
3159 }
3160 
3161 static u32 vmx_segment_access_rights(struct kvm_segment *var)
3162 {
3163 	u32 ar;
3164 
3165 	if (var->unusable || !var->present)
3166 		ar = 1 << 16;
3167 	else {
3168 		ar = var->type & 15;
3169 		ar |= (var->s & 1) << 4;
3170 		ar |= (var->dpl & 3) << 5;
3171 		ar |= (var->present & 1) << 7;
3172 		ar |= (var->avl & 1) << 12;
3173 		ar |= (var->l & 1) << 13;
3174 		ar |= (var->db & 1) << 14;
3175 		ar |= (var->g & 1) << 15;
3176 	}
3177 
3178 	return ar;
3179 }
3180 
3181 void vmx_set_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg)
3182 {
3183 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3184 	const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3185 
3186 	vmx_segment_cache_clear(vmx);
3187 
3188 	if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3189 		vmx->rmode.segs[seg] = *var;
3190 		if (seg == VCPU_SREG_TR)
3191 			vmcs_write16(sf->selector, var->selector);
3192 		else if (var->s)
3193 			fix_rmode_seg(seg, &vmx->rmode.segs[seg]);
3194 		goto out;
3195 	}
3196 
3197 	vmcs_writel(sf->base, var->base);
3198 	vmcs_write32(sf->limit, var->limit);
3199 	vmcs_write16(sf->selector, var->selector);
3200 
3201 	/*
3202 	 *   Fix the "Accessed" bit in AR field of segment registers for older
3203 	 * qemu binaries.
3204 	 *   IA32 arch specifies that at the time of processor reset the
3205 	 * "Accessed" bit in the AR field of segment registers is 1. And qemu
3206 	 * is setting it to 0 in the userland code. This causes invalid guest
3207 	 * state vmexit when "unrestricted guest" mode is turned on.
3208 	 *    Fix for this setup issue in cpu_reset is being pushed in the qemu
3209 	 * tree. Newer qemu binaries with that qemu fix would not need this
3210 	 * kvm hack.
3211 	 */
3212 	if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
3213 		var->type |= 0x1; /* Accessed */
3214 
3215 	vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(var));
3216 
3217 out:
3218 	vmx->emulation_required = emulation_required(vcpu);
3219 }
3220 
3221 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3222 {
3223 	u32 ar = vmx_read_guest_seg_ar(to_vmx(vcpu), VCPU_SREG_CS);
3224 
3225 	*db = (ar >> 14) & 1;
3226 	*l = (ar >> 13) & 1;
3227 }
3228 
3229 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3230 {
3231 	dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
3232 	dt->address = vmcs_readl(GUEST_IDTR_BASE);
3233 }
3234 
3235 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3236 {
3237 	vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
3238 	vmcs_writel(GUEST_IDTR_BASE, dt->address);
3239 }
3240 
3241 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3242 {
3243 	dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
3244 	dt->address = vmcs_readl(GUEST_GDTR_BASE);
3245 }
3246 
3247 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3248 {
3249 	vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
3250 	vmcs_writel(GUEST_GDTR_BASE, dt->address);
3251 }
3252 
3253 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
3254 {
3255 	struct kvm_segment var;
3256 	u32 ar;
3257 
3258 	vmx_get_segment(vcpu, &var, seg);
3259 	var.dpl = 0x3;
3260 	if (seg == VCPU_SREG_CS)
3261 		var.type = 0x3;
3262 	ar = vmx_segment_access_rights(&var);
3263 
3264 	if (var.base != (var.selector << 4))
3265 		return false;
3266 	if (var.limit != 0xffff)
3267 		return false;
3268 	if (ar != 0xf3)
3269 		return false;
3270 
3271 	return true;
3272 }
3273 
3274 static bool code_segment_valid(struct kvm_vcpu *vcpu)
3275 {
3276 	struct kvm_segment cs;
3277 	unsigned int cs_rpl;
3278 
3279 	vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3280 	cs_rpl = cs.selector & SEGMENT_RPL_MASK;
3281 
3282 	if (cs.unusable)
3283 		return false;
3284 	if (~cs.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_ACCESSES_MASK))
3285 		return false;
3286 	if (!cs.s)
3287 		return false;
3288 	if (cs.type & VMX_AR_TYPE_WRITEABLE_MASK) {
3289 		if (cs.dpl > cs_rpl)
3290 			return false;
3291 	} else {
3292 		if (cs.dpl != cs_rpl)
3293 			return false;
3294 	}
3295 	if (!cs.present)
3296 		return false;
3297 
3298 	/* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
3299 	return true;
3300 }
3301 
3302 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
3303 {
3304 	struct kvm_segment ss;
3305 	unsigned int ss_rpl;
3306 
3307 	vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3308 	ss_rpl = ss.selector & SEGMENT_RPL_MASK;
3309 
3310 	if (ss.unusable)
3311 		return true;
3312 	if (ss.type != 3 && ss.type != 7)
3313 		return false;
3314 	if (!ss.s)
3315 		return false;
3316 	if (ss.dpl != ss_rpl) /* DPL != RPL */
3317 		return false;
3318 	if (!ss.present)
3319 		return false;
3320 
3321 	return true;
3322 }
3323 
3324 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
3325 {
3326 	struct kvm_segment var;
3327 	unsigned int rpl;
3328 
3329 	vmx_get_segment(vcpu, &var, seg);
3330 	rpl = var.selector & SEGMENT_RPL_MASK;
3331 
3332 	if (var.unusable)
3333 		return true;
3334 	if (!var.s)
3335 		return false;
3336 	if (!var.present)
3337 		return false;
3338 	if (~var.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_WRITEABLE_MASK)) {
3339 		if (var.dpl < rpl) /* DPL < RPL */
3340 			return false;
3341 	}
3342 
3343 	/* TODO: Add other members to kvm_segment_field to allow checking for other access
3344 	 * rights flags
3345 	 */
3346 	return true;
3347 }
3348 
3349 static bool tr_valid(struct kvm_vcpu *vcpu)
3350 {
3351 	struct kvm_segment tr;
3352 
3353 	vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
3354 
3355 	if (tr.unusable)
3356 		return false;
3357 	if (tr.selector & SEGMENT_TI_MASK)	/* TI = 1 */
3358 		return false;
3359 	if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
3360 		return false;
3361 	if (!tr.present)
3362 		return false;
3363 
3364 	return true;
3365 }
3366 
3367 static bool ldtr_valid(struct kvm_vcpu *vcpu)
3368 {
3369 	struct kvm_segment ldtr;
3370 
3371 	vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
3372 
3373 	if (ldtr.unusable)
3374 		return true;
3375 	if (ldtr.selector & SEGMENT_TI_MASK)	/* TI = 1 */
3376 		return false;
3377 	if (ldtr.type != 2)
3378 		return false;
3379 	if (!ldtr.present)
3380 		return false;
3381 
3382 	return true;
3383 }
3384 
3385 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
3386 {
3387 	struct kvm_segment cs, ss;
3388 
3389 	vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3390 	vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3391 
3392 	return ((cs.selector & SEGMENT_RPL_MASK) ==
3393 		 (ss.selector & SEGMENT_RPL_MASK));
3394 }
3395 
3396 /*
3397  * Check if guest state is valid. Returns true if valid, false if
3398  * not.
3399  * We assume that registers are always usable
3400  */
3401 static bool guest_state_valid(struct kvm_vcpu *vcpu)
3402 {
3403 	if (enable_unrestricted_guest)
3404 		return true;
3405 
3406 	/* real mode guest state checks */
3407 	if (!is_protmode(vcpu) || (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
3408 		if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
3409 			return false;
3410 		if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
3411 			return false;
3412 		if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
3413 			return false;
3414 		if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
3415 			return false;
3416 		if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
3417 			return false;
3418 		if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
3419 			return false;
3420 	} else {
3421 	/* protected mode guest state checks */
3422 		if (!cs_ss_rpl_check(vcpu))
3423 			return false;
3424 		if (!code_segment_valid(vcpu))
3425 			return false;
3426 		if (!stack_segment_valid(vcpu))
3427 			return false;
3428 		if (!data_segment_valid(vcpu, VCPU_SREG_DS))
3429 			return false;
3430 		if (!data_segment_valid(vcpu, VCPU_SREG_ES))
3431 			return false;
3432 		if (!data_segment_valid(vcpu, VCPU_SREG_FS))
3433 			return false;
3434 		if (!data_segment_valid(vcpu, VCPU_SREG_GS))
3435 			return false;
3436 		if (!tr_valid(vcpu))
3437 			return false;
3438 		if (!ldtr_valid(vcpu))
3439 			return false;
3440 	}
3441 	/* TODO:
3442 	 * - Add checks on RIP
3443 	 * - Add checks on RFLAGS
3444 	 */
3445 
3446 	return true;
3447 }
3448 
3449 static int init_rmode_tss(struct kvm *kvm)
3450 {
3451 	gfn_t fn;
3452 	u16 data = 0;
3453 	int idx, r;
3454 
3455 	idx = srcu_read_lock(&kvm->srcu);
3456 	fn = to_kvm_vmx(kvm)->tss_addr >> PAGE_SHIFT;
3457 	r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3458 	if (r < 0)
3459 		goto out;
3460 	data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
3461 	r = kvm_write_guest_page(kvm, fn++, &data,
3462 			TSS_IOPB_BASE_OFFSET, sizeof(u16));
3463 	if (r < 0)
3464 		goto out;
3465 	r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
3466 	if (r < 0)
3467 		goto out;
3468 	r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3469 	if (r < 0)
3470 		goto out;
3471 	data = ~0;
3472 	r = kvm_write_guest_page(kvm, fn, &data,
3473 				 RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
3474 				 sizeof(u8));
3475 out:
3476 	srcu_read_unlock(&kvm->srcu, idx);
3477 	return r;
3478 }
3479 
3480 static int init_rmode_identity_map(struct kvm *kvm)
3481 {
3482 	struct kvm_vmx *kvm_vmx = to_kvm_vmx(kvm);
3483 	int i, r = 0;
3484 	kvm_pfn_t identity_map_pfn;
3485 	u32 tmp;
3486 
3487 	/* Protect kvm_vmx->ept_identity_pagetable_done. */
3488 	mutex_lock(&kvm->slots_lock);
3489 
3490 	if (likely(kvm_vmx->ept_identity_pagetable_done))
3491 		goto out;
3492 
3493 	if (!kvm_vmx->ept_identity_map_addr)
3494 		kvm_vmx->ept_identity_map_addr = VMX_EPT_IDENTITY_PAGETABLE_ADDR;
3495 	identity_map_pfn = kvm_vmx->ept_identity_map_addr >> PAGE_SHIFT;
3496 
3497 	r = __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
3498 				    kvm_vmx->ept_identity_map_addr, PAGE_SIZE);
3499 	if (r < 0)
3500 		goto out;
3501 
3502 	r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
3503 	if (r < 0)
3504 		goto out;
3505 	/* Set up identity-mapping pagetable for EPT in real mode */
3506 	for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
3507 		tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
3508 			_PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
3509 		r = kvm_write_guest_page(kvm, identity_map_pfn,
3510 				&tmp, i * sizeof(tmp), sizeof(tmp));
3511 		if (r < 0)
3512 			goto out;
3513 	}
3514 	kvm_vmx->ept_identity_pagetable_done = true;
3515 
3516 out:
3517 	mutex_unlock(&kvm->slots_lock);
3518 	return r;
3519 }
3520 
3521 static void seg_setup(int seg)
3522 {
3523 	const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3524 	unsigned int ar;
3525 
3526 	vmcs_write16(sf->selector, 0);
3527 	vmcs_writel(sf->base, 0);
3528 	vmcs_write32(sf->limit, 0xffff);
3529 	ar = 0x93;
3530 	if (seg == VCPU_SREG_CS)
3531 		ar |= 0x08; /* code segment */
3532 
3533 	vmcs_write32(sf->ar_bytes, ar);
3534 }
3535 
3536 static int alloc_apic_access_page(struct kvm *kvm)
3537 {
3538 	struct page *page;
3539 	int r = 0;
3540 
3541 	mutex_lock(&kvm->slots_lock);
3542 	if (kvm->arch.apic_access_page_done)
3543 		goto out;
3544 	r = __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
3545 				    APIC_DEFAULT_PHYS_BASE, PAGE_SIZE);
3546 	if (r)
3547 		goto out;
3548 
3549 	page = gfn_to_page(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
3550 	if (is_error_page(page)) {
3551 		r = -EFAULT;
3552 		goto out;
3553 	}
3554 
3555 	/*
3556 	 * Do not pin the page in memory, so that memory hot-unplug
3557 	 * is able to migrate it.
3558 	 */
3559 	put_page(page);
3560 	kvm->arch.apic_access_page_done = true;
3561 out:
3562 	mutex_unlock(&kvm->slots_lock);
3563 	return r;
3564 }
3565 
3566 int allocate_vpid(void)
3567 {
3568 	int vpid;
3569 
3570 	if (!enable_vpid)
3571 		return 0;
3572 	spin_lock(&vmx_vpid_lock);
3573 	vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
3574 	if (vpid < VMX_NR_VPIDS)
3575 		__set_bit(vpid, vmx_vpid_bitmap);
3576 	else
3577 		vpid = 0;
3578 	spin_unlock(&vmx_vpid_lock);
3579 	return vpid;
3580 }
3581 
3582 void free_vpid(int vpid)
3583 {
3584 	if (!enable_vpid || vpid == 0)
3585 		return;
3586 	spin_lock(&vmx_vpid_lock);
3587 	__clear_bit(vpid, vmx_vpid_bitmap);
3588 	spin_unlock(&vmx_vpid_lock);
3589 }
3590 
3591 static __always_inline void vmx_disable_intercept_for_msr(unsigned long *msr_bitmap,
3592 							  u32 msr, int type)
3593 {
3594 	int f = sizeof(unsigned long);
3595 
3596 	if (!cpu_has_vmx_msr_bitmap())
3597 		return;
3598 
3599 	if (static_branch_unlikely(&enable_evmcs))
3600 		evmcs_touch_msr_bitmap();
3601 
3602 	/*
3603 	 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
3604 	 * have the write-low and read-high bitmap offsets the wrong way round.
3605 	 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
3606 	 */
3607 	if (msr <= 0x1fff) {
3608 		if (type & MSR_TYPE_R)
3609 			/* read-low */
3610 			__clear_bit(msr, msr_bitmap + 0x000 / f);
3611 
3612 		if (type & MSR_TYPE_W)
3613 			/* write-low */
3614 			__clear_bit(msr, msr_bitmap + 0x800 / f);
3615 
3616 	} else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
3617 		msr &= 0x1fff;
3618 		if (type & MSR_TYPE_R)
3619 			/* read-high */
3620 			__clear_bit(msr, msr_bitmap + 0x400 / f);
3621 
3622 		if (type & MSR_TYPE_W)
3623 			/* write-high */
3624 			__clear_bit(msr, msr_bitmap + 0xc00 / f);
3625 
3626 	}
3627 }
3628 
3629 static __always_inline void vmx_enable_intercept_for_msr(unsigned long *msr_bitmap,
3630 							 u32 msr, int type)
3631 {
3632 	int f = sizeof(unsigned long);
3633 
3634 	if (!cpu_has_vmx_msr_bitmap())
3635 		return;
3636 
3637 	if (static_branch_unlikely(&enable_evmcs))
3638 		evmcs_touch_msr_bitmap();
3639 
3640 	/*
3641 	 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
3642 	 * have the write-low and read-high bitmap offsets the wrong way round.
3643 	 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
3644 	 */
3645 	if (msr <= 0x1fff) {
3646 		if (type & MSR_TYPE_R)
3647 			/* read-low */
3648 			__set_bit(msr, msr_bitmap + 0x000 / f);
3649 
3650 		if (type & MSR_TYPE_W)
3651 			/* write-low */
3652 			__set_bit(msr, msr_bitmap + 0x800 / f);
3653 
3654 	} else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
3655 		msr &= 0x1fff;
3656 		if (type & MSR_TYPE_R)
3657 			/* read-high */
3658 			__set_bit(msr, msr_bitmap + 0x400 / f);
3659 
3660 		if (type & MSR_TYPE_W)
3661 			/* write-high */
3662 			__set_bit(msr, msr_bitmap + 0xc00 / f);
3663 
3664 	}
3665 }
3666 
3667 static __always_inline void vmx_set_intercept_for_msr(unsigned long *msr_bitmap,
3668 			     			      u32 msr, int type, bool value)
3669 {
3670 	if (value)
3671 		vmx_enable_intercept_for_msr(msr_bitmap, msr, type);
3672 	else
3673 		vmx_disable_intercept_for_msr(msr_bitmap, msr, type);
3674 }
3675 
3676 static u8 vmx_msr_bitmap_mode(struct kvm_vcpu *vcpu)
3677 {
3678 	u8 mode = 0;
3679 
3680 	if (cpu_has_secondary_exec_ctrls() &&
3681 	    (secondary_exec_controls_get(to_vmx(vcpu)) &
3682 	     SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE)) {
3683 		mode |= MSR_BITMAP_MODE_X2APIC;
3684 		if (enable_apicv && kvm_vcpu_apicv_active(vcpu))
3685 			mode |= MSR_BITMAP_MODE_X2APIC_APICV;
3686 	}
3687 
3688 	return mode;
3689 }
3690 
3691 static void vmx_update_msr_bitmap_x2apic(unsigned long *msr_bitmap,
3692 					 u8 mode)
3693 {
3694 	int msr;
3695 
3696 	for (msr = 0x800; msr <= 0x8ff; msr += BITS_PER_LONG) {
3697 		unsigned word = msr / BITS_PER_LONG;
3698 		msr_bitmap[word] = (mode & MSR_BITMAP_MODE_X2APIC_APICV) ? 0 : ~0;
3699 		msr_bitmap[word + (0x800 / sizeof(long))] = ~0;
3700 	}
3701 
3702 	if (mode & MSR_BITMAP_MODE_X2APIC) {
3703 		/*
3704 		 * TPR reads and writes can be virtualized even if virtual interrupt
3705 		 * delivery is not in use.
3706 		 */
3707 		vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_TASKPRI), MSR_TYPE_RW);
3708 		if (mode & MSR_BITMAP_MODE_X2APIC_APICV) {
3709 			vmx_enable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_TMCCT), MSR_TYPE_R);
3710 			vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_EOI), MSR_TYPE_W);
3711 			vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_SELF_IPI), MSR_TYPE_W);
3712 		}
3713 	}
3714 }
3715 
3716 void vmx_update_msr_bitmap(struct kvm_vcpu *vcpu)
3717 {
3718 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3719 	unsigned long *msr_bitmap = vmx->vmcs01.msr_bitmap;
3720 	u8 mode = vmx_msr_bitmap_mode(vcpu);
3721 	u8 changed = mode ^ vmx->msr_bitmap_mode;
3722 
3723 	if (!changed)
3724 		return;
3725 
3726 	if (changed & (MSR_BITMAP_MODE_X2APIC | MSR_BITMAP_MODE_X2APIC_APICV))
3727 		vmx_update_msr_bitmap_x2apic(msr_bitmap, mode);
3728 
3729 	vmx->msr_bitmap_mode = mode;
3730 }
3731 
3732 void pt_update_intercept_for_msr(struct vcpu_vmx *vmx)
3733 {
3734 	unsigned long *msr_bitmap = vmx->vmcs01.msr_bitmap;
3735 	bool flag = !(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN);
3736 	u32 i;
3737 
3738 	vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_STATUS,
3739 							MSR_TYPE_RW, flag);
3740 	vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_OUTPUT_BASE,
3741 							MSR_TYPE_RW, flag);
3742 	vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_OUTPUT_MASK,
3743 							MSR_TYPE_RW, flag);
3744 	vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_CR3_MATCH,
3745 							MSR_TYPE_RW, flag);
3746 	for (i = 0; i < vmx->pt_desc.addr_range; i++) {
3747 		vmx_set_intercept_for_msr(msr_bitmap,
3748 			MSR_IA32_RTIT_ADDR0_A + i * 2, MSR_TYPE_RW, flag);
3749 		vmx_set_intercept_for_msr(msr_bitmap,
3750 			MSR_IA32_RTIT_ADDR0_B + i * 2, MSR_TYPE_RW, flag);
3751 	}
3752 }
3753 
3754 static bool vmx_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
3755 {
3756 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3757 	void *vapic_page;
3758 	u32 vppr;
3759 	int rvi;
3760 
3761 	if (WARN_ON_ONCE(!is_guest_mode(vcpu)) ||
3762 		!nested_cpu_has_vid(get_vmcs12(vcpu)) ||
3763 		WARN_ON_ONCE(!vmx->nested.virtual_apic_map.gfn))
3764 		return false;
3765 
3766 	rvi = vmx_get_rvi();
3767 
3768 	vapic_page = vmx->nested.virtual_apic_map.hva;
3769 	vppr = *((u32 *)(vapic_page + APIC_PROCPRI));
3770 
3771 	return ((rvi & 0xf0) > (vppr & 0xf0));
3772 }
3773 
3774 static inline bool kvm_vcpu_trigger_posted_interrupt(struct kvm_vcpu *vcpu,
3775 						     bool nested)
3776 {
3777 #ifdef CONFIG_SMP
3778 	int pi_vec = nested ? POSTED_INTR_NESTED_VECTOR : POSTED_INTR_VECTOR;
3779 
3780 	if (vcpu->mode == IN_GUEST_MODE) {
3781 		/*
3782 		 * The vector of interrupt to be delivered to vcpu had
3783 		 * been set in PIR before this function.
3784 		 *
3785 		 * Following cases will be reached in this block, and
3786 		 * we always send a notification event in all cases as
3787 		 * explained below.
3788 		 *
3789 		 * Case 1: vcpu keeps in non-root mode. Sending a
3790 		 * notification event posts the interrupt to vcpu.
3791 		 *
3792 		 * Case 2: vcpu exits to root mode and is still
3793 		 * runnable. PIR will be synced to vIRR before the
3794 		 * next vcpu entry. Sending a notification event in
3795 		 * this case has no effect, as vcpu is not in root
3796 		 * mode.
3797 		 *
3798 		 * Case 3: vcpu exits to root mode and is blocked.
3799 		 * vcpu_block() has already synced PIR to vIRR and
3800 		 * never blocks vcpu if vIRR is not cleared. Therefore,
3801 		 * a blocked vcpu here does not wait for any requested
3802 		 * interrupts in PIR, and sending a notification event
3803 		 * which has no effect is safe here.
3804 		 */
3805 
3806 		apic->send_IPI_mask(get_cpu_mask(vcpu->cpu), pi_vec);
3807 		return true;
3808 	}
3809 #endif
3810 	return false;
3811 }
3812 
3813 static int vmx_deliver_nested_posted_interrupt(struct kvm_vcpu *vcpu,
3814 						int vector)
3815 {
3816 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3817 
3818 	if (is_guest_mode(vcpu) &&
3819 	    vector == vmx->nested.posted_intr_nv) {
3820 		/*
3821 		 * If a posted intr is not recognized by hardware,
3822 		 * we will accomplish it in the next vmentry.
3823 		 */
3824 		vmx->nested.pi_pending = true;
3825 		kvm_make_request(KVM_REQ_EVENT, vcpu);
3826 		/* the PIR and ON have been set by L1. */
3827 		if (!kvm_vcpu_trigger_posted_interrupt(vcpu, true))
3828 			kvm_vcpu_kick(vcpu);
3829 		return 0;
3830 	}
3831 	return -1;
3832 }
3833 /*
3834  * Send interrupt to vcpu via posted interrupt way.
3835  * 1. If target vcpu is running(non-root mode), send posted interrupt
3836  * notification to vcpu and hardware will sync PIR to vIRR atomically.
3837  * 2. If target vcpu isn't running(root mode), kick it to pick up the
3838  * interrupt from PIR in next vmentry.
3839  */
3840 static int vmx_deliver_posted_interrupt(struct kvm_vcpu *vcpu, int vector)
3841 {
3842 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3843 	int r;
3844 
3845 	r = vmx_deliver_nested_posted_interrupt(vcpu, vector);
3846 	if (!r)
3847 		return 0;
3848 
3849 	if (!vcpu->arch.apicv_active)
3850 		return -1;
3851 
3852 	if (pi_test_and_set_pir(vector, &vmx->pi_desc))
3853 		return 0;
3854 
3855 	/* If a previous notification has sent the IPI, nothing to do.  */
3856 	if (pi_test_and_set_on(&vmx->pi_desc))
3857 		return 0;
3858 
3859 	if (!kvm_vcpu_trigger_posted_interrupt(vcpu, false))
3860 		kvm_vcpu_kick(vcpu);
3861 
3862 	return 0;
3863 }
3864 
3865 /*
3866  * Set up the vmcs's constant host-state fields, i.e., host-state fields that
3867  * will not change in the lifetime of the guest.
3868  * Note that host-state that does change is set elsewhere. E.g., host-state
3869  * that is set differently for each CPU is set in vmx_vcpu_load(), not here.
3870  */
3871 void vmx_set_constant_host_state(struct vcpu_vmx *vmx)
3872 {
3873 	u32 low32, high32;
3874 	unsigned long tmpl;
3875 	unsigned long cr0, cr3, cr4;
3876 
3877 	cr0 = read_cr0();
3878 	WARN_ON(cr0 & X86_CR0_TS);
3879 	vmcs_writel(HOST_CR0, cr0);  /* 22.2.3 */
3880 
3881 	/*
3882 	 * Save the most likely value for this task's CR3 in the VMCS.
3883 	 * We can't use __get_current_cr3_fast() because we're not atomic.
