xref: /openbmc/linux/arch/x86/kvm/vmx/vmx.c (revision 0a671dc5)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Kernel-based Virtual Machine driver for Linux
4  *
5  * This module enables machines with Intel VT-x extensions to run virtual
6  * machines without emulation or binary translation.
7  *
8  * Copyright (C) 2006 Qumranet, Inc.
9  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
10  *
11  * Authors:
12  *   Avi Kivity   <avi@qumranet.com>
13  *   Yaniv Kamay  <yaniv@qumranet.com>
14  */
15 
16 #include <linux/frame.h>
17 #include <linux/highmem.h>
18 #include <linux/hrtimer.h>
19 #include <linux/kernel.h>
20 #include <linux/kvm_host.h>
21 #include <linux/module.h>
22 #include <linux/moduleparam.h>
23 #include <linux/mod_devicetable.h>
24 #include <linux/mm.h>
25 #include <linux/sched.h>
26 #include <linux/sched/smt.h>
27 #include <linux/slab.h>
28 #include <linux/tboot.h>
29 #include <linux/trace_events.h>
30 
31 #include <asm/apic.h>
32 #include <asm/asm.h>
33 #include <asm/cpu.h>
34 #include <asm/debugreg.h>
35 #include <asm/desc.h>
36 #include <asm/fpu/internal.h>
37 #include <asm/io.h>
38 #include <asm/irq_remapping.h>
39 #include <asm/kexec.h>
40 #include <asm/perf_event.h>
41 #include <asm/mce.h>
42 #include <asm/mmu_context.h>
43 #include <asm/mshyperv.h>
44 #include <asm/spec-ctrl.h>
45 #include <asm/virtext.h>
46 #include <asm/vmx.h>
47 
48 #include "capabilities.h"
49 #include "cpuid.h"
50 #include "evmcs.h"
51 #include "irq.h"
52 #include "kvm_cache_regs.h"
53 #include "lapic.h"
54 #include "mmu.h"
55 #include "nested.h"
56 #include "ops.h"
57 #include "pmu.h"
58 #include "trace.h"
59 #include "vmcs.h"
60 #include "vmcs12.h"
61 #include "vmx.h"
62 #include "x86.h"
63 
64 MODULE_AUTHOR("Qumranet");
65 MODULE_LICENSE("GPL");
66 
67 static const struct x86_cpu_id vmx_cpu_id[] = {
68 	X86_FEATURE_MATCH(X86_FEATURE_VMX),
69 	{}
70 };
71 MODULE_DEVICE_TABLE(x86cpu, vmx_cpu_id);
72 
73 bool __read_mostly enable_vpid = 1;
74 module_param_named(vpid, enable_vpid, bool, 0444);
75 
76 static bool __read_mostly enable_vnmi = 1;
77 module_param_named(vnmi, enable_vnmi, bool, S_IRUGO);
78 
79 bool __read_mostly flexpriority_enabled = 1;
80 module_param_named(flexpriority, flexpriority_enabled, bool, S_IRUGO);
81 
82 bool __read_mostly enable_ept = 1;
83 module_param_named(ept, enable_ept, bool, S_IRUGO);
84 
85 bool __read_mostly enable_unrestricted_guest = 1;
86 module_param_named(unrestricted_guest,
87 			enable_unrestricted_guest, bool, S_IRUGO);
88 
89 bool __read_mostly enable_ept_ad_bits = 1;
90 module_param_named(eptad, enable_ept_ad_bits, bool, S_IRUGO);
91 
92 static bool __read_mostly emulate_invalid_guest_state = true;
93 module_param(emulate_invalid_guest_state, bool, S_IRUGO);
94 
95 static bool __read_mostly fasteoi = 1;
96 module_param(fasteoi, bool, S_IRUGO);
97 
98 static bool __read_mostly enable_apicv = 1;
99 module_param(enable_apicv, bool, S_IRUGO);
100 
101 /*
102  * If nested=1, nested virtualization is supported, i.e., guests may use
103  * VMX and be a hypervisor for its own guests. If nested=0, guests may not
104  * use VMX instructions.
105  */
106 static bool __read_mostly nested = 1;
107 module_param(nested, bool, S_IRUGO);
108 
109 static u64 __read_mostly host_xss;
110 
111 bool __read_mostly enable_pml = 1;
112 module_param_named(pml, enable_pml, bool, S_IRUGO);
113 
114 static bool __read_mostly dump_invalid_vmcs = 0;
115 module_param(dump_invalid_vmcs, bool, 0644);
116 
117 #define MSR_BITMAP_MODE_X2APIC		1
118 #define MSR_BITMAP_MODE_X2APIC_APICV	2
119 
120 #define KVM_VMX_TSC_MULTIPLIER_MAX     0xffffffffffffffffULL
121 
122 /* Guest_tsc -> host_tsc conversion requires 64-bit division.  */
123 static int __read_mostly cpu_preemption_timer_multi;
124 static bool __read_mostly enable_preemption_timer = 1;
125 #ifdef CONFIG_X86_64
126 module_param_named(preemption_timer, enable_preemption_timer, bool, S_IRUGO);
127 #endif
128 
129 #define KVM_VM_CR0_ALWAYS_OFF (X86_CR0_NW | X86_CR0_CD)
130 #define KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST X86_CR0_NE
131 #define KVM_VM_CR0_ALWAYS_ON				\
132 	(KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST | 	\
133 	 X86_CR0_WP | X86_CR0_PG | X86_CR0_PE)
134 #define KVM_CR4_GUEST_OWNED_BITS				      \
135 	(X86_CR4_PVI | X86_CR4_DE | X86_CR4_PCE | X86_CR4_OSFXSR      \
136 	 | X86_CR4_OSXMMEXCPT | X86_CR4_LA57 | X86_CR4_TSD)
137 
138 #define KVM_VM_CR4_ALWAYS_ON_UNRESTRICTED_GUEST X86_CR4_VMXE
139 #define KVM_PMODE_VM_CR4_ALWAYS_ON (X86_CR4_PAE | X86_CR4_VMXE)
140 #define KVM_RMODE_VM_CR4_ALWAYS_ON (X86_CR4_VME | X86_CR4_PAE | X86_CR4_VMXE)
141 
142 #define RMODE_GUEST_OWNED_EFLAGS_BITS (~(X86_EFLAGS_IOPL | X86_EFLAGS_VM))
143 
144 #define MSR_IA32_RTIT_STATUS_MASK (~(RTIT_STATUS_FILTEREN | \
145 	RTIT_STATUS_CONTEXTEN | RTIT_STATUS_TRIGGEREN | \
146 	RTIT_STATUS_ERROR | RTIT_STATUS_STOPPED | \
147 	RTIT_STATUS_BYTECNT))
148 
149 #define MSR_IA32_RTIT_OUTPUT_BASE_MASK \
150 	(~((1UL << cpuid_query_maxphyaddr(vcpu)) - 1) | 0x7f)
151 
152 /*
153  * These 2 parameters are used to config the controls for Pause-Loop Exiting:
154  * ple_gap:    upper bound on the amount of time between two successive
155  *             executions of PAUSE in a loop. Also indicate if ple enabled.
156  *             According to test, this time is usually smaller than 128 cycles.
157  * ple_window: upper bound on the amount of time a guest is allowed to execute
158  *             in a PAUSE loop. Tests indicate that most spinlocks are held for
159  *             less than 2^12 cycles
160  * Time is measured based on a counter that runs at the same rate as the TSC,
161  * refer SDM volume 3b section 21.6.13 & 22.1.3.
162  */
163 static unsigned int ple_gap = KVM_DEFAULT_PLE_GAP;
164 module_param(ple_gap, uint, 0444);
165 
166 static unsigned int ple_window = KVM_VMX_DEFAULT_PLE_WINDOW;
167 module_param(ple_window, uint, 0444);
168 
169 /* Default doubles per-vcpu window every exit. */
170 static unsigned int ple_window_grow = KVM_DEFAULT_PLE_WINDOW_GROW;
171 module_param(ple_window_grow, uint, 0444);
172 
173 /* Default resets per-vcpu window every exit to ple_window. */
174 static unsigned int ple_window_shrink = KVM_DEFAULT_PLE_WINDOW_SHRINK;
175 module_param(ple_window_shrink, uint, 0444);
176 
177 /* Default is to compute the maximum so we can never overflow. */
178 static unsigned int ple_window_max        = KVM_VMX_DEFAULT_PLE_WINDOW_MAX;
179 module_param(ple_window_max, uint, 0444);
180 
181 /* Default is SYSTEM mode, 1 for host-guest mode */
182 int __read_mostly pt_mode = PT_MODE_SYSTEM;
183 module_param(pt_mode, int, S_IRUGO);
184 
185 static DEFINE_STATIC_KEY_FALSE(vmx_l1d_should_flush);
186 static DEFINE_STATIC_KEY_FALSE(vmx_l1d_flush_cond);
187 static DEFINE_MUTEX(vmx_l1d_flush_mutex);
188 
189 /* Storage for pre module init parameter parsing */
190 static enum vmx_l1d_flush_state __read_mostly vmentry_l1d_flush_param = VMENTER_L1D_FLUSH_AUTO;
191 
192 static const struct {
193 	const char *option;
194 	bool for_parse;
195 } vmentry_l1d_param[] = {
196 	[VMENTER_L1D_FLUSH_AUTO]	 = {"auto", true},
197 	[VMENTER_L1D_FLUSH_NEVER]	 = {"never", true},
198 	[VMENTER_L1D_FLUSH_COND]	 = {"cond", true},
199 	[VMENTER_L1D_FLUSH_ALWAYS]	 = {"always", true},
200 	[VMENTER_L1D_FLUSH_EPT_DISABLED] = {"EPT disabled", false},
201 	[VMENTER_L1D_FLUSH_NOT_REQUIRED] = {"not required", false},
202 };
203 
204 #define L1D_CACHE_ORDER 4
205 static void *vmx_l1d_flush_pages;
206 
207 static int vmx_setup_l1d_flush(enum vmx_l1d_flush_state l1tf)
208 {
209 	struct page *page;
210 	unsigned int i;
211 
212 	if (!boot_cpu_has_bug(X86_BUG_L1TF)) {
213 		l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_NOT_REQUIRED;
214 		return 0;
215 	}
216 
217 	if (!enable_ept) {
218 		l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_EPT_DISABLED;
219 		return 0;
220 	}
221 
222 	if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES)) {
223 		u64 msr;
224 
225 		rdmsrl(MSR_IA32_ARCH_CAPABILITIES, msr);
226 		if (msr & ARCH_CAP_SKIP_VMENTRY_L1DFLUSH) {
227 			l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_NOT_REQUIRED;
228 			return 0;
229 		}
230 	}
231 
232 	/* If set to auto use the default l1tf mitigation method */
233 	if (l1tf == VMENTER_L1D_FLUSH_AUTO) {
234 		switch (l1tf_mitigation) {
235 		case L1TF_MITIGATION_OFF:
236 			l1tf = VMENTER_L1D_FLUSH_NEVER;
237 			break;
238 		case L1TF_MITIGATION_FLUSH_NOWARN:
239 		case L1TF_MITIGATION_FLUSH:
240 		case L1TF_MITIGATION_FLUSH_NOSMT:
241 			l1tf = VMENTER_L1D_FLUSH_COND;
242 			break;
243 		case L1TF_MITIGATION_FULL:
244 		case L1TF_MITIGATION_FULL_FORCE:
245 			l1tf = VMENTER_L1D_FLUSH_ALWAYS;
246 			break;
247 		}
248 	} else if (l1tf_mitigation == L1TF_MITIGATION_FULL_FORCE) {
249 		l1tf = VMENTER_L1D_FLUSH_ALWAYS;
250 	}
251 
252 	if (l1tf != VMENTER_L1D_FLUSH_NEVER && !vmx_l1d_flush_pages &&
253 	    !boot_cpu_has(X86_FEATURE_FLUSH_L1D)) {
254 		/*
255 		 * This allocation for vmx_l1d_flush_pages is not tied to a VM
256 		 * lifetime and so should not be charged to a memcg.
257 		 */
258 		page = alloc_pages(GFP_KERNEL, L1D_CACHE_ORDER);
259 		if (!page)
260 			return -ENOMEM;
261 		vmx_l1d_flush_pages = page_address(page);
262 
263 		/*
264 		 * Initialize each page with a different pattern in
265 		 * order to protect against KSM in the nested
266 		 * virtualization case.
267 		 */
268 		for (i = 0; i < 1u << L1D_CACHE_ORDER; ++i) {
269 			memset(vmx_l1d_flush_pages + i * PAGE_SIZE, i + 1,
270 			       PAGE_SIZE);
271 		}
272 	}
273 
274 	l1tf_vmx_mitigation = l1tf;
275 
276 	if (l1tf != VMENTER_L1D_FLUSH_NEVER)
277 		static_branch_enable(&vmx_l1d_should_flush);
278 	else
279 		static_branch_disable(&vmx_l1d_should_flush);
280 
281 	if (l1tf == VMENTER_L1D_FLUSH_COND)
282 		static_branch_enable(&vmx_l1d_flush_cond);
283 	else
284 		static_branch_disable(&vmx_l1d_flush_cond);
285 	return 0;
286 }
287 
288 static int vmentry_l1d_flush_parse(const char *s)
289 {
290 	unsigned int i;
291 
292 	if (s) {
293 		for (i = 0; i < ARRAY_SIZE(vmentry_l1d_param); i++) {
294 			if (vmentry_l1d_param[i].for_parse &&
295 			    sysfs_streq(s, vmentry_l1d_param[i].option))
296 				return i;
297 		}
298 	}
299 	return -EINVAL;
300 }
301 
302 static int vmentry_l1d_flush_set(const char *s, const struct kernel_param *kp)
303 {
304 	int l1tf, ret;
305 
306 	l1tf = vmentry_l1d_flush_parse(s);
307 	if (l1tf < 0)
308 		return l1tf;
309 
310 	if (!boot_cpu_has(X86_BUG_L1TF))
311 		return 0;
312 
313 	/*
314 	 * Has vmx_init() run already? If not then this is the pre init
315 	 * parameter parsing. In that case just store the value and let
316 	 * vmx_init() do the proper setup after enable_ept has been
317 	 * established.
318 	 */
319 	if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_AUTO) {
320 		vmentry_l1d_flush_param = l1tf;
321 		return 0;
322 	}
323 
324 	mutex_lock(&vmx_l1d_flush_mutex);
325 	ret = vmx_setup_l1d_flush(l1tf);
326 	mutex_unlock(&vmx_l1d_flush_mutex);
327 	return ret;
328 }
329 
330 static int vmentry_l1d_flush_get(char *s, const struct kernel_param *kp)
331 {
332 	if (WARN_ON_ONCE(l1tf_vmx_mitigation >= ARRAY_SIZE(vmentry_l1d_param)))
333 		return sprintf(s, "???\n");
334 
335 	return sprintf(s, "%s\n", vmentry_l1d_param[l1tf_vmx_mitigation].option);
336 }
337 
338 static const struct kernel_param_ops vmentry_l1d_flush_ops = {
339 	.set = vmentry_l1d_flush_set,
340 	.get = vmentry_l1d_flush_get,
341 };
342 module_param_cb(vmentry_l1d_flush, &vmentry_l1d_flush_ops, NULL, 0644);
343 
344 static bool guest_state_valid(struct kvm_vcpu *vcpu);
345 static u32 vmx_segment_access_rights(struct kvm_segment *var);
346 static __always_inline void vmx_disable_intercept_for_msr(unsigned long *msr_bitmap,
347 							  u32 msr, int type);
348 
349 void vmx_vmexit(void);
350 
351 #define vmx_insn_failed(fmt...)		\
352 do {					\
353 	WARN_ONCE(1, fmt);		\
354 	pr_warn_ratelimited(fmt);	\
355 } while (0)
356 
357 asmlinkage void vmread_error(unsigned long field, bool fault)
358 {
359 	if (fault)
360 		kvm_spurious_fault();
361 	else
362 		vmx_insn_failed("kvm: vmread failed: field=%lx\n", field);
363 }
364 
365 noinline void vmwrite_error(unsigned long field, unsigned long value)
366 {
367 	vmx_insn_failed("kvm: vmwrite failed: field=%lx val=%lx err=%d\n",
368 			field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
369 }
370 
371 noinline void vmclear_error(struct vmcs *vmcs, u64 phys_addr)
372 {
373 	vmx_insn_failed("kvm: vmclear failed: %p/%llx\n", vmcs, phys_addr);
374 }
375 
376 noinline void vmptrld_error(struct vmcs *vmcs, u64 phys_addr)
377 {
378 	vmx_insn_failed("kvm: vmptrld failed: %p/%llx\n", vmcs, phys_addr);
379 }
380 
381 noinline void invvpid_error(unsigned long ext, u16 vpid, gva_t gva)
382 {
383 	vmx_insn_failed("kvm: invvpid failed: ext=0x%lx vpid=%u gva=0x%lx\n",
384 			ext, vpid, gva);
385 }
386 
387 noinline void invept_error(unsigned long ext, u64 eptp, gpa_t gpa)
388 {
389 	vmx_insn_failed("kvm: invept failed: ext=0x%lx eptp=%llx gpa=0x%llx\n",
390 			ext, eptp, gpa);
391 }
392 
393 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
394 DEFINE_PER_CPU(struct vmcs *, current_vmcs);
395 /*
396  * We maintain a per-CPU linked-list of VMCS loaded on that CPU. This is needed
397  * when a CPU is brought down, and we need to VMCLEAR all VMCSs loaded on it.
398  */
399 static DEFINE_PER_CPU(struct list_head, loaded_vmcss_on_cpu);
400 
401 /*
402  * We maintian a per-CPU linked-list of vCPU, so in wakeup_handler() we
403  * can find which vCPU should be waken up.
404  */
405 static DEFINE_PER_CPU(struct list_head, blocked_vcpu_on_cpu);
406 static DEFINE_PER_CPU(spinlock_t, blocked_vcpu_on_cpu_lock);
407 
408 static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
409 static DEFINE_SPINLOCK(vmx_vpid_lock);
410 
411 struct vmcs_config vmcs_config;
412 struct vmx_capability vmx_capability;
413 
414 #define VMX_SEGMENT_FIELD(seg)					\
415 	[VCPU_SREG_##seg] = {                                   \
416 		.selector = GUEST_##seg##_SELECTOR,		\
417 		.base = GUEST_##seg##_BASE,		   	\
418 		.limit = GUEST_##seg##_LIMIT,		   	\
419 		.ar_bytes = GUEST_##seg##_AR_BYTES,	   	\
420 	}
421 
422 static const struct kvm_vmx_segment_field {
423 	unsigned selector;
424 	unsigned base;
425 	unsigned limit;
426 	unsigned ar_bytes;
427 } kvm_vmx_segment_fields[] = {
428 	VMX_SEGMENT_FIELD(CS),
429 	VMX_SEGMENT_FIELD(DS),
430 	VMX_SEGMENT_FIELD(ES),
431 	VMX_SEGMENT_FIELD(FS),
432 	VMX_SEGMENT_FIELD(GS),
433 	VMX_SEGMENT_FIELD(SS),
434 	VMX_SEGMENT_FIELD(TR),
435 	VMX_SEGMENT_FIELD(LDTR),
436 };
437 
438 u64 host_efer;
439 static unsigned long host_idt_base;
440 
441 /*
442  * Though SYSCALL is only supported in 64-bit mode on Intel CPUs, kvm
443  * will emulate SYSCALL in legacy mode if the vendor string in guest
444  * CPUID.0:{EBX,ECX,EDX} is "AuthenticAMD" or "AMDisbetter!" To
445  * support this emulation, IA32_STAR must always be included in
446  * vmx_msr_index[], even in i386 builds.
447  */
448 const u32 vmx_msr_index[] = {
449 #ifdef CONFIG_X86_64
450 	MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
451 #endif
452 	MSR_EFER, MSR_TSC_AUX, MSR_STAR,
453 };
454 
455 #if IS_ENABLED(CONFIG_HYPERV)
456 static bool __read_mostly enlightened_vmcs = true;
457 module_param(enlightened_vmcs, bool, 0444);
458 
459 /* check_ept_pointer() should be under protection of ept_pointer_lock. */
460 static void check_ept_pointer_match(struct kvm *kvm)
461 {
462 	struct kvm_vcpu *vcpu;
463 	u64 tmp_eptp = INVALID_PAGE;
464 	int i;
465 
466 	kvm_for_each_vcpu(i, vcpu, kvm) {
467 		if (!VALID_PAGE(tmp_eptp)) {
468 			tmp_eptp = to_vmx(vcpu)->ept_pointer;
469 		} else if (tmp_eptp != to_vmx(vcpu)->ept_pointer) {
470 			to_kvm_vmx(kvm)->ept_pointers_match
471 				= EPT_POINTERS_MISMATCH;
472 			return;
473 		}
474 	}
475 
476 	to_kvm_vmx(kvm)->ept_pointers_match = EPT_POINTERS_MATCH;
477 }
478 
479 static int kvm_fill_hv_flush_list_func(struct hv_guest_mapping_flush_list *flush,
480 		void *data)
481 {
482 	struct kvm_tlb_range *range = data;
483 
484 	return hyperv_fill_flush_guest_mapping_list(flush, range->start_gfn,
485 			range->pages);
486 }
487 
488 static inline int __hv_remote_flush_tlb_with_range(struct kvm *kvm,
489 		struct kvm_vcpu *vcpu, struct kvm_tlb_range *range)
490 {
491 	u64 ept_pointer = to_vmx(vcpu)->ept_pointer;
492 
493 	/*
494 	 * FLUSH_GUEST_PHYSICAL_ADDRESS_SPACE hypercall needs address
495 	 * of the base of EPT PML4 table, strip off EPT configuration
496 	 * information.
497 	 */
498 	if (range)
499 		return hyperv_flush_guest_mapping_range(ept_pointer & PAGE_MASK,
500 				kvm_fill_hv_flush_list_func, (void *)range);
501 	else
502 		return hyperv_flush_guest_mapping(ept_pointer & PAGE_MASK);
503 }
504 
505 static int hv_remote_flush_tlb_with_range(struct kvm *kvm,
506 		struct kvm_tlb_range *range)
507 {
508 	struct kvm_vcpu *vcpu;
509 	int ret = 0, i;
510 
511 	spin_lock(&to_kvm_vmx(kvm)->ept_pointer_lock);
512 
513 	if (to_kvm_vmx(kvm)->ept_pointers_match == EPT_POINTERS_CHECK)
514 		check_ept_pointer_match(kvm);
515 
516 	if (to_kvm_vmx(kvm)->ept_pointers_match != EPT_POINTERS_MATCH) {
517 		kvm_for_each_vcpu(i, vcpu, kvm) {
518 			/* If ept_pointer is invalid pointer, bypass flush request. */
519 			if (VALID_PAGE(to_vmx(vcpu)->ept_pointer))
520 				ret |= __hv_remote_flush_tlb_with_range(
521 					kvm, vcpu, range);
522 		}
523 	} else {
524 		ret = __hv_remote_flush_tlb_with_range(kvm,
525 				kvm_get_vcpu(kvm, 0), range);
526 	}
527 
528 	spin_unlock(&to_kvm_vmx(kvm)->ept_pointer_lock);
529 	return ret;
530 }
531 static int hv_remote_flush_tlb(struct kvm *kvm)
532 {
533 	return hv_remote_flush_tlb_with_range(kvm, NULL);
534 }
535 
536 static int hv_enable_direct_tlbflush(struct kvm_vcpu *vcpu)
537 {
538 	struct hv_enlightened_vmcs *evmcs;
539 	struct hv_partition_assist_pg **p_hv_pa_pg =
540 			&vcpu->kvm->arch.hyperv.hv_pa_pg;
541 	/*
542 	 * Synthetic VM-Exit is not enabled in current code and so All
543 	 * evmcs in singe VM shares same assist page.
544 	 */
545 	if (!*p_hv_pa_pg)
546 		*p_hv_pa_pg = kzalloc(PAGE_SIZE, GFP_KERNEL);
547 
548 	if (!*p_hv_pa_pg)
549 		return -ENOMEM;
550 
551 	evmcs = (struct hv_enlightened_vmcs *)to_vmx(vcpu)->loaded_vmcs->vmcs;
552 
553 	evmcs->partition_assist_page =
554 		__pa(*p_hv_pa_pg);
555 	evmcs->hv_vm_id = (unsigned long)vcpu->kvm;
556 	evmcs->hv_enlightenments_control.nested_flush_hypercall = 1;
557 
558 	return 0;
559 }
560 
561 #endif /* IS_ENABLED(CONFIG_HYPERV) */
562 
563 /*
564  * Comment's format: document - errata name - stepping - processor name.
565  * Refer from
566  * https://www.virtualbox.org/svn/vbox/trunk/src/VBox/VMM/VMMR0/HMR0.cpp
567  */
568 static u32 vmx_preemption_cpu_tfms[] = {
569 /* 323344.pdf - BA86   - D0 - Xeon 7500 Series */
570 0x000206E6,
571 /* 323056.pdf - AAX65  - C2 - Xeon L3406 */
572 /* 322814.pdf - AAT59  - C2 - i7-600, i5-500, i5-400 and i3-300 Mobile */
573 /* 322911.pdf - AAU65  - C2 - i5-600, i3-500 Desktop and Pentium G6950 */
574 0x00020652,
575 /* 322911.pdf - AAU65  - K0 - i5-600, i3-500 Desktop and Pentium G6950 */
576 0x00020655,
577 /* 322373.pdf - AAO95  - B1 - Xeon 3400 Series */
578 /* 322166.pdf - AAN92  - B1 - i7-800 and i5-700 Desktop */
579 /*
580  * 320767.pdf - AAP86  - B1 -
581  * i7-900 Mobile Extreme, i7-800 and i7-700 Mobile
582  */
583 0x000106E5,
584 /* 321333.pdf - AAM126 - C0 - Xeon 3500 */
585 0x000106A0,
586 /* 321333.pdf - AAM126 - C1 - Xeon 3500 */
587 0x000106A1,
588 /* 320836.pdf - AAJ124 - C0 - i7-900 Desktop Extreme and i7-900 Desktop */
589 0x000106A4,
590  /* 321333.pdf - AAM126 - D0 - Xeon 3500 */
591  /* 321324.pdf - AAK139 - D0 - Xeon 5500 */
592  /* 320836.pdf - AAJ124 - D0 - i7-900 Extreme and i7-900 Desktop */
593 0x000106A5,
594  /* Xeon E3-1220 V2 */
595 0x000306A8,
596 };
597 
598 static inline bool cpu_has_broken_vmx_preemption_timer(void)
599 {
600 	u32 eax = cpuid_eax(0x00000001), i;
601 
602 	/* Clear the reserved bits */
603 	eax &= ~(0x3U << 14 | 0xfU << 28);
604 	for (i = 0; i < ARRAY_SIZE(vmx_preemption_cpu_tfms); i++)
605 		if (eax == vmx_preemption_cpu_tfms[i])
606 			return true;
607 
608 	return false;
609 }
610 
611 static inline bool cpu_need_virtualize_apic_accesses(struct kvm_vcpu *vcpu)
612 {
613 	return flexpriority_enabled && lapic_in_kernel(vcpu);
614 }
615 
616 static inline bool report_flexpriority(void)
617 {
618 	return flexpriority_enabled;
619 }
620 
621 static inline int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
622 {
623 	int i;
624 
625 	for (i = 0; i < vmx->nmsrs; ++i)
626 		if (vmx_msr_index[vmx->guest_msrs[i].index] == msr)
627 			return i;
628 	return -1;
629 }
630 
631 struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
632 {
633 	int i;
634 
635 	i = __find_msr_index(vmx, msr);
636 	if (i >= 0)
637 		return &vmx->guest_msrs[i];
638 	return NULL;
639 }
640 
641 void loaded_vmcs_init(struct loaded_vmcs *loaded_vmcs)
642 {
643 	vmcs_clear(loaded_vmcs->vmcs);
644 	if (loaded_vmcs->shadow_vmcs && loaded_vmcs->launched)
645 		vmcs_clear(loaded_vmcs->shadow_vmcs);
646 	loaded_vmcs->cpu = -1;
647 	loaded_vmcs->launched = 0;
648 }
649 
650 #ifdef CONFIG_KEXEC_CORE
651 /*
652  * This bitmap is used to indicate whether the vmclear
653  * operation is enabled on all cpus. All disabled by
654  * default.
655  */
656 static cpumask_t crash_vmclear_enabled_bitmap = CPU_MASK_NONE;
657 
658 static inline void crash_enable_local_vmclear(int cpu)
659 {
660 	cpumask_set_cpu(cpu, &crash_vmclear_enabled_bitmap);
661 }
662 
663 static inline void crash_disable_local_vmclear(int cpu)
664 {
665 	cpumask_clear_cpu(cpu, &crash_vmclear_enabled_bitmap);
666 }
667 
668 static inline int crash_local_vmclear_enabled(int cpu)
669 {
670 	return cpumask_test_cpu(cpu, &crash_vmclear_enabled_bitmap);
671 }
672 
673 static void crash_vmclear_local_loaded_vmcss(void)
674 {
675 	int cpu = raw_smp_processor_id();
676 	struct loaded_vmcs *v;
677 
678 	if (!crash_local_vmclear_enabled(cpu))
679 		return;
680 
681 	list_for_each_entry(v, &per_cpu(loaded_vmcss_on_cpu, cpu),
682 			    loaded_vmcss_on_cpu_link)
683 		vmcs_clear(v->vmcs);
684 }
685 #else
686 static inline void crash_enable_local_vmclear(int cpu) { }
687 static inline void crash_disable_local_vmclear(int cpu) { }
688 #endif /* CONFIG_KEXEC_CORE */
689 
690 static void __loaded_vmcs_clear(void *arg)
691 {
692 	struct loaded_vmcs *loaded_vmcs = arg;
693 	int cpu = raw_smp_processor_id();
694 
695 	if (loaded_vmcs->cpu != cpu)
696 		return; /* vcpu migration can race with cpu offline */
697 	if (per_cpu(current_vmcs, cpu) == loaded_vmcs->vmcs)
698 		per_cpu(current_vmcs, cpu) = NULL;
699 	crash_disable_local_vmclear(cpu);
700 	list_del(&loaded_vmcs->loaded_vmcss_on_cpu_link);
701 
702 	/*
703 	 * we should ensure updating loaded_vmcs->loaded_vmcss_on_cpu_link
704 	 * is before setting loaded_vmcs->vcpu to -1 which is done in
705 	 * loaded_vmcs_init. Otherwise, other cpu can see vcpu = -1 fist
706 	 * then adds the vmcs into percpu list before it is deleted.
707 	 */
708 	smp_wmb();
709 
710 	loaded_vmcs_init(loaded_vmcs);
711 	crash_enable_local_vmclear(cpu);
712 }
713 
714 void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs)
715 {
716 	int cpu = loaded_vmcs->cpu;
717 
718 	if (cpu != -1)
719 		smp_call_function_single(cpu,
720 			 __loaded_vmcs_clear, loaded_vmcs, 1);
721 }
722 
723 static bool vmx_segment_cache_test_set(struct vcpu_vmx *vmx, unsigned seg,
724 				       unsigned field)
725 {
726 	bool ret;
727 	u32 mask = 1 << (seg * SEG_FIELD_NR + field);
728 
729 	if (!(vmx->vcpu.arch.regs_avail & (1 << VCPU_EXREG_SEGMENTS))) {
730 		vmx->vcpu.arch.regs_avail |= (1 << VCPU_EXREG_SEGMENTS);
731 		vmx->segment_cache.bitmask = 0;
732 	}
733 	ret = vmx->segment_cache.bitmask & mask;
734 	vmx->segment_cache.bitmask |= mask;
735 	return ret;
736 }
737 
738 static u16 vmx_read_guest_seg_selector(struct vcpu_vmx *vmx, unsigned seg)
739 {
740 	u16 *p = &vmx->segment_cache.seg[seg].selector;
741 
742 	if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_SEL))
743 		*p = vmcs_read16(kvm_vmx_segment_fields[seg].selector);
744 	return *p;
745 }
746 
747 static ulong vmx_read_guest_seg_base(struct vcpu_vmx *vmx, unsigned seg)
748 {
749 	ulong *p = &vmx->segment_cache.seg[seg].base;
750 
751 	if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_BASE))
752 		*p = vmcs_readl(kvm_vmx_segment_fields[seg].base);
753 	return *p;
754 }
755 
756 static u32 vmx_read_guest_seg_limit(struct vcpu_vmx *vmx, unsigned seg)
757 {
758 	u32 *p = &vmx->segment_cache.seg[seg].limit;
759 
760 	if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_LIMIT))
761 		*p = vmcs_read32(kvm_vmx_segment_fields[seg].limit);
762 	return *p;
763 }
764 
765 static u32 vmx_read_guest_seg_ar(struct vcpu_vmx *vmx, unsigned seg)
766 {
767 	u32 *p = &vmx->segment_cache.seg[seg].ar;
768 
769 	if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_AR))
770 		*p = vmcs_read32(kvm_vmx_segment_fields[seg].ar_bytes);
771 	return *p;
772 }
773 
774 void update_exception_bitmap(struct kvm_vcpu *vcpu)
775 {
776 	u32 eb;
777 
778 	eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR) |
779 	     (1u << DB_VECTOR) | (1u << AC_VECTOR);
780 	/*
781 	 * Guest access to VMware backdoor ports could legitimately
782 	 * trigger #GP because of TSS I/O permission bitmap.
783 	 * We intercept those #GP and allow access to them anyway
784 	 * as VMware does.
785 	 */
786 	if (enable_vmware_backdoor)
787 		eb |= (1u << GP_VECTOR);
788 	if ((vcpu->guest_debug &
789 	     (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) ==
790 	    (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP))
791 		eb |= 1u << BP_VECTOR;
792 	if (to_vmx(vcpu)->rmode.vm86_active)
793 		eb = ~0;
794 	if (enable_ept)
795 		eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */
796 
797 	/* When we are running a nested L2 guest and L1 specified for it a
798 	 * certain exception bitmap, we must trap the same exceptions and pass
799 	 * them to L1. When running L2, we will only handle the exceptions
800 	 * specified above if L1 did not want them.
801 	 */
802 	if (is_guest_mode(vcpu))
803 		eb |= get_vmcs12(vcpu)->exception_bitmap;
804 
805 	vmcs_write32(EXCEPTION_BITMAP, eb);
806 }
807 
808 /*
809  * Check if MSR is intercepted for currently loaded MSR bitmap.
810  */
811 static bool msr_write_intercepted(struct kvm_vcpu *vcpu, u32 msr)
812 {
813 	unsigned long *msr_bitmap;
814 	int f = sizeof(unsigned long);
815 
816 	if (!cpu_has_vmx_msr_bitmap())
817 		return true;
818 
819 	msr_bitmap = to_vmx(vcpu)->loaded_vmcs->msr_bitmap;
820 
821 	if (msr <= 0x1fff) {
822 		return !!test_bit(msr, msr_bitmap + 0x800 / f);
823 	} else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
824 		msr &= 0x1fff;
825 		return !!test_bit(msr, msr_bitmap + 0xc00 / f);
826 	}
827 
828 	return true;
829 }
830 
831 static void clear_atomic_switch_msr_special(struct vcpu_vmx *vmx,
832 		unsigned long entry, unsigned long exit)
833 {
834 	vm_entry_controls_clearbit(vmx, entry);
835 	vm_exit_controls_clearbit(vmx, exit);
836 }
837 
838 static int find_msr(struct vmx_msrs *m, unsigned int msr)
839 {
840 	unsigned int i;
841 
842 	for (i = 0; i < m->nr; ++i) {
843 		if (m->val[i].index == msr)
844 			return i;
845 	}
846 	return -ENOENT;
847 }
848 
849 static void clear_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr)
850 {
851 	int i;
852 	struct msr_autoload *m = &vmx->msr_autoload;
853 
854 	switch (msr) {
855 	case MSR_EFER:
856 		if (cpu_has_load_ia32_efer()) {
857 			clear_atomic_switch_msr_special(vmx,
858 					VM_ENTRY_LOAD_IA32_EFER,
859 					VM_EXIT_LOAD_IA32_EFER);
860 			return;
861 		}
862 		break;
863 	case MSR_CORE_PERF_GLOBAL_CTRL:
864 		if (cpu_has_load_perf_global_ctrl()) {
865 			clear_atomic_switch_msr_special(vmx,
866 					VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
867 					VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
868 			return;
869 		}
870 		break;
871 	}
872 	i = find_msr(&m->guest, msr);
873 	if (i < 0)
874 		goto skip_guest;
875 	--m->guest.nr;
876 	m->guest.val[i] = m->guest.val[m->guest.nr];
877 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->guest.nr);
878 
879 skip_guest:
880 	i = find_msr(&m->host, msr);
881 	if (i < 0)
882 		return;
883 
884 	--m->host.nr;
885 	m->host.val[i] = m->host.val[m->host.nr];
886 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->host.nr);
887 }
888 
889 static void add_atomic_switch_msr_special(struct vcpu_vmx *vmx,
890 		unsigned long entry, unsigned long exit,
891 		unsigned long guest_val_vmcs, unsigned long host_val_vmcs,
892 		u64 guest_val, u64 host_val)
893 {
894 	vmcs_write64(guest_val_vmcs, guest_val);
895 	if (host_val_vmcs != HOST_IA32_EFER)
896 		vmcs_write64(host_val_vmcs, host_val);
897 	vm_entry_controls_setbit(vmx, entry);
898 	vm_exit_controls_setbit(vmx, exit);
899 }
900 
901 static void add_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr,
902 				  u64 guest_val, u64 host_val, bool entry_only)
903 {
904 	int i, j = 0;
905 	struct msr_autoload *m = &vmx->msr_autoload;
906 
907 	switch (msr) {
908 	case MSR_EFER:
909 		if (cpu_has_load_ia32_efer()) {
910 			add_atomic_switch_msr_special(vmx,
911 					VM_ENTRY_LOAD_IA32_EFER,
912 					VM_EXIT_LOAD_IA32_EFER,
913 					GUEST_IA32_EFER,
914 					HOST_IA32_EFER,
915 					guest_val, host_val);
916 			return;
917 		}
918 		break;
919 	case MSR_CORE_PERF_GLOBAL_CTRL:
920 		if (cpu_has_load_perf_global_ctrl()) {
921 			add_atomic_switch_msr_special(vmx,
922 					VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
923 					VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL,
924 					GUEST_IA32_PERF_GLOBAL_CTRL,
925 					HOST_IA32_PERF_GLOBAL_CTRL,
926 					guest_val, host_val);
927 			return;
928 		}
929 		break;
930 	case MSR_IA32_PEBS_ENABLE:
931 		/* PEBS needs a quiescent period after being disabled (to write
932 		 * a record).  Disabling PEBS through VMX MSR swapping doesn't
933 		 * provide that period, so a CPU could write host's record into
934 		 * guest's memory.
935 		 */
936 		wrmsrl(MSR_IA32_PEBS_ENABLE, 0);
937 	}
938 
939 	i = find_msr(&m->guest, msr);
940 	if (!entry_only)
941 		j = find_msr(&m->host, msr);
942 
943 	if ((i < 0 && m->guest.nr == NR_AUTOLOAD_MSRS) ||
944 		(j < 0 &&  m->host.nr == NR_AUTOLOAD_MSRS)) {
945 		printk_once(KERN_WARNING "Not enough msr switch entries. "
946 				"Can't add msr %x\n", msr);
947 		return;
948 	}
949 	if (i < 0) {
950 		i = m->guest.nr++;
951 		vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->guest.nr);
952 	}
953 	m->guest.val[i].index = msr;
954 	m->guest.val[i].value = guest_val;
955 
956 	if (entry_only)
957 		return;
958 
959 	if (j < 0) {
960 		j = m->host.nr++;
961 		vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->host.nr);
962 	}
963 	m->host.val[j].index = msr;
964 	m->host.val[j].value = host_val;
965 }
966 
967 static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset)
968 {
969 	u64 guest_efer = vmx->vcpu.arch.efer;
970 	u64 ignore_bits = 0;
971 
972 	if (!enable_ept) {
973 		/*
974 		 * NX is needed to handle CR0.WP=1, CR4.SMEP=1.  Testing
975 		 * host CPUID is more efficient than testing guest CPUID
976 		 * or CR4.  Host SMEP is anyway a requirement for guest SMEP.
977 		 */
978 		if (boot_cpu_has(X86_FEATURE_SMEP))
979 			guest_efer |= EFER_NX;
980 		else if (!(guest_efer & EFER_NX))
981 			ignore_bits |= EFER_NX;
982 	}
983 
984 	/*
985 	 * LMA and LME handled by hardware; SCE meaningless outside long mode.
986 	 */
987 	ignore_bits |= EFER_SCE;
988 #ifdef CONFIG_X86_64
989 	ignore_bits |= EFER_LMA | EFER_LME;
990 	/* SCE is meaningful only in long mode on Intel */
991 	if (guest_efer & EFER_LMA)
992 		ignore_bits &= ~(u64)EFER_SCE;
993 #endif
994 
995 	/*
996 	 * On EPT, we can't emulate NX, so we must switch EFER atomically.
