xref: /openbmc/linux/arch/x86/kvm/vmx/vmx.c (revision f9834f18)
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 #ifdef MODULE
68 static const struct x86_cpu_id vmx_cpu_id[] = {
69 	X86_FEATURE_MATCH(X86_FEATURE_VMX),
70 	{}
71 };
72 MODULE_DEVICE_TABLE(x86cpu, vmx_cpu_id);
73 #endif
74 
75 bool __read_mostly enable_vpid = 1;
76 module_param_named(vpid, enable_vpid, bool, 0444);
77 
78 static bool __read_mostly enable_vnmi = 1;
79 module_param_named(vnmi, enable_vnmi, bool, S_IRUGO);
80 
81 bool __read_mostly flexpriority_enabled = 1;
82 module_param_named(flexpriority, flexpriority_enabled, bool, S_IRUGO);
83 
84 bool __read_mostly enable_ept = 1;
85 module_param_named(ept, enable_ept, bool, S_IRUGO);
86 
87 bool __read_mostly enable_unrestricted_guest = 1;
88 module_param_named(unrestricted_guest,
89 			enable_unrestricted_guest, bool, S_IRUGO);
90 
91 bool __read_mostly enable_ept_ad_bits = 1;
92 module_param_named(eptad, enable_ept_ad_bits, bool, S_IRUGO);
93 
94 static bool __read_mostly emulate_invalid_guest_state = true;
95 module_param(emulate_invalid_guest_state, bool, S_IRUGO);
96 
97 static bool __read_mostly fasteoi = 1;
98 module_param(fasteoi, bool, S_IRUGO);
99 
100 bool __read_mostly enable_apicv = 1;
101 module_param(enable_apicv, bool, S_IRUGO);
102 
103 /*
104  * If nested=1, nested virtualization is supported, i.e., guests may use
105  * VMX and be a hypervisor for its own guests. If nested=0, guests may not
106  * use VMX instructions.
107  */
108 static bool __read_mostly nested = 1;
109 module_param(nested, bool, S_IRUGO);
110 
111 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 	MSR_IA32_TSX_CTRL,
454 };
455 
456 #if IS_ENABLED(CONFIG_HYPERV)
457 static bool __read_mostly enlightened_vmcs = true;
458 module_param(enlightened_vmcs, bool, 0444);
459 
460 /* check_ept_pointer() should be under protection of ept_pointer_lock. */
461 static void check_ept_pointer_match(struct kvm *kvm)
462 {
463 	struct kvm_vcpu *vcpu;
464 	u64 tmp_eptp = INVALID_PAGE;
465 	int i;
466 
467 	kvm_for_each_vcpu(i, vcpu, kvm) {
468 		if (!VALID_PAGE(tmp_eptp)) {
469 			tmp_eptp = to_vmx(vcpu)->ept_pointer;
470 		} else if (tmp_eptp != to_vmx(vcpu)->ept_pointer) {
471 			to_kvm_vmx(kvm)->ept_pointers_match
472 				= EPT_POINTERS_MISMATCH;
473 			return;
474 		}
475 	}
476 
477 	to_kvm_vmx(kvm)->ept_pointers_match = EPT_POINTERS_MATCH;
478 }
479 
480 static int kvm_fill_hv_flush_list_func(struct hv_guest_mapping_flush_list *flush,
481 		void *data)
482 {
483 	struct kvm_tlb_range *range = data;
484 
485 	return hyperv_fill_flush_guest_mapping_list(flush, range->start_gfn,
486 			range->pages);
487 }
488 
489 static inline int __hv_remote_flush_tlb_with_range(struct kvm *kvm,
490 		struct kvm_vcpu *vcpu, struct kvm_tlb_range *range)
491 {
492 	u64 ept_pointer = to_vmx(vcpu)->ept_pointer;
493 
494 	/*
495 	 * FLUSH_GUEST_PHYSICAL_ADDRESS_SPACE hypercall needs address
496 	 * of the base of EPT PML4 table, strip off EPT configuration
497 	 * information.
498 	 */
499 	if (range)
500 		return hyperv_flush_guest_mapping_range(ept_pointer & PAGE_MASK,
501 				kvm_fill_hv_flush_list_func, (void *)range);
502 	else
503 		return hyperv_flush_guest_mapping(ept_pointer & PAGE_MASK);
504 }
505 
506 static int hv_remote_flush_tlb_with_range(struct kvm *kvm,
507 		struct kvm_tlb_range *range)
508 {
509 	struct kvm_vcpu *vcpu;
510 	int ret = 0, i;
511 
512 	spin_lock(&to_kvm_vmx(kvm)->ept_pointer_lock);
513 
514 	if (to_kvm_vmx(kvm)->ept_pointers_match == EPT_POINTERS_CHECK)
515 		check_ept_pointer_match(kvm);
516 
517 	if (to_kvm_vmx(kvm)->ept_pointers_match != EPT_POINTERS_MATCH) {
518 		kvm_for_each_vcpu(i, vcpu, kvm) {
519 			/* If ept_pointer is invalid pointer, bypass flush request. */
520 			if (VALID_PAGE(to_vmx(vcpu)->ept_pointer))
521 				ret |= __hv_remote_flush_tlb_with_range(
522 					kvm, vcpu, range);
523 		}
524 	} else {
525 		ret = __hv_remote_flush_tlb_with_range(kvm,
526 				kvm_get_vcpu(kvm, 0), range);
527 	}
528 
529 	spin_unlock(&to_kvm_vmx(kvm)->ept_pointer_lock);
530 	return ret;
531 }
532 static int hv_remote_flush_tlb(struct kvm *kvm)
533 {
534 	return hv_remote_flush_tlb_with_range(kvm, NULL);
535 }
536 
537 static int hv_enable_direct_tlbflush(struct kvm_vcpu *vcpu)
538 {
539 	struct hv_enlightened_vmcs *evmcs;
540 	struct hv_partition_assist_pg **p_hv_pa_pg =
541 			&vcpu->kvm->arch.hyperv.hv_pa_pg;
542 	/*
543 	 * Synthetic VM-Exit is not enabled in current code and so All
544 	 * evmcs in singe VM shares same assist page.
545 	 */
546 	if (!*p_hv_pa_pg)
547 		*p_hv_pa_pg = kzalloc(PAGE_SIZE, GFP_KERNEL);
548 
549 	if (!*p_hv_pa_pg)
550 		return -ENOMEM;
551 
552 	evmcs = (struct hv_enlightened_vmcs *)to_vmx(vcpu)->loaded_vmcs->vmcs;
553 
554 	evmcs->partition_assist_page =
555 		__pa(*p_hv_pa_pg);
556 	evmcs->hv_vm_id = (unsigned long)vcpu->kvm;
557 	evmcs->hv_enlightenments_control.nested_flush_hypercall = 1;
558 
559 	return 0;
560 }
561 
562 #endif /* IS_ENABLED(CONFIG_HYPERV) */
563 
564 /*
565  * Comment's format: document - errata name - stepping - processor name.
566  * Refer from
567  * https://www.virtualbox.org/svn/vbox/trunk/src/VBox/VMM/VMMR0/HMR0.cpp
568  */
569 static u32 vmx_preemption_cpu_tfms[] = {
570 /* 323344.pdf - BA86   - D0 - Xeon 7500 Series */
571 0x000206E6,
572 /* 323056.pdf - AAX65  - C2 - Xeon L3406 */
573 /* 322814.pdf - AAT59  - C2 - i7-600, i5-500, i5-400 and i3-300 Mobile */
574 /* 322911.pdf - AAU65  - C2 - i5-600, i3-500 Desktop and Pentium G6950 */
575 0x00020652,
576 /* 322911.pdf - AAU65  - K0 - i5-600, i3-500 Desktop and Pentium G6950 */
577 0x00020655,
578 /* 322373.pdf - AAO95  - B1 - Xeon 3400 Series */
579 /* 322166.pdf - AAN92  - B1 - i7-800 and i5-700 Desktop */
580 /*
581  * 320767.pdf - AAP86  - B1 -
582  * i7-900 Mobile Extreme, i7-800 and i7-700 Mobile
583  */
584 0x000106E5,
585 /* 321333.pdf - AAM126 - C0 - Xeon 3500 */
586 0x000106A0,
587 /* 321333.pdf - AAM126 - C1 - Xeon 3500 */
588 0x000106A1,
589 /* 320836.pdf - AAJ124 - C0 - i7-900 Desktop Extreme and i7-900 Desktop */
590 0x000106A4,
591  /* 321333.pdf - AAM126 - D0 - Xeon 3500 */
592  /* 321324.pdf - AAK139 - D0 - Xeon 5500 */
593  /* 320836.pdf - AAJ124 - D0 - i7-900 Extreme and i7-900 Desktop */
594 0x000106A5,
595  /* Xeon E3-1220 V2 */
596 0x000306A8,
597 };
598 
599 static inline bool cpu_has_broken_vmx_preemption_timer(void)
600 {
601 	u32 eax = cpuid_eax(0x00000001), i;
602 
603 	/* Clear the reserved bits */
604 	eax &= ~(0x3U << 14 | 0xfU << 28);
605 	for (i = 0; i < ARRAY_SIZE(vmx_preemption_cpu_tfms); i++)
606 		if (eax == vmx_preemption_cpu_tfms[i])
607 			return true;
608 
609 	return false;
610 }
611 
612 static inline bool cpu_need_virtualize_apic_accesses(struct kvm_vcpu *vcpu)
613 {
614 	return flexpriority_enabled && lapic_in_kernel(vcpu);
615 }
616 
617 static inline bool report_flexpriority(void)
618 {
619 	return flexpriority_enabled;
620 }
621 
622 static inline int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
623 {
624 	int i;
625 
626 	for (i = 0; i < vmx->nmsrs; ++i)
627 		if (vmx_msr_index[vmx->guest_msrs[i].index] == msr)
628 			return i;
629 	return -1;
630 }
631 
632 struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
633 {
634 	int i;
635 
636 	i = __find_msr_index(vmx, msr);
637 	if (i >= 0)
638 		return &vmx->guest_msrs[i];
639 	return NULL;
640 }
641 
642 static int vmx_set_guest_msr(struct vcpu_vmx *vmx, struct shared_msr_entry *msr, u64 data)
643 {
644 	int ret = 0;
645 
646 	u64 old_msr_data = msr->data;
647 	msr->data = data;
648 	if (msr - vmx->guest_msrs < vmx->save_nmsrs) {
649 		preempt_disable();
650 		ret = kvm_set_shared_msr(msr->index, msr->data,
651 					 msr->mask);
652 		preempt_enable();
653 		if (ret)
654 			msr->data = old_msr_data;
655 	}
656 	return ret;
657 }
658 
659 void loaded_vmcs_init(struct loaded_vmcs *loaded_vmcs)
660 {
661 	vmcs_clear(loaded_vmcs->vmcs);
662 	if (loaded_vmcs->shadow_vmcs && loaded_vmcs->launched)
663 		vmcs_clear(loaded_vmcs->shadow_vmcs);
664 	loaded_vmcs->cpu = -1;
665 	loaded_vmcs->launched = 0;
666 }
667 
668 #ifdef CONFIG_KEXEC_CORE
669 /*
670  * This bitmap is used to indicate whether the vmclear
671  * operation is enabled on all cpus. All disabled by
672  * default.
673  */
674 static cpumask_t crash_vmclear_enabled_bitmap = CPU_MASK_NONE;
675 
676 static inline void crash_enable_local_vmclear(int cpu)
677 {
678 	cpumask_set_cpu(cpu, &crash_vmclear_enabled_bitmap);
679 }
680 
681 static inline void crash_disable_local_vmclear(int cpu)
682 {
683 	cpumask_clear_cpu(cpu, &crash_vmclear_enabled_bitmap);
684 }
685 
686 static inline int crash_local_vmclear_enabled(int cpu)
687 {
688 	return cpumask_test_cpu(cpu, &crash_vmclear_enabled_bitmap);
689 }
690 
691 static void crash_vmclear_local_loaded_vmcss(void)
692 {
693 	int cpu = raw_smp_processor_id();
694 	struct loaded_vmcs *v;
695 
696 	if (!crash_local_vmclear_enabled(cpu))
697 		return;
698 
699 	list_for_each_entry(v, &per_cpu(loaded_vmcss_on_cpu, cpu),
700 			    loaded_vmcss_on_cpu_link)
701 		vmcs_clear(v->vmcs);
702 }
703 #else
704 static inline void crash_enable_local_vmclear(int cpu) { }
705 static inline void crash_disable_local_vmclear(int cpu) { }
706 #endif /* CONFIG_KEXEC_CORE */
707 
708 static void __loaded_vmcs_clear(void *arg)
709 {
710 	struct loaded_vmcs *loaded_vmcs = arg;
711 	int cpu = raw_smp_processor_id();
712 
713 	if (loaded_vmcs->cpu != cpu)
714 		return; /* vcpu migration can race with cpu offline */
715 	if (per_cpu(current_vmcs, cpu) == loaded_vmcs->vmcs)
716 		per_cpu(current_vmcs, cpu) = NULL;
717 	crash_disable_local_vmclear(cpu);
718 	list_del(&loaded_vmcs->loaded_vmcss_on_cpu_link);
719 
720 	/*
721 	 * we should ensure updating loaded_vmcs->loaded_vmcss_on_cpu_link
722 	 * is before setting loaded_vmcs->vcpu to -1 which is done in
723 	 * loaded_vmcs_init. Otherwise, other cpu can see vcpu = -1 fist
724 	 * then adds the vmcs into percpu list before it is deleted.
725 	 */
726 	smp_wmb();
727 
728 	loaded_vmcs_init(loaded_vmcs);
729 	crash_enable_local_vmclear(cpu);
730 }
731 
732 void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs)
733 {
734 	int cpu = loaded_vmcs->cpu;
735 
736 	if (cpu != -1)
737 		smp_call_function_single(cpu,
738 			 __loaded_vmcs_clear, loaded_vmcs, 1);
739 }
740 
741 static bool vmx_segment_cache_test_set(struct vcpu_vmx *vmx, unsigned seg,
742 				       unsigned field)
743 {
744 	bool ret;
745 	u32 mask = 1 << (seg * SEG_FIELD_NR + field);
746 
747 	if (!kvm_register_is_available(&vmx->vcpu, VCPU_EXREG_SEGMENTS)) {
748 		kvm_register_mark_available(&vmx->vcpu, VCPU_EXREG_SEGMENTS);
749 		vmx->segment_cache.bitmask = 0;
750 	}
751 	ret = vmx->segment_cache.bitmask & mask;
752 	vmx->segment_cache.bitmask |= mask;
753 	return ret;
754 }
755 
756 static u16 vmx_read_guest_seg_selector(struct vcpu_vmx *vmx, unsigned seg)
757 {
758 	u16 *p = &vmx->segment_cache.seg[seg].selector;
759 
760 	if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_SEL))
761 		*p = vmcs_read16(kvm_vmx_segment_fields[seg].selector);
762 	return *p;
763 }
764 
765 static ulong vmx_read_guest_seg_base(struct vcpu_vmx *vmx, unsigned seg)
766 {
767 	ulong *p = &vmx->segment_cache.seg[seg].base;
768 
769 	if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_BASE))
770 		*p = vmcs_readl(kvm_vmx_segment_fields[seg].base);
771 	return *p;
772 }
773 
774 static u32 vmx_read_guest_seg_limit(struct vcpu_vmx *vmx, unsigned seg)
775 {
776 	u32 *p = &vmx->segment_cache.seg[seg].limit;
777 
778 	if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_LIMIT))
779 		*p = vmcs_read32(kvm_vmx_segment_fields[seg].limit);
780 	return *p;
781 }
782 
783 static u32 vmx_read_guest_seg_ar(struct vcpu_vmx *vmx, unsigned seg)
784 {
785 	u32 *p = &vmx->segment_cache.seg[seg].ar;
786 
787 	if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_AR))
788 		*p = vmcs_read32(kvm_vmx_segment_fields[seg].ar_bytes);
789 	return *p;
790 }
791 
792 void update_exception_bitmap(struct kvm_vcpu *vcpu)
793 {
794 	u32 eb;
795 
796 	eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR) |
797 	     (1u << DB_VECTOR) | (1u << AC_VECTOR);
798 	/*
799 	 * Guest access to VMware backdoor ports could legitimately
800 	 * trigger #GP because of TSS I/O permission bitmap.
801 	 * We intercept those #GP and allow access to them anyway
802 	 * as VMware does.
803 	 */
804 	if (enable_vmware_backdoor)
805 		eb |= (1u << GP_VECTOR);
806 	if ((vcpu->guest_debug &
807 	     (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) ==
808 	    (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP))
809 		eb |= 1u << BP_VECTOR;
810 	if (to_vmx(vcpu)->rmode.vm86_active)
811 		eb = ~0;
812 	if (enable_ept)
813 		eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */
814 
815 	/* When we are running a nested L2 guest and L1 specified for it a
816 	 * certain exception bitmap, we must trap the same exceptions and pass
817 	 * them to L1. When running L2, we will only handle the exceptions
818 	 * specified above if L1 did not want them.
819 	 */
820 	if (is_guest_mode(vcpu))
821 		eb |= get_vmcs12(vcpu)->exception_bitmap;
822 
823 	vmcs_write32(EXCEPTION_BITMAP, eb);
824 }
825 
826 /*
827  * Check if MSR is intercepted for currently loaded MSR bitmap.
828  */
829 static bool msr_write_intercepted(struct kvm_vcpu *vcpu, u32 msr)
830 {
831 	unsigned long *msr_bitmap;
832 	int f = sizeof(unsigned long);
833 
834 	if (!cpu_has_vmx_msr_bitmap())
835 		return true;
836 
837 	msr_bitmap = to_vmx(vcpu)->loaded_vmcs->msr_bitmap;
838 
839 	if (msr <= 0x1fff) {
840 		return !!test_bit(msr, msr_bitmap + 0x800 / f);
841 	} else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
842 		msr &= 0x1fff;
843 		return !!test_bit(msr, msr_bitmap + 0xc00 / f);
844 	}
845 
846 	return true;
847 }
848 
849 static void clear_atomic_switch_msr_special(struct vcpu_vmx *vmx,
850 		unsigned long entry, unsigned long exit)
851 {
852 	vm_entry_controls_clearbit(vmx, entry);
853 	vm_exit_controls_clearbit(vmx, exit);
854 }
855 
856 int vmx_find_msr_index(struct vmx_msrs *m, u32 msr)
857 {
858 	unsigned int i;
859 
860 	for (i = 0; i < m->nr; ++i) {
861 		if (m->val[i].index == msr)
862 			return i;
863 	}
864 	return -ENOENT;
865 }
866 
867 static void clear_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr)
868 {
869 	int i;
870 	struct msr_autoload *m = &vmx->msr_autoload;
871 
872 	switch (msr) {
873 	case MSR_EFER:
874 		if (cpu_has_load_ia32_efer()) {
875 			clear_atomic_switch_msr_special(vmx,
876 					VM_ENTRY_LOAD_IA32_EFER,
877 					VM_EXIT_LOAD_IA32_EFER);
878 			return;
879 		}
880 		break;
881 	case MSR_CORE_PERF_GLOBAL_CTRL:
882 		if (cpu_has_load_perf_global_ctrl()) {
883 			clear_atomic_switch_msr_special(vmx,
884 					VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
885 					VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
886 			return;
887 		}
888 		break;
889 	}
890 	i = vmx_find_msr_index(&m->guest, msr);
891 	if (i < 0)
892 		goto skip_guest;
893 	--m->guest.nr;
894 	m->guest.val[i] = m->guest.val[m->guest.nr];
895 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->guest.nr);
896 
897 skip_guest:
898 	i = vmx_find_msr_index(&m->host, msr);
899 	if (i < 0)
900 		return;
901 
902 	--m->host.nr;
903 	m->host.val[i] = m->host.val[m->host.nr];
904 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->host.nr);
905 }
906 
907 static void add_atomic_switch_msr_special(struct vcpu_vmx *vmx,
908 		unsigned long entry, unsigned long exit,
909 		unsigned long guest_val_vmcs, unsigned long host_val_vmcs,
910 		u64 guest_val, u64 host_val)
911 {
912 	vmcs_write64(guest_val_vmcs, guest_val);
913 	if (host_val_vmcs != HOST_IA32_EFER)
914 		vmcs_write64(host_val_vmcs, host_val);
915 	vm_entry_controls_setbit(vmx, entry);
916 	vm_exit_controls_setbit(vmx, exit);
917 }
918 
919 static void add_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr,
920 				  u64 guest_val, u64 host_val, bool entry_only)
921 {
922 	int i, j = 0;
923 	struct msr_autoload *m = &vmx->msr_autoload;
924 
925 	switch (msr) {
926 	case MSR_EFER:
927 		if (cpu_has_load_ia32_efer()) {
928 			add_atomic_switch_msr_special(vmx,
929 					VM_ENTRY_LOAD_IA32_EFER,
930 					VM_EXIT_LOAD_IA32_EFER,
931 					GUEST_IA32_EFER,
932 					HOST_IA32_EFER,
933 					guest_val, host_val);
934 			return;
935 		}
936 		break;
937 	case MSR_CORE_PERF_GLOBAL_CTRL:
938 		if (cpu_has_load_perf_global_ctrl()) {
939 			add_atomic_switch_msr_special(vmx,
940 					VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
941 					VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL,
942 					GUEST_IA32_PERF_GLOBAL_CTRL,
943 					HOST_IA32_PERF_GLOBAL_CTRL,
944 					guest_val, host_val);
945 			return;
946 		}
947 		break;
948 	case MSR_IA32_PEBS_ENABLE:
949 		/* PEBS needs a quiescent period after being disabled (to write
950 		 * a record).  Disabling PEBS through VMX MSR swapping doesn't
951 		 * provide that period, so a CPU could write host's record into
952 		 * guest's memory.
953 		 */
954 		wrmsrl(MSR_IA32_PEBS_ENABLE, 0);
955 	}
956 
957 	i = vmx_find_msr_index(&m->guest, msr);
958 	if (!entry_only)
959 		j = vmx_find_msr_index(&m->host, msr);
960 
961 	if ((i < 0 && m->guest.nr == NR_LOADSTORE_MSRS) ||
962 		(j < 0 &&  m->host.nr == NR_LOADSTORE_MSRS)) {
963 		printk_once(KERN_WARNING "Not enough msr switch entries. "
964 				"Can't add msr %x\n", msr);
965 		return;
966 	}
967 	if (i < 0) {
968 		i = m->guest.nr++;
969 		vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->guest.nr);
970 	}
971 	m->guest.val[i].index = msr;
972 	m->guest.val[i].value = guest_val;
973 
974 	if (entry_only)
975 		return;
976 
977 	if (j < 0) {
978 		j = m->host.nr++;
979 		vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->host.nr);
980 	}
981 	m->host.val[j].index = msr;
982 	m->host.val[j].value = host_val;
983 }
984 
985 static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset)
986 {
987 	u64 guest_efer = vmx->vcpu.arch.efer;
988 	u64 ignore_bits = 0;
989 
990 	/* Shadow paging assumes NX to be available.  */
991 	if (!enable_ept)
992 		guest_efer |= EFER_NX;
993 
994 	/*
995 	 * LMA and LME handled by hardware; SCE meaningless outside long mode.
996 	 */
997 	ignore_bits |= EFER_SCE;
998 #ifdef CONFIG_X86_64
999 	ignore_bits |= EFER_LMA | EFER_LME;
1000 	/* SCE is meaningful only in long mode on Intel */
1001 	if (guest_efer & EFER_LMA)
1002 		ignore_bits &= ~(u64)EFER_SCE;
1003 #endif
1004 
1005 	/*
1006 	 * On EPT, we can't emulate NX, so we must switch EFER atomically.
1007 	 * On CPUs that support "load IA32_EFER", always switch EFER
1008 	 * atomically, since it's faster than switching it manually.
1009 	 */
1010 	if (cpu_has_load_ia32_efer() ||
1011 	    (enable_ept && ((vmx->vcpu.arch.efer ^ host_efer) & EFER_NX))) {
1012 		if (!(guest_efer & EFER_LMA))
1013 			guest_efer &= ~EFER_LME;
1014 		if (guest_efer != host_efer)
1015 			add_atomic_switch_msr(vmx, MSR_EFER,
1016 					      guest_efer, host_efer, false);
1017 		else
1018 			clear_atomic_switch_msr(vmx, MSR_EFER);
1019 		return false;
1020 	} else {
1021 		clear_atomic_switch_msr(vmx, MSR_EFER);
1022 
1023 		guest_efer &= ~ignore_bits;
1024 		guest_efer |= host_efer & ignore_bits;
1025 
1026 		vmx->guest_msrs[efer_offset].data = guest_efer;
1027 		vmx->guest_msrs[efer_offset].mask = ~ignore_bits;
1028 
1029 		return true;
1030 	}
1031 }
1032 
1033 #ifdef CONFIG_X86_32
1034 /*
1035  * On 32-bit kernels, VM exits still load the FS and GS bases from the
1036  * VMCS rather than the segment table.  KVM uses this helper to figure
1037  * out the current bases to poke them into the VMCS before entry.
1038  */
1039 static unsigned long segment_base(u16 selector)
1040 {
1041 	struct desc_struct *table;
1042 	unsigned long v;
1043 
1044 	if (!(selector & ~SEGMENT_RPL_MASK))
1045 		return 0;
1046 
1047 	table = get_current_gdt_ro();
1048 
1049 	if ((selector & SEGMENT_TI_MASK) == SEGMENT_LDT) {
1050 		u16 ldt_selector = kvm_read_ldt();
1051 
1052 		if (!(ldt_selector & ~SEGMENT_RPL_MASK))
1053 			return 0;
1054 
1055 		table = (struct desc_struct *)segment_base(ldt_selector);
1056 	}
1057 	v = get_desc_base(&table[selector >> 3]);
1058 	return v;
1059 }
1060 #endif
1061 
1062 static inline bool pt_can_write_msr(struct vcpu_vmx *vmx)
1063 {
1064 	return (pt_mode == PT_MODE_HOST_GUEST) &&
1065 	       !(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN);
1066 }
1067 
1068 static inline void pt_load_msr(struct pt_ctx *ctx, u32 addr_range)
1069 {
1070 	u32 i;
1071 
1072 	wrmsrl(MSR_IA32_RTIT_STATUS, ctx->status);
1073 	wrmsrl(MSR_IA32_RTIT_OUTPUT_BASE, ctx->output_base);
1074 	wrmsrl(MSR_IA32_RTIT_OUTPUT_MASK, ctx->output_mask);
1075 	wrmsrl(MSR_IA32_RTIT_CR3_MATCH, ctx->cr3_match);
1076 	for (i = 0; i < addr_range; i++) {
1077 		wrmsrl(MSR_IA32_RTIT_ADDR0_A + i * 2, ctx->addr_a[i]);
1078 		wrmsrl(MSR_IA32_RTIT_ADDR0_B + i * 2, ctx->addr_b[i]);
1079 	}
1080 }
1081 
1082 static inline void pt_save_msr(struct pt_ctx *ctx, u32 addr_range)
1083 {
1084 	u32 i;
1085 
1086 	rdmsrl(MSR_IA32_RTIT_STATUS, ctx->status);
1087 	rdmsrl(MSR_IA32_RTIT_OUTPUT_BASE, ctx->output_base);
1088 	rdmsrl(MSR_IA32_RTIT_OUTPUT_MASK, ctx->output_mask);
1089 	rdmsrl(MSR_IA32_RTIT_CR3_MATCH, ctx->cr3_match);
1090 	for (i = 0; i < addr_range; i++) {
1091 		rdmsrl(MSR_IA32_RTIT_ADDR0_A + i * 2, ctx->addr_a[i]);
1092 		rdmsrl(MSR_IA32_RTIT_ADDR0_B + i * 2, ctx->addr_b[i]);
1093 	}
1094 }
1095 
1096 static void pt_guest_enter(struct vcpu_vmx *vmx)
1097 {
1098 	if (pt_mode == PT_MODE_SYSTEM)
1099 		return;
1100 
1101 	/*
1102 	 * GUEST_IA32_RTIT_CTL is already set in the VMCS.
1103 	 * Save host state before VM entry.
1104 	 */
1105 	rdmsrl(MSR_IA32_RTIT_CTL, vmx->pt_desc.host.ctl);
1106 	if (vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) {
1107 		wrmsrl(MSR_IA32_RTIT_CTL, 0);
1108 		pt_save_msr(&vmx->pt_desc.host, vmx->pt_desc.addr_range);
1109 		pt_load_msr(&vmx->pt_desc.guest, vmx->pt_desc.addr_range);
1110 	}
1111 }
1112 
1113 static void pt_guest_exit(struct vcpu_vmx *vmx)
1114 {
1115 	if (pt_mode == PT_MODE_SYSTEM)
1116 		return;
1117 
1118 	if (vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) {
1119 		pt_save_msr(&vmx->pt_desc.guest, vmx->pt_desc.addr_range);
1120 		pt_load_msr(&vmx->pt_desc.host, vmx->pt_desc.addr_range);
1121 	}
1122 
1123 	/* Reload host state (IA32_RTIT_CTL will be cleared on VM exit). */
1124 	wrmsrl(MSR_IA32_RTIT_CTL, vmx->pt_desc.host.ctl);
1125 }
1126 
1127 void vmx_set_host_fs_gs(struct vmcs_host_state *host, u16 fs_sel, u16 gs_sel,
1128 			unsigned long fs_base, unsigned long gs_base)
1129 {
1130 	if (unlikely(fs_sel != host->fs_sel)) {
1131 		if (!(fs_sel & 7))
1132 			vmcs_write16(HOST_FS_SELECTOR, fs_sel);
1133 		else
1134 			vmcs_write16(HOST_FS_SELECTOR, 0);
1135 		host->fs_sel = fs_sel;
1136 	}
1137 	if (unlikely(gs_sel != host->gs_sel)) {
1138 		if (!(gs_sel & 7))
1139 			vmcs_write16(HOST_GS_SELECTOR, gs_sel);
1140 		else
1141 			vmcs_write16(HOST_GS_SELECTOR, 0);
1142 		host->gs_sel = gs_sel;
1143 	}
1144 	if (unlikely(fs_base != host->fs_base)) {
1145 		vmcs_writel(HOST_FS_BASE, fs_base);
1146 		host->fs_base = fs_base;
1147 	}
1148 	if (unlikely(gs_base != host->gs_base)) {
1149 		vmcs_writel(HOST_GS_BASE, gs_base);
1150 		host->gs_base = gs_base;
1151 	}
1152 }
1153 
1154 void vmx_prepare_switch_to_guest(struct kvm_vcpu *vcpu)
1155 {
1156 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1157 	struct vmcs_host_state *host_state;
1158 #ifdef CONFIG_X86_64
1159 	int cpu = raw_smp_processor_id();
1160 #endif
1161 	unsigned long fs_base, gs_base;
1162 	u16 fs_sel, gs_sel;
1163 	int i;
1164 
1165 	vmx->req_immediate_exit = false;
1166 
1167 	/*
1168 	 * Note that guest MSRs to be saved/restored can also be changed
1169 	 * when guest state is loaded. This happens when guest transitions
1170 	 * to/from long-mode by setting MSR_EFER.LMA.
1171 	 */
1172 	if (!vmx->guest_msrs_ready) {
1173 		vmx->guest_msrs_ready = true;
1174 		for (i = 0; i < vmx->save_nmsrs; ++i)
1175 			kvm_set_shared_msr(vmx->guest_msrs[i].index,
1176 					   vmx->guest_msrs[i].data,
1177 					   vmx->guest_msrs[i].mask);
1178 
1179 	}
1180 
1181     	if (vmx->nested.need_vmcs12_to_shadow_sync)
1182 		nested_sync_vmcs12_to_shadow(vcpu);
1183 
1184 	if (vmx->guest_state_loaded)
1185 		return;
1186 
1187 	host_state = &vmx->loaded_vmcs->host_state;
1188 
1189 	/*
1190 	 * Set host fs and gs selectors.  Unfortunately, 22.2.3 does not
1191 	 * allow segment selectors with cpl > 0 or ti == 1.
1192 	 */
1193 	host_state->ldt_sel = kvm_read_ldt();
1194 
1195 #ifdef CONFIG_X86_64
1196 	savesegment(ds, host_state->ds_sel);
1197 	savesegment(es, host_state->es_sel);
1198 
1199 	gs_base = cpu_kernelmode_gs_base(cpu);
1200 	if (likely(is_64bit_mm(current->mm))) {
1201 		save_fsgs_for_kvm();
1202 		fs_sel = current->thread.fsindex;
1203 		gs_sel = current->thread.gsindex;
1204 		fs_base = current->thread.fsbase;
1205 		vmx->msr_host_kernel_gs_base = current->thread.gsbase;
1206 	} else {
1207 		savesegment(fs, fs_sel);
1208 		savesegment(gs, gs_sel);
1209 		fs_base = read_msr(MSR_FS_BASE);
1210 		vmx->msr_host_kernel_gs_base = read_msr(MSR_KERNEL_GS_BASE);
1211 	}
1212 
1213 	wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1214 #else
1215 	savesegment(fs, fs_sel);
1216 	savesegment(gs, gs_sel);
1217 	fs_base = segment_base(fs_sel);
1218 	gs_base = segment_base(gs_sel);
1219 #endif
1220 
1221 	vmx_set_host_fs_gs(host_state, fs_sel, gs_sel, fs_base, gs_base);
1222 	vmx->guest_state_loaded = true;
1223 }
1224 
1225 static void vmx_prepare_switch_to_host(struct vcpu_vmx *vmx)
1226 {
1227 	struct vmcs_host_state *host_state;
1228 
1229 	if (!vmx->guest_state_loaded)
1230 		return;
1231 
1232 	host_state = &vmx->loaded_vmcs->host_state;
1233 
1234 	++vmx->vcpu.stat.host_state_reload;
1235 
1236 #ifdef CONFIG_X86_64
1237 	rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1238 #endif
1239 	if (host_state->ldt_sel || (host_state->gs_sel & 7)) {
1240 		kvm_load_ldt(host_state->ldt_sel);
1241 #ifdef CONFIG_X86_64
1242 		load_gs_index(host_state->gs_sel);
1243 #else
1244 		loadsegment(gs, host_state->gs_sel);
1245 #endif
1246 	}
1247 	if (host_state->fs_sel & 7)
1248 		loadsegment(fs, host_state->fs_sel);
1249 #ifdef CONFIG_X86_64
1250 	if (unlikely(host_state->ds_sel | host_state->es_sel)) {
1251 		loadsegment(ds, host_state->ds_sel);
1252 		loadsegment(es, host_state->es_sel);
1253 	}
1254 #endif
1255 	invalidate_tss_limit();
1256 #ifdef CONFIG_X86_64
1257 	wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
1258 #endif
1259 	load_fixmap_gdt(raw_smp_processor_id());
1260 	vmx->guest_state_loaded = false;
1261 	vmx->guest_msrs_ready = false;
1262 }
1263 
1264 #ifdef CONFIG_X86_64
1265 static u64 vmx_read_guest_kernel_gs_base(struct vcpu_vmx *vmx)
1266 {
1267 	preempt_disable();
1268 	if (vmx->guest_state_loaded)
1269 		rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1270 	preempt_enable();
1271 	return vmx->msr_guest_kernel_gs_base;
1272 }
1273 
1274 static void vmx_write_guest_kernel_gs_base(struct vcpu_vmx *vmx, u64 data)
1275 {
1276 	preempt_disable();
1277 	if (vmx->guest_state_loaded)
1278 		wrmsrl(MSR_KERNEL_GS_BASE, data);
1279 	preempt_enable();
1280 	vmx->msr_guest_kernel_gs_base = data;
1281 }
1282 #endif
1283 
1284 static void vmx_vcpu_pi_load(struct kvm_vcpu *vcpu, int cpu)
1285 {
1286 	struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
1287 	struct pi_desc old, new;
1288 	unsigned int dest;
1289 
1290 	/*
1291 	 * In case of hot-plug or hot-unplug, we may have to undo
1292 	 * vmx_vcpu_pi_put even if there is no assigned device.  And we
1293 	 * always keep PI.NDST up to date for simplicity: it makes the
1294 	 * code easier, and CPU migration is not a fast path.
1295 	 */
1296 	if (!pi_test_sn(pi_desc) && vcpu->cpu == cpu)
1297 		return;
1298 
1299 	/*
1300 	 * If the 'nv' field is POSTED_INTR_WAKEUP_VECTOR, do not change
1301 	 * PI.NDST: pi_post_block is the one expected to change PID.NDST and the
1302 	 * wakeup handler expects the vCPU to be on the blocked_vcpu_list that
1303 	 * matches PI.NDST. Otherwise, a vcpu may not be able to be woken up
1304 	 * correctly.
1305 	 */
1306 	if (pi_desc->nv == POSTED_INTR_WAKEUP_VECTOR || vcpu->cpu == cpu) {
1307 		pi_clear_sn(pi_desc);
1308 		goto after_clear_sn;
1309 	}
1310 
1311 	/* The full case.  */
1312 	do {
1313 		old.control = new.control = pi_desc->control;
1314 
1315 		dest = cpu_physical_id(cpu);
1316 
1317 		if (x2apic_enabled())
1318 			new.ndst = dest;
1319 		else
1320 			new.ndst = (dest << 8) & 0xFF00;
1321 
1322 		new.sn = 0;
1323 	} while (cmpxchg64(&pi_desc->control, old.control,
1324 			   new.control) != old.control);
1325 
1326 after_clear_sn:
1327 
1328 	/*
1329 	 * Clear SN before reading the bitmap.  The VT-d firmware
1330 	 * writes the bitmap and reads SN atomically (5.2.3 in the
1331 	 * spec), so it doesn't really have a memory barrier that
1332 	 * pairs with this, but we cannot do that and we need one.
1333 	 */
1334 	smp_mb__after_atomic();
1335 
1336 	if (!pi_is_pir_empty(pi_desc))
1337 		pi_set_on(pi_desc);
1338 }
1339 
1340 void vmx_vcpu_load_vmcs(struct kvm_vcpu *vcpu, int cpu)
1341 {
1342 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1343 	bool already_loaded = vmx->loaded_vmcs->cpu == cpu;
1344 
1345 	if (!already_loaded) {
1346 		loaded_vmcs_clear(vmx->loaded_vmcs);
1347 		local_irq_disable();
1348 		crash_disable_local_vmclear(cpu);
1349 
1350 		/*
1351 		 * Read loaded_vmcs->cpu should be before fetching
1352 		 * loaded_vmcs->loaded_vmcss_on_cpu_link.
1353 		 * See the comments in __loaded_vmcs_clear().
1354 		 */
1355 		smp_rmb();
1356 
1357 		list_add(&vmx->loaded_vmcs->loaded_vmcss_on_cpu_link,
1358 			 &per_cpu(loaded_vmcss_on_cpu, cpu));
1359 		crash_enable_local_vmclear(cpu);
1360 		local_irq_enable();
1361 	}
1362 
1363 	if (per_cpu(current_vmcs, cpu) != vmx->loaded_vmcs->vmcs) {
1364 		per_cpu(current_vmcs, cpu) = vmx->loaded_vmcs->vmcs;
1365 		vmcs_load(vmx->loaded_vmcs->vmcs);
1366 		indirect_branch_prediction_barrier();
1367 	}
1368 
1369 	if (!already_loaded) {
1370 		void *gdt = get_current_gdt_ro();
1371 		unsigned long sysenter_esp;
1372 
1373 		kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1374 
1375 		/*
1376 		 * Linux uses per-cpu TSS and GDT, so set these when switching
1377 		 * processors.  See 22.2.4.