3884 	 */
3885 	cr3 = __read_cr3();
3886 	vmcs_writel(HOST_CR3, cr3);		/* 22.2.3  FIXME: shadow tables */
3887 	vmx->loaded_vmcs->host_state.cr3 = cr3;
3888 
3889 	/* Save the most likely value for this task's CR4 in the VMCS. */
3890 	cr4 = cr4_read_shadow();
3891 	vmcs_writel(HOST_CR4, cr4);			/* 22.2.3, 22.2.5 */
3892 	vmx->loaded_vmcs->host_state.cr4 = cr4;
3893 
3894 	vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS);  /* 22.2.4 */
3895 #ifdef CONFIG_X86_64
3896 	/*
3897 	 * Load null selectors, so we can avoid reloading them in
3898 	 * vmx_prepare_switch_to_host(), in case userspace uses
3899 	 * the null selectors too (the expected case).
3900 	 */
3901 	vmcs_write16(HOST_DS_SELECTOR, 0);
3902 	vmcs_write16(HOST_ES_SELECTOR, 0);
3903 #else
3904 	vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
3905 	vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
3906 #endif
3907 	vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
3908 	vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8);  /* 22.2.4 */
3909 
3910 	vmcs_writel(HOST_IDTR_BASE, host_idt_base);   /* 22.2.4 */
3911 
3912 	vmcs_writel(HOST_RIP, (unsigned long)vmx_vmexit); /* 22.2.5 */
3913 
3914 	rdmsr(MSR_IA32_SYSENTER_CS, low32, high32);
3915 	vmcs_write32(HOST_IA32_SYSENTER_CS, low32);
3916 	rdmsrl(MSR_IA32_SYSENTER_EIP, tmpl);
3917 	vmcs_writel(HOST_IA32_SYSENTER_EIP, tmpl);   /* 22.2.3 */
3918 
3919 	if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
3920 		rdmsr(MSR_IA32_CR_PAT, low32, high32);
3921 		vmcs_write64(HOST_IA32_PAT, low32 | ((u64) high32 << 32));
3922 	}
3923 
3924 	if (cpu_has_load_ia32_efer())
3925 		vmcs_write64(HOST_IA32_EFER, host_efer);
3926 }
3927 
3928 void set_cr4_guest_host_mask(struct vcpu_vmx *vmx)
3929 {
3930 	vmx->vcpu.arch.cr4_guest_owned_bits = KVM_CR4_GUEST_OWNED_BITS;
3931 	if (enable_ept)
3932 		vmx->vcpu.arch.cr4_guest_owned_bits |= X86_CR4_PGE;
3933 	if (is_guest_mode(&vmx->vcpu))
3934 		vmx->vcpu.arch.cr4_guest_owned_bits &=
3935 			~get_vmcs12(&vmx->vcpu)->cr4_guest_host_mask;
3936 	vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits);
3937 }
3938 
3939 u32 vmx_pin_based_exec_ctrl(struct vcpu_vmx *vmx)
3940 {
3941 	u32 pin_based_exec_ctrl = vmcs_config.pin_based_exec_ctrl;
3942 
3943 	if (!kvm_vcpu_apicv_active(&vmx->vcpu))
3944 		pin_based_exec_ctrl &= ~PIN_BASED_POSTED_INTR;
3945 
3946 	if (!enable_vnmi)
3947 		pin_based_exec_ctrl &= ~PIN_BASED_VIRTUAL_NMIS;
3948 
3949 	if (!enable_preemption_timer)
3950 		pin_based_exec_ctrl &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
3951 
3952 	return pin_based_exec_ctrl;
3953 }
3954 
3955 static void vmx_refresh_apicv_exec_ctrl(struct kvm_vcpu *vcpu)
3956 {
3957 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3958 
3959 	pin_controls_set(vmx, vmx_pin_based_exec_ctrl(vmx));
3960 	if (cpu_has_secondary_exec_ctrls()) {
3961 		if (kvm_vcpu_apicv_active(vcpu))
3962 			secondary_exec_controls_setbit(vmx,
3963 				      SECONDARY_EXEC_APIC_REGISTER_VIRT |
3964 				      SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
3965 		else
3966 			secondary_exec_controls_clearbit(vmx,
3967 					SECONDARY_EXEC_APIC_REGISTER_VIRT |
3968 					SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
3969 	}
3970 
3971 	if (cpu_has_vmx_msr_bitmap())
3972 		vmx_update_msr_bitmap(vcpu);
3973 }
3974 
3975 u32 vmx_exec_control(struct vcpu_vmx *vmx)
3976 {
3977 	u32 exec_control = vmcs_config.cpu_based_exec_ctrl;
3978 
3979 	if (vmx->vcpu.arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)
3980 		exec_control &= ~CPU_BASED_MOV_DR_EXITING;
3981 
3982 	if (!cpu_need_tpr_shadow(&vmx->vcpu)) {
3983 		exec_control &= ~CPU_BASED_TPR_SHADOW;
3984 #ifdef CONFIG_X86_64
3985 		exec_control |= CPU_BASED_CR8_STORE_EXITING |
3986 				CPU_BASED_CR8_LOAD_EXITING;
3987 #endif
3988 	}
3989 	if (!enable_ept)
3990 		exec_control |= CPU_BASED_CR3_STORE_EXITING |
3991 				CPU_BASED_CR3_LOAD_EXITING  |
3992 				CPU_BASED_INVLPG_EXITING;
3993 	if (kvm_mwait_in_guest(vmx->vcpu.kvm))
3994 		exec_control &= ~(CPU_BASED_MWAIT_EXITING |
3995 				CPU_BASED_MONITOR_EXITING);
3996 	if (kvm_hlt_in_guest(vmx->vcpu.kvm))
3997 		exec_control &= ~CPU_BASED_HLT_EXITING;
3998 	return exec_control;
3999 }
4000 
4001 
4002 static void vmx_compute_secondary_exec_control(struct vcpu_vmx *vmx)
4003 {
4004 	struct kvm_vcpu *vcpu = &vmx->vcpu;
4005 
4006 	u32 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
4007 
4008 	if (vmx_pt_mode_is_system())
4009 		exec_control &= ~(SECONDARY_EXEC_PT_USE_GPA | SECONDARY_EXEC_PT_CONCEAL_VMX);
4010 	if (!cpu_need_virtualize_apic_accesses(vcpu))
4011 		exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
4012 	if (vmx->vpid == 0)
4013 		exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
4014 	if (!enable_ept) {
4015 		exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
4016 		enable_unrestricted_guest = 0;
4017 	}
4018 	if (!enable_unrestricted_guest)
4019 		exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
4020 	if (kvm_pause_in_guest(vmx->vcpu.kvm))
4021 		exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
4022 	if (!kvm_vcpu_apicv_active(vcpu))
4023 		exec_control &= ~(SECONDARY_EXEC_APIC_REGISTER_VIRT |
4024 				  SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
4025 	exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
4026 
4027 	/* SECONDARY_EXEC_DESC is enabled/disabled on writes to CR4.UMIP,
4028 	 * in vmx_set_cr4.  */
4029 	exec_control &= ~SECONDARY_EXEC_DESC;
4030 
4031 	/* SECONDARY_EXEC_SHADOW_VMCS is enabled when L1 executes VMPTRLD
4032 	   (handle_vmptrld).
4033 	   We can NOT enable shadow_vmcs here because we don't have yet
4034 	   a current VMCS12
4035 	*/
4036 	exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
4037 
4038 	if (!enable_pml)
4039 		exec_control &= ~SECONDARY_EXEC_ENABLE_PML;
4040 
4041 	if (vmx_xsaves_supported()) {
4042 		/* Exposing XSAVES only when XSAVE is exposed */
4043 		bool xsaves_enabled =
4044 			boot_cpu_has(X86_FEATURE_XSAVE) &&
4045 			guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
4046 			guest_cpuid_has(vcpu, X86_FEATURE_XSAVES);
4047 
4048 		vcpu->arch.xsaves_enabled = xsaves_enabled;
4049 
4050 		if (!xsaves_enabled)
4051 			exec_control &= ~SECONDARY_EXEC_XSAVES;
4052 
4053 		if (nested) {
4054 			if (xsaves_enabled)
4055 				vmx->nested.msrs.secondary_ctls_high |=
4056 					SECONDARY_EXEC_XSAVES;
4057 			else
4058 				vmx->nested.msrs.secondary_ctls_high &=
4059 					~SECONDARY_EXEC_XSAVES;
4060 		}
4061 	}
4062 
4063 	if (cpu_has_vmx_rdtscp()) {
4064 		bool rdtscp_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP);
4065 		if (!rdtscp_enabled)
4066 			exec_control &= ~SECONDARY_EXEC_RDTSCP;
4067 
4068 		if (nested) {
4069 			if (rdtscp_enabled)
4070 				vmx->nested.msrs.secondary_ctls_high |=
4071 					SECONDARY_EXEC_RDTSCP;
4072 			else
4073 				vmx->nested.msrs.secondary_ctls_high &=
4074 					~SECONDARY_EXEC_RDTSCP;
4075 		}
4076 	}
4077 
4078 	if (cpu_has_vmx_invpcid()) {
4079 		/* Exposing INVPCID only when PCID is exposed */
4080 		bool invpcid_enabled =
4081 			guest_cpuid_has(vcpu, X86_FEATURE_INVPCID) &&
4082 			guest_cpuid_has(vcpu, X86_FEATURE_PCID);
4083 
4084 		if (!invpcid_enabled) {
4085 			exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID;
4086 			guest_cpuid_clear(vcpu, X86_FEATURE_INVPCID);
4087 		}
4088 
4089 		if (nested) {
4090 			if (invpcid_enabled)
4091 				vmx->nested.msrs.secondary_ctls_high |=
4092 					SECONDARY_EXEC_ENABLE_INVPCID;
4093 			else
4094 				vmx->nested.msrs.secondary_ctls_high &=
4095 					~SECONDARY_EXEC_ENABLE_INVPCID;
4096 		}
4097 	}
4098 
4099 	if (vmx_rdrand_supported()) {
4100 		bool rdrand_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDRAND);
4101 		if (rdrand_enabled)
4102 			exec_control &= ~SECONDARY_EXEC_RDRAND_EXITING;
4103 
4104 		if (nested) {
4105 			if (rdrand_enabled)
4106 				vmx->nested.msrs.secondary_ctls_high |=
4107 					SECONDARY_EXEC_RDRAND_EXITING;
4108 			else
4109 				vmx->nested.msrs.secondary_ctls_high &=
4110 					~SECONDARY_EXEC_RDRAND_EXITING;
4111 		}
4112 	}
4113 
4114 	if (vmx_rdseed_supported()) {
4115 		bool rdseed_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDSEED);
4116 		if (rdseed_enabled)
4117 			exec_control &= ~SECONDARY_EXEC_RDSEED_EXITING;
4118 
4119 		if (nested) {
4120 			if (rdseed_enabled)
4121 				vmx->nested.msrs.secondary_ctls_high |=
4122 					SECONDARY_EXEC_RDSEED_EXITING;
4123 			else
4124 				vmx->nested.msrs.secondary_ctls_high &=
4125 					~SECONDARY_EXEC_RDSEED_EXITING;
4126 		}
4127 	}
4128 
4129 	if (vmx_waitpkg_supported()) {
4130 		bool waitpkg_enabled =
4131 			guest_cpuid_has(vcpu, X86_FEATURE_WAITPKG);
4132 
4133 		if (!waitpkg_enabled)
4134 			exec_control &= ~SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE;
4135 
4136 		if (nested) {
4137 			if (waitpkg_enabled)
4138 				vmx->nested.msrs.secondary_ctls_high |=
4139 					SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE;
4140 			else
4141 				vmx->nested.msrs.secondary_ctls_high &=
4142 					~SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE;
4143 		}
4144 	}
4145 
4146 	vmx->secondary_exec_control = exec_control;
4147 }
4148 
4149 static void ept_set_mmio_spte_mask(void)
4150 {
4151 	/*
4152 	 * EPT Misconfigurations can be generated if the value of bits 2:0
4153 	 * of an EPT paging-structure entry is 110b (write/execute).
4154 	 */
4155 	kvm_mmu_set_mmio_spte_mask(VMX_EPT_RWX_MASK,
4156 				   VMX_EPT_MISCONFIG_WX_VALUE, 0);
4157 }
4158 
4159 #define VMX_XSS_EXIT_BITMAP 0
4160 
4161 /*
4162  * Noting that the initialization of Guest-state Area of VMCS is in
4163  * vmx_vcpu_reset().
4164  */
4165 static void init_vmcs(struct vcpu_vmx *vmx)
4166 {
4167 	if (nested)
4168 		nested_vmx_set_vmcs_shadowing_bitmap();
4169 
4170 	if (cpu_has_vmx_msr_bitmap())
4171 		vmcs_write64(MSR_BITMAP, __pa(vmx->vmcs01.msr_bitmap));
4172 
4173 	vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
4174 
4175 	/* Control */
4176 	pin_controls_set(vmx, vmx_pin_based_exec_ctrl(vmx));
4177 
4178 	exec_controls_set(vmx, vmx_exec_control(vmx));
4179 
4180 	if (cpu_has_secondary_exec_ctrls()) {
4181 		vmx_compute_secondary_exec_control(vmx);
4182 		secondary_exec_controls_set(vmx, vmx->secondary_exec_control);
4183 	}
4184 
4185 	if (kvm_vcpu_apicv_active(&vmx->vcpu)) {
4186 		vmcs_write64(EOI_EXIT_BITMAP0, 0);
4187 		vmcs_write64(EOI_EXIT_BITMAP1, 0);
4188 		vmcs_write64(EOI_EXIT_BITMAP2, 0);
4189 		vmcs_write64(EOI_EXIT_BITMAP3, 0);
4190 
4191 		vmcs_write16(GUEST_INTR_STATUS, 0);
4192 
4193 		vmcs_write16(POSTED_INTR_NV, POSTED_INTR_VECTOR);
4194 		vmcs_write64(POSTED_INTR_DESC_ADDR, __pa((&vmx->pi_desc)));
4195 	}
4196 
4197 	if (!kvm_pause_in_guest(vmx->vcpu.kvm)) {
4198 		vmcs_write32(PLE_GAP, ple_gap);
4199 		vmx->ple_window = ple_window;
4200 		vmx->ple_window_dirty = true;
4201 	}
4202 
4203 	vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
4204 	vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
4205 	vmcs_write32(CR3_TARGET_COUNT, 0);           /* 22.2.1 */
4206 
4207 	vmcs_write16(HOST_FS_SELECTOR, 0);            /* 22.2.4 */
4208 	vmcs_write16(HOST_GS_SELECTOR, 0);            /* 22.2.4 */
4209 	vmx_set_constant_host_state(vmx);
4210 	vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
4211 	vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
4212 
4213 	if (cpu_has_vmx_vmfunc())
4214 		vmcs_write64(VM_FUNCTION_CONTROL, 0);
4215 
4216 	vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
4217 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
4218 	vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host.val));
4219 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
4220 	vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest.val));
4221 
4222 	if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
4223 		vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
4224 
4225 	vm_exit_controls_set(vmx, vmx_vmexit_ctrl());
4226 
4227 	/* 22.2.1, 20.8.1 */
4228 	vm_entry_controls_set(vmx, vmx_vmentry_ctrl());
4229 
4230 	vmx->vcpu.arch.cr0_guest_owned_bits = X86_CR0_TS;
4231 	vmcs_writel(CR0_GUEST_HOST_MASK, ~X86_CR0_TS);
4232 
4233 	set_cr4_guest_host_mask(vmx);
4234 
4235 	if (vmx->vpid != 0)
4236 		vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
4237 
4238 	if (vmx_xsaves_supported())
4239 		vmcs_write64(XSS_EXIT_BITMAP, VMX_XSS_EXIT_BITMAP);
4240 
4241 	if (enable_pml) {
4242 		vmcs_write64(PML_ADDRESS, page_to_phys(vmx->pml_pg));
4243 		vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
4244 	}
4245 
4246 	if (cpu_has_vmx_encls_vmexit())
4247 		vmcs_write64(ENCLS_EXITING_BITMAP, -1ull);
4248 
4249 	if (vmx_pt_mode_is_host_guest()) {
4250 		memset(&vmx->pt_desc, 0, sizeof(vmx->pt_desc));
4251 		/* Bit[6~0] are forced to 1, writes are ignored. */
4252 		vmx->pt_desc.guest.output_mask = 0x7F;
4253 		vmcs_write64(GUEST_IA32_RTIT_CTL, 0);
4254 	}
4255 }
4256 
4257 static void vmx_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
4258 {
4259 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4260 	struct msr_data apic_base_msr;
4261 	u64 cr0;
4262 
4263 	vmx->rmode.vm86_active = 0;
4264 	vmx->spec_ctrl = 0;
4265 
4266 	vmx->msr_ia32_umwait_control = 0;
4267 
4268 	vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
4269 	vmx->hv_deadline_tsc = -1;
4270 	kvm_set_cr8(vcpu, 0);
4271 
4272 	if (!init_event) {
4273 		apic_base_msr.data = APIC_DEFAULT_PHYS_BASE |
4274 				     MSR_IA32_APICBASE_ENABLE;
4275 		if (kvm_vcpu_is_reset_bsp(vcpu))
4276 			apic_base_msr.data |= MSR_IA32_APICBASE_BSP;
4277 		apic_base_msr.host_initiated = true;
4278 		kvm_set_apic_base(vcpu, &apic_base_msr);
4279 	}
4280 
4281 	vmx_segment_cache_clear(vmx);
4282 
4283 	seg_setup(VCPU_SREG_CS);
4284 	vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
4285 	vmcs_writel(GUEST_CS_BASE, 0xffff0000ul);
4286 
4287 	seg_setup(VCPU_SREG_DS);
4288 	seg_setup(VCPU_SREG_ES);
4289 	seg_setup(VCPU_SREG_FS);
4290 	seg_setup(VCPU_SREG_GS);
4291 	seg_setup(VCPU_SREG_SS);
4292 
4293 	vmcs_write16(GUEST_TR_SELECTOR, 0);
4294 	vmcs_writel(GUEST_TR_BASE, 0);
4295 	vmcs_write32(GUEST_TR_LIMIT, 0xffff);
4296 	vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
4297 
4298 	vmcs_write16(GUEST_LDTR_SELECTOR, 0);
4299 	vmcs_writel(GUEST_LDTR_BASE, 0);
4300 	vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
4301 	vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
4302 
4303 	if (!init_event) {
4304 		vmcs_write32(GUEST_SYSENTER_CS, 0);
4305 		vmcs_writel(GUEST_SYSENTER_ESP, 0);
4306 		vmcs_writel(GUEST_SYSENTER_EIP, 0);
4307 		vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
4308 	}
4309 
4310 	kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
4311 	kvm_rip_write(vcpu, 0xfff0);
4312 
4313 	vmcs_writel(GUEST_GDTR_BASE, 0);
4314 	vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
4315 
4316 	vmcs_writel(GUEST_IDTR_BASE, 0);
4317 	vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
4318 
4319 	vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
4320 	vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
4321 	vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS, 0);
4322 	if (kvm_mpx_supported())
4323 		vmcs_write64(GUEST_BNDCFGS, 0);
4324 
4325 	setup_msrs(vmx);
4326 
4327 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);  /* 22.2.1 */
4328 
4329 	if (cpu_has_vmx_tpr_shadow() && !init_event) {
4330 		vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
4331 		if (cpu_need_tpr_shadow(vcpu))
4332 			vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
4333 				     __pa(vcpu->arch.apic->regs));
4334 		vmcs_write32(TPR_THRESHOLD, 0);
4335 	}
4336 
4337 	kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
4338 
4339 	cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
4340 	vmx->vcpu.arch.cr0 = cr0;
4341 	vmx_set_cr0(vcpu, cr0); /* enter rmode */
4342 	vmx_set_cr4(vcpu, 0);
4343 	vmx_set_efer(vcpu, 0);
4344 
4345 	update_exception_bitmap(vcpu);
4346 
4347 	vpid_sync_context(vmx->vpid);
4348 	if (init_event)
4349 		vmx_clear_hlt(vcpu);
4350 }
4351 
4352 static void enable_irq_window(struct kvm_vcpu *vcpu)
4353 {
4354 	exec_controls_setbit(to_vmx(vcpu), CPU_BASED_INTR_WINDOW_EXITING);
4355 }
4356 
4357 static void enable_nmi_window(struct kvm_vcpu *vcpu)
4358 {
4359 	if (!enable_vnmi ||
4360 	    vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_STI) {
4361 		enable_irq_window(vcpu);
4362 		return;
4363 	}
4364 
4365 	exec_controls_setbit(to_vmx(vcpu), CPU_BASED_NMI_WINDOW_EXITING);
4366 }
4367 
4368 static void vmx_inject_irq(struct kvm_vcpu *vcpu)
4369 {
4370 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4371 	uint32_t intr;
4372 	int irq = vcpu->arch.interrupt.nr;
4373 
4374 	trace_kvm_inj_virq(irq);
4375 
4376 	++vcpu->stat.irq_injections;
4377 	if (vmx->rmode.vm86_active) {
4378 		int inc_eip = 0;
4379 		if (vcpu->arch.interrupt.soft)
4380 			inc_eip = vcpu->arch.event_exit_inst_len;
4381 		kvm_inject_realmode_interrupt(vcpu, irq, inc_eip);
4382 		return;
4383 	}
4384 	intr = irq | INTR_INFO_VALID_MASK;
4385 	if (vcpu->arch.interrupt.soft) {
4386 		intr |= INTR_TYPE_SOFT_INTR;
4387 		vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
4388 			     vmx->vcpu.arch.event_exit_inst_len);
4389 	} else
4390 		intr |= INTR_TYPE_EXT_INTR;
4391 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
4392 
4393 	vmx_clear_hlt(vcpu);
4394 }
4395 
4396 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
4397 {
4398 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4399 
4400 	if (!enable_vnmi) {
4401 		/*
4402 		 * Tracking the NMI-blocked state in software is built upon
4403 		 * finding the next open IRQ window. This, in turn, depends on
4404 		 * well-behaving guests: They have to keep IRQs disabled at
4405 		 * least as long as the NMI handler runs. Otherwise we may
4406 		 * cause NMI nesting, maybe breaking the guest. But as this is
4407 		 * highly unlikely, we can live with the residual risk.
4408 		 */
4409 		vmx->loaded_vmcs->soft_vnmi_blocked = 1;
4410 		vmx->loaded_vmcs->vnmi_blocked_time = 0;
4411 	}
4412 
4413 	++vcpu->stat.nmi_injections;
4414 	vmx->loaded_vmcs->nmi_known_unmasked = false;
4415 
4416 	if (vmx->rmode.vm86_active) {
4417 		kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR, 0);
4418 		return;
4419 	}
4420 
4421 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
4422 			INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
4423 
4424 	vmx_clear_hlt(vcpu);
4425 }
4426 
4427 bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
4428 {
4429 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4430 	bool masked;
4431 
4432 	if (!enable_vnmi)
4433 		return vmx->loaded_vmcs->soft_vnmi_blocked;
4434 	if (vmx->loaded_vmcs->nmi_known_unmasked)
4435 		return false;
4436 	masked = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI;
4437 	vmx->loaded_vmcs->nmi_known_unmasked = !masked;
4438 	return masked;
4439 }
4440 
4441 void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
4442 {
4443 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4444 
4445 	if (!enable_vnmi) {
4446 		if (vmx->loaded_vmcs->soft_vnmi_blocked != masked) {
4447 			vmx->loaded_vmcs->soft_vnmi_blocked = masked;
4448 			vmx->loaded_vmcs->vnmi_blocked_time = 0;
4449 		}
4450 	} else {
4451 		vmx->loaded_vmcs->nmi_known_unmasked = !masked;
4452 		if (masked)
4453 			vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
4454 				      GUEST_INTR_STATE_NMI);
4455 		else
4456 			vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
4457 					GUEST_INTR_STATE_NMI);
4458 	}
4459 }
4460 
4461 static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
4462 {
4463 	if (to_vmx(vcpu)->nested.nested_run_pending)
4464 		return 0;
4465 
4466 	if (!enable_vnmi &&
4467 	    to_vmx(vcpu)->loaded_vmcs->soft_vnmi_blocked)
4468 		return 0;
4469 
4470 	return	!(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4471 		  (GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_STI
4472 		   | GUEST_INTR_STATE_NMI));
4473 }
4474 
4475 static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
4476 {
4477 	if (to_vmx(vcpu)->nested.nested_run_pending)
4478 		return false;
4479 
4480 	if (is_guest_mode(vcpu) && nested_exit_on_intr(vcpu))
4481 		return true;
4482 
4483 	return (vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
4484 		!(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4485 			(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
4486 }
4487 
4488 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
4489 {
4490 	int ret;
4491 
4492 	if (enable_unrestricted_guest)
4493 		return 0;
4494 
4495 	mutex_lock(&kvm->slots_lock);
4496 	ret = __x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, addr,
4497 				      PAGE_SIZE * 3);
4498 	mutex_unlock(&kvm->slots_lock);
4499 
4500 	if (ret)
4501 		return ret;
4502 	to_kvm_vmx(kvm)->tss_addr = addr;
4503 	return init_rmode_tss(kvm);
4504 }
4505 
4506 static int vmx_set_identity_map_addr(struct kvm *kvm, u64 ident_addr)
4507 {
4508 	to_kvm_vmx(kvm)->ept_identity_map_addr = ident_addr;
4509 	return 0;
4510 }
4511 
4512 static bool rmode_exception(struct kvm_vcpu *vcpu, int vec)
4513 {
4514 	switch (vec) {
4515 	case BP_VECTOR:
4516 		/*
4517 		 * Update instruction length as we may reinject the exception
4518 		 * from user space while in guest debugging mode.
4519 		 */
4520 		to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
4521 			vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4522 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
4523 			return false;
4524 		/* fall through */
4525 	case DB_VECTOR:
4526 		if (vcpu->guest_debug &
4527 			(KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
4528 			return false;
4529 		/* fall through */
4530 	case DE_VECTOR:
4531 	case OF_VECTOR:
4532 	case BR_VECTOR:
4533 	case UD_VECTOR:
4534 	case DF_VECTOR:
4535 	case SS_VECTOR:
4536 	case GP_VECTOR:
4537 	case MF_VECTOR:
4538 		return true;
4539 	}
4540 	return false;
4541 }
4542 
4543 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
4544 				  int vec, u32 err_code)
4545 {
4546 	/*
4547 	 * Instruction with address size override prefix opcode 0x67
4548 	 * Cause the #SS fault with 0 error code in VM86 mode.