997 	 * On CPUs that support "load IA32_EFER", always switch EFER
998 	 * atomically, since it's faster than switching it manually.
999 	 */
1000 	if (cpu_has_load_ia32_efer() ||
1001 	    (enable_ept && ((vmx->vcpu.arch.efer ^ host_efer) & EFER_NX))) {
1002 		if (!(guest_efer & EFER_LMA))
1003 			guest_efer &= ~EFER_LME;
1004 		if (guest_efer != host_efer)
1005 			add_atomic_switch_msr(vmx, MSR_EFER,
1006 					      guest_efer, host_efer, false);
1007 		else
1008 			clear_atomic_switch_msr(vmx, MSR_EFER);
1009 		return false;
1010 	} else {
1011 		clear_atomic_switch_msr(vmx, MSR_EFER);
1012 
1013 		guest_efer &= ~ignore_bits;
1014 		guest_efer |= host_efer & ignore_bits;
1015 
1016 		vmx->guest_msrs[efer_offset].data = guest_efer;
1017 		vmx->guest_msrs[efer_offset].mask = ~ignore_bits;
1018 
1019 		return true;
1020 	}
1021 }
1022 
1023 #ifdef CONFIG_X86_32
1024 /*
1025  * On 32-bit kernels, VM exits still load the FS and GS bases from the
1026  * VMCS rather than the segment table.  KVM uses this helper to figure
1027  * out the current bases to poke them into the VMCS before entry.
1028  */
1029 static unsigned long segment_base(u16 selector)
1030 {
1031 	struct desc_struct *table;
1032 	unsigned long v;
1033 
1034 	if (!(selector & ~SEGMENT_RPL_MASK))
1035 		return 0;
1036 
1037 	table = get_current_gdt_ro();
1038 
1039 	if ((selector & SEGMENT_TI_MASK) == SEGMENT_LDT) {
1040 		u16 ldt_selector = kvm_read_ldt();
1041 
1042 		if (!(ldt_selector & ~SEGMENT_RPL_MASK))
1043 			return 0;
1044 
1045 		table = (struct desc_struct *)segment_base(ldt_selector);
1046 	}
1047 	v = get_desc_base(&table[selector >> 3]);
1048 	return v;
1049 }
1050 #endif
1051 
1052 static inline void pt_load_msr(struct pt_ctx *ctx, u32 addr_range)
1053 {
1054 	u32 i;
1055 
1056 	wrmsrl(MSR_IA32_RTIT_STATUS, ctx->status);
1057 	wrmsrl(MSR_IA32_RTIT_OUTPUT_BASE, ctx->output_base);
1058 	wrmsrl(MSR_IA32_RTIT_OUTPUT_MASK, ctx->output_mask);
1059 	wrmsrl(MSR_IA32_RTIT_CR3_MATCH, ctx->cr3_match);
1060 	for (i = 0; i < addr_range; i++) {
1061 		wrmsrl(MSR_IA32_RTIT_ADDR0_A + i * 2, ctx->addr_a[i]);
1062 		wrmsrl(MSR_IA32_RTIT_ADDR0_B + i * 2, ctx->addr_b[i]);
1063 	}
1064 }
1065 
1066 static inline void pt_save_msr(struct pt_ctx *ctx, u32 addr_range)
1067 {
1068 	u32 i;
1069 
1070 	rdmsrl(MSR_IA32_RTIT_STATUS, ctx->status);
1071 	rdmsrl(MSR_IA32_RTIT_OUTPUT_BASE, ctx->output_base);
1072 	rdmsrl(MSR_IA32_RTIT_OUTPUT_MASK, ctx->output_mask);
1073 	rdmsrl(MSR_IA32_RTIT_CR3_MATCH, ctx->cr3_match);
1074 	for (i = 0; i < addr_range; i++) {
1075 		rdmsrl(MSR_IA32_RTIT_ADDR0_A + i * 2, ctx->addr_a[i]);
1076 		rdmsrl(MSR_IA32_RTIT_ADDR0_B + i * 2, ctx->addr_b[i]);
1077 	}
1078 }
1079 
1080 static void pt_guest_enter(struct vcpu_vmx *vmx)
1081 {
1082 	if (pt_mode == PT_MODE_SYSTEM)
1083 		return;
1084 
1085 	/*
1086 	 * GUEST_IA32_RTIT_CTL is already set in the VMCS.
1087 	 * Save host state before VM entry.
1088 	 */
1089 	rdmsrl(MSR_IA32_RTIT_CTL, vmx->pt_desc.host.ctl);
1090 	if (vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) {
1091 		wrmsrl(MSR_IA32_RTIT_CTL, 0);
1092 		pt_save_msr(&vmx->pt_desc.host, vmx->pt_desc.addr_range);
1093 		pt_load_msr(&vmx->pt_desc.guest, vmx->pt_desc.addr_range);
1094 	}
1095 }
1096 
1097 static void pt_guest_exit(struct vcpu_vmx *vmx)
1098 {
1099 	if (pt_mode == PT_MODE_SYSTEM)
1100 		return;
1101 
1102 	if (vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) {
1103 		pt_save_msr(&vmx->pt_desc.guest, vmx->pt_desc.addr_range);
1104 		pt_load_msr(&vmx->pt_desc.host, vmx->pt_desc.addr_range);
1105 	}
1106 
1107 	/* Reload host state (IA32_RTIT_CTL will be cleared on VM exit). */
1108 	wrmsrl(MSR_IA32_RTIT_CTL, vmx->pt_desc.host.ctl);
1109 }
1110 
1111 void vmx_set_host_fs_gs(struct vmcs_host_state *host, u16 fs_sel, u16 gs_sel,
1112 			unsigned long fs_base, unsigned long gs_base)
1113 {
1114 	if (unlikely(fs_sel != host->fs_sel)) {
1115 		if (!(fs_sel & 7))
1116 			vmcs_write16(HOST_FS_SELECTOR, fs_sel);
1117 		else
1118 			vmcs_write16(HOST_FS_SELECTOR, 0);
1119 		host->fs_sel = fs_sel;
1120 	}
1121 	if (unlikely(gs_sel != host->gs_sel)) {
1122 		if (!(gs_sel & 7))
1123 			vmcs_write16(HOST_GS_SELECTOR, gs_sel);
1124 		else
1125 			vmcs_write16(HOST_GS_SELECTOR, 0);
1126 		host->gs_sel = gs_sel;
1127 	}
1128 	if (unlikely(fs_base != host->fs_base)) {
1129 		vmcs_writel(HOST_FS_BASE, fs_base);
1130 		host->fs_base = fs_base;
1131 	}
1132 	if (unlikely(gs_base != host->gs_base)) {
1133 		vmcs_writel(HOST_GS_BASE, gs_base);
1134 		host->gs_base = gs_base;
1135 	}
1136 }
1137 
1138 void vmx_prepare_switch_to_guest(struct kvm_vcpu *vcpu)
1139 {
1140 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1141 	struct vmcs_host_state *host_state;
1142 #ifdef CONFIG_X86_64
1143 	int cpu = raw_smp_processor_id();
1144 #endif
1145 	unsigned long fs_base, gs_base;
1146 	u16 fs_sel, gs_sel;
1147 	int i;
1148 
1149 	vmx->req_immediate_exit = false;
1150 
1151 	/*
1152 	 * Note that guest MSRs to be saved/restored can also be changed
1153 	 * when guest state is loaded. This happens when guest transitions
1154 	 * to/from long-mode by setting MSR_EFER.LMA.
1155 	 */
1156 	if (!vmx->guest_msrs_ready) {
1157 		vmx->guest_msrs_ready = true;
1158 		for (i = 0; i < vmx->save_nmsrs; ++i)
1159 			kvm_set_shared_msr(vmx->guest_msrs[i].index,
1160 					   vmx->guest_msrs[i].data,
1161 					   vmx->guest_msrs[i].mask);
1162 
1163 	}
1164 	if (vmx->guest_state_loaded)
1165 		return;
1166 
1167 	host_state = &vmx->loaded_vmcs->host_state;
1168 
1169 	/*
1170 	 * Set host fs and gs selectors.  Unfortunately, 22.2.3 does not
1171 	 * allow segment selectors with cpl > 0 or ti == 1.
1172 	 */
1173 	host_state->ldt_sel = kvm_read_ldt();
1174 
1175 #ifdef CONFIG_X86_64
1176 	savesegment(ds, host_state->ds_sel);
1177 	savesegment(es, host_state->es_sel);
1178 
1179 	gs_base = cpu_kernelmode_gs_base(cpu);
1180 	if (likely(is_64bit_mm(current->mm))) {
1181 		save_fsgs_for_kvm();
1182 		fs_sel = current->thread.fsindex;
1183 		gs_sel = current->thread.gsindex;
1184 		fs_base = current->thread.fsbase;
1185 		vmx->msr_host_kernel_gs_base = current->thread.gsbase;
1186 	} else {
1187 		savesegment(fs, fs_sel);
1188 		savesegment(gs, gs_sel);
1189 		fs_base = read_msr(MSR_FS_BASE);
1190 		vmx->msr_host_kernel_gs_base = read_msr(MSR_KERNEL_GS_BASE);
1191 	}
1192 
1193 	wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1194 #else
1195 	savesegment(fs, fs_sel);
1196 	savesegment(gs, gs_sel);
1197 	fs_base = segment_base(fs_sel);
1198 	gs_base = segment_base(gs_sel);
1199 #endif
1200 
1201 	vmx_set_host_fs_gs(host_state, fs_sel, gs_sel, fs_base, gs_base);
1202 	vmx->guest_state_loaded = true;
1203 }
1204 
1205 static void vmx_prepare_switch_to_host(struct vcpu_vmx *vmx)
1206 {
1207 	struct vmcs_host_state *host_state;
1208 
1209 	if (!vmx->guest_state_loaded)
1210 		return;
1211 
1212 	host_state = &vmx->loaded_vmcs->host_state;
1213 
1214 	++vmx->vcpu.stat.host_state_reload;
1215 
1216 #ifdef CONFIG_X86_64
1217 	rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1218 #endif
1219 	if (host_state->ldt_sel || (host_state->gs_sel & 7)) {
1220 		kvm_load_ldt(host_state->ldt_sel);
1221 #ifdef CONFIG_X86_64
1222 		load_gs_index(host_state->gs_sel);
1223 #else
1224 		loadsegment(gs, host_state->gs_sel);
1225 #endif
1226 	}
1227 	if (host_state->fs_sel & 7)
1228 		loadsegment(fs, host_state->fs_sel);
1229 #ifdef CONFIG_X86_64
1230 	if (unlikely(host_state->ds_sel | host_state->es_sel)) {
1231 		loadsegment(ds, host_state->ds_sel);
1232 		loadsegment(es, host_state->es_sel);
1233 	}
1234 #endif
1235 	invalidate_tss_limit();
1236 #ifdef CONFIG_X86_64
1237 	wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
1238 #endif
1239 	load_fixmap_gdt(raw_smp_processor_id());
1240 	vmx->guest_state_loaded = false;
1241 	vmx->guest_msrs_ready = false;
1242 }
1243 
1244 #ifdef CONFIG_X86_64
1245 static u64 vmx_read_guest_kernel_gs_base(struct vcpu_vmx *vmx)
1246 {
1247 	preempt_disable();
1248 	if (vmx->guest_state_loaded)
1249 		rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1250 	preempt_enable();
1251 	return vmx->msr_guest_kernel_gs_base;
1252 }
1253 
1254 static void vmx_write_guest_kernel_gs_base(struct vcpu_vmx *vmx, u64 data)
1255 {
1256 	preempt_disable();
1257 	if (vmx->guest_state_loaded)
1258 		wrmsrl(MSR_KERNEL_GS_BASE, data);
1259 	preempt_enable();
1260 	vmx->msr_guest_kernel_gs_base = data;
1261 }
1262 #endif
1263 
1264 static void vmx_vcpu_pi_load(struct kvm_vcpu *vcpu, int cpu)
1265 {
1266 	struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
1267 	struct pi_desc old, new;
1268 	unsigned int dest;
1269 
1270 	/*
1271 	 * In case of hot-plug or hot-unplug, we may have to undo
1272 	 * vmx_vcpu_pi_put even if there is no assigned device.  And we
1273 	 * always keep PI.NDST up to date for simplicity: it makes the
1274 	 * code easier, and CPU migration is not a fast path.
1275 	 */
1276 	if (!pi_test_sn(pi_desc) && vcpu->cpu == cpu)
1277 		return;
1278 
1279 	/* The full case.  */
1280 	do {
1281 		old.control = new.control = pi_desc->control;
1282 
1283 		dest = cpu_physical_id(cpu);
1284 
1285 		if (x2apic_enabled())
1286 			new.ndst = dest;
1287 		else
1288 			new.ndst = (dest << 8) & 0xFF00;
1289 
1290 		new.sn = 0;
1291 	} while (cmpxchg64(&pi_desc->control, old.control,
1292 			   new.control) != old.control);
1293 
1294 	/*
1295 	 * Clear SN before reading the bitmap.  The VT-d firmware
1296 	 * writes the bitmap and reads SN atomically (5.2.3 in the
1297 	 * spec), so it doesn't really have a memory barrier that
1298 	 * pairs with this, but we cannot do that and we need one.
1299 	 */
1300 	smp_mb__after_atomic();
1301 
1302 	if (!bitmap_empty((unsigned long *)pi_desc->pir, NR_VECTORS))
1303 		pi_set_on(pi_desc);
1304 }
1305 
1306 void vmx_vcpu_load_vmcs(struct kvm_vcpu *vcpu, int cpu)
1307 {
1308 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1309 	bool already_loaded = vmx->loaded_vmcs->cpu == cpu;
1310 
1311 	if (!already_loaded) {
1312 		loaded_vmcs_clear(vmx->loaded_vmcs);
1313 		local_irq_disable();
1314 		crash_disable_local_vmclear(cpu);
1315 
1316 		/*
1317 		 * Read loaded_vmcs->cpu should be before fetching
1318 		 * loaded_vmcs->loaded_vmcss_on_cpu_link.
1319 		 * See the comments in __loaded_vmcs_clear().
1320 		 */
1321 		smp_rmb();
1322 
1323 		list_add(&vmx->loaded_vmcs->loaded_vmcss_on_cpu_link,
1324 			 &per_cpu(loaded_vmcss_on_cpu, cpu));
1325 		crash_enable_local_vmclear(cpu);
1326 		local_irq_enable();
1327 	}
1328 
1329 	if (per_cpu(current_vmcs, cpu) != vmx->loaded_vmcs->vmcs) {
1330 		per_cpu(current_vmcs, cpu) = vmx->loaded_vmcs->vmcs;
1331 		vmcs_load(vmx->loaded_vmcs->vmcs);
1332 		indirect_branch_prediction_barrier();
1333 	}
1334 
1335 	if (!already_loaded) {
1336 		void *gdt = get_current_gdt_ro();
1337 		unsigned long sysenter_esp;
1338 
1339 		kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1340 
1341 		/*
1342 		 * Linux uses per-cpu TSS and GDT, so set these when switching
1343 		 * processors.  See 22.2.4.
1344 		 */
1345 		vmcs_writel(HOST_TR_BASE,
1346 			    (unsigned long)&get_cpu_entry_area(cpu)->tss.x86_tss);
1347 		vmcs_writel(HOST_GDTR_BASE, (unsigned long)gdt);   /* 22.2.4 */
1348 
1349 		/*
1350 		 * VM exits change the host TR limit to 0x67 after a VM
1351 		 * exit.  This is okay, since 0x67 covers everything except
1352 		 * the IO bitmap and have have code to handle the IO bitmap
1353 		 * being lost after a VM exit.
1354 		 */
1355 		BUILD_BUG_ON(IO_BITMAP_OFFSET - 1 != 0x67);
1356 
1357 		rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
1358 		vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
1359 
1360 		vmx->loaded_vmcs->cpu = cpu;
1361 	}
1362 
1363 	/* Setup TSC multiplier */
1364 	if (kvm_has_tsc_control &&
1365 	    vmx->current_tsc_ratio != vcpu->arch.tsc_scaling_ratio)
1366 		decache_tsc_multiplier(vmx);
1367 }
1368 
1369 /*
1370  * Switches to specified vcpu, until a matching vcpu_put(), but assumes
1371  * vcpu mutex is already taken.
1372  */
1373 void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1374 {
1375 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1376 
1377 	vmx_vcpu_load_vmcs(vcpu, cpu);
1378 
1379 	vmx_vcpu_pi_load(vcpu, cpu);
1380 
1381 	vmx->host_pkru = read_pkru();
1382 	vmx->host_debugctlmsr = get_debugctlmsr();
1383 }
1384 
1385 static void vmx_vcpu_pi_put(struct kvm_vcpu *vcpu)
1386 {
1387 	struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
1388 
1389 	if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
1390 		!irq_remapping_cap(IRQ_POSTING_CAP)  ||
1391 		!kvm_vcpu_apicv_active(vcpu))
1392 		return;
1393 
1394 	/* Set SN when the vCPU is preempted */
1395 	if (vcpu->preempted)
1396 		pi_set_sn(pi_desc);
1397 }
1398 
1399 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
1400 {
1401 	vmx_vcpu_pi_put(vcpu);
1402 
1403 	vmx_prepare_switch_to_host(to_vmx(vcpu));
1404 }
1405 
1406 static bool emulation_required(struct kvm_vcpu *vcpu)
1407 {
1408 	return emulate_invalid_guest_state && !guest_state_valid(vcpu);
1409 }
1410 
1411 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu);
1412 
1413 unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
1414 {
1415 	unsigned long rflags, save_rflags;
1416 
1417 	if (!test_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail)) {
1418 		__set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
1419 		rflags = vmcs_readl(GUEST_RFLAGS);
1420 		if (to_vmx(vcpu)->rmode.vm86_active) {
1421 			rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
1422 			save_rflags = to_vmx(vcpu)->rmode.save_rflags;
1423 			rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
1424 		}
1425 		to_vmx(vcpu)->rflags = rflags;
1426 	}
1427 	return to_vmx(vcpu)->rflags;
1428 }
1429 
1430 void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
1431 {
1432 	unsigned long old_rflags = vmx_get_rflags(vcpu);
1433 
1434 	__set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
1435 	to_vmx(vcpu)->rflags = rflags;
1436 	if (to_vmx(vcpu)->rmode.vm86_active) {
1437 		to_vmx(vcpu)->rmode.save_rflags = rflags;
1438 		rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
1439 	}
1440 	vmcs_writel(GUEST_RFLAGS, rflags);
1441 
1442 	if ((old_rflags ^ to_vmx(vcpu)->rflags) & X86_EFLAGS_VM)
1443 		to_vmx(vcpu)->emulation_required = emulation_required(vcpu);
1444 }
1445 
1446 u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu)
1447 {
1448 	u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1449 	int ret = 0;
1450 
1451 	if (interruptibility & GUEST_INTR_STATE_STI)
1452 		ret |= KVM_X86_SHADOW_INT_STI;
1453 	if (interruptibility & GUEST_INTR_STATE_MOV_SS)
1454 		ret |= KVM_X86_SHADOW_INT_MOV_SS;
1455 
1456 	return ret;
1457 }
1458 
1459 void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
1460 {
1461 	u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1462 	u32 interruptibility = interruptibility_old;
1463 
1464 	interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
1465 
1466 	if (mask & KVM_X86_SHADOW_INT_MOV_SS)
1467 		interruptibility |= GUEST_INTR_STATE_MOV_SS;
1468 	else if (mask & KVM_X86_SHADOW_INT_STI)
1469 		interruptibility |= GUEST_INTR_STATE_STI;
1470 
1471 	if ((interruptibility != interruptibility_old))
1472 		vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
1473 }
1474 
1475 static int vmx_rtit_ctl_check(struct kvm_vcpu *vcpu, u64 data)
1476 {
1477 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1478 	unsigned long value;
1479 
1480 	/*
1481 	 * Any MSR write that attempts to change bits marked reserved will
1482 	 * case a #GP fault.
1483 	 */
1484 	if (data & vmx->pt_desc.ctl_bitmask)
1485 		return 1;
1486 
1487 	/*
1488 	 * Any attempt to modify IA32_RTIT_CTL while TraceEn is set will
1489 	 * result in a #GP unless the same write also clears TraceEn.
1490 	 */
1491 	if ((vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) &&
1492 		((vmx->pt_desc.guest.ctl ^ data) & ~RTIT_CTL_TRACEEN))
1493 		return 1;
1494 
1495 	/*
1496 	 * WRMSR to IA32_RTIT_CTL that sets TraceEn but clears this bit
1497 	 * and FabricEn would cause #GP, if
1498 	 * CPUID.(EAX=14H, ECX=0):ECX.SNGLRGNOUT[bit 2] = 0
1499 	 */
1500 	if ((data & RTIT_CTL_TRACEEN) && !(data & RTIT_CTL_TOPA) &&
1501 		!(data & RTIT_CTL_FABRIC_EN) &&
1502 		!intel_pt_validate_cap(vmx->pt_desc.caps,
1503 					PT_CAP_single_range_output))
1504 		return 1;
1505 
1506 	/*
1507 	 * MTCFreq, CycThresh and PSBFreq encodings check, any MSR write that
1508 	 * utilize encodings marked reserved will casue a #GP fault.
1509 	 */
1510 	value = intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc_periods);
1511 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc) &&
1512 			!test_bit((data & RTIT_CTL_MTC_RANGE) >>
1513 			RTIT_CTL_MTC_RANGE_OFFSET, &value))
1514 		return 1;
1515 	value = intel_pt_validate_cap(vmx->pt_desc.caps,
1516 						PT_CAP_cycle_thresholds);
1517 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc) &&
1518 			!test_bit((data & RTIT_CTL_CYC_THRESH) >>
1519 			RTIT_CTL_CYC_THRESH_OFFSET, &value))
1520 		return 1;
1521 	value = intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_periods);
1522 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc) &&
1523 			!test_bit((data & RTIT_CTL_PSB_FREQ) >>
1524 			RTIT_CTL_PSB_FREQ_OFFSET, &value))
1525 		return 1;
1526 
1527 	/*
1528 	 * If ADDRx_CFG is reserved or the encodings is >2 will
1529 	 * cause a #GP fault.
1530 	 */
1531 	value = (data & RTIT_CTL_ADDR0) >> RTIT_CTL_ADDR0_OFFSET;
1532 	if ((value && (vmx->pt_desc.addr_range < 1)) || (value > 2))
1533 		return 1;
1534 	value = (data & RTIT_CTL_ADDR1) >> RTIT_CTL_ADDR1_OFFSET;
1535 	if ((value && (vmx->pt_desc.addr_range < 2)) || (value > 2))
1536 		return 1;
1537 	value = (data & RTIT_CTL_ADDR2) >> RTIT_CTL_ADDR2_OFFSET;
1538 	if ((value && (vmx->pt_desc.addr_range < 3)) || (value > 2))
1539 		return 1;
1540 	value = (data & RTIT_CTL_ADDR3) >> RTIT_CTL_ADDR3_OFFSET;
1541 	if ((value && (vmx->pt_desc.addr_range < 4)) || (value > 2))
1542 		return 1;
1543 
1544 	return 0;
1545 }
1546 
1547 static int skip_emulated_instruction(struct kvm_vcpu *vcpu)
1548 {
1549 	unsigned long rip;
1550 
1551 	/*
1552 	 * Using VMCS.VM_EXIT_INSTRUCTION_LEN on EPT misconfig depends on
1553 	 * undefined behavior: Intel's SDM doesn't mandate the VMCS field be
1554 	 * set when EPT misconfig occurs.  In practice, real hardware updates
1555 	 * VM_EXIT_INSTRUCTION_LEN on EPT misconfig, but other hypervisors
1556 	 * (namely Hyper-V) don't set it due to it being undefined behavior,
1557 	 * i.e. we end up advancing IP with some random value.
1558 	 */
1559 	if (!static_cpu_has(X86_FEATURE_HYPERVISOR) ||
1560 	    to_vmx(vcpu)->exit_reason != EXIT_REASON_EPT_MISCONFIG) {
1561 		rip = kvm_rip_read(vcpu);
1562 		rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
1563 		kvm_rip_write(vcpu, rip);
1564 	} else {
1565 		if (!kvm_emulate_instruction(vcpu, EMULTYPE_SKIP))
1566 			return 0;
1567 	}
1568 
1569 	/* skipping an emulated instruction also counts */
1570 	vmx_set_interrupt_shadow(vcpu, 0);
1571 
1572 	return 1;
1573 }
1574 
1575 static void vmx_clear_hlt(struct kvm_vcpu *vcpu)
1576 {
1577 	/*
1578 	 * Ensure that we clear the HLT state in the VMCS.  We don't need to
1579 	 * explicitly skip the instruction because if the HLT state is set,
1580 	 * then the instruction is already executing and RIP has already been
1581 	 * advanced.
1582 	 */
1583 	if (kvm_hlt_in_guest(vcpu->kvm) &&
1584 			vmcs_read32(GUEST_ACTIVITY_STATE) == GUEST_ACTIVITY_HLT)
1585 		vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
1586 }
1587 
1588 static void vmx_queue_exception(struct kvm_vcpu *vcpu)
1589 {
1590 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1591 	unsigned nr = vcpu->arch.exception.nr;
1592 	bool has_error_code = vcpu->arch.exception.has_error_code;
1593 	u32 error_code = vcpu->arch.exception.error_code;
1594 	u32 intr_info = nr | INTR_INFO_VALID_MASK;
1595 
1596 	kvm_deliver_exception_payload(vcpu);
1597 
1598 	if (has_error_code) {
1599 		vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
1600 		intr_info |= INTR_INFO_DELIVER_CODE_MASK;
1601 	}
1602 
1603 	if (vmx->rmode.vm86_active) {
1604 		int inc_eip = 0;
1605 		if (kvm_exception_is_soft(nr))
1606 			inc_eip = vcpu->arch.event_exit_inst_len;
1607 		kvm_inject_realmode_interrupt(vcpu, nr, inc_eip);
1608 		return;
1609 	}
1610 
1611 	WARN_ON_ONCE(vmx->emulation_required);
1612 
1613 	if (kvm_exception_is_soft(nr)) {
1614 		vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
1615 			     vmx->vcpu.arch.event_exit_inst_len);
1616 		intr_info |= INTR_TYPE_SOFT_EXCEPTION;
1617 	} else
1618 		intr_info |= INTR_TYPE_HARD_EXCEPTION;
1619 
1620 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
1621 
1622 	vmx_clear_hlt(vcpu);
1623 }
1624 
1625 static bool vmx_rdtscp_supported(void)
1626 {
1627 	return cpu_has_vmx_rdtscp();
1628 }
1629 
1630 static bool vmx_invpcid_supported(void)
1631 {
1632 	return cpu_has_vmx_invpcid();
1633 }
1634 
1635 /*
1636  * Swap MSR entry in host/guest MSR entry array.
1637  */
1638 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
1639 {
1640 	struct shared_msr_entry tmp;
1641 
1642 	tmp = vmx->guest_msrs[to];
1643 	vmx->guest_msrs[to] = vmx->guest_msrs[from];
1644 	vmx->guest_msrs[from] = tmp;
1645 }
1646 
1647 /*
1648  * Set up the vmcs to automatically save and restore system
1649  * msrs.  Don't touch the 64-bit msrs if the guest is in legacy
1650  * mode, as fiddling with msrs is very expensive.
1651  */
1652 static void setup_msrs(struct vcpu_vmx *vmx)
1653 {
1654 	int save_nmsrs, index;
1655 
1656 	save_nmsrs = 0;
1657 #ifdef CONFIG_X86_64
1658 	/*
1659 	 * The SYSCALL MSRs are only needed on long mode guests, and only
1660 	 * when EFER.SCE is set.
1661 	 */
1662 	if (is_long_mode(&vmx->vcpu) && (vmx->vcpu.arch.efer & EFER_SCE)) {
1663 		index = __find_msr_index(vmx, MSR_STAR);
1664 		if (index >= 0)
1665 			move_msr_up(vmx, index, save_nmsrs++);
1666 		index = __find_msr_index(vmx, MSR_LSTAR);
1667 		if (index >= 0)
1668 			move_msr_up(vmx, index, save_nmsrs++);
1669 		index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
1670 		if (index >= 0)
1671 			move_msr_up(vmx, index, save_nmsrs++);
1672 	}
1673 #endif
1674 	index = __find_msr_index(vmx, MSR_EFER);
1675 	if (index >= 0 && update_transition_efer(vmx, index))
1676 		move_msr_up(vmx, index, save_nmsrs++);
1677 	index = __find_msr_index(vmx, MSR_TSC_AUX);
1678 	if (index >= 0 && guest_cpuid_has(&vmx->vcpu, X86_FEATURE_RDTSCP))
1679 		move_msr_up(vmx, index, save_nmsrs++);
1680 
1681 	vmx->save_nmsrs = save_nmsrs;
1682 	vmx->guest_msrs_ready = false;
1683 
1684 	if (cpu_has_vmx_msr_bitmap())
1685 		vmx_update_msr_bitmap(&vmx->vcpu);
1686 }
1687 
1688 static u64 vmx_read_l1_tsc_offset(struct kvm_vcpu *vcpu)
1689 {
1690 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1691 
1692 	if (is_guest_mode(vcpu) &&
1693 	    (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING))
1694 		return vcpu->arch.tsc_offset - vmcs12->tsc_offset;
1695 
1696 	return vcpu->arch.tsc_offset;
1697 }
1698 
1699 static u64 vmx_write_l1_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1700 {
1701 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1702 	u64 g_tsc_offset = 0;
1703 
1704 	/*
1705 	 * We're here if L1 chose not to trap WRMSR to TSC. According
1706 	 * to the spec, this should set L1's TSC; The offset that L1
1707 	 * set for L2 remains unchanged, and still needs to be added
1708 	 * to the newly set TSC to get L2's TSC.
1709 	 */
1710 	if (is_guest_mode(vcpu) &&
1711 	    (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING))
1712 		g_tsc_offset = vmcs12->tsc_offset;
1713 
1714 	trace_kvm_write_tsc_offset(vcpu->vcpu_id,
1715 				   vcpu->arch.tsc_offset - g_tsc_offset,
1716 				   offset);
1717 	vmcs_write64(TSC_OFFSET, offset + g_tsc_offset);
1718 	return offset + g_tsc_offset;
1719 }
1720 
1721 /*
1722  * nested_vmx_allowed() checks whether a guest should be allowed to use VMX
1723  * instructions and MSRs (i.e., nested VMX). Nested VMX is disabled for
1724  * all guests if the "nested" module option is off, and can also be disabled
1725  * for a single guest by disabling its VMX cpuid bit.
1726  */
1727 bool nested_vmx_allowed(struct kvm_vcpu *vcpu)
1728 {
1729 	return nested && guest_cpuid_has(vcpu, X86_FEATURE_VMX);
1730 }
1731 
1732 static inline bool vmx_feature_control_msr_valid(struct kvm_vcpu *vcpu,
1733 						 uint64_t val)
1734 {
1735 	uint64_t valid_bits = to_vmx(vcpu)->msr_ia32_feature_control_valid_bits;
1736 
1737 	return !(val & ~valid_bits);
1738 }
1739 
1740 static int vmx_get_msr_feature(struct kvm_msr_entry *msr)
1741 {
1742 	switch (msr->index) {
1743 	case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
1744 		if (!nested)
1745 			return 1;
1746 		return vmx_get_vmx_msr(&vmcs_config.nested, msr->index, &msr->data);
1747 	default:
1748 		return 1;
1749 	}
1750 
1751 	return 0;
1752 }
1753 
1754 /*
1755  * Reads an msr value (of 'msr_index') into 'pdata'.
1756  * Returns 0 on success, non-0 otherwise.
1757  * Assumes vcpu_load() was already called.
1758  */
1759 static int vmx_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1760 {
1761 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1762 	struct shared_msr_entry *msr;
1763 	u32 index;
1764 
1765 	switch (msr_info->index) {
1766 #ifdef CONFIG_X86_64
1767 	case MSR_FS_BASE:
1768 		msr_info->data = vmcs_readl(GUEST_FS_BASE);
1769 		break;
1770 	case MSR_GS_BASE:
1771 		msr_info->data = vmcs_readl(GUEST_GS_BASE);
1772 		break;
1773 	case MSR_KERNEL_GS_BASE:
1774 		msr_info->data = vmx_read_guest_kernel_gs_base(vmx);
1775 		break;
1776 #endif
1777 	case MSR_EFER:
1778 		return kvm_get_msr_common(vcpu, msr_info);
1779 	case MSR_IA32_UMWAIT_CONTROL:
1780 		if (!msr_info->host_initiated && !vmx_has_waitpkg(vmx))
1781 			return 1;
1782 
1783 		msr_info->data = vmx->msr_ia32_umwait_control;
1784 		break;
1785 	case MSR_IA32_SPEC_CTRL:
1786 		if (!msr_info->host_initiated &&
1787 		    !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
1788 			return 1;
1789 
1790 		msr_info->data = to_vmx(vcpu)->spec_ctrl;
1791 		break;
1792 	case MSR_IA32_SYSENTER_CS:
1793 		msr_info->data = vmcs_read32(GUEST_SYSENTER_CS);
1794 		break;
1795 	case MSR_IA32_SYSENTER_EIP:
1796 		msr_info->data = vmcs_readl(GUEST_SYSENTER_EIP);
1797 		break;
1798 	case MSR_IA32_SYSENTER_ESP:
1799 		msr_info->data = vmcs_readl(GUEST_SYSENTER_ESP);
1800 		break;
1801 	case MSR_IA32_BNDCFGS:
1802 		if (!kvm_mpx_supported() ||
1803 		    (!msr_info->host_initiated &&
1804 		     !guest_cpuid_has(vcpu, X86_FEATURE_MPX)))
1805 			return 1;
1806 		msr_info->data = vmcs_read64(GUEST_BNDCFGS);
1807 		break;
1808 	case MSR_IA32_MCG_EXT_CTL:
1809 		if (!msr_info->host_initiated &&
1810 		    !(vmx->msr_ia32_feature_control &
1811 		      FEATURE_CONTROL_LMCE))
1812 			return 1;
1813 		msr_info->data = vcpu->arch.mcg_ext_ctl;
1814 		break;
1815 	case MSR_IA32_FEATURE_CONTROL:
1816 		msr_info->data = vmx->msr_ia32_feature_control;
1817 		break;
1818 	case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
1819 		if (!nested_vmx_allowed(vcpu))
1820 			return 1;
1821 		return vmx_get_vmx_msr(&vmx->nested.msrs, msr_info->index,
1822 				       &msr_info->data);
1823 	case MSR_IA32_XSS:
1824 		if (!vmx_xsaves_supported() ||
1825 		    (!msr_info->host_initiated &&
1826 		     !(guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
1827 		       guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))))
1828 			return 1;
1829 		msr_info->data = vcpu->arch.ia32_xss;
1830 		break;
1831 	case MSR_IA32_RTIT_CTL:
1832 		if (pt_mode != PT_MODE_HOST_GUEST)
1833 			return 1;
1834 		msr_info->data = vmx->pt_desc.guest.ctl;
1835 		break;
1836 	case MSR_IA32_RTIT_STATUS:
1837 		if (pt_mode != PT_MODE_HOST_GUEST)
1838 			return 1;
1839 		msr_info->data = vmx->pt_desc.guest.status;
1840 		break;
1841 	case MSR_IA32_RTIT_CR3_MATCH:
1842 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
1843 			!intel_pt_validate_cap(vmx->pt_desc.caps,
1844 						PT_CAP_cr3_filtering))
1845 			return 1;
1846 		msr_info->data = vmx->pt_desc.guest.cr3_match;
1847 		break;
1848 	case MSR_IA32_RTIT_OUTPUT_BASE:
1849 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
1850 			(!intel_pt_validate_cap(vmx->pt_desc.caps,
1851 					PT_CAP_topa_output) &&
1852 			 !intel_pt_validate_cap(vmx->pt_desc.caps,
1853 					PT_CAP_single_range_output)))
1854 			return 1;
1855 		msr_info->data = vmx->pt_desc.guest.output_base;
1856 		break;
1857 	case MSR_IA32_RTIT_OUTPUT_MASK:
1858 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
1859 			(!intel_pt_validate_cap(vmx->pt_desc.caps,
1860 					PT_CAP_topa_output) &&
1861 			 !intel_pt_validate_cap(vmx->pt_desc.caps,
1862 					PT_CAP_single_range_output)))
1863 			return 1;
1864 		msr_info->data = vmx->pt_desc.guest.output_mask;
1865 		break;
1866 	case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
1867 		index = msr_info->index - MSR_IA32_RTIT_ADDR0_A;
1868 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
1869 			(index >= 2 * intel_pt_validate_cap(vmx->pt_desc.caps,
1870 					PT_CAP_num_address_ranges)))
1871 			return 1;
1872 		if (index % 2)
1873 			msr_info->data = vmx->pt_desc.guest.addr_b[index / 2];
1874 		else
1875 			msr_info->data = vmx->pt_desc.guest.addr_a[index / 2];
1876 		break;
1877 	case MSR_TSC_AUX:
1878 		if (!msr_info->host_initiated &&
1879 		    !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP))
1880 			return 1;
1881 		/* Else, falls through */
1882 	default:
1883 		msr = find_msr_entry(vmx, msr_info->index);
1884 		if (msr) {
1885 			msr_info->data = msr->data;
1886 			break;
1887 		}
1888 		return kvm_get_msr_common(vcpu, msr_info);
1889 	}
1890 
1891 	return 0;
1892 }
1893 
1894 /*
1895  * Writes msr value into into the appropriate "register".
1896  * Returns 0 on success, non-0 otherwise.
1897  * Assumes vcpu_load() was already called.
1898  */
1899 static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1900 {
1901 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1902 	struct shared_msr_entry *msr;
1903 	int ret = 0;
1904 	u32 msr_index = msr_info->index;
1905 	u64 data = msr_info->data;
1906 	u32 index;
1907 
1908 	switch (msr_index) {
1909 	case MSR_EFER:
1910 		ret = kvm_set_msr_common(vcpu, msr_info);
1911 		break;
1912 #ifdef CONFIG_X86_64
1913 	case MSR_FS_BASE:
1914 		vmx_segment_cache_clear(vmx);
1915 		vmcs_writel(GUEST_FS_BASE, data);
1916 		break;
1917 	case MSR_GS_BASE:
1918 		vmx_segment_cache_clear(vmx);
1919 		vmcs_writel(GUEST_GS_BASE, data);
1920 		break;
1921 	case MSR_KERNEL_GS_BASE:
1922 		vmx_write_guest_kernel_gs_base(vmx, data);
1923 		break;
1924 #endif
1925 	case MSR_IA32_SYSENTER_CS:
1926 		if (is_guest_mode(vcpu))
1927 			get_vmcs12(vcpu)->guest_sysenter_cs = data;
1928 		vmcs_write32(GUEST_SYSENTER_CS, data);
1929 		break;
1930 	case MSR_IA32_SYSENTER_EIP:
1931 		if (is_guest_mode(vcpu))
1932 			get_vmcs12(vcpu)->guest_sysenter_eip = data;
1933 		vmcs_writel(GUEST_SYSENTER_EIP, data);
1934 		break;
1935 	case MSR_IA32_SYSENTER_ESP:
1936 		if (is_guest_mode(vcpu))
1937 			get_vmcs12(vcpu)->guest_sysenter_esp = data;
1938 		vmcs_writel(GUEST_SYSENTER_ESP, data);
1939 		break;
1940 	case MSR_IA32_DEBUGCTLMSR:
1941 		if (is_guest_mode(vcpu) && get_vmcs12(vcpu)->vm_exit_controls &
1942 						VM_EXIT_SAVE_DEBUG_CONTROLS)
1943 			get_vmcs12(vcpu)->guest_ia32_debugctl = data;
1944 
1945 		ret = kvm_set_msr_common(vcpu, msr_info);
1946 		break;
1947 
1948 	case MSR_IA32_BNDCFGS:
1949 		if (!kvm_mpx_supported() ||
1950 		    (!msr_info->host_initiated &&
1951 		     !guest_cpuid_has(vcpu, X86_FEATURE_MPX)))
1952 			return 1;
1953 		if (is_noncanonical_address(data & PAGE_MASK, vcpu) ||
1954 		    (data & MSR_IA32_BNDCFGS_RSVD))
1955 			return 1;
1956 		vmcs_write64(GUEST_BNDCFGS, data);
1957 		break;
1958 	case MSR_IA32_UMWAIT_CONTROL:
1959 		if (!msr_info->host_initiated && !vmx_has_waitpkg(vmx))
1960 			return 1;
1961 
1962 		/* The reserved bit 1 and non-32 bit [63:32] should be zero */
1963 		if (data & (BIT_ULL(1) | GENMASK_ULL(63, 32)))
1964 			return 1;
1965 
1966 		vmx->msr_ia32_umwait_control = data;
1967 		break;
1968 	case MSR_IA32_SPEC_CTRL:
1969 		if (!msr_info->host_initiated &&
1970 		    !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
1971 			return 1;
1972 
1973 		/* The STIBP bit doesn't fault even if it's not advertised */
1974 		if (data & ~(SPEC_CTRL_IBRS | SPEC_CTRL_STIBP | SPEC_CTRL_SSBD))
1975 			return 1;
1976 
1977 		vmx->spec_ctrl = data;
1978 
1979 		if (!data)
1980 			break;
1981 
1982 		/*
1983 		 * For non-nested:
1984 		 * When it's written (to non-zero) for the first time, pass
1985 		 * it through.