1378 		 */
1379 		vmcs_writel(HOST_TR_BASE,
1380 			    (unsigned long)&get_cpu_entry_area(cpu)->tss.x86_tss);
1381 		vmcs_writel(HOST_GDTR_BASE, (unsigned long)gdt);   /* 22.2.4 */
1382 
1383 		rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
1384 		vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
1385 
1386 		vmx->loaded_vmcs->cpu = cpu;
1387 	}
1388 
1389 	/* Setup TSC multiplier */
1390 	if (kvm_has_tsc_control &&
1391 	    vmx->current_tsc_ratio != vcpu->arch.tsc_scaling_ratio)
1392 		decache_tsc_multiplier(vmx);
1393 }
1394 
1395 /*
1396  * Switches to specified vcpu, until a matching vcpu_put(), but assumes
1397  * vcpu mutex is already taken.
1398  */
1399 void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1400 {
1401 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1402 
1403 	vmx_vcpu_load_vmcs(vcpu, cpu);
1404 
1405 	vmx_vcpu_pi_load(vcpu, cpu);
1406 
1407 	vmx->host_pkru = read_pkru();
1408 	vmx->host_debugctlmsr = get_debugctlmsr();
1409 }
1410 
1411 static void vmx_vcpu_pi_put(struct kvm_vcpu *vcpu)
1412 {
1413 	struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
1414 
1415 	if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
1416 		!irq_remapping_cap(IRQ_POSTING_CAP)  ||
1417 		!kvm_vcpu_apicv_active(vcpu))
1418 		return;
1419 
1420 	/* Set SN when the vCPU is preempted */
1421 	if (vcpu->preempted)
1422 		pi_set_sn(pi_desc);
1423 }
1424 
1425 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
1426 {
1427 	vmx_vcpu_pi_put(vcpu);
1428 
1429 	vmx_prepare_switch_to_host(to_vmx(vcpu));
1430 }
1431 
1432 static bool emulation_required(struct kvm_vcpu *vcpu)
1433 {
1434 	return emulate_invalid_guest_state && !guest_state_valid(vcpu);
1435 }
1436 
1437 unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
1438 {
1439 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1440 	unsigned long rflags, save_rflags;
1441 
1442 	if (!kvm_register_is_available(vcpu, VCPU_EXREG_RFLAGS)) {
1443 		kvm_register_mark_available(vcpu, VCPU_EXREG_RFLAGS);
1444 		rflags = vmcs_readl(GUEST_RFLAGS);
1445 		if (vmx->rmode.vm86_active) {
1446 			rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
1447 			save_rflags = vmx->rmode.save_rflags;
1448 			rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
1449 		}
1450 		vmx->rflags = rflags;
1451 	}
1452 	return vmx->rflags;
1453 }
1454 
1455 void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
1456 {
1457 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1458 	unsigned long old_rflags;
1459 
1460 	if (enable_unrestricted_guest) {
1461 		kvm_register_mark_available(vcpu, VCPU_EXREG_RFLAGS);
1462 		vmx->rflags = rflags;
1463 		vmcs_writel(GUEST_RFLAGS, rflags);
1464 		return;
1465 	}
1466 
1467 	old_rflags = vmx_get_rflags(vcpu);
1468 	vmx->rflags = rflags;
1469 	if (vmx->rmode.vm86_active) {
1470 		vmx->rmode.save_rflags = rflags;
1471 		rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
1472 	}
1473 	vmcs_writel(GUEST_RFLAGS, rflags);
1474 
1475 	if ((old_rflags ^ vmx->rflags) & X86_EFLAGS_VM)
1476 		vmx->emulation_required = emulation_required(vcpu);
1477 }
1478 
1479 u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu)
1480 {
1481 	u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1482 	int ret = 0;
1483 
1484 	if (interruptibility & GUEST_INTR_STATE_STI)
1485 		ret |= KVM_X86_SHADOW_INT_STI;
1486 	if (interruptibility & GUEST_INTR_STATE_MOV_SS)
1487 		ret |= KVM_X86_SHADOW_INT_MOV_SS;
1488 
1489 	return ret;
1490 }
1491 
1492 void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
1493 {
1494 	u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1495 	u32 interruptibility = interruptibility_old;
1496 
1497 	interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
1498 
1499 	if (mask & KVM_X86_SHADOW_INT_MOV_SS)
1500 		interruptibility |= GUEST_INTR_STATE_MOV_SS;
1501 	else if (mask & KVM_X86_SHADOW_INT_STI)
1502 		interruptibility |= GUEST_INTR_STATE_STI;
1503 
1504 	if ((interruptibility != interruptibility_old))
1505 		vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
1506 }
1507 
1508 static int vmx_rtit_ctl_check(struct kvm_vcpu *vcpu, u64 data)
1509 {
1510 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1511 	unsigned long value;
1512 
1513 	/*
1514 	 * Any MSR write that attempts to change bits marked reserved will
1515 	 * case a #GP fault.
1516 	 */
1517 	if (data & vmx->pt_desc.ctl_bitmask)
1518 		return 1;
1519 
1520 	/*
1521 	 * Any attempt to modify IA32_RTIT_CTL while TraceEn is set will
1522 	 * result in a #GP unless the same write also clears TraceEn.
1523 	 */
1524 	if ((vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) &&
1525 		((vmx->pt_desc.guest.ctl ^ data) & ~RTIT_CTL_TRACEEN))
1526 		return 1;
1527 
1528 	/*
1529 	 * WRMSR to IA32_RTIT_CTL that sets TraceEn but clears this bit
1530 	 * and FabricEn would cause #GP, if
1531 	 * CPUID.(EAX=14H, ECX=0):ECX.SNGLRGNOUT[bit 2] = 0
1532 	 */
1533 	if ((data & RTIT_CTL_TRACEEN) && !(data & RTIT_CTL_TOPA) &&
1534 		!(data & RTIT_CTL_FABRIC_EN) &&
1535 		!intel_pt_validate_cap(vmx->pt_desc.caps,
1536 					PT_CAP_single_range_output))
1537 		return 1;
1538 
1539 	/*
1540 	 * MTCFreq, CycThresh and PSBFreq encodings check, any MSR write that
1541 	 * utilize encodings marked reserved will casue a #GP fault.
1542 	 */
1543 	value = intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc_periods);
1544 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc) &&
1545 			!test_bit((data & RTIT_CTL_MTC_RANGE) >>
1546 			RTIT_CTL_MTC_RANGE_OFFSET, &value))
1547 		return 1;
1548 	value = intel_pt_validate_cap(vmx->pt_desc.caps,
1549 						PT_CAP_cycle_thresholds);
1550 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc) &&
1551 			!test_bit((data & RTIT_CTL_CYC_THRESH) >>
1552 			RTIT_CTL_CYC_THRESH_OFFSET, &value))
1553 		return 1;
1554 	value = intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_periods);
1555 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc) &&
1556 			!test_bit((data & RTIT_CTL_PSB_FREQ) >>
1557 			RTIT_CTL_PSB_FREQ_OFFSET, &value))
1558 		return 1;
1559 
1560 	/*
1561 	 * If ADDRx_CFG is reserved or the encodings is >2 will
1562 	 * cause a #GP fault.
1563 	 */
1564 	value = (data & RTIT_CTL_ADDR0) >> RTIT_CTL_ADDR0_OFFSET;
1565 	if ((value && (vmx->pt_desc.addr_range < 1)) || (value > 2))
1566 		return 1;
1567 	value = (data & RTIT_CTL_ADDR1) >> RTIT_CTL_ADDR1_OFFSET;
1568 	if ((value && (vmx->pt_desc.addr_range < 2)) || (value > 2))
1569 		return 1;
1570 	value = (data & RTIT_CTL_ADDR2) >> RTIT_CTL_ADDR2_OFFSET;
1571 	if ((value && (vmx->pt_desc.addr_range < 3)) || (value > 2))
1572 		return 1;
1573 	value = (data & RTIT_CTL_ADDR3) >> RTIT_CTL_ADDR3_OFFSET;
1574 	if ((value && (vmx->pt_desc.addr_range < 4)) || (value > 2))
1575 		return 1;
1576 
1577 	return 0;
1578 }
1579 
1580 static int skip_emulated_instruction(struct kvm_vcpu *vcpu)
1581 {
1582 	unsigned long rip;
1583 
1584 	/*
1585 	 * Using VMCS.VM_EXIT_INSTRUCTION_LEN on EPT misconfig depends on
1586 	 * undefined behavior: Intel's SDM doesn't mandate the VMCS field be
1587 	 * set when EPT misconfig occurs.  In practice, real hardware updates
1588 	 * VM_EXIT_INSTRUCTION_LEN on EPT misconfig, but other hypervisors
1589 	 * (namely Hyper-V) don't set it due to it being undefined behavior,
1590 	 * i.e. we end up advancing IP with some random value.
1591 	 */
1592 	if (!static_cpu_has(X86_FEATURE_HYPERVISOR) ||
1593 	    to_vmx(vcpu)->exit_reason != EXIT_REASON_EPT_MISCONFIG) {
1594 		rip = kvm_rip_read(vcpu);
1595 		rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
1596 		kvm_rip_write(vcpu, rip);
1597 	} else {
1598 		if (!kvm_emulate_instruction(vcpu, EMULTYPE_SKIP))
1599 			return 0;
1600 	}
1601 
1602 	/* skipping an emulated instruction also counts */
1603 	vmx_set_interrupt_shadow(vcpu, 0);
1604 
1605 	return 1;
1606 }
1607 
1608 
1609 /*
1610  * Recognizes a pending MTF VM-exit and records the nested state for later
1611  * delivery.
1612  */
1613 static void vmx_update_emulated_instruction(struct kvm_vcpu *vcpu)
1614 {
1615 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1616 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1617 
1618 	if (!is_guest_mode(vcpu))
1619 		return;
1620 
1621 	/*
1622 	 * Per the SDM, MTF takes priority over debug-trap exceptions besides
1623 	 * T-bit traps. As instruction emulation is completed (i.e. at the
1624 	 * instruction boundary), any #DB exception pending delivery must be a
1625 	 * debug-trap. Record the pending MTF state to be delivered in
1626 	 * vmx_check_nested_events().
1627 	 */
1628 	if (nested_cpu_has_mtf(vmcs12) &&
1629 	    (!vcpu->arch.exception.pending ||
1630 	     vcpu->arch.exception.nr == DB_VECTOR))
1631 		vmx->nested.mtf_pending = true;
1632 	else
1633 		vmx->nested.mtf_pending = false;
1634 }
1635 
1636 static int vmx_skip_emulated_instruction(struct kvm_vcpu *vcpu)
1637 {
1638 	vmx_update_emulated_instruction(vcpu);
1639 	return skip_emulated_instruction(vcpu);
1640 }
1641 
1642 static void vmx_clear_hlt(struct kvm_vcpu *vcpu)
1643 {
1644 	/*
1645 	 * Ensure that we clear the HLT state in the VMCS.  We don't need to
1646 	 * explicitly skip the instruction because if the HLT state is set,
1647 	 * then the instruction is already executing and RIP has already been
1648 	 * advanced.
1649 	 */
1650 	if (kvm_hlt_in_guest(vcpu->kvm) &&
1651 			vmcs_read32(GUEST_ACTIVITY_STATE) == GUEST_ACTIVITY_HLT)
1652 		vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
1653 }
1654 
1655 static void vmx_queue_exception(struct kvm_vcpu *vcpu)
1656 {
1657 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1658 	unsigned nr = vcpu->arch.exception.nr;
1659 	bool has_error_code = vcpu->arch.exception.has_error_code;
1660 	u32 error_code = vcpu->arch.exception.error_code;
1661 	u32 intr_info = nr | INTR_INFO_VALID_MASK;
1662 
1663 	kvm_deliver_exception_payload(vcpu);
1664 
1665 	if (has_error_code) {
1666 		vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
1667 		intr_info |= INTR_INFO_DELIVER_CODE_MASK;
1668 	}
1669 
1670 	if (vmx->rmode.vm86_active) {
1671 		int inc_eip = 0;
1672 		if (kvm_exception_is_soft(nr))
1673 			inc_eip = vcpu->arch.event_exit_inst_len;
1674 		kvm_inject_realmode_interrupt(vcpu, nr, inc_eip);
1675 		return;
1676 	}
1677 
1678 	WARN_ON_ONCE(vmx->emulation_required);
1679 
1680 	if (kvm_exception_is_soft(nr)) {
1681 		vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
1682 			     vmx->vcpu.arch.event_exit_inst_len);
1683 		intr_info |= INTR_TYPE_SOFT_EXCEPTION;
1684 	} else
1685 		intr_info |= INTR_TYPE_HARD_EXCEPTION;
1686 
1687 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
1688 
1689 	vmx_clear_hlt(vcpu);
1690 }
1691 
1692 static bool vmx_rdtscp_supported(void)
1693 {
1694 	return cpu_has_vmx_rdtscp();
1695 }
1696 
1697 static bool vmx_invpcid_supported(void)
1698 {
1699 	return cpu_has_vmx_invpcid();
1700 }
1701 
1702 /*
1703  * Swap MSR entry in host/guest MSR entry array.
1704  */
1705 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
1706 {
1707 	struct shared_msr_entry tmp;
1708 
1709 	tmp = vmx->guest_msrs[to];
1710 	vmx->guest_msrs[to] = vmx->guest_msrs[from];
1711 	vmx->guest_msrs[from] = tmp;
1712 }
1713 
1714 /*
1715  * Set up the vmcs to automatically save and restore system
1716  * msrs.  Don't touch the 64-bit msrs if the guest is in legacy
1717  * mode, as fiddling with msrs is very expensive.
1718  */
1719 static void setup_msrs(struct vcpu_vmx *vmx)
1720 {
1721 	int save_nmsrs, index;
1722 
1723 	save_nmsrs = 0;
1724 #ifdef CONFIG_X86_64
1725 	/*
1726 	 * The SYSCALL MSRs are only needed on long mode guests, and only
1727 	 * when EFER.SCE is set.
1728 	 */
1729 	if (is_long_mode(&vmx->vcpu) && (vmx->vcpu.arch.efer & EFER_SCE)) {
1730 		index = __find_msr_index(vmx, MSR_STAR);
1731 		if (index >= 0)
1732 			move_msr_up(vmx, index, save_nmsrs++);
1733 		index = __find_msr_index(vmx, MSR_LSTAR);
1734 		if (index >= 0)
1735 			move_msr_up(vmx, index, save_nmsrs++);
1736 		index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
1737 		if (index >= 0)
1738 			move_msr_up(vmx, index, save_nmsrs++);
1739 	}
1740 #endif
1741 	index = __find_msr_index(vmx, MSR_EFER);
1742 	if (index >= 0 && update_transition_efer(vmx, index))
1743 		move_msr_up(vmx, index, save_nmsrs++);
1744 	index = __find_msr_index(vmx, MSR_TSC_AUX);
1745 	if (index >= 0 && guest_cpuid_has(&vmx->vcpu, X86_FEATURE_RDTSCP))
1746 		move_msr_up(vmx, index, save_nmsrs++);
1747 	index = __find_msr_index(vmx, MSR_IA32_TSX_CTRL);
1748 	if (index >= 0)
1749 		move_msr_up(vmx, index, save_nmsrs++);
1750 
1751 	vmx->save_nmsrs = save_nmsrs;
1752 	vmx->guest_msrs_ready = false;
1753 
1754 	if (cpu_has_vmx_msr_bitmap())
1755 		vmx_update_msr_bitmap(&vmx->vcpu);
1756 }
1757 
1758 static u64 vmx_read_l1_tsc_offset(struct kvm_vcpu *vcpu)
1759 {
1760 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1761 
1762 	if (is_guest_mode(vcpu) &&
1763 	    (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETTING))
1764 		return vcpu->arch.tsc_offset - vmcs12->tsc_offset;
1765 
1766 	return vcpu->arch.tsc_offset;
1767 }
1768 
1769 static u64 vmx_write_l1_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1770 {
1771 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1772 	u64 g_tsc_offset = 0;
1773 
1774 	/*
1775 	 * We're here if L1 chose not to trap WRMSR to TSC. According
1776 	 * to the spec, this should set L1's TSC; The offset that L1
1777 	 * set for L2 remains unchanged, and still needs to be added
1778 	 * to the newly set TSC to get L2's TSC.
1779 	 */
1780 	if (is_guest_mode(vcpu) &&
1781 	    (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETTING))
1782 		g_tsc_offset = vmcs12->tsc_offset;
1783 
1784 	trace_kvm_write_tsc_offset(vcpu->vcpu_id,
1785 				   vcpu->arch.tsc_offset - g_tsc_offset,
1786 				   offset);
1787 	vmcs_write64(TSC_OFFSET, offset + g_tsc_offset);
1788 	return offset + g_tsc_offset;
1789 }
1790 
1791 /*
1792  * nested_vmx_allowed() checks whether a guest should be allowed to use VMX
1793  * instructions and MSRs (i.e., nested VMX). Nested VMX is disabled for
1794  * all guests if the "nested" module option is off, and can also be disabled
1795  * for a single guest by disabling its VMX cpuid bit.
1796  */
1797 bool nested_vmx_allowed(struct kvm_vcpu *vcpu)
1798 {
1799 	return nested && guest_cpuid_has(vcpu, X86_FEATURE_VMX);
1800 }
1801 
1802 static inline bool vmx_feature_control_msr_valid(struct kvm_vcpu *vcpu,
1803 						 uint64_t val)
1804 {
1805 	uint64_t valid_bits = to_vmx(vcpu)->msr_ia32_feature_control_valid_bits;
1806 
1807 	return !(val & ~valid_bits);
1808 }
1809 
1810 static int vmx_get_msr_feature(struct kvm_msr_entry *msr)
1811 {
1812 	switch (msr->index) {
1813 	case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
1814 		if (!nested)
1815 			return 1;
1816 		return vmx_get_vmx_msr(&vmcs_config.nested, msr->index, &msr->data);
1817 	default:
1818 		return 1;
1819 	}
1820 }
1821 
1822 /*
1823  * Reads an msr value (of 'msr_index') into 'pdata'.
1824  * Returns 0 on success, non-0 otherwise.
1825  * Assumes vcpu_load() was already called.
1826  */
1827 static int vmx_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1828 {
1829 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1830 	struct shared_msr_entry *msr;
1831 	u32 index;
1832 
1833 	switch (msr_info->index) {
1834 #ifdef CONFIG_X86_64
1835 	case MSR_FS_BASE:
1836 		msr_info->data = vmcs_readl(GUEST_FS_BASE);
1837 		break;
1838 	case MSR_GS_BASE:
1839 		msr_info->data = vmcs_readl(GUEST_GS_BASE);
1840 		break;
1841 	case MSR_KERNEL_GS_BASE:
1842 		msr_info->data = vmx_read_guest_kernel_gs_base(vmx);
1843 		break;
1844 #endif
1845 	case MSR_EFER:
1846 		return kvm_get_msr_common(vcpu, msr_info);
1847 	case MSR_IA32_TSX_CTRL:
1848 		if (!msr_info->host_initiated &&
1849 		    !(vcpu->arch.arch_capabilities & ARCH_CAP_TSX_CTRL_MSR))
1850 			return 1;
1851 		goto find_shared_msr;
1852 	case MSR_IA32_UMWAIT_CONTROL:
1853 		if (!msr_info->host_initiated && !vmx_has_waitpkg(vmx))
1854 			return 1;
1855 
1856 		msr_info->data = vmx->msr_ia32_umwait_control;
1857 		break;
1858 	case MSR_IA32_SPEC_CTRL:
1859 		if (!msr_info->host_initiated &&
1860 		    !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
1861 			return 1;
1862 
1863 		msr_info->data = to_vmx(vcpu)->spec_ctrl;
1864 		break;
1865 	case MSR_IA32_SYSENTER_CS:
1866 		msr_info->data = vmcs_read32(GUEST_SYSENTER_CS);
1867 		break;
1868 	case MSR_IA32_SYSENTER_EIP:
1869 		msr_info->data = vmcs_readl(GUEST_SYSENTER_EIP);
1870 		break;
1871 	case MSR_IA32_SYSENTER_ESP:
1872 		msr_info->data = vmcs_readl(GUEST_SYSENTER_ESP);
1873 		break;
1874 	case MSR_IA32_BNDCFGS:
1875 		if (!kvm_mpx_supported() ||
1876 		    (!msr_info->host_initiated &&
1877 		     !guest_cpuid_has(vcpu, X86_FEATURE_MPX)))
1878 			return 1;
1879 		msr_info->data = vmcs_read64(GUEST_BNDCFGS);
1880 		break;
1881 	case MSR_IA32_MCG_EXT_CTL:
1882 		if (!msr_info->host_initiated &&
1883 		    !(vmx->msr_ia32_feature_control &
1884 		      FEAT_CTL_LMCE_ENABLED))
1885 			return 1;
1886 		msr_info->data = vcpu->arch.mcg_ext_ctl;
1887 		break;
1888 	case MSR_IA32_FEAT_CTL:
1889 		msr_info->data = vmx->msr_ia32_feature_control;
1890 		break;
1891 	case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
1892 		if (!nested_vmx_allowed(vcpu))
1893 			return 1;
1894 		if (vmx_get_vmx_msr(&vmx->nested.msrs, msr_info->index,
1895 				    &msr_info->data))
1896 			return 1;
1897 		/*
1898 		 * Enlightened VMCS v1 doesn't have certain fields, but buggy
1899 		 * Hyper-V versions are still trying to use corresponding
1900 		 * features when they are exposed. Filter out the essential
1901 		 * minimum.
1902 		 */
1903 		if (!msr_info->host_initiated &&
1904 		    vmx->nested.enlightened_vmcs_enabled)
1905 			nested_evmcs_filter_control_msr(msr_info->index,
1906 							&msr_info->data);
1907 		break;
1908 	case MSR_IA32_RTIT_CTL:
1909 		if (pt_mode != PT_MODE_HOST_GUEST)
1910 			return 1;
1911 		msr_info->data = vmx->pt_desc.guest.ctl;
1912 		break;
1913 	case MSR_IA32_RTIT_STATUS:
1914 		if (pt_mode != PT_MODE_HOST_GUEST)
1915 			return 1;
1916 		msr_info->data = vmx->pt_desc.guest.status;
1917 		break;
1918 	case MSR_IA32_RTIT_CR3_MATCH:
1919 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
1920 			!intel_pt_validate_cap(vmx->pt_desc.caps,
1921 						PT_CAP_cr3_filtering))
1922 			return 1;
1923 		msr_info->data = vmx->pt_desc.guest.cr3_match;
1924 		break;
1925 	case MSR_IA32_RTIT_OUTPUT_BASE:
1926 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
1927 			(!intel_pt_validate_cap(vmx->pt_desc.caps,
1928 					PT_CAP_topa_output) &&
1929 			 !intel_pt_validate_cap(vmx->pt_desc.caps,
1930 					PT_CAP_single_range_output)))
1931 			return 1;
1932 		msr_info->data = vmx->pt_desc.guest.output_base;
1933 		break;
1934 	case MSR_IA32_RTIT_OUTPUT_MASK:
1935 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
1936 			(!intel_pt_validate_cap(vmx->pt_desc.caps,
1937 					PT_CAP_topa_output) &&
1938 			 !intel_pt_validate_cap(vmx->pt_desc.caps,
1939 					PT_CAP_single_range_output)))
1940 			return 1;
1941 		msr_info->data = vmx->pt_desc.guest.output_mask;
1942 		break;
1943 	case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
1944 		index = msr_info->index - MSR_IA32_RTIT_ADDR0_A;
1945 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
1946 			(index >= 2 * intel_pt_validate_cap(vmx->pt_desc.caps,
1947 					PT_CAP_num_address_ranges)))
1948 			return 1;
1949 		if (index % 2)
1950 			msr_info->data = vmx->pt_desc.guest.addr_b[index / 2];
1951 		else
1952 			msr_info->data = vmx->pt_desc.guest.addr_a[index / 2];
1953 		break;
1954 	case MSR_TSC_AUX:
1955 		if (!msr_info->host_initiated &&
1956 		    !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP))
1957 			return 1;
1958 		goto find_shared_msr;
1959 	default:
1960 	find_shared_msr:
1961 		msr = find_msr_entry(vmx, msr_info->index);
1962 		if (msr) {
1963 			msr_info->data = msr->data;
1964 			break;
1965 		}
1966 		return kvm_get_msr_common(vcpu, msr_info);
1967 	}
1968 
1969 	return 0;
1970 }
1971 
1972 /*
1973  * Writes msr value into the appropriate "register".
1974  * Returns 0 on success, non-0 otherwise.
1975  * Assumes vcpu_load() was already called.
1976  */
1977 static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1978 {
1979 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1980 	struct shared_msr_entry *msr;
1981 	int ret = 0;
1982 	u32 msr_index = msr_info->index;
1983 	u64 data = msr_info->data;
1984 	u32 index;
1985 
1986 	switch (msr_index) {
1987 	case MSR_EFER:
1988 		ret = kvm_set_msr_common(vcpu, msr_info);
1989 		break;
1990 #ifdef CONFIG_X86_64
1991 	case MSR_FS_BASE:
1992 		vmx_segment_cache_clear(vmx);
1993 		vmcs_writel(GUEST_FS_BASE, data);
1994 		break;
1995 	case MSR_GS_BASE:
1996 		vmx_segment_cache_clear(vmx);
1997 		vmcs_writel(GUEST_GS_BASE, data);
1998 		break;
1999 	case MSR_KERNEL_GS_BASE:
2000 		vmx_write_guest_kernel_gs_base(vmx, data);
2001 		break;
2002 #endif
2003 	case MSR_IA32_SYSENTER_CS:
2004 		if (is_guest_mode(vcpu))
2005 			get_vmcs12(vcpu)->guest_sysenter_cs = data;
2006 		vmcs_write32(GUEST_SYSENTER_CS, data);
2007 		break;
2008 	case MSR_IA32_SYSENTER_EIP:
2009 		if (is_guest_mode(vcpu))
2010 			get_vmcs12(vcpu)->guest_sysenter_eip = data;
2011 		vmcs_writel(GUEST_SYSENTER_EIP, data);
2012 		break;
2013 	case MSR_IA32_SYSENTER_ESP:
2014 		if (is_guest_mode(vcpu))
2015 			get_vmcs12(vcpu)->guest_sysenter_esp = data;
2016 		vmcs_writel(GUEST_SYSENTER_ESP, data);
2017 		break;
2018 	case MSR_IA32_DEBUGCTLMSR:
2019 		if (is_guest_mode(vcpu) && get_vmcs12(vcpu)->vm_exit_controls &
2020 						VM_EXIT_SAVE_DEBUG_CONTROLS)
2021 			get_vmcs12(vcpu)->guest_ia32_debugctl = data;
2022 
2023 		ret = kvm_set_msr_common(vcpu, msr_info);
2024 		break;
2025 
2026 	case MSR_IA32_BNDCFGS:
2027 		if (!kvm_mpx_supported() ||
2028 		    (!msr_info->host_initiated &&
2029 		     !guest_cpuid_has(vcpu, X86_FEATURE_MPX)))
2030 			return 1;
2031 		if (is_noncanonical_address(data & PAGE_MASK, vcpu) ||
2032 		    (data & MSR_IA32_BNDCFGS_RSVD))
2033 			return 1;
2034 		vmcs_write64(GUEST_BNDCFGS, data);
2035 		break;
2036 	case MSR_IA32_UMWAIT_CONTROL:
2037 		if (!msr_info->host_initiated && !vmx_has_waitpkg(vmx))
2038 			return 1;
2039 
2040 		/* The reserved bit 1 and non-32 bit [63:32] should be zero */
2041 		if (data & (BIT_ULL(1) | GENMASK_ULL(63, 32)))
2042 			return 1;
2043 
2044 		vmx->msr_ia32_umwait_control = data;
2045 		break;
2046 	case MSR_IA32_SPEC_CTRL:
2047 		if (!msr_info->host_initiated &&
2048 		    !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
2049 			return 1;
2050 
2051 		if (data & ~kvm_spec_ctrl_valid_bits(vcpu))
2052 			return 1;
2053 
2054 		vmx->spec_ctrl = data;
2055 		if (!data)
2056 			break;
2057 
2058 		/*
2059 		 * For non-nested:
2060 		 * When it's written (to non-zero) for the first time, pass
2061 		 * it through.
2062 		 *
2063 		 * For nested:
2064 		 * The handling of the MSR bitmap for L2 guests is done in
2065 		 * nested_vmx_prepare_msr_bitmap. We should not touch the
2066 		 * vmcs02.msr_bitmap here since it gets completely overwritten
2067 		 * in the merging. We update the vmcs01 here for L1 as well
2068 		 * since it will end up touching the MSR anyway now.
2069 		 */
2070 		vmx_disable_intercept_for_msr(vmx->vmcs01.msr_bitmap,
2071 					      MSR_IA32_SPEC_CTRL,
2072 					      MSR_TYPE_RW);
2073 		break;
2074 	case MSR_IA32_TSX_CTRL:
2075 		if (!msr_info->host_initiated &&
2076 		    !(vcpu->arch.arch_capabilities & ARCH_CAP_TSX_CTRL_MSR))
2077 			return 1;
2078 		if (data & ~(TSX_CTRL_RTM_DISABLE | TSX_CTRL_CPUID_CLEAR))
2079 			return 1;
2080 		goto find_shared_msr;
2081 	case MSR_IA32_PRED_CMD:
2082 		if (!msr_info->host_initiated &&
2083 		    !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
2084 			return 1;
2085 
2086 		if (data & ~PRED_CMD_IBPB)
2087 			return 1;
2088 		if (!boot_cpu_has(X86_FEATURE_SPEC_CTRL))
2089 			return 1;
2090 		if (!data)
2091 			break;
2092 
2093 		wrmsrl(MSR_IA32_PRED_CMD, PRED_CMD_IBPB);
2094 
2095 		/*
2096 		 * For non-nested:
2097 		 * When it's written (to non-zero) for the first time, pass
2098 		 * it through.
2099 		 *
2100 		 * For nested:
2101 		 * The handling of the MSR bitmap for L2 guests is done in
2102 		 * nested_vmx_prepare_msr_bitmap. We should not touch the
2103 		 * vmcs02.msr_bitmap here since it gets completely overwritten
2104 		 * in the merging.
2105 		 */
2106 		vmx_disable_intercept_for_msr(vmx->vmcs01.msr_bitmap, MSR_IA32_PRED_CMD,
2107 					      MSR_TYPE_W);
2108 		break;
2109 	case MSR_IA32_CR_PAT:
2110 		if (!kvm_pat_valid(data))
2111 			return 1;
2112 
2113 		if (is_guest_mode(vcpu) &&
2114 		    get_vmcs12(vcpu)->vm_exit_controls & VM_EXIT_SAVE_IA32_PAT)
2115 			get_vmcs12(vcpu)->guest_ia32_pat = data;
2116 
2117 		if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2118 			vmcs_write64(GUEST_IA32_PAT, data);
2119 			vcpu->arch.pat = data;
2120 			break;
2121 		}
2122 		ret = kvm_set_msr_common(vcpu, msr_info);
2123 		break;
2124 	case MSR_IA32_TSC_ADJUST:
2125 		ret = kvm_set_msr_common(vcpu, msr_info);
2126 		break;
2127 	case MSR_IA32_MCG_EXT_CTL:
2128 		if ((!msr_info->host_initiated &&
2129 		     !(to_vmx(vcpu)->msr_ia32_feature_control &
2130 		       FEAT_CTL_LMCE_ENABLED)) ||
2131 		    (data & ~MCG_EXT_CTL_LMCE_EN))
2132 			return 1;
2133 		vcpu->arch.mcg_ext_ctl = data;
2134 		break;
2135 	case MSR_IA32_FEAT_CTL:
2136 		if (!vmx_feature_control_msr_valid(vcpu, data) ||
2137 		    (to_vmx(vcpu)->msr_ia32_feature_control &
2138 		     FEAT_CTL_LOCKED && !msr_info->host_initiated))
2139 			return 1;
2140 		vmx->msr_ia32_feature_control = data;
2141 		if (msr_info->host_initiated && data == 0)
2142 			vmx_leave_nested(vcpu);
2143 		break;
2144 	case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
2145 		if (!msr_info->host_initiated)
2146 			return 1; /* they are read-only */
2147 		if (!nested_vmx_allowed(vcpu))
2148 			return 1;
2149 		return vmx_set_vmx_msr(vcpu, msr_index, data);
2150 	case MSR_IA32_RTIT_CTL:
2151 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
2152 			vmx_rtit_ctl_check(vcpu, data) ||
2153 			vmx->nested.vmxon)
2154 			return 1;
2155 		vmcs_write64(GUEST_IA32_RTIT_CTL, data);
2156 		vmx->pt_desc.guest.ctl = data;
2157 		pt_update_intercept_for_msr(vmx);
2158 		break;
2159 	case MSR_IA32_RTIT_STATUS:
2160 		if (!pt_can_write_msr(vmx))
2161 			return 1;
2162 		if (data & MSR_IA32_RTIT_STATUS_MASK)
2163 			return 1;
2164 		vmx->pt_desc.guest.status = data;
2165 		break;
2166 	case MSR_IA32_RTIT_CR3_MATCH:
2167 		if (!pt_can_write_msr(vmx))
2168 			return 1;
2169 		if (!intel_pt_validate_cap(vmx->pt_desc.caps,
2170 					   PT_CAP_cr3_filtering))
2171 			return 1;
2172 		vmx->pt_desc.guest.cr3_match = data;
2173 		break;
2174 	case MSR_IA32_RTIT_OUTPUT_BASE:
2175 		if (!pt_can_write_msr(vmx))
2176 			return 1;
2177 		if (!intel_pt_validate_cap(vmx->pt_desc.caps,
2178 					   PT_CAP_topa_output) &&
2179 		    !intel_pt_validate_cap(vmx->pt_desc.caps,
2180 					   PT_CAP_single_range_output))
2181 			return 1;
2182 		if (data & MSR_IA32_RTIT_OUTPUT_BASE_MASK)
2183 			return 1;
2184 		vmx->pt_desc.guest.output_base = data;
2185 		break;
2186 	case MSR_IA32_RTIT_OUTPUT_MASK:
2187 		if (!pt_can_write_msr(vmx))
2188 			return 1;
2189 		if (!intel_pt_validate_cap(vmx->pt_desc.caps,
2190 					   PT_CAP_topa_output) &&
2191 		    !intel_pt_validate_cap(vmx->pt_desc.caps,
2192 					   PT_CAP_single_range_output))
2193 			return 1;
2194 		vmx->pt_desc.guest.output_mask = data;
2195 		break;
2196 	case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
2197 		if (!pt_can_write_msr(vmx))
2198 			return 1;
2199 		index = msr_info->index - MSR_IA32_RTIT_ADDR0_A;
2200 		if (index >= 2 * intel_pt_validate_cap(vmx->pt_desc.caps,
2201 						       PT_CAP_num_address_ranges))
2202 			return 1;
2203 		if (is_noncanonical_address(data, vcpu))
2204 			return 1;
2205 		if (index % 2)
2206 			vmx->pt_desc.guest.addr_b[index / 2] = data;
2207 		else
2208 			vmx->pt_desc.guest.addr_a[index / 2] = data;
2209 		break;
2210 	case MSR_TSC_AUX:
2211 		if (!msr_info->host_initiated &&
2212 		    !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP))
2213 			return 1;
2214 		/* Check reserved bit, higher 32 bits should be zero */
2215 		if ((data >> 32) != 0)
2216 			return 1;
2217 		goto find_shared_msr;
2218 
2219 	default:
2220 	find_shared_msr:
2221 		msr = find_msr_entry(vmx, msr_index);
2222 		if (msr)
2223 			ret = vmx_set_guest_msr(vmx, msr, data);
2224 		else
2225 			ret = kvm_set_msr_common(vcpu, msr_info);
2226 	}
2227 
2228 	return ret;
2229 }
2230 
2231 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
2232 {
2233 	kvm_register_mark_available(vcpu, reg);
2234 
2235 	switch (reg) {
2236 	case VCPU_REGS_RSP:
2237 		vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
2238 		break;
2239 	case VCPU_REGS_RIP:
2240 		vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
2241 		break;
2242 	case VCPU_EXREG_PDPTR:
2243 		if (enable_ept)
2244 			ept_save_pdptrs(vcpu);
2245 		break;
2246 	case VCPU_EXREG_CR3:
2247 		if (enable_unrestricted_guest || (enable_ept && is_paging(vcpu)))
2248 			vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
2249 		break;
2250 	default:
2251 		WARN_ON_ONCE(1);
2252 		break;
2253 	}
2254 }
2255 
2256 static __init int cpu_has_kvm_support(void)
2257 {
2258 	return cpu_has_vmx();
2259 }
2260 
2261 static __init int vmx_disabled_by_bios(void)
2262 {
2263 	return !boot_cpu_has(X86_FEATURE_MSR_IA32_FEAT_CTL) ||
2264 	       !boot_cpu_has(X86_FEATURE_VMX);
2265 }
2266 
2267 static void kvm_cpu_vmxon(u64 addr)
2268 {
2269 	cr4_set_bits(X86_CR4_VMXE);
2270 	intel_pt_handle_vmx(1);
2271 
2272 	asm volatile ("vmxon %0" : : "m"(addr));
2273 }
2274 
2275 static int hardware_enable(void)
2276 {
2277 	int cpu = raw_smp_processor_id();
2278 	u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
2279 
2280 	if (cr4_read_shadow() & X86_CR4_VMXE)
2281 		return -EBUSY;
2282 
2283 	/*
2284 	 * This can happen if we hot-added a CPU but failed to allocate
2285 	 * VP assist page for it.
2286 	 */
2287 	if (static_branch_unlikely(&enable_evmcs) &&
2288 	    !hv_get_vp_assist_page(cpu))
2289 		return -EFAULT;
2290 
2291 	INIT_LIST_HEAD(&per_cpu(loaded_vmcss_on_cpu, cpu));
2292 	INIT_LIST_HEAD(&per_cpu(blocked_vcpu_on_cpu, cpu));
2293 	spin_lock_init(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
2294 
2295 	/*
2296 	 * Now we can enable the vmclear operation in kdump
2297 	 * since the loaded_vmcss_on_cpu list on this cpu
2298 	 * has been initialized.
2299 	 *
2300 	 * Though the cpu is not in VMX operation now, there
2301 	 * is no problem to enable the vmclear operation
2302 	 * for the loaded_vmcss_on_cpu list is empty!
2303 	 */
2304 	crash_enable_local_vmclear(cpu);
2305 
2306 	kvm_cpu_vmxon(phys_addr);
2307 	if (enable_ept)
2308 		ept_sync_global();
2309 
2310 	return 0;
2311 }
2312 
2313 static void vmclear_local_loaded_vmcss(void)
2314 {
2315 	int cpu = raw_smp_processor_id();
2316 	struct loaded_vmcs *v, *n;
2317 
2318 	list_for_each_entry_safe(v, n, &per_cpu(loaded_vmcss_on_cpu, cpu),
2319 				 loaded_vmcss_on_cpu_link)
2320 		__loaded_vmcs_clear(v);
2321 }
2322 
2323 
2324 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
2325  * tricks.