4549 	 */
4550 	if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0) {
4551 		if (kvm_emulate_instruction(vcpu, 0)) {
4552 			if (vcpu->arch.halt_request) {
4553 				vcpu->arch.halt_request = 0;
4554 				return kvm_vcpu_halt(vcpu);
4555 			}
4556 			return 1;
4557 		}
4558 		return 0;
4559 	}
4560 
4561 	/*
4562 	 * Forward all other exceptions that are valid in real mode.
4563 	 * FIXME: Breaks guest debugging in real mode, needs to be fixed with
4564 	 *        the required debugging infrastructure rework.
4565 	 */
4566 	kvm_queue_exception(vcpu, vec);
4567 	return 1;
4568 }
4569 
4570 /*
4571  * Trigger machine check on the host. We assume all the MSRs are already set up
4572  * by the CPU and that we still run on the same CPU as the MCE occurred on.
4573  * We pass a fake environment to the machine check handler because we want
4574  * the guest to be always treated like user space, no matter what context
4575  * it used internally.
4576  */
4577 static void kvm_machine_check(void)
4578 {
4579 #if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
4580 	struct pt_regs regs = {
4581 		.cs = 3, /* Fake ring 3 no matter what the guest ran on */
4582 		.flags = X86_EFLAGS_IF,
4583 	};
4584 
4585 	do_machine_check(&regs, 0);
4586 #endif
4587 }
4588 
4589 static int handle_machine_check(struct kvm_vcpu *vcpu)
4590 {
4591 	/* handled by vmx_vcpu_run() */
4592 	return 1;
4593 }
4594 
4595 static int handle_exception_nmi(struct kvm_vcpu *vcpu)
4596 {
4597 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4598 	struct kvm_run *kvm_run = vcpu->run;
4599 	u32 intr_info, ex_no, error_code;
4600 	unsigned long cr2, rip, dr6;
4601 	u32 vect_info;
4602 
4603 	vect_info = vmx->idt_vectoring_info;
4604 	intr_info = vmx->exit_intr_info;
4605 
4606 	if (is_machine_check(intr_info) || is_nmi(intr_info))
4607 		return 1; /* handled by handle_exception_nmi_irqoff() */
4608 
4609 	if (is_invalid_opcode(intr_info))
4610 		return handle_ud(vcpu);
4611 
4612 	error_code = 0;
4613 	if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
4614 		error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
4615 
4616 	if (!vmx->rmode.vm86_active && is_gp_fault(intr_info)) {
4617 		WARN_ON_ONCE(!enable_vmware_backdoor);
4618 
4619 		/*
4620 		 * VMware backdoor emulation on #GP interception only handles
4621 		 * IN{S}, OUT{S}, and RDPMC, none of which generate a non-zero
4622 		 * error code on #GP.
4623 		 */
4624 		if (error_code) {
4625 			kvm_queue_exception_e(vcpu, GP_VECTOR, error_code);
4626 			return 1;
4627 		}
4628 		return kvm_emulate_instruction(vcpu, EMULTYPE_VMWARE_GP);
4629 	}
4630 
4631 	/*
4632 	 * The #PF with PFEC.RSVD = 1 indicates the guest is accessing
4633 	 * MMIO, it is better to report an internal error.
4634 	 * See the comments in vmx_handle_exit.
4635 	 */
4636 	if ((vect_info & VECTORING_INFO_VALID_MASK) &&
4637 	    !(is_page_fault(intr_info) && !(error_code & PFERR_RSVD_MASK))) {
4638 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4639 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
4640 		vcpu->run->internal.ndata = 3;
4641 		vcpu->run->internal.data[0] = vect_info;
4642 		vcpu->run->internal.data[1] = intr_info;
4643 		vcpu->run->internal.data[2] = error_code;
4644 		return 0;
4645 	}
4646 
4647 	if (is_page_fault(intr_info)) {
4648 		cr2 = vmcs_readl(EXIT_QUALIFICATION);
4649 		/* EPT won't cause page fault directly */
4650 		WARN_ON_ONCE(!vcpu->arch.apf.host_apf_reason && enable_ept);
4651 		return kvm_handle_page_fault(vcpu, error_code, cr2, NULL, 0);
4652 	}
4653 
4654 	ex_no = intr_info & INTR_INFO_VECTOR_MASK;
4655 
4656 	if (vmx->rmode.vm86_active && rmode_exception(vcpu, ex_no))
4657 		return handle_rmode_exception(vcpu, ex_no, error_code);
4658 
4659 	switch (ex_no) {
4660 	case AC_VECTOR:
4661 		kvm_queue_exception_e(vcpu, AC_VECTOR, error_code);
4662 		return 1;
4663 	case DB_VECTOR:
4664 		dr6 = vmcs_readl(EXIT_QUALIFICATION);
4665 		if (!(vcpu->guest_debug &
4666 		      (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
4667 			vcpu->arch.dr6 &= ~DR_TRAP_BITS;
4668 			vcpu->arch.dr6 |= dr6 | DR6_RTM;
4669 			if (is_icebp(intr_info))
4670 				WARN_ON(!skip_emulated_instruction(vcpu));
4671 
4672 			kvm_queue_exception(vcpu, DB_VECTOR);
4673 			return 1;
4674 		}
4675 		kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
4676 		kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
4677 		/* fall through */
4678 	case BP_VECTOR:
4679 		/*
4680 		 * Update instruction length as we may reinject #BP from
4681 		 * user space while in guest debugging mode. Reading it for
4682 		 * #DB as well causes no harm, it is not used in that case.
4683 		 */
4684 		vmx->vcpu.arch.event_exit_inst_len =
4685 			vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4686 		kvm_run->exit_reason = KVM_EXIT_DEBUG;
4687 		rip = kvm_rip_read(vcpu);
4688 		kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
4689 		kvm_run->debug.arch.exception = ex_no;
4690 		break;
4691 	default:
4692 		kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
4693 		kvm_run->ex.exception = ex_no;
4694 		kvm_run->ex.error_code = error_code;
4695 		break;
4696 	}
4697 	return 0;
4698 }
4699 
4700 static __always_inline int handle_external_interrupt(struct kvm_vcpu *vcpu)
4701 {
4702 	++vcpu->stat.irq_exits;
4703 	return 1;
4704 }
4705 
4706 static int handle_triple_fault(struct kvm_vcpu *vcpu)
4707 {
4708 	vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
4709 	vcpu->mmio_needed = 0;
4710 	return 0;
4711 }
4712 
4713 static int handle_io(struct kvm_vcpu *vcpu)
4714 {
4715 	unsigned long exit_qualification;
4716 	int size, in, string;
4717 	unsigned port;
4718 
4719 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4720 	string = (exit_qualification & 16) != 0;
4721 
4722 	++vcpu->stat.io_exits;
4723 
4724 	if (string)
4725 		return kvm_emulate_instruction(vcpu, 0);
4726 
4727 	port = exit_qualification >> 16;
4728 	size = (exit_qualification & 7) + 1;
4729 	in = (exit_qualification & 8) != 0;
4730 
4731 	return kvm_fast_pio(vcpu, size, port, in);
4732 }
4733 
4734 static void
4735 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
4736 {
4737 	/*
4738 	 * Patch in the VMCALL instruction:
4739 	 */
4740 	hypercall[0] = 0x0f;
4741 	hypercall[1] = 0x01;
4742 	hypercall[2] = 0xc1;
4743 }
4744 
4745 /* called to set cr0 as appropriate for a mov-to-cr0 exit. */
4746 static int handle_set_cr0(struct kvm_vcpu *vcpu, unsigned long val)
4747 {
4748 	if (is_guest_mode(vcpu)) {
4749 		struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4750 		unsigned long orig_val = val;
4751 
4752 		/*
4753 		 * We get here when L2 changed cr0 in a way that did not change
4754 		 * any of L1's shadowed bits (see nested_vmx_exit_handled_cr),
4755 		 * but did change L0 shadowed bits. So we first calculate the
4756 		 * effective cr0 value that L1 would like to write into the
4757 		 * hardware. It consists of the L2-owned bits from the new
4758 		 * value combined with the L1-owned bits from L1's guest_cr0.
4759 		 */
4760 		val = (val & ~vmcs12->cr0_guest_host_mask) |
4761 			(vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask);
4762 
4763 		if (!nested_guest_cr0_valid(vcpu, val))
4764 			return 1;
4765 
4766 		if (kvm_set_cr0(vcpu, val))
4767 			return 1;
4768 		vmcs_writel(CR0_READ_SHADOW, orig_val);
4769 		return 0;
4770 	} else {
4771 		if (to_vmx(vcpu)->nested.vmxon &&
4772 		    !nested_host_cr0_valid(vcpu, val))
4773 			return 1;
4774 
4775 		return kvm_set_cr0(vcpu, val);
4776 	}
4777 }
4778 
4779 static int handle_set_cr4(struct kvm_vcpu *vcpu, unsigned long val)
4780 {
4781 	if (is_guest_mode(vcpu)) {
4782 		struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4783 		unsigned long orig_val = val;
4784 
4785 		/* analogously to handle_set_cr0 */
4786 		val = (val & ~vmcs12->cr4_guest_host_mask) |
4787 			(vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask);
4788 		if (kvm_set_cr4(vcpu, val))
4789 			return 1;
4790 		vmcs_writel(CR4_READ_SHADOW, orig_val);
4791 		return 0;
4792 	} else
4793 		return kvm_set_cr4(vcpu, val);
4794 }
4795 
4796 static int handle_desc(struct kvm_vcpu *vcpu)
4797 {
4798 	WARN_ON(!(vcpu->arch.cr4 & X86_CR4_UMIP));
4799 	return kvm_emulate_instruction(vcpu, 0);
4800 }
4801 
4802 static int handle_cr(struct kvm_vcpu *vcpu)
4803 {
4804 	unsigned long exit_qualification, val;
4805 	int cr;
4806 	int reg;
4807 	int err;
4808 	int ret;
4809 
4810 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4811 	cr = exit_qualification & 15;
4812 	reg = (exit_qualification >> 8) & 15;
4813 	switch ((exit_qualification >> 4) & 3) {
4814 	case 0: /* mov to cr */
4815 		val = kvm_register_readl(vcpu, reg);
4816 		trace_kvm_cr_write(cr, val);
4817 		switch (cr) {
4818 		case 0:
4819 			err = handle_set_cr0(vcpu, val);
4820 			return kvm_complete_insn_gp(vcpu, err);
4821 		case 3:
4822 			WARN_ON_ONCE(enable_unrestricted_guest);
4823 			err = kvm_set_cr3(vcpu, val);
4824 			return kvm_complete_insn_gp(vcpu, err);
4825 		case 4:
4826 			err = handle_set_cr4(vcpu, val);
4827 			return kvm_complete_insn_gp(vcpu, err);
4828 		case 8: {
4829 				u8 cr8_prev = kvm_get_cr8(vcpu);
4830 				u8 cr8 = (u8)val;
4831 				err = kvm_set_cr8(vcpu, cr8);
4832 				ret = kvm_complete_insn_gp(vcpu, err);
4833 				if (lapic_in_kernel(vcpu))
4834 					return ret;
4835 				if (cr8_prev <= cr8)
4836 					return ret;
4837 				/*
4838 				 * TODO: we might be squashing a
4839 				 * KVM_GUESTDBG_SINGLESTEP-triggered
4840 				 * KVM_EXIT_DEBUG here.
4841 				 */
4842 				vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
4843 				return 0;
4844 			}
4845 		}
4846 		break;
4847 	case 2: /* clts */
4848 		WARN_ONCE(1, "Guest should always own CR0.TS");
4849 		vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
4850 		trace_kvm_cr_write(0, kvm_read_cr0(vcpu));
4851 		return kvm_skip_emulated_instruction(vcpu);
4852 	case 1: /*mov from cr*/
4853 		switch (cr) {
4854 		case 3:
4855 			WARN_ON_ONCE(enable_unrestricted_guest);
4856 			val = kvm_read_cr3(vcpu);
4857 			kvm_register_write(vcpu, reg, val);
4858 			trace_kvm_cr_read(cr, val);
4859 			return kvm_skip_emulated_instruction(vcpu);
4860 		case 8:
4861 			val = kvm_get_cr8(vcpu);
4862 			kvm_register_write(vcpu, reg, val);
4863 			trace_kvm_cr_read(cr, val);
4864 			return kvm_skip_emulated_instruction(vcpu);
4865 		}
4866 		break;
4867 	case 3: /* lmsw */
4868 		val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
4869 		trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val);
4870 		kvm_lmsw(vcpu, val);
4871 
4872 		return kvm_skip_emulated_instruction(vcpu);
4873 	default:
4874 		break;
4875 	}
4876 	vcpu->run->exit_reason = 0;
4877 	vcpu_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
4878 	       (int)(exit_qualification >> 4) & 3, cr);
4879 	return 0;
4880 }
4881 
4882 static int handle_dr(struct kvm_vcpu *vcpu)
4883 {
4884 	unsigned long exit_qualification;
4885 	int dr, dr7, reg;
4886 
4887 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4888 	dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
4889 
4890 	/* First, if DR does not exist, trigger UD */
4891 	if (!kvm_require_dr(vcpu, dr))
4892 		return 1;
4893 
4894 	/* Do not handle if the CPL > 0, will trigger GP on re-entry */
4895 	if (!kvm_require_cpl(vcpu, 0))
4896 		return 1;
4897 	dr7 = vmcs_readl(GUEST_DR7);
4898 	if (dr7 & DR7_GD) {
4899 		/*
4900 		 * As the vm-exit takes precedence over the debug trap, we
4901 		 * need to emulate the latter, either for the host or the
4902 		 * guest debugging itself.
4903 		 */
4904 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
4905 			vcpu->run->debug.arch.dr6 = vcpu->arch.dr6;
4906 			vcpu->run->debug.arch.dr7 = dr7;
4907 			vcpu->run->debug.arch.pc = kvm_get_linear_rip(vcpu);
4908 			vcpu->run->debug.arch.exception = DB_VECTOR;
4909 			vcpu->run->exit_reason = KVM_EXIT_DEBUG;
4910 			return 0;
4911 		} else {
4912 			vcpu->arch.dr6 &= ~DR_TRAP_BITS;
4913 			vcpu->arch.dr6 |= DR6_BD | DR6_RTM;
4914 			kvm_queue_exception(vcpu, DB_VECTOR);
4915 			return 1;
4916 		}
4917 	}
4918 
4919 	if (vcpu->guest_debug == 0) {
4920 		exec_controls_clearbit(to_vmx(vcpu), CPU_BASED_MOV_DR_EXITING);
4921 
4922 		/*
4923 		 * No more DR vmexits; force a reload of the debug registers
4924 		 * and reenter on this instruction.  The next vmexit will
4925 		 * retrieve the full state of the debug registers.
4926 		 */
4927 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
4928 		return 1;
4929 	}
4930 
4931 	reg = DEBUG_REG_ACCESS_REG(exit_qualification);
4932 	if (exit_qualification & TYPE_MOV_FROM_DR) {
4933 		unsigned long val;
4934 
4935 		if (kvm_get_dr(vcpu, dr, &val))
4936 			return 1;
4937 		kvm_register_write(vcpu, reg, val);
4938 	} else
4939 		if (kvm_set_dr(vcpu, dr, kvm_register_readl(vcpu, reg)))
4940 			return 1;
4941 
4942 	return kvm_skip_emulated_instruction(vcpu);
4943 }
4944 
4945 static u64 vmx_get_dr6(struct kvm_vcpu *vcpu)
4946 {
4947 	return vcpu->arch.dr6;
4948 }
4949 
4950 static void vmx_set_dr6(struct kvm_vcpu *vcpu, unsigned long val)
4951 {
4952 }
4953 
4954 static void vmx_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
4955 {
4956 	get_debugreg(vcpu->arch.db[0], 0);
4957 	get_debugreg(vcpu->arch.db[1], 1);
4958 	get_debugreg(vcpu->arch.db[2], 2);
4959 	get_debugreg(vcpu->arch.db[3], 3);
4960 	get_debugreg(vcpu->arch.dr6, 6);
4961 	vcpu->arch.dr7 = vmcs_readl(GUEST_DR7);
4962 
4963 	vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
4964 	exec_controls_setbit(to_vmx(vcpu), CPU_BASED_MOV_DR_EXITING);
4965 }
4966 
4967 static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
4968 {
4969 	vmcs_writel(GUEST_DR7, val);
4970 }
4971 
4972 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
4973 {
4974 	kvm_apic_update_ppr(vcpu);
4975 	return 1;
4976 }
4977 
4978 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
4979 {
4980 	exec_controls_clearbit(to_vmx(vcpu), CPU_BASED_INTR_WINDOW_EXITING);
4981 
4982 	kvm_make_request(KVM_REQ_EVENT, vcpu);
4983 
4984 	++vcpu->stat.irq_window_exits;
4985 	return 1;
4986 }
4987 
4988 static int handle_vmcall(struct kvm_vcpu *vcpu)
4989 {
4990 	return kvm_emulate_hypercall(vcpu);
4991 }
4992 
4993 static int handle_invd(struct kvm_vcpu *vcpu)
4994 {
4995 	return kvm_emulate_instruction(vcpu, 0);
4996 }
4997 
4998 static int handle_invlpg(struct kvm_vcpu *vcpu)
4999 {
5000 	unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5001 
5002 	kvm_mmu_invlpg(vcpu, exit_qualification);
5003 	return kvm_skip_emulated_instruction(vcpu);
5004 }
5005 
5006 static int handle_rdpmc(struct kvm_vcpu *vcpu)
5007 {
5008 	int err;
5009 
5010 	err = kvm_rdpmc(vcpu);
5011 	return kvm_complete_insn_gp(vcpu, err);
5012 }
5013 
5014 static int handle_wbinvd(struct kvm_vcpu *vcpu)
5015 {
5016 	return kvm_emulate_wbinvd(vcpu);
5017 }
5018 
5019 static int handle_xsetbv(struct kvm_vcpu *vcpu)
5020 {
5021 	u64 new_bv = kvm_read_edx_eax(vcpu);
5022 	u32 index = kvm_rcx_read(vcpu);
5023 
5024 	if (kvm_set_xcr(vcpu, index, new_bv) == 0)
5025 		return kvm_skip_emulated_instruction(vcpu);
5026 	return 1;
5027 }
5028 
5029 static int handle_apic_access(struct kvm_vcpu *vcpu)
5030 {
5031 	if (likely(fasteoi)) {
5032 		unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5033 		int access_type, offset;
5034 
5035 		access_type = exit_qualification & APIC_ACCESS_TYPE;
5036 		offset = exit_qualification & APIC_ACCESS_OFFSET;
5037 		/*
5038 		 * Sane guest uses MOV to write EOI, with written value
5039 		 * not cared. So make a short-circuit here by avoiding
5040 		 * heavy instruction emulation.
5041 		 */
5042 		if ((access_type == TYPE_LINEAR_APIC_INST_WRITE) &&
5043 		    (offset == APIC_EOI)) {
5044 			kvm_lapic_set_eoi(vcpu);
5045 			return kvm_skip_emulated_instruction(vcpu);
5046 		}
5047 	}
5048 	return kvm_emulate_instruction(vcpu, 0);
5049 }
5050 
5051 static int handle_apic_eoi_induced(struct kvm_vcpu *vcpu)
5052 {
5053 	unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5054 	int vector = exit_qualification & 0xff;
5055 
5056 	/* EOI-induced VM exit is trap-like and thus no need to adjust IP */
5057 	kvm_apic_set_eoi_accelerated(vcpu, vector);
5058 	return 1;
5059 }
5060 
5061 static int handle_apic_write(struct kvm_vcpu *vcpu)
5062 {
5063 	unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5064 	u32 offset = exit_qualification & 0xfff;
5065 
5066 	/* APIC-write VM exit is trap-like and thus no need to adjust IP */
5067 	kvm_apic_write_nodecode(vcpu, offset);
5068 	return 1;
5069 }
5070 
5071 static int handle_task_switch(struct kvm_vcpu *vcpu)
5072 {
5073 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5074 	unsigned long exit_qualification;
5075 	bool has_error_code = false;
5076 	u32 error_code = 0;
5077 	u16 tss_selector;
5078 	int reason, type, idt_v, idt_index;
5079 
5080 	idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
5081 	idt_index = (vmx->idt_vectoring_info & VECTORING_INFO_VECTOR_MASK);
5082 	type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
5083 
5084 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5085 
5086 	reason = (u32)exit_qualification >> 30;
5087 	if (reason == TASK_SWITCH_GATE && idt_v) {
5088 		switch (type) {
5089 		case INTR_TYPE_NMI_INTR:
5090 			vcpu->arch.nmi_injected = false;
5091 			vmx_set_nmi_mask(vcpu, true);
5092 			break;
5093 		case INTR_TYPE_EXT_INTR:
5094 		case INTR_TYPE_SOFT_INTR:
5095 			kvm_clear_interrupt_queue(vcpu);
5096 			break;
5097 		case INTR_TYPE_HARD_EXCEPTION:
5098 			if (vmx->idt_vectoring_info &
5099 			    VECTORING_INFO_DELIVER_CODE_MASK) {
5100 				has_error_code = true;
5101 				error_code =
5102 					vmcs_read32(IDT_VECTORING_ERROR_CODE);
5103 			}
5104 			/* fall through */
5105 		case INTR_TYPE_SOFT_EXCEPTION:
5106 			kvm_clear_exception_queue(vcpu);
5107 			break;
5108 		default:
5109 			break;
5110 		}
5111 	}
5112 	tss_selector = exit_qualification;
5113 
5114 	if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
5115 		       type != INTR_TYPE_EXT_INTR &&
5116 		       type != INTR_TYPE_NMI_INTR))
5117 		WARN_ON(!skip_emulated_instruction(vcpu));
5118 
5119 	/*
5120 	 * TODO: What about debug traps on tss switch?
5121 	 *       Are we supposed to inject them and update dr6?
5122 	 */
5123 	return kvm_task_switch(vcpu, tss_selector,
5124 			       type == INTR_TYPE_SOFT_INTR ? idt_index : -1,
5125 			       reason, has_error_code, error_code);
5126 }
5127 
5128 static int handle_ept_violation(struct kvm_vcpu *vcpu)
5129 {
5130 	unsigned long exit_qualification;
5131 	gpa_t gpa;
5132 	u64 error_code;
5133 
5134 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5135 
5136 	/*
5137 	 * EPT violation happened while executing iret from NMI,
5138 	 * "blocked by NMI" bit has to be set before next VM entry.
5139 	 * There are errata that may cause this bit to not be set:
5140 	 * AAK134, BY25.
5141 	 */
5142 	if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
5143 			enable_vnmi &&
5144 			(exit_qualification & INTR_INFO_UNBLOCK_NMI))
5145 		vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO, GUEST_INTR_STATE_NMI);
5146 
5147 	gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5148 	trace_kvm_page_fault(gpa, exit_qualification);
5149 
5150 	/* Is it a read fault? */
5151 	error_code = (exit_qualification & EPT_VIOLATION_ACC_READ)
5152 		     ? PFERR_USER_MASK : 0;
5153 	/* Is it a write fault? */
5154 	error_code |= (exit_qualification & EPT_VIOLATION_ACC_WRITE)
5155 		      ? PFERR_WRITE_MASK : 0;
5156 	/* Is it a fetch fault? */
5157 	error_code |= (exit_qualification & EPT_VIOLATION_ACC_INSTR)
5158 		      ? PFERR_FETCH_MASK : 0;
5159 	/* ept page table entry is present? */
5160 	error_code |= (exit_qualification &
5161 		       (EPT_VIOLATION_READABLE | EPT_VIOLATION_WRITABLE |
5162 			EPT_VIOLATION_EXECUTABLE))
5163 		      ? PFERR_PRESENT_MASK : 0;
5164 
5165 	error_code |= (exit_qualification & 0x100) != 0 ?
5166 	       PFERR_GUEST_FINAL_MASK : PFERR_GUEST_PAGE_MASK;
5167 
5168 	vcpu->arch.exit_qualification = exit_qualification;
5169 	return kvm_mmu_page_fault(vcpu, gpa, error_code, NULL, 0);
5170 }
5171 
5172 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
5173 {
5174 	gpa_t gpa;
5175 
5176 	/*
5177 	 * A nested guest cannot optimize MMIO vmexits, because we have an
5178 	 * nGPA here instead of the required GPA.
5179 	 */
5180 	gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5181 	if (!is_guest_mode(vcpu) &&
5182 	    !kvm_io_bus_write(vcpu, KVM_FAST_MMIO_BUS, gpa, 0, NULL)) {
5183 		trace_kvm_fast_mmio(gpa);
5184 		return kvm_skip_emulated_instruction(vcpu);
5185 	}
5186 
5187 	return kvm_mmu_page_fault(vcpu, gpa, PFERR_RSVD_MASK, NULL, 0);
5188 }
5189 
5190 static int handle_nmi_window(struct kvm_vcpu *vcpu)
5191 {
5192 	WARN_ON_ONCE(!enable_vnmi);
5193 	exec_controls_clearbit(to_vmx(vcpu), CPU_BASED_NMI_WINDOW_EXITING);
5194 	++vcpu->stat.nmi_window_exits;
5195 	kvm_make_request(KVM_REQ_EVENT, vcpu);
5196 
5197 	return 1;
5198 }
5199 
5200 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
5201 {
5202 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5203 	bool intr_window_requested;
5204 	unsigned count = 130;
5205 
5206 	/*
5207 	 * We should never reach the point where we are emulating L2
5208 	 * due to invalid guest state as that means we incorrectly
5209 	 * allowed a nested VMEntry with an invalid vmcs12.