1986 		 *
1987 		 * For nested:
1988 		 * The handling of the MSR bitmap for L2 guests is done in
1989 		 * nested_vmx_merge_msr_bitmap. We should not touch the
1990 		 * vmcs02.msr_bitmap here since it gets completely overwritten
1991 		 * in the merging. We update the vmcs01 here for L1 as well
1992 		 * since it will end up touching the MSR anyway now.
1993 		 */
1994 		vmx_disable_intercept_for_msr(vmx->vmcs01.msr_bitmap,
1995 					      MSR_IA32_SPEC_CTRL,
1996 					      MSR_TYPE_RW);
1997 		break;
1998 	case MSR_IA32_PRED_CMD:
1999 		if (!msr_info->host_initiated &&
2000 		    !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
2001 			return 1;
2002 
2003 		if (data & ~PRED_CMD_IBPB)
2004 			return 1;
2005 
2006 		if (!data)
2007 			break;
2008 
2009 		wrmsrl(MSR_IA32_PRED_CMD, PRED_CMD_IBPB);
2010 
2011 		/*
2012 		 * For non-nested:
2013 		 * When it's written (to non-zero) for the first time, pass
2014 		 * it through.
2015 		 *
2016 		 * For nested:
2017 		 * The handling of the MSR bitmap for L2 guests is done in
2018 		 * nested_vmx_merge_msr_bitmap. We should not touch the
2019 		 * vmcs02.msr_bitmap here since it gets completely overwritten
2020 		 * in the merging.
2021 		 */
2022 		vmx_disable_intercept_for_msr(vmx->vmcs01.msr_bitmap, MSR_IA32_PRED_CMD,
2023 					      MSR_TYPE_W);
2024 		break;
2025 	case MSR_IA32_CR_PAT:
2026 		if (!kvm_pat_valid(data))
2027 			return 1;
2028 
2029 		if (is_guest_mode(vcpu) &&
2030 		    get_vmcs12(vcpu)->vm_exit_controls & VM_EXIT_SAVE_IA32_PAT)
2031 			get_vmcs12(vcpu)->guest_ia32_pat = data;
2032 
2033 		if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2034 			vmcs_write64(GUEST_IA32_PAT, data);
2035 			vcpu->arch.pat = data;
2036 			break;
2037 		}
2038 		ret = kvm_set_msr_common(vcpu, msr_info);
2039 		break;
2040 	case MSR_IA32_TSC_ADJUST:
2041 		ret = kvm_set_msr_common(vcpu, msr_info);
2042 		break;
2043 	case MSR_IA32_MCG_EXT_CTL:
2044 		if ((!msr_info->host_initiated &&
2045 		     !(to_vmx(vcpu)->msr_ia32_feature_control &
2046 		       FEATURE_CONTROL_LMCE)) ||
2047 		    (data & ~MCG_EXT_CTL_LMCE_EN))
2048 			return 1;
2049 		vcpu->arch.mcg_ext_ctl = data;
2050 		break;
2051 	case MSR_IA32_FEATURE_CONTROL:
2052 		if (!vmx_feature_control_msr_valid(vcpu, data) ||
2053 		    (to_vmx(vcpu)->msr_ia32_feature_control &
2054 		     FEATURE_CONTROL_LOCKED && !msr_info->host_initiated))
2055 			return 1;
2056 		vmx->msr_ia32_feature_control = data;
2057 		if (msr_info->host_initiated && data == 0)
2058 			vmx_leave_nested(vcpu);
2059 		break;
2060 	case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
2061 		if (!msr_info->host_initiated)
2062 			return 1; /* they are read-only */
2063 		if (!nested_vmx_allowed(vcpu))
2064 			return 1;
2065 		return vmx_set_vmx_msr(vcpu, msr_index, data);
2066 	case MSR_IA32_XSS:
2067 		if (!vmx_xsaves_supported() ||
2068 		    (!msr_info->host_initiated &&
2069 		     !(guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
2070 		       guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))))
2071 			return 1;
2072 		/*
2073 		 * The only supported bit as of Skylake is bit 8, but
2074 		 * it is not supported on KVM.
2075 		 */
2076 		if (data != 0)
2077 			return 1;
2078 		vcpu->arch.ia32_xss = data;
2079 		if (vcpu->arch.ia32_xss != host_xss)
2080 			add_atomic_switch_msr(vmx, MSR_IA32_XSS,
2081 				vcpu->arch.ia32_xss, host_xss, false);
2082 		else
2083 			clear_atomic_switch_msr(vmx, MSR_IA32_XSS);
2084 		break;
2085 	case MSR_IA32_RTIT_CTL:
2086 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
2087 			vmx_rtit_ctl_check(vcpu, data) ||
2088 			vmx->nested.vmxon)
2089 			return 1;
2090 		vmcs_write64(GUEST_IA32_RTIT_CTL, data);
2091 		vmx->pt_desc.guest.ctl = data;
2092 		pt_update_intercept_for_msr(vmx);
2093 		break;
2094 	case MSR_IA32_RTIT_STATUS:
2095 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
2096 			(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) ||
2097 			(data & MSR_IA32_RTIT_STATUS_MASK))
2098 			return 1;
2099 		vmx->pt_desc.guest.status = data;
2100 		break;
2101 	case MSR_IA32_RTIT_CR3_MATCH:
2102 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
2103 			(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) ||
2104 			!intel_pt_validate_cap(vmx->pt_desc.caps,
2105 						PT_CAP_cr3_filtering))
2106 			return 1;
2107 		vmx->pt_desc.guest.cr3_match = data;
2108 		break;
2109 	case MSR_IA32_RTIT_OUTPUT_BASE:
2110 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
2111 			(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) ||
2112 			(!intel_pt_validate_cap(vmx->pt_desc.caps,
2113 					PT_CAP_topa_output) &&
2114 			 !intel_pt_validate_cap(vmx->pt_desc.caps,
2115 					PT_CAP_single_range_output)) ||
2116 			(data & MSR_IA32_RTIT_OUTPUT_BASE_MASK))
2117 			return 1;
2118 		vmx->pt_desc.guest.output_base = data;
2119 		break;
2120 	case MSR_IA32_RTIT_OUTPUT_MASK:
2121 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
2122 			(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) ||
2123 			(!intel_pt_validate_cap(vmx->pt_desc.caps,
2124 					PT_CAP_topa_output) &&
2125 			 !intel_pt_validate_cap(vmx->pt_desc.caps,
2126 					PT_CAP_single_range_output)))
2127 			return 1;
2128 		vmx->pt_desc.guest.output_mask = data;
2129 		break;
2130 	case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
2131 		index = msr_info->index - MSR_IA32_RTIT_ADDR0_A;
2132 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
2133 			(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) ||
2134 			(index >= 2 * intel_pt_validate_cap(vmx->pt_desc.caps,
2135 					PT_CAP_num_address_ranges)))
2136 			return 1;
2137 		if (index % 2)
2138 			vmx->pt_desc.guest.addr_b[index / 2] = data;
2139 		else
2140 			vmx->pt_desc.guest.addr_a[index / 2] = data;
2141 		break;
2142 	case MSR_TSC_AUX:
2143 		if (!msr_info->host_initiated &&
2144 		    !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP))
2145 			return 1;
2146 		/* Check reserved bit, higher 32 bits should be zero */
2147 		if ((data >> 32) != 0)
2148 			return 1;
2149 		/* Else, falls through */
2150 	default:
2151 		msr = find_msr_entry(vmx, msr_index);
2152 		if (msr) {
2153 			u64 old_msr_data = msr->data;
2154 			msr->data = data;
2155 			if (msr - vmx->guest_msrs < vmx->save_nmsrs) {
2156 				preempt_disable();
2157 				ret = kvm_set_shared_msr(msr->index, msr->data,
2158 							 msr->mask);
2159 				preempt_enable();
2160 				if (ret)
2161 					msr->data = old_msr_data;
2162 			}
2163 			break;
2164 		}
2165 		ret = kvm_set_msr_common(vcpu, msr_info);
2166 	}
2167 
2168 	return ret;
2169 }
2170 
2171 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
2172 {
2173 	__set_bit(reg, (unsigned long *)&vcpu->arch.regs_avail);
2174 	switch (reg) {
2175 	case VCPU_REGS_RSP:
2176 		vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
2177 		break;
2178 	case VCPU_REGS_RIP:
2179 		vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
2180 		break;
2181 	case VCPU_EXREG_PDPTR:
2182 		if (enable_ept)
2183 			ept_save_pdptrs(vcpu);
2184 		break;
2185 	default:
2186 		break;
2187 	}
2188 }
2189 
2190 static __init int cpu_has_kvm_support(void)
2191 {
2192 	return cpu_has_vmx();
2193 }
2194 
2195 static __init int vmx_disabled_by_bios(void)
2196 {
2197 	u64 msr;
2198 
2199 	rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
2200 	if (msr & FEATURE_CONTROL_LOCKED) {
2201 		/* launched w/ TXT and VMX disabled */
2202 		if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
2203 			&& tboot_enabled())
2204 			return 1;
2205 		/* launched w/o TXT and VMX only enabled w/ TXT */
2206 		if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
2207 			&& (msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
2208 			&& !tboot_enabled()) {
2209 			printk(KERN_WARNING "kvm: disable TXT in the BIOS or "
2210 				"activate TXT before enabling KVM\n");
2211 			return 1;
2212 		}
2213 		/* launched w/o TXT and VMX disabled */
2214 		if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
2215 			&& !tboot_enabled())
2216 			return 1;
2217 	}
2218 
2219 	return 0;
2220 }
2221 
2222 static void kvm_cpu_vmxon(u64 addr)
2223 {
2224 	cr4_set_bits(X86_CR4_VMXE);
2225 	intel_pt_handle_vmx(1);
2226 
2227 	asm volatile ("vmxon %0" : : "m"(addr));
2228 }
2229 
2230 static int hardware_enable(void)
2231 {
2232 	int cpu = raw_smp_processor_id();
2233 	u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
2234 	u64 old, test_bits;
2235 
2236 	if (cr4_read_shadow() & X86_CR4_VMXE)
2237 		return -EBUSY;
2238 
2239 	/*
2240 	 * This can happen if we hot-added a CPU but failed to allocate
2241 	 * VP assist page for it.
2242 	 */
2243 	if (static_branch_unlikely(&enable_evmcs) &&
2244 	    !hv_get_vp_assist_page(cpu))
2245 		return -EFAULT;
2246 
2247 	INIT_LIST_HEAD(&per_cpu(loaded_vmcss_on_cpu, cpu));
2248 	INIT_LIST_HEAD(&per_cpu(blocked_vcpu_on_cpu, cpu));
2249 	spin_lock_init(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
2250 
2251 	/*
2252 	 * Now we can enable the vmclear operation in kdump
2253 	 * since the loaded_vmcss_on_cpu list on this cpu
2254 	 * has been initialized.
2255 	 *
2256 	 * Though the cpu is not in VMX operation now, there
2257 	 * is no problem to enable the vmclear operation
2258 	 * for the loaded_vmcss_on_cpu list is empty!
2259 	 */
2260 	crash_enable_local_vmclear(cpu);
2261 
2262 	rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
2263 
2264 	test_bits = FEATURE_CONTROL_LOCKED;
2265 	test_bits |= FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
2266 	if (tboot_enabled())
2267 		test_bits |= FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX;
2268 
2269 	if ((old & test_bits) != test_bits) {
2270 		/* enable and lock */
2271 		wrmsrl(MSR_IA32_FEATURE_CONTROL, old | test_bits);
2272 	}
2273 	kvm_cpu_vmxon(phys_addr);
2274 	if (enable_ept)
2275 		ept_sync_global();
2276 
2277 	return 0;
2278 }
2279 
2280 static void vmclear_local_loaded_vmcss(void)
2281 {
2282 	int cpu = raw_smp_processor_id();
2283 	struct loaded_vmcs *v, *n;
2284 
2285 	list_for_each_entry_safe(v, n, &per_cpu(loaded_vmcss_on_cpu, cpu),
2286 				 loaded_vmcss_on_cpu_link)
2287 		__loaded_vmcs_clear(v);
2288 }
2289 
2290 
2291 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
2292  * tricks.
2293  */
2294 static void kvm_cpu_vmxoff(void)
2295 {
2296 	asm volatile (__ex("vmxoff"));
2297 
2298 	intel_pt_handle_vmx(0);
2299 	cr4_clear_bits(X86_CR4_VMXE);
2300 }
2301 
2302 static void hardware_disable(void)
2303 {
2304 	vmclear_local_loaded_vmcss();
2305 	kvm_cpu_vmxoff();
2306 }
2307 
2308 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
2309 				      u32 msr, u32 *result)
2310 {
2311 	u32 vmx_msr_low, vmx_msr_high;
2312 	u32 ctl = ctl_min | ctl_opt;
2313 
2314 	rdmsr(msr, vmx_msr_low, vmx_msr_high);
2315 
2316 	ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
2317 	ctl |= vmx_msr_low;  /* bit == 1 in low word  ==> must be one  */
2318 
2319 	/* Ensure minimum (required) set of control bits are supported. */
2320 	if (ctl_min & ~ctl)
2321 		return -EIO;
2322 
2323 	*result = ctl;
2324 	return 0;
2325 }
2326 
2327 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf,
2328 				    struct vmx_capability *vmx_cap)
2329 {
2330 	u32 vmx_msr_low, vmx_msr_high;
2331 	u32 min, opt, min2, opt2;
2332 	u32 _pin_based_exec_control = 0;
2333 	u32 _cpu_based_exec_control = 0;
2334 	u32 _cpu_based_2nd_exec_control = 0;
2335 	u32 _vmexit_control = 0;
2336 	u32 _vmentry_control = 0;
2337 
2338 	memset(vmcs_conf, 0, sizeof(*vmcs_conf));
2339 	min = CPU_BASED_HLT_EXITING |
2340 #ifdef CONFIG_X86_64
2341 	      CPU_BASED_CR8_LOAD_EXITING |
2342 	      CPU_BASED_CR8_STORE_EXITING |
2343 #endif
2344 	      CPU_BASED_CR3_LOAD_EXITING |
2345 	      CPU_BASED_CR3_STORE_EXITING |
2346 	      CPU_BASED_UNCOND_IO_EXITING |
2347 	      CPU_BASED_MOV_DR_EXITING |
2348 	      CPU_BASED_USE_TSC_OFFSETING |
2349 	      CPU_BASED_MWAIT_EXITING |
2350 	      CPU_BASED_MONITOR_EXITING |
2351 	      CPU_BASED_INVLPG_EXITING |
2352 	      CPU_BASED_RDPMC_EXITING;
2353 
2354 	opt = CPU_BASED_TPR_SHADOW |
2355 	      CPU_BASED_USE_MSR_BITMAPS |
2356 	      CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
2357 	if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
2358 				&_cpu_based_exec_control) < 0)
2359 		return -EIO;
2360 #ifdef CONFIG_X86_64
2361 	if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2362 		_cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
2363 					   ~CPU_BASED_CR8_STORE_EXITING;
2364 #endif
2365 	if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
2366 		min2 = 0;
2367 		opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2368 			SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2369 			SECONDARY_EXEC_WBINVD_EXITING |
2370 			SECONDARY_EXEC_ENABLE_VPID |
2371 			SECONDARY_EXEC_ENABLE_EPT |
2372 			SECONDARY_EXEC_UNRESTRICTED_GUEST |
2373 			SECONDARY_EXEC_PAUSE_LOOP_EXITING |
2374 			SECONDARY_EXEC_DESC |
2375 			SECONDARY_EXEC_RDTSCP |
2376 			SECONDARY_EXEC_ENABLE_INVPCID |
2377 			SECONDARY_EXEC_APIC_REGISTER_VIRT |
2378 			SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
2379 			SECONDARY_EXEC_SHADOW_VMCS |
2380 			SECONDARY_EXEC_XSAVES |
2381 			SECONDARY_EXEC_RDSEED_EXITING |
2382 			SECONDARY_EXEC_RDRAND_EXITING |
2383 			SECONDARY_EXEC_ENABLE_PML |
2384 			SECONDARY_EXEC_TSC_SCALING |
2385 			SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE |
2386 			SECONDARY_EXEC_PT_USE_GPA |
2387 			SECONDARY_EXEC_PT_CONCEAL_VMX |
2388 			SECONDARY_EXEC_ENABLE_VMFUNC |
2389 			SECONDARY_EXEC_ENCLS_EXITING;
2390 		if (adjust_vmx_controls(min2, opt2,
2391 					MSR_IA32_VMX_PROCBASED_CTLS2,
2392 					&_cpu_based_2nd_exec_control) < 0)
2393 			return -EIO;
2394 	}
2395 #ifndef CONFIG_X86_64
2396 	if (!(_cpu_based_2nd_exec_control &
2397 				SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
2398 		_cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
2399 #endif
2400 
2401 	if (!(_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2402 		_cpu_based_2nd_exec_control &= ~(
2403 				SECONDARY_EXEC_APIC_REGISTER_VIRT |
2404 				SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2405 				SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
2406 
2407 	rdmsr_safe(MSR_IA32_VMX_EPT_VPID_CAP,
2408 		&vmx_cap->ept, &vmx_cap->vpid);
2409 
2410 	if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
2411 		/* CR3 accesses and invlpg don't need to cause VM Exits when EPT
2412 		   enabled */
2413 		_cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
2414 					     CPU_BASED_CR3_STORE_EXITING |
2415 					     CPU_BASED_INVLPG_EXITING);
2416 	} else if (vmx_cap->ept) {
2417 		vmx_cap->ept = 0;
2418 		pr_warn_once("EPT CAP should not exist if not support "
2419 				"1-setting enable EPT VM-execution control\n");
2420 	}
2421 	if (!(_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_VPID) &&
2422 		vmx_cap->vpid) {
2423 		vmx_cap->vpid = 0;
2424 		pr_warn_once("VPID CAP should not exist if not support "
2425 				"1-setting enable VPID VM-execution control\n");
2426 	}
2427 
2428 	min = VM_EXIT_SAVE_DEBUG_CONTROLS | VM_EXIT_ACK_INTR_ON_EXIT;
2429 #ifdef CONFIG_X86_64
2430 	min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
2431 #endif
2432 	opt = VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL |
2433 	      VM_EXIT_LOAD_IA32_PAT |
2434 	      VM_EXIT_LOAD_IA32_EFER |
2435 	      VM_EXIT_CLEAR_BNDCFGS |
2436 	      VM_EXIT_PT_CONCEAL_PIP |
2437 	      VM_EXIT_CLEAR_IA32_RTIT_CTL;
2438 	if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
2439 				&_vmexit_control) < 0)
2440 		return -EIO;
2441 
2442 	min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
2443 	opt = PIN_BASED_VIRTUAL_NMIS | PIN_BASED_POSTED_INTR |
2444 		 PIN_BASED_VMX_PREEMPTION_TIMER;
2445 	if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
2446 				&_pin_based_exec_control) < 0)
2447 		return -EIO;
2448 
2449 	if (cpu_has_broken_vmx_preemption_timer())
2450 		_pin_based_exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
2451 	if (!(_cpu_based_2nd_exec_control &
2452 		SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY))
2453 		_pin_based_exec_control &= ~PIN_BASED_POSTED_INTR;
2454 
2455 	min = VM_ENTRY_LOAD_DEBUG_CONTROLS;
2456 	opt = VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL |
2457 	      VM_ENTRY_LOAD_IA32_PAT |
2458 	      VM_ENTRY_LOAD_IA32_EFER |
2459 	      VM_ENTRY_LOAD_BNDCFGS |
2460 	      VM_ENTRY_PT_CONCEAL_PIP |
2461 	      VM_ENTRY_LOAD_IA32_RTIT_CTL;
2462 	if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
2463 				&_vmentry_control) < 0)
2464 		return -EIO;
2465 
2466 	/*
2467 	 * Some cpus support VM_{ENTRY,EXIT}_IA32_PERF_GLOBAL_CTRL but they
2468 	 * can't be used due to an errata where VM Exit may incorrectly clear
2469 	 * IA32_PERF_GLOBAL_CTRL[34:32].  Workaround the errata by using the
2470 	 * MSR load mechanism to switch IA32_PERF_GLOBAL_CTRL.
2471 	 */
2472 	if (boot_cpu_data.x86 == 0x6) {
2473 		switch (boot_cpu_data.x86_model) {
2474 		case 26: /* AAK155 */
2475 		case 30: /* AAP115 */
2476 		case 37: /* AAT100 */
2477 		case 44: /* BC86,AAY89,BD102 */
2478 		case 46: /* BA97 */
2479 			_vmentry_control &= ~VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL;
2480 			_vmexit_control &= ~VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL;
2481 			pr_warn_once("kvm: VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL "
2482 					"does not work properly. Using workaround\n");
2483 			break;
2484 		default:
2485 			break;
2486 		}
2487 	}
2488 
2489 
2490 	rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
2491 
2492 	/* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
2493 	if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
2494 		return -EIO;
2495 
2496 #ifdef CONFIG_X86_64
2497 	/* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
2498 	if (vmx_msr_high & (1u<<16))
2499 		return -EIO;
2500 #endif
2501 
2502 	/* Require Write-Back (WB) memory type for VMCS accesses. */
2503 	if (((vmx_msr_high >> 18) & 15) != 6)
2504 		return -EIO;
2505 
2506 	vmcs_conf->size = vmx_msr_high & 0x1fff;
2507 	vmcs_conf->order = get_order(vmcs_conf->size);
2508 	vmcs_conf->basic_cap = vmx_msr_high & ~0x1fff;
2509 
2510 	vmcs_conf->revision_id = vmx_msr_low;
2511 
2512 	vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
2513 	vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
2514 	vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
2515 	vmcs_conf->vmexit_ctrl         = _vmexit_control;
2516 	vmcs_conf->vmentry_ctrl        = _vmentry_control;
2517 
2518 	if (static_branch_unlikely(&enable_evmcs))
2519 		evmcs_sanitize_exec_ctrls(vmcs_conf);
2520 
2521 	return 0;
2522 }
2523 
2524 struct vmcs *alloc_vmcs_cpu(bool shadow, int cpu, gfp_t flags)
2525 {
2526 	int node = cpu_to_node(cpu);
2527 	struct page *pages;
2528 	struct vmcs *vmcs;
2529 
2530 	pages = __alloc_pages_node(node, flags, vmcs_config.order);
2531 	if (!pages)
2532 		return NULL;
2533 	vmcs = page_address(pages);
2534 	memset(vmcs, 0, vmcs_config.size);
2535 
2536 	/* KVM supports Enlightened VMCS v1 only */
2537 	if (static_branch_unlikely(&enable_evmcs))
2538 		vmcs->hdr.revision_id = KVM_EVMCS_VERSION;
2539 	else
2540 		vmcs->hdr.revision_id = vmcs_config.revision_id;
2541 
2542 	if (shadow)
2543 		vmcs->hdr.shadow_vmcs = 1;
2544 	return vmcs;
2545 }
2546 
2547 void free_vmcs(struct vmcs *vmcs)
2548 {
2549 	free_pages((unsigned long)vmcs, vmcs_config.order);
2550 }
2551 
2552 /*
2553  * Free a VMCS, but before that VMCLEAR it on the CPU where it was last loaded
2554  */
2555 void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
2556 {
2557 	if (!loaded_vmcs->vmcs)
2558 		return;
2559 	loaded_vmcs_clear(loaded_vmcs);
2560 	free_vmcs(loaded_vmcs->vmcs);
2561 	loaded_vmcs->vmcs = NULL;
2562 	if (loaded_vmcs->msr_bitmap)
2563 		free_page((unsigned long)loaded_vmcs->msr_bitmap);
2564 	WARN_ON(loaded_vmcs->shadow_vmcs != NULL);
2565 }
2566 
2567 int alloc_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
2568 {
2569 	loaded_vmcs->vmcs = alloc_vmcs(false);
2570 	if (!loaded_vmcs->vmcs)
2571 		return -ENOMEM;
2572 
2573 	loaded_vmcs->shadow_vmcs = NULL;
2574 	loaded_vmcs->hv_timer_soft_disabled = false;
2575 	loaded_vmcs_init(loaded_vmcs);
2576 
2577 	if (cpu_has_vmx_msr_bitmap()) {
2578 		loaded_vmcs->msr_bitmap = (unsigned long *)
2579 				__get_free_page(GFP_KERNEL_ACCOUNT);
2580 		if (!loaded_vmcs->msr_bitmap)
2581 			goto out_vmcs;
2582 		memset(loaded_vmcs->msr_bitmap, 0xff, PAGE_SIZE);
2583 
2584 		if (IS_ENABLED(CONFIG_HYPERV) &&
2585 		    static_branch_unlikely(&enable_evmcs) &&
2586 		    (ms_hyperv.nested_features & HV_X64_NESTED_MSR_BITMAP)) {
2587 			struct hv_enlightened_vmcs *evmcs =
2588 				(struct hv_enlightened_vmcs *)loaded_vmcs->vmcs;
2589 
2590 			evmcs->hv_enlightenments_control.msr_bitmap = 1;
2591 		}
2592 	}
2593 
2594 	memset(&loaded_vmcs->host_state, 0, sizeof(struct vmcs_host_state));
2595 	memset(&loaded_vmcs->controls_shadow, 0,
2596 		sizeof(struct vmcs_controls_shadow));
2597 
2598 	return 0;
2599 
2600 out_vmcs:
2601 	free_loaded_vmcs(loaded_vmcs);
2602 	return -ENOMEM;
2603 }
2604 
2605 static void free_kvm_area(void)
2606 {
2607 	int cpu;
2608 
2609 	for_each_possible_cpu(cpu) {
2610 		free_vmcs(per_cpu(vmxarea, cpu));
2611 		per_cpu(vmxarea, cpu) = NULL;
2612 	}
2613 }
2614 
2615 static __init int alloc_kvm_area(void)
2616 {
2617 	int cpu;
2618 
2619 	for_each_possible_cpu(cpu) {
2620 		struct vmcs *vmcs;
2621 
2622 		vmcs = alloc_vmcs_cpu(false, cpu, GFP_KERNEL);
2623 		if (!vmcs) {
2624 			free_kvm_area();
2625 			return -ENOMEM;
2626 		}
2627 
2628 		/*
2629 		 * When eVMCS is enabled, alloc_vmcs_cpu() sets
2630 		 * vmcs->revision_id to KVM_EVMCS_VERSION instead of
2631 		 * revision_id reported by MSR_IA32_VMX_BASIC.
2632 		 *
2633 		 * However, even though not explicitly documented by
2634 		 * TLFS, VMXArea passed as VMXON argument should
2635 		 * still be marked with revision_id reported by
2636 		 * physical CPU.
2637 		 */
2638 		if (static_branch_unlikely(&enable_evmcs))
2639 			vmcs->hdr.revision_id = vmcs_config.revision_id;
2640 
2641 		per_cpu(vmxarea, cpu) = vmcs;
2642 	}
2643 	return 0;
2644 }
2645 
2646 static void fix_pmode_seg(struct kvm_vcpu *vcpu, int seg,
2647 		struct kvm_segment *save)
2648 {
2649 	if (!emulate_invalid_guest_state) {
2650 		/*
2651 		 * CS and SS RPL should be equal during guest entry according
2652 		 * to VMX spec, but in reality it is not always so. Since vcpu
2653 		 * is in the middle of the transition from real mode to
2654 		 * protected mode it is safe to assume that RPL 0 is a good
2655 		 * default value.
2656 		 */
2657 		if (seg == VCPU_SREG_CS || seg == VCPU_SREG_SS)
2658 			save->selector &= ~SEGMENT_RPL_MASK;
2659 		save->dpl = save->selector & SEGMENT_RPL_MASK;
2660 		save->s = 1;
2661 	}
2662 	vmx_set_segment(vcpu, save, seg);
2663 }
2664 
2665 static void enter_pmode(struct kvm_vcpu *vcpu)
2666 {
2667 	unsigned long flags;
2668 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2669 
2670 	/*
2671 	 * Update real mode segment cache. It may be not up-to-date if sement
2672 	 * register was written while vcpu was in a guest mode.
2673 	 */
2674 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
2675 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
2676 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
2677 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
2678 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
2679 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
2680 
2681 	vmx->rmode.vm86_active = 0;
2682 
2683 	vmx_segment_cache_clear(vmx);
2684 
2685 	vmx_set_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
2686 
2687 	flags = vmcs_readl(GUEST_RFLAGS);
2688 	flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
2689 	flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
2690 	vmcs_writel(GUEST_RFLAGS, flags);
2691 
2692 	vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
2693 			(vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
2694 
2695 	update_exception_bitmap(vcpu);
2696 
2697 	fix_pmode_seg(vcpu, VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
2698 	fix_pmode_seg(vcpu, VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
2699 	fix_pmode_seg(vcpu, VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
2700 	fix_pmode_seg(vcpu, VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
2701 	fix_pmode_seg(vcpu, VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
2702 	fix_pmode_seg(vcpu, VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
2703 }
2704 
2705 static void fix_rmode_seg(int seg, struct kvm_segment *save)
2706 {
2707 	const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2708 	struct kvm_segment var = *save;
2709 
2710 	var.dpl = 0x3;
2711 	if (seg == VCPU_SREG_CS)
2712 		var.type = 0x3;
2713 
2714 	if (!emulate_invalid_guest_state) {
2715 		var.selector = var.base >> 4;
2716 		var.base = var.base & 0xffff0;
2717 		var.limit = 0xffff;
2718 		var.g = 0;
2719 		var.db = 0;
2720 		var.present = 1;
2721 		var.s = 1;
2722 		var.l = 0;
2723 		var.unusable = 0;
2724 		var.type = 0x3;
2725 		var.avl = 0;
2726 		if (save->base & 0xf)
2727 			printk_once(KERN_WARNING "kvm: segment base is not "
2728 					"paragraph aligned when entering "
2729 					"protected mode (seg=%d)", seg);
2730 	}
2731 
2732 	vmcs_write16(sf->selector, var.selector);
2733 	vmcs_writel(sf->base, var.base);
2734 	vmcs_write32(sf->limit, var.limit);
2735 	vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(&var));
2736 }
2737 
2738 static void enter_rmode(struct kvm_vcpu *vcpu)
2739 {
2740 	unsigned long flags;
2741 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2742 	struct kvm_vmx *kvm_vmx = to_kvm_vmx(vcpu->kvm);
2743 
2744 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
2745 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
2746 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
2747 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
2748 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
2749 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
2750 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
2751 
2752 	vmx->rmode.vm86_active = 1;
2753 
2754 	/*
2755 	 * Very old userspace does not call KVM_SET_TSS_ADDR before entering
2756 	 * vcpu. Warn the user that an update is overdue.
2757 	 */
2758 	if (!kvm_vmx->tss_addr)
2759 		printk_once(KERN_WARNING "kvm: KVM_SET_TSS_ADDR need to be "
2760 			     "called before entering vcpu\n");
2761 
2762 	vmx_segment_cache_clear(vmx);
2763 
2764 	vmcs_writel(GUEST_TR_BASE, kvm_vmx->tss_addr);
2765 	vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
2766 	vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
2767 
2768 	flags = vmcs_readl(GUEST_RFLAGS);
2769 	vmx->rmode.save_rflags = flags;
2770 
2771 	flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
2772 
2773 	vmcs_writel(GUEST_RFLAGS, flags);
2774 	vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
2775 	update_exception_bitmap(vcpu);
2776 
2777 	fix_rmode_seg(VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
2778 	fix_rmode_seg(VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
2779 	fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
2780 	fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
2781 	fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
2782 	fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
2783 
2784 	kvm_mmu_reset_context(vcpu);
2785 }
2786 
2787 void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
2788 {
2789 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2790 	struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
2791 
2792 	if (!msr)
2793 		return;
2794 
2795 	vcpu->arch.efer = efer;
2796 	if (efer & EFER_LMA) {
2797 		vm_entry_controls_setbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
2798 		msr->data = efer;
2799 	} else {
2800 		vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
2801 
2802 		msr->data = efer & ~EFER_LME;
2803 	}
2804 	setup_msrs(vmx);
2805 }
2806 
2807 #ifdef CONFIG_X86_64
2808 
2809 static void enter_lmode(struct kvm_vcpu *vcpu)
2810 {
2811 	u32 guest_tr_ar;
2812 
2813 	vmx_segment_cache_clear(to_vmx(vcpu));
2814 
2815 	guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
2816 	if ((guest_tr_ar & VMX_AR_TYPE_MASK) != VMX_AR_TYPE_BUSY_64_TSS) {
2817 		pr_debug_ratelimited("%s: tss fixup for long mode. \n",
2818 				     __func__);
2819 		vmcs_write32(GUEST_TR_AR_BYTES,
2820 			     (guest_tr_ar & ~VMX_AR_TYPE_MASK)
2821 			     | VMX_AR_TYPE_BUSY_64_TSS);
2822 	}
2823 	vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
2824 }
2825 
2826 static void exit_lmode(struct kvm_vcpu *vcpu)
2827 {
2828 	vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
2829 	vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
2830 }
2831 
2832 #endif
2833 
2834 static void vmx_flush_tlb_gva(struct kvm_vcpu *vcpu, gva_t addr)
2835 {
2836 	int vpid = to_vmx(vcpu)->vpid;
2837 
2838 	if (!vpid_sync_vcpu_addr(vpid, addr))
2839 		vpid_sync_context(vpid);
2840 
2841 	/*
2842 	 * If VPIDs are not supported or enabled, then the above is a no-op.
2843 	 * But we don't really need a TLB flush in that case anyway, because
2844 	 * each VM entry/exit includes an implicit flush when VPID is 0.
2845 	 */
2846 }
2847 
2848 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
2849 {
2850 	ulong cr0_guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
2851 
2852 	vcpu->arch.cr0 &= ~cr0_guest_owned_bits;
2853 	vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & cr0_guest_owned_bits;
2854 }
2855 
2856 static void vmx_decache_cr3(struct kvm_vcpu *vcpu)
2857 {
2858 	if (enable_unrestricted_guest || (enable_ept && is_paging(vcpu)))
2859 		vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
2860 	__set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
2861 }
2862 
2863 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
2864 {
2865 	ulong cr4_guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
2866 
2867 	vcpu->arch.cr4 &= ~cr4_guest_owned_bits;
2868 	vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & cr4_guest_owned_bits;
2869 }
2870 
2871 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
2872 {
2873 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
2874 
2875 	if (!test_bit(VCPU_EXREG_PDPTR,
2876 		      (unsigned long *)&vcpu->arch.regs_dirty))
2877 		return;
2878 
2879 	if (is_pae_paging(vcpu)) {
2880 		vmcs_write64(GUEST_PDPTR0, mmu->pdptrs[0]);
2881 		vmcs_write64(GUEST_PDPTR1, mmu->pdptrs[1]);
2882 		vmcs_write64(GUEST_PDPTR2, mmu->pdptrs[2]);
2883 		vmcs_write64(GUEST_PDPTR3, mmu->pdptrs[3]);
2884 	}
2885 }
2886 
2887 void ept_save_pdptrs(struct kvm_vcpu *vcpu)
2888 {
2889 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
2890 
2891 	if (is_pae_paging(vcpu)) {
2892 		mmu->pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
2893 		mmu->pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
2894 		mmu->pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
2895 		mmu->pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
2896 	}
2897 
2898 	__set_bit(VCPU_EXREG_PDPTR,
2899 		  (unsigned long *)&vcpu->arch.regs_avail);
2900 	__set_bit(VCPU_EXREG_PDPTR,
2901 		  (unsigned long *)&vcpu->arch.regs_dirty);
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 (!test_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail))
2911 		vmx_decache_cr3(vcpu);
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 (cpu_has_vmx_ept_5levels() && (cpuid_maxphyaddr(vcpu) > 48))
2971 		return 5;
2972 	return 4;
2973 }
2974 
2975 u64 construct_eptp(struct kvm_vcpu *vcpu, unsigned long root_hpa)
2976 {
2977 	u64 eptp = VMX_EPTP_MT_WB;
2978 
2979 	eptp |= (get_ept_level(vcpu) == 5) ? VMX_EPTP_PWL_5 : VMX_EPTP_PWL_4;
2980 
2981 	if (enable_ept_ad_bits &&
2982 	    (!is_guest_mode(vcpu) || nested_ept_ad_enabled(vcpu)))
2983 		eptp |= VMX_EPTP_AD_ENABLE_BIT;
2984 	eptp |= (root_hpa & PAGE_MASK);
2985 
2986 	return eptp;
2987 }
2988 
2989 void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
2990 {
2991 	struct kvm *kvm = vcpu->kvm;
2992 	unsigned long guest_cr3;
2993 	u64 eptp;
2994 
2995 	guest_cr3 = cr3;
2996 	if (enable_ept) {
2997 		eptp = construct_eptp(vcpu, cr3);
2998 		vmcs_write64(EPT_POINTER, eptp);
2999 
3000 		if (kvm_x86_ops->tlb_remote_flush) {
3001 			spin_lock(&to_kvm_vmx(kvm)->ept_pointer_lock);
3002 			to_vmx(vcpu)->ept_pointer = eptp;
3003 			to_kvm_vmx(kvm)->ept_pointers_match
3004 				= EPT_POINTERS_CHECK;
3005 			spin_unlock(&to_kvm_vmx(kvm)->ept_pointer_lock);
3006 		}
3007 
3008 		if (enable_unrestricted_guest || is_paging(vcpu) ||
3009 		    is_guest_mode(vcpu))
3010 			guest_cr3 = kvm_read_cr3(vcpu);
3011 		else
3012 			guest_cr3 = to_kvm_vmx(kvm)->ept_identity_map_addr;
3013 		ept_load_pdptrs(vcpu);
3014 	}
3015 
3016 	vmcs_writel(GUEST_CR3, guest_cr3);
3017 }
3018 
3019 int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
3020 {
3021 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3022 	/*
3023 	 * Pass through host's Machine Check Enable value to hw_cr4, which
3024 	 * is in force while we are in guest mode.  Do not let guests control
3025 	 * this bit, even if host CR4.MCE == 0.
3026 	 */
3027 	unsigned long hw_cr4;
3028 
3029 	hw_cr4 = (cr4_read_shadow() & X86_CR4_MCE) | (cr4 & ~X86_CR4_MCE);
3030 	if (enable_unrestricted_guest)
3031 		hw_cr4 |= KVM_VM_CR4_ALWAYS_ON_UNRESTRICTED_GUEST;
3032 	else if (vmx->rmode.vm86_active)
3033 		hw_cr4 |= KVM_RMODE_VM_CR4_ALWAYS_ON;
3034 	else
3035 		hw_cr4 |= KVM_PMODE_VM_CR4_ALWAYS_ON;
3036 
3037 	if (!boot_cpu_has(X86_FEATURE_UMIP) && vmx_umip_emulated()) {
3038 		if (cr4 & X86_CR4_UMIP) {
3039 			secondary_exec_controls_setbit(vmx, SECONDARY_EXEC_DESC);
3040 			hw_cr4 &= ~X86_CR4_UMIP;
3041 		} else if (!is_guest_mode(vcpu) ||
3042 			!nested_cpu_has2(get_vmcs12(vcpu), SECONDARY_EXEC_DESC)) {
3043 			secondary_exec_controls_clearbit(vmx, SECONDARY_EXEC_DESC);
3044 		}
3045 	}
3046 
3047 	if (cr4 & X86_CR4_VMXE) {
3048 		/*
3049 		 * To use VMXON (and later other VMX instructions), a guest
3050 		 * must first be able to turn on cr4.VMXE (see handle_vmon()).
3051 		 * So basically the check on whether to allow nested VMX
3052 		 * is here.  We operate under the default treatment of SMM,
3053 		 * so VMX cannot be enabled under SMM.
3054 		 */
3055 		if (!nested_vmx_allowed(vcpu) || is_smm(vcpu))
3056 			return 1;
3057 	}
3058 
3059 	if (vmx->nested.vmxon && !nested_cr4_valid(vcpu, cr4))
3060 		return 1;
3061 
3062 	vcpu->arch.cr4 = cr4;
3063 
3064 	if (!enable_unrestricted_guest) {
3065 		if (enable_ept) {
3066 			if (!is_paging(vcpu)) {
3067 				hw_cr4 &= ~X86_CR4_PAE;
3068 				hw_cr4 |= X86_CR4_PSE;
3069 			} else if (!(cr4 & X86_CR4_PAE)) {
3070 				hw_cr4 &= ~X86_CR4_PAE;
3071 			}
3072 		}
3073 
3074 		/*
3075 		 * SMEP/SMAP/PKU is disabled if CPU is in non-paging mode in
3076 		 * hardware.  To emulate this behavior, SMEP/SMAP/PKU needs
3077 		 * to be manually disabled when guest switches to non-paging
3078 		 * mode.