2326  */
2327 static void kvm_cpu_vmxoff(void)
2328 {
2329 	asm volatile (__ex("vmxoff"));
2330 
2331 	intel_pt_handle_vmx(0);
2332 	cr4_clear_bits(X86_CR4_VMXE);
2333 }
2334 
2335 static void hardware_disable(void)
2336 {
2337 	vmclear_local_loaded_vmcss();
2338 	kvm_cpu_vmxoff();
2339 }
2340 
2341 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
2342 				      u32 msr, u32 *result)
2343 {
2344 	u32 vmx_msr_low, vmx_msr_high;
2345 	u32 ctl = ctl_min | ctl_opt;
2346 
2347 	rdmsr(msr, vmx_msr_low, vmx_msr_high);
2348 
2349 	ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
2350 	ctl |= vmx_msr_low;  /* bit == 1 in low word  ==> must be one  */
2351 
2352 	/* Ensure minimum (required) set of control bits are supported. */
2353 	if (ctl_min & ~ctl)
2354 		return -EIO;
2355 
2356 	*result = ctl;
2357 	return 0;
2358 }
2359 
2360 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf,
2361 				    struct vmx_capability *vmx_cap)
2362 {
2363 	u32 vmx_msr_low, vmx_msr_high;
2364 	u32 min, opt, min2, opt2;
2365 	u32 _pin_based_exec_control = 0;
2366 	u32 _cpu_based_exec_control = 0;
2367 	u32 _cpu_based_2nd_exec_control = 0;
2368 	u32 _vmexit_control = 0;
2369 	u32 _vmentry_control = 0;
2370 
2371 	memset(vmcs_conf, 0, sizeof(*vmcs_conf));
2372 	min = CPU_BASED_HLT_EXITING |
2373 #ifdef CONFIG_X86_64
2374 	      CPU_BASED_CR8_LOAD_EXITING |
2375 	      CPU_BASED_CR8_STORE_EXITING |
2376 #endif
2377 	      CPU_BASED_CR3_LOAD_EXITING |
2378 	      CPU_BASED_CR3_STORE_EXITING |
2379 	      CPU_BASED_UNCOND_IO_EXITING |
2380 	      CPU_BASED_MOV_DR_EXITING |
2381 	      CPU_BASED_USE_TSC_OFFSETTING |
2382 	      CPU_BASED_MWAIT_EXITING |
2383 	      CPU_BASED_MONITOR_EXITING |
2384 	      CPU_BASED_INVLPG_EXITING |
2385 	      CPU_BASED_RDPMC_EXITING;
2386 
2387 	opt = CPU_BASED_TPR_SHADOW |
2388 	      CPU_BASED_USE_MSR_BITMAPS |
2389 	      CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
2390 	if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
2391 				&_cpu_based_exec_control) < 0)
2392 		return -EIO;
2393 #ifdef CONFIG_X86_64
2394 	if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2395 		_cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
2396 					   ~CPU_BASED_CR8_STORE_EXITING;
2397 #endif
2398 	if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
2399 		min2 = 0;
2400 		opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2401 			SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2402 			SECONDARY_EXEC_WBINVD_EXITING |
2403 			SECONDARY_EXEC_ENABLE_VPID |
2404 			SECONDARY_EXEC_ENABLE_EPT |
2405 			SECONDARY_EXEC_UNRESTRICTED_GUEST |
2406 			SECONDARY_EXEC_PAUSE_LOOP_EXITING |
2407 			SECONDARY_EXEC_DESC |
2408 			SECONDARY_EXEC_RDTSCP |
2409 			SECONDARY_EXEC_ENABLE_INVPCID |
2410 			SECONDARY_EXEC_APIC_REGISTER_VIRT |
2411 			SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
2412 			SECONDARY_EXEC_SHADOW_VMCS |
2413 			SECONDARY_EXEC_XSAVES |
2414 			SECONDARY_EXEC_RDSEED_EXITING |
2415 			SECONDARY_EXEC_RDRAND_EXITING |
2416 			SECONDARY_EXEC_ENABLE_PML |
2417 			SECONDARY_EXEC_TSC_SCALING |
2418 			SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE |
2419 			SECONDARY_EXEC_PT_USE_GPA |
2420 			SECONDARY_EXEC_PT_CONCEAL_VMX |
2421 			SECONDARY_EXEC_ENABLE_VMFUNC |
2422 			SECONDARY_EXEC_ENCLS_EXITING;
2423 		if (adjust_vmx_controls(min2, opt2,
2424 					MSR_IA32_VMX_PROCBASED_CTLS2,
2425 					&_cpu_based_2nd_exec_control) < 0)
2426 			return -EIO;
2427 	}
2428 #ifndef CONFIG_X86_64
2429 	if (!(_cpu_based_2nd_exec_control &
2430 				SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
2431 		_cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
2432 #endif
2433 
2434 	if (!(_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2435 		_cpu_based_2nd_exec_control &= ~(
2436 				SECONDARY_EXEC_APIC_REGISTER_VIRT |
2437 				SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2438 				SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
2439 
2440 	rdmsr_safe(MSR_IA32_VMX_EPT_VPID_CAP,
2441 		&vmx_cap->ept, &vmx_cap->vpid);
2442 
2443 	if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
2444 		/* CR3 accesses and invlpg don't need to cause VM Exits when EPT
2445 		   enabled */
2446 		_cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
2447 					     CPU_BASED_CR3_STORE_EXITING |
2448 					     CPU_BASED_INVLPG_EXITING);
2449 	} else if (vmx_cap->ept) {
2450 		vmx_cap->ept = 0;
2451 		pr_warn_once("EPT CAP should not exist if not support "
2452 				"1-setting enable EPT VM-execution control\n");
2453 	}
2454 	if (!(_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_VPID) &&
2455 		vmx_cap->vpid) {
2456 		vmx_cap->vpid = 0;
2457 		pr_warn_once("VPID CAP should not exist if not support "
2458 				"1-setting enable VPID VM-execution control\n");
2459 	}
2460 
2461 	min = VM_EXIT_SAVE_DEBUG_CONTROLS | VM_EXIT_ACK_INTR_ON_EXIT;
2462 #ifdef CONFIG_X86_64
2463 	min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
2464 #endif
2465 	opt = VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL |
2466 	      VM_EXIT_LOAD_IA32_PAT |
2467 	      VM_EXIT_LOAD_IA32_EFER |
2468 	      VM_EXIT_CLEAR_BNDCFGS |
2469 	      VM_EXIT_PT_CONCEAL_PIP |
2470 	      VM_EXIT_CLEAR_IA32_RTIT_CTL;
2471 	if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
2472 				&_vmexit_control) < 0)
2473 		return -EIO;
2474 
2475 	min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
2476 	opt = PIN_BASED_VIRTUAL_NMIS | PIN_BASED_POSTED_INTR |
2477 		 PIN_BASED_VMX_PREEMPTION_TIMER;
2478 	if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
2479 				&_pin_based_exec_control) < 0)
2480 		return -EIO;
2481 
2482 	if (cpu_has_broken_vmx_preemption_timer())
2483 		_pin_based_exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
2484 	if (!(_cpu_based_2nd_exec_control &
2485 		SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY))
2486 		_pin_based_exec_control &= ~PIN_BASED_POSTED_INTR;
2487 
2488 	min = VM_ENTRY_LOAD_DEBUG_CONTROLS;
2489 	opt = VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL |
2490 	      VM_ENTRY_LOAD_IA32_PAT |
2491 	      VM_ENTRY_LOAD_IA32_EFER |
2492 	      VM_ENTRY_LOAD_BNDCFGS |
2493 	      VM_ENTRY_PT_CONCEAL_PIP |
2494 	      VM_ENTRY_LOAD_IA32_RTIT_CTL;
2495 	if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
2496 				&_vmentry_control) < 0)
2497 		return -EIO;
2498 
2499 	/*
2500 	 * Some cpus support VM_{ENTRY,EXIT}_IA32_PERF_GLOBAL_CTRL but they
2501 	 * can't be used due to an errata where VM Exit may incorrectly clear
2502 	 * IA32_PERF_GLOBAL_CTRL[34:32].  Workaround the errata by using the
2503 	 * MSR load mechanism to switch IA32_PERF_GLOBAL_CTRL.
2504 	 */
2505 	if (boot_cpu_data.x86 == 0x6) {
2506 		switch (boot_cpu_data.x86_model) {
2507 		case 26: /* AAK155 */
2508 		case 30: /* AAP115 */
2509 		case 37: /* AAT100 */
2510 		case 44: /* BC86,AAY89,BD102 */
2511 		case 46: /* BA97 */
2512 			_vmentry_control &= ~VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL;
2513 			_vmexit_control &= ~VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL;
2514 			pr_warn_once("kvm: VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL "
2515 					"does not work properly. Using workaround\n");
2516 			break;
2517 		default:
2518 			break;
2519 		}
2520 	}
2521 
2522 
2523 	rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
2524 
2525 	/* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
2526 	if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
2527 		return -EIO;
2528 
2529 #ifdef CONFIG_X86_64
2530 	/* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
2531 	if (vmx_msr_high & (1u<<16))
2532 		return -EIO;
2533 #endif
2534 
2535 	/* Require Write-Back (WB) memory type for VMCS accesses. */
2536 	if (((vmx_msr_high >> 18) & 15) != 6)
2537 		return -EIO;
2538 
2539 	vmcs_conf->size = vmx_msr_high & 0x1fff;
2540 	vmcs_conf->order = get_order(vmcs_conf->size);
2541 	vmcs_conf->basic_cap = vmx_msr_high & ~0x1fff;
2542 
2543 	vmcs_conf->revision_id = vmx_msr_low;
2544 
2545 	vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
2546 	vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
2547 	vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
2548 	vmcs_conf->vmexit_ctrl         = _vmexit_control;
2549 	vmcs_conf->vmentry_ctrl        = _vmentry_control;
2550 
2551 	if (static_branch_unlikely(&enable_evmcs))
2552 		evmcs_sanitize_exec_ctrls(vmcs_conf);
2553 
2554 	return 0;
2555 }
2556 
2557 struct vmcs *alloc_vmcs_cpu(bool shadow, int cpu, gfp_t flags)
2558 {
2559 	int node = cpu_to_node(cpu);
2560 	struct page *pages;
2561 	struct vmcs *vmcs;
2562 
2563 	pages = __alloc_pages_node(node, flags, vmcs_config.order);
2564 	if (!pages)
2565 		return NULL;
2566 	vmcs = page_address(pages);
2567 	memset(vmcs, 0, vmcs_config.size);
2568 
2569 	/* KVM supports Enlightened VMCS v1 only */
2570 	if (static_branch_unlikely(&enable_evmcs))
2571 		vmcs->hdr.revision_id = KVM_EVMCS_VERSION;
2572 	else
2573 		vmcs->hdr.revision_id = vmcs_config.revision_id;
2574 
2575 	if (shadow)
2576 		vmcs->hdr.shadow_vmcs = 1;
2577 	return vmcs;
2578 }
2579 
2580 void free_vmcs(struct vmcs *vmcs)
2581 {
2582 	free_pages((unsigned long)vmcs, vmcs_config.order);
2583 }
2584 
2585 /*
2586  * Free a VMCS, but before that VMCLEAR it on the CPU where it was last loaded
2587  */
2588 void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
2589 {
2590 	if (!loaded_vmcs->vmcs)
2591 		return;
2592 	loaded_vmcs_clear(loaded_vmcs);
2593 	free_vmcs(loaded_vmcs->vmcs);
2594 	loaded_vmcs->vmcs = NULL;
2595 	if (loaded_vmcs->msr_bitmap)
2596 		free_page((unsigned long)loaded_vmcs->msr_bitmap);
2597 	WARN_ON(loaded_vmcs->shadow_vmcs != NULL);
2598 }
2599 
2600 int alloc_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
2601 {
2602 	loaded_vmcs->vmcs = alloc_vmcs(false);
2603 	if (!loaded_vmcs->vmcs)
2604 		return -ENOMEM;
2605 
2606 	loaded_vmcs->shadow_vmcs = NULL;
2607 	loaded_vmcs->hv_timer_soft_disabled = false;
2608 	loaded_vmcs_init(loaded_vmcs);
2609 
2610 	if (cpu_has_vmx_msr_bitmap()) {
2611 		loaded_vmcs->msr_bitmap = (unsigned long *)
2612 				__get_free_page(GFP_KERNEL_ACCOUNT);
2613 		if (!loaded_vmcs->msr_bitmap)
2614 			goto out_vmcs;
2615 		memset(loaded_vmcs->msr_bitmap, 0xff, PAGE_SIZE);
2616 
2617 		if (IS_ENABLED(CONFIG_HYPERV) &&
2618 		    static_branch_unlikely(&enable_evmcs) &&
2619 		    (ms_hyperv.nested_features & HV_X64_NESTED_MSR_BITMAP)) {
2620 			struct hv_enlightened_vmcs *evmcs =
2621 				(struct hv_enlightened_vmcs *)loaded_vmcs->vmcs;
2622 
2623 			evmcs->hv_enlightenments_control.msr_bitmap = 1;
2624 		}
2625 	}
2626 
2627 	memset(&loaded_vmcs->host_state, 0, sizeof(struct vmcs_host_state));
2628 	memset(&loaded_vmcs->controls_shadow, 0,
2629 		sizeof(struct vmcs_controls_shadow));
2630 
2631 	return 0;
2632 
2633 out_vmcs:
2634 	free_loaded_vmcs(loaded_vmcs);
2635 	return -ENOMEM;
2636 }
2637 
2638 static void free_kvm_area(void)
2639 {
2640 	int cpu;
2641 
2642 	for_each_possible_cpu(cpu) {
2643 		free_vmcs(per_cpu(vmxarea, cpu));
2644 		per_cpu(vmxarea, cpu) = NULL;
2645 	}
2646 }
2647 
2648 static __init int alloc_kvm_area(void)
2649 {
2650 	int cpu;
2651 
2652 	for_each_possible_cpu(cpu) {
2653 		struct vmcs *vmcs;
2654 
2655 		vmcs = alloc_vmcs_cpu(false, cpu, GFP_KERNEL);
2656 		if (!vmcs) {
2657 			free_kvm_area();
2658 			return -ENOMEM;
2659 		}
2660 
2661 		/*
2662 		 * When eVMCS is enabled, alloc_vmcs_cpu() sets
2663 		 * vmcs->revision_id to KVM_EVMCS_VERSION instead of
2664 		 * revision_id reported by MSR_IA32_VMX_BASIC.
2665 		 *
2666 		 * However, even though not explicitly documented by
2667 		 * TLFS, VMXArea passed as VMXON argument should
2668 		 * still be marked with revision_id reported by
2669 		 * physical CPU.
2670 		 */
2671 		if (static_branch_unlikely(&enable_evmcs))
2672 			vmcs->hdr.revision_id = vmcs_config.revision_id;
2673 
2674 		per_cpu(vmxarea, cpu) = vmcs;
2675 	}
2676 	return 0;
2677 }
2678 
2679 static void fix_pmode_seg(struct kvm_vcpu *vcpu, int seg,
2680 		struct kvm_segment *save)
2681 {
2682 	if (!emulate_invalid_guest_state) {
2683 		/*
2684 		 * CS and SS RPL should be equal during guest entry according
2685 		 * to VMX spec, but in reality it is not always so. Since vcpu
2686 		 * is in the middle of the transition from real mode to
2687 		 * protected mode it is safe to assume that RPL 0 is a good
2688 		 * default value.
2689 		 */
2690 		if (seg == VCPU_SREG_CS || seg == VCPU_SREG_SS)
2691 			save->selector &= ~SEGMENT_RPL_MASK;
2692 		save->dpl = save->selector & SEGMENT_RPL_MASK;
2693 		save->s = 1;
2694 	}
2695 	vmx_set_segment(vcpu, save, seg);
2696 }
2697 
2698 static void enter_pmode(struct kvm_vcpu *vcpu)
2699 {
2700 	unsigned long flags;
2701 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2702 
2703 	/*
2704 	 * Update real mode segment cache. It may be not up-to-date if sement
2705 	 * register was written while vcpu was in a guest mode.
2706 	 */
2707 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
2708 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
2709 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
2710 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
2711 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
2712 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
2713 
2714 	vmx->rmode.vm86_active = 0;
2715 
2716 	vmx_set_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
2717 
2718 	flags = vmcs_readl(GUEST_RFLAGS);
2719 	flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
2720 	flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
2721 	vmcs_writel(GUEST_RFLAGS, flags);
2722 
2723 	vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
2724 			(vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
2725 
2726 	update_exception_bitmap(vcpu);
2727 
2728 	fix_pmode_seg(vcpu, VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
2729 	fix_pmode_seg(vcpu, VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
2730 	fix_pmode_seg(vcpu, VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
2731 	fix_pmode_seg(vcpu, VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
2732 	fix_pmode_seg(vcpu, VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
2733 	fix_pmode_seg(vcpu, VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
2734 }
2735 
2736 static void fix_rmode_seg(int seg, struct kvm_segment *save)
2737 {
2738 	const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2739 	struct kvm_segment var = *save;
2740 
2741 	var.dpl = 0x3;
2742 	if (seg == VCPU_SREG_CS)
2743 		var.type = 0x3;
2744 
2745 	if (!emulate_invalid_guest_state) {
2746 		var.selector = var.base >> 4;
2747 		var.base = var.base & 0xffff0;
2748 		var.limit = 0xffff;
2749 		var.g = 0;
2750 		var.db = 0;
2751 		var.present = 1;
2752 		var.s = 1;
2753 		var.l = 0;
2754 		var.unusable = 0;
2755 		var.type = 0x3;
2756 		var.avl = 0;
2757 		if (save->base & 0xf)
2758 			printk_once(KERN_WARNING "kvm: segment base is not "
2759 					"paragraph aligned when entering "
2760 					"protected mode (seg=%d)", seg);
2761 	}
2762 
2763 	vmcs_write16(sf->selector, var.selector);
2764 	vmcs_writel(sf->base, var.base);
2765 	vmcs_write32(sf->limit, var.limit);
2766 	vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(&var));
2767 }
2768 
2769 static void enter_rmode(struct kvm_vcpu *vcpu)
2770 {
2771 	unsigned long flags;
2772 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2773 	struct kvm_vmx *kvm_vmx = to_kvm_vmx(vcpu->kvm);
2774 
2775 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
2776 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
2777 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
2778 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
2779 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
2780 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
2781 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
2782 
2783 	vmx->rmode.vm86_active = 1;
2784 
2785 	/*
2786 	 * Very old userspace does not call KVM_SET_TSS_ADDR before entering
2787 	 * vcpu. Warn the user that an update is overdue.
2788 	 */
2789 	if (!kvm_vmx->tss_addr)
2790 		printk_once(KERN_WARNING "kvm: KVM_SET_TSS_ADDR need to be "
2791 			     "called before entering vcpu\n");
2792 
2793 	vmx_segment_cache_clear(vmx);
2794 
2795 	vmcs_writel(GUEST_TR_BASE, kvm_vmx->tss_addr);
2796 	vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
2797 	vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
2798 
2799 	flags = vmcs_readl(GUEST_RFLAGS);
2800 	vmx->rmode.save_rflags = flags;
2801 
2802 	flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
2803 
2804 	vmcs_writel(GUEST_RFLAGS, flags);
2805 	vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
2806 	update_exception_bitmap(vcpu);
2807 
2808 	fix_rmode_seg(VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
2809 	fix_rmode_seg(VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
2810 	fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
2811 	fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
2812 	fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
2813 	fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
2814 
2815 	kvm_mmu_reset_context(vcpu);
2816 }
2817 
2818 void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
2819 {
2820 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2821 	struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
2822 
2823 	if (!msr)
2824 		return;
2825 
2826 	vcpu->arch.efer = efer;
2827 	if (efer & EFER_LMA) {
2828 		vm_entry_controls_setbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
2829 		msr->data = efer;
2830 	} else {
2831 		vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
2832 
2833 		msr->data = efer & ~EFER_LME;
2834 	}
2835 	setup_msrs(vmx);
2836 }
2837 
2838 #ifdef CONFIG_X86_64
2839 
2840 static void enter_lmode(struct kvm_vcpu *vcpu)
2841 {
2842 	u32 guest_tr_ar;
2843 
2844 	vmx_segment_cache_clear(to_vmx(vcpu));
2845 
2846 	guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
2847 	if ((guest_tr_ar & VMX_AR_TYPE_MASK) != VMX_AR_TYPE_BUSY_64_TSS) {
2848 		pr_debug_ratelimited("%s: tss fixup for long mode. \n",
2849 				     __func__);
2850 		vmcs_write32(GUEST_TR_AR_BYTES,
2851 			     (guest_tr_ar & ~VMX_AR_TYPE_MASK)
2852 			     | VMX_AR_TYPE_BUSY_64_TSS);
2853 	}
2854 	vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
2855 }
2856 
2857 static void exit_lmode(struct kvm_vcpu *vcpu)
2858 {
2859 	vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
2860 	vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
2861 }
2862 
2863 #endif
2864 
2865 static void vmx_flush_tlb_gva(struct kvm_vcpu *vcpu, gva_t addr)
2866 {
2867 	int vpid = to_vmx(vcpu)->vpid;
2868 
2869 	if (!vpid_sync_vcpu_addr(vpid, addr))
2870 		vpid_sync_context(vpid);
2871 
2872 	/*
2873 	 * If VPIDs are not supported or enabled, then the above is a no-op.
2874 	 * But we don't really need a TLB flush in that case anyway, because
2875 	 * each VM entry/exit includes an implicit flush when VPID is 0.
2876 	 */
2877 }
2878 
2879 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
2880 {
2881 	ulong cr0_guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
2882 
2883 	vcpu->arch.cr0 &= ~cr0_guest_owned_bits;
2884 	vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & cr0_guest_owned_bits;
2885 }
2886 
2887 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
2888 {
2889 	ulong cr4_guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
2890 
2891 	vcpu->arch.cr4 &= ~cr4_guest_owned_bits;
2892 	vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & cr4_guest_owned_bits;
2893 }
2894 
2895 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
2896 {
2897 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
2898 
2899 	if (!kvm_register_is_dirty(vcpu, VCPU_EXREG_PDPTR))
2900 		return;
2901 
2902 	if (is_pae_paging(vcpu)) {
2903 		vmcs_write64(GUEST_PDPTR0, mmu->pdptrs[0]);
2904 		vmcs_write64(GUEST_PDPTR1, mmu->pdptrs[1]);
2905 		vmcs_write64(GUEST_PDPTR2, mmu->pdptrs[2]);
2906 		vmcs_write64(GUEST_PDPTR3, mmu->pdptrs[3]);
2907 	}
2908 }
2909 
2910 void ept_save_pdptrs(struct kvm_vcpu *vcpu)
2911 {
2912 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
2913 
2914 	if (is_pae_paging(vcpu)) {
2915 		mmu->pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
2916 		mmu->pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
2917 		mmu->pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
2918 		mmu->pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
2919 	}
2920 
2921 	kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
2922 }
2923 
2924 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
2925 					unsigned long cr0,
2926 					struct kvm_vcpu *vcpu)
2927 {
2928 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2929 
2930 	if (!kvm_register_is_available(vcpu, VCPU_EXREG_CR3))
2931 		vmx_cache_reg(vcpu, VCPU_EXREG_CR3);
2932 	if (!(cr0 & X86_CR0_PG)) {
2933 		/* From paging/starting to nonpaging */
2934 		exec_controls_setbit(vmx, CPU_BASED_CR3_LOAD_EXITING |
2935 					  CPU_BASED_CR3_STORE_EXITING);
2936 		vcpu->arch.cr0 = cr0;
2937 		vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
2938 	} else if (!is_paging(vcpu)) {
2939 		/* From nonpaging to paging */
2940 		exec_controls_clearbit(vmx, CPU_BASED_CR3_LOAD_EXITING |
2941 					    CPU_BASED_CR3_STORE_EXITING);
2942 		vcpu->arch.cr0 = cr0;
2943 		vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
2944 	}
2945 
2946 	if (!(cr0 & X86_CR0_WP))
2947 		*hw_cr0 &= ~X86_CR0_WP;
2948 }
2949 
2950 void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
2951 {
2952 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2953 	unsigned long hw_cr0;
2954 
2955 	hw_cr0 = (cr0 & ~KVM_VM_CR0_ALWAYS_OFF);
2956 	if (enable_unrestricted_guest)
2957 		hw_cr0 |= KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
2958 	else {
2959 		hw_cr0 |= KVM_VM_CR0_ALWAYS_ON;
2960 
2961 		if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
2962 			enter_pmode(vcpu);
2963 
2964 		if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
2965 			enter_rmode(vcpu);
2966 	}
2967 
2968 #ifdef CONFIG_X86_64
2969 	if (vcpu->arch.efer & EFER_LME) {
2970 		if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
2971 			enter_lmode(vcpu);
2972 		if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
2973 			exit_lmode(vcpu);
2974 	}
2975 #endif
2976 
2977 	if (enable_ept && !enable_unrestricted_guest)
2978 		ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
2979 
2980 	vmcs_writel(CR0_READ_SHADOW, cr0);
2981 	vmcs_writel(GUEST_CR0, hw_cr0);
2982 	vcpu->arch.cr0 = cr0;
2983 
2984 	/* depends on vcpu->arch.cr0 to be set to a new value */
2985 	vmx->emulation_required = emulation_required(vcpu);
2986 }
2987 
2988 static int get_ept_level(struct kvm_vcpu *vcpu)
2989 {
2990 	/* Nested EPT currently only supports 4-level walks. */
2991 	if (is_guest_mode(vcpu) && nested_cpu_has_ept(get_vmcs12(vcpu)))
2992 		return 4;
2993 	if (cpu_has_vmx_ept_5levels() && (cpuid_maxphyaddr(vcpu) > 48))
2994 		return 5;
2995 	return 4;
2996 }
2997 
2998 u64 construct_eptp(struct kvm_vcpu *vcpu, unsigned long root_hpa)
2999 {
3000 	u64 eptp = VMX_EPTP_MT_WB;
3001 
3002 	eptp |= (get_ept_level(vcpu) == 5) ? VMX_EPTP_PWL_5 : VMX_EPTP_PWL_4;
3003 
3004 	if (enable_ept_ad_bits &&
3005 	    (!is_guest_mode(vcpu) || nested_ept_ad_enabled(vcpu)))
3006 		eptp |= VMX_EPTP_AD_ENABLE_BIT;
3007 	eptp |= (root_hpa & PAGE_MASK);
3008 
3009 	return eptp;
3010 }
3011 
3012 void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
3013 {
3014 	struct kvm *kvm = vcpu->kvm;
3015 	bool update_guest_cr3 = true;
3016 	unsigned long guest_cr3;
3017 	u64 eptp;
3018 
3019 	guest_cr3 = cr3;
3020 	if (enable_ept) {
3021 		eptp = construct_eptp(vcpu, cr3);
3022 		vmcs_write64(EPT_POINTER, eptp);
3023 
3024 		if (kvm_x86_ops->tlb_remote_flush) {
3025 			spin_lock(&to_kvm_vmx(kvm)->ept_pointer_lock);
3026 			to_vmx(vcpu)->ept_pointer = eptp;
3027 			to_kvm_vmx(kvm)->ept_pointers_match
3028 				= EPT_POINTERS_CHECK;
3029 			spin_unlock(&to_kvm_vmx(kvm)->ept_pointer_lock);
3030 		}
3031 
3032 		/* Loading vmcs02.GUEST_CR3 is handled by nested VM-Enter. */
3033 		if (is_guest_mode(vcpu))
3034 			update_guest_cr3 = false;
3035 		else if (!enable_unrestricted_guest && !is_paging(vcpu))
3036 			guest_cr3 = to_kvm_vmx(kvm)->ept_identity_map_addr;
3037 		else if (test_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail))
3038 			guest_cr3 = vcpu->arch.cr3;
3039 		else /* vmcs01.GUEST_CR3 is already up-to-date. */
3040 			update_guest_cr3 = false;
3041 		ept_load_pdptrs(vcpu);
3042 	}
3043 
3044 	if (update_guest_cr3)
3045 		vmcs_writel(GUEST_CR3, guest_cr3);
3046 }
3047 
3048 int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
3049 {
3050 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3051 	/*
3052 	 * Pass through host's Machine Check Enable value to hw_cr4, which
3053 	 * is in force while we are in guest mode.  Do not let guests control
3054 	 * this bit, even if host CR4.MCE == 0.
3055 	 */
3056 	unsigned long hw_cr4;
3057 
3058 	hw_cr4 = (cr4_read_shadow() & X86_CR4_MCE) | (cr4 & ~X86_CR4_MCE);
3059 	if (enable_unrestricted_guest)
3060 		hw_cr4 |= KVM_VM_CR4_ALWAYS_ON_UNRESTRICTED_GUEST;
3061 	else if (vmx->rmode.vm86_active)
3062 		hw_cr4 |= KVM_RMODE_VM_CR4_ALWAYS_ON;
3063 	else
3064 		hw_cr4 |= KVM_PMODE_VM_CR4_ALWAYS_ON;
3065 
3066 	if (!boot_cpu_has(X86_FEATURE_UMIP) && vmx_umip_emulated()) {
3067 		if (cr4 & X86_CR4_UMIP) {
3068 			secondary_exec_controls_setbit(vmx, SECONDARY_EXEC_DESC);
3069 			hw_cr4 &= ~X86_CR4_UMIP;
3070 		} else if (!is_guest_mode(vcpu) ||
3071 			!nested_cpu_has2(get_vmcs12(vcpu), SECONDARY_EXEC_DESC)) {
3072 			secondary_exec_controls_clearbit(vmx, SECONDARY_EXEC_DESC);
3073 		}
3074 	}
3075 
3076 	if (cr4 & X86_CR4_VMXE) {
3077 		/*
3078 		 * To use VMXON (and later other VMX instructions), a guest
3079 		 * must first be able to turn on cr4.VMXE (see handle_vmon()).
3080 		 * So basically the check on whether to allow nested VMX
3081 		 * is here.  We operate under the default treatment of SMM,
3082 		 * so VMX cannot be enabled under SMM.
3083 		 */
3084 		if (!nested_vmx_allowed(vcpu) || is_smm(vcpu))
3085 			return 1;
3086 	}
3087 
3088 	if (vmx->nested.vmxon && !nested_cr4_valid(vcpu, cr4))
3089 		return 1;
3090 
3091 	vcpu->arch.cr4 = cr4;
3092 
3093 	if (!enable_unrestricted_guest) {
3094 		if (enable_ept) {
3095 			if (!is_paging(vcpu)) {
3096 				hw_cr4 &= ~X86_CR4_PAE;
3097 				hw_cr4 |= X86_CR4_PSE;
3098 			} else if (!(cr4 & X86_CR4_PAE)) {
3099 				hw_cr4 &= ~X86_CR4_PAE;
3100 			}
3101 		}
3102 
3103 		/*
3104 		 * SMEP/SMAP/PKU is disabled if CPU is in non-paging mode in
3105 		 * hardware.  To emulate this behavior, SMEP/SMAP/PKU needs
3106 		 * to be manually disabled when guest switches to non-paging
3107 		 * mode.
3108 		 *
3109 		 * If !enable_unrestricted_guest, the CPU is always running
3110 		 * with CR0.PG=1 and CR4 needs to be modified.
3111 		 * If enable_unrestricted_guest, the CPU automatically
3112 		 * disables SMEP/SMAP/PKU when the guest sets CR0.PG=0.
3113 		 */
3114 		if (!is_paging(vcpu))
3115 			hw_cr4 &= ~(X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE);
3116 	}
3117 
3118 	vmcs_writel(CR4_READ_SHADOW, cr4);
3119 	vmcs_writel(GUEST_CR4, hw_cr4);
3120 	return 0;
3121 }
3122 
3123 void vmx_get_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg)
3124 {
3125 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3126 	u32 ar;
3127 
3128 	if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3129 		*var = vmx->rmode.segs[seg];
3130 		if (seg == VCPU_SREG_TR
3131 		    || var->selector == vmx_read_guest_seg_selector(vmx, seg))
3132 			return;
3133 		var->base = vmx_read_guest_seg_base(vmx, seg);
3134 		var->selector = vmx_read_guest_seg_selector(vmx, seg);
3135 		return;
3136 	}
3137 	var->base = vmx_read_guest_seg_base(vmx, seg);
3138 	var->limit = vmx_read_guest_seg_limit(vmx, seg);
3139 	var->selector = vmx_read_guest_seg_selector(vmx, seg);
3140 	ar = vmx_read_guest_seg_ar(vmx, seg);
3141 	var->unusable = (ar >> 16) & 1;
3142 	var->type = ar & 15;
3143 	var->s = (ar >> 4) & 1;
3144 	var->dpl = (ar >> 5) & 3;
3145 	/*
3146 	 * Some userspaces do not preserve unusable property. Since usable
3147 	 * segment has to be present according to VMX spec we can use present
3148 	 * property to amend userspace bug by making unusable segment always
3149 	 * nonpresent. vmx_segment_access_rights() already marks nonpresent
3150 	 * segment as unusable.
3151 	 */
3152 	var->present = !var->unusable;
3153 	var->avl = (ar >> 12) & 1;
3154 	var->l = (ar >> 13) & 1;
3155 	var->db = (ar >> 14) & 1;
3156 	var->g = (ar >> 15) & 1;
3157 }
3158 
3159 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
3160 {
3161 	struct kvm_segment s;
3162 
3163 	if (to_vmx(vcpu)->rmode.vm86_active) {
3164 		vmx_get_segment(vcpu, &s, seg);
3165 		return s.base;
3166 	}
3167 	return vmx_read_guest_seg_base(to_vmx(vcpu), seg);
3168 }
3169 
3170 int vmx_get_cpl(struct kvm_vcpu *vcpu)
3171 {
3172 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3173 
3174 	if (unlikely(vmx->rmode.vm86_active))
3175 		return 0;
3176 	else {
3177 		int ar = vmx_read_guest_seg_ar(vmx, VCPU_SREG_SS);
3178 		return VMX_AR_DPL(ar);
3179 	}
3180 }
3181 
3182 static u32 vmx_segment_access_rights(struct kvm_segment *var)
3183 {
3184 	u32 ar;
3185 
3186 	if (var->unusable || !var->present)
3187 		ar = 1 << 16;
3188 	else {
3189 		ar = var->type & 15;
3190 		ar |= (var->s & 1) << 4;
3191 		ar |= (var->dpl & 3) << 5;
3192 		ar |= (var->present & 1) << 7;
3193 		ar |= (var->avl & 1) << 12;
3194 		ar |= (var->l & 1) << 13;
3195 		ar |= (var->db & 1) << 14;
3196 		ar |= (var->g & 1) << 15;
3197 	}
3198 
3199 	return ar;
3200 }
3201 
3202 void vmx_set_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg)
3203 {
3204 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3205 	const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3206 
3207 	vmx_segment_cache_clear(vmx);
3208 
3209 	if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3210 		vmx->rmode.segs[seg] = *var;
3211 		if (seg == VCPU_SREG_TR)
3212 			vmcs_write16(sf->selector, var->selector);
3213 		else if (var->s)
3214 			fix_rmode_seg(seg, &vmx->rmode.segs[seg]);
3215 		goto out;
3216 	}
3217 
3218 	vmcs_writel(sf->base, var->base);
3219 	vmcs_write32(sf->limit, var->limit);
3220 	vmcs_write16(sf->selector, var->selector);
3221 
3222 	/*
3223 	 *   Fix the "Accessed" bit in AR field of segment registers for older
3224 	 * qemu binaries.
3225 	 *   IA32 arch specifies that at the time of processor reset the
3226 	 * "Accessed" bit in the AR field of segment registers is 1. And qemu
3227 	 * is setting it to 0 in the userland code. This causes invalid guest
3228 	 * state vmexit when "unrestricted guest" mode is turned on.
3229 	 *    Fix for this setup issue in cpu_reset is being pushed in the qemu
3230 	 * tree. Newer qemu binaries with that qemu fix would not need this
3231 	 * kvm hack.
3232 	 */
3233 	if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
3234 		var->type |= 0x1; /* Accessed */
3235 
3236 	vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(var));
3237 
3238 out:
3239 	vmx->emulation_required = emulation_required(vcpu);
3240 }
3241 
3242 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3243 {
3244 	u32 ar = vmx_read_guest_seg_ar(to_vmx(vcpu), VCPU_SREG_CS);
3245 
3246 	*db = (ar >> 14) & 1;
3247 	*l = (ar >> 13) & 1;
3248 }
3249 
3250 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3251 {
3252 	dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
3253 	dt->address = vmcs_readl(GUEST_IDTR_BASE);
3254 }
3255 
3256 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3257 {
3258 	vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
3259 	vmcs_writel(GUEST_IDTR_BASE, dt->address);
3260 }
3261 
3262 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3263 {
3264 	dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
3265 	dt->address = vmcs_readl(GUEST_GDTR_BASE);
3266 }
3267 
3268 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3269 {
3270 	vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
3271 	vmcs_writel(GUEST_GDTR_BASE, dt->address);
3272 }
3273 
3274 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
3275 {
3276 	struct kvm_segment var;
3277 	u32 ar;
3278 
3279 	vmx_get_segment(vcpu, &var, seg);
3280 	var.dpl = 0x3;
3281 	if (seg == VCPU_SREG_CS)
3282 		var.type = 0x3;
3283 	ar = vmx_segment_access_rights(&var);
3284 
3285 	if (var.base != (var.selector << 4))
3286 		return false;
3287 	if (var.limit != 0xffff)
3288 		return false;
3289 	if (ar != 0xf3)
3290 		return false;
3291 
3292 	return true;
3293 }
3294 
3295 static bool code_segment_valid(struct kvm_vcpu *vcpu)
3296 {
3297 	struct kvm_segment cs;
3298 	unsigned int cs_rpl;
3299 
3300 	vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3301 	cs_rpl = cs.selector & SEGMENT_RPL_MASK;
3302 
3303 	if (cs.unusable)
3304 		return false;
3305 	if (~cs.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_ACCESSES_MASK))
3306 		return false;
3307 	if (!cs.s)
3308 		return false;
3309 	if (cs.type & VMX_AR_TYPE_WRITEABLE_MASK) {
3310 		if (cs.dpl > cs_rpl)
3311 			return false;
3312 	} else {
3313 		if (cs.dpl != cs_rpl)
3314 			return false;
3315 	}
3316 	if (!cs.present)
3317 		return false;
3318 
3319 	/* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
3320 	return true;
3321 }
3322 
3323 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
3324 {
3325 	struct kvm_segment ss;
3326 	unsigned int ss_rpl;
3327 
3328 	vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3329 	ss_rpl = ss.selector & SEGMENT_RPL_MASK;
3330 
3331 	if (ss.unusable)
3332 		return true;
3333 	if (ss.type != 3 && ss.type != 7)
3334 		return false;
3335 	if (!ss.s)
3336 		return false;
3337 	if (ss.dpl != ss_rpl) /* DPL != RPL */
3338 		return false;
3339 	if (!ss.present)
3340 		return false;
3341 
3342 	return true;
3343 }
3344 
3345 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
3346 {
3347 	struct kvm_segment var;
3348 	unsigned int rpl;
3349 
3350 	vmx_get_segment(vcpu, &var, seg);
3351 	rpl = var.selector & SEGMENT_RPL_MASK;
3352 
3353 	if (var.unusable)
3354 		return true;
3355 	if (!var.s)
3356 		return false;
3357 	if (!var.present)
3358 		return false;
3359 	if (~var.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_WRITEABLE_MASK)) {
3360 		if (var.dpl < rpl) /* DPL < RPL */
3361 			return false;
3362 	}
3363 
3364 	/* TODO: Add other members to kvm_segment_field to allow checking for other access
3365 	 * rights flags
3366 	 */
3367 	return true;
3368 }
3369 
3370 static bool tr_valid(struct kvm_vcpu *vcpu)
3371 {
3372 	struct kvm_segment tr;
3373 
3374 	vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
3375 
3376 	if (tr.unusable)
3377 		return false;
3378 	if (tr.selector & SEGMENT_TI_MASK)	/* TI = 1 */
3379 		return false;
3380 	if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
3381 		return false;
3382 	if (!tr.present)
3383 		return false;
3384 
3385 	return true;
3386 }
3387 
3388 static bool ldtr_valid(struct kvm_vcpu *vcpu)
3389 {
3390 	struct kvm_segment ldtr;
3391 
3392 	vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
3393 
3394 	if (ldtr.unusable)
3395 		return true;
3396 	if (ldtr.selector & SEGMENT_TI_MASK)	/* TI = 1 */
3397 		return false;
3398 	if (ldtr.type != 2)
3399 		return false;
3400 	if (!ldtr.present)
3401 		return false;
3402 
3403 	return true;
3404 }
3405 
3406 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
3407 {
3408 	struct kvm_segment cs, ss;
3409 
3410 	vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3411 	vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3412 
3413 	return ((cs.selector & SEGMENT_RPL_MASK) ==
3414 		 (ss.selector & SEGMENT_RPL_MASK));
3415 }
3416 
3417 /*
3418  * Check if guest state is valid. Returns true if valid, false if
3419  * not.