5210 	 */
5211 	WARN_ON_ONCE(vmx->emulation_required && vmx->nested.nested_run_pending);
5212 
5213 	intr_window_requested = exec_controls_get(vmx) &
5214 				CPU_BASED_INTR_WINDOW_EXITING;
5215 
5216 	while (vmx->emulation_required && count-- != 0) {
5217 		if (intr_window_requested && vmx_interrupt_allowed(vcpu))
5218 			return handle_interrupt_window(&vmx->vcpu);
5219 
5220 		if (kvm_test_request(KVM_REQ_EVENT, vcpu))
5221 			return 1;
5222 
5223 		if (!kvm_emulate_instruction(vcpu, 0))
5224 			return 0;
5225 
5226 		if (vmx->emulation_required && !vmx->rmode.vm86_active &&
5227 		    vcpu->arch.exception.pending) {
5228 			vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5229 			vcpu->run->internal.suberror =
5230 						KVM_INTERNAL_ERROR_EMULATION;
5231 			vcpu->run->internal.ndata = 0;
5232 			return 0;
5233 		}
5234 
5235 		if (vcpu->arch.halt_request) {
5236 			vcpu->arch.halt_request = 0;
5237 			return kvm_vcpu_halt(vcpu);
5238 		}
5239 
5240 		/*
5241 		 * Note, return 1 and not 0, vcpu_run() is responsible for
5242 		 * morphing the pending signal into the proper return code.
5243 		 */
5244 		if (signal_pending(current))
5245 			return 1;
5246 
5247 		if (need_resched())
5248 			schedule();
5249 	}
5250 
5251 	return 1;
5252 }
5253 
5254 static void grow_ple_window(struct kvm_vcpu *vcpu)
5255 {
5256 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5257 	unsigned int old = vmx->ple_window;
5258 
5259 	vmx->ple_window = __grow_ple_window(old, ple_window,
5260 					    ple_window_grow,
5261 					    ple_window_max);
5262 
5263 	if (vmx->ple_window != old) {
5264 		vmx->ple_window_dirty = true;
5265 		trace_kvm_ple_window_update(vcpu->vcpu_id,
5266 					    vmx->ple_window, old);
5267 	}
5268 }
5269 
5270 static void shrink_ple_window(struct kvm_vcpu *vcpu)
5271 {
5272 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5273 	unsigned int old = vmx->ple_window;
5274 
5275 	vmx->ple_window = __shrink_ple_window(old, ple_window,
5276 					      ple_window_shrink,
5277 					      ple_window);
5278 
5279 	if (vmx->ple_window != old) {
5280 		vmx->ple_window_dirty = true;
5281 		trace_kvm_ple_window_update(vcpu->vcpu_id,
5282 					    vmx->ple_window, old);
5283 	}
5284 }
5285 
5286 /*
5287  * Handler for POSTED_INTERRUPT_WAKEUP_VECTOR.
5288  */
5289 static void wakeup_handler(void)
5290 {
5291 	struct kvm_vcpu *vcpu;
5292 	int cpu = smp_processor_id();
5293 
5294 	spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
5295 	list_for_each_entry(vcpu, &per_cpu(blocked_vcpu_on_cpu, cpu),
5296 			blocked_vcpu_list) {
5297 		struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
5298 
5299 		if (pi_test_on(pi_desc) == 1)
5300 			kvm_vcpu_kick(vcpu);
5301 	}
5302 	spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
5303 }
5304 
5305 static void vmx_enable_tdp(void)
5306 {
5307 	kvm_mmu_set_mask_ptes(VMX_EPT_READABLE_MASK,
5308 		enable_ept_ad_bits ? VMX_EPT_ACCESS_BIT : 0ull,
5309 		enable_ept_ad_bits ? VMX_EPT_DIRTY_BIT : 0ull,
5310 		0ull, VMX_EPT_EXECUTABLE_MASK,
5311 		cpu_has_vmx_ept_execute_only() ? 0ull : VMX_EPT_READABLE_MASK,
5312 		VMX_EPT_RWX_MASK, 0ull);
5313 
5314 	ept_set_mmio_spte_mask();
5315 }
5316 
5317 /*
5318  * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
5319  * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
5320  */
5321 static int handle_pause(struct kvm_vcpu *vcpu)
5322 {
5323 	if (!kvm_pause_in_guest(vcpu->kvm))
5324 		grow_ple_window(vcpu);
5325 
5326 	/*
5327 	 * Intel sdm vol3 ch-25.1.3 says: The "PAUSE-loop exiting"
5328 	 * VM-execution control is ignored if CPL > 0. OTOH, KVM
5329 	 * never set PAUSE_EXITING and just set PLE if supported,
5330 	 * so the vcpu must be CPL=0 if it gets a PAUSE exit.
5331 	 */
5332 	kvm_vcpu_on_spin(vcpu, true);
5333 	return kvm_skip_emulated_instruction(vcpu);
5334 }
5335 
5336 static int handle_nop(struct kvm_vcpu *vcpu)
5337 {
5338 	return kvm_skip_emulated_instruction(vcpu);
5339 }
5340 
5341 static int handle_mwait(struct kvm_vcpu *vcpu)
5342 {
5343 	printk_once(KERN_WARNING "kvm: MWAIT instruction emulated as NOP!\n");
5344 	return handle_nop(vcpu);
5345 }
5346 
5347 static int handle_invalid_op(struct kvm_vcpu *vcpu)
5348 {
5349 	kvm_queue_exception(vcpu, UD_VECTOR);
5350 	return 1;
5351 }
5352 
5353 static int handle_monitor_trap(struct kvm_vcpu *vcpu)
5354 {
5355 	return 1;
5356 }
5357 
5358 static int handle_monitor(struct kvm_vcpu *vcpu)
5359 {
5360 	printk_once(KERN_WARNING "kvm: MONITOR instruction emulated as NOP!\n");
5361 	return handle_nop(vcpu);
5362 }
5363 
5364 static int handle_invpcid(struct kvm_vcpu *vcpu)
5365 {
5366 	u32 vmx_instruction_info;
5367 	unsigned long type;
5368 	bool pcid_enabled;
5369 	gva_t gva;
5370 	struct x86_exception e;
5371 	unsigned i;
5372 	unsigned long roots_to_free = 0;
5373 	struct {
5374 		u64 pcid;
5375 		u64 gla;
5376 	} operand;
5377 
5378 	if (!guest_cpuid_has(vcpu, X86_FEATURE_INVPCID)) {
5379 		kvm_queue_exception(vcpu, UD_VECTOR);
5380 		return 1;
5381 	}
5382 
5383 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5384 	type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
5385 
5386 	if (type > 3) {
5387 		kvm_inject_gp(vcpu, 0);
5388 		return 1;
5389 	}
5390 
5391 	/* According to the Intel instruction reference, the memory operand
5392 	 * is read even if it isn't needed (e.g., for type==all)
5393 	 */
5394 	if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
5395 				vmx_instruction_info, false,
5396 				sizeof(operand), &gva))
5397 		return 1;
5398 
5399 	if (kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e)) {
5400 		kvm_inject_page_fault(vcpu, &e);
5401 		return 1;
5402 	}
5403 
5404 	if (operand.pcid >> 12 != 0) {
5405 		kvm_inject_gp(vcpu, 0);
5406 		return 1;
5407 	}
5408 
5409 	pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
5410 
5411 	switch (type) {
5412 	case INVPCID_TYPE_INDIV_ADDR:
5413 		if ((!pcid_enabled && (operand.pcid != 0)) ||
5414 		    is_noncanonical_address(operand.gla, vcpu)) {
5415 			kvm_inject_gp(vcpu, 0);
5416 			return 1;
5417 		}
5418 		kvm_mmu_invpcid_gva(vcpu, operand.gla, operand.pcid);
5419 		return kvm_skip_emulated_instruction(vcpu);
5420 
5421 	case INVPCID_TYPE_SINGLE_CTXT:
5422 		if (!pcid_enabled && (operand.pcid != 0)) {
5423 			kvm_inject_gp(vcpu, 0);
5424 			return 1;
5425 		}
5426 
5427 		if (kvm_get_active_pcid(vcpu) == operand.pcid) {
5428 			kvm_mmu_sync_roots(vcpu);
5429 			kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
5430 		}
5431 
5432 		for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++)
5433 			if (kvm_get_pcid(vcpu, vcpu->arch.mmu->prev_roots[i].cr3)
5434 			    == operand.pcid)
5435 				roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i);
5436 
5437 		kvm_mmu_free_roots(vcpu, vcpu->arch.mmu, roots_to_free);
5438 		/*
5439 		 * If neither the current cr3 nor any of the prev_roots use the
5440 		 * given PCID, then nothing needs to be done here because a
5441 		 * resync will happen anyway before switching to any other CR3.
5442 		 */
5443 
5444 		return kvm_skip_emulated_instruction(vcpu);
5445 
5446 	case INVPCID_TYPE_ALL_NON_GLOBAL:
5447 		/*
5448 		 * Currently, KVM doesn't mark global entries in the shadow
5449 		 * page tables, so a non-global flush just degenerates to a
5450 		 * global flush. If needed, we could optimize this later by
5451 		 * keeping track of global entries in shadow page tables.
5452 		 */
5453 
5454 		/* fall-through */
5455 	case INVPCID_TYPE_ALL_INCL_GLOBAL:
5456 		kvm_mmu_unload(vcpu);
5457 		return kvm_skip_emulated_instruction(vcpu);
5458 
5459 	default:
5460 		BUG(); /* We have already checked above that type <= 3 */
5461 	}
5462 }
5463 
5464 static int handle_pml_full(struct kvm_vcpu *vcpu)
5465 {
5466 	unsigned long exit_qualification;
5467 
5468 	trace_kvm_pml_full(vcpu->vcpu_id);
5469 
5470 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5471 
5472 	/*
5473 	 * PML buffer FULL happened while executing iret from NMI,
5474 	 * "blocked by NMI" bit has to be set before next VM entry.
5475 	 */
5476 	if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
5477 			enable_vnmi &&
5478 			(exit_qualification & INTR_INFO_UNBLOCK_NMI))
5479 		vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
5480 				GUEST_INTR_STATE_NMI);
5481 
5482 	/*
5483 	 * PML buffer already flushed at beginning of VMEXIT. Nothing to do
5484 	 * here.., and there's no userspace involvement needed for PML.
5485 	 */
5486 	return 1;
5487 }
5488 
5489 static int handle_preemption_timer(struct kvm_vcpu *vcpu)
5490 {
5491 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5492 
5493 	if (!vmx->req_immediate_exit &&
5494 	    !unlikely(vmx->loaded_vmcs->hv_timer_soft_disabled))
5495 		kvm_lapic_expired_hv_timer(vcpu);
5496 
5497 	return 1;
5498 }
5499 
5500 /*
5501  * When nested=0, all VMX instruction VM Exits filter here.  The handlers
5502  * are overwritten by nested_vmx_setup() when nested=1.
5503  */
5504 static int handle_vmx_instruction(struct kvm_vcpu *vcpu)
5505 {
5506 	kvm_queue_exception(vcpu, UD_VECTOR);
5507 	return 1;
5508 }
5509 
5510 static int handle_encls(struct kvm_vcpu *vcpu)
5511 {
5512 	/*
5513 	 * SGX virtualization is not yet supported.  There is no software
5514 	 * enable bit for SGX, so we have to trap ENCLS and inject a #UD
5515 	 * to prevent the guest from executing ENCLS.
5516 	 */
5517 	kvm_queue_exception(vcpu, UD_VECTOR);
5518 	return 1;
5519 }
5520 
5521 /*
5522  * The exit handlers return 1 if the exit was handled fully and guest execution
5523  * may resume.  Otherwise they set the kvm_run parameter to indicate what needs
5524  * to be done to userspace and return 0.
5525  */
5526 static int (*kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
5527 	[EXIT_REASON_EXCEPTION_NMI]           = handle_exception_nmi,
5528 	[EXIT_REASON_EXTERNAL_INTERRUPT]      = handle_external_interrupt,
5529 	[EXIT_REASON_TRIPLE_FAULT]            = handle_triple_fault,
5530 	[EXIT_REASON_NMI_WINDOW]	      = handle_nmi_window,
5531 	[EXIT_REASON_IO_INSTRUCTION]          = handle_io,
5532 	[EXIT_REASON_CR_ACCESS]               = handle_cr,
5533 	[EXIT_REASON_DR_ACCESS]               = handle_dr,
5534 	[EXIT_REASON_CPUID]                   = kvm_emulate_cpuid,
5535 	[EXIT_REASON_MSR_READ]                = kvm_emulate_rdmsr,
5536 	[EXIT_REASON_MSR_WRITE]               = kvm_emulate_wrmsr,
5537 	[EXIT_REASON_INTERRUPT_WINDOW]        = handle_interrupt_window,
5538 	[EXIT_REASON_HLT]                     = kvm_emulate_halt,
5539 	[EXIT_REASON_INVD]		      = handle_invd,
5540 	[EXIT_REASON_INVLPG]		      = handle_invlpg,
5541 	[EXIT_REASON_RDPMC]                   = handle_rdpmc,
5542 	[EXIT_REASON_VMCALL]                  = handle_vmcall,
5543 	[EXIT_REASON_VMCLEAR]		      = handle_vmx_instruction,
5544 	[EXIT_REASON_VMLAUNCH]		      = handle_vmx_instruction,
5545 	[EXIT_REASON_VMPTRLD]		      = handle_vmx_instruction,
5546 	[EXIT_REASON_VMPTRST]		      = handle_vmx_instruction,
5547 	[EXIT_REASON_VMREAD]		      = handle_vmx_instruction,
5548 	[EXIT_REASON_VMRESUME]		      = handle_vmx_instruction,
5549 	[EXIT_REASON_VMWRITE]		      = handle_vmx_instruction,
5550 	[EXIT_REASON_VMOFF]		      = handle_vmx_instruction,
5551 	[EXIT_REASON_VMON]		      = handle_vmx_instruction,
5552 	[EXIT_REASON_TPR_BELOW_THRESHOLD]     = handle_tpr_below_threshold,
5553 	[EXIT_REASON_APIC_ACCESS]             = handle_apic_access,
5554 	[EXIT_REASON_APIC_WRITE]              = handle_apic_write,
5555 	[EXIT_REASON_EOI_INDUCED]             = handle_apic_eoi_induced,
5556 	[EXIT_REASON_WBINVD]                  = handle_wbinvd,
5557 	[EXIT_REASON_XSETBV]                  = handle_xsetbv,
5558 	[EXIT_REASON_TASK_SWITCH]             = handle_task_switch,
5559 	[EXIT_REASON_MCE_DURING_VMENTRY]      = handle_machine_check,
5560 	[EXIT_REASON_GDTR_IDTR]		      = handle_desc,
5561 	[EXIT_REASON_LDTR_TR]		      = handle_desc,
5562 	[EXIT_REASON_EPT_VIOLATION]	      = handle_ept_violation,
5563 	[EXIT_REASON_EPT_MISCONFIG]           = handle_ept_misconfig,
5564 	[EXIT_REASON_PAUSE_INSTRUCTION]       = handle_pause,
5565 	[EXIT_REASON_MWAIT_INSTRUCTION]	      = handle_mwait,
5566 	[EXIT_REASON_MONITOR_TRAP_FLAG]       = handle_monitor_trap,
5567 	[EXIT_REASON_MONITOR_INSTRUCTION]     = handle_monitor,
5568 	[EXIT_REASON_INVEPT]                  = handle_vmx_instruction,
5569 	[EXIT_REASON_INVVPID]                 = handle_vmx_instruction,
5570 	[EXIT_REASON_RDRAND]                  = handle_invalid_op,
5571 	[EXIT_REASON_RDSEED]                  = handle_invalid_op,
5572 	[EXIT_REASON_PML_FULL]		      = handle_pml_full,
5573 	[EXIT_REASON_INVPCID]                 = handle_invpcid,
5574 	[EXIT_REASON_VMFUNC]		      = handle_vmx_instruction,
5575 	[EXIT_REASON_PREEMPTION_TIMER]	      = handle_preemption_timer,
5576 	[EXIT_REASON_ENCLS]		      = handle_encls,
5577 };
5578 
5579 static const int kvm_vmx_max_exit_handlers =
5580 	ARRAY_SIZE(kvm_vmx_exit_handlers);
5581 
5582 static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
5583 {
5584 	*info1 = vmcs_readl(EXIT_QUALIFICATION);
5585 	*info2 = vmcs_read32(VM_EXIT_INTR_INFO);
5586 }
5587 
5588 static void vmx_destroy_pml_buffer(struct vcpu_vmx *vmx)
5589 {
5590 	if (vmx->pml_pg) {
5591 		__free_page(vmx->pml_pg);
5592 		vmx->pml_pg = NULL;
5593 	}
5594 }
5595 
5596 static void vmx_flush_pml_buffer(struct kvm_vcpu *vcpu)
5597 {
5598 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5599 	u64 *pml_buf;
5600 	u16 pml_idx;
5601 
5602 	pml_idx = vmcs_read16(GUEST_PML_INDEX);
5603 
5604 	/* Do nothing if PML buffer is empty */
5605 	if (pml_idx == (PML_ENTITY_NUM - 1))
5606 		return;
5607 
5608 	/* PML index always points to next available PML buffer entity */
5609 	if (pml_idx >= PML_ENTITY_NUM)
5610 		pml_idx = 0;
5611 	else
5612 		pml_idx++;
5613 
5614 	pml_buf = page_address(vmx->pml_pg);
5615 	for (; pml_idx < PML_ENTITY_NUM; pml_idx++) {
5616 		u64 gpa;
5617 
5618 		gpa = pml_buf[pml_idx];
5619 		WARN_ON(gpa & (PAGE_SIZE - 1));
5620 		kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
5621 	}
5622 
5623 	/* reset PML index */
5624 	vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
5625 }
5626 
5627 /*
5628  * Flush all vcpus' PML buffer and update logged GPAs to dirty_bitmap.
5629  * Called before reporting dirty_bitmap to userspace.
5630  */
5631 static void kvm_flush_pml_buffers(struct kvm *kvm)
5632 {
5633 	int i;
5634 	struct kvm_vcpu *vcpu;
5635 	/*
5636 	 * We only need to kick vcpu out of guest mode here, as PML buffer
5637 	 * is flushed at beginning of all VMEXITs, and it's obvious that only
5638 	 * vcpus running in guest are possible to have unflushed GPAs in PML
5639 	 * buffer.
5640 	 */
5641 	kvm_for_each_vcpu(i, vcpu, kvm)
5642 		kvm_vcpu_kick(vcpu);
5643 }
5644 
5645 static void vmx_dump_sel(char *name, uint32_t sel)
5646 {
5647 	pr_err("%s sel=0x%04x, attr=0x%05x, limit=0x%08x, base=0x%016lx\n",
5648 	       name, vmcs_read16(sel),
5649 	       vmcs_read32(sel + GUEST_ES_AR_BYTES - GUEST_ES_SELECTOR),
5650 	       vmcs_read32(sel + GUEST_ES_LIMIT - GUEST_ES_SELECTOR),
5651 	       vmcs_readl(sel + GUEST_ES_BASE - GUEST_ES_SELECTOR));
5652 }
5653 
5654 static void vmx_dump_dtsel(char *name, uint32_t limit)
5655 {
5656 	pr_err("%s                           limit=0x%08x, base=0x%016lx\n",
5657 	       name, vmcs_read32(limit),
5658 	       vmcs_readl(limit + GUEST_GDTR_BASE - GUEST_GDTR_LIMIT));
5659 }
5660 
5661 void dump_vmcs(void)
5662 {
5663 	u32 vmentry_ctl, vmexit_ctl;
5664 	u32 cpu_based_exec_ctrl, pin_based_exec_ctrl, secondary_exec_control;
5665 	unsigned long cr4;
5666 	u64 efer;
5667 	int i, n;
5668 
5669 	if (!dump_invalid_vmcs) {
5670 		pr_warn_ratelimited("set kvm_intel.dump_invalid_vmcs=1 to dump internal KVM state.\n");
5671 		return;
5672 	}
5673 
5674 	vmentry_ctl = vmcs_read32(VM_ENTRY_CONTROLS);
5675 	vmexit_ctl = vmcs_read32(VM_EXIT_CONTROLS);
5676 	cpu_based_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5677 	pin_based_exec_ctrl = vmcs_read32(PIN_BASED_VM_EXEC_CONTROL);
5678 	cr4 = vmcs_readl(GUEST_CR4);
5679 	efer = vmcs_read64(GUEST_IA32_EFER);
5680 	secondary_exec_control = 0;
5681 	if (cpu_has_secondary_exec_ctrls())
5682 		secondary_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
5683 
5684 	pr_err("*** Guest State ***\n");
5685 	pr_err("CR0: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
5686 	       vmcs_readl(GUEST_CR0), vmcs_readl(CR0_READ_SHADOW),
5687 	       vmcs_readl(CR0_GUEST_HOST_MASK));
5688 	pr_err("CR4: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
5689 	       cr4, vmcs_readl(CR4_READ_SHADOW), vmcs_readl(CR4_GUEST_HOST_MASK));
5690 	pr_err("CR3 = 0x%016lx\n", vmcs_readl(GUEST_CR3));
5691 	if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT) &&
5692 	    (cr4 & X86_CR4_PAE) && !(efer & EFER_LMA))
5693 	{
5694 		pr_err("PDPTR0 = 0x%016llx  PDPTR1 = 0x%016llx\n",
5695 		       vmcs_read64(GUEST_PDPTR0), vmcs_read64(GUEST_PDPTR1));
5696 		pr_err("PDPTR2 = 0x%016llx  PDPTR3 = 0x%016llx\n",
5697 		       vmcs_read64(GUEST_PDPTR2), vmcs_read64(GUEST_PDPTR3));
5698 	}
5699 	pr_err("RSP = 0x%016lx  RIP = 0x%016lx\n",
5700 	       vmcs_readl(GUEST_RSP), vmcs_readl(GUEST_RIP));
5701 	pr_err("RFLAGS=0x%08lx         DR7 = 0x%016lx\n",
5702 	       vmcs_readl(GUEST_RFLAGS), vmcs_readl(GUEST_DR7));
5703 	pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
5704 	       vmcs_readl(GUEST_SYSENTER_ESP),
5705 	       vmcs_read32(GUEST_SYSENTER_CS), vmcs_readl(GUEST_SYSENTER_EIP));
5706 	vmx_dump_sel("CS:  ", GUEST_CS_SELECTOR);
5707 	vmx_dump_sel("DS:  ", GUEST_DS_SELECTOR);
5708 	vmx_dump_sel("SS:  ", GUEST_SS_SELECTOR);
5709 	vmx_dump_sel("ES:  ", GUEST_ES_SELECTOR);
5710 	vmx_dump_sel("FS:  ", GUEST_FS_SELECTOR);
5711 	vmx_dump_sel("GS:  ", GUEST_GS_SELECTOR);
5712 	vmx_dump_dtsel("GDTR:", GUEST_GDTR_LIMIT);
5713 	vmx_dump_sel("LDTR:", GUEST_LDTR_SELECTOR);
5714 	vmx_dump_dtsel("IDTR:", GUEST_IDTR_LIMIT);
5715 	vmx_dump_sel("TR:  ", GUEST_TR_SELECTOR);
5716 	if ((vmexit_ctl & (VM_EXIT_SAVE_IA32_PAT | VM_EXIT_SAVE_IA32_EFER)) ||
5717 	    (vmentry_ctl & (VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_IA32_EFER)))
5718 		pr_err("EFER =     0x%016llx  PAT = 0x%016llx\n",
5719 		       efer, vmcs_read64(GUEST_IA32_PAT));
5720 	pr_err("DebugCtl = 0x%016llx  DebugExceptions = 0x%016lx\n",
5721 	       vmcs_read64(GUEST_IA32_DEBUGCTL),
5722 	       vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS));
5723 	if (cpu_has_load_perf_global_ctrl() &&
5724 	    vmentry_ctl & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
5725 		pr_err("PerfGlobCtl = 0x%016llx\n",
5726 		       vmcs_read64(GUEST_IA32_PERF_GLOBAL_CTRL));
5727 	if (vmentry_ctl & VM_ENTRY_LOAD_BNDCFGS)
5728 		pr_err("BndCfgS = 0x%016llx\n", vmcs_read64(GUEST_BNDCFGS));
5729 	pr_err("Interruptibility = %08x  ActivityState = %08x\n",
5730 	       vmcs_read32(GUEST_INTERRUPTIBILITY_INFO),
5731 	       vmcs_read32(GUEST_ACTIVITY_STATE));
5732 	if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)
5733 		pr_err("InterruptStatus = %04x\n",
5734 		       vmcs_read16(GUEST_INTR_STATUS));
5735 
5736 	pr_err("*** Host State ***\n");
5737 	pr_err("RIP = 0x%016lx  RSP = 0x%016lx\n",
5738 	       vmcs_readl(HOST_RIP), vmcs_readl(HOST_RSP));
5739 	pr_err("CS=%04x SS=%04x DS=%04x ES=%04x FS=%04x GS=%04x TR=%04x\n",
5740 	       vmcs_read16(HOST_CS_SELECTOR), vmcs_read16(HOST_SS_SELECTOR),
5741 	       vmcs_read16(HOST_DS_SELECTOR), vmcs_read16(HOST_ES_SELECTOR),
5742 	       vmcs_read16(HOST_FS_SELECTOR), vmcs_read16(HOST_GS_SELECTOR),
5743 	       vmcs_read16(HOST_TR_SELECTOR));
5744 	pr_err("FSBase=%016lx GSBase=%016lx TRBase=%016lx\n",
5745 	       vmcs_readl(HOST_FS_BASE), vmcs_readl(HOST_GS_BASE),
5746 	       vmcs_readl(HOST_TR_BASE));
5747 	pr_err("GDTBase=%016lx IDTBase=%016lx\n",
5748 	       vmcs_readl(HOST_GDTR_BASE), vmcs_readl(HOST_IDTR_BASE));
5749 	pr_err("CR0=%016lx CR3=%016lx CR4=%016lx\n",
5750 	       vmcs_readl(HOST_CR0), vmcs_readl(HOST_CR3),
5751 	       vmcs_readl(HOST_CR4));
5752 	pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
5753 	       vmcs_readl(HOST_IA32_SYSENTER_ESP),
5754 	       vmcs_read32(HOST_IA32_SYSENTER_CS),
5755 	       vmcs_readl(HOST_IA32_SYSENTER_EIP));
5756 	if (vmexit_ctl & (VM_EXIT_LOAD_IA32_PAT | VM_EXIT_LOAD_IA32_EFER))
5757 		pr_err("EFER = 0x%016llx  PAT = 0x%016llx\n",
5758 		       vmcs_read64(HOST_IA32_EFER),
5759 		       vmcs_read64(HOST_IA32_PAT));
5760 	if (cpu_has_load_perf_global_ctrl() &&
5761 	    vmexit_ctl & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
5762 		pr_err("PerfGlobCtl = 0x%016llx\n",
5763 		       vmcs_read64(HOST_IA32_PERF_GLOBAL_CTRL));
5764 
5765 	pr_err("*** Control State ***\n");
5766 	pr_err("PinBased=%08x CPUBased=%08x SecondaryExec=%08x\n",
5767 	       pin_based_exec_ctrl, cpu_based_exec_ctrl, secondary_exec_control);
5768 	pr_err("EntryControls=%08x ExitControls=%08x\n", vmentry_ctl, vmexit_ctl);
5769 	pr_err("ExceptionBitmap=%08x PFECmask=%08x PFECmatch=%08x\n",
5770 	       vmcs_read32(EXCEPTION_BITMAP),
5771 	       vmcs_read32(PAGE_FAULT_ERROR_CODE_MASK),
5772 	       vmcs_read32(PAGE_FAULT_ERROR_CODE_MATCH));
5773 	pr_err("VMEntry: intr_info=%08x errcode=%08x ilen=%08x\n",
5774 	       vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
5775 	       vmcs_read32(VM_ENTRY_EXCEPTION_ERROR_CODE),
5776 	       vmcs_read32(VM_ENTRY_INSTRUCTION_LEN));
5777 	pr_err("VMExit: intr_info=%08x errcode=%08x ilen=%08x\n",
5778 	       vmcs_read32(VM_EXIT_INTR_INFO),
5779 	       vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
5780 	       vmcs_read32(VM_EXIT_INSTRUCTION_LEN));
5781 	pr_err("        reason=%08x qualification=%016lx\n",
5782 	       vmcs_read32(VM_EXIT_REASON), vmcs_readl(EXIT_QUALIFICATION));
5783 	pr_err("IDTVectoring: info=%08x errcode=%08x\n",
5784 	       vmcs_read32(IDT_VECTORING_INFO_FIELD),
5785 	       vmcs_read32(IDT_VECTORING_ERROR_CODE));
5786 	pr_err("TSC Offset = 0x%016llx\n", vmcs_read64(TSC_OFFSET));
5787 	if (secondary_exec_control & SECONDARY_EXEC_TSC_SCALING)
5788 		pr_err("TSC Multiplier = 0x%016llx\n",
5789 		       vmcs_read64(TSC_MULTIPLIER));
5790 	if (cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW) {
5791 		if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY) {
5792 			u16 status = vmcs_read16(GUEST_INTR_STATUS);
5793 			pr_err("SVI|RVI = %02x|%02x ", status >> 8, status & 0xff);
5794 		}
5795 		pr_cont("TPR Threshold = 0x%02x\n", vmcs_read32(TPR_THRESHOLD));
5796 		if (secondary_exec_control & SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)
5797 			pr_err("APIC-access addr = 0x%016llx ", vmcs_read64(APIC_ACCESS_ADDR));
5798 		pr_cont("virt-APIC addr = 0x%016llx\n", vmcs_read64(VIRTUAL_APIC_PAGE_ADDR));
5799 	}
5800 	if (pin_based_exec_ctrl & PIN_BASED_POSTED_INTR)
5801 		pr_err("PostedIntrVec = 0x%02x\n", vmcs_read16(POSTED_INTR_NV));
5802 	if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT))
5803 		pr_err("EPT pointer = 0x%016llx\n", vmcs_read64(EPT_POINTER));
5804 	n = vmcs_read32(CR3_TARGET_COUNT);
5805 	for (i = 0; i + 1 < n; i += 4)
5806 		pr_err("CR3 target%u=%016lx target%u=%016lx\n",
5807 		       i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2),
5808 		       i + 1, vmcs_readl(CR3_TARGET_VALUE0 + i * 2 + 2));
5809 	if (i < n)
5810 		pr_err("CR3 target%u=%016lx\n",
5811 		       i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2));
5812 	if (secondary_exec_control & SECONDARY_EXEC_PAUSE_LOOP_EXITING)
5813 		pr_err("PLE Gap=%08x Window=%08x\n",
5814 		       vmcs_read32(PLE_GAP), vmcs_read32(PLE_WINDOW));
5815 	if (secondary_exec_control & SECONDARY_EXEC_ENABLE_VPID)
5816 		pr_err("Virtual processor ID = 0x%04x\n",
5817 		       vmcs_read16(VIRTUAL_PROCESSOR_ID));
5818 }
5819 
5820 /*
5821  * The guest has exited.  See if we can fix it or if we need userspace
5822  * assistance.