3079 		 *
3080 		 * If !enable_unrestricted_guest, the CPU is always running
3081 		 * with CR0.PG=1 and CR4 needs to be modified.
3082 		 * If enable_unrestricted_guest, the CPU automatically
3083 		 * disables SMEP/SMAP/PKU when the guest sets CR0.PG=0.
3084 		 */
3085 		if (!is_paging(vcpu))
3086 			hw_cr4 &= ~(X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE);
3087 	}
3088 
3089 	vmcs_writel(CR4_READ_SHADOW, cr4);
3090 	vmcs_writel(GUEST_CR4, hw_cr4);
3091 	return 0;
3092 }
3093 
3094 void vmx_get_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg)
3095 {
3096 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3097 	u32 ar;
3098 
3099 	if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3100 		*var = vmx->rmode.segs[seg];
3101 		if (seg == VCPU_SREG_TR
3102 		    || var->selector == vmx_read_guest_seg_selector(vmx, seg))
3103 			return;
3104 		var->base = vmx_read_guest_seg_base(vmx, seg);
3105 		var->selector = vmx_read_guest_seg_selector(vmx, seg);
3106 		return;
3107 	}
3108 	var->base = vmx_read_guest_seg_base(vmx, seg);
3109 	var->limit = vmx_read_guest_seg_limit(vmx, seg);
3110 	var->selector = vmx_read_guest_seg_selector(vmx, seg);
3111 	ar = vmx_read_guest_seg_ar(vmx, seg);
3112 	var->unusable = (ar >> 16) & 1;
3113 	var->type = ar & 15;
3114 	var->s = (ar >> 4) & 1;
3115 	var->dpl = (ar >> 5) & 3;
3116 	/*
3117 	 * Some userspaces do not preserve unusable property. Since usable
3118 	 * segment has to be present according to VMX spec we can use present
3119 	 * property to amend userspace bug by making unusable segment always
3120 	 * nonpresent. vmx_segment_access_rights() already marks nonpresent
3121 	 * segment as unusable.
3122 	 */
3123 	var->present = !var->unusable;
3124 	var->avl = (ar >> 12) & 1;
3125 	var->l = (ar >> 13) & 1;
3126 	var->db = (ar >> 14) & 1;
3127 	var->g = (ar >> 15) & 1;
3128 }
3129 
3130 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
3131 {
3132 	struct kvm_segment s;
3133 
3134 	if (to_vmx(vcpu)->rmode.vm86_active) {
3135 		vmx_get_segment(vcpu, &s, seg);
3136 		return s.base;
3137 	}
3138 	return vmx_read_guest_seg_base(to_vmx(vcpu), seg);
3139 }
3140 
3141 int vmx_get_cpl(struct kvm_vcpu *vcpu)
3142 {
3143 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3144 
3145 	if (unlikely(vmx->rmode.vm86_active))
3146 		return 0;
3147 	else {
3148 		int ar = vmx_read_guest_seg_ar(vmx, VCPU_SREG_SS);
3149 		return VMX_AR_DPL(ar);
3150 	}
3151 }
3152 
3153 static u32 vmx_segment_access_rights(struct kvm_segment *var)
3154 {
3155 	u32 ar;
3156 
3157 	if (var->unusable || !var->present)
3158 		ar = 1 << 16;
3159 	else {
3160 		ar = var->type & 15;
3161 		ar |= (var->s & 1) << 4;
3162 		ar |= (var->dpl & 3) << 5;
3163 		ar |= (var->present & 1) << 7;
3164 		ar |= (var->avl & 1) << 12;
3165 		ar |= (var->l & 1) << 13;
3166 		ar |= (var->db & 1) << 14;
3167 		ar |= (var->g & 1) << 15;
3168 	}
3169 
3170 	return ar;
3171 }
3172 
3173 void vmx_set_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg)
3174 {
3175 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3176 	const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3177 
3178 	vmx_segment_cache_clear(vmx);
3179 
3180 	if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3181 		vmx->rmode.segs[seg] = *var;
3182 		if (seg == VCPU_SREG_TR)
3183 			vmcs_write16(sf->selector, var->selector);
3184 		else if (var->s)
3185 			fix_rmode_seg(seg, &vmx->rmode.segs[seg]);
3186 		goto out;
3187 	}
3188 
3189 	vmcs_writel(sf->base, var->base);
3190 	vmcs_write32(sf->limit, var->limit);
3191 	vmcs_write16(sf->selector, var->selector);
3192 
3193 	/*
3194 	 *   Fix the "Accessed" bit in AR field of segment registers for older
3195 	 * qemu binaries.
3196 	 *   IA32 arch specifies that at the time of processor reset the
3197 	 * "Accessed" bit in the AR field of segment registers is 1. And qemu
3198 	 * is setting it to 0 in the userland code. This causes invalid guest
3199 	 * state vmexit when "unrestricted guest" mode is turned on.
3200 	 *    Fix for this setup issue in cpu_reset is being pushed in the qemu
3201 	 * tree. Newer qemu binaries with that qemu fix would not need this
3202 	 * kvm hack.
3203 	 */
3204 	if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
3205 		var->type |= 0x1; /* Accessed */
3206 
3207 	vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(var));
3208 
3209 out:
3210 	vmx->emulation_required = emulation_required(vcpu);
3211 }
3212 
3213 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3214 {
3215 	u32 ar = vmx_read_guest_seg_ar(to_vmx(vcpu), VCPU_SREG_CS);
3216 
3217 	*db = (ar >> 14) & 1;
3218 	*l = (ar >> 13) & 1;
3219 }
3220 
3221 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3222 {
3223 	dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
3224 	dt->address = vmcs_readl(GUEST_IDTR_BASE);
3225 }
3226 
3227 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3228 {
3229 	vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
3230 	vmcs_writel(GUEST_IDTR_BASE, dt->address);
3231 }
3232 
3233 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3234 {
3235 	dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
3236 	dt->address = vmcs_readl(GUEST_GDTR_BASE);
3237 }
3238 
3239 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3240 {
3241 	vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
3242 	vmcs_writel(GUEST_GDTR_BASE, dt->address);
3243 }
3244 
3245 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
3246 {
3247 	struct kvm_segment var;
3248 	u32 ar;
3249 
3250 	vmx_get_segment(vcpu, &var, seg);
3251 	var.dpl = 0x3;
3252 	if (seg == VCPU_SREG_CS)
3253 		var.type = 0x3;
3254 	ar = vmx_segment_access_rights(&var);
3255 
3256 	if (var.base != (var.selector << 4))
3257 		return false;
3258 	if (var.limit != 0xffff)
3259 		return false;
3260 	if (ar != 0xf3)
3261 		return false;
3262 
3263 	return true;
3264 }
3265 
3266 static bool code_segment_valid(struct kvm_vcpu *vcpu)
3267 {
3268 	struct kvm_segment cs;
3269 	unsigned int cs_rpl;
3270 
3271 	vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3272 	cs_rpl = cs.selector & SEGMENT_RPL_MASK;
3273 
3274 	if (cs.unusable)
3275 		return false;
3276 	if (~cs.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_ACCESSES_MASK))
3277 		return false;
3278 	if (!cs.s)
3279 		return false;
3280 	if (cs.type & VMX_AR_TYPE_WRITEABLE_MASK) {
3281 		if (cs.dpl > cs_rpl)
3282 			return false;
3283 	} else {
3284 		if (cs.dpl != cs_rpl)
3285 			return false;
3286 	}
3287 	if (!cs.present)
3288 		return false;
3289 
3290 	/* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
3291 	return true;
3292 }
3293 
3294 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
3295 {
3296 	struct kvm_segment ss;
3297 	unsigned int ss_rpl;
3298 
3299 	vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3300 	ss_rpl = ss.selector & SEGMENT_RPL_MASK;
3301 
3302 	if (ss.unusable)
3303 		return true;
3304 	if (ss.type != 3 && ss.type != 7)
3305 		return false;
3306 	if (!ss.s)
3307 		return false;
3308 	if (ss.dpl != ss_rpl) /* DPL != RPL */
3309 		return false;
3310 	if (!ss.present)
3311 		return false;
3312 
3313 	return true;
3314 }
3315 
3316 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
3317 {
3318 	struct kvm_segment var;
3319 	unsigned int rpl;
3320 
3321 	vmx_get_segment(vcpu, &var, seg);
3322 	rpl = var.selector & SEGMENT_RPL_MASK;
3323 
3324 	if (var.unusable)
3325 		return true;
3326 	if (!var.s)
3327 		return false;
3328 	if (!var.present)
3329 		return false;
3330 	if (~var.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_WRITEABLE_MASK)) {
3331 		if (var.dpl < rpl) /* DPL < RPL */
3332 			return false;
3333 	}
3334 
3335 	/* TODO: Add other members to kvm_segment_field to allow checking for other access
3336 	 * rights flags
3337 	 */
3338 	return true;
3339 }
3340 
3341 static bool tr_valid(struct kvm_vcpu *vcpu)
3342 {
3343 	struct kvm_segment tr;
3344 
3345 	vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
3346 
3347 	if (tr.unusable)
3348 		return false;
3349 	if (tr.selector & SEGMENT_TI_MASK)	/* TI = 1 */
3350 		return false;
3351 	if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
3352 		return false;
3353 	if (!tr.present)
3354 		return false;
3355 
3356 	return true;
3357 }
3358 
3359 static bool ldtr_valid(struct kvm_vcpu *vcpu)
3360 {
3361 	struct kvm_segment ldtr;
3362 
3363 	vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
3364 
3365 	if (ldtr.unusable)
3366 		return true;
3367 	if (ldtr.selector & SEGMENT_TI_MASK)	/* TI = 1 */
3368 		return false;
3369 	if (ldtr.type != 2)
3370 		return false;
3371 	if (!ldtr.present)
3372 		return false;
3373 
3374 	return true;
3375 }
3376 
3377 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
3378 {
3379 	struct kvm_segment cs, ss;
3380 
3381 	vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3382 	vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3383 
3384 	return ((cs.selector & SEGMENT_RPL_MASK) ==
3385 		 (ss.selector & SEGMENT_RPL_MASK));
3386 }
3387 
3388 /*
3389  * Check if guest state is valid. Returns true if valid, false if
3390  * not.
3391  * We assume that registers are always usable
3392  */
3393 static bool guest_state_valid(struct kvm_vcpu *vcpu)
3394 {
3395 	if (enable_unrestricted_guest)
3396 		return true;
3397 
3398 	/* real mode guest state checks */
3399 	if (!is_protmode(vcpu) || (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
3400 		if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
3401 			return false;
3402 		if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
3403 			return false;
3404 		if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
3405 			return false;
3406 		if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
3407 			return false;
3408 		if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
3409 			return false;
3410 		if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
3411 			return false;
3412 	} else {
3413 	/* protected mode guest state checks */
3414 		if (!cs_ss_rpl_check(vcpu))
3415 			return false;
3416 		if (!code_segment_valid(vcpu))
3417 			return false;
3418 		if (!stack_segment_valid(vcpu))
3419 			return false;
3420 		if (!data_segment_valid(vcpu, VCPU_SREG_DS))
3421 			return false;
3422 		if (!data_segment_valid(vcpu, VCPU_SREG_ES))
3423 			return false;
3424 		if (!data_segment_valid(vcpu, VCPU_SREG_FS))
3425 			return false;
3426 		if (!data_segment_valid(vcpu, VCPU_SREG_GS))
3427 			return false;
3428 		if (!tr_valid(vcpu))
3429 			return false;
3430 		if (!ldtr_valid(vcpu))
3431 			return false;
3432 	}
3433 	/* TODO:
3434 	 * - Add checks on RIP
3435 	 * - Add checks on RFLAGS
3436 	 */
3437 
3438 	return true;
3439 }
3440 
3441 static int init_rmode_tss(struct kvm *kvm)
3442 {
3443 	gfn_t fn;
3444 	u16 data = 0;
3445 	int idx, r;
3446 
3447 	idx = srcu_read_lock(&kvm->srcu);
3448 	fn = to_kvm_vmx(kvm)->tss_addr >> PAGE_SHIFT;
3449 	r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3450 	if (r < 0)
3451 		goto out;
3452 	data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
3453 	r = kvm_write_guest_page(kvm, fn++, &data,
3454 			TSS_IOPB_BASE_OFFSET, sizeof(u16));
3455 	if (r < 0)
3456 		goto out;
3457 	r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
3458 	if (r < 0)
3459 		goto out;
3460 	r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3461 	if (r < 0)
3462 		goto out;
3463 	data = ~0;
3464 	r = kvm_write_guest_page(kvm, fn, &data,
3465 				 RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
3466 				 sizeof(u8));
3467 out:
3468 	srcu_read_unlock(&kvm->srcu, idx);
3469 	return r;
3470 }
3471 
3472 static int init_rmode_identity_map(struct kvm *kvm)
3473 {
3474 	struct kvm_vmx *kvm_vmx = to_kvm_vmx(kvm);
3475 	int i, idx, r = 0;
3476 	kvm_pfn_t identity_map_pfn;
3477 	u32 tmp;
3478 
3479 	/* Protect kvm_vmx->ept_identity_pagetable_done. */
3480 	mutex_lock(&kvm->slots_lock);
3481 
3482 	if (likely(kvm_vmx->ept_identity_pagetable_done))
3483 		goto out2;
3484 
3485 	if (!kvm_vmx->ept_identity_map_addr)
3486 		kvm_vmx->ept_identity_map_addr = VMX_EPT_IDENTITY_PAGETABLE_ADDR;
3487 	identity_map_pfn = kvm_vmx->ept_identity_map_addr >> PAGE_SHIFT;
3488 
3489 	r = __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
3490 				    kvm_vmx->ept_identity_map_addr, PAGE_SIZE);
3491 	if (r < 0)
3492 		goto out2;
3493 
3494 	idx = srcu_read_lock(&kvm->srcu);
3495 	r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
3496 	if (r < 0)
3497 		goto out;
3498 	/* Set up identity-mapping pagetable for EPT in real mode */
3499 	for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
3500 		tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
3501 			_PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
3502 		r = kvm_write_guest_page(kvm, identity_map_pfn,
3503 				&tmp, i * sizeof(tmp), sizeof(tmp));
3504 		if (r < 0)
3505 			goto out;
3506 	}
3507 	kvm_vmx->ept_identity_pagetable_done = true;
3508 
3509 out:
3510 	srcu_read_unlock(&kvm->srcu, idx);
3511 
3512 out2:
3513 	mutex_unlock(&kvm->slots_lock);
3514 	return r;
3515 }
3516 
3517 static void seg_setup(int seg)
3518 {
3519 	const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3520 	unsigned int ar;
3521 
3522 	vmcs_write16(sf->selector, 0);
3523 	vmcs_writel(sf->base, 0);
3524 	vmcs_write32(sf->limit, 0xffff);
3525 	ar = 0x93;
3526 	if (seg == VCPU_SREG_CS)
3527 		ar |= 0x08; /* code segment */
3528 
3529 	vmcs_write32(sf->ar_bytes, ar);
3530 }
3531 
3532 static int alloc_apic_access_page(struct kvm *kvm)
3533 {
3534 	struct page *page;
3535 	int r = 0;
3536 
3537 	mutex_lock(&kvm->slots_lock);
3538 	if (kvm->arch.apic_access_page_done)
3539 		goto out;
3540 	r = __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
3541 				    APIC_DEFAULT_PHYS_BASE, PAGE_SIZE);
3542 	if (r)
3543 		goto out;
3544 
3545 	page = gfn_to_page(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
3546 	if (is_error_page(page)) {
3547 		r = -EFAULT;
3548 		goto out;
3549 	}
3550 
3551 	/*
3552 	 * Do not pin the page in memory, so that memory hot-unplug
3553 	 * is able to migrate it.
3554 	 */
3555 	put_page(page);
3556 	kvm->arch.apic_access_page_done = true;
3557 out:
3558 	mutex_unlock(&kvm->slots_lock);
3559 	return r;
3560 }
3561 
3562 int allocate_vpid(void)
3563 {
3564 	int vpid;
3565 
3566 	if (!enable_vpid)
3567 		return 0;
3568 	spin_lock(&vmx_vpid_lock);
3569 	vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
3570 	if (vpid < VMX_NR_VPIDS)
3571 		__set_bit(vpid, vmx_vpid_bitmap);
3572 	else
3573 		vpid = 0;
3574 	spin_unlock(&vmx_vpid_lock);
3575 	return vpid;
3576 }
3577 
3578 void free_vpid(int vpid)
3579 {
3580 	if (!enable_vpid || vpid == 0)
3581 		return;
3582 	spin_lock(&vmx_vpid_lock);
3583 	__clear_bit(vpid, vmx_vpid_bitmap);
3584 	spin_unlock(&vmx_vpid_lock);
3585 }
3586 
3587 static __always_inline void vmx_disable_intercept_for_msr(unsigned long *msr_bitmap,
3588 							  u32 msr, int type)
3589 {
3590 	int f = sizeof(unsigned long);
3591 
3592 	if (!cpu_has_vmx_msr_bitmap())
3593 		return;
3594 
3595 	if (static_branch_unlikely(&enable_evmcs))
3596 		evmcs_touch_msr_bitmap();
3597 
3598 	/*
3599 	 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
3600 	 * have the write-low and read-high bitmap offsets the wrong way round.
3601 	 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
3602 	 */
3603 	if (msr <= 0x1fff) {
3604 		if (type & MSR_TYPE_R)
3605 			/* read-low */
3606 			__clear_bit(msr, msr_bitmap + 0x000 / f);
3607 
3608 		if (type & MSR_TYPE_W)
3609 			/* write-low */
3610 			__clear_bit(msr, msr_bitmap + 0x800 / f);
3611 
3612 	} else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
3613 		msr &= 0x1fff;
3614 		if (type & MSR_TYPE_R)
3615 			/* read-high */
3616 			__clear_bit(msr, msr_bitmap + 0x400 / f);
3617 
3618 		if (type & MSR_TYPE_W)
3619 			/* write-high */
3620 			__clear_bit(msr, msr_bitmap + 0xc00 / f);
3621 
3622 	}
3623 }
3624 
3625 static __always_inline void vmx_enable_intercept_for_msr(unsigned long *msr_bitmap,
3626 							 u32 msr, int type)
3627 {
3628 	int f = sizeof(unsigned long);
3629 
3630 	if (!cpu_has_vmx_msr_bitmap())
3631 		return;
3632 
3633 	if (static_branch_unlikely(&enable_evmcs))
3634 		evmcs_touch_msr_bitmap();
3635 
3636 	/*
3637 	 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
3638 	 * have the write-low and read-high bitmap offsets the wrong way round.
3639 	 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
3640 	 */
3641 	if (msr <= 0x1fff) {
3642 		if (type & MSR_TYPE_R)
3643 			/* read-low */
3644 			__set_bit(msr, msr_bitmap + 0x000 / f);
3645 
3646 		if (type & MSR_TYPE_W)
3647 			/* write-low */
3648 			__set_bit(msr, msr_bitmap + 0x800 / f);
3649 
3650 	} else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
3651 		msr &= 0x1fff;
3652 		if (type & MSR_TYPE_R)
3653 			/* read-high */
3654 			__set_bit(msr, msr_bitmap + 0x400 / f);
3655 
3656 		if (type & MSR_TYPE_W)
3657 			/* write-high */
3658 			__set_bit(msr, msr_bitmap + 0xc00 / f);
3659 
3660 	}
3661 }
3662 
3663 static __always_inline void vmx_set_intercept_for_msr(unsigned long *msr_bitmap,
3664 			     			      u32 msr, int type, bool value)
3665 {
3666 	if (value)
3667 		vmx_enable_intercept_for_msr(msr_bitmap, msr, type);
3668 	else
3669 		vmx_disable_intercept_for_msr(msr_bitmap, msr, type);
3670 }
3671 
3672 static u8 vmx_msr_bitmap_mode(struct kvm_vcpu *vcpu)
3673 {
3674 	u8 mode = 0;
3675 
3676 	if (cpu_has_secondary_exec_ctrls() &&
3677 	    (secondary_exec_controls_get(to_vmx(vcpu)) &
3678 	     SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE)) {
3679 		mode |= MSR_BITMAP_MODE_X2APIC;
3680 		if (enable_apicv && kvm_vcpu_apicv_active(vcpu))
3681 			mode |= MSR_BITMAP_MODE_X2APIC_APICV;
3682 	}
3683 
3684 	return mode;
3685 }
3686 
3687 static void vmx_update_msr_bitmap_x2apic(unsigned long *msr_bitmap,
3688 					 u8 mode)
3689 {
3690 	int msr;
3691 
3692 	for (msr = 0x800; msr <= 0x8ff; msr += BITS_PER_LONG) {
3693 		unsigned word = msr / BITS_PER_LONG;
3694 		msr_bitmap[word] = (mode & MSR_BITMAP_MODE_X2APIC_APICV) ? 0 : ~0;
3695 		msr_bitmap[word + (0x800 / sizeof(long))] = ~0;
3696 	}
3697 
3698 	if (mode & MSR_BITMAP_MODE_X2APIC) {
3699 		/*
3700 		 * TPR reads and writes can be virtualized even if virtual interrupt
3701 		 * delivery is not in use.
3702 		 */
3703 		vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_TASKPRI), MSR_TYPE_RW);
3704 		if (mode & MSR_BITMAP_MODE_X2APIC_APICV) {
3705 			vmx_enable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_TMCCT), MSR_TYPE_R);
3706 			vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_EOI), MSR_TYPE_W);
3707 			vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_SELF_IPI), MSR_TYPE_W);
3708 		}
3709 	}
3710 }
3711 
3712 void vmx_update_msr_bitmap(struct kvm_vcpu *vcpu)
3713 {
3714 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3715 	unsigned long *msr_bitmap = vmx->vmcs01.msr_bitmap;
3716 	u8 mode = vmx_msr_bitmap_mode(vcpu);
3717 	u8 changed = mode ^ vmx->msr_bitmap_mode;
3718 
3719 	if (!changed)
3720 		return;
3721 
3722 	if (changed & (MSR_BITMAP_MODE_X2APIC | MSR_BITMAP_MODE_X2APIC_APICV))
3723 		vmx_update_msr_bitmap_x2apic(msr_bitmap, mode);
3724 
3725 	vmx->msr_bitmap_mode = mode;
3726 }
3727 
3728 void pt_update_intercept_for_msr(struct vcpu_vmx *vmx)
3729 {
3730 	unsigned long *msr_bitmap = vmx->vmcs01.msr_bitmap;
3731 	bool flag = !(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN);
3732 	u32 i;
3733 
3734 	vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_STATUS,
3735 							MSR_TYPE_RW, flag);
3736 	vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_OUTPUT_BASE,
3737 							MSR_TYPE_RW, flag);
3738 	vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_OUTPUT_MASK,
3739 							MSR_TYPE_RW, flag);
3740 	vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_CR3_MATCH,
3741 							MSR_TYPE_RW, flag);
3742 	for (i = 0; i < vmx->pt_desc.addr_range; i++) {
3743 		vmx_set_intercept_for_msr(msr_bitmap,
3744 			MSR_IA32_RTIT_ADDR0_A + i * 2, MSR_TYPE_RW, flag);
3745 		vmx_set_intercept_for_msr(msr_bitmap,
3746 			MSR_IA32_RTIT_ADDR0_B + i * 2, MSR_TYPE_RW, flag);
3747 	}
3748 }
3749 
3750 static bool vmx_get_enable_apicv(struct kvm_vcpu *vcpu)
3751 {
3752 	return enable_apicv;
3753 }
3754 
3755 static bool vmx_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
3756 {
3757 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3758 	void *vapic_page;
3759 	u32 vppr;
3760 	int rvi;
3761 
3762 	if (WARN_ON_ONCE(!is_guest_mode(vcpu)) ||
3763 		!nested_cpu_has_vid(get_vmcs12(vcpu)) ||
3764 		WARN_ON_ONCE(!vmx->nested.virtual_apic_map.gfn))
3765 		return false;
3766 
3767 	rvi = vmx_get_rvi();
3768 
3769 	vapic_page = vmx->nested.virtual_apic_map.hva;
3770 	vppr = *((u32 *)(vapic_page + APIC_PROCPRI));
3771 
3772 	return ((rvi & 0xf0) > (vppr & 0xf0));
3773 }
3774 
3775 static inline bool kvm_vcpu_trigger_posted_interrupt(struct kvm_vcpu *vcpu,
3776 						     bool nested)
3777 {
3778 #ifdef CONFIG_SMP
3779 	int pi_vec = nested ? POSTED_INTR_NESTED_VECTOR : POSTED_INTR_VECTOR;
3780 
3781 	if (vcpu->mode == IN_GUEST_MODE) {
3782 		/*
3783 		 * The vector of interrupt to be delivered to vcpu had
3784 		 * been set in PIR before this function.
3785 		 *
3786 		 * Following cases will be reached in this block, and
3787 		 * we always send a notification event in all cases as
3788 		 * explained below.
3789 		 *
3790 		 * Case 1: vcpu keeps in non-root mode. Sending a
3791 		 * notification event posts the interrupt to vcpu.
3792 		 *
3793 		 * Case 2: vcpu exits to root mode and is still
3794 		 * runnable. PIR will be synced to vIRR before the
3795 		 * next vcpu entry. Sending a notification event in
3796 		 * this case has no effect, as vcpu is not in root
3797 		 * mode.
3798 		 *
3799 		 * Case 3: vcpu exits to root mode and is blocked.
3800 		 * vcpu_block() has already synced PIR to vIRR and
3801 		 * never blocks vcpu if vIRR is not cleared. Therefore,
3802 		 * a blocked vcpu here does not wait for any requested
3803 		 * interrupts in PIR, and sending a notification event
3804 		 * which has no effect is safe here.
3805 		 */
3806 
3807 		apic->send_IPI_mask(get_cpu_mask(vcpu->cpu), pi_vec);
3808 		return true;
3809 	}
3810 #endif
3811 	return false;
3812 }
3813 
3814 static int vmx_deliver_nested_posted_interrupt(struct kvm_vcpu *vcpu,
3815 						int vector)
3816 {
3817 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3818 
3819 	if (is_guest_mode(vcpu) &&
3820 	    vector == vmx->nested.posted_intr_nv) {
3821 		/*
3822 		 * If a posted intr is not recognized by hardware,
3823 		 * we will accomplish it in the next vmentry.
3824 		 */
3825 		vmx->nested.pi_pending = true;
3826 		kvm_make_request(KVM_REQ_EVENT, vcpu);
3827 		/* the PIR and ON have been set by L1. */
3828 		if (!kvm_vcpu_trigger_posted_interrupt(vcpu, true))
3829 			kvm_vcpu_kick(vcpu);
3830 		return 0;
3831 	}
3832 	return -1;
3833 }
3834 /*
3835  * Send interrupt to vcpu via posted interrupt way.
3836  * 1. If target vcpu is running(non-root mode), send posted interrupt
3837  * notification to vcpu and hardware will sync PIR to vIRR atomically.
3838  * 2. If target vcpu isn't running(root mode), kick it to pick up the
3839  * interrupt from PIR in next vmentry.
3840  */
3841 static void vmx_deliver_posted_interrupt(struct kvm_vcpu *vcpu, int vector)
3842 {
3843 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3844 	int r;
3845 
3846 	r = vmx_deliver_nested_posted_interrupt(vcpu, vector);
3847 	if (!r)
3848 		return;
3849 
3850 	if (pi_test_and_set_pir(vector, &vmx->pi_desc))
3851 		return;
3852 
3853 	/* If a previous notification has sent the IPI, nothing to do.  */
3854 	if (pi_test_and_set_on(&vmx->pi_desc))
3855 		return;
3856 
3857 	if (!kvm_vcpu_trigger_posted_interrupt(vcpu, false))
3858 		kvm_vcpu_kick(vcpu);
3859 }
3860 
3861 /*
3862  * Set up the vmcs's constant host-state fields, i.e., host-state fields that
3863  * will not change in the lifetime of the guest.
3864  * Note that host-state that does change is set elsewhere. E.g., host-state
3865  * that is set differently for each CPU is set in vmx_vcpu_load(), not here.
3866  */
3867 void vmx_set_constant_host_state(struct vcpu_vmx *vmx)
3868 {
3869 	u32 low32, high32;
3870 	unsigned long tmpl;
3871 	unsigned long cr0, cr3, cr4;
3872 
3873 	cr0 = read_cr0();
3874 	WARN_ON(cr0 & X86_CR0_TS);
3875 	vmcs_writel(HOST_CR0, cr0);  /* 22.2.3 */
3876 
3877 	/*
3878 	 * Save the most likely value for this task's CR3 in the VMCS.
3879 	 * We can't use __get_current_cr3_fast() because we're not atomic.
3880 	 */
3881 	cr3 = __read_cr3();
3882 	vmcs_writel(HOST_CR3, cr3);		/* 22.2.3  FIXME: shadow tables */
3883 	vmx->loaded_vmcs->host_state.cr3 = cr3;
3884 
3885 	/* Save the most likely value for this task's CR4 in the VMCS. */
3886 	cr4 = cr4_read_shadow();
3887 	vmcs_writel(HOST_CR4, cr4);			/* 22.2.3, 22.2.5 */
3888 	vmx->loaded_vmcs->host_state.cr4 = cr4;
3889 
3890 	vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS);  /* 22.2.4 */
3891 #ifdef CONFIG_X86_64
3892 	/*
3893 	 * Load null selectors, so we can avoid reloading them in
3894 	 * vmx_prepare_switch_to_host(), in case userspace uses
3895 	 * the null selectors too (the expected case).
3896 	 */
3897 	vmcs_write16(HOST_DS_SELECTOR, 0);
3898 	vmcs_write16(HOST_ES_SELECTOR, 0);
3899 #else
3900 	vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
3901 	vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
3902 #endif
3903 	vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
3904 	vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8);  /* 22.2.4 */
3905 
3906 	vmcs_writel(HOST_IDTR_BASE, host_idt_base);   /* 22.2.4 */
3907 
3908 	vmcs_writel(HOST_RIP, (unsigned long)vmx_vmexit); /* 22.2.5 */
3909 
3910 	rdmsr(MSR_IA32_SYSENTER_CS, low32, high32);
3911 	vmcs_write32(HOST_IA32_SYSENTER_CS, low32);
3912 	rdmsrl(MSR_IA32_SYSENTER_EIP, tmpl);
3913 	vmcs_writel(HOST_IA32_SYSENTER_EIP, tmpl);   /* 22.2.3 */
3914 
3915 	if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
3916 		rdmsr(MSR_IA32_CR_PAT, low32, high32);
3917 		vmcs_write64(HOST_IA32_PAT, low32 | ((u64) high32 << 32));
3918 	}
3919 
3920 	if (cpu_has_load_ia32_efer())
3921 		vmcs_write64(HOST_IA32_EFER, host_efer);
3922 }
3923 
3924 void set_cr4_guest_host_mask(struct vcpu_vmx *vmx)
3925 {
3926 	vmx->vcpu.arch.cr4_guest_owned_bits = KVM_CR4_GUEST_OWNED_BITS;
3927 	if (enable_ept)
3928 		vmx->vcpu.arch.cr4_guest_owned_bits |= X86_CR4_PGE;
3929 	if (is_guest_mode(&vmx->vcpu))
3930 		vmx->vcpu.arch.cr4_guest_owned_bits &=
3931 			~get_vmcs12(&vmx->vcpu)->cr4_guest_host_mask;
3932 	vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits);
3933 }
3934 
3935 u32 vmx_pin_based_exec_ctrl(struct vcpu_vmx *vmx)
3936 {
3937 	u32 pin_based_exec_ctrl = vmcs_config.pin_based_exec_ctrl;
3938 
3939 	if (!kvm_vcpu_apicv_active(&vmx->vcpu))
3940 		pin_based_exec_ctrl &= ~PIN_BASED_POSTED_INTR;
3941 
3942 	if (!enable_vnmi)
3943 		pin_based_exec_ctrl &= ~PIN_BASED_VIRTUAL_NMIS;
3944 
3945 	if (!enable_preemption_timer)
3946 		pin_based_exec_ctrl &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
3947 
3948 	return pin_based_exec_ctrl;
3949 }
3950 
3951 static void vmx_refresh_apicv_exec_ctrl(struct kvm_vcpu *vcpu)
3952 {
3953 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3954 
3955 	pin_controls_set(vmx, vmx_pin_based_exec_ctrl(vmx));
3956 	if (cpu_has_secondary_exec_ctrls()) {
3957 		if (kvm_vcpu_apicv_active(vcpu))
3958 			secondary_exec_controls_setbit(vmx,
3959 				      SECONDARY_EXEC_APIC_REGISTER_VIRT |
3960 				      SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
3961 		else
3962 			secondary_exec_controls_clearbit(vmx,
3963 					SECONDARY_EXEC_APIC_REGISTER_VIRT |
3964 					SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
3965 	}
3966 
3967 	if (cpu_has_vmx_msr_bitmap())
3968 		vmx_update_msr_bitmap(vcpu);
3969 }
3970 
3971 u32 vmx_exec_control(struct vcpu_vmx *vmx)
3972 {
3973 	u32 exec_control = vmcs_config.cpu_based_exec_ctrl;
3974 
3975 	if (vmx->vcpu.arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)
3976 		exec_control &= ~CPU_BASED_MOV_DR_EXITING;
3977 
3978 	if (!cpu_need_tpr_shadow(&vmx->vcpu)) {
3979 		exec_control &= ~CPU_BASED_TPR_SHADOW;
3980 #ifdef CONFIG_X86_64
3981 		exec_control |= CPU_BASED_CR8_STORE_EXITING |
3982 				CPU_BASED_CR8_LOAD_EXITING;
3983 #endif
3984 	}
3985 	if (!enable_ept)
3986 		exec_control |= CPU_BASED_CR3_STORE_EXITING |
3987 				CPU_BASED_CR3_LOAD_EXITING  |
3988 				CPU_BASED_INVLPG_EXITING;
3989 	if (kvm_mwait_in_guest(vmx->vcpu.kvm))
3990 		exec_control &= ~(CPU_BASED_MWAIT_EXITING |
3991 				CPU_BASED_MONITOR_EXITING);
3992 	if (kvm_hlt_in_guest(vmx->vcpu.kvm))
3993 		exec_control &= ~CPU_BASED_HLT_EXITING;
3994 	return exec_control;
3995 }
3996 
3997 
3998 static void vmx_compute_secondary_exec_control(struct vcpu_vmx *vmx)
3999 {
4000 	struct kvm_vcpu *vcpu = &vmx->vcpu;
4001 
4002 	u32 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
4003 
4004 	if (pt_mode == PT_MODE_SYSTEM)
4005 		exec_control &= ~(SECONDARY_EXEC_PT_USE_GPA | SECONDARY_EXEC_PT_CONCEAL_VMX);
4006 	if (!cpu_need_virtualize_apic_accesses(vcpu))
4007 		exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
4008 	if (vmx->vpid == 0)
4009 		exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
4010 	if (!enable_ept) {
4011 		exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
4012 		enable_unrestricted_guest = 0;
4013 	}
4014 	if (!enable_unrestricted_guest)
4015 		exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
4016 	if (kvm_pause_in_guest(vmx->vcpu.kvm))
4017 		exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
4018 	if (!kvm_vcpu_apicv_active(vcpu))
4019 		exec_control &= ~(SECONDARY_EXEC_APIC_REGISTER_VIRT |
4020 				  SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
4021 	exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
4022 
4023 	/* SECONDARY_EXEC_DESC is enabled/disabled on writes to CR4.UMIP,
4024 	 * in vmx_set_cr4.  */
4025 	exec_control &= ~SECONDARY_EXEC_DESC;
4026 
4027 	/* SECONDARY_EXEC_SHADOW_VMCS is enabled when L1 executes VMPTRLD
4028 	   (handle_vmptrld).
4029 	   We can NOT enable shadow_vmcs here because we don't have yet
4030 	   a current VMCS12
4031 	*/
4032 	exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
4033 
4034 	if (!enable_pml)
4035 		exec_control &= ~SECONDARY_EXEC_ENABLE_PML;
4036 
4037 	if (vmx_xsaves_supported()) {
4038 		/* Exposing XSAVES only when XSAVE is exposed */
4039 		bool xsaves_enabled =
4040 			guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
4041 			guest_cpuid_has(vcpu, X86_FEATURE_XSAVES);
4042 
4043 		if (!xsaves_enabled)
4044 			exec_control &= ~SECONDARY_EXEC_XSAVES;
4045 
4046 		if (nested) {
4047 			if (xsaves_enabled)
4048 				vmx->nested.msrs.secondary_ctls_high |=
4049 					SECONDARY_EXEC_XSAVES;
4050 			else
4051 				vmx->nested.msrs.secondary_ctls_high &=
4052 					~SECONDARY_EXEC_XSAVES;
4053 		}
4054 	}
4055 
4056 	if (vmx_rdtscp_supported()) {
4057 		bool rdtscp_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP);
4058 		if (!rdtscp_enabled)
4059 			exec_control &= ~SECONDARY_EXEC_RDTSCP;
4060 
4061 		if (nested) {
4062 			if (rdtscp_enabled)
4063 				vmx->nested.msrs.secondary_ctls_high |=
4064 					SECONDARY_EXEC_RDTSCP;
4065 			else
4066 				vmx->nested.msrs.secondary_ctls_high &=
4067 					~SECONDARY_EXEC_RDTSCP;
4068 		}
4069 	}
4070 
4071 	if (vmx_invpcid_supported()) {
4072 		/* Exposing INVPCID only when PCID is exposed */
4073 		bool invpcid_enabled =
4074 			guest_cpuid_has(vcpu, X86_FEATURE_INVPCID) &&
4075 			guest_cpuid_has(vcpu, X86_FEATURE_PCID);
4076 
4077 		if (!invpcid_enabled) {
4078 			exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID;
4079 			guest_cpuid_clear(vcpu, X86_FEATURE_INVPCID);
4080 		}
4081 
4082 		if (nested) {
4083 			if (invpcid_enabled)
4084 				vmx->nested.msrs.secondary_ctls_high |=
4085 					SECONDARY_EXEC_ENABLE_INVPCID;
4086 			else
4087 				vmx->nested.msrs.secondary_ctls_high &=
4088 					~SECONDARY_EXEC_ENABLE_INVPCID;
4089 		}
4090 	}
4091 
4092 	if (vmx_rdrand_supported()) {
4093 		bool rdrand_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDRAND);
4094 		if (rdrand_enabled)
4095 			exec_control &= ~SECONDARY_EXEC_RDRAND_EXITING;
4096 
4097 		if (nested) {
4098 			if (rdrand_enabled)
4099 				vmx->nested.msrs.secondary_ctls_high |=
4100 					SECONDARY_EXEC_RDRAND_EXITING;
4101 			else
4102 				vmx->nested.msrs.secondary_ctls_high &=
4103 					~SECONDARY_EXEC_RDRAND_EXITING;
4104 		}
4105 	}
4106 
4107 	if (vmx_rdseed_supported()) {
4108 		bool rdseed_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDSEED);
4109 		if (rdseed_enabled)
4110 			exec_control &= ~SECONDARY_EXEC_RDSEED_EXITING;
4111 
4112 		if (nested) {
4113 			if (rdseed_enabled)
4114 				vmx->nested.msrs.secondary_ctls_high |=
4115 					SECONDARY_EXEC_RDSEED_EXITING;
4116 			else
4117 				vmx->nested.msrs.secondary_ctls_high &=
4118 					~SECONDARY_EXEC_RDSEED_EXITING;
4119 		}
4120 	}
4121 
4122 	if (vmx_waitpkg_supported()) {
4123 		bool waitpkg_enabled =
4124 			guest_cpuid_has(vcpu, X86_FEATURE_WAITPKG);
4125 
4126 		if (!waitpkg_enabled)
4127 			exec_control &= ~SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE;
4128 
4129 		if (nested) {
4130 			if (waitpkg_enabled)
4131 				vmx->nested.msrs.secondary_ctls_high |=
4132 					SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE;
4133 			else
4134 				vmx->nested.msrs.secondary_ctls_high &=
4135 					~SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE;
4136 		}
4137 	}
4138 
4139 	vmx->secondary_exec_control = exec_control;
4140 }
4141 
4142 static void ept_set_mmio_spte_mask(void)
4143 {
4144 	/*
4145 	 * EPT Misconfigurations can be generated if the value of bits 2:0
4146 	 * of an EPT paging-structure entry is 110b (write/execute).
4147 	 */
4148 	kvm_mmu_set_mmio_spte_mask(VMX_EPT_RWX_MASK,
4149 				   VMX_EPT_MISCONFIG_WX_VALUE, 0);
4150 }
4151 
4152 #define VMX_XSS_EXIT_BITMAP 0
4153 
4154 /*
4155  * Sets up the vmcs for emulated real mode.