3420  * We assume that registers are always usable
3421  */
3422 static bool guest_state_valid(struct kvm_vcpu *vcpu)
3423 {
3424 	if (enable_unrestricted_guest)
3425 		return true;
3426 
3427 	/* real mode guest state checks */
3428 	if (!is_protmode(vcpu) || (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
3429 		if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
3430 			return false;
3431 		if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
3432 			return false;
3433 		if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
3434 			return false;
3435 		if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
3436 			return false;
3437 		if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
3438 			return false;
3439 		if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
3440 			return false;
3441 	} else {
3442 	/* protected mode guest state checks */
3443 		if (!cs_ss_rpl_check(vcpu))
3444 			return false;
3445 		if (!code_segment_valid(vcpu))
3446 			return false;
3447 		if (!stack_segment_valid(vcpu))
3448 			return false;
3449 		if (!data_segment_valid(vcpu, VCPU_SREG_DS))
3450 			return false;
3451 		if (!data_segment_valid(vcpu, VCPU_SREG_ES))
3452 			return false;
3453 		if (!data_segment_valid(vcpu, VCPU_SREG_FS))
3454 			return false;
3455 		if (!data_segment_valid(vcpu, VCPU_SREG_GS))
3456 			return false;
3457 		if (!tr_valid(vcpu))
3458 			return false;
3459 		if (!ldtr_valid(vcpu))
3460 			return false;
3461 	}
3462 	/* TODO:
3463 	 * - Add checks on RIP
3464 	 * - Add checks on RFLAGS
3465 	 */
3466 
3467 	return true;
3468 }
3469 
3470 static int init_rmode_tss(struct kvm *kvm)
3471 {
3472 	gfn_t fn;
3473 	u16 data = 0;
3474 	int idx, r;
3475 
3476 	idx = srcu_read_lock(&kvm->srcu);
3477 	fn = to_kvm_vmx(kvm)->tss_addr >> PAGE_SHIFT;
3478 	r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3479 	if (r < 0)
3480 		goto out;
3481 	data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
3482 	r = kvm_write_guest_page(kvm, fn++, &data,
3483 			TSS_IOPB_BASE_OFFSET, sizeof(u16));
3484 	if (r < 0)
3485 		goto out;
3486 	r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
3487 	if (r < 0)
3488 		goto out;
3489 	r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3490 	if (r < 0)
3491 		goto out;
3492 	data = ~0;
3493 	r = kvm_write_guest_page(kvm, fn, &data,
3494 				 RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
3495 				 sizeof(u8));
3496 out:
3497 	srcu_read_unlock(&kvm->srcu, idx);
3498 	return r;
3499 }
3500 
3501 static int init_rmode_identity_map(struct kvm *kvm)
3502 {
3503 	struct kvm_vmx *kvm_vmx = to_kvm_vmx(kvm);
3504 	int i, r = 0;
3505 	kvm_pfn_t identity_map_pfn;
3506 	u32 tmp;
3507 
3508 	/* Protect kvm_vmx->ept_identity_pagetable_done. */
3509 	mutex_lock(&kvm->slots_lock);
3510 
3511 	if (likely(kvm_vmx->ept_identity_pagetable_done))
3512 		goto out;
3513 
3514 	if (!kvm_vmx->ept_identity_map_addr)
3515 		kvm_vmx->ept_identity_map_addr = VMX_EPT_IDENTITY_PAGETABLE_ADDR;
3516 	identity_map_pfn = kvm_vmx->ept_identity_map_addr >> PAGE_SHIFT;
3517 
3518 	r = __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
3519 				    kvm_vmx->ept_identity_map_addr, PAGE_SIZE);
3520 	if (r < 0)
3521 		goto out;
3522 
3523 	r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
3524 	if (r < 0)
3525 		goto out;
3526 	/* Set up identity-mapping pagetable for EPT in real mode */
3527 	for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
3528 		tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
3529 			_PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
3530 		r = kvm_write_guest_page(kvm, identity_map_pfn,
3531 				&tmp, i * sizeof(tmp), sizeof(tmp));
3532 		if (r < 0)
3533 			goto out;
3534 	}
3535 	kvm_vmx->ept_identity_pagetable_done = true;
3536 
3537 out:
3538 	mutex_unlock(&kvm->slots_lock);
3539 	return r;
3540 }
3541 
3542 static void seg_setup(int seg)
3543 {
3544 	const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3545 	unsigned int ar;
3546 
3547 	vmcs_write16(sf->selector, 0);
3548 	vmcs_writel(sf->base, 0);
3549 	vmcs_write32(sf->limit, 0xffff);
3550 	ar = 0x93;
3551 	if (seg == VCPU_SREG_CS)
3552 		ar |= 0x08; /* code segment */
3553 
3554 	vmcs_write32(sf->ar_bytes, ar);
3555 }
3556 
3557 static int alloc_apic_access_page(struct kvm *kvm)
3558 {
3559 	struct page *page;
3560 	int r = 0;
3561 
3562 	mutex_lock(&kvm->slots_lock);
3563 	if (kvm->arch.apic_access_page_done)
3564 		goto out;
3565 	r = __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
3566 				    APIC_DEFAULT_PHYS_BASE, PAGE_SIZE);
3567 	if (r)
3568 		goto out;
3569 
3570 	page = gfn_to_page(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
3571 	if (is_error_page(page)) {
3572 		r = -EFAULT;
3573 		goto out;
3574 	}
3575 
3576 	/*
3577 	 * Do not pin the page in memory, so that memory hot-unplug
3578 	 * is able to migrate it.
3579 	 */
3580 	put_page(page);
3581 	kvm->arch.apic_access_page_done = true;
3582 out:
3583 	mutex_unlock(&kvm->slots_lock);
3584 	return r;
3585 }
3586 
3587 int allocate_vpid(void)
3588 {
3589 	int vpid;
3590 
3591 	if (!enable_vpid)
3592 		return 0;
3593 	spin_lock(&vmx_vpid_lock);
3594 	vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
3595 	if (vpid < VMX_NR_VPIDS)
3596 		__set_bit(vpid, vmx_vpid_bitmap);
3597 	else
3598 		vpid = 0;
3599 	spin_unlock(&vmx_vpid_lock);
3600 	return vpid;
3601 }
3602 
3603 void free_vpid(int vpid)
3604 {
3605 	if (!enable_vpid || vpid == 0)
3606 		return;
3607 	spin_lock(&vmx_vpid_lock);
3608 	__clear_bit(vpid, vmx_vpid_bitmap);
3609 	spin_unlock(&vmx_vpid_lock);
3610 }
3611 
3612 static __always_inline void vmx_disable_intercept_for_msr(unsigned long *msr_bitmap,
3613 							  u32 msr, int type)
3614 {
3615 	int f = sizeof(unsigned long);
3616 
3617 	if (!cpu_has_vmx_msr_bitmap())
3618 		return;
3619 
3620 	if (static_branch_unlikely(&enable_evmcs))
3621 		evmcs_touch_msr_bitmap();
3622 
3623 	/*
3624 	 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
3625 	 * have the write-low and read-high bitmap offsets the wrong way round.
3626 	 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
3627 	 */
3628 	if (msr <= 0x1fff) {
3629 		if (type & MSR_TYPE_R)
3630 			/* read-low */
3631 			__clear_bit(msr, msr_bitmap + 0x000 / f);
3632 
3633 		if (type & MSR_TYPE_W)
3634 			/* write-low */
3635 			__clear_bit(msr, msr_bitmap + 0x800 / f);
3636 
3637 	} else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
3638 		msr &= 0x1fff;
3639 		if (type & MSR_TYPE_R)
3640 			/* read-high */
3641 			__clear_bit(msr, msr_bitmap + 0x400 / f);
3642 
3643 		if (type & MSR_TYPE_W)
3644 			/* write-high */
3645 			__clear_bit(msr, msr_bitmap + 0xc00 / f);
3646 
3647 	}
3648 }
3649 
3650 static __always_inline void vmx_enable_intercept_for_msr(unsigned long *msr_bitmap,
3651 							 u32 msr, int type)
3652 {
3653 	int f = sizeof(unsigned long);
3654 
3655 	if (!cpu_has_vmx_msr_bitmap())
3656 		return;
3657 
3658 	if (static_branch_unlikely(&enable_evmcs))
3659 		evmcs_touch_msr_bitmap();
3660 
3661 	/*
3662 	 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
3663 	 * have the write-low and read-high bitmap offsets the wrong way round.
3664 	 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
3665 	 */
3666 	if (msr <= 0x1fff) {
3667 		if (type & MSR_TYPE_R)
3668 			/* read-low */
3669 			__set_bit(msr, msr_bitmap + 0x000 / f);
3670 
3671 		if (type & MSR_TYPE_W)
3672 			/* write-low */
3673 			__set_bit(msr, msr_bitmap + 0x800 / f);
3674 
3675 	} else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
3676 		msr &= 0x1fff;
3677 		if (type & MSR_TYPE_R)
3678 			/* read-high */
3679 			__set_bit(msr, msr_bitmap + 0x400 / f);
3680 
3681 		if (type & MSR_TYPE_W)
3682 			/* write-high */
3683 			__set_bit(msr, msr_bitmap + 0xc00 / f);
3684 
3685 	}
3686 }
3687 
3688 static __always_inline void vmx_set_intercept_for_msr(unsigned long *msr_bitmap,
3689 			     			      u32 msr, int type, bool value)
3690 {
3691 	if (value)
3692 		vmx_enable_intercept_for_msr(msr_bitmap, msr, type);
3693 	else
3694 		vmx_disable_intercept_for_msr(msr_bitmap, msr, type);
3695 }
3696 
3697 static u8 vmx_msr_bitmap_mode(struct kvm_vcpu *vcpu)
3698 {
3699 	u8 mode = 0;
3700 
3701 	if (cpu_has_secondary_exec_ctrls() &&
3702 	    (secondary_exec_controls_get(to_vmx(vcpu)) &
3703 	     SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE)) {
3704 		mode |= MSR_BITMAP_MODE_X2APIC;
3705 		if (enable_apicv && kvm_vcpu_apicv_active(vcpu))
3706 			mode |= MSR_BITMAP_MODE_X2APIC_APICV;
3707 	}
3708 
3709 	return mode;
3710 }
3711 
3712 static void vmx_update_msr_bitmap_x2apic(unsigned long *msr_bitmap,
3713 					 u8 mode)
3714 {
3715 	int msr;
3716 
3717 	for (msr = 0x800; msr <= 0x8ff; msr += BITS_PER_LONG) {
3718 		unsigned word = msr / BITS_PER_LONG;
3719 		msr_bitmap[word] = (mode & MSR_BITMAP_MODE_X2APIC_APICV) ? 0 : ~0;
3720 		msr_bitmap[word + (0x800 / sizeof(long))] = ~0;
3721 	}
3722 
3723 	if (mode & MSR_BITMAP_MODE_X2APIC) {
3724 		/*
3725 		 * TPR reads and writes can be virtualized even if virtual interrupt
3726 		 * delivery is not in use.
3727 		 */
3728 		vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_TASKPRI), MSR_TYPE_RW);
3729 		if (mode & MSR_BITMAP_MODE_X2APIC_APICV) {
3730 			vmx_enable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_TMCCT), MSR_TYPE_R);
3731 			vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_EOI), MSR_TYPE_W);
3732 			vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_SELF_IPI), MSR_TYPE_W);
3733 		}
3734 	}
3735 }
3736 
3737 void vmx_update_msr_bitmap(struct kvm_vcpu *vcpu)
3738 {
3739 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3740 	unsigned long *msr_bitmap = vmx->vmcs01.msr_bitmap;
3741 	u8 mode = vmx_msr_bitmap_mode(vcpu);
3742 	u8 changed = mode ^ vmx->msr_bitmap_mode;
3743 
3744 	if (!changed)
3745 		return;
3746 
3747 	if (changed & (MSR_BITMAP_MODE_X2APIC | MSR_BITMAP_MODE_X2APIC_APICV))
3748 		vmx_update_msr_bitmap_x2apic(msr_bitmap, mode);
3749 
3750 	vmx->msr_bitmap_mode = mode;
3751 }
3752 
3753 void pt_update_intercept_for_msr(struct vcpu_vmx *vmx)
3754 {
3755 	unsigned long *msr_bitmap = vmx->vmcs01.msr_bitmap;
3756 	bool flag = !(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN);
3757 	u32 i;
3758 
3759 	vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_STATUS,
3760 							MSR_TYPE_RW, flag);
3761 	vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_OUTPUT_BASE,
3762 							MSR_TYPE_RW, flag);
3763 	vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_OUTPUT_MASK,
3764 							MSR_TYPE_RW, flag);
3765 	vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_CR3_MATCH,
3766 							MSR_TYPE_RW, flag);
3767 	for (i = 0; i < vmx->pt_desc.addr_range; i++) {
3768 		vmx_set_intercept_for_msr(msr_bitmap,
3769 			MSR_IA32_RTIT_ADDR0_A + i * 2, MSR_TYPE_RW, flag);
3770 		vmx_set_intercept_for_msr(msr_bitmap,
3771 			MSR_IA32_RTIT_ADDR0_B + i * 2, MSR_TYPE_RW, flag);
3772 	}
3773 }
3774 
3775 static bool vmx_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
3776 {
3777 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3778 	void *vapic_page;
3779 	u32 vppr;
3780 	int rvi;
3781 
3782 	if (WARN_ON_ONCE(!is_guest_mode(vcpu)) ||
3783 		!nested_cpu_has_vid(get_vmcs12(vcpu)) ||
3784 		WARN_ON_ONCE(!vmx->nested.virtual_apic_map.gfn))
3785 		return false;
3786 
3787 	rvi = vmx_get_rvi();
3788 
3789 	vapic_page = vmx->nested.virtual_apic_map.hva;
3790 	vppr = *((u32 *)(vapic_page + APIC_PROCPRI));
3791 
3792 	return ((rvi & 0xf0) > (vppr & 0xf0));
3793 }
3794 
3795 static inline bool kvm_vcpu_trigger_posted_interrupt(struct kvm_vcpu *vcpu,
3796 						     bool nested)
3797 {
3798 #ifdef CONFIG_SMP
3799 	int pi_vec = nested ? POSTED_INTR_NESTED_VECTOR : POSTED_INTR_VECTOR;
3800 
3801 	if (vcpu->mode == IN_GUEST_MODE) {
3802 		/*
3803 		 * The vector of interrupt to be delivered to vcpu had
3804 		 * been set in PIR before this function.
3805 		 *
3806 		 * Following cases will be reached in this block, and
3807 		 * we always send a notification event in all cases as
3808 		 * explained below.
3809 		 *
3810 		 * Case 1: vcpu keeps in non-root mode. Sending a
3811 		 * notification event posts the interrupt to vcpu.
3812 		 *
3813 		 * Case 2: vcpu exits to root mode and is still
3814 		 * runnable. PIR will be synced to vIRR before the
3815 		 * next vcpu entry. Sending a notification event in
3816 		 * this case has no effect, as vcpu is not in root
3817 		 * mode.
3818 		 *
3819 		 * Case 3: vcpu exits to root mode and is blocked.
3820 		 * vcpu_block() has already synced PIR to vIRR and
3821 		 * never blocks vcpu if vIRR is not cleared. Therefore,
3822 		 * a blocked vcpu here does not wait for any requested
3823 		 * interrupts in PIR, and sending a notification event
3824 		 * which has no effect is safe here.
3825 		 */
3826 
3827 		apic->send_IPI_mask(get_cpu_mask(vcpu->cpu), pi_vec);
3828 		return true;
3829 	}
3830 #endif
3831 	return false;
3832 }
3833 
3834 static int vmx_deliver_nested_posted_interrupt(struct kvm_vcpu *vcpu,
3835 						int vector)
3836 {
3837 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3838 
3839 	if (is_guest_mode(vcpu) &&
3840 	    vector == vmx->nested.posted_intr_nv) {
3841 		/*
3842 		 * If a posted intr is not recognized by hardware,
3843 		 * we will accomplish it in the next vmentry.
3844 		 */
3845 		vmx->nested.pi_pending = true;
3846 		kvm_make_request(KVM_REQ_EVENT, vcpu);
3847 		/* the PIR and ON have been set by L1. */
3848 		if (!kvm_vcpu_trigger_posted_interrupt(vcpu, true))
3849 			kvm_vcpu_kick(vcpu);
3850 		return 0;
3851 	}
3852 	return -1;
3853 }
3854 /*
3855  * Send interrupt to vcpu via posted interrupt way.
3856  * 1. If target vcpu is running(non-root mode), send posted interrupt
3857  * notification to vcpu and hardware will sync PIR to vIRR atomically.
3858  * 2. If target vcpu isn't running(root mode), kick it to pick up the
3859  * interrupt from PIR in next vmentry.
3860  */
3861 static int vmx_deliver_posted_interrupt(struct kvm_vcpu *vcpu, int vector)
3862 {
3863 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3864 	int r;
3865 
3866 	r = vmx_deliver_nested_posted_interrupt(vcpu, vector);
3867 	if (!r)
3868 		return 0;
3869 
3870 	if (!vcpu->arch.apicv_active)
3871 		return -1;
3872 
3873 	if (pi_test_and_set_pir(vector, &vmx->pi_desc))
3874 		return 0;
3875 
3876 	/* If a previous notification has sent the IPI, nothing to do.  */
3877 	if (pi_test_and_set_on(&vmx->pi_desc))
3878 		return 0;
3879 
3880 	if (!kvm_vcpu_trigger_posted_interrupt(vcpu, false))
3881 		kvm_vcpu_kick(vcpu);
3882 
3883 	return 0;
3884 }
3885 
3886 /*
3887  * Set up the vmcs's constant host-state fields, i.e., host-state fields that
3888  * will not change in the lifetime of the guest.
3889  * Note that host-state that does change is set elsewhere. E.g., host-state
3890  * that is set differently for each CPU is set in vmx_vcpu_load(), not here.
3891  */
3892 void vmx_set_constant_host_state(struct vcpu_vmx *vmx)
3893 {
3894 	u32 low32, high32;
3895 	unsigned long tmpl;
3896 	unsigned long cr0, cr3, cr4;
3897 
3898 	cr0 = read_cr0();
3899 	WARN_ON(cr0 & X86_CR0_TS);
3900 	vmcs_writel(HOST_CR0, cr0);  /* 22.2.3 */
3901 
3902 	/*
3903 	 * Save the most likely value for this task's CR3 in the VMCS.
3904 	 * We can't use __get_current_cr3_fast() because we're not atomic.
3905 	 */
3906 	cr3 = __read_cr3();
3907 	vmcs_writel(HOST_CR3, cr3);		/* 22.2.3  FIXME: shadow tables */
3908 	vmx->loaded_vmcs->host_state.cr3 = cr3;
3909 
3910 	/* Save the most likely value for this task's CR4 in the VMCS. */
3911 	cr4 = cr4_read_shadow();
3912 	vmcs_writel(HOST_CR4, cr4);			/* 22.2.3, 22.2.5 */
3913 	vmx->loaded_vmcs->host_state.cr4 = cr4;
3914 
3915 	vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS);  /* 22.2.4 */
3916 #ifdef CONFIG_X86_64
3917 	/*
3918 	 * Load null selectors, so we can avoid reloading them in
3919 	 * vmx_prepare_switch_to_host(), in case userspace uses
3920 	 * the null selectors too (the expected case).
3921 	 */
3922 	vmcs_write16(HOST_DS_SELECTOR, 0);
3923 	vmcs_write16(HOST_ES_SELECTOR, 0);
3924 #else
3925 	vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
3926 	vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
3927 #endif
3928 	vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
3929 	vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8);  /* 22.2.4 */
3930 
3931 	vmcs_writel(HOST_IDTR_BASE, host_idt_base);   /* 22.2.4 */
3932 
3933 	vmcs_writel(HOST_RIP, (unsigned long)vmx_vmexit); /* 22.2.5 */
3934 
3935 	rdmsr(MSR_IA32_SYSENTER_CS, low32, high32);
3936 	vmcs_write32(HOST_IA32_SYSENTER_CS, low32);
3937 	rdmsrl(MSR_IA32_SYSENTER_EIP, tmpl);
3938 	vmcs_writel(HOST_IA32_SYSENTER_EIP, tmpl);   /* 22.2.3 */
3939 
3940 	if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
3941 		rdmsr(MSR_IA32_CR_PAT, low32, high32);
3942 		vmcs_write64(HOST_IA32_PAT, low32 | ((u64) high32 << 32));
3943 	}
3944 
3945 	if (cpu_has_load_ia32_efer())
3946 		vmcs_write64(HOST_IA32_EFER, host_efer);
3947 }
3948 
3949 void set_cr4_guest_host_mask(struct vcpu_vmx *vmx)
3950 {
3951 	vmx->vcpu.arch.cr4_guest_owned_bits = KVM_CR4_GUEST_OWNED_BITS;
3952 	if (enable_ept)
3953 		vmx->vcpu.arch.cr4_guest_owned_bits |= X86_CR4_PGE;
3954 	if (is_guest_mode(&vmx->vcpu))
3955 		vmx->vcpu.arch.cr4_guest_owned_bits &=
3956 			~get_vmcs12(&vmx->vcpu)->cr4_guest_host_mask;
3957 	vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits);
3958 }
3959 
3960 u32 vmx_pin_based_exec_ctrl(struct vcpu_vmx *vmx)
3961 {
3962 	u32 pin_based_exec_ctrl = vmcs_config.pin_based_exec_ctrl;
3963 
3964 	if (!kvm_vcpu_apicv_active(&vmx->vcpu))
3965 		pin_based_exec_ctrl &= ~PIN_BASED_POSTED_INTR;
3966 
3967 	if (!enable_vnmi)
3968 		pin_based_exec_ctrl &= ~PIN_BASED_VIRTUAL_NMIS;
3969 
3970 	if (!enable_preemption_timer)
3971 		pin_based_exec_ctrl &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
3972 
3973 	return pin_based_exec_ctrl;
3974 }
3975 
3976 static void vmx_refresh_apicv_exec_ctrl(struct kvm_vcpu *vcpu)
3977 {
3978 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3979 
3980 	pin_controls_set(vmx, vmx_pin_based_exec_ctrl(vmx));
3981 	if (cpu_has_secondary_exec_ctrls()) {
3982 		if (kvm_vcpu_apicv_active(vcpu))
3983 			secondary_exec_controls_setbit(vmx,
3984 				      SECONDARY_EXEC_APIC_REGISTER_VIRT |
3985 				      SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
3986 		else
3987 			secondary_exec_controls_clearbit(vmx,
3988 					SECONDARY_EXEC_APIC_REGISTER_VIRT |
3989 					SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
3990 	}
3991 
3992 	if (cpu_has_vmx_msr_bitmap())
3993 		vmx_update_msr_bitmap(vcpu);
3994 }
3995 
3996 u32 vmx_exec_control(struct vcpu_vmx *vmx)
3997 {
3998 	u32 exec_control = vmcs_config.cpu_based_exec_ctrl;
3999 
4000 	if (vmx->vcpu.arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)
4001 		exec_control &= ~CPU_BASED_MOV_DR_EXITING;
4002 
4003 	if (!cpu_need_tpr_shadow(&vmx->vcpu)) {
4004 		exec_control &= ~CPU_BASED_TPR_SHADOW;
4005 #ifdef CONFIG_X86_64
4006 		exec_control |= CPU_BASED_CR8_STORE_EXITING |
4007 				CPU_BASED_CR8_LOAD_EXITING;
4008 #endif
4009 	}
4010 	if (!enable_ept)
4011 		exec_control |= CPU_BASED_CR3_STORE_EXITING |
4012 				CPU_BASED_CR3_LOAD_EXITING  |
4013 				CPU_BASED_INVLPG_EXITING;
4014 	if (kvm_mwait_in_guest(vmx->vcpu.kvm))
4015 		exec_control &= ~(CPU_BASED_MWAIT_EXITING |
4016 				CPU_BASED_MONITOR_EXITING);
4017 	if (kvm_hlt_in_guest(vmx->vcpu.kvm))
4018 		exec_control &= ~CPU_BASED_HLT_EXITING;
4019 	return exec_control;
4020 }
4021 
4022 
4023 static void vmx_compute_secondary_exec_control(struct vcpu_vmx *vmx)
4024 {
4025 	struct kvm_vcpu *vcpu = &vmx->vcpu;
4026 
4027 	u32 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
4028 
4029 	if (pt_mode == PT_MODE_SYSTEM)
4030 		exec_control &= ~(SECONDARY_EXEC_PT_USE_GPA | SECONDARY_EXEC_PT_CONCEAL_VMX);
4031 	if (!cpu_need_virtualize_apic_accesses(vcpu))
4032 		exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
4033 	if (vmx->vpid == 0)
4034 		exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
4035 	if (!enable_ept) {
4036 		exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
4037 		enable_unrestricted_guest = 0;
4038 	}
4039 	if (!enable_unrestricted_guest)
4040 		exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
4041 	if (kvm_pause_in_guest(vmx->vcpu.kvm))
4042 		exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
4043 	if (!kvm_vcpu_apicv_active(vcpu))
4044 		exec_control &= ~(SECONDARY_EXEC_APIC_REGISTER_VIRT |
4045 				  SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
4046 	exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
4047 
4048 	/* SECONDARY_EXEC_DESC is enabled/disabled on writes to CR4.UMIP,
4049 	 * in vmx_set_cr4.  */
4050 	exec_control &= ~SECONDARY_EXEC_DESC;
4051 
4052 	/* SECONDARY_EXEC_SHADOW_VMCS is enabled when L1 executes VMPTRLD
4053 	   (handle_vmptrld).
4054 	   We can NOT enable shadow_vmcs here because we don't have yet
4055 	   a current VMCS12
4056 	*/
4057 	exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
4058 
4059 	if (!enable_pml)
4060 		exec_control &= ~SECONDARY_EXEC_ENABLE_PML;
4061 
4062 	if (vmx_xsaves_supported()) {
4063 		/* Exposing XSAVES only when XSAVE is exposed */
4064 		bool xsaves_enabled =
4065 			boot_cpu_has(X86_FEATURE_XSAVE) &&
4066 			guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
4067 			guest_cpuid_has(vcpu, X86_FEATURE_XSAVES);
4068 
4069 		vcpu->arch.xsaves_enabled = xsaves_enabled;
4070 
4071 		if (!xsaves_enabled)
4072 			exec_control &= ~SECONDARY_EXEC_XSAVES;
4073 
4074 		if (nested) {
4075 			if (xsaves_enabled)
4076 				vmx->nested.msrs.secondary_ctls_high |=
4077 					SECONDARY_EXEC_XSAVES;
4078 			else
4079 				vmx->nested.msrs.secondary_ctls_high &=
4080 					~SECONDARY_EXEC_XSAVES;
4081 		}
4082 	}
4083 
4084 	if (vmx_rdtscp_supported()) {
4085 		bool rdtscp_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP);
4086 		if (!rdtscp_enabled)
4087 			exec_control &= ~SECONDARY_EXEC_RDTSCP;
4088 
4089 		if (nested) {
4090 			if (rdtscp_enabled)
4091 				vmx->nested.msrs.secondary_ctls_high |=
4092 					SECONDARY_EXEC_RDTSCP;
4093 			else
4094 				vmx->nested.msrs.secondary_ctls_high &=
4095 					~SECONDARY_EXEC_RDTSCP;
4096 		}
4097 	}
4098 
4099 	if (vmx_invpcid_supported()) {
4100 		/* Exposing INVPCID only when PCID is exposed */
4101 		bool invpcid_enabled =
4102 			guest_cpuid_has(vcpu, X86_FEATURE_INVPCID) &&
4103 			guest_cpuid_has(vcpu, X86_FEATURE_PCID);
4104 
4105 		if (!invpcid_enabled) {
4106 			exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID;
4107 			guest_cpuid_clear(vcpu, X86_FEATURE_INVPCID);
4108 		}
4109 
4110 		if (nested) {
4111 			if (invpcid_enabled)
4112 				vmx->nested.msrs.secondary_ctls_high |=
4113 					SECONDARY_EXEC_ENABLE_INVPCID;
4114 			else
4115 				vmx->nested.msrs.secondary_ctls_high &=
4116 					~SECONDARY_EXEC_ENABLE_INVPCID;
4117 		}
4118 	}
4119 
4120 	if (vmx_rdrand_supported()) {
4121 		bool rdrand_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDRAND);
4122 		if (rdrand_enabled)
4123 			exec_control &= ~SECONDARY_EXEC_RDRAND_EXITING;
4124 
4125 		if (nested) {
4126 			if (rdrand_enabled)
4127 				vmx->nested.msrs.secondary_ctls_high |=
4128 					SECONDARY_EXEC_RDRAND_EXITING;
4129 			else
4130 				vmx->nested.msrs.secondary_ctls_high &=
4131 					~SECONDARY_EXEC_RDRAND_EXITING;
4132 		}
4133 	}
4134 
4135 	if (vmx_rdseed_supported()) {
4136 		bool rdseed_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDSEED);
4137 		if (rdseed_enabled)
4138 			exec_control &= ~SECONDARY_EXEC_RDSEED_EXITING;
4139 
4140 		if (nested) {
4141 			if (rdseed_enabled)
4142 				vmx->nested.msrs.secondary_ctls_high |=
4143 					SECONDARY_EXEC_RDSEED_EXITING;
4144 			else
4145 				vmx->nested.msrs.secondary_ctls_high &=
4146 					~SECONDARY_EXEC_RDSEED_EXITING;
4147 		}
4148 	}
4149 
4150 	if (vmx_waitpkg_supported()) {
4151 		bool waitpkg_enabled =
4152 			guest_cpuid_has(vcpu, X86_FEATURE_WAITPKG);
4153 
4154 		if (!waitpkg_enabled)
4155 			exec_control &= ~SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE;
4156 
4157 		if (nested) {
4158 			if (waitpkg_enabled)
4159 				vmx->nested.msrs.secondary_ctls_high |=
4160 					SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE;
4161 			else
4162 				vmx->nested.msrs.secondary_ctls_high &=
4163 					~SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE;
4164 		}
4165 	}
4166 
4167 	vmx->secondary_exec_control = exec_control;
4168 }
4169 
4170 static void ept_set_mmio_spte_mask(void)
4171 {
4172 	/*
4173 	 * EPT Misconfigurations can be generated if the value of bits 2:0
4174 	 * of an EPT paging-structure entry is 110b (write/execute).
4175 	 */
4176 	kvm_mmu_set_mmio_spte_mask(VMX_EPT_RWX_MASK,
4177 				   VMX_EPT_MISCONFIG_WX_VALUE, 0);
4178 }
4179 
4180 #define VMX_XSS_EXIT_BITMAP 0
4181 
4182 /*
4183  * Noting that the initialization of Guest-state Area of VMCS is in
4184  * vmx_vcpu_reset().
4185  */
4186 static void init_vmcs(struct vcpu_vmx *vmx)
4187 {
4188 	if (nested)
4189 		nested_vmx_set_vmcs_shadowing_bitmap();
4190 
4191 	if (cpu_has_vmx_msr_bitmap())
4192 		vmcs_write64(MSR_BITMAP, __pa(vmx->vmcs01.msr_bitmap));
4193 
4194 	vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
4195 
4196 	/* Control */
4197 	pin_controls_set(vmx, vmx_pin_based_exec_ctrl(vmx));
4198 
4199 	exec_controls_set(vmx, vmx_exec_control(vmx));
4200 
4201 	if (cpu_has_secondary_exec_ctrls()) {
4202 		vmx_compute_secondary_exec_control(vmx);
4203 		secondary_exec_controls_set(vmx, vmx->secondary_exec_control);
4204 	}
4205 
4206 	if (kvm_vcpu_apicv_active(&vmx->vcpu)) {
4207 		vmcs_write64(EOI_EXIT_BITMAP0, 0);
4208 		vmcs_write64(EOI_EXIT_BITMAP1, 0);
4209 		vmcs_write64(EOI_EXIT_BITMAP2, 0);
4210 		vmcs_write64(EOI_EXIT_BITMAP3, 0);
4211 
4212 		vmcs_write16(GUEST_INTR_STATUS, 0);
4213 
4214 		vmcs_write16(POSTED_INTR_NV, POSTED_INTR_VECTOR);
4215 		vmcs_write64(POSTED_INTR_DESC_ADDR, __pa((&vmx->pi_desc)));
4216 	}
4217 
4218 	if (!kvm_pause_in_guest(vmx->vcpu.kvm)) {
4219 		vmcs_write32(PLE_GAP, ple_gap);
4220 		vmx->ple_window = ple_window;
4221 		vmx->ple_window_dirty = true;
4222 	}
4223 
4224 	vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
4225 	vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
4226 	vmcs_write32(CR3_TARGET_COUNT, 0);           /* 22.2.1 */
4227 
4228 	vmcs_write16(HOST_FS_SELECTOR, 0);            /* 22.2.4 */
4229 	vmcs_write16(HOST_GS_SELECTOR, 0);            /* 22.2.4 */
4230 	vmx_set_constant_host_state(vmx);
4231 	vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
4232 	vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
4233 
4234 	if (cpu_has_vmx_vmfunc())
4235 		vmcs_write64(VM_FUNCTION_CONTROL, 0);
4236 
4237 	vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
4238 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
4239 	vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host.val));
4240 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
4241 	vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest.val));
4242 
4243 	if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
4244 		vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
4245 
4246 	vm_exit_controls_set(vmx, vmx_vmexit_ctrl());
4247 
4248 	/* 22.2.1, 20.8.1 */
4249 	vm_entry_controls_set(vmx, vmx_vmentry_ctrl());
4250 
4251 	vmx->vcpu.arch.cr0_guest_owned_bits = X86_CR0_TS;
4252 	vmcs_writel(CR0_GUEST_HOST_MASK, ~X86_CR0_TS);
4253 
4254 	set_cr4_guest_host_mask(vmx);
4255 
4256 	if (vmx->vpid != 0)
4257 		vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
4258 
4259 	if (vmx_xsaves_supported())
4260 		vmcs_write64(XSS_EXIT_BITMAP, VMX_XSS_EXIT_BITMAP);
4261 
4262 	if (enable_pml) {
4263 		vmcs_write64(PML_ADDRESS, page_to_phys(vmx->pml_pg));
4264 		vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
4265 	}
4266 
4267 	if (cpu_has_vmx_encls_vmexit())
4268 		vmcs_write64(ENCLS_EXITING_BITMAP, -1ull);
4269 
4270 	if (pt_mode == PT_MODE_HOST_GUEST) {
4271 		memset(&vmx->pt_desc, 0, sizeof(vmx->pt_desc));
4272 		/* Bit[6~0] are forced to 1, writes are ignored. */
4273 		vmx->pt_desc.guest.output_mask = 0x7F;
4274 		vmcs_write64(GUEST_IA32_RTIT_CTL, 0);
4275 	}
4276 }
4277 
4278 static void vmx_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
4279 {
4280 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4281 	struct msr_data apic_base_msr;
4282 	u64 cr0;
4283 
4284 	vmx->rmode.vm86_active = 0;
4285 	vmx->spec_ctrl = 0;
4286 
4287 	vmx->msr_ia32_umwait_control = 0;
4288 
4289 	vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
4290 	vmx->hv_deadline_tsc = -1;
4291 	kvm_set_cr8(vcpu, 0);
4292 
4293 	if (!init_event) {
4294 		apic_base_msr.data = APIC_DEFAULT_PHYS_BASE |
4295 				     MSR_IA32_APICBASE_ENABLE;
4296 		if (kvm_vcpu_is_reset_bsp(vcpu))
4297 			apic_base_msr.data |= MSR_IA32_APICBASE_BSP;
4298 		apic_base_msr.host_initiated = true;
4299 		kvm_set_apic_base(vcpu, &apic_base_msr);
4300 	}
4301 
4302 	vmx_segment_cache_clear(vmx);
4303 
4304 	seg_setup(VCPU_SREG_CS);
4305 	vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
4306 	vmcs_writel(GUEST_CS_BASE, 0xffff0000ul);
4307 
4308 	seg_setup(VCPU_SREG_DS);
4309 	seg_setup(VCPU_SREG_ES);
4310 	seg_setup(VCPU_SREG_FS);
4311 	seg_setup(VCPU_SREG_GS);
4312 	seg_setup(VCPU_SREG_SS);
4313 
4314 	vmcs_write16(GUEST_TR_SELECTOR, 0);
4315 	vmcs_writel(GUEST_TR_BASE, 0);
4316 	vmcs_write32(GUEST_TR_LIMIT, 0xffff);
4317 	vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
4318 
4319 	vmcs_write16(GUEST_LDTR_SELECTOR, 0);
4320 	vmcs_writel(GUEST_LDTR_BASE, 0);
4321 	vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
4322 	vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
4323 
4324 	if (!init_event) {
4325 		vmcs_write32(GUEST_SYSENTER_CS, 0);
4326 		vmcs_writel(GUEST_SYSENTER_ESP, 0);
4327 		vmcs_writel(GUEST_SYSENTER_EIP, 0);
4328 		vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
4329 	}
4330 
4331 	kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
4332 	kvm_rip_write(vcpu, 0xfff0);
4333 
4334 	vmcs_writel(GUEST_GDTR_BASE, 0);
4335 	vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
4336 
4337 	vmcs_writel(GUEST_IDTR_BASE, 0);
4338 	vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
4339 
4340 	vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
4341 	vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
4342 	vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS, 0);
4343 	if (kvm_mpx_supported())
4344 		vmcs_write64(GUEST_BNDCFGS, 0);
4345 
4346 	setup_msrs(vmx);
4347 
4348 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);  /* 22.2.1 */
4349 
4350 	if (cpu_has_vmx_tpr_shadow() && !init_event) {
4351 		vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
4352 		if (cpu_need_tpr_shadow(vcpu))
4353 			vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
4354 				     __pa(vcpu->arch.apic->regs));
4355 		vmcs_write32(TPR_THRESHOLD, 0);
4356 	}
4357 
4358 	kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
4359 
4360 	cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
4361 	vmx->vcpu.arch.cr0 = cr0;
4362 	vmx_set_cr0(vcpu, cr0); /* enter rmode */
4363 	vmx_set_cr4(vcpu, 0);
4364 	vmx_set_efer(vcpu, 0);
4365 
4366 	update_exception_bitmap(vcpu);
4367 
4368 	vpid_sync_context(vmx->vpid);
4369 	if (init_event)
4370 		vmx_clear_hlt(vcpu);
4371 }
4372 
4373 static void enable_irq_window(struct kvm_vcpu *vcpu)
4374 {
4375 	exec_controls_setbit(to_vmx(vcpu), CPU_BASED_INTR_WINDOW_EXITING);
4376 }
4377 
4378 static void enable_nmi_window(struct kvm_vcpu *vcpu)
4379 {
4380 	if (!enable_vnmi ||
4381 	    vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_STI) {
4382 		enable_irq_window(vcpu);
4383 		return;
4384 	}
4385 
4386 	exec_controls_setbit(to_vmx(vcpu), CPU_BASED_NMI_WINDOW_EXITING);
4387 }
4388 
4389 static void vmx_inject_irq(struct kvm_vcpu *vcpu)
4390 {
4391 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4392 	uint32_t intr;
4393 	int irq = vcpu->arch.interrupt.nr;
4394 
4395 	trace_kvm_inj_virq(irq);
4396 
4397 	++vcpu->stat.irq_injections;
4398 	if (vmx->rmode.vm86_active) {
4399 		int inc_eip = 0;
4400 		if (vcpu->arch.interrupt.soft)
4401 			inc_eip = vcpu->arch.event_exit_inst_len;
4402 		kvm_inject_realmode_interrupt(vcpu, irq, inc_eip);
4403 		return;
4404 	}
4405 	intr = irq | INTR_INFO_VALID_MASK;
4406 	if (vcpu->arch.interrupt.soft) {
4407 		intr |= INTR_TYPE_SOFT_INTR;
4408 		vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
4409 			     vmx->vcpu.arch.event_exit_inst_len);
4410 	} else
4411 		intr |= INTR_TYPE_EXT_INTR;
4412 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
4413 
4414 	vmx_clear_hlt(vcpu);
4415 }
4416 
4417 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
4418 {
4419 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4420 
4421 	if (!enable_vnmi) {
4422 		/*
4423 		 * Tracking the NMI-blocked state in software is built upon
4424 		 * finding the next open IRQ window. This, in turn, depends on
4425 		 * well-behaving guests: They have to keep IRQs disabled at
4426 		 * least as long as the NMI handler runs. Otherwise we may
4427 		 * cause NMI nesting, maybe breaking the guest. But as this is
4428 		 * highly unlikely, we can live with the residual risk.