5823  */
5824 static int vmx_handle_exit(struct kvm_vcpu *vcpu,
5825 	enum exit_fastpath_completion exit_fastpath)
5826 {
5827 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5828 	u32 exit_reason = vmx->exit_reason;
5829 	u32 vectoring_info = vmx->idt_vectoring_info;
5830 
5831 	trace_kvm_exit(exit_reason, vcpu, KVM_ISA_VMX);
5832 
5833 	/*
5834 	 * Flush logged GPAs PML buffer, this will make dirty_bitmap more
5835 	 * updated. Another good is, in kvm_vm_ioctl_get_dirty_log, before
5836 	 * querying dirty_bitmap, we only need to kick all vcpus out of guest
5837 	 * mode as if vcpus is in root mode, the PML buffer must has been
5838 	 * flushed already.
5839 	 */
5840 	if (enable_pml)
5841 		vmx_flush_pml_buffer(vcpu);
5842 
5843 	/* If guest state is invalid, start emulating */
5844 	if (vmx->emulation_required)
5845 		return handle_invalid_guest_state(vcpu);
5846 
5847 	if (is_guest_mode(vcpu)) {
5848 		/*
5849 		 * The host physical addresses of some pages of guest memory
5850 		 * are loaded into the vmcs02 (e.g. vmcs12's Virtual APIC
5851 		 * Page). The CPU may write to these pages via their host
5852 		 * physical address while L2 is running, bypassing any
5853 		 * address-translation-based dirty tracking (e.g. EPT write
5854 		 * protection).
5855 		 *
5856 		 * Mark them dirty on every exit from L2 to prevent them from
5857 		 * getting out of sync with dirty tracking.
5858 		 */
5859 		nested_mark_vmcs12_pages_dirty(vcpu);
5860 
5861 		if (nested_vmx_exit_reflected(vcpu, exit_reason))
5862 			return nested_vmx_reflect_vmexit(vcpu, exit_reason);
5863 	}
5864 
5865 	if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) {
5866 		dump_vmcs();
5867 		vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
5868 		vcpu->run->fail_entry.hardware_entry_failure_reason
5869 			= exit_reason;
5870 		return 0;
5871 	}
5872 
5873 	if (unlikely(vmx->fail)) {
5874 		dump_vmcs();
5875 		vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
5876 		vcpu->run->fail_entry.hardware_entry_failure_reason
5877 			= vmcs_read32(VM_INSTRUCTION_ERROR);
5878 		return 0;
5879 	}
5880 
5881 	/*
5882 	 * Note:
5883 	 * Do not try to fix EXIT_REASON_EPT_MISCONFIG if it caused by
5884 	 * delivery event since it indicates guest is accessing MMIO.
5885 	 * The vm-exit can be triggered again after return to guest that
5886 	 * will cause infinite loop.
5887 	 */
5888 	if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
5889 			(exit_reason != EXIT_REASON_EXCEPTION_NMI &&
5890 			exit_reason != EXIT_REASON_EPT_VIOLATION &&
5891 			exit_reason != EXIT_REASON_PML_FULL &&
5892 			exit_reason != EXIT_REASON_TASK_SWITCH)) {
5893 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5894 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_DELIVERY_EV;
5895 		vcpu->run->internal.ndata = 3;
5896 		vcpu->run->internal.data[0] = vectoring_info;
5897 		vcpu->run->internal.data[1] = exit_reason;
5898 		vcpu->run->internal.data[2] = vcpu->arch.exit_qualification;
5899 		if (exit_reason == EXIT_REASON_EPT_MISCONFIG) {
5900 			vcpu->run->internal.ndata++;
5901 			vcpu->run->internal.data[3] =
5902 				vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5903 		}
5904 		return 0;
5905 	}
5906 
5907 	if (unlikely(!enable_vnmi &&
5908 		     vmx->loaded_vmcs->soft_vnmi_blocked)) {
5909 		if (vmx_interrupt_allowed(vcpu)) {
5910 			vmx->loaded_vmcs->soft_vnmi_blocked = 0;
5911 		} else if (vmx->loaded_vmcs->vnmi_blocked_time > 1000000000LL &&
5912 			   vcpu->arch.nmi_pending) {
5913 			/*
5914 			 * This CPU don't support us in finding the end of an
5915 			 * NMI-blocked window if the guest runs with IRQs
5916 			 * disabled. So we pull the trigger after 1 s of
5917 			 * futile waiting, but inform the user about this.
5918 			 */
5919 			printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
5920 			       "state on VCPU %d after 1 s timeout\n",
5921 			       __func__, vcpu->vcpu_id);
5922 			vmx->loaded_vmcs->soft_vnmi_blocked = 0;
5923 		}
5924 	}
5925 
5926 	if (exit_fastpath == EXIT_FASTPATH_SKIP_EMUL_INS) {
5927 		kvm_skip_emulated_instruction(vcpu);
5928 		return 1;
5929 	}
5930 
5931 	if (exit_reason >= kvm_vmx_max_exit_handlers)
5932 		goto unexpected_vmexit;
5933 #ifdef CONFIG_RETPOLINE
5934 	if (exit_reason == EXIT_REASON_MSR_WRITE)
5935 		return kvm_emulate_wrmsr(vcpu);
5936 	else if (exit_reason == EXIT_REASON_PREEMPTION_TIMER)
5937 		return handle_preemption_timer(vcpu);
5938 	else if (exit_reason == EXIT_REASON_INTERRUPT_WINDOW)
5939 		return handle_interrupt_window(vcpu);
5940 	else if (exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT)
5941 		return handle_external_interrupt(vcpu);
5942 	else if (exit_reason == EXIT_REASON_HLT)
5943 		return kvm_emulate_halt(vcpu);
5944 	else if (exit_reason == EXIT_REASON_EPT_MISCONFIG)
5945 		return handle_ept_misconfig(vcpu);
5946 #endif
5947 
5948 	exit_reason = array_index_nospec(exit_reason,
5949 					 kvm_vmx_max_exit_handlers);
5950 	if (!kvm_vmx_exit_handlers[exit_reason])
5951 		goto unexpected_vmexit;
5952 
5953 	return kvm_vmx_exit_handlers[exit_reason](vcpu);
5954 
5955 unexpected_vmexit:
5956 	vcpu_unimpl(vcpu, "vmx: unexpected exit reason 0x%x\n", exit_reason);
5957 	dump_vmcs();
5958 	vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5959 	vcpu->run->internal.suberror =
5960 			KVM_INTERNAL_ERROR_UNEXPECTED_EXIT_REASON;
5961 	vcpu->run->internal.ndata = 1;
5962 	vcpu->run->internal.data[0] = exit_reason;
5963 	return 0;
5964 }
5965 
5966 /*
5967  * Software based L1D cache flush which is used when microcode providing
5968  * the cache control MSR is not loaded.
5969  *
5970  * The L1D cache is 32 KiB on Nehalem and later microarchitectures, but to
5971  * flush it is required to read in 64 KiB because the replacement algorithm
5972  * is not exactly LRU. This could be sized at runtime via topology
5973  * information but as all relevant affected CPUs have 32KiB L1D cache size
5974  * there is no point in doing so.
5975  */
5976 static void vmx_l1d_flush(struct kvm_vcpu *vcpu)
5977 {
5978 	int size = PAGE_SIZE << L1D_CACHE_ORDER;
5979 
5980 	/*
5981 	 * This code is only executed when the the flush mode is 'cond' or
5982 	 * 'always'
5983 	 */
5984 	if (static_branch_likely(&vmx_l1d_flush_cond)) {
5985 		bool flush_l1d;
5986 
5987 		/*
5988 		 * Clear the per-vcpu flush bit, it gets set again
5989 		 * either from vcpu_run() or from one of the unsafe
5990 		 * VMEXIT handlers.
5991 		 */
5992 		flush_l1d = vcpu->arch.l1tf_flush_l1d;
5993 		vcpu->arch.l1tf_flush_l1d = false;
5994 
5995 		/*
5996 		 * Clear the per-cpu flush bit, it gets set again from
5997 		 * the interrupt handlers.
5998 		 */
5999 		flush_l1d |= kvm_get_cpu_l1tf_flush_l1d();
6000 		kvm_clear_cpu_l1tf_flush_l1d();
6001 
6002 		if (!flush_l1d)
6003 			return;
6004 	}
6005 
6006 	vcpu->stat.l1d_flush++;
6007 
6008 	if (static_cpu_has(X86_FEATURE_FLUSH_L1D)) {
6009 		wrmsrl(MSR_IA32_FLUSH_CMD, L1D_FLUSH);
6010 		return;
6011 	}
6012 
6013 	asm volatile(
6014 		/* First ensure the pages are in the TLB */
6015 		"xorl	%%eax, %%eax\n"
6016 		".Lpopulate_tlb:\n\t"
6017 		"movzbl	(%[flush_pages], %%" _ASM_AX "), %%ecx\n\t"
6018 		"addl	$4096, %%eax\n\t"
6019 		"cmpl	%%eax, %[size]\n\t"
6020 		"jne	.Lpopulate_tlb\n\t"
6021 		"xorl	%%eax, %%eax\n\t"
6022 		"cpuid\n\t"
6023 		/* Now fill the cache */
6024 		"xorl	%%eax, %%eax\n"
6025 		".Lfill_cache:\n"
6026 		"movzbl	(%[flush_pages], %%" _ASM_AX "), %%ecx\n\t"
6027 		"addl	$64, %%eax\n\t"
6028 		"cmpl	%%eax, %[size]\n\t"
6029 		"jne	.Lfill_cache\n\t"
6030 		"lfence\n"
6031 		:: [flush_pages] "r" (vmx_l1d_flush_pages),
6032 		    [size] "r" (size)
6033 		: "eax", "ebx", "ecx", "edx");
6034 }
6035 
6036 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
6037 {
6038 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6039 	int tpr_threshold;
6040 
6041 	if (is_guest_mode(vcpu) &&
6042 		nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
6043 		return;
6044 
6045 	tpr_threshold = (irr == -1 || tpr < irr) ? 0 : irr;
6046 	if (is_guest_mode(vcpu))
6047 		to_vmx(vcpu)->nested.l1_tpr_threshold = tpr_threshold;
6048 	else
6049 		vmcs_write32(TPR_THRESHOLD, tpr_threshold);
6050 }
6051 
6052 void vmx_set_virtual_apic_mode(struct kvm_vcpu *vcpu)
6053 {
6054 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6055 	u32 sec_exec_control;
6056 
6057 	if (!lapic_in_kernel(vcpu))
6058 		return;
6059 
6060 	if (!flexpriority_enabled &&
6061 	    !cpu_has_vmx_virtualize_x2apic_mode())
6062 		return;
6063 
6064 	/* Postpone execution until vmcs01 is the current VMCS. */
6065 	if (is_guest_mode(vcpu)) {
6066 		vmx->nested.change_vmcs01_virtual_apic_mode = true;
6067 		return;
6068 	}
6069 
6070 	sec_exec_control = secondary_exec_controls_get(vmx);
6071 	sec_exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
6072 			      SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE);
6073 
6074 	switch (kvm_get_apic_mode(vcpu)) {
6075 	case LAPIC_MODE_INVALID:
6076 		WARN_ONCE(true, "Invalid local APIC state");
6077 	case LAPIC_MODE_DISABLED:
6078 		break;
6079 	case LAPIC_MODE_XAPIC:
6080 		if (flexpriority_enabled) {
6081 			sec_exec_control |=
6082 				SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
6083 			vmx_flush_tlb(vcpu, true);
6084 		}
6085 		break;
6086 	case LAPIC_MODE_X2APIC:
6087 		if (cpu_has_vmx_virtualize_x2apic_mode())
6088 			sec_exec_control |=
6089 				SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
6090 		break;
6091 	}
6092 	secondary_exec_controls_set(vmx, sec_exec_control);
6093 
6094 	vmx_update_msr_bitmap(vcpu);
6095 }
6096 
6097 static void vmx_set_apic_access_page_addr(struct kvm_vcpu *vcpu, hpa_t hpa)
6098 {
6099 	if (!is_guest_mode(vcpu)) {
6100 		vmcs_write64(APIC_ACCESS_ADDR, hpa);
6101 		vmx_flush_tlb(vcpu, true);
6102 	}
6103 }
6104 
6105 static void vmx_hwapic_isr_update(struct kvm_vcpu *vcpu, int max_isr)
6106 {
6107 	u16 status;
6108 	u8 old;
6109 
6110 	if (max_isr == -1)
6111 		max_isr = 0;
6112 
6113 	status = vmcs_read16(GUEST_INTR_STATUS);
6114 	old = status >> 8;
6115 	if (max_isr != old) {
6116 		status &= 0xff;
6117 		status |= max_isr << 8;
6118 		vmcs_write16(GUEST_INTR_STATUS, status);
6119 	}
6120 }
6121 
6122 static void vmx_set_rvi(int vector)
6123 {
6124 	u16 status;
6125 	u8 old;
6126 
6127 	if (vector == -1)
6128 		vector = 0;
6129 
6130 	status = vmcs_read16(GUEST_INTR_STATUS);
6131 	old = (u8)status & 0xff;
6132 	if ((u8)vector != old) {
6133 		status &= ~0xff;
6134 		status |= (u8)vector;
6135 		vmcs_write16(GUEST_INTR_STATUS, status);
6136 	}
6137 }
6138 
6139 static void vmx_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr)
6140 {
6141 	/*
6142 	 * When running L2, updating RVI is only relevant when
6143 	 * vmcs12 virtual-interrupt-delivery enabled.
6144 	 * However, it can be enabled only when L1 also
6145 	 * intercepts external-interrupts and in that case
6146 	 * we should not update vmcs02 RVI but instead intercept
6147 	 * interrupt. Therefore, do nothing when running L2.
6148 	 */
6149 	if (!is_guest_mode(vcpu))
6150 		vmx_set_rvi(max_irr);
6151 }
6152 
6153 static int vmx_sync_pir_to_irr(struct kvm_vcpu *vcpu)
6154 {
6155 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6156 	int max_irr;
6157 	bool max_irr_updated;
6158 
6159 	WARN_ON(!vcpu->arch.apicv_active);
6160 	if (pi_test_on(&vmx->pi_desc)) {
6161 		pi_clear_on(&vmx->pi_desc);
6162 		/*
6163 		 * IOMMU can write to PID.ON, so the barrier matters even on UP.
6164 		 * But on x86 this is just a compiler barrier anyway.
6165 		 */
6166 		smp_mb__after_atomic();
6167 		max_irr_updated =
6168 			kvm_apic_update_irr(vcpu, vmx->pi_desc.pir, &max_irr);
6169 
6170 		/*
6171 		 * If we are running L2 and L1 has a new pending interrupt
6172 		 * which can be injected, we should re-evaluate
6173 		 * what should be done with this new L1 interrupt.
6174 		 * If L1 intercepts external-interrupts, we should
6175 		 * exit from L2 to L1. Otherwise, interrupt should be
6176 		 * delivered directly to L2.
6177 		 */
6178 		if (is_guest_mode(vcpu) && max_irr_updated) {
6179 			if (nested_exit_on_intr(vcpu))
6180 				kvm_vcpu_exiting_guest_mode(vcpu);
6181 			else
6182 				kvm_make_request(KVM_REQ_EVENT, vcpu);
6183 		}
6184 	} else {
6185 		max_irr = kvm_lapic_find_highest_irr(vcpu);
6186 	}
6187 	vmx_hwapic_irr_update(vcpu, max_irr);
6188 	return max_irr;
6189 }
6190 
6191 static bool vmx_dy_apicv_has_pending_interrupt(struct kvm_vcpu *vcpu)
6192 {
6193 	struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
6194 
6195 	return pi_test_on(pi_desc) ||
6196 		(pi_test_sn(pi_desc) && !pi_is_pir_empty(pi_desc));
6197 }
6198 
6199 static void vmx_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap)
6200 {
6201 	if (!kvm_vcpu_apicv_active(vcpu))
6202 		return;
6203 
6204 	vmcs_write64(EOI_EXIT_BITMAP0, eoi_exit_bitmap[0]);
6205 	vmcs_write64(EOI_EXIT_BITMAP1, eoi_exit_bitmap[1]);
6206 	vmcs_write64(EOI_EXIT_BITMAP2, eoi_exit_bitmap[2]);
6207 	vmcs_write64(EOI_EXIT_BITMAP3, eoi_exit_bitmap[3]);
6208 }
6209 
6210 static void vmx_apicv_post_state_restore(struct kvm_vcpu *vcpu)
6211 {
6212 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6213 
6214 	pi_clear_on(&vmx->pi_desc);
6215 	memset(vmx->pi_desc.pir, 0, sizeof(vmx->pi_desc.pir));
6216 }
6217 
6218 static void handle_exception_nmi_irqoff(struct vcpu_vmx *vmx)
6219 {
6220 	vmx->exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6221 
6222 	/* if exit due to PF check for async PF */
6223 	if (is_page_fault(vmx->exit_intr_info)) {
6224 		vmx->vcpu.arch.apf.host_apf_reason = kvm_read_and_reset_pf_reason();
6225 	/* Handle machine checks before interrupts are enabled */
6226 	} else if (is_machine_check(vmx->exit_intr_info)) {
6227 		kvm_machine_check();
6228 	/* We need to handle NMIs before interrupts are enabled */
6229 	} else if (is_nmi(vmx->exit_intr_info)) {
6230 		kvm_before_interrupt(&vmx->vcpu);
6231 		asm("int $2");
6232 		kvm_after_interrupt(&vmx->vcpu);
6233 	}
6234 }
6235 
6236 static void handle_external_interrupt_irqoff(struct kvm_vcpu *vcpu)
6237 {
6238 	unsigned int vector;
6239 	unsigned long entry;
6240 #ifdef CONFIG_X86_64
6241 	unsigned long tmp;
6242 #endif
6243 	gate_desc *desc;
6244 	u32 intr_info;
6245 
6246 	intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6247 	if (WARN_ONCE(!is_external_intr(intr_info),
6248 	    "KVM: unexpected VM-Exit interrupt info: 0x%x", intr_info))
6249 		return;
6250 
6251 	vector = intr_info & INTR_INFO_VECTOR_MASK;
6252 	desc = (gate_desc *)host_idt_base + vector;
6253 	entry = gate_offset(desc);
6254 
6255 	kvm_before_interrupt(vcpu);
6256 
6257 	asm volatile(
6258 #ifdef CONFIG_X86_64
6259 		"mov %%" _ASM_SP ", %[sp]\n\t"
6260 		"and $0xfffffffffffffff0, %%" _ASM_SP "\n\t"
6261 		"push $%c[ss]\n\t"
6262 		"push %[sp]\n\t"
6263 #endif
6264 		"pushf\n\t"
6265 		__ASM_SIZE(push) " $%c[cs]\n\t"
6266 		CALL_NOSPEC
6267 		:
6268 #ifdef CONFIG_X86_64
6269 		[sp]"=&r"(tmp),
6270 #endif
6271 		ASM_CALL_CONSTRAINT
6272 		:
6273 		[thunk_target]"r"(entry),
6274 		[ss]"i"(__KERNEL_DS),
6275 		[cs]"i"(__KERNEL_CS)
6276 	);
6277 
6278 	kvm_after_interrupt(vcpu);
6279 }
6280 STACK_FRAME_NON_STANDARD(handle_external_interrupt_irqoff);
6281 
6282 static void vmx_handle_exit_irqoff(struct kvm_vcpu *vcpu,
6283 	enum exit_fastpath_completion *exit_fastpath)
6284 {
6285 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6286 
6287 	if (vmx->exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT)
6288 		handle_external_interrupt_irqoff(vcpu);
6289 	else if (vmx->exit_reason == EXIT_REASON_EXCEPTION_NMI)
6290 		handle_exception_nmi_irqoff(vmx);
6291 	else if (!is_guest_mode(vcpu) &&
6292 		vmx->exit_reason == EXIT_REASON_MSR_WRITE)
6293 		*exit_fastpath = handle_fastpath_set_msr_irqoff(vcpu);
6294 }
6295 
6296 static bool vmx_has_emulated_msr(int index)
6297 {
6298 	switch (index) {
6299 	case MSR_IA32_SMBASE:
6300 		/*
6301 		 * We cannot do SMM unless we can run the guest in big
6302 		 * real mode.
6303 		 */
6304 		return enable_unrestricted_guest || emulate_invalid_guest_state;
6305 	case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
6306 		return nested;
6307 	case MSR_AMD64_VIRT_SPEC_CTRL:
6308 		/* This is AMD only.  */
6309 		return false;
6310 	default:
6311 		return true;
6312 	}
6313 }
6314 
6315 static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx)
6316 {
6317 	u32 exit_intr_info;
6318 	bool unblock_nmi;
6319 	u8 vector;
6320 	bool idtv_info_valid;
6321 
6322 	idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK;
6323 
6324 	if (enable_vnmi) {
6325 		if (vmx->loaded_vmcs->nmi_known_unmasked)
6326 			return;
6327 		/*
6328 		 * Can't use vmx->exit_intr_info since we're not sure what
6329 		 * the exit reason is.