4156  */
4157 static void vmx_vcpu_setup(struct vcpu_vmx *vmx)
4158 {
4159 	int i;
4160 
4161 	if (nested)
4162 		nested_vmx_vcpu_setup();
4163 
4164 	if (cpu_has_vmx_msr_bitmap())
4165 		vmcs_write64(MSR_BITMAP, __pa(vmx->vmcs01.msr_bitmap));
4166 
4167 	vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
4168 
4169 	/* Control */
4170 	pin_controls_set(vmx, vmx_pin_based_exec_ctrl(vmx));
4171 	vmx->hv_deadline_tsc = -1;
4172 
4173 	exec_controls_set(vmx, vmx_exec_control(vmx));
4174 
4175 	if (cpu_has_secondary_exec_ctrls()) {
4176 		vmx_compute_secondary_exec_control(vmx);
4177 		secondary_exec_controls_set(vmx, vmx->secondary_exec_control);
4178 	}
4179 
4180 	if (kvm_vcpu_apicv_active(&vmx->vcpu)) {
4181 		vmcs_write64(EOI_EXIT_BITMAP0, 0);
4182 		vmcs_write64(EOI_EXIT_BITMAP1, 0);
4183 		vmcs_write64(EOI_EXIT_BITMAP2, 0);
4184 		vmcs_write64(EOI_EXIT_BITMAP3, 0);
4185 
4186 		vmcs_write16(GUEST_INTR_STATUS, 0);
4187 
4188 		vmcs_write16(POSTED_INTR_NV, POSTED_INTR_VECTOR);
4189 		vmcs_write64(POSTED_INTR_DESC_ADDR, __pa((&vmx->pi_desc)));
4190 	}
4191 
4192 	if (!kvm_pause_in_guest(vmx->vcpu.kvm)) {
4193 		vmcs_write32(PLE_GAP, ple_gap);
4194 		vmx->ple_window = ple_window;
4195 		vmx->ple_window_dirty = true;
4196 	}
4197 
4198 	vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
4199 	vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
4200 	vmcs_write32(CR3_TARGET_COUNT, 0);           /* 22.2.1 */
4201 
4202 	vmcs_write16(HOST_FS_SELECTOR, 0);            /* 22.2.4 */
4203 	vmcs_write16(HOST_GS_SELECTOR, 0);            /* 22.2.4 */
4204 	vmx_set_constant_host_state(vmx);
4205 	vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
4206 	vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
4207 
4208 	if (cpu_has_vmx_vmfunc())
4209 		vmcs_write64(VM_FUNCTION_CONTROL, 0);
4210 
4211 	vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
4212 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
4213 	vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host.val));
4214 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
4215 	vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest.val));
4216 
4217 	if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
4218 		vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
4219 
4220 	for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i) {
4221 		u32 index = vmx_msr_index[i];
4222 		u32 data_low, data_high;
4223 		int j = vmx->nmsrs;
4224 
4225 		if (rdmsr_safe(index, &data_low, &data_high) < 0)
4226 			continue;
4227 		if (wrmsr_safe(index, data_low, data_high) < 0)
4228 			continue;
4229 		vmx->guest_msrs[j].index = i;
4230 		vmx->guest_msrs[j].data = 0;
4231 		vmx->guest_msrs[j].mask = -1ull;
4232 		++vmx->nmsrs;
4233 	}
4234 
4235 	vm_exit_controls_set(vmx, vmx_vmexit_ctrl());
4236 
4237 	/* 22.2.1, 20.8.1 */
4238 	vm_entry_controls_set(vmx, vmx_vmentry_ctrl());
4239 
4240 	vmx->vcpu.arch.cr0_guest_owned_bits = X86_CR0_TS;
4241 	vmcs_writel(CR0_GUEST_HOST_MASK, ~X86_CR0_TS);
4242 
4243 	set_cr4_guest_host_mask(vmx);
4244 
4245 	if (vmx_xsaves_supported())
4246 		vmcs_write64(XSS_EXIT_BITMAP, VMX_XSS_EXIT_BITMAP);
4247 
4248 	if (enable_pml) {
4249 		vmcs_write64(PML_ADDRESS, page_to_phys(vmx->pml_pg));
4250 		vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
4251 	}
4252 
4253 	if (cpu_has_vmx_encls_vmexit())
4254 		vmcs_write64(ENCLS_EXITING_BITMAP, -1ull);
4255 
4256 	if (pt_mode == PT_MODE_HOST_GUEST) {
4257 		memset(&vmx->pt_desc, 0, sizeof(vmx->pt_desc));
4258 		/* Bit[6~0] are forced to 1, writes are ignored. */
4259 		vmx->pt_desc.guest.output_mask = 0x7F;
4260 		vmcs_write64(GUEST_IA32_RTIT_CTL, 0);
4261 	}
4262 }
4263 
4264 static void vmx_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
4265 {
4266 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4267 	struct msr_data apic_base_msr;
4268 	u64 cr0;
4269 
4270 	vmx->rmode.vm86_active = 0;
4271 	vmx->spec_ctrl = 0;
4272 
4273 	vmx->msr_ia32_umwait_control = 0;
4274 
4275 	vcpu->arch.microcode_version = 0x100000000ULL;
4276 	vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
4277 	vmx->hv_deadline_tsc = -1;
4278 	kvm_set_cr8(vcpu, 0);
4279 
4280 	if (!init_event) {
4281 		apic_base_msr.data = APIC_DEFAULT_PHYS_BASE |
4282 				     MSR_IA32_APICBASE_ENABLE;
4283 		if (kvm_vcpu_is_reset_bsp(vcpu))
4284 			apic_base_msr.data |= MSR_IA32_APICBASE_BSP;
4285 		apic_base_msr.host_initiated = true;
4286 		kvm_set_apic_base(vcpu, &apic_base_msr);
4287 	}
4288 
4289 	vmx_segment_cache_clear(vmx);
4290 
4291 	seg_setup(VCPU_SREG_CS);
4292 	vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
4293 	vmcs_writel(GUEST_CS_BASE, 0xffff0000ul);
4294 
4295 	seg_setup(VCPU_SREG_DS);
4296 	seg_setup(VCPU_SREG_ES);
4297 	seg_setup(VCPU_SREG_FS);
4298 	seg_setup(VCPU_SREG_GS);
4299 	seg_setup(VCPU_SREG_SS);
4300 
4301 	vmcs_write16(GUEST_TR_SELECTOR, 0);
4302 	vmcs_writel(GUEST_TR_BASE, 0);
4303 	vmcs_write32(GUEST_TR_LIMIT, 0xffff);
4304 	vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
4305 
4306 	vmcs_write16(GUEST_LDTR_SELECTOR, 0);
4307 	vmcs_writel(GUEST_LDTR_BASE, 0);
4308 	vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
4309 	vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
4310 
4311 	if (!init_event) {
4312 		vmcs_write32(GUEST_SYSENTER_CS, 0);
4313 		vmcs_writel(GUEST_SYSENTER_ESP, 0);
4314 		vmcs_writel(GUEST_SYSENTER_EIP, 0);
4315 		vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
4316 	}
4317 
4318 	kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
4319 	kvm_rip_write(vcpu, 0xfff0);
4320 
4321 	vmcs_writel(GUEST_GDTR_BASE, 0);
4322 	vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
4323 
4324 	vmcs_writel(GUEST_IDTR_BASE, 0);
4325 	vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
4326 
4327 	vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
4328 	vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
4329 	vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS, 0);
4330 	if (kvm_mpx_supported())
4331 		vmcs_write64(GUEST_BNDCFGS, 0);
4332 
4333 	setup_msrs(vmx);
4334 
4335 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);  /* 22.2.1 */
4336 
4337 	if (cpu_has_vmx_tpr_shadow() && !init_event) {
4338 		vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
4339 		if (cpu_need_tpr_shadow(vcpu))
4340 			vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
4341 				     __pa(vcpu->arch.apic->regs));
4342 		vmcs_write32(TPR_THRESHOLD, 0);
4343 	}
4344 
4345 	kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
4346 
4347 	if (vmx->vpid != 0)
4348 		vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
4349 
4350 	cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
4351 	vmx->vcpu.arch.cr0 = cr0;
4352 	vmx_set_cr0(vcpu, cr0); /* enter rmode */
4353 	vmx_set_cr4(vcpu, 0);
4354 	vmx_set_efer(vcpu, 0);
4355 
4356 	update_exception_bitmap(vcpu);
4357 
4358 	vpid_sync_context(vmx->vpid);
4359 	if (init_event)
4360 		vmx_clear_hlt(vcpu);
4361 }
4362 
4363 static void enable_irq_window(struct kvm_vcpu *vcpu)
4364 {
4365 	exec_controls_setbit(to_vmx(vcpu), CPU_BASED_VIRTUAL_INTR_PENDING);
4366 }
4367 
4368 static void enable_nmi_window(struct kvm_vcpu *vcpu)
4369 {
4370 	if (!enable_vnmi ||
4371 	    vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_STI) {
4372 		enable_irq_window(vcpu);
4373 		return;
4374 	}
4375 
4376 	exec_controls_setbit(to_vmx(vcpu), CPU_BASED_VIRTUAL_NMI_PENDING);
4377 }
4378 
4379 static void vmx_inject_irq(struct kvm_vcpu *vcpu)
4380 {
4381 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4382 	uint32_t intr;
4383 	int irq = vcpu->arch.interrupt.nr;
4384 
4385 	trace_kvm_inj_virq(irq);
4386 
4387 	++vcpu->stat.irq_injections;
4388 	if (vmx->rmode.vm86_active) {
4389 		int inc_eip = 0;
4390 		if (vcpu->arch.interrupt.soft)
4391 			inc_eip = vcpu->arch.event_exit_inst_len;
4392 		kvm_inject_realmode_interrupt(vcpu, irq, inc_eip);
4393 		return;
4394 	}
4395 	intr = irq | INTR_INFO_VALID_MASK;
4396 	if (vcpu->arch.interrupt.soft) {
4397 		intr |= INTR_TYPE_SOFT_INTR;
4398 		vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
4399 			     vmx->vcpu.arch.event_exit_inst_len);
4400 	} else
4401 		intr |= INTR_TYPE_EXT_INTR;
4402 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
4403 
4404 	vmx_clear_hlt(vcpu);
4405 }
4406 
4407 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
4408 {
4409 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4410 
4411 	if (!enable_vnmi) {
4412 		/*
4413 		 * Tracking the NMI-blocked state in software is built upon
4414 		 * finding the next open IRQ window. This, in turn, depends on
4415 		 * well-behaving guests: They have to keep IRQs disabled at
4416 		 * least as long as the NMI handler runs. Otherwise we may
4417 		 * cause NMI nesting, maybe breaking the guest. But as this is
4418 		 * highly unlikely, we can live with the residual risk.
4419 		 */
4420 		vmx->loaded_vmcs->soft_vnmi_blocked = 1;
4421 		vmx->loaded_vmcs->vnmi_blocked_time = 0;
4422 	}
4423 
4424 	++vcpu->stat.nmi_injections;
4425 	vmx->loaded_vmcs->nmi_known_unmasked = false;
4426 
4427 	if (vmx->rmode.vm86_active) {
4428 		kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR, 0);
4429 		return;
4430 	}
4431 
4432 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
4433 			INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
4434 
4435 	vmx_clear_hlt(vcpu);
4436 }
4437 
4438 bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
4439 {
4440 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4441 	bool masked;
4442 
4443 	if (!enable_vnmi)
4444 		return vmx->loaded_vmcs->soft_vnmi_blocked;
4445 	if (vmx->loaded_vmcs->nmi_known_unmasked)
4446 		return false;
4447 	masked = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI;
4448 	vmx->loaded_vmcs->nmi_known_unmasked = !masked;
4449 	return masked;
4450 }
4451 
4452 void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
4453 {
4454 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4455 
4456 	if (!enable_vnmi) {
4457 		if (vmx->loaded_vmcs->soft_vnmi_blocked != masked) {
4458 			vmx->loaded_vmcs->soft_vnmi_blocked = masked;
4459 			vmx->loaded_vmcs->vnmi_blocked_time = 0;
4460 		}
4461 	} else {
4462 		vmx->loaded_vmcs->nmi_known_unmasked = !masked;
4463 		if (masked)
4464 			vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
4465 				      GUEST_INTR_STATE_NMI);
4466 		else
4467 			vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
4468 					GUEST_INTR_STATE_NMI);
4469 	}
4470 }
4471 
4472 static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
4473 {
4474 	if (to_vmx(vcpu)->nested.nested_run_pending)
4475 		return 0;
4476 
4477 	if (!enable_vnmi &&
4478 	    to_vmx(vcpu)->loaded_vmcs->soft_vnmi_blocked)
4479 		return 0;
4480 
4481 	return	!(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4482 		  (GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_STI
4483 		   | GUEST_INTR_STATE_NMI));
4484 }
4485 
4486 static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
4487 {
4488 	return (!to_vmx(vcpu)->nested.nested_run_pending &&
4489 		vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
4490 		!(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4491 			(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
4492 }
4493 
4494 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
4495 {
4496 	int ret;
4497 
4498 	if (enable_unrestricted_guest)
4499 		return 0;
4500 
4501 	ret = x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, addr,
4502 				    PAGE_SIZE * 3);
4503 	if (ret)
4504 		return ret;
4505 	to_kvm_vmx(kvm)->tss_addr = addr;
4506 	return init_rmode_tss(kvm);
4507 }
4508 
4509 static int vmx_set_identity_map_addr(struct kvm *kvm, u64 ident_addr)
4510 {
4511 	to_kvm_vmx(kvm)->ept_identity_map_addr = ident_addr;
4512 	return 0;
4513 }
4514 
4515 static bool rmode_exception(struct kvm_vcpu *vcpu, int vec)
4516 {
4517 	switch (vec) {
4518 	case BP_VECTOR:
4519 		/*
4520 		 * Update instruction length as we may reinject the exception
4521 		 * from user space while in guest debugging mode.
4522 		 */
4523 		to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
4524 			vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4525 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
4526 			return false;
4527 		/* fall through */
4528 	case DB_VECTOR:
4529 		if (vcpu->guest_debug &
4530 			(KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
4531 			return false;
4532 		/* fall through */
4533 	case DE_VECTOR:
4534 	case OF_VECTOR:
4535 	case BR_VECTOR:
4536 	case UD_VECTOR:
4537 	case DF_VECTOR:
4538 	case SS_VECTOR:
4539 	case GP_VECTOR:
4540 	case MF_VECTOR:
4541 		return true;
4542 	break;
4543 	}
4544 	return false;
4545 }
4546 
4547 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
4548 				  int vec, u32 err_code)
4549 {
4550 	/*
4551 	 * Instruction with address size override prefix opcode 0x67
4552 	 * Cause the #SS fault with 0 error code in VM86 mode.
4553 	 */
4554 	if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0) {
4555 		if (kvm_emulate_instruction(vcpu, 0)) {
4556 			if (vcpu->arch.halt_request) {
4557 				vcpu->arch.halt_request = 0;
4558 				return kvm_vcpu_halt(vcpu);
4559 			}
4560 			return 1;
4561 		}
4562 		return 0;
4563 	}
4564 
4565 	/*
4566 	 * Forward all other exceptions that are valid in real mode.
4567 	 * FIXME: Breaks guest debugging in real mode, needs to be fixed with
4568 	 *        the required debugging infrastructure rework.
4569 	 */
4570 	kvm_queue_exception(vcpu, vec);
4571 	return 1;
4572 }
4573 
4574 /*
4575  * Trigger machine check on the host. We assume all the MSRs are already set up
4576  * by the CPU and that we still run on the same CPU as the MCE occurred on.
4577  * We pass a fake environment to the machine check handler because we want
4578  * the guest to be always treated like user space, no matter what context
4579  * it used internally.
4580  */
4581 static void kvm_machine_check(void)
4582 {
4583 #if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
4584 	struct pt_regs regs = {
4585 		.cs = 3, /* Fake ring 3 no matter what the guest ran on */
4586 		.flags = X86_EFLAGS_IF,
4587 	};
4588 
4589 	do_machine_check(&regs, 0);
4590 #endif
4591 }
4592 
4593 static int handle_machine_check(struct kvm_vcpu *vcpu)
4594 {
4595 	/* handled by vmx_vcpu_run() */
4596 	return 1;
4597 }
4598 
4599 static int handle_exception_nmi(struct kvm_vcpu *vcpu)
4600 {
4601 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4602 	struct kvm_run *kvm_run = vcpu->run;
4603 	u32 intr_info, ex_no, error_code;
4604 	unsigned long cr2, rip, dr6;
4605 	u32 vect_info;
4606 
4607 	vect_info = vmx->idt_vectoring_info;
4608 	intr_info = vmx->exit_intr_info;
4609 
4610 	if (is_machine_check(intr_info) || is_nmi(intr_info))
4611 		return 1; /* handled by handle_exception_nmi_irqoff() */
4612 
4613 	if (is_invalid_opcode(intr_info))
4614 		return handle_ud(vcpu);
4615 
4616 	error_code = 0;
4617 	if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
4618 		error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
4619 
4620 	if (!vmx->rmode.vm86_active && is_gp_fault(intr_info)) {
4621 		WARN_ON_ONCE(!enable_vmware_backdoor);
4622 
4623 		/*
4624 		 * VMware backdoor emulation on #GP interception only handles
4625 		 * IN{S}, OUT{S}, and RDPMC, none of which generate a non-zero
4626 		 * error code on #GP.
4627 		 */
4628 		if (error_code) {
4629 			kvm_queue_exception_e(vcpu, GP_VECTOR, error_code);
4630 			return 1;
4631 		}
4632 		return kvm_emulate_instruction(vcpu, EMULTYPE_VMWARE_GP);
4633 	}
4634 
4635 	/*
4636 	 * The #PF with PFEC.RSVD = 1 indicates the guest is accessing
4637 	 * MMIO, it is better to report an internal error.
4638 	 * See the comments in vmx_handle_exit.
4639 	 */
4640 	if ((vect_info & VECTORING_INFO_VALID_MASK) &&
4641 	    !(is_page_fault(intr_info) && !(error_code & PFERR_RSVD_MASK))) {
4642 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4643 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
4644 		vcpu->run->internal.ndata = 3;
4645 		vcpu->run->internal.data[0] = vect_info;
4646 		vcpu->run->internal.data[1] = intr_info;
4647 		vcpu->run->internal.data[2] = error_code;
4648 		return 0;
4649 	}
4650 
4651 	if (is_page_fault(intr_info)) {
4652 		cr2 = vmcs_readl(EXIT_QUALIFICATION);
4653 		/* EPT won't cause page fault directly */
4654 		WARN_ON_ONCE(!vcpu->arch.apf.host_apf_reason && enable_ept);
4655 		return kvm_handle_page_fault(vcpu, error_code, cr2, NULL, 0);
4656 	}
4657 
4658 	ex_no = intr_info & INTR_INFO_VECTOR_MASK;
4659 
4660 	if (vmx->rmode.vm86_active && rmode_exception(vcpu, ex_no))
4661 		return handle_rmode_exception(vcpu, ex_no, error_code);
4662 
4663 	switch (ex_no) {
4664 	case AC_VECTOR:
4665 		kvm_queue_exception_e(vcpu, AC_VECTOR, error_code);
4666 		return 1;
4667 	case DB_VECTOR:
4668 		dr6 = vmcs_readl(EXIT_QUALIFICATION);
4669 		if (!(vcpu->guest_debug &
4670 		      (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
4671 			vcpu->arch.dr6 &= ~DR_TRAP_BITS;
4672 			vcpu->arch.dr6 |= dr6 | DR6_RTM;
4673 			if (is_icebp(intr_info))
4674 				WARN_ON(!skip_emulated_instruction(vcpu));
4675 
4676 			kvm_queue_exception(vcpu, DB_VECTOR);
4677 			return 1;
4678 		}
4679 		kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
4680 		kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
4681 		/* fall through */
4682 	case BP_VECTOR:
4683 		/*
4684 		 * Update instruction length as we may reinject #BP from
4685 		 * user space while in guest debugging mode. Reading it for
4686 		 * #DB as well causes no harm, it is not used in that case.
4687 		 */
4688 		vmx->vcpu.arch.event_exit_inst_len =
4689 			vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4690 		kvm_run->exit_reason = KVM_EXIT_DEBUG;
4691 		rip = kvm_rip_read(vcpu);
4692 		kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
4693 		kvm_run->debug.arch.exception = ex_no;
4694 		break;
4695 	default:
4696 		kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
4697 		kvm_run->ex.exception = ex_no;
4698 		kvm_run->ex.error_code = error_code;
4699 		break;
4700 	}
4701 	return 0;
4702 }
4703 
4704 static int handle_external_interrupt(struct kvm_vcpu *vcpu)
4705 {
4706 	++vcpu->stat.irq_exits;
4707 	return 1;
4708 }
4709 
4710 static int handle_triple_fault(struct kvm_vcpu *vcpu)
4711 {
4712 	vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
4713 	vcpu->mmio_needed = 0;
4714 	return 0;
4715 }
4716 
4717 static int handle_io(struct kvm_vcpu *vcpu)
4718 {
4719 	unsigned long exit_qualification;
4720 	int size, in, string;
4721 	unsigned port;
4722 
4723 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4724 	string = (exit_qualification & 16) != 0;
4725 
4726 	++vcpu->stat.io_exits;
4727 
4728 	if (string)
4729 		return kvm_emulate_instruction(vcpu, 0);
4730 
4731 	port = exit_qualification >> 16;
4732 	size = (exit_qualification & 7) + 1;
4733 	in = (exit_qualification & 8) != 0;
4734 
4735 	return kvm_fast_pio(vcpu, size, port, in);
4736 }
4737 
4738 static void
4739 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
4740 {
4741 	/*
4742 	 * Patch in the VMCALL instruction:
4743 	 */
4744 	hypercall[0] = 0x0f;
4745 	hypercall[1] = 0x01;
4746 	hypercall[2] = 0xc1;
4747 }
4748 
4749 /* called to set cr0 as appropriate for a mov-to-cr0 exit. */
4750 static int handle_set_cr0(struct kvm_vcpu *vcpu, unsigned long val)
4751 {
4752 	if (is_guest_mode(vcpu)) {
4753 		struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4754 		unsigned long orig_val = val;
4755 
4756 		/*
4757 		 * We get here when L2 changed cr0 in a way that did not change
4758 		 * any of L1's shadowed bits (see nested_vmx_exit_handled_cr),
4759 		 * but did change L0 shadowed bits. So we first calculate the
4760 		 * effective cr0 value that L1 would like to write into the
4761 		 * hardware. It consists of the L2-owned bits from the new
4762 		 * value combined with the L1-owned bits from L1's guest_cr0.
4763 		 */
4764 		val = (val & ~vmcs12->cr0_guest_host_mask) |
4765 			(vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask);
4766 
4767 		if (!nested_guest_cr0_valid(vcpu, val))
4768 			return 1;
4769 
4770 		if (kvm_set_cr0(vcpu, val))
4771 			return 1;
4772 		vmcs_writel(CR0_READ_SHADOW, orig_val);
4773 		return 0;
4774 	} else {
4775 		if (to_vmx(vcpu)->nested.vmxon &&
4776 		    !nested_host_cr0_valid(vcpu, val))
4777 			return 1;
4778 
4779 		return kvm_set_cr0(vcpu, val);
4780 	}
4781 }
4782 
4783 static int handle_set_cr4(struct kvm_vcpu *vcpu, unsigned long val)
4784 {
4785 	if (is_guest_mode(vcpu)) {
4786 		struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4787 		unsigned long orig_val = val;
4788 
4789 		/* analogously to handle_set_cr0 */
4790 		val = (val & ~vmcs12->cr4_guest_host_mask) |
4791 			(vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask);
4792 		if (kvm_set_cr4(vcpu, val))
4793 			return 1;
4794 		vmcs_writel(CR4_READ_SHADOW, orig_val);
4795 		return 0;
4796 	} else
4797 		return kvm_set_cr4(vcpu, val);
4798 }
4799 
4800 static int handle_desc(struct kvm_vcpu *vcpu)
4801 {
4802 	WARN_ON(!(vcpu->arch.cr4 & X86_CR4_UMIP));
4803 	return kvm_emulate_instruction(vcpu, 0);
4804 }
4805 
4806 static int handle_cr(struct kvm_vcpu *vcpu)
4807 {
4808 	unsigned long exit_qualification, val;
4809 	int cr;
4810 	int reg;
4811 	int err;
4812 	int ret;
4813 
4814 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4815 	cr = exit_qualification & 15;
4816 	reg = (exit_qualification >> 8) & 15;
4817 	switch ((exit_qualification >> 4) & 3) {
4818 	case 0: /* mov to cr */
4819 		val = kvm_register_readl(vcpu, reg);
4820 		trace_kvm_cr_write(cr, val);
4821 		switch (cr) {
4822 		case 0:
4823 			err = handle_set_cr0(vcpu, val);
4824 			return kvm_complete_insn_gp(vcpu, err);
4825 		case 3:
4826 			WARN_ON_ONCE(enable_unrestricted_guest);
4827 			err = kvm_set_cr3(vcpu, val);
4828 			return kvm_complete_insn_gp(vcpu, err);
4829 		case 4:
4830 			err = handle_set_cr4(vcpu, val);
4831 			return kvm_complete_insn_gp(vcpu, err);
4832 		case 8: {
4833 				u8 cr8_prev = kvm_get_cr8(vcpu);
4834 				u8 cr8 = (u8)val;
4835 				err = kvm_set_cr8(vcpu, cr8);
4836 				ret = kvm_complete_insn_gp(vcpu, err);
4837 				if (lapic_in_kernel(vcpu))
4838 					return ret;
4839 				if (cr8_prev <= cr8)
4840 					return ret;
4841 				/*
4842 				 * TODO: we might be squashing a
4843 				 * KVM_GUESTDBG_SINGLESTEP-triggered
4844 				 * KVM_EXIT_DEBUG here.
4845 				 */
4846 				vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
4847 				return 0;
4848 			}
4849 		}
4850 		break;
4851 	case 2: /* clts */
4852 		WARN_ONCE(1, "Guest should always own CR0.TS");
4853 		vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
4854 		trace_kvm_cr_write(0, kvm_read_cr0(vcpu));
4855 		return kvm_skip_emulated_instruction(vcpu);
4856 	case 1: /*mov from cr*/
4857 		switch (cr) {
4858 		case 3:
4859 			WARN_ON_ONCE(enable_unrestricted_guest);
4860 			val = kvm_read_cr3(vcpu);
4861 			kvm_register_write(vcpu, reg, val);
4862 			trace_kvm_cr_read(cr, val);
4863 			return kvm_skip_emulated_instruction(vcpu);
4864 		case 8:
4865 			val = kvm_get_cr8(vcpu);
4866 			kvm_register_write(vcpu, reg, val);
4867 			trace_kvm_cr_read(cr, val);
4868 			return kvm_skip_emulated_instruction(vcpu);
4869 		}
4870 		break;
4871 	case 3: /* lmsw */
4872 		val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
4873 		trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val);
4874 		kvm_lmsw(vcpu, val);
4875 
4876 		return kvm_skip_emulated_instruction(vcpu);
4877 	default:
4878 		break;
4879 	}
4880 	vcpu->run->exit_reason = 0;
4881 	vcpu_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
4882 	       (int)(exit_qualification >> 4) & 3, cr);
4883 	return 0;
4884 }
4885 
4886 static int handle_dr(struct kvm_vcpu *vcpu)
4887 {
4888 	unsigned long exit_qualification;
4889 	int dr, dr7, reg;
4890 
4891 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4892 	dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
4893 
4894 	/* First, if DR does not exist, trigger UD */
4895 	if (!kvm_require_dr(vcpu, dr))
4896 		return 1;
4897 
4898 	/* Do not handle if the CPL > 0, will trigger GP on re-entry */
4899 	if (!kvm_require_cpl(vcpu, 0))
4900 		return 1;
4901 	dr7 = vmcs_readl(GUEST_DR7);
4902 	if (dr7 & DR7_GD) {
4903 		/*
4904 		 * As the vm-exit takes precedence over the debug trap, we
4905 		 * need to emulate the latter, either for the host or the
4906 		 * guest debugging itself.
4907 		 */
4908 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
4909 			vcpu->run->debug.arch.dr6 = vcpu->arch.dr6;
4910 			vcpu->run->debug.arch.dr7 = dr7;
4911 			vcpu->run->debug.arch.pc = kvm_get_linear_rip(vcpu);
4912 			vcpu->run->debug.arch.exception = DB_VECTOR;
4913 			vcpu->run->exit_reason = KVM_EXIT_DEBUG;
4914 			return 0;
4915 		} else {
4916 			vcpu->arch.dr6 &= ~DR_TRAP_BITS;
4917 			vcpu->arch.dr6 |= DR6_BD | DR6_RTM;
4918 			kvm_queue_exception(vcpu, DB_VECTOR);
4919 			return 1;
4920 		}
4921 	}
4922 
4923 	if (vcpu->guest_debug == 0) {
4924 		exec_controls_clearbit(to_vmx(vcpu), CPU_BASED_MOV_DR_EXITING);
4925 
4926 		/*
4927 		 * No more DR vmexits; force a reload of the debug registers
4928 		 * and reenter on this instruction.  The next vmexit will
4929 		 * retrieve the full state of the debug registers.
4930 		 */
4931 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
4932 		return 1;
4933 	}
4934 
4935 	reg = DEBUG_REG_ACCESS_REG(exit_qualification);
4936 	if (exit_qualification & TYPE_MOV_FROM_DR) {
4937 		unsigned long val;
4938 
4939 		if (kvm_get_dr(vcpu, dr, &val))
4940 			return 1;
4941 		kvm_register_write(vcpu, reg, val);
4942 	} else
4943 		if (kvm_set_dr(vcpu, dr, kvm_register_readl(vcpu, reg)))
4944 			return 1;
4945 
4946 	return kvm_skip_emulated_instruction(vcpu);
4947 }
4948 
4949 static u64 vmx_get_dr6(struct kvm_vcpu *vcpu)
4950 {
4951 	return vcpu->arch.dr6;
4952 }
4953 
4954 static void vmx_set_dr6(struct kvm_vcpu *vcpu, unsigned long val)
4955 {
4956 }
4957 
4958 static void vmx_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
4959 {
4960 	get_debugreg(vcpu->arch.db[0], 0);
4961 	get_debugreg(vcpu->arch.db[1], 1);
4962 	get_debugreg(vcpu->arch.db[2], 2);
4963 	get_debugreg(vcpu->arch.db[3], 3);
4964 	get_debugreg(vcpu->arch.dr6, 6);
4965 	vcpu->arch.dr7 = vmcs_readl(GUEST_DR7);
4966 
4967 	vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
4968 	exec_controls_setbit(to_vmx(vcpu), CPU_BASED_MOV_DR_EXITING);
4969 }
4970 
4971 static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
4972 {
4973 	vmcs_writel(GUEST_DR7, val);
4974 }
4975 
4976 static int handle_cpuid(struct kvm_vcpu *vcpu)
4977 {
4978 	return kvm_emulate_cpuid(vcpu);
4979 }
4980 
4981 static int handle_rdmsr(struct kvm_vcpu *vcpu)
4982 {
4983 	return kvm_emulate_rdmsr(vcpu);
4984 }
4985 
4986 static int handle_wrmsr(struct kvm_vcpu *vcpu)
4987 {
4988 	return kvm_emulate_wrmsr(vcpu);
4989 }
4990 
4991 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
4992 {
4993 	kvm_apic_update_ppr(vcpu);
4994 	return 1;
4995 }
4996 
4997 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
4998 {
4999 	exec_controls_clearbit(to_vmx(vcpu), CPU_BASED_VIRTUAL_INTR_PENDING);
5000 
5001 	kvm_make_request(KVM_REQ_EVENT, vcpu);
5002 
5003 	++vcpu->stat.irq_window_exits;
5004 	return 1;
5005 }
5006 
5007 static int handle_halt(struct kvm_vcpu *vcpu)
5008 {
5009 	return kvm_emulate_halt(vcpu);
5010 }
5011 
5012 static int handle_vmcall(struct kvm_vcpu *vcpu)
5013 {
5014 	return kvm_emulate_hypercall(vcpu);
5015 }
5016 
5017 static int handle_invd(struct kvm_vcpu *vcpu)
5018 {
5019 	return kvm_emulate_instruction(vcpu, 0);
5020 }
5021 
5022 static int handle_invlpg(struct kvm_vcpu *vcpu)
5023 {
5024 	unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5025 
5026 	kvm_mmu_invlpg(vcpu, exit_qualification);
5027 	return kvm_skip_emulated_instruction(vcpu);
5028 }
5029 
5030 static int handle_rdpmc(struct kvm_vcpu *vcpu)
5031 {
5032 	int err;
5033 
5034 	err = kvm_rdpmc(vcpu);
5035 	return kvm_complete_insn_gp(vcpu, err);
5036 }
5037 
5038 static int handle_wbinvd(struct kvm_vcpu *vcpu)
5039 {
5040 	return kvm_emulate_wbinvd(vcpu);
5041 }
5042 
5043 static int handle_xsetbv(struct kvm_vcpu *vcpu)
5044 {
5045 	u64 new_bv = kvm_read_edx_eax(vcpu);
5046 	u32 index = kvm_rcx_read(vcpu);
5047 
5048 	if (kvm_set_xcr(vcpu, index, new_bv) == 0)
5049 		return kvm_skip_emulated_instruction(vcpu);
5050 	return 1;
5051 }
5052 
5053 static int handle_apic_access(struct kvm_vcpu *vcpu)
5054 {
5055 	if (likely(fasteoi)) {
5056 		unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5057 		int access_type, offset;
5058 
5059 		access_type = exit_qualification & APIC_ACCESS_TYPE;
5060 		offset = exit_qualification & APIC_ACCESS_OFFSET;
5061 		/*
5062 		 * Sane guest uses MOV to write EOI, with written value
5063 		 * not cared. So make a short-circuit here by avoiding
5064 		 * heavy instruction emulation.
5065 		 */
5066 		if ((access_type == TYPE_LINEAR_APIC_INST_WRITE) &&
5067 		    (offset == APIC_EOI)) {
5068 			kvm_lapic_set_eoi(vcpu);
5069 			return kvm_skip_emulated_instruction(vcpu);
5070 		}
5071 	}
5072 	return kvm_emulate_instruction(vcpu, 0);
5073 }
5074 
5075 static int handle_apic_eoi_induced(struct kvm_vcpu *vcpu)
5076 {
5077 	unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5078 	int vector = exit_qualification & 0xff;
5079 
5080 	/* EOI-induced VM exit is trap-like and thus no need to adjust IP */
5081 	kvm_apic_set_eoi_accelerated(vcpu, vector);
5082 	return 1;
5083 }
5084 
5085 static int handle_apic_write(struct kvm_vcpu *vcpu)
5086 {
5087 	unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5088 	u32 offset = exit_qualification & 0xfff;
5089 
5090 	/* APIC-write VM exit is trap-like and thus no need to adjust IP */
5091 	kvm_apic_write_nodecode(vcpu, offset);
5092 	return 1;
5093 }
5094 
5095 static int handle_task_switch(struct kvm_vcpu *vcpu)
5096 {
5097 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5098 	unsigned long exit_qualification;
5099 	bool has_error_code = false;
5100 	u32 error_code = 0;
5101 	u16 tss_selector;
5102 	int reason, type, idt_v, idt_index;
5103 
5104 	idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
5105 	idt_index = (vmx->idt_vectoring_info & VECTORING_INFO_VECTOR_MASK);
5106 	type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
5107 
5108 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5109 
5110 	reason = (u32)exit_qualification >> 30;
5111 	if (reason == TASK_SWITCH_GATE && idt_v) {
5112 		switch (type) {
5113 		case INTR_TYPE_NMI_INTR:
5114 			vcpu->arch.nmi_injected = false;
5115 			vmx_set_nmi_mask(vcpu, true);
5116 			break;
5117 		case INTR_TYPE_EXT_INTR:
5118 		case INTR_TYPE_SOFT_INTR:
5119 			kvm_clear_interrupt_queue(vcpu);
5120 			break;
5121 		case INTR_TYPE_HARD_EXCEPTION:
5122 			if (vmx->idt_vectoring_info &
5123 			    VECTORING_INFO_DELIVER_CODE_MASK) {
5124 				has_error_code = true;
5125 				error_code =
5126 					vmcs_read32(IDT_VECTORING_ERROR_CODE);
5127 			}
5128 			/* fall through */
5129 		case INTR_TYPE_SOFT_EXCEPTION:
5130 			kvm_clear_exception_queue(vcpu);
5131 			break;
5132 		default:
5133 			break;
5134 		}
5135 	}
5136 	tss_selector = exit_qualification;
5137 
5138 	if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
5139 		       type != INTR_TYPE_EXT_INTR &&
5140 		       type != INTR_TYPE_NMI_INTR))
5141 		WARN_ON(!skip_emulated_instruction(vcpu));
5142 
5143 	/*
5144 	 * TODO: What about debug traps on tss switch?
5145 	 *       Are we supposed to inject them and update dr6?
5146 	 */
5147 	return kvm_task_switch(vcpu, tss_selector,
5148 			       type == INTR_TYPE_SOFT_INTR ? idt_index : -1,
5149 			       reason, has_error_code, error_code);
5150 }
5151 
5152 static int handle_ept_violation(struct kvm_vcpu *vcpu)
5153 {
5154 	unsigned long exit_qualification;
5155 	gpa_t gpa;
5156 	u64 error_code;
5157 
5158 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5159 
5160 	/*
5161 	 * EPT violation happened while executing iret from NMI,
5162 	 * "blocked by NMI" bit has to be set before next VM entry.
5163 	 * There are errata that may cause this bit to not be set:
5164 	 * AAK134, BY25.
5165 	 */
5166 	if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
5167 			enable_vnmi &&
5168 			(exit_qualification & INTR_INFO_UNBLOCK_NMI))
5169 		vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO, GUEST_INTR_STATE_NMI);
5170 
5171 	gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5172 	trace_kvm_page_fault(gpa, exit_qualification);
5173 
5174 	/* Is it a read fault? */
5175 	error_code = (exit_qualification & EPT_VIOLATION_ACC_READ)
5176 		     ? PFERR_USER_MASK : 0;
5177 	/* Is it a write fault? */
5178 	error_code |= (exit_qualification & EPT_VIOLATION_ACC_WRITE)
5179 		      ? PFERR_WRITE_MASK : 0;
5180 	/* Is it a fetch fault? */
5181 	error_code |= (exit_qualification & EPT_VIOLATION_ACC_INSTR)
5182 		      ? PFERR_FETCH_MASK : 0;
5183 	/* ept page table entry is present? */
5184 	error_code |= (exit_qualification &
5185 		       (EPT_VIOLATION_READABLE | EPT_VIOLATION_WRITABLE |
5186 			EPT_VIOLATION_EXECUTABLE))
5187 		      ? PFERR_PRESENT_MASK : 0;
5188 
5189 	error_code |= (exit_qualification & 0x100) != 0 ?
5190 	       PFERR_GUEST_FINAL_MASK : PFERR_GUEST_PAGE_MASK;
5191 
5192 	vcpu->arch.exit_qualification = exit_qualification;
5193 	return kvm_mmu_page_fault(vcpu, gpa, error_code, NULL, 0);
5194 }
5195 
5196 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
5197 {
5198 	gpa_t gpa;
5199 
5200 	/*
5201 	 * A nested guest cannot optimize MMIO vmexits, because we have an
5202 	 * nGPA here instead of the required GPA.
5203 	 */
5204 	gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5205 	if (!is_guest_mode(vcpu) &&
5206 	    !kvm_io_bus_write(vcpu, KVM_FAST_MMIO_BUS, gpa, 0, NULL)) {
5207 		trace_kvm_fast_mmio(gpa);
5208 		return kvm_skip_emulated_instruction(vcpu);
5209 	}
5210 
5211 	return kvm_mmu_page_fault(vcpu, gpa, PFERR_RSVD_MASK, NULL, 0);
5212 }
5213 
5214 static int handle_nmi_window(struct kvm_vcpu *vcpu)
5215 {
5216 	WARN_ON_ONCE(!enable_vnmi);
5217 	exec_controls_clearbit(to_vmx(vcpu), CPU_BASED_VIRTUAL_NMI_PENDING);
5218 	++vcpu->stat.nmi_window_exits;
5219 	kvm_make_request(KVM_REQ_EVENT, vcpu);
5220 
5221 	return 1;
5222 }
5223 
5224 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
5225 {
5226 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5227 	bool intr_window_requested;
5228 	unsigned count = 130;
5229 
5230 	/*
5231 	 * We should never reach the point where we are emulating L2
5232 	 * due to invalid guest state as that means we incorrectly
5233 	 * allowed a nested VMEntry with an invalid vmcs12.