4429 		 */
4430 		vmx->loaded_vmcs->soft_vnmi_blocked = 1;
4431 		vmx->loaded_vmcs->vnmi_blocked_time = 0;
4432 	}
4433 
4434 	++vcpu->stat.nmi_injections;
4435 	vmx->loaded_vmcs->nmi_known_unmasked = false;
4436 
4437 	if (vmx->rmode.vm86_active) {
4438 		kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR, 0);
4439 		return;
4440 	}
4441 
4442 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
4443 			INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
4444 
4445 	vmx_clear_hlt(vcpu);
4446 }
4447 
4448 bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
4449 {
4450 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4451 	bool masked;
4452 
4453 	if (!enable_vnmi)
4454 		return vmx->loaded_vmcs->soft_vnmi_blocked;
4455 	if (vmx->loaded_vmcs->nmi_known_unmasked)
4456 		return false;
4457 	masked = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI;
4458 	vmx->loaded_vmcs->nmi_known_unmasked = !masked;
4459 	return masked;
4460 }
4461 
4462 void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
4463 {
4464 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4465 
4466 	if (!enable_vnmi) {
4467 		if (vmx->loaded_vmcs->soft_vnmi_blocked != masked) {
4468 			vmx->loaded_vmcs->soft_vnmi_blocked = masked;
4469 			vmx->loaded_vmcs->vnmi_blocked_time = 0;
4470 		}
4471 	} else {
4472 		vmx->loaded_vmcs->nmi_known_unmasked = !masked;
4473 		if (masked)
4474 			vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
4475 				      GUEST_INTR_STATE_NMI);
4476 		else
4477 			vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
4478 					GUEST_INTR_STATE_NMI);
4479 	}
4480 }
4481 
4482 static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
4483 {
4484 	if (to_vmx(vcpu)->nested.nested_run_pending)
4485 		return 0;
4486 
4487 	if (!enable_vnmi &&
4488 	    to_vmx(vcpu)->loaded_vmcs->soft_vnmi_blocked)
4489 		return 0;
4490 
4491 	return	!(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4492 		  (GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_STI
4493 		   | GUEST_INTR_STATE_NMI));
4494 }
4495 
4496 static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
4497 {
4498 	return (!to_vmx(vcpu)->nested.nested_run_pending &&
4499 		vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
4500 		!(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4501 			(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
4502 }
4503 
4504 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
4505 {
4506 	int ret;
4507 
4508 	if (enable_unrestricted_guest)
4509 		return 0;
4510 
4511 	mutex_lock(&kvm->slots_lock);
4512 	ret = __x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, addr,
4513 				      PAGE_SIZE * 3);
4514 	mutex_unlock(&kvm->slots_lock);
4515 
4516 	if (ret)
4517 		return ret;
4518 	to_kvm_vmx(kvm)->tss_addr = addr;
4519 	return init_rmode_tss(kvm);
4520 }
4521 
4522 static int vmx_set_identity_map_addr(struct kvm *kvm, u64 ident_addr)
4523 {
4524 	to_kvm_vmx(kvm)->ept_identity_map_addr = ident_addr;
4525 	return 0;
4526 }
4527 
4528 static bool rmode_exception(struct kvm_vcpu *vcpu, int vec)
4529 {
4530 	switch (vec) {
4531 	case BP_VECTOR:
4532 		/*
4533 		 * Update instruction length as we may reinject the exception
4534 		 * from user space while in guest debugging mode.
4535 		 */
4536 		to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
4537 			vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4538 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
4539 			return false;
4540 		/* fall through */
4541 	case DB_VECTOR:
4542 		if (vcpu->guest_debug &
4543 			(KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
4544 			return false;
4545 		/* fall through */
4546 	case DE_VECTOR:
4547 	case OF_VECTOR:
4548 	case BR_VECTOR:
4549 	case UD_VECTOR:
4550 	case DF_VECTOR:
4551 	case SS_VECTOR:
4552 	case GP_VECTOR:
4553 	case MF_VECTOR:
4554 		return true;
4555 	break;
4556 	}
4557 	return false;
4558 }
4559 
4560 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
4561 				  int vec, u32 err_code)
4562 {
4563 	/*
4564 	 * Instruction with address size override prefix opcode 0x67
4565 	 * Cause the #SS fault with 0 error code in VM86 mode.
4566 	 */
4567 	if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0) {
4568 		if (kvm_emulate_instruction(vcpu, 0)) {
4569 			if (vcpu->arch.halt_request) {
4570 				vcpu->arch.halt_request = 0;
4571 				return kvm_vcpu_halt(vcpu);
4572 			}
4573 			return 1;
4574 		}
4575 		return 0;
4576 	}
4577 
4578 	/*
4579 	 * Forward all other exceptions that are valid in real mode.
4580 	 * FIXME: Breaks guest debugging in real mode, needs to be fixed with
4581 	 *        the required debugging infrastructure rework.
4582 	 */
4583 	kvm_queue_exception(vcpu, vec);
4584 	return 1;
4585 }
4586 
4587 /*
4588  * Trigger machine check on the host. We assume all the MSRs are already set up
4589  * by the CPU and that we still run on the same CPU as the MCE occurred on.
4590  * We pass a fake environment to the machine check handler because we want
4591  * the guest to be always treated like user space, no matter what context
4592  * it used internally.
4593  */
4594 static void kvm_machine_check(void)
4595 {
4596 #if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
4597 	struct pt_regs regs = {
4598 		.cs = 3, /* Fake ring 3 no matter what the guest ran on */
4599 		.flags = X86_EFLAGS_IF,
4600 	};
4601 
4602 	do_machine_check(&regs, 0);
4603 #endif
4604 }
4605 
4606 static int handle_machine_check(struct kvm_vcpu *vcpu)
4607 {
4608 	/* handled by vmx_vcpu_run() */
4609 	return 1;
4610 }
4611 
4612 static int handle_exception_nmi(struct kvm_vcpu *vcpu)
4613 {
4614 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4615 	struct kvm_run *kvm_run = vcpu->run;
4616 	u32 intr_info, ex_no, error_code;
4617 	unsigned long cr2, rip, dr6;
4618 	u32 vect_info;
4619 
4620 	vect_info = vmx->idt_vectoring_info;
4621 	intr_info = vmx->exit_intr_info;
4622 
4623 	if (is_machine_check(intr_info) || is_nmi(intr_info))
4624 		return 1; /* handled by handle_exception_nmi_irqoff() */
4625 
4626 	if (is_invalid_opcode(intr_info))
4627 		return handle_ud(vcpu);
4628 
4629 	error_code = 0;
4630 	if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
4631 		error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
4632 
4633 	if (!vmx->rmode.vm86_active && is_gp_fault(intr_info)) {
4634 		WARN_ON_ONCE(!enable_vmware_backdoor);
4635 
4636 		/*
4637 		 * VMware backdoor emulation on #GP interception only handles
4638 		 * IN{S}, OUT{S}, and RDPMC, none of which generate a non-zero
4639 		 * error code on #GP.
4640 		 */
4641 		if (error_code) {
4642 			kvm_queue_exception_e(vcpu, GP_VECTOR, error_code);
4643 			return 1;
4644 		}
4645 		return kvm_emulate_instruction(vcpu, EMULTYPE_VMWARE_GP);
4646 	}
4647 
4648 	/*
4649 	 * The #PF with PFEC.RSVD = 1 indicates the guest is accessing
4650 	 * MMIO, it is better to report an internal error.
4651 	 * See the comments in vmx_handle_exit.
4652 	 */
4653 	if ((vect_info & VECTORING_INFO_VALID_MASK) &&
4654 	    !(is_page_fault(intr_info) && !(error_code & PFERR_RSVD_MASK))) {
4655 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4656 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
4657 		vcpu->run->internal.ndata = 3;
4658 		vcpu->run->internal.data[0] = vect_info;
4659 		vcpu->run->internal.data[1] = intr_info;
4660 		vcpu->run->internal.data[2] = error_code;
4661 		return 0;
4662 	}
4663 
4664 	if (is_page_fault(intr_info)) {
4665 		cr2 = vmcs_readl(EXIT_QUALIFICATION);
4666 		/* EPT won't cause page fault directly */
4667 		WARN_ON_ONCE(!vcpu->arch.apf.host_apf_reason && enable_ept);
4668 		return kvm_handle_page_fault(vcpu, error_code, cr2, NULL, 0);
4669 	}
4670 
4671 	ex_no = intr_info & INTR_INFO_VECTOR_MASK;
4672 
4673 	if (vmx->rmode.vm86_active && rmode_exception(vcpu, ex_no))
4674 		return handle_rmode_exception(vcpu, ex_no, error_code);
4675 
4676 	switch (ex_no) {
4677 	case AC_VECTOR:
4678 		kvm_queue_exception_e(vcpu, AC_VECTOR, error_code);
4679 		return 1;
4680 	case DB_VECTOR:
4681 		dr6 = vmcs_readl(EXIT_QUALIFICATION);
4682 		if (!(vcpu->guest_debug &
4683 		      (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
4684 			vcpu->arch.dr6 &= ~DR_TRAP_BITS;
4685 			vcpu->arch.dr6 |= dr6 | DR6_RTM;
4686 			if (is_icebp(intr_info))
4687 				WARN_ON(!skip_emulated_instruction(vcpu));
4688 
4689 			kvm_queue_exception(vcpu, DB_VECTOR);
4690 			return 1;
4691 		}
4692 		kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
4693 		kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
4694 		/* fall through */
4695 	case BP_VECTOR:
4696 		/*
4697 		 * Update instruction length as we may reinject #BP from
4698 		 * user space while in guest debugging mode. Reading it for
4699 		 * #DB as well causes no harm, it is not used in that case.
4700 		 */
4701 		vmx->vcpu.arch.event_exit_inst_len =
4702 			vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4703 		kvm_run->exit_reason = KVM_EXIT_DEBUG;
4704 		rip = kvm_rip_read(vcpu);
4705 		kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
4706 		kvm_run->debug.arch.exception = ex_no;
4707 		break;
4708 	default:
4709 		kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
4710 		kvm_run->ex.exception = ex_no;
4711 		kvm_run->ex.error_code = error_code;
4712 		break;
4713 	}
4714 	return 0;
4715 }
4716 
4717 static __always_inline int handle_external_interrupt(struct kvm_vcpu *vcpu)
4718 {
4719 	++vcpu->stat.irq_exits;
4720 	return 1;
4721 }
4722 
4723 static int handle_triple_fault(struct kvm_vcpu *vcpu)
4724 {
4725 	vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
4726 	vcpu->mmio_needed = 0;
4727 	return 0;
4728 }
4729 
4730 static int handle_io(struct kvm_vcpu *vcpu)
4731 {
4732 	unsigned long exit_qualification;
4733 	int size, in, string;
4734 	unsigned port;
4735 
4736 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4737 	string = (exit_qualification & 16) != 0;
4738 
4739 	++vcpu->stat.io_exits;
4740 
4741 	if (string)
4742 		return kvm_emulate_instruction(vcpu, 0);
4743 
4744 	port = exit_qualification >> 16;
4745 	size = (exit_qualification & 7) + 1;
4746 	in = (exit_qualification & 8) != 0;
4747 
4748 	return kvm_fast_pio(vcpu, size, port, in);
4749 }
4750 
4751 static void
4752 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
4753 {
4754 	/*
4755 	 * Patch in the VMCALL instruction:
4756 	 */
4757 	hypercall[0] = 0x0f;
4758 	hypercall[1] = 0x01;
4759 	hypercall[2] = 0xc1;
4760 }
4761 
4762 /* called to set cr0 as appropriate for a mov-to-cr0 exit. */
4763 static int handle_set_cr0(struct kvm_vcpu *vcpu, unsigned long val)
4764 {
4765 	if (is_guest_mode(vcpu)) {
4766 		struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4767 		unsigned long orig_val = val;
4768 
4769 		/*
4770 		 * We get here when L2 changed cr0 in a way that did not change
4771 		 * any of L1's shadowed bits (see nested_vmx_exit_handled_cr),
4772 		 * but did change L0 shadowed bits. So we first calculate the
4773 		 * effective cr0 value that L1 would like to write into the
4774 		 * hardware. It consists of the L2-owned bits from the new
4775 		 * value combined with the L1-owned bits from L1's guest_cr0.
4776 		 */
4777 		val = (val & ~vmcs12->cr0_guest_host_mask) |
4778 			(vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask);
4779 
4780 		if (!nested_guest_cr0_valid(vcpu, val))
4781 			return 1;
4782 
4783 		if (kvm_set_cr0(vcpu, val))
4784 			return 1;
4785 		vmcs_writel(CR0_READ_SHADOW, orig_val);
4786 		return 0;
4787 	} else {
4788 		if (to_vmx(vcpu)->nested.vmxon &&
4789 		    !nested_host_cr0_valid(vcpu, val))
4790 			return 1;
4791 
4792 		return kvm_set_cr0(vcpu, val);
4793 	}
4794 }
4795 
4796 static int handle_set_cr4(struct kvm_vcpu *vcpu, unsigned long val)
4797 {
4798 	if (is_guest_mode(vcpu)) {
4799 		struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4800 		unsigned long orig_val = val;
4801 
4802 		/* analogously to handle_set_cr0 */
4803 		val = (val & ~vmcs12->cr4_guest_host_mask) |
4804 			(vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask);
4805 		if (kvm_set_cr4(vcpu, val))
4806 			return 1;
4807 		vmcs_writel(CR4_READ_SHADOW, orig_val);
4808 		return 0;
4809 	} else
4810 		return kvm_set_cr4(vcpu, val);
4811 }
4812 
4813 static int handle_desc(struct kvm_vcpu *vcpu)
4814 {
4815 	WARN_ON(!(vcpu->arch.cr4 & X86_CR4_UMIP));
4816 	return kvm_emulate_instruction(vcpu, 0);
4817 }
4818 
4819 static int handle_cr(struct kvm_vcpu *vcpu)
4820 {
4821 	unsigned long exit_qualification, val;
4822 	int cr;
4823 	int reg;
4824 	int err;
4825 	int ret;
4826 
4827 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4828 	cr = exit_qualification & 15;
4829 	reg = (exit_qualification >> 8) & 15;
4830 	switch ((exit_qualification >> 4) & 3) {
4831 	case 0: /* mov to cr */
4832 		val = kvm_register_readl(vcpu, reg);
4833 		trace_kvm_cr_write(cr, val);
4834 		switch (cr) {
4835 		case 0:
4836 			err = handle_set_cr0(vcpu, val);
4837 			return kvm_complete_insn_gp(vcpu, err);
4838 		case 3:
4839 			WARN_ON_ONCE(enable_unrestricted_guest);
4840 			err = kvm_set_cr3(vcpu, val);
4841 			return kvm_complete_insn_gp(vcpu, err);
4842 		case 4:
4843 			err = handle_set_cr4(vcpu, val);
4844 			return kvm_complete_insn_gp(vcpu, err);
4845 		case 8: {
4846 				u8 cr8_prev = kvm_get_cr8(vcpu);
4847 				u8 cr8 = (u8)val;
4848 				err = kvm_set_cr8(vcpu, cr8);
4849 				ret = kvm_complete_insn_gp(vcpu, err);
4850 				if (lapic_in_kernel(vcpu))
4851 					return ret;
4852 				if (cr8_prev <= cr8)
4853 					return ret;
4854 				/*
4855 				 * TODO: we might be squashing a
4856 				 * KVM_GUESTDBG_SINGLESTEP-triggered
4857 				 * KVM_EXIT_DEBUG here.
4858 				 */
4859 				vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
4860 				return 0;
4861 			}
4862 		}
4863 		break;
4864 	case 2: /* clts */
4865 		WARN_ONCE(1, "Guest should always own CR0.TS");
4866 		vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
4867 		trace_kvm_cr_write(0, kvm_read_cr0(vcpu));
4868 		return kvm_skip_emulated_instruction(vcpu);
4869 	case 1: /*mov from cr*/
4870 		switch (cr) {
4871 		case 3:
4872 			WARN_ON_ONCE(enable_unrestricted_guest);
4873 			val = kvm_read_cr3(vcpu);
4874 			kvm_register_write(vcpu, reg, val);
4875 			trace_kvm_cr_read(cr, val);
4876 			return kvm_skip_emulated_instruction(vcpu);
4877 		case 8:
4878 			val = kvm_get_cr8(vcpu);
4879 			kvm_register_write(vcpu, reg, val);
4880 			trace_kvm_cr_read(cr, val);
4881 			return kvm_skip_emulated_instruction(vcpu);
4882 		}
4883 		break;
4884 	case 3: /* lmsw */
4885 		val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
4886 		trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val);
4887 		kvm_lmsw(vcpu, val);
4888 
4889 		return kvm_skip_emulated_instruction(vcpu);
4890 	default:
4891 		break;
4892 	}
4893 	vcpu->run->exit_reason = 0;
4894 	vcpu_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
4895 	       (int)(exit_qualification >> 4) & 3, cr);
4896 	return 0;
4897 }
4898 
4899 static int handle_dr(struct kvm_vcpu *vcpu)
4900 {
4901 	unsigned long exit_qualification;
4902 	int dr, dr7, reg;
4903 
4904 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4905 	dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
4906 
4907 	/* First, if DR does not exist, trigger UD */
4908 	if (!kvm_require_dr(vcpu, dr))
4909 		return 1;
4910 
4911 	/* Do not handle if the CPL > 0, will trigger GP on re-entry */
4912 	if (!kvm_require_cpl(vcpu, 0))
4913 		return 1;
4914 	dr7 = vmcs_readl(GUEST_DR7);
4915 	if (dr7 & DR7_GD) {
4916 		/*
4917 		 * As the vm-exit takes precedence over the debug trap, we
4918 		 * need to emulate the latter, either for the host or the
4919 		 * guest debugging itself.
4920 		 */
4921 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
4922 			vcpu->run->debug.arch.dr6 = vcpu->arch.dr6;
4923 			vcpu->run->debug.arch.dr7 = dr7;
4924 			vcpu->run->debug.arch.pc = kvm_get_linear_rip(vcpu);
4925 			vcpu->run->debug.arch.exception = DB_VECTOR;
4926 			vcpu->run->exit_reason = KVM_EXIT_DEBUG;
4927 			return 0;
4928 		} else {
4929 			vcpu->arch.dr6 &= ~DR_TRAP_BITS;
4930 			vcpu->arch.dr6 |= DR6_BD | DR6_RTM;
4931 			kvm_queue_exception(vcpu, DB_VECTOR);
4932 			return 1;
4933 		}
4934 	}
4935 
4936 	if (vcpu->guest_debug == 0) {
4937 		exec_controls_clearbit(to_vmx(vcpu), CPU_BASED_MOV_DR_EXITING);
4938 
4939 		/*
4940 		 * No more DR vmexits; force a reload of the debug registers
4941 		 * and reenter on this instruction.  The next vmexit will
4942 		 * retrieve the full state of the debug registers.
4943 		 */
4944 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
4945 		return 1;
4946 	}
4947 
4948 	reg = DEBUG_REG_ACCESS_REG(exit_qualification);
4949 	if (exit_qualification & TYPE_MOV_FROM_DR) {
4950 		unsigned long val;
4951 
4952 		if (kvm_get_dr(vcpu, dr, &val))
4953 			return 1;
4954 		kvm_register_write(vcpu, reg, val);
4955 	} else
4956 		if (kvm_set_dr(vcpu, dr, kvm_register_readl(vcpu, reg)))
4957 			return 1;
4958 
4959 	return kvm_skip_emulated_instruction(vcpu);
4960 }
4961 
4962 static u64 vmx_get_dr6(struct kvm_vcpu *vcpu)
4963 {
4964 	return vcpu->arch.dr6;
4965 }
4966 
4967 static void vmx_set_dr6(struct kvm_vcpu *vcpu, unsigned long val)
4968 {
4969 }
4970 
4971 static void vmx_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
4972 {
4973 	get_debugreg(vcpu->arch.db[0], 0);
4974 	get_debugreg(vcpu->arch.db[1], 1);
4975 	get_debugreg(vcpu->arch.db[2], 2);
4976 	get_debugreg(vcpu->arch.db[3], 3);
4977 	get_debugreg(vcpu->arch.dr6, 6);
4978 	vcpu->arch.dr7 = vmcs_readl(GUEST_DR7);
4979 
4980 	vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
4981 	exec_controls_setbit(to_vmx(vcpu), CPU_BASED_MOV_DR_EXITING);
4982 }
4983 
4984 static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
4985 {
4986 	vmcs_writel(GUEST_DR7, val);
4987 }
4988 
4989 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
4990 {
4991 	kvm_apic_update_ppr(vcpu);
4992 	return 1;
4993 }
4994 
4995 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
4996 {
4997 	exec_controls_clearbit(to_vmx(vcpu), CPU_BASED_INTR_WINDOW_EXITING);
4998 
4999 	kvm_make_request(KVM_REQ_EVENT, vcpu);
5000 
5001 	++vcpu->stat.irq_window_exits;
5002 	return 1;
5003 }
5004 
5005 static int handle_vmcall(struct kvm_vcpu *vcpu)
5006 {
5007 	return kvm_emulate_hypercall(vcpu);
5008 }
5009 
5010 static int handle_invd(struct kvm_vcpu *vcpu)
5011 {
5012 	return kvm_emulate_instruction(vcpu, 0);
5013 }
5014 
5015 static int handle_invlpg(struct kvm_vcpu *vcpu)
5016 {
5017 	unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5018 
5019 	kvm_mmu_invlpg(vcpu, exit_qualification);
5020 	return kvm_skip_emulated_instruction(vcpu);
5021 }
5022 
5023 static int handle_rdpmc(struct kvm_vcpu *vcpu)
5024 {
5025 	int err;
5026 
5027 	err = kvm_rdpmc(vcpu);
5028 	return kvm_complete_insn_gp(vcpu, err);
5029 }
5030 
5031 static int handle_wbinvd(struct kvm_vcpu *vcpu)
5032 {
5033 	return kvm_emulate_wbinvd(vcpu);
5034 }
5035 
5036 static int handle_xsetbv(struct kvm_vcpu *vcpu)
5037 {
5038 	u64 new_bv = kvm_read_edx_eax(vcpu);
5039 	u32 index = kvm_rcx_read(vcpu);
5040 
5041 	if (kvm_set_xcr(vcpu, index, new_bv) == 0)
5042 		return kvm_skip_emulated_instruction(vcpu);
5043 	return 1;
5044 }
5045 
5046 static int handle_apic_access(struct kvm_vcpu *vcpu)
5047 {
5048 	if (likely(fasteoi)) {
5049 		unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5050 		int access_type, offset;
5051 
5052 		access_type = exit_qualification & APIC_ACCESS_TYPE;
5053 		offset = exit_qualification & APIC_ACCESS_OFFSET;
5054 		/*
5055 		 * Sane guest uses MOV to write EOI, with written value
5056 		 * not cared. So make a short-circuit here by avoiding
5057 		 * heavy instruction emulation.
5058 		 */
5059 		if ((access_type == TYPE_LINEAR_APIC_INST_WRITE) &&
5060 		    (offset == APIC_EOI)) {
5061 			kvm_lapic_set_eoi(vcpu);
5062 			return kvm_skip_emulated_instruction(vcpu);
5063 		}
5064 	}
5065 	return kvm_emulate_instruction(vcpu, 0);
5066 }
5067 
5068 static int handle_apic_eoi_induced(struct kvm_vcpu *vcpu)
5069 {
5070 	unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5071 	int vector = exit_qualification & 0xff;
5072 
5073 	/* EOI-induced VM exit is trap-like and thus no need to adjust IP */
5074 	kvm_apic_set_eoi_accelerated(vcpu, vector);
5075 	return 1;
5076 }
5077 
5078 static int handle_apic_write(struct kvm_vcpu *vcpu)
5079 {
5080 	unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5081 	u32 offset = exit_qualification & 0xfff;
5082 
5083 	/* APIC-write VM exit is trap-like and thus no need to adjust IP */
5084 	kvm_apic_write_nodecode(vcpu, offset);
5085 	return 1;
5086 }
5087 
5088 static int handle_task_switch(struct kvm_vcpu *vcpu)
5089 {
5090 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5091 	unsigned long exit_qualification;
5092 	bool has_error_code = false;
5093 	u32 error_code = 0;
5094 	u16 tss_selector;
5095 	int reason, type, idt_v, idt_index;
5096 
5097 	idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
5098 	idt_index = (vmx->idt_vectoring_info & VECTORING_INFO_VECTOR_MASK);
5099 	type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
5100 
5101 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5102 
5103 	reason = (u32)exit_qualification >> 30;
5104 	if (reason == TASK_SWITCH_GATE && idt_v) {
5105 		switch (type) {
5106 		case INTR_TYPE_NMI_INTR:
5107 			vcpu->arch.nmi_injected = false;
5108 			vmx_set_nmi_mask(vcpu, true);
5109 			break;
5110 		case INTR_TYPE_EXT_INTR:
5111 		case INTR_TYPE_SOFT_INTR:
5112 			kvm_clear_interrupt_queue(vcpu);
5113 			break;
5114 		case INTR_TYPE_HARD_EXCEPTION:
5115 			if (vmx->idt_vectoring_info &
5116 			    VECTORING_INFO_DELIVER_CODE_MASK) {
5117 				has_error_code = true;
5118 				error_code =
5119 					vmcs_read32(IDT_VECTORING_ERROR_CODE);
5120 			}
5121 			/* fall through */
5122 		case INTR_TYPE_SOFT_EXCEPTION:
5123 			kvm_clear_exception_queue(vcpu);
5124 			break;
5125 		default:
5126 			break;
5127 		}
5128 	}
5129 	tss_selector = exit_qualification;
5130 
5131 	if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
5132 		       type != INTR_TYPE_EXT_INTR &&
5133 		       type != INTR_TYPE_NMI_INTR))
5134 		WARN_ON(!skip_emulated_instruction(vcpu));
5135 
5136 	/*
5137 	 * TODO: What about debug traps on tss switch?
5138 	 *       Are we supposed to inject them and update dr6?
5139 	 */
5140 	return kvm_task_switch(vcpu, tss_selector,
5141 			       type == INTR_TYPE_SOFT_INTR ? idt_index : -1,
5142 			       reason, has_error_code, error_code);
5143 }
5144 
5145 static int handle_ept_violation(struct kvm_vcpu *vcpu)
5146 {
5147 	unsigned long exit_qualification;
5148 	gpa_t gpa;
5149 	u64 error_code;
5150 
5151 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5152 
5153 	/*
5154 	 * EPT violation happened while executing iret from NMI,
5155 	 * "blocked by NMI" bit has to be set before next VM entry.
5156 	 * There are errata that may cause this bit to not be set:
5157 	 * AAK134, BY25.
5158 	 */
5159 	if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
5160 			enable_vnmi &&
5161 			(exit_qualification & INTR_INFO_UNBLOCK_NMI))
5162 		vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO, GUEST_INTR_STATE_NMI);
5163 
5164 	gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5165 	trace_kvm_page_fault(gpa, exit_qualification);
5166 
5167 	/* Is it a read fault? */
5168 	error_code = (exit_qualification & EPT_VIOLATION_ACC_READ)
5169 		     ? PFERR_USER_MASK : 0;
5170 	/* Is it a write fault? */
5171 	error_code |= (exit_qualification & EPT_VIOLATION_ACC_WRITE)
5172 		      ? PFERR_WRITE_MASK : 0;
5173 	/* Is it a fetch fault? */
5174 	error_code |= (exit_qualification & EPT_VIOLATION_ACC_INSTR)
5175 		      ? PFERR_FETCH_MASK : 0;
5176 	/* ept page table entry is present? */
5177 	error_code |= (exit_qualification &
5178 		       (EPT_VIOLATION_READABLE | EPT_VIOLATION_WRITABLE |
5179 			EPT_VIOLATION_EXECUTABLE))
5180 		      ? PFERR_PRESENT_MASK : 0;
5181 
5182 	error_code |= (exit_qualification & 0x100) != 0 ?
5183 	       PFERR_GUEST_FINAL_MASK : PFERR_GUEST_PAGE_MASK;
5184 
5185 	vcpu->arch.exit_qualification = exit_qualification;
5186 	return kvm_mmu_page_fault(vcpu, gpa, error_code, NULL, 0);
5187 }
5188 
5189 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
5190 {
5191 	gpa_t gpa;
5192 
5193 	/*
5194 	 * A nested guest cannot optimize MMIO vmexits, because we have an
5195 	 * nGPA here instead of the required GPA.
5196 	 */
5197 	gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5198 	if (!is_guest_mode(vcpu) &&
5199 	    !kvm_io_bus_write(vcpu, KVM_FAST_MMIO_BUS, gpa, 0, NULL)) {
5200 		trace_kvm_fast_mmio(gpa);
5201 		return kvm_skip_emulated_instruction(vcpu);
5202 	}
5203 
5204 	return kvm_mmu_page_fault(vcpu, gpa, PFERR_RSVD_MASK, NULL, 0);
5205 }
5206 
5207 static int handle_nmi_window(struct kvm_vcpu *vcpu)
5208 {
5209 	WARN_ON_ONCE(!enable_vnmi);
5210 	exec_controls_clearbit(to_vmx(vcpu), CPU_BASED_NMI_WINDOW_EXITING);
5211 	++vcpu->stat.nmi_window_exits;
5212 	kvm_make_request(KVM_REQ_EVENT, vcpu);
5213 
5214 	return 1;
5215 }
5216 
5217 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
5218 {
5219 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5220 	bool intr_window_requested;
5221 	unsigned count = 130;
5222 
5223 	/*
5224 	 * We should never reach the point where we are emulating L2
5225 	 * due to invalid guest state as that means we incorrectly
5226 	 * allowed a nested VMEntry with an invalid vmcs12.
5227 	 */
5228 	WARN_ON_ONCE(vmx->emulation_required && vmx->nested.nested_run_pending);
5229 
5230 	intr_window_requested = exec_controls_get(vmx) &
5231 				CPU_BASED_INTR_WINDOW_EXITING;
5232 
5233 	while (vmx->emulation_required && count-- != 0) {
5234 		if (intr_window_requested && vmx_interrupt_allowed(vcpu))
5235 			return handle_interrupt_window(&vmx->vcpu);
5236 
5237 		if (kvm_test_request(KVM_REQ_EVENT, vcpu))
5238 			return 1;
5239 
5240 		if (!kvm_emulate_instruction(vcpu, 0))
5241 			return 0;
5242 
5243 		if (vmx->emulation_required && !vmx->rmode.vm86_active &&
5244 		    vcpu->arch.exception.pending) {
5245 			vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5246 			vcpu->run->internal.suberror =
5247 						KVM_INTERNAL_ERROR_EMULATION;
5248 			vcpu->run->internal.ndata = 0;
5249 			return 0;
5250 		}
5251 
5252 		if (vcpu->arch.halt_request) {
5253 			vcpu->arch.halt_request = 0;
5254 			return kvm_vcpu_halt(vcpu);
5255 		}
5256 
5257 		/*
5258 		 * Note, return 1 and not 0, vcpu_run() is responsible for
5259 		 * morphing the pending signal into the proper return code.
5260 		 */
5261 		if (signal_pending(current))
5262 			return 1;
5263 
5264 		if (need_resched())
5265 			schedule();
5266 	}
5267 
5268 	return 1;
5269 }
5270 
5271 static void grow_ple_window(struct kvm_vcpu *vcpu)
5272 {
5273 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5274 	unsigned int old = vmx->ple_window;
5275 
5276 	vmx->ple_window = __grow_ple_window(old, ple_window,
5277 					    ple_window_grow,
5278 					    ple_window_max);
5279 
5280 	if (vmx->ple_window != old) {
5281 		vmx->ple_window_dirty = true;
5282 		trace_kvm_ple_window_update(vcpu->vcpu_id,
5283 					    vmx->ple_window, old);
5284 	}
5285 }
5286 
5287 static void shrink_ple_window(struct kvm_vcpu *vcpu)
5288 {
5289 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5290 	unsigned int old = vmx->ple_window;
5291 
5292 	vmx->ple_window = __shrink_ple_window(old, ple_window,
5293 					      ple_window_shrink,
5294 					      ple_window);
5295 
5296 	if (vmx->ple_window != old) {
5297 		vmx->ple_window_dirty = true;
5298 		trace_kvm_ple_window_update(vcpu->vcpu_id,
5299 					    vmx->ple_window, old);
5300 	}
5301 }
5302 
5303 /*
5304  * Handler for POSTED_INTERRUPT_WAKEUP_VECTOR.
5305  */
5306 static void wakeup_handler(void)
5307 {
5308 	struct kvm_vcpu *vcpu;
5309 	int cpu = smp_processor_id();
5310 
5311 	spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
5312 	list_for_each_entry(vcpu, &per_cpu(blocked_vcpu_on_cpu, cpu),
5313 			blocked_vcpu_list) {
5314 		struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
5315 
5316 		if (pi_test_on(pi_desc) == 1)
5317 			kvm_vcpu_kick(vcpu);
5318 	}
5319 	spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
5320 }
5321 
5322 static void vmx_enable_tdp(void)
5323 {
5324 	kvm_mmu_set_mask_ptes(VMX_EPT_READABLE_MASK,
5325 		enable_ept_ad_bits ? VMX_EPT_ACCESS_BIT : 0ull,
5326 		enable_ept_ad_bits ? VMX_EPT_DIRTY_BIT : 0ull,
5327 		0ull, VMX_EPT_EXECUTABLE_MASK,
5328 		cpu_has_vmx_ept_execute_only() ? 0ull : VMX_EPT_READABLE_MASK,
5329 		VMX_EPT_RWX_MASK, 0ull);
5330 
5331 	ept_set_mmio_spte_mask();
5332 	kvm_enable_tdp();
5333 }
5334 
5335 /*
5336  * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
5337  * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
5338  */
5339 static int handle_pause(struct kvm_vcpu *vcpu)
5340 {
5341 	if (!kvm_pause_in_guest(vcpu->kvm))
5342 		grow_ple_window(vcpu);
5343 
5344 	/*
5345 	 * Intel sdm vol3 ch-25.1.3 says: The "PAUSE-loop exiting"
5346 	 * VM-execution control is ignored if CPL > 0. OTOH, KVM
5347 	 * never set PAUSE_EXITING and just set PLE if supported,
5348 	 * so the vcpu must be CPL=0 if it gets a PAUSE exit.
5349 	 */
5350 	kvm_vcpu_on_spin(vcpu, true);
5351 	return kvm_skip_emulated_instruction(vcpu);
5352 }
5353 
5354 static int handle_nop(struct kvm_vcpu *vcpu)
5355 {
5356 	return kvm_skip_emulated_instruction(vcpu);
5357 }
5358 
5359 static int handle_mwait(struct kvm_vcpu *vcpu)
5360 {
5361 	printk_once(KERN_WARNING "kvm: MWAIT instruction emulated as NOP!\n");
5362 	return handle_nop(vcpu);
5363 }
5364 
5365 static int handle_invalid_op(struct kvm_vcpu *vcpu)
5366 {
5367 	kvm_queue_exception(vcpu, UD_VECTOR);
5368 	return 1;
5369 }
5370 
5371 static int handle_monitor_trap(struct kvm_vcpu *vcpu)
5372 {
5373 	return 1;
5374 }
5375 
5376 static int handle_monitor(struct kvm_vcpu *vcpu)
5377 {
5378 	printk_once(KERN_WARNING "kvm: MONITOR instruction emulated as NOP!\n");
5379 	return handle_nop(vcpu);
5380 }
5381 
5382 static int handle_invpcid(struct kvm_vcpu *vcpu)
5383 {
5384 	u32 vmx_instruction_info;
5385 	unsigned long type;
5386 	bool pcid_enabled;
5387 	gva_t gva;
5388 	struct x86_exception e;
5389 	unsigned i;
5390 	unsigned long roots_to_free = 0;
5391 	struct {
5392 		u64 pcid;
5393 		u64 gla;
5394 	} operand;
5395 
5396 	if (!guest_cpuid_has(vcpu, X86_FEATURE_INVPCID)) {
5397 		kvm_queue_exception(vcpu, UD_VECTOR);
5398 		return 1;
5399 	}
5400 
5401 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5402 	type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
5403 
5404 	if (type > 3) {
5405 		kvm_inject_gp(vcpu, 0);
5406 		return 1;
5407 	}
5408 
5409 	/* According to the Intel instruction reference, the memory operand
5410 	 * is read even if it isn't needed (e.g., for type==all)
5411 	 */
5412 	if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
5413 				vmx_instruction_info, false,
5414 				sizeof(operand), &gva))
5415 		return 1;
5416 
5417 	if (kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e)) {
5418 		kvm_inject_page_fault(vcpu, &e);
5419 		return 1;
5420 	}
5421 
5422 	if (operand.pcid >> 12 != 0) {
5423 		kvm_inject_gp(vcpu, 0);
5424 		return 1;
5425 	}
5426 
5427 	pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
5428 
5429 	switch (type) {
5430 	case INVPCID_TYPE_INDIV_ADDR:
5431 		if ((!pcid_enabled && (operand.pcid != 0)) ||
5432 		    is_noncanonical_address(operand.gla, vcpu)) {
5433 			kvm_inject_gp(vcpu, 0);
5434 			return 1;
5435 		}
5436 		kvm_mmu_invpcid_gva(vcpu, operand.gla, operand.pcid);
5437 		return kvm_skip_emulated_instruction(vcpu);
5438 
5439 	case INVPCID_TYPE_SINGLE_CTXT:
5440 		if (!pcid_enabled && (operand.pcid != 0)) {
5441 			kvm_inject_gp(vcpu, 0);
5442 			return 1;
5443 		}
5444 
5445 		if (kvm_get_active_pcid(vcpu) == operand.pcid) {
5446 			kvm_mmu_sync_roots(vcpu);
5447 			kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
5448 		}
5449 
5450 		for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++)
5451 			if (kvm_get_pcid(vcpu, vcpu->arch.mmu->prev_roots[i].cr3)
5452 			    == operand.pcid)
5453 				roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i);
5454 
5455 		kvm_mmu_free_roots(vcpu, vcpu->arch.mmu, roots_to_free);
5456 		/*
5457 		 * If neither the current cr3 nor any of the prev_roots use the
5458 		 * given PCID, then nothing needs to be done here because a
5459 		 * resync will happen anyway before switching to any other CR3.