6330 		 */
6331 		exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6332 		unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
6333 		vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
6334 		/*
6335 		 * SDM 3: 27.7.1.2 (September 2008)
6336 		 * Re-set bit "block by NMI" before VM entry if vmexit caused by
6337 		 * a guest IRET fault.
6338 		 * SDM 3: 23.2.2 (September 2008)
6339 		 * Bit 12 is undefined in any of the following cases:
6340 		 *  If the VM exit sets the valid bit in the IDT-vectoring
6341 		 *   information field.
6342 		 *  If the VM exit is due to a double fault.
6343 		 */
6344 		if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
6345 		    vector != DF_VECTOR && !idtv_info_valid)
6346 			vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
6347 				      GUEST_INTR_STATE_NMI);
6348 		else
6349 			vmx->loaded_vmcs->nmi_known_unmasked =
6350 				!(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO)
6351 				  & GUEST_INTR_STATE_NMI);
6352 	} else if (unlikely(vmx->loaded_vmcs->soft_vnmi_blocked))
6353 		vmx->loaded_vmcs->vnmi_blocked_time +=
6354 			ktime_to_ns(ktime_sub(ktime_get(),
6355 					      vmx->loaded_vmcs->entry_time));
6356 }
6357 
6358 static void __vmx_complete_interrupts(struct kvm_vcpu *vcpu,
6359 				      u32 idt_vectoring_info,
6360 				      int instr_len_field,
6361 				      int error_code_field)
6362 {
6363 	u8 vector;
6364 	int type;
6365 	bool idtv_info_valid;
6366 
6367 	idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
6368 
6369 	vcpu->arch.nmi_injected = false;
6370 	kvm_clear_exception_queue(vcpu);
6371 	kvm_clear_interrupt_queue(vcpu);
6372 
6373 	if (!idtv_info_valid)
6374 		return;
6375 
6376 	kvm_make_request(KVM_REQ_EVENT, vcpu);
6377 
6378 	vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
6379 	type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
6380 
6381 	switch (type) {
6382 	case INTR_TYPE_NMI_INTR:
6383 		vcpu->arch.nmi_injected = true;
6384 		/*
6385 		 * SDM 3: 27.7.1.2 (September 2008)
6386 		 * Clear bit "block by NMI" before VM entry if a NMI
6387 		 * delivery faulted.
6388 		 */
6389 		vmx_set_nmi_mask(vcpu, false);
6390 		break;
6391 	case INTR_TYPE_SOFT_EXCEPTION:
6392 		vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
6393 		/* fall through */
6394 	case INTR_TYPE_HARD_EXCEPTION:
6395 		if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
6396 			u32 err = vmcs_read32(error_code_field);
6397 			kvm_requeue_exception_e(vcpu, vector, err);
6398 		} else
6399 			kvm_requeue_exception(vcpu, vector);
6400 		break;
6401 	case INTR_TYPE_SOFT_INTR:
6402 		vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
6403 		/* fall through */
6404 	case INTR_TYPE_EXT_INTR:
6405 		kvm_queue_interrupt(vcpu, vector, type == INTR_TYPE_SOFT_INTR);
6406 		break;
6407 	default:
6408 		break;
6409 	}
6410 }
6411 
6412 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
6413 {
6414 	__vmx_complete_interrupts(&vmx->vcpu, vmx->idt_vectoring_info,
6415 				  VM_EXIT_INSTRUCTION_LEN,
6416 				  IDT_VECTORING_ERROR_CODE);
6417 }
6418 
6419 static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
6420 {
6421 	__vmx_complete_interrupts(vcpu,
6422 				  vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
6423 				  VM_ENTRY_INSTRUCTION_LEN,
6424 				  VM_ENTRY_EXCEPTION_ERROR_CODE);
6425 
6426 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
6427 }
6428 
6429 static void atomic_switch_perf_msrs(struct vcpu_vmx *vmx)
6430 {
6431 	int i, nr_msrs;
6432 	struct perf_guest_switch_msr *msrs;
6433 
6434 	msrs = perf_guest_get_msrs(&nr_msrs);
6435 
6436 	if (!msrs)
6437 		return;
6438 
6439 	for (i = 0; i < nr_msrs; i++)
6440 		if (msrs[i].host == msrs[i].guest)
6441 			clear_atomic_switch_msr(vmx, msrs[i].msr);
6442 		else
6443 			add_atomic_switch_msr(vmx, msrs[i].msr, msrs[i].guest,
6444 					msrs[i].host, false);
6445 }
6446 
6447 static void atomic_switch_umwait_control_msr(struct vcpu_vmx *vmx)
6448 {
6449 	u32 host_umwait_control;
6450 
6451 	if (!vmx_has_waitpkg(vmx))
6452 		return;
6453 
6454 	host_umwait_control = get_umwait_control_msr();
6455 
6456 	if (vmx->msr_ia32_umwait_control != host_umwait_control)
6457 		add_atomic_switch_msr(vmx, MSR_IA32_UMWAIT_CONTROL,
6458 			vmx->msr_ia32_umwait_control,
6459 			host_umwait_control, false);
6460 	else
6461 		clear_atomic_switch_msr(vmx, MSR_IA32_UMWAIT_CONTROL);
6462 }
6463 
6464 static void vmx_update_hv_timer(struct kvm_vcpu *vcpu)
6465 {
6466 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6467 	u64 tscl;
6468 	u32 delta_tsc;
6469 
6470 	if (vmx->req_immediate_exit) {
6471 		vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, 0);
6472 		vmx->loaded_vmcs->hv_timer_soft_disabled = false;
6473 	} else if (vmx->hv_deadline_tsc != -1) {
6474 		tscl = rdtsc();
6475 		if (vmx->hv_deadline_tsc > tscl)
6476 			/* set_hv_timer ensures the delta fits in 32-bits */
6477 			delta_tsc = (u32)((vmx->hv_deadline_tsc - tscl) >>
6478 				cpu_preemption_timer_multi);
6479 		else
6480 			delta_tsc = 0;
6481 
6482 		vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, delta_tsc);
6483 		vmx->loaded_vmcs->hv_timer_soft_disabled = false;
6484 	} else if (!vmx->loaded_vmcs->hv_timer_soft_disabled) {
6485 		vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, -1);
6486 		vmx->loaded_vmcs->hv_timer_soft_disabled = true;
6487 	}
6488 }
6489 
6490 void vmx_update_host_rsp(struct vcpu_vmx *vmx, unsigned long host_rsp)
6491 {
6492 	if (unlikely(host_rsp != vmx->loaded_vmcs->host_state.rsp)) {
6493 		vmx->loaded_vmcs->host_state.rsp = host_rsp;
6494 		vmcs_writel(HOST_RSP, host_rsp);
6495 	}
6496 }
6497 
6498 bool __vmx_vcpu_run(struct vcpu_vmx *vmx, unsigned long *regs, bool launched);
6499 
6500 static void vmx_vcpu_run(struct kvm_vcpu *vcpu)
6501 {
6502 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6503 	unsigned long cr3, cr4;
6504 
6505 	/* Record the guest's net vcpu time for enforced NMI injections. */
6506 	if (unlikely(!enable_vnmi &&
6507 		     vmx->loaded_vmcs->soft_vnmi_blocked))
6508 		vmx->loaded_vmcs->entry_time = ktime_get();
6509 
6510 	/* Don't enter VMX if guest state is invalid, let the exit handler
6511 	   start emulation until we arrive back to a valid state */
6512 	if (vmx->emulation_required)
6513 		return;
6514 
6515 	if (vmx->ple_window_dirty) {
6516 		vmx->ple_window_dirty = false;
6517 		vmcs_write32(PLE_WINDOW, vmx->ple_window);
6518 	}
6519 
6520 	/*
6521 	 * We did this in prepare_switch_to_guest, because it needs to
6522 	 * be within srcu_read_lock.
6523 	 */
6524 	WARN_ON_ONCE(vmx->nested.need_vmcs12_to_shadow_sync);
6525 
6526 	if (kvm_register_is_dirty(vcpu, VCPU_REGS_RSP))
6527 		vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
6528 	if (kvm_register_is_dirty(vcpu, VCPU_REGS_RIP))
6529 		vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
6530 
6531 	cr3 = __get_current_cr3_fast();
6532 	if (unlikely(cr3 != vmx->loaded_vmcs->host_state.cr3)) {
6533 		vmcs_writel(HOST_CR3, cr3);
6534 		vmx->loaded_vmcs->host_state.cr3 = cr3;
6535 	}
6536 
6537 	cr4 = cr4_read_shadow();
6538 	if (unlikely(cr4 != vmx->loaded_vmcs->host_state.cr4)) {
6539 		vmcs_writel(HOST_CR4, cr4);
6540 		vmx->loaded_vmcs->host_state.cr4 = cr4;
6541 	}
6542 
6543 	/* When single-stepping over STI and MOV SS, we must clear the
6544 	 * corresponding interruptibility bits in the guest state. Otherwise
6545 	 * vmentry fails as it then expects bit 14 (BS) in pending debug
6546 	 * exceptions being set, but that's not correct for the guest debugging
6547 	 * case. */
6548 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6549 		vmx_set_interrupt_shadow(vcpu, 0);
6550 
6551 	kvm_load_guest_xsave_state(vcpu);
6552 
6553 	if (static_cpu_has(X86_FEATURE_PKU) &&
6554 	    kvm_read_cr4_bits(vcpu, X86_CR4_PKE) &&
6555 	    vcpu->arch.pkru != vmx->host_pkru)
6556 		__write_pkru(vcpu->arch.pkru);
6557 
6558 	pt_guest_enter(vmx);
6559 
6560 	if (vcpu_to_pmu(vcpu)->version)
6561 		atomic_switch_perf_msrs(vmx);
6562 	atomic_switch_umwait_control_msr(vmx);
6563 
6564 	if (enable_preemption_timer)
6565 		vmx_update_hv_timer(vcpu);
6566 
6567 	if (lapic_in_kernel(vcpu) &&
6568 		vcpu->arch.apic->lapic_timer.timer_advance_ns)
6569 		kvm_wait_lapic_expire(vcpu);
6570 
6571 	/*
6572 	 * If this vCPU has touched SPEC_CTRL, restore the guest's value if
6573 	 * it's non-zero. Since vmentry is serialising on affected CPUs, there
6574 	 * is no need to worry about the conditional branch over the wrmsr
6575 	 * being speculatively taken.
6576 	 */
6577 	x86_spec_ctrl_set_guest(vmx->spec_ctrl, 0);
6578 
6579 	/* L1D Flush includes CPU buffer clear to mitigate MDS */
6580 	if (static_branch_unlikely(&vmx_l1d_should_flush))
6581 		vmx_l1d_flush(vcpu);
6582 	else if (static_branch_unlikely(&mds_user_clear))
6583 		mds_clear_cpu_buffers();
6584 
6585 	if (vcpu->arch.cr2 != read_cr2())
6586 		write_cr2(vcpu->arch.cr2);
6587 
6588 	vmx->fail = __vmx_vcpu_run(vmx, (unsigned long *)&vcpu->arch.regs,
6589 				   vmx->loaded_vmcs->launched);
6590 
6591 	vcpu->arch.cr2 = read_cr2();
6592 
6593 	/*
6594 	 * We do not use IBRS in the kernel. If this vCPU has used the
6595 	 * SPEC_CTRL MSR it may have left it on; save the value and
6596 	 * turn it off. This is much more efficient than blindly adding
6597 	 * it to the atomic save/restore list. Especially as the former
6598 	 * (Saving guest MSRs on vmexit) doesn't even exist in KVM.
6599 	 *
6600 	 * For non-nested case:
6601 	 * If the L01 MSR bitmap does not intercept the MSR, then we need to
6602 	 * save it.
6603 	 *
6604 	 * For nested case:
6605 	 * If the L02 MSR bitmap does not intercept the MSR, then we need to
6606 	 * save it.
6607 	 */
6608 	if (unlikely(!msr_write_intercepted(vcpu, MSR_IA32_SPEC_CTRL)))
6609 		vmx->spec_ctrl = native_read_msr(MSR_IA32_SPEC_CTRL);
6610 
6611 	x86_spec_ctrl_restore_host(vmx->spec_ctrl, 0);
6612 
6613 	/* All fields are clean at this point */
6614 	if (static_branch_unlikely(&enable_evmcs))
6615 		current_evmcs->hv_clean_fields |=
6616 			HV_VMX_ENLIGHTENED_CLEAN_FIELD_ALL;
6617 
6618 	if (static_branch_unlikely(&enable_evmcs))
6619 		current_evmcs->hv_vp_id = vcpu->arch.hyperv.vp_index;
6620 
6621 	/* MSR_IA32_DEBUGCTLMSR is zeroed on vmexit. Restore it if needed */
6622 	if (vmx->host_debugctlmsr)
6623 		update_debugctlmsr(vmx->host_debugctlmsr);
6624 
6625 #ifndef CONFIG_X86_64
6626 	/*
6627 	 * The sysexit path does not restore ds/es, so we must set them to
6628 	 * a reasonable value ourselves.
6629 	 *
6630 	 * We can't defer this to vmx_prepare_switch_to_host() since that
6631 	 * function may be executed in interrupt context, which saves and
6632 	 * restore segments around it, nullifying its effect.
6633 	 */
6634 	loadsegment(ds, __USER_DS);
6635 	loadsegment(es, __USER_DS);
6636 #endif
6637 
6638 	vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
6639 				  | (1 << VCPU_EXREG_RFLAGS)
6640 				  | (1 << VCPU_EXREG_PDPTR)
6641 				  | (1 << VCPU_EXREG_SEGMENTS)
6642 				  | (1 << VCPU_EXREG_CR3));
6643 	vcpu->arch.regs_dirty = 0;
6644 
6645 	pt_guest_exit(vmx);
6646 
6647 	/*
6648 	 * eager fpu is enabled if PKEY is supported and CR4 is switched
6649 	 * back on host, so it is safe to read guest PKRU from current
6650 	 * XSAVE.
6651 	 */
6652 	if (static_cpu_has(X86_FEATURE_PKU) &&
6653 	    kvm_read_cr4_bits(vcpu, X86_CR4_PKE)) {
6654 		vcpu->arch.pkru = rdpkru();
6655 		if (vcpu->arch.pkru != vmx->host_pkru)
6656 			__write_pkru(vmx->host_pkru);
6657 	}
6658 
6659 	kvm_load_host_xsave_state(vcpu);
6660 
6661 	vmx->nested.nested_run_pending = 0;
6662 	vmx->idt_vectoring_info = 0;
6663 
6664 	vmx->exit_reason = vmx->fail ? 0xdead : vmcs_read32(VM_EXIT_REASON);
6665 	if ((u16)vmx->exit_reason == EXIT_REASON_MCE_DURING_VMENTRY)
6666 		kvm_machine_check();
6667 
6668 	if (vmx->fail || (vmx->exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY))
6669 		return;
6670 
6671 	vmx->loaded_vmcs->launched = 1;
6672 	vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
6673 
6674 	vmx_recover_nmi_blocking(vmx);
6675 	vmx_complete_interrupts(vmx);
6676 }
6677 
6678 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
6679 {
6680 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6681 
6682 	if (enable_pml)
6683 		vmx_destroy_pml_buffer(vmx);
6684 	free_vpid(vmx->vpid);
6685 	nested_vmx_free_vcpu(vcpu);
6686 	free_loaded_vmcs(vmx->loaded_vmcs);
6687 }
6688 
6689 static int vmx_create_vcpu(struct kvm_vcpu *vcpu)
6690 {
6691 	struct vcpu_vmx *vmx;
6692 	unsigned long *msr_bitmap;
6693 	int i, cpu, err;
6694 
6695 	BUILD_BUG_ON(offsetof(struct vcpu_vmx, vcpu) != 0);
6696 	vmx = to_vmx(vcpu);
6697 
6698 	err = -ENOMEM;
6699 
6700 	vmx->vpid = allocate_vpid();
6701 
6702 	/*
6703 	 * If PML is turned on, failure on enabling PML just results in failure
6704 	 * of creating the vcpu, therefore we can simplify PML logic (by
6705 	 * avoiding dealing with cases, such as enabling PML partially on vcpus
6706 	 * for the guest), etc.
6707 	 */
6708 	if (enable_pml) {
6709 		vmx->pml_pg = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
6710 		if (!vmx->pml_pg)
6711 			goto free_vpid;
6712 	}
6713 
6714 	BUILD_BUG_ON(ARRAY_SIZE(vmx_msr_index) != NR_SHARED_MSRS);
6715 
6716 	for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i) {
6717 		u32 index = vmx_msr_index[i];
6718 		u32 data_low, data_high;
6719 		int j = vmx->nmsrs;
6720 
6721 		if (rdmsr_safe(index, &data_low, &data_high) < 0)
6722 			continue;
6723 		if (wrmsr_safe(index, data_low, data_high) < 0)
6724 			continue;
6725 
6726 		vmx->guest_msrs[j].index = i;
6727 		vmx->guest_msrs[j].data = 0;
6728 		switch (index) {
6729 		case MSR_IA32_TSX_CTRL:
6730 			/*
6731 			 * No need to pass TSX_CTRL_CPUID_CLEAR through, so
6732 			 * let's avoid changing CPUID bits under the host
6733 			 * kernel's feet.
6734 			 */
6735 			vmx->guest_msrs[j].mask = ~(u64)TSX_CTRL_CPUID_CLEAR;
6736 			break;
6737 		default:
6738 			vmx->guest_msrs[j].mask = -1ull;
6739 			break;
6740 		}
6741 		++vmx->nmsrs;
6742 	}
6743 
6744 	err = alloc_loaded_vmcs(&vmx->vmcs01);
6745 	if (err < 0)
6746 		goto free_pml;
6747 
6748 	msr_bitmap = vmx->vmcs01.msr_bitmap;
6749 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_TSC, MSR_TYPE_R);
6750 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_FS_BASE, MSR_TYPE_RW);
6751 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_GS_BASE, MSR_TYPE_RW);
6752 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_KERNEL_GS_BASE, MSR_TYPE_RW);
6753 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_CS, MSR_TYPE_RW);
6754 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_ESP, MSR_TYPE_RW);
6755 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_EIP, MSR_TYPE_RW);
6756 	if (kvm_cstate_in_guest(vcpu->kvm)) {
6757 		vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C1_RES, MSR_TYPE_R);
6758 		vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C3_RESIDENCY, MSR_TYPE_R);
6759 		vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C6_RESIDENCY, MSR_TYPE_R);
6760 		vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C7_RESIDENCY, MSR_TYPE_R);
6761 	}
6762 	vmx->msr_bitmap_mode = 0;
6763 
6764 	vmx->loaded_vmcs = &vmx->vmcs01;
6765 	cpu = get_cpu();
6766 	vmx_vcpu_load(vcpu, cpu);
6767 	vcpu->cpu = cpu;
6768 	init_vmcs(vmx);
6769 	vmx_vcpu_put(vcpu);
6770 	put_cpu();
6771 	if (cpu_need_virtualize_apic_accesses(vcpu)) {
6772 		err = alloc_apic_access_page(vcpu->kvm);
6773 		if (err)
6774 			goto free_vmcs;
6775 	}
6776 
6777 	if (enable_ept && !enable_unrestricted_guest) {
6778 		err = init_rmode_identity_map(vcpu->kvm);
6779 		if (err)
6780 			goto free_vmcs;
6781 	}
6782 
6783 	if (nested)
6784 		nested_vmx_setup_ctls_msrs(&vmx->nested.msrs,
6785 					   vmx_capability.ept);
6786 	else
6787 		memset(&vmx->nested.msrs, 0, sizeof(vmx->nested.msrs));
6788 
6789 	vmx->nested.posted_intr_nv = -1;
6790 	vmx->nested.current_vmptr = -1ull;
6791 
6792 	vcpu->arch.microcode_version = 0x100000000ULL;
6793 	vmx->msr_ia32_feature_control_valid_bits = FEAT_CTL_LOCKED;
6794 
6795 	/*
6796 	 * Enforce invariant: pi_desc.nv is always either POSTED_INTR_VECTOR
6797 	 * or POSTED_INTR_WAKEUP_VECTOR.
6798 	 */
6799 	vmx->pi_desc.nv = POSTED_INTR_VECTOR;
6800 	vmx->pi_desc.sn = 1;
6801 
6802 	vmx->ept_pointer = INVALID_PAGE;
6803 
6804 	return 0;
6805 
6806 free_vmcs:
6807 	free_loaded_vmcs(vmx->loaded_vmcs);
6808 free_pml:
6809 	vmx_destroy_pml_buffer(vmx);
6810 free_vpid:
6811 	free_vpid(vmx->vpid);
6812 	return err;
6813 }
6814 
6815 #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"
6816 #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"
6817 
6818 static int vmx_vm_init(struct kvm *kvm)
6819 {
6820 	spin_lock_init(&to_kvm_vmx(kvm)->ept_pointer_lock);
6821 
6822 	if (!ple_gap)
6823 		kvm->arch.pause_in_guest = true;
6824 
6825 	if (boot_cpu_has(X86_BUG_L1TF) && enable_ept) {
6826 		switch (l1tf_mitigation) {
6827 		case L1TF_MITIGATION_OFF:
6828 		case L1TF_MITIGATION_FLUSH_NOWARN:
6829 			/* 'I explicitly don't care' is set */
6830 			break;
6831 		case L1TF_MITIGATION_FLUSH:
6832 		case L1TF_MITIGATION_FLUSH_NOSMT:
6833 		case L1TF_MITIGATION_FULL:
6834 			/*
6835 			 * Warn upon starting the first VM in a potentially
6836 			 * insecure environment.
6837 			 */
6838 			if (sched_smt_active())
6839 				pr_warn_once(L1TF_MSG_SMT);
6840 			if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_NEVER)
6841 				pr_warn_once(L1TF_MSG_L1D);
6842 			break;
6843 		case L1TF_MITIGATION_FULL_FORCE:
6844 			/* Flush is enforced */
6845 			break;
6846 		}
6847 	}
6848 	kvm_apicv_init(kvm, enable_apicv);
6849 	return 0;
6850 }
6851 
6852 static int __init vmx_check_processor_compat(void)
6853 {
6854 	struct vmcs_config vmcs_conf;
6855 	struct vmx_capability vmx_cap;
6856 
6857 	if (!this_cpu_has(X86_FEATURE_MSR_IA32_FEAT_CTL) ||
6858 	    !this_cpu_has(X86_FEATURE_VMX)) {
6859 		pr_err("kvm: VMX is disabled on CPU %d\n", smp_processor_id());
6860 		return -EIO;
6861 	}
6862 
6863 	if (setup_vmcs_config(&vmcs_conf, &vmx_cap) < 0)
6864 		return -EIO;
6865 	if (nested)
6866 		nested_vmx_setup_ctls_msrs(&vmcs_conf.nested, vmx_cap.ept);
6867 	if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
6868 		printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
6869 				smp_processor_id());
6870 		return -EIO;
6871 	}
6872 	return 0;
6873 }
6874 
6875 static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
6876 {
6877 	u8 cache;
6878 	u64 ipat = 0;
6879 
6880 	/* We wanted to honor guest CD/MTRR/PAT, but doing so could result in
6881 	 * memory aliases with conflicting memory types and sometimes MCEs.
6882 	 * We have to be careful as to what are honored and when.
6883 	 *
6884 	 * For MMIO, guest CD/MTRR are ignored.  The EPT memory type is set to
6885 	 * UC.  The effective memory type is UC or WC depending on guest PAT.
6886 	 * This was historically the source of MCEs and we want to be
6887 	 * conservative.
6888 	 *
6889 	 * When there is no need to deal with noncoherent DMA (e.g., no VT-d
6890 	 * or VT-d has snoop control), guest CD/MTRR/PAT are all ignored.  The
6891 	 * EPT memory type is set to WB.  The effective memory type is forced
6892 	 * WB.
6893 	 *
6894 	 * Otherwise, we trust guest.  Guest CD/MTRR/PAT are all honored.  The
6895 	 * EPT memory type is used to emulate guest CD/MTRR.
6896 	 */
6897 
6898 	if (is_mmio) {
6899 		cache = MTRR_TYPE_UNCACHABLE;
6900 		goto exit;
6901 	}
6902 
6903 	if (!kvm_arch_has_noncoherent_dma(vcpu->kvm)) {
6904 		ipat = VMX_EPT_IPAT_BIT;
6905 		cache = MTRR_TYPE_WRBACK;
6906 		goto exit;
6907 	}
6908 
6909 	if (kvm_read_cr0(vcpu) & X86_CR0_CD) {
6910 		ipat = VMX_EPT_IPAT_BIT;
6911 		if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
6912 			cache = MTRR_TYPE_WRBACK;
6913 		else
6914 			cache = MTRR_TYPE_UNCACHABLE;
6915 		goto exit;
6916 	}
6917 
6918 	cache = kvm_mtrr_get_guest_memory_type(vcpu, gfn);
6919 
6920 exit:
6921 	return (cache << VMX_EPT_MT_EPTE_SHIFT) | ipat;
6922 }
6923 
6924 static void vmcs_set_secondary_exec_control(struct vcpu_vmx *vmx)
6925 {
6926 	/*
6927 	 * These bits in the secondary execution controls field
6928 	 * are dynamic, the others are mostly based on the hypervisor
6929 	 * architecture and the guest's CPUID.  Do not touch the
6930 	 * dynamic bits.