5234 	 */
5235 	WARN_ON_ONCE(vmx->emulation_required && vmx->nested.nested_run_pending);
5236 
5237 	intr_window_requested = exec_controls_get(vmx) &
5238 				CPU_BASED_VIRTUAL_INTR_PENDING;
5239 
5240 	while (vmx->emulation_required && count-- != 0) {
5241 		if (intr_window_requested && vmx_interrupt_allowed(vcpu))
5242 			return handle_interrupt_window(&vmx->vcpu);
5243 
5244 		if (kvm_test_request(KVM_REQ_EVENT, vcpu))
5245 			return 1;
5246 
5247 		if (!kvm_emulate_instruction(vcpu, 0))
5248 			return 0;
5249 
5250 		if (vmx->emulation_required && !vmx->rmode.vm86_active &&
5251 		    vcpu->arch.exception.pending) {
5252 			vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5253 			vcpu->run->internal.suberror =
5254 						KVM_INTERNAL_ERROR_EMULATION;
5255 			vcpu->run->internal.ndata = 0;
5256 			return 0;
5257 		}
5258 
5259 		if (vcpu->arch.halt_request) {
5260 			vcpu->arch.halt_request = 0;
5261 			return kvm_vcpu_halt(vcpu);
5262 		}
5263 
5264 		/*
5265 		 * Note, return 1 and not 0, vcpu_run() is responsible for
5266 		 * morphing the pending signal into the proper return code.
5267 		 */
5268 		if (signal_pending(current))
5269 			return 1;
5270 
5271 		if (need_resched())
5272 			schedule();
5273 	}
5274 
5275 	return 1;
5276 }
5277 
5278 static void grow_ple_window(struct kvm_vcpu *vcpu)
5279 {
5280 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5281 	unsigned int old = vmx->ple_window;
5282 
5283 	vmx->ple_window = __grow_ple_window(old, ple_window,
5284 					    ple_window_grow,
5285 					    ple_window_max);
5286 
5287 	if (vmx->ple_window != old) {
5288 		vmx->ple_window_dirty = true;
5289 		trace_kvm_ple_window_update(vcpu->vcpu_id,
5290 					    vmx->ple_window, old);
5291 	}
5292 }
5293 
5294 static void shrink_ple_window(struct kvm_vcpu *vcpu)
5295 {
5296 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5297 	unsigned int old = vmx->ple_window;
5298 
5299 	vmx->ple_window = __shrink_ple_window(old, ple_window,
5300 					      ple_window_shrink,
5301 					      ple_window);
5302 
5303 	if (vmx->ple_window != old) {
5304 		vmx->ple_window_dirty = true;
5305 		trace_kvm_ple_window_update(vcpu->vcpu_id,
5306 					    vmx->ple_window, old);
5307 	}
5308 }
5309 
5310 /*
5311  * Handler for POSTED_INTERRUPT_WAKEUP_VECTOR.
5312  */
5313 static void wakeup_handler(void)
5314 {
5315 	struct kvm_vcpu *vcpu;
5316 	int cpu = smp_processor_id();
5317 
5318 	spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
5319 	list_for_each_entry(vcpu, &per_cpu(blocked_vcpu_on_cpu, cpu),
5320 			blocked_vcpu_list) {
5321 		struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
5322 
5323 		if (pi_test_on(pi_desc) == 1)
5324 			kvm_vcpu_kick(vcpu);
5325 	}
5326 	spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
5327 }
5328 
5329 static void vmx_enable_tdp(void)
5330 {
5331 	kvm_mmu_set_mask_ptes(VMX_EPT_READABLE_MASK,
5332 		enable_ept_ad_bits ? VMX_EPT_ACCESS_BIT : 0ull,
5333 		enable_ept_ad_bits ? VMX_EPT_DIRTY_BIT : 0ull,
5334 		0ull, VMX_EPT_EXECUTABLE_MASK,
5335 		cpu_has_vmx_ept_execute_only() ? 0ull : VMX_EPT_READABLE_MASK,
5336 		VMX_EPT_RWX_MASK, 0ull);
5337 
5338 	ept_set_mmio_spte_mask();
5339 	kvm_enable_tdp();
5340 }
5341 
5342 /*
5343  * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
5344  * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
5345  */
5346 static int handle_pause(struct kvm_vcpu *vcpu)
5347 {
5348 	if (!kvm_pause_in_guest(vcpu->kvm))
5349 		grow_ple_window(vcpu);
5350 
5351 	/*
5352 	 * Intel sdm vol3 ch-25.1.3 says: The "PAUSE-loop exiting"
5353 	 * VM-execution control is ignored if CPL > 0. OTOH, KVM
5354 	 * never set PAUSE_EXITING and just set PLE if supported,
5355 	 * so the vcpu must be CPL=0 if it gets a PAUSE exit.
5356 	 */
5357 	kvm_vcpu_on_spin(vcpu, true);
5358 	return kvm_skip_emulated_instruction(vcpu);
5359 }
5360 
5361 static int handle_nop(struct kvm_vcpu *vcpu)
5362 {
5363 	return kvm_skip_emulated_instruction(vcpu);
5364 }
5365 
5366 static int handle_mwait(struct kvm_vcpu *vcpu)
5367 {
5368 	printk_once(KERN_WARNING "kvm: MWAIT instruction emulated as NOP!\n");
5369 	return handle_nop(vcpu);
5370 }
5371 
5372 static int handle_invalid_op(struct kvm_vcpu *vcpu)
5373 {
5374 	kvm_queue_exception(vcpu, UD_VECTOR);
5375 	return 1;
5376 }
5377 
5378 static int handle_monitor_trap(struct kvm_vcpu *vcpu)
5379 {
5380 	return 1;
5381 }
5382 
5383 static int handle_monitor(struct kvm_vcpu *vcpu)
5384 {
5385 	printk_once(KERN_WARNING "kvm: MONITOR instruction emulated as NOP!\n");
5386 	return handle_nop(vcpu);
5387 }
5388 
5389 static int handle_invpcid(struct kvm_vcpu *vcpu)
5390 {
5391 	u32 vmx_instruction_info;
5392 	unsigned long type;
5393 	bool pcid_enabled;
5394 	gva_t gva;
5395 	struct x86_exception e;
5396 	unsigned i;
5397 	unsigned long roots_to_free = 0;
5398 	struct {
5399 		u64 pcid;
5400 		u64 gla;
5401 	} operand;
5402 
5403 	if (!guest_cpuid_has(vcpu, X86_FEATURE_INVPCID)) {
5404 		kvm_queue_exception(vcpu, UD_VECTOR);
5405 		return 1;
5406 	}
5407 
5408 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5409 	type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
5410 
5411 	if (type > 3) {
5412 		kvm_inject_gp(vcpu, 0);
5413 		return 1;
5414 	}
5415 
5416 	/* According to the Intel instruction reference, the memory operand
5417 	 * is read even if it isn't needed (e.g., for type==all)
5418 	 */
5419 	if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
5420 				vmx_instruction_info, false,
5421 				sizeof(operand), &gva))
5422 		return 1;
5423 
5424 	if (kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e)) {
5425 		kvm_inject_page_fault(vcpu, &e);
5426 		return 1;
5427 	}
5428 
5429 	if (operand.pcid >> 12 != 0) {
5430 		kvm_inject_gp(vcpu, 0);
5431 		return 1;
5432 	}
5433 
5434 	pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
5435 
5436 	switch (type) {
5437 	case INVPCID_TYPE_INDIV_ADDR:
5438 		if ((!pcid_enabled && (operand.pcid != 0)) ||
5439 		    is_noncanonical_address(operand.gla, vcpu)) {
5440 			kvm_inject_gp(vcpu, 0);
5441 			return 1;
5442 		}
5443 		kvm_mmu_invpcid_gva(vcpu, operand.gla, operand.pcid);
5444 		return kvm_skip_emulated_instruction(vcpu);
5445 
5446 	case INVPCID_TYPE_SINGLE_CTXT:
5447 		if (!pcid_enabled && (operand.pcid != 0)) {
5448 			kvm_inject_gp(vcpu, 0);
5449 			return 1;
5450 		}
5451 
5452 		if (kvm_get_active_pcid(vcpu) == operand.pcid) {
5453 			kvm_mmu_sync_roots(vcpu);
5454 			kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
5455 		}
5456 
5457 		for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++)
5458 			if (kvm_get_pcid(vcpu, vcpu->arch.mmu->prev_roots[i].cr3)
5459 			    == operand.pcid)
5460 				roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i);
5461 
5462 		kvm_mmu_free_roots(vcpu, vcpu->arch.mmu, roots_to_free);
5463 		/*
5464 		 * If neither the current cr3 nor any of the prev_roots use the
5465 		 * given PCID, then nothing needs to be done here because a
5466 		 * resync will happen anyway before switching to any other CR3.
5467 		 */
5468 
5469 		return kvm_skip_emulated_instruction(vcpu);
5470 
5471 	case INVPCID_TYPE_ALL_NON_GLOBAL:
5472 		/*
5473 		 * Currently, KVM doesn't mark global entries in the shadow
5474 		 * page tables, so a non-global flush just degenerates to a
5475 		 * global flush. If needed, we could optimize this later by
5476 		 * keeping track of global entries in shadow page tables.
5477 		 */
5478 
5479 		/* fall-through */
5480 	case INVPCID_TYPE_ALL_INCL_GLOBAL:
5481 		kvm_mmu_unload(vcpu);
5482 		return kvm_skip_emulated_instruction(vcpu);
5483 
5484 	default:
5485 		BUG(); /* We have already checked above that type <= 3 */
5486 	}
5487 }
5488 
5489 static int handle_pml_full(struct kvm_vcpu *vcpu)
5490 {
5491 	unsigned long exit_qualification;
5492 
5493 	trace_kvm_pml_full(vcpu->vcpu_id);
5494 
5495 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5496 
5497 	/*
5498 	 * PML buffer FULL happened while executing iret from NMI,
5499 	 * "blocked by NMI" bit has to be set before next VM entry.
5500 	 */
5501 	if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
5502 			enable_vnmi &&
5503 			(exit_qualification & INTR_INFO_UNBLOCK_NMI))
5504 		vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
5505 				GUEST_INTR_STATE_NMI);
5506 
5507 	/*
5508 	 * PML buffer already flushed at beginning of VMEXIT. Nothing to do
5509 	 * here.., and there's no userspace involvement needed for PML.
5510 	 */
5511 	return 1;
5512 }
5513 
5514 static int handle_preemption_timer(struct kvm_vcpu *vcpu)
5515 {
5516 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5517 
5518 	if (!vmx->req_immediate_exit &&
5519 	    !unlikely(vmx->loaded_vmcs->hv_timer_soft_disabled))
5520 		kvm_lapic_expired_hv_timer(vcpu);
5521 
5522 	return 1;
5523 }
5524 
5525 /*
5526  * When nested=0, all VMX instruction VM Exits filter here.  The handlers
5527  * are overwritten by nested_vmx_setup() when nested=1.
5528  */
5529 static int handle_vmx_instruction(struct kvm_vcpu *vcpu)
5530 {
5531 	kvm_queue_exception(vcpu, UD_VECTOR);
5532 	return 1;
5533 }
5534 
5535 static int handle_encls(struct kvm_vcpu *vcpu)
5536 {
5537 	/*
5538 	 * SGX virtualization is not yet supported.  There is no software
5539 	 * enable bit for SGX, so we have to trap ENCLS and inject a #UD
5540 	 * to prevent the guest from executing ENCLS.
5541 	 */
5542 	kvm_queue_exception(vcpu, UD_VECTOR);
5543 	return 1;
5544 }
5545 
5546 static int handle_unexpected_vmexit(struct kvm_vcpu *vcpu)
5547 {
5548 	kvm_skip_emulated_instruction(vcpu);
5549 	WARN_ONCE(1, "Unexpected VM-Exit Reason = 0x%x",
5550 		vmcs_read32(VM_EXIT_REASON));
5551 	return 1;
5552 }
5553 
5554 /*
5555  * The exit handlers return 1 if the exit was handled fully and guest execution
5556  * may resume.  Otherwise they set the kvm_run parameter to indicate what needs
5557  * to be done to userspace and return 0.
5558  */
5559 static int (*kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
5560 	[EXIT_REASON_EXCEPTION_NMI]           = handle_exception_nmi,
5561 	[EXIT_REASON_EXTERNAL_INTERRUPT]      = handle_external_interrupt,
5562 	[EXIT_REASON_TRIPLE_FAULT]            = handle_triple_fault,
5563 	[EXIT_REASON_NMI_WINDOW]	      = handle_nmi_window,
5564 	[EXIT_REASON_IO_INSTRUCTION]          = handle_io,
5565 	[EXIT_REASON_CR_ACCESS]               = handle_cr,
5566 	[EXIT_REASON_DR_ACCESS]               = handle_dr,
5567 	[EXIT_REASON_CPUID]                   = handle_cpuid,
5568 	[EXIT_REASON_MSR_READ]                = handle_rdmsr,
5569 	[EXIT_REASON_MSR_WRITE]               = handle_wrmsr,
5570 	[EXIT_REASON_PENDING_INTERRUPT]       = handle_interrupt_window,
5571 	[EXIT_REASON_HLT]                     = handle_halt,
5572 	[EXIT_REASON_INVD]		      = handle_invd,
5573 	[EXIT_REASON_INVLPG]		      = handle_invlpg,
5574 	[EXIT_REASON_RDPMC]                   = handle_rdpmc,
5575 	[EXIT_REASON_VMCALL]                  = handle_vmcall,
5576 	[EXIT_REASON_VMCLEAR]		      = handle_vmx_instruction,
5577 	[EXIT_REASON_VMLAUNCH]		      = handle_vmx_instruction,
5578 	[EXIT_REASON_VMPTRLD]		      = handle_vmx_instruction,
5579 	[EXIT_REASON_VMPTRST]		      = handle_vmx_instruction,
5580 	[EXIT_REASON_VMREAD]		      = handle_vmx_instruction,
5581 	[EXIT_REASON_VMRESUME]		      = handle_vmx_instruction,
5582 	[EXIT_REASON_VMWRITE]		      = handle_vmx_instruction,
5583 	[EXIT_REASON_VMOFF]		      = handle_vmx_instruction,
5584 	[EXIT_REASON_VMON]		      = handle_vmx_instruction,
5585 	[EXIT_REASON_TPR_BELOW_THRESHOLD]     = handle_tpr_below_threshold,
5586 	[EXIT_REASON_APIC_ACCESS]             = handle_apic_access,
5587 	[EXIT_REASON_APIC_WRITE]              = handle_apic_write,
5588 	[EXIT_REASON_EOI_INDUCED]             = handle_apic_eoi_induced,
5589 	[EXIT_REASON_WBINVD]                  = handle_wbinvd,
5590 	[EXIT_REASON_XSETBV]                  = handle_xsetbv,
5591 	[EXIT_REASON_TASK_SWITCH]             = handle_task_switch,
5592 	[EXIT_REASON_MCE_DURING_VMENTRY]      = handle_machine_check,
5593 	[EXIT_REASON_GDTR_IDTR]		      = handle_desc,
5594 	[EXIT_REASON_LDTR_TR]		      = handle_desc,
5595 	[EXIT_REASON_EPT_VIOLATION]	      = handle_ept_violation,
5596 	[EXIT_REASON_EPT_MISCONFIG]           = handle_ept_misconfig,
5597 	[EXIT_REASON_PAUSE_INSTRUCTION]       = handle_pause,
5598 	[EXIT_REASON_MWAIT_INSTRUCTION]	      = handle_mwait,
5599 	[EXIT_REASON_MONITOR_TRAP_FLAG]       = handle_monitor_trap,
5600 	[EXIT_REASON_MONITOR_INSTRUCTION]     = handle_monitor,
5601 	[EXIT_REASON_INVEPT]                  = handle_vmx_instruction,
5602 	[EXIT_REASON_INVVPID]                 = handle_vmx_instruction,
5603 	[EXIT_REASON_RDRAND]                  = handle_invalid_op,
5604 	[EXIT_REASON_RDSEED]                  = handle_invalid_op,
5605 	[EXIT_REASON_XSAVES]                  = handle_unexpected_vmexit,
5606 	[EXIT_REASON_XRSTORS]                 = handle_unexpected_vmexit,
5607 	[EXIT_REASON_PML_FULL]		      = handle_pml_full,
5608 	[EXIT_REASON_INVPCID]                 = handle_invpcid,
5609 	[EXIT_REASON_VMFUNC]		      = handle_vmx_instruction,
5610 	[EXIT_REASON_PREEMPTION_TIMER]	      = handle_preemption_timer,
5611 	[EXIT_REASON_ENCLS]		      = handle_encls,
5612 	[EXIT_REASON_UMWAIT]                  = handle_unexpected_vmexit,
5613 	[EXIT_REASON_TPAUSE]                  = handle_unexpected_vmexit,
5614 };
5615 
5616 static const int kvm_vmx_max_exit_handlers =
5617 	ARRAY_SIZE(kvm_vmx_exit_handlers);
5618 
5619 static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
5620 {
5621 	*info1 = vmcs_readl(EXIT_QUALIFICATION);
5622 	*info2 = vmcs_read32(VM_EXIT_INTR_INFO);
5623 }
5624 
5625 static void vmx_destroy_pml_buffer(struct vcpu_vmx *vmx)
5626 {
5627 	if (vmx->pml_pg) {
5628 		__free_page(vmx->pml_pg);
5629 		vmx->pml_pg = NULL;
5630 	}
5631 }
5632 
5633 static void vmx_flush_pml_buffer(struct kvm_vcpu *vcpu)
5634 {
5635 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5636 	u64 *pml_buf;
5637 	u16 pml_idx;
5638 
5639 	pml_idx = vmcs_read16(GUEST_PML_INDEX);
5640 
5641 	/* Do nothing if PML buffer is empty */
5642 	if (pml_idx == (PML_ENTITY_NUM - 1))
5643 		return;
5644 
5645 	/* PML index always points to next available PML buffer entity */
5646 	if (pml_idx >= PML_ENTITY_NUM)
5647 		pml_idx = 0;
5648 	else
5649 		pml_idx++;
5650 
5651 	pml_buf = page_address(vmx->pml_pg);
5652 	for (; pml_idx < PML_ENTITY_NUM; pml_idx++) {
5653 		u64 gpa;
5654 
5655 		gpa = pml_buf[pml_idx];
5656 		WARN_ON(gpa & (PAGE_SIZE - 1));
5657 		kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
5658 	}
5659 
5660 	/* reset PML index */
5661 	vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
5662 }
5663 
5664 /*
5665  * Flush all vcpus' PML buffer and update logged GPAs to dirty_bitmap.
5666  * Called before reporting dirty_bitmap to userspace.
5667  */
5668 static void kvm_flush_pml_buffers(struct kvm *kvm)
5669 {
5670 	int i;
5671 	struct kvm_vcpu *vcpu;
5672 	/*
5673 	 * We only need to kick vcpu out of guest mode here, as PML buffer
5674 	 * is flushed at beginning of all VMEXITs, and it's obvious that only
5675 	 * vcpus running in guest are possible to have unflushed GPAs in PML
5676 	 * buffer.
5677 	 */
5678 	kvm_for_each_vcpu(i, vcpu, kvm)
5679 		kvm_vcpu_kick(vcpu);
5680 }
5681 
5682 static void vmx_dump_sel(char *name, uint32_t sel)
5683 {
5684 	pr_err("%s sel=0x%04x, attr=0x%05x, limit=0x%08x, base=0x%016lx\n",
5685 	       name, vmcs_read16(sel),
5686 	       vmcs_read32(sel + GUEST_ES_AR_BYTES - GUEST_ES_SELECTOR),
5687 	       vmcs_read32(sel + GUEST_ES_LIMIT - GUEST_ES_SELECTOR),
5688 	       vmcs_readl(sel + GUEST_ES_BASE - GUEST_ES_SELECTOR));
5689 }
5690 
5691 static void vmx_dump_dtsel(char *name, uint32_t limit)
5692 {
5693 	pr_err("%s                           limit=0x%08x, base=0x%016lx\n",
5694 	       name, vmcs_read32(limit),
5695 	       vmcs_readl(limit + GUEST_GDTR_BASE - GUEST_GDTR_LIMIT));
5696 }
5697 
5698 void dump_vmcs(void)
5699 {
5700 	u32 vmentry_ctl, vmexit_ctl;
5701 	u32 cpu_based_exec_ctrl, pin_based_exec_ctrl, secondary_exec_control;
5702 	unsigned long cr4;
5703 	u64 efer;
5704 	int i, n;
5705 
5706 	if (!dump_invalid_vmcs) {
5707 		pr_warn_ratelimited("set kvm_intel.dump_invalid_vmcs=1 to dump internal KVM state.\n");
5708 		return;
5709 	}
5710 
5711 	vmentry_ctl = vmcs_read32(VM_ENTRY_CONTROLS);
5712 	vmexit_ctl = vmcs_read32(VM_EXIT_CONTROLS);
5713 	cpu_based_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5714 	pin_based_exec_ctrl = vmcs_read32(PIN_BASED_VM_EXEC_CONTROL);
5715 	cr4 = vmcs_readl(GUEST_CR4);
5716 	efer = vmcs_read64(GUEST_IA32_EFER);
5717 	secondary_exec_control = 0;
5718 	if (cpu_has_secondary_exec_ctrls())
5719 		secondary_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
5720 
5721 	pr_err("*** Guest State ***\n");
5722 	pr_err("CR0: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
5723 	       vmcs_readl(GUEST_CR0), vmcs_readl(CR0_READ_SHADOW),
5724 	       vmcs_readl(CR0_GUEST_HOST_MASK));
5725 	pr_err("CR4: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
5726 	       cr4, vmcs_readl(CR4_READ_SHADOW), vmcs_readl(CR4_GUEST_HOST_MASK));
5727 	pr_err("CR3 = 0x%016lx\n", vmcs_readl(GUEST_CR3));
5728 	if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT) &&
5729 	    (cr4 & X86_CR4_PAE) && !(efer & EFER_LMA))
5730 	{
5731 		pr_err("PDPTR0 = 0x%016llx  PDPTR1 = 0x%016llx\n",
5732 		       vmcs_read64(GUEST_PDPTR0), vmcs_read64(GUEST_PDPTR1));
5733 		pr_err("PDPTR2 = 0x%016llx  PDPTR3 = 0x%016llx\n",
5734 		       vmcs_read64(GUEST_PDPTR2), vmcs_read64(GUEST_PDPTR3));
5735 	}
5736 	pr_err("RSP = 0x%016lx  RIP = 0x%016lx\n",
5737 	       vmcs_readl(GUEST_RSP), vmcs_readl(GUEST_RIP));
5738 	pr_err("RFLAGS=0x%08lx         DR7 = 0x%016lx\n",
5739 	       vmcs_readl(GUEST_RFLAGS), vmcs_readl(GUEST_DR7));
5740 	pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
5741 	       vmcs_readl(GUEST_SYSENTER_ESP),
5742 	       vmcs_read32(GUEST_SYSENTER_CS), vmcs_readl(GUEST_SYSENTER_EIP));
5743 	vmx_dump_sel("CS:  ", GUEST_CS_SELECTOR);
5744 	vmx_dump_sel("DS:  ", GUEST_DS_SELECTOR);
5745 	vmx_dump_sel("SS:  ", GUEST_SS_SELECTOR);
5746 	vmx_dump_sel("ES:  ", GUEST_ES_SELECTOR);
5747 	vmx_dump_sel("FS:  ", GUEST_FS_SELECTOR);
5748 	vmx_dump_sel("GS:  ", GUEST_GS_SELECTOR);
5749 	vmx_dump_dtsel("GDTR:", GUEST_GDTR_LIMIT);
5750 	vmx_dump_sel("LDTR:", GUEST_LDTR_SELECTOR);
5751 	vmx_dump_dtsel("IDTR:", GUEST_IDTR_LIMIT);
5752 	vmx_dump_sel("TR:  ", GUEST_TR_SELECTOR);
5753 	if ((vmexit_ctl & (VM_EXIT_SAVE_IA32_PAT | VM_EXIT_SAVE_IA32_EFER)) ||
5754 	    (vmentry_ctl & (VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_IA32_EFER)))
5755 		pr_err("EFER =     0x%016llx  PAT = 0x%016llx\n",
5756 		       efer, vmcs_read64(GUEST_IA32_PAT));
5757 	pr_err("DebugCtl = 0x%016llx  DebugExceptions = 0x%016lx\n",
5758 	       vmcs_read64(GUEST_IA32_DEBUGCTL),
5759 	       vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS));
5760 	if (cpu_has_load_perf_global_ctrl() &&
5761 	    vmentry_ctl & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
5762 		pr_err("PerfGlobCtl = 0x%016llx\n",
5763 		       vmcs_read64(GUEST_IA32_PERF_GLOBAL_CTRL));
5764 	if (vmentry_ctl & VM_ENTRY_LOAD_BNDCFGS)
5765 		pr_err("BndCfgS = 0x%016llx\n", vmcs_read64(GUEST_BNDCFGS));
5766 	pr_err("Interruptibility = %08x  ActivityState = %08x\n",
5767 	       vmcs_read32(GUEST_INTERRUPTIBILITY_INFO),
5768 	       vmcs_read32(GUEST_ACTIVITY_STATE));
5769 	if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)
5770 		pr_err("InterruptStatus = %04x\n",
5771 		       vmcs_read16(GUEST_INTR_STATUS));
5772 
5773 	pr_err("*** Host State ***\n");
5774 	pr_err("RIP = 0x%016lx  RSP = 0x%016lx\n",
5775 	       vmcs_readl(HOST_RIP), vmcs_readl(HOST_RSP));
5776 	pr_err("CS=%04x SS=%04x DS=%04x ES=%04x FS=%04x GS=%04x TR=%04x\n",
5777 	       vmcs_read16(HOST_CS_SELECTOR), vmcs_read16(HOST_SS_SELECTOR),
5778 	       vmcs_read16(HOST_DS_SELECTOR), vmcs_read16(HOST_ES_SELECTOR),
5779 	       vmcs_read16(HOST_FS_SELECTOR), vmcs_read16(HOST_GS_SELECTOR),
5780 	       vmcs_read16(HOST_TR_SELECTOR));
5781 	pr_err("FSBase=%016lx GSBase=%016lx TRBase=%016lx\n",
5782 	       vmcs_readl(HOST_FS_BASE), vmcs_readl(HOST_GS_BASE),
5783 	       vmcs_readl(HOST_TR_BASE));
5784 	pr_err("GDTBase=%016lx IDTBase=%016lx\n",
5785 	       vmcs_readl(HOST_GDTR_BASE), vmcs_readl(HOST_IDTR_BASE));
5786 	pr_err("CR0=%016lx CR3=%016lx CR4=%016lx\n",
5787 	       vmcs_readl(HOST_CR0), vmcs_readl(HOST_CR3),
5788 	       vmcs_readl(HOST_CR4));
5789 	pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
5790 	       vmcs_readl(HOST_IA32_SYSENTER_ESP),
5791 	       vmcs_read32(HOST_IA32_SYSENTER_CS),
5792 	       vmcs_readl(HOST_IA32_SYSENTER_EIP));
5793 	if (vmexit_ctl & (VM_EXIT_LOAD_IA32_PAT | VM_EXIT_LOAD_IA32_EFER))
5794 		pr_err("EFER = 0x%016llx  PAT = 0x%016llx\n",
5795 		       vmcs_read64(HOST_IA32_EFER),
5796 		       vmcs_read64(HOST_IA32_PAT));
5797 	if (cpu_has_load_perf_global_ctrl() &&
5798 	    vmexit_ctl & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
5799 		pr_err("PerfGlobCtl = 0x%016llx\n",
5800 		       vmcs_read64(HOST_IA32_PERF_GLOBAL_CTRL));
5801 
5802 	pr_err("*** Control State ***\n");
5803 	pr_err("PinBased=%08x CPUBased=%08x SecondaryExec=%08x\n",
5804 	       pin_based_exec_ctrl, cpu_based_exec_ctrl, secondary_exec_control);
5805 	pr_err("EntryControls=%08x ExitControls=%08x\n", vmentry_ctl, vmexit_ctl);
5806 	pr_err("ExceptionBitmap=%08x PFECmask=%08x PFECmatch=%08x\n",
5807 	       vmcs_read32(EXCEPTION_BITMAP),
5808 	       vmcs_read32(PAGE_FAULT_ERROR_CODE_MASK),
5809 	       vmcs_read32(PAGE_FAULT_ERROR_CODE_MATCH));
5810 	pr_err("VMEntry: intr_info=%08x errcode=%08x ilen=%08x\n",
5811 	       vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
5812 	       vmcs_read32(VM_ENTRY_EXCEPTION_ERROR_CODE),
5813 	       vmcs_read32(VM_ENTRY_INSTRUCTION_LEN));
5814 	pr_err("VMExit: intr_info=%08x errcode=%08x ilen=%08x\n",
5815 	       vmcs_read32(VM_EXIT_INTR_INFO),
5816 	       vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
5817 	       vmcs_read32(VM_EXIT_INSTRUCTION_LEN));
5818 	pr_err("        reason=%08x qualification=%016lx\n",
5819 	       vmcs_read32(VM_EXIT_REASON), vmcs_readl(EXIT_QUALIFICATION));
5820 	pr_err("IDTVectoring: info=%08x errcode=%08x\n",
5821 	       vmcs_read32(IDT_VECTORING_INFO_FIELD),
5822 	       vmcs_read32(IDT_VECTORING_ERROR_CODE));
5823 	pr_err("TSC Offset = 0x%016llx\n", vmcs_read64(TSC_OFFSET));
5824 	if (secondary_exec_control & SECONDARY_EXEC_TSC_SCALING)
5825 		pr_err("TSC Multiplier = 0x%016llx\n",
5826 		       vmcs_read64(TSC_MULTIPLIER));
5827 	if (cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW) {
5828 		if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY) {
5829 			u16 status = vmcs_read16(GUEST_INTR_STATUS);
5830 			pr_err("SVI|RVI = %02x|%02x ", status >> 8, status & 0xff);
5831 		}
5832 		pr_cont("TPR Threshold = 0x%02x\n", vmcs_read32(TPR_THRESHOLD));
5833 		if (secondary_exec_control & SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)
5834 			pr_err("APIC-access addr = 0x%016llx ", vmcs_read64(APIC_ACCESS_ADDR));
5835 		pr_cont("virt-APIC addr = 0x%016llx\n", vmcs_read64(VIRTUAL_APIC_PAGE_ADDR));
5836 	}
5837 	if (pin_based_exec_ctrl & PIN_BASED_POSTED_INTR)
5838 		pr_err("PostedIntrVec = 0x%02x\n", vmcs_read16(POSTED_INTR_NV));
5839 	if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT))
5840 		pr_err("EPT pointer = 0x%016llx\n", vmcs_read64(EPT_POINTER));
5841 	n = vmcs_read32(CR3_TARGET_COUNT);
5842 	for (i = 0; i + 1 < n; i += 4)
5843 		pr_err("CR3 target%u=%016lx target%u=%016lx\n",
5844 		       i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2),
5845 		       i + 1, vmcs_readl(CR3_TARGET_VALUE0 + i * 2 + 2));
5846 	if (i < n)
5847 		pr_err("CR3 target%u=%016lx\n",
5848 		       i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2));
5849 	if (secondary_exec_control & SECONDARY_EXEC_PAUSE_LOOP_EXITING)
5850 		pr_err("PLE Gap=%08x Window=%08x\n",
5851 		       vmcs_read32(PLE_GAP), vmcs_read32(PLE_WINDOW));
5852 	if (secondary_exec_control & SECONDARY_EXEC_ENABLE_VPID)
5853 		pr_err("Virtual processor ID = 0x%04x\n",
5854 		       vmcs_read16(VIRTUAL_PROCESSOR_ID));
5855 }
5856 
5857 /*
5858  * The guest has exited.  See if we can fix it or if we need userspace
5859  * assistance.
5860  */
5861 static int vmx_handle_exit(struct kvm_vcpu *vcpu)
5862 {
5863 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5864 	u32 exit_reason = vmx->exit_reason;
5865 	u32 vectoring_info = vmx->idt_vectoring_info;
5866 
5867 	trace_kvm_exit(exit_reason, vcpu, KVM_ISA_VMX);
5868 
5869 	/*
5870 	 * Flush logged GPAs PML buffer, this will make dirty_bitmap more
5871 	 * updated. Another good is, in kvm_vm_ioctl_get_dirty_log, before
5872 	 * querying dirty_bitmap, we only need to kick all vcpus out of guest
5873 	 * mode as if vcpus is in root mode, the PML buffer must has been
5874 	 * flushed already.
5875 	 */
5876 	if (enable_pml)
5877 		vmx_flush_pml_buffer(vcpu);
5878 
5879 	/* If guest state is invalid, start emulating */
5880 	if (vmx->emulation_required)
5881 		return handle_invalid_guest_state(vcpu);
5882 
5883 	if (is_guest_mode(vcpu) && nested_vmx_exit_reflected(vcpu, exit_reason))
5884 		return nested_vmx_reflect_vmexit(vcpu, exit_reason);
5885 
5886 	if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) {
5887 		dump_vmcs();
5888 		vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
5889 		vcpu->run->fail_entry.hardware_entry_failure_reason
5890 			= exit_reason;
5891 		return 0;
5892 	}
5893 
5894 	if (unlikely(vmx->fail)) {
5895 		dump_vmcs();
5896 		vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
5897 		vcpu->run->fail_entry.hardware_entry_failure_reason
5898 			= vmcs_read32(VM_INSTRUCTION_ERROR);
5899 		return 0;
5900 	}
5901 
5902 	/*
5903 	 * Note:
5904 	 * Do not try to fix EXIT_REASON_EPT_MISCONFIG if it caused by
5905 	 * delivery event since it indicates guest is accessing MMIO.
5906 	 * The vm-exit can be triggered again after return to guest that
5907 	 * will cause infinite loop.
5908 	 */
5909 	if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
5910 			(exit_reason != EXIT_REASON_EXCEPTION_NMI &&
5911 			exit_reason != EXIT_REASON_EPT_VIOLATION &&
5912 			exit_reason != EXIT_REASON_PML_FULL &&
5913 			exit_reason != EXIT_REASON_TASK_SWITCH)) {
5914 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5915 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_DELIVERY_EV;
5916 		vcpu->run->internal.ndata = 3;
5917 		vcpu->run->internal.data[0] = vectoring_info;
5918 		vcpu->run->internal.data[1] = exit_reason;
5919 		vcpu->run->internal.data[2] = vcpu->arch.exit_qualification;
5920 		if (exit_reason == EXIT_REASON_EPT_MISCONFIG) {
5921 			vcpu->run->internal.ndata++;
5922 			vcpu->run->internal.data[3] =
5923 				vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5924 		}
5925 		return 0;
5926 	}
5927 
5928 	if (unlikely(!enable_vnmi &&
5929 		     vmx->loaded_vmcs->soft_vnmi_blocked)) {
5930 		if (vmx_interrupt_allowed(vcpu)) {
5931 			vmx->loaded_vmcs->soft_vnmi_blocked = 0;
5932 		} else if (vmx->loaded_vmcs->vnmi_blocked_time > 1000000000LL &&
5933 			   vcpu->arch.nmi_pending) {
5934 			/*
5935 			 * This CPU don't support us in finding the end of an
5936 			 * NMI-blocked window if the guest runs with IRQs
5937 			 * disabled. So we pull the trigger after 1 s of
5938 			 * futile waiting, but inform the user about this.
5939 			 */
5940 			printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
5941 			       "state on VCPU %d after 1 s timeout\n",
5942 			       __func__, vcpu->vcpu_id);
5943 			vmx->loaded_vmcs->soft_vnmi_blocked = 0;
5944 		}
5945 	}
5946 
5947 	if (exit_reason < kvm_vmx_max_exit_handlers
5948 	    && kvm_vmx_exit_handlers[exit_reason])
5949 		return kvm_vmx_exit_handlers[exit_reason](vcpu);
5950 	else {
5951 		vcpu_unimpl(vcpu, "vmx: unexpected exit reason 0x%x\n",
5952 				exit_reason);
5953 		dump_vmcs();
5954 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5955 		vcpu->run->internal.suberror =
5956 			KVM_INTERNAL_ERROR_UNEXPECTED_EXIT_REASON;
5957 		vcpu->run->internal.ndata = 1;
5958 		vcpu->run->internal.data[0] = exit_reason;
5959 		return 0;
5960 	}
5961 }
5962 
5963 /*
5964  * Software based L1D cache flush which is used when microcode providing
5965  * the cache control MSR is not loaded.
5966  *
5967  * The L1D cache is 32 KiB on Nehalem and later microarchitectures, but to
5968  * flush it is required to read in 64 KiB because the replacement algorithm
5969  * is not exactly LRU. This could be sized at runtime via topology
5970  * information but as all relevant affected CPUs have 32KiB L1D cache size
5971  * there is no point in doing so.
5972  */
5973 static void vmx_l1d_flush(struct kvm_vcpu *vcpu)
5974 {
5975 	int size = PAGE_SIZE << L1D_CACHE_ORDER;
5976 
5977 	/*
5978 	 * This code is only executed when the the flush mode is 'cond' or
5979 	 * 'always'
5980 	 */
5981 	if (static_branch_likely(&vmx_l1d_flush_cond)) {
5982 		bool flush_l1d;
5983 
5984 		/*
5985 		 * Clear the per-vcpu flush bit, it gets set again
5986 		 * either from vcpu_run() or from one of the unsafe
5987 		 * VMEXIT handlers.
5988 		 */
5989 		flush_l1d = vcpu->arch.l1tf_flush_l1d;
5990 		vcpu->arch.l1tf_flush_l1d = false;
5991 
5992 		/*
5993 		 * Clear the per-cpu flush bit, it gets set again from
5994 		 * the interrupt handlers.
5995 		 */
5996 		flush_l1d |= kvm_get_cpu_l1tf_flush_l1d();
5997 		kvm_clear_cpu_l1tf_flush_l1d();
5998 
5999 		if (!flush_l1d)
6000 			return;
6001 	}
6002 
6003 	vcpu->stat.l1d_flush++;
6004 
6005 	if (static_cpu_has(X86_FEATURE_FLUSH_L1D)) {
6006 		wrmsrl(MSR_IA32_FLUSH_CMD, L1D_FLUSH);
6007 		return;
6008 	}
6009 
6010 	asm volatile(
6011 		/* First ensure the pages are in the TLB */
6012 		"xorl	%%eax, %%eax\n"
6013 		".Lpopulate_tlb:\n\t"
6014 		"movzbl	(%[flush_pages], %%" _ASM_AX "), %%ecx\n\t"
6015 		"addl	$4096, %%eax\n\t"
6016 		"cmpl	%%eax, %[size]\n\t"
6017 		"jne	.Lpopulate_tlb\n\t"
6018 		"xorl	%%eax, %%eax\n\t"
6019 		"cpuid\n\t"
6020 		/* Now fill the cache */
6021 		"xorl	%%eax, %%eax\n"
6022 		".Lfill_cache:\n"
6023 		"movzbl	(%[flush_pages], %%" _ASM_AX "), %%ecx\n\t"
6024 		"addl	$64, %%eax\n\t"
6025 		"cmpl	%%eax, %[size]\n\t"
6026 		"jne	.Lfill_cache\n\t"
6027 		"lfence\n"
6028 		:: [flush_pages] "r" (vmx_l1d_flush_pages),
6029 		    [size] "r" (size)
6030 		: "eax", "ebx", "ecx", "edx");
6031 }
6032 
6033 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
6034 {
6035 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6036 
6037 	if (is_guest_mode(vcpu) &&
6038 		nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
6039 		return;
6040 
6041 	if (irr == -1 || tpr < irr) {
6042 		vmcs_write32(TPR_THRESHOLD, 0);
6043 		return;
6044 	}
6045 
6046 	vmcs_write32(TPR_THRESHOLD, irr);
6047 }
6048 
6049 void vmx_set_virtual_apic_mode(struct kvm_vcpu *vcpu)
6050 {
6051 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6052 	u32 sec_exec_control;
6053 
6054 	if (!lapic_in_kernel(vcpu))
6055 		return;
6056 
6057 	if (!flexpriority_enabled &&
6058 	    !cpu_has_vmx_virtualize_x2apic_mode())
6059 		return;
6060 
6061 	/* Postpone execution until vmcs01 is the current VMCS. */
6062 	if (is_guest_mode(vcpu)) {
6063 		vmx->nested.change_vmcs01_virtual_apic_mode = true;
6064 		return;
6065 	}
6066 
6067 	sec_exec_control = secondary_exec_controls_get(vmx);
6068 	sec_exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
6069 			      SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE);
6070 
6071 	switch (kvm_get_apic_mode(vcpu)) {
6072 	case LAPIC_MODE_INVALID:
6073 		WARN_ONCE(true, "Invalid local APIC state");
6074 	case LAPIC_MODE_DISABLED:
6075 		break;
6076 	case LAPIC_MODE_XAPIC:
6077 		if (flexpriority_enabled) {
6078 			sec_exec_control |=
6079 				SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
6080 			vmx_flush_tlb(vcpu, true);
6081 		}
6082 		break;
6083 	case LAPIC_MODE_X2APIC:
6084 		if (cpu_has_vmx_virtualize_x2apic_mode())
6085 			sec_exec_control |=
6086 				SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
6087 		break;
6088 	}
6089 	secondary_exec_controls_set(vmx, sec_exec_control);
6090 
6091 	vmx_update_msr_bitmap(vcpu);
6092 }
6093 
6094 static void vmx_set_apic_access_page_addr(struct kvm_vcpu *vcpu, hpa_t hpa)
6095 {
6096 	if (!is_guest_mode(vcpu)) {
6097 		vmcs_write64(APIC_ACCESS_ADDR, hpa);
6098 		vmx_flush_tlb(vcpu, true);
6099 	}
6100 }
6101 
6102 static void vmx_hwapic_isr_update(struct kvm_vcpu *vcpu, int max_isr)
6103 {
6104 	u16 status;
6105 	u8 old;
6106 
6107 	if (max_isr == -1)
6108 		max_isr = 0;
6109 
6110 	status = vmcs_read16(GUEST_INTR_STATUS);
6111 	old = status >> 8;
6112 	if (max_isr != old) {
6113 		status &= 0xff;
6114 		status |= max_isr << 8;
6115 		vmcs_write16(GUEST_INTR_STATUS, status);
6116 	}
6117 }
6118 
6119 static void vmx_set_rvi(int vector)
6120 {
6121 	u16 status;
6122 	u8 old;
6123 
6124 	if (vector == -1)
6125 		vector = 0;
6126 
6127 	status = vmcs_read16(GUEST_INTR_STATUS);
6128 	old = (u8)status & 0xff;
6129 	if ((u8)vector != old) {
6130 		status &= ~0xff;
6131 		status |= (u8)vector;
6132 		vmcs_write16(GUEST_INTR_STATUS, status);
6133 	}
6134 }
6135 
6136 static void vmx_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr)
6137 {
6138 	/*
6139 	 * When running L2, updating RVI is only relevant when
6140 	 * vmcs12 virtual-interrupt-delivery enabled.