5460 		 */
5461 
5462 		return kvm_skip_emulated_instruction(vcpu);
5463 
5464 	case INVPCID_TYPE_ALL_NON_GLOBAL:
5465 		/*
5466 		 * Currently, KVM doesn't mark global entries in the shadow
5467 		 * page tables, so a non-global flush just degenerates to a
5468 		 * global flush. If needed, we could optimize this later by
5469 		 * keeping track of global entries in shadow page tables.
5470 		 */
5471 
5472 		/* fall-through */
5473 	case INVPCID_TYPE_ALL_INCL_GLOBAL:
5474 		kvm_mmu_unload(vcpu);
5475 		return kvm_skip_emulated_instruction(vcpu);
5476 
5477 	default:
5478 		BUG(); /* We have already checked above that type <= 3 */
5479 	}
5480 }
5481 
5482 static int handle_pml_full(struct kvm_vcpu *vcpu)
5483 {
5484 	unsigned long exit_qualification;
5485 
5486 	trace_kvm_pml_full(vcpu->vcpu_id);
5487 
5488 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5489 
5490 	/*
5491 	 * PML buffer FULL happened while executing iret from NMI,
5492 	 * "blocked by NMI" bit has to be set before next VM entry.
5493 	 */
5494 	if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
5495 			enable_vnmi &&
5496 			(exit_qualification & INTR_INFO_UNBLOCK_NMI))
5497 		vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
5498 				GUEST_INTR_STATE_NMI);
5499 
5500 	/*
5501 	 * PML buffer already flushed at beginning of VMEXIT. Nothing to do
5502 	 * here.., and there's no userspace involvement needed for PML.
5503 	 */
5504 	return 1;
5505 }
5506 
5507 static int handle_preemption_timer(struct kvm_vcpu *vcpu)
5508 {
5509 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5510 
5511 	if (!vmx->req_immediate_exit &&
5512 	    !unlikely(vmx->loaded_vmcs->hv_timer_soft_disabled))
5513 		kvm_lapic_expired_hv_timer(vcpu);
5514 
5515 	return 1;
5516 }
5517 
5518 /*
5519  * When nested=0, all VMX instruction VM Exits filter here.  The handlers
5520  * are overwritten by nested_vmx_setup() when nested=1.
5521  */
5522 static int handle_vmx_instruction(struct kvm_vcpu *vcpu)
5523 {
5524 	kvm_queue_exception(vcpu, UD_VECTOR);
5525 	return 1;
5526 }
5527 
5528 static int handle_encls(struct kvm_vcpu *vcpu)
5529 {
5530 	/*
5531 	 * SGX virtualization is not yet supported.  There is no software
5532 	 * enable bit for SGX, so we have to trap ENCLS and inject a #UD
5533 	 * to prevent the guest from executing ENCLS.
5534 	 */
5535 	kvm_queue_exception(vcpu, UD_VECTOR);
5536 	return 1;
5537 }
5538 
5539 /*
5540  * The exit handlers return 1 if the exit was handled fully and guest execution
5541  * may resume.  Otherwise they set the kvm_run parameter to indicate what needs
5542  * to be done to userspace and return 0.
5543  */
5544 static int (*kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
5545 	[EXIT_REASON_EXCEPTION_NMI]           = handle_exception_nmi,
5546 	[EXIT_REASON_EXTERNAL_INTERRUPT]      = handle_external_interrupt,
5547 	[EXIT_REASON_TRIPLE_FAULT]            = handle_triple_fault,
5548 	[EXIT_REASON_NMI_WINDOW]	      = handle_nmi_window,
5549 	[EXIT_REASON_IO_INSTRUCTION]          = handle_io,
5550 	[EXIT_REASON_CR_ACCESS]               = handle_cr,
5551 	[EXIT_REASON_DR_ACCESS]               = handle_dr,
5552 	[EXIT_REASON_CPUID]                   = kvm_emulate_cpuid,
5553 	[EXIT_REASON_MSR_READ]                = kvm_emulate_rdmsr,
5554 	[EXIT_REASON_MSR_WRITE]               = kvm_emulate_wrmsr,
5555 	[EXIT_REASON_INTERRUPT_WINDOW]        = handle_interrupt_window,
5556 	[EXIT_REASON_HLT]                     = kvm_emulate_halt,
5557 	[EXIT_REASON_INVD]		      = handle_invd,
5558 	[EXIT_REASON_INVLPG]		      = handle_invlpg,
5559 	[EXIT_REASON_RDPMC]                   = handle_rdpmc,
5560 	[EXIT_REASON_VMCALL]                  = handle_vmcall,
5561 	[EXIT_REASON_VMCLEAR]		      = handle_vmx_instruction,
5562 	[EXIT_REASON_VMLAUNCH]		      = handle_vmx_instruction,
5563 	[EXIT_REASON_VMPTRLD]		      = handle_vmx_instruction,
5564 	[EXIT_REASON_VMPTRST]		      = handle_vmx_instruction,
5565 	[EXIT_REASON_VMREAD]		      = handle_vmx_instruction,
5566 	[EXIT_REASON_VMRESUME]		      = handle_vmx_instruction,
5567 	[EXIT_REASON_VMWRITE]		      = handle_vmx_instruction,
5568 	[EXIT_REASON_VMOFF]		      = handle_vmx_instruction,
5569 	[EXIT_REASON_VMON]		      = handle_vmx_instruction,
5570 	[EXIT_REASON_TPR_BELOW_THRESHOLD]     = handle_tpr_below_threshold,
5571 	[EXIT_REASON_APIC_ACCESS]             = handle_apic_access,
5572 	[EXIT_REASON_APIC_WRITE]              = handle_apic_write,
5573 	[EXIT_REASON_EOI_INDUCED]             = handle_apic_eoi_induced,
5574 	[EXIT_REASON_WBINVD]                  = handle_wbinvd,
5575 	[EXIT_REASON_XSETBV]                  = handle_xsetbv,
5576 	[EXIT_REASON_TASK_SWITCH]             = handle_task_switch,
5577 	[EXIT_REASON_MCE_DURING_VMENTRY]      = handle_machine_check,
5578 	[EXIT_REASON_GDTR_IDTR]		      = handle_desc,
5579 	[EXIT_REASON_LDTR_TR]		      = handle_desc,
5580 	[EXIT_REASON_EPT_VIOLATION]	      = handle_ept_violation,
5581 	[EXIT_REASON_EPT_MISCONFIG]           = handle_ept_misconfig,
5582 	[EXIT_REASON_PAUSE_INSTRUCTION]       = handle_pause,
5583 	[EXIT_REASON_MWAIT_INSTRUCTION]	      = handle_mwait,
5584 	[EXIT_REASON_MONITOR_TRAP_FLAG]       = handle_monitor_trap,
5585 	[EXIT_REASON_MONITOR_INSTRUCTION]     = handle_monitor,
5586 	[EXIT_REASON_INVEPT]                  = handle_vmx_instruction,
5587 	[EXIT_REASON_INVVPID]                 = handle_vmx_instruction,
5588 	[EXIT_REASON_RDRAND]                  = handle_invalid_op,
5589 	[EXIT_REASON_RDSEED]                  = handle_invalid_op,
5590 	[EXIT_REASON_PML_FULL]		      = handle_pml_full,
5591 	[EXIT_REASON_INVPCID]                 = handle_invpcid,
5592 	[EXIT_REASON_VMFUNC]		      = handle_vmx_instruction,
5593 	[EXIT_REASON_PREEMPTION_TIMER]	      = handle_preemption_timer,
5594 	[EXIT_REASON_ENCLS]		      = handle_encls,
5595 };
5596 
5597 static const int kvm_vmx_max_exit_handlers =
5598 	ARRAY_SIZE(kvm_vmx_exit_handlers);
5599 
5600 static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
5601 {
5602 	*info1 = vmcs_readl(EXIT_QUALIFICATION);
5603 	*info2 = vmcs_read32(VM_EXIT_INTR_INFO);
5604 }
5605 
5606 static void vmx_destroy_pml_buffer(struct vcpu_vmx *vmx)
5607 {
5608 	if (vmx->pml_pg) {
5609 		__free_page(vmx->pml_pg);
5610 		vmx->pml_pg = NULL;
5611 	}
5612 }
5613 
5614 static void vmx_flush_pml_buffer(struct kvm_vcpu *vcpu)
5615 {
5616 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5617 	u64 *pml_buf;
5618 	u16 pml_idx;
5619 
5620 	pml_idx = vmcs_read16(GUEST_PML_INDEX);
5621 
5622 	/* Do nothing if PML buffer is empty */
5623 	if (pml_idx == (PML_ENTITY_NUM - 1))
5624 		return;
5625 
5626 	/* PML index always points to next available PML buffer entity */
5627 	if (pml_idx >= PML_ENTITY_NUM)
5628 		pml_idx = 0;
5629 	else
5630 		pml_idx++;
5631 
5632 	pml_buf = page_address(vmx->pml_pg);
5633 	for (; pml_idx < PML_ENTITY_NUM; pml_idx++) {
5634 		u64 gpa;
5635 
5636 		gpa = pml_buf[pml_idx];
5637 		WARN_ON(gpa & (PAGE_SIZE - 1));
5638 		kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
5639 	}
5640 
5641 	/* reset PML index */
5642 	vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
5643 }
5644 
5645 /*
5646  * Flush all vcpus' PML buffer and update logged GPAs to dirty_bitmap.
5647  * Called before reporting dirty_bitmap to userspace.
5648  */
5649 static void kvm_flush_pml_buffers(struct kvm *kvm)
5650 {
5651 	int i;
5652 	struct kvm_vcpu *vcpu;
5653 	/*
5654 	 * We only need to kick vcpu out of guest mode here, as PML buffer
5655 	 * is flushed at beginning of all VMEXITs, and it's obvious that only
5656 	 * vcpus running in guest are possible to have unflushed GPAs in PML
5657 	 * buffer.
5658 	 */
5659 	kvm_for_each_vcpu(i, vcpu, kvm)
5660 		kvm_vcpu_kick(vcpu);
5661 }
5662 
5663 static void vmx_dump_sel(char *name, uint32_t sel)
5664 {
5665 	pr_err("%s sel=0x%04x, attr=0x%05x, limit=0x%08x, base=0x%016lx\n",
5666 	       name, vmcs_read16(sel),
5667 	       vmcs_read32(sel + GUEST_ES_AR_BYTES - GUEST_ES_SELECTOR),
5668 	       vmcs_read32(sel + GUEST_ES_LIMIT - GUEST_ES_SELECTOR),
5669 	       vmcs_readl(sel + GUEST_ES_BASE - GUEST_ES_SELECTOR));
5670 }
5671 
5672 static void vmx_dump_dtsel(char *name, uint32_t limit)
5673 {
5674 	pr_err("%s                           limit=0x%08x, base=0x%016lx\n",
5675 	       name, vmcs_read32(limit),
5676 	       vmcs_readl(limit + GUEST_GDTR_BASE - GUEST_GDTR_LIMIT));
5677 }
5678 
5679 void dump_vmcs(void)
5680 {
5681 	u32 vmentry_ctl, vmexit_ctl;
5682 	u32 cpu_based_exec_ctrl, pin_based_exec_ctrl, secondary_exec_control;
5683 	unsigned long cr4;
5684 	u64 efer;
5685 	int i, n;
5686 
5687 	if (!dump_invalid_vmcs) {
5688 		pr_warn_ratelimited("set kvm_intel.dump_invalid_vmcs=1 to dump internal KVM state.\n");
5689 		return;
5690 	}
5691 
5692 	vmentry_ctl = vmcs_read32(VM_ENTRY_CONTROLS);
5693 	vmexit_ctl = vmcs_read32(VM_EXIT_CONTROLS);
5694 	cpu_based_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5695 	pin_based_exec_ctrl = vmcs_read32(PIN_BASED_VM_EXEC_CONTROL);
5696 	cr4 = vmcs_readl(GUEST_CR4);
5697 	efer = vmcs_read64(GUEST_IA32_EFER);
5698 	secondary_exec_control = 0;
5699 	if (cpu_has_secondary_exec_ctrls())
5700 		secondary_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
5701 
5702 	pr_err("*** Guest State ***\n");
5703 	pr_err("CR0: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
5704 	       vmcs_readl(GUEST_CR0), vmcs_readl(CR0_READ_SHADOW),
5705 	       vmcs_readl(CR0_GUEST_HOST_MASK));
5706 	pr_err("CR4: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
5707 	       cr4, vmcs_readl(CR4_READ_SHADOW), vmcs_readl(CR4_GUEST_HOST_MASK));
5708 	pr_err("CR3 = 0x%016lx\n", vmcs_readl(GUEST_CR3));
5709 	if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT) &&
5710 	    (cr4 & X86_CR4_PAE) && !(efer & EFER_LMA))
5711 	{
5712 		pr_err("PDPTR0 = 0x%016llx  PDPTR1 = 0x%016llx\n",
5713 		       vmcs_read64(GUEST_PDPTR0), vmcs_read64(GUEST_PDPTR1));
5714 		pr_err("PDPTR2 = 0x%016llx  PDPTR3 = 0x%016llx\n",
5715 		       vmcs_read64(GUEST_PDPTR2), vmcs_read64(GUEST_PDPTR3));
5716 	}
5717 	pr_err("RSP = 0x%016lx  RIP = 0x%016lx\n",
5718 	       vmcs_readl(GUEST_RSP), vmcs_readl(GUEST_RIP));
5719 	pr_err("RFLAGS=0x%08lx         DR7 = 0x%016lx\n",
5720 	       vmcs_readl(GUEST_RFLAGS), vmcs_readl(GUEST_DR7));
5721 	pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
5722 	       vmcs_readl(GUEST_SYSENTER_ESP),
5723 	       vmcs_read32(GUEST_SYSENTER_CS), vmcs_readl(GUEST_SYSENTER_EIP));
5724 	vmx_dump_sel("CS:  ", GUEST_CS_SELECTOR);
5725 	vmx_dump_sel("DS:  ", GUEST_DS_SELECTOR);
5726 	vmx_dump_sel("SS:  ", GUEST_SS_SELECTOR);
5727 	vmx_dump_sel("ES:  ", GUEST_ES_SELECTOR);
5728 	vmx_dump_sel("FS:  ", GUEST_FS_SELECTOR);
5729 	vmx_dump_sel("GS:  ", GUEST_GS_SELECTOR);
5730 	vmx_dump_dtsel("GDTR:", GUEST_GDTR_LIMIT);
5731 	vmx_dump_sel("LDTR:", GUEST_LDTR_SELECTOR);
5732 	vmx_dump_dtsel("IDTR:", GUEST_IDTR_LIMIT);
5733 	vmx_dump_sel("TR:  ", GUEST_TR_SELECTOR);
5734 	if ((vmexit_ctl & (VM_EXIT_SAVE_IA32_PAT | VM_EXIT_SAVE_IA32_EFER)) ||
5735 	    (vmentry_ctl & (VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_IA32_EFER)))
5736 		pr_err("EFER =     0x%016llx  PAT = 0x%016llx\n",
5737 		       efer, vmcs_read64(GUEST_IA32_PAT));
5738 	pr_err("DebugCtl = 0x%016llx  DebugExceptions = 0x%016lx\n",
5739 	       vmcs_read64(GUEST_IA32_DEBUGCTL),
5740 	       vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS));
5741 	if (cpu_has_load_perf_global_ctrl() &&
5742 	    vmentry_ctl & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
5743 		pr_err("PerfGlobCtl = 0x%016llx\n",
5744 		       vmcs_read64(GUEST_IA32_PERF_GLOBAL_CTRL));
5745 	if (vmentry_ctl & VM_ENTRY_LOAD_BNDCFGS)
5746 		pr_err("BndCfgS = 0x%016llx\n", vmcs_read64(GUEST_BNDCFGS));
5747 	pr_err("Interruptibility = %08x  ActivityState = %08x\n",
5748 	       vmcs_read32(GUEST_INTERRUPTIBILITY_INFO),
5749 	       vmcs_read32(GUEST_ACTIVITY_STATE));
5750 	if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)
5751 		pr_err("InterruptStatus = %04x\n",
5752 		       vmcs_read16(GUEST_INTR_STATUS));
5753 
5754 	pr_err("*** Host State ***\n");
5755 	pr_err("RIP = 0x%016lx  RSP = 0x%016lx\n",
5756 	       vmcs_readl(HOST_RIP), vmcs_readl(HOST_RSP));
5757 	pr_err("CS=%04x SS=%04x DS=%04x ES=%04x FS=%04x GS=%04x TR=%04x\n",
5758 	       vmcs_read16(HOST_CS_SELECTOR), vmcs_read16(HOST_SS_SELECTOR),
5759 	       vmcs_read16(HOST_DS_SELECTOR), vmcs_read16(HOST_ES_SELECTOR),
5760 	       vmcs_read16(HOST_FS_SELECTOR), vmcs_read16(HOST_GS_SELECTOR),
5761 	       vmcs_read16(HOST_TR_SELECTOR));
5762 	pr_err("FSBase=%016lx GSBase=%016lx TRBase=%016lx\n",
5763 	       vmcs_readl(HOST_FS_BASE), vmcs_readl(HOST_GS_BASE),
5764 	       vmcs_readl(HOST_TR_BASE));
5765 	pr_err("GDTBase=%016lx IDTBase=%016lx\n",
5766 	       vmcs_readl(HOST_GDTR_BASE), vmcs_readl(HOST_IDTR_BASE));
5767 	pr_err("CR0=%016lx CR3=%016lx CR4=%016lx\n",
5768 	       vmcs_readl(HOST_CR0), vmcs_readl(HOST_CR3),
5769 	       vmcs_readl(HOST_CR4));
5770 	pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
5771 	       vmcs_readl(HOST_IA32_SYSENTER_ESP),
5772 	       vmcs_read32(HOST_IA32_SYSENTER_CS),
5773 	       vmcs_readl(HOST_IA32_SYSENTER_EIP));
5774 	if (vmexit_ctl & (VM_EXIT_LOAD_IA32_PAT | VM_EXIT_LOAD_IA32_EFER))
5775 		pr_err("EFER = 0x%016llx  PAT = 0x%016llx\n",
5776 		       vmcs_read64(HOST_IA32_EFER),
5777 		       vmcs_read64(HOST_IA32_PAT));
5778 	if (cpu_has_load_perf_global_ctrl() &&
5779 	    vmexit_ctl & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
5780 		pr_err("PerfGlobCtl = 0x%016llx\n",
5781 		       vmcs_read64(HOST_IA32_PERF_GLOBAL_CTRL));
5782 
5783 	pr_err("*** Control State ***\n");
5784 	pr_err("PinBased=%08x CPUBased=%08x SecondaryExec=%08x\n",
5785 	       pin_based_exec_ctrl, cpu_based_exec_ctrl, secondary_exec_control);
5786 	pr_err("EntryControls=%08x ExitControls=%08x\n", vmentry_ctl, vmexit_ctl);
5787 	pr_err("ExceptionBitmap=%08x PFECmask=%08x PFECmatch=%08x\n",
5788 	       vmcs_read32(EXCEPTION_BITMAP),
5789 	       vmcs_read32(PAGE_FAULT_ERROR_CODE_MASK),
5790 	       vmcs_read32(PAGE_FAULT_ERROR_CODE_MATCH));
5791 	pr_err("VMEntry: intr_info=%08x errcode=%08x ilen=%08x\n",
5792 	       vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
5793 	       vmcs_read32(VM_ENTRY_EXCEPTION_ERROR_CODE),
5794 	       vmcs_read32(VM_ENTRY_INSTRUCTION_LEN));
5795 	pr_err("VMExit: intr_info=%08x errcode=%08x ilen=%08x\n",
5796 	       vmcs_read32(VM_EXIT_INTR_INFO),
5797 	       vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
5798 	       vmcs_read32(VM_EXIT_INSTRUCTION_LEN));
5799 	pr_err("        reason=%08x qualification=%016lx\n",
5800 	       vmcs_read32(VM_EXIT_REASON), vmcs_readl(EXIT_QUALIFICATION));
5801 	pr_err("IDTVectoring: info=%08x errcode=%08x\n",
5802 	       vmcs_read32(IDT_VECTORING_INFO_FIELD),
5803 	       vmcs_read32(IDT_VECTORING_ERROR_CODE));
5804 	pr_err("TSC Offset = 0x%016llx\n", vmcs_read64(TSC_OFFSET));
5805 	if (secondary_exec_control & SECONDARY_EXEC_TSC_SCALING)
5806 		pr_err("TSC Multiplier = 0x%016llx\n",
5807 		       vmcs_read64(TSC_MULTIPLIER));
5808 	if (cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW) {
5809 		if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY) {
5810 			u16 status = vmcs_read16(GUEST_INTR_STATUS);
5811 			pr_err("SVI|RVI = %02x|%02x ", status >> 8, status & 0xff);
5812 		}
5813 		pr_cont("TPR Threshold = 0x%02x\n", vmcs_read32(TPR_THRESHOLD));
5814 		if (secondary_exec_control & SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)
5815 			pr_err("APIC-access addr = 0x%016llx ", vmcs_read64(APIC_ACCESS_ADDR));
5816 		pr_cont("virt-APIC addr = 0x%016llx\n", vmcs_read64(VIRTUAL_APIC_PAGE_ADDR));
5817 	}
5818 	if (pin_based_exec_ctrl & PIN_BASED_POSTED_INTR)
5819 		pr_err("PostedIntrVec = 0x%02x\n", vmcs_read16(POSTED_INTR_NV));
5820 	if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT))
5821 		pr_err("EPT pointer = 0x%016llx\n", vmcs_read64(EPT_POINTER));
5822 	n = vmcs_read32(CR3_TARGET_COUNT);
5823 	for (i = 0; i + 1 < n; i += 4)
5824 		pr_err("CR3 target%u=%016lx target%u=%016lx\n",
5825 		       i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2),
5826 		       i + 1, vmcs_readl(CR3_TARGET_VALUE0 + i * 2 + 2));
5827 	if (i < n)
5828 		pr_err("CR3 target%u=%016lx\n",
5829 		       i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2));
5830 	if (secondary_exec_control & SECONDARY_EXEC_PAUSE_LOOP_EXITING)
5831 		pr_err("PLE Gap=%08x Window=%08x\n",
5832 		       vmcs_read32(PLE_GAP), vmcs_read32(PLE_WINDOW));
5833 	if (secondary_exec_control & SECONDARY_EXEC_ENABLE_VPID)
5834 		pr_err("Virtual processor ID = 0x%04x\n",
5835 		       vmcs_read16(VIRTUAL_PROCESSOR_ID));
5836 }
5837 
5838 /*
5839  * The guest has exited.  See if we can fix it or if we need userspace
5840  * assistance.
5841  */
5842 static int vmx_handle_exit(struct kvm_vcpu *vcpu,
5843 	enum exit_fastpath_completion exit_fastpath)
5844 {
5845 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5846 	u32 exit_reason = vmx->exit_reason;
5847 	u32 vectoring_info = vmx->idt_vectoring_info;
5848 
5849 	trace_kvm_exit(exit_reason, vcpu, KVM_ISA_VMX);
5850 
5851 	/*
5852 	 * Flush logged GPAs PML buffer, this will make dirty_bitmap more
5853 	 * updated. Another good is, in kvm_vm_ioctl_get_dirty_log, before
5854 	 * querying dirty_bitmap, we only need to kick all vcpus out of guest
5855 	 * mode as if vcpus is in root mode, the PML buffer must has been
5856 	 * flushed already.
5857 	 */
5858 	if (enable_pml)
5859 		vmx_flush_pml_buffer(vcpu);
5860 
5861 	/* If guest state is invalid, start emulating */
5862 	if (vmx->emulation_required)
5863 		return handle_invalid_guest_state(vcpu);
5864 
5865 	if (is_guest_mode(vcpu) && nested_vmx_exit_reflected(vcpu, exit_reason))
5866 		return nested_vmx_reflect_vmexit(vcpu, exit_reason);
5867 
5868 	if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) {
5869 		dump_vmcs();
5870 		vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
5871 		vcpu->run->fail_entry.hardware_entry_failure_reason
5872 			= exit_reason;
5873 		return 0;
5874 	}
5875 
5876 	if (unlikely(vmx->fail)) {
5877 		dump_vmcs();
5878 		vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
5879 		vcpu->run->fail_entry.hardware_entry_failure_reason
5880 			= vmcs_read32(VM_INSTRUCTION_ERROR);
5881 		return 0;
5882 	}
5883 
5884 	/*
5885 	 * Note:
5886 	 * Do not try to fix EXIT_REASON_EPT_MISCONFIG if it caused by
5887 	 * delivery event since it indicates guest is accessing MMIO.
5888 	 * The vm-exit can be triggered again after return to guest that
5889 	 * will cause infinite loop.
5890 	 */
5891 	if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
5892 			(exit_reason != EXIT_REASON_EXCEPTION_NMI &&
5893 			exit_reason != EXIT_REASON_EPT_VIOLATION &&
5894 			exit_reason != EXIT_REASON_PML_FULL &&
5895 			exit_reason != EXIT_REASON_TASK_SWITCH)) {
5896 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5897 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_DELIVERY_EV;
5898 		vcpu->run->internal.ndata = 3;
5899 		vcpu->run->internal.data[0] = vectoring_info;
5900 		vcpu->run->internal.data[1] = exit_reason;
5901 		vcpu->run->internal.data[2] = vcpu->arch.exit_qualification;
5902 		if (exit_reason == EXIT_REASON_EPT_MISCONFIG) {
5903 			vcpu->run->internal.ndata++;
5904 			vcpu->run->internal.data[3] =
5905 				vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5906 		}
5907 		return 0;
5908 	}
5909 
5910 	if (unlikely(!enable_vnmi &&
5911 		     vmx->loaded_vmcs->soft_vnmi_blocked)) {
5912 		if (vmx_interrupt_allowed(vcpu)) {
5913 			vmx->loaded_vmcs->soft_vnmi_blocked = 0;
5914 		} else if (vmx->loaded_vmcs->vnmi_blocked_time > 1000000000LL &&
5915 			   vcpu->arch.nmi_pending) {
5916 			/*
5917 			 * This CPU don't support us in finding the end of an
5918 			 * NMI-blocked window if the guest runs with IRQs
5919 			 * disabled. So we pull the trigger after 1 s of
5920 			 * futile waiting, but inform the user about this.
5921 			 */
5922 			printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
5923 			       "state on VCPU %d after 1 s timeout\n",
5924 			       __func__, vcpu->vcpu_id);
5925 			vmx->loaded_vmcs->soft_vnmi_blocked = 0;
5926 		}
5927 	}
5928 
5929 	if (exit_fastpath == EXIT_FASTPATH_SKIP_EMUL_INS) {
5930 		kvm_skip_emulated_instruction(vcpu);
5931 		return 1;
5932 	}
5933 
5934 	if (exit_reason >= kvm_vmx_max_exit_handlers)
5935 		goto unexpected_vmexit;
5936 #ifdef CONFIG_RETPOLINE
5937 	if (exit_reason == EXIT_REASON_MSR_WRITE)
5938 		return kvm_emulate_wrmsr(vcpu);
5939 	else if (exit_reason == EXIT_REASON_PREEMPTION_TIMER)
5940 		return handle_preemption_timer(vcpu);
5941 	else if (exit_reason == EXIT_REASON_INTERRUPT_WINDOW)
5942 		return handle_interrupt_window(vcpu);
5943 	else if (exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT)
5944 		return handle_external_interrupt(vcpu);
5945 	else if (exit_reason == EXIT_REASON_HLT)
5946 		return kvm_emulate_halt(vcpu);
5947 	else if (exit_reason == EXIT_REASON_EPT_MISCONFIG)
5948 		return handle_ept_misconfig(vcpu);
5949 #endif
5950 
5951 	exit_reason = array_index_nospec(exit_reason,
5952 					 kvm_vmx_max_exit_handlers);
5953 	if (!kvm_vmx_exit_handlers[exit_reason])
5954 		goto unexpected_vmexit;
5955 
5956 	return kvm_vmx_exit_handlers[exit_reason](vcpu);
5957 
5958 unexpected_vmexit:
5959 	vcpu_unimpl(vcpu, "vmx: unexpected exit reason 0x%x\n", exit_reason);
5960 	dump_vmcs();
5961 	vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5962 	vcpu->run->internal.suberror =
5963 			KVM_INTERNAL_ERROR_UNEXPECTED_EXIT_REASON;
5964 	vcpu->run->internal.ndata = 1;
5965 	vcpu->run->internal.data[0] = exit_reason;
5966 	return 0;
5967 }
5968 
5969 /*
5970  * Software based L1D cache flush which is used when microcode providing
5971  * the cache control MSR is not loaded.
5972  *
5973  * The L1D cache is 32 KiB on Nehalem and later microarchitectures, but to
5974  * flush it is required to read in 64 KiB because the replacement algorithm
5975  * is not exactly LRU. This could be sized at runtime via topology
5976  * information but as all relevant affected CPUs have 32KiB L1D cache size
5977  * there is no point in doing so.
5978  */
5979 static void vmx_l1d_flush(struct kvm_vcpu *vcpu)
5980 {
5981 	int size = PAGE_SIZE << L1D_CACHE_ORDER;
5982 
5983 	/*
5984 	 * This code is only executed when the the flush mode is 'cond' or
5985 	 * 'always'
5986 	 */
5987 	if (static_branch_likely(&vmx_l1d_flush_cond)) {
5988 		bool flush_l1d;
5989 
5990 		/*
5991 		 * Clear the per-vcpu flush bit, it gets set again
5992 		 * either from vcpu_run() or from one of the unsafe
5993 		 * VMEXIT handlers.
5994 		 */
5995 		flush_l1d = vcpu->arch.l1tf_flush_l1d;
5996 		vcpu->arch.l1tf_flush_l1d = false;
5997 
5998 		/*
5999 		 * Clear the per-cpu flush bit, it gets set again from
6000 		 * the interrupt handlers.
6001 		 */
6002 		flush_l1d |= kvm_get_cpu_l1tf_flush_l1d();
6003 		kvm_clear_cpu_l1tf_flush_l1d();
6004 
6005 		if (!flush_l1d)
6006 			return;
6007 	}
6008 
6009 	vcpu->stat.l1d_flush++;
6010 
6011 	if (static_cpu_has(X86_FEATURE_FLUSH_L1D)) {
6012 		wrmsrl(MSR_IA32_FLUSH_CMD, L1D_FLUSH);
6013 		return;
6014 	}
6015 
6016 	asm volatile(
6017 		/* First ensure the pages are in the TLB */
6018 		"xorl	%%eax, %%eax\n"
6019 		".Lpopulate_tlb:\n\t"
6020 		"movzbl	(%[flush_pages], %%" _ASM_AX "), %%ecx\n\t"
6021 		"addl	$4096, %%eax\n\t"
6022 		"cmpl	%%eax, %[size]\n\t"
6023 		"jne	.Lpopulate_tlb\n\t"
6024 		"xorl	%%eax, %%eax\n\t"
6025 		"cpuid\n\t"
6026 		/* Now fill the cache */
6027 		"xorl	%%eax, %%eax\n"
6028 		".Lfill_cache:\n"
6029 		"movzbl	(%[flush_pages], %%" _ASM_AX "), %%ecx\n\t"
6030 		"addl	$64, %%eax\n\t"
6031 		"cmpl	%%eax, %[size]\n\t"
6032 		"jne	.Lfill_cache\n\t"
6033 		"lfence\n"
6034 		:: [flush_pages] "r" (vmx_l1d_flush_pages),
6035 		    [size] "r" (size)
6036 		: "eax", "ebx", "ecx", "edx");
6037 }
6038 
6039 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
6040 {
6041 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6042 	int tpr_threshold;
6043 
6044 	if (is_guest_mode(vcpu) &&
6045 		nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
6046 		return;
6047 
6048 	tpr_threshold = (irr == -1 || tpr < irr) ? 0 : irr;
6049 	if (is_guest_mode(vcpu))
6050 		to_vmx(vcpu)->nested.l1_tpr_threshold = tpr_threshold;
6051 	else
6052 		vmcs_write32(TPR_THRESHOLD, tpr_threshold);
6053 }
6054 
6055 void vmx_set_virtual_apic_mode(struct kvm_vcpu *vcpu)
6056 {
6057 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6058 	u32 sec_exec_control;
6059 
6060 	if (!lapic_in_kernel(vcpu))
6061 		return;
6062 
6063 	if (!flexpriority_enabled &&
6064 	    !cpu_has_vmx_virtualize_x2apic_mode())
6065 		return;
6066 
6067 	/* Postpone execution until vmcs01 is the current VMCS. */
6068 	if (is_guest_mode(vcpu)) {
6069 		vmx->nested.change_vmcs01_virtual_apic_mode = true;
6070 		return;
6071 	}
6072 
6073 	sec_exec_control = secondary_exec_controls_get(vmx);
6074 	sec_exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
6075 			      SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE);
6076 
6077 	switch (kvm_get_apic_mode(vcpu)) {
6078 	case LAPIC_MODE_INVALID:
6079 		WARN_ONCE(true, "Invalid local APIC state");
6080 	case LAPIC_MODE_DISABLED:
6081 		break;
6082 	case LAPIC_MODE_XAPIC:
6083 		if (flexpriority_enabled) {
6084 			sec_exec_control |=
6085 				SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
6086 			vmx_flush_tlb(vcpu, true);
6087 		}
6088 		break;
6089 	case LAPIC_MODE_X2APIC:
6090 		if (cpu_has_vmx_virtualize_x2apic_mode())
6091 			sec_exec_control |=
6092 				SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
6093 		break;
6094 	}
6095 	secondary_exec_controls_set(vmx, sec_exec_control);
6096 
6097 	vmx_update_msr_bitmap(vcpu);
6098 }
6099 
6100 static void vmx_set_apic_access_page_addr(struct kvm_vcpu *vcpu, hpa_t hpa)
6101 {
6102 	if (!is_guest_mode(vcpu)) {
6103 		vmcs_write64(APIC_ACCESS_ADDR, hpa);
6104 		vmx_flush_tlb(vcpu, true);
6105 	}
6106 }
6107 
6108 static void vmx_hwapic_isr_update(struct kvm_vcpu *vcpu, int max_isr)
6109 {
6110 	u16 status;
6111 	u8 old;
6112 
6113 	if (max_isr == -1)
6114 		max_isr = 0;
6115 
6116 	status = vmcs_read16(GUEST_INTR_STATUS);
6117 	old = status >> 8;
6118 	if (max_isr != old) {
6119 		status &= 0xff;
6120 		status |= max_isr << 8;
6121 		vmcs_write16(GUEST_INTR_STATUS, status);
6122 	}
6123 }
6124 
6125 static void vmx_set_rvi(int vector)
6126 {
6127 	u16 status;
6128 	u8 old;
6129 
6130 	if (vector == -1)
6131 		vector = 0;
6132 
6133 	status = vmcs_read16(GUEST_INTR_STATUS);
6134 	old = (u8)status & 0xff;
6135 	if ((u8)vector != old) {
6136 		status &= ~0xff;
6137 		status |= (u8)vector;
6138 		vmcs_write16(GUEST_INTR_STATUS, status);
6139 	}
6140 }
6141 
6142 static void vmx_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr)
6143 {
6144 	/*
6145 	 * When running L2, updating RVI is only relevant when
6146 	 * vmcs12 virtual-interrupt-delivery enabled.
6147 	 * However, it can be enabled only when L1 also
6148 	 * intercepts external-interrupts and in that case
6149 	 * we should not update vmcs02 RVI but instead intercept
6150 	 * interrupt. Therefore, do nothing when running L2.
6151 	 */
6152 	if (!is_guest_mode(vcpu))
6153 		vmx_set_rvi(max_irr);
6154 }
6155 
6156 static int vmx_sync_pir_to_irr(struct kvm_vcpu *vcpu)
6157 {
6158 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6159 	int max_irr;
6160 	bool max_irr_updated;
6161 
6162 	WARN_ON(!vcpu->arch.apicv_active);
6163 	if (pi_test_on(&vmx->pi_desc)) {
6164 		pi_clear_on(&vmx->pi_desc);
6165 		/*
6166 		 * IOMMU can write to PID.ON, so the barrier matters even on UP.
6167 		 * But on x86 this is just a compiler barrier anyway.
6168 		 */
6169 		smp_mb__after_atomic();
6170 		max_irr_updated =
6171 			kvm_apic_update_irr(vcpu, vmx->pi_desc.pir, &max_irr);
6172 
6173 		/*
6174 		 * If we are running L2 and L1 has a new pending interrupt
6175 		 * which can be injected, we should re-evaluate
6176 		 * what should be done with this new L1 interrupt.
6177 		 * If L1 intercepts external-interrupts, we should
6178 		 * exit from L2 to L1. Otherwise, interrupt should be
6179 		 * delivered directly to L2.