6931 	 */
6932 	u32 mask =
6933 		SECONDARY_EXEC_SHADOW_VMCS |
6934 		SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
6935 		SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
6936 		SECONDARY_EXEC_DESC;
6937 
6938 	u32 new_ctl = vmx->secondary_exec_control;
6939 	u32 cur_ctl = secondary_exec_controls_get(vmx);
6940 
6941 	secondary_exec_controls_set(vmx, (new_ctl & ~mask) | (cur_ctl & mask));
6942 }
6943 
6944 /*
6945  * Generate MSR_IA32_VMX_CR{0,4}_FIXED1 according to CPUID. Only set bits
6946  * (indicating "allowed-1") if they are supported in the guest's CPUID.
6947  */
6948 static void nested_vmx_cr_fixed1_bits_update(struct kvm_vcpu *vcpu)
6949 {
6950 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6951 	struct kvm_cpuid_entry2 *entry;
6952 
6953 	vmx->nested.msrs.cr0_fixed1 = 0xffffffff;
6954 	vmx->nested.msrs.cr4_fixed1 = X86_CR4_PCE;
6955 
6956 #define cr4_fixed1_update(_cr4_mask, _reg, _cpuid_mask) do {		\
6957 	if (entry && (entry->_reg & (_cpuid_mask)))			\
6958 		vmx->nested.msrs.cr4_fixed1 |= (_cr4_mask);	\
6959 } while (0)
6960 
6961 	entry = kvm_find_cpuid_entry(vcpu, 0x1, 0);
6962 	cr4_fixed1_update(X86_CR4_VME,        edx, feature_bit(VME));
6963 	cr4_fixed1_update(X86_CR4_PVI,        edx, feature_bit(VME));
6964 	cr4_fixed1_update(X86_CR4_TSD,        edx, feature_bit(TSC));
6965 	cr4_fixed1_update(X86_CR4_DE,         edx, feature_bit(DE));
6966 	cr4_fixed1_update(X86_CR4_PSE,        edx, feature_bit(PSE));
6967 	cr4_fixed1_update(X86_CR4_PAE,        edx, feature_bit(PAE));
6968 	cr4_fixed1_update(X86_CR4_MCE,        edx, feature_bit(MCE));
6969 	cr4_fixed1_update(X86_CR4_PGE,        edx, feature_bit(PGE));
6970 	cr4_fixed1_update(X86_CR4_OSFXSR,     edx, feature_bit(FXSR));
6971 	cr4_fixed1_update(X86_CR4_OSXMMEXCPT, edx, feature_bit(XMM));
6972 	cr4_fixed1_update(X86_CR4_VMXE,       ecx, feature_bit(VMX));
6973 	cr4_fixed1_update(X86_CR4_SMXE,       ecx, feature_bit(SMX));
6974 	cr4_fixed1_update(X86_CR4_PCIDE,      ecx, feature_bit(PCID));
6975 	cr4_fixed1_update(X86_CR4_OSXSAVE,    ecx, feature_bit(XSAVE));
6976 
6977 	entry = kvm_find_cpuid_entry(vcpu, 0x7, 0);
6978 	cr4_fixed1_update(X86_CR4_FSGSBASE,   ebx, feature_bit(FSGSBASE));
6979 	cr4_fixed1_update(X86_CR4_SMEP,       ebx, feature_bit(SMEP));
6980 	cr4_fixed1_update(X86_CR4_SMAP,       ebx, feature_bit(SMAP));
6981 	cr4_fixed1_update(X86_CR4_PKE,        ecx, feature_bit(PKU));
6982 	cr4_fixed1_update(X86_CR4_UMIP,       ecx, feature_bit(UMIP));
6983 	cr4_fixed1_update(X86_CR4_LA57,       ecx, feature_bit(LA57));
6984 
6985 #undef cr4_fixed1_update
6986 }
6987 
6988 static void nested_vmx_entry_exit_ctls_update(struct kvm_vcpu *vcpu)
6989 {
6990 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6991 
6992 	if (kvm_mpx_supported()) {
6993 		bool mpx_enabled = guest_cpuid_has(vcpu, X86_FEATURE_MPX);
6994 
6995 		if (mpx_enabled) {
6996 			vmx->nested.msrs.entry_ctls_high |= VM_ENTRY_LOAD_BNDCFGS;
6997 			vmx->nested.msrs.exit_ctls_high |= VM_EXIT_CLEAR_BNDCFGS;
6998 		} else {
6999 			vmx->nested.msrs.entry_ctls_high &= ~VM_ENTRY_LOAD_BNDCFGS;
7000 			vmx->nested.msrs.exit_ctls_high &= ~VM_EXIT_CLEAR_BNDCFGS;
7001 		}
7002 	}
7003 }
7004 
7005 static void update_intel_pt_cfg(struct kvm_vcpu *vcpu)
7006 {
7007 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7008 	struct kvm_cpuid_entry2 *best = NULL;
7009 	int i;
7010 
7011 	for (i = 0; i < PT_CPUID_LEAVES; i++) {
7012 		best = kvm_find_cpuid_entry(vcpu, 0x14, i);
7013 		if (!best)
7014 			return;
7015 		vmx->pt_desc.caps[CPUID_EAX + i*PT_CPUID_REGS_NUM] = best->eax;
7016 		vmx->pt_desc.caps[CPUID_EBX + i*PT_CPUID_REGS_NUM] = best->ebx;
7017 		vmx->pt_desc.caps[CPUID_ECX + i*PT_CPUID_REGS_NUM] = best->ecx;
7018 		vmx->pt_desc.caps[CPUID_EDX + i*PT_CPUID_REGS_NUM] = best->edx;
7019 	}
7020 
7021 	/* Get the number of configurable Address Ranges for filtering */
7022 	vmx->pt_desc.addr_range = intel_pt_validate_cap(vmx->pt_desc.caps,
7023 						PT_CAP_num_address_ranges);
7024 
7025 	/* Initialize and clear the no dependency bits */
7026 	vmx->pt_desc.ctl_bitmask = ~(RTIT_CTL_TRACEEN | RTIT_CTL_OS |
7027 			RTIT_CTL_USR | RTIT_CTL_TSC_EN | RTIT_CTL_DISRETC);
7028 
7029 	/*
7030 	 * If CPUID.(EAX=14H,ECX=0):EBX[0]=1 CR3Filter can be set otherwise
7031 	 * will inject an #GP
7032 	 */
7033 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_cr3_filtering))
7034 		vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_CR3EN;
7035 
7036 	/*
7037 	 * If CPUID.(EAX=14H,ECX=0):EBX[1]=1 CYCEn, CycThresh and
7038 	 * PSBFreq can be set
7039 	 */
7040 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc))
7041 		vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_CYCLEACC |
7042 				RTIT_CTL_CYC_THRESH | RTIT_CTL_PSB_FREQ);
7043 
7044 	/*
7045 	 * If CPUID.(EAX=14H,ECX=0):EBX[3]=1 MTCEn BranchEn and
7046 	 * MTCFreq can be set
7047 	 */
7048 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc))
7049 		vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_MTC_EN |
7050 				RTIT_CTL_BRANCH_EN | RTIT_CTL_MTC_RANGE);
7051 
7052 	/* If CPUID.(EAX=14H,ECX=0):EBX[4]=1 FUPonPTW and PTWEn can be set */
7053 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_ptwrite))
7054 		vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_FUP_ON_PTW |
7055 							RTIT_CTL_PTW_EN);
7056 
7057 	/* If CPUID.(EAX=14H,ECX=0):EBX[5]=1 PwrEvEn can be set */
7058 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_power_event_trace))
7059 		vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_PWR_EVT_EN;
7060 
7061 	/* If CPUID.(EAX=14H,ECX=0):ECX[0]=1 ToPA can be set */
7062 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_topa_output))
7063 		vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_TOPA;
7064 
7065 	/* If CPUID.(EAX=14H,ECX=0):ECX[3]=1 FabircEn can be set */
7066 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_output_subsys))
7067 		vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_FABRIC_EN;
7068 
7069 	/* unmask address range configure area */
7070 	for (i = 0; i < vmx->pt_desc.addr_range; i++)
7071 		vmx->pt_desc.ctl_bitmask &= ~(0xfULL << (32 + i * 4));
7072 }
7073 
7074 static void vmx_cpuid_update(struct kvm_vcpu *vcpu)
7075 {
7076 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7077 
7078 	/* xsaves_enabled is recomputed in vmx_compute_secondary_exec_control(). */
7079 	vcpu->arch.xsaves_enabled = false;
7080 
7081 	if (cpu_has_secondary_exec_ctrls()) {
7082 		vmx_compute_secondary_exec_control(vmx);
7083 		vmcs_set_secondary_exec_control(vmx);
7084 	}
7085 
7086 	if (nested_vmx_allowed(vcpu))
7087 		to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
7088 			FEAT_CTL_VMX_ENABLED_INSIDE_SMX |
7089 			FEAT_CTL_VMX_ENABLED_OUTSIDE_SMX;
7090 	else
7091 		to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
7092 			~(FEAT_CTL_VMX_ENABLED_INSIDE_SMX |
7093 			  FEAT_CTL_VMX_ENABLED_OUTSIDE_SMX);
7094 
7095 	if (nested_vmx_allowed(vcpu)) {
7096 		nested_vmx_cr_fixed1_bits_update(vcpu);
7097 		nested_vmx_entry_exit_ctls_update(vcpu);
7098 	}
7099 
7100 	if (boot_cpu_has(X86_FEATURE_INTEL_PT) &&
7101 			guest_cpuid_has(vcpu, X86_FEATURE_INTEL_PT))
7102 		update_intel_pt_cfg(vcpu);
7103 
7104 	if (boot_cpu_has(X86_FEATURE_RTM)) {
7105 		struct shared_msr_entry *msr;
7106 		msr = find_msr_entry(vmx, MSR_IA32_TSX_CTRL);
7107 		if (msr) {
7108 			bool enabled = guest_cpuid_has(vcpu, X86_FEATURE_RTM);
7109 			vmx_set_guest_msr(vmx, msr, enabled ? 0 : TSX_CTRL_RTM_DISABLE);
7110 		}
7111 	}
7112 }
7113 
7114 static __init void vmx_set_cpu_caps(void)
7115 {
7116 	kvm_set_cpu_caps();
7117 
7118 	/* CPUID 0x1 */
7119 	if (nested)
7120 		kvm_cpu_cap_set(X86_FEATURE_VMX);
7121 
7122 	/* CPUID 0x7 */
7123 	if (kvm_mpx_supported())
7124 		kvm_cpu_cap_check_and_set(X86_FEATURE_MPX);
7125 	if (cpu_has_vmx_invpcid())
7126 		kvm_cpu_cap_check_and_set(X86_FEATURE_INVPCID);
7127 	if (vmx_pt_mode_is_host_guest())
7128 		kvm_cpu_cap_check_and_set(X86_FEATURE_INTEL_PT);
7129 
7130 	/* PKU is not yet implemented for shadow paging. */
7131 	if (enable_ept && boot_cpu_has(X86_FEATURE_OSPKE))
7132 		kvm_cpu_cap_check_and_set(X86_FEATURE_PKU);
7133 
7134 	if (vmx_umip_emulated())
7135 		kvm_cpu_cap_set(X86_FEATURE_UMIP);
7136 
7137 	/* CPUID 0xD.1 */
7138 	supported_xss = 0;
7139 	if (!vmx_xsaves_supported())
7140 		kvm_cpu_cap_clear(X86_FEATURE_XSAVES);
7141 
7142 	/* CPUID 0x80000001 */
7143 	if (!cpu_has_vmx_rdtscp())
7144 		kvm_cpu_cap_clear(X86_FEATURE_RDTSCP);
7145 }
7146 
7147 static void vmx_request_immediate_exit(struct kvm_vcpu *vcpu)
7148 {
7149 	to_vmx(vcpu)->req_immediate_exit = true;
7150 }
7151 
7152 static int vmx_check_intercept_io(struct kvm_vcpu *vcpu,
7153 				  struct x86_instruction_info *info)
7154 {
7155 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
7156 	unsigned short port;
7157 	bool intercept;
7158 	int size;
7159 
7160 	if (info->intercept == x86_intercept_in ||
7161 	    info->intercept == x86_intercept_ins) {
7162 		port = info->src_val;
7163 		size = info->dst_bytes;
7164 	} else {
7165 		port = info->dst_val;
7166 		size = info->src_bytes;
7167 	}
7168 
7169 	/*
7170 	 * If the 'use IO bitmaps' VM-execution control is 0, IO instruction
7171 	 * VM-exits depend on the 'unconditional IO exiting' VM-execution
7172 	 * control.
7173 	 *
7174 	 * Otherwise, IO instruction VM-exits are controlled by the IO bitmaps.
7175 	 */
7176 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
7177 		intercept = nested_cpu_has(vmcs12,
7178 					   CPU_BASED_UNCOND_IO_EXITING);
7179 	else
7180 		intercept = nested_vmx_check_io_bitmaps(vcpu, port, size);
7181 
7182 	/* FIXME: produce nested vmexit and return X86EMUL_INTERCEPTED.  */
7183 	return intercept ? X86EMUL_UNHANDLEABLE : X86EMUL_CONTINUE;
7184 }
7185 
7186 static int vmx_check_intercept(struct kvm_vcpu *vcpu,
7187 			       struct x86_instruction_info *info,
7188 			       enum x86_intercept_stage stage,
7189 			       struct x86_exception *exception)
7190 {
7191 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
7192 
7193 	switch (info->intercept) {
7194 	/*
7195 	 * RDPID causes #UD if disabled through secondary execution controls.
7196 	 * Because it is marked as EmulateOnUD, we need to intercept it here.
7197 	 */
7198 	case x86_intercept_rdtscp:
7199 		if (!nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDTSCP)) {
7200 			exception->vector = UD_VECTOR;
7201 			exception->error_code_valid = false;
7202 			return X86EMUL_PROPAGATE_FAULT;
7203 		}
7204 		break;
7205 
7206 	case x86_intercept_in:
7207 	case x86_intercept_ins:
7208 	case x86_intercept_out:
7209 	case x86_intercept_outs:
7210 		return vmx_check_intercept_io(vcpu, info);
7211 
7212 	case x86_intercept_lgdt:
7213 	case x86_intercept_lidt:
7214 	case x86_intercept_lldt:
7215 	case x86_intercept_ltr:
7216 	case x86_intercept_sgdt:
7217 	case x86_intercept_sidt:
7218 	case x86_intercept_sldt:
7219 	case x86_intercept_str:
7220 		if (!nested_cpu_has2(vmcs12, SECONDARY_EXEC_DESC))
7221 			return X86EMUL_CONTINUE;
7222 
7223 		/* FIXME: produce nested vmexit and return X86EMUL_INTERCEPTED.  */
7224 		break;
7225 
7226 	/* TODO: check more intercepts... */
7227 	default:
7228 		break;
7229 	}
7230 
7231 	return X86EMUL_UNHANDLEABLE;
7232 }
7233 
7234 #ifdef CONFIG_X86_64
7235 /* (a << shift) / divisor, return 1 if overflow otherwise 0 */
7236 static inline int u64_shl_div_u64(u64 a, unsigned int shift,
7237 				  u64 divisor, u64 *result)
7238 {
7239 	u64 low = a << shift, high = a >> (64 - shift);
7240 
7241 	/* To avoid the overflow on divq */
7242 	if (high >= divisor)
7243 		return 1;
7244 
7245 	/* Low hold the result, high hold rem which is discarded */
7246 	asm("divq %2\n\t" : "=a" (low), "=d" (high) :
7247 	    "rm" (divisor), "0" (low), "1" (high));
7248 	*result = low;
7249 
7250 	return 0;
7251 }
7252 
7253 static int vmx_set_hv_timer(struct kvm_vcpu *vcpu, u64 guest_deadline_tsc,
7254 			    bool *expired)
7255 {
7256 	struct vcpu_vmx *vmx;
7257 	u64 tscl, guest_tscl, delta_tsc, lapic_timer_advance_cycles;
7258 	struct kvm_timer *ktimer = &vcpu->arch.apic->lapic_timer;
7259 
7260 	if (kvm_mwait_in_guest(vcpu->kvm) ||
7261 		kvm_can_post_timer_interrupt(vcpu))
7262 		return -EOPNOTSUPP;
7263 
7264 	vmx = to_vmx(vcpu);
7265 	tscl = rdtsc();
7266 	guest_tscl = kvm_read_l1_tsc(vcpu, tscl);
7267 	delta_tsc = max(guest_deadline_tsc, guest_tscl) - guest_tscl;
7268 	lapic_timer_advance_cycles = nsec_to_cycles(vcpu,
7269 						    ktimer->timer_advance_ns);
7270 
7271 	if (delta_tsc > lapic_timer_advance_cycles)
7272 		delta_tsc -= lapic_timer_advance_cycles;
7273 	else
7274 		delta_tsc = 0;
7275 
7276 	/* Convert to host delta tsc if tsc scaling is enabled */
7277 	if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio &&
7278 	    delta_tsc && u64_shl_div_u64(delta_tsc,
7279 				kvm_tsc_scaling_ratio_frac_bits,
7280 				vcpu->arch.tsc_scaling_ratio, &delta_tsc))
7281 		return -ERANGE;
7282 
7283 	/*
7284 	 * If the delta tsc can't fit in the 32 bit after the multi shift,
7285 	 * we can't use the preemption timer.
7286 	 * It's possible that it fits on later vmentries, but checking
7287 	 * on every vmentry is costly so we just use an hrtimer.
7288 	 */
7289 	if (delta_tsc >> (cpu_preemption_timer_multi + 32))
7290 		return -ERANGE;
7291 
7292 	vmx->hv_deadline_tsc = tscl + delta_tsc;
7293 	*expired = !delta_tsc;
7294 	return 0;
7295 }
7296 
7297 static void vmx_cancel_hv_timer(struct kvm_vcpu *vcpu)
7298 {
7299 	to_vmx(vcpu)->hv_deadline_tsc = -1;
7300 }
7301 #endif
7302 
7303 static void vmx_sched_in(struct kvm_vcpu *vcpu, int cpu)
7304 {
7305 	if (!kvm_pause_in_guest(vcpu->kvm))
7306 		shrink_ple_window(vcpu);
7307 }
7308 
7309 static void vmx_slot_enable_log_dirty(struct kvm *kvm,
7310 				     struct kvm_memory_slot *slot)
7311 {
7312 	if (!kvm_dirty_log_manual_protect_and_init_set(kvm))
7313 		kvm_mmu_slot_leaf_clear_dirty(kvm, slot);
7314 	kvm_mmu_slot_largepage_remove_write_access(kvm, slot);
7315 }
7316 
7317 static void vmx_slot_disable_log_dirty(struct kvm *kvm,
7318 				       struct kvm_memory_slot *slot)
7319 {
7320 	kvm_mmu_slot_set_dirty(kvm, slot);
7321 }
7322 
7323 static void vmx_flush_log_dirty(struct kvm *kvm)
7324 {
7325 	kvm_flush_pml_buffers(kvm);
7326 }
7327 
7328 static int vmx_write_pml_buffer(struct kvm_vcpu *vcpu)
7329 {
7330 	struct vmcs12 *vmcs12;
7331 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7332 	gpa_t gpa, dst;
7333 
7334 	if (is_guest_mode(vcpu)) {
7335 		WARN_ON_ONCE(vmx->nested.pml_full);
7336 
7337 		/*
7338 		 * Check if PML is enabled for the nested guest.
7339 		 * Whether eptp bit 6 is set is already checked
7340 		 * as part of A/D emulation.
7341 		 */
7342 		vmcs12 = get_vmcs12(vcpu);
7343 		if (!nested_cpu_has_pml(vmcs12))
7344 			return 0;
7345 
7346 		if (vmcs12->guest_pml_index >= PML_ENTITY_NUM) {
7347 			vmx->nested.pml_full = true;
7348 			return 1;
7349 		}
7350 
7351 		gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS) & ~0xFFFull;
7352 		dst = vmcs12->pml_address + sizeof(u64) * vmcs12->guest_pml_index;
7353 
7354 		if (kvm_write_guest_page(vcpu->kvm, gpa_to_gfn(dst), &gpa,
7355 					 offset_in_page(dst), sizeof(gpa)))
7356 			return 0;
7357 
7358 		vmcs12->guest_pml_index--;
7359 	}
7360 
7361 	return 0;
7362 }
7363 
7364 static void vmx_enable_log_dirty_pt_masked(struct kvm *kvm,
7365 					   struct kvm_memory_slot *memslot,
7366 					   gfn_t offset, unsigned long mask)
7367 {
7368 	kvm_mmu_clear_dirty_pt_masked(kvm, memslot, offset, mask);
7369 }
7370 
7371 static void __pi_post_block(struct kvm_vcpu *vcpu)
7372 {
7373 	struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
7374 	struct pi_desc old, new;
7375 	unsigned int dest;
7376 
7377 	do {
7378 		old.control = new.control = pi_desc->control;
7379 		WARN(old.nv != POSTED_INTR_WAKEUP_VECTOR,
7380 		     "Wakeup handler not enabled while the VCPU is blocked\n");
7381 
7382 		dest = cpu_physical_id(vcpu->cpu);
7383 
7384 		if (x2apic_enabled())
7385 			new.ndst = dest;
7386 		else
7387 			new.ndst = (dest << 8) & 0xFF00;
7388 
7389 		/* set 'NV' to 'notification vector' */
7390 		new.nv = POSTED_INTR_VECTOR;
7391 	} while (cmpxchg64(&pi_desc->control, old.control,
7392 			   new.control) != old.control);
7393 
7394 	if (!WARN_ON_ONCE(vcpu->pre_pcpu == -1)) {
7395 		spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7396 		list_del(&vcpu->blocked_vcpu_list);
7397 		spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7398 		vcpu->pre_pcpu = -1;
7399 	}
7400 }
7401 
7402 /*
7403  * This routine does the following things for vCPU which is going
7404  * to be blocked if VT-d PI is enabled.
7405  * - Store the vCPU to the wakeup list, so when interrupts happen
7406  *   we can find the right vCPU to wake up.
7407  * - Change the Posted-interrupt descriptor as below:
7408  *      'NDST' <-- vcpu->pre_pcpu
7409  *      'NV' <-- POSTED_INTR_WAKEUP_VECTOR
7410  * - If 'ON' is set during this process, which means at least one
7411  *   interrupt is posted for this vCPU, we cannot block it, in
7412  *   this case, return 1, otherwise, return 0.
7413  *
7414  */
7415 static int pi_pre_block(struct kvm_vcpu *vcpu)
7416 {
7417 	unsigned int dest;
7418 	struct pi_desc old, new;
7419 	struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
7420 
7421 	if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
7422 		!irq_remapping_cap(IRQ_POSTING_CAP)  ||
7423 		!kvm_vcpu_apicv_active(vcpu))
7424 		return 0;
7425 
7426 	WARN_ON(irqs_disabled());
7427 	local_irq_disable();
7428 	if (!WARN_ON_ONCE(vcpu->pre_pcpu != -1)) {
7429 		vcpu->pre_pcpu = vcpu->cpu;
7430 		spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7431 		list_add_tail(&vcpu->blocked_vcpu_list,
7432 			      &per_cpu(blocked_vcpu_on_cpu,
7433 				       vcpu->pre_pcpu));
7434 		spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7435 	}
7436 
7437 	do {
7438 		old.control = new.control = pi_desc->control;
7439 
7440 		WARN((pi_desc->sn == 1),
7441 		     "Warning: SN field of posted-interrupts "
7442 		     "is set before blocking\n");
7443 
7444 		/*
7445 		 * Since vCPU can be preempted during this process,
7446 		 * vcpu->cpu could be different with pre_pcpu, we
7447 		 * need to set pre_pcpu as the destination of wakeup
7448 		 * notification event, then we can find the right vCPU
7449 		 * to wakeup in wakeup handler if interrupts happen
7450 		 * when the vCPU is in blocked state.
7451 		 */
7452 		dest = cpu_physical_id(vcpu->pre_pcpu);
7453 
7454 		if (x2apic_enabled())
7455 			new.ndst = dest;
7456 		else
7457 			new.ndst = (dest << 8) & 0xFF00;
7458 
7459 		/* set 'NV' to 'wakeup vector' */
7460 		new.nv = POSTED_INTR_WAKEUP_VECTOR;
7461 	} while (cmpxchg64(&pi_desc->control, old.control,
7462 			   new.control) != old.control);
7463 
7464 	/* We should not block the vCPU if an interrupt is posted for it.  */
7465 	if (pi_test_on(pi_desc) == 1)
7466 		__pi_post_block(vcpu);
7467 
7468 	local_irq_enable();
7469 	return (vcpu->pre_pcpu == -1);
7470 }
7471 
7472 static int vmx_pre_block(struct kvm_vcpu *vcpu)
7473 {
7474 	if (pi_pre_block(vcpu))
7475 		return 1;
7476 
7477 	if (kvm_lapic_hv_timer_in_use(vcpu))
7478 		kvm_lapic_switch_to_sw_timer(vcpu);
7479 
7480 	return 0;
7481 }
7482 
7483 static void pi_post_block(struct kvm_vcpu *vcpu)
7484 {
7485 	if (vcpu->pre_pcpu == -1)
7486 		return;
7487 
7488 	WARN_ON(irqs_disabled());
7489 	local_irq_disable();
7490 	__pi_post_block(vcpu);
7491 	local_irq_enable();
7492 }
7493 
7494 static void vmx_post_block(struct kvm_vcpu *vcpu)
7495 {
7496 	if (kvm_x86_ops.set_hv_timer)
7497 		kvm_lapic_switch_to_hv_timer(vcpu);
7498 
7499 	pi_post_block(vcpu);
7500 }
7501 
7502 /*
7503  * vmx_update_pi_irte - set IRTE for Posted-Interrupts
7504  *
7505  * @kvm: kvm
7506  * @host_irq: host irq of the interrupt
7507  * @guest_irq: gsi of the interrupt
7508  * @set: set or unset PI
7509  * returns 0 on success, < 0 on failure
7510  */
7511 static int vmx_update_pi_irte(struct kvm *kvm, unsigned int host_irq,
7512 			      uint32_t guest_irq, bool set)
7513 {
7514 	struct kvm_kernel_irq_routing_entry *e;
7515 	struct kvm_irq_routing_table *irq_rt;
7516 	struct kvm_lapic_irq irq;
7517 	struct kvm_vcpu *vcpu;
7518 	struct vcpu_data vcpu_info;
7519 	int idx, ret = 0;
7520 
7521 	if (!kvm_arch_has_assigned_device(kvm) ||
7522 		!irq_remapping_cap(IRQ_POSTING_CAP) ||
7523 		!kvm_vcpu_apicv_active(kvm->vcpus[0]))
7524 		return 0;
7525 
7526 	idx = srcu_read_lock(&kvm->irq_srcu);
7527 	irq_rt = srcu_dereference(kvm->irq_routing, &kvm->irq_srcu);
7528 	if (guest_irq >= irq_rt->nr_rt_entries ||
7529 	    hlist_empty(&irq_rt->map[guest_irq])) {
7530 		pr_warn_once("no route for guest_irq %u/%u (broken user space?)\n",
7531 			     guest_irq, irq_rt->nr_rt_entries);
7532 		goto out;
7533 	}
7534 
7535 	hlist_for_each_entry(e, &irq_rt->map[guest_irq], link) {
7536 		if (e->type != KVM_IRQ_ROUTING_MSI)
7537 			continue;
7538 		/*
7539 		 * VT-d PI cannot support posting multicast/broadcast
7540 		 * interrupts to a vCPU, we still use interrupt remapping
7541 		 * for these kind of interrupts.