6141 	 * However, it can be enabled only when L1 also
6142 	 * intercepts external-interrupts and in that case
6143 	 * we should not update vmcs02 RVI but instead intercept
6144 	 * interrupt. Therefore, do nothing when running L2.
6145 	 */
6146 	if (!is_guest_mode(vcpu))
6147 		vmx_set_rvi(max_irr);
6148 }
6149 
6150 static int vmx_sync_pir_to_irr(struct kvm_vcpu *vcpu)
6151 {
6152 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6153 	int max_irr;
6154 	bool max_irr_updated;
6155 
6156 	WARN_ON(!vcpu->arch.apicv_active);
6157 	if (pi_test_on(&vmx->pi_desc)) {
6158 		pi_clear_on(&vmx->pi_desc);
6159 		/*
6160 		 * IOMMU can write to PIR.ON, so the barrier matters even on UP.
6161 		 * But on x86 this is just a compiler barrier anyway.
6162 		 */
6163 		smp_mb__after_atomic();
6164 		max_irr_updated =
6165 			kvm_apic_update_irr(vcpu, vmx->pi_desc.pir, &max_irr);
6166 
6167 		/*
6168 		 * If we are running L2 and L1 has a new pending interrupt
6169 		 * which can be injected, we should re-evaluate
6170 		 * what should be done with this new L1 interrupt.
6171 		 * If L1 intercepts external-interrupts, we should
6172 		 * exit from L2 to L1. Otherwise, interrupt should be
6173 		 * delivered directly to L2.
6174 		 */
6175 		if (is_guest_mode(vcpu) && max_irr_updated) {
6176 			if (nested_exit_on_intr(vcpu))
6177 				kvm_vcpu_exiting_guest_mode(vcpu);
6178 			else
6179 				kvm_make_request(KVM_REQ_EVENT, vcpu);
6180 		}
6181 	} else {
6182 		max_irr = kvm_lapic_find_highest_irr(vcpu);
6183 	}
6184 	vmx_hwapic_irr_update(vcpu, max_irr);
6185 	return max_irr;
6186 }
6187 
6188 static bool vmx_dy_apicv_has_pending_interrupt(struct kvm_vcpu *vcpu)
6189 {
6190 	return pi_test_on(vcpu_to_pi_desc(vcpu));
6191 }
6192 
6193 static void vmx_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap)
6194 {
6195 	if (!kvm_vcpu_apicv_active(vcpu))
6196 		return;
6197 
6198 	vmcs_write64(EOI_EXIT_BITMAP0, eoi_exit_bitmap[0]);
6199 	vmcs_write64(EOI_EXIT_BITMAP1, eoi_exit_bitmap[1]);
6200 	vmcs_write64(EOI_EXIT_BITMAP2, eoi_exit_bitmap[2]);
6201 	vmcs_write64(EOI_EXIT_BITMAP3, eoi_exit_bitmap[3]);
6202 }
6203 
6204 static void vmx_apicv_post_state_restore(struct kvm_vcpu *vcpu)
6205 {
6206 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6207 
6208 	pi_clear_on(&vmx->pi_desc);
6209 	memset(vmx->pi_desc.pir, 0, sizeof(vmx->pi_desc.pir));
6210 }
6211 
6212 static void handle_exception_nmi_irqoff(struct vcpu_vmx *vmx)
6213 {
6214 	vmx->exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6215 
6216 	/* if exit due to PF check for async PF */
6217 	if (is_page_fault(vmx->exit_intr_info))
6218 		vmx->vcpu.arch.apf.host_apf_reason = kvm_read_and_reset_pf_reason();
6219 
6220 	/* Handle machine checks before interrupts are enabled */
6221 	if (is_machine_check(vmx->exit_intr_info))
6222 		kvm_machine_check();
6223 
6224 	/* We need to handle NMIs before interrupts are enabled */
6225 	if (is_nmi(vmx->exit_intr_info)) {
6226 		kvm_before_interrupt(&vmx->vcpu);
6227 		asm("int $2");
6228 		kvm_after_interrupt(&vmx->vcpu);
6229 	}
6230 }
6231 
6232 static void handle_external_interrupt_irqoff(struct kvm_vcpu *vcpu)
6233 {
6234 	unsigned int vector;
6235 	unsigned long entry;
6236 #ifdef CONFIG_X86_64
6237 	unsigned long tmp;
6238 #endif
6239 	gate_desc *desc;
6240 	u32 intr_info;
6241 
6242 	intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6243 	if (WARN_ONCE(!is_external_intr(intr_info),
6244 	    "KVM: unexpected VM-Exit interrupt info: 0x%x", intr_info))
6245 		return;
6246 
6247 	vector = intr_info & INTR_INFO_VECTOR_MASK;
6248 	desc = (gate_desc *)host_idt_base + vector;
6249 	entry = gate_offset(desc);
6250 
6251 	kvm_before_interrupt(vcpu);
6252 
6253 	asm volatile(
6254 #ifdef CONFIG_X86_64
6255 		"mov %%" _ASM_SP ", %[sp]\n\t"
6256 		"and $0xfffffffffffffff0, %%" _ASM_SP "\n\t"
6257 		"push $%c[ss]\n\t"
6258 		"push %[sp]\n\t"
6259 #endif
6260 		"pushf\n\t"
6261 		__ASM_SIZE(push) " $%c[cs]\n\t"
6262 		CALL_NOSPEC
6263 		:
6264 #ifdef CONFIG_X86_64
6265 		[sp]"=&r"(tmp),
6266 #endif
6267 		ASM_CALL_CONSTRAINT
6268 		:
6269 		THUNK_TARGET(entry),
6270 		[ss]"i"(__KERNEL_DS),
6271 		[cs]"i"(__KERNEL_CS)
6272 	);
6273 
6274 	kvm_after_interrupt(vcpu);
6275 }
6276 STACK_FRAME_NON_STANDARD(handle_external_interrupt_irqoff);
6277 
6278 static void vmx_handle_exit_irqoff(struct kvm_vcpu *vcpu)
6279 {
6280 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6281 
6282 	if (vmx->exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT)
6283 		handle_external_interrupt_irqoff(vcpu);
6284 	else if (vmx->exit_reason == EXIT_REASON_EXCEPTION_NMI)
6285 		handle_exception_nmi_irqoff(vmx);
6286 }
6287 
6288 static bool vmx_has_emulated_msr(int index)
6289 {
6290 	switch (index) {
6291 	case MSR_IA32_SMBASE:
6292 		/*
6293 		 * We cannot do SMM unless we can run the guest in big
6294 		 * real mode.
6295 		 */
6296 		return enable_unrestricted_guest || emulate_invalid_guest_state;
6297 	case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
6298 		return nested;
6299 	case MSR_AMD64_VIRT_SPEC_CTRL:
6300 		/* This is AMD only.  */
6301 		return false;
6302 	default:
6303 		return true;
6304 	}
6305 }
6306 
6307 static bool vmx_pt_supported(void)
6308 {
6309 	return pt_mode == PT_MODE_HOST_GUEST;
6310 }
6311 
6312 static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx)
6313 {
6314 	u32 exit_intr_info;
6315 	bool unblock_nmi;
6316 	u8 vector;
6317 	bool idtv_info_valid;
6318 
6319 	idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK;
6320 
6321 	if (enable_vnmi) {
6322 		if (vmx->loaded_vmcs->nmi_known_unmasked)
6323 			return;
6324 		/*
6325 		 * Can't use vmx->exit_intr_info since we're not sure what
6326 		 * the exit reason is.
6327 		 */
6328 		exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6329 		unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
6330 		vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
6331 		/*
6332 		 * SDM 3: 27.7.1.2 (September 2008)
6333 		 * Re-set bit "block by NMI" before VM entry if vmexit caused by
6334 		 * a guest IRET fault.
6335 		 * SDM 3: 23.2.2 (September 2008)
6336 		 * Bit 12 is undefined in any of the following cases:
6337 		 *  If the VM exit sets the valid bit in the IDT-vectoring
6338 		 *   information field.
6339 		 *  If the VM exit is due to a double fault.
6340 		 */
6341 		if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
6342 		    vector != DF_VECTOR && !idtv_info_valid)
6343 			vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
6344 				      GUEST_INTR_STATE_NMI);
6345 		else
6346 			vmx->loaded_vmcs->nmi_known_unmasked =
6347 				!(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO)
6348 				  & GUEST_INTR_STATE_NMI);
6349 	} else if (unlikely(vmx->loaded_vmcs->soft_vnmi_blocked))
6350 		vmx->loaded_vmcs->vnmi_blocked_time +=
6351 			ktime_to_ns(ktime_sub(ktime_get(),
6352 					      vmx->loaded_vmcs->entry_time));
6353 }
6354 
6355 static void __vmx_complete_interrupts(struct kvm_vcpu *vcpu,
6356 				      u32 idt_vectoring_info,
6357 				      int instr_len_field,
6358 				      int error_code_field)
6359 {
6360 	u8 vector;
6361 	int type;
6362 	bool idtv_info_valid;
6363 
6364 	idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
6365 
6366 	vcpu->arch.nmi_injected = false;
6367 	kvm_clear_exception_queue(vcpu);
6368 	kvm_clear_interrupt_queue(vcpu);
6369 
6370 	if (!idtv_info_valid)
6371 		return;
6372 
6373 	kvm_make_request(KVM_REQ_EVENT, vcpu);
6374 
6375 	vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
6376 	type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
6377 
6378 	switch (type) {
6379 	case INTR_TYPE_NMI_INTR:
6380 		vcpu->arch.nmi_injected = true;
6381 		/*
6382 		 * SDM 3: 27.7.1.2 (September 2008)
6383 		 * Clear bit "block by NMI" before VM entry if a NMI
6384 		 * delivery faulted.
6385 		 */
6386 		vmx_set_nmi_mask(vcpu, false);
6387 		break;
6388 	case INTR_TYPE_SOFT_EXCEPTION:
6389 		vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
6390 		/* fall through */
6391 	case INTR_TYPE_HARD_EXCEPTION:
6392 		if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
6393 			u32 err = vmcs_read32(error_code_field);
6394 			kvm_requeue_exception_e(vcpu, vector, err);
6395 		} else
6396 			kvm_requeue_exception(vcpu, vector);
6397 		break;
6398 	case INTR_TYPE_SOFT_INTR:
6399 		vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
6400 		/* fall through */
6401 	case INTR_TYPE_EXT_INTR:
6402 		kvm_queue_interrupt(vcpu, vector, type == INTR_TYPE_SOFT_INTR);
6403 		break;
6404 	default:
6405 		break;
6406 	}
6407 }
6408 
6409 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
6410 {
6411 	__vmx_complete_interrupts(&vmx->vcpu, vmx->idt_vectoring_info,
6412 				  VM_EXIT_INSTRUCTION_LEN,
6413 				  IDT_VECTORING_ERROR_CODE);
6414 }
6415 
6416 static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
6417 {
6418 	__vmx_complete_interrupts(vcpu,
6419 				  vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
6420 				  VM_ENTRY_INSTRUCTION_LEN,
6421 				  VM_ENTRY_EXCEPTION_ERROR_CODE);
6422 
6423 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
6424 }
6425 
6426 static void atomic_switch_perf_msrs(struct vcpu_vmx *vmx)
6427 {
6428 	int i, nr_msrs;
6429 	struct perf_guest_switch_msr *msrs;
6430 
6431 	msrs = perf_guest_get_msrs(&nr_msrs);
6432 
6433 	if (!msrs)
6434 		return;
6435 
6436 	for (i = 0; i < nr_msrs; i++)
6437 		if (msrs[i].host == msrs[i].guest)
6438 			clear_atomic_switch_msr(vmx, msrs[i].msr);
6439 		else
6440 			add_atomic_switch_msr(vmx, msrs[i].msr, msrs[i].guest,
6441 					msrs[i].host, false);
6442 }
6443 
6444 static void atomic_switch_umwait_control_msr(struct vcpu_vmx *vmx)
6445 {
6446 	u32 host_umwait_control;
6447 
6448 	if (!vmx_has_waitpkg(vmx))
6449 		return;
6450 
6451 	host_umwait_control = get_umwait_control_msr();
6452 
6453 	if (vmx->msr_ia32_umwait_control != host_umwait_control)
6454 		add_atomic_switch_msr(vmx, MSR_IA32_UMWAIT_CONTROL,
6455 			vmx->msr_ia32_umwait_control,
6456 			host_umwait_control, false);
6457 	else
6458 		clear_atomic_switch_msr(vmx, MSR_IA32_UMWAIT_CONTROL);
6459 }
6460 
6461 static void vmx_update_hv_timer(struct kvm_vcpu *vcpu)
6462 {
6463 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6464 	u64 tscl;
6465 	u32 delta_tsc;
6466 
6467 	if (vmx->req_immediate_exit) {
6468 		vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, 0);
6469 		vmx->loaded_vmcs->hv_timer_soft_disabled = false;
6470 	} else if (vmx->hv_deadline_tsc != -1) {
6471 		tscl = rdtsc();
6472 		if (vmx->hv_deadline_tsc > tscl)
6473 			/* set_hv_timer ensures the delta fits in 32-bits */
6474 			delta_tsc = (u32)((vmx->hv_deadline_tsc - tscl) >>
6475 				cpu_preemption_timer_multi);
6476 		else
6477 			delta_tsc = 0;
6478 
6479 		vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, delta_tsc);
6480 		vmx->loaded_vmcs->hv_timer_soft_disabled = false;
6481 	} else if (!vmx->loaded_vmcs->hv_timer_soft_disabled) {
6482 		vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, -1);
6483 		vmx->loaded_vmcs->hv_timer_soft_disabled = true;
6484 	}
6485 }
6486 
6487 void vmx_update_host_rsp(struct vcpu_vmx *vmx, unsigned long host_rsp)
6488 {
6489 	if (unlikely(host_rsp != vmx->loaded_vmcs->host_state.rsp)) {
6490 		vmx->loaded_vmcs->host_state.rsp = host_rsp;
6491 		vmcs_writel(HOST_RSP, host_rsp);
6492 	}
6493 }
6494 
6495 bool __vmx_vcpu_run(struct vcpu_vmx *vmx, unsigned long *regs, bool launched);
6496 
6497 static void vmx_vcpu_run(struct kvm_vcpu *vcpu)
6498 {
6499 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6500 	unsigned long cr3, cr4;
6501 
6502 	/* Record the guest's net vcpu time for enforced NMI injections. */
6503 	if (unlikely(!enable_vnmi &&
6504 		     vmx->loaded_vmcs->soft_vnmi_blocked))
6505 		vmx->loaded_vmcs->entry_time = ktime_get();
6506 
6507 	/* Don't enter VMX if guest state is invalid, let the exit handler
6508 	   start emulation until we arrive back to a valid state */
6509 	if (vmx->emulation_required)
6510 		return;
6511 
6512 	if (vmx->ple_window_dirty) {
6513 		vmx->ple_window_dirty = false;
6514 		vmcs_write32(PLE_WINDOW, vmx->ple_window);
6515 	}
6516 
6517 	if (vmx->nested.need_vmcs12_to_shadow_sync)
6518 		nested_sync_vmcs12_to_shadow(vcpu);
6519 
6520 	if (test_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_dirty))
6521 		vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
6522 	if (test_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_dirty))
6523 		vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
6524 
6525 	cr3 = __get_current_cr3_fast();
6526 	if (unlikely(cr3 != vmx->loaded_vmcs->host_state.cr3)) {
6527 		vmcs_writel(HOST_CR3, cr3);
6528 		vmx->loaded_vmcs->host_state.cr3 = cr3;
6529 	}
6530 
6531 	cr4 = cr4_read_shadow();
6532 	if (unlikely(cr4 != vmx->loaded_vmcs->host_state.cr4)) {
6533 		vmcs_writel(HOST_CR4, cr4);
6534 		vmx->loaded_vmcs->host_state.cr4 = cr4;
6535 	}
6536 
6537 	/* When single-stepping over STI and MOV SS, we must clear the
6538 	 * corresponding interruptibility bits in the guest state. Otherwise
6539 	 * vmentry fails as it then expects bit 14 (BS) in pending debug
6540 	 * exceptions being set, but that's not correct for the guest debugging
6541 	 * case. */
6542 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6543 		vmx_set_interrupt_shadow(vcpu, 0);
6544 
6545 	kvm_load_guest_xcr0(vcpu);
6546 
6547 	if (static_cpu_has(X86_FEATURE_PKU) &&
6548 	    kvm_read_cr4_bits(vcpu, X86_CR4_PKE) &&
6549 	    vcpu->arch.pkru != vmx->host_pkru)
6550 		__write_pkru(vcpu->arch.pkru);
6551 
6552 	pt_guest_enter(vmx);
6553 
6554 	atomic_switch_perf_msrs(vmx);
6555 	atomic_switch_umwait_control_msr(vmx);
6556 
6557 	if (enable_preemption_timer)
6558 		vmx_update_hv_timer(vcpu);
6559 
6560 	if (lapic_in_kernel(vcpu) &&
6561 		vcpu->arch.apic->lapic_timer.timer_advance_ns)
6562 		kvm_wait_lapic_expire(vcpu);
6563 
6564 	/*
6565 	 * If this vCPU has touched SPEC_CTRL, restore the guest's value if
6566 	 * it's non-zero. Since vmentry is serialising on affected CPUs, there
6567 	 * is no need to worry about the conditional branch over the wrmsr
6568 	 * being speculatively taken.
6569 	 */
6570 	x86_spec_ctrl_set_guest(vmx->spec_ctrl, 0);
6571 
6572 	/* L1D Flush includes CPU buffer clear to mitigate MDS */
6573 	if (static_branch_unlikely(&vmx_l1d_should_flush))
6574 		vmx_l1d_flush(vcpu);
6575 	else if (static_branch_unlikely(&mds_user_clear))
6576 		mds_clear_cpu_buffers();
6577 
6578 	if (vcpu->arch.cr2 != read_cr2())
6579 		write_cr2(vcpu->arch.cr2);
6580 
6581 	vmx->fail = __vmx_vcpu_run(vmx, (unsigned long *)&vcpu->arch.regs,
6582 				   vmx->loaded_vmcs->launched);
6583 
6584 	vcpu->arch.cr2 = read_cr2();
6585 
6586 	/*
6587 	 * We do not use IBRS in the kernel. If this vCPU has used the
6588 	 * SPEC_CTRL MSR it may have left it on; save the value and
6589 	 * turn it off. This is much more efficient than blindly adding
6590 	 * it to the atomic save/restore list. Especially as the former
6591 	 * (Saving guest MSRs on vmexit) doesn't even exist in KVM.
6592 	 *
6593 	 * For non-nested case:
6594 	 * If the L01 MSR bitmap does not intercept the MSR, then we need to
6595 	 * save it.
6596 	 *
6597 	 * For nested case:
6598 	 * If the L02 MSR bitmap does not intercept the MSR, then we need to
6599 	 * save it.
6600 	 */
6601 	if (unlikely(!msr_write_intercepted(vcpu, MSR_IA32_SPEC_CTRL)))
6602 		vmx->spec_ctrl = native_read_msr(MSR_IA32_SPEC_CTRL);
6603 
6604 	x86_spec_ctrl_restore_host(vmx->spec_ctrl, 0);
6605 
6606 	/* All fields are clean at this point */
6607 	if (static_branch_unlikely(&enable_evmcs))
6608 		current_evmcs->hv_clean_fields |=
6609 			HV_VMX_ENLIGHTENED_CLEAN_FIELD_ALL;
6610 
6611 	if (static_branch_unlikely(&enable_evmcs))
6612 		current_evmcs->hv_vp_id = vcpu->arch.hyperv.vp_index;
6613 
6614 	/* MSR_IA32_DEBUGCTLMSR is zeroed on vmexit. Restore it if needed */
6615 	if (vmx->host_debugctlmsr)
6616 		update_debugctlmsr(vmx->host_debugctlmsr);
6617 
6618 #ifndef CONFIG_X86_64
6619 	/*
6620 	 * The sysexit path does not restore ds/es, so we must set them to
6621 	 * a reasonable value ourselves.
6622 	 *
6623 	 * We can't defer this to vmx_prepare_switch_to_host() since that
6624 	 * function may be executed in interrupt context, which saves and
6625 	 * restore segments around it, nullifying its effect.
6626 	 */
6627 	loadsegment(ds, __USER_DS);
6628 	loadsegment(es, __USER_DS);
6629 #endif
6630 
6631 	vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
6632 				  | (1 << VCPU_EXREG_RFLAGS)
6633 				  | (1 << VCPU_EXREG_PDPTR)
6634 				  | (1 << VCPU_EXREG_SEGMENTS)
6635 				  | (1 << VCPU_EXREG_CR3));
6636 	vcpu->arch.regs_dirty = 0;
6637 
6638 	pt_guest_exit(vmx);
6639 
6640 	/*
6641 	 * eager fpu is enabled if PKEY is supported and CR4 is switched
6642 	 * back on host, so it is safe to read guest PKRU from current
6643 	 * XSAVE.
6644 	 */
6645 	if (static_cpu_has(X86_FEATURE_PKU) &&
6646 	    kvm_read_cr4_bits(vcpu, X86_CR4_PKE)) {
6647 		vcpu->arch.pkru = rdpkru();
6648 		if (vcpu->arch.pkru != vmx->host_pkru)
6649 			__write_pkru(vmx->host_pkru);
6650 	}
6651 
6652 	kvm_put_guest_xcr0(vcpu);
6653 
6654 	vmx->nested.nested_run_pending = 0;
6655 	vmx->idt_vectoring_info = 0;
6656 
6657 	vmx->exit_reason = vmx->fail ? 0xdead : vmcs_read32(VM_EXIT_REASON);
6658 	if ((u16)vmx->exit_reason == EXIT_REASON_MCE_DURING_VMENTRY)
6659 		kvm_machine_check();
6660 
6661 	if (vmx->fail || (vmx->exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY))
6662 		return;
6663 
6664 	vmx->loaded_vmcs->launched = 1;
6665 	vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
6666 
6667 	vmx_recover_nmi_blocking(vmx);
6668 	vmx_complete_interrupts(vmx);
6669 }
6670 
6671 static struct kvm *vmx_vm_alloc(void)
6672 {
6673 	struct kvm_vmx *kvm_vmx = __vmalloc(sizeof(struct kvm_vmx),
6674 					    GFP_KERNEL_ACCOUNT | __GFP_ZERO,
6675 					    PAGE_KERNEL);
6676 	return &kvm_vmx->kvm;
6677 }
6678 
6679 static void vmx_vm_free(struct kvm *kvm)
6680 {
6681 	kfree(kvm->arch.hyperv.hv_pa_pg);
6682 	vfree(to_kvm_vmx(kvm));
6683 }
6684 
6685 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
6686 {
6687 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6688 
6689 	if (enable_pml)
6690 		vmx_destroy_pml_buffer(vmx);
6691 	free_vpid(vmx->vpid);
6692 	nested_vmx_free_vcpu(vcpu);
6693 	free_loaded_vmcs(vmx->loaded_vmcs);
6694 	kfree(vmx->guest_msrs);
6695 	kvm_vcpu_uninit(vcpu);
6696 	kmem_cache_free(x86_fpu_cache, vmx->vcpu.arch.user_fpu);
6697 	kmem_cache_free(x86_fpu_cache, vmx->vcpu.arch.guest_fpu);
6698 	kmem_cache_free(kvm_vcpu_cache, vmx);
6699 }
6700 
6701 static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
6702 {
6703 	int err;
6704 	struct vcpu_vmx *vmx;
6705 	unsigned long *msr_bitmap;
6706 	int cpu;
6707 
6708 	BUILD_BUG_ON_MSG(offsetof(struct vcpu_vmx, vcpu) != 0,
6709 		"struct kvm_vcpu must be at offset 0 for arch usercopy region");
6710 
6711 	vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL_ACCOUNT);
6712 	if (!vmx)
6713 		return ERR_PTR(-ENOMEM);
6714 
6715 	vmx->vcpu.arch.user_fpu = kmem_cache_zalloc(x86_fpu_cache,
6716 			GFP_KERNEL_ACCOUNT);
6717 	if (!vmx->vcpu.arch.user_fpu) {
6718 		printk(KERN_ERR "kvm: failed to allocate kvm userspace's fpu\n");
6719 		err = -ENOMEM;
6720 		goto free_partial_vcpu;
6721 	}
6722 
6723 	vmx->vcpu.arch.guest_fpu = kmem_cache_zalloc(x86_fpu_cache,
6724 			GFP_KERNEL_ACCOUNT);
6725 	if (!vmx->vcpu.arch.guest_fpu) {
6726 		printk(KERN_ERR "kvm: failed to allocate vcpu's fpu\n");
6727 		err = -ENOMEM;
6728 		goto free_user_fpu;
6729 	}
6730 
6731 	vmx->vpid = allocate_vpid();
6732 
6733 	err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
6734 	if (err)
6735 		goto free_vcpu;
6736 
6737 	err = -ENOMEM;
6738 
6739 	/*
6740 	 * If PML is turned on, failure on enabling PML just results in failure
6741 	 * of creating the vcpu, therefore we can simplify PML logic (by
6742 	 * avoiding dealing with cases, such as enabling PML partially on vcpus
6743 	 * for the guest, etc.
6744 	 */
6745 	if (enable_pml) {
6746 		vmx->pml_pg = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
6747 		if (!vmx->pml_pg)
6748 			goto uninit_vcpu;
6749 	}
6750 
6751 	vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL_ACCOUNT);
6752 	BUILD_BUG_ON(ARRAY_SIZE(vmx_msr_index) * sizeof(vmx->guest_msrs[0])
6753 		     > PAGE_SIZE);
6754 
6755 	if (!vmx->guest_msrs)
6756 		goto free_pml;
6757 
6758 	err = alloc_loaded_vmcs(&vmx->vmcs01);
6759 	if (err < 0)
6760 		goto free_msrs;
6761 
6762 	msr_bitmap = vmx->vmcs01.msr_bitmap;
6763 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_TSC, MSR_TYPE_R);
6764 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_FS_BASE, MSR_TYPE_RW);
6765 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_GS_BASE, MSR_TYPE_RW);
6766 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_KERNEL_GS_BASE, MSR_TYPE_RW);
6767 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_CS, MSR_TYPE_RW);
6768 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_ESP, MSR_TYPE_RW);
6769 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_EIP, MSR_TYPE_RW);
6770 	if (kvm_cstate_in_guest(kvm)) {
6771 		vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C1_RES, MSR_TYPE_R);
6772 		vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C3_RESIDENCY, MSR_TYPE_R);
6773 		vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C6_RESIDENCY, MSR_TYPE_R);
6774 		vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C7_RESIDENCY, MSR_TYPE_R);
6775 	}
6776 	vmx->msr_bitmap_mode = 0;
6777 
6778 	vmx->loaded_vmcs = &vmx->vmcs01;
6779 	cpu = get_cpu();
6780 	vmx_vcpu_load(&vmx->vcpu, cpu);
6781 	vmx->vcpu.cpu = cpu;
6782 	vmx_vcpu_setup(vmx);
6783 	vmx_vcpu_put(&vmx->vcpu);
6784 	put_cpu();
6785 	if (cpu_need_virtualize_apic_accesses(&vmx->vcpu)) {
6786 		err = alloc_apic_access_page(kvm);
6787 		if (err)
6788 			goto free_vmcs;
6789 	}
6790 
6791 	if (enable_ept && !enable_unrestricted_guest) {
6792 		err = init_rmode_identity_map(kvm);
6793 		if (err)
6794 			goto free_vmcs;
6795 	}
6796 
6797 	if (nested)
6798 		nested_vmx_setup_ctls_msrs(&vmx->nested.msrs,
6799 					   vmx_capability.ept,
6800 					   kvm_vcpu_apicv_active(&vmx->vcpu));
6801 	else
6802 		memset(&vmx->nested.msrs, 0, sizeof(vmx->nested.msrs));
6803 
6804 	vmx->nested.posted_intr_nv = -1;
6805 	vmx->nested.current_vmptr = -1ull;
6806 
6807 	vmx->msr_ia32_feature_control_valid_bits = FEATURE_CONTROL_LOCKED;
6808 
6809 	/*
6810 	 * Enforce invariant: pi_desc.nv is always either POSTED_INTR_VECTOR
6811 	 * or POSTED_INTR_WAKEUP_VECTOR.
6812 	 */
6813 	vmx->pi_desc.nv = POSTED_INTR_VECTOR;
6814 	vmx->pi_desc.sn = 1;
6815 
6816 	vmx->ept_pointer = INVALID_PAGE;
6817 
6818 	return &vmx->vcpu;
6819 
6820 free_vmcs:
6821 	free_loaded_vmcs(vmx->loaded_vmcs);
6822 free_msrs:
6823 	kfree(vmx->guest_msrs);
6824 free_pml:
6825 	vmx_destroy_pml_buffer(vmx);
6826 uninit_vcpu:
6827 	kvm_vcpu_uninit(&vmx->vcpu);
6828 free_vcpu:
6829 	free_vpid(vmx->vpid);
6830 	kmem_cache_free(x86_fpu_cache, vmx->vcpu.arch.guest_fpu);
6831 free_user_fpu:
6832 	kmem_cache_free(x86_fpu_cache, vmx->vcpu.arch.user_fpu);
6833 free_partial_vcpu:
6834 	kmem_cache_free(kvm_vcpu_cache, vmx);
6835 	return ERR_PTR(err);
6836 }
6837 
6838 #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"
6839 #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"
6840 
6841 static int vmx_vm_init(struct kvm *kvm)
6842 {
6843 	spin_lock_init(&to_kvm_vmx(kvm)->ept_pointer_lock);
6844 
6845 	if (!ple_gap)
6846 		kvm->arch.pause_in_guest = true;
6847 
6848 	if (boot_cpu_has(X86_BUG_L1TF) && enable_ept) {
6849 		switch (l1tf_mitigation) {
6850 		case L1TF_MITIGATION_OFF:
6851 		case L1TF_MITIGATION_FLUSH_NOWARN:
6852 			/* 'I explicitly don't care' is set */
6853 			break;
6854 		case L1TF_MITIGATION_FLUSH:
6855 		case L1TF_MITIGATION_FLUSH_NOSMT:
6856 		case L1TF_MITIGATION_FULL:
6857 			/*
6858 			 * Warn upon starting the first VM in a potentially
6859 			 * insecure environment.
6860 			 */
6861 			if (sched_smt_active())
6862 				pr_warn_once(L1TF_MSG_SMT);
6863 			if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_NEVER)
6864 				pr_warn_once(L1TF_MSG_L1D);
6865 			break;
6866 		case L1TF_MITIGATION_FULL_FORCE:
6867 			/* Flush is enforced */
6868 			break;
6869 		}
6870 	}
6871 	return 0;
6872 }
6873 
6874 static int __init vmx_check_processor_compat(void)
6875 {
6876 	struct vmcs_config vmcs_conf;
6877 	struct vmx_capability vmx_cap;
6878 
6879 	if (setup_vmcs_config(&vmcs_conf, &vmx_cap) < 0)
6880 		return -EIO;
6881 	if (nested)
6882 		nested_vmx_setup_ctls_msrs(&vmcs_conf.nested, vmx_cap.ept,
6883 					   enable_apicv);
6884 	if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
6885 		printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
6886 				smp_processor_id());
6887 		return -EIO;
6888 	}
6889 	return 0;
6890 }
6891 
6892 static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
6893 {
6894 	u8 cache;
6895 	u64 ipat = 0;
6896 
6897 	/* For VT-d and EPT combination
6898 	 * 1. MMIO: always map as UC
6899 	 * 2. EPT with VT-d:
6900 	 *   a. VT-d without snooping control feature: can't guarantee the
6901 	 *	result, try to trust guest.
6902 	 *   b. VT-d with snooping control feature: snooping control feature of
6903 	 *	VT-d engine can guarantee the cache correctness. Just set it
6904 	 *	to WB to keep consistent with host. So the same as item 3.
6905 	 * 3. EPT without VT-d: always map as WB and set IPAT=1 to keep
6906 	 *    consistent with host MTRR
6907 	 */
6908 	if (is_mmio) {
6909 		cache = MTRR_TYPE_UNCACHABLE;
6910 		goto exit;
6911 	}
6912 
6913 	if (!kvm_arch_has_noncoherent_dma(vcpu->kvm)) {
6914 		ipat = VMX_EPT_IPAT_BIT;
6915 		cache = MTRR_TYPE_WRBACK;
6916 		goto exit;
6917 	}
6918 
6919 	if (kvm_read_cr0(vcpu) & X86_CR0_CD) {
6920 		ipat = VMX_EPT_IPAT_BIT;
6921 		if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
6922 			cache = MTRR_TYPE_WRBACK;
6923 		else
6924 			cache = MTRR_TYPE_UNCACHABLE;
6925 		goto exit;
6926 	}
6927 
6928 	cache = kvm_mtrr_get_guest_memory_type(vcpu, gfn);
6929 
6930 exit:
6931 	return (cache << VMX_EPT_MT_EPTE_SHIFT) | ipat;
6932 }
6933 
6934 static int vmx_get_lpage_level(void)
6935 {
6936 	if (enable_ept && !cpu_has_vmx_ept_1g_page())
6937 		return PT_DIRECTORY_LEVEL;
6938 	else
6939 		/* For shadow and EPT supported 1GB page */
6940 		return PT_PDPE_LEVEL;
6941 }
6942 
6943 static void vmcs_set_secondary_exec_control(struct vcpu_vmx *vmx)
6944 {
6945 	/*
6946 	 * These bits in the secondary execution controls field
6947 	 * are dynamic, the others are mostly based on the hypervisor
6948 	 * architecture and the guest's CPUID.  Do not touch the
6949 	 * dynamic bits.
6950 	 */
6951 	u32 mask =
6952 		SECONDARY_EXEC_SHADOW_VMCS |
6953 		SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
6954 		SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
6955 		SECONDARY_EXEC_DESC;
6956 
6957 	u32 new_ctl = vmx->secondary_exec_control;
6958 	u32 cur_ctl = secondary_exec_controls_get(vmx);
6959 
6960 	secondary_exec_controls_set(vmx, (new_ctl & ~mask) | (cur_ctl & mask));
6961 }
6962 
6963 /*
6964  * Generate MSR_IA32_VMX_CR{0,4}_FIXED1 according to CPUID. Only set bits
6965  * (indicating "allowed-1") if they are supported in the guest's CPUID.
6966  */
6967 static void nested_vmx_cr_fixed1_bits_update(struct kvm_vcpu *vcpu)
6968 {
6969 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6970 	struct kvm_cpuid_entry2 *entry;
6971 
6972 	vmx->nested.msrs.cr0_fixed1 = 0xffffffff;
6973 	vmx->nested.msrs.cr4_fixed1 = X86_CR4_PCE;
6974 
6975 #define cr4_fixed1_update(_cr4_mask, _reg, _cpuid_mask) do {		\
6976 	if (entry && (entry->_reg & (_cpuid_mask)))			\
6977 		vmx->nested.msrs.cr4_fixed1 |= (_cr4_mask);	\
6978 } while (0)
6979 
6980 	entry = kvm_find_cpuid_entry(vcpu, 0x1, 0);
6981 	cr4_fixed1_update(X86_CR4_VME,        edx, bit(X86_FEATURE_VME));
6982 	cr4_fixed1_update(X86_CR4_PVI,        edx, bit(X86_FEATURE_VME));
6983 	cr4_fixed1_update(X86_CR4_TSD,        edx, bit(X86_FEATURE_TSC));
6984 	cr4_fixed1_update(X86_CR4_DE,         edx, bit(X86_FEATURE_DE));
6985 	cr4_fixed1_update(X86_CR4_PSE,        edx, bit(X86_FEATURE_PSE));
6986 	cr4_fixed1_update(X86_CR4_PAE,        edx, bit(X86_FEATURE_PAE));
6987 	cr4_fixed1_update(X86_CR4_MCE,        edx, bit(X86_FEATURE_MCE));
6988 	cr4_fixed1_update(X86_CR4_PGE,        edx, bit(X86_FEATURE_PGE));
6989 	cr4_fixed1_update(X86_CR4_OSFXSR,     edx, bit(X86_FEATURE_FXSR));
6990 	cr4_fixed1_update(X86_CR4_OSXMMEXCPT, edx, bit(X86_FEATURE_XMM));
6991 	cr4_fixed1_update(X86_CR4_VMXE,       ecx, bit(X86_FEATURE_VMX));
6992 	cr4_fixed1_update(X86_CR4_SMXE,       ecx, bit(X86_FEATURE_SMX));
6993 	cr4_fixed1_update(X86_CR4_PCIDE,      ecx, bit(X86_FEATURE_PCID));
6994 	cr4_fixed1_update(X86_CR4_OSXSAVE,    ecx, bit(X86_FEATURE_XSAVE));
6995 
6996 	entry = kvm_find_cpuid_entry(vcpu, 0x7, 0);
6997 	cr4_fixed1_update(X86_CR4_FSGSBASE,   ebx, bit(X86_FEATURE_FSGSBASE));
6998 	cr4_fixed1_update(X86_CR4_SMEP,       ebx, bit(X86_FEATURE_SMEP));
6999 	cr4_fixed1_update(X86_CR4_SMAP,       ebx, bit(X86_FEATURE_SMAP));
7000 	cr4_fixed1_update(X86_CR4_PKE,        ecx, bit(X86_FEATURE_PKU));
7001 	cr4_fixed1_update(X86_CR4_UMIP,       ecx, bit(X86_FEATURE_UMIP));
7002 
7003 #undef cr4_fixed1_update
7004 }
7005 
7006 static void nested_vmx_entry_exit_ctls_update(struct kvm_vcpu *vcpu)
7007 {
7008 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7009 
7010 	if (kvm_mpx_supported()) {
7011 		bool mpx_enabled = guest_cpuid_has(vcpu, X86_FEATURE_MPX);
7012 
7013 		if (mpx_enabled) {
7014 			vmx->nested.msrs.entry_ctls_high |= VM_ENTRY_LOAD_BNDCFGS;
7015 			vmx->nested.msrs.exit_ctls_high |= VM_EXIT_CLEAR_BNDCFGS;
7016 		} else {
7017 			vmx->nested.msrs.entry_ctls_high &= ~VM_ENTRY_LOAD_BNDCFGS;
7018 			vmx->nested.msrs.exit_ctls_high &= ~VM_EXIT_CLEAR_BNDCFGS;
7019 		}
7020 	}
7021 }
7022 
7023 static void update_intel_pt_cfg(struct kvm_vcpu *vcpu)
7024 {
7025 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7026 	struct kvm_cpuid_entry2 *best = NULL;
7027 	int i;
7028 
7029 	for (i = 0; i < PT_CPUID_LEAVES; i++) {
7030 		best = kvm_find_cpuid_entry(vcpu, 0x14, i);
7031 		if (!best)
7032 			return;
7033 		vmx->pt_desc.caps[CPUID_EAX + i*PT_CPUID_REGS_NUM] = best->eax;
7034 		vmx->pt_desc.caps[CPUID_EBX + i*PT_CPUID_REGS_NUM] = best->ebx;
7035 		vmx->pt_desc.caps[CPUID_ECX + i*PT_CPUID_REGS_NUM] = best->ecx;
7036 		vmx->pt_desc.caps[CPUID_EDX + i*PT_CPUID_REGS_NUM] = best->edx;
7037 	}
7038 
7039 	/* Get the number of configurable Address Ranges for filtering */
7040 	vmx->pt_desc.addr_range = intel_pt_validate_cap(vmx->pt_desc.caps,
7041 						PT_CAP_num_address_ranges);
7042 
7043 	/* Initialize and clear the no dependency bits */
7044 	vmx->pt_desc.ctl_bitmask = ~(RTIT_CTL_TRACEEN | RTIT_CTL_OS |
7045 			RTIT_CTL_USR | RTIT_CTL_TSC_EN | RTIT_CTL_DISRETC);
7046 
7047 	/*
7048 	 * If CPUID.(EAX=14H,ECX=0):EBX[0]=1 CR3Filter can be set otherwise
7049 	 * will inject an #GP
7050 	 */
7051 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_cr3_filtering))
7052 		vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_CR3EN;
7053 
7054 	/*
7055 	 * If CPUID.(EAX=14H,ECX=0):EBX[1]=1 CYCEn, CycThresh and
7056 	 * PSBFreq can be set
7057 	 */
7058 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc))
7059 		vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_CYCLEACC |
7060 				RTIT_CTL_CYC_THRESH | RTIT_CTL_PSB_FREQ);
7061 
7062 	/*
7063 	 * If CPUID.(EAX=14H,ECX=0):EBX[3]=1 MTCEn BranchEn and
7064 	 * MTCFreq can be set
7065 	 */
7066 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc))
7067 		vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_MTC_EN |
7068 				RTIT_CTL_BRANCH_EN | RTIT_CTL_MTC_RANGE);
7069 
7070 	/* If CPUID.(EAX=14H,ECX=0):EBX[4]=1 FUPonPTW and PTWEn can be set */
7071 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_ptwrite))
7072 		vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_FUP_ON_PTW |
7073 							RTIT_CTL_PTW_EN);
7074 
7075 	/* If CPUID.(EAX=14H,ECX=0):EBX[5]=1 PwrEvEn can be set */
7076 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_power_event_trace))
7077 		vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_PWR_EVT_EN;
7078 
7079 	/* If CPUID.(EAX=14H,ECX=0):ECX[0]=1 ToPA can be set */
7080 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_topa_output))
7081 		vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_TOPA;
7082 
7083 	/* If CPUID.(EAX=14H,ECX=0):ECX[3]=1 FabircEn can be set */
7084 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_output_subsys))
7085 		vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_FABRIC_EN;
7086 
7087 	/* unmask address range configure area */
7088 	for (i = 0; i < vmx->pt_desc.addr_range; i++)
7089 		vmx->pt_desc.ctl_bitmask &= ~(0xfULL << (32 + i * 4));
7090 }
7091 
7092 static void vmx_cpuid_update(struct kvm_vcpu *vcpu)
7093 {
7094 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7095 
7096 	if (cpu_has_secondary_exec_ctrls()) {
7097 		vmx_compute_secondary_exec_control(vmx);
7098 		vmcs_set_secondary_exec_control(vmx);
7099 	}
7100 
7101 	if (nested_vmx_allowed(vcpu))
7102 		to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
7103 			FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
7104 	else
7105 		to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
7106 			~FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
7107 
7108 	if (nested_vmx_allowed(vcpu)) {
7109 		nested_vmx_cr_fixed1_bits_update(vcpu);
7110 		nested_vmx_entry_exit_ctls_update(vcpu);
7111 	}
7112 
7113 	if (boot_cpu_has(X86_FEATURE_INTEL_PT) &&
7114 			guest_cpuid_has(vcpu, X86_FEATURE_INTEL_PT))
7115 		update_intel_pt_cfg(vcpu);
7116 }
7117 
7118 static void vmx_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
7119 {
7120 	if (func == 1 && nested)
7121 		entry->ecx |= bit(X86_FEATURE_VMX);
7122 }
7123 
7124 static void vmx_request_immediate_exit(struct kvm_vcpu *vcpu)
7125 {
7126 	to_vmx(vcpu)->req_immediate_exit = true;
7127 }
7128 
7129 static int vmx_check_intercept(struct kvm_vcpu *vcpu,
7130 			       struct x86_instruction_info *info,
7131 			       enum x86_intercept_stage stage)
7132 {
7133 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
7134 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
7135 
7136 	/*
7137 	 * RDPID causes #UD if disabled through secondary execution controls.