6180 		 */
6181 		if (is_guest_mode(vcpu) && max_irr_updated) {
6182 			if (nested_exit_on_intr(vcpu))
6183 				kvm_vcpu_exiting_guest_mode(vcpu);
6184 			else
6185 				kvm_make_request(KVM_REQ_EVENT, vcpu);
6186 		}
6187 	} else {
6188 		max_irr = kvm_lapic_find_highest_irr(vcpu);
6189 	}
6190 	vmx_hwapic_irr_update(vcpu, max_irr);
6191 	return max_irr;
6192 }
6193 
6194 static bool vmx_dy_apicv_has_pending_interrupt(struct kvm_vcpu *vcpu)
6195 {
6196 	struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
6197 
6198 	return pi_test_on(pi_desc) ||
6199 		(pi_test_sn(pi_desc) && !pi_is_pir_empty(pi_desc));
6200 }
6201 
6202 static void vmx_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap)
6203 {
6204 	if (!kvm_vcpu_apicv_active(vcpu))
6205 		return;
6206 
6207 	vmcs_write64(EOI_EXIT_BITMAP0, eoi_exit_bitmap[0]);
6208 	vmcs_write64(EOI_EXIT_BITMAP1, eoi_exit_bitmap[1]);
6209 	vmcs_write64(EOI_EXIT_BITMAP2, eoi_exit_bitmap[2]);
6210 	vmcs_write64(EOI_EXIT_BITMAP3, eoi_exit_bitmap[3]);
6211 }
6212 
6213 static void vmx_apicv_post_state_restore(struct kvm_vcpu *vcpu)
6214 {
6215 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6216 
6217 	pi_clear_on(&vmx->pi_desc);
6218 	memset(vmx->pi_desc.pir, 0, sizeof(vmx->pi_desc.pir));
6219 }
6220 
6221 static void handle_exception_nmi_irqoff(struct vcpu_vmx *vmx)
6222 {
6223 	vmx->exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6224 
6225 	/* if exit due to PF check for async PF */
6226 	if (is_page_fault(vmx->exit_intr_info))
6227 		vmx->vcpu.arch.apf.host_apf_reason = kvm_read_and_reset_pf_reason();
6228 
6229 	/* Handle machine checks before interrupts are enabled */
6230 	if (is_machine_check(vmx->exit_intr_info))
6231 		kvm_machine_check();
6232 
6233 	/* We need to handle NMIs before interrupts are enabled */
6234 	if (is_nmi(vmx->exit_intr_info)) {
6235 		kvm_before_interrupt(&vmx->vcpu);
6236 		asm("int $2");
6237 		kvm_after_interrupt(&vmx->vcpu);
6238 	}
6239 }
6240 
6241 static void handle_external_interrupt_irqoff(struct kvm_vcpu *vcpu)
6242 {
6243 	unsigned int vector;
6244 	unsigned long entry;
6245 #ifdef CONFIG_X86_64
6246 	unsigned long tmp;
6247 #endif
6248 	gate_desc *desc;
6249 	u32 intr_info;
6250 
6251 	intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6252 	if (WARN_ONCE(!is_external_intr(intr_info),
6253 	    "KVM: unexpected VM-Exit interrupt info: 0x%x", intr_info))
6254 		return;
6255 
6256 	vector = intr_info & INTR_INFO_VECTOR_MASK;
6257 	desc = (gate_desc *)host_idt_base + vector;
6258 	entry = gate_offset(desc);
6259 
6260 	kvm_before_interrupt(vcpu);
6261 
6262 	asm volatile(
6263 #ifdef CONFIG_X86_64
6264 		"mov %%" _ASM_SP ", %[sp]\n\t"
6265 		"and $0xfffffffffffffff0, %%" _ASM_SP "\n\t"
6266 		"push $%c[ss]\n\t"
6267 		"push %[sp]\n\t"
6268 #endif
6269 		"pushf\n\t"
6270 		__ASM_SIZE(push) " $%c[cs]\n\t"
6271 		CALL_NOSPEC
6272 		:
6273 #ifdef CONFIG_X86_64
6274 		[sp]"=&r"(tmp),
6275 #endif
6276 		ASM_CALL_CONSTRAINT
6277 		:
6278 		THUNK_TARGET(entry),
6279 		[ss]"i"(__KERNEL_DS),
6280 		[cs]"i"(__KERNEL_CS)
6281 	);
6282 
6283 	kvm_after_interrupt(vcpu);
6284 }
6285 STACK_FRAME_NON_STANDARD(handle_external_interrupt_irqoff);
6286 
6287 static void vmx_handle_exit_irqoff(struct kvm_vcpu *vcpu,
6288 	enum exit_fastpath_completion *exit_fastpath)
6289 {
6290 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6291 
6292 	if (vmx->exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT)
6293 		handle_external_interrupt_irqoff(vcpu);
6294 	else if (vmx->exit_reason == EXIT_REASON_EXCEPTION_NMI)
6295 		handle_exception_nmi_irqoff(vmx);
6296 	else if (!is_guest_mode(vcpu) &&
6297 		vmx->exit_reason == EXIT_REASON_MSR_WRITE)
6298 		*exit_fastpath = handle_fastpath_set_msr_irqoff(vcpu);
6299 }
6300 
6301 static bool vmx_has_emulated_msr(int index)
6302 {
6303 	switch (index) {
6304 	case MSR_IA32_SMBASE:
6305 		/*
6306 		 * We cannot do SMM unless we can run the guest in big
6307 		 * real mode.
6308 		 */
6309 		return enable_unrestricted_guest || emulate_invalid_guest_state;
6310 	case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
6311 		return nested;
6312 	case MSR_AMD64_VIRT_SPEC_CTRL:
6313 		/* This is AMD only.  */
6314 		return false;
6315 	default:
6316 		return true;
6317 	}
6318 }
6319 
6320 static bool vmx_pt_supported(void)
6321 {
6322 	return pt_mode == PT_MODE_HOST_GUEST;
6323 }
6324 
6325 static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx)
6326 {
6327 	u32 exit_intr_info;
6328 	bool unblock_nmi;
6329 	u8 vector;
6330 	bool idtv_info_valid;
6331 
6332 	idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK;
6333 
6334 	if (enable_vnmi) {
6335 		if (vmx->loaded_vmcs->nmi_known_unmasked)
6336 			return;
6337 		/*
6338 		 * Can't use vmx->exit_intr_info since we're not sure what
6339 		 * the exit reason is.
6340 		 */
6341 		exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6342 		unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
6343 		vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
6344 		/*
6345 		 * SDM 3: 27.7.1.2 (September 2008)
6346 		 * Re-set bit "block by NMI" before VM entry if vmexit caused by
6347 		 * a guest IRET fault.
6348 		 * SDM 3: 23.2.2 (September 2008)
6349 		 * Bit 12 is undefined in any of the following cases:
6350 		 *  If the VM exit sets the valid bit in the IDT-vectoring
6351 		 *   information field.
6352 		 *  If the VM exit is due to a double fault.
6353 		 */
6354 		if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
6355 		    vector != DF_VECTOR && !idtv_info_valid)
6356 			vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
6357 				      GUEST_INTR_STATE_NMI);
6358 		else
6359 			vmx->loaded_vmcs->nmi_known_unmasked =
6360 				!(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO)
6361 				  & GUEST_INTR_STATE_NMI);
6362 	} else if (unlikely(vmx->loaded_vmcs->soft_vnmi_blocked))
6363 		vmx->loaded_vmcs->vnmi_blocked_time +=
6364 			ktime_to_ns(ktime_sub(ktime_get(),
6365 					      vmx->loaded_vmcs->entry_time));
6366 }
6367 
6368 static void __vmx_complete_interrupts(struct kvm_vcpu *vcpu,
6369 				      u32 idt_vectoring_info,
6370 				      int instr_len_field,
6371 				      int error_code_field)
6372 {
6373 	u8 vector;
6374 	int type;
6375 	bool idtv_info_valid;
6376 
6377 	idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
6378 
6379 	vcpu->arch.nmi_injected = false;
6380 	kvm_clear_exception_queue(vcpu);
6381 	kvm_clear_interrupt_queue(vcpu);
6382 
6383 	if (!idtv_info_valid)
6384 		return;
6385 
6386 	kvm_make_request(KVM_REQ_EVENT, vcpu);
6387 
6388 	vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
6389 	type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
6390 
6391 	switch (type) {
6392 	case INTR_TYPE_NMI_INTR:
6393 		vcpu->arch.nmi_injected = true;
6394 		/*
6395 		 * SDM 3: 27.7.1.2 (September 2008)
6396 		 * Clear bit "block by NMI" before VM entry if a NMI
6397 		 * delivery faulted.
6398 		 */
6399 		vmx_set_nmi_mask(vcpu, false);
6400 		break;
6401 	case INTR_TYPE_SOFT_EXCEPTION:
6402 		vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
6403 		/* fall through */
6404 	case INTR_TYPE_HARD_EXCEPTION:
6405 		if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
6406 			u32 err = vmcs_read32(error_code_field);
6407 			kvm_requeue_exception_e(vcpu, vector, err);
6408 		} else
6409 			kvm_requeue_exception(vcpu, vector);
6410 		break;
6411 	case INTR_TYPE_SOFT_INTR:
6412 		vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
6413 		/* fall through */
6414 	case INTR_TYPE_EXT_INTR:
6415 		kvm_queue_interrupt(vcpu, vector, type == INTR_TYPE_SOFT_INTR);
6416 		break;
6417 	default:
6418 		break;
6419 	}
6420 }
6421 
6422 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
6423 {
6424 	__vmx_complete_interrupts(&vmx->vcpu, vmx->idt_vectoring_info,
6425 				  VM_EXIT_INSTRUCTION_LEN,
6426 				  IDT_VECTORING_ERROR_CODE);
6427 }
6428 
6429 static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
6430 {
6431 	__vmx_complete_interrupts(vcpu,
6432 				  vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
6433 				  VM_ENTRY_INSTRUCTION_LEN,
6434 				  VM_ENTRY_EXCEPTION_ERROR_CODE);
6435 
6436 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
6437 }
6438 
6439 static void atomic_switch_perf_msrs(struct vcpu_vmx *vmx)
6440 {
6441 	int i, nr_msrs;
6442 	struct perf_guest_switch_msr *msrs;
6443 
6444 	msrs = perf_guest_get_msrs(&nr_msrs);
6445 
6446 	if (!msrs)
6447 		return;
6448 
6449 	for (i = 0; i < nr_msrs; i++)
6450 		if (msrs[i].host == msrs[i].guest)
6451 			clear_atomic_switch_msr(vmx, msrs[i].msr);
6452 		else
6453 			add_atomic_switch_msr(vmx, msrs[i].msr, msrs[i].guest,
6454 					msrs[i].host, false);
6455 }
6456 
6457 static void atomic_switch_umwait_control_msr(struct vcpu_vmx *vmx)
6458 {
6459 	u32 host_umwait_control;
6460 
6461 	if (!vmx_has_waitpkg(vmx))
6462 		return;
6463 
6464 	host_umwait_control = get_umwait_control_msr();
6465 
6466 	if (vmx->msr_ia32_umwait_control != host_umwait_control)
6467 		add_atomic_switch_msr(vmx, MSR_IA32_UMWAIT_CONTROL,
6468 			vmx->msr_ia32_umwait_control,
6469 			host_umwait_control, false);
6470 	else
6471 		clear_atomic_switch_msr(vmx, MSR_IA32_UMWAIT_CONTROL);
6472 }
6473 
6474 static void vmx_update_hv_timer(struct kvm_vcpu *vcpu)
6475 {
6476 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6477 	u64 tscl;
6478 	u32 delta_tsc;
6479 
6480 	if (vmx->req_immediate_exit) {
6481 		vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, 0);
6482 		vmx->loaded_vmcs->hv_timer_soft_disabled = false;
6483 	} else if (vmx->hv_deadline_tsc != -1) {
6484 		tscl = rdtsc();
6485 		if (vmx->hv_deadline_tsc > tscl)
6486 			/* set_hv_timer ensures the delta fits in 32-bits */
6487 			delta_tsc = (u32)((vmx->hv_deadline_tsc - tscl) >>
6488 				cpu_preemption_timer_multi);
6489 		else
6490 			delta_tsc = 0;
6491 
6492 		vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, delta_tsc);
6493 		vmx->loaded_vmcs->hv_timer_soft_disabled = false;
6494 	} else if (!vmx->loaded_vmcs->hv_timer_soft_disabled) {
6495 		vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, -1);
6496 		vmx->loaded_vmcs->hv_timer_soft_disabled = true;
6497 	}
6498 }
6499 
6500 void vmx_update_host_rsp(struct vcpu_vmx *vmx, unsigned long host_rsp)
6501 {
6502 	if (unlikely(host_rsp != vmx->loaded_vmcs->host_state.rsp)) {
6503 		vmx->loaded_vmcs->host_state.rsp = host_rsp;
6504 		vmcs_writel(HOST_RSP, host_rsp);
6505 	}
6506 }
6507 
6508 bool __vmx_vcpu_run(struct vcpu_vmx *vmx, unsigned long *regs, bool launched);
6509 
6510 static void vmx_vcpu_run(struct kvm_vcpu *vcpu)
6511 {
6512 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6513 	unsigned long cr3, cr4;
6514 
6515 	/* Record the guest's net vcpu time for enforced NMI injections. */
6516 	if (unlikely(!enable_vnmi &&
6517 		     vmx->loaded_vmcs->soft_vnmi_blocked))
6518 		vmx->loaded_vmcs->entry_time = ktime_get();
6519 
6520 	/* Don't enter VMX if guest state is invalid, let the exit handler
6521 	   start emulation until we arrive back to a valid state */
6522 	if (vmx->emulation_required)
6523 		return;
6524 
6525 	if (vmx->ple_window_dirty) {
6526 		vmx->ple_window_dirty = false;
6527 		vmcs_write32(PLE_WINDOW, vmx->ple_window);
6528 	}
6529 
6530 	/*
6531 	 * We did this in prepare_switch_to_guest, because it needs to
6532 	 * be within srcu_read_lock.
6533 	 */
6534 	WARN_ON_ONCE(vmx->nested.need_vmcs12_to_shadow_sync);
6535 
6536 	if (kvm_register_is_dirty(vcpu, VCPU_REGS_RSP))
6537 		vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
6538 	if (kvm_register_is_dirty(vcpu, VCPU_REGS_RIP))
6539 		vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
6540 
6541 	cr3 = __get_current_cr3_fast();
6542 	if (unlikely(cr3 != vmx->loaded_vmcs->host_state.cr3)) {
6543 		vmcs_writel(HOST_CR3, cr3);
6544 		vmx->loaded_vmcs->host_state.cr3 = cr3;
6545 	}
6546 
6547 	cr4 = cr4_read_shadow();
6548 	if (unlikely(cr4 != vmx->loaded_vmcs->host_state.cr4)) {
6549 		vmcs_writel(HOST_CR4, cr4);
6550 		vmx->loaded_vmcs->host_state.cr4 = cr4;
6551 	}
6552 
6553 	/* When single-stepping over STI and MOV SS, we must clear the
6554 	 * corresponding interruptibility bits in the guest state. Otherwise
6555 	 * vmentry fails as it then expects bit 14 (BS) in pending debug
6556 	 * exceptions being set, but that's not correct for the guest debugging
6557 	 * case. */
6558 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6559 		vmx_set_interrupt_shadow(vcpu, 0);
6560 
6561 	kvm_load_guest_xsave_state(vcpu);
6562 
6563 	if (static_cpu_has(X86_FEATURE_PKU) &&
6564 	    kvm_read_cr4_bits(vcpu, X86_CR4_PKE) &&
6565 	    vcpu->arch.pkru != vmx->host_pkru)
6566 		__write_pkru(vcpu->arch.pkru);
6567 
6568 	pt_guest_enter(vmx);
6569 
6570 	atomic_switch_perf_msrs(vmx);
6571 	atomic_switch_umwait_control_msr(vmx);
6572 
6573 	if (enable_preemption_timer)
6574 		vmx_update_hv_timer(vcpu);
6575 
6576 	if (lapic_in_kernel(vcpu) &&
6577 		vcpu->arch.apic->lapic_timer.timer_advance_ns)
6578 		kvm_wait_lapic_expire(vcpu);
6579 
6580 	/*
6581 	 * If this vCPU has touched SPEC_CTRL, restore the guest's value if
6582 	 * it's non-zero. Since vmentry is serialising on affected CPUs, there
6583 	 * is no need to worry about the conditional branch over the wrmsr
6584 	 * being speculatively taken.
6585 	 */
6586 	x86_spec_ctrl_set_guest(vmx->spec_ctrl, 0);
6587 
6588 	/* L1D Flush includes CPU buffer clear to mitigate MDS */
6589 	if (static_branch_unlikely(&vmx_l1d_should_flush))
6590 		vmx_l1d_flush(vcpu);
6591 	else if (static_branch_unlikely(&mds_user_clear))
6592 		mds_clear_cpu_buffers();
6593 
6594 	if (vcpu->arch.cr2 != read_cr2())
6595 		write_cr2(vcpu->arch.cr2);
6596 
6597 	vmx->fail = __vmx_vcpu_run(vmx, (unsigned long *)&vcpu->arch.regs,
6598 				   vmx->loaded_vmcs->launched);
6599 
6600 	vcpu->arch.cr2 = read_cr2();
6601 
6602 	/*
6603 	 * We do not use IBRS in the kernel. If this vCPU has used the
6604 	 * SPEC_CTRL MSR it may have left it on; save the value and
6605 	 * turn it off. This is much more efficient than blindly adding
6606 	 * it to the atomic save/restore list. Especially as the former
6607 	 * (Saving guest MSRs on vmexit) doesn't even exist in KVM.
6608 	 *
6609 	 * For non-nested case:
6610 	 * If the L01 MSR bitmap does not intercept the MSR, then we need to
6611 	 * save it.
6612 	 *
6613 	 * For nested case:
6614 	 * If the L02 MSR bitmap does not intercept the MSR, then we need to
6615 	 * save it.
6616 	 */
6617 	if (unlikely(!msr_write_intercepted(vcpu, MSR_IA32_SPEC_CTRL)))
6618 		vmx->spec_ctrl = native_read_msr(MSR_IA32_SPEC_CTRL);
6619 
6620 	x86_spec_ctrl_restore_host(vmx->spec_ctrl, 0);
6621 
6622 	/* All fields are clean at this point */
6623 	if (static_branch_unlikely(&enable_evmcs))
6624 		current_evmcs->hv_clean_fields |=
6625 			HV_VMX_ENLIGHTENED_CLEAN_FIELD_ALL;
6626 
6627 	if (static_branch_unlikely(&enable_evmcs))
6628 		current_evmcs->hv_vp_id = vcpu->arch.hyperv.vp_index;
6629 
6630 	/* MSR_IA32_DEBUGCTLMSR is zeroed on vmexit. Restore it if needed */
6631 	if (vmx->host_debugctlmsr)
6632 		update_debugctlmsr(vmx->host_debugctlmsr);
6633 
6634 #ifndef CONFIG_X86_64
6635 	/*
6636 	 * The sysexit path does not restore ds/es, so we must set them to
6637 	 * a reasonable value ourselves.
6638 	 *
6639 	 * We can't defer this to vmx_prepare_switch_to_host() since that
6640 	 * function may be executed in interrupt context, which saves and
6641 	 * restore segments around it, nullifying its effect.
6642 	 */
6643 	loadsegment(ds, __USER_DS);
6644 	loadsegment(es, __USER_DS);
6645 #endif
6646 
6647 	vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
6648 				  | (1 << VCPU_EXREG_RFLAGS)
6649 				  | (1 << VCPU_EXREG_PDPTR)
6650 				  | (1 << VCPU_EXREG_SEGMENTS)
6651 				  | (1 << VCPU_EXREG_CR3));
6652 	vcpu->arch.regs_dirty = 0;
6653 
6654 	pt_guest_exit(vmx);
6655 
6656 	/*
6657 	 * eager fpu is enabled if PKEY is supported and CR4 is switched
6658 	 * back on host, so it is safe to read guest PKRU from current
6659 	 * XSAVE.
6660 	 */
6661 	if (static_cpu_has(X86_FEATURE_PKU) &&
6662 	    kvm_read_cr4_bits(vcpu, X86_CR4_PKE)) {
6663 		vcpu->arch.pkru = rdpkru();
6664 		if (vcpu->arch.pkru != vmx->host_pkru)
6665 			__write_pkru(vmx->host_pkru);
6666 	}
6667 
6668 	kvm_load_host_xsave_state(vcpu);
6669 
6670 	vmx->nested.nested_run_pending = 0;
6671 	vmx->idt_vectoring_info = 0;
6672 
6673 	vmx->exit_reason = vmx->fail ? 0xdead : vmcs_read32(VM_EXIT_REASON);
6674 	if ((u16)vmx->exit_reason == EXIT_REASON_MCE_DURING_VMENTRY)
6675 		kvm_machine_check();
6676 
6677 	if (vmx->fail || (vmx->exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY))
6678 		return;
6679 
6680 	vmx->loaded_vmcs->launched = 1;
6681 	vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
6682 
6683 	vmx_recover_nmi_blocking(vmx);
6684 	vmx_complete_interrupts(vmx);
6685 }
6686 
6687 static struct kvm *vmx_vm_alloc(void)
6688 {
6689 	struct kvm_vmx *kvm_vmx = __vmalloc(sizeof(struct kvm_vmx),
6690 					    GFP_KERNEL_ACCOUNT | __GFP_ZERO,
6691 					    PAGE_KERNEL);
6692 	return &kvm_vmx->kvm;
6693 }
6694 
6695 static void vmx_vm_free(struct kvm *kvm)
6696 {
6697 	kfree(kvm->arch.hyperv.hv_pa_pg);
6698 	vfree(to_kvm_vmx(kvm));
6699 }
6700 
6701 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
6702 {
6703 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6704 
6705 	if (enable_pml)
6706 		vmx_destroy_pml_buffer(vmx);
6707 	free_vpid(vmx->vpid);
6708 	nested_vmx_free_vcpu(vcpu);
6709 	free_loaded_vmcs(vmx->loaded_vmcs);
6710 }
6711 
6712 static int vmx_create_vcpu(struct kvm_vcpu *vcpu)
6713 {
6714 	struct vcpu_vmx *vmx;
6715 	unsigned long *msr_bitmap;
6716 	int i, cpu, err;
6717 
6718 	BUILD_BUG_ON(offsetof(struct vcpu_vmx, vcpu) != 0);
6719 	vmx = to_vmx(vcpu);
6720 
6721 	err = -ENOMEM;
6722 
6723 	vmx->vpid = allocate_vpid();
6724 
6725 	/*
6726 	 * If PML is turned on, failure on enabling PML just results in failure
6727 	 * of creating the vcpu, therefore we can simplify PML logic (by
6728 	 * avoiding dealing with cases, such as enabling PML partially on vcpus
6729 	 * for the guest), etc.
6730 	 */
6731 	if (enable_pml) {
6732 		vmx->pml_pg = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
6733 		if (!vmx->pml_pg)
6734 			goto free_vpid;
6735 	}
6736 
6737 	BUILD_BUG_ON(ARRAY_SIZE(vmx_msr_index) != NR_SHARED_MSRS);
6738 
6739 	for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i) {
6740 		u32 index = vmx_msr_index[i];
6741 		u32 data_low, data_high;
6742 		int j = vmx->nmsrs;
6743 
6744 		if (rdmsr_safe(index, &data_low, &data_high) < 0)
6745 			continue;
6746 		if (wrmsr_safe(index, data_low, data_high) < 0)
6747 			continue;
6748 
6749 		vmx->guest_msrs[j].index = i;
6750 		vmx->guest_msrs[j].data = 0;
6751 		switch (index) {
6752 		case MSR_IA32_TSX_CTRL:
6753 			/*
6754 			 * No need to pass TSX_CTRL_CPUID_CLEAR through, so
6755 			 * let's avoid changing CPUID bits under the host
6756 			 * kernel's feet.
6757 			 */
6758 			vmx->guest_msrs[j].mask = ~(u64)TSX_CTRL_CPUID_CLEAR;
6759 			break;
6760 		default:
6761 			vmx->guest_msrs[j].mask = -1ull;
6762 			break;
6763 		}
6764 		++vmx->nmsrs;
6765 	}
6766 
6767 	err = alloc_loaded_vmcs(&vmx->vmcs01);
6768 	if (err < 0)
6769 		goto free_pml;
6770 
6771 	msr_bitmap = vmx->vmcs01.msr_bitmap;
6772 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_TSC, MSR_TYPE_R);
6773 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_FS_BASE, MSR_TYPE_RW);
6774 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_GS_BASE, MSR_TYPE_RW);
6775 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_KERNEL_GS_BASE, MSR_TYPE_RW);
6776 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_CS, MSR_TYPE_RW);
6777 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_ESP, MSR_TYPE_RW);
6778 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_EIP, MSR_TYPE_RW);
6779 	if (kvm_cstate_in_guest(vcpu->kvm)) {
6780 		vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C1_RES, MSR_TYPE_R);
6781 		vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C3_RESIDENCY, MSR_TYPE_R);
6782 		vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C6_RESIDENCY, MSR_TYPE_R);
6783 		vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C7_RESIDENCY, MSR_TYPE_R);
6784 	}
6785 	vmx->msr_bitmap_mode = 0;
6786 
6787 	vmx->loaded_vmcs = &vmx->vmcs01;
6788 	cpu = get_cpu();
6789 	vmx_vcpu_load(vcpu, cpu);
6790 	vcpu->cpu = cpu;
6791 	init_vmcs(vmx);
6792 	vmx_vcpu_put(vcpu);
6793 	put_cpu();
6794 	if (cpu_need_virtualize_apic_accesses(vcpu)) {
6795 		err = alloc_apic_access_page(vcpu->kvm);
6796 		if (err)
6797 			goto free_vmcs;
6798 	}
6799 
6800 	if (enable_ept && !enable_unrestricted_guest) {
6801 		err = init_rmode_identity_map(vcpu->kvm);
6802 		if (err)
6803 			goto free_vmcs;
6804 	}
6805 
6806 	if (nested)
6807 		nested_vmx_setup_ctls_msrs(&vmx->nested.msrs,
6808 					   vmx_capability.ept);
6809 	else
6810 		memset(&vmx->nested.msrs, 0, sizeof(vmx->nested.msrs));
6811 
6812 	vmx->nested.posted_intr_nv = -1;
6813 	vmx->nested.current_vmptr = -1ull;
6814 
6815 	vcpu->arch.microcode_version = 0x100000000ULL;
6816 	vmx->msr_ia32_feature_control_valid_bits = FEAT_CTL_LOCKED;
6817 
6818 	/*
6819 	 * Enforce invariant: pi_desc.nv is always either POSTED_INTR_VECTOR
6820 	 * or POSTED_INTR_WAKEUP_VECTOR.
6821 	 */
6822 	vmx->pi_desc.nv = POSTED_INTR_VECTOR;
6823 	vmx->pi_desc.sn = 1;
6824 
6825 	vmx->ept_pointer = INVALID_PAGE;
6826 
6827 	return 0;
6828 
6829 free_vmcs:
6830 	free_loaded_vmcs(vmx->loaded_vmcs);
6831 free_pml:
6832 	vmx_destroy_pml_buffer(vmx);
6833 free_vpid:
6834 	free_vpid(vmx->vpid);
6835 	return 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 	kvm_apicv_init(kvm, enable_apicv);
6872 	return 0;
6873 }
6874 
6875 static int __init vmx_check_processor_compat(void)
6876 {
6877 	struct vmcs_config vmcs_conf;
6878 	struct vmx_capability vmx_cap;
6879 
6880 	if (!this_cpu_has(X86_FEATURE_MSR_IA32_FEAT_CTL) ||
6881 	    !this_cpu_has(X86_FEATURE_VMX)) {
6882 		pr_err("kvm: VMX is disabled on CPU %d\n", smp_processor_id());
6883 		return -EIO;
6884 	}
6885 
6886 	if (setup_vmcs_config(&vmcs_conf, &vmx_cap) < 0)
6887 		return -EIO;
6888 	if (nested)
6889 		nested_vmx_setup_ctls_msrs(&vmcs_conf.nested, vmx_cap.ept);
6890 	if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
6891 		printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
6892 				smp_processor_id());
6893 		return -EIO;
6894 	}
6895 	return 0;
6896 }
6897 
6898 static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
6899 {
6900 	u8 cache;
6901 	u64 ipat = 0;
6902 
6903 	/* For VT-d and EPT combination
6904 	 * 1. MMIO: always map as UC
6905 	 * 2. EPT with VT-d:
6906 	 *   a. VT-d without snooping control feature: can't guarantee the
6907 	 *	result, try to trust guest.
6908 	 *   b. VT-d with snooping control feature: snooping control feature of
6909 	 *	VT-d engine can guarantee the cache correctness. Just set it
6910 	 *	to WB to keep consistent with host. So the same as item 3.
6911 	 * 3. EPT without VT-d: always map as WB and set IPAT=1 to keep
6912 	 *    consistent with host MTRR
6913 	 */
6914 	if (is_mmio) {
6915 		cache = MTRR_TYPE_UNCACHABLE;
6916 		goto exit;
6917 	}
6918 
6919 	if (!kvm_arch_has_noncoherent_dma(vcpu->kvm)) {
6920 		ipat = VMX_EPT_IPAT_BIT;
6921 		cache = MTRR_TYPE_WRBACK;
6922 		goto exit;
6923 	}
6924 
6925 	if (kvm_read_cr0(vcpu) & X86_CR0_CD) {
6926 		ipat = VMX_EPT_IPAT_BIT;
6927 		if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
6928 			cache = MTRR_TYPE_WRBACK;
6929 		else
6930 			cache = MTRR_TYPE_UNCACHABLE;
6931 		goto exit;
6932 	}
6933 
6934 	cache = kvm_mtrr_get_guest_memory_type(vcpu, gfn);
6935 
6936 exit:
6937 	return (cache << VMX_EPT_MT_EPTE_SHIFT) | ipat;
6938 }
6939 
6940 static int vmx_get_lpage_level(void)
6941 {
6942 	if (enable_ept && !cpu_has_vmx_ept_1g_page())
6943 		return PT_DIRECTORY_LEVEL;
6944 	else
6945 		/* For shadow and EPT supported 1GB page */
6946 		return PT_PDPE_LEVEL;
6947 }
6948 
6949 static void vmcs_set_secondary_exec_control(struct vcpu_vmx *vmx)
6950 {
6951 	/*
6952 	 * These bits in the secondary execution controls field
6953 	 * are dynamic, the others are mostly based on the hypervisor
6954 	 * architecture and the guest's CPUID.  Do not touch the
6955 	 * dynamic bits.
6956 	 */
6957 	u32 mask =
6958 		SECONDARY_EXEC_SHADOW_VMCS |
6959 		SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
6960 		SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
6961 		SECONDARY_EXEC_DESC;
6962 
6963 	u32 new_ctl = vmx->secondary_exec_control;
6964 	u32 cur_ctl = secondary_exec_controls_get(vmx);
6965 
6966 	secondary_exec_controls_set(vmx, (new_ctl & ~mask) | (cur_ctl & mask));
6967 }
6968 
6969 /*
6970  * Generate MSR_IA32_VMX_CR{0,4}_FIXED1 according to CPUID. Only set bits
6971  * (indicating "allowed-1") if they are supported in the guest's CPUID.
6972  */
6973 static void nested_vmx_cr_fixed1_bits_update(struct kvm_vcpu *vcpu)
6974 {
6975 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6976 	struct kvm_cpuid_entry2 *entry;
6977 
6978 	vmx->nested.msrs.cr0_fixed1 = 0xffffffff;
6979 	vmx->nested.msrs.cr4_fixed1 = X86_CR4_PCE;
6980 
6981 #define cr4_fixed1_update(_cr4_mask, _reg, _cpuid_mask) do {		\
6982 	if (entry && (entry->_reg & (_cpuid_mask)))			\
6983 		vmx->nested.msrs.cr4_fixed1 |= (_cr4_mask);	\
6984 } while (0)
6985 
6986 	entry = kvm_find_cpuid_entry(vcpu, 0x1, 0);
6987 	cr4_fixed1_update(X86_CR4_VME,        edx, feature_bit(VME));
6988 	cr4_fixed1_update(X86_CR4_PVI,        edx, feature_bit(VME));
6989 	cr4_fixed1_update(X86_CR4_TSD,        edx, feature_bit(TSC));
6990 	cr4_fixed1_update(X86_CR4_DE,         edx, feature_bit(DE));
6991 	cr4_fixed1_update(X86_CR4_PSE,        edx, feature_bit(PSE));
6992 	cr4_fixed1_update(X86_CR4_PAE,        edx, feature_bit(PAE));
6993 	cr4_fixed1_update(X86_CR4_MCE,        edx, feature_bit(MCE));
6994 	cr4_fixed1_update(X86_CR4_PGE,        edx, feature_bit(PGE));
6995 	cr4_fixed1_update(X86_CR4_OSFXSR,     edx, feature_bit(FXSR));
6996 	cr4_fixed1_update(X86_CR4_OSXMMEXCPT, edx, feature_bit(XMM));
6997 	cr4_fixed1_update(X86_CR4_VMXE,       ecx, feature_bit(VMX));
6998 	cr4_fixed1_update(X86_CR4_SMXE,       ecx, feature_bit(SMX));
6999 	cr4_fixed1_update(X86_CR4_PCIDE,      ecx, feature_bit(PCID));
7000 	cr4_fixed1_update(X86_CR4_OSXSAVE,    ecx, feature_bit(XSAVE));
7001 
7002 	entry = kvm_find_cpuid_entry(vcpu, 0x7, 0);
7003 	cr4_fixed1_update(X86_CR4_FSGSBASE,   ebx, feature_bit(FSGSBASE));
7004 	cr4_fixed1_update(X86_CR4_SMEP,       ebx, feature_bit(SMEP));
7005 	cr4_fixed1_update(X86_CR4_SMAP,       ebx, feature_bit(SMAP));
7006 	cr4_fixed1_update(X86_CR4_PKE,        ecx, feature_bit(PKU));
7007 	cr4_fixed1_update(X86_CR4_UMIP,       ecx, feature_bit(UMIP));
7008 	cr4_fixed1_update(X86_CR4_LA57,       ecx, feature_bit(LA57));
7009 
7010 #undef cr4_fixed1_update
7011 }
7012 
7013 static void nested_vmx_entry_exit_ctls_update(struct kvm_vcpu *vcpu)
7014 {
7015 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7016 
7017 	if (kvm_mpx_supported()) {
7018 		bool mpx_enabled = guest_cpuid_has(vcpu, X86_FEATURE_MPX);
7019 
7020 		if (mpx_enabled) {
7021 			vmx->nested.msrs.entry_ctls_high |= VM_ENTRY_LOAD_BNDCFGS;
7022 			vmx->nested.msrs.exit_ctls_high |= VM_EXIT_CLEAR_BNDCFGS;
7023 		} else {
7024 			vmx->nested.msrs.entry_ctls_high &= ~VM_ENTRY_LOAD_BNDCFGS;
7025 			vmx->nested.msrs.exit_ctls_high &= ~VM_EXIT_CLEAR_BNDCFGS;
7026 		}
7027 	}
7028 }
7029 
7030 static void update_intel_pt_cfg(struct kvm_vcpu *vcpu)
7031 {
7032 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7033 	struct kvm_cpuid_entry2 *best = NULL;
7034 	int i;
7035 
7036 	for (i = 0; i < PT_CPUID_LEAVES; i++) {
7037 		best = kvm_find_cpuid_entry(vcpu, 0x14, i);
7038 		if (!best)
7039 			return;
7040 		vmx->pt_desc.caps[CPUID_EAX + i*PT_CPUID_REGS_NUM] = best->eax;
7041 		vmx->pt_desc.caps[CPUID_EBX + i*PT_CPUID_REGS_NUM] = best->ebx;
7042 		vmx->pt_desc.caps[CPUID_ECX + i*PT_CPUID_REGS_NUM] = best->ecx;
7043 		vmx->pt_desc.caps[CPUID_EDX + i*PT_CPUID_REGS_NUM] = best->edx;
7044 	}
7045 
7046 	/* Get the number of configurable Address Ranges for filtering */
7047 	vmx->pt_desc.addr_range = intel_pt_validate_cap(vmx->pt_desc.caps,
7048 						PT_CAP_num_address_ranges);
7049 
7050 	/* Initialize and clear the no dependency bits */
7051 	vmx->pt_desc.ctl_bitmask = ~(RTIT_CTL_TRACEEN | RTIT_CTL_OS |
7052 			RTIT_CTL_USR | RTIT_CTL_TSC_EN | RTIT_CTL_DISRETC);
7053 
7054 	/*
7055 	 * If CPUID.(EAX=14H,ECX=0):EBX[0]=1 CR3Filter can be set otherwise
7056 	 * will inject an #GP
7057 	 */
7058 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_cr3_filtering))
7059 		vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_CR3EN;
7060 
7061 	/*
7062 	 * If CPUID.(EAX=14H,ECX=0):EBX[1]=1 CYCEn, CycThresh and
7063 	 * PSBFreq can be set
7064 	 */
7065 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc))
7066 		vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_CYCLEACC |
7067 				RTIT_CTL_CYC_THRESH | RTIT_CTL_PSB_FREQ);
7068 
7069 	/*
7070 	 * If CPUID.(EAX=14H,ECX=0):EBX[3]=1 MTCEn BranchEn and
7071 	 * MTCFreq can be set
7072 	 */
7073 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc))
7074 		vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_MTC_EN |
7075 				RTIT_CTL_BRANCH_EN | RTIT_CTL_MTC_RANGE);
7076 
7077 	/* If CPUID.(EAX=14H,ECX=0):EBX[4]=1 FUPonPTW and PTWEn can be set */
7078 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_ptwrite))
7079 		vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_FUP_ON_PTW |
7080 							RTIT_CTL_PTW_EN);
7081 
7082 	/* If CPUID.(EAX=14H,ECX=0):EBX[5]=1 PwrEvEn can be set */
7083 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_power_event_trace))
7084 		vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_PWR_EVT_EN;
7085 
7086 	/* If CPUID.(EAX=14H,ECX=0):ECX[0]=1 ToPA can be set */
7087 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_topa_output))
7088 		vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_TOPA;
7089 
7090 	/* If CPUID.(EAX=14H,ECX=0):ECX[3]=1 FabircEn can be set */
7091 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_output_subsys))
7092 		vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_FABRIC_EN;
7093 
7094 	/* unmask address range configure area */
7095 	for (i = 0; i < vmx->pt_desc.addr_range; i++)
7096 		vmx->pt_desc.ctl_bitmask &= ~(0xfULL << (32 + i * 4));
7097 }
7098 
7099 static void vmx_cpuid_update(struct kvm_vcpu *vcpu)
7100 {
7101 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7102 
7103 	/* xsaves_enabled is recomputed in vmx_compute_secondary_exec_control(). */
7104 	vcpu->arch.xsaves_enabled = false;
7105 
7106 	if (cpu_has_secondary_exec_ctrls()) {
7107 		vmx_compute_secondary_exec_control(vmx);
7108 		vmcs_set_secondary_exec_control(vmx);
7109 	}
7110 
7111 	if (nested_vmx_allowed(vcpu))
7112 		to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
7113 			FEAT_CTL_VMX_ENABLED_INSIDE_SMX |
7114 			FEAT_CTL_VMX_ENABLED_OUTSIDE_SMX;
7115 	else
7116 		to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
7117 			~(FEAT_CTL_VMX_ENABLED_INSIDE_SMX |
7118 			  FEAT_CTL_VMX_ENABLED_OUTSIDE_SMX);
7119 
7120 	if (nested_vmx_allowed(vcpu)) {
7121 		nested_vmx_cr_fixed1_bits_update(vcpu);
7122 		nested_vmx_entry_exit_ctls_update(vcpu);
7123 	}
7124 
7125 	if (boot_cpu_has(X86_FEATURE_INTEL_PT) &&
7126 			guest_cpuid_has(vcpu, X86_FEATURE_INTEL_PT))
7127 		update_intel_pt_cfg(vcpu);
7128 
7129 	if (boot_cpu_has(X86_FEATURE_RTM)) {
7130 		struct shared_msr_entry *msr;
7131 		msr = find_msr_entry(vmx, MSR_IA32_TSX_CTRL);
7132 		if (msr) {
7133 			bool enabled = guest_cpuid_has(vcpu, X86_FEATURE_RTM);
7134 			vmx_set_guest_msr(vmx, msr, enabled ? 0 : TSX_CTRL_RTM_DISABLE);
7135 		}
7136 	}
7137 }
7138 
7139 static void vmx_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
7140 {
7141 	if (func == 1 && nested)
7142 		entry->ecx |= feature_bit(VMX);
7143 }
7144 
7145 static void vmx_request_immediate_exit(struct kvm_vcpu *vcpu)
7146 {
7147 	to_vmx(vcpu)->req_immediate_exit = true;
7148 }
7149 
7150 static int vmx_check_intercept_io(struct kvm_vcpu *vcpu,
7151 				  struct x86_instruction_info *info)
7152 {
7153 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
7154 	unsigned short port;
7155 	bool intercept;
7156 	int size;
7157 
7158 	if (info->intercept == x86_intercept_in ||
7159 	    info->intercept == x86_intercept_ins) {
7160 		port = info->src_val;
7161 		size = info->dst_bytes;
7162 	} else {
7163 		port = info->dst_val;
7164 		size = info->src_bytes;
7165 	}
7166 
7167 	/*
7168 	 * If the 'use IO bitmaps' VM-execution control is 0, IO instruction
7169 	 * VM-exits depend on the 'unconditional IO exiting' VM-execution
7170 	 * control.