7542 		 *
7543 		 * For lowest-priority interrupts, we only support
7544 		 * those with single CPU as the destination, e.g. user
7545 		 * configures the interrupts via /proc/irq or uses
7546 		 * irqbalance to make the interrupts single-CPU.
7547 		 *
7548 		 * We will support full lowest-priority interrupt later.
7549 		 *
7550 		 * In addition, we can only inject generic interrupts using
7551 		 * the PI mechanism, refuse to route others through it.
7552 		 */
7553 
7554 		kvm_set_msi_irq(kvm, e, &irq);
7555 		if (!kvm_intr_is_single_vcpu(kvm, &irq, &vcpu) ||
7556 		    !kvm_irq_is_postable(&irq)) {
7557 			/*
7558 			 * Make sure the IRTE is in remapped mode if
7559 			 * we don't handle it in posted mode.
7560 			 */
7561 			ret = irq_set_vcpu_affinity(host_irq, NULL);
7562 			if (ret < 0) {
7563 				printk(KERN_INFO
7564 				   "failed to back to remapped mode, irq: %u\n",
7565 				   host_irq);
7566 				goto out;
7567 			}
7568 
7569 			continue;
7570 		}
7571 
7572 		vcpu_info.pi_desc_addr = __pa(vcpu_to_pi_desc(vcpu));
7573 		vcpu_info.vector = irq.vector;
7574 
7575 		trace_kvm_pi_irte_update(host_irq, vcpu->vcpu_id, e->gsi,
7576 				vcpu_info.vector, vcpu_info.pi_desc_addr, set);
7577 
7578 		if (set)
7579 			ret = irq_set_vcpu_affinity(host_irq, &vcpu_info);
7580 		else
7581 			ret = irq_set_vcpu_affinity(host_irq, NULL);
7582 
7583 		if (ret < 0) {
7584 			printk(KERN_INFO "%s: failed to update PI IRTE\n",
7585 					__func__);
7586 			goto out;
7587 		}
7588 	}
7589 
7590 	ret = 0;
7591 out:
7592 	srcu_read_unlock(&kvm->irq_srcu, idx);
7593 	return ret;
7594 }
7595 
7596 static void vmx_setup_mce(struct kvm_vcpu *vcpu)
7597 {
7598 	if (vcpu->arch.mcg_cap & MCG_LMCE_P)
7599 		to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
7600 			FEAT_CTL_LMCE_ENABLED;
7601 	else
7602 		to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
7603 			~FEAT_CTL_LMCE_ENABLED;
7604 }
7605 
7606 static int vmx_smi_allowed(struct kvm_vcpu *vcpu)
7607 {
7608 	/* we need a nested vmexit to enter SMM, postpone if run is pending */
7609 	if (to_vmx(vcpu)->nested.nested_run_pending)
7610 		return 0;
7611 	return 1;
7612 }
7613 
7614 static int vmx_pre_enter_smm(struct kvm_vcpu *vcpu, char *smstate)
7615 {
7616 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7617 
7618 	vmx->nested.smm.guest_mode = is_guest_mode(vcpu);
7619 	if (vmx->nested.smm.guest_mode)
7620 		nested_vmx_vmexit(vcpu, -1, 0, 0);
7621 
7622 	vmx->nested.smm.vmxon = vmx->nested.vmxon;
7623 	vmx->nested.vmxon = false;
7624 	vmx_clear_hlt(vcpu);
7625 	return 0;
7626 }
7627 
7628 static int vmx_pre_leave_smm(struct kvm_vcpu *vcpu, const char *smstate)
7629 {
7630 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7631 	int ret;
7632 
7633 	if (vmx->nested.smm.vmxon) {
7634 		vmx->nested.vmxon = true;
7635 		vmx->nested.smm.vmxon = false;
7636 	}
7637 
7638 	if (vmx->nested.smm.guest_mode) {
7639 		ret = nested_vmx_enter_non_root_mode(vcpu, false);
7640 		if (ret)
7641 			return ret;
7642 
7643 		vmx->nested.smm.guest_mode = false;
7644 	}
7645 	return 0;
7646 }
7647 
7648 static int enable_smi_window(struct kvm_vcpu *vcpu)
7649 {
7650 	return 0;
7651 }
7652 
7653 static bool vmx_need_emulation_on_page_fault(struct kvm_vcpu *vcpu)
7654 {
7655 	return false;
7656 }
7657 
7658 static bool vmx_apic_init_signal_blocked(struct kvm_vcpu *vcpu)
7659 {
7660 	return to_vmx(vcpu)->nested.vmxon;
7661 }
7662 
7663 static void hardware_unsetup(void)
7664 {
7665 	if (nested)
7666 		nested_vmx_hardware_unsetup();
7667 
7668 	free_kvm_area();
7669 }
7670 
7671 static bool vmx_check_apicv_inhibit_reasons(ulong bit)
7672 {
7673 	ulong supported = BIT(APICV_INHIBIT_REASON_DISABLE) |
7674 			  BIT(APICV_INHIBIT_REASON_HYPERV);
7675 
7676 	return supported & BIT(bit);
7677 }
7678 
7679 static struct kvm_x86_ops vmx_x86_ops __initdata = {
7680 	.hardware_unsetup = hardware_unsetup,
7681 
7682 	.hardware_enable = hardware_enable,
7683 	.hardware_disable = hardware_disable,
7684 	.cpu_has_accelerated_tpr = report_flexpriority,
7685 	.has_emulated_msr = vmx_has_emulated_msr,
7686 
7687 	.vm_size = sizeof(struct kvm_vmx),
7688 	.vm_init = vmx_vm_init,
7689 
7690 	.vcpu_create = vmx_create_vcpu,
7691 	.vcpu_free = vmx_free_vcpu,
7692 	.vcpu_reset = vmx_vcpu_reset,
7693 
7694 	.prepare_guest_switch = vmx_prepare_switch_to_guest,
7695 	.vcpu_load = vmx_vcpu_load,
7696 	.vcpu_put = vmx_vcpu_put,
7697 
7698 	.update_bp_intercept = update_exception_bitmap,
7699 	.get_msr_feature = vmx_get_msr_feature,
7700 	.get_msr = vmx_get_msr,
7701 	.set_msr = vmx_set_msr,
7702 	.get_segment_base = vmx_get_segment_base,
7703 	.get_segment = vmx_get_segment,
7704 	.set_segment = vmx_set_segment,
7705 	.get_cpl = vmx_get_cpl,
7706 	.get_cs_db_l_bits = vmx_get_cs_db_l_bits,
7707 	.decache_cr0_guest_bits = vmx_decache_cr0_guest_bits,
7708 	.decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
7709 	.set_cr0 = vmx_set_cr0,
7710 	.set_cr4 = vmx_set_cr4,
7711 	.set_efer = vmx_set_efer,
7712 	.get_idt = vmx_get_idt,
7713 	.set_idt = vmx_set_idt,
7714 	.get_gdt = vmx_get_gdt,
7715 	.set_gdt = vmx_set_gdt,
7716 	.get_dr6 = vmx_get_dr6,
7717 	.set_dr6 = vmx_set_dr6,
7718 	.set_dr7 = vmx_set_dr7,
7719 	.sync_dirty_debug_regs = vmx_sync_dirty_debug_regs,
7720 	.cache_reg = vmx_cache_reg,
7721 	.get_rflags = vmx_get_rflags,
7722 	.set_rflags = vmx_set_rflags,
7723 
7724 	.tlb_flush = vmx_flush_tlb,
7725 	.tlb_flush_gva = vmx_flush_tlb_gva,
7726 
7727 	.run = vmx_vcpu_run,
7728 	.handle_exit = vmx_handle_exit,
7729 	.skip_emulated_instruction = vmx_skip_emulated_instruction,
7730 	.update_emulated_instruction = vmx_update_emulated_instruction,
7731 	.set_interrupt_shadow = vmx_set_interrupt_shadow,
7732 	.get_interrupt_shadow = vmx_get_interrupt_shadow,
7733 	.patch_hypercall = vmx_patch_hypercall,
7734 	.set_irq = vmx_inject_irq,
7735 	.set_nmi = vmx_inject_nmi,
7736 	.queue_exception = vmx_queue_exception,
7737 	.cancel_injection = vmx_cancel_injection,
7738 	.interrupt_allowed = vmx_interrupt_allowed,
7739 	.nmi_allowed = vmx_nmi_allowed,
7740 	.get_nmi_mask = vmx_get_nmi_mask,
7741 	.set_nmi_mask = vmx_set_nmi_mask,
7742 	.enable_nmi_window = enable_nmi_window,
7743 	.enable_irq_window = enable_irq_window,
7744 	.update_cr8_intercept = update_cr8_intercept,
7745 	.set_virtual_apic_mode = vmx_set_virtual_apic_mode,
7746 	.set_apic_access_page_addr = vmx_set_apic_access_page_addr,
7747 	.refresh_apicv_exec_ctrl = vmx_refresh_apicv_exec_ctrl,
7748 	.load_eoi_exitmap = vmx_load_eoi_exitmap,
7749 	.apicv_post_state_restore = vmx_apicv_post_state_restore,
7750 	.check_apicv_inhibit_reasons = vmx_check_apicv_inhibit_reasons,
7751 	.hwapic_irr_update = vmx_hwapic_irr_update,
7752 	.hwapic_isr_update = vmx_hwapic_isr_update,
7753 	.guest_apic_has_interrupt = vmx_guest_apic_has_interrupt,
7754 	.sync_pir_to_irr = vmx_sync_pir_to_irr,
7755 	.deliver_posted_interrupt = vmx_deliver_posted_interrupt,
7756 	.dy_apicv_has_pending_interrupt = vmx_dy_apicv_has_pending_interrupt,
7757 
7758 	.set_tss_addr = vmx_set_tss_addr,
7759 	.set_identity_map_addr = vmx_set_identity_map_addr,
7760 	.get_tdp_level = get_ept_level,
7761 	.get_mt_mask = vmx_get_mt_mask,
7762 
7763 	.get_exit_info = vmx_get_exit_info,
7764 
7765 	.cpuid_update = vmx_cpuid_update,
7766 
7767 	.has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
7768 
7769 	.read_l1_tsc_offset = vmx_read_l1_tsc_offset,
7770 	.write_l1_tsc_offset = vmx_write_l1_tsc_offset,
7771 
7772 	.load_mmu_pgd = vmx_load_mmu_pgd,
7773 
7774 	.check_intercept = vmx_check_intercept,
7775 	.handle_exit_irqoff = vmx_handle_exit_irqoff,
7776 
7777 	.request_immediate_exit = vmx_request_immediate_exit,
7778 
7779 	.sched_in = vmx_sched_in,
7780 
7781 	.slot_enable_log_dirty = vmx_slot_enable_log_dirty,
7782 	.slot_disable_log_dirty = vmx_slot_disable_log_dirty,
7783 	.flush_log_dirty = vmx_flush_log_dirty,
7784 	.enable_log_dirty_pt_masked = vmx_enable_log_dirty_pt_masked,
7785 	.write_log_dirty = vmx_write_pml_buffer,
7786 
7787 	.pre_block = vmx_pre_block,
7788 	.post_block = vmx_post_block,
7789 
7790 	.pmu_ops = &intel_pmu_ops,
7791 
7792 	.update_pi_irte = vmx_update_pi_irte,
7793 
7794 #ifdef CONFIG_X86_64
7795 	.set_hv_timer = vmx_set_hv_timer,
7796 	.cancel_hv_timer = vmx_cancel_hv_timer,
7797 #endif
7798 
7799 	.setup_mce = vmx_setup_mce,
7800 
7801 	.smi_allowed = vmx_smi_allowed,
7802 	.pre_enter_smm = vmx_pre_enter_smm,
7803 	.pre_leave_smm = vmx_pre_leave_smm,
7804 	.enable_smi_window = enable_smi_window,
7805 
7806 	.check_nested_events = NULL,
7807 	.get_nested_state = NULL,
7808 	.set_nested_state = NULL,
7809 	.get_vmcs12_pages = NULL,
7810 	.nested_enable_evmcs = NULL,
7811 	.nested_get_evmcs_version = NULL,
7812 	.need_emulation_on_page_fault = vmx_need_emulation_on_page_fault,
7813 	.apic_init_signal_blocked = vmx_apic_init_signal_blocked,
7814 };
7815 
7816 static __init int hardware_setup(void)
7817 {
7818 	unsigned long host_bndcfgs;
7819 	struct desc_ptr dt;
7820 	int r, i, ept_lpage_level;
7821 
7822 	store_idt(&dt);
7823 	host_idt_base = dt.address;
7824 
7825 	for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i)
7826 		kvm_define_shared_msr(i, vmx_msr_index[i]);
7827 
7828 	if (setup_vmcs_config(&vmcs_config, &vmx_capability) < 0)
7829 		return -EIO;
7830 
7831 	if (boot_cpu_has(X86_FEATURE_NX))
7832 		kvm_enable_efer_bits(EFER_NX);
7833 
7834 	if (boot_cpu_has(X86_FEATURE_MPX)) {
7835 		rdmsrl(MSR_IA32_BNDCFGS, host_bndcfgs);
7836 		WARN_ONCE(host_bndcfgs, "KVM: BNDCFGS in host will be lost");
7837 	}
7838 
7839 	if (!cpu_has_vmx_mpx())
7840 		supported_xcr0 &= ~(XFEATURE_MASK_BNDREGS |
7841 				    XFEATURE_MASK_BNDCSR);
7842 
7843 	if (!cpu_has_vmx_vpid() || !cpu_has_vmx_invvpid() ||
7844 	    !(cpu_has_vmx_invvpid_single() || cpu_has_vmx_invvpid_global()))
7845 		enable_vpid = 0;
7846 
7847 	if (!cpu_has_vmx_ept() ||
7848 	    !cpu_has_vmx_ept_4levels() ||
7849 	    !cpu_has_vmx_ept_mt_wb() ||
7850 	    !cpu_has_vmx_invept_global())
7851 		enable_ept = 0;
7852 
7853 	if (!cpu_has_vmx_ept_ad_bits() || !enable_ept)
7854 		enable_ept_ad_bits = 0;
7855 
7856 	if (!cpu_has_vmx_unrestricted_guest() || !enable_ept)
7857 		enable_unrestricted_guest = 0;
7858 
7859 	if (!cpu_has_vmx_flexpriority())
7860 		flexpriority_enabled = 0;
7861 
7862 	if (!cpu_has_virtual_nmis())
7863 		enable_vnmi = 0;
7864 
7865 	/*
7866 	 * set_apic_access_page_addr() is used to reload apic access
7867 	 * page upon invalidation.  No need to do anything if not
7868 	 * using the APIC_ACCESS_ADDR VMCS field.
7869 	 */
7870 	if (!flexpriority_enabled)
7871 		vmx_x86_ops.set_apic_access_page_addr = NULL;
7872 
7873 	if (!cpu_has_vmx_tpr_shadow())
7874 		vmx_x86_ops.update_cr8_intercept = NULL;
7875 
7876 #if IS_ENABLED(CONFIG_HYPERV)
7877 	if (ms_hyperv.nested_features & HV_X64_NESTED_GUEST_MAPPING_FLUSH
7878 	    && enable_ept) {
7879 		vmx_x86_ops.tlb_remote_flush = hv_remote_flush_tlb;
7880 		vmx_x86_ops.tlb_remote_flush_with_range =
7881 				hv_remote_flush_tlb_with_range;
7882 	}
7883 #endif
7884 
7885 	if (!cpu_has_vmx_ple()) {
7886 		ple_gap = 0;
7887 		ple_window = 0;
7888 		ple_window_grow = 0;
7889 		ple_window_max = 0;
7890 		ple_window_shrink = 0;
7891 	}
7892 
7893 	if (!cpu_has_vmx_apicv()) {
7894 		enable_apicv = 0;
7895 		vmx_x86_ops.sync_pir_to_irr = NULL;
7896 	}
7897 
7898 	if (cpu_has_vmx_tsc_scaling()) {
7899 		kvm_has_tsc_control = true;
7900 		kvm_max_tsc_scaling_ratio = KVM_VMX_TSC_MULTIPLIER_MAX;
7901 		kvm_tsc_scaling_ratio_frac_bits = 48;
7902 	}
7903 
7904 	set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
7905 
7906 	if (enable_ept)
7907 		vmx_enable_tdp();
7908 
7909 	if (!enable_ept)
7910 		ept_lpage_level = 0;
7911 	else if (cpu_has_vmx_ept_1g_page())
7912 		ept_lpage_level = PT_PDPE_LEVEL;
7913 	else if (cpu_has_vmx_ept_2m_page())
7914 		ept_lpage_level = PT_DIRECTORY_LEVEL;
7915 	else
7916 		ept_lpage_level = PT_PAGE_TABLE_LEVEL;
7917 	kvm_configure_mmu(enable_ept, ept_lpage_level);
7918 
7919 	/*
7920 	 * Only enable PML when hardware supports PML feature, and both EPT
7921 	 * and EPT A/D bit features are enabled -- PML depends on them to work.
7922 	 */
7923 	if (!enable_ept || !enable_ept_ad_bits || !cpu_has_vmx_pml())
7924 		enable_pml = 0;
7925 
7926 	if (!enable_pml) {
7927 		vmx_x86_ops.slot_enable_log_dirty = NULL;
7928 		vmx_x86_ops.slot_disable_log_dirty = NULL;
7929 		vmx_x86_ops.flush_log_dirty = NULL;
7930 		vmx_x86_ops.enable_log_dirty_pt_masked = NULL;
7931 	}
7932 
7933 	if (!cpu_has_vmx_preemption_timer())
7934 		enable_preemption_timer = false;
7935 
7936 	if (enable_preemption_timer) {
7937 		u64 use_timer_freq = 5000ULL * 1000 * 1000;
7938 		u64 vmx_msr;
7939 
7940 		rdmsrl(MSR_IA32_VMX_MISC, vmx_msr);
7941 		cpu_preemption_timer_multi =
7942 			vmx_msr & VMX_MISC_PREEMPTION_TIMER_RATE_MASK;
7943 
7944 		if (tsc_khz)
7945 			use_timer_freq = (u64)tsc_khz * 1000;
7946 		use_timer_freq >>= cpu_preemption_timer_multi;
7947 
7948 		/*
7949 		 * KVM "disables" the preemption timer by setting it to its max
7950 		 * value.  Don't use the timer if it might cause spurious exits
7951 		 * at a rate faster than 0.1 Hz (of uninterrupted guest time).
7952 		 */
7953 		if (use_timer_freq > 0xffffffffu / 10)
7954 			enable_preemption_timer = false;
7955 	}
7956 
7957 	if (!enable_preemption_timer) {
7958 		vmx_x86_ops.set_hv_timer = NULL;
7959 		vmx_x86_ops.cancel_hv_timer = NULL;
7960 		vmx_x86_ops.request_immediate_exit = __kvm_request_immediate_exit;
7961 	}
7962 
7963 	kvm_set_posted_intr_wakeup_handler(wakeup_handler);
7964 
7965 	kvm_mce_cap_supported |= MCG_LMCE_P;
7966 
7967 	if (pt_mode != PT_MODE_SYSTEM && pt_mode != PT_MODE_HOST_GUEST)
7968 		return -EINVAL;
7969 	if (!enable_ept || !cpu_has_vmx_intel_pt())
7970 		pt_mode = PT_MODE_SYSTEM;
7971 
7972 	if (nested) {
7973 		nested_vmx_setup_ctls_msrs(&vmcs_config.nested,
7974 					   vmx_capability.ept);
7975 
7976 		r = nested_vmx_hardware_setup(&vmx_x86_ops,
7977 					      kvm_vmx_exit_handlers);
7978 		if (r)
7979 			return r;
7980 	}
7981 
7982 	vmx_set_cpu_caps();
7983 
7984 	r = alloc_kvm_area();
7985 	if (r)
7986 		nested_vmx_hardware_unsetup();
7987 	return r;
7988 }
7989 
7990 static struct kvm_x86_init_ops vmx_init_ops __initdata = {
7991 	.cpu_has_kvm_support = cpu_has_kvm_support,
7992 	.disabled_by_bios = vmx_disabled_by_bios,
7993 	.check_processor_compatibility = vmx_check_processor_compat,
7994 	.hardware_setup = hardware_setup,
7995 
7996 	.runtime_ops = &vmx_x86_ops,
7997 };
7998 
7999 static void vmx_cleanup_l1d_flush(void)
8000 {
8001 	if (vmx_l1d_flush_pages) {
8002 		free_pages((unsigned long)vmx_l1d_flush_pages, L1D_CACHE_ORDER);
8003 		vmx_l1d_flush_pages = NULL;
8004 	}
8005 	/* Restore state so sysfs ignores VMX */
8006 	l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_AUTO;
8007 }
8008 
8009 static void vmx_exit(void)
8010 {
8011 #ifdef CONFIG_KEXEC_CORE
8012 	RCU_INIT_POINTER(crash_vmclear_loaded_vmcss, NULL);
8013 	synchronize_rcu();
8014 #endif
8015 
8016 	kvm_exit();
8017 
8018 #if IS_ENABLED(CONFIG_HYPERV)
8019 	if (static_branch_unlikely(&enable_evmcs)) {
8020 		int cpu;
8021 		struct hv_vp_assist_page *vp_ap;
8022 		/*
8023 		 * Reset everything to support using non-enlightened VMCS
8024 		 * access later (e.g. when we reload the module with
8025 		 * enlightened_vmcs=0)
8026 		 */
8027 		for_each_online_cpu(cpu) {
8028 			vp_ap =	hv_get_vp_assist_page(cpu);
8029 
8030 			if (!vp_ap)
8031 				continue;
8032 
8033 			vp_ap->nested_control.features.directhypercall = 0;
8034 			vp_ap->current_nested_vmcs = 0;
8035 			vp_ap->enlighten_vmentry = 0;
8036 		}
8037 
8038 		static_branch_disable(&enable_evmcs);
8039 	}
8040 #endif
8041 	vmx_cleanup_l1d_flush();
8042 }
8043 module_exit(vmx_exit);
8044 
8045 static int __init vmx_init(void)
8046 {
8047 	int r;
8048 
8049 #if IS_ENABLED(CONFIG_HYPERV)
8050 	/*
8051 	 * Enlightened VMCS usage should be recommended and the host needs
8052 	 * to support eVMCS v1 or above. We can also disable eVMCS support
8053 	 * with module parameter.
8054 	 */
8055 	if (enlightened_vmcs &&
8056 	    ms_hyperv.hints & HV_X64_ENLIGHTENED_VMCS_RECOMMENDED &&
8057 	    (ms_hyperv.nested_features & HV_X64_ENLIGHTENED_VMCS_VERSION) >=
8058 	    KVM_EVMCS_VERSION) {
8059 		int cpu;
8060 
8061 		/* Check that we have assist pages on all online CPUs */
8062 		for_each_online_cpu(cpu) {
8063 			if (!hv_get_vp_assist_page(cpu)) {
8064 				enlightened_vmcs = false;
8065 				break;
8066 			}
8067 		}
8068 
8069 		if (enlightened_vmcs) {
8070 			pr_info("KVM: vmx: using Hyper-V Enlightened VMCS\n");
8071 			static_branch_enable(&enable_evmcs);
8072 		}
8073 
8074 		if (ms_hyperv.nested_features & HV_X64_NESTED_DIRECT_FLUSH)
8075 			vmx_x86_ops.enable_direct_tlbflush
8076 				= hv_enable_direct_tlbflush;
8077 
8078 	} else {
8079 		enlightened_vmcs = false;
8080 	}
8081 #endif
8082 
8083 	r = kvm_init(&vmx_init_ops, sizeof(struct vcpu_vmx),
8084 		     __alignof__(struct vcpu_vmx), THIS_MODULE);
8085 	if (r)
8086 		return r;
8087 
8088 	/*
8089 	 * Must be called after kvm_init() so enable_ept is properly set
8090 	 * up. Hand the parameter mitigation value in which was stored in
8091 	 * the pre module init parser. If no parameter was given, it will
8092 	 * contain 'auto' which will be turned into the default 'cond'
8093 	 * mitigation mode.
8094 	 */
8095 	r = vmx_setup_l1d_flush(vmentry_l1d_flush_param);
8096 	if (r) {
8097 		vmx_exit();
8098 		return r;
8099 	}
8100 
8101 #ifdef CONFIG_KEXEC_CORE
8102 	rcu_assign_pointer(crash_vmclear_loaded_vmcss,
8103 			   crash_vmclear_local_loaded_vmcss);
8104 #endif
8105 	vmx_check_vmcs12_offsets();
8106 
8107 	return 0;
8108 }
8109 module_init(vmx_init);
8110