7138 	 * Because it is marked as EmulateOnUD, we need to intercept it here.
7139 	 */
7140 	if (info->intercept == x86_intercept_rdtscp &&
7141 	    !nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDTSCP)) {
7142 		ctxt->exception.vector = UD_VECTOR;
7143 		ctxt->exception.error_code_valid = false;
7144 		return X86EMUL_PROPAGATE_FAULT;
7145 	}
7146 
7147 	/* TODO: check more intercepts... */
7148 	return X86EMUL_CONTINUE;
7149 }
7150 
7151 #ifdef CONFIG_X86_64
7152 /* (a << shift) / divisor, return 1 if overflow otherwise 0 */
7153 static inline int u64_shl_div_u64(u64 a, unsigned int shift,
7154 				  u64 divisor, u64 *result)
7155 {
7156 	u64 low = a << shift, high = a >> (64 - shift);
7157 
7158 	/* To avoid the overflow on divq */
7159 	if (high >= divisor)
7160 		return 1;
7161 
7162 	/* Low hold the result, high hold rem which is discarded */
7163 	asm("divq %2\n\t" : "=a" (low), "=d" (high) :
7164 	    "rm" (divisor), "0" (low), "1" (high));
7165 	*result = low;
7166 
7167 	return 0;
7168 }
7169 
7170 static int vmx_set_hv_timer(struct kvm_vcpu *vcpu, u64 guest_deadline_tsc,
7171 			    bool *expired)
7172 {
7173 	struct vcpu_vmx *vmx;
7174 	u64 tscl, guest_tscl, delta_tsc, lapic_timer_advance_cycles;
7175 	struct kvm_timer *ktimer = &vcpu->arch.apic->lapic_timer;
7176 
7177 	if (kvm_mwait_in_guest(vcpu->kvm) ||
7178 		kvm_can_post_timer_interrupt(vcpu))
7179 		return -EOPNOTSUPP;
7180 
7181 	vmx = to_vmx(vcpu);
7182 	tscl = rdtsc();
7183 	guest_tscl = kvm_read_l1_tsc(vcpu, tscl);
7184 	delta_tsc = max(guest_deadline_tsc, guest_tscl) - guest_tscl;
7185 	lapic_timer_advance_cycles = nsec_to_cycles(vcpu,
7186 						    ktimer->timer_advance_ns);
7187 
7188 	if (delta_tsc > lapic_timer_advance_cycles)
7189 		delta_tsc -= lapic_timer_advance_cycles;
7190 	else
7191 		delta_tsc = 0;
7192 
7193 	/* Convert to host delta tsc if tsc scaling is enabled */
7194 	if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio &&
7195 	    delta_tsc && u64_shl_div_u64(delta_tsc,
7196 				kvm_tsc_scaling_ratio_frac_bits,
7197 				vcpu->arch.tsc_scaling_ratio, &delta_tsc))
7198 		return -ERANGE;
7199 
7200 	/*
7201 	 * If the delta tsc can't fit in the 32 bit after the multi shift,
7202 	 * we can't use the preemption timer.
7203 	 * It's possible that it fits on later vmentries, but checking
7204 	 * on every vmentry is costly so we just use an hrtimer.
7205 	 */
7206 	if (delta_tsc >> (cpu_preemption_timer_multi + 32))
7207 		return -ERANGE;
7208 
7209 	vmx->hv_deadline_tsc = tscl + delta_tsc;
7210 	*expired = !delta_tsc;
7211 	return 0;
7212 }
7213 
7214 static void vmx_cancel_hv_timer(struct kvm_vcpu *vcpu)
7215 {
7216 	to_vmx(vcpu)->hv_deadline_tsc = -1;
7217 }
7218 #endif
7219 
7220 static void vmx_sched_in(struct kvm_vcpu *vcpu, int cpu)
7221 {
7222 	if (!kvm_pause_in_guest(vcpu->kvm))
7223 		shrink_ple_window(vcpu);
7224 }
7225 
7226 static void vmx_slot_enable_log_dirty(struct kvm *kvm,
7227 				     struct kvm_memory_slot *slot)
7228 {
7229 	kvm_mmu_slot_leaf_clear_dirty(kvm, slot);
7230 	kvm_mmu_slot_largepage_remove_write_access(kvm, slot);
7231 }
7232 
7233 static void vmx_slot_disable_log_dirty(struct kvm *kvm,
7234 				       struct kvm_memory_slot *slot)
7235 {
7236 	kvm_mmu_slot_set_dirty(kvm, slot);
7237 }
7238 
7239 static void vmx_flush_log_dirty(struct kvm *kvm)
7240 {
7241 	kvm_flush_pml_buffers(kvm);
7242 }
7243 
7244 static int vmx_write_pml_buffer(struct kvm_vcpu *vcpu)
7245 {
7246 	struct vmcs12 *vmcs12;
7247 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7248 	gpa_t gpa, dst;
7249 
7250 	if (is_guest_mode(vcpu)) {
7251 		WARN_ON_ONCE(vmx->nested.pml_full);
7252 
7253 		/*
7254 		 * Check if PML is enabled for the nested guest.
7255 		 * Whether eptp bit 6 is set is already checked
7256 		 * as part of A/D emulation.
7257 		 */
7258 		vmcs12 = get_vmcs12(vcpu);
7259 		if (!nested_cpu_has_pml(vmcs12))
7260 			return 0;
7261 
7262 		if (vmcs12->guest_pml_index >= PML_ENTITY_NUM) {
7263 			vmx->nested.pml_full = true;
7264 			return 1;
7265 		}
7266 
7267 		gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS) & ~0xFFFull;
7268 		dst = vmcs12->pml_address + sizeof(u64) * vmcs12->guest_pml_index;
7269 
7270 		if (kvm_write_guest_page(vcpu->kvm, gpa_to_gfn(dst), &gpa,
7271 					 offset_in_page(dst), sizeof(gpa)))
7272 			return 0;
7273 
7274 		vmcs12->guest_pml_index--;
7275 	}
7276 
7277 	return 0;
7278 }
7279 
7280 static void vmx_enable_log_dirty_pt_masked(struct kvm *kvm,
7281 					   struct kvm_memory_slot *memslot,
7282 					   gfn_t offset, unsigned long mask)
7283 {
7284 	kvm_mmu_clear_dirty_pt_masked(kvm, memslot, offset, mask);
7285 }
7286 
7287 static void __pi_post_block(struct kvm_vcpu *vcpu)
7288 {
7289 	struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
7290 	struct pi_desc old, new;
7291 	unsigned int dest;
7292 
7293 	do {
7294 		old.control = new.control = pi_desc->control;
7295 		WARN(old.nv != POSTED_INTR_WAKEUP_VECTOR,
7296 		     "Wakeup handler not enabled while the VCPU is blocked\n");
7297 
7298 		dest = cpu_physical_id(vcpu->cpu);
7299 
7300 		if (x2apic_enabled())
7301 			new.ndst = dest;
7302 		else
7303 			new.ndst = (dest << 8) & 0xFF00;
7304 
7305 		/* set 'NV' to 'notification vector' */
7306 		new.nv = POSTED_INTR_VECTOR;
7307 	} while (cmpxchg64(&pi_desc->control, old.control,
7308 			   new.control) != old.control);
7309 
7310 	if (!WARN_ON_ONCE(vcpu->pre_pcpu == -1)) {
7311 		spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7312 		list_del(&vcpu->blocked_vcpu_list);
7313 		spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7314 		vcpu->pre_pcpu = -1;
7315 	}
7316 }
7317 
7318 /*
7319  * This routine does the following things for vCPU which is going
7320  * to be blocked if VT-d PI is enabled.
7321  * - Store the vCPU to the wakeup list, so when interrupts happen
7322  *   we can find the right vCPU to wake up.
7323  * - Change the Posted-interrupt descriptor as below:
7324  *      'NDST' <-- vcpu->pre_pcpu
7325  *      'NV' <-- POSTED_INTR_WAKEUP_VECTOR
7326  * - If 'ON' is set during this process, which means at least one
7327  *   interrupt is posted for this vCPU, we cannot block it, in
7328  *   this case, return 1, otherwise, return 0.
7329  *
7330  */
7331 static int pi_pre_block(struct kvm_vcpu *vcpu)
7332 {
7333 	unsigned int dest;
7334 	struct pi_desc old, new;
7335 	struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
7336 
7337 	if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
7338 		!irq_remapping_cap(IRQ_POSTING_CAP)  ||
7339 		!kvm_vcpu_apicv_active(vcpu))
7340 		return 0;
7341 
7342 	WARN_ON(irqs_disabled());
7343 	local_irq_disable();
7344 	if (!WARN_ON_ONCE(vcpu->pre_pcpu != -1)) {
7345 		vcpu->pre_pcpu = vcpu->cpu;
7346 		spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7347 		list_add_tail(&vcpu->blocked_vcpu_list,
7348 			      &per_cpu(blocked_vcpu_on_cpu,
7349 				       vcpu->pre_pcpu));
7350 		spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7351 	}
7352 
7353 	do {
7354 		old.control = new.control = pi_desc->control;
7355 
7356 		WARN((pi_desc->sn == 1),
7357 		     "Warning: SN field of posted-interrupts "
7358 		     "is set before blocking\n");
7359 
7360 		/*
7361 		 * Since vCPU can be preempted during this process,
7362 		 * vcpu->cpu could be different with pre_pcpu, we
7363 		 * need to set pre_pcpu as the destination of wakeup
7364 		 * notification event, then we can find the right vCPU
7365 		 * to wakeup in wakeup handler if interrupts happen
7366 		 * when the vCPU is in blocked state.
7367 		 */
7368 		dest = cpu_physical_id(vcpu->pre_pcpu);
7369 
7370 		if (x2apic_enabled())
7371 			new.ndst = dest;
7372 		else
7373 			new.ndst = (dest << 8) & 0xFF00;
7374 
7375 		/* set 'NV' to 'wakeup vector' */
7376 		new.nv = POSTED_INTR_WAKEUP_VECTOR;
7377 	} while (cmpxchg64(&pi_desc->control, old.control,
7378 			   new.control) != old.control);
7379 
7380 	/* We should not block the vCPU if an interrupt is posted for it.  */
7381 	if (pi_test_on(pi_desc) == 1)
7382 		__pi_post_block(vcpu);
7383 
7384 	local_irq_enable();
7385 	return (vcpu->pre_pcpu == -1);
7386 }
7387 
7388 static int vmx_pre_block(struct kvm_vcpu *vcpu)
7389 {
7390 	if (pi_pre_block(vcpu))
7391 		return 1;
7392 
7393 	if (kvm_lapic_hv_timer_in_use(vcpu))
7394 		kvm_lapic_switch_to_sw_timer(vcpu);
7395 
7396 	return 0;
7397 }
7398 
7399 static void pi_post_block(struct kvm_vcpu *vcpu)
7400 {
7401 	if (vcpu->pre_pcpu == -1)
7402 		return;
7403 
7404 	WARN_ON(irqs_disabled());
7405 	local_irq_disable();
7406 	__pi_post_block(vcpu);
7407 	local_irq_enable();
7408 }
7409 
7410 static void vmx_post_block(struct kvm_vcpu *vcpu)
7411 {
7412 	if (kvm_x86_ops->set_hv_timer)
7413 		kvm_lapic_switch_to_hv_timer(vcpu);
7414 
7415 	pi_post_block(vcpu);
7416 }
7417 
7418 /*
7419  * vmx_update_pi_irte - set IRTE for Posted-Interrupts
7420  *
7421  * @kvm: kvm
7422  * @host_irq: host irq of the interrupt
7423  * @guest_irq: gsi of the interrupt
7424  * @set: set or unset PI
7425  * returns 0 on success, < 0 on failure
7426  */
7427 static int vmx_update_pi_irte(struct kvm *kvm, unsigned int host_irq,
7428 			      uint32_t guest_irq, bool set)
7429 {
7430 	struct kvm_kernel_irq_routing_entry *e;
7431 	struct kvm_irq_routing_table *irq_rt;
7432 	struct kvm_lapic_irq irq;
7433 	struct kvm_vcpu *vcpu;
7434 	struct vcpu_data vcpu_info;
7435 	int idx, ret = 0;
7436 
7437 	if (!kvm_arch_has_assigned_device(kvm) ||
7438 		!irq_remapping_cap(IRQ_POSTING_CAP) ||
7439 		!kvm_vcpu_apicv_active(kvm->vcpus[0]))
7440 		return 0;
7441 
7442 	idx = srcu_read_lock(&kvm->irq_srcu);
7443 	irq_rt = srcu_dereference(kvm->irq_routing, &kvm->irq_srcu);
7444 	if (guest_irq >= irq_rt->nr_rt_entries ||
7445 	    hlist_empty(&irq_rt->map[guest_irq])) {
7446 		pr_warn_once("no route for guest_irq %u/%u (broken user space?)\n",
7447 			     guest_irq, irq_rt->nr_rt_entries);
7448 		goto out;
7449 	}
7450 
7451 	hlist_for_each_entry(e, &irq_rt->map[guest_irq], link) {
7452 		if (e->type != KVM_IRQ_ROUTING_MSI)
7453 			continue;
7454 		/*
7455 		 * VT-d PI cannot support posting multicast/broadcast
7456 		 * interrupts to a vCPU, we still use interrupt remapping
7457 		 * for these kind of interrupts.
7458 		 *
7459 		 * For lowest-priority interrupts, we only support
7460 		 * those with single CPU as the destination, e.g. user
7461 		 * configures the interrupts via /proc/irq or uses
7462 		 * irqbalance to make the interrupts single-CPU.
7463 		 *
7464 		 * We will support full lowest-priority interrupt later.
7465 		 *
7466 		 * In addition, we can only inject generic interrupts using
7467 		 * the PI mechanism, refuse to route others through it.
7468 		 */
7469 
7470 		kvm_set_msi_irq(kvm, e, &irq);
7471 		if (!kvm_intr_is_single_vcpu(kvm, &irq, &vcpu) ||
7472 		    !kvm_irq_is_postable(&irq)) {
7473 			/*
7474 			 * Make sure the IRTE is in remapped mode if
7475 			 * we don't handle it in posted mode.
7476 			 */
7477 			ret = irq_set_vcpu_affinity(host_irq, NULL);
7478 			if (ret < 0) {
7479 				printk(KERN_INFO
7480 				   "failed to back to remapped mode, irq: %u\n",
7481 				   host_irq);
7482 				goto out;
7483 			}
7484 
7485 			continue;
7486 		}
7487 
7488 		vcpu_info.pi_desc_addr = __pa(vcpu_to_pi_desc(vcpu));
7489 		vcpu_info.vector = irq.vector;
7490 
7491 		trace_kvm_pi_irte_update(host_irq, vcpu->vcpu_id, e->gsi,
7492 				vcpu_info.vector, vcpu_info.pi_desc_addr, set);
7493 
7494 		if (set)
7495 			ret = irq_set_vcpu_affinity(host_irq, &vcpu_info);
7496 		else
7497 			ret = irq_set_vcpu_affinity(host_irq, NULL);
7498 
7499 		if (ret < 0) {
7500 			printk(KERN_INFO "%s: failed to update PI IRTE\n",
7501 					__func__);
7502 			goto out;
7503 		}
7504 	}
7505 
7506 	ret = 0;
7507 out:
7508 	srcu_read_unlock(&kvm->irq_srcu, idx);
7509 	return ret;
7510 }
7511 
7512 static void vmx_setup_mce(struct kvm_vcpu *vcpu)
7513 {
7514 	if (vcpu->arch.mcg_cap & MCG_LMCE_P)
7515 		to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
7516 			FEATURE_CONTROL_LMCE;
7517 	else
7518 		to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
7519 			~FEATURE_CONTROL_LMCE;
7520 }
7521 
7522 static int vmx_smi_allowed(struct kvm_vcpu *vcpu)
7523 {
7524 	/* we need a nested vmexit to enter SMM, postpone if run is pending */
7525 	if (to_vmx(vcpu)->nested.nested_run_pending)
7526 		return 0;
7527 	return 1;
7528 }
7529 
7530 static int vmx_pre_enter_smm(struct kvm_vcpu *vcpu, char *smstate)
7531 {
7532 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7533 
7534 	vmx->nested.smm.guest_mode = is_guest_mode(vcpu);
7535 	if (vmx->nested.smm.guest_mode)
7536 		nested_vmx_vmexit(vcpu, -1, 0, 0);
7537 
7538 	vmx->nested.smm.vmxon = vmx->nested.vmxon;
7539 	vmx->nested.vmxon = false;
7540 	vmx_clear_hlt(vcpu);
7541 	return 0;
7542 }
7543 
7544 static int vmx_pre_leave_smm(struct kvm_vcpu *vcpu, const char *smstate)
7545 {
7546 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7547 	int ret;
7548 
7549 	if (vmx->nested.smm.vmxon) {
7550 		vmx->nested.vmxon = true;
7551 		vmx->nested.smm.vmxon = false;
7552 	}
7553 
7554 	if (vmx->nested.smm.guest_mode) {
7555 		ret = nested_vmx_enter_non_root_mode(vcpu, false);
7556 		if (ret)
7557 			return ret;
7558 
7559 		vmx->nested.smm.guest_mode = false;
7560 	}
7561 	return 0;
7562 }
7563 
7564 static int enable_smi_window(struct kvm_vcpu *vcpu)
7565 {
7566 	return 0;
7567 }
7568 
7569 static bool vmx_need_emulation_on_page_fault(struct kvm_vcpu *vcpu)
7570 {
7571 	return false;
7572 }
7573 
7574 static bool vmx_apic_init_signal_blocked(struct kvm_vcpu *vcpu)
7575 {
7576 	return to_vmx(vcpu)->nested.vmxon;
7577 }
7578 
7579 static __init int hardware_setup(void)
7580 {
7581 	unsigned long host_bndcfgs;
7582 	struct desc_ptr dt;
7583 	int r, i;
7584 
7585 	rdmsrl_safe(MSR_EFER, &host_efer);
7586 
7587 	store_idt(&dt);
7588 	host_idt_base = dt.address;
7589 
7590 	for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i)
7591 		kvm_define_shared_msr(i, vmx_msr_index[i]);
7592 
7593 	if (setup_vmcs_config(&vmcs_config, &vmx_capability) < 0)
7594 		return -EIO;
7595 
7596 	if (boot_cpu_has(X86_FEATURE_NX))
7597 		kvm_enable_efer_bits(EFER_NX);
7598 
7599 	if (boot_cpu_has(X86_FEATURE_MPX)) {
7600 		rdmsrl(MSR_IA32_BNDCFGS, host_bndcfgs);
7601 		WARN_ONCE(host_bndcfgs, "KVM: BNDCFGS in host will be lost");
7602 	}
7603 
7604 	if (boot_cpu_has(X86_FEATURE_XSAVES))
7605 		rdmsrl(MSR_IA32_XSS, host_xss);
7606 
7607 	if (!cpu_has_vmx_vpid() || !cpu_has_vmx_invvpid() ||
7608 	    !(cpu_has_vmx_invvpid_single() || cpu_has_vmx_invvpid_global()))
7609 		enable_vpid = 0;
7610 
7611 	if (!cpu_has_vmx_ept() ||
7612 	    !cpu_has_vmx_ept_4levels() ||
7613 	    !cpu_has_vmx_ept_mt_wb() ||
7614 	    !cpu_has_vmx_invept_global())
7615 		enable_ept = 0;
7616 
7617 	if (!cpu_has_vmx_ept_ad_bits() || !enable_ept)
7618 		enable_ept_ad_bits = 0;
7619 
7620 	if (!cpu_has_vmx_unrestricted_guest() || !enable_ept)
7621 		enable_unrestricted_guest = 0;
7622 
7623 	if (!cpu_has_vmx_flexpriority())
7624 		flexpriority_enabled = 0;
7625 
7626 	if (!cpu_has_virtual_nmis())
7627 		enable_vnmi = 0;
7628 
7629 	/*
7630 	 * set_apic_access_page_addr() is used to reload apic access
7631 	 * page upon invalidation.  No need to do anything if not
7632 	 * using the APIC_ACCESS_ADDR VMCS field.
7633 	 */
7634 	if (!flexpriority_enabled)
7635 		kvm_x86_ops->set_apic_access_page_addr = NULL;
7636 
7637 	if (!cpu_has_vmx_tpr_shadow())
7638 		kvm_x86_ops->update_cr8_intercept = NULL;
7639 
7640 	if (enable_ept && !cpu_has_vmx_ept_2m_page())
7641 		kvm_disable_largepages();
7642 
7643 #if IS_ENABLED(CONFIG_HYPERV)
7644 	if (ms_hyperv.nested_features & HV_X64_NESTED_GUEST_MAPPING_FLUSH
7645 	    && enable_ept) {
7646 		kvm_x86_ops->tlb_remote_flush = hv_remote_flush_tlb;
7647 		kvm_x86_ops->tlb_remote_flush_with_range =
7648 				hv_remote_flush_tlb_with_range;
7649 	}
7650 #endif
7651 
7652 	if (!cpu_has_vmx_ple()) {
7653 		ple_gap = 0;
7654 		ple_window = 0;
7655 		ple_window_grow = 0;
7656 		ple_window_max = 0;
7657 		ple_window_shrink = 0;
7658 	}
7659 
7660 	if (!cpu_has_vmx_apicv()) {
7661 		enable_apicv = 0;
7662 		kvm_x86_ops->sync_pir_to_irr = NULL;
7663 	}
7664 
7665 	if (cpu_has_vmx_tsc_scaling()) {
7666 		kvm_has_tsc_control = true;
7667 		kvm_max_tsc_scaling_ratio = KVM_VMX_TSC_MULTIPLIER_MAX;
7668 		kvm_tsc_scaling_ratio_frac_bits = 48;
7669 	}
7670 
7671 	set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
7672 
7673 	if (enable_ept)
7674 		vmx_enable_tdp();
7675 	else
7676 		kvm_disable_tdp();
7677 
7678 	/*
7679 	 * Only enable PML when hardware supports PML feature, and both EPT
7680 	 * and EPT A/D bit features are enabled -- PML depends on them to work.
7681 	 */
7682 	if (!enable_ept || !enable_ept_ad_bits || !cpu_has_vmx_pml())
7683 		enable_pml = 0;
7684 
7685 	if (!enable_pml) {
7686 		kvm_x86_ops->slot_enable_log_dirty = NULL;
7687 		kvm_x86_ops->slot_disable_log_dirty = NULL;
7688 		kvm_x86_ops->flush_log_dirty = NULL;
7689 		kvm_x86_ops->enable_log_dirty_pt_masked = NULL;
7690 	}
7691 
7692 	if (!cpu_has_vmx_preemption_timer())
7693 		enable_preemption_timer = false;
7694 
7695 	if (enable_preemption_timer) {
7696 		u64 use_timer_freq = 5000ULL * 1000 * 1000;
7697 		u64 vmx_msr;
7698 
7699 		rdmsrl(MSR_IA32_VMX_MISC, vmx_msr);
7700 		cpu_preemption_timer_multi =
7701 			vmx_msr & VMX_MISC_PREEMPTION_TIMER_RATE_MASK;
7702 
7703 		if (tsc_khz)
7704 			use_timer_freq = (u64)tsc_khz * 1000;
7705 		use_timer_freq >>= cpu_preemption_timer_multi;
7706 
7707 		/*
7708 		 * KVM "disables" the preemption timer by setting it to its max
7709 		 * value.  Don't use the timer if it might cause spurious exits
7710 		 * at a rate faster than 0.1 Hz (of uninterrupted guest time).
7711 		 */
7712 		if (use_timer_freq > 0xffffffffu / 10)
7713 			enable_preemption_timer = false;
7714 	}
7715 
7716 	if (!enable_preemption_timer) {
7717 		kvm_x86_ops->set_hv_timer = NULL;
7718 		kvm_x86_ops->cancel_hv_timer = NULL;
7719 		kvm_x86_ops->request_immediate_exit = __kvm_request_immediate_exit;
7720 	}
7721 
7722 	kvm_set_posted_intr_wakeup_handler(wakeup_handler);
7723 
7724 	kvm_mce_cap_supported |= MCG_LMCE_P;
7725 
7726 	if (pt_mode != PT_MODE_SYSTEM && pt_mode != PT_MODE_HOST_GUEST)
7727 		return -EINVAL;
7728 	if (!enable_ept || !cpu_has_vmx_intel_pt())
7729 		pt_mode = PT_MODE_SYSTEM;
7730 
7731 	if (nested) {
7732 		nested_vmx_setup_ctls_msrs(&vmcs_config.nested,
7733 					   vmx_capability.ept, enable_apicv);
7734 
7735 		r = nested_vmx_hardware_setup(kvm_vmx_exit_handlers);
7736 		if (r)
7737 			return r;
7738 	}
7739 
7740 	r = alloc_kvm_area();
7741 	if (r)
7742 		nested_vmx_hardware_unsetup();
7743 	return r;
7744 }
7745 
7746 static __exit void hardware_unsetup(void)
7747 {
7748 	if (nested)
7749 		nested_vmx_hardware_unsetup();
7750 
7751 	free_kvm_area();
7752 }
7753 
7754 static struct kvm_x86_ops vmx_x86_ops __ro_after_init = {
7755 	.cpu_has_kvm_support = cpu_has_kvm_support,
7756 	.disabled_by_bios = vmx_disabled_by_bios,
7757 	.hardware_setup = hardware_setup,
7758 	.hardware_unsetup = hardware_unsetup,
7759 	.check_processor_compatibility = vmx_check_processor_compat,
7760 	.hardware_enable = hardware_enable,
7761 	.hardware_disable = hardware_disable,
7762 	.cpu_has_accelerated_tpr = report_flexpriority,
7763 	.has_emulated_msr = vmx_has_emulated_msr,
7764 
7765 	.vm_init = vmx_vm_init,
7766 	.vm_alloc = vmx_vm_alloc,
7767 	.vm_free = vmx_vm_free,
7768 
7769 	.vcpu_create = vmx_create_vcpu,
7770 	.vcpu_free = vmx_free_vcpu,
7771 	.vcpu_reset = vmx_vcpu_reset,
7772 
7773 	.prepare_guest_switch = vmx_prepare_switch_to_guest,
7774 	.vcpu_load = vmx_vcpu_load,
7775 	.vcpu_put = vmx_vcpu_put,
7776 
7777 	.update_bp_intercept = update_exception_bitmap,
7778 	.get_msr_feature = vmx_get_msr_feature,
7779 	.get_msr = vmx_get_msr,
7780 	.set_msr = vmx_set_msr,
7781 	.get_segment_base = vmx_get_segment_base,
7782 	.get_segment = vmx_get_segment,
7783 	.set_segment = vmx_set_segment,
7784 	.get_cpl = vmx_get_cpl,
7785 	.get_cs_db_l_bits = vmx_get_cs_db_l_bits,
7786 	.decache_cr0_guest_bits = vmx_decache_cr0_guest_bits,
7787 	.decache_cr3 = vmx_decache_cr3,
7788 	.decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
7789 	.set_cr0 = vmx_set_cr0,
7790 	.set_cr3 = vmx_set_cr3,
7791 	.set_cr4 = vmx_set_cr4,
7792 	.set_efer = vmx_set_efer,
7793 	.get_idt = vmx_get_idt,
7794 	.set_idt = vmx_set_idt,
7795 	.get_gdt = vmx_get_gdt,
7796 	.set_gdt = vmx_set_gdt,
7797 	.get_dr6 = vmx_get_dr6,
7798 	.set_dr6 = vmx_set_dr6,
7799 	.set_dr7 = vmx_set_dr7,
7800 	.sync_dirty_debug_regs = vmx_sync_dirty_debug_regs,
7801 	.cache_reg = vmx_cache_reg,
7802 	.get_rflags = vmx_get_rflags,
7803 	.set_rflags = vmx_set_rflags,
7804 
7805 	.tlb_flush = vmx_flush_tlb,
7806 	.tlb_flush_gva = vmx_flush_tlb_gva,
7807 
7808 	.run = vmx_vcpu_run,
7809 	.handle_exit = vmx_handle_exit,
7810 	.skip_emulated_instruction = skip_emulated_instruction,
7811 	.set_interrupt_shadow = vmx_set_interrupt_shadow,
7812 	.get_interrupt_shadow = vmx_get_interrupt_shadow,
7813 	.patch_hypercall = vmx_patch_hypercall,
7814 	.set_irq = vmx_inject_irq,
7815 	.set_nmi = vmx_inject_nmi,
7816 	.queue_exception = vmx_queue_exception,
7817 	.cancel_injection = vmx_cancel_injection,
7818 	.interrupt_allowed = vmx_interrupt_allowed,
7819 	.nmi_allowed = vmx_nmi_allowed,
7820 	.get_nmi_mask = vmx_get_nmi_mask,
7821 	.set_nmi_mask = vmx_set_nmi_mask,
7822 	.enable_nmi_window = enable_nmi_window,
7823 	.enable_irq_window = enable_irq_window,
7824 	.update_cr8_intercept = update_cr8_intercept,
7825 	.set_virtual_apic_mode = vmx_set_virtual_apic_mode,
7826 	.set_apic_access_page_addr = vmx_set_apic_access_page_addr,
7827 	.get_enable_apicv = vmx_get_enable_apicv,
7828 	.refresh_apicv_exec_ctrl = vmx_refresh_apicv_exec_ctrl,
7829 	.load_eoi_exitmap = vmx_load_eoi_exitmap,
7830 	.apicv_post_state_restore = vmx_apicv_post_state_restore,
7831 	.hwapic_irr_update = vmx_hwapic_irr_update,
7832 	.hwapic_isr_update = vmx_hwapic_isr_update,
7833 	.guest_apic_has_interrupt = vmx_guest_apic_has_interrupt,
7834 	.sync_pir_to_irr = vmx_sync_pir_to_irr,
7835 	.deliver_posted_interrupt = vmx_deliver_posted_interrupt,
7836 	.dy_apicv_has_pending_interrupt = vmx_dy_apicv_has_pending_interrupt,
7837 
7838 	.set_tss_addr = vmx_set_tss_addr,
7839 	.set_identity_map_addr = vmx_set_identity_map_addr,
7840 	.get_tdp_level = get_ept_level,
7841 	.get_mt_mask = vmx_get_mt_mask,
7842 
7843 	.get_exit_info = vmx_get_exit_info,
7844 
7845 	.get_lpage_level = vmx_get_lpage_level,
7846 
7847 	.cpuid_update = vmx_cpuid_update,
7848 
7849 	.rdtscp_supported = vmx_rdtscp_supported,
7850 	.invpcid_supported = vmx_invpcid_supported,
7851 
7852 	.set_supported_cpuid = vmx_set_supported_cpuid,
7853 
7854 	.has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
7855 
7856 	.read_l1_tsc_offset = vmx_read_l1_tsc_offset,
7857 	.write_l1_tsc_offset = vmx_write_l1_tsc_offset,
7858 
7859 	.set_tdp_cr3 = vmx_set_cr3,
7860 
7861 	.check_intercept = vmx_check_intercept,
7862 	.handle_exit_irqoff = vmx_handle_exit_irqoff,
7863 	.mpx_supported = vmx_mpx_supported,
7864 	.xsaves_supported = vmx_xsaves_supported,
7865 	.umip_emulated = vmx_umip_emulated,
7866 	.pt_supported = vmx_pt_supported,
7867 
7868 	.request_immediate_exit = vmx_request_immediate_exit,
7869 
7870 	.sched_in = vmx_sched_in,
7871 
7872 	.slot_enable_log_dirty = vmx_slot_enable_log_dirty,
7873 	.slot_disable_log_dirty = vmx_slot_disable_log_dirty,
7874 	.flush_log_dirty = vmx_flush_log_dirty,
7875 	.enable_log_dirty_pt_masked = vmx_enable_log_dirty_pt_masked,
7876 	.write_log_dirty = vmx_write_pml_buffer,
7877 
7878 	.pre_block = vmx_pre_block,
7879 	.post_block = vmx_post_block,
7880 
7881 	.pmu_ops = &intel_pmu_ops,
7882 
7883 	.update_pi_irte = vmx_update_pi_irte,
7884 
7885 #ifdef CONFIG_X86_64
7886 	.set_hv_timer = vmx_set_hv_timer,
7887 	.cancel_hv_timer = vmx_cancel_hv_timer,
7888 #endif
7889 
7890 	.setup_mce = vmx_setup_mce,
7891 
7892 	.smi_allowed = vmx_smi_allowed,
7893 	.pre_enter_smm = vmx_pre_enter_smm,
7894 	.pre_leave_smm = vmx_pre_leave_smm,
7895 	.enable_smi_window = enable_smi_window,
7896 
7897 	.check_nested_events = NULL,
7898 	.get_nested_state = NULL,
7899 	.set_nested_state = NULL,
7900 	.get_vmcs12_pages = NULL,
7901 	.nested_enable_evmcs = NULL,
7902 	.nested_get_evmcs_version = NULL,
7903 	.need_emulation_on_page_fault = vmx_need_emulation_on_page_fault,
7904 	.apic_init_signal_blocked = vmx_apic_init_signal_blocked,
7905 };
7906 
7907 static void vmx_cleanup_l1d_flush(void)
7908 {
7909 	if (vmx_l1d_flush_pages) {
7910 		free_pages((unsigned long)vmx_l1d_flush_pages, L1D_CACHE_ORDER);
7911 		vmx_l1d_flush_pages = NULL;
7912 	}
7913 	/* Restore state so sysfs ignores VMX */
7914 	l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_AUTO;
7915 }
7916 
7917 static void vmx_exit(void)
7918 {
7919 #ifdef CONFIG_KEXEC_CORE
7920 	RCU_INIT_POINTER(crash_vmclear_loaded_vmcss, NULL);
7921 	synchronize_rcu();
7922 #endif
7923 
7924 	kvm_exit();
7925 
7926 #if IS_ENABLED(CONFIG_HYPERV)
7927 	if (static_branch_unlikely(&enable_evmcs)) {
7928 		int cpu;
7929 		struct hv_vp_assist_page *vp_ap;
7930 		/*
7931 		 * Reset everything to support using non-enlightened VMCS
7932 		 * access later (e.g. when we reload the module with
7933 		 * enlightened_vmcs=0)
7934 		 */
7935 		for_each_online_cpu(cpu) {
7936 			vp_ap =	hv_get_vp_assist_page(cpu);
7937 
7938 			if (!vp_ap)
7939 				continue;
7940 
7941 			vp_ap->nested_control.features.directhypercall = 0;
7942 			vp_ap->current_nested_vmcs = 0;
7943 			vp_ap->enlighten_vmentry = 0;
7944 		}
7945 
7946 		static_branch_disable(&enable_evmcs);
7947 	}
7948 #endif
7949 	vmx_cleanup_l1d_flush();
7950 }
7951 module_exit(vmx_exit);
7952 
7953 static int __init vmx_init(void)
7954 {
7955 	int r;
7956 
7957 #if IS_ENABLED(CONFIG_HYPERV)
7958 	/*
7959 	 * Enlightened VMCS usage should be recommended and the host needs
7960 	 * to support eVMCS v1 or above. We can also disable eVMCS support
7961 	 * with module parameter.
7962 	 */
7963 	if (enlightened_vmcs &&
7964 	    ms_hyperv.hints & HV_X64_ENLIGHTENED_VMCS_RECOMMENDED &&
7965 	    (ms_hyperv.nested_features & HV_X64_ENLIGHTENED_VMCS_VERSION) >=
7966 	    KVM_EVMCS_VERSION) {
7967 		int cpu;
7968 
7969 		/* Check that we have assist pages on all online CPUs */
7970 		for_each_online_cpu(cpu) {
7971 			if (!hv_get_vp_assist_page(cpu)) {
7972 				enlightened_vmcs = false;
7973 				break;
7974 			}
7975 		}
7976 
7977 		if (enlightened_vmcs) {
7978 			pr_info("KVM: vmx: using Hyper-V Enlightened VMCS\n");
7979 			static_branch_enable(&enable_evmcs);
7980 		}
7981 
7982 		if (ms_hyperv.nested_features & HV_X64_NESTED_DIRECT_FLUSH)
7983 			vmx_x86_ops.enable_direct_tlbflush
7984 				= hv_enable_direct_tlbflush;
7985 
7986 	} else {
7987 		enlightened_vmcs = false;
7988 	}
7989 #endif
7990 
7991 	r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx),
7992 		     __alignof__(struct vcpu_vmx), THIS_MODULE);
7993 	if (r)
7994 		return r;
7995 
7996 	/*
7997 	 * Must be called after kvm_init() so enable_ept is properly set
7998 	 * up. Hand the parameter mitigation value in which was stored in
7999 	 * the pre module init parser. If no parameter was given, it will
8000 	 * contain 'auto' which will be turned into the default 'cond'
8001 	 * mitigation mode.
8002 	 */
8003 	r = vmx_setup_l1d_flush(vmentry_l1d_flush_param);
8004 	if (r) {
8005 		vmx_exit();
8006 		return r;
8007 	}
8008 
8009 #ifdef CONFIG_KEXEC_CORE
8010 	rcu_assign_pointer(crash_vmclear_loaded_vmcss,
8011 			   crash_vmclear_local_loaded_vmcss);
8012 #endif
8013 	vmx_check_vmcs12_offsets();
8014 
8015 	return 0;
8016 }
8017 module_init(vmx_init);
8018