7171 	 *
7172 	 * Otherwise, IO instruction VM-exits are controlled by the IO bitmaps.
7173 	 */
7174 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
7175 		intercept = nested_cpu_has(vmcs12,
7176 					   CPU_BASED_UNCOND_IO_EXITING);
7177 	else
7178 		intercept = nested_vmx_check_io_bitmaps(vcpu, port, size);
7179 
7180 	/* FIXME: produce nested vmexit and return X86EMUL_INTERCEPTED.  */
7181 	return intercept ? X86EMUL_UNHANDLEABLE : X86EMUL_CONTINUE;
7182 }
7183 
7184 static int vmx_check_intercept(struct kvm_vcpu *vcpu,
7185 			       struct x86_instruction_info *info,
7186 			       enum x86_intercept_stage stage)
7187 {
7188 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
7189 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
7190 
7191 	switch (info->intercept) {
7192 	/*
7193 	 * RDPID causes #UD if disabled through secondary execution controls.
7194 	 * Because it is marked as EmulateOnUD, we need to intercept it here.
7195 	 */
7196 	case x86_intercept_rdtscp:
7197 		if (!nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDTSCP)) {
7198 			ctxt->exception.vector = UD_VECTOR;
7199 			ctxt->exception.error_code_valid = false;
7200 			return X86EMUL_PROPAGATE_FAULT;
7201 		}
7202 		break;
7203 
7204 	case x86_intercept_in:
7205 	case x86_intercept_ins:
7206 	case x86_intercept_out:
7207 	case x86_intercept_outs:
7208 		return vmx_check_intercept_io(vcpu, info);
7209 
7210 	case x86_intercept_lgdt:
7211 	case x86_intercept_lidt:
7212 	case x86_intercept_lldt:
7213 	case x86_intercept_ltr:
7214 	case x86_intercept_sgdt:
7215 	case x86_intercept_sidt:
7216 	case x86_intercept_sldt:
7217 	case x86_intercept_str:
7218 		if (!nested_cpu_has2(vmcs12, SECONDARY_EXEC_DESC))
7219 			return X86EMUL_CONTINUE;
7220 
7221 		/* FIXME: produce nested vmexit and return X86EMUL_INTERCEPTED.  */
7222 		break;
7223 
7224 	/* TODO: check more intercepts... */
7225 	default:
7226 		break;
7227 	}
7228 
7229 	return X86EMUL_UNHANDLEABLE;
7230 }
7231 
7232 #ifdef CONFIG_X86_64
7233 /* (a << shift) / divisor, return 1 if overflow otherwise 0 */
7234 static inline int u64_shl_div_u64(u64 a, unsigned int shift,
7235 				  u64 divisor, u64 *result)
7236 {
7237 	u64 low = a << shift, high = a >> (64 - shift);
7238 
7239 	/* To avoid the overflow on divq */
7240 	if (high >= divisor)
7241 		return 1;
7242 
7243 	/* Low hold the result, high hold rem which is discarded */
7244 	asm("divq %2\n\t" : "=a" (low), "=d" (high) :
7245 	    "rm" (divisor), "0" (low), "1" (high));
7246 	*result = low;
7247 
7248 	return 0;
7249 }
7250 
7251 static int vmx_set_hv_timer(struct kvm_vcpu *vcpu, u64 guest_deadline_tsc,
7252 			    bool *expired)
7253 {
7254 	struct vcpu_vmx *vmx;
7255 	u64 tscl, guest_tscl, delta_tsc, lapic_timer_advance_cycles;
7256 	struct kvm_timer *ktimer = &vcpu->arch.apic->lapic_timer;
7257 
7258 	if (kvm_mwait_in_guest(vcpu->kvm) ||
7259 		kvm_can_post_timer_interrupt(vcpu))
7260 		return -EOPNOTSUPP;
7261 
7262 	vmx = to_vmx(vcpu);
7263 	tscl = rdtsc();
7264 	guest_tscl = kvm_read_l1_tsc(vcpu, tscl);
7265 	delta_tsc = max(guest_deadline_tsc, guest_tscl) - guest_tscl;
7266 	lapic_timer_advance_cycles = nsec_to_cycles(vcpu,
7267 						    ktimer->timer_advance_ns);
7268 
7269 	if (delta_tsc > lapic_timer_advance_cycles)
7270 		delta_tsc -= lapic_timer_advance_cycles;
7271 	else
7272 		delta_tsc = 0;
7273 
7274 	/* Convert to host delta tsc if tsc scaling is enabled */
7275 	if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio &&
7276 	    delta_tsc && u64_shl_div_u64(delta_tsc,
7277 				kvm_tsc_scaling_ratio_frac_bits,
7278 				vcpu->arch.tsc_scaling_ratio, &delta_tsc))
7279 		return -ERANGE;
7280 
7281 	/*
7282 	 * If the delta tsc can't fit in the 32 bit after the multi shift,
7283 	 * we can't use the preemption timer.
7284 	 * It's possible that it fits on later vmentries, but checking
7285 	 * on every vmentry is costly so we just use an hrtimer.
7286 	 */
7287 	if (delta_tsc >> (cpu_preemption_timer_multi + 32))
7288 		return -ERANGE;
7289 
7290 	vmx->hv_deadline_tsc = tscl + delta_tsc;
7291 	*expired = !delta_tsc;
7292 	return 0;
7293 }
7294 
7295 static void vmx_cancel_hv_timer(struct kvm_vcpu *vcpu)
7296 {
7297 	to_vmx(vcpu)->hv_deadline_tsc = -1;
7298 }
7299 #endif
7300 
7301 static void vmx_sched_in(struct kvm_vcpu *vcpu, int cpu)
7302 {
7303 	if (!kvm_pause_in_guest(vcpu->kvm))
7304 		shrink_ple_window(vcpu);
7305 }
7306 
7307 static void vmx_slot_enable_log_dirty(struct kvm *kvm,
7308 				     struct kvm_memory_slot *slot)
7309 {
7310 	kvm_mmu_slot_leaf_clear_dirty(kvm, slot);
7311 	kvm_mmu_slot_largepage_remove_write_access(kvm, slot);
7312 }
7313 
7314 static void vmx_slot_disable_log_dirty(struct kvm *kvm,
7315 				       struct kvm_memory_slot *slot)
7316 {
7317 	kvm_mmu_slot_set_dirty(kvm, slot);
7318 }
7319 
7320 static void vmx_flush_log_dirty(struct kvm *kvm)
7321 {
7322 	kvm_flush_pml_buffers(kvm);
7323 }
7324 
7325 static int vmx_write_pml_buffer(struct kvm_vcpu *vcpu)
7326 {
7327 	struct vmcs12 *vmcs12;
7328 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7329 	gpa_t gpa, dst;
7330 
7331 	if (is_guest_mode(vcpu)) {
7332 		WARN_ON_ONCE(vmx->nested.pml_full);
7333 
7334 		/*
7335 		 * Check if PML is enabled for the nested guest.
7336 		 * Whether eptp bit 6 is set is already checked
7337 		 * as part of A/D emulation.
7338 		 */
7339 		vmcs12 = get_vmcs12(vcpu);
7340 		if (!nested_cpu_has_pml(vmcs12))
7341 			return 0;
7342 
7343 		if (vmcs12->guest_pml_index >= PML_ENTITY_NUM) {
7344 			vmx->nested.pml_full = true;
7345 			return 1;
7346 		}
7347 
7348 		gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS) & ~0xFFFull;
7349 		dst = vmcs12->pml_address + sizeof(u64) * vmcs12->guest_pml_index;
7350 
7351 		if (kvm_write_guest_page(vcpu->kvm, gpa_to_gfn(dst), &gpa,
7352 					 offset_in_page(dst), sizeof(gpa)))
7353 			return 0;
7354 
7355 		vmcs12->guest_pml_index--;
7356 	}
7357 
7358 	return 0;
7359 }
7360 
7361 static void vmx_enable_log_dirty_pt_masked(struct kvm *kvm,
7362 					   struct kvm_memory_slot *memslot,
7363 					   gfn_t offset, unsigned long mask)
7364 {
7365 	kvm_mmu_clear_dirty_pt_masked(kvm, memslot, offset, mask);
7366 }
7367 
7368 static void __pi_post_block(struct kvm_vcpu *vcpu)
7369 {
7370 	struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
7371 	struct pi_desc old, new;
7372 	unsigned int dest;
7373 
7374 	do {
7375 		old.control = new.control = pi_desc->control;
7376 		WARN(old.nv != POSTED_INTR_WAKEUP_VECTOR,
7377 		     "Wakeup handler not enabled while the VCPU is blocked\n");
7378 
7379 		dest = cpu_physical_id(vcpu->cpu);
7380 
7381 		if (x2apic_enabled())
7382 			new.ndst = dest;
7383 		else
7384 			new.ndst = (dest << 8) & 0xFF00;
7385 
7386 		/* set 'NV' to 'notification vector' */
7387 		new.nv = POSTED_INTR_VECTOR;
7388 	} while (cmpxchg64(&pi_desc->control, old.control,
7389 			   new.control) != old.control);
7390 
7391 	if (!WARN_ON_ONCE(vcpu->pre_pcpu == -1)) {
7392 		spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7393 		list_del(&vcpu->blocked_vcpu_list);
7394 		spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7395 		vcpu->pre_pcpu = -1;
7396 	}
7397 }
7398 
7399 /*
7400  * This routine does the following things for vCPU which is going
7401  * to be blocked if VT-d PI is enabled.
7402  * - Store the vCPU to the wakeup list, so when interrupts happen
7403  *   we can find the right vCPU to wake up.
7404  * - Change the Posted-interrupt descriptor as below:
7405  *      'NDST' <-- vcpu->pre_pcpu
7406  *      'NV' <-- POSTED_INTR_WAKEUP_VECTOR
7407  * - If 'ON' is set during this process, which means at least one
7408  *   interrupt is posted for this vCPU, we cannot block it, in
7409  *   this case, return 1, otherwise, return 0.
7410  *
7411  */
7412 static int pi_pre_block(struct kvm_vcpu *vcpu)
7413 {
7414 	unsigned int dest;
7415 	struct pi_desc old, new;
7416 	struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
7417 
7418 	if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
7419 		!irq_remapping_cap(IRQ_POSTING_CAP)  ||
7420 		!kvm_vcpu_apicv_active(vcpu))
7421 		return 0;
7422 
7423 	WARN_ON(irqs_disabled());
7424 	local_irq_disable();
7425 	if (!WARN_ON_ONCE(vcpu->pre_pcpu != -1)) {
7426 		vcpu->pre_pcpu = vcpu->cpu;
7427 		spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7428 		list_add_tail(&vcpu->blocked_vcpu_list,
7429 			      &per_cpu(blocked_vcpu_on_cpu,
7430 				       vcpu->pre_pcpu));
7431 		spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7432 	}
7433 
7434 	do {
7435 		old.control = new.control = pi_desc->control;
7436 
7437 		WARN((pi_desc->sn == 1),
7438 		     "Warning: SN field of posted-interrupts "
7439 		     "is set before blocking\n");
7440 
7441 		/*
7442 		 * Since vCPU can be preempted during this process,
7443 		 * vcpu->cpu could be different with pre_pcpu, we
7444 		 * need to set pre_pcpu as the destination of wakeup
7445 		 * notification event, then we can find the right vCPU
7446 		 * to wakeup in wakeup handler if interrupts happen
7447 		 * when the vCPU is in blocked state.
7448 		 */
7449 		dest = cpu_physical_id(vcpu->pre_pcpu);
7450 
7451 		if (x2apic_enabled())
7452 			new.ndst = dest;
7453 		else
7454 			new.ndst = (dest << 8) & 0xFF00;
7455 
7456 		/* set 'NV' to 'wakeup vector' */
7457 		new.nv = POSTED_INTR_WAKEUP_VECTOR;
7458 	} while (cmpxchg64(&pi_desc->control, old.control,
7459 			   new.control) != old.control);
7460 
7461 	/* We should not block the vCPU if an interrupt is posted for it.  */
7462 	if (pi_test_on(pi_desc) == 1)
7463 		__pi_post_block(vcpu);
7464 
7465 	local_irq_enable();
7466 	return (vcpu->pre_pcpu == -1);
7467 }
7468 
7469 static int vmx_pre_block(struct kvm_vcpu *vcpu)
7470 {
7471 	if (pi_pre_block(vcpu))
7472 		return 1;
7473 
7474 	if (kvm_lapic_hv_timer_in_use(vcpu))
7475 		kvm_lapic_switch_to_sw_timer(vcpu);
7476 
7477 	return 0;
7478 }
7479 
7480 static void pi_post_block(struct kvm_vcpu *vcpu)
7481 {
7482 	if (vcpu->pre_pcpu == -1)
7483 		return;
7484 
7485 	WARN_ON(irqs_disabled());
7486 	local_irq_disable();
7487 	__pi_post_block(vcpu);
7488 	local_irq_enable();
7489 }
7490 
7491 static void vmx_post_block(struct kvm_vcpu *vcpu)
7492 {
7493 	if (kvm_x86_ops->set_hv_timer)
7494 		kvm_lapic_switch_to_hv_timer(vcpu);
7495 
7496 	pi_post_block(vcpu);
7497 }
7498 
7499 /*
7500  * vmx_update_pi_irte - set IRTE for Posted-Interrupts
7501  *
7502  * @kvm: kvm
7503  * @host_irq: host irq of the interrupt
7504  * @guest_irq: gsi of the interrupt
7505  * @set: set or unset PI
7506  * returns 0 on success, < 0 on failure
7507  */
7508 static int vmx_update_pi_irte(struct kvm *kvm, unsigned int host_irq,
7509 			      uint32_t guest_irq, bool set)
7510 {
7511 	struct kvm_kernel_irq_routing_entry *e;
7512 	struct kvm_irq_routing_table *irq_rt;
7513 	struct kvm_lapic_irq irq;
7514 	struct kvm_vcpu *vcpu;
7515 	struct vcpu_data vcpu_info;
7516 	int idx, ret = 0;
7517 
7518 	if (!kvm_arch_has_assigned_device(kvm) ||
7519 		!irq_remapping_cap(IRQ_POSTING_CAP) ||
7520 		!kvm_vcpu_apicv_active(kvm->vcpus[0]))
7521 		return 0;
7522 
7523 	idx = srcu_read_lock(&kvm->irq_srcu);
7524 	irq_rt = srcu_dereference(kvm->irq_routing, &kvm->irq_srcu);
7525 	if (guest_irq >= irq_rt->nr_rt_entries ||
7526 	    hlist_empty(&irq_rt->map[guest_irq])) {
7527 		pr_warn_once("no route for guest_irq %u/%u (broken user space?)\n",
7528 			     guest_irq, irq_rt->nr_rt_entries);
7529 		goto out;
7530 	}
7531 
7532 	hlist_for_each_entry(e, &irq_rt->map[guest_irq], link) {
7533 		if (e->type != KVM_IRQ_ROUTING_MSI)
7534 			continue;
7535 		/*
7536 		 * VT-d PI cannot support posting multicast/broadcast
7537 		 * interrupts to a vCPU, we still use interrupt remapping
7538 		 * for these kind of interrupts.
7539 		 *
7540 		 * For lowest-priority interrupts, we only support
7541 		 * those with single CPU as the destination, e.g. user
7542 		 * configures the interrupts via /proc/irq or uses
7543 		 * irqbalance to make the interrupts single-CPU.
7544 		 *
7545 		 * We will support full lowest-priority interrupt later.
7546 		 *
7547 		 * In addition, we can only inject generic interrupts using
7548 		 * the PI mechanism, refuse to route others through it.
7549 		 */
7550 
7551 		kvm_set_msi_irq(kvm, e, &irq);
7552 		if (!kvm_intr_is_single_vcpu(kvm, &irq, &vcpu) ||
7553 		    !kvm_irq_is_postable(&irq)) {
7554 			/*
7555 			 * Make sure the IRTE is in remapped mode if
7556 			 * we don't handle it in posted mode.
7557 			 */
7558 			ret = irq_set_vcpu_affinity(host_irq, NULL);
7559 			if (ret < 0) {
7560 				printk(KERN_INFO
7561 				   "failed to back to remapped mode, irq: %u\n",
7562 				   host_irq);
7563 				goto out;
7564 			}
7565 
7566 			continue;
7567 		}
7568 
7569 		vcpu_info.pi_desc_addr = __pa(vcpu_to_pi_desc(vcpu));
7570 		vcpu_info.vector = irq.vector;
7571 
7572 		trace_kvm_pi_irte_update(host_irq, vcpu->vcpu_id, e->gsi,
7573 				vcpu_info.vector, vcpu_info.pi_desc_addr, set);
7574 
7575 		if (set)
7576 			ret = irq_set_vcpu_affinity(host_irq, &vcpu_info);
7577 		else
7578 			ret = irq_set_vcpu_affinity(host_irq, NULL);
7579 
7580 		if (ret < 0) {
7581 			printk(KERN_INFO "%s: failed to update PI IRTE\n",
7582 					__func__);
7583 			goto out;
7584 		}
7585 	}
7586 
7587 	ret = 0;
7588 out:
7589 	srcu_read_unlock(&kvm->irq_srcu, idx);
7590 	return ret;
7591 }
7592 
7593 static void vmx_setup_mce(struct kvm_vcpu *vcpu)
7594 {
7595 	if (vcpu->arch.mcg_cap & MCG_LMCE_P)
7596 		to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
7597 			FEAT_CTL_LMCE_ENABLED;
7598 	else
7599 		to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
7600 			~FEAT_CTL_LMCE_ENABLED;
7601 }
7602 
7603 static int vmx_smi_allowed(struct kvm_vcpu *vcpu)
7604 {
7605 	/* we need a nested vmexit to enter SMM, postpone if run is pending */
7606 	if (to_vmx(vcpu)->nested.nested_run_pending)
7607 		return 0;
7608 	return 1;
7609 }
7610 
7611 static int vmx_pre_enter_smm(struct kvm_vcpu *vcpu, char *smstate)
7612 {
7613 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7614 
7615 	vmx->nested.smm.guest_mode = is_guest_mode(vcpu);
7616 	if (vmx->nested.smm.guest_mode)
7617 		nested_vmx_vmexit(vcpu, -1, 0, 0);
7618 
7619 	vmx->nested.smm.vmxon = vmx->nested.vmxon;
7620 	vmx->nested.vmxon = false;
7621 	vmx_clear_hlt(vcpu);
7622 	return 0;
7623 }
7624 
7625 static int vmx_pre_leave_smm(struct kvm_vcpu *vcpu, const char *smstate)
7626 {
7627 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7628 	int ret;
7629 
7630 	if (vmx->nested.smm.vmxon) {
7631 		vmx->nested.vmxon = true;
7632 		vmx->nested.smm.vmxon = false;
7633 	}
7634 
7635 	if (vmx->nested.smm.guest_mode) {
7636 		ret = nested_vmx_enter_non_root_mode(vcpu, false);
7637 		if (ret)
7638 			return ret;
7639 
7640 		vmx->nested.smm.guest_mode = false;
7641 	}
7642 	return 0;
7643 }
7644 
7645 static int enable_smi_window(struct kvm_vcpu *vcpu)
7646 {
7647 	return 0;
7648 }
7649 
7650 static bool vmx_need_emulation_on_page_fault(struct kvm_vcpu *vcpu)
7651 {
7652 	return false;
7653 }
7654 
7655 static bool vmx_apic_init_signal_blocked(struct kvm_vcpu *vcpu)
7656 {
7657 	return to_vmx(vcpu)->nested.vmxon;
7658 }
7659 
7660 static __init int hardware_setup(void)
7661 {
7662 	unsigned long host_bndcfgs;
7663 	struct desc_ptr dt;
7664 	int r, i;
7665 
7666 	rdmsrl_safe(MSR_EFER, &host_efer);
7667 
7668 	store_idt(&dt);
7669 	host_idt_base = dt.address;
7670 
7671 	for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i)
7672 		kvm_define_shared_msr(i, vmx_msr_index[i]);
7673 
7674 	if (setup_vmcs_config(&vmcs_config, &vmx_capability) < 0)
7675 		return -EIO;
7676 
7677 	if (boot_cpu_has(X86_FEATURE_NX))
7678 		kvm_enable_efer_bits(EFER_NX);
7679 
7680 	if (boot_cpu_has(X86_FEATURE_MPX)) {
7681 		rdmsrl(MSR_IA32_BNDCFGS, host_bndcfgs);
7682 		WARN_ONCE(host_bndcfgs, "KVM: BNDCFGS in host will be lost");
7683 	}
7684 
7685 	if (!cpu_has_vmx_vpid() || !cpu_has_vmx_invvpid() ||
7686 	    !(cpu_has_vmx_invvpid_single() || cpu_has_vmx_invvpid_global()))
7687 		enable_vpid = 0;
7688 
7689 	if (!cpu_has_vmx_ept() ||
7690 	    !cpu_has_vmx_ept_4levels() ||
7691 	    !cpu_has_vmx_ept_mt_wb() ||
7692 	    !cpu_has_vmx_invept_global())
7693 		enable_ept = 0;
7694 
7695 	if (!cpu_has_vmx_ept_ad_bits() || !enable_ept)
7696 		enable_ept_ad_bits = 0;
7697 
7698 	if (!cpu_has_vmx_unrestricted_guest() || !enable_ept)
7699 		enable_unrestricted_guest = 0;
7700 
7701 	if (!cpu_has_vmx_flexpriority())
7702 		flexpriority_enabled = 0;
7703 
7704 	if (!cpu_has_virtual_nmis())
7705 		enable_vnmi = 0;
7706 
7707 	/*
7708 	 * set_apic_access_page_addr() is used to reload apic access
7709 	 * page upon invalidation.  No need to do anything if not
7710 	 * using the APIC_ACCESS_ADDR VMCS field.
7711 	 */
7712 	if (!flexpriority_enabled)
7713 		kvm_x86_ops->set_apic_access_page_addr = NULL;
7714 
7715 	if (!cpu_has_vmx_tpr_shadow())
7716 		kvm_x86_ops->update_cr8_intercept = NULL;
7717 
7718 	if (enable_ept && !cpu_has_vmx_ept_2m_page())
7719 		kvm_disable_largepages();
7720 
7721 #if IS_ENABLED(CONFIG_HYPERV)
7722 	if (ms_hyperv.nested_features & HV_X64_NESTED_GUEST_MAPPING_FLUSH
7723 	    && enable_ept) {
7724 		kvm_x86_ops->tlb_remote_flush = hv_remote_flush_tlb;
7725 		kvm_x86_ops->tlb_remote_flush_with_range =
7726 				hv_remote_flush_tlb_with_range;
7727 	}
7728 #endif
7729 
7730 	if (!cpu_has_vmx_ple()) {
7731 		ple_gap = 0;
7732 		ple_window = 0;
7733 		ple_window_grow = 0;
7734 		ple_window_max = 0;
7735 		ple_window_shrink = 0;
7736 	}
7737 
7738 	if (!cpu_has_vmx_apicv()) {
7739 		enable_apicv = 0;
7740 		kvm_x86_ops->sync_pir_to_irr = NULL;
7741 	}
7742 
7743 	if (cpu_has_vmx_tsc_scaling()) {
7744 		kvm_has_tsc_control = true;
7745 		kvm_max_tsc_scaling_ratio = KVM_VMX_TSC_MULTIPLIER_MAX;
7746 		kvm_tsc_scaling_ratio_frac_bits = 48;
7747 	}
7748 
7749 	set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
7750 
7751 	if (enable_ept)
7752 		vmx_enable_tdp();
7753 	else
7754 		kvm_disable_tdp();
7755 
7756 	/*
7757 	 * Only enable PML when hardware supports PML feature, and both EPT
7758 	 * and EPT A/D bit features are enabled -- PML depends on them to work.
7759 	 */
7760 	if (!enable_ept || !enable_ept_ad_bits || !cpu_has_vmx_pml())
7761 		enable_pml = 0;
7762 
7763 	if (!enable_pml) {
7764 		kvm_x86_ops->slot_enable_log_dirty = NULL;
7765 		kvm_x86_ops->slot_disable_log_dirty = NULL;
7766 		kvm_x86_ops->flush_log_dirty = NULL;
7767 		kvm_x86_ops->enable_log_dirty_pt_masked = NULL;
7768 	}
7769 
7770 	if (!cpu_has_vmx_preemption_timer())
7771 		enable_preemption_timer = false;
7772 
7773 	if (enable_preemption_timer) {
7774 		u64 use_timer_freq = 5000ULL * 1000 * 1000;
7775 		u64 vmx_msr;
7776 
7777 		rdmsrl(MSR_IA32_VMX_MISC, vmx_msr);
7778 		cpu_preemption_timer_multi =
7779 			vmx_msr & VMX_MISC_PREEMPTION_TIMER_RATE_MASK;
7780 
7781 		if (tsc_khz)
7782 			use_timer_freq = (u64)tsc_khz * 1000;
7783 		use_timer_freq >>= cpu_preemption_timer_multi;
7784 
7785 		/*
7786 		 * KVM "disables" the preemption timer by setting it to its max
7787 		 * value.  Don't use the timer if it might cause spurious exits
7788 		 * at a rate faster than 0.1 Hz (of uninterrupted guest time).
7789 		 */
7790 		if (use_timer_freq > 0xffffffffu / 10)
7791 			enable_preemption_timer = false;
7792 	}
7793 
7794 	if (!enable_preemption_timer) {
7795 		kvm_x86_ops->set_hv_timer = NULL;
7796 		kvm_x86_ops->cancel_hv_timer = NULL;
7797 		kvm_x86_ops->request_immediate_exit = __kvm_request_immediate_exit;
7798 	}
7799 
7800 	kvm_set_posted_intr_wakeup_handler(wakeup_handler);
7801 
7802 	kvm_mce_cap_supported |= MCG_LMCE_P;
7803 
7804 	if (pt_mode != PT_MODE_SYSTEM && pt_mode != PT_MODE_HOST_GUEST)
7805 		return -EINVAL;
7806 	if (!enable_ept || !cpu_has_vmx_intel_pt())
7807 		pt_mode = PT_MODE_SYSTEM;
7808 
7809 	if (nested) {
7810 		nested_vmx_setup_ctls_msrs(&vmcs_config.nested,
7811 					   vmx_capability.ept);
7812 
7813 		r = nested_vmx_hardware_setup(kvm_vmx_exit_handlers);
7814 		if (r)
7815 			return r;
7816 	}
7817 
7818 	r = alloc_kvm_area();
7819 	if (r)
7820 		nested_vmx_hardware_unsetup();
7821 	return r;
7822 }
7823 
7824 static __exit void hardware_unsetup(void)
7825 {
7826 	if (nested)
7827 		nested_vmx_hardware_unsetup();
7828 
7829 	free_kvm_area();
7830 }
7831 
7832 static bool vmx_check_apicv_inhibit_reasons(ulong bit)
7833 {
7834 	ulong supported = BIT(APICV_INHIBIT_REASON_DISABLE) |
7835 			  BIT(APICV_INHIBIT_REASON_HYPERV);
7836 
7837 	return supported & BIT(bit);
7838 }
7839 
7840 static struct kvm_x86_ops vmx_x86_ops __ro_after_init = {
7841 	.cpu_has_kvm_support = cpu_has_kvm_support,
7842 	.disabled_by_bios = vmx_disabled_by_bios,
7843 	.hardware_setup = hardware_setup,
7844 	.hardware_unsetup = hardware_unsetup,
7845 	.check_processor_compatibility = vmx_check_processor_compat,
7846 	.hardware_enable = hardware_enable,
7847 	.hardware_disable = hardware_disable,
7848 	.cpu_has_accelerated_tpr = report_flexpriority,
7849 	.has_emulated_msr = vmx_has_emulated_msr,
7850 
7851 	.vm_init = vmx_vm_init,
7852 	.vm_alloc = vmx_vm_alloc,
7853 	.vm_free = vmx_vm_free,
7854 
7855 	.vcpu_create = vmx_create_vcpu,
7856 	.vcpu_free = vmx_free_vcpu,
7857 	.vcpu_reset = vmx_vcpu_reset,
7858 
7859 	.prepare_guest_switch = vmx_prepare_switch_to_guest,
7860 	.vcpu_load = vmx_vcpu_load,
7861 	.vcpu_put = vmx_vcpu_put,
7862 
7863 	.update_bp_intercept = update_exception_bitmap,
7864 	.get_msr_feature = vmx_get_msr_feature,
7865 	.get_msr = vmx_get_msr,
7866 	.set_msr = vmx_set_msr,
7867 	.get_segment_base = vmx_get_segment_base,
7868 	.get_segment = vmx_get_segment,
7869 	.set_segment = vmx_set_segment,
7870 	.get_cpl = vmx_get_cpl,
7871 	.get_cs_db_l_bits = vmx_get_cs_db_l_bits,
7872 	.decache_cr0_guest_bits = vmx_decache_cr0_guest_bits,
7873 	.decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
7874 	.set_cr0 = vmx_set_cr0,
7875 	.set_cr3 = vmx_set_cr3,
7876 	.set_cr4 = vmx_set_cr4,
7877 	.set_efer = vmx_set_efer,
7878 	.get_idt = vmx_get_idt,
7879 	.set_idt = vmx_set_idt,
7880 	.get_gdt = vmx_get_gdt,
7881 	.set_gdt = vmx_set_gdt,
7882 	.get_dr6 = vmx_get_dr6,
7883 	.set_dr6 = vmx_set_dr6,
7884 	.set_dr7 = vmx_set_dr7,
7885 	.sync_dirty_debug_regs = vmx_sync_dirty_debug_regs,
7886 	.cache_reg = vmx_cache_reg,
7887 	.get_rflags = vmx_get_rflags,
7888 	.set_rflags = vmx_set_rflags,
7889 
7890 	.tlb_flush = vmx_flush_tlb,
7891 	.tlb_flush_gva = vmx_flush_tlb_gva,
7892 
7893 	.run = vmx_vcpu_run,
7894 	.handle_exit = vmx_handle_exit,
7895 	.skip_emulated_instruction = vmx_skip_emulated_instruction,
7896 	.update_emulated_instruction = vmx_update_emulated_instruction,
7897 	.set_interrupt_shadow = vmx_set_interrupt_shadow,
7898 	.get_interrupt_shadow = vmx_get_interrupt_shadow,
7899 	.patch_hypercall = vmx_patch_hypercall,
7900 	.set_irq = vmx_inject_irq,
7901 	.set_nmi = vmx_inject_nmi,
7902 	.queue_exception = vmx_queue_exception,
7903 	.cancel_injection = vmx_cancel_injection,
7904 	.interrupt_allowed = vmx_interrupt_allowed,
7905 	.nmi_allowed = vmx_nmi_allowed,
7906 	.get_nmi_mask = vmx_get_nmi_mask,
7907 	.set_nmi_mask = vmx_set_nmi_mask,
7908 	.enable_nmi_window = enable_nmi_window,
7909 	.enable_irq_window = enable_irq_window,
7910 	.update_cr8_intercept = update_cr8_intercept,
7911 	.set_virtual_apic_mode = vmx_set_virtual_apic_mode,
7912 	.set_apic_access_page_addr = vmx_set_apic_access_page_addr,
7913 	.refresh_apicv_exec_ctrl = vmx_refresh_apicv_exec_ctrl,
7914 	.load_eoi_exitmap = vmx_load_eoi_exitmap,
7915 	.apicv_post_state_restore = vmx_apicv_post_state_restore,
7916 	.check_apicv_inhibit_reasons = vmx_check_apicv_inhibit_reasons,
7917 	.hwapic_irr_update = vmx_hwapic_irr_update,
7918 	.hwapic_isr_update = vmx_hwapic_isr_update,
7919 	.guest_apic_has_interrupt = vmx_guest_apic_has_interrupt,
7920 	.sync_pir_to_irr = vmx_sync_pir_to_irr,
7921 	.deliver_posted_interrupt = vmx_deliver_posted_interrupt,
7922 	.dy_apicv_has_pending_interrupt = vmx_dy_apicv_has_pending_interrupt,
7923 
7924 	.set_tss_addr = vmx_set_tss_addr,
7925 	.set_identity_map_addr = vmx_set_identity_map_addr,
7926 	.get_tdp_level = get_ept_level,
7927 	.get_mt_mask = vmx_get_mt_mask,
7928 
7929 	.get_exit_info = vmx_get_exit_info,
7930 
7931 	.get_lpage_level = vmx_get_lpage_level,
7932 
7933 	.cpuid_update = vmx_cpuid_update,
7934 
7935 	.rdtscp_supported = vmx_rdtscp_supported,
7936 	.invpcid_supported = vmx_invpcid_supported,
7937 
7938 	.set_supported_cpuid = vmx_set_supported_cpuid,
7939 
7940 	.has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
7941 
7942 	.read_l1_tsc_offset = vmx_read_l1_tsc_offset,
7943 	.write_l1_tsc_offset = vmx_write_l1_tsc_offset,
7944 
7945 	.set_tdp_cr3 = vmx_set_cr3,
7946 
7947 	.check_intercept = vmx_check_intercept,
7948 	.handle_exit_irqoff = vmx_handle_exit_irqoff,
7949 	.mpx_supported = vmx_mpx_supported,
7950 	.xsaves_supported = vmx_xsaves_supported,
7951 	.umip_emulated = vmx_umip_emulated,
7952 	.pt_supported = vmx_pt_supported,
7953 	.pku_supported = vmx_pku_supported,
7954 
7955 	.request_immediate_exit = vmx_request_immediate_exit,
7956 
7957 	.sched_in = vmx_sched_in,
7958 
7959 	.slot_enable_log_dirty = vmx_slot_enable_log_dirty,
7960 	.slot_disable_log_dirty = vmx_slot_disable_log_dirty,
7961 	.flush_log_dirty = vmx_flush_log_dirty,
7962 	.enable_log_dirty_pt_masked = vmx_enable_log_dirty_pt_masked,
7963 	.write_log_dirty = vmx_write_pml_buffer,
7964 
7965 	.pre_block = vmx_pre_block,
7966 	.post_block = vmx_post_block,
7967 
7968 	.pmu_ops = &intel_pmu_ops,
7969 
7970 	.update_pi_irte = vmx_update_pi_irte,
7971 
7972 #ifdef CONFIG_X86_64
7973 	.set_hv_timer = vmx_set_hv_timer,
7974 	.cancel_hv_timer = vmx_cancel_hv_timer,
7975 #endif
7976 
7977 	.setup_mce = vmx_setup_mce,
7978 
7979 	.smi_allowed = vmx_smi_allowed,
7980 	.pre_enter_smm = vmx_pre_enter_smm,
7981 	.pre_leave_smm = vmx_pre_leave_smm,
7982 	.enable_smi_window = enable_smi_window,
7983 
7984 	.check_nested_events = NULL,
7985 	.get_nested_state = NULL,
7986 	.set_nested_state = NULL,
7987 	.get_vmcs12_pages = NULL,
7988 	.nested_enable_evmcs = NULL,
7989 	.nested_get_evmcs_version = NULL,
7990 	.need_emulation_on_page_fault = vmx_need_emulation_on_page_fault,
7991 	.apic_init_signal_blocked = vmx_apic_init_signal_blocked,
7992 };
7993 
7994 static void vmx_cleanup_l1d_flush(void)
7995 {
7996 	if (vmx_l1d_flush_pages) {
7997 		free_pages((unsigned long)vmx_l1d_flush_pages, L1D_CACHE_ORDER);
7998 		vmx_l1d_flush_pages = NULL;
7999 	}
8000 	/* Restore state so sysfs ignores VMX */
8001 	l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_AUTO;
8002 }
8003 
8004 static void vmx_exit(void)
8005 {
8006 #ifdef CONFIG_KEXEC_CORE
8007 	RCU_INIT_POINTER(crash_vmclear_loaded_vmcss, NULL);
8008 	synchronize_rcu();
8009 #endif
8010 
8011 	kvm_exit();
8012 
8013 #if IS_ENABLED(CONFIG_HYPERV)
8014 	if (static_branch_unlikely(&enable_evmcs)) {
8015 		int cpu;
8016 		struct hv_vp_assist_page *vp_ap;
8017 		/*
8018 		 * Reset everything to support using non-enlightened VMCS
8019 		 * access later (e.g. when we reload the module with
8020 		 * enlightened_vmcs=0)
8021 		 */
8022 		for_each_online_cpu(cpu) {
8023 			vp_ap =	hv_get_vp_assist_page(cpu);
8024 
8025 			if (!vp_ap)
8026 				continue;
8027 
8028 			vp_ap->nested_control.features.directhypercall = 0;
8029 			vp_ap->current_nested_vmcs = 0;
8030 			vp_ap->enlighten_vmentry = 0;
8031 		}
8032 
8033 		static_branch_disable(&enable_evmcs);
8034 	}
8035 #endif
8036 	vmx_cleanup_l1d_flush();
8037 }
8038 module_exit(vmx_exit);
8039 
8040 static int __init vmx_init(void)
8041 {
8042 	int r;
8043 
8044 #if IS_ENABLED(CONFIG_HYPERV)
8045 	/*
8046 	 * Enlightened VMCS usage should be recommended and the host needs
8047 	 * to support eVMCS v1 or above. We can also disable eVMCS support
8048 	 * with module parameter.
8049 	 */
8050 	if (enlightened_vmcs &&
8051 	    ms_hyperv.hints & HV_X64_ENLIGHTENED_VMCS_RECOMMENDED &&
8052 	    (ms_hyperv.nested_features & HV_X64_ENLIGHTENED_VMCS_VERSION) >=
8053 	    KVM_EVMCS_VERSION) {
8054 		int cpu;
8055 
8056 		/* Check that we have assist pages on all online CPUs */
8057 		for_each_online_cpu(cpu) {
8058 			if (!hv_get_vp_assist_page(cpu)) {
8059 				enlightened_vmcs = false;
8060 				break;
8061 			}
8062 		}
8063 
8064 		if (enlightened_vmcs) {
8065 			pr_info("KVM: vmx: using Hyper-V Enlightened VMCS\n");
8066 			static_branch_enable(&enable_evmcs);
8067 		}
8068 
8069 		if (ms_hyperv.nested_features & HV_X64_NESTED_DIRECT_FLUSH)
8070 			vmx_x86_ops.enable_direct_tlbflush
8071 				= hv_enable_direct_tlbflush;
8072 
8073 	} else {
8074 		enlightened_vmcs = false;
8075 	}
8076 #endif
8077 
8078 	r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx),
8079 		     __alignof__(struct vcpu_vmx), THIS_MODULE);
8080 	if (r)
8081 		return r;
8082 
8083 	/*
8084 	 * Must be called after kvm_init() so enable_ept is properly set
8085 	 * up. Hand the parameter mitigation value in which was stored in
8086 	 * the pre module init parser. If no parameter was given, it will
8087 	 * contain 'auto' which will be turned into the default 'cond'
8088 	 * mitigation mode.
8089 	 */
8090 	r = vmx_setup_l1d_flush(vmentry_l1d_flush_param);
8091 	if (r) {
8092 		vmx_exit();
8093 		return r;
8094 	}
8095 
8096 #ifdef CONFIG_KEXEC_CORE
8097 	rcu_assign_pointer(crash_vmclear_loaded_vmcss,
8098 			   crash_vmclear_local_loaded_vmcss);
8099 #endif
8100 	vmx_check_vmcs12_offsets();
8101 
8102 	return 0;
8103 }
8104 module_init(vmx_init);
